Root/
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/kthread.h> |
31 | #include <linux/pagemap.h> |
32 | #include <linux/debugfs.h> |
33 | #include <linux/seq_file.h> |
34 | #include <linux/time.h> |
35 | #include <linux/quotaops.h> |
36 | |
37 | #define MLOG_MASK_PREFIX ML_DLM_GLUE |
38 | #include <cluster/masklog.h> |
39 | |
40 | #include "ocfs2.h" |
41 | #include "ocfs2_lockingver.h" |
42 | |
43 | #include "alloc.h" |
44 | #include "dcache.h" |
45 | #include "dlmglue.h" |
46 | #include "extent_map.h" |
47 | #include "file.h" |
48 | #include "heartbeat.h" |
49 | #include "inode.h" |
50 | #include "journal.h" |
51 | #include "stackglue.h" |
52 | #include "slot_map.h" |
53 | #include "super.h" |
54 | #include "uptodate.h" |
55 | #include "quota.h" |
56 | |
57 | #include "buffer_head_io.h" |
58 | |
59 | struct ocfs2_mask_waiter { |
60 | struct list_head mw_item; |
61 | int mw_status; |
62 | struct completion mw_complete; |
63 | unsigned long mw_mask; |
64 | unsigned long mw_goal; |
65 | #ifdef CONFIG_OCFS2_FS_STATS |
66 | unsigned long long mw_lock_start; |
67 | #endif |
68 | }; |
69 | |
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); |
72 | static struct ocfs2_super *ocfs2_get_file_osb(struct ocfs2_lock_res *lockres); |
73 | static struct ocfs2_super *ocfs2_get_qinfo_osb(struct ocfs2_lock_res *lockres); |
74 | |
75 | /* |
76 | * Return value from ->downconvert_worker functions. |
77 | * |
78 | * These control the precise actions of ocfs2_unblock_lock() |
79 | * and ocfs2_process_blocked_lock() |
80 | * |
81 | */ |
82 | enum ocfs2_unblock_action { |
83 | UNBLOCK_CONTINUE = 0, /* Continue downconvert */ |
84 | UNBLOCK_CONTINUE_POST = 1, /* Continue downconvert, fire |
85 | * ->post_unlock callback */ |
86 | UNBLOCK_STOP_POST = 2, /* Do not downconvert, fire |
87 | * ->post_unlock() callback. */ |
88 | }; |
89 | |
90 | struct ocfs2_unblock_ctl { |
91 | int requeue; |
92 | enum ocfs2_unblock_action unblock_action; |
93 | }; |
94 | |
95 | /* Lockdep class keys */ |
96 | struct lock_class_key lockdep_keys[OCFS2_NUM_LOCK_TYPES]; |
97 | |
98 | static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres, |
99 | int new_level); |
100 | static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres); |
101 | |
102 | static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres, |
103 | int blocking); |
104 | |
105 | static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres, |
106 | int blocking); |
107 | |
108 | static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb, |
109 | struct ocfs2_lock_res *lockres); |
110 | |
111 | static void ocfs2_set_qinfo_lvb(struct ocfs2_lock_res *lockres); |
112 | |
113 | #define mlog_meta_lvb(__level, __lockres) ocfs2_dump_meta_lvb_info(__level, __PRETTY_FUNCTION__, __LINE__, __lockres) |
114 | |
115 | /* This aids in debugging situations where a bad LVB might be involved. */ |
116 | static void ocfs2_dump_meta_lvb_info(u64 level, |
117 | const char *function, |
118 | unsigned int line, |
119 | struct ocfs2_lock_res *lockres) |
120 | { |
121 | struct ocfs2_meta_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb); |
122 | |
123 | mlog(level, "LVB information for %s (called from %s:%u):\n", |
124 | lockres->l_name, function, line); |
125 | mlog(level, "version: %u, clusters: %u, generation: 0x%x\n", |
126 | lvb->lvb_version, be32_to_cpu(lvb->lvb_iclusters), |
127 | be32_to_cpu(lvb->lvb_igeneration)); |
128 | mlog(level, "size: %llu, uid %u, gid %u, mode 0x%x\n", |
129 | (unsigned long long)be64_to_cpu(lvb->lvb_isize), |
130 | be32_to_cpu(lvb->lvb_iuid), be32_to_cpu(lvb->lvb_igid), |
131 | be16_to_cpu(lvb->lvb_imode)); |
132 | mlog(level, "nlink %u, atime_packed 0x%llx, ctime_packed 0x%llx, " |
133 | "mtime_packed 0x%llx iattr 0x%x\n", be16_to_cpu(lvb->lvb_inlink), |
134 | (long long)be64_to_cpu(lvb->lvb_iatime_packed), |
135 | (long long)be64_to_cpu(lvb->lvb_ictime_packed), |
136 | (long long)be64_to_cpu(lvb->lvb_imtime_packed), |
137 | be32_to_cpu(lvb->lvb_iattr)); |
138 | } |
139 | |
140 | |
141 | /* |
142 | * OCFS2 Lock Resource Operations |
143 | * |
144 | * These fine tune the behavior of the generic dlmglue locking infrastructure. |
145 | * |
146 | * The most basic of lock types can point ->l_priv to their respective |
147 | * struct ocfs2_super and allow the default actions to manage things. |
148 | * |
149 | * Right now, each lock type also needs to implement an init function, |
150 | * and trivial lock/unlock wrappers. ocfs2_simple_drop_lockres() |
151 | * should be called when the lock is no longer needed (i.e., object |
152 | * destruction time). |
153 | */ |
154 | struct ocfs2_lock_res_ops { |
155 | /* |
156 | * Translate an ocfs2_lock_res * into an ocfs2_super *. Define |
157 | * this callback if ->l_priv is not an ocfs2_super pointer |
158 | */ |
159 | struct ocfs2_super * (*get_osb)(struct ocfs2_lock_res *); |
160 | |
161 | /* |
162 | * Optionally called in the downconvert thread after a |
163 | * successful downconvert. The lockres will not be referenced |
164 | * after this callback is called, so it is safe to free |
165 | * memory, etc. |
166 | * |
167 | * The exact semantics of when this is called are controlled |
168 | * by ->downconvert_worker() |
169 | */ |
170 | void (*post_unlock)(struct ocfs2_super *, struct ocfs2_lock_res *); |
171 | |
172 | /* |
173 | * Allow a lock type to add checks to determine whether it is |
174 | * safe to downconvert a lock. Return 0 to re-queue the |
175 | * downconvert at a later time, nonzero to continue. |
176 | * |
177 | * For most locks, the default checks that there are no |
178 | * incompatible holders are sufficient. |
179 | * |
180 | * Called with the lockres spinlock held. |
181 | */ |
182 | int (*check_downconvert)(struct ocfs2_lock_res *, int); |
183 | |
184 | /* |
185 | * Allows a lock type to populate the lock value block. This |
186 | * is called on downconvert, and when we drop a lock. |
187 | * |
188 | * Locks that want to use this should set LOCK_TYPE_USES_LVB |
189 | * in the flags field. |
190 | * |
191 | * Called with the lockres spinlock held. |
192 | */ |
193 | void (*set_lvb)(struct ocfs2_lock_res *); |
194 | |
195 | /* |
196 | * Called from the downconvert thread when it is determined |
197 | * that a lock will be downconverted. This is called without |
198 | * any locks held so the function can do work that might |
199 | * schedule (syncing out data, etc). |
200 | * |
201 | * This should return any one of the ocfs2_unblock_action |
202 | * values, depending on what it wants the thread to do. |
203 | */ |
204 | int (*downconvert_worker)(struct ocfs2_lock_res *, int); |
205 | |
206 | /* |
207 | * LOCK_TYPE_* flags which describe the specific requirements |
208 | * of a lock type. Descriptions of each individual flag follow. |
209 | */ |
210 | int flags; |
211 | }; |
212 | |
213 | /* |
214 | * Some locks want to "refresh" potentially stale data when a |
215 | * meaningful (PRMODE or EXMODE) lock level is first obtained. If this |
216 | * flag is set, the OCFS2_LOCK_NEEDS_REFRESH flag will be set on the |
217 | * individual lockres l_flags member from the ast function. It is |
218 | * expected that the locking wrapper will clear the |
219 | * OCFS2_LOCK_NEEDS_REFRESH flag when done. |
220 | */ |
221 | #define LOCK_TYPE_REQUIRES_REFRESH 0x1 |
222 | |
223 | /* |
224 | * Indicate that a lock type makes use of the lock value block. The |
225 | * ->set_lvb lock type callback must be defined. |
226 | */ |
227 | #define LOCK_TYPE_USES_LVB 0x2 |
228 | |
229 | static struct ocfs2_lock_res_ops ocfs2_inode_rw_lops = { |
230 | .get_osb = ocfs2_get_inode_osb, |
231 | .flags = 0, |
232 | }; |
233 | |
234 | static struct ocfs2_lock_res_ops ocfs2_inode_inode_lops = { |
235 | .get_osb = ocfs2_get_inode_osb, |
236 | .check_downconvert = ocfs2_check_meta_downconvert, |
237 | .set_lvb = ocfs2_set_meta_lvb, |
238 | .downconvert_worker = ocfs2_data_convert_worker, |
239 | .flags = LOCK_TYPE_REQUIRES_REFRESH|LOCK_TYPE_USES_LVB, |
240 | }; |
241 | |
242 | static struct ocfs2_lock_res_ops ocfs2_super_lops = { |
243 | .flags = LOCK_TYPE_REQUIRES_REFRESH, |
244 | }; |
245 | |
246 | static struct ocfs2_lock_res_ops ocfs2_rename_lops = { |
247 | .flags = 0, |
248 | }; |
249 | |
250 | static struct ocfs2_lock_res_ops ocfs2_nfs_sync_lops = { |
251 | .flags = 0, |
252 | }; |
253 | |
254 | static struct ocfs2_lock_res_ops ocfs2_orphan_scan_lops = { |
255 | .flags = LOCK_TYPE_REQUIRES_REFRESH|LOCK_TYPE_USES_LVB, |
256 | }; |
257 | |
258 | static struct ocfs2_lock_res_ops ocfs2_dentry_lops = { |
259 | .get_osb = ocfs2_get_dentry_osb, |
260 | .post_unlock = ocfs2_dentry_post_unlock, |
261 | .downconvert_worker = ocfs2_dentry_convert_worker, |
262 | .flags = 0, |
263 | }; |
264 | |
265 | static struct ocfs2_lock_res_ops ocfs2_inode_open_lops = { |
266 | .get_osb = ocfs2_get_inode_osb, |
267 | .flags = 0, |
268 | }; |
269 | |
270 | static struct ocfs2_lock_res_ops ocfs2_flock_lops = { |
271 | .get_osb = ocfs2_get_file_osb, |
272 | .flags = 0, |
273 | }; |
274 | |
275 | static struct ocfs2_lock_res_ops ocfs2_qinfo_lops = { |
276 | .set_lvb = ocfs2_set_qinfo_lvb, |
277 | .get_osb = ocfs2_get_qinfo_osb, |
278 | .flags = LOCK_TYPE_REQUIRES_REFRESH | LOCK_TYPE_USES_LVB, |
279 | }; |
280 | |
281 | static inline int ocfs2_is_inode_lock(struct ocfs2_lock_res *lockres) |
282 | { |
283 | return lockres->l_type == OCFS2_LOCK_TYPE_META || |
284 | lockres->l_type == OCFS2_LOCK_TYPE_RW || |
285 | lockres->l_type == OCFS2_LOCK_TYPE_OPEN; |
286 | } |
287 | |
288 | static inline struct inode *ocfs2_lock_res_inode(struct ocfs2_lock_res *lockres) |
289 | { |
290 | BUG_ON(!ocfs2_is_inode_lock(lockres)); |
291 | |
292 | return (struct inode *) lockres->l_priv; |
293 | } |
294 | |
295 | static inline struct ocfs2_dentry_lock *ocfs2_lock_res_dl(struct ocfs2_lock_res *lockres) |
296 | { |
297 | BUG_ON(lockres->l_type != OCFS2_LOCK_TYPE_DENTRY); |
298 | |
299 | return (struct ocfs2_dentry_lock *)lockres->l_priv; |
300 | } |
301 | |
302 | static inline struct ocfs2_mem_dqinfo *ocfs2_lock_res_qinfo(struct ocfs2_lock_res *lockres) |
303 | { |
304 | BUG_ON(lockres->l_type != OCFS2_LOCK_TYPE_QINFO); |
305 | |
306 | return (struct ocfs2_mem_dqinfo *)lockres->l_priv; |
307 | } |
308 | |
309 | static inline struct ocfs2_super *ocfs2_get_lockres_osb(struct ocfs2_lock_res *lockres) |
310 | { |
311 | if (lockres->l_ops->get_osb) |
312 | return lockres->l_ops->get_osb(lockres); |
313 | |
314 | return (struct ocfs2_super *)lockres->l_priv; |
315 | } |
316 | |
317 | static int ocfs2_lock_create(struct ocfs2_super *osb, |
318 | struct ocfs2_lock_res *lockres, |
319 | int level, |
320 | u32 dlm_flags); |
321 | static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres, |
322 | int wanted); |
323 | static void __ocfs2_cluster_unlock(struct ocfs2_super *osb, |
324 | struct ocfs2_lock_res *lockres, |
325 | int level, unsigned long caller_ip); |
326 | static inline void ocfs2_cluster_unlock(struct ocfs2_super *osb, |
327 | struct ocfs2_lock_res *lockres, |
328 | int level) |
329 | { |
330 | __ocfs2_cluster_unlock(osb, lockres, level, _RET_IP_); |
331 | } |
332 | |
333 | static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres); |
334 | static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres); |
335 | static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres); |
336 | static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres, int level); |
337 | static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb, |
338 | struct ocfs2_lock_res *lockres); |
339 | static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres, |
340 | int convert); |
341 | #define ocfs2_log_dlm_error(_func, _err, _lockres) do { \ |
342 | if ((_lockres)->l_type != OCFS2_LOCK_TYPE_DENTRY) \ |
343 | mlog(ML_ERROR, "DLM error %d while calling %s on resource %s\n", \ |
344 | _err, _func, _lockres->l_name); \ |
345 | else \ |
346 | mlog(ML_ERROR, "DLM error %d while calling %s on resource %.*s%08x\n", \ |
347 | _err, _func, OCFS2_DENTRY_LOCK_INO_START - 1, (_lockres)->l_name, \ |
348 | (unsigned int)ocfs2_get_dentry_lock_ino(_lockres)); \ |
349 | } while (0) |
350 | static int ocfs2_downconvert_thread(void *arg); |
351 | static void ocfs2_downconvert_on_unlock(struct ocfs2_super *osb, |
352 | struct ocfs2_lock_res *lockres); |
353 | static int ocfs2_inode_lock_update(struct inode *inode, |
354 | struct buffer_head **bh); |
355 | static void ocfs2_drop_osb_locks(struct ocfs2_super *osb); |
356 | static inline int ocfs2_highest_compat_lock_level(int level); |
357 | static unsigned int ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres, |
358 | int new_level); |
359 | static int ocfs2_downconvert_lock(struct ocfs2_super *osb, |
360 | struct ocfs2_lock_res *lockres, |
361 | int new_level, |
362 | int lvb, |
363 | unsigned int generation); |
364 | static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb, |
365 | struct ocfs2_lock_res *lockres); |
366 | static int ocfs2_cancel_convert(struct ocfs2_super *osb, |
367 | struct ocfs2_lock_res *lockres); |
368 | |
369 | |
370 | static void ocfs2_build_lock_name(enum ocfs2_lock_type type, |
371 | u64 blkno, |
372 | u32 generation, |
373 | char *name) |
374 | { |
375 | int len; |
376 | |
377 | mlog_entry_void(); |
378 | |
379 | BUG_ON(type >= OCFS2_NUM_LOCK_TYPES); |
380 | |
381 | len = snprintf(name, OCFS2_LOCK_ID_MAX_LEN, "%c%s%016llx%08x", |
382 | ocfs2_lock_type_char(type), OCFS2_LOCK_ID_PAD, |
383 | (long long)blkno, generation); |
384 | |
385 | BUG_ON(len != (OCFS2_LOCK_ID_MAX_LEN - 1)); |
386 | |
387 | mlog(0, "built lock resource with name: %s\n", name); |
388 | |
389 | mlog_exit_void(); |
390 | } |
391 | |
392 | static DEFINE_SPINLOCK(ocfs2_dlm_tracking_lock); |
393 | |
394 | static void ocfs2_add_lockres_tracking(struct ocfs2_lock_res *res, |
395 | struct ocfs2_dlm_debug *dlm_debug) |
396 | { |
397 | mlog(0, "Add tracking for lockres %s\n", res->l_name); |
398 | |
399 | spin_lock(&ocfs2_dlm_tracking_lock); |
400 | list_add(&res->l_debug_list, &dlm_debug->d_lockres_tracking); |
401 | spin_unlock(&ocfs2_dlm_tracking_lock); |
402 | } |
403 | |
404 | static void ocfs2_remove_lockres_tracking(struct ocfs2_lock_res *res) |
405 | { |
406 | spin_lock(&ocfs2_dlm_tracking_lock); |
407 | if (!list_empty(&res->l_debug_list)) |
408 | list_del_init(&res->l_debug_list); |
409 | spin_unlock(&ocfs2_dlm_tracking_lock); |
410 | } |
411 | |
412 | #ifdef CONFIG_OCFS2_FS_STATS |
413 | static void ocfs2_init_lock_stats(struct ocfs2_lock_res *res) |
414 | { |
415 | res->l_lock_num_prmode = 0; |
416 | res->l_lock_num_prmode_failed = 0; |
417 | res->l_lock_total_prmode = 0; |
418 | res->l_lock_max_prmode = 0; |
419 | res->l_lock_num_exmode = 0; |
420 | res->l_lock_num_exmode_failed = 0; |
421 | res->l_lock_total_exmode = 0; |
422 | res->l_lock_max_exmode = 0; |
423 | res->l_lock_refresh = 0; |
424 | } |
425 | |
426 | static void ocfs2_update_lock_stats(struct ocfs2_lock_res *res, int level, |
427 | struct ocfs2_mask_waiter *mw, int ret) |
428 | { |
429 | unsigned long long *num, *sum; |
430 | unsigned int *max, *failed; |
431 | struct timespec ts = current_kernel_time(); |
432 | unsigned long long time = timespec_to_ns(&ts) - mw->mw_lock_start; |
433 | |
434 | if (level == LKM_PRMODE) { |
435 | num = &res->l_lock_num_prmode; |
436 | sum = &res->l_lock_total_prmode; |
437 | max = &res->l_lock_max_prmode; |
438 | failed = &res->l_lock_num_prmode_failed; |
439 | } else if (level == LKM_EXMODE) { |
440 | num = &res->l_lock_num_exmode; |
441 | sum = &res->l_lock_total_exmode; |
442 | max = &res->l_lock_max_exmode; |
443 | failed = &res->l_lock_num_exmode_failed; |
444 | } else |
445 | return; |
446 | |
447 | (*num)++; |
448 | (*sum) += time; |
449 | if (time > *max) |
450 | *max = time; |
451 | if (ret) |
452 | (*failed)++; |
453 | } |
454 | |
455 | static inline void ocfs2_track_lock_refresh(struct ocfs2_lock_res *lockres) |
456 | { |
457 | lockres->l_lock_refresh++; |
458 | } |
459 | |
460 | static inline void ocfs2_init_start_time(struct ocfs2_mask_waiter *mw) |
461 | { |
462 | struct timespec ts = current_kernel_time(); |
463 | mw->mw_lock_start = timespec_to_ns(&ts); |
464 | } |
465 | #else |
466 | static inline void ocfs2_init_lock_stats(struct ocfs2_lock_res *res) |
467 | { |
468 | } |
469 | static inline void ocfs2_update_lock_stats(struct ocfs2_lock_res *res, |
470 | int level, struct ocfs2_mask_waiter *mw, int ret) |
471 | { |
472 | } |
473 | static inline void ocfs2_track_lock_refresh(struct ocfs2_lock_res *lockres) |
474 | { |
475 | } |
476 | static inline void ocfs2_init_start_time(struct ocfs2_mask_waiter *mw) |
477 | { |
478 | } |
479 | #endif |
480 | |
481 | static void ocfs2_lock_res_init_common(struct ocfs2_super *osb, |
482 | struct ocfs2_lock_res *res, |
483 | enum ocfs2_lock_type type, |
484 | struct ocfs2_lock_res_ops *ops, |
485 | void *priv) |
486 | { |
487 | res->l_type = type; |
488 | res->l_ops = ops; |
489 | res->l_priv = priv; |
490 | |
491 | res->l_level = DLM_LOCK_IV; |
492 | res->l_requested = DLM_LOCK_IV; |
493 | res->l_blocking = DLM_LOCK_IV; |
494 | res->l_action = OCFS2_AST_INVALID; |
495 | res->l_unlock_action = OCFS2_UNLOCK_INVALID; |
496 | |
497 | res->l_flags = OCFS2_LOCK_INITIALIZED; |
498 | |
499 | ocfs2_add_lockres_tracking(res, osb->osb_dlm_debug); |
500 | |
501 | ocfs2_init_lock_stats(res); |
502 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
503 | if (type != OCFS2_LOCK_TYPE_OPEN) |
504 | lockdep_init_map(&res->l_lockdep_map, ocfs2_lock_type_strings[type], |
505 | &lockdep_keys[type], 0); |
506 | else |
507 | res->l_lockdep_map.key = NULL; |
508 | #endif |
509 | } |
510 | |
511 | void ocfs2_lock_res_init_once(struct ocfs2_lock_res *res) |
512 | { |
513 | /* This also clears out the lock status block */ |
514 | memset(res, 0, sizeof(struct ocfs2_lock_res)); |
515 | spin_lock_init(&res->l_lock); |
516 | init_waitqueue_head(&res->l_event); |
517 | INIT_LIST_HEAD(&res->l_blocked_list); |
518 | INIT_LIST_HEAD(&res->l_mask_waiters); |
519 | } |
520 | |
521 | void ocfs2_inode_lock_res_init(struct ocfs2_lock_res *res, |
522 | enum ocfs2_lock_type type, |
523 | unsigned int generation, |
524 | struct inode *inode) |
525 | { |
526 | struct ocfs2_lock_res_ops *ops; |
527 | |
528 | switch(type) { |
529 | case OCFS2_LOCK_TYPE_RW: |
530 | ops = &ocfs2_inode_rw_lops; |
531 | break; |
532 | case OCFS2_LOCK_TYPE_META: |
533 | ops = &ocfs2_inode_inode_lops; |
534 | break; |
535 | case OCFS2_LOCK_TYPE_OPEN: |
536 | ops = &ocfs2_inode_open_lops; |
537 | break; |
538 | default: |
539 | mlog_bug_on_msg(1, "type: %d\n", type); |
540 | ops = NULL; /* thanks, gcc */ |
541 | break; |
542 | }; |
543 | |
544 | ocfs2_build_lock_name(type, OCFS2_I(inode)->ip_blkno, |
545 | generation, res->l_name); |
546 | ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), res, type, ops, inode); |
547 | } |
548 | |
549 | static struct ocfs2_super *ocfs2_get_inode_osb(struct ocfs2_lock_res *lockres) |
550 | { |
551 | struct inode *inode = ocfs2_lock_res_inode(lockres); |
552 | |
553 | return OCFS2_SB(inode->i_sb); |
554 | } |
555 | |
556 | static struct ocfs2_super *ocfs2_get_qinfo_osb(struct ocfs2_lock_res *lockres) |
557 | { |
558 | struct ocfs2_mem_dqinfo *info = lockres->l_priv; |
559 | |
560 | return OCFS2_SB(info->dqi_gi.dqi_sb); |
561 | } |
562 | |
563 | static struct ocfs2_super *ocfs2_get_file_osb(struct ocfs2_lock_res *lockres) |
564 | { |
565 | struct ocfs2_file_private *fp = lockres->l_priv; |
566 | |
567 | return OCFS2_SB(fp->fp_file->f_mapping->host->i_sb); |
568 | } |
569 | |
570 | static __u64 ocfs2_get_dentry_lock_ino(struct ocfs2_lock_res *lockres) |
571 | { |
572 | __be64 inode_blkno_be; |
573 | |
574 | memcpy(&inode_blkno_be, &lockres->l_name[OCFS2_DENTRY_LOCK_INO_START], |
575 | sizeof(__be64)); |
576 | |
577 | return be64_to_cpu(inode_blkno_be); |
578 | } |
579 | |
580 | static struct ocfs2_super *ocfs2_get_dentry_osb(struct ocfs2_lock_res *lockres) |
581 | { |
582 | struct ocfs2_dentry_lock *dl = lockres->l_priv; |
583 | |
584 | return OCFS2_SB(dl->dl_inode->i_sb); |
585 | } |
586 | |
587 | void ocfs2_dentry_lock_res_init(struct ocfs2_dentry_lock *dl, |
588 | u64 parent, struct inode *inode) |
589 | { |
590 | int len; |
591 | u64 inode_blkno = OCFS2_I(inode)->ip_blkno; |
592 | __be64 inode_blkno_be = cpu_to_be64(inode_blkno); |
593 | struct ocfs2_lock_res *lockres = &dl->dl_lockres; |
594 | |
595 | ocfs2_lock_res_init_once(lockres); |
596 | |
597 | /* |
598 | * Unfortunately, the standard lock naming scheme won't work |
599 | * here because we have two 16 byte values to use. Instead, |
600 | * we'll stuff the inode number as a binary value. We still |
601 | * want error prints to show something without garbling the |
602 | * display, so drop a null byte in there before the inode |
603 | * number. A future version of OCFS2 will likely use all |
604 | * binary lock names. The stringified names have been a |
605 | * tremendous aid in debugging, but now that the debugfs |
606 | * interface exists, we can mangle things there if need be. |
607 | * |
608 | * NOTE: We also drop the standard "pad" value (the total lock |
609 | * name size stays the same though - the last part is all |
610 | * zeros due to the memset in ocfs2_lock_res_init_once() |
611 | */ |
612 | len = snprintf(lockres->l_name, OCFS2_DENTRY_LOCK_INO_START, |
613 | "%c%016llx", |
614 | ocfs2_lock_type_char(OCFS2_LOCK_TYPE_DENTRY), |
615 | (long long)parent); |
616 | |
617 | BUG_ON(len != (OCFS2_DENTRY_LOCK_INO_START - 1)); |
618 | |
619 | memcpy(&lockres->l_name[OCFS2_DENTRY_LOCK_INO_START], &inode_blkno_be, |
620 | sizeof(__be64)); |
621 | |
622 | ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), lockres, |
623 | OCFS2_LOCK_TYPE_DENTRY, &ocfs2_dentry_lops, |
624 | dl); |
625 | } |
626 | |
627 | static void ocfs2_super_lock_res_init(struct ocfs2_lock_res *res, |
628 | struct ocfs2_super *osb) |
629 | { |
630 | /* Superblock lockres doesn't come from a slab so we call init |
631 | * once on it manually. */ |
632 | ocfs2_lock_res_init_once(res); |
633 | ocfs2_build_lock_name(OCFS2_LOCK_TYPE_SUPER, OCFS2_SUPER_BLOCK_BLKNO, |
634 | 0, res->l_name); |
635 | ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_SUPER, |
636 | &ocfs2_super_lops, osb); |
637 | } |
638 | |
639 | static void ocfs2_rename_lock_res_init(struct ocfs2_lock_res *res, |
640 | struct ocfs2_super *osb) |
641 | { |
642 | /* Rename lockres doesn't come from a slab so we call init |
643 | * once on it manually. */ |
644 | ocfs2_lock_res_init_once(res); |
645 | ocfs2_build_lock_name(OCFS2_LOCK_TYPE_RENAME, 0, 0, res->l_name); |
646 | ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_RENAME, |
647 | &ocfs2_rename_lops, osb); |
648 | } |
649 | |
650 | static void ocfs2_nfs_sync_lock_res_init(struct ocfs2_lock_res *res, |
651 | struct ocfs2_super *osb) |
652 | { |
653 | /* nfs_sync lockres doesn't come from a slab so we call init |
654 | * once on it manually. */ |
655 | ocfs2_lock_res_init_once(res); |
656 | ocfs2_build_lock_name(OCFS2_LOCK_TYPE_NFS_SYNC, 0, 0, res->l_name); |
657 | ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_NFS_SYNC, |
658 | &ocfs2_nfs_sync_lops, osb); |
659 | } |
660 | |
661 | static void ocfs2_orphan_scan_lock_res_init(struct ocfs2_lock_res *res, |
662 | struct ocfs2_super *osb) |
663 | { |
664 | ocfs2_lock_res_init_once(res); |
665 | ocfs2_build_lock_name(OCFS2_LOCK_TYPE_ORPHAN_SCAN, 0, 0, res->l_name); |
666 | ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_ORPHAN_SCAN, |
667 | &ocfs2_orphan_scan_lops, osb); |
668 | } |
669 | |
670 | void ocfs2_file_lock_res_init(struct ocfs2_lock_res *lockres, |
671 | struct ocfs2_file_private *fp) |
672 | { |
673 | struct inode *inode = fp->fp_file->f_mapping->host; |
674 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
675 | |
676 | ocfs2_lock_res_init_once(lockres); |
677 | ocfs2_build_lock_name(OCFS2_LOCK_TYPE_FLOCK, oi->ip_blkno, |
678 | inode->i_generation, lockres->l_name); |
679 | ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), lockres, |
680 | OCFS2_LOCK_TYPE_FLOCK, &ocfs2_flock_lops, |
681 | fp); |
682 | lockres->l_flags |= OCFS2_LOCK_NOCACHE; |
683 | } |
684 | |
685 | void ocfs2_qinfo_lock_res_init(struct ocfs2_lock_res *lockres, |
686 | struct ocfs2_mem_dqinfo *info) |
687 | { |
688 | ocfs2_lock_res_init_once(lockres); |
689 | ocfs2_build_lock_name(OCFS2_LOCK_TYPE_QINFO, info->dqi_gi.dqi_type, |
690 | 0, lockres->l_name); |
691 | ocfs2_lock_res_init_common(OCFS2_SB(info->dqi_gi.dqi_sb), lockres, |
692 | OCFS2_LOCK_TYPE_QINFO, &ocfs2_qinfo_lops, |
693 | info); |
694 | } |
695 | |
696 | void ocfs2_lock_res_free(struct ocfs2_lock_res *res) |
697 | { |
698 | mlog_entry_void(); |
699 | |
700 | if (!(res->l_flags & OCFS2_LOCK_INITIALIZED)) |
701 | return; |
702 | |
703 | ocfs2_remove_lockres_tracking(res); |
704 | |
705 | mlog_bug_on_msg(!list_empty(&res->l_blocked_list), |
706 | "Lockres %s is on the blocked list\n", |
707 | res->l_name); |
708 | mlog_bug_on_msg(!list_empty(&res->l_mask_waiters), |
709 | "Lockres %s has mask waiters pending\n", |
710 | res->l_name); |
711 | mlog_bug_on_msg(spin_is_locked(&res->l_lock), |
712 | "Lockres %s is locked\n", |
713 | res->l_name); |
714 | mlog_bug_on_msg(res->l_ro_holders, |
715 | "Lockres %s has %u ro holders\n", |
716 | res->l_name, res->l_ro_holders); |
717 | mlog_bug_on_msg(res->l_ex_holders, |
718 | "Lockres %s has %u ex holders\n", |
719 | res->l_name, res->l_ex_holders); |
720 | |
721 | /* Need to clear out the lock status block for the dlm */ |
722 | memset(&res->l_lksb, 0, sizeof(res->l_lksb)); |
723 | |
724 | res->l_flags = 0UL; |
725 | mlog_exit_void(); |
726 | } |
727 | |
728 | static inline void ocfs2_inc_holders(struct ocfs2_lock_res *lockres, |
729 | int level) |
730 | { |
731 | mlog_entry_void(); |
732 | |
733 | BUG_ON(!lockres); |
734 | |
735 | switch(level) { |
736 | case DLM_LOCK_EX: |
737 | lockres->l_ex_holders++; |
738 | break; |
739 | case DLM_LOCK_PR: |
740 | lockres->l_ro_holders++; |
741 | break; |
742 | default: |
743 | BUG(); |
744 | } |
745 | |
746 | mlog_exit_void(); |
747 | } |
748 | |
749 | static inline void ocfs2_dec_holders(struct ocfs2_lock_res *lockres, |
750 | int level) |
751 | { |
752 | mlog_entry_void(); |
753 | |
754 | BUG_ON(!lockres); |
755 | |
756 | switch(level) { |
757 | case DLM_LOCK_EX: |
758 | BUG_ON(!lockres->l_ex_holders); |
759 | lockres->l_ex_holders--; |
760 | break; |
761 | case DLM_LOCK_PR: |
762 | BUG_ON(!lockres->l_ro_holders); |
763 | lockres->l_ro_holders--; |
764 | break; |
765 | default: |
766 | BUG(); |
767 | } |
768 | mlog_exit_void(); |
769 | } |
770 | |
771 | /* WARNING: This function lives in a world where the only three lock |
772 | * levels are EX, PR, and NL. It *will* have to be adjusted when more |
773 | * lock types are added. */ |
774 | static inline int ocfs2_highest_compat_lock_level(int level) |
775 | { |
776 | int new_level = DLM_LOCK_EX; |
777 | |
778 | if (level == DLM_LOCK_EX) |
779 | new_level = DLM_LOCK_NL; |
780 | else if (level == DLM_LOCK_PR) |
781 | new_level = DLM_LOCK_PR; |
782 | return new_level; |
783 | } |
784 | |
785 | static void lockres_set_flags(struct ocfs2_lock_res *lockres, |
786 | unsigned long newflags) |
787 | { |
788 | struct ocfs2_mask_waiter *mw, *tmp; |
789 | |
790 | assert_spin_locked(&lockres->l_lock); |
791 | |
792 | lockres->l_flags = newflags; |
793 | |
794 | list_for_each_entry_safe(mw, tmp, &lockres->l_mask_waiters, mw_item) { |
795 | if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal) |
796 | continue; |
797 | |
798 | list_del_init(&mw->mw_item); |
799 | mw->mw_status = 0; |
800 | complete(&mw->mw_complete); |
801 | } |
802 | } |
803 | static void lockres_or_flags(struct ocfs2_lock_res *lockres, unsigned long or) |
804 | { |
805 | lockres_set_flags(lockres, lockres->l_flags | or); |
806 | } |
807 | static void lockres_clear_flags(struct ocfs2_lock_res *lockres, |
808 | unsigned long clear) |
809 | { |
810 | lockres_set_flags(lockres, lockres->l_flags & ~clear); |
811 | } |
812 | |
813 | static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres) |
814 | { |
815 | mlog_entry_void(); |
816 | |
817 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY)); |
818 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED)); |
819 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED)); |
820 | BUG_ON(lockres->l_blocking <= DLM_LOCK_NL); |
821 | |
822 | lockres->l_level = lockres->l_requested; |
823 | if (lockres->l_level <= |
824 | ocfs2_highest_compat_lock_level(lockres->l_blocking)) { |
825 | lockres->l_blocking = DLM_LOCK_NL; |
826 | lockres_clear_flags(lockres, OCFS2_LOCK_BLOCKED); |
827 | } |
828 | lockres_clear_flags(lockres, OCFS2_LOCK_BUSY); |
829 | |
830 | mlog_exit_void(); |
831 | } |
832 | |
833 | static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres) |
834 | { |
835 | mlog_entry_void(); |
836 | |
837 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY)); |
838 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED)); |
839 | |
840 | /* Convert from RO to EX doesn't really need anything as our |
841 | * information is already up to data. Convert from NL to |
842 | * *anything* however should mark ourselves as needing an |
843 | * update */ |
844 | if (lockres->l_level == DLM_LOCK_NL && |
845 | lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH) |
846 | lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH); |
847 | |
848 | lockres->l_level = lockres->l_requested; |
849 | lockres_clear_flags(lockres, OCFS2_LOCK_BUSY); |
850 | |
851 | mlog_exit_void(); |
852 | } |
853 | |
854 | static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres) |
855 | { |
856 | mlog_entry_void(); |
857 | |
858 | BUG_ON((!(lockres->l_flags & OCFS2_LOCK_BUSY))); |
859 | BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED); |
860 | |
861 | if (lockres->l_requested > DLM_LOCK_NL && |
862 | !(lockres->l_flags & OCFS2_LOCK_LOCAL) && |
863 | lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH) |
864 | lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH); |
865 | |
866 | lockres->l_level = lockres->l_requested; |
867 | lockres_or_flags(lockres, OCFS2_LOCK_ATTACHED); |
868 | lockres_clear_flags(lockres, OCFS2_LOCK_BUSY); |
869 | |
870 | mlog_exit_void(); |
871 | } |
872 | |
873 | static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres, |
874 | int level) |
875 | { |
876 | int needs_downconvert = 0; |
877 | mlog_entry_void(); |
878 | |
879 | assert_spin_locked(&lockres->l_lock); |
880 | |
881 | lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED); |
882 | |
883 | if (level > lockres->l_blocking) { |
884 | /* only schedule a downconvert if we haven't already scheduled |
885 | * one that goes low enough to satisfy the level we're |
886 | * blocking. this also catches the case where we get |
887 | * duplicate BASTs */ |
888 | if (ocfs2_highest_compat_lock_level(level) < |
889 | ocfs2_highest_compat_lock_level(lockres->l_blocking)) |
890 | needs_downconvert = 1; |
891 | |
892 | lockres->l_blocking = level; |
893 | } |
894 | |
895 | mlog_exit(needs_downconvert); |
896 | return needs_downconvert; |
897 | } |
898 | |
899 | /* |
900 | * OCFS2_LOCK_PENDING and l_pending_gen. |
901 | * |
902 | * Why does OCFS2_LOCK_PENDING exist? To close a race between setting |
903 | * OCFS2_LOCK_BUSY and calling ocfs2_dlm_lock(). See ocfs2_unblock_lock() |
904 | * for more details on the race. |
905 | * |
906 | * OCFS2_LOCK_PENDING closes the race quite nicely. However, it introduces |
907 | * a race on itself. In o2dlm, we can get the ast before ocfs2_dlm_lock() |
908 | * returns. The ast clears OCFS2_LOCK_BUSY, and must therefore clear |
909 | * OCFS2_LOCK_PENDING at the same time. When ocfs2_dlm_lock() returns, |
910 | * the caller is going to try to clear PENDING again. If nothing else is |
911 | * happening, __lockres_clear_pending() sees PENDING is unset and does |
912 | * nothing. |
913 | * |
914 | * But what if another path (eg downconvert thread) has just started a |
915 | * new locking action? The other path has re-set PENDING. Our path |
916 | * cannot clear PENDING, because that will re-open the original race |
917 | * window. |
918 | * |
919 | * [Example] |
920 | * |
921 | * ocfs2_meta_lock() |
922 | * ocfs2_cluster_lock() |
923 | * set BUSY |
924 | * set PENDING |
925 | * drop l_lock |
926 | * ocfs2_dlm_lock() |
927 | * ocfs2_locking_ast() ocfs2_downconvert_thread() |
928 | * clear PENDING ocfs2_unblock_lock() |
929 | * take_l_lock |
930 | * !BUSY |
931 | * ocfs2_prepare_downconvert() |
932 | * set BUSY |
933 | * set PENDING |
934 | * drop l_lock |
935 | * take l_lock |
936 | * clear PENDING |
937 | * drop l_lock |
938 | * <window> |
939 | * ocfs2_dlm_lock() |
940 | * |
941 | * So as you can see, we now have a window where l_lock is not held, |
942 | * PENDING is not set, and ocfs2_dlm_lock() has not been called. |
943 | * |
944 | * The core problem is that ocfs2_cluster_lock() has cleared the PENDING |
945 | * set by ocfs2_prepare_downconvert(). That wasn't nice. |
946 | * |
947 | * To solve this we introduce l_pending_gen. A call to |
948 | * lockres_clear_pending() will only do so when it is passed a generation |
949 | * number that matches the lockres. lockres_set_pending() will return the |
950 | * current generation number. When ocfs2_cluster_lock() goes to clear |
951 | * PENDING, it passes the generation it got from set_pending(). In our |
952 | * example above, the generation numbers will *not* match. Thus, |
953 | * ocfs2_cluster_lock() will not clear the PENDING set by |
954 | * ocfs2_prepare_downconvert(). |
955 | */ |
956 | |
957 | /* Unlocked version for ocfs2_locking_ast() */ |
958 | static void __lockres_clear_pending(struct ocfs2_lock_res *lockres, |
959 | unsigned int generation, |
960 | struct ocfs2_super *osb) |
961 | { |
962 | assert_spin_locked(&lockres->l_lock); |
963 | |
964 | /* |
965 | * The ast and locking functions can race us here. The winner |
966 | * will clear pending, the loser will not. |
967 | */ |
968 | if (!(lockres->l_flags & OCFS2_LOCK_PENDING) || |
969 | (lockres->l_pending_gen != generation)) |
970 | return; |
971 | |
972 | lockres_clear_flags(lockres, OCFS2_LOCK_PENDING); |
973 | lockres->l_pending_gen++; |
974 | |
975 | /* |
976 | * The downconvert thread may have skipped us because we |
977 | * were PENDING. Wake it up. |
978 | */ |
979 | if (lockres->l_flags & OCFS2_LOCK_BLOCKED) |
980 | ocfs2_wake_downconvert_thread(osb); |
981 | } |
982 | |
983 | /* Locked version for callers of ocfs2_dlm_lock() */ |
984 | static void lockres_clear_pending(struct ocfs2_lock_res *lockres, |
985 | unsigned int generation, |
986 | struct ocfs2_super *osb) |
987 | { |
988 | unsigned long flags; |
989 | |
990 | spin_lock_irqsave(&lockres->l_lock, flags); |
991 | __lockres_clear_pending(lockres, generation, osb); |
992 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
993 | } |
994 | |
995 | static unsigned int lockres_set_pending(struct ocfs2_lock_res *lockres) |
996 | { |
997 | assert_spin_locked(&lockres->l_lock); |
998 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY)); |
999 | |
1000 | lockres_or_flags(lockres, OCFS2_LOCK_PENDING); |
1001 | |
1002 | return lockres->l_pending_gen; |
1003 | } |
1004 | |
1005 | |
1006 | static void ocfs2_blocking_ast(void *opaque, int level) |
1007 | { |
1008 | struct ocfs2_lock_res *lockres = opaque; |
1009 | struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres); |
1010 | int needs_downconvert; |
1011 | unsigned long flags; |
1012 | |
1013 | BUG_ON(level <= DLM_LOCK_NL); |
1014 | |
1015 | mlog(0, "BAST fired for lockres %s, blocking %d, level %d type %s\n", |
1016 | lockres->l_name, level, lockres->l_level, |
1017 | ocfs2_lock_type_string(lockres->l_type)); |
1018 | |
1019 | /* |
1020 | * We can skip the bast for locks which don't enable caching - |
1021 | * they'll be dropped at the earliest possible time anyway. |
1022 | */ |
1023 | if (lockres->l_flags & OCFS2_LOCK_NOCACHE) |
1024 | return; |
1025 | |
1026 | spin_lock_irqsave(&lockres->l_lock, flags); |
1027 | needs_downconvert = ocfs2_generic_handle_bast(lockres, level); |
1028 | if (needs_downconvert) |
1029 | ocfs2_schedule_blocked_lock(osb, lockres); |
1030 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
1031 | |
1032 | wake_up(&lockres->l_event); |
1033 | |
1034 | ocfs2_wake_downconvert_thread(osb); |
1035 | } |
1036 | |
1037 | static void ocfs2_locking_ast(void *opaque) |
1038 | { |
1039 | struct ocfs2_lock_res *lockres = opaque; |
1040 | struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres); |
1041 | unsigned long flags; |
1042 | int status; |
1043 | |
1044 | spin_lock_irqsave(&lockres->l_lock, flags); |
1045 | |
1046 | status = ocfs2_dlm_lock_status(&lockres->l_lksb); |
1047 | |
1048 | if (status == -EAGAIN) { |
1049 | lockres_clear_flags(lockres, OCFS2_LOCK_BUSY); |
1050 | goto out; |
1051 | } |
1052 | |
1053 | if (status) { |
1054 | mlog(ML_ERROR, "lockres %s: lksb status value of %d!\n", |
1055 | lockres->l_name, status); |
1056 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
1057 | return; |
1058 | } |
1059 | |
1060 | switch(lockres->l_action) { |
1061 | case OCFS2_AST_ATTACH: |
1062 | ocfs2_generic_handle_attach_action(lockres); |
1063 | lockres_clear_flags(lockres, OCFS2_LOCK_LOCAL); |
1064 | break; |
1065 | case OCFS2_AST_CONVERT: |
1066 | ocfs2_generic_handle_convert_action(lockres); |
1067 | break; |
1068 | case OCFS2_AST_DOWNCONVERT: |
1069 | ocfs2_generic_handle_downconvert_action(lockres); |
1070 | break; |
1071 | default: |
1072 | mlog(ML_ERROR, "lockres %s: ast fired with invalid action: %u " |
1073 | "lockres flags = 0x%lx, unlock action: %u\n", |
1074 | lockres->l_name, lockres->l_action, lockres->l_flags, |
1075 | lockres->l_unlock_action); |
1076 | BUG(); |
1077 | } |
1078 | out: |
1079 | /* set it to something invalid so if we get called again we |
1080 | * can catch it. */ |
1081 | lockres->l_action = OCFS2_AST_INVALID; |
1082 | |
1083 | /* Did we try to cancel this lock? Clear that state */ |
1084 | if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) |
1085 | lockres->l_unlock_action = OCFS2_UNLOCK_INVALID; |
1086 | |
1087 | /* |
1088 | * We may have beaten the locking functions here. We certainly |
1089 | * know that dlm_lock() has been called :-) |
1090 | * Because we can't have two lock calls in flight at once, we |
1091 | * can use lockres->l_pending_gen. |
1092 | */ |
1093 | __lockres_clear_pending(lockres, lockres->l_pending_gen, osb); |
1094 | |
1095 | wake_up(&lockres->l_event); |
1096 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
1097 | } |
1098 | |
1099 | static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres, |
1100 | int convert) |
1101 | { |
1102 | unsigned long flags; |
1103 | |
1104 | mlog_entry_void(); |
1105 | spin_lock_irqsave(&lockres->l_lock, flags); |
1106 | lockres_clear_flags(lockres, OCFS2_LOCK_BUSY); |
1107 | if (convert) |
1108 | lockres->l_action = OCFS2_AST_INVALID; |
1109 | else |
1110 | lockres->l_unlock_action = OCFS2_UNLOCK_INVALID; |
1111 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
1112 | |
1113 | wake_up(&lockres->l_event); |
1114 | mlog_exit_void(); |
1115 | } |
1116 | |
1117 | /* Note: If we detect another process working on the lock (i.e., |
1118 | * OCFS2_LOCK_BUSY), we'll bail out returning 0. It's up to the caller |
1119 | * to do the right thing in that case. |
1120 | */ |
1121 | static int ocfs2_lock_create(struct ocfs2_super *osb, |
1122 | struct ocfs2_lock_res *lockres, |
1123 | int level, |
1124 | u32 dlm_flags) |
1125 | { |
1126 | int ret = 0; |
1127 | unsigned long flags; |
1128 | unsigned int gen; |
1129 | |
1130 | mlog_entry_void(); |
1131 | |
1132 | mlog(0, "lock %s, level = %d, flags = %u\n", lockres->l_name, level, |
1133 | dlm_flags); |
1134 | |
1135 | spin_lock_irqsave(&lockres->l_lock, flags); |
1136 | if ((lockres->l_flags & OCFS2_LOCK_ATTACHED) || |
1137 | (lockres->l_flags & OCFS2_LOCK_BUSY)) { |
1138 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
1139 | goto bail; |
1140 | } |
1141 | |
1142 | lockres->l_action = OCFS2_AST_ATTACH; |
1143 | lockres->l_requested = level; |
1144 | lockres_or_flags(lockres, OCFS2_LOCK_BUSY); |
1145 | gen = lockres_set_pending(lockres); |
1146 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
1147 | |
1148 | ret = ocfs2_dlm_lock(osb->cconn, |
1149 | level, |
1150 | &lockres->l_lksb, |
1151 | dlm_flags, |
1152 | lockres->l_name, |
1153 | OCFS2_LOCK_ID_MAX_LEN - 1, |
1154 | lockres); |
1155 | lockres_clear_pending(lockres, gen, osb); |
1156 | if (ret) { |
1157 | ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres); |
1158 | ocfs2_recover_from_dlm_error(lockres, 1); |
1159 | } |
1160 | |
1161 | mlog(0, "lock %s, return from ocfs2_dlm_lock\n", lockres->l_name); |
1162 | |
1163 | bail: |
1164 | mlog_exit(ret); |
1165 | return ret; |
1166 | } |
1167 | |
1168 | static inline int ocfs2_check_wait_flag(struct ocfs2_lock_res *lockres, |
1169 | int flag) |
1170 | { |
1171 | unsigned long flags; |
1172 | int ret; |
1173 | |
1174 | spin_lock_irqsave(&lockres->l_lock, flags); |
1175 | ret = lockres->l_flags & flag; |
1176 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
1177 | |
1178 | return ret; |
1179 | } |
1180 | |
1181 | static inline void ocfs2_wait_on_busy_lock(struct ocfs2_lock_res *lockres) |
1182 | |
1183 | { |
1184 | wait_event(lockres->l_event, |
1185 | !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_BUSY)); |
1186 | } |
1187 | |
1188 | static inline void ocfs2_wait_on_refreshing_lock(struct ocfs2_lock_res *lockres) |
1189 | |
1190 | { |
1191 | wait_event(lockres->l_event, |
1192 | !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_REFRESHING)); |
1193 | } |
1194 | |
1195 | /* predict what lock level we'll be dropping down to on behalf |
1196 | * of another node, and return true if the currently wanted |
1197 | * level will be compatible with it. */ |
1198 | static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres, |
1199 | int wanted) |
1200 | { |
1201 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED)); |
1202 | |
1203 | return wanted <= ocfs2_highest_compat_lock_level(lockres->l_blocking); |
1204 | } |
1205 | |
1206 | static void ocfs2_init_mask_waiter(struct ocfs2_mask_waiter *mw) |
1207 | { |
1208 | INIT_LIST_HEAD(&mw->mw_item); |
1209 | init_completion(&mw->mw_complete); |
1210 | ocfs2_init_start_time(mw); |
1211 | } |
1212 | |
1213 | static int ocfs2_wait_for_mask(struct ocfs2_mask_waiter *mw) |
1214 | { |
1215 | wait_for_completion(&mw->mw_complete); |
1216 | /* Re-arm the completion in case we want to wait on it again */ |
1217 | INIT_COMPLETION(mw->mw_complete); |
1218 | return mw->mw_status; |
1219 | } |
1220 | |
1221 | static void lockres_add_mask_waiter(struct ocfs2_lock_res *lockres, |
1222 | struct ocfs2_mask_waiter *mw, |
1223 | unsigned long mask, |
1224 | unsigned long goal) |
1225 | { |
1226 | BUG_ON(!list_empty(&mw->mw_item)); |
1227 | |
1228 | assert_spin_locked(&lockres->l_lock); |
1229 | |
1230 | list_add_tail(&mw->mw_item, &lockres->l_mask_waiters); |
1231 | mw->mw_mask = mask; |
1232 | mw->mw_goal = goal; |
1233 | } |
1234 | |
1235 | /* returns 0 if the mw that was removed was already satisfied, -EBUSY |
1236 | * if the mask still hadn't reached its goal */ |
1237 | static int lockres_remove_mask_waiter(struct ocfs2_lock_res *lockres, |
1238 | struct ocfs2_mask_waiter *mw) |
1239 | { |
1240 | unsigned long flags; |
1241 | int ret = 0; |
1242 | |
1243 | spin_lock_irqsave(&lockres->l_lock, flags); |
1244 | if (!list_empty(&mw->mw_item)) { |
1245 | if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal) |
1246 | ret = -EBUSY; |
1247 | |
1248 | list_del_init(&mw->mw_item); |
1249 | init_completion(&mw->mw_complete); |
1250 | } |
1251 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
1252 | |
1253 | return ret; |
1254 | |
1255 | } |
1256 | |
1257 | static int ocfs2_wait_for_mask_interruptible(struct ocfs2_mask_waiter *mw, |
1258 | struct ocfs2_lock_res *lockres) |
1259 | { |
1260 | int ret; |
1261 | |
1262 | ret = wait_for_completion_interruptible(&mw->mw_complete); |
1263 | if (ret) |
1264 | lockres_remove_mask_waiter(lockres, mw); |
1265 | else |
1266 | ret = mw->mw_status; |
1267 | /* Re-arm the completion in case we want to wait on it again */ |
1268 | INIT_COMPLETION(mw->mw_complete); |
1269 | return ret; |
1270 | } |
1271 | |
1272 | static int __ocfs2_cluster_lock(struct ocfs2_super *osb, |
1273 | struct ocfs2_lock_res *lockres, |
1274 | int level, |
1275 | u32 lkm_flags, |
1276 | int arg_flags, |
1277 | int l_subclass, |
1278 | unsigned long caller_ip) |
1279 | { |
1280 | struct ocfs2_mask_waiter mw; |
1281 | int wait, catch_signals = !(osb->s_mount_opt & OCFS2_MOUNT_NOINTR); |
1282 | int ret = 0; /* gcc doesn't realize wait = 1 guarantees ret is set */ |
1283 | unsigned long flags; |
1284 | unsigned int gen; |
1285 | int noqueue_attempted = 0; |
1286 | |
1287 | mlog_entry_void(); |
1288 | |
1289 | ocfs2_init_mask_waiter(&mw); |
1290 | |
1291 | if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) |
1292 | lkm_flags |= DLM_LKF_VALBLK; |
1293 | |
1294 | again: |
1295 | wait = 0; |
1296 | |
1297 | if (catch_signals && signal_pending(current)) { |
1298 | ret = -ERESTARTSYS; |
1299 | goto out; |
1300 | } |
1301 | |
1302 | spin_lock_irqsave(&lockres->l_lock, flags); |
1303 | |
1304 | mlog_bug_on_msg(lockres->l_flags & OCFS2_LOCK_FREEING, |
1305 | "Cluster lock called on freeing lockres %s! flags " |
1306 | "0x%lx\n", lockres->l_name, lockres->l_flags); |
1307 | |
1308 | /* We only compare against the currently granted level |
1309 | * here. If the lock is blocked waiting on a downconvert, |
1310 | * we'll get caught below. */ |
1311 | if (lockres->l_flags & OCFS2_LOCK_BUSY && |
1312 | level > lockres->l_level) { |
1313 | /* is someone sitting in dlm_lock? If so, wait on |
1314 | * them. */ |
1315 | lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0); |
1316 | wait = 1; |
1317 | goto unlock; |
1318 | } |
1319 | |
1320 | if (lockres->l_flags & OCFS2_LOCK_BLOCKED && |
1321 | !ocfs2_may_continue_on_blocked_lock(lockres, level)) { |
1322 | /* is the lock is currently blocked on behalf of |
1323 | * another node */ |
1324 | lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BLOCKED, 0); |
1325 | wait = 1; |
1326 | goto unlock; |
1327 | } |
1328 | |
1329 | if (level > lockres->l_level) { |
1330 | if (noqueue_attempted > 0) { |
1331 | ret = -EAGAIN; |
1332 | goto unlock; |
1333 | } |
1334 | if (lkm_flags & DLM_LKF_NOQUEUE) |
1335 | noqueue_attempted = 1; |
1336 | |
1337 | if (lockres->l_action != OCFS2_AST_INVALID) |
1338 | mlog(ML_ERROR, "lockres %s has action %u pending\n", |
1339 | lockres->l_name, lockres->l_action); |
1340 | |
1341 | if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) { |
1342 | lockres->l_action = OCFS2_AST_ATTACH; |
1343 | lkm_flags &= ~DLM_LKF_CONVERT; |
1344 | } else { |
1345 | lockres->l_action = OCFS2_AST_CONVERT; |
1346 | lkm_flags |= DLM_LKF_CONVERT; |
1347 | } |
1348 | |
1349 | lockres->l_requested = level; |
1350 | lockres_or_flags(lockres, OCFS2_LOCK_BUSY); |
1351 | gen = lockres_set_pending(lockres); |
1352 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
1353 | |
1354 | BUG_ON(level == DLM_LOCK_IV); |
1355 | BUG_ON(level == DLM_LOCK_NL); |
1356 | |
1357 | mlog(0, "lock %s, convert from %d to level = %d\n", |
1358 | lockres->l_name, lockres->l_level, level); |
1359 | |
1360 | /* call dlm_lock to upgrade lock now */ |
1361 | ret = ocfs2_dlm_lock(osb->cconn, |
1362 | level, |
1363 | &lockres->l_lksb, |
1364 | lkm_flags, |
1365 | lockres->l_name, |
1366 | OCFS2_LOCK_ID_MAX_LEN - 1, |
1367 | lockres); |
1368 | lockres_clear_pending(lockres, gen, osb); |
1369 | if (ret) { |
1370 | if (!(lkm_flags & DLM_LKF_NOQUEUE) || |
1371 | (ret != -EAGAIN)) { |
1372 | ocfs2_log_dlm_error("ocfs2_dlm_lock", |
1373 | ret, lockres); |
1374 | } |
1375 | ocfs2_recover_from_dlm_error(lockres, 1); |
1376 | goto out; |
1377 | } |
1378 | |
1379 | mlog(0, "lock %s, successful return from ocfs2_dlm_lock\n", |
1380 | lockres->l_name); |
1381 | |
1382 | /* At this point we've gone inside the dlm and need to |
1383 | * complete our work regardless. */ |
1384 | catch_signals = 0; |
1385 | |
1386 | /* wait for busy to clear and carry on */ |
1387 | goto again; |
1388 | } |
1389 | |
1390 | /* Ok, if we get here then we're good to go. */ |
1391 | ocfs2_inc_holders(lockres, level); |
1392 | |
1393 | ret = 0; |
1394 | unlock: |
1395 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
1396 | out: |
1397 | /* |
1398 | * This is helping work around a lock inversion between the page lock |
1399 | * and dlm locks. One path holds the page lock while calling aops |
1400 | * which block acquiring dlm locks. The voting thread holds dlm |
1401 | * locks while acquiring page locks while down converting data locks. |
1402 | * This block is helping an aop path notice the inversion and back |
1403 | * off to unlock its page lock before trying the dlm lock again. |
1404 | */ |
1405 | if (wait && arg_flags & OCFS2_LOCK_NONBLOCK && |
1406 | mw.mw_mask & (OCFS2_LOCK_BUSY|OCFS2_LOCK_BLOCKED)) { |
1407 | wait = 0; |
1408 | if (lockres_remove_mask_waiter(lockres, &mw)) |
1409 | ret = -EAGAIN; |
1410 | else |
1411 | goto again; |
1412 | } |
1413 | if (wait) { |
1414 | ret = ocfs2_wait_for_mask(&mw); |
1415 | if (ret == 0) |
1416 | goto again; |
1417 | mlog_errno(ret); |
1418 | } |
1419 | ocfs2_update_lock_stats(lockres, level, &mw, ret); |
1420 | |
1421 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
1422 | if (!ret && lockres->l_lockdep_map.key != NULL) { |
1423 | if (level == DLM_LOCK_PR) |
1424 | rwsem_acquire_read(&lockres->l_lockdep_map, l_subclass, |
1425 | !!(arg_flags & OCFS2_META_LOCK_NOQUEUE), |
1426 | caller_ip); |
1427 | else |
1428 | rwsem_acquire(&lockres->l_lockdep_map, l_subclass, |
1429 | !!(arg_flags & OCFS2_META_LOCK_NOQUEUE), |
1430 | caller_ip); |
1431 | } |
1432 | #endif |
1433 | mlog_exit(ret); |
1434 | return ret; |
1435 | } |
1436 | |
1437 | static inline int ocfs2_cluster_lock(struct ocfs2_super *osb, |
1438 | struct ocfs2_lock_res *lockres, |
1439 | int level, |
1440 | u32 lkm_flags, |
1441 | int arg_flags) |
1442 | { |
1443 | return __ocfs2_cluster_lock(osb, lockres, level, lkm_flags, arg_flags, |
1444 | 0, _RET_IP_); |
1445 | } |
1446 | |
1447 | |
1448 | static void __ocfs2_cluster_unlock(struct ocfs2_super *osb, |
1449 | struct ocfs2_lock_res *lockres, |
1450 | int level, |
1451 | unsigned long caller_ip) |
1452 | { |
1453 | unsigned long flags; |
1454 | |
1455 | mlog_entry_void(); |
1456 | spin_lock_irqsave(&lockres->l_lock, flags); |
1457 | ocfs2_dec_holders(lockres, level); |
1458 | ocfs2_downconvert_on_unlock(osb, lockres); |
1459 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
1460 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
1461 | if (lockres->l_lockdep_map.key != NULL) |
1462 | rwsem_release(&lockres->l_lockdep_map, 1, caller_ip); |
1463 | #endif |
1464 | mlog_exit_void(); |
1465 | } |
1466 | |
1467 | static int ocfs2_create_new_lock(struct ocfs2_super *osb, |
1468 | struct ocfs2_lock_res *lockres, |
1469 | int ex, |
1470 | int local) |
1471 | { |
1472 | int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR; |
1473 | unsigned long flags; |
1474 | u32 lkm_flags = local ? DLM_LKF_LOCAL : 0; |
1475 | |
1476 | spin_lock_irqsave(&lockres->l_lock, flags); |
1477 | BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED); |
1478 | lockres_or_flags(lockres, OCFS2_LOCK_LOCAL); |
1479 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
1480 | |
1481 | return ocfs2_lock_create(osb, lockres, level, lkm_flags); |
1482 | } |
1483 | |
1484 | /* Grants us an EX lock on the data and metadata resources, skipping |
1485 | * the normal cluster directory lookup. Use this ONLY on newly created |
1486 | * inodes which other nodes can't possibly see, and which haven't been |
1487 | * hashed in the inode hash yet. This can give us a good performance |
1488 | * increase as it'll skip the network broadcast normally associated |
1489 | * with creating a new lock resource. */ |
1490 | int ocfs2_create_new_inode_locks(struct inode *inode) |
1491 | { |
1492 | int ret; |
1493 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
1494 | |
1495 | BUG_ON(!inode); |
1496 | BUG_ON(!ocfs2_inode_is_new(inode)); |
1497 | |
1498 | mlog_entry_void(); |
1499 | |
1500 | mlog(0, "Inode %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno); |
1501 | |
1502 | /* NOTE: That we don't increment any of the holder counts, nor |
1503 | * do we add anything to a journal handle. Since this is |
1504 | * supposed to be a new inode which the cluster doesn't know |
1505 | * about yet, there is no need to. As far as the LVB handling |
1506 | * is concerned, this is basically like acquiring an EX lock |
1507 | * on a resource which has an invalid one -- we'll set it |
1508 | * valid when we release the EX. */ |
1509 | |
1510 | ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_rw_lockres, 1, 1); |
1511 | if (ret) { |
1512 | mlog_errno(ret); |
1513 | goto bail; |
1514 | } |
1515 | |
1516 | /* |
1517 | * We don't want to use DLM_LKF_LOCAL on a meta data lock as they |
1518 | * don't use a generation in their lock names. |
1519 | */ |
1520 | ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_inode_lockres, 1, 0); |
1521 | if (ret) { |
1522 | mlog_errno(ret); |
1523 | goto bail; |
1524 | } |
1525 | |
1526 | ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_open_lockres, 0, 0); |
1527 | if (ret) { |
1528 | mlog_errno(ret); |
1529 | goto bail; |
1530 | } |
1531 | |
1532 | bail: |
1533 | mlog_exit(ret); |
1534 | return ret; |
1535 | } |
1536 | |
1537 | int ocfs2_rw_lock(struct inode *inode, int write) |
1538 | { |
1539 | int status, level; |
1540 | struct ocfs2_lock_res *lockres; |
1541 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
1542 | |
1543 | BUG_ON(!inode); |
1544 | |
1545 | mlog_entry_void(); |
1546 | |
1547 | mlog(0, "inode %llu take %s RW lock\n", |
1548 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
1549 | write ? "EXMODE" : "PRMODE"); |
1550 | |
1551 | if (ocfs2_mount_local(osb)) |
1552 | return 0; |
1553 | |
1554 | lockres = &OCFS2_I(inode)->ip_rw_lockres; |
1555 | |
1556 | level = write ? DLM_LOCK_EX : DLM_LOCK_PR; |
1557 | |
1558 | status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres, level, 0, |
1559 | 0); |
1560 | if (status < 0) |
1561 | mlog_errno(status); |
1562 | |
1563 | mlog_exit(status); |
1564 | return status; |
1565 | } |
1566 | |
1567 | void ocfs2_rw_unlock(struct inode *inode, int write) |
1568 | { |
1569 | int level = write ? DLM_LOCK_EX : DLM_LOCK_PR; |
1570 | struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_rw_lockres; |
1571 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
1572 | |
1573 | mlog_entry_void(); |
1574 | |
1575 | mlog(0, "inode %llu drop %s RW lock\n", |
1576 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
1577 | write ? "EXMODE" : "PRMODE"); |
1578 | |
1579 | if (!ocfs2_mount_local(osb)) |
1580 | ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level); |
1581 | |
1582 | mlog_exit_void(); |
1583 | } |
1584 | |
1585 | /* |
1586 | * ocfs2_open_lock always get PR mode lock. |
1587 | */ |
1588 | int ocfs2_open_lock(struct inode *inode) |
1589 | { |
1590 | int status = 0; |
1591 | struct ocfs2_lock_res *lockres; |
1592 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
1593 | |
1594 | BUG_ON(!inode); |
1595 | |
1596 | mlog_entry_void(); |
1597 | |
1598 | mlog(0, "inode %llu take PRMODE open lock\n", |
1599 | (unsigned long long)OCFS2_I(inode)->ip_blkno); |
1600 | |
1601 | if (ocfs2_mount_local(osb)) |
1602 | goto out; |
1603 | |
1604 | lockres = &OCFS2_I(inode)->ip_open_lockres; |
1605 | |
1606 | status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres, |
1607 | DLM_LOCK_PR, 0, 0); |
1608 | if (status < 0) |
1609 | mlog_errno(status); |
1610 | |
1611 | out: |
1612 | mlog_exit(status); |
1613 | return status; |
1614 | } |
1615 | |
1616 | int ocfs2_try_open_lock(struct inode *inode, int write) |
1617 | { |
1618 | int status = 0, level; |
1619 | struct ocfs2_lock_res *lockres; |
1620 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
1621 | |
1622 | BUG_ON(!inode); |
1623 | |
1624 | mlog_entry_void(); |
1625 | |
1626 | mlog(0, "inode %llu try to take %s open lock\n", |
1627 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
1628 | write ? "EXMODE" : "PRMODE"); |
1629 | |
1630 | if (ocfs2_mount_local(osb)) |
1631 | goto out; |
1632 | |
1633 | lockres = &OCFS2_I(inode)->ip_open_lockres; |
1634 | |
1635 | level = write ? DLM_LOCK_EX : DLM_LOCK_PR; |
1636 | |
1637 | /* |
1638 | * The file system may already holding a PRMODE/EXMODE open lock. |
1639 | * Since we pass DLM_LKF_NOQUEUE, the request won't block waiting on |
1640 | * other nodes and the -EAGAIN will indicate to the caller that |
1641 | * this inode is still in use. |
1642 | */ |
1643 | status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres, |
1644 | level, DLM_LKF_NOQUEUE, 0); |
1645 | |
1646 | out: |
1647 | mlog_exit(status); |
1648 | return status; |
1649 | } |
1650 | |
1651 | /* |
1652 | * ocfs2_open_unlock unlock PR and EX mode open locks. |
1653 | */ |
1654 | void ocfs2_open_unlock(struct inode *inode) |
1655 | { |
1656 | struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_open_lockres; |
1657 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
1658 | |
1659 | mlog_entry_void(); |
1660 | |
1661 | mlog(0, "inode %llu drop open lock\n", |
1662 | (unsigned long long)OCFS2_I(inode)->ip_blkno); |
1663 | |
1664 | if (ocfs2_mount_local(osb)) |
1665 | goto out; |
1666 | |
1667 | if(lockres->l_ro_holders) |
1668 | ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, |
1669 | DLM_LOCK_PR); |
1670 | if(lockres->l_ex_holders) |
1671 | ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, |
1672 | DLM_LOCK_EX); |
1673 | |
1674 | out: |
1675 | mlog_exit_void(); |
1676 | } |
1677 | |
1678 | static int ocfs2_flock_handle_signal(struct ocfs2_lock_res *lockres, |
1679 | int level) |
1680 | { |
1681 | int ret; |
1682 | struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres); |
1683 | unsigned long flags; |
1684 | struct ocfs2_mask_waiter mw; |
1685 | |
1686 | ocfs2_init_mask_waiter(&mw); |
1687 | |
1688 | retry_cancel: |
1689 | spin_lock_irqsave(&lockres->l_lock, flags); |
1690 | if (lockres->l_flags & OCFS2_LOCK_BUSY) { |
1691 | ret = ocfs2_prepare_cancel_convert(osb, lockres); |
1692 | if (ret) { |
1693 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
1694 | ret = ocfs2_cancel_convert(osb, lockres); |
1695 | if (ret < 0) { |
1696 | mlog_errno(ret); |
1697 | goto out; |
1698 | } |
1699 | goto retry_cancel; |
1700 | } |
1701 | lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0); |
1702 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
1703 | |
1704 | ocfs2_wait_for_mask(&mw); |
1705 | goto retry_cancel; |
1706 | } |
1707 | |
1708 | ret = -ERESTARTSYS; |
1709 | /* |
1710 | * We may still have gotten the lock, in which case there's no |
1711 | * point to restarting the syscall. |
1712 | */ |
1713 | if (lockres->l_level == level) |
1714 | ret = 0; |
1715 | |
1716 | mlog(0, "Cancel returning %d. flags: 0x%lx, level: %d, act: %d\n", ret, |
1717 | lockres->l_flags, lockres->l_level, lockres->l_action); |
1718 | |
1719 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
1720 | |
1721 | out: |
1722 | return ret; |
1723 | } |
1724 | |
1725 | /* |
1726 | * ocfs2_file_lock() and ocfs2_file_unlock() map to a single pair of |
1727 | * flock() calls. The locking approach this requires is sufficiently |
1728 | * different from all other cluster lock types that we implement a |
1729 | * seperate path to the "low-level" dlm calls. In particular: |
1730 | * |
1731 | * - No optimization of lock levels is done - we take at exactly |
1732 | * what's been requested. |
1733 | * |
1734 | * - No lock caching is employed. We immediately downconvert to |
1735 | * no-lock at unlock time. This also means flock locks never go on |
1736 | * the blocking list). |
1737 | * |
1738 | * - Since userspace can trivially deadlock itself with flock, we make |
1739 | * sure to allow cancellation of a misbehaving applications flock() |
1740 | * request. |
1741 | * |
1742 | * - Access to any flock lockres doesn't require concurrency, so we |
1743 | * can simplify the code by requiring the caller to guarantee |
1744 | * serialization of dlmglue flock calls. |
1745 | */ |
1746 | int ocfs2_file_lock(struct file *file, int ex, int trylock) |
1747 | { |
1748 | int ret, level = ex ? DLM_LOCK_EX : DLM_LOCK_PR; |
1749 | unsigned int lkm_flags = trylock ? DLM_LKF_NOQUEUE : 0; |
1750 | unsigned long flags; |
1751 | struct ocfs2_file_private *fp = file->private_data; |
1752 | struct ocfs2_lock_res *lockres = &fp->fp_flock; |
1753 | struct ocfs2_super *osb = OCFS2_SB(file->f_mapping->host->i_sb); |
1754 | struct ocfs2_mask_waiter mw; |
1755 | |
1756 | ocfs2_init_mask_waiter(&mw); |
1757 | |
1758 | if ((lockres->l_flags & OCFS2_LOCK_BUSY) || |
1759 | (lockres->l_level > DLM_LOCK_NL)) { |
1760 | mlog(ML_ERROR, |
1761 | "File lock \"%s\" has busy or locked state: flags: 0x%lx, " |
1762 | "level: %u\n", lockres->l_name, lockres->l_flags, |
1763 | lockres->l_level); |
1764 | return -EINVAL; |
1765 | } |
1766 | |
1767 | spin_lock_irqsave(&lockres->l_lock, flags); |
1768 | if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) { |
1769 | lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0); |
1770 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
1771 | |
1772 | /* |
1773 | * Get the lock at NLMODE to start - that way we |
1774 | * can cancel the upconvert request if need be. |
1775 | */ |
1776 | ret = ocfs2_lock_create(osb, lockres, DLM_LOCK_NL, 0); |
1777 | if (ret < 0) { |
1778 | mlog_errno(ret); |
1779 | goto out; |
1780 | } |
1781 | |
1782 | ret = ocfs2_wait_for_mask(&mw); |
1783 | if (ret) { |
1784 | mlog_errno(ret); |
1785 | goto out; |
1786 | } |
1787 | spin_lock_irqsave(&lockres->l_lock, flags); |
1788 | } |
1789 | |
1790 | lockres->l_action = OCFS2_AST_CONVERT; |
1791 | lkm_flags |= DLM_LKF_CONVERT; |
1792 | lockres->l_requested = level; |
1793 | lockres_or_flags(lockres, OCFS2_LOCK_BUSY); |
1794 | |
1795 | lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0); |
1796 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
1797 | |
1798 | ret = ocfs2_dlm_lock(osb->cconn, level, &lockres->l_lksb, lkm_flags, |
1799 | lockres->l_name, OCFS2_LOCK_ID_MAX_LEN - 1, |
1800 | lockres); |
1801 | if (ret) { |
1802 | if (!trylock || (ret != -EAGAIN)) { |
1803 | ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres); |
1804 | ret = -EINVAL; |
1805 | } |
1806 | |
1807 | ocfs2_recover_from_dlm_error(lockres, 1); |
1808 | lockres_remove_mask_waiter(lockres, &mw); |
1809 | goto out; |
1810 | } |
1811 | |
1812 | ret = ocfs2_wait_for_mask_interruptible(&mw, lockres); |
1813 | if (ret == -ERESTARTSYS) { |
1814 | /* |
1815 | * Userspace can cause deadlock itself with |
1816 | * flock(). Current behavior locally is to allow the |
1817 | * deadlock, but abort the system call if a signal is |
1818 | * received. We follow this example, otherwise a |
1819 | * poorly written program could sit in kernel until |
1820 | * reboot. |
1821 | * |
1822 | * Handling this is a bit more complicated for Ocfs2 |
1823 | * though. We can't exit this function with an |
1824 | * outstanding lock request, so a cancel convert is |
1825 | * required. We intentionally overwrite 'ret' - if the |
1826 | * cancel fails and the lock was granted, it's easier |
1827 | * to just bubble sucess back up to the user. |
1828 | */ |
1829 | ret = ocfs2_flock_handle_signal(lockres, level); |
1830 | } else if (!ret && (level > lockres->l_level)) { |
1831 | /* Trylock failed asynchronously */ |
1832 | BUG_ON(!trylock); |
1833 | ret = -EAGAIN; |
1834 | } |
1835 | |
1836 | out: |
1837 | |
1838 | mlog(0, "Lock: \"%s\" ex: %d, trylock: %d, returns: %d\n", |
1839 | lockres->l_name, ex, trylock, ret); |
1840 | return ret; |
1841 | } |
1842 | |
1843 | void ocfs2_file_unlock(struct file *file) |
1844 | { |
1845 | int ret; |
1846 | unsigned int gen; |
1847 | unsigned long flags; |
1848 | struct ocfs2_file_private *fp = file->private_data; |
1849 | struct ocfs2_lock_res *lockres = &fp->fp_flock; |
1850 | struct ocfs2_super *osb = OCFS2_SB(file->f_mapping->host->i_sb); |
1851 | struct ocfs2_mask_waiter mw; |
1852 | |
1853 | ocfs2_init_mask_waiter(&mw); |
1854 | |
1855 | if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) |
1856 | return; |
1857 | |
1858 | if (lockres->l_level == DLM_LOCK_NL) |
1859 | return; |
1860 | |
1861 | mlog(0, "Unlock: \"%s\" flags: 0x%lx, level: %d, act: %d\n", |
1862 | lockres->l_name, lockres->l_flags, lockres->l_level, |
1863 | lockres->l_action); |
1864 | |
1865 | spin_lock_irqsave(&lockres->l_lock, flags); |
1866 | /* |
1867 | * Fake a blocking ast for the downconvert code. |
1868 | */ |
1869 | lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED); |
1870 | lockres->l_blocking = DLM_LOCK_EX; |
1871 | |
1872 | gen = ocfs2_prepare_downconvert(lockres, DLM_LOCK_NL); |
1873 | lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0); |
1874 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
1875 | |
1876 | ret = ocfs2_downconvert_lock(osb, lockres, DLM_LOCK_NL, 0, gen); |
1877 | if (ret) { |
1878 | mlog_errno(ret); |
1879 | return; |
1880 | } |
1881 | |
1882 | ret = ocfs2_wait_for_mask(&mw); |
1883 | if (ret) |
1884 | mlog_errno(ret); |
1885 | } |
1886 | |
1887 | static void ocfs2_downconvert_on_unlock(struct ocfs2_super *osb, |
1888 | struct ocfs2_lock_res *lockres) |
1889 | { |
1890 | int kick = 0; |
1891 | |
1892 | mlog_entry_void(); |
1893 | |
1894 | /* If we know that another node is waiting on our lock, kick |
1895 | * the downconvert thread * pre-emptively when we reach a release |
1896 | * condition. */ |
1897 | if (lockres->l_flags & OCFS2_LOCK_BLOCKED) { |
1898 | switch(lockres->l_blocking) { |
1899 | case DLM_LOCK_EX: |
1900 | if (!lockres->l_ex_holders && !lockres->l_ro_holders) |
1901 | kick = 1; |
1902 | break; |
1903 | case DLM_LOCK_PR: |
1904 | if (!lockres->l_ex_holders) |
1905 | kick = 1; |
1906 | break; |
1907 | default: |
1908 | BUG(); |
1909 | } |
1910 | } |
1911 | |
1912 | if (kick) |
1913 | ocfs2_wake_downconvert_thread(osb); |
1914 | |
1915 | mlog_exit_void(); |
1916 | } |
1917 | |
1918 | #define OCFS2_SEC_BITS 34 |
1919 | #define OCFS2_SEC_SHIFT (64 - 34) |
1920 | #define OCFS2_NSEC_MASK ((1ULL << OCFS2_SEC_SHIFT) - 1) |
1921 | |
1922 | /* LVB only has room for 64 bits of time here so we pack it for |
1923 | * now. */ |
1924 | static u64 ocfs2_pack_timespec(struct timespec *spec) |
1925 | { |
1926 | u64 res; |
1927 | u64 sec = spec->tv_sec; |
1928 | u32 nsec = spec->tv_nsec; |
1929 | |
1930 | res = (sec << OCFS2_SEC_SHIFT) | (nsec & OCFS2_NSEC_MASK); |
1931 | |
1932 | return res; |
1933 | } |
1934 | |
1935 | /* Call this with the lockres locked. I am reasonably sure we don't |
1936 | * need ip_lock in this function as anyone who would be changing those |
1937 | * values is supposed to be blocked in ocfs2_inode_lock right now. */ |
1938 | static void __ocfs2_stuff_meta_lvb(struct inode *inode) |
1939 | { |
1940 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
1941 | struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres; |
1942 | struct ocfs2_meta_lvb *lvb; |
1943 | |
1944 | mlog_entry_void(); |
1945 | |
1946 | lvb = ocfs2_dlm_lvb(&lockres->l_lksb); |
1947 | |
1948 | /* |
1949 | * Invalidate the LVB of a deleted inode - this way other |
1950 | * nodes are forced to go to disk and discover the new inode |
1951 | * status. |
1952 | */ |
1953 | if (oi->ip_flags & OCFS2_INODE_DELETED) { |
1954 | lvb->lvb_version = 0; |
1955 | goto out; |
1956 | } |
1957 | |
1958 | lvb->lvb_version = OCFS2_LVB_VERSION; |
1959 | lvb->lvb_isize = cpu_to_be64(i_size_read(inode)); |
1960 | lvb->lvb_iclusters = cpu_to_be32(oi->ip_clusters); |
1961 | lvb->lvb_iuid = cpu_to_be32(inode->i_uid); |
1962 | lvb->lvb_igid = cpu_to_be32(inode->i_gid); |
1963 | lvb->lvb_imode = cpu_to_be16(inode->i_mode); |
1964 | lvb->lvb_inlink = cpu_to_be16(inode->i_nlink); |
1965 | lvb->lvb_iatime_packed = |
1966 | cpu_to_be64(ocfs2_pack_timespec(&inode->i_atime)); |
1967 | lvb->lvb_ictime_packed = |
1968 | cpu_to_be64(ocfs2_pack_timespec(&inode->i_ctime)); |
1969 | lvb->lvb_imtime_packed = |
1970 | cpu_to_be64(ocfs2_pack_timespec(&inode->i_mtime)); |
1971 | lvb->lvb_iattr = cpu_to_be32(oi->ip_attr); |
1972 | lvb->lvb_idynfeatures = cpu_to_be16(oi->ip_dyn_features); |
1973 | lvb->lvb_igeneration = cpu_to_be32(inode->i_generation); |
1974 | |
1975 | out: |
1976 | mlog_meta_lvb(0, lockres); |
1977 | |
1978 | mlog_exit_void(); |
1979 | } |
1980 | |
1981 | static void ocfs2_unpack_timespec(struct timespec *spec, |
1982 | u64 packed_time) |
1983 | { |
1984 | spec->tv_sec = packed_time >> OCFS2_SEC_SHIFT; |
1985 | spec->tv_nsec = packed_time & OCFS2_NSEC_MASK; |
1986 | } |
1987 | |
1988 | static void ocfs2_refresh_inode_from_lvb(struct inode *inode) |
1989 | { |
1990 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
1991 | struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres; |
1992 | struct ocfs2_meta_lvb *lvb; |
1993 | |
1994 | mlog_entry_void(); |
1995 | |
1996 | mlog_meta_lvb(0, lockres); |
1997 | |
1998 | lvb = ocfs2_dlm_lvb(&lockres->l_lksb); |
1999 | |
2000 | /* We're safe here without the lockres lock... */ |
2001 | spin_lock(&oi->ip_lock); |
2002 | oi->ip_clusters = be32_to_cpu(lvb->lvb_iclusters); |
2003 | i_size_write(inode, be64_to_cpu(lvb->lvb_isize)); |
2004 | |
2005 | oi->ip_attr = be32_to_cpu(lvb->lvb_iattr); |
2006 | oi->ip_dyn_features = be16_to_cpu(lvb->lvb_idynfeatures); |
2007 | ocfs2_set_inode_flags(inode); |
2008 | |
2009 | /* fast-symlinks are a special case */ |
2010 | if (S_ISLNK(inode->i_mode) && !oi->ip_clusters) |
2011 | inode->i_blocks = 0; |
2012 | else |
2013 | inode->i_blocks = ocfs2_inode_sector_count(inode); |
2014 | |
2015 | inode->i_uid = be32_to_cpu(lvb->lvb_iuid); |
2016 | inode->i_gid = be32_to_cpu(lvb->lvb_igid); |
2017 | inode->i_mode = be16_to_cpu(lvb->lvb_imode); |
2018 | inode->i_nlink = be16_to_cpu(lvb->lvb_inlink); |
2019 | ocfs2_unpack_timespec(&inode->i_atime, |
2020 | be64_to_cpu(lvb->lvb_iatime_packed)); |
2021 | ocfs2_unpack_timespec(&inode->i_mtime, |
2022 | be64_to_cpu(lvb->lvb_imtime_packed)); |
2023 | ocfs2_unpack_timespec(&inode->i_ctime, |
2024 | be64_to_cpu(lvb->lvb_ictime_packed)); |
2025 | spin_unlock(&oi->ip_lock); |
2026 | |
2027 | mlog_exit_void(); |
2028 | } |
2029 | |
2030 | static inline int ocfs2_meta_lvb_is_trustable(struct inode *inode, |
2031 | struct ocfs2_lock_res *lockres) |
2032 | { |
2033 | struct ocfs2_meta_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb); |
2034 | |
2035 | if (ocfs2_dlm_lvb_valid(&lockres->l_lksb) |
2036 | && lvb->lvb_version == OCFS2_LVB_VERSION |
2037 | && be32_to_cpu(lvb->lvb_igeneration) == inode->i_generation) |
2038 | return 1; |
2039 | return 0; |
2040 | } |
2041 | |
2042 | /* Determine whether a lock resource needs to be refreshed, and |
2043 | * arbitrate who gets to refresh it. |
2044 | * |
2045 | * 0 means no refresh needed. |
2046 | * |
2047 | * > 0 means you need to refresh this and you MUST call |
2048 | * ocfs2_complete_lock_res_refresh afterwards. */ |
2049 | static int ocfs2_should_refresh_lock_res(struct ocfs2_lock_res *lockres) |
2050 | { |
2051 | unsigned long flags; |
2052 | int status = 0; |
2053 | |
2054 | mlog_entry_void(); |
2055 | |
2056 | refresh_check: |
2057 | spin_lock_irqsave(&lockres->l_lock, flags); |
2058 | if (!(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) { |
2059 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
2060 | goto bail; |
2061 | } |
2062 | |
2063 | if (lockres->l_flags & OCFS2_LOCK_REFRESHING) { |
2064 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
2065 | |
2066 | ocfs2_wait_on_refreshing_lock(lockres); |
2067 | goto refresh_check; |
2068 | } |
2069 | |
2070 | /* Ok, I'll be the one to refresh this lock. */ |
2071 | lockres_or_flags(lockres, OCFS2_LOCK_REFRESHING); |
2072 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
2073 | |
2074 | status = 1; |
2075 | bail: |
2076 | mlog_exit(status); |
2077 | return status; |
2078 | } |
2079 | |
2080 | /* If status is non zero, I'll mark it as not being in refresh |
2081 | * anymroe, but i won't clear the needs refresh flag. */ |
2082 | static inline void ocfs2_complete_lock_res_refresh(struct ocfs2_lock_res *lockres, |
2083 | int status) |
2084 | { |
2085 | unsigned long flags; |
2086 | mlog_entry_void(); |
2087 | |
2088 | spin_lock_irqsave(&lockres->l_lock, flags); |
2089 | lockres_clear_flags(lockres, OCFS2_LOCK_REFRESHING); |
2090 | if (!status) |
2091 | lockres_clear_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH); |
2092 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
2093 | |
2094 | wake_up(&lockres->l_event); |
2095 | |
2096 | mlog_exit_void(); |
2097 | } |
2098 | |
2099 | /* may or may not return a bh if it went to disk. */ |
2100 | static int ocfs2_inode_lock_update(struct inode *inode, |
2101 | struct buffer_head **bh) |
2102 | { |
2103 | int status = 0; |
2104 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
2105 | struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres; |
2106 | struct ocfs2_dinode *fe; |
2107 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
2108 | |
2109 | mlog_entry_void(); |
2110 | |
2111 | if (ocfs2_mount_local(osb)) |
2112 | goto bail; |
2113 | |
2114 | spin_lock(&oi->ip_lock); |
2115 | if (oi->ip_flags & OCFS2_INODE_DELETED) { |
2116 | mlog(0, "Orphaned inode %llu was deleted while we " |
2117 | "were waiting on a lock. ip_flags = 0x%x\n", |
2118 | (unsigned long long)oi->ip_blkno, oi->ip_flags); |
2119 | spin_unlock(&oi->ip_lock); |
2120 | status = -ENOENT; |
2121 | goto bail; |
2122 | } |
2123 | spin_unlock(&oi->ip_lock); |
2124 | |
2125 | if (!ocfs2_should_refresh_lock_res(lockres)) |
2126 | goto bail; |
2127 | |
2128 | /* This will discard any caching information we might have had |
2129 | * for the inode metadata. */ |
2130 | ocfs2_metadata_cache_purge(inode); |
2131 | |
2132 | ocfs2_extent_map_trunc(inode, 0); |
2133 | |
2134 | if (ocfs2_meta_lvb_is_trustable(inode, lockres)) { |
2135 | mlog(0, "Trusting LVB on inode %llu\n", |
2136 | (unsigned long long)oi->ip_blkno); |
2137 | ocfs2_refresh_inode_from_lvb(inode); |
2138 | } else { |
2139 | /* Boo, we have to go to disk. */ |
2140 | /* read bh, cast, ocfs2_refresh_inode */ |
2141 | status = ocfs2_read_inode_block(inode, bh); |
2142 | if (status < 0) { |
2143 | mlog_errno(status); |
2144 | goto bail_refresh; |
2145 | } |
2146 | fe = (struct ocfs2_dinode *) (*bh)->b_data; |
2147 | |
2148 | /* This is a good chance to make sure we're not |
2149 | * locking an invalid object. ocfs2_read_inode_block() |
2150 | * already checked that the inode block is sane. |
2151 | * |
2152 | * We bug on a stale inode here because we checked |
2153 | * above whether it was wiped from disk. The wiping |
2154 | * node provides a guarantee that we receive that |
2155 | * message and can mark the inode before dropping any |
2156 | * locks associated with it. */ |
2157 | mlog_bug_on_msg(inode->i_generation != |
2158 | le32_to_cpu(fe->i_generation), |
2159 | "Invalid dinode %llu disk generation: %u " |
2160 | "inode->i_generation: %u\n", |
2161 | (unsigned long long)oi->ip_blkno, |
2162 | le32_to_cpu(fe->i_generation), |
2163 | inode->i_generation); |
2164 | mlog_bug_on_msg(le64_to_cpu(fe->i_dtime) || |
2165 | !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)), |
2166 | "Stale dinode %llu dtime: %llu flags: 0x%x\n", |
2167 | (unsigned long long)oi->ip_blkno, |
2168 | (unsigned long long)le64_to_cpu(fe->i_dtime), |
2169 | le32_to_cpu(fe->i_flags)); |
2170 | |
2171 | ocfs2_refresh_inode(inode, fe); |
2172 | ocfs2_track_lock_refresh(lockres); |
2173 | } |
2174 | |
2175 | status = 0; |
2176 | bail_refresh: |
2177 | ocfs2_complete_lock_res_refresh(lockres, status); |
2178 | bail: |
2179 | mlog_exit(status); |
2180 | return status; |
2181 | } |
2182 | |
2183 | static int ocfs2_assign_bh(struct inode *inode, |
2184 | struct buffer_head **ret_bh, |
2185 | struct buffer_head *passed_bh) |
2186 | { |
2187 | int status; |
2188 | |
2189 | if (passed_bh) { |
2190 | /* Ok, the update went to disk for us, use the |
2191 | * returned bh. */ |
2192 | *ret_bh = passed_bh; |
2193 | get_bh(*ret_bh); |
2194 | |
2195 | return 0; |
2196 | } |
2197 | |
2198 | status = ocfs2_read_inode_block(inode, ret_bh); |
2199 | if (status < 0) |
2200 | mlog_errno(status); |
2201 | |
2202 | return status; |
2203 | } |
2204 | |
2205 | /* |
2206 | * returns < 0 error if the callback will never be called, otherwise |
2207 | * the result of the lock will be communicated via the callback. |
2208 | */ |
2209 | int ocfs2_inode_lock_full_nested(struct inode *inode, |
2210 | struct buffer_head **ret_bh, |
2211 | int ex, |
2212 | int arg_flags, |
2213 | int subclass) |
2214 | { |
2215 | int status, level, acquired; |
2216 | u32 dlm_flags; |
2217 | struct ocfs2_lock_res *lockres = NULL; |
2218 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
2219 | struct buffer_head *local_bh = NULL; |
2220 | |
2221 | BUG_ON(!inode); |
2222 | |
2223 | mlog_entry_void(); |
2224 | |
2225 | mlog(0, "inode %llu, take %s META lock\n", |
2226 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
2227 | ex ? "EXMODE" : "PRMODE"); |
2228 | |
2229 | status = 0; |
2230 | acquired = 0; |
2231 | /* We'll allow faking a readonly metadata lock for |
2232 | * rodevices. */ |
2233 | if (ocfs2_is_hard_readonly(osb)) { |
2234 | if (ex) |
2235 | status = -EROFS; |
2236 | goto bail; |
2237 | } |
2238 | |
2239 | if (ocfs2_mount_local(osb)) |
2240 | goto local; |
2241 | |
2242 | if (!(arg_flags & OCFS2_META_LOCK_RECOVERY)) |
2243 | ocfs2_wait_for_recovery(osb); |
2244 | |
2245 | lockres = &OCFS2_I(inode)->ip_inode_lockres; |
2246 | level = ex ? DLM_LOCK_EX : DLM_LOCK_PR; |
2247 | dlm_flags = 0; |
2248 | if (arg_flags & OCFS2_META_LOCK_NOQUEUE) |
2249 | dlm_flags |= DLM_LKF_NOQUEUE; |
2250 | |
2251 | status = __ocfs2_cluster_lock(osb, lockres, level, dlm_flags, |
2252 | arg_flags, subclass, _RET_IP_); |
2253 | if (status < 0) { |
2254 | if (status != -EAGAIN && status != -EIOCBRETRY) |
2255 | mlog_errno(status); |
2256 | goto bail; |
2257 | } |
2258 | |
2259 | /* Notify the error cleanup path to drop the cluster lock. */ |
2260 | acquired = 1; |
2261 | |
2262 | /* We wait twice because a node may have died while we were in |
2263 | * the lower dlm layers. The second time though, we've |
2264 | * committed to owning this lock so we don't allow signals to |
2265 | * abort the operation. */ |
2266 | if (!(arg_flags & OCFS2_META_LOCK_RECOVERY)) |
2267 | ocfs2_wait_for_recovery(osb); |
2268 | |
2269 | local: |
2270 | /* |
2271 | * We only see this flag if we're being called from |
2272 | * ocfs2_read_locked_inode(). It means we're locking an inode |
2273 | * which hasn't been populated yet, so clear the refresh flag |
2274 | * and let the caller handle it. |
2275 | */ |
2276 | if (inode->i_state & I_NEW) { |
2277 | status = 0; |
2278 | if (lockres) |
2279 | ocfs2_complete_lock_res_refresh(lockres, 0); |
2280 | goto bail; |
2281 | } |
2282 | |
2283 | /* This is fun. The caller may want a bh back, or it may |
2284 | * not. ocfs2_inode_lock_update definitely wants one in, but |
2285 | * may or may not read one, depending on what's in the |
2286 | * LVB. The result of all of this is that we've *only* gone to |
2287 | * disk if we have to, so the complexity is worthwhile. */ |
2288 | status = ocfs2_inode_lock_update(inode, &local_bh); |
2289 | if (status < 0) { |
2290 | if (status != -ENOENT) |
2291 | mlog_errno(status); |
2292 | goto bail; |
2293 | } |
2294 | |
2295 | if (ret_bh) { |
2296 | status = ocfs2_assign_bh(inode, ret_bh, local_bh); |
2297 | if (status < 0) { |
2298 | mlog_errno(status); |
2299 | goto bail; |
2300 | } |
2301 | } |
2302 | |
2303 | bail: |
2304 | if (status < 0) { |
2305 | if (ret_bh && (*ret_bh)) { |
2306 | brelse(*ret_bh); |
2307 | *ret_bh = NULL; |
2308 | } |
2309 | if (acquired) |
2310 | ocfs2_inode_unlock(inode, ex); |
2311 | } |
2312 | |
2313 | if (local_bh) |
2314 | brelse(local_bh); |
2315 | |
2316 | mlog_exit(status); |
2317 | return status; |
2318 | } |
2319 | |
2320 | /* |
2321 | * This is working around a lock inversion between tasks acquiring DLM |
2322 | * locks while holding a page lock and the downconvert thread which |
2323 | * blocks dlm lock acquiry while acquiring page locks. |
2324 | * |
2325 | * ** These _with_page variantes are only intended to be called from aop |
2326 | * methods that hold page locks and return a very specific *positive* error |
2327 | * code that aop methods pass up to the VFS -- test for errors with != 0. ** |
2328 | * |
2329 | * The DLM is called such that it returns -EAGAIN if it would have |
2330 | * blocked waiting for the downconvert thread. In that case we unlock |
2331 | * our page so the downconvert thread can make progress. Once we've |
2332 | * done this we have to return AOP_TRUNCATED_PAGE so the aop method |
2333 | * that called us can bubble that back up into the VFS who will then |
2334 | * immediately retry the aop call. |
2335 | * |
2336 | * We do a blocking lock and immediate unlock before returning, though, so that |
2337 | * the lock has a great chance of being cached on this node by the time the VFS |
2338 | * calls back to retry the aop. This has a potential to livelock as nodes |
2339 | * ping locks back and forth, but that's a risk we're willing to take to avoid |
2340 | * the lock inversion simply. |
2341 | */ |
2342 | int ocfs2_inode_lock_with_page(struct inode *inode, |
2343 | struct buffer_head **ret_bh, |
2344 | int ex, |
2345 | struct page *page) |
2346 | { |
2347 | int ret; |
2348 | |
2349 | ret = ocfs2_inode_lock_full(inode, ret_bh, ex, OCFS2_LOCK_NONBLOCK); |
2350 | if (ret == -EAGAIN) { |
2351 | unlock_page(page); |
2352 | if (ocfs2_inode_lock(inode, ret_bh, ex) == 0) |
2353 | ocfs2_inode_unlock(inode, ex); |
2354 | ret = AOP_TRUNCATED_PAGE; |
2355 | } |
2356 | |
2357 | return ret; |
2358 | } |
2359 | |
2360 | int ocfs2_inode_lock_atime(struct inode *inode, |
2361 | struct vfsmount *vfsmnt, |
2362 | int *level) |
2363 | { |
2364 | int ret; |
2365 | |
2366 | mlog_entry_void(); |
2367 | ret = ocfs2_inode_lock(inode, NULL, 0); |
2368 | if (ret < 0) { |
2369 | mlog_errno(ret); |
2370 | return ret; |
2371 | } |
2372 | |
2373 | /* |
2374 | * If we should update atime, we will get EX lock, |
2375 | * otherwise we just get PR lock. |
2376 | */ |
2377 | if (ocfs2_should_update_atime(inode, vfsmnt)) { |
2378 | struct buffer_head *bh = NULL; |
2379 | |
2380 | ocfs2_inode_unlock(inode, 0); |
2381 | ret = ocfs2_inode_lock(inode, &bh, 1); |
2382 | if (ret < 0) { |
2383 | mlog_errno(ret); |
2384 | return ret; |
2385 | } |
2386 | *level = 1; |
2387 | if (ocfs2_should_update_atime(inode, vfsmnt)) |
2388 | ocfs2_update_inode_atime(inode, bh); |
2389 | if (bh) |
2390 | brelse(bh); |
2391 | } else |
2392 | *level = 0; |
2393 | |
2394 | mlog_exit(ret); |
2395 | return ret; |
2396 | } |
2397 | |
2398 | void ocfs2_inode_unlock(struct inode *inode, |
2399 | int ex) |
2400 | { |
2401 | int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR; |
2402 | struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_inode_lockres; |
2403 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
2404 | |
2405 | mlog_entry_void(); |
2406 | |
2407 | mlog(0, "inode %llu drop %s META lock\n", |
2408 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
2409 | ex ? "EXMODE" : "PRMODE"); |
2410 | |
2411 | if (!ocfs2_is_hard_readonly(OCFS2_SB(inode->i_sb)) && |
2412 | !ocfs2_mount_local(osb)) |
2413 | ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level); |
2414 | |
2415 | mlog_exit_void(); |
2416 | } |
2417 | |
2418 | int ocfs2_orphan_scan_lock(struct ocfs2_super *osb, u32 *seqno) |
2419 | { |
2420 | struct ocfs2_lock_res *lockres; |
2421 | struct ocfs2_orphan_scan_lvb *lvb; |
2422 | int status = 0; |
2423 | |
2424 | if (ocfs2_is_hard_readonly(osb)) |
2425 | return -EROFS; |
2426 | |
2427 | if (ocfs2_mount_local(osb)) |
2428 | return 0; |
2429 | |
2430 | lockres = &osb->osb_orphan_scan.os_lockres; |
2431 | status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_EX, 0, 0); |
2432 | if (status < 0) |
2433 | return status; |
2434 | |
2435 | lvb = ocfs2_dlm_lvb(&lockres->l_lksb); |
2436 | if (ocfs2_dlm_lvb_valid(&lockres->l_lksb) && |
2437 | lvb->lvb_version == OCFS2_ORPHAN_LVB_VERSION) |
2438 | *seqno = be32_to_cpu(lvb->lvb_os_seqno); |
2439 | else |
2440 | *seqno = osb->osb_orphan_scan.os_seqno + 1; |
2441 | |
2442 | return status; |
2443 | } |
2444 | |
2445 | void ocfs2_orphan_scan_unlock(struct ocfs2_super *osb, u32 seqno) |
2446 | { |
2447 | struct ocfs2_lock_res *lockres; |
2448 | struct ocfs2_orphan_scan_lvb *lvb; |
2449 | |
2450 | if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb)) { |
2451 | lockres = &osb->osb_orphan_scan.os_lockres; |
2452 | lvb = ocfs2_dlm_lvb(&lockres->l_lksb); |
2453 | lvb->lvb_version = OCFS2_ORPHAN_LVB_VERSION; |
2454 | lvb->lvb_os_seqno = cpu_to_be32(seqno); |
2455 | ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX); |
2456 | } |
2457 | } |
2458 | |
2459 | int ocfs2_super_lock(struct ocfs2_super *osb, |
2460 | int ex) |
2461 | { |
2462 | int status = 0; |
2463 | int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR; |
2464 | struct ocfs2_lock_res *lockres = &osb->osb_super_lockres; |
2465 | |
2466 | mlog_entry_void(); |
2467 | |
2468 | if (ocfs2_is_hard_readonly(osb)) |
2469 | return -EROFS; |
2470 | |
2471 | if (ocfs2_mount_local(osb)) |
2472 | goto bail; |
2473 | |
2474 | status = ocfs2_cluster_lock(osb, lockres, level, 0, 0); |
2475 | if (status < 0) { |
2476 | mlog_errno(status); |
2477 | goto bail; |
2478 | } |
2479 | |
2480 | /* The super block lock path is really in the best position to |
2481 | * know when resources covered by the lock need to be |
2482 | * refreshed, so we do it here. Of course, making sense of |
2483 | * everything is up to the caller :) */ |
2484 | status = ocfs2_should_refresh_lock_res(lockres); |
2485 | if (status < 0) { |
2486 | mlog_errno(status); |
2487 | goto bail; |
2488 | } |
2489 | if (status) { |
2490 | status = ocfs2_refresh_slot_info(osb); |
2491 | |
2492 | ocfs2_complete_lock_res_refresh(lockres, status); |
2493 | |
2494 | if (status < 0) |
2495 | mlog_errno(status); |
2496 | ocfs2_track_lock_refresh(lockres); |
2497 | } |
2498 | bail: |
2499 | mlog_exit(status); |
2500 | return status; |
2501 | } |
2502 | |
2503 | void ocfs2_super_unlock(struct ocfs2_super *osb, |
2504 | int ex) |
2505 | { |
2506 | int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR; |
2507 | struct ocfs2_lock_res *lockres = &osb->osb_super_lockres; |
2508 | |
2509 | if (!ocfs2_mount_local(osb)) |
2510 | ocfs2_cluster_unlock(osb, lockres, level); |
2511 | } |
2512 | |
2513 | int ocfs2_rename_lock(struct ocfs2_super *osb) |
2514 | { |
2515 | int status; |
2516 | struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres; |
2517 | |
2518 | if (ocfs2_is_hard_readonly(osb)) |
2519 | return -EROFS; |
2520 | |
2521 | if (ocfs2_mount_local(osb)) |
2522 | return 0; |
2523 | |
2524 | status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_EX, 0, 0); |
2525 | if (status < 0) |
2526 | mlog_errno(status); |
2527 | |
2528 | return status; |
2529 | } |
2530 | |
2531 | void ocfs2_rename_unlock(struct ocfs2_super *osb) |
2532 | { |
2533 | struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres; |
2534 | |
2535 | if (!ocfs2_mount_local(osb)) |
2536 | ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX); |
2537 | } |
2538 | |
2539 | int ocfs2_nfs_sync_lock(struct ocfs2_super *osb, int ex) |
2540 | { |
2541 | int status; |
2542 | struct ocfs2_lock_res *lockres = &osb->osb_nfs_sync_lockres; |
2543 | |
2544 | if (ocfs2_is_hard_readonly(osb)) |
2545 | return -EROFS; |
2546 | |
2547 | if (ocfs2_mount_local(osb)) |
2548 | return 0; |
2549 | |
2550 | status = ocfs2_cluster_lock(osb, lockres, ex ? LKM_EXMODE : LKM_PRMODE, |
2551 | 0, 0); |
2552 | if (status < 0) |
2553 | mlog(ML_ERROR, "lock on nfs sync lock failed %d\n", status); |
2554 | |
2555 | return status; |
2556 | } |
2557 | |
2558 | void ocfs2_nfs_sync_unlock(struct ocfs2_super *osb, int ex) |
2559 | { |
2560 | struct ocfs2_lock_res *lockres = &osb->osb_nfs_sync_lockres; |
2561 | |
2562 | if (!ocfs2_mount_local(osb)) |
2563 | ocfs2_cluster_unlock(osb, lockres, |
2564 | ex ? LKM_EXMODE : LKM_PRMODE); |
2565 | } |
2566 | |
2567 | int ocfs2_dentry_lock(struct dentry *dentry, int ex) |
2568 | { |
2569 | int ret; |
2570 | int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR; |
2571 | struct ocfs2_dentry_lock *dl = dentry->d_fsdata; |
2572 | struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb); |
2573 | |
2574 | BUG_ON(!dl); |
2575 | |
2576 | if (ocfs2_is_hard_readonly(osb)) |
2577 | return -EROFS; |
2578 | |
2579 | if (ocfs2_mount_local(osb)) |
2580 | return 0; |
2581 | |
2582 | ret = ocfs2_cluster_lock(osb, &dl->dl_lockres, level, 0, 0); |
2583 | if (ret < 0) |
2584 | mlog_errno(ret); |
2585 | |
2586 | return ret; |
2587 | } |
2588 | |
2589 | void ocfs2_dentry_unlock(struct dentry *dentry, int ex) |
2590 | { |
2591 | int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR; |
2592 | struct ocfs2_dentry_lock *dl = dentry->d_fsdata; |
2593 | struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb); |
2594 | |
2595 | if (!ocfs2_mount_local(osb)) |
2596 | ocfs2_cluster_unlock(osb, &dl->dl_lockres, level); |
2597 | } |
2598 | |
2599 | /* Reference counting of the dlm debug structure. We want this because |
2600 | * open references on the debug inodes can live on after a mount, so |
2601 | * we can't rely on the ocfs2_super to always exist. */ |
2602 | static void ocfs2_dlm_debug_free(struct kref *kref) |
2603 | { |
2604 | struct ocfs2_dlm_debug *dlm_debug; |
2605 | |
2606 | dlm_debug = container_of(kref, struct ocfs2_dlm_debug, d_refcnt); |
2607 | |
2608 | kfree(dlm_debug); |
2609 | } |
2610 | |
2611 | void ocfs2_put_dlm_debug(struct ocfs2_dlm_debug *dlm_debug) |
2612 | { |
2613 | if (dlm_debug) |
2614 | kref_put(&dlm_debug->d_refcnt, ocfs2_dlm_debug_free); |
2615 | } |
2616 | |
2617 | static void ocfs2_get_dlm_debug(struct ocfs2_dlm_debug *debug) |
2618 | { |
2619 | kref_get(&debug->d_refcnt); |
2620 | } |
2621 | |
2622 | struct ocfs2_dlm_debug *ocfs2_new_dlm_debug(void) |
2623 | { |
2624 | struct ocfs2_dlm_debug *dlm_debug; |
2625 | |
2626 | dlm_debug = kmalloc(sizeof(struct ocfs2_dlm_debug), GFP_KERNEL); |
2627 | if (!dlm_debug) { |
2628 | mlog_errno(-ENOMEM); |
2629 | goto out; |
2630 | } |
2631 | |
2632 | kref_init(&dlm_debug->d_refcnt); |
2633 | INIT_LIST_HEAD(&dlm_debug->d_lockres_tracking); |
2634 | dlm_debug->d_locking_state = NULL; |
2635 | out: |
2636 | return dlm_debug; |
2637 | } |
2638 | |
2639 | /* Access to this is arbitrated for us via seq_file->sem. */ |
2640 | struct ocfs2_dlm_seq_priv { |
2641 | struct ocfs2_dlm_debug *p_dlm_debug; |
2642 | struct ocfs2_lock_res p_iter_res; |
2643 | struct ocfs2_lock_res p_tmp_res; |
2644 | }; |
2645 | |
2646 | static struct ocfs2_lock_res *ocfs2_dlm_next_res(struct ocfs2_lock_res *start, |
2647 | struct ocfs2_dlm_seq_priv *priv) |
2648 | { |
2649 | struct ocfs2_lock_res *iter, *ret = NULL; |
2650 | struct ocfs2_dlm_debug *dlm_debug = priv->p_dlm_debug; |
2651 | |
2652 | assert_spin_locked(&ocfs2_dlm_tracking_lock); |
2653 | |
2654 | list_for_each_entry(iter, &start->l_debug_list, l_debug_list) { |
2655 | /* discover the head of the list */ |
2656 | if (&iter->l_debug_list == &dlm_debug->d_lockres_tracking) { |
2657 | mlog(0, "End of list found, %p\n", ret); |
2658 | break; |
2659 | } |
2660 | |
2661 | /* We track our "dummy" iteration lockres' by a NULL |
2662 | * l_ops field. */ |
2663 | if (iter->l_ops != NULL) { |
2664 | ret = iter; |
2665 | break; |
2666 | } |
2667 | } |
2668 | |
2669 | return ret; |
2670 | } |
2671 | |
2672 | static void *ocfs2_dlm_seq_start(struct seq_file *m, loff_t *pos) |
2673 | { |
2674 | struct ocfs2_dlm_seq_priv *priv = m->private; |
2675 | struct ocfs2_lock_res *iter; |
2676 | |
2677 | spin_lock(&ocfs2_dlm_tracking_lock); |
2678 | iter = ocfs2_dlm_next_res(&priv->p_iter_res, priv); |
2679 | if (iter) { |
2680 | /* Since lockres' have the lifetime of their container |
2681 | * (which can be inodes, ocfs2_supers, etc) we want to |
2682 | * copy this out to a temporary lockres while still |
2683 | * under the spinlock. Obviously after this we can't |
2684 | * trust any pointers on the copy returned, but that's |
2685 | * ok as the information we want isn't typically held |
2686 | * in them. */ |
2687 | priv->p_tmp_res = *iter; |
2688 | iter = &priv->p_tmp_res; |
2689 | } |
2690 | spin_unlock(&ocfs2_dlm_tracking_lock); |
2691 | |
2692 | return iter; |
2693 | } |
2694 | |
2695 | static void ocfs2_dlm_seq_stop(struct seq_file *m, void *v) |
2696 | { |
2697 | } |
2698 | |
2699 | static void *ocfs2_dlm_seq_next(struct seq_file *m, void *v, loff_t *pos) |
2700 | { |
2701 | struct ocfs2_dlm_seq_priv *priv = m->private; |
2702 | struct ocfs2_lock_res *iter = v; |
2703 | struct ocfs2_lock_res *dummy = &priv->p_iter_res; |
2704 | |
2705 | spin_lock(&ocfs2_dlm_tracking_lock); |
2706 | iter = ocfs2_dlm_next_res(iter, priv); |
2707 | list_del_init(&dummy->l_debug_list); |
2708 | if (iter) { |
2709 | list_add(&dummy->l_debug_list, &iter->l_debug_list); |
2710 | priv->p_tmp_res = *iter; |
2711 | iter = &priv->p_tmp_res; |
2712 | } |
2713 | spin_unlock(&ocfs2_dlm_tracking_lock); |
2714 | |
2715 | return iter; |
2716 | } |
2717 | |
2718 | /* So that debugfs.ocfs2 can determine which format is being used */ |
2719 | #define OCFS2_DLM_DEBUG_STR_VERSION 2 |
2720 | static int ocfs2_dlm_seq_show(struct seq_file *m, void *v) |
2721 | { |
2722 | int i; |
2723 | char *lvb; |
2724 | struct ocfs2_lock_res *lockres = v; |
2725 | |
2726 | if (!lockres) |
2727 | return -EINVAL; |
2728 | |
2729 | seq_printf(m, "0x%x\t", OCFS2_DLM_DEBUG_STR_VERSION); |
2730 | |
2731 | if (lockres->l_type == OCFS2_LOCK_TYPE_DENTRY) |
2732 | seq_printf(m, "%.*s%08x\t", OCFS2_DENTRY_LOCK_INO_START - 1, |
2733 | lockres->l_name, |
2734 | (unsigned int)ocfs2_get_dentry_lock_ino(lockres)); |
2735 | else |
2736 | seq_printf(m, "%.*s\t", OCFS2_LOCK_ID_MAX_LEN, lockres->l_name); |
2737 | |
2738 | seq_printf(m, "%d\t" |
2739 | "0x%lx\t" |
2740 | "0x%x\t" |
2741 | "0x%x\t" |
2742 | "%u\t" |
2743 | "%u\t" |
2744 | "%d\t" |
2745 | "%d\t", |
2746 | lockres->l_level, |
2747 | lockres->l_flags, |
2748 | lockres->l_action, |
2749 | lockres->l_unlock_action, |
2750 | lockres->l_ro_holders, |
2751 | lockres->l_ex_holders, |
2752 | lockres->l_requested, |
2753 | lockres->l_blocking); |
2754 | |
2755 | /* Dump the raw LVB */ |
2756 | lvb = ocfs2_dlm_lvb(&lockres->l_lksb); |
2757 | for(i = 0; i < DLM_LVB_LEN; i++) |
2758 | seq_printf(m, "0x%x\t", lvb[i]); |
2759 | |
2760 | #ifdef CONFIG_OCFS2_FS_STATS |
2761 | # define lock_num_prmode(_l) (_l)->l_lock_num_prmode |
2762 | # define lock_num_exmode(_l) (_l)->l_lock_num_exmode |
2763 | # define lock_num_prmode_failed(_l) (_l)->l_lock_num_prmode_failed |
2764 | # define lock_num_exmode_failed(_l) (_l)->l_lock_num_exmode_failed |
2765 | # define lock_total_prmode(_l) (_l)->l_lock_total_prmode |
2766 | # define lock_total_exmode(_l) (_l)->l_lock_total_exmode |
2767 | # define lock_max_prmode(_l) (_l)->l_lock_max_prmode |
2768 | # define lock_max_exmode(_l) (_l)->l_lock_max_exmode |
2769 | # define lock_refresh(_l) (_l)->l_lock_refresh |
2770 | #else |
2771 | # define lock_num_prmode(_l) (0ULL) |
2772 | # define lock_num_exmode(_l) (0ULL) |
2773 | # define lock_num_prmode_failed(_l) (0) |
2774 | # define lock_num_exmode_failed(_l) (0) |
2775 | # define lock_total_prmode(_l) (0ULL) |
2776 | # define lock_total_exmode(_l) (0ULL) |
2777 | # define lock_max_prmode(_l) (0) |
2778 | # define lock_max_exmode(_l) (0) |
2779 | # define lock_refresh(_l) (0) |
2780 | #endif |
2781 | /* The following seq_print was added in version 2 of this output */ |
2782 | seq_printf(m, "%llu\t" |
2783 | "%llu\t" |
2784 | "%u\t" |
2785 | "%u\t" |
2786 | "%llu\t" |
2787 | "%llu\t" |
2788 | "%u\t" |
2789 | "%u\t" |
2790 | "%u\t", |
2791 | lock_num_prmode(lockres), |
2792 | lock_num_exmode(lockres), |
2793 | lock_num_prmode_failed(lockres), |
2794 | lock_num_exmode_failed(lockres), |
2795 | lock_total_prmode(lockres), |
2796 | lock_total_exmode(lockres), |
2797 | lock_max_prmode(lockres), |
2798 | lock_max_exmode(lockres), |
2799 | lock_refresh(lockres)); |
2800 | |
2801 | /* End the line */ |
2802 | seq_printf(m, "\n"); |
2803 | return 0; |
2804 | } |
2805 | |
2806 | static const struct seq_operations ocfs2_dlm_seq_ops = { |
2807 | .start = ocfs2_dlm_seq_start, |
2808 | .stop = ocfs2_dlm_seq_stop, |
2809 | .next = ocfs2_dlm_seq_next, |
2810 | .show = ocfs2_dlm_seq_show, |
2811 | }; |
2812 | |
2813 | static int ocfs2_dlm_debug_release(struct inode *inode, struct file *file) |
2814 | { |
2815 | struct seq_file *seq = (struct seq_file *) file->private_data; |
2816 | struct ocfs2_dlm_seq_priv *priv = seq->private; |
2817 | struct ocfs2_lock_res *res = &priv->p_iter_res; |
2818 | |
2819 | ocfs2_remove_lockres_tracking(res); |
2820 | ocfs2_put_dlm_debug(priv->p_dlm_debug); |
2821 | return seq_release_private(inode, file); |
2822 | } |
2823 | |
2824 | static int ocfs2_dlm_debug_open(struct inode *inode, struct file *file) |
2825 | { |
2826 | int ret; |
2827 | struct ocfs2_dlm_seq_priv *priv; |
2828 | struct seq_file *seq; |
2829 | struct ocfs2_super *osb; |
2830 | |
2831 | priv = kzalloc(sizeof(struct ocfs2_dlm_seq_priv), GFP_KERNEL); |
2832 | if (!priv) { |
2833 | ret = -ENOMEM; |
2834 | mlog_errno(ret); |
2835 | goto out; |
2836 | } |
2837 | osb = inode->i_private; |
2838 | ocfs2_get_dlm_debug(osb->osb_dlm_debug); |
2839 | priv->p_dlm_debug = osb->osb_dlm_debug; |
2840 | INIT_LIST_HEAD(&priv->p_iter_res.l_debug_list); |
2841 | |
2842 | ret = seq_open(file, &ocfs2_dlm_seq_ops); |
2843 | if (ret) { |
2844 | kfree(priv); |
2845 | mlog_errno(ret); |
2846 | goto out; |
2847 | } |
2848 | |
2849 | seq = (struct seq_file *) file->private_data; |
2850 | seq->private = priv; |
2851 | |
2852 | ocfs2_add_lockres_tracking(&priv->p_iter_res, |
2853 | priv->p_dlm_debug); |
2854 | |
2855 | out: |
2856 | return ret; |
2857 | } |
2858 | |
2859 | static const struct file_operations ocfs2_dlm_debug_fops = { |
2860 | .open = ocfs2_dlm_debug_open, |
2861 | .release = ocfs2_dlm_debug_release, |
2862 | .read = seq_read, |
2863 | .llseek = seq_lseek, |
2864 | }; |
2865 | |
2866 | static int ocfs2_dlm_init_debug(struct ocfs2_super *osb) |
2867 | { |
2868 | int ret = 0; |
2869 | struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug; |
2870 | |
2871 | dlm_debug->d_locking_state = debugfs_create_file("locking_state", |
2872 | S_IFREG|S_IRUSR, |
2873 | osb->osb_debug_root, |
2874 | osb, |
2875 | &ocfs2_dlm_debug_fops); |
2876 | if (!dlm_debug->d_locking_state) { |
2877 | ret = -EINVAL; |
2878 | mlog(ML_ERROR, |
2879 | "Unable to create locking state debugfs file.\n"); |
2880 | goto out; |
2881 | } |
2882 | |
2883 | ocfs2_get_dlm_debug(dlm_debug); |
2884 | out: |
2885 | return ret; |
2886 | } |
2887 | |
2888 | static void ocfs2_dlm_shutdown_debug(struct ocfs2_super *osb) |
2889 | { |
2890 | struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug; |
2891 | |
2892 | if (dlm_debug) { |
2893 | debugfs_remove(dlm_debug->d_locking_state); |
2894 | ocfs2_put_dlm_debug(dlm_debug); |
2895 | } |
2896 | } |
2897 | |
2898 | int ocfs2_dlm_init(struct ocfs2_super *osb) |
2899 | { |
2900 | int status = 0; |
2901 | struct ocfs2_cluster_connection *conn = NULL; |
2902 | |
2903 | mlog_entry_void(); |
2904 | |
2905 | if (ocfs2_mount_local(osb)) { |
2906 | osb->node_num = 0; |
2907 | goto local; |
2908 | } |
2909 | |
2910 | status = ocfs2_dlm_init_debug(osb); |
2911 | if (status < 0) { |
2912 | mlog_errno(status); |
2913 | goto bail; |
2914 | } |
2915 | |
2916 | /* launch downconvert thread */ |
2917 | osb->dc_task = kthread_run(ocfs2_downconvert_thread, osb, "ocfs2dc"); |
2918 | if (IS_ERR(osb->dc_task)) { |
2919 | status = PTR_ERR(osb->dc_task); |
2920 | osb->dc_task = NULL; |
2921 | mlog_errno(status); |
2922 | goto bail; |
2923 | } |
2924 | |
2925 | /* for now, uuid == domain */ |
2926 | status = ocfs2_cluster_connect(osb->osb_cluster_stack, |
2927 | osb->uuid_str, |
2928 | strlen(osb->uuid_str), |
2929 | ocfs2_do_node_down, osb, |
2930 | &conn); |
2931 | if (status) { |
2932 | mlog_errno(status); |
2933 | goto bail; |
2934 | } |
2935 | |
2936 | status = ocfs2_cluster_this_node(&osb->node_num); |
2937 | if (status < 0) { |
2938 | mlog_errno(status); |
2939 | mlog(ML_ERROR, |
2940 | "could not find this host's node number\n"); |
2941 | ocfs2_cluster_disconnect(conn, 0); |
2942 | goto bail; |
2943 | } |
2944 | |
2945 | local: |
2946 | ocfs2_super_lock_res_init(&osb->osb_super_lockres, osb); |
2947 | ocfs2_rename_lock_res_init(&osb->osb_rename_lockres, osb); |
2948 | ocfs2_nfs_sync_lock_res_init(&osb->osb_nfs_sync_lockres, osb); |
2949 | ocfs2_orphan_scan_lock_res_init(&osb->osb_orphan_scan.os_lockres, osb); |
2950 | |
2951 | osb->cconn = conn; |
2952 | |
2953 | status = 0; |
2954 | bail: |
2955 | if (status < 0) { |
2956 | ocfs2_dlm_shutdown_debug(osb); |
2957 | if (osb->dc_task) |
2958 | kthread_stop(osb->dc_task); |
2959 | } |
2960 | |
2961 | mlog_exit(status); |
2962 | return status; |
2963 | } |
2964 | |
2965 | void ocfs2_dlm_shutdown(struct ocfs2_super *osb, |
2966 | int hangup_pending) |
2967 | { |
2968 | mlog_entry_void(); |
2969 | |
2970 | ocfs2_drop_osb_locks(osb); |
2971 | |
2972 | /* |
2973 | * Now that we have dropped all locks and ocfs2_dismount_volume() |
2974 | * has disabled recovery, the DLM won't be talking to us. It's |
2975 | * safe to tear things down before disconnecting the cluster. |
2976 | */ |
2977 | |
2978 | if (osb->dc_task) { |
2979 | kthread_stop(osb->dc_task); |
2980 | osb->dc_task = NULL; |
2981 | } |
2982 | |
2983 | ocfs2_lock_res_free(&osb->osb_super_lockres); |
2984 | ocfs2_lock_res_free(&osb->osb_rename_lockres); |
2985 | ocfs2_lock_res_free(&osb->osb_nfs_sync_lockres); |
2986 | ocfs2_lock_res_free(&osb->osb_orphan_scan.os_lockres); |
2987 | |
2988 | ocfs2_cluster_disconnect(osb->cconn, hangup_pending); |
2989 | osb->cconn = NULL; |
2990 | |
2991 | ocfs2_dlm_shutdown_debug(osb); |
2992 | |
2993 | mlog_exit_void(); |
2994 | } |
2995 | |
2996 | static void ocfs2_unlock_ast(void *opaque, int error) |
2997 | { |
2998 | struct ocfs2_lock_res *lockres = opaque; |
2999 | unsigned long flags; |
3000 | |
3001 | mlog_entry_void(); |
3002 | |
3003 | mlog(0, "UNLOCK AST called on lock %s, action = %d\n", lockres->l_name, |
3004 | lockres->l_unlock_action); |
3005 | |
3006 | spin_lock_irqsave(&lockres->l_lock, flags); |
3007 | if (error) { |
3008 | mlog(ML_ERROR, "Dlm passes error %d for lock %s, " |
3009 | "unlock_action %d\n", error, lockres->l_name, |
3010 | lockres->l_unlock_action); |
3011 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
3012 | return; |
3013 | } |
3014 | |
3015 | switch(lockres->l_unlock_action) { |
3016 | case OCFS2_UNLOCK_CANCEL_CONVERT: |
3017 | mlog(0, "Cancel convert success for %s\n", lockres->l_name); |
3018 | lockres->l_action = OCFS2_AST_INVALID; |
3019 | /* Downconvert thread may have requeued this lock, we |
3020 | * need to wake it. */ |
3021 | if (lockres->l_flags & OCFS2_LOCK_BLOCKED) |
3022 | ocfs2_wake_downconvert_thread(ocfs2_get_lockres_osb(lockres)); |
3023 | break; |
3024 | case OCFS2_UNLOCK_DROP_LOCK: |
3025 | lockres->l_level = DLM_LOCK_IV; |
3026 | break; |
3027 | default: |
3028 | BUG(); |
3029 | } |
3030 | |
3031 | lockres_clear_flags(lockres, OCFS2_LOCK_BUSY); |
3032 | lockres->l_unlock_action = OCFS2_UNLOCK_INVALID; |
3033 | wake_up(&lockres->l_event); |
3034 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
3035 | |
3036 | mlog_exit_void(); |
3037 | } |
3038 | |
3039 | static int ocfs2_drop_lock(struct ocfs2_super *osb, |
3040 | struct ocfs2_lock_res *lockres) |
3041 | { |
3042 | int ret; |
3043 | unsigned long flags; |
3044 | u32 lkm_flags = 0; |
3045 | |
3046 | /* We didn't get anywhere near actually using this lockres. */ |
3047 | if (!(lockres->l_flags & OCFS2_LOCK_INITIALIZED)) |
3048 | goto out; |
3049 | |
3050 | if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) |
3051 | lkm_flags |= DLM_LKF_VALBLK; |
3052 | |
3053 | spin_lock_irqsave(&lockres->l_lock, flags); |
3054 | |
3055 | mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_FREEING), |
3056 | "lockres %s, flags 0x%lx\n", |
3057 | lockres->l_name, lockres->l_flags); |
3058 | |
3059 | while (lockres->l_flags & OCFS2_LOCK_BUSY) { |
3060 | mlog(0, "waiting on busy lock \"%s\": flags = %lx, action = " |
3061 | "%u, unlock_action = %u\n", |
3062 | lockres->l_name, lockres->l_flags, lockres->l_action, |
3063 | lockres->l_unlock_action); |
3064 | |
3065 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
3066 | |
3067 | /* XXX: Today we just wait on any busy |
3068 | * locks... Perhaps we need to cancel converts in the |
3069 | * future? */ |
3070 | ocfs2_wait_on_busy_lock(lockres); |
3071 | |
3072 | spin_lock_irqsave(&lockres->l_lock, flags); |
3073 | } |
3074 | |
3075 | if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) { |
3076 | if (lockres->l_flags & OCFS2_LOCK_ATTACHED && |
3077 | lockres->l_level == DLM_LOCK_EX && |
3078 | !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) |
3079 | lockres->l_ops->set_lvb(lockres); |
3080 | } |
3081 | |
3082 | if (lockres->l_flags & OCFS2_LOCK_BUSY) |
3083 | mlog(ML_ERROR, "destroying busy lock: \"%s\"\n", |
3084 | lockres->l_name); |
3085 | if (lockres->l_flags & OCFS2_LOCK_BLOCKED) |
3086 | mlog(0, "destroying blocked lock: \"%s\"\n", lockres->l_name); |
3087 | |
3088 | if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) { |
3089 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
3090 | goto out; |
3091 | } |
3092 | |
3093 | lockres_clear_flags(lockres, OCFS2_LOCK_ATTACHED); |
3094 | |
3095 | /* make sure we never get here while waiting for an ast to |
3096 | * fire. */ |
3097 | BUG_ON(lockres->l_action != OCFS2_AST_INVALID); |
3098 | |
3099 | /* is this necessary? */ |
3100 | lockres_or_flags(lockres, OCFS2_LOCK_BUSY); |
3101 | lockres->l_unlock_action = OCFS2_UNLOCK_DROP_LOCK; |
3102 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
3103 | |
3104 | mlog(0, "lock %s\n", lockres->l_name); |
3105 | |
3106 | ret = ocfs2_dlm_unlock(osb->cconn, &lockres->l_lksb, lkm_flags, |
3107 | lockres); |
3108 | if (ret) { |
3109 | ocfs2_log_dlm_error("ocfs2_dlm_unlock", ret, lockres); |
3110 | mlog(ML_ERROR, "lockres flags: %lu\n", lockres->l_flags); |
3111 | ocfs2_dlm_dump_lksb(&lockres->l_lksb); |
3112 | BUG(); |
3113 | } |
3114 | mlog(0, "lock %s, successful return from ocfs2_dlm_unlock\n", |
3115 | lockres->l_name); |
3116 | |
3117 | ocfs2_wait_on_busy_lock(lockres); |
3118 | out: |
3119 | mlog_exit(0); |
3120 | return 0; |
3121 | } |
3122 | |
3123 | /* Mark the lockres as being dropped. It will no longer be |
3124 | * queued if blocking, but we still may have to wait on it |
3125 | * being dequeued from the downconvert thread before we can consider |
3126 | * it safe to drop. |
3127 | * |
3128 | * You can *not* attempt to call cluster_lock on this lockres anymore. */ |
3129 | void ocfs2_mark_lockres_freeing(struct ocfs2_lock_res *lockres) |
3130 | { |
3131 | int status; |
3132 | struct ocfs2_mask_waiter mw; |
3133 | unsigned long flags; |
3134 | |
3135 | ocfs2_init_mask_waiter(&mw); |
3136 | |
3137 | spin_lock_irqsave(&lockres->l_lock, flags); |
3138 | lockres->l_flags |= OCFS2_LOCK_FREEING; |
3139 | while (lockres->l_flags & OCFS2_LOCK_QUEUED) { |
3140 | lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_QUEUED, 0); |
3141 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
3142 | |
3143 | mlog(0, "Waiting on lockres %s\n", lockres->l_name); |
3144 | |
3145 | status = ocfs2_wait_for_mask(&mw); |
3146 | if (status) |
3147 | mlog_errno(status); |
3148 | |
3149 | spin_lock_irqsave(&lockres->l_lock, flags); |
3150 | } |
3151 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
3152 | } |
3153 | |
3154 | void ocfs2_simple_drop_lockres(struct ocfs2_super *osb, |
3155 | struct ocfs2_lock_res *lockres) |
3156 | { |
3157 | int ret; |
3158 | |
3159 | ocfs2_mark_lockres_freeing(lockres); |
3160 | ret = ocfs2_drop_lock(osb, lockres); |
3161 | if (ret) |
3162 | mlog_errno(ret); |
3163 | } |
3164 | |
3165 | static void ocfs2_drop_osb_locks(struct ocfs2_super *osb) |
3166 | { |
3167 | ocfs2_simple_drop_lockres(osb, &osb->osb_super_lockres); |
3168 | ocfs2_simple_drop_lockres(osb, &osb->osb_rename_lockres); |
3169 | ocfs2_simple_drop_lockres(osb, &osb->osb_nfs_sync_lockres); |
3170 | ocfs2_simple_drop_lockres(osb, &osb->osb_orphan_scan.os_lockres); |
3171 | } |
3172 | |
3173 | int ocfs2_drop_inode_locks(struct inode *inode) |
3174 | { |
3175 | int status, err; |
3176 | |
3177 | mlog_entry_void(); |
3178 | |
3179 | /* No need to call ocfs2_mark_lockres_freeing here - |
3180 | * ocfs2_clear_inode has done it for us. */ |
3181 | |
3182 | err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb), |
3183 | &OCFS2_I(inode)->ip_open_lockres); |
3184 | if (err < 0) |
3185 | mlog_errno(err); |
3186 | |
3187 | status = err; |
3188 | |
3189 | err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb), |
3190 | &OCFS2_I(inode)->ip_inode_lockres); |
3191 | if (err < 0) |
3192 | mlog_errno(err); |
3193 | if (err < 0 && !status) |
3194 | status = err; |
3195 | |
3196 | err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb), |
3197 | &OCFS2_I(inode)->ip_rw_lockres); |
3198 | if (err < 0) |
3199 | mlog_errno(err); |
3200 | if (err < 0 && !status) |
3201 | status = err; |
3202 | |
3203 | mlog_exit(status); |
3204 | return status; |
3205 | } |
3206 | |
3207 | static unsigned int ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres, |
3208 | int new_level) |
3209 | { |
3210 | assert_spin_locked(&lockres->l_lock); |
3211 | |
3212 | BUG_ON(lockres->l_blocking <= DLM_LOCK_NL); |
3213 | |
3214 | if (lockres->l_level <= new_level) { |
3215 | mlog(ML_ERROR, "lockres->l_level (%d) <= new_level (%d)\n", |
3216 | lockres->l_level, new_level); |
3217 | BUG(); |
3218 | } |
3219 | |
3220 | mlog(0, "lock %s, new_level = %d, l_blocking = %d\n", |
3221 | lockres->l_name, new_level, lockres->l_blocking); |
3222 | |
3223 | lockres->l_action = OCFS2_AST_DOWNCONVERT; |
3224 | lockres->l_requested = new_level; |
3225 | lockres_or_flags(lockres, OCFS2_LOCK_BUSY); |
3226 | return lockres_set_pending(lockres); |
3227 | } |
3228 | |
3229 | static int ocfs2_downconvert_lock(struct ocfs2_super *osb, |
3230 | struct ocfs2_lock_res *lockres, |
3231 | int new_level, |
3232 | int lvb, |
3233 | unsigned int generation) |
3234 | { |
3235 | int ret; |
3236 | u32 dlm_flags = DLM_LKF_CONVERT; |
3237 | |
3238 | mlog_entry_void(); |
3239 | |
3240 | if (lvb) |
3241 | dlm_flags |= DLM_LKF_VALBLK; |
3242 | |
3243 | ret = ocfs2_dlm_lock(osb->cconn, |
3244 | new_level, |
3245 | &lockres->l_lksb, |
3246 | dlm_flags, |
3247 | lockres->l_name, |
3248 | OCFS2_LOCK_ID_MAX_LEN - 1, |
3249 | lockres); |
3250 | lockres_clear_pending(lockres, generation, osb); |
3251 | if (ret) { |
3252 | ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres); |
3253 | ocfs2_recover_from_dlm_error(lockres, 1); |
3254 | goto bail; |
3255 | } |
3256 | |
3257 | ret = 0; |
3258 | bail: |
3259 | mlog_exit(ret); |
3260 | return ret; |
3261 | } |
3262 | |
3263 | /* returns 1 when the caller should unlock and call ocfs2_dlm_unlock */ |
3264 | static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb, |
3265 | struct ocfs2_lock_res *lockres) |
3266 | { |
3267 | assert_spin_locked(&lockres->l_lock); |
3268 | |
3269 | mlog_entry_void(); |
3270 | mlog(0, "lock %s\n", lockres->l_name); |
3271 | |
3272 | if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) { |
3273 | /* If we're already trying to cancel a lock conversion |
3274 | * then just drop the spinlock and allow the caller to |
3275 | * requeue this lock. */ |
3276 | |
3277 | mlog(0, "Lockres %s, skip convert\n", lockres->l_name); |
3278 | return 0; |
3279 | } |
3280 | |
3281 | /* were we in a convert when we got the bast fire? */ |
3282 | BUG_ON(lockres->l_action != OCFS2_AST_CONVERT && |
3283 | lockres->l_action != OCFS2_AST_DOWNCONVERT); |
3284 | /* set things up for the unlockast to know to just |
3285 | * clear out the ast_action and unset busy, etc. */ |
3286 | lockres->l_unlock_action = OCFS2_UNLOCK_CANCEL_CONVERT; |
3287 | |
3288 | mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_BUSY), |
3289 | "lock %s, invalid flags: 0x%lx\n", |
3290 | lockres->l_name, lockres->l_flags); |
3291 | |
3292 | return 1; |
3293 | } |
3294 | |
3295 | static int ocfs2_cancel_convert(struct ocfs2_super *osb, |
3296 | struct ocfs2_lock_res *lockres) |
3297 | { |
3298 | int ret; |
3299 | |
3300 | mlog_entry_void(); |
3301 | mlog(0, "lock %s\n", lockres->l_name); |
3302 | |
3303 | ret = ocfs2_dlm_unlock(osb->cconn, &lockres->l_lksb, |
3304 | DLM_LKF_CANCEL, lockres); |
3305 | if (ret) { |
3306 | ocfs2_log_dlm_error("ocfs2_dlm_unlock", ret, lockres); |
3307 | ocfs2_recover_from_dlm_error(lockres, 0); |
3308 | } |
3309 | |
3310 | mlog(0, "lock %s return from ocfs2_dlm_unlock\n", lockres->l_name); |
3311 | |
3312 | mlog_exit(ret); |
3313 | return ret; |
3314 | } |
3315 | |
3316 | static int ocfs2_unblock_lock(struct ocfs2_super *osb, |
3317 | struct ocfs2_lock_res *lockres, |
3318 | struct ocfs2_unblock_ctl *ctl) |
3319 | { |
3320 | unsigned long flags; |
3321 | int blocking; |
3322 | int new_level; |
3323 | int ret = 0; |
3324 | int set_lvb = 0; |
3325 | unsigned int gen; |
3326 | |
3327 | mlog_entry_void(); |
3328 | |
3329 | spin_lock_irqsave(&lockres->l_lock, flags); |
3330 | |
3331 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED)); |
3332 | |
3333 | recheck: |
3334 | if (lockres->l_flags & OCFS2_LOCK_BUSY) { |
3335 | /* XXX |
3336 | * This is a *big* race. The OCFS2_LOCK_PENDING flag |
3337 | * exists entirely for one reason - another thread has set |
3338 | * OCFS2_LOCK_BUSY, but has *NOT* yet called dlm_lock(). |
3339 | * |
3340 | * If we do ocfs2_cancel_convert() before the other thread |
3341 | * calls dlm_lock(), our cancel will do nothing. We will |
3342 | * get no ast, and we will have no way of knowing the |
3343 | * cancel failed. Meanwhile, the other thread will call |
3344 | * into dlm_lock() and wait...forever. |
3345 | * |
3346 | * Why forever? Because another node has asked for the |
3347 | * lock first; that's why we're here in unblock_lock(). |
3348 | * |
3349 | * The solution is OCFS2_LOCK_PENDING. When PENDING is |
3350 | * set, we just requeue the unblock. Only when the other |
3351 | * thread has called dlm_lock() and cleared PENDING will |
3352 | * we then cancel their request. |
3353 | * |
3354 | * All callers of dlm_lock() must set OCFS2_DLM_PENDING |
3355 | * at the same time they set OCFS2_DLM_BUSY. They must |
3356 | * clear OCFS2_DLM_PENDING after dlm_lock() returns. |
3357 | */ |
3358 | if (lockres->l_flags & OCFS2_LOCK_PENDING) |
3359 | goto leave_requeue; |
3360 | |
3361 | ctl->requeue = 1; |
3362 | ret = ocfs2_prepare_cancel_convert(osb, lockres); |
3363 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
3364 | if (ret) { |
3365 | ret = ocfs2_cancel_convert(osb, lockres); |
3366 | if (ret < 0) |
3367 | mlog_errno(ret); |
3368 | } |
3369 | goto leave; |
3370 | } |
3371 | |
3372 | /* if we're blocking an exclusive and we have *any* holders, |
3373 | * then requeue. */ |
3374 | if ((lockres->l_blocking == DLM_LOCK_EX) |
3375 | && (lockres->l_ex_holders || lockres->l_ro_holders)) |
3376 | goto leave_requeue; |
3377 | |
3378 | /* If it's a PR we're blocking, then only |
3379 | * requeue if we've got any EX holders */ |
3380 | if (lockres->l_blocking == DLM_LOCK_PR && |
3381 | lockres->l_ex_holders) |
3382 | goto leave_requeue; |
3383 | |
3384 | /* |
3385 | * Can we get a lock in this state if the holder counts are |
3386 | * zero? The meta data unblock code used to check this. |
3387 | */ |
3388 | if ((lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH) |
3389 | && (lockres->l_flags & OCFS2_LOCK_REFRESHING)) |
3390 | goto leave_requeue; |
3391 | |
3392 | new_level = ocfs2_highest_compat_lock_level(lockres->l_blocking); |
3393 | |
3394 | if (lockres->l_ops->check_downconvert |
3395 | && !lockres->l_ops->check_downconvert(lockres, new_level)) |
3396 | goto leave_requeue; |
3397 | |
3398 | /* If we get here, then we know that there are no more |
3399 | * incompatible holders (and anyone asking for an incompatible |
3400 | * lock is blocked). We can now downconvert the lock */ |
3401 | if (!lockres->l_ops->downconvert_worker) |
3402 | goto downconvert; |
3403 | |
3404 | /* Some lockres types want to do a bit of work before |
3405 | * downconverting a lock. Allow that here. The worker function |
3406 | * may sleep, so we save off a copy of what we're blocking as |
3407 | * it may change while we're not holding the spin lock. */ |
3408 | blocking = lockres->l_blocking; |
3409 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
3410 | |
3411 | ctl->unblock_action = lockres->l_ops->downconvert_worker(lockres, blocking); |
3412 | |
3413 | if (ctl->unblock_action == UNBLOCK_STOP_POST) |
3414 | goto leave; |
3415 | |
3416 | spin_lock_irqsave(&lockres->l_lock, flags); |
3417 | if (blocking != lockres->l_blocking) { |
3418 | /* If this changed underneath us, then we can't drop |
3419 | * it just yet. */ |
3420 | goto recheck; |
3421 | } |
3422 | |
3423 | downconvert: |
3424 | ctl->requeue = 0; |
3425 | |
3426 | if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) { |
3427 | if (lockres->l_level == DLM_LOCK_EX) |
3428 | set_lvb = 1; |
3429 | |
3430 | /* |
3431 | * We only set the lvb if the lock has been fully |
3432 | * refreshed - otherwise we risk setting stale |
3433 | * data. Otherwise, there's no need to actually clear |
3434 | * out the lvb here as it's value is still valid. |
3435 | */ |
3436 | if (set_lvb && !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) |
3437 | lockres->l_ops->set_lvb(lockres); |
3438 | } |
3439 | |
3440 | gen = ocfs2_prepare_downconvert(lockres, new_level); |
3441 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
3442 | ret = ocfs2_downconvert_lock(osb, lockres, new_level, set_lvb, |
3443 | gen); |
3444 | |
3445 | leave: |
3446 | mlog_exit(ret); |
3447 | return ret; |
3448 | |
3449 | leave_requeue: |
3450 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
3451 | ctl->requeue = 1; |
3452 | |
3453 | mlog_exit(0); |
3454 | return 0; |
3455 | } |
3456 | |
3457 | static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres, |
3458 | int blocking) |
3459 | { |
3460 | struct inode *inode; |
3461 | struct address_space *mapping; |
3462 | |
3463 | inode = ocfs2_lock_res_inode(lockres); |
3464 | mapping = inode->i_mapping; |
3465 | |
3466 | if (!S_ISREG(inode->i_mode)) |
3467 | goto out; |
3468 | |
3469 | /* |
3470 | * We need this before the filemap_fdatawrite() so that it can |
3471 | * transfer the dirty bit from the PTE to the |
3472 | * page. Unfortunately this means that even for EX->PR |
3473 | * downconverts, we'll lose our mappings and have to build |
3474 | * them up again. |
3475 | */ |
3476 | unmap_mapping_range(mapping, 0, 0, 0); |
3477 | |
3478 | if (filemap_fdatawrite(mapping)) { |
3479 | mlog(ML_ERROR, "Could not sync inode %llu for downconvert!", |
3480 | (unsigned long long)OCFS2_I(inode)->ip_blkno); |
3481 | } |
3482 | sync_mapping_buffers(mapping); |
3483 | if (blocking == DLM_LOCK_EX) { |
3484 | truncate_inode_pages(mapping, 0); |
3485 | } else { |
3486 | /* We only need to wait on the I/O if we're not also |
3487 | * truncating pages because truncate_inode_pages waits |
3488 | * for us above. We don't truncate pages if we're |
3489 | * blocking anything < EXMODE because we want to keep |
3490 | * them around in that case. */ |
3491 | filemap_fdatawait(mapping); |
3492 | } |
3493 | |
3494 | out: |
3495 | return UNBLOCK_CONTINUE; |
3496 | } |
3497 | |
3498 | static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres, |
3499 | int new_level) |
3500 | { |
3501 | struct inode *inode = ocfs2_lock_res_inode(lockres); |
3502 | int checkpointed = ocfs2_inode_fully_checkpointed(inode); |
3503 | |
3504 | BUG_ON(new_level != DLM_LOCK_NL && new_level != DLM_LOCK_PR); |
3505 | BUG_ON(lockres->l_level != DLM_LOCK_EX && !checkpointed); |
3506 | |
3507 | if (checkpointed) |
3508 | return 1; |
3509 | |
3510 | ocfs2_start_checkpoint(OCFS2_SB(inode->i_sb)); |
3511 | return 0; |
3512 | } |
3513 | |
3514 | static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres) |
3515 | { |
3516 | struct inode *inode = ocfs2_lock_res_inode(lockres); |
3517 | |
3518 | __ocfs2_stuff_meta_lvb(inode); |
3519 | } |
3520 | |
3521 | /* |
3522 | * Does the final reference drop on our dentry lock. Right now this |
3523 | * happens in the downconvert thread, but we could choose to simplify the |
3524 | * dlmglue API and push these off to the ocfs2_wq in the future. |
3525 | */ |
3526 | static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb, |
3527 | struct ocfs2_lock_res *lockres) |
3528 | { |
3529 | struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres); |
3530 | ocfs2_dentry_lock_put(osb, dl); |
3531 | } |
3532 | |
3533 | /* |
3534 | * d_delete() matching dentries before the lock downconvert. |
3535 | * |
3536 | * At this point, any process waiting to destroy the |
3537 | * dentry_lock due to last ref count is stopped by the |
3538 | * OCFS2_LOCK_QUEUED flag. |
3539 | * |
3540 | * We have two potential problems |
3541 | * |
3542 | * 1) If we do the last reference drop on our dentry_lock (via dput) |
3543 | * we'll wind up in ocfs2_release_dentry_lock(), waiting on |
3544 | * the downconvert to finish. Instead we take an elevated |
3545 | * reference and push the drop until after we've completed our |
3546 | * unblock processing. |
3547 | * |
3548 | * 2) There might be another process with a final reference, |
3549 | * waiting on us to finish processing. If this is the case, we |
3550 | * detect it and exit out - there's no more dentries anyway. |
3551 | */ |
3552 | static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres, |
3553 | int blocking) |
3554 | { |
3555 | struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres); |
3556 | struct ocfs2_inode_info *oi = OCFS2_I(dl->dl_inode); |
3557 | struct dentry *dentry; |
3558 | unsigned long flags; |
3559 | int extra_ref = 0; |
3560 | |
3561 | /* |
3562 | * This node is blocking another node from getting a read |
3563 | * lock. This happens when we've renamed within a |
3564 | * directory. We've forced the other nodes to d_delete(), but |
3565 | * we never actually dropped our lock because it's still |
3566 | * valid. The downconvert code will retain a PR for this node, |
3567 | * so there's no further work to do. |
3568 | */ |
3569 | if (blocking == DLM_LOCK_PR) |
3570 | return UNBLOCK_CONTINUE; |
3571 | |
3572 | /* |
3573 | * Mark this inode as potentially orphaned. The code in |
3574 | * ocfs2_delete_inode() will figure out whether it actually |
3575 | * needs to be freed or not. |
3576 | */ |
3577 | spin_lock(&oi->ip_lock); |
3578 | oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED; |
3579 | spin_unlock(&oi->ip_lock); |
3580 | |
3581 | /* |
3582 | * Yuck. We need to make sure however that the check of |
3583 | * OCFS2_LOCK_FREEING and the extra reference are atomic with |
3584 | * respect to a reference decrement or the setting of that |
3585 | * flag. |
3586 | */ |
3587 | spin_lock_irqsave(&lockres->l_lock, flags); |
3588 | spin_lock(&dentry_attach_lock); |
3589 | if (!(lockres->l_flags & OCFS2_LOCK_FREEING) |
3590 | && dl->dl_count) { |
3591 | dl->dl_count++; |
3592 | extra_ref = 1; |
3593 | } |
3594 | spin_unlock(&dentry_attach_lock); |
3595 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
3596 | |
3597 | mlog(0, "extra_ref = %d\n", extra_ref); |
3598 | |
3599 | /* |
3600 | * We have a process waiting on us in ocfs2_dentry_iput(), |
3601 | * which means we can't have any more outstanding |
3602 | * aliases. There's no need to do any more work. |
3603 | */ |
3604 | if (!extra_ref) |
3605 | return UNBLOCK_CONTINUE; |
3606 | |
3607 | spin_lock(&dentry_attach_lock); |
3608 | while (1) { |
3609 | dentry = ocfs2_find_local_alias(dl->dl_inode, |
3610 | dl->dl_parent_blkno, 1); |
3611 | if (!dentry) |
3612 | break; |
3613 | spin_unlock(&dentry_attach_lock); |
3614 | |
3615 | mlog(0, "d_delete(%.*s);\n", dentry->d_name.len, |
3616 | dentry->d_name.name); |
3617 | |
3618 | /* |
3619 | * The following dcache calls may do an |
3620 | * iput(). Normally we don't want that from the |
3621 | * downconverting thread, but in this case it's ok |
3622 | * because the requesting node already has an |
3623 | * exclusive lock on the inode, so it can't be queued |
3624 | * for a downconvert. |
3625 | */ |
3626 | d_delete(dentry); |
3627 | dput(dentry); |
3628 | |
3629 | spin_lock(&dentry_attach_lock); |
3630 | } |
3631 | spin_unlock(&dentry_attach_lock); |
3632 | |
3633 | /* |
3634 | * If we are the last holder of this dentry lock, there is no |
3635 | * reason to downconvert so skip straight to the unlock. |
3636 | */ |
3637 | if (dl->dl_count == 1) |
3638 | return UNBLOCK_STOP_POST; |
3639 | |
3640 | return UNBLOCK_CONTINUE_POST; |
3641 | } |
3642 | |
3643 | static void ocfs2_set_qinfo_lvb(struct ocfs2_lock_res *lockres) |
3644 | { |
3645 | struct ocfs2_qinfo_lvb *lvb; |
3646 | struct ocfs2_mem_dqinfo *oinfo = ocfs2_lock_res_qinfo(lockres); |
3647 | struct mem_dqinfo *info = sb_dqinfo(oinfo->dqi_gi.dqi_sb, |
3648 | oinfo->dqi_gi.dqi_type); |
3649 | |
3650 | mlog_entry_void(); |
3651 | |
3652 | lvb = ocfs2_dlm_lvb(&lockres->l_lksb); |
3653 | lvb->lvb_version = OCFS2_QINFO_LVB_VERSION; |
3654 | lvb->lvb_bgrace = cpu_to_be32(info->dqi_bgrace); |
3655 | lvb->lvb_igrace = cpu_to_be32(info->dqi_igrace); |
3656 | lvb->lvb_syncms = cpu_to_be32(oinfo->dqi_syncms); |
3657 | lvb->lvb_blocks = cpu_to_be32(oinfo->dqi_gi.dqi_blocks); |
3658 | lvb->lvb_free_blk = cpu_to_be32(oinfo->dqi_gi.dqi_free_blk); |
3659 | lvb->lvb_free_entry = cpu_to_be32(oinfo->dqi_gi.dqi_free_entry); |
3660 | |
3661 | mlog_exit_void(); |
3662 | } |
3663 | |
3664 | void ocfs2_qinfo_unlock(struct ocfs2_mem_dqinfo *oinfo, int ex) |
3665 | { |
3666 | struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock; |
3667 | struct ocfs2_super *osb = OCFS2_SB(oinfo->dqi_gi.dqi_sb); |
3668 | int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR; |
3669 | |
3670 | mlog_entry_void(); |
3671 | if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb)) |
3672 | ocfs2_cluster_unlock(osb, lockres, level); |
3673 | mlog_exit_void(); |
3674 | } |
3675 | |
3676 | static int ocfs2_refresh_qinfo(struct ocfs2_mem_dqinfo *oinfo) |
3677 | { |
3678 | struct mem_dqinfo *info = sb_dqinfo(oinfo->dqi_gi.dqi_sb, |
3679 | oinfo->dqi_gi.dqi_type); |
3680 | struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock; |
3681 | struct ocfs2_qinfo_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb); |
3682 | struct buffer_head *bh = NULL; |
3683 | struct ocfs2_global_disk_dqinfo *gdinfo; |
3684 | int status = 0; |
3685 | |
3686 | if (ocfs2_dlm_lvb_valid(&lockres->l_lksb) && |
3687 | lvb->lvb_version == OCFS2_QINFO_LVB_VERSION) { |
3688 | info->dqi_bgrace = be32_to_cpu(lvb->lvb_bgrace); |
3689 | info->dqi_igrace = be32_to_cpu(lvb->lvb_igrace); |
3690 | oinfo->dqi_syncms = be32_to_cpu(lvb->lvb_syncms); |
3691 | oinfo->dqi_gi.dqi_blocks = be32_to_cpu(lvb->lvb_blocks); |
3692 | oinfo->dqi_gi.dqi_free_blk = be32_to_cpu(lvb->lvb_free_blk); |
3693 | oinfo->dqi_gi.dqi_free_entry = |
3694 | be32_to_cpu(lvb->lvb_free_entry); |
3695 | } else { |
3696 | status = ocfs2_read_quota_block(oinfo->dqi_gqinode, 0, &bh); |
3697 | if (status) { |
3698 | mlog_errno(status); |
3699 | goto bail; |
3700 | } |
3701 | gdinfo = (struct ocfs2_global_disk_dqinfo *) |
3702 | (bh->b_data + OCFS2_GLOBAL_INFO_OFF); |
3703 | info->dqi_bgrace = le32_to_cpu(gdinfo->dqi_bgrace); |
3704 | info->dqi_igrace = le32_to_cpu(gdinfo->dqi_igrace); |
3705 | oinfo->dqi_syncms = le32_to_cpu(gdinfo->dqi_syncms); |
3706 | oinfo->dqi_gi.dqi_blocks = le32_to_cpu(gdinfo->dqi_blocks); |
3707 | oinfo->dqi_gi.dqi_free_blk = le32_to_cpu(gdinfo->dqi_free_blk); |
3708 | oinfo->dqi_gi.dqi_free_entry = |
3709 | le32_to_cpu(gdinfo->dqi_free_entry); |
3710 | brelse(bh); |
3711 | ocfs2_track_lock_refresh(lockres); |
3712 | } |
3713 | |
3714 | bail: |
3715 | return status; |
3716 | } |
3717 | |
3718 | /* Lock quota info, this function expects at least shared lock on the quota file |
3719 | * so that we can safely refresh quota info from disk. */ |
3720 | int ocfs2_qinfo_lock(struct ocfs2_mem_dqinfo *oinfo, int ex) |
3721 | { |
3722 | struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock; |
3723 | struct ocfs2_super *osb = OCFS2_SB(oinfo->dqi_gi.dqi_sb); |
3724 | int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR; |
3725 | int status = 0; |
3726 | |
3727 | mlog_entry_void(); |
3728 | |
3729 | /* On RO devices, locking really isn't needed... */ |
3730 | if (ocfs2_is_hard_readonly(osb)) { |
3731 | if (ex) |
3732 | status = -EROFS; |
3733 | goto bail; |
3734 | } |
3735 | if (ocfs2_mount_local(osb)) |
3736 | goto bail; |
3737 | |
3738 | status = ocfs2_cluster_lock(osb, lockres, level, 0, 0); |
3739 | if (status < 0) { |
3740 | mlog_errno(status); |
3741 | goto bail; |
3742 | } |
3743 | if (!ocfs2_should_refresh_lock_res(lockres)) |
3744 | goto bail; |
3745 | /* OK, we have the lock but we need to refresh the quota info */ |
3746 | status = ocfs2_refresh_qinfo(oinfo); |
3747 | if (status) |
3748 | ocfs2_qinfo_unlock(oinfo, ex); |
3749 | ocfs2_complete_lock_res_refresh(lockres, status); |
3750 | bail: |
3751 | mlog_exit(status); |
3752 | return status; |
3753 | } |
3754 | |
3755 | /* |
3756 | * This is the filesystem locking protocol. It provides the lock handling |
3757 | * hooks for the underlying DLM. It has a maximum version number. |
3758 | * The version number allows interoperability with systems running at |
3759 | * the same major number and an equal or smaller minor number. |
3760 | * |
3761 | * Whenever the filesystem does new things with locks (adds or removes a |
3762 | * lock, orders them differently, does different things underneath a lock), |
3763 | * the version must be changed. The protocol is negotiated when joining |
3764 | * the dlm domain. A node may join the domain if its major version is |
3765 | * identical to all other nodes and its minor version is greater than |
3766 | * or equal to all other nodes. When its minor version is greater than |
3767 | * the other nodes, it will run at the minor version specified by the |
3768 | * other nodes. |
3769 | * |
3770 | * If a locking change is made that will not be compatible with older |
3771 | * versions, the major number must be increased and the minor version set |
3772 | * to zero. If a change merely adds a behavior that can be disabled when |
3773 | * speaking to older versions, the minor version must be increased. If a |
3774 | * change adds a fully backwards compatible change (eg, LVB changes that |
3775 | * are just ignored by older versions), the version does not need to be |
3776 | * updated. |
3777 | */ |
3778 | static struct ocfs2_locking_protocol lproto = { |
3779 | .lp_max_version = { |
3780 | .pv_major = OCFS2_LOCKING_PROTOCOL_MAJOR, |
3781 | .pv_minor = OCFS2_LOCKING_PROTOCOL_MINOR, |
3782 | }, |
3783 | .lp_lock_ast = ocfs2_locking_ast, |
3784 | .lp_blocking_ast = ocfs2_blocking_ast, |
3785 | .lp_unlock_ast = ocfs2_unlock_ast, |
3786 | }; |
3787 | |
3788 | void ocfs2_set_locking_protocol(void) |
3789 | { |
3790 | ocfs2_stack_glue_set_locking_protocol(&lproto); |
3791 | } |
3792 | |
3793 | |
3794 | static void ocfs2_process_blocked_lock(struct ocfs2_super *osb, |
3795 | struct ocfs2_lock_res *lockres) |
3796 | { |
3797 | int status; |
3798 | struct ocfs2_unblock_ctl ctl = {0, 0,}; |
3799 | unsigned long flags; |
3800 | |
3801 | /* Our reference to the lockres in this function can be |
3802 | * considered valid until we remove the OCFS2_LOCK_QUEUED |
3803 | * flag. */ |
3804 | |
3805 | mlog_entry_void(); |
3806 | |
3807 | BUG_ON(!lockres); |
3808 | BUG_ON(!lockres->l_ops); |
3809 | |
3810 | mlog(0, "lockres %s blocked.\n", lockres->l_name); |
3811 | |
3812 | /* Detect whether a lock has been marked as going away while |
3813 | * the downconvert thread was processing other things. A lock can |
3814 | * still be marked with OCFS2_LOCK_FREEING after this check, |
3815 | * but short circuiting here will still save us some |
3816 | * performance. */ |
3817 | spin_lock_irqsave(&lockres->l_lock, flags); |
3818 | if (lockres->l_flags & OCFS2_LOCK_FREEING) |
3819 | goto unqueue; |
3820 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
3821 | |
3822 | status = ocfs2_unblock_lock(osb, lockres, &ctl); |
3823 | if (status < 0) |
3824 | mlog_errno(status); |
3825 | |
3826 | spin_lock_irqsave(&lockres->l_lock, flags); |
3827 | unqueue: |
3828 | if (lockres->l_flags & OCFS2_LOCK_FREEING || !ctl.requeue) { |
3829 | lockres_clear_flags(lockres, OCFS2_LOCK_QUEUED); |
3830 | } else |
3831 | ocfs2_schedule_blocked_lock(osb, lockres); |
3832 | |
3833 | mlog(0, "lockres %s, requeue = %s.\n", lockres->l_name, |
3834 | ctl.requeue ? "yes" : "no"); |
3835 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
3836 | |
3837 | if (ctl.unblock_action != UNBLOCK_CONTINUE |
3838 | && lockres->l_ops->post_unlock) |
3839 | lockres->l_ops->post_unlock(osb, lockres); |
3840 | |
3841 | mlog_exit_void(); |
3842 | } |
3843 | |
3844 | static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb, |
3845 | struct ocfs2_lock_res *lockres) |
3846 | { |
3847 | mlog_entry_void(); |
3848 | |
3849 | assert_spin_locked(&lockres->l_lock); |
3850 | |
3851 | if (lockres->l_flags & OCFS2_LOCK_FREEING) { |
3852 | /* Do not schedule a lock for downconvert when it's on |
3853 | * the way to destruction - any nodes wanting access |
3854 | * to the resource will get it soon. */ |
3855 | mlog(0, "Lockres %s won't be scheduled: flags 0x%lx\n", |
3856 | lockres->l_name, lockres->l_flags); |
3857 | return; |
3858 | } |
3859 | |
3860 | lockres_or_flags(lockres, OCFS2_LOCK_QUEUED); |
3861 | |
3862 | spin_lock(&osb->dc_task_lock); |
3863 | if (list_empty(&lockres->l_blocked_list)) { |
3864 | list_add_tail(&lockres->l_blocked_list, |
3865 | &osb->blocked_lock_list); |
3866 | osb->blocked_lock_count++; |
3867 | } |
3868 | spin_unlock(&osb->dc_task_lock); |
3869 | |
3870 | mlog_exit_void(); |
3871 | } |
3872 | |
3873 | static void ocfs2_downconvert_thread_do_work(struct ocfs2_super *osb) |
3874 | { |
3875 | unsigned long processed; |
3876 | struct ocfs2_lock_res *lockres; |
3877 | |
3878 | mlog_entry_void(); |
3879 | |
3880 | spin_lock(&osb->dc_task_lock); |
3881 | /* grab this early so we know to try again if a state change and |
3882 | * wake happens part-way through our work */ |
3883 | osb->dc_work_sequence = osb->dc_wake_sequence; |
3884 | |
3885 | processed = osb->blocked_lock_count; |
3886 | while (processed) { |
3887 | BUG_ON(list_empty(&osb->blocked_lock_list)); |
3888 | |
3889 | lockres = list_entry(osb->blocked_lock_list.next, |
3890 | struct ocfs2_lock_res, l_blocked_list); |
3891 | list_del_init(&lockres->l_blocked_list); |
3892 | osb->blocked_lock_count--; |
3893 | spin_unlock(&osb->dc_task_lock); |
3894 | |
3895 | BUG_ON(!processed); |
3896 | processed--; |
3897 | |
3898 | ocfs2_process_blocked_lock(osb, lockres); |
3899 | |
3900 | spin_lock(&osb->dc_task_lock); |
3901 | } |
3902 | spin_unlock(&osb->dc_task_lock); |
3903 | |
3904 | mlog_exit_void(); |
3905 | } |
3906 | |
3907 | static int ocfs2_downconvert_thread_lists_empty(struct ocfs2_super *osb) |
3908 | { |
3909 | int empty = 0; |
3910 | |
3911 | spin_lock(&osb->dc_task_lock); |
3912 | if (list_empty(&osb->blocked_lock_list)) |
3913 | empty = 1; |
3914 | |
3915 | spin_unlock(&osb->dc_task_lock); |
3916 | return empty; |
3917 | } |
3918 | |
3919 | static int ocfs2_downconvert_thread_should_wake(struct ocfs2_super *osb) |
3920 | { |
3921 | int should_wake = 0; |
3922 | |
3923 | spin_lock(&osb->dc_task_lock); |
3924 | if (osb->dc_work_sequence != osb->dc_wake_sequence) |
3925 | should_wake = 1; |
3926 | spin_unlock(&osb->dc_task_lock); |
3927 | |
3928 | return should_wake; |
3929 | } |
3930 | |
3931 | static int ocfs2_downconvert_thread(void *arg) |
3932 | { |
3933 | int status = 0; |
3934 | struct ocfs2_super *osb = arg; |
3935 | |
3936 | /* only quit once we've been asked to stop and there is no more |
3937 | * work available */ |
3938 | while (!(kthread_should_stop() && |
3939 | ocfs2_downconvert_thread_lists_empty(osb))) { |
3940 | |
3941 | wait_event_interruptible(osb->dc_event, |
3942 | ocfs2_downconvert_thread_should_wake(osb) || |
3943 | kthread_should_stop()); |
3944 | |
3945 | mlog(0, "downconvert_thread: awoken\n"); |
3946 | |
3947 | ocfs2_downconvert_thread_do_work(osb); |
3948 | } |
3949 | |
3950 | osb->dc_task = NULL; |
3951 | return status; |
3952 | } |
3953 | |
3954 | void ocfs2_wake_downconvert_thread(struct ocfs2_super *osb) |
3955 | { |
3956 | spin_lock(&osb->dc_task_lock); |
3957 | /* make sure the voting thread gets a swipe at whatever changes |
3958 | * the caller may have made to the voting state */ |
3959 | osb->dc_wake_sequence++; |
3960 | spin_unlock(&osb->dc_task_lock); |
3961 | wake_up(&osb->dc_event); |
3962 | } |
3963 |
Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master
Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9