Commit | Line | Data |
---|---|---|
ccd979bd MF |
1 | /* -*- mode: c; c-basic-offset: 8; -*- |
2 | * vim: noexpandtab sw=8 ts=8 sts=0: | |
3 | * | |
4 | * dlmglue.c | |
5 | * | |
6 | * Code which implements an OCFS2 specific interface to our DLM. | |
7 | * | |
8 | * Copyright (C) 2003, 2004 Oracle. All rights reserved. | |
9 | * | |
10 | * This program is free software; you can redistribute it and/or | |
11 | * modify it under the terms of the GNU General Public | |
12 | * License as published by the Free Software Foundation; either | |
13 | * version 2 of the License, or (at your option) any later version. | |
14 | * | |
15 | * This program is distributed in the hope that it will be useful, | |
16 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
18 | * General Public License for more details. | |
19 | * | |
20 | * You should have received a copy of the GNU General Public | |
21 | * License along with this program; if not, write to the | |
22 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
23 | * Boston, MA 021110-1307, USA. | |
24 | */ | |
25 | ||
26 | #include <linux/types.h> | |
27 | #include <linux/slab.h> | |
28 | #include <linux/highmem.h> | |
29 | #include <linux/mm.h> | |
ccd979bd MF |
30 | #include <linux/kthread.h> |
31 | #include <linux/pagemap.h> | |
32 | #include <linux/debugfs.h> | |
33 | #include <linux/seq_file.h> | |
8ddb7b00 | 34 | #include <linux/time.h> |
9e33d69f | 35 | #include <linux/quotaops.h> |
ccd979bd | 36 | |
ccd979bd MF |
37 | #define MLOG_MASK_PREFIX ML_DLM_GLUE |
38 | #include <cluster/masklog.h> | |
39 | ||
40 | #include "ocfs2.h" | |
d24fbcda | 41 | #include "ocfs2_lockingver.h" |
ccd979bd MF |
42 | |
43 | #include "alloc.h" | |
d680efe9 | 44 | #include "dcache.h" |
ccd979bd MF |
45 | #include "dlmglue.h" |
46 | #include "extent_map.h" | |
7f1a37e3 | 47 | #include "file.h" |
ccd979bd MF |
48 | #include "heartbeat.h" |
49 | #include "inode.h" | |
50 | #include "journal.h" | |
24ef1815 | 51 | #include "stackglue.h" |
ccd979bd MF |
52 | #include "slot_map.h" |
53 | #include "super.h" | |
54 | #include "uptodate.h" | |
9e33d69f | 55 | #include "quota.h" |
ccd979bd MF |
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; | |
8ddb7b00 SM |
65 | #ifdef CONFIG_OCFS2_FS_STATS |
66 | unsigned long long mw_lock_start; | |
67 | #endif | |
ccd979bd MF |
68 | }; |
69 | ||
54a7e755 MF |
70 | static struct ocfs2_super *ocfs2_get_dentry_osb(struct ocfs2_lock_res *lockres); |
71 | static struct ocfs2_super *ocfs2_get_inode_osb(struct ocfs2_lock_res *lockres); | |
cf8e06f1 | 72 | static struct ocfs2_super *ocfs2_get_file_osb(struct ocfs2_lock_res *lockres); |
9e33d69f | 73 | static struct ocfs2_super *ocfs2_get_qinfo_osb(struct ocfs2_lock_res *lockres); |
ccd979bd | 74 | |
d680efe9 | 75 | /* |
cc567d89 | 76 | * Return value from ->downconvert_worker functions. |
d680efe9 | 77 | * |
b5e500e2 | 78 | * These control the precise actions of ocfs2_unblock_lock() |
d680efe9 MF |
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 | ||
810d5aeb MF |
95 | static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres, |
96 | int new_level); | |
97 | static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres); | |
98 | ||
cc567d89 MF |
99 | static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres, |
100 | int blocking); | |
101 | ||
cc567d89 MF |
102 | static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres, |
103 | int blocking); | |
d680efe9 MF |
104 | |
105 | static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb, | |
106 | struct ocfs2_lock_res *lockres); | |
ccd979bd | 107 | |
9e33d69f | 108 | static void ocfs2_set_qinfo_lvb(struct ocfs2_lock_res *lockres); |
6cb129f5 AB |
109 | |
110 | #define mlog_meta_lvb(__level, __lockres) ocfs2_dump_meta_lvb_info(__level, __PRETTY_FUNCTION__, __LINE__, __lockres) | |
111 | ||
112 | /* This aids in debugging situations where a bad LVB might be involved. */ | |
113 | static void ocfs2_dump_meta_lvb_info(u64 level, | |
114 | const char *function, | |
115 | unsigned int line, | |
116 | struct ocfs2_lock_res *lockres) | |
117 | { | |
8f2c9c1b JB |
118 | struct ocfs2_meta_lvb *lvb = |
119 | (struct ocfs2_meta_lvb *)ocfs2_dlm_lvb(&lockres->l_lksb); | |
6cb129f5 AB |
120 | |
121 | mlog(level, "LVB information for %s (called from %s:%u):\n", | |
122 | lockres->l_name, function, line); | |
123 | mlog(level, "version: %u, clusters: %u, generation: 0x%x\n", | |
124 | lvb->lvb_version, be32_to_cpu(lvb->lvb_iclusters), | |
125 | be32_to_cpu(lvb->lvb_igeneration)); | |
126 | mlog(level, "size: %llu, uid %u, gid %u, mode 0x%x\n", | |
127 | (unsigned long long)be64_to_cpu(lvb->lvb_isize), | |
128 | be32_to_cpu(lvb->lvb_iuid), be32_to_cpu(lvb->lvb_igid), | |
129 | be16_to_cpu(lvb->lvb_imode)); | |
130 | mlog(level, "nlink %u, atime_packed 0x%llx, ctime_packed 0x%llx, " | |
131 | "mtime_packed 0x%llx iattr 0x%x\n", be16_to_cpu(lvb->lvb_inlink), | |
132 | (long long)be64_to_cpu(lvb->lvb_iatime_packed), | |
133 | (long long)be64_to_cpu(lvb->lvb_ictime_packed), | |
134 | (long long)be64_to_cpu(lvb->lvb_imtime_packed), | |
135 | be32_to_cpu(lvb->lvb_iattr)); | |
136 | } | |
137 | ||
138 | ||
f625c979 MF |
139 | /* |
140 | * OCFS2 Lock Resource Operations | |
141 | * | |
142 | * These fine tune the behavior of the generic dlmglue locking infrastructure. | |
0d5dc6c2 MF |
143 | * |
144 | * The most basic of lock types can point ->l_priv to their respective | |
145 | * struct ocfs2_super and allow the default actions to manage things. | |
146 | * | |
147 | * Right now, each lock type also needs to implement an init function, | |
148 | * and trivial lock/unlock wrappers. ocfs2_simple_drop_lockres() | |
149 | * should be called when the lock is no longer needed (i.e., object | |
150 | * destruction time). | |
f625c979 | 151 | */ |
ccd979bd | 152 | struct ocfs2_lock_res_ops { |
54a7e755 MF |
153 | /* |
154 | * Translate an ocfs2_lock_res * into an ocfs2_super *. Define | |
155 | * this callback if ->l_priv is not an ocfs2_super pointer | |
156 | */ | |
157 | struct ocfs2_super * (*get_osb)(struct ocfs2_lock_res *); | |
b5e500e2 | 158 | |
0d5dc6c2 | 159 | /* |
34d024f8 MF |
160 | * Optionally called in the downconvert thread after a |
161 | * successful downconvert. The lockres will not be referenced | |
162 | * after this callback is called, so it is safe to free | |
163 | * memory, etc. | |
0d5dc6c2 MF |
164 | * |
165 | * The exact semantics of when this is called are controlled | |
166 | * by ->downconvert_worker() | |
167 | */ | |
d680efe9 | 168 | void (*post_unlock)(struct ocfs2_super *, struct ocfs2_lock_res *); |
f625c979 | 169 | |
16d5b956 MF |
170 | /* |
171 | * Allow a lock type to add checks to determine whether it is | |
172 | * safe to downconvert a lock. Return 0 to re-queue the | |
173 | * downconvert at a later time, nonzero to continue. | |
174 | * | |
175 | * For most locks, the default checks that there are no | |
176 | * incompatible holders are sufficient. | |
177 | * | |
178 | * Called with the lockres spinlock held. | |
179 | */ | |
180 | int (*check_downconvert)(struct ocfs2_lock_res *, int); | |
181 | ||
5ef0d4ea MF |
182 | /* |
183 | * Allows a lock type to populate the lock value block. This | |
184 | * is called on downconvert, and when we drop a lock. | |
185 | * | |
186 | * Locks that want to use this should set LOCK_TYPE_USES_LVB | |
187 | * in the flags field. | |
188 | * | |
189 | * Called with the lockres spinlock held. | |
190 | */ | |
191 | void (*set_lvb)(struct ocfs2_lock_res *); | |
192 | ||
cc567d89 MF |
193 | /* |
194 | * Called from the downconvert thread when it is determined | |
195 | * that a lock will be downconverted. This is called without | |
196 | * any locks held so the function can do work that might | |
197 | * schedule (syncing out data, etc). | |
198 | * | |
199 | * This should return any one of the ocfs2_unblock_action | |
200 | * values, depending on what it wants the thread to do. | |
201 | */ | |
202 | int (*downconvert_worker)(struct ocfs2_lock_res *, int); | |
203 | ||
f625c979 MF |
204 | /* |
205 | * LOCK_TYPE_* flags which describe the specific requirements | |
206 | * of a lock type. Descriptions of each individual flag follow. | |
207 | */ | |
208 | int flags; | |
ccd979bd MF |
209 | }; |
210 | ||
f625c979 MF |
211 | /* |
212 | * Some locks want to "refresh" potentially stale data when a | |
213 | * meaningful (PRMODE or EXMODE) lock level is first obtained. If this | |
214 | * flag is set, the OCFS2_LOCK_NEEDS_REFRESH flag will be set on the | |
215 | * individual lockres l_flags member from the ast function. It is | |
216 | * expected that the locking wrapper will clear the | |
217 | * OCFS2_LOCK_NEEDS_REFRESH flag when done. | |
218 | */ | |
219 | #define LOCK_TYPE_REQUIRES_REFRESH 0x1 | |
220 | ||
b80fc012 | 221 | /* |
5ef0d4ea MF |
222 | * Indicate that a lock type makes use of the lock value block. The |
223 | * ->set_lvb lock type callback must be defined. | |
b80fc012 MF |
224 | */ |
225 | #define LOCK_TYPE_USES_LVB 0x2 | |
226 | ||
ccd979bd | 227 | static struct ocfs2_lock_res_ops ocfs2_inode_rw_lops = { |
54a7e755 | 228 | .get_osb = ocfs2_get_inode_osb, |
f625c979 | 229 | .flags = 0, |
ccd979bd MF |
230 | }; |
231 | ||
e63aecb6 | 232 | static struct ocfs2_lock_res_ops ocfs2_inode_inode_lops = { |
54a7e755 | 233 | .get_osb = ocfs2_get_inode_osb, |
810d5aeb MF |
234 | .check_downconvert = ocfs2_check_meta_downconvert, |
235 | .set_lvb = ocfs2_set_meta_lvb, | |
f1f54068 | 236 | .downconvert_worker = ocfs2_data_convert_worker, |
b80fc012 | 237 | .flags = LOCK_TYPE_REQUIRES_REFRESH|LOCK_TYPE_USES_LVB, |
ccd979bd MF |
238 | }; |
239 | ||
ccd979bd | 240 | static struct ocfs2_lock_res_ops ocfs2_super_lops = { |
f625c979 | 241 | .flags = LOCK_TYPE_REQUIRES_REFRESH, |
ccd979bd MF |
242 | }; |
243 | ||
244 | static struct ocfs2_lock_res_ops ocfs2_rename_lops = { | |
f625c979 | 245 | .flags = 0, |
ccd979bd MF |
246 | }; |
247 | ||
d680efe9 | 248 | static struct ocfs2_lock_res_ops ocfs2_dentry_lops = { |
54a7e755 | 249 | .get_osb = ocfs2_get_dentry_osb, |
d680efe9 | 250 | .post_unlock = ocfs2_dentry_post_unlock, |
cc567d89 | 251 | .downconvert_worker = ocfs2_dentry_convert_worker, |
f625c979 | 252 | .flags = 0, |
d680efe9 MF |
253 | }; |
254 | ||
50008630 TY |
255 | static struct ocfs2_lock_res_ops ocfs2_inode_open_lops = { |
256 | .get_osb = ocfs2_get_inode_osb, | |
257 | .flags = 0, | |
258 | }; | |
259 | ||
cf8e06f1 MF |
260 | static struct ocfs2_lock_res_ops ocfs2_flock_lops = { |
261 | .get_osb = ocfs2_get_file_osb, | |
262 | .flags = 0, | |
263 | }; | |
264 | ||
9e33d69f JK |
265 | static struct ocfs2_lock_res_ops ocfs2_qinfo_lops = { |
266 | .set_lvb = ocfs2_set_qinfo_lvb, | |
267 | .get_osb = ocfs2_get_qinfo_osb, | |
268 | .flags = LOCK_TYPE_REQUIRES_REFRESH | LOCK_TYPE_USES_LVB, | |
269 | }; | |
270 | ||
ccd979bd MF |
271 | static inline int ocfs2_is_inode_lock(struct ocfs2_lock_res *lockres) |
272 | { | |
273 | return lockres->l_type == OCFS2_LOCK_TYPE_META || | |
50008630 TY |
274 | lockres->l_type == OCFS2_LOCK_TYPE_RW || |
275 | lockres->l_type == OCFS2_LOCK_TYPE_OPEN; | |
ccd979bd MF |
276 | } |
277 | ||
ccd979bd MF |
278 | static inline struct inode *ocfs2_lock_res_inode(struct ocfs2_lock_res *lockres) |
279 | { | |
280 | BUG_ON(!ocfs2_is_inode_lock(lockres)); | |
281 | ||
282 | return (struct inode *) lockres->l_priv; | |
283 | } | |
284 | ||
d680efe9 MF |
285 | static inline struct ocfs2_dentry_lock *ocfs2_lock_res_dl(struct ocfs2_lock_res *lockres) |
286 | { | |
287 | BUG_ON(lockres->l_type != OCFS2_LOCK_TYPE_DENTRY); | |
288 | ||
289 | return (struct ocfs2_dentry_lock *)lockres->l_priv; | |
290 | } | |
291 | ||
9e33d69f JK |
292 | static inline struct ocfs2_mem_dqinfo *ocfs2_lock_res_qinfo(struct ocfs2_lock_res *lockres) |
293 | { | |
294 | BUG_ON(lockres->l_type != OCFS2_LOCK_TYPE_QINFO); | |
295 | ||
296 | return (struct ocfs2_mem_dqinfo *)lockres->l_priv; | |
297 | } | |
298 | ||
54a7e755 MF |
299 | static inline struct ocfs2_super *ocfs2_get_lockres_osb(struct ocfs2_lock_res *lockres) |
300 | { | |
301 | if (lockres->l_ops->get_osb) | |
302 | return lockres->l_ops->get_osb(lockres); | |
303 | ||
304 | return (struct ocfs2_super *)lockres->l_priv; | |
305 | } | |
306 | ||
ccd979bd MF |
307 | static int ocfs2_lock_create(struct ocfs2_super *osb, |
308 | struct ocfs2_lock_res *lockres, | |
309 | int level, | |
bd3e7610 | 310 | u32 dlm_flags); |
ccd979bd MF |
311 | static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres, |
312 | int wanted); | |
313 | static void ocfs2_cluster_unlock(struct ocfs2_super *osb, | |
314 | struct ocfs2_lock_res *lockres, | |
315 | int level); | |
316 | static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres); | |
317 | static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres); | |
318 | static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres); | |
319 | static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres, int level); | |
320 | static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb, | |
321 | struct ocfs2_lock_res *lockres); | |
322 | static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres, | |
323 | int convert); | |
7431cd7e JB |
324 | #define ocfs2_log_dlm_error(_func, _err, _lockres) do { \ |
325 | mlog(ML_ERROR, "DLM error %d while calling %s on resource %s\n", \ | |
326 | _err, _func, _lockres->l_name); \ | |
ccd979bd | 327 | } while (0) |
34d024f8 MF |
328 | static int ocfs2_downconvert_thread(void *arg); |
329 | static void ocfs2_downconvert_on_unlock(struct ocfs2_super *osb, | |
330 | struct ocfs2_lock_res *lockres); | |
e63aecb6 | 331 | static int ocfs2_inode_lock_update(struct inode *inode, |
ccd979bd MF |
332 | struct buffer_head **bh); |
333 | static void ocfs2_drop_osb_locks(struct ocfs2_super *osb); | |
334 | static inline int ocfs2_highest_compat_lock_level(int level); | |
de551246 JB |
335 | static unsigned int ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres, |
336 | int new_level); | |
cf8e06f1 MF |
337 | static int ocfs2_downconvert_lock(struct ocfs2_super *osb, |
338 | struct ocfs2_lock_res *lockres, | |
339 | int new_level, | |
de551246 JB |
340 | int lvb, |
341 | unsigned int generation); | |
cf8e06f1 MF |
342 | static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb, |
343 | struct ocfs2_lock_res *lockres); | |
344 | static int ocfs2_cancel_convert(struct ocfs2_super *osb, | |
345 | struct ocfs2_lock_res *lockres); | |
346 | ||
ccd979bd | 347 | |
ccd979bd MF |
348 | static void ocfs2_build_lock_name(enum ocfs2_lock_type type, |
349 | u64 blkno, | |
350 | u32 generation, | |
351 | char *name) | |
352 | { | |
353 | int len; | |
354 | ||
355 | mlog_entry_void(); | |
356 | ||
357 | BUG_ON(type >= OCFS2_NUM_LOCK_TYPES); | |
358 | ||
b0697053 MF |
359 | len = snprintf(name, OCFS2_LOCK_ID_MAX_LEN, "%c%s%016llx%08x", |
360 | ocfs2_lock_type_char(type), OCFS2_LOCK_ID_PAD, | |
361 | (long long)blkno, generation); | |
ccd979bd MF |
362 | |
363 | BUG_ON(len != (OCFS2_LOCK_ID_MAX_LEN - 1)); | |
364 | ||
365 | mlog(0, "built lock resource with name: %s\n", name); | |
366 | ||
367 | mlog_exit_void(); | |
368 | } | |
369 | ||
34af946a | 370 | static DEFINE_SPINLOCK(ocfs2_dlm_tracking_lock); |
ccd979bd MF |
371 | |
372 | static void ocfs2_add_lockres_tracking(struct ocfs2_lock_res *res, | |
373 | struct ocfs2_dlm_debug *dlm_debug) | |
374 | { | |
375 | mlog(0, "Add tracking for lockres %s\n", res->l_name); | |
376 | ||
377 | spin_lock(&ocfs2_dlm_tracking_lock); | |
378 | list_add(&res->l_debug_list, &dlm_debug->d_lockres_tracking); | |
379 | spin_unlock(&ocfs2_dlm_tracking_lock); | |
380 | } | |
381 | ||
382 | static void ocfs2_remove_lockres_tracking(struct ocfs2_lock_res *res) | |
383 | { | |
384 | spin_lock(&ocfs2_dlm_tracking_lock); | |
385 | if (!list_empty(&res->l_debug_list)) | |
386 | list_del_init(&res->l_debug_list); | |
387 | spin_unlock(&ocfs2_dlm_tracking_lock); | |
388 | } | |
389 | ||
8ddb7b00 SM |
390 | #ifdef CONFIG_OCFS2_FS_STATS |
391 | static void ocfs2_init_lock_stats(struct ocfs2_lock_res *res) | |
392 | { | |
393 | res->l_lock_num_prmode = 0; | |
394 | res->l_lock_num_prmode_failed = 0; | |
395 | res->l_lock_total_prmode = 0; | |
396 | res->l_lock_max_prmode = 0; | |
397 | res->l_lock_num_exmode = 0; | |
398 | res->l_lock_num_exmode_failed = 0; | |
399 | res->l_lock_total_exmode = 0; | |
400 | res->l_lock_max_exmode = 0; | |
401 | res->l_lock_refresh = 0; | |
402 | } | |
403 | ||
404 | static void ocfs2_update_lock_stats(struct ocfs2_lock_res *res, int level, | |
405 | struct ocfs2_mask_waiter *mw, int ret) | |
406 | { | |
407 | unsigned long long *num, *sum; | |
408 | unsigned int *max, *failed; | |
409 | struct timespec ts = current_kernel_time(); | |
410 | unsigned long long time = timespec_to_ns(&ts) - mw->mw_lock_start; | |
411 | ||
412 | if (level == LKM_PRMODE) { | |
413 | num = &res->l_lock_num_prmode; | |
414 | sum = &res->l_lock_total_prmode; | |
415 | max = &res->l_lock_max_prmode; | |
416 | failed = &res->l_lock_num_prmode_failed; | |
417 | } else if (level == LKM_EXMODE) { | |
418 | num = &res->l_lock_num_exmode; | |
419 | sum = &res->l_lock_total_exmode; | |
420 | max = &res->l_lock_max_exmode; | |
421 | failed = &res->l_lock_num_exmode_failed; | |
422 | } else | |
423 | return; | |
424 | ||
425 | (*num)++; | |
426 | (*sum) += time; | |
427 | if (time > *max) | |
428 | *max = time; | |
429 | if (ret) | |
430 | (*failed)++; | |
431 | } | |
432 | ||
433 | static inline void ocfs2_track_lock_refresh(struct ocfs2_lock_res *lockres) | |
434 | { | |
435 | lockres->l_lock_refresh++; | |
436 | } | |
437 | ||
438 | static inline void ocfs2_init_start_time(struct ocfs2_mask_waiter *mw) | |
439 | { | |
440 | struct timespec ts = current_kernel_time(); | |
441 | mw->mw_lock_start = timespec_to_ns(&ts); | |
442 | } | |
443 | #else | |
444 | static inline void ocfs2_init_lock_stats(struct ocfs2_lock_res *res) | |
445 | { | |
446 | } | |
447 | static inline void ocfs2_update_lock_stats(struct ocfs2_lock_res *res, | |
448 | int level, struct ocfs2_mask_waiter *mw, int ret) | |
449 | { | |
450 | } | |
451 | static inline void ocfs2_track_lock_refresh(struct ocfs2_lock_res *lockres) | |
452 | { | |
453 | } | |
454 | static inline void ocfs2_init_start_time(struct ocfs2_mask_waiter *mw) | |
455 | { | |
456 | } | |
457 | #endif | |
458 | ||
ccd979bd MF |
459 | static void ocfs2_lock_res_init_common(struct ocfs2_super *osb, |
460 | struct ocfs2_lock_res *res, | |
461 | enum ocfs2_lock_type type, | |
ccd979bd MF |
462 | struct ocfs2_lock_res_ops *ops, |
463 | void *priv) | |
464 | { | |
ccd979bd MF |
465 | res->l_type = type; |
466 | res->l_ops = ops; | |
467 | res->l_priv = priv; | |
468 | ||
bd3e7610 JB |
469 | res->l_level = DLM_LOCK_IV; |
470 | res->l_requested = DLM_LOCK_IV; | |
471 | res->l_blocking = DLM_LOCK_IV; | |
ccd979bd MF |
472 | res->l_action = OCFS2_AST_INVALID; |
473 | res->l_unlock_action = OCFS2_UNLOCK_INVALID; | |
474 | ||
475 | res->l_flags = OCFS2_LOCK_INITIALIZED; | |
476 | ||
477 | ocfs2_add_lockres_tracking(res, osb->osb_dlm_debug); | |
8ddb7b00 SM |
478 | |
479 | ocfs2_init_lock_stats(res); | |
ccd979bd MF |
480 | } |
481 | ||
482 | void ocfs2_lock_res_init_once(struct ocfs2_lock_res *res) | |
483 | { | |
484 | /* This also clears out the lock status block */ | |
485 | memset(res, 0, sizeof(struct ocfs2_lock_res)); | |
486 | spin_lock_init(&res->l_lock); | |
487 | init_waitqueue_head(&res->l_event); | |
488 | INIT_LIST_HEAD(&res->l_blocked_list); | |
489 | INIT_LIST_HEAD(&res->l_mask_waiters); | |
490 | } | |
491 | ||
492 | void ocfs2_inode_lock_res_init(struct ocfs2_lock_res *res, | |
493 | enum ocfs2_lock_type type, | |
24c19ef4 | 494 | unsigned int generation, |
ccd979bd MF |
495 | struct inode *inode) |
496 | { | |
497 | struct ocfs2_lock_res_ops *ops; | |
498 | ||
499 | switch(type) { | |
500 | case OCFS2_LOCK_TYPE_RW: | |
501 | ops = &ocfs2_inode_rw_lops; | |
502 | break; | |
503 | case OCFS2_LOCK_TYPE_META: | |
e63aecb6 | 504 | ops = &ocfs2_inode_inode_lops; |
ccd979bd | 505 | break; |
50008630 TY |
506 | case OCFS2_LOCK_TYPE_OPEN: |
507 | ops = &ocfs2_inode_open_lops; | |
508 | break; | |
ccd979bd MF |
509 | default: |
510 | mlog_bug_on_msg(1, "type: %d\n", type); | |
511 | ops = NULL; /* thanks, gcc */ | |
512 | break; | |
513 | }; | |
514 | ||
d680efe9 | 515 | ocfs2_build_lock_name(type, OCFS2_I(inode)->ip_blkno, |
24c19ef4 | 516 | generation, res->l_name); |
d680efe9 MF |
517 | ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), res, type, ops, inode); |
518 | } | |
519 | ||
54a7e755 MF |
520 | static struct ocfs2_super *ocfs2_get_inode_osb(struct ocfs2_lock_res *lockres) |
521 | { | |
522 | struct inode *inode = ocfs2_lock_res_inode(lockres); | |
523 | ||
524 | return OCFS2_SB(inode->i_sb); | |
525 | } | |
526 | ||
9e33d69f JK |
527 | static struct ocfs2_super *ocfs2_get_qinfo_osb(struct ocfs2_lock_res *lockres) |
528 | { | |
529 | struct ocfs2_mem_dqinfo *info = lockres->l_priv; | |
530 | ||
531 | return OCFS2_SB(info->dqi_gi.dqi_sb); | |
532 | } | |
533 | ||
cf8e06f1 MF |
534 | static struct ocfs2_super *ocfs2_get_file_osb(struct ocfs2_lock_res *lockres) |
535 | { | |
536 | struct ocfs2_file_private *fp = lockres->l_priv; | |
537 | ||
538 | return OCFS2_SB(fp->fp_file->f_mapping->host->i_sb); | |
539 | } | |
540 | ||
d680efe9 MF |
541 | static __u64 ocfs2_get_dentry_lock_ino(struct ocfs2_lock_res *lockres) |
542 | { | |
543 | __be64 inode_blkno_be; | |
544 | ||
545 | memcpy(&inode_blkno_be, &lockres->l_name[OCFS2_DENTRY_LOCK_INO_START], | |
546 | sizeof(__be64)); | |
547 | ||
548 | return be64_to_cpu(inode_blkno_be); | |
549 | } | |
550 | ||
54a7e755 MF |
551 | static struct ocfs2_super *ocfs2_get_dentry_osb(struct ocfs2_lock_res *lockres) |
552 | { | |
553 | struct ocfs2_dentry_lock *dl = lockres->l_priv; | |
554 | ||
555 | return OCFS2_SB(dl->dl_inode->i_sb); | |
556 | } | |
557 | ||
d680efe9 MF |
558 | void ocfs2_dentry_lock_res_init(struct ocfs2_dentry_lock *dl, |
559 | u64 parent, struct inode *inode) | |
560 | { | |
561 | int len; | |
562 | u64 inode_blkno = OCFS2_I(inode)->ip_blkno; | |
563 | __be64 inode_blkno_be = cpu_to_be64(inode_blkno); | |
564 | struct ocfs2_lock_res *lockres = &dl->dl_lockres; | |
565 | ||
566 | ocfs2_lock_res_init_once(lockres); | |
567 | ||
568 | /* | |
569 | * Unfortunately, the standard lock naming scheme won't work | |
570 | * here because we have two 16 byte values to use. Instead, | |
571 | * we'll stuff the inode number as a binary value. We still | |
572 | * want error prints to show something without garbling the | |
573 | * display, so drop a null byte in there before the inode | |
574 | * number. A future version of OCFS2 will likely use all | |
575 | * binary lock names. The stringified names have been a | |
576 | * tremendous aid in debugging, but now that the debugfs | |
577 | * interface exists, we can mangle things there if need be. | |
578 | * | |
579 | * NOTE: We also drop the standard "pad" value (the total lock | |
580 | * name size stays the same though - the last part is all | |
581 | * zeros due to the memset in ocfs2_lock_res_init_once() | |
582 | */ | |
583 | len = snprintf(lockres->l_name, OCFS2_DENTRY_LOCK_INO_START, | |
584 | "%c%016llx", | |
585 | ocfs2_lock_type_char(OCFS2_LOCK_TYPE_DENTRY), | |
586 | (long long)parent); | |
587 | ||
588 | BUG_ON(len != (OCFS2_DENTRY_LOCK_INO_START - 1)); | |
589 | ||
590 | memcpy(&lockres->l_name[OCFS2_DENTRY_LOCK_INO_START], &inode_blkno_be, | |
591 | sizeof(__be64)); | |
592 | ||
593 | ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), lockres, | |
594 | OCFS2_LOCK_TYPE_DENTRY, &ocfs2_dentry_lops, | |
595 | dl); | |
ccd979bd MF |
596 | } |
597 | ||
598 | static void ocfs2_super_lock_res_init(struct ocfs2_lock_res *res, | |
599 | struct ocfs2_super *osb) | |
600 | { | |
601 | /* Superblock lockres doesn't come from a slab so we call init | |
602 | * once on it manually. */ | |
603 | ocfs2_lock_res_init_once(res); | |
d680efe9 MF |
604 | ocfs2_build_lock_name(OCFS2_LOCK_TYPE_SUPER, OCFS2_SUPER_BLOCK_BLKNO, |
605 | 0, res->l_name); | |
ccd979bd | 606 | ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_SUPER, |
ccd979bd MF |
607 | &ocfs2_super_lops, osb); |
608 | } | |
609 | ||
610 | static void ocfs2_rename_lock_res_init(struct ocfs2_lock_res *res, | |
611 | struct ocfs2_super *osb) | |
612 | { | |
613 | /* Rename lockres doesn't come from a slab so we call init | |
614 | * once on it manually. */ | |
615 | ocfs2_lock_res_init_once(res); | |
d680efe9 MF |
616 | ocfs2_build_lock_name(OCFS2_LOCK_TYPE_RENAME, 0, 0, res->l_name); |
617 | ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_RENAME, | |
ccd979bd MF |
618 | &ocfs2_rename_lops, osb); |
619 | } | |
620 | ||
cf8e06f1 MF |
621 | void ocfs2_file_lock_res_init(struct ocfs2_lock_res *lockres, |
622 | struct ocfs2_file_private *fp) | |
623 | { | |
624 | struct inode *inode = fp->fp_file->f_mapping->host; | |
625 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
626 | ||
627 | ocfs2_lock_res_init_once(lockres); | |
628 | ocfs2_build_lock_name(OCFS2_LOCK_TYPE_FLOCK, oi->ip_blkno, | |
629 | inode->i_generation, lockres->l_name); | |
630 | ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), lockres, | |
631 | OCFS2_LOCK_TYPE_FLOCK, &ocfs2_flock_lops, | |
632 | fp); | |
633 | lockres->l_flags |= OCFS2_LOCK_NOCACHE; | |
634 | } | |
635 | ||
9e33d69f JK |
636 | void ocfs2_qinfo_lock_res_init(struct ocfs2_lock_res *lockres, |
637 | struct ocfs2_mem_dqinfo *info) | |
638 | { | |
639 | ocfs2_lock_res_init_once(lockres); | |
640 | ocfs2_build_lock_name(OCFS2_LOCK_TYPE_QINFO, info->dqi_gi.dqi_type, | |
641 | 0, lockres->l_name); | |
642 | ocfs2_lock_res_init_common(OCFS2_SB(info->dqi_gi.dqi_sb), lockres, | |
643 | OCFS2_LOCK_TYPE_QINFO, &ocfs2_qinfo_lops, | |
644 | info); | |
645 | } | |
646 | ||
ccd979bd MF |
647 | void ocfs2_lock_res_free(struct ocfs2_lock_res *res) |
648 | { | |
649 | mlog_entry_void(); | |
650 | ||
651 | if (!(res->l_flags & OCFS2_LOCK_INITIALIZED)) | |
652 | return; | |
653 | ||
654 | ocfs2_remove_lockres_tracking(res); | |
655 | ||
656 | mlog_bug_on_msg(!list_empty(&res->l_blocked_list), | |
657 | "Lockres %s is on the blocked list\n", | |
658 | res->l_name); | |
659 | mlog_bug_on_msg(!list_empty(&res->l_mask_waiters), | |
660 | "Lockres %s has mask waiters pending\n", | |
661 | res->l_name); | |
662 | mlog_bug_on_msg(spin_is_locked(&res->l_lock), | |
663 | "Lockres %s is locked\n", | |
664 | res->l_name); | |
665 | mlog_bug_on_msg(res->l_ro_holders, | |
666 | "Lockres %s has %u ro holders\n", | |
667 | res->l_name, res->l_ro_holders); | |
668 | mlog_bug_on_msg(res->l_ex_holders, | |
669 | "Lockres %s has %u ex holders\n", | |
670 | res->l_name, res->l_ex_holders); | |
671 | ||
672 | /* Need to clear out the lock status block for the dlm */ | |
673 | memset(&res->l_lksb, 0, sizeof(res->l_lksb)); | |
674 | ||
675 | res->l_flags = 0UL; | |
676 | mlog_exit_void(); | |
677 | } | |
678 | ||
679 | static inline void ocfs2_inc_holders(struct ocfs2_lock_res *lockres, | |
680 | int level) | |
681 | { | |
682 | mlog_entry_void(); | |
683 | ||
684 | BUG_ON(!lockres); | |
685 | ||
686 | switch(level) { | |
bd3e7610 | 687 | case DLM_LOCK_EX: |
ccd979bd MF |
688 | lockres->l_ex_holders++; |
689 | break; | |
bd3e7610 | 690 | case DLM_LOCK_PR: |
ccd979bd MF |
691 | lockres->l_ro_holders++; |
692 | break; | |
693 | default: | |
694 | BUG(); | |
695 | } | |
696 | ||
697 | mlog_exit_void(); | |
698 | } | |
699 | ||
700 | static inline void ocfs2_dec_holders(struct ocfs2_lock_res *lockres, | |
701 | int level) | |
702 | { | |
703 | mlog_entry_void(); | |
704 | ||
705 | BUG_ON(!lockres); | |
706 | ||
707 | switch(level) { | |
bd3e7610 | 708 | case DLM_LOCK_EX: |
ccd979bd MF |
709 | BUG_ON(!lockres->l_ex_holders); |
710 | lockres->l_ex_holders--; | |
711 | break; | |
bd3e7610 | 712 | case DLM_LOCK_PR: |
ccd979bd MF |
713 | BUG_ON(!lockres->l_ro_holders); |
714 | lockres->l_ro_holders--; | |
715 | break; | |
716 | default: | |
717 | BUG(); | |
718 | } | |
719 | mlog_exit_void(); | |
720 | } | |
721 | ||
722 | /* WARNING: This function lives in a world where the only three lock | |
723 | * levels are EX, PR, and NL. It *will* have to be adjusted when more | |
724 | * lock types are added. */ | |
725 | static inline int ocfs2_highest_compat_lock_level(int level) | |
726 | { | |
bd3e7610 | 727 | int new_level = DLM_LOCK_EX; |
ccd979bd | 728 | |
bd3e7610 JB |
729 | if (level == DLM_LOCK_EX) |
730 | new_level = DLM_LOCK_NL; | |
731 | else if (level == DLM_LOCK_PR) | |
732 | new_level = DLM_LOCK_PR; | |
ccd979bd MF |
733 | return new_level; |
734 | } | |
735 | ||
736 | static void lockres_set_flags(struct ocfs2_lock_res *lockres, | |
737 | unsigned long newflags) | |
738 | { | |
800deef3 | 739 | struct ocfs2_mask_waiter *mw, *tmp; |
ccd979bd MF |
740 | |
741 | assert_spin_locked(&lockres->l_lock); | |
742 | ||
743 | lockres->l_flags = newflags; | |
744 | ||
800deef3 | 745 | list_for_each_entry_safe(mw, tmp, &lockres->l_mask_waiters, mw_item) { |
ccd979bd MF |
746 | if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal) |
747 | continue; | |
748 | ||
749 | list_del_init(&mw->mw_item); | |
750 | mw->mw_status = 0; | |
751 | complete(&mw->mw_complete); | |
752 | } | |
753 | } | |
754 | static void lockres_or_flags(struct ocfs2_lock_res *lockres, unsigned long or) | |
755 | { | |
756 | lockres_set_flags(lockres, lockres->l_flags | or); | |
757 | } | |
758 | static void lockres_clear_flags(struct ocfs2_lock_res *lockres, | |
759 | unsigned long clear) | |
760 | { | |
761 | lockres_set_flags(lockres, lockres->l_flags & ~clear); | |
762 | } | |
763 | ||
764 | static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres) | |
765 | { | |
766 | mlog_entry_void(); | |
767 | ||
768 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY)); | |
769 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED)); | |
770 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED)); | |
bd3e7610 | 771 | BUG_ON(lockres->l_blocking <= DLM_LOCK_NL); |
ccd979bd MF |
772 | |
773 | lockres->l_level = lockres->l_requested; | |
774 | if (lockres->l_level <= | |
775 | ocfs2_highest_compat_lock_level(lockres->l_blocking)) { | |
bd3e7610 | 776 | lockres->l_blocking = DLM_LOCK_NL; |
ccd979bd MF |
777 | lockres_clear_flags(lockres, OCFS2_LOCK_BLOCKED); |
778 | } | |
779 | lockres_clear_flags(lockres, OCFS2_LOCK_BUSY); | |
780 | ||
781 | mlog_exit_void(); | |
782 | } | |
783 | ||
784 | static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres) | |
785 | { | |
786 | mlog_entry_void(); | |
787 | ||
788 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY)); | |
789 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED)); | |
790 | ||
791 | /* Convert from RO to EX doesn't really need anything as our | |
792 | * information is already up to data. Convert from NL to | |
793 | * *anything* however should mark ourselves as needing an | |
794 | * update */ | |
bd3e7610 | 795 | if (lockres->l_level == DLM_LOCK_NL && |
f625c979 | 796 | lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH) |
ccd979bd MF |
797 | lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH); |
798 | ||
799 | lockres->l_level = lockres->l_requested; | |
800 | lockres_clear_flags(lockres, OCFS2_LOCK_BUSY); | |
801 | ||
802 | mlog_exit_void(); | |
803 | } | |
804 | ||
805 | static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres) | |
806 | { | |
807 | mlog_entry_void(); | |
808 | ||
3cf0c507 | 809 | BUG_ON((!(lockres->l_flags & OCFS2_LOCK_BUSY))); |
ccd979bd MF |
810 | BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED); |
811 | ||
bd3e7610 | 812 | if (lockres->l_requested > DLM_LOCK_NL && |
f625c979 MF |
813 | !(lockres->l_flags & OCFS2_LOCK_LOCAL) && |
814 | lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH) | |
ccd979bd MF |
815 | lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH); |
816 | ||
817 | lockres->l_level = lockres->l_requested; | |
818 | lockres_or_flags(lockres, OCFS2_LOCK_ATTACHED); | |
819 | lockres_clear_flags(lockres, OCFS2_LOCK_BUSY); | |
820 | ||
821 | mlog_exit_void(); | |
822 | } | |
823 | ||
ccd979bd MF |
824 | static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres, |
825 | int level) | |
826 | { | |
827 | int needs_downconvert = 0; | |
828 | mlog_entry_void(); | |
829 | ||
830 | assert_spin_locked(&lockres->l_lock); | |
831 | ||
832 | lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED); | |
833 | ||
834 | if (level > lockres->l_blocking) { | |
835 | /* only schedule a downconvert if we haven't already scheduled | |
836 | * one that goes low enough to satisfy the level we're | |
837 | * blocking. this also catches the case where we get | |
838 | * duplicate BASTs */ | |
839 | if (ocfs2_highest_compat_lock_level(level) < | |
840 | ocfs2_highest_compat_lock_level(lockres->l_blocking)) | |
841 | needs_downconvert = 1; | |
842 | ||
843 | lockres->l_blocking = level; | |
844 | } | |
845 | ||
846 | mlog_exit(needs_downconvert); | |
847 | return needs_downconvert; | |
848 | } | |
849 | ||
de551246 JB |
850 | /* |
851 | * OCFS2_LOCK_PENDING and l_pending_gen. | |
852 | * | |
853 | * Why does OCFS2_LOCK_PENDING exist? To close a race between setting | |
854 | * OCFS2_LOCK_BUSY and calling ocfs2_dlm_lock(). See ocfs2_unblock_lock() | |
855 | * for more details on the race. | |
856 | * | |
857 | * OCFS2_LOCK_PENDING closes the race quite nicely. However, it introduces | |
858 | * a race on itself. In o2dlm, we can get the ast before ocfs2_dlm_lock() | |
859 | * returns. The ast clears OCFS2_LOCK_BUSY, and must therefore clear | |
860 | * OCFS2_LOCK_PENDING at the same time. When ocfs2_dlm_lock() returns, | |
861 | * the caller is going to try to clear PENDING again. If nothing else is | |
862 | * happening, __lockres_clear_pending() sees PENDING is unset and does | |
863 | * nothing. | |
864 | * | |
865 | * But what if another path (eg downconvert thread) has just started a | |
866 | * new locking action? The other path has re-set PENDING. Our path | |
867 | * cannot clear PENDING, because that will re-open the original race | |
868 | * window. | |
869 | * | |
870 | * [Example] | |
871 | * | |
872 | * ocfs2_meta_lock() | |
873 | * ocfs2_cluster_lock() | |
874 | * set BUSY | |
875 | * set PENDING | |
876 | * drop l_lock | |
877 | * ocfs2_dlm_lock() | |
878 | * ocfs2_locking_ast() ocfs2_downconvert_thread() | |
879 | * clear PENDING ocfs2_unblock_lock() | |
880 | * take_l_lock | |
881 | * !BUSY | |
882 | * ocfs2_prepare_downconvert() | |
883 | * set BUSY | |
884 | * set PENDING | |
885 | * drop l_lock | |
886 | * take l_lock | |
887 | * clear PENDING | |
888 | * drop l_lock | |
889 | * <window> | |
890 | * ocfs2_dlm_lock() | |
891 | * | |
892 | * So as you can see, we now have a window where l_lock is not held, | |
893 | * PENDING is not set, and ocfs2_dlm_lock() has not been called. | |
894 | * | |
895 | * The core problem is that ocfs2_cluster_lock() has cleared the PENDING | |
896 | * set by ocfs2_prepare_downconvert(). That wasn't nice. | |
897 | * | |
898 | * To solve this we introduce l_pending_gen. A call to | |
899 | * lockres_clear_pending() will only do so when it is passed a generation | |
900 | * number that matches the lockres. lockres_set_pending() will return the | |
901 | * current generation number. When ocfs2_cluster_lock() goes to clear | |
902 | * PENDING, it passes the generation it got from set_pending(). In our | |
903 | * example above, the generation numbers will *not* match. Thus, | |
904 | * ocfs2_cluster_lock() will not clear the PENDING set by | |
905 | * ocfs2_prepare_downconvert(). | |
906 | */ | |
907 | ||
908 | /* Unlocked version for ocfs2_locking_ast() */ | |
909 | static void __lockres_clear_pending(struct ocfs2_lock_res *lockres, | |
910 | unsigned int generation, | |
911 | struct ocfs2_super *osb) | |
912 | { | |
913 | assert_spin_locked(&lockres->l_lock); | |
914 | ||
915 | /* | |
916 | * The ast and locking functions can race us here. The winner | |
917 | * will clear pending, the loser will not. | |
918 | */ | |
919 | if (!(lockres->l_flags & OCFS2_LOCK_PENDING) || | |
920 | (lockres->l_pending_gen != generation)) | |
921 | return; | |
922 | ||
923 | lockres_clear_flags(lockres, OCFS2_LOCK_PENDING); | |
924 | lockres->l_pending_gen++; | |
925 | ||
926 | /* | |
927 | * The downconvert thread may have skipped us because we | |
928 | * were PENDING. Wake it up. | |
929 | */ | |
930 | if (lockres->l_flags & OCFS2_LOCK_BLOCKED) | |
931 | ocfs2_wake_downconvert_thread(osb); | |
932 | } | |
933 | ||
934 | /* Locked version for callers of ocfs2_dlm_lock() */ | |
935 | static void lockres_clear_pending(struct ocfs2_lock_res *lockres, | |
936 | unsigned int generation, | |
937 | struct ocfs2_super *osb) | |
938 | { | |
939 | unsigned long flags; | |
940 | ||
941 | spin_lock_irqsave(&lockres->l_lock, flags); | |
942 | __lockres_clear_pending(lockres, generation, osb); | |
943 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
944 | } | |
945 | ||
946 | static unsigned int lockres_set_pending(struct ocfs2_lock_res *lockres) | |
947 | { | |
948 | assert_spin_locked(&lockres->l_lock); | |
949 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY)); | |
950 | ||
951 | lockres_or_flags(lockres, OCFS2_LOCK_PENDING); | |
952 | ||
953 | return lockres->l_pending_gen; | |
954 | } | |
955 | ||
956 | ||
aa2623ad | 957 | static void ocfs2_blocking_ast(void *opaque, int level) |
ccd979bd | 958 | { |
aa2623ad MF |
959 | struct ocfs2_lock_res *lockres = opaque; |
960 | struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres); | |
ccd979bd MF |
961 | int needs_downconvert; |
962 | unsigned long flags; | |
963 | ||
bd3e7610 | 964 | BUG_ON(level <= DLM_LOCK_NL); |
ccd979bd | 965 | |
aa2623ad MF |
966 | mlog(0, "BAST fired for lockres %s, blocking %d, level %d type %s\n", |
967 | lockres->l_name, level, lockres->l_level, | |
968 | ocfs2_lock_type_string(lockres->l_type)); | |
969 | ||
cf8e06f1 MF |
970 | /* |
971 | * We can skip the bast for locks which don't enable caching - | |
972 | * they'll be dropped at the earliest possible time anyway. | |
973 | */ | |
974 | if (lockres->l_flags & OCFS2_LOCK_NOCACHE) | |
975 | return; | |
976 | ||
ccd979bd MF |
977 | spin_lock_irqsave(&lockres->l_lock, flags); |
978 | needs_downconvert = ocfs2_generic_handle_bast(lockres, level); | |
979 | if (needs_downconvert) | |
980 | ocfs2_schedule_blocked_lock(osb, lockres); | |
981 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
982 | ||
d680efe9 MF |
983 | wake_up(&lockres->l_event); |
984 | ||
34d024f8 | 985 | ocfs2_wake_downconvert_thread(osb); |
ccd979bd MF |
986 | } |
987 | ||
e92d57df | 988 | static void ocfs2_locking_ast(void *opaque) |
ccd979bd | 989 | { |
e92d57df | 990 | struct ocfs2_lock_res *lockres = opaque; |
de551246 | 991 | struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres); |
ccd979bd | 992 | unsigned long flags; |
1693a5c0 | 993 | int status; |
ccd979bd MF |
994 | |
995 | spin_lock_irqsave(&lockres->l_lock, flags); | |
996 | ||
1693a5c0 DT |
997 | status = ocfs2_dlm_lock_status(&lockres->l_lksb); |
998 | ||
999 | if (status == -EAGAIN) { | |
1000 | lockres_clear_flags(lockres, OCFS2_LOCK_BUSY); | |
1001 | goto out; | |
1002 | } | |
1003 | ||
1004 | if (status) { | |
8f2c9c1b | 1005 | mlog(ML_ERROR, "lockres %s: lksb status value of %d!\n", |
1693a5c0 | 1006 | lockres->l_name, status); |
ccd979bd MF |
1007 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
1008 | return; | |
1009 | } | |
1010 | ||
1011 | switch(lockres->l_action) { | |
1012 | case OCFS2_AST_ATTACH: | |
1013 | ocfs2_generic_handle_attach_action(lockres); | |
e92d57df | 1014 | lockres_clear_flags(lockres, OCFS2_LOCK_LOCAL); |
ccd979bd MF |
1015 | break; |
1016 | case OCFS2_AST_CONVERT: | |
1017 | ocfs2_generic_handle_convert_action(lockres); | |
1018 | break; | |
1019 | case OCFS2_AST_DOWNCONVERT: | |
1020 | ocfs2_generic_handle_downconvert_action(lockres); | |
1021 | break; | |
1022 | default: | |
e92d57df MF |
1023 | mlog(ML_ERROR, "lockres %s: ast fired with invalid action: %u " |
1024 | "lockres flags = 0x%lx, unlock action: %u\n", | |
1025 | lockres->l_name, lockres->l_action, lockres->l_flags, | |
1026 | lockres->l_unlock_action); | |
ccd979bd MF |
1027 | BUG(); |
1028 | } | |
1693a5c0 | 1029 | out: |
ccd979bd MF |
1030 | /* set it to something invalid so if we get called again we |
1031 | * can catch it. */ | |
1032 | lockres->l_action = OCFS2_AST_INVALID; | |
ccd979bd | 1033 | |
de551246 JB |
1034 | /* Did we try to cancel this lock? Clear that state */ |
1035 | if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) | |
1036 | lockres->l_unlock_action = OCFS2_UNLOCK_INVALID; | |
1037 | ||
1038 | /* | |
1039 | * We may have beaten the locking functions here. We certainly | |
1040 | * know that dlm_lock() has been called :-) | |
1041 | * Because we can't have two lock calls in flight at once, we | |
1042 | * can use lockres->l_pending_gen. | |
1043 | */ | |
1044 | __lockres_clear_pending(lockres, lockres->l_pending_gen, osb); | |
1045 | ||
ccd979bd | 1046 | wake_up(&lockres->l_event); |
d680efe9 | 1047 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
ccd979bd MF |
1048 | } |
1049 | ||
ccd979bd MF |
1050 | static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres, |
1051 | int convert) | |
1052 | { | |
1053 | unsigned long flags; | |
1054 | ||
1055 | mlog_entry_void(); | |
1056 | spin_lock_irqsave(&lockres->l_lock, flags); | |
1057 | lockres_clear_flags(lockres, OCFS2_LOCK_BUSY); | |
1058 | if (convert) | |
1059 | lockres->l_action = OCFS2_AST_INVALID; | |
1060 | else | |
1061 | lockres->l_unlock_action = OCFS2_UNLOCK_INVALID; | |
1062 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1063 | ||
1064 | wake_up(&lockres->l_event); | |
1065 | mlog_exit_void(); | |
1066 | } | |
1067 | ||
1068 | /* Note: If we detect another process working on the lock (i.e., | |
1069 | * OCFS2_LOCK_BUSY), we'll bail out returning 0. It's up to the caller | |
1070 | * to do the right thing in that case. | |
1071 | */ | |
1072 | static int ocfs2_lock_create(struct ocfs2_super *osb, | |
1073 | struct ocfs2_lock_res *lockres, | |
1074 | int level, | |
bd3e7610 | 1075 | u32 dlm_flags) |
ccd979bd MF |
1076 | { |
1077 | int ret = 0; | |
ccd979bd | 1078 | unsigned long flags; |
de551246 | 1079 | unsigned int gen; |
ccd979bd MF |
1080 | |
1081 | mlog_entry_void(); | |
1082 | ||
bd3e7610 | 1083 | mlog(0, "lock %s, level = %d, flags = %u\n", lockres->l_name, level, |
ccd979bd MF |
1084 | dlm_flags); |
1085 | ||
1086 | spin_lock_irqsave(&lockres->l_lock, flags); | |
1087 | if ((lockres->l_flags & OCFS2_LOCK_ATTACHED) || | |
1088 | (lockres->l_flags & OCFS2_LOCK_BUSY)) { | |
1089 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1090 | goto bail; | |
1091 | } | |
1092 | ||
1093 | lockres->l_action = OCFS2_AST_ATTACH; | |
1094 | lockres->l_requested = level; | |
1095 | lockres_or_flags(lockres, OCFS2_LOCK_BUSY); | |
de551246 | 1096 | gen = lockres_set_pending(lockres); |
ccd979bd MF |
1097 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
1098 | ||
4670c46d | 1099 | ret = ocfs2_dlm_lock(osb->cconn, |
7431cd7e JB |
1100 | level, |
1101 | &lockres->l_lksb, | |
1102 | dlm_flags, | |
1103 | lockres->l_name, | |
1104 | OCFS2_LOCK_ID_MAX_LEN - 1, | |
1105 | lockres); | |
de551246 | 1106 | lockres_clear_pending(lockres, gen, osb); |
7431cd7e JB |
1107 | if (ret) { |
1108 | ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres); | |
ccd979bd MF |
1109 | ocfs2_recover_from_dlm_error(lockres, 1); |
1110 | } | |
1111 | ||
7431cd7e | 1112 | mlog(0, "lock %s, return from ocfs2_dlm_lock\n", lockres->l_name); |
ccd979bd MF |
1113 | |
1114 | bail: | |
1115 | mlog_exit(ret); | |
1116 | return ret; | |
1117 | } | |
1118 | ||
1119 | static inline int ocfs2_check_wait_flag(struct ocfs2_lock_res *lockres, | |
1120 | int flag) | |
1121 | { | |
1122 | unsigned long flags; | |
1123 | int ret; | |
1124 | ||
1125 | spin_lock_irqsave(&lockres->l_lock, flags); | |
1126 | ret = lockres->l_flags & flag; | |
1127 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1128 | ||
1129 | return ret; | |
1130 | } | |
1131 | ||
1132 | static inline void ocfs2_wait_on_busy_lock(struct ocfs2_lock_res *lockres) | |
1133 | ||
1134 | { | |
1135 | wait_event(lockres->l_event, | |
1136 | !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_BUSY)); | |
1137 | } | |
1138 | ||
1139 | static inline void ocfs2_wait_on_refreshing_lock(struct ocfs2_lock_res *lockres) | |
1140 | ||
1141 | { | |
1142 | wait_event(lockres->l_event, | |
1143 | !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_REFRESHING)); | |
1144 | } | |
1145 | ||
1146 | /* predict what lock level we'll be dropping down to on behalf | |
1147 | * of another node, and return true if the currently wanted | |
1148 | * level will be compatible with it. */ | |
1149 | static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres, | |
1150 | int wanted) | |
1151 | { | |
1152 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED)); | |
1153 | ||
1154 | return wanted <= ocfs2_highest_compat_lock_level(lockres->l_blocking); | |
1155 | } | |
1156 | ||
1157 | static void ocfs2_init_mask_waiter(struct ocfs2_mask_waiter *mw) | |
1158 | { | |
1159 | INIT_LIST_HEAD(&mw->mw_item); | |
1160 | init_completion(&mw->mw_complete); | |
8ddb7b00 | 1161 | ocfs2_init_start_time(mw); |
ccd979bd MF |
1162 | } |
1163 | ||
1164 | static int ocfs2_wait_for_mask(struct ocfs2_mask_waiter *mw) | |
1165 | { | |
1166 | wait_for_completion(&mw->mw_complete); | |
1167 | /* Re-arm the completion in case we want to wait on it again */ | |
1168 | INIT_COMPLETION(mw->mw_complete); | |
1169 | return mw->mw_status; | |
1170 | } | |
1171 | ||
1172 | static void lockres_add_mask_waiter(struct ocfs2_lock_res *lockres, | |
1173 | struct ocfs2_mask_waiter *mw, | |
1174 | unsigned long mask, | |
1175 | unsigned long goal) | |
1176 | { | |
1177 | BUG_ON(!list_empty(&mw->mw_item)); | |
1178 | ||
1179 | assert_spin_locked(&lockres->l_lock); | |
1180 | ||
1181 | list_add_tail(&mw->mw_item, &lockres->l_mask_waiters); | |
1182 | mw->mw_mask = mask; | |
1183 | mw->mw_goal = goal; | |
1184 | } | |
1185 | ||
1186 | /* returns 0 if the mw that was removed was already satisfied, -EBUSY | |
1187 | * if the mask still hadn't reached its goal */ | |
1188 | static int lockres_remove_mask_waiter(struct ocfs2_lock_res *lockres, | |
1189 | struct ocfs2_mask_waiter *mw) | |
1190 | { | |
1191 | unsigned long flags; | |
1192 | int ret = 0; | |
1193 | ||
1194 | spin_lock_irqsave(&lockres->l_lock, flags); | |
1195 | if (!list_empty(&mw->mw_item)) { | |
1196 | if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal) | |
1197 | ret = -EBUSY; | |
1198 | ||
1199 | list_del_init(&mw->mw_item); | |
1200 | init_completion(&mw->mw_complete); | |
1201 | } | |
1202 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1203 | ||
1204 | return ret; | |
1205 | ||
1206 | } | |
1207 | ||
cf8e06f1 MF |
1208 | static int ocfs2_wait_for_mask_interruptible(struct ocfs2_mask_waiter *mw, |
1209 | struct ocfs2_lock_res *lockres) | |
1210 | { | |
1211 | int ret; | |
1212 | ||
1213 | ret = wait_for_completion_interruptible(&mw->mw_complete); | |
1214 | if (ret) | |
1215 | lockres_remove_mask_waiter(lockres, mw); | |
1216 | else | |
1217 | ret = mw->mw_status; | |
1218 | /* Re-arm the completion in case we want to wait on it again */ | |
1219 | INIT_COMPLETION(mw->mw_complete); | |
1220 | return ret; | |
1221 | } | |
1222 | ||
ccd979bd MF |
1223 | static int ocfs2_cluster_lock(struct ocfs2_super *osb, |
1224 | struct ocfs2_lock_res *lockres, | |
1225 | int level, | |
bd3e7610 | 1226 | u32 lkm_flags, |
ccd979bd MF |
1227 | int arg_flags) |
1228 | { | |
1229 | struct ocfs2_mask_waiter mw; | |
ccd979bd MF |
1230 | int wait, catch_signals = !(osb->s_mount_opt & OCFS2_MOUNT_NOINTR); |
1231 | int ret = 0; /* gcc doesn't realize wait = 1 guarantees ret is set */ | |
1232 | unsigned long flags; | |
de551246 | 1233 | unsigned int gen; |
1693a5c0 | 1234 | int noqueue_attempted = 0; |
ccd979bd MF |
1235 | |
1236 | mlog_entry_void(); | |
1237 | ||
1238 | ocfs2_init_mask_waiter(&mw); | |
1239 | ||
b80fc012 | 1240 | if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) |
bd3e7610 | 1241 | lkm_flags |= DLM_LKF_VALBLK; |
b80fc012 | 1242 | |
ccd979bd MF |
1243 | again: |
1244 | wait = 0; | |
1245 | ||
1246 | if (catch_signals && signal_pending(current)) { | |
1247 | ret = -ERESTARTSYS; | |
1248 | goto out; | |
1249 | } | |
1250 | ||
1251 | spin_lock_irqsave(&lockres->l_lock, flags); | |
1252 | ||
1253 | mlog_bug_on_msg(lockres->l_flags & OCFS2_LOCK_FREEING, | |
1254 | "Cluster lock called on freeing lockres %s! flags " | |
1255 | "0x%lx\n", lockres->l_name, lockres->l_flags); | |
1256 | ||
1257 | /* We only compare against the currently granted level | |
1258 | * here. If the lock is blocked waiting on a downconvert, | |
1259 | * we'll get caught below. */ | |
1260 | if (lockres->l_flags & OCFS2_LOCK_BUSY && | |
1261 | level > lockres->l_level) { | |
1262 | /* is someone sitting in dlm_lock? If so, wait on | |
1263 | * them. */ | |
1264 | lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0); | |
1265 | wait = 1; | |
1266 | goto unlock; | |
1267 | } | |
1268 | ||
ccd979bd MF |
1269 | if (lockres->l_flags & OCFS2_LOCK_BLOCKED && |
1270 | !ocfs2_may_continue_on_blocked_lock(lockres, level)) { | |
1271 | /* is the lock is currently blocked on behalf of | |
1272 | * another node */ | |
1273 | lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BLOCKED, 0); | |
1274 | wait = 1; | |
1275 | goto unlock; | |
1276 | } | |
1277 | ||
1278 | if (level > lockres->l_level) { | |
1693a5c0 DT |
1279 | if (noqueue_attempted > 0) { |
1280 | ret = -EAGAIN; | |
1281 | goto unlock; | |
1282 | } | |
1283 | if (lkm_flags & DLM_LKF_NOQUEUE) | |
1284 | noqueue_attempted = 1; | |
1285 | ||
ccd979bd MF |
1286 | if (lockres->l_action != OCFS2_AST_INVALID) |
1287 | mlog(ML_ERROR, "lockres %s has action %u pending\n", | |
1288 | lockres->l_name, lockres->l_action); | |
1289 | ||
019d1b22 MF |
1290 | if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) { |
1291 | lockres->l_action = OCFS2_AST_ATTACH; | |
bd3e7610 | 1292 | lkm_flags &= ~DLM_LKF_CONVERT; |
019d1b22 MF |
1293 | } else { |
1294 | lockres->l_action = OCFS2_AST_CONVERT; | |
bd3e7610 | 1295 | lkm_flags |= DLM_LKF_CONVERT; |
019d1b22 MF |
1296 | } |
1297 | ||
ccd979bd MF |
1298 | lockres->l_requested = level; |
1299 | lockres_or_flags(lockres, OCFS2_LOCK_BUSY); | |
de551246 | 1300 | gen = lockres_set_pending(lockres); |
ccd979bd MF |
1301 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
1302 | ||
bd3e7610 JB |
1303 | BUG_ON(level == DLM_LOCK_IV); |
1304 | BUG_ON(level == DLM_LOCK_NL); | |
ccd979bd MF |
1305 | |
1306 | mlog(0, "lock %s, convert from %d to level = %d\n", | |
1307 | lockres->l_name, lockres->l_level, level); | |
1308 | ||
1309 | /* call dlm_lock to upgrade lock now */ | |
4670c46d | 1310 | ret = ocfs2_dlm_lock(osb->cconn, |
7431cd7e JB |
1311 | level, |
1312 | &lockres->l_lksb, | |
1313 | lkm_flags, | |
1314 | lockres->l_name, | |
1315 | OCFS2_LOCK_ID_MAX_LEN - 1, | |
1316 | lockres); | |
de551246 | 1317 | lockres_clear_pending(lockres, gen, osb); |
7431cd7e JB |
1318 | if (ret) { |
1319 | if (!(lkm_flags & DLM_LKF_NOQUEUE) || | |
1320 | (ret != -EAGAIN)) { | |
24ef1815 | 1321 | ocfs2_log_dlm_error("ocfs2_dlm_lock", |
7431cd7e | 1322 | ret, lockres); |
ccd979bd MF |
1323 | } |
1324 | ocfs2_recover_from_dlm_error(lockres, 1); | |
1325 | goto out; | |
1326 | } | |
1327 | ||
24ef1815 | 1328 | mlog(0, "lock %s, successfull return from ocfs2_dlm_lock\n", |
ccd979bd MF |
1329 | lockres->l_name); |
1330 | ||
1331 | /* At this point we've gone inside the dlm and need to | |
1332 | * complete our work regardless. */ | |
1333 | catch_signals = 0; | |
1334 | ||
1335 | /* wait for busy to clear and carry on */ | |
1336 | goto again; | |
1337 | } | |
1338 | ||
1339 | /* Ok, if we get here then we're good to go. */ | |
1340 | ocfs2_inc_holders(lockres, level); | |
1341 | ||
1342 | ret = 0; | |
1343 | unlock: | |
1344 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1345 | out: | |
1346 | /* | |
1347 | * This is helping work around a lock inversion between the page lock | |
1348 | * and dlm locks. One path holds the page lock while calling aops | |
1349 | * which block acquiring dlm locks. The voting thread holds dlm | |
1350 | * locks while acquiring page locks while down converting data locks. | |
1351 | * This block is helping an aop path notice the inversion and back | |
1352 | * off to unlock its page lock before trying the dlm lock again. | |
1353 | */ | |
1354 | if (wait && arg_flags & OCFS2_LOCK_NONBLOCK && | |
1355 | mw.mw_mask & (OCFS2_LOCK_BUSY|OCFS2_LOCK_BLOCKED)) { | |
1356 | wait = 0; | |
1357 | if (lockres_remove_mask_waiter(lockres, &mw)) | |
1358 | ret = -EAGAIN; | |
1359 | else | |
1360 | goto again; | |
1361 | } | |
1362 | if (wait) { | |
1363 | ret = ocfs2_wait_for_mask(&mw); | |
1364 | if (ret == 0) | |
1365 | goto again; | |
1366 | mlog_errno(ret); | |
1367 | } | |
8ddb7b00 | 1368 | ocfs2_update_lock_stats(lockres, level, &mw, ret); |
ccd979bd MF |
1369 | |
1370 | mlog_exit(ret); | |
1371 | return ret; | |
1372 | } | |
1373 | ||
1374 | static void ocfs2_cluster_unlock(struct ocfs2_super *osb, | |
1375 | struct ocfs2_lock_res *lockres, | |
1376 | int level) | |
1377 | { | |
1378 | unsigned long flags; | |
1379 | ||
1380 | mlog_entry_void(); | |
1381 | spin_lock_irqsave(&lockres->l_lock, flags); | |
1382 | ocfs2_dec_holders(lockres, level); | |
34d024f8 | 1383 | ocfs2_downconvert_on_unlock(osb, lockres); |
ccd979bd MF |
1384 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
1385 | mlog_exit_void(); | |
1386 | } | |
1387 | ||
da66116e AB |
1388 | static int ocfs2_create_new_lock(struct ocfs2_super *osb, |
1389 | struct ocfs2_lock_res *lockres, | |
1390 | int ex, | |
1391 | int local) | |
ccd979bd | 1392 | { |
bd3e7610 | 1393 | int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR; |
ccd979bd | 1394 | unsigned long flags; |
bd3e7610 | 1395 | u32 lkm_flags = local ? DLM_LKF_LOCAL : 0; |
ccd979bd MF |
1396 | |
1397 | spin_lock_irqsave(&lockres->l_lock, flags); | |
1398 | BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED); | |
1399 | lockres_or_flags(lockres, OCFS2_LOCK_LOCAL); | |
1400 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1401 | ||
24c19ef4 | 1402 | return ocfs2_lock_create(osb, lockres, level, lkm_flags); |
ccd979bd MF |
1403 | } |
1404 | ||
1405 | /* Grants us an EX lock on the data and metadata resources, skipping | |
1406 | * the normal cluster directory lookup. Use this ONLY on newly created | |
1407 | * inodes which other nodes can't possibly see, and which haven't been | |
1408 | * hashed in the inode hash yet. This can give us a good performance | |
1409 | * increase as it'll skip the network broadcast normally associated | |
1410 | * with creating a new lock resource. */ | |
1411 | int ocfs2_create_new_inode_locks(struct inode *inode) | |
1412 | { | |
1413 | int ret; | |
d680efe9 | 1414 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
ccd979bd MF |
1415 | |
1416 | BUG_ON(!inode); | |
1417 | BUG_ON(!ocfs2_inode_is_new(inode)); | |
1418 | ||
1419 | mlog_entry_void(); | |
1420 | ||
b0697053 | 1421 | mlog(0, "Inode %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno); |
ccd979bd MF |
1422 | |
1423 | /* NOTE: That we don't increment any of the holder counts, nor | |
1424 | * do we add anything to a journal handle. Since this is | |
1425 | * supposed to be a new inode which the cluster doesn't know | |
1426 | * about yet, there is no need to. As far as the LVB handling | |
1427 | * is concerned, this is basically like acquiring an EX lock | |
1428 | * on a resource which has an invalid one -- we'll set it | |
1429 | * valid when we release the EX. */ | |
1430 | ||
24c19ef4 | 1431 | ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_rw_lockres, 1, 1); |
ccd979bd MF |
1432 | if (ret) { |
1433 | mlog_errno(ret); | |
1434 | goto bail; | |
1435 | } | |
1436 | ||
24c19ef4 | 1437 | /* |
bd3e7610 | 1438 | * We don't want to use DLM_LKF_LOCAL on a meta data lock as they |
24c19ef4 MF |
1439 | * don't use a generation in their lock names. |
1440 | */ | |
e63aecb6 | 1441 | ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_inode_lockres, 1, 0); |
ccd979bd MF |
1442 | if (ret) { |
1443 | mlog_errno(ret); | |
1444 | goto bail; | |
1445 | } | |
1446 | ||
50008630 TY |
1447 | ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_open_lockres, 0, 0); |
1448 | if (ret) { | |
1449 | mlog_errno(ret); | |
1450 | goto bail; | |
1451 | } | |
1452 | ||
ccd979bd MF |
1453 | bail: |
1454 | mlog_exit(ret); | |
1455 | return ret; | |
1456 | } | |
1457 | ||
1458 | int ocfs2_rw_lock(struct inode *inode, int write) | |
1459 | { | |
1460 | int status, level; | |
1461 | struct ocfs2_lock_res *lockres; | |
c271c5c2 | 1462 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
ccd979bd MF |
1463 | |
1464 | BUG_ON(!inode); | |
1465 | ||
1466 | mlog_entry_void(); | |
1467 | ||
b0697053 MF |
1468 | mlog(0, "inode %llu take %s RW lock\n", |
1469 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
ccd979bd MF |
1470 | write ? "EXMODE" : "PRMODE"); |
1471 | ||
c271c5c2 SM |
1472 | if (ocfs2_mount_local(osb)) |
1473 | return 0; | |
1474 | ||
ccd979bd MF |
1475 | lockres = &OCFS2_I(inode)->ip_rw_lockres; |
1476 | ||
bd3e7610 | 1477 | level = write ? DLM_LOCK_EX : DLM_LOCK_PR; |
ccd979bd MF |
1478 | |
1479 | status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres, level, 0, | |
1480 | 0); | |
1481 | if (status < 0) | |
1482 | mlog_errno(status); | |
1483 | ||
1484 | mlog_exit(status); | |
1485 | return status; | |
1486 | } | |
1487 | ||
1488 | void ocfs2_rw_unlock(struct inode *inode, int write) | |
1489 | { | |
bd3e7610 | 1490 | int level = write ? DLM_LOCK_EX : DLM_LOCK_PR; |
ccd979bd | 1491 | struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_rw_lockres; |
c271c5c2 | 1492 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
ccd979bd MF |
1493 | |
1494 | mlog_entry_void(); | |
1495 | ||
b0697053 MF |
1496 | mlog(0, "inode %llu drop %s RW lock\n", |
1497 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
ccd979bd MF |
1498 | write ? "EXMODE" : "PRMODE"); |
1499 | ||
c271c5c2 SM |
1500 | if (!ocfs2_mount_local(osb)) |
1501 | ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level); | |
ccd979bd MF |
1502 | |
1503 | mlog_exit_void(); | |
1504 | } | |
1505 | ||
50008630 TY |
1506 | /* |
1507 | * ocfs2_open_lock always get PR mode lock. | |
1508 | */ | |
1509 | int ocfs2_open_lock(struct inode *inode) | |
1510 | { | |
1511 | int status = 0; | |
1512 | struct ocfs2_lock_res *lockres; | |
1513 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1514 | ||
1515 | BUG_ON(!inode); | |
1516 | ||
1517 | mlog_entry_void(); | |
1518 | ||
1519 | mlog(0, "inode %llu take PRMODE open lock\n", | |
1520 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
1521 | ||
1522 | if (ocfs2_mount_local(osb)) | |
1523 | goto out; | |
1524 | ||
1525 | lockres = &OCFS2_I(inode)->ip_open_lockres; | |
1526 | ||
1527 | status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres, | |
bd3e7610 | 1528 | DLM_LOCK_PR, 0, 0); |
50008630 TY |
1529 | if (status < 0) |
1530 | mlog_errno(status); | |
1531 | ||
1532 | out: | |
1533 | mlog_exit(status); | |
1534 | return status; | |
1535 | } | |
1536 | ||
1537 | int ocfs2_try_open_lock(struct inode *inode, int write) | |
1538 | { | |
1539 | int status = 0, 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 try to take %s open lock\n", | |
1548 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1549 | write ? "EXMODE" : "PRMODE"); | |
1550 | ||
1551 | if (ocfs2_mount_local(osb)) | |
1552 | goto out; | |
1553 | ||
1554 | lockres = &OCFS2_I(inode)->ip_open_lockres; | |
1555 | ||
bd3e7610 | 1556 | level = write ? DLM_LOCK_EX : DLM_LOCK_PR; |
50008630 TY |
1557 | |
1558 | /* | |
1559 | * The file system may already holding a PRMODE/EXMODE open lock. | |
bd3e7610 | 1560 | * Since we pass DLM_LKF_NOQUEUE, the request won't block waiting on |
50008630 TY |
1561 | * other nodes and the -EAGAIN will indicate to the caller that |
1562 | * this inode is still in use. | |
1563 | */ | |
1564 | status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres, | |
bd3e7610 | 1565 | level, DLM_LKF_NOQUEUE, 0); |
50008630 TY |
1566 | |
1567 | out: | |
1568 | mlog_exit(status); | |
1569 | return status; | |
1570 | } | |
1571 | ||
1572 | /* | |
1573 | * ocfs2_open_unlock unlock PR and EX mode open locks. | |
1574 | */ | |
1575 | void ocfs2_open_unlock(struct inode *inode) | |
1576 | { | |
1577 | struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_open_lockres; | |
1578 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1579 | ||
1580 | mlog_entry_void(); | |
1581 | ||
1582 | mlog(0, "inode %llu drop open lock\n", | |
1583 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
1584 | ||
1585 | if (ocfs2_mount_local(osb)) | |
1586 | goto out; | |
1587 | ||
1588 | if(lockres->l_ro_holders) | |
1589 | ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, | |
bd3e7610 | 1590 | DLM_LOCK_PR); |
50008630 TY |
1591 | if(lockres->l_ex_holders) |
1592 | ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, | |
bd3e7610 | 1593 | DLM_LOCK_EX); |
50008630 TY |
1594 | |
1595 | out: | |
1596 | mlog_exit_void(); | |
1597 | } | |
1598 | ||
cf8e06f1 MF |
1599 | static int ocfs2_flock_handle_signal(struct ocfs2_lock_res *lockres, |
1600 | int level) | |
1601 | { | |
1602 | int ret; | |
1603 | struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres); | |
1604 | unsigned long flags; | |
1605 | struct ocfs2_mask_waiter mw; | |
1606 | ||
1607 | ocfs2_init_mask_waiter(&mw); | |
1608 | ||
1609 | retry_cancel: | |
1610 | spin_lock_irqsave(&lockres->l_lock, flags); | |
1611 | if (lockres->l_flags & OCFS2_LOCK_BUSY) { | |
1612 | ret = ocfs2_prepare_cancel_convert(osb, lockres); | |
1613 | if (ret) { | |
1614 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1615 | ret = ocfs2_cancel_convert(osb, lockres); | |
1616 | if (ret < 0) { | |
1617 | mlog_errno(ret); | |
1618 | goto out; | |
1619 | } | |
1620 | goto retry_cancel; | |
1621 | } | |
1622 | lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0); | |
1623 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1624 | ||
1625 | ocfs2_wait_for_mask(&mw); | |
1626 | goto retry_cancel; | |
1627 | } | |
1628 | ||
1629 | ret = -ERESTARTSYS; | |
1630 | /* | |
1631 | * We may still have gotten the lock, in which case there's no | |
1632 | * point to restarting the syscall. | |
1633 | */ | |
1634 | if (lockres->l_level == level) | |
1635 | ret = 0; | |
1636 | ||
1637 | mlog(0, "Cancel returning %d. flags: 0x%lx, level: %d, act: %d\n", ret, | |
1638 | lockres->l_flags, lockres->l_level, lockres->l_action); | |
1639 | ||
1640 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1641 | ||
1642 | out: | |
1643 | return ret; | |
1644 | } | |
1645 | ||
1646 | /* | |
1647 | * ocfs2_file_lock() and ocfs2_file_unlock() map to a single pair of | |
1648 | * flock() calls. The locking approach this requires is sufficiently | |
1649 | * different from all other cluster lock types that we implement a | |
1650 | * seperate path to the "low-level" dlm calls. In particular: | |
1651 | * | |
1652 | * - No optimization of lock levels is done - we take at exactly | |
1653 | * what's been requested. | |
1654 | * | |
1655 | * - No lock caching is employed. We immediately downconvert to | |
1656 | * no-lock at unlock time. This also means flock locks never go on | |
1657 | * the blocking list). | |
1658 | * | |
1659 | * - Since userspace can trivially deadlock itself with flock, we make | |
1660 | * sure to allow cancellation of a misbehaving applications flock() | |
1661 | * request. | |
1662 | * | |
1663 | * - Access to any flock lockres doesn't require concurrency, so we | |
1664 | * can simplify the code by requiring the caller to guarantee | |
1665 | * serialization of dlmglue flock calls. | |
1666 | */ | |
1667 | int ocfs2_file_lock(struct file *file, int ex, int trylock) | |
1668 | { | |
e988cf1c MF |
1669 | int ret, level = ex ? DLM_LOCK_EX : DLM_LOCK_PR; |
1670 | unsigned int lkm_flags = trylock ? DLM_LKF_NOQUEUE : 0; | |
cf8e06f1 MF |
1671 | unsigned long flags; |
1672 | struct ocfs2_file_private *fp = file->private_data; | |
1673 | struct ocfs2_lock_res *lockres = &fp->fp_flock; | |
1674 | struct ocfs2_super *osb = OCFS2_SB(file->f_mapping->host->i_sb); | |
1675 | struct ocfs2_mask_waiter mw; | |
1676 | ||
1677 | ocfs2_init_mask_waiter(&mw); | |
1678 | ||
1679 | if ((lockres->l_flags & OCFS2_LOCK_BUSY) || | |
bd3e7610 | 1680 | (lockres->l_level > DLM_LOCK_NL)) { |
cf8e06f1 MF |
1681 | mlog(ML_ERROR, |
1682 | "File lock \"%s\" has busy or locked state: flags: 0x%lx, " | |
1683 | "level: %u\n", lockres->l_name, lockres->l_flags, | |
1684 | lockres->l_level); | |
1685 | return -EINVAL; | |
1686 | } | |
1687 | ||
1688 | spin_lock_irqsave(&lockres->l_lock, flags); | |
1689 | if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) { | |
1690 | lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0); | |
1691 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1692 | ||
1693 | /* | |
1694 | * Get the lock at NLMODE to start - that way we | |
1695 | * can cancel the upconvert request if need be. | |
1696 | */ | |
e988cf1c | 1697 | ret = ocfs2_lock_create(osb, lockres, DLM_LOCK_NL, 0); |
cf8e06f1 MF |
1698 | if (ret < 0) { |
1699 | mlog_errno(ret); | |
1700 | goto out; | |
1701 | } | |
1702 | ||
1703 | ret = ocfs2_wait_for_mask(&mw); | |
1704 | if (ret) { | |
1705 | mlog_errno(ret); | |
1706 | goto out; | |
1707 | } | |
1708 | spin_lock_irqsave(&lockres->l_lock, flags); | |
1709 | } | |
1710 | ||
1711 | lockres->l_action = OCFS2_AST_CONVERT; | |
e988cf1c | 1712 | lkm_flags |= DLM_LKF_CONVERT; |
cf8e06f1 MF |
1713 | lockres->l_requested = level; |
1714 | lockres_or_flags(lockres, OCFS2_LOCK_BUSY); | |
1715 | ||
1716 | lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0); | |
1717 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1718 | ||
4670c46d | 1719 | ret = ocfs2_dlm_lock(osb->cconn, level, &lockres->l_lksb, lkm_flags, |
24ef1815 JB |
1720 | lockres->l_name, OCFS2_LOCK_ID_MAX_LEN - 1, |
1721 | lockres); | |
7431cd7e JB |
1722 | if (ret) { |
1723 | if (!trylock || (ret != -EAGAIN)) { | |
24ef1815 | 1724 | ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres); |
cf8e06f1 MF |
1725 | ret = -EINVAL; |
1726 | } | |
1727 | ||
1728 | ocfs2_recover_from_dlm_error(lockres, 1); | |
1729 | lockres_remove_mask_waiter(lockres, &mw); | |
1730 | goto out; | |
1731 | } | |
1732 | ||
1733 | ret = ocfs2_wait_for_mask_interruptible(&mw, lockres); | |
1734 | if (ret == -ERESTARTSYS) { | |
1735 | /* | |
1736 | * Userspace can cause deadlock itself with | |
1737 | * flock(). Current behavior locally is to allow the | |
1738 | * deadlock, but abort the system call if a signal is | |
1739 | * received. We follow this example, otherwise a | |
1740 | * poorly written program could sit in kernel until | |
1741 | * reboot. | |
1742 | * | |
1743 | * Handling this is a bit more complicated for Ocfs2 | |
1744 | * though. We can't exit this function with an | |
1745 | * outstanding lock request, so a cancel convert is | |
1746 | * required. We intentionally overwrite 'ret' - if the | |
1747 | * cancel fails and the lock was granted, it's easier | |
1748 | * to just bubble sucess back up to the user. | |
1749 | */ | |
1750 | ret = ocfs2_flock_handle_signal(lockres, level); | |
1693a5c0 DT |
1751 | } else if (!ret && (level > lockres->l_level)) { |
1752 | /* Trylock failed asynchronously */ | |
1753 | BUG_ON(!trylock); | |
1754 | ret = -EAGAIN; | |
cf8e06f1 MF |
1755 | } |
1756 | ||
1757 | out: | |
1758 | ||
1759 | mlog(0, "Lock: \"%s\" ex: %d, trylock: %d, returns: %d\n", | |
1760 | lockres->l_name, ex, trylock, ret); | |
1761 | return ret; | |
1762 | } | |
1763 | ||
1764 | void ocfs2_file_unlock(struct file *file) | |
1765 | { | |
1766 | int ret; | |
de551246 | 1767 | unsigned int gen; |
cf8e06f1 MF |
1768 | unsigned long flags; |
1769 | struct ocfs2_file_private *fp = file->private_data; | |
1770 | struct ocfs2_lock_res *lockres = &fp->fp_flock; | |
1771 | struct ocfs2_super *osb = OCFS2_SB(file->f_mapping->host->i_sb); | |
1772 | struct ocfs2_mask_waiter mw; | |
1773 | ||
1774 | ocfs2_init_mask_waiter(&mw); | |
1775 | ||
1776 | if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) | |
1777 | return; | |
1778 | ||
e988cf1c | 1779 | if (lockres->l_level == DLM_LOCK_NL) |
cf8e06f1 MF |
1780 | return; |
1781 | ||
1782 | mlog(0, "Unlock: \"%s\" flags: 0x%lx, level: %d, act: %d\n", | |
1783 | lockres->l_name, lockres->l_flags, lockres->l_level, | |
1784 | lockres->l_action); | |
1785 | ||
1786 | spin_lock_irqsave(&lockres->l_lock, flags); | |
1787 | /* | |
1788 | * Fake a blocking ast for the downconvert code. | |
1789 | */ | |
1790 | lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED); | |
bd3e7610 | 1791 | lockres->l_blocking = DLM_LOCK_EX; |
cf8e06f1 | 1792 | |
e988cf1c | 1793 | gen = ocfs2_prepare_downconvert(lockres, DLM_LOCK_NL); |
cf8e06f1 MF |
1794 | lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0); |
1795 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1796 | ||
e988cf1c | 1797 | ret = ocfs2_downconvert_lock(osb, lockres, DLM_LOCK_NL, 0, gen); |
cf8e06f1 MF |
1798 | if (ret) { |
1799 | mlog_errno(ret); | |
1800 | return; | |
1801 | } | |
1802 | ||
1803 | ret = ocfs2_wait_for_mask(&mw); | |
1804 | if (ret) | |
1805 | mlog_errno(ret); | |
1806 | } | |
1807 | ||
34d024f8 MF |
1808 | static void ocfs2_downconvert_on_unlock(struct ocfs2_super *osb, |
1809 | struct ocfs2_lock_res *lockres) | |
ccd979bd MF |
1810 | { |
1811 | int kick = 0; | |
1812 | ||
1813 | mlog_entry_void(); | |
1814 | ||
1815 | /* If we know that another node is waiting on our lock, kick | |
34d024f8 | 1816 | * the downconvert thread * pre-emptively when we reach a release |
ccd979bd MF |
1817 | * condition. */ |
1818 | if (lockres->l_flags & OCFS2_LOCK_BLOCKED) { | |
1819 | switch(lockres->l_blocking) { | |
bd3e7610 | 1820 | case DLM_LOCK_EX: |
ccd979bd MF |
1821 | if (!lockres->l_ex_holders && !lockres->l_ro_holders) |
1822 | kick = 1; | |
1823 | break; | |
bd3e7610 | 1824 | case DLM_LOCK_PR: |
ccd979bd MF |
1825 | if (!lockres->l_ex_holders) |
1826 | kick = 1; | |
1827 | break; | |
1828 | default: | |
1829 | BUG(); | |
1830 | } | |
1831 | } | |
1832 | ||
1833 | if (kick) | |
34d024f8 | 1834 | ocfs2_wake_downconvert_thread(osb); |
ccd979bd MF |
1835 | |
1836 | mlog_exit_void(); | |
1837 | } | |
1838 | ||
ccd979bd MF |
1839 | #define OCFS2_SEC_BITS 34 |
1840 | #define OCFS2_SEC_SHIFT (64 - 34) | |
1841 | #define OCFS2_NSEC_MASK ((1ULL << OCFS2_SEC_SHIFT) - 1) | |
1842 | ||
1843 | /* LVB only has room for 64 bits of time here so we pack it for | |
1844 | * now. */ | |
1845 | static u64 ocfs2_pack_timespec(struct timespec *spec) | |
1846 | { | |
1847 | u64 res; | |
1848 | u64 sec = spec->tv_sec; | |
1849 | u32 nsec = spec->tv_nsec; | |
1850 | ||
1851 | res = (sec << OCFS2_SEC_SHIFT) | (nsec & OCFS2_NSEC_MASK); | |
1852 | ||
1853 | return res; | |
1854 | } | |
1855 | ||
1856 | /* Call this with the lockres locked. I am reasonably sure we don't | |
1857 | * need ip_lock in this function as anyone who would be changing those | |
e63aecb6 | 1858 | * values is supposed to be blocked in ocfs2_inode_lock right now. */ |
ccd979bd MF |
1859 | static void __ocfs2_stuff_meta_lvb(struct inode *inode) |
1860 | { | |
1861 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
e63aecb6 | 1862 | struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres; |
ccd979bd MF |
1863 | struct ocfs2_meta_lvb *lvb; |
1864 | ||
1865 | mlog_entry_void(); | |
1866 | ||
8f2c9c1b | 1867 | lvb = (struct ocfs2_meta_lvb *)ocfs2_dlm_lvb(&lockres->l_lksb); |
ccd979bd | 1868 | |
24c19ef4 MF |
1869 | /* |
1870 | * Invalidate the LVB of a deleted inode - this way other | |
1871 | * nodes are forced to go to disk and discover the new inode | |
1872 | * status. | |
1873 | */ | |
1874 | if (oi->ip_flags & OCFS2_INODE_DELETED) { | |
1875 | lvb->lvb_version = 0; | |
1876 | goto out; | |
1877 | } | |
1878 | ||
4d3b83f7 | 1879 | lvb->lvb_version = OCFS2_LVB_VERSION; |
ccd979bd MF |
1880 | lvb->lvb_isize = cpu_to_be64(i_size_read(inode)); |
1881 | lvb->lvb_iclusters = cpu_to_be32(oi->ip_clusters); | |
1882 | lvb->lvb_iuid = cpu_to_be32(inode->i_uid); | |
1883 | lvb->lvb_igid = cpu_to_be32(inode->i_gid); | |
1884 | lvb->lvb_imode = cpu_to_be16(inode->i_mode); | |
1885 | lvb->lvb_inlink = cpu_to_be16(inode->i_nlink); | |
1886 | lvb->lvb_iatime_packed = | |
1887 | cpu_to_be64(ocfs2_pack_timespec(&inode->i_atime)); | |
1888 | lvb->lvb_ictime_packed = | |
1889 | cpu_to_be64(ocfs2_pack_timespec(&inode->i_ctime)); | |
1890 | lvb->lvb_imtime_packed = | |
1891 | cpu_to_be64(ocfs2_pack_timespec(&inode->i_mtime)); | |
ca4d147e | 1892 | lvb->lvb_iattr = cpu_to_be32(oi->ip_attr); |
15b1e36b | 1893 | lvb->lvb_idynfeatures = cpu_to_be16(oi->ip_dyn_features); |
f9e2d82e | 1894 | lvb->lvb_igeneration = cpu_to_be32(inode->i_generation); |
ccd979bd | 1895 | |
24c19ef4 | 1896 | out: |
ccd979bd MF |
1897 | mlog_meta_lvb(0, lockres); |
1898 | ||
1899 | mlog_exit_void(); | |
1900 | } | |
1901 | ||
1902 | static void ocfs2_unpack_timespec(struct timespec *spec, | |
1903 | u64 packed_time) | |
1904 | { | |
1905 | spec->tv_sec = packed_time >> OCFS2_SEC_SHIFT; | |
1906 | spec->tv_nsec = packed_time & OCFS2_NSEC_MASK; | |
1907 | } | |
1908 | ||
1909 | static void ocfs2_refresh_inode_from_lvb(struct inode *inode) | |
1910 | { | |
1911 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
e63aecb6 | 1912 | struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres; |
ccd979bd MF |
1913 | struct ocfs2_meta_lvb *lvb; |
1914 | ||
1915 | mlog_entry_void(); | |
1916 | ||
1917 | mlog_meta_lvb(0, lockres); | |
1918 | ||
8f2c9c1b | 1919 | lvb = (struct ocfs2_meta_lvb *)ocfs2_dlm_lvb(&lockres->l_lksb); |
ccd979bd MF |
1920 | |
1921 | /* We're safe here without the lockres lock... */ | |
1922 | spin_lock(&oi->ip_lock); | |
1923 | oi->ip_clusters = be32_to_cpu(lvb->lvb_iclusters); | |
1924 | i_size_write(inode, be64_to_cpu(lvb->lvb_isize)); | |
1925 | ||
ca4d147e | 1926 | oi->ip_attr = be32_to_cpu(lvb->lvb_iattr); |
15b1e36b | 1927 | oi->ip_dyn_features = be16_to_cpu(lvb->lvb_idynfeatures); |
ca4d147e HP |
1928 | ocfs2_set_inode_flags(inode); |
1929 | ||
ccd979bd MF |
1930 | /* fast-symlinks are a special case */ |
1931 | if (S_ISLNK(inode->i_mode) && !oi->ip_clusters) | |
1932 | inode->i_blocks = 0; | |
1933 | else | |
8110b073 | 1934 | inode->i_blocks = ocfs2_inode_sector_count(inode); |
ccd979bd MF |
1935 | |
1936 | inode->i_uid = be32_to_cpu(lvb->lvb_iuid); | |
1937 | inode->i_gid = be32_to_cpu(lvb->lvb_igid); | |
1938 | inode->i_mode = be16_to_cpu(lvb->lvb_imode); | |
1939 | inode->i_nlink = be16_to_cpu(lvb->lvb_inlink); | |
1940 | ocfs2_unpack_timespec(&inode->i_atime, | |
1941 | be64_to_cpu(lvb->lvb_iatime_packed)); | |
1942 | ocfs2_unpack_timespec(&inode->i_mtime, | |
1943 | be64_to_cpu(lvb->lvb_imtime_packed)); | |
1944 | ocfs2_unpack_timespec(&inode->i_ctime, | |
1945 | be64_to_cpu(lvb->lvb_ictime_packed)); | |
1946 | spin_unlock(&oi->ip_lock); | |
1947 | ||
1948 | mlog_exit_void(); | |
1949 | } | |
1950 | ||
f9e2d82e MF |
1951 | static inline int ocfs2_meta_lvb_is_trustable(struct inode *inode, |
1952 | struct ocfs2_lock_res *lockres) | |
ccd979bd | 1953 | { |
8f2c9c1b JB |
1954 | struct ocfs2_meta_lvb *lvb = |
1955 | (struct ocfs2_meta_lvb *)ocfs2_dlm_lvb(&lockres->l_lksb); | |
ccd979bd | 1956 | |
f9e2d82e MF |
1957 | if (lvb->lvb_version == OCFS2_LVB_VERSION |
1958 | && be32_to_cpu(lvb->lvb_igeneration) == inode->i_generation) | |
ccd979bd MF |
1959 | return 1; |
1960 | return 0; | |
1961 | } | |
1962 | ||
1963 | /* Determine whether a lock resource needs to be refreshed, and | |
1964 | * arbitrate who gets to refresh it. | |
1965 | * | |
1966 | * 0 means no refresh needed. | |
1967 | * | |
1968 | * > 0 means you need to refresh this and you MUST call | |
1969 | * ocfs2_complete_lock_res_refresh afterwards. */ | |
1970 | static int ocfs2_should_refresh_lock_res(struct ocfs2_lock_res *lockres) | |
1971 | { | |
1972 | unsigned long flags; | |
1973 | int status = 0; | |
1974 | ||
1975 | mlog_entry_void(); | |
1976 | ||
1977 | refresh_check: | |
1978 | spin_lock_irqsave(&lockres->l_lock, flags); | |
1979 | if (!(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) { | |
1980 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1981 | goto bail; | |
1982 | } | |
1983 | ||
1984 | if (lockres->l_flags & OCFS2_LOCK_REFRESHING) { | |
1985 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1986 | ||
1987 | ocfs2_wait_on_refreshing_lock(lockres); | |
1988 | goto refresh_check; | |
1989 | } | |
1990 | ||
1991 | /* Ok, I'll be the one to refresh this lock. */ | |
1992 | lockres_or_flags(lockres, OCFS2_LOCK_REFRESHING); | |
1993 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
1994 | ||
1995 | status = 1; | |
1996 | bail: | |
1997 | mlog_exit(status); | |
1998 | return status; | |
1999 | } | |
2000 | ||
2001 | /* If status is non zero, I'll mark it as not being in refresh | |
2002 | * anymroe, but i won't clear the needs refresh flag. */ | |
2003 | static inline void ocfs2_complete_lock_res_refresh(struct ocfs2_lock_res *lockres, | |
2004 | int status) | |
2005 | { | |
2006 | unsigned long flags; | |
2007 | mlog_entry_void(); | |
2008 | ||
2009 | spin_lock_irqsave(&lockres->l_lock, flags); | |
2010 | lockres_clear_flags(lockres, OCFS2_LOCK_REFRESHING); | |
2011 | if (!status) | |
2012 | lockres_clear_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH); | |
2013 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2014 | ||
2015 | wake_up(&lockres->l_event); | |
2016 | ||
2017 | mlog_exit_void(); | |
2018 | } | |
2019 | ||
2020 | /* may or may not return a bh if it went to disk. */ | |
e63aecb6 | 2021 | static int ocfs2_inode_lock_update(struct inode *inode, |
ccd979bd MF |
2022 | struct buffer_head **bh) |
2023 | { | |
2024 | int status = 0; | |
2025 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
e63aecb6 | 2026 | struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres; |
ccd979bd | 2027 | struct ocfs2_dinode *fe; |
c271c5c2 | 2028 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
ccd979bd MF |
2029 | |
2030 | mlog_entry_void(); | |
2031 | ||
be9e986b MF |
2032 | if (ocfs2_mount_local(osb)) |
2033 | goto bail; | |
2034 | ||
ccd979bd MF |
2035 | spin_lock(&oi->ip_lock); |
2036 | if (oi->ip_flags & OCFS2_INODE_DELETED) { | |
b0697053 | 2037 | mlog(0, "Orphaned inode %llu was deleted while we " |
ccd979bd | 2038 | "were waiting on a lock. ip_flags = 0x%x\n", |
b0697053 | 2039 | (unsigned long long)oi->ip_blkno, oi->ip_flags); |
ccd979bd MF |
2040 | spin_unlock(&oi->ip_lock); |
2041 | status = -ENOENT; | |
2042 | goto bail; | |
2043 | } | |
2044 | spin_unlock(&oi->ip_lock); | |
2045 | ||
be9e986b MF |
2046 | if (!ocfs2_should_refresh_lock_res(lockres)) |
2047 | goto bail; | |
ccd979bd MF |
2048 | |
2049 | /* This will discard any caching information we might have had | |
2050 | * for the inode metadata. */ | |
2051 | ocfs2_metadata_cache_purge(inode); | |
2052 | ||
83418978 MF |
2053 | ocfs2_extent_map_trunc(inode, 0); |
2054 | ||
be9e986b | 2055 | if (ocfs2_meta_lvb_is_trustable(inode, lockres)) { |
b0697053 MF |
2056 | mlog(0, "Trusting LVB on inode %llu\n", |
2057 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
2058 | ocfs2_refresh_inode_from_lvb(inode); |
2059 | } else { | |
2060 | /* Boo, we have to go to disk. */ | |
2061 | /* read bh, cast, ocfs2_refresh_inode */ | |
b657c95c | 2062 | status = ocfs2_read_inode_block(inode, bh); |
ccd979bd MF |
2063 | if (status < 0) { |
2064 | mlog_errno(status); | |
2065 | goto bail_refresh; | |
2066 | } | |
2067 | fe = (struct ocfs2_dinode *) (*bh)->b_data; | |
2068 | ||
2069 | /* This is a good chance to make sure we're not | |
b657c95c JB |
2070 | * locking an invalid object. ocfs2_read_inode_block() |
2071 | * already checked that the inode block is sane. | |
ccd979bd MF |
2072 | * |
2073 | * We bug on a stale inode here because we checked | |
2074 | * above whether it was wiped from disk. The wiping | |
2075 | * node provides a guarantee that we receive that | |
2076 | * message and can mark the inode before dropping any | |
2077 | * locks associated with it. */ | |
ccd979bd MF |
2078 | mlog_bug_on_msg(inode->i_generation != |
2079 | le32_to_cpu(fe->i_generation), | |
b0697053 | 2080 | "Invalid dinode %llu disk generation: %u " |
ccd979bd | 2081 | "inode->i_generation: %u\n", |
b0697053 MF |
2082 | (unsigned long long)oi->ip_blkno, |
2083 | le32_to_cpu(fe->i_generation), | |
ccd979bd MF |
2084 | inode->i_generation); |
2085 | mlog_bug_on_msg(le64_to_cpu(fe->i_dtime) || | |
2086 | !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)), | |
b0697053 MF |
2087 | "Stale dinode %llu dtime: %llu flags: 0x%x\n", |
2088 | (unsigned long long)oi->ip_blkno, | |
2089 | (unsigned long long)le64_to_cpu(fe->i_dtime), | |
ccd979bd MF |
2090 | le32_to_cpu(fe->i_flags)); |
2091 | ||
2092 | ocfs2_refresh_inode(inode, fe); | |
8ddb7b00 | 2093 | ocfs2_track_lock_refresh(lockres); |
ccd979bd MF |
2094 | } |
2095 | ||
2096 | status = 0; | |
2097 | bail_refresh: | |
be9e986b | 2098 | ocfs2_complete_lock_res_refresh(lockres, status); |
ccd979bd MF |
2099 | bail: |
2100 | mlog_exit(status); | |
2101 | return status; | |
2102 | } | |
2103 | ||
2104 | static int ocfs2_assign_bh(struct inode *inode, | |
2105 | struct buffer_head **ret_bh, | |
2106 | struct buffer_head *passed_bh) | |
2107 | { | |
2108 | int status; | |
2109 | ||
2110 | if (passed_bh) { | |
2111 | /* Ok, the update went to disk for us, use the | |
2112 | * returned bh. */ | |
2113 | *ret_bh = passed_bh; | |
2114 | get_bh(*ret_bh); | |
2115 | ||
2116 | return 0; | |
2117 | } | |
2118 | ||
b657c95c | 2119 | status = ocfs2_read_inode_block(inode, ret_bh); |
ccd979bd MF |
2120 | if (status < 0) |
2121 | mlog_errno(status); | |
2122 | ||
2123 | return status; | |
2124 | } | |
2125 | ||
2126 | /* | |
2127 | * returns < 0 error if the callback will never be called, otherwise | |
2128 | * the result of the lock will be communicated via the callback. | |
2129 | */ | |
e63aecb6 | 2130 | int ocfs2_inode_lock_full(struct inode *inode, |
ccd979bd MF |
2131 | struct buffer_head **ret_bh, |
2132 | int ex, | |
2133 | int arg_flags) | |
2134 | { | |
bd3e7610 JB |
2135 | int status, level, acquired; |
2136 | u32 dlm_flags; | |
c271c5c2 | 2137 | struct ocfs2_lock_res *lockres = NULL; |
ccd979bd MF |
2138 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
2139 | struct buffer_head *local_bh = NULL; | |
2140 | ||
2141 | BUG_ON(!inode); | |
2142 | ||
2143 | mlog_entry_void(); | |
2144 | ||
b0697053 MF |
2145 | mlog(0, "inode %llu, take %s META lock\n", |
2146 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
ccd979bd MF |
2147 | ex ? "EXMODE" : "PRMODE"); |
2148 | ||
2149 | status = 0; | |
2150 | acquired = 0; | |
2151 | /* We'll allow faking a readonly metadata lock for | |
2152 | * rodevices. */ | |
2153 | if (ocfs2_is_hard_readonly(osb)) { | |
2154 | if (ex) | |
2155 | status = -EROFS; | |
2156 | goto bail; | |
2157 | } | |
2158 | ||
c271c5c2 SM |
2159 | if (ocfs2_mount_local(osb)) |
2160 | goto local; | |
2161 | ||
ccd979bd | 2162 | if (!(arg_flags & OCFS2_META_LOCK_RECOVERY)) |
553abd04 | 2163 | ocfs2_wait_for_recovery(osb); |
ccd979bd | 2164 | |
e63aecb6 | 2165 | lockres = &OCFS2_I(inode)->ip_inode_lockres; |
bd3e7610 | 2166 | level = ex ? DLM_LOCK_EX : DLM_LOCK_PR; |
ccd979bd MF |
2167 | dlm_flags = 0; |
2168 | if (arg_flags & OCFS2_META_LOCK_NOQUEUE) | |
bd3e7610 | 2169 | dlm_flags |= DLM_LKF_NOQUEUE; |
ccd979bd MF |
2170 | |
2171 | status = ocfs2_cluster_lock(osb, lockres, level, dlm_flags, arg_flags); | |
2172 | if (status < 0) { | |
2173 | if (status != -EAGAIN && status != -EIOCBRETRY) | |
2174 | mlog_errno(status); | |
2175 | goto bail; | |
2176 | } | |
2177 | ||
2178 | /* Notify the error cleanup path to drop the cluster lock. */ | |
2179 | acquired = 1; | |
2180 | ||
2181 | /* We wait twice because a node may have died while we were in | |
2182 | * the lower dlm layers. The second time though, we've | |
2183 | * committed to owning this lock so we don't allow signals to | |
2184 | * abort the operation. */ | |
2185 | if (!(arg_flags & OCFS2_META_LOCK_RECOVERY)) | |
553abd04 | 2186 | ocfs2_wait_for_recovery(osb); |
ccd979bd | 2187 | |
c271c5c2 | 2188 | local: |
24c19ef4 MF |
2189 | /* |
2190 | * We only see this flag if we're being called from | |
2191 | * ocfs2_read_locked_inode(). It means we're locking an inode | |
2192 | * which hasn't been populated yet, so clear the refresh flag | |
2193 | * and let the caller handle it. | |
2194 | */ | |
2195 | if (inode->i_state & I_NEW) { | |
2196 | status = 0; | |
c271c5c2 SM |
2197 | if (lockres) |
2198 | ocfs2_complete_lock_res_refresh(lockres, 0); | |
24c19ef4 MF |
2199 | goto bail; |
2200 | } | |
2201 | ||
ccd979bd | 2202 | /* This is fun. The caller may want a bh back, or it may |
e63aecb6 | 2203 | * not. ocfs2_inode_lock_update definitely wants one in, but |
ccd979bd MF |
2204 | * may or may not read one, depending on what's in the |
2205 | * LVB. The result of all of this is that we've *only* gone to | |
2206 | * disk if we have to, so the complexity is worthwhile. */ | |
e63aecb6 | 2207 | status = ocfs2_inode_lock_update(inode, &local_bh); |
ccd979bd MF |
2208 | if (status < 0) { |
2209 | if (status != -ENOENT) | |
2210 | mlog_errno(status); | |
2211 | goto bail; | |
2212 | } | |
2213 | ||
2214 | if (ret_bh) { | |
2215 | status = ocfs2_assign_bh(inode, ret_bh, local_bh); | |
2216 | if (status < 0) { | |
2217 | mlog_errno(status); | |
2218 | goto bail; | |
2219 | } | |
2220 | } | |
2221 | ||
ccd979bd MF |
2222 | bail: |
2223 | if (status < 0) { | |
2224 | if (ret_bh && (*ret_bh)) { | |
2225 | brelse(*ret_bh); | |
2226 | *ret_bh = NULL; | |
2227 | } | |
2228 | if (acquired) | |
e63aecb6 | 2229 | ocfs2_inode_unlock(inode, ex); |
ccd979bd MF |
2230 | } |
2231 | ||
2232 | if (local_bh) | |
2233 | brelse(local_bh); | |
2234 | ||
2235 | mlog_exit(status); | |
2236 | return status; | |
2237 | } | |
2238 | ||
2239 | /* | |
34d024f8 MF |
2240 | * This is working around a lock inversion between tasks acquiring DLM |
2241 | * locks while holding a page lock and the downconvert thread which | |
2242 | * blocks dlm lock acquiry while acquiring page locks. | |
ccd979bd MF |
2243 | * |
2244 | * ** These _with_page variantes are only intended to be called from aop | |
2245 | * methods that hold page locks and return a very specific *positive* error | |
2246 | * code that aop methods pass up to the VFS -- test for errors with != 0. ** | |
2247 | * | |
34d024f8 MF |
2248 | * The DLM is called such that it returns -EAGAIN if it would have |
2249 | * blocked waiting for the downconvert thread. In that case we unlock | |
2250 | * our page so the downconvert thread can make progress. Once we've | |
2251 | * done this we have to return AOP_TRUNCATED_PAGE so the aop method | |
2252 | * that called us can bubble that back up into the VFS who will then | |
2253 | * immediately retry the aop call. | |
ccd979bd MF |
2254 | * |
2255 | * We do a blocking lock and immediate unlock before returning, though, so that | |
2256 | * the lock has a great chance of being cached on this node by the time the VFS | |
2257 | * calls back to retry the aop. This has a potential to livelock as nodes | |
2258 | * ping locks back and forth, but that's a risk we're willing to take to avoid | |
2259 | * the lock inversion simply. | |
2260 | */ | |
e63aecb6 | 2261 | int ocfs2_inode_lock_with_page(struct inode *inode, |
ccd979bd MF |
2262 | struct buffer_head **ret_bh, |
2263 | int ex, | |
2264 | struct page *page) | |
2265 | { | |
2266 | int ret; | |
2267 | ||
e63aecb6 | 2268 | ret = ocfs2_inode_lock_full(inode, ret_bh, ex, OCFS2_LOCK_NONBLOCK); |
ccd979bd MF |
2269 | if (ret == -EAGAIN) { |
2270 | unlock_page(page); | |
e63aecb6 MF |
2271 | if (ocfs2_inode_lock(inode, ret_bh, ex) == 0) |
2272 | ocfs2_inode_unlock(inode, ex); | |
ccd979bd MF |
2273 | ret = AOP_TRUNCATED_PAGE; |
2274 | } | |
2275 | ||
2276 | return ret; | |
2277 | } | |
2278 | ||
e63aecb6 | 2279 | int ocfs2_inode_lock_atime(struct inode *inode, |
7f1a37e3 TY |
2280 | struct vfsmount *vfsmnt, |
2281 | int *level) | |
2282 | { | |
2283 | int ret; | |
2284 | ||
2285 | mlog_entry_void(); | |
e63aecb6 | 2286 | ret = ocfs2_inode_lock(inode, NULL, 0); |
7f1a37e3 TY |
2287 | if (ret < 0) { |
2288 | mlog_errno(ret); | |
2289 | return ret; | |
2290 | } | |
2291 | ||
2292 | /* | |
2293 | * If we should update atime, we will get EX lock, | |
2294 | * otherwise we just get PR lock. | |
2295 | */ | |
2296 | if (ocfs2_should_update_atime(inode, vfsmnt)) { | |
2297 | struct buffer_head *bh = NULL; | |
2298 | ||
e63aecb6 MF |
2299 | ocfs2_inode_unlock(inode, 0); |
2300 | ret = ocfs2_inode_lock(inode, &bh, 1); | |
7f1a37e3 TY |
2301 | if (ret < 0) { |
2302 | mlog_errno(ret); | |
2303 | return ret; | |
2304 | } | |
2305 | *level = 1; | |
2306 | if (ocfs2_should_update_atime(inode, vfsmnt)) | |
2307 | ocfs2_update_inode_atime(inode, bh); | |
2308 | if (bh) | |
2309 | brelse(bh); | |
2310 | } else | |
2311 | *level = 0; | |
2312 | ||
2313 | mlog_exit(ret); | |
2314 | return ret; | |
2315 | } | |
2316 | ||
e63aecb6 | 2317 | void ocfs2_inode_unlock(struct inode *inode, |
ccd979bd MF |
2318 | int ex) |
2319 | { | |
bd3e7610 | 2320 | int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR; |
e63aecb6 | 2321 | struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_inode_lockres; |
c271c5c2 | 2322 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
ccd979bd MF |
2323 | |
2324 | mlog_entry_void(); | |
2325 | ||
b0697053 MF |
2326 | mlog(0, "inode %llu drop %s META lock\n", |
2327 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
ccd979bd MF |
2328 | ex ? "EXMODE" : "PRMODE"); |
2329 | ||
c271c5c2 SM |
2330 | if (!ocfs2_is_hard_readonly(OCFS2_SB(inode->i_sb)) && |
2331 | !ocfs2_mount_local(osb)) | |
ccd979bd MF |
2332 | ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level); |
2333 | ||
2334 | mlog_exit_void(); | |
2335 | } | |
2336 | ||
2337 | int ocfs2_super_lock(struct ocfs2_super *osb, | |
2338 | int ex) | |
2339 | { | |
c271c5c2 | 2340 | int status = 0; |
bd3e7610 | 2341 | int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR; |
ccd979bd | 2342 | struct ocfs2_lock_res *lockres = &osb->osb_super_lockres; |
ccd979bd MF |
2343 | |
2344 | mlog_entry_void(); | |
2345 | ||
2346 | if (ocfs2_is_hard_readonly(osb)) | |
2347 | return -EROFS; | |
2348 | ||
c271c5c2 SM |
2349 | if (ocfs2_mount_local(osb)) |
2350 | goto bail; | |
2351 | ||
ccd979bd MF |
2352 | status = ocfs2_cluster_lock(osb, lockres, level, 0, 0); |
2353 | if (status < 0) { | |
2354 | mlog_errno(status); | |
2355 | goto bail; | |
2356 | } | |
2357 | ||
2358 | /* The super block lock path is really in the best position to | |
2359 | * know when resources covered by the lock need to be | |
2360 | * refreshed, so we do it here. Of course, making sense of | |
2361 | * everything is up to the caller :) */ | |
2362 | status = ocfs2_should_refresh_lock_res(lockres); | |
2363 | if (status < 0) { | |
2364 | mlog_errno(status); | |
2365 | goto bail; | |
2366 | } | |
2367 | if (status) { | |
8e8a4603 | 2368 | status = ocfs2_refresh_slot_info(osb); |
ccd979bd MF |
2369 | |
2370 | ocfs2_complete_lock_res_refresh(lockres, status); | |
2371 | ||
2372 | if (status < 0) | |
2373 | mlog_errno(status); | |
8ddb7b00 | 2374 | ocfs2_track_lock_refresh(lockres); |
ccd979bd MF |
2375 | } |
2376 | bail: | |
2377 | mlog_exit(status); | |
2378 | return status; | |
2379 | } | |
2380 | ||
2381 | void ocfs2_super_unlock(struct ocfs2_super *osb, | |
2382 | int ex) | |
2383 | { | |
bd3e7610 | 2384 | int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR; |
ccd979bd MF |
2385 | struct ocfs2_lock_res *lockres = &osb->osb_super_lockres; |
2386 | ||
c271c5c2 SM |
2387 | if (!ocfs2_mount_local(osb)) |
2388 | ocfs2_cluster_unlock(osb, lockres, level); | |
ccd979bd MF |
2389 | } |
2390 | ||
2391 | int ocfs2_rename_lock(struct ocfs2_super *osb) | |
2392 | { | |
2393 | int status; | |
2394 | struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres; | |
2395 | ||
2396 | if (ocfs2_is_hard_readonly(osb)) | |
2397 | return -EROFS; | |
2398 | ||
c271c5c2 SM |
2399 | if (ocfs2_mount_local(osb)) |
2400 | return 0; | |
2401 | ||
bd3e7610 | 2402 | status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_EX, 0, 0); |
ccd979bd MF |
2403 | if (status < 0) |
2404 | mlog_errno(status); | |
2405 | ||
2406 | return status; | |
2407 | } | |
2408 | ||
2409 | void ocfs2_rename_unlock(struct ocfs2_super *osb) | |
2410 | { | |
2411 | struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres; | |
2412 | ||
c271c5c2 | 2413 | if (!ocfs2_mount_local(osb)) |
bd3e7610 | 2414 | ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX); |
ccd979bd MF |
2415 | } |
2416 | ||
d680efe9 MF |
2417 | int ocfs2_dentry_lock(struct dentry *dentry, int ex) |
2418 | { | |
2419 | int ret; | |
bd3e7610 | 2420 | int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR; |
d680efe9 MF |
2421 | struct ocfs2_dentry_lock *dl = dentry->d_fsdata; |
2422 | struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb); | |
2423 | ||
2424 | BUG_ON(!dl); | |
2425 | ||
2426 | if (ocfs2_is_hard_readonly(osb)) | |
2427 | return -EROFS; | |
2428 | ||
c271c5c2 SM |
2429 | if (ocfs2_mount_local(osb)) |
2430 | return 0; | |
2431 | ||
d680efe9 MF |
2432 | ret = ocfs2_cluster_lock(osb, &dl->dl_lockres, level, 0, 0); |
2433 | if (ret < 0) | |
2434 | mlog_errno(ret); | |
2435 | ||
2436 | return ret; | |
2437 | } | |
2438 | ||
2439 | void ocfs2_dentry_unlock(struct dentry *dentry, int ex) | |
2440 | { | |
bd3e7610 | 2441 | int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR; |
d680efe9 MF |
2442 | struct ocfs2_dentry_lock *dl = dentry->d_fsdata; |
2443 | struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb); | |
2444 | ||
c271c5c2 SM |
2445 | if (!ocfs2_mount_local(osb)) |
2446 | ocfs2_cluster_unlock(osb, &dl->dl_lockres, level); | |
d680efe9 MF |
2447 | } |
2448 | ||
ccd979bd MF |
2449 | /* Reference counting of the dlm debug structure. We want this because |
2450 | * open references on the debug inodes can live on after a mount, so | |
2451 | * we can't rely on the ocfs2_super to always exist. */ | |
2452 | static void ocfs2_dlm_debug_free(struct kref *kref) | |
2453 | { | |
2454 | struct ocfs2_dlm_debug *dlm_debug; | |
2455 | ||
2456 | dlm_debug = container_of(kref, struct ocfs2_dlm_debug, d_refcnt); | |
2457 | ||
2458 | kfree(dlm_debug); | |
2459 | } | |
2460 | ||
2461 | void ocfs2_put_dlm_debug(struct ocfs2_dlm_debug *dlm_debug) | |
2462 | { | |
2463 | if (dlm_debug) | |
2464 | kref_put(&dlm_debug->d_refcnt, ocfs2_dlm_debug_free); | |
2465 | } | |
2466 | ||
2467 | static void ocfs2_get_dlm_debug(struct ocfs2_dlm_debug *debug) | |
2468 | { | |
2469 | kref_get(&debug->d_refcnt); | |
2470 | } | |
2471 | ||
2472 | struct ocfs2_dlm_debug *ocfs2_new_dlm_debug(void) | |
2473 | { | |
2474 | struct ocfs2_dlm_debug *dlm_debug; | |
2475 | ||
2476 | dlm_debug = kmalloc(sizeof(struct ocfs2_dlm_debug), GFP_KERNEL); | |
2477 | if (!dlm_debug) { | |
2478 | mlog_errno(-ENOMEM); | |
2479 | goto out; | |
2480 | } | |
2481 | ||
2482 | kref_init(&dlm_debug->d_refcnt); | |
2483 | INIT_LIST_HEAD(&dlm_debug->d_lockres_tracking); | |
2484 | dlm_debug->d_locking_state = NULL; | |
2485 | out: | |
2486 | return dlm_debug; | |
2487 | } | |
2488 | ||
2489 | /* Access to this is arbitrated for us via seq_file->sem. */ | |
2490 | struct ocfs2_dlm_seq_priv { | |
2491 | struct ocfs2_dlm_debug *p_dlm_debug; | |
2492 | struct ocfs2_lock_res p_iter_res; | |
2493 | struct ocfs2_lock_res p_tmp_res; | |
2494 | }; | |
2495 | ||
2496 | static struct ocfs2_lock_res *ocfs2_dlm_next_res(struct ocfs2_lock_res *start, | |
2497 | struct ocfs2_dlm_seq_priv *priv) | |
2498 | { | |
2499 | struct ocfs2_lock_res *iter, *ret = NULL; | |
2500 | struct ocfs2_dlm_debug *dlm_debug = priv->p_dlm_debug; | |
2501 | ||
2502 | assert_spin_locked(&ocfs2_dlm_tracking_lock); | |
2503 | ||
2504 | list_for_each_entry(iter, &start->l_debug_list, l_debug_list) { | |
2505 | /* discover the head of the list */ | |
2506 | if (&iter->l_debug_list == &dlm_debug->d_lockres_tracking) { | |
2507 | mlog(0, "End of list found, %p\n", ret); | |
2508 | break; | |
2509 | } | |
2510 | ||
2511 | /* We track our "dummy" iteration lockres' by a NULL | |
2512 | * l_ops field. */ | |
2513 | if (iter->l_ops != NULL) { | |
2514 | ret = iter; | |
2515 | break; | |
2516 | } | |
2517 | } | |
2518 | ||
2519 | return ret; | |
2520 | } | |
2521 | ||
2522 | static void *ocfs2_dlm_seq_start(struct seq_file *m, loff_t *pos) | |
2523 | { | |
2524 | struct ocfs2_dlm_seq_priv *priv = m->private; | |
2525 | struct ocfs2_lock_res *iter; | |
2526 | ||
2527 | spin_lock(&ocfs2_dlm_tracking_lock); | |
2528 | iter = ocfs2_dlm_next_res(&priv->p_iter_res, priv); | |
2529 | if (iter) { | |
2530 | /* Since lockres' have the lifetime of their container | |
2531 | * (which can be inodes, ocfs2_supers, etc) we want to | |
2532 | * copy this out to a temporary lockres while still | |
2533 | * under the spinlock. Obviously after this we can't | |
2534 | * trust any pointers on the copy returned, but that's | |
2535 | * ok as the information we want isn't typically held | |
2536 | * in them. */ | |
2537 | priv->p_tmp_res = *iter; | |
2538 | iter = &priv->p_tmp_res; | |
2539 | } | |
2540 | spin_unlock(&ocfs2_dlm_tracking_lock); | |
2541 | ||
2542 | return iter; | |
2543 | } | |
2544 | ||
2545 | static void ocfs2_dlm_seq_stop(struct seq_file *m, void *v) | |
2546 | { | |
2547 | } | |
2548 | ||
2549 | static void *ocfs2_dlm_seq_next(struct seq_file *m, void *v, loff_t *pos) | |
2550 | { | |
2551 | struct ocfs2_dlm_seq_priv *priv = m->private; | |
2552 | struct ocfs2_lock_res *iter = v; | |
2553 | struct ocfs2_lock_res *dummy = &priv->p_iter_res; | |
2554 | ||
2555 | spin_lock(&ocfs2_dlm_tracking_lock); | |
2556 | iter = ocfs2_dlm_next_res(iter, priv); | |
2557 | list_del_init(&dummy->l_debug_list); | |
2558 | if (iter) { | |
2559 | list_add(&dummy->l_debug_list, &iter->l_debug_list); | |
2560 | priv->p_tmp_res = *iter; | |
2561 | iter = &priv->p_tmp_res; | |
2562 | } | |
2563 | spin_unlock(&ocfs2_dlm_tracking_lock); | |
2564 | ||
2565 | return iter; | |
2566 | } | |
2567 | ||
2568 | /* So that debugfs.ocfs2 can determine which format is being used */ | |
8ddb7b00 | 2569 | #define OCFS2_DLM_DEBUG_STR_VERSION 2 |
ccd979bd MF |
2570 | static int ocfs2_dlm_seq_show(struct seq_file *m, void *v) |
2571 | { | |
2572 | int i; | |
2573 | char *lvb; | |
2574 | struct ocfs2_lock_res *lockres = v; | |
2575 | ||
2576 | if (!lockres) | |
2577 | return -EINVAL; | |
2578 | ||
d680efe9 MF |
2579 | seq_printf(m, "0x%x\t", OCFS2_DLM_DEBUG_STR_VERSION); |
2580 | ||
2581 | if (lockres->l_type == OCFS2_LOCK_TYPE_DENTRY) | |
2582 | seq_printf(m, "%.*s%08x\t", OCFS2_DENTRY_LOCK_INO_START - 1, | |
2583 | lockres->l_name, | |
2584 | (unsigned int)ocfs2_get_dentry_lock_ino(lockres)); | |
2585 | else | |
2586 | seq_printf(m, "%.*s\t", OCFS2_LOCK_ID_MAX_LEN, lockres->l_name); | |
2587 | ||
2588 | seq_printf(m, "%d\t" | |
ccd979bd MF |
2589 | "0x%lx\t" |
2590 | "0x%x\t" | |
2591 | "0x%x\t" | |
2592 | "%u\t" | |
2593 | "%u\t" | |
2594 | "%d\t" | |
2595 | "%d\t", | |
ccd979bd MF |
2596 | lockres->l_level, |
2597 | lockres->l_flags, | |
2598 | lockres->l_action, | |
2599 | lockres->l_unlock_action, | |
2600 | lockres->l_ro_holders, | |
2601 | lockres->l_ex_holders, | |
2602 | lockres->l_requested, | |
2603 | lockres->l_blocking); | |
2604 | ||
2605 | /* Dump the raw LVB */ | |
8f2c9c1b | 2606 | lvb = ocfs2_dlm_lvb(&lockres->l_lksb); |
ccd979bd MF |
2607 | for(i = 0; i < DLM_LVB_LEN; i++) |
2608 | seq_printf(m, "0x%x\t", lvb[i]); | |
2609 | ||
8ddb7b00 SM |
2610 | #ifdef CONFIG_OCFS2_FS_STATS |
2611 | # define lock_num_prmode(_l) (_l)->l_lock_num_prmode | |
2612 | # define lock_num_exmode(_l) (_l)->l_lock_num_exmode | |
2613 | # define lock_num_prmode_failed(_l) (_l)->l_lock_num_prmode_failed | |
2614 | # define lock_num_exmode_failed(_l) (_l)->l_lock_num_exmode_failed | |
2615 | # define lock_total_prmode(_l) (_l)->l_lock_total_prmode | |
2616 | # define lock_total_exmode(_l) (_l)->l_lock_total_exmode | |
2617 | # define lock_max_prmode(_l) (_l)->l_lock_max_prmode | |
2618 | # define lock_max_exmode(_l) (_l)->l_lock_max_exmode | |
2619 | # define lock_refresh(_l) (_l)->l_lock_refresh | |
2620 | #else | |
dd25e55e RD |
2621 | # define lock_num_prmode(_l) (0ULL) |
2622 | # define lock_num_exmode(_l) (0ULL) | |
8ddb7b00 SM |
2623 | # define lock_num_prmode_failed(_l) (0) |
2624 | # define lock_num_exmode_failed(_l) (0) | |
dd25e55e RD |
2625 | # define lock_total_prmode(_l) (0ULL) |
2626 | # define lock_total_exmode(_l) (0ULL) | |
8ddb7b00 SM |
2627 | # define lock_max_prmode(_l) (0) |
2628 | # define lock_max_exmode(_l) (0) | |
2629 | # define lock_refresh(_l) (0) | |
2630 | #endif | |
2631 | /* The following seq_print was added in version 2 of this output */ | |
2632 | seq_printf(m, "%llu\t" | |
2633 | "%llu\t" | |
2634 | "%u\t" | |
2635 | "%u\t" | |
2636 | "%llu\t" | |
2637 | "%llu\t" | |
2638 | "%u\t" | |
2639 | "%u\t" | |
2640 | "%u\t", | |
2641 | lock_num_prmode(lockres), | |
2642 | lock_num_exmode(lockres), | |
2643 | lock_num_prmode_failed(lockres), | |
2644 | lock_num_exmode_failed(lockres), | |
2645 | lock_total_prmode(lockres), | |
2646 | lock_total_exmode(lockres), | |
2647 | lock_max_prmode(lockres), | |
2648 | lock_max_exmode(lockres), | |
2649 | lock_refresh(lockres)); | |
2650 | ||
ccd979bd MF |
2651 | /* End the line */ |
2652 | seq_printf(m, "\n"); | |
2653 | return 0; | |
2654 | } | |
2655 | ||
90d99779 | 2656 | static const struct seq_operations ocfs2_dlm_seq_ops = { |
ccd979bd MF |
2657 | .start = ocfs2_dlm_seq_start, |
2658 | .stop = ocfs2_dlm_seq_stop, | |
2659 | .next = ocfs2_dlm_seq_next, | |
2660 | .show = ocfs2_dlm_seq_show, | |
2661 | }; | |
2662 | ||
2663 | static int ocfs2_dlm_debug_release(struct inode *inode, struct file *file) | |
2664 | { | |
2665 | struct seq_file *seq = (struct seq_file *) file->private_data; | |
2666 | struct ocfs2_dlm_seq_priv *priv = seq->private; | |
2667 | struct ocfs2_lock_res *res = &priv->p_iter_res; | |
2668 | ||
2669 | ocfs2_remove_lockres_tracking(res); | |
2670 | ocfs2_put_dlm_debug(priv->p_dlm_debug); | |
2671 | return seq_release_private(inode, file); | |
2672 | } | |
2673 | ||
2674 | static int ocfs2_dlm_debug_open(struct inode *inode, struct file *file) | |
2675 | { | |
2676 | int ret; | |
2677 | struct ocfs2_dlm_seq_priv *priv; | |
2678 | struct seq_file *seq; | |
2679 | struct ocfs2_super *osb; | |
2680 | ||
2681 | priv = kzalloc(sizeof(struct ocfs2_dlm_seq_priv), GFP_KERNEL); | |
2682 | if (!priv) { | |
2683 | ret = -ENOMEM; | |
2684 | mlog_errno(ret); | |
2685 | goto out; | |
2686 | } | |
8e18e294 | 2687 | osb = inode->i_private; |
ccd979bd MF |
2688 | ocfs2_get_dlm_debug(osb->osb_dlm_debug); |
2689 | priv->p_dlm_debug = osb->osb_dlm_debug; | |
2690 | INIT_LIST_HEAD(&priv->p_iter_res.l_debug_list); | |
2691 | ||
2692 | ret = seq_open(file, &ocfs2_dlm_seq_ops); | |
2693 | if (ret) { | |
2694 | kfree(priv); | |
2695 | mlog_errno(ret); | |
2696 | goto out; | |
2697 | } | |
2698 | ||
2699 | seq = (struct seq_file *) file->private_data; | |
2700 | seq->private = priv; | |
2701 | ||
2702 | ocfs2_add_lockres_tracking(&priv->p_iter_res, | |
2703 | priv->p_dlm_debug); | |
2704 | ||
2705 | out: | |
2706 | return ret; | |
2707 | } | |
2708 | ||
4b6f5d20 | 2709 | static const struct file_operations ocfs2_dlm_debug_fops = { |
ccd979bd MF |
2710 | .open = ocfs2_dlm_debug_open, |
2711 | .release = ocfs2_dlm_debug_release, | |
2712 | .read = seq_read, | |
2713 | .llseek = seq_lseek, | |
2714 | }; | |
2715 | ||
2716 | static int ocfs2_dlm_init_debug(struct ocfs2_super *osb) | |
2717 | { | |
2718 | int ret = 0; | |
2719 | struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug; | |
2720 | ||
2721 | dlm_debug->d_locking_state = debugfs_create_file("locking_state", | |
2722 | S_IFREG|S_IRUSR, | |
2723 | osb->osb_debug_root, | |
2724 | osb, | |
2725 | &ocfs2_dlm_debug_fops); | |
2726 | if (!dlm_debug->d_locking_state) { | |
2727 | ret = -EINVAL; | |
2728 | mlog(ML_ERROR, | |
2729 | "Unable to create locking state debugfs file.\n"); | |
2730 | goto out; | |
2731 | } | |
2732 | ||
2733 | ocfs2_get_dlm_debug(dlm_debug); | |
2734 | out: | |
2735 | return ret; | |
2736 | } | |
2737 | ||
2738 | static void ocfs2_dlm_shutdown_debug(struct ocfs2_super *osb) | |
2739 | { | |
2740 | struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug; | |
2741 | ||
2742 | if (dlm_debug) { | |
2743 | debugfs_remove(dlm_debug->d_locking_state); | |
2744 | ocfs2_put_dlm_debug(dlm_debug); | |
2745 | } | |
2746 | } | |
2747 | ||
2748 | int ocfs2_dlm_init(struct ocfs2_super *osb) | |
2749 | { | |
c271c5c2 | 2750 | int status = 0; |
4670c46d | 2751 | struct ocfs2_cluster_connection *conn = NULL; |
ccd979bd MF |
2752 | |
2753 | mlog_entry_void(); | |
2754 | ||
0abd6d18 MF |
2755 | if (ocfs2_mount_local(osb)) { |
2756 | osb->node_num = 0; | |
c271c5c2 | 2757 | goto local; |
0abd6d18 | 2758 | } |
c271c5c2 | 2759 | |
ccd979bd MF |
2760 | status = ocfs2_dlm_init_debug(osb); |
2761 | if (status < 0) { | |
2762 | mlog_errno(status); | |
2763 | goto bail; | |
2764 | } | |
2765 | ||
34d024f8 MF |
2766 | /* launch downconvert thread */ |
2767 | osb->dc_task = kthread_run(ocfs2_downconvert_thread, osb, "ocfs2dc"); | |
2768 | if (IS_ERR(osb->dc_task)) { | |
2769 | status = PTR_ERR(osb->dc_task); | |
2770 | osb->dc_task = NULL; | |
ccd979bd MF |
2771 | mlog_errno(status); |
2772 | goto bail; | |
2773 | } | |
2774 | ||
ccd979bd | 2775 | /* for now, uuid == domain */ |
9c6c877c JB |
2776 | status = ocfs2_cluster_connect(osb->osb_cluster_stack, |
2777 | osb->uuid_str, | |
4670c46d JB |
2778 | strlen(osb->uuid_str), |
2779 | ocfs2_do_node_down, osb, | |
2780 | &conn); | |
2781 | if (status) { | |
ccd979bd MF |
2782 | mlog_errno(status); |
2783 | goto bail; | |
2784 | } | |
2785 | ||
0abd6d18 MF |
2786 | status = ocfs2_cluster_this_node(&osb->node_num); |
2787 | if (status < 0) { | |
2788 | mlog_errno(status); | |
2789 | mlog(ML_ERROR, | |
2790 | "could not find this host's node number\n"); | |
286eaa95 | 2791 | ocfs2_cluster_disconnect(conn, 0); |
0abd6d18 MF |
2792 | goto bail; |
2793 | } | |
2794 | ||
c271c5c2 | 2795 | local: |
ccd979bd MF |
2796 | ocfs2_super_lock_res_init(&osb->osb_super_lockres, osb); |
2797 | ocfs2_rename_lock_res_init(&osb->osb_rename_lockres, osb); | |
2798 | ||
4670c46d | 2799 | osb->cconn = conn; |
ccd979bd MF |
2800 | |
2801 | status = 0; | |
2802 | bail: | |
2803 | if (status < 0) { | |
2804 | ocfs2_dlm_shutdown_debug(osb); | |
34d024f8 MF |
2805 | if (osb->dc_task) |
2806 | kthread_stop(osb->dc_task); | |
ccd979bd MF |
2807 | } |
2808 | ||
2809 | mlog_exit(status); | |
2810 | return status; | |
2811 | } | |
2812 | ||
286eaa95 JB |
2813 | void ocfs2_dlm_shutdown(struct ocfs2_super *osb, |
2814 | int hangup_pending) | |
ccd979bd MF |
2815 | { |
2816 | mlog_entry_void(); | |
2817 | ||
ccd979bd MF |
2818 | ocfs2_drop_osb_locks(osb); |
2819 | ||
4670c46d JB |
2820 | /* |
2821 | * Now that we have dropped all locks and ocfs2_dismount_volume() | |
2822 | * has disabled recovery, the DLM won't be talking to us. It's | |
2823 | * safe to tear things down before disconnecting the cluster. | |
2824 | */ | |
2825 | ||
34d024f8 MF |
2826 | if (osb->dc_task) { |
2827 | kthread_stop(osb->dc_task); | |
2828 | osb->dc_task = NULL; | |
ccd979bd MF |
2829 | } |
2830 | ||
2831 | ocfs2_lock_res_free(&osb->osb_super_lockres); | |
2832 | ocfs2_lock_res_free(&osb->osb_rename_lockres); | |
2833 | ||
286eaa95 | 2834 | ocfs2_cluster_disconnect(osb->cconn, hangup_pending); |
4670c46d | 2835 | osb->cconn = NULL; |
ccd979bd MF |
2836 | |
2837 | ocfs2_dlm_shutdown_debug(osb); | |
2838 | ||
2839 | mlog_exit_void(); | |
2840 | } | |
2841 | ||
7431cd7e | 2842 | static void ocfs2_unlock_ast(void *opaque, int error) |
ccd979bd MF |
2843 | { |
2844 | struct ocfs2_lock_res *lockres = opaque; | |
2845 | unsigned long flags; | |
2846 | ||
2847 | mlog_entry_void(); | |
2848 | ||
2849 | mlog(0, "UNLOCK AST called on lock %s, action = %d\n", lockres->l_name, | |
2850 | lockres->l_unlock_action); | |
2851 | ||
2852 | spin_lock_irqsave(&lockres->l_lock, flags); | |
de551246 | 2853 | if (error) { |
7431cd7e JB |
2854 | mlog(ML_ERROR, "Dlm passes error %d for lock %s, " |
2855 | "unlock_action %d\n", error, lockres->l_name, | |
ccd979bd MF |
2856 | lockres->l_unlock_action); |
2857 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2858 | return; | |
2859 | } | |
2860 | ||
2861 | switch(lockres->l_unlock_action) { | |
2862 | case OCFS2_UNLOCK_CANCEL_CONVERT: | |
2863 | mlog(0, "Cancel convert success for %s\n", lockres->l_name); | |
2864 | lockres->l_action = OCFS2_AST_INVALID; | |
2865 | break; | |
2866 | case OCFS2_UNLOCK_DROP_LOCK: | |
bd3e7610 | 2867 | lockres->l_level = DLM_LOCK_IV; |
ccd979bd MF |
2868 | break; |
2869 | default: | |
2870 | BUG(); | |
2871 | } | |
2872 | ||
2873 | lockres_clear_flags(lockres, OCFS2_LOCK_BUSY); | |
ccd979bd | 2874 | lockres->l_unlock_action = OCFS2_UNLOCK_INVALID; |
ccd979bd | 2875 | wake_up(&lockres->l_event); |
07f9eebc | 2876 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
ccd979bd MF |
2877 | |
2878 | mlog_exit_void(); | |
2879 | } | |
2880 | ||
ccd979bd | 2881 | static int ocfs2_drop_lock(struct ocfs2_super *osb, |
0d5dc6c2 | 2882 | struct ocfs2_lock_res *lockres) |
ccd979bd | 2883 | { |
7431cd7e | 2884 | int ret; |
ccd979bd | 2885 | unsigned long flags; |
bd3e7610 | 2886 | u32 lkm_flags = 0; |
ccd979bd MF |
2887 | |
2888 | /* We didn't get anywhere near actually using this lockres. */ | |
2889 | if (!(lockres->l_flags & OCFS2_LOCK_INITIALIZED)) | |
2890 | goto out; | |
2891 | ||
b80fc012 | 2892 | if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) |
bd3e7610 | 2893 | lkm_flags |= DLM_LKF_VALBLK; |
b80fc012 | 2894 | |
ccd979bd MF |
2895 | spin_lock_irqsave(&lockres->l_lock, flags); |
2896 | ||
2897 | mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_FREEING), | |
2898 | "lockres %s, flags 0x%lx\n", | |
2899 | lockres->l_name, lockres->l_flags); | |
2900 | ||
2901 | while (lockres->l_flags & OCFS2_LOCK_BUSY) { | |
2902 | mlog(0, "waiting on busy lock \"%s\": flags = %lx, action = " | |
2903 | "%u, unlock_action = %u\n", | |
2904 | lockres->l_name, lockres->l_flags, lockres->l_action, | |
2905 | lockres->l_unlock_action); | |
2906 | ||
2907 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2908 | ||
2909 | /* XXX: Today we just wait on any busy | |
2910 | * locks... Perhaps we need to cancel converts in the | |
2911 | * future? */ | |
2912 | ocfs2_wait_on_busy_lock(lockres); | |
2913 | ||
2914 | spin_lock_irqsave(&lockres->l_lock, flags); | |
2915 | } | |
2916 | ||
0d5dc6c2 MF |
2917 | if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) { |
2918 | if (lockres->l_flags & OCFS2_LOCK_ATTACHED && | |
bd3e7610 | 2919 | lockres->l_level == DLM_LOCK_EX && |
0d5dc6c2 MF |
2920 | !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) |
2921 | lockres->l_ops->set_lvb(lockres); | |
2922 | } | |
ccd979bd MF |
2923 | |
2924 | if (lockres->l_flags & OCFS2_LOCK_BUSY) | |
2925 | mlog(ML_ERROR, "destroying busy lock: \"%s\"\n", | |
2926 | lockres->l_name); | |
2927 | if (lockres->l_flags & OCFS2_LOCK_BLOCKED) | |
2928 | mlog(0, "destroying blocked lock: \"%s\"\n", lockres->l_name); | |
2929 | ||
2930 | if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) { | |
2931 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2932 | goto out; | |
2933 | } | |
2934 | ||
2935 | lockres_clear_flags(lockres, OCFS2_LOCK_ATTACHED); | |
2936 | ||
2937 | /* make sure we never get here while waiting for an ast to | |
2938 | * fire. */ | |
2939 | BUG_ON(lockres->l_action != OCFS2_AST_INVALID); | |
2940 | ||
2941 | /* is this necessary? */ | |
2942 | lockres_or_flags(lockres, OCFS2_LOCK_BUSY); | |
2943 | lockres->l_unlock_action = OCFS2_UNLOCK_DROP_LOCK; | |
2944 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2945 | ||
2946 | mlog(0, "lock %s\n", lockres->l_name); | |
2947 | ||
4670c46d | 2948 | ret = ocfs2_dlm_unlock(osb->cconn, &lockres->l_lksb, lkm_flags, |
7431cd7e JB |
2949 | lockres); |
2950 | if (ret) { | |
2951 | ocfs2_log_dlm_error("ocfs2_dlm_unlock", ret, lockres); | |
ccd979bd | 2952 | mlog(ML_ERROR, "lockres flags: %lu\n", lockres->l_flags); |
cf0acdcd | 2953 | ocfs2_dlm_dump_lksb(&lockres->l_lksb); |
ccd979bd MF |
2954 | BUG(); |
2955 | } | |
24ef1815 | 2956 | mlog(0, "lock %s, successfull return from ocfs2_dlm_unlock\n", |
ccd979bd MF |
2957 | lockres->l_name); |
2958 | ||
2959 | ocfs2_wait_on_busy_lock(lockres); | |
2960 | out: | |
2961 | mlog_exit(0); | |
2962 | return 0; | |
2963 | } | |
2964 | ||
2965 | /* Mark the lockres as being dropped. It will no longer be | |
2966 | * queued if blocking, but we still may have to wait on it | |
34d024f8 | 2967 | * being dequeued from the downconvert thread before we can consider |
ccd979bd MF |
2968 | * it safe to drop. |
2969 | * | |
2970 | * You can *not* attempt to call cluster_lock on this lockres anymore. */ | |
2971 | void ocfs2_mark_lockres_freeing(struct ocfs2_lock_res *lockres) | |
2972 | { | |
2973 | int status; | |
2974 | struct ocfs2_mask_waiter mw; | |
2975 | unsigned long flags; | |
2976 | ||
2977 | ocfs2_init_mask_waiter(&mw); | |
2978 | ||
2979 | spin_lock_irqsave(&lockres->l_lock, flags); | |
2980 | lockres->l_flags |= OCFS2_LOCK_FREEING; | |
2981 | while (lockres->l_flags & OCFS2_LOCK_QUEUED) { | |
2982 | lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_QUEUED, 0); | |
2983 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2984 | ||
2985 | mlog(0, "Waiting on lockres %s\n", lockres->l_name); | |
2986 | ||
2987 | status = ocfs2_wait_for_mask(&mw); | |
2988 | if (status) | |
2989 | mlog_errno(status); | |
2990 | ||
2991 | spin_lock_irqsave(&lockres->l_lock, flags); | |
2992 | } | |
2993 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
2994 | } | |
2995 | ||
d680efe9 MF |
2996 | void ocfs2_simple_drop_lockres(struct ocfs2_super *osb, |
2997 | struct ocfs2_lock_res *lockres) | |
ccd979bd | 2998 | { |
d680efe9 | 2999 | int ret; |
ccd979bd | 3000 | |
d680efe9 | 3001 | ocfs2_mark_lockres_freeing(lockres); |
0d5dc6c2 | 3002 | ret = ocfs2_drop_lock(osb, lockres); |
d680efe9 MF |
3003 | if (ret) |
3004 | mlog_errno(ret); | |
3005 | } | |
ccd979bd | 3006 | |
d680efe9 MF |
3007 | static void ocfs2_drop_osb_locks(struct ocfs2_super *osb) |
3008 | { | |
3009 | ocfs2_simple_drop_lockres(osb, &osb->osb_super_lockres); | |
3010 | ocfs2_simple_drop_lockres(osb, &osb->osb_rename_lockres); | |
ccd979bd MF |
3011 | } |
3012 | ||
ccd979bd MF |
3013 | int ocfs2_drop_inode_locks(struct inode *inode) |
3014 | { | |
3015 | int status, err; | |
ccd979bd MF |
3016 | |
3017 | mlog_entry_void(); | |
3018 | ||
3019 | /* No need to call ocfs2_mark_lockres_freeing here - | |
3020 | * ocfs2_clear_inode has done it for us. */ | |
3021 | ||
3022 | err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb), | |
50008630 | 3023 | &OCFS2_I(inode)->ip_open_lockres); |
ccd979bd MF |
3024 | if (err < 0) |
3025 | mlog_errno(err); | |
3026 | ||
3027 | status = err; | |
3028 | ||
3029 | err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb), | |
e63aecb6 | 3030 | &OCFS2_I(inode)->ip_inode_lockres); |
ccd979bd MF |
3031 | if (err < 0) |
3032 | mlog_errno(err); | |
3033 | if (err < 0 && !status) | |
3034 | status = err; | |
3035 | ||
3036 | err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb), | |
0d5dc6c2 | 3037 | &OCFS2_I(inode)->ip_rw_lockres); |
ccd979bd MF |
3038 | if (err < 0) |
3039 | mlog_errno(err); | |
3040 | if (err < 0 && !status) | |
3041 | status = err; | |
3042 | ||
3043 | mlog_exit(status); | |
3044 | return status; | |
3045 | } | |
3046 | ||
de551246 JB |
3047 | static unsigned int ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres, |
3048 | int new_level) | |
ccd979bd MF |
3049 | { |
3050 | assert_spin_locked(&lockres->l_lock); | |
3051 | ||
bd3e7610 | 3052 | BUG_ON(lockres->l_blocking <= DLM_LOCK_NL); |
ccd979bd MF |
3053 | |
3054 | if (lockres->l_level <= new_level) { | |
bd3e7610 | 3055 | mlog(ML_ERROR, "lockres->l_level (%d) <= new_level (%d)\n", |
ccd979bd MF |
3056 | lockres->l_level, new_level); |
3057 | BUG(); | |
3058 | } | |
3059 | ||
3060 | mlog(0, "lock %s, new_level = %d, l_blocking = %d\n", | |
3061 | lockres->l_name, new_level, lockres->l_blocking); | |
3062 | ||
3063 | lockres->l_action = OCFS2_AST_DOWNCONVERT; | |
3064 | lockres->l_requested = new_level; | |
3065 | lockres_or_flags(lockres, OCFS2_LOCK_BUSY); | |
de551246 | 3066 | return lockres_set_pending(lockres); |
ccd979bd MF |
3067 | } |
3068 | ||
3069 | static int ocfs2_downconvert_lock(struct ocfs2_super *osb, | |
3070 | struct ocfs2_lock_res *lockres, | |
3071 | int new_level, | |
de551246 JB |
3072 | int lvb, |
3073 | unsigned int generation) | |
ccd979bd | 3074 | { |
bd3e7610 JB |
3075 | int ret; |
3076 | u32 dlm_flags = DLM_LKF_CONVERT; | |
ccd979bd MF |
3077 | |
3078 | mlog_entry_void(); | |
3079 | ||
3080 | if (lvb) | |
bd3e7610 | 3081 | dlm_flags |= DLM_LKF_VALBLK; |
ccd979bd | 3082 | |
4670c46d | 3083 | ret = ocfs2_dlm_lock(osb->cconn, |
7431cd7e JB |
3084 | new_level, |
3085 | &lockres->l_lksb, | |
3086 | dlm_flags, | |
3087 | lockres->l_name, | |
3088 | OCFS2_LOCK_ID_MAX_LEN - 1, | |
3089 | lockres); | |
de551246 | 3090 | lockres_clear_pending(lockres, generation, osb); |
7431cd7e JB |
3091 | if (ret) { |
3092 | ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres); | |
ccd979bd MF |
3093 | ocfs2_recover_from_dlm_error(lockres, 1); |
3094 | goto bail; | |
3095 | } | |
3096 | ||
3097 | ret = 0; | |
3098 | bail: | |
3099 | mlog_exit(ret); | |
3100 | return ret; | |
3101 | } | |
3102 | ||
24ef1815 | 3103 | /* returns 1 when the caller should unlock and call ocfs2_dlm_unlock */ |
ccd979bd MF |
3104 | static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb, |
3105 | struct ocfs2_lock_res *lockres) | |
3106 | { | |
3107 | assert_spin_locked(&lockres->l_lock); | |
3108 | ||
3109 | mlog_entry_void(); | |
3110 | mlog(0, "lock %s\n", lockres->l_name); | |
3111 | ||
3112 | if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) { | |
3113 | /* If we're already trying to cancel a lock conversion | |
3114 | * then just drop the spinlock and allow the caller to | |
3115 | * requeue this lock. */ | |
3116 | ||
3117 | mlog(0, "Lockres %s, skip convert\n", lockres->l_name); | |
3118 | return 0; | |
3119 | } | |
3120 | ||
3121 | /* were we in a convert when we got the bast fire? */ | |
3122 | BUG_ON(lockres->l_action != OCFS2_AST_CONVERT && | |
3123 | lockres->l_action != OCFS2_AST_DOWNCONVERT); | |
3124 | /* set things up for the unlockast to know to just | |
3125 | * clear out the ast_action and unset busy, etc. */ | |
3126 | lockres->l_unlock_action = OCFS2_UNLOCK_CANCEL_CONVERT; | |
3127 | ||
3128 | mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_BUSY), | |
3129 | "lock %s, invalid flags: 0x%lx\n", | |
3130 | lockres->l_name, lockres->l_flags); | |
3131 | ||
3132 | return 1; | |
3133 | } | |
3134 | ||
3135 | static int ocfs2_cancel_convert(struct ocfs2_super *osb, | |
3136 | struct ocfs2_lock_res *lockres) | |
3137 | { | |
3138 | int ret; | |
ccd979bd MF |
3139 | |
3140 | mlog_entry_void(); | |
3141 | mlog(0, "lock %s\n", lockres->l_name); | |
3142 | ||
4670c46d | 3143 | ret = ocfs2_dlm_unlock(osb->cconn, &lockres->l_lksb, |
7431cd7e JB |
3144 | DLM_LKF_CANCEL, lockres); |
3145 | if (ret) { | |
3146 | ocfs2_log_dlm_error("ocfs2_dlm_unlock", ret, lockres); | |
ccd979bd MF |
3147 | ocfs2_recover_from_dlm_error(lockres, 0); |
3148 | } | |
3149 | ||
24ef1815 | 3150 | mlog(0, "lock %s return from ocfs2_dlm_unlock\n", lockres->l_name); |
ccd979bd | 3151 | |
ccd979bd MF |
3152 | mlog_exit(ret); |
3153 | return ret; | |
3154 | } | |
3155 | ||
b5e500e2 MF |
3156 | static int ocfs2_unblock_lock(struct ocfs2_super *osb, |
3157 | struct ocfs2_lock_res *lockres, | |
3158 | struct ocfs2_unblock_ctl *ctl) | |
ccd979bd MF |
3159 | { |
3160 | unsigned long flags; | |
3161 | int blocking; | |
3162 | int new_level; | |
3163 | int ret = 0; | |
5ef0d4ea | 3164 | int set_lvb = 0; |
de551246 | 3165 | unsigned int gen; |
ccd979bd MF |
3166 | |
3167 | mlog_entry_void(); | |
3168 | ||
3169 | spin_lock_irqsave(&lockres->l_lock, flags); | |
3170 | ||
3171 | BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED)); | |
3172 | ||
3173 | recheck: | |
3174 | if (lockres->l_flags & OCFS2_LOCK_BUSY) { | |
de551246 JB |
3175 | /* XXX |
3176 | * This is a *big* race. The OCFS2_LOCK_PENDING flag | |
3177 | * exists entirely for one reason - another thread has set | |
3178 | * OCFS2_LOCK_BUSY, but has *NOT* yet called dlm_lock(). | |
3179 | * | |
3180 | * If we do ocfs2_cancel_convert() before the other thread | |
3181 | * calls dlm_lock(), our cancel will do nothing. We will | |
3182 | * get no ast, and we will have no way of knowing the | |
3183 | * cancel failed. Meanwhile, the other thread will call | |
3184 | * into dlm_lock() and wait...forever. | |
3185 | * | |
3186 | * Why forever? Because another node has asked for the | |
3187 | * lock first; that's why we're here in unblock_lock(). | |
3188 | * | |
3189 | * The solution is OCFS2_LOCK_PENDING. When PENDING is | |
3190 | * set, we just requeue the unblock. Only when the other | |
3191 | * thread has called dlm_lock() and cleared PENDING will | |
3192 | * we then cancel their request. | |
3193 | * | |
3194 | * All callers of dlm_lock() must set OCFS2_DLM_PENDING | |
3195 | * at the same time they set OCFS2_DLM_BUSY. They must | |
3196 | * clear OCFS2_DLM_PENDING after dlm_lock() returns. | |
3197 | */ | |
3198 | if (lockres->l_flags & OCFS2_LOCK_PENDING) | |
3199 | goto leave_requeue; | |
3200 | ||
d680efe9 | 3201 | ctl->requeue = 1; |
ccd979bd MF |
3202 | ret = ocfs2_prepare_cancel_convert(osb, lockres); |
3203 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
3204 | if (ret) { | |
3205 | ret = ocfs2_cancel_convert(osb, lockres); | |
3206 | if (ret < 0) | |
3207 | mlog_errno(ret); | |
3208 | } | |
3209 | goto leave; | |
3210 | } | |
3211 | ||
3212 | /* if we're blocking an exclusive and we have *any* holders, | |
3213 | * then requeue. */ | |
bd3e7610 | 3214 | if ((lockres->l_blocking == DLM_LOCK_EX) |
f7fbfdd1 MF |
3215 | && (lockres->l_ex_holders || lockres->l_ro_holders)) |
3216 | goto leave_requeue; | |
ccd979bd MF |
3217 | |
3218 | /* If it's a PR we're blocking, then only | |
3219 | * requeue if we've got any EX holders */ | |
bd3e7610 | 3220 | if (lockres->l_blocking == DLM_LOCK_PR && |
f7fbfdd1 MF |
3221 | lockres->l_ex_holders) |
3222 | goto leave_requeue; | |
3223 | ||
3224 | /* | |
3225 | * Can we get a lock in this state if the holder counts are | |
3226 | * zero? The meta data unblock code used to check this. | |
3227 | */ | |
3228 | if ((lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH) | |
3229 | && (lockres->l_flags & OCFS2_LOCK_REFRESHING)) | |
3230 | goto leave_requeue; | |
ccd979bd | 3231 | |
16d5b956 MF |
3232 | new_level = ocfs2_highest_compat_lock_level(lockres->l_blocking); |
3233 | ||
3234 | if (lockres->l_ops->check_downconvert | |
3235 | && !lockres->l_ops->check_downconvert(lockres, new_level)) | |
3236 | goto leave_requeue; | |
3237 | ||
ccd979bd MF |
3238 | /* If we get here, then we know that there are no more |
3239 | * incompatible holders (and anyone asking for an incompatible | |
3240 | * lock is blocked). We can now downconvert the lock */ | |
cc567d89 | 3241 | if (!lockres->l_ops->downconvert_worker) |
ccd979bd MF |
3242 | goto downconvert; |
3243 | ||
3244 | /* Some lockres types want to do a bit of work before | |
3245 | * downconverting a lock. Allow that here. The worker function | |
3246 | * may sleep, so we save off a copy of what we're blocking as | |
3247 | * it may change while we're not holding the spin lock. */ | |
3248 | blocking = lockres->l_blocking; | |
3249 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
3250 | ||
cc567d89 | 3251 | ctl->unblock_action = lockres->l_ops->downconvert_worker(lockres, blocking); |
d680efe9 MF |
3252 | |
3253 | if (ctl->unblock_action == UNBLOCK_STOP_POST) | |
3254 | goto leave; | |
ccd979bd MF |
3255 | |
3256 | spin_lock_irqsave(&lockres->l_lock, flags); | |
3257 | if (blocking != lockres->l_blocking) { | |
3258 | /* If this changed underneath us, then we can't drop | |
3259 | * it just yet. */ | |
3260 | goto recheck; | |
3261 | } | |
3262 | ||
3263 | downconvert: | |
d680efe9 | 3264 | ctl->requeue = 0; |
ccd979bd | 3265 | |
5ef0d4ea | 3266 | if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) { |
bd3e7610 | 3267 | if (lockres->l_level == DLM_LOCK_EX) |
5ef0d4ea MF |
3268 | set_lvb = 1; |
3269 | ||
3270 | /* | |
3271 | * We only set the lvb if the lock has been fully | |
3272 | * refreshed - otherwise we risk setting stale | |
3273 | * data. Otherwise, there's no need to actually clear | |
3274 | * out the lvb here as it's value is still valid. | |
3275 | */ | |
3276 | if (set_lvb && !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) | |
3277 | lockres->l_ops->set_lvb(lockres); | |
3278 | } | |
3279 | ||
de551246 | 3280 | gen = ocfs2_prepare_downconvert(lockres, new_level); |
ccd979bd | 3281 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
de551246 JB |
3282 | ret = ocfs2_downconvert_lock(osb, lockres, new_level, set_lvb, |
3283 | gen); | |
3284 | ||
ccd979bd MF |
3285 | leave: |
3286 | mlog_exit(ret); | |
3287 | return ret; | |
f7fbfdd1 MF |
3288 | |
3289 | leave_requeue: | |
3290 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
3291 | ctl->requeue = 1; | |
3292 | ||
3293 | mlog_exit(0); | |
3294 | return 0; | |
ccd979bd MF |
3295 | } |
3296 | ||
d680efe9 MF |
3297 | static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres, |
3298 | int blocking) | |
ccd979bd MF |
3299 | { |
3300 | struct inode *inode; | |
3301 | struct address_space *mapping; | |
3302 | ||
ccd979bd MF |
3303 | inode = ocfs2_lock_res_inode(lockres); |
3304 | mapping = inode->i_mapping; | |
3305 | ||
1044e401 | 3306 | if (!S_ISREG(inode->i_mode)) |
f1f54068 MF |
3307 | goto out; |
3308 | ||
7f4a2a97 MF |
3309 | /* |
3310 | * We need this before the filemap_fdatawrite() so that it can | |
3311 | * transfer the dirty bit from the PTE to the | |
3312 | * page. Unfortunately this means that even for EX->PR | |
3313 | * downconverts, we'll lose our mappings and have to build | |
3314 | * them up again. | |
3315 | */ | |
3316 | unmap_mapping_range(mapping, 0, 0, 0); | |
3317 | ||
ccd979bd | 3318 | if (filemap_fdatawrite(mapping)) { |
b0697053 MF |
3319 | mlog(ML_ERROR, "Could not sync inode %llu for downconvert!", |
3320 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
ccd979bd MF |
3321 | } |
3322 | sync_mapping_buffers(mapping); | |
bd3e7610 | 3323 | if (blocking == DLM_LOCK_EX) { |
ccd979bd | 3324 | truncate_inode_pages(mapping, 0); |
ccd979bd MF |
3325 | } else { |
3326 | /* We only need to wait on the I/O if we're not also | |
3327 | * truncating pages because truncate_inode_pages waits | |
3328 | * for us above. We don't truncate pages if we're | |
3329 | * blocking anything < EXMODE because we want to keep | |
3330 | * them around in that case. */ | |
3331 | filemap_fdatawait(mapping); | |
3332 | } | |
3333 | ||
f1f54068 | 3334 | out: |
d680efe9 | 3335 | return UNBLOCK_CONTINUE; |
ccd979bd MF |
3336 | } |
3337 | ||
810d5aeb MF |
3338 | static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres, |
3339 | int new_level) | |
3340 | { | |
3341 | struct inode *inode = ocfs2_lock_res_inode(lockres); | |
3342 | int checkpointed = ocfs2_inode_fully_checkpointed(inode); | |
3343 | ||
bd3e7610 JB |
3344 | BUG_ON(new_level != DLM_LOCK_NL && new_level != DLM_LOCK_PR); |
3345 | BUG_ON(lockres->l_level != DLM_LOCK_EX && !checkpointed); | |
810d5aeb MF |
3346 | |
3347 | if (checkpointed) | |
3348 | return 1; | |
3349 | ||
3350 | ocfs2_start_checkpoint(OCFS2_SB(inode->i_sb)); | |
3351 | return 0; | |
3352 | } | |
3353 | ||
3354 | static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres) | |
3355 | { | |
3356 | struct inode *inode = ocfs2_lock_res_inode(lockres); | |
3357 | ||
3358 | __ocfs2_stuff_meta_lvb(inode); | |
3359 | } | |
3360 | ||
d680efe9 MF |
3361 | /* |
3362 | * Does the final reference drop on our dentry lock. Right now this | |
34d024f8 | 3363 | * happens in the downconvert thread, but we could choose to simplify the |
d680efe9 MF |
3364 | * dlmglue API and push these off to the ocfs2_wq in the future. |
3365 | */ | |
3366 | static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb, | |
3367 | struct ocfs2_lock_res *lockres) | |
3368 | { | |
3369 | struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres); | |
3370 | ocfs2_dentry_lock_put(osb, dl); | |
3371 | } | |
3372 | ||
3373 | /* | |
3374 | * d_delete() matching dentries before the lock downconvert. | |
3375 | * | |
3376 | * At this point, any process waiting to destroy the | |
3377 | * dentry_lock due to last ref count is stopped by the | |
3378 | * OCFS2_LOCK_QUEUED flag. | |
3379 | * | |
3380 | * We have two potential problems | |
3381 | * | |
3382 | * 1) If we do the last reference drop on our dentry_lock (via dput) | |
3383 | * we'll wind up in ocfs2_release_dentry_lock(), waiting on | |
3384 | * the downconvert to finish. Instead we take an elevated | |
3385 | * reference and push the drop until after we've completed our | |
3386 | * unblock processing. | |
3387 | * | |
3388 | * 2) There might be another process with a final reference, | |
3389 | * waiting on us to finish processing. If this is the case, we | |
3390 | * detect it and exit out - there's no more dentries anyway. | |
3391 | */ | |
3392 | static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres, | |
3393 | int blocking) | |
3394 | { | |
3395 | struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres); | |
3396 | struct ocfs2_inode_info *oi = OCFS2_I(dl->dl_inode); | |
3397 | struct dentry *dentry; | |
3398 | unsigned long flags; | |
3399 | int extra_ref = 0; | |
3400 | ||
3401 | /* | |
3402 | * This node is blocking another node from getting a read | |
3403 | * lock. This happens when we've renamed within a | |
3404 | * directory. We've forced the other nodes to d_delete(), but | |
3405 | * we never actually dropped our lock because it's still | |
3406 | * valid. The downconvert code will retain a PR for this node, | |
3407 | * so there's no further work to do. | |
3408 | */ | |
bd3e7610 | 3409 | if (blocking == DLM_LOCK_PR) |
d680efe9 MF |
3410 | return UNBLOCK_CONTINUE; |
3411 | ||
3412 | /* | |
3413 | * Mark this inode as potentially orphaned. The code in | |
3414 | * ocfs2_delete_inode() will figure out whether it actually | |
3415 | * needs to be freed or not. | |
3416 | */ | |
3417 | spin_lock(&oi->ip_lock); | |
3418 | oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED; | |
3419 | spin_unlock(&oi->ip_lock); | |
3420 | ||
3421 | /* | |
3422 | * Yuck. We need to make sure however that the check of | |
3423 | * OCFS2_LOCK_FREEING and the extra reference are atomic with | |
3424 | * respect to a reference decrement or the setting of that | |
3425 | * flag. | |
3426 | */ | |
3427 | spin_lock_irqsave(&lockres->l_lock, flags); | |
3428 | spin_lock(&dentry_attach_lock); | |
3429 | if (!(lockres->l_flags & OCFS2_LOCK_FREEING) | |
3430 | && dl->dl_count) { | |
3431 | dl->dl_count++; | |
3432 | extra_ref = 1; | |
3433 | } | |
3434 | spin_unlock(&dentry_attach_lock); | |
3435 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
3436 | ||
3437 | mlog(0, "extra_ref = %d\n", extra_ref); | |
3438 | ||
3439 | /* | |
3440 | * We have a process waiting on us in ocfs2_dentry_iput(), | |
3441 | * which means we can't have any more outstanding | |
3442 | * aliases. There's no need to do any more work. | |
3443 | */ | |
3444 | if (!extra_ref) | |
3445 | return UNBLOCK_CONTINUE; | |
3446 | ||
3447 | spin_lock(&dentry_attach_lock); | |
3448 | while (1) { | |
3449 | dentry = ocfs2_find_local_alias(dl->dl_inode, | |
3450 | dl->dl_parent_blkno, 1); | |
3451 | if (!dentry) | |
3452 | break; | |
3453 | spin_unlock(&dentry_attach_lock); | |
3454 | ||
3455 | mlog(0, "d_delete(%.*s);\n", dentry->d_name.len, | |
3456 | dentry->d_name.name); | |
3457 | ||
3458 | /* | |
3459 | * The following dcache calls may do an | |
3460 | * iput(). Normally we don't want that from the | |
3461 | * downconverting thread, but in this case it's ok | |
3462 | * because the requesting node already has an | |
3463 | * exclusive lock on the inode, so it can't be queued | |
3464 | * for a downconvert. | |
3465 | */ | |
3466 | d_delete(dentry); | |
3467 | dput(dentry); | |
3468 | ||
3469 | spin_lock(&dentry_attach_lock); | |
3470 | } | |
3471 | spin_unlock(&dentry_attach_lock); | |
3472 | ||
3473 | /* | |
3474 | * If we are the last holder of this dentry lock, there is no | |
3475 | * reason to downconvert so skip straight to the unlock. | |
3476 | */ | |
3477 | if (dl->dl_count == 1) | |
3478 | return UNBLOCK_STOP_POST; | |
3479 | ||
3480 | return UNBLOCK_CONTINUE_POST; | |
3481 | } | |
3482 | ||
9e33d69f JK |
3483 | static void ocfs2_set_qinfo_lvb(struct ocfs2_lock_res *lockres) |
3484 | { | |
3485 | struct ocfs2_qinfo_lvb *lvb; | |
3486 | struct ocfs2_mem_dqinfo *oinfo = ocfs2_lock_res_qinfo(lockres); | |
3487 | struct mem_dqinfo *info = sb_dqinfo(oinfo->dqi_gi.dqi_sb, | |
3488 | oinfo->dqi_gi.dqi_type); | |
3489 | ||
3490 | mlog_entry_void(); | |
3491 | ||
3492 | lvb = (struct ocfs2_qinfo_lvb *)ocfs2_dlm_lvb(&lockres->l_lksb); | |
3493 | lvb->lvb_version = OCFS2_QINFO_LVB_VERSION; | |
3494 | lvb->lvb_bgrace = cpu_to_be32(info->dqi_bgrace); | |
3495 | lvb->lvb_igrace = cpu_to_be32(info->dqi_igrace); | |
3496 | lvb->lvb_syncms = cpu_to_be32(oinfo->dqi_syncms); | |
3497 | lvb->lvb_blocks = cpu_to_be32(oinfo->dqi_gi.dqi_blocks); | |
3498 | lvb->lvb_free_blk = cpu_to_be32(oinfo->dqi_gi.dqi_free_blk); | |
3499 | lvb->lvb_free_entry = cpu_to_be32(oinfo->dqi_gi.dqi_free_entry); | |
3500 | ||
3501 | mlog_exit_void(); | |
3502 | } | |
3503 | ||
3504 | void ocfs2_qinfo_unlock(struct ocfs2_mem_dqinfo *oinfo, int ex) | |
3505 | { | |
3506 | struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock; | |
3507 | struct ocfs2_super *osb = OCFS2_SB(oinfo->dqi_gi.dqi_sb); | |
3508 | int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR; | |
3509 | ||
3510 | mlog_entry_void(); | |
3511 | if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb)) | |
3512 | ocfs2_cluster_unlock(osb, lockres, level); | |
3513 | mlog_exit_void(); | |
3514 | } | |
3515 | ||
3516 | static int ocfs2_refresh_qinfo(struct ocfs2_mem_dqinfo *oinfo) | |
3517 | { | |
3518 | struct mem_dqinfo *info = sb_dqinfo(oinfo->dqi_gi.dqi_sb, | |
3519 | oinfo->dqi_gi.dqi_type); | |
3520 | struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock; | |
3521 | struct ocfs2_qinfo_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb); | |
3522 | struct buffer_head *bh; | |
3523 | struct ocfs2_global_disk_dqinfo *gdinfo; | |
3524 | int status = 0; | |
3525 | ||
3526 | if (lvb->lvb_version == OCFS2_QINFO_LVB_VERSION) { | |
3527 | info->dqi_bgrace = be32_to_cpu(lvb->lvb_bgrace); | |
3528 | info->dqi_igrace = be32_to_cpu(lvb->lvb_igrace); | |
3529 | oinfo->dqi_syncms = be32_to_cpu(lvb->lvb_syncms); | |
3530 | oinfo->dqi_gi.dqi_blocks = be32_to_cpu(lvb->lvb_blocks); | |
3531 | oinfo->dqi_gi.dqi_free_blk = be32_to_cpu(lvb->lvb_free_blk); | |
3532 | oinfo->dqi_gi.dqi_free_entry = | |
3533 | be32_to_cpu(lvb->lvb_free_entry); | |
3534 | } else { | |
3535 | bh = ocfs2_read_quota_block(oinfo->dqi_gqinode, 0, &status); | |
3536 | if (!bh) { | |
3537 | mlog_errno(status); | |
3538 | goto bail; | |
3539 | } | |
3540 | gdinfo = (struct ocfs2_global_disk_dqinfo *) | |
3541 | (bh->b_data + OCFS2_GLOBAL_INFO_OFF); | |
3542 | info->dqi_bgrace = le32_to_cpu(gdinfo->dqi_bgrace); | |
3543 | info->dqi_igrace = le32_to_cpu(gdinfo->dqi_igrace); | |
3544 | oinfo->dqi_syncms = le32_to_cpu(gdinfo->dqi_syncms); | |
3545 | oinfo->dqi_gi.dqi_blocks = le32_to_cpu(gdinfo->dqi_blocks); | |
3546 | oinfo->dqi_gi.dqi_free_blk = le32_to_cpu(gdinfo->dqi_free_blk); | |
3547 | oinfo->dqi_gi.dqi_free_entry = | |
3548 | le32_to_cpu(gdinfo->dqi_free_entry); | |
3549 | brelse(bh); | |
3550 | ocfs2_track_lock_refresh(lockres); | |
3551 | } | |
3552 | ||
3553 | bail: | |
3554 | return status; | |
3555 | } | |
3556 | ||
3557 | /* Lock quota info, this function expects at least shared lock on the quota file | |
3558 | * so that we can safely refresh quota info from disk. */ | |
3559 | int ocfs2_qinfo_lock(struct ocfs2_mem_dqinfo *oinfo, int ex) | |
3560 | { | |
3561 | struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock; | |
3562 | struct ocfs2_super *osb = OCFS2_SB(oinfo->dqi_gi.dqi_sb); | |
3563 | int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR; | |
3564 | int status = 0; | |
3565 | ||
3566 | mlog_entry_void(); | |
3567 | ||
3568 | /* On RO devices, locking really isn't needed... */ | |
3569 | if (ocfs2_is_hard_readonly(osb)) { | |
3570 | if (ex) | |
3571 | status = -EROFS; | |
3572 | goto bail; | |
3573 | } | |
3574 | if (ocfs2_mount_local(osb)) | |
3575 | goto bail; | |
3576 | ||
3577 | status = ocfs2_cluster_lock(osb, lockres, level, 0, 0); | |
3578 | if (status < 0) { | |
3579 | mlog_errno(status); | |
3580 | goto bail; | |
3581 | } | |
3582 | if (!ocfs2_should_refresh_lock_res(lockres)) | |
3583 | goto bail; | |
3584 | /* OK, we have the lock but we need to refresh the quota info */ | |
3585 | status = ocfs2_refresh_qinfo(oinfo); | |
3586 | if (status) | |
3587 | ocfs2_qinfo_unlock(oinfo, ex); | |
3588 | ocfs2_complete_lock_res_refresh(lockres, status); | |
3589 | bail: | |
3590 | mlog_exit(status); | |
3591 | return status; | |
3592 | } | |
3593 | ||
4670c46d JB |
3594 | /* |
3595 | * This is the filesystem locking protocol. It provides the lock handling | |
3596 | * hooks for the underlying DLM. It has a maximum version number. | |
3597 | * The version number allows interoperability with systems running at | |
3598 | * the same major number and an equal or smaller minor number. | |
3599 | * | |
3600 | * Whenever the filesystem does new things with locks (adds or removes a | |
3601 | * lock, orders them differently, does different things underneath a lock), | |
3602 | * the version must be changed. The protocol is negotiated when joining | |
3603 | * the dlm domain. A node may join the domain if its major version is | |
3604 | * identical to all other nodes and its minor version is greater than | |
3605 | * or equal to all other nodes. When its minor version is greater than | |
3606 | * the other nodes, it will run at the minor version specified by the | |
3607 | * other nodes. | |
3608 | * | |
3609 | * If a locking change is made that will not be compatible with older | |
3610 | * versions, the major number must be increased and the minor version set | |
3611 | * to zero. If a change merely adds a behavior that can be disabled when | |
3612 | * speaking to older versions, the minor version must be increased. If a | |
3613 | * change adds a fully backwards compatible change (eg, LVB changes that | |
3614 | * are just ignored by older versions), the version does not need to be | |
3615 | * updated. | |
3616 | */ | |
24ef1815 | 3617 | static struct ocfs2_locking_protocol lproto = { |
4670c46d JB |
3618 | .lp_max_version = { |
3619 | .pv_major = OCFS2_LOCKING_PROTOCOL_MAJOR, | |
3620 | .pv_minor = OCFS2_LOCKING_PROTOCOL_MINOR, | |
3621 | }, | |
24ef1815 JB |
3622 | .lp_lock_ast = ocfs2_locking_ast, |
3623 | .lp_blocking_ast = ocfs2_blocking_ast, | |
3624 | .lp_unlock_ast = ocfs2_unlock_ast, | |
3625 | }; | |
3626 | ||
63e0c48a | 3627 | void ocfs2_set_locking_protocol(void) |
24ef1815 | 3628 | { |
63e0c48a | 3629 | ocfs2_stack_glue_set_locking_protocol(&lproto); |
24ef1815 JB |
3630 | } |
3631 | ||
24ef1815 | 3632 | |
00600056 AB |
3633 | static void ocfs2_process_blocked_lock(struct ocfs2_super *osb, |
3634 | struct ocfs2_lock_res *lockres) | |
ccd979bd MF |
3635 | { |
3636 | int status; | |
d680efe9 | 3637 | struct ocfs2_unblock_ctl ctl = {0, 0,}; |
ccd979bd MF |
3638 | unsigned long flags; |
3639 | ||
3640 | /* Our reference to the lockres in this function can be | |
3641 | * considered valid until we remove the OCFS2_LOCK_QUEUED | |
3642 | * flag. */ | |
3643 | ||
3644 | mlog_entry_void(); | |
3645 | ||
3646 | BUG_ON(!lockres); | |
3647 | BUG_ON(!lockres->l_ops); | |
ccd979bd MF |
3648 | |
3649 | mlog(0, "lockres %s blocked.\n", lockres->l_name); | |
3650 | ||
3651 | /* Detect whether a lock has been marked as going away while | |
34d024f8 | 3652 | * the downconvert thread was processing other things. A lock can |
ccd979bd MF |
3653 | * still be marked with OCFS2_LOCK_FREEING after this check, |
3654 | * but short circuiting here will still save us some | |
3655 | * performance. */ | |
3656 | spin_lock_irqsave(&lockres->l_lock, flags); | |
3657 | if (lockres->l_flags & OCFS2_LOCK_FREEING) | |
3658 | goto unqueue; | |
3659 | spin_unlock_irqrestore(&lockres->l_lock, flags); | |
3660 | ||
b5e500e2 | 3661 | status = ocfs2_unblock_lock(osb, lockres, &ctl); |
ccd979bd MF |
3662 | if (status < 0) |
3663 | mlog_errno(status); | |
3664 | ||
3665 | spin_lock_irqsave(&lockres->l_lock, flags); | |
3666 | unqueue: | |
d680efe9 | 3667 | if (lockres->l_flags & OCFS2_LOCK_FREEING || !ctl.requeue) { |
ccd979bd MF |
3668 | lockres_clear_flags(lockres, OCFS2_LOCK_QUEUED); |
3669 | } else | |
3670 | ocfs2_schedule_blocked_lock(osb, lockres); | |
3671 | ||
3672 | mlog(0, "lockres %s, requeue = %s.\n", lockres->l_name, | |
d680efe9 | 3673 | ctl.requeue ? "yes" : "no"); |
ccd979bd MF |
3674 | spin_unlock_irqrestore(&lockres->l_lock, flags); |
3675 | ||
d680efe9 MF |
3676 | if (ctl.unblock_action != UNBLOCK_CONTINUE |
3677 | && lockres->l_ops->post_unlock) | |
3678 | lockres->l_ops->post_unlock(osb, lockres); | |
3679 | ||
ccd979bd MF |
3680 | mlog_exit_void(); |
3681 | } | |
3682 | ||
3683 | static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb, | |
3684 | struct ocfs2_lock_res *lockres) | |
3685 | { | |
3686 | mlog_entry_void(); | |
3687 | ||
3688 | assert_spin_locked(&lockres->l_lock); | |
3689 | ||
3690 | if (lockres->l_flags & OCFS2_LOCK_FREEING) { | |
3691 | /* Do not schedule a lock for downconvert when it's on | |
3692 | * the way to destruction - any nodes wanting access | |
3693 | * to the resource will get it soon. */ | |
3694 | mlog(0, "Lockres %s won't be scheduled: flags 0x%lx\n", | |
3695 | lockres->l_name, lockres->l_flags); | |
3696 | return; | |
3697 | } | |
3698 | ||
3699 | lockres_or_flags(lockres, OCFS2_LOCK_QUEUED); | |
3700 | ||
34d024f8 | 3701 | spin_lock(&osb->dc_task_lock); |
ccd979bd MF |
3702 | if (list_empty(&lockres->l_blocked_list)) { |
3703 | list_add_tail(&lockres->l_blocked_list, | |
3704 | &osb->blocked_lock_list); | |
3705 | osb->blocked_lock_count++; | |
3706 | } | |
34d024f8 | 3707 | spin_unlock(&osb->dc_task_lock); |
ccd979bd MF |
3708 | |
3709 | mlog_exit_void(); | |
3710 | } | |
34d024f8 MF |
3711 | |
3712 | static void ocfs2_downconvert_thread_do_work(struct ocfs2_super *osb) | |
3713 | { | |
3714 | unsigned long processed; | |
3715 | struct ocfs2_lock_res *lockres; | |
3716 | ||
3717 | mlog_entry_void(); | |
3718 | ||
3719 | spin_lock(&osb->dc_task_lock); | |
3720 | /* grab this early so we know to try again if a state change and | |
3721 | * wake happens part-way through our work */ | |
3722 | osb->dc_work_sequence = osb->dc_wake_sequence; | |
3723 | ||
3724 | processed = osb->blocked_lock_count; | |
3725 | while (processed) { | |
3726 | BUG_ON(list_empty(&osb->blocked_lock_list)); | |
3727 | ||
3728 | lockres = list_entry(osb->blocked_lock_list.next, | |
3729 | struct ocfs2_lock_res, l_blocked_list); | |
3730 | list_del_init(&lockres->l_blocked_list); | |
3731 | osb->blocked_lock_count--; | |
3732 | spin_unlock(&osb->dc_task_lock); | |
3733 | ||
3734 | BUG_ON(!processed); | |
3735 | processed--; | |
3736 | ||
3737 | ocfs2_process_blocked_lock(osb, lockres); | |
3738 | ||
3739 | spin_lock(&osb->dc_task_lock); | |
3740 | } | |
3741 | spin_unlock(&osb->dc_task_lock); | |
3742 | ||
3743 | mlog_exit_void(); | |
3744 | } | |
3745 | ||
3746 | static int ocfs2_downconvert_thread_lists_empty(struct ocfs2_super *osb) | |
3747 | { | |
3748 | int empty = 0; | |
3749 | ||
3750 | spin_lock(&osb->dc_task_lock); | |
3751 | if (list_empty(&osb->blocked_lock_list)) | |
3752 | empty = 1; | |
3753 | ||
3754 | spin_unlock(&osb->dc_task_lock); | |
3755 | return empty; | |
3756 | } | |
3757 | ||
3758 | static int ocfs2_downconvert_thread_should_wake(struct ocfs2_super *osb) | |
3759 | { | |
3760 | int should_wake = 0; | |
3761 | ||
3762 | spin_lock(&osb->dc_task_lock); | |
3763 | if (osb->dc_work_sequence != osb->dc_wake_sequence) | |
3764 | should_wake = 1; | |
3765 | spin_unlock(&osb->dc_task_lock); | |
3766 | ||
3767 | return should_wake; | |
3768 | } | |
3769 | ||
200bfae3 | 3770 | static int ocfs2_downconvert_thread(void *arg) |
34d024f8 MF |
3771 | { |
3772 | int status = 0; | |
3773 | struct ocfs2_super *osb = arg; | |
3774 | ||
3775 | /* only quit once we've been asked to stop and there is no more | |
3776 | * work available */ | |
3777 | while (!(kthread_should_stop() && | |
3778 | ocfs2_downconvert_thread_lists_empty(osb))) { | |
3779 | ||
3780 | wait_event_interruptible(osb->dc_event, | |
3781 | ocfs2_downconvert_thread_should_wake(osb) || | |
3782 | kthread_should_stop()); | |
3783 | ||
3784 | mlog(0, "downconvert_thread: awoken\n"); | |
3785 | ||
3786 | ocfs2_downconvert_thread_do_work(osb); | |
3787 | } | |
3788 | ||
3789 | osb->dc_task = NULL; | |
3790 | return status; | |
3791 | } | |
3792 | ||
3793 | void ocfs2_wake_downconvert_thread(struct ocfs2_super *osb) | |
3794 | { | |
3795 | spin_lock(&osb->dc_task_lock); | |
3796 | /* make sure the voting thread gets a swipe at whatever changes | |
3797 | * the caller may have made to the voting state */ | |
3798 | osb->dc_wake_sequence++; | |
3799 | spin_unlock(&osb->dc_task_lock); | |
3800 | wake_up(&osb->dc_event); | |
3801 | } |