Initiate recovery on node failure
[deliverable/linux.git] / drivers / md / md-cluster.c
1 /*
2 * Copyright (C) 2015, SUSE
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2, or (at your option)
7 * any later version.
8 *
9 */
10
11
12 #include <linux/module.h>
13 #include <linux/dlm.h>
14 #include <linux/sched.h>
15 #include "md.h"
16 #include "bitmap.h"
17 #include "md-cluster.h"
18
19 #define LVB_SIZE 64
20
21 struct dlm_lock_resource {
22 dlm_lockspace_t *ls;
23 struct dlm_lksb lksb;
24 char *name; /* lock name. */
25 uint32_t flags; /* flags to pass to dlm_lock() */
26 struct completion completion; /* completion for synchronized locking */
27 void (*bast)(void *arg, int mode); /* blocking AST function pointer*/
28 struct mddev *mddev; /* pointing back to mddev. */
29 };
30
31 struct suspend_info {
32 int slot;
33 sector_t lo;
34 sector_t hi;
35 struct list_head list;
36 };
37
38 struct resync_info {
39 __le64 lo;
40 __le64 hi;
41 };
42
43 struct md_cluster_info {
44 /* dlm lock space and resources for clustered raid. */
45 dlm_lockspace_t *lockspace;
46 int slot_number;
47 struct completion completion;
48 struct dlm_lock_resource *sb_lock;
49 struct mutex sb_mutex;
50 struct dlm_lock_resource *bitmap_lockres;
51 struct list_head suspend_list;
52 spinlock_t suspend_lock;
53 struct md_thread *recovery_thread;
54 unsigned long recovery_map;
55 };
56
57 static void sync_ast(void *arg)
58 {
59 struct dlm_lock_resource *res;
60
61 res = (struct dlm_lock_resource *) arg;
62 complete(&res->completion);
63 }
64
65 static int dlm_lock_sync(struct dlm_lock_resource *res, int mode)
66 {
67 int ret = 0;
68
69 init_completion(&res->completion);
70 ret = dlm_lock(res->ls, mode, &res->lksb,
71 res->flags, res->name, strlen(res->name),
72 0, sync_ast, res, res->bast);
73 if (ret)
74 return ret;
75 wait_for_completion(&res->completion);
76 return res->lksb.sb_status;
77 }
78
79 static int dlm_unlock_sync(struct dlm_lock_resource *res)
80 {
81 return dlm_lock_sync(res, DLM_LOCK_NL);
82 }
83
84 static struct dlm_lock_resource *lockres_init(struct mddev *mddev,
85 char *name, void (*bastfn)(void *arg, int mode), int with_lvb)
86 {
87 struct dlm_lock_resource *res = NULL;
88 int ret, namelen;
89 struct md_cluster_info *cinfo = mddev->cluster_info;
90
91 res = kzalloc(sizeof(struct dlm_lock_resource), GFP_KERNEL);
92 if (!res)
93 return NULL;
94 res->ls = cinfo->lockspace;
95 res->mddev = mddev;
96 namelen = strlen(name);
97 res->name = kzalloc(namelen + 1, GFP_KERNEL);
98 if (!res->name) {
99 pr_err("md-cluster: Unable to allocate resource name for resource %s\n", name);
100 goto out_err;
101 }
102 strlcpy(res->name, name, namelen + 1);
103 if (with_lvb) {
104 res->lksb.sb_lvbptr = kzalloc(LVB_SIZE, GFP_KERNEL);
105 if (!res->lksb.sb_lvbptr) {
106 pr_err("md-cluster: Unable to allocate LVB for resource %s\n", name);
107 goto out_err;
108 }
109 res->flags = DLM_LKF_VALBLK;
110 }
111
112 if (bastfn)
113 res->bast = bastfn;
114
115 res->flags |= DLM_LKF_EXPEDITE;
116
117 ret = dlm_lock_sync(res, DLM_LOCK_NL);
118 if (ret) {
119 pr_err("md-cluster: Unable to lock NL on new lock resource %s\n", name);
120 goto out_err;
121 }
122 res->flags &= ~DLM_LKF_EXPEDITE;
123 res->flags |= DLM_LKF_CONVERT;
124
125 return res;
126 out_err:
127 kfree(res->lksb.sb_lvbptr);
128 kfree(res->name);
129 kfree(res);
130 return NULL;
131 }
132
133 static void lockres_free(struct dlm_lock_resource *res)
134 {
135 if (!res)
136 return;
137
138 init_completion(&res->completion);
139 dlm_unlock(res->ls, res->lksb.sb_lkid, 0, &res->lksb, res);
140 wait_for_completion(&res->completion);
141
142 kfree(res->name);
143 kfree(res->lksb.sb_lvbptr);
144 kfree(res);
145 }
146
147 static char *pretty_uuid(char *dest, char *src)
148 {
149 int i, len = 0;
150
151 for (i = 0; i < 16; i++) {
152 if (i == 4 || i == 6 || i == 8 || i == 10)
153 len += sprintf(dest + len, "-");
154 len += sprintf(dest + len, "%02x", (__u8)src[i]);
155 }
156 return dest;
157 }
158
159 static void add_resync_info(struct mddev *mddev, struct dlm_lock_resource *lockres,
160 sector_t lo, sector_t hi)
161 {
162 struct resync_info *ri;
163
164 ri = (struct resync_info *)lockres->lksb.sb_lvbptr;
165 ri->lo = cpu_to_le64(lo);
166 ri->hi = cpu_to_le64(hi);
167 }
168
169 static struct suspend_info *read_resync_info(struct mddev *mddev, struct dlm_lock_resource *lockres)
170 {
171 struct resync_info ri;
172 struct suspend_info *s = NULL;
173 sector_t hi = 0;
174
175 dlm_lock_sync(lockres, DLM_LOCK_CR);
176 memcpy(&ri, lockres->lksb.sb_lvbptr, sizeof(struct resync_info));
177 hi = le64_to_cpu(ri.hi);
178 if (ri.hi > 0) {
179 s = kzalloc(sizeof(struct suspend_info), GFP_KERNEL);
180 if (!s)
181 goto out;
182 s->hi = hi;
183 s->lo = le64_to_cpu(ri.lo);
184 }
185 dlm_unlock_sync(lockres);
186 out:
187 return s;
188 }
189
190 void recover_bitmaps(struct md_thread *thread)
191 {
192 struct mddev *mddev = thread->mddev;
193 struct md_cluster_info *cinfo = mddev->cluster_info;
194 struct dlm_lock_resource *bm_lockres;
195 char str[64];
196 int slot, ret;
197 struct suspend_info *s, *tmp;
198 sector_t lo, hi;
199
200 while (cinfo->recovery_map) {
201 slot = fls64((u64)cinfo->recovery_map) - 1;
202
203 /* Clear suspend_area associated with the bitmap */
204 spin_lock_irq(&cinfo->suspend_lock);
205 list_for_each_entry_safe(s, tmp, &cinfo->suspend_list, list)
206 if (slot == s->slot) {
207 list_del(&s->list);
208 kfree(s);
209 }
210 spin_unlock_irq(&cinfo->suspend_lock);
211
212 snprintf(str, 64, "bitmap%04d", slot);
213 bm_lockres = lockres_init(mddev, str, NULL, 1);
214 if (!bm_lockres) {
215 pr_err("md-cluster: Cannot initialize bitmaps\n");
216 goto clear_bit;
217 }
218
219 ret = dlm_lock_sync(bm_lockres, DLM_LOCK_PW);
220 if (ret) {
221 pr_err("md-cluster: Could not DLM lock %s: %d\n",
222 str, ret);
223 goto clear_bit;
224 }
225 ret = bitmap_copy_from_slot(mddev, slot, &lo, &hi);
226 if (ret)
227 pr_err("md-cluster: Could not copy data from bitmap %d\n", slot);
228 dlm_unlock_sync(bm_lockres);
229 clear_bit:
230 clear_bit(slot, &cinfo->recovery_map);
231 }
232 }
233
234 static void recover_prep(void *arg)
235 {
236 }
237
238 static void recover_slot(void *arg, struct dlm_slot *slot)
239 {
240 struct mddev *mddev = arg;
241 struct md_cluster_info *cinfo = mddev->cluster_info;
242
243 pr_info("md-cluster: %s Node %d/%d down. My slot: %d. Initiating recovery.\n",
244 mddev->bitmap_info.cluster_name,
245 slot->nodeid, slot->slot,
246 cinfo->slot_number);
247 set_bit(slot->slot - 1, &cinfo->recovery_map);
248 if (!cinfo->recovery_thread) {
249 cinfo->recovery_thread = md_register_thread(recover_bitmaps,
250 mddev, "recover");
251 if (!cinfo->recovery_thread) {
252 pr_warn("md-cluster: Could not create recovery thread\n");
253 return;
254 }
255 }
256 md_wakeup_thread(cinfo->recovery_thread);
257 }
258
259 static void recover_done(void *arg, struct dlm_slot *slots,
260 int num_slots, int our_slot,
261 uint32_t generation)
262 {
263 struct mddev *mddev = arg;
264 struct md_cluster_info *cinfo = mddev->cluster_info;
265
266 cinfo->slot_number = our_slot;
267 complete(&cinfo->completion);
268 }
269
270 static const struct dlm_lockspace_ops md_ls_ops = {
271 .recover_prep = recover_prep,
272 .recover_slot = recover_slot,
273 .recover_done = recover_done,
274 };
275
276 static int gather_all_resync_info(struct mddev *mddev, int total_slots)
277 {
278 struct md_cluster_info *cinfo = mddev->cluster_info;
279 int i, ret = 0;
280 struct dlm_lock_resource *bm_lockres;
281 struct suspend_info *s;
282 char str[64];
283
284
285 for (i = 0; i < total_slots; i++) {
286 memset(str, '\0', 64);
287 snprintf(str, 64, "bitmap%04d", i);
288 bm_lockres = lockres_init(mddev, str, NULL, 1);
289 if (!bm_lockres)
290 return -ENOMEM;
291 if (i == (cinfo->slot_number - 1))
292 continue;
293
294 bm_lockres->flags |= DLM_LKF_NOQUEUE;
295 ret = dlm_lock_sync(bm_lockres, DLM_LOCK_PW);
296 if (ret == -EAGAIN) {
297 memset(bm_lockres->lksb.sb_lvbptr, '\0', LVB_SIZE);
298 s = read_resync_info(mddev, bm_lockres);
299 if (s) {
300 pr_info("%s:%d Resync[%llu..%llu] in progress on %d\n",
301 __func__, __LINE__,
302 (unsigned long long) s->lo,
303 (unsigned long long) s->hi, i);
304 spin_lock_irq(&cinfo->suspend_lock);
305 s->slot = i;
306 list_add(&s->list, &cinfo->suspend_list);
307 spin_unlock_irq(&cinfo->suspend_lock);
308 }
309 ret = 0;
310 lockres_free(bm_lockres);
311 continue;
312 }
313 if (ret)
314 goto out;
315 /* TODO: Read the disk bitmap sb and check if it needs recovery */
316 dlm_unlock_sync(bm_lockres);
317 lockres_free(bm_lockres);
318 }
319 out:
320 return ret;
321 }
322
323 static int join(struct mddev *mddev, int nodes)
324 {
325 struct md_cluster_info *cinfo;
326 int ret, ops_rv;
327 char str[64];
328
329 if (!try_module_get(THIS_MODULE))
330 return -ENOENT;
331
332 cinfo = kzalloc(sizeof(struct md_cluster_info), GFP_KERNEL);
333 if (!cinfo)
334 return -ENOMEM;
335
336 init_completion(&cinfo->completion);
337
338 mutex_init(&cinfo->sb_mutex);
339 mddev->cluster_info = cinfo;
340
341 memset(str, 0, 64);
342 pretty_uuid(str, mddev->uuid);
343 ret = dlm_new_lockspace(str, mddev->bitmap_info.cluster_name,
344 DLM_LSFL_FS, LVB_SIZE,
345 &md_ls_ops, mddev, &ops_rv, &cinfo->lockspace);
346 if (ret)
347 goto err;
348 wait_for_completion(&cinfo->completion);
349 if (nodes <= cinfo->slot_number) {
350 pr_err("md-cluster: Slot allotted(%d) greater than available slots(%d)", cinfo->slot_number - 1,
351 nodes);
352 ret = -ERANGE;
353 goto err;
354 }
355 cinfo->sb_lock = lockres_init(mddev, "cmd-super",
356 NULL, 0);
357 if (!cinfo->sb_lock) {
358 ret = -ENOMEM;
359 goto err;
360 }
361
362 pr_info("md-cluster: Joined cluster %s slot %d\n", str, cinfo->slot_number);
363 snprintf(str, 64, "bitmap%04d", cinfo->slot_number - 1);
364 cinfo->bitmap_lockres = lockres_init(mddev, str, NULL, 1);
365 if (!cinfo->bitmap_lockres)
366 goto err;
367 if (dlm_lock_sync(cinfo->bitmap_lockres, DLM_LOCK_PW)) {
368 pr_err("Failed to get bitmap lock\n");
369 ret = -EINVAL;
370 goto err;
371 }
372
373 INIT_LIST_HEAD(&cinfo->suspend_list);
374 spin_lock_init(&cinfo->suspend_lock);
375
376 ret = gather_all_resync_info(mddev, nodes);
377 if (ret)
378 goto err;
379
380 return 0;
381 err:
382 lockres_free(cinfo->bitmap_lockres);
383 lockres_free(cinfo->sb_lock);
384 if (cinfo->lockspace)
385 dlm_release_lockspace(cinfo->lockspace, 2);
386 mddev->cluster_info = NULL;
387 kfree(cinfo);
388 module_put(THIS_MODULE);
389 return ret;
390 }
391
392 static int leave(struct mddev *mddev)
393 {
394 struct md_cluster_info *cinfo = mddev->cluster_info;
395
396 if (!cinfo)
397 return 0;
398 md_unregister_thread(&cinfo->recovery_thread);
399 lockres_free(cinfo->sb_lock);
400 lockres_free(cinfo->bitmap_lockres);
401 dlm_release_lockspace(cinfo->lockspace, 2);
402 return 0;
403 }
404
405 /* slot_number(): Returns the MD slot number to use
406 * DLM starts the slot numbers from 1, wheras cluster-md
407 * wants the number to be from zero, so we deduct one
408 */
409 static int slot_number(struct mddev *mddev)
410 {
411 struct md_cluster_info *cinfo = mddev->cluster_info;
412
413 return cinfo->slot_number - 1;
414 }
415
416 static void resync_info_update(struct mddev *mddev, sector_t lo, sector_t hi)
417 {
418 struct md_cluster_info *cinfo = mddev->cluster_info;
419
420 add_resync_info(mddev, cinfo->bitmap_lockres, lo, hi);
421 /* Re-acquire the lock to refresh LVB */
422 dlm_lock_sync(cinfo->bitmap_lockres, DLM_LOCK_PW);
423 }
424
425 static struct md_cluster_operations cluster_ops = {
426 .join = join,
427 .leave = leave,
428 .slot_number = slot_number,
429 .resync_info_update = resync_info_update,
430 };
431
432 static int __init cluster_init(void)
433 {
434 pr_warn("md-cluster: EXPERIMENTAL. Use with caution\n");
435 pr_info("Registering Cluster MD functions\n");
436 register_md_cluster_operations(&cluster_ops, THIS_MODULE);
437 return 0;
438 }
439
440 static void cluster_exit(void)
441 {
442 unregister_md_cluster_operations();
443 }
444
445 module_init(cluster_init);
446 module_exit(cluster_exit);
447 MODULE_LICENSE("GPL");
448 MODULE_DESCRIPTION("Clustering support for MD");
This page took 0.040297 seconds and 6 git commands to generate.