1 // SPDX-License-Identifier: GPL-2.0-only 1 // SPDX-License-Identifier: GPL-2.0-only
2 /********************************************* 2 /******************************************************************************
3 ********************************************** 3 *******************************************************************************
4 ** 4 **
5 ** Copyright (C) 2005-2011 Red Hat, Inc. All 5 ** Copyright (C) 2005-2011 Red Hat, Inc. All rights reserved.
6 ** 6 **
7 ** 7 **
8 ********************************************** 8 *******************************************************************************
9 ********************************************** 9 ******************************************************************************/
10 10
11 #include "dlm_internal.h" 11 #include "dlm_internal.h"
12 #include "lockspace.h" 12 #include "lockspace.h"
13 #include "member.h" 13 #include "member.h"
14 #include "recoverd.h" 14 #include "recoverd.h"
15 #include "recover.h" 15 #include "recover.h"
16 #include "rcom.h" 16 #include "rcom.h"
17 #include "config.h" 17 #include "config.h"
18 #include "midcomms.h" 18 #include "midcomms.h"
19 #include "lowcomms.h" 19 #include "lowcomms.h"
20 20
21 int dlm_slots_version(const struct dlm_header 21 int dlm_slots_version(const struct dlm_header *h)
22 { 22 {
23 if ((le32_to_cpu(h->h_version) & 0x000 23 if ((le32_to_cpu(h->h_version) & 0x0000FFFF) < DLM_HEADER_SLOTS)
24 return 0; 24 return 0;
25 return 1; 25 return 1;
26 } 26 }
27 27
28 void dlm_slot_save(struct dlm_ls *ls, struct d 28 void dlm_slot_save(struct dlm_ls *ls, struct dlm_rcom *rc,
29 struct dlm_member *memb) 29 struct dlm_member *memb)
30 { 30 {
31 struct rcom_config *rf = (struct rcom_ 31 struct rcom_config *rf = (struct rcom_config *)rc->rc_buf;
32 32
33 if (!dlm_slots_version(&rc->rc_header) 33 if (!dlm_slots_version(&rc->rc_header))
34 return; 34 return;
35 35
36 memb->slot = le16_to_cpu(rf->rf_our_sl 36 memb->slot = le16_to_cpu(rf->rf_our_slot);
37 memb->generation = le32_to_cpu(rf->rf_ 37 memb->generation = le32_to_cpu(rf->rf_generation);
38 } 38 }
39 39
40 void dlm_slots_copy_out(struct dlm_ls *ls, str 40 void dlm_slots_copy_out(struct dlm_ls *ls, struct dlm_rcom *rc)
41 { 41 {
42 struct dlm_slot *slot; 42 struct dlm_slot *slot;
43 struct rcom_slot *ro; 43 struct rcom_slot *ro;
44 int i; 44 int i;
45 45
46 ro = (struct rcom_slot *)(rc->rc_buf + 46 ro = (struct rcom_slot *)(rc->rc_buf + sizeof(struct rcom_config));
47 47
48 /* ls_slots array is sparse, but not r 48 /* ls_slots array is sparse, but not rcom_slots */
49 49
50 for (i = 0; i < ls->ls_slots_size; i++ 50 for (i = 0; i < ls->ls_slots_size; i++) {
51 slot = &ls->ls_slots[i]; 51 slot = &ls->ls_slots[i];
52 if (!slot->nodeid) 52 if (!slot->nodeid)
53 continue; 53 continue;
54 ro->ro_nodeid = cpu_to_le32(sl 54 ro->ro_nodeid = cpu_to_le32(slot->nodeid);
55 ro->ro_slot = cpu_to_le16(slot 55 ro->ro_slot = cpu_to_le16(slot->slot);
56 ro++; 56 ro++;
57 } 57 }
58 } 58 }
59 59
60 #define SLOT_DEBUG_LINE 128 60 #define SLOT_DEBUG_LINE 128
61 61
62 static void log_slots(struct dlm_ls *ls, uint3 62 static void log_slots(struct dlm_ls *ls, uint32_t gen, int num_slots,
63 struct rcom_slot *ro0, s 63 struct rcom_slot *ro0, struct dlm_slot *array,
64 int array_size) 64 int array_size)
65 { 65 {
66 char line[SLOT_DEBUG_LINE]; 66 char line[SLOT_DEBUG_LINE];
67 int len = SLOT_DEBUG_LINE - 1; 67 int len = SLOT_DEBUG_LINE - 1;
68 int pos = 0; 68 int pos = 0;
69 int ret, i; 69 int ret, i;
70 70
71 memset(line, 0, sizeof(line)); 71 memset(line, 0, sizeof(line));
72 72
73 if (array) { 73 if (array) {
74 for (i = 0; i < array_size; i+ 74 for (i = 0; i < array_size; i++) {
75 if (!array[i].nodeid) 75 if (!array[i].nodeid)
76 continue; 76 continue;
77 77
78 ret = snprintf(line + 78 ret = snprintf(line + pos, len - pos, " %d:%d",
79 array[i 79 array[i].slot, array[i].nodeid);
80 if (ret >= len - pos) 80 if (ret >= len - pos)
81 break; 81 break;
82 pos += ret; 82 pos += ret;
83 } 83 }
84 } else if (ro0) { 84 } else if (ro0) {
85 for (i = 0; i < num_slots; i++ 85 for (i = 0; i < num_slots; i++) {
86 ret = snprintf(line + 86 ret = snprintf(line + pos, len - pos, " %d:%d",
87 ro0[i]. 87 ro0[i].ro_slot, ro0[i].ro_nodeid);
88 if (ret >= len - pos) 88 if (ret >= len - pos)
89 break; 89 break;
90 pos += ret; 90 pos += ret;
91 } 91 }
92 } 92 }
93 93
94 log_rinfo(ls, "generation %u slots %d% 94 log_rinfo(ls, "generation %u slots %d%s", gen, num_slots, line);
95 } 95 }
96 96
97 int dlm_slots_copy_in(struct dlm_ls *ls) 97 int dlm_slots_copy_in(struct dlm_ls *ls)
98 { 98 {
99 struct dlm_member *memb; 99 struct dlm_member *memb;
100 struct dlm_rcom *rc = ls->ls_recover_b 100 struct dlm_rcom *rc = ls->ls_recover_buf;
101 struct rcom_config *rf = (struct rcom_ 101 struct rcom_config *rf = (struct rcom_config *)rc->rc_buf;
102 struct rcom_slot *ro0, *ro; 102 struct rcom_slot *ro0, *ro;
103 int our_nodeid = dlm_our_nodeid(); 103 int our_nodeid = dlm_our_nodeid();
104 int i, num_slots; 104 int i, num_slots;
105 uint32_t gen; 105 uint32_t gen;
106 106
107 if (!dlm_slots_version(&rc->rc_header) 107 if (!dlm_slots_version(&rc->rc_header))
108 return -1; 108 return -1;
109 109
110 gen = le32_to_cpu(rf->rf_generation); 110 gen = le32_to_cpu(rf->rf_generation);
111 if (gen <= ls->ls_generation) { 111 if (gen <= ls->ls_generation) {
112 log_error(ls, "dlm_slots_copy_ 112 log_error(ls, "dlm_slots_copy_in gen %u old %u",
113 gen, ls->ls_generati 113 gen, ls->ls_generation);
114 } 114 }
115 ls->ls_generation = gen; 115 ls->ls_generation = gen;
116 116
117 num_slots = le16_to_cpu(rf->rf_num_slo 117 num_slots = le16_to_cpu(rf->rf_num_slots);
118 if (!num_slots) 118 if (!num_slots)
119 return -1; 119 return -1;
120 120
121 ro0 = (struct rcom_slot *)(rc->rc_buf 121 ro0 = (struct rcom_slot *)(rc->rc_buf + sizeof(struct rcom_config));
122 122
123 log_slots(ls, gen, num_slots, ro0, NUL 123 log_slots(ls, gen, num_slots, ro0, NULL, 0);
124 124
125 list_for_each_entry(memb, &ls->ls_node 125 list_for_each_entry(memb, &ls->ls_nodes, list) {
126 for (i = 0, ro = ro0; i < num_ 126 for (i = 0, ro = ro0; i < num_slots; i++, ro++) {
127 if (le32_to_cpu(ro->ro 127 if (le32_to_cpu(ro->ro_nodeid) != memb->nodeid)
128 continue; 128 continue;
129 memb->slot = le16_to_c 129 memb->slot = le16_to_cpu(ro->ro_slot);
130 memb->slot_prev = memb 130 memb->slot_prev = memb->slot;
131 break; 131 break;
132 } 132 }
133 133
134 if (memb->nodeid == our_nodeid 134 if (memb->nodeid == our_nodeid) {
135 if (ls->ls_slot && ls- 135 if (ls->ls_slot && ls->ls_slot != memb->slot) {
136 log_error(ls, 136 log_error(ls, "dlm_slots_copy_in our slot "
137 "cha 137 "changed %d %d", ls->ls_slot,
138 memb 138 memb->slot);
139 return -1; 139 return -1;
140 } 140 }
141 141
142 if (!ls->ls_slot) 142 if (!ls->ls_slot)
143 ls->ls_slot = 143 ls->ls_slot = memb->slot;
144 } 144 }
145 145
146 if (!memb->slot) { 146 if (!memb->slot) {
147 log_error(ls, "dlm_slo 147 log_error(ls, "dlm_slots_copy_in nodeid %d no slot",
148 memb->nodei 148 memb->nodeid);
149 return -1; 149 return -1;
150 } 150 }
151 } 151 }
152 152
153 return 0; 153 return 0;
154 } 154 }
155 155
156 /* for any nodes that do not support slots, we 156 /* for any nodes that do not support slots, we will not have set memb->slot
157 in wait_status_all(), so memb->slot will re 157 in wait_status_all(), so memb->slot will remain -1, and we will not
158 assign slots or set ls_num_slots here */ 158 assign slots or set ls_num_slots here */
159 159
160 int dlm_slots_assign(struct dlm_ls *ls, int *n 160 int dlm_slots_assign(struct dlm_ls *ls, int *num_slots, int *slots_size,
161 struct dlm_slot **slots_o 161 struct dlm_slot **slots_out, uint32_t *gen_out)
162 { 162 {
163 struct dlm_member *memb; 163 struct dlm_member *memb;
164 struct dlm_slot *array; 164 struct dlm_slot *array;
165 int our_nodeid = dlm_our_nodeid(); 165 int our_nodeid = dlm_our_nodeid();
166 int array_size, max_slots, i; 166 int array_size, max_slots, i;
167 int need = 0; 167 int need = 0;
168 int max = 0; 168 int max = 0;
169 int num = 0; 169 int num = 0;
170 uint32_t gen = 0; 170 uint32_t gen = 0;
171 171
172 /* our own memb struct will have slot 172 /* our own memb struct will have slot -1 gen 0 */
173 173
174 list_for_each_entry(memb, &ls->ls_node 174 list_for_each_entry(memb, &ls->ls_nodes, list) {
175 if (memb->nodeid == our_nodeid 175 if (memb->nodeid == our_nodeid) {
176 memb->slot = ls->ls_sl 176 memb->slot = ls->ls_slot;
177 memb->generation = ls- 177 memb->generation = ls->ls_generation;
178 break; 178 break;
179 } 179 }
180 } 180 }
181 181
182 list_for_each_entry(memb, &ls->ls_node 182 list_for_each_entry(memb, &ls->ls_nodes, list) {
183 if (memb->generation > gen) 183 if (memb->generation > gen)
184 gen = memb->generation 184 gen = memb->generation;
185 185
186 /* node doesn't support slots 186 /* node doesn't support slots */
187 187
188 if (memb->slot == -1) 188 if (memb->slot == -1)
189 return -1; 189 return -1;
190 190
191 /* node needs a slot assigned 191 /* node needs a slot assigned */
192 192
193 if (!memb->slot) 193 if (!memb->slot)
194 need++; 194 need++;
195 195
196 /* node has a slot assigned */ 196 /* node has a slot assigned */
197 197
198 num++; 198 num++;
199 199
200 if (!max || max < memb->slot) 200 if (!max || max < memb->slot)
201 max = memb->slot; 201 max = memb->slot;
202 202
203 /* sanity check, once slot is 203 /* sanity check, once slot is assigned it shouldn't change */
204 204
205 if (memb->slot_prev && memb->s 205 if (memb->slot_prev && memb->slot && memb->slot_prev != memb->slot) {
206 log_error(ls, "nodeid 206 log_error(ls, "nodeid %d slot changed %d %d",
207 memb->nodeid 207 memb->nodeid, memb->slot_prev, memb->slot);
208 return -1; 208 return -1;
209 } 209 }
210 memb->slot_prev = memb->slot; 210 memb->slot_prev = memb->slot;
211 } 211 }
212 212
213 array_size = max + need; 213 array_size = max + need;
214 array = kcalloc(array_size, sizeof(*ar 214 array = kcalloc(array_size, sizeof(*array), GFP_NOFS);
215 if (!array) 215 if (!array)
216 return -ENOMEM; 216 return -ENOMEM;
217 217
218 num = 0; 218 num = 0;
219 219
220 /* fill in slots (offsets) that are us 220 /* fill in slots (offsets) that are used */
221 221
222 list_for_each_entry(memb, &ls->ls_node 222 list_for_each_entry(memb, &ls->ls_nodes, list) {
223 if (!memb->slot) 223 if (!memb->slot)
224 continue; 224 continue;
225 225
226 if (memb->slot > array_size) { 226 if (memb->slot > array_size) {
227 log_error(ls, "invalid 227 log_error(ls, "invalid slot number %d", memb->slot);
228 kfree(array); 228 kfree(array);
229 return -1; 229 return -1;
230 } 230 }
231 231
232 array[memb->slot - 1].nodeid = 232 array[memb->slot - 1].nodeid = memb->nodeid;
233 array[memb->slot - 1].slot = m 233 array[memb->slot - 1].slot = memb->slot;
234 num++; 234 num++;
235 } 235 }
236 236
237 /* assign new slots from unused offset 237 /* assign new slots from unused offsets */
238 238
239 list_for_each_entry(memb, &ls->ls_node 239 list_for_each_entry(memb, &ls->ls_nodes, list) {
240 if (memb->slot) 240 if (memb->slot)
241 continue; 241 continue;
242 242
243 for (i = 0; i < array_size; i+ 243 for (i = 0; i < array_size; i++) {
244 if (array[i].nodeid) 244 if (array[i].nodeid)
245 continue; 245 continue;
246 246
247 memb->slot = i + 1; 247 memb->slot = i + 1;
248 memb->slot_prev = memb 248 memb->slot_prev = memb->slot;
249 array[i].nodeid = memb 249 array[i].nodeid = memb->nodeid;
250 array[i].slot = memb-> 250 array[i].slot = memb->slot;
251 num++; 251 num++;
252 252
253 if (!ls->ls_slot && me 253 if (!ls->ls_slot && memb->nodeid == our_nodeid)
254 ls->ls_slot = 254 ls->ls_slot = memb->slot;
255 break; 255 break;
256 } 256 }
257 257
258 if (!memb->slot) { 258 if (!memb->slot) {
259 log_error(ls, "no free 259 log_error(ls, "no free slot found");
260 kfree(array); 260 kfree(array);
261 return -1; 261 return -1;
262 } 262 }
263 } 263 }
264 264
265 gen++; 265 gen++;
266 266
267 log_slots(ls, gen, num, NULL, array, a 267 log_slots(ls, gen, num, NULL, array, array_size);
268 268
269 max_slots = (DLM_MAX_APP_BUFSIZE - siz 269 max_slots = (DLM_MAX_APP_BUFSIZE - sizeof(struct dlm_rcom) -
270 sizeof(struct rcom_config 270 sizeof(struct rcom_config)) / sizeof(struct rcom_slot);
271 271
272 if (num > max_slots) { 272 if (num > max_slots) {
273 log_error(ls, "num_slots %d ex 273 log_error(ls, "num_slots %d exceeds max_slots %d",
274 num, max_slots); 274 num, max_slots);
275 kfree(array); 275 kfree(array);
276 return -1; 276 return -1;
277 } 277 }
278 278
279 *gen_out = gen; 279 *gen_out = gen;
280 *slots_out = array; 280 *slots_out = array;
281 *slots_size = array_size; 281 *slots_size = array_size;
282 *num_slots = num; 282 *num_slots = num;
283 return 0; 283 return 0;
284 } 284 }
285 285
286 static void add_ordered_member(struct dlm_ls * 286 static void add_ordered_member(struct dlm_ls *ls, struct dlm_member *new)
287 { 287 {
288 struct dlm_member *memb = NULL; 288 struct dlm_member *memb = NULL;
289 struct list_head *tmp; 289 struct list_head *tmp;
290 struct list_head *newlist = &new->list 290 struct list_head *newlist = &new->list;
291 struct list_head *head = &ls->ls_nodes 291 struct list_head *head = &ls->ls_nodes;
292 292
293 list_for_each(tmp, head) { 293 list_for_each(tmp, head) {
294 memb = list_entry(tmp, struct 294 memb = list_entry(tmp, struct dlm_member, list);
295 if (new->nodeid < memb->nodeid 295 if (new->nodeid < memb->nodeid)
296 break; 296 break;
297 } 297 }
298 298
299 if (!memb) 299 if (!memb)
300 list_add_tail(newlist, head); 300 list_add_tail(newlist, head);
301 else { 301 else {
302 /* FIXME: can use list macro h 302 /* FIXME: can use list macro here */
303 newlist->prev = tmp->prev; 303 newlist->prev = tmp->prev;
304 newlist->next = tmp; 304 newlist->next = tmp;
305 tmp->prev->next = newlist; 305 tmp->prev->next = newlist;
306 tmp->prev = newlist; 306 tmp->prev = newlist;
307 } 307 }
308 } 308 }
309 309
310 static int add_remote_member(int nodeid) 310 static int add_remote_member(int nodeid)
311 { 311 {
312 int error; 312 int error;
313 313
314 if (nodeid == dlm_our_nodeid()) 314 if (nodeid == dlm_our_nodeid())
315 return 0; 315 return 0;
316 316
317 error = dlm_lowcomms_connect_node(node 317 error = dlm_lowcomms_connect_node(nodeid);
318 if (error < 0) 318 if (error < 0)
319 return error; 319 return error;
320 320
321 dlm_midcomms_add_member(nodeid); 321 dlm_midcomms_add_member(nodeid);
322 return 0; 322 return 0;
323 } 323 }
324 324
325 static int dlm_add_member(struct dlm_ls *ls, s 325 static int dlm_add_member(struct dlm_ls *ls, struct dlm_config_node *node)
326 { 326 {
327 struct dlm_member *memb; 327 struct dlm_member *memb;
328 int error; 328 int error;
329 329
330 memb = kzalloc(sizeof(*memb), GFP_NOFS 330 memb = kzalloc(sizeof(*memb), GFP_NOFS);
331 if (!memb) 331 if (!memb)
332 return -ENOMEM; 332 return -ENOMEM;
333 333
334 memb->nodeid = node->nodeid; 334 memb->nodeid = node->nodeid;
335 memb->weight = node->weight; 335 memb->weight = node->weight;
336 memb->comm_seq = node->comm_seq; 336 memb->comm_seq = node->comm_seq;
337 337
338 error = add_remote_member(node->nodeid 338 error = add_remote_member(node->nodeid);
339 if (error < 0) { 339 if (error < 0) {
340 kfree(memb); 340 kfree(memb);
341 return error; 341 return error;
342 } 342 }
343 343
344 add_ordered_member(ls, memb); 344 add_ordered_member(ls, memb);
345 ls->ls_num_nodes++; 345 ls->ls_num_nodes++;
346 return 0; 346 return 0;
347 } 347 }
348 348
349 static struct dlm_member *find_memb(struct lis 349 static struct dlm_member *find_memb(struct list_head *head, int nodeid)
350 { 350 {
351 struct dlm_member *memb; 351 struct dlm_member *memb;
352 352
353 list_for_each_entry(memb, head, list) 353 list_for_each_entry(memb, head, list) {
354 if (memb->nodeid == nodeid) 354 if (memb->nodeid == nodeid)
355 return memb; 355 return memb;
356 } 356 }
357 return NULL; 357 return NULL;
358 } 358 }
359 359
360 int dlm_is_member(struct dlm_ls *ls, int nodei 360 int dlm_is_member(struct dlm_ls *ls, int nodeid)
361 { 361 {
362 if (find_memb(&ls->ls_nodes, nodeid)) 362 if (find_memb(&ls->ls_nodes, nodeid))
363 return 1; 363 return 1;
364 return 0; 364 return 0;
365 } 365 }
366 366
367 int dlm_is_removed(struct dlm_ls *ls, int node 367 int dlm_is_removed(struct dlm_ls *ls, int nodeid)
368 { 368 {
369 WARN_ON_ONCE(!nodeid || nodeid == -1); <<
370 <<
371 if (find_memb(&ls->ls_nodes_gone, node 369 if (find_memb(&ls->ls_nodes_gone, nodeid))
372 return 1; 370 return 1;
373 return 0; 371 return 0;
374 } 372 }
375 373
376 static void clear_memb_list(struct list_head * 374 static void clear_memb_list(struct list_head *head,
377 void (*after_del)( 375 void (*after_del)(int nodeid))
378 { 376 {
379 struct dlm_member *memb; 377 struct dlm_member *memb;
380 378
381 while (!list_empty(head)) { 379 while (!list_empty(head)) {
382 memb = list_entry(head->next, 380 memb = list_entry(head->next, struct dlm_member, list);
383 list_del(&memb->list); 381 list_del(&memb->list);
384 if (after_del) 382 if (after_del)
385 after_del(memb->nodeid 383 after_del(memb->nodeid);
386 kfree(memb); 384 kfree(memb);
387 } 385 }
388 } 386 }
389 387
390 static void remove_remote_member(int nodeid) 388 static void remove_remote_member(int nodeid)
391 { 389 {
392 if (nodeid == dlm_our_nodeid()) 390 if (nodeid == dlm_our_nodeid())
393 return; 391 return;
394 392
395 dlm_midcomms_remove_member(nodeid); 393 dlm_midcomms_remove_member(nodeid);
396 } 394 }
397 395
398 void dlm_clear_members(struct dlm_ls *ls) 396 void dlm_clear_members(struct dlm_ls *ls)
399 { 397 {
400 clear_memb_list(&ls->ls_nodes, remove_ 398 clear_memb_list(&ls->ls_nodes, remove_remote_member);
401 ls->ls_num_nodes = 0; 399 ls->ls_num_nodes = 0;
402 } 400 }
403 401
404 void dlm_clear_members_gone(struct dlm_ls *ls) 402 void dlm_clear_members_gone(struct dlm_ls *ls)
405 { 403 {
406 clear_memb_list(&ls->ls_nodes_gone, NU 404 clear_memb_list(&ls->ls_nodes_gone, NULL);
407 } 405 }
408 406
409 static void make_member_array(struct dlm_ls *l 407 static void make_member_array(struct dlm_ls *ls)
410 { 408 {
411 struct dlm_member *memb; 409 struct dlm_member *memb;
412 int i, w, x = 0, total = 0, all_zero = 410 int i, w, x = 0, total = 0, all_zero = 0, *array;
413 411
414 kfree(ls->ls_node_array); 412 kfree(ls->ls_node_array);
415 ls->ls_node_array = NULL; 413 ls->ls_node_array = NULL;
416 414
417 list_for_each_entry(memb, &ls->ls_node 415 list_for_each_entry(memb, &ls->ls_nodes, list) {
418 if (memb->weight) 416 if (memb->weight)
419 total += memb->weight; 417 total += memb->weight;
420 } 418 }
421 419
422 /* all nodes revert to weight of 1 if 420 /* all nodes revert to weight of 1 if all have weight 0 */
423 421
424 if (!total) { 422 if (!total) {
425 total = ls->ls_num_nodes; 423 total = ls->ls_num_nodes;
426 all_zero = 1; 424 all_zero = 1;
427 } 425 }
428 426
429 ls->ls_total_weight = total; 427 ls->ls_total_weight = total;
430 array = kmalloc_array(total, sizeof(*a 428 array = kmalloc_array(total, sizeof(*array), GFP_NOFS);
431 if (!array) 429 if (!array)
432 return; 430 return;
433 431
434 list_for_each_entry(memb, &ls->ls_node 432 list_for_each_entry(memb, &ls->ls_nodes, list) {
435 if (!all_zero && !memb->weight 433 if (!all_zero && !memb->weight)
436 continue; 434 continue;
437 435
438 if (all_zero) 436 if (all_zero)
439 w = 1; 437 w = 1;
440 else 438 else
441 w = memb->weight; 439 w = memb->weight;
442 440
443 DLM_ASSERT(x < total, printk(" 441 DLM_ASSERT(x < total, printk("total %d x %d\n", total, x););
444 442
445 for (i = 0; i < w; i++) 443 for (i = 0; i < w; i++)
446 array[x++] = memb->nod 444 array[x++] = memb->nodeid;
447 } 445 }
448 446
449 ls->ls_node_array = array; 447 ls->ls_node_array = array;
450 } 448 }
451 449
452 /* send a status request to all members just t 450 /* send a status request to all members just to establish comms connections */
453 451
454 static int ping_members(struct dlm_ls *ls, uin 452 static int ping_members(struct dlm_ls *ls, uint64_t seq)
455 { 453 {
456 struct dlm_member *memb; 454 struct dlm_member *memb;
457 int error = 0; 455 int error = 0;
458 456
459 list_for_each_entry(memb, &ls->ls_node 457 list_for_each_entry(memb, &ls->ls_nodes, list) {
460 if (dlm_recovery_stopped(ls)) 458 if (dlm_recovery_stopped(ls)) {
461 error = -EINTR; 459 error = -EINTR;
462 break; 460 break;
463 } 461 }
464 error = dlm_rcom_status(ls, me 462 error = dlm_rcom_status(ls, memb->nodeid, 0, seq);
465 if (error) 463 if (error)
466 break; 464 break;
467 } 465 }
468 if (error) 466 if (error)
469 log_rinfo(ls, "ping_members ab 467 log_rinfo(ls, "ping_members aborted %d last nodeid %d",
470 error, ls->ls_recove 468 error, ls->ls_recover_nodeid);
471 return error; 469 return error;
472 } 470 }
473 471
474 static void dlm_lsop_recover_prep(struct dlm_l 472 static void dlm_lsop_recover_prep(struct dlm_ls *ls)
475 { 473 {
476 if (!ls->ls_ops || !ls->ls_ops->recove 474 if (!ls->ls_ops || !ls->ls_ops->recover_prep)
477 return; 475 return;
478 ls->ls_ops->recover_prep(ls->ls_ops_ar 476 ls->ls_ops->recover_prep(ls->ls_ops_arg);
479 } 477 }
480 478
481 static void dlm_lsop_recover_slot(struct dlm_l 479 static void dlm_lsop_recover_slot(struct dlm_ls *ls, struct dlm_member *memb)
482 { 480 {
483 struct dlm_slot slot; 481 struct dlm_slot slot;
484 uint32_t seq; 482 uint32_t seq;
485 int error; 483 int error;
486 484
487 if (!ls->ls_ops || !ls->ls_ops->recove 485 if (!ls->ls_ops || !ls->ls_ops->recover_slot)
488 return; 486 return;
489 487
490 /* if there is no comms connection wit 488 /* if there is no comms connection with this node
491 or the present comms connection is 489 or the present comms connection is newer
492 than the one when this member was a 490 than the one when this member was added, then
493 we consider the node to have failed 491 we consider the node to have failed (versus
494 being removed due to dlm_release_lo 492 being removed due to dlm_release_lockspace) */
495 493
496 error = dlm_comm_seq(memb->nodeid, &se 494 error = dlm_comm_seq(memb->nodeid, &seq);
497 495
498 if (!error && seq == memb->comm_seq) 496 if (!error && seq == memb->comm_seq)
499 return; 497 return;
500 498
501 slot.nodeid = memb->nodeid; 499 slot.nodeid = memb->nodeid;
502 slot.slot = memb->slot; 500 slot.slot = memb->slot;
503 501
504 ls->ls_ops->recover_slot(ls->ls_ops_ar 502 ls->ls_ops->recover_slot(ls->ls_ops_arg, &slot);
505 } 503 }
506 504
507 void dlm_lsop_recover_done(struct dlm_ls *ls) 505 void dlm_lsop_recover_done(struct dlm_ls *ls)
508 { 506 {
509 struct dlm_member *memb; 507 struct dlm_member *memb;
510 struct dlm_slot *slots; 508 struct dlm_slot *slots;
511 int i, num; 509 int i, num;
512 510
513 if (!ls->ls_ops || !ls->ls_ops->recove 511 if (!ls->ls_ops || !ls->ls_ops->recover_done)
514 return; 512 return;
515 513
516 num = ls->ls_num_nodes; 514 num = ls->ls_num_nodes;
517 slots = kcalloc(num, sizeof(*slots), G 515 slots = kcalloc(num, sizeof(*slots), GFP_KERNEL);
518 if (!slots) 516 if (!slots)
519 return; 517 return;
520 518
521 i = 0; 519 i = 0;
522 list_for_each_entry(memb, &ls->ls_node 520 list_for_each_entry(memb, &ls->ls_nodes, list) {
523 if (i == num) { 521 if (i == num) {
524 log_error(ls, "dlm_lso 522 log_error(ls, "dlm_lsop_recover_done bad num %d", num);
525 goto out; 523 goto out;
526 } 524 }
527 slots[i].nodeid = memb->nodeid 525 slots[i].nodeid = memb->nodeid;
528 slots[i].slot = memb->slot; 526 slots[i].slot = memb->slot;
529 i++; 527 i++;
530 } 528 }
531 529
532 ls->ls_ops->recover_done(ls->ls_ops_ar 530 ls->ls_ops->recover_done(ls->ls_ops_arg, slots, num,
533 ls->ls_slot, 531 ls->ls_slot, ls->ls_generation);
534 out: 532 out:
535 kfree(slots); 533 kfree(slots);
536 } 534 }
537 535
538 static struct dlm_config_node *find_config_nod 536 static struct dlm_config_node *find_config_node(struct dlm_recover *rv,
539 537 int nodeid)
540 { 538 {
541 int i; 539 int i;
542 540
543 for (i = 0; i < rv->nodes_count; i++) 541 for (i = 0; i < rv->nodes_count; i++) {
544 if (rv->nodes[i].nodeid == nod 542 if (rv->nodes[i].nodeid == nodeid)
545 return &rv->nodes[i]; 543 return &rv->nodes[i];
546 } 544 }
547 return NULL; 545 return NULL;
548 } 546 }
549 547
550 int dlm_recover_members(struct dlm_ls *ls, str 548 int dlm_recover_members(struct dlm_ls *ls, struct dlm_recover *rv, int *neg_out)
551 { 549 {
552 struct dlm_member *memb, *safe; 550 struct dlm_member *memb, *safe;
553 struct dlm_config_node *node; 551 struct dlm_config_node *node;
554 int i, error, neg = 0, low = -1; 552 int i, error, neg = 0, low = -1;
555 553
556 /* previously removed members that we' 554 /* previously removed members that we've not finished removing need to
557 * count as a negative change so the " 555 * count as a negative change so the "neg" recovery steps will happen
558 * 556 *
559 * This functionality must report all 557 * This functionality must report all member changes to lsops or
560 * midcomms layer and must never retur 558 * midcomms layer and must never return before.
561 */ 559 */
562 560
563 list_for_each_entry(memb, &ls->ls_node 561 list_for_each_entry(memb, &ls->ls_nodes_gone, list) {
564 log_rinfo(ls, "prev removed me 562 log_rinfo(ls, "prev removed member %d", memb->nodeid);
565 neg++; 563 neg++;
566 } 564 }
567 565
568 /* move departed members from ls_nodes 566 /* move departed members from ls_nodes to ls_nodes_gone */
569 567
570 list_for_each_entry_safe(memb, safe, & 568 list_for_each_entry_safe(memb, safe, &ls->ls_nodes, list) {
571 node = find_config_node(rv, me 569 node = find_config_node(rv, memb->nodeid);
572 if (node && !node->new) 570 if (node && !node->new)
573 continue; 571 continue;
574 572
575 if (!node) { 573 if (!node) {
576 log_rinfo(ls, "remove 574 log_rinfo(ls, "remove member %d", memb->nodeid);
577 } else { 575 } else {
578 /* removed and re-adde 576 /* removed and re-added */
579 log_rinfo(ls, "remove 577 log_rinfo(ls, "remove member %d comm_seq %u %u",
580 memb->nodeid 578 memb->nodeid, memb->comm_seq, node->comm_seq);
581 } 579 }
582 580
583 neg++; 581 neg++;
584 list_move(&memb->list, &ls->ls 582 list_move(&memb->list, &ls->ls_nodes_gone);
585 remove_remote_member(memb->nod 583 remove_remote_member(memb->nodeid);
586 ls->ls_num_nodes--; 584 ls->ls_num_nodes--;
587 dlm_lsop_recover_slot(ls, memb 585 dlm_lsop_recover_slot(ls, memb);
588 } 586 }
589 587
590 /* add new members to ls_nodes */ 588 /* add new members to ls_nodes */
591 589
592 for (i = 0; i < rv->nodes_count; i++) 590 for (i = 0; i < rv->nodes_count; i++) {
593 node = &rv->nodes[i]; 591 node = &rv->nodes[i];
594 if (dlm_is_member(ls, node->no 592 if (dlm_is_member(ls, node->nodeid))
595 continue; 593 continue;
596 error = dlm_add_member(ls, nod 594 error = dlm_add_member(ls, node);
597 if (error) 595 if (error)
598 return error; 596 return error;
599 597
600 log_rinfo(ls, "add member %d", 598 log_rinfo(ls, "add member %d", node->nodeid);
601 } 599 }
602 600
603 list_for_each_entry(memb, &ls->ls_node 601 list_for_each_entry(memb, &ls->ls_nodes, list) {
604 if (low == -1 || memb->nodeid 602 if (low == -1 || memb->nodeid < low)
605 low = memb->nodeid; 603 low = memb->nodeid;
606 } 604 }
607 ls->ls_low_nodeid = low; 605 ls->ls_low_nodeid = low;
608 606
609 make_member_array(ls); 607 make_member_array(ls);
610 *neg_out = neg; 608 *neg_out = neg;
611 609
612 error = ping_members(ls, rv->seq); 610 error = ping_members(ls, rv->seq);
613 log_rinfo(ls, "dlm_recover_members %d 611 log_rinfo(ls, "dlm_recover_members %d nodes", ls->ls_num_nodes);
614 return error; 612 return error;
615 } 613 }
616 614
617 /* Userspace guarantees that dlm_ls_stop() has 615 /* Userspace guarantees that dlm_ls_stop() has completed on all nodes before
618 dlm_ls_start() is called on any of them to 616 dlm_ls_start() is called on any of them to start the new recovery. */
619 617
620 int dlm_ls_stop(struct dlm_ls *ls) 618 int dlm_ls_stop(struct dlm_ls *ls)
621 { 619 {
622 int new; 620 int new;
623 621
624 /* 622 /*
625 * Prevent dlm_recv from being in the 623 * Prevent dlm_recv from being in the middle of something when we do
626 * the stop. This includes ensuring d 624 * the stop. This includes ensuring dlm_recv isn't processing a
627 * recovery message (rcom), while dlm_ 625 * recovery message (rcom), while dlm_recoverd is aborting and
628 * resetting things from an in-progres 626 * resetting things from an in-progress recovery. i.e. we want
629 * dlm_recoverd to abort its recovery 627 * dlm_recoverd to abort its recovery without worrying about dlm_recv
630 * processing an rcom at the same time 628 * processing an rcom at the same time. Stopping dlm_recv also makes
631 * it easy for dlm_receive_message() t 629 * it easy for dlm_receive_message() to check locking stopped and add a
632 * message to the requestqueue without 630 * message to the requestqueue without races.
633 */ 631 */
634 632
635 write_lock_bh(&ls->ls_recv_active); 633 write_lock_bh(&ls->ls_recv_active);
636 634
637 /* 635 /*
638 * Abort any recovery that's in progre 636 * Abort any recovery that's in progress (see RECOVER_STOP,
639 * dlm_recovery_stopped()) and tell an 637 * dlm_recovery_stopped()) and tell any other threads running in the
640 * dlm to quit any processing (see RUN 638 * dlm to quit any processing (see RUNNING, dlm_locking_stopped()).
641 */ 639 */
642 640
643 spin_lock_bh(&ls->ls_recover_lock); 641 spin_lock_bh(&ls->ls_recover_lock);
644 set_bit(LSFL_RECOVER_STOP, &ls->ls_fla 642 set_bit(LSFL_RECOVER_STOP, &ls->ls_flags);
645 new = test_and_clear_bit(LSFL_RUNNING, 643 new = test_and_clear_bit(LSFL_RUNNING, &ls->ls_flags);
646 if (new) 644 if (new)
647 timer_delete_sync(&ls->ls_scan 645 timer_delete_sync(&ls->ls_scan_timer);
648 ls->ls_recover_seq++; 646 ls->ls_recover_seq++;
649 647
650 /* activate requestqueue and stop proc 648 /* activate requestqueue and stop processing */
651 write_lock_bh(&ls->ls_requestqueue_loc 649 write_lock_bh(&ls->ls_requestqueue_lock);
652 set_bit(LSFL_RECV_MSG_BLOCKED, &ls->ls 650 set_bit(LSFL_RECV_MSG_BLOCKED, &ls->ls_flags);
653 write_unlock_bh(&ls->ls_requestqueue_l 651 write_unlock_bh(&ls->ls_requestqueue_lock);
654 spin_unlock_bh(&ls->ls_recover_lock); 652 spin_unlock_bh(&ls->ls_recover_lock);
655 653
656 /* 654 /*
657 * Let dlm_recv run again, now any nor 655 * Let dlm_recv run again, now any normal messages will be saved on the
658 * requestqueue for later. 656 * requestqueue for later.
659 */ 657 */
660 658
661 write_unlock_bh(&ls->ls_recv_active); 659 write_unlock_bh(&ls->ls_recv_active);
662 660
663 /* 661 /*
664 * This in_recovery lock does two thin 662 * This in_recovery lock does two things:
665 * 1) Keeps this function from returni 663 * 1) Keeps this function from returning until all threads are out
666 * of locking routines and locking 664 * of locking routines and locking is truly stopped.
667 * 2) Keeps any new requests from bein 665 * 2) Keeps any new requests from being processed until it's unlocked
668 * when recovery is complete. 666 * when recovery is complete.
669 */ 667 */
670 668
671 if (new) { 669 if (new) {
672 set_bit(LSFL_RECOVER_DOWN, &ls 670 set_bit(LSFL_RECOVER_DOWN, &ls->ls_flags);
673 wake_up_process(ls->ls_recover 671 wake_up_process(ls->ls_recoverd_task);
674 wait_event(ls->ls_recover_lock 672 wait_event(ls->ls_recover_lock_wait,
675 test_bit(LSFL_RECOV 673 test_bit(LSFL_RECOVER_LOCK, &ls->ls_flags));
676 } 674 }
677 675
678 /* 676 /*
679 * The recoverd suspend/resume makes s 677 * The recoverd suspend/resume makes sure that dlm_recoverd (if
680 * running) has noticed RECOVER_STOP a 678 * running) has noticed RECOVER_STOP above and quit processing the
681 * previous recovery. 679 * previous recovery.
682 */ 680 */
683 681
684 dlm_recoverd_suspend(ls); 682 dlm_recoverd_suspend(ls);
685 683
686 spin_lock_bh(&ls->ls_recover_lock); 684 spin_lock_bh(&ls->ls_recover_lock);
687 kfree(ls->ls_slots); 685 kfree(ls->ls_slots);
688 ls->ls_slots = NULL; 686 ls->ls_slots = NULL;
689 ls->ls_num_slots = 0; 687 ls->ls_num_slots = 0;
690 ls->ls_slots_size = 0; 688 ls->ls_slots_size = 0;
691 ls->ls_recover_status = 0; 689 ls->ls_recover_status = 0;
692 spin_unlock_bh(&ls->ls_recover_lock); 690 spin_unlock_bh(&ls->ls_recover_lock);
693 691
694 dlm_recoverd_resume(ls); 692 dlm_recoverd_resume(ls);
695 693
696 if (!ls->ls_recover_begin) 694 if (!ls->ls_recover_begin)
697 ls->ls_recover_begin = jiffies 695 ls->ls_recover_begin = jiffies;
698 696
699 /* call recover_prep ops only once and 697 /* call recover_prep ops only once and not multiple times
700 * for each possible dlm_ls_stop() whe 698 * for each possible dlm_ls_stop() when recovery is already
701 * stopped. 699 * stopped.
702 * 700 *
703 * If we successful was able to clear 701 * If we successful was able to clear LSFL_RUNNING bit and
704 * it was set we know it is the first 702 * it was set we know it is the first dlm_ls_stop() call.
705 */ 703 */
706 if (new) 704 if (new)
707 dlm_lsop_recover_prep(ls); 705 dlm_lsop_recover_prep(ls);
708 706
709 return 0; 707 return 0;
710 } 708 }
711 709
712 int dlm_ls_start(struct dlm_ls *ls) 710 int dlm_ls_start(struct dlm_ls *ls)
713 { 711 {
714 struct dlm_recover *rv, *rv_old; 712 struct dlm_recover *rv, *rv_old;
715 struct dlm_config_node *nodes = NULL; 713 struct dlm_config_node *nodes = NULL;
716 int error, count; 714 int error, count;
717 715
718 rv = kzalloc(sizeof(*rv), GFP_NOFS); 716 rv = kzalloc(sizeof(*rv), GFP_NOFS);
719 if (!rv) 717 if (!rv)
720 return -ENOMEM; 718 return -ENOMEM;
721 719
722 error = dlm_config_nodes(ls->ls_name, 720 error = dlm_config_nodes(ls->ls_name, &nodes, &count);
723 if (error < 0) 721 if (error < 0)
724 goto fail_rv; 722 goto fail_rv;
725 723
726 spin_lock_bh(&ls->ls_recover_lock); 724 spin_lock_bh(&ls->ls_recover_lock);
727 725
728 /* the lockspace needs to be stopped b 726 /* the lockspace needs to be stopped before it can be started */
729 727
730 if (!dlm_locking_stopped(ls)) { 728 if (!dlm_locking_stopped(ls)) {
731 spin_unlock_bh(&ls->ls_recover 729 spin_unlock_bh(&ls->ls_recover_lock);
732 log_error(ls, "start ignored: 730 log_error(ls, "start ignored: lockspace running");
733 error = -EINVAL; 731 error = -EINVAL;
734 goto fail; 732 goto fail;
735 } 733 }
736 734
737 rv->nodes = nodes; 735 rv->nodes = nodes;
738 rv->nodes_count = count; 736 rv->nodes_count = count;
739 rv->seq = ++ls->ls_recover_seq; 737 rv->seq = ++ls->ls_recover_seq;
740 rv_old = ls->ls_recover_args; 738 rv_old = ls->ls_recover_args;
741 ls->ls_recover_args = rv; 739 ls->ls_recover_args = rv;
742 spin_unlock_bh(&ls->ls_recover_lock); 740 spin_unlock_bh(&ls->ls_recover_lock);
743 741
744 if (rv_old) { 742 if (rv_old) {
745 log_error(ls, "unused recovery 743 log_error(ls, "unused recovery %llx %d",
746 (unsigned long long) 744 (unsigned long long)rv_old->seq, rv_old->nodes_count);
747 kfree(rv_old->nodes); 745 kfree(rv_old->nodes);
748 kfree(rv_old); 746 kfree(rv_old);
749 } 747 }
750 748
751 set_bit(LSFL_RECOVER_WORK, &ls->ls_fla 749 set_bit(LSFL_RECOVER_WORK, &ls->ls_flags);
752 wake_up_process(ls->ls_recoverd_task); 750 wake_up_process(ls->ls_recoverd_task);
753 return 0; 751 return 0;
754 752
755 fail: 753 fail:
756 kfree(nodes); 754 kfree(nodes);
757 fail_rv: 755 fail_rv:
758 kfree(rv); 756 kfree(rv);
759 return error; 757 return error;
760 } 758 }
761 759
762 760
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.