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