| 1 |
// SPDX-License-Identifier: GPL-2.0-only |
| 2 |
/****************************************************************************** |
| 3 |
******************************************************************************* |
| 4 |
** |
| 5 |
** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. |
| 6 |
** Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved. |
| 7 |
** |
| 8 |
** |
| 9 |
******************************************************************************* |
| 10 |
******************************************************************************/ |
| 11 |
|
| 12 |
#include "dlm_internal.h" |
| 13 |
#include "lockspace.h" |
| 14 |
#include "member.h" |
| 15 |
#include "lowcomms.h" |
| 16 |
#include "rcom.h" |
| 17 |
#include "config.h" |
| 18 |
#include "memory.h" |
| 19 |
#include "recover.h" |
| 20 |
#include "util.h" |
| 21 |
#include "lock.h" |
| 22 |
#include "dir.h" |
| 23 |
|
| 24 |
/* |
| 25 |
* We use the upper 16 bits of the hash value to select the directory node. |
| 26 |
* Low bits are used for distribution of rsb's among hash buckets on each node. |
| 27 |
* |
| 28 |
* To give the exact range wanted (0 to num_nodes-1), we apply a modulus of |
| 29 |
* num_nodes to the hash value. This value in the desired range is used as an |
| 30 |
* offset into the sorted list of nodeid's to give the particular nodeid. |
| 31 |
*/ |
| 32 |
|
| 33 |
int dlm_hash2nodeid(struct dlm_ls *ls, uint32_t hash) |
| 34 |
{ |
| 35 |
uint32_t node; |
| 36 |
|
| 37 |
if (ls->ls_num_nodes == 1) |
| 38 |
return dlm_our_nodeid(); |
| 39 |
else { |
| 40 |
node = (hash >> 16) % ls->ls_total_weight; |
| 41 |
return ls->ls_node_array[node]; |
| 42 |
} |
| 43 |
} |
| 44 |
|
| 45 |
int dlm_dir_nodeid(struct dlm_rsb *r) |
| 46 |
{ |
| 47 |
return r->res_dir_nodeid; |
| 48 |
} |
| 49 |
|
| 50 |
void dlm_recover_dir_nodeid(struct dlm_ls *ls, const struct list_head *root_list) |
| 51 |
{ |
| 52 |
struct dlm_rsb *r; |
| 53 |
|
| 54 |
list_for_each_entry(r, root_list, res_root_list) { |
| 55 |
r->res_dir_nodeid = dlm_hash2nodeid(ls, r->res_hash); |
| 56 |
} |
| 57 |
} |
| 58 |
|
| 59 |
int dlm_recover_directory(struct dlm_ls *ls, uint64_t seq) |
| 60 |
{ |
| 61 |
struct dlm_member *memb; |
| 62 |
char *b, *last_name = NULL; |
| 63 |
int error = -ENOMEM, last_len, nodeid, result; |
| 64 |
uint16_t namelen; |
| 65 |
unsigned int count = 0, count_bad = 0, count_add = 0; |
| 66 |
|
| 67 |
log_rinfo(ls, "dlm_recover_directory"); |
| 68 |
|
| 69 |
if (dlm_no_directory(ls)) |
| 70 |
goto out_status; |
| 71 |
|
| 72 |
last_name = kmalloc(DLM_RESNAME_MAXLEN, GFP_NOFS); |
| 73 |
if (!last_name) |
| 74 |
goto out; |
| 75 |
|
| 76 |
list_for_each_entry(memb, &ls->ls_nodes, list) { |
| 77 |
if (memb->nodeid == dlm_our_nodeid()) |
| 78 |
continue; |
| 79 |
|
| 80 |
memset(last_name, 0, DLM_RESNAME_MAXLEN); |
| 81 |
last_len = 0; |
| 82 |
|
| 83 |
for (;;) { |
| 84 |
int left; |
| 85 |
if (dlm_recovery_stopped(ls)) { |
| 86 |
error = -EINTR; |
| 87 |
goto out_free; |
| 88 |
} |
| 89 |
|
| 90 |
error = dlm_rcom_names(ls, memb->nodeid, |
| 91 |
last_name, last_len, seq); |
| 92 |
if (error) |
| 93 |
goto out_free; |
| 94 |
|
| 95 |
cond_resched(); |
| 96 |
|
| 97 |
/* |
| 98 |
* pick namelen/name pairs out of received buffer |
| 99 |
*/ |
| 100 |
|
| 101 |
b = ls->ls_recover_buf->rc_buf; |
| 102 |
left = le16_to_cpu(ls->ls_recover_buf->rc_header.h_length); |
| 103 |
left -= sizeof(struct dlm_rcom); |
| 104 |
|
| 105 |
for (;;) { |
| 106 |
__be16 v; |
| 107 |
|
| 108 |
error = -EINVAL; |
| 109 |
if (left < sizeof(__be16)) |
| 110 |
goto out_free; |
| 111 |
|
| 112 |
memcpy(&v, b, sizeof(__be16)); |
| 113 |
namelen = be16_to_cpu(v); |
| 114 |
b += sizeof(__be16); |
| 115 |
left -= sizeof(__be16); |
| 116 |
|
| 117 |
/* namelen of 0xFFFFF marks end of names for |
| 118 |
this node; namelen of 0 marks end of the |
| 119 |
buffer */ |
| 120 |
|
| 121 |
if (namelen == 0xFFFF) |
| 122 |
goto done; |
| 123 |
if (!namelen) |
| 124 |
break; |
| 125 |
|
| 126 |
if (namelen > left) |
| 127 |
goto out_free; |
| 128 |
|
| 129 |
if (namelen > DLM_RESNAME_MAXLEN) |
| 130 |
goto out_free; |
| 131 |
|
| 132 |
error = dlm_master_lookup(ls, memb->nodeid, |
| 133 |
b, namelen, |
| 134 |
DLM_LU_RECOVER_DIR, |
| 135 |
&nodeid, &result); |
| 136 |
if (error) { |
| 137 |
log_error(ls, "recover_dir lookup %d", |
| 138 |
error); |
| 139 |
goto out_free; |
| 140 |
} |
| 141 |
|
| 142 |
/* The name was found in rsbtbl, but the |
| 143 |
* master nodeid is different from |
| 144 |
* memb->nodeid which says it is the master. |
| 145 |
* This should not happen. */ |
| 146 |
|
| 147 |
if (result == DLM_LU_MATCH && |
| 148 |
nodeid != memb->nodeid) { |
| 149 |
count_bad++; |
| 150 |
log_error(ls, "recover_dir lookup %d " |
| 151 |
"nodeid %d memb %d bad %u", |
| 152 |
result, nodeid, memb->nodeid, |
| 153 |
count_bad); |
| 154 |
print_hex_dump_bytes("dlm_recover_dir ", |
| 155 |
DUMP_PREFIX_NONE, |
| 156 |
b, namelen); |
| 157 |
} |
| 158 |
|
| 159 |
/* The name was found in rsbtbl, and the |
| 160 |
* master nodeid matches memb->nodeid. |
| 161 |
|
| 162 |
if (result == DLM_LU_MATCH && |
| 163 |
nodeid == memb->nodeid) { |
| 164 |
count_match++; |
| 165 |
}*/ |
| 166 |
|
| 167 |
/* The name was not found in rsbtbl and was |
| 168 |
* added with memb->nodeid as the master. */ |
| 169 |
|
| 170 |
if (result == DLM_LU_ADD) { |
| 171 |
count_add++; |
| 172 |
} |
| 173 |
|
| 174 |
last_len = namelen; |
| 175 |
memcpy(last_name, b, namelen); |
| 176 |
b += namelen; |
| 177 |
left -= namelen; |
| 178 |
count++; |
| 179 |
} |
| 180 |
} |
| 181 |
done: |
| 182 |
; |
| 183 |
} |
| 184 |
|
| 185 |
out_status: |
| 186 |
error = 0; |
| 187 |
dlm_set_recover_status(ls, DLM_RS_DIR); |
| 188 |
|
| 189 |
log_rinfo(ls, "dlm_recover_directory %u in %u new", |
| 190 |
count, count_add); |
| 191 |
out_free: |
| 192 |
kfree(last_name); |
| 193 |
out: |
| 194 |
return error; |
| 195 |
} |
| 196 |
|
| 197 |
static struct dlm_rsb *find_rsb_root(struct dlm_ls *ls, const char *name, |
| 198 |
int len) |
| 199 |
{ |
| 200 |
struct dlm_rsb *r; |
| 201 |
int rv; |
| 202 |
|
| 203 |
read_lock_bh(&ls->ls_rsbtbl_lock); |
| 204 |
rv = dlm_search_rsb_tree(&ls->ls_rsbtbl, name, len, &r); |
| 205 |
read_unlock_bh(&ls->ls_rsbtbl_lock); |
| 206 |
if (!rv) |
| 207 |
return r; |
| 208 |
|
| 209 |
list_for_each_entry(r, &ls->ls_masters_list, res_masters_list) { |
| 210 |
if (len == r->res_length && !memcmp(name, r->res_name, len)) { |
| 211 |
log_debug(ls, "find_rsb_root revert to root_list %s", |
| 212 |
r->res_name); |
| 213 |
return r; |
| 214 |
} |
| 215 |
} |
| 216 |
return NULL; |
| 217 |
} |
| 218 |
|
| 219 |
struct dlm_dir_dump { |
| 220 |
/* init values to match if whole |
| 221 |
* dump fits to one seq. Sanity check only. |
| 222 |
*/ |
| 223 |
uint64_t seq_init; |
| 224 |
uint64_t nodeid_init; |
| 225 |
/* compare local pointer with last lookup, |
| 226 |
* just a sanity check. |
| 227 |
*/ |
| 228 |
struct list_head *last; |
| 229 |
|
| 230 |
unsigned int sent_res; /* for log info */ |
| 231 |
unsigned int sent_msg; /* for log info */ |
| 232 |
|
| 233 |
struct list_head list; |
| 234 |
}; |
| 235 |
|
| 236 |
static void drop_dir_ctx(struct dlm_ls *ls, int nodeid) |
| 237 |
{ |
| 238 |
struct dlm_dir_dump *dd, *safe; |
| 239 |
|
| 240 |
write_lock_bh(&ls->ls_dir_dump_lock); |
| 241 |
list_for_each_entry_safe(dd, safe, &ls->ls_dir_dump_list, list) { |
| 242 |
if (dd->nodeid_init == nodeid) { |
| 243 |
log_error(ls, "drop dump seq %llu", |
| 244 |
(unsigned long long)dd->seq_init); |
| 245 |
list_del(&dd->list); |
| 246 |
kfree(dd); |
| 247 |
} |
| 248 |
} |
| 249 |
write_unlock_bh(&ls->ls_dir_dump_lock); |
| 250 |
} |
| 251 |
|
| 252 |
static struct dlm_dir_dump *lookup_dir_dump(struct dlm_ls *ls, int nodeid) |
| 253 |
{ |
| 254 |
struct dlm_dir_dump *iter, *dd = NULL; |
| 255 |
|
| 256 |
read_lock_bh(&ls->ls_dir_dump_lock); |
| 257 |
list_for_each_entry(iter, &ls->ls_dir_dump_list, list) { |
| 258 |
if (iter->nodeid_init == nodeid) { |
| 259 |
dd = iter; |
| 260 |
break; |
| 261 |
} |
| 262 |
} |
| 263 |
read_unlock_bh(&ls->ls_dir_dump_lock); |
| 264 |
|
| 265 |
return dd; |
| 266 |
} |
| 267 |
|
| 268 |
static struct dlm_dir_dump *init_dir_dump(struct dlm_ls *ls, int nodeid) |
| 269 |
{ |
| 270 |
struct dlm_dir_dump *dd; |
| 271 |
|
| 272 |
dd = lookup_dir_dump(ls, nodeid); |
| 273 |
if (dd) { |
| 274 |
log_error(ls, "found ongoing dir dump for node %d, will drop it", |
| 275 |
nodeid); |
| 276 |
drop_dir_ctx(ls, nodeid); |
| 277 |
} |
| 278 |
|
| 279 |
dd = kzalloc_obj(*dd, GFP_ATOMIC); |
| 280 |
if (!dd) |
| 281 |
return NULL; |
| 282 |
|
| 283 |
dd->seq_init = ls->ls_recover_seq; |
| 284 |
dd->nodeid_init = nodeid; |
| 285 |
|
| 286 |
write_lock_bh(&ls->ls_dir_dump_lock); |
| 287 |
list_add(&dd->list, &ls->ls_dir_dump_list); |
| 288 |
write_unlock_bh(&ls->ls_dir_dump_lock); |
| 289 |
|
| 290 |
return dd; |
| 291 |
} |
| 292 |
|
| 293 |
/* Find the rsb where we left off (or start again), then send rsb names |
| 294 |
for rsb's we're master of and whose directory node matches the requesting |
| 295 |
node. inbuf is the rsb name last sent, inlen is the name's length */ |
| 296 |
|
| 297 |
void dlm_copy_master_names(struct dlm_ls *ls, const char *inbuf, int inlen, |
| 298 |
char *outbuf, int outlen, int nodeid) |
| 299 |
{ |
| 300 |
struct list_head *list; |
| 301 |
struct dlm_rsb *r; |
| 302 |
int offset = 0, dir_nodeid; |
| 303 |
struct dlm_dir_dump *dd; |
| 304 |
__be16 be_namelen; |
| 305 |
|
| 306 |
read_lock_bh(&ls->ls_masters_lock); |
| 307 |
|
| 308 |
if (inlen > 1) { |
| 309 |
dd = lookup_dir_dump(ls, nodeid); |
| 310 |
if (!dd) { |
| 311 |
log_error(ls, "failed to lookup dir dump context nodeid: %d", |
| 312 |
nodeid); |
| 313 |
goto out; |
| 314 |
} |
| 315 |
|
| 316 |
/* next chunk in dump */ |
| 317 |
r = find_rsb_root(ls, inbuf, inlen); |
| 318 |
if (!r) { |
| 319 |
log_error(ls, "copy_master_names from %d start %d %.*s", |
| 320 |
nodeid, inlen, inlen, inbuf); |
| 321 |
goto out; |
| 322 |
} |
| 323 |
list = r->res_masters_list.next; |
| 324 |
|
| 325 |
/* sanity checks */ |
| 326 |
if (dd->last != &r->res_masters_list || |
| 327 |
dd->seq_init != ls->ls_recover_seq) { |
| 328 |
log_error(ls, "failed dir dump sanity check seq_init: %llu seq: %llu", |
| 329 |
(unsigned long long)dd->seq_init, |
| 330 |
(unsigned long long)ls->ls_recover_seq); |
| 331 |
goto out; |
| 332 |
} |
| 333 |
} else { |
| 334 |
dd = init_dir_dump(ls, nodeid); |
| 335 |
if (!dd) { |
| 336 |
log_error(ls, "failed to allocate dir dump context"); |
| 337 |
goto out; |
| 338 |
} |
| 339 |
|
| 340 |
/* start dump */ |
| 341 |
list = ls->ls_masters_list.next; |
| 342 |
dd->last = list; |
| 343 |
} |
| 344 |
|
| 345 |
for (offset = 0; list != &ls->ls_masters_list; list = list->next) { |
| 346 |
r = list_entry(list, struct dlm_rsb, res_masters_list); |
| 347 |
dir_nodeid = dlm_dir_nodeid(r); |
| 348 |
if (dir_nodeid != nodeid) |
| 349 |
continue; |
| 350 |
|
| 351 |
/* |
| 352 |
* The block ends when we can't fit the following in the |
| 353 |
* remaining buffer space: |
| 354 |
* namelen (uint16_t) + |
| 355 |
* name (r->res_length) + |
| 356 |
* end-of-block record 0x0000 (uint16_t) |
| 357 |
*/ |
| 358 |
|
| 359 |
if (offset + sizeof(uint16_t)*2 + r->res_length > outlen) { |
| 360 |
/* Write end-of-block record */ |
| 361 |
be_namelen = cpu_to_be16(0); |
| 362 |
memcpy(outbuf + offset, &be_namelen, sizeof(__be16)); |
| 363 |
offset += sizeof(__be16); |
| 364 |
dd->sent_msg++; |
| 365 |
goto out; |
| 366 |
} |
| 367 |
|
| 368 |
be_namelen = cpu_to_be16(r->res_length); |
| 369 |
memcpy(outbuf + offset, &be_namelen, sizeof(__be16)); |
| 370 |
offset += sizeof(__be16); |
| 371 |
memcpy(outbuf + offset, r->res_name, r->res_length); |
| 372 |
offset += r->res_length; |
| 373 |
dd->sent_res++; |
| 374 |
dd->last = list; |
| 375 |
} |
| 376 |
|
| 377 |
/* |
| 378 |
* If we've reached the end of the list (and there's room) write a |
| 379 |
* terminating record. |
| 380 |
*/ |
| 381 |
|
| 382 |
if ((list == &ls->ls_masters_list) && |
| 383 |
(offset + sizeof(uint16_t) <= outlen)) { |
| 384 |
/* end dump */ |
| 385 |
be_namelen = cpu_to_be16(0xFFFF); |
| 386 |
memcpy(outbuf + offset, &be_namelen, sizeof(__be16)); |
| 387 |
offset += sizeof(__be16); |
| 388 |
dd->sent_msg++; |
| 389 |
log_rinfo(ls, "dlm_recover_directory nodeid %d sent %u res out %u messages", |
| 390 |
nodeid, dd->sent_res, dd->sent_msg); |
| 391 |
|
| 392 |
write_lock_bh(&ls->ls_dir_dump_lock); |
| 393 |
list_del_init(&dd->list); |
| 394 |
write_unlock_bh(&ls->ls_dir_dump_lock); |
| 395 |
kfree(dd); |
| 396 |
} |
| 397 |
out: |
| 398 |
read_unlock_bh(&ls->ls_masters_lock); |
| 399 |
} |
| 400 |
|