1// SPDX-License-Identifier: GPL-2.0-or-later 2/* AFS server record management 3 * 4 * Copyright (C) 2002, 2007 Red Hat, Inc. All Rights Reserved. 5 * Written by David Howells (dhowells@redhat.com) 6 */ 7 8#include <linux/sched.h> 9#include <linux/slab.h> 10#include "afs_fs.h" 11#include "internal.h" 12#include "protocol_yfs.h" 13 14static unsigned afs_server_gc_delay = 10; /* Server record timeout in seconds */ 15static atomic_t afs_server_debug_id; 16 17static struct afs_server *afs_maybe_use_server(struct afs_server *, 18 enum afs_server_trace); 19static void __afs_put_server(struct afs_net *, struct afs_server *); 20 21/* 22 * Find a server by one of its addresses. 23 */ 24struct afs_server *afs_find_server(struct afs_net *net, 25 const struct sockaddr_rxrpc *srx) 26{ 27 const struct afs_addr_list *alist; 28 struct afs_server *server = NULL; 29 unsigned int i; 30 int seq = 1, diff; 31 32 rcu_read_lock(); 33 34 do { 35 if (server) 36 afs_unuse_server_notime(net, server, afs_server_trace_put_find_rsq); 37 server = NULL; 38 seq++; /* 2 on the 1st/lockless path, otherwise odd */ 39 read_seqbegin_or_lock(&net->fs_addr_lock, &seq); 40 41 if (srx->transport.family == AF_INET6) { 42 const struct sockaddr_in6 *a = &srx->transport.sin6, *b; 43 hlist_for_each_entry_rcu(server, &net->fs_addresses6, addr6_link) { 44 alist = rcu_dereference(server->addresses); 45 for (i = alist->nr_ipv4; i < alist->nr_addrs; i++) { 46 b = &alist->addrs[i].transport.sin6; 47 diff = ((u16 __force)a->sin6_port - 48 (u16 __force)b->sin6_port); 49 if (diff == 0) 50 diff = memcmp(&a->sin6_addr, 51 &b->sin6_addr, 52 sizeof(struct in6_addr)); 53 if (diff == 0) 54 goto found; 55 } 56 } 57 } else { 58 const struct sockaddr_in *a = &srx->transport.sin, *b; 59 hlist_for_each_entry_rcu(server, &net->fs_addresses4, addr4_link) { 60 alist = rcu_dereference(server->addresses); 61 for (i = 0; i < alist->nr_ipv4; i++) { 62 b = &alist->addrs[i].transport.sin; 63 diff = ((u16 __force)a->sin_port - 64 (u16 __force)b->sin_port); 65 if (diff == 0) 66 diff = ((u32 __force)a->sin_addr.s_addr - 67 (u32 __force)b->sin_addr.s_addr); 68 if (diff == 0) 69 goto found; 70 } 71 } 72 } 73 74 server = NULL; 75 continue; 76 found: 77 server = afs_maybe_use_server(server, afs_server_trace_get_by_addr); 78 79 } while (need_seqretry(&net->fs_addr_lock, seq)); 80 81 done_seqretry(&net->fs_addr_lock, seq); 82 83 rcu_read_unlock(); 84 return server; 85} 86 87/* 88 * Look up a server by its UUID and mark it active. 89 */ 90struct afs_server *afs_find_server_by_uuid(struct afs_net *net, const uuid_t *uuid) 91{ 92 struct afs_server *server = NULL; 93 struct rb_node *p; 94 int diff, seq = 1; 95 96 _enter("%pU", uuid); 97 98 do { 99 /* Unfortunately, rbtree walking doesn't give reliable results 100 * under just the RCU read lock, so we have to check for 101 * changes. 102 */ 103 if (server) 104 afs_unuse_server(net, server, afs_server_trace_put_uuid_rsq); 105 server = NULL; 106 seq++; /* 2 on the 1st/lockless path, otherwise odd */ 107 read_seqbegin_or_lock(&net->fs_lock, &seq); 108 109 p = net->fs_servers.rb_node; 110 while (p) { 111 server = rb_entry(p, struct afs_server, uuid_rb); 112 113 diff = memcmp(uuid, &server->uuid, sizeof(*uuid)); 114 if (diff < 0) { 115 p = p->rb_left; 116 } else if (diff > 0) { 117 p = p->rb_right; 118 } else { 119 afs_use_server(server, afs_server_trace_get_by_uuid); 120 break; 121 } 122 123 server = NULL; 124 } 125 } while (need_seqretry(&net->fs_lock, seq)); 126 127 done_seqretry(&net->fs_lock, seq); 128 129 _leave(" = %p", server); 130 return server; 131} 132 133/* 134 * Install a server record in the namespace tree. If there's a clash, we stick 135 * it into a list anchored on whichever afs_server struct is actually in the 136 * tree. 137 */ 138static struct afs_server *afs_install_server(struct afs_cell *cell, 139 struct afs_server *candidate) 140{ 141 const struct afs_addr_list *alist; 142 struct afs_server *server, *next; 143 struct afs_net *net = cell->net; 144 struct rb_node **pp, *p; 145 int diff; 146 147 _enter("%p", candidate); 148 149 write_seqlock(&net->fs_lock); 150 151 /* Firstly install the server in the UUID lookup tree */ 152 pp = &net->fs_servers.rb_node; 153 p = NULL; 154 while (*pp) { 155 p = *pp; 156 _debug("- consider %p", p); 157 server = rb_entry(p, struct afs_server, uuid_rb); 158 diff = memcmp(&candidate->uuid, &server->uuid, sizeof(uuid_t)); 159 if (diff < 0) { 160 pp = &(*pp)->rb_left; 161 } else if (diff > 0) { 162 pp = &(*pp)->rb_right; 163 } else { 164 if (server->cell == cell) 165 goto exists; 166 167 /* We have the same UUID representing servers in 168 * different cells. Append the new server to the list. 169 */ 170 for (;;) { 171 next = rcu_dereference_protected( 172 server->uuid_next, 173 lockdep_is_held(&net->fs_lock.lock)); 174 if (!next) 175 break; 176 server = next; 177 } 178 rcu_assign_pointer(server->uuid_next, candidate); 179 candidate->uuid_prev = server; 180 server = candidate; 181 goto added_dup; 182 } 183 } 184 185 server = candidate; 186 rb_link_node(&server->uuid_rb, p, pp); 187 rb_insert_color(&server->uuid_rb, &net->fs_servers); 188 hlist_add_head_rcu(&server->proc_link, &net->fs_proc); 189 190added_dup: 191 write_seqlock(&net->fs_addr_lock); 192 alist = rcu_dereference_protected(server->addresses, 193 lockdep_is_held(&net->fs_addr_lock.lock)); 194 195 /* Secondly, if the server has any IPv4 and/or IPv6 addresses, install 196 * it in the IPv4 and/or IPv6 reverse-map lists. 197 * 198 * TODO: For speed we want to use something other than a flat list 199 * here; even sorting the list in terms of lowest address would help a 200 * bit, but anything we might want to do gets messy and memory 201 * intensive. 202 */ 203 if (alist->nr_ipv4 > 0) 204 hlist_add_head_rcu(&server->addr4_link, &net->fs_addresses4); 205 if (alist->nr_addrs > alist->nr_ipv4) 206 hlist_add_head_rcu(&server->addr6_link, &net->fs_addresses6); 207 208 write_sequnlock(&net->fs_addr_lock); 209 210exists: 211 afs_get_server(server, afs_server_trace_get_install); 212 write_sequnlock(&net->fs_lock); 213 return server; 214} 215 216/* 217 * Allocate a new server record and mark it active. 218 */ 219static struct afs_server *afs_alloc_server(struct afs_cell *cell, 220 const uuid_t *uuid, 221 struct afs_addr_list *alist) 222{ 223 struct afs_server *server; 224 struct afs_net *net = cell->net; 225 226 _enter(""); 227 228 server = kzalloc(sizeof(struct afs_server), GFP_KERNEL); 229 if (!server) 230 goto enomem; 231 232 refcount_set(&server->ref, 1); 233 atomic_set(&server->active, 1); 234 server->debug_id = atomic_inc_return(&afs_server_debug_id); 235 RCU_INIT_POINTER(server->addresses, alist); 236 server->addr_version = alist->version; 237 server->uuid = *uuid; 238 rwlock_init(&server->fs_lock); 239 INIT_WORK(&server->initcb_work, afs_server_init_callback_work); 240 init_waitqueue_head(&server->probe_wq); 241 INIT_LIST_HEAD(&server->probe_link); 242 spin_lock_init(&server->probe_lock); 243 server->cell = cell; 244 server->rtt = UINT_MAX; 245 246 afs_inc_servers_outstanding(net); 247 trace_afs_server(server->debug_id, 1, 1, afs_server_trace_alloc); 248 _leave(" = %p", server); 249 return server; 250 251enomem: 252 _leave(" = NULL [nomem]"); 253 return NULL; 254} 255 256/* 257 * Look up an address record for a server 258 */ 259static struct afs_addr_list *afs_vl_lookup_addrs(struct afs_cell *cell, 260 struct key *key, const uuid_t *uuid) 261{ 262 struct afs_vl_cursor vc; 263 struct afs_addr_list *alist = NULL; 264 int ret; 265 266 ret = -ERESTARTSYS; 267 if (afs_begin_vlserver_operation(&vc, cell, key)) { 268 while (afs_select_vlserver(&vc)) { 269 if (test_bit(AFS_VLSERVER_FL_IS_YFS, &vc.server->flags)) 270 alist = afs_yfsvl_get_endpoints(&vc, uuid); 271 else 272 alist = afs_vl_get_addrs_u(&vc, uuid); 273 } 274 275 ret = afs_end_vlserver_operation(&vc); 276 } 277 278 return ret < 0 ? ERR_PTR(ret) : alist; 279} 280 281/* 282 * Get or create a fileserver record. 283 */ 284struct afs_server *afs_lookup_server(struct afs_cell *cell, struct key *key, 285 const uuid_t *uuid, u32 addr_version) 286{ 287 struct afs_addr_list *alist; 288 struct afs_server *server, *candidate; 289 290 _enter("%p,%pU", cell->net, uuid); 291 292 server = afs_find_server_by_uuid(cell->net, uuid); 293 if (server) { 294 if (server->addr_version != addr_version) 295 set_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags); 296 return server; 297 } 298 299 alist = afs_vl_lookup_addrs(cell, key, uuid); 300 if (IS_ERR(alist)) 301 return ERR_CAST(alist); 302 303 candidate = afs_alloc_server(cell, uuid, alist); 304 if (!candidate) { 305 afs_put_addrlist(alist); 306 return ERR_PTR(-ENOMEM); 307 } 308 309 server = afs_install_server(cell, candidate); 310 if (server != candidate) { 311 afs_put_addrlist(alist); 312 kfree(candidate); 313 } else { 314 /* Immediately dispatch an asynchronous probe to each interface 315 * on the fileserver. This will make sure the repeat-probing 316 * service is started. 317 */ 318 afs_fs_probe_fileserver(cell->net, server, key, true); 319 } 320 321 return server; 322} 323 324/* 325 * Set the server timer to fire after a given delay, assuming it's not already 326 * set for an earlier time. 327 */ 328static void afs_set_server_timer(struct afs_net *net, time64_t delay) 329{ 330 if (net->live) { 331 afs_inc_servers_outstanding(net); 332 if (timer_reduce(&net->fs_timer, jiffies + delay * HZ)) 333 afs_dec_servers_outstanding(net); 334 } 335} 336 337/* 338 * Server management timer. We have an increment on fs_outstanding that we 339 * need to pass along to the work item. 340 */ 341void afs_servers_timer(struct timer_list *timer) 342{ 343 struct afs_net *net = container_of(timer, struct afs_net, fs_timer); 344 345 _enter(""); 346 if (!queue_work(afs_wq, &net->fs_manager)) 347 afs_dec_servers_outstanding(net); 348} 349 350/* 351 * Get a reference on a server object. 352 */ 353struct afs_server *afs_get_server(struct afs_server *server, 354 enum afs_server_trace reason) 355{ 356 unsigned int a; 357 int r; 358 359 __refcount_inc(&server->ref, &r); 360 a = atomic_read(&server->active); 361 trace_afs_server(server->debug_id, r + 1, a, reason); 362 return server; 363} 364 365/* 366 * Try to get a reference on a server object. 367 */ 368static struct afs_server *afs_maybe_use_server(struct afs_server *server, 369 enum afs_server_trace reason) 370{ 371 unsigned int a; 372 int r; 373 374 if (!__refcount_inc_not_zero(&server->ref, &r)) 375 return NULL; 376 377 a = atomic_inc_return(&server->active); 378 trace_afs_server(server->debug_id, r + 1, a, reason); 379 return server; 380} 381 382/* 383 * Get an active count on a server object. 384 */ 385struct afs_server *afs_use_server(struct afs_server *server, enum afs_server_trace reason) 386{ 387 unsigned int a; 388 int r; 389 390 __refcount_inc(&server->ref, &r); 391 a = atomic_inc_return(&server->active); 392 393 trace_afs_server(server->debug_id, r + 1, a, reason); 394 return server; 395} 396 397/* 398 * Release a reference on a server record. 399 */ 400void afs_put_server(struct afs_net *net, struct afs_server *server, 401 enum afs_server_trace reason) 402{ 403 unsigned int a, debug_id = server->debug_id; 404 bool zero; 405 int r; 406 407 if (!server) 408 return; 409 410 a = atomic_read(&server->active); 411 zero = __refcount_dec_and_test(&server->ref, &r); 412 trace_afs_server(debug_id, r - 1, a, reason); 413 if (unlikely(zero)) 414 __afs_put_server(net, server); 415} 416 417/* 418 * Drop an active count on a server object without updating the last-unused 419 * time. 420 */ 421void afs_unuse_server_notime(struct afs_net *net, struct afs_server *server, 422 enum afs_server_trace reason) 423{ 424 if (server) { 425 unsigned int active = atomic_dec_return(&server->active); 426 427 if (active == 0) 428 afs_set_server_timer(net, afs_server_gc_delay); 429 afs_put_server(net, server, reason); 430 } 431} 432 433/* 434 * Drop an active count on a server object. 435 */ 436void afs_unuse_server(struct afs_net *net, struct afs_server *server, 437 enum afs_server_trace reason) 438{ 439 if (server) { 440 server->unuse_time = ktime_get_real_seconds(); 441 afs_unuse_server_notime(net, server, reason); 442 } 443} 444 445static void afs_server_rcu(struct rcu_head *rcu) 446{ 447 struct afs_server *server = container_of(rcu, struct afs_server, rcu); 448 449 trace_afs_server(server->debug_id, refcount_read(&server->ref), 450 atomic_read(&server->active), afs_server_trace_free); 451 afs_put_addrlist(rcu_access_pointer(server->addresses)); 452 kfree(server); 453} 454 455static void __afs_put_server(struct afs_net *net, struct afs_server *server) 456{ 457 call_rcu(&server->rcu, afs_server_rcu); 458 afs_dec_servers_outstanding(net); 459} 460 461static void afs_give_up_callbacks(struct afs_net *net, struct afs_server *server) 462{ 463 struct afs_addr_list *alist = rcu_access_pointer(server->addresses); 464 struct afs_addr_cursor ac = { 465 .alist = alist, 466 .index = alist->preferred, 467 .error = 0, 468 }; 469 470 afs_fs_give_up_all_callbacks(net, server, &ac, NULL); 471} 472 473/* 474 * destroy a dead server 475 */ 476static void afs_destroy_server(struct afs_net *net, struct afs_server *server) 477{ 478 if (test_bit(AFS_SERVER_FL_MAY_HAVE_CB, &server->flags)) 479 afs_give_up_callbacks(net, server); 480 481 flush_work(&server->initcb_work); 482 afs_put_server(net, server, afs_server_trace_destroy); 483} 484 485/* 486 * Garbage collect any expired servers. 487 */ 488static void afs_gc_servers(struct afs_net *net, struct afs_server *gc_list) 489{ 490 struct afs_server *server, *next, *prev; 491 int active; 492 493 while ((server = gc_list)) { 494 gc_list = server->gc_next; 495 496 write_seqlock(&net->fs_lock); 497 498 active = atomic_read(&server->active); 499 if (active == 0) { 500 trace_afs_server(server->debug_id, refcount_read(&server->ref), 501 active, afs_server_trace_gc); 502 next = rcu_dereference_protected( 503 server->uuid_next, lockdep_is_held(&net->fs_lock.lock)); 504 prev = server->uuid_prev; 505 if (!prev) { 506 /* The one at the front is in the tree */ 507 if (!next) { 508 rb_erase(&server->uuid_rb, &net->fs_servers); 509 } else { 510 rb_replace_node_rcu(&server->uuid_rb, 511 &next->uuid_rb, 512 &net->fs_servers); 513 next->uuid_prev = NULL; 514 } 515 } else { 516 /* This server is not at the front */ 517 rcu_assign_pointer(prev->uuid_next, next); 518 if (next) 519 next->uuid_prev = prev; 520 } 521 522 list_del(&server->probe_link); 523 hlist_del_rcu(&server->proc_link); 524 if (!hlist_unhashed(&server->addr4_link)) 525 hlist_del_rcu(&server->addr4_link); 526 if (!hlist_unhashed(&server->addr6_link)) 527 hlist_del_rcu(&server->addr6_link); 528 } 529 write_sequnlock(&net->fs_lock); 530 531 if (active == 0) 532 afs_destroy_server(net, server); 533 } 534} 535 536/* 537 * Manage the records of servers known to be within a network namespace. This 538 * includes garbage collecting unused servers. 539 * 540 * Note also that we were given an increment on net->servers_outstanding by 541 * whoever queued us that we need to deal with before returning. 542 */ 543void afs_manage_servers(struct work_struct *work) 544{ 545 struct afs_net *net = container_of(work, struct afs_net, fs_manager); 546 struct afs_server *gc_list = NULL; 547 struct rb_node *cursor; 548 time64_t now = ktime_get_real_seconds(), next_manage = TIME64_MAX; 549 bool purging = !net->live; 550 551 _enter(""); 552 553 /* Trawl the server list looking for servers that have expired from 554 * lack of use. 555 */ 556 read_seqlock_excl(&net->fs_lock); 557 558 for (cursor = rb_first(&net->fs_servers); cursor; cursor = rb_next(cursor)) { 559 struct afs_server *server = 560 rb_entry(cursor, struct afs_server, uuid_rb); 561 int active = atomic_read(&server->active); 562 563 _debug("manage %pU %u", &server->uuid, active); 564 565 if (purging) { 566 trace_afs_server(server->debug_id, refcount_read(&server->ref), 567 active, afs_server_trace_purging); 568 if (active != 0) 569 pr_notice("Can't purge s=%08x\n", server->debug_id); 570 } 571 572 if (active == 0) { 573 time64_t expire_at = server->unuse_time; 574 575 if (!test_bit(AFS_SERVER_FL_VL_FAIL, &server->flags) && 576 !test_bit(AFS_SERVER_FL_NOT_FOUND, &server->flags)) 577 expire_at += afs_server_gc_delay; 578 if (purging || expire_at <= now) { 579 server->gc_next = gc_list; 580 gc_list = server; 581 } else if (expire_at < next_manage) { 582 next_manage = expire_at; 583 } 584 } 585 } 586 587 read_sequnlock_excl(&net->fs_lock); 588 589 /* Update the timer on the way out. We have to pass an increment on 590 * servers_outstanding in the namespace that we are in to the timer or 591 * the work scheduler. 592 */ 593 if (!purging && next_manage < TIME64_MAX) { 594 now = ktime_get_real_seconds(); 595 596 if (next_manage - now <= 0) { 597 if (queue_work(afs_wq, &net->fs_manager)) 598 afs_inc_servers_outstanding(net); 599 } else { 600 afs_set_server_timer(net, next_manage - now); 601 } 602 } 603 604 afs_gc_servers(net, gc_list); 605 606 afs_dec_servers_outstanding(net); 607 _leave(" [%d]", atomic_read(&net->servers_outstanding)); 608} 609 610static void afs_queue_server_manager(struct afs_net *net) 611{ 612 afs_inc_servers_outstanding(net); 613 if (!queue_work(afs_wq, &net->fs_manager)) 614 afs_dec_servers_outstanding(net); 615} 616 617/* 618 * Purge list of servers. 619 */ 620void afs_purge_servers(struct afs_net *net) 621{ 622 _enter(""); 623 624 if (del_timer_sync(&net->fs_timer)) 625 afs_dec_servers_outstanding(net); 626 627 afs_queue_server_manager(net); 628 629 _debug("wait"); 630 atomic_dec(&net->servers_outstanding); 631 wait_var_event(&net->servers_outstanding, 632 !atomic_read(&net->servers_outstanding)); 633 _leave(""); 634} 635 636/* 637 * Get an update for a server's address list. 638 */ 639static noinline bool afs_update_server_record(struct afs_operation *op, 640 struct afs_server *server) 641{ 642 struct afs_addr_list *alist, *discard; 643 644 _enter(""); 645 646 trace_afs_server(server->debug_id, refcount_read(&server->ref), 647 atomic_read(&server->active), 648 afs_server_trace_update); 649 650 alist = afs_vl_lookup_addrs(op->volume->cell, op->key, &server->uuid); 651 if (IS_ERR(alist)) { 652 if ((PTR_ERR(alist) == -ERESTARTSYS || 653 PTR_ERR(alist) == -EINTR) && 654 (op->flags & AFS_OPERATION_UNINTR) && 655 server->addresses) { 656 _leave(" = t [intr]"); 657 return true; 658 } 659 op->error = PTR_ERR(alist); 660 _leave(" = f [%d]", op->error); 661 return false; 662 } 663 664 discard = alist; 665 if (server->addr_version != alist->version) { 666 write_lock(&server->fs_lock); 667 discard = rcu_dereference_protected(server->addresses, 668 lockdep_is_held(&server->fs_lock)); 669 rcu_assign_pointer(server->addresses, alist); 670 server->addr_version = alist->version; 671 write_unlock(&server->fs_lock); 672 } 673 674 afs_put_addrlist(discard); 675 _leave(" = t"); 676 return true; 677} 678 679/* 680 * See if a server's address list needs updating. 681 */ 682bool afs_check_server_record(struct afs_operation *op, struct afs_server *server) 683{ 684 bool success; 685 int ret, retries = 0; 686 687 _enter(""); 688 689 ASSERT(server); 690 691retry: 692 if (test_bit(AFS_SERVER_FL_UPDATING, &server->flags)) 693 goto wait; 694 if (test_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags)) 695 goto update; 696 _leave(" = t [good]"); 697 return true; 698 699update: 700 if (!test_and_set_bit_lock(AFS_SERVER_FL_UPDATING, &server->flags)) { 701 clear_bit(AFS_SERVER_FL_NEEDS_UPDATE, &server->flags); 702 success = afs_update_server_record(op, server); 703 clear_bit_unlock(AFS_SERVER_FL_UPDATING, &server->flags); 704 wake_up_bit(&server->flags, AFS_SERVER_FL_UPDATING); 705 _leave(" = %d", success); 706 return success; 707 } 708 709wait: 710 ret = wait_on_bit(&server->flags, AFS_SERVER_FL_UPDATING, 711 (op->flags & AFS_OPERATION_UNINTR) ? 712 TASK_UNINTERRUPTIBLE : TASK_INTERRUPTIBLE); 713 if (ret == -ERESTARTSYS) { 714 op->error = ret; 715 _leave(" = f [intr]"); 716 return false; 717 } 718 719 retries++; 720 if (retries == 4) { 721 _leave(" = f [stale]"); 722 ret = -ESTALE; 723 return false; 724 } 725 goto retry; 726} 727