1 .. _list_rcu_doc: 2 3 Using RCU to Protect Read-Mostly Linked Lists 4 ============================================= 5 6 One of the most common uses of RCU is protecti 7 (``struct list_head`` in list.h). One big adv 8 that all of the required memory ordering is pr 9 This document describes several list-based RCU 10 11 When iterating a list while holding the rcu_re 12 modify the list. The reader is guaranteed to 13 which were added to the list before they acqui 14 and are still on the list when they drop the r 15 Elements which are added to, or removed from t 16 be seen. If the writer calls list_replace_rcu 17 either the old element or the new element; the 18 nor will they see neither. 19 20 21 Example 1: Read-mostly list: Deferred Destruct 22 ---------------------------------------------- 23 24 A widely used usecase for RCU lists in the ker 25 all processes in the system. ``task_struct::ta 26 links all the processes. The list can be trave 27 additions or removals. 28 29 The traversal of the list is done using ``for_ 30 by the 2 macros:: 31 32 #define next_task(p) \ 33 list_entry_rcu((p)->tasks.next 34 35 #define for_each_process(p) \ 36 for (p = &init_task ; (p = nex 37 38 The code traversing the list of all processes 39 40 rcu_read_lock(); 41 for_each_process(p) { 42 /* Do something with p */ 43 } 44 rcu_read_unlock(); 45 46 The simplified and heavily inlined code for re 47 task list is:: 48 49 void release_task(struct task_struct * 50 { 51 write_lock(&tasklist_lock); 52 list_del_rcu(&p->tasks); 53 write_unlock(&tasklist_lock); 54 call_rcu(&p->rcu, delayed_put_ 55 } 56 57 When a process exits, ``release_task()`` calls 58 via __exit_signal() and __unhash_process() und 59 writer lock protection. The list_del_rcu() in 60 the task from the list of all tasks. The ``tas 61 prevents concurrent list additions/removals fr 62 list. Readers using ``for_each_process()`` are 63 ``tasklist_lock``. To prevent readers from not 64 pointers, the ``task_struct`` object is freed 65 grace periods elapse, with the help of call_rc 66 put_task_struct_rcu_user(). This deferring of 67 any readers traversing the list will see valid 68 and deletion/freeing can happen in parallel wi 69 This pattern is also called an **existence loc 70 from invoking the delayed_put_task_struct() ca 71 all existing readers finish, which guarantees 72 object in question will remain in existence un 73 of all RCU readers that might possibly have a 74 75 76 Example 2: Read-Side Action Taken Outside of L 77 ---------------------------------------------- 78 79 Some reader-writer locking use cases compute a 80 the read-side lock, but continue to use that v 81 released. These use cases are often good cand 82 to RCU. One prominent example involves networ 83 Because the packet-routing data tracks the sta 84 of the computer, it will at times contain stal 85 the route has been computed, there is no need 86 static during transmission of the packet. Aft 87 routing table static all you want, but that wo 88 Internet from changing, and it is the state of 89 that really matters. In addition, routing ent 90 or deleted, rather than being modified in plac 91 of the finite speed of light and the non-zero 92 helping make synchronization be lighter weight 93 94 A straightforward example of this type of RCU 95 the system-call auditing support. For example 96 implementation of ``audit_filter_task()`` migh 97 98 static enum audit_state audit_filter_t 99 { 100 struct audit_entry *e; 101 enum audit_state state; 102 103 read_lock(&auditsc_lock); 104 /* Note: audit_filter_mutex he 105 list_for_each_entry(e, &audit_ 106 if (audit_filter_rules 107 if (state == A 108 *key = 109 read_unlock(&a 110 return state; 111 } 112 } 113 read_unlock(&auditsc_lock); 114 return AUDIT_BUILD_CONTEXT; 115 } 116 117 Here the list is searched under the lock, but 118 the corresponding value is returned. By the t 119 on, the list may well have been modified. Thi 120 you are turning auditing off, it is OK to audi 121 122 This means that RCU can be easily applied to t 123 124 static enum audit_state audit_filter_t 125 { 126 struct audit_entry *e; 127 enum audit_state state; 128 129 rcu_read_lock(); 130 /* Note: audit_filter_mutex he 131 list_for_each_entry_rcu(e, &au 132 if (audit_filter_rules 133 if (state == A 134 *key = 135 rcu_read_unloc 136 return state; 137 } 138 } 139 rcu_read_unlock(); 140 return AUDIT_BUILD_CONTEXT; 141 } 142 143 The read_lock() and read_unlock() calls have b 144 and rcu_read_unlock(), respectively, and the l 145 has become list_for_each_entry_rcu(). The **_ 146 primitives add READ_ONCE() and diagnostic chec 147 outside of an RCU read-side critical section. 148 149 The changes to the update side are also straig 150 might be used as follows for deletion and inse 151 versions of audit_del_rule() and audit_add_rul 152 153 static inline int audit_del_rule(struc 154 struc 155 { 156 struct audit_entry *e; 157 158 write_lock(&auditsc_lock); 159 list_for_each_entry(e, list, l 160 if (!audit_compare_rul 161 list_del(&e->l 162 write_unlock(& 163 return 0; 164 } 165 } 166 write_unlock(&auditsc_lock); 167 return -EFAULT; /* No 168 } 169 170 static inline int audit_add_rule(struc 171 struc 172 { 173 write_lock(&auditsc_lock); 174 if (entry->rule.flags & AUDIT_ 175 entry->rule.flags &= ~ 176 list_add(&entry->list, 177 } else { 178 list_add_tail(&entry-> 179 } 180 write_unlock(&auditsc_lock); 181 return 0; 182 } 183 184 Following are the RCU equivalents for these tw 185 186 static inline int audit_del_rule(struc 187 struc 188 { 189 struct audit_entry *e; 190 191 /* No need to use the _rcu ite 192 * deletion routine. */ 193 list_for_each_entry(e, list, l 194 if (!audit_compare_rul 195 list_del_rcu(& 196 call_rcu(&e->r 197 return 0; 198 } 199 } 200 return -EFAULT; /* No 201 } 202 203 static inline int audit_add_rule(struc 204 struc 205 { 206 if (entry->rule.flags & AUDIT_ 207 entry->rule.flags &= ~ 208 list_add_rcu(&entry->l 209 } else { 210 list_add_tail_rcu(&ent 211 } 212 return 0; 213 } 214 215 Normally, the write_lock() and write_unlock() 216 spin_lock() and a spin_unlock(). But in this c 217 ``audit_filter_mutex``, so no additional locki 218 auditsc_lock can therefore be eliminated, sinc 219 need for writers to exclude readers. 220 221 The list_del(), list_add(), and list_add_tail( 222 replaced by list_del_rcu(), list_add_rcu(), an 223 The **_rcu()** list-manipulation primitives ad 224 needed on weakly ordered CPUs. The list_del_r 225 pointer poisoning debug-assist code that would 226 readers to fail spectacularly. 227 228 So, when readers can tolerate stale data and w 229 deleted, without in-place modification, it is 230 231 232 Example 3: Handling In-Place Updates 233 ------------------------------------ 234 235 The system-call auditing code does not update 236 if it did, the reader-writer-locked code to do 237 (assuming only ``field_count`` is updated, oth 238 need to be filled in):: 239 240 static inline int audit_upd_rule(struc 241 struc 242 __u32 243 __u32 244 { 245 struct audit_entry *e; 246 struct audit_entry *ne; 247 248 write_lock(&auditsc_lock); 249 /* Note: audit_filter_mutex he 250 list_for_each_entry(e, list, l 251 if (!audit_compare_rul 252 e->rule.action 253 e->rule.field_ 254 write_unlock(& 255 return 0; 256 } 257 } 258 write_unlock(&auditsc_lock); 259 return -EFAULT; /* No 260 } 261 262 The RCU version creates a copy, updates the co 263 entry with the newly updated entry. This sequ 264 concurrent reads while making a copy to perfor 265 RCU (*read-copy update*) its name. 266 267 The RCU version of audit_upd_rule() is as foll 268 269 static inline int audit_upd_rule(struc 270 struc 271 __u32 272 __u32 273 { 274 struct audit_entry *e; 275 struct audit_entry *ne; 276 277 list_for_each_entry(e, list, l 278 if (!audit_compare_rul 279 ne = kmalloc(s 280 if (ne == NULL 281 return 282 audit_copy_rul 283 ne->rule.actio 284 ne->rule.field 285 list_replace_r 286 call_rcu(&e->r 287 return 0; 288 } 289 } 290 return -EFAULT; /* No 291 } 292 293 Again, this assumes that the caller holds ``au 294 writer lock would become a spinlock in this so 295 296 The update_lsm_rule() does something very simi 297 prefer to look at real Linux-kernel code. 298 299 Another use of this pattern can be found in th 300 tracking table* code in ``ct_limit_set()``. T 301 entries and has a limit on the maximum entries 302 per-zone and hence one *limit* per zone. The 303 through a hashtable using an RCU-managed hlist 304 limit is set, a new limit object is allocated 305 to replace the old limit object with the new o 306 The old limit object is then freed after a gra 307 308 309 Example 4: Eliminating Stale Data 310 --------------------------------- 311 312 The auditing example above tolerates stale dat 313 that are tracking external state. After all, 314 from the time the external state changes befor 315 of the change, and so as noted earlier, a smal 316 RCU-induced staleness is generally not a probl 317 318 However, there are many examples where stale d 319 One example in the Linux kernel is the System 320 function in ipc/shm.c). This code checks a *d 321 per-entry spinlock, and, if the *deleted* flag 322 entry does not exist. For this to be helpful, 323 return holding the per-entry spinlock, as shm_ 324 325 .. _quick_quiz: 326 327 Quick Quiz: 328 For the deleted-flag technique to be h 329 to hold the per-entry lock while retur 330 331 :ref:`Answer to Quick Quiz <quick_quiz_answer> 332 333 If the system-call audit module were to ever n 334 to accomplish this would be to add a ``deleted 335 ``audit_entry`` structure, and modify audit_fi 336 337 static enum audit_state audit_filter_t 338 { 339 struct audit_entry *e; 340 enum audit_state state; 341 342 rcu_read_lock(); 343 list_for_each_entry_rcu(e, &au 344 if (audit_filter_rules 345 spin_lock(&e-> 346 if (e->deleted 347 spin_u 348 rcu_re 349 return 350 } 351 rcu_read_unloc 352 if (state == A 353 *key = 354 return state; 355 } 356 } 357 rcu_read_unlock(); 358 return AUDIT_BUILD_CONTEXT; 359 } 360 361 The ``audit_del_rule()`` function would need t 362 spinlock as follows:: 363 364 static inline int audit_del_rule(struc 365 struc 366 { 367 struct audit_entry *e; 368 369 /* No need to use the _rcu ite 370 * is the only deletion routin 371 list_for_each_entry(e, list, l 372 if (!audit_compare_rul 373 spin_lock(&e-> 374 list_del_rcu(& 375 e->deleted = 1 376 spin_unlock(&e 377 call_rcu(&e->r 378 return 0; 379 } 380 } 381 return -EFAULT; /* No 382 } 383 384 This too assumes that the caller holds ``audit 385 386 Note that this example assumes that entries ar 387 Additional mechanism is required to deal corre 388 performed by audit_upd_rule(). For one thing, 389 need to hold the locks of both the old ``audit 390 while executing the list_replace_rcu(). 391 392 393 Example 5: Skipping Stale Objects 394 --------------------------------- 395 396 For some use cases, reader performance can be 397 stale objects during read-side list traversal, 398 are those that will be removed and destroyed a 399 periods. One such example can be found in the 400 ``CLOCK_REALTIME`` clock is reprogrammed (for 401 of the system time) then all programmed ``time 402 this clock get triggered and processes waiting 403 advance of their scheduled expiry. To facilita 404 are added to an RCU-managed ``cancel_list`` wh 405 ``timerfd_setup_cancel()``:: 406 407 static void timerfd_setup_cancel(struc 408 { 409 spin_lock(&ctx->cancel_lock); 410 if ((ctx->clockid == CLOCK_REA 411 ctx->clockid == CLOCK_REA 412 (flags & TFD_TIMER_ABSTIME 413 if (!ctx->might_cancel 414 ctx->might_can 415 spin_lock(&can 416 list_add_rcu(& 417 spin_unlock(&c 418 } 419 } else { 420 __timerfd_remove_cance 421 } 422 spin_unlock(&ctx->cancel_lock) 423 } 424 425 When a timerfd is freed (fd is closed), then t 426 flag of the timerfd object is cleared, the obj 427 ``cancel_list`` and destroyed, as shown in thi 428 version of timerfd_release():: 429 430 int timerfd_release(struct inode *inod 431 { 432 struct timerfd_ctx *ctx = file 433 434 spin_lock(&ctx->cancel_lock); 435 if (ctx->might_cancel) { 436 ctx->might_cancel = fa 437 spin_lock(&cancel_lock 438 list_del_rcu(&ctx->cli 439 spin_unlock(&cancel_lo 440 } 441 spin_unlock(&ctx->cancel_lock) 442 443 if (isalarm(ctx)) 444 alarm_cancel(&ctx->t.a 445 else 446 hrtimer_cancel(&ctx->t 447 kfree_rcu(ctx, rcu); 448 return 0; 449 } 450 451 If the ``CLOCK_REALTIME`` clock is set, for ex 452 hrtimer framework calls ``timerfd_clock_was_se 453 ``cancel_list`` and wakes up processes waiting 454 the ``cancel_list``, the ``might_cancel`` flag 455 objects:: 456 457 void timerfd_clock_was_set(void) 458 { 459 ktime_t moffs = ktime_mono_to_ 460 struct timerfd_ctx *ctx; 461 unsigned long flags; 462 463 rcu_read_lock(); 464 list_for_each_entry_rcu(ctx, & 465 if (!ctx->might_cancel 466 continue; 467 spin_lock_irqsave(&ctx 468 if (ctx->moffs != moff 469 ctx->moffs = K 470 ctx->ticks++; 471 wake_up_locked 472 } 473 spin_unlock_irqrestore 474 } 475 rcu_read_unlock(); 476 } 477 478 The key point is that because RCU-protected tr 479 ``cancel_list`` happens concurrently with obje 480 sometimes the traversal can access an object t 481 the list. In this example, a flag is used to s 482 483 484 Summary 485 ------- 486 487 Read-mostly list-based data structures that ca 488 the most amenable to use of RCU. The simplest 489 either added or deleted from the data structur 490 in place), but non-atomic in-place modificatio 491 a copy, updating the copy, then replacing the 492 If stale data cannot be tolerated, then a *del 493 in conjunction with a per-entry spinlock in or 494 function to reject newly deleted data. 495 496 .. _quick_quiz_answer: 497 498 Answer to Quick Quiz: 499 For the deleted-flag technique to be h 500 to hold the per-entry lock while retur 501 502 If the search function drops the per-e 503 then the caller will be processing sta 504 is really OK to be processing stale da 505 *deleted* flag. If processing stale d 506 then you need to hold the per-entry lo 507 that uses the value that was returned. 508 509 :ref:`Back to Quick Quiz <quick_quiz>`
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.