David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2 | /* |
| 3 | * Copyright (C) 2008-2014 Mathieu Desnoyers |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 4 | */ |
| 5 | #include <linux/module.h> |
| 6 | #include <linux/mutex.h> |
| 7 | #include <linux/types.h> |
| 8 | #include <linux/jhash.h> |
| 9 | #include <linux/list.h> |
| 10 | #include <linux/rcupdate.h> |
| 11 | #include <linux/tracepoint.h> |
| 12 | #include <linux/err.h> |
| 13 | #include <linux/slab.h> |
| 14 | #include <linux/sched/signal.h> |
| 15 | #include <linux/sched/task.h> |
| 16 | #include <linux/static_key.h> |
| 17 | |
Olivier Deprez | 157378f | 2022-04-04 15:47:50 +0200 | [diff] [blame^] | 18 | enum tp_func_state { |
| 19 | TP_FUNC_0, |
| 20 | TP_FUNC_1, |
| 21 | TP_FUNC_2, |
| 22 | TP_FUNC_N, |
| 23 | }; |
| 24 | |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 25 | extern tracepoint_ptr_t __start___tracepoints_ptrs[]; |
| 26 | extern tracepoint_ptr_t __stop___tracepoints_ptrs[]; |
| 27 | |
| 28 | DEFINE_SRCU(tracepoint_srcu); |
| 29 | EXPORT_SYMBOL_GPL(tracepoint_srcu); |
| 30 | |
Olivier Deprez | 157378f | 2022-04-04 15:47:50 +0200 | [diff] [blame^] | 31 | enum tp_transition_sync { |
| 32 | TP_TRANSITION_SYNC_1_0_1, |
| 33 | TP_TRANSITION_SYNC_N_2_1, |
| 34 | |
| 35 | _NR_TP_TRANSITION_SYNC, |
| 36 | }; |
| 37 | |
| 38 | struct tp_transition_snapshot { |
| 39 | unsigned long rcu; |
| 40 | unsigned long srcu; |
| 41 | bool ongoing; |
| 42 | }; |
| 43 | |
| 44 | /* Protected by tracepoints_mutex */ |
| 45 | static struct tp_transition_snapshot tp_transition_snapshot[_NR_TP_TRANSITION_SYNC]; |
| 46 | |
| 47 | static void tp_rcu_get_state(enum tp_transition_sync sync) |
| 48 | { |
| 49 | struct tp_transition_snapshot *snapshot = &tp_transition_snapshot[sync]; |
| 50 | |
| 51 | /* Keep the latest get_state snapshot. */ |
| 52 | snapshot->rcu = get_state_synchronize_rcu(); |
| 53 | snapshot->srcu = start_poll_synchronize_srcu(&tracepoint_srcu); |
| 54 | snapshot->ongoing = true; |
| 55 | } |
| 56 | |
| 57 | static void tp_rcu_cond_sync(enum tp_transition_sync sync) |
| 58 | { |
| 59 | struct tp_transition_snapshot *snapshot = &tp_transition_snapshot[sync]; |
| 60 | |
| 61 | if (!snapshot->ongoing) |
| 62 | return; |
| 63 | cond_synchronize_rcu(snapshot->rcu); |
| 64 | if (!poll_state_synchronize_srcu(&tracepoint_srcu, snapshot->srcu)) |
| 65 | synchronize_srcu(&tracepoint_srcu); |
| 66 | snapshot->ongoing = false; |
| 67 | } |
| 68 | |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 69 | /* Set to 1 to enable tracepoint debug output */ |
| 70 | static const int tracepoint_debug; |
| 71 | |
| 72 | #ifdef CONFIG_MODULES |
| 73 | /* |
| 74 | * Tracepoint module list mutex protects the local module list. |
| 75 | */ |
| 76 | static DEFINE_MUTEX(tracepoint_module_list_mutex); |
| 77 | |
| 78 | /* Local list of struct tp_module */ |
| 79 | static LIST_HEAD(tracepoint_module_list); |
| 80 | #endif /* CONFIG_MODULES */ |
| 81 | |
| 82 | /* |
| 83 | * tracepoints_mutex protects the builtin and module tracepoints. |
| 84 | * tracepoints_mutex nests inside tracepoint_module_list_mutex. |
| 85 | */ |
| 86 | static DEFINE_MUTEX(tracepoints_mutex); |
| 87 | |
| 88 | static struct rcu_head *early_probes; |
| 89 | static bool ok_to_free_tracepoints; |
| 90 | |
| 91 | /* |
| 92 | * Note about RCU : |
| 93 | * It is used to delay the free of multiple probes array until a quiescent |
| 94 | * state is reached. |
| 95 | */ |
| 96 | struct tp_probes { |
| 97 | struct rcu_head rcu; |
Olivier Deprez | 157378f | 2022-04-04 15:47:50 +0200 | [diff] [blame^] | 98 | struct tracepoint_func probes[]; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 99 | }; |
| 100 | |
Olivier Deprez | 0e64123 | 2021-09-23 10:07:05 +0200 | [diff] [blame] | 101 | /* Called in removal of a func but failed to allocate a new tp_funcs */ |
| 102 | static void tp_stub_func(void) |
| 103 | { |
| 104 | return; |
| 105 | } |
| 106 | |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 107 | static inline void *allocate_probes(int count) |
| 108 | { |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 109 | struct tp_probes *p = kmalloc(struct_size(p, probes, count), |
| 110 | GFP_KERNEL); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 111 | return p == NULL ? NULL : p->probes; |
| 112 | } |
| 113 | |
| 114 | static void srcu_free_old_probes(struct rcu_head *head) |
| 115 | { |
| 116 | kfree(container_of(head, struct tp_probes, rcu)); |
| 117 | } |
| 118 | |
| 119 | static void rcu_free_old_probes(struct rcu_head *head) |
| 120 | { |
| 121 | call_srcu(&tracepoint_srcu, head, srcu_free_old_probes); |
| 122 | } |
| 123 | |
| 124 | static __init int release_early_probes(void) |
| 125 | { |
| 126 | struct rcu_head *tmp; |
| 127 | |
| 128 | ok_to_free_tracepoints = true; |
| 129 | |
| 130 | while (early_probes) { |
| 131 | tmp = early_probes; |
| 132 | early_probes = tmp->next; |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 133 | call_rcu(tmp, rcu_free_old_probes); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 134 | } |
| 135 | |
| 136 | return 0; |
| 137 | } |
| 138 | |
| 139 | /* SRCU is initialized at core_initcall */ |
| 140 | postcore_initcall(release_early_probes); |
| 141 | |
| 142 | static inline void release_probes(struct tracepoint_func *old) |
| 143 | { |
| 144 | if (old) { |
| 145 | struct tp_probes *tp_probes = container_of(old, |
| 146 | struct tp_probes, probes[0]); |
| 147 | |
| 148 | /* |
| 149 | * We can't free probes if SRCU is not initialized yet. |
| 150 | * Postpone the freeing till after SRCU is initialized. |
| 151 | */ |
| 152 | if (unlikely(!ok_to_free_tracepoints)) { |
| 153 | tp_probes->rcu.next = early_probes; |
| 154 | early_probes = &tp_probes->rcu; |
| 155 | return; |
| 156 | } |
| 157 | |
| 158 | /* |
| 159 | * Tracepoint probes are protected by both sched RCU and SRCU, |
| 160 | * by calling the SRCU callback in the sched RCU callback we |
| 161 | * cover both cases. So let us chain the SRCU and sched RCU |
| 162 | * callbacks to wait for both grace periods. |
| 163 | */ |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 164 | call_rcu(&tp_probes->rcu, rcu_free_old_probes); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 165 | } |
| 166 | } |
| 167 | |
| 168 | static void debug_print_probes(struct tracepoint_func *funcs) |
| 169 | { |
| 170 | int i; |
| 171 | |
| 172 | if (!tracepoint_debug || !funcs) |
| 173 | return; |
| 174 | |
| 175 | for (i = 0; funcs[i].func; i++) |
| 176 | printk(KERN_DEBUG "Probe %d : %p\n", i, funcs[i].func); |
| 177 | } |
| 178 | |
| 179 | static struct tracepoint_func * |
| 180 | func_add(struct tracepoint_func **funcs, struct tracepoint_func *tp_func, |
| 181 | int prio) |
| 182 | { |
| 183 | struct tracepoint_func *old, *new; |
| 184 | int nr_probes = 0; |
Olivier Deprez | 0e64123 | 2021-09-23 10:07:05 +0200 | [diff] [blame] | 185 | int stub_funcs = 0; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 186 | int pos = -1; |
| 187 | |
| 188 | if (WARN_ON(!tp_func->func)) |
| 189 | return ERR_PTR(-EINVAL); |
| 190 | |
| 191 | debug_print_probes(*funcs); |
| 192 | old = *funcs; |
| 193 | if (old) { |
| 194 | /* (N -> N+1), (N != 0, 1) probes */ |
| 195 | for (nr_probes = 0; old[nr_probes].func; nr_probes++) { |
| 196 | /* Insert before probes of lower priority */ |
| 197 | if (pos < 0 && old[nr_probes].prio < prio) |
| 198 | pos = nr_probes; |
| 199 | if (old[nr_probes].func == tp_func->func && |
| 200 | old[nr_probes].data == tp_func->data) |
| 201 | return ERR_PTR(-EEXIST); |
Olivier Deprez | 0e64123 | 2021-09-23 10:07:05 +0200 | [diff] [blame] | 202 | if (old[nr_probes].func == tp_stub_func) |
| 203 | stub_funcs++; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 204 | } |
| 205 | } |
Olivier Deprez | 0e64123 | 2021-09-23 10:07:05 +0200 | [diff] [blame] | 206 | /* + 2 : one for new probe, one for NULL func - stub functions */ |
| 207 | new = allocate_probes(nr_probes + 2 - stub_funcs); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 208 | if (new == NULL) |
| 209 | return ERR_PTR(-ENOMEM); |
| 210 | if (old) { |
Olivier Deprez | 0e64123 | 2021-09-23 10:07:05 +0200 | [diff] [blame] | 211 | if (stub_funcs) { |
| 212 | /* Need to copy one at a time to remove stubs */ |
| 213 | int probes = 0; |
| 214 | |
| 215 | pos = -1; |
| 216 | for (nr_probes = 0; old[nr_probes].func; nr_probes++) { |
| 217 | if (old[nr_probes].func == tp_stub_func) |
| 218 | continue; |
| 219 | if (pos < 0 && old[nr_probes].prio < prio) |
| 220 | pos = probes++; |
| 221 | new[probes++] = old[nr_probes]; |
| 222 | } |
| 223 | nr_probes = probes; |
| 224 | if (pos < 0) |
| 225 | pos = probes; |
| 226 | else |
| 227 | nr_probes--; /* Account for insertion */ |
| 228 | |
| 229 | } else if (pos < 0) { |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 230 | pos = nr_probes; |
| 231 | memcpy(new, old, nr_probes * sizeof(struct tracepoint_func)); |
| 232 | } else { |
| 233 | /* Copy higher priority probes ahead of the new probe */ |
| 234 | memcpy(new, old, pos * sizeof(struct tracepoint_func)); |
| 235 | /* Copy the rest after it. */ |
| 236 | memcpy(new + pos + 1, old + pos, |
| 237 | (nr_probes - pos) * sizeof(struct tracepoint_func)); |
| 238 | } |
| 239 | } else |
| 240 | pos = 0; |
| 241 | new[pos] = *tp_func; |
| 242 | new[nr_probes + 1].func = NULL; |
| 243 | *funcs = new; |
| 244 | debug_print_probes(*funcs); |
| 245 | return old; |
| 246 | } |
| 247 | |
| 248 | static void *func_remove(struct tracepoint_func **funcs, |
| 249 | struct tracepoint_func *tp_func) |
| 250 | { |
| 251 | int nr_probes = 0, nr_del = 0, i; |
| 252 | struct tracepoint_func *old, *new; |
| 253 | |
| 254 | old = *funcs; |
| 255 | |
| 256 | if (!old) |
| 257 | return ERR_PTR(-ENOENT); |
| 258 | |
| 259 | debug_print_probes(*funcs); |
| 260 | /* (N -> M), (N > 1, M >= 0) probes */ |
| 261 | if (tp_func->func) { |
| 262 | for (nr_probes = 0; old[nr_probes].func; nr_probes++) { |
Olivier Deprez | 0e64123 | 2021-09-23 10:07:05 +0200 | [diff] [blame] | 263 | if ((old[nr_probes].func == tp_func->func && |
| 264 | old[nr_probes].data == tp_func->data) || |
| 265 | old[nr_probes].func == tp_stub_func) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 266 | nr_del++; |
| 267 | } |
| 268 | } |
| 269 | |
| 270 | /* |
| 271 | * If probe is NULL, then nr_probes = nr_del = 0, and then the |
| 272 | * entire entry will be removed. |
| 273 | */ |
| 274 | if (nr_probes - nr_del == 0) { |
| 275 | /* N -> 0, (N > 1) */ |
| 276 | *funcs = NULL; |
| 277 | debug_print_probes(*funcs); |
| 278 | return old; |
| 279 | } else { |
| 280 | int j = 0; |
| 281 | /* N -> M, (N > 1, M > 0) */ |
| 282 | /* + 1 for NULL */ |
| 283 | new = allocate_probes(nr_probes - nr_del + 1); |
Olivier Deprez | 0e64123 | 2021-09-23 10:07:05 +0200 | [diff] [blame] | 284 | if (new) { |
| 285 | for (i = 0; old[i].func; i++) |
| 286 | if ((old[i].func != tp_func->func |
| 287 | || old[i].data != tp_func->data) |
| 288 | && old[i].func != tp_stub_func) |
| 289 | new[j++] = old[i]; |
| 290 | new[nr_probes - nr_del].func = NULL; |
| 291 | *funcs = new; |
| 292 | } else { |
| 293 | /* |
| 294 | * Failed to allocate, replace the old function |
| 295 | * with calls to tp_stub_func. |
| 296 | */ |
| 297 | for (i = 0; old[i].func; i++) |
| 298 | if (old[i].func == tp_func->func && |
| 299 | old[i].data == tp_func->data) { |
| 300 | old[i].func = tp_stub_func; |
| 301 | /* Set the prio to the next event. */ |
| 302 | if (old[i + 1].func) |
| 303 | old[i].prio = |
| 304 | old[i + 1].prio; |
| 305 | else |
| 306 | old[i].prio = -1; |
| 307 | } |
| 308 | *funcs = old; |
| 309 | } |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 310 | } |
| 311 | debug_print_probes(*funcs); |
| 312 | return old; |
| 313 | } |
| 314 | |
| 315 | /* |
Olivier Deprez | 157378f | 2022-04-04 15:47:50 +0200 | [diff] [blame^] | 316 | * Count the number of functions (enum tp_func_state) in a tp_funcs array. |
| 317 | */ |
| 318 | static enum tp_func_state nr_func_state(const struct tracepoint_func *tp_funcs) |
| 319 | { |
| 320 | if (!tp_funcs) |
| 321 | return TP_FUNC_0; |
| 322 | if (!tp_funcs[1].func) |
| 323 | return TP_FUNC_1; |
| 324 | if (!tp_funcs[2].func) |
| 325 | return TP_FUNC_2; |
| 326 | return TP_FUNC_N; /* 3 or more */ |
| 327 | } |
| 328 | |
| 329 | static void tracepoint_update_call(struct tracepoint *tp, struct tracepoint_func *tp_funcs) |
| 330 | { |
| 331 | void *func = tp->iterator; |
| 332 | |
| 333 | /* Synthetic events do not have static call sites */ |
| 334 | if (!tp->static_call_key) |
| 335 | return; |
| 336 | if (nr_func_state(tp_funcs) == TP_FUNC_1) |
| 337 | func = tp_funcs[0].func; |
| 338 | __static_call_update(tp->static_call_key, tp->static_call_tramp, func); |
| 339 | } |
| 340 | |
| 341 | /* |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 342 | * Add the probe function to a tracepoint. |
| 343 | */ |
| 344 | static int tracepoint_add_func(struct tracepoint *tp, |
Olivier Deprez | 0e64123 | 2021-09-23 10:07:05 +0200 | [diff] [blame] | 345 | struct tracepoint_func *func, int prio, |
| 346 | bool warn) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 347 | { |
| 348 | struct tracepoint_func *old, *tp_funcs; |
| 349 | int ret; |
| 350 | |
| 351 | if (tp->regfunc && !static_key_enabled(&tp->key)) { |
| 352 | ret = tp->regfunc(); |
| 353 | if (ret < 0) |
| 354 | return ret; |
| 355 | } |
| 356 | |
| 357 | tp_funcs = rcu_dereference_protected(tp->funcs, |
| 358 | lockdep_is_held(&tracepoints_mutex)); |
| 359 | old = func_add(&tp_funcs, func, prio); |
| 360 | if (IS_ERR(old)) { |
Olivier Deprez | 0e64123 | 2021-09-23 10:07:05 +0200 | [diff] [blame] | 361 | WARN_ON_ONCE(warn && PTR_ERR(old) != -ENOMEM); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 362 | return PTR_ERR(old); |
| 363 | } |
| 364 | |
| 365 | /* |
| 366 | * rcu_assign_pointer has as smp_store_release() which makes sure |
| 367 | * that the new probe callbacks array is consistent before setting |
| 368 | * a pointer to it. This array is referenced by __DO_TRACE from |
| 369 | * include/linux/tracepoint.h using rcu_dereference_sched(). |
| 370 | */ |
Olivier Deprez | 157378f | 2022-04-04 15:47:50 +0200 | [diff] [blame^] | 371 | switch (nr_func_state(tp_funcs)) { |
| 372 | case TP_FUNC_1: /* 0->1 */ |
| 373 | /* |
| 374 | * Make sure new static func never uses old data after a |
| 375 | * 1->0->1 transition sequence. |
| 376 | */ |
| 377 | tp_rcu_cond_sync(TP_TRANSITION_SYNC_1_0_1); |
| 378 | /* Set static call to first function */ |
| 379 | tracepoint_update_call(tp, tp_funcs); |
| 380 | /* Both iterator and static call handle NULL tp->funcs */ |
| 381 | rcu_assign_pointer(tp->funcs, tp_funcs); |
| 382 | static_key_enable(&tp->key); |
| 383 | break; |
| 384 | case TP_FUNC_2: /* 1->2 */ |
| 385 | /* Set iterator static call */ |
| 386 | tracepoint_update_call(tp, tp_funcs); |
| 387 | /* |
| 388 | * Iterator callback installed before updating tp->funcs. |
| 389 | * Requires ordering between RCU assign/dereference and |
| 390 | * static call update/call. |
| 391 | */ |
| 392 | fallthrough; |
| 393 | case TP_FUNC_N: /* N->N+1 (N>1) */ |
| 394 | rcu_assign_pointer(tp->funcs, tp_funcs); |
| 395 | /* |
| 396 | * Make sure static func never uses incorrect data after a |
| 397 | * N->...->2->1 (N>1) transition sequence. |
| 398 | */ |
| 399 | if (tp_funcs[0].data != old[0].data) |
| 400 | tp_rcu_get_state(TP_TRANSITION_SYNC_N_2_1); |
| 401 | break; |
| 402 | default: |
| 403 | WARN_ON_ONCE(1); |
| 404 | break; |
| 405 | } |
| 406 | |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 407 | release_probes(old); |
| 408 | return 0; |
| 409 | } |
| 410 | |
| 411 | /* |
| 412 | * Remove a probe function from a tracepoint. |
| 413 | * Note: only waiting an RCU period after setting elem->call to the empty |
| 414 | * function insures that the original callback is not used anymore. This insured |
| 415 | * by preempt_disable around the call site. |
| 416 | */ |
| 417 | static int tracepoint_remove_func(struct tracepoint *tp, |
| 418 | struct tracepoint_func *func) |
| 419 | { |
| 420 | struct tracepoint_func *old, *tp_funcs; |
| 421 | |
| 422 | tp_funcs = rcu_dereference_protected(tp->funcs, |
| 423 | lockdep_is_held(&tracepoints_mutex)); |
| 424 | old = func_remove(&tp_funcs, func); |
Olivier Deprez | 0e64123 | 2021-09-23 10:07:05 +0200 | [diff] [blame] | 425 | if (WARN_ON_ONCE(IS_ERR(old))) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 426 | return PTR_ERR(old); |
Olivier Deprez | 0e64123 | 2021-09-23 10:07:05 +0200 | [diff] [blame] | 427 | |
| 428 | if (tp_funcs == old) |
| 429 | /* Failed allocating new tp_funcs, replaced func with stub */ |
| 430 | return 0; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 431 | |
Olivier Deprez | 157378f | 2022-04-04 15:47:50 +0200 | [diff] [blame^] | 432 | switch (nr_func_state(tp_funcs)) { |
| 433 | case TP_FUNC_0: /* 1->0 */ |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 434 | /* Removed last function */ |
| 435 | if (tp->unregfunc && static_key_enabled(&tp->key)) |
| 436 | tp->unregfunc(); |
| 437 | |
Olivier Deprez | 157378f | 2022-04-04 15:47:50 +0200 | [diff] [blame^] | 438 | static_key_disable(&tp->key); |
| 439 | /* Set iterator static call */ |
| 440 | tracepoint_update_call(tp, tp_funcs); |
| 441 | /* Both iterator and static call handle NULL tp->funcs */ |
| 442 | rcu_assign_pointer(tp->funcs, NULL); |
| 443 | /* |
| 444 | * Make sure new static func never uses old data after a |
| 445 | * 1->0->1 transition sequence. |
| 446 | */ |
| 447 | tp_rcu_get_state(TP_TRANSITION_SYNC_1_0_1); |
| 448 | break; |
| 449 | case TP_FUNC_1: /* 2->1 */ |
| 450 | rcu_assign_pointer(tp->funcs, tp_funcs); |
| 451 | /* |
| 452 | * Make sure static func never uses incorrect data after a |
| 453 | * N->...->2->1 (N>2) transition sequence. If the first |
| 454 | * element's data has changed, then force the synchronization |
| 455 | * to prevent current readers that have loaded the old data |
| 456 | * from calling the new function. |
| 457 | */ |
| 458 | if (tp_funcs[0].data != old[0].data) |
| 459 | tp_rcu_get_state(TP_TRANSITION_SYNC_N_2_1); |
| 460 | tp_rcu_cond_sync(TP_TRANSITION_SYNC_N_2_1); |
| 461 | /* Set static call to first function */ |
| 462 | tracepoint_update_call(tp, tp_funcs); |
| 463 | break; |
| 464 | case TP_FUNC_2: /* N->N-1 (N>2) */ |
| 465 | fallthrough; |
| 466 | case TP_FUNC_N: |
| 467 | rcu_assign_pointer(tp->funcs, tp_funcs); |
| 468 | /* |
| 469 | * Make sure static func never uses incorrect data after a |
| 470 | * N->...->2->1 (N>2) transition sequence. |
| 471 | */ |
| 472 | if (tp_funcs[0].data != old[0].data) |
| 473 | tp_rcu_get_state(TP_TRANSITION_SYNC_N_2_1); |
| 474 | break; |
| 475 | default: |
| 476 | WARN_ON_ONCE(1); |
| 477 | break; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 478 | } |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 479 | release_probes(old); |
| 480 | return 0; |
| 481 | } |
| 482 | |
| 483 | /** |
Olivier Deprez | 0e64123 | 2021-09-23 10:07:05 +0200 | [diff] [blame] | 484 | * tracepoint_probe_register_prio_may_exist - Connect a probe to a tracepoint with priority |
| 485 | * @tp: tracepoint |
| 486 | * @probe: probe handler |
| 487 | * @data: tracepoint data |
| 488 | * @prio: priority of this function over other registered functions |
| 489 | * |
| 490 | * Same as tracepoint_probe_register_prio() except that it will not warn |
| 491 | * if the tracepoint is already registered. |
| 492 | */ |
| 493 | int tracepoint_probe_register_prio_may_exist(struct tracepoint *tp, void *probe, |
| 494 | void *data, int prio) |
| 495 | { |
| 496 | struct tracepoint_func tp_func; |
| 497 | int ret; |
| 498 | |
| 499 | mutex_lock(&tracepoints_mutex); |
| 500 | tp_func.func = probe; |
| 501 | tp_func.data = data; |
| 502 | tp_func.prio = prio; |
| 503 | ret = tracepoint_add_func(tp, &tp_func, prio, false); |
| 504 | mutex_unlock(&tracepoints_mutex); |
| 505 | return ret; |
| 506 | } |
| 507 | EXPORT_SYMBOL_GPL(tracepoint_probe_register_prio_may_exist); |
| 508 | |
| 509 | /** |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 510 | * tracepoint_probe_register_prio - Connect a probe to a tracepoint with priority |
| 511 | * @tp: tracepoint |
| 512 | * @probe: probe handler |
| 513 | * @data: tracepoint data |
| 514 | * @prio: priority of this function over other registered functions |
| 515 | * |
| 516 | * Returns 0 if ok, error value on error. |
| 517 | * Note: if @tp is within a module, the caller is responsible for |
| 518 | * unregistering the probe before the module is gone. This can be |
| 519 | * performed either with a tracepoint module going notifier, or from |
| 520 | * within module exit functions. |
| 521 | */ |
| 522 | int tracepoint_probe_register_prio(struct tracepoint *tp, void *probe, |
| 523 | void *data, int prio) |
| 524 | { |
| 525 | struct tracepoint_func tp_func; |
| 526 | int ret; |
| 527 | |
| 528 | mutex_lock(&tracepoints_mutex); |
| 529 | tp_func.func = probe; |
| 530 | tp_func.data = data; |
| 531 | tp_func.prio = prio; |
Olivier Deprez | 0e64123 | 2021-09-23 10:07:05 +0200 | [diff] [blame] | 532 | ret = tracepoint_add_func(tp, &tp_func, prio, true); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 533 | mutex_unlock(&tracepoints_mutex); |
| 534 | return ret; |
| 535 | } |
| 536 | EXPORT_SYMBOL_GPL(tracepoint_probe_register_prio); |
| 537 | |
| 538 | /** |
| 539 | * tracepoint_probe_register - Connect a probe to a tracepoint |
| 540 | * @tp: tracepoint |
| 541 | * @probe: probe handler |
| 542 | * @data: tracepoint data |
| 543 | * |
| 544 | * Returns 0 if ok, error value on error. |
| 545 | * Note: if @tp is within a module, the caller is responsible for |
| 546 | * unregistering the probe before the module is gone. This can be |
| 547 | * performed either with a tracepoint module going notifier, or from |
| 548 | * within module exit functions. |
| 549 | */ |
| 550 | int tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data) |
| 551 | { |
| 552 | return tracepoint_probe_register_prio(tp, probe, data, TRACEPOINT_DEFAULT_PRIO); |
| 553 | } |
| 554 | EXPORT_SYMBOL_GPL(tracepoint_probe_register); |
| 555 | |
| 556 | /** |
| 557 | * tracepoint_probe_unregister - Disconnect a probe from a tracepoint |
| 558 | * @tp: tracepoint |
| 559 | * @probe: probe function pointer |
| 560 | * @data: tracepoint data |
| 561 | * |
| 562 | * Returns 0 if ok, error value on error. |
| 563 | */ |
| 564 | int tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data) |
| 565 | { |
| 566 | struct tracepoint_func tp_func; |
| 567 | int ret; |
| 568 | |
| 569 | mutex_lock(&tracepoints_mutex); |
| 570 | tp_func.func = probe; |
| 571 | tp_func.data = data; |
| 572 | ret = tracepoint_remove_func(tp, &tp_func); |
| 573 | mutex_unlock(&tracepoints_mutex); |
| 574 | return ret; |
| 575 | } |
| 576 | EXPORT_SYMBOL_GPL(tracepoint_probe_unregister); |
| 577 | |
| 578 | static void for_each_tracepoint_range( |
| 579 | tracepoint_ptr_t *begin, tracepoint_ptr_t *end, |
| 580 | void (*fct)(struct tracepoint *tp, void *priv), |
| 581 | void *priv) |
| 582 | { |
| 583 | tracepoint_ptr_t *iter; |
| 584 | |
| 585 | if (!begin) |
| 586 | return; |
| 587 | for (iter = begin; iter < end; iter++) |
| 588 | fct(tracepoint_ptr_deref(iter), priv); |
| 589 | } |
| 590 | |
| 591 | #ifdef CONFIG_MODULES |
| 592 | bool trace_module_has_bad_taint(struct module *mod) |
| 593 | { |
| 594 | return mod->taints & ~((1 << TAINT_OOT_MODULE) | (1 << TAINT_CRAP) | |
| 595 | (1 << TAINT_UNSIGNED_MODULE)); |
| 596 | } |
| 597 | |
| 598 | static BLOCKING_NOTIFIER_HEAD(tracepoint_notify_list); |
| 599 | |
| 600 | /** |
| 601 | * register_tracepoint_notifier - register tracepoint coming/going notifier |
| 602 | * @nb: notifier block |
| 603 | * |
| 604 | * Notifiers registered with this function are called on module |
| 605 | * coming/going with the tracepoint_module_list_mutex held. |
| 606 | * The notifier block callback should expect a "struct tp_module" data |
| 607 | * pointer. |
| 608 | */ |
| 609 | int register_tracepoint_module_notifier(struct notifier_block *nb) |
| 610 | { |
| 611 | struct tp_module *tp_mod; |
| 612 | int ret; |
| 613 | |
| 614 | mutex_lock(&tracepoint_module_list_mutex); |
| 615 | ret = blocking_notifier_chain_register(&tracepoint_notify_list, nb); |
| 616 | if (ret) |
| 617 | goto end; |
| 618 | list_for_each_entry(tp_mod, &tracepoint_module_list, list) |
| 619 | (void) nb->notifier_call(nb, MODULE_STATE_COMING, tp_mod); |
| 620 | end: |
| 621 | mutex_unlock(&tracepoint_module_list_mutex); |
| 622 | return ret; |
| 623 | } |
| 624 | EXPORT_SYMBOL_GPL(register_tracepoint_module_notifier); |
| 625 | |
| 626 | /** |
| 627 | * unregister_tracepoint_notifier - unregister tracepoint coming/going notifier |
| 628 | * @nb: notifier block |
| 629 | * |
| 630 | * The notifier block callback should expect a "struct tp_module" data |
| 631 | * pointer. |
| 632 | */ |
| 633 | int unregister_tracepoint_module_notifier(struct notifier_block *nb) |
| 634 | { |
| 635 | struct tp_module *tp_mod; |
| 636 | int ret; |
| 637 | |
| 638 | mutex_lock(&tracepoint_module_list_mutex); |
| 639 | ret = blocking_notifier_chain_unregister(&tracepoint_notify_list, nb); |
| 640 | if (ret) |
| 641 | goto end; |
| 642 | list_for_each_entry(tp_mod, &tracepoint_module_list, list) |
| 643 | (void) nb->notifier_call(nb, MODULE_STATE_GOING, tp_mod); |
| 644 | end: |
| 645 | mutex_unlock(&tracepoint_module_list_mutex); |
| 646 | return ret; |
| 647 | |
| 648 | } |
| 649 | EXPORT_SYMBOL_GPL(unregister_tracepoint_module_notifier); |
| 650 | |
| 651 | /* |
| 652 | * Ensure the tracer unregistered the module's probes before the module |
| 653 | * teardown is performed. Prevents leaks of probe and data pointers. |
| 654 | */ |
| 655 | static void tp_module_going_check_quiescent(struct tracepoint *tp, void *priv) |
| 656 | { |
| 657 | WARN_ON_ONCE(tp->funcs); |
| 658 | } |
| 659 | |
| 660 | static int tracepoint_module_coming(struct module *mod) |
| 661 | { |
| 662 | struct tp_module *tp_mod; |
| 663 | int ret = 0; |
| 664 | |
| 665 | if (!mod->num_tracepoints) |
| 666 | return 0; |
| 667 | |
| 668 | /* |
| 669 | * We skip modules that taint the kernel, especially those with different |
| 670 | * module headers (for forced load), to make sure we don't cause a crash. |
| 671 | * Staging, out-of-tree, and unsigned GPL modules are fine. |
| 672 | */ |
| 673 | if (trace_module_has_bad_taint(mod)) |
| 674 | return 0; |
| 675 | mutex_lock(&tracepoint_module_list_mutex); |
| 676 | tp_mod = kmalloc(sizeof(struct tp_module), GFP_KERNEL); |
| 677 | if (!tp_mod) { |
| 678 | ret = -ENOMEM; |
| 679 | goto end; |
| 680 | } |
| 681 | tp_mod->mod = mod; |
| 682 | list_add_tail(&tp_mod->list, &tracepoint_module_list); |
| 683 | blocking_notifier_call_chain(&tracepoint_notify_list, |
| 684 | MODULE_STATE_COMING, tp_mod); |
| 685 | end: |
| 686 | mutex_unlock(&tracepoint_module_list_mutex); |
| 687 | return ret; |
| 688 | } |
| 689 | |
| 690 | static void tracepoint_module_going(struct module *mod) |
| 691 | { |
| 692 | struct tp_module *tp_mod; |
| 693 | |
| 694 | if (!mod->num_tracepoints) |
| 695 | return; |
| 696 | |
| 697 | mutex_lock(&tracepoint_module_list_mutex); |
| 698 | list_for_each_entry(tp_mod, &tracepoint_module_list, list) { |
| 699 | if (tp_mod->mod == mod) { |
| 700 | blocking_notifier_call_chain(&tracepoint_notify_list, |
| 701 | MODULE_STATE_GOING, tp_mod); |
| 702 | list_del(&tp_mod->list); |
| 703 | kfree(tp_mod); |
| 704 | /* |
| 705 | * Called the going notifier before checking for |
| 706 | * quiescence. |
| 707 | */ |
| 708 | for_each_tracepoint_range(mod->tracepoints_ptrs, |
| 709 | mod->tracepoints_ptrs + mod->num_tracepoints, |
| 710 | tp_module_going_check_quiescent, NULL); |
| 711 | break; |
| 712 | } |
| 713 | } |
| 714 | /* |
| 715 | * In the case of modules that were tainted at "coming", we'll simply |
| 716 | * walk through the list without finding it. We cannot use the "tainted" |
| 717 | * flag on "going", in case a module taints the kernel only after being |
| 718 | * loaded. |
| 719 | */ |
| 720 | mutex_unlock(&tracepoint_module_list_mutex); |
| 721 | } |
| 722 | |
| 723 | static int tracepoint_module_notify(struct notifier_block *self, |
| 724 | unsigned long val, void *data) |
| 725 | { |
| 726 | struct module *mod = data; |
| 727 | int ret = 0; |
| 728 | |
| 729 | switch (val) { |
| 730 | case MODULE_STATE_COMING: |
| 731 | ret = tracepoint_module_coming(mod); |
| 732 | break; |
| 733 | case MODULE_STATE_LIVE: |
| 734 | break; |
| 735 | case MODULE_STATE_GOING: |
| 736 | tracepoint_module_going(mod); |
| 737 | break; |
| 738 | case MODULE_STATE_UNFORMED: |
| 739 | break; |
| 740 | } |
Olivier Deprez | 157378f | 2022-04-04 15:47:50 +0200 | [diff] [blame^] | 741 | return notifier_from_errno(ret); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 742 | } |
| 743 | |
| 744 | static struct notifier_block tracepoint_module_nb = { |
| 745 | .notifier_call = tracepoint_module_notify, |
| 746 | .priority = 0, |
| 747 | }; |
| 748 | |
| 749 | static __init int init_tracepoints(void) |
| 750 | { |
| 751 | int ret; |
| 752 | |
| 753 | ret = register_module_notifier(&tracepoint_module_nb); |
| 754 | if (ret) |
| 755 | pr_warn("Failed to register tracepoint module enter notifier\n"); |
| 756 | |
| 757 | return ret; |
| 758 | } |
| 759 | __initcall(init_tracepoints); |
| 760 | #endif /* CONFIG_MODULES */ |
| 761 | |
| 762 | /** |
| 763 | * for_each_kernel_tracepoint - iteration on all kernel tracepoints |
| 764 | * @fct: callback |
| 765 | * @priv: private data |
| 766 | */ |
| 767 | void for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv), |
| 768 | void *priv) |
| 769 | { |
| 770 | for_each_tracepoint_range(__start___tracepoints_ptrs, |
| 771 | __stop___tracepoints_ptrs, fct, priv); |
| 772 | } |
| 773 | EXPORT_SYMBOL_GPL(for_each_kernel_tracepoint); |
| 774 | |
| 775 | #ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS |
| 776 | |
| 777 | /* NB: reg/unreg are called while guarded with the tracepoints_mutex */ |
| 778 | static int sys_tracepoint_refcount; |
| 779 | |
| 780 | int syscall_regfunc(void) |
| 781 | { |
| 782 | struct task_struct *p, *t; |
| 783 | |
| 784 | if (!sys_tracepoint_refcount) { |
| 785 | read_lock(&tasklist_lock); |
| 786 | for_each_process_thread(p, t) { |
| 787 | set_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT); |
| 788 | } |
| 789 | read_unlock(&tasklist_lock); |
| 790 | } |
| 791 | sys_tracepoint_refcount++; |
| 792 | |
| 793 | return 0; |
| 794 | } |
| 795 | |
| 796 | void syscall_unregfunc(void) |
| 797 | { |
| 798 | struct task_struct *p, *t; |
| 799 | |
| 800 | sys_tracepoint_refcount--; |
| 801 | if (!sys_tracepoint_refcount) { |
| 802 | read_lock(&tasklist_lock); |
| 803 | for_each_process_thread(p, t) { |
| 804 | clear_tsk_thread_flag(t, TIF_SYSCALL_TRACEPOINT); |
| 805 | } |
| 806 | read_unlock(&tasklist_lock); |
| 807 | } |
| 808 | } |
| 809 | #endif |