| 1 | /* |
| 2 | * lttng-tracepoint.c |
| 3 | * |
| 4 | * LTTng adaptation layer for Linux kernel 3.15+ tracepoints. |
| 5 | * |
| 6 | * Copyright (C) 2014 Mathieu Desnoyers <mathieu.desnoyers@efficios.com> |
| 7 | * |
| 8 | * This library is free software; you can redistribute it and/or |
| 9 | * modify it under the terms of the GNU Lesser General Public |
| 10 | * License as published by the Free Software Foundation; only |
| 11 | * version 2.1 of the License. |
| 12 | * |
| 13 | * This library is distributed in the hope that it will be useful, |
| 14 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 16 | * Lesser General Public License for more details. |
| 17 | * |
| 18 | * You should have received a copy of the GNU Lesser General Public |
| 19 | * License along with this library; if not, write to the Free Software |
| 20 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 21 | */ |
| 22 | |
| 23 | #include <linux/mutex.h> |
| 24 | #include <linux/err.h> |
| 25 | #include <linux/notifier.h> |
| 26 | #include <linux/tracepoint.h> |
| 27 | #include <linux/slab.h> |
| 28 | #include <linux/jhash.h> |
| 29 | #include <linux/module.h> |
| 30 | |
| 31 | #include <lttng-tracepoint.h> |
| 32 | #include <wrapper/list.h> |
| 33 | #include <wrapper/tracepoint.h> |
| 34 | |
| 35 | /* |
| 36 | * Protect the tracepoint table. lttng_tracepoint_mutex nests within |
| 37 | * kernel/tracepoint.c tp_modlist_mutex. kernel/tracepoint.c |
| 38 | * tracepoint_mutex nests within lttng_tracepoint_mutex. |
| 39 | */ |
| 40 | static |
| 41 | DEFINE_MUTEX(lttng_tracepoint_mutex); |
| 42 | |
| 43 | #define TRACEPOINT_HASH_BITS 6 |
| 44 | #define TRACEPOINT_TABLE_SIZE (1 << TRACEPOINT_HASH_BITS) |
| 45 | static |
| 46 | struct hlist_head tracepoint_table[TRACEPOINT_TABLE_SIZE]; |
| 47 | |
| 48 | /* |
| 49 | * The tracepoint entry is the node contained within the hash table. It |
| 50 | * is a mapping from the "string" key to the struct tracepoint pointer. |
| 51 | */ |
| 52 | struct tracepoint_entry { |
| 53 | struct hlist_node hlist; |
| 54 | struct tracepoint *tp; |
| 55 | int refcount; |
| 56 | struct list_head probes; |
| 57 | char name[0]; |
| 58 | }; |
| 59 | |
| 60 | struct lttng_tp_probe { |
| 61 | struct tracepoint_func tp_func; |
| 62 | struct list_head list; |
| 63 | }; |
| 64 | |
| 65 | static |
| 66 | int add_probe(struct tracepoint_entry *e, void *probe, void *data) |
| 67 | { |
| 68 | struct lttng_tp_probe *p; |
| 69 | int found = 0; |
| 70 | |
| 71 | list_for_each_entry(p, &e->probes, list) { |
| 72 | if (p->tp_func.func == probe && p->tp_func.data == data) { |
| 73 | found = 1; |
| 74 | break; |
| 75 | } |
| 76 | } |
| 77 | if (found) |
| 78 | return -EEXIST; |
| 79 | p = kmalloc(sizeof(struct lttng_tp_probe), GFP_KERNEL); |
| 80 | if (!p) |
| 81 | return -ENOMEM; |
| 82 | p->tp_func.func = probe; |
| 83 | p->tp_func.data = data; |
| 84 | list_add(&p->list, &e->probes); |
| 85 | return 0; |
| 86 | } |
| 87 | |
| 88 | static |
| 89 | int remove_probe(struct tracepoint_entry *e, void *probe, void *data) |
| 90 | { |
| 91 | struct lttng_tp_probe *p; |
| 92 | int found = 0; |
| 93 | |
| 94 | list_for_each_entry(p, &e->probes, list) { |
| 95 | if (p->tp_func.func == probe && p->tp_func.data == data) { |
| 96 | found = 1; |
| 97 | break; |
| 98 | } |
| 99 | } |
| 100 | if (found) { |
| 101 | list_del(&p->list); |
| 102 | kfree(p); |
| 103 | return 0; |
| 104 | } else { |
| 105 | WARN_ON(1); |
| 106 | return -ENOENT; |
| 107 | } |
| 108 | } |
| 109 | |
| 110 | /* |
| 111 | * Get tracepoint if the tracepoint is present in the tracepoint hash table. |
| 112 | * Must be called with lttng_tracepoint_mutex held. |
| 113 | * Returns NULL if not present. |
| 114 | */ |
| 115 | static |
| 116 | struct tracepoint_entry *get_tracepoint(const char *name) |
| 117 | { |
| 118 | struct hlist_head *head; |
| 119 | struct tracepoint_entry *e; |
| 120 | u32 hash = jhash(name, strlen(name), 0); |
| 121 | |
| 122 | head = &tracepoint_table[hash & (TRACEPOINT_TABLE_SIZE - 1)]; |
| 123 | lttng_hlist_for_each_entry(e, head, hlist) { |
| 124 | if (!strcmp(name, e->name)) |
| 125 | return e; |
| 126 | } |
| 127 | return NULL; |
| 128 | } |
| 129 | |
| 130 | /* |
| 131 | * Add the tracepoint to the tracepoint hash table. Must be called with |
| 132 | * lttng_tracepoint_mutex held. |
| 133 | */ |
| 134 | static |
| 135 | struct tracepoint_entry *add_tracepoint(const char *name) |
| 136 | { |
| 137 | struct hlist_head *head; |
| 138 | struct tracepoint_entry *e; |
| 139 | size_t name_len = strlen(name) + 1; |
| 140 | u32 hash = jhash(name, name_len - 1, 0); |
| 141 | |
| 142 | head = &tracepoint_table[hash & (TRACEPOINT_TABLE_SIZE - 1)]; |
| 143 | lttng_hlist_for_each_entry(e, head, hlist) { |
| 144 | if (!strcmp(name, e->name)) { |
| 145 | printk(KERN_NOTICE |
| 146 | "tracepoint %s busy\n", name); |
| 147 | return ERR_PTR(-EEXIST); /* Already there */ |
| 148 | } |
| 149 | } |
| 150 | /* |
| 151 | * Using kmalloc here to allocate a variable length element. Could |
| 152 | * cause some memory fragmentation if overused. |
| 153 | */ |
| 154 | e = kmalloc(sizeof(struct tracepoint_entry) + name_len, GFP_KERNEL); |
| 155 | if (!e) |
| 156 | return ERR_PTR(-ENOMEM); |
| 157 | memcpy(&e->name[0], name, name_len); |
| 158 | e->tp = NULL; |
| 159 | e->refcount = 0; |
| 160 | INIT_LIST_HEAD(&e->probes); |
| 161 | hlist_add_head(&e->hlist, head); |
| 162 | return e; |
| 163 | } |
| 164 | |
| 165 | /* |
| 166 | * Remove the tracepoint from the tracepoint hash table. Must be called |
| 167 | * with lttng_tracepoint_mutex held. |
| 168 | */ |
| 169 | static |
| 170 | void remove_tracepoint(struct tracepoint_entry *e) |
| 171 | { |
| 172 | hlist_del(&e->hlist); |
| 173 | kfree(e); |
| 174 | } |
| 175 | |
| 176 | int lttng_tracepoint_probe_register(const char *name, void *probe, void *data) |
| 177 | { |
| 178 | struct tracepoint_entry *e; |
| 179 | int ret = 0; |
| 180 | |
| 181 | mutex_lock(<tng_tracepoint_mutex); |
| 182 | e = get_tracepoint(name); |
| 183 | if (!e) { |
| 184 | e = add_tracepoint(name); |
| 185 | if (IS_ERR(e)) { |
| 186 | ret = PTR_ERR(e); |
| 187 | goto end; |
| 188 | } |
| 189 | } |
| 190 | /* add (probe, data) to entry */ |
| 191 | ret = add_probe(e, probe, data); |
| 192 | if (ret) |
| 193 | goto end; |
| 194 | e->refcount++; |
| 195 | if (e->tp) { |
| 196 | ret = tracepoint_probe_register(e->tp, probe, data); |
| 197 | WARN_ON_ONCE(ret); |
| 198 | ret = 0; |
| 199 | } |
| 200 | end: |
| 201 | mutex_unlock(<tng_tracepoint_mutex); |
| 202 | return ret; |
| 203 | } |
| 204 | |
| 205 | int lttng_tracepoint_probe_unregister(const char *name, void *probe, void *data) |
| 206 | { |
| 207 | struct tracepoint_entry *e; |
| 208 | int ret = 0; |
| 209 | |
| 210 | mutex_lock(<tng_tracepoint_mutex); |
| 211 | e = get_tracepoint(name); |
| 212 | if (!e) { |
| 213 | ret = -ENOENT; |
| 214 | goto end; |
| 215 | } |
| 216 | /* remove (probe, data) from entry */ |
| 217 | ret = remove_probe(e, probe, data); |
| 218 | if (ret) |
| 219 | goto end; |
| 220 | if (e->tp) { |
| 221 | ret = tracepoint_probe_unregister(e->tp, probe, data); |
| 222 | WARN_ON_ONCE(ret); |
| 223 | ret = 0; |
| 224 | } |
| 225 | if (!--e->refcount) |
| 226 | remove_tracepoint(e); |
| 227 | end: |
| 228 | mutex_unlock(<tng_tracepoint_mutex); |
| 229 | return ret; |
| 230 | } |
| 231 | |
| 232 | #ifdef CONFIG_MODULES |
| 233 | |
| 234 | static |
| 235 | int lttng_tracepoint_coming(struct tp_module *tp_mod) |
| 236 | { |
| 237 | int i; |
| 238 | |
| 239 | mutex_lock(<tng_tracepoint_mutex); |
| 240 | for (i = 0; i < tp_mod->mod->num_tracepoints; i++) { |
| 241 | struct tracepoint *tp; |
| 242 | struct tracepoint_entry *e; |
| 243 | struct lttng_tp_probe *p; |
| 244 | |
| 245 | tp = lttng_tracepoint_ptr_deref(&tp_mod->mod->tracepoints_ptrs[i]); |
| 246 | e = get_tracepoint(tp->name); |
| 247 | if (!e) { |
| 248 | e = add_tracepoint(tp->name); |
| 249 | if (IS_ERR(e)) { |
| 250 | pr_warn("LTTng: error (%ld) adding tracepoint\n", |
| 251 | PTR_ERR(e)); |
| 252 | continue; |
| 253 | } |
| 254 | } |
| 255 | /* If already enabled, just check consistency */ |
| 256 | if (e->tp) { |
| 257 | WARN_ON(e->tp != tp); |
| 258 | continue; |
| 259 | } |
| 260 | e->tp = tp; |
| 261 | e->refcount++; |
| 262 | /* register each (probe, data) */ |
| 263 | list_for_each_entry(p, &e->probes, list) { |
| 264 | int ret; |
| 265 | |
| 266 | ret = tracepoint_probe_register(e->tp, |
| 267 | p->tp_func.func, p->tp_func.data); |
| 268 | WARN_ON_ONCE(ret); |
| 269 | } |
| 270 | } |
| 271 | mutex_unlock(<tng_tracepoint_mutex); |
| 272 | return NOTIFY_OK; |
| 273 | } |
| 274 | |
| 275 | static |
| 276 | int lttng_tracepoint_going(struct tp_module *tp_mod) |
| 277 | { |
| 278 | int i; |
| 279 | |
| 280 | mutex_lock(<tng_tracepoint_mutex); |
| 281 | for (i = 0; i < tp_mod->mod->num_tracepoints; i++) { |
| 282 | struct tracepoint *tp; |
| 283 | struct tracepoint_entry *e; |
| 284 | struct lttng_tp_probe *p; |
| 285 | |
| 286 | tp = lttng_tracepoint_ptr_deref(&tp_mod->mod->tracepoints_ptrs[i]); |
| 287 | e = get_tracepoint(tp->name); |
| 288 | if (!e || !e->tp) |
| 289 | continue; |
| 290 | /* unregister each (probe, data) */ |
| 291 | list_for_each_entry(p, &e->probes, list) { |
| 292 | int ret; |
| 293 | |
| 294 | ret = tracepoint_probe_unregister(e->tp, |
| 295 | p->tp_func.func, p->tp_func.data); |
| 296 | WARN_ON_ONCE(ret); |
| 297 | } |
| 298 | e->tp = NULL; |
| 299 | if (!--e->refcount) |
| 300 | remove_tracepoint(e); |
| 301 | } |
| 302 | mutex_unlock(<tng_tracepoint_mutex); |
| 303 | return 0; |
| 304 | } |
| 305 | |
| 306 | static |
| 307 | int lttng_tracepoint_notify(struct notifier_block *self, |
| 308 | unsigned long val, void *data) |
| 309 | { |
| 310 | struct tp_module *tp_mod = data; |
| 311 | int ret = 0; |
| 312 | |
| 313 | switch (val) { |
| 314 | case MODULE_STATE_COMING: |
| 315 | ret = lttng_tracepoint_coming(tp_mod); |
| 316 | break; |
| 317 | case MODULE_STATE_GOING: |
| 318 | ret = lttng_tracepoint_going(tp_mod); |
| 319 | break; |
| 320 | default: |
| 321 | break; |
| 322 | } |
| 323 | return ret; |
| 324 | } |
| 325 | |
| 326 | static |
| 327 | struct notifier_block lttng_tracepoint_notifier = { |
| 328 | .notifier_call = lttng_tracepoint_notify, |
| 329 | .priority = 0, |
| 330 | }; |
| 331 | |
| 332 | static |
| 333 | int lttng_tracepoint_module_init(void) |
| 334 | { |
| 335 | return register_tracepoint_module_notifier(<tng_tracepoint_notifier); |
| 336 | } |
| 337 | |
| 338 | static |
| 339 | void lttng_tracepoint_module_exit(void) |
| 340 | { |
| 341 | WARN_ON(unregister_tracepoint_module_notifier(<tng_tracepoint_notifier)); |
| 342 | } |
| 343 | |
| 344 | #else /* #ifdef CONFIG_MODULES */ |
| 345 | |
| 346 | static |
| 347 | int lttng_tracepoint_module_init(void) |
| 348 | { |
| 349 | return 0; |
| 350 | } |
| 351 | |
| 352 | static |
| 353 | void lttng_tracepoint_module_exit(void) |
| 354 | { |
| 355 | } |
| 356 | |
| 357 | #endif /* #else #ifdef CONFIG_MODULES */ |
| 358 | |
| 359 | static |
| 360 | void lttng_kernel_tracepoint_add(struct tracepoint *tp, void *priv) |
| 361 | { |
| 362 | struct tracepoint_entry *e; |
| 363 | struct lttng_tp_probe *p; |
| 364 | int *ret = priv; |
| 365 | |
| 366 | mutex_lock(<tng_tracepoint_mutex); |
| 367 | e = get_tracepoint(tp->name); |
| 368 | if (!e) { |
| 369 | e = add_tracepoint(tp->name); |
| 370 | if (IS_ERR(e)) { |
| 371 | pr_warn("LTTng: error (%ld) adding tracepoint\n", |
| 372 | PTR_ERR(e)); |
| 373 | *ret = (int) PTR_ERR(e); |
| 374 | goto end; |
| 375 | } |
| 376 | } |
| 377 | /* If already enabled, just check consistency */ |
| 378 | if (e->tp) { |
| 379 | WARN_ON(e->tp != tp); |
| 380 | goto end; |
| 381 | } |
| 382 | e->tp = tp; |
| 383 | e->refcount++; |
| 384 | /* register each (probe, data) */ |
| 385 | list_for_each_entry(p, &e->probes, list) { |
| 386 | int ret; |
| 387 | |
| 388 | ret = tracepoint_probe_register(e->tp, |
| 389 | p->tp_func.func, p->tp_func.data); |
| 390 | WARN_ON_ONCE(ret); |
| 391 | } |
| 392 | end: |
| 393 | mutex_unlock(<tng_tracepoint_mutex); |
| 394 | } |
| 395 | |
| 396 | static |
| 397 | void lttng_kernel_tracepoint_remove(struct tracepoint *tp, void *priv) |
| 398 | { |
| 399 | struct tracepoint_entry *e; |
| 400 | int *ret = priv; |
| 401 | |
| 402 | mutex_lock(<tng_tracepoint_mutex); |
| 403 | e = get_tracepoint(tp->name); |
| 404 | if (!e || e->refcount != 1 || !list_empty(&e->probes)) { |
| 405 | *ret = -EINVAL; |
| 406 | goto end; |
| 407 | } |
| 408 | remove_tracepoint(e); |
| 409 | end: |
| 410 | mutex_unlock(<tng_tracepoint_mutex); |
| 411 | } |
| 412 | |
| 413 | int __init lttng_tracepoint_init(void) |
| 414 | { |
| 415 | int ret = 0; |
| 416 | |
| 417 | for_each_kernel_tracepoint(lttng_kernel_tracepoint_add, &ret); |
| 418 | if (ret) |
| 419 | goto error; |
| 420 | ret = lttng_tracepoint_module_init(); |
| 421 | if (ret) |
| 422 | goto error_module; |
| 423 | return 0; |
| 424 | |
| 425 | error_module: |
| 426 | { |
| 427 | int error_ret = 0; |
| 428 | |
| 429 | for_each_kernel_tracepoint(lttng_kernel_tracepoint_remove, |
| 430 | &error_ret); |
| 431 | WARN_ON(error_ret); |
| 432 | } |
| 433 | error: |
| 434 | return ret; |
| 435 | } |
| 436 | |
| 437 | void lttng_tracepoint_exit(void) |
| 438 | { |
| 439 | int i, ret = 0; |
| 440 | |
| 441 | lttng_tracepoint_module_exit(); |
| 442 | for_each_kernel_tracepoint(lttng_kernel_tracepoint_remove, &ret); |
| 443 | WARN_ON(ret); |
| 444 | mutex_lock(<tng_tracepoint_mutex); |
| 445 | for (i = 0; i < TRACEPOINT_TABLE_SIZE; i++) { |
| 446 | struct hlist_head *head = &tracepoint_table[i]; |
| 447 | |
| 448 | /* All tracepoints should be removed */ |
| 449 | WARN_ON(!hlist_empty(head)); |
| 450 | } |
| 451 | mutex_unlock(<tng_tracepoint_mutex); |
| 452 | } |