Andrew Geissler | c9f7865 | 2020-09-18 14:11:35 -0500 | [diff] [blame^] | 1 | From ad594e3a953db1b0c3c059fde45b5a5494f6be78 Mon Sep 17 00:00:00 2001 |
| 2 | From: Mathieu Desnoyers <mathieu.desnoyers@efficios.com> |
| 3 | Date: Tue, 28 Jan 2020 16:02:44 -0500 |
| 4 | Subject: [PATCH 10/10] Fix: system call filter table |
| 5 | |
| 6 | The system call filter table has effectively been unused for a long |
| 7 | time due to system call name prefix mismatch. This means the overhead of |
| 8 | selective system call tracing was larger than it should have been because |
| 9 | the event payload preparation would be done for all system calls as soon |
| 10 | as a single system call is traced. |
| 11 | |
| 12 | However, fixing this underlying issue unearths several issues that crept |
| 13 | unnoticed when the "enabler" concept was introduced (after the original |
| 14 | implementation of the system call filter table). |
| 15 | |
| 16 | Here is a list of the issues which are resolved here: |
| 17 | |
| 18 | - Split lttng_syscalls_unregister into an unregister and destroy |
| 19 | function, thus awaiting for a grace period (and therefore quiescence |
| 20 | of the users) after unregistering the system call tracepoints before |
| 21 | freeing the system call filter data structures. This effectively fixes |
| 22 | a use-after-free. |
| 23 | |
| 24 | - The state for enabling "all" system calls vs enabling specific system |
| 25 | calls (and sequences of enable-disable) was incorrect with respect to |
| 26 | the "enablers" semantic. This is solved by always tracking the |
| 27 | bitmap of enabled system calls, and keeping this bitmap even when |
| 28 | enabling all system calls. The sc_filter is now always allocated |
| 29 | before system call tracing is registered to tracepoints, which means |
| 30 | it does not need to be RCU dereferenced anymore. |
| 31 | |
| 32 | Padding fields in the ABI are reserved to select whether to: |
| 33 | |
| 34 | - Trace either native or compat system call (or both, which is the |
| 35 | behavior currently implemented), |
| 36 | - Trace either system call entry or exit (or both, which is the |
| 37 | behavior currently implemented), |
| 38 | - Select the system call to trace by name (behavior currently |
| 39 | implemented) or by system call number, |
| 40 | |
| 41 | Upstream-Status: Backport |
| 42 | |
| 43 | Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com> |
| 44 | --- |
| 45 | lttng-abi.c | 43 ++++++ |
| 46 | lttng-abi.h | 26 ++++ |
| 47 | lttng-events.c | 112 +++++++++++++-- |
| 48 | lttng-events.h | 31 ++++- |
| 49 | lttng-syscalls.c | 348 +++++++++++++++++++++++++---------------------- |
| 50 | 5 files changed, 380 insertions(+), 180 deletions(-) |
| 51 | |
| 52 | diff --git a/lttng-abi.c b/lttng-abi.c |
| 53 | index 64ea99d..b33879d 100644 |
| 54 | --- a/lttng-abi.c |
| 55 | +++ b/lttng-abi.c |
| 56 | @@ -1264,6 +1264,46 @@ nomem: |
| 57 | return ret; |
| 58 | } |
| 59 | |
| 60 | +static |
| 61 | +int lttng_abi_validate_event_param(struct lttng_kernel_event *event_param) |
| 62 | +{ |
| 63 | + /* Limit ABI to implemented features. */ |
| 64 | + switch (event_param->instrumentation) { |
| 65 | + case LTTNG_KERNEL_SYSCALL: |
| 66 | + switch (event_param->u.syscall.entryexit) { |
| 67 | + case LTTNG_KERNEL_SYSCALL_ENTRYEXIT: |
| 68 | + break; |
| 69 | + default: |
| 70 | + return -EINVAL; |
| 71 | + } |
| 72 | + switch (event_param->u.syscall.abi) { |
| 73 | + case LTTNG_KERNEL_SYSCALL_ABI_ALL: |
| 74 | + break; |
| 75 | + default: |
| 76 | + return -EINVAL; |
| 77 | + } |
| 78 | + switch (event_param->u.syscall.match) { |
| 79 | + case LTTNG_SYSCALL_MATCH_NAME: |
| 80 | + break; |
| 81 | + default: |
| 82 | + return -EINVAL; |
| 83 | + } |
| 84 | + break; |
| 85 | + |
| 86 | + case LTTNG_KERNEL_TRACEPOINT: /* Fallthrough */ |
| 87 | + case LTTNG_KERNEL_KPROBE: /* Fallthrough */ |
| 88 | + case LTTNG_KERNEL_KRETPROBE: /* Fallthrough */ |
| 89 | + case LTTNG_KERNEL_NOOP: /* Fallthrough */ |
| 90 | + case LTTNG_KERNEL_UPROBE: |
| 91 | + break; |
| 92 | + |
| 93 | + case LTTNG_KERNEL_FUNCTION: /* Fallthrough */ |
| 94 | + default: |
| 95 | + return -EINVAL; |
| 96 | + } |
| 97 | + return 0; |
| 98 | +} |
| 99 | + |
| 100 | static |
| 101 | int lttng_abi_create_event(struct file *channel_file, |
| 102 | struct lttng_kernel_event *event_param) |
| 103 | @@ -1305,6 +1345,9 @@ int lttng_abi_create_event(struct file *channel_file, |
| 104 | ret = -EOVERFLOW; |
| 105 | goto refcount_error; |
| 106 | } |
| 107 | + ret = lttng_abi_validate_event_param(event_param); |
| 108 | + if (ret) |
| 109 | + goto event_error; |
| 110 | if (event_param->instrumentation == LTTNG_KERNEL_TRACEPOINT |
| 111 | || event_param->instrumentation == LTTNG_KERNEL_SYSCALL) { |
| 112 | struct lttng_enabler *enabler; |
| 113 | diff --git a/lttng-abi.h b/lttng-abi.h |
| 114 | index 1d356ab..51d60e5 100644 |
| 115 | --- a/lttng-abi.h |
| 116 | +++ b/lttng-abi.h |
| 117 | @@ -90,6 +90,31 @@ struct lttng_kernel_event_callsite { |
| 118 | } u; |
| 119 | } __attribute__((packed)); |
| 120 | |
| 121 | +enum lttng_kernel_syscall_entryexit { |
| 122 | + LTTNG_KERNEL_SYSCALL_ENTRYEXIT = 0, |
| 123 | + LTTNG_KERNEL_SYSCALL_ENTRY = 1, /* Not implemented. */ |
| 124 | + LTTNG_KERNEL_SYSCALL_EXIT = 2, /* Not implemented. */ |
| 125 | +}; |
| 126 | + |
| 127 | +enum lttng_kernel_syscall_abi { |
| 128 | + LTTNG_KERNEL_SYSCALL_ABI_ALL = 0, |
| 129 | + LTTNG_KERNEL_SYSCALL_ABI_NATIVE = 1, /* Not implemented. */ |
| 130 | + LTTNG_KERNEL_SYSCALL_ABI_COMPAT = 2, /* Not implemented. */ |
| 131 | +}; |
| 132 | + |
| 133 | +enum lttng_kernel_syscall_match { |
| 134 | + LTTNG_SYSCALL_MATCH_NAME = 0, |
| 135 | + LTTNG_SYSCALL_MATCH_NR = 1, /* Not implemented. */ |
| 136 | +}; |
| 137 | + |
| 138 | +struct lttng_kernel_syscall { |
| 139 | + uint8_t entryexit; /* enum lttng_kernel_syscall_entryexit */ |
| 140 | + uint8_t abi; /* enum lttng_kernel_syscall_abi */ |
| 141 | + uint8_t match; /* enum lttng_kernel_syscall_match */ |
| 142 | + uint8_t padding; |
| 143 | + uint32_t nr; /* For LTTNG_SYSCALL_MATCH_NR */ |
| 144 | +} __attribute__((packed)); |
| 145 | + |
| 146 | /* |
| 147 | * For syscall tracing, name = "*" means "enable all". |
| 148 | */ |
| 149 | @@ -106,6 +131,7 @@ struct lttng_kernel_event { |
| 150 | struct lttng_kernel_kprobe kprobe; |
| 151 | struct lttng_kernel_function_tracer ftrace; |
| 152 | struct lttng_kernel_uprobe uprobe; |
| 153 | + struct lttng_kernel_syscall syscall; |
| 154 | char padding[LTTNG_KERNEL_EVENT_PADDING2]; |
| 155 | } u; |
| 156 | } __attribute__((packed)); |
| 157 | diff --git a/lttng-events.c b/lttng-events.c |
| 158 | index d719294..4c0b04a 100644 |
| 159 | --- a/lttng-events.c |
| 160 | +++ b/lttng-events.c |
| 161 | @@ -201,6 +201,10 @@ void lttng_session_destroy(struct lttng_session *session) |
| 162 | WARN_ON(ret); |
| 163 | } |
| 164 | synchronize_trace(); /* Wait for in-flight events to complete */ |
| 165 | + list_for_each_entry(chan, &session->chan, list) { |
| 166 | + ret = lttng_syscalls_destroy(chan); |
| 167 | + WARN_ON(ret); |
| 168 | + } |
| 169 | list_for_each_entry_safe(enabler, tmpenabler, |
| 170 | &session->enablers_head, node) |
| 171 | lttng_enabler_destroy(enabler); |
| 172 | @@ -740,6 +744,28 @@ struct lttng_event *_lttng_event_create(struct lttng_channel *chan, |
| 173 | event->enabled = 0; |
| 174 | event->registered = 0; |
| 175 | event->desc = event_desc; |
| 176 | + switch (event_param->u.syscall.entryexit) { |
| 177 | + case LTTNG_KERNEL_SYSCALL_ENTRYEXIT: |
| 178 | + ret = -EINVAL; |
| 179 | + goto register_error; |
| 180 | + case LTTNG_KERNEL_SYSCALL_ENTRY: |
| 181 | + event->u.syscall.entryexit = LTTNG_SYSCALL_ENTRY; |
| 182 | + break; |
| 183 | + case LTTNG_KERNEL_SYSCALL_EXIT: |
| 184 | + event->u.syscall.entryexit = LTTNG_SYSCALL_EXIT; |
| 185 | + break; |
| 186 | + } |
| 187 | + switch (event_param->u.syscall.abi) { |
| 188 | + case LTTNG_KERNEL_SYSCALL_ABI_ALL: |
| 189 | + ret = -EINVAL; |
| 190 | + goto register_error; |
| 191 | + case LTTNG_KERNEL_SYSCALL_ABI_NATIVE: |
| 192 | + event->u.syscall.abi = LTTNG_SYSCALL_ABI_NATIVE; |
| 193 | + break; |
| 194 | + case LTTNG_KERNEL_SYSCALL_ABI_COMPAT: |
| 195 | + event->u.syscall.abi = LTTNG_SYSCALL_ABI_COMPAT; |
| 196 | + break; |
| 197 | + } |
| 198 | if (!event->desc) { |
| 199 | ret = -EINVAL; |
| 200 | goto register_error; |
| 201 | @@ -826,8 +852,7 @@ void register_event(struct lttng_event *event) |
| 202 | event); |
| 203 | break; |
| 204 | case LTTNG_KERNEL_SYSCALL: |
| 205 | - ret = lttng_syscall_filter_enable(event->chan, |
| 206 | - desc->name); |
| 207 | + ret = lttng_syscall_filter_enable(event->chan, event); |
| 208 | break; |
| 209 | case LTTNG_KERNEL_KPROBE: |
| 210 | case LTTNG_KERNEL_UPROBE: |
| 211 | @@ -870,8 +895,7 @@ int _lttng_event_unregister(struct lttng_event *event) |
| 212 | ret = 0; |
| 213 | break; |
| 214 | case LTTNG_KERNEL_SYSCALL: |
| 215 | - ret = lttng_syscall_filter_disable(event->chan, |
| 216 | - desc->name); |
| 217 | + ret = lttng_syscall_filter_disable(event->chan, event); |
| 218 | break; |
| 219 | case LTTNG_KERNEL_NOOP: |
| 220 | ret = 0; |
| 221 | @@ -1203,39 +1227,87 @@ int lttng_desc_match_enabler(const struct lttng_event_desc *desc, |
| 222 | struct lttng_enabler *enabler) |
| 223 | { |
| 224 | const char *desc_name, *enabler_name; |
| 225 | + bool compat = false, entry = false; |
| 226 | |
| 227 | enabler_name = enabler->event_param.name; |
| 228 | switch (enabler->event_param.instrumentation) { |
| 229 | case LTTNG_KERNEL_TRACEPOINT: |
| 230 | desc_name = desc->name; |
| 231 | + switch (enabler->type) { |
| 232 | + case LTTNG_ENABLER_STAR_GLOB: |
| 233 | + return lttng_match_enabler_star_glob(desc_name, enabler_name); |
| 234 | + case LTTNG_ENABLER_NAME: |
| 235 | + return lttng_match_enabler_name(desc_name, enabler_name); |
| 236 | + default: |
| 237 | + return -EINVAL; |
| 238 | + } |
| 239 | break; |
| 240 | case LTTNG_KERNEL_SYSCALL: |
| 241 | desc_name = desc->name; |
| 242 | - if (!strncmp(desc_name, "compat_", strlen("compat_"))) |
| 243 | + if (!strncmp(desc_name, "compat_", strlen("compat_"))) { |
| 244 | desc_name += strlen("compat_"); |
| 245 | + compat = true; |
| 246 | + } |
| 247 | if (!strncmp(desc_name, "syscall_exit_", |
| 248 | strlen("syscall_exit_"))) { |
| 249 | desc_name += strlen("syscall_exit_"); |
| 250 | } else if (!strncmp(desc_name, "syscall_entry_", |
| 251 | strlen("syscall_entry_"))) { |
| 252 | desc_name += strlen("syscall_entry_"); |
| 253 | + entry = true; |
| 254 | } else { |
| 255 | WARN_ON_ONCE(1); |
| 256 | return -EINVAL; |
| 257 | } |
| 258 | + switch (enabler->event_param.u.syscall.entryexit) { |
| 259 | + case LTTNG_KERNEL_SYSCALL_ENTRYEXIT: |
| 260 | + break; |
| 261 | + case LTTNG_KERNEL_SYSCALL_ENTRY: |
| 262 | + if (!entry) |
| 263 | + return 0; |
| 264 | + break; |
| 265 | + case LTTNG_KERNEL_SYSCALL_EXIT: |
| 266 | + if (entry) |
| 267 | + return 0; |
| 268 | + break; |
| 269 | + default: |
| 270 | + return -EINVAL; |
| 271 | + } |
| 272 | + switch (enabler->event_param.u.syscall.abi) { |
| 273 | + case LTTNG_KERNEL_SYSCALL_ABI_ALL: |
| 274 | + break; |
| 275 | + case LTTNG_KERNEL_SYSCALL_ABI_NATIVE: |
| 276 | + if (compat) |
| 277 | + return 0; |
| 278 | + break; |
| 279 | + case LTTNG_KERNEL_SYSCALL_ABI_COMPAT: |
| 280 | + if (!compat) |
| 281 | + return 0; |
| 282 | + break; |
| 283 | + default: |
| 284 | + return -EINVAL; |
| 285 | + } |
| 286 | + switch (enabler->event_param.u.syscall.match) { |
| 287 | + case LTTNG_SYSCALL_MATCH_NAME: |
| 288 | + switch (enabler->type) { |
| 289 | + case LTTNG_ENABLER_STAR_GLOB: |
| 290 | + return lttng_match_enabler_star_glob(desc_name, enabler_name); |
| 291 | + case LTTNG_ENABLER_NAME: |
| 292 | + return lttng_match_enabler_name(desc_name, enabler_name); |
| 293 | + default: |
| 294 | + return -EINVAL; |
| 295 | + } |
| 296 | + break; |
| 297 | + case LTTNG_SYSCALL_MATCH_NR: |
| 298 | + return -EINVAL; /* Not implemented. */ |
| 299 | + default: |
| 300 | + return -EINVAL; |
| 301 | + } |
| 302 | break; |
| 303 | default: |
| 304 | WARN_ON_ONCE(1); |
| 305 | return -EINVAL; |
| 306 | } |
| 307 | - switch (enabler->type) { |
| 308 | - case LTTNG_ENABLER_STAR_GLOB: |
| 309 | - return lttng_match_enabler_star_glob(desc_name, enabler_name); |
| 310 | - case LTTNG_ENABLER_NAME: |
| 311 | - return lttng_match_enabler_name(desc_name, enabler_name); |
| 312 | - default: |
| 313 | - return -EINVAL; |
| 314 | - } |
| 315 | } |
| 316 | |
| 317 | static |
| 318 | @@ -1361,9 +1433,21 @@ void lttng_create_event_if_missing(struct lttng_enabler *enabler) |
| 319 | static |
| 320 | int lttng_enabler_ref_events(struct lttng_enabler *enabler) |
| 321 | { |
| 322 | - struct lttng_session *session = enabler->chan->session; |
| 323 | + struct lttng_channel *chan = enabler->chan; |
| 324 | + struct lttng_session *session = chan->session; |
| 325 | struct lttng_event *event; |
| 326 | |
| 327 | + if (enabler->event_param.instrumentation == LTTNG_KERNEL_SYSCALL && |
| 328 | + enabler->event_param.u.syscall.entryexit == LTTNG_KERNEL_SYSCALL_ENTRYEXIT && |
| 329 | + enabler->event_param.u.syscall.abi == LTTNG_KERNEL_SYSCALL_ABI_ALL && |
| 330 | + enabler->event_param.u.syscall.match == LTTNG_SYSCALL_MATCH_NAME && |
| 331 | + !strcmp(enabler->event_param.name, "*")) { |
| 332 | + if (enabler->enabled) |
| 333 | + WRITE_ONCE(chan->syscall_all, 1); |
| 334 | + else |
| 335 | + WRITE_ONCE(chan->syscall_all, 0); |
| 336 | + } |
| 337 | + |
| 338 | /* First ensure that probe events are created for this enabler. */ |
| 339 | lttng_create_event_if_missing(enabler); |
| 340 | |
| 341 | diff --git a/lttng-events.h b/lttng-events.h |
| 342 | index a36a312..d4d9976 100644 |
| 343 | --- a/lttng-events.h |
| 344 | +++ b/lttng-events.h |
| 345 | @@ -292,6 +292,16 @@ struct lttng_uprobe_handler { |
| 346 | struct list_head node; |
| 347 | }; |
| 348 | |
| 349 | +enum lttng_syscall_entryexit { |
| 350 | + LTTNG_SYSCALL_ENTRY, |
| 351 | + LTTNG_SYSCALL_EXIT, |
| 352 | +}; |
| 353 | + |
| 354 | +enum lttng_syscall_abi { |
| 355 | + LTTNG_SYSCALL_ABI_NATIVE, |
| 356 | + LTTNG_SYSCALL_ABI_COMPAT, |
| 357 | +}; |
| 358 | + |
| 359 | /* |
| 360 | * lttng_event structure is referred to by the tracing fast path. It must be |
| 361 | * kept small. |
| 362 | @@ -318,6 +328,11 @@ struct lttng_event { |
| 363 | struct inode *inode; |
| 364 | struct list_head head; |
| 365 | } uprobe; |
| 366 | + struct { |
| 367 | + char *syscall_name; |
| 368 | + enum lttng_syscall_entryexit entryexit; |
| 369 | + enum lttng_syscall_abi abi; |
| 370 | + } syscall; |
| 371 | } u; |
| 372 | struct list_head list; /* Event list in session */ |
| 373 | unsigned int metadata_dumped:1; |
| 374 | @@ -457,10 +472,10 @@ struct lttng_channel { |
| 375 | struct lttng_syscall_filter *sc_filter; |
| 376 | int header_type; /* 0: unset, 1: compact, 2: large */ |
| 377 | enum channel_type channel_type; |
| 378 | + int syscall_all; |
| 379 | unsigned int metadata_dumped:1, |
| 380 | sys_enter_registered:1, |
| 381 | sys_exit_registered:1, |
| 382 | - syscall_all:1, |
| 383 | tstate:1; /* Transient enable state */ |
| 384 | }; |
| 385 | |
| 386 | @@ -653,10 +668,11 @@ void lttng_clock_unref(void); |
| 387 | #if defined(CONFIG_HAVE_SYSCALL_TRACEPOINTS) |
| 388 | int lttng_syscalls_register(struct lttng_channel *chan, void *filter); |
| 389 | int lttng_syscalls_unregister(struct lttng_channel *chan); |
| 390 | +int lttng_syscalls_destroy(struct lttng_channel *chan); |
| 391 | int lttng_syscall_filter_enable(struct lttng_channel *chan, |
| 392 | - const char *name); |
| 393 | + struct lttng_event *event); |
| 394 | int lttng_syscall_filter_disable(struct lttng_channel *chan, |
| 395 | - const char *name); |
| 396 | + struct lttng_event *event); |
| 397 | long lttng_channel_syscall_mask(struct lttng_channel *channel, |
| 398 | struct lttng_kernel_syscall_mask __user *usyscall_mask); |
| 399 | #else |
| 400 | @@ -670,14 +686,19 @@ static inline int lttng_syscalls_unregister(struct lttng_channel *chan) |
| 401 | return 0; |
| 402 | } |
| 403 | |
| 404 | +static inline int lttng_syscalls_destroy(struct lttng_channel *chan) |
| 405 | +{ |
| 406 | + return 0; |
| 407 | +} |
| 408 | + |
| 409 | static inline int lttng_syscall_filter_enable(struct lttng_channel *chan, |
| 410 | - const char *name) |
| 411 | + struct lttng_event *event); |
| 412 | { |
| 413 | return -ENOSYS; |
| 414 | } |
| 415 | |
| 416 | static inline int lttng_syscall_filter_disable(struct lttng_channel *chan, |
| 417 | - const char *name) |
| 418 | + struct lttng_event *event); |
| 419 | { |
| 420 | return -ENOSYS; |
| 421 | } |
| 422 | diff --git a/lttng-syscalls.c b/lttng-syscalls.c |
| 423 | index 97f1ba9..26cead6 100644 |
| 424 | --- a/lttng-syscalls.c |
| 425 | +++ b/lttng-syscalls.c |
| 426 | @@ -367,8 +367,10 @@ const struct trace_syscall_entry compat_sc_exit_table[] = { |
| 427 | #undef CREATE_SYSCALL_TABLE |
| 428 | |
| 429 | struct lttng_syscall_filter { |
| 430 | - DECLARE_BITMAP(sc, NR_syscalls); |
| 431 | - DECLARE_BITMAP(sc_compat, NR_compat_syscalls); |
| 432 | + DECLARE_BITMAP(sc_entry, NR_syscalls); |
| 433 | + DECLARE_BITMAP(sc_exit, NR_syscalls); |
| 434 | + DECLARE_BITMAP(sc_compat_entry, NR_compat_syscalls); |
| 435 | + DECLARE_BITMAP(sc_compat_exit, NR_compat_syscalls); |
| 436 | }; |
| 437 | |
| 438 | static void syscall_entry_unknown(struct lttng_event *event, |
| 439 | @@ -391,29 +393,23 @@ void syscall_entry_probe(void *__data, struct pt_regs *regs, long id) |
| 440 | size_t table_len; |
| 441 | |
| 442 | if (unlikely(in_compat_syscall())) { |
| 443 | - struct lttng_syscall_filter *filter; |
| 444 | - |
| 445 | - filter = lttng_rcu_dereference(chan->sc_filter); |
| 446 | - if (filter) { |
| 447 | - if (id < 0 || id >= NR_compat_syscalls |
| 448 | - || !test_bit(id, filter->sc_compat)) { |
| 449 | - /* System call filtered out. */ |
| 450 | - return; |
| 451 | - } |
| 452 | + struct lttng_syscall_filter *filter = chan->sc_filter; |
| 453 | + |
| 454 | + if (id < 0 || id >= NR_compat_syscalls |
| 455 | + || (!READ_ONCE(chan->syscall_all) && !test_bit(id, filter->sc_compat_entry))) { |
| 456 | + /* System call filtered out. */ |
| 457 | + return; |
| 458 | } |
| 459 | table = compat_sc_table; |
| 460 | table_len = ARRAY_SIZE(compat_sc_table); |
| 461 | unknown_event = chan->sc_compat_unknown; |
| 462 | } else { |
| 463 | - struct lttng_syscall_filter *filter; |
| 464 | - |
| 465 | - filter = lttng_rcu_dereference(chan->sc_filter); |
| 466 | - if (filter) { |
| 467 | - if (id < 0 || id >= NR_syscalls |
| 468 | - || !test_bit(id, filter->sc)) { |
| 469 | - /* System call filtered out. */ |
| 470 | - return; |
| 471 | - } |
| 472 | + struct lttng_syscall_filter *filter = chan->sc_filter; |
| 473 | + |
| 474 | + if (id < 0 || id >= NR_syscalls |
| 475 | + || (!READ_ONCE(chan->syscall_all) && !test_bit(id, filter->sc_entry))) { |
| 476 | + /* System call filtered out. */ |
| 477 | + return; |
| 478 | } |
| 479 | table = sc_table; |
| 480 | table_len = ARRAY_SIZE(sc_table); |
| 481 | @@ -545,29 +541,23 @@ void syscall_exit_probe(void *__data, struct pt_regs *regs, long ret) |
| 482 | |
| 483 | id = syscall_get_nr(current, regs); |
| 484 | if (unlikely(in_compat_syscall())) { |
| 485 | - struct lttng_syscall_filter *filter; |
| 486 | - |
| 487 | - filter = lttng_rcu_dereference(chan->sc_filter); |
| 488 | - if (filter) { |
| 489 | - if (id < 0 || id >= NR_compat_syscalls |
| 490 | - || !test_bit(id, filter->sc_compat)) { |
| 491 | - /* System call filtered out. */ |
| 492 | - return; |
| 493 | - } |
| 494 | + struct lttng_syscall_filter *filter = chan->sc_filter; |
| 495 | + |
| 496 | + if (id < 0 || id >= NR_compat_syscalls |
| 497 | + || (!READ_ONCE(chan->syscall_all) && !test_bit(id, filter->sc_compat_exit))) { |
| 498 | + /* System call filtered out. */ |
| 499 | + return; |
| 500 | } |
| 501 | table = compat_sc_exit_table; |
| 502 | table_len = ARRAY_SIZE(compat_sc_exit_table); |
| 503 | unknown_event = chan->compat_sc_exit_unknown; |
| 504 | } else { |
| 505 | - struct lttng_syscall_filter *filter; |
| 506 | - |
| 507 | - filter = lttng_rcu_dereference(chan->sc_filter); |
| 508 | - if (filter) { |
| 509 | - if (id < 0 || id >= NR_syscalls |
| 510 | - || !test_bit(id, filter->sc)) { |
| 511 | - /* System call filtered out. */ |
| 512 | - return; |
| 513 | - } |
| 514 | + struct lttng_syscall_filter *filter = chan->sc_filter; |
| 515 | + |
| 516 | + if (id < 0 || id >= NR_syscalls |
| 517 | + || (!READ_ONCE(chan->syscall_all) && !test_bit(id, filter->sc_exit))) { |
| 518 | + /* System call filtered out. */ |
| 519 | + return; |
| 520 | } |
| 521 | table = sc_exit_table; |
| 522 | table_len = ARRAY_SIZE(sc_exit_table); |
| 523 | @@ -713,27 +703,23 @@ int fill_table(const struct trace_syscall_entry *table, size_t table_len, |
| 524 | memset(&ev, 0, sizeof(ev)); |
| 525 | switch (type) { |
| 526 | case SC_TYPE_ENTRY: |
| 527 | - strncpy(ev.name, SYSCALL_ENTRY_STR, |
| 528 | - LTTNG_KERNEL_SYM_NAME_LEN); |
| 529 | + ev.u.syscall.entryexit = LTTNG_KERNEL_SYSCALL_ENTRY; |
| 530 | + ev.u.syscall.abi = LTTNG_KERNEL_SYSCALL_ABI_NATIVE; |
| 531 | break; |
| 532 | case SC_TYPE_EXIT: |
| 533 | - strncpy(ev.name, SYSCALL_EXIT_STR, |
| 534 | - LTTNG_KERNEL_SYM_NAME_LEN); |
| 535 | + ev.u.syscall.entryexit = LTTNG_KERNEL_SYSCALL_EXIT; |
| 536 | + ev.u.syscall.abi = LTTNG_KERNEL_SYSCALL_ABI_NATIVE; |
| 537 | break; |
| 538 | case SC_TYPE_COMPAT_ENTRY: |
| 539 | - strncpy(ev.name, COMPAT_SYSCALL_ENTRY_STR, |
| 540 | - LTTNG_KERNEL_SYM_NAME_LEN); |
| 541 | + ev.u.syscall.entryexit = LTTNG_KERNEL_SYSCALL_ENTRY; |
| 542 | + ev.u.syscall.abi = LTTNG_KERNEL_SYSCALL_ABI_COMPAT; |
| 543 | break; |
| 544 | case SC_TYPE_COMPAT_EXIT: |
| 545 | - strncpy(ev.name, COMPAT_SYSCALL_EXIT_STR, |
| 546 | - LTTNG_KERNEL_SYM_NAME_LEN); |
| 547 | - break; |
| 548 | - default: |
| 549 | - BUG_ON(1); |
| 550 | + ev.u.syscall.entryexit = LTTNG_KERNEL_SYSCALL_EXIT; |
| 551 | + ev.u.syscall.abi = LTTNG_KERNEL_SYSCALL_ABI_COMPAT; |
| 552 | break; |
| 553 | } |
| 554 | - strncat(ev.name, desc->name, |
| 555 | - LTTNG_KERNEL_SYM_NAME_LEN - strlen(ev.name) - 1); |
| 556 | + strncpy(ev.name, desc->name, LTTNG_KERNEL_SYM_NAME_LEN); |
| 557 | ev.name[LTTNG_KERNEL_SYM_NAME_LEN - 1] = '\0'; |
| 558 | ev.instrumentation = LTTNG_KERNEL_SYSCALL; |
| 559 | chan_table[i] = _lttng_event_create(chan, &ev, filter, |
| 560 | @@ -803,6 +789,8 @@ int lttng_syscalls_register(struct lttng_channel *chan, void *filter) |
| 561 | strncpy(ev.name, desc->name, LTTNG_KERNEL_SYM_NAME_LEN); |
| 562 | ev.name[LTTNG_KERNEL_SYM_NAME_LEN - 1] = '\0'; |
| 563 | ev.instrumentation = LTTNG_KERNEL_SYSCALL; |
| 564 | + ev.u.syscall.entryexit = LTTNG_KERNEL_SYSCALL_ENTRY; |
| 565 | + ev.u.syscall.abi = LTTNG_KERNEL_SYSCALL_ABI_NATIVE; |
| 566 | chan->sc_unknown = _lttng_event_create(chan, &ev, filter, |
| 567 | desc, |
| 568 | ev.instrumentation); |
| 569 | @@ -820,6 +808,8 @@ int lttng_syscalls_register(struct lttng_channel *chan, void *filter) |
| 570 | strncpy(ev.name, desc->name, LTTNG_KERNEL_SYM_NAME_LEN); |
| 571 | ev.name[LTTNG_KERNEL_SYM_NAME_LEN - 1] = '\0'; |
| 572 | ev.instrumentation = LTTNG_KERNEL_SYSCALL; |
| 573 | + ev.u.syscall.entryexit = LTTNG_KERNEL_SYSCALL_ENTRY; |
| 574 | + ev.u.syscall.abi = LTTNG_KERNEL_SYSCALL_ABI_COMPAT; |
| 575 | chan->sc_compat_unknown = _lttng_event_create(chan, &ev, filter, |
| 576 | desc, |
| 577 | ev.instrumentation); |
| 578 | @@ -837,6 +827,8 @@ int lttng_syscalls_register(struct lttng_channel *chan, void *filter) |
| 579 | strncpy(ev.name, desc->name, LTTNG_KERNEL_SYM_NAME_LEN); |
| 580 | ev.name[LTTNG_KERNEL_SYM_NAME_LEN - 1] = '\0'; |
| 581 | ev.instrumentation = LTTNG_KERNEL_SYSCALL; |
| 582 | + ev.u.syscall.entryexit = LTTNG_KERNEL_SYSCALL_EXIT; |
| 583 | + ev.u.syscall.abi = LTTNG_KERNEL_SYSCALL_ABI_COMPAT; |
| 584 | chan->compat_sc_exit_unknown = _lttng_event_create(chan, &ev, |
| 585 | filter, desc, |
| 586 | ev.instrumentation); |
| 587 | @@ -854,6 +846,8 @@ int lttng_syscalls_register(struct lttng_channel *chan, void *filter) |
| 588 | strncpy(ev.name, desc->name, LTTNG_KERNEL_SYM_NAME_LEN); |
| 589 | ev.name[LTTNG_KERNEL_SYM_NAME_LEN - 1] = '\0'; |
| 590 | ev.instrumentation = LTTNG_KERNEL_SYSCALL; |
| 591 | + ev.u.syscall.entryexit = LTTNG_KERNEL_SYSCALL_EXIT; |
| 592 | + ev.u.syscall.abi = LTTNG_KERNEL_SYSCALL_ABI_NATIVE; |
| 593 | chan->sc_exit_unknown = _lttng_event_create(chan, &ev, filter, |
| 594 | desc, ev.instrumentation); |
| 595 | WARN_ON_ONCE(!chan->sc_exit_unknown); |
| 596 | @@ -883,6 +877,14 @@ int lttng_syscalls_register(struct lttng_channel *chan, void *filter) |
| 597 | if (ret) |
| 598 | return ret; |
| 599 | #endif |
| 600 | + |
| 601 | + if (!chan->sc_filter) { |
| 602 | + chan->sc_filter = kzalloc(sizeof(struct lttng_syscall_filter), |
| 603 | + GFP_KERNEL); |
| 604 | + if (!chan->sc_filter) |
| 605 | + return -ENOMEM; |
| 606 | + } |
| 607 | + |
| 608 | if (!chan->sys_enter_registered) { |
| 609 | ret = lttng_wrapper_tracepoint_probe_register("sys_enter", |
| 610 | (void *) syscall_entry_probe, chan); |
| 611 | @@ -930,7 +932,11 @@ int lttng_syscalls_unregister(struct lttng_channel *chan) |
| 612 | return ret; |
| 613 | chan->sys_exit_registered = 0; |
| 614 | } |
| 615 | - /* lttng_event destroy will be performed by lttng_session_destroy() */ |
| 616 | + return 0; |
| 617 | +} |
| 618 | + |
| 619 | +int lttng_syscalls_destroy(struct lttng_channel *chan) |
| 620 | +{ |
| 621 | kfree(chan->sc_table); |
| 622 | kfree(chan->sc_exit_table); |
| 623 | #ifdef CONFIG_COMPAT |
| 624 | @@ -993,136 +999,150 @@ uint32_t get_sc_tables_len(void) |
| 625 | return ARRAY_SIZE(sc_table) + ARRAY_SIZE(compat_sc_table); |
| 626 | } |
| 627 | |
| 628 | -int lttng_syscall_filter_enable(struct lttng_channel *chan, |
| 629 | - const char *name) |
| 630 | +static |
| 631 | +const char *get_syscall_name(struct lttng_event *event) |
| 632 | { |
| 633 | - int syscall_nr, compat_syscall_nr, ret; |
| 634 | - struct lttng_syscall_filter *filter; |
| 635 | + size_t prefix_len = 0; |
| 636 | |
| 637 | - WARN_ON_ONCE(!chan->sc_table); |
| 638 | + WARN_ON_ONCE(event->instrumentation != LTTNG_KERNEL_SYSCALL); |
| 639 | |
| 640 | - if (!name) { |
| 641 | - /* Enable all system calls by removing filter */ |
| 642 | - if (chan->sc_filter) { |
| 643 | - filter = chan->sc_filter; |
| 644 | - rcu_assign_pointer(chan->sc_filter, NULL); |
| 645 | - synchronize_trace(); |
| 646 | - kfree(filter); |
| 647 | + switch (event->u.syscall.entryexit) { |
| 648 | + case LTTNG_SYSCALL_ENTRY: |
| 649 | + switch (event->u.syscall.abi) { |
| 650 | + case LTTNG_SYSCALL_ABI_NATIVE: |
| 651 | + prefix_len = strlen(SYSCALL_ENTRY_STR); |
| 652 | + break; |
| 653 | + case LTTNG_SYSCALL_ABI_COMPAT: |
| 654 | + prefix_len = strlen(COMPAT_SYSCALL_ENTRY_STR); |
| 655 | + break; |
| 656 | } |
| 657 | - chan->syscall_all = 1; |
| 658 | - return 0; |
| 659 | - } |
| 660 | - |
| 661 | - if (!chan->sc_filter) { |
| 662 | - if (chan->syscall_all) { |
| 663 | - /* |
| 664 | - * All syscalls are already enabled. |
| 665 | - */ |
| 666 | - return -EEXIST; |
| 667 | + break; |
| 668 | + case LTTNG_SYSCALL_EXIT: |
| 669 | + switch (event->u.syscall.abi) { |
| 670 | + case LTTNG_SYSCALL_ABI_NATIVE: |
| 671 | + prefix_len = strlen(SYSCALL_EXIT_STR); |
| 672 | + break; |
| 673 | + case LTTNG_SYSCALL_ABI_COMPAT: |
| 674 | + prefix_len = strlen(COMPAT_SYSCALL_EXIT_STR); |
| 675 | + break; |
| 676 | } |
| 677 | - filter = kzalloc(sizeof(struct lttng_syscall_filter), |
| 678 | - GFP_KERNEL); |
| 679 | - if (!filter) |
| 680 | - return -ENOMEM; |
| 681 | - } else { |
| 682 | - filter = chan->sc_filter; |
| 683 | + break; |
| 684 | } |
| 685 | - syscall_nr = get_syscall_nr(name); |
| 686 | - compat_syscall_nr = get_compat_syscall_nr(name); |
| 687 | - if (syscall_nr < 0 && compat_syscall_nr < 0) { |
| 688 | - ret = -ENOENT; |
| 689 | - goto error; |
| 690 | + WARN_ON_ONCE(prefix_len == 0); |
| 691 | + return event->desc->name + prefix_len; |
| 692 | +} |
| 693 | + |
| 694 | +int lttng_syscall_filter_enable(struct lttng_channel *chan, |
| 695 | + struct lttng_event *event) |
| 696 | +{ |
| 697 | + struct lttng_syscall_filter *filter = chan->sc_filter; |
| 698 | + const char *syscall_name; |
| 699 | + unsigned long *bitmap; |
| 700 | + int syscall_nr; |
| 701 | + |
| 702 | + WARN_ON_ONCE(!chan->sc_table); |
| 703 | + |
| 704 | + syscall_name = get_syscall_name(event); |
| 705 | + |
| 706 | + switch (event->u.syscall.abi) { |
| 707 | + case LTTNG_SYSCALL_ABI_NATIVE: |
| 708 | + syscall_nr = get_syscall_nr(syscall_name); |
| 709 | + break; |
| 710 | + case LTTNG_SYSCALL_ABI_COMPAT: |
| 711 | + syscall_nr = get_compat_syscall_nr(syscall_name); |
| 712 | + break; |
| 713 | + default: |
| 714 | + return -EINVAL; |
| 715 | } |
| 716 | - if (syscall_nr >= 0) { |
| 717 | - if (test_bit(syscall_nr, filter->sc)) { |
| 718 | - ret = -EEXIST; |
| 719 | - goto error; |
| 720 | + if (syscall_nr < 0) |
| 721 | + return -ENOENT; |
| 722 | + |
| 723 | + |
| 724 | + switch (event->u.syscall.entryexit) { |
| 725 | + case LTTNG_SYSCALL_ENTRY: |
| 726 | + switch (event->u.syscall.abi) { |
| 727 | + case LTTNG_SYSCALL_ABI_NATIVE: |
| 728 | + bitmap = filter->sc_entry; |
| 729 | + break; |
| 730 | + case LTTNG_SYSCALL_ABI_COMPAT: |
| 731 | + bitmap = filter->sc_compat_entry; |
| 732 | + break; |
| 733 | } |
| 734 | - bitmap_set(filter->sc, syscall_nr, 1); |
| 735 | - } |
| 736 | - if (compat_syscall_nr >= 0) { |
| 737 | - if (test_bit(compat_syscall_nr, filter->sc_compat)) { |
| 738 | - ret = -EEXIST; |
| 739 | - goto error; |
| 740 | + break; |
| 741 | + case LTTNG_SYSCALL_EXIT: |
| 742 | + switch (event->u.syscall.abi) { |
| 743 | + case LTTNG_SYSCALL_ABI_NATIVE: |
| 744 | + bitmap = filter->sc_exit; |
| 745 | + break; |
| 746 | + case LTTNG_SYSCALL_ABI_COMPAT: |
| 747 | + bitmap = filter->sc_compat_exit; |
| 748 | + break; |
| 749 | } |
| 750 | - bitmap_set(filter->sc_compat, compat_syscall_nr, 1); |
| 751 | + break; |
| 752 | + default: |
| 753 | + return -EINVAL; |
| 754 | } |
| 755 | - if (!chan->sc_filter) |
| 756 | - rcu_assign_pointer(chan->sc_filter, filter); |
| 757 | + if (test_bit(syscall_nr, bitmap)) |
| 758 | + return -EEXIST; |
| 759 | + bitmap_set(bitmap, syscall_nr, 1); |
| 760 | return 0; |
| 761 | - |
| 762 | -error: |
| 763 | - if (!chan->sc_filter) |
| 764 | - kfree(filter); |
| 765 | - return ret; |
| 766 | } |
| 767 | |
| 768 | int lttng_syscall_filter_disable(struct lttng_channel *chan, |
| 769 | - const char *name) |
| 770 | + struct lttng_event *event) |
| 771 | { |
| 772 | - int syscall_nr, compat_syscall_nr, ret; |
| 773 | - struct lttng_syscall_filter *filter; |
| 774 | + struct lttng_syscall_filter *filter = chan->sc_filter; |
| 775 | + const char *syscall_name; |
| 776 | + unsigned long *bitmap; |
| 777 | + int syscall_nr; |
| 778 | |
| 779 | WARN_ON_ONCE(!chan->sc_table); |
| 780 | |
| 781 | - if (!chan->sc_filter) { |
| 782 | - if (!chan->syscall_all) |
| 783 | - return -EEXIST; |
| 784 | - filter = kzalloc(sizeof(struct lttng_syscall_filter), |
| 785 | - GFP_KERNEL); |
| 786 | - if (!filter) |
| 787 | - return -ENOMEM; |
| 788 | - /* Trace all system calls, then apply disable. */ |
| 789 | - bitmap_set(filter->sc, 0, NR_syscalls); |
| 790 | - bitmap_set(filter->sc_compat, 0, NR_compat_syscalls); |
| 791 | - } else { |
| 792 | - filter = chan->sc_filter; |
| 793 | + syscall_name = get_syscall_name(event); |
| 794 | + |
| 795 | + switch (event->u.syscall.abi) { |
| 796 | + case LTTNG_SYSCALL_ABI_NATIVE: |
| 797 | + syscall_nr = get_syscall_nr(syscall_name); |
| 798 | + break; |
| 799 | + case LTTNG_SYSCALL_ABI_COMPAT: |
| 800 | + syscall_nr = get_compat_syscall_nr(syscall_name); |
| 801 | + break; |
| 802 | + default: |
| 803 | + return -EINVAL; |
| 804 | } |
| 805 | + if (syscall_nr < 0) |
| 806 | + return -ENOENT; |
| 807 | |
| 808 | - if (!name) { |
| 809 | - /* Fail if all syscalls are already disabled. */ |
| 810 | - if (bitmap_empty(filter->sc, NR_syscalls) |
| 811 | - && bitmap_empty(filter->sc_compat, |
| 812 | - NR_compat_syscalls)) { |
| 813 | - ret = -EEXIST; |
| 814 | - goto error; |
| 815 | - } |
| 816 | |
| 817 | - /* Disable all system calls */ |
| 818 | - bitmap_clear(filter->sc, 0, NR_syscalls); |
| 819 | - bitmap_clear(filter->sc_compat, 0, NR_compat_syscalls); |
| 820 | - goto apply_filter; |
| 821 | - } |
| 822 | - syscall_nr = get_syscall_nr(name); |
| 823 | - compat_syscall_nr = get_compat_syscall_nr(name); |
| 824 | - if (syscall_nr < 0 && compat_syscall_nr < 0) { |
| 825 | - ret = -ENOENT; |
| 826 | - goto error; |
| 827 | - } |
| 828 | - if (syscall_nr >= 0) { |
| 829 | - if (!test_bit(syscall_nr, filter->sc)) { |
| 830 | - ret = -EEXIST; |
| 831 | - goto error; |
| 832 | + switch (event->u.syscall.entryexit) { |
| 833 | + case LTTNG_SYSCALL_ENTRY: |
| 834 | + switch (event->u.syscall.abi) { |
| 835 | + case LTTNG_SYSCALL_ABI_NATIVE: |
| 836 | + bitmap = filter->sc_entry; |
| 837 | + break; |
| 838 | + case LTTNG_SYSCALL_ABI_COMPAT: |
| 839 | + bitmap = filter->sc_compat_entry; |
| 840 | + break; |
| 841 | } |
| 842 | - bitmap_clear(filter->sc, syscall_nr, 1); |
| 843 | - } |
| 844 | - if (compat_syscall_nr >= 0) { |
| 845 | - if (!test_bit(compat_syscall_nr, filter->sc_compat)) { |
| 846 | - ret = -EEXIST; |
| 847 | - goto error; |
| 848 | + break; |
| 849 | + case LTTNG_SYSCALL_EXIT: |
| 850 | + switch (event->u.syscall.abi) { |
| 851 | + case LTTNG_SYSCALL_ABI_NATIVE: |
| 852 | + bitmap = filter->sc_exit; |
| 853 | + break; |
| 854 | + case LTTNG_SYSCALL_ABI_COMPAT: |
| 855 | + bitmap = filter->sc_compat_exit; |
| 856 | + break; |
| 857 | } |
| 858 | - bitmap_clear(filter->sc_compat, compat_syscall_nr, 1); |
| 859 | + break; |
| 860 | + default: |
| 861 | + return -EINVAL; |
| 862 | } |
| 863 | -apply_filter: |
| 864 | - if (!chan->sc_filter) |
| 865 | - rcu_assign_pointer(chan->sc_filter, filter); |
| 866 | - chan->syscall_all = 0; |
| 867 | - return 0; |
| 868 | + if (!test_bit(syscall_nr, bitmap)) |
| 869 | + return -EEXIST; |
| 870 | + bitmap_clear(bitmap, syscall_nr, 1); |
| 871 | |
| 872 | -error: |
| 873 | - if (!chan->sc_filter) |
| 874 | - kfree(filter); |
| 875 | - return ret; |
| 876 | + return 0; |
| 877 | } |
| 878 | |
| 879 | static |
| 880 | @@ -1236,6 +1256,9 @@ const struct file_operations lttng_syscall_list_fops = { |
| 881 | .release = seq_release, |
| 882 | }; |
| 883 | |
| 884 | +/* |
| 885 | + * A syscall is enabled if it is traced for either entry or exit. |
| 886 | + */ |
| 887 | long lttng_channel_syscall_mask(struct lttng_channel *channel, |
| 888 | struct lttng_kernel_syscall_mask __user *usyscall_mask) |
| 889 | { |
| 890 | @@ -1262,8 +1285,9 @@ long lttng_channel_syscall_mask(struct lttng_channel *channel, |
| 891 | char state; |
| 892 | |
| 893 | if (channel->sc_table) { |
| 894 | - if (filter) |
| 895 | - state = test_bit(bit, filter->sc); |
| 896 | + if (!READ_ONCE(channel->syscall_all) && filter) |
| 897 | + state = test_bit(bit, filter->sc_entry) |
| 898 | + || test_bit(bit, filter->sc_exit); |
| 899 | else |
| 900 | state = 1; |
| 901 | } else { |
| 902 | @@ -1275,9 +1299,11 @@ long lttng_channel_syscall_mask(struct lttng_channel *channel, |
| 903 | char state; |
| 904 | |
| 905 | if (channel->compat_sc_table) { |
| 906 | - if (filter) |
| 907 | + if (!READ_ONCE(channel->syscall_all) && filter) |
| 908 | state = test_bit(bit - ARRAY_SIZE(sc_table), |
| 909 | - filter->sc_compat); |
| 910 | + filter->sc_compat_entry) |
| 911 | + || test_bit(bit - ARRAY_SIZE(sc_table), |
| 912 | + filter->sc_compat_exit); |
| 913 | else |
| 914 | state = 1; |
| 915 | } else { |
| 916 | -- |
| 917 | 2.19.1 |
| 918 | |