diff options
Diffstat (limited to 'meta/recipes-kernel/lttng/lttng-modules/0010-Fix-system-call-filter-table.patch')
| -rw-r--r-- | meta/recipes-kernel/lttng/lttng-modules/0010-Fix-system-call-filter-table.patch | 918 |
1 files changed, 918 insertions, 0 deletions
diff --git a/meta/recipes-kernel/lttng/lttng-modules/0010-Fix-system-call-filter-table.patch b/meta/recipes-kernel/lttng/lttng-modules/0010-Fix-system-call-filter-table.patch new file mode 100644 index 0000000000..a16750ddb3 --- /dev/null +++ b/meta/recipes-kernel/lttng/lttng-modules/0010-Fix-system-call-filter-table.patch @@ -0,0 +1,918 @@ +From ad594e3a953db1b0c3c059fde45b5a5494f6be78 Mon Sep 17 00:00:00 2001 +From: Mathieu Desnoyers <mathieu.desnoyers@efficios.com> +Date: Tue, 28 Jan 2020 16:02:44 -0500 +Subject: [PATCH 10/10] Fix: system call filter table + +The system call filter table has effectively been unused for a long +time due to system call name prefix mismatch. This means the overhead of +selective system call tracing was larger than it should have been because +the event payload preparation would be done for all system calls as soon +as a single system call is traced. + +However, fixing this underlying issue unearths several issues that crept +unnoticed when the "enabler" concept was introduced (after the original +implementation of the system call filter table). + +Here is a list of the issues which are resolved here: + +- Split lttng_syscalls_unregister into an unregister and destroy + function, thus awaiting for a grace period (and therefore quiescence + of the users) after unregistering the system call tracepoints before + freeing the system call filter data structures. This effectively fixes + a use-after-free. + +- The state for enabling "all" system calls vs enabling specific system + calls (and sequences of enable-disable) was incorrect with respect to + the "enablers" semantic. This is solved by always tracking the + bitmap of enabled system calls, and keeping this bitmap even when + enabling all system calls. The sc_filter is now always allocated + before system call tracing is registered to tracepoints, which means + it does not need to be RCU dereferenced anymore. + +Padding fields in the ABI are reserved to select whether to: + +- Trace either native or compat system call (or both, which is the + behavior currently implemented), +- Trace either system call entry or exit (or both, which is the + behavior currently implemented), +- Select the system call to trace by name (behavior currently + implemented) or by system call number, + +Upstream-Status: Backport + +Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com> +--- + lttng-abi.c | 43 ++++++ + lttng-abi.h | 26 ++++ + lttng-events.c | 112 +++++++++++++-- + lttng-events.h | 31 ++++- + lttng-syscalls.c | 348 +++++++++++++++++++++++++---------------------- + 5 files changed, 380 insertions(+), 180 deletions(-) + +diff --git a/lttng-abi.c b/lttng-abi.c +index 64ea99d..b33879d 100644 +--- a/lttng-abi.c ++++ b/lttng-abi.c +@@ -1264,6 +1264,46 @@ nomem: + return ret; + } + ++static ++int lttng_abi_validate_event_param(struct lttng_kernel_event *event_param) ++{ ++ /* Limit ABI to implemented features. */ ++ switch (event_param->instrumentation) { ++ case LTTNG_KERNEL_SYSCALL: ++ switch (event_param->u.syscall.entryexit) { ++ case LTTNG_KERNEL_SYSCALL_ENTRYEXIT: ++ break; ++ default: ++ return -EINVAL; ++ } ++ switch (event_param->u.syscall.abi) { ++ case LTTNG_KERNEL_SYSCALL_ABI_ALL: ++ break; ++ default: ++ return -EINVAL; ++ } ++ switch (event_param->u.syscall.match) { ++ case LTTNG_SYSCALL_MATCH_NAME: ++ break; ++ default: ++ return -EINVAL; ++ } ++ break; ++ ++ case LTTNG_KERNEL_TRACEPOINT: /* Fallthrough */ ++ case LTTNG_KERNEL_KPROBE: /* Fallthrough */ ++ case LTTNG_KERNEL_KRETPROBE: /* Fallthrough */ ++ case LTTNG_KERNEL_NOOP: /* Fallthrough */ ++ case LTTNG_KERNEL_UPROBE: ++ break; ++ ++ case LTTNG_KERNEL_FUNCTION: /* Fallthrough */ ++ default: ++ return -EINVAL; ++ } ++ return 0; ++} ++ + static + int lttng_abi_create_event(struct file *channel_file, + struct lttng_kernel_event *event_param) +@@ -1305,6 +1345,9 @@ int lttng_abi_create_event(struct file *channel_file, + ret = -EOVERFLOW; + goto refcount_error; + } ++ ret = lttng_abi_validate_event_param(event_param); ++ if (ret) ++ goto event_error; + if (event_param->instrumentation == LTTNG_KERNEL_TRACEPOINT + || event_param->instrumentation == LTTNG_KERNEL_SYSCALL) { + struct lttng_enabler *enabler; +diff --git a/lttng-abi.h b/lttng-abi.h +index 1d356ab..51d60e5 100644 +--- a/lttng-abi.h ++++ b/lttng-abi.h +@@ -90,6 +90,31 @@ struct lttng_kernel_event_callsite { + } u; + } __attribute__((packed)); + ++enum lttng_kernel_syscall_entryexit { ++ LTTNG_KERNEL_SYSCALL_ENTRYEXIT = 0, ++ LTTNG_KERNEL_SYSCALL_ENTRY = 1, /* Not implemented. */ ++ LTTNG_KERNEL_SYSCALL_EXIT = 2, /* Not implemented. */ ++}; ++ ++enum lttng_kernel_syscall_abi { ++ LTTNG_KERNEL_SYSCALL_ABI_ALL = 0, ++ LTTNG_KERNEL_SYSCALL_ABI_NATIVE = 1, /* Not implemented. */ ++ LTTNG_KERNEL_SYSCALL_ABI_COMPAT = 2, /* Not implemented. */ ++}; ++ ++enum lttng_kernel_syscall_match { ++ LTTNG_SYSCALL_MATCH_NAME = 0, ++ LTTNG_SYSCALL_MATCH_NR = 1, /* Not implemented. */ ++}; ++ ++struct lttng_kernel_syscall { ++ uint8_t entryexit; /* enum lttng_kernel_syscall_entryexit */ ++ uint8_t abi; /* enum lttng_kernel_syscall_abi */ ++ uint8_t match; /* enum lttng_kernel_syscall_match */ ++ uint8_t padding; ++ uint32_t nr; /* For LTTNG_SYSCALL_MATCH_NR */ ++} __attribute__((packed)); ++ + /* + * For syscall tracing, name = "*" means "enable all". + */ +@@ -106,6 +131,7 @@ struct lttng_kernel_event { + struct lttng_kernel_kprobe kprobe; + struct lttng_kernel_function_tracer ftrace; + struct lttng_kernel_uprobe uprobe; ++ struct lttng_kernel_syscall syscall; + char padding[LTTNG_KERNEL_EVENT_PADDING2]; + } u; + } __attribute__((packed)); +diff --git a/lttng-events.c b/lttng-events.c +index d719294..4c0b04a 100644 +--- a/lttng-events.c ++++ b/lttng-events.c +@@ -201,6 +201,10 @@ void lttng_session_destroy(struct lttng_session *session) + WARN_ON(ret); + } + synchronize_trace(); /* Wait for in-flight events to complete */ ++ list_for_each_entry(chan, &session->chan, list) { ++ ret = lttng_syscalls_destroy(chan); ++ WARN_ON(ret); ++ } + list_for_each_entry_safe(enabler, tmpenabler, + &session->enablers_head, node) + lttng_enabler_destroy(enabler); +@@ -740,6 +744,28 @@ struct lttng_event *_lttng_event_create(struct lttng_channel *chan, + event->enabled = 0; + event->registered = 0; + event->desc = event_desc; ++ switch (event_param->u.syscall.entryexit) { ++ case LTTNG_KERNEL_SYSCALL_ENTRYEXIT: ++ ret = -EINVAL; ++ goto register_error; ++ case LTTNG_KERNEL_SYSCALL_ENTRY: ++ event->u.syscall.entryexit = LTTNG_SYSCALL_ENTRY; ++ break; ++ case LTTNG_KERNEL_SYSCALL_EXIT: ++ event->u.syscall.entryexit = LTTNG_SYSCALL_EXIT; ++ break; ++ } ++ switch (event_param->u.syscall.abi) { ++ case LTTNG_KERNEL_SYSCALL_ABI_ALL: ++ ret = -EINVAL; ++ goto register_error; ++ case LTTNG_KERNEL_SYSCALL_ABI_NATIVE: ++ event->u.syscall.abi = LTTNG_SYSCALL_ABI_NATIVE; ++ break; ++ case LTTNG_KERNEL_SYSCALL_ABI_COMPAT: ++ event->u.syscall.abi = LTTNG_SYSCALL_ABI_COMPAT; ++ break; ++ } + if (!event->desc) { + ret = -EINVAL; + goto register_error; +@@ -826,8 +852,7 @@ void register_event(struct lttng_event *event) + event); + break; + case LTTNG_KERNEL_SYSCALL: +- ret = lttng_syscall_filter_enable(event->chan, +- desc->name); ++ ret = lttng_syscall_filter_enable(event->chan, event); + break; + case LTTNG_KERNEL_KPROBE: + case LTTNG_KERNEL_UPROBE: +@@ -870,8 +895,7 @@ int _lttng_event_unregister(struct lttng_event *event) + ret = 0; + break; + case LTTNG_KERNEL_SYSCALL: +- ret = lttng_syscall_filter_disable(event->chan, +- desc->name); ++ ret = lttng_syscall_filter_disable(event->chan, event); + break; + case LTTNG_KERNEL_NOOP: + ret = 0; +@@ -1203,39 +1227,87 @@ int lttng_desc_match_enabler(const struct lttng_event_desc *desc, + struct lttng_enabler *enabler) + { + const char *desc_name, *enabler_name; ++ bool compat = false, entry = false; + + enabler_name = enabler->event_param.name; + switch (enabler->event_param.instrumentation) { + case LTTNG_KERNEL_TRACEPOINT: + desc_name = desc->name; ++ switch (enabler->type) { ++ case LTTNG_ENABLER_STAR_GLOB: ++ return lttng_match_enabler_star_glob(desc_name, enabler_name); ++ case LTTNG_ENABLER_NAME: ++ return lttng_match_enabler_name(desc_name, enabler_name); ++ default: ++ return -EINVAL; ++ } + break; + case LTTNG_KERNEL_SYSCALL: + desc_name = desc->name; +- if (!strncmp(desc_name, "compat_", strlen("compat_"))) ++ if (!strncmp(desc_name, "compat_", strlen("compat_"))) { + desc_name += strlen("compat_"); ++ compat = true; ++ } + if (!strncmp(desc_name, "syscall_exit_", + strlen("syscall_exit_"))) { + desc_name += strlen("syscall_exit_"); + } else if (!strncmp(desc_name, "syscall_entry_", + strlen("syscall_entry_"))) { + desc_name += strlen("syscall_entry_"); ++ entry = true; + } else { + WARN_ON_ONCE(1); + return -EINVAL; + } ++ switch (enabler->event_param.u.syscall.entryexit) { ++ case LTTNG_KERNEL_SYSCALL_ENTRYEXIT: ++ break; ++ case LTTNG_KERNEL_SYSCALL_ENTRY: ++ if (!entry) ++ return 0; ++ break; ++ case LTTNG_KERNEL_SYSCALL_EXIT: ++ if (entry) ++ return 0; ++ break; ++ default: ++ return -EINVAL; ++ } ++ switch (enabler->event_param.u.syscall.abi) { ++ case LTTNG_KERNEL_SYSCALL_ABI_ALL: ++ break; ++ case LTTNG_KERNEL_SYSCALL_ABI_NATIVE: ++ if (compat) ++ return 0; ++ break; ++ case LTTNG_KERNEL_SYSCALL_ABI_COMPAT: ++ if (!compat) ++ return 0; ++ break; ++ default: ++ return -EINVAL; ++ } ++ switch (enabler->event_param.u.syscall.match) { ++ case LTTNG_SYSCALL_MATCH_NAME: ++ switch (enabler->type) { ++ case LTTNG_ENABLER_STAR_GLOB: ++ return lttng_match_enabler_star_glob(desc_name, enabler_name); ++ case LTTNG_ENABLER_NAME: ++ return lttng_match_enabler_name(desc_name, enabler_name); ++ default: ++ return -EINVAL; ++ } ++ break; ++ case LTTNG_SYSCALL_MATCH_NR: ++ return -EINVAL; /* Not implemented. */ ++ default: ++ return -EINVAL; ++ } + break; + default: + WARN_ON_ONCE(1); + return -EINVAL; + } +- switch (enabler->type) { +- case LTTNG_ENABLER_STAR_GLOB: +- return lttng_match_enabler_star_glob(desc_name, enabler_name); +- case LTTNG_ENABLER_NAME: +- return lttng_match_enabler_name(desc_name, enabler_name); +- default: +- return -EINVAL; +- } + } + + static +@@ -1361,9 +1433,21 @@ void lttng_create_event_if_missing(struct lttng_enabler *enabler) + static + int lttng_enabler_ref_events(struct lttng_enabler *enabler) + { +- struct lttng_session *session = enabler->chan->session; ++ struct lttng_channel *chan = enabler->chan; ++ struct lttng_session *session = chan->session; + struct lttng_event *event; + ++ if (enabler->event_param.instrumentation == LTTNG_KERNEL_SYSCALL && ++ enabler->event_param.u.syscall.entryexit == LTTNG_KERNEL_SYSCALL_ENTRYEXIT && ++ enabler->event_param.u.syscall.abi == LTTNG_KERNEL_SYSCALL_ABI_ALL && ++ enabler->event_param.u.syscall.match == LTTNG_SYSCALL_MATCH_NAME && ++ !strcmp(enabler->event_param.name, "*")) { ++ if (enabler->enabled) ++ WRITE_ONCE(chan->syscall_all, 1); ++ else ++ WRITE_ONCE(chan->syscall_all, 0); ++ } ++ + /* First ensure that probe events are created for this enabler. */ + lttng_create_event_if_missing(enabler); + +diff --git a/lttng-events.h b/lttng-events.h +index a36a312..d4d9976 100644 +--- a/lttng-events.h ++++ b/lttng-events.h +@@ -292,6 +292,16 @@ struct lttng_uprobe_handler { + struct list_head node; + }; + ++enum lttng_syscall_entryexit { ++ LTTNG_SYSCALL_ENTRY, ++ LTTNG_SYSCALL_EXIT, ++}; ++ ++enum lttng_syscall_abi { ++ LTTNG_SYSCALL_ABI_NATIVE, ++ LTTNG_SYSCALL_ABI_COMPAT, ++}; ++ + /* + * lttng_event structure is referred to by the tracing fast path. It must be + * kept small. +@@ -318,6 +328,11 @@ struct lttng_event { + struct inode *inode; + struct list_head head; + } uprobe; ++ struct { ++ char *syscall_name; ++ enum lttng_syscall_entryexit entryexit; ++ enum lttng_syscall_abi abi; ++ } syscall; + } u; + struct list_head list; /* Event list in session */ + unsigned int metadata_dumped:1; +@@ -457,10 +472,10 @@ struct lttng_channel { + struct lttng_syscall_filter *sc_filter; + int header_type; /* 0: unset, 1: compact, 2: large */ + enum channel_type channel_type; ++ int syscall_all; + unsigned int metadata_dumped:1, + sys_enter_registered:1, + sys_exit_registered:1, +- syscall_all:1, + tstate:1; /* Transient enable state */ + }; + +@@ -653,10 +668,11 @@ void lttng_clock_unref(void); + #if defined(CONFIG_HAVE_SYSCALL_TRACEPOINTS) + int lttng_syscalls_register(struct lttng_channel *chan, void *filter); + int lttng_syscalls_unregister(struct lttng_channel *chan); ++int lttng_syscalls_destroy(struct lttng_channel *chan); + int lttng_syscall_filter_enable(struct lttng_channel *chan, +- const char *name); ++ struct lttng_event *event); + int lttng_syscall_filter_disable(struct lttng_channel *chan, +- const char *name); ++ struct lttng_event *event); + long lttng_channel_syscall_mask(struct lttng_channel *channel, + struct lttng_kernel_syscall_mask __user *usyscall_mask); + #else +@@ -670,14 +686,19 @@ static inline int lttng_syscalls_unregister(struct lttng_channel *chan) + return 0; + } + ++static inline int lttng_syscalls_destroy(struct lttng_channel *chan) ++{ ++ return 0; ++} ++ + static inline int lttng_syscall_filter_enable(struct lttng_channel *chan, +- const char *name) ++ struct lttng_event *event); + { + return -ENOSYS; + } + + static inline int lttng_syscall_filter_disable(struct lttng_channel *chan, +- const char *name) ++ struct lttng_event *event); + { + return -ENOSYS; + } +diff --git a/lttng-syscalls.c b/lttng-syscalls.c +index 97f1ba9..26cead6 100644 +--- a/lttng-syscalls.c ++++ b/lttng-syscalls.c +@@ -367,8 +367,10 @@ const struct trace_syscall_entry compat_sc_exit_table[] = { + #undef CREATE_SYSCALL_TABLE + + struct lttng_syscall_filter { +- DECLARE_BITMAP(sc, NR_syscalls); +- DECLARE_BITMAP(sc_compat, NR_compat_syscalls); ++ DECLARE_BITMAP(sc_entry, NR_syscalls); ++ DECLARE_BITMAP(sc_exit, NR_syscalls); ++ DECLARE_BITMAP(sc_compat_entry, NR_compat_syscalls); ++ DECLARE_BITMAP(sc_compat_exit, NR_compat_syscalls); + }; + + static void syscall_entry_unknown(struct lttng_event *event, +@@ -391,29 +393,23 @@ void syscall_entry_probe(void *__data, struct pt_regs *regs, long id) + size_t table_len; + + if (unlikely(in_compat_syscall())) { +- struct lttng_syscall_filter *filter; +- +- filter = lttng_rcu_dereference(chan->sc_filter); +- if (filter) { +- if (id < 0 || id >= NR_compat_syscalls +- || !test_bit(id, filter->sc_compat)) { +- /* System call filtered out. */ +- return; +- } ++ struct lttng_syscall_filter *filter = chan->sc_filter; ++ ++ if (id < 0 || id >= NR_compat_syscalls ++ || (!READ_ONCE(chan->syscall_all) && !test_bit(id, filter->sc_compat_entry))) { ++ /* System call filtered out. */ ++ return; + } + table = compat_sc_table; + table_len = ARRAY_SIZE(compat_sc_table); + unknown_event = chan->sc_compat_unknown; + } else { +- struct lttng_syscall_filter *filter; +- +- filter = lttng_rcu_dereference(chan->sc_filter); +- if (filter) { +- if (id < 0 || id >= NR_syscalls +- || !test_bit(id, filter->sc)) { +- /* System call filtered out. */ +- return; +- } ++ struct lttng_syscall_filter *filter = chan->sc_filter; ++ ++ if (id < 0 || id >= NR_syscalls ++ || (!READ_ONCE(chan->syscall_all) && !test_bit(id, filter->sc_entry))) { ++ /* System call filtered out. */ ++ return; + } + table = sc_table; + table_len = ARRAY_SIZE(sc_table); +@@ -545,29 +541,23 @@ void syscall_exit_probe(void *__data, struct pt_regs *regs, long ret) + + id = syscall_get_nr(current, regs); + if (unlikely(in_compat_syscall())) { +- struct lttng_syscall_filter *filter; +- +- filter = lttng_rcu_dereference(chan->sc_filter); +- if (filter) { +- if (id < 0 || id >= NR_compat_syscalls +- || !test_bit(id, filter->sc_compat)) { +- /* System call filtered out. */ +- return; +- } ++ struct lttng_syscall_filter *filter = chan->sc_filter; ++ ++ if (id < 0 || id >= NR_compat_syscalls ++ || (!READ_ONCE(chan->syscall_all) && !test_bit(id, filter->sc_compat_exit))) { ++ /* System call filtered out. */ ++ return; + } + table = compat_sc_exit_table; + table_len = ARRAY_SIZE(compat_sc_exit_table); + unknown_event = chan->compat_sc_exit_unknown; + } else { +- struct lttng_syscall_filter *filter; +- +- filter = lttng_rcu_dereference(chan->sc_filter); +- if (filter) { +- if (id < 0 || id >= NR_syscalls +- || !test_bit(id, filter->sc)) { +- /* System call filtered out. */ +- return; +- } ++ struct lttng_syscall_filter *filter = chan->sc_filter; ++ ++ if (id < 0 || id >= NR_syscalls ++ || (!READ_ONCE(chan->syscall_all) && !test_bit(id, filter->sc_exit))) { ++ /* System call filtered out. */ ++ return; + } + table = sc_exit_table; + table_len = ARRAY_SIZE(sc_exit_table); +@@ -713,27 +703,23 @@ int fill_table(const struct trace_syscall_entry *table, size_t table_len, + memset(&ev, 0, sizeof(ev)); + switch (type) { + case SC_TYPE_ENTRY: +- strncpy(ev.name, SYSCALL_ENTRY_STR, +- LTTNG_KERNEL_SYM_NAME_LEN); ++ ev.u.syscall.entryexit = LTTNG_KERNEL_SYSCALL_ENTRY; ++ ev.u.syscall.abi = LTTNG_KERNEL_SYSCALL_ABI_NATIVE; + break; + case SC_TYPE_EXIT: +- strncpy(ev.name, SYSCALL_EXIT_STR, +- LTTNG_KERNEL_SYM_NAME_LEN); ++ ev.u.syscall.entryexit = LTTNG_KERNEL_SYSCALL_EXIT; ++ ev.u.syscall.abi = LTTNG_KERNEL_SYSCALL_ABI_NATIVE; + break; + case SC_TYPE_COMPAT_ENTRY: +- strncpy(ev.name, COMPAT_SYSCALL_ENTRY_STR, +- LTTNG_KERNEL_SYM_NAME_LEN); ++ ev.u.syscall.entryexit = LTTNG_KERNEL_SYSCALL_ENTRY; ++ ev.u.syscall.abi = LTTNG_KERNEL_SYSCALL_ABI_COMPAT; + break; + case SC_TYPE_COMPAT_EXIT: +- strncpy(ev.name, COMPAT_SYSCALL_EXIT_STR, +- LTTNG_KERNEL_SYM_NAME_LEN); +- break; +- default: +- BUG_ON(1); ++ ev.u.syscall.entryexit = LTTNG_KERNEL_SYSCALL_EXIT; ++ ev.u.syscall.abi = LTTNG_KERNEL_SYSCALL_ABI_COMPAT; + break; + } +- strncat(ev.name, desc->name, +- LTTNG_KERNEL_SYM_NAME_LEN - strlen(ev.name) - 1); ++ strncpy(ev.name, desc->name, LTTNG_KERNEL_SYM_NAME_LEN); + ev.name[LTTNG_KERNEL_SYM_NAME_LEN - 1] = '\0'; + ev.instrumentation = LTTNG_KERNEL_SYSCALL; + chan_table[i] = _lttng_event_create(chan, &ev, filter, +@@ -803,6 +789,8 @@ int lttng_syscalls_register(struct lttng_channel *chan, void *filter) + strncpy(ev.name, desc->name, LTTNG_KERNEL_SYM_NAME_LEN); + ev.name[LTTNG_KERNEL_SYM_NAME_LEN - 1] = '\0'; + ev.instrumentation = LTTNG_KERNEL_SYSCALL; ++ ev.u.syscall.entryexit = LTTNG_KERNEL_SYSCALL_ENTRY; ++ ev.u.syscall.abi = LTTNG_KERNEL_SYSCALL_ABI_NATIVE; + chan->sc_unknown = _lttng_event_create(chan, &ev, filter, + desc, + ev.instrumentation); +@@ -820,6 +808,8 @@ int lttng_syscalls_register(struct lttng_channel *chan, void *filter) + strncpy(ev.name, desc->name, LTTNG_KERNEL_SYM_NAME_LEN); + ev.name[LTTNG_KERNEL_SYM_NAME_LEN - 1] = '\0'; + ev.instrumentation = LTTNG_KERNEL_SYSCALL; ++ ev.u.syscall.entryexit = LTTNG_KERNEL_SYSCALL_ENTRY; ++ ev.u.syscall.abi = LTTNG_KERNEL_SYSCALL_ABI_COMPAT; + chan->sc_compat_unknown = _lttng_event_create(chan, &ev, filter, + desc, + ev.instrumentation); +@@ -837,6 +827,8 @@ int lttng_syscalls_register(struct lttng_channel *chan, void *filter) + strncpy(ev.name, desc->name, LTTNG_KERNEL_SYM_NAME_LEN); + ev.name[LTTNG_KERNEL_SYM_NAME_LEN - 1] = '\0'; + ev.instrumentation = LTTNG_KERNEL_SYSCALL; ++ ev.u.syscall.entryexit = LTTNG_KERNEL_SYSCALL_EXIT; ++ ev.u.syscall.abi = LTTNG_KERNEL_SYSCALL_ABI_COMPAT; + chan->compat_sc_exit_unknown = _lttng_event_create(chan, &ev, + filter, desc, + ev.instrumentation); +@@ -854,6 +846,8 @@ int lttng_syscalls_register(struct lttng_channel *chan, void *filter) + strncpy(ev.name, desc->name, LTTNG_KERNEL_SYM_NAME_LEN); + ev.name[LTTNG_KERNEL_SYM_NAME_LEN - 1] = '\0'; + ev.instrumentation = LTTNG_KERNEL_SYSCALL; ++ ev.u.syscall.entryexit = LTTNG_KERNEL_SYSCALL_EXIT; ++ ev.u.syscall.abi = LTTNG_KERNEL_SYSCALL_ABI_NATIVE; + chan->sc_exit_unknown = _lttng_event_create(chan, &ev, filter, + desc, ev.instrumentation); + WARN_ON_ONCE(!chan->sc_exit_unknown); +@@ -883,6 +877,14 @@ int lttng_syscalls_register(struct lttng_channel *chan, void *filter) + if (ret) + return ret; + #endif ++ ++ if (!chan->sc_filter) { ++ chan->sc_filter = kzalloc(sizeof(struct lttng_syscall_filter), ++ GFP_KERNEL); ++ if (!chan->sc_filter) ++ return -ENOMEM; ++ } ++ + if (!chan->sys_enter_registered) { + ret = lttng_wrapper_tracepoint_probe_register("sys_enter", + (void *) syscall_entry_probe, chan); +@@ -930,7 +932,11 @@ int lttng_syscalls_unregister(struct lttng_channel *chan) + return ret; + chan->sys_exit_registered = 0; + } +- /* lttng_event destroy will be performed by lttng_session_destroy() */ ++ return 0; ++} ++ ++int lttng_syscalls_destroy(struct lttng_channel *chan) ++{ + kfree(chan->sc_table); + kfree(chan->sc_exit_table); + #ifdef CONFIG_COMPAT +@@ -993,136 +999,150 @@ uint32_t get_sc_tables_len(void) + return ARRAY_SIZE(sc_table) + ARRAY_SIZE(compat_sc_table); + } + +-int lttng_syscall_filter_enable(struct lttng_channel *chan, +- const char *name) ++static ++const char *get_syscall_name(struct lttng_event *event) + { +- int syscall_nr, compat_syscall_nr, ret; +- struct lttng_syscall_filter *filter; ++ size_t prefix_len = 0; + +- WARN_ON_ONCE(!chan->sc_table); ++ WARN_ON_ONCE(event->instrumentation != LTTNG_KERNEL_SYSCALL); + +- if (!name) { +- /* Enable all system calls by removing filter */ +- if (chan->sc_filter) { +- filter = chan->sc_filter; +- rcu_assign_pointer(chan->sc_filter, NULL); +- synchronize_trace(); +- kfree(filter); ++ switch (event->u.syscall.entryexit) { ++ case LTTNG_SYSCALL_ENTRY: ++ switch (event->u.syscall.abi) { ++ case LTTNG_SYSCALL_ABI_NATIVE: ++ prefix_len = strlen(SYSCALL_ENTRY_STR); ++ break; ++ case LTTNG_SYSCALL_ABI_COMPAT: ++ prefix_len = strlen(COMPAT_SYSCALL_ENTRY_STR); ++ break; + } +- chan->syscall_all = 1; +- return 0; +- } +- +- if (!chan->sc_filter) { +- if (chan->syscall_all) { +- /* +- * All syscalls are already enabled. +- */ +- return -EEXIST; ++ break; ++ case LTTNG_SYSCALL_EXIT: ++ switch (event->u.syscall.abi) { ++ case LTTNG_SYSCALL_ABI_NATIVE: ++ prefix_len = strlen(SYSCALL_EXIT_STR); ++ break; ++ case LTTNG_SYSCALL_ABI_COMPAT: ++ prefix_len = strlen(COMPAT_SYSCALL_EXIT_STR); ++ break; + } +- filter = kzalloc(sizeof(struct lttng_syscall_filter), +- GFP_KERNEL); +- if (!filter) +- return -ENOMEM; +- } else { +- filter = chan->sc_filter; ++ break; + } +- syscall_nr = get_syscall_nr(name); +- compat_syscall_nr = get_compat_syscall_nr(name); +- if (syscall_nr < 0 && compat_syscall_nr < 0) { +- ret = -ENOENT; +- goto error; ++ WARN_ON_ONCE(prefix_len == 0); ++ return event->desc->name + prefix_len; ++} ++ ++int lttng_syscall_filter_enable(struct lttng_channel *chan, ++ struct lttng_event *event) ++{ ++ struct lttng_syscall_filter *filter = chan->sc_filter; ++ const char *syscall_name; ++ unsigned long *bitmap; ++ int syscall_nr; ++ ++ WARN_ON_ONCE(!chan->sc_table); ++ ++ syscall_name = get_syscall_name(event); ++ ++ switch (event->u.syscall.abi) { ++ case LTTNG_SYSCALL_ABI_NATIVE: ++ syscall_nr = get_syscall_nr(syscall_name); ++ break; ++ case LTTNG_SYSCALL_ABI_COMPAT: ++ syscall_nr = get_compat_syscall_nr(syscall_name); ++ break; ++ default: ++ return -EINVAL; + } +- if (syscall_nr >= 0) { +- if (test_bit(syscall_nr, filter->sc)) { +- ret = -EEXIST; +- goto error; ++ if (syscall_nr < 0) ++ return -ENOENT; ++ ++ ++ switch (event->u.syscall.entryexit) { ++ case LTTNG_SYSCALL_ENTRY: ++ switch (event->u.syscall.abi) { ++ case LTTNG_SYSCALL_ABI_NATIVE: ++ bitmap = filter->sc_entry; ++ break; ++ case LTTNG_SYSCALL_ABI_COMPAT: ++ bitmap = filter->sc_compat_entry; ++ break; + } +- bitmap_set(filter->sc, syscall_nr, 1); +- } +- if (compat_syscall_nr >= 0) { +- if (test_bit(compat_syscall_nr, filter->sc_compat)) { +- ret = -EEXIST; +- goto error; ++ break; ++ case LTTNG_SYSCALL_EXIT: ++ switch (event->u.syscall.abi) { ++ case LTTNG_SYSCALL_ABI_NATIVE: ++ bitmap = filter->sc_exit; ++ break; ++ case LTTNG_SYSCALL_ABI_COMPAT: ++ bitmap = filter->sc_compat_exit; ++ break; + } +- bitmap_set(filter->sc_compat, compat_syscall_nr, 1); ++ break; ++ default: ++ return -EINVAL; + } +- if (!chan->sc_filter) +- rcu_assign_pointer(chan->sc_filter, filter); ++ if (test_bit(syscall_nr, bitmap)) ++ return -EEXIST; ++ bitmap_set(bitmap, syscall_nr, 1); + return 0; +- +-error: +- if (!chan->sc_filter) +- kfree(filter); +- return ret; + } + + int lttng_syscall_filter_disable(struct lttng_channel *chan, +- const char *name) ++ struct lttng_event *event) + { +- int syscall_nr, compat_syscall_nr, ret; +- struct lttng_syscall_filter *filter; ++ struct lttng_syscall_filter *filter = chan->sc_filter; ++ const char *syscall_name; ++ unsigned long *bitmap; ++ int syscall_nr; + + WARN_ON_ONCE(!chan->sc_table); + +- if (!chan->sc_filter) { +- if (!chan->syscall_all) +- return -EEXIST; +- filter = kzalloc(sizeof(struct lttng_syscall_filter), +- GFP_KERNEL); +- if (!filter) +- return -ENOMEM; +- /* Trace all system calls, then apply disable. */ +- bitmap_set(filter->sc, 0, NR_syscalls); +- bitmap_set(filter->sc_compat, 0, NR_compat_syscalls); +- } else { +- filter = chan->sc_filter; ++ syscall_name = get_syscall_name(event); ++ ++ switch (event->u.syscall.abi) { ++ case LTTNG_SYSCALL_ABI_NATIVE: ++ syscall_nr = get_syscall_nr(syscall_name); ++ break; ++ case LTTNG_SYSCALL_ABI_COMPAT: ++ syscall_nr = get_compat_syscall_nr(syscall_name); ++ break; ++ default: ++ return -EINVAL; + } ++ if (syscall_nr < 0) ++ return -ENOENT; + +- if (!name) { +- /* Fail if all syscalls are already disabled. */ +- if (bitmap_empty(filter->sc, NR_syscalls) +- && bitmap_empty(filter->sc_compat, +- NR_compat_syscalls)) { +- ret = -EEXIST; +- goto error; +- } + +- /* Disable all system calls */ +- bitmap_clear(filter->sc, 0, NR_syscalls); +- bitmap_clear(filter->sc_compat, 0, NR_compat_syscalls); +- goto apply_filter; +- } +- syscall_nr = get_syscall_nr(name); +- compat_syscall_nr = get_compat_syscall_nr(name); +- if (syscall_nr < 0 && compat_syscall_nr < 0) { +- ret = -ENOENT; +- goto error; +- } +- if (syscall_nr >= 0) { +- if (!test_bit(syscall_nr, filter->sc)) { +- ret = -EEXIST; +- goto error; ++ switch (event->u.syscall.entryexit) { ++ case LTTNG_SYSCALL_ENTRY: ++ switch (event->u.syscall.abi) { ++ case LTTNG_SYSCALL_ABI_NATIVE: ++ bitmap = filter->sc_entry; ++ break; ++ case LTTNG_SYSCALL_ABI_COMPAT: ++ bitmap = filter->sc_compat_entry; ++ break; + } +- bitmap_clear(filter->sc, syscall_nr, 1); +- } +- if (compat_syscall_nr >= 0) { +- if (!test_bit(compat_syscall_nr, filter->sc_compat)) { +- ret = -EEXIST; +- goto error; ++ break; ++ case LTTNG_SYSCALL_EXIT: ++ switch (event->u.syscall.abi) { ++ case LTTNG_SYSCALL_ABI_NATIVE: ++ bitmap = filter->sc_exit; ++ break; ++ case LTTNG_SYSCALL_ABI_COMPAT: ++ bitmap = filter->sc_compat_exit; ++ break; + } +- bitmap_clear(filter->sc_compat, compat_syscall_nr, 1); ++ break; ++ default: ++ return -EINVAL; + } +-apply_filter: +- if (!chan->sc_filter) +- rcu_assign_pointer(chan->sc_filter, filter); +- chan->syscall_all = 0; +- return 0; ++ if (!test_bit(syscall_nr, bitmap)) ++ return -EEXIST; ++ bitmap_clear(bitmap, syscall_nr, 1); + +-error: +- if (!chan->sc_filter) +- kfree(filter); +- return ret; ++ return 0; + } + + static +@@ -1236,6 +1256,9 @@ const struct file_operations lttng_syscall_list_fops = { + .release = seq_release, + }; + ++/* ++ * A syscall is enabled if it is traced for either entry or exit. ++ */ + long lttng_channel_syscall_mask(struct lttng_channel *channel, + struct lttng_kernel_syscall_mask __user *usyscall_mask) + { +@@ -1262,8 +1285,9 @@ long lttng_channel_syscall_mask(struct lttng_channel *channel, + char state; + + if (channel->sc_table) { +- if (filter) +- state = test_bit(bit, filter->sc); ++ if (!READ_ONCE(channel->syscall_all) && filter) ++ state = test_bit(bit, filter->sc_entry) ++ || test_bit(bit, filter->sc_exit); + else + state = 1; + } else { +@@ -1275,9 +1299,11 @@ long lttng_channel_syscall_mask(struct lttng_channel *channel, + char state; + + if (channel->compat_sc_table) { +- if (filter) ++ if (!READ_ONCE(channel->syscall_all) && filter) + state = test_bit(bit - ARRAY_SIZE(sc_table), +- filter->sc_compat); ++ filter->sc_compat_entry) ++ || test_bit(bit - ARRAY_SIZE(sc_table), ++ filter->sc_compat_exit); + else + state = 1; + } else { +-- +2.19.1 + |