]> git.proxmox.com Git - mirror_ubuntu-focal-kernel.git/commitdiff
perf trace: Put the per-syscall entry/exit prog_array BPF map infrastructure in place
authorArnaldo Carvalho de Melo <acme@redhat.com>
Tue, 16 Jul 2019 14:27:03 +0000 (11:27 -0300)
committerArnaldo Carvalho de Melo <acme@redhat.com>
Mon, 29 Jul 2019 21:34:41 +0000 (18:34 -0300)
I.e. look for "syscalls_sys_enter" and "syscalls_sys_exit" BPF maps of
type PROG_ARRAY and populate it with the handlers as specified per
syscall, for now only 'open' is wiring it to something, in time all
syscalls that need to copy arguments entering a syscall or returning
from one will set these to the right handlers, reusing when possible
pre-existing ones.

Next step is to use bpf_tail_call() into that.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Luis Cláudio Gonçalves <lclaudio@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: https://lkml.kernel.org/n/tip-t0p4u43i9vbpzs1xtowna3gb@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
tools/perf/builtin-trace.c
tools/perf/examples/bpf/augmented_raw_syscalls.c

index 6cc696edf24af65bc215824775ece45894380189..fb8b8e78d7b500aea4ae75f8742b3bb3ef27a4ba 100644 (file)
@@ -1,4 +1,3 @@
-// SPDX-License-Identifier: GPL-2.0-only
 /*
  * builtin-trace.c
  *
@@ -83,6 +82,10 @@ struct trace {
                int             max;
                struct syscall  *table;
                struct bpf_map  *map;
+               struct { // per syscall BPF_MAP_TYPE_PROG_ARRAY
+                       struct bpf_map  *sys_enter,
+                                       *sys_exit;
+               }               prog_array;
                struct {
                        struct perf_evsel *sys_enter,
                                          *sys_exit,
@@ -1619,6 +1622,22 @@ out_free:
        goto out;
 }
 
+static __maybe_unused bool trace__syscall_enabled(struct trace *trace, int id)
+{
+       bool in_ev_qualifier;
+
+       if (trace->ev_qualifier_ids.nr == 0)
+               return true;
+
+       in_ev_qualifier = bsearch(&id, trace->ev_qualifier_ids.entries,
+                                 trace->ev_qualifier_ids.nr, sizeof(int), intcmp) != NULL;
+
+       if (in_ev_qualifier)
+              return !trace->not_ev_qualifier;
+
+       return trace->not_ev_qualifier;
+}
+
 /*
  * args is to be interpreted as a series of longs but we need to handle
  * 8-byte unaligned accesses. args points to raw_data within the event
@@ -2784,6 +2803,18 @@ static void trace__init_syscall_bpf_progs(struct trace *trace, int id)
        }
 }
 
+static int trace__bpf_prog_sys_enter_fd(struct trace *trace, int id)
+{
+       struct syscall *sc = trace__syscall_info(trace, NULL, id);
+       return sc ? bpf_program__fd(sc->bpf_prog.sys_enter) : bpf_program__fd(trace->syscalls.unaugmented_prog);
+}
+
+static int trace__bpf_prog_sys_exit_fd(struct trace *trace, int id)
+{
+       struct syscall *sc = trace__syscall_info(trace, NULL, id);
+       return sc ? bpf_program__fd(sc->bpf_prog.sys_exit) : bpf_program__fd(trace->syscalls.unaugmented_prog);
+}
+
 static void trace__init_bpf_map_syscall_args(struct trace *trace, int id, struct bpf_map_syscall_entry *entry)
 {
        struct syscall *sc = trace__syscall_info(trace, NULL, id);
@@ -2837,10 +2868,8 @@ static int __trace__init_syscalls_bpf_map(struct trace *trace, bool enabled)
        int err = 0, key;
 
        for (key = 0; key < trace->sctbl->syscalls.nr_entries; ++key) {
-               if (enabled) {
+               if (enabled)
                        trace__init_bpf_map_syscall_args(trace, key, &value);
-                       trace__init_syscall_bpf_progs(trace, key);
-               }
 
                err = bpf_map_update_elem(fd, &key, &value, BPF_ANY);
                if (err)
@@ -2859,6 +2888,34 @@ static int trace__init_syscalls_bpf_map(struct trace *trace)
 
        return __trace__init_syscalls_bpf_map(trace, enabled);
 }
+
+static int trace__init_syscalls_bpf_prog_array_maps(struct trace *trace)
+{
+       int map_enter_fd = bpf_map__fd(trace->syscalls.prog_array.sys_enter),
+           map_exit_fd  = bpf_map__fd(trace->syscalls.prog_array.sys_exit);
+       int err = 0, key;
+
+       for (key = 0; key < trace->sctbl->syscalls.nr_entries; ++key) {
+               int prog_fd;
+
+               if (!trace__syscall_enabled(trace, key))
+                       continue;
+
+               trace__init_syscall_bpf_progs(trace, key);
+
+               // It'll get at least the "!raw_syscalls:unaugmented"
+               prog_fd = trace__bpf_prog_sys_enter_fd(trace, key);
+               err = bpf_map_update_elem(map_enter_fd, &key, &prog_fd, BPF_ANY);
+               if (err)
+                       break;
+               prog_fd = trace__bpf_prog_sys_exit_fd(trace, key);
+               err = bpf_map_update_elem(map_exit_fd, &key, &prog_fd, BPF_ANY);
+               if (err)
+                       break;
+       }
+
+       return err;
+}
 #else
 static int trace__set_ev_qualifier_bpf_filter(struct trace *trace __maybe_unused)
 {
@@ -2875,6 +2932,11 @@ static struct bpf_program *trace__find_bpf_program_by_title(struct trace *trace
 {
        return NULL;
 }
+
+static int trace__init_syscalls_bpf_prog_array_maps(struct trace *trace __maybe_unused)
+{
+       return 0;
+}
 #endif // HAVE_LIBBPF_SUPPORT
 
 static int trace__set_ev_qualifier_filter(struct trace *trace)
@@ -3129,6 +3191,10 @@ static int trace__run(struct trace *trace, int argc, const char **argv)
        if (trace->syscalls.map)
                trace__init_syscalls_bpf_map(trace);
 
+       if (trace->syscalls.prog_array.sys_enter)
+               trace__init_syscalls_bpf_prog_array_maps(trace);
+
+
        if (trace->ev_qualifier_ids.nr > 0) {
                err = trace__set_ev_qualifier_filter(trace);
                if (err < 0)
@@ -3754,6 +3820,8 @@ static void trace__set_bpf_map_filtered_pids(struct trace *trace)
 static void trace__set_bpf_map_syscalls(struct trace *trace)
 {
        trace->syscalls.map = trace__find_bpf_map_by_name(trace, "syscalls");
+       trace->syscalls.prog_array.sys_enter = trace__find_bpf_map_by_name(trace, "syscalls_sys_enter");
+       trace->syscalls.prog_array.sys_exit  = trace__find_bpf_map_by_name(trace, "syscalls_sys_exit");
 }
 
 static int trace__config(const char *var, const char *value, void *arg)
index 66b33b29934993983d7fc20720e9a7b49572931a..c66474a6ccf4289b321f802d6dede090345ab1c3 100644 (file)
@@ -33,6 +33,20 @@ struct syscall {
 
 bpf_map(syscalls, ARRAY, int, struct syscall, 512);
 
+/*
+ * What to augment at entry?
+ *
+ * Pointer arg payloads (filenames, etc) passed from userspace to the kernel
+ */
+bpf_map(syscalls_sys_enter, PROG_ARRAY, u32, u32, 512);
+
+/*
+ * What to augment at exit?
+ *
+ * Pointer arg payloads returned from the kernel (struct stat, etc) to userspace.
+ */
+bpf_map(syscalls_sys_exit, PROG_ARRAY, u32, u32, 512);
+
 struct syscall_enter_args {
        unsigned long long common_tp_fields;
        long               syscall_nr;