Merge pull request #1607 from palmtenor/no_reader

Remove PERF_SAMPLE_CALLCHAIN logic from TRACEPOINT events

src/cc/api/BPF.cc

@@ -159,8 +159,7 @@ StatusTuple BPF::detach_all() {
 
 StatusTuple BPF::attach_kprobe(const std::string& kernel_func,
                                const std::string& probe_func,
-                               bpf_probe_attach_type attach_type,
-                               perf_reader_cb cb, void* cb_cookie) {
+                               bpf_probe_attach_type attach_type) {
   std::string probe_event = get_kprobe_event(kernel_func, attach_type);
   if (kprobes_.find(probe_event) != kprobes_.end())
     return StatusTuple(-1, "kprobe %s already attached", probe_event.c_str());
@@ -168,10 +167,10 @@ StatusTuple BPF::attach_kprobe(const std::string& kernel_func,
   int probe_fd;
   TRY2(load_func(probe_func, BPF_PROG_TYPE_KPROBE, probe_fd));
 
-  void* res = bpf_attach_kprobe(probe_fd, attach_type, probe_event.c_str(),
-                                kernel_func.c_str(), cb, cb_cookie);
+  int res_fd = bpf_attach_kprobe(probe_fd, attach_type, probe_event.c_str(),
+                                 kernel_func.c_str());
 
-  if (!res) {
+  if (res_fd < 0) {
     TRY2(unload_func(probe_func));
     return StatusTuple(-1, "Unable to attach %skprobe for %s using %s",
                        attach_type_debug(attach_type).c_str(),
@@ -179,7 +178,7 @@ StatusTuple BPF::attach_kprobe(const std::string& kernel_func,
   }
 
   open_probe_t p = {};
-  p.reader_ptr = res;
+  p.perf_event_fd = res_fd;
   p.func = probe_func;
   kprobes_[probe_event] = std::move(p);
   return StatusTuple(0);
@@ -189,8 +188,7 @@ StatusTuple BPF::attach_uprobe(const std::string& binary_path,
                                const std::string& symbol,
                                const std::string& probe_func,
                                uint64_t symbol_addr,
-                               bpf_probe_attach_type attach_type, pid_t pid,
-                               perf_reader_cb cb, void* cb_cookie) {
+                               bpf_probe_attach_type attach_type, pid_t pid) {
   std::string module;
   uint64_t offset;
   TRY2(check_binary_symbol(binary_path, symbol, symbol_addr, module, offset));
@@ -202,11 +200,10 @@ StatusTuple BPF::attach_uprobe(const std::string& binary_path,
   int probe_fd;
   TRY2(load_func(probe_func, BPF_PROG_TYPE_KPROBE, probe_fd));
 
-  void* res =
-      bpf_attach_uprobe(probe_fd, attach_type, probe_event.c_str(),
-                        binary_path.c_str(), offset, pid, cb, cb_cookie);
+  int res_fd = bpf_attach_uprobe(probe_fd, attach_type, probe_event.c_str(),
+                                 binary_path.c_str(), offset, pid);
 
-  if (!res) {
+  if (res_fd < 0) {
     TRY2(unload_func(probe_func));
     return StatusTuple(
         -1,
@@ -216,7 +213,7 @@ StatusTuple BPF::attach_uprobe(const std::string& binary_path,
   }
 
   open_probe_t p = {};
-  p.reader_ptr = res;
+  p.perf_event_fd = res_fd;
   p.func = probe_func;
   uprobes_[probe_event] = std::move(p);
   return StatusTuple(0);
@@ -253,8 +250,7 @@ StatusTuple BPF::attach_usdt(const USDT& usdt, pid_t pid) {
 }
 
 StatusTuple BPF::attach_tracepoint(const std::string& tracepoint,
-                                   const std::string& probe_func,
-                                   perf_reader_cb cb, void* cb_cookie) {
+                                   const std::string& probe_func) {
   if (tracepoints_.find(tracepoint) != tracepoints_.end())
     return StatusTuple(-1, "Tracepoint %s already attached",
                        tracepoint.c_str());
@@ -268,17 +264,17 @@ StatusTuple BPF::attach_tracepoint(const std::string& tracepoint,
   int probe_fd;
   TRY2(load_func(probe_func, BPF_PROG_TYPE_TRACEPOINT, probe_fd));
 
-  void* res = bpf_attach_tracepoint(probe_fd, tp_category.c_str(),
-                                    tp_name.c_str(), cb, cb_cookie);
+  int res_fd = bpf_attach_tracepoint(probe_fd, tp_category.c_str(),
+                                    tp_name.c_str());
 
-  if (!res) {
+  if (res_fd < 0) {
     TRY2(unload_func(probe_func));
     return StatusTuple(-1, "Unable to attach Tracepoint %s using %s",
                        tracepoint.c_str(), probe_func.c_str());
   }
 
   open_probe_t p = {};
-  p.reader_ptr = res;
+  p.perf_event_fd = res_fd;
   p.func = probe_func;
   tracepoints_[tracepoint] = std::move(p);
   return StatusTuple(0);
@@ -558,10 +554,7 @@ std::string BPF::get_uprobe_event(const std::string& binary_path,
 
 StatusTuple BPF::detach_kprobe_event(const std::string& event,
                                      open_probe_t& attr) {
-  if (attr.reader_ptr) {
-    perf_reader_free(attr.reader_ptr);
-    attr.reader_ptr = nullptr;
-  }
+  bpf_close_perf_event_fd(attr.perf_event_fd);
   TRY2(unload_func(attr.func));
   if (bpf_detach_kprobe(event.c_str()) < 0)
     return StatusTuple(-1, "Unable to detach kprobe %s", event.c_str());
@@ -570,10 +563,7 @@ StatusTuple BPF::detach_kprobe_event(const std::string& event,
 
 StatusTuple BPF::detach_uprobe_event(const std::string& event,
                                      open_probe_t& attr) {
-  if (attr.reader_ptr) {
-    perf_reader_free(attr.reader_ptr);
-    attr.reader_ptr = nullptr;
-  }
+  bpf_close_perf_event_fd(attr.perf_event_fd);
   TRY2(unload_func(attr.func));
   if (bpf_detach_uprobe(event.c_str()) < 0)
     return StatusTuple(-1, "Unable to detach uprobe %s", event.c_str());
@@ -582,10 +572,7 @@ StatusTuple BPF::detach_uprobe_event(const std::string& event,
 
 StatusTuple BPF::detach_tracepoint_event(const std::string& tracepoint,
                                          open_probe_t& attr) {
-  if (attr.reader_ptr) {
-    perf_reader_free(attr.reader_ptr);
-    attr.reader_ptr = nullptr;
-  }
+  bpf_close_perf_event_fd(attr.perf_event_fd);
   TRY2(unload_func(attr.func));
 
   // TODO: bpf_detach_tracepoint currently does nothing.

src/cc/api/BPF.h

@@ -34,7 +34,7 @@ static const int DEFAULT_PERF_BUFFER_PAGE_CNT = 8;
 namespace ebpf {
 
 struct open_probe_t {
-  void* reader_ptr;
+  int perf_event_fd;
   std::string func;
   std::map<int, int>* per_cpu_fd;
 };
@@ -57,9 +57,7 @@ class BPF {
 
   StatusTuple attach_kprobe(const std::string& kernel_func,
                             const std::string& probe_func,
-                            bpf_probe_attach_type = BPF_PROBE_ENTRY,
-                            perf_reader_cb cb = nullptr,
-                            void* cb_cookie = nullptr);
+                            bpf_probe_attach_type = BPF_PROBE_ENTRY);
   StatusTuple detach_kprobe(
       const std::string& kernel_func,
       bpf_probe_attach_type attach_type = BPF_PROBE_ENTRY);
@@ -69,8 +67,7 @@ class BPF {
                             const std::string& probe_func,
                             uint64_t symbol_addr = 0,
                             bpf_probe_attach_type attach_type = BPF_PROBE_ENTRY,
-                            pid_t pid = -1, perf_reader_cb cb = nullptr,
-                            void* cb_cookie = nullptr);
+                            pid_t pid = -1);
   StatusTuple detach_uprobe(const std::string& binary_path,
                             const std::string& symbol, uint64_t symbol_addr = 0,
                             bpf_probe_attach_type attach_type = BPF_PROBE_ENTRY,
@@ -79,9 +76,7 @@ class BPF {
   StatusTuple detach_usdt(const USDT& usdt);
 
   StatusTuple attach_tracepoint(const std::string& tracepoint,
-                                const std::string& probe_func,
-                                perf_reader_cb cb = nullptr,
-                                void* cb_cookie = nullptr);
+                                const std::string& probe_func);
   StatusTuple detach_tracepoint(const std::string& tracepoint);
 
   StatusTuple attach_perf_event(uint32_t ev_type, uint32_t ev_config,

src/cc/libbpf.c

@@ -151,8 +151,6 @@ static struct bpf_helper helpers[] = {
   {"sock_ops_cb_flags_set", "4.16"},
 };
 
-static int probe_perf_reader_page_cnt = 8;
-
 static uint64_t ptr_to_u64(void *ptr)
 {
   return (uint64_t) (unsigned long) ptr;
@@ -682,7 +680,6 @@ static int bpf_try_perf_event_open_with_probe(const char *name, uint64_t offs,
 
   if (type < 0 || is_return_bit < 0)
     return -1;
-  attr.sample_type = PERF_SAMPLE_RAW | PERF_SAMPLE_CALLCHAIN;
   attr.sample_period = 1;
   attr.wakeup_events = 1;
   if (is_return)
@@ -724,16 +721,23 @@ static int bpf_try_perf_event_open_with_probe(const char *name, uint64_t offs,
                  PERF_FLAG_FD_CLOEXEC);
 }
 
-static int bpf_attach_tracing_event(int progfd, const char *event_path,
-                                    struct perf_reader *reader, int pid,
-                                    int pfd)
+// When a valid Perf Event FD provided through pfd, it will be used to enable
+// and attach BPF program to the event, and event_path will be ignored.
+// Otherwise, event_path is expected to contain the path to the event in debugfs
+// and it will be used to open the Perf Event FD.
+// In either case, if the attach partially failed (such as issue with the
+// ioctl operations), the **caller** need to clean up the Perf Event FD, either
+// provided by the caller or opened here.
+static int bpf_attach_tracing_event(int progfd, const char *event_path, int pid,
+                                    int *pfd)
 {
   int efd, cpu = 0;
   ssize_t bytes;
   char buf[PATH_MAX];
   struct perf_event_attr attr = {};
-
-  if (pfd < 0) {
+  // Caller did not provided a valid Perf Event FD. Create one with the debugfs
+  // event path provided.
+  if (*pfd < 0) {
     snprintf(buf, sizeof(buf), "%s/id", event_path);
     efd = open(buf, O_RDONLY, 0);
     if (efd < 0) {
@@ -751,7 +755,6 @@ static int bpf_attach_tracing_event(int progfd, const char *event_path,
     buf[bytes] = '\0';
     attr.config = strtol(buf, NULL, 0);
     attr.type = PERF_TYPE_TRACEPOINT;
-    attr.sample_type = PERF_SAMPLE_RAW | PERF_SAMPLE_CALLCHAIN;
     attr.sample_period = 1;
     attr.wakeup_events = 1;
     // PID filter is only possible for uprobe events.
@@ -762,22 +765,18 @@ static int bpf_attach_tracing_event(int progfd, const char *event_path,
     // Tracing events do not do CPU filtering in any cases.
     if (pid != -1)
       cpu = -1;
-    pfd = syscall(__NR_perf_event_open, &attr, pid, cpu, -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
-    if (pfd < 0) {
+    *pfd = syscall(__NR_perf_event_open, &attr, pid, cpu, -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
+    if (*pfd < 0) {
       fprintf(stderr, "perf_event_open(%s/id): %s\n", event_path, strerror(errno));
       return -1;
     }
   }
-  perf_reader_set_fd(reader, pfd);
-
-  if (perf_reader_mmap(reader, attr.type, attr.sample_type) < 0)
-    return -1;
 
-  if (ioctl(pfd, PERF_EVENT_IOC_SET_BPF, progfd) < 0) {
+  if (ioctl(*pfd, PERF_EVENT_IOC_SET_BPF, progfd) < 0) {
     perror("ioctl(PERF_EVENT_IOC_SET_BPF)");
     return -1;
   }
-  if (ioctl(pfd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
+  if (ioctl(*pfd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
     perror("ioctl(PERF_EVENT_IOC_ENABLE)");
     return -1;
   }
@@ -785,24 +784,19 @@ static int bpf_attach_tracing_event(int progfd, const char *event_path,
   return 0;
 }
 
-void *bpf_attach_kprobe(int progfd, enum bpf_probe_attach_type attach_type,
-                        const char *ev_name, const char *fn_name,
-                        perf_reader_cb cb, void *cb_cookie)
+int bpf_attach_kprobe(int progfd, enum bpf_probe_attach_type attach_type,
+                      const char *ev_name, const char *fn_name)
 {
-  int kfd;
-  int pfd;
+  int kfd, pfd = -1;
   char buf[256];
   char event_alias[128];
-  struct perf_reader *reader = NULL;
   static char *event_type = "kprobe";
 
-  reader = perf_reader_new(cb, NULL, NULL, cb_cookie, probe_perf_reader_page_cnt);
-  if (!reader)
-    goto error;
-
+  // Try create the kprobe Perf Event with perf_event_open API.
   pfd = bpf_try_perf_event_open_with_probe(fn_name, 0, -1, event_type,
                                            attach_type != BPF_PROBE_ENTRY);
-
+  // If failed, most likely Kernel doesn't support the new perf_event_open API
+  // yet. Try create the event using debugfs.
   if (pfd < 0) {
     snprintf(buf, sizeof(buf), "/sys/kernel/debug/tracing/%s_events", event_type);
     kfd = open(buf, O_WRONLY | O_APPEND, 0);
@@ -823,15 +817,16 @@ void *bpf_attach_kprobe(int progfd, enum bpf_probe_attach_type attach_type,
     close(kfd);
     snprintf(buf, sizeof(buf), "/sys/kernel/debug/tracing/events/%ss/%s", event_type, event_alias);
   }
-  if (bpf_attach_tracing_event(progfd, buf, reader, -1 /* PID */, pfd) < 0)
-    goto error;
-
-  return reader;
+  // If perf_event_open succeeded, bpf_attach_tracing_event will use the created
+  // Perf Event FD directly and buf would be empty and unused.
+  // Otherwise it will read the event ID from the path in buf, create the
+  // Perf Event event using that ID, and updated value of pfd.
+  if (bpf_attach_tracing_event(progfd, buf, -1 /* PID */, &pfd) == 0)
+    return pfd;
 
 error:
-  perf_reader_free(reader);
-  return NULL;
-
+  bpf_close_perf_event_fd(pfd);
+  return -1;
 }
 
 static int enter_mount_ns(int pid) {
@@ -895,24 +890,19 @@ static void exit_mount_ns(int fd) {
     perror("setns");
 }
 
-void *bpf_attach_uprobe(int progfd, enum bpf_probe_attach_type attach_type,
-                        const char *ev_name, const char *binary_path,
-                        uint64_t offset, pid_t pid, perf_reader_cb cb,
-                        void *cb_cookie)
+int bpf_attach_uprobe(int progfd, enum bpf_probe_attach_type attach_type,
+                      const char *ev_name, const char *binary_path,
+                      uint64_t offset, pid_t pid)
 {
   char buf[PATH_MAX];
   char event_alias[PATH_MAX];
-  struct perf_reader *reader = NULL;
   static char *event_type = "uprobe";
-  int res, kfd = -1, ns_fd = -1;
-  int pfd;
-
-  reader = perf_reader_new(cb, NULL, NULL, cb_cookie, probe_perf_reader_page_cnt);
-  if (!reader)
-    goto error;
-
+  int res, kfd = -1, pfd = -1, ns_fd = -1;
+  // Try create the uprobe Perf Event with perf_event_open API.
   pfd = bpf_try_perf_event_open_with_probe(binary_path, offset, pid, event_type,
                                            attach_type != BPF_PROBE_ENTRY);
+  // If failed, most likely Kernel doesn't support the new perf_event_open API
+  // yet. Try create the event using debugfs.
   if (pfd < 0) {
     snprintf(buf, sizeof(buf), "/sys/kernel/debug/tracing/%s_events", event_type);
     kfd = open(buf, O_WRONLY | O_APPEND, 0);
@@ -940,22 +930,25 @@ void *bpf_attach_uprobe(int progfd, enum bpf_probe_attach_type attach_type,
       goto error;
     }
     close(kfd);
+    kfd = -1;
     exit_mount_ns(ns_fd);
     ns_fd = -1;
 
     snprintf(buf, sizeof(buf), "/sys/kernel/debug/tracing/events/%ss/%s", event_type, event_alias);
   }
-  if (bpf_attach_tracing_event(progfd, buf, reader, pid, pfd) < 0)
-    goto error;
-
-  return reader;
+  // If perf_event_open succeeded, bpf_attach_tracing_event will use the created
+  // Perf Event FD directly and buf would be empty and unused.
+  // Otherwise it will read the event ID from the path in buf, create the
+  // Perf Event event using that ID, and updated value of pfd.
+  if (bpf_attach_tracing_event(progfd, buf, pid, &pfd) == 0)
+    return pfd;
 
 error:
   if (kfd >= 0)
     close(kfd);
   exit_mount_ns(ns_fd);
-  perf_reader_free(reader);
-  return NULL;
+  bpf_close_perf_event_fd(pfd);
+  return -1;
 }
 
 static int bpf_detach_probe(const char *ev_name, const char *event_type)
@@ -1036,26 +1029,19 @@ int bpf_detach_uprobe(const char *ev_name)
 }
 
 
-void *bpf_attach_tracepoint(int progfd, const char *tp_category,
-                            const char *tp_name, perf_reader_cb cb,
-                            void *cb_cookie) {
+int bpf_attach_tracepoint(int progfd, const char *tp_category,
+                          const char *tp_name)
+{
   char buf[256];
-  struct perf_reader *reader = NULL;
-
-  reader = perf_reader_new(cb, NULL, NULL, cb_cookie, probe_perf_reader_page_cnt);
-  if (!reader)
-    goto error;
+  int pfd = -1;
 
   snprintf(buf, sizeof(buf), "/sys/kernel/debug/tracing/events/%s/%s",
            tp_category, tp_name);
-  if (bpf_attach_tracing_event(progfd, buf, reader, -1 /* PID */, -1 /* pfd */) < 0)
-    goto error;
+  if (bpf_attach_tracing_event(progfd, buf, -1 /* PID */, &pfd) == 0)
+    return pfd;
 
-  return reader;
-
-error:
-  perf_reader_free(reader);
-  return NULL;
+  bpf_close_perf_event_fd(pfd);
+  return -1;
 }
 
 int bpf_detach_tracepoint(const char *tp_category, const char *tp_name) {
@@ -1071,7 +1057,7 @@ void * bpf_open_perf_buffer(perf_reader_raw_cb raw_cb,
   struct perf_event_attr attr = {};
   struct perf_reader *reader = NULL;
 
-  reader = perf_reader_new(NULL, raw_cb, lost_cb, cb_cookie, page_cnt);
+  reader = perf_reader_new(raw_cb, lost_cb, cb_cookie, page_cnt);
   if (!reader)
     goto error;
 
@@ -1088,7 +1074,7 @@ void * bpf_open_perf_buffer(perf_reader_raw_cb raw_cb,
   }
   perf_reader_set_fd(reader, pfd);
 
-  if (perf_reader_mmap(reader, attr.type, attr.sample_type) < 0)
+  if (perf_reader_mmap(reader) < 0)
     goto error;
 
   if (ioctl(pfd, PERF_EVENT_IOC_ENABLE, 0) < 0) {

src/cc/libbpf.h

@@ -65,27 +65,20 @@ int bpf_attach_socket(int sockfd, int progfd);
  * bind the raw socket to the interface 'name' */
 int bpf_open_raw_sock(const char *name);
 
-typedef void (*perf_reader_cb)(void *cb_cookie, int pid, uint64_t callchain_num,
-                               void *callchain);
 typedef void (*perf_reader_raw_cb)(void *cb_cookie, void *raw, int raw_size);
 typedef void (*perf_reader_lost_cb)(void *cb_cookie, uint64_t lost);
 
-void *bpf_attach_kprobe(int progfd, enum bpf_probe_attach_type attach_type,
-                        const char *ev_name, const char *fn_name,
-                        perf_reader_cb cb, void *cb_cookie);
-
+int bpf_attach_kprobe(int progfd, enum bpf_probe_attach_type attach_type,
+                      const char *ev_name, const char *fn_name);
 int bpf_detach_kprobe(const char *ev_name);
 
-void *bpf_attach_uprobe(int progfd, enum bpf_probe_attach_type attach_type,
-                        const char *ev_name, const char *binary_path,
-                        uint64_t offset, pid_t pid, perf_reader_cb cb,
-                        void *cb_cookie);
-
+int bpf_attach_uprobe(int progfd, enum bpf_probe_attach_type attach_type,
+                      const char *ev_name, const char *binary_path,
+                      uint64_t offset, pid_t pid);
 int bpf_detach_uprobe(const char *ev_name);
 
-void *bpf_attach_tracepoint(int progfd, const char *tp_category,
-                            const char *tp_name, perf_reader_cb cb,
-                            void *cb_cookie);
+int bpf_attach_tracepoint(int progfd, const char *tp_category,
+                          const char *tp_name);
 int bpf_detach_tracepoint(const char *tp_category, const char *tp_name);
 
 void * bpf_open_perf_buffer(perf_reader_raw_cb raw_cb,

src/cc/perf_reader.c

@@ -37,7 +37,6 @@ enum {
 };
 
 struct perf_reader {
-  perf_reader_cb cb;
   perf_reader_raw_cb raw_cb;
   perf_reader_lost_cb lost_cb;
   void *cb_cookie; // to be returned in the cb
@@ -49,18 +48,14 @@ struct perf_reader {
   int page_size;
   int page_cnt;
   int fd;
-  uint32_t type;
-  uint64_t sample_type;
 };
 
-struct perf_reader * perf_reader_new(perf_reader_cb cb,
-                                     perf_reader_raw_cb raw_cb,
+struct perf_reader * perf_reader_new(perf_reader_raw_cb raw_cb,
                                      perf_reader_lost_cb lost_cb,
                                      void *cb_cookie, int page_cnt) {
   struct perf_reader *reader = calloc(1, sizeof(struct perf_reader));
   if (!reader)
     return NULL;
-  reader->cb = cb;
   reader->raw_cb = raw_cb;
   reader->lost_cb = lost_cb;
   reader->cb_cookie = cb_cookie;
@@ -89,7 +84,7 @@ void perf_reader_free(void *ptr) {
   }
 }
 
-int perf_reader_mmap(struct perf_reader *reader, unsigned type, unsigned long sample_type) {
+int perf_reader_mmap(struct perf_reader *reader) {
   int mmap_size = reader->page_size * (reader->page_cnt + 1);
 
   if (reader->fd < 0) {
@@ -102,8 +97,6 @@ int perf_reader_mmap(struct perf_reader *reader, unsigned type, unsigned long sa
     perror("mmap");
     return -1;
   }
-  reader->type = type;
-  reader->sample_type = sample_type;
 
   return 0;
 }
@@ -120,69 +113,6 @@ struct perf_sample_trace_kprobe {
   uint64_t ip;
 };
 
-static void parse_tracepoint(struct perf_reader *reader, void *data, int size) {
-  uint8_t *ptr = data;
-  struct perf_event_header *header = (void *)data;
-
-  struct perf_sample_trace_kprobe *tk = NULL;
-  uint64_t *callchain = NULL;
-  uint64_t num_callchain = 0;
-
-  ptr += sizeof(*header);
-  if (ptr > (uint8_t *)data + size) {
-    fprintf(stderr, "%s: corrupt sample header\n", __FUNCTION__);
-    return;
-  }
-
-  if (reader->sample_type & PERF_SAMPLE_CALLCHAIN) {
-    struct {
-      uint64_t nr;
-      uint64_t ips[0];
-    } *cc = (void *)ptr;
-    ptr += sizeof(cc->nr) + sizeof(*cc->ips) * cc->nr;
-    // size sanity check
-    if (ptr > (uint8_t *)data + size) {
-      fprintf(stderr, "%s: corrupt callchain sample\n", __FUNCTION__);
-      return;
-    }
-    int i;
-    // don't include magic numbers in the call chain
-    for (i = 0; i < cc->nr; ++i) {
-      if (cc->ips[i] == PERF_CONTEXT_USER)
-        break;
-      if (cc->ips[i] >= PERF_CONTEXT_MAX)
-        continue;
-      if (!callchain)
-        callchain = &cc->ips[i];
-      ++num_callchain;
-    }
-  }
-  // for kprobes, raw samples just include the common data structure and the
-  // instruction pointer
-  if (reader->sample_type & PERF_SAMPLE_RAW) {
-    struct {
-      uint32_t size;
-      char data[0];
-    } *raw = (void *)ptr;
-    ptr += sizeof(raw->size) + raw->size;
-    if (ptr > (uint8_t *)data + size) {
-      fprintf(stderr, "%s: corrupt raw sample\n", __FUNCTION__);
-      return;
-    }
-    tk = (void *)raw->data;
-  }
-
-  // sanity check
-  if (ptr != (uint8_t *)data + size) {
-    fprintf(stderr, "%s: extra data at end of sample\n", __FUNCTION__);
-    return;
-  }
-
-  // call out to the user with the parsed data
-  if (reader->cb)
-    reader->cb(reader->cb_cookie, tk ? tk->common.pid : -1, num_callchain, callchain);
-}
-
 static void parse_sw(struct perf_reader *reader, void *data, int size) {
   uint8_t *ptr = data;
   struct perf_event_header *header = (void *)data;
@@ -198,13 +128,11 @@ static void parse_sw(struct perf_reader *reader, void *data, int size) {
     return;
   }
 
-  if (reader->sample_type & PERF_SAMPLE_RAW) {
-    raw = (void *)ptr;
-    ptr += sizeof(raw->size) + raw->size;
-    if (ptr > (uint8_t *)data + size) {
-      fprintf(stderr, "%s: corrupt raw sample\n", __FUNCTION__);
-      return;
-    }
+  raw = (void *)ptr;
+  ptr += sizeof(raw->size) + raw->size;
+  if (ptr > (uint8_t *)data + size) {
+    fprintf(stderr, "%s: corrupt raw sample\n", __FUNCTION__);
+    return;
   }
 
   // sanity check
@@ -278,10 +206,7 @@ void perf_reader_event_read(struct perf_reader *reader) {
         fprintf(stderr, "Possibly lost %" PRIu64 " samples\n", lost);
       }
     } else if (e->type == PERF_RECORD_SAMPLE) {
-      if (reader->type == PERF_TYPE_TRACEPOINT)
-        parse_tracepoint(reader, ptr, e->size);
-      else if (reader->type == PERF_TYPE_SOFTWARE)
-        parse_sw(reader, ptr, e->size);
+      parse_sw(reader, ptr, e->size);
     } else {
       fprintf(stderr, "%s: unknown sample type %d\n", __FUNCTION__, e->type);
     }

src/cc/perf_reader.h

@@ -25,12 +25,11 @@ extern "C" {
 
 struct perf_reader;
 
-struct perf_reader * perf_reader_new(perf_reader_cb cb,
-                                     perf_reader_raw_cb raw_cb,
+struct perf_reader * perf_reader_new(perf_reader_raw_cb raw_cb,
                                      perf_reader_lost_cb lost_cb,
                                      void *cb_cookie, int page_cnt);
 void perf_reader_free(void *ptr);
-int perf_reader_mmap(struct perf_reader *reader, unsigned type, unsigned long sample_type);
+int perf_reader_mmap(struct perf_reader *reader);
 void perf_reader_event_read(struct perf_reader *reader);
 int perf_reader_poll(int num_readers, struct perf_reader **readers, int timeout);
 int perf_reader_fd(struct perf_reader *reader);

src/lua/bcc/bpf.lua

@@ -24,6 +24,7 @@ local Bpf = class("BPF")
 
 Bpf.static.open_kprobes = {}
 Bpf.static.open_uprobes = {}
+Bpf.static.perf_buffers = {}
 Bpf.static.KPROBE_LIMIT = 1000
 Bpf.static.tracer_pipe = nil
 Bpf.static.DEFAULT_CFLAGS = {
@@ -39,8 +40,8 @@ end
 
 function Bpf.static.cleanup()
   local function detach_all(probe_type, all_probes)
-    for key, probe in pairs(all_probes) do
-      libbcc.perf_reader_free(probe)
+    for key, fd in pairs(all_probes) do
+      libbcc.bpf_close_perf_event_fd(fd)
       -- skip bcc-specific kprobes
       if not key:starts("bcc:") then
         if probe_type == "kprobes" then
@@ -55,6 +56,12 @@ function Bpf.static.cleanup()
 
   detach_all("kprobes", Bpf.static.open_kprobes)
   detach_all("uprobes", Bpf.static.open_uprobes)
+
+  for key, perf_buffer in pairs(Bpf.static.perf_buffers) do
+    libbcc.perf_reader_free(perf_buffer)
+    Bpf.static.perf_buffers[key] = nil
+  end
+
   if Bpf.static.tracer_pipe ~= nil then
     Bpf.static.tracer_pipe:close()
   end
@@ -189,9 +196,9 @@ function Bpf:attach_uprobe(args)
   local retprobe = args.retprobe and 1 or 0
 
   local res = libbcc.bpf_attach_uprobe(fn.fd, retprobe, ev_name, path, addr,
-    args.pid or -1, nil, nil) -- TODO; reader callback
+    args.pid or -1)
 
-  assert(res ~= nil, "failed to attach BPF to uprobe")
+  assert(res >= 0, "failed to attach BPF to uprobe")
   self:probe_store("uprobe", ev_name, res)
   return self
 end
@@ -206,10 +213,9 @@ function Bpf:attach_kprobe(args)
   local ev_name = string.format("%s_%s", ptype, event:gsub("[%+%.]", "_"))
   local retprobe = args.retprobe and 1 or 0
 
-  local res = libbcc.bpf_attach_kprobe(fn.fd, retprobe, ev_name, event,
-    nil, nil) -- TODO; reader callback
+  local res = libbcc.bpf_attach_kprobe(fn.fd, retprobe, ev_name, event)
 
-  assert(res ~= nil, "failed to attach BPF to kprobe")
+  assert(res >= 0, "failed to attach BPF to kprobe")
   self:probe_store("kprobe", ev_name, res)
   return self
 end
@@ -228,16 +234,22 @@ function Bpf:get_table(name, key_type, leaf_type)
   return self.tables[name]
 end
 
-function Bpf:probe_store(t, id, reader)
+function Bpf:probe_store(t, id, fd)
   if t == "kprobe" then
-    Bpf.open_kprobes[id] = reader
+    Bpf.open_kprobes[id] = fd
   elseif t == "uprobe" then
-    Bpf.open_uprobes[id] = reader
+    Bpf.open_uprobes[id] = fd
   else
     error("unknown probe type '%s'" % t)
   end
 
-  log.info("%s -> %s", id, reader)
+  log.info("%s -> %s", id, fd)
+end
+
+function Bpf:perf_buffer_store(t, id, reader)
+    Bpf.perf_buffers[id] = reader
+
+    log.info("%s -> %s", id, fd)
 end
 
 function Bpf:probe_lookup(t, id)
@@ -250,32 +262,32 @@ function Bpf:probe_lookup(t, id)
   end
 end
 
-function Bpf:_kprobe_array()
-  local kprobe_count = table.count(Bpf.open_kprobes)
-  local readers = ffi.new("struct perf_reader*[?]", kprobe_count)
+function Bpf:_perf_buffer_array()
+  local perf_buffer_count = table.count(Bpf.perf_buffers)
+  local readers = ffi.new("struct perf_reader*[?]", perf_buffer_count)
   local n = 0
 
-  for _, r in pairs(Bpf.open_kprobes) do
+  for _, r in pairs(Bpf.perf_buffers) do
     readers[n] = r
     n = n + 1
   end
 
-  assert(n == kprobe_count)
+  assert(n == perf_buffer_count)
   return readers, n
 end
 
 function Bpf:kprobe_poll_loop()
-  local probes, probe_count = self:_kprobe_array()
+  local perf_buffer, perf_buffer_count = self:_perf_buffer_array()
   return pcall(function()
     while true do
-      libbcc.perf_reader_poll(probe_count, probes, -1)
+      libbcc.perf_reader_poll(perf_buffer_count, perf_buffers, -1)
     end
   end)
 end
 
 function Bpf:kprobe_poll(timeout)
-  local probes, probe_count = self:_kprobe_array()
-  libbcc.perf_reader_poll(probe_count, probes, timeout or -1)
+  local perf_buffer, perf_buffer_count = self:_perf_buffer_array()
+  libbcc.perf_reader_poll(perf_buffer_count, perf_buffers, timeout or -1)
 end
 
 return Bpf

src/lua/bcc/libbcc.lua

@@ -39,23 +39,22 @@ int bpf_attach_socket(int sockfd, int progfd);
 /* create RAW socket and bind to interface 'name' */
 int bpf_open_raw_sock(const char *name);
 
-typedef void (*perf_reader_cb)(void *cb_cookie, int pid, uint64_t callchain_num, void *callchain);
 typedef void (*perf_reader_raw_cb)(void *cb_cookie, void *raw, int raw_size);
 typedef void (*perf_reader_lost_cb)(void *cb_cookie, uint64_t lost);
 
-void *bpf_attach_kprobe(int progfd, int attach_type, const char *ev_name,
-                        const char *fn_name, perf_reader_cb cb,
-                        void *cb_cookie);
+int bpf_attach_kprobe(int progfd, int attach_type, const char *ev_name,
+                      const char *fn_name);
 
 int bpf_detach_kprobe(const char *ev_name);
 
-void *bpf_attach_uprobe(int progfd, int attach_type, const char *ev_name,
-                        const char *binary_path, uint64_t offset, int pid,
-                        perf_reader_cb cb, void *cb_cookie);
+int bpf_attach_uprobe(int progfd, int attach_type, const char *ev_name,
+                      const char *binary_path, uint64_t offset, int pid);
 
 int bpf_detach_uprobe(const char *ev_name);
 
 void * bpf_open_perf_buffer(perf_reader_raw_cb raw_cb, perf_reader_lost_cb lost_cb, void *cb_cookie, int pid, int cpu, int page_cnt);
+
+int bpf_close_perf_event_fd(int fd);
 ]]
 
 ffi.cdef[[
@@ -100,7 +99,7 @@ ffi.cdef[[
 struct perf_reader;
 
 void perf_reader_free(void *ptr);
-int perf_reader_mmap(struct perf_reader *reader, unsigned type, unsigned long sample_type);
+int perf_reader_mmap(struct perf_reader *reader);
 int perf_reader_poll(int num_readers, struct perf_reader **readers, int timeout);
 int perf_reader_fd(struct perf_reader *reader);
 void perf_reader_set_fd(struct perf_reader *reader, int fd);

src/lua/bcc/table.lua

@@ -263,7 +263,7 @@ function PerfEventArray:_open_perf_buffer(cpu, callback, ctype, page_cnt, lost_c
 
   local fd = libbcc.perf_reader_fd(reader)
   self:set(cpu, fd)
-  self.bpf:probe_store("kprobe", _perf_id(self.map_id, cpu), reader)
+  self.bpf:perf_buffer_store(_perf_id(self.map_id, cpu), reader)
   self._callbacks[cpu] = _cb
 end
 

src/python/bcc/libbcc.py

@@ -85,23 +85,19 @@ lib.bpf_attach_socket.argtypes = [ct.c_int, ct.c_int]
 lib.bpf_prog_load.restype = ct.c_int
 lib.bpf_prog_load.argtypes = [ct.c_int, ct.c_char_p, ct.c_void_p,
         ct.c_size_t, ct.c_char_p, ct.c_uint, ct.c_int, ct.c_char_p, ct.c_uint]
-lib.bpf_attach_kprobe.restype = ct.c_void_p
-_CB_TYPE = ct.CFUNCTYPE(None, ct.py_object, ct.c_int,
-        ct.c_ulonglong, ct.POINTER(ct.c_ulonglong))
 _RAW_CB_TYPE = ct.CFUNCTYPE(None, ct.py_object, ct.c_void_p, ct.c_int)
 _LOST_CB_TYPE = ct.CFUNCTYPE(None, ct.py_object, ct.c_ulonglong)
-lib.bpf_attach_kprobe.argtypes = [ct.c_int, ct.c_int, ct.c_char_p, ct.c_char_p,
-        _CB_TYPE, ct.py_object]
+lib.bpf_attach_kprobe.restype = ct.c_int
+lib.bpf_attach_kprobe.argtypes = [ct.c_int, ct.c_int, ct.c_char_p, ct.c_char_p]
 lib.bpf_detach_kprobe.restype = ct.c_int
 lib.bpf_detach_kprobe.argtypes = [ct.c_char_p]
-lib.bpf_attach_uprobe.restype = ct.c_void_p
+lib.bpf_attach_uprobe.restype = ct.c_int
 lib.bpf_attach_uprobe.argtypes = [ct.c_int, ct.c_int, ct.c_char_p, ct.c_char_p,
-        ct.c_ulonglong, ct.c_int, _CB_TYPE, ct.py_object]
+        ct.c_ulonglong, ct.c_int]
 lib.bpf_detach_uprobe.restype = ct.c_int
 lib.bpf_detach_uprobe.argtypes = [ct.c_char_p]
-lib.bpf_attach_tracepoint.restype = ct.c_void_p
-lib.bpf_attach_tracepoint.argtypes = [ct.c_int, ct.c_char_p, ct.c_char_p,
-        _CB_TYPE, ct.py_object]
+lib.bpf_attach_tracepoint.restype = ct.c_int
+lib.bpf_attach_tracepoint.argtypes = [ct.c_int, ct.c_char_p, ct.c_char_p]
 lib.bpf_detach_tracepoint.restype = ct.c_int
 lib.bpf_detach_tracepoint.argtypes = [ct.c_char_p, ct.c_char_p]
 lib.bpf_open_perf_buffer.restype = ct.c_void_p

src/python/bcc/table.py

@@ -494,10 +494,10 @@ class PerfEventArray(ArrayBase):
         # Delete entry from the array
         super(PerfEventArray, self).__delitem__(key)
         key_id = (id(self), key)
-        if key_id in self.bpf.open_kprobes:
+        if key_id in self.bpf.perf_buffers:
             # The key is opened for perf ring buffer
-            lib.perf_reader_free(self.bpf.open_kprobes[key_id])
-            self.bpf._del_kprobe(key_id)
+            lib.perf_reader_free(self.bpf.perf_buffers[key_id])
+            del self.bpf.perf_buffers[key_id]
             del self._cbs[key]
         else:
             # The key is opened for perf event read
@@ -544,7 +544,7 @@ class PerfEventArray(ArrayBase):
             raise Exception("Could not open perf buffer")
         fd = lib.perf_reader_fd(reader)
         self[self.Key(cpu)] = self.Leaf(fd)
-        self.bpf._add_kprobe((id(self), cpu), reader)
+        self.bpf.perf_buffers[(id(self), cpu)] = reader
         # keep a refcnt
         self._cbs[cpu] = (fn, lost_fn)
         # The actual fd is held by the perf reader, add to track opened keys

tests/python/CMakeLists.txt

@@ -46,8 +46,6 @@ add_test(NAME py_test_clang WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}
   COMMAND ${TEST_WRAPPER} py_clang sudo ${CMAKE_CURRENT_SOURCE_DIR}/test_clang.py)
 add_test(NAME py_test_histogram WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}
   COMMAND ${TEST_WRAPPER} py_histogram sudo ${CMAKE_CURRENT_SOURCE_DIR}/test_histogram.py)
-add_test(NAME py_test_callchain WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}
-  COMMAND ${TEST_WRAPPER} py_callchain sudo ${CMAKE_CURRENT_SOURCE_DIR}/test_callchain.py)
 add_test(NAME py_array WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}
   COMMAND ${TEST_WRAPPER} py_array sudo ${CMAKE_CURRENT_SOURCE_DIR}/test_array.py)
 add_test(NAME py_uprobes WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}

tests/python/test_callchain.py

-#!/usr/bin/env python
-# Copyright (c) PLUMgrid, Inc.
-# Licensed under the Apache License, Version 2.0 (the "License")
-
-from bcc import BPF
-import time
-from unittest import main, TestCase
-
-class TestCallchain(TestCase):
-    def test_callchain1(self):
-        hist = {}
-        def cb(pid, callchain):
-            counter = hist.get(callchain, 0)
-            counter += 1
-            hist[callchain] = counter
-
-        b = BPF(text="""
-#include <linux/ptrace.h>
-int kprobe__finish_task_switch(struct pt_regs *ctx) {
-    return 1;
-}
-""", cb=cb)
-        start = time.time()
-        while time.time() < start + 1:
-            b.kprobe_poll()
-
-        for k, v in hist.items():
-            syms = [b.ksym(addr) for addr in k]
-            print("%-08d:" % v, syms)
-
-if __name__ == "__main__":
-    main()