perf report: Add validation of call-chain entries
authorIngo Molnar <mingo@elte.hu>
Thu, 18 Jun 2009 06:00:17 +0000 (08:00 +0200)
committerIngo Molnar <mingo@elte.hu>
Thu, 18 Jun 2009 06:15:47 +0000 (08:15 +0200)
Add boundary checks for call-chain events. In case of corrupted
entries we could crash otherwise.

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
include/linux/perf_counter.h
tools/perf/builtin-report.c

index eccae43..a7d3a61 100644 (file)
@@ -337,6 +337,16 @@ enum perf_event_type {
         */
 };
 
+#define MAX_STACK_DEPTH                        255
+
+struct perf_callchain_entry {
+       __u16                           nr;
+       __u16                           hv;
+       __u16                           kernel;
+       __u16                           user;
+       __u64                           ip[MAX_STACK_DEPTH];
+};
+
 #ifdef __KERNEL__
 /*
  * Kernel-internal data types and definitions:
@@ -652,16 +662,6 @@ extern void perf_counter_fork(struct task_struct *tsk);
 
 extern void perf_counter_task_migration(struct task_struct *task, int cpu);
 
-#define MAX_STACK_DEPTH                        255
-
-struct perf_callchain_entry {
-       u16                             nr;
-       u16                             hv;
-       u16                             kernel;
-       u16                             user;
-       u64                             ip[MAX_STACK_DEPTH];
-};
-
 extern struct perf_callchain_entry *perf_callchain(struct pt_regs *regs);
 
 extern int sysctl_perf_counter_paranoid;
index 9868346..e14e986 100644 (file)
@@ -39,6 +39,8 @@ static int            dump_trace = 0;
 #define cdprintf(x...) do { if (dump_trace) color_fprintf(stdout, color, x); } while (0)
 
 static int             verbose;
+#define eprintf(x...)  do { if (verbose) fprintf(stderr, x); } while (0)
+
 static int             full_paths;
 
 static unsigned long   page_size;
@@ -47,14 +49,6 @@ static unsigned long mmap_window = 32;
 static char            *parent_pattern = "^sys_|^do_page_fault";
 static regex_t         parent_regex;
 
-struct ip_chain_event {
-       __u16 nr;
-       __u16 hv;
-       __u16 kernel;
-       __u16 user;
-       __u64 ips[];
-};
-
 struct ip_event {
        struct perf_event_header header;
        __u64 ip;
@@ -131,15 +125,11 @@ static struct dso *dsos__findnew(const char *name)
 
        nr = dso__load(dso, NULL, verbose);
        if (nr < 0) {
-               if (verbose)
-                       fprintf(stderr, "Failed to open: %s\n", name);
+               eprintf("Failed to open: %s\n", name);
                goto out_delete_dso;
        }
-       if (!nr && verbose) {
-               fprintf(stderr,
-               "No symbols found in: %s, maybe install a debug package?\n",
-                               name);
-       }
+       if (!nr)
+               eprintf("No symbols found in: %s, maybe install a debug package?\n", name);
 
        dsos__add(dso);
 
@@ -844,7 +834,7 @@ static struct symbol *call__match(struct symbol *sym)
 
 static int
 hist_entry__add(struct thread *thread, struct map *map, struct dso *dso,
-               struct symbol *sym, __u64 ip, struct ip_chain_event *chain,
+               struct symbol *sym, __u64 ip, struct perf_callchain_entry *chain,
                char level, __u64 count)
 {
        struct rb_node **p = &hist.rb_node;
@@ -868,7 +858,7 @@ hist_entry__add(struct thread *thread, struct map *map, struct dso *dso,
                __u64 ip;
 
                for (i = 0; i < chain->kernel; i++) {
-                       ip = chain->ips[nr + i];
+                       ip = chain->ip[nr + i];
                        dso = kernel_dso;
                        sym = resolve_symbol(thread, NULL, &dso, &ip);
                        entry.parent = call__match(sym);
@@ -878,7 +868,7 @@ hist_entry__add(struct thread *thread, struct map *map, struct dso *dso,
                nr += i;
 
                for (i = 0; i < chain->user; i++) {
-                       ip = chain->ips[nr + i];
+                       ip = chain->ip[nr + i];
                        sym = resolve_symbol(thread, NULL, NULL, &ip);
                        entry.parent = call__match(sym);
                        if (entry.parent)
@@ -1080,6 +1070,30 @@ static unsigned long total = 0,
                     total_fork = 0,
                     total_unknown = 0;
 
+static int validate_chain(struct perf_callchain_entry *chain, event_t *event)
+{
+       unsigned int chain_size;
+
+       if (chain->nr > MAX_STACK_DEPTH)
+               return -1;
+       if (chain->hv > MAX_STACK_DEPTH)
+               return -1;
+       if (chain->kernel > MAX_STACK_DEPTH)
+               return -1;
+       if (chain->user > MAX_STACK_DEPTH)
+               return -1;
+       if (chain->hv + chain->kernel + chain->user != chain->nr)
+               return -1;
+
+       chain_size = event->header.size;
+       chain_size -= (unsigned long)&event->ip.__more_data - (unsigned long)event;
+
+       if (chain->nr*sizeof(__u64) > chain_size)
+               return -1;
+
+       return 0;
+}
+
 static int
 process_overflow_event(event_t *event, unsigned long offset, unsigned long head)
 {
@@ -1091,7 +1105,7 @@ process_overflow_event(event_t *event, unsigned long offset, unsigned long head)
        __u64 period = 1;
        struct map *map = NULL;
        void *more_data = event->ip.__more_data;
-       struct ip_chain_event *chain = NULL;
+       struct perf_callchain_entry *chain = NULL;
 
        if (event->header.type & PERF_SAMPLE_PERIOD) {
                period = *(__u64 *)more_data;
@@ -1111,21 +1125,26 @@ process_overflow_event(event_t *event, unsigned long offset, unsigned long head)
 
                chain = (void *)more_data;
 
-               if (dump_trace) {
-                       dprintf("... chain: u:%d, k:%d, nr:%d\n",
-                               chain->user,
-                               chain->kernel,
-                               chain->nr);
+               dprintf("... chain: u:%d, k:%d, nr:%d\n",
+                       chain->user,
+                       chain->kernel,
+                       chain->nr);
 
+               if (validate_chain(chain, event) < 0) {
+                       eprintf("call-chain problem with event, skipping it.\n");
+                       return 0;
+               }
+
+               if (dump_trace) {
                        for (i = 0; i < chain->nr; i++)
-                               dprintf("..... %2d: %016Lx\n", i, chain->ips[i]);
+                               dprintf("..... %2d: %016Lx\n", i, chain->ip[i]);
                }
        }
 
        dprintf(" ... thread: %s:%d\n", thread->comm, thread->pid);
 
        if (thread == NULL) {
-               fprintf(stderr, "problem processing %d event, skipping it.\n",
+               eprintf("problem processing %d event, skipping it.\n",
                        event->header.type);
                return -1;
        }
@@ -1153,8 +1172,7 @@ process_overflow_event(event_t *event, unsigned long offset, unsigned long head)
                struct symbol *sym = resolve_symbol(thread, &map, &dso, &ip);
 
                if (hist_entry__add(thread, map, dso, sym, ip, chain, level, period)) {
-                       fprintf(stderr,
-               "problem incrementing symbol count, skipping event\n");
+                       eprintf("problem incrementing symbol count, skipping event\n");
                        return -1;
                }
        }