Linux: gather permitted capabilities via capget(2)

#1211 has shown reading /proc/<pid>/status might have a significant
performance impact.  It was started to be read by default to gather the
permitted capabilities of the process.

Gather permitted capabilities via the syscall capget(2) instead.
cap_get_proc(3) is not used to avoid linking with -lcap.

3 files changed, 24 insertions, 10 deletions

diff --git a/Process.c b/Process.c
index c343e77d..527fc201 100644
--- a/Process.c
+++ b/Process.c
@@ -736,7 +736,7 @@ void Process_writeField(const Process* this, RichString* str, RowField field) {
       }
       break;
    case USER:
-      if (this->elevated_priv)
+      if (this->elevated_priv == TRI_ON)
          attr = CRT_colors[PROCESS_PRIV];
       else if (host->htopUserId != this->st_uid)
          attr = CRT_colors[PROCESS_SHADOW];
diff --git a/Process.h b/Process.h
index 6a80bc51..066c1402 100644
--- a/Process.h
+++ b/Process.h
@@ -24,6 +24,13 @@ in the source distribution for its full text.
 
 #define DEFAULT_HIGHLIGHT_SECS 5
 
+typedef enum Tristate_ {
+   TRI_INITIAL = 0,
+   TRI_OFF = -1,
+   TRI_ON = 1,
+} Tristate;
+
+
 /* Core process states (shared by platforms)
  * NOTE: The enum has an ordering that is important!
  * See processStateChar in process.c for ProcessSate -> letter mapping */
@@ -106,7 +113,7 @@ typedef struct Process_ {
     *   - from file capabilities
     *   - inherited from the ambient set
     */
-   bool elevated_priv;
+   Tristate elevated_priv;
 
    /* Process runtime (in hundredth of a second) */
    unsigned long long int time;
diff --git a/linux/LinuxProcessTable.c b/linux/LinuxProcessTable.c
index 9a5ffdb4..57f93227 100644
--- a/linux/LinuxProcessTable.c
+++ b/linux/LinuxProcessTable.c
@@ -20,7 +20,9 @@ in the source distribution for its full text.
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
+#include <syscall.h>
 #include <unistd.h>
+#include <linux/capability.h> // raw syscall, no libcap  // IWYU pragma: keep // IWYU pragma: no_include <sys/capability.h>
 #include <sys/stat.h>
 
 #include "Compat.h"
@@ -391,14 +393,6 @@ static bool LinuxProcessTable_readStatusFile(Process* process, openat_arg_t proc
          if (pid_ns_count > 1)
             process->isRunningInContainer = true;
 
-      } else if (String_startsWith(buffer, "CapPrm:")) {
-         char* ptr = buffer + strlen("CapPrm:");
-         while (*ptr == ' ' || *ptr == '\t')
-            ptr++;
-
-         uint64_t cap_permitted = fast_strtoull_hex(&ptr, 16);
-         process->elevated_priv = cap_permitted != 0 && process->st_uid != 0;
-
       } else if (String_startsWith(buffer, "voluntary_ctxt_switches:")) {
          unsigned long vctxt;
          int ok = sscanf(buffer, "voluntary_ctxt_switches:\t%lu", &vctxt);
@@ -1481,6 +1475,19 @@ static bool LinuxProcessTable_recurseProcTree(LinuxProcessTable* this, openat_ar
          }
       }
 
+      /* Gather permitted capabilities (thread-specific data) for non-root process. */
+      if (proc->st_uid != 0 && proc->elevated_priv != TRI_OFF) {
+         struct __user_cap_header_struct header = { .version = _LINUX_CAPABILITY_VERSION_3, .pid = Process_getPid(proc) };
+         struct __user_cap_data_struct data;
+
+         long res = syscall(SYS_capget, &header, &data);
+         if (res == 0) {
+            proc->elevated_priv = (data.permitted != 0) ? TRI_ON : TRI_OFF;
+         } else {
+            proc->elevated_priv = TRI_OFF;
+         }
+      }
+
       if (ss->flags & PROCESS_FLAG_LINUX_CGROUP)
          LinuxProcessTable_readCGroupFile(lp, procFd);