Repo sync

Cargo.toml

@@ -26,7 +26,7 @@ name = "aperf"
 path = "src/bin/aperf.rs"
 
 [dependencies]
-nix = { version = "0.29.0", features = ["signal", "poll"] }
+nix = { version = "0.29.0", features = ["signal", "poll", "user"] }
 clap = { version = "4.2.5", features = ["derive"] }
 clap_complete = "4.2.3"
 serde = { version = "1.0", features = ["derive"] }

src/data/perf_profile.rs

@@ -9,8 +9,8 @@ use std::{process::Child, sync::Mutex};
 use {
     crate::data::{CollectData, CollectorParams},
     crate::PDError,
-    log::{debug, error},
-    nix::{sys::signal, unistd::Pid},
+    log::{debug, error, warn},
+    nix::{sys::signal, unistd, unistd::Pid},
     std::io::Write,
     std::process::{Command, Stdio},
 };
@@ -38,27 +38,38 @@ impl PerfProfileRaw {
 #[cfg(target_os = "linux")]
 impl CollectData for PerfProfileRaw {
     fn prepare_data_collector(&mut self, params: &CollectorParams) -> Result<()> {
-        // Check kernel.perf_event_paranoid
-        let paranoid_value = fs::read_to_string("/proc/sys/kernel/perf_event_paranoid")?
-            .trim()
-            .parse::<i32>()
-            .unwrap_or(4);
-        if paranoid_value != -1 {
-            return Err(PDError::DependencyError(format!(
-                "kernel.perf_event_paranoid is not -1. Run `sudo sysctl -w kernel.perf_event_paranoid=-1`"
-            )).into());
-        }
+        let is_root = unistd::geteuid().is_root();
+
+        // Check kernel configs for the following perf command
+        if !is_root {
+            //  Check kernel.perf_event_paranoid
+            // -1: Allow use of almost all events by all users (mmap perf_event_open for hotline)
+            //  0: Allow CPU event Data
+            //          Disallow ftrace function tracepoint by users without CAP_SYS_ADMIN
+            //          Disallow raw tracepoint access by users without CAP_SYS_ADMIN
+            //  1: Allow Kernel Profiling (perf will fail, but generate 0 byte perf.data)
+            //          Disallow CPU event access by users without CAP_SYS_ADMIN
+            //  2: Disallow everything (perf will fail, but generate 0 byte perf.data)
+            //          Disallow kernel profiling by users without CAP_SYS_ADMIN
+            let paranoid_value = fs::read_to_string("/proc/sys/kernel/perf_event_paranoid")?
+                .trim()
+                .parse::<i32>()
+                .unwrap_or(4);
+            if paranoid_value > 0 {
+                warn!("kernel.perf_event_paranoid is not <=0, which disallows access to CPU event data. Run `sudo sysctl -w kernel.perf_event_paranoid=-1`");
+            }
 
-        // Check kernel.kptr_restrict
-        let kptr_value = fs::read_to_string("/proc/sys/kernel/kptr_restrict")?
-            .trim()
-            .parse::<i32>()
-            .unwrap_or(-1);
-        if kptr_value != 0 {
-            return Err(PDError::DependencyError(format!(
-                "kernel.kptr_restrict is not 0. Run `sudo sysctl -w kernel.kptr_restrict=0`"
-            ))
-            .into());
+            //  Check kernel.kptr_restrict
+            //  0: the address is hashed before printing. (This is the equivalent to %p.)
+            //  1: (symbols skewed if not root) kernel pointers replaced with 0's unless the user has CAP_SYSLOG.
+            //  2: (symbols may be skewed, perf may fail even with root on older perf) kernel pointers replaced with 0's regardless of privileges.
+            let kptr_value = fs::read_to_string("/proc/sys/kernel/kptr_restrict")?
+                .trim()
+                .parse::<i32>()
+                .unwrap_or(-1);
+            if kptr_value > 0 {
+                warn!("kernel.kptr_restrict is not 0, which may result in missing kernel symbols. Run `sudo sysctl -w kernel.kptr_restrict=0`");
+            }
         }
 
         match Command::new("perf")

src/data/perf_stat.rs

@@ -244,7 +244,7 @@ impl CollectData for PerfStatRaw {
                     Err(e) => {
                         match e.kind() {
                             ErrorKind::PermissionDenied => {
-                                warn!("kernel.perf_event_paranoid needs to be -1. Run `sudo sysctl -w kernel.perf_event_paranoid=-1`")
+                                warn!("kernel.perf_event_paranoid is not <=0. Run `sudo sysctl -w kernel.perf_event_paranoid=-1`")
                             }
                             ErrorKind::NotFound => warn!("PMU counters not available on this instance type. Refer to APerf documentation for supported instances"),
                             _ => match e.raw_os_error().unwrap() {