[sled-diagnostics] Add pfiles output (#7258)

This PR adds support for gathering `pfiles` output for all Oxide
processes on a sled.

This is on top of:
- #7228 
- #7194

openapi/sled-agent.json

@@ -754,6 +754,33 @@
         }
       }
     },
+    "/support/pfiles-info": {
+      "get": {
+        "operationId": "support_pfiles_info",
+        "responses": {
+          "200": {
+            "description": "successful operation",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "title": "Array_of_SledDiagnosticsQueryOutput",
+                  "type": "array",
+                  "items": {
+                    "$ref": "#/components/schemas/SledDiagnosticsQueryOutput"
+                  }
+                }
+              }
+            }
+          },
+          "4XX": {
+            "$ref": "#/components/responses/Error"
+          },
+          "5XX": {
+            "$ref": "#/components/responses/Error"
+          }
+        }
+      }
+    },
     "/support/pstack-info": {
       "get": {
         "operationId": "support_pstack_info",

sled-agent/api/src/lib.rs

@@ -664,6 +664,14 @@ pub trait SledAgentApi {
     async fn support_pstack_info(
         request_context: RequestContext<Self::Context>,
     ) -> Result<HttpResponseOk<Vec<SledDiagnosticsQueryOutput>>, HttpError>;
+
+    #[endpoint {
+        method = GET,
+        path = "/support/pfiles-info",
+    }]
+    async fn support_pfiles_info(
+        request_context: RequestContext<Self::Context>,
+    ) -> Result<HttpResponseOk<Vec<SledDiagnosticsQueryOutput>>, HttpError>;
 }
 
 #[derive(Clone, Debug, Deserialize, JsonSchema, Serialize)]

sled-agent/src/http_entrypoints.rs

@@ -1027,4 +1027,18 @@ impl SledAgentApi for SledAgentImpl {
                 .collect::<Vec<_>>(),
         ))
     }
+
+    async fn support_pfiles_info(
+        request_context: RequestContext<Self::Context>,
+    ) -> Result<HttpResponseOk<Vec<SledDiagnosticsQueryOutput>>, HttpError>
+    {
+        let sa = request_context.context();
+        Ok(HttpResponseOk(
+            sa.support_pfiles_info()
+                .await
+                .into_iter()
+                .map(|cmd| cmd.get_output())
+                .collect::<Vec<_>>(),
+        ))
+    }
 }

sled-agent/src/sim/http_entrypoints.rs

@@ -754,6 +754,13 @@ impl SledAgentApi for SledAgentSimImpl {
     {
         method_unimplemented()
     }
+
+    async fn support_pfiles_info(
+        _request_context: RequestContext<Self::Context>,
+    ) -> Result<HttpResponseOk<Vec<SledDiagnosticsQueryOutput>>, HttpError>
+    {
+        method_unimplemented()
+    }
 }
 
 fn method_unimplemented<T>() -> Result<T, HttpError> {

sled-agent/src/sled_agent.rs

@@ -1400,6 +1400,12 @@ impl SledAgent {
     ) -> Vec<Result<SledDiagnosticsCmdOutput, SledDiagnosticsCmdError>> {
         sled_diagnostics::pstack_oxide_processes(&self.log).await
     }
+
+    pub(crate) async fn support_pfiles_info(
+        &self,
+    ) -> Vec<Result<SledDiagnosticsCmdOutput, SledDiagnosticsCmdError>> {
+        sled_diagnostics::pfiles_oxide_processes(&self.log).await
+    }
 }
 
 #[derive(From, thiserror::Error, Debug)]

sled-diagnostics/src/lib.rs

@@ -102,3 +102,24 @@ pub async fn pstack_oxide_processes(
         .collect::<Vec<Result<_, _>>>()
         .await
 }
+
+pub async fn pfiles_oxide_processes(
+    log: &Logger,
+) -> Vec<Result<SledDiagnosticsCmdOutput, SledDiagnosticsCmdError>> {
+    // In a diagnostics context we care about looping over every pid we find,
+    // but on failure we should just return a single error in a vec that
+    // represents the entire failed operation.
+    let pids = match contract::find_oxide_pids(log) {
+        Ok(pids) => pids,
+        Err(e) => return vec![Err(e.into())],
+    };
+
+    pids.iter()
+        .map(|pid| pfiles_process(*pid))
+        .map(|c| async move {
+            execute_command_with_timeout(c, DEFAULT_TIMEOUT).await
+        })
+        .collect::<FuturesUnordered<_>>()
+        .collect::<Vec<Result<_, _>>>()
+        .await
+}

sled-diagnostics/src/queries.rs

@@ -23,6 +23,7 @@ use crate::contract_stub::ContractError;
 const DLADM: &str = "/usr/sbin/dladm";
 const IPADM: &str = "/usr/sbin/ipadm";
 const PFEXEC: &str = "/usr/bin/pfexec";
+const PFILES: &str = "/usr/bin/pfiles";
 const PSTACK: &str = "/usr/bin/pstack";
 const PARGS: &str = "/usr/bin/pargs";
 const ZONEADM: &str = "/usr/sbin/zoneadm";
@@ -247,6 +248,12 @@ pub fn pstack_process(pid: i32) -> Command {
     cmd
 }
 
+pub fn pfiles_process(pid: i32) -> Command {
+    let mut cmd = std::process::Command::new(PFEXEC);
+    cmd.env_clear().arg(PFILES).arg(pid.to_string());
+    cmd
+}
+
 #[cfg(test)]
 mod test {
     use super::*;