Merge pull request #77938 from andrurogerz/swift-inspect-linux

[swift-inspect] implement basic GNU/Linux support

Author: compnerd

tools/swift-inspect/Package.swift

@@ -19,9 +19,19 @@ let package = Package(
                 .product(name: "ArgumentParser", package: "swift-argument-parser"),
                 .target(name: "SwiftInspectClient", condition: .when(platforms: [.windows])),
                 .target(name: "SwiftInspectClientInterface", condition: .when(platforms: [.windows])),
+                .target(name: "SwiftInspectLinux", condition: .when(platforms: [.linux])),
             ],
             swiftSettings: [.unsafeFlags(["-parse-as-library"])]),
         .target(name: "SwiftInspectClient"),
+        .target(
+            name: "SwiftInspectLinux",
+            dependencies: ["LinuxSystemHeaders"],
+            path: "Sources/SwiftInspectLinux",
+            exclude: ["SystemHeaders"],
+            cSettings: [.define("_GNU_SOURCE", to: "1")]),
+        .systemLibrary(
+            name: "LinuxSystemHeaders",
+            path: "Sources/SwiftInspectLinux/SystemHeaders"),
         .systemLibrary(
             name: "SwiftInspectClientInterface"),
         .testTarget(

tools/swift-inspect/README.md

@@ -16,6 +16,14 @@ In order to build on Windows, some additional parameters must be passed to the b
 swift build -Xcc -I%SDKROOT%\usr\include\swift\SwiftRemoteMirror -Xlinker %SDKROOT%\usr\lib\swift\windows\x86_64\swiftRemoteMirror.lib
 ~~~
 
+#### Linux
+
+In order to build on Linux, some additional parameters must be passed to the build tool to locate the necessary includes and libraries.
+
+~~~
+swift build -Xswiftc -I$(git rev-parse --show-toplevel)/include/swift/SwiftRemoteMirror -Xlinker -lswiftRemoteMirror
+~~~
+
 #### CMake
 
 In order to build on Windows with CMake, some additional parameters must be passed to the build tool to locate the necessary Swift modules.
@@ -30,9 +38,6 @@ The following inspection operations are available currently.
 
 ##### All Platforms
 
-dump-arrays <name-or-pid>
-: Print information about array objects in the target
-
 dump-cache-nodes <name-or-pid>
 : Print the metadata cache nodes.
 
@@ -45,6 +50,11 @@ dump-generic-metadata <name-or-pid> [--backtrace] [--backtrace-long]
 dump-raw-metadata <name-or-pid> [--backtrace] [--backtrace-long]
 : Print metadata allocations.
 
+#### Darwin and Windows Only
+
+dump-arrays <name-or-pid>
+: Print information about array objects in the target
+
 ##### Darwin Only
 
 dump-concurrency <name-or-pid>

tools/swift-inspect/Sources/SwiftInspectLinux/ElfFile.swift

@@ -0,0 +1,142 @@
+//===----------------------------------------------------------------------===//
+//
+// This source file is part of the Swift.org open source project
+//
+// Copyright (c) 2014 - 2024 Apple Inc. and the Swift project authors
+// Licensed under Apache License v2.0 with Runtime Library Exception
+//
+// See https://swift.org/LICENSE.txt for license information
+// See https://swift.org/CONTRIBUTORS.txt for the list of Swift project authors
+//
+//===----------------------------------------------------------------------===//
+
+import Foundation
+import LinuxSystemHeaders
+
+// TODO: replace this implementation with general purpose ELF parsing support
+// currently private to swift/stdlib/public/Backtrace.
+class ElfFile {
+  public enum ELFError: Error {
+    case notELF64(_ filePath: String, _ description: String = "")
+    case malformedFile(_ filePath: String, _ description: String = "")
+  }
+
+  public typealias SymbolMap = [String: (start: UInt64, end: UInt64)]
+
+  let filePath: String
+  let fileData: Data
+  let ehdr: Elf64_Ehdr
+
+  public init(filePath: String) throws {
+    self.filePath = filePath
+
+    self.fileData = try Data(contentsOf: URL(fileURLWithPath: filePath), options: .alwaysMapped)
+
+    let ident = fileData.prefix(upTo: Int(EI_NIDENT))
+
+    guard String(bytes: ident.prefix(Int(SELFMAG)), encoding: .utf8) == ELFMAG else {
+      throw ELFError.notELF64(filePath, "\(ident.prefix(Int(SELFMAG))) != ELFMAG")
+    }
+
+    guard ident[Int(EI_CLASS)] == ELFCLASS64 else {
+      throw ELFError.notELF64(filePath, "\(ident[Int(EI_CLASS)]) != ELFCLASS64")
+    }
+
+    let ehdrSize = MemoryLayout<Elf64_Ehdr>.size
+    self.ehdr = fileData[0..<ehdrSize].withUnsafeBytes { $0.load(as: Elf64_Ehdr.self) }
+  }
+
+  // returns a map of symbol names to their offset range in file (+ baseAddress)
+  public func loadSymbols(baseAddress: UInt64 = 0) throws -> SymbolMap {
+    guard let sectionCount = UInt(exactly: self.ehdr.e_shnum) else {
+      throw ELFError.malformedFile(
+        self.filePath, "invalid Elf64_Ehdr.e_shnum: \(self.ehdr.e_shnum)")
+    }
+
+    var symbols: SymbolMap = [:]
+    for sectionIndex in 0..<sectionCount {
+      let shdr: Elf64_Shdr = try self.loadShdr(index: sectionIndex)
+      guard shdr.sh_type == SHT_SYMTAB || shdr.sh_type == SHT_DYNSYM else { continue }
+
+      let symTableData: Data = try self.loadSection(shdr)
+      let symTable: [Elf64_Sym] = symTableData.withUnsafeBytes {
+        Array($0.bindMemory(to: Elf64_Sym.self))
+      }
+
+      guard shdr.sh_entsize == MemoryLayout<Elf64_Sym>.size else {
+        throw ELFError.malformedFile(self.filePath, "invalid Elf64_Shdr.sh_entsize")
+      }
+
+      // the link field in the section header for a symbol table section refers
+      // to the index of the string table section containing the symbol names
+      guard let linkIndex = UInt(exactly: shdr.sh_link) else {
+        throw ELFError.malformedFile(self.filePath, "invalid Elf64_Shdr.sh_link: \(shdr.sh_link)")
+      }
+
+      let shdrLink: Elf64_Shdr = try self.loadShdr(index: UInt(linkIndex))
+      guard shdrLink.sh_type == SHT_STRTAB else {
+        throw ELFError.malformedFile(self.filePath, "linked section not SHT_STRTAB")
+      }
+
+      // load the entire contents of the string table into memory
+      let strTableData: Data = try self.loadSection(shdrLink)
+      let strTable: [UInt8] = strTableData.withUnsafeBytes {
+        Array($0.bindMemory(to: UInt8.self))
+      }
+
+      let symCount = Int(shdr.sh_size / shdr.sh_entsize)
+      for symIndex in 0..<symCount {
+        let sym = symTable[symIndex]
+        guard sym.st_shndx != SHN_UNDEF, sym.st_value != 0, sym.st_size != 0 else { continue }
+
+        // sym.st_name is a byte offset into the string table
+        guard let strStart = Int(exactly: sym.st_name), strStart < strTable.count else {
+          throw ELFError.malformedFile(self.filePath, "invalid string table offset: \(sym.st_name)")
+        }
+
+        guard let strEnd = strTable[strStart...].firstIndex(of: 0),
+          let symName = String(bytes: strTable[strStart..<strEnd], encoding: .utf8)
+        else {
+          throw ELFError.malformedFile(self.filePath, "invalid string @ offset \(strStart)")
+        }
+
+        // rebase the symbol value on the base address provided by the caller
+        let symStart = sym.st_value + baseAddress
+        symbols[symName] = (start: symStart, end: symStart + sym.st_size)
+      }
+    }
+
+    return symbols
+  }
+
+  // returns the Elf64_Shdr at the specified index
+  internal func loadShdr(index: UInt) throws -> Elf64_Shdr {
+    guard index < self.ehdr.e_shnum else {
+      throw ELFError.malformedFile(
+        self.filePath, "section index \(index) >= Elf64_Ehdr.e_shnum \(self.ehdr.e_shnum))")
+    }
+
+    let shdrSize = MemoryLayout<Elf64_Shdr>.size
+    guard shdrSize == self.ehdr.e_shentsize else {
+      throw ELFError.malformedFile(self.filePath, "Elf64_Ehdr.e_shentsize != \(shdrSize)")
+    }
+
+    let shdrOffset = Int(self.ehdr.e_shoff) + Int(index) * shdrSize
+    let shdrData = self.fileData[shdrOffset..<(shdrOffset + shdrSize)]
+    return shdrData.withUnsafeBytes { $0.load(as: Elf64_Shdr.self) }
+  }
+
+  // returns all data in the specified section
+  internal func loadSection(_ shdr: Elf64_Shdr) throws -> Data {
+    guard let sectionSize = Int(exactly: shdr.sh_size) else {
+      throw ELFError.malformedFile(self.filePath, "Elf64_Shdr.sh_size too large \(shdr.sh_size)")
+    }
+
+    guard let fileOffset = Int(exactly: shdr.sh_offset) else {
+      throw ELFError.malformedFile(
+        self.filePath, "Elf64_Shdr.sh_offset too large \(shdr.sh_offset)")
+    }
+
+    return self.fileData[fileOffset..<(fileOffset + sectionSize)]
+  }
+}

tools/swift-inspect/Sources/SwiftInspectLinux/LinkMap.swift

@@ -0,0 +1,144 @@
+//===----------------------------------------------------------------------===//
+//
+// This source file is part of the Swift.org open source project
+//
+// Copyright (c) 2014 - 2024 Apple Inc. and the Swift project authors
+// Licensed under Apache License v2.0 with Runtime Library Exception
+//
+// See https://swift.org/LICENSE.txt for license information
+// See https://swift.org/CONTRIBUTORS.txt for the list of Swift project authors
+//
+//===----------------------------------------------------------------------===//
+
+import Foundation
+import LinuxSystemHeaders
+
+class LinkMap {
+  public enum LinkMapError: Error {
+    case failedLoadingAuxVec(for: pid_t)
+    case missingAuxVecEntry(for: pid_t, _ tag: Int32)
+    case malformedELF(for: pid_t, _ description: String)
+  }
+
+  public struct Entry {
+    let baseAddress: UInt64
+    let moduleName: String
+  }
+
+  public let entries: [Entry]
+
+  public init(for process: Process) throws {
+    let auxVec = try Self.loadAuxVec(for: process.pid)
+    guard let phdrAddr = auxVec[AT_PHDR] else {
+      throw LinkMapError.missingAuxVecEntry(for: process.pid, AT_PHDR)
+    }
+
+    guard let phdrSize = auxVec[AT_PHENT] else {
+      throw LinkMapError.missingAuxVecEntry(for: process.pid, AT_PHENT)
+    }
+
+    guard let phdrCount = auxVec[AT_PHNUM] else {
+      throw LinkMapError.missingAuxVecEntry(for: process.pid, AT_PHNUM)
+    }
+
+    guard phdrSize == MemoryLayout<Elf64_Phdr>.size else {
+      throw LinkMapError.malformedELF(for: process.pid, "AT_PHENT invalid size: \(phdrSize)")
+    }
+
+    // determine the base load address for the executable file and locate the
+    // dynamic segment
+    var dynamicSegment: Elf64_Phdr? = nil
+    var baseLoadSegment: Elf64_Phdr? = nil
+    for i in 0...phdrCount {
+      let address: UInt64 = phdrAddr + i * phdrSize
+      let phdr: Elf64_Phdr = try process.readStruct(address: address)
+
+      switch phdr.p_type {
+      case UInt32(PT_LOAD):
+        // chose the PT_LOAD segment with the lowest p_vaddr value, which will
+        // typically be zero
+        if let loadSegment = baseLoadSegment {
+          if phdr.p_vaddr < loadSegment.p_vaddr { baseLoadSegment = phdr }
+        } else {
+          baseLoadSegment = phdr
+        }
+
+      case UInt32(PT_DYNAMIC):
+        guard dynamicSegment == nil else {
+          throw LinkMapError.malformedELF(for: process.pid, "multiple PT_DYNAMIC segments found")
+        }
+        dynamicSegment = phdr
+
+      default: continue
+      }
+    }
+
+    guard let dynamicSegment = dynamicSegment else {
+      throw LinkMapError.malformedELF(for: process.pid, "PT_DYNAMIC segment not found")
+    }
+
+    guard let baseLoadSegment = baseLoadSegment else {
+      throw LinkMapError.malformedELF(for: process.pid, "PT_LOAD segment not found")
+    }
+
+    let ehdrSize = MemoryLayout<Elf64_Ehdr>.size
+    let loadAddr: UInt64 = phdrAddr - UInt64(ehdrSize)
+    let baseAddr: UInt64 = loadAddr - baseLoadSegment.p_vaddr
+    let dynamicSegmentAddr: UInt64 = baseAddr + dynamicSegment.p_vaddr
+
+    // parse through the dynamic segment to find the location of the .debug section
+    var rDebugEntry: Elf64_Dyn? = nil
+    let entrySize = MemoryLayout<Elf64_Dyn>.size
+    let dynamicEntryCount = UInt(dynamicSegment.p_memsz / UInt64(entrySize))
+    for i in 0...dynamicEntryCount {
+      let address: UInt64 = dynamicSegmentAddr + UInt64(i) * UInt64(entrySize)
+      let dyn: Elf64_Dyn = try process.readStruct(address: address)
+      if dyn.d_tag == DT_DEBUG {
+        rDebugEntry = dyn
+        break
+      }
+    }
+
+    guard let rDebugEntry = rDebugEntry else {
+      throw LinkMapError.malformedELF(for: process.pid, "DT_DEBUG not found in dynamic segment")
+    }
+
+    let rDebugAddr: UInt64 = rDebugEntry.d_un.d_val
+    let rDebug: r_debug = try process.readStruct(address: rDebugAddr)
+
+    var entries: [Entry] = []
+    var linkMapAddr = UInt(bitPattern: rDebug.r_map)
+    while linkMapAddr != 0 {
+      let linkMap: link_map = try process.readStruct(address: UInt64(linkMapAddr))
+      let nameAddr = UInt(bitPattern: linkMap.l_name)
+      let name = try process.readString(address: UInt64(nameAddr))
+      entries.append(Entry(baseAddress: linkMap.l_addr, moduleName: name))
+
+      linkMapAddr = UInt(bitPattern: linkMap.l_next)
+    }
+
+    self.entries = entries
+  }
+
+  // loads the auxiliary vector for a 64-bit process
+  static func loadAuxVec(for pid: pid_t) throws -> [Int32: UInt64] {
+    guard let data = ProcFS.loadFile(for: pid, "auxv") else {
+      throw LinkMapError.failedLoadingAuxVec(for: pid)
+    }
+
+    return data.withUnsafeBytes {
+      // in a 64-bit process, aux vector is an array of 8-byte pairs
+      let count = $0.count / MemoryLayout<(UInt64, UInt64)>.stride
+      let auxVec = Array($0.bindMemory(to: (UInt64, UInt64).self)[..<count])
+
+      var entries: [Int32: UInt64] = [:]
+      for (rawTag, value) in auxVec {
+        // the AT_ constants defined in linux/auxv.h are imported as Int32
+        guard let tag = Int32(exactly: rawTag) else { continue }
+        entries[tag] = UInt64(value)
+      }
+
+      return entries
+    }
+  }
+}