[profile] Implement a non-mmap path when reading profile files from a non-local filesystem

Author: Wael Yehia

compiler-rt/lib/profile/InstrProfilingFile.c

@@ -428,17 +428,18 @@ static int getProfileFileSizeForMerging(FILE *ProfileFile,
  * \p ProfileBuffer. Returns -1 on failure. On success, the caller is
  * responsible for unmapping the mmap'd buffer in \p ProfileBuffer. */
 static int mmapProfileForMerging(FILE *ProfileFile, uint64_t ProfileFileSize,
-                                 char **ProfileBuffer) {
-  *ProfileBuffer = mmap(NULL, ProfileFileSize, PROT_READ, MAP_SHARED | MAP_FILE,
-                        fileno(ProfileFile), 0);
-  if (*ProfileBuffer == MAP_FAILED) {
+                                 ManagedMemory *ProfileBuffer) {
+  lprofGetFileContentBuffer(ProfileFile, ProfileFileSize, ProfileBuffer);
+
+  if (ProfileBuffer->Status == MM_INVALID) {
     PROF_ERR("Unable to merge profile data, mmap failed: %s\n",
              strerror(errno));
     return -1;
   }
 
-  if (__llvm_profile_check_compatibility(*ProfileBuffer, ProfileFileSize)) {
-    (void)munmap(*ProfileBuffer, ProfileFileSize);
+  if (__llvm_profile_check_compatibility(ProfileBuffer->Addr,
+                                         ProfileFileSize)) {
+    (void)lprofReleaseBuffer(ProfileBuffer, ProfileFileSize);
     PROF_WARN("Unable to merge profile data: %s\n",
               "source profile file is not compatible.");
     return -1;
@@ -453,7 +454,7 @@ static int mmapProfileForMerging(FILE *ProfileFile, uint64_t ProfileFileSize,
 */
 static int doProfileMerging(FILE *ProfileFile, int *MergeDone) {
   uint64_t ProfileFileSize;
-  char *ProfileBuffer;
+  ManagedMemory ProfileBuffer;
 
   /* Get the size of the profile on disk. */
   if (getProfileFileSizeForMerging(ProfileFile, &ProfileFileSize) == -1)
@@ -469,9 +470,9 @@ static int doProfileMerging(FILE *ProfileFile, int *MergeDone) {
     return -1;
 
   /* Now start merging */
-  if (__llvm_profile_merge_from_buffer(ProfileBuffer, ProfileFileSize)) {
+  if (__llvm_profile_merge_from_buffer(ProfileBuffer.Addr, ProfileFileSize)) {
     PROF_ERR("%s\n", "Invalid profile data to merge");
-    (void)munmap(ProfileBuffer, ProfileFileSize);
+    (void)lprofReleaseBuffer(&ProfileBuffer, ProfileFileSize);
     return -1;
   }
 
@@ -480,7 +481,7 @@ static int doProfileMerging(FILE *ProfileFile, int *MergeDone) {
   (void)COMPILER_RT_FTRUNCATE(ProfileFile,
                               __llvm_profile_get_size_for_buffer());
 
-  (void)munmap(ProfileBuffer, ProfileFileSize);
+  (void)lprofReleaseBuffer(&ProfileBuffer, ProfileFileSize);
   *MergeDone = 1;
 
   return 0;
@@ -702,13 +703,13 @@ static void initializeProfileForContinuousMode(void) {
     } else {
       /* The merged profile has a non-zero length. Check that it is compatible
        * with the data in this process. */
-      char *ProfileBuffer;
+      ManagedMemory ProfileBuffer;
       if (mmapProfileForMerging(File, ProfileFileSize, &ProfileBuffer) == -1) {
         lprofUnlockFileHandle(File);
         fclose(File);
         return;
       }
-      (void)munmap(ProfileBuffer, ProfileFileSize);
+      (void)lprofReleaseBuffer(&ProfileBuffer, ProfileFileSize);
     }
   } else {
     File = fopen(Filename, FileOpenMode);
@@ -1346,12 +1347,12 @@ COMPILER_RT_VISIBILITY int __llvm_profile_set_file_object(FILE *File,
     } else {
       /* The merged profile has a non-zero length. Check that it is compatible
        * with the data in this process. */
-      char *ProfileBuffer;
+      ManagedMemory ProfileBuffer;
       if (mmapProfileForMerging(File, ProfileFileSize, &ProfileBuffer) == -1) {
         lprofUnlockFileHandle(File);
         return 1;
       }
-      (void)munmap(ProfileBuffer, ProfileFileSize);
+      (void)lprofReleaseBuffer(&ProfileBuffer, ProfileFileSize);
     }
     mmapForContinuousMode(0, File);
     lprofUnlockFileHandle(File);

compiler-rt/lib/profile/InstrProfilingUtil.c

@@ -21,6 +21,15 @@
 #include <unistd.h>
 #endif
 
+#ifdef _AIX
+#include <sys/statfs.h>
+// <sys/vmount.h> depends on `uint` to be a typedef from <sys/types.h> to
+// `uint_t`; however, <sys/types.h> does not always declare `uint`. We provide
+// the typedef prior to including <sys/vmount.h> to work around this issue.
+typedef uint_t uint;
+#include <sys/vmount.h>
+#endif
+
 #ifdef COMPILER_RT_HAS_UNAME
 #include <sys/utsname.h>
 #endif
@@ -258,6 +267,113 @@ COMPILER_RT_VISIBILITY FILE *lprofOpenFileEx(const char *ProfileName) {
   return f;
 }
 
+// Return 1 (true) if the file descriptor Fd represents a file that is on a
+// local filesystem, otherwise return 0.
+static int is_local_filesystem(int Fd) {
+#if defined(_AIX)
+  struct statfs Vfs;
+  if (fstatfs(Fd, &Vfs) != 0) {
+    PROF_ERR("%s: fstatfs(%d) failed: %s\n", __func__, Fd, strerror(errno));
+    return 0;
+  }
+
+  int Ret;
+  size_t BufSize = 2048u;
+  char *Buf;
+  int Tries = 3;
+  while (Tries--) {
+    Buf = malloc(BufSize);
+    Ret = mntctl(MCTL_QUERY, BufSize, Buf);
+    if (Ret != 0)
+      break;
+    BufSize = *(unsigned int *)Buf;
+    free(Buf);
+  }
+
+  if (Ret != -1) {
+    // Look for the correct vmount entry.
+    char *CurObjPtr = Buf;
+    while (Ret--) {
+      struct vmount *Vp = (struct vmount *)CurObjPtr;
+      _Static_assert(sizeof(Vfs.f_fsid) == sizeof(Vp->vmt_fsid),
+                     "fsid length mismatch");
+      if (memcmp(&Vfs.f_fsid, &Vp->vmt_fsid, sizeof Vfs.f_fsid) == 0) {
+        free(Buf);
+        return (Vp->vmt_flags & MNT_REMOTE) == 0;
+      }
+      CurObjPtr += Vp->vmt_length;
+    }
+  }
+
+  free(Buf);
+  // There was an error in mntctl or vmount entry not found; "remote" is the
+  // conservative answer.
+#endif
+  return 0;
+}
+
+static int isMmapSafe(int Fd) {
+  if (getenv("LLVM_NO_MMAP")) // For testing purposes.
+    return 0;
+  (void)&is_local_filesystem; // a fake reference to satisfy -Wunused-function
+#ifdef _AIX
+  return is_local_filesystem(Fd);
+#endif
+  return 1;
+}
+
+COMPILER_RT_VISIBILITY void lprofGetFileContentBuffer(FILE *F, uint64_t Length,
+                                                      ManagedMemory *Buf) {
+  Buf->Status = MM_INVALID;
+
+  if (!F || isMmapSafe(fileno(F))) {
+    Buf->Addr = mmap(NULL, Length, PROT_READ, MAP_SHARED | MAP_FILE,
+                     F ? fileno(F) : -1, 0);
+    if (Buf->Addr != MAP_FAILED)
+      Buf->Status = MM_MMAP;
+    return;
+  }
+
+  if (getenv("LLVM_PROFILE_VERBOSE"))
+    PROF_NOTE("Could not use mmap; using fread instead.%s\n", "");
+
+  void *Buffer = malloc(Length);
+
+  if (!Buffer) {
+    PROF_ERR("%s: malloc failed: %s\n", __func__, strerror(errno));
+    return;
+  }
+  if (fseek(F, 0L, SEEK_SET) != 0) {
+    PROF_ERR("%s: fseek(0, SEEK_SET) failed: %s\n", __func__, strerror(errno));
+    return;
+  }
+
+  // Read the entire file into memory.
+  size_t BytesRead = fread(Buffer, 1, Length, F);
+  if (BytesRead != (size_t)Length || ferror(F)) {
+    PROF_ERR("%s: fread failed: %s\n", __func__, strerror(errno));
+    return;
+  }
+
+  // Reading was successful, record the result in the Buf parameter.
+  Buf->Addr = Buffer;
+  Buf->Status = MM_MALLOC;
+}
+
+void lprofReleaseBuffer(ManagedMemory *Buf, size_t Length) {
+  switch (Buf->Status) {
+  case MM_MALLOC:
+    free(Buf->Addr);
+    break;
+  case MM_MMAP:
+    munmap(Buf->Addr, Length);
+    break;
+  case MM_INVALID:
+    PROF_ERR("%s: Buffer has invalid state: %d", __func__, Buf->Status);
+    break;
+  }
+}
+
 COMPILER_RT_VISIBILITY const char *lprofGetPathPrefix(int *PrefixStrip,
                                                       size_t *PrefixLen) {
   const char *Prefix = getenv("GCOV_PREFIX");

compiler-rt/lib/profile/InstrProfilingUtil.h

@@ -31,6 +31,25 @@ int lprofUnlockFileHandle(FILE *F);
  * lock for exclusive access. The caller will block
  * if the lock is already held by another process. */
 FILE *lprofOpenFileEx(const char *Filename);
+
+enum MemoryStatus {
+  MM_INVALID = 0x1, // Addr is not a valid address
+  MM_MMAP = 0x2,    // Addr was mmap'ed
+  MM_MALLOC = 0x4   // Addr was malloc'ed
+};
+typedef struct {
+  void *Addr;
+  enum MemoryStatus Status;
+} ManagedMemory;
+
+/* Read the content of a file using mmap or fread into a buffer.
+ * Certain files (e.g. NFS mounted) cannot be opened reliably with mmap,
+ * so we use fread in those cases. The corresponding lprofReleaseBuffer
+ * will free/munmap the buffer.
+ */
+void lprofGetFileContentBuffer(FILE *F, uint64_t FileSize, ManagedMemory *Buf);
+void lprofReleaseBuffer(ManagedMemory *FileBuffer, size_t Length);
+
 /* PS4 doesn't have setenv/getenv/fork. Define a shim. */
 #if __ORBIS__
 #include <sys/types.h>

compiler-rt/test/profile/Posix/instrprof-fork.c

@@ -3,11 +3,15 @@
 // RUN: %clang_pgogen=%t.profdir -o %t -O2 %s
 // RUN: %run %t
 // RUN: llvm-profdata show --all-functions --counts %t.profdir/default_*.profraw  | FileCheck %s
+// RUN: rm -fr %t.profdir
+// RUN: env LLVM_NO_MMAP=1 %run %t
+// RUN: llvm-profdata show --all-functions --counts %t.profdir/default_*.profraw  | FileCheck %s
+
 //
 // CHECK: func1:
-// CHECK: Block counts: [4]
+// CHECK: Block counts: [21]
 // CHECK:  func2:
-// CHECK: Block counts: [1]
+// CHECK: Block counts: [10]
 
 #include <sys/wait.h>
 #include <unistd.h>
@@ -16,17 +20,21 @@ __attribute__((noinline)) void func1() {}
 __attribute__((noinline)) void func2() {}
 
 int main(void) {
-  //                       child     | parent
-  int status;         // func1 func2 | func1 func2
-  func1();            //   +1        |   +1        (*)
-  pid_t pid = fork(); //             |
-  if (pid == -1)      //             |
-    return 1;         //             |
-  if (pid == 0)       //             |
-    func2();          //         +1  |
-  func1();            //   +1        |   +1
-  if (pid)            // ------------+------------
-    wait(&status);    //    2     1  |    2    0
-  return 0;           // (*)  the child inherits counter values prior to fork
-                      //      from the parent in non-continuous mode.
+  //                           child     | parent
+  //                         func1 func2 | func1 func2
+  func1();              //   +10       |   +1        (*)
+  int i = 10;           //             |
+  while (i-- > 0) {     //             |
+    pid_t pid = fork(); //             |
+    if (pid == -1)      //             |
+      return 1;         //             |
+    if (pid == 0) {     //             |
+      func2();          //         +10 |
+      func1();          //   +10       |
+      return 0;         //             |
+    }                   //             |
+  }                     // ------------+------------
+  int status;           //   20     10 |   1     0
+  wait(&status);        // (*)  the child inherits counter values prior to fork
+  return 0;             //      from the parent in non-continuous mode.
 }

compiler-rt/test/profile/instrprof-no-mmap-during-merging.c

@@ -0,0 +1,25 @@
+// RUN: mkdir -p %t.d && cd %t.d
+// RUN: rm -f *.profraw
+// RUN: %clang_pgogen %s -o a.out
+
+// Need to run a.out twice, the second time a merge will occur which will trigger an mmap.
+// RUN: ./a.out
+// RUN: llvm-profdata show default_*.profraw --all-functions --counts --memop-sizes 2>&1 | FileCheck %s -check-prefix=PROFDATA
+// RUN: env LLVM_NO_MMAP=1 LLVM_PROFILE_VERBOSE=1 ./a.out 2>&1 | FileCheck %s
+// RUN: llvm-profdata show default_*.profraw --all-functions --counts --memop-sizes 2>&1 | FileCheck %s -check-prefix=PROFDATA2
+
+// CHECK: Could not use mmap; using fread instead.
+// PROFDATA: Block counts: [1]
+// PROFDATA: [  0,    0,          1 ]
+// PROFDATA: Maximum function count: 1
+// PROFDATA2: Block counts: [2]
+// PROFDATA2: [  0,    0,          2 ]
+// PROFDATA2: Maximum function count: 2
+
+#include <string.h>
+int ar[8];
+int main() {
+  memcpy(ar, ar + 2, ar[0]);
+  memcpy(ar, ar + 2, ar[2]);
+  return ar[2];
+}