From 38a44fec0347cd14bf379ad7e2a523499bd160fb Mon Sep 17 00:00:00 2001
From: Dariusz Sosnowski <dsosnowski@dsosnowski.pl>
Date: Tue, 15 Sep 2020 23:21:51 +0200
Subject: [PATCH] MonitoredSharedMemory: PoC - snapshot hugepages

---
 src/MonitoredSharedMemory.cc | 39 ++++++++++++++++++++++++++++++------
 src/Session.cc               | 24 ++++++++++++++++++----
 src/record_syscall.cc        |  6 ++++++
 3 files changed, 59 insertions(+), 10 deletions(-)

diff --git a/src/MonitoredSharedMemory.cc b/src/MonitoredSharedMemory.cc
index 29aac3183ca..74eaa51a402 100644
--- a/src/MonitoredSharedMemory.cc
+++ b/src/MonitoredSharedMemory.cc
@@ -19,13 +19,39 @@ MonitoredSharedMemory::~MonitoredSharedMemory() { munmap(real_mem, size); }
 
 static const char dconf_suffix[] = "/dconf/user";
 
+// static const char dpdk_prefix[] = "/run/dpdk/rte";
+
+static const char rtemap_prefix[] = "/dev/hugepages/rtemap";
+
 void MonitoredSharedMemory::maybe_monitor(RecordTask* t,
                                           const string& file_name,
                                           const AddressSpace::Mapping& m,
-                                          int tracee_fd, uint64_t offset) {
+                                          int tracee_fd, uint64_t offset)
+{
+  bool is_dconf = true;
+  // bool is_dpdk = true;
+  bool is_rtemap = true;
+
   size_t dconf_suffix_len = sizeof(dconf_suffix) - 1;
   if (file_name.size() < dconf_suffix_len ||
       file_name.substr(file_name.size() - dconf_suffix_len) != dconf_suffix) {
+    is_dconf = false;
+  }
+
+  // size_t dpdk_prefix_len = sizeof(dpdk_prefix) - 1;
+  // if (file_name.size() < dpdk_prefix_len ||
+  //     file_name.substr(0, dpdk_prefix_len) != dpdk_prefix) {
+  //   is_dpdk = false;
+  // }
+
+  size_t rtemap_prefix_len = sizeof(rtemap_prefix) - 1;
+  if (file_name.size() < rtemap_prefix_len ||
+      file_name.substr(0, rtemap_prefix_len) != rtemap_prefix) {
+    is_rtemap = false;
+  }
+
+  // if (!is_dconf && !is_dpdk && !is_rtemap) {
+  if (!is_dconf && !is_rtemap) {
     return;
   }
 
@@ -33,7 +59,7 @@ void MonitoredSharedMemory::maybe_monitor(RecordTask* t,
 
   ScopedFd fd = remote.retrieve_fd(tracee_fd);
   uint8_t* real_mem = static_cast<uint8_t*>(
-      mmap(NULL, m.map.size(), PROT_READ, MAP_SHARED, fd, offset));
+      mmap(NULL, m.map.size(), PROT_READ | PROT_WRITE, MAP_SHARED, fd, offset));
   ASSERT(t, real_mem != MAP_FAILED);
 
   auto result = shared_ptr<MonitoredSharedMemory>(
@@ -74,10 +100,11 @@ void MonitoredSharedMemory::check_for_changes(RecordTask* t,
     m = Session::recreate_shared_mmap(remote, m, Session::DISCARD_CONTENTS,
                                       move(msm));
   }
-  if (!memcmp(m.local_addr, real_mem, size)) {
-    return;
-  }
-  memcpy(m.local_addr, real_mem, size);
+  // if (!memcmp(m.local_addr, real_mem, size)) {
+  //   return;
+  // }
+  // memcpy(m.local_addr, real_mem, size);
+  memcpy(real_mem, m.local_addr, size);
   t->record_local(m.map.start(), size, real_mem);
 }
 }
diff --git a/src/Session.cc b/src/Session.cc
index 77a492391e7..7598e414341 100644
--- a/src/Session.cc
+++ b/src/Session.cc
@@ -429,20 +429,36 @@ static void remap_shared_mmap(AutoRemoteSyscalls& remote, EmuFs& emu_fs,
 KernelMapping Session::create_shared_mmap(
     AutoRemoteSyscalls& remote, size_t size, remote_ptr<void> map_hint,
     const char* name, int tracee_prot, int tracee_flags,
-    MonitoredSharedMemory::shr_ptr&& monitored) {
+    MonitoredSharedMemory::shr_ptr&& monitored)
+{
+  bool is_huge_backed = false;
+
+  if (monitored) {
+    is_huge_backed = true;
+  }
+
   static int nonce = 0;
   // Create the segment we'll share with the tracee.
   char path[PATH_MAX];
-  snprintf(path, sizeof(path) - 1, "%s%s%s-%d-%d", tmp_dir(),
+  if (!is_huge_backed) {
+    snprintf(path, sizeof(path) - 1, "%s%s%s-%d-%d", tmp_dir(),
            rr_mapping_prefix(), name, remote.task()->real_tgid(), nonce++);
+  } else {
+    snprintf(path, sizeof(path) - 1, "%s%s%s-%d-%d", "/dev/hugepages",
+           rr_mapping_prefix(), name, remote.task()->real_tgid(), nonce++);
+  }
 
   ScopedFd shmem_fd(path, O_CREAT | O_EXCL | O_RDWR);
   /* Remove the fs name so that we don't have to worry about
    * cleaning up this segment in error conditions. */
-  unlink(path);
+  if (!is_huge_backed) {
+    unlink(path);
+  }
 
   int child_shmem_fd = remote.send_fd(shmem_fd);
-  resize_shmem_segment(shmem_fd, size);
+  if (!is_huge_backed) {
+    resize_shmem_segment(shmem_fd, size);
+  }
   LOG(debug) << "created shmem segment " << path;
 
   // Map the segment in ours and the tracee's address spaces.
diff --git a/src/record_syscall.cc b/src/record_syscall.cc
index 3927c0f8cc3..2d814e2ac6c 100644
--- a/src/record_syscall.cc
+++ b/src/record_syscall.cc
@@ -5524,6 +5524,12 @@ static void process_mmap(RecordTask* t, size_t length, int prot, int flags,
     MonitoredSharedMemory::maybe_monitor(t, tracee_file_name,
                                          t->vm()->mapping_of(addr), fd, offset);
   }
+
+  if ((prot & (PROT_WRITE | PROT_READ)) == (PROT_WRITE | PROT_READ) && (flags & MAP_SHARED) &&
+      !effectively_anonymous) {
+    MonitoredSharedMemory::maybe_monitor(t, tracee_file_name,
+                                         t->vm()->mapping_of(addr), fd, offset);
+  }
 }
 
 static void process_mremap(RecordTask* t, remote_ptr<void> old_addr,