path: root/mozglue/linker/Mappable.cpp

/* This Source Code Form is subject to the terms of the Mozilla Public
 * License, v. 2.0. If a copy of the MPL was not distributed with this file,
 * You can obtain one at http://mozilla.org/MPL/2.0/. */

#include <fcntl.h>
#include <unistd.h>
#include <sys/mman.h>
#include <sys/stat.h>
#include <cstring>
#include <cstdlib>
#include <cstdio>
#include <string>

#include "Mappable.h"

#include "mozilla/UniquePtr.h"

#include <sys/stat.h>
#include <errno.h>
#include "ElfLoader.h"
#include "SeekableZStream.h"
#include "XZStream.h"
#include "Logging.h"

using mozilla::MakeUnique;
using mozilla::UniquePtr;

class CacheValidator
{
public:
  CacheValidator(const char* aCachedLibPath, Zip* aZip, Zip::Stream* aStream)
    : mCachedLibPath(aCachedLibPath)
  {
    static const char kChecksumSuffix[] = ".crc";

    mCachedChecksumPath =
      MakeUnique<char[]>(strlen(aCachedLibPath) + sizeof(kChecksumSuffix));
    sprintf(mCachedChecksumPath.get(), "%s%s", aCachedLibPath, kChecksumSuffix);
    DEBUG_LOG("mCachedChecksumPath: %s", mCachedChecksumPath.get());

    mChecksum = aStream->GetCRC32();
    DEBUG_LOG("mChecksum: %x", mChecksum);
  }

  // Returns whether the cache is valid and up-to-date.
  bool IsValid() const
  {
    // Validate based on checksum.
    RefPtr<Mappable> checksumMap = MappableFile::Create(mCachedChecksumPath.get());
    if (!checksumMap) {
      // Force caching if checksum is missing in cache.
      return false;
    }

    DEBUG_LOG("Comparing %x with %s", mChecksum, mCachedChecksumPath.get());
    MappedPtr checksumBuf = checksumMap->mmap(nullptr, checksumMap->GetLength(),
                                              PROT_READ, MAP_PRIVATE, 0);
    if (checksumBuf == MAP_FAILED) {
      WARN("Couldn't map %s to validate checksum", mCachedChecksumPath.get());
      return false;
    }
    if (memcmp(checksumBuf, &mChecksum, sizeof(mChecksum))) {
      return false;
    }
    return !access(mCachedLibPath.c_str(), R_OK);
  }

  // Caches the APK-provided checksum used in future cache validations.
  void CacheChecksum() const
  {
    AutoCloseFD fd(open(mCachedChecksumPath.get(),
                        O_TRUNC | O_RDWR | O_CREAT | O_NOATIME,
                        S_IRUSR | S_IWUSR));
    if (fd == -1) {
      WARN("Couldn't open %s to update checksum", mCachedChecksumPath.get());
      return;
    }

    DEBUG_LOG("Updating checksum %s", mCachedChecksumPath.get());

    const size_t size = sizeof(mChecksum);
    size_t written = 0;
    while (written < size) {
      ssize_t ret = write(fd,
                          reinterpret_cast<const uint8_t*>(&mChecksum) + written,
                          size - written);
      if (ret >= 0) {
        written += ret;
      } else if (errno != EINTR) {
        WARN("Writing checksum %s failed with errno %d",
             mCachedChecksumPath.get(), errno);
        break;
      }
    }
  }

private:
  const std::string mCachedLibPath;
  UniquePtr<char[]> mCachedChecksumPath;
  uint32_t mChecksum;
};

Mappable *
MappableFile::Create(const char *path)
{
  int fd = open(path, O_RDONLY);
  if (fd != -1)
    return new MappableFile(fd);
  return nullptr;
}

MemoryRange
MappableFile::mmap(const void *addr, size_t length, int prot, int flags,
                   off_t offset)
{
  MOZ_ASSERT(fd != -1);
  MOZ_ASSERT(!(flags & MAP_SHARED));
  flags |= MAP_PRIVATE;

  return MemoryRange::mmap(const_cast<void *>(addr), length, prot, flags,
                           fd, offset);
}

void
MappableFile::finalize()
{
  /* Close file ; equivalent to close(fd.forget()) */
  fd = -1;
}

size_t
MappableFile::GetLength() const
{
  struct stat st;
  return fstat(fd, &st) ? 0 : st.st_size;
}

Mappable *
MappableExtractFile::Create(const char *name, Zip *zip, Zip::Stream *stream)
{
  MOZ_ASSERT(zip && stream);

  const char *cachePath = getenv("MOZ_LINKER_CACHE");
  if (!cachePath || !*cachePath) {
    WARN("MOZ_LINKER_EXTRACT is set, but not MOZ_LINKER_CACHE; "
        "not extracting");
    return nullptr;
  }

  // Ensure that the cache dir is private.
  chmod(cachePath, 0770);

  UniquePtr<char[]> path =
    MakeUnique<char[]>(strlen(cachePath) + strlen(name) + 2);
  sprintf(path.get(), "%s/%s", cachePath, name);

  CacheValidator validator(path.get(), zip, stream);
  if (validator.IsValid()) {
    DEBUG_LOG("Reusing %s", static_cast<char *>(path.get()));
    return MappableFile::Create(path.get());
  }
  DEBUG_LOG("Extracting to %s", static_cast<char *>(path.get()));
  AutoCloseFD fd;
  fd = open(path.get(), O_TRUNC | O_RDWR | O_CREAT | O_NOATIME,
                        S_IRUSR | S_IWUSR);
  if (fd == -1) {
    ERROR("Couldn't open %s to decompress library", path.get());
    return nullptr;
  }
  AutoUnlinkFile file(path.release());
  if (stream->GetType() == Zip::Stream::DEFLATE) {
    if (ftruncate(fd, stream->GetUncompressedSize()) == -1) {
      ERROR("Couldn't ftruncate %s to decompress library", file.get());
      return nullptr;
    }
    /* Map the temporary file for use as inflate buffer */
    MappedPtr buffer(MemoryRange::mmap(nullptr, stream->GetUncompressedSize(),
                                       PROT_WRITE, MAP_SHARED, fd, 0));
    if (buffer == MAP_FAILED) {
      ERROR("Couldn't map %s to decompress library", file.get());
      return nullptr;
    }

    zxx_stream zStream = stream->GetZStream(buffer);

    /* Decompress */
    if (inflateInit2(&zStream, -MAX_WBITS) != Z_OK) {
      ERROR("inflateInit failed: %s", zStream.msg);
      return nullptr;
    }
    if (inflate(&zStream, Z_FINISH) != Z_STREAM_END) {
      ERROR("inflate failed: %s", zStream.msg);
      return nullptr;
    }
    if (inflateEnd(&zStream) != Z_OK) {
      ERROR("inflateEnd failed: %s", zStream.msg);
      return nullptr;
    }
    if (zStream.total_out != stream->GetUncompressedSize()) {
      ERROR("File not fully uncompressed! %ld / %d", zStream.total_out,
          static_cast<unsigned int>(stream->GetUncompressedSize()));
      return nullptr;
    }
  } else if (XZStream::IsXZ(stream->GetBuffer(), stream->GetSize())) {
    XZStream xzStream(stream->GetBuffer(), stream->GetSize());

    if (!xzStream.Init()) {
      ERROR("Couldn't initialize XZ decoder");
      return nullptr;
    }
    DEBUG_LOG("XZStream created, compressed=%u, uncompressed=%u",
              xzStream.Size(), xzStream.UncompressedSize());

    if (ftruncate(fd, xzStream.UncompressedSize()) == -1) {
      ERROR("Couldn't ftruncate %s to decompress library", file.get());
      return nullptr;
    }
    MappedPtr buffer(MemoryRange::mmap(nullptr, xzStream.UncompressedSize(),
                                       PROT_WRITE, MAP_SHARED, fd, 0));
    if (buffer == MAP_FAILED) {
      ERROR("Couldn't map %s to decompress library", file.get());
      return nullptr;
    }
    const size_t written = xzStream.Decode(buffer, buffer.GetLength());
    DEBUG_LOG("XZStream decoded %u", written);
    if (written != buffer.GetLength()) {
      ERROR("Error decoding XZ file %s", file.get());
      return nullptr;
    }
  } else if (stream->GetType() == Zip::Stream::STORE) {
    SeekableZStream zStream;
    if (!zStream.Init(stream->GetBuffer(), stream->GetSize())) {
      ERROR("Couldn't initialize SeekableZStream for %s", name);
      return nullptr;
    }
    if (ftruncate(fd, zStream.GetUncompressedSize()) == -1) {
      ERROR("Couldn't ftruncate %s to decompress library", file.get());
      return nullptr;
    }
    MappedPtr buffer(MemoryRange::mmap(nullptr, zStream.GetUncompressedSize(),
                                       PROT_WRITE, MAP_SHARED, fd, 0));
    if (buffer == MAP_FAILED) {
      ERROR("Couldn't map %s to decompress library", file.get());
      return nullptr;
    }

    if (!zStream.Decompress(buffer, 0, zStream.GetUncompressedSize())) {
      ERROR("%s: failed to decompress", name);
      return nullptr;
    }
  } else {
    return nullptr;
  }

  validator.CacheChecksum();
  return new MappableExtractFile(fd.forget(), file.release());
}

/**
 * _MappableBuffer is a buffer which content can be mapped at different
 * locations in the virtual address space.
 * On Linux, uses a (deleted) temporary file on a tmpfs for sharable content.
 */
class _MappableBuffer: public MappedPtr
{
public:
  /**
   * Returns a _MappableBuffer instance with the given name and the given
   * length.
   */
  static _MappableBuffer *Create(const char *name, size_t length)
  {
    AutoCloseFD fd;

    /* On Linux, use /dev/shm as base directory for temporary files, assuming
     * it's on tmpfs */
    /* TODO: check that /dev/shm is tmpfs */
    char path[256];
    sprintf(path, "/dev/shm/%s.XXXXXX", name);
    fd = mkstemp(path);
    if (fd == -1)
      return nullptr;
    unlink(path);
    ftruncate(fd, length);

    void *buf = ::mmap(nullptr, length, PROT_READ | PROT_WRITE,
                       MAP_SHARED, fd, 0);
    if (buf != MAP_FAILED) {
      DEBUG_LOG("Decompression buffer of size %ld in \"%s\", mapped @%p",
                length, path, buf);
      return new _MappableBuffer(fd.forget(), buf, length);
    }

    return nullptr;
  }

  void *mmap(const void *addr, size_t length, int prot, int flags, off_t offset)
  {
    MOZ_ASSERT(fd != -1);
    return ::mmap(const_cast<void *>(addr), length, prot, flags, fd, offset);
  }

private:
  _MappableBuffer(int fd, void *buf, size_t length)
  : MappedPtr(buf, length), fd(fd) { }

  /* File descriptor for the temporary file or ashmem */
  AutoCloseFD fd;
};


Mappable *
MappableDeflate::Create(const char *name, Zip *zip, Zip::Stream *stream)
{
  MOZ_ASSERT(stream->GetType() == Zip::Stream::DEFLATE);
  _MappableBuffer *buf = _MappableBuffer::Create(name, stream->GetUncompressedSize());
  if (buf)
    return new MappableDeflate(buf, zip, stream);
  return nullptr;
}

MappableDeflate::MappableDeflate(_MappableBuffer *buf, Zip *zip,
                                 Zip::Stream *stream)
: zip(zip), buffer(buf), zStream(stream->GetZStream(*buf)) { }

MappableDeflate::~MappableDeflate() { }

MemoryRange
MappableDeflate::mmap(const void *addr, size_t length, int prot, int flags, off_t offset)
{
  MOZ_ASSERT(buffer);
  MOZ_ASSERT(!(flags & MAP_SHARED));
  flags |= MAP_PRIVATE;

  /* The deflate stream is uncompressed up to the required offset + length, if
   * it hasn't previously been uncompressed */
  ssize_t missing = offset + length + zStream.avail_out - buffer->GetLength();
  if (missing > 0) {
    uInt avail_out = zStream.avail_out;
    zStream.avail_out = missing;
    if ((*buffer == zStream.next_out) &&
        (inflateInit2(&zStream, -MAX_WBITS) != Z_OK)) {
      ERROR("inflateInit failed: %s", zStream.msg);
      return MemoryRange(MAP_FAILED, 0);
    }
    int ret = inflate(&zStream, Z_SYNC_FLUSH);
    if (ret < 0) {
      ERROR("inflate failed: %s", zStream.msg);
      return MemoryRange(MAP_FAILED, 0);
    }
    if (ret == Z_NEED_DICT) {
      ERROR("zstream requires a dictionary. %s", zStream.msg);
      return MemoryRange(MAP_FAILED, 0);
    }
    zStream.avail_out = avail_out - missing + zStream.avail_out;
    if (ret == Z_STREAM_END) {
      if (inflateEnd(&zStream) != Z_OK) {
        ERROR("inflateEnd failed: %s", zStream.msg);
        return MemoryRange(MAP_FAILED, 0);
      }
      if (zStream.total_out != buffer->GetLength()) {
        ERROR("File not fully uncompressed! %ld / %d", zStream.total_out,
            static_cast<unsigned int>(buffer->GetLength()));
        return MemoryRange(MAP_FAILED, 0);
      }
    }
  }

  return MemoryRange(buffer->mmap(addr, length, prot, flags, offset), length);
}

void
MappableDeflate::finalize()
{
  /* Free zlib internal buffers */
  inflateEnd(&zStream);
  /* Free decompression buffer */
  buffer = nullptr;
  /* Remove reference to Zip archive */
  zip = nullptr;
}

size_t
MappableDeflate::GetLength() const
{
  return buffer->GetLength();
}

Mappable *
MappableSeekableZStream::Create(const char *name, Zip *zip,
                                Zip::Stream *stream)
{
  MOZ_ASSERT(stream->GetType() == Zip::Stream::STORE);
  UniquePtr<MappableSeekableZStream> mappable(new MappableSeekableZStream(zip));

  pthread_mutexattr_t recursiveAttr;
  pthread_mutexattr_init(&recursiveAttr);
  pthread_mutexattr_settype(&recursiveAttr, PTHREAD_MUTEX_RECURSIVE);

  if (pthread_mutex_init(&mappable->mutex, &recursiveAttr))
    return nullptr;

  if (!mappable->zStream.Init(stream->GetBuffer(), stream->GetSize()))
    return nullptr;

  mappable->buffer.reset(_MappableBuffer::Create(name,
                              mappable->zStream.GetUncompressedSize()));
  if (!mappable->buffer)
    return nullptr;

  mappable->chunkAvail = MakeUnique<unsigned char[]>(mappable->zStream.GetChunksNum());

  return mappable.release();
}

MappableSeekableZStream::MappableSeekableZStream(Zip *zip)
: zip(zip), chunkAvailNum(0) { }

MappableSeekableZStream::~MappableSeekableZStream()
{
  pthread_mutex_destroy(&mutex);
}

MemoryRange
MappableSeekableZStream::mmap(const void *addr, size_t length, int prot,
                              int flags, off_t offset)
{
  /* Map with PROT_NONE so that accessing the mapping would segfault, and
   * bring us to ensure() */
  void *res = buffer->mmap(addr, length, PROT_NONE, flags, offset);
  if (res == MAP_FAILED)
    return MemoryRange(MAP_FAILED, 0);

  /* Store the mapping, ordered by offset and length */
  std::vector<LazyMap>::reverse_iterator it;
  for (it = lazyMaps.rbegin(); it < lazyMaps.rend(); ++it) {
    if ((it->offset < offset) ||
        ((it->offset == offset) && (it->length < length)))
      break;
  }
  LazyMap map = { res, length, prot, offset };
  lazyMaps.insert(it.base(), map);
  return MemoryRange(res, length);
}

void
MappableSeekableZStream::munmap(void *addr, size_t length)
{
  std::vector<LazyMap>::iterator it;
  for (it = lazyMaps.begin(); it < lazyMaps.end(); ++it)
    if ((it->addr = addr) && (it->length == length)) {
      lazyMaps.erase(it);
      ::munmap(addr, length);
      return;
    }
  MOZ_CRASH("munmap called with unknown mapping");
}

void
MappableSeekableZStream::finalize() { }

bool
MappableSeekableZStream::ensure(const void *addr)
{
  DEBUG_LOG("ensure @%p", addr);
  const void *addrPage = PageAlignedPtr(addr);
  /* Find the mapping corresponding to the given page */
  std::vector<LazyMap>::iterator map;
  for (map = lazyMaps.begin(); map < lazyMaps.end(); ++map) {
    if (map->Contains(addrPage))
      break;
  }
  if (map == lazyMaps.end())
    return false;

  /* Find corresponding chunk */
  off_t mapOffset = map->offsetOf(addrPage);
  off_t chunk = mapOffset / zStream.GetChunkSize();

  /* In the typical case, we just need to decompress the chunk entirely. But
   * when the current mapping ends in the middle of the chunk, we want to
   * stop at the end of the corresponding page.
   * However, if another mapping needs the last part of the chunk, we still
   * need to continue. As mappings are ordered by offset and length, we don't
   * need to scan the entire list of mappings.
   * It is safe to run through lazyMaps here because the linker is never
   * going to call mmap (which adds lazyMaps) while this function is
   * called. */
  size_t length = zStream.GetChunkSize(chunk);
  off_t chunkStart = chunk * zStream.GetChunkSize();
  off_t chunkEnd = chunkStart + length;
  std::vector<LazyMap>::iterator it;
  for (it = map; it < lazyMaps.end(); ++it) {
    if (chunkEnd <= it->endOffset())
      break;
  }
  if ((it == lazyMaps.end()) || (chunkEnd > it->endOffset())) {
    /* The mapping "it" points at now is past the interesting one */
    --it;
    length = it->endOffset() - chunkStart;
  }

  length = PageAlignedSize(length);

  /* The following lock can be re-acquired by the thread holding it.
   * If this happens, it means the following code is interrupted somehow by
   * some signal, and ends up retriggering a chunk decompression for the
   * same MappableSeekableZStream.
   * If the chunk to decompress is different the second time, then everything
   * is safe as the only common data touched below is chunkAvailNum, and it is
   * atomically updated (leaving out any chance of an interruption while it is
   * updated affecting the result). If the chunk to decompress is the same, the
   * worst thing that can happen is chunkAvailNum being incremented one too
   * many times, which doesn't affect functionality. The chances of it
   * happening being pretty slim, and the effect being harmless, we can just
   * ignore the issue. Other than that, we'd just be wasting time decompressing
   * the same chunk twice. */
  AutoLock lock(&mutex);

  /* The very first page is mapped and accessed separately of the rest, and
   * as such, only the first page of the first chunk is decompressed this way.
   * When we fault in the remaining pages of that chunk, we want to decompress
   * the complete chunk again. Short of doing that, we would end up with
   * no data between PageSize() and chunkSize, which would effectively corrupt
   * symbol resolution in the underlying library. */
  if (chunkAvail[chunk] < PageNumber(length)) {
    if (!zStream.DecompressChunk(*buffer + chunkStart, chunk, length))
      return false;

    /* Only count if we haven't already decompressed parts of the chunk */
    if (chunkAvail[chunk] == 0)
      chunkAvailNum++;

    chunkAvail[chunk] = PageNumber(length);
  }

  /* Flip the chunk mapping protection to the recorded flags. We could
   * also flip the protection for other mappings of the same chunk,
   * but it's easier to skip that and let further segfaults call
   * ensure again. */
  const void *chunkAddr = reinterpret_cast<const void *>
                          (reinterpret_cast<uintptr_t>(addrPage)
                           - mapOffset % zStream.GetChunkSize());
  const void *chunkEndAddr = reinterpret_cast<const void *>
                             (reinterpret_cast<uintptr_t>(chunkAddr) + length);
  
  const void *start = std::max(map->addr, chunkAddr);
  const void *end = std::min(map->end(), chunkEndAddr);
  length = reinterpret_cast<uintptr_t>(end)
           - reinterpret_cast<uintptr_t>(start);

  if (mprotect(const_cast<void *>(start), length, map->prot) == 0) {
    DEBUG_LOG("mprotect @%p, 0x%" PRIxSize ", 0x%x", start, length, map->prot);
    return true;
  }

  ERROR("mprotect @%p, 0x%" PRIxSize ", 0x%x failed with errno %d",
      start, length, map->prot, errno);
  return false;
}

void
MappableSeekableZStream::stats(const char *when, const char *name) const
{
  size_t nEntries = zStream.GetChunksNum();
  DEBUG_LOG("%s: %s; %" PRIdSize "/%" PRIdSize " chunks decompressed",
            name, when, static_cast<size_t>(chunkAvailNum), nEntries);

  size_t len = 64;
  UniquePtr<char[]> map = MakeUnique<char[]>(len + 3);
  map[0] = '[';

  for (size_t i = 0, j = 1; i < nEntries; i++, j++) {
    map[j] = chunkAvail[i] ? '*' : '_';
    if ((j == len) || (i == nEntries - 1)) {
      map[j + 1] = ']';
      map[j + 2] = '\0';
      DEBUG_LOG("%s", static_cast<char *>(map.get()));
      j = 0;
    }
  }
}

size_t
MappableSeekableZStream::GetLength() const
{
  return buffer->GetLength();
}