@@ -33,6 +33,22 @@
//#define DEBUG_MMAP
+/*
+ * On x86_64 we can tell mmap that we only want to map within the first 32
+ * bits to not get pointers that potentially exceed the return size. Without
+ * this flag set mmap will eventually break for users when running 32-on-64.
+ *
+ * However, Linux doesn't implement this for non-x86_64 systems. So we have
+ * to safeguard the bit with an empty flag which will be ignore on other
+ * architectures. At least we fixed the "common case" this way :).
+ *
+ * - agraf
+ */
+#if !defined(MAP_32BIT) || !defined(__x86_64__) || (TARGET_LONG_BITS != 32)
+#undef MAP_32BIT
+#define MAP_32BIT 0
+#endif
+
#if defined(CONFIG_USE_NPTL)
static pthread_mutex_t mmap_mutex = PTHREAD_MUTEX_INITIALIZER;
static __thread int mmap_lock_count;
@@ -169,7 +185,7 @@ static int mmap_frag(abi_ulong real_start,
if (prot1 == 0) {
/* no page was there, so we allocate one */
void *p = mmap(host_start, qemu_host_page_size, prot,
- flags | MAP_ANONYMOUS, -1, 0);
+ flags | MAP_ANONYMOUS | MAP_32BIT, -1, 0);
if (p == MAP_FAILED)
return -1;
prot1 = prot;
@@ -292,7 +308,7 @@ abi_ulong mmap_find_vma(abi_ulong start, abi_ulong size)
* - shmat() with SHM_REMAP flag
*/
ptr = mmap(g2h(addr), size, PROT_NONE,
- MAP_ANONYMOUS|MAP_PRIVATE|MAP_NORESERVE, -1, 0);
+ MAP_ANONYMOUS|MAP_PRIVATE|MAP_NORESERVE|MAP_32BIT, -1, 0);
/* ENOMEM, if host address space has no memory */
if (ptr == MAP_FAILED) {
@@ -454,14 +470,15 @@ abi_long target_mmap(abi_ulong start, abi_ulong len, int prot,
especially important if qemu_host_page_size >
qemu_real_host_page_size */
p = mmap(g2h(mmap_start),
- host_len, prot, flags | MAP_FIXED | MAP_ANONYMOUS, -1, 0);
+ host_len, prot, flags | MAP_FIXED | MAP_ANONYMOUS | MAP_32BIT,
+ -1, 0);
if (p == MAP_FAILED)
goto fail;
/* update start so that it points to the file position at 'offset' */
host_start = (unsigned long)p;
if (!(flags & MAP_ANONYMOUS)) {
p = mmap(g2h(mmap_start), len, prot,
- flags | MAP_FIXED, fd, host_offset);
+ flags | MAP_FIXED | MAP_32BIT, fd, host_offset);
host_start += offset - host_offset;
}
start = h2g(host_start);
@@ -495,8 +512,8 @@ abi_long target_mmap(abi_ulong start, abi_ulong len, int prot,
goto fail;
}
retaddr = target_mmap(start, len, prot | PROT_WRITE,
- MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS,
- -1, 0);
+ MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS |
+ MAP_32BIT, -1, 0);
if (retaddr == -1)
goto fail;
if (pread(fd, g2h(start), len, offset) == -1)
@@ -547,7 +564,7 @@ abi_long target_mmap(abi_ulong start, abi_ulong len, int prot,
else
offset1 = offset + real_start - start;
p = mmap(g2h(real_start), real_end - real_start,
- prot, flags, fd, offset1);
+ prot, flags | MAP_32BIT, fd, offset1);
if (p == MAP_FAILED)
goto fail;
}
@@ -603,8 +620,8 @@ static void mmap_reserve(abi_ulong start, abi_ulong size)
}
if (real_start != real_end) {
mmap(g2h(real_start), real_end - real_start, PROT_NONE,
- MAP_FIXED | MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE,
- -1, 0);
+ MAP_FIXED | MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE |
+ MAP_32BIT, -1, 0);
}
}
When running a 32 bit guest on a 64 bit host, we can run into trouble while calling the host's mmap() because it could potentially give us a 64 bit return value which the guest can't interpret. There are 2 ways of dealing with this: 1) Only do MAP_FIXED mmap calls and implement our own vm management in QEMU 2) Tell the kernel that we only want mappings in the lower 32 bits Way 1 is very involved and hard to do. It's been advocated forever now but nobody sat down to actually implement it. Way 2 is easy. It's what this patch does. However, it only works on x86_64 because that's the only platform implementing the MAP_32BIT flag. Since most people are on x86_64 though, I think it's a good enough compromise for now though Signed-off-by: Alexander Graf <agraf@suse.de> --- linux-user/mmap.c | 35 ++++++++++++++++++++++++++--------- 1 files changed, 26 insertions(+), 9 deletions(-)