linux-user: Avoid mmap of the last byte of the reserved_va
There is an overflow problem in mmap_find_vma_reserved:
when reserved_va == UINT32_MAX, end may overflow to 0.
Rather than a larger rewrite at this time, simply avoid
the final byte of the VA, which avoids searching the
final page, which avoids the overflow.
Cc: qemu-stable@nongnu.org
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/1741
Fixes: 95059f9c ("include/exec: Change reserved_va semantics to last byte")
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Tested-by: Michael Tokarev <mjt@tls.msk.ru>
Reviewed-by: Michael Tokarev <mjt@tls.msk.ru>
Message-Id: <20230629080835.71371-1-richard.henderson@linaro.org>
diff --git a/linux-user/mmap.c b/linux-user/mmap.c
index 0aa8ae7..2692936 100644
--- a/linux-user/mmap.c
+++ b/linux-user/mmap.c
@@ -281,9 +281,15 @@
/* Note that start and size have already been aligned by mmap_find_vma. */
end_addr = start + size;
+ /*
+ * Start at the top of the address space, ignoring the last page.
+ * If reserved_va == UINT32_MAX, then end_addr wraps to 0,
+ * throwing the rest of the calculations off.
+ * TODO: rewrite using last_addr instead.
+ * TODO: use the interval tree instead of probing every page.
+ */
if (start > reserved_va - size) {
- /* Start at the top of the address space. */
- end_addr = ((reserved_va + 1 - size) & -align) + size;
+ end_addr = ((reserved_va - size) & -align) + size;
looped = true;
}
@@ -296,8 +302,8 @@
/* Failure. The entire address space has been searched. */
return (abi_ulong)-1;
}
- /* Re-start at the top of the address space. */
- addr = end_addr = ((reserved_va + 1 - size) & -align) + size;
+ /* Re-start at the top of the address space (see above). */
+ addr = end_addr = ((reserved_va - size) & -align) + size;
looped = true;
} else {
prot = page_get_flags(addr);