[Bug 259076] pthread_mutex_init fails with limited AS

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=259076

--- Comment #9 from Konstantin Belousov <kib@FreeBSD.org> ---
(In reply to Denis Koreshkov from comment #8)
Why do not they help?  When mmap(2) returns MMAP_FAILED, morepages() returns 0.
Then, morecore would not touch pagepool_start.

Hmm, It is probably wrong that pagepool_end is not reset there.  I think I can
agree to some part of your note, that we should e.g. set both pagepool_start
and pagepool_end to 0 to avoid munmap(2)-ing something not own by us on the
next allocation attempt.

Like this (didn't tested)

diff --git a/libexec/rtld-elf/rtld_malloc.c b/libexec/rtld-elf/rtld_malloc.c
index 64218b5bb786..54159bec8e4e 100644
--- a/libexec/rtld-elf/rtld_malloc.c
+++ b/libexec/rtld-elf/rtld_malloc.c
@@ -271,21 +271,21 @@ morepages(int n)
                }
        }

-       if (pagepool_start == MAP_FAILED)
-               pagepool_start = 0;
        offset = (uintptr_t)pagepool_start - rounddown2(
            (uintptr_t)pagepool_start, pagesz);

-       pagepool_start = mmap(0, n * pagesz, PROT_READ | PROT_WRITE,
+       addr = mmap(0, n * pagesz, PROT_READ | PROT_WRITE,
            MAP_ANON | MAP_PRIVATE, -1, 0);
-       if (pagepool_start == MAP_FAILED) {
+       if (addr == MAP_FAILED) {
 #ifdef IN_RTLD
                rtld_fdprintf(STDERR_FILENO, _BASENAME_RTLD ": morepages: "
                    "cannot mmap anonymous memory: %s\n",
                    rtld_strerror(errno));
 #endif
+               pagepool_start = pagepool_end = 0;
                return (0);
        }
+       pagepool_start = addr;
        pagepool_end = pagepool_start + n * pagesz;
        pagepool_start += offset;

-- 
You are receiving this mail because:
You are the assignee for the bug.