Date: Tue, 25 Aug 2015 03:21:04 +0000 From: bugzilla-noreply@freebsd.org To: freebsd-threads@FreeBSD.org Subject: [Bug 202636] race in lib/libc/nls/msgcat.c Message-ID: <bug-202636-16@https.bugs.freebsd.org/bugzilla/>
next in thread | raw e-mail | index | archive | help
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=3D202636 Bug ID: 202636 Summary: race in lib/libc/nls/msgcat.c Product: Base System Version: 11.0-CURRENT Hardware: Any OS: Any Status: New Severity: Affects Only Me Priority: --- Component: threads Assignee: freebsd-threads@FreeBSD.org Reporter: henry.hu.sh@gmail.com Tracing from crashing xfdesktop, I found that the following program crashes= in seconds (in environment, set LANG=3Dzh_CN.UTF-8): #include <nl_types.h> #include <stdio.h> #include <errno.h> #include <string.h> #include <locale.h> #include <pthread.h> void* work(void *arg) { while (1) { nl_catd catd =3D catopen("libc", NL_CAT_LOCALE); catgets(catd, 1, 2, "No such file or directory"); catclose(catd); } } int main() { setlocale(LC_MESSAGES, ""); pthread_t thr1, thr2, thr3, thr4; pthread_create(&thr1, NULL, &work, NULL); pthread_create(&thr2, NULL, &work, NULL); pthread_create(&thr3, NULL, &work, NULL); pthread_create(&thr4, NULL, &work, NULL); pthread_join(thr1, NULL); pthread_join(thr2, NULL); pthread_join(thr3, NULL); pthread_join(thr4, NULL); } It always crashes somewhere in catgets: Program received signal SIGSEGV, Segmentation fault. [Switching to Thread 801016000 (LWP 101899/test)] catgets (catd=3D0x801615170, set_id=3D1, msg_id=3D2, s=3D0x400a9b "No such = file or directory") at /usr/src/lib/libc/nls/msgcat.c:280 280 u =3D ntohl((u_int32_t)cat_hdr->__nsets) - 1; and it looks like that catd is freed: (gdb) p *catd $2 =3D {__data =3D 0x5a5a5a5a5a5a5a5a, __size =3D 1515870810} After a closer look, it looks like that the increments to np->refcount is r= acy. See https://svnweb.freebsd.org/base/head/lib/libc/nls/msgcat.c?annotate=3D2= 78530. It is only protected by a read lock, so multiple threads may change the refcount at the same time, thus it is a race condition and may corrupt the refcount field.=20 Proposed fix: Index: lib/libc/nls/msgcat.c =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D --- lib/libc/nls/msgcat.c (=E7=89=88=E6=9C=AC 287028) +++ lib/libc/nls/msgcat.c (=E5=B7=A5=E4=BD=9C=E5=89=AF=E6=9C=AC) @@ -141,7 +141,7 @@ } /* Try to get it from the cache first */ - RLOCK(NLERR); + WLOCK(NLERR); SLIST_FOREACH(np, &cache, list) { if ((strcmp(np->name, name) =3D=3D 0) && ((lang !=3D NULL && np->lang !=3D NULL && @@ -376,7 +376,7 @@ * One more try in cache; if it was not found by name, * it might still be found by absolute path. */ - RLOCK(NLERR); + WLOCK(NLERR); SLIST_FOREACH(np, &cache, list) { if ((np->path !=3D NULL) && (strcmp(np->path, path) =3D=3D = 0)) { np->refcount++; This patch fixes the case in catopen(). This also changes the access in load_msgcat(), which also seems to be incorrect. --=20 You are receiving this mail because: You are the assignee for the bug.=
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?bug-202636-16>