#define R_SYM(x) ((x)>>32)
#endif
+#define MAXP2(a,b) (-(-(a)&-(b)))
+#define ALIGN(x,y) ((x)+(y)-1 & -(y))
+
struct debug {
int ver;
void *head;
void **new_dtv;
unsigned char *new_tls;
int new_dtv_idx, new_tls_idx;
+ struct dso *fini_next;
char *shortname;
char buf[];
};
void __init_ssp(size_t *);
void *__install_initial_tls(void *);
-static struct dso *head, *tail, *libc;
+static struct dso *head, *tail, *libc, *fini_head;
static char *env_path, *sys_path, *r_path;
static int ssp_used;
static int runtime;
static pthread_rwlock_t lock;
static struct debug debug;
static size_t *auxv;
-static size_t tls_cnt, tls_size;
+static size_t tls_cnt, tls_offset, tls_start, tls_align = 4*sizeof(size_t);
+static pthread_mutex_t init_fini_lock = { ._m_type = PTHREAD_MUTEX_RECURSIVE };
struct debug *_dl_debug_addr = &debug;
if (!h) h = sysv_hash(s);
sym = sysv_lookup(s, h, dso);
}
- if (sym && (!need_def || sym->st_shndx) && sym->st_value
- && (1<<(sym->st_info&0xf) & OK_TYPES)
- && (1<<(sym->st_info>>4) & OK_BINDS)) {
- if (def.sym && sym->st_info>>4 == STB_WEAK) continue;
- def.sym = sym;
- def.dso = dso;
- if (sym->st_info>>4 == STB_GLOBAL) break;
- }
+ if (!sym) continue;
+ if (!sym->st_shndx)
+ if (need_def || (sym->st_info&0xf) == STT_TLS)
+ continue;
+ if (!sym->st_value)
+ if ((sym->st_info&0xf) != STT_TLS)
+ continue;
+ if (!(1<<(sym->st_info&0xf) & OK_TYPES)) continue;
+ if (!(1<<(sym->st_info>>4) & OK_BINDS)) continue;
+
+ if (def.sym && sym->st_info>>4 == STB_WEAK) continue;
+ def.sym = sym;
+ def.dso = dso;
+ if (sym->st_info>>4 == STB_GLOBAL) break;
}
return def;
}
if (pathname != name) p->shortname = strrchr(p->name, '/')+1;
if (p->tls_image) {
p->tls_id = ++tls_cnt;
- tls_size += p->tls_size + p->tls_align + 8*sizeof(size_t) - 1
- & -4*sizeof(size_t);
+ tls_align = MAXP2(tls_align, p->tls_align);
+ tls_offset += p->tls_size + p->tls_align - 1;
+ tls_offset -= (tls_offset + (uintptr_t)p->tls_image)
+ & (p->tls_align-1);
+ p->tls_offset = tls_offset;
p->new_dtv = (void *)(-sizeof(size_t) &
(uintptr_t)(p->name+strlen(p->name)+sizeof(size_t)));
p->new_tls = (void *)(p->new_dtv + n_th*(tls_cnt+1));
p->map_len = max_addr - min_addr;
}
+static void do_fini()
+{
+ struct dso *p;
+ size_t dyn[DYN_CNT] = {0};
+ for (p=fini_head; p; p=p->fini_next) {
+ if (!p->constructed) continue;
+ decode_vec(p->dynv, dyn, DYN_CNT);
+ ((void (*)(void))(p->base + dyn[DT_FINI]))();
+ }
+}
+
static void do_init_fini(struct dso *p)
{
size_t dyn[DYN_CNT] = {0};
+ int need_locking = __libc.threads_minus_1;
+ /* Allow recursive calls that arise when a library calls
+ * dlopen from one of its constructors, but block any
+ * other threads until all ctors have finished. */
+ if (need_locking) pthread_mutex_lock(&init_fini_lock);
for (; p; p=p->prev) {
- if (p->constructed) return;
+ if (p->constructed) continue;
+ p->constructed = 1;
decode_vec(p->dynv, dyn, DYN_CNT);
- if (dyn[0] & (1<<DT_FINI))
- atexit((void (*)(void))(p->base + dyn[DT_FINI]));
+ if (dyn[0] & (1<<DT_FINI)) {
+ p->fini_next = fini_head;
+ fini_head = p;
+ }
if (dyn[0] & (1<<DT_INIT))
((void (*)(void))(p->base + dyn[DT_INIT]))();
- p->constructed = 1;
}
+ if (need_locking) pthread_mutex_unlock(&init_fini_lock);
}
void _dl_debug_state(void)
void *__copy_tls(unsigned char *mem)
{
+ pthread_t td;
struct dso *p;
+
+ if (!tls_cnt) return mem;
+
void **dtv = (void *)mem;
dtv[0] = (void *)tls_cnt;
- mem = (void *)(dtv + tls_cnt + 1);
- for (p=tail; p; p=p->prev) {
+
+ mem += __libc.tls_size - sizeof(struct pthread);
+ mem -= (uintptr_t)mem & (tls_align-1);
+ mem -= tls_start;
+ td = (pthread_t)mem;
+
+ for (p=head; p; p=p->next) {
if (!p->tls_id) continue;
- mem += -p->tls_len & (4*sizeof(size_t)-1);
- mem += ((uintptr_t)p->tls_image - (uintptr_t)mem)
- & (p->tls_align-1);
- dtv[p->tls_id] = mem;
- memcpy(mem, p->tls_image, p->tls_len);
- mem += p->tls_size;
+ dtv[p->tls_id] = mem - p->tls_offset;
+ memcpy(dtv[p->tls_id], p->tls_image, p->tls_len);
}
- ((pthread_t)mem)->dtv = dtv;
- return mem;
+ td->dtv = dtv;
+ return td;
}
void *__tls_get_addr(size_t *v)
(p->tls_size + p->tls_align) * a_fetch_add(&p->new_tls_idx,1);
mem += ((uintptr_t)p->tls_image - (uintptr_t)mem) & (p->tls_align-1);
self->dtv[v[0]] = mem;
- memcpy(mem, p->tls_image, p->tls_size);
+ memcpy(mem, p->tls_image, p->tls_len);
pthread_sigmask(SIG_SETMASK, &set, 0);
return mem + v[1];
}
+static void update_tls_size()
+{
+ size_t below_tp = (1+tls_cnt) * sizeof(void *) + tls_offset;
+ size_t above_tp = sizeof(struct pthread) + tls_start + tls_align;
+ __libc.tls_size = ALIGN(below_tp + above_tp, tls_align);
+}
+
void *__dynlink(int argc, char **argv)
{
size_t aux[AUX_CNT] = {0};
aux[AT_ENTRY] = ehdr->e_entry;
}
if (app->tls_size) {
- app->tls_id = ++tls_cnt;
- tls_size += app->tls_size+app->tls_align + 8*sizeof(size_t)-1
- & -4*sizeof(size_t);
+ app->tls_id = tls_cnt = 1;
+ tls_offset = app->tls_offset = app->tls_size;
+ tls_start = -((uintptr_t)app->tls_image + app->tls_size)
+ & (app->tls_align-1);
+ tls_align = MAXP2(tls_align, app->tls_align);
}
app->global = 1;
app->constructed = 1;
load_deps(app);
make_global(app);
- /* Make an initial pass setting up TLS before performing relocs.
- * This provides the TP-based offset of each DSO's TLS for
- * use in TP-relative relocations. After relocations, we need
- * to copy the TLS images again in case they had relocs. */
- tls_size += sizeof(struct pthread) + 4*sizeof(size_t);
- __libc.tls_size = tls_size;
+ reloc_all(app->next);
+ reloc_all(app);
+
+ update_tls_size();
if (tls_cnt) {
struct dso *p;
void *mem = mmap(0, __libc.tls_size, PROT_READ|PROT_WRITE,
if (mem==MAP_FAILED ||
!__install_initial_tls(__copy_tls(mem))) {
dprintf(2, "%s: Error getting %zu bytes thread-local storage: %m\n",
- argv[0], tls_size);
+ argv[0], __libc.tls_size);
_exit(127);
}
- for (p=head; p; p=p->next) {
- if (!p->tls_id) continue;
- p->tls_offset = (char *)__pthread_self()
- - (char *)__pthread_self()->dtv[p->tls_id];
- }
}
- reloc_all(app->next);
- reloc_all(app);
-
- /* The initial DTV is located at the base of the memory
- * allocated for TLS. Repeat copying TLS to pick up relocs. */
- if (tls_cnt) __copy_tls((void *)__pthread_self()->dtv);
-
if (ldso_fail) _exit(127);
if (ldd_mode) _exit(0);
if (ssp_used) __init_ssp(auxv);
+ atexit(do_fini);
do_init_fini(tail);
errno = 0;
void *dlopen(const char *file, int mode)
{
struct dso *volatile p, *orig_tail, *next;
- size_t orig_tls_cnt;
+ size_t orig_tls_cnt, orig_tls_offset, orig_tls_align;
size_t i;
int cs;
p = 0;
orig_tls_cnt = tls_cnt;
+ orig_tls_offset = tls_offset;
+ orig_tls_align = tls_align;
orig_tail = tail;
if (setjmp(rtld_fail)) {
free(p);
}
tls_cnt = orig_tls_cnt;
- tls_size = __libc.tls_size;
+ tls_offset = orig_tls_offset;
+ tls_align = orig_tls_align;
tail = orig_tail;
tail->next = 0;
p = 0;
p->global = 1;
}
- __libc.tls_size = tls_size;
+ update_tls_size();
if (ssp_used) __init_ssp(auxv);
_dl_debug_state();
-
- do_init_fini(tail);
+ orig_tail = tail;
end:
__release_ptc();
pthread_rwlock_unlock(&lock);
+ if (p) do_init_fini(orig_tail);
pthread_setcancelstate(cs, 0);
return p;
}