From 83b5323261bb72313bffcf37476c1b8f0847c736 Mon Sep 17 00:00:00 2001 From: Szabolcs Nagy Date: Wed, 15 Sep 2021 15:16:19 +0100 Subject: [PATCH] elf: Avoid deadlock between pthread_create and ctors [BZ #28357] The fix for bug 19329 caused a regression such that pthread_create can deadlock when concurrent ctors from dlopen are waiting for it to finish. Use a new GL(dl_load_tls_lock) in pthread_create that is not taken around ctors in dlopen. The new lock is also used in __tls_get_addr instead of GL(dl_load_lock). The new lock is held in _dl_open_worker and _dl_close_worker around most of the logic before/after the init/fini routines. When init/fini routines are running then TLS is in a consistent, usable state. In _dl_open_worker the new lock requires catching and reraising dlopen failures that happen in the critical section. The new lock is reinitialized in a fork child, to keep the existing behaviour and it is kept recursive in case malloc interposition or TLS access from signal handlers can retake it. It is not obvious if this is necessary or helps, but avoids changing the preexisting behaviour. The new lock may be more appropriate for dl_iterate_phdr too than GL(dl_load_write_lock), since TLS state of an incompletely loaded module may be accessed. If the new lock can replace the old one, that can be a separate change. Fixes bug 28357. Reviewed-by: Adhemerval Zanella --- elf/dl-close.c | 6 ++ elf/dl-open.c | 35 ++++++++- elf/dl-support.c | 7 ++ elf/dl-tls.c | 16 ++--- elf/rtld.c | 1 + sysdeps/nptl/fork.c | 3 + sysdeps/generic/ldsodefs.h | 9 ++- 10 files changed, 235 insertions(+), 12 deletions(-) --- Upstream-Status: Backport [https://sourceware.org/git/?p=glibc.git;a=patch;h=024a7640ab9ecea80e527f4e4d7f7a1868e952c5] Comment: This patch is refreshed for glibc 2.31. In upstream glibc 2.34 multiple src files are shuffled, updated this patch as per the code present in glibc 2.31. Removed test case. Signed-off-by: Akash Hadke Signed-off-by: Akash Hadke --- diff --git a/elf/dl-close.c b/elf/dl-close.c index 93ff5c96e9..cfe0f1c0c9 100644 --- a/elf/dl-close.c +++ b/elf/dl-close.c @@ -551,6 +551,9 @@ size_t tls_free_end; tls_free_start = tls_free_end = NO_TLS_OFFSET; + /* Protects global and module specitic TLS state. */ + __rtld_lock_lock_recursive (GL(dl_load_tls_lock)); + /* We modify the list of loaded objects. */ __rtld_lock_lock_recursive (GL(dl_load_write_lock)); @@ -786,6 +789,9 @@ GL(dl_tls_static_used) = tls_free_start; } + /* TLS is cleaned up for the unloaded modules. */ + __rtld_lock_unlock_recursive (GL(dl_load_tls_lock)); + #ifdef SHARED /* Auditing checkpoint: we have deleted all objects. */ if (__glibc_unlikely (do_audit)) diff --git a/elf/dl-open.c b/elf/dl-open.c index 5295e931b0..6ea5dd2457 100644 --- a/elf/dl-open.c +++ b/elf/dl-open.c @@ -57,6 +57,9 @@ (non-negative). */ unsigned int original_global_scope_pending_adds; + /* Set to true if the end of dl_open_worker_begin was reached. */ + bool worker_continue; + /* Original parameters to the program and the current environment. */ int argc; char **argv; @@ -473,7 +473,7 @@ } static void -dl_open_worker (void *a) +dl_open_worker_begin (void *a) { struct dl_open_args *args = a; const char *file = args->file; @@ -747,6 +747,36 @@ if (mode & RTLD_GLOBAL) add_to_global_resize (new); + args->worker_continue = true; +} + +static void +dl_open_worker (void *a) +{ + struct dl_open_args *args = a; + + args->worker_continue = false; + + { + /* Protects global and module specific TLS state. */ + __rtld_lock_lock_recursive (GL(dl_load_tls_lock)); + + struct dl_exception ex; + int err = _dl_catch_exception (&ex, dl_open_worker_begin, args); + + __rtld_lock_unlock_recursive (GL(dl_load_tls_lock)); + + if (__glibc_unlikely (ex.errstring != NULL)) + /* Reraise the error. */ + _dl_signal_exception (err, &ex, NULL); + } + + if (!args->worker_continue) + return; + + int mode = args->mode; + struct link_map *new = args->map; + /* Run the initializer functions of new objects. Temporarily disable the exception handler, so that lazy binding failures are fatal. */ diff --git a/elf/dl-support.c b/elf/dl-support.c index 02e2ed72f5..d99c1f1d62 100644 --- a/elf/dl-support.c +++ b/elf/dl-support.c @@ -219,6 +219,13 @@ list of loaded objects while an object is added to or removed from that list. */ __rtld_lock_define_initialized_recursive (, _dl_load_write_lock) +/* This lock protects global and module specific TLS related data. + E.g. it is held in dlopen and dlclose when GL(dl_tls_generation), + GL(dl_tls_max_dtv_idx) or GL(dl_tls_dtv_slotinfo_list) are + accessed and when TLS related relocations are processed for a + module. It was introduced to keep pthread_create accessing TLS + state that is being set up. */ +__rtld_lock_define_initialized_recursive (, _dl_load_tls_lock) #ifdef HAVE_AUX_VECTOR diff --git a/elf/dl-tls.c b/elf/dl-tls.c index d554ae4497..9260d2d696 100644 --- a/elf/dl-tls.c +++ b/elf/dl-tls.c @@ -443,7 +443,7 @@ size_t maxgen = 0; /* Protects global dynamic TLS related state. */ - __rtld_lock_lock_recursive (GL(dl_load_lock)); + __rtld_lock_lock_recursive (GL(dl_load_tls_lock)); /* Check if the current dtv is big enough. */ if (dtv[-1].counter < GL(dl_tls_max_dtv_idx)) @@ -517,7 +517,7 @@ listp = listp->next; assert (listp != NULL); } - __rtld_lock_unlock_recursive (GL(dl_load_lock)); + __rtld_lock_unlock_recursive (GL(dl_load_tls_lock)); /* The DTV version is up-to-date now. */ dtv[0].counter = maxgen; @@ -656,7 +656,7 @@ Here the dtv needs to be updated to new_gen generation count. - This code may be called during TLS access when GL(dl_load_lock) + This code may be called during TLS access when GL(dl_load_tls_lock) is not held. In that case the user code has to synchronize with dlopen and dlclose calls of relevant modules. A module m is relevant if the generation of m <= new_gen and dlclose of m is @@ -778,11 +778,11 @@ if (__glibc_unlikely (the_map->l_tls_offset != FORCED_DYNAMIC_TLS_OFFSET)) { - __rtld_lock_lock_recursive (GL(dl_load_lock)); + __rtld_lock_lock_recursive (GL(dl_load_tls_lock)); if (__glibc_likely (the_map->l_tls_offset == NO_TLS_OFFSET)) { the_map->l_tls_offset = FORCED_DYNAMIC_TLS_OFFSET; - __rtld_lock_unlock_recursive (GL(dl_load_lock)); + __rtld_lock_unlock_recursive (GL(dl_load_tls_lock)); } else if (__glibc_likely (the_map->l_tls_offset != FORCED_DYNAMIC_TLS_OFFSET)) @@ -794,7 +794,7 @@ #else # error "Either TLS_TCB_AT_TP or TLS_DTV_AT_TP must be defined" #endif - __rtld_lock_unlock_recursive (GL(dl_load_lock)); + __rtld_lock_unlock_recursive (GL(dl_load_tls_lock)); dtv[GET_ADDR_MODULE].pointer.to_free = NULL; dtv[GET_ADDR_MODULE].pointer.val = p; @@ -802,7 +802,7 @@ return (char *) p + GET_ADDR_OFFSET; } else - __rtld_lock_unlock_recursive (GL(dl_load_lock)); + __rtld_lock_unlock_recursive (GL(dl_load_tls_lock)); } struct dtv_pointer result = allocate_and_init (the_map); dtv[GET_ADDR_MODULE].pointer = result; @@ -873,7 +873,7 @@ return NULL; dtv_t *dtv = THREAD_DTV (); - /* This may be called without holding the GL(dl_load_lock). Reading + /* This may be called without holding the GL(dl_load_tls_lock). Reading arbitrary gen value is fine since this is best effort code. */ size_t gen = atomic_load_relaxed (&GL(dl_tls_generation)); if (__glibc_unlikely (dtv[0].counter != gen)) diff --git a/elf/rtld.c b/elf/rtld.c index 8d2bba3d43..9642eb9c92 100644 --- a/elf/rtld.c +++ b/elf/rtld.c @@ -283,6 +283,7 @@ #ifdef _LIBC_REENTRANT ._dl_load_lock = _RTLD_LOCK_RECURSIVE_INITIALIZER, ._dl_load_write_lock = _RTLD_LOCK_RECURSIVE_INITIALIZER, + ._dl_load_tls_lock = _RTLD_LOCK_RECURSIVE_INITIALIZER, #endif ._dl_nns = 1, ._dl_ns = diff --git a/sysdeps/nptl/fork.c b/sysdeps/nptl/fork.c index c471f7b15f..021691b9b7 100644 --- a/sysdeps/nptl/fork.c +++ b/sysdeps/nptl/fork.c @@ -125,6 +125,9 @@ /* Reset the lock the dynamic loader uses to protect its data. */ __rtld_lock_initialize (GL(dl_load_lock)); + /* Reset the lock protecting dynamic TLS related data. */ + __rtld_lock_initialize (GL(dl_load_tls_lock)); + /* Run the handlers registered for the child. */ __run_fork_handlers (atfork_run_child, multiple_threads); } diff --git a/sysdeps/generic/ldsodefs.h b/sysdeps/generic/ldsodefs.h index d49529da0d..9ec1511bb0 100644 --- a/sysdeps/generic/ldsodefs.h +++ b/sysdeps/generic/ldsodefs.h @@ -369,6 +369,13 @@ list of loaded objects while an object is added to or removed from that list. */ __rtld_lock_define_recursive (EXTERN, _dl_load_write_lock) + /* This lock protects global and module specific TLS related data. + E.g. it is held in dlopen and dlclose when GL(dl_tls_generation), + GL(dl_tls_max_dtv_idx) or GL(dl_tls_dtv_slotinfo_list) are + accessed and when TLS related relocations are processed for a + module. It was introduced to keep pthread_create accessing TLS + state that is being set up. */ + __rtld_lock_define_recursive (EXTERN, _dl_load_tls_lock) /* Incremented whenever something may have been added to dl_loaded. */ EXTERN unsigned long long _dl_load_adds; @@ -1153,7 +1160,7 @@ /* Add module to slot information data. If DO_ADD is false, only the required memory is allocated. Must be called with GL - (dl_load_lock) acquired. If the function has already been called + (dl_load_tls_lock) acquired. If the function has already been called for the link map L with !do_add, then this function will not raise an exception, otherwise it is possible that it encounters a memory allocation failure. */ -- 2.27.0