Fix a misaligned load inside ptrace_attach()
cover Fix a misaligned load inside ptrace_attach()
I ran across what I believe is a bug in some asm-generic code while working on the RISC-V Linux port. Essentially the problem is that wait_on_bit() takes a void *, but then perfroms long-aligned operation. As far as I can tell, this bug could manifest on any other architecture that doesn't support misaligned operations and uses this particular asm-generic implementation.
The patch set is split into three parts:
* #1 fixes the bug by making task_struct.jobctl an unsigned long, which ensures wait_on_bit() always ends up with a long-aligned argument.
* #2 changes the prototype of wait_on_bit() and friends to take a "unsigned long *" instead of a "void *", with the intent of ensuring these problems don't happen again.
* #3 is a bit more intrusive: it goes and changes all uses of task_struct.jobctl from int to long.
I'm not sure if #3 has gone too far, but I think #1 and #2 are sane. The cost is making task_struct larger on machines where sizeof(long)>sizeof(int), but since it's so big already this isn't too much cost. I thought about making test_bit() perform byte-aligned accesses to avoid this cost, but since there are very similar looking atomic functions I thought that would be too odd.
PATCH 1/3 Fix a misaligned load inside ptrace_attach()
The misaligned load exception arises when running ptrace_attach() on the RISC-V (which hasn't been upstreamed yet). The problem is that wait_on_bit() takes a void* but then proceeds to call test_bit(), which takes a long*. This allows an int-aligned pointer to be passed to test_bit(), which promptly fails. This will manifest on any other asm-generic port where unaligned loads trap, where sizeof(long) > sizeof(int), and where task_struct.jobctl ends up not being long-aligned.
This patch changes task_struct.jobctl to be a long, which ensures it has the correct alignment.
--- include/linux/sched.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
include/linux/sched.h +1 -1
@@ -1369,7 +1369,7 @@ struct task_struct {int exit_state;int exit_code, exit_signal;int pdeath_signal;/* The signal sent when the parent dies */ -unsigned int jobctl;/* JOBCTL_*, siglock protected */ +unsigned long jobctl;/* JOBCTL_*, siglock protected */ /* Used for emulating ABI behavior of previous Linux versions */ unsigned int personality;
Signature · 1 line
2.0.5
tip-bot for Palmer Dabbelt
merge bot[tip:sched/core] signals, ptrace, sched: Fix a misaligned load inside ptrace_attach()
signals, ptrace, sched: Fix a misaligned load inside ptrace_attach()
Commit-ID: e7cc4173115347bcdaa5de2824dd46ef2c58425f Gitweb: http://git.kernel.org/tip/e7cc4173115347bcdaa5de2824dd46ef2c58425f Author: Palmer Dabbelt <[email protected]> AuthorDate: Thu, 30 Apr 2015 21:19:55 -0700 Committer: Ingo Molnar <[email protected]> CommitDate: Fri, 8 May 2015 12:06:57 +0200
signals, ptrace, sched: Fix a misaligned load inside ptrace_attach()
The misaligned load exception arises when running ptrace_attach() on the RISC-V (which hasn't been upstreamed yet). The problem is that wait_on_bit() takes a void* but then proceeds to call test_bit(), which takes a long*. This allows an int-aligned pointer to be passed to test_bit(), which promptly fails. This will manifest on any other asm-generic port where unaligned loads trap, where sizeof(long) > sizeof(int), and where task_struct.jobctl ends up not being long-aligned.
This patch changes task_struct.jobctl to be a long, which ensures it has the correct alignment.
--- include/linux/sched.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
include/linux/sched.h +1 -1
@@ -1374,7 +1374,7 @@ struct task_struct {int exit_state;int exit_code, exit_signal;int pdeath_signal;/* The signal sent when the parent dies */ -unsigned int jobctl;/* JOBCTL_*, siglock protected */ +unsigned long jobctl;/* JOBCTL_*, siglock protected */ /* Used for emulating ABI behavior of previous Linux versions */ unsigned int personality;
PATCH 2/3 Change wait_on_bit*() to take an unsigned long*, not a void*
In reply to Palmer DabbeltThe implementations of wait_on_bit*() will only work with long-aligned memory on systems that don't support misaligned loads and stores. This patch changes the function prototypes to ensure that the compiler will enforce alignment.
Running
make defconfig
make KFLAGS="-Werror"
seems to indicate that, as of c56fb6564dcd ("Fix a misaligned load inside ptrace_attach()"), there are now no users of non-long-aligned calls to wait_on_bit*(). I additionally tried a few "make randconfig" attempts, none of which failed to compile for this reason.
--- include/linux/wait.h | 17 ++++++++++------- 1 file changed, 10 insertions(+), 7 deletions(-)
include/linux/wait.h +10 -7
@@ -969,7 +969,7 @@ extern int bit_wait_io_timeout(struct wait_bit_key *);* on that signal. */ static inline int -wait_on_bit (void *word,int bit,unsigned mode) +wait_on_bit (unsigned long *word,int bit,unsigned mode) {might_sleep ();if (!test_bit (bit, word))
@@ -994,7 +994,7 @@ wait_on_bit(void *word, int bit, unsigned mode)* on that signal. */ static inline int -wait_on_bit_io (void *word,int bit,unsigned mode) +wait_on_bit_io (unsigned long *word,int bit,unsigned mode) {might_sleep ();if (!test_bit (bit, word))
@@ -1020,7 +1020,8 @@ wait_on_bit_io(void *word, int bit, unsigned mode)* received a signal and the mode permitted wakeup on that signal. */ static inline int -wait_on_bit_timeout (void *word,int bit,unsigned mode,unsigned long timeout) +wait_on_bit_timeout (unsigned long *word,int bit,unsigned mode, +unsigned long timeout) {might_sleep ();if (!test_bit (bit, word))
@@ -1047,7 +1048,8 @@ wait_on_bit_timeout(void *word, int bit, unsigned mode, unsigned long timeout)* on that signal. */ static inline int -wait_on_bit_action (void *word,int bit,wait_bit_action_f *action,unsigned mode) +wait_on_bit_action (unsigned long *word,int bit,wait_bit_action_f *action, +unsigned mode) {might_sleep ();if (!test_bit (bit, word))
@@ -1075,7 +1077,7 @@ wait_on_bit_action(void *word, int bit, wait_bit_action_f *action, unsigned mode* the @mode allows that signal to wake the process. */ static inline int -wait_on_bit_lock (void *word,int bit,unsigned mode) +wait_on_bit_lock (unsigned long *word,int bit,unsigned mode) {might_sleep ();if (!test_and_set_bit (bit, word))
@@ -1099,7 +1101,7 @@ wait_on_bit_lock(void *word, int bit, unsigned mode)* the @mode allows that signal to wake the process. */ static inline int -wait_on_bit_lock_io (void *word,int bit,unsigned mode) +wait_on_bit_lock_io (unsigned long *word,int bit,unsigned mode) {might_sleep ();if (!test_and_set_bit (bit, word))
@@ -1125,7 +1127,8 @@ wait_on_bit_lock_io(void *word, int bit, unsigned mode)* the @mode allows that signal to wake the process. */ static inline int -wait_on_bit_lock_action (void *word,int bit,wait_bit_action_f *action,unsigned mode) +wait_on_bit_lock_action (unsigned long *word,int bit,wait_bit_action_f *action, +unsigned mode) {might_sleep ();if (!test_and_set_bit (bit, word))
Signature · 1 line
2.0.5
tip-bot for Palmer Dabbelt
merge bot[tip:sched/core] sched/wait: Change wait_on_bit*() to take an unsigned long *, not a void *
sched/wait: Change wait_on_bit*() to take an unsigned long *, not a void *
Commit-ID: 7e60598785f30cf3dc9e476cc0fc3feeb37a0c63 Gitweb: http://git.kernel.org/tip/7e60598785f30cf3dc9e476cc0fc3feeb37a0c63 Author: Palmer Dabbelt <[email protected]> AuthorDate: Thu, 30 Apr 2015 21:19:56 -0700 Committer: Ingo Molnar <[email protected]> CommitDate: Fri, 8 May 2015 12:05:41 +0200
sched/wait: Change wait_on_bit*() to take an unsigned long *, not a void *
The implementations of wait_on_bit*() will only work with long-aligned memory on systems that don't support misaligned loads and stores.
This patch changes the function prototypes to ensure that the compiler will enforce alignment.
Running
make defconfig
make KFLAGS="-Werror"
seems to indicate that, as of c56fb6564dcd ("Fix a misaligned load inside ptrace_attach()"), there are now no users of non-long-aligned calls to wait_on_bit*(). I additionally tried a few "make randconfig" attempts, none of which failed to compile for this reason.
--- include/linux/wait.h | 17 ++++++++++------- 1 file changed, 10 insertions(+), 7 deletions(-)
include/linux/wait.h +10 -7
@@ -969,7 +969,7 @@ extern int bit_wait_io_timeout(struct wait_bit_key *);* on that signal. */ static inline int -wait_on_bit (void *word,int bit,unsigned mode) +wait_on_bit (unsigned long *word,int bit,unsigned mode) {might_sleep ();if (!test_bit (bit, word))
@@ -994,7 +994,7 @@ wait_on_bit(void *word, int bit, unsigned mode)* on that signal. */ static inline int -wait_on_bit_io (void *word,int bit,unsigned mode) +wait_on_bit_io (unsigned long *word,int bit,unsigned mode) {might_sleep ();if (!test_bit (bit, word))
@@ -1020,7 +1020,8 @@ wait_on_bit_io(void *word, int bit, unsigned mode)* received a signal and the mode permitted wakeup on that signal. */ static inline int -wait_on_bit_timeout (void *word,int bit,unsigned mode,unsigned long timeout) +wait_on_bit_timeout (unsigned long *word,int bit,unsigned mode, +unsigned long timeout) {might_sleep ();if (!test_bit (bit, word))
@@ -1047,7 +1048,8 @@ wait_on_bit_timeout(void *word, int bit, unsigned mode, unsigned long timeout)* on that signal. */ static inline int -wait_on_bit_action (void *word,int bit,wait_bit_action_f *action,unsigned mode) +wait_on_bit_action (unsigned long *word,int bit,wait_bit_action_f *action, +unsigned mode) {might_sleep ();if (!test_bit (bit, word))
@@ -1075,7 +1077,7 @@ wait_on_bit_action(void *word, int bit, wait_bit_action_f *action, unsigned mode* the @mode allows that signal to wake the process. */ static inline int -wait_on_bit_lock (void *word,int bit,unsigned mode) +wait_on_bit_lock (unsigned long *word,int bit,unsigned mode) {might_sleep ();if (!test_and_set_bit (bit, word))
@@ -1099,7 +1101,7 @@ wait_on_bit_lock(void *word, int bit, unsigned mode)* the @mode allows that signal to wake the process. */ static inline int -wait_on_bit_lock_io (void *word,int bit,unsigned mode) +wait_on_bit_lock_io (unsigned long *word,int bit,unsigned mode) {might_sleep ();if (!test_and_set_bit (bit, word))
@@ -1125,7 +1127,8 @@ wait_on_bit_lock_io(void *word, int bit, unsigned mode)* the @mode allows that signal to wake the process. */ static inline int -wait_on_bit_lock_action (void *word,int bit,wait_bit_action_f *action,unsigned mode) +wait_on_bit_lock_action (unsigned long *word,int bit,wait_bit_action_f *action, +unsigned mode) {might_sleep ();if (!test_and_set_bit (bit, word))
PATCH 3/3 Change all uses of JOBCTL_* from int to long
In reply to Palmer Dabbeltc56fb6564dcd ("Fix a misaligned load inside ptrace_attach()") makes jobctl an "unsigned long". It makes sense to have the masks applied to it match that type. This is currently just a cosmetic change, but it will prevent the mask from being unexpectedly truncated if we ever end up with masks with more bits.
One instance of "signr" is an int, but I left this alone because the mask ensures that it will never overflow.
--- include/linux/sched.h | 18 +++++++++--------- kernel/signal.c | 6 +++--- 2 files changed, 12 insertions(+), 12 deletions(-)
include/linux/sched.h +9 -9
@@ -2077,22 +2077,22 @@ TASK_PFA_CLEAR(SPREAD_SLAB, spread_slab)#define JOBCTL_TRAPPING_BIT 21/* switching to TRACED */ #define JOBCTL_LISTENING_BIT 22/* ptracer is listening for events */ -#define JOBCTL_STOP_DEQUEUED (1 << JOBCTL_STOP_DEQUEUED_BIT) -#define JOBCTL_STOP_PENDING (1 << JOBCTL_STOP_PENDING_BIT) -#define JOBCTL_STOP_CONSUME (1 << JOBCTL_STOP_CONSUME_BIT) -#define JOBCTL_TRAP_STOP (1 << JOBCTL_TRAP_STOP_BIT) -#define JOBCTL_TRAP_NOTIFY (1 << JOBCTL_TRAP_NOTIFY_BIT) -#define JOBCTL_TRAPPING (1 << JOBCTL_TRAPPING_BIT) -#define JOBCTL_LISTENING (1 << JOBCTL_LISTENING_BIT) +#define JOBCTL_STOP_DEQUEUED (1UL << JOBCTL_STOP_DEQUEUED_BIT) +#define JOBCTL_STOP_PENDING (1UL << JOBCTL_STOP_PENDING_BIT) +#define JOBCTL_STOP_CONSUME (1UL << JOBCTL_STOP_CONSUME_BIT) +#define JOBCTL_TRAP_STOP (1UL << JOBCTL_TRAP_STOP_BIT) +#define JOBCTL_TRAP_NOTIFY (1UL << JOBCTL_TRAP_NOTIFY_BIT) +#define JOBCTL_TRAPPING (1UL << JOBCTL_TRAPPING_BIT) +#define JOBCTL_LISTENING (1UL << JOBCTL_LISTENING_BIT)#define JOBCTL_TRAP_MASK (JOBCTL_TRAP_STOP | JOBCTL_TRAP_NOTIFY)#define JOBCTL_PENDING_MASK (JOBCTL_STOP_PENDING | JOBCTL_TRAP_MASK)extern bool task_set_jobctl_pending (struct task_struct *task, -unsigned int mask); +unsigned long mask);extern void task_clear_jobctl_trapping (struct task_struct *task);extern void task_clear_jobctl_pending (struct task_struct *task, -unsigned int mask); +unsigned long mask);static inline void rcu_copy_process (struct task_struct *p) {
kernel/signal.c +3 -3
@@ -245,7 +245,7 @@ static inline void print_dropped_signal(int sig)* RETURNS: * %true if @mask is set, %false if made noop because @task was dying. */ -bool task_set_jobctl_pending (struct task_struct *task,unsigned int mask) +bool task_set_jobctl_pending (struct task_struct *task,unsigned long mask) {BUG_ON (mask & ~(JOBCTL_PENDING_MASK |JOBCTL_STOP_CONSUME |JOBCTL_STOP_SIGMASK |JOBCTL_TRAPPING ));
@@ -297,7 +297,7 @@ void task_clear_jobctl_trapping(struct task_struct *task)* CONTEXT: * Must be called with @task->sighand->siglock held. */ -void task_clear_jobctl_pending (struct task_struct *task,unsigned int mask) +void task_clear_jobctl_pending (struct task_struct *task,unsigned long mask) {BUG_ON (mask & ~JOBCTL_PENDING_MASK );
@@ -2000,7 +2000,7 @@ static bool do_signal_stop(int signr)struct signal_struct *sig = current->signal;if (!(current->jobctl &JOBCTL_STOP_PENDING )) { -unsigned int gstop =JOBCTL_STOP_PENDING |JOBCTL_STOP_CONSUME ; +unsigned long gstop =JOBCTL_STOP_PENDING |JOBCTL_STOP_CONSUME ;struct task_struct *t;/* signr will be recorded in task->jobctl for retries */
Signature · 1 line
2.0.5
tip-bot for Palmer Dabbelt
merge bot[tip:sched/core] signals, sched: Change all uses of JOBCTL_* from 'int' to 'long'
signals, sched: Change all uses of JOBCTL_* from 'int' to 'long'
Commit-ID: b76808e6808e34e7e78131d2b8cb0535622b8e9f Gitweb: http://git.kernel.org/tip/b76808e6808e34e7e78131d2b8cb0535622b8e9f Author: Palmer Dabbelt <[email protected]> AuthorDate: Thu, 30 Apr 2015 21:19:57 -0700 Committer: Ingo Molnar <[email protected]> CommitDate: Fri, 8 May 2015 12:04:36 +0200
signals, sched: Change all uses of JOBCTL_* from 'int' to 'long'
c56fb6564dcd ("Fix a misaligned load inside ptrace_attach()") makes jobctl an "unsigned long". It makes sense to have the masks applied to it match that type. This is currently just a cosmetic change, but it will prevent the mask from being unexpectedly truncated if we ever end up with masks with more bits.
One instance of "signr" is an int, but I left this alone because the mask ensures that it will never overflow.
--- include/linux/sched.h | 18 +++++++++--------- kernel/signal.c | 6 +++--- 2 files changed, 12 insertions(+), 12 deletions(-)
include/linux/sched.h +9 -9
@@ -2082,22 +2082,22 @@ TASK_PFA_CLEAR(SPREAD_SLAB, spread_slab)#define JOBCTL_TRAPPING_BIT 21/* switching to TRACED */ #define JOBCTL_LISTENING_BIT 22/* ptracer is listening for events */ -#define JOBCTL_STOP_DEQUEUED (1 << JOBCTL_STOP_DEQUEUED_BIT) -#define JOBCTL_STOP_PENDING (1 << JOBCTL_STOP_PENDING_BIT) -#define JOBCTL_STOP_CONSUME (1 << JOBCTL_STOP_CONSUME_BIT) -#define JOBCTL_TRAP_STOP (1 << JOBCTL_TRAP_STOP_BIT) -#define JOBCTL_TRAP_NOTIFY (1 << JOBCTL_TRAP_NOTIFY_BIT) -#define JOBCTL_TRAPPING (1 << JOBCTL_TRAPPING_BIT) -#define JOBCTL_LISTENING (1 << JOBCTL_LISTENING_BIT) +#define JOBCTL_STOP_DEQUEUED (1UL << JOBCTL_STOP_DEQUEUED_BIT) +#define JOBCTL_STOP_PENDING (1UL << JOBCTL_STOP_PENDING_BIT) +#define JOBCTL_STOP_CONSUME (1UL << JOBCTL_STOP_CONSUME_BIT) +#define JOBCTL_TRAP_STOP (1UL << JOBCTL_TRAP_STOP_BIT) +#define JOBCTL_TRAP_NOTIFY (1UL << JOBCTL_TRAP_NOTIFY_BIT) +#define JOBCTL_TRAPPING (1UL << JOBCTL_TRAPPING_BIT) +#define JOBCTL_LISTENING (1UL << JOBCTL_LISTENING_BIT)#define JOBCTL_TRAP_MASK (JOBCTL_TRAP_STOP | JOBCTL_TRAP_NOTIFY)#define JOBCTL_PENDING_MASK (JOBCTL_STOP_PENDING | JOBCTL_TRAP_MASK)extern bool task_set_jobctl_pending (struct task_struct *task, -unsigned int mask); +unsigned long mask);extern void task_clear_jobctl_trapping (struct task_struct *task);extern void task_clear_jobctl_pending (struct task_struct *task, -unsigned int mask); +unsigned long mask);static inline void rcu_copy_process (struct task_struct *p) {
kernel/signal.c +3 -3
@@ -245,7 +245,7 @@ static inline void print_dropped_signal(int sig)* RETURNS: * %true if @mask is set, %false if made noop because @task was dying. */ -bool task_set_jobctl_pending (struct task_struct *task,unsigned int mask) +bool task_set_jobctl_pending (struct task_struct *task,unsigned long mask) {BUG_ON (mask & ~(JOBCTL_PENDING_MASK |JOBCTL_STOP_CONSUME |JOBCTL_STOP_SIGMASK |JOBCTL_TRAPPING ));
@@ -297,7 +297,7 @@ void task_clear_jobctl_trapping(struct task_struct *task)* CONTEXT: * Must be called with @task->sighand->siglock held. */ -void task_clear_jobctl_pending (struct task_struct *task,unsigned int mask) +void task_clear_jobctl_pending (struct task_struct *task,unsigned long mask) {BUG_ON (mask & ~JOBCTL_PENDING_MASK );
@@ -2000,7 +2000,7 @@ static bool do_signal_stop(int signr)struct signal_struct *sig = current->signal;if (!(current->jobctl &JOBCTL_STOP_PENDING )) { -unsigned int gstop =JOBCTL_STOP_PENDING |JOBCTL_STOP_CONSUME ; +unsigned long gstop =JOBCTL_STOP_PENDING |JOBCTL_STOP_CONSUME ;struct task_struct *t;/* signr will be recorded in task->jobctl for retries */
Peter Zijlstra
maintainerQuoted 20 lines · Palmer Dabbelt
I ran across what I believe is a bug in some asm-generic code while working on the RISC-V Linux port. Essentially the problem is that
wait_on_bit()takes a void *, but then perfroms long-aligned operation. As far as I can tell, this bug could manifest on any other architecture that doesn't support misaligned operations and uses this particular asm-generic implementation.The patch set is split into three parts:
* #1 fixes the bug by making task_struct.jobctl an unsigned long, which ensures
wait_on_bit()always ends up with a long-aligned argument.* #2 changes the prototype of
wait_on_bit()and friends to take a "unsigned long *" instead of a "void *", with the intent of ensuring these problems don't happen again.* #3 is a bit more intrusive: it goes and changes all uses of task_struct.jobctl from int to long.
Quoted 6 lines
I'm not sure if #3 has gone too far, but I think #1 and #2 are sane. The cost is making
task_structlarger on machines where sizeof(long)>sizeof(int), but since it's so big already this isn't too much cost. I thought about makingtest_bit()perform byte-aligned accesses to avoid this cost, but since there are very similar looking atomic functions I thought that would be too odd.
Fair enough. Thanks!