lockdep cmp fn conversions
Subsystems (7)
cover lockdep cmp fn conversions
rationale:*_lock_nested() is fundamentally broken - in order for lockdep to work we need to be able to check that we're following some defined ordering, and it's not possible to define a total ordering of an arbitrary number of objects with only a small fixed size enum.
so it needs to go. awhile back I added the ability to set a comparison function for a lock class, and this is the start of hopefully a slow steady trickle of patches as time allows to convert code to use it.
Kent Overstreet (4):
fs/pipe: Convert to lockdep_cmp_fn
pktcdvd: kill mutex_lock_nested() usage
net: Convert sk->sk_peer_lock to lock_set_cmp_fn_ptr_order()
af_unix: convert to lock_cmp_fn
drivers/block/pktcdvd.c | 8 ++--- fs/pipe.c | 77 ++++++++++++++++------------------------ include/linux/lockdep.h | 3 ++ include/net/af_unix.h | 3 -- kernel/locking/lockdep.c | 6 ++++ net/core/sock.c | 1 + net/unix/af_unix.c | 24 ++++++------- net/unix/diag.c | 2 +- 8 files changed, 55 insertions(+), 69 deletions(-)
Signature · 1 line
2.43.0
PATCH 1/4 fs/pipe: Convert to lockdep_cmp_fn
*_lock_nested() is fundamentally broken; lockdep needs to check lock ordering, but we cannot device a total ordering on an unbounded number of elements with only a few subclasses.
the replacement is to define lock ordering with a proper comparison function.
fs/pipe.c was already doing everything correctly otherwise, nothing much changes here.
--- fs/pipe.c | 81 +++++++++++++++++++++++++------------------------------ 1 file changed, 36 insertions(+), 45 deletions(-)
fs/pipe.c +36 -45
@@ -76,18 +76,20 @@ static unsigned long pipe_user_pages_soft = PIPE_DEF_BUFFERS * INR_OPEN_CUR;* -- Manfred Spraul <[email protected]> 2002-05-09 */ -static void pipe_lock_nested (struct pipe_inode_info *pipe,int subclass) +#define cmp_int (l, r) ((l > r) - (l < r)) + +#ifdef CONFIG_PROVE_LOCKING +static int pipe_lock_cmp_fn (const struct lockdep_map *a, +const struct lockdep_map *b) { -if (pipe->files) -mutex_lock_nested (&pipe->mutex, subclass); +return cmp_int ((unsigned long ) a, (unsigned long ) b); } +#endif void pipe_lock (struct pipe_inode_info *pipe) { -/* -* pipe_lock() nests non-pipe inode locks (for writing to a file) -*/ -pipe_lock_nested (pipe,I_MUTEX_PARENT ); +if (pipe->files) +mutex_lock (&pipe->mutex); }EXPORT_SYMBOL (pipe_lock);
@@ -98,28 +100,16 @@ void pipe_unlock(struct pipe_inode_info *pipe) }EXPORT_SYMBOL (pipe_unlock); -static inline void __pipe_lock (struct pipe_inode_info *pipe) -{ -mutex_lock_nested (&pipe->mutex,I_MUTEX_PARENT ); -} - -static inline void __pipe_unlock (struct pipe_inode_info *pipe) -{ -mutex_unlock (&pipe->mutex); -} -void pipe_double_lock (struct pipe_inode_info *pipe1,struct pipe_inode_info *pipe2) {BUG_ON (pipe1 == pipe2); -if (pipe1 < pipe2) { -pipe_lock_nested (pipe1,I_MUTEX_PARENT ); -pipe_lock_nested (pipe2,I_MUTEX_CHILD ); - }else { -pipe_lock_nested (pipe2,I_MUTEX_PARENT ); -pipe_lock_nested (pipe1,I_MUTEX_CHILD ); - } +if (pipe1 > pipe2) +swap (pipe1, pipe2); + +pipe_lock (pipe1); +pipe_lock (pipe2); } static void anon_pipe_buf_release(struct pipe_inode_info *pipe,
@@ -271,7 +261,7 @@ pipe_read(struct kiocb *iocb, struct iov_iter *to)return 0 ; ret =0 ; -__pipe_lock (pipe); +mutex_lock (&pipe->mutex);/* * We only wake up writers if the pipe was full when we started
@@ -368,7 +358,7 @@ pipe_read(struct kiocb *iocb, struct iov_iter *to) ret = -EAGAIN ;break ; } -__pipe_unlock (pipe); +mutex_unlock (&pipe->mutex);/* * We only get here if we didn't actually read anything.
@@ -400,13 +390,13 @@ pipe_read(struct kiocb *iocb, struct iov_iter *to)if (wait_event_interruptible_exclusive (pipe->rd_wait,pipe_readable (pipe)) <0 )return -ERESTARTSYS ; -__pipe_lock (pipe); +mutex_lock (&pipe->mutex); was_full =pipe_full (pipe->head, pipe->tail, pipe->max_usage); wake_next_reader = true; }if (pipe_empty (pipe->head, pipe->tail)) wake_next_reader = false; -__pipe_unlock (pipe); +mutex_unlock (&pipe->mutex);if (was_full)wake_up_interruptible_sync_poll (&pipe->wr_wait,EPOLLOUT |EPOLLWRNORM );
@@ -462,7 +452,7 @@ pipe_write(struct kiocb *iocb, struct iov_iter *from)if (unlikely (total_len ==0 ))return 0 ; -__pipe_lock (pipe); +mutex_lock (&pipe->mutex);if (!pipe->readers) {send_sig (SIGPIPE , current,0 );
@@ -582,19 +572,19 @@ pipe_write(struct kiocb *iocb, struct iov_iter *from)* after waiting we need to re-check whether the pipe * become empty while we dropped the lock. */ -__pipe_unlock (pipe); +mutex_unlock (&pipe->mutex);if (was_empty)wake_up_interruptible_sync_poll (&pipe->rd_wait,EPOLLIN |EPOLLRDNORM );kill_fasync (&pipe->fasync_readers,SIGIO ,POLL_IN );wait_event_interruptible_exclusive (pipe->wr_wait,pipe_writable (pipe)); -__pipe_lock (pipe); +mutex_lock (&pipe->mutex); was_empty =pipe_empty (pipe->head, pipe->tail); wake_next_writer = true; } out:if (pipe_full (pipe->head, pipe->tail, pipe->max_usage)) wake_next_writer = false; -__pipe_unlock (pipe); +mutex_unlock (&pipe->mutex);/* * If we do do a wakeup event, we do a 'sync' wakeup, because we
@@ -629,7 +619,7 @@ static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)switch (cmd) {case FIONREAD : -__pipe_lock (pipe); +mutex_lock (&pipe->mutex); count =0 ; head = pipe->head; tail = pipe->tail;
@@ -639,16 +629,16 @@ static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) count += pipe->bufs[tail & mask].len; tail++; } -__pipe_unlock (pipe); +mutex_unlock (&pipe->mutex);return put_user (count, (int __user *)arg);#ifdef CONFIG_WATCH_QUEUE case IOC_WATCH_QUEUE_SET_SIZE : {int ret; -__pipe_lock (pipe); +mutex_lock (&pipe->mutex); ret =watch_queue_set_size (pipe, arg); -__pipe_unlock (pipe); +mutex_unlock (&pipe->mutex);return ret; }
@@ -734,7 +724,7 @@ pipe_release(struct inode *inode, struct file *file) {struct pipe_inode_info *pipe = file->private_data; -__pipe_lock (pipe); +mutex_lock (&pipe->mutex);if (file->f_mode &FMODE_READ ) pipe->readers--;if (file->f_mode &FMODE_WRITE )
@@ -747,7 +737,7 @@ pipe_release(struct inode *inode, struct file *file)kill_fasync (&pipe->fasync_readers,SIGIO ,POLL_IN );kill_fasync (&pipe->fasync_writers,SIGIO ,POLL_OUT ); } -__pipe_unlock (pipe); +mutex_unlock (&pipe->mutex);put_pipe_info (inode, pipe);return 0 ;
@@ -759,7 +749,7 @@ pipe_fasync(int fd, struct file *filp, int on)struct pipe_inode_info *pipe = filp->private_data;int retval =0 ; -__pipe_lock (pipe); +mutex_lock (&pipe->mutex);if (filp->f_mode &FMODE_READ ) retval =fasync_helper (fd, filp, on, &pipe->fasync_readers);if ((filp->f_mode &FMODE_WRITE ) && retval >=0 ) {
@@ -768,7 +758,7 @@ pipe_fasync(int fd, struct file *filp, int on)/* this can happen only if on == T */ fasync_helper (-1 , filp,0 , &pipe->fasync_readers); } -__pipe_unlock (pipe); +mutex_unlock (&pipe->mutex);return retval; }
@@ -834,6 +824,7 @@ struct pipe_inode_info *alloc_pipe_info(void) pipe->nr_accounted = pipe_bufs; pipe->user = user;mutex_init (&pipe->mutex); +lock_set_cmp_fn (&pipe->mutex, pipe_lock_cmp_fn,NULL );return pipe; }
@@ -1144,7 +1135,7 @@ static int fifo_open(struct inode *inode, struct file *filp) filp->private_data = pipe;/* OK, we have a pipe and it's pinned down */ -__pipe_lock (pipe); +mutex_lock (&pipe->mutex);/* We can only do regular read/write on fifos */ stream_open (inode, filp);
@@ -1214,7 +1205,7 @@ static int fifo_open(struct inode *inode, struct file *filp) }/* Ok! */ -__pipe_unlock (pipe); +mutex_unlock (&pipe->mutex);return 0 ; err_rd:
@@ -1230,7 +1221,7 @@ static int fifo_open(struct inode *inode, struct file *filp) goto err; err: -__pipe_unlock (pipe); +mutex_unlock (&pipe->mutex);put_pipe_info (inode, pipe);return ret;
@@ -1411,7 +1402,7 @@ long pipe_fcntl(struct file *file, unsigned int cmd, unsigned int arg)if (!pipe)return -EBADF ; -__pipe_lock (pipe); +mutex_lock (&pipe->mutex);switch (cmd) {case F_SETPIPE_SZ :
@@ -1425,7 +1416,7 @@ long pipe_fcntl(struct file *file, unsigned int cmd, unsigned int arg)break ; } -__pipe_unlock (pipe); +mutex_unlock (&pipe->mutex);return ret; }
Signature · 1 line
2.43.0
PATCH 2/4 pktcdvd: kill mutex_lock_nested() usage
In reply to Kent OverstreetUnecessary, we're not actually taking nested locks of the same type.
--- drivers/block/pktcdvd.c | 8 ++++---- fs/pipe.c | 10 +--------- include/linux/lockdep.h | 3 +++ kernel/locking/lockdep.c | 6 ++++++ 4 files changed, 14 insertions(+), 13 deletions(-)
drivers/block/pktcdvd.c +4 -4
@@ -332,7 +332,7 @@ static ssize_t device_map_show(const struct class *c, const struct class_attribu {int n =0 ;int idx; -mutex_lock_nested (&ctl_mutex,SINGLE_DEPTH_NESTING ); +mutex_lock (&ctl_mutex);for (idx =0 ; idx <MAX_WRITERS ; idx++) {struct pktcdvd_device *pd = pkt_devs[idx];if (!pd)
@@ -2639,7 +2639,7 @@ static int pkt_setup_dev(dev_t dev, dev_t* pkt_dev)struct pktcdvd_device *pd;struct gendisk *disk; -mutex_lock_nested (&ctl_mutex,SINGLE_DEPTH_NESTING ); +mutex_lock (&ctl_mutex);for (idx =0 ; idx <MAX_WRITERS ; idx++)if (!pkt_devs[idx])
@@ -2729,7 +2729,7 @@ static int pkt_remove_dev(dev_t pkt_dev)int idx;int ret =0 ; -mutex_lock_nested (&ctl_mutex,SINGLE_DEPTH_NESTING ); +mutex_lock (&ctl_mutex);for (idx =0 ; idx <MAX_WRITERS ; idx++) { pd = pkt_devs[idx];
@@ -2780,7 +2780,7 @@ static void pkt_get_status(struct pkt_ctrl_command *ctrl_cmd) {struct pktcdvd_device *pd; -mutex_lock_nested (&ctl_mutex,SINGLE_DEPTH_NESTING ); +mutex_lock (&ctl_mutex); pd =pkt_find_dev_from_minor (ctrl_cmd->dev_index);if (pd) {
fs/pipe.c +1 -9
@@ -78,14 +78,6 @@ static unsigned long pipe_user_pages_soft = PIPE_DEF_BUFFERS * INR_OPEN_CUR;#define cmp_int (l, r) ((l > r) - (l < r)) -#ifdef CONFIG_PROVE_LOCKING -static int pipe_lock_cmp_fn (const struct lockdep_map *a, -const struct lockdep_map *b) -{ -return cmp_int ((unsigned long ) a, (unsigned long ) b); -} -#endif -void pipe_lock (struct pipe_inode_info *pipe) {if (pipe->files)
@@ -824,7 +816,7 @@ struct pipe_inode_info *alloc_pipe_info(void) pipe->nr_accounted = pipe_bufs; pipe->user = user;mutex_init (&pipe->mutex); -lock_set_cmp_fn (&pipe->mutex, pipe_lock_cmp_fn,NULL ); +lock_set_cmp_fn_ptr_order (&pipe->mutex);return pipe; }
include/linux/lockdep.h +3 -0
@@ -391,6 +391,7 @@ extern int lockdep_is_held(const void *);#endif /* !LOCKDEP */ #ifdef CONFIG_PROVE_LOCKING +int lockdep_ptr_order_cmp_fn (const struct lockdep_map *,const struct lockdep_map *);void lockdep_set_lock_cmp_fn (struct lockdep_map *,lock_cmp_fn ,lock_print_fn );#define lock_set_cmp_fn (lock, ...) lockdep_set_lock_cmp_fn(&(lock)->dep_map, __VA_ARGS__)
@@ -398,6 +399,8 @@ void lockdep_set_lock_cmp_fn(struct lockdep_map *, lock_cmp_fn, lock_print_fn);#define lock_set_cmp_fn (lock, ...) do { } while (0)#endif +#define lock_set_cmp_fn_ptr_order (lock) lock_set_cmp_fn(lock, lockdep_ptr_order_cmp_fn); +enum xhlock_context_t {XHLOCK_HARD ,XHLOCK_SOFT ,
kernel/locking/lockdep.c +6 -0
@@ -4919,6 +4919,12 @@ struct lock_class_key __lockdep_no_validate__;EXPORT_SYMBOL_GPL (__lockdep_no_validate__);#ifdef CONFIG_PROVE_LOCKING +int lockdep_ptr_order_cmp_fn (const struct lockdep_map *a, +const struct lockdep_map *b) +{ +return cmp_int ((unsigned long ) a, (unsigned long ) b); +} +void lockdep_set_lock_cmp_fn (struct lockdep_map *lock,lock_cmp_fn cmp_fn,lock_print_fn print_fn) {
Signature · 1 line
2.43.0
PATCH 3/4 net: Convert sk->sk_peer_lock to lock_set_cmp_fn_ptr_order()
In reply to Kent Overstreet--- net/core/sock.c | 1 + net/unix/af_unix.c | 4 ++-- 2 files changed, 3 insertions(+), 2 deletions(-)
net/core/sock.c +1 -0
@@ -3474,6 +3474,7 @@ void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid) sk->sk_peer_pid =NULL ; sk->sk_peer_cred =NULL ;spin_lock_init (&sk->sk_peer_lock); +lock_set_cmp_fn_ptr_order (&sk->sk_peer_lock); sk->sk_write_pending =0 ; sk->sk_rcvlowat =1 ;
net/unix/af_unix.c +2 -2
@@ -706,10 +706,10 @@ static void copy_peercred(struct sock *sk, struct sock *peersk)if (sk < peersk) {spin_lock (&sk->sk_peer_lock); -spin_lock_nested (&peersk->sk_peer_lock,SINGLE_DEPTH_NESTING ); +spin_lock (&peersk->sk_peer_lock); }else {spin_lock (&peersk->sk_peer_lock); -spin_lock_nested (&sk->sk_peer_lock,SINGLE_DEPTH_NESTING ); +spin_lock (&sk->sk_peer_lock); } old_pid = sk->sk_peer_pid; old_cred = sk->sk_peer_cred;
Signature · 1 line
2.43.0
PATCH 4/4 af_unix: convert to lock_cmp_fn
In reply to Kent OverstreetKill
- unix_state_lock_nested
- _nested usage for net->unx.table.locks[].
replace both with lock_set_cmp_fn_ptr_order(&u->lock).
The lock ordering in sk_diag_dump_icons() looks suspicious; this may turn up a real issue.
--- include/net/af_unix.h | 3 --- net/unix/af_unix.c | 20 ++++++++------------ net/unix/diag.c | 2 +- 3 files changed, 9 insertions(+), 16 deletions(-)
include/net/af_unix.h +0 -3
@@ -48,9 +48,6 @@ struct scm_stat {#define unix_state_lock (s) spin_lock(&unix_sk(s)->lock)#define unix_state_unlock (s) spin_unlock(&unix_sk(s)->lock) -#define unix_state_lock_nested (s) \ - spin_lock_nested(&unix_sk(s)->lock, \ - SINGLE_DEPTH_NESTING)/* The AF_UNIX socket */ struct unix_sock {
net/unix/af_unix.c +8 -12
@@ -170,7 +170,7 @@ static void unix_table_double_lock(struct net *net,swap (hash1, hash2);spin_lock (&net->unx.table.locks[hash1]); -spin_lock_nested (&net->unx.table.locks[hash2],SINGLE_DEPTH_NESTING ); +spin_lock (&net->unx.table.locks[hash2]); } static void unix_table_double_unlock(struct net *net,
@@ -997,6 +997,7 @@ static struct sock *unix_create1(struct net *net, struct socket *sock, int kern, u->path.dentry =NULL ; u->path.mnt =NULL ;spin_lock_init (&u->lock); +lock_set_cmp_fn_ptr_order (&u->lock);atomic_long_set (&u->inflight,0 );INIT_LIST_HEAD (&u->link);mutex_init (&u->iolock);/* single task reading lock */
@@ -1340,17 +1341,11 @@ static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)static void unix_state_double_lock (struct sock *sk1,struct sock *sk2) { -if (unlikely (sk1 == sk2) || !sk2) { -unix_state_lock (sk1); -return ; - } -if (sk1 < sk2) { +if (sk1 > sk2) +swap (sk1, sk2); +if (sk1 && sk1 != sk2)unix_state_lock (sk1); -unix_state_lock_nested (sk2); - }else { -unix_state_lock (sk2); -unix_state_lock_nested (sk1); - } +unix_state_lock (sk2); }static void unix_state_double_unlock (struct sock *sk1,struct sock *sk2)
@@ -1591,7 +1586,7 @@ static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr, goto out_unlock; } -unix_state_lock_nested (sk); +unix_state_lock (sk);if (sk->sk_state != st) {unix_state_unlock (sk);
@@ -3575,6 +3570,7 @@ static int __net_init unix_net_init(struct net *net)for (i =0 ; i <UNIX_HASH_SIZE ; i++) {spin_lock_init (&net->unx.table.locks[i]); +lock_set_cmp_fn_ptr_order (&net->unx.table.locks[i]);INIT_HLIST_HEAD (&net->unx.table.buckets[i]); }
net/unix/diag.c +1 -1
@@ -84,7 +84,7 @@ static int sk_diag_dump_icons(struct sock *sk, struct sk_buff *nlskb)* queue lock. With the other's queue locked it's * OK to lock the state. */ -unix_state_lock_nested (req); +unix_state_lock (req); peer =unix_sk (req)->peer; buf[i++] = (peer ?sock_i_ino (peer) :0 );unix_state_unlock (req);
Signature · 1 line
2.43.0
kernel test robot
build botRe: [PATCH 2/4] pktcdvd: kill mutex_lock_nested() usage
In reply to Kent Overstreetkernel test robot noticed the following build errors:
Hi Kent,
kernel test robot noticed the following build errors:
[auto build test ERROR on tip/locking/core]
[also build test ERROR on net/main net-next/main linus/master v6.8-rc1 next-20240125] [If your patch is applied to the wrong git tree, kindly drop us a note. And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]
url: https://github.com/intel-lab-lkp/linux/commits/Kent-Overstreet/fs-pipe-Convert-to-lockdep_cmp_fn/20240127-101832 base: tip/locking/core patch link: https://lore.kernel.org/r/20240127020833.487907-3-kent.overstreet%40linux.dev patch subject: [PATCH 2/4] pktcdvd: kill mutex_lock_nested() usage config: mips-randconfig-r063-20240127 (https://download.01.org/0day-ci/archive/20240128/[email protected]/config) compiler: clang version 15.0.7 (https://github.com/llvm/llvm-project.git 8dfdcc7b7bf66834a761bd8de445840ef68e4d1a) reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20240128/[email protected]/reproduce)
If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <[email protected]>
| Closes: https://lore.kernel.org/oe-kbuild-all/[email protected]/
All errors (new ones prefixed by >>):
>> kernel/locking/lockdep.c:4925:9: error: call to undeclared function 'cmp_int'; ISO C99 and later do not support implicit function declarations [-Werror,-Wimplicit-function-declaration]
return cmp_int((unsigned long) a, (unsigned long) b);
^
kernel/locking/lockdep.c:4925:9: note: did you mean 'smp_init'?
include/linux/smp.h:225:20: note: 'smp_init' declared here
static inline void smp_init(void) { }
^1 error generated.
vim +/cmp_int +4925 kernel/locking/lockdep.c
4920
4921 #ifdef CONFIG_PROVE_LOCKING
4922 int lockdep_ptr_order_cmp_fn(const struct lockdep_map *a,
4923 const struct lockdep_map *b)
4924 {
> 4925 return cmp_int((unsigned long) a, (unsigned long) b);
4926 }
4927
Signature · 2 lines
0-DAY CI Kernel Test Service https://github.com/intel/lkp-tests/wiki
kernel test robot
build botRe: [PATCH 2/4] pktcdvd: kill mutex_lock_nested() usage
In reply to Kent Overstreetkernel test robot noticed the following build errors:
Hi Kent,
kernel test robot noticed the following build errors:
[auto build test ERROR on tip/locking/core]
[also build test ERROR on net/main net-next/main linus/master v6.8-rc1 next-20240125] [cannot apply to horms-ipvs/master]
[If your patch is applied to the wrong git tree, kindly drop us a note. And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]
url: https://github.com/intel-lab-lkp/linux/commits/Kent-Overstreet/fs-pipe-Convert-to-lockdep_cmp_fn/20240127-101832 base: tip/locking/core patch link: https://lore.kernel.org/r/20240127020833.487907-3-kent.overstreet%40linux.dev patch subject: [PATCH 2/4] pktcdvd: kill mutex_lock_nested() usage config: x86_64-randconfig-161-20240127 (https://download.01.org/0day-ci/archive/20240128/[email protected]/config) compiler: gcc-9 (Debian 9.3.0-22) 9.3.0 reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20240128/[email protected]/reproduce)
If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <[email protected]>
| Closes: https://lore.kernel.org/oe-kbuild-all/[email protected]/
All errors (new ones prefixed by >>):
kernel/locking/lockdep.c: In function 'lockdep_ptr_order_cmp_fn':
>> kernel/locking/lockdep.c:4925:9: error: implicit declaration of function 'cmp_int'; did you mean 'smp_init'? [-Werror=implicit-function-declaration]
4925 | return cmp_int((unsigned long) a, (unsigned long) b);
| ^~~~~~~
| smp_initcc1: some warnings being treated as errors
vim +4925 kernel/locking/lockdep.c
4920
4921 #ifdef CONFIG_PROVE_LOCKING
4922 int lockdep_ptr_order_cmp_fn(const struct lockdep_map *a,
4923 const struct lockdep_map *b)
4924 {
> 4925 return cmp_int((unsigned long) a, (unsigned long) b);
4926 }
4927
Signature · 2 lines
0-DAY CI Kernel Test Service https://github.com/intel/lkp-tests/wiki
Re: [PATCH 4/4] af_unix: convert to lock_cmp_fn
In reply to Kent OverstreetFrom: Kent Overstreet <[email protected]>
Date: Fri, 26 Jan 2024 21:08:31 -0500
Quoted 8 lines
Kill
-unix_state_lock_nested
-_nestedusage for net->unx.table.locks[].replace both with
lock_set_cmp_fn_ptr_order(&u->lock).The lock ordering in
sk_diag_dump_icons()looks suspicious; this may turn up a real issue.
Yes, you cannot use lock_cmp_fn() for unix_state_lock_nested().
The lock order in sk_diag_dump_icons() is
listening socket -> child socket in the listener's queue
, and the inverse order never happens. ptr comparison does not make sense in this case, and lockdep will complain about false positive.
Quoted 98 lines
Signed-off-by Kent Overstreet <[email protected]>--- include/net/af_unix.h | 3 --- net/unix/af_unix.c | 20 ++++++++------------ net/unix/diag.c | 2 +- 3 files changed, 9 insertions(+), 16 deletions(-)include/net/af_unix.h +0
-3@@ -48,9 +48,6 @@ struct scm_stat {#define unix_state_lock (s) spin_lock(&unix_sk(s)->lock)#define unix_state_unlock (s) spin_unlock(&unix_sk(s)->lock) -#define unix_state_lock_nested (s) \ - spin_lock_nested(&unix_sk(s)->lock, \ - SINGLE_DEPTH_NESTING)/* The AF_UNIX socket */ struct unix_sock {net/unix/af_unix.c +8
-12@@ -170,7 +170,7 @@ static void unix_table_double_lock(struct net *net,swap (hash1, hash2);spin_lock (&net->unx.table.locks[hash1]); -spin_lock_nested (&net->unx.table.locks[hash2],SINGLE_DEPTH_NESTING ); +spin_lock (&net->unx.table.locks[hash2]); } static void unix_table_double_unlock(struct net *net,@@ -997,6 +997,7 @@ static struct sock *unix_create1(struct net *net, struct socket *sock, int kern, u->path.dentry =NULL ; u->path.mnt =NULL ;spin_lock_init (&u->lock); +lock_set_cmp_fn_ptr_order (&u->lock);atomic_long_set (&u->inflight,0 );INIT_LIST_HEAD (&u->link);mutex_init (&u->iolock);/* single task reading lock */ @@ -1340,17 +1341,11 @@ static int unix_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)static void unix_state_double_lock (struct sock *sk1,struct sock *sk2) { -if (unlikely (sk1 == sk2) || !sk2) { -unix_state_lock (sk1); -return ; - } -if (sk1 < sk2) { +if (sk1 > sk2) +swap (sk1, sk2); +if (sk1 && sk1 != sk2)unix_state_lock (sk1); -unix_state_lock_nested (sk2); - }else { -unix_state_lock (sk2); -unix_state_lock_nested (sk1); - } +unix_state_lock (sk2); }static void unix_state_double_unlock (struct sock *sk1,struct sock *sk2)@@ -1591,7 +1586,7 @@ static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr, goto out_unlock; } -unix_state_lock_nested (sk); +unix_state_lock (sk);if (sk->sk_state != st) {unix_state_unlock (sk);@@ -3575,6 +3570,7 @@ static int __net_init unix_net_init(struct net *net)for (i =0 ; i <UNIX_HASH_SIZE ; i++) {spin_lock_init (&net->unx.table.locks[i]); +lock_set_cmp_fn_ptr_order (&net->unx.table.locks[i]);INIT_HLIST_HEAD (&net->unx.table.buckets[i]); }net/unix/diag.c +1
-1@@ -84,7 +84,7 @@ static int sk_diag_dump_icons(struct sock *sk, struct sk_buff *nlskb)* queue lock. With the other's queue locked it's * OK to lock the state. */ -unix_state_lock_nested (req); +unix_state_lock (req); peer =unix_sk (req)->peer; buf[i++] = (peer ?sock_i_ino (peer) :0 );unix_state_unlock (req);Signature · 1 line
2.43.0
Re: [PATCH 3/4] net: Convert sk->sk_peer_lock to lock_set_cmp_fn_ptr_order()
In reply to Kent OverstreetFrom: Kent Overstreet <[email protected]>
Date: Fri, 26 Jan 2024 21:08:30 -0500
Quoted 24 lines
Signed-off-by Kent Overstreet <[email protected]>--- net/core/sock.c | 1 + net/unix/af_unix.c | 4 ++-- 2 files changed, 3 insertions(+), 2 deletions(-)net/core/sock.c +1
-0@@ -3474,6 +3474,7 @@ void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid) sk->sk_peer_pid =NULL ; sk->sk_peer_cred =NULL ;spin_lock_init (&sk->sk_peer_lock); +lock_set_cmp_fn_ptr_order (&sk->sk_peer_lock); sk->sk_write_pending =0 ; sk->sk_rcvlowat =1 ;net/unix/af_unix.c +0
-0
net/unix/af_unix.c static void copy_peercred(struct sock *sk, struct sock *peersk) · line 713
Quoted 10 lines
net/unix/af_unix.c +2
-2@@ -706,10 +706,10 @@ static void copy_peercred(struct sock *sk, struct sock *peersk)if (sk < peersk) {spin_lock (&sk->sk_peer_lock); -spin_lock_nested (&peersk->sk_peer_lock,SINGLE_DEPTH_NESTING ); +spin_lock (&peersk->sk_peer_lock); }else {spin_lock (&peersk->sk_peer_lock); -spin_lock_nested (&sk->sk_peer_lock,SINGLE_DEPTH_NESTING ); +spin_lock (&sk->sk_peer_lock); }
hmm.. I think we need not hold two locks here in the first place. Let me post patches.
Thanks!
Quoted 4 lines
net/unix/af_unix.c +0
-0old_pid = sk->sk_peer_pid; old_cred = sk->sk_peer_cred;Signature · 1 line
2.43.0
Re: [PATCH 4/4] af_unix: convert to lock_cmp_fn
In reply to Kuniyuki IwashimaQuoted 11 lines · Kuniyuki Iwashima
From: Kent Overstreet <[email protected]>
Date: Fri, 26 Jan 2024 21:08:31 -0500Quoted 8 lines
Kill
-unix_state_lock_nested
-_nestedusage for net->unx.table.locks[].replace both with
lock_set_cmp_fn_ptr_order(&u->lock).The lock ordering in
sk_diag_dump_icons()looks suspicious; this may turn up a real issue.
Quoted 8 lines
Yes, you cannot use
lock_cmp_fn()forunix_state_lock_nested().The lock order in
sk_diag_dump_icons()islistening socket -> child socket in the listener's queue
, and the inverse order never happens. ptr comparison does not make sense in this case, and lockdep will complain about false positive.
Is that a real lock ordering? Is this parent -> child relationship well defined?
If it is, we should be able to write a lock_cmp_fn for it, as long as it's not some handwavy "this will never happen but _nested won't check for it" like I saw elsewhere in the net code... :)
Re: [PATCH 4/4] af_unix: convert to lock_cmp_fn
From: Kent Overstreet <[email protected]>
Date: Sun, 28 Jan 2024 14:38:02 -0500
Quoted 21 lines · Kuniyuki Iwashima
Quoted 11 lines
From: Kent Overstreet <[email protected]>
Date: Fri, 26 Jan 2024 21:08:31 -0500Quoted 8 lines
Kill
-unix_state_lock_nested
-_nestedusage for net->unx.table.locks[].replace both with
lock_set_cmp_fn_ptr_order(&u->lock).The lock ordering in
sk_diag_dump_icons()looks suspicious; this may turn up a real issue.Quoted 8 lines
Yes, you cannot use
lock_cmp_fn()forunix_state_lock_nested().The lock order in
sk_diag_dump_icons()islistening socket -> child socket in the listener's queue
, and the inverse order never happens. ptr comparison does not make sense in this case, and lockdep will complain about false positive.
Quoted 6 lines
Is that a real lock ordering? Is this parent -> child relationship well defined?
If it is, we should be able to write a
lock_cmp_fnfor it, as long as it's not some handwavy "this will never happen but_nestedwon't check for it" like I saw elsewhere in the net code... :)
The problem would be there's no handy way to detect the relationship except for iterating the queue again.
---8<---
static int unix_state_lock_cmp_fn(const struct lockdep_map *_a,
const struct lockdep_map *_b)
{
const struct unix_sock *a = container_of(_a, struct unix_sock, lock.dep_map);
const struct unix_sock *b = container_of(_b, struct unix_sock, lock.dep_map);
if (a->sk.sk_state == TCP_LISTEN && b->sk.sk_state == TCP_ESTABLISHED) {
/* check if b is a's cihld */
}
/* otherwise, ptr comparison here. */
}---8<---
This can be resolved by a patch like this, which is in my local tree for another series.
So, after posting the series, I can revisit this and write lock_cmp_fn for u->lock.
---8<---
commit 12d39766b06068fda5987f4e7b4827e8c3273137
Author: Kuniyuki Iwashima <[email protected]> Date: Thu Jan 11 22:36:58 2024 +0000
af_unix: Save listener for embryo socket.
include/net/af_unix.h +1 -0
@@ -82,6 +82,7 @@ struct scm_stat {struct unix_sock {/* WARNING: sk has to be the first member */ struct sock sk; +#define usk_listener sk.__sk_common.skc_unix_listenerstruct unix_address *addr;struct path path;struct mutex iolock, bindlock;
include/net/sock.h +3 -0
@@ -142,6 +142,8 @@ typedef __u64 __bitwise __addrpair;* [union with @skc_incoming_cpu] * @skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number * [union with @skc_incoming_cpu] +* @skc_unix_listener: connection request listener socket for AF_UNIX +* [union with @skc_rxhash] * @skc_refcnt: reference count * * This is the minimal network layer representation of sockets, the header
@@ -227,6 +229,7 @@ struct sock_common {u32 skc_rxhash;u32 skc_window_clamp;u32 skc_tw_snd_nxt;/* struct tcp_timewait_sock */ +struct sock *skc_unix_listener;/* struct unix_sock */ };/* public: */ };
net/unix/af_unix.c +4 -1
@@ -991,6 +991,7 @@ static struct sock *unix_create1(struct net *net, struct socket *sock, int kern, sk->sk_max_ack_backlog = net->unx.sysctl_max_dgram_qlen; sk->sk_destruct = unix_sock_destructor; u =unix_sk (sk); + u->usk_listener =NULL ; u->inflight =0 ; u->path.dentry =NULL ; u->path.mnt =NULL ;
@@ -1612,6 +1613,7 @@ static int unix_stream_connect(struct socket *sock, struct sockaddr *uaddr, newsk->sk_type = sk->sk_type;init_peercred (newsk); newu =unix_sk (newsk); + newu->usk_listener = other;RCU_INIT_POINTER (newsk->sk_wq, &newu->peer_wq); otheru =unix_sk (other);
@@ -1707,8 +1709,8 @@ static int unix_accept(struct socket *sock, struct socket *newsock, int flags, bool kern) {struct sock *sk = sock->sk; -struct sock *tsk;struct sk_buff *skb; +struct sock *tsk;int err; err = -EOPNOTSUPP ;
@@ -1733,6 +1735,7 @@ static int unix_accept(struct socket *sock, struct socket *newsock, int flags, } tsk = skb->sk; +unix_sk (tsk)->usk_listener =NULL ;skb_free_datagram (sk, skb);wake_up_interruptible (&unix_sk (sk)->peer_wait);
---8<---
Re: [PATCH 4/4] af_unix: convert to lock_cmp_fn
Quoted 46 lines · Kuniyuki Iwashima
From: Kent Overstreet <[email protected]>
Date: Sun, 28 Jan 2024 14:38:02 -0500Quoted 21 lines · Kuniyuki Iwashima
Quoted 11 lines
From: Kent Overstreet <[email protected]>
Date: Fri, 26 Jan 2024 21:08:31 -0500Quoted 8 lines
Kill
-unix_state_lock_nested
-_nestedusage for net->unx.table.locks[].replace both with
lock_set_cmp_fn_ptr_order(&u->lock).The lock ordering in
sk_diag_dump_icons()looks suspicious; this may turn up a real issue.Quoted 8 lines
Yes, you cannot use
lock_cmp_fn()forunix_state_lock_nested().The lock order in
sk_diag_dump_icons()islistening socket -> child socket in the listener's queue
, and the inverse order never happens. ptr comparison does not make sense in this case, and lockdep will complain about false positive.
Quoted 6 lines
Is that a real lock ordering? Is this parent -> child relationship well defined?
If it is, we should be able to write a
lock_cmp_fnfor it, as long as it's not some handwavy "this will never happen but_nestedwon't check for it" like I saw elsewhere in the net code... :)The problem would be there's no handy way to detect the relationship except for iterating the queue again.
---8<---
static int unix_state_lock_cmp_fn(const structlockdep_map*_a,
const structlockdep_map*_b){ const struct unix_sock *a = container_of(_a, struct unix_sock, lock.dep_map); const struct unix_sock *b = container_of(_b, struct unix_sock, lock.dep_map); if (a->sk.sk_state == TCP_LISTEN && b->sk.sk_state == TCP_ESTABLISHED) { /* check if b is a's cihld */ } /* otherwise, ptr comparison here. */ }
Quoted 8 lines
---8<---
This can be resolved by a patch like this, which is in my local tree for another series.
So, after posting the series, I can revisit this and write
lock_cmp_fnforu->lock.
Sounds good! Please CC me when you do.
Jan Kara
reviewerRe: [PATCH 1/4] fs/pipe: Convert to lockdep_cmp_fn
In reply to Kent OverstreetQuoted 7 lines · Kent Overstreet
*_lock_nested()is fundamentally broken; lockdep needs to check lock ordering, but we cannot device a total ordering on an unbounded number of elements with only a few subclasses.the replacement is to define lock ordering with a proper comparison function.
Quoted 7 lines
fs/pipe.cwas already doing everything correctly otherwise, nothing much changes here.Cc Alexander Viro <[email protected]>Cc Christian Brauner <[email protected]>Cc Jan Kara <[email protected]>Signed-off-by Kent Overstreet <[email protected]>
I had to digest for a while what this new lockdep lock ordering feature is about. I have one pending question - what is the motivation of this conversion of pipe code? AFAIU we don't have any problems with lockdep annotations on pipe->mutex because there are always only two subclasses?
Honza
Quoted 257 lines
--- fs/pipe.c | 81 +++++++++++++++++++++++++------------------------------ 1 file changed, 36 insertions(+), 45 deletions(-)fs/pipe.c +36
-45@@ -76,18 +76,20 @@ static unsigned long pipe_user_pages_soft = PIPE_DEF_BUFFERS * INR_OPEN_CUR;* -- Manfred Spraul <[email protected]> 2002-05-09 */ -static void pipe_lock_nested (struct pipe_inode_info *pipe,int subclass) +#define cmp_int (l, r) ((l > r) - (l < r)) + +#ifdef CONFIG_PROVE_LOCKING +static int pipe_lock_cmp_fn (const struct lockdep_map *a, +const struct lockdep_map *b) { -if (pipe->files) -mutex_lock_nested (&pipe->mutex, subclass); +return cmp_int ((unsigned long ) a, (unsigned long ) b); } +#endif void pipe_lock (struct pipe_inode_info *pipe) { -/* -* pipe_lock() nests non-pipe inode locks (for writing to a file) -*/ -pipe_lock_nested (pipe,I_MUTEX_PARENT ); +if (pipe->files) +mutex_lock (&pipe->mutex); }EXPORT_SYMBOL (pipe_lock);@@ -98,28 +100,16 @@ void pipe_unlock(struct pipe_inode_info *pipe) }EXPORT_SYMBOL (pipe_unlock); -static inline void __pipe_lock (struct pipe_inode_info *pipe) -{ -mutex_lock_nested (&pipe->mutex,I_MUTEX_PARENT ); -} - -static inline void __pipe_unlock (struct pipe_inode_info *pipe) -{ -mutex_unlock (&pipe->mutex); -} -void pipe_double_lock (struct pipe_inode_info *pipe1,struct pipe_inode_info *pipe2) {BUG_ON (pipe1 == pipe2); -if (pipe1 < pipe2) { -pipe_lock_nested (pipe1,I_MUTEX_PARENT ); -pipe_lock_nested (pipe2,I_MUTEX_CHILD ); - }else { -pipe_lock_nested (pipe2,I_MUTEX_PARENT ); -pipe_lock_nested (pipe1,I_MUTEX_CHILD ); - } +if (pipe1 > pipe2) +swap (pipe1, pipe2); + +pipe_lock (pipe1); +pipe_lock (pipe2); } static void anon_pipe_buf_release(struct pipe_inode_info *pipe,@@ -271,7 +261,7 @@ pipe_read(struct kiocb *iocb, struct iov_iter *to)return 0 ; ret =0 ; -__pipe_lock (pipe); +mutex_lock (&pipe->mutex);/* * We only wake up writers if the pipe was full when we started @@ -368,7 +358,7 @@ pipe_read(struct kiocb *iocb, struct iov_iter *to) ret = -EAGAIN ;break ; } -__pipe_unlock (pipe); +mutex_unlock (&pipe->mutex);/* * We only get here if we didn't actually read anything. @@ -400,13 +390,13 @@ pipe_read(struct kiocb *iocb, struct iov_iter *to)if (wait_event_interruptible_exclusive (pipe->rd_wait,pipe_readable (pipe)) <0 )return -ERESTARTSYS ; -__pipe_lock (pipe); +mutex_lock (&pipe->mutex); was_full =pipe_full (pipe->head, pipe->tail, pipe->max_usage); wake_next_reader = true; }if (pipe_empty (pipe->head, pipe->tail)) wake_next_reader = false; -__pipe_unlock (pipe); +mutex_unlock (&pipe->mutex);if (was_full)wake_up_interruptible_sync_poll (&pipe->wr_wait,EPOLLOUT |EPOLLWRNORM );@@ -462,7 +452,7 @@ pipe_write(struct kiocb *iocb, struct iov_iter *from)if (unlikely (total_len ==0 ))return 0 ; -__pipe_lock (pipe); +mutex_lock (&pipe->mutex);if (!pipe->readers) {send_sig (SIGPIPE , current,0 );@@ -582,19 +572,19 @@ pipe_write(struct kiocb *iocb, struct iov_iter *from)* after waiting we need to re-check whether the pipe * become empty while we dropped the lock. */ -__pipe_unlock (pipe); +mutex_unlock (&pipe->mutex);if (was_empty)wake_up_interruptible_sync_poll (&pipe->rd_wait,EPOLLIN |EPOLLRDNORM );kill_fasync (&pipe->fasync_readers,SIGIO ,POLL_IN );wait_event_interruptible_exclusive (pipe->wr_wait,pipe_writable (pipe)); -__pipe_lock (pipe); +mutex_lock (&pipe->mutex); was_empty =pipe_empty (pipe->head, pipe->tail); wake_next_writer = true; } out:if (pipe_full (pipe->head, pipe->tail, pipe->max_usage)) wake_next_writer = false; -__pipe_unlock (pipe); +mutex_unlock (&pipe->mutex);/* * If we do do a wakeup event, we do a 'sync' wakeup, because we @@ -629,7 +619,7 @@ static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)switch (cmd) {case FIONREAD : -__pipe_lock (pipe); +mutex_lock (&pipe->mutex); count =0 ; head = pipe->head; tail = pipe->tail;@@ -639,16 +629,16 @@ static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) count += pipe->bufs[tail & mask].len; tail++; } -__pipe_unlock (pipe); +mutex_unlock (&pipe->mutex);return put_user (count, (int __user *)arg);#ifdef CONFIG_WATCH_QUEUE case IOC_WATCH_QUEUE_SET_SIZE : {int ret; -__pipe_lock (pipe); +mutex_lock (&pipe->mutex); ret =watch_queue_set_size (pipe, arg); -__pipe_unlock (pipe); +mutex_unlock (&pipe->mutex);return ret; }@@ -734,7 +724,7 @@ pipe_release(struct inode *inode, struct file *file) {struct pipe_inode_info *pipe = file->private_data; -__pipe_lock (pipe); +mutex_lock (&pipe->mutex);if (file->f_mode &FMODE_READ ) pipe->readers--;if (file->f_mode &FMODE_WRITE )@@ -747,7 +737,7 @@ pipe_release(struct inode *inode, struct file *file)kill_fasync (&pipe->fasync_readers,SIGIO ,POLL_IN );kill_fasync (&pipe->fasync_writers,SIGIO ,POLL_OUT ); } -__pipe_unlock (pipe); +mutex_unlock (&pipe->mutex);put_pipe_info (inode, pipe);return 0 ;@@ -759,7 +749,7 @@ pipe_fasync(int fd, struct file *filp, int on)struct pipe_inode_info *pipe = filp->private_data;int retval =0 ; -__pipe_lock (pipe); +mutex_lock (&pipe->mutex);if (filp->f_mode &FMODE_READ ) retval =fasync_helper (fd, filp, on, &pipe->fasync_readers);if ((filp->f_mode &FMODE_WRITE ) && retval >=0 ) {@@ -768,7 +758,7 @@ pipe_fasync(int fd, struct file *filp, int on)/* this can happen only if on == T */ fasync_helper (-1 , filp,0 , &pipe->fasync_readers); } -__pipe_unlock (pipe); +mutex_unlock (&pipe->mutex);return retval; }@@ -834,6 +824,7 @@ struct pipe_inode_info *alloc_pipe_info(void) pipe->nr_accounted = pipe_bufs; pipe->user = user;mutex_init (&pipe->mutex); +lock_set_cmp_fn (&pipe->mutex, pipe_lock_cmp_fn,NULL );return pipe; }@@ -1144,7 +1135,7 @@ static int fifo_open(struct inode *inode, struct file *filp) filp->private_data = pipe;/* OK, we have a pipe and it's pinned down */ -__pipe_lock (pipe); +mutex_lock (&pipe->mutex);/* We can only do regular read/write on fifos */ stream_open (inode, filp);@@ -1214,7 +1205,7 @@ static int fifo_open(struct inode *inode, struct file *filp) }/* Ok! */ -__pipe_unlock (pipe); +mutex_unlock (&pipe->mutex);return 0 ; err_rd:@@ -1230,7 +1221,7 @@ static int fifo_open(struct inode *inode, struct file *filp) goto err; err: -__pipe_unlock (pipe); +mutex_unlock (&pipe->mutex);put_pipe_info (inode, pipe);return ret;@@ -1411,7 +1402,7 @@ long pipe_fcntl(struct file *file, unsigned int cmd, unsigned int arg)if (!pipe)return -EBADF ; -__pipe_lock (pipe); +mutex_lock (&pipe->mutex);switch (cmd) {case F_SETPIPE_SZ :@@ -1425,7 +1416,7 @@ long pipe_fcntl(struct file *file, unsigned int cmd, unsigned int arg)break ; } -__pipe_unlock (pipe); +mutex_unlock (&pipe->mutex);return ret; }Signature · 1 line
2.43.0
Signature · 2 lines
Jan Kara <[email protected]> SUSE Labs, CR
Re: [PATCH 1/4] fs/pipe: Convert to lockdep_cmp_fn
Quoted 17 lines · Jan Kara
Quoted 7 lines · Kent Overstreet
*_lock_nested()is fundamentally broken; lockdep needs to check lock ordering, but we cannot device a total ordering on an unbounded number of elements with only a few subclasses.the replacement is to define lock ordering with a proper comparison function.
Quoted 7 lines
fs/pipe.cwas already doing everything correctly otherwise, nothing much changes here.Cc Alexander Viro <[email protected]>Cc Christian Brauner <[email protected]>Cc Jan Kara <[email protected]>Signed-off-by Kent Overstreet <[email protected]>
Quoted 6 lines
I had to digest for a while what this new lockdep lock ordering feature is about. I have one pending question - what is the motivation of this conversion of pipe code? AFAIU we don't have any problems with lockdep annotations on
pipe->mutexbecause there are always only two subclasses?Honza
Hi,
"Numbers talk - Bullshit walks." (Linus Torvalds)
In things of pipes - I normally benchmark like this (example):
root# cat /dev/sdc | pipebench > /dev/null
Do you have numbers for your patch-series?
Thanks.
BG,
-Sedat-
[1] https://packages.debian.org/pipebench
Quoted 262 lines
Quoted 256 lines
--- fs/pipe.c | 81 +++++++++++++++++++++++++------------------------------ 1 file changed, 36 insertions(+), 45 deletions(-)fs/pipe.c +36
-46@@ -76,18 +76,20 @@ static unsigned long pipe_user_pages_soft = PIPE_DEF_BUFFERS * INR_OPEN_CUR;* -- Manfred Spraul <[email protected]> 2002-05-09 */ -static void pipe_lock_nested (struct pipe_inode_info *pipe,int subclass) +#define cmp_int (l, r) ((l > r) - (l < r)) + +#ifdef CONFIG_PROVE_LOCKING +static int pipe_lock_cmp_fn (const struct lockdep_map *a, +const struct lockdep_map *b) { -if (pipe->files) -mutex_lock_nested (&pipe->mutex, subclass); +return cmp_int ((unsigned long ) a, (unsigned long ) b); } +#endif void pipe_lock (struct pipe_inode_info *pipe) { -/* -* pipe_lock() nests non-pipe inode locks (for writing to a file) -*/ -pipe_lock_nested (pipe,I_MUTEX_PARENT ); +if (pipe->files) +mutex_lock (&pipe->mutex); }EXPORT_SYMBOL (pipe_lock);@@ -98,28 +100,16 @@ void pipe_unlock(struct pipe_inode_info *pipe) }EXPORT_SYMBOL (pipe_unlock); -static inline void __pipe_lock (struct pipe_inode_info *pipe) -{ -mutex_lock_nested (&pipe->mutex,I_MUTEX_PARENT ); -} - -static inline void __pipe_unlock (struct pipe_inode_info *pipe) -{ -mutex_unlock (&pipe->mutex); -} -void pipe_double_lock (struct pipe_inode_info *pipe1,struct pipe_inode_info *pipe2) {BUG_ON (pipe1 == pipe2); -if (pipe1 < pipe2) { -pipe_lock_nested (pipe1,I_MUTEX_PARENT ); -pipe_lock_nested (pipe2,I_MUTEX_CHILD ); - }else { -pipe_lock_nested (pipe2,I_MUTEX_PARENT ); -pipe_lock_nested (pipe1,I_MUTEX_CHILD ); - } +if (pipe1 > pipe2) +swap (pipe1, pipe2); + +pipe_lock (pipe1); +pipe_lock (pipe2); } static void anon_pipe_buf_release(struct pipe_inode_info *pipe,@@ -271,7 +261,7 @@ pipe_read(struct kiocb *iocb, struct iov_iter *to)return 0 ; ret =0 ; -__pipe_lock (pipe); +mutex_lock (&pipe->mutex);/* * We only wake up writers if the pipe was full when we started @@ -368,7 +358,7 @@ pipe_read(struct kiocb *iocb, struct iov_iter *to) ret = -EAGAIN ;break ; } -__pipe_unlock (pipe); +mutex_unlock (&pipe->mutex);/* * We only get here if we didn't actually read anything. @@ -400,13 +390,13 @@ pipe_read(struct kiocb *iocb, struct iov_iter *to)if (wait_event_interruptible_exclusive (pipe->rd_wait,pipe_readable (pipe)) <0 )return -ERESTARTSYS ; -__pipe_lock (pipe); +mutex_lock (&pipe->mutex); was_full =pipe_full (pipe->head, pipe->tail, pipe->max_usage); wake_next_reader = true; }if (pipe_empty (pipe->head, pipe->tail)) wake_next_reader = false; -__pipe_unlock (pipe); +mutex_unlock (&pipe->mutex);if (was_full)wake_up_interruptible_sync_poll (&pipe->wr_wait,EPOLLOUT |EPOLLWRNORM );@@ -462,7 +452,7 @@ pipe_write(struct kiocb *iocb, struct iov_iter *from)if (unlikely (total_len ==0 ))return 0 ; -__pipe_lock (pipe); +mutex_lock (&pipe->mutex);if (!pipe->readers) {send_sig (SIGPIPE , current,0 );@@ -582,19 +572,19 @@ pipe_write(struct kiocb *iocb, struct iov_iter *from)* after waiting we need to re-check whether the pipe * become empty while we dropped the lock. */ -__pipe_unlock (pipe); +mutex_unlock (&pipe->mutex);if (was_empty)wake_up_interruptible_sync_poll (&pipe->rd_wait,EPOLLIN |EPOLLRDNORM );kill_fasync (&pipe->fasync_readers,SIGIO ,POLL_IN );wait_event_interruptible_exclusive (pipe->wr_wait,pipe_writable (pipe)); -__pipe_lock (pipe); +mutex_lock (&pipe->mutex); was_empty =pipe_empty (pipe->head, pipe->tail); wake_next_writer = true; } out:if (pipe_full (pipe->head, pipe->tail, pipe->max_usage)) wake_next_writer = false; -__pipe_unlock (pipe); +mutex_unlock (&pipe->mutex);/* * If we do do a wakeup event, we do a 'sync' wakeup, because we @@ -629,7 +619,7 @@ static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)switch (cmd) {case FIONREAD : -__pipe_lock (pipe); +mutex_lock (&pipe->mutex); count =0 ; head = pipe->head; tail = pipe->tail;@@ -639,16 +629,16 @@ static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) count += pipe->bufs[tail & mask].len; tail++; } -__pipe_unlock (pipe); +mutex_unlock (&pipe->mutex);return put_user (count, (int __user *)arg);#ifdef CONFIG_WATCH_QUEUE case IOC_WATCH_QUEUE_SET_SIZE : {int ret; -__pipe_lock (pipe); +mutex_lock (&pipe->mutex); ret =watch_queue_set_size (pipe, arg); -__pipe_unlock (pipe); +mutex_unlock (&pipe->mutex);return ret; }@@ -734,7 +724,7 @@ pipe_release(struct inode *inode, struct file *file) {struct pipe_inode_info *pipe = file->private_data; -__pipe_lock (pipe); +mutex_lock (&pipe->mutex);if (file->f_mode &FMODE_READ ) pipe->readers--;if (file->f_mode &FMODE_WRITE )@@ -747,7 +737,7 @@ pipe_release(struct inode *inode, struct file *file)kill_fasync (&pipe->fasync_readers,SIGIO ,POLL_IN );kill_fasync (&pipe->fasync_writers,SIGIO ,POLL_OUT ); } -__pipe_unlock (pipe); +mutex_unlock (&pipe->mutex);put_pipe_info (inode, pipe);return 0 ;@@ -759,7 +749,7 @@ pipe_fasync(int fd, struct file *filp, int on)struct pipe_inode_info *pipe = filp->private_data;int retval =0 ; -__pipe_lock (pipe); +mutex_lock (&pipe->mutex);if (filp->f_mode &FMODE_READ ) retval =fasync_helper (fd, filp, on, &pipe->fasync_readers);if ((filp->f_mode &FMODE_WRITE ) && retval >=0 ) {@@ -768,7 +758,7 @@ pipe_fasync(int fd, struct file *filp, int on)/* this can happen only if on == T */ fasync_helper (-1 , filp,0 , &pipe->fasync_readers); } -__pipe_unlock (pipe); +mutex_unlock (&pipe->mutex);return retval; }@@ -834,6 +824,7 @@ struct pipe_inode_info *alloc_pipe_info(void) pipe->nr_accounted = pipe_bufs; pipe->user = user;mutex_init (&pipe->mutex); +lock_set_cmp_fn (&pipe->mutex, pipe_lock_cmp_fn,NULL );return pipe; }@@ -1144,7 +1135,7 @@ static int fifo_open(struct inode *inode, struct file *filp) filp->private_data = pipe;/* OK, we have a pipe and it's pinned down */ -__pipe_lock (pipe); +mutex_lock (&pipe->mutex);/* We can only do regular read/write on fifos */ stream_open (inode, filp);@@ -1214,7 +1205,7 @@ static int fifo_open(struct inode *inode, struct file *filp) }/* Ok! */ -__pipe_unlock (pipe); +mutex_unlock (&pipe->mutex);return 0 ; err_rd:@@ -1230,7 +1221,7 @@ static int fifo_open(struct inode *inode, struct file *filp) goto err; err: -__pipe_unlock (pipe); +mutex_unlock (&pipe->mutex);put_pipe_info (inode, pipe);return ret;@@ -1411,7 +1402,7 @@ long pipe_fcntl(struct file *file, unsigned int cmd, unsigned int arg)if (!pipe)return -EBADF ; -__pipe_lock (pipe); +mutex_lock (&pipe->mutex);switch (cmd) {case F_SETPIPE_SZ :@@ -1425,7 +1416,7 @@ long pipe_fcntl(struct file *file, unsigned int cmd, unsigned int arg)break ; } -__pipe_unlock (pipe); +mutex_unlock (&pipe->mutex);return ret; } --Quoted 2 lines
2.43.0
--
Jan Kara <[email protected]>
SUSE Labs, CR
Re: [PATCH 1/4] fs/pipe: Convert to lockdep_cmp_fn
In reply to Jan KaraQuoted 16 lines · Jan Kara
Quoted 7 lines · Kent Overstreet
*_lock_nested()is fundamentally broken; lockdep needs to check lock ordering, but we cannot device a total ordering on an unbounded number of elements with only a few subclasses.the replacement is to define lock ordering with a proper comparison function.
Quoted 7 lines
fs/pipe.cwas already doing everything correctly otherwise, nothing much changes here.Cc Alexander Viro <[email protected]>Cc Christian Brauner <[email protected]>Cc Jan Kara <[email protected]>Signed-off-by Kent Overstreet <[email protected]>
Quoted 4 lines
I had to digest for a while what this new lockdep lock ordering feature is about. I have one pending question - what is the motivation of this conversion of pipe code? AFAIU we don't have any problems with lockdep annotations on
pipe->mutexbecause there are always only two subclasses?
It's one of the easier conversions to do, and ideally /all/ users of subclasses would go away.
Start with the easier ones, figure out those patterns, then the harder...
Jan Kara
reviewerRe: [PATCH 1/4] fs/pipe: Convert to lockdep_cmp_fn
In reply to Sedat DilekQuoted 27 lines · Sedat Dilek
Quoted 17 lines · Jan Kara
Quoted 7 lines · Kent Overstreet
*_lock_nested()is fundamentally broken; lockdep needs to check lock ordering, but we cannot device a total ordering on an unbounded number of elements with only a few subclasses.the replacement is to define lock ordering with a proper comparison function.
Quoted 7 lines
fs/pipe.cwas already doing everything correctly otherwise, nothing much changes here.Cc Alexander Viro <[email protected]>Cc Christian Brauner <[email protected]>Cc Jan Kara <[email protected]>Signed-off-by Kent Overstreet <[email protected]>Quoted 6 lines
I had to digest for a while what this new lockdep lock ordering feature is about. I have one pending question - what is the motivation of this conversion of pipe code? AFAIU we don't have any problems with lockdep annotations on
pipe->mutexbecause there are always only two subclasses?Honza
Hi,
Quoted 7 lines
"Numbers talk - Bullshit walks." (Linus Torvalds)
In things of pipes - I normally benchmark like this (example):
root# cat
/dev/sdc| pipebench >/dev/nullDo you have numbers for your patch-series?
Sedat AFAIU this patch is not about performance at all but rather about lockdep instrumentation... But maybe I'm missing your point?
Honza
Signature · 2 lines
Jan Kara <[email protected]> SUSE Labs, CR
Re: [PATCH 1/4] fs/pipe: Convert to lockdep_cmp_fn
Quoted 31 lines · Jan Kara
Quoted 29 lines · Sedat Dilek
Quoted 17 lines · Jan Kara
Quoted 7 lines · Kent Overstreet
*_lock_nested()is fundamentally broken; lockdep needs to check lock ordering, but we cannot device a total ordering on an unbounded number of elements with only a few subclasses.the replacement is to define lock ordering with a proper comparison function.
Quoted 7 lines
fs/pipe.cwas already doing everything correctly otherwise, nothing much changes here.Cc Alexander Viro <[email protected]>Cc Christian Brauner <[email protected]>Cc Jan Kara <[email protected]>Signed-off-by Kent Overstreet <[email protected]>Quoted 6 lines
I had to digest for a while what this new lockdep lock ordering feature is about. I have one pending question - what is the motivation of this conversion of pipe code? AFAIU we don't have any problems with lockdep annotations on
pipe->mutexbecause there are always only two subclasses?Honza
Hi,
"Numbers talk - Bullshit walks." (Linus Torvalds)
Quoted 9 lines
Quoted 5 lines
In things of pipes - I normally benchmark like this (example):
root# cat
/dev/sdc| pipebench >/dev/nullDo you have numbers for your patch-series?
Sedat AFAIU this patch is not about performance at all but rather about lockdep instrumentation... But maybe I'm missing your point?
Sorry, I missed the point, Jan.
-Sedat-
Re: [PATCH 1/4] fs/pipe: Convert to lockdep_cmp_fn
In reply to Kent OverstreetQuoted 22 lines · Kent Overstreet
Quoted 16 lines · Jan Kara
Quoted 7 lines · Kent Overstreet
*_lock_nested()is fundamentally broken; lockdep needs to check lock ordering, but we cannot device a total ordering on an unbounded number of elements with only a few subclasses.the replacement is to define lock ordering with a proper comparison function.
Quoted 7 lines
fs/pipe.cwas already doing everything correctly otherwise, nothing much changes here.Cc Alexander Viro <[email protected]>Cc Christian Brauner <[email protected]>Cc Jan Kara <[email protected]>Signed-off-by Kent Overstreet <[email protected]>Quoted 4 lines
I had to digest for a while what this new lockdep lock ordering feature is about. I have one pending question - what is the motivation of this conversion of pipe code? AFAIU we don't have any problems with lockdep annotations on
pipe->mutexbecause there are always only two subclasses?
Quoted 5 lines
It's one of the easier conversions to do, and ideally /all/ users of subclasses would go away.
Start with the easier ones, figure out those patterns, then the harder...
I see, thanks for explanation. So in the pipes case I actually like that the patch makes the locking less obfuscated with lockdep details (to which I'm mostly used to but still ;)). So feel free to add:
for this patch. I'm not 100% convinced it will be always possible to replace subclasses with the new ordering mechanism but I guess time will show.
Honza
Signature · 2 lines
Jan Kara <[email protected]> SUSE Labs, CR