Skip to contentSkip to conversation

fasync() BKL pushdown

RFC PATCH In review Jonathan Corbet 3 participants 8 messages Jun 20, 2008 – Jun 27, 2008

Subsystems (10)
lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 fasync() BKL pushdown

When I put together the bkl-removal tree, I didn't add Andi's fasync() patches because the addition of an unlocked_fasync() function was not universally accepted. Recently I decided to take a look and see what would be involved in just pushing the BKL down into fasync() implementations. It turns out that it wasn't too hard.

The majority of fasync() functions just call fasync_helper() with a pointer to an fasync_struct reachable from the file structure. Given that (1) the struct file will not go away while fasync() is running, and (2) the VFS-level fasync() stuff is protected with the Big Fasync Lock, I contend that these simple implementations have no need for the BKL. Once those are filtered out, there's really only a half-dozen places which need to be fixed.

(As an aside, fasync_helper() can return a positive value on success. A number of fasync() implementations remap that to zero, but quite a few others return it directly. That, in turn, will cause fcntl() to return said positive value to user space, which is contrary to what the man page says. There is one user of the positive return value (tty_io.c), so maybe some aspiring janitor would like to clean up the various fasync() functions that just do "return fasync_helper(...);"?)

The set of patches is this:

Jonathan Corbet (7):
mpt: fasync BKL pushdown
i2o: fasync BKL pushdown
tun: fasync BKL pushdown
tty_io: fasync BKL pushdown
Bluetooth VHCI: fasync BKL pushdown
ecryptfs: fasync BKL pushdown
Call fasync() functions without the BKL

And the actual patch is appended. If nobody objects, I'll pull these into the bkl-removal tree shortly.

jon

drivers/bluetooth/hci_vhci.c +6 -3
index 7734bc9..d97700a 100644
@@ -318,18 +318,21 @@ static int vhci_release(struct inode *inode, struct file *file)
 static int vhci_fasync(int fd, struct file *file, int on)
 {
 	struct vhci_data *data = file->private_data;
-	int err;
+	int err = 0;
 
+	lock_kernel();
 	err = fasync_helper(fd, file, on, &data->fasync);
 	if (err < 0)
-		return err;
+		goto out;
 
 	if (on)
 		data->flags |= VHCI_FASYNC;
 	else
 		data->flags &= ~VHCI_FASYNC;
 
-	return 0;
+out:
+	unlock_kernel();
+	return err;
 }
 
 static const struct file_operations vhci_fops = {
drivers/char/tty_io.c +9 -5
index fd182f2..eda2789 100644
@@ -2909,15 +2909,16 @@ static int tty_fasync(int fd, struct file *filp, int on)
 {
 	struct tty_struct *tty;
 	unsigned long flags;
-	int retval;
+	int retval = 0;
 
+	lock_kernel();
 	tty = (struct tty_struct *)filp->private_data;
 	if (tty_paranoia_check(tty, filp->f_path.dentry->d_inode, "tty_fasync"))
-		return 0;
+		goto out;
 
 	retval = fasync_helper(fd, filp, on, &tty->fasync);
 	if (retval <= 0)
-		return retval;
+		goto out;
 
 	if (on) {
 		enum pid_type type;
@@ -2935,12 +2936,15 @@ static int tty_fasync(int fd, struct file *filp, int on)
 		spin_unlock_irqrestore(&tty->ctrl_lock, flags);
 		retval = __f_setown(filp, pid, type, 0);
 		if (retval)
-			return retval;
+			goto out;
 	} else {
 		if (!tty->fasync && !waitqueue_active(&tty->read_wait))
 			tty->minimum_to_wake = N_TTY_BUF_SIZE;
 	}
-	return 0;
+	retval = 0;
+out:
+	unlock_kernel();
+	return retval;
 }
 
 /**
drivers/message/fusion/mptctl.c +5 -1
index e630b50..c594656 100644
@@ -548,11 +548,15 @@ static int
 mptctl_fasync(int fd, struct file *filep, int mode)
 {
 	MPT_ADAPTER	*ioc;
+	int ret;
 
+	lock_kernel();
 	list_for_each_entry(ioc, &ioc_list, list)
 		ioc->aen_event_read_flag=0;
 
-	return fasync_helper(fd, filep, mode, &async_queue);
+	ret = fasync_helper(fd, filep, mode, &async_queue);
+	unlock_kernel();
+	return ret;
 }
 
 static int
drivers/message/i2o/i2o_config.c +6 -4
index 95b4c10..4238de9 100644
@@ -1084,15 +1084,17 @@ static int cfg_fasync(int fd, struct file *fp, int on)
 {
 	ulong id = (ulong) fp->private_data;
 	struct i2o_cfg_info *p;
+	int ret = -EBADF;
 
+	lock_kernel();
 	for (p = open_files; p; p = p->next)
 		if (p->q_id == id)
 			break;
 
-	if (!p)
-		return -EBADF;
-
-	return fasync_helper(fd, fp, on, &p->fasync);
+	if (p)
+		ret = fasync_helper(fd, fp, on, &p->fasync);
+	unlock_kernel();
+	return ret;
 }
 
 static int cfg_release(struct inode *inode, struct file *file)
drivers/net/tun.c +7 -4
index ce5af2a..4c0c597 100644
@@ -782,18 +782,21 @@ static int tun_chr_fasync(int fd, struct file *file, int on)
 
 	DBG(KERN_INFO "%s: tun_chr_fasync %d\n", tun->dev->name, on);
 
+	lock_kernel();
 	if ((ret = fasync_helper(fd, file, on, &tun->fasync)) < 0)
-		return ret;
+		goto out;
 
 	if (on) {
 		ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
 		if (ret)
-			return ret;
+			goto out;
 		tun->flags |= TUN_FASYNC;
 	} else
 		tun->flags &= ~TUN_FASYNC;
-
-	return 0;
+	ret = 0;
+out:
+	unlock_kernel();
+	return ret;
 }
 
 static int tun_chr_open(struct inode *inode, struct file * file)
fs/ecryptfs/file.c +3 -0
index 2258b8f..24749bf 100644
@@ -30,6 +30,7 @@
 #include <linux/security.h>
 #include <linux/compat.h>
 #include <linux/fs_stack.h>
+#include <linux/smp_lock.h>
 #include "ecryptfs_kernel.h"
 
 /**
@@ -277,9 +278,11 @@ static int ecryptfs_fasync(int fd, struct file *file, int flag)
 	int rc = 0;
 	struct file *lower_file = NULL;
 
+	lock_kernel();
 	lower_file = ecryptfs_file_to_lower(file);
 	if (lower_file->f_op && lower_file->f_op->fasync)
 		rc = lower_file->f_op->fasync(fd, lower_file, flag);
+	unlock_kernel();
 	return rc;
 }
 
fs/fcntl.c +0 -3
index bfd7765..330a7d7 100644
@@ -12,7 +12,6 @@
 #include <linux/fdtable.h>
 #include <linux/capability.h>
 #include <linux/dnotify.h>
-#include <linux/smp_lock.h>
 #include <linux/slab.h>
 #include <linux/module.h>
 #include <linux/security.h>
@@ -227,7 +226,6 @@ static int setfl(int fd, struct file * filp, unsigned long arg)
 	if (error)
 		return error;
 
-	lock_kernel();
 	if ((arg ^ filp->f_flags) & FASYNC) {
 		if (filp->f_op && filp->f_op->fasync) {
 			error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
@@ -238,7 +236,6 @@ static int setfl(int fd, struct file * filp, unsigned long arg)
 
 	filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
  out:
-	unlock_kernel();
 	return error;
 }
 
#1 lore raw
Quoted 5 lines · Jonathan Corbet

The majority of fasync() functions just call fasync_helper() with a pointer to an fasync_struct reachable from the file structure. Given that (1) the struct file will not go away while fasync() is running, and (2) the VFS-level fasync() stuff is protected with the Big Fasync Lock, I contend that these simple implementations have no need for the BKL.

Not necessarily true, they might require BKL still for fd live time issues.

-Andi

#2 lore raw
Quoted 11 lines · On Fri, 20 Jun 2008 19:55:03 +0200 Andi Kleen <[email protected]> wrote:
Quoted 6 lines · Jonathan Corbet

The majority of fasync() functions just call fasync_helper() with a pointer to an fasync_struct reachable from the file structure. Given that (1) the struct file will not go away while fasync() is running, and (2) the VFS-level fasync() stuff is protected with the Big Fasync Lock, I contend that these simple implementations have no need for the BKL.

Not necessarily true, they might require BKL still for fd live time issues.

Could you help me out a bit here? I'm even slower than usual when it comes to VFS stuff. As far as I can tell, the given file cannot go away during the call to fasync(), as sys_fcntl() holds a reference on it. Are you saying that something else can happen during that time?

Thanks,

jon

#3 lore raw
Quoted 11 lines · Jonathan Corbet
Quoted 8 lines · On Fri, 20 Jun 2008 19:55:03 +0200 Andi Kleen <[email protected]> wrote:
Quoted 6 lines · Jonathan Corbet

The majority of fasync() functions just call fasync_helper() with a pointer to an fasync_struct reachable from the file structure. Given that (1) the struct file will not go away while fasync() is running, and (2) the VFS-level fasync() stuff is protected with the Big Fasync Lock, I contend that these simple implementations have no need for the BKL.

Quoted 7 lines
Quoted 2 lines

Not necessarily true, they might require BKL still for fd live time issues.

Could you help me out a bit here? I'm even slower than usual when it comes to VFS stuff. As far as I can tell, the given file cannot go away during the call to fasync(), as sys_fcntl() holds a reference on it. Are you saying that something else can happen during that time?

Some devices do state change even when the reference count is > 0. Would need to double check it's all ok with the fasync list.

Anyways I did this auditing for the cases where I used unlocked_ioctl [but I think I wanted to redo it because i wasn't 100% sure anymore] and I haven't done it at all for the cases that weren't converted.

-Andi

#4 lore raw

RFC PATCH 1/1 fasync() BKL pushdown (take 2)

Quoted 2 lines · On Fri, 20 Jun 2008 21:12:51 +0200 Andi Kleen <[email protected]> wrote:

Some devices do state change even when the reference count is > 0. Would need to double check it's all ok with the fasync list.

OK, I've gone over all of the fasync() definitions again with an eye toward convincing myself that the fasync list would not get cleared, freed, or otherwise molested if fasync() runs without BKL protection. I focused especially on other code (open(), ioctl()) which might still run with the BKL. The result was two more pushdowns in spots where I wasn't sure; chance are both are unnecessary.

The new fasync pushdown tree contains:

Jonathan Corbet (9):
mpt: fasync BKL pushdown
i2o: fasync BKL pushdown
tun: fasync BKL pushdown
tty_io: fasync BKL pushdown
Bluetooth VHCI: fasync BKL pushdown
ecryptfs: fasync BKL pushdown
ipmi: fasync BKL pushdown
snd/PCM: fasync BKL pushdown
Call fasync() functions without the BKL

 drivers/bluetooth/hci_vhci.c     |    9 ++++++---
 drivers/char/ipmi/ipmi_devintf.c |    2 ++
 drivers/char/tty_io.c            |   14 +++++++++-----
 drivers/message/fusion/mptctl.c  |    6 +++++-
 drivers/message/i2o/i2o_config.c |   10 ++++++----
 drivers/net/tun.c                |   11 +++++++----
 fs/ecryptfs/file.c               |    3 +++
 fs/fcntl.c                       |    3 ---
 sound/core/pcm_native.c          |    8 ++++++--
 9 files changed, 44 insertions(+), 22 deletions(-)

Once again, the combined patch is small, so I'll just attach it. Anybody have a reason why I should not fold this work into the bkl-removal tree?

Thanks,

jon

drivers/bluetooth/hci_vhci.c +0 -0
index 7734bc9..d97700a 100644
@@ -318,18 +318,21 @@ static int vhci_release(struct inode *inode,

struct file *file) static int vhci_fasync(int fd, struct file *file, int on) {
struct vhci_data *data = file->private_data;
- int err;
+ int err = 0;

+ lock_kernel();
err = fasync_helper(fd, file, on, &data->fasync); if (err < 0)
- return err;
+ goto out;

if (on)
data->flags |= VHCI_FASYNC;
else
data->flags &= ~VHCI_FASYNC;

- return 0;
+out:
+ unlock_kernel();
+ return err;
}

static const struct file_operations vhci_fops = {

(quoted diff) +0 -0

b/drivers/char/ipmi/ipmi_devintf.c index c816656..c11a404 100644

drivers/char/ipmi/ipmi_devintf.c +0 -0
@@ -101,7 +101,9 @@ static int ipmi_fasync(int fd, struct file *file,

int on) struct ipmi_file_private *priv = file->private_data;
int result;

+ lock_kernel(); /* could race against open() otherwise */
result = fasync_helper(fd, file, on, &priv->fasync_queue);
+ unlock_kernel();

return (result);
}

drivers/char/tty_io.c +0 -0
index fd182f2..eda2789 100644
@@ -2909,15 +2909,16 @@ static int tty_fasync(int fd, struct file

*filp, int on) {
struct tty_struct *tty;
unsigned long flags;
- int retval;
+ int retval = 0;

+ lock_kernel();
tty = (struct tty_struct *)filp->private_data; if (tty_paranoia_check(tty, filp->f_path.dentry->d_inode,
"tty_fasync"))
- return 0;
+ goto out;

retval = fasync_helper(fd, filp, on, &tty->fasync); if (retval <= 0)
- return retval;
+ goto out;

if (on) {
enum pid_type type;
@@ -2935,12 +2936,15 @@ static int tty_fasync(int fd, struct file *filp, int on) spin_unlock_irqrestore(&tty->ctrl_lock, flags);
retval = __f_setown(filp, pid, type, 0);
if (retval)
- return retval;
+ goto out;
} else {
if (!tty->fasync && !waitqueue_active(&tty->read_wait))
tty->minimum_to_wake = N_TTY_BUF_SIZE;
}
- return 0;
+ retval = 0;
+out:
+ unlock_kernel();
+ return retval;
}

/**

(quoted diff) +0 -0

b/drivers/message/fusion/mptctl.c index e630b50..c594656 100644

drivers/message/fusion/mptctl.c +5 -1
@@ -548,11 +548,15 @@ static int
 mptctl_fasync(int fd, struct file *filep, int mode)
 {
 	MPT_ADAPTER	*ioc;
+	int ret;
 
+	lock_kernel();
 	list_for_each_entry(ioc, &ioc_list, list)
 		ioc->aen_event_read_flag=0;
 
-	return fasync_helper(fd, filep, mode, &async_queue);
+	ret = fasync_helper(fd, filep, mode, &async_queue);
+	unlock_kernel();
+	return ret;
 }
 
 static int
(quoted diff) +0 -0

b/drivers/message/i2o/i2o_config.c index 95b4c10..4238de9 100644

drivers/message/i2o/i2o_config.c +0 -0
@@ -1084,15 +1084,17 @@ static int cfg_fasync(int fd, struct file *fp,

int on) {
ulong id = (ulong) fp->private_data;
struct i2o_cfg_info *p;
+ int ret = -EBADF;

+ lock_kernel();
for (p = open_files; p; p = p->next)
if (p->q_id == id)
break;

- if (!p)
- return -EBADF;
-
- return fasync_helper(fd, fp, on, &p->fasync);
+ if (p)
+ ret = fasync_helper(fd, fp, on, &p->fasync);
+ unlock_kernel();
+ return ret;
}

static int cfg_release(struct inode *inode, struct file *file)

drivers/net/tun.c +0 -0
index ce5af2a..4c0c597 100644
@@ -782,18 +782,21 @@ static int tun_chr_fasync(int fd, struct file

*file, int on)
DBG(KERN_INFO "%s: tun_chr_fasync %d\n", tun->dev->name, on);

+ lock_kernel();
if ((ret = fasync_helper(fd, file, on, &tun->fasync)) < 0)
- return ret;
+ goto out;

if (on) {
ret = __f_setown(file, task_pid(current), PIDTYPE_PID,
0); if (ret)
- return ret;
+ goto out;
tun->flags |= TUN_FASYNC;
} else
tun->flags &= ~TUN_FASYNC;
-
- return 0;
+ ret = 0;
+out:
+ unlock_kernel();
+ return ret;
}

static int tun_chr_open(struct inode *inode, struct file * file)

fs/ecryptfs/file.c +1 -0
index 2258b8f..24749bf 100644
@@ -30,6 +30,7 @@
 #include <linux/security.h>
 #include <linux/compat.h>
 #include <linux/fs_stack.h>
+#include <linux/smp_lock.h>
 #include "ecryptfs_kernel.h"
 
 /**
@@ -277,9 +278,11 @@ static int ecryptfs_fasync(int fd, struct file

*file, int flag) int rc = 0;
struct file *lower_file = NULL;

+ lock_kernel();
lower_file = ecryptfs_file_to_lower(file);
if (lower_file->f_op && lower_file->f_op->fasync)
rc = lower_file->f_op->fasync(fd, lower_file, flag);
+ unlock_kernel();
return rc;
}

fs/fcntl.c +0 -1
index bfd7765..330a7d7 100644
@@ -12,7 +12,6 @@
 #include <linux/fdtable.h>
 #include <linux/capability.h>
 #include <linux/dnotify.h>
-#include <linux/smp_lock.h>
 #include <linux/slab.h>
 #include <linux/module.h>
 #include <linux/security.h>
@@ -227,7 +226,6 @@ static int setfl(int fd, struct file * filp,

unsigned long arg) if (error)
return error;

- lock_kernel();
if ((arg ^ filp->f_flags) & FASYNC) {
if (filp->f_op && filp->f_op->fasync) {
error = filp->f_op->fasync(fd, filp, (arg &
FASYNC) != 0); @@ -238,7 +236,6 @@ static int setfl(int fd, struct file
* filp, unsigned long arg)
filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags &
~SETFL_MASK); out:
- unlock_kernel();
return error;
}

sound/core/pcm_native.c +1 -0
index 61f5d42..c49b9d9 100644
@@ -22,6 +22,7 @@
 #include <linux/mm.h>
 #include <linux/file.h>
 #include <linux/slab.h>
+#include <linux/smp_lock.h>
 #include <linux/time.h>
 #include <linux/pm_qos_params.h>
 #include <linux/uio.h>
@@ -3249,14 +3250,17 @@ static int snd_pcm_fasync(int fd, struct file *

file, int on) struct snd_pcm_file * pcm_file;
struct snd_pcm_substream *substream;
struct snd_pcm_runtime *runtime;
- int err;
+ int err = -ENXIO;

+ lock_kernel();
pcm_file = file->private_data;
substream = pcm_file->substream;
- snd_assert(substream != NULL, return -ENXIO);
+ snd_assert(substream != NULL, goto out);
runtime = substream->runtime;

err = fasync_helper(fd, file, on, &runtime->fasync);
+out:
+ unlock_kernel();
if (err < 0)
return err;
return 0;

Acked-by on v1 1/1
#5 lore raw

Re: [PATCH, RFC] fasync() BKL pushdown (take 2)

Quoted 6 lines · Jonathan Corbet
Quoted 2 lines · On Fri, 20 Jun 2008 21:12:51 +0200 Andi Kleen <[email protected]> wrote:

Some devices do state change even when the reference count is > 0. Would need to double check it's all ok with the fasync list.

Quoted 6 lines

OK, I've gone over all of the fasync() definitions again with an eye toward convincing myself that the fasync list would not get cleared, freed, or otherwise molested if fasync() runs without BKL protection. I focused especially on other code (open(), ioctl()) which might still run with the BKL. The result was two more pushdowns in spots where I wasn't sure; chance are both are unnecessary.

Ok fine for me then. I haven't read it again in detail, but it sounds good now.

I still think it would be better to somehow compile break external users, but that would be only the icing on the cake.

-Andi

#6 lore raw
In reply to Jonathan Corbet
Quoted 19 lines · On Fri, 20 Jun 2008 11:29:14 -0600 Jonathan Corbet <[email protected]> wrote:

The majority of fasync() functions just call fasync_helper() with a pointer to an fasync_struct reachable from the file structure. Given that (1) the struct file will not go away while fasync() is running, and (2) the VFS-level fasync() stuff is protected with the Big Fasync Lock, I contend that these simple implementations have no need for the BKL. Once those are filtered out, there's really only a half-dozen places which need to be fixed.

Jonathan Corbet (7):
mpt: fasync BKL pushdown
i2o: fasync BKL pushdown
tun: fasync BKL pushdown
tty_io: fasync BKL pushdown
Bluetooth VHCI: fasync BKL pushdown
ecryptfs: fasync BKL pushdown
Call fasync() functions without the BKL

Quoted 3 lines

And the actual patch is appended. If nobody objects, I'll pull these into the bkl-removal tree shortly.

looks good to me

only question is if this one is really needed:
+ lock_kernel();
tty = (struct tty_struct *)filp->private_data; if (tty_paranoia_check(tty, filp->f_path.dentry->d_inode,
"tty_fasync"))
- return 0;
+ goto out;

retval = fasync_helper(fd, filp, on, &tty->fasync); if (retval <= 0)

(given that it just calls the helper.. mostly and that Alan has been busy removing the BKL from the tty layer)

Signature · 3 lines
If you want to reach me at my work email, use [email protected]
For development, discussion and tips for power savings, 
visit http://www.lesswatts.org
#7 lore raw
Quoted 11 lines · On Fri, 20 Jun 2008 13:58:40 -0700 Arjan van de Ven <[email protected]> wrote:

only question is if this one is really needed:

drivers/char/tty_io.c +1 -0
+	lock_kernel();
 	tty = (struct tty_struct *)filp->private_data;
 	if (tty_paranoia_check(tty, filp->f_path.dentry->d_inode,

"tty_fasync"))

drivers/char/tty_io.c +1 -1
-		return 0;
+		goto out;
 
 	retval = fasync_helper(fd, filp, on, &tty->fasync);
 	if (retval <= 0)
Quoted 2 lines

(given that it just calls the helper.. mostly and that Alan has been busy removing the BKL from the tty layer)

There's enough other stuff going on there, including manipulations of the tty_struct, waitqueue_active() checks, etc., that I figured I should really leave the BKL in there and let Alan figure it out. I've learned that it's best to tread with great care in that part of the kernel...

jon

Jonathan Corbet / LWN.net / [email protected]