Skip to contentSkip to conversation

tty: move pty count limiting into devpts

PATCH In review Konstantin Khlebnikov 4 participants 8 messages Jan 2, 2012 – Jan 5, 2012 TTY layer and serial drivers

lore
Share & tools
lore

Konstantin Khlebnikov

author
#0 lore raw

PATCH 1/2 tty: move pty count limiting into devpts

Let's move this stuff to the better place, where we can account pty right in tty-indexes managing code.

Signed-off-by Konstantin Khlebnikov <[email protected]>
---
 drivers/tty/pty.c |   63 -----------------------------------------------------
 fs/devpts/inode.c |   53 ++++++++++++++++++++++++++++++++++++++++++++-
 2 files changed, 52 insertions(+), 64 deletions(-)
drivers/tty/pty.c +0 -63
index d2bf3c1..33f7e18 100644
@@ -21,7 +21,6 @@
 #include <linux/major.h>
 #include <linux/mm.h>
 #include <linux/init.h>
-#include <linux/sysctl.h>
 #include <linux/device.h>
 #include <linux/uaccess.h>
 #include <linux/bitops.h>

@@ -439,66 +438,9 @@ static inline void legacy_pty_init(void) { }
 
 /* Unix98 devices */
 #ifdef CONFIG_UNIX98_PTYS
-/*
- * sysctl support for setting limits on the number of Unix98 ptys allocated.
- * Otherwise one can eat up all kernel memory by opening /dev/ptmx repeatedly.
- */
-int pty_limit = NR_UNIX98_PTY_DEFAULT;
-static int pty_limit_min;
-static int pty_limit_max = NR_UNIX98_PTY_MAX;
-static int tty_count;
-static int pty_count;
-
-static inline void pty_inc_count(void)
-{
-	pty_count = (++tty_count) / 2;
-}
-
-static inline void pty_dec_count(void)
-{
-	pty_count = (--tty_count) / 2;
-}
 
 static struct cdev ptmx_cdev;
 
-static struct ctl_table pty_table[] = {
-	{
-		.procname	= "max",
-		.maxlen		= sizeof(int),
-		.mode		= 0644,
-		.data		= &pty_limit,
-		.proc_handler	= proc_dointvec_minmax,
-		.extra1		= &pty_limit_min,
-		.extra2		= &pty_limit_max,
-	}, {
-		.procname	= "nr",
-		.maxlen		= sizeof(int),
-		.mode		= 0444,
-		.data		= &pty_count,
-		.proc_handler	= proc_dointvec,
-	}, 
-	{}
-};
-
-static struct ctl_table pty_kern_table[] = {
-	{
-		.procname	= "pty",
-		.mode		= 0555,
-		.child		= pty_table,
-	},
-	{}
-};
-
-static struct ctl_table pty_root_table[] = {
-	{
-		.procname	= "kernel",
-		.mode		= 0555,
-		.child		= pty_kern_table,
-	},
-	{}
-};
-
-
 static int pty_unix98_ioctl(struct tty_struct *tty,
 			    unsigned int cmd, unsigned long arg)
 {

@@ -598,8 +540,6 @@ static int pty_unix98_install(struct tty_driver *driver, struct tty_struct *tty)
 	 */
 	tty_driver_kref_get(driver);
 	tty->count++;
-	pty_inc_count(); /* tty */
-	pty_inc_count(); /* tty->link */
 	return 0;
 err_free_mem:
 	deinitialize_tty_struct(o_tty);

@@ -613,7 +553,6 @@ err_free_tty:
 
 static void pty_unix98_remove(struct tty_driver *driver, struct tty_struct *tty)
 {
-	pty_dec_count();
 }
 
 static const struct tty_operations ptm_unix98_ops = {

@@ -768,8 +707,6 @@ static void __init unix98_pty_init(void)
 	if (tty_register_driver(pts_driver))
 		panic("Couldn't register Unix98 pts driver");
 
-	register_sysctl_table(pty_root_table);
-
 	/* Now create the /dev/ptmx special device */
 	tty_default_fops(&ptmx_fops);
 	ptmx_fops.open = ptmx_open;
fs/devpts/inode.c +52 -1

82 lines not on this page · raw message

author
#1 lore raw

PATCH 2/2 tty: rework pty count limiting

After adding devpts multiple-insrances sysctl kernel.pty.max limit pty count for each devpts instance independently, while kernel.pty.nr shows total pty count.

This patch restores sysctl kernel.pty.max as global limit (4096 by default), adds pty reseve for main devpts (mounted without "newinstance" argument), and new sysctl to tune it: kernel.pty.reserve (1024 by default)

Also it adds devpts mount option "max=%d" to limit pty count for each devpts instance independently. (by default NR_UNIX98_PTY_MAX == 2^20)

Thus devpts instances in containers cannot eat up all available pty even if we didn't set any limits, while with "max" argument we can adjust limits more precisely.

Plus, now open("/dev/ptmx") return -ENOSPC in case lack of pty indexes, this is more informative than -EIO.

Signed-off-by Konstantin Khlebnikov <[email protected]>
---
 fs/devpts/inode.c   |   34 ++++++++++++++++++++++++++++++----
 include/linux/tty.h |    1 +
 2 files changed, 31 insertions(+), 4 deletions(-)
fs/devpts/inode.c +30 -4
index dd37e85..ff84832 100644
@@ -41,8 +41,9 @@
  * Otherwise one can eat up all kernel memory by opening /dev/ptmx repeatedly.
  */
 static int pty_limit = NR_UNIX98_PTY_DEFAULT;
+static int pty_reserve = NR_UNIX98_PTY_RESERVE;
 static int pty_limit_min;
-static int pty_limit_max = NR_UNIX98_PTY_MAX;
+static int pty_limit_max = INT_MAX;
 static int pty_count;
 
 static struct ctl_table pty_table[] = {

@@ -55,6 +56,14 @@ static struct ctl_table pty_table[] = {
 		.extra1		= &pty_limit_min,
 		.extra2		= &pty_limit_max,
 	}, {
+		.procname	= "reserve",
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.data		= &pty_reserve,
+		.proc_handler	= proc_dointvec_minmax,
+		.extra1		= &pty_limit_min,
+		.extra2		= &pty_limit_max,
+	}, {
 		.procname	= "nr",
 		.maxlen		= sizeof(int),
 		.mode		= 0444,

@@ -94,10 +103,11 @@ struct pts_mount_opts {
 	umode_t mode;
 	umode_t ptmxmode;
 	int newinstance;
+	int max;
 };
 
 enum {
-	Opt_uid, Opt_gid, Opt_mode, Opt_ptmxmode, Opt_newinstance,
+	Opt_uid, Opt_gid, Opt_mode, Opt_ptmxmode, Opt_newinstance,  Opt_max,
 	Opt_err
 };
 

@@ -108,6 +118,7 @@ static const match_table_t tokens = {
 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
 	{Opt_ptmxmode, "ptmxmode=%o"},
 	{Opt_newinstance, "newinstance"},
+	{Opt_max, "max=%d"},
 #endif
 	{Opt_err, NULL}
 };

@@ -154,6 +165,7 @@ static int parse_mount_options(char *data, int op, struct pts_mount_opts *opts)
 	opts->gid     = 0;
 	opts->mode    = DEVPTS_DEFAULT_MODE;
 	opts->ptmxmode = DEVPTS_DEFAULT_PTMX_MODE;
+	opts->max     = NR_UNIX98_PTY_MAX;
 
 	/* newinstance makes sense only on initial mount */
 	if (op == PARSE_MOUNT)

@@ -197,6 +209,12 @@ static int parse_mount_options(char *data, int op, struct pts_mount_opts *opts)
 			if (op == PARSE_MOUNT)
 				opts->newinstance = 1;
 			break;
+		case Opt_max:
+			if (match_int(&args[0], &option) ||
+			    option < 0 || option > NR_UNIX98_PTY_MAX)
+				return -EINVAL;
+			opts->max = option;
+			break;
 #endif
 		default:
 			printk(KERN_ERR "devpts: called with bogus options\n");

@@ -303,6 +321,8 @@ static int devpts_show_options(struct seq_file *seq, struct vfsmount *vfs)
 	seq_printf(seq, ",mode=%03o", opts->mode);
 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
 	seq_printf(seq, ",ptmxmode=%03o", opts->ptmxmode);
+	if (opts->max < NR_UNIX98_PTY_MAX)
+		seq_printf(seq, ",max=%d", opts->max);
 #endif
 
 	return 0;

@@ -485,6 +505,12 @@ retry:
 		return -ENOMEM;
 
 	mutex_lock(&allocated_ptys_lock);
+	if (pty_count >= pty_limit -
+			(fsi->mount_opts.newinstance ? pty_reserve : 0)) {
+		mutex_unlock(&allocated_ptys_lock);
+		return -ENOSPC;
+	}
+
 	ida_ret = ida_get_new(&fsi->allocated_ptys, &index);
 	if (ida_ret < 0) {
 		mutex_unlock(&allocated_ptys_lock);

@@ -493,10 +519,10 @@ retry:
 		return -EIO;
 	}
 
-	if (index >= pty_limit) {
+	if (index >= fsi->mount_opts.max) {
 		ida_remove(&fsi->allocated_ptys, index);
 		mutex_unlock(&allocated_ptys_lock);
-		return -EIO;
+		return -ENOSPC;
 	}
 	pty_count++;
 	mutex_unlock(&allocated_ptys_lock);
include/linux/tty.h +1 -0
index d3ebd76..d407741 100644
@@ -52,6 +52,7 @@
  * hardcoded at present.)
  */
 #define NR_UNIX98_PTY_DEFAULT	4096      /* Default maximum for Unix98 ptys */
+#define NR_UNIX98_PTY_RESERVE	1024	  /* Default reserve for main devpts */
 #define NR_UNIX98_PTY_MAX	(1 << MINORBITS) /* Absolute limit */
 
 /*
#2 lore raw

Re: [PATCH 2/2] tty: rework pty count limiting

Quoted 7 lines · Konstantin Khlebnikov

After adding devpts multiple-insrances sysctl kernel.pty.max limit pty count for each devpts instance independently, while kernel.pty.nr shows total pty count.

This patch restores sysctl kernel.pty.max as global limit (4096 by default), adds pty reseve for main devpts (mounted without "newinstance" argument), and new sysctl to tune it: kernel.pty.reserve (1024 by default)

Quoted 8 lines

Also it adds devpts mount option "max=%d" to limit pty count for each devpts instance independently. (by default NR_UNIX98_PTY_MAX == 2^20)

Thus devpts instances in containers cannot eat up all available pty even if we didn't set any limits, while with "max" argument we can adjust limits more precisely.

Plus, now open("/dev/ptmx") return -ENOSPC in case lack of pty indexes, this is more informative than -EIO.

That's a userspace api change, why is this going to be allowed?

greg k-h

author
#3 lore raw

Re: [PATCH 2/2] tty: rework pty count limiting

Quoted 11 lines · Greg KH
Quoted 10 lines · Konstantin Khlebnikov

After adding devpts multiple-insrances sysctl kernel.pty.max limit pty count for each devpts instance independently, while kernel.pty.nr shows total pty count.

This patch restores sysctl kernel.pty.max as global limit (4096 by default), adds pty reseve for main devpts (mounted without "newinstance" argument), and new sysctl to tune it: kernel.pty.reserve (1024 by default)

Also it adds devpts mount option "max=%d" to limit pty count for each devpts instance independently. (by default NR_UNIX98_PTY_MAX == 2^20)

Quoted 7 lines
Quoted 5 lines

Thus devpts instances in containers cannot eat up all available pty even if we didn't set any limits, while with "max" argument we can adjust limits more precisely.

Plus, now open("/dev/ptmx") return -ENOSPC in case lack of pty indexes, this is more informative than -EIO.

That's a userspace api change, why is this going to be allowed?

open() can return -ENOSPC in case normal fs, meanwhile error codes for /dev/ptmx wasn't documented at all:
there is nothing in manpage pts, pty and ptmx. manpage posix_openpt and getpt refers to manpage open(2). I can return EIO back, if this is critical.

Quoted 2 lines

greg k-h

#4 lore raw

Re: [PATCH 2/2] tty: rework pty count limiting

Quoted 7 lines
Quoted 1 line

That's a userspace api change, why is this going to be allowed?

open() can return -ENOSPC in case normal fs, meanwhile error codes for /dev/ptmx wasn't documented at all:
there is nothing in manpage pts, pty and ptmx. manpage posix_openpt and getpt refers to manpage open(2). I can return EIO back, if this is critical.

I prefer the -ENOSPC change - its perfectly within POSIX and while its vaguely an ABI change I don't believe its one that a single piece of software actually cares about. So nothing will break (famous last words).

Lets try it and see.

Alan

author
#6 lore raw

PATCH 1/2 tty: move pty count limiting into devpts

Let's move this stuff to the better place, where we can account pty right in tty-indexes managing code.

Signed-off-by Konstantin Khlebnikov <[email protected]>
---
 drivers/tty/pty.c |   51 ---------------------------------------------------
 fs/devpts/inode.c |   53 ++++++++++++++++++++++++++++++++++++++++++++++++++++-
 2 files changed, 52 insertions(+), 52 deletions(-)
drivers/tty/pty.c +0 -51
index 03147fa..d505837 100644
@@ -21,7 +21,6 @@
 #include <linux/major.h>
 #include <linux/mm.h>
 #include <linux/init.h>
-#include <linux/sysctl.h>
 #include <linux/device.h>
 #include <linux/uaccess.h>
 #include <linux/bitops.h>

@@ -439,55 +438,9 @@ static inline void legacy_pty_init(void) { }
 
 /* Unix98 devices */
 #ifdef CONFIG_UNIX98_PTYS
-/*
- * sysctl support for setting limits on the number of Unix98 ptys allocated.
- * Otherwise one can eat up all kernel memory by opening /dev/ptmx repeatedly.
- */
-int pty_limit = NR_UNIX98_PTY_DEFAULT;
-static int pty_limit_min;
-static int pty_limit_max = NR_UNIX98_PTY_MAX;
-static int pty_count;
 
 static struct cdev ptmx_cdev;
 
-static struct ctl_table pty_table[] = {
-	{
-		.procname	= "max",
-		.maxlen		= sizeof(int),
-		.mode		= 0644,
-		.data		= &pty_limit,
-		.proc_handler	= proc_dointvec_minmax,
-		.extra1		= &pty_limit_min,
-		.extra2		= &pty_limit_max,
-	}, {
-		.procname	= "nr",
-		.maxlen		= sizeof(int),
-		.mode		= 0444,
-		.data		= &pty_count,
-		.proc_handler	= proc_dointvec,
-	}, 
-	{}
-};
-
-static struct ctl_table pty_kern_table[] = {
-	{
-		.procname	= "pty",
-		.mode		= 0555,
-		.child		= pty_table,
-	},
-	{}
-};
-
-static struct ctl_table pty_root_table[] = {
-	{
-		.procname	= "kernel",
-		.mode		= 0555,
-		.child		= pty_kern_table,
-	},
-	{}
-};
-
-
 static int pty_unix98_ioctl(struct tty_struct *tty,
 			    unsigned int cmd, unsigned long arg)
 {

@@ -587,7 +540,6 @@ static int pty_unix98_install(struct tty_driver *driver, struct tty_struct *tty)
 	 */
 	tty_driver_kref_get(driver);
 	tty->count++;
-	pty_count++;
 	return 0;
 err_free_mem:
 	deinitialize_tty_struct(o_tty);

@@ -601,7 +553,6 @@ err_free_tty:
 
 static void ptm_unix98_remove(struct tty_driver *driver, struct tty_struct *tty)
 {
-	pty_count--;
 }
 
 static void pts_unix98_remove(struct tty_driver *driver, struct tty_struct *tty)

@@ -760,8 +711,6 @@ static void __init unix98_pty_init(void)
 	if (tty_register_driver(pts_driver))
 		panic("Couldn't register Unix98 pts driver");
 
-	register_sysctl_table(pty_root_table);
-
 	/* Now create the /dev/ptmx special device */
 	tty_default_fops(&ptmx_fops);
 	ptmx_fops.open = ptmx_open;
fs/devpts/inode.c +52 -1

82 lines not on this page · raw message

author
#7 lore raw

PATCH 2/2 tty: rework pty count limiting

In reply to Greg KH

After adding devpts multiple-insrances sysctl kernel.pty.max limit pty count for each devpts instance independently, while kernel.pty.nr shows total pty count.

This patch restores sysctl kernel.pty.max as global limit (4096 by default), adds pty reseve for main devpts (mounted without "newinstance" argument), and new sysctl to tune it: kernel.pty.reserve (1024 by default)

Also it adds devpts mount option "max=%d" to limit pty count for each devpts instance independently. (by default NR_UNIX98_PTY_MAX == 2^20)

Thus devpts instances in containers cannot eat up all available pty even if we didn't set any limits, while with "max" argument we can adjust limits more precisely.

Plus, now open("/dev/ptmx") return -ENOSPC in case lack of pty indexes, this is more informative than -EIO.

Signed-off-by Konstantin Khlebnikov <[email protected]>
---
 fs/devpts/inode.c   |   34 ++++++++++++++++++++++++++++++----
 include/linux/tty.h |    1 +
 2 files changed, 31 insertions(+), 4 deletions(-)
fs/devpts/inode.c +30 -4
index dd37e85..ff84832 100644
@@ -41,8 +41,9 @@
  * Otherwise one can eat up all kernel memory by opening /dev/ptmx repeatedly.
  */
 static int pty_limit = NR_UNIX98_PTY_DEFAULT;
+static int pty_reserve = NR_UNIX98_PTY_RESERVE;
 static int pty_limit_min;
-static int pty_limit_max = NR_UNIX98_PTY_MAX;
+static int pty_limit_max = INT_MAX;
 static int pty_count;
 
 static struct ctl_table pty_table[] = {

@@ -55,6 +56,14 @@ static struct ctl_table pty_table[] = {
 		.extra1		= &pty_limit_min,
 		.extra2		= &pty_limit_max,
 	}, {
+		.procname	= "reserve",
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.data		= &pty_reserve,
+		.proc_handler	= proc_dointvec_minmax,
+		.extra1		= &pty_limit_min,
+		.extra2		= &pty_limit_max,
+	}, {
 		.procname	= "nr",
 		.maxlen		= sizeof(int),
 		.mode		= 0444,

@@ -94,10 +103,11 @@ struct pts_mount_opts {
 	umode_t mode;
 	umode_t ptmxmode;
 	int newinstance;
+	int max;
 };
 
 enum {
-	Opt_uid, Opt_gid, Opt_mode, Opt_ptmxmode, Opt_newinstance,
+	Opt_uid, Opt_gid, Opt_mode, Opt_ptmxmode, Opt_newinstance,  Opt_max,
 	Opt_err
 };
 

@@ -108,6 +118,7 @@ static const match_table_t tokens = {
 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
 	{Opt_ptmxmode, "ptmxmode=%o"},
 	{Opt_newinstance, "newinstance"},
+	{Opt_max, "max=%d"},
 #endif
 	{Opt_err, NULL}
 };

@@ -154,6 +165,7 @@ static int parse_mount_options(char *data, int op, struct pts_mount_opts *opts)
 	opts->gid     = 0;
 	opts->mode    = DEVPTS_DEFAULT_MODE;
 	opts->ptmxmode = DEVPTS_DEFAULT_PTMX_MODE;
+	opts->max     = NR_UNIX98_PTY_MAX;
 
 	/* newinstance makes sense only on initial mount */
 	if (op == PARSE_MOUNT)

@@ -197,6 +209,12 @@ static int parse_mount_options(char *data, int op, struct pts_mount_opts *opts)
 			if (op == PARSE_MOUNT)
 				opts->newinstance = 1;
 			break;
+		case Opt_max:
+			if (match_int(&args[0], &option) ||
+			    option < 0 || option > NR_UNIX98_PTY_MAX)
+				return -EINVAL;
+			opts->max = option;
+			break;
 #endif
 		default:
 			printk(KERN_ERR "devpts: called with bogus options\n");

@@ -303,6 +321,8 @@ static int devpts_show_options(struct seq_file *seq, struct vfsmount *vfs)
 	seq_printf(seq, ",mode=%03o", opts->mode);
 #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES
 	seq_printf(seq, ",ptmxmode=%03o", opts->ptmxmode);
+	if (opts->max < NR_UNIX98_PTY_MAX)
+		seq_printf(seq, ",max=%d", opts->max);
 #endif
 
 	return 0;

@@ -485,6 +505,12 @@ retry:
 		return -ENOMEM;
 
 	mutex_lock(&allocated_ptys_lock);
+	if (pty_count >= pty_limit -
+			(fsi->mount_opts.newinstance ? pty_reserve : 0)) {
+		mutex_unlock(&allocated_ptys_lock);
+		return -ENOSPC;
+	}
+
 	ida_ret = ida_get_new(&fsi->allocated_ptys, &index);
 	if (ida_ret < 0) {
 		mutex_unlock(&allocated_ptys_lock);

@@ -493,10 +519,10 @@ retry:
 		return -EIO;
 	}
 
-	if (index >= pty_limit) {
+	if (index >= fsi->mount_opts.max) {
 		ida_remove(&fsi->allocated_ptys, index);
 		mutex_unlock(&allocated_ptys_lock);
-		return -EIO;
+		return -ENOSPC;
 	}
 	pty_count++;
 	mutex_unlock(&allocated_ptys_lock);
include/linux/tty.h +1 -0
index d3ebd76..d407741 100644
@@ -52,6 +52,7 @@
  * hardcoded at present.)
  */
 #define NR_UNIX98_PTY_DEFAULT	4096      /* Default maximum for Unix98 ptys */
+#define NR_UNIX98_PTY_RESERVE	1024	  /* Default reserve for main devpts */
 #define NR_UNIX98_PTY_MAX	(1 << MINORBITS) /* Absolute limit */
 
 /*