When booting a 16TB system, unix_create1 fails due to integer overflow.

[Robin Holt]

I do not know which direction to take, but here is the summary of the
problem.

We recently started trying to boot a customer's two new machines which
are configured with 384GB short of 16TB of memory.

We were seeing a failure which prevented boot. The kernel was incapable
of creating either a named pipe or unix domain socket. This comes down
to a common kernel function called unix_create1() which does:

atomic_inc(&unix_nr_socks);
if (atomic_read(&unix_nr_socks) > 2 * get_max_files())
goto out;

The function get_max_files() is a simple return of files_stat.max_files.
files_stat.max_files is a signed integer and is computed in
fs/file_table.c's files_init().

n = (mempages * (PAGE_SIZE / 1024)) / 10;
files_stat.max_files = n;

In our case, mempages (total_ram_pages) is approx 3,758,096,384
(0xe0000000). That leaves max_files at approximately 1,503,238,553.
This causes 2 * get_max_files() to integer overflow.

We came up with a few possible solutions:

Our first response was to limit max_files to (INT_MAX / 2) This at
least got us past the problem and seemed reasonable.

We could also have changed the 2 * get_max_files() to 2UL *
get_max_files() and gotten past this point in boot. That was not tested.

We could also have changed the definition of max_files to at least an
unsigned int instead of an int and gotten past the problem, but again,
not tested.

Any suggestions for a direction would be appreciated.

Thank you,
Robin Holt
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majo...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

[Eric Dumazet]

Hi Robin

I would say : We can use atomic_long_t instead of atomic_t

And make get_max_files(void) return a long ?

Something like :

fs/file_table.c | 10 +++++-----
include/linux/fs.h | 2 +-
net/unix/af_unix.c | 14 +++++++-------
3 files changed, 13 insertions(+), 13 deletions(-)

diff --git a/fs/file_table.c b/fs/file_table.c
index a04bdd8..a2d2189 100644
--- a/fs/file_table.c
+++ b/fs/file_table.c
@@ -68,9 +68,9 @@ static int get_nr_files(void)
/*
* Return the maximum number of open files in the system
*/
-int get_max_files(void)
+unsigned long get_max_files(void)
{
- return files_stat.max_files;
+ return (unsigned long)(unsigned int)files_stat.max_files;
}
EXPORT_SYMBOL_GPL(get_max_files);

@@ -140,7 +140,7 @@ struct file *get_empty_filp(void)
over:
/* Ran out of filps - report that */
if (get_nr_files() > old_max) {
- printk(KERN_INFO "VFS: file-max limit %d reached\n",
+ printk(KERN_INFO "VFS: file-max limit %lu reached\n",
get_max_files());
old_max = get_nr_files();
}
@@ -487,7 +487,7 @@ retry:

void __init files_init(unsigned long mempages)
{
- int n;
+ unsigned long n;

filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
@@ -498,7 +498,7 @@ void __init files_init(unsigned long mempages)
*/

n = (mempages * (PAGE_SIZE / 1024)) / 10;

- files_stat.max_files = n;
+ files_stat.max_files = min(n, 0x7FFFFFFFUL);
if (files_stat.max_files < NR_FILE)
files_stat.max_files = NR_FILE;
files_defer_init();
diff --git a/include/linux/fs.h b/include/linux/fs.h
index 63d069b..0de4989 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -404,7 +404,7 @@ extern void __init inode_init_early(void);
extern void __init files_init(unsigned long);

extern struct files_stat_struct files_stat;
-extern int get_max_files(void);
+extern unsigned long get_max_files(void);
extern int sysctl_nr_open;
extern struct inodes_stat_t inodes_stat;
extern int leases_enable, lease_break_time;
diff --git a/net/unix/af_unix.c b/net/unix/af_unix.c
index 0b39b24..b3c70ac 100644
--- a/net/unix/af_unix.c
+++ b/net/unix/af_unix.c
@@ -117,7 +117,7 @@

static struct hlist_head unix_socket_table[UNIX_HASH_SIZE + 1];
static DEFINE_SPINLOCK(unix_table_lock);
-static atomic_t unix_nr_socks = ATOMIC_INIT(0);
+static atomic_long_t unix_nr_socks = ATOMIC_INIT(0);

#define unix_sockets_unbound (&unix_socket_table[UNIX_HASH_SIZE])

@@ -360,13 +360,13 @@ static void unix_sock_destructor(struct sock *sk)
if (u->addr)
unix_release_addr(u->addr);

- atomic_dec(&unix_nr_socks);
+ atomic_long_dec(&unix_nr_socks);
local_bh_disable();
sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
local_bh_enable();
#ifdef UNIX_REFCNT_DEBUG
- printk(KERN_DEBUG "UNIX %p is destroyed, %d are still alive.\n", sk,
- atomic_read(&unix_nr_socks));
+ printk(KERN_DEBUG "UNIX %p is destroyed, %ld are still alive.\n", sk,
+ atomic_long_read(&unix_nr_socks));
#endif
}

@@ -606,8 +606,8 @@ static struct sock *unix_create1(struct net *net, struct socket *sock)
struct sock *sk = NULL;
struct unix_sock *u;

- atomic_inc(&unix_nr_socks);
- if (atomic_read(&unix_nr_socks) > 2 * get_max_files())
+ atomic_long_inc(&unix_nr_socks);
+ if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files())
goto out;

sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
@@ -632,7 +632,7 @@ static struct sock *unix_create1(struct net *net, struct socket *sock)
unix_insert_socket(unix_sockets_unbound, sk);
out:
if (sk == NULL)
- atomic_dec(&unix_nr_socks);
+ atomic_long_dec(&unix_nr_socks);
else {
local_bh_disable();
sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);

[Eric Dumazet]

Le jeudi 23 septembre 2010 à 14:53 +0200, Eric Dumazet a écrit :

> Hi Robin
>
> I would say : We can use atomic_long_t instead of atomic_t
>
> And make get_max_files(void) return a long ?
>

AFAIK we can move to long everywhere
(/proc/sys/fs/file-nr & /proc/sys/fs/file-max) ...

So you can open more than 2^31 files ;)

Here is the patch I tested (unfortunately not on a 16TB machine )

[PATCH] fs: allow for more than 2^31 files

Robin Holt tried to boot a 16TB system and found af_unix was overflowing
a 32bit value :

<quote>

We were seeing a failure which prevented boot. The kernel was incapable
of creating either a named pipe or unix domain socket. This comes down
to a common kernel function called unix_create1() which does:

atomic_inc(&unix_nr_socks);
if (atomic_read(&unix_nr_socks) > 2 * get_max_files())
goto out;

The function get_max_files() is a simple return of files_stat.max_files.
files_stat.max_files is a signed integer and is computed in
fs/file_table.c's files_init().

n = (mempages * (PAGE_SIZE / 1024)) / 10;
files_stat.max_files = n;

In our case, mempages (total_ram_pages) is approx 3,758,096,384
(0xe0000000). That leaves max_files at approximately 1,503,238,553.
This causes 2 * get_max_files() to integer overflow.

</quote>

Fix is to let /proc/sys/fs/file-nr & /proc/sys/fs/file-max use long
integers, and change af_unix to use an atomic_long_t instead of
atomic_t.

get_max_files() is changed to return an unsigned long.

before patch :
# echo 2147483648 >/proc/sys/fs/file-max
# cat /proc/sys/fs/file-max
-18446744071562067968

after patch:
# echo 2147483648 >/proc/sys/fs/file-max
# cat /proc/sys/fs/file-max
2147483648

Reported-by: Robin Holt <ho...@sgi.com>
Signed-off-by: Eric Dumazet <eric.d...@gmail.com>
---
fs/file_table.c | 15 ++++++---------
include/linux/fs.h | 8 ++++----
kernel/sysctl.c | 8 ++++----
net/unix/af_unix.c | 14 +++++++-------
4 files changed, 21 insertions(+), 24 deletions(-)

diff --git a/fs/file_table.c b/fs/file_table.c
index a04bdd8..46457ba 100644
--- a/fs/file_table.c
+++ b/fs/file_table.c
@@ -60,7 +60,7 @@ static inline void file_free(struct file *f)
/*
* Return the total number of open files in the system
*/
-static int get_nr_files(void)
+static long get_nr_files(void)
{
return percpu_counter_read_positive(&nr_files);
}
@@ -68,7 +68,7 @@ static int get_nr_files(void)

/*
* Return the maximum number of open files in the system
*/
-int get_max_files(void)
+unsigned long get_max_files(void)
{

return files_stat.max_files;
}
@@ -105,7 +105,7 @@ int proc_nr_files(ctl_table *table, int write,
struct file *get_empty_filp(void)
{
const struct cred *cred = current_cred();
- static int old_max;
+ static long old_max;
struct file * f;

/*
@@ -140,8 +140,7 @@ struct file *get_empty_filp(void)

over:
/* Ran out of filps - report that */
if (get_nr_files() > old_max) {
- printk(KERN_INFO "VFS: file-max limit %d reached\n",

- get_max_files());
+ pr_info("VFS: file-max limit %lu reached\n", get_max_files());
old_max = get_nr_files();
}
goto fail;
@@ -487,7 +486,7 @@ retry:

void __init files_init(unsigned long mempages)
{
- int n;
+ unsigned long n;

filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);

@@ -498,9 +497,7 @@ void __init files_init(unsigned long mempages)

*/

n = (mempages * (PAGE_SIZE / 1024)) / 10;
- files_stat.max_files = n;

- if (files_stat.max_files < NR_FILE)
- files_stat.max_files = NR_FILE;
+ files_stat.max_files = max_t(unsigned long, n, NR_FILE);
files_defer_init();
lg_lock_init(files_lglock);
percpu_counter_init(&nr_files, 0);
diff --git a/include/linux/fs.h b/include/linux/fs.h
index 63d069b..8c06590 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -34,9 +34,9 @@

/* And dynamically-tunable limits and defaults: */
struct files_stat_struct {
- int nr_files; /* read only */
- int nr_free_files; /* read only */
- int max_files; /* tunable */
+ unsigned long nr_files; /* read only */
+ unsigned long nr_free_files; /* read only */
+ unsigned long max_files; /* tunable */
};

struct inodes_stat_t {

@@ -404,7 +404,7 @@ extern void __init inode_init_early(void);
extern void __init files_init(unsigned long);

extern struct files_stat_struct files_stat;
-extern int get_max_files(void);
+extern unsigned long get_max_files(void);
extern int sysctl_nr_open;
extern struct inodes_stat_t inodes_stat;
extern int leases_enable, lease_break_time;

diff --git a/kernel/sysctl.c b/kernel/sysctl.c
index f88552c..fc667bf 100644
--- a/kernel/sysctl.c
+++ b/kernel/sysctl.c
@@ -1352,16 +1352,16 @@ static struct ctl_table fs_table[] = {
{
.procname = "file-nr",
.data = &files_stat,
- .maxlen = 3*sizeof(int),
+ .maxlen = sizeof(files_stat),
.mode = 0444,
- .proc_handler = proc_nr_files,
+ .proc_handler = proc_doulongvec_minmax,
},
{
.procname = "file-max",
.data = &files_stat.max_files,
- .maxlen = sizeof(int),
+ .maxlen = sizeof(files_stat.max_files),
.mode = 0644,
- .proc_handler = proc_dointvec,
+ .proc_handler = proc_doulongvec_minmax,
},
{
.procname = "nr_open",

[Dipankar Sarma]

On Thu, Sep 23, 2010 at 02:53:04PM +0200, Eric Dumazet wrote:
> Le jeudi 23 septembre 2010 à 07:17 -0500, Robin Holt a écrit :
> > I do not know which direction to take, but here is the summary of the
> > problem.
> >
> > We recently started trying to boot a customer's two new machines which
> > are configured with 384GB short of 16TB of memory.
> >
> > We were seeing a failure which prevented boot. The kernel was incapable
> > of creating either a named pipe or unix domain socket. This comes down
> > to a common kernel function called unix_create1() which does:
> >
> > atomic_inc(&unix_nr_socks);
> > if (atomic_read(&unix_nr_socks) > 2 * get_max_files())
> > goto out;
> >
>

> Hi Robin
>
> I would say : We can use atomic_long_t instead of atomic_t
>
> And make get_max_files(void) return a long ?
>
> Something like :
>
>
> fs/file_table.c | 10 +++++-----
> include/linux/fs.h | 2 +-
> net/unix/af_unix.c | 14 +++++++-------
> 3 files changed, 13 insertions(+), 13 deletions(-)
>
> diff --git a/fs/file_table.c b/fs/file_table.c
>

> n = (mempages * (PAGE_SIZE / 1024)) / 10;
> - files_stat.max_files = n;
> + files_stat.max_files = min(n, 0x7FFFFFFFUL);

It may be cleaner to just convert both the file counters and
the file limits to usnsigned long.

Other than that, this seems like a reasonable thing to do.

Thanks
Dipankar

[David Miller]

From: Dipankar Sarma <dipa...@in.ibm.com>
Date: Thu, 23 Sep 2010 19:40:37 +0530

Is someone following up on integrating this upstream so this thing
gets fixed?

Thanks.

[Eric Dumazet]

Le lundi 27 septembre 2010 à 15:36 -0700, David Miller a écrit :

> Is someone following up on integrating this upstream so this thing
> gets fixed?
>

Thanks for the heads-up.

I am not sure V2 of my patch was reviewed, maybe it did not reach the
list.

Here is V3 of it. I removed the ATOMIC_INIT(0) I left in V2.
It should be an ATOMIC_LONG_INIT(0), but then, it can be avoided.

CC netdev

Thanks

[PATCH V3] fs: allow for more than 2^31 files

Robin Holt tried to boot a 16TB system and found af_unix was overflowing
a 32bit value :

<quote>

We were seeing a failure which prevented boot. The kernel was incapable

of creating either a named pipe or unix domain socket. This comes down
to a common kernel function called unix_create1() which does:

atomic_inc(&unix_nr_socks);
if (atomic_read(&unix_nr_socks) > 2 * get_max_files())
goto out;

The function get_max_files() is a simple return of files_stat.max_files.

files_stat.max_files is a signed integer and is computed in
fs/file_table.c's files_init().

n = (mempages * (PAGE_SIZE / 1024)) / 10;
files_stat.max_files = n;

In our case, mempages (total_ram_pages) is approx 3,758,096,384
(0xe0000000). That leaves max_files at approximately 1,503,238,553.
This causes 2 * get_max_files() to integer overflow.

</quote>

Fix is to let /proc/sys/fs/file-nr & /proc/sys/fs/file-max use long
integers, and change af_unix to use an atomic_long_t instead of
atomic_t.

get_max_files() is changed to return an unsigned long.

get_nr_files() is changed to return a long.

unix_nr_socks is changed from atomic_t to atomic_long_t, while not
strictly needed to address Robin problem.

Before patch (on a 64bit kernel) :

# echo 2147483648 >/proc/sys/fs/file-max
# cat /proc/sys/fs/file-max
-18446744071562067968

After patch:

# echo 2147483648 >/proc/sys/fs/file-max
# cat /proc/sys/fs/file-max
2147483648

Reported-by: Robin Holt <ho...@sgi.com>
Signed-off-by: Eric Dumazet <eric.d...@gmail.com>
---
fs/file_table.c | 15 ++++++---------
include/linux/fs.h | 8 ++++----
kernel/sysctl.c | 8 ++++----
net/unix/af_unix.c | 14 +++++++-------

4 files changed, 21 insertions(+), 24 deletions(-)

diff --git a/fs/file_table.c b/fs/file_table.c
index a04bdd8..46457ba 100644
--- a/fs/file_table.c
+++ b/fs/file_table.c

@@ -60,7 +60,7 @@ static inline void file_free(struct file *f)
/*

* Return the total number of open files in the system
*/

-static int get_nr_files(void)
+static long get_nr_files(void)
{
return percpu_counter_read_positive(&nr_files);
}

@@ -68,7 +68,7 @@ static int get_nr_files(void)

/*
* Return the maximum number of open files in the system
*/
-int get_max_files(void)
+unsigned long get_max_files(void)
{

return files_stat.max_files;
}
@@ -105,7 +105,7 @@ int proc_nr_files(ctl_table *table, int write,
struct file *get_empty_filp(void)
{
const struct cred *cred = current_cred();
- static int old_max;
+ static long old_max;
struct file * f;

/*

@@ -140,8 +140,7 @@ struct file *get_empty_filp(void)

over:
/* Ran out of filps - report that */
if (get_nr_files() > old_max) {
- printk(KERN_INFO "VFS: file-max limit %d reached\n",

- get_max_files());
+ pr_info("VFS: file-max limit %lu reached\n", get_max_files());
old_max = get_nr_files();
}
goto fail;
@@ -487,7 +486,7 @@ retry:

void __init files_init(unsigned long mempages)
{
- int n;
+ unsigned long n;

filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);

@@ -498,9 +497,7 @@ void __init files_init(unsigned long mempages)
*/

n = (mempages * (PAGE_SIZE / 1024)) / 10;
- files_stat.max_files = n;

- if (files_stat.max_files < NR_FILE)
- files_stat.max_files = NR_FILE;
+ files_stat.max_files = max_t(unsigned long, n, NR_FILE);
files_defer_init();
lg_lock_init(files_lglock);
percpu_counter_init(&nr_files, 0);

diff --git a/include/linux/fs.h b/include/linux/fs.h
index 63d069b..8c06590 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h

@@ -34,9 +34,9 @@

/* And dynamically-tunable limits and defaults: */
struct files_stat_struct {
- int nr_files; /* read only */
- int nr_free_files; /* read only */
- int max_files; /* tunable */
+ unsigned long nr_files; /* read only */
+ unsigned long nr_free_files; /* read only */
+ unsigned long max_files; /* tunable */
};

struct inodes_stat_t {

@@ -404,7 +404,7 @@ extern void __init inode_init_early(void);
extern void __init files_init(unsigned long);

extern struct files_stat_struct files_stat;
-extern int get_max_files(void);
+extern unsigned long get_max_files(void);
extern int sysctl_nr_open;
extern struct inodes_stat_t inodes_stat;
extern int leases_enable, lease_break_time;

diff --git a/kernel/sysctl.c b/kernel/sysctl.c
index f88552c..fc667bf 100644
--- a/kernel/sysctl.c
+++ b/kernel/sysctl.c
@@ -1352,16 +1352,16 @@ static struct ctl_table fs_table[] = {
{
.procname = "file-nr",
.data = &files_stat,
- .maxlen = 3*sizeof(int),
+ .maxlen = sizeof(files_stat),
.mode = 0444,
- .proc_handler = proc_nr_files,
+ .proc_handler = proc_doulongvec_minmax,
},
{
.procname = "file-max",
.data = &files_stat.max_files,
- .maxlen = sizeof(int),
+ .maxlen = sizeof(files_stat.max_files),
.mode = 0644,
- .proc_handler = proc_dointvec,
+ .proc_handler = proc_doulongvec_minmax,
},
{
.procname = "nr_open",

diff --git a/net/unix/af_unix.c b/net/unix/af_unix.c
index 0b39b24..3e1d7d1 100644

--- a/net/unix/af_unix.c
+++ b/net/unix/af_unix.c
@@ -117,7 +117,7 @@

static struct hlist_head unix_socket_table[UNIX_HASH_SIZE + 1];
static DEFINE_SPINLOCK(unix_table_lock);
-static atomic_t unix_nr_socks = ATOMIC_INIT(0);

+static atomic_long_t unix_nr_socks;

#define unix_sockets_unbound (&unix_socket_table[UNIX_HASH_SIZE])

@@ -360,13 +360,13 @@ static void unix_sock_destructor(struct sock *sk)
if (u->addr)
unix_release_addr(u->addr);

- atomic_dec(&unix_nr_socks);
+ atomic_long_dec(&unix_nr_socks);
local_bh_disable();
sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
local_bh_enable();
#ifdef UNIX_REFCNT_DEBUG
- printk(KERN_DEBUG "UNIX %p is destroyed, %d are still alive.\n", sk,
- atomic_read(&unix_nr_socks));
+ printk(KERN_DEBUG "UNIX %p is destroyed, %ld are still alive.\n", sk,
+ atomic_long_read(&unix_nr_socks));
#endif
}

@@ -606,8 +606,8 @@ static struct sock *unix_create1(struct net *net, struct socket *sock)
struct sock *sk = NULL;
struct unix_sock *u;

- atomic_inc(&unix_nr_socks);
- if (atomic_read(&unix_nr_socks) > 2 * get_max_files())
+ atomic_long_inc(&unix_nr_socks);

+ if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files())
goto out;

sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
@@ -632,7 +632,7 @@ static struct sock *unix_create1(struct net *net, struct socket *sock)
unix_insert_socket(unix_sockets_unbound, sk);
out:
if (sk == NULL)
- atomic_dec(&unix_nr_socks);
+ atomic_long_dec(&unix_nr_socks);
else {
local_bh_disable();
sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);

[David Miller]

From: Eric Dumazet <eric.d...@gmail.com>
Date: Tue, 28 Sep 2010 05:46:51 +0200

> [PATCH V3] fs: allow for more than 2^31 files

..

> Reported-by: Robin Holt <ho...@sgi.com>
> Signed-off-by: Eric Dumazet <eric.d...@gmail.com>

Acked-by: David S. Miller <da...@davemloft.net>

[Robin Holt]

On Tue, Sep 28, 2010 at 05:46:51AM +0200, Eric Dumazet wrote:
> Le lundi 27 septembre 2010 à 15:36 -0700, David Miller a écrit :

..

> Fix is to let /proc/sys/fs/file-nr & /proc/sys/fs/file-max use long
> integers, and change af_unix to use an atomic_long_t instead of
> atomic_t.
>
> get_max_files() is changed to return an unsigned long.

I _THINK_ we actually want get_max_files to return a long and have
the files_stat_struct definitions be longs. If we do not have it that
way, we could theoretically open enough files on a single cpu to make
get_nr_files return a negative without overflowing max_files. That,
of course, would require an insane amount of memory, but I think it is
technically more correct.

> --- a/kernel/sysctl.c
> +++ b/kernel/sysctl.c
> @@ -1352,16 +1352,16 @@ static struct ctl_table fs_table[] = {
> {
> .procname = "file-nr",
> .data = &files_stat,
> - .maxlen = 3*sizeof(int),
> + .maxlen = sizeof(files_stat),
> .mode = 0444,
> - .proc_handler = proc_nr_files,
> + .proc_handler = proc_doulongvec_minmax,

With this change, don't we lose the current nr_files value? I think
you need proc_nr_files to stay as it was. If you disagree, we should
probably eliminate the definitions for proc_nr_files as I don't believe
they are used anywhere else.

Thanks,
Robin

[Eric Dumazet]

Le jeudi 30 septembre 2010 à 15:26 -0500, Robin Holt a écrit :
> On Tue, Sep 28, 2010 at 05:46:51AM +0200, Eric Dumazet wrote:
> > Le lundi 27 septembre 2010 à 15:36 -0700, David Miller a écrit :

> ...

>
> > Fix is to let /proc/sys/fs/file-nr & /proc/sys/fs/file-max use long
> > integers, and change af_unix to use an atomic_long_t instead of
> > atomic_t.
> >
> > get_max_files() is changed to return an unsigned long.
>
> I _THINK_ we actually want get_max_files to return a long and have
> the files_stat_struct definitions be longs. If we do not have it that
> way, we could theoretically open enough files on a single cpu to make
> get_nr_files return a negative without overflowing max_files. That,
> of course, would require an insane amount of memory, but I think it is
> technically more correct.
>

Number of opened file is technically a positive (or null) value, I have
no idea why you want it being signed.

>
> > --- a/kernel/sysctl.c
> > +++ b/kernel/sysctl.c
> > @@ -1352,16 +1352,16 @@ static struct ctl_table fs_table[] = {
> > {
> > .procname = "file-nr",
> > .data = &files_stat,
> > - .maxlen = 3*sizeof(int),
> > + .maxlen = sizeof(files_stat),
> > .mode = 0444,
> > - .proc_handler = proc_nr_files,
> > + .proc_handler = proc_doulongvec_minmax,
>
> With this change, don't we lose the current nr_files value? I think
> you need proc_nr_files to stay as it was. If you disagree, we should
> probably eliminate the definitions for proc_nr_files as I don't believe
> they are used anywhere else.
>

I have no idea why you think I changed something. I only made the value
use 64bit on 64bit arches, so that we are not anymore limited to 2^31
files.

[Robin Holt]

The proc_handler used to be proc_nr_files() which would call
get_nr_files() and deposit the result in files_stat.nr_files then cascade
to proc_dointvec() which would dump the 3 values. Now it will dump the
three values, but not update the middle (nr_files) value first.

Robin

[Eric Dumazet]

Le jeudi 30 septembre 2010 à 23:34 -0500, Robin Holt a écrit :

> The proc_handler used to be proc_nr_files() which would call
> get_nr_files() and deposit the result in files_stat.nr_files then cascade
> to proc_dointvec() which would dump the 3 values. Now it will dump the
> three values, but not update the middle (nr_files) value first.
>

Ah I get it now, thanks !

I'll send a V4 shortly.

[Eric Dumazet]

Le vendredi 01 octobre 2010 à 07:03 +0200, Eric Dumazet a écrit :
> Le jeudi 30 septembre 2010 à 23:34 -0500, Robin Holt a écrit :
>
> > The proc_handler used to be proc_nr_files() which would call
> > get_nr_files() and deposit the result in files_stat.nr_files then cascade
> > to proc_dointvec() which would dump the 3 values. Now it will dump the
> > three values, but not update the middle (nr_files) value first.
> >
>
> Ah I get it now, thanks !
>
> I'll send a V4 shortly.
>
>

In this v4, I call proc_nr_files() again, and proc_nr_files() calls
proc_doulongvec_minmax() instead of proc_dointvec()

Added the "cat /proc/sys/fs/file-nr" in Changelog

Thanks again Robin

[PATCH V3] fs: allow for more than 2^31 files

Robin Holt tried to boot a 16TB system and found af_unix was overflowing
a 32bit value :

<quote>

We were seeing a failure which prevented boot. The kernel was incapable

of creating either a named pipe or unix domain socket. This comes down
to a common kernel function called unix_create1() which does:

atomic_inc(&unix_nr_socks);
if (atomic_read(&unix_nr_socks) > 2 * get_max_files())
goto out;

The function get_max_files() is a simple return of files_stat.max_files.
files_stat.max_files is a signed integer and is computed in
fs/file_table.c's files_init().

n = (mempages * (PAGE_SIZE / 1024)) / 10;
files_stat.max_files = n;

In our case, mempages (total_ram_pages) is approx 3,758,096,384
(0xe0000000). That leaves max_files at approximately 1,503,238,553.
This causes 2 * get_max_files() to integer overflow.

</quote>

Fix is to let /proc/sys/fs/file-nr & /proc/sys/fs/file-max use long
integers, and change af_unix to use an atomic_long_t instead of
atomic_t.

get_max_files() is changed to return an unsigned long.

get_nr_files() is changed to return a long.

unix_nr_socks is changed from atomic_t to atomic_long_t, while not
strictly needed to address Robin problem.

Before patch (on a 64bit kernel) :
# echo 2147483648 >/proc/sys/fs/file-max
# cat /proc/sys/fs/file-max
-18446744071562067968

After patch:
# echo 2147483648 >/proc/sys/fs/file-max
# cat /proc/sys/fs/file-max
2147483648

# cat /proc/sys/fs/file-nr
704 0 2147483648

Reported-by: Robin Holt <ho...@sgi.com>
Signed-off-by: Eric Dumazet <eric.d...@gmail.com>

---
fs/file_table.c | 17 +++++++----------
include/linux/fs.h | 8 ++++----
kernel/sysctl.c | 6 +++---

net/unix/af_unix.c | 14 +++++++-------
4 files changed, 21 insertions(+), 24 deletions(-)

diff --git a/fs/file_table.c b/fs/file_table.c
index a04bdd8..c3dee38 100644

--- a/fs/file_table.c
+++ b/fs/file_table.c
@@ -60,7 +60,7 @@ static inline void file_free(struct file *f)
/*
* Return the total number of open files in the system
*/
-static int get_nr_files(void)
+static long get_nr_files(void)
{
return percpu_counter_read_positive(&nr_files);
}
@@ -68,7 +68,7 @@ static int get_nr_files(void)
/*
* Return the maximum number of open files in the system
*/
-int get_max_files(void)
+unsigned long get_max_files(void)
{
return files_stat.max_files;
}

@@ -82,7 +82,7 @@ int proc_nr_files(ctl_table *table, int write,
void __user *buffer, size_t *lenp, loff_t *ppos)
{
files_stat.nr_files = get_nr_files();
- return proc_dointvec(table, write, buffer, lenp, ppos);
+ return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
}
#else

n = (mempages * (PAGE_SIZE / 1024)) / 10;

index f88552c..f789a0a 100644

--- a/kernel/sysctl.c
+++ b/kernel/sysctl.c
@@ -1352,16 +1352,16 @@ static struct ctl_table fs_table[] = {
{
.procname = "file-nr",
.data = &files_stat,
- .maxlen = 3*sizeof(int),
+ .maxlen = sizeof(files_stat),
.mode = 0444,

.proc_handler = proc_nr_files,

+ if (atomic_long_read(&unix_nr_socks) > 2 * get_max_files())
goto out;

sk = sk_alloc(net, PF_UNIX, GFP_KERNEL, &unix_proto);
@@ -632,7 +632,7 @@ static struct sock *unix_create1(struct net *net, struct socket *sock)
unix_insert_socket(unix_sockets_unbound, sk);
out:
if (sk == NULL)
- atomic_dec(&unix_nr_socks);
+ atomic_long_dec(&unix_nr_socks);
else {
local_bh_disable();
sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);

[Robin Holt]

Looks good.

Reviewed-by: Robin Holt <ho...@sgi.com>
Tested-by: Robin Holt <ho...@sgi.com>

I don't mean to flood this with my name, merely that I do find this
patch acceptable, worthy, and have tested it. Feel free to lop off any
of these lines that are offensive.

Robin

[Eric Dumazet]

Andrew,

Could you please review this patch, you probably are the right guy to
take it, because it crosses fs and net trees.

Note : /proc/sys/fs/file-nr is a read-only file, so this patch doesnt
depend on previous patch (sysctl: fix min/max handling in
__do_proc_doulongvec_minmax())

Thanks !

[PATCH V4] fs: allow for more than 2^31 files

<quote>

</quote>

Acked-by: David Miller <da...@davemloft.net>

Reviewed-by: Robin Holt <ho...@sgi.com>
Tested-by: Robin Holt <ho...@sgi.com>