MIME-version: 1.0
Content-transfer-encoding: 7BIT
Content-type: text/plain; CHARSET=US-ASCII; delsp=yes; format=flowed
Date: Mon, 11 Jan 2010 08:40:52 -0400
From: Andreas Dilger <adilger@sun.com>
Subject: Re: [RFC][PATCHv3] List per-process file descriptor consumption when
 hitting file-max
In-reply-to: <1263202687-18529-1-git-send-email-virtuoso@slind.org>
To: Alexander Shishkin <virtuoso@slind.org>
Cc: Valdis.Kletnieks@vt.edu, linux-fsdevel@vger.kernel.org,
       akpm@linux-foundation.org, linux-kernel@vger.kernel.org,
       viro@zeniv.linux.org.uk
Message-id: <64A9A867-32E6-430C-A1A5-C515102D069D@sun.com>
References: <28675.1248957636@turing-police.cc.vt.edu>
 <1263202687-18529-1-git-send-email-virtuoso@slind.org>
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 6515
Lines: 235

On 2010-01-11, at 05:38, Alexander Shishkin wrote:
> When a file descriptor limit is hit, display the top consumers of
> descriptors so that it is possible to identify and fix those which
> leak them.
>
> Two new sysctl tunables are introduced:
>  * file-max-consumers -- number of processes to display (defaults
>    to 10);
>  * file-max-rate-limit -- time interval between subsequent dumps
>    (defaults to 10 seconds).

This should default to max_consumers=0 to avoid spamming the logs, IMHO.

> Signed-off-by: Alexander Shishkin <virtuoso@slind.org>
> CC: viro@zeniv.linux.org.uk
> CC: linux-fsdevel@vger.kernel.org
> ---
> Changes:
> v3 -- fix a couple of silly checkpatch errors
> v2 -- add rate-limiting and reduce number of processes to be output
> v1 -- initial implementation.
>
> fs/file_table.c    |   89 +++++++++++++++++++++++++++++++++++++++++++ 
> ++++++++-
> include/linux/fs.h |    5 +++
> kernel/sysctl.c    |   14 ++++++++
> 3 files changed, 107 insertions(+), 1 deletions(-)
>
> diff --git a/fs/file_table.c b/fs/file_table.c
> index 69652c5..26666fd 100644
> --- a/fs/file_table.c
> +++ b/fs/file_table.c
> @@ -9,6 +9,7 @@
> #include <linux/slab.h>
> #include <linux/file.h>
> #include <linux/fdtable.h>
> +#include <linux/sort.h>
> #include <linux/init.h>
> #include <linux/module.h>
> #include <linux/fs.h>
> @@ -29,7 +30,8 @@
>
> /* sysctl tunables... */
> struct files_stat_struct files_stat = {
> -	.max_files = NR_FILE
> +	.max_files = NR_FILE,
> +	.max_consumers = NR_CONSUMERS,
> };
>
> /* public. Not pretty! */
> @@ -90,6 +92,80 @@ int proc_nr_files(ctl_table *table, int write,
> }
> #endif
>
> +/*
> + * Number of open file descriptors per task_struct
> + */
> +struct fd_consumer {
> +	struct task_struct *task;
> +	int fd_count;
> +};
> +
> +static int cmp_fd_consumers(const void *a, const void *b)
> +{
> +	const struct fd_consumer *x = a, *y = b;
> +
> +	return y->fd_count - x->fd_count;
> +}
> +
> +static void dump_fd_consumers(void)
> +{
> +	struct task_struct *p;
> +	struct files_struct *files;
> +	struct fdtable *fdt;
> +	int proc_limit = files_stat.max_consumers;
> +	int i, nproc;
> +	struct fd_consumer *procs, *tmp;
> +
> +	if (!files_stat.max_consumers)
> +		return;
> +
> +	read_lock(&tasklist_lock);
> +
> +	/* build an array of per-task file descriptor usage */
> +	nproc = nr_processes();
> +	procs = kzalloc(nproc * sizeof(struct fd_consumer), GFP_KERNEL);
> +	if (!procs)
> +		goto out;
> +
> +	tmp = procs;
> +
> +	for_each_process(p) {
> +		tmp->task = p;
> +
> +		files = get_files_struct(p);
> +		if (!files)
> +			continue;
> +
> +		spin_lock(&files->file_lock);
> +		fdt = files_fdtable(files);
> +
> +		/* we have to actually *count* the fds */
> +		for (tmp->fd_count = i = 0; i < fdt->max_fds; i++)
> +			tmp->fd_count += !!fcheck_files(files, i);
> +
> +		spin_unlock(&files->file_lock);
> +		put_files_struct(files);
> +
> +		tmp++;
> +	}
> +
> +	/* sort by number of used descriptor in descending order */
> +	sort(procs, nproc, sizeof(struct fd_consumer), cmp_fd_consumers,  
> NULL);
> +
> +	if (proc_limit > nproc)
> +		proc_limit = nproc;
> +
> +	/* output the 'proc_limit' first entries */
> +	for (i = 0, tmp = procs; i < proc_limit; i++, tmp++)
> +		printk(KERN_INFO "=> %s [%d]: open=%d\n", tmp->task->comm,
> +		       tmp->task->pid, tmp->fd_count);
> +
> +	kfree(procs);
> +
> +out:
> +	read_unlock(&tasklist_lock);
> +}
> +
> /* Find an unused file structure and return a pointer to it.
>  * Returns NULL, if there are no more free file structures or
>  * we run out of memory.
> @@ -105,6 +181,7 @@ struct file *get_empty_filp(void)
> 	const struct cred *cred = current_cred();
> 	static int old_max;
> 	struct file * f;
> +	static unsigned long next_dump;
>
> 	/*
> 	 * Privileged users can go above max_files
> @@ -140,6 +217,14 @@ over:
> 	if (get_nr_files() > old_max) {
> 		printk(KERN_INFO "VFS: file-max limit %d reached\n",
> 					get_max_files());
> +
> +		/* dump the biggest file descriptor users */
> +		if (!next_dump || time_after(jiffies, next_dump)) {
> +			next_dump = jiffies + files_stat.rate_limit;
> +
> +			dump_fd_consumers();
> +		}
> +
> 		old_max = get_nr_files();
> 	}
> 	goto fail;
> @@ -425,6 +510,8 @@ void __init files_init(unsigned long mempages)
> 	files_stat.max_files = n;
> 	if (files_stat.max_files < NR_FILE)
> 		files_stat.max_files = NR_FILE;
> +
> +	files_stat.rate_limit = DUMP_RATE_LIMIT;
> 	files_defer_init();
> 	percpu_counter_init(&nr_files, 0);
> }
> diff --git a/include/linux/fs.h b/include/linux/fs.h
> index 9147ca8..291beb3 100644
> --- a/include/linux/fs.h
> +++ b/include/linux/fs.h
> @@ -36,6 +36,8 @@ struct files_stat_struct {
> 	int nr_files;		/* read only */
> 	int nr_free_files;	/* read only */
> 	int max_files;		/* tunable */
> +	int max_consumers;	/* tunable */
> +	unsigned long rate_limit;	/* tunable */
> };
>
> struct inodes_stat_t {
> @@ -46,6 +48,9 @@ struct inodes_stat_t {
>
>
> #define NR_FILE  8192	/* this can well be larger on a larger system */
> +#define NR_CONSUMERS 10 /* dump this many tasks when file-max is  
> hit */
> +#define DUMP_RATE_LIMIT msecs_to_jiffies(10000) /* wait this long  
> between
> +						   dumps */
>
> #define MAY_EXEC 1
> #define MAY_WRITE 2
> diff --git a/kernel/sysctl.c b/kernel/sysctl.c
> index 8a68b24..dfb08fc 100644
> --- a/kernel/sysctl.c
> +++ b/kernel/sysctl.c
> @@ -1325,6 +1325,20 @@ static struct ctl_table fs_table[] = {
> 		.proc_handler	= proc_dointvec,
> 	},
> 	{
> +		.procname	= "file-max-consumers",
> +		.data		= &files_stat.max_consumers,
> +		.maxlen		= sizeof(int),
> +		.mode		= 0644,
> +		.proc_handler	= proc_dointvec,
> +	},
> +	{
> +		.procname	= "file-max-rate-limit",
> +		.data		= &files_stat.rate_limit,
> +		.maxlen		= sizeof(unsigned long),
> +		.mode		= 0644,
> +		.proc_handler	= proc_doulongvec_ms_jiffies_minmax,
> +	},
> +	{
> 		.procname	= "nr_open",
> 		.data		= &sysctl_nr_open,
> 		.maxlen		= sizeof(int),
> -- 
> 1.6.5
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux- 
> fsdevel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html


Cheers, Andreas
--
Andreas Dilger
Sr. Staff Engineer, Lustre Group
Sun Microsystems of Canada, Inc.

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