All of lore.kernel.org
 help / color / mirror / Atom feed
From: "J. Bruce Fields" <bfields@fieldses.org>
To: Sean Finney <sean.finney@sonyericsson.com>
Cc: Steve Dickson <SteveD@redhat.com>,
	"linux-nfs@vger.kernel.org" <linux-nfs@vger.kernel.org>
Subject: Re: [PATCH v2] nfs-utils: Increase the stdio file buffer size for procfs files
Date: Fri, 8 Apr 2011 14:04:44 -0400	[thread overview]
Message-ID: <20110408180444.GA18677@fieldses.org> (raw)
In-Reply-To: <1302261406-9048-2-git-send-email-sean.finney@sonyericsson.com>

Looks sensible to me.--b.

On Fri, Apr 08, 2011 at 01:16:46PM +0200, Sean Finney wrote:
> Previously, when writing to /proc/net/rpc/*/channel, if a cache line
> were larger than the default buffer size (likely 1024 bytes), mountd
> and svcgssd would split writes into a number of buffer-sized writes.
> Each of these writes would get an EINVAL error back from the kernel
> procfs handle (it expects line-oriented input and does not account for
> multiple/split writes), and no cache update would occur.
> 
> When such behavior occurs, NFS clients depending on mountd to finish
> the cache operation would block/hang, or receive EPERM, depending on
> the context of the operation.  This is likely to happen if a user is a
> member of a large (~100-200) number of groups.
> 
> Instead, every fopen() on the procfs files in question is followed by
> a call to setvbuf(), using a per-file dedicated buffer of
> RPC_CHAN_BUF_SIZE length.
> 
> Really, mountd should not be using stdio-style buffered file operations
> on files in /proc to begin with.  A better solution would be to use
> internally managed buffers and calls to write() instead of these stdio
> calls, but that would be a more extensive change; so this is proposed
> as a quick and not-so-dirty fix in the meantime.
> 
> Signed-off-by: Sean Finney <sean.finney@sonyericsson.com>
> ---
>  support/include/misc.h    |    3 +++
>  utils/gssd/svcgssd_proc.c |    3 +++
>  utils/mountd/cache.c      |    2 ++
>  3 files changed, 8 insertions(+), 0 deletions(-)
> 
> diff --git a/support/include/misc.h b/support/include/misc.h
> index 9a1b25d..7e3874e 100644
> --- a/support/include/misc.h
> +++ b/support/include/misc.h
> @@ -24,4 +24,7 @@ struct hostent *get_reliable_hostbyaddr(const char *addr, int len, int type);
>  
>  extern int is_mountpoint(char *path);
>  
> +/* size of the file pointer buffers for rpc procfs files */
> +#define RPC_CHAN_BUF_SIZE 32768
> +
>  #endif /* MISC_H */
> diff --git a/utils/gssd/svcgssd_proc.c b/utils/gssd/svcgssd_proc.c
> index 6f2ba61..8f6548e 100644
> --- a/utils/gssd/svcgssd_proc.c
> +++ b/utils/gssd/svcgssd_proc.c
> @@ -56,6 +56,7 @@
>  #include "gss_util.h"
>  #include "err_util.h"
>  #include "context.h"
> +#include "misc.h"
>  
>  extern char * mech2file(gss_OID mech);
>  #define SVCGSSD_CONTEXT_CHANNEL "/proc/net/rpc/auth.rpcsec.context/channel"
> @@ -78,6 +79,7 @@ do_svc_downcall(gss_buffer_desc *out_handle, struct svc_cred *cred,
>  	FILE *f;
>  	int i;
>  	char *fname = NULL;
> +	char vbuf[RPC_CHAN_BUF_SIZE];
>  	int err;
>  
>  	printerr(1, "doing downcall\n");
> @@ -90,6 +92,7 @@ do_svc_downcall(gss_buffer_desc *out_handle, struct svc_cred *cred,
>  			     SVCGSSD_CONTEXT_CHANNEL, strerror(errno));
>  		goto out_err;
>  	}
> +	setvbuf(f, vbuf, _IOLBF, RPC_CHAN_BUF_SIZE);
>  	qword_printhex(f, out_handle->value, out_handle->length);
>  	/* XXX are types OK for the rest of this? */
>  	/* For context cache, use the actual context endtime */
> diff --git a/utils/mountd/cache.c b/utils/mountd/cache.c
> index 34808cd..bd67157 100644
> --- a/utils/mountd/cache.c
> +++ b/utils/mountd/cache.c
> @@ -739,6 +739,7 @@ struct {
>  	char *cache_name;
>  	void (*cache_handle)(FILE *f);
>  	FILE *f;
> +	char vbuf[RPC_CHAN_BUF_SIZE];
>  } cachelist[] = {
>  	{ "auth.unix.ip", auth_unix_ip},
>  	{ "auth.unix.gid", auth_unix_gid},
> @@ -757,6 +758,7 @@ void cache_open(void)
>  			continue;
>  		sprintf(path, "/proc/net/rpc/%s/channel", cachelist[i].cache_name);
>  		cachelist[i].f = fopen(path, "r+");
> +		setvbuf(cachelist[i].f, cachelist[i].vbuf, _IOLBF, RPC_CHAN_BUF_SIZE);
>  	}
>  }
>  
> -- 
> 1.7.4.1
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

  reply	other threads:[~2011-04-08 18:04 UTC|newest]

Thread overview: 5+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2011-04-08 11:16 [PATCH v2] nfs-utils: mountd: Use a dynamic buffer for storing lists of gid's Sean Finney
2011-04-08 11:16 ` [PATCH v2] nfs-utils: Increase the stdio file buffer size for procfs files Sean Finney
2011-04-08 18:04   ` J. Bruce Fields [this message]
2011-04-21 12:28   ` Steve Dickson
2011-04-21 12:26 ` [PATCH v2] nfs-utils: mountd: Use a dynamic buffer for storing lists of gid's Steve Dickson

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20110408180444.GA18677@fieldses.org \
    --to=bfields@fieldses.org \
    --cc=SteveD@redhat.com \
    --cc=linux-nfs@vger.kernel.org \
    --cc=sean.finney@sonyericsson.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.