WSL2-Linux-Kernel/net/sunrpc/svcauth_unix.c

889 строки
20 KiB
C
Исходник Обычный вид История

#include <linux/types.h>
#include <linux/sched.h>
#include <linux/module.h>
#include <linux/sunrpc/types.h>
#include <linux/sunrpc/xdr.h>
#include <linux/sunrpc/svcsock.h>
#include <linux/sunrpc/svcauth.h>
#include <linux/sunrpc/gss_api.h>
#include <linux/err.h>
#include <linux/seq_file.h>
#include <linux/hash.h>
#include <linux/string.h>
include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit slab.h inclusion from percpu.h percpu.h is included by sched.h and module.h and thus ends up being included when building most .c files. percpu.h includes slab.h which in turn includes gfp.h making everything defined by the two files universally available and complicating inclusion dependencies. percpu.h -> slab.h dependency is about to be removed. Prepare for this change by updating users of gfp and slab facilities include those headers directly instead of assuming availability. As this conversion needs to touch large number of source files, the following script is used as the basis of conversion. http://userweb.kernel.org/~tj/misc/slabh-sweep.py The script does the followings. * Scan files for gfp and slab usages and update includes such that only the necessary includes are there. ie. if only gfp is used, gfp.h, if slab is used, slab.h. * When the script inserts a new include, it looks at the include blocks and try to put the new include such that its order conforms to its surrounding. It's put in the include block which contains core kernel includes, in the same order that the rest are ordered - alphabetical, Christmas tree, rev-Xmas-tree or at the end if there doesn't seem to be any matching order. * If the script can't find a place to put a new include (mostly because the file doesn't have fitting include block), it prints out an error message indicating which .h file needs to be added to the file. The conversion was done in the following steps. 1. The initial automatic conversion of all .c files updated slightly over 4000 files, deleting around 700 includes and adding ~480 gfp.h and ~3000 slab.h inclusions. The script emitted errors for ~400 files. 2. Each error was manually checked. Some didn't need the inclusion, some needed manual addition while adding it to implementation .h or embedding .c file was more appropriate for others. This step added inclusions to around 150 files. 3. The script was run again and the output was compared to the edits from #2 to make sure no file was left behind. 4. Several build tests were done and a couple of problems were fixed. e.g. lib/decompress_*.c used malloc/free() wrappers around slab APIs requiring slab.h to be added manually. 5. The script was run on all .h files but without automatically editing them as sprinkling gfp.h and slab.h inclusions around .h files could easily lead to inclusion dependency hell. Most gfp.h inclusion directives were ignored as stuff from gfp.h was usually wildly available and often used in preprocessor macros. Each slab.h inclusion directive was examined and added manually as necessary. 6. percpu.h was updated not to include slab.h. 7. Build test were done on the following configurations and failures were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my distributed build env didn't work with gcov compiles) and a few more options had to be turned off depending on archs to make things build (like ipr on powerpc/64 which failed due to missing writeq). * x86 and x86_64 UP and SMP allmodconfig and a custom test config. * powerpc and powerpc64 SMP allmodconfig * sparc and sparc64 SMP allmodconfig * ia64 SMP allmodconfig * s390 SMP allmodconfig * alpha SMP allmodconfig * um on x86_64 SMP allmodconfig 8. percpu.h modifications were reverted so that it could be applied as a separate patch and serve as bisection point. Given the fact that I had only a couple of failures from tests on step 6, I'm fairly confident about the coverage of this conversion patch. If there is a breakage, it's likely to be something in one of the arch headers which should be easily discoverable easily on most builds of the specific arch. Signed-off-by: Tejun Heo <tj@kernel.org> Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
2010-03-24 11:04:11 +03:00
#include <linux/slab.h>
[PATCH] knfsd: knfsd: cache ipmap per TCP socket Speed up high call-rate workloads by caching the struct ip_map for the peer on the connected struct svc_sock instead of looking it up in the ip_map cache hashtable on every call. This helps workloads using AUTH_SYS authentication over TCP. Testing was on a 4 CPU 4 NIC Altix using 4 IRIX clients, each with 16 synthetic client threads simulating an rsync (i.e. recursive directory listing) workload reading from an i386 RH9 install image (161480 regular files in 10841 directories) on the server. That tree is small enough to fill in the server's RAM so no disk traffic was involved. This setup gives a sustained call rate in excess of 60000 calls/sec before being CPU-bound on the server. Profiling showed strcmp(), called from ip_map_match(), was taking 4.8% of each CPU, and ip_map_lookup() was taking 2.9%. This patch drops both contribution into the profile noise. Note that the above result overstates this value of this patch for most workloads. The synthetic clients are all using separate IP addresses, so there are 64 entries in the ip_map cache hash. Because the kernel measured contained the bug fixed in commit commit 1f1e030bf75774b6a283518e1534d598e14147d4 and was running on 64bit little-endian machine, probably all of those 64 entries were on a single chain, thus increasing the cost of ip_map_lookup(). With a modern kernel you would need more clients to see the same amount of performance improvement. This patch has helped to scale knfsd to handle a deployment with 2000 NFS clients. Signed-off-by: Greg Banks <gnb@melbourne.sgi.com> Signed-off-by: Neil Brown <neilb@suse.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-10-04 13:15:50 +04:00
#include <net/sock.h>
#include <net/ipv6.h>
#include <linux/kernel.h>
#define RPCDBG_FACILITY RPCDBG_AUTH
#include <linux/sunrpc/clnt.h>
/*
* AUTHUNIX and AUTHNULL credentials are both handled here.
* AUTHNULL is treated just like AUTHUNIX except that the uid/gid
* are always nobody (-2). i.e. we do the same IP address checks for
* AUTHNULL as for AUTHUNIX, and that is done here.
*/
struct unix_domain {
struct auth_domain h;
int addr_changes;
/* other stuff later */
};
extern struct auth_ops svcauth_unix;
struct auth_domain *unix_domain_find(char *name)
{
struct auth_domain *rv;
struct unix_domain *new = NULL;
rv = auth_domain_lookup(name, NULL);
while(1) {
if (rv) {
if (new && rv != &new->h)
auth_domain_put(&new->h);
if (rv->flavour != &svcauth_unix) {
auth_domain_put(rv);
return NULL;
}
return rv;
}
new = kmalloc(sizeof(*new), GFP_KERNEL);
if (new == NULL)
return NULL;
kref_init(&new->h.ref);
new->h.name = kstrdup(name, GFP_KERNEL);
if (new->h.name == NULL) {
kfree(new);
return NULL;
}
new->h.flavour = &svcauth_unix;
new->addr_changes = 0;
rv = auth_domain_lookup(name, &new->h);
}
}
EXPORT_SYMBOL_GPL(unix_domain_find);
static void svcauth_unix_domain_release(struct auth_domain *dom)
{
struct unix_domain *ud = container_of(dom, struct unix_domain, h);
kfree(dom->name);
kfree(ud);
}
/**************************************************
* cache for IP address to unix_domain
* as needed by AUTH_UNIX
*/
#define IP_HASHBITS 8
#define IP_HASHMAX (1<<IP_HASHBITS)
#define IP_HASHMASK (IP_HASHMAX-1)
struct ip_map {
struct cache_head h;
char m_class[8]; /* e.g. "nfsd" */
struct in6_addr m_addr;
struct unix_domain *m_client;
int m_add_change;
};
static struct cache_head *ip_table[IP_HASHMAX];
static void ip_map_put(struct kref *kref)
{
struct cache_head *item = container_of(kref, struct cache_head, ref);
struct ip_map *im = container_of(item, struct ip_map,h);
if (test_bit(CACHE_VALID, &item->flags) &&
!test_bit(CACHE_NEGATIVE, &item->flags))
auth_domain_put(&im->m_client->h);
kfree(im);
}
#if IP_HASHBITS == 8
/* hash_long on a 64 bit machine is currently REALLY BAD for
* IP addresses in reverse-endian (i.e. on a little-endian machine).
* So use a trivial but reliable hash instead
*/
static inline int hash_ip(__be32 ip)
{
int hash = (__force u32)ip ^ ((__force u32)ip>>16);
return (hash ^ (hash>>8)) & 0xff;
}
#endif
static inline int hash_ip6(struct in6_addr ip)
{
return (hash_ip(ip.s6_addr32[0]) ^
hash_ip(ip.s6_addr32[1]) ^
hash_ip(ip.s6_addr32[2]) ^
hash_ip(ip.s6_addr32[3]));
}
static int ip_map_match(struct cache_head *corig, struct cache_head *cnew)
{
struct ip_map *orig = container_of(corig, struct ip_map, h);
struct ip_map *new = container_of(cnew, struct ip_map, h);
return strcmp(orig->m_class, new->m_class) == 0 &&
ipv6_addr_equal(&orig->m_addr, &new->m_addr);
}
static void ip_map_init(struct cache_head *cnew, struct cache_head *citem)
{
struct ip_map *new = container_of(cnew, struct ip_map, h);
struct ip_map *item = container_of(citem, struct ip_map, h);
strcpy(new->m_class, item->m_class);
ipv6_addr_copy(&new->m_addr, &item->m_addr);
}
static void update(struct cache_head *cnew, struct cache_head *citem)
{
struct ip_map *new = container_of(cnew, struct ip_map, h);
struct ip_map *item = container_of(citem, struct ip_map, h);
kref_get(&item->m_client->h.ref);
new->m_client = item->m_client;
new->m_add_change = item->m_add_change;
}
static struct cache_head *ip_map_alloc(void)
{
struct ip_map *i = kmalloc(sizeof(*i), GFP_KERNEL);
if (i)
return &i->h;
else
return NULL;
}
static void ip_map_request(struct cache_detail *cd,
struct cache_head *h,
char **bpp, int *blen)
{
char text_addr[40];
struct ip_map *im = container_of(h, struct ip_map, h);
if (ipv6_addr_v4mapped(&(im->m_addr))) {
snprintf(text_addr, 20, "%pI4", &im->m_addr.s6_addr32[3]);
} else {
snprintf(text_addr, 40, "%pI6", &im->m_addr);
}
qword_add(bpp, blen, im->m_class);
qword_add(bpp, blen, text_addr);
(*bpp)[-1] = '\n';
}
static int ip_map_upcall(struct cache_detail *cd, struct cache_head *h)
{
return sunrpc_cache_pipe_upcall(cd, h, ip_map_request);
}
static struct ip_map *ip_map_lookup(char *class, struct in6_addr *addr);
static int ip_map_update(struct ip_map *ipm, struct unix_domain *udom, time_t expiry);
static int ip_map_parse(struct cache_detail *cd,
char *mesg, int mlen)
{
/* class ipaddress [domainname] */
/* should be safe just to use the start of the input buffer
* for scratch: */
char *buf = mesg;
int len;
char class[8];
union {
struct sockaddr sa;
struct sockaddr_in s4;
struct sockaddr_in6 s6;
} address;
struct sockaddr_in6 sin6;
int err;
struct ip_map *ipmp;
struct auth_domain *dom;
time_t expiry;
if (mesg[mlen-1] != '\n')
return -EINVAL;
mesg[mlen-1] = 0;
/* class */
len = qword_get(&mesg, class, sizeof(class));
if (len <= 0) return -EINVAL;
/* ip address */
len = qword_get(&mesg, buf, mlen);
if (len <= 0) return -EINVAL;
if (rpc_pton(buf, len, &address.sa, sizeof(address)) == 0)
return -EINVAL;
switch (address.sa.sa_family) {
case AF_INET:
/* Form a mapped IPv4 address in sin6 */
memset(&sin6, 0, sizeof(sin6));
sin6.sin6_family = AF_INET6;
sin6.sin6_addr.s6_addr32[2] = htonl(0xffff);
sin6.sin6_addr.s6_addr32[3] = address.s4.sin_addr.s_addr;
break;
#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
case AF_INET6:
memcpy(&sin6, &address.s6, sizeof(sin6));
break;
#endif
default:
return -EINVAL;
}
expiry = get_expiry(&mesg);
if (expiry ==0)
return -EINVAL;
/* domainname, or empty for NEGATIVE */
len = qword_get(&mesg, buf, mlen);
if (len < 0) return -EINVAL;
if (len) {
dom = unix_domain_find(buf);
if (dom == NULL)
return -ENOENT;
} else
dom = NULL;
/* IPv6 scope IDs are ignored for now */
ipmp = ip_map_lookup(class, &sin6.sin6_addr);
if (ipmp) {
err = ip_map_update(ipmp,
container_of(dom, struct unix_domain, h),
expiry);
} else
err = -ENOMEM;
if (dom)
auth_domain_put(dom);
cache_flush();
return err;
}
static int ip_map_show(struct seq_file *m,
struct cache_detail *cd,
struct cache_head *h)
{
struct ip_map *im;
struct in6_addr addr;
char *dom = "-no-domain-";
if (h == NULL) {
seq_puts(m, "#class IP domain\n");
return 0;
}
im = container_of(h, struct ip_map, h);
/* class addr domain */
ipv6_addr_copy(&addr, &im->m_addr);
if (test_bit(CACHE_VALID, &h->flags) &&
!test_bit(CACHE_NEGATIVE, &h->flags))
dom = im->m_client->h.name;
if (ipv6_addr_v4mapped(&addr)) {
seq_printf(m, "%s %pI4 %s\n",
im->m_class, &addr.s6_addr32[3], dom);
} else {
seq_printf(m, "%s %pI6 %s\n", im->m_class, &addr, dom);
}
return 0;
}
struct cache_detail ip_map_cache = {
.owner = THIS_MODULE,
.hash_size = IP_HASHMAX,
.hash_table = ip_table,
.name = "auth.unix.ip",
.cache_put = ip_map_put,
.cache_upcall = ip_map_upcall,
.cache_parse = ip_map_parse,
.cache_show = ip_map_show,
.match = ip_map_match,
.init = ip_map_init,
.update = update,
.alloc = ip_map_alloc,
};
static struct ip_map *ip_map_lookup(char *class, struct in6_addr *addr)
{
struct ip_map ip;
struct cache_head *ch;
strcpy(ip.m_class, class);
ipv6_addr_copy(&ip.m_addr, addr);
ch = sunrpc_cache_lookup(&ip_map_cache, &ip.h,
hash_str(class, IP_HASHBITS) ^
hash_ip6(*addr));
if (ch)
return container_of(ch, struct ip_map, h);
else
return NULL;
}
static int ip_map_update(struct ip_map *ipm, struct unix_domain *udom, time_t expiry)
{
struct ip_map ip;
struct cache_head *ch;
ip.m_client = udom;
ip.h.flags = 0;
if (!udom)
set_bit(CACHE_NEGATIVE, &ip.h.flags);
else {
ip.m_add_change = udom->addr_changes;
/* if this is from the legacy set_client system call,
* we need m_add_change to be one higher
*/
if (expiry == NEVER)
ip.m_add_change++;
}
ip.h.expiry_time = expiry;
ch = sunrpc_cache_update(&ip_map_cache,
&ip.h, &ipm->h,
hash_str(ipm->m_class, IP_HASHBITS) ^
hash_ip6(ipm->m_addr));
if (!ch)
return -ENOMEM;
cache_put(ch, &ip_map_cache);
return 0;
}
int auth_unix_add_addr(struct in6_addr *addr, struct auth_domain *dom)
{
struct unix_domain *udom;
struct ip_map *ipmp;
if (dom->flavour != &svcauth_unix)
return -EINVAL;
udom = container_of(dom, struct unix_domain, h);
ipmp = ip_map_lookup("nfsd", addr);
if (ipmp)
return ip_map_update(ipmp, udom, NEVER);
else
return -ENOMEM;
}
EXPORT_SYMBOL_GPL(auth_unix_add_addr);
int auth_unix_forget_old(struct auth_domain *dom)
{
struct unix_domain *udom;
if (dom->flavour != &svcauth_unix)
return -EINVAL;
udom = container_of(dom, struct unix_domain, h);
udom->addr_changes++;
return 0;
}
EXPORT_SYMBOL_GPL(auth_unix_forget_old);
struct auth_domain *auth_unix_lookup(struct in6_addr *addr)
{
struct ip_map *ipm;
struct auth_domain *rv;
ipm = ip_map_lookup("nfsd", addr);
if (!ipm)
return NULL;
if (cache_check(&ip_map_cache, &ipm->h, NULL))
return NULL;
if ((ipm->m_client->addr_changes - ipm->m_add_change) >0) {
if (test_and_set_bit(CACHE_NEGATIVE, &ipm->h.flags) == 0)
auth_domain_put(&ipm->m_client->h);
rv = NULL;
} else {
rv = &ipm->m_client->h;
kref_get(&rv->ref);
}
cache_put(&ipm->h, &ip_map_cache);
return rv;
}
EXPORT_SYMBOL_GPL(auth_unix_lookup);
void svcauth_unix_purge(void)
{
cache_purge(&ip_map_cache);
}
EXPORT_SYMBOL_GPL(svcauth_unix_purge);
[PATCH] knfsd: knfsd: cache ipmap per TCP socket Speed up high call-rate workloads by caching the struct ip_map for the peer on the connected struct svc_sock instead of looking it up in the ip_map cache hashtable on every call. This helps workloads using AUTH_SYS authentication over TCP. Testing was on a 4 CPU 4 NIC Altix using 4 IRIX clients, each with 16 synthetic client threads simulating an rsync (i.e. recursive directory listing) workload reading from an i386 RH9 install image (161480 regular files in 10841 directories) on the server. That tree is small enough to fill in the server's RAM so no disk traffic was involved. This setup gives a sustained call rate in excess of 60000 calls/sec before being CPU-bound on the server. Profiling showed strcmp(), called from ip_map_match(), was taking 4.8% of each CPU, and ip_map_lookup() was taking 2.9%. This patch drops both contribution into the profile noise. Note that the above result overstates this value of this patch for most workloads. The synthetic clients are all using separate IP addresses, so there are 64 entries in the ip_map cache hash. Because the kernel measured contained the bug fixed in commit commit 1f1e030bf75774b6a283518e1534d598e14147d4 and was running on 64bit little-endian machine, probably all of those 64 entries were on a single chain, thus increasing the cost of ip_map_lookup(). With a modern kernel you would need more clients to see the same amount of performance improvement. This patch has helped to scale knfsd to handle a deployment with 2000 NFS clients. Signed-off-by: Greg Banks <gnb@melbourne.sgi.com> Signed-off-by: Neil Brown <neilb@suse.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-10-04 13:15:50 +04:00
static inline struct ip_map *
ip_map_cached_get(struct svc_rqst *rqstp)
{
struct ip_map *ipm = NULL;
struct svc_xprt *xprt = rqstp->rq_xprt;
if (test_bit(XPT_CACHE_AUTH, &xprt->xpt_flags)) {
spin_lock(&xprt->xpt_lock);
ipm = xprt->xpt_auth_cache;
if (ipm != NULL) {
if (!cache_valid(&ipm->h)) {
/*
* The entry has been invalidated since it was
* remembered, e.g. by a second mount from the
* same IP address.
*/
xprt->xpt_auth_cache = NULL;
spin_unlock(&xprt->xpt_lock);
cache_put(&ipm->h, &ip_map_cache);
return NULL;
}
cache_get(&ipm->h);
[PATCH] knfsd: knfsd: cache ipmap per TCP socket Speed up high call-rate workloads by caching the struct ip_map for the peer on the connected struct svc_sock instead of looking it up in the ip_map cache hashtable on every call. This helps workloads using AUTH_SYS authentication over TCP. Testing was on a 4 CPU 4 NIC Altix using 4 IRIX clients, each with 16 synthetic client threads simulating an rsync (i.e. recursive directory listing) workload reading from an i386 RH9 install image (161480 regular files in 10841 directories) on the server. That tree is small enough to fill in the server's RAM so no disk traffic was involved. This setup gives a sustained call rate in excess of 60000 calls/sec before being CPU-bound on the server. Profiling showed strcmp(), called from ip_map_match(), was taking 4.8% of each CPU, and ip_map_lookup() was taking 2.9%. This patch drops both contribution into the profile noise. Note that the above result overstates this value of this patch for most workloads. The synthetic clients are all using separate IP addresses, so there are 64 entries in the ip_map cache hash. Because the kernel measured contained the bug fixed in commit commit 1f1e030bf75774b6a283518e1534d598e14147d4 and was running on 64bit little-endian machine, probably all of those 64 entries were on a single chain, thus increasing the cost of ip_map_lookup(). With a modern kernel you would need more clients to see the same amount of performance improvement. This patch has helped to scale knfsd to handle a deployment with 2000 NFS clients. Signed-off-by: Greg Banks <gnb@melbourne.sgi.com> Signed-off-by: Neil Brown <neilb@suse.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-10-04 13:15:50 +04:00
}
spin_unlock(&xprt->xpt_lock);
[PATCH] knfsd: knfsd: cache ipmap per TCP socket Speed up high call-rate workloads by caching the struct ip_map for the peer on the connected struct svc_sock instead of looking it up in the ip_map cache hashtable on every call. This helps workloads using AUTH_SYS authentication over TCP. Testing was on a 4 CPU 4 NIC Altix using 4 IRIX clients, each with 16 synthetic client threads simulating an rsync (i.e. recursive directory listing) workload reading from an i386 RH9 install image (161480 regular files in 10841 directories) on the server. That tree is small enough to fill in the server's RAM so no disk traffic was involved. This setup gives a sustained call rate in excess of 60000 calls/sec before being CPU-bound on the server. Profiling showed strcmp(), called from ip_map_match(), was taking 4.8% of each CPU, and ip_map_lookup() was taking 2.9%. This patch drops both contribution into the profile noise. Note that the above result overstates this value of this patch for most workloads. The synthetic clients are all using separate IP addresses, so there are 64 entries in the ip_map cache hash. Because the kernel measured contained the bug fixed in commit commit 1f1e030bf75774b6a283518e1534d598e14147d4 and was running on 64bit little-endian machine, probably all of those 64 entries were on a single chain, thus increasing the cost of ip_map_lookup(). With a modern kernel you would need more clients to see the same amount of performance improvement. This patch has helped to scale knfsd to handle a deployment with 2000 NFS clients. Signed-off-by: Greg Banks <gnb@melbourne.sgi.com> Signed-off-by: Neil Brown <neilb@suse.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-10-04 13:15:50 +04:00
}
return ipm;
}
static inline void
ip_map_cached_put(struct svc_rqst *rqstp, struct ip_map *ipm)
{
struct svc_xprt *xprt = rqstp->rq_xprt;
if (test_bit(XPT_CACHE_AUTH, &xprt->xpt_flags)) {
spin_lock(&xprt->xpt_lock);
if (xprt->xpt_auth_cache == NULL) {
/* newly cached, keep the reference */
xprt->xpt_auth_cache = ipm;
ipm = NULL;
}
spin_unlock(&xprt->xpt_lock);
}
if (ipm)
[PATCH] knfsd: knfsd: cache ipmap per TCP socket Speed up high call-rate workloads by caching the struct ip_map for the peer on the connected struct svc_sock instead of looking it up in the ip_map cache hashtable on every call. This helps workloads using AUTH_SYS authentication over TCP. Testing was on a 4 CPU 4 NIC Altix using 4 IRIX clients, each with 16 synthetic client threads simulating an rsync (i.e. recursive directory listing) workload reading from an i386 RH9 install image (161480 regular files in 10841 directories) on the server. That tree is small enough to fill in the server's RAM so no disk traffic was involved. This setup gives a sustained call rate in excess of 60000 calls/sec before being CPU-bound on the server. Profiling showed strcmp(), called from ip_map_match(), was taking 4.8% of each CPU, and ip_map_lookup() was taking 2.9%. This patch drops both contribution into the profile noise. Note that the above result overstates this value of this patch for most workloads. The synthetic clients are all using separate IP addresses, so there are 64 entries in the ip_map cache hash. Because the kernel measured contained the bug fixed in commit commit 1f1e030bf75774b6a283518e1534d598e14147d4 and was running on 64bit little-endian machine, probably all of those 64 entries were on a single chain, thus increasing the cost of ip_map_lookup(). With a modern kernel you would need more clients to see the same amount of performance improvement. This patch has helped to scale knfsd to handle a deployment with 2000 NFS clients. Signed-off-by: Greg Banks <gnb@melbourne.sgi.com> Signed-off-by: Neil Brown <neilb@suse.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-10-04 13:15:50 +04:00
cache_put(&ipm->h, &ip_map_cache);
}
void
svcauth_unix_info_release(void *info)
{
struct ip_map *ipm = info;
cache_put(&ipm->h, &ip_map_cache);
}
/****************************************************************************
* auth.unix.gid cache
* simple cache to map a UID to a list of GIDs
* because AUTH_UNIX aka AUTH_SYS has a max of 16
*/
#define GID_HASHBITS 8
#define GID_HASHMAX (1<<GID_HASHBITS)
#define GID_HASHMASK (GID_HASHMAX - 1)
struct unix_gid {
struct cache_head h;
uid_t uid;
struct group_info *gi;
};
static struct cache_head *gid_table[GID_HASHMAX];
static void unix_gid_put(struct kref *kref)
{
struct cache_head *item = container_of(kref, struct cache_head, ref);
struct unix_gid *ug = container_of(item, struct unix_gid, h);
if (test_bit(CACHE_VALID, &item->flags) &&
!test_bit(CACHE_NEGATIVE, &item->flags))
put_group_info(ug->gi);
kfree(ug);
}
static int unix_gid_match(struct cache_head *corig, struct cache_head *cnew)
{
struct unix_gid *orig = container_of(corig, struct unix_gid, h);
struct unix_gid *new = container_of(cnew, struct unix_gid, h);
return orig->uid == new->uid;
}
static void unix_gid_init(struct cache_head *cnew, struct cache_head *citem)
{
struct unix_gid *new = container_of(cnew, struct unix_gid, h);
struct unix_gid *item = container_of(citem, struct unix_gid, h);
new->uid = item->uid;
}
static void unix_gid_update(struct cache_head *cnew, struct cache_head *citem)
{
struct unix_gid *new = container_of(cnew, struct unix_gid, h);
struct unix_gid *item = container_of(citem, struct unix_gid, h);
get_group_info(item->gi);
new->gi = item->gi;
}
static struct cache_head *unix_gid_alloc(void)
{
struct unix_gid *g = kmalloc(sizeof(*g), GFP_KERNEL);
if (g)
return &g->h;
else
return NULL;
}
static void unix_gid_request(struct cache_detail *cd,
struct cache_head *h,
char **bpp, int *blen)
{
char tuid[20];
struct unix_gid *ug = container_of(h, struct unix_gid, h);
snprintf(tuid, 20, "%u", ug->uid);
qword_add(bpp, blen, tuid);
(*bpp)[-1] = '\n';
}
static int unix_gid_upcall(struct cache_detail *cd, struct cache_head *h)
{
return sunrpc_cache_pipe_upcall(cd, h, unix_gid_request);
}
static struct unix_gid *unix_gid_lookup(uid_t uid);
extern struct cache_detail unix_gid_cache;
static int unix_gid_parse(struct cache_detail *cd,
char *mesg, int mlen)
{
/* uid expiry Ngid gid0 gid1 ... gidN-1 */
int uid;
int gids;
int rv;
int i;
int err;
time_t expiry;
struct unix_gid ug, *ugp;
if (mlen <= 0 || mesg[mlen-1] != '\n')
return -EINVAL;
mesg[mlen-1] = 0;
rv = get_int(&mesg, &uid);
if (rv)
return -EINVAL;
ug.uid = uid;
expiry = get_expiry(&mesg);
if (expiry == 0)
return -EINVAL;
rv = get_int(&mesg, &gids);
if (rv || gids < 0 || gids > 8192)
return -EINVAL;
ug.gi = groups_alloc(gids);
if (!ug.gi)
return -ENOMEM;
for (i = 0 ; i < gids ; i++) {
int gid;
rv = get_int(&mesg, &gid);
err = -EINVAL;
if (rv)
goto out;
GROUP_AT(ug.gi, i) = gid;
}
ugp = unix_gid_lookup(uid);
if (ugp) {
struct cache_head *ch;
ug.h.flags = 0;
ug.h.expiry_time = expiry;
ch = sunrpc_cache_update(&unix_gid_cache,
&ug.h, &ugp->h,
hash_long(uid, GID_HASHBITS));
if (!ch)
err = -ENOMEM;
else {
err = 0;
cache_put(ch, &unix_gid_cache);
}
} else
err = -ENOMEM;
out:
if (ug.gi)
put_group_info(ug.gi);
return err;
}
static int unix_gid_show(struct seq_file *m,
struct cache_detail *cd,
struct cache_head *h)
{
struct unix_gid *ug;
int i;
int glen;
if (h == NULL) {
seq_puts(m, "#uid cnt: gids...\n");
return 0;
}
ug = container_of(h, struct unix_gid, h);
if (test_bit(CACHE_VALID, &h->flags) &&
!test_bit(CACHE_NEGATIVE, &h->flags))
glen = ug->gi->ngroups;
else
glen = 0;
seq_printf(m, "%u %d:", ug->uid, glen);
for (i = 0; i < glen; i++)
seq_printf(m, " %d", GROUP_AT(ug->gi, i));
seq_printf(m, "\n");
return 0;
}
struct cache_detail unix_gid_cache = {
.owner = THIS_MODULE,
.hash_size = GID_HASHMAX,
.hash_table = gid_table,
.name = "auth.unix.gid",
.cache_put = unix_gid_put,
.cache_upcall = unix_gid_upcall,
.cache_parse = unix_gid_parse,
.cache_show = unix_gid_show,
.match = unix_gid_match,
.init = unix_gid_init,
.update = unix_gid_update,
.alloc = unix_gid_alloc,
};
static struct unix_gid *unix_gid_lookup(uid_t uid)
{
struct unix_gid ug;
struct cache_head *ch;
ug.uid = uid;
ch = sunrpc_cache_lookup(&unix_gid_cache, &ug.h,
hash_long(uid, GID_HASHBITS));
if (ch)
return container_of(ch, struct unix_gid, h);
else
return NULL;
}
static struct group_info *unix_gid_find(uid_t uid, struct svc_rqst *rqstp)
{
struct unix_gid *ug;
struct group_info *gi;
int ret;
ug = unix_gid_lookup(uid);
if (!ug)
return ERR_PTR(-EAGAIN);
ret = cache_check(&unix_gid_cache, &ug->h, &rqstp->rq_chandle);
switch (ret) {
case -ENOENT:
return ERR_PTR(-ENOENT);
case 0:
gi = get_group_info(ug->gi);
cache_put(&ug->h, &unix_gid_cache);
return gi;
default:
return ERR_PTR(-EAGAIN);
}
}
knfsd: nfsd: set rq_client to ip-address-determined-domain We want it to be possible for users to restrict exports both by IP address and by pseudoflavor. The pseudoflavor information has previously been passed using special auth_domains stored in the rq_client field. After the preceding patch that stored the pseudoflavor in rq_pflavor, that's now superfluous; so now we use rq_client for the ip information, as auth_null and auth_unix do. However, we keep around the special auth_domain in the rq_gssclient field for backwards compatibility purposes, so we can still do upcalls using the old "gss/pseudoflavor" auth_domain if upcalls using the unix domain to give us an appropriate export. This allows us to continue supporting old mountd. In fact, for this first patch, we always use the "gss/pseudoflavor" auth_domain (and only it) if it is available; thus rq_client is ignored in the auth_gss case, and this patch on its own makes no change in behavior; that will be left to later patches. Note on idmap: I'm almost tempted to just replace the auth_domain in the idmap upcall by a dummy value--no version of idmapd has ever used it, and it's unlikely anyone really wants to perform idmapping differently depending on the where the client is (they may want to perform *credential* mapping differently, but that's a different matter--the idmapper just handles id's used in getattr and setattr). But I'm updating the idmapd code anyway, just out of general backwards-compatibility paranoia. Signed-off-by: "J. Bruce Fields" <bfields@citi.umich.edu> Signed-off-by: Neil Brown <neilb@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-17 15:04:46 +04:00
int
svcauth_unix_set_client(struct svc_rqst *rqstp)
{
struct sockaddr_in *sin;
struct sockaddr_in6 *sin6, sin6_storage;
struct ip_map *ipm;
struct group_info *gi;
struct svc_cred *cred = &rqstp->rq_cred;
switch (rqstp->rq_addr.ss_family) {
case AF_INET:
sin = svc_addr_in(rqstp);
sin6 = &sin6_storage;
ipv6_addr_set_v4mapped(sin->sin_addr.s_addr, &sin6->sin6_addr);
break;
case AF_INET6:
sin6 = svc_addr_in6(rqstp);
break;
default:
BUG();
}
rqstp->rq_client = NULL;
if (rqstp->rq_proc == 0)
return SVC_OK;
[PATCH] knfsd: knfsd: cache ipmap per TCP socket Speed up high call-rate workloads by caching the struct ip_map for the peer on the connected struct svc_sock instead of looking it up in the ip_map cache hashtable on every call. This helps workloads using AUTH_SYS authentication over TCP. Testing was on a 4 CPU 4 NIC Altix using 4 IRIX clients, each with 16 synthetic client threads simulating an rsync (i.e. recursive directory listing) workload reading from an i386 RH9 install image (161480 regular files in 10841 directories) on the server. That tree is small enough to fill in the server's RAM so no disk traffic was involved. This setup gives a sustained call rate in excess of 60000 calls/sec before being CPU-bound on the server. Profiling showed strcmp(), called from ip_map_match(), was taking 4.8% of each CPU, and ip_map_lookup() was taking 2.9%. This patch drops both contribution into the profile noise. Note that the above result overstates this value of this patch for most workloads. The synthetic clients are all using separate IP addresses, so there are 64 entries in the ip_map cache hash. Because the kernel measured contained the bug fixed in commit commit 1f1e030bf75774b6a283518e1534d598e14147d4 and was running on 64bit little-endian machine, probably all of those 64 entries were on a single chain, thus increasing the cost of ip_map_lookup(). With a modern kernel you would need more clients to see the same amount of performance improvement. This patch has helped to scale knfsd to handle a deployment with 2000 NFS clients. Signed-off-by: Greg Banks <gnb@melbourne.sgi.com> Signed-off-by: Neil Brown <neilb@suse.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-10-04 13:15:50 +04:00
ipm = ip_map_cached_get(rqstp);
if (ipm == NULL)
ipm = ip_map_lookup(rqstp->rq_server->sv_program->pg_class,
&sin6->sin6_addr);
if (ipm == NULL)
return SVC_DENIED;
switch (cache_check(&ip_map_cache, &ipm->h, &rqstp->rq_chandle)) {
default:
BUG();
case -EAGAIN:
case -ETIMEDOUT:
return SVC_DROP;
case -ENOENT:
return SVC_DENIED;
case 0:
rqstp->rq_client = &ipm->m_client->h;
kref_get(&rqstp->rq_client->ref);
[PATCH] knfsd: knfsd: cache ipmap per TCP socket Speed up high call-rate workloads by caching the struct ip_map for the peer on the connected struct svc_sock instead of looking it up in the ip_map cache hashtable on every call. This helps workloads using AUTH_SYS authentication over TCP. Testing was on a 4 CPU 4 NIC Altix using 4 IRIX clients, each with 16 synthetic client threads simulating an rsync (i.e. recursive directory listing) workload reading from an i386 RH9 install image (161480 regular files in 10841 directories) on the server. That tree is small enough to fill in the server's RAM so no disk traffic was involved. This setup gives a sustained call rate in excess of 60000 calls/sec before being CPU-bound on the server. Profiling showed strcmp(), called from ip_map_match(), was taking 4.8% of each CPU, and ip_map_lookup() was taking 2.9%. This patch drops both contribution into the profile noise. Note that the above result overstates this value of this patch for most workloads. The synthetic clients are all using separate IP addresses, so there are 64 entries in the ip_map cache hash. Because the kernel measured contained the bug fixed in commit commit 1f1e030bf75774b6a283518e1534d598e14147d4 and was running on 64bit little-endian machine, probably all of those 64 entries were on a single chain, thus increasing the cost of ip_map_lookup(). With a modern kernel you would need more clients to see the same amount of performance improvement. This patch has helped to scale knfsd to handle a deployment with 2000 NFS clients. Signed-off-by: Greg Banks <gnb@melbourne.sgi.com> Signed-off-by: Neil Brown <neilb@suse.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-10-04 13:15:50 +04:00
ip_map_cached_put(rqstp, ipm);
break;
}
gi = unix_gid_find(cred->cr_uid, rqstp);
switch (PTR_ERR(gi)) {
case -EAGAIN:
return SVC_DROP;
case -ENOENT:
break;
default:
put_group_info(cred->cr_group_info);
cred->cr_group_info = gi;
}
return SVC_OK;
}
EXPORT_SYMBOL_GPL(svcauth_unix_set_client);
knfsd: nfsd: set rq_client to ip-address-determined-domain We want it to be possible for users to restrict exports both by IP address and by pseudoflavor. The pseudoflavor information has previously been passed using special auth_domains stored in the rq_client field. After the preceding patch that stored the pseudoflavor in rq_pflavor, that's now superfluous; so now we use rq_client for the ip information, as auth_null and auth_unix do. However, we keep around the special auth_domain in the rq_gssclient field for backwards compatibility purposes, so we can still do upcalls using the old "gss/pseudoflavor" auth_domain if upcalls using the unix domain to give us an appropriate export. This allows us to continue supporting old mountd. In fact, for this first patch, we always use the "gss/pseudoflavor" auth_domain (and only it) if it is available; thus rq_client is ignored in the auth_gss case, and this patch on its own makes no change in behavior; that will be left to later patches. Note on idmap: I'm almost tempted to just replace the auth_domain in the idmap upcall by a dummy value--no version of idmapd has ever used it, and it's unlikely anyone really wants to perform idmapping differently depending on the where the client is (they may want to perform *credential* mapping differently, but that's a different matter--the idmapper just handles id's used in getattr and setattr). But I'm updating the idmapd code anyway, just out of general backwards-compatibility paranoia. Signed-off-by: "J. Bruce Fields" <bfields@citi.umich.edu> Signed-off-by: Neil Brown <neilb@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-07-17 15:04:46 +04:00
static int
svcauth_null_accept(struct svc_rqst *rqstp, __be32 *authp)
{
struct kvec *argv = &rqstp->rq_arg.head[0];
struct kvec *resv = &rqstp->rq_res.head[0];
struct svc_cred *cred = &rqstp->rq_cred;
cred->cr_group_info = NULL;
rqstp->rq_client = NULL;
if (argv->iov_len < 3*4)
return SVC_GARBAGE;
if (svc_getu32(argv) != 0) {
dprintk("svc: bad null cred\n");
*authp = rpc_autherr_badcred;
return SVC_DENIED;
}
if (svc_getu32(argv) != htonl(RPC_AUTH_NULL) || svc_getu32(argv) != 0) {
dprintk("svc: bad null verf\n");
*authp = rpc_autherr_badverf;
return SVC_DENIED;
}
/* Signal that mapping to nobody uid/gid is required */
cred->cr_uid = (uid_t) -1;
cred->cr_gid = (gid_t) -1;
cred->cr_group_info = groups_alloc(0);
if (cred->cr_group_info == NULL)
return SVC_DROP; /* kmalloc failure - client must retry */
/* Put NULL verifier */
svc_putnl(resv, RPC_AUTH_NULL);
svc_putnl(resv, 0);
rqstp->rq_flavor = RPC_AUTH_NULL;
return SVC_OK;
}
static int
svcauth_null_release(struct svc_rqst *rqstp)
{
if (rqstp->rq_client)
auth_domain_put(rqstp->rq_client);
rqstp->rq_client = NULL;
if (rqstp->rq_cred.cr_group_info)
put_group_info(rqstp->rq_cred.cr_group_info);
rqstp->rq_cred.cr_group_info = NULL;
return 0; /* don't drop */
}
struct auth_ops svcauth_null = {
.name = "null",
.owner = THIS_MODULE,
.flavour = RPC_AUTH_NULL,
.accept = svcauth_null_accept,
.release = svcauth_null_release,
.set_client = svcauth_unix_set_client,
};
static int
svcauth_unix_accept(struct svc_rqst *rqstp, __be32 *authp)
{
struct kvec *argv = &rqstp->rq_arg.head[0];
struct kvec *resv = &rqstp->rq_res.head[0];
struct svc_cred *cred = &rqstp->rq_cred;
u32 slen, i;
int len = argv->iov_len;
cred->cr_group_info = NULL;
rqstp->rq_client = NULL;
if ((len -= 3*4) < 0)
return SVC_GARBAGE;
svc_getu32(argv); /* length */
svc_getu32(argv); /* time stamp */
slen = XDR_QUADLEN(svc_getnl(argv)); /* machname length */
if (slen > 64 || (len -= (slen + 3)*4) < 0)
goto badcred;
argv->iov_base = (void*)((__be32*)argv->iov_base + slen); /* skip machname */
argv->iov_len -= slen*4;
cred->cr_uid = svc_getnl(argv); /* uid */
cred->cr_gid = svc_getnl(argv); /* gid */
slen = svc_getnl(argv); /* gids length */
if (slen > 16 || (len -= (slen + 2)*4) < 0)
goto badcred;
cred->cr_group_info = groups_alloc(slen);
if (cred->cr_group_info == NULL)
return SVC_DROP;
for (i = 0; i < slen; i++)
GROUP_AT(cred->cr_group_info, i) = svc_getnl(argv);
if (svc_getu32(argv) != htonl(RPC_AUTH_NULL) || svc_getu32(argv) != 0) {
*authp = rpc_autherr_badverf;
return SVC_DENIED;
}
/* Put NULL verifier */
svc_putnl(resv, RPC_AUTH_NULL);
svc_putnl(resv, 0);
rqstp->rq_flavor = RPC_AUTH_UNIX;
return SVC_OK;
badcred:
*authp = rpc_autherr_badcred;
return SVC_DENIED;
}
static int
svcauth_unix_release(struct svc_rqst *rqstp)
{
/* Verifier (such as it is) is already in place.
*/
if (rqstp->rq_client)
auth_domain_put(rqstp->rq_client);
rqstp->rq_client = NULL;
if (rqstp->rq_cred.cr_group_info)
put_group_info(rqstp->rq_cred.cr_group_info);
rqstp->rq_cred.cr_group_info = NULL;
return 0;
}
struct auth_ops svcauth_unix = {
.name = "unix",
.owner = THIS_MODULE,
.flavour = RPC_AUTH_UNIX,
.accept = svcauth_unix_accept,
.release = svcauth_unix_release,
.domain_release = svcauth_unix_domain_release,
.set_client = svcauth_unix_set_client,
};