nsproxy: incorporate fs namespace
Subsystems (5)
A message that is not in the archive (sent off-list or not yet received): [email protected]
PATCH 2/9 nsproxy: incorporate fs namespace
Incorporate fs namespace into nsproxy.
---
fs/namespace.c | 21 +++++++++++++-------- fs/proc/base.c | 5 +++-- include/linux/init_task.h | 1 + include/linux/namespace.h | 6 ++---- include/linux/nsproxy.h | 3 +++ include/linux/sched.h | 4 +--- kernel/exit.c | 7 +++---- kernel/fork.c | 6 +++--- 8 files changed, 29 insertions(+), 24 deletions(-)
8095ae8032ce2164e7177fb2d254629e7851947c
fs/namespace.c +13 -8
@@ -133,7 +133,7 @@ struct vfsmount *lookup_mnt(struct vfsmostatic inline int check_mnt (struct vfsmount *mnt) { -return mnt->mnt_namespace == current->namespace; +return mnt->mnt_namespace == current->nsproxy->namespace; }static void touch_namespace (struct namespace *ns)
@@ -832,7 +832,7 @@ static int attach_recursive_mnt(struct vif (parent_nd) {detach_mnt (source_mnt, parent_nd);attach_mnt (source_mnt, nd); -touch_namespace (current->namespace); +touch_namespace (current->nsproxy->namespace); }else {mnt_set_mountpoint (dest_mnt, dest_dentry, source_mnt);commit_tree (source_mnt);
@@ -1372,7 +1372,7 @@ dput_out:*/ struct namespace *dup_namespace (struct task_struct *tsk,struct fs_struct *fs) { -struct namespace *namespace = tsk->namespace; +struct namespace *namespace = tsk->nsproxy->namespace;struct namespace *new_ns;struct vfsmount *rootmnt =NULL , *pwdmnt =NULL , *altrootmnt =NULL ;struct vfsmount *p, *q;
@@ -1439,7 +1439,7 @@ struct namespace *dup_namespace(struct tint copy_namespace (int flags,struct task_struct *tsk) { -struct namespace *namespace = tsk->namespace; +struct namespace *namespace = tsk->nsproxy->namespace;struct namespace *new_ns;int err =0 ;
@@ -1462,7 +1462,7 @@ int copy_namespace(int flags, struct tas goto out; } - tsk->namespace = new_ns; + tsk->nsproxy->namespace = new_ns; out:put_namespace (namespace);
@@ -1685,7 +1685,7 @@ asmlinkage long sys_pivot_root(const chadetach_mnt (user_nd.mnt, &root_parent);attach_mnt (user_nd.mnt, &old_nd);/* mount old root on put_old */ attach_mnt (new_nd.mnt, &root_parent);/* mount new_root on / */ -touch_namespace (current->namespace); +touch_namespace (current->nsproxy->namespace);spin_unlock (&vfsmount_lock);chroot_fs_refs (&user_nd, &new_nd);security_sb_post_pivotroot (&user_nd, &new_nd);
@@ -1727,11 +1727,16 @@ static void __init init_mount_tree(void) namespace->root = mnt; mnt->mnt_namespace = namespace; - init_task.namespace = namespace; + init_task.nsproxy->namespace = namespace;read_lock (&tasklist_lock); do_each_thread(g, p) { +/* do we want namespace count to be #nsproxies, +* or # processes pointing to the namespace? */ get_namespace (namespace); - p->namespace = namespace; +#if 0 +/* should only be 1 nsproxy so far */ + p->nsproxy->namespace = namespace; +#endif }while_each_thread (g, p);read_unlock (&tasklist_lock);
fs/proc/base.c +3 -2
@@ -72,6 +72,7 @@ #include <linux/seccomp.h>#include <linux/cpuset.h> #include <linux/audit.h> #include <linux/poll.h> +#include <linux/nsproxy.h> #include "internal.h" /*
@@ -685,7 +686,7 @@ static int mounts_open(struct inode *inoint ret = -EINVAL ;task_lock (task); - namespace = task->namespace; + namespace = task->nsproxy->namespace;if (namespace)get_namespace (namespace);task_unlock (task);
@@ -752,7 +753,7 @@ static int mountstats_open(struct inodestruct seq_file *m = file->private_data;struct namespace *namespace;task_lock (task); - namespace = task->namespace; + namespace = task->nsproxy->namespace;if (namespace)get_namespace (namespace);task_unlock (task);
include/linux/init_task.h +1 -0
@@ -70,6 +70,7 @@ extern struct nsproxy init_nsproxy;#define INIT_NSPROXY (nsproxy) { \ .count = ATOMIC_INIT(1), \ .nslock = SPIN_LOCK_UNLOCKED, \ + .namespace = NULL, \ }#define INIT_SIGHAND (sighand) { \
include/linux/namespace.h +2 -4
@@ -4,6 +4,7 @@ #ifdef __KERNEL__#include <linux/mount.h> #include <linux/sched.h> +#include <linux/nsproxy.h> struct namespace {atomic_t count;
@@ -26,11 +27,8 @@ static inline void put_namespace(structstatic inline void exit_namespace (struct task_struct *p) { -struct namespace *namespace = p->namespace; +struct namespace *namespace = p->nsproxy->namespace;if (namespace) { -task_lock (p); - p->namespace =NULL ; -task_unlock (p);put_namespace (namespace); } }
include/linux/nsproxy.h +3 -0
@@ -4,9 +4,12 @@ #define _LINUX_NSPROXY_H#include <linux/spinlock.h> #include <linux/sched.h> +struct namespace ; +struct nsproxy {atomic_t count;spinlock_t nslock; +struct namespace *namespace; };extern struct nsproxy init_nsproxy;
include/linux/sched.h +1 -3
@@ -235,7 +235,6 @@ extern signed long schedule_timeout_inteextern signed long schedule_timeout_uninterruptible (signed long timeout); asmlinkage void schedule(void); -struct namespace ;struct nsproxy ;/* Maximum number of active map areas.. This is a random (large) number */
@@ -807,8 +806,7 @@ #endifstruct fs_struct *fs;/* open file information */ struct files_struct *files; -/* namespace */ -struct namespace *namespace; +/* namespaces */ struct nsproxy *nsproxy;/* signal handlers */ struct signal_struct *signal;
kernel/exit.c +3 -4
@@ -14,6 +14,7 @@ #include <linux/capability.h>#include <linux/completion.h> #include <linux/personality.h> #include <linux/tty.h> +#include <linux/nsproxy.h> #include <linux/namespace.h> #include <linux/key.h> #include <linux/security.h>
@@ -36,7 +37,6 @@ #include <linux/futex.h>#include <linux/compat.h> #include <linux/pipe_fs_i.h> #include <linux/audit.h> /* for audit_free() */ -#include <linux/nsproxy.h> #include <asm/uaccess.h> #include <asm/unistd.h>
@@ -416,11 +416,10 @@ void daemonize(const char *name, ...) current->fs = fs;atomic_inc (&fs->count);exit_namespace (current); - current->namespace = init_task.namespace; -get_namespace (current->namespace);exit_nsproxy (current); current->nsproxy = init_task.nsproxy;get_nsproxy (current->nsproxy); +get_namespace (current->nsproxy->namespace);exit_files (current); current->files = init_task.files;atomic_inc (¤t->files->count);
@@ -923,7 +922,7 @@ #endifexit_sem (tsk);__exit_files (tsk);__exit_fs (tsk); -exit_namespace (tsk); +exit_namespace (current);exit_nsproxy (current);exit_thread ();cpuset_exit (tsk);
kernel/fork.c +3 -3
@@ -1472,7 +1472,7 @@ static int unshare_fs(unsigned long unsh*/ static int unshare_namespace (unsigned long unshare_flags,struct namespace **new_nsp,struct fs_struct *new_fs) { -struct namespace *ns = current->namespace; +struct namespace *ns = current->nsproxy->namespace;if ((unshare_flags &CLONE_NEWNS ) && (ns &&atomic_read (&ns->count) >1 )) {
@@ -1609,8 +1609,8 @@ asmlinkage long sys_unshare(unsigned lon }if (new_ns) { - ns = current->namespace; - current->namespace = new_ns; + ns = current->nsproxy->namespace; + current->nsproxy->namespace = new_ns; new_ns = ns; }
Signature · 1 line
1.3.0
fs/namespace.c static void __init init_mount_tree(void) · line 1734
Quoted 10 lines · Serge E. Hallyn
fs/namespace.c +3
-1@@ -1727,11 +1727,16 @@ static void __init init_mount_tree(void) namespace->root = mnt; mnt->mnt_namespace = namespace; - init_task.namespace = namespace; + init_task.nsproxy->namespace = namespace;read_lock (&tasklist_lock); do_each_thread(g, p) { +/* do we want namespace count to be #nsproxies, +* or # processes pointing to the namespace? */
I am fairly certain we want the count to be #nsproxies.
fs/namespace.c static void __init init_mount_tree(void) · line 1741
Quoted 8 lines
fs/namespace.c +4
-1get_namespace (namespace); - p->namespace = namespace; +#if 0 +/* should only be 1 nsproxy so far */ + p->nsproxy->namespace = namespace; +#endif }while_each_thread (g, p);read_unlock (&tasklist_lock);
So I think this bit is wrong.
Quoted 16 lines · Eric W. Biederman
fs/namespace.c static void __init init_mount_tree(void) · line 1734
Quoted 10 lines · Serge E. Hallyn
fs/namespace.c +3
-1@@ -1727,11 +1727,16 @@ static void __init init_mount_tree(void) namespace->root = mnt; mnt->mnt_namespace = namespace; - init_task.namespace = namespace; + init_task.nsproxy->namespace = namespace;read_lock (&tasklist_lock); do_each_thread(g, p) { +/* do we want namespace count to be #nsproxies, +* or # processes pointing to the namespace? */ I am fairly certain we want the count to be #nsproxies.
Quoted 10 lines
fs/namespace.c static void __init init_mount_tree(void) · line 1741
Quoted 8 lines
fs/namespace.c +4
-1get_namespace (namespace); - p->namespace = namespace; +#if 0 +/* should only be 1 nsproxy so far */ + p->nsproxy->namespace = namespace; +#endif }while_each_thread (g, p);read_unlock (&tasklist_lock);So I think this bit is wrong.
Ok - in that case I need to audit the rest of namespace usage to make certain nothing depends on the count being #tasks.
BTW - a first set of comparison results showed nsproxy to have better dbench and tbench throughput, and worse kernbench performance. Which may make sense given that nsproxy results in lower memory usage but likely increased cache misses due to extra pointer dereference.
-serge
Quoted 28 lines · Serge E. Hallyn
Quoted 16 lines · Eric W. Biederman
fs/namespace.c static void __init init_mount_tree(void) · line 1734
Quoted 10 lines · Serge E. Hallyn
fs/namespace.c +3
-1@@ -1727,11 +1727,16 @@ static void __init init_mount_tree(void) namespace->root = mnt; mnt->mnt_namespace = namespace; - init_task.namespace = namespace; + init_task.nsproxy->namespace = namespace;read_lock (&tasklist_lock); do_each_thread(g, p) { +/* do we want namespace count to be #nsproxies, +* or # processes pointing to the namespace? */ I am fairly certain we want the count to be #nsproxies.
Quoted 10 lines
fs/namespace.c static void __init init_mount_tree(void) · line 1741
Quoted 8 lines
fs/namespace.c +4
-1get_namespace (namespace); - p->namespace = namespace; +#if 0 +/* should only be 1 nsproxy so far */ + p->nsproxy->namespace = namespace; +#endif }while_each_thread (g, p);read_unlock (&tasklist_lock);So I think this bit is wrong.
Quoted 2 lines
Ok - in that case I need to audit the rest of namespace usage to make certain nothing depends on the count being #tasks.
Ok. Thats makes sense.
Quoted 4 lines
BTW - a first set of comparison results showed nsproxy to have better dbench and tbench throughput, and worse kernbench performance. Which may make sense given that nsproxy results in lower memory usage but likely increased cache misses due to extra pointer dereference.
There are two additional things I can think of that are worth looking at:
- moving copy_uts_namespace, and copy_namespace inside of copy_nsproxy so we only run those we create a new nsproxy instance.
- Attempting to optimize cache line utilization by placing the structures in line in struct ns_proxy:
struct nsproxy {
atomic_t count;
struct namespace *namespace;
struct uts_namespace *uts_ns;
struct namespace namespace_data;
struct new_utsname uts_data;
};With the nsproxy count severing as a count for both the embedded data and for the nsproxy itself. I think it is a long shot but it could be interesting.
Given the frequency of use of the uts namespace and the filesystem namespace simply I think not accessing those namespaces on fork is likely to reduce the additional cache line miss rate enough so that it is lost in the noise.
Eric
Quoted 4 lines · Eric W. Biederman
There are two additional things I can think of that are worth looking at:
- movingcopy_uts_namespace, andcopy_namespaceinside ofcopy_nsproxyso we only run those we create a new nsproxy instance.
Was about to do that when I stopped because I was thinking I'd need to keep track of which namespace had been copied before a failaure, for the sake of clone.
But of course I don't have to - copy_nsproxy could do the cleanup itself on failure.
So this should be a nice little cleanup - especially as # namespaces increases.
Quoted 9 lines
- Attempting to optimize cache line utilization by placing the structures in line in struct
ns_proxy:struct nsproxy { atomic_t count; struct namespace *namespace; struct uts_namespace *uts_ns; struct namespace namespace_data; struct new_utsname uts_data; };
Quoted 8 lines
With the nsproxy count severing as a count for both the embedded data and for the nsproxy itself. I think it is a long shot but it could be interesting.
Given the frequency of use of the uts namespace and the filesystem namespace simply I think not accessing those namespaces on fork is likely to reduce the additional cache line miss rate enough so that it is lost in the noise.
Not getting this. Are you saying the uts_data would be a copy of the contents of *uts_ns, or that uts_ns points to nsproxy->uts_data? If the latter, then just unsharing uts_ns but not mounts namespace is no longer possible, right?
thanks,
-serge
Quoted 10 lines · Serge E. Hallyn
Quoted 4 lines · Eric W. Biederman
There are two additional things I can think of that are worth looking at:
- movingcopy_uts_namespace, andcopy_namespaceinside ofcopy_nsproxyso we only run those we create a new nsproxy instance.Was about to do that when I stopped because I was thinking I'd need to keep track of which namespace had been copied before a failaure, for the sake of clone.
Quoted 5 lines
But of course I don't have to -
copy_nsproxycould do the cleanup itself on failure.So this should be a nice little cleanup - especially as # namespaces increases.
Yes. At least if nsproxy doesn't show a performance degradation...
Quoted 18 lines
Quoted 9 lines
- Attempting to optimize cache line utilization by placing the structures in line in struct
ns_proxy:struct nsproxy { atomic_t count; struct namespace *namespace; struct uts_namespace *uts_ns; struct namespace namespace_data; struct new_utsname uts_data; };Quoted 8 lines
With the nsproxy count severing as a count for both the embedded data and for the nsproxy itself. I think it is a long shot but it could be interesting.
Given the frequency of use of the uts namespace and the filesystem namespace simply I think not accessing those namespaces on fork is likely to reduce the additional cache line miss rate enough so that it is lost in the noise.
Quoted 4 lines
Not getting this. Are you saying the
uts_datawould be a copy of the contents of*uts_ns, or thatuts_nspoints tonsproxy->uts_data? If the latter, then just unsharinguts_nsbut not mounts namespace is no longer possible, right?
The latter, uts_ns normally points to nsproxy->uts_data. But it still remains possible to unshare just the mounts namespace by simply coping the pointer when we clone nsproxy, and incrementing the previous ns_proxies count.
Like I said I think it is a long shot but if the data for namespaces really does remain small and they are usually all unshared in a group it could be a win.
Eric
Quoted 12 lines · Eric W. Biederman
Quoted 10 lines · Serge E. Hallyn
Quoted 4 lines · Eric W. Biederman
There are two additional things I can think of that are worth looking at:
- movingcopy_uts_namespace, andcopy_namespaceinside ofcopy_nsproxyso we only run those we create a new nsproxy instance.Was about to do that when I stopped because I was thinking I'd need to keep track of which namespace had been copied before a failaure, for the sake of clone.
Quoted 7 lines
Quoted 5 lines
But of course I don't have to -
copy_nsproxycould do the cleanup itself on failure.So this should be a nice little cleanup - especially as # namespaces increases.
Yes. At least if nsproxy doesn't show a performance degradation...
Ok, at least here is a patch to do a bit of this. It moves things around enough that didn't want to combine this with setting namespace->refcount to #nsproxies, or your inlining trick.
-serge
Subject: [PATCH] nsproxy: code cleanup
consolidate nsproxy and namespace copy/get/exit.
---
include/linux/namespace.h | 8 ---- include/linux/nsproxy.h | 26 ++++++------- kernel/exit.c | 13 ++----- kernel/fork.c | 39 +++++++++----------- kernel/nsproxy.c | 87 ++++++++++++++++++++++++++++++++++++++++++--- 5 files changed, 113 insertions(+), 60 deletions(-)
5492e3bdc0245ab27e70fe8dc6c004cb437dc8f6
include/linux/namespace.h +0 -8
@@ -25,14 +25,6 @@ static inline void put_namespace(struct__put_namespace (namespace); } -static inline void exit_namespace (struct task_struct *p) -{ -struct namespace *namespace = p->nsproxy->namespace; -if (namespace) { -put_namespace (namespace); - } -} -static inline void get_namespace (struct namespace *namespace) {atomic_inc (&namespace->count);
include/linux/nsproxy.h +11 -15
@@ -14,28 +14,24 @@ struct nsproxy { };extern struct nsproxy init_nsproxy; -static inline void get_nsproxy (struct nsproxy *nsp) +struct nsproxy *dup_namespaces (struct nsproxy *orig); +int copy_namespaces (int flags,struct task_struct *tsk); +void get_namespaces (struct task_struct *tsk); +void exit_namespaces (struct nsproxy *ns); + +static inline void exit_task_namespaces (struct task_struct *p) { -atomic_inc (&nsp->count); +struct nsproxy *ns = p->nsproxy; +if (ns) { +exit_namespaces (p->nsproxy); + p->nsproxy =NULL ; + } } -struct nsproxy *clone_nsproxy (struct nsproxy *orig); -int copy_nsproxy (int flags,struct task_struct *tsk); -void free_nsproxy (struct nsproxy *nsp); -static inline void put_nsproxy (struct nsproxy *nsp) {if (atomic_dec_and_test (&nsp->count)) {kfree (nsp); } } - -static inline void exit_nsproxy (struct task_struct *p) -{ -struct nsproxy *nsp = p->nsproxy; -if (nsp) { - p->nsproxy =NULL ; -put_nsproxy (nsp); - } -}#endif
kernel/exit.c +3 -10
@@ -175,7 +175,6 @@ repeat:write_unlock_irq (&tasklist_lock);spin_unlock (&p->proc_lock);proc_pid_flush (proc_dentry); -exit_nsproxy (p);release_thread (p);call_rcu (&p->rcu, delayed_put_task_struct);
@@ -416,13 +415,9 @@ void daemonize(const char *name, ...) current->fs = fs;atomic_inc (&fs->count); -exit_utsname (current); -exit_namespace (current); -exit_nsproxy (current); +exit_task_namespaces (current); current->nsproxy = init_task.nsproxy; -get_nsproxy (current->nsproxy); -get_namespace (current->nsproxy->namespace); -get_uts_ns (current->nsproxy->uts_ns); +get_namespaces (current);exit_files (current); current->files = init_task.files;
@@ -926,9 +921,7 @@ fastcall NORET_TYPE void do_exit(long coexit_sem (tsk);__exit_files (tsk);__exit_fs (tsk); -exit_utsname (current); -exit_namespace (current); -exit_nsproxy (current); +exit_task_namespaces (current);exit_thread ();cpuset_exit (tsk);exit_keys (tsk);
kernel/fork.c +18 -21
@@ -45,7 +45,6 @@#include <linux/acct.h> #include <linux/cn_proc.h> #include <linux/nsproxy.h> -#include <linux/utsname.h> #include <asm/pgtable.h> #include <asm/pgalloc.h>
@@ -1062,15 +1061,11 @@ static task_t *copy_process(unsigned lon goto bad_fork_cleanup_signal;if ((retval =copy_keys (clone_flags, p))) goto bad_fork_cleanup_mm; -if ((retval =copy_nsproxy (clone_flags, p))) +if ((retval =copy_namespaces (clone_flags, p))) goto bad_fork_cleanup_keys; -if ((retval =copy_utsname (clone_flags, p))) - goto bad_fork_cleanup_nsproxy; -if ((retval =copy_namespace (clone_flags, p))) - goto bad_fork_cleanup_utsname; retval =copy_thread (0 , clone_flags, stack_start, stack_size, p, regs);if (retval) - goto bad_fork_cleanup_namespace; + goto bad_fork_cleanup_namespaces; p->set_child_tid = (clone_flags &CLONE_CHILD_SETTID ) ? child_tidptr :NULL ;/*
@@ -1157,7 +1152,7 @@ static task_t *copy_process(unsigned lonspin_unlock (¤t->sighand->siglock);write_unlock_irq (&tasklist_lock); retval = -ERESTARTNOINTR ; - goto bad_fork_cleanup_namespace; + goto bad_fork_cleanup_namespaces; }if (clone_flags &CLONE_THREAD ) {
@@ -1170,7 +1165,7 @@ static task_t *copy_process(unsigned lonspin_unlock (¤t->sighand->siglock);write_unlock_irq (&tasklist_lock); retval = -EAGAIN ; - goto bad_fork_cleanup_namespace; + goto bad_fork_cleanup_namespaces; } p->group_leader = current->group_leader;
@@ -1222,12 +1217,8 @@ static task_t *copy_process(unsigned lonproc_fork_connector (p);return p; -bad_fork_cleanup_namespace: -exit_namespace (p); -bad_fork_cleanup_utsname: -exit_utsname (p); -bad_fork_cleanup_nsproxy: -exit_nsproxy (p); +bad_fork_cleanup_namespaces: +exit_task_namespaces (p); bad_fork_cleanup_keys:exit_keys (p); bad_fork_cleanup_mm:
@@ -1570,7 +1561,7 @@ asmlinkage long sys_unshare(unsigned lonstruct files_struct *fd, *new_fd =NULL ;struct sem_undo_list *new_ulist =NULL ;struct nsproxy *new_nsproxy, *old_nsproxy; -struct uts_namespace *new_uts =NULL ; +struct uts_namespace *uts, *new_uts =NULL ;check_unshare_flags (&unshare_flags);
@@ -1595,21 +1586,20 @@ asmlinkage long sys_unshare(unsigned lonif ((err =unshare_semundo (unshare_flags, &new_ulist))) goto bad_unshare_cleanup_fd;if ((err =unshare_utsname (unshare_flags, &new_uts))) - goto bad_unshare_cleanup_fd; + goto bad_unshare_cleanup_semundo;if (new_fs || new_ns || new_sigh || new_mm || new_fd || new_ulist || new_uts) { old_nsproxy = current->nsproxy; - new_nsproxy =clone_nsproxy (old_nsproxy); + new_nsproxy =dup_namespaces (old_nsproxy);if (!new_nsproxy) { err = -ENOMEM ; - goto bad_unshare_cleanup_semundo; + goto bad_unshare_cleanup_uts; }task_lock (current); current->nsproxy = new_nsproxy; -put_nsproxy (old_nsproxy);if (new_fs) { fs = current->fs;
@@ -1645,13 +1635,20 @@ asmlinkage long sys_unshare(unsigned lon }if (new_uts) { -put_uts_ns (current->nsproxy->uts_ns); + uts = current->nsproxy->uts_ns; current->nsproxy->uts_ns = new_uts; + new_uts = uts; }task_unlock (current); + +put_nsproxy (old_nsproxy); } +bad_unshare_cleanup_uts: +if (new_uts) +put_uts_ns (new_uts); + bad_unshare_cleanup_semundo: bad_unshare_cleanup_fd:if (new_fd)
kernel/nsproxy.c +81 -6
115 lines not on this page · raw message
Signature · 1 line
1.1.6
kernel/exit.c void daemonize(const char *name, ...) · line 420
Quoted 10 lines · Serge E. Hallyn
kernel/exit.c +2
-6-exit_utsname (current); -exit_namespace (current); -exit_nsproxy (current); +exit_task_namespaces (current); current->nsproxy = init_task.nsproxy; -get_nsproxy (current->nsproxy); -get_namespace (current->nsproxy->namespace); -get_uts_ns (current->nsproxy->uts_ns); +get_namespaces (current);
That really cleans up the main path quite a bit. Very nice.
For parity with exit_task_namespaces(), should that be called get_task_namespaces()?
-- Dave
Quoted 12 lines · Dave Hansen
kernel/exit.c void daemonize(const char *name, ...) · line 420
Quoted 10 lines · Serge E. Hallyn
kernel/exit.c +2
-6-exit_utsname (current); -exit_namespace (current); -exit_nsproxy (current); +exit_task_namespaces (current); current->nsproxy = init_task.nsproxy; -get_nsproxy (current->nsproxy); -get_namespace (current->nsproxy->namespace); -get_uts_ns (current->nsproxy->uts_ns); +get_namespaces (current);
Quoted 4 lines
That really cleans up the main path quite a bit. Very nice.
For parity with
exit_task_namespaces(), should that be calledget_task_namespaces()?
Yup, that would make sense. Will update the patch in my git tree.
Though those functions be renamed when the namespaces->count becomes #nsproxies. Getting a weird regression with that patch, though, so won't be sending that out today.
thanks,
-serge
Quoted 12 lines · Dave Hansen
Review
Quoted 10 lines · Serge E. Hallyn
(quoted diff) +2
-6- exit_utsname(current); - exit_namespace(current); - exit_nsproxy(current); + exit_task_namespaces(current); current->nsproxy = init_task.nsproxy; - get_nsproxy(current->nsproxy); - get_namespace(current->nsproxy->namespace); - get_uts_ns(current->nsproxy->uts_ns); + get_namespaces(current);
Quoted 4 lines
That really cleans up the main path quite a bit. Very nice.
For parity with
exit_task_namespaces(), should that be calledget_task_namespaces()?
Here's a new patch:
Subject: [PATCH 10/11] nsproxy: code cleanup
consolidate nsproxy and namespace copy/get/exit.
Changelog:
renamed get_namespaces(tsk) to get_task_namespaces(tsk).
---
include/linux/namespace.h | 8 ---- include/linux/nsproxy.h | 26 +++++-------- kernel/exit.c | 13 +----- kernel/fork.c | 39 +++++++++---------- kernel/nsproxy.c | 92 ++++++++++++++++++++++++++++++++++++++++++--- 5 files changed, 118 insertions(+), 60 deletions(-)
3f3ce98f9c6fc8c3ce31cb170bf281c8eb0d4c6a
include/linux/namespace.h +0 -8
@@ -25,14 +25,6 @@ static inline void put_namespace(struct__put_namespace (namespace); } -static inline void exit_namespace (struct task_struct *p) -{ -struct namespace *namespace = p->nsproxy->namespace; -if (namespace) { -put_namespace (namespace); - } -} -static inline void get_namespace (struct namespace *namespace) {atomic_inc (&namespace->count);
include/linux/nsproxy.h +11 -15
@@ -14,28 +14,24 @@ struct nsproxy { };extern struct nsproxy init_nsproxy; -static inline void get_nsproxy (struct nsproxy *nsp) +struct nsproxy *dup_namespaces (struct nsproxy *orig); +int copy_namespaces (int flags,struct task_struct *tsk); +void get_task_namespaces (struct task_struct *tsk); +void exit_namespaces (struct nsproxy *ns); + +static inline void exit_task_namespaces (struct task_struct *p) { -atomic_inc (&nsp->count); +struct nsproxy *ns = p->nsproxy; +if (ns) { +exit_namespaces (p->nsproxy); + p->nsproxy =NULL ; + } } -struct nsproxy *clone_nsproxy (struct nsproxy *orig); -int copy_nsproxy (int flags,struct task_struct *tsk); -void free_nsproxy (struct nsproxy *nsp); -static inline void put_nsproxy (struct nsproxy *nsp) {if (atomic_dec_and_test (&nsp->count)) {kfree (nsp); } } - -static inline void exit_nsproxy (struct task_struct *p) -{ -struct nsproxy *nsp = p->nsproxy; -if (nsp) { - p->nsproxy =NULL ; -put_nsproxy (nsp); - } -}#endif
kernel/exit.c +3 -10
@@ -175,7 +175,6 @@ repeat:write_unlock_irq (&tasklist_lock);spin_unlock (&p->proc_lock);proc_pid_flush (proc_dentry); -exit_nsproxy (p);release_thread (p);call_rcu (&p->rcu, delayed_put_task_struct);
@@ -416,13 +415,9 @@ void daemonize(const char *name, ...) current->fs = fs;atomic_inc (&fs->count); -exit_utsname (current); -exit_namespace (current); -exit_nsproxy (current); +exit_task_namespaces (current); current->nsproxy = init_task.nsproxy; -get_nsproxy (current->nsproxy); -get_namespace (current->nsproxy->namespace); -get_uts_ns (current->nsproxy->uts_ns); +get_task_namespaces (current);exit_files (current); current->files = init_task.files;
@@ -926,9 +921,7 @@ fastcall NORET_TYPE void do_exit(long coexit_sem (tsk);__exit_files (tsk);__exit_fs (tsk); -exit_utsname (current); -exit_namespace (current); -exit_nsproxy (current); +exit_task_namespaces (current);exit_thread ();cpuset_exit (tsk);exit_keys (tsk);
kernel/fork.c +18 -21
@@ -45,7 +45,6 @@#include <linux/acct.h> #include <linux/cn_proc.h> #include <linux/nsproxy.h> -#include <linux/utsname.h> #include <asm/pgtable.h> #include <asm/pgalloc.h>
@@ -1062,15 +1061,11 @@ static task_t *copy_process(unsigned lon goto bad_fork_cleanup_signal;if ((retval =copy_keys (clone_flags, p))) goto bad_fork_cleanup_mm; -if ((retval =copy_nsproxy (clone_flags, p))) +if ((retval =copy_namespaces (clone_flags, p))) goto bad_fork_cleanup_keys; -if ((retval =copy_utsname (clone_flags, p))) - goto bad_fork_cleanup_nsproxy; -if ((retval =copy_namespace (clone_flags, p))) - goto bad_fork_cleanup_utsname; retval =copy_thread (0 , clone_flags, stack_start, stack_size, p, regs);if (retval) - goto bad_fork_cleanup_namespace; + goto bad_fork_cleanup_namespaces; p->set_child_tid = (clone_flags &CLONE_CHILD_SETTID ) ? child_tidptr :NULL ;/*
@@ -1157,7 +1152,7 @@ static task_t *copy_process(unsigned lonspin_unlock (¤t->sighand->siglock);write_unlock_irq (&tasklist_lock); retval = -ERESTARTNOINTR ; - goto bad_fork_cleanup_namespace; + goto bad_fork_cleanup_namespaces; }if (clone_flags &CLONE_THREAD ) {
@@ -1170,7 +1165,7 @@ static task_t *copy_process(unsigned lonspin_unlock (¤t->sighand->siglock);write_unlock_irq (&tasklist_lock); retval = -EAGAIN ; - goto bad_fork_cleanup_namespace; + goto bad_fork_cleanup_namespaces; } p->group_leader = current->group_leader;
@@ -1222,12 +1217,8 @@ static task_t *copy_process(unsigned lonproc_fork_connector (p);return p; -bad_fork_cleanup_namespace: -exit_namespace (p); -bad_fork_cleanup_utsname: -exit_utsname (p); -bad_fork_cleanup_nsproxy: -exit_nsproxy (p); +bad_fork_cleanup_namespaces: +exit_task_namespaces (p); bad_fork_cleanup_keys:exit_keys (p); bad_fork_cleanup_mm:
@@ -1570,7 +1561,7 @@ asmlinkage long sys_unshare(unsigned lonstruct files_struct *fd, *new_fd =NULL ;struct sem_undo_list *new_ulist =NULL ;struct nsproxy *new_nsproxy, *old_nsproxy; -struct uts_namespace *new_uts =NULL ; +struct uts_namespace *uts, *new_uts =NULL ;check_unshare_flags (&unshare_flags);
@@ -1595,21 +1586,20 @@ asmlinkage long sys_unshare(unsigned lonif ((err =unshare_semundo (unshare_flags, &new_ulist))) goto bad_unshare_cleanup_fd;if ((err =unshare_utsname (unshare_flags, &new_uts))) - goto bad_unshare_cleanup_fd; + goto bad_unshare_cleanup_semundo;if (new_fs || new_ns || new_sigh || new_mm || new_fd || new_ulist || new_uts) { old_nsproxy = current->nsproxy; - new_nsproxy =clone_nsproxy (old_nsproxy); + new_nsproxy =dup_namespaces (old_nsproxy);if (!new_nsproxy) { err = -ENOMEM ; - goto bad_unshare_cleanup_semundo; + goto bad_unshare_cleanup_uts; }task_lock (current); current->nsproxy = new_nsproxy; -put_nsproxy (old_nsproxy);if (new_fs) { fs = current->fs;
@@ -1645,13 +1635,20 @@ asmlinkage long sys_unshare(unsigned lon }if (new_uts) { -put_uts_ns (current->nsproxy->uts_ns); + uts = current->nsproxy->uts_ns; current->nsproxy->uts_ns = new_uts; + new_uts = uts; }task_unlock (current); + +put_nsproxy (old_nsproxy); } +bad_unshare_cleanup_uts: +if (new_uts) +put_uts_ns (new_uts); + bad_unshare_cleanup_semundo: bad_unshare_cleanup_fd:if (new_fd)
kernel/nsproxy.c +86 -6
120 lines not on this page · raw message
Signature · 1 line
1.1.6
Quoted 16 lines · Eric W. Biederman
Review static void __init init_mount_tree(void) · line 1734
Quoted 10 lines · Serge E. Hallyn
(quoted diff) +3
-1@@ -1727,11 +1727,16 @@ static void __init init_mount_tree(void) namespace->root = mnt; mnt->mnt_namespace = namespace; - init_task.namespace = namespace; + init_task.nsproxy->namespace = namespace; read_lock(&tasklist_lock); do_each_thread(g, p) { + /* do we want namespace count to be #nsproxies, + * or # processes pointing to the namespace? */I am fairly certain we want the count to be #nsproxies.
Quoted 10 lines
Review static void __init init_mount_tree(void) · line 1741
Quoted 8 lines
(quoted diff) +4
-1get_namespace(namespace); - p->namespace = namespace; +#if 0 + /* should only be 1 nsproxy so far */ + p->nsproxy->namespace = namespace; +#endif } while_each_thread(g, p); read_unlock(&tasklist_lock);So I think this bit is wrong.
Here is a patch (on top of the patchset + the patch I sent in response to Dave) to change the fs namespace and utsname ->counts to being the number of nsproxies holding a reference.
thanks,
-serge
Subject: [PATCH 11/11] nsproxy: change meaning of namespace refcount
switch namespace+utsname refcount to count nsproxies
---
fs/namespace.c | 13 +------------ include/linux/nsproxy.h | 18 +++++++++--------- kernel/fork.c | 3 +-- kernel/nsproxy.c | 39 ++++++++++++++------------------------- 4 files changed, 25 insertions(+), 48 deletions(-)
41b3b9a9df03156627adc34b88c041dd3ade1236
fs/namespace.c +1 -12
@@ -1711,7 +1711,6 @@ static void __init init_mount_tree(void) {struct vfsmount *mnt;struct namespace *namespace; -struct task_struct *g, *p; mnt =do_kern_mount ("rootfs" ,0 ,"rootfs" ,NULL );if (IS_ERR (mnt))
@@ -1728,17 +1727,7 @@ static void __init init_mount_tree(void) mnt->mnt_namespace = namespace; init_task.nsproxy->namespace = namespace; -read_lock (&tasklist_lock); - do_each_thread(g, p) { -/* do we want namespace count to be #nsproxies, -* or # processes pointing to the namespace? */ -get_namespace (namespace); -#if 0 -/* should only be 1 nsproxy so far */ - p->nsproxy->namespace = namespace; -#endif - }while_each_thread (g, p); -read_unlock (&tasklist_lock); +get_namespace (namespace);set_fs_pwd (current->fs, namespace->root, namespace->root->mnt_root);set_fs_root (current->fs, namespace->root, namespace->root->mnt_root);
include/linux/nsproxy.h +9 -9
@@ -17,21 +17,21 @@ extern struct nsproxy init_nsproxy;struct nsproxy *dup_namespaces (struct nsproxy *orig);int copy_namespaces (int flags,struct task_struct *tsk);void get_task_namespaces (struct task_struct *tsk); -void exit_namespaces (struct nsproxy *ns); +void free_nsproxy (struct nsproxy *ns); -static inline void exit_task_namespaces (struct task_struct *p) +static inline void put_nsproxy (struct nsproxy *ns) { -struct nsproxy *ns = p->nsproxy; -if (ns) { -exit_namespaces (p->nsproxy); - p->nsproxy =NULL ; +if (atomic_dec_and_test (&ns->count)) { +free_nsproxy (ns); } } -static inline void put_nsproxy (struct nsproxy *nsp) +static inline void exit_task_namespaces (struct task_struct *p) { -if (atomic_dec_and_test (&nsp->count)) { -kfree (nsp); +struct nsproxy *ns = p->nsproxy; +if (ns) { +put_nsproxy (ns); + p->nsproxy =NULL ; } }#endif
kernel/fork.c +1 -2
@@ -1470,8 +1470,7 @@ static int unshare_namespace(unsigned lo {struct namespace *ns = current->nsproxy->namespace; -if ((unshare_flags &CLONE_NEWNS ) && - (ns &&atomic_read (&ns->count) >1 )) { +if ((unshare_flags &CLONE_NEWNS ) && ns) {if (!capable (CAP_SYS_ADMIN ))return -EPERM ;
kernel/nsproxy.c +14 -25
@@ -21,26 +21,18 @@ static inline void get_nsproxy(struct nsatomic_inc (&ns->count); } -static inline void get_namespaces (struct nsproxy *ns) -{ -get_nsproxy (ns); -if (ns->namespace) -get_namespace (ns->namespace); -if (ns->uts_ns) -get_uts_ns (ns->uts_ns); -} -void get_task_namespaces (struct task_struct *tsk) {struct nsproxy *ns = tsk->nsproxy;if (ns) { -get_namespaces (ns); +get_nsproxy (ns); } }/* * creates a copy of "orig" with refcount 1. -* This does not grab references to the contained namespaces. +* This does not grab references to the contained namespaces, +* so that needs to be done by dup_namespaces. */ static inline struct nsproxy *clone_namespaces (struct nsproxy *orig) {
@@ -74,18 +66,6 @@ struct nsproxy *dup_namespaces(struct ns }/* -* Put refcount on nsproxy and each namespace therein -*/ -void exit_namespaces (struct nsproxy *ns) -{ -if (ns->namespace) -put_namespace (ns->namespace); -if (ns->uts_ns) -put_uts_ns (ns->uts_ns); -put_nsproxy (ns); -} - -/* * called from clone. This now handles copy for nsproxy and all * namespaces therein. */
@@ -98,7 +78,7 @@ int copy_namespaces(int flags, struct taif (!old_ns)return 0 ; -get_namespaces (old_ns); +get_nsproxy (old_ns);if (!(flags & (CLONE_NEWNS |CLONE_NEWUTS )))return 0 ;
@@ -128,6 +108,15 @@ int copy_namespaces(int flags, struct ta } out: -exit_namespaces (old_ns); +put_nsproxy (old_ns);return err; } + +void free_nsproxy (struct nsproxy *ns) +{ +if (ns->namespace) +put_namespace (ns->namespace); +if (ns->uts_ns) +put_uts_ns (ns->uts_ns); +kfree (ns); +}
Signature · 1 line
1.1.6