Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
Merge branch 'work.namespace' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
Pull ipc namespace update from Al Viro:
"Rik's patches reducing the amount of synchronize_rcu() triggered by
ipc namespace destruction.
I've some pending stuff reducing that on the normal umount side, but
it's nowhere near ready and Rik's stuff shouldn't be held back due to
conflicts - I'll just redo the parts of my series that stray into
ipc/*"
* 'work.namespace' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
ipc,namespace: batch free ipc_namespace structures
ipc,namespace: make ipc namespace allocation wait for pending free
+41
-20
+14
-4
fs/namespace.c
+14
-4
fs/namespace.c
···
1283
1283
}
1284
1284
EXPORT_SYMBOL(mntget);
1285
1285
1286
+
/*
1287
+
* Make a mount point inaccessible to new lookups.
1288
+
* Because there may still be current users, the caller MUST WAIT
1289
+
* for an RCU grace period before destroying the mount point.
1290
+
*/
1291
+
void mnt_make_shortterm(struct vfsmount *mnt)
1292
+
{
1293
+
if (mnt)
1294
+
real_mount(mnt)->mnt_ns = NULL;
1295
+
}
1296
+
1286
1297
/**
1287
1298
* path_is_mountpoint() - Check if path is a mount in the current namespace.
1288
1299
* @path: path to check
···
4470
4459
void kern_unmount(struct vfsmount *mnt)
4471
4460
{
4472
4461
/* release long term mount so mount point can be released */
4473
-
if (!IS_ERR_OR_NULL(mnt)) {
4474
-
real_mount(mnt)->mnt_ns = NULL;
4462
+
if (!IS_ERR(mnt)) {
4463
+
mnt_make_shortterm(mnt);
4475
4464
synchronize_rcu(); /* yecchhh... */
4476
4465
mntput(mnt);
4477
4466
}
···
4483
4472
unsigned int i;
4484
4473
4485
4474
for (i = 0; i < num; i++)
4486
-
if (mnt[i])
4487
-
real_mount(mnt[i])->mnt_ns = NULL;
4475
+
mnt_make_shortterm(mnt[i]);
4488
4476
synchronize_rcu_expedited();
4489
4477
for (i = 0; i < num; i++)
4490
4478
mntput(mnt[i]);
+1
include/linux/mount.h
+1
include/linux/mount.h
···
86
86
extern void mnt_drop_write_file(struct file *file);
87
87
extern void mntput(struct vfsmount *mnt);
88
88
extern struct vfsmount *mntget(struct vfsmount *mnt);
89
+
extern void mnt_make_shortterm(struct vfsmount *mnt);
89
90
extern struct vfsmount *mnt_clone_internal(const struct path *path);
90
91
extern bool __mnt_is_readonly(struct vfsmount *mnt);
91
92
extern bool mnt_may_suid(struct vfsmount *mnt);
-5
ipc/mqueue.c
-5
ipc/mqueue.c
+26
-9
ipc/namespace.c
+26
-9
ipc/namespace.c
···
19
19
20
20
#include "util.h"
21
21
22
+
/*
23
+
* The work queue is used to avoid the cost of synchronize_rcu in kern_unmount.
24
+
*/
25
+
static void free_ipc(struct work_struct *unused);
26
+
static DECLARE_WORK(free_ipc_work, free_ipc);
27
+
22
28
static struct ucounts *inc_ipc_namespaces(struct user_namespace *ns)
23
29
{
24
30
return inc_ucount(ns, current_euid(), UCOUNT_IPC_NAMESPACES);
···
43
37
int err;
44
38
45
39
err = -ENOSPC;
40
+
again:
46
41
ucounts = inc_ipc_namespaces(user_ns);
47
-
if (!ucounts)
42
+
if (!ucounts) {
43
+
/*
44
+
* IPC namespaces are freed asynchronously, by free_ipc_work.
45
+
* If frees were pending, flush_work will wait, and
46
+
* return true. Fail the allocation if no frees are pending.
47
+
*/
48
+
if (flush_work(&free_ipc_work))
49
+
goto again;
48
50
goto fail;
51
+
}
49
52
50
53
err = -ENOMEM;
51
54
ns = kzalloc(sizeof(struct ipc_namespace), GFP_KERNEL_ACCOUNT);
···
145
130
146
131
static void free_ipc_ns(struct ipc_namespace *ns)
147
132
{
148
-
/* mq_put_mnt() waits for a grace period as kern_unmount()
149
-
* uses synchronize_rcu().
133
+
/*
134
+
* Caller needs to wait for an RCU grace period to have passed
135
+
* after making the mount point inaccessible to new accesses.
150
136
*/
151
-
mq_put_mnt(ns);
137
+
mntput(ns->mq_mnt);
152
138
sem_exit_ns(ns);
153
139
msg_exit_ns(ns);
154
140
shm_exit_ns(ns);
···
170
154
struct ipc_namespace *n, *t;
171
155
172
156
llist_for_each_entry_safe(n, t, node, mnt_llist)
157
+
mnt_make_shortterm(n->mq_mnt);
158
+
159
+
/* Wait for any last users to have gone away. */
160
+
synchronize_rcu();
161
+
162
+
llist_for_each_entry_safe(n, t, node, mnt_llist)
173
163
free_ipc_ns(n);
174
164
}
175
-
176
-
/*
177
-
* The work queue is used to avoid the cost of synchronize_rcu in kern_unmount.
178
-
*/
179
-
static DECLARE_WORK(free_ipc_work, free_ipc);
180
165
181
166
/*
182
167
* put_ipc_ns - drop a reference to an ipc namespace.
-2
ipc/util.h
-2
ipc/util.h
···
56
56
57
57
#ifdef CONFIG_POSIX_MQUEUE
58
58
extern void mq_clear_sbinfo(struct ipc_namespace *ns);
59
-
extern void mq_put_mnt(struct ipc_namespace *ns);
60
59
#else
61
60
static inline void mq_clear_sbinfo(struct ipc_namespace *ns) { }
62
-
static inline void mq_put_mnt(struct ipc_namespace *ns) { }
63
61
#endif
64
62
65
63
#ifdef CONFIG_SYSVIPC