Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
pidfs: adapt to rhashtable-based simple_xattrs
Adapt pidfs to use the rhashtable-based xattr path by switching from a
dedicated slab cache to simple_xattrs_alloc().
Previously pidfs used a custom kmem_cache (pidfs_xattr_cachep) that
allocated a struct containing an embedded simple_xattrs plus
simple_xattrs_init(). Replace this with simple_xattrs_alloc() which
combines kzalloc + rhashtable_init, and drop the dedicated slab cache
entirely.
Use simple_xattr_free_rcu() for replaced xattr entries to allow
concurrent RCU readers to finish.
Link: https://patch.msgid.link/20260216-work-xattr-socket-v1-5-c2efa4f74cb7@kernel.org
Acked-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Christian Brauner <brauner@kernel.org>
+40
-24
+40
-24
fs/pidfs.c
+40
-24
fs/pidfs.c
···
22
22
#include <net/net_namespace.h>
23
23
#include <linux/coredump.h>
24
24
#include <linux/rhashtable.h>
25
+
#include <linux/llist.h>
25
26
#include <linux/xattr.h>
26
27
#include <linux/cookie.h>
27
28
···
32
31
#define PIDFS_PID_DEAD ERR_PTR(-ESRCH)
33
32
34
33
static struct kmem_cache *pidfs_attr_cachep __ro_after_init;
35
-
static struct kmem_cache *pidfs_xattr_cachep __ro_after_init;
36
34
37
35
static struct path pidfs_root_path = {};
38
36
···
46
46
PIDFS_ATTR_BIT_COREDUMP = 1,
47
47
};
48
48
49
-
struct pidfs_attr {
49
+
struct pidfs_anon_attr {
50
50
unsigned long attr_mask;
51
-
struct simple_xattrs *xattrs;
52
51
struct /* exit info */ {
53
52
__u64 cgroupid;
54
53
__s32 exit_code;
···
92
93
* inode number and the inode generation number to compare or
93
94
* use file handles.
94
95
*/
96
+
struct pidfs_attr {
97
+
struct simple_xattrs *xattrs;
98
+
union {
99
+
struct pidfs_anon_attr;
100
+
struct llist_node pidfs_llist;
101
+
};
102
+
};
95
103
96
104
#if BITS_PER_LONG == 32
97
105
···
184
178
pidfs_ino_ht_params);
185
179
}
186
180
181
+
static LLIST_HEAD(pidfs_free_list);
182
+
183
+
static void pidfs_free_attr_work(struct work_struct *work)
184
+
{
185
+
struct pidfs_attr *attr, *next;
186
+
struct llist_node *head;
187
+
188
+
head = llist_del_all(&pidfs_free_list);
189
+
llist_for_each_entry_safe(attr, next, head, pidfs_llist) {
190
+
struct simple_xattrs *xattrs = attr->xattrs;
191
+
192
+
if (xattrs) {
193
+
simple_xattrs_free(xattrs, NULL);
194
+
kfree(xattrs);
195
+
}
196
+
kfree(attr);
197
+
}
198
+
}
199
+
200
+
static DECLARE_WORK(pidfs_free_work, pidfs_free_attr_work);
201
+
187
202
void pidfs_free_pid(struct pid *pid)
188
203
{
189
-
struct pidfs_attr *attr __free(kfree) = no_free_ptr(pid->attr);
190
-
struct simple_xattrs *xattrs __free(kfree) = NULL;
204
+
struct pidfs_attr *attr = pid->attr;
191
205
192
206
/*
193
207
* Any dentry must've been wiped from the pid by now.
···
226
200
if (IS_ERR(attr))
227
201
return;
228
202
229
-
xattrs = no_free_ptr(attr->xattrs);
230
-
if (xattrs)
231
-
simple_xattrs_free(xattrs, NULL);
203
+
if (likely(!attr->xattrs))
204
+
kfree(attr);
205
+
else if (llist_add(&attr->pidfs_llist, &pidfs_free_list))
206
+
schedule_work(&pidfs_free_work);
232
207
}
233
208
234
209
#ifdef CONFIG_PROC_FS
···
1038
1011
1039
1012
xattrs = READ_ONCE(attr->xattrs);
1040
1013
if (!xattrs)
1041
-
return 0;
1014
+
return -ENODATA;
1042
1015
1043
1016
name = xattr_full_name(handler, suffix);
1044
1017
return simple_xattr_get(xattrs, name, value, size);
···
1058
1031
/* Ensure we're the only one to set @attr->xattrs. */
1059
1032
WARN_ON_ONCE(!inode_is_locked(inode));
1060
1033
1061
-
xattrs = READ_ONCE(attr->xattrs);
1062
-
if (!xattrs) {
1063
-
xattrs = kmem_cache_zalloc(pidfs_xattr_cachep, GFP_KERNEL);
1064
-
if (!xattrs)
1065
-
return -ENOMEM;
1066
-
1067
-
simple_xattrs_init(xattrs);
1068
-
smp_store_release(&pid->attr->xattrs, xattrs);
1069
-
}
1034
+
xattrs = simple_xattrs_lazy_alloc(&attr->xattrs, value, flags);
1035
+
if (IS_ERR_OR_NULL(xattrs))
1036
+
return PTR_ERR(xattrs);
1070
1037
1071
1038
name = xattr_full_name(handler, suffix);
1072
1039
old_xattr = simple_xattr_set(xattrs, name, value, size, flags);
1073
1040
if (IS_ERR(old_xattr))
1074
1041
return PTR_ERR(old_xattr);
1075
1042
1076
-
simple_xattr_free(old_xattr);
1043
+
simple_xattr_free_rcu(old_xattr);
1077
1044
return 0;
1078
1045
}
1079
1046
···
1144
1123
pidfs_attr_cachep = kmem_cache_create("pidfs_attr_cache", sizeof(struct pidfs_attr), 0,
1145
1124
(SLAB_HWCACHE_ALIGN | SLAB_RECLAIM_ACCOUNT |
1146
1125
SLAB_ACCOUNT | SLAB_PANIC), NULL);
1147
-
1148
-
pidfs_xattr_cachep = kmem_cache_create("pidfs_xattr_cache",
1149
-
sizeof(struct simple_xattrs), 0,
1150
-
(SLAB_HWCACHE_ALIGN | SLAB_RECLAIM_ACCOUNT |
1151
-
SLAB_ACCOUNT | SLAB_PANIC), NULL);
1152
1126
1153
1127
pidfs_mnt = kern_mount(&pidfs_type);
1154
1128
if (IS_ERR(pidfs_mnt))