Commit | Line | Data |
---|---|---|
f4e65870 JA |
1 | // SPDX-License-Identifier: GPL-2.0 |
2 | #include <linux/module.h> | |
3 | #include <linux/kernel.h> | |
4 | #include <linux/string.h> | |
5 | #include <linux/socket.h> | |
6 | #include <linux/net.h> | |
7 | #include <linux/fs.h> | |
8 | #include <net/af_unix.h> | |
9 | #include <net/scm.h> | |
10 | #include <linux/init.h> | |
a3ec6005 | 11 | #include <linux/io_uring.h> |
f4e65870 JA |
12 | |
13 | #include "scm.h" | |
14 | ||
15 | unsigned int unix_tot_inflight; | |
16 | EXPORT_SYMBOL(unix_tot_inflight); | |
17 | ||
18 | LIST_HEAD(gc_inflight_list); | |
19 | EXPORT_SYMBOL(gc_inflight_list); | |
20 | ||
21 | DEFINE_SPINLOCK(unix_gc_lock); | |
22 | EXPORT_SYMBOL(unix_gc_lock); | |
23 | ||
24 | struct sock *unix_get_socket(struct file *filp) | |
25 | { | |
26 | struct sock *u_sock = NULL; | |
27 | struct inode *inode = file_inode(filp); | |
28 | ||
29 | /* Socket ? */ | |
30 | if (S_ISSOCK(inode->i_mode) && !(filp->f_mode & FMODE_PATH)) { | |
31 | struct socket *sock = SOCKET_I(inode); | |
32 | struct sock *s = sock->sk; | |
33 | ||
34 | /* PF_UNIX ? */ | |
35 | if (s && sock->ops && sock->ops->family == PF_UNIX) | |
36 | u_sock = s; | |
37 | } else { | |
38 | /* Could be an io_uring instance */ | |
39 | u_sock = io_uring_get_socket(filp); | |
40 | } | |
41 | return u_sock; | |
42 | } | |
43 | EXPORT_SYMBOL(unix_get_socket); | |
44 | ||
45 | /* Keep the number of times in flight count for the file | |
46 | * descriptor if it is for an AF_UNIX socket. | |
47 | */ | |
48 | void unix_inflight(struct user_struct *user, struct file *fp) | |
49 | { | |
50 | struct sock *s = unix_get_socket(fp); | |
51 | ||
52 | spin_lock(&unix_gc_lock); | |
53 | ||
54 | if (s) { | |
55 | struct unix_sock *u = unix_sk(s); | |
56 | ||
57 | if (atomic_long_inc_return(&u->inflight) == 1) { | |
58 | BUG_ON(!list_empty(&u->link)); | |
59 | list_add_tail(&u->link, &gc_inflight_list); | |
60 | } else { | |
61 | BUG_ON(list_empty(&u->link)); | |
62 | } | |
9d6d7f1c ED |
63 | /* Paired with READ_ONCE() in wait_for_unix_gc() */ |
64 | WRITE_ONCE(unix_tot_inflight, unix_tot_inflight + 1); | |
f4e65870 JA |
65 | } |
66 | user->unix_inflight++; | |
67 | spin_unlock(&unix_gc_lock); | |
68 | } | |
69 | ||
70 | void unix_notinflight(struct user_struct *user, struct file *fp) | |
71 | { | |
72 | struct sock *s = unix_get_socket(fp); | |
73 | ||
74 | spin_lock(&unix_gc_lock); | |
75 | ||
76 | if (s) { | |
77 | struct unix_sock *u = unix_sk(s); | |
78 | ||
79 | BUG_ON(!atomic_long_read(&u->inflight)); | |
80 | BUG_ON(list_empty(&u->link)); | |
81 | ||
82 | if (atomic_long_dec_and_test(&u->inflight)) | |
83 | list_del_init(&u->link); | |
9d6d7f1c ED |
84 | /* Paired with READ_ONCE() in wait_for_unix_gc() */ |
85 | WRITE_ONCE(unix_tot_inflight, unix_tot_inflight - 1); | |
f4e65870 JA |
86 | } |
87 | user->unix_inflight--; | |
88 | spin_unlock(&unix_gc_lock); | |
89 | } | |
90 | ||
91 | /* | |
92 | * The "user->unix_inflight" variable is protected by the garbage | |
93 | * collection lock, and we just read it locklessly here. If you go | |
94 | * over the limit, there might be a tiny race in actually noticing | |
95 | * it across threads. Tough. | |
96 | */ | |
97 | static inline bool too_many_unix_fds(struct task_struct *p) | |
98 | { | |
99 | struct user_struct *user = current_user(); | |
100 | ||
101 | if (unlikely(user->unix_inflight > task_rlimit(p, RLIMIT_NOFILE))) | |
102 | return !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN); | |
103 | return false; | |
104 | } | |
105 | ||
106 | int unix_attach_fds(struct scm_cookie *scm, struct sk_buff *skb) | |
107 | { | |
108 | int i; | |
109 | ||
110 | if (too_many_unix_fds(current)) | |
111 | return -ETOOMANYREFS; | |
112 | ||
113 | /* | |
114 | * Need to duplicate file references for the sake of garbage | |
115 | * collection. Otherwise a socket in the fps might become a | |
116 | * candidate for GC while the skb is not yet queued. | |
117 | */ | |
118 | UNIXCB(skb).fp = scm_fp_dup(scm->fp); | |
119 | if (!UNIXCB(skb).fp) | |
120 | return -ENOMEM; | |
121 | ||
122 | for (i = scm->fp->count - 1; i >= 0; i--) | |
123 | unix_inflight(scm->fp->user, scm->fp->fp[i]); | |
124 | return 0; | |
125 | } | |
126 | EXPORT_SYMBOL(unix_attach_fds); | |
127 | ||
128 | void unix_detach_fds(struct scm_cookie *scm, struct sk_buff *skb) | |
129 | { | |
130 | int i; | |
131 | ||
132 | scm->fp = UNIXCB(skb).fp; | |
133 | UNIXCB(skb).fp = NULL; | |
134 | ||
135 | for (i = scm->fp->count-1; i >= 0; i--) | |
136 | unix_notinflight(scm->fp->user, scm->fp->fp[i]); | |
137 | } | |
138 | EXPORT_SYMBOL(unix_detach_fds); | |
139 | ||
140 | void unix_destruct_scm(struct sk_buff *skb) | |
141 | { | |
142 | struct scm_cookie scm; | |
143 | ||
144 | memset(&scm, 0, sizeof(scm)); | |
145 | scm.pid = UNIXCB(skb).pid; | |
146 | if (UNIXCB(skb).fp) | |
147 | unix_detach_fds(&scm, skb); | |
148 | ||
149 | /* Alas, it calls VFS */ | |
150 | /* So fscking what? fput() had been SMP-safe since the last Summer */ | |
151 | scm_destroy(&scm); | |
152 | sock_wfree(skb); | |
153 | } | |
154 | EXPORT_SYMBOL(unix_destruct_scm); |