1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/anon_inodes.h>
3 #include <linux/file.h>
5 #include <linux/magic.h>
6 #include <linux/mount.h>
8 #include <linux/pidfs.h>
9 #include <linux/pid_namespace.h>
10 #include <linux/poll.h>
11 #include <linux/proc_fs.h>
12 #include <linux/proc_ns.h>
13 #include <linux/pseudo_fs.h>
14 #include <linux/ptrace.h>
15 #include <linux/seq_file.h>
16 #include <uapi/linux/pidfd.h>
17 #include <linux/ipc_namespace.h>
18 #include <linux/time_namespace.h>
19 #include <linux/utsname.h>
20 #include <net/net_namespace.h>
27 * pidfd_show_fdinfo - print information about a pidfd
28 * @m: proc fdinfo file
29 * @f: file referencing a pidfd
32 * This function will print the pid that a given pidfd refers to in the
33 * pid namespace of the procfs instance.
34 * If the pid namespace of the process is not a descendant of the pid
35 * namespace of the procfs instance 0 will be shown as its pid. This is
36 * similar to calling getppid() on a process whose parent is outside of
40 * If pid namespaces are supported then this function will also print
41 * the pid of a given pidfd refers to for all descendant pid namespaces
42 * starting from the current pid namespace of the instance, i.e. the
43 * Pid field and the first entry in the NSpid field will be identical.
44 * If the pid namespace of the process is not a descendant of the pid
45 * namespace of the procfs instance 0 will be shown as its first NSpid
46 * entry and no others will be shown.
47 * Note that this differs from the Pid and NSpid fields in
48 * /proc/<pid>/status where Pid and NSpid are always shown relative to
49 * the pid namespace of the procfs instance. The difference becomes
50 * obvious when sending around a pidfd between pid namespaces from a
51 * different branch of the tree, i.e. where no ancestral relation is
52 * present between the pid namespaces:
53 * - create two new pid namespaces ns1 and ns2 in the initial pid
54 * namespace (also take care to create new mount namespaces in the
55 * new pid namespace and mount procfs)
56 * - create a process with a pidfd in ns1
57 * - send pidfd from ns1 to ns2
58 * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
59 * have exactly one entry, which is 0
61 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
63 struct pid *pid = pidfd_pid(f);
64 struct pid_namespace *ns;
67 if (likely(pid_has_task(pid, PIDTYPE_PID))) {
68 ns = proc_pid_ns(file_inode(m->file)->i_sb);
69 nr = pid_nr_ns(pid, ns);
72 seq_put_decimal_ll(m, "Pid:\t", nr);
75 seq_put_decimal_ll(m, "\nNSpid:\t", nr);
79 /* If nr is non-zero it means that 'pid' is valid and that
80 * ns, i.e. the pid namespace associated with the procfs
81 * instance, is in the pid namespace hierarchy of pid.
82 * Start at one below the already printed level.
84 for (i = ns->level + 1; i <= pid->level; i++)
85 seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
93 * Poll support for process exit notification.
95 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
97 struct pid *pid = pidfd_pid(file);
98 bool thread = file->f_flags & PIDFD_THREAD;
99 struct task_struct *task;
100 __poll_t poll_flags = 0;
102 poll_wait(file, &pid->wait_pidfd, pts);
104 * Depending on PIDFD_THREAD, inform pollers when the thread
105 * or the whole thread-group exits.
108 task = pid_task(pid, PIDTYPE_PID);
110 poll_flags = EPOLLIN | EPOLLRDNORM | EPOLLHUP;
111 else if (task->exit_state && (thread || thread_group_empty(task)))
112 poll_flags = EPOLLIN | EPOLLRDNORM;
117 static long pidfd_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
119 struct task_struct *task __free(put_task) = NULL;
120 struct nsproxy *nsp __free(put_nsproxy) = NULL;
121 struct pid *pid = pidfd_pid(file);
122 struct ns_common *ns_common = NULL;
123 struct pid_namespace *pid_ns;
128 task = get_pid_task(pid, PIDTYPE_PID);
132 scoped_guard(task_lock, task) {
138 return -ESRCH; /* just pretend it didn't exist */
141 * We're trying to open a file descriptor to the namespace so perform a
142 * filesystem cred ptrace check. Also, we mirror nsfs behavior.
144 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
148 /* Namespaces that hang of nsproxy. */
149 case PIDFD_GET_CGROUP_NAMESPACE:
150 if (IS_ENABLED(CONFIG_CGROUPS)) {
151 get_cgroup_ns(nsp->cgroup_ns);
152 ns_common = to_ns_common(nsp->cgroup_ns);
155 case PIDFD_GET_IPC_NAMESPACE:
156 if (IS_ENABLED(CONFIG_IPC_NS)) {
157 get_ipc_ns(nsp->ipc_ns);
158 ns_common = to_ns_common(nsp->ipc_ns);
161 case PIDFD_GET_MNT_NAMESPACE:
162 get_mnt_ns(nsp->mnt_ns);
163 ns_common = to_ns_common(nsp->mnt_ns);
165 case PIDFD_GET_NET_NAMESPACE:
166 if (IS_ENABLED(CONFIG_NET_NS)) {
167 ns_common = to_ns_common(nsp->net_ns);
168 get_net_ns(ns_common);
171 case PIDFD_GET_PID_FOR_CHILDREN_NAMESPACE:
172 if (IS_ENABLED(CONFIG_PID_NS)) {
173 get_pid_ns(nsp->pid_ns_for_children);
174 ns_common = to_ns_common(nsp->pid_ns_for_children);
177 case PIDFD_GET_TIME_NAMESPACE:
178 if (IS_ENABLED(CONFIG_TIME_NS)) {
179 get_time_ns(nsp->time_ns);
180 ns_common = to_ns_common(nsp->time_ns);
183 case PIDFD_GET_TIME_FOR_CHILDREN_NAMESPACE:
184 if (IS_ENABLED(CONFIG_TIME_NS)) {
185 get_time_ns(nsp->time_ns_for_children);
186 ns_common = to_ns_common(nsp->time_ns_for_children);
189 case PIDFD_GET_UTS_NAMESPACE:
190 if (IS_ENABLED(CONFIG_UTS_NS)) {
191 get_uts_ns(nsp->uts_ns);
192 ns_common = to_ns_common(nsp->uts_ns);
195 /* Namespaces that don't hang of nsproxy. */
196 case PIDFD_GET_USER_NAMESPACE:
197 if (IS_ENABLED(CONFIG_USER_NS)) {
199 ns_common = to_ns_common(get_user_ns(task_cred_xxx(task, user_ns)));
203 case PIDFD_GET_PID_NAMESPACE:
204 if (IS_ENABLED(CONFIG_PID_NS)) {
206 pid_ns = task_active_pid_ns(task);
208 ns_common = to_ns_common(get_pid_ns(pid_ns));
219 /* open_namespace() unconditionally consumes the reference */
220 return open_namespace(ns_common);
223 static const struct file_operations pidfs_file_operations = {
225 #ifdef CONFIG_PROC_FS
226 .show_fdinfo = pidfd_show_fdinfo,
228 .unlocked_ioctl = pidfd_ioctl,
229 .compat_ioctl = compat_ptr_ioctl,
232 struct pid *pidfd_pid(const struct file *file)
234 if (file->f_op != &pidfs_file_operations)
235 return ERR_PTR(-EBADF);
236 return file_inode(file)->i_private;
239 static struct vfsmount *pidfs_mnt __ro_after_init;
241 #if BITS_PER_LONG == 32
243 * Provide a fallback mechanism for 32-bit systems so processes remain
244 * reliably comparable by inode number even on those systems.
246 static DEFINE_IDA(pidfd_inum_ida);
248 static int pidfs_inum(struct pid *pid, unsigned long *ino)
252 ret = ida_alloc_range(&pidfd_inum_ida, RESERVED_PIDS + 1,
253 UINT_MAX, GFP_ATOMIC);
261 static inline void pidfs_free_inum(unsigned long ino)
264 ida_free(&pidfd_inum_ida, ino);
267 static inline int pidfs_inum(struct pid *pid, unsigned long *ino)
272 #define pidfs_free_inum(ino) ((void)(ino))
276 * The vfs falls back to simple_setattr() if i_op->setattr() isn't
277 * implemented. Let's reject it completely until we have a clean
278 * permission concept for pidfds.
280 static int pidfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
288 * User space expects pidfs inodes to have no file type in st_mode.
290 * In particular, 'lsof' has this legacy logic:
292 * type = s->st_mode & S_IFMT;
296 * if (!strcmp(p, "anon_inode"))
297 * Lf->ntype = Ntype = N_ANON_INODE;
299 * to detect our old anon_inode logic.
301 * Rather than mess with our internal sane inode data, just fix it
302 * up here in getattr() by masking off the format bits.
304 static int pidfs_getattr(struct mnt_idmap *idmap, const struct path *path,
305 struct kstat *stat, u32 request_mask,
306 unsigned int query_flags)
308 struct inode *inode = d_inode(path->dentry);
310 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
311 stat->mode &= ~S_IFMT;
315 static const struct inode_operations pidfs_inode_operations = {
316 .getattr = pidfs_getattr,
317 .setattr = pidfs_setattr,
320 static void pidfs_evict_inode(struct inode *inode)
322 struct pid *pid = inode->i_private;
326 pidfs_free_inum(inode->i_ino);
329 static const struct super_operations pidfs_sops = {
330 .drop_inode = generic_delete_inode,
331 .evict_inode = pidfs_evict_inode,
332 .statfs = simple_statfs,
336 * 'lsof' has knowledge of out historical anon_inode use, and expects
337 * the pidfs dentry name to start with 'anon_inode'.
339 static char *pidfs_dname(struct dentry *dentry, char *buffer, int buflen)
341 return dynamic_dname(buffer, buflen, "anon_inode:[pidfd]");
344 static const struct dentry_operations pidfs_dentry_operations = {
345 .d_delete = always_delete_dentry,
346 .d_dname = pidfs_dname,
347 .d_prune = stashed_dentry_prune,
350 static int pidfs_init_inode(struct inode *inode, void *data)
352 inode->i_private = data;
353 inode->i_flags |= S_PRIVATE;
354 inode->i_mode |= S_IRWXU;
355 inode->i_op = &pidfs_inode_operations;
356 inode->i_fop = &pidfs_file_operations;
358 * Inode numbering for pidfs start at RESERVED_PIDS + 1. This
359 * avoids collisions with the root inode which is 1 for pseudo
362 return pidfs_inum(data, &inode->i_ino);
365 static void pidfs_put_data(void *data)
367 struct pid *pid = data;
371 static const struct stashed_operations pidfs_stashed_ops = {
372 .init_inode = pidfs_init_inode,
373 .put_data = pidfs_put_data,
376 static int pidfs_init_fs_context(struct fs_context *fc)
378 struct pseudo_fs_context *ctx;
380 ctx = init_pseudo(fc, PID_FS_MAGIC);
384 ctx->ops = &pidfs_sops;
385 ctx->dops = &pidfs_dentry_operations;
386 fc->s_fs_info = (void *)&pidfs_stashed_ops;
390 static struct file_system_type pidfs_type = {
392 .init_fs_context = pidfs_init_fs_context,
393 .kill_sb = kill_anon_super,
396 struct file *pidfs_alloc_file(struct pid *pid, unsigned int flags)
399 struct file *pidfd_file;
403 ret = path_from_stashed(&pid->stashed, pidfs_mnt, get_pid(pid), &path);
407 pidfd_file = dentry_open(&path, flags, current_cred());
412 void __init pidfs_init(void)
414 pidfs_mnt = kern_mount(&pidfs_type);
415 if (IS_ERR(pidfs_mnt))
416 panic("Failed to mount pidfs pseudo filesystem");