1/*
2 *  linux/fs/proc/base.c
3 *
4 *  Copyright (C) 1991, 1992 Linus Torvalds
5 *
6 *  proc base directory handling functions
7 *
8 *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
9 *  Instead of using magical inumbers to determine the kind of object
10 *  we allocate and fill in-core inodes upon lookup. They don't even
11 *  go into icache. We cache the reference to task_struct upon lookup too.
12 *  Eventually it should become a filesystem in its own. We don't use the
13 *  rest of procfs anymore.
14 *
15 *
16 *  Changelog:
17 *  17-Jan-2005
18 *  Allan Bezerra
19 *  Bruna Moreira <bruna.moreira@indt.org.br>
20 *  Edjard Mota <edjard.mota@indt.org.br>
21 *  Ilias Biris <ilias.biris@indt.org.br>
22 *  Mauricio Lin <mauricio.lin@indt.org.br>
23 *
24 *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
25 *
26 *  A new process specific entry (smaps) included in /proc. It shows the
27 *  size of rss for each memory area. The maps entry lacks information
28 *  about physical memory size (rss) for each mapped file, i.e.,
29 *  rss information for executables and library files.
30 *  This additional information is useful for any tools that need to know
31 *  about physical memory consumption for a process specific library.
32 *
33 *  Changelog:
34 *  21-Feb-2005
35 *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
36 *  Pud inclusion in the page table walking.
37 *
38 *  ChangeLog:
39 *  10-Mar-2005
40 *  10LE Instituto Nokia de Tecnologia - INdT:
41 *  A better way to walks through the page table as suggested by Hugh Dickins.
42 *
43 *  Simo Piiroinen <simo.piiroinen@nokia.com>:
44 *  Smaps information related to shared, private, clean and dirty pages.
45 *
46 *  Paul Mundt <paul.mundt@nokia.com>:
47 *  Overall revision about smaps.
48 */
49
50#include <asm/uaccess.h>
51
52#include <linux/errno.h>
53#include <linux/time.h>
54#include <linux/proc_fs.h>
55#include <linux/stat.h>
56#include <linux/task_io_accounting_ops.h>
57#include <linux/init.h>
58#include <linux/capability.h>
59#include <linux/file.h>
60#include <linux/fdtable.h>
61#include <linux/string.h>
62#include <linux/seq_file.h>
63#include <linux/namei.h>
64#include <linux/mnt_namespace.h>
65#include <linux/mm.h>
66#include <linux/swap.h>
67#include <linux/rcupdate.h>
68#include <linux/kallsyms.h>
69#include <linux/stacktrace.h>
70#include <linux/resource.h>
71#include <linux/module.h>
72#include <linux/mount.h>
73#include <linux/security.h>
74#include <linux/ptrace.h>
75#include <linux/tracehook.h>
76#include <linux/printk.h>
77#include <linux/cgroup.h>
78#include <linux/cpuset.h>
79#include <linux/audit.h>
80#include <linux/poll.h>
81#include <linux/nsproxy.h>
82#include <linux/oom.h>
83#include <linux/elf.h>
84#include <linux/pid_namespace.h>
85#include <linux/user_namespace.h>
86#include <linux/fs_struct.h>
87#include <linux/slab.h>
88#include <linux/flex_array.h>
89#include <linux/posix-timers.h>
90#ifdef CONFIG_HARDWALL
91#include <asm/hardwall.h>
92#endif
93#include <trace/events/oom.h>
94#include "internal.h"
95#include "fd.h"
96
97/* NOTE:
98 *	Implementing inode permission operations in /proc is almost
99 *	certainly an error.  Permission checks need to happen during
100 *	each system call not at open time.  The reason is that most of
101 *	what we wish to check for permissions in /proc varies at runtime.
102 *
103 *	The classic example of a problem is opening file descriptors
104 *	in /proc for a task before it execs a suid executable.
105 */
106
107struct pid_entry {
108	const char *name;
109	int len;
110	umode_t mode;
111	const struct inode_operations *iop;
112	const struct file_operations *fop;
113	union proc_op op;
114};
115
116#define NOD(NAME, MODE, IOP, FOP, OP) {			\
117	.name = (NAME),					\
118	.len  = sizeof(NAME) - 1,			\
119	.mode = MODE,					\
120	.iop  = IOP,					\
121	.fop  = FOP,					\
122	.op   = OP,					\
123}
124
125#define DIR(NAME, MODE, iops, fops)	\
126	NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
127#define LNK(NAME, get_link)					\
128	NOD(NAME, (S_IFLNK|S_IRWXUGO),				\
129		&proc_pid_link_inode_operations, NULL,		\
130		{ .proc_get_link = get_link } )
131#define REG(NAME, MODE, fops)				\
132	NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
133#define ONE(NAME, MODE, show)				\
134	NOD(NAME, (S_IFREG|(MODE)), 			\
135		NULL, &proc_single_file_operations,	\
136		{ .proc_show = show } )
137
138/*
139 * Count the number of hardlinks for the pid_entry table, excluding the .
140 * and .. links.
141 */
142static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
143	unsigned int n)
144{
145	unsigned int i;
146	unsigned int count;
147
148	count = 0;
149	for (i = 0; i < n; ++i) {
150		if (S_ISDIR(entries[i].mode))
151			++count;
152	}
153
154	return count;
155}
156
157static int get_task_root(struct task_struct *task, struct path *root)
158{
159	int result = -ENOENT;
160
161	task_lock(task);
162	if (task->fs) {
163		get_fs_root(task->fs, root);
164		result = 0;
165	}
166	task_unlock(task);
167	return result;
168}
169
170static int proc_cwd_link(struct dentry *dentry, struct path *path)
171{
172	struct task_struct *task = get_proc_task(d_inode(dentry));
173	int result = -ENOENT;
174
175	if (task) {
176		task_lock(task);
177		if (task->fs) {
178			get_fs_pwd(task->fs, path);
179			result = 0;
180		}
181		task_unlock(task);
182		put_task_struct(task);
183	}
184	return result;
185}
186
187static int proc_root_link(struct dentry *dentry, struct path *path)
188{
189	struct task_struct *task = get_proc_task(d_inode(dentry));
190	int result = -ENOENT;
191
192	if (task) {
193		result = get_task_root(task, path);
194		put_task_struct(task);
195	}
196	return result;
197}
198
199static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
200				     size_t _count, loff_t *pos)
201{
202	struct task_struct *tsk;
203	struct mm_struct *mm;
204	char *page;
205	unsigned long count = _count;
206	unsigned long arg_start, arg_end, env_start, env_end;
207	unsigned long len1, len2, len;
208	unsigned long p;
209	char c;
210	ssize_t rv;
211
212	BUG_ON(*pos < 0);
213
214	tsk = get_proc_task(file_inode(file));
215	if (!tsk)
216		return -ESRCH;
217	mm = get_task_mm(tsk);
218	put_task_struct(tsk);
219	if (!mm)
220		return 0;
221	/* Check if process spawned far enough to have cmdline. */
222	if (!mm->env_end) {
223		rv = 0;
224		goto out_mmput;
225	}
226
227	page = (char *)__get_free_page(GFP_TEMPORARY);
228	if (!page) {
229		rv = -ENOMEM;
230		goto out_mmput;
231	}
232
233	down_read(&mm->mmap_sem);
234	arg_start = mm->arg_start;
235	arg_end = mm->arg_end;
236	env_start = mm->env_start;
237	env_end = mm->env_end;
238	up_read(&mm->mmap_sem);
239
240	BUG_ON(arg_start > arg_end);
241	BUG_ON(env_start > env_end);
242
243	len1 = arg_end - arg_start;
244	len2 = env_end - env_start;
245
246	/* Empty ARGV. */
247	if (len1 == 0) {
248		rv = 0;
249		goto out_free_page;
250	}
251	/*
252	 * Inherently racy -- command line shares address space
253	 * with code and data.
254	 */
255	rv = access_remote_vm(mm, arg_end - 1, &c, 1, 0);
256	if (rv <= 0)
257		goto out_free_page;
258
259	rv = 0;
260
261	if (c == '\0') {
262		/* Command line (set of strings) occupies whole ARGV. */
263		if (len1 <= *pos)
264			goto out_free_page;
265
266		p = arg_start + *pos;
267		len = len1 - *pos;
268		while (count > 0 && len > 0) {
269			unsigned int _count;
270			int nr_read;
271
272			_count = min3(count, len, PAGE_SIZE);
273			nr_read = access_remote_vm(mm, p, page, _count, 0);
274			if (nr_read < 0)
275				rv = nr_read;
276			if (nr_read <= 0)
277				goto out_free_page;
278
279			if (copy_to_user(buf, page, nr_read)) {
280				rv = -EFAULT;
281				goto out_free_page;
282			}
283
284			p	+= nr_read;
285			len	-= nr_read;
286			buf	+= nr_read;
287			count	-= nr_read;
288			rv	+= nr_read;
289		}
290	} else {
291		/*
292		 * Command line (1 string) occupies ARGV and maybe
293		 * extends into ENVP.
294		 */
295		if (len1 + len2 <= *pos)
296			goto skip_argv_envp;
297		if (len1 <= *pos)
298			goto skip_argv;
299
300		p = arg_start + *pos;
301		len = len1 - *pos;
302		while (count > 0 && len > 0) {
303			unsigned int _count, l;
304			int nr_read;
305			bool final;
306
307			_count = min3(count, len, PAGE_SIZE);
308			nr_read = access_remote_vm(mm, p, page, _count, 0);
309			if (nr_read < 0)
310				rv = nr_read;
311			if (nr_read <= 0)
312				goto out_free_page;
313
314			/*
315			 * Command line can be shorter than whole ARGV
316			 * even if last "marker" byte says it is not.
317			 */
318			final = false;
319			l = strnlen(page, nr_read);
320			if (l < nr_read) {
321				nr_read = l;
322				final = true;
323			}
324
325			if (copy_to_user(buf, page, nr_read)) {
326				rv = -EFAULT;
327				goto out_free_page;
328			}
329
330			p	+= nr_read;
331			len	-= nr_read;
332			buf	+= nr_read;
333			count	-= nr_read;
334			rv	+= nr_read;
335
336			if (final)
337				goto out_free_page;
338		}
339skip_argv:
340		/*
341		 * Command line (1 string) occupies ARGV and
342		 * extends into ENVP.
343		 */
344		if (len1 <= *pos) {
345			p = env_start + *pos - len1;
346			len = len1 + len2 - *pos;
347		} else {
348			p = env_start;
349			len = len2;
350		}
351		while (count > 0 && len > 0) {
352			unsigned int _count, l;
353			int nr_read;
354			bool final;
355
356			_count = min3(count, len, PAGE_SIZE);
357			nr_read = access_remote_vm(mm, p, page, _count, 0);
358			if (nr_read < 0)
359				rv = nr_read;
360			if (nr_read <= 0)
361				goto out_free_page;
362
363			/* Find EOS. */
364			final = false;
365			l = strnlen(page, nr_read);
366			if (l < nr_read) {
367				nr_read = l;
368				final = true;
369			}
370
371			if (copy_to_user(buf, page, nr_read)) {
372				rv = -EFAULT;
373				goto out_free_page;
374			}
375
376			p	+= nr_read;
377			len	-= nr_read;
378			buf	+= nr_read;
379			count	-= nr_read;
380			rv	+= nr_read;
381
382			if (final)
383				goto out_free_page;
384		}
385skip_argv_envp:
386		;
387	}
388
389out_free_page:
390	free_page((unsigned long)page);
391out_mmput:
392	mmput(mm);
393	if (rv > 0)
394		*pos += rv;
395	return rv;
396}
397
398static const struct file_operations proc_pid_cmdline_ops = {
399	.read	= proc_pid_cmdline_read,
400	.llseek	= generic_file_llseek,
401};
402
403static int proc_pid_auxv(struct seq_file *m, struct pid_namespace *ns,
404			 struct pid *pid, struct task_struct *task)
405{
406	struct mm_struct *mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
407	if (mm && !IS_ERR(mm)) {
408		unsigned int nwords = 0;
409		do {
410			nwords += 2;
411		} while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
412		seq_write(m, mm->saved_auxv, nwords * sizeof(mm->saved_auxv[0]));
413		mmput(mm);
414		return 0;
415	} else
416		return PTR_ERR(mm);
417}
418
419
420#ifdef CONFIG_KALLSYMS
421/*
422 * Provides a wchan file via kallsyms in a proper one-value-per-file format.
423 * Returns the resolved symbol.  If that fails, simply return the address.
424 */
425static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
426			  struct pid *pid, struct task_struct *task)
427{
428	unsigned long wchan;
429	char symname[KSYM_NAME_LEN];
430
431	wchan = get_wchan(task);
432
433	if (wchan && ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)
434			&& !lookup_symbol_name(wchan, symname))
435		seq_printf(m, "%s", symname);
436	else
437		seq_putc(m, '0');
438
439	return 0;
440}
441#endif /* CONFIG_KALLSYMS */
442
443static int lock_trace(struct task_struct *task)
444{
445	int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
446	if (err)
447		return err;
448	if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
449		mutex_unlock(&task->signal->cred_guard_mutex);
450		return -EPERM;
451	}
452	return 0;
453}
454
455static void unlock_trace(struct task_struct *task)
456{
457	mutex_unlock(&task->signal->cred_guard_mutex);
458}
459
460#ifdef CONFIG_STACKTRACE
461
462#define MAX_STACK_TRACE_DEPTH	64
463
464static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
465			  struct pid *pid, struct task_struct *task)
466{
467	struct stack_trace trace;
468	unsigned long *entries;
469	int err;
470	int i;
471
472	entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
473	if (!entries)
474		return -ENOMEM;
475
476	trace.nr_entries	= 0;
477	trace.max_entries	= MAX_STACK_TRACE_DEPTH;
478	trace.entries		= entries;
479	trace.skip		= 0;
480
481	err = lock_trace(task);
482	if (!err) {
483		save_stack_trace_tsk(task, &trace);
484
485		for (i = 0; i < trace.nr_entries; i++) {
486			seq_printf(m, "[<%pK>] %pS\n",
487				   (void *)entries[i], (void *)entries[i]);
488		}
489		unlock_trace(task);
490	}
491	kfree(entries);
492
493	return err;
494}
495#endif
496
497#ifdef CONFIG_SCHED_INFO
498/*
499 * Provides /proc/PID/schedstat
500 */
501static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
502			      struct pid *pid, struct task_struct *task)
503{
504	if (unlikely(!sched_info_on()))
505		seq_printf(m, "0 0 0\n");
506	else
507		seq_printf(m, "%llu %llu %lu\n",
508		   (unsigned long long)task->se.sum_exec_runtime,
509		   (unsigned long long)task->sched_info.run_delay,
510		   task->sched_info.pcount);
511
512	return 0;
513}
514#endif
515
516#ifdef CONFIG_LATENCYTOP
517static int lstats_show_proc(struct seq_file *m, void *v)
518{
519	int i;
520	struct inode *inode = m->private;
521	struct task_struct *task = get_proc_task(inode);
522
523	if (!task)
524		return -ESRCH;
525	seq_puts(m, "Latency Top version : v0.1\n");
526	for (i = 0; i < 32; i++) {
527		struct latency_record *lr = &task->latency_record[i];
528		if (lr->backtrace[0]) {
529			int q;
530			seq_printf(m, "%i %li %li",
531				   lr->count, lr->time, lr->max);
532			for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
533				unsigned long bt = lr->backtrace[q];
534				if (!bt)
535					break;
536				if (bt == ULONG_MAX)
537					break;
538				seq_printf(m, " %ps", (void *)bt);
539			}
540			seq_putc(m, '\n');
541		}
542
543	}
544	put_task_struct(task);
545	return 0;
546}
547
548static int lstats_open(struct inode *inode, struct file *file)
549{
550	return single_open(file, lstats_show_proc, inode);
551}
552
553static ssize_t lstats_write(struct file *file, const char __user *buf,
554			    size_t count, loff_t *offs)
555{
556	struct task_struct *task = get_proc_task(file_inode(file));
557
558	if (!task)
559		return -ESRCH;
560	clear_all_latency_tracing(task);
561	put_task_struct(task);
562
563	return count;
564}
565
566static const struct file_operations proc_lstats_operations = {
567	.open		= lstats_open,
568	.read		= seq_read,
569	.write		= lstats_write,
570	.llseek		= seq_lseek,
571	.release	= single_release,
572};
573
574#endif
575
576static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
577			  struct pid *pid, struct task_struct *task)
578{
579	unsigned long totalpages = totalram_pages + total_swap_pages;
580	unsigned long points = 0;
581
582	read_lock(&tasklist_lock);
583	if (pid_alive(task))
584		points = oom_badness(task, NULL, NULL, totalpages) *
585						1000 / totalpages;
586	read_unlock(&tasklist_lock);
587	seq_printf(m, "%lu\n", points);
588
589	return 0;
590}
591
592struct limit_names {
593	const char *name;
594	const char *unit;
595};
596
597static const struct limit_names lnames[RLIM_NLIMITS] = {
598	[RLIMIT_CPU] = {"Max cpu time", "seconds"},
599	[RLIMIT_FSIZE] = {"Max file size", "bytes"},
600	[RLIMIT_DATA] = {"Max data size", "bytes"},
601	[RLIMIT_STACK] = {"Max stack size", "bytes"},
602	[RLIMIT_CORE] = {"Max core file size", "bytes"},
603	[RLIMIT_RSS] = {"Max resident set", "bytes"},
604	[RLIMIT_NPROC] = {"Max processes", "processes"},
605	[RLIMIT_NOFILE] = {"Max open files", "files"},
606	[RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
607	[RLIMIT_AS] = {"Max address space", "bytes"},
608	[RLIMIT_LOCKS] = {"Max file locks", "locks"},
609	[RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
610	[RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
611	[RLIMIT_NICE] = {"Max nice priority", NULL},
612	[RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
613	[RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
614};
615
616/* Display limits for a process */
617static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
618			   struct pid *pid, struct task_struct *task)
619{
620	unsigned int i;
621	unsigned long flags;
622
623	struct rlimit rlim[RLIM_NLIMITS];
624
625	if (!lock_task_sighand(task, &flags))
626		return 0;
627	memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
628	unlock_task_sighand(task, &flags);
629
630	/*
631	 * print the file header
632	 */
633       seq_printf(m, "%-25s %-20s %-20s %-10s\n",
634		  "Limit", "Soft Limit", "Hard Limit", "Units");
635
636	for (i = 0; i < RLIM_NLIMITS; i++) {
637		if (rlim[i].rlim_cur == RLIM_INFINITY)
638			seq_printf(m, "%-25s %-20s ",
639				   lnames[i].name, "unlimited");
640		else
641			seq_printf(m, "%-25s %-20lu ",
642				   lnames[i].name, rlim[i].rlim_cur);
643
644		if (rlim[i].rlim_max == RLIM_INFINITY)
645			seq_printf(m, "%-20s ", "unlimited");
646		else
647			seq_printf(m, "%-20lu ", rlim[i].rlim_max);
648
649		if (lnames[i].unit)
650			seq_printf(m, "%-10s\n", lnames[i].unit);
651		else
652			seq_putc(m, '\n');
653	}
654
655	return 0;
656}
657
658#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
659static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
660			    struct pid *pid, struct task_struct *task)
661{
662	long nr;
663	unsigned long args[6], sp, pc;
664	int res;
665
666	res = lock_trace(task);
667	if (res)
668		return res;
669
670	if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
671		seq_puts(m, "running\n");
672	else if (nr < 0)
673		seq_printf(m, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
674	else
675		seq_printf(m,
676		       "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
677		       nr,
678		       args[0], args[1], args[2], args[3], args[4], args[5],
679		       sp, pc);
680	unlock_trace(task);
681
682	return 0;
683}
684#endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
685
686/************************************************************************/
687/*                       Here the fs part begins                        */
688/************************************************************************/
689
690/* permission checks */
691static int proc_fd_access_allowed(struct inode *inode)
692{
693	struct task_struct *task;
694	int allowed = 0;
695	/* Allow access to a task's file descriptors if it is us or we
696	 * may use ptrace attach to the process and find out that
697	 * information.
698	 */
699	task = get_proc_task(inode);
700	if (task) {
701		allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
702		put_task_struct(task);
703	}
704	return allowed;
705}
706
707int proc_setattr(struct dentry *dentry, struct iattr *attr)
708{
709	int error;
710	struct inode *inode = d_inode(dentry);
711
712	if (attr->ia_valid & ATTR_MODE)
713		return -EPERM;
714
715	error = inode_change_ok(inode, attr);
716	if (error)
717		return error;
718
719	setattr_copy(inode, attr);
720	mark_inode_dirty(inode);
721	return 0;
722}
723
724/*
725 * May current process learn task's sched/cmdline info (for hide_pid_min=1)
726 * or euid/egid (for hide_pid_min=2)?
727 */
728static bool has_pid_permissions(struct pid_namespace *pid,
729				 struct task_struct *task,
730				 int hide_pid_min)
731{
732	if (pid->hide_pid < hide_pid_min)
733		return true;
734	if (in_group_p(pid->pid_gid))
735		return true;
736	return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
737}
738
739
740static int proc_pid_permission(struct inode *inode, int mask)
741{
742	struct pid_namespace *pid = inode->i_sb->s_fs_info;
743	struct task_struct *task;
744	bool has_perms;
745
746	task = get_proc_task(inode);
747	if (!task)
748		return -ESRCH;
749	has_perms = has_pid_permissions(pid, task, 1);
750	put_task_struct(task);
751
752	if (!has_perms) {
753		if (pid->hide_pid == 2) {
754			/*
755			 * Let's make getdents(), stat(), and open()
756			 * consistent with each other.  If a process
757			 * may not stat() a file, it shouldn't be seen
758			 * in procfs at all.
759			 */
760			return -ENOENT;
761		}
762
763		return -EPERM;
764	}
765	return generic_permission(inode, mask);
766}
767
768
769
770static const struct inode_operations proc_def_inode_operations = {
771	.setattr	= proc_setattr,
772};
773
774static int proc_single_show(struct seq_file *m, void *v)
775{
776	struct inode *inode = m->private;
777	struct pid_namespace *ns;
778	struct pid *pid;
779	struct task_struct *task;
780	int ret;
781
782	ns = inode->i_sb->s_fs_info;
783	pid = proc_pid(inode);
784	task = get_pid_task(pid, PIDTYPE_PID);
785	if (!task)
786		return -ESRCH;
787
788	ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
789
790	put_task_struct(task);
791	return ret;
792}
793
794static int proc_single_open(struct inode *inode, struct file *filp)
795{
796	return single_open(filp, proc_single_show, inode);
797}
798
799static const struct file_operations proc_single_file_operations = {
800	.open		= proc_single_open,
801	.read		= seq_read,
802	.llseek		= seq_lseek,
803	.release	= single_release,
804};
805
806
807struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
808{
809	struct task_struct *task = get_proc_task(inode);
810	struct mm_struct *mm = ERR_PTR(-ESRCH);
811
812	if (task) {
813		mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
814		put_task_struct(task);
815
816		if (!IS_ERR_OR_NULL(mm)) {
817			/* ensure this mm_struct can't be freed */
818			atomic_inc(&mm->mm_count);
819			/* but do not pin its memory */
820			mmput(mm);
821		}
822	}
823
824	return mm;
825}
826
827static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
828{
829	struct mm_struct *mm = proc_mem_open(inode, mode);
830
831	if (IS_ERR(mm))
832		return PTR_ERR(mm);
833
834	file->private_data = mm;
835	return 0;
836}
837
838static int mem_open(struct inode *inode, struct file *file)
839{
840	int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
841
842	/* OK to pass negative loff_t, we can catch out-of-range */
843	file->f_mode |= FMODE_UNSIGNED_OFFSET;
844
845	return ret;
846}
847
848static ssize_t mem_rw(struct file *file, char __user *buf,
849			size_t count, loff_t *ppos, int write)
850{
851	struct mm_struct *mm = file->private_data;
852	unsigned long addr = *ppos;
853	ssize_t copied;
854	char *page;
855
856	if (!mm)
857		return 0;
858
859	page = (char *)__get_free_page(GFP_TEMPORARY);
860	if (!page)
861		return -ENOMEM;
862
863	copied = 0;
864	if (!atomic_inc_not_zero(&mm->mm_users))
865		goto free;
866
867	while (count > 0) {
868		int this_len = min_t(int, count, PAGE_SIZE);
869
870		if (write && copy_from_user(page, buf, this_len)) {
871			copied = -EFAULT;
872			break;
873		}
874
875		this_len = access_remote_vm(mm, addr, page, this_len, write);
876		if (!this_len) {
877			if (!copied)
878				copied = -EIO;
879			break;
880		}
881
882		if (!write && copy_to_user(buf, page, this_len)) {
883			copied = -EFAULT;
884			break;
885		}
886
887		buf += this_len;
888		addr += this_len;
889		copied += this_len;
890		count -= this_len;
891	}
892	*ppos = addr;
893
894	mmput(mm);
895free:
896	free_page((unsigned long) page);
897	return copied;
898}
899
900static ssize_t mem_read(struct file *file, char __user *buf,
901			size_t count, loff_t *ppos)
902{
903	return mem_rw(file, buf, count, ppos, 0);
904}
905
906static ssize_t mem_write(struct file *file, const char __user *buf,
907			 size_t count, loff_t *ppos)
908{
909	return mem_rw(file, (char __user*)buf, count, ppos, 1);
910}
911
912loff_t mem_lseek(struct file *file, loff_t offset, int orig)
913{
914	switch (orig) {
915	case 0:
916		file->f_pos = offset;
917		break;
918	case 1:
919		file->f_pos += offset;
920		break;
921	default:
922		return -EINVAL;
923	}
924	force_successful_syscall_return();
925	return file->f_pos;
926}
927
928static int mem_release(struct inode *inode, struct file *file)
929{
930	struct mm_struct *mm = file->private_data;
931	if (mm)
932		mmdrop(mm);
933	return 0;
934}
935
936static const struct file_operations proc_mem_operations = {
937	.llseek		= mem_lseek,
938	.read		= mem_read,
939	.write		= mem_write,
940	.open		= mem_open,
941	.release	= mem_release,
942};
943
944static int environ_open(struct inode *inode, struct file *file)
945{
946	return __mem_open(inode, file, PTRACE_MODE_READ);
947}
948
949static ssize_t environ_read(struct file *file, char __user *buf,
950			size_t count, loff_t *ppos)
951{
952	char *page;
953	unsigned long src = *ppos;
954	int ret = 0;
955	struct mm_struct *mm = file->private_data;
956
957	/* Ensure the process spawned far enough to have an environment. */
958	if (!mm || !mm->env_end)
959		return 0;
960
961	page = (char *)__get_free_page(GFP_TEMPORARY);
962	if (!page)
963		return -ENOMEM;
964
965	ret = 0;
966	if (!atomic_inc_not_zero(&mm->mm_users))
967		goto free;
968	while (count > 0) {
969		size_t this_len, max_len;
970		int retval;
971
972		if (src >= (mm->env_end - mm->env_start))
973			break;
974
975		this_len = mm->env_end - (mm->env_start + src);
976
977		max_len = min_t(size_t, PAGE_SIZE, count);
978		this_len = min(max_len, this_len);
979
980		retval = access_remote_vm(mm, (mm->env_start + src),
981			page, this_len, 0);
982
983		if (retval <= 0) {
984			ret = retval;
985			break;
986		}
987
988		if (copy_to_user(buf, page, retval)) {
989			ret = -EFAULT;
990			break;
991		}
992
993		ret += retval;
994		src += retval;
995		buf += retval;
996		count -= retval;
997	}
998	*ppos = src;
999	mmput(mm);
1000
1001free:
1002	free_page((unsigned long) page);
1003	return ret;
1004}
1005
1006static const struct file_operations proc_environ_operations = {
1007	.open		= environ_open,
1008	.read		= environ_read,
1009	.llseek		= generic_file_llseek,
1010	.release	= mem_release,
1011};
1012
1013static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1014			    loff_t *ppos)
1015{
1016	struct task_struct *task = get_proc_task(file_inode(file));
1017	char buffer[PROC_NUMBUF];
1018	int oom_adj = OOM_ADJUST_MIN;
1019	size_t len;
1020	unsigned long flags;
1021
1022	if (!task)
1023		return -ESRCH;
1024	if (lock_task_sighand(task, &flags)) {
1025		if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1026			oom_adj = OOM_ADJUST_MAX;
1027		else
1028			oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1029				  OOM_SCORE_ADJ_MAX;
1030		unlock_task_sighand(task, &flags);
1031	}
1032	put_task_struct(task);
1033	len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1034	return simple_read_from_buffer(buf, count, ppos, buffer, len);
1035}
1036
1037/*
1038 * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1039 * kernels.  The effective policy is defined by oom_score_adj, which has a
1040 * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1041 * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1042 * Processes that become oom disabled via oom_adj will still be oom disabled
1043 * with this implementation.
1044 *
1045 * oom_adj cannot be removed since existing userspace binaries use it.
1046 */
1047static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1048			     size_t count, loff_t *ppos)
1049{
1050	struct task_struct *task;
1051	char buffer[PROC_NUMBUF];
1052	int oom_adj;
1053	unsigned long flags;
1054	int err;
1055
1056	memset(buffer, 0, sizeof(buffer));
1057	if (count > sizeof(buffer) - 1)
1058		count = sizeof(buffer) - 1;
1059	if (copy_from_user(buffer, buf, count)) {
1060		err = -EFAULT;
1061		goto out;
1062	}
1063
1064	err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1065	if (err)
1066		goto out;
1067	if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1068	     oom_adj != OOM_DISABLE) {
1069		err = -EINVAL;
1070		goto out;
1071	}
1072
1073	task = get_proc_task(file_inode(file));
1074	if (!task) {
1075		err = -ESRCH;
1076		goto out;
1077	}
1078
1079	task_lock(task);
1080	if (!task->mm) {
1081		err = -EINVAL;
1082		goto err_task_lock;
1083	}
1084
1085	if (!lock_task_sighand(task, &flags)) {
1086		err = -ESRCH;
1087		goto err_task_lock;
1088	}
1089
1090	/*
1091	 * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1092	 * value is always attainable.
1093	 */
1094	if (oom_adj == OOM_ADJUST_MAX)
1095		oom_adj = OOM_SCORE_ADJ_MAX;
1096	else
1097		oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1098
1099	if (oom_adj < task->signal->oom_score_adj &&
1100	    !capable(CAP_SYS_RESOURCE)) {
1101		err = -EACCES;
1102		goto err_sighand;
1103	}
1104
1105	/*
1106	 * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1107	 * /proc/pid/oom_score_adj instead.
1108	 */
1109	pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1110		  current->comm, task_pid_nr(current), task_pid_nr(task),
1111		  task_pid_nr(task));
1112
1113	task->signal->oom_score_adj = oom_adj;
1114	trace_oom_score_adj_update(task);
1115err_sighand:
1116	unlock_task_sighand(task, &flags);
1117err_task_lock:
1118	task_unlock(task);
1119	put_task_struct(task);
1120out:
1121	return err < 0 ? err : count;
1122}
1123
1124static const struct file_operations proc_oom_adj_operations = {
1125	.read		= oom_adj_read,
1126	.write		= oom_adj_write,
1127	.llseek		= generic_file_llseek,
1128};
1129
1130static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1131					size_t count, loff_t *ppos)
1132{
1133	struct task_struct *task = get_proc_task(file_inode(file));
1134	char buffer[PROC_NUMBUF];
1135	short oom_score_adj = OOM_SCORE_ADJ_MIN;
1136	unsigned long flags;
1137	size_t len;
1138
1139	if (!task)
1140		return -ESRCH;
1141	if (lock_task_sighand(task, &flags)) {
1142		oom_score_adj = task->signal->oom_score_adj;
1143		unlock_task_sighand(task, &flags);
1144	}
1145	put_task_struct(task);
1146	len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1147	return simple_read_from_buffer(buf, count, ppos, buffer, len);
1148}
1149
1150static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1151					size_t count, loff_t *ppos)
1152{
1153	struct task_struct *task;
1154	char buffer[PROC_NUMBUF];
1155	unsigned long flags;
1156	int oom_score_adj;
1157	int err;
1158
1159	memset(buffer, 0, sizeof(buffer));
1160	if (count > sizeof(buffer) - 1)
1161		count = sizeof(buffer) - 1;
1162	if (copy_from_user(buffer, buf, count)) {
1163		err = -EFAULT;
1164		goto out;
1165	}
1166
1167	err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1168	if (err)
1169		goto out;
1170	if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1171			oom_score_adj > OOM_SCORE_ADJ_MAX) {
1172		err = -EINVAL;
1173		goto out;
1174	}
1175
1176	task = get_proc_task(file_inode(file));
1177	if (!task) {
1178		err = -ESRCH;
1179		goto out;
1180	}
1181
1182	task_lock(task);
1183	if (!task->mm) {
1184		err = -EINVAL;
1185		goto err_task_lock;
1186	}
1187
1188	if (!lock_task_sighand(task, &flags)) {
1189		err = -ESRCH;
1190		goto err_task_lock;
1191	}
1192
1193	if ((short)oom_score_adj < task->signal->oom_score_adj_min &&
1194			!capable(CAP_SYS_RESOURCE)) {
1195		err = -EACCES;
1196		goto err_sighand;
1197	}
1198
1199	task->signal->oom_score_adj = (short)oom_score_adj;
1200	if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1201		task->signal->oom_score_adj_min = (short)oom_score_adj;
1202	trace_oom_score_adj_update(task);
1203
1204err_sighand:
1205	unlock_task_sighand(task, &flags);
1206err_task_lock:
1207	task_unlock(task);
1208	put_task_struct(task);
1209out:
1210	return err < 0 ? err : count;
1211}
1212
1213static const struct file_operations proc_oom_score_adj_operations = {
1214	.read		= oom_score_adj_read,
1215	.write		= oom_score_adj_write,
1216	.llseek		= default_llseek,
1217};
1218
1219#ifdef CONFIG_AUDITSYSCALL
1220#define TMPBUFLEN 21
1221static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1222				  size_t count, loff_t *ppos)
1223{
1224	struct inode * inode = file_inode(file);
1225	struct task_struct *task = get_proc_task(inode);
1226	ssize_t length;
1227	char tmpbuf[TMPBUFLEN];
1228
1229	if (!task)
1230		return -ESRCH;
1231	length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1232			   from_kuid(file->f_cred->user_ns,
1233				     audit_get_loginuid(task)));
1234	put_task_struct(task);
1235	return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1236}
1237
1238static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1239				   size_t count, loff_t *ppos)
1240{
1241	struct inode * inode = file_inode(file);
1242	uid_t loginuid;
1243	kuid_t kloginuid;
1244	int rv;
1245
1246	rcu_read_lock();
1247	if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1248		rcu_read_unlock();
1249		return -EPERM;
1250	}
1251	rcu_read_unlock();
1252
1253	if (*ppos != 0) {
1254		/* No partial writes. */
1255		return -EINVAL;
1256	}
1257
1258	rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1259	if (rv < 0)
1260		return rv;
1261
1262	/* is userspace tring to explicitly UNSET the loginuid? */
1263	if (loginuid == AUDIT_UID_UNSET) {
1264		kloginuid = INVALID_UID;
1265	} else {
1266		kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1267		if (!uid_valid(kloginuid))
1268			return -EINVAL;
1269	}
1270
1271	rv = audit_set_loginuid(kloginuid);
1272	if (rv < 0)
1273		return rv;
1274	return count;
1275}
1276
1277static const struct file_operations proc_loginuid_operations = {
1278	.read		= proc_loginuid_read,
1279	.write		= proc_loginuid_write,
1280	.llseek		= generic_file_llseek,
1281};
1282
1283static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1284				  size_t count, loff_t *ppos)
1285{
1286	struct inode * inode = file_inode(file);
1287	struct task_struct *task = get_proc_task(inode);
1288	ssize_t length;
1289	char tmpbuf[TMPBUFLEN];
1290
1291	if (!task)
1292		return -ESRCH;
1293	length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1294				audit_get_sessionid(task));
1295	put_task_struct(task);
1296	return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1297}
1298
1299static const struct file_operations proc_sessionid_operations = {
1300	.read		= proc_sessionid_read,
1301	.llseek		= generic_file_llseek,
1302};
1303#endif
1304
1305#ifdef CONFIG_FAULT_INJECTION
1306static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1307				      size_t count, loff_t *ppos)
1308{
1309	struct task_struct *task = get_proc_task(file_inode(file));
1310	char buffer[PROC_NUMBUF];
1311	size_t len;
1312	int make_it_fail;
1313
1314	if (!task)
1315		return -ESRCH;
1316	make_it_fail = task->make_it_fail;
1317	put_task_struct(task);
1318
1319	len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1320
1321	return simple_read_from_buffer(buf, count, ppos, buffer, len);
1322}
1323
1324static ssize_t proc_fault_inject_write(struct file * file,
1325			const char __user * buf, size_t count, loff_t *ppos)
1326{
1327	struct task_struct *task;
1328	char buffer[PROC_NUMBUF];
1329	int make_it_fail;
1330	int rv;
1331
1332	if (!capable(CAP_SYS_RESOURCE))
1333		return -EPERM;
1334	memset(buffer, 0, sizeof(buffer));
1335	if (count > sizeof(buffer) - 1)
1336		count = sizeof(buffer) - 1;
1337	if (copy_from_user(buffer, buf, count))
1338		return -EFAULT;
1339	rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1340	if (rv < 0)
1341		return rv;
1342	if (make_it_fail < 0 || make_it_fail > 1)
1343		return -EINVAL;
1344
1345	task = get_proc_task(file_inode(file));
1346	if (!task)
1347		return -ESRCH;
1348	task->make_it_fail = make_it_fail;
1349	put_task_struct(task);
1350
1351	return count;
1352}
1353
1354static const struct file_operations proc_fault_inject_operations = {
1355	.read		= proc_fault_inject_read,
1356	.write		= proc_fault_inject_write,
1357	.llseek		= generic_file_llseek,
1358};
1359#endif
1360
1361
1362#ifdef CONFIG_SCHED_DEBUG
1363/*
1364 * Print out various scheduling related per-task fields:
1365 */
1366static int sched_show(struct seq_file *m, void *v)
1367{
1368	struct inode *inode = m->private;
1369	struct task_struct *p;
1370
1371	p = get_proc_task(inode);
1372	if (!p)
1373		return -ESRCH;
1374	proc_sched_show_task(p, m);
1375
1376	put_task_struct(p);
1377
1378	return 0;
1379}
1380
1381static ssize_t
1382sched_write(struct file *file, const char __user *buf,
1383	    size_t count, loff_t *offset)
1384{
1385	struct inode *inode = file_inode(file);
1386	struct task_struct *p;
1387
1388	p = get_proc_task(inode);
1389	if (!p)
1390		return -ESRCH;
1391	proc_sched_set_task(p);
1392
1393	put_task_struct(p);
1394
1395	return count;
1396}
1397
1398static int sched_open(struct inode *inode, struct file *filp)
1399{
1400	return single_open(filp, sched_show, inode);
1401}
1402
1403static const struct file_operations proc_pid_sched_operations = {
1404	.open		= sched_open,
1405	.read		= seq_read,
1406	.write		= sched_write,
1407	.llseek		= seq_lseek,
1408	.release	= single_release,
1409};
1410
1411#endif
1412
1413#ifdef CONFIG_SCHED_AUTOGROUP
1414/*
1415 * Print out autogroup related information:
1416 */
1417static int sched_autogroup_show(struct seq_file *m, void *v)
1418{
1419	struct inode *inode = m->private;
1420	struct task_struct *p;
1421
1422	p = get_proc_task(inode);
1423	if (!p)
1424		return -ESRCH;
1425	proc_sched_autogroup_show_task(p, m);
1426
1427	put_task_struct(p);
1428
1429	return 0;
1430}
1431
1432static ssize_t
1433sched_autogroup_write(struct file *file, const char __user *buf,
1434	    size_t count, loff_t *offset)
1435{
1436	struct inode *inode = file_inode(file);
1437	struct task_struct *p;
1438	char buffer[PROC_NUMBUF];
1439	int nice;
1440	int err;
1441
1442	memset(buffer, 0, sizeof(buffer));
1443	if (count > sizeof(buffer) - 1)
1444		count = sizeof(buffer) - 1;
1445	if (copy_from_user(buffer, buf, count))
1446		return -EFAULT;
1447
1448	err = kstrtoint(strstrip(buffer), 0, &nice);
1449	if (err < 0)
1450		return err;
1451
1452	p = get_proc_task(inode);
1453	if (!p)
1454		return -ESRCH;
1455
1456	err = proc_sched_autogroup_set_nice(p, nice);
1457	if (err)
1458		count = err;
1459
1460	put_task_struct(p);
1461
1462	return count;
1463}
1464
1465static int sched_autogroup_open(struct inode *inode, struct file *filp)
1466{
1467	int ret;
1468
1469	ret = single_open(filp, sched_autogroup_show, NULL);
1470	if (!ret) {
1471		struct seq_file *m = filp->private_data;
1472
1473		m->private = inode;
1474	}
1475	return ret;
1476}
1477
1478static const struct file_operations proc_pid_sched_autogroup_operations = {
1479	.open		= sched_autogroup_open,
1480	.read		= seq_read,
1481	.write		= sched_autogroup_write,
1482	.llseek		= seq_lseek,
1483	.release	= single_release,
1484};
1485
1486#endif /* CONFIG_SCHED_AUTOGROUP */
1487
1488static ssize_t comm_write(struct file *file, const char __user *buf,
1489				size_t count, loff_t *offset)
1490{
1491	struct inode *inode = file_inode(file);
1492	struct task_struct *p;
1493	char buffer[TASK_COMM_LEN];
1494	const size_t maxlen = sizeof(buffer) - 1;
1495
1496	memset(buffer, 0, sizeof(buffer));
1497	if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1498		return -EFAULT;
1499
1500	p = get_proc_task(inode);
1501	if (!p)
1502		return -ESRCH;
1503
1504	if (same_thread_group(current, p))
1505		set_task_comm(p, buffer);
1506	else
1507		count = -EINVAL;
1508
1509	put_task_struct(p);
1510
1511	return count;
1512}
1513
1514static int comm_show(struct seq_file *m, void *v)
1515{
1516	struct inode *inode = m->private;
1517	struct task_struct *p;
1518
1519	p = get_proc_task(inode);
1520	if (!p)
1521		return -ESRCH;
1522
1523	task_lock(p);
1524	seq_printf(m, "%s\n", p->comm);
1525	task_unlock(p);
1526
1527	put_task_struct(p);
1528
1529	return 0;
1530}
1531
1532static int comm_open(struct inode *inode, struct file *filp)
1533{
1534	return single_open(filp, comm_show, inode);
1535}
1536
1537static const struct file_operations proc_pid_set_comm_operations = {
1538	.open		= comm_open,
1539	.read		= seq_read,
1540	.write		= comm_write,
1541	.llseek		= seq_lseek,
1542	.release	= single_release,
1543};
1544
1545static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1546{
1547	struct task_struct *task;
1548	struct mm_struct *mm;
1549	struct file *exe_file;
1550
1551	task = get_proc_task(d_inode(dentry));
1552	if (!task)
1553		return -ENOENT;
1554	mm = get_task_mm(task);
1555	put_task_struct(task);
1556	if (!mm)
1557		return -ENOENT;
1558	exe_file = get_mm_exe_file(mm);
1559	mmput(mm);
1560	if (exe_file) {
1561		*exe_path = exe_file->f_path;
1562		path_get(&exe_file->f_path);
1563		fput(exe_file);
1564		return 0;
1565	} else
1566		return -ENOENT;
1567}
1568
1569static const char *proc_pid_follow_link(struct dentry *dentry, void **cookie)
1570{
1571	struct inode *inode = d_inode(dentry);
1572	struct path path;
1573	int error = -EACCES;
1574
1575	/* Are we allowed to snoop on the tasks file descriptors? */
1576	if (!proc_fd_access_allowed(inode))
1577		goto out;
1578
1579	error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1580	if (error)
1581		goto out;
1582
1583	nd_jump_link(&path);
1584	return NULL;
1585out:
1586	return ERR_PTR(error);
1587}
1588
1589static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1590{
1591	char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1592	char *pathname;
1593	int len;
1594
1595	if (!tmp)
1596		return -ENOMEM;
1597
1598	pathname = d_path(path, tmp, PAGE_SIZE);
1599	len = PTR_ERR(pathname);
1600	if (IS_ERR(pathname))
1601		goto out;
1602	len = tmp + PAGE_SIZE - 1 - pathname;
1603
1604	if (len > buflen)
1605		len = buflen;
1606	if (copy_to_user(buffer, pathname, len))
1607		len = -EFAULT;
1608 out:
1609	free_page((unsigned long)tmp);
1610	return len;
1611}
1612
1613static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1614{
1615	int error = -EACCES;
1616	struct inode *inode = d_inode(dentry);
1617	struct path path;
1618
1619	/* Are we allowed to snoop on the tasks file descriptors? */
1620	if (!proc_fd_access_allowed(inode))
1621		goto out;
1622
1623	error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1624	if (error)
1625		goto out;
1626
1627	error = do_proc_readlink(&path, buffer, buflen);
1628	path_put(&path);
1629out:
1630	return error;
1631}
1632
1633const struct inode_operations proc_pid_link_inode_operations = {
1634	.readlink	= proc_pid_readlink,
1635	.follow_link	= proc_pid_follow_link,
1636	.setattr	= proc_setattr,
1637};
1638
1639
1640/* building an inode */
1641
1642struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1643{
1644	struct inode * inode;
1645	struct proc_inode *ei;
1646	const struct cred *cred;
1647
1648	/* We need a new inode */
1649
1650	inode = new_inode(sb);
1651	if (!inode)
1652		goto out;
1653
1654	/* Common stuff */
1655	ei = PROC_I(inode);
1656	inode->i_ino = get_next_ino();
1657	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1658	inode->i_op = &proc_def_inode_operations;
1659
1660	/*
1661	 * grab the reference to task.
1662	 */
1663	ei->pid = get_task_pid(task, PIDTYPE_PID);
1664	if (!ei->pid)
1665		goto out_unlock;
1666
1667	if (task_dumpable(task)) {
1668		rcu_read_lock();
1669		cred = __task_cred(task);
1670		inode->i_uid = cred->euid;
1671		inode->i_gid = cred->egid;
1672		rcu_read_unlock();
1673	}
1674	security_task_to_inode(task, inode);
1675
1676out:
1677	return inode;
1678
1679out_unlock:
1680	iput(inode);
1681	return NULL;
1682}
1683
1684int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1685{
1686	struct inode *inode = d_inode(dentry);
1687	struct task_struct *task;
1688	const struct cred *cred;
1689	struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1690
1691	generic_fillattr(inode, stat);
1692
1693	rcu_read_lock();
1694	stat->uid = GLOBAL_ROOT_UID;
1695	stat->gid = GLOBAL_ROOT_GID;
1696	task = pid_task(proc_pid(inode), PIDTYPE_PID);
1697	if (task) {
1698		if (!has_pid_permissions(pid, task, 2)) {
1699			rcu_read_unlock();
1700			/*
1701			 * This doesn't prevent learning whether PID exists,
1702			 * it only makes getattr() consistent with readdir().
1703			 */
1704			return -ENOENT;
1705		}
1706		if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1707		    task_dumpable(task)) {
1708			cred = __task_cred(task);
1709			stat->uid = cred->euid;
1710			stat->gid = cred->egid;
1711		}
1712	}
1713	rcu_read_unlock();
1714	return 0;
1715}
1716
1717/* dentry stuff */
1718
1719/*
1720 *	Exceptional case: normally we are not allowed to unhash a busy
1721 * directory. In this case, however, we can do it - no aliasing problems
1722 * due to the way we treat inodes.
1723 *
1724 * Rewrite the inode's ownerships here because the owning task may have
1725 * performed a setuid(), etc.
1726 *
1727 * Before the /proc/pid/status file was created the only way to read
1728 * the effective uid of a /process was to stat /proc/pid.  Reading
1729 * /proc/pid/status is slow enough that procps and other packages
1730 * kept stating /proc/pid.  To keep the rules in /proc simple I have
1731 * made this apply to all per process world readable and executable
1732 * directories.
1733 */
1734int pid_revalidate(struct dentry *dentry, unsigned int flags)
1735{
1736	struct inode *inode;
1737	struct task_struct *task;
1738	const struct cred *cred;
1739
1740	if (flags & LOOKUP_RCU)
1741		return -ECHILD;
1742
1743	inode = d_inode(dentry);
1744	task = get_proc_task(inode);
1745
1746	if (task) {
1747		if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1748		    task_dumpable(task)) {
1749			rcu_read_lock();
1750			cred = __task_cred(task);
1751			inode->i_uid = cred->euid;
1752			inode->i_gid = cred->egid;
1753			rcu_read_unlock();
1754		} else {
1755			inode->i_uid = GLOBAL_ROOT_UID;
1756			inode->i_gid = GLOBAL_ROOT_GID;
1757		}
1758		inode->i_mode &= ~(S_ISUID | S_ISGID);
1759		security_task_to_inode(task, inode);
1760		put_task_struct(task);
1761		return 1;
1762	}
1763	return 0;
1764}
1765
1766static inline bool proc_inode_is_dead(struct inode *inode)
1767{
1768	return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1769}
1770
1771int pid_delete_dentry(const struct dentry *dentry)
1772{
1773	/* Is the task we represent dead?
1774	 * If so, then don't put the dentry on the lru list,
1775	 * kill it immediately.
1776	 */
1777	return proc_inode_is_dead(d_inode(dentry));
1778}
1779
1780const struct dentry_operations pid_dentry_operations =
1781{
1782	.d_revalidate	= pid_revalidate,
1783	.d_delete	= pid_delete_dentry,
1784};
1785
1786/* Lookups */
1787
1788/*
1789 * Fill a directory entry.
1790 *
1791 * If possible create the dcache entry and derive our inode number and
1792 * file type from dcache entry.
1793 *
1794 * Since all of the proc inode numbers are dynamically generated, the inode
1795 * numbers do not exist until the inode is cache.  This means creating the
1796 * the dcache entry in readdir is necessary to keep the inode numbers
1797 * reported by readdir in sync with the inode numbers reported
1798 * by stat.
1799 */
1800bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1801	const char *name, int len,
1802	instantiate_t instantiate, struct task_struct *task, const void *ptr)
1803{
1804	struct dentry *child, *dir = file->f_path.dentry;
1805	struct qstr qname = QSTR_INIT(name, len);
1806	struct inode *inode;
1807	unsigned type;
1808	ino_t ino;
1809
1810	child = d_hash_and_lookup(dir, &qname);
1811	if (!child) {
1812		child = d_alloc(dir, &qname);
1813		if (!child)
1814			goto end_instantiate;
1815		if (instantiate(d_inode(dir), child, task, ptr) < 0) {
1816			dput(child);
1817			goto end_instantiate;
1818		}
1819	}
1820	inode = d_inode(child);
1821	ino = inode->i_ino;
1822	type = inode->i_mode >> 12;
1823	dput(child);
1824	return dir_emit(ctx, name, len, ino, type);
1825
1826end_instantiate:
1827	return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1828}
1829
1830/*
1831 * dname_to_vma_addr - maps a dentry name into two unsigned longs
1832 * which represent vma start and end addresses.
1833 */
1834static int dname_to_vma_addr(struct dentry *dentry,
1835			     unsigned long *start, unsigned long *end)
1836{
1837	if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1838		return -EINVAL;
1839
1840	return 0;
1841}
1842
1843static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1844{
1845	unsigned long vm_start, vm_end;
1846	bool exact_vma_exists = false;
1847	struct mm_struct *mm = NULL;
1848	struct task_struct *task;
1849	const struct cred *cred;
1850	struct inode *inode;
1851	int status = 0;
1852
1853	if (flags & LOOKUP_RCU)
1854		return -ECHILD;
1855
1856	inode = d_inode(dentry);
1857	task = get_proc_task(inode);
1858	if (!task)
1859		goto out_notask;
1860
1861	mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1862	if (IS_ERR_OR_NULL(mm))
1863		goto out;
1864
1865	if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1866		down_read(&mm->mmap_sem);
1867		exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1868		up_read(&mm->mmap_sem);
1869	}
1870
1871	mmput(mm);
1872
1873	if (exact_vma_exists) {
1874		if (task_dumpable(task)) {
1875			rcu_read_lock();
1876			cred = __task_cred(task);
1877			inode->i_uid = cred->euid;
1878			inode->i_gid = cred->egid;
1879			rcu_read_unlock();
1880		} else {
1881			inode->i_uid = GLOBAL_ROOT_UID;
1882			inode->i_gid = GLOBAL_ROOT_GID;
1883		}
1884		security_task_to_inode(task, inode);
1885		status = 1;
1886	}
1887
1888out:
1889	put_task_struct(task);
1890
1891out_notask:
1892	return status;
1893}
1894
1895static const struct dentry_operations tid_map_files_dentry_operations = {
1896	.d_revalidate	= map_files_d_revalidate,
1897	.d_delete	= pid_delete_dentry,
1898};
1899
1900static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
1901{
1902	unsigned long vm_start, vm_end;
1903	struct vm_area_struct *vma;
1904	struct task_struct *task;
1905	struct mm_struct *mm;
1906	int rc;
1907
1908	rc = -ENOENT;
1909	task = get_proc_task(d_inode(dentry));
1910	if (!task)
1911		goto out;
1912
1913	mm = get_task_mm(task);
1914	put_task_struct(task);
1915	if (!mm)
1916		goto out;
1917
1918	rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1919	if (rc)
1920		goto out_mmput;
1921
1922	rc = -ENOENT;
1923	down_read(&mm->mmap_sem);
1924	vma = find_exact_vma(mm, vm_start, vm_end);
1925	if (vma && vma->vm_file) {
1926		*path = vma->vm_file->f_path;
1927		path_get(path);
1928		rc = 0;
1929	}
1930	up_read(&mm->mmap_sem);
1931
1932out_mmput:
1933	mmput(mm);
1934out:
1935	return rc;
1936}
1937
1938struct map_files_info {
1939	fmode_t		mode;
1940	unsigned long	len;
1941	unsigned char	name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1942};
1943
1944/*
1945 * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
1946 * symlinks may be used to bypass permissions on ancestor directories in the
1947 * path to the file in question.
1948 */
1949static const char *
1950proc_map_files_follow_link(struct dentry *dentry, void **cookie)
1951{
1952	if (!capable(CAP_SYS_ADMIN))
1953		return ERR_PTR(-EPERM);
1954
1955	return proc_pid_follow_link(dentry, NULL);
1956}
1957
1958/*
1959 * Identical to proc_pid_link_inode_operations except for follow_link()
1960 */
1961static const struct inode_operations proc_map_files_link_inode_operations = {
1962	.readlink	= proc_pid_readlink,
1963	.follow_link	= proc_map_files_follow_link,
1964	.setattr	= proc_setattr,
1965};
1966
1967static int
1968proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
1969			   struct task_struct *task, const void *ptr)
1970{
1971	fmode_t mode = (fmode_t)(unsigned long)ptr;
1972	struct proc_inode *ei;
1973	struct inode *inode;
1974
1975	inode = proc_pid_make_inode(dir->i_sb, task);
1976	if (!inode)
1977		return -ENOENT;
1978
1979	ei = PROC_I(inode);
1980	ei->op.proc_get_link = proc_map_files_get_link;
1981
1982	inode->i_op = &proc_map_files_link_inode_operations;
1983	inode->i_size = 64;
1984	inode->i_mode = S_IFLNK;
1985
1986	if (mode & FMODE_READ)
1987		inode->i_mode |= S_IRUSR;
1988	if (mode & FMODE_WRITE)
1989		inode->i_mode |= S_IWUSR;
1990
1991	d_set_d_op(dentry, &tid_map_files_dentry_operations);
1992	d_add(dentry, inode);
1993
1994	return 0;
1995}
1996
1997static struct dentry *proc_map_files_lookup(struct inode *dir,
1998		struct dentry *dentry, unsigned int flags)
1999{
2000	unsigned long vm_start, vm_end;
2001	struct vm_area_struct *vma;
2002	struct task_struct *task;
2003	int result;
2004	struct mm_struct *mm;
2005
2006	result = -ENOENT;
2007	task = get_proc_task(dir);
2008	if (!task)
2009		goto out;
2010
2011	result = -EACCES;
2012	if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2013		goto out_put_task;
2014
2015	result = -ENOENT;
2016	if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2017		goto out_put_task;
2018
2019	mm = get_task_mm(task);
2020	if (!mm)
2021		goto out_put_task;
2022
2023	down_read(&mm->mmap_sem);
2024	vma = find_exact_vma(mm, vm_start, vm_end);
2025	if (!vma)
2026		goto out_no_vma;
2027
2028	if (vma->vm_file)
2029		result = proc_map_files_instantiate(dir, dentry, task,
2030				(void *)(unsigned long)vma->vm_file->f_mode);
2031
2032out_no_vma:
2033	up_read(&mm->mmap_sem);
2034	mmput(mm);
2035out_put_task:
2036	put_task_struct(task);
2037out:
2038	return ERR_PTR(result);
2039}
2040
2041static const struct inode_operations proc_map_files_inode_operations = {
2042	.lookup		= proc_map_files_lookup,
2043	.permission	= proc_fd_permission,
2044	.setattr	= proc_setattr,
2045};
2046
2047static int
2048proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2049{
2050	struct vm_area_struct *vma;
2051	struct task_struct *task;
2052	struct mm_struct *mm;
2053	unsigned long nr_files, pos, i;
2054	struct flex_array *fa = NULL;
2055	struct map_files_info info;
2056	struct map_files_info *p;
2057	int ret;
2058
2059	ret = -ENOENT;
2060	task = get_proc_task(file_inode(file));
2061	if (!task)
2062		goto out;
2063
2064	ret = -EACCES;
2065	if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2066		goto out_put_task;
2067
2068	ret = 0;
2069	if (!dir_emit_dots(file, ctx))
2070		goto out_put_task;
2071
2072	mm = get_task_mm(task);
2073	if (!mm)
2074		goto out_put_task;
2075	down_read(&mm->mmap_sem);
2076
2077	nr_files = 0;
2078
2079	/*
2080	 * We need two passes here:
2081	 *
2082	 *  1) Collect vmas of mapped files with mmap_sem taken
2083	 *  2) Release mmap_sem and instantiate entries
2084	 *
2085	 * otherwise we get lockdep complained, since filldir()
2086	 * routine might require mmap_sem taken in might_fault().
2087	 */
2088
2089	for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2090		if (vma->vm_file && ++pos > ctx->pos)
2091			nr_files++;
2092	}
2093
2094	if (nr_files) {
2095		fa = flex_array_alloc(sizeof(info), nr_files,
2096					GFP_KERNEL);
2097		if (!fa || flex_array_prealloc(fa, 0, nr_files,
2098						GFP_KERNEL)) {
2099			ret = -ENOMEM;
2100			if (fa)
2101				flex_array_free(fa);
2102			up_read(&mm->mmap_sem);
2103			mmput(mm);
2104			goto out_put_task;
2105		}
2106		for (i = 0, vma = mm->mmap, pos = 2; vma;
2107				vma = vma->vm_next) {
2108			if (!vma->vm_file)
2109				continue;
2110			if (++pos <= ctx->pos)
2111				continue;
2112
2113			info.mode = vma->vm_file->f_mode;
2114			info.len = snprintf(info.name,
2115					sizeof(info.name), "%lx-%lx",
2116					vma->vm_start, vma->vm_end);
2117			if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2118				BUG();
2119		}
2120	}
2121	up_read(&mm->mmap_sem);
2122
2123	for (i = 0; i < nr_files; i++) {
2124		p = flex_array_get(fa, i);
2125		if (!proc_fill_cache(file, ctx,
2126				      p->name, p->len,
2127				      proc_map_files_instantiate,
2128				      task,
2129				      (void *)(unsigned long)p->mode))
2130			break;
2131		ctx->pos++;
2132	}
2133	if (fa)
2134		flex_array_free(fa);
2135	mmput(mm);
2136
2137out_put_task:
2138	put_task_struct(task);
2139out:
2140	return ret;
2141}
2142
2143static const struct file_operations proc_map_files_operations = {
2144	.read		= generic_read_dir,
2145	.iterate	= proc_map_files_readdir,
2146	.llseek		= default_llseek,
2147};
2148
2149struct timers_private {
2150	struct pid *pid;
2151	struct task_struct *task;
2152	struct sighand_struct *sighand;
2153	struct pid_namespace *ns;
2154	unsigned long flags;
2155};
2156
2157static void *timers_start(struct seq_file *m, loff_t *pos)
2158{
2159	struct timers_private *tp = m->private;
2160
2161	tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2162	if (!tp->task)
2163		return ERR_PTR(-ESRCH);
2164
2165	tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2166	if (!tp->sighand)
2167		return ERR_PTR(-ESRCH);
2168
2169	return seq_list_start(&tp->task->signal->posix_timers, *pos);
2170}
2171
2172static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2173{
2174	struct timers_private *tp = m->private;
2175	return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2176}
2177
2178static void timers_stop(struct seq_file *m, void *v)
2179{
2180	struct timers_private *tp = m->private;
2181
2182	if (tp->sighand) {
2183		unlock_task_sighand(tp->task, &tp->flags);
2184		tp->sighand = NULL;
2185	}
2186
2187	if (tp->task) {
2188		put_task_struct(tp->task);
2189		tp->task = NULL;
2190	}
2191}
2192
2193static int show_timer(struct seq_file *m, void *v)
2194{
2195	struct k_itimer *timer;
2196	struct timers_private *tp = m->private;
2197	int notify;
2198	static const char * const nstr[] = {
2199		[SIGEV_SIGNAL] = "signal",
2200		[SIGEV_NONE] = "none",
2201		[SIGEV_THREAD] = "thread",
2202	};
2203
2204	timer = list_entry((struct list_head *)v, struct k_itimer, list);
2205	notify = timer->it_sigev_notify;
2206
2207	seq_printf(m, "ID: %d\n", timer->it_id);
2208	seq_printf(m, "signal: %d/%p\n",
2209		   timer->sigq->info.si_signo,
2210		   timer->sigq->info.si_value.sival_ptr);
2211	seq_printf(m, "notify: %s/%s.%d\n",
2212		   nstr[notify & ~SIGEV_THREAD_ID],
2213		   (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2214		   pid_nr_ns(timer->it_pid, tp->ns));
2215	seq_printf(m, "ClockID: %d\n", timer->it_clock);
2216
2217	return 0;
2218}
2219
2220static const struct seq_operations proc_timers_seq_ops = {
2221	.start	= timers_start,
2222	.next	= timers_next,
2223	.stop	= timers_stop,
2224	.show	= show_timer,
2225};
2226
2227static int proc_timers_open(struct inode *inode, struct file *file)
2228{
2229	struct timers_private *tp;
2230
2231	tp = __seq_open_private(file, &proc_timers_seq_ops,
2232			sizeof(struct timers_private));
2233	if (!tp)
2234		return -ENOMEM;
2235
2236	tp->pid = proc_pid(inode);
2237	tp->ns = inode->i_sb->s_fs_info;
2238	return 0;
2239}
2240
2241static const struct file_operations proc_timers_operations = {
2242	.open		= proc_timers_open,
2243	.read		= seq_read,
2244	.llseek		= seq_lseek,
2245	.release	= seq_release_private,
2246};
2247
2248static int proc_pident_instantiate(struct inode *dir,
2249	struct dentry *dentry, struct task_struct *task, const void *ptr)
2250{
2251	const struct pid_entry *p = ptr;
2252	struct inode *inode;
2253	struct proc_inode *ei;
2254
2255	inode = proc_pid_make_inode(dir->i_sb, task);
2256	if (!inode)
2257		goto out;
2258
2259	ei = PROC_I(inode);
2260	inode->i_mode = p->mode;
2261	if (S_ISDIR(inode->i_mode))
2262		set_nlink(inode, 2);	/* Use getattr to fix if necessary */
2263	if (p->iop)
2264		inode->i_op = p->iop;
2265	if (p->fop)
2266		inode->i_fop = p->fop;
2267	ei->op = p->op;
2268	d_set_d_op(dentry, &pid_dentry_operations);
2269	d_add(dentry, inode);
2270	/* Close the race of the process dying before we return the dentry */
2271	if (pid_revalidate(dentry, 0))
2272		return 0;
2273out:
2274	return -ENOENT;
2275}
2276
2277static struct dentry *proc_pident_lookup(struct inode *dir,
2278					 struct dentry *dentry,
2279					 const struct pid_entry *ents,
2280					 unsigned int nents)
2281{
2282	int error;
2283	struct task_struct *task = get_proc_task(dir);
2284	const struct pid_entry *p, *last;
2285
2286	error = -ENOENT;
2287
2288	if (!task)
2289		goto out_no_task;
2290
2291	/*
2292	 * Yes, it does not scale. And it should not. Don't add
2293	 * new entries into /proc/<tgid>/ without very good reasons.
2294	 */
2295	last = &ents[nents - 1];
2296	for (p = ents; p <= last; p++) {
2297		if (p->len != dentry->d_name.len)
2298			continue;
2299		if (!memcmp(dentry->d_name.name, p->name, p->len))
2300			break;
2301	}
2302	if (p > last)
2303		goto out;
2304
2305	error = proc_pident_instantiate(dir, dentry, task, p);
2306out:
2307	put_task_struct(task);
2308out_no_task:
2309	return ERR_PTR(error);
2310}
2311
2312static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2313		const struct pid_entry *ents, unsigned int nents)
2314{
2315	struct task_struct *task = get_proc_task(file_inode(file));
2316	const struct pid_entry *p;
2317
2318	if (!task)
2319		return -ENOENT;
2320
2321	if (!dir_emit_dots(file, ctx))
2322		goto out;
2323
2324	if (ctx->pos >= nents + 2)
2325		goto out;
2326
2327	for (p = ents + (ctx->pos - 2); p <= ents + nents - 1; p++) {
2328		if (!proc_fill_cache(file, ctx, p->name, p->len,
2329				proc_pident_instantiate, task, p))
2330			break;
2331		ctx->pos++;
2332	}
2333out:
2334	put_task_struct(task);
2335	return 0;
2336}
2337
2338#ifdef CONFIG_SECURITY
2339static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2340				  size_t count, loff_t *ppos)
2341{
2342	struct inode * inode = file_inode(file);
2343	char *p = NULL;
2344	ssize_t length;
2345	struct task_struct *task = get_proc_task(inode);
2346
2347	if (!task)
2348		return -ESRCH;
2349
2350	length = security_getprocattr(task,
2351				      (char*)file->f_path.dentry->d_name.name,
2352				      &p);
2353	put_task_struct(task);
2354	if (length > 0)
2355		length = simple_read_from_buffer(buf, count, ppos, p, length);
2356	kfree(p);
2357	return length;
2358}
2359
2360static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2361				   size_t count, loff_t *ppos)
2362{
2363	struct inode * inode = file_inode(file);
2364	char *page;
2365	ssize_t length;
2366	struct task_struct *task = get_proc_task(inode);
2367
2368	length = -ESRCH;
2369	if (!task)
2370		goto out_no_task;
2371	if (count > PAGE_SIZE)
2372		count = PAGE_SIZE;
2373
2374	/* No partial writes. */
2375	length = -EINVAL;
2376	if (*ppos != 0)
2377		goto out;
2378
2379	length = -ENOMEM;
2380	page = (char*)__get_free_page(GFP_TEMPORARY);
2381	if (!page)
2382		goto out;
2383
2384	length = -EFAULT;
2385	if (copy_from_user(page, buf, count))
2386		goto out_free;
2387
2388	/* Guard against adverse ptrace interaction */
2389	length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2390	if (length < 0)
2391		goto out_free;
2392
2393	length = security_setprocattr(task,
2394				      (char*)file->f_path.dentry->d_name.name,
2395				      (void*)page, count);
2396	mutex_unlock(&task->signal->cred_guard_mutex);
2397out_free:
2398	free_page((unsigned long) page);
2399out:
2400	put_task_struct(task);
2401out_no_task:
2402	return length;
2403}
2404
2405static const struct file_operations proc_pid_attr_operations = {
2406	.read		= proc_pid_attr_read,
2407	.write		= proc_pid_attr_write,
2408	.llseek		= generic_file_llseek,
2409};
2410
2411static const struct pid_entry attr_dir_stuff[] = {
2412	REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2413	REG("prev",       S_IRUGO,	   proc_pid_attr_operations),
2414	REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2415	REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2416	REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2417	REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2418};
2419
2420static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2421{
2422	return proc_pident_readdir(file, ctx,
2423				   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2424}
2425
2426static const struct file_operations proc_attr_dir_operations = {
2427	.read		= generic_read_dir,
2428	.iterate	= proc_attr_dir_readdir,
2429	.llseek		= default_llseek,
2430};
2431
2432static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2433				struct dentry *dentry, unsigned int flags)
2434{
2435	return proc_pident_lookup(dir, dentry,
2436				  attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2437}
2438
2439static const struct inode_operations proc_attr_dir_inode_operations = {
2440	.lookup		= proc_attr_dir_lookup,
2441	.getattr	= pid_getattr,
2442	.setattr	= proc_setattr,
2443};
2444
2445#endif
2446
2447#ifdef CONFIG_ELF_CORE
2448static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2449					 size_t count, loff_t *ppos)
2450{
2451	struct task_struct *task = get_proc_task(file_inode(file));
2452	struct mm_struct *mm;
2453	char buffer[PROC_NUMBUF];
2454	size_t len;
2455	int ret;
2456
2457	if (!task)
2458		return -ESRCH;
2459
2460	ret = 0;
2461	mm = get_task_mm(task);
2462	if (mm) {
2463		len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2464			       ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2465				MMF_DUMP_FILTER_SHIFT));
2466		mmput(mm);
2467		ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2468	}
2469
2470	put_task_struct(task);
2471
2472	return ret;
2473}
2474
2475static ssize_t proc_coredump_filter_write(struct file *file,
2476					  const char __user *buf,
2477					  size_t count,
2478					  loff_t *ppos)
2479{
2480	struct task_struct *task;
2481	struct mm_struct *mm;
2482	unsigned int val;
2483	int ret;
2484	int i;
2485	unsigned long mask;
2486
2487	ret = kstrtouint_from_user(buf, count, 0, &val);
2488	if (ret < 0)
2489		return ret;
2490
2491	ret = -ESRCH;
2492	task = get_proc_task(file_inode(file));
2493	if (!task)
2494		goto out_no_task;
2495
2496	mm = get_task_mm(task);
2497	if (!mm)
2498		goto out_no_mm;
2499	ret = 0;
2500
2501	for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2502		if (val & mask)
2503			set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2504		else
2505			clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2506	}
2507
2508	mmput(mm);
2509 out_no_mm:
2510	put_task_struct(task);
2511 out_no_task:
2512	if (ret < 0)
2513		return ret;
2514	return count;
2515}
2516
2517static const struct file_operations proc_coredump_filter_operations = {
2518	.read		= proc_coredump_filter_read,
2519	.write		= proc_coredump_filter_write,
2520	.llseek		= generic_file_llseek,
2521};
2522#endif
2523
2524#ifdef CONFIG_TASK_IO_ACCOUNTING
2525static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2526{
2527	struct task_io_accounting acct = task->ioac;
2528	unsigned long flags;
2529	int result;
2530
2531	result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2532	if (result)
2533		return result;
2534
2535	if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2536		result = -EACCES;
2537		goto out_unlock;
2538	}
2539
2540	if (whole && lock_task_sighand(task, &flags)) {
2541		struct task_struct *t = task;
2542
2543		task_io_accounting_add(&acct, &task->signal->ioac);
2544		while_each_thread(task, t)
2545			task_io_accounting_add(&acct, &t->ioac);
2546
2547		unlock_task_sighand(task, &flags);
2548	}
2549	seq_printf(m,
2550		   "rchar: %llu\n"
2551		   "wchar: %llu\n"
2552		   "syscr: %llu\n"
2553		   "syscw: %llu\n"
2554		   "read_bytes: %llu\n"
2555		   "write_bytes: %llu\n"
2556		   "cancelled_write_bytes: %llu\n",
2557		   (unsigned long long)acct.rchar,
2558		   (unsigned long long)acct.wchar,
2559		   (unsigned long long)acct.syscr,
2560		   (unsigned long long)acct.syscw,
2561		   (unsigned long long)acct.read_bytes,
2562		   (unsigned long long)acct.write_bytes,
2563		   (unsigned long long)acct.cancelled_write_bytes);
2564	result = 0;
2565
2566out_unlock:
2567	mutex_unlock(&task->signal->cred_guard_mutex);
2568	return result;
2569}
2570
2571static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2572				  struct pid *pid, struct task_struct *task)
2573{
2574	return do_io_accounting(task, m, 0);
2575}
2576
2577static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2578				   struct pid *pid, struct task_struct *task)
2579{
2580	return do_io_accounting(task, m, 1);
2581}
2582#endif /* CONFIG_TASK_IO_ACCOUNTING */
2583
2584#ifdef CONFIG_USER_NS
2585static int proc_id_map_open(struct inode *inode, struct file *file,
2586	const struct seq_operations *seq_ops)
2587{
2588	struct user_namespace *ns = NULL;
2589	struct task_struct *task;
2590	struct seq_file *seq;
2591	int ret = -EINVAL;
2592
2593	task = get_proc_task(inode);
2594	if (task) {
2595		rcu_read_lock();
2596		ns = get_user_ns(task_cred_xxx(task, user_ns));
2597		rcu_read_unlock();
2598		put_task_struct(task);
2599	}
2600	if (!ns)
2601		goto err;
2602
2603	ret = seq_open(file, seq_ops);
2604	if (ret)
2605		goto err_put_ns;
2606
2607	seq = file->private_data;
2608	seq->private = ns;
2609
2610	return 0;
2611err_put_ns:
2612	put_user_ns(ns);
2613err:
2614	return ret;
2615}
2616
2617static int proc_id_map_release(struct inode *inode, struct file *file)
2618{
2619	struct seq_file *seq = file->private_data;
2620	struct user_namespace *ns = seq->private;
2621	put_user_ns(ns);
2622	return seq_release(inode, file);
2623}
2624
2625static int proc_uid_map_open(struct inode *inode, struct file *file)
2626{
2627	return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2628}
2629
2630static int proc_gid_map_open(struct inode *inode, struct file *file)
2631{
2632	return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2633}
2634
2635static int proc_projid_map_open(struct inode *inode, struct file *file)
2636{
2637	return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2638}
2639
2640static const struct file_operations proc_uid_map_operations = {
2641	.open		= proc_uid_map_open,
2642	.write		= proc_uid_map_write,
2643	.read		= seq_read,
2644	.llseek		= seq_lseek,
2645	.release	= proc_id_map_release,
2646};
2647
2648static const struct file_operations proc_gid_map_operations = {
2649	.open		= proc_gid_map_open,
2650	.write		= proc_gid_map_write,
2651	.read		= seq_read,
2652	.llseek		= seq_lseek,
2653	.release	= proc_id_map_release,
2654};
2655
2656static const struct file_operations proc_projid_map_operations = {
2657	.open		= proc_projid_map_open,
2658	.write		= proc_projid_map_write,
2659	.read		= seq_read,
2660	.llseek		= seq_lseek,
2661	.release	= proc_id_map_release,
2662};
2663
2664static int proc_setgroups_open(struct inode *inode, struct file *file)
2665{
2666	struct user_namespace *ns = NULL;
2667	struct task_struct *task;
2668	int ret;
2669
2670	ret = -ESRCH;
2671	task = get_proc_task(inode);
2672	if (task) {
2673		rcu_read_lock();
2674		ns = get_user_ns(task_cred_xxx(task, user_ns));
2675		rcu_read_unlock();
2676		put_task_struct(task);
2677	}
2678	if (!ns)
2679		goto err;
2680
2681	if (file->f_mode & FMODE_WRITE) {
2682		ret = -EACCES;
2683		if (!ns_capable(ns, CAP_SYS_ADMIN))
2684			goto err_put_ns;
2685	}
2686
2687	ret = single_open(file, &proc_setgroups_show, ns);
2688	if (ret)
2689		goto err_put_ns;
2690
2691	return 0;
2692err_put_ns:
2693	put_user_ns(ns);
2694err:
2695	return ret;
2696}
2697
2698static int proc_setgroups_release(struct inode *inode, struct file *file)
2699{
2700	struct seq_file *seq = file->private_data;
2701	struct user_namespace *ns = seq->private;
2702	int ret = single_release(inode, file);
2703	put_user_ns(ns);
2704	return ret;
2705}
2706
2707static const struct file_operations proc_setgroups_operations = {
2708	.open		= proc_setgroups_open,
2709	.write		= proc_setgroups_write,
2710	.read		= seq_read,
2711	.llseek		= seq_lseek,
2712	.release	= proc_setgroups_release,
2713};
2714#endif /* CONFIG_USER_NS */
2715
2716static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2717				struct pid *pid, struct task_struct *task)
2718{
2719	int err = lock_trace(task);
2720	if (!err) {
2721		seq_printf(m, "%08x\n", task->personality);
2722		unlock_trace(task);
2723	}
2724	return err;
2725}
2726
2727/*
2728 * Thread groups
2729 */
2730static const struct file_operations proc_task_operations;
2731static const struct inode_operations proc_task_inode_operations;
2732
2733static const struct pid_entry tgid_base_stuff[] = {
2734	DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2735	DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2736	DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2737	DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2738	DIR("ns",	  S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2739#ifdef CONFIG_NET
2740	DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2741#endif
2742	REG("environ",    S_IRUSR, proc_environ_operations),
2743	ONE("auxv",       S_IRUSR, proc_pid_auxv),
2744	ONE("status",     S_IRUGO, proc_pid_status),
2745	ONE("personality", S_IRUSR, proc_pid_personality),
2746	ONE("limits",	  S_IRUGO, proc_pid_limits),
2747#ifdef CONFIG_SCHED_DEBUG
2748	REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2749#endif
2750#ifdef CONFIG_SCHED_AUTOGROUP
2751	REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2752#endif
2753	REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2754#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2755	ONE("syscall",    S_IRUSR, proc_pid_syscall),
2756#endif
2757	REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
2758	ONE("stat",       S_IRUGO, proc_tgid_stat),
2759	ONE("statm",      S_IRUGO, proc_pid_statm),
2760	REG("maps",       S_IRUGO, proc_pid_maps_operations),
2761#ifdef CONFIG_NUMA
2762	REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2763#endif
2764	REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2765	LNK("cwd",        proc_cwd_link),
2766	LNK("root",       proc_root_link),
2767	LNK("exe",        proc_exe_link),
2768	REG("mounts",     S_IRUGO, proc_mounts_operations),
2769	REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2770	REG("mountstats", S_IRUSR, proc_mountstats_operations),
2771#ifdef CONFIG_PROC_PAGE_MONITOR
2772	REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2773	REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2774	REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2775#endif
2776#ifdef CONFIG_SECURITY
2777	DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2778#endif
2779#ifdef CONFIG_KALLSYMS
2780	ONE("wchan",      S_IRUGO, proc_pid_wchan),
2781#endif
2782#ifdef CONFIG_STACKTRACE
2783	ONE("stack",      S_IRUSR, proc_pid_stack),
2784#endif
2785#ifdef CONFIG_SCHED_INFO
2786	ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
2787#endif
2788#ifdef CONFIG_LATENCYTOP
2789	REG("latency",  S_IRUGO, proc_lstats_operations),
2790#endif
2791#ifdef CONFIG_PROC_PID_CPUSET
2792	ONE("cpuset",     S_IRUGO, proc_cpuset_show),
2793#endif
2794#ifdef CONFIG_CGROUPS
2795	ONE("cgroup",  S_IRUGO, proc_cgroup_show),
2796#endif
2797	ONE("oom_score",  S_IRUGO, proc_oom_score),
2798	REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2799	REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2800#ifdef CONFIG_AUDITSYSCALL
2801	REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2802	REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2803#endif
2804#ifdef CONFIG_FAULT_INJECTION
2805	REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2806#endif
2807#ifdef CONFIG_ELF_CORE
2808	REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2809#endif
2810#ifdef CONFIG_TASK_IO_ACCOUNTING
2811	ONE("io",	S_IRUSR, proc_tgid_io_accounting),
2812#endif
2813#ifdef CONFIG_HARDWALL
2814	ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
2815#endif
2816#ifdef CONFIG_USER_NS
2817	REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2818	REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2819	REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2820	REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
2821#endif
2822#ifdef CONFIG_CHECKPOINT_RESTORE
2823	REG("timers",	  S_IRUGO, proc_timers_operations),
2824#endif
2825};
2826
2827static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2828{
2829	return proc_pident_readdir(file, ctx,
2830				   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2831}
2832
2833static const struct file_operations proc_tgid_base_operations = {
2834	.read		= generic_read_dir,
2835	.iterate	= proc_tgid_base_readdir,
2836	.llseek		= default_llseek,
2837};
2838
2839static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2840{
2841	return proc_pident_lookup(dir, dentry,
2842				  tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2843}
2844
2845static const struct inode_operations proc_tgid_base_inode_operations = {
2846	.lookup		= proc_tgid_base_lookup,
2847	.getattr	= pid_getattr,
2848	.setattr	= proc_setattr,
2849	.permission	= proc_pid_permission,
2850};
2851
2852static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2853{
2854	struct dentry *dentry, *leader, *dir;
2855	char buf[PROC_NUMBUF];
2856	struct qstr name;
2857
2858	name.name = buf;
2859	name.len = snprintf(buf, sizeof(buf), "%d", pid);
2860	/* no ->d_hash() rejects on procfs */
2861	dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2862	if (dentry) {
2863		d_invalidate(dentry);
2864		dput(dentry);
2865	}
2866
2867	if (pid == tgid)
2868		return;
2869
2870	name.name = buf;
2871	name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2872	leader = d_hash_and_lookup(mnt->mnt_root, &name);
2873	if (!leader)
2874		goto out;
2875
2876	name.name = "task";
2877	name.len = strlen(name.name);
2878	dir = d_hash_and_lookup(leader, &name);
2879	if (!dir)
2880		goto out_put_leader;
2881
2882	name.name = buf;
2883	name.len = snprintf(buf, sizeof(buf), "%d", pid);
2884	dentry = d_hash_and_lookup(dir, &name);
2885	if (dentry) {
2886		d_invalidate(dentry);
2887		dput(dentry);
2888	}
2889
2890	dput(dir);
2891out_put_leader:
2892	dput(leader);
2893out:
2894	return;
2895}
2896
2897/**
2898 * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2899 * @task: task that should be flushed.
2900 *
2901 * When flushing dentries from proc, one needs to flush them from global
2902 * proc (proc_mnt) and from all the namespaces' procs this task was seen
2903 * in. This call is supposed to do all of this job.
2904 *
2905 * Looks in the dcache for
2906 * /proc/@pid
2907 * /proc/@tgid/task/@pid
2908 * if either directory is present flushes it and all of it'ts children
2909 * from the dcache.
2910 *
2911 * It is safe and reasonable to cache /proc entries for a task until
2912 * that task exits.  After that they just clog up the dcache with
2913 * useless entries, possibly causing useful dcache entries to be
2914 * flushed instead.  This routine is proved to flush those useless
2915 * dcache entries at process exit time.
2916 *
2917 * NOTE: This routine is just an optimization so it does not guarantee
2918 *       that no dcache entries will exist at process exit time it
2919 *       just makes it very unlikely that any will persist.
2920 */
2921
2922void proc_flush_task(struct task_struct *task)
2923{
2924	int i;
2925	struct pid *pid, *tgid;
2926	struct upid *upid;
2927
2928	pid = task_pid(task);
2929	tgid = task_tgid(task);
2930
2931	for (i = 0; i <= pid->level; i++) {
2932		upid = &pid->numbers[i];
2933		proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2934					tgid->numbers[i].nr);
2935	}
2936}
2937
2938static int proc_pid_instantiate(struct inode *dir,
2939				   struct dentry * dentry,
2940				   struct task_struct *task, const void *ptr)
2941{
2942	struct inode *inode;
2943
2944	inode = proc_pid_make_inode(dir->i_sb, task);
2945	if (!inode)
2946		goto out;
2947
2948	inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2949	inode->i_op = &proc_tgid_base_inode_operations;
2950	inode->i_fop = &proc_tgid_base_operations;
2951	inode->i_flags|=S_IMMUTABLE;
2952
2953	set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
2954						  ARRAY_SIZE(tgid_base_stuff)));
2955
2956	d_set_d_op(dentry, &pid_dentry_operations);
2957
2958	d_add(dentry, inode);
2959	/* Close the race of the process dying before we return the dentry */
2960	if (pid_revalidate(dentry, 0))
2961		return 0;
2962out:
2963	return -ENOENT;
2964}
2965
2966struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
2967{
2968	int result = -ENOENT;
2969	struct task_struct *task;
2970	unsigned tgid;
2971	struct pid_namespace *ns;
2972
2973	tgid = name_to_int(&dentry->d_name);
2974	if (tgid == ~0U)
2975		goto out;
2976
2977	ns = dentry->d_sb->s_fs_info;
2978	rcu_read_lock();
2979	task = find_task_by_pid_ns(tgid, ns);
2980	if (task)
2981		get_task_struct(task);
2982	rcu_read_unlock();
2983	if (!task)
2984		goto out;
2985
2986	result = proc_pid_instantiate(dir, dentry, task, NULL);
2987	put_task_struct(task);
2988out:
2989	return ERR_PTR(result);
2990}
2991
2992/*
2993 * Find the first task with tgid >= tgid
2994 *
2995 */
2996struct tgid_iter {
2997	unsigned int tgid;
2998	struct task_struct *task;
2999};
3000static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3001{
3002	struct pid *pid;
3003
3004	if (iter.task)
3005		put_task_struct(iter.task);
3006	rcu_read_lock();
3007retry:
3008	iter.task = NULL;
3009	pid = find_ge_pid(iter.tgid, ns);
3010	if (pid) {
3011		iter.tgid = pid_nr_ns(pid, ns);
3012		iter.task = pid_task(pid, PIDTYPE_PID);
3013		/* What we to know is if the pid we have find is the
3014		 * pid of a thread_group_leader.  Testing for task
3015		 * being a thread_group_leader is the obvious thing
3016		 * todo but there is a window when it fails, due to
3017		 * the pid transfer logic in de_thread.
3018		 *
3019		 * So we perform the straight forward test of seeing
3020		 * if the pid we have found is the pid of a thread
3021		 * group leader, and don't worry if the task we have
3022		 * found doesn't happen to be a thread group leader.
3023		 * As we don't care in the case of readdir.
3024		 */
3025		if (!iter.task || !has_group_leader_pid(iter.task)) {
3026			iter.tgid += 1;
3027			goto retry;
3028		}
3029		get_task_struct(iter.task);
3030	}
3031	rcu_read_unlock();
3032	return iter;
3033}
3034
3035#define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3036
3037/* for the /proc/ directory itself, after non-process stuff has been done */
3038int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3039{
3040	struct tgid_iter iter;
3041	struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
3042	loff_t pos = ctx->pos;
3043
3044	if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3045		return 0;
3046
3047	if (pos == TGID_OFFSET - 2) {
3048		struct inode *inode = d_inode(ns->proc_self);
3049		if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3050			return 0;
3051		ctx->pos = pos = pos + 1;
3052	}
3053	if (pos == TGID_OFFSET - 1) {
3054		struct inode *inode = d_inode(ns->proc_thread_self);
3055		if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3056			return 0;
3057		ctx->pos = pos = pos + 1;
3058	}
3059	iter.tgid = pos - TGID_OFFSET;
3060	iter.task = NULL;
3061	for (iter = next_tgid(ns, iter);
3062	     iter.task;
3063	     iter.tgid += 1, iter = next_tgid(ns, iter)) {
3064		char name[PROC_NUMBUF];
3065		int len;
3066		if (!has_pid_permissions(ns, iter.task, 2))
3067			continue;
3068
3069		len = snprintf(name, sizeof(name), "%d", iter.tgid);
3070		ctx->pos = iter.tgid + TGID_OFFSET;
3071		if (!proc_fill_cache(file, ctx, name, len,
3072				     proc_pid_instantiate, iter.task, NULL)) {
3073			put_task_struct(iter.task);
3074			return 0;
3075		}
3076	}
3077	ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3078	return 0;
3079}
3080
3081/*
3082 * Tasks
3083 */
3084static const struct pid_entry tid_base_stuff[] = {
3085	DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3086	DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3087	DIR("ns",	 S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3088#ifdef CONFIG_NET
3089	DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3090#endif
3091	REG("environ",   S_IRUSR, proc_environ_operations),
3092	ONE("auxv",      S_IRUSR, proc_pid_auxv),
3093	ONE("status",    S_IRUGO, proc_pid_status),
3094	ONE("personality", S_IRUSR, proc_pid_personality),
3095	ONE("limits",	 S_IRUGO, proc_pid_limits),
3096#ifdef CONFIG_SCHED_DEBUG
3097	REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3098#endif
3099	REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3100#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3101	ONE("syscall",   S_IRUSR, proc_pid_syscall),
3102#endif
3103	REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3104	ONE("stat",      S_IRUGO, proc_tid_stat),
3105	ONE("statm",     S_IRUGO, proc_pid_statm),
3106	REG("maps",      S_IRUGO, proc_tid_maps_operations),
3107#ifdef CONFIG_PROC_CHILDREN
3108	REG("children",  S_IRUGO, proc_tid_children_operations),
3109#endif
3110#ifdef CONFIG_NUMA
3111	REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3112#endif
3113	REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3114	LNK("cwd",       proc_cwd_link),
3115	LNK("root",      proc_root_link),
3116	LNK("exe",       proc_exe_link),
3117	REG("mounts",    S_IRUGO, proc_mounts_operations),
3118	REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3119#ifdef CONFIG_PROC_PAGE_MONITOR
3120	REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3121	REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3122	REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3123#endif
3124#ifdef CONFIG_SECURITY
3125	DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3126#endif
3127#ifdef CONFIG_KALLSYMS
3128	ONE("wchan",     S_IRUGO, proc_pid_wchan),
3129#endif
3130#ifdef CONFIG_STACKTRACE
3131	ONE("stack",      S_IRUSR, proc_pid_stack),
3132#endif
3133#ifdef CONFIG_SCHED_INFO
3134	ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3135#endif
3136#ifdef CONFIG_LATENCYTOP
3137	REG("latency",  S_IRUGO, proc_lstats_operations),
3138#endif
3139#ifdef CONFIG_PROC_PID_CPUSET
3140	ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3141#endif
3142#ifdef CONFIG_CGROUPS
3143	ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3144#endif
3145	ONE("oom_score", S_IRUGO, proc_oom_score),
3146	REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3147	REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3148#ifdef CONFIG_AUDITSYSCALL
3149	REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3150	REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3151#endif
3152#ifdef CONFIG_FAULT_INJECTION
3153	REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3154#endif
3155#ifdef CONFIG_TASK_IO_ACCOUNTING
3156	ONE("io",	S_IRUSR, proc_tid_io_accounting),
3157#endif
3158#ifdef CONFIG_HARDWALL
3159	ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
3160#endif
3161#ifdef CONFIG_USER_NS
3162	REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3163	REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3164	REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3165	REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3166#endif
3167};
3168
3169static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3170{
3171	return proc_pident_readdir(file, ctx,
3172				   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3173}
3174
3175static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3176{
3177	return proc_pident_lookup(dir, dentry,
3178				  tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3179}
3180
3181static const struct file_operations proc_tid_base_operations = {
3182	.read		= generic_read_dir,
3183	.iterate	= proc_tid_base_readdir,
3184	.llseek		= default_llseek,
3185};
3186
3187static const struct inode_operations proc_tid_base_inode_operations = {
3188	.lookup		= proc_tid_base_lookup,
3189	.getattr	= pid_getattr,
3190	.setattr	= proc_setattr,
3191};
3192
3193static int proc_task_instantiate(struct inode *dir,
3194	struct dentry *dentry, struct task_struct *task, const void *ptr)
3195{
3196	struct inode *inode;
3197	inode = proc_pid_make_inode(dir->i_sb, task);
3198
3199	if (!inode)
3200		goto out;
3201	inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3202	inode->i_op = &proc_tid_base_inode_operations;
3203	inode->i_fop = &proc_tid_base_operations;
3204	inode->i_flags|=S_IMMUTABLE;
3205
3206	set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3207						  ARRAY_SIZE(tid_base_stuff)));
3208
3209	d_set_d_op(dentry, &pid_dentry_operations);
3210
3211	d_add(dentry, inode);
3212	/* Close the race of the process dying before we return the dentry */
3213	if (pid_revalidate(dentry, 0))
3214		return 0;
3215out:
3216	return -ENOENT;
3217}
3218
3219static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3220{
3221	int result = -ENOENT;
3222	struct task_struct *task;
3223	struct task_struct *leader = get_proc_task(dir);
3224	unsigned tid;
3225	struct pid_namespace *ns;
3226
3227	if (!leader)
3228		goto out_no_task;
3229
3230	tid = name_to_int(&dentry->d_name);
3231	if (tid == ~0U)
3232		goto out;
3233
3234	ns = dentry->d_sb->s_fs_info;
3235	rcu_read_lock();
3236	task = find_task_by_pid_ns(tid, ns);
3237	if (task)
3238		get_task_struct(task);
3239	rcu_read_unlock();
3240	if (!task)
3241		goto out;
3242	if (!same_thread_group(leader, task))
3243		goto out_drop_task;
3244
3245	result = proc_task_instantiate(dir, dentry, task, NULL);
3246out_drop_task:
3247	put_task_struct(task);
3248out:
3249	put_task_struct(leader);
3250out_no_task:
3251	return ERR_PTR(result);
3252}
3253
3254/*
3255 * Find the first tid of a thread group to return to user space.
3256 *
3257 * Usually this is just the thread group leader, but if the users
3258 * buffer was too small or there was a seek into the middle of the
3259 * directory we have more work todo.
3260 *
3261 * In the case of a short read we start with find_task_by_pid.
3262 *
3263 * In the case of a seek we start with the leader and walk nr
3264 * threads past it.
3265 */
3266static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3267					struct pid_namespace *ns)
3268{
3269	struct task_struct *pos, *task;
3270	unsigned long nr = f_pos;
3271
3272	if (nr != f_pos)	/* 32bit overflow? */
3273		return NULL;
3274
3275	rcu_read_lock();
3276	task = pid_task(pid, PIDTYPE_PID);
3277	if (!task)
3278		goto fail;
3279
3280	/* Attempt to start with the tid of a thread */
3281	if (tid && nr) {
3282		pos = find_task_by_pid_ns(tid, ns);
3283		if (pos && same_thread_group(pos, task))
3284			goto found;
3285	}
3286
3287	/* If nr exceeds the number of threads there is nothing todo */
3288	if (nr >= get_nr_threads(task))
3289		goto fail;
3290
3291	/* If we haven't found our starting place yet start
3292	 * with the leader and walk nr threads forward.
3293	 */
3294	pos = task = task->group_leader;
3295	do {
3296		if (!nr--)
3297			goto found;
3298	} while_each_thread(task, pos);
3299fail:
3300	pos = NULL;
3301	goto out;
3302found:
3303	get_task_struct(pos);
3304out:
3305	rcu_read_unlock();
3306	return pos;
3307}
3308
3309/*
3310 * Find the next thread in the thread list.
3311 * Return NULL if there is an error or no next thread.
3312 *
3313 * The reference to the input task_struct is released.
3314 */
3315static struct task_struct *next_tid(struct task_struct *start)
3316{
3317	struct task_struct *pos = NULL;
3318	rcu_read_lock();
3319	if (pid_alive(start)) {
3320		pos = next_thread(start);
3321		if (thread_group_leader(pos))
3322			pos = NULL;
3323		else
3324			get_task_struct(pos);
3325	}
3326	rcu_read_unlock();
3327	put_task_struct(start);
3328	return pos;
3329}
3330
3331/* for the /proc/TGID/task/ directories */
3332static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3333{
3334	struct inode *inode = file_inode(file);
3335	struct task_struct *task;
3336	struct pid_namespace *ns;
3337	int tid;
3338
3339	if (proc_inode_is_dead(inode))
3340		return -ENOENT;
3341
3342	if (!dir_emit_dots(file, ctx))
3343		return 0;
3344
3345	/* f_version caches the tgid value that the last readdir call couldn't
3346	 * return. lseek aka telldir automagically resets f_version to 0.
3347	 */
3348	ns = inode->i_sb->s_fs_info;
3349	tid = (int)file->f_version;
3350	file->f_version = 0;
3351	for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3352	     task;
3353	     task = next_tid(task), ctx->pos++) {
3354		char name[PROC_NUMBUF];
3355		int len;
3356		tid = task_pid_nr_ns(task, ns);
3357		len = snprintf(name, sizeof(name), "%d", tid);
3358		if (!proc_fill_cache(file, ctx, name, len,
3359				proc_task_instantiate, task, NULL)) {
3360			/* returning this tgid failed, save it as the first
3361			 * pid for the next readir call */
3362			file->f_version = (u64)tid;
3363			put_task_struct(task);
3364			break;
3365		}
3366	}
3367
3368	return 0;
3369}
3370
3371static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3372{
3373	struct inode *inode = d_inode(dentry);
3374	struct task_struct *p = get_proc_task(inode);
3375	generic_fillattr(inode, stat);
3376
3377	if (p) {
3378		stat->nlink += get_nr_threads(p);
3379		put_task_struct(p);
3380	}
3381
3382	return 0;
3383}
3384
3385static const struct inode_operations proc_task_inode_operations = {
3386	.lookup		= proc_task_lookup,
3387	.getattr	= proc_task_getattr,
3388	.setattr	= proc_setattr,
3389	.permission	= proc_pid_permission,
3390};
3391
3392static const struct file_operations proc_task_operations = {
3393	.read		= generic_read_dir,
3394	.iterate	= proc_task_readdir,
3395	.llseek		= default_llseek,
3396};
3397