• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 #include <linux/cpufreq_times.h>
91 #ifdef CONFIG_HARDWALL
92 #include <asm/hardwall.h>
93 #endif
94 #include <trace/events/oom.h>
95 #include "internal.h"
96 #include "fd.h"
97 
98 /* NOTE:
99  *	Implementing inode permission operations in /proc is almost
100  *	certainly an error.  Permission checks need to happen during
101  *	each system call not at open time.  The reason is that most of
102  *	what we wish to check for permissions in /proc varies at runtime.
103  *
104  *	The classic example of a problem is opening file descriptors
105  *	in /proc for a task before it execs a suid executable.
106  */
107 
108 struct pid_entry {
109 	const char *name;
110 	int len;
111 	umode_t mode;
112 	const struct inode_operations *iop;
113 	const struct file_operations *fop;
114 	union proc_op op;
115 };
116 
117 #define NOD(NAME, MODE, IOP, FOP, OP) {			\
118 	.name = (NAME),					\
119 	.len  = sizeof(NAME) - 1,			\
120 	.mode = MODE,					\
121 	.iop  = IOP,					\
122 	.fop  = FOP,					\
123 	.op   = OP,					\
124 }
125 
126 #define DIR(NAME, MODE, iops, fops)	\
127 	NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
128 #define LNK(NAME, get_link)					\
129 	NOD(NAME, (S_IFLNK|S_IRWXUGO),				\
130 		&proc_pid_link_inode_operations, NULL,		\
131 		{ .proc_get_link = get_link } )
132 #define REG(NAME, MODE, fops)				\
133 	NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
134 #define ONE(NAME, MODE, show)				\
135 	NOD(NAME, (S_IFREG|(MODE)), 			\
136 		NULL, &proc_single_file_operations,	\
137 		{ .proc_show = show } )
138 
139 /*
140  * Count the number of hardlinks for the pid_entry table, excluding the .
141  * and .. links.
142  */
pid_entry_count_dirs(const struct pid_entry * entries,unsigned int n)143 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
144 	unsigned int n)
145 {
146 	unsigned int i;
147 	unsigned int count;
148 
149 	count = 0;
150 	for (i = 0; i < n; ++i) {
151 		if (S_ISDIR(entries[i].mode))
152 			++count;
153 	}
154 
155 	return count;
156 }
157 
get_task_root(struct task_struct * task,struct path * root)158 static int get_task_root(struct task_struct *task, struct path *root)
159 {
160 	int result = -ENOENT;
161 
162 	task_lock(task);
163 	if (task->fs) {
164 		get_fs_root(task->fs, root);
165 		result = 0;
166 	}
167 	task_unlock(task);
168 	return result;
169 }
170 
proc_cwd_link(struct dentry * dentry,struct path * path)171 static int proc_cwd_link(struct dentry *dentry, struct path *path)
172 {
173 	struct task_struct *task = get_proc_task(d_inode(dentry));
174 	int result = -ENOENT;
175 
176 	if (task) {
177 		task_lock(task);
178 		if (task->fs) {
179 			get_fs_pwd(task->fs, path);
180 			result = 0;
181 		}
182 		task_unlock(task);
183 		put_task_struct(task);
184 	}
185 	return result;
186 }
187 
proc_root_link(struct dentry * dentry,struct path * path)188 static int proc_root_link(struct dentry *dentry, struct path *path)
189 {
190 	struct task_struct *task = get_proc_task(d_inode(dentry));
191 	int result = -ENOENT;
192 
193 	if (task) {
194 		result = get_task_root(task, path);
195 		put_task_struct(task);
196 	}
197 	return result;
198 }
199 
proc_pid_cmdline_read(struct file * file,char __user * buf,size_t _count,loff_t * pos)200 static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
201 				     size_t _count, loff_t *pos)
202 {
203 	struct task_struct *tsk;
204 	struct mm_struct *mm;
205 	char *page;
206 	unsigned long count = _count;
207 	unsigned long arg_start, arg_end, env_start, env_end;
208 	unsigned long len1, len2, len;
209 	unsigned long p;
210 	char c;
211 	ssize_t rv;
212 
213 	BUG_ON(*pos < 0);
214 
215 	tsk = get_proc_task(file_inode(file));
216 	if (!tsk)
217 		return -ESRCH;
218 	mm = get_task_mm(tsk);
219 	put_task_struct(tsk);
220 	if (!mm)
221 		return 0;
222 	/* Check if process spawned far enough to have cmdline. */
223 	if (!mm->env_end) {
224 		rv = 0;
225 		goto out_mmput;
226 	}
227 
228 	page = (char *)__get_free_page(GFP_TEMPORARY);
229 	if (!page) {
230 		rv = -ENOMEM;
231 		goto out_mmput;
232 	}
233 
234 	down_read(&mm->mmap_sem);
235 	arg_start = mm->arg_start;
236 	arg_end = mm->arg_end;
237 	env_start = mm->env_start;
238 	env_end = mm->env_end;
239 	up_read(&mm->mmap_sem);
240 
241 	BUG_ON(arg_start > arg_end);
242 	BUG_ON(env_start > env_end);
243 
244 	len1 = arg_end - arg_start;
245 	len2 = env_end - env_start;
246 
247 	/* Empty ARGV. */
248 	if (len1 == 0) {
249 		rv = 0;
250 		goto out_free_page;
251 	}
252 	/*
253 	 * Inherently racy -- command line shares address space
254 	 * with code and data.
255 	 */
256 	rv = access_remote_vm(mm, arg_end - 1, &c, 1, 0);
257 	if (rv <= 0)
258 		goto out_free_page;
259 
260 	rv = 0;
261 
262 	if (c == '\0') {
263 		/* Command line (set of strings) occupies whole ARGV. */
264 		if (len1 <= *pos)
265 			goto out_free_page;
266 
267 		p = arg_start + *pos;
268 		len = len1 - *pos;
269 		while (count > 0 && len > 0) {
270 			unsigned int _count;
271 			int nr_read;
272 
273 			_count = min3(count, len, PAGE_SIZE);
274 			nr_read = access_remote_vm(mm, p, page, _count, 0);
275 			if (nr_read < 0)
276 				rv = nr_read;
277 			if (nr_read <= 0)
278 				goto out_free_page;
279 
280 			if (copy_to_user(buf, page, nr_read)) {
281 				rv = -EFAULT;
282 				goto out_free_page;
283 			}
284 
285 			p	+= nr_read;
286 			len	-= nr_read;
287 			buf	+= nr_read;
288 			count	-= nr_read;
289 			rv	+= nr_read;
290 		}
291 	} else {
292 		/*
293 		 * Command line (1 string) occupies ARGV and maybe
294 		 * extends into ENVP.
295 		 */
296 		if (len1 + len2 <= *pos)
297 			goto skip_argv_envp;
298 		if (len1 <= *pos)
299 			goto skip_argv;
300 
301 		p = arg_start + *pos;
302 		len = len1 - *pos;
303 		while (count > 0 && len > 0) {
304 			unsigned int _count, l;
305 			int nr_read;
306 			bool final;
307 
308 			_count = min3(count, len, PAGE_SIZE);
309 			nr_read = access_remote_vm(mm, p, page, _count, 0);
310 			if (nr_read < 0)
311 				rv = nr_read;
312 			if (nr_read <= 0)
313 				goto out_free_page;
314 
315 			/*
316 			 * Command line can be shorter than whole ARGV
317 			 * even if last "marker" byte says it is not.
318 			 */
319 			final = false;
320 			l = strnlen(page, nr_read);
321 			if (l < nr_read) {
322 				nr_read = l;
323 				final = true;
324 			}
325 
326 			if (copy_to_user(buf, page, nr_read)) {
327 				rv = -EFAULT;
328 				goto out_free_page;
329 			}
330 
331 			p	+= nr_read;
332 			len	-= nr_read;
333 			buf	+= nr_read;
334 			count	-= nr_read;
335 			rv	+= nr_read;
336 
337 			if (final)
338 				goto out_free_page;
339 		}
340 skip_argv:
341 		/*
342 		 * Command line (1 string) occupies ARGV and
343 		 * extends into ENVP.
344 		 */
345 		if (len1 <= *pos) {
346 			p = env_start + *pos - len1;
347 			len = len1 + len2 - *pos;
348 		} else {
349 			p = env_start;
350 			len = len2;
351 		}
352 		while (count > 0 && len > 0) {
353 			unsigned int _count, l;
354 			int nr_read;
355 			bool final;
356 
357 			_count = min3(count, len, PAGE_SIZE);
358 			nr_read = access_remote_vm(mm, p, page, _count, 0);
359 			if (nr_read < 0)
360 				rv = nr_read;
361 			if (nr_read <= 0)
362 				goto out_free_page;
363 
364 			/* Find EOS. */
365 			final = false;
366 			l = strnlen(page, nr_read);
367 			if (l < nr_read) {
368 				nr_read = l;
369 				final = true;
370 			}
371 
372 			if (copy_to_user(buf, page, nr_read)) {
373 				rv = -EFAULT;
374 				goto out_free_page;
375 			}
376 
377 			p	+= nr_read;
378 			len	-= nr_read;
379 			buf	+= nr_read;
380 			count	-= nr_read;
381 			rv	+= nr_read;
382 
383 			if (final)
384 				goto out_free_page;
385 		}
386 skip_argv_envp:
387 		;
388 	}
389 
390 out_free_page:
391 	free_page((unsigned long)page);
392 out_mmput:
393 	mmput(mm);
394 	if (rv > 0)
395 		*pos += rv;
396 	return rv;
397 }
398 
399 static const struct file_operations proc_pid_cmdline_ops = {
400 	.read	= proc_pid_cmdline_read,
401 	.llseek	= generic_file_llseek,
402 };
403 
404 #ifdef CONFIG_KALLSYMS
405 /*
406  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
407  * Returns the resolved symbol.  If that fails, simply return the address.
408  */
proc_pid_wchan(struct seq_file * m,struct pid_namespace * ns,struct pid * pid,struct task_struct * task)409 static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
410 			  struct pid *pid, struct task_struct *task)
411 {
412 	unsigned long wchan;
413 	char symname[KSYM_NAME_LEN];
414 
415 	wchan = get_wchan(task);
416 
417 	if (wchan && ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)
418 			&& !lookup_symbol_name(wchan, symname))
419 		seq_printf(m, "%s", symname);
420 	else
421 		seq_putc(m, '0');
422 
423 	return 0;
424 }
425 #endif /* CONFIG_KALLSYMS */
426 
lock_trace(struct task_struct * task)427 static int lock_trace(struct task_struct *task)
428 {
429 	int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
430 	if (err)
431 		return err;
432 	if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
433 		mutex_unlock(&task->signal->cred_guard_mutex);
434 		return -EPERM;
435 	}
436 	return 0;
437 }
438 
unlock_trace(struct task_struct * task)439 static void unlock_trace(struct task_struct *task)
440 {
441 	mutex_unlock(&task->signal->cred_guard_mutex);
442 }
443 
444 #ifdef CONFIG_STACKTRACE
445 
446 #define MAX_STACK_TRACE_DEPTH	64
447 
proc_pid_stack(struct seq_file * m,struct pid_namespace * ns,struct pid * pid,struct task_struct * task)448 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
449 			  struct pid *pid, struct task_struct *task)
450 {
451 	struct stack_trace trace;
452 	unsigned long *entries;
453 	int err;
454 	int i;
455 
456 	entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
457 	if (!entries)
458 		return -ENOMEM;
459 
460 	trace.nr_entries	= 0;
461 	trace.max_entries	= MAX_STACK_TRACE_DEPTH;
462 	trace.entries		= entries;
463 	trace.skip		= 0;
464 
465 	err = lock_trace(task);
466 	if (!err) {
467 		save_stack_trace_tsk(task, &trace);
468 
469 		for (i = 0; i < trace.nr_entries; i++) {
470 			seq_printf(m, "[<%pK>] %pB\n",
471 				   (void *)entries[i], (void *)entries[i]);
472 		}
473 		unlock_trace(task);
474 	}
475 	kfree(entries);
476 
477 	return err;
478 }
479 #endif
480 
481 #ifdef CONFIG_SCHED_INFO
482 /*
483  * Provides /proc/PID/schedstat
484  */
proc_pid_schedstat(struct seq_file * m,struct pid_namespace * ns,struct pid * pid,struct task_struct * task)485 static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
486 			      struct pid *pid, struct task_struct *task)
487 {
488 	if (unlikely(!sched_info_on()))
489 		seq_printf(m, "0 0 0\n");
490 	else
491 		seq_printf(m, "%llu %llu %lu\n",
492 		   (unsigned long long)task->se.sum_exec_runtime,
493 		   (unsigned long long)task->sched_info.run_delay,
494 		   task->sched_info.pcount);
495 
496 	return 0;
497 }
498 #endif
499 
500 #ifdef CONFIG_LATENCYTOP
lstats_show_proc(struct seq_file * m,void * v)501 static int lstats_show_proc(struct seq_file *m, void *v)
502 {
503 	int i;
504 	struct inode *inode = m->private;
505 	struct task_struct *task = get_proc_task(inode);
506 
507 	if (!task)
508 		return -ESRCH;
509 	seq_puts(m, "Latency Top version : v0.1\n");
510 	for (i = 0; i < 32; i++) {
511 		struct latency_record *lr = &task->latency_record[i];
512 		if (lr->backtrace[0]) {
513 			int q;
514 			seq_printf(m, "%i %li %li",
515 				   lr->count, lr->time, lr->max);
516 			for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
517 				unsigned long bt = lr->backtrace[q];
518 				if (!bt)
519 					break;
520 				if (bt == ULONG_MAX)
521 					break;
522 				seq_printf(m, " %ps", (void *)bt);
523 			}
524 			seq_putc(m, '\n');
525 		}
526 
527 	}
528 	put_task_struct(task);
529 	return 0;
530 }
531 
lstats_open(struct inode * inode,struct file * file)532 static int lstats_open(struct inode *inode, struct file *file)
533 {
534 	return single_open(file, lstats_show_proc, inode);
535 }
536 
lstats_write(struct file * file,const char __user * buf,size_t count,loff_t * offs)537 static ssize_t lstats_write(struct file *file, const char __user *buf,
538 			    size_t count, loff_t *offs)
539 {
540 	struct task_struct *task = get_proc_task(file_inode(file));
541 
542 	if (!task)
543 		return -ESRCH;
544 	clear_all_latency_tracing(task);
545 	put_task_struct(task);
546 
547 	return count;
548 }
549 
550 static const struct file_operations proc_lstats_operations = {
551 	.open		= lstats_open,
552 	.read		= seq_read,
553 	.write		= lstats_write,
554 	.llseek		= seq_lseek,
555 	.release	= single_release,
556 };
557 
558 #endif
559 
proc_oom_score(struct seq_file * m,struct pid_namespace * ns,struct pid * pid,struct task_struct * task)560 static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
561 			  struct pid *pid, struct task_struct *task)
562 {
563 	unsigned long totalpages = totalram_pages + total_swap_pages;
564 	unsigned long points = 0;
565 
566 	points = oom_badness(task, NULL, NULL, totalpages) *
567 					1000 / totalpages;
568 	seq_printf(m, "%lu\n", points);
569 
570 	return 0;
571 }
572 
573 struct limit_names {
574 	const char *name;
575 	const char *unit;
576 };
577 
578 static const struct limit_names lnames[RLIM_NLIMITS] = {
579 	[RLIMIT_CPU] = {"Max cpu time", "seconds"},
580 	[RLIMIT_FSIZE] = {"Max file size", "bytes"},
581 	[RLIMIT_DATA] = {"Max data size", "bytes"},
582 	[RLIMIT_STACK] = {"Max stack size", "bytes"},
583 	[RLIMIT_CORE] = {"Max core file size", "bytes"},
584 	[RLIMIT_RSS] = {"Max resident set", "bytes"},
585 	[RLIMIT_NPROC] = {"Max processes", "processes"},
586 	[RLIMIT_NOFILE] = {"Max open files", "files"},
587 	[RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
588 	[RLIMIT_AS] = {"Max address space", "bytes"},
589 	[RLIMIT_LOCKS] = {"Max file locks", "locks"},
590 	[RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
591 	[RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
592 	[RLIMIT_NICE] = {"Max nice priority", NULL},
593 	[RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
594 	[RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
595 };
596 
597 /* Display limits for a process */
proc_pid_limits(struct seq_file * m,struct pid_namespace * ns,struct pid * pid,struct task_struct * task)598 static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
599 			   struct pid *pid, struct task_struct *task)
600 {
601 	unsigned int i;
602 	unsigned long flags;
603 
604 	struct rlimit rlim[RLIM_NLIMITS];
605 
606 	if (!lock_task_sighand(task, &flags))
607 		return 0;
608 	memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
609 	unlock_task_sighand(task, &flags);
610 
611 	/*
612 	 * print the file header
613 	 */
614        seq_printf(m, "%-25s %-20s %-20s %-10s\n",
615 		  "Limit", "Soft Limit", "Hard Limit", "Units");
616 
617 	for (i = 0; i < RLIM_NLIMITS; i++) {
618 		if (rlim[i].rlim_cur == RLIM_INFINITY)
619 			seq_printf(m, "%-25s %-20s ",
620 				   lnames[i].name, "unlimited");
621 		else
622 			seq_printf(m, "%-25s %-20lu ",
623 				   lnames[i].name, rlim[i].rlim_cur);
624 
625 		if (rlim[i].rlim_max == RLIM_INFINITY)
626 			seq_printf(m, "%-20s ", "unlimited");
627 		else
628 			seq_printf(m, "%-20lu ", rlim[i].rlim_max);
629 
630 		if (lnames[i].unit)
631 			seq_printf(m, "%-10s\n", lnames[i].unit);
632 		else
633 			seq_putc(m, '\n');
634 	}
635 
636 	return 0;
637 }
638 
639 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
proc_pid_syscall(struct seq_file * m,struct pid_namespace * ns,struct pid * pid,struct task_struct * task)640 static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
641 			    struct pid *pid, struct task_struct *task)
642 {
643 	long nr;
644 	unsigned long args[6], sp, pc;
645 	int res;
646 
647 	res = lock_trace(task);
648 	if (res)
649 		return res;
650 
651 	if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
652 		seq_puts(m, "running\n");
653 	else if (nr < 0)
654 		seq_printf(m, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
655 	else
656 		seq_printf(m,
657 		       "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
658 		       nr,
659 		       args[0], args[1], args[2], args[3], args[4], args[5],
660 		       sp, pc);
661 	unlock_trace(task);
662 
663 	return 0;
664 }
665 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
666 
667 /************************************************************************/
668 /*                       Here the fs part begins                        */
669 /************************************************************************/
670 
671 /* permission checks */
proc_fd_access_allowed(struct inode * inode)672 static int proc_fd_access_allowed(struct inode *inode)
673 {
674 	struct task_struct *task;
675 	int allowed = 0;
676 	/* Allow access to a task's file descriptors if it is us or we
677 	 * may use ptrace attach to the process and find out that
678 	 * information.
679 	 */
680 	task = get_proc_task(inode);
681 	if (task) {
682 		allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
683 		put_task_struct(task);
684 	}
685 	return allowed;
686 }
687 
proc_setattr(struct dentry * dentry,struct iattr * attr)688 int proc_setattr(struct dentry *dentry, struct iattr *attr)
689 {
690 	int error;
691 	struct inode *inode = d_inode(dentry);
692 
693 	if (attr->ia_valid & ATTR_MODE)
694 		return -EPERM;
695 
696 	error = setattr_prepare(dentry, attr);
697 	if (error)
698 		return error;
699 
700 	setattr_copy(inode, attr);
701 	mark_inode_dirty(inode);
702 	return 0;
703 }
704 
705 /*
706  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
707  * or euid/egid (for hide_pid_min=2)?
708  */
has_pid_permissions(struct pid_namespace * pid,struct task_struct * task,int hide_pid_min)709 static bool has_pid_permissions(struct pid_namespace *pid,
710 				 struct task_struct *task,
711 				 int hide_pid_min)
712 {
713 	if (pid->hide_pid < hide_pid_min)
714 		return true;
715 	if (in_group_p(pid->pid_gid))
716 		return true;
717 	return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
718 }
719 
720 
proc_pid_permission(struct inode * inode,int mask)721 static int proc_pid_permission(struct inode *inode, int mask)
722 {
723 	struct pid_namespace *pid = inode->i_sb->s_fs_info;
724 	struct task_struct *task;
725 	bool has_perms;
726 
727 	task = get_proc_task(inode);
728 	if (!task)
729 		return -ESRCH;
730 	has_perms = has_pid_permissions(pid, task, 1);
731 	put_task_struct(task);
732 
733 	if (!has_perms) {
734 		if (pid->hide_pid == 2) {
735 			/*
736 			 * Let's make getdents(), stat(), and open()
737 			 * consistent with each other.  If a process
738 			 * may not stat() a file, it shouldn't be seen
739 			 * in procfs at all.
740 			 */
741 			return -ENOENT;
742 		}
743 
744 		return -EPERM;
745 	}
746 	return generic_permission(inode, mask);
747 }
748 
749 
750 
751 static const struct inode_operations proc_def_inode_operations = {
752 	.setattr	= proc_setattr,
753 };
754 
proc_single_show(struct seq_file * m,void * v)755 static int proc_single_show(struct seq_file *m, void *v)
756 {
757 	struct inode *inode = m->private;
758 	struct pid_namespace *ns;
759 	struct pid *pid;
760 	struct task_struct *task;
761 	int ret;
762 
763 	ns = inode->i_sb->s_fs_info;
764 	pid = proc_pid(inode);
765 	task = get_pid_task(pid, PIDTYPE_PID);
766 	if (!task)
767 		return -ESRCH;
768 
769 	ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
770 
771 	put_task_struct(task);
772 	return ret;
773 }
774 
proc_single_open(struct inode * inode,struct file * filp)775 static int proc_single_open(struct inode *inode, struct file *filp)
776 {
777 	return single_open(filp, proc_single_show, inode);
778 }
779 
780 static const struct file_operations proc_single_file_operations = {
781 	.open		= proc_single_open,
782 	.read		= seq_read,
783 	.llseek		= seq_lseek,
784 	.release	= single_release,
785 };
786 
787 
proc_mem_open(struct inode * inode,unsigned int mode)788 struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
789 {
790 	struct task_struct *task = get_proc_task(inode);
791 	struct mm_struct *mm = ERR_PTR(-ESRCH);
792 
793 	if (task) {
794 		mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
795 		put_task_struct(task);
796 
797 		if (!IS_ERR_OR_NULL(mm)) {
798 			/* ensure this mm_struct can't be freed */
799 			atomic_inc(&mm->mm_count);
800 			/* but do not pin its memory */
801 			mmput(mm);
802 		}
803 	}
804 
805 	return mm;
806 }
807 
__mem_open(struct inode * inode,struct file * file,unsigned int mode)808 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
809 {
810 	struct mm_struct *mm = proc_mem_open(inode, mode);
811 
812 	if (IS_ERR(mm))
813 		return PTR_ERR(mm);
814 
815 	file->private_data = mm;
816 	return 0;
817 }
818 
mem_open(struct inode * inode,struct file * file)819 static int mem_open(struct inode *inode, struct file *file)
820 {
821 	int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
822 
823 	/* OK to pass negative loff_t, we can catch out-of-range */
824 	file->f_mode |= FMODE_UNSIGNED_OFFSET;
825 
826 	return ret;
827 }
828 
mem_rw(struct file * file,char __user * buf,size_t count,loff_t * ppos,int write)829 static ssize_t mem_rw(struct file *file, char __user *buf,
830 			size_t count, loff_t *ppos, int write)
831 {
832 	struct mm_struct *mm = file->private_data;
833 	unsigned long addr = *ppos;
834 	ssize_t copied;
835 	char *page;
836 	unsigned int flags;
837 
838 	if (!mm)
839 		return 0;
840 
841 	page = (char *)__get_free_page(GFP_TEMPORARY);
842 	if (!page)
843 		return -ENOMEM;
844 
845 	copied = 0;
846 	if (!atomic_inc_not_zero(&mm->mm_users))
847 		goto free;
848 
849 	/* Maybe we should limit FOLL_FORCE to actual ptrace users? */
850 	flags = FOLL_FORCE;
851 	if (write)
852 		flags |= FOLL_WRITE;
853 
854 	while (count > 0) {
855 		int this_len = min_t(int, count, PAGE_SIZE);
856 
857 		if (write && copy_from_user(page, buf, this_len)) {
858 			copied = -EFAULT;
859 			break;
860 		}
861 
862 		this_len = access_remote_vm(mm, addr, page, this_len, flags);
863 		if (!this_len) {
864 			if (!copied)
865 				copied = -EIO;
866 			break;
867 		}
868 
869 		if (!write && copy_to_user(buf, page, this_len)) {
870 			copied = -EFAULT;
871 			break;
872 		}
873 
874 		buf += this_len;
875 		addr += this_len;
876 		copied += this_len;
877 		count -= this_len;
878 	}
879 	*ppos = addr;
880 
881 	mmput(mm);
882 free:
883 	free_page((unsigned long) page);
884 	return copied;
885 }
886 
mem_read(struct file * file,char __user * buf,size_t count,loff_t * ppos)887 static ssize_t mem_read(struct file *file, char __user *buf,
888 			size_t count, loff_t *ppos)
889 {
890 	return mem_rw(file, buf, count, ppos, 0);
891 }
892 
mem_write(struct file * file,const char __user * buf,size_t count,loff_t * ppos)893 static ssize_t mem_write(struct file *file, const char __user *buf,
894 			 size_t count, loff_t *ppos)
895 {
896 	return mem_rw(file, (char __user*)buf, count, ppos, 1);
897 }
898 
mem_lseek(struct file * file,loff_t offset,int orig)899 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
900 {
901 	switch (orig) {
902 	case 0:
903 		file->f_pos = offset;
904 		break;
905 	case 1:
906 		file->f_pos += offset;
907 		break;
908 	default:
909 		return -EINVAL;
910 	}
911 	force_successful_syscall_return();
912 	return file->f_pos;
913 }
914 
mem_release(struct inode * inode,struct file * file)915 static int mem_release(struct inode *inode, struct file *file)
916 {
917 	struct mm_struct *mm = file->private_data;
918 	if (mm)
919 		mmdrop(mm);
920 	return 0;
921 }
922 
923 static const struct file_operations proc_mem_operations = {
924 	.llseek		= mem_lseek,
925 	.read		= mem_read,
926 	.write		= mem_write,
927 	.open		= mem_open,
928 	.release	= mem_release,
929 };
930 
environ_open(struct inode * inode,struct file * file)931 static int environ_open(struct inode *inode, struct file *file)
932 {
933 	return __mem_open(inode, file, PTRACE_MODE_READ);
934 }
935 
environ_read(struct file * file,char __user * buf,size_t count,loff_t * ppos)936 static ssize_t environ_read(struct file *file, char __user *buf,
937 			size_t count, loff_t *ppos)
938 {
939 	char *page;
940 	unsigned long src = *ppos;
941 	int ret = 0;
942 	struct mm_struct *mm = file->private_data;
943 	unsigned long env_start, env_end;
944 
945 	/* Ensure the process spawned far enough to have an environment. */
946 	if (!mm || !mm->env_end)
947 		return 0;
948 
949 	page = (char *)__get_free_page(GFP_TEMPORARY);
950 	if (!page)
951 		return -ENOMEM;
952 
953 	ret = 0;
954 	if (!atomic_inc_not_zero(&mm->mm_users))
955 		goto free;
956 
957 	down_read(&mm->mmap_sem);
958 	env_start = mm->env_start;
959 	env_end = mm->env_end;
960 	up_read(&mm->mmap_sem);
961 
962 	while (count > 0) {
963 		size_t this_len, max_len;
964 		int retval;
965 
966 		if (src >= (env_end - env_start))
967 			break;
968 
969 		this_len = env_end - (env_start + src);
970 
971 		max_len = min_t(size_t, PAGE_SIZE, count);
972 		this_len = min(max_len, this_len);
973 
974 		retval = access_remote_vm(mm, (env_start + src), page, this_len, 0);
975 
976 		if (retval <= 0) {
977 			ret = retval;
978 			break;
979 		}
980 
981 		if (copy_to_user(buf, page, retval)) {
982 			ret = -EFAULT;
983 			break;
984 		}
985 
986 		ret += retval;
987 		src += retval;
988 		buf += retval;
989 		count -= retval;
990 	}
991 	*ppos = src;
992 	mmput(mm);
993 
994 free:
995 	free_page((unsigned long) page);
996 	return ret;
997 }
998 
999 static const struct file_operations proc_environ_operations = {
1000 	.open		= environ_open,
1001 	.read		= environ_read,
1002 	.llseek		= generic_file_llseek,
1003 	.release	= mem_release,
1004 };
1005 
auxv_open(struct inode * inode,struct file * file)1006 static int auxv_open(struct inode *inode, struct file *file)
1007 {
1008 	return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
1009 }
1010 
auxv_read(struct file * file,char __user * buf,size_t count,loff_t * ppos)1011 static ssize_t auxv_read(struct file *file, char __user *buf,
1012 			size_t count, loff_t *ppos)
1013 {
1014 	struct mm_struct *mm = file->private_data;
1015 	unsigned int nwords = 0;
1016 
1017 	if (!mm)
1018 		return 0;
1019 	do {
1020 		nwords += 2;
1021 	} while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
1022 	return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
1023 				       nwords * sizeof(mm->saved_auxv[0]));
1024 }
1025 
1026 static const struct file_operations proc_auxv_operations = {
1027 	.open		= auxv_open,
1028 	.read		= auxv_read,
1029 	.llseek		= generic_file_llseek,
1030 	.release	= mem_release,
1031 };
1032 
oom_adj_read(struct file * file,char __user * buf,size_t count,loff_t * ppos)1033 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1034 			    loff_t *ppos)
1035 {
1036 	struct task_struct *task = get_proc_task(file_inode(file));
1037 	char buffer[PROC_NUMBUF];
1038 	int oom_adj = OOM_ADJUST_MIN;
1039 	size_t len;
1040 
1041 	if (!task)
1042 		return -ESRCH;
1043 	if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1044 		oom_adj = OOM_ADJUST_MAX;
1045 	else
1046 		oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1047 			  OOM_SCORE_ADJ_MAX;
1048 	put_task_struct(task);
1049 	len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1050 	return simple_read_from_buffer(buf, count, ppos, buffer, len);
1051 }
1052 
__set_oom_adj(struct file * file,int oom_adj,bool legacy)1053 static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1054 {
1055 	static DEFINE_MUTEX(oom_adj_mutex);
1056 	struct mm_struct *mm = NULL;
1057 	struct task_struct *task;
1058 	int err = 0;
1059 
1060 	task = get_proc_task(file_inode(file));
1061 	if (!task)
1062 		return -ESRCH;
1063 
1064 	mutex_lock(&oom_adj_mutex);
1065 	if (legacy) {
1066 		if (oom_adj < task->signal->oom_score_adj &&
1067 				!capable(CAP_SYS_RESOURCE)) {
1068 			err = -EACCES;
1069 			goto err_unlock;
1070 		}
1071 		/*
1072 		 * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1073 		 * /proc/pid/oom_score_adj instead.
1074 		 */
1075 		pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1076 			  current->comm, task_pid_nr(current), task_pid_nr(task),
1077 			  task_pid_nr(task));
1078 	} else {
1079 		if ((short)oom_adj < task->signal->oom_score_adj_min &&
1080 				!capable(CAP_SYS_RESOURCE)) {
1081 			err = -EACCES;
1082 			goto err_unlock;
1083 		}
1084 	}
1085 
1086 	/*
1087 	 * Make sure we will check other processes sharing the mm if this is
1088 	 * not vfrok which wants its own oom_score_adj.
1089 	 * pin the mm so it doesn't go away and get reused after task_unlock
1090 	 */
1091 	if (!task->vfork_done) {
1092 		struct task_struct *p = find_lock_task_mm(task);
1093 
1094 		if (p) {
1095 			if (atomic_read(&p->mm->mm_users) > 1) {
1096 				mm = p->mm;
1097 				atomic_inc(&mm->mm_count);
1098 			}
1099 			task_unlock(p);
1100 		}
1101 	}
1102 
1103 	task->signal->oom_score_adj = oom_adj;
1104 	if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1105 		task->signal->oom_score_adj_min = (short)oom_adj;
1106 	trace_oom_score_adj_update(task);
1107 
1108 	if (mm) {
1109 		struct task_struct *p;
1110 
1111 		rcu_read_lock();
1112 		for_each_process(p) {
1113 			if (same_thread_group(task, p))
1114 				continue;
1115 
1116 			/* do not touch kernel threads or the global init */
1117 			if (p->flags & PF_KTHREAD || is_global_init(p))
1118 				continue;
1119 
1120 			task_lock(p);
1121 			if (!p->vfork_done && process_shares_mm(p, mm)) {
1122 				pr_info("updating oom_score_adj for %d (%s) from %d to %d because it shares mm with %d (%s). Report if this is unexpected.\n",
1123 						task_pid_nr(p), p->comm,
1124 						p->signal->oom_score_adj, oom_adj,
1125 						task_pid_nr(task), task->comm);
1126 				p->signal->oom_score_adj = oom_adj;
1127 				if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1128 					p->signal->oom_score_adj_min = (short)oom_adj;
1129 			}
1130 			task_unlock(p);
1131 		}
1132 		rcu_read_unlock();
1133 		mmdrop(mm);
1134 	}
1135 err_unlock:
1136 	mutex_unlock(&oom_adj_mutex);
1137 	put_task_struct(task);
1138 	return err;
1139 }
1140 
1141 /*
1142  * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1143  * kernels.  The effective policy is defined by oom_score_adj, which has a
1144  * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1145  * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1146  * Processes that become oom disabled via oom_adj will still be oom disabled
1147  * with this implementation.
1148  *
1149  * oom_adj cannot be removed since existing userspace binaries use it.
1150  */
oom_adj_write(struct file * file,const char __user * buf,size_t count,loff_t * ppos)1151 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1152 			     size_t count, loff_t *ppos)
1153 {
1154 	char buffer[PROC_NUMBUF];
1155 	int oom_adj;
1156 	int err;
1157 
1158 	memset(buffer, 0, sizeof(buffer));
1159 	if (count > sizeof(buffer) - 1)
1160 		count = sizeof(buffer) - 1;
1161 	if (copy_from_user(buffer, buf, count)) {
1162 		err = -EFAULT;
1163 		goto out;
1164 	}
1165 
1166 	err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1167 	if (err)
1168 		goto out;
1169 	if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1170 	     oom_adj != OOM_DISABLE) {
1171 		err = -EINVAL;
1172 		goto out;
1173 	}
1174 
1175 	/*
1176 	 * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1177 	 * value is always attainable.
1178 	 */
1179 	if (oom_adj == OOM_ADJUST_MAX)
1180 		oom_adj = OOM_SCORE_ADJ_MAX;
1181 	else
1182 		oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1183 
1184 	err = __set_oom_adj(file, oom_adj, true);
1185 out:
1186 	return err < 0 ? err : count;
1187 }
1188 
1189 static const struct file_operations proc_oom_adj_operations = {
1190 	.read		= oom_adj_read,
1191 	.write		= oom_adj_write,
1192 	.llseek		= generic_file_llseek,
1193 };
1194 
oom_score_adj_read(struct file * file,char __user * buf,size_t count,loff_t * ppos)1195 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1196 					size_t count, loff_t *ppos)
1197 {
1198 	struct task_struct *task = get_proc_task(file_inode(file));
1199 	char buffer[PROC_NUMBUF];
1200 	short oom_score_adj = OOM_SCORE_ADJ_MIN;
1201 	size_t len;
1202 
1203 	if (!task)
1204 		return -ESRCH;
1205 	oom_score_adj = task->signal->oom_score_adj;
1206 	put_task_struct(task);
1207 	len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1208 	return simple_read_from_buffer(buf, count, ppos, buffer, len);
1209 }
1210 
oom_score_adj_write(struct file * file,const char __user * buf,size_t count,loff_t * ppos)1211 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1212 					size_t count, loff_t *ppos)
1213 {
1214 	char buffer[PROC_NUMBUF];
1215 	int oom_score_adj;
1216 	int err;
1217 
1218 	memset(buffer, 0, sizeof(buffer));
1219 	if (count > sizeof(buffer) - 1)
1220 		count = sizeof(buffer) - 1;
1221 	if (copy_from_user(buffer, buf, count)) {
1222 		err = -EFAULT;
1223 		goto out;
1224 	}
1225 
1226 	err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1227 	if (err)
1228 		goto out;
1229 	if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1230 			oom_score_adj > OOM_SCORE_ADJ_MAX) {
1231 		err = -EINVAL;
1232 		goto out;
1233 	}
1234 
1235 	err = __set_oom_adj(file, oom_score_adj, false);
1236 out:
1237 	return err < 0 ? err : count;
1238 }
1239 
1240 static const struct file_operations proc_oom_score_adj_operations = {
1241 	.read		= oom_score_adj_read,
1242 	.write		= oom_score_adj_write,
1243 	.llseek		= default_llseek,
1244 };
1245 
1246 #ifdef CONFIG_AUDITSYSCALL
1247 #define TMPBUFLEN 21
proc_loginuid_read(struct file * file,char __user * buf,size_t count,loff_t * ppos)1248 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1249 				  size_t count, loff_t *ppos)
1250 {
1251 	struct inode * inode = file_inode(file);
1252 	struct task_struct *task = get_proc_task(inode);
1253 	ssize_t length;
1254 	char tmpbuf[TMPBUFLEN];
1255 
1256 	if (!task)
1257 		return -ESRCH;
1258 	length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1259 			   from_kuid(file->f_cred->user_ns,
1260 				     audit_get_loginuid(task)));
1261 	put_task_struct(task);
1262 	return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1263 }
1264 
proc_loginuid_write(struct file * file,const char __user * buf,size_t count,loff_t * ppos)1265 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1266 				   size_t count, loff_t *ppos)
1267 {
1268 	struct inode * inode = file_inode(file);
1269 	uid_t loginuid;
1270 	kuid_t kloginuid;
1271 	int rv;
1272 
1273 	rcu_read_lock();
1274 	if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1275 		rcu_read_unlock();
1276 		return -EPERM;
1277 	}
1278 	rcu_read_unlock();
1279 
1280 	if (*ppos != 0) {
1281 		/* No partial writes. */
1282 		return -EINVAL;
1283 	}
1284 
1285 	rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1286 	if (rv < 0)
1287 		return rv;
1288 
1289 	/* is userspace tring to explicitly UNSET the loginuid? */
1290 	if (loginuid == AUDIT_UID_UNSET) {
1291 		kloginuid = INVALID_UID;
1292 	} else {
1293 		kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1294 		if (!uid_valid(kloginuid))
1295 			return -EINVAL;
1296 	}
1297 
1298 	rv = audit_set_loginuid(kloginuid);
1299 	if (rv < 0)
1300 		return rv;
1301 	return count;
1302 }
1303 
1304 static const struct file_operations proc_loginuid_operations = {
1305 	.read		= proc_loginuid_read,
1306 	.write		= proc_loginuid_write,
1307 	.llseek		= generic_file_llseek,
1308 };
1309 
proc_sessionid_read(struct file * file,char __user * buf,size_t count,loff_t * ppos)1310 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1311 				  size_t count, loff_t *ppos)
1312 {
1313 	struct inode * inode = file_inode(file);
1314 	struct task_struct *task = get_proc_task(inode);
1315 	ssize_t length;
1316 	char tmpbuf[TMPBUFLEN];
1317 
1318 	if (!task)
1319 		return -ESRCH;
1320 	length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1321 				audit_get_sessionid(task));
1322 	put_task_struct(task);
1323 	return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1324 }
1325 
1326 static const struct file_operations proc_sessionid_operations = {
1327 	.read		= proc_sessionid_read,
1328 	.llseek		= generic_file_llseek,
1329 };
1330 #endif
1331 
1332 #ifdef CONFIG_FAULT_INJECTION
proc_fault_inject_read(struct file * file,char __user * buf,size_t count,loff_t * ppos)1333 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1334 				      size_t count, loff_t *ppos)
1335 {
1336 	struct task_struct *task = get_proc_task(file_inode(file));
1337 	char buffer[PROC_NUMBUF];
1338 	size_t len;
1339 	int make_it_fail;
1340 
1341 	if (!task)
1342 		return -ESRCH;
1343 	make_it_fail = task->make_it_fail;
1344 	put_task_struct(task);
1345 
1346 	len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1347 
1348 	return simple_read_from_buffer(buf, count, ppos, buffer, len);
1349 }
1350 
proc_fault_inject_write(struct file * file,const char __user * buf,size_t count,loff_t * ppos)1351 static ssize_t proc_fault_inject_write(struct file * file,
1352 			const char __user * buf, size_t count, loff_t *ppos)
1353 {
1354 	struct task_struct *task;
1355 	char buffer[PROC_NUMBUF];
1356 	int make_it_fail;
1357 	int rv;
1358 
1359 	if (!capable(CAP_SYS_RESOURCE))
1360 		return -EPERM;
1361 	memset(buffer, 0, sizeof(buffer));
1362 	if (count > sizeof(buffer) - 1)
1363 		count = sizeof(buffer) - 1;
1364 	if (copy_from_user(buffer, buf, count))
1365 		return -EFAULT;
1366 	rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1367 	if (rv < 0)
1368 		return rv;
1369 	if (make_it_fail < 0 || make_it_fail > 1)
1370 		return -EINVAL;
1371 
1372 	task = get_proc_task(file_inode(file));
1373 	if (!task)
1374 		return -ESRCH;
1375 	task->make_it_fail = make_it_fail;
1376 	put_task_struct(task);
1377 
1378 	return count;
1379 }
1380 
1381 static const struct file_operations proc_fault_inject_operations = {
1382 	.read		= proc_fault_inject_read,
1383 	.write		= proc_fault_inject_write,
1384 	.llseek		= generic_file_llseek,
1385 };
1386 #endif
1387 
1388 
1389 #ifdef CONFIG_SCHED_DEBUG
1390 /*
1391  * Print out various scheduling related per-task fields:
1392  */
sched_show(struct seq_file * m,void * v)1393 static int sched_show(struct seq_file *m, void *v)
1394 {
1395 	struct inode *inode = m->private;
1396 	struct task_struct *p;
1397 
1398 	p = get_proc_task(inode);
1399 	if (!p)
1400 		return -ESRCH;
1401 	proc_sched_show_task(p, m);
1402 
1403 	put_task_struct(p);
1404 
1405 	return 0;
1406 }
1407 
1408 static ssize_t
sched_write(struct file * file,const char __user * buf,size_t count,loff_t * offset)1409 sched_write(struct file *file, const char __user *buf,
1410 	    size_t count, loff_t *offset)
1411 {
1412 	struct inode *inode = file_inode(file);
1413 	struct task_struct *p;
1414 
1415 	p = get_proc_task(inode);
1416 	if (!p)
1417 		return -ESRCH;
1418 	proc_sched_set_task(p);
1419 
1420 	put_task_struct(p);
1421 
1422 	return count;
1423 }
1424 
sched_open(struct inode * inode,struct file * filp)1425 static int sched_open(struct inode *inode, struct file *filp)
1426 {
1427 	return single_open(filp, sched_show, inode);
1428 }
1429 
1430 static const struct file_operations proc_pid_sched_operations = {
1431 	.open		= sched_open,
1432 	.read		= seq_read,
1433 	.write		= sched_write,
1434 	.llseek		= seq_lseek,
1435 	.release	= single_release,
1436 };
1437 
1438 #endif
1439 
1440 #ifdef CONFIG_SCHED_AUTOGROUP
1441 /*
1442  * Print out autogroup related information:
1443  */
sched_autogroup_show(struct seq_file * m,void * v)1444 static int sched_autogroup_show(struct seq_file *m, void *v)
1445 {
1446 	struct inode *inode = m->private;
1447 	struct task_struct *p;
1448 
1449 	p = get_proc_task(inode);
1450 	if (!p)
1451 		return -ESRCH;
1452 	proc_sched_autogroup_show_task(p, m);
1453 
1454 	put_task_struct(p);
1455 
1456 	return 0;
1457 }
1458 
1459 static ssize_t
sched_autogroup_write(struct file * file,const char __user * buf,size_t count,loff_t * offset)1460 sched_autogroup_write(struct file *file, const char __user *buf,
1461 	    size_t count, loff_t *offset)
1462 {
1463 	struct inode *inode = file_inode(file);
1464 	struct task_struct *p;
1465 	char buffer[PROC_NUMBUF];
1466 	int nice;
1467 	int err;
1468 
1469 	memset(buffer, 0, sizeof(buffer));
1470 	if (count > sizeof(buffer) - 1)
1471 		count = sizeof(buffer) - 1;
1472 	if (copy_from_user(buffer, buf, count))
1473 		return -EFAULT;
1474 
1475 	err = kstrtoint(strstrip(buffer), 0, &nice);
1476 	if (err < 0)
1477 		return err;
1478 
1479 	p = get_proc_task(inode);
1480 	if (!p)
1481 		return -ESRCH;
1482 
1483 	err = proc_sched_autogroup_set_nice(p, nice);
1484 	if (err)
1485 		count = err;
1486 
1487 	put_task_struct(p);
1488 
1489 	return count;
1490 }
1491 
sched_autogroup_open(struct inode * inode,struct file * filp)1492 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1493 {
1494 	int ret;
1495 
1496 	ret = single_open(filp, sched_autogroup_show, NULL);
1497 	if (!ret) {
1498 		struct seq_file *m = filp->private_data;
1499 
1500 		m->private = inode;
1501 	}
1502 	return ret;
1503 }
1504 
1505 static const struct file_operations proc_pid_sched_autogroup_operations = {
1506 	.open		= sched_autogroup_open,
1507 	.read		= seq_read,
1508 	.write		= sched_autogroup_write,
1509 	.llseek		= seq_lseek,
1510 	.release	= single_release,
1511 };
1512 
1513 #endif /* CONFIG_SCHED_AUTOGROUP */
1514 
comm_write(struct file * file,const char __user * buf,size_t count,loff_t * offset)1515 static ssize_t comm_write(struct file *file, const char __user *buf,
1516 				size_t count, loff_t *offset)
1517 {
1518 	struct inode *inode = file_inode(file);
1519 	struct task_struct *p;
1520 	char buffer[TASK_COMM_LEN];
1521 	const size_t maxlen = sizeof(buffer) - 1;
1522 
1523 	memset(buffer, 0, sizeof(buffer));
1524 	if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1525 		return -EFAULT;
1526 
1527 	p = get_proc_task(inode);
1528 	if (!p)
1529 		return -ESRCH;
1530 
1531 	if (same_thread_group(current, p))
1532 		set_task_comm(p, buffer);
1533 	else
1534 		count = -EINVAL;
1535 
1536 	put_task_struct(p);
1537 
1538 	return count;
1539 }
1540 
comm_show(struct seq_file * m,void * v)1541 static int comm_show(struct seq_file *m, void *v)
1542 {
1543 	struct inode *inode = m->private;
1544 	struct task_struct *p;
1545 
1546 	p = get_proc_task(inode);
1547 	if (!p)
1548 		return -ESRCH;
1549 
1550 	task_lock(p);
1551 	seq_printf(m, "%s\n", p->comm);
1552 	task_unlock(p);
1553 
1554 	put_task_struct(p);
1555 
1556 	return 0;
1557 }
1558 
comm_open(struct inode * inode,struct file * filp)1559 static int comm_open(struct inode *inode, struct file *filp)
1560 {
1561 	return single_open(filp, comm_show, inode);
1562 }
1563 
1564 static const struct file_operations proc_pid_set_comm_operations = {
1565 	.open		= comm_open,
1566 	.read		= seq_read,
1567 	.write		= comm_write,
1568 	.llseek		= seq_lseek,
1569 	.release	= single_release,
1570 };
1571 
proc_exe_link(struct dentry * dentry,struct path * exe_path)1572 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1573 {
1574 	struct task_struct *task;
1575 	struct file *exe_file;
1576 
1577 	task = get_proc_task(d_inode(dentry));
1578 	if (!task)
1579 		return -ENOENT;
1580 	exe_file = get_task_exe_file(task);
1581 	put_task_struct(task);
1582 	if (exe_file) {
1583 		*exe_path = exe_file->f_path;
1584 		path_get(&exe_file->f_path);
1585 		fput(exe_file);
1586 		return 0;
1587 	} else
1588 		return -ENOENT;
1589 }
1590 
proc_pid_get_link(struct dentry * dentry,struct inode * inode,struct delayed_call * done)1591 static const char *proc_pid_get_link(struct dentry *dentry,
1592 				     struct inode *inode,
1593 				     struct delayed_call *done)
1594 {
1595 	struct path path;
1596 	int error = -EACCES;
1597 
1598 	if (!dentry)
1599 		return ERR_PTR(-ECHILD);
1600 
1601 	/* Are we allowed to snoop on the tasks file descriptors? */
1602 	if (!proc_fd_access_allowed(inode))
1603 		goto out;
1604 
1605 	error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1606 	if (error)
1607 		goto out;
1608 
1609 	nd_jump_link(&path);
1610 	return NULL;
1611 out:
1612 	return ERR_PTR(error);
1613 }
1614 
do_proc_readlink(struct path * path,char __user * buffer,int buflen)1615 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1616 {
1617 	char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1618 	char *pathname;
1619 	int len;
1620 
1621 	if (!tmp)
1622 		return -ENOMEM;
1623 
1624 	pathname = d_path(path, tmp, PAGE_SIZE);
1625 	len = PTR_ERR(pathname);
1626 	if (IS_ERR(pathname))
1627 		goto out;
1628 	len = tmp + PAGE_SIZE - 1 - pathname;
1629 
1630 	if (len > buflen)
1631 		len = buflen;
1632 	if (copy_to_user(buffer, pathname, len))
1633 		len = -EFAULT;
1634  out:
1635 	free_page((unsigned long)tmp);
1636 	return len;
1637 }
1638 
proc_pid_readlink(struct dentry * dentry,char __user * buffer,int buflen)1639 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1640 {
1641 	int error = -EACCES;
1642 	struct inode *inode = d_inode(dentry);
1643 	struct path path;
1644 
1645 	/* Are we allowed to snoop on the tasks file descriptors? */
1646 	if (!proc_fd_access_allowed(inode))
1647 		goto out;
1648 
1649 	error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1650 	if (error)
1651 		goto out;
1652 
1653 	error = do_proc_readlink(&path, buffer, buflen);
1654 	path_put(&path);
1655 out:
1656 	return error;
1657 }
1658 
1659 const struct inode_operations proc_pid_link_inode_operations = {
1660 	.readlink	= proc_pid_readlink,
1661 	.get_link	= proc_pid_get_link,
1662 	.setattr	= proc_setattr,
1663 };
1664 
1665 
1666 /* building an inode */
1667 
proc_pid_make_inode(struct super_block * sb,struct task_struct * task)1668 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1669 {
1670 	struct inode * inode;
1671 	struct proc_inode *ei;
1672 	const struct cred *cred;
1673 
1674 	/* We need a new inode */
1675 
1676 	inode = new_inode(sb);
1677 	if (!inode)
1678 		goto out;
1679 
1680 	/* Common stuff */
1681 	ei = PROC_I(inode);
1682 	inode->i_ino = get_next_ino();
1683 	inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1684 	inode->i_op = &proc_def_inode_operations;
1685 
1686 	/*
1687 	 * grab the reference to task.
1688 	 */
1689 	ei->pid = get_task_pid(task, PIDTYPE_PID);
1690 	if (!ei->pid)
1691 		goto out_unlock;
1692 
1693 	if (task_dumpable(task)) {
1694 		rcu_read_lock();
1695 		cred = __task_cred(task);
1696 		inode->i_uid = cred->euid;
1697 		inode->i_gid = cred->egid;
1698 		rcu_read_unlock();
1699 	}
1700 	security_task_to_inode(task, inode);
1701 
1702 out:
1703 	return inode;
1704 
1705 out_unlock:
1706 	iput(inode);
1707 	return NULL;
1708 }
1709 
pid_getattr(struct vfsmount * mnt,struct dentry * dentry,struct kstat * stat)1710 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1711 {
1712 	struct inode *inode = d_inode(dentry);
1713 	struct task_struct *task;
1714 	const struct cred *cred;
1715 	struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1716 
1717 	generic_fillattr(inode, stat);
1718 
1719 	rcu_read_lock();
1720 	stat->uid = GLOBAL_ROOT_UID;
1721 	stat->gid = GLOBAL_ROOT_GID;
1722 	task = pid_task(proc_pid(inode), PIDTYPE_PID);
1723 	if (task) {
1724 		if (!has_pid_permissions(pid, task, 2)) {
1725 			rcu_read_unlock();
1726 			/*
1727 			 * This doesn't prevent learning whether PID exists,
1728 			 * it only makes getattr() consistent with readdir().
1729 			 */
1730 			return -ENOENT;
1731 		}
1732 		if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1733 		    task_dumpable(task)) {
1734 			cred = __task_cred(task);
1735 			stat->uid = cred->euid;
1736 			stat->gid = cred->egid;
1737 		}
1738 	}
1739 	rcu_read_unlock();
1740 	return 0;
1741 }
1742 
1743 /* dentry stuff */
1744 
1745 /*
1746  *	Exceptional case: normally we are not allowed to unhash a busy
1747  * directory. In this case, however, we can do it - no aliasing problems
1748  * due to the way we treat inodes.
1749  *
1750  * Rewrite the inode's ownerships here because the owning task may have
1751  * performed a setuid(), etc.
1752  *
1753  * Before the /proc/pid/status file was created the only way to read
1754  * the effective uid of a /process was to stat /proc/pid.  Reading
1755  * /proc/pid/status is slow enough that procps and other packages
1756  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1757  * made this apply to all per process world readable and executable
1758  * directories.
1759  */
pid_revalidate(struct dentry * dentry,unsigned int flags)1760 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1761 {
1762 	struct inode *inode;
1763 	struct task_struct *task;
1764 	const struct cred *cred;
1765 
1766 	if (flags & LOOKUP_RCU)
1767 		return -ECHILD;
1768 
1769 	inode = d_inode(dentry);
1770 	task = get_proc_task(inode);
1771 
1772 	if (task) {
1773 		if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1774 		    task_dumpable(task)) {
1775 			rcu_read_lock();
1776 			cred = __task_cred(task);
1777 			inode->i_uid = cred->euid;
1778 			inode->i_gid = cred->egid;
1779 			rcu_read_unlock();
1780 		} else {
1781 			inode->i_uid = GLOBAL_ROOT_UID;
1782 			inode->i_gid = GLOBAL_ROOT_GID;
1783 		}
1784 		inode->i_mode &= ~(S_ISUID | S_ISGID);
1785 		security_task_to_inode(task, inode);
1786 		put_task_struct(task);
1787 		return 1;
1788 	}
1789 	return 0;
1790 }
1791 
proc_inode_is_dead(struct inode * inode)1792 static inline bool proc_inode_is_dead(struct inode *inode)
1793 {
1794 	return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1795 }
1796 
pid_delete_dentry(const struct dentry * dentry)1797 int pid_delete_dentry(const struct dentry *dentry)
1798 {
1799 	/* Is the task we represent dead?
1800 	 * If so, then don't put the dentry on the lru list,
1801 	 * kill it immediately.
1802 	 */
1803 	return proc_inode_is_dead(d_inode(dentry));
1804 }
1805 
1806 const struct dentry_operations pid_dentry_operations =
1807 {
1808 	.d_revalidate	= pid_revalidate,
1809 	.d_delete	= pid_delete_dentry,
1810 };
1811 
1812 /* Lookups */
1813 
1814 /*
1815  * Fill a directory entry.
1816  *
1817  * If possible create the dcache entry and derive our inode number and
1818  * file type from dcache entry.
1819  *
1820  * Since all of the proc inode numbers are dynamically generated, the inode
1821  * numbers do not exist until the inode is cache.  This means creating the
1822  * the dcache entry in readdir is necessary to keep the inode numbers
1823  * reported by readdir in sync with the inode numbers reported
1824  * by stat.
1825  */
proc_fill_cache(struct file * file,struct dir_context * ctx,const char * name,int len,instantiate_t instantiate,struct task_struct * task,const void * ptr)1826 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1827 	const char *name, int len,
1828 	instantiate_t instantiate, struct task_struct *task, const void *ptr)
1829 {
1830 	struct dentry *child, *dir = file->f_path.dentry;
1831 	struct qstr qname = QSTR_INIT(name, len);
1832 	struct inode *inode;
1833 	unsigned type;
1834 	ino_t ino;
1835 
1836 	child = d_hash_and_lookup(dir, &qname);
1837 	if (!child) {
1838 		DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1839 		child = d_alloc_parallel(dir, &qname, &wq);
1840 		if (IS_ERR(child))
1841 			goto end_instantiate;
1842 		if (d_in_lookup(child)) {
1843 			int err = instantiate(d_inode(dir), child, task, ptr);
1844 			d_lookup_done(child);
1845 			if (err < 0) {
1846 				dput(child);
1847 				goto end_instantiate;
1848 			}
1849 		}
1850 	}
1851 	inode = d_inode(child);
1852 	ino = inode->i_ino;
1853 	type = inode->i_mode >> 12;
1854 	dput(child);
1855 	return dir_emit(ctx, name, len, ino, type);
1856 
1857 end_instantiate:
1858 	return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1859 }
1860 
1861 /*
1862  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1863  * which represent vma start and end addresses.
1864  */
dname_to_vma_addr(struct dentry * dentry,unsigned long * start,unsigned long * end)1865 static int dname_to_vma_addr(struct dentry *dentry,
1866 			     unsigned long *start, unsigned long *end)
1867 {
1868 	if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1869 		return -EINVAL;
1870 
1871 	return 0;
1872 }
1873 
map_files_d_revalidate(struct dentry * dentry,unsigned int flags)1874 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1875 {
1876 	unsigned long vm_start, vm_end;
1877 	bool exact_vma_exists = false;
1878 	struct mm_struct *mm = NULL;
1879 	struct task_struct *task;
1880 	const struct cred *cred;
1881 	struct inode *inode;
1882 	int status = 0;
1883 
1884 	if (flags & LOOKUP_RCU)
1885 		return -ECHILD;
1886 
1887 	inode = d_inode(dentry);
1888 	task = get_proc_task(inode);
1889 	if (!task)
1890 		goto out_notask;
1891 
1892 	mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1893 	if (IS_ERR_OR_NULL(mm))
1894 		goto out;
1895 
1896 	if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1897 		down_read(&mm->mmap_sem);
1898 		exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1899 		up_read(&mm->mmap_sem);
1900 	}
1901 
1902 	mmput(mm);
1903 
1904 	if (exact_vma_exists) {
1905 		if (task_dumpable(task)) {
1906 			rcu_read_lock();
1907 			cred = __task_cred(task);
1908 			inode->i_uid = cred->euid;
1909 			inode->i_gid = cred->egid;
1910 			rcu_read_unlock();
1911 		} else {
1912 			inode->i_uid = GLOBAL_ROOT_UID;
1913 			inode->i_gid = GLOBAL_ROOT_GID;
1914 		}
1915 		security_task_to_inode(task, inode);
1916 		status = 1;
1917 	}
1918 
1919 out:
1920 	put_task_struct(task);
1921 
1922 out_notask:
1923 	return status;
1924 }
1925 
1926 static const struct dentry_operations tid_map_files_dentry_operations = {
1927 	.d_revalidate	= map_files_d_revalidate,
1928 	.d_delete	= pid_delete_dentry,
1929 };
1930 
map_files_get_link(struct dentry * dentry,struct path * path)1931 static int map_files_get_link(struct dentry *dentry, struct path *path)
1932 {
1933 	unsigned long vm_start, vm_end;
1934 	struct vm_area_struct *vma;
1935 	struct task_struct *task;
1936 	struct mm_struct *mm;
1937 	int rc;
1938 
1939 	rc = -ENOENT;
1940 	task = get_proc_task(d_inode(dentry));
1941 	if (!task)
1942 		goto out;
1943 
1944 	mm = get_task_mm(task);
1945 	put_task_struct(task);
1946 	if (!mm)
1947 		goto out;
1948 
1949 	rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1950 	if (rc)
1951 		goto out_mmput;
1952 
1953 	rc = -ENOENT;
1954 	down_read(&mm->mmap_sem);
1955 	vma = find_exact_vma(mm, vm_start, vm_end);
1956 	if (vma && vma->vm_file) {
1957 		*path = vma->vm_file->f_path;
1958 		path_get(path);
1959 		rc = 0;
1960 	}
1961 	up_read(&mm->mmap_sem);
1962 
1963 out_mmput:
1964 	mmput(mm);
1965 out:
1966 	return rc;
1967 }
1968 
1969 struct map_files_info {
1970 	fmode_t		mode;
1971 	unsigned long	len;
1972 	unsigned char	name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1973 };
1974 
1975 /*
1976  * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
1977  * symlinks may be used to bypass permissions on ancestor directories in the
1978  * path to the file in question.
1979  */
1980 static const char *
proc_map_files_get_link(struct dentry * dentry,struct inode * inode,struct delayed_call * done)1981 proc_map_files_get_link(struct dentry *dentry,
1982 			struct inode *inode,
1983 		        struct delayed_call *done)
1984 {
1985 	if (!capable(CAP_SYS_ADMIN))
1986 		return ERR_PTR(-EPERM);
1987 
1988 	return proc_pid_get_link(dentry, inode, done);
1989 }
1990 
1991 /*
1992  * Identical to proc_pid_link_inode_operations except for get_link()
1993  */
1994 static const struct inode_operations proc_map_files_link_inode_operations = {
1995 	.readlink	= proc_pid_readlink,
1996 	.get_link	= proc_map_files_get_link,
1997 	.setattr	= proc_setattr,
1998 };
1999 
2000 static int
proc_map_files_instantiate(struct inode * dir,struct dentry * dentry,struct task_struct * task,const void * ptr)2001 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
2002 			   struct task_struct *task, const void *ptr)
2003 {
2004 	fmode_t mode = (fmode_t)(unsigned long)ptr;
2005 	struct proc_inode *ei;
2006 	struct inode *inode;
2007 
2008 	inode = proc_pid_make_inode(dir->i_sb, task);
2009 	if (!inode)
2010 		return -ENOENT;
2011 
2012 	ei = PROC_I(inode);
2013 	ei->op.proc_get_link = map_files_get_link;
2014 
2015 	inode->i_op = &proc_map_files_link_inode_operations;
2016 	inode->i_size = 64;
2017 	inode->i_mode = S_IFLNK;
2018 
2019 	if (mode & FMODE_READ)
2020 		inode->i_mode |= S_IRUSR;
2021 	if (mode & FMODE_WRITE)
2022 		inode->i_mode |= S_IWUSR;
2023 
2024 	d_set_d_op(dentry, &tid_map_files_dentry_operations);
2025 	d_add(dentry, inode);
2026 
2027 	return 0;
2028 }
2029 
proc_map_files_lookup(struct inode * dir,struct dentry * dentry,unsigned int flags)2030 static struct dentry *proc_map_files_lookup(struct inode *dir,
2031 		struct dentry *dentry, unsigned int flags)
2032 {
2033 	unsigned long vm_start, vm_end;
2034 	struct vm_area_struct *vma;
2035 	struct task_struct *task;
2036 	int result;
2037 	struct mm_struct *mm;
2038 
2039 	result = -ENOENT;
2040 	task = get_proc_task(dir);
2041 	if (!task)
2042 		goto out;
2043 
2044 	result = -EACCES;
2045 	if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2046 		goto out_put_task;
2047 
2048 	result = -ENOENT;
2049 	if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2050 		goto out_put_task;
2051 
2052 	mm = get_task_mm(task);
2053 	if (!mm)
2054 		goto out_put_task;
2055 
2056 	down_read(&mm->mmap_sem);
2057 	vma = find_exact_vma(mm, vm_start, vm_end);
2058 	if (!vma)
2059 		goto out_no_vma;
2060 
2061 	if (vma->vm_file)
2062 		result = proc_map_files_instantiate(dir, dentry, task,
2063 				(void *)(unsigned long)vma->vm_file->f_mode);
2064 
2065 out_no_vma:
2066 	up_read(&mm->mmap_sem);
2067 	mmput(mm);
2068 out_put_task:
2069 	put_task_struct(task);
2070 out:
2071 	return ERR_PTR(result);
2072 }
2073 
2074 static const struct inode_operations proc_map_files_inode_operations = {
2075 	.lookup		= proc_map_files_lookup,
2076 	.permission	= proc_fd_permission,
2077 	.setattr	= proc_setattr,
2078 };
2079 
2080 static int
proc_map_files_readdir(struct file * file,struct dir_context * ctx)2081 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2082 {
2083 	struct vm_area_struct *vma;
2084 	struct task_struct *task;
2085 	struct mm_struct *mm;
2086 	unsigned long nr_files, pos, i;
2087 	struct flex_array *fa = NULL;
2088 	struct map_files_info info;
2089 	struct map_files_info *p;
2090 	int ret;
2091 
2092 	ret = -ENOENT;
2093 	task = get_proc_task(file_inode(file));
2094 	if (!task)
2095 		goto out;
2096 
2097 	ret = -EACCES;
2098 	if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2099 		goto out_put_task;
2100 
2101 	ret = 0;
2102 	if (!dir_emit_dots(file, ctx))
2103 		goto out_put_task;
2104 
2105 	mm = get_task_mm(task);
2106 	if (!mm)
2107 		goto out_put_task;
2108 	down_read(&mm->mmap_sem);
2109 
2110 	nr_files = 0;
2111 
2112 	/*
2113 	 * We need two passes here:
2114 	 *
2115 	 *  1) Collect vmas of mapped files with mmap_sem taken
2116 	 *  2) Release mmap_sem and instantiate entries
2117 	 *
2118 	 * otherwise we get lockdep complained, since filldir()
2119 	 * routine might require mmap_sem taken in might_fault().
2120 	 */
2121 
2122 	for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2123 		if (vma->vm_file && ++pos > ctx->pos)
2124 			nr_files++;
2125 	}
2126 
2127 	if (nr_files) {
2128 		fa = flex_array_alloc(sizeof(info), nr_files,
2129 					GFP_KERNEL);
2130 		if (!fa || flex_array_prealloc(fa, 0, nr_files,
2131 						GFP_KERNEL)) {
2132 			ret = -ENOMEM;
2133 			if (fa)
2134 				flex_array_free(fa);
2135 			up_read(&mm->mmap_sem);
2136 			mmput(mm);
2137 			goto out_put_task;
2138 		}
2139 		for (i = 0, vma = mm->mmap, pos = 2; vma;
2140 				vma = vma->vm_next) {
2141 			if (!vma->vm_file)
2142 				continue;
2143 			if (++pos <= ctx->pos)
2144 				continue;
2145 
2146 			info.mode = vma->vm_file->f_mode;
2147 			info.len = snprintf(info.name,
2148 					sizeof(info.name), "%lx-%lx",
2149 					vma->vm_start, vma->vm_end);
2150 			if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2151 				BUG();
2152 		}
2153 	}
2154 	up_read(&mm->mmap_sem);
2155 
2156 	for (i = 0; i < nr_files; i++) {
2157 		p = flex_array_get(fa, i);
2158 		if (!proc_fill_cache(file, ctx,
2159 				      p->name, p->len,
2160 				      proc_map_files_instantiate,
2161 				      task,
2162 				      (void *)(unsigned long)p->mode))
2163 			break;
2164 		ctx->pos++;
2165 	}
2166 	if (fa)
2167 		flex_array_free(fa);
2168 	mmput(mm);
2169 
2170 out_put_task:
2171 	put_task_struct(task);
2172 out:
2173 	return ret;
2174 }
2175 
2176 static const struct file_operations proc_map_files_operations = {
2177 	.read		= generic_read_dir,
2178 	.iterate_shared	= proc_map_files_readdir,
2179 	.llseek		= generic_file_llseek,
2180 };
2181 
2182 #ifdef CONFIG_CHECKPOINT_RESTORE
2183 struct timers_private {
2184 	struct pid *pid;
2185 	struct task_struct *task;
2186 	struct sighand_struct *sighand;
2187 	struct pid_namespace *ns;
2188 	unsigned long flags;
2189 };
2190 
timers_start(struct seq_file * m,loff_t * pos)2191 static void *timers_start(struct seq_file *m, loff_t *pos)
2192 {
2193 	struct timers_private *tp = m->private;
2194 
2195 	tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2196 	if (!tp->task)
2197 		return ERR_PTR(-ESRCH);
2198 
2199 	tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2200 	if (!tp->sighand)
2201 		return ERR_PTR(-ESRCH);
2202 
2203 	return seq_list_start(&tp->task->signal->posix_timers, *pos);
2204 }
2205 
timers_next(struct seq_file * m,void * v,loff_t * pos)2206 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2207 {
2208 	struct timers_private *tp = m->private;
2209 	return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2210 }
2211 
timers_stop(struct seq_file * m,void * v)2212 static void timers_stop(struct seq_file *m, void *v)
2213 {
2214 	struct timers_private *tp = m->private;
2215 
2216 	if (tp->sighand) {
2217 		unlock_task_sighand(tp->task, &tp->flags);
2218 		tp->sighand = NULL;
2219 	}
2220 
2221 	if (tp->task) {
2222 		put_task_struct(tp->task);
2223 		tp->task = NULL;
2224 	}
2225 }
2226 
show_timer(struct seq_file * m,void * v)2227 static int show_timer(struct seq_file *m, void *v)
2228 {
2229 	struct k_itimer *timer;
2230 	struct timers_private *tp = m->private;
2231 	int notify;
2232 	static const char * const nstr[] = {
2233 		[SIGEV_SIGNAL] = "signal",
2234 		[SIGEV_NONE] = "none",
2235 		[SIGEV_THREAD] = "thread",
2236 	};
2237 
2238 	timer = list_entry((struct list_head *)v, struct k_itimer, list);
2239 	notify = timer->it_sigev_notify;
2240 
2241 	seq_printf(m, "ID: %d\n", timer->it_id);
2242 	seq_printf(m, "signal: %d/%p\n",
2243 		   timer->sigq->info.si_signo,
2244 		   timer->sigq->info.si_value.sival_ptr);
2245 	seq_printf(m, "notify: %s/%s.%d\n",
2246 		   nstr[notify & ~SIGEV_THREAD_ID],
2247 		   (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2248 		   pid_nr_ns(timer->it_pid, tp->ns));
2249 	seq_printf(m, "ClockID: %d\n", timer->it_clock);
2250 
2251 	return 0;
2252 }
2253 
2254 static const struct seq_operations proc_timers_seq_ops = {
2255 	.start	= timers_start,
2256 	.next	= timers_next,
2257 	.stop	= timers_stop,
2258 	.show	= show_timer,
2259 };
2260 
proc_timers_open(struct inode * inode,struct file * file)2261 static int proc_timers_open(struct inode *inode, struct file *file)
2262 {
2263 	struct timers_private *tp;
2264 
2265 	tp = __seq_open_private(file, &proc_timers_seq_ops,
2266 			sizeof(struct timers_private));
2267 	if (!tp)
2268 		return -ENOMEM;
2269 
2270 	tp->pid = proc_pid(inode);
2271 	tp->ns = inode->i_sb->s_fs_info;
2272 	return 0;
2273 }
2274 
2275 static const struct file_operations proc_timers_operations = {
2276 	.open		= proc_timers_open,
2277 	.read		= seq_read,
2278 	.llseek		= seq_lseek,
2279 	.release	= seq_release_private,
2280 };
2281 #endif
2282 
timerslack_ns_write(struct file * file,const char __user * buf,size_t count,loff_t * offset)2283 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2284 					size_t count, loff_t *offset)
2285 {
2286 	struct inode *inode = file_inode(file);
2287 	struct task_struct *p;
2288 	u64 slack_ns;
2289 	int err;
2290 
2291 	err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2292 	if (err < 0)
2293 		return err;
2294 
2295 	p = get_proc_task(inode);
2296 	if (!p)
2297 		return -ESRCH;
2298 
2299 	if (p != current) {
2300 		if (!capable(CAP_SYS_NICE)) {
2301 			count = -EPERM;
2302 			goto out;
2303 		}
2304 
2305 		err = security_task_setscheduler(p);
2306 		if (err) {
2307 			count = err;
2308 			goto out;
2309 		}
2310 	}
2311 
2312 	task_lock(p);
2313 	if (slack_ns == 0)
2314 		p->timer_slack_ns = p->default_timer_slack_ns;
2315 	else
2316 		p->timer_slack_ns = slack_ns;
2317 	task_unlock(p);
2318 
2319 out:
2320 	put_task_struct(p);
2321 
2322 	return count;
2323 }
2324 
timerslack_ns_show(struct seq_file * m,void * v)2325 static int timerslack_ns_show(struct seq_file *m, void *v)
2326 {
2327 	struct inode *inode = m->private;
2328 	struct task_struct *p;
2329 	int err = 0;
2330 
2331 	p = get_proc_task(inode);
2332 	if (!p)
2333 		return -ESRCH;
2334 
2335 	if (p != current) {
2336 
2337 		if (!capable(CAP_SYS_NICE)) {
2338 			err = -EPERM;
2339 			goto out;
2340 		}
2341 		err = security_task_getscheduler(p);
2342 		if (err)
2343 			goto out;
2344 	}
2345 
2346 	task_lock(p);
2347 	seq_printf(m, "%llu\n", p->timer_slack_ns);
2348 	task_unlock(p);
2349 
2350 out:
2351 	put_task_struct(p);
2352 
2353 	return err;
2354 }
2355 
timerslack_ns_open(struct inode * inode,struct file * filp)2356 static int timerslack_ns_open(struct inode *inode, struct file *filp)
2357 {
2358 	return single_open(filp, timerslack_ns_show, inode);
2359 }
2360 
2361 static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2362 	.open		= timerslack_ns_open,
2363 	.read		= seq_read,
2364 	.write		= timerslack_ns_write,
2365 	.llseek		= seq_lseek,
2366 	.release	= single_release,
2367 };
2368 
proc_pident_instantiate(struct inode * dir,struct dentry * dentry,struct task_struct * task,const void * ptr)2369 static int proc_pident_instantiate(struct inode *dir,
2370 	struct dentry *dentry, struct task_struct *task, const void *ptr)
2371 {
2372 	const struct pid_entry *p = ptr;
2373 	struct inode *inode;
2374 	struct proc_inode *ei;
2375 
2376 	inode = proc_pid_make_inode(dir->i_sb, task);
2377 	if (!inode)
2378 		goto out;
2379 
2380 	ei = PROC_I(inode);
2381 	inode->i_mode = p->mode;
2382 	if (S_ISDIR(inode->i_mode))
2383 		set_nlink(inode, 2);	/* Use getattr to fix if necessary */
2384 	if (p->iop)
2385 		inode->i_op = p->iop;
2386 	if (p->fop)
2387 		inode->i_fop = p->fop;
2388 	ei->op = p->op;
2389 	d_set_d_op(dentry, &pid_dentry_operations);
2390 	d_add(dentry, inode);
2391 	/* Close the race of the process dying before we return the dentry */
2392 	if (pid_revalidate(dentry, 0))
2393 		return 0;
2394 out:
2395 	return -ENOENT;
2396 }
2397 
proc_pident_lookup(struct inode * dir,struct dentry * dentry,const struct pid_entry * ents,unsigned int nents)2398 static struct dentry *proc_pident_lookup(struct inode *dir,
2399 					 struct dentry *dentry,
2400 					 const struct pid_entry *ents,
2401 					 unsigned int nents)
2402 {
2403 	int error;
2404 	struct task_struct *task = get_proc_task(dir);
2405 	const struct pid_entry *p, *last;
2406 
2407 	error = -ENOENT;
2408 
2409 	if (!task)
2410 		goto out_no_task;
2411 
2412 	/*
2413 	 * Yes, it does not scale. And it should not. Don't add
2414 	 * new entries into /proc/<tgid>/ without very good reasons.
2415 	 */
2416 	last = &ents[nents - 1];
2417 	for (p = ents; p <= last; p++) {
2418 		if (p->len != dentry->d_name.len)
2419 			continue;
2420 		if (!memcmp(dentry->d_name.name, p->name, p->len))
2421 			break;
2422 	}
2423 	if (p > last)
2424 		goto out;
2425 
2426 	error = proc_pident_instantiate(dir, dentry, task, p);
2427 out:
2428 	put_task_struct(task);
2429 out_no_task:
2430 	return ERR_PTR(error);
2431 }
2432 
proc_pident_readdir(struct file * file,struct dir_context * ctx,const struct pid_entry * ents,unsigned int nents)2433 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2434 		const struct pid_entry *ents, unsigned int nents)
2435 {
2436 	struct task_struct *task = get_proc_task(file_inode(file));
2437 	const struct pid_entry *p;
2438 
2439 	if (!task)
2440 		return -ENOENT;
2441 
2442 	if (!dir_emit_dots(file, ctx))
2443 		goto out;
2444 
2445 	if (ctx->pos >= nents + 2)
2446 		goto out;
2447 
2448 	for (p = ents + (ctx->pos - 2); p <= ents + nents - 1; p++) {
2449 		if (!proc_fill_cache(file, ctx, p->name, p->len,
2450 				proc_pident_instantiate, task, p))
2451 			break;
2452 		ctx->pos++;
2453 	}
2454 out:
2455 	put_task_struct(task);
2456 	return 0;
2457 }
2458 
2459 #ifdef CONFIG_SECURITY
proc_pid_attr_read(struct file * file,char __user * buf,size_t count,loff_t * ppos)2460 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2461 				  size_t count, loff_t *ppos)
2462 {
2463 	struct inode * inode = file_inode(file);
2464 	char *p = NULL;
2465 	ssize_t length;
2466 	struct task_struct *task = get_proc_task(inode);
2467 
2468 	if (!task)
2469 		return -ESRCH;
2470 
2471 	length = security_getprocattr(task,
2472 				      (char*)file->f_path.dentry->d_name.name,
2473 				      &p);
2474 	put_task_struct(task);
2475 	if (length > 0)
2476 		length = simple_read_from_buffer(buf, count, ppos, p, length);
2477 	kfree(p);
2478 	return length;
2479 }
2480 
proc_pid_attr_write(struct file * file,const char __user * buf,size_t count,loff_t * ppos)2481 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2482 				   size_t count, loff_t *ppos)
2483 {
2484 	struct inode * inode = file_inode(file);
2485 	void *page;
2486 	ssize_t length;
2487 	struct task_struct *task = get_proc_task(inode);
2488 
2489 	length = -ESRCH;
2490 	if (!task)
2491 		goto out_no_task;
2492 	if (count > PAGE_SIZE)
2493 		count = PAGE_SIZE;
2494 
2495 	/* No partial writes. */
2496 	length = -EINVAL;
2497 	if (*ppos != 0)
2498 		goto out;
2499 
2500 	page = memdup_user(buf, count);
2501 	if (IS_ERR(page)) {
2502 		length = PTR_ERR(page);
2503 		goto out;
2504 	}
2505 
2506 	/* Guard against adverse ptrace interaction */
2507 	length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2508 	if (length < 0)
2509 		goto out_free;
2510 
2511 	length = security_setprocattr(task,
2512 				      (char*)file->f_path.dentry->d_name.name,
2513 				      page, count);
2514 	mutex_unlock(&task->signal->cred_guard_mutex);
2515 out_free:
2516 	kfree(page);
2517 out:
2518 	put_task_struct(task);
2519 out_no_task:
2520 	return length;
2521 }
2522 
2523 static const struct file_operations proc_pid_attr_operations = {
2524 	.read		= proc_pid_attr_read,
2525 	.write		= proc_pid_attr_write,
2526 	.llseek		= generic_file_llseek,
2527 };
2528 
2529 static const struct pid_entry attr_dir_stuff[] = {
2530 	REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2531 	REG("prev",       S_IRUGO,	   proc_pid_attr_operations),
2532 	REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2533 	REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2534 	REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2535 	REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2536 };
2537 
proc_attr_dir_readdir(struct file * file,struct dir_context * ctx)2538 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2539 {
2540 	return proc_pident_readdir(file, ctx,
2541 				   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2542 }
2543 
2544 static const struct file_operations proc_attr_dir_operations = {
2545 	.read		= generic_read_dir,
2546 	.iterate_shared	= proc_attr_dir_readdir,
2547 	.llseek		= generic_file_llseek,
2548 };
2549 
proc_attr_dir_lookup(struct inode * dir,struct dentry * dentry,unsigned int flags)2550 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2551 				struct dentry *dentry, unsigned int flags)
2552 {
2553 	return proc_pident_lookup(dir, dentry,
2554 				  attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2555 }
2556 
2557 static const struct inode_operations proc_attr_dir_inode_operations = {
2558 	.lookup		= proc_attr_dir_lookup,
2559 	.getattr	= pid_getattr,
2560 	.setattr	= proc_setattr,
2561 };
2562 
2563 #endif
2564 
2565 #ifdef CONFIG_ELF_CORE
proc_coredump_filter_read(struct file * file,char __user * buf,size_t count,loff_t * ppos)2566 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2567 					 size_t count, loff_t *ppos)
2568 {
2569 	struct task_struct *task = get_proc_task(file_inode(file));
2570 	struct mm_struct *mm;
2571 	char buffer[PROC_NUMBUF];
2572 	size_t len;
2573 	int ret;
2574 
2575 	if (!task)
2576 		return -ESRCH;
2577 
2578 	ret = 0;
2579 	mm = get_task_mm(task);
2580 	if (mm) {
2581 		len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2582 			       ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2583 				MMF_DUMP_FILTER_SHIFT));
2584 		mmput(mm);
2585 		ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2586 	}
2587 
2588 	put_task_struct(task);
2589 
2590 	return ret;
2591 }
2592 
proc_coredump_filter_write(struct file * file,const char __user * buf,size_t count,loff_t * ppos)2593 static ssize_t proc_coredump_filter_write(struct file *file,
2594 					  const char __user *buf,
2595 					  size_t count,
2596 					  loff_t *ppos)
2597 {
2598 	struct task_struct *task;
2599 	struct mm_struct *mm;
2600 	unsigned int val;
2601 	int ret;
2602 	int i;
2603 	unsigned long mask;
2604 
2605 	ret = kstrtouint_from_user(buf, count, 0, &val);
2606 	if (ret < 0)
2607 		return ret;
2608 
2609 	ret = -ESRCH;
2610 	task = get_proc_task(file_inode(file));
2611 	if (!task)
2612 		goto out_no_task;
2613 
2614 	mm = get_task_mm(task);
2615 	if (!mm)
2616 		goto out_no_mm;
2617 	ret = 0;
2618 
2619 	for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2620 		if (val & mask)
2621 			set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2622 		else
2623 			clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2624 	}
2625 
2626 	mmput(mm);
2627  out_no_mm:
2628 	put_task_struct(task);
2629  out_no_task:
2630 	if (ret < 0)
2631 		return ret;
2632 	return count;
2633 }
2634 
2635 static const struct file_operations proc_coredump_filter_operations = {
2636 	.read		= proc_coredump_filter_read,
2637 	.write		= proc_coredump_filter_write,
2638 	.llseek		= generic_file_llseek,
2639 };
2640 #endif
2641 
2642 #ifdef CONFIG_TASK_IO_ACCOUNTING
do_io_accounting(struct task_struct * task,struct seq_file * m,int whole)2643 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2644 {
2645 	struct task_io_accounting acct = task->ioac;
2646 	unsigned long flags;
2647 	int result;
2648 
2649 	result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2650 	if (result)
2651 		return result;
2652 
2653 	if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2654 		result = -EACCES;
2655 		goto out_unlock;
2656 	}
2657 
2658 	if (whole && lock_task_sighand(task, &flags)) {
2659 		struct task_struct *t = task;
2660 
2661 		task_io_accounting_add(&acct, &task->signal->ioac);
2662 		while_each_thread(task, t)
2663 			task_io_accounting_add(&acct, &t->ioac);
2664 
2665 		unlock_task_sighand(task, &flags);
2666 	}
2667 	seq_printf(m,
2668 		   "rchar: %llu\n"
2669 		   "wchar: %llu\n"
2670 		   "syscr: %llu\n"
2671 		   "syscw: %llu\n"
2672 		   "read_bytes: %llu\n"
2673 		   "write_bytes: %llu\n"
2674 		   "cancelled_write_bytes: %llu\n",
2675 		   (unsigned long long)acct.rchar,
2676 		   (unsigned long long)acct.wchar,
2677 		   (unsigned long long)acct.syscr,
2678 		   (unsigned long long)acct.syscw,
2679 		   (unsigned long long)acct.read_bytes,
2680 		   (unsigned long long)acct.write_bytes,
2681 		   (unsigned long long)acct.cancelled_write_bytes);
2682 	result = 0;
2683 
2684 out_unlock:
2685 	mutex_unlock(&task->signal->cred_guard_mutex);
2686 	return result;
2687 }
2688 
proc_tid_io_accounting(struct seq_file * m,struct pid_namespace * ns,struct pid * pid,struct task_struct * task)2689 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2690 				  struct pid *pid, struct task_struct *task)
2691 {
2692 	return do_io_accounting(task, m, 0);
2693 }
2694 
proc_tgid_io_accounting(struct seq_file * m,struct pid_namespace * ns,struct pid * pid,struct task_struct * task)2695 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2696 				   struct pid *pid, struct task_struct *task)
2697 {
2698 	return do_io_accounting(task, m, 1);
2699 }
2700 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2701 
2702 #ifdef CONFIG_USER_NS
proc_id_map_open(struct inode * inode,struct file * file,const struct seq_operations * seq_ops)2703 static int proc_id_map_open(struct inode *inode, struct file *file,
2704 	const struct seq_operations *seq_ops)
2705 {
2706 	struct user_namespace *ns = NULL;
2707 	struct task_struct *task;
2708 	struct seq_file *seq;
2709 	int ret = -EINVAL;
2710 
2711 	task = get_proc_task(inode);
2712 	if (task) {
2713 		rcu_read_lock();
2714 		ns = get_user_ns(task_cred_xxx(task, user_ns));
2715 		rcu_read_unlock();
2716 		put_task_struct(task);
2717 	}
2718 	if (!ns)
2719 		goto err;
2720 
2721 	ret = seq_open(file, seq_ops);
2722 	if (ret)
2723 		goto err_put_ns;
2724 
2725 	seq = file->private_data;
2726 	seq->private = ns;
2727 
2728 	return 0;
2729 err_put_ns:
2730 	put_user_ns(ns);
2731 err:
2732 	return ret;
2733 }
2734 
proc_id_map_release(struct inode * inode,struct file * file)2735 static int proc_id_map_release(struct inode *inode, struct file *file)
2736 {
2737 	struct seq_file *seq = file->private_data;
2738 	struct user_namespace *ns = seq->private;
2739 	put_user_ns(ns);
2740 	return seq_release(inode, file);
2741 }
2742 
proc_uid_map_open(struct inode * inode,struct file * file)2743 static int proc_uid_map_open(struct inode *inode, struct file *file)
2744 {
2745 	return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2746 }
2747 
proc_gid_map_open(struct inode * inode,struct file * file)2748 static int proc_gid_map_open(struct inode *inode, struct file *file)
2749 {
2750 	return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2751 }
2752 
proc_projid_map_open(struct inode * inode,struct file * file)2753 static int proc_projid_map_open(struct inode *inode, struct file *file)
2754 {
2755 	return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2756 }
2757 
2758 static const struct file_operations proc_uid_map_operations = {
2759 	.open		= proc_uid_map_open,
2760 	.write		= proc_uid_map_write,
2761 	.read		= seq_read,
2762 	.llseek		= seq_lseek,
2763 	.release	= proc_id_map_release,
2764 };
2765 
2766 static const struct file_operations proc_gid_map_operations = {
2767 	.open		= proc_gid_map_open,
2768 	.write		= proc_gid_map_write,
2769 	.read		= seq_read,
2770 	.llseek		= seq_lseek,
2771 	.release	= proc_id_map_release,
2772 };
2773 
2774 static const struct file_operations proc_projid_map_operations = {
2775 	.open		= proc_projid_map_open,
2776 	.write		= proc_projid_map_write,
2777 	.read		= seq_read,
2778 	.llseek		= seq_lseek,
2779 	.release	= proc_id_map_release,
2780 };
2781 
proc_setgroups_open(struct inode * inode,struct file * file)2782 static int proc_setgroups_open(struct inode *inode, struct file *file)
2783 {
2784 	struct user_namespace *ns = NULL;
2785 	struct task_struct *task;
2786 	int ret;
2787 
2788 	ret = -ESRCH;
2789 	task = get_proc_task(inode);
2790 	if (task) {
2791 		rcu_read_lock();
2792 		ns = get_user_ns(task_cred_xxx(task, user_ns));
2793 		rcu_read_unlock();
2794 		put_task_struct(task);
2795 	}
2796 	if (!ns)
2797 		goto err;
2798 
2799 	if (file->f_mode & FMODE_WRITE) {
2800 		ret = -EACCES;
2801 		if (!ns_capable(ns, CAP_SYS_ADMIN))
2802 			goto err_put_ns;
2803 	}
2804 
2805 	ret = single_open(file, &proc_setgroups_show, ns);
2806 	if (ret)
2807 		goto err_put_ns;
2808 
2809 	return 0;
2810 err_put_ns:
2811 	put_user_ns(ns);
2812 err:
2813 	return ret;
2814 }
2815 
proc_setgroups_release(struct inode * inode,struct file * file)2816 static int proc_setgroups_release(struct inode *inode, struct file *file)
2817 {
2818 	struct seq_file *seq = file->private_data;
2819 	struct user_namespace *ns = seq->private;
2820 	int ret = single_release(inode, file);
2821 	put_user_ns(ns);
2822 	return ret;
2823 }
2824 
2825 static const struct file_operations proc_setgroups_operations = {
2826 	.open		= proc_setgroups_open,
2827 	.write		= proc_setgroups_write,
2828 	.read		= seq_read,
2829 	.llseek		= seq_lseek,
2830 	.release	= proc_setgroups_release,
2831 };
2832 #endif /* CONFIG_USER_NS */
2833 
proc_pid_personality(struct seq_file * m,struct pid_namespace * ns,struct pid * pid,struct task_struct * task)2834 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2835 				struct pid *pid, struct task_struct *task)
2836 {
2837 	int err = lock_trace(task);
2838 	if (!err) {
2839 		seq_printf(m, "%08x\n", task->personality);
2840 		unlock_trace(task);
2841 	}
2842 	return err;
2843 }
2844 
2845 /*
2846  * Thread groups
2847  */
2848 static const struct file_operations proc_task_operations;
2849 static const struct inode_operations proc_task_inode_operations;
2850 
2851 static const struct pid_entry tgid_base_stuff[] = {
2852 	DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2853 	DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2854 	DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2855 	DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2856 	DIR("ns",	  S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2857 #ifdef CONFIG_NET
2858 	DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2859 #endif
2860 	REG("environ",    S_IRUSR, proc_environ_operations),
2861 	REG("auxv",       S_IRUSR, proc_auxv_operations),
2862 	ONE("status",     S_IRUGO, proc_pid_status),
2863 	ONE("personality", S_IRUSR, proc_pid_personality),
2864 	ONE("limits",	  S_IRUGO, proc_pid_limits),
2865 #ifdef CONFIG_SCHED_DEBUG
2866 	REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2867 #endif
2868 #ifdef CONFIG_SCHED_AUTOGROUP
2869 	REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2870 #endif
2871 	REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2872 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2873 	ONE("syscall",    S_IRUSR, proc_pid_syscall),
2874 #endif
2875 	REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
2876 	ONE("stat",       S_IRUGO, proc_tgid_stat),
2877 	ONE("statm",      S_IRUGO, proc_pid_statm),
2878 	REG("maps",       S_IRUGO, proc_pid_maps_operations),
2879 #ifdef CONFIG_NUMA
2880 	REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2881 #endif
2882 	REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2883 	LNK("cwd",        proc_cwd_link),
2884 	LNK("root",       proc_root_link),
2885 	LNK("exe",        proc_exe_link),
2886 	REG("mounts",     S_IRUGO, proc_mounts_operations),
2887 	REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2888 	REG("mountstats", S_IRUSR, proc_mountstats_operations),
2889 #ifdef CONFIG_PROC_PAGE_MONITOR
2890 	REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2891 	REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2892 	REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2893 #endif
2894 #ifdef CONFIG_SECURITY
2895 	DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2896 #endif
2897 #ifdef CONFIG_KALLSYMS
2898 	ONE("wchan",      S_IRUGO, proc_pid_wchan),
2899 #endif
2900 #ifdef CONFIG_STACKTRACE
2901 	ONE("stack",      S_IRUSR, proc_pid_stack),
2902 #endif
2903 #ifdef CONFIG_SCHED_INFO
2904 	ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
2905 #endif
2906 #ifdef CONFIG_LATENCYTOP
2907 	REG("latency",  S_IRUGO, proc_lstats_operations),
2908 #endif
2909 #ifdef CONFIG_PROC_PID_CPUSET
2910 	ONE("cpuset",     S_IRUGO, proc_cpuset_show),
2911 #endif
2912 #ifdef CONFIG_CGROUPS
2913 	ONE("cgroup",  S_IRUGO, proc_cgroup_show),
2914 #endif
2915 	ONE("oom_score",  S_IRUGO, proc_oom_score),
2916 	REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2917 	REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2918 #ifdef CONFIG_AUDITSYSCALL
2919 	REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2920 	REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2921 #endif
2922 #ifdef CONFIG_FAULT_INJECTION
2923 	REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2924 #endif
2925 #ifdef CONFIG_ELF_CORE
2926 	REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2927 #endif
2928 #ifdef CONFIG_TASK_IO_ACCOUNTING
2929 	ONE("io",	S_IRUSR, proc_tgid_io_accounting),
2930 #endif
2931 #ifdef CONFIG_HARDWALL
2932 	ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
2933 #endif
2934 #ifdef CONFIG_USER_NS
2935 	REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2936 	REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2937 	REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2938 	REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
2939 #endif
2940 #ifdef CONFIG_CHECKPOINT_RESTORE
2941 	REG("timers",	  S_IRUGO, proc_timers_operations),
2942 #endif
2943 	REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
2944 #ifdef CONFIG_CPU_FREQ_TIMES
2945 	ONE("time_in_state", 0444, proc_time_in_state_show),
2946 #endif
2947 };
2948 
proc_tgid_base_readdir(struct file * file,struct dir_context * ctx)2949 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2950 {
2951 	return proc_pident_readdir(file, ctx,
2952 				   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2953 }
2954 
2955 static const struct file_operations proc_tgid_base_operations = {
2956 	.read		= generic_read_dir,
2957 	.iterate_shared	= proc_tgid_base_readdir,
2958 	.llseek		= generic_file_llseek,
2959 };
2960 
proc_tgid_base_lookup(struct inode * dir,struct dentry * dentry,unsigned int flags)2961 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2962 {
2963 	return proc_pident_lookup(dir, dentry,
2964 				  tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2965 }
2966 
2967 static const struct inode_operations proc_tgid_base_inode_operations = {
2968 	.lookup		= proc_tgid_base_lookup,
2969 	.getattr	= pid_getattr,
2970 	.setattr	= proc_setattr,
2971 	.permission	= proc_pid_permission,
2972 };
2973 
proc_flush_task_mnt(struct vfsmount * mnt,pid_t pid,pid_t tgid)2974 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2975 {
2976 	struct dentry *dentry, *leader, *dir;
2977 	char buf[PROC_NUMBUF];
2978 	struct qstr name;
2979 
2980 	name.name = buf;
2981 	name.len = snprintf(buf, sizeof(buf), "%d", pid);
2982 	/* no ->d_hash() rejects on procfs */
2983 	dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2984 	if (dentry) {
2985 		d_invalidate(dentry);
2986 		dput(dentry);
2987 	}
2988 
2989 	if (pid == tgid)
2990 		return;
2991 
2992 	name.name = buf;
2993 	name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2994 	leader = d_hash_and_lookup(mnt->mnt_root, &name);
2995 	if (!leader)
2996 		goto out;
2997 
2998 	name.name = "task";
2999 	name.len = strlen(name.name);
3000 	dir = d_hash_and_lookup(leader, &name);
3001 	if (!dir)
3002 		goto out_put_leader;
3003 
3004 	name.name = buf;
3005 	name.len = snprintf(buf, sizeof(buf), "%d", pid);
3006 	dentry = d_hash_and_lookup(dir, &name);
3007 	if (dentry) {
3008 		d_invalidate(dentry);
3009 		dput(dentry);
3010 	}
3011 
3012 	dput(dir);
3013 out_put_leader:
3014 	dput(leader);
3015 out:
3016 	return;
3017 }
3018 
3019 /**
3020  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
3021  * @task: task that should be flushed.
3022  *
3023  * When flushing dentries from proc, one needs to flush them from global
3024  * proc (proc_mnt) and from all the namespaces' procs this task was seen
3025  * in. This call is supposed to do all of this job.
3026  *
3027  * Looks in the dcache for
3028  * /proc/@pid
3029  * /proc/@tgid/task/@pid
3030  * if either directory is present flushes it and all of it'ts children
3031  * from the dcache.
3032  *
3033  * It is safe and reasonable to cache /proc entries for a task until
3034  * that task exits.  After that they just clog up the dcache with
3035  * useless entries, possibly causing useful dcache entries to be
3036  * flushed instead.  This routine is proved to flush those useless
3037  * dcache entries at process exit time.
3038  *
3039  * NOTE: This routine is just an optimization so it does not guarantee
3040  *       that no dcache entries will exist at process exit time it
3041  *       just makes it very unlikely that any will persist.
3042  */
3043 
proc_flush_task(struct task_struct * task)3044 void proc_flush_task(struct task_struct *task)
3045 {
3046 	int i;
3047 	struct pid *pid, *tgid;
3048 	struct upid *upid;
3049 
3050 	pid = task_pid(task);
3051 	tgid = task_tgid(task);
3052 
3053 	for (i = 0; i <= pid->level; i++) {
3054 		upid = &pid->numbers[i];
3055 		proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3056 					tgid->numbers[i].nr);
3057 	}
3058 }
3059 
proc_pid_instantiate(struct inode * dir,struct dentry * dentry,struct task_struct * task,const void * ptr)3060 static int proc_pid_instantiate(struct inode *dir,
3061 				   struct dentry * dentry,
3062 				   struct task_struct *task, const void *ptr)
3063 {
3064 	struct inode *inode;
3065 
3066 	inode = proc_pid_make_inode(dir->i_sb, task);
3067 	if (!inode)
3068 		goto out;
3069 
3070 	inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3071 	inode->i_op = &proc_tgid_base_inode_operations;
3072 	inode->i_fop = &proc_tgid_base_operations;
3073 	inode->i_flags|=S_IMMUTABLE;
3074 
3075 	set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
3076 						  ARRAY_SIZE(tgid_base_stuff)));
3077 
3078 	d_set_d_op(dentry, &pid_dentry_operations);
3079 
3080 	d_add(dentry, inode);
3081 	/* Close the race of the process dying before we return the dentry */
3082 	if (pid_revalidate(dentry, 0))
3083 		return 0;
3084 out:
3085 	return -ENOENT;
3086 }
3087 
proc_pid_lookup(struct inode * dir,struct dentry * dentry,unsigned int flags)3088 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3089 {
3090 	int result = -ENOENT;
3091 	struct task_struct *task;
3092 	unsigned tgid;
3093 	struct pid_namespace *ns;
3094 
3095 	tgid = name_to_int(&dentry->d_name);
3096 	if (tgid == ~0U)
3097 		goto out;
3098 
3099 	ns = dentry->d_sb->s_fs_info;
3100 	rcu_read_lock();
3101 	task = find_task_by_pid_ns(tgid, ns);
3102 	if (task)
3103 		get_task_struct(task);
3104 	rcu_read_unlock();
3105 	if (!task)
3106 		goto out;
3107 
3108 	result = proc_pid_instantiate(dir, dentry, task, NULL);
3109 	put_task_struct(task);
3110 out:
3111 	return ERR_PTR(result);
3112 }
3113 
3114 /*
3115  * Find the first task with tgid >= tgid
3116  *
3117  */
3118 struct tgid_iter {
3119 	unsigned int tgid;
3120 	struct task_struct *task;
3121 };
next_tgid(struct pid_namespace * ns,struct tgid_iter iter)3122 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3123 {
3124 	struct pid *pid;
3125 
3126 	if (iter.task)
3127 		put_task_struct(iter.task);
3128 	rcu_read_lock();
3129 retry:
3130 	iter.task = NULL;
3131 	pid = find_ge_pid(iter.tgid, ns);
3132 	if (pid) {
3133 		iter.tgid = pid_nr_ns(pid, ns);
3134 		iter.task = pid_task(pid, PIDTYPE_PID);
3135 		/* What we to know is if the pid we have find is the
3136 		 * pid of a thread_group_leader.  Testing for task
3137 		 * being a thread_group_leader is the obvious thing
3138 		 * todo but there is a window when it fails, due to
3139 		 * the pid transfer logic in de_thread.
3140 		 *
3141 		 * So we perform the straight forward test of seeing
3142 		 * if the pid we have found is the pid of a thread
3143 		 * group leader, and don't worry if the task we have
3144 		 * found doesn't happen to be a thread group leader.
3145 		 * As we don't care in the case of readdir.
3146 		 */
3147 		if (!iter.task || !has_group_leader_pid(iter.task)) {
3148 			iter.tgid += 1;
3149 			goto retry;
3150 		}
3151 		get_task_struct(iter.task);
3152 	}
3153 	rcu_read_unlock();
3154 	return iter;
3155 }
3156 
3157 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3158 
3159 /* for the /proc/ directory itself, after non-process stuff has been done */
proc_pid_readdir(struct file * file,struct dir_context * ctx)3160 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3161 {
3162 	struct tgid_iter iter;
3163 	struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
3164 	loff_t pos = ctx->pos;
3165 
3166 	if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3167 		return 0;
3168 
3169 	if (pos == TGID_OFFSET - 2) {
3170 		struct inode *inode = d_inode(ns->proc_self);
3171 		if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3172 			return 0;
3173 		ctx->pos = pos = pos + 1;
3174 	}
3175 	if (pos == TGID_OFFSET - 1) {
3176 		struct inode *inode = d_inode(ns->proc_thread_self);
3177 		if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3178 			return 0;
3179 		ctx->pos = pos = pos + 1;
3180 	}
3181 	iter.tgid = pos - TGID_OFFSET;
3182 	iter.task = NULL;
3183 	for (iter = next_tgid(ns, iter);
3184 	     iter.task;
3185 	     iter.tgid += 1, iter = next_tgid(ns, iter)) {
3186 		char name[PROC_NUMBUF];
3187 		int len;
3188 
3189 		cond_resched();
3190 		if (!has_pid_permissions(ns, iter.task, 2))
3191 			continue;
3192 
3193 		len = snprintf(name, sizeof(name), "%d", iter.tgid);
3194 		ctx->pos = iter.tgid + TGID_OFFSET;
3195 		if (!proc_fill_cache(file, ctx, name, len,
3196 				     proc_pid_instantiate, iter.task, NULL)) {
3197 			put_task_struct(iter.task);
3198 			return 0;
3199 		}
3200 	}
3201 	ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3202 	return 0;
3203 }
3204 
3205 /*
3206  * proc_tid_comm_permission is a special permission function exclusively
3207  * used for the node /proc/<pid>/task/<tid>/comm.
3208  * It bypasses generic permission checks in the case where a task of the same
3209  * task group attempts to access the node.
3210  * The rationale behind this is that glibc and bionic access this node for
3211  * cross thread naming (pthread_set/getname_np(!self)). However, if
3212  * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3213  * which locks out the cross thread naming implementation.
3214  * This function makes sure that the node is always accessible for members of
3215  * same thread group.
3216  */
proc_tid_comm_permission(struct inode * inode,int mask)3217 static int proc_tid_comm_permission(struct inode *inode, int mask)
3218 {
3219 	bool is_same_tgroup;
3220 	struct task_struct *task;
3221 
3222 	task = get_proc_task(inode);
3223 	if (!task)
3224 		return -ESRCH;
3225 	is_same_tgroup = same_thread_group(current, task);
3226 	put_task_struct(task);
3227 
3228 	if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3229 		/* This file (/proc/<pid>/task/<tid>/comm) can always be
3230 		 * read or written by the members of the corresponding
3231 		 * thread group.
3232 		 */
3233 		return 0;
3234 	}
3235 
3236 	return generic_permission(inode, mask);
3237 }
3238 
3239 static const struct inode_operations proc_tid_comm_inode_operations = {
3240 		.permission = proc_tid_comm_permission,
3241 };
3242 
3243 /*
3244  * Tasks
3245  */
3246 static const struct pid_entry tid_base_stuff[] = {
3247 	DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3248 	DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3249 	DIR("ns",	 S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3250 #ifdef CONFIG_NET
3251 	DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3252 #endif
3253 	REG("environ",   S_IRUSR, proc_environ_operations),
3254 	REG("auxv",      S_IRUSR, proc_auxv_operations),
3255 	ONE("status",    S_IRUGO, proc_pid_status),
3256 	ONE("personality", S_IRUSR, proc_pid_personality),
3257 	ONE("limits",	 S_IRUGO, proc_pid_limits),
3258 #ifdef CONFIG_SCHED_DEBUG
3259 	REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3260 #endif
3261 	NOD("comm",      S_IFREG|S_IRUGO|S_IWUSR,
3262 			 &proc_tid_comm_inode_operations,
3263 			 &proc_pid_set_comm_operations, {}),
3264 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3265 	ONE("syscall",   S_IRUSR, proc_pid_syscall),
3266 #endif
3267 	REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3268 	ONE("stat",      S_IRUGO, proc_tid_stat),
3269 	ONE("statm",     S_IRUGO, proc_pid_statm),
3270 	REG("maps",      S_IRUGO, proc_tid_maps_operations),
3271 #ifdef CONFIG_PROC_CHILDREN
3272 	REG("children",  S_IRUGO, proc_tid_children_operations),
3273 #endif
3274 #ifdef CONFIG_NUMA
3275 	REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3276 #endif
3277 	REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3278 	LNK("cwd",       proc_cwd_link),
3279 	LNK("root",      proc_root_link),
3280 	LNK("exe",       proc_exe_link),
3281 	REG("mounts",    S_IRUGO, proc_mounts_operations),
3282 	REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3283 #ifdef CONFIG_PROC_PAGE_MONITOR
3284 	REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3285 	REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3286 	REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3287 #endif
3288 #ifdef CONFIG_SECURITY
3289 	DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3290 #endif
3291 #ifdef CONFIG_KALLSYMS
3292 	ONE("wchan",     S_IRUGO, proc_pid_wchan),
3293 #endif
3294 #ifdef CONFIG_STACKTRACE
3295 	ONE("stack",      S_IRUSR, proc_pid_stack),
3296 #endif
3297 #ifdef CONFIG_SCHED_INFO
3298 	ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3299 #endif
3300 #ifdef CONFIG_LATENCYTOP
3301 	REG("latency",  S_IRUGO, proc_lstats_operations),
3302 #endif
3303 #ifdef CONFIG_PROC_PID_CPUSET
3304 	ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3305 #endif
3306 #ifdef CONFIG_CGROUPS
3307 	ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3308 #endif
3309 	ONE("oom_score", S_IRUGO, proc_oom_score),
3310 	REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3311 	REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3312 #ifdef CONFIG_AUDITSYSCALL
3313 	REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3314 	REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3315 #endif
3316 #ifdef CONFIG_FAULT_INJECTION
3317 	REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3318 #endif
3319 #ifdef CONFIG_TASK_IO_ACCOUNTING
3320 	ONE("io",	S_IRUSR, proc_tid_io_accounting),
3321 #endif
3322 #ifdef CONFIG_HARDWALL
3323 	ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
3324 #endif
3325 #ifdef CONFIG_USER_NS
3326 	REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3327 	REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3328 	REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3329 	REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3330 #endif
3331 #ifdef CONFIG_CPU_FREQ_TIMES
3332 	ONE("time_in_state", 0444, proc_time_in_state_show),
3333 #endif
3334 };
3335 
proc_tid_base_readdir(struct file * file,struct dir_context * ctx)3336 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3337 {
3338 	return proc_pident_readdir(file, ctx,
3339 				   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3340 }
3341 
proc_tid_base_lookup(struct inode * dir,struct dentry * dentry,unsigned int flags)3342 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3343 {
3344 	return proc_pident_lookup(dir, dentry,
3345 				  tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3346 }
3347 
3348 static const struct file_operations proc_tid_base_operations = {
3349 	.read		= generic_read_dir,
3350 	.iterate_shared	= proc_tid_base_readdir,
3351 	.llseek		= generic_file_llseek,
3352 };
3353 
3354 static const struct inode_operations proc_tid_base_inode_operations = {
3355 	.lookup		= proc_tid_base_lookup,
3356 	.getattr	= pid_getattr,
3357 	.setattr	= proc_setattr,
3358 };
3359 
proc_task_instantiate(struct inode * dir,struct dentry * dentry,struct task_struct * task,const void * ptr)3360 static int proc_task_instantiate(struct inode *dir,
3361 	struct dentry *dentry, struct task_struct *task, const void *ptr)
3362 {
3363 	struct inode *inode;
3364 	inode = proc_pid_make_inode(dir->i_sb, task);
3365 
3366 	if (!inode)
3367 		goto out;
3368 	inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3369 	inode->i_op = &proc_tid_base_inode_operations;
3370 	inode->i_fop = &proc_tid_base_operations;
3371 	inode->i_flags|=S_IMMUTABLE;
3372 
3373 	set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3374 						  ARRAY_SIZE(tid_base_stuff)));
3375 
3376 	d_set_d_op(dentry, &pid_dentry_operations);
3377 
3378 	d_add(dentry, inode);
3379 	/* Close the race of the process dying before we return the dentry */
3380 	if (pid_revalidate(dentry, 0))
3381 		return 0;
3382 out:
3383 	return -ENOENT;
3384 }
3385 
proc_task_lookup(struct inode * dir,struct dentry * dentry,unsigned int flags)3386 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3387 {
3388 	int result = -ENOENT;
3389 	struct task_struct *task;
3390 	struct task_struct *leader = get_proc_task(dir);
3391 	unsigned tid;
3392 	struct pid_namespace *ns;
3393 
3394 	if (!leader)
3395 		goto out_no_task;
3396 
3397 	tid = name_to_int(&dentry->d_name);
3398 	if (tid == ~0U)
3399 		goto out;
3400 
3401 	ns = dentry->d_sb->s_fs_info;
3402 	rcu_read_lock();
3403 	task = find_task_by_pid_ns(tid, ns);
3404 	if (task)
3405 		get_task_struct(task);
3406 	rcu_read_unlock();
3407 	if (!task)
3408 		goto out;
3409 	if (!same_thread_group(leader, task))
3410 		goto out_drop_task;
3411 
3412 	result = proc_task_instantiate(dir, dentry, task, NULL);
3413 out_drop_task:
3414 	put_task_struct(task);
3415 out:
3416 	put_task_struct(leader);
3417 out_no_task:
3418 	return ERR_PTR(result);
3419 }
3420 
3421 /*
3422  * Find the first tid of a thread group to return to user space.
3423  *
3424  * Usually this is just the thread group leader, but if the users
3425  * buffer was too small or there was a seek into the middle of the
3426  * directory we have more work todo.
3427  *
3428  * In the case of a short read we start with find_task_by_pid.
3429  *
3430  * In the case of a seek we start with the leader and walk nr
3431  * threads past it.
3432  */
first_tid(struct pid * pid,int tid,loff_t f_pos,struct pid_namespace * ns)3433 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3434 					struct pid_namespace *ns)
3435 {
3436 	struct task_struct *pos, *task;
3437 	unsigned long nr = f_pos;
3438 
3439 	if (nr != f_pos)	/* 32bit overflow? */
3440 		return NULL;
3441 
3442 	rcu_read_lock();
3443 	task = pid_task(pid, PIDTYPE_PID);
3444 	if (!task)
3445 		goto fail;
3446 
3447 	/* Attempt to start with the tid of a thread */
3448 	if (tid && nr) {
3449 		pos = find_task_by_pid_ns(tid, ns);
3450 		if (pos && same_thread_group(pos, task))
3451 			goto found;
3452 	}
3453 
3454 	/* If nr exceeds the number of threads there is nothing todo */
3455 	if (nr >= get_nr_threads(task))
3456 		goto fail;
3457 
3458 	/* If we haven't found our starting place yet start
3459 	 * with the leader and walk nr threads forward.
3460 	 */
3461 	pos = task = task->group_leader;
3462 	do {
3463 		if (!nr--)
3464 			goto found;
3465 	} while_each_thread(task, pos);
3466 fail:
3467 	pos = NULL;
3468 	goto out;
3469 found:
3470 	get_task_struct(pos);
3471 out:
3472 	rcu_read_unlock();
3473 	return pos;
3474 }
3475 
3476 /*
3477  * Find the next thread in the thread list.
3478  * Return NULL if there is an error or no next thread.
3479  *
3480  * The reference to the input task_struct is released.
3481  */
next_tid(struct task_struct * start)3482 static struct task_struct *next_tid(struct task_struct *start)
3483 {
3484 	struct task_struct *pos = NULL;
3485 	rcu_read_lock();
3486 	if (pid_alive(start)) {
3487 		pos = next_thread(start);
3488 		if (thread_group_leader(pos))
3489 			pos = NULL;
3490 		else
3491 			get_task_struct(pos);
3492 	}
3493 	rcu_read_unlock();
3494 	put_task_struct(start);
3495 	return pos;
3496 }
3497 
3498 /* for the /proc/TGID/task/ directories */
proc_task_readdir(struct file * file,struct dir_context * ctx)3499 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3500 {
3501 	struct inode *inode = file_inode(file);
3502 	struct task_struct *task;
3503 	struct pid_namespace *ns;
3504 	int tid;
3505 
3506 	if (proc_inode_is_dead(inode))
3507 		return -ENOENT;
3508 
3509 	if (!dir_emit_dots(file, ctx))
3510 		return 0;
3511 
3512 	/* f_version caches the tgid value that the last readdir call couldn't
3513 	 * return. lseek aka telldir automagically resets f_version to 0.
3514 	 */
3515 	ns = inode->i_sb->s_fs_info;
3516 	tid = (int)file->f_version;
3517 	file->f_version = 0;
3518 	for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3519 	     task;
3520 	     task = next_tid(task), ctx->pos++) {
3521 		char name[PROC_NUMBUF];
3522 		int len;
3523 		tid = task_pid_nr_ns(task, ns);
3524 		len = snprintf(name, sizeof(name), "%d", tid);
3525 		if (!proc_fill_cache(file, ctx, name, len,
3526 				proc_task_instantiate, task, NULL)) {
3527 			/* returning this tgid failed, save it as the first
3528 			 * pid for the next readir call */
3529 			file->f_version = (u64)tid;
3530 			put_task_struct(task);
3531 			break;
3532 		}
3533 	}
3534 
3535 	return 0;
3536 }
3537 
proc_task_getattr(struct vfsmount * mnt,struct dentry * dentry,struct kstat * stat)3538 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3539 {
3540 	struct inode *inode = d_inode(dentry);
3541 	struct task_struct *p = get_proc_task(inode);
3542 	generic_fillattr(inode, stat);
3543 
3544 	if (p) {
3545 		stat->nlink += get_nr_threads(p);
3546 		put_task_struct(p);
3547 	}
3548 
3549 	return 0;
3550 }
3551 
3552 static const struct inode_operations proc_task_inode_operations = {
3553 	.lookup		= proc_task_lookup,
3554 	.getattr	= proc_task_getattr,
3555 	.setattr	= proc_setattr,
3556 	.permission	= proc_pid_permission,
3557 };
3558 
3559 static const struct file_operations proc_task_operations = {
3560 	.read		= generic_read_dir,
3561 	.iterate_shared	= proc_task_readdir,
3562 	.llseek		= generic_file_llseek,
3563 };
3564