4 * Copyright (C) 1991, 1992 Linus Torvalds
6 * proc base directory handling functions
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.
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>
24 * Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
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.
35 * Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
36 * Pud inclusion in the page table walking.
40 * 10LE Instituto Nokia de Tecnologia - INdT:
41 * A better way to walks through the page table as suggested by Hugh Dickins.
43 * Simo Piiroinen <simo.piiroinen@nokia.com>:
44 * Smaps information related to shared, private, clean and dirty pages.
46 * Paul Mundt <paul.mundt@nokia.com>:
47 * Overall revision about smaps.
50 #include <asm/uaccess.h>
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>
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/cgroup.h>
77 #include <linux/cpuset.h>
78 #include <linux/audit.h>
79 #include <linux/poll.h>
80 #include <linux/nsproxy.h>
81 #include <linux/oom.h>
82 #include <linux/elf.h>
83 #include <linux/pid_namespace.h>
84 #include <linux/fs_struct.h>
85 #include <linux/slab.h>
86 #ifdef CONFIG_HARDWALL
87 #include <asm/hardwall.h>
92 * Implementing inode permission operations in /proc is almost
93 * certainly an error. Permission checks need to happen during
94 * each system call not at open time. The reason is that most of
95 * what we wish to check for permissions in /proc varies at runtime.
97 * The classic example of a problem is opening file descriptors
98 * in /proc for a task before it execs a suid executable.
105 const struct inode_operations *iop;
106 const struct file_operations *fop;
110 #define NOD(NAME, MODE, IOP, FOP, OP) { \
112 .len = sizeof(NAME) - 1, \
119 #define DIR(NAME, MODE, iops, fops) \
120 NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
121 #define LNK(NAME, get_link) \
122 NOD(NAME, (S_IFLNK|S_IRWXUGO), \
123 &proc_pid_link_inode_operations, NULL, \
124 { .proc_get_link = get_link } )
125 #define REG(NAME, MODE, fops) \
126 NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
127 #define INF(NAME, MODE, read) \
128 NOD(NAME, (S_IFREG|(MODE)), \
129 NULL, &proc_info_file_operations, \
130 { .proc_read = read } )
131 #define ONE(NAME, MODE, show) \
132 NOD(NAME, (S_IFREG|(MODE)), \
133 NULL, &proc_single_file_operations, \
134 { .proc_show = show } )
137 * Count the number of hardlinks for the pid_entry table, excluding the .
140 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
147 for (i = 0; i < n; ++i) {
148 if (S_ISDIR(entries[i].mode))
155 static int get_task_root(struct task_struct *task, struct path *root)
157 int result = -ENOENT;
161 get_fs_root(task->fs, root);
168 static int proc_cwd_link(struct inode *inode, struct path *path)
170 struct task_struct *task = get_proc_task(inode);
171 int result = -ENOENT;
176 get_fs_pwd(task->fs, path);
180 put_task_struct(task);
185 static int proc_root_link(struct inode *inode, struct path *path)
187 struct task_struct *task = get_proc_task(inode);
188 int result = -ENOENT;
191 result = get_task_root(task, path);
192 put_task_struct(task);
197 static struct mm_struct *__check_mem_permission(struct task_struct *task)
199 struct mm_struct *mm;
201 mm = get_task_mm(task);
203 return ERR_PTR(-EINVAL);
206 * A task can always look at itself, in case it chooses
207 * to use system calls instead of load instructions.
213 * If current is actively ptrace'ing, and would also be
214 * permitted to freshly attach with ptrace now, permit it.
216 if (task_is_stopped_or_traced(task)) {
219 match = (tracehook_tracer_task(task) == current);
221 if (match && ptrace_may_access(task, PTRACE_MODE_ATTACH))
226 * No one else is allowed.
229 return ERR_PTR(-EPERM);
233 * If current may access user memory in @task return a reference to the
234 * corresponding mm, otherwise ERR_PTR.
236 static struct mm_struct *check_mem_permission(struct task_struct *task)
238 struct mm_struct *mm;
242 * Avoid racing if task exec's as we might get a new mm but validate
243 * against old credentials.
245 err = mutex_lock_killable(&task->signal->cred_guard_mutex);
249 mm = __check_mem_permission(task);
250 mutex_unlock(&task->signal->cred_guard_mutex);
255 struct mm_struct *mm_for_maps(struct task_struct *task)
257 struct mm_struct *mm;
260 err = mutex_lock_killable(&task->signal->cred_guard_mutex);
264 mm = get_task_mm(task);
265 if (mm && mm != current->mm &&
266 !ptrace_may_access(task, PTRACE_MODE_READ)) {
268 mm = ERR_PTR(-EACCES);
270 mutex_unlock(&task->signal->cred_guard_mutex);
275 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
279 struct mm_struct *mm = get_task_mm(task);
283 goto out_mm; /* Shh! No looking before we're done */
285 len = mm->arg_end - mm->arg_start;
290 res = access_process_vm(task, mm->arg_start, buffer, len, 0);
292 // If the nul at the end of args has been overwritten, then
293 // assume application is using setproctitle(3).
294 if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
295 len = strnlen(buffer, res);
299 len = mm->env_end - mm->env_start;
300 if (len > PAGE_SIZE - res)
301 len = PAGE_SIZE - res;
302 res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
303 res = strnlen(buffer, res);
312 static int proc_pid_auxv(struct task_struct *task, char *buffer)
314 struct mm_struct *mm = mm_for_maps(task);
315 int res = PTR_ERR(mm);
316 if (mm && !IS_ERR(mm)) {
317 unsigned int nwords = 0;
320 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
321 res = nwords * sizeof(mm->saved_auxv[0]);
324 memcpy(buffer, mm->saved_auxv, res);
331 #ifdef CONFIG_KALLSYMS
333 * Provides a wchan file via kallsyms in a proper one-value-per-file format.
334 * Returns the resolved symbol. If that fails, simply return the address.
336 static int proc_pid_wchan(struct task_struct *task, char *buffer)
339 char symname[KSYM_NAME_LEN];
341 wchan = get_wchan(task);
343 if (lookup_symbol_name(wchan, symname) < 0)
344 if (!ptrace_may_access(task, PTRACE_MODE_READ))
347 return sprintf(buffer, "%lu", wchan);
349 return sprintf(buffer, "%s", symname);
351 #endif /* CONFIG_KALLSYMS */
353 static int lock_trace(struct task_struct *task)
355 int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
358 if (!ptrace_may_access(task, PTRACE_MODE_ATTACH)) {
359 mutex_unlock(&task->signal->cred_guard_mutex);
365 static void unlock_trace(struct task_struct *task)
367 mutex_unlock(&task->signal->cred_guard_mutex);
370 #ifdef CONFIG_STACKTRACE
372 #define MAX_STACK_TRACE_DEPTH 64
374 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
375 struct pid *pid, struct task_struct *task)
377 struct stack_trace trace;
378 unsigned long *entries;
382 entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
386 trace.nr_entries = 0;
387 trace.max_entries = MAX_STACK_TRACE_DEPTH;
388 trace.entries = entries;
391 err = lock_trace(task);
393 save_stack_trace_tsk(task, &trace);
395 for (i = 0; i < trace.nr_entries; i++) {
396 seq_printf(m, "[<%pK>] %pS\n",
397 (void *)entries[i], (void *)entries[i]);
407 #ifdef CONFIG_SCHEDSTATS
409 * Provides /proc/PID/schedstat
411 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
413 return sprintf(buffer, "%llu %llu %lu\n",
414 (unsigned long long)task->se.sum_exec_runtime,
415 (unsigned long long)task->sched_info.run_delay,
416 task->sched_info.pcount);
420 #ifdef CONFIG_LATENCYTOP
421 static int lstats_show_proc(struct seq_file *m, void *v)
424 struct inode *inode = m->private;
425 struct task_struct *task = get_proc_task(inode);
429 seq_puts(m, "Latency Top version : v0.1\n");
430 for (i = 0; i < 32; i++) {
431 struct latency_record *lr = &task->latency_record[i];
432 if (lr->backtrace[0]) {
434 seq_printf(m, "%i %li %li",
435 lr->count, lr->time, lr->max);
436 for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
437 unsigned long bt = lr->backtrace[q];
442 seq_printf(m, " %ps", (void *)bt);
448 put_task_struct(task);
452 static int lstats_open(struct inode *inode, struct file *file)
454 return single_open(file, lstats_show_proc, inode);
457 static ssize_t lstats_write(struct file *file, const char __user *buf,
458 size_t count, loff_t *offs)
460 struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
464 clear_all_latency_tracing(task);
465 put_task_struct(task);
470 static const struct file_operations proc_lstats_operations = {
473 .write = lstats_write,
475 .release = single_release,
480 static int proc_oom_score(struct task_struct *task, char *buffer)
482 unsigned long points = 0;
484 read_lock(&tasklist_lock);
486 points = oom_badness(task, NULL, NULL,
487 totalram_pages + total_swap_pages);
488 read_unlock(&tasklist_lock);
489 return sprintf(buffer, "%lu\n", points);
497 static const struct limit_names lnames[RLIM_NLIMITS] = {
498 [RLIMIT_CPU] = {"Max cpu time", "seconds"},
499 [RLIMIT_FSIZE] = {"Max file size", "bytes"},
500 [RLIMIT_DATA] = {"Max data size", "bytes"},
501 [RLIMIT_STACK] = {"Max stack size", "bytes"},
502 [RLIMIT_CORE] = {"Max core file size", "bytes"},
503 [RLIMIT_RSS] = {"Max resident set", "bytes"},
504 [RLIMIT_NPROC] = {"Max processes", "processes"},
505 [RLIMIT_NOFILE] = {"Max open files", "files"},
506 [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
507 [RLIMIT_AS] = {"Max address space", "bytes"},
508 [RLIMIT_LOCKS] = {"Max file locks", "locks"},
509 [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
510 [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
511 [RLIMIT_NICE] = {"Max nice priority", NULL},
512 [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
513 [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
516 /* Display limits for a process */
517 static int proc_pid_limits(struct task_struct *task, char *buffer)
522 char *bufptr = buffer;
524 struct rlimit rlim[RLIM_NLIMITS];
526 if (!lock_task_sighand(task, &flags))
528 memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
529 unlock_task_sighand(task, &flags);
532 * print the file header
534 count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
535 "Limit", "Soft Limit", "Hard Limit", "Units");
537 for (i = 0; i < RLIM_NLIMITS; i++) {
538 if (rlim[i].rlim_cur == RLIM_INFINITY)
539 count += sprintf(&bufptr[count], "%-25s %-20s ",
540 lnames[i].name, "unlimited");
542 count += sprintf(&bufptr[count], "%-25s %-20lu ",
543 lnames[i].name, rlim[i].rlim_cur);
545 if (rlim[i].rlim_max == RLIM_INFINITY)
546 count += sprintf(&bufptr[count], "%-20s ", "unlimited");
548 count += sprintf(&bufptr[count], "%-20lu ",
552 count += sprintf(&bufptr[count], "%-10s\n",
555 count += sprintf(&bufptr[count], "\n");
561 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
562 static int proc_pid_syscall(struct task_struct *task, char *buffer)
565 unsigned long args[6], sp, pc;
566 int res = lock_trace(task);
570 if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
571 res = sprintf(buffer, "running\n");
573 res = sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
575 res = sprintf(buffer,
576 "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
578 args[0], args[1], args[2], args[3], args[4], args[5],
583 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
585 /************************************************************************/
586 /* Here the fs part begins */
587 /************************************************************************/
589 /* permission checks */
590 static int proc_fd_access_allowed(struct inode *inode)
592 struct task_struct *task;
594 /* Allow access to a task's file descriptors if it is us or we
595 * may use ptrace attach to the process and find out that
598 task = get_proc_task(inode);
600 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
601 put_task_struct(task);
606 static int proc_setattr(struct dentry *dentry, struct iattr *attr)
609 struct inode *inode = dentry->d_inode;
611 if (attr->ia_valid & ATTR_MODE)
614 error = inode_change_ok(inode, attr);
618 if ((attr->ia_valid & ATTR_SIZE) &&
619 attr->ia_size != i_size_read(inode)) {
620 error = vmtruncate(inode, attr->ia_size);
625 setattr_copy(inode, attr);
626 mark_inode_dirty(inode);
630 static const struct inode_operations proc_def_inode_operations = {
631 .setattr = proc_setattr,
634 static int mounts_open_common(struct inode *inode, struct file *file,
635 const struct seq_operations *op)
637 struct task_struct *task = get_proc_task(inode);
639 struct mnt_namespace *ns = NULL;
641 struct proc_mounts *p;
646 nsp = task_nsproxy(task);
653 if (ns && get_task_root(task, &root) == 0)
655 put_task_struct(task);
664 p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
668 file->private_data = &p->m;
669 ret = seq_open(file, op);
676 p->event = ns->event;
690 static int mounts_release(struct inode *inode, struct file *file)
692 struct proc_mounts *p = file->private_data;
695 return seq_release(inode, file);
698 static unsigned mounts_poll(struct file *file, poll_table *wait)
700 struct proc_mounts *p = file->private_data;
701 unsigned res = POLLIN | POLLRDNORM;
703 poll_wait(file, &p->ns->poll, wait);
704 if (mnt_had_events(p))
705 res |= POLLERR | POLLPRI;
710 static int mounts_open(struct inode *inode, struct file *file)
712 return mounts_open_common(inode, file, &mounts_op);
715 static const struct file_operations proc_mounts_operations = {
719 .release = mounts_release,
723 static int mountinfo_open(struct inode *inode, struct file *file)
725 return mounts_open_common(inode, file, &mountinfo_op);
728 static const struct file_operations proc_mountinfo_operations = {
729 .open = mountinfo_open,
732 .release = mounts_release,
736 static int mountstats_open(struct inode *inode, struct file *file)
738 return mounts_open_common(inode, file, &mountstats_op);
741 static const struct file_operations proc_mountstats_operations = {
742 .open = mountstats_open,
745 .release = mounts_release,
748 #define PROC_BLOCK_SIZE (3*1024) /* 4K page size but our output routines use some slack for overruns */
750 static ssize_t proc_info_read(struct file * file, char __user * buf,
751 size_t count, loff_t *ppos)
753 struct inode * inode = file->f_path.dentry->d_inode;
756 struct task_struct *task = get_proc_task(inode);
762 if (count > PROC_BLOCK_SIZE)
763 count = PROC_BLOCK_SIZE;
766 if (!(page = __get_free_page(GFP_TEMPORARY)))
769 length = PROC_I(inode)->op.proc_read(task, (char*)page);
772 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
775 put_task_struct(task);
780 static const struct file_operations proc_info_file_operations = {
781 .read = proc_info_read,
782 .llseek = generic_file_llseek,
785 static int proc_single_show(struct seq_file *m, void *v)
787 struct inode *inode = m->private;
788 struct pid_namespace *ns;
790 struct task_struct *task;
793 ns = inode->i_sb->s_fs_info;
794 pid = proc_pid(inode);
795 task = get_pid_task(pid, PIDTYPE_PID);
799 ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
801 put_task_struct(task);
805 static int proc_single_open(struct inode *inode, struct file *filp)
807 return single_open(filp, proc_single_show, inode);
810 static const struct file_operations proc_single_file_operations = {
811 .open = proc_single_open,
814 .release = single_release,
817 static int mem_open(struct inode* inode, struct file* file)
819 file->private_data = (void*)((long)current->self_exec_id);
820 /* OK to pass negative loff_t, we can catch out-of-range */
821 file->f_mode |= FMODE_UNSIGNED_OFFSET;
825 static ssize_t mem_read(struct file * file, char __user * buf,
826 size_t count, loff_t *ppos)
828 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
830 unsigned long src = *ppos;
832 struct mm_struct *mm;
838 page = (char *)__get_free_page(GFP_TEMPORARY);
842 mm = check_mem_permission(task);
849 if (file->private_data != (void*)((long)current->self_exec_id))
855 int this_len, retval;
857 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
858 retval = access_remote_vm(mm, src, page, this_len, 0);
865 if (copy_to_user(buf, page, retval)) {
880 free_page((unsigned long) page);
882 put_task_struct(task);
887 static ssize_t mem_write(struct file * file, const char __user *buf,
888 size_t count, loff_t *ppos)
892 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
893 unsigned long dst = *ppos;
894 struct mm_struct *mm;
900 mm = check_mem_permission(task);
901 copied = PTR_ERR(mm);
906 if (file->private_data != (void *)((long)current->self_exec_id))
910 page = (char *)__get_free_page(GFP_TEMPORARY);
916 int this_len, retval;
918 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
919 if (copy_from_user(page, buf, this_len)) {
923 retval = access_remote_vm(mm, dst, page, this_len, 1);
935 free_page((unsigned long) page);
939 put_task_struct(task);
944 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
948 file->f_pos = offset;
951 file->f_pos += offset;
956 force_successful_syscall_return();
960 static const struct file_operations proc_mem_operations = {
967 static ssize_t environ_read(struct file *file, char __user *buf,
968 size_t count, loff_t *ppos)
970 struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
972 unsigned long src = *ppos;
974 struct mm_struct *mm;
980 page = (char *)__get_free_page(GFP_TEMPORARY);
985 mm = mm_for_maps(task);
987 if (!mm || IS_ERR(mm))
992 int this_len, retval, max_len;
994 this_len = mm->env_end - (mm->env_start + src);
999 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
1000 this_len = (this_len > max_len) ? max_len : this_len;
1002 retval = access_process_vm(task, (mm->env_start + src),
1010 if (copy_to_user(buf, page, retval)) {
1024 free_page((unsigned long) page);
1026 put_task_struct(task);
1031 static const struct file_operations proc_environ_operations = {
1032 .read = environ_read,
1033 .llseek = generic_file_llseek,
1036 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
1037 size_t count, loff_t *ppos)
1039 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
1040 char buffer[PROC_NUMBUF];
1042 int oom_adjust = OOM_DISABLE;
1043 unsigned long flags;
1048 if (lock_task_sighand(task, &flags)) {
1049 oom_adjust = task->signal->oom_adj;
1050 unlock_task_sighand(task, &flags);
1053 put_task_struct(task);
1055 len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
1057 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1060 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
1061 size_t count, loff_t *ppos)
1063 struct task_struct *task;
1064 char buffer[PROC_NUMBUF];
1066 unsigned long flags;
1069 memset(buffer, 0, sizeof(buffer));
1070 if (count > sizeof(buffer) - 1)
1071 count = sizeof(buffer) - 1;
1072 if (copy_from_user(buffer, buf, count)) {
1077 err = strict_strtol(strstrip(buffer), 0, &oom_adjust);
1080 if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
1081 oom_adjust != OOM_DISABLE) {
1086 task = get_proc_task(file->f_path.dentry->d_inode);
1098 if (!lock_task_sighand(task, &flags)) {
1103 if (oom_adjust < task->signal->oom_adj && !capable(CAP_SYS_RESOURCE)) {
1108 if (oom_adjust != task->signal->oom_adj) {
1109 if (oom_adjust == OOM_DISABLE)
1110 atomic_inc(&task->mm->oom_disable_count);
1111 if (task->signal->oom_adj == OOM_DISABLE)
1112 atomic_dec(&task->mm->oom_disable_count);
1116 * Warn that /proc/pid/oom_adj is deprecated, see
1117 * Documentation/feature-removal-schedule.txt.
1119 printk_once(KERN_WARNING "%s (%d): /proc/%d/oom_adj is deprecated, "
1120 "please use /proc/%d/oom_score_adj instead.\n",
1121 current->comm, task_pid_nr(current),
1122 task_pid_nr(task), task_pid_nr(task));
1123 task->signal->oom_adj = oom_adjust;
1125 * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1126 * value is always attainable.
1128 if (task->signal->oom_adj == OOM_ADJUST_MAX)
1129 task->signal->oom_score_adj = OOM_SCORE_ADJ_MAX;
1131 task->signal->oom_score_adj = (oom_adjust * OOM_SCORE_ADJ_MAX) /
1134 unlock_task_sighand(task, &flags);
1137 put_task_struct(task);
1139 return err < 0 ? err : count;
1142 static const struct file_operations proc_oom_adjust_operations = {
1143 .read = oom_adjust_read,
1144 .write = oom_adjust_write,
1145 .llseek = generic_file_llseek,
1148 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1149 size_t count, loff_t *ppos)
1151 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
1152 char buffer[PROC_NUMBUF];
1153 int oom_score_adj = OOM_SCORE_ADJ_MIN;
1154 unsigned long flags;
1159 if (lock_task_sighand(task, &flags)) {
1160 oom_score_adj = task->signal->oom_score_adj;
1161 unlock_task_sighand(task, &flags);
1163 put_task_struct(task);
1164 len = snprintf(buffer, sizeof(buffer), "%d\n", oom_score_adj);
1165 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1168 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1169 size_t count, loff_t *ppos)
1171 struct task_struct *task;
1172 char buffer[PROC_NUMBUF];
1173 unsigned long flags;
1177 memset(buffer, 0, sizeof(buffer));
1178 if (count > sizeof(buffer) - 1)
1179 count = sizeof(buffer) - 1;
1180 if (copy_from_user(buffer, buf, count)) {
1185 err = strict_strtol(strstrip(buffer), 0, &oom_score_adj);
1188 if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1189 oom_score_adj > OOM_SCORE_ADJ_MAX) {
1194 task = get_proc_task(file->f_path.dentry->d_inode);
1206 if (!lock_task_sighand(task, &flags)) {
1211 if (oom_score_adj < task->signal->oom_score_adj_min &&
1212 !capable(CAP_SYS_RESOURCE)) {
1217 if (oom_score_adj != task->signal->oom_score_adj) {
1218 if (oom_score_adj == OOM_SCORE_ADJ_MIN)
1219 atomic_inc(&task->mm->oom_disable_count);
1220 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1221 atomic_dec(&task->mm->oom_disable_count);
1223 task->signal->oom_score_adj = oom_score_adj;
1224 if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1225 task->signal->oom_score_adj_min = oom_score_adj;
1227 * Scale /proc/pid/oom_adj appropriately ensuring that OOM_DISABLE is
1228 * always attainable.
1230 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1231 task->signal->oom_adj = OOM_DISABLE;
1233 task->signal->oom_adj = (oom_score_adj * OOM_ADJUST_MAX) /
1236 unlock_task_sighand(task, &flags);
1239 put_task_struct(task);
1241 return err < 0 ? err : count;
1244 static const struct file_operations proc_oom_score_adj_operations = {
1245 .read = oom_score_adj_read,
1246 .write = oom_score_adj_write,
1247 .llseek = default_llseek,
1250 #ifdef CONFIG_AUDITSYSCALL
1251 #define TMPBUFLEN 21
1252 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1253 size_t count, loff_t *ppos)
1255 struct inode * inode = file->f_path.dentry->d_inode;
1256 struct task_struct *task = get_proc_task(inode);
1258 char tmpbuf[TMPBUFLEN];
1262 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1263 audit_get_loginuid(task));
1264 put_task_struct(task);
1265 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1268 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1269 size_t count, loff_t *ppos)
1271 struct inode * inode = file->f_path.dentry->d_inode;
1276 if (!capable(CAP_AUDIT_CONTROL))
1280 if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1286 if (count >= PAGE_SIZE)
1287 count = PAGE_SIZE - 1;
1290 /* No partial writes. */
1293 page = (char*)__get_free_page(GFP_TEMPORARY);
1297 if (copy_from_user(page, buf, count))
1301 loginuid = simple_strtoul(page, &tmp, 10);
1307 length = audit_set_loginuid(current, loginuid);
1308 if (likely(length == 0))
1312 free_page((unsigned long) page);
1316 static const struct file_operations proc_loginuid_operations = {
1317 .read = proc_loginuid_read,
1318 .write = proc_loginuid_write,
1319 .llseek = generic_file_llseek,
1322 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1323 size_t count, loff_t *ppos)
1325 struct inode * inode = file->f_path.dentry->d_inode;
1326 struct task_struct *task = get_proc_task(inode);
1328 char tmpbuf[TMPBUFLEN];
1332 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1333 audit_get_sessionid(task));
1334 put_task_struct(task);
1335 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1338 static const struct file_operations proc_sessionid_operations = {
1339 .read = proc_sessionid_read,
1340 .llseek = generic_file_llseek,
1344 #ifdef CONFIG_FAULT_INJECTION
1345 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1346 size_t count, loff_t *ppos)
1348 struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1349 char buffer[PROC_NUMBUF];
1355 make_it_fail = task->make_it_fail;
1356 put_task_struct(task);
1358 len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1360 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1363 static ssize_t proc_fault_inject_write(struct file * file,
1364 const char __user * buf, size_t count, loff_t *ppos)
1366 struct task_struct *task;
1367 char buffer[PROC_NUMBUF], *end;
1370 if (!capable(CAP_SYS_RESOURCE))
1372 memset(buffer, 0, sizeof(buffer));
1373 if (count > sizeof(buffer) - 1)
1374 count = sizeof(buffer) - 1;
1375 if (copy_from_user(buffer, buf, count))
1377 make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1380 task = get_proc_task(file->f_dentry->d_inode);
1383 task->make_it_fail = make_it_fail;
1384 put_task_struct(task);
1389 static const struct file_operations proc_fault_inject_operations = {
1390 .read = proc_fault_inject_read,
1391 .write = proc_fault_inject_write,
1392 .llseek = generic_file_llseek,
1397 #ifdef CONFIG_SCHED_DEBUG
1399 * Print out various scheduling related per-task fields:
1401 static int sched_show(struct seq_file *m, void *v)
1403 struct inode *inode = m->private;
1404 struct task_struct *p;
1406 p = get_proc_task(inode);
1409 proc_sched_show_task(p, m);
1417 sched_write(struct file *file, const char __user *buf,
1418 size_t count, loff_t *offset)
1420 struct inode *inode = file->f_path.dentry->d_inode;
1421 struct task_struct *p;
1423 p = get_proc_task(inode);
1426 proc_sched_set_task(p);
1433 static int sched_open(struct inode *inode, struct file *filp)
1435 return single_open(filp, sched_show, inode);
1438 static const struct file_operations proc_pid_sched_operations = {
1441 .write = sched_write,
1442 .llseek = seq_lseek,
1443 .release = single_release,
1448 #ifdef CONFIG_SCHED_AUTOGROUP
1450 * Print out autogroup related information:
1452 static int sched_autogroup_show(struct seq_file *m, void *v)
1454 struct inode *inode = m->private;
1455 struct task_struct *p;
1457 p = get_proc_task(inode);
1460 proc_sched_autogroup_show_task(p, m);
1468 sched_autogroup_write(struct file *file, const char __user *buf,
1469 size_t count, loff_t *offset)
1471 struct inode *inode = file->f_path.dentry->d_inode;
1472 struct task_struct *p;
1473 char buffer[PROC_NUMBUF];
1477 memset(buffer, 0, sizeof(buffer));
1478 if (count > sizeof(buffer) - 1)
1479 count = sizeof(buffer) - 1;
1480 if (copy_from_user(buffer, buf, count))
1483 err = strict_strtol(strstrip(buffer), 0, &nice);
1487 p = get_proc_task(inode);
1492 err = proc_sched_autogroup_set_nice(p, &err);
1501 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1505 ret = single_open(filp, sched_autogroup_show, NULL);
1507 struct seq_file *m = filp->private_data;
1514 static const struct file_operations proc_pid_sched_autogroup_operations = {
1515 .open = sched_autogroup_open,
1517 .write = sched_autogroup_write,
1518 .llseek = seq_lseek,
1519 .release = single_release,
1522 #endif /* CONFIG_SCHED_AUTOGROUP */
1524 static ssize_t comm_write(struct file *file, const char __user *buf,
1525 size_t count, loff_t *offset)
1527 struct inode *inode = file->f_path.dentry->d_inode;
1528 struct task_struct *p;
1529 char buffer[TASK_COMM_LEN];
1531 memset(buffer, 0, sizeof(buffer));
1532 if (count > sizeof(buffer) - 1)
1533 count = sizeof(buffer) - 1;
1534 if (copy_from_user(buffer, buf, count))
1537 p = get_proc_task(inode);
1541 if (same_thread_group(current, p))
1542 set_task_comm(p, buffer);
1551 static int comm_show(struct seq_file *m, void *v)
1553 struct inode *inode = m->private;
1554 struct task_struct *p;
1556 p = get_proc_task(inode);
1561 seq_printf(m, "%s\n", p->comm);
1569 static int comm_open(struct inode *inode, struct file *filp)
1571 return single_open(filp, comm_show, inode);
1574 static const struct file_operations proc_pid_set_comm_operations = {
1577 .write = comm_write,
1578 .llseek = seq_lseek,
1579 .release = single_release,
1583 * We added or removed a vma mapping the executable. The vmas are only mapped
1584 * during exec and are not mapped with the mmap system call.
1585 * Callers must hold down_write() on the mm's mmap_sem for these
1587 void added_exe_file_vma(struct mm_struct *mm)
1589 mm->num_exe_file_vmas++;
1592 void removed_exe_file_vma(struct mm_struct *mm)
1594 mm->num_exe_file_vmas--;
1595 if ((mm->num_exe_file_vmas == 0) && mm->exe_file){
1597 mm->exe_file = NULL;
1602 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1605 get_file(new_exe_file);
1608 mm->exe_file = new_exe_file;
1609 mm->num_exe_file_vmas = 0;
1612 struct file *get_mm_exe_file(struct mm_struct *mm)
1614 struct file *exe_file;
1616 /* We need mmap_sem to protect against races with removal of
1617 * VM_EXECUTABLE vmas */
1618 down_read(&mm->mmap_sem);
1619 exe_file = mm->exe_file;
1622 up_read(&mm->mmap_sem);
1626 void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
1628 /* It's safe to write the exe_file pointer without exe_file_lock because
1629 * this is called during fork when the task is not yet in /proc */
1630 newmm->exe_file = get_mm_exe_file(oldmm);
1633 static int proc_exe_link(struct inode *inode, struct path *exe_path)
1635 struct task_struct *task;
1636 struct mm_struct *mm;
1637 struct file *exe_file;
1639 task = get_proc_task(inode);
1642 mm = get_task_mm(task);
1643 put_task_struct(task);
1646 exe_file = get_mm_exe_file(mm);
1649 *exe_path = exe_file->f_path;
1650 path_get(&exe_file->f_path);
1657 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1659 struct inode *inode = dentry->d_inode;
1660 int error = -EACCES;
1662 /* We don't need a base pointer in the /proc filesystem */
1663 path_put(&nd->path);
1665 /* Are we allowed to snoop on the tasks file descriptors? */
1666 if (!proc_fd_access_allowed(inode))
1669 error = PROC_I(inode)->op.proc_get_link(inode, &nd->path);
1671 return ERR_PTR(error);
1674 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1676 char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1683 pathname = d_path(path, tmp, PAGE_SIZE);
1684 len = PTR_ERR(pathname);
1685 if (IS_ERR(pathname))
1687 len = tmp + PAGE_SIZE - 1 - pathname;
1691 if (copy_to_user(buffer, pathname, len))
1694 free_page((unsigned long)tmp);
1698 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1700 int error = -EACCES;
1701 struct inode *inode = dentry->d_inode;
1704 /* Are we allowed to snoop on the tasks file descriptors? */
1705 if (!proc_fd_access_allowed(inode))
1708 error = PROC_I(inode)->op.proc_get_link(inode, &path);
1712 error = do_proc_readlink(&path, buffer, buflen);
1718 static const struct inode_operations proc_pid_link_inode_operations = {
1719 .readlink = proc_pid_readlink,
1720 .follow_link = proc_pid_follow_link,
1721 .setattr = proc_setattr,
1725 /* building an inode */
1727 static int task_dumpable(struct task_struct *task)
1730 struct mm_struct *mm;
1735 dumpable = get_dumpable(mm);
1743 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1745 struct inode * inode;
1746 struct proc_inode *ei;
1747 const struct cred *cred;
1749 /* We need a new inode */
1751 inode = new_inode(sb);
1757 inode->i_ino = get_next_ino();
1758 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1759 inode->i_op = &proc_def_inode_operations;
1762 * grab the reference to task.
1764 ei->pid = get_task_pid(task, PIDTYPE_PID);
1768 if (task_dumpable(task)) {
1770 cred = __task_cred(task);
1771 inode->i_uid = cred->euid;
1772 inode->i_gid = cred->egid;
1775 security_task_to_inode(task, inode);
1785 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1787 struct inode *inode = dentry->d_inode;
1788 struct task_struct *task;
1789 const struct cred *cred;
1791 generic_fillattr(inode, stat);
1796 task = pid_task(proc_pid(inode), PIDTYPE_PID);
1798 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1799 task_dumpable(task)) {
1800 cred = __task_cred(task);
1801 stat->uid = cred->euid;
1802 stat->gid = cred->egid;
1812 * Exceptional case: normally we are not allowed to unhash a busy
1813 * directory. In this case, however, we can do it - no aliasing problems
1814 * due to the way we treat inodes.
1816 * Rewrite the inode's ownerships here because the owning task may have
1817 * performed a setuid(), etc.
1819 * Before the /proc/pid/status file was created the only way to read
1820 * the effective uid of a /process was to stat /proc/pid. Reading
1821 * /proc/pid/status is slow enough that procps and other packages
1822 * kept stating /proc/pid. To keep the rules in /proc simple I have
1823 * made this apply to all per process world readable and executable
1826 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1828 struct inode *inode;
1829 struct task_struct *task;
1830 const struct cred *cred;
1832 if (nd && nd->flags & LOOKUP_RCU)
1835 inode = dentry->d_inode;
1836 task = get_proc_task(inode);
1839 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1840 task_dumpable(task)) {
1842 cred = __task_cred(task);
1843 inode->i_uid = cred->euid;
1844 inode->i_gid = cred->egid;
1850 inode->i_mode &= ~(S_ISUID | S_ISGID);
1851 security_task_to_inode(task, inode);
1852 put_task_struct(task);
1859 static int pid_delete_dentry(const struct dentry * dentry)
1861 /* Is the task we represent dead?
1862 * If so, then don't put the dentry on the lru list,
1863 * kill it immediately.
1865 return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1868 static const struct dentry_operations pid_dentry_operations =
1870 .d_revalidate = pid_revalidate,
1871 .d_delete = pid_delete_dentry,
1876 typedef struct dentry *instantiate_t(struct inode *, struct dentry *,
1877 struct task_struct *, const void *);
1880 * Fill a directory entry.
1882 * If possible create the dcache entry and derive our inode number and
1883 * file type from dcache entry.
1885 * Since all of the proc inode numbers are dynamically generated, the inode
1886 * numbers do not exist until the inode is cache. This means creating the
1887 * the dcache entry in readdir is necessary to keep the inode numbers
1888 * reported by readdir in sync with the inode numbers reported
1891 static int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1892 char *name, int len,
1893 instantiate_t instantiate, struct task_struct *task, const void *ptr)
1895 struct dentry *child, *dir = filp->f_path.dentry;
1896 struct inode *inode;
1899 unsigned type = DT_UNKNOWN;
1903 qname.hash = full_name_hash(name, len);
1905 child = d_lookup(dir, &qname);
1908 new = d_alloc(dir, &qname);
1910 child = instantiate(dir->d_inode, new, task, ptr);
1917 if (!child || IS_ERR(child) || !child->d_inode)
1918 goto end_instantiate;
1919 inode = child->d_inode;
1922 type = inode->i_mode >> 12;
1927 ino = find_inode_number(dir, &qname);
1930 return filldir(dirent, name, len, filp->f_pos, ino, type);
1933 static unsigned name_to_int(struct dentry *dentry)
1935 const char *name = dentry->d_name.name;
1936 int len = dentry->d_name.len;
1939 if (len > 1 && *name == '0')
1942 unsigned c = *name++ - '0';
1945 if (n >= (~0U-9)/10)
1955 #define PROC_FDINFO_MAX 64
1957 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1959 struct task_struct *task = get_proc_task(inode);
1960 struct files_struct *files = NULL;
1962 int fd = proc_fd(inode);
1965 files = get_files_struct(task);
1966 put_task_struct(task);
1970 * We are not taking a ref to the file structure, so we must
1973 spin_lock(&files->file_lock);
1974 file = fcheck_files(files, fd);
1977 *path = file->f_path;
1978 path_get(&file->f_path);
1981 snprintf(info, PROC_FDINFO_MAX,
1984 (long long) file->f_pos,
1986 spin_unlock(&files->file_lock);
1987 put_files_struct(files);
1990 spin_unlock(&files->file_lock);
1991 put_files_struct(files);
1996 static int proc_fd_link(struct inode *inode, struct path *path)
1998 return proc_fd_info(inode, path, NULL);
2001 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
2003 struct inode *inode;
2004 struct task_struct *task;
2006 struct files_struct *files;
2007 const struct cred *cred;
2009 if (nd && nd->flags & LOOKUP_RCU)
2012 inode = dentry->d_inode;
2013 task = get_proc_task(inode);
2014 fd = proc_fd(inode);
2017 files = get_files_struct(task);
2020 if (fcheck_files(files, fd)) {
2022 put_files_struct(files);
2023 if (task_dumpable(task)) {
2025 cred = __task_cred(task);
2026 inode->i_uid = cred->euid;
2027 inode->i_gid = cred->egid;
2033 inode->i_mode &= ~(S_ISUID | S_ISGID);
2034 security_task_to_inode(task, inode);
2035 put_task_struct(task);
2039 put_files_struct(files);
2041 put_task_struct(task);
2047 static const struct dentry_operations tid_fd_dentry_operations =
2049 .d_revalidate = tid_fd_revalidate,
2050 .d_delete = pid_delete_dentry,
2053 static struct dentry *proc_fd_instantiate(struct inode *dir,
2054 struct dentry *dentry, struct task_struct *task, const void *ptr)
2056 unsigned fd = *(const unsigned *)ptr;
2058 struct files_struct *files;
2059 struct inode *inode;
2060 struct proc_inode *ei;
2061 struct dentry *error = ERR_PTR(-ENOENT);
2063 inode = proc_pid_make_inode(dir->i_sb, task);
2068 files = get_files_struct(task);
2071 inode->i_mode = S_IFLNK;
2074 * We are not taking a ref to the file structure, so we must
2077 spin_lock(&files->file_lock);
2078 file = fcheck_files(files, fd);
2081 if (file->f_mode & FMODE_READ)
2082 inode->i_mode |= S_IRUSR | S_IXUSR;
2083 if (file->f_mode & FMODE_WRITE)
2084 inode->i_mode |= S_IWUSR | S_IXUSR;
2085 spin_unlock(&files->file_lock);
2086 put_files_struct(files);
2088 inode->i_op = &proc_pid_link_inode_operations;
2090 ei->op.proc_get_link = proc_fd_link;
2091 d_set_d_op(dentry, &tid_fd_dentry_operations);
2092 d_add(dentry, inode);
2093 /* Close the race of the process dying before we return the dentry */
2094 if (tid_fd_revalidate(dentry, NULL))
2100 spin_unlock(&files->file_lock);
2101 put_files_struct(files);
2107 static struct dentry *proc_lookupfd_common(struct inode *dir,
2108 struct dentry *dentry,
2109 instantiate_t instantiate)
2111 struct task_struct *task = get_proc_task(dir);
2112 unsigned fd = name_to_int(dentry);
2113 struct dentry *result = ERR_PTR(-ENOENT);
2120 result = instantiate(dir, dentry, task, &fd);
2122 put_task_struct(task);
2127 static int proc_readfd_common(struct file * filp, void * dirent,
2128 filldir_t filldir, instantiate_t instantiate)
2130 struct dentry *dentry = filp->f_path.dentry;
2131 struct inode *inode = dentry->d_inode;
2132 struct task_struct *p = get_proc_task(inode);
2133 unsigned int fd, ino;
2135 struct files_struct * files;
2145 if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
2149 ino = parent_ino(dentry);
2150 if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
2154 files = get_files_struct(p);
2158 for (fd = filp->f_pos-2;
2159 fd < files_fdtable(files)->max_fds;
2160 fd++, filp->f_pos++) {
2161 char name[PROC_NUMBUF];
2164 if (!fcheck_files(files, fd))
2168 len = snprintf(name, sizeof(name), "%d", fd);
2169 if (proc_fill_cache(filp, dirent, filldir,
2170 name, len, instantiate,
2178 put_files_struct(files);
2186 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
2187 struct nameidata *nd)
2189 return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
2192 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
2194 return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
2197 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
2198 size_t len, loff_t *ppos)
2200 char tmp[PROC_FDINFO_MAX];
2201 int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
2203 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
2207 static const struct file_operations proc_fdinfo_file_operations = {
2208 .open = nonseekable_open,
2209 .read = proc_fdinfo_read,
2210 .llseek = no_llseek,
2213 static const struct file_operations proc_fd_operations = {
2214 .read = generic_read_dir,
2215 .readdir = proc_readfd,
2216 .llseek = default_llseek,
2220 * /proc/pid/fd needs a special permission handler so that a process can still
2221 * access /proc/self/fd after it has executed a setuid().
2223 static int proc_fd_permission(struct inode *inode, int mask, unsigned int flags)
2227 if (flags & IPERM_FLAG_RCU)
2229 rv = generic_permission(inode, mask, flags, NULL);
2232 if (task_pid(current) == proc_pid(inode))
2238 * proc directories can do almost nothing..
2240 static const struct inode_operations proc_fd_inode_operations = {
2241 .lookup = proc_lookupfd,
2242 .permission = proc_fd_permission,
2243 .setattr = proc_setattr,
2246 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
2247 struct dentry *dentry, struct task_struct *task, const void *ptr)
2249 unsigned fd = *(unsigned *)ptr;
2250 struct inode *inode;
2251 struct proc_inode *ei;
2252 struct dentry *error = ERR_PTR(-ENOENT);
2254 inode = proc_pid_make_inode(dir->i_sb, task);
2259 inode->i_mode = S_IFREG | S_IRUSR;
2260 inode->i_fop = &proc_fdinfo_file_operations;
2261 d_set_d_op(dentry, &tid_fd_dentry_operations);
2262 d_add(dentry, inode);
2263 /* Close the race of the process dying before we return the dentry */
2264 if (tid_fd_revalidate(dentry, NULL))
2271 static struct dentry *proc_lookupfdinfo(struct inode *dir,
2272 struct dentry *dentry,
2273 struct nameidata *nd)
2275 return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
2278 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
2280 return proc_readfd_common(filp, dirent, filldir,
2281 proc_fdinfo_instantiate);
2284 static const struct file_operations proc_fdinfo_operations = {
2285 .read = generic_read_dir,
2286 .readdir = proc_readfdinfo,
2287 .llseek = default_llseek,
2291 * proc directories can do almost nothing..
2293 static const struct inode_operations proc_fdinfo_inode_operations = {
2294 .lookup = proc_lookupfdinfo,
2295 .setattr = proc_setattr,
2299 static struct dentry *proc_pident_instantiate(struct inode *dir,
2300 struct dentry *dentry, struct task_struct *task, const void *ptr)
2302 const struct pid_entry *p = ptr;
2303 struct inode *inode;
2304 struct proc_inode *ei;
2305 struct dentry *error = ERR_PTR(-ENOENT);
2307 inode = proc_pid_make_inode(dir->i_sb, task);
2312 inode->i_mode = p->mode;
2313 if (S_ISDIR(inode->i_mode))
2314 inode->i_nlink = 2; /* Use getattr to fix if necessary */
2316 inode->i_op = p->iop;
2318 inode->i_fop = p->fop;
2320 d_set_d_op(dentry, &pid_dentry_operations);
2321 d_add(dentry, inode);
2322 /* Close the race of the process dying before we return the dentry */
2323 if (pid_revalidate(dentry, NULL))
2329 static struct dentry *proc_pident_lookup(struct inode *dir,
2330 struct dentry *dentry,
2331 const struct pid_entry *ents,
2334 struct dentry *error;
2335 struct task_struct *task = get_proc_task(dir);
2336 const struct pid_entry *p, *last;
2338 error = ERR_PTR(-ENOENT);
2344 * Yes, it does not scale. And it should not. Don't add
2345 * new entries into /proc/<tgid>/ without very good reasons.
2347 last = &ents[nents - 1];
2348 for (p = ents; p <= last; p++) {
2349 if (p->len != dentry->d_name.len)
2351 if (!memcmp(dentry->d_name.name, p->name, p->len))
2357 error = proc_pident_instantiate(dir, dentry, task, p);
2359 put_task_struct(task);
2364 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2365 filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2367 return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2368 proc_pident_instantiate, task, p);
2371 static int proc_pident_readdir(struct file *filp,
2372 void *dirent, filldir_t filldir,
2373 const struct pid_entry *ents, unsigned int nents)
2376 struct dentry *dentry = filp->f_path.dentry;
2377 struct inode *inode = dentry->d_inode;
2378 struct task_struct *task = get_proc_task(inode);
2379 const struct pid_entry *p, *last;
2392 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2398 ino = parent_ino(dentry);
2399 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2411 last = &ents[nents - 1];
2413 if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2422 put_task_struct(task);
2427 #ifdef CONFIG_SECURITY
2428 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2429 size_t count, loff_t *ppos)
2431 struct inode * inode = file->f_path.dentry->d_inode;
2434 struct task_struct *task = get_proc_task(inode);
2439 length = security_getprocattr(task,
2440 (char*)file->f_path.dentry->d_name.name,
2442 put_task_struct(task);
2444 length = simple_read_from_buffer(buf, count, ppos, p, length);
2449 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2450 size_t count, loff_t *ppos)
2452 struct inode * inode = file->f_path.dentry->d_inode;
2455 struct task_struct *task = get_proc_task(inode);
2460 if (count > PAGE_SIZE)
2463 /* No partial writes. */
2469 page = (char*)__get_free_page(GFP_TEMPORARY);
2474 if (copy_from_user(page, buf, count))
2477 /* Guard against adverse ptrace interaction */
2478 length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2482 length = security_setprocattr(task,
2483 (char*)file->f_path.dentry->d_name.name,
2484 (void*)page, count);
2485 mutex_unlock(&task->signal->cred_guard_mutex);
2487 free_page((unsigned long) page);
2489 put_task_struct(task);
2494 static const struct file_operations proc_pid_attr_operations = {
2495 .read = proc_pid_attr_read,
2496 .write = proc_pid_attr_write,
2497 .llseek = generic_file_llseek,
2500 static const struct pid_entry attr_dir_stuff[] = {
2501 REG("current", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2502 REG("prev", S_IRUGO, proc_pid_attr_operations),
2503 REG("exec", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2504 REG("fscreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2505 REG("keycreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2506 REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2509 static int proc_attr_dir_readdir(struct file * filp,
2510 void * dirent, filldir_t filldir)
2512 return proc_pident_readdir(filp,dirent,filldir,
2513 attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2516 static const struct file_operations proc_attr_dir_operations = {
2517 .read = generic_read_dir,
2518 .readdir = proc_attr_dir_readdir,
2519 .llseek = default_llseek,
2522 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2523 struct dentry *dentry, struct nameidata *nd)
2525 return proc_pident_lookup(dir, dentry,
2526 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2529 static const struct inode_operations proc_attr_dir_inode_operations = {
2530 .lookup = proc_attr_dir_lookup,
2531 .getattr = pid_getattr,
2532 .setattr = proc_setattr,
2537 #ifdef CONFIG_ELF_CORE
2538 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2539 size_t count, loff_t *ppos)
2541 struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2542 struct mm_struct *mm;
2543 char buffer[PROC_NUMBUF];
2551 mm = get_task_mm(task);
2553 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2554 ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2555 MMF_DUMP_FILTER_SHIFT));
2557 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2560 put_task_struct(task);
2565 static ssize_t proc_coredump_filter_write(struct file *file,
2566 const char __user *buf,
2570 struct task_struct *task;
2571 struct mm_struct *mm;
2572 char buffer[PROC_NUMBUF], *end;
2579 memset(buffer, 0, sizeof(buffer));
2580 if (count > sizeof(buffer) - 1)
2581 count = sizeof(buffer) - 1;
2582 if (copy_from_user(buffer, buf, count))
2586 val = (unsigned int)simple_strtoul(buffer, &end, 0);
2589 if (end - buffer == 0)
2593 task = get_proc_task(file->f_dentry->d_inode);
2598 mm = get_task_mm(task);
2602 for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2604 set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2606 clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2611 put_task_struct(task);
2616 static const struct file_operations proc_coredump_filter_operations = {
2617 .read = proc_coredump_filter_read,
2618 .write = proc_coredump_filter_write,
2619 .llseek = generic_file_llseek,
2626 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2629 struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2630 pid_t tgid = task_tgid_nr_ns(current, ns);
2631 char tmp[PROC_NUMBUF];
2634 sprintf(tmp, "%d", tgid);
2635 return vfs_readlink(dentry,buffer,buflen,tmp);
2638 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2640 struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2641 pid_t tgid = task_tgid_nr_ns(current, ns);
2642 char *name = ERR_PTR(-ENOENT);
2646 name = ERR_PTR(-ENOMEM);
2648 sprintf(name, "%d", tgid);
2650 nd_set_link(nd, name);
2654 static void proc_self_put_link(struct dentry *dentry, struct nameidata *nd,
2657 char *s = nd_get_link(nd);
2662 static const struct inode_operations proc_self_inode_operations = {
2663 .readlink = proc_self_readlink,
2664 .follow_link = proc_self_follow_link,
2665 .put_link = proc_self_put_link,
2671 * These are the directory entries in the root directory of /proc
2672 * that properly belong to the /proc filesystem, as they describe
2673 * describe something that is process related.
2675 static const struct pid_entry proc_base_stuff[] = {
2676 NOD("self", S_IFLNK|S_IRWXUGO,
2677 &proc_self_inode_operations, NULL, {}),
2680 static struct dentry *proc_base_instantiate(struct inode *dir,
2681 struct dentry *dentry, struct task_struct *task, const void *ptr)
2683 const struct pid_entry *p = ptr;
2684 struct inode *inode;
2685 struct proc_inode *ei;
2686 struct dentry *error;
2688 /* Allocate the inode */
2689 error = ERR_PTR(-ENOMEM);
2690 inode = new_inode(dir->i_sb);
2694 /* Initialize the inode */
2696 inode->i_ino = get_next_ino();
2697 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2700 * grab the reference to the task.
2702 ei->pid = get_task_pid(task, PIDTYPE_PID);
2706 inode->i_mode = p->mode;
2707 if (S_ISDIR(inode->i_mode))
2709 if (S_ISLNK(inode->i_mode))
2712 inode->i_op = p->iop;
2714 inode->i_fop = p->fop;
2716 d_add(dentry, inode);
2725 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2727 struct dentry *error;
2728 struct task_struct *task = get_proc_task(dir);
2729 const struct pid_entry *p, *last;
2731 error = ERR_PTR(-ENOENT);
2736 /* Lookup the directory entry */
2737 last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2738 for (p = proc_base_stuff; p <= last; p++) {
2739 if (p->len != dentry->d_name.len)
2741 if (!memcmp(dentry->d_name.name, p->name, p->len))
2747 error = proc_base_instantiate(dir, dentry, task, p);
2750 put_task_struct(task);
2755 static int proc_base_fill_cache(struct file *filp, void *dirent,
2756 filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2758 return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2759 proc_base_instantiate, task, p);
2762 #ifdef CONFIG_TASK_IO_ACCOUNTING
2763 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2765 struct task_io_accounting acct = task->ioac;
2766 unsigned long flags;
2768 if (whole && lock_task_sighand(task, &flags)) {
2769 struct task_struct *t = task;
2771 task_io_accounting_add(&acct, &task->signal->ioac);
2772 while_each_thread(task, t)
2773 task_io_accounting_add(&acct, &t->ioac);
2775 unlock_task_sighand(task, &flags);
2777 return sprintf(buffer,
2782 "read_bytes: %llu\n"
2783 "write_bytes: %llu\n"
2784 "cancelled_write_bytes: %llu\n",
2785 (unsigned long long)acct.rchar,
2786 (unsigned long long)acct.wchar,
2787 (unsigned long long)acct.syscr,
2788 (unsigned long long)acct.syscw,
2789 (unsigned long long)acct.read_bytes,
2790 (unsigned long long)acct.write_bytes,
2791 (unsigned long long)acct.cancelled_write_bytes);
2794 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2796 return do_io_accounting(task, buffer, 0);
2799 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2801 return do_io_accounting(task, buffer, 1);
2803 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2805 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2806 struct pid *pid, struct task_struct *task)
2808 int err = lock_trace(task);
2810 seq_printf(m, "%08x\n", task->personality);
2819 static const struct file_operations proc_task_operations;
2820 static const struct inode_operations proc_task_inode_operations;
2822 static const struct pid_entry tgid_base_stuff[] = {
2823 DIR("task", S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2824 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2825 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2827 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2829 REG("environ", S_IRUSR, proc_environ_operations),
2830 INF("auxv", S_IRUSR, proc_pid_auxv),
2831 ONE("status", S_IRUGO, proc_pid_status),
2832 ONE("personality", S_IRUGO, proc_pid_personality),
2833 INF("limits", S_IRUGO, proc_pid_limits),
2834 #ifdef CONFIG_SCHED_DEBUG
2835 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2837 #ifdef CONFIG_SCHED_AUTOGROUP
2838 REG("autogroup", S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2840 REG("comm", S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2841 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2842 INF("syscall", S_IRUGO, proc_pid_syscall),
2844 INF("cmdline", S_IRUGO, proc_pid_cmdline),
2845 ONE("stat", S_IRUGO, proc_tgid_stat),
2846 ONE("statm", S_IRUGO, proc_pid_statm),
2847 REG("maps", S_IRUGO, proc_maps_operations),
2849 REG("numa_maps", S_IRUGO, proc_numa_maps_operations),
2851 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
2852 LNK("cwd", proc_cwd_link),
2853 LNK("root", proc_root_link),
2854 LNK("exe", proc_exe_link),
2855 REG("mounts", S_IRUGO, proc_mounts_operations),
2856 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
2857 REG("mountstats", S_IRUSR, proc_mountstats_operations),
2858 #ifdef CONFIG_PROC_PAGE_MONITOR
2859 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2860 REG("smaps", S_IRUGO, proc_smaps_operations),
2861 REG("pagemap", S_IRUGO, proc_pagemap_operations),
2863 #ifdef CONFIG_SECURITY
2864 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2866 #ifdef CONFIG_KALLSYMS
2867 INF("wchan", S_IRUGO, proc_pid_wchan),
2869 #ifdef CONFIG_STACKTRACE
2870 ONE("stack", S_IRUGO, proc_pid_stack),
2872 #ifdef CONFIG_SCHEDSTATS
2873 INF("schedstat", S_IRUGO, proc_pid_schedstat),
2875 #ifdef CONFIG_LATENCYTOP
2876 REG("latency", S_IRUGO, proc_lstats_operations),
2878 #ifdef CONFIG_PROC_PID_CPUSET
2879 REG("cpuset", S_IRUGO, proc_cpuset_operations),
2881 #ifdef CONFIG_CGROUPS
2882 REG("cgroup", S_IRUGO, proc_cgroup_operations),
2884 INF("oom_score", S_IRUGO, proc_oom_score),
2885 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2886 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2887 #ifdef CONFIG_AUDITSYSCALL
2888 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
2889 REG("sessionid", S_IRUGO, proc_sessionid_operations),
2891 #ifdef CONFIG_FAULT_INJECTION
2892 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2894 #ifdef CONFIG_ELF_CORE
2895 REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2897 #ifdef CONFIG_TASK_IO_ACCOUNTING
2898 INF("io", S_IRUGO, proc_tgid_io_accounting),
2900 #ifdef CONFIG_HARDWALL
2901 INF("hardwall", S_IRUGO, proc_pid_hardwall),
2905 static int proc_tgid_base_readdir(struct file * filp,
2906 void * dirent, filldir_t filldir)
2908 return proc_pident_readdir(filp,dirent,filldir,
2909 tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2912 static const struct file_operations proc_tgid_base_operations = {
2913 .read = generic_read_dir,
2914 .readdir = proc_tgid_base_readdir,
2915 .llseek = default_llseek,
2918 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2919 return proc_pident_lookup(dir, dentry,
2920 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2923 static const struct inode_operations proc_tgid_base_inode_operations = {
2924 .lookup = proc_tgid_base_lookup,
2925 .getattr = pid_getattr,
2926 .setattr = proc_setattr,
2929 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2931 struct dentry *dentry, *leader, *dir;
2932 char buf[PROC_NUMBUF];
2936 name.len = snprintf(buf, sizeof(buf), "%d", pid);
2937 dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2939 shrink_dcache_parent(dentry);
2945 name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2946 leader = d_hash_and_lookup(mnt->mnt_root, &name);
2951 name.len = strlen(name.name);
2952 dir = d_hash_and_lookup(leader, &name);
2954 goto out_put_leader;
2957 name.len = snprintf(buf, sizeof(buf), "%d", pid);
2958 dentry = d_hash_and_lookup(dir, &name);
2960 shrink_dcache_parent(dentry);
2973 * proc_flush_task - Remove dcache entries for @task from the /proc dcache.
2974 * @task: task that should be flushed.
2976 * When flushing dentries from proc, one needs to flush them from global
2977 * proc (proc_mnt) and from all the namespaces' procs this task was seen
2978 * in. This call is supposed to do all of this job.
2980 * Looks in the dcache for
2982 * /proc/@tgid/task/@pid
2983 * if either directory is present flushes it and all of it'ts children
2986 * It is safe and reasonable to cache /proc entries for a task until
2987 * that task exits. After that they just clog up the dcache with
2988 * useless entries, possibly causing useful dcache entries to be
2989 * flushed instead. This routine is proved to flush those useless
2990 * dcache entries at process exit time.
2992 * NOTE: This routine is just an optimization so it does not guarantee
2993 * that no dcache entries will exist at process exit time it
2994 * just makes it very unlikely that any will persist.
2997 void proc_flush_task(struct task_struct *task)
3000 struct pid *pid, *tgid;
3003 pid = task_pid(task);
3004 tgid = task_tgid(task);
3006 for (i = 0; i <= pid->level; i++) {
3007 upid = &pid->numbers[i];
3008 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3009 tgid->numbers[i].nr);
3012 upid = &pid->numbers[pid->level];
3014 pid_ns_release_proc(upid->ns);
3017 static struct dentry *proc_pid_instantiate(struct inode *dir,
3018 struct dentry * dentry,
3019 struct task_struct *task, const void *ptr)
3021 struct dentry *error = ERR_PTR(-ENOENT);
3022 struct inode *inode;
3024 inode = proc_pid_make_inode(dir->i_sb, task);
3028 inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3029 inode->i_op = &proc_tgid_base_inode_operations;
3030 inode->i_fop = &proc_tgid_base_operations;
3031 inode->i_flags|=S_IMMUTABLE;
3033 inode->i_nlink = 2 + pid_entry_count_dirs(tgid_base_stuff,
3034 ARRAY_SIZE(tgid_base_stuff));
3036 d_set_d_op(dentry, &pid_dentry_operations);
3038 d_add(dentry, inode);
3039 /* Close the race of the process dying before we return the dentry */
3040 if (pid_revalidate(dentry, NULL))
3046 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3048 struct dentry *result;
3049 struct task_struct *task;
3051 struct pid_namespace *ns;
3053 result = proc_base_lookup(dir, dentry);
3054 if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
3057 tgid = name_to_int(dentry);
3061 ns = dentry->d_sb->s_fs_info;
3063 task = find_task_by_pid_ns(tgid, ns);
3065 get_task_struct(task);
3070 result = proc_pid_instantiate(dir, dentry, task, NULL);
3071 put_task_struct(task);
3077 * Find the first task with tgid >= tgid
3082 struct task_struct *task;
3084 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3089 put_task_struct(iter.task);
3093 pid = find_ge_pid(iter.tgid, ns);
3095 iter.tgid = pid_nr_ns(pid, ns);
3096 iter.task = pid_task(pid, PIDTYPE_PID);
3097 /* What we to know is if the pid we have find is the
3098 * pid of a thread_group_leader. Testing for task
3099 * being a thread_group_leader is the obvious thing
3100 * todo but there is a window when it fails, due to
3101 * the pid transfer logic in de_thread.
3103 * So we perform the straight forward test of seeing
3104 * if the pid we have found is the pid of a thread
3105 * group leader, and don't worry if the task we have
3106 * found doesn't happen to be a thread group leader.
3107 * As we don't care in the case of readdir.
3109 if (!iter.task || !has_group_leader_pid(iter.task)) {
3113 get_task_struct(iter.task);
3119 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
3121 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3122 struct tgid_iter iter)
3124 char name[PROC_NUMBUF];
3125 int len = snprintf(name, sizeof(name), "%d", iter.tgid);
3126 return proc_fill_cache(filp, dirent, filldir, name, len,
3127 proc_pid_instantiate, iter.task, NULL);
3130 /* for the /proc/ directory itself, after non-process stuff has been done */
3131 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
3134 struct task_struct *reaper;
3135 struct tgid_iter iter;
3136 struct pid_namespace *ns;
3138 if (filp->f_pos >= PID_MAX_LIMIT + TGID_OFFSET)
3140 nr = filp->f_pos - FIRST_PROCESS_ENTRY;
3142 reaper = get_proc_task(filp->f_path.dentry->d_inode);
3146 for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
3147 const struct pid_entry *p = &proc_base_stuff[nr];
3148 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
3152 ns = filp->f_dentry->d_sb->s_fs_info;
3154 iter.tgid = filp->f_pos - TGID_OFFSET;
3155 for (iter = next_tgid(ns, iter);
3157 iter.tgid += 1, iter = next_tgid(ns, iter)) {
3158 filp->f_pos = iter.tgid + TGID_OFFSET;
3159 if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
3160 put_task_struct(iter.task);
3164 filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
3166 put_task_struct(reaper);
3174 static const struct pid_entry tid_base_stuff[] = {
3175 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3176 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3177 REG("environ", S_IRUSR, proc_environ_operations),
3178 INF("auxv", S_IRUSR, proc_pid_auxv),
3179 ONE("status", S_IRUGO, proc_pid_status),
3180 ONE("personality", S_IRUGO, proc_pid_personality),
3181 INF("limits", S_IRUGO, proc_pid_limits),
3182 #ifdef CONFIG_SCHED_DEBUG
3183 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3185 REG("comm", S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3186 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3187 INF("syscall", S_IRUGO, proc_pid_syscall),
3189 INF("cmdline", S_IRUGO, proc_pid_cmdline),
3190 ONE("stat", S_IRUGO, proc_tid_stat),
3191 ONE("statm", S_IRUGO, proc_pid_statm),
3192 REG("maps", S_IRUGO, proc_maps_operations),
3194 REG("numa_maps", S_IRUGO, proc_numa_maps_operations),
3196 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
3197 LNK("cwd", proc_cwd_link),
3198 LNK("root", proc_root_link),
3199 LNK("exe", proc_exe_link),
3200 REG("mounts", S_IRUGO, proc_mounts_operations),
3201 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
3202 #ifdef CONFIG_PROC_PAGE_MONITOR
3203 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3204 REG("smaps", S_IRUGO, proc_smaps_operations),
3205 REG("pagemap", S_IRUGO, proc_pagemap_operations),
3207 #ifdef CONFIG_SECURITY
3208 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3210 #ifdef CONFIG_KALLSYMS
3211 INF("wchan", S_IRUGO, proc_pid_wchan),
3213 #ifdef CONFIG_STACKTRACE
3214 ONE("stack", S_IRUGO, proc_pid_stack),
3216 #ifdef CONFIG_SCHEDSTATS
3217 INF("schedstat", S_IRUGO, proc_pid_schedstat),
3219 #ifdef CONFIG_LATENCYTOP
3220 REG("latency", S_IRUGO, proc_lstats_operations),
3222 #ifdef CONFIG_PROC_PID_CPUSET
3223 REG("cpuset", S_IRUGO, proc_cpuset_operations),
3225 #ifdef CONFIG_CGROUPS
3226 REG("cgroup", S_IRUGO, proc_cgroup_operations),
3228 INF("oom_score", S_IRUGO, proc_oom_score),
3229 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3230 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3231 #ifdef CONFIG_AUDITSYSCALL
3232 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
3233 REG("sessionid", S_IRUGO, proc_sessionid_operations),
3235 #ifdef CONFIG_FAULT_INJECTION
3236 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3238 #ifdef CONFIG_TASK_IO_ACCOUNTING
3239 INF("io", S_IRUGO, proc_tid_io_accounting),
3241 #ifdef CONFIG_HARDWALL
3242 INF("hardwall", S_IRUGO, proc_pid_hardwall),
3246 static int proc_tid_base_readdir(struct file * filp,
3247 void * dirent, filldir_t filldir)
3249 return proc_pident_readdir(filp,dirent,filldir,
3250 tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
3253 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3254 return proc_pident_lookup(dir, dentry,
3255 tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3258 static const struct file_operations proc_tid_base_operations = {
3259 .read = generic_read_dir,
3260 .readdir = proc_tid_base_readdir,
3261 .llseek = default_llseek,
3264 static const struct inode_operations proc_tid_base_inode_operations = {
3265 .lookup = proc_tid_base_lookup,
3266 .getattr = pid_getattr,
3267 .setattr = proc_setattr,
3270 static struct dentry *proc_task_instantiate(struct inode *dir,
3271 struct dentry *dentry, struct task_struct *task, const void *ptr)
3273 struct dentry *error = ERR_PTR(-ENOENT);
3274 struct inode *inode;
3275 inode = proc_pid_make_inode(dir->i_sb, task);
3279 inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3280 inode->i_op = &proc_tid_base_inode_operations;
3281 inode->i_fop = &proc_tid_base_operations;
3282 inode->i_flags|=S_IMMUTABLE;
3284 inode->i_nlink = 2 + pid_entry_count_dirs(tid_base_stuff,
3285 ARRAY_SIZE(tid_base_stuff));
3287 d_set_d_op(dentry, &pid_dentry_operations);
3289 d_add(dentry, inode);
3290 /* Close the race of the process dying before we return the dentry */
3291 if (pid_revalidate(dentry, NULL))
3297 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3299 struct dentry *result = ERR_PTR(-ENOENT);
3300 struct task_struct *task;
3301 struct task_struct *leader = get_proc_task(dir);
3303 struct pid_namespace *ns;
3308 tid = name_to_int(dentry);
3312 ns = dentry->d_sb->s_fs_info;
3314 task = find_task_by_pid_ns(tid, ns);
3316 get_task_struct(task);
3320 if (!same_thread_group(leader, task))
3323 result = proc_task_instantiate(dir, dentry, task, NULL);
3325 put_task_struct(task);
3327 put_task_struct(leader);
3333 * Find the first tid of a thread group to return to user space.
3335 * Usually this is just the thread group leader, but if the users
3336 * buffer was too small or there was a seek into the middle of the
3337 * directory we have more work todo.
3339 * In the case of a short read we start with find_task_by_pid.
3341 * In the case of a seek we start with the leader and walk nr
3344 static struct task_struct *first_tid(struct task_struct *leader,
3345 int tid, int nr, struct pid_namespace *ns)
3347 struct task_struct *pos;
3350 /* Attempt to start with the pid of a thread */
3351 if (tid && (nr > 0)) {
3352 pos = find_task_by_pid_ns(tid, ns);
3353 if (pos && (pos->group_leader == leader))
3357 /* If nr exceeds the number of threads there is nothing todo */
3359 if (nr && nr >= get_nr_threads(leader))
3362 /* If we haven't found our starting place yet start
3363 * with the leader and walk nr threads forward.
3365 for (pos = leader; nr > 0; --nr) {
3366 pos = next_thread(pos);
3367 if (pos == leader) {
3373 get_task_struct(pos);
3380 * Find the next thread in the thread list.
3381 * Return NULL if there is an error or no next thread.
3383 * The reference to the input task_struct is released.
3385 static struct task_struct *next_tid(struct task_struct *start)
3387 struct task_struct *pos = NULL;
3389 if (pid_alive(start)) {
3390 pos = next_thread(start);
3391 if (thread_group_leader(pos))
3394 get_task_struct(pos);
3397 put_task_struct(start);
3401 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3402 struct task_struct *task, int tid)
3404 char name[PROC_NUMBUF];
3405 int len = snprintf(name, sizeof(name), "%d", tid);
3406 return proc_fill_cache(filp, dirent, filldir, name, len,
3407 proc_task_instantiate, task, NULL);
3410 /* for the /proc/TGID/task/ directories */
3411 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3413 struct dentry *dentry = filp->f_path.dentry;
3414 struct inode *inode = dentry->d_inode;
3415 struct task_struct *leader = NULL;
3416 struct task_struct *task;
3417 int retval = -ENOENT;
3420 struct pid_namespace *ns;
3422 task = get_proc_task(inode);
3426 if (pid_alive(task)) {
3427 leader = task->group_leader;
3428 get_task_struct(leader);
3431 put_task_struct(task);
3436 switch ((unsigned long)filp->f_pos) {
3439 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3444 ino = parent_ino(dentry);
3445 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3451 /* f_version caches the tgid value that the last readdir call couldn't
3452 * return. lseek aka telldir automagically resets f_version to 0.
3454 ns = filp->f_dentry->d_sb->s_fs_info;
3455 tid = (int)filp->f_version;
3456 filp->f_version = 0;
3457 for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3459 task = next_tid(task), filp->f_pos++) {
3460 tid = task_pid_nr_ns(task, ns);
3461 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3462 /* returning this tgid failed, save it as the first
3463 * pid for the next readir call */
3464 filp->f_version = (u64)tid;
3465 put_task_struct(task);
3470 put_task_struct(leader);
3475 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3477 struct inode *inode = dentry->d_inode;
3478 struct task_struct *p = get_proc_task(inode);
3479 generic_fillattr(inode, stat);
3482 stat->nlink += get_nr_threads(p);
3489 static const struct inode_operations proc_task_inode_operations = {
3490 .lookup = proc_task_lookup,
3491 .getattr = proc_task_getattr,
3492 .setattr = proc_setattr,
3495 static const struct file_operations proc_task_operations = {
3496 .read = generic_read_dir,
3497 .readdir = proc_task_readdir,
3498 .llseek = default_llseek,