fs/proc/base.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/proc/base.c
   4  *
   5  *  Copyright (C) 1991, 1992 Linus Torvalds
   6  *
   7  *  proc base directory handling functions
   8  *
   9  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
  10  *  Instead of using magical inumbers to determine the kind of object
  11  *  we allocate and fill in-core inodes upon lookup. They don't even
  12  *  go into icache. We cache the reference to task_struct upon lookup too.
  13  *  Eventually it should become a filesystem in its own. We don't use the
  14  *  rest of procfs anymore.
  15  *
  16  *
  17  *  Changelog:
  18  *  17-Jan-2005
  19  *  Allan Bezerra
  20  *  Bruna Moreira <bruna.moreira@indt.org.br>
  21  *  Edjard Mota <edjard.mota@indt.org.br>
  22  *  Ilias Biris <ilias.biris@indt.org.br>
  23  *  Mauricio Lin <mauricio.lin@indt.org.br>
  24  *
  25  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  26  *
  27  *  A new process specific entry (smaps) included in /proc. It shows the
  28  *  size of rss for each memory area. The maps entry lacks information
  29  *  about physical memory size (rss) for each mapped file, i.e.,
  30  *  rss information for executables and library files.
  31  *  This additional information is useful for any tools that need to know
  32  *  about physical memory consumption for a process specific library.
  33  *
  34  *  Changelog:
  35  *  21-Feb-2005
  36  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  37  *  Pud inclusion in the page table walking.
  38  *
  39  *  ChangeLog:
  40  *  10-Mar-2005
  41  *  10LE Instituto Nokia de Tecnologia - INdT:
  42  *  A better way to walks through the page table as suggested by Hugh Dickins.
  43  *
  44  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
  45  *  Smaps information related to shared, private, clean and dirty pages.
  46  *
  47  *  Paul Mundt <paul.mundt@nokia.com>:
  48  *  Overall revision about smaps.
  49  */
  50
  51 #include <linux/uaccess.h>
  52
  53 #include <linux/errno.h>
  54 #include <linux/time.h>
  55 #include <linux/proc_fs.h>
  56 #include <linux/stat.h>
  57 #include <linux/task_io_accounting_ops.h>
  58 #include <linux/init.h>
  59 #include <linux/capability.h>
  60 #include <linux/file.h>
  61 #include <linux/fdtable.h>
  62 #include <linux/string.h>
  63 #include <linux/seq_file.h>
  64 #include <linux/namei.h>
  65 #include <linux/mnt_namespace.h>
  66 #include <linux/mm.h>
  67 #include <linux/swap.h>
  68 #include <linux/rcupdate.h>
  69 #include <linux/kallsyms.h>
  70 #include <linux/stacktrace.h>
  71 #include <linux/resource.h>
  72 #include <linux/module.h>
  73 #include <linux/mount.h>
  74 #include <linux/security.h>
  75 #include <linux/ptrace.h>
  76 #include <linux/tracehook.h>
  77 #include <linux/printk.h>
  78 #include <linux/cache.h>
  79 #include <linux/cgroup.h>
  80 #include <linux/cpuset.h>
  81 #include <linux/audit.h>
  82 #include <linux/poll.h>
  83 #include <linux/nsproxy.h>
  84 #include <linux/oom.h>
  85 #include <linux/elf.h>
  86 #include <linux/pid_namespace.h>
  87 #include <linux/user_namespace.h>
  88 #include <linux/fs_struct.h>
  89 #include <linux/slab.h>
  90 #include <linux/sched/autogroup.h>
  91 #include <linux/sched/mm.h>
  92 #include <linux/sched/coredump.h>
  93 #include <linux/sched/debug.h>
  94 #include <linux/sched/stat.h>
  95 #include <linux/flex_array.h>
  96 #include <linux/posix-timers.h>
  97 #include <trace/events/oom.h>
  98 #include "internal.h"
  99 #include "fd.h"
 100
 101 #include "../../lib/kstrtox.h"
 102
 103 /* NOTE:
 104  *      Implementing inode permission operations in /proc is almost
 105  *      certainly an error.  Permission checks need to happen during
 106  *      each system call not at open time.  The reason is that most of
 107  *      what we wish to check for permissions in /proc varies at runtime.
 108  *
 109  *      The classic example of a problem is opening file descriptors
 110  *      in /proc for a task before it execs a suid executable.
 111  */
 112
 113 static u8 nlink_tid __ro_after_init;
 114 static u8 nlink_tgid __ro_after_init;
 115
 116 struct pid_entry {
 117         const char *name;
 118         unsigned int len;
 119         umode_t mode;
 120         const struct inode_operations *iop;
 121         const struct file_operations *fop;
 122         union proc_op op;
 123 };
 124
 125 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 126         .name = (NAME),                                 \
 127         .len  = sizeof(NAME) - 1,                       \
 128         .mode = MODE,                                   \
 129         .iop  = IOP,                                    \
 130         .fop  = FOP,                                    \
 131         .op   = OP,                                     \
 132 }
 133
 134 #define DIR(NAME, MODE, iops, fops)     \
 135         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 136 #define LNK(NAME, get_link)                                     \
 137         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 138                 &proc_pid_link_inode_operations, NULL,          \
 139                 { .proc_get_link = get_link } )
 140 #define REG(NAME, MODE, fops)                           \
 141         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 142 #define ONE(NAME, MODE, show)                           \
 143         NOD(NAME, (S_IFREG|(MODE)),                     \
 144                 NULL, &proc_single_file_operations,     \
 145                 { .proc_show = show } )
 146
 147 /*
 148  * Count the number of hardlinks for the pid_entry table, excluding the .
 149  * and .. links.
 150  */
 151 static unsigned int __init pid_entry_nlink(const struct pid_entry *entries,
 152         unsigned int n)
 153 {
 154         unsigned int i;
 155         unsigned int count;
 156
 157         count = 2;
 158         for (i = 0; i < n; ++i) {
 159                 if (S_ISDIR(entries[i].mode))
 160                         ++count;
 161         }
 162
 163         return count;
 164 }
 165
 166 static int get_task_root(struct task_struct *task, struct path *root)
 167 {
 168         int result = -ENOENT;
 169
 170         task_lock(task);
 171         if (task->fs) {
 172                 get_fs_root(task->fs, root);
 173                 result = 0;
 174         }
 175         task_unlock(task);
 176         return result;
 177 }
 178
 179 static int proc_cwd_link(struct dentry *dentry, struct path *path)
 180 {
 181         struct task_struct *task = get_proc_task(d_inode(dentry));
 182         int result = -ENOENT;
 183
 184         if (task) {
 185                 task_lock(task);
 186                 if (task->fs) {
 187                         get_fs_pwd(task->fs, path);
 188                         result = 0;
 189                 }
 190                 task_unlock(task);
 191                 put_task_struct(task);
 192         }
 193         return result;
 194 }
 195
 196 static int proc_root_link(struct dentry *dentry, struct path *path)
 197 {
 198         struct task_struct *task = get_proc_task(d_inode(dentry));
 199         int result = -ENOENT;
 200
 201         if (task) {
 202                 result = get_task_root(task, path);
 203                 put_task_struct(task);
 204         }
 205         return result;
 206 }
 207
 208 static ssize_t get_mm_cmdline(struct mm_struct *mm, char __user *buf,
 209                               size_t count, loff_t *ppos)
 210 {
 211         unsigned long arg_start, arg_end, env_start, env_end;
 212         unsigned long pos, len;
 213         char *page;
 214
 215         /* Check if process spawned far enough to have cmdline. */
 216         if (!mm->env_end)
 217                 return 0;
 218
 219         spin_lock(&mm->arg_lock);
 220         arg_start = mm->arg_start;
 221         arg_end = mm->arg_end;
 222         env_start = mm->env_start;
 223         env_end = mm->env_end;
 224         spin_unlock(&mm->arg_lock);
 225
 226         if (arg_start >= arg_end)
 227                 return 0;
 228
 229         /*
 230          * We have traditionally allowed the user to re-write
 231          * the argument strings and overflow the end result
 232          * into the environment section. But only do that if
 233          * the environment area is contiguous to the arguments.
 234          */
 235         if (env_start != arg_end || env_start >= env_end)
 236                 env_start = env_end = arg_end;
 237
 238         /* .. and limit it to a maximum of one page of slop */
 239         if (env_end >= arg_end + PAGE_SIZE)
 240                 env_end = arg_end + PAGE_SIZE - 1;
 241
 242         /* We're not going to care if "*ppos" has high bits set */
 243         pos = arg_start + *ppos;
 244
 245         /* .. but we do check the result is in the proper range */
 246         if (pos < arg_start || pos >= env_end)
 247                 return 0;
 248
 249         /* .. and we never go past env_end */
 250         if (env_end - pos < count)
 251                 count = env_end - pos;
 252
 253         page = (char *)__get_free_page(GFP_KERNEL);
 254         if (!page)
 255                 return -ENOMEM;
 256
 257         len = 0;
 258         while (count) {
 259                 int got;
 260                 size_t size = min_t(size_t, PAGE_SIZE, count);
 261                 long offset;
 262
 263                 /*
 264                  * Are we already starting past the official end?
 265                  * We always include the last byte that is *supposed*
 266                  * to be NUL
 267                  */
 268                 offset = (pos >= arg_end) ? pos - arg_end + 1 : 0;
 269
 270                 got = access_remote_vm(mm, pos - offset, page, size + offset, FOLL_ANON);
 271                 if (got <= offset)
 272                         break;
 273                 got -= offset;
 274
 275                 /* Don't walk past a NUL character once you hit arg_end */
 276                 if (pos + got >= arg_end) {
 277                         int n = 0;
 278
 279                         /*
 280                          * If we started before 'arg_end' but ended up
 281                          * at or after it, we start the NUL character
 282                          * check at arg_end-1 (where we expect the normal
 283                          * EOF to be).
 284                          *
 285                          * NOTE! This is smaller than 'got', because
 286                          * pos + got >= arg_end
 287                          */
 288                         if (pos < arg_end)
 289                                 n = arg_end - pos - 1;
 290
 291                         /* Cut off at first NUL after 'n' */
 292                         got = n + strnlen(page+n, offset+got-n);
 293                         if (got < offset)
 294                                 break;
 295                         got -= offset;
 296
 297                         /* Include the NUL if it existed */
 298                         if (got < size)
 299                                 got++;
 300                 }
 301
 302                 got -= copy_to_user(buf, page+offset, got);
 303                 if (unlikely(!got)) {
 304                         if (!len)
 305                                 len = -EFAULT;
 306                         break;
 307                 }
 308                 pos += got;
 309                 buf += got;
 310                 len += got;
 311                 count -= got;
 312         }
 313
 314         free_page((unsigned long)page);
 315         return len;
 316 }
 317
 318 static ssize_t get_task_cmdline(struct task_struct *tsk, char __user *buf,
 319                                 size_t count, loff_t *pos)
 320 {
 321         struct mm_struct *mm;
 322         ssize_t ret;
 323
 324         mm = get_task_mm(tsk);
 325         if (!mm)
 326                 return 0;
 327
 328         ret = get_mm_cmdline(mm, buf, count, pos);
 329         mmput(mm);
 330         return ret;
 331 }
 332
 333 static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
 334                                      size_t count, loff_t *pos)
 335 {
 336         struct task_struct *tsk;
 337         ssize_t ret;
 338
 339         BUG_ON(*pos < 0);
 340
 341         tsk = get_proc_task(file_inode(file));
 342         if (!tsk)
 343                 return -ESRCH;
 344         ret = get_task_cmdline(tsk, buf, count, pos);
 345         put_task_struct(tsk);
 346         if (ret > 0)
 347                 *pos += ret;
 348         return ret;
 349 }
 350
 351 static const struct file_operations proc_pid_cmdline_ops = {
 352         .read   = proc_pid_cmdline_read,
 353         .llseek = generic_file_llseek,
 354 };
 355
 356 #ifdef CONFIG_KALLSYMS
 357 /*
 358  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 359  * Returns the resolved symbol.  If that fails, simply return the address.
 360  */
 361 static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
 362                           struct pid *pid, struct task_struct *task)
 363 {
 364         unsigned long wchan;
 365         char symname[KSYM_NAME_LEN];
 366
 367         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
 368                 goto print0;
 369
 370         wchan = get_wchan(task);
 371         if (wchan && !lookup_symbol_name(wchan, symname)) {
 372                 seq_puts(m, symname);
 373                 return 0;
 374         }
 375
 376 print0:
 377         seq_putc(m, '0');
 378         return 0;
 379 }
 380 #endif /* CONFIG_KALLSYMS */
 381
 382 static int lock_trace(struct task_struct *task)
 383 {
 384         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
 385         if (err)
 386                 return err;
 387         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
 388                 mutex_unlock(&task->signal->cred_guard_mutex);
 389                 return -EPERM;
 390         }
 391         return 0;
 392 }
 393
 394 static void unlock_trace(struct task_struct *task)
 395 {
 396         mutex_unlock(&task->signal->cred_guard_mutex);
 397 }
 398
 399 #ifdef CONFIG_STACKTRACE
 400
 401 #define MAX_STACK_TRACE_DEPTH   64
 402
 403 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 404                           struct pid *pid, struct task_struct *task)
 405 {
 406         struct stack_trace trace;
 407         unsigned long *entries;
 408         int err;
 409
 410         /*
 411          * The ability to racily run the kernel stack unwinder on a running task
 412          * and then observe the unwinder output is scary; while it is useful for
 413          * debugging kernel issues, it can also allow an attacker to leak kernel
 414          * stack contents.
 415          * Doing this in a manner that is at least safe from races would require
 416          * some work to ensure that the remote task can not be scheduled; and
 417          * even then, this would still expose the unwinder as local attack
 418          * surface.
 419          * Therefore, this interface is restricted to root.
 420          */
 421         if (!file_ns_capable(m->file, &init_user_ns, CAP_SYS_ADMIN))
 422                 return -EACCES;
 423
 424         entries = kmalloc_array(MAX_STACK_TRACE_DEPTH, sizeof(*entries),
 425                                 GFP_KERNEL);
 426         if (!entries)
 427                 return -ENOMEM;
 428
 429         trace.nr_entries        = 0;
 430         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 431         trace.entries           = entries;
 432         trace.skip              = 0;
 433
 434         err = lock_trace(task);
 435         if (!err) {
 436                 unsigned int i;
 437
 438                 save_stack_trace_tsk(task, &trace);
 439
 440                 for (i = 0; i < trace.nr_entries; i++) {
 441                         seq_printf(m, "[<0>] %pB\n", (void *)entries[i]);
 442                 }
 443                 unlock_trace(task);
 444         }
 445         kfree(entries);
 446
 447         return err;
 448 }
 449 #endif
 450
 451 #ifdef CONFIG_SCHED_INFO
 452 /*
 453  * Provides /proc/PID/schedstat
 454  */
 455 static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
 456                               struct pid *pid, struct task_struct *task)
 457 {
 458         if (unlikely(!sched_info_on()))
 459                 seq_printf(m, "0 0 0\n");
 460         else
 461                 seq_printf(m, "%llu %llu %lu\n",
 462                    (unsigned long long)task->se.sum_exec_runtime,
 463                    (unsigned long long)task->sched_info.run_delay,
 464                    task->sched_info.pcount);
 465
 466         return 0;
 467 }
 468 #endif
 469
 470 #ifdef CONFIG_LATENCYTOP
 471 static int lstats_show_proc(struct seq_file *m, void *v)
 472 {
 473         int i;
 474         struct inode *inode = m->private;
 475         struct task_struct *task = get_proc_task(inode);
 476
 477         if (!task)
 478                 return -ESRCH;
 479         seq_puts(m, "Latency Top version : v0.1\n");
 480         for (i = 0; i < LT_SAVECOUNT; i++) {
 481                 struct latency_record *lr = &task->latency_record[i];
 482                 if (lr->backtrace[0]) {
 483                         int q;
 484                         seq_printf(m, "%i %li %li",
 485                                    lr->count, lr->time, lr->max);
 486                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 487                                 unsigned long bt = lr->backtrace[q];
 488                                 if (!bt)
 489                                         break;
 490                                 if (bt == ULONG_MAX)
 491                                         break;
 492                                 seq_printf(m, " %ps", (void *)bt);
 493                         }
 494                         seq_putc(m, '\n');
 495                 }
 496
 497         }
 498         put_task_struct(task);
 499         return 0;
 500 }
 501
 502 static int lstats_open(struct inode *inode, struct file *file)
 503 {
 504         return single_open(file, lstats_show_proc, inode);
 505 }
 506
 507 static ssize_t lstats_write(struct file *file, const char __user *buf,
 508                             size_t count, loff_t *offs)
 509 {
 510         struct task_struct *task = get_proc_task(file_inode(file));
 511
 512         if (!task)
 513                 return -ESRCH;
 514         clear_all_latency_tracing(task);
 515         put_task_struct(task);
 516
 517         return count;
 518 }
 519
 520 static const struct file_operations proc_lstats_operations = {
 521         .open           = lstats_open,
 522         .read           = seq_read,
 523         .write          = lstats_write,
 524         .llseek         = seq_lseek,
 525         .release        = single_release,
 526 };
 527
 528 #endif
 529
 530 static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
 531                           struct pid *pid, struct task_struct *task)
 532 {
 533         unsigned long totalpages = totalram_pages() + total_swap_pages;
 534         unsigned long points = 0;
 535
 536         points = oom_badness(task, NULL, NULL, totalpages) *
 537                                         1000 / totalpages;
 538         seq_printf(m, "%lu\n", points);
 539
 540         return 0;
 541 }
 542
 543 struct limit_names {
 544         const char *name;
 545         const char *unit;
 546 };
 547
 548 static const struct limit_names lnames[RLIM_NLIMITS] = {
 549         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 550         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 551         [RLIMIT_DATA] = {"Max data size", "bytes"},
 552         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 553         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 554         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 555         [RLIMIT_NPROC] = {"Max processes", "processes"},
 556         [RLIMIT_NOFILE] = {"Max open files", "files"},
 557         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 558         [RLIMIT_AS] = {"Max address space", "bytes"},
 559         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 560         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 561         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 562         [RLIMIT_NICE] = {"Max nice priority", NULL},
 563         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 564         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 565 };
 566
 567 /* Display limits for a process */
 568 static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
 569                            struct pid *pid, struct task_struct *task)
 570 {
 571         unsigned int i;
 572         unsigned long flags;
 573
 574         struct rlimit rlim[RLIM_NLIMITS];
 575
 576         if (!lock_task_sighand(task, &flags))
 577                 return 0;
 578         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 579         unlock_task_sighand(task, &flags);
 580
 581         /*
 582          * print the file header
 583          */
 584         seq_puts(m, "Limit                     "
 585                 "Soft Limit           "
 586                 "Hard Limit           "
 587                 "Units     \n");
 588
 589         for (i = 0; i < RLIM_NLIMITS; i++) {
 590                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 591                         seq_printf(m, "%-25s %-20s ",
 592                                    lnames[i].name, "unlimited");
 593                 else
 594                         seq_printf(m, "%-25s %-20lu ",
 595                                    lnames[i].name, rlim[i].rlim_cur);
 596
 597                 if (rlim[i].rlim_max == RLIM_INFINITY)
 598                         seq_printf(m, "%-20s ", "unlimited");
 599                 else
 600                         seq_printf(m, "%-20lu ", rlim[i].rlim_max);
 601
 602                 if (lnames[i].unit)
 603                         seq_printf(m, "%-10s\n", lnames[i].unit);
 604                 else
 605                         seq_putc(m, '\n');
 606         }
 607
 608         return 0;
 609 }
 610
 611 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 612 static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
 613                             struct pid *pid, struct task_struct *task)
 614 {
 615         long nr;
 616         unsigned long args[6], sp, pc;
 617         int res;
 618
 619         res = lock_trace(task);
 620         if (res)
 621                 return res;
 622
 623         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 624                 seq_puts(m, "running\n");
 625         else if (nr < 0)
 626                 seq_printf(m, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 627         else
 628                 seq_printf(m,
 629                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 630                        nr,
 631                        args[0], args[1], args[2], args[3], args[4], args[5],
 632                        sp, pc);
 633         unlock_trace(task);
 634
 635         return 0;
 636 }
 637 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 638
 639 /************************************************************************/
 640 /*                       Here the fs part begins                        */
 641 /************************************************************************/
 642
 643 /* permission checks */
 644 static int proc_fd_access_allowed(struct inode *inode)
 645 {
 646         struct task_struct *task;
 647         int allowed = 0;
 648         /* Allow access to a task's file descriptors if it is us or we
 649          * may use ptrace attach to the process and find out that
 650          * information.
 651          */
 652         task = get_proc_task(inode);
 653         if (task) {
 654                 allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 655                 put_task_struct(task);
 656         }
 657         return allowed;
 658 }
 659
 660 int proc_setattr(struct dentry *dentry, struct iattr *attr)
 661 {
 662         int error;
 663         struct inode *inode = d_inode(dentry);
 664
 665         if (attr->ia_valid & ATTR_MODE)
 666                 return -EPERM;
 667
 668         error = setattr_prepare(dentry, attr);
 669         if (error)
 670                 return error;
 671
 672         setattr_copy(inode, attr);
 673         mark_inode_dirty(inode);
 674         return 0;
 675 }
 676
 677 /*
 678  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
 679  * or euid/egid (for hide_pid_min=2)?
 680  */
 681 static bool has_pid_permissions(struct pid_namespace *pid,
 682                                  struct task_struct *task,
 683                                  int hide_pid_min)
 684 {
 685         if (pid->hide_pid < hide_pid_min)
 686                 return true;
 687         if (in_group_p(pid->pid_gid))
 688                 return true;
 689         return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 690 }
 691
 692
 693 static int proc_pid_permission(struct inode *inode, int mask)
 694 {
 695         struct pid_namespace *pid = proc_pid_ns(inode);
 696         struct task_struct *task;
 697         bool has_perms;
 698
 699         task = get_proc_task(inode);
 700         if (!task)
 701                 return -ESRCH;
 702         has_perms = has_pid_permissions(pid, task, HIDEPID_NO_ACCESS);
 703         put_task_struct(task);
 704
 705         if (!has_perms) {
 706                 if (pid->hide_pid == HIDEPID_INVISIBLE) {
 707                         /*
 708                          * Let's make getdents(), stat(), and open()
 709                          * consistent with each other.  If a process
 710                          * may not stat() a file, it shouldn't be seen
 711                          * in procfs at all.
 712                          */
 713                         return -ENOENT;
 714                 }
 715
 716                 return -EPERM;
 717         }
 718         return generic_permission(inode, mask);
 719 }
 720
 721
 722
 723 static const struct inode_operations proc_def_inode_operations = {
 724         .setattr        = proc_setattr,
 725 };
 726
 727 static int proc_single_show(struct seq_file *m, void *v)
 728 {
 729         struct inode *inode = m->private;
 730         struct pid_namespace *ns = proc_pid_ns(inode);
 731         struct pid *pid = proc_pid(inode);
 732         struct task_struct *task;
 733         int ret;
 734
 735         task = get_pid_task(pid, PIDTYPE_PID);
 736         if (!task)
 737                 return -ESRCH;
 738
 739         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 740
 741         put_task_struct(task);
 742         return ret;
 743 }
 744
 745 static int proc_single_open(struct inode *inode, struct file *filp)
 746 {
 747         return single_open(filp, proc_single_show, inode);
 748 }
 749
 750 static const struct file_operations proc_single_file_operations = {
 751         .open           = proc_single_open,
 752         .read           = seq_read,
 753         .llseek         = seq_lseek,
 754         .release        = single_release,
 755 };
 756
 757
 758 struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
 759 {
 760         struct task_struct *task = get_proc_task(inode);
 761         struct mm_struct *mm = ERR_PTR(-ESRCH);
 762
 763         if (task) {
 764                 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
 765                 put_task_struct(task);
 766
 767                 if (!IS_ERR_OR_NULL(mm)) {
 768                         /* ensure this mm_struct can't be freed */
 769                         mmgrab(mm);
 770                         /* but do not pin its memory */
 771                         mmput(mm);
 772                 }
 773         }
 774
 775         return mm;
 776 }
 777
 778 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
 779 {
 780         struct mm_struct *mm = proc_mem_open(inode, mode);
 781
 782         if (IS_ERR(mm))
 783                 return PTR_ERR(mm);
 784
 785         file->private_data = mm;
 786         return 0;
 787 }
 788
 789 static int mem_open(struct inode *inode, struct file *file)
 790 {
 791         int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
 792
 793         /* OK to pass negative loff_t, we can catch out-of-range */
 794         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 795
 796         return ret;
 797 }
 798
 799 static ssize_t mem_rw(struct file *file, char __user *buf,
 800                         size_t count, loff_t *ppos, int write)
 801 {
 802         struct mm_struct *mm = file->private_data;
 803         unsigned long addr = *ppos;
 804         ssize_t copied;
 805         char *page;
 806         unsigned int flags;
 807
 808         if (!mm)
 809                 return 0;
 810
 811         page = (char *)__get_free_page(GFP_KERNEL);
 812         if (!page)
 813                 return -ENOMEM;
 814
 815         copied = 0;
 816         if (!mmget_not_zero(mm))
 817                 goto free;
 818
 819         flags = FOLL_FORCE | (write ? FOLL_WRITE : 0);
 820
 821         while (count > 0) {
 822                 int this_len = min_t(int, count, PAGE_SIZE);
 823
 824                 if (write && copy_from_user(page, buf, this_len)) {
 825                         copied = -EFAULT;
 826                         break;
 827                 }
 828
 829                 this_len = access_remote_vm(mm, addr, page, this_len, flags);
 830                 if (!this_len) {
 831                         if (!copied)
 832                                 copied = -EIO;
 833                         break;
 834                 }
 835
 836                 if (!write && copy_to_user(buf, page, this_len)) {
 837                         copied = -EFAULT;
 838                         break;
 839                 }
 840
 841                 buf += this_len;
 842                 addr += this_len;
 843                 copied += this_len;
 844                 count -= this_len;
 845         }
 846         *ppos = addr;
 847
 848         mmput(mm);
 849 free:
 850         free_page((unsigned long) page);
 851         return copied;
 852 }
 853
 854 static ssize_t mem_read(struct file *file, char __user *buf,
 855                         size_t count, loff_t *ppos)
 856 {
 857         return mem_rw(file, buf, count, ppos, 0);
 858 }
 859
 860 static ssize_t mem_write(struct file *file, const char __user *buf,
 861                          size_t count, loff_t *ppos)
 862 {
 863         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 864 }
 865
 866 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 867 {
 868         switch (orig) {
 869         case 0:
 870                 file->f_pos = offset;
 871                 break;
 872         case 1:
 873                 file->f_pos += offset;
 874                 break;
 875         default:
 876                 return -EINVAL;
 877         }
 878         force_successful_syscall_return();
 879         return file->f_pos;
 880 }
 881
 882 static int mem_release(struct inode *inode, struct file *file)
 883 {
 884         struct mm_struct *mm = file->private_data;
 885         if (mm)
 886                 mmdrop(mm);
 887         return 0;
 888 }
 889
 890 static const struct file_operations proc_mem_operations = {
 891         .llseek         = mem_lseek,
 892         .read           = mem_read,
 893         .write          = mem_write,
 894         .open           = mem_open,
 895         .release        = mem_release,
 896 };
 897
 898 static int environ_open(struct inode *inode, struct file *file)
 899 {
 900         return __mem_open(inode, file, PTRACE_MODE_READ);
 901 }
 902
 903 static ssize_t environ_read(struct file *file, char __user *buf,
 904                         size_t count, loff_t *ppos)
 905 {
 906         char *page;
 907         unsigned long src = *ppos;
 908         int ret = 0;
 909         struct mm_struct *mm = file->private_data;
 910         unsigned long env_start, env_end;
 911
 912         /* Ensure the process spawned far enough to have an environment. */
 913         if (!mm || !mm->env_end)
 914                 return 0;
 915
 916         page = (char *)__get_free_page(GFP_KERNEL);
 917         if (!page)
 918                 return -ENOMEM;
 919
 920         ret = 0;
 921         if (!mmget_not_zero(mm))
 922                 goto free;
 923
 924         spin_lock(&mm->arg_lock);
 925         env_start = mm->env_start;
 926         env_end = mm->env_end;
 927         spin_unlock(&mm->arg_lock);
 928
 929         while (count > 0) {
 930                 size_t this_len, max_len;
 931                 int retval;
 932
 933                 if (src >= (env_end - env_start))
 934                         break;
 935
 936                 this_len = env_end - (env_start + src);
 937
 938                 max_len = min_t(size_t, PAGE_SIZE, count);
 939                 this_len = min(max_len, this_len);
 940
 941                 retval = access_remote_vm(mm, (env_start + src), page, this_len, FOLL_ANON);
 942
 943                 if (retval <= 0) {
 944                         ret = retval;
 945                         break;
 946                 }
 947
 948                 if (copy_to_user(buf, page, retval)) {
 949                         ret = -EFAULT;
 950                         break;
 951                 }
 952
 953                 ret += retval;
 954                 src += retval;
 955                 buf += retval;
 956                 count -= retval;
 957         }
 958         *ppos = src;
 959         mmput(mm);
 960
 961 free:
 962         free_page((unsigned long) page);
 963         return ret;
 964 }
 965
 966 static const struct file_operations proc_environ_operations = {
 967         .open           = environ_open,
 968         .read           = environ_read,
 969         .llseek         = generic_file_llseek,
 970         .release        = mem_release,
 971 };
 972
 973 static int auxv_open(struct inode *inode, struct file *file)
 974 {
 975         return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
 976 }
 977
 978 static ssize_t auxv_read(struct file *file, char __user *buf,
 979                         size_t count, loff_t *ppos)
 980 {
 981         struct mm_struct *mm = file->private_data;
 982         unsigned int nwords = 0;
 983
 984         if (!mm)
 985                 return 0;
 986         do {
 987                 nwords += 2;
 988         } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 989         return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
 990                                        nwords * sizeof(mm->saved_auxv[0]));
 991 }
 992
 993 static const struct file_operations proc_auxv_operations = {
 994         .open           = auxv_open,
 995         .read           = auxv_read,
 996         .llseek         = generic_file_llseek,
 997         .release        = mem_release,
 998 };
 999
1000 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1001                             loff_t *ppos)
1002 {
1003         struct task_struct *task = get_proc_task(file_inode(file));
1004         char buffer[PROC_NUMBUF];
1005         int oom_adj = OOM_ADJUST_MIN;
1006         size_t len;
1007
1008         if (!task)
1009                 return -ESRCH;
1010         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1011                 oom_adj = OOM_ADJUST_MAX;
1012         else
1013                 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1014                           OOM_SCORE_ADJ_MAX;
1015         put_task_struct(task);
1016         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1017         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1018 }
1019
1020 static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1021 {
1022         static DEFINE_MUTEX(oom_adj_mutex);
1023         struct mm_struct *mm = NULL;
1024         struct task_struct *task;
1025         int err = 0;
1026
1027         task = get_proc_task(file_inode(file));
1028         if (!task)
1029                 return -ESRCH;
1030
1031         mutex_lock(&oom_adj_mutex);
1032         if (legacy) {
1033                 if (oom_adj < task->signal->oom_score_adj &&
1034                                 !capable(CAP_SYS_RESOURCE)) {
1035                         err = -EACCES;
1036                         goto err_unlock;
1037                 }
1038                 /*
1039                  * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1040                  * /proc/pid/oom_score_adj instead.
1041                  */
1042                 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1043                           current->comm, task_pid_nr(current), task_pid_nr(task),
1044                           task_pid_nr(task));
1045         } else {
1046                 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1047                                 !capable(CAP_SYS_RESOURCE)) {
1048                         err = -EACCES;
1049                         goto err_unlock;
1050                 }
1051         }
1052
1053         /*
1054          * Make sure we will check other processes sharing the mm if this is
1055          * not vfrok which wants its own oom_score_adj.
1056          * pin the mm so it doesn't go away and get reused after task_unlock
1057          */
1058         if (!task->vfork_done) {
1059                 struct task_struct *p = find_lock_task_mm(task);
1060
1061                 if (p) {
1062                         if (atomic_read(&p->mm->mm_users) > 1) {
1063                                 mm = p->mm;
1064                                 mmgrab(mm);
1065                         }
1066                         task_unlock(p);
1067                 }
1068         }
1069
1070         task->signal->oom_score_adj = oom_adj;
1071         if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1072                 task->signal->oom_score_adj_min = (short)oom_adj;
1073         trace_oom_score_adj_update(task);
1074
1075         if (mm) {
1076                 struct task_struct *p;
1077
1078                 rcu_read_lock();
1079                 for_each_process(p) {
1080                         if (same_thread_group(task, p))
1081                                 continue;
1082
1083                         /* do not touch kernel threads or the global init */
1084                         if (p->flags & PF_KTHREAD || is_global_init(p))
1085                                 continue;
1086
1087                         task_lock(p);
1088                         if (!p->vfork_done && process_shares_mm(p, mm)) {
1089                                 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",
1090                                                 task_pid_nr(p), p->comm,
1091                                                 p->signal->oom_score_adj, oom_adj,
1092                                                 task_pid_nr(task), task->comm);
1093                                 p->signal->oom_score_adj = oom_adj;
1094                                 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1095                                         p->signal->oom_score_adj_min = (short)oom_adj;
1096                         }
1097                         task_unlock(p);
1098                 }
1099                 rcu_read_unlock();
1100                 mmdrop(mm);
1101         }
1102 err_unlock:
1103         mutex_unlock(&oom_adj_mutex);
1104         put_task_struct(task);
1105         return err;
1106 }
1107
1108 /*
1109  * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1110  * kernels.  The effective policy is defined by oom_score_adj, which has a
1111  * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1112  * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1113  * Processes that become oom disabled via oom_adj will still be oom disabled
1114  * with this implementation.
1115  *
1116  * oom_adj cannot be removed since existing userspace binaries use it.
1117  */
1118 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1119                              size_t count, loff_t *ppos)
1120 {
1121         char buffer[PROC_NUMBUF];
1122         int oom_adj;
1123         int err;
1124
1125         memset(buffer, 0, sizeof(buffer));
1126         if (count > sizeof(buffer) - 1)
1127                 count = sizeof(buffer) - 1;
1128         if (copy_from_user(buffer, buf, count)) {
1129                 err = -EFAULT;
1130                 goto out;
1131         }
1132
1133         err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1134         if (err)
1135                 goto out;
1136         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1137              oom_adj != OOM_DISABLE) {
1138                 err = -EINVAL;
1139                 goto out;
1140         }
1141
1142         /*
1143          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1144          * value is always attainable.
1145          */
1146         if (oom_adj == OOM_ADJUST_MAX)
1147                 oom_adj = OOM_SCORE_ADJ_MAX;
1148         else
1149                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1150
1151         err = __set_oom_adj(file, oom_adj, true);
1152 out:
1153         return err < 0 ? err : count;
1154 }
1155
1156 static const struct file_operations proc_oom_adj_operations = {
1157         .read           = oom_adj_read,
1158         .write          = oom_adj_write,
1159         .llseek         = generic_file_llseek,
1160 };
1161
1162 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1163                                         size_t count, loff_t *ppos)
1164 {
1165         struct task_struct *task = get_proc_task(file_inode(file));
1166         char buffer[PROC_NUMBUF];
1167         short oom_score_adj = OOM_SCORE_ADJ_MIN;
1168         size_t len;
1169
1170         if (!task)
1171                 return -ESRCH;
1172         oom_score_adj = task->signal->oom_score_adj;
1173         put_task_struct(task);
1174         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1175         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1176 }
1177
1178 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1179                                         size_t count, loff_t *ppos)
1180 {
1181         char buffer[PROC_NUMBUF];
1182         int oom_score_adj;
1183         int err;
1184
1185         memset(buffer, 0, sizeof(buffer));
1186         if (count > sizeof(buffer) - 1)
1187                 count = sizeof(buffer) - 1;
1188         if (copy_from_user(buffer, buf, count)) {
1189                 err = -EFAULT;
1190                 goto out;
1191         }
1192
1193         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1194         if (err)
1195                 goto out;
1196         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1197                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1198                 err = -EINVAL;
1199                 goto out;
1200         }
1201
1202         err = __set_oom_adj(file, oom_score_adj, false);
1203 out:
1204         return err < 0 ? err : count;
1205 }
1206
1207 static const struct file_operations proc_oom_score_adj_operations = {
1208         .read           = oom_score_adj_read,
1209         .write          = oom_score_adj_write,
1210         .llseek         = default_llseek,
1211 };
1212
1213 #ifdef CONFIG_AUDITSYSCALL
1214 #define TMPBUFLEN 11
1215 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1216                                   size_t count, loff_t *ppos)
1217 {
1218         struct inode * inode = file_inode(file);
1219         struct task_struct *task = get_proc_task(inode);
1220         ssize_t length;
1221         char tmpbuf[TMPBUFLEN];
1222
1223         if (!task)
1224                 return -ESRCH;
1225         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1226                            from_kuid(file->f_cred->user_ns,
1227                                      audit_get_loginuid(task)));
1228         put_task_struct(task);
1229         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1230 }
1231
1232 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1233                                    size_t count, loff_t *ppos)
1234 {
1235         struct inode * inode = file_inode(file);
1236         uid_t loginuid;
1237         kuid_t kloginuid;
1238         int rv;
1239
1240         rcu_read_lock();
1241         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1242                 rcu_read_unlock();
1243                 return -EPERM;
1244         }
1245         rcu_read_unlock();
1246
1247         if (*ppos != 0) {
1248                 /* No partial writes. */
1249                 return -EINVAL;
1250         }
1251
1252         rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1253         if (rv < 0)
1254                 return rv;
1255
1256         /* is userspace tring to explicitly UNSET the loginuid? */
1257         if (loginuid == AUDIT_UID_UNSET) {
1258                 kloginuid = INVALID_UID;
1259         } else {
1260                 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1261                 if (!uid_valid(kloginuid))
1262                         return -EINVAL;
1263         }
1264
1265         rv = audit_set_loginuid(kloginuid);
1266         if (rv < 0)
1267                 return rv;
1268         return count;
1269 }
1270
1271 static const struct file_operations proc_loginuid_operations = {
1272         .read           = proc_loginuid_read,
1273         .write          = proc_loginuid_write,
1274         .llseek         = generic_file_llseek,
1275 };
1276
1277 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1278                                   size_t count, loff_t *ppos)
1279 {
1280         struct inode * inode = file_inode(file);
1281         struct task_struct *task = get_proc_task(inode);
1282         ssize_t length;
1283         char tmpbuf[TMPBUFLEN];
1284
1285         if (!task)
1286                 return -ESRCH;
1287         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1288                                 audit_get_sessionid(task));
1289         put_task_struct(task);
1290         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1291 }
1292
1293 static const struct file_operations proc_sessionid_operations = {
1294         .read           = proc_sessionid_read,
1295         .llseek         = generic_file_llseek,
1296 };
1297 #endif
1298
1299 #ifdef CONFIG_FAULT_INJECTION
1300 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1301                                       size_t count, loff_t *ppos)
1302 {
1303         struct task_struct *task = get_proc_task(file_inode(file));
1304         char buffer[PROC_NUMBUF];
1305         size_t len;
1306         int make_it_fail;
1307
1308         if (!task)
1309                 return -ESRCH;
1310         make_it_fail = task->make_it_fail;
1311         put_task_struct(task);
1312
1313         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1314
1315         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1316 }
1317
1318 static ssize_t proc_fault_inject_write(struct file * file,
1319                         const char __user * buf, size_t count, loff_t *ppos)
1320 {
1321         struct task_struct *task;
1322         char buffer[PROC_NUMBUF];
1323         int make_it_fail;
1324         int rv;
1325
1326         if (!capable(CAP_SYS_RESOURCE))
1327                 return -EPERM;
1328         memset(buffer, 0, sizeof(buffer));
1329         if (count > sizeof(buffer) - 1)
1330                 count = sizeof(buffer) - 1;
1331         if (copy_from_user(buffer, buf, count))
1332                 return -EFAULT;
1333         rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1334         if (rv < 0)
1335                 return rv;
1336         if (make_it_fail < 0 || make_it_fail > 1)
1337                 return -EINVAL;
1338
1339         task = get_proc_task(file_inode(file));
1340         if (!task)
1341                 return -ESRCH;
1342         task->make_it_fail = make_it_fail;
1343         put_task_struct(task);
1344
1345         return count;
1346 }
1347
1348 static const struct file_operations proc_fault_inject_operations = {
1349         .read           = proc_fault_inject_read,
1350         .write          = proc_fault_inject_write,
1351         .llseek         = generic_file_llseek,
1352 };
1353
1354 static ssize_t proc_fail_nth_write(struct file *file, const char __user *buf,
1355                                    size_t count, loff_t *ppos)
1356 {
1357         struct task_struct *task;
1358         int err;
1359         unsigned int n;
1360
1361         err = kstrtouint_from_user(buf, count, 0, &n);
1362         if (err)
1363                 return err;
1364
1365         task = get_proc_task(file_inode(file));
1366         if (!task)
1367                 return -ESRCH;
1368         task->fail_nth = n;
1369         put_task_struct(task);
1370
1371         return count;
1372 }
1373
1374 static ssize_t proc_fail_nth_read(struct file *file, char __user *buf,
1375                                   size_t count, loff_t *ppos)
1376 {
1377         struct task_struct *task;
1378         char numbuf[PROC_NUMBUF];
1379         ssize_t len;
1380
1381         task = get_proc_task(file_inode(file));
1382         if (!task)
1383                 return -ESRCH;
1384         len = snprintf(numbuf, sizeof(numbuf), "%u\n", task->fail_nth);
1385         put_task_struct(task);
1386         return simple_read_from_buffer(buf, count, ppos, numbuf, len);
1387 }
1388
1389 static const struct file_operations proc_fail_nth_operations = {
1390         .read           = proc_fail_nth_read,
1391         .write          = proc_fail_nth_write,
1392 };
1393 #endif
1394
1395
1396 #ifdef CONFIG_SCHED_DEBUG
1397 /*
1398  * Print out various scheduling related per-task fields:
1399  */
1400 static int sched_show(struct seq_file *m, void *v)
1401 {
1402         struct inode *inode = m->private;
1403         struct pid_namespace *ns = proc_pid_ns(inode);
1404         struct task_struct *p;
1405
1406         p = get_proc_task(inode);
1407         if (!p)
1408                 return -ESRCH;
1409         proc_sched_show_task(p, ns, m);
1410
1411         put_task_struct(p);
1412
1413         return 0;
1414 }
1415
1416 static ssize_t
1417 sched_write(struct file *file, const char __user *buf,
1418             size_t count, loff_t *offset)
1419 {
1420         struct inode *inode = file_inode(file);
1421         struct task_struct *p;
1422
1423         p = get_proc_task(inode);
1424         if (!p)
1425                 return -ESRCH;
1426         proc_sched_set_task(p);
1427
1428         put_task_struct(p);
1429
1430         return count;
1431 }
1432
1433 static int sched_open(struct inode *inode, struct file *filp)
1434 {
1435         return single_open(filp, sched_show, inode);
1436 }
1437
1438 static const struct file_operations proc_pid_sched_operations = {
1439         .open           = sched_open,
1440         .read           = seq_read,
1441         .write          = sched_write,
1442         .llseek         = seq_lseek,
1443         .release        = single_release,
1444 };
1445
1446 #endif
1447
1448 #ifdef CONFIG_SCHED_AUTOGROUP
1449 /*
1450  * Print out autogroup related information:
1451  */
1452 static int sched_autogroup_show(struct seq_file *m, void *v)
1453 {
1454         struct inode *inode = m->private;
1455         struct task_struct *p;
1456
1457         p = get_proc_task(inode);
1458         if (!p)
1459                 return -ESRCH;
1460         proc_sched_autogroup_show_task(p, m);
1461
1462         put_task_struct(p);
1463
1464         return 0;
1465 }
1466
1467 static ssize_t
1468 sched_autogroup_write(struct file *file, const char __user *buf,
1469             size_t count, loff_t *offset)
1470 {
1471         struct inode *inode = file_inode(file);
1472         struct task_struct *p;
1473         char buffer[PROC_NUMBUF];
1474         int nice;
1475         int err;
1476
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))
1481                 return -EFAULT;
1482
1483         err = kstrtoint(strstrip(buffer), 0, &nice);
1484         if (err < 0)
1485                 return err;
1486
1487         p = get_proc_task(inode);
1488         if (!p)
1489                 return -ESRCH;
1490
1491         err = proc_sched_autogroup_set_nice(p, nice);
1492         if (err)
1493                 count = err;
1494
1495         put_task_struct(p);
1496
1497         return count;
1498 }
1499
1500 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1501 {
1502         int ret;
1503
1504         ret = single_open(filp, sched_autogroup_show, NULL);
1505         if (!ret) {
1506                 struct seq_file *m = filp->private_data;
1507
1508                 m->private = inode;
1509         }
1510         return ret;
1511 }
1512
1513 static const struct file_operations proc_pid_sched_autogroup_operations = {
1514         .open           = sched_autogroup_open,
1515         .read           = seq_read,
1516         .write          = sched_autogroup_write,
1517         .llseek         = seq_lseek,
1518         .release        = single_release,
1519 };
1520
1521 #endif /* CONFIG_SCHED_AUTOGROUP */
1522
1523 static ssize_t comm_write(struct file *file, const char __user *buf,
1524                                 size_t count, loff_t *offset)
1525 {
1526         struct inode *inode = file_inode(file);
1527         struct task_struct *p;
1528         char buffer[TASK_COMM_LEN];
1529         const size_t maxlen = sizeof(buffer) - 1;
1530
1531         memset(buffer, 0, sizeof(buffer));
1532         if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1533                 return -EFAULT;
1534
1535         p = get_proc_task(inode);
1536         if (!p)
1537                 return -ESRCH;
1538
1539         if (same_thread_group(current, p))
1540                 set_task_comm(p, buffer);
1541         else
1542                 count = -EINVAL;
1543
1544         put_task_struct(p);
1545
1546         return count;
1547 }
1548
1549 static int comm_show(struct seq_file *m, void *v)
1550 {
1551         struct inode *inode = m->private;
1552         struct task_struct *p;
1553
1554         p = get_proc_task(inode);
1555         if (!p)
1556                 return -ESRCH;
1557
1558         proc_task_name(m, p, false);
1559         seq_putc(m, '\n');
1560
1561         put_task_struct(p);
1562
1563         return 0;
1564 }
1565
1566 static int comm_open(struct inode *inode, struct file *filp)
1567 {
1568         return single_open(filp, comm_show, inode);
1569 }
1570
1571 static const struct file_operations proc_pid_set_comm_operations = {
1572         .open           = comm_open,
1573         .read           = seq_read,
1574         .write          = comm_write,
1575         .llseek         = seq_lseek,
1576         .release        = single_release,
1577 };
1578
1579 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1580 {
1581         struct task_struct *task;
1582         struct file *exe_file;
1583
1584         task = get_proc_task(d_inode(dentry));
1585         if (!task)
1586                 return -ENOENT;
1587         exe_file = get_task_exe_file(task);
1588         put_task_struct(task);
1589         if (exe_file) {
1590                 *exe_path = exe_file->f_path;
1591                 path_get(&exe_file->f_path);
1592                 fput(exe_file);
1593                 return 0;
1594         } else
1595                 return -ENOENT;
1596 }
1597
1598 static const char *proc_pid_get_link(struct dentry *dentry,
1599                                      struct inode *inode,
1600                                      struct delayed_call *done)
1601 {
1602         struct path path;
1603         int error = -EACCES;
1604
1605         if (!dentry)
1606                 return ERR_PTR(-ECHILD);
1607
1608         /* Are we allowed to snoop on the tasks file descriptors? */
1609         if (!proc_fd_access_allowed(inode))
1610                 goto out;
1611
1612         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1613         if (error)
1614                 goto out;
1615
1616         nd_jump_link(&path);
1617         return NULL;
1618 out:
1619         return ERR_PTR(error);
1620 }
1621
1622 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1623 {
1624         char *tmp = (char *)__get_free_page(GFP_KERNEL);
1625         char *pathname;
1626         int len;
1627
1628         if (!tmp)
1629                 return -ENOMEM;
1630
1631         pathname = d_path(path, tmp, PAGE_SIZE);
1632         len = PTR_ERR(pathname);
1633         if (IS_ERR(pathname))
1634                 goto out;
1635         len = tmp + PAGE_SIZE - 1 - pathname;
1636
1637         if (len > buflen)
1638                 len = buflen;
1639         if (copy_to_user(buffer, pathname, len))
1640                 len = -EFAULT;
1641  out:
1642         free_page((unsigned long)tmp);
1643         return len;
1644 }
1645
1646 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1647 {
1648         int error = -EACCES;
1649         struct inode *inode = d_inode(dentry);
1650         struct path path;
1651
1652         /* Are we allowed to snoop on the tasks file descriptors? */
1653         if (!proc_fd_access_allowed(inode))
1654                 goto out;
1655
1656         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1657         if (error)
1658                 goto out;
1659
1660         error = do_proc_readlink(&path, buffer, buflen);
1661         path_put(&path);
1662 out:
1663         return error;
1664 }
1665
1666 const struct inode_operations proc_pid_link_inode_operations = {
1667         .readlink       = proc_pid_readlink,
1668         .get_link       = proc_pid_get_link,
1669         .setattr        = proc_setattr,
1670 };
1671
1672
1673 /* building an inode */
1674
1675 void task_dump_owner(struct task_struct *task, umode_t mode,
1676                      kuid_t *ruid, kgid_t *rgid)
1677 {
1678         /* Depending on the state of dumpable compute who should own a
1679          * proc file for a task.
1680          */
1681         const struct cred *cred;
1682         kuid_t uid;
1683         kgid_t gid;
1684
1685         if (unlikely(task->flags & PF_KTHREAD)) {
1686                 *ruid = GLOBAL_ROOT_UID;
1687                 *rgid = GLOBAL_ROOT_GID;
1688                 return;
1689         }
1690
1691         /* Default to the tasks effective ownership */
1692         rcu_read_lock();
1693         cred = __task_cred(task);
1694         uid = cred->euid;
1695         gid = cred->egid;
1696         rcu_read_unlock();
1697
1698         /*
1699          * Before the /proc/pid/status file was created the only way to read
1700          * the effective uid of a /process was to stat /proc/pid.  Reading
1701          * /proc/pid/status is slow enough that procps and other packages
1702          * kept stating /proc/pid.  To keep the rules in /proc simple I have
1703          * made this apply to all per process world readable and executable
1704          * directories.
1705          */
1706         if (mode != (S_IFDIR|S_IRUGO|S_IXUGO)) {
1707                 struct mm_struct *mm;
1708                 task_lock(task);
1709                 mm = task->mm;
1710                 /* Make non-dumpable tasks owned by some root */
1711                 if (mm) {
1712                         if (get_dumpable(mm) != SUID_DUMP_USER) {
1713                                 struct user_namespace *user_ns = mm->user_ns;
1714
1715                                 uid = make_kuid(user_ns, 0);
1716                                 if (!uid_valid(uid))
1717                                         uid = GLOBAL_ROOT_UID;
1718
1719                                 gid = make_kgid(user_ns, 0);
1720                                 if (!gid_valid(gid))
1721                                         gid = GLOBAL_ROOT_GID;
1722                         }
1723                 } else {
1724                         uid = GLOBAL_ROOT_UID;
1725                         gid = GLOBAL_ROOT_GID;
1726                 }
1727                 task_unlock(task);
1728         }
1729         *ruid = uid;
1730         *rgid = gid;
1731 }
1732
1733 struct inode *proc_pid_make_inode(struct super_block * sb,
1734                                   struct task_struct *task, umode_t mode)
1735 {
1736         struct inode * inode;
1737         struct proc_inode *ei;
1738
1739         /* We need a new inode */
1740
1741         inode = new_inode(sb);
1742         if (!inode)
1743                 goto out;
1744
1745         /* Common stuff */
1746         ei = PROC_I(inode);
1747         inode->i_mode = mode;
1748         inode->i_ino = get_next_ino();
1749         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1750         inode->i_op = &proc_def_inode_operations;
1751
1752         /*
1753          * grab the reference to task.
1754          */
1755         ei->pid = get_task_pid(task, PIDTYPE_PID);
1756         if (!ei->pid)
1757                 goto out_unlock;
1758
1759         task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1760         security_task_to_inode(task, inode);
1761
1762 out:
1763         return inode;
1764
1765 out_unlock:
1766         iput(inode);
1767         return NULL;
1768 }
1769
1770 int pid_getattr(const struct path *path, struct kstat *stat,
1771                 u32 request_mask, unsigned int query_flags)
1772 {
1773         struct inode *inode = d_inode(path->dentry);
1774         struct pid_namespace *pid = proc_pid_ns(inode);
1775         struct task_struct *task;
1776
1777         generic_fillattr(inode, stat);
1778
1779         stat->uid = GLOBAL_ROOT_UID;
1780         stat->gid = GLOBAL_ROOT_GID;
1781         rcu_read_lock();
1782         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1783         if (task) {
1784                 if (!has_pid_permissions(pid, task, HIDEPID_INVISIBLE)) {
1785                         rcu_read_unlock();
1786                         /*
1787                          * This doesn't prevent learning whether PID exists,
1788                          * it only makes getattr() consistent with readdir().
1789                          */
1790                         return -ENOENT;
1791                 }
1792                 task_dump_owner(task, inode->i_mode, &stat->uid, &stat->gid);
1793         }
1794         rcu_read_unlock();
1795         return 0;
1796 }
1797
1798 /* dentry stuff */
1799
1800 /*
1801  * Set <pid>/... inode ownership (can change due to setuid(), etc.)
1802  */
1803 void pid_update_inode(struct task_struct *task, struct inode *inode)
1804 {
1805         task_dump_owner(task, inode->i_mode, &inode->i_uid, &inode->i_gid);
1806
1807         inode->i_mode &= ~(S_ISUID | S_ISGID);
1808         security_task_to_inode(task, inode);
1809 }
1810
1811 /*
1812  * Rewrite the inode's ownerships here because the owning task may have
1813  * performed a setuid(), etc.
1814  *
1815  */
1816 static int pid_revalidate(struct dentry *dentry, unsigned int flags)
1817 {
1818         struct inode *inode;
1819         struct task_struct *task;
1820
1821         if (flags & LOOKUP_RCU)
1822                 return -ECHILD;
1823
1824         inode = d_inode(dentry);
1825         task = get_proc_task(inode);
1826
1827         if (task) {
1828                 pid_update_inode(task, inode);
1829                 put_task_struct(task);
1830                 return 1;
1831         }
1832         return 0;
1833 }
1834
1835 static inline bool proc_inode_is_dead(struct inode *inode)
1836 {
1837         return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1838 }
1839
1840 int pid_delete_dentry(const struct dentry *dentry)
1841 {
1842         /* Is the task we represent dead?
1843          * If so, then don't put the dentry on the lru list,
1844          * kill it immediately.
1845          */
1846         return proc_inode_is_dead(d_inode(dentry));
1847 }
1848
1849 const struct dentry_operations pid_dentry_operations =
1850 {
1851         .d_revalidate   = pid_revalidate,
1852         .d_delete       = pid_delete_dentry,
1853 };
1854
1855 /* Lookups */
1856
1857 /*
1858  * Fill a directory entry.
1859  *
1860  * If possible create the dcache entry and derive our inode number and
1861  * file type from dcache entry.
1862  *
1863  * Since all of the proc inode numbers are dynamically generated, the inode
1864  * numbers do not exist until the inode is cache.  This means creating the
1865  * the dcache entry in readdir is necessary to keep the inode numbers
1866  * reported by readdir in sync with the inode numbers reported
1867  * by stat.
1868  */
1869 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1870         const char *name, unsigned int len,
1871         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1872 {
1873         struct dentry *child, *dir = file->f_path.dentry;
1874         struct qstr qname = QSTR_INIT(name, len);
1875         struct inode *inode;
1876         unsigned type = DT_UNKNOWN;
1877         ino_t ino = 1;
1878
1879         child = d_hash_and_lookup(dir, &qname);
1880         if (!child) {
1881                 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1882                 child = d_alloc_parallel(dir, &qname, &wq);
1883                 if (IS_ERR(child))
1884                         goto end_instantiate;
1885                 if (d_in_lookup(child)) {
1886                         struct dentry *res;
1887                         res = instantiate(child, task, ptr);
1888                         d_lookup_done(child);
1889                         if (unlikely(res)) {
1890                                 dput(child);
1891                                 child = res;
1892                                 if (IS_ERR(child))
1893                                         goto end_instantiate;
1894                         }
1895                 }
1896         }
1897         inode = d_inode(child);
1898         ino = inode->i_ino;
1899         type = inode->i_mode >> 12;
1900         dput(child);
1901 end_instantiate:
1902         return dir_emit(ctx, name, len, ino, type);
1903 }
1904
1905 /*
1906  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1907  * which represent vma start and end addresses.
1908  */
1909 static int dname_to_vma_addr(struct dentry *dentry,
1910                              unsigned long *start, unsigned long *end)
1911 {
1912         const char *str = dentry->d_name.name;
1913         unsigned long long sval, eval;
1914         unsigned int len;
1915
1916         if (str[0] == '0' && str[1] != '-')
1917                 return -EINVAL;
1918         len = _parse_integer(str, 16, &sval);
1919         if (len & KSTRTOX_OVERFLOW)
1920                 return -EINVAL;
1921         if (sval != (unsigned long)sval)
1922                 return -EINVAL;
1923         str += len;
1924
1925         if (*str != '-')
1926                 return -EINVAL;
1927         str++;
1928
1929         if (str[0] == '0' && str[1])
1930                 return -EINVAL;
1931         len = _parse_integer(str, 16, &eval);
1932         if (len & KSTRTOX_OVERFLOW)
1933                 return -EINVAL;
1934         if (eval != (unsigned long)eval)
1935                 return -EINVAL;
1936         str += len;
1937
1938         if (*str != '\0')
1939                 return -EINVAL;
1940
1941         *start = sval;
1942         *end = eval;
1943
1944         return 0;
1945 }
1946
1947 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1948 {
1949         unsigned long vm_start, vm_end;
1950         bool exact_vma_exists = false;
1951         struct mm_struct *mm = NULL;
1952         struct task_struct *task;
1953         struct inode *inode;
1954         int status = 0;
1955
1956         if (flags & LOOKUP_RCU)
1957                 return -ECHILD;
1958
1959         inode = d_inode(dentry);
1960         task = get_proc_task(inode);
1961         if (!task)
1962                 goto out_notask;
1963
1964         mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1965         if (IS_ERR_OR_NULL(mm))
1966                 goto out;
1967
1968         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1969                 down_read(&mm->mmap_sem);
1970                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1971                 up_read(&mm->mmap_sem);
1972         }
1973
1974         mmput(mm);
1975
1976         if (exact_vma_exists) {
1977                 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1978
1979                 security_task_to_inode(task, inode);
1980                 status = 1;
1981         }
1982
1983 out:
1984         put_task_struct(task);
1985
1986 out_notask:
1987         return status;
1988 }
1989
1990 static const struct dentry_operations tid_map_files_dentry_operations = {
1991         .d_revalidate   = map_files_d_revalidate,
1992         .d_delete       = pid_delete_dentry,
1993 };
1994
1995 static int map_files_get_link(struct dentry *dentry, struct path *path)
1996 {
1997         unsigned long vm_start, vm_end;
1998         struct vm_area_struct *vma;
1999         struct task_struct *task;
2000         struct mm_struct *mm;
2001         int rc;
2002
2003         rc = -ENOENT;
2004         task = get_proc_task(d_inode(dentry));
2005         if (!task)
2006                 goto out;
2007
2008         mm = get_task_mm(task);
2009         put_task_struct(task);
2010         if (!mm)
2011                 goto out;
2012
2013         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2014         if (rc)
2015                 goto out_mmput;
2016
2017         rc = -ENOENT;
2018         down_read(&mm->mmap_sem);
2019         vma = find_exact_vma(mm, vm_start, vm_end);
2020         if (vma && vma->vm_file) {
2021                 *path = vma->vm_file->f_path;
2022                 path_get(path);
2023                 rc = 0;
2024         }
2025         up_read(&mm->mmap_sem);
2026
2027 out_mmput:
2028         mmput(mm);
2029 out:
2030         return rc;
2031 }
2032
2033 struct map_files_info {
2034         unsigned long   start;
2035         unsigned long   end;
2036         fmode_t         mode;
2037 };
2038
2039 /*
2040  * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
2041  * symlinks may be used to bypass permissions on ancestor directories in the
2042  * path to the file in question.
2043  */
2044 static const char *
2045 proc_map_files_get_link(struct dentry *dentry,
2046                         struct inode *inode,
2047                         struct delayed_call *done)
2048 {
2049         if (!capable(CAP_SYS_ADMIN))
2050                 return ERR_PTR(-EPERM);
2051
2052         return proc_pid_get_link(dentry, inode, done);
2053 }
2054
2055 /*
2056  * Identical to proc_pid_link_inode_operations except for get_link()
2057  */
2058 static const struct inode_operations proc_map_files_link_inode_operations = {
2059         .readlink       = proc_pid_readlink,
2060         .get_link       = proc_map_files_get_link,
2061         .setattr        = proc_setattr,
2062 };
2063
2064 static struct dentry *
2065 proc_map_files_instantiate(struct dentry *dentry,
2066                            struct task_struct *task, const void *ptr)
2067 {
2068         fmode_t mode = (fmode_t)(unsigned long)ptr;
2069         struct proc_inode *ei;
2070         struct inode *inode;
2071
2072         inode = proc_pid_make_inode(dentry->d_sb, task, S_IFLNK |
2073                                     ((mode & FMODE_READ ) ? S_IRUSR : 0) |
2074                                     ((mode & FMODE_WRITE) ? S_IWUSR : 0));
2075         if (!inode)
2076                 return ERR_PTR(-ENOENT);
2077
2078         ei = PROC_I(inode);
2079         ei->op.proc_get_link = map_files_get_link;
2080
2081         inode->i_op = &proc_map_files_link_inode_operations;
2082         inode->i_size = 64;
2083
2084         d_set_d_op(dentry, &tid_map_files_dentry_operations);
2085         return d_splice_alias(inode, dentry);
2086 }
2087
2088 static struct dentry *proc_map_files_lookup(struct inode *dir,
2089                 struct dentry *dentry, unsigned int flags)
2090 {
2091         unsigned long vm_start, vm_end;
2092         struct vm_area_struct *vma;
2093         struct task_struct *task;
2094         struct dentry *result;
2095         struct mm_struct *mm;
2096
2097         result = ERR_PTR(-ENOENT);
2098         task = get_proc_task(dir);
2099         if (!task)
2100                 goto out;
2101
2102         result = ERR_PTR(-EACCES);
2103         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2104                 goto out_put_task;
2105
2106         result = ERR_PTR(-ENOENT);
2107         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2108                 goto out_put_task;
2109
2110         mm = get_task_mm(task);
2111         if (!mm)
2112                 goto out_put_task;
2113
2114         down_read(&mm->mmap_sem);
2115         vma = find_exact_vma(mm, vm_start, vm_end);
2116         if (!vma)
2117                 goto out_no_vma;
2118
2119         if (vma->vm_file)
2120                 result = proc_map_files_instantiate(dentry, task,
2121                                 (void *)(unsigned long)vma->vm_file->f_mode);
2122
2123 out_no_vma:
2124         up_read(&mm->mmap_sem);
2125         mmput(mm);
2126 out_put_task:
2127         put_task_struct(task);
2128 out:
2129         return result;
2130 }
2131
2132 static const struct inode_operations proc_map_files_inode_operations = {
2133         .lookup         = proc_map_files_lookup,
2134         .permission     = proc_fd_permission,
2135         .setattr        = proc_setattr,
2136 };
2137
2138 static int
2139 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2140 {
2141         struct vm_area_struct *vma;
2142         struct task_struct *task;
2143         struct mm_struct *mm;
2144         unsigned long nr_files, pos, i;
2145         struct flex_array *fa = NULL;
2146         struct map_files_info info;
2147         struct map_files_info *p;
2148         int ret;
2149
2150         ret = -ENOENT;
2151         task = get_proc_task(file_inode(file));
2152         if (!task)
2153                 goto out;
2154
2155         ret = -EACCES;
2156         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2157                 goto out_put_task;
2158
2159         ret = 0;
2160         if (!dir_emit_dots(file, ctx))
2161                 goto out_put_task;
2162
2163         mm = get_task_mm(task);
2164         if (!mm)
2165                 goto out_put_task;
2166         down_read(&mm->mmap_sem);
2167
2168         nr_files = 0;
2169
2170         /*
2171          * We need two passes here:
2172          *
2173          *  1) Collect vmas of mapped files with mmap_sem taken
2174          *  2) Release mmap_sem and instantiate entries
2175          *
2176          * otherwise we get lockdep complained, since filldir()
2177          * routine might require mmap_sem taken in might_fault().
2178          */
2179
2180         for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2181                 if (vma->vm_file && ++pos > ctx->pos)
2182                         nr_files++;
2183         }
2184
2185         if (nr_files) {
2186                 fa = flex_array_alloc(sizeof(info), nr_files,
2187                                         GFP_KERNEL);
2188                 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2189                                                 GFP_KERNEL)) {
2190                         ret = -ENOMEM;
2191                         if (fa)
2192                                 flex_array_free(fa);
2193                         up_read(&mm->mmap_sem);
2194                         mmput(mm);
2195                         goto out_put_task;
2196                 }
2197                 for (i = 0, vma = mm->mmap, pos = 2; vma;
2198                                 vma = vma->vm_next) {
2199                         if (!vma->vm_file)
2200                                 continue;
2201                         if (++pos <= ctx->pos)
2202                                 continue;
2203
2204                         info.start = vma->vm_start;
2205                         info.end = vma->vm_end;
2206                         info.mode = vma->vm_file->f_mode;
2207                         if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2208                                 BUG();
2209                 }
2210         }
2211         up_read(&mm->mmap_sem);
2212         mmput(mm);
2213
2214         for (i = 0; i < nr_files; i++) {
2215                 char buf[4 * sizeof(long) + 2]; /* max: %lx-%lx\0 */
2216                 unsigned int len;
2217
2218                 p = flex_array_get(fa, i);
2219                 len = snprintf(buf, sizeof(buf), "%lx-%lx", p->start, p->end);
2220                 if (!proc_fill_cache(file, ctx,
2221                                       buf, len,
2222                                       proc_map_files_instantiate,
2223                                       task,
2224                                       (void *)(unsigned long)p->mode))
2225                         break;
2226                 ctx->pos++;
2227         }
2228         if (fa)
2229                 flex_array_free(fa);
2230
2231 out_put_task:
2232         put_task_struct(task);
2233 out:
2234         return ret;
2235 }
2236
2237 static const struct file_operations proc_map_files_operations = {
2238         .read           = generic_read_dir,
2239         .iterate_shared = proc_map_files_readdir,
2240         .llseek         = generic_file_llseek,
2241 };
2242
2243 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2244 struct timers_private {
2245         struct pid *pid;
2246         struct task_struct *task;
2247         struct sighand_struct *sighand;
2248         struct pid_namespace *ns;
2249         unsigned long flags;
2250 };
2251
2252 static void *timers_start(struct seq_file *m, loff_t *pos)
2253 {
2254         struct timers_private *tp = m->private;
2255
2256         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2257         if (!tp->task)
2258                 return ERR_PTR(-ESRCH);
2259
2260         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2261         if (!tp->sighand)
2262                 return ERR_PTR(-ESRCH);
2263
2264         return seq_list_start(&tp->task->signal->posix_timers, *pos);
2265 }
2266
2267 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2268 {
2269         struct timers_private *tp = m->private;
2270         return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2271 }
2272
2273 static void timers_stop(struct seq_file *m, void *v)
2274 {
2275         struct timers_private *tp = m->private;
2276
2277         if (tp->sighand) {
2278                 unlock_task_sighand(tp->task, &tp->flags);
2279                 tp->sighand = NULL;
2280         }
2281
2282         if (tp->task) {
2283                 put_task_struct(tp->task);
2284                 tp->task = NULL;
2285         }
2286 }
2287
2288 static int show_timer(struct seq_file *m, void *v)
2289 {
2290         struct k_itimer *timer;
2291         struct timers_private *tp = m->private;
2292         int notify;
2293         static const char * const nstr[] = {
2294                 [SIGEV_SIGNAL] = "signal",
2295                 [SIGEV_NONE] = "none",
2296                 [SIGEV_THREAD] = "thread",
2297         };
2298
2299         timer = list_entry((struct list_head *)v, struct k_itimer, list);
2300         notify = timer->it_sigev_notify;
2301
2302         seq_printf(m, "ID: %d\n", timer->it_id);
2303         seq_printf(m, "signal: %d/%px\n",
2304                    timer->sigq->info.si_signo,
2305                    timer->sigq->info.si_value.sival_ptr);
2306         seq_printf(m, "notify: %s/%s.%d\n",
2307                    nstr[notify & ~SIGEV_THREAD_ID],
2308                    (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2309                    pid_nr_ns(timer->it_pid, tp->ns));
2310         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2311
2312         return 0;
2313 }
2314
2315 static const struct seq_operations proc_timers_seq_ops = {
2316         .start  = timers_start,
2317         .next   = timers_next,
2318         .stop   = timers_stop,
2319         .show   = show_timer,
2320 };
2321
2322 static int proc_timers_open(struct inode *inode, struct file *file)
2323 {
2324         struct timers_private *tp;
2325
2326         tp = __seq_open_private(file, &proc_timers_seq_ops,
2327                         sizeof(struct timers_private));
2328         if (!tp)
2329                 return -ENOMEM;
2330
2331         tp->pid = proc_pid(inode);
2332         tp->ns = proc_pid_ns(inode);
2333         return 0;
2334 }
2335
2336 static const struct file_operations proc_timers_operations = {
2337         .open           = proc_timers_open,
2338         .read           = seq_read,
2339         .llseek         = seq_lseek,
2340         .release        = seq_release_private,
2341 };
2342 #endif
2343
2344 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2345                                         size_t count, loff_t *offset)
2346 {
2347         struct inode *inode = file_inode(file);
2348         struct task_struct *p;
2349         u64 slack_ns;
2350         int err;
2351
2352         err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2353         if (err < 0)
2354                 return err;
2355
2356         p = get_proc_task(inode);
2357         if (!p)
2358                 return -ESRCH;
2359
2360         if (p != current) {
2361                 rcu_read_lock();
2362                 if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
2363                         rcu_read_unlock();
2364                         count = -EPERM;
2365                         goto out;
2366                 }
2367                 rcu_read_unlock();
2368
2369                 err = security_task_setscheduler(p);
2370                 if (err) {
2371                         count = err;
2372                         goto out;
2373                 }
2374         }
2375
2376         task_lock(p);
2377         if (slack_ns == 0)
2378                 p->timer_slack_ns = p->default_timer_slack_ns;
2379         else
2380                 p->timer_slack_ns = slack_ns;
2381         task_unlock(p);
2382
2383 out:
2384         put_task_struct(p);
2385
2386         return count;
2387 }
2388
2389 static int timerslack_ns_show(struct seq_file *m, void *v)
2390 {
2391         struct inode *inode = m->private;
2392         struct task_struct *p;
2393         int err = 0;
2394
2395         p = get_proc_task(inode);
2396         if (!p)
2397                 return -ESRCH;
2398
2399         if (p != current) {
2400                 rcu_read_lock();
2401                 if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
2402                         rcu_read_unlock();
2403                         err = -EPERM;
2404                         goto out;
2405                 }
2406                 rcu_read_unlock();
2407
2408                 err = security_task_getscheduler(p);
2409                 if (err)
2410                         goto out;
2411         }
2412
2413         task_lock(p);
2414         seq_printf(m, "%llu\n", p->timer_slack_ns);
2415         task_unlock(p);
2416
2417 out:
2418         put_task_struct(p);
2419
2420         return err;
2421 }
2422
2423 static int timerslack_ns_open(struct inode *inode, struct file *filp)
2424 {
2425         return single_open(filp, timerslack_ns_show, inode);
2426 }
2427
2428 static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2429         .open           = timerslack_ns_open,
2430         .read           = seq_read,
2431         .write          = timerslack_ns_write,
2432         .llseek         = seq_lseek,
2433         .release        = single_release,
2434 };
2435
2436 static struct dentry *proc_pident_instantiate(struct dentry *dentry,
2437         struct task_struct *task, const void *ptr)
2438 {
2439         const struct pid_entry *p = ptr;
2440         struct inode *inode;
2441         struct proc_inode *ei;
2442
2443         inode = proc_pid_make_inode(dentry->d_sb, task, p->mode);
2444         if (!inode)
2445                 return ERR_PTR(-ENOENT);
2446
2447         ei = PROC_I(inode);
2448         if (S_ISDIR(inode->i_mode))
2449                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2450         if (p->iop)
2451                 inode->i_op = p->iop;
2452         if (p->fop)
2453                 inode->i_fop = p->fop;
2454         ei->op = p->op;
2455         pid_update_inode(task, inode);
2456         d_set_d_op(dentry, &pid_dentry_operations);
2457         return d_splice_alias(inode, dentry);
2458 }
2459
2460 static struct dentry *proc_pident_lookup(struct inode *dir,
2461                                          struct dentry *dentry,
2462                                          const struct pid_entry *ents,
2463                                          unsigned int nents)
2464 {
2465         struct task_struct *task = get_proc_task(dir);
2466         const struct pid_entry *p, *last;
2467         struct dentry *res = ERR_PTR(-ENOENT);
2468
2469         if (!task)
2470                 goto out_no_task;
2471
2472         /*
2473          * Yes, it does not scale. And it should not. Don't add
2474          * new entries into /proc/<tgid>/ without very good reasons.
2475          */
2476         last = &ents[nents];
2477         for (p = ents; p < last; p++) {
2478                 if (p->len != dentry->d_name.len)
2479                         continue;
2480                 if (!memcmp(dentry->d_name.name, p->name, p->len)) {
2481                         res = proc_pident_instantiate(dentry, task, p);
2482                         break;
2483                 }
2484         }
2485         put_task_struct(task);
2486 out_no_task:
2487         return res;
2488 }
2489
2490 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2491                 const struct pid_entry *ents, unsigned int nents)
2492 {
2493         struct task_struct *task = get_proc_task(file_inode(file));
2494         const struct pid_entry *p;
2495
2496         if (!task)
2497                 return -ENOENT;
2498
2499         if (!dir_emit_dots(file, ctx))
2500                 goto out;
2501
2502         if (ctx->pos >= nents + 2)
2503                 goto out;
2504
2505         for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2506                 if (!proc_fill_cache(file, ctx, p->name, p->len,
2507                                 proc_pident_instantiate, task, p))
2508                         break;
2509                 ctx->pos++;
2510         }
2511 out:
2512         put_task_struct(task);
2513         return 0;
2514 }
2515
2516 #ifdef CONFIG_SECURITY
2517 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2518                                   size_t count, loff_t *ppos)
2519 {
2520         struct inode * inode = file_inode(file);
2521         char *p = NULL;
2522         ssize_t length;
2523         struct task_struct *task = get_proc_task(inode);
2524
2525         if (!task)
2526                 return -ESRCH;
2527
2528         length = security_getprocattr(task,
2529                                       (char*)file->f_path.dentry->d_name.name,
2530                                       &p);
2531         put_task_struct(task);
2532         if (length > 0)
2533                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2534         kfree(p);
2535         return length;
2536 }
2537
2538 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2539                                    size_t count, loff_t *ppos)
2540 {
2541         struct inode * inode = file_inode(file);
2542         struct task_struct *task;
2543         void *page;
2544         int rv;
2545
2546         rcu_read_lock();
2547         task = pid_task(proc_pid(inode), PIDTYPE_PID);
2548         if (!task) {
2549                 rcu_read_unlock();
2550                 return -ESRCH;
2551         }
2552         /* A task may only write its own attributes. */
2553         if (current != task) {
2554                 rcu_read_unlock();
2555                 return -EACCES;
2556         }
2557         rcu_read_unlock();
2558
2559         if (count > PAGE_SIZE)
2560                 count = PAGE_SIZE;
2561
2562         /* No partial writes. */
2563         if (*ppos != 0)
2564                 return -EINVAL;
2565
2566         page = memdup_user(buf, count);
2567         if (IS_ERR(page)) {
2568                 rv = PTR_ERR(page);
2569                 goto out;
2570         }
2571
2572         /* Guard against adverse ptrace interaction */
2573         rv = mutex_lock_interruptible(&current->signal->cred_guard_mutex);
2574         if (rv < 0)
2575                 goto out_free;
2576
2577         rv = security_setprocattr(file->f_path.dentry->d_name.name, page, count);
2578         mutex_unlock(&current->signal->cred_guard_mutex);
2579 out_free:
2580         kfree(page);
2581 out:
2582         return rv;
2583 }
2584
2585 static const struct file_operations proc_pid_attr_operations = {
2586         .read           = proc_pid_attr_read,
2587         .write          = proc_pid_attr_write,
2588         .llseek         = generic_file_llseek,
2589 };
2590
2591 static const struct pid_entry attr_dir_stuff[] = {
2592         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2593         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2594         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2595         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2596         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2597         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2598 };
2599
2600 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2601 {
2602         return proc_pident_readdir(file, ctx,
2603                                    attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2604 }
2605
2606 static const struct file_operations proc_attr_dir_operations = {
2607         .read           = generic_read_dir,
2608         .iterate_shared = proc_attr_dir_readdir,
2609         .llseek         = generic_file_llseek,
2610 };
2611
2612 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2613                                 struct dentry *dentry, unsigned int flags)
2614 {
2615         return proc_pident_lookup(dir, dentry,
2616                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2617 }
2618
2619 static const struct inode_operations proc_attr_dir_inode_operations = {
2620         .lookup         = proc_attr_dir_lookup,
2621         .getattr        = pid_getattr,
2622         .setattr        = proc_setattr,
2623 };
2624
2625 #endif
2626
2627 #ifdef CONFIG_ELF_CORE
2628 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2629                                          size_t count, loff_t *ppos)
2630 {
2631         struct task_struct *task = get_proc_task(file_inode(file));
2632         struct mm_struct *mm;
2633         char buffer[PROC_NUMBUF];
2634         size_t len;
2635         int ret;
2636
2637         if (!task)
2638                 return -ESRCH;
2639
2640         ret = 0;
2641         mm = get_task_mm(task);
2642         if (mm) {
2643                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2644                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2645                                 MMF_DUMP_FILTER_SHIFT));
2646                 mmput(mm);
2647                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2648         }
2649
2650         put_task_struct(task);
2651
2652         return ret;
2653 }
2654
2655 static ssize_t proc_coredump_filter_write(struct file *file,
2656                                           const char __user *buf,
2657                                           size_t count,
2658                                           loff_t *ppos)
2659 {
2660         struct task_struct *task;
2661         struct mm_struct *mm;
2662         unsigned int val;
2663         int ret;
2664         int i;
2665         unsigned long mask;
2666
2667         ret = kstrtouint_from_user(buf, count, 0, &val);
2668         if (ret < 0)
2669                 return ret;
2670
2671         ret = -ESRCH;
2672         task = get_proc_task(file_inode(file));
2673         if (!task)
2674                 goto out_no_task;
2675
2676         mm = get_task_mm(task);
2677         if (!mm)
2678                 goto out_no_mm;
2679         ret = 0;
2680
2681         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2682                 if (val & mask)
2683                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2684                 else
2685                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2686         }
2687
2688         mmput(mm);
2689  out_no_mm:
2690         put_task_struct(task);
2691  out_no_task:
2692         if (ret < 0)
2693                 return ret;
2694         return count;
2695 }
2696
2697 static const struct file_operations proc_coredump_filter_operations = {
2698         .read           = proc_coredump_filter_read,
2699         .write          = proc_coredump_filter_write,
2700         .llseek         = generic_file_llseek,
2701 };
2702 #endif
2703
2704 #ifdef CONFIG_TASK_IO_ACCOUNTING
2705 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2706 {
2707         struct task_io_accounting acct = task->ioac;
2708         unsigned long flags;
2709         int result;
2710
2711         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2712         if (result)
2713                 return result;
2714
2715         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2716                 result = -EACCES;
2717                 goto out_unlock;
2718         }
2719
2720         if (whole && lock_task_sighand(task, &flags)) {
2721                 struct task_struct *t = task;
2722
2723                 task_io_accounting_add(&acct, &task->signal->ioac);
2724                 while_each_thread(task, t)
2725                         task_io_accounting_add(&acct, &t->ioac);
2726
2727                 unlock_task_sighand(task, &flags);
2728         }
2729         seq_printf(m,
2730                    "rchar: %llu\n"
2731                    "wchar: %llu\n"
2732                    "syscr: %llu\n"
2733                    "syscw: %llu\n"
2734                    "read_bytes: %llu\n"
2735                    "write_bytes: %llu\n"
2736                    "cancelled_write_bytes: %llu\n",
2737                    (unsigned long long)acct.rchar,
2738                    (unsigned long long)acct.wchar,
2739                    (unsigned long long)acct.syscr,
2740                    (unsigned long long)acct.syscw,
2741                    (unsigned long long)acct.read_bytes,
2742                    (unsigned long long)acct.write_bytes,
2743                    (unsigned long long)acct.cancelled_write_bytes);
2744         result = 0;
2745
2746 out_unlock:
2747         mutex_unlock(&task->signal->cred_guard_mutex);
2748         return result;
2749 }
2750
2751 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2752                                   struct pid *pid, struct task_struct *task)
2753 {
2754         return do_io_accounting(task, m, 0);
2755 }
2756
2757 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2758                                    struct pid *pid, struct task_struct *task)
2759 {
2760         return do_io_accounting(task, m, 1);
2761 }
2762 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2763
2764 #ifdef CONFIG_USER_NS
2765 static int proc_id_map_open(struct inode *inode, struct file *file,
2766         const struct seq_operations *seq_ops)
2767 {
2768         struct user_namespace *ns = NULL;
2769         struct task_struct *task;
2770         struct seq_file *seq;
2771         int ret = -EINVAL;
2772
2773         task = get_proc_task(inode);
2774         if (task) {
2775                 rcu_read_lock();
2776                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2777                 rcu_read_unlock();
2778                 put_task_struct(task);
2779         }
2780         if (!ns)
2781                 goto err;
2782
2783         ret = seq_open(file, seq_ops);
2784         if (ret)
2785                 goto err_put_ns;
2786
2787         seq = file->private_data;
2788         seq->private = ns;
2789
2790         return 0;
2791 err_put_ns:
2792         put_user_ns(ns);
2793 err:
2794         return ret;
2795 }
2796
2797 static int proc_id_map_release(struct inode *inode, struct file *file)
2798 {
2799         struct seq_file *seq = file->private_data;
2800         struct user_namespace *ns = seq->private;
2801         put_user_ns(ns);
2802         return seq_release(inode, file);
2803 }
2804
2805 static int proc_uid_map_open(struct inode *inode, struct file *file)
2806 {
2807         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2808 }
2809
2810 static int proc_gid_map_open(struct inode *inode, struct file *file)
2811 {
2812         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2813 }
2814
2815 static int proc_projid_map_open(struct inode *inode, struct file *file)
2816 {
2817         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2818 }
2819
2820 static const struct file_operations proc_uid_map_operations = {
2821         .open           = proc_uid_map_open,
2822         .write          = proc_uid_map_write,
2823         .read           = seq_read,
2824         .llseek         = seq_lseek,
2825         .release        = proc_id_map_release,
2826 };
2827
2828 static const struct file_operations proc_gid_map_operations = {
2829         .open           = proc_gid_map_open,
2830         .write          = proc_gid_map_write,
2831         .read           = seq_read,
2832         .llseek         = seq_lseek,
2833         .release        = proc_id_map_release,
2834 };
2835
2836 static const struct file_operations proc_projid_map_operations = {
2837         .open           = proc_projid_map_open,
2838         .write          = proc_projid_map_write,
2839         .read           = seq_read,
2840         .llseek         = seq_lseek,
2841         .release        = proc_id_map_release,
2842 };
2843
2844 static int proc_setgroups_open(struct inode *inode, struct file *file)
2845 {
2846         struct user_namespace *ns = NULL;
2847         struct task_struct *task;
2848         int ret;
2849
2850         ret = -ESRCH;
2851         task = get_proc_task(inode);
2852         if (task) {
2853                 rcu_read_lock();
2854                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2855                 rcu_read_unlock();
2856                 put_task_struct(task);
2857         }
2858         if (!ns)
2859                 goto err;
2860
2861         if (file->f_mode & FMODE_WRITE) {
2862                 ret = -EACCES;
2863                 if (!ns_capable(ns, CAP_SYS_ADMIN))
2864                         goto err_put_ns;
2865         }
2866
2867         ret = single_open(file, &proc_setgroups_show, ns);
2868         if (ret)
2869                 goto err_put_ns;
2870
2871         return 0;
2872 err_put_ns:
2873         put_user_ns(ns);
2874 err:
2875         return ret;
2876 }
2877
2878 static int proc_setgroups_release(struct inode *inode, struct file *file)
2879 {
2880         struct seq_file *seq = file->private_data;
2881         struct user_namespace *ns = seq->private;
2882         int ret = single_release(inode, file);
2883         put_user_ns(ns);
2884         return ret;
2885 }
2886
2887 static const struct file_operations proc_setgroups_operations = {
2888         .open           = proc_setgroups_open,
2889         .write          = proc_setgroups_write,
2890         .read           = seq_read,
2891         .llseek         = seq_lseek,
2892         .release        = proc_setgroups_release,
2893 };
2894 #endif /* CONFIG_USER_NS */
2895
2896 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2897                                 struct pid *pid, struct task_struct *task)
2898 {
2899         int err = lock_trace(task);
2900         if (!err) {
2901                 seq_printf(m, "%08x\n", task->personality);
2902                 unlock_trace(task);
2903         }
2904         return err;
2905 }
2906
2907 #ifdef CONFIG_LIVEPATCH
2908 static int proc_pid_patch_state(struct seq_file *m, struct pid_namespace *ns,
2909                                 struct pid *pid, struct task_struct *task)
2910 {
2911         seq_printf(m, "%d\n", task->patch_state);
2912         return 0;
2913 }
2914 #endif /* CONFIG_LIVEPATCH */
2915
2916 #ifdef CONFIG_STACKLEAK_METRICS
2917 static int proc_stack_depth(struct seq_file *m, struct pid_namespace *ns,
2918                                 struct pid *pid, struct task_struct *task)
2919 {
2920         unsigned long prev_depth = THREAD_SIZE -
2921                                 (task->prev_lowest_stack & (THREAD_SIZE - 1));
2922         unsigned long depth = THREAD_SIZE -
2923                                 (task->lowest_stack & (THREAD_SIZE - 1));
2924
2925         seq_printf(m, "previous stack depth: %lu\nstack depth: %lu\n",
2926                                                         prev_depth, depth);
2927         return 0;
2928 }
2929 #endif /* CONFIG_STACKLEAK_METRICS */
2930
2931 /*
2932  * Thread groups
2933  */
2934 static const struct file_operations proc_task_operations;
2935 static const struct inode_operations proc_task_inode_operations;
2936
2937 static const struct pid_entry tgid_base_stuff[] = {
2938         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2939         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2940         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2941         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2942         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2943 #ifdef CONFIG_NET
2944         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2945 #endif
2946         REG("environ",    S_IRUSR, proc_environ_operations),
2947         REG("auxv",       S_IRUSR, proc_auxv_operations),
2948         ONE("status",     S_IRUGO, proc_pid_status),
2949         ONE("personality", S_IRUSR, proc_pid_personality),
2950         ONE("limits",     S_IRUGO, proc_pid_limits),
2951 #ifdef CONFIG_SCHED_DEBUG
2952         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2953 #endif
2954 #ifdef CONFIG_SCHED_AUTOGROUP
2955         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2956 #endif
2957         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2958 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2959         ONE("syscall",    S_IRUSR, proc_pid_syscall),
2960 #endif
2961         REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
2962         ONE("stat",       S_IRUGO, proc_tgid_stat),
2963         ONE("statm",      S_IRUGO, proc_pid_statm),
2964         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2965 #ifdef CONFIG_NUMA
2966         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2967 #endif
2968         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2969         LNK("cwd",        proc_cwd_link),
2970         LNK("root",       proc_root_link),
2971         LNK("exe",        proc_exe_link),
2972         REG("mounts",     S_IRUGO, proc_mounts_operations),
2973         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2974         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2975 #ifdef CONFIG_PROC_PAGE_MONITOR
2976         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2977         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2978         REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
2979         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2980 #endif
2981 #ifdef CONFIG_SECURITY
2982         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2983 #endif
2984 #ifdef CONFIG_KALLSYMS
2985         ONE("wchan",      S_IRUGO, proc_pid_wchan),
2986 #endif
2987 #ifdef CONFIG_STACKTRACE
2988         ONE("stack",      S_IRUSR, proc_pid_stack),
2989 #endif
2990 #ifdef CONFIG_SCHED_INFO
2991         ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
2992 #endif
2993 #ifdef CONFIG_LATENCYTOP
2994         REG("latency",  S_IRUGO, proc_lstats_operations),
2995 #endif
2996 #ifdef CONFIG_PROC_PID_CPUSET
2997         ONE("cpuset",     S_IRUGO, proc_cpuset_show),
2998 #endif
2999 #ifdef CONFIG_CGROUPS
3000         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3001 #endif
3002         ONE("oom_score",  S_IRUGO, proc_oom_score),
3003         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3004         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3005 #ifdef CONFIG_AUDITSYSCALL
3006         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
3007         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3008 #endif
3009 #ifdef CONFIG_FAULT_INJECTION
3010         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3011         REG("fail-nth", 0644, proc_fail_nth_operations),
3012 #endif
3013 #ifdef CONFIG_ELF_CORE
3014         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3015 #endif
3016 #ifdef CONFIG_TASK_IO_ACCOUNTING
3017         ONE("io",       S_IRUSR, proc_tgid_io_accounting),
3018 #endif
3019 #ifdef CONFIG_USER_NS
3020         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3021         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3022         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3023         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3024 #endif
3025 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
3026         REG("timers",     S_IRUGO, proc_timers_operations),
3027 #endif
3028         REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
3029 #ifdef CONFIG_LIVEPATCH
3030         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
3031 #endif
3032 #ifdef CONFIG_STACKLEAK_METRICS
3033         ONE("stack_depth", S_IRUGO, proc_stack_depth),
3034 #endif
3035 };
3036
3037 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
3038 {
3039         return proc_pident_readdir(file, ctx,
3040                                    tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3041 }
3042
3043 static const struct file_operations proc_tgid_base_operations = {
3044         .read           = generic_read_dir,
3045         .iterate_shared = proc_tgid_base_readdir,
3046         .llseek         = generic_file_llseek,
3047 };
3048
3049 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3050 {
3051         return proc_pident_lookup(dir, dentry,
3052                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3053 }
3054
3055 static const struct inode_operations proc_tgid_base_inode_operations = {
3056         .lookup         = proc_tgid_base_lookup,
3057         .getattr        = pid_getattr,
3058         .setattr        = proc_setattr,
3059         .permission     = proc_pid_permission,
3060 };
3061
3062 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
3063 {
3064         struct dentry *dentry, *leader, *dir;
3065         char buf[10 + 1];
3066         struct qstr name;
3067
3068         name.name = buf;
3069         name.len = snprintf(buf, sizeof(buf), "%u", pid);
3070         /* no ->d_hash() rejects on procfs */
3071         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
3072         if (dentry) {
3073                 d_invalidate(dentry);
3074                 dput(dentry);
3075         }
3076
3077         if (pid == tgid)
3078                 return;
3079
3080         name.name = buf;
3081         name.len = snprintf(buf, sizeof(buf), "%u", tgid);
3082         leader = d_hash_and_lookup(mnt->mnt_root, &name);
3083         if (!leader)
3084                 goto out;
3085
3086         name.name = "task";
3087         name.len = strlen(name.name);
3088         dir = d_hash_and_lookup(leader, &name);
3089         if (!dir)
3090                 goto out_put_leader;
3091
3092         name.name = buf;
3093         name.len = snprintf(buf, sizeof(buf), "%u", pid);
3094         dentry = d_hash_and_lookup(dir, &name);
3095         if (dentry) {
3096                 d_invalidate(dentry);
3097                 dput(dentry);
3098         }
3099
3100         dput(dir);
3101 out_put_leader:
3102         dput(leader);
3103 out:
3104         return;
3105 }
3106
3107 /**
3108  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
3109  * @task: task that should be flushed.
3110  *
3111  * When flushing dentries from proc, one needs to flush them from global
3112  * proc (proc_mnt) and from all the namespaces' procs this task was seen
3113  * in. This call is supposed to do all of this job.
3114  *
3115  * Looks in the dcache for
3116  * /proc/@pid
3117  * /proc/@tgid/task/@pid
3118  * if either directory is present flushes it and all of it'ts children
3119  * from the dcache.
3120  *
3121  * It is safe and reasonable to cache /proc entries for a task until
3122  * that task exits.  After that they just clog up the dcache with
3123  * useless entries, possibly causing useful dcache entries to be
3124  * flushed instead.  This routine is proved to flush those useless
3125  * dcache entries at process exit time.
3126  *
3127  * NOTE: This routine is just an optimization so it does not guarantee
3128  *       that no dcache entries will exist at process exit time it
3129  *       just makes it very unlikely that any will persist.
3130  */
3131
3132 void proc_flush_task(struct task_struct *task)
3133 {
3134         int i;
3135         struct pid *pid, *tgid;
3136         struct upid *upid;
3137
3138         pid = task_pid(task);
3139         tgid = task_tgid(task);
3140
3141         for (i = 0; i <= pid->level; i++) {
3142                 upid = &pid->numbers[i];
3143                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3144                                         tgid->numbers[i].nr);
3145         }
3146 }
3147
3148 static struct dentry *proc_pid_instantiate(struct dentry * dentry,
3149                                    struct task_struct *task, const void *ptr)
3150 {
3151         struct inode *inode;
3152
3153         inode = proc_pid_make_inode(dentry->d_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3154         if (!inode)
3155                 return ERR_PTR(-ENOENT);
3156
3157         inode->i_op = &proc_tgid_base_inode_operations;
3158         inode->i_fop = &proc_tgid_base_operations;
3159         inode->i_flags|=S_IMMUTABLE;
3160
3161         set_nlink(inode, nlink_tgid);
3162         pid_update_inode(task, inode);
3163
3164         d_set_d_op(dentry, &pid_dentry_operations);
3165         return d_splice_alias(inode, dentry);
3166 }
3167
3168 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3169 {
3170         struct task_struct *task;
3171         unsigned tgid;
3172         struct pid_namespace *ns;
3173         struct dentry *result = ERR_PTR(-ENOENT);
3174
3175         tgid = name_to_int(&dentry->d_name);
3176         if (tgid == ~0U)
3177                 goto out;
3178
3179         ns = dentry->d_sb->s_fs_info;
3180         rcu_read_lock();
3181         task = find_task_by_pid_ns(tgid, ns);
3182         if (task)
3183                 get_task_struct(task);
3184         rcu_read_unlock();
3185         if (!task)
3186                 goto out;
3187
3188         result = proc_pid_instantiate(dentry, task, NULL);
3189         put_task_struct(task);
3190 out:
3191         return result;
3192 }
3193
3194 /*
3195  * Find the first task with tgid >= tgid
3196  *
3197  */
3198 struct tgid_iter {
3199         unsigned int tgid;
3200         struct task_struct *task;
3201 };
3202 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3203 {
3204         struct pid *pid;
3205
3206         if (iter.task)
3207                 put_task_struct(iter.task);
3208         rcu_read_lock();
3209 retry:
3210         iter.task = NULL;
3211         pid = find_ge_pid(iter.tgid, ns);
3212         if (pid) {
3213                 iter.tgid = pid_nr_ns(pid, ns);
3214                 iter.task = pid_task(pid, PIDTYPE_PID);
3215                 /* What we to know is if the pid we have find is the
3216                  * pid of a thread_group_leader.  Testing for task
3217                  * being a thread_group_leader is the obvious thing
3218                  * todo but there is a window when it fails, due to
3219                  * the pid transfer logic in de_thread.
3220                  *
3221                  * So we perform the straight forward test of seeing
3222                  * if the pid we have found is the pid of a thread
3223                  * group leader, and don't worry if the task we have
3224                  * found doesn't happen to be a thread group leader.
3225                  * As we don't care in the case of readdir.
3226                  */
3227                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3228                         iter.tgid += 1;
3229                         goto retry;
3230                 }
3231                 get_task_struct(iter.task);
3232         }
3233         rcu_read_unlock();
3234         return iter;
3235 }
3236
3237 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3238
3239 /* for the /proc/ directory itself, after non-process stuff has been done */
3240 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3241 {
3242         struct tgid_iter iter;
3243         struct pid_namespace *ns = proc_pid_ns(file_inode(file));
3244         loff_t pos = ctx->pos;
3245
3246         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3247                 return 0;
3248
3249         if (pos == TGID_OFFSET - 2) {
3250                 struct inode *inode = d_inode(ns->proc_self);
3251                 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3252                         return 0;
3253                 ctx->pos = pos = pos + 1;
3254         }
3255         if (pos == TGID_OFFSET - 1) {
3256                 struct inode *inode = d_inode(ns->proc_thread_self);
3257                 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3258                         return 0;
3259                 ctx->pos = pos = pos + 1;
3260         }
3261         iter.tgid = pos - TGID_OFFSET;
3262         iter.task = NULL;
3263         for (iter = next_tgid(ns, iter);
3264              iter.task;
3265              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3266                 char name[10 + 1];
3267                 unsigned int len;
3268
3269                 cond_resched();
3270                 if (!has_pid_permissions(ns, iter.task, HIDEPID_INVISIBLE))
3271                         continue;
3272
3273                 len = snprintf(name, sizeof(name), "%u", iter.tgid);
3274                 ctx->pos = iter.tgid + TGID_OFFSET;
3275                 if (!proc_fill_cache(file, ctx, name, len,
3276                                      proc_pid_instantiate, iter.task, NULL)) {
3277                         put_task_struct(iter.task);
3278                         return 0;
3279                 }
3280         }
3281         ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3282         return 0;
3283 }
3284
3285 /*
3286  * proc_tid_comm_permission is a special permission function exclusively
3287  * used for the node /proc/<pid>/task/<tid>/comm.
3288  * It bypasses generic permission checks in the case where a task of the same
3289  * task group attempts to access the node.
3290  * The rationale behind this is that glibc and bionic access this node for
3291  * cross thread naming (pthread_set/getname_np(!self)). However, if
3292  * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3293  * which locks out the cross thread naming implementation.
3294  * This function makes sure that the node is always accessible for members of
3295  * same thread group.
3296  */
3297 static int proc_tid_comm_permission(struct inode *inode, int mask)
3298 {
3299         bool is_same_tgroup;
3300         struct task_struct *task;
3301
3302         task = get_proc_task(inode);
3303         if (!task)
3304                 return -ESRCH;
3305         is_same_tgroup = same_thread_group(current, task);
3306         put_task_struct(task);
3307
3308         if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3309                 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3310                  * read or written by the members of the corresponding
3311                  * thread group.
3312                  */
3313                 return 0;
3314         }
3315
3316         return generic_permission(inode, mask);
3317 }
3318
3319 static const struct inode_operations proc_tid_comm_inode_operations = {
3320                 .permission = proc_tid_comm_permission,
3321 };
3322
3323 /*
3324  * Tasks
3325  */
3326 static const struct pid_entry tid_base_stuff[] = {
3327         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3328         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3329         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3330 #ifdef CONFIG_NET
3331         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3332 #endif
3333         REG("environ",   S_IRUSR, proc_environ_operations),
3334         REG("auxv",      S_IRUSR, proc_auxv_operations),
3335         ONE("status",    S_IRUGO, proc_pid_status),
3336         ONE("personality", S_IRUSR, proc_pid_personality),
3337         ONE("limits",    S_IRUGO, proc_pid_limits),
3338 #ifdef CONFIG_SCHED_DEBUG
3339         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3340 #endif
3341         NOD("comm",      S_IFREG|S_IRUGO|S_IWUSR,
3342                          &proc_tid_comm_inode_operations,
3343                          &proc_pid_set_comm_operations, {}),
3344 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3345         ONE("syscall",   S_IRUSR, proc_pid_syscall),
3346 #endif
3347         REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3348         ONE("stat",      S_IRUGO, proc_tid_stat),
3349         ONE("statm",     S_IRUGO, proc_pid_statm),
3350         REG("maps",      S_IRUGO, proc_pid_maps_operations),
3351 #ifdef CONFIG_PROC_CHILDREN
3352         REG("children",  S_IRUGO, proc_tid_children_operations),
3353 #endif
3354 #ifdef CONFIG_NUMA
3355         REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
3356 #endif
3357         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3358         LNK("cwd",       proc_cwd_link),
3359         LNK("root",      proc_root_link),
3360         LNK("exe",       proc_exe_link),
3361         REG("mounts",    S_IRUGO, proc_mounts_operations),
3362         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3363 #ifdef CONFIG_PROC_PAGE_MONITOR
3364         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3365         REG("smaps",     S_IRUGO, proc_pid_smaps_operations),
3366         REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3367         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3368 #endif
3369 #ifdef CONFIG_SECURITY
3370         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3371 #endif
3372 #ifdef CONFIG_KALLSYMS
3373         ONE("wchan",     S_IRUGO, proc_pid_wchan),
3374 #endif
3375 #ifdef CONFIG_STACKTRACE
3376         ONE("stack",      S_IRUSR, proc_pid_stack),
3377 #endif
3378 #ifdef CONFIG_SCHED_INFO
3379         ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3380 #endif
3381 #ifdef CONFIG_LATENCYTOP
3382         REG("latency",  S_IRUGO, proc_lstats_operations),
3383 #endif
3384 #ifdef CONFIG_PROC_PID_CPUSET
3385         ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3386 #endif
3387 #ifdef CONFIG_CGROUPS
3388         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3389 #endif
3390         ONE("oom_score", S_IRUGO, proc_oom_score),
3391         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3392         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3393 #ifdef CONFIG_AUDITSYSCALL
3394         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3395         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3396 #endif
3397 #ifdef CONFIG_FAULT_INJECTION
3398         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3399         REG("fail-nth", 0644, proc_fail_nth_operations),
3400 #endif
3401 #ifdef CONFIG_TASK_IO_ACCOUNTING
3402         ONE("io",       S_IRUSR, proc_tid_io_accounting),
3403 #endif
3404 #ifdef CONFIG_USER_NS
3405         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3406         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3407         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3408         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3409 #endif
3410 #ifdef CONFIG_LIVEPATCH
3411         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
3412 #endif
3413 };
3414
3415 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3416 {
3417         return proc_pident_readdir(file, ctx,
3418                                    tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3419 }
3420
3421 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3422 {
3423         return proc_pident_lookup(dir, dentry,
3424                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3425 }
3426
3427 static const struct file_operations proc_tid_base_operations = {
3428         .read           = generic_read_dir,
3429         .iterate_shared = proc_tid_base_readdir,
3430         .llseek         = generic_file_llseek,
3431 };
3432
3433 static const struct inode_operations proc_tid_base_inode_operations = {
3434         .lookup         = proc_tid_base_lookup,
3435         .getattr        = pid_getattr,
3436         .setattr        = proc_setattr,
3437 };
3438
3439 static struct dentry *proc_task_instantiate(struct dentry *dentry,
3440         struct task_struct *task, const void *ptr)
3441 {
3442         struct inode *inode;
3443         inode = proc_pid_make_inode(dentry->d_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3444         if (!inode)
3445                 return ERR_PTR(-ENOENT);
3446
3447         inode->i_op = &proc_tid_base_inode_operations;
3448         inode->i_fop = &proc_tid_base_operations;
3449         inode->i_flags |= S_IMMUTABLE;
3450
3451         set_nlink(inode, nlink_tid);
3452         pid_update_inode(task, inode);
3453
3454         d_set_d_op(dentry, &pid_dentry_operations);
3455         return d_splice_alias(inode, dentry);
3456 }
3457
3458 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3459 {
3460         struct task_struct *task;
3461         struct task_struct *leader = get_proc_task(dir);
3462         unsigned tid;
3463         struct pid_namespace *ns;
3464         struct dentry *result = ERR_PTR(-ENOENT);
3465
3466         if (!leader)
3467                 goto out_no_task;
3468
3469         tid = name_to_int(&dentry->d_name);
3470         if (tid == ~0U)
3471                 goto out;
3472
3473         ns = dentry->d_sb->s_fs_info;
3474         rcu_read_lock();
3475         task = find_task_by_pid_ns(tid, ns);
3476         if (task)
3477                 get_task_struct(task);
3478         rcu_read_unlock();
3479         if (!task)
3480                 goto out;
3481         if (!same_thread_group(leader, task))
3482                 goto out_drop_task;
3483
3484         result = proc_task_instantiate(dentry, task, NULL);
3485 out_drop_task:
3486         put_task_struct(task);
3487 out:
3488         put_task_struct(leader);
3489 out_no_task:
3490         return result;
3491 }
3492
3493 /*
3494  * Find the first tid of a thread group to return to user space.
3495  *
3496  * Usually this is just the thread group leader, but if the users
3497  * buffer was too small or there was a seek into the middle of the
3498  * directory we have more work todo.
3499  *
3500  * In the case of a short read we start with find_task_by_pid.
3501  *
3502  * In the case of a seek we start with the leader and walk nr
3503  * threads past it.
3504  */
3505 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3506                                         struct pid_namespace *ns)
3507 {
3508         struct task_struct *pos, *task;
3509         unsigned long nr = f_pos;
3510
3511         if (nr != f_pos)        /* 32bit overflow? */
3512                 return NULL;
3513
3514         rcu_read_lock();
3515         task = pid_task(pid, PIDTYPE_PID);
3516         if (!task)
3517                 goto fail;
3518
3519         /* Attempt to start with the tid of a thread */
3520         if (tid && nr) {
3521                 pos = find_task_by_pid_ns(tid, ns);
3522                 if (pos && same_thread_group(pos, task))
3523                         goto found;
3524         }
3525
3526         /* If nr exceeds the number of threads there is nothing todo */
3527         if (nr >= get_nr_threads(task))
3528                 goto fail;
3529
3530         /* If we haven't found our starting place yet start
3531          * with the leader and walk nr threads forward.
3532          */
3533         pos = task = task->group_leader;
3534         do {
3535                 if (!nr--)
3536                         goto found;
3537         } while_each_thread(task, pos);
3538 fail:
3539         pos = NULL;
3540         goto out;
3541 found:
3542         get_task_struct(pos);
3543 out:
3544         rcu_read_unlock();
3545         return pos;
3546 }
3547
3548 /*
3549  * Find the next thread in the thread list.
3550  * Return NULL if there is an error or no next thread.
3551  *
3552  * The reference to the input task_struct is released.
3553  */
3554 static struct task_struct *next_tid(struct task_struct *start)
3555 {
3556         struct task_struct *pos = NULL;
3557         rcu_read_lock();
3558         if (pid_alive(start)) {
3559                 pos = next_thread(start);
3560                 if (thread_group_leader(pos))
3561                         pos = NULL;
3562                 else
3563                         get_task_struct(pos);
3564         }
3565         rcu_read_unlock();
3566         put_task_struct(start);
3567         return pos;
3568 }
3569
3570 /* for the /proc/TGID/task/ directories */
3571 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3572 {
3573         struct inode *inode = file_inode(file);
3574         struct task_struct *task;
3575         struct pid_namespace *ns;
3576         int tid;
3577
3578         if (proc_inode_is_dead(inode))
3579                 return -ENOENT;
3580
3581         if (!dir_emit_dots(file, ctx))
3582                 return 0;
3583
3584         /* f_version caches the tgid value that the last readdir call couldn't
3585          * return. lseek aka telldir automagically resets f_version to 0.
3586          */
3587         ns = proc_pid_ns(inode);
3588         tid = (int)file->f_version;
3589         file->f_version = 0;
3590         for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3591              task;
3592              task = next_tid(task), ctx->pos++) {
3593                 char name[10 + 1];
3594                 unsigned int len;
3595                 tid = task_pid_nr_ns(task, ns);
3596                 len = snprintf(name, sizeof(name), "%u", tid);
3597                 if (!proc_fill_cache(file, ctx, name, len,
3598                                 proc_task_instantiate, task, NULL)) {
3599                         /* returning this tgid failed, save it as the first
3600                          * pid for the next readir call */
3601                         file->f_version = (u64)tid;
3602                         put_task_struct(task);
3603                         break;
3604                 }
3605         }
3606
3607         return 0;
3608 }
3609
3610 static int proc_task_getattr(const struct path *path, struct kstat *stat,
3611                              u32 request_mask, unsigned int query_flags)
3612 {
3613         struct inode *inode = d_inode(path->dentry);
3614         struct task_struct *p = get_proc_task(inode);
3615         generic_fillattr(inode, stat);
3616
3617         if (p) {
3618                 stat->nlink += get_nr_threads(p);
3619                 put_task_struct(p);
3620         }
3621
3622         return 0;
3623 }
3624
3625 static const struct inode_operations proc_task_inode_operations = {
3626         .lookup         = proc_task_lookup,
3627         .getattr        = proc_task_getattr,
3628         .setattr        = proc_setattr,
3629         .permission     = proc_pid_permission,
3630 };
3631
3632 static const struct file_operations proc_task_operations = {
3633         .read           = generic_read_dir,
3634         .iterate_shared = proc_task_readdir,
3635         .llseek         = generic_file_llseek,
3636 };
3637
3638 void __init set_proc_pid_nlink(void)
3639 {
3640         nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3641         nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3642 }