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                                 p->signal->oom_score_adj = oom_adj;
1090                                 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1091                                         p->signal->oom_score_adj_min = (short)oom_adj;
1092                         }
1093                         task_unlock(p);
1094                 }
1095                 rcu_read_unlock();
1096                 mmdrop(mm);
1097         }
1098 err_unlock:
1099         mutex_unlock(&oom_adj_mutex);
1100         put_task_struct(task);
1101         return err;
1102 }
1103
1104 /*
1105  * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1106  * kernels.  The effective policy is defined by oom_score_adj, which has a
1107  * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1108  * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1109  * Processes that become oom disabled via oom_adj will still be oom disabled
1110  * with this implementation.
1111  *
1112  * oom_adj cannot be removed since existing userspace binaries use it.
1113  */
1114 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1115                              size_t count, loff_t *ppos)
1116 {
1117         char buffer[PROC_NUMBUF];
1118         int oom_adj;
1119         int err;
1120
1121         memset(buffer, 0, sizeof(buffer));
1122         if (count > sizeof(buffer) - 1)
1123                 count = sizeof(buffer) - 1;
1124         if (copy_from_user(buffer, buf, count)) {
1125                 err = -EFAULT;
1126                 goto out;
1127         }
1128
1129         err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1130         if (err)
1131                 goto out;
1132         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1133              oom_adj != OOM_DISABLE) {
1134                 err = -EINVAL;
1135                 goto out;
1136         }
1137
1138         /*
1139          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1140          * value is always attainable.
1141          */
1142         if (oom_adj == OOM_ADJUST_MAX)
1143                 oom_adj = OOM_SCORE_ADJ_MAX;
1144         else
1145                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1146
1147         err = __set_oom_adj(file, oom_adj, true);
1148 out:
1149         return err < 0 ? err : count;
1150 }
1151
1152 static const struct file_operations proc_oom_adj_operations = {
1153         .read           = oom_adj_read,
1154         .write          = oom_adj_write,
1155         .llseek         = generic_file_llseek,
1156 };
1157
1158 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1159                                         size_t count, loff_t *ppos)
1160 {
1161         struct task_struct *task = get_proc_task(file_inode(file));
1162         char buffer[PROC_NUMBUF];
1163         short oom_score_adj = OOM_SCORE_ADJ_MIN;
1164         size_t len;
1165
1166         if (!task)
1167                 return -ESRCH;
1168         oom_score_adj = task->signal->oom_score_adj;
1169         put_task_struct(task);
1170         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1171         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1172 }
1173
1174 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1175                                         size_t count, loff_t *ppos)
1176 {
1177         char buffer[PROC_NUMBUF];
1178         int oom_score_adj;
1179         int err;
1180
1181         memset(buffer, 0, sizeof(buffer));
1182         if (count > sizeof(buffer) - 1)
1183                 count = sizeof(buffer) - 1;
1184         if (copy_from_user(buffer, buf, count)) {
1185                 err = -EFAULT;
1186                 goto out;
1187         }
1188
1189         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1190         if (err)
1191                 goto out;
1192         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1193                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1194                 err = -EINVAL;
1195                 goto out;
1196         }
1197
1198         err = __set_oom_adj(file, oom_score_adj, false);
1199 out:
1200         return err < 0 ? err : count;
1201 }
1202
1203 static const struct file_operations proc_oom_score_adj_operations = {
1204         .read           = oom_score_adj_read,
1205         .write          = oom_score_adj_write,
1206         .llseek         = default_llseek,
1207 };
1208
1209 #ifdef CONFIG_AUDITSYSCALL
1210 #define TMPBUFLEN 11
1211 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1212                                   size_t count, loff_t *ppos)
1213 {
1214         struct inode * inode = file_inode(file);
1215         struct task_struct *task = get_proc_task(inode);
1216         ssize_t length;
1217         char tmpbuf[TMPBUFLEN];
1218
1219         if (!task)
1220                 return -ESRCH;
1221         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1222                            from_kuid(file->f_cred->user_ns,
1223                                      audit_get_loginuid(task)));
1224         put_task_struct(task);
1225         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1226 }
1227
1228 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1229                                    size_t count, loff_t *ppos)
1230 {
1231         struct inode * inode = file_inode(file);
1232         uid_t loginuid;
1233         kuid_t kloginuid;
1234         int rv;
1235
1236         rcu_read_lock();
1237         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1238                 rcu_read_unlock();
1239                 return -EPERM;
1240         }
1241         rcu_read_unlock();
1242
1243         if (*ppos != 0) {
1244                 /* No partial writes. */
1245                 return -EINVAL;
1246         }
1247
1248         rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1249         if (rv < 0)
1250                 return rv;
1251
1252         /* is userspace tring to explicitly UNSET the loginuid? */
1253         if (loginuid == AUDIT_UID_UNSET) {
1254                 kloginuid = INVALID_UID;
1255         } else {
1256                 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1257                 if (!uid_valid(kloginuid))
1258                         return -EINVAL;
1259         }
1260
1261         rv = audit_set_loginuid(kloginuid);
1262         if (rv < 0)
1263                 return rv;
1264         return count;
1265 }
1266
1267 static const struct file_operations proc_loginuid_operations = {
1268         .read           = proc_loginuid_read,
1269         .write          = proc_loginuid_write,
1270         .llseek         = generic_file_llseek,
1271 };
1272
1273 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1274                                   size_t count, loff_t *ppos)
1275 {
1276         struct inode * inode = file_inode(file);
1277         struct task_struct *task = get_proc_task(inode);
1278         ssize_t length;
1279         char tmpbuf[TMPBUFLEN];
1280
1281         if (!task)
1282                 return -ESRCH;
1283         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1284                                 audit_get_sessionid(task));
1285         put_task_struct(task);
1286         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1287 }
1288
1289 static const struct file_operations proc_sessionid_operations = {
1290         .read           = proc_sessionid_read,
1291         .llseek         = generic_file_llseek,
1292 };
1293 #endif
1294
1295 #ifdef CONFIG_FAULT_INJECTION
1296 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1297                                       size_t count, loff_t *ppos)
1298 {
1299         struct task_struct *task = get_proc_task(file_inode(file));
1300         char buffer[PROC_NUMBUF];
1301         size_t len;
1302         int make_it_fail;
1303
1304         if (!task)
1305                 return -ESRCH;
1306         make_it_fail = task->make_it_fail;
1307         put_task_struct(task);
1308
1309         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1310
1311         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1312 }
1313
1314 static ssize_t proc_fault_inject_write(struct file * file,
1315                         const char __user * buf, size_t count, loff_t *ppos)
1316 {
1317         struct task_struct *task;
1318         char buffer[PROC_NUMBUF];
1319         int make_it_fail;
1320         int rv;
1321
1322         if (!capable(CAP_SYS_RESOURCE))
1323                 return -EPERM;
1324         memset(buffer, 0, sizeof(buffer));
1325         if (count > sizeof(buffer) - 1)
1326                 count = sizeof(buffer) - 1;
1327         if (copy_from_user(buffer, buf, count))
1328                 return -EFAULT;
1329         rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1330         if (rv < 0)
1331                 return rv;
1332         if (make_it_fail < 0 || make_it_fail > 1)
1333                 return -EINVAL;
1334
1335         task = get_proc_task(file_inode(file));
1336         if (!task)
1337                 return -ESRCH;
1338         task->make_it_fail = make_it_fail;
1339         put_task_struct(task);
1340
1341         return count;
1342 }
1343
1344 static const struct file_operations proc_fault_inject_operations = {
1345         .read           = proc_fault_inject_read,
1346         .write          = proc_fault_inject_write,
1347         .llseek         = generic_file_llseek,
1348 };
1349
1350 static ssize_t proc_fail_nth_write(struct file *file, const char __user *buf,
1351                                    size_t count, loff_t *ppos)
1352 {
1353         struct task_struct *task;
1354         int err;
1355         unsigned int n;
1356
1357         err = kstrtouint_from_user(buf, count, 0, &n);
1358         if (err)
1359                 return err;
1360
1361         task = get_proc_task(file_inode(file));
1362         if (!task)
1363                 return -ESRCH;
1364         task->fail_nth = n;
1365         put_task_struct(task);
1366
1367         return count;
1368 }
1369
1370 static ssize_t proc_fail_nth_read(struct file *file, char __user *buf,
1371                                   size_t count, loff_t *ppos)
1372 {
1373         struct task_struct *task;
1374         char numbuf[PROC_NUMBUF];
1375         ssize_t len;
1376
1377         task = get_proc_task(file_inode(file));
1378         if (!task)
1379                 return -ESRCH;
1380         len = snprintf(numbuf, sizeof(numbuf), "%u\n", task->fail_nth);
1381         put_task_struct(task);
1382         return simple_read_from_buffer(buf, count, ppos, numbuf, len);
1383 }
1384
1385 static const struct file_operations proc_fail_nth_operations = {
1386         .read           = proc_fail_nth_read,
1387         .write          = proc_fail_nth_write,
1388 };
1389 #endif
1390
1391
1392 #ifdef CONFIG_SCHED_DEBUG
1393 /*
1394  * Print out various scheduling related per-task fields:
1395  */
1396 static int sched_show(struct seq_file *m, void *v)
1397 {
1398         struct inode *inode = m->private;
1399         struct pid_namespace *ns = proc_pid_ns(inode);
1400         struct task_struct *p;
1401
1402         p = get_proc_task(inode);
1403         if (!p)
1404                 return -ESRCH;
1405         proc_sched_show_task(p, ns, m);
1406
1407         put_task_struct(p);
1408
1409         return 0;
1410 }
1411
1412 static ssize_t
1413 sched_write(struct file *file, const char __user *buf,
1414             size_t count, loff_t *offset)
1415 {
1416         struct inode *inode = file_inode(file);
1417         struct task_struct *p;
1418
1419         p = get_proc_task(inode);
1420         if (!p)
1421                 return -ESRCH;
1422         proc_sched_set_task(p);
1423
1424         put_task_struct(p);
1425
1426         return count;
1427 }
1428
1429 static int sched_open(struct inode *inode, struct file *filp)
1430 {
1431         return single_open(filp, sched_show, inode);
1432 }
1433
1434 static const struct file_operations proc_pid_sched_operations = {
1435         .open           = sched_open,
1436         .read           = seq_read,
1437         .write          = sched_write,
1438         .llseek         = seq_lseek,
1439         .release        = single_release,
1440 };
1441
1442 #endif
1443
1444 #ifdef CONFIG_SCHED_AUTOGROUP
1445 /*
1446  * Print out autogroup related information:
1447  */
1448 static int sched_autogroup_show(struct seq_file *m, void *v)
1449 {
1450         struct inode *inode = m->private;
1451         struct task_struct *p;
1452
1453         p = get_proc_task(inode);
1454         if (!p)
1455                 return -ESRCH;
1456         proc_sched_autogroup_show_task(p, m);
1457
1458         put_task_struct(p);
1459
1460         return 0;
1461 }
1462
1463 static ssize_t
1464 sched_autogroup_write(struct file *file, const char __user *buf,
1465             size_t count, loff_t *offset)
1466 {
1467         struct inode *inode = file_inode(file);
1468         struct task_struct *p;
1469         char buffer[PROC_NUMBUF];
1470         int nice;
1471         int err;
1472
1473         memset(buffer, 0, sizeof(buffer));
1474         if (count > sizeof(buffer) - 1)
1475                 count = sizeof(buffer) - 1;
1476         if (copy_from_user(buffer, buf, count))
1477                 return -EFAULT;
1478
1479         err = kstrtoint(strstrip(buffer), 0, &nice);
1480         if (err < 0)
1481                 return err;
1482
1483         p = get_proc_task(inode);
1484         if (!p)
1485                 return -ESRCH;
1486
1487         err = proc_sched_autogroup_set_nice(p, nice);
1488         if (err)
1489                 count = err;
1490
1491         put_task_struct(p);
1492
1493         return count;
1494 }
1495
1496 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1497 {
1498         int ret;
1499
1500         ret = single_open(filp, sched_autogroup_show, NULL);
1501         if (!ret) {
1502                 struct seq_file *m = filp->private_data;
1503
1504                 m->private = inode;
1505         }
1506         return ret;
1507 }
1508
1509 static const struct file_operations proc_pid_sched_autogroup_operations = {
1510         .open           = sched_autogroup_open,
1511         .read           = seq_read,
1512         .write          = sched_autogroup_write,
1513         .llseek         = seq_lseek,
1514         .release        = single_release,
1515 };
1516
1517 #endif /* CONFIG_SCHED_AUTOGROUP */
1518
1519 static ssize_t comm_write(struct file *file, const char __user *buf,
1520                                 size_t count, loff_t *offset)
1521 {
1522         struct inode *inode = file_inode(file);
1523         struct task_struct *p;
1524         char buffer[TASK_COMM_LEN];
1525         const size_t maxlen = sizeof(buffer) - 1;
1526
1527         memset(buffer, 0, sizeof(buffer));
1528         if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1529                 return -EFAULT;
1530
1531         p = get_proc_task(inode);
1532         if (!p)
1533                 return -ESRCH;
1534
1535         if (same_thread_group(current, p))
1536                 set_task_comm(p, buffer);
1537         else
1538                 count = -EINVAL;
1539
1540         put_task_struct(p);
1541
1542         return count;
1543 }
1544
1545 static int comm_show(struct seq_file *m, void *v)
1546 {
1547         struct inode *inode = m->private;
1548         struct task_struct *p;
1549
1550         p = get_proc_task(inode);
1551         if (!p)
1552                 return -ESRCH;
1553
1554         proc_task_name(m, p, false);
1555         seq_putc(m, '\n');
1556
1557         put_task_struct(p);
1558
1559         return 0;
1560 }
1561
1562 static int comm_open(struct inode *inode, struct file *filp)
1563 {
1564         return single_open(filp, comm_show, inode);
1565 }
1566
1567 static const struct file_operations proc_pid_set_comm_operations = {
1568         .open           = comm_open,
1569         .read           = seq_read,
1570         .write          = comm_write,
1571         .llseek         = seq_lseek,
1572         .release        = single_release,
1573 };
1574
1575 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1576 {
1577         struct task_struct *task;
1578         struct file *exe_file;
1579
1580         task = get_proc_task(d_inode(dentry));
1581         if (!task)
1582                 return -ENOENT;
1583         exe_file = get_task_exe_file(task);
1584         put_task_struct(task);
1585         if (exe_file) {
1586                 *exe_path = exe_file->f_path;
1587                 path_get(&exe_file->f_path);
1588                 fput(exe_file);
1589                 return 0;
1590         } else
1591                 return -ENOENT;
1592 }
1593
1594 static const char *proc_pid_get_link(struct dentry *dentry,
1595                                      struct inode *inode,
1596                                      struct delayed_call *done)
1597 {
1598         struct path path;
1599         int error = -EACCES;
1600
1601         if (!dentry)
1602                 return ERR_PTR(-ECHILD);
1603
1604         /* Are we allowed to snoop on the tasks file descriptors? */
1605         if (!proc_fd_access_allowed(inode))
1606                 goto out;
1607
1608         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1609         if (error)
1610                 goto out;
1611
1612         nd_jump_link(&path);
1613         return NULL;
1614 out:
1615         return ERR_PTR(error);
1616 }
1617
1618 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1619 {
1620         char *tmp = (char *)__get_free_page(GFP_KERNEL);
1621         char *pathname;
1622         int len;
1623
1624         if (!tmp)
1625                 return -ENOMEM;
1626
1627         pathname = d_path(path, tmp, PAGE_SIZE);
1628         len = PTR_ERR(pathname);
1629         if (IS_ERR(pathname))
1630                 goto out;
1631         len = tmp + PAGE_SIZE - 1 - pathname;
1632
1633         if (len > buflen)
1634                 len = buflen;
1635         if (copy_to_user(buffer, pathname, len))
1636                 len = -EFAULT;
1637  out:
1638         free_page((unsigned long)tmp);
1639         return len;
1640 }
1641
1642 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1643 {
1644         int error = -EACCES;
1645         struct inode *inode = d_inode(dentry);
1646         struct path path;
1647
1648         /* Are we allowed to snoop on the tasks file descriptors? */
1649         if (!proc_fd_access_allowed(inode))
1650                 goto out;
1651
1652         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1653         if (error)
1654                 goto out;
1655
1656         error = do_proc_readlink(&path, buffer, buflen);
1657         path_put(&path);
1658 out:
1659         return error;
1660 }
1661
1662 const struct inode_operations proc_pid_link_inode_operations = {
1663         .readlink       = proc_pid_readlink,
1664         .get_link       = proc_pid_get_link,
1665         .setattr        = proc_setattr,
1666 };
1667
1668
1669 /* building an inode */
1670
1671 void task_dump_owner(struct task_struct *task, umode_t mode,
1672                      kuid_t *ruid, kgid_t *rgid)
1673 {
1674         /* Depending on the state of dumpable compute who should own a
1675          * proc file for a task.
1676          */
1677         const struct cred *cred;
1678         kuid_t uid;
1679         kgid_t gid;
1680
1681         if (unlikely(task->flags & PF_KTHREAD)) {
1682                 *ruid = GLOBAL_ROOT_UID;
1683                 *rgid = GLOBAL_ROOT_GID;
1684                 return;
1685         }
1686
1687         /* Default to the tasks effective ownership */
1688         rcu_read_lock();
1689         cred = __task_cred(task);
1690         uid = cred->euid;
1691         gid = cred->egid;
1692         rcu_read_unlock();
1693
1694         /*
1695          * Before the /proc/pid/status file was created the only way to read
1696          * the effective uid of a /process was to stat /proc/pid.  Reading
1697          * /proc/pid/status is slow enough that procps and other packages
1698          * kept stating /proc/pid.  To keep the rules in /proc simple I have
1699          * made this apply to all per process world readable and executable
1700          * directories.
1701          */
1702         if (mode != (S_IFDIR|S_IRUGO|S_IXUGO)) {
1703                 struct mm_struct *mm;
1704                 task_lock(task);
1705                 mm = task->mm;
1706                 /* Make non-dumpable tasks owned by some root */
1707                 if (mm) {
1708                         if (get_dumpable(mm) != SUID_DUMP_USER) {
1709                                 struct user_namespace *user_ns = mm->user_ns;
1710
1711                                 uid = make_kuid(user_ns, 0);
1712                                 if (!uid_valid(uid))
1713                                         uid = GLOBAL_ROOT_UID;
1714
1715                                 gid = make_kgid(user_ns, 0);
1716                                 if (!gid_valid(gid))
1717                                         gid = GLOBAL_ROOT_GID;
1718                         }
1719                 } else {
1720                         uid = GLOBAL_ROOT_UID;
1721                         gid = GLOBAL_ROOT_GID;
1722                 }
1723                 task_unlock(task);
1724         }
1725         *ruid = uid;
1726         *rgid = gid;
1727 }
1728
1729 struct inode *proc_pid_make_inode(struct super_block * sb,
1730                                   struct task_struct *task, umode_t mode)
1731 {
1732         struct inode * inode;
1733         struct proc_inode *ei;
1734
1735         /* We need a new inode */
1736
1737         inode = new_inode(sb);
1738         if (!inode)
1739                 goto out;
1740
1741         /* Common stuff */
1742         ei = PROC_I(inode);
1743         inode->i_mode = mode;
1744         inode->i_ino = get_next_ino();
1745         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1746         inode->i_op = &proc_def_inode_operations;
1747
1748         /*
1749          * grab the reference to task.
1750          */
1751         ei->pid = get_task_pid(task, PIDTYPE_PID);
1752         if (!ei->pid)
1753                 goto out_unlock;
1754
1755         task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1756         security_task_to_inode(task, inode);
1757
1758 out:
1759         return inode;
1760
1761 out_unlock:
1762         iput(inode);
1763         return NULL;
1764 }
1765
1766 int pid_getattr(const struct path *path, struct kstat *stat,
1767                 u32 request_mask, unsigned int query_flags)
1768 {
1769         struct inode *inode = d_inode(path->dentry);
1770         struct pid_namespace *pid = proc_pid_ns(inode);
1771         struct task_struct *task;
1772
1773         generic_fillattr(inode, stat);
1774
1775         stat->uid = GLOBAL_ROOT_UID;
1776         stat->gid = GLOBAL_ROOT_GID;
1777         rcu_read_lock();
1778         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1779         if (task) {
1780                 if (!has_pid_permissions(pid, task, HIDEPID_INVISIBLE)) {
1781                         rcu_read_unlock();
1782                         /*
1783                          * This doesn't prevent learning whether PID exists,
1784                          * it only makes getattr() consistent with readdir().
1785                          */
1786                         return -ENOENT;
1787                 }
1788                 task_dump_owner(task, inode->i_mode, &stat->uid, &stat->gid);
1789         }
1790         rcu_read_unlock();
1791         return 0;
1792 }
1793
1794 /* dentry stuff */
1795
1796 /*
1797  * Set <pid>/... inode ownership (can change due to setuid(), etc.)
1798  */
1799 void pid_update_inode(struct task_struct *task, struct inode *inode)
1800 {
1801         task_dump_owner(task, inode->i_mode, &inode->i_uid, &inode->i_gid);
1802
1803         inode->i_mode &= ~(S_ISUID | S_ISGID);
1804         security_task_to_inode(task, inode);
1805 }
1806
1807 /*
1808  * Rewrite the inode's ownerships here because the owning task may have
1809  * performed a setuid(), etc.
1810  *
1811  */
1812 static int pid_revalidate(struct dentry *dentry, unsigned int flags)
1813 {
1814         struct inode *inode;
1815         struct task_struct *task;
1816
1817         if (flags & LOOKUP_RCU)
1818                 return -ECHILD;
1819
1820         inode = d_inode(dentry);
1821         task = get_proc_task(inode);
1822
1823         if (task) {
1824                 pid_update_inode(task, inode);
1825                 put_task_struct(task);
1826                 return 1;
1827         }
1828         return 0;
1829 }
1830
1831 static inline bool proc_inode_is_dead(struct inode *inode)
1832 {
1833         return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1834 }
1835
1836 int pid_delete_dentry(const struct dentry *dentry)
1837 {
1838         /* Is the task we represent dead?
1839          * If so, then don't put the dentry on the lru list,
1840          * kill it immediately.
1841          */
1842         return proc_inode_is_dead(d_inode(dentry));
1843 }
1844
1845 const struct dentry_operations pid_dentry_operations =
1846 {
1847         .d_revalidate   = pid_revalidate,
1848         .d_delete       = pid_delete_dentry,
1849 };
1850
1851 /* Lookups */
1852
1853 /*
1854  * Fill a directory entry.
1855  *
1856  * If possible create the dcache entry and derive our inode number and
1857  * file type from dcache entry.
1858  *
1859  * Since all of the proc inode numbers are dynamically generated, the inode
1860  * numbers do not exist until the inode is cache.  This means creating the
1861  * the dcache entry in readdir is necessary to keep the inode numbers
1862  * reported by readdir in sync with the inode numbers reported
1863  * by stat.
1864  */
1865 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1866         const char *name, unsigned int len,
1867         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1868 {
1869         struct dentry *child, *dir = file->f_path.dentry;
1870         struct qstr qname = QSTR_INIT(name, len);
1871         struct inode *inode;
1872         unsigned type = DT_UNKNOWN;
1873         ino_t ino = 1;
1874
1875         child = d_hash_and_lookup(dir, &qname);
1876         if (!child) {
1877                 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1878                 child = d_alloc_parallel(dir, &qname, &wq);
1879                 if (IS_ERR(child))
1880                         goto end_instantiate;
1881                 if (d_in_lookup(child)) {
1882                         struct dentry *res;
1883                         res = instantiate(child, task, ptr);
1884                         d_lookup_done(child);
1885                         if (unlikely(res)) {
1886                                 dput(child);
1887                                 child = res;
1888                                 if (IS_ERR(child))
1889                                         goto end_instantiate;
1890                         }
1891                 }
1892         }
1893         inode = d_inode(child);
1894         ino = inode->i_ino;
1895         type = inode->i_mode >> 12;
1896         dput(child);
1897 end_instantiate:
1898         return dir_emit(ctx, name, len, ino, type);
1899 }
1900
1901 /*
1902  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1903  * which represent vma start and end addresses.
1904  */
1905 static int dname_to_vma_addr(struct dentry *dentry,
1906                              unsigned long *start, unsigned long *end)
1907 {
1908         const char *str = dentry->d_name.name;
1909         unsigned long long sval, eval;
1910         unsigned int len;
1911
1912         if (str[0] == '0' && str[1] != '-')
1913                 return -EINVAL;
1914         len = _parse_integer(str, 16, &sval);
1915         if (len & KSTRTOX_OVERFLOW)
1916                 return -EINVAL;
1917         if (sval != (unsigned long)sval)
1918                 return -EINVAL;
1919         str += len;
1920
1921         if (*str != '-')
1922                 return -EINVAL;
1923         str++;
1924
1925         if (str[0] == '0' && str[1])
1926                 return -EINVAL;
1927         len = _parse_integer(str, 16, &eval);
1928         if (len & KSTRTOX_OVERFLOW)
1929                 return -EINVAL;
1930         if (eval != (unsigned long)eval)
1931                 return -EINVAL;
1932         str += len;
1933
1934         if (*str != '\0')
1935                 return -EINVAL;
1936
1937         *start = sval;
1938         *end = eval;
1939
1940         return 0;
1941 }
1942
1943 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1944 {
1945         unsigned long vm_start, vm_end;
1946         bool exact_vma_exists = false;
1947         struct mm_struct *mm = NULL;
1948         struct task_struct *task;
1949         struct inode *inode;
1950         int status = 0;
1951
1952         if (flags & LOOKUP_RCU)
1953                 return -ECHILD;
1954
1955         inode = d_inode(dentry);
1956         task = get_proc_task(inode);
1957         if (!task)
1958                 goto out_notask;
1959
1960         mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1961         if (IS_ERR_OR_NULL(mm))
1962                 goto out;
1963
1964         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1965                 down_read(&mm->mmap_sem);
1966                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1967                 up_read(&mm->mmap_sem);
1968         }
1969
1970         mmput(mm);
1971
1972         if (exact_vma_exists) {
1973                 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1974
1975                 security_task_to_inode(task, inode);
1976                 status = 1;
1977         }
1978
1979 out:
1980         put_task_struct(task);
1981
1982 out_notask:
1983         return status;
1984 }
1985
1986 static const struct dentry_operations tid_map_files_dentry_operations = {
1987         .d_revalidate   = map_files_d_revalidate,
1988         .d_delete       = pid_delete_dentry,
1989 };
1990
1991 static int map_files_get_link(struct dentry *dentry, struct path *path)
1992 {
1993         unsigned long vm_start, vm_end;
1994         struct vm_area_struct *vma;
1995         struct task_struct *task;
1996         struct mm_struct *mm;
1997         int rc;
1998
1999         rc = -ENOENT;
2000         task = get_proc_task(d_inode(dentry));
2001         if (!task)
2002                 goto out;
2003
2004         mm = get_task_mm(task);
2005         put_task_struct(task);
2006         if (!mm)
2007                 goto out;
2008
2009         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2010         if (rc)
2011                 goto out_mmput;
2012
2013         rc = -ENOENT;
2014         down_read(&mm->mmap_sem);
2015         vma = find_exact_vma(mm, vm_start, vm_end);
2016         if (vma && vma->vm_file) {
2017                 *path = vma->vm_file->f_path;
2018                 path_get(path);
2019                 rc = 0;
2020         }
2021         up_read(&mm->mmap_sem);
2022
2023 out_mmput:
2024         mmput(mm);
2025 out:
2026         return rc;
2027 }
2028
2029 struct map_files_info {
2030         unsigned long   start;
2031         unsigned long   end;
2032         fmode_t         mode;
2033 };
2034
2035 /*
2036  * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
2037  * symlinks may be used to bypass permissions on ancestor directories in the
2038  * path to the file in question.
2039  */
2040 static const char *
2041 proc_map_files_get_link(struct dentry *dentry,
2042                         struct inode *inode,
2043                         struct delayed_call *done)
2044 {
2045         if (!capable(CAP_SYS_ADMIN))
2046                 return ERR_PTR(-EPERM);
2047
2048         return proc_pid_get_link(dentry, inode, done);
2049 }
2050
2051 /*
2052  * Identical to proc_pid_link_inode_operations except for get_link()
2053  */
2054 static const struct inode_operations proc_map_files_link_inode_operations = {
2055         .readlink       = proc_pid_readlink,
2056         .get_link       = proc_map_files_get_link,
2057         .setattr        = proc_setattr,
2058 };
2059
2060 static struct dentry *
2061 proc_map_files_instantiate(struct dentry *dentry,
2062                            struct task_struct *task, const void *ptr)
2063 {
2064         fmode_t mode = (fmode_t)(unsigned long)ptr;
2065         struct proc_inode *ei;
2066         struct inode *inode;
2067
2068         inode = proc_pid_make_inode(dentry->d_sb, task, S_IFLNK |
2069                                     ((mode & FMODE_READ ) ? S_IRUSR : 0) |
2070                                     ((mode & FMODE_WRITE) ? S_IWUSR : 0));
2071         if (!inode)
2072                 return ERR_PTR(-ENOENT);
2073
2074         ei = PROC_I(inode);
2075         ei->op.proc_get_link = map_files_get_link;
2076
2077         inode->i_op = &proc_map_files_link_inode_operations;
2078         inode->i_size = 64;
2079
2080         d_set_d_op(dentry, &tid_map_files_dentry_operations);
2081         return d_splice_alias(inode, dentry);
2082 }
2083
2084 static struct dentry *proc_map_files_lookup(struct inode *dir,
2085                 struct dentry *dentry, unsigned int flags)
2086 {
2087         unsigned long vm_start, vm_end;
2088         struct vm_area_struct *vma;
2089         struct task_struct *task;
2090         struct dentry *result;
2091         struct mm_struct *mm;
2092
2093         result = ERR_PTR(-ENOENT);
2094         task = get_proc_task(dir);
2095         if (!task)
2096                 goto out;
2097
2098         result = ERR_PTR(-EACCES);
2099         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2100                 goto out_put_task;
2101
2102         result = ERR_PTR(-ENOENT);
2103         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2104                 goto out_put_task;
2105
2106         mm = get_task_mm(task);
2107         if (!mm)
2108                 goto out_put_task;
2109
2110         down_read(&mm->mmap_sem);
2111         vma = find_exact_vma(mm, vm_start, vm_end);
2112         if (!vma)
2113                 goto out_no_vma;
2114
2115         if (vma->vm_file)
2116                 result = proc_map_files_instantiate(dentry, task,
2117                                 (void *)(unsigned long)vma->vm_file->f_mode);
2118
2119 out_no_vma:
2120         up_read(&mm->mmap_sem);
2121         mmput(mm);
2122 out_put_task:
2123         put_task_struct(task);
2124 out:
2125         return result;
2126 }
2127
2128 static const struct inode_operations proc_map_files_inode_operations = {
2129         .lookup         = proc_map_files_lookup,
2130         .permission     = proc_fd_permission,
2131         .setattr        = proc_setattr,
2132 };
2133
2134 static int
2135 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2136 {
2137         struct vm_area_struct *vma;
2138         struct task_struct *task;
2139         struct mm_struct *mm;
2140         unsigned long nr_files, pos, i;
2141         struct flex_array *fa = NULL;
2142         struct map_files_info info;
2143         struct map_files_info *p;
2144         int ret;
2145
2146         ret = -ENOENT;
2147         task = get_proc_task(file_inode(file));
2148         if (!task)
2149                 goto out;
2150
2151         ret = -EACCES;
2152         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2153                 goto out_put_task;
2154
2155         ret = 0;
2156         if (!dir_emit_dots(file, ctx))
2157                 goto out_put_task;
2158
2159         mm = get_task_mm(task);
2160         if (!mm)
2161                 goto out_put_task;
2162         down_read(&mm->mmap_sem);
2163
2164         nr_files = 0;
2165
2166         /*
2167          * We need two passes here:
2168          *
2169          *  1) Collect vmas of mapped files with mmap_sem taken
2170          *  2) Release mmap_sem and instantiate entries
2171          *
2172          * otherwise we get lockdep complained, since filldir()
2173          * routine might require mmap_sem taken in might_fault().
2174          */
2175
2176         for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2177                 if (vma->vm_file && ++pos > ctx->pos)
2178                         nr_files++;
2179         }
2180
2181         if (nr_files) {
2182                 fa = flex_array_alloc(sizeof(info), nr_files,
2183                                         GFP_KERNEL);
2184                 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2185                                                 GFP_KERNEL)) {
2186                         ret = -ENOMEM;
2187                         if (fa)
2188                                 flex_array_free(fa);
2189                         up_read(&mm->mmap_sem);
2190                         mmput(mm);
2191                         goto out_put_task;
2192                 }
2193                 for (i = 0, vma = mm->mmap, pos = 2; vma;
2194                                 vma = vma->vm_next) {
2195                         if (!vma->vm_file)
2196                                 continue;
2197                         if (++pos <= ctx->pos)
2198                                 continue;
2199
2200                         info.start = vma->vm_start;
2201                         info.end = vma->vm_end;
2202                         info.mode = vma->vm_file->f_mode;
2203                         if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2204                                 BUG();
2205                 }
2206         }
2207         up_read(&mm->mmap_sem);
2208         mmput(mm);
2209
2210         for (i = 0; i < nr_files; i++) {
2211                 char buf[4 * sizeof(long) + 2]; /* max: %lx-%lx\0 */
2212                 unsigned int len;
2213
2214                 p = flex_array_get(fa, i);
2215                 len = snprintf(buf, sizeof(buf), "%lx-%lx", p->start, p->end);
2216                 if (!proc_fill_cache(file, ctx,
2217                                       buf, len,
2218                                       proc_map_files_instantiate,
2219                                       task,
2220                                       (void *)(unsigned long)p->mode))
2221                         break;
2222                 ctx->pos++;
2223         }
2224         if (fa)
2225                 flex_array_free(fa);
2226
2227 out_put_task:
2228         put_task_struct(task);
2229 out:
2230         return ret;
2231 }
2232
2233 static const struct file_operations proc_map_files_operations = {
2234         .read           = generic_read_dir,
2235         .iterate_shared = proc_map_files_readdir,
2236         .llseek         = generic_file_llseek,
2237 };
2238
2239 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2240 struct timers_private {
2241         struct pid *pid;
2242         struct task_struct *task;
2243         struct sighand_struct *sighand;
2244         struct pid_namespace *ns;
2245         unsigned long flags;
2246 };
2247
2248 static void *timers_start(struct seq_file *m, loff_t *pos)
2249 {
2250         struct timers_private *tp = m->private;
2251
2252         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2253         if (!tp->task)
2254                 return ERR_PTR(-ESRCH);
2255
2256         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2257         if (!tp->sighand)
2258                 return ERR_PTR(-ESRCH);
2259
2260         return seq_list_start(&tp->task->signal->posix_timers, *pos);
2261 }
2262
2263 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2264 {
2265         struct timers_private *tp = m->private;
2266         return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2267 }
2268
2269 static void timers_stop(struct seq_file *m, void *v)
2270 {
2271         struct timers_private *tp = m->private;
2272
2273         if (tp->sighand) {
2274                 unlock_task_sighand(tp->task, &tp->flags);
2275                 tp->sighand = NULL;
2276         }
2277
2278         if (tp->task) {
2279                 put_task_struct(tp->task);
2280                 tp->task = NULL;
2281         }
2282 }
2283
2284 static int show_timer(struct seq_file *m, void *v)
2285 {
2286         struct k_itimer *timer;
2287         struct timers_private *tp = m->private;
2288         int notify;
2289         static const char * const nstr[] = {
2290                 [SIGEV_SIGNAL] = "signal",
2291                 [SIGEV_NONE] = "none",
2292                 [SIGEV_THREAD] = "thread",
2293         };
2294
2295         timer = list_entry((struct list_head *)v, struct k_itimer, list);
2296         notify = timer->it_sigev_notify;
2297
2298         seq_printf(m, "ID: %d\n", timer->it_id);
2299         seq_printf(m, "signal: %d/%px\n",
2300                    timer->sigq->info.si_signo,
2301                    timer->sigq->info.si_value.sival_ptr);
2302         seq_printf(m, "notify: %s/%s.%d\n",
2303                    nstr[notify & ~SIGEV_THREAD_ID],
2304                    (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2305                    pid_nr_ns(timer->it_pid, tp->ns));
2306         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2307
2308         return 0;
2309 }
2310
2311 static const struct seq_operations proc_timers_seq_ops = {
2312         .start  = timers_start,
2313         .next   = timers_next,
2314         .stop   = timers_stop,
2315         .show   = show_timer,
2316 };
2317
2318 static int proc_timers_open(struct inode *inode, struct file *file)
2319 {
2320         struct timers_private *tp;
2321
2322         tp = __seq_open_private(file, &proc_timers_seq_ops,
2323                         sizeof(struct timers_private));
2324         if (!tp)
2325                 return -ENOMEM;
2326
2327         tp->pid = proc_pid(inode);
2328         tp->ns = proc_pid_ns(inode);
2329         return 0;
2330 }
2331
2332 static const struct file_operations proc_timers_operations = {
2333         .open           = proc_timers_open,
2334         .read           = seq_read,
2335         .llseek         = seq_lseek,
2336         .release        = seq_release_private,
2337 };
2338 #endif
2339
2340 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2341                                         size_t count, loff_t *offset)
2342 {
2343         struct inode *inode = file_inode(file);
2344         struct task_struct *p;
2345         u64 slack_ns;
2346         int err;
2347
2348         err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2349         if (err < 0)
2350                 return err;
2351
2352         p = get_proc_task(inode);
2353         if (!p)
2354                 return -ESRCH;
2355
2356         if (p != current) {
2357                 rcu_read_lock();
2358                 if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
2359                         rcu_read_unlock();
2360                         count = -EPERM;
2361                         goto out;
2362                 }
2363                 rcu_read_unlock();
2364
2365                 err = security_task_setscheduler(p);
2366                 if (err) {
2367                         count = err;
2368                         goto out;
2369                 }
2370         }
2371
2372         task_lock(p);
2373         if (slack_ns == 0)
2374                 p->timer_slack_ns = p->default_timer_slack_ns;
2375         else
2376                 p->timer_slack_ns = slack_ns;
2377         task_unlock(p);
2378
2379 out:
2380         put_task_struct(p);
2381
2382         return count;
2383 }
2384
2385 static int timerslack_ns_show(struct seq_file *m, void *v)
2386 {
2387         struct inode *inode = m->private;
2388         struct task_struct *p;
2389         int err = 0;
2390
2391         p = get_proc_task(inode);
2392         if (!p)
2393                 return -ESRCH;
2394
2395         if (p != current) {
2396                 rcu_read_lock();
2397                 if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
2398                         rcu_read_unlock();
2399                         err = -EPERM;
2400                         goto out;
2401                 }
2402                 rcu_read_unlock();
2403
2404                 err = security_task_getscheduler(p);
2405                 if (err)
2406                         goto out;
2407         }
2408
2409         task_lock(p);
2410         seq_printf(m, "%llu\n", p->timer_slack_ns);
2411         task_unlock(p);
2412
2413 out:
2414         put_task_struct(p);
2415
2416         return err;
2417 }
2418
2419 static int timerslack_ns_open(struct inode *inode, struct file *filp)
2420 {
2421         return single_open(filp, timerslack_ns_show, inode);
2422 }
2423
2424 static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2425         .open           = timerslack_ns_open,
2426         .read           = seq_read,
2427         .write          = timerslack_ns_write,
2428         .llseek         = seq_lseek,
2429         .release        = single_release,
2430 };
2431
2432 static struct dentry *proc_pident_instantiate(struct dentry *dentry,
2433         struct task_struct *task, const void *ptr)
2434 {
2435         const struct pid_entry *p = ptr;
2436         struct inode *inode;
2437         struct proc_inode *ei;
2438
2439         inode = proc_pid_make_inode(dentry->d_sb, task, p->mode);
2440         if (!inode)
2441                 return ERR_PTR(-ENOENT);
2442
2443         ei = PROC_I(inode);
2444         if (S_ISDIR(inode->i_mode))
2445                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2446         if (p->iop)
2447                 inode->i_op = p->iop;
2448         if (p->fop)
2449                 inode->i_fop = p->fop;
2450         ei->op = p->op;
2451         pid_update_inode(task, inode);
2452         d_set_d_op(dentry, &pid_dentry_operations);
2453         return d_splice_alias(inode, dentry);
2454 }
2455
2456 static struct dentry *proc_pident_lookup(struct inode *dir,
2457                                          struct dentry *dentry,
2458                                          const struct pid_entry *ents,
2459                                          unsigned int nents)
2460 {
2461         struct task_struct *task = get_proc_task(dir);
2462         const struct pid_entry *p, *last;
2463         struct dentry *res = ERR_PTR(-ENOENT);
2464
2465         if (!task)
2466                 goto out_no_task;
2467
2468         /*
2469          * Yes, it does not scale. And it should not. Don't add
2470          * new entries into /proc/<tgid>/ without very good reasons.
2471          */
2472         last = &ents[nents];
2473         for (p = ents; p < last; p++) {
2474                 if (p->len != dentry->d_name.len)
2475                         continue;
2476                 if (!memcmp(dentry->d_name.name, p->name, p->len)) {
2477                         res = proc_pident_instantiate(dentry, task, p);
2478                         break;
2479                 }
2480         }
2481         put_task_struct(task);
2482 out_no_task:
2483         return res;
2484 }
2485
2486 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2487                 const struct pid_entry *ents, unsigned int nents)
2488 {
2489         struct task_struct *task = get_proc_task(file_inode(file));
2490         const struct pid_entry *p;
2491
2492         if (!task)
2493                 return -ENOENT;
2494
2495         if (!dir_emit_dots(file, ctx))
2496                 goto out;
2497
2498         if (ctx->pos >= nents + 2)
2499                 goto out;
2500
2501         for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2502                 if (!proc_fill_cache(file, ctx, p->name, p->len,
2503                                 proc_pident_instantiate, task, p))
2504                         break;
2505                 ctx->pos++;
2506         }
2507 out:
2508         put_task_struct(task);
2509         return 0;
2510 }
2511
2512 #ifdef CONFIG_SECURITY
2513 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2514                                   size_t count, loff_t *ppos)
2515 {
2516         struct inode * inode = file_inode(file);
2517         char *p = NULL;
2518         ssize_t length;
2519         struct task_struct *task = get_proc_task(inode);
2520
2521         if (!task)
2522                 return -ESRCH;
2523
2524         length = security_getprocattr(task,
2525                                       (char*)file->f_path.dentry->d_name.name,
2526                                       &p);
2527         put_task_struct(task);
2528         if (length > 0)
2529                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2530         kfree(p);
2531         return length;
2532 }
2533
2534 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2535                                    size_t count, loff_t *ppos)
2536 {
2537         struct inode * inode = file_inode(file);
2538         struct task_struct *task;
2539         void *page;
2540         int rv;
2541
2542         rcu_read_lock();
2543         task = pid_task(proc_pid(inode), PIDTYPE_PID);
2544         if (!task) {
2545                 rcu_read_unlock();
2546                 return -ESRCH;
2547         }
2548         /* A task may only write its own attributes. */
2549         if (current != task) {
2550                 rcu_read_unlock();
2551                 return -EACCES;
2552         }
2553         rcu_read_unlock();
2554
2555         if (count > PAGE_SIZE)
2556                 count = PAGE_SIZE;
2557
2558         /* No partial writes. */
2559         if (*ppos != 0)
2560                 return -EINVAL;
2561
2562         page = memdup_user(buf, count);
2563         if (IS_ERR(page)) {
2564                 rv = PTR_ERR(page);
2565                 goto out;
2566         }
2567
2568         /* Guard against adverse ptrace interaction */
2569         rv = mutex_lock_interruptible(&current->signal->cred_guard_mutex);
2570         if (rv < 0)
2571                 goto out_free;
2572
2573         rv = security_setprocattr(file->f_path.dentry->d_name.name, page, count);
2574         mutex_unlock(&current->signal->cred_guard_mutex);
2575 out_free:
2576         kfree(page);
2577 out:
2578         return rv;
2579 }
2580
2581 static const struct file_operations proc_pid_attr_operations = {
2582         .read           = proc_pid_attr_read,
2583         .write          = proc_pid_attr_write,
2584         .llseek         = generic_file_llseek,
2585 };
2586
2587 static const struct pid_entry attr_dir_stuff[] = {
2588         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2589         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2590         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2591         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2592         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2593         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2594 };
2595
2596 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2597 {
2598         return proc_pident_readdir(file, ctx,
2599                                    attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2600 }
2601
2602 static const struct file_operations proc_attr_dir_operations = {
2603         .read           = generic_read_dir,
2604         .iterate_shared = proc_attr_dir_readdir,
2605         .llseek         = generic_file_llseek,
2606 };
2607
2608 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2609                                 struct dentry *dentry, unsigned int flags)
2610 {
2611         return proc_pident_lookup(dir, dentry,
2612                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2613 }
2614
2615 static const struct inode_operations proc_attr_dir_inode_operations = {
2616         .lookup         = proc_attr_dir_lookup,
2617         .getattr        = pid_getattr,
2618         .setattr        = proc_setattr,
2619 };
2620
2621 #endif
2622
2623 #ifdef CONFIG_ELF_CORE
2624 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2625                                          size_t count, loff_t *ppos)
2626 {
2627         struct task_struct *task = get_proc_task(file_inode(file));
2628         struct mm_struct *mm;
2629         char buffer[PROC_NUMBUF];
2630         size_t len;
2631         int ret;
2632
2633         if (!task)
2634                 return -ESRCH;
2635
2636         ret = 0;
2637         mm = get_task_mm(task);
2638         if (mm) {
2639                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2640                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2641                                 MMF_DUMP_FILTER_SHIFT));
2642                 mmput(mm);
2643                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2644         }
2645
2646         put_task_struct(task);
2647
2648         return ret;
2649 }
2650
2651 static ssize_t proc_coredump_filter_write(struct file *file,
2652                                           const char __user *buf,
2653                                           size_t count,
2654                                           loff_t *ppos)
2655 {
2656         struct task_struct *task;
2657         struct mm_struct *mm;
2658         unsigned int val;
2659         int ret;
2660         int i;
2661         unsigned long mask;
2662
2663         ret = kstrtouint_from_user(buf, count, 0, &val);
2664         if (ret < 0)
2665                 return ret;
2666
2667         ret = -ESRCH;
2668         task = get_proc_task(file_inode(file));
2669         if (!task)
2670                 goto out_no_task;
2671
2672         mm = get_task_mm(task);
2673         if (!mm)
2674                 goto out_no_mm;
2675         ret = 0;
2676
2677         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2678                 if (val & mask)
2679                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2680                 else
2681                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2682         }
2683
2684         mmput(mm);
2685  out_no_mm:
2686         put_task_struct(task);
2687  out_no_task:
2688         if (ret < 0)
2689                 return ret;
2690         return count;
2691 }
2692
2693 static const struct file_operations proc_coredump_filter_operations = {
2694         .read           = proc_coredump_filter_read,
2695         .write          = proc_coredump_filter_write,
2696         .llseek         = generic_file_llseek,
2697 };
2698 #endif
2699
2700 #ifdef CONFIG_TASK_IO_ACCOUNTING
2701 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2702 {
2703         struct task_io_accounting acct = task->ioac;
2704         unsigned long flags;
2705         int result;
2706
2707         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2708         if (result)
2709                 return result;
2710
2711         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2712                 result = -EACCES;
2713                 goto out_unlock;
2714         }
2715
2716         if (whole && lock_task_sighand(task, &flags)) {
2717                 struct task_struct *t = task;
2718
2719                 task_io_accounting_add(&acct, &task->signal->ioac);
2720                 while_each_thread(task, t)
2721                         task_io_accounting_add(&acct, &t->ioac);
2722
2723                 unlock_task_sighand(task, &flags);
2724         }
2725         seq_printf(m,
2726                    "rchar: %llu\n"
2727                    "wchar: %llu\n"
2728                    "syscr: %llu\n"
2729                    "syscw: %llu\n"
2730                    "read_bytes: %llu\n"
2731                    "write_bytes: %llu\n"
2732                    "cancelled_write_bytes: %llu\n",
2733                    (unsigned long long)acct.rchar,
2734                    (unsigned long long)acct.wchar,
2735                    (unsigned long long)acct.syscr,
2736                    (unsigned long long)acct.syscw,
2737                    (unsigned long long)acct.read_bytes,
2738                    (unsigned long long)acct.write_bytes,
2739                    (unsigned long long)acct.cancelled_write_bytes);
2740         result = 0;
2741
2742 out_unlock:
2743         mutex_unlock(&task->signal->cred_guard_mutex);
2744         return result;
2745 }
2746
2747 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2748                                   struct pid *pid, struct task_struct *task)
2749 {
2750         return do_io_accounting(task, m, 0);
2751 }
2752
2753 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2754                                    struct pid *pid, struct task_struct *task)
2755 {
2756         return do_io_accounting(task, m, 1);
2757 }
2758 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2759
2760 #ifdef CONFIG_USER_NS
2761 static int proc_id_map_open(struct inode *inode, struct file *file,
2762         const struct seq_operations *seq_ops)
2763 {
2764         struct user_namespace *ns = NULL;
2765         struct task_struct *task;
2766         struct seq_file *seq;
2767         int ret = -EINVAL;
2768
2769         task = get_proc_task(inode);
2770         if (task) {
2771                 rcu_read_lock();
2772                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2773                 rcu_read_unlock();
2774                 put_task_struct(task);
2775         }
2776         if (!ns)
2777                 goto err;
2778
2779         ret = seq_open(file, seq_ops);
2780         if (ret)
2781                 goto err_put_ns;
2782
2783         seq = file->private_data;
2784         seq->private = ns;
2785
2786         return 0;
2787 err_put_ns:
2788         put_user_ns(ns);
2789 err:
2790         return ret;
2791 }
2792
2793 static int proc_id_map_release(struct inode *inode, struct file *file)
2794 {
2795         struct seq_file *seq = file->private_data;
2796         struct user_namespace *ns = seq->private;
2797         put_user_ns(ns);
2798         return seq_release(inode, file);
2799 }
2800
2801 static int proc_uid_map_open(struct inode *inode, struct file *file)
2802 {
2803         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2804 }
2805
2806 static int proc_gid_map_open(struct inode *inode, struct file *file)
2807 {
2808         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2809 }
2810
2811 static int proc_projid_map_open(struct inode *inode, struct file *file)
2812 {
2813         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2814 }
2815
2816 static const struct file_operations proc_uid_map_operations = {
2817         .open           = proc_uid_map_open,
2818         .write          = proc_uid_map_write,
2819         .read           = seq_read,
2820         .llseek         = seq_lseek,
2821         .release        = proc_id_map_release,
2822 };
2823
2824 static const struct file_operations proc_gid_map_operations = {
2825         .open           = proc_gid_map_open,
2826         .write          = proc_gid_map_write,
2827         .read           = seq_read,
2828         .llseek         = seq_lseek,
2829         .release        = proc_id_map_release,
2830 };
2831
2832 static const struct file_operations proc_projid_map_operations = {
2833         .open           = proc_projid_map_open,
2834         .write          = proc_projid_map_write,
2835         .read           = seq_read,
2836         .llseek         = seq_lseek,
2837         .release        = proc_id_map_release,
2838 };
2839
2840 static int proc_setgroups_open(struct inode *inode, struct file *file)
2841 {
2842         struct user_namespace *ns = NULL;
2843         struct task_struct *task;
2844         int ret;
2845
2846         ret = -ESRCH;
2847         task = get_proc_task(inode);
2848         if (task) {
2849                 rcu_read_lock();
2850                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2851                 rcu_read_unlock();
2852                 put_task_struct(task);
2853         }
2854         if (!ns)
2855                 goto err;
2856
2857         if (file->f_mode & FMODE_WRITE) {
2858                 ret = -EACCES;
2859                 if (!ns_capable(ns, CAP_SYS_ADMIN))
2860                         goto err_put_ns;
2861         }
2862
2863         ret = single_open(file, &proc_setgroups_show, ns);
2864         if (ret)
2865                 goto err_put_ns;
2866
2867         return 0;
2868 err_put_ns:
2869         put_user_ns(ns);
2870 err:
2871         return ret;
2872 }
2873
2874 static int proc_setgroups_release(struct inode *inode, struct file *file)
2875 {
2876         struct seq_file *seq = file->private_data;
2877         struct user_namespace *ns = seq->private;
2878         int ret = single_release(inode, file);
2879         put_user_ns(ns);
2880         return ret;
2881 }
2882
2883 static const struct file_operations proc_setgroups_operations = {
2884         .open           = proc_setgroups_open,
2885         .write          = proc_setgroups_write,
2886         .read           = seq_read,
2887         .llseek         = seq_lseek,
2888         .release        = proc_setgroups_release,
2889 };
2890 #endif /* CONFIG_USER_NS */
2891
2892 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2893                                 struct pid *pid, struct task_struct *task)
2894 {
2895         int err = lock_trace(task);
2896         if (!err) {
2897                 seq_printf(m, "%08x\n", task->personality);
2898                 unlock_trace(task);
2899         }
2900         return err;
2901 }
2902
2903 #ifdef CONFIG_LIVEPATCH
2904 static int proc_pid_patch_state(struct seq_file *m, struct pid_namespace *ns,
2905                                 struct pid *pid, struct task_struct *task)
2906 {
2907         seq_printf(m, "%d\n", task->patch_state);
2908         return 0;
2909 }
2910 #endif /* CONFIG_LIVEPATCH */
2911
2912 #ifdef CONFIG_STACKLEAK_METRICS
2913 static int proc_stack_depth(struct seq_file *m, struct pid_namespace *ns,
2914                                 struct pid *pid, struct task_struct *task)
2915 {
2916         unsigned long prev_depth = THREAD_SIZE -
2917                                 (task->prev_lowest_stack & (THREAD_SIZE - 1));
2918         unsigned long depth = THREAD_SIZE -
2919                                 (task->lowest_stack & (THREAD_SIZE - 1));
2920
2921         seq_printf(m, "previous stack depth: %lu\nstack depth: %lu\n",
2922                                                         prev_depth, depth);
2923         return 0;
2924 }
2925 #endif /* CONFIG_STACKLEAK_METRICS */
2926
2927 /*
2928  * Thread groups
2929  */
2930 static const struct file_operations proc_task_operations;
2931 static const struct inode_operations proc_task_inode_operations;
2932
2933 static const struct pid_entry tgid_base_stuff[] = {
2934         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2935         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2936         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2937         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2938         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2939 #ifdef CONFIG_NET
2940         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2941 #endif
2942         REG("environ",    S_IRUSR, proc_environ_operations),
2943         REG("auxv",       S_IRUSR, proc_auxv_operations),
2944         ONE("status",     S_IRUGO, proc_pid_status),
2945         ONE("personality", S_IRUSR, proc_pid_personality),
2946         ONE("limits",     S_IRUGO, proc_pid_limits),
2947 #ifdef CONFIG_SCHED_DEBUG
2948         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2949 #endif
2950 #ifdef CONFIG_SCHED_AUTOGROUP
2951         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2952 #endif
2953         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2954 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2955         ONE("syscall",    S_IRUSR, proc_pid_syscall),
2956 #endif
2957         REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
2958         ONE("stat",       S_IRUGO, proc_tgid_stat),
2959         ONE("statm",      S_IRUGO, proc_pid_statm),
2960         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2961 #ifdef CONFIG_NUMA
2962         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2963 #endif
2964         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2965         LNK("cwd",        proc_cwd_link),
2966         LNK("root",       proc_root_link),
2967         LNK("exe",        proc_exe_link),
2968         REG("mounts",     S_IRUGO, proc_mounts_operations),
2969         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2970         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2971 #ifdef CONFIG_PROC_PAGE_MONITOR
2972         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2973         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2974         REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
2975         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2976 #endif
2977 #ifdef CONFIG_SECURITY
2978         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2979 #endif
2980 #ifdef CONFIG_KALLSYMS
2981         ONE("wchan",      S_IRUGO, proc_pid_wchan),
2982 #endif
2983 #ifdef CONFIG_STACKTRACE
2984         ONE("stack",      S_IRUSR, proc_pid_stack),
2985 #endif
2986 #ifdef CONFIG_SCHED_INFO
2987         ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
2988 #endif
2989 #ifdef CONFIG_LATENCYTOP
2990         REG("latency",  S_IRUGO, proc_lstats_operations),
2991 #endif
2992 #ifdef CONFIG_PROC_PID_CPUSET
2993         ONE("cpuset",     S_IRUGO, proc_cpuset_show),
2994 #endif
2995 #ifdef CONFIG_CGROUPS
2996         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
2997 #endif
2998         ONE("oom_score",  S_IRUGO, proc_oom_score),
2999         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3000         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3001 #ifdef CONFIG_AUDITSYSCALL
3002         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
3003         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3004 #endif
3005 #ifdef CONFIG_FAULT_INJECTION
3006         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3007         REG("fail-nth", 0644, proc_fail_nth_operations),
3008 #endif
3009 #ifdef CONFIG_ELF_CORE
3010         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3011 #endif
3012 #ifdef CONFIG_TASK_IO_ACCOUNTING
3013         ONE("io",       S_IRUSR, proc_tgid_io_accounting),
3014 #endif
3015 #ifdef CONFIG_USER_NS
3016         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3017         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3018         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3019         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3020 #endif
3021 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
3022         REG("timers",     S_IRUGO, proc_timers_operations),
3023 #endif
3024         REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
3025 #ifdef CONFIG_LIVEPATCH
3026         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
3027 #endif
3028 #ifdef CONFIG_STACKLEAK_METRICS
3029         ONE("stack_depth", S_IRUGO, proc_stack_depth),
3030 #endif
3031 };
3032
3033 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
3034 {
3035         return proc_pident_readdir(file, ctx,
3036                                    tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3037 }
3038
3039 static const struct file_operations proc_tgid_base_operations = {
3040         .read           = generic_read_dir,
3041         .iterate_shared = proc_tgid_base_readdir,
3042         .llseek         = generic_file_llseek,
3043 };
3044
3045 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3046 {
3047         return proc_pident_lookup(dir, dentry,
3048                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3049 }
3050
3051 static const struct inode_operations proc_tgid_base_inode_operations = {
3052         .lookup         = proc_tgid_base_lookup,
3053         .getattr        = pid_getattr,
3054         .setattr        = proc_setattr,
3055         .permission     = proc_pid_permission,
3056 };
3057
3058 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
3059 {
3060         struct dentry *dentry, *leader, *dir;
3061         char buf[10 + 1];
3062         struct qstr name;
3063
3064         name.name = buf;
3065         name.len = snprintf(buf, sizeof(buf), "%u", pid);
3066         /* no ->d_hash() rejects on procfs */
3067         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
3068         if (dentry) {
3069                 d_invalidate(dentry);
3070                 dput(dentry);
3071         }
3072
3073         if (pid == tgid)
3074                 return;
3075
3076         name.name = buf;
3077         name.len = snprintf(buf, sizeof(buf), "%u", tgid);
3078         leader = d_hash_and_lookup(mnt->mnt_root, &name);
3079         if (!leader)
3080                 goto out;
3081
3082         name.name = "task";
3083         name.len = strlen(name.name);
3084         dir = d_hash_and_lookup(leader, &name);
3085         if (!dir)
3086                 goto out_put_leader;
3087
3088         name.name = buf;
3089         name.len = snprintf(buf, sizeof(buf), "%u", pid);
3090         dentry = d_hash_and_lookup(dir, &name);
3091         if (dentry) {
3092                 d_invalidate(dentry);
3093                 dput(dentry);
3094         }
3095
3096         dput(dir);
3097 out_put_leader:
3098         dput(leader);
3099 out:
3100         return;
3101 }
3102
3103 /**
3104  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
3105  * @task: task that should be flushed.
3106  *
3107  * When flushing dentries from proc, one needs to flush them from global
3108  * proc (proc_mnt) and from all the namespaces' procs this task was seen
3109  * in. This call is supposed to do all of this job.
3110  *
3111  * Looks in the dcache for
3112  * /proc/@pid
3113  * /proc/@tgid/task/@pid
3114  * if either directory is present flushes it and all of it'ts children
3115  * from the dcache.
3116  *
3117  * It is safe and reasonable to cache /proc entries for a task until
3118  * that task exits.  After that they just clog up the dcache with
3119  * useless entries, possibly causing useful dcache entries to be
3120  * flushed instead.  This routine is proved to flush those useless
3121  * dcache entries at process exit time.
3122  *
3123  * NOTE: This routine is just an optimization so it does not guarantee
3124  *       that no dcache entries will exist at process exit time it
3125  *       just makes it very unlikely that any will persist.
3126  */
3127
3128 void proc_flush_task(struct task_struct *task)
3129 {
3130         int i;
3131         struct pid *pid, *tgid;
3132         struct upid *upid;
3133
3134         pid = task_pid(task);
3135         tgid = task_tgid(task);
3136
3137         for (i = 0; i <= pid->level; i++) {
3138                 upid = &pid->numbers[i];
3139                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3140                                         tgid->numbers[i].nr);
3141         }
3142 }
3143
3144 static struct dentry *proc_pid_instantiate(struct dentry * dentry,
3145                                    struct task_struct *task, const void *ptr)
3146 {
3147         struct inode *inode;
3148
3149         inode = proc_pid_make_inode(dentry->d_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3150         if (!inode)
3151                 return ERR_PTR(-ENOENT);
3152
3153         inode->i_op = &proc_tgid_base_inode_operations;
3154         inode->i_fop = &proc_tgid_base_operations;
3155         inode->i_flags|=S_IMMUTABLE;
3156
3157         set_nlink(inode, nlink_tgid);
3158         pid_update_inode(task, inode);
3159
3160         d_set_d_op(dentry, &pid_dentry_operations);
3161         return d_splice_alias(inode, dentry);
3162 }
3163
3164 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3165 {
3166         struct task_struct *task;
3167         unsigned tgid;
3168         struct pid_namespace *ns;
3169         struct dentry *result = ERR_PTR(-ENOENT);
3170
3171         tgid = name_to_int(&dentry->d_name);
3172         if (tgid == ~0U)
3173                 goto out;
3174
3175         ns = dentry->d_sb->s_fs_info;
3176         rcu_read_lock();
3177         task = find_task_by_pid_ns(tgid, ns);
3178         if (task)
3179                 get_task_struct(task);
3180         rcu_read_unlock();
3181         if (!task)
3182                 goto out;
3183
3184         result = proc_pid_instantiate(dentry, task, NULL);
3185         put_task_struct(task);
3186 out:
3187         return result;
3188 }
3189
3190 /*
3191  * Find the first task with tgid >= tgid
3192  *
3193  */
3194 struct tgid_iter {
3195         unsigned int tgid;
3196         struct task_struct *task;
3197 };
3198 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3199 {
3200         struct pid *pid;
3201
3202         if (iter.task)
3203                 put_task_struct(iter.task);
3204         rcu_read_lock();
3205 retry:
3206         iter.task = NULL;
3207         pid = find_ge_pid(iter.tgid, ns);
3208         if (pid) {
3209                 iter.tgid = pid_nr_ns(pid, ns);
3210                 iter.task = pid_task(pid, PIDTYPE_PID);
3211                 /* What we to know is if the pid we have find is the
3212                  * pid of a thread_group_leader.  Testing for task
3213                  * being a thread_group_leader is the obvious thing
3214                  * todo but there is a window when it fails, due to
3215                  * the pid transfer logic in de_thread.
3216                  *
3217                  * So we perform the straight forward test of seeing
3218                  * if the pid we have found is the pid of a thread
3219                  * group leader, and don't worry if the task we have
3220                  * found doesn't happen to be a thread group leader.
3221                  * As we don't care in the case of readdir.
3222                  */
3223                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3224                         iter.tgid += 1;
3225                         goto retry;
3226                 }
3227                 get_task_struct(iter.task);
3228         }
3229         rcu_read_unlock();
3230         return iter;
3231 }
3232
3233 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3234
3235 /* for the /proc/ directory itself, after non-process stuff has been done */
3236 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3237 {
3238         struct tgid_iter iter;
3239         struct pid_namespace *ns = proc_pid_ns(file_inode(file));
3240         loff_t pos = ctx->pos;
3241
3242         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3243                 return 0;
3244
3245         if (pos == TGID_OFFSET - 2) {
3246                 struct inode *inode = d_inode(ns->proc_self);
3247                 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3248                         return 0;
3249                 ctx->pos = pos = pos + 1;
3250         }
3251         if (pos == TGID_OFFSET - 1) {
3252                 struct inode *inode = d_inode(ns->proc_thread_self);
3253                 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3254                         return 0;
3255                 ctx->pos = pos = pos + 1;
3256         }
3257         iter.tgid = pos - TGID_OFFSET;
3258         iter.task = NULL;
3259         for (iter = next_tgid(ns, iter);
3260              iter.task;
3261              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3262                 char name[10 + 1];
3263                 unsigned int len;
3264
3265                 cond_resched();
3266                 if (!has_pid_permissions(ns, iter.task, HIDEPID_INVISIBLE))
3267                         continue;
3268
3269                 len = snprintf(name, sizeof(name), "%u", iter.tgid);
3270                 ctx->pos = iter.tgid + TGID_OFFSET;
3271                 if (!proc_fill_cache(file, ctx, name, len,
3272                                      proc_pid_instantiate, iter.task, NULL)) {
3273                         put_task_struct(iter.task);
3274                         return 0;
3275                 }
3276         }
3277         ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3278         return 0;
3279 }
3280
3281 /*
3282  * proc_tid_comm_permission is a special permission function exclusively
3283  * used for the node /proc/<pid>/task/<tid>/comm.
3284  * It bypasses generic permission checks in the case where a task of the same
3285  * task group attempts to access the node.
3286  * The rationale behind this is that glibc and bionic access this node for
3287  * cross thread naming (pthread_set/getname_np(!self)). However, if
3288  * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3289  * which locks out the cross thread naming implementation.
3290  * This function makes sure that the node is always accessible for members of
3291  * same thread group.
3292  */
3293 static int proc_tid_comm_permission(struct inode *inode, int mask)
3294 {
3295         bool is_same_tgroup;
3296         struct task_struct *task;
3297
3298         task = get_proc_task(inode);
3299         if (!task)
3300                 return -ESRCH;
3301         is_same_tgroup = same_thread_group(current, task);
3302         put_task_struct(task);
3303
3304         if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3305                 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3306                  * read or written by the members of the corresponding
3307                  * thread group.
3308                  */
3309                 return 0;
3310         }
3311
3312         return generic_permission(inode, mask);
3313 }
3314
3315 static const struct inode_operations proc_tid_comm_inode_operations = {
3316                 .permission = proc_tid_comm_permission,
3317 };
3318
3319 /*
3320  * Tasks
3321  */
3322 static const struct pid_entry tid_base_stuff[] = {
3323         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3324         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3325         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3326 #ifdef CONFIG_NET
3327         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3328 #endif
3329         REG("environ",   S_IRUSR, proc_environ_operations),
3330         REG("auxv",      S_IRUSR, proc_auxv_operations),
3331         ONE("status",    S_IRUGO, proc_pid_status),
3332         ONE("personality", S_IRUSR, proc_pid_personality),
3333         ONE("limits",    S_IRUGO, proc_pid_limits),
3334 #ifdef CONFIG_SCHED_DEBUG
3335         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3336 #endif
3337         NOD("comm",      S_IFREG|S_IRUGO|S_IWUSR,
3338                          &proc_tid_comm_inode_operations,
3339                          &proc_pid_set_comm_operations, {}),
3340 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3341         ONE("syscall",   S_IRUSR, proc_pid_syscall),
3342 #endif
3343         REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3344         ONE("stat",      S_IRUGO, proc_tid_stat),
3345         ONE("statm",     S_IRUGO, proc_pid_statm),
3346         REG("maps",      S_IRUGO, proc_pid_maps_operations),
3347 #ifdef CONFIG_PROC_CHILDREN
3348         REG("children",  S_IRUGO, proc_tid_children_operations),
3349 #endif
3350 #ifdef CONFIG_NUMA
3351         REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
3352 #endif
3353         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3354         LNK("cwd",       proc_cwd_link),
3355         LNK("root",      proc_root_link),
3356         LNK("exe",       proc_exe_link),
3357         REG("mounts",    S_IRUGO, proc_mounts_operations),
3358         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3359 #ifdef CONFIG_PROC_PAGE_MONITOR
3360         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3361         REG("smaps",     S_IRUGO, proc_pid_smaps_operations),
3362         REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3363         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3364 #endif
3365 #ifdef CONFIG_SECURITY
3366         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3367 #endif
3368 #ifdef CONFIG_KALLSYMS
3369         ONE("wchan",     S_IRUGO, proc_pid_wchan),
3370 #endif
3371 #ifdef CONFIG_STACKTRACE
3372         ONE("stack",      S_IRUSR, proc_pid_stack),
3373 #endif
3374 #ifdef CONFIG_SCHED_INFO
3375         ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3376 #endif
3377 #ifdef CONFIG_LATENCYTOP
3378         REG("latency",  S_IRUGO, proc_lstats_operations),
3379 #endif
3380 #ifdef CONFIG_PROC_PID_CPUSET
3381         ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3382 #endif
3383 #ifdef CONFIG_CGROUPS
3384         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3385 #endif
3386         ONE("oom_score", S_IRUGO, proc_oom_score),
3387         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3388         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3389 #ifdef CONFIG_AUDITSYSCALL
3390         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3391         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3392 #endif
3393 #ifdef CONFIG_FAULT_INJECTION
3394         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3395         REG("fail-nth", 0644, proc_fail_nth_operations),
3396 #endif
3397 #ifdef CONFIG_TASK_IO_ACCOUNTING
3398         ONE("io",       S_IRUSR, proc_tid_io_accounting),
3399 #endif
3400 #ifdef CONFIG_USER_NS
3401         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3402         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3403         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3404         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3405 #endif
3406 #ifdef CONFIG_LIVEPATCH
3407         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
3408 #endif
3409 };
3410
3411 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3412 {
3413         return proc_pident_readdir(file, ctx,
3414                                    tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3415 }
3416
3417 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3418 {
3419         return proc_pident_lookup(dir, dentry,
3420                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3421 }
3422
3423 static const struct file_operations proc_tid_base_operations = {
3424         .read           = generic_read_dir,
3425         .iterate_shared = proc_tid_base_readdir,
3426         .llseek         = generic_file_llseek,
3427 };
3428
3429 static const struct inode_operations proc_tid_base_inode_operations = {
3430         .lookup         = proc_tid_base_lookup,
3431         .getattr        = pid_getattr,
3432         .setattr        = proc_setattr,
3433 };
3434
3435 static struct dentry *proc_task_instantiate(struct dentry *dentry,
3436         struct task_struct *task, const void *ptr)
3437 {
3438         struct inode *inode;
3439         inode = proc_pid_make_inode(dentry->d_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3440         if (!inode)
3441                 return ERR_PTR(-ENOENT);
3442
3443         inode->i_op = &proc_tid_base_inode_operations;
3444         inode->i_fop = &proc_tid_base_operations;
3445         inode->i_flags |= S_IMMUTABLE;
3446
3447         set_nlink(inode, nlink_tid);
3448         pid_update_inode(task, inode);
3449
3450         d_set_d_op(dentry, &pid_dentry_operations);
3451         return d_splice_alias(inode, dentry);
3452 }
3453
3454 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3455 {
3456         struct task_struct *task;
3457         struct task_struct *leader = get_proc_task(dir);
3458         unsigned tid;
3459         struct pid_namespace *ns;
3460         struct dentry *result = ERR_PTR(-ENOENT);
3461
3462         if (!leader)
3463                 goto out_no_task;
3464
3465         tid = name_to_int(&dentry->d_name);
3466         if (tid == ~0U)
3467                 goto out;
3468
3469         ns = dentry->d_sb->s_fs_info;
3470         rcu_read_lock();
3471         task = find_task_by_pid_ns(tid, ns);
3472         if (task)
3473                 get_task_struct(task);
3474         rcu_read_unlock();
3475         if (!task)
3476                 goto out;
3477         if (!same_thread_group(leader, task))
3478                 goto out_drop_task;
3479
3480         result = proc_task_instantiate(dentry, task, NULL);
3481 out_drop_task:
3482         put_task_struct(task);
3483 out:
3484         put_task_struct(leader);
3485 out_no_task:
3486         return result;
3487 }
3488
3489 /*
3490  * Find the first tid of a thread group to return to user space.
3491  *
3492  * Usually this is just the thread group leader, but if the users
3493  * buffer was too small or there was a seek into the middle of the
3494  * directory we have more work todo.
3495  *
3496  * In the case of a short read we start with find_task_by_pid.
3497  *
3498  * In the case of a seek we start with the leader and walk nr
3499  * threads past it.
3500  */
3501 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3502                                         struct pid_namespace *ns)
3503 {
3504         struct task_struct *pos, *task;
3505         unsigned long nr = f_pos;
3506
3507         if (nr != f_pos)        /* 32bit overflow? */
3508                 return NULL;
3509
3510         rcu_read_lock();
3511         task = pid_task(pid, PIDTYPE_PID);
3512         if (!task)
3513                 goto fail;
3514
3515         /* Attempt to start with the tid of a thread */
3516         if (tid && nr) {
3517                 pos = find_task_by_pid_ns(tid, ns);
3518                 if (pos && same_thread_group(pos, task))
3519                         goto found;
3520         }
3521
3522         /* If nr exceeds the number of threads there is nothing todo */
3523         if (nr >= get_nr_threads(task))
3524                 goto fail;
3525
3526         /* If we haven't found our starting place yet start
3527          * with the leader and walk nr threads forward.
3528          */
3529         pos = task = task->group_leader;
3530         do {
3531                 if (!nr--)
3532                         goto found;
3533         } while_each_thread(task, pos);
3534 fail:
3535         pos = NULL;
3536         goto out;
3537 found:
3538         get_task_struct(pos);
3539 out:
3540         rcu_read_unlock();
3541         return pos;
3542 }
3543
3544 /*
3545  * Find the next thread in the thread list.
3546  * Return NULL if there is an error or no next thread.
3547  *
3548  * The reference to the input task_struct is released.
3549  */
3550 static struct task_struct *next_tid(struct task_struct *start)
3551 {
3552         struct task_struct *pos = NULL;
3553         rcu_read_lock();
3554         if (pid_alive(start)) {
3555                 pos = next_thread(start);
3556                 if (thread_group_leader(pos))
3557                         pos = NULL;
3558                 else
3559                         get_task_struct(pos);
3560         }
3561         rcu_read_unlock();
3562         put_task_struct(start);
3563         return pos;
3564 }
3565
3566 /* for the /proc/TGID/task/ directories */
3567 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3568 {
3569         struct inode *inode = file_inode(file);
3570         struct task_struct *task;
3571         struct pid_namespace *ns;
3572         int tid;
3573
3574         if (proc_inode_is_dead(inode))
3575                 return -ENOENT;
3576
3577         if (!dir_emit_dots(file, ctx))
3578                 return 0;
3579
3580         /* f_version caches the tgid value that the last readdir call couldn't
3581          * return. lseek aka telldir automagically resets f_version to 0.
3582          */
3583         ns = proc_pid_ns(inode);
3584         tid = (int)file->f_version;
3585         file->f_version = 0;
3586         for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3587              task;
3588              task = next_tid(task), ctx->pos++) {
3589                 char name[10 + 1];
3590                 unsigned int len;
3591                 tid = task_pid_nr_ns(task, ns);
3592                 len = snprintf(name, sizeof(name), "%u", tid);
3593                 if (!proc_fill_cache(file, ctx, name, len,
3594                                 proc_task_instantiate, task, NULL)) {
3595                         /* returning this tgid failed, save it as the first
3596                          * pid for the next readir call */
3597                         file->f_version = (u64)tid;
3598                         put_task_struct(task);
3599                         break;
3600                 }
3601         }
3602
3603         return 0;
3604 }
3605
3606 static int proc_task_getattr(const struct path *path, struct kstat *stat,
3607                              u32 request_mask, unsigned int query_flags)
3608 {
3609         struct inode *inode = d_inode(path->dentry);
3610         struct task_struct *p = get_proc_task(inode);
3611         generic_fillattr(inode, stat);
3612
3613         if (p) {
3614                 stat->nlink += get_nr_threads(p);
3615                 put_task_struct(p);
3616         }
3617
3618         return 0;
3619 }
3620
3621 static const struct inode_operations proc_task_inode_operations = {
3622         .lookup         = proc_task_lookup,
3623         .getattr        = proc_task_getattr,
3624         .setattr        = proc_setattr,
3625         .permission     = proc_pid_permission,
3626 };
3627
3628 static const struct file_operations proc_task_operations = {
3629         .read           = generic_read_dir,
3630         .iterate_shared = proc_task_readdir,
3631         .llseek         = generic_file_llseek,
3632 };
3633
3634 void __init set_proc_pid_nlink(void)
3635 {
3636         nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3637         nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3638 }