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 <[email protected]>
  21  *  Edjard Mota <[email protected]>
  22  *  Ilias Biris <[email protected]>
  23  *  Mauricio Lin <[email protected]>
  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 <[email protected]>:
  45  *  Smaps information related to shared, private, clean and dirty pages.
  46  *
  47  *  Paul Mundt <[email protected]>:
  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_printf(m, "%-25s %-20s %-20s %-10s\n",
 585                   "Limit", "Soft Limit", "Hard Limit", "Units");
 586
 587         for (i = 0; i < RLIM_NLIMITS; i++) {
 588                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 589                         seq_printf(m, "%-25s %-20s ",
 590                                    lnames[i].name, "unlimited");
 591                 else
 592                         seq_printf(m, "%-25s %-20lu ",
 593                                    lnames[i].name, rlim[i].rlim_cur);
 594
 595                 if (rlim[i].rlim_max == RLIM_INFINITY)
 596                         seq_printf(m, "%-20s ", "unlimited");
 597                 else
 598                         seq_printf(m, "%-20lu ", rlim[i].rlim_max);
 599
 600                 if (lnames[i].unit)
 601                         seq_printf(m, "%-10s\n", lnames[i].unit);
 602                 else
 603                         seq_putc(m, '\n');
 604         }
 605
 606         return 0;
 607 }
 608
 609 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 610 static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
 611                             struct pid *pid, struct task_struct *task)
 612 {
 613         long nr;
 614         unsigned long args[6], sp, pc;
 615         int res;
 616
 617         res = lock_trace(task);
 618         if (res)
 619                 return res;
 620
 621         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 622                 seq_puts(m, "running\n");
 623         else if (nr < 0)
 624                 seq_printf(m, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 625         else
 626                 seq_printf(m,
 627                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 628                        nr,
 629                        args[0], args[1], args[2], args[3], args[4], args[5],
 630                        sp, pc);
 631         unlock_trace(task);
 632
 633         return 0;
 634 }
 635 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 636
 637 /************************************************************************/
 638 /*                       Here the fs part begins                        */
 639 /************************************************************************/
 640
 641 /* permission checks */
 642 static int proc_fd_access_allowed(struct inode *inode)
 643 {
 644         struct task_struct *task;
 645         int allowed = 0;
 646         /* Allow access to a task's file descriptors if it is us or we
 647          * may use ptrace attach to the process and find out that
 648          * information.
 649          */
 650         task = get_proc_task(inode);
 651         if (task) {
 652                 allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 653                 put_task_struct(task);
 654         }
 655         return allowed;
 656 }
 657
 658 int proc_setattr(struct dentry *dentry, struct iattr *attr)
 659 {
 660         int error;
 661         struct inode *inode = d_inode(dentry);
 662
 663         if (attr->ia_valid & ATTR_MODE)
 664                 return -EPERM;
 665
 666         error = setattr_prepare(dentry, attr);
 667         if (error)
 668                 return error;
 669
 670         setattr_copy(inode, attr);
 671         mark_inode_dirty(inode);
 672         return 0;
 673 }
 674
 675 /*
 676  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
 677  * or euid/egid (for hide_pid_min=2)?
 678  */
 679 static bool has_pid_permissions(struct pid_namespace *pid,
 680                                  struct task_struct *task,
 681                                  int hide_pid_min)
 682 {
 683         if (pid->hide_pid < hide_pid_min)
 684                 return true;
 685         if (in_group_p(pid->pid_gid))
 686                 return true;
 687         return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 688 }
 689
 690
 691 static int proc_pid_permission(struct inode *inode, int mask)
 692 {
 693         struct pid_namespace *pid = proc_pid_ns(inode);
 694         struct task_struct *task;
 695         bool has_perms;
 696
 697         task = get_proc_task(inode);
 698         if (!task)
 699                 return -ESRCH;
 700         has_perms = has_pid_permissions(pid, task, HIDEPID_NO_ACCESS);
 701         put_task_struct(task);
 702
 703         if (!has_perms) {
 704                 if (pid->hide_pid == HIDEPID_INVISIBLE) {
 705                         /*
 706                          * Let's make getdents(), stat(), and open()
 707                          * consistent with each other.  If a process
 708                          * may not stat() a file, it shouldn't be seen
 709                          * in procfs at all.
 710                          */
 711                         return -ENOENT;
 712                 }
 713
 714                 return -EPERM;
 715         }
 716         return generic_permission(inode, mask);
 717 }
 718
 719
 720
 721 static const struct inode_operations proc_def_inode_operations = {
 722         .setattr        = proc_setattr,
 723 };
 724
 725 static int proc_single_show(struct seq_file *m, void *v)
 726 {
 727         struct inode *inode = m->private;
 728         struct pid_namespace *ns = proc_pid_ns(inode);
 729         struct pid *pid = proc_pid(inode);
 730         struct task_struct *task;
 731         int ret;
 732
 733         task = get_pid_task(pid, PIDTYPE_PID);
 734         if (!task)
 735                 return -ESRCH;
 736
 737         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 738
 739         put_task_struct(task);
 740         return ret;
 741 }
 742
 743 static int proc_single_open(struct inode *inode, struct file *filp)
 744 {
 745         return single_open(filp, proc_single_show, inode);
 746 }
 747
 748 static const struct file_operations proc_single_file_operations = {
 749         .open           = proc_single_open,
 750         .read           = seq_read,
 751         .llseek         = seq_lseek,
 752         .release        = single_release,
 753 };
 754
 755
 756 struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
 757 {
 758         struct task_struct *task = get_proc_task(inode);
 759         struct mm_struct *mm = ERR_PTR(-ESRCH);
 760
 761         if (task) {
 762                 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
 763                 put_task_struct(task);
 764
 765                 if (!IS_ERR_OR_NULL(mm)) {
 766                         /* ensure this mm_struct can't be freed */
 767                         mmgrab(mm);
 768                         /* but do not pin its memory */
 769                         mmput(mm);
 770                 }
 771         }
 772
 773         return mm;
 774 }
 775
 776 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
 777 {
 778         struct mm_struct *mm = proc_mem_open(inode, mode);
 779
 780         if (IS_ERR(mm))
 781                 return PTR_ERR(mm);
 782
 783         file->private_data = mm;
 784         return 0;
 785 }
 786
 787 static int mem_open(struct inode *inode, struct file *file)
 788 {
 789         int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
 790
 791         /* OK to pass negative loff_t, we can catch out-of-range */
 792         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 793
 794         return ret;
 795 }
 796
 797 static ssize_t mem_rw(struct file *file, char __user *buf,
 798                         size_t count, loff_t *ppos, int write)
 799 {
 800         struct mm_struct *mm = file->private_data;
 801         unsigned long addr = *ppos;
 802         ssize_t copied;
 803         char *page;
 804         unsigned int flags;
 805
 806         if (!mm)
 807                 return 0;
 808
 809         page = (char *)__get_free_page(GFP_KERNEL);
 810         if (!page)
 811                 return -ENOMEM;
 812
 813         copied = 0;
 814         if (!mmget_not_zero(mm))
 815                 goto free;
 816
 817         flags = FOLL_FORCE | (write ? FOLL_WRITE : 0);
 818
 819         while (count > 0) {
 820                 int this_len = min_t(int, count, PAGE_SIZE);
 821
 822                 if (write && copy_from_user(page, buf, this_len)) {
 823                         copied = -EFAULT;
 824                         break;
 825                 }
 826
 827                 this_len = access_remote_vm(mm, addr, page, this_len, flags);
 828                 if (!this_len) {
 829                         if (!copied)
 830                                 copied = -EIO;
 831                         break;
 832                 }
 833
 834                 if (!write && copy_to_user(buf, page, this_len)) {
 835                         copied = -EFAULT;
 836                         break;
 837                 }
 838
 839                 buf += this_len;
 840                 addr += this_len;
 841                 copied += this_len;
 842                 count -= this_len;
 843         }
 844         *ppos = addr;
 845
 846         mmput(mm);
 847 free:
 848         free_page((unsigned long) page);
 849         return copied;
 850 }
 851
 852 static ssize_t mem_read(struct file *file, char __user *buf,
 853                         size_t count, loff_t *ppos)
 854 {
 855         return mem_rw(file, buf, count, ppos, 0);
 856 }
 857
 858 static ssize_t mem_write(struct file *file, const char __user *buf,
 859                          size_t count, loff_t *ppos)
 860 {
 861         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 862 }
 863
 864 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 865 {
 866         switch (orig) {
 867         case 0:
 868                 file->f_pos = offset;
 869                 break;
 870         case 1:
 871                 file->f_pos += offset;
 872                 break;
 873         default:
 874                 return -EINVAL;
 875         }
 876         force_successful_syscall_return();
 877         return file->f_pos;
 878 }
 879
 880 static int mem_release(struct inode *inode, struct file *file)
 881 {
 882         struct mm_struct *mm = file->private_data;
 883         if (mm)
 884                 mmdrop(mm);
 885         return 0;
 886 }
 887
 888 static const struct file_operations proc_mem_operations = {
 889         .llseek         = mem_lseek,
 890         .read           = mem_read,
 891         .write          = mem_write,
 892         .open           = mem_open,
 893         .release        = mem_release,
 894 };
 895
 896 static int environ_open(struct inode *inode, struct file *file)
 897 {
 898         return __mem_open(inode, file, PTRACE_MODE_READ);
 899 }
 900
 901 static ssize_t environ_read(struct file *file, char __user *buf,
 902                         size_t count, loff_t *ppos)
 903 {
 904         char *page;
 905         unsigned long src = *ppos;
 906         int ret = 0;
 907         struct mm_struct *mm = file->private_data;
 908         unsigned long env_start, env_end;
 909
 910         /* Ensure the process spawned far enough to have an environment. */
 911         if (!mm || !mm->env_end)
 912                 return 0;
 913
 914         page = (char *)__get_free_page(GFP_KERNEL);
 915         if (!page)
 916                 return -ENOMEM;
 917
 918         ret = 0;
 919         if (!mmget_not_zero(mm))
 920                 goto free;
 921
 922         spin_lock(&mm->arg_lock);
 923         env_start = mm->env_start;
 924         env_end = mm->env_end;
 925         spin_unlock(&mm->arg_lock);
 926
 927         while (count > 0) {
 928                 size_t this_len, max_len;
 929                 int retval;
 930
 931                 if (src >= (env_end - env_start))
 932                         break;
 933
 934                 this_len = env_end - (env_start + src);
 935
 936                 max_len = min_t(size_t, PAGE_SIZE, count);
 937                 this_len = min(max_len, this_len);
 938
 939                 retval = access_remote_vm(mm, (env_start + src), page, this_len, FOLL_ANON);
 940
 941                 if (retval <= 0) {
 942                         ret = retval;
 943                         break;
 944                 }
 945
 946                 if (copy_to_user(buf, page, retval)) {
 947                         ret = -EFAULT;
 948                         break;
 949                 }
 950
 951                 ret += retval;
 952                 src += retval;
 953                 buf += retval;
 954                 count -= retval;
 955         }
 956         *ppos = src;
 957         mmput(mm);
 958
 959 free:
 960         free_page((unsigned long) page);
 961         return ret;
 962 }
 963
 964 static const struct file_operations proc_environ_operations = {
 965         .open           = environ_open,
 966         .read           = environ_read,
 967         .llseek         = generic_file_llseek,
 968         .release        = mem_release,
 969 };
 970
 971 static int auxv_open(struct inode *inode, struct file *file)
 972 {
 973         return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
 974 }
 975
 976 static ssize_t auxv_read(struct file *file, char __user *buf,
 977                         size_t count, loff_t *ppos)
 978 {
 979         struct mm_struct *mm = file->private_data;
 980         unsigned int nwords = 0;
 981
 982         if (!mm)
 983                 return 0;
 984         do {
 985                 nwords += 2;
 986         } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 987         return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
 988                                        nwords * sizeof(mm->saved_auxv[0]));
 989 }
 990
 991 static const struct file_operations proc_auxv_operations = {
 992         .open           = auxv_open,
 993         .read           = auxv_read,
 994         .llseek         = generic_file_llseek,
 995         .release        = mem_release,
 996 };
 997
 998 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
 999                             loff_t *ppos)
1000 {
1001         struct task_struct *task = get_proc_task(file_inode(file));
1002         char buffer[PROC_NUMBUF];
1003         int oom_adj = OOM_ADJUST_MIN;
1004         size_t len;
1005
1006         if (!task)
1007                 return -ESRCH;
1008         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1009                 oom_adj = OOM_ADJUST_MAX;
1010         else
1011                 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1012                           OOM_SCORE_ADJ_MAX;
1013         put_task_struct(task);
1014         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1015         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1016 }
1017
1018 static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1019 {
1020         static DEFINE_MUTEX(oom_adj_mutex);
1021         struct mm_struct *mm = NULL;
1022         struct task_struct *task;
1023         int err = 0;
1024
1025         task = get_proc_task(file_inode(file));
1026         if (!task)
1027                 return -ESRCH;
1028
1029         mutex_lock(&oom_adj_mutex);
1030         if (legacy) {
1031                 if (oom_adj < task->signal->oom_score_adj &&
1032                                 !capable(CAP_SYS_RESOURCE)) {
1033                         err = -EACCES;
1034                         goto err_unlock;
1035                 }
1036                 /*
1037                  * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1038                  * /proc/pid/oom_score_adj instead.
1039                  */
1040                 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1041                           current->comm, task_pid_nr(current), task_pid_nr(task),
1042                           task_pid_nr(task));
1043         } else {
1044                 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1045                                 !capable(CAP_SYS_RESOURCE)) {
1046                         err = -EACCES;
1047                         goto err_unlock;
1048                 }
1049         }
1050
1051         /*
1052          * Make sure we will check other processes sharing the mm if this is
1053          * not vfrok which wants its own oom_score_adj.
1054          * pin the mm so it doesn't go away and get reused after task_unlock
1055          */
1056         if (!task->vfork_done) {
1057                 struct task_struct *p = find_lock_task_mm(task);
1058
1059                 if (p) {
1060                         if (atomic_read(&p->mm->mm_users) > 1) {
1061                                 mm = p->mm;
1062                                 mmgrab(mm);
1063                         }
1064                         task_unlock(p);
1065                 }
1066         }
1067
1068         task->signal->oom_score_adj = oom_adj;
1069         if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1070                 task->signal->oom_score_adj_min = (short)oom_adj;
1071         trace_oom_score_adj_update(task);
1072
1073         if (mm) {
1074                 struct task_struct *p;
1075
1076                 rcu_read_lock();
1077                 for_each_process(p) {
1078                         if (same_thread_group(task, p))
1079                                 continue;
1080
1081                         /* do not touch kernel threads or the global init */
1082                         if (p->flags & PF_KTHREAD || is_global_init(p))
1083                                 continue;
1084
1085                         task_lock(p);
1086                         if (!p->vfork_done && process_shares_mm(p, mm)) {
1087                                 pr_info("updating oom_score_adj for %d (%s) from %d to %d because it shares mm with %d (%s). Report if this is unexpected.\n",
1088                                                 task_pid_nr(p), p->comm,
1089                                                 p->signal->oom_score_adj, oom_adj,
1090                                                 task_pid_nr(task), task->comm);
1091                                 p->signal->oom_score_adj = oom_adj;
1092                                 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1093                                         p->signal->oom_score_adj_min = (short)oom_adj;
1094                         }
1095                         task_unlock(p);
1096                 }
1097                 rcu_read_unlock();
1098                 mmdrop(mm);
1099         }
1100 err_unlock:
1101         mutex_unlock(&oom_adj_mutex);
1102         put_task_struct(task);
1103         return err;
1104 }
1105
1106 /*
1107  * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1108  * kernels.  The effective policy is defined by oom_score_adj, which has a
1109  * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1110  * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1111  * Processes that become oom disabled via oom_adj will still be oom disabled
1112  * with this implementation.
1113  *
1114  * oom_adj cannot be removed since existing userspace binaries use it.
1115  */
1116 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1117                              size_t count, loff_t *ppos)
1118 {
1119         char buffer[PROC_NUMBUF];
1120         int oom_adj;
1121         int err;
1122
1123         memset(buffer, 0, sizeof(buffer));
1124         if (count > sizeof(buffer) - 1)
1125                 count = sizeof(buffer) - 1;
1126         if (copy_from_user(buffer, buf, count)) {
1127                 err = -EFAULT;
1128                 goto out;
1129         }
1130
1131         err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1132         if (err)
1133                 goto out;
1134         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1135              oom_adj != OOM_DISABLE) {
1136                 err = -EINVAL;
1137                 goto out;
1138         }
1139
1140         /*
1141          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1142          * value is always attainable.
1143          */
1144         if (oom_adj == OOM_ADJUST_MAX)
1145                 oom_adj = OOM_SCORE_ADJ_MAX;
1146         else
1147                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1148
1149         err = __set_oom_adj(file, oom_adj, true);
1150 out:
1151         return err < 0 ? err : count;
1152 }
1153
1154 static const struct file_operations proc_oom_adj_operations = {
1155         .read           = oom_adj_read,
1156         .write          = oom_adj_write,
1157         .llseek         = generic_file_llseek,
1158 };
1159
1160 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1161                                         size_t count, loff_t *ppos)
1162 {
1163         struct task_struct *task = get_proc_task(file_inode(file));
1164         char buffer[PROC_NUMBUF];
1165         short oom_score_adj = OOM_SCORE_ADJ_MIN;
1166         size_t len;
1167
1168         if (!task)
1169                 return -ESRCH;
1170         oom_score_adj = task->signal->oom_score_adj;
1171         put_task_struct(task);
1172         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1173         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1174 }
1175
1176 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1177                                         size_t count, loff_t *ppos)
1178 {
1179         char buffer[PROC_NUMBUF];
1180         int oom_score_adj;
1181         int err;
1182
1183         memset(buffer, 0, sizeof(buffer));
1184         if (count > sizeof(buffer) - 1)
1185                 count = sizeof(buffer) - 1;
1186         if (copy_from_user(buffer, buf, count)) {
1187                 err = -EFAULT;
1188                 goto out;
1189         }
1190
1191         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1192         if (err)
1193                 goto out;
1194         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1195                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1196                 err = -EINVAL;
1197                 goto out;
1198         }
1199
1200         err = __set_oom_adj(file, oom_score_adj, false);
1201 out:
1202         return err < 0 ? err : count;
1203 }
1204
1205 static const struct file_operations proc_oom_score_adj_operations = {
1206         .read           = oom_score_adj_read,
1207         .write          = oom_score_adj_write,
1208         .llseek         = default_llseek,
1209 };
1210
1211 #ifdef CONFIG_AUDITSYSCALL
1212 #define TMPBUFLEN 11
1213 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1214                                   size_t count, loff_t *ppos)
1215 {
1216         struct inode * inode = file_inode(file);
1217         struct task_struct *task = get_proc_task(inode);
1218         ssize_t length;
1219         char tmpbuf[TMPBUFLEN];
1220
1221         if (!task)
1222                 return -ESRCH;
1223         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1224                            from_kuid(file->f_cred->user_ns,
1225                                      audit_get_loginuid(task)));
1226         put_task_struct(task);
1227         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1228 }
1229
1230 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1231                                    size_t count, loff_t *ppos)
1232 {
1233         struct inode * inode = file_inode(file);
1234         uid_t loginuid;
1235         kuid_t kloginuid;
1236         int rv;
1237
1238         rcu_read_lock();
1239         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1240                 rcu_read_unlock();
1241                 return -EPERM;
1242         }
1243         rcu_read_unlock();
1244
1245         if (*ppos != 0) {
1246                 /* No partial writes. */
1247                 return -EINVAL;
1248         }
1249
1250         rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1251         if (rv < 0)
1252                 return rv;
1253
1254         /* is userspace tring to explicitly UNSET the loginuid? */
1255         if (loginuid == AUDIT_UID_UNSET) {
1256                 kloginuid = INVALID_UID;
1257         } else {
1258                 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1259                 if (!uid_valid(kloginuid))
1260                         return -EINVAL;
1261         }
1262
1263         rv = audit_set_loginuid(kloginuid);
1264         if (rv < 0)
1265                 return rv;
1266         return count;
1267 }
1268
1269 static const struct file_operations proc_loginuid_operations = {
1270         .read           = proc_loginuid_read,
1271         .write          = proc_loginuid_write,
1272         .llseek         = generic_file_llseek,
1273 };
1274
1275 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1276                                   size_t count, loff_t *ppos)
1277 {
1278         struct inode * inode = file_inode(file);
1279         struct task_struct *task = get_proc_task(inode);
1280         ssize_t length;
1281         char tmpbuf[TMPBUFLEN];
1282
1283         if (!task)
1284                 return -ESRCH;
1285         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1286                                 audit_get_sessionid(task));
1287         put_task_struct(task);
1288         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1289 }
1290
1291 static const struct file_operations proc_sessionid_operations = {
1292         .read           = proc_sessionid_read,
1293         .llseek         = generic_file_llseek,
1294 };
1295 #endif
1296
1297 #ifdef CONFIG_FAULT_INJECTION
1298 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1299                                       size_t count, loff_t *ppos)
1300 {
1301         struct task_struct *task = get_proc_task(file_inode(file));
1302         char buffer[PROC_NUMBUF];
1303         size_t len;
1304         int make_it_fail;
1305
1306         if (!task)
1307                 return -ESRCH;
1308         make_it_fail = task->make_it_fail;
1309         put_task_struct(task);
1310
1311         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1312
1313         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1314 }
1315
1316 static ssize_t proc_fault_inject_write(struct file * file,
1317                         const char __user * buf, size_t count, loff_t *ppos)
1318 {
1319         struct task_struct *task;
1320         char buffer[PROC_NUMBUF];
1321         int make_it_fail;
1322         int rv;
1323
1324         if (!capable(CAP_SYS_RESOURCE))
1325                 return -EPERM;
1326         memset(buffer, 0, sizeof(buffer));
1327         if (count > sizeof(buffer) - 1)
1328                 count = sizeof(buffer) - 1;
1329         if (copy_from_user(buffer, buf, count))
1330                 return -EFAULT;
1331         rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1332         if (rv < 0)
1333                 return rv;
1334         if (make_it_fail < 0 || make_it_fail > 1)
1335                 return -EINVAL;
1336
1337         task = get_proc_task(file_inode(file));
1338         if (!task)
1339                 return -ESRCH;
1340         task->make_it_fail = make_it_fail;
1341         put_task_struct(task);
1342
1343         return count;
1344 }
1345
1346 static const struct file_operations proc_fault_inject_operations = {
1347         .read           = proc_fault_inject_read,
1348         .write          = proc_fault_inject_write,
1349         .llseek         = generic_file_llseek,
1350 };
1351
1352 static ssize_t proc_fail_nth_write(struct file *file, const char __user *buf,
1353                                    size_t count, loff_t *ppos)
1354 {
1355         struct task_struct *task;
1356         int err;
1357         unsigned int n;
1358
1359         err = kstrtouint_from_user(buf, count, 0, &n);
1360         if (err)
1361                 return err;
1362
1363         task = get_proc_task(file_inode(file));
1364         if (!task)
1365                 return -ESRCH;
1366         task->fail_nth = n;
1367         put_task_struct(task);
1368
1369         return count;
1370 }
1371
1372 static ssize_t proc_fail_nth_read(struct file *file, char __user *buf,
1373                                   size_t count, loff_t *ppos)
1374 {
1375         struct task_struct *task;
1376         char numbuf[PROC_NUMBUF];
1377         ssize_t len;
1378
1379         task = get_proc_task(file_inode(file));
1380         if (!task)
1381                 return -ESRCH;
1382         len = snprintf(numbuf, sizeof(numbuf), "%u\n", task->fail_nth);
1383         put_task_struct(task);
1384         return simple_read_from_buffer(buf, count, ppos, numbuf, len);
1385 }
1386
1387 static const struct file_operations proc_fail_nth_operations = {
1388         .read           = proc_fail_nth_read,
1389         .write          = proc_fail_nth_write,
1390 };
1391 #endif
1392
1393
1394 #ifdef CONFIG_SCHED_DEBUG
1395 /*
1396  * Print out various scheduling related per-task fields:
1397  */
1398 static int sched_show(struct seq_file *m, void *v)
1399 {
1400         struct inode *inode = m->private;
1401         struct pid_namespace *ns = proc_pid_ns(inode);
1402         struct task_struct *p;
1403
1404         p = get_proc_task(inode);
1405         if (!p)
1406                 return -ESRCH;
1407         proc_sched_show_task(p, ns, m);
1408
1409         put_task_struct(p);
1410
1411         return 0;
1412 }
1413
1414 static ssize_t
1415 sched_write(struct file *file, const char __user *buf,
1416             size_t count, loff_t *offset)
1417 {
1418         struct inode *inode = file_inode(file);
1419         struct task_struct *p;
1420
1421         p = get_proc_task(inode);
1422         if (!p)
1423                 return -ESRCH;
1424         proc_sched_set_task(p);
1425
1426         put_task_struct(p);
1427
1428         return count;
1429 }
1430
1431 static int sched_open(struct inode *inode, struct file *filp)
1432 {
1433         return single_open(filp, sched_show, inode);
1434 }
1435
1436 static const struct file_operations proc_pid_sched_operations = {
1437         .open           = sched_open,
1438         .read           = seq_read,
1439         .write          = sched_write,
1440         .llseek         = seq_lseek,
1441         .release        = single_release,
1442 };
1443
1444 #endif
1445
1446 #ifdef CONFIG_SCHED_AUTOGROUP
1447 /*
1448  * Print out autogroup related information:
1449  */
1450 static int sched_autogroup_show(struct seq_file *m, void *v)
1451 {
1452         struct inode *inode = m->private;
1453         struct task_struct *p;
1454
1455         p = get_proc_task(inode);
1456         if (!p)
1457                 return -ESRCH;
1458         proc_sched_autogroup_show_task(p, m);
1459
1460         put_task_struct(p);
1461
1462         return 0;
1463 }
1464
1465 static ssize_t
1466 sched_autogroup_write(struct file *file, const char __user *buf,
1467             size_t count, loff_t *offset)
1468 {
1469         struct inode *inode = file_inode(file);
1470         struct task_struct *p;
1471         char buffer[PROC_NUMBUF];
1472         int nice;
1473         int err;
1474
1475         memset(buffer, 0, sizeof(buffer));
1476         if (count > sizeof(buffer) - 1)
1477                 count = sizeof(buffer) - 1;
1478         if (copy_from_user(buffer, buf, count))
1479                 return -EFAULT;
1480
1481         err = kstrtoint(strstrip(buffer), 0, &nice);
1482         if (err < 0)
1483                 return err;
1484
1485         p = get_proc_task(inode);
1486         if (!p)
1487                 return -ESRCH;
1488
1489         err = proc_sched_autogroup_set_nice(p, nice);
1490         if (err)
1491                 count = err;
1492
1493         put_task_struct(p);
1494
1495         return count;
1496 }
1497
1498 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1499 {
1500         int ret;
1501
1502         ret = single_open(filp, sched_autogroup_show, NULL);
1503         if (!ret) {
1504                 struct seq_file *m = filp->private_data;
1505
1506                 m->private = inode;
1507         }
1508         return ret;
1509 }
1510
1511 static const struct file_operations proc_pid_sched_autogroup_operations = {
1512         .open           = sched_autogroup_open,
1513         .read           = seq_read,
1514         .write          = sched_autogroup_write,
1515         .llseek         = seq_lseek,
1516         .release        = single_release,
1517 };
1518
1519 #endif /* CONFIG_SCHED_AUTOGROUP */
1520
1521 static ssize_t comm_write(struct file *file, const char __user *buf,
1522                                 size_t count, loff_t *offset)
1523 {
1524         struct inode *inode = file_inode(file);
1525         struct task_struct *p;
1526         char buffer[TASK_COMM_LEN];
1527         const size_t maxlen = sizeof(buffer) - 1;
1528
1529         memset(buffer, 0, sizeof(buffer));
1530         if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1531                 return -EFAULT;
1532
1533         p = get_proc_task(inode);
1534         if (!p)
1535                 return -ESRCH;
1536
1537         if (same_thread_group(current, p))
1538                 set_task_comm(p, buffer);
1539         else
1540                 count = -EINVAL;
1541
1542         put_task_struct(p);
1543
1544         return count;
1545 }
1546
1547 static int comm_show(struct seq_file *m, void *v)
1548 {
1549         struct inode *inode = m->private;
1550         struct task_struct *p;
1551
1552         p = get_proc_task(inode);
1553         if (!p)
1554                 return -ESRCH;
1555
1556         proc_task_name(m, p, false);
1557         seq_putc(m, '\n');
1558
1559         put_task_struct(p);
1560
1561         return 0;
1562 }
1563
1564 static int comm_open(struct inode *inode, struct file *filp)
1565 {
1566         return single_open(filp, comm_show, inode);
1567 }
1568
1569 static const struct file_operations proc_pid_set_comm_operations = {
1570         .open           = comm_open,
1571         .read           = seq_read,
1572         .write          = comm_write,
1573         .llseek         = seq_lseek,
1574         .release        = single_release,
1575 };
1576
1577 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1578 {
1579         struct task_struct *task;
1580         struct file *exe_file;
1581
1582         task = get_proc_task(d_inode(dentry));
1583         if (!task)
1584                 return -ENOENT;
1585         exe_file = get_task_exe_file(task);
1586         put_task_struct(task);
1587         if (exe_file) {
1588                 *exe_path = exe_file->f_path;
1589                 path_get(&exe_file->f_path);
1590                 fput(exe_file);
1591                 return 0;
1592         } else
1593                 return -ENOENT;
1594 }
1595
1596 static const char *proc_pid_get_link(struct dentry *dentry,
1597                                      struct inode *inode,
1598                                      struct delayed_call *done)
1599 {
1600         struct path path;
1601         int error = -EACCES;
1602
1603         if (!dentry)
1604                 return ERR_PTR(-ECHILD);
1605
1606         /* Are we allowed to snoop on the tasks file descriptors? */
1607         if (!proc_fd_access_allowed(inode))
1608                 goto out;
1609
1610         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1611         if (error)
1612                 goto out;
1613
1614         nd_jump_link(&path);
1615         return NULL;
1616 out:
1617         return ERR_PTR(error);
1618 }
1619
1620 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1621 {
1622         char *tmp = (char *)__get_free_page(GFP_KERNEL);
1623         char *pathname;
1624         int len;
1625
1626         if (!tmp)
1627                 return -ENOMEM;
1628
1629         pathname = d_path(path, tmp, PAGE_SIZE);
1630         len = PTR_ERR(pathname);
1631         if (IS_ERR(pathname))
1632                 goto out;
1633         len = tmp + PAGE_SIZE - 1 - pathname;
1634
1635         if (len > buflen)
1636                 len = buflen;
1637         if (copy_to_user(buffer, pathname, len))
1638                 len = -EFAULT;
1639  out:
1640         free_page((unsigned long)tmp);
1641         return len;
1642 }
1643
1644 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1645 {
1646         int error = -EACCES;
1647         struct inode *inode = d_inode(dentry);
1648         struct path path;
1649
1650         /* Are we allowed to snoop on the tasks file descriptors? */
1651         if (!proc_fd_access_allowed(inode))
1652                 goto out;
1653
1654         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1655         if (error)
1656                 goto out;
1657
1658         error = do_proc_readlink(&path, buffer, buflen);
1659         path_put(&path);
1660 out:
1661         return error;
1662 }
1663
1664 const struct inode_operations proc_pid_link_inode_operations = {
1665         .readlink       = proc_pid_readlink,
1666         .get_link       = proc_pid_get_link,
1667         .setattr        = proc_setattr,
1668 };
1669
1670
1671 /* building an inode */
1672
1673 void task_dump_owner(struct task_struct *task, umode_t mode,
1674                      kuid_t *ruid, kgid_t *rgid)
1675 {
1676         /* Depending on the state of dumpable compute who should own a
1677          * proc file for a task.
1678          */
1679         const struct cred *cred;
1680         kuid_t uid;
1681         kgid_t gid;
1682
1683         if (unlikely(task->flags & PF_KTHREAD)) {
1684                 *ruid = GLOBAL_ROOT_UID;
1685                 *rgid = GLOBAL_ROOT_GID;
1686                 return;
1687         }
1688
1689         /* Default to the tasks effective ownership */
1690         rcu_read_lock();
1691         cred = __task_cred(task);
1692         uid = cred->euid;
1693         gid = cred->egid;
1694         rcu_read_unlock();
1695
1696         /*
1697          * Before the /proc/pid/status file was created the only way to read
1698          * the effective uid of a /process was to stat /proc/pid.  Reading
1699          * /proc/pid/status is slow enough that procps and other packages
1700          * kept stating /proc/pid.  To keep the rules in /proc simple I have
1701          * made this apply to all per process world readable and executable
1702          * directories.
1703          */
1704         if (mode != (S_IFDIR|S_IRUGO|S_IXUGO)) {
1705                 struct mm_struct *mm;
1706                 task_lock(task);
1707                 mm = task->mm;
1708                 /* Make non-dumpable tasks owned by some root */
1709                 if (mm) {
1710                         if (get_dumpable(mm) != SUID_DUMP_USER) {
1711                                 struct user_namespace *user_ns = mm->user_ns;
1712
1713                                 uid = make_kuid(user_ns, 0);
1714                                 if (!uid_valid(uid))
1715                                         uid = GLOBAL_ROOT_UID;
1716
1717                                 gid = make_kgid(user_ns, 0);
1718                                 if (!gid_valid(gid))
1719                                         gid = GLOBAL_ROOT_GID;
1720                         }
1721                 } else {
1722                         uid = GLOBAL_ROOT_UID;
1723                         gid = GLOBAL_ROOT_GID;
1724                 }
1725                 task_unlock(task);
1726         }
1727         *ruid = uid;
1728         *rgid = gid;
1729 }
1730
1731 struct inode *proc_pid_make_inode(struct super_block * sb,
1732                                   struct task_struct *task, umode_t mode)
1733 {
1734         struct inode * inode;
1735         struct proc_inode *ei;
1736
1737         /* We need a new inode */
1738
1739         inode = new_inode(sb);
1740         if (!inode)
1741                 goto out;
1742
1743         /* Common stuff */
1744         ei = PROC_I(inode);
1745         inode->i_mode = mode;
1746         inode->i_ino = get_next_ino();
1747         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1748         inode->i_op = &proc_def_inode_operations;
1749
1750         /*
1751          * grab the reference to task.
1752          */
1753         ei->pid = get_task_pid(task, PIDTYPE_PID);
1754         if (!ei->pid)
1755                 goto out_unlock;
1756
1757         task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1758         security_task_to_inode(task, inode);
1759
1760 out:
1761         return inode;
1762
1763 out_unlock:
1764         iput(inode);
1765         return NULL;
1766 }
1767
1768 int pid_getattr(const struct path *path, struct kstat *stat,
1769                 u32 request_mask, unsigned int query_flags)
1770 {
1771         struct inode *inode = d_inode(path->dentry);
1772         struct pid_namespace *pid = proc_pid_ns(inode);
1773         struct task_struct *task;
1774
1775         generic_fillattr(inode, stat);
1776
1777         stat->uid = GLOBAL_ROOT_UID;
1778         stat->gid = GLOBAL_ROOT_GID;
1779         rcu_read_lock();
1780         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1781         if (task) {
1782                 if (!has_pid_permissions(pid, task, HIDEPID_INVISIBLE)) {
1783                         rcu_read_unlock();
1784                         /*
1785                          * This doesn't prevent learning whether PID exists,
1786                          * it only makes getattr() consistent with readdir().
1787                          */
1788                         return -ENOENT;
1789                 }
1790                 task_dump_owner(task, inode->i_mode, &stat->uid, &stat->gid);
1791         }
1792         rcu_read_unlock();
1793         return 0;
1794 }
1795
1796 /* dentry stuff */
1797
1798 /*
1799  * Set <pid>/... inode ownership (can change due to setuid(), etc.)
1800  */
1801 void pid_update_inode(struct task_struct *task, struct inode *inode)
1802 {
1803         task_dump_owner(task, inode->i_mode, &inode->i_uid, &inode->i_gid);
1804
1805         inode->i_mode &= ~(S_ISUID | S_ISGID);
1806         security_task_to_inode(task, inode);
1807 }
1808
1809 /*
1810  * Rewrite the inode's ownerships here because the owning task may have
1811  * performed a setuid(), etc.
1812  *
1813  */
1814 static int pid_revalidate(struct dentry *dentry, unsigned int flags)
1815 {
1816         struct inode *inode;
1817         struct task_struct *task;
1818
1819         if (flags & LOOKUP_RCU)
1820                 return -ECHILD;
1821
1822         inode = d_inode(dentry);
1823         task = get_proc_task(inode);
1824
1825         if (task) {
1826                 pid_update_inode(task, inode);
1827                 put_task_struct(task);
1828                 return 1;
1829         }
1830         return 0;
1831 }
1832
1833 static inline bool proc_inode_is_dead(struct inode *inode)
1834 {
1835         return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1836 }
1837
1838 int pid_delete_dentry(const struct dentry *dentry)
1839 {
1840         /* Is the task we represent dead?
1841          * If so, then don't put the dentry on the lru list,
1842          * kill it immediately.
1843          */
1844         return proc_inode_is_dead(d_inode(dentry));
1845 }
1846
1847 const struct dentry_operations pid_dentry_operations =
1848 {
1849         .d_revalidate   = pid_revalidate,
1850         .d_delete       = pid_delete_dentry,
1851 };
1852
1853 /* Lookups */
1854
1855 /*
1856  * Fill a directory entry.
1857  *
1858  * If possible create the dcache entry and derive our inode number and
1859  * file type from dcache entry.
1860  *
1861  * Since all of the proc inode numbers are dynamically generated, the inode
1862  * numbers do not exist until the inode is cache.  This means creating the
1863  * the dcache entry in readdir is necessary to keep the inode numbers
1864  * reported by readdir in sync with the inode numbers reported
1865  * by stat.
1866  */
1867 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1868         const char *name, unsigned int len,
1869         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1870 {
1871         struct dentry *child, *dir = file->f_path.dentry;
1872         struct qstr qname = QSTR_INIT(name, len);
1873         struct inode *inode;
1874         unsigned type = DT_UNKNOWN;
1875         ino_t ino = 1;
1876
1877         child = d_hash_and_lookup(dir, &qname);
1878         if (!child) {
1879                 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1880                 child = d_alloc_parallel(dir, &qname, &wq);
1881                 if (IS_ERR(child))
1882                         goto end_instantiate;
1883                 if (d_in_lookup(child)) {
1884                         struct dentry *res;
1885                         res = instantiate(child, task, ptr);
1886                         d_lookup_done(child);
1887                         if (unlikely(res)) {
1888                                 dput(child);
1889                                 child = res;
1890                                 if (IS_ERR(child))
1891                                         goto end_instantiate;
1892                         }
1893                 }
1894         }
1895         inode = d_inode(child);
1896         ino = inode->i_ino;
1897         type = inode->i_mode >> 12;
1898         dput(child);
1899 end_instantiate:
1900         return dir_emit(ctx, name, len, ino, type);
1901 }
1902
1903 /*
1904  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1905  * which represent vma start and end addresses.
1906  */
1907 static int dname_to_vma_addr(struct dentry *dentry,
1908                              unsigned long *start, unsigned long *end)
1909 {
1910         const char *str = dentry->d_name.name;
1911         unsigned long long sval, eval;
1912         unsigned int len;
1913
1914         if (str[0] == '0' && str[1] != '-')
1915                 return -EINVAL;
1916         len = _parse_integer(str, 16, &sval);
1917         if (len & KSTRTOX_OVERFLOW)
1918                 return -EINVAL;
1919         if (sval != (unsigned long)sval)
1920                 return -EINVAL;
1921         str += len;
1922
1923         if (*str != '-')
1924                 return -EINVAL;
1925         str++;
1926
1927         if (str[0] == '0' && str[1])
1928                 return -EINVAL;
1929         len = _parse_integer(str, 16, &eval);
1930         if (len & KSTRTOX_OVERFLOW)
1931                 return -EINVAL;
1932         if (eval != (unsigned long)eval)
1933                 return -EINVAL;
1934         str += len;
1935
1936         if (*str != '\0')
1937                 return -EINVAL;
1938
1939         *start = sval;
1940         *end = eval;
1941
1942         return 0;
1943 }
1944
1945 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1946 {
1947         unsigned long vm_start, vm_end;
1948         bool exact_vma_exists = false;
1949         struct mm_struct *mm = NULL;
1950         struct task_struct *task;
1951         struct inode *inode;
1952         int status = 0;
1953
1954         if (flags & LOOKUP_RCU)
1955                 return -ECHILD;
1956
1957         inode = d_inode(dentry);
1958         task = get_proc_task(inode);
1959         if (!task)
1960                 goto out_notask;
1961
1962         mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1963         if (IS_ERR_OR_NULL(mm))
1964                 goto out;
1965
1966         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1967                 down_read(&mm->mmap_sem);
1968                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1969                 up_read(&mm->mmap_sem);
1970         }
1971
1972         mmput(mm);
1973
1974         if (exact_vma_exists) {
1975                 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1976
1977                 security_task_to_inode(task, inode);
1978                 status = 1;
1979         }
1980
1981 out:
1982         put_task_struct(task);
1983
1984 out_notask:
1985         return status;
1986 }
1987
1988 static const struct dentry_operations tid_map_files_dentry_operations = {
1989         .d_revalidate   = map_files_d_revalidate,
1990         .d_delete       = pid_delete_dentry,
1991 };
1992
1993 static int map_files_get_link(struct dentry *dentry, struct path *path)
1994 {
1995         unsigned long vm_start, vm_end;
1996         struct vm_area_struct *vma;
1997         struct task_struct *task;
1998         struct mm_struct *mm;
1999         int rc;
2000
2001         rc = -ENOENT;
2002         task = get_proc_task(d_inode(dentry));
2003         if (!task)
2004                 goto out;
2005
2006         mm = get_task_mm(task);
2007         put_task_struct(task);
2008         if (!mm)
2009                 goto out;
2010
2011         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2012         if (rc)
2013                 goto out_mmput;
2014
2015         rc = -ENOENT;
2016         down_read(&mm->mmap_sem);
2017         vma = find_exact_vma(mm, vm_start, vm_end);
2018         if (vma && vma->vm_file) {
2019                 *path = vma->vm_file->f_path;
2020                 path_get(path);
2021                 rc = 0;
2022         }
2023         up_read(&mm->mmap_sem);
2024
2025 out_mmput:
2026         mmput(mm);
2027 out:
2028         return rc;
2029 }
2030
2031 struct map_files_info {
2032         unsigned long   start;
2033         unsigned long   end;
2034         fmode_t         mode;
2035 };
2036
2037 /*
2038  * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
2039  * symlinks may be used to bypass permissions on ancestor directories in the
2040  * path to the file in question.
2041  */
2042 static const char *
2043 proc_map_files_get_link(struct dentry *dentry,
2044                         struct inode *inode,
2045                         struct delayed_call *done)
2046 {
2047         if (!capable(CAP_SYS_ADMIN))
2048                 return ERR_PTR(-EPERM);
2049
2050         return proc_pid_get_link(dentry, inode, done);
2051 }
2052
2053 /*
2054  * Identical to proc_pid_link_inode_operations except for get_link()
2055  */
2056 static const struct inode_operations proc_map_files_link_inode_operations = {
2057         .readlink       = proc_pid_readlink,
2058         .get_link       = proc_map_files_get_link,
2059         .setattr        = proc_setattr,
2060 };
2061
2062 static struct dentry *
2063 proc_map_files_instantiate(struct dentry *dentry,
2064                            struct task_struct *task, const void *ptr)
2065 {
2066         fmode_t mode = (fmode_t)(unsigned long)ptr;
2067         struct proc_inode *ei;
2068         struct inode *inode;
2069
2070         inode = proc_pid_make_inode(dentry->d_sb, task, S_IFLNK |
2071                                     ((mode & FMODE_READ ) ? S_IRUSR : 0) |
2072                                     ((mode & FMODE_WRITE) ? S_IWUSR : 0));
2073         if (!inode)
2074                 return ERR_PTR(-ENOENT);
2075
2076         ei = PROC_I(inode);
2077         ei->op.proc_get_link = map_files_get_link;
2078
2079         inode->i_op = &proc_map_files_link_inode_operations;
2080         inode->i_size = 64;
2081
2082         d_set_d_op(dentry, &tid_map_files_dentry_operations);
2083         return d_splice_alias(inode, dentry);
2084 }
2085
2086 static struct dentry *proc_map_files_lookup(struct inode *dir,
2087                 struct dentry *dentry, unsigned int flags)
2088 {
2089         unsigned long vm_start, vm_end;
2090         struct vm_area_struct *vma;
2091         struct task_struct *task;
2092         struct dentry *result;
2093         struct mm_struct *mm;
2094
2095         result = ERR_PTR(-ENOENT);
2096         task = get_proc_task(dir);
2097         if (!task)
2098                 goto out;
2099
2100         result = ERR_PTR(-EACCES);
2101         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2102                 goto out_put_task;
2103
2104         result = ERR_PTR(-ENOENT);
2105         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2106                 goto out_put_task;
2107
2108         mm = get_task_mm(task);
2109         if (!mm)
2110                 goto out_put_task;
2111
2112         down_read(&mm->mmap_sem);
2113         vma = find_exact_vma(mm, vm_start, vm_end);
2114         if (!vma)
2115                 goto out_no_vma;
2116
2117         if (vma->vm_file)
2118                 result = proc_map_files_instantiate(dentry, task,
2119                                 (void *)(unsigned long)vma->vm_file->f_mode);
2120
2121 out_no_vma:
2122         up_read(&mm->mmap_sem);
2123         mmput(mm);
2124 out_put_task:
2125         put_task_struct(task);
2126 out:
2127         return result;
2128 }
2129
2130 static const struct inode_operations proc_map_files_inode_operations = {
2131         .lookup         = proc_map_files_lookup,
2132         .permission     = proc_fd_permission,
2133         .setattr        = proc_setattr,
2134 };
2135
2136 static int
2137 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2138 {
2139         struct vm_area_struct *vma;
2140         struct task_struct *task;
2141         struct mm_struct *mm;
2142         unsigned long nr_files, pos, i;
2143         struct flex_array *fa = NULL;
2144         struct map_files_info info;
2145         struct map_files_info *p;
2146         int ret;
2147
2148         ret = -ENOENT;
2149         task = get_proc_task(file_inode(file));
2150         if (!task)
2151                 goto out;
2152
2153         ret = -EACCES;
2154         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2155                 goto out_put_task;
2156
2157         ret = 0;
2158         if (!dir_emit_dots(file, ctx))
2159                 goto out_put_task;
2160
2161         mm = get_task_mm(task);
2162         if (!mm)
2163                 goto out_put_task;
2164         down_read(&mm->mmap_sem);
2165
2166         nr_files = 0;
2167
2168         /*
2169          * We need two passes here:
2170          *
2171          *  1) Collect vmas of mapped files with mmap_sem taken
2172          *  2) Release mmap_sem and instantiate entries
2173          *
2174          * otherwise we get lockdep complained, since filldir()
2175          * routine might require mmap_sem taken in might_fault().
2176          */
2177
2178         for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2179                 if (vma->vm_file && ++pos > ctx->pos)
2180                         nr_files++;
2181         }
2182
2183         if (nr_files) {
2184                 fa = flex_array_alloc(sizeof(info), nr_files,
2185                                         GFP_KERNEL);
2186                 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2187                                                 GFP_KERNEL)) {
2188                         ret = -ENOMEM;
2189                         if (fa)
2190                                 flex_array_free(fa);
2191                         up_read(&mm->mmap_sem);
2192                         mmput(mm);
2193                         goto out_put_task;
2194                 }
2195                 for (i = 0, vma = mm->mmap, pos = 2; vma;
2196                                 vma = vma->vm_next) {
2197                         if (!vma->vm_file)
2198                                 continue;
2199                         if (++pos <= ctx->pos)
2200                                 continue;
2201
2202                         info.start = vma->vm_start;
2203                         info.end = vma->vm_end;
2204                         info.mode = vma->vm_file->f_mode;
2205                         if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2206                                 BUG();
2207                 }
2208         }
2209         up_read(&mm->mmap_sem);
2210         mmput(mm);
2211
2212         for (i = 0; i < nr_files; i++) {
2213                 char buf[4 * sizeof(long) + 2]; /* max: %lx-%lx\0 */
2214                 unsigned int len;
2215
2216                 p = flex_array_get(fa, i);
2217                 len = snprintf(buf, sizeof(buf), "%lx-%lx", p->start, p->end);
2218                 if (!proc_fill_cache(file, ctx,
2219                                       buf, len,
2220                                       proc_map_files_instantiate,
2221                                       task,
2222                                       (void *)(unsigned long)p->mode))
2223                         break;
2224                 ctx->pos++;
2225         }
2226         if (fa)
2227                 flex_array_free(fa);
2228
2229 out_put_task:
2230         put_task_struct(task);
2231 out:
2232         return ret;
2233 }
2234
2235 static const struct file_operations proc_map_files_operations = {
2236         .read           = generic_read_dir,
2237         .iterate_shared = proc_map_files_readdir,
2238         .llseek         = generic_file_llseek,
2239 };
2240
2241 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2242 struct timers_private {
2243         struct pid *pid;
2244         struct task_struct *task;
2245         struct sighand_struct *sighand;
2246         struct pid_namespace *ns;
2247         unsigned long flags;
2248 };
2249
2250 static void *timers_start(struct seq_file *m, loff_t *pos)
2251 {
2252         struct timers_private *tp = m->private;
2253
2254         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2255         if (!tp->task)
2256                 return ERR_PTR(-ESRCH);
2257
2258         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2259         if (!tp->sighand)
2260                 return ERR_PTR(-ESRCH);
2261
2262         return seq_list_start(&tp->task->signal->posix_timers, *pos);
2263 }
2264
2265 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2266 {
2267         struct timers_private *tp = m->private;
2268         return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2269 }
2270
2271 static void timers_stop(struct seq_file *m, void *v)
2272 {
2273         struct timers_private *tp = m->private;
2274
2275         if (tp->sighand) {
2276                 unlock_task_sighand(tp->task, &tp->flags);
2277                 tp->sighand = NULL;
2278         }
2279
2280         if (tp->task) {
2281                 put_task_struct(tp->task);
2282                 tp->task = NULL;
2283         }
2284 }
2285
2286 static int show_timer(struct seq_file *m, void *v)
2287 {
2288         struct k_itimer *timer;
2289         struct timers_private *tp = m->private;
2290         int notify;
2291         static const char * const nstr[] = {
2292                 [SIGEV_SIGNAL] = "signal",
2293                 [SIGEV_NONE] = "none",
2294                 [SIGEV_THREAD] = "thread",
2295         };
2296
2297         timer = list_entry((struct list_head *)v, struct k_itimer, list);
2298         notify = timer->it_sigev_notify;
2299
2300         seq_printf(m, "ID: %d\n", timer->it_id);
2301         seq_printf(m, "signal: %d/%px\n",
2302                    timer->sigq->info.si_signo,
2303                    timer->sigq->info.si_value.sival_ptr);
2304         seq_printf(m, "notify: %s/%s.%d\n",
2305                    nstr[notify & ~SIGEV_THREAD_ID],
2306                    (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2307                    pid_nr_ns(timer->it_pid, tp->ns));
2308         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2309
2310         return 0;
2311 }
2312
2313 static const struct seq_operations proc_timers_seq_ops = {
2314         .start  = timers_start,
2315         .next   = timers_next,
2316         .stop   = timers_stop,
2317         .show   = show_timer,
2318 };
2319
2320 static int proc_timers_open(struct inode *inode, struct file *file)
2321 {
2322         struct timers_private *tp;
2323
2324         tp = __seq_open_private(file, &proc_timers_seq_ops,
2325                         sizeof(struct timers_private));
2326         if (!tp)
2327                 return -ENOMEM;
2328
2329         tp->pid = proc_pid(inode);
2330         tp->ns = proc_pid_ns(inode);
2331         return 0;
2332 }
2333
2334 static const struct file_operations proc_timers_operations = {
2335         .open           = proc_timers_open,
2336         .read           = seq_read,
2337         .llseek         = seq_lseek,
2338         .release        = seq_release_private,
2339 };
2340 #endif
2341
2342 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2343                                         size_t count, loff_t *offset)
2344 {
2345         struct inode *inode = file_inode(file);
2346         struct task_struct *p;
2347         u64 slack_ns;
2348         int err;
2349
2350         err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2351         if (err < 0)
2352                 return err;
2353
2354         p = get_proc_task(inode);
2355         if (!p)
2356                 return -ESRCH;
2357
2358         if (p != current) {
2359                 if (!capable(CAP_SYS_NICE)) {
2360                         count = -EPERM;
2361                         goto out;
2362                 }
2363
2364                 err = security_task_setscheduler(p);
2365                 if (err) {
2366                         count = err;
2367                         goto out;
2368                 }
2369         }
2370
2371         task_lock(p);
2372         if (slack_ns == 0)
2373                 p->timer_slack_ns = p->default_timer_slack_ns;
2374         else
2375                 p->timer_slack_ns = slack_ns;
2376         task_unlock(p);
2377
2378 out:
2379         put_task_struct(p);
2380
2381         return count;
2382 }
2383
2384 static int timerslack_ns_show(struct seq_file *m, void *v)
2385 {
2386         struct inode *inode = m->private;
2387         struct task_struct *p;
2388         int err = 0;
2389
2390         p = get_proc_task(inode);
2391         if (!p)
2392                 return -ESRCH;
2393
2394         if (p != current) {
2395
2396                 if (!capable(CAP_SYS_NICE)) {
2397                         err = -EPERM;
2398                         goto out;
2399                 }
2400                 err = security_task_getscheduler(p);
2401                 if (err)
2402                         goto out;
2403         }
2404
2405         task_lock(p);
2406         seq_printf(m, "%llu\n", p->timer_slack_ns);
2407         task_unlock(p);
2408
2409 out:
2410         put_task_struct(p);
2411
2412         return err;
2413 }
2414
2415 static int timerslack_ns_open(struct inode *inode, struct file *filp)
2416 {
2417         return single_open(filp, timerslack_ns_show, inode);
2418 }
2419
2420 static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2421         .open           = timerslack_ns_open,
2422         .read           = seq_read,
2423         .write          = timerslack_ns_write,
2424         .llseek         = seq_lseek,
2425         .release        = single_release,
2426 };
2427
2428 static struct dentry *proc_pident_instantiate(struct dentry *dentry,
2429         struct task_struct *task, const void *ptr)
2430 {
2431         const struct pid_entry *p = ptr;
2432         struct inode *inode;
2433         struct proc_inode *ei;
2434
2435         inode = proc_pid_make_inode(dentry->d_sb, task, p->mode);
2436         if (!inode)
2437                 return ERR_PTR(-ENOENT);
2438
2439         ei = PROC_I(inode);
2440         if (S_ISDIR(inode->i_mode))
2441                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2442         if (p->iop)
2443                 inode->i_op = p->iop;
2444         if (p->fop)
2445                 inode->i_fop = p->fop;
2446         ei->op = p->op;
2447         pid_update_inode(task, inode);
2448         d_set_d_op(dentry, &pid_dentry_operations);
2449         return d_splice_alias(inode, dentry);
2450 }
2451
2452 static struct dentry *proc_pident_lookup(struct inode *dir,
2453                                          struct dentry *dentry,
2454                                          const struct pid_entry *ents,
2455                                          unsigned int nents)
2456 {
2457         struct task_struct *task = get_proc_task(dir);
2458         const struct pid_entry *p, *last;
2459         struct dentry *res = ERR_PTR(-ENOENT);
2460
2461         if (!task)
2462                 goto out_no_task;
2463
2464         /*
2465          * Yes, it does not scale. And it should not. Don't add
2466          * new entries into /proc/<tgid>/ without very good reasons.
2467          */
2468         last = &ents[nents];
2469         for (p = ents; p < last; p++) {
2470                 if (p->len != dentry->d_name.len)
2471                         continue;
2472                 if (!memcmp(dentry->d_name.name, p->name, p->len)) {
2473                         res = proc_pident_instantiate(dentry, task, p);
2474                         break;
2475                 }
2476         }
2477         put_task_struct(task);
2478 out_no_task:
2479         return res;
2480 }
2481
2482 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2483                 const struct pid_entry *ents, unsigned int nents)
2484 {
2485         struct task_struct *task = get_proc_task(file_inode(file));
2486         const struct pid_entry *p;
2487
2488         if (!task)
2489                 return -ENOENT;
2490
2491         if (!dir_emit_dots(file, ctx))
2492                 goto out;
2493
2494         if (ctx->pos >= nents + 2)
2495                 goto out;
2496
2497         for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2498                 if (!proc_fill_cache(file, ctx, p->name, p->len,
2499                                 proc_pident_instantiate, task, p))
2500                         break;
2501                 ctx->pos++;
2502         }
2503 out:
2504         put_task_struct(task);
2505         return 0;
2506 }
2507
2508 #ifdef CONFIG_SECURITY
2509 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2510                                   size_t count, loff_t *ppos)
2511 {
2512         struct inode * inode = file_inode(file);
2513         char *p = NULL;
2514         ssize_t length;
2515         struct task_struct *task = get_proc_task(inode);
2516
2517         if (!task)
2518                 return -ESRCH;
2519
2520         length = security_getprocattr(task,
2521                                       (char*)file->f_path.dentry->d_name.name,
2522                                       &p);
2523         put_task_struct(task);
2524         if (length > 0)
2525                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2526         kfree(p);
2527         return length;
2528 }
2529
2530 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2531                                    size_t count, loff_t *ppos)
2532 {
2533         struct inode * inode = file_inode(file);
2534         struct task_struct *task;
2535         void *page;
2536         int rv;
2537
2538         rcu_read_lock();
2539         task = pid_task(proc_pid(inode), PIDTYPE_PID);
2540         if (!task) {
2541                 rcu_read_unlock();
2542                 return -ESRCH;
2543         }
2544         /* A task may only write its own attributes. */
2545         if (current != task) {
2546                 rcu_read_unlock();
2547                 return -EACCES;
2548         }
2549         rcu_read_unlock();
2550
2551         if (count > PAGE_SIZE)
2552                 count = PAGE_SIZE;
2553
2554         /* No partial writes. */
2555         if (*ppos != 0)
2556                 return -EINVAL;
2557
2558         page = memdup_user(buf, count);
2559         if (IS_ERR(page)) {
2560                 rv = PTR_ERR(page);
2561                 goto out;
2562         }
2563
2564         /* Guard against adverse ptrace interaction */
2565         rv = mutex_lock_interruptible(&current->signal->cred_guard_mutex);
2566         if (rv < 0)
2567                 goto out_free;
2568
2569         rv = security_setprocattr(file->f_path.dentry->d_name.name, page, count);
2570         mutex_unlock(&current->signal->cred_guard_mutex);
2571 out_free:
2572         kfree(page);
2573 out:
2574         return rv;
2575 }
2576
2577 static const struct file_operations proc_pid_attr_operations = {
2578         .read           = proc_pid_attr_read,
2579         .write          = proc_pid_attr_write,
2580         .llseek         = generic_file_llseek,
2581 };
2582
2583 static const struct pid_entry attr_dir_stuff[] = {
2584         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2585         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2586         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2587         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2588         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2589         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2590 };
2591
2592 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2593 {
2594         return proc_pident_readdir(file, ctx,
2595                                    attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2596 }
2597
2598 static const struct file_operations proc_attr_dir_operations = {
2599         .read           = generic_read_dir,
2600         .iterate_shared = proc_attr_dir_readdir,
2601         .llseek         = generic_file_llseek,
2602 };
2603
2604 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2605                                 struct dentry *dentry, unsigned int flags)
2606 {
2607         return proc_pident_lookup(dir, dentry,
2608                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2609 }
2610
2611 static const struct inode_operations proc_attr_dir_inode_operations = {
2612         .lookup         = proc_attr_dir_lookup,
2613         .getattr        = pid_getattr,
2614         .setattr        = proc_setattr,
2615 };
2616
2617 #endif
2618
2619 #ifdef CONFIG_ELF_CORE
2620 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2621                                          size_t count, loff_t *ppos)
2622 {
2623         struct task_struct *task = get_proc_task(file_inode(file));
2624         struct mm_struct *mm;
2625         char buffer[PROC_NUMBUF];
2626         size_t len;
2627         int ret;
2628
2629         if (!task)
2630                 return -ESRCH;
2631
2632         ret = 0;
2633         mm = get_task_mm(task);
2634         if (mm) {
2635                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2636                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2637                                 MMF_DUMP_FILTER_SHIFT));
2638                 mmput(mm);
2639                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2640         }
2641
2642         put_task_struct(task);
2643
2644         return ret;
2645 }
2646
2647 static ssize_t proc_coredump_filter_write(struct file *file,
2648                                           const char __user *buf,
2649                                           size_t count,
2650                                           loff_t *ppos)
2651 {
2652         struct task_struct *task;
2653         struct mm_struct *mm;
2654         unsigned int val;
2655         int ret;
2656         int i;
2657         unsigned long mask;
2658
2659         ret = kstrtouint_from_user(buf, count, 0, &val);
2660         if (ret < 0)
2661                 return ret;
2662
2663         ret = -ESRCH;
2664         task = get_proc_task(file_inode(file));
2665         if (!task)
2666                 goto out_no_task;
2667
2668         mm = get_task_mm(task);
2669         if (!mm)
2670                 goto out_no_mm;
2671         ret = 0;
2672
2673         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2674                 if (val & mask)
2675                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2676                 else
2677                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2678         }
2679
2680         mmput(mm);
2681  out_no_mm:
2682         put_task_struct(task);
2683  out_no_task:
2684         if (ret < 0)
2685                 return ret;
2686         return count;
2687 }
2688
2689 static const struct file_operations proc_coredump_filter_operations = {
2690         .read           = proc_coredump_filter_read,
2691         .write          = proc_coredump_filter_write,
2692         .llseek         = generic_file_llseek,
2693 };
2694 #endif
2695
2696 #ifdef CONFIG_TASK_IO_ACCOUNTING
2697 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2698 {
2699         struct task_io_accounting acct = task->ioac;
2700         unsigned long flags;
2701         int result;
2702
2703         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2704         if (result)
2705                 return result;
2706
2707         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2708                 result = -EACCES;
2709                 goto out_unlock;
2710         }
2711
2712         if (whole && lock_task_sighand(task, &flags)) {
2713                 struct task_struct *t = task;
2714
2715                 task_io_accounting_add(&acct, &task->signal->ioac);
2716                 while_each_thread(task, t)
2717                         task_io_accounting_add(&acct, &t->ioac);
2718
2719                 unlock_task_sighand(task, &flags);
2720         }
2721         seq_printf(m,
2722                    "rchar: %llu\n"
2723                    "wchar: %llu\n"
2724                    "syscr: %llu\n"
2725                    "syscw: %llu\n"
2726                    "read_bytes: %llu\n"
2727                    "write_bytes: %llu\n"
2728                    "cancelled_write_bytes: %llu\n",
2729                    (unsigned long long)acct.rchar,
2730                    (unsigned long long)acct.wchar,
2731                    (unsigned long long)acct.syscr,
2732                    (unsigned long long)acct.syscw,
2733                    (unsigned long long)acct.read_bytes,
2734                    (unsigned long long)acct.write_bytes,
2735                    (unsigned long long)acct.cancelled_write_bytes);
2736         result = 0;
2737
2738 out_unlock:
2739         mutex_unlock(&task->signal->cred_guard_mutex);
2740         return result;
2741 }
2742
2743 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2744                                   struct pid *pid, struct task_struct *task)
2745 {
2746         return do_io_accounting(task, m, 0);
2747 }
2748
2749 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2750                                    struct pid *pid, struct task_struct *task)
2751 {
2752         return do_io_accounting(task, m, 1);
2753 }
2754 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2755
2756 #ifdef CONFIG_USER_NS
2757 static int proc_id_map_open(struct inode *inode, struct file *file,
2758         const struct seq_operations *seq_ops)
2759 {
2760         struct user_namespace *ns = NULL;
2761         struct task_struct *task;
2762         struct seq_file *seq;
2763         int ret = -EINVAL;
2764
2765         task = get_proc_task(inode);
2766         if (task) {
2767                 rcu_read_lock();
2768                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2769                 rcu_read_unlock();
2770                 put_task_struct(task);
2771         }
2772         if (!ns)
2773                 goto err;
2774
2775         ret = seq_open(file, seq_ops);
2776         if (ret)
2777                 goto err_put_ns;
2778
2779         seq = file->private_data;
2780         seq->private = ns;
2781
2782         return 0;
2783 err_put_ns:
2784         put_user_ns(ns);
2785 err:
2786         return ret;
2787 }
2788
2789 static int proc_id_map_release(struct inode *inode, struct file *file)
2790 {
2791         struct seq_file *seq = file->private_data;
2792         struct user_namespace *ns = seq->private;
2793         put_user_ns(ns);
2794         return seq_release(inode, file);
2795 }
2796
2797 static int proc_uid_map_open(struct inode *inode, struct file *file)
2798 {
2799         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2800 }
2801
2802 static int proc_gid_map_open(struct inode *inode, struct file *file)
2803 {
2804         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2805 }
2806
2807 static int proc_projid_map_open(struct inode *inode, struct file *file)
2808 {
2809         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2810 }
2811
2812 static const struct file_operations proc_uid_map_operations = {
2813         .open           = proc_uid_map_open,
2814         .write          = proc_uid_map_write,
2815         .read           = seq_read,
2816         .llseek         = seq_lseek,
2817         .release        = proc_id_map_release,
2818 };
2819
2820 static const struct file_operations proc_gid_map_operations = {
2821         .open           = proc_gid_map_open,
2822         .write          = proc_gid_map_write,
2823         .read           = seq_read,
2824         .llseek         = seq_lseek,
2825         .release        = proc_id_map_release,
2826 };
2827
2828 static const struct file_operations proc_projid_map_operations = {
2829         .open           = proc_projid_map_open,
2830         .write          = proc_projid_map_write,
2831         .read           = seq_read,
2832         .llseek         = seq_lseek,
2833         .release        = proc_id_map_release,
2834 };
2835
2836 static int proc_setgroups_open(struct inode *inode, struct file *file)
2837 {
2838         struct user_namespace *ns = NULL;
2839         struct task_struct *task;
2840         int ret;
2841
2842         ret = -ESRCH;
2843         task = get_proc_task(inode);
2844         if (task) {
2845                 rcu_read_lock();
2846                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2847                 rcu_read_unlock();
2848                 put_task_struct(task);
2849         }
2850         if (!ns)
2851                 goto err;
2852
2853         if (file->f_mode & FMODE_WRITE) {
2854                 ret = -EACCES;
2855                 if (!ns_capable(ns, CAP_SYS_ADMIN))
2856                         goto err_put_ns;
2857         }
2858
2859         ret = single_open(file, &proc_setgroups_show, ns);
2860         if (ret)
2861                 goto err_put_ns;
2862
2863         return 0;
2864 err_put_ns:
2865         put_user_ns(ns);
2866 err:
2867         return ret;
2868 }
2869
2870 static int proc_setgroups_release(struct inode *inode, struct file *file)
2871 {
2872         struct seq_file *seq = file->private_data;
2873         struct user_namespace *ns = seq->private;
2874         int ret = single_release(inode, file);
2875         put_user_ns(ns);
2876         return ret;
2877 }
2878
2879 static const struct file_operations proc_setgroups_operations = {
2880         .open           = proc_setgroups_open,
2881         .write          = proc_setgroups_write,
2882         .read           = seq_read,
2883         .llseek         = seq_lseek,
2884         .release        = proc_setgroups_release,
2885 };
2886 #endif /* CONFIG_USER_NS */
2887
2888 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2889                                 struct pid *pid, struct task_struct *task)
2890 {
2891         int err = lock_trace(task);
2892         if (!err) {
2893                 seq_printf(m, "%08x\n", task->personality);
2894                 unlock_trace(task);
2895         }
2896         return err;
2897 }
2898
2899 #ifdef CONFIG_LIVEPATCH
2900 static int proc_pid_patch_state(struct seq_file *m, struct pid_namespace *ns,
2901                                 struct pid *pid, struct task_struct *task)
2902 {
2903         seq_printf(m, "%d\n", task->patch_state);
2904         return 0;
2905 }
2906 #endif /* CONFIG_LIVEPATCH */
2907
2908 #ifdef CONFIG_STACKLEAK_METRICS
2909 static int proc_stack_depth(struct seq_file *m, struct pid_namespace *ns,
2910                                 struct pid *pid, struct task_struct *task)
2911 {
2912         unsigned long prev_depth = THREAD_SIZE -
2913                                 (task->prev_lowest_stack & (THREAD_SIZE - 1));
2914         unsigned long depth = THREAD_SIZE -
2915                                 (task->lowest_stack & (THREAD_SIZE - 1));
2916
2917         seq_printf(m, "previous stack depth: %lu\nstack depth: %lu\n",
2918                                                         prev_depth, depth);
2919         return 0;
2920 }
2921 #endif /* CONFIG_STACKLEAK_METRICS */
2922
2923 /*
2924  * Thread groups
2925  */
2926 static const struct file_operations proc_task_operations;
2927 static const struct inode_operations proc_task_inode_operations;
2928
2929 static const struct pid_entry tgid_base_stuff[] = {
2930         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2931         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2932         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2933         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2934         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2935 #ifdef CONFIG_NET
2936         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2937 #endif
2938         REG("environ",    S_IRUSR, proc_environ_operations),
2939         REG("auxv",       S_IRUSR, proc_auxv_operations),
2940         ONE("status",     S_IRUGO, proc_pid_status),
2941         ONE("personality", S_IRUSR, proc_pid_personality),
2942         ONE("limits",     S_IRUGO, proc_pid_limits),
2943 #ifdef CONFIG_SCHED_DEBUG
2944         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2945 #endif
2946 #ifdef CONFIG_SCHED_AUTOGROUP
2947         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2948 #endif
2949         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2950 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2951         ONE("syscall",    S_IRUSR, proc_pid_syscall),
2952 #endif
2953         REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
2954         ONE("stat",       S_IRUGO, proc_tgid_stat),
2955         ONE("statm",      S_IRUGO, proc_pid_statm),
2956         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2957 #ifdef CONFIG_NUMA
2958         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2959 #endif
2960         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2961         LNK("cwd",        proc_cwd_link),
2962         LNK("root",       proc_root_link),
2963         LNK("exe",        proc_exe_link),
2964         REG("mounts",     S_IRUGO, proc_mounts_operations),
2965         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2966         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2967 #ifdef CONFIG_PROC_PAGE_MONITOR
2968         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2969         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2970         REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
2971         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2972 #endif
2973 #ifdef CONFIG_SECURITY
2974         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2975 #endif
2976 #ifdef CONFIG_KALLSYMS
2977         ONE("wchan",      S_IRUGO, proc_pid_wchan),
2978 #endif
2979 #ifdef CONFIG_STACKTRACE
2980         ONE("stack",      S_IRUSR, proc_pid_stack),
2981 #endif
2982 #ifdef CONFIG_SCHED_INFO
2983         ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
2984 #endif
2985 #ifdef CONFIG_LATENCYTOP
2986         REG("latency",  S_IRUGO, proc_lstats_operations),
2987 #endif
2988 #ifdef CONFIG_PROC_PID_CPUSET
2989         ONE("cpuset",     S_IRUGO, proc_cpuset_show),
2990 #endif
2991 #ifdef CONFIG_CGROUPS
2992         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
2993 #endif
2994         ONE("oom_score",  S_IRUGO, proc_oom_score),
2995         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2996         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2997 #ifdef CONFIG_AUDITSYSCALL
2998         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2999         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3000 #endif
3001 #ifdef CONFIG_FAULT_INJECTION
3002         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3003         REG("fail-nth", 0644, proc_fail_nth_operations),
3004 #endif
3005 #ifdef CONFIG_ELF_CORE
3006         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3007 #endif
3008 #ifdef CONFIG_TASK_IO_ACCOUNTING
3009         ONE("io",       S_IRUSR, proc_tgid_io_accounting),
3010 #endif
3011 #ifdef CONFIG_USER_NS
3012         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3013         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3014         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3015         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3016 #endif
3017 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
3018         REG("timers",     S_IRUGO, proc_timers_operations),
3019 #endif
3020         REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
3021 #ifdef CONFIG_LIVEPATCH
3022         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
3023 #endif
3024 #ifdef CONFIG_STACKLEAK_METRICS
3025         ONE("stack_depth", S_IRUGO, proc_stack_depth),
3026 #endif
3027 };
3028
3029 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
3030 {
3031         return proc_pident_readdir(file, ctx,
3032                                    tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3033 }
3034
3035 static const struct file_operations proc_tgid_base_operations = {
3036         .read           = generic_read_dir,
3037         .iterate_shared = proc_tgid_base_readdir,
3038         .llseek         = generic_file_llseek,
3039 };
3040
3041 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3042 {
3043         return proc_pident_lookup(dir, dentry,
3044                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3045 }
3046
3047 static const struct inode_operations proc_tgid_base_inode_operations = {
3048         .lookup         = proc_tgid_base_lookup,
3049         .getattr        = pid_getattr,
3050         .setattr        = proc_setattr,
3051         .permission     = proc_pid_permission,
3052 };
3053
3054 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
3055 {
3056         struct dentry *dentry, *leader, *dir;
3057         char buf[10 + 1];
3058         struct qstr name;
3059
3060         name.name = buf;
3061         name.len = snprintf(buf, sizeof(buf), "%u", pid);
3062         /* no ->d_hash() rejects on procfs */
3063         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
3064         if (dentry) {
3065                 d_invalidate(dentry);
3066                 dput(dentry);
3067         }
3068
3069         if (pid == tgid)
3070                 return;
3071
3072         name.name = buf;
3073         name.len = snprintf(buf, sizeof(buf), "%u", tgid);
3074         leader = d_hash_and_lookup(mnt->mnt_root, &name);
3075         if (!leader)
3076                 goto out;
3077
3078         name.name = "task";
3079         name.len = strlen(name.name);
3080         dir = d_hash_and_lookup(leader, &name);
3081         if (!dir)
3082                 goto out_put_leader;
3083
3084         name.name = buf;
3085         name.len = snprintf(buf, sizeof(buf), "%u", pid);
3086         dentry = d_hash_and_lookup(dir, &name);
3087         if (dentry) {
3088                 d_invalidate(dentry);
3089                 dput(dentry);
3090         }
3091
3092         dput(dir);
3093 out_put_leader:
3094         dput(leader);
3095 out:
3096         return;
3097 }
3098
3099 /**
3100  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
3101  * @task: task that should be flushed.
3102  *
3103  * When flushing dentries from proc, one needs to flush them from global
3104  * proc (proc_mnt) and from all the namespaces' procs this task was seen
3105  * in. This call is supposed to do all of this job.
3106  *
3107  * Looks in the dcache for
3108  * /proc/@pid
3109  * /proc/@tgid/task/@pid
3110  * if either directory is present flushes it and all of it'ts children
3111  * from the dcache.
3112  *
3113  * It is safe and reasonable to cache /proc entries for a task until
3114  * that task exits.  After that they just clog up the dcache with
3115  * useless entries, possibly causing useful dcache entries to be
3116  * flushed instead.  This routine is proved to flush those useless
3117  * dcache entries at process exit time.
3118  *
3119  * NOTE: This routine is just an optimization so it does not guarantee
3120  *       that no dcache entries will exist at process exit time it
3121  *       just makes it very unlikely that any will persist.
3122  */
3123
3124 void proc_flush_task(struct task_struct *task)
3125 {
3126         int i;
3127         struct pid *pid, *tgid;
3128         struct upid *upid;
3129
3130         pid = task_pid(task);
3131         tgid = task_tgid(task);
3132
3133         for (i = 0; i <= pid->level; i++) {
3134                 upid = &pid->numbers[i];
3135                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3136                                         tgid->numbers[i].nr);
3137         }
3138 }
3139
3140 static struct dentry *proc_pid_instantiate(struct dentry * dentry,
3141                                    struct task_struct *task, const void *ptr)
3142 {
3143         struct inode *inode;
3144
3145         inode = proc_pid_make_inode(dentry->d_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3146         if (!inode)
3147                 return ERR_PTR(-ENOENT);
3148
3149         inode->i_op = &proc_tgid_base_inode_operations;
3150         inode->i_fop = &proc_tgid_base_operations;
3151         inode->i_flags|=S_IMMUTABLE;
3152
3153         set_nlink(inode, nlink_tgid);
3154         pid_update_inode(task, inode);
3155
3156         d_set_d_op(dentry, &pid_dentry_operations);
3157         return d_splice_alias(inode, dentry);
3158 }
3159
3160 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3161 {
3162         struct task_struct *task;
3163         unsigned tgid;
3164         struct pid_namespace *ns;
3165         struct dentry *result = ERR_PTR(-ENOENT);
3166
3167         tgid = name_to_int(&dentry->d_name);
3168         if (tgid == ~0U)
3169                 goto out;
3170
3171         ns = dentry->d_sb->s_fs_info;
3172         rcu_read_lock();
3173         task = find_task_by_pid_ns(tgid, ns);
3174         if (task)
3175                 get_task_struct(task);
3176         rcu_read_unlock();
3177         if (!task)
3178                 goto out;
3179
3180         result = proc_pid_instantiate(dentry, task, NULL);
3181         put_task_struct(task);
3182 out:
3183         return result;
3184 }
3185
3186 /*
3187  * Find the first task with tgid >= tgid
3188  *
3189  */
3190 struct tgid_iter {
3191         unsigned int tgid;
3192         struct task_struct *task;
3193 };
3194 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3195 {
3196         struct pid *pid;
3197
3198         if (iter.task)
3199                 put_task_struct(iter.task);
3200         rcu_read_lock();
3201 retry:
3202         iter.task = NULL;
3203         pid = find_ge_pid(iter.tgid, ns);
3204         if (pid) {
3205                 iter.tgid = pid_nr_ns(pid, ns);
3206                 iter.task = pid_task(pid, PIDTYPE_PID);
3207                 /* What we to know is if the pid we have find is the
3208                  * pid of a thread_group_leader.  Testing for task
3209                  * being a thread_group_leader is the obvious thing
3210                  * todo but there is a window when it fails, due to
3211                  * the pid transfer logic in de_thread.
3212                  *
3213                  * So we perform the straight forward test of seeing
3214                  * if the pid we have found is the pid of a thread
3215                  * group leader, and don't worry if the task we have
3216                  * found doesn't happen to be a thread group leader.
3217                  * As we don't care in the case of readdir.
3218                  */
3219                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3220                         iter.tgid += 1;
3221                         goto retry;
3222                 }
3223                 get_task_struct(iter.task);
3224         }
3225         rcu_read_unlock();
3226         return iter;
3227 }
3228
3229 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3230
3231 /* for the /proc/ directory itself, after non-process stuff has been done */
3232 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3233 {
3234         struct tgid_iter iter;
3235         struct pid_namespace *ns = proc_pid_ns(file_inode(file));
3236         loff_t pos = ctx->pos;
3237
3238         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3239                 return 0;
3240
3241         if (pos == TGID_OFFSET - 2) {
3242                 struct inode *inode = d_inode(ns->proc_self);
3243                 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3244                         return 0;
3245                 ctx->pos = pos = pos + 1;
3246         }
3247         if (pos == TGID_OFFSET - 1) {
3248                 struct inode *inode = d_inode(ns->proc_thread_self);
3249                 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3250                         return 0;
3251                 ctx->pos = pos = pos + 1;
3252         }
3253         iter.tgid = pos - TGID_OFFSET;
3254         iter.task = NULL;
3255         for (iter = next_tgid(ns, iter);
3256              iter.task;
3257              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3258                 char name[10 + 1];
3259                 unsigned int len;
3260
3261                 cond_resched();
3262                 if (!has_pid_permissions(ns, iter.task, HIDEPID_INVISIBLE))
3263                         continue;
3264
3265                 len = snprintf(name, sizeof(name), "%u", iter.tgid);
3266                 ctx->pos = iter.tgid + TGID_OFFSET;
3267                 if (!proc_fill_cache(file, ctx, name, len,
3268                                      proc_pid_instantiate, iter.task, NULL)) {
3269                         put_task_struct(iter.task);
3270                         return 0;
3271                 }
3272         }
3273         ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3274         return 0;
3275 }
3276
3277 /*
3278  * proc_tid_comm_permission is a special permission function exclusively
3279  * used for the node /proc/<pid>/task/<tid>/comm.
3280  * It bypasses generic permission checks in the case where a task of the same
3281  * task group attempts to access the node.
3282  * The rationale behind this is that glibc and bionic access this node for
3283  * cross thread naming (pthread_set/getname_np(!self)). However, if
3284  * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3285  * which locks out the cross thread naming implementation.
3286  * This function makes sure that the node is always accessible for members of
3287  * same thread group.
3288  */
3289 static int proc_tid_comm_permission(struct inode *inode, int mask)
3290 {
3291         bool is_same_tgroup;
3292         struct task_struct *task;
3293
3294         task = get_proc_task(inode);
3295         if (!task)
3296                 return -ESRCH;
3297         is_same_tgroup = same_thread_group(current, task);
3298         put_task_struct(task);
3299
3300         if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3301                 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3302                  * read or written by the members of the corresponding
3303                  * thread group.
3304                  */
3305                 return 0;
3306         }
3307
3308         return generic_permission(inode, mask);
3309 }
3310
3311 static const struct inode_operations proc_tid_comm_inode_operations = {
3312                 .permission = proc_tid_comm_permission,
3313 };
3314
3315 /*
3316  * Tasks
3317  */
3318 static const struct pid_entry tid_base_stuff[] = {
3319         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3320         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3321         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3322 #ifdef CONFIG_NET
3323         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3324 #endif
3325         REG("environ",   S_IRUSR, proc_environ_operations),
3326         REG("auxv",      S_IRUSR, proc_auxv_operations),
3327         ONE("status",    S_IRUGO, proc_pid_status),
3328         ONE("personality", S_IRUSR, proc_pid_personality),
3329         ONE("limits",    S_IRUGO, proc_pid_limits),
3330 #ifdef CONFIG_SCHED_DEBUG
3331         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3332 #endif
3333         NOD("comm",      S_IFREG|S_IRUGO|S_IWUSR,
3334                          &proc_tid_comm_inode_operations,
3335                          &proc_pid_set_comm_operations, {}),
3336 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3337         ONE("syscall",   S_IRUSR, proc_pid_syscall),
3338 #endif
3339         REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3340         ONE("stat",      S_IRUGO, proc_tid_stat),
3341         ONE("statm",     S_IRUGO, proc_pid_statm),
3342         REG("maps",      S_IRUGO, proc_pid_maps_operations),
3343 #ifdef CONFIG_PROC_CHILDREN
3344         REG("children",  S_IRUGO, proc_tid_children_operations),
3345 #endif
3346 #ifdef CONFIG_NUMA
3347         REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
3348 #endif
3349         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3350         LNK("cwd",       proc_cwd_link),
3351         LNK("root",      proc_root_link),
3352         LNK("exe",       proc_exe_link),
3353         REG("mounts",    S_IRUGO, proc_mounts_operations),
3354         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3355 #ifdef CONFIG_PROC_PAGE_MONITOR
3356         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3357         REG("smaps",     S_IRUGO, proc_pid_smaps_operations),
3358         REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3359         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3360 #endif
3361 #ifdef CONFIG_SECURITY
3362         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3363 #endif
3364 #ifdef CONFIG_KALLSYMS
3365         ONE("wchan",     S_IRUGO, proc_pid_wchan),
3366 #endif
3367 #ifdef CONFIG_STACKTRACE
3368         ONE("stack",      S_IRUSR, proc_pid_stack),
3369 #endif
3370 #ifdef CONFIG_SCHED_INFO
3371         ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3372 #endif
3373 #ifdef CONFIG_LATENCYTOP
3374         REG("latency",  S_IRUGO, proc_lstats_operations),
3375 #endif
3376 #ifdef CONFIG_PROC_PID_CPUSET
3377         ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3378 #endif
3379 #ifdef CONFIG_CGROUPS
3380         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3381 #endif
3382         ONE("oom_score", S_IRUGO, proc_oom_score),
3383         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3384         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3385 #ifdef CONFIG_AUDITSYSCALL
3386         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3387         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3388 #endif
3389 #ifdef CONFIG_FAULT_INJECTION
3390         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3391         REG("fail-nth", 0644, proc_fail_nth_operations),
3392 #endif
3393 #ifdef CONFIG_TASK_IO_ACCOUNTING
3394         ONE("io",       S_IRUSR, proc_tid_io_accounting),
3395 #endif
3396 #ifdef CONFIG_USER_NS
3397         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3398         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3399         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3400         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3401 #endif
3402 #ifdef CONFIG_LIVEPATCH
3403         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
3404 #endif
3405 };
3406
3407 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3408 {
3409         return proc_pident_readdir(file, ctx,
3410                                    tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3411 }
3412
3413 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3414 {
3415         return proc_pident_lookup(dir, dentry,
3416                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3417 }
3418
3419 static const struct file_operations proc_tid_base_operations = {
3420         .read           = generic_read_dir,
3421         .iterate_shared = proc_tid_base_readdir,
3422         .llseek         = generic_file_llseek,
3423 };
3424
3425 static const struct inode_operations proc_tid_base_inode_operations = {
3426         .lookup         = proc_tid_base_lookup,
3427         .getattr        = pid_getattr,
3428         .setattr        = proc_setattr,
3429 };
3430
3431 static struct dentry *proc_task_instantiate(struct dentry *dentry,
3432         struct task_struct *task, const void *ptr)
3433 {
3434         struct inode *inode;
3435         inode = proc_pid_make_inode(dentry->d_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3436         if (!inode)
3437                 return ERR_PTR(-ENOENT);
3438
3439         inode->i_op = &proc_tid_base_inode_operations;
3440         inode->i_fop = &proc_tid_base_operations;
3441         inode->i_flags |= S_IMMUTABLE;
3442
3443         set_nlink(inode, nlink_tid);
3444         pid_update_inode(task, inode);
3445
3446         d_set_d_op(dentry, &pid_dentry_operations);
3447         return d_splice_alias(inode, dentry);
3448 }
3449
3450 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3451 {
3452         struct task_struct *task;
3453         struct task_struct *leader = get_proc_task(dir);
3454         unsigned tid;
3455         struct pid_namespace *ns;
3456         struct dentry *result = ERR_PTR(-ENOENT);
3457
3458         if (!leader)
3459                 goto out_no_task;
3460
3461         tid = name_to_int(&dentry->d_name);
3462         if (tid == ~0U)
3463                 goto out;
3464
3465         ns = dentry->d_sb->s_fs_info;
3466         rcu_read_lock();
3467         task = find_task_by_pid_ns(tid, ns);
3468         if (task)
3469                 get_task_struct(task);
3470         rcu_read_unlock();
3471         if (!task)
3472                 goto out;
3473         if (!same_thread_group(leader, task))
3474                 goto out_drop_task;
3475
3476         result = proc_task_instantiate(dentry, task, NULL);
3477 out_drop_task:
3478         put_task_struct(task);
3479 out:
3480         put_task_struct(leader);
3481 out_no_task:
3482         return result;
3483 }
3484
3485 /*
3486  * Find the first tid of a thread group to return to user space.
3487  *
3488  * Usually this is just the thread group leader, but if the users
3489  * buffer was too small or there was a seek into the middle of the
3490  * directory we have more work todo.
3491  *
3492  * In the case of a short read we start with find_task_by_pid.
3493  *
3494  * In the case of a seek we start with the leader and walk nr
3495  * threads past it.
3496  */
3497 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3498                                         struct pid_namespace *ns)
3499 {
3500         struct task_struct *pos, *task;
3501         unsigned long nr = f_pos;
3502
3503         if (nr != f_pos)        /* 32bit overflow? */
3504                 return NULL;
3505
3506         rcu_read_lock();
3507         task = pid_task(pid, PIDTYPE_PID);
3508         if (!task)
3509                 goto fail;
3510
3511         /* Attempt to start with the tid of a thread */
3512         if (tid && nr) {
3513                 pos = find_task_by_pid_ns(tid, ns);
3514                 if (pos && same_thread_group(pos, task))
3515                         goto found;
3516         }
3517
3518         /* If nr exceeds the number of threads there is nothing todo */
3519         if (nr >= get_nr_threads(task))
3520                 goto fail;
3521
3522         /* If we haven't found our starting place yet start
3523          * with the leader and walk nr threads forward.
3524          */
3525         pos = task = task->group_leader;
3526         do {
3527                 if (!nr--)
3528                         goto found;
3529         } while_each_thread(task, pos);
3530 fail:
3531         pos = NULL;
3532         goto out;
3533 found:
3534         get_task_struct(pos);
3535 out:
3536         rcu_read_unlock();
3537         return pos;
3538 }
3539
3540 /*
3541  * Find the next thread in the thread list.
3542  * Return NULL if there is an error or no next thread.
3543  *
3544  * The reference to the input task_struct is released.
3545  */
3546 static struct task_struct *next_tid(struct task_struct *start)
3547 {
3548         struct task_struct *pos = NULL;
3549         rcu_read_lock();
3550         if (pid_alive(start)) {
3551                 pos = next_thread(start);
3552                 if (thread_group_leader(pos))
3553                         pos = NULL;
3554                 else
3555                         get_task_struct(pos);
3556         }
3557         rcu_read_unlock();
3558         put_task_struct(start);
3559         return pos;
3560 }
3561
3562 /* for the /proc/TGID/task/ directories */
3563 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3564 {
3565         struct inode *inode = file_inode(file);
3566         struct task_struct *task;
3567         struct pid_namespace *ns;
3568         int tid;
3569
3570         if (proc_inode_is_dead(inode))
3571                 return -ENOENT;
3572
3573         if (!dir_emit_dots(file, ctx))
3574                 return 0;
3575
3576         /* f_version caches the tgid value that the last readdir call couldn't
3577          * return. lseek aka telldir automagically resets f_version to 0.
3578          */
3579         ns = proc_pid_ns(inode);
3580         tid = (int)file->f_version;
3581         file->f_version = 0;
3582         for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3583              task;
3584              task = next_tid(task), ctx->pos++) {
3585                 char name[10 + 1];
3586                 unsigned int len;
3587                 tid = task_pid_nr_ns(task, ns);
3588                 len = snprintf(name, sizeof(name), "%u", tid);
3589                 if (!proc_fill_cache(file, ctx, name, len,
3590                                 proc_task_instantiate, task, NULL)) {
3591                         /* returning this tgid failed, save it as the first
3592                          * pid for the next readir call */
3593                         file->f_version = (u64)tid;
3594                         put_task_struct(task);
3595                         break;
3596                 }
3597         }
3598
3599         return 0;
3600 }
3601
3602 static int proc_task_getattr(const struct path *path, struct kstat *stat,
3603                              u32 request_mask, unsigned int query_flags)
3604 {
3605         struct inode *inode = d_inode(path->dentry);
3606         struct task_struct *p = get_proc_task(inode);
3607         generic_fillattr(inode, stat);
3608
3609         if (p) {
3610                 stat->nlink += get_nr_threads(p);
3611                 put_task_struct(p);
3612         }
3613
3614         return 0;
3615 }
3616
3617 static const struct inode_operations proc_task_inode_operations = {
3618         .lookup         = proc_task_lookup,
3619         .getattr        = proc_task_getattr,
3620         .setattr        = proc_setattr,
3621         .permission     = proc_pid_permission,
3622 };
3623
3624 static const struct file_operations proc_task_operations = {
3625         .read           = generic_read_dir,
3626         .iterate_shared = proc_task_readdir,
3627         .llseek         = generic_file_llseek,
3628 };
3629
3630 void __init set_proc_pid_nlink(void)
3631 {
3632         nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3633         nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3634 }