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 proc_pid_cmdline_read(struct file *file, char __user *buf,
 209                                      size_t _count, loff_t *pos)
 210 {
 211         struct task_struct *tsk;
 212         struct mm_struct *mm;
 213         char *page;
 214         unsigned long count = _count;
 215         unsigned long arg_start, arg_end, env_start, env_end;
 216         unsigned long len1, len2, len;
 217         unsigned long p;
 218         char c;
 219         ssize_t rv;
 220
 221         BUG_ON(*pos < 0);
 222
 223         tsk = get_proc_task(file_inode(file));
 224         if (!tsk)
 225                 return -ESRCH;
 226         mm = get_task_mm(tsk);
 227         put_task_struct(tsk);
 228         if (!mm)
 229                 return 0;
 230         /* Check if process spawned far enough to have cmdline. */
 231         if (!mm->env_end) {
 232                 rv = 0;
 233                 goto out_mmput;
 234         }
 235
 236         page = (char *)__get_free_page(GFP_KERNEL);
 237         if (!page) {
 238                 rv = -ENOMEM;
 239                 goto out_mmput;
 240         }
 241
 242         down_read(&mm->mmap_sem);
 243         arg_start = mm->arg_start;
 244         arg_end = mm->arg_end;
 245         env_start = mm->env_start;
 246         env_end = mm->env_end;
 247         up_read(&mm->mmap_sem);
 248
 249         BUG_ON(arg_start > arg_end);
 250         BUG_ON(env_start > env_end);
 251
 252         len1 = arg_end - arg_start;
 253         len2 = env_end - env_start;
 254
 255         /* Empty ARGV. */
 256         if (len1 == 0) {
 257                 rv = 0;
 258                 goto out_free_page;
 259         }
 260         /*
 261          * Inherently racy -- command line shares address space
 262          * with code and data.
 263          */
 264         rv = access_remote_vm(mm, arg_end - 1, &c, 1, 0);
 265         if (rv <= 0)
 266                 goto out_free_page;
 267
 268         rv = 0;
 269
 270         if (c == '\0') {
 271                 /* Command line (set of strings) occupies whole ARGV. */
 272                 if (len1 <= *pos)
 273                         goto out_free_page;
 274
 275                 p = arg_start + *pos;
 276                 len = len1 - *pos;
 277                 while (count > 0 && len > 0) {
 278                         unsigned int _count;
 279                         int nr_read;
 280
 281                         _count = min3(count, len, PAGE_SIZE);
 282                         nr_read = access_remote_vm(mm, p, page, _count, 0);
 283                         if (nr_read < 0)
 284                                 rv = nr_read;
 285                         if (nr_read <= 0)
 286                                 goto out_free_page;
 287
 288                         if (copy_to_user(buf, page, nr_read)) {
 289                                 rv = -EFAULT;
 290                                 goto out_free_page;
 291                         }
 292
 293                         p       += nr_read;
 294                         len     -= nr_read;
 295                         buf     += nr_read;
 296                         count   -= nr_read;
 297                         rv      += nr_read;
 298                 }
 299         } else {
 300                 /*
 301                  * Command line (1 string) occupies ARGV and
 302                  * extends into ENVP.
 303                  */
 304                 struct {
 305                         unsigned long p;
 306                         unsigned long len;
 307                 } cmdline[2] = {
 308                         { .p = arg_start, .len = len1 },
 309                         { .p = env_start, .len = len2 },
 310                 };
 311                 loff_t pos1 = *pos;
 312                 unsigned int i;
 313
 314                 i = 0;
 315                 while (i < 2 && pos1 >= cmdline[i].len) {
 316                         pos1 -= cmdline[i].len;
 317                         i++;
 318                 }
 319                 while (i < 2) {
 320                         p = cmdline[i].p + pos1;
 321                         len = cmdline[i].len - pos1;
 322                         while (count > 0 && len > 0) {
 323                                 unsigned int _count, l;
 324                                 int nr_read;
 325                                 bool final;
 326
 327                                 _count = min3(count, len, PAGE_SIZE);
 328                                 nr_read = access_remote_vm(mm, p, page, _count, 0);
 329                                 if (nr_read < 0)
 330                                         rv = nr_read;
 331                                 if (nr_read <= 0)
 332                                         goto out_free_page;
 333
 334                                 /*
 335                                  * Command line can be shorter than whole ARGV
 336                                  * even if last "marker" byte says it is not.
 337                                  */
 338                                 final = false;
 339                                 l = strnlen(page, nr_read);
 340                                 if (l < nr_read) {
 341                                         nr_read = l;
 342                                         final = true;
 343                                 }
 344
 345                                 if (copy_to_user(buf, page, nr_read)) {
 346                                         rv = -EFAULT;
 347                                         goto out_free_page;
 348                                 }
 349
 350                                 p       += nr_read;
 351                                 len     -= nr_read;
 352                                 buf     += nr_read;
 353                                 count   -= nr_read;
 354                                 rv      += nr_read;
 355
 356                                 if (final)
 357                                         goto out_free_page;
 358                         }
 359
 360                         /* Only first chunk can be read partially. */
 361                         pos1 = 0;
 362                         i++;
 363                 }
 364         }
 365
 366 out_free_page:
 367         free_page((unsigned long)page);
 368 out_mmput:
 369         mmput(mm);
 370         if (rv > 0)
 371                 *pos += rv;
 372         return rv;
 373 }
 374
 375 static const struct file_operations proc_pid_cmdline_ops = {
 376         .read   = proc_pid_cmdline_read,
 377         .llseek = generic_file_llseek,
 378 };
 379
 380 #ifdef CONFIG_KALLSYMS
 381 /*
 382  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 383  * Returns the resolved symbol.  If that fails, simply return the address.
 384  */
 385 static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
 386                           struct pid *pid, struct task_struct *task)
 387 {
 388         unsigned long wchan;
 389         char symname[KSYM_NAME_LEN];
 390
 391         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
 392                 goto print0;
 393
 394         wchan = get_wchan(task);
 395         if (wchan && !lookup_symbol_name(wchan, symname)) {
 396                 seq_puts(m, symname);
 397                 return 0;
 398         }
 399
 400 print0:
 401         seq_putc(m, '0');
 402         return 0;
 403 }
 404 #endif /* CONFIG_KALLSYMS */
 405
 406 static int lock_trace(struct task_struct *task)
 407 {
 408         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
 409         if (err)
 410                 return err;
 411         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
 412                 mutex_unlock(&task->signal->cred_guard_mutex);
 413                 return -EPERM;
 414         }
 415         return 0;
 416 }
 417
 418 static void unlock_trace(struct task_struct *task)
 419 {
 420         mutex_unlock(&task->signal->cred_guard_mutex);
 421 }
 422
 423 #ifdef CONFIG_STACKTRACE
 424
 425 #define MAX_STACK_TRACE_DEPTH   64
 426
 427 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 428                           struct pid *pid, struct task_struct *task)
 429 {
 430         struct stack_trace trace;
 431         unsigned long *entries;
 432         int err;
 433         int i;
 434
 435         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 436         if (!entries)
 437                 return -ENOMEM;
 438
 439         trace.nr_entries        = 0;
 440         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 441         trace.entries           = entries;
 442         trace.skip              = 0;
 443
 444         err = lock_trace(task);
 445         if (!err) {
 446                 save_stack_trace_tsk(task, &trace);
 447
 448                 for (i = 0; i < trace.nr_entries; i++) {
 449                         seq_printf(m, "[<0>] %pB\n", (void *)entries[i]);
 450                 }
 451                 unlock_trace(task);
 452         }
 453         kfree(entries);
 454
 455         return err;
 456 }
 457 #endif
 458
 459 #ifdef CONFIG_SCHED_INFO
 460 /*
 461  * Provides /proc/PID/schedstat
 462  */
 463 static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
 464                               struct pid *pid, struct task_struct *task)
 465 {
 466         if (unlikely(!sched_info_on()))
 467                 seq_printf(m, "0 0 0\n");
 468         else
 469                 seq_printf(m, "%llu %llu %lu\n",
 470                    (unsigned long long)task->se.sum_exec_runtime,
 471                    (unsigned long long)task->sched_info.run_delay,
 472                    task->sched_info.pcount);
 473
 474         return 0;
 475 }
 476 #endif
 477
 478 #ifdef CONFIG_LATENCYTOP
 479 static int lstats_show_proc(struct seq_file *m, void *v)
 480 {
 481         int i;
 482         struct inode *inode = m->private;
 483         struct task_struct *task = get_proc_task(inode);
 484
 485         if (!task)
 486                 return -ESRCH;
 487         seq_puts(m, "Latency Top version : v0.1\n");
 488         for (i = 0; i < 32; i++) {
 489                 struct latency_record *lr = &task->latency_record[i];
 490                 if (lr->backtrace[0]) {
 491                         int q;
 492                         seq_printf(m, "%i %li %li",
 493                                    lr->count, lr->time, lr->max);
 494                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 495                                 unsigned long bt = lr->backtrace[q];
 496                                 if (!bt)
 497                                         break;
 498                                 if (bt == ULONG_MAX)
 499                                         break;
 500                                 seq_printf(m, " %ps", (void *)bt);
 501                         }
 502                         seq_putc(m, '\n');
 503                 }
 504
 505         }
 506         put_task_struct(task);
 507         return 0;
 508 }
 509
 510 static int lstats_open(struct inode *inode, struct file *file)
 511 {
 512         return single_open(file, lstats_show_proc, inode);
 513 }
 514
 515 static ssize_t lstats_write(struct file *file, const char __user *buf,
 516                             size_t count, loff_t *offs)
 517 {
 518         struct task_struct *task = get_proc_task(file_inode(file));
 519
 520         if (!task)
 521                 return -ESRCH;
 522         clear_all_latency_tracing(task);
 523         put_task_struct(task);
 524
 525         return count;
 526 }
 527
 528 static const struct file_operations proc_lstats_operations = {
 529         .open           = lstats_open,
 530         .read           = seq_read,
 531         .write          = lstats_write,
 532         .llseek         = seq_lseek,
 533         .release        = single_release,
 534 };
 535
 536 #endif
 537
 538 static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
 539                           struct pid *pid, struct task_struct *task)
 540 {
 541         unsigned long totalpages = totalram_pages + total_swap_pages;
 542         unsigned long points = 0;
 543
 544         points = oom_badness(task, NULL, NULL, totalpages) *
 545                                         1000 / totalpages;
 546         seq_printf(m, "%lu\n", points);
 547
 548         return 0;
 549 }
 550
 551 struct limit_names {
 552         const char *name;
 553         const char *unit;
 554 };
 555
 556 static const struct limit_names lnames[RLIM_NLIMITS] = {
 557         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 558         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 559         [RLIMIT_DATA] = {"Max data size", "bytes"},
 560         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 561         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 562         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 563         [RLIMIT_NPROC] = {"Max processes", "processes"},
 564         [RLIMIT_NOFILE] = {"Max open files", "files"},
 565         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 566         [RLIMIT_AS] = {"Max address space", "bytes"},
 567         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 568         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 569         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 570         [RLIMIT_NICE] = {"Max nice priority", NULL},
 571         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 572         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 573 };
 574
 575 /* Display limits for a process */
 576 static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
 577                            struct pid *pid, struct task_struct *task)
 578 {
 579         unsigned int i;
 580         unsigned long flags;
 581
 582         struct rlimit rlim[RLIM_NLIMITS];
 583
 584         if (!lock_task_sighand(task, &flags))
 585                 return 0;
 586         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 587         unlock_task_sighand(task, &flags);
 588
 589         /*
 590          * print the file header
 591          */
 592        seq_printf(m, "%-25s %-20s %-20s %-10s\n",
 593                   "Limit", "Soft Limit", "Hard Limit", "Units");
 594
 595         for (i = 0; i < RLIM_NLIMITS; i++) {
 596                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 597                         seq_printf(m, "%-25s %-20s ",
 598                                    lnames[i].name, "unlimited");
 599                 else
 600                         seq_printf(m, "%-25s %-20lu ",
 601                                    lnames[i].name, rlim[i].rlim_cur);
 602
 603                 if (rlim[i].rlim_max == RLIM_INFINITY)
 604                         seq_printf(m, "%-20s ", "unlimited");
 605                 else
 606                         seq_printf(m, "%-20lu ", rlim[i].rlim_max);
 607
 608                 if (lnames[i].unit)
 609                         seq_printf(m, "%-10s\n", lnames[i].unit);
 610                 else
 611                         seq_putc(m, '\n');
 612         }
 613
 614         return 0;
 615 }
 616
 617 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 618 static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
 619                             struct pid *pid, struct task_struct *task)
 620 {
 621         long nr;
 622         unsigned long args[6], sp, pc;
 623         int res;
 624
 625         res = lock_trace(task);
 626         if (res)
 627                 return res;
 628
 629         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 630                 seq_puts(m, "running\n");
 631         else if (nr < 0)
 632                 seq_printf(m, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 633         else
 634                 seq_printf(m,
 635                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 636                        nr,
 637                        args[0], args[1], args[2], args[3], args[4], args[5],
 638                        sp, pc);
 639         unlock_trace(task);
 640
 641         return 0;
 642 }
 643 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 644
 645 /************************************************************************/
 646 /*                       Here the fs part begins                        */
 647 /************************************************************************/
 648
 649 /* permission checks */
 650 static int proc_fd_access_allowed(struct inode *inode)
 651 {
 652         struct task_struct *task;
 653         int allowed = 0;
 654         /* Allow access to a task's file descriptors if it is us or we
 655          * may use ptrace attach to the process and find out that
 656          * information.
 657          */
 658         task = get_proc_task(inode);
 659         if (task) {
 660                 allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 661                 put_task_struct(task);
 662         }
 663         return allowed;
 664 }
 665
 666 int proc_setattr(struct dentry *dentry, struct iattr *attr)
 667 {
 668         int error;
 669         struct inode *inode = d_inode(dentry);
 670
 671         if (attr->ia_valid & ATTR_MODE)
 672                 return -EPERM;
 673
 674         error = setattr_prepare(dentry, attr);
 675         if (error)
 676                 return error;
 677
 678         setattr_copy(inode, attr);
 679         mark_inode_dirty(inode);
 680         return 0;
 681 }
 682
 683 /*
 684  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
 685  * or euid/egid (for hide_pid_min=2)?
 686  */
 687 static bool has_pid_permissions(struct pid_namespace *pid,
 688                                  struct task_struct *task,
 689                                  int hide_pid_min)
 690 {
 691         if (pid->hide_pid < hide_pid_min)
 692                 return true;
 693         if (in_group_p(pid->pid_gid))
 694                 return true;
 695         return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 696 }
 697
 698
 699 static int proc_pid_permission(struct inode *inode, int mask)
 700 {
 701         struct pid_namespace *pid = inode->i_sb->s_fs_info;
 702         struct task_struct *task;
 703         bool has_perms;
 704
 705         task = get_proc_task(inode);
 706         if (!task)
 707                 return -ESRCH;
 708         has_perms = has_pid_permissions(pid, task, HIDEPID_NO_ACCESS);
 709         put_task_struct(task);
 710
 711         if (!has_perms) {
 712                 if (pid->hide_pid == HIDEPID_INVISIBLE) {
 713                         /*
 714                          * Let's make getdents(), stat(), and open()
 715                          * consistent with each other.  If a process
 716                          * may not stat() a file, it shouldn't be seen
 717                          * in procfs at all.
 718                          */
 719                         return -ENOENT;
 720                 }
 721
 722                 return -EPERM;
 723         }
 724         return generic_permission(inode, mask);
 725 }
 726
 727
 728
 729 static const struct inode_operations proc_def_inode_operations = {
 730         .setattr        = proc_setattr,
 731 };
 732
 733 static int proc_single_show(struct seq_file *m, void *v)
 734 {
 735         struct inode *inode = m->private;
 736         struct pid_namespace *ns;
 737         struct pid *pid;
 738         struct task_struct *task;
 739         int ret;
 740
 741         ns = inode->i_sb->s_fs_info;
 742         pid = proc_pid(inode);
 743         task = get_pid_task(pid, PIDTYPE_PID);
 744         if (!task)
 745                 return -ESRCH;
 746
 747         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 748
 749         put_task_struct(task);
 750         return ret;
 751 }
 752
 753 static int proc_single_open(struct inode *inode, struct file *filp)
 754 {
 755         return single_open(filp, proc_single_show, inode);
 756 }
 757
 758 static const struct file_operations proc_single_file_operations = {
 759         .open           = proc_single_open,
 760         .read           = seq_read,
 761         .llseek         = seq_lseek,
 762         .release        = single_release,
 763 };
 764
 765
 766 struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
 767 {
 768         struct task_struct *task = get_proc_task(inode);
 769         struct mm_struct *mm = ERR_PTR(-ESRCH);
 770
 771         if (task) {
 772                 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
 773                 put_task_struct(task);
 774
 775                 if (!IS_ERR_OR_NULL(mm)) {
 776                         /* ensure this mm_struct can't be freed */
 777                         mmgrab(mm);
 778                         /* but do not pin its memory */
 779                         mmput(mm);
 780                 }
 781         }
 782
 783         return mm;
 784 }
 785
 786 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
 787 {
 788         struct mm_struct *mm = proc_mem_open(inode, mode);
 789
 790         if (IS_ERR(mm))
 791                 return PTR_ERR(mm);
 792
 793         file->private_data = mm;
 794         return 0;
 795 }
 796
 797 static int mem_open(struct inode *inode, struct file *file)
 798 {
 799         int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
 800
 801         /* OK to pass negative loff_t, we can catch out-of-range */
 802         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 803
 804         return ret;
 805 }
 806
 807 static ssize_t mem_rw(struct file *file, char __user *buf,
 808                         size_t count, loff_t *ppos, int write)
 809 {
 810         struct mm_struct *mm = file->private_data;
 811         unsigned long addr = *ppos;
 812         ssize_t copied;
 813         char *page;
 814         unsigned int flags;
 815
 816         if (!mm)
 817                 return 0;
 818
 819         page = (char *)__get_free_page(GFP_KERNEL);
 820         if (!page)
 821                 return -ENOMEM;
 822
 823         copied = 0;
 824         if (!mmget_not_zero(mm))
 825                 goto free;
 826
 827         flags = FOLL_FORCE | (write ? FOLL_WRITE : 0);
 828
 829         while (count > 0) {
 830                 int this_len = min_t(int, count, PAGE_SIZE);
 831
 832                 if (write && copy_from_user(page, buf, this_len)) {
 833                         copied = -EFAULT;
 834                         break;
 835                 }
 836
 837                 this_len = access_remote_vm(mm, addr, page, this_len, flags);
 838                 if (!this_len) {
 839                         if (!copied)
 840                                 copied = -EIO;
 841                         break;
 842                 }
 843
 844                 if (!write && copy_to_user(buf, page, this_len)) {
 845                         copied = -EFAULT;
 846                         break;
 847                 }
 848
 849                 buf += this_len;
 850                 addr += this_len;
 851                 copied += this_len;
 852                 count -= this_len;
 853         }
 854         *ppos = addr;
 855
 856         mmput(mm);
 857 free:
 858         free_page((unsigned long) page);
 859         return copied;
 860 }
 861
 862 static ssize_t mem_read(struct file *file, char __user *buf,
 863                         size_t count, loff_t *ppos)
 864 {
 865         return mem_rw(file, buf, count, ppos, 0);
 866 }
 867
 868 static ssize_t mem_write(struct file *file, const char __user *buf,
 869                          size_t count, loff_t *ppos)
 870 {
 871         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 872 }
 873
 874 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 875 {
 876         switch (orig) {
 877         case 0:
 878                 file->f_pos = offset;
 879                 break;
 880         case 1:
 881                 file->f_pos += offset;
 882                 break;
 883         default:
 884                 return -EINVAL;
 885         }
 886         force_successful_syscall_return();
 887         return file->f_pos;
 888 }
 889
 890 static int mem_release(struct inode *inode, struct file *file)
 891 {
 892         struct mm_struct *mm = file->private_data;
 893         if (mm)
 894                 mmdrop(mm);
 895         return 0;
 896 }
 897
 898 static const struct file_operations proc_mem_operations = {
 899         .llseek         = mem_lseek,
 900         .read           = mem_read,
 901         .write          = mem_write,
 902         .open           = mem_open,
 903         .release        = mem_release,
 904 };
 905
 906 static int environ_open(struct inode *inode, struct file *file)
 907 {
 908         return __mem_open(inode, file, PTRACE_MODE_READ);
 909 }
 910
 911 static ssize_t environ_read(struct file *file, char __user *buf,
 912                         size_t count, loff_t *ppos)
 913 {
 914         char *page;
 915         unsigned long src = *ppos;
 916         int ret = 0;
 917         struct mm_struct *mm = file->private_data;
 918         unsigned long env_start, env_end;
 919
 920         /* Ensure the process spawned far enough to have an environment. */
 921         if (!mm || !mm->env_end)
 922                 return 0;
 923
 924         page = (char *)__get_free_page(GFP_KERNEL);
 925         if (!page)
 926                 return -ENOMEM;
 927
 928         ret = 0;
 929         if (!mmget_not_zero(mm))
 930                 goto free;
 931
 932         down_read(&mm->mmap_sem);
 933         env_start = mm->env_start;
 934         env_end = mm->env_end;
 935         up_read(&mm->mmap_sem);
 936
 937         while (count > 0) {
 938                 size_t this_len, max_len;
 939                 int retval;
 940
 941                 if (src >= (env_end - env_start))
 942                         break;
 943
 944                 this_len = env_end - (env_start + src);
 945
 946                 max_len = min_t(size_t, PAGE_SIZE, count);
 947                 this_len = min(max_len, this_len);
 948
 949                 retval = access_remote_vm(mm, (env_start + src), page, this_len, 0);
 950
 951                 if (retval <= 0) {
 952                         ret = retval;
 953                         break;
 954                 }
 955
 956                 if (copy_to_user(buf, page, retval)) {
 957                         ret = -EFAULT;
 958                         break;
 959                 }
 960
 961                 ret += retval;
 962                 src += retval;
 963                 buf += retval;
 964                 count -= retval;
 965         }
 966         *ppos = src;
 967         mmput(mm);
 968
 969 free:
 970         free_page((unsigned long) page);
 971         return ret;
 972 }
 973
 974 static const struct file_operations proc_environ_operations = {
 975         .open           = environ_open,
 976         .read           = environ_read,
 977         .llseek         = generic_file_llseek,
 978         .release        = mem_release,
 979 };
 980
 981 static int auxv_open(struct inode *inode, struct file *file)
 982 {
 983         return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
 984 }
 985
 986 static ssize_t auxv_read(struct file *file, char __user *buf,
 987                         size_t count, loff_t *ppos)
 988 {
 989         struct mm_struct *mm = file->private_data;
 990         unsigned int nwords = 0;
 991
 992         if (!mm)
 993                 return 0;
 994         do {
 995                 nwords += 2;
 996         } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 997         return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
 998                                        nwords * sizeof(mm->saved_auxv[0]));
 999 }
1000
1001 static const struct file_operations proc_auxv_operations = {
1002         .open           = auxv_open,
1003         .read           = auxv_read,
1004         .llseek         = generic_file_llseek,
1005         .release        = mem_release,
1006 };
1007
1008 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1009                             loff_t *ppos)
1010 {
1011         struct task_struct *task = get_proc_task(file_inode(file));
1012         char buffer[PROC_NUMBUF];
1013         int oom_adj = OOM_ADJUST_MIN;
1014         size_t len;
1015
1016         if (!task)
1017                 return -ESRCH;
1018         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1019                 oom_adj = OOM_ADJUST_MAX;
1020         else
1021                 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1022                           OOM_SCORE_ADJ_MAX;
1023         put_task_struct(task);
1024         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1025         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1026 }
1027
1028 static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1029 {
1030         static DEFINE_MUTEX(oom_adj_mutex);
1031         struct mm_struct *mm = NULL;
1032         struct task_struct *task;
1033         int err = 0;
1034
1035         task = get_proc_task(file_inode(file));
1036         if (!task)
1037                 return -ESRCH;
1038
1039         mutex_lock(&oom_adj_mutex);
1040         if (legacy) {
1041                 if (oom_adj < task->signal->oom_score_adj &&
1042                                 !capable(CAP_SYS_RESOURCE)) {
1043                         err = -EACCES;
1044                         goto err_unlock;
1045                 }
1046                 /*
1047                  * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1048                  * /proc/pid/oom_score_adj instead.
1049                  */
1050                 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1051                           current->comm, task_pid_nr(current), task_pid_nr(task),
1052                           task_pid_nr(task));
1053         } else {
1054                 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1055                                 !capable(CAP_SYS_RESOURCE)) {
1056                         err = -EACCES;
1057                         goto err_unlock;
1058                 }
1059         }
1060
1061         /*
1062          * Make sure we will check other processes sharing the mm if this is
1063          * not vfrok which wants its own oom_score_adj.
1064          * pin the mm so it doesn't go away and get reused after task_unlock
1065          */
1066         if (!task->vfork_done) {
1067                 struct task_struct *p = find_lock_task_mm(task);
1068
1069                 if (p) {
1070                         if (atomic_read(&p->mm->mm_users) > 1) {
1071                                 mm = p->mm;
1072                                 mmgrab(mm);
1073                         }
1074                         task_unlock(p);
1075                 }
1076         }
1077
1078         task->signal->oom_score_adj = oom_adj;
1079         if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1080                 task->signal->oom_score_adj_min = (short)oom_adj;
1081         trace_oom_score_adj_update(task);
1082
1083         if (mm) {
1084                 struct task_struct *p;
1085
1086                 rcu_read_lock();
1087                 for_each_process(p) {
1088                         if (same_thread_group(task, p))
1089                                 continue;
1090
1091                         /* do not touch kernel threads or the global init */
1092                         if (p->flags & PF_KTHREAD || is_global_init(p))
1093                                 continue;
1094
1095                         task_lock(p);
1096                         if (!p->vfork_done && process_shares_mm(p, mm)) {
1097                                 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",
1098                                                 task_pid_nr(p), p->comm,
1099                                                 p->signal->oom_score_adj, oom_adj,
1100                                                 task_pid_nr(task), task->comm);
1101                                 p->signal->oom_score_adj = oom_adj;
1102                                 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1103                                         p->signal->oom_score_adj_min = (short)oom_adj;
1104                         }
1105                         task_unlock(p);
1106                 }
1107                 rcu_read_unlock();
1108                 mmdrop(mm);
1109         }
1110 err_unlock:
1111         mutex_unlock(&oom_adj_mutex);
1112         put_task_struct(task);
1113         return err;
1114 }
1115
1116 /*
1117  * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1118  * kernels.  The effective policy is defined by oom_score_adj, which has a
1119  * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1120  * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1121  * Processes that become oom disabled via oom_adj will still be oom disabled
1122  * with this implementation.
1123  *
1124  * oom_adj cannot be removed since existing userspace binaries use it.
1125  */
1126 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1127                              size_t count, loff_t *ppos)
1128 {
1129         char buffer[PROC_NUMBUF];
1130         int oom_adj;
1131         int err;
1132
1133         memset(buffer, 0, sizeof(buffer));
1134         if (count > sizeof(buffer) - 1)
1135                 count = sizeof(buffer) - 1;
1136         if (copy_from_user(buffer, buf, count)) {
1137                 err = -EFAULT;
1138                 goto out;
1139         }
1140
1141         err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1142         if (err)
1143                 goto out;
1144         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1145              oom_adj != OOM_DISABLE) {
1146                 err = -EINVAL;
1147                 goto out;
1148         }
1149
1150         /*
1151          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1152          * value is always attainable.
1153          */
1154         if (oom_adj == OOM_ADJUST_MAX)
1155                 oom_adj = OOM_SCORE_ADJ_MAX;
1156         else
1157                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1158
1159         err = __set_oom_adj(file, oom_adj, true);
1160 out:
1161         return err < 0 ? err : count;
1162 }
1163
1164 static const struct file_operations proc_oom_adj_operations = {
1165         .read           = oom_adj_read,
1166         .write          = oom_adj_write,
1167         .llseek         = generic_file_llseek,
1168 };
1169
1170 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1171                                         size_t count, loff_t *ppos)
1172 {
1173         struct task_struct *task = get_proc_task(file_inode(file));
1174         char buffer[PROC_NUMBUF];
1175         short oom_score_adj = OOM_SCORE_ADJ_MIN;
1176         size_t len;
1177
1178         if (!task)
1179                 return -ESRCH;
1180         oom_score_adj = task->signal->oom_score_adj;
1181         put_task_struct(task);
1182         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1183         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1184 }
1185
1186 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1187                                         size_t count, loff_t *ppos)
1188 {
1189         char buffer[PROC_NUMBUF];
1190         int oom_score_adj;
1191         int err;
1192
1193         memset(buffer, 0, sizeof(buffer));
1194         if (count > sizeof(buffer) - 1)
1195                 count = sizeof(buffer) - 1;
1196         if (copy_from_user(buffer, buf, count)) {
1197                 err = -EFAULT;
1198                 goto out;
1199         }
1200
1201         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1202         if (err)
1203                 goto out;
1204         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1205                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1206                 err = -EINVAL;
1207                 goto out;
1208         }
1209
1210         err = __set_oom_adj(file, oom_score_adj, false);
1211 out:
1212         return err < 0 ? err : count;
1213 }
1214
1215 static const struct file_operations proc_oom_score_adj_operations = {
1216         .read           = oom_score_adj_read,
1217         .write          = oom_score_adj_write,
1218         .llseek         = default_llseek,
1219 };
1220
1221 #ifdef CONFIG_AUDITSYSCALL
1222 #define TMPBUFLEN 11
1223 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1224                                   size_t count, loff_t *ppos)
1225 {
1226         struct inode * inode = file_inode(file);
1227         struct task_struct *task = get_proc_task(inode);
1228         ssize_t length;
1229         char tmpbuf[TMPBUFLEN];
1230
1231         if (!task)
1232                 return -ESRCH;
1233         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1234                            from_kuid(file->f_cred->user_ns,
1235                                      audit_get_loginuid(task)));
1236         put_task_struct(task);
1237         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1238 }
1239
1240 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1241                                    size_t count, loff_t *ppos)
1242 {
1243         struct inode * inode = file_inode(file);
1244         uid_t loginuid;
1245         kuid_t kloginuid;
1246         int rv;
1247
1248         rcu_read_lock();
1249         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1250                 rcu_read_unlock();
1251                 return -EPERM;
1252         }
1253         rcu_read_unlock();
1254
1255         if (*ppos != 0) {
1256                 /* No partial writes. */
1257                 return -EINVAL;
1258         }
1259
1260         rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1261         if (rv < 0)
1262                 return rv;
1263
1264         /* is userspace tring to explicitly UNSET the loginuid? */
1265         if (loginuid == AUDIT_UID_UNSET) {
1266                 kloginuid = INVALID_UID;
1267         } else {
1268                 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1269                 if (!uid_valid(kloginuid))
1270                         return -EINVAL;
1271         }
1272
1273         rv = audit_set_loginuid(kloginuid);
1274         if (rv < 0)
1275                 return rv;
1276         return count;
1277 }
1278
1279 static const struct file_operations proc_loginuid_operations = {
1280         .read           = proc_loginuid_read,
1281         .write          = proc_loginuid_write,
1282         .llseek         = generic_file_llseek,
1283 };
1284
1285 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1286                                   size_t count, loff_t *ppos)
1287 {
1288         struct inode * inode = file_inode(file);
1289         struct task_struct *task = get_proc_task(inode);
1290         ssize_t length;
1291         char tmpbuf[TMPBUFLEN];
1292
1293         if (!task)
1294                 return -ESRCH;
1295         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1296                                 audit_get_sessionid(task));
1297         put_task_struct(task);
1298         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1299 }
1300
1301 static const struct file_operations proc_sessionid_operations = {
1302         .read           = proc_sessionid_read,
1303         .llseek         = generic_file_llseek,
1304 };
1305 #endif
1306
1307 #ifdef CONFIG_FAULT_INJECTION
1308 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1309                                       size_t count, loff_t *ppos)
1310 {
1311         struct task_struct *task = get_proc_task(file_inode(file));
1312         char buffer[PROC_NUMBUF];
1313         size_t len;
1314         int make_it_fail;
1315
1316         if (!task)
1317                 return -ESRCH;
1318         make_it_fail = task->make_it_fail;
1319         put_task_struct(task);
1320
1321         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1322
1323         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1324 }
1325
1326 static ssize_t proc_fault_inject_write(struct file * file,
1327                         const char __user * buf, size_t count, loff_t *ppos)
1328 {
1329         struct task_struct *task;
1330         char buffer[PROC_NUMBUF];
1331         int make_it_fail;
1332         int rv;
1333
1334         if (!capable(CAP_SYS_RESOURCE))
1335                 return -EPERM;
1336         memset(buffer, 0, sizeof(buffer));
1337         if (count > sizeof(buffer) - 1)
1338                 count = sizeof(buffer) - 1;
1339         if (copy_from_user(buffer, buf, count))
1340                 return -EFAULT;
1341         rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1342         if (rv < 0)
1343                 return rv;
1344         if (make_it_fail < 0 || make_it_fail > 1)
1345                 return -EINVAL;
1346
1347         task = get_proc_task(file_inode(file));
1348         if (!task)
1349                 return -ESRCH;
1350         task->make_it_fail = make_it_fail;
1351         put_task_struct(task);
1352
1353         return count;
1354 }
1355
1356 static const struct file_operations proc_fault_inject_operations = {
1357         .read           = proc_fault_inject_read,
1358         .write          = proc_fault_inject_write,
1359         .llseek         = generic_file_llseek,
1360 };
1361
1362 static ssize_t proc_fail_nth_write(struct file *file, const char __user *buf,
1363                                    size_t count, loff_t *ppos)
1364 {
1365         struct task_struct *task;
1366         int err;
1367         unsigned int n;
1368
1369         err = kstrtouint_from_user(buf, count, 0, &n);
1370         if (err)
1371                 return err;
1372
1373         task = get_proc_task(file_inode(file));
1374         if (!task)
1375                 return -ESRCH;
1376         task->fail_nth = n;
1377         put_task_struct(task);
1378
1379         return count;
1380 }
1381
1382 static ssize_t proc_fail_nth_read(struct file *file, char __user *buf,
1383                                   size_t count, loff_t *ppos)
1384 {
1385         struct task_struct *task;
1386         char numbuf[PROC_NUMBUF];
1387         ssize_t len;
1388
1389         task = get_proc_task(file_inode(file));
1390         if (!task)
1391                 return -ESRCH;
1392         len = snprintf(numbuf, sizeof(numbuf), "%u\n", task->fail_nth);
1393         len = simple_read_from_buffer(buf, count, ppos, numbuf, len);
1394         put_task_struct(task);
1395
1396         return len;
1397 }
1398
1399 static const struct file_operations proc_fail_nth_operations = {
1400         .read           = proc_fail_nth_read,
1401         .write          = proc_fail_nth_write,
1402 };
1403 #endif
1404
1405
1406 #ifdef CONFIG_SCHED_DEBUG
1407 /*
1408  * Print out various scheduling related per-task fields:
1409  */
1410 static int sched_show(struct seq_file *m, void *v)
1411 {
1412         struct inode *inode = m->private;
1413         struct pid_namespace *ns = inode->i_sb->s_fs_info;
1414         struct task_struct *p;
1415
1416         p = get_proc_task(inode);
1417         if (!p)
1418                 return -ESRCH;
1419         proc_sched_show_task(p, ns, m);
1420
1421         put_task_struct(p);
1422
1423         return 0;
1424 }
1425
1426 static ssize_t
1427 sched_write(struct file *file, const char __user *buf,
1428             size_t count, loff_t *offset)
1429 {
1430         struct inode *inode = file_inode(file);
1431         struct task_struct *p;
1432
1433         p = get_proc_task(inode);
1434         if (!p)
1435                 return -ESRCH;
1436         proc_sched_set_task(p);
1437
1438         put_task_struct(p);
1439
1440         return count;
1441 }
1442
1443 static int sched_open(struct inode *inode, struct file *filp)
1444 {
1445         return single_open(filp, sched_show, inode);
1446 }
1447
1448 static const struct file_operations proc_pid_sched_operations = {
1449         .open           = sched_open,
1450         .read           = seq_read,
1451         .write          = sched_write,
1452         .llseek         = seq_lseek,
1453         .release        = single_release,
1454 };
1455
1456 #endif
1457
1458 #ifdef CONFIG_SCHED_AUTOGROUP
1459 /*
1460  * Print out autogroup related information:
1461  */
1462 static int sched_autogroup_show(struct seq_file *m, void *v)
1463 {
1464         struct inode *inode = m->private;
1465         struct task_struct *p;
1466
1467         p = get_proc_task(inode);
1468         if (!p)
1469                 return -ESRCH;
1470         proc_sched_autogroup_show_task(p, m);
1471
1472         put_task_struct(p);
1473
1474         return 0;
1475 }
1476
1477 static ssize_t
1478 sched_autogroup_write(struct file *file, const char __user *buf,
1479             size_t count, loff_t *offset)
1480 {
1481         struct inode *inode = file_inode(file);
1482         struct task_struct *p;
1483         char buffer[PROC_NUMBUF];
1484         int nice;
1485         int err;
1486
1487         memset(buffer, 0, sizeof(buffer));
1488         if (count > sizeof(buffer) - 1)
1489                 count = sizeof(buffer) - 1;
1490         if (copy_from_user(buffer, buf, count))
1491                 return -EFAULT;
1492
1493         err = kstrtoint(strstrip(buffer), 0, &nice);
1494         if (err < 0)
1495                 return err;
1496
1497         p = get_proc_task(inode);
1498         if (!p)
1499                 return -ESRCH;
1500
1501         err = proc_sched_autogroup_set_nice(p, nice);
1502         if (err)
1503                 count = err;
1504
1505         put_task_struct(p);
1506
1507         return count;
1508 }
1509
1510 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1511 {
1512         int ret;
1513
1514         ret = single_open(filp, sched_autogroup_show, NULL);
1515         if (!ret) {
1516                 struct seq_file *m = filp->private_data;
1517
1518                 m->private = inode;
1519         }
1520         return ret;
1521 }
1522
1523 static const struct file_operations proc_pid_sched_autogroup_operations = {
1524         .open           = sched_autogroup_open,
1525         .read           = seq_read,
1526         .write          = sched_autogroup_write,
1527         .llseek         = seq_lseek,
1528         .release        = single_release,
1529 };
1530
1531 #endif /* CONFIG_SCHED_AUTOGROUP */
1532
1533 static ssize_t comm_write(struct file *file, const char __user *buf,
1534                                 size_t count, loff_t *offset)
1535 {
1536         struct inode *inode = file_inode(file);
1537         struct task_struct *p;
1538         char buffer[TASK_COMM_LEN];
1539         const size_t maxlen = sizeof(buffer) - 1;
1540
1541         memset(buffer, 0, sizeof(buffer));
1542         if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1543                 return -EFAULT;
1544
1545         p = get_proc_task(inode);
1546         if (!p)
1547                 return -ESRCH;
1548
1549         if (same_thread_group(current, p))
1550                 set_task_comm(p, buffer);
1551         else
1552                 count = -EINVAL;
1553
1554         put_task_struct(p);
1555
1556         return count;
1557 }
1558
1559 static int comm_show(struct seq_file *m, void *v)
1560 {
1561         struct inode *inode = m->private;
1562         struct task_struct *p;
1563
1564         p = get_proc_task(inode);
1565         if (!p)
1566                 return -ESRCH;
1567
1568         task_lock(p);
1569         seq_printf(m, "%s\n", p->comm);
1570         task_unlock(p);
1571
1572         put_task_struct(p);
1573
1574         return 0;
1575 }
1576
1577 static int comm_open(struct inode *inode, struct file *filp)
1578 {
1579         return single_open(filp, comm_show, inode);
1580 }
1581
1582 static const struct file_operations proc_pid_set_comm_operations = {
1583         .open           = comm_open,
1584         .read           = seq_read,
1585         .write          = comm_write,
1586         .llseek         = seq_lseek,
1587         .release        = single_release,
1588 };
1589
1590 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1591 {
1592         struct task_struct *task;
1593         struct file *exe_file;
1594
1595         task = get_proc_task(d_inode(dentry));
1596         if (!task)
1597                 return -ENOENT;
1598         exe_file = get_task_exe_file(task);
1599         put_task_struct(task);
1600         if (exe_file) {
1601                 *exe_path = exe_file->f_path;
1602                 path_get(&exe_file->f_path);
1603                 fput(exe_file);
1604                 return 0;
1605         } else
1606                 return -ENOENT;
1607 }
1608
1609 static const char *proc_pid_get_link(struct dentry *dentry,
1610                                      struct inode *inode,
1611                                      struct delayed_call *done)
1612 {
1613         struct path path;
1614         int error = -EACCES;
1615
1616         if (!dentry)
1617                 return ERR_PTR(-ECHILD);
1618
1619         /* Are we allowed to snoop on the tasks file descriptors? */
1620         if (!proc_fd_access_allowed(inode))
1621                 goto out;
1622
1623         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1624         if (error)
1625                 goto out;
1626
1627         nd_jump_link(&path);
1628         return NULL;
1629 out:
1630         return ERR_PTR(error);
1631 }
1632
1633 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1634 {
1635         char *tmp = (char *)__get_free_page(GFP_KERNEL);
1636         char *pathname;
1637         int len;
1638
1639         if (!tmp)
1640                 return -ENOMEM;
1641
1642         pathname = d_path(path, tmp, PAGE_SIZE);
1643         len = PTR_ERR(pathname);
1644         if (IS_ERR(pathname))
1645                 goto out;
1646         len = tmp + PAGE_SIZE - 1 - pathname;
1647
1648         if (len > buflen)
1649                 len = buflen;
1650         if (copy_to_user(buffer, pathname, len))
1651                 len = -EFAULT;
1652  out:
1653         free_page((unsigned long)tmp);
1654         return len;
1655 }
1656
1657 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1658 {
1659         int error = -EACCES;
1660         struct inode *inode = d_inode(dentry);
1661         struct path path;
1662
1663         /* Are we allowed to snoop on the tasks file descriptors? */
1664         if (!proc_fd_access_allowed(inode))
1665                 goto out;
1666
1667         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1668         if (error)
1669                 goto out;
1670
1671         error = do_proc_readlink(&path, buffer, buflen);
1672         path_put(&path);
1673 out:
1674         return error;
1675 }
1676
1677 const struct inode_operations proc_pid_link_inode_operations = {
1678         .readlink       = proc_pid_readlink,
1679         .get_link       = proc_pid_get_link,
1680         .setattr        = proc_setattr,
1681 };
1682
1683
1684 /* building an inode */
1685
1686 void task_dump_owner(struct task_struct *task, umode_t mode,
1687                      kuid_t *ruid, kgid_t *rgid)
1688 {
1689         /* Depending on the state of dumpable compute who should own a
1690          * proc file for a task.
1691          */
1692         const struct cred *cred;
1693         kuid_t uid;
1694         kgid_t gid;
1695
1696         /* Default to the tasks effective ownership */
1697         rcu_read_lock();
1698         cred = __task_cred(task);
1699         uid = cred->euid;
1700         gid = cred->egid;
1701         rcu_read_unlock();
1702
1703         /*
1704          * Before the /proc/pid/status file was created the only way to read
1705          * the effective uid of a /process was to stat /proc/pid.  Reading
1706          * /proc/pid/status is slow enough that procps and other packages
1707          * kept stating /proc/pid.  To keep the rules in /proc simple I have
1708          * made this apply to all per process world readable and executable
1709          * directories.
1710          */
1711         if (mode != (S_IFDIR|S_IRUGO|S_IXUGO)) {
1712                 struct mm_struct *mm;
1713                 task_lock(task);
1714                 mm = task->mm;
1715                 /* Make non-dumpable tasks owned by some root */
1716                 if (mm) {
1717                         if (get_dumpable(mm) != SUID_DUMP_USER) {
1718                                 struct user_namespace *user_ns = mm->user_ns;
1719
1720                                 uid = make_kuid(user_ns, 0);
1721                                 if (!uid_valid(uid))
1722                                         uid = GLOBAL_ROOT_UID;
1723
1724                                 gid = make_kgid(user_ns, 0);
1725                                 if (!gid_valid(gid))
1726                                         gid = GLOBAL_ROOT_GID;
1727                         }
1728                 } else {
1729                         uid = GLOBAL_ROOT_UID;
1730                         gid = GLOBAL_ROOT_GID;
1731                 }
1732                 task_unlock(task);
1733         }
1734         *ruid = uid;
1735         *rgid = gid;
1736 }
1737
1738 struct inode *proc_pid_make_inode(struct super_block * sb,
1739                                   struct task_struct *task, umode_t mode)
1740 {
1741         struct inode * inode;
1742         struct proc_inode *ei;
1743
1744         /* We need a new inode */
1745
1746         inode = new_inode(sb);
1747         if (!inode)
1748                 goto out;
1749
1750         /* Common stuff */
1751         ei = PROC_I(inode);
1752         inode->i_mode = mode;
1753         inode->i_ino = get_next_ino();
1754         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1755         inode->i_op = &proc_def_inode_operations;
1756
1757         /*
1758          * grab the reference to task.
1759          */
1760         ei->pid = get_task_pid(task, PIDTYPE_PID);
1761         if (!ei->pid)
1762                 goto out_unlock;
1763
1764         task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1765         security_task_to_inode(task, inode);
1766
1767 out:
1768         return inode;
1769
1770 out_unlock:
1771         iput(inode);
1772         return NULL;
1773 }
1774
1775 int pid_getattr(const struct path *path, struct kstat *stat,
1776                 u32 request_mask, unsigned int query_flags)
1777 {
1778         struct inode *inode = d_inode(path->dentry);
1779         struct task_struct *task;
1780         struct pid_namespace *pid = path->dentry->d_sb->s_fs_info;
1781
1782         generic_fillattr(inode, stat);
1783
1784         rcu_read_lock();
1785         stat->uid = GLOBAL_ROOT_UID;
1786         stat->gid = GLOBAL_ROOT_GID;
1787         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1788         if (task) {
1789                 if (!has_pid_permissions(pid, task, HIDEPID_INVISIBLE)) {
1790                         rcu_read_unlock();
1791                         /*
1792                          * This doesn't prevent learning whether PID exists,
1793                          * it only makes getattr() consistent with readdir().
1794                          */
1795                         return -ENOENT;
1796                 }
1797                 task_dump_owner(task, inode->i_mode, &stat->uid, &stat->gid);
1798         }
1799         rcu_read_unlock();
1800         return 0;
1801 }
1802
1803 /* dentry stuff */
1804
1805 /*
1806  *      Exceptional case: normally we are not allowed to unhash a busy
1807  * directory. In this case, however, we can do it - no aliasing problems
1808  * due to the way we treat inodes.
1809  *
1810  * Rewrite the inode's ownerships here because the owning task may have
1811  * performed a setuid(), etc.
1812  *
1813  */
1814 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                 task_dump_owner(task, inode->i_mode, &inode->i_uid, &inode->i_gid);
1827
1828                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1829                 security_task_to_inode(task, inode);
1830                 put_task_struct(task);
1831                 return 1;
1832         }
1833         return 0;
1834 }
1835
1836 static inline bool proc_inode_is_dead(struct inode *inode)
1837 {
1838         return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1839 }
1840
1841 int pid_delete_dentry(const struct dentry *dentry)
1842 {
1843         /* Is the task we represent dead?
1844          * If so, then don't put the dentry on the lru list,
1845          * kill it immediately.
1846          */
1847         return proc_inode_is_dead(d_inode(dentry));
1848 }
1849
1850 const struct dentry_operations pid_dentry_operations =
1851 {
1852         .d_revalidate   = pid_revalidate,
1853         .d_delete       = pid_delete_dentry,
1854 };
1855
1856 /* Lookups */
1857
1858 /*
1859  * Fill a directory entry.
1860  *
1861  * If possible create the dcache entry and derive our inode number and
1862  * file type from dcache entry.
1863  *
1864  * Since all of the proc inode numbers are dynamically generated, the inode
1865  * numbers do not exist until the inode is cache.  This means creating the
1866  * the dcache entry in readdir is necessary to keep the inode numbers
1867  * reported by readdir in sync with the inode numbers reported
1868  * by stat.
1869  */
1870 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1871         const char *name, int len,
1872         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1873 {
1874         struct dentry *child, *dir = file->f_path.dentry;
1875         struct qstr qname = QSTR_INIT(name, len);
1876         struct inode *inode;
1877         unsigned type;
1878         ino_t ino;
1879
1880         child = d_hash_and_lookup(dir, &qname);
1881         if (!child) {
1882                 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1883                 child = d_alloc_parallel(dir, &qname, &wq);
1884                 if (IS_ERR(child))
1885                         goto end_instantiate;
1886                 if (d_in_lookup(child)) {
1887                         int err = instantiate(d_inode(dir), child, task, ptr);
1888                         d_lookup_done(child);
1889                         if (err < 0) {
1890                                 dput(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         return dir_emit(ctx, name, len, ino, type);
1900
1901 end_instantiate:
1902         return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1903 }
1904
1905 /*
1906  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1907  * which represent vma start and end addresses.
1908  */
1909 static int dname_to_vma_addr(struct dentry *dentry,
1910                              unsigned long *start, unsigned long *end)
1911 {
1912         const char *str = dentry->d_name.name;
1913         unsigned long long sval, eval;
1914         unsigned int len;
1915
1916         if (str[0] == '0' && str[1] != '-')
1917                 return -EINVAL;
1918         len = _parse_integer(str, 16, &sval);
1919         if (len & KSTRTOX_OVERFLOW)
1920                 return -EINVAL;
1921         if (sval != (unsigned long)sval)
1922                 return -EINVAL;
1923         str += len;
1924
1925         if (*str != '-')
1926                 return -EINVAL;
1927         str++;
1928
1929         if (str[0] == '0' && str[1])
1930                 return -EINVAL;
1931         len = _parse_integer(str, 16, &eval);
1932         if (len & KSTRTOX_OVERFLOW)
1933                 return -EINVAL;
1934         if (eval != (unsigned long)eval)
1935                 return -EINVAL;
1936         str += len;
1937
1938         if (*str != '\0')
1939                 return -EINVAL;
1940
1941         *start = sval;
1942         *end = eval;
1943
1944         return 0;
1945 }
1946
1947 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1948 {
1949         unsigned long vm_start, vm_end;
1950         bool exact_vma_exists = false;
1951         struct mm_struct *mm = NULL;
1952         struct task_struct *task;
1953         struct inode *inode;
1954         int status = 0;
1955
1956         if (flags & LOOKUP_RCU)
1957                 return -ECHILD;
1958
1959         inode = d_inode(dentry);
1960         task = get_proc_task(inode);
1961         if (!task)
1962                 goto out_notask;
1963
1964         mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1965         if (IS_ERR_OR_NULL(mm))
1966                 goto out;
1967
1968         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1969                 down_read(&mm->mmap_sem);
1970                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1971                 up_read(&mm->mmap_sem);
1972         }
1973
1974         mmput(mm);
1975
1976         if (exact_vma_exists) {
1977                 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1978
1979                 security_task_to_inode(task, inode);
1980                 status = 1;
1981         }
1982
1983 out:
1984         put_task_struct(task);
1985
1986 out_notask:
1987         return status;
1988 }
1989
1990 static const struct dentry_operations tid_map_files_dentry_operations = {
1991         .d_revalidate   = map_files_d_revalidate,
1992         .d_delete       = pid_delete_dentry,
1993 };
1994
1995 static int map_files_get_link(struct dentry *dentry, struct path *path)
1996 {
1997         unsigned long vm_start, vm_end;
1998         struct vm_area_struct *vma;
1999         struct task_struct *task;
2000         struct mm_struct *mm;
2001         int rc;
2002
2003         rc = -ENOENT;
2004         task = get_proc_task(d_inode(dentry));
2005         if (!task)
2006                 goto out;
2007
2008         mm = get_task_mm(task);
2009         put_task_struct(task);
2010         if (!mm)
2011                 goto out;
2012
2013         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2014         if (rc)
2015                 goto out_mmput;
2016
2017         rc = -ENOENT;
2018         down_read(&mm->mmap_sem);
2019         vma = find_exact_vma(mm, vm_start, vm_end);
2020         if (vma && vma->vm_file) {
2021                 *path = vma->vm_file->f_path;
2022                 path_get(path);
2023                 rc = 0;
2024         }
2025         up_read(&mm->mmap_sem);
2026
2027 out_mmput:
2028         mmput(mm);
2029 out:
2030         return rc;
2031 }
2032
2033 struct map_files_info {
2034         unsigned long   start;
2035         unsigned long   end;
2036         fmode_t         mode;
2037 };
2038
2039 /*
2040  * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
2041  * symlinks may be used to bypass permissions on ancestor directories in the
2042  * path to the file in question.
2043  */
2044 static const char *
2045 proc_map_files_get_link(struct dentry *dentry,
2046                         struct inode *inode,
2047                         struct delayed_call *done)
2048 {
2049         if (!capable(CAP_SYS_ADMIN))
2050                 return ERR_PTR(-EPERM);
2051
2052         return proc_pid_get_link(dentry, inode, done);
2053 }
2054
2055 /*
2056  * Identical to proc_pid_link_inode_operations except for get_link()
2057  */
2058 static const struct inode_operations proc_map_files_link_inode_operations = {
2059         .readlink       = proc_pid_readlink,
2060         .get_link       = proc_map_files_get_link,
2061         .setattr        = proc_setattr,
2062 };
2063
2064 static int
2065 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
2066                            struct task_struct *task, const void *ptr)
2067 {
2068         fmode_t mode = (fmode_t)(unsigned long)ptr;
2069         struct proc_inode *ei;
2070         struct inode *inode;
2071
2072         inode = proc_pid_make_inode(dir->i_sb, task, S_IFLNK |
2073                                     ((mode & FMODE_READ ) ? S_IRUSR : 0) |
2074                                     ((mode & FMODE_WRITE) ? S_IWUSR : 0));
2075         if (!inode)
2076                 return -ENOENT;
2077
2078         ei = PROC_I(inode);
2079         ei->op.proc_get_link = map_files_get_link;
2080
2081         inode->i_op = &proc_map_files_link_inode_operations;
2082         inode->i_size = 64;
2083
2084         d_set_d_op(dentry, &tid_map_files_dentry_operations);
2085         d_add(dentry, inode);
2086
2087         return 0;
2088 }
2089
2090 static struct dentry *proc_map_files_lookup(struct inode *dir,
2091                 struct dentry *dentry, unsigned int flags)
2092 {
2093         unsigned long vm_start, vm_end;
2094         struct vm_area_struct *vma;
2095         struct task_struct *task;
2096         int result;
2097         struct mm_struct *mm;
2098
2099         result = -ENOENT;
2100         task = get_proc_task(dir);
2101         if (!task)
2102                 goto out;
2103
2104         result = -EACCES;
2105         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2106                 goto out_put_task;
2107
2108         result = -ENOENT;
2109         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2110                 goto out_put_task;
2111
2112         mm = get_task_mm(task);
2113         if (!mm)
2114                 goto out_put_task;
2115
2116         down_read(&mm->mmap_sem);
2117         vma = find_exact_vma(mm, vm_start, vm_end);
2118         if (!vma)
2119                 goto out_no_vma;
2120
2121         if (vma->vm_file)
2122                 result = proc_map_files_instantiate(dir, dentry, task,
2123                                 (void *)(unsigned long)vma->vm_file->f_mode);
2124
2125 out_no_vma:
2126         up_read(&mm->mmap_sem);
2127         mmput(mm);
2128 out_put_task:
2129         put_task_struct(task);
2130 out:
2131         return ERR_PTR(result);
2132 }
2133
2134 static const struct inode_operations proc_map_files_inode_operations = {
2135         .lookup         = proc_map_files_lookup,
2136         .permission     = proc_fd_permission,
2137         .setattr        = proc_setattr,
2138 };
2139
2140 static int
2141 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2142 {
2143         struct vm_area_struct *vma;
2144         struct task_struct *task;
2145         struct mm_struct *mm;
2146         unsigned long nr_files, pos, i;
2147         struct flex_array *fa = NULL;
2148         struct map_files_info info;
2149         struct map_files_info *p;
2150         int ret;
2151
2152         ret = -ENOENT;
2153         task = get_proc_task(file_inode(file));
2154         if (!task)
2155                 goto out;
2156
2157         ret = -EACCES;
2158         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2159                 goto out_put_task;
2160
2161         ret = 0;
2162         if (!dir_emit_dots(file, ctx))
2163                 goto out_put_task;
2164
2165         mm = get_task_mm(task);
2166         if (!mm)
2167                 goto out_put_task;
2168         down_read(&mm->mmap_sem);
2169
2170         nr_files = 0;
2171
2172         /*
2173          * We need two passes here:
2174          *
2175          *  1) Collect vmas of mapped files with mmap_sem taken
2176          *  2) Release mmap_sem and instantiate entries
2177          *
2178          * otherwise we get lockdep complained, since filldir()
2179          * routine might require mmap_sem taken in might_fault().
2180          */
2181
2182         for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2183                 if (vma->vm_file && ++pos > ctx->pos)
2184                         nr_files++;
2185         }
2186
2187         if (nr_files) {
2188                 fa = flex_array_alloc(sizeof(info), nr_files,
2189                                         GFP_KERNEL);
2190                 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2191                                                 GFP_KERNEL)) {
2192                         ret = -ENOMEM;
2193                         if (fa)
2194                                 flex_array_free(fa);
2195                         up_read(&mm->mmap_sem);
2196                         mmput(mm);
2197                         goto out_put_task;
2198                 }
2199                 for (i = 0, vma = mm->mmap, pos = 2; vma;
2200                                 vma = vma->vm_next) {
2201                         if (!vma->vm_file)
2202                                 continue;
2203                         if (++pos <= ctx->pos)
2204                                 continue;
2205
2206                         info.start = vma->vm_start;
2207                         info.end = vma->vm_end;
2208                         info.mode = vma->vm_file->f_mode;
2209                         if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2210                                 BUG();
2211                 }
2212         }
2213         up_read(&mm->mmap_sem);
2214         mmput(mm);
2215
2216         for (i = 0; i < nr_files; i++) {
2217                 char buf[4 * sizeof(long) + 2]; /* max: %lx-%lx\0 */
2218                 unsigned int len;
2219
2220                 p = flex_array_get(fa, i);
2221                 len = snprintf(buf, sizeof(buf), "%lx-%lx", p->start, p->end);
2222                 if (!proc_fill_cache(file, ctx,
2223                                       buf, len,
2224                                       proc_map_files_instantiate,
2225                                       task,
2226                                       (void *)(unsigned long)p->mode))
2227                         break;
2228                 ctx->pos++;
2229         }
2230         if (fa)
2231                 flex_array_free(fa);
2232
2233 out_put_task:
2234         put_task_struct(task);
2235 out:
2236         return ret;
2237 }
2238
2239 static const struct file_operations proc_map_files_operations = {
2240         .read           = generic_read_dir,
2241         .iterate_shared = proc_map_files_readdir,
2242         .llseek         = generic_file_llseek,
2243 };
2244
2245 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2246 struct timers_private {
2247         struct pid *pid;
2248         struct task_struct *task;
2249         struct sighand_struct *sighand;
2250         struct pid_namespace *ns;
2251         unsigned long flags;
2252 };
2253
2254 static void *timers_start(struct seq_file *m, loff_t *pos)
2255 {
2256         struct timers_private *tp = m->private;
2257
2258         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2259         if (!tp->task)
2260                 return ERR_PTR(-ESRCH);
2261
2262         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2263         if (!tp->sighand)
2264                 return ERR_PTR(-ESRCH);
2265
2266         return seq_list_start(&tp->task->signal->posix_timers, *pos);
2267 }
2268
2269 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2270 {
2271         struct timers_private *tp = m->private;
2272         return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2273 }
2274
2275 static void timers_stop(struct seq_file *m, void *v)
2276 {
2277         struct timers_private *tp = m->private;
2278
2279         if (tp->sighand) {
2280                 unlock_task_sighand(tp->task, &tp->flags);
2281                 tp->sighand = NULL;
2282         }
2283
2284         if (tp->task) {
2285                 put_task_struct(tp->task);
2286                 tp->task = NULL;
2287         }
2288 }
2289
2290 static int show_timer(struct seq_file *m, void *v)
2291 {
2292         struct k_itimer *timer;
2293         struct timers_private *tp = m->private;
2294         int notify;
2295         static const char * const nstr[] = {
2296                 [SIGEV_SIGNAL] = "signal",
2297                 [SIGEV_NONE] = "none",
2298                 [SIGEV_THREAD] = "thread",
2299         };
2300
2301         timer = list_entry((struct list_head *)v, struct k_itimer, list);
2302         notify = timer->it_sigev_notify;
2303
2304         seq_printf(m, "ID: %d\n", timer->it_id);
2305         seq_printf(m, "signal: %d/%px\n",
2306                    timer->sigq->info.si_signo,
2307                    timer->sigq->info.si_value.sival_ptr);
2308         seq_printf(m, "notify: %s/%s.%d\n",
2309                    nstr[notify & ~SIGEV_THREAD_ID],
2310                    (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2311                    pid_nr_ns(timer->it_pid, tp->ns));
2312         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2313
2314         return 0;
2315 }
2316
2317 static const struct seq_operations proc_timers_seq_ops = {
2318         .start  = timers_start,
2319         .next   = timers_next,
2320         .stop   = timers_stop,
2321         .show   = show_timer,
2322 };
2323
2324 static int proc_timers_open(struct inode *inode, struct file *file)
2325 {
2326         struct timers_private *tp;
2327
2328         tp = __seq_open_private(file, &proc_timers_seq_ops,
2329                         sizeof(struct timers_private));
2330         if (!tp)
2331                 return -ENOMEM;
2332
2333         tp->pid = proc_pid(inode);
2334         tp->ns = inode->i_sb->s_fs_info;
2335         return 0;
2336 }
2337
2338 static const struct file_operations proc_timers_operations = {
2339         .open           = proc_timers_open,
2340         .read           = seq_read,
2341         .llseek         = seq_lseek,
2342         .release        = seq_release_private,
2343 };
2344 #endif
2345
2346 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2347                                         size_t count, loff_t *offset)
2348 {
2349         struct inode *inode = file_inode(file);
2350         struct task_struct *p;
2351         u64 slack_ns;
2352         int err;
2353
2354         err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2355         if (err < 0)
2356                 return err;
2357
2358         p = get_proc_task(inode);
2359         if (!p)
2360                 return -ESRCH;
2361
2362         if (p != current) {
2363                 if (!capable(CAP_SYS_NICE)) {
2364                         count = -EPERM;
2365                         goto out;
2366                 }
2367
2368                 err = security_task_setscheduler(p);
2369                 if (err) {
2370                         count = err;
2371                         goto out;
2372                 }
2373         }
2374
2375         task_lock(p);
2376         if (slack_ns == 0)
2377                 p->timer_slack_ns = p->default_timer_slack_ns;
2378         else
2379                 p->timer_slack_ns = slack_ns;
2380         task_unlock(p);
2381
2382 out:
2383         put_task_struct(p);
2384
2385         return count;
2386 }
2387
2388 static int timerslack_ns_show(struct seq_file *m, void *v)
2389 {
2390         struct inode *inode = m->private;
2391         struct task_struct *p;
2392         int err = 0;
2393
2394         p = get_proc_task(inode);
2395         if (!p)
2396                 return -ESRCH;
2397
2398         if (p != current) {
2399
2400                 if (!capable(CAP_SYS_NICE)) {
2401                         err = -EPERM;
2402                         goto out;
2403                 }
2404                 err = security_task_getscheduler(p);
2405                 if (err)
2406                         goto out;
2407         }
2408
2409         task_lock(p);
2410         seq_printf(m, "%llu\n", p->timer_slack_ns);
2411         task_unlock(p);
2412
2413 out:
2414         put_task_struct(p);
2415
2416         return err;
2417 }
2418
2419 static int timerslack_ns_open(struct inode *inode, struct file *filp)
2420 {
2421         return single_open(filp, timerslack_ns_show, inode);
2422 }
2423
2424 static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2425         .open           = timerslack_ns_open,
2426         .read           = seq_read,
2427         .write          = timerslack_ns_write,
2428         .llseek         = seq_lseek,
2429         .release        = single_release,
2430 };
2431
2432 static int proc_pident_instantiate(struct inode *dir,
2433         struct dentry *dentry, struct task_struct *task, const void *ptr)
2434 {
2435         const struct pid_entry *p = ptr;
2436         struct inode *inode;
2437         struct proc_inode *ei;
2438
2439         inode = proc_pid_make_inode(dir->i_sb, task, p->mode);
2440         if (!inode)
2441                 goto out;
2442
2443         ei = PROC_I(inode);
2444         if (S_ISDIR(inode->i_mode))
2445                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2446         if (p->iop)
2447                 inode->i_op = p->iop;
2448         if (p->fop)
2449                 inode->i_fop = p->fop;
2450         ei->op = p->op;
2451         d_set_d_op(dentry, &pid_dentry_operations);
2452         d_add(dentry, inode);
2453         /* Close the race of the process dying before we return the dentry */
2454         if (pid_revalidate(dentry, 0))
2455                 return 0;
2456 out:
2457         return -ENOENT;
2458 }
2459
2460 static struct dentry *proc_pident_lookup(struct inode *dir,
2461                                          struct dentry *dentry,
2462                                          const struct pid_entry *ents,
2463                                          unsigned int nents)
2464 {
2465         int error;
2466         struct task_struct *task = get_proc_task(dir);
2467         const struct pid_entry *p, *last;
2468
2469         error = -ENOENT;
2470
2471         if (!task)
2472                 goto out_no_task;
2473
2474         /*
2475          * Yes, it does not scale. And it should not. Don't add
2476          * new entries into /proc/<tgid>/ without very good reasons.
2477          */
2478         last = &ents[nents];
2479         for (p = ents; p < last; p++) {
2480                 if (p->len != dentry->d_name.len)
2481                         continue;
2482                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2483                         break;
2484         }
2485         if (p >= last)
2486                 goto out;
2487
2488         error = proc_pident_instantiate(dir, dentry, task, p);
2489 out:
2490         put_task_struct(task);
2491 out_no_task:
2492         return ERR_PTR(error);
2493 }
2494
2495 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2496                 const struct pid_entry *ents, unsigned int nents)
2497 {
2498         struct task_struct *task = get_proc_task(file_inode(file));
2499         const struct pid_entry *p;
2500
2501         if (!task)
2502                 return -ENOENT;
2503
2504         if (!dir_emit_dots(file, ctx))
2505                 goto out;
2506
2507         if (ctx->pos >= nents + 2)
2508                 goto out;
2509
2510         for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2511                 if (!proc_fill_cache(file, ctx, p->name, p->len,
2512                                 proc_pident_instantiate, task, p))
2513                         break;
2514                 ctx->pos++;
2515         }
2516 out:
2517         put_task_struct(task);
2518         return 0;
2519 }
2520
2521 #ifdef CONFIG_SECURITY
2522 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2523                                   size_t count, loff_t *ppos)
2524 {
2525         struct inode * inode = file_inode(file);
2526         char *p = NULL;
2527         ssize_t length;
2528         struct task_struct *task = get_proc_task(inode);
2529
2530         if (!task)
2531                 return -ESRCH;
2532
2533         length = security_getprocattr(task,
2534                                       (char*)file->f_path.dentry->d_name.name,
2535                                       &p);
2536         put_task_struct(task);
2537         if (length > 0)
2538                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2539         kfree(p);
2540         return length;
2541 }
2542
2543 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2544                                    size_t count, loff_t *ppos)
2545 {
2546         struct inode * inode = file_inode(file);
2547         void *page;
2548         ssize_t length;
2549         struct task_struct *task = get_proc_task(inode);
2550
2551         length = -ESRCH;
2552         if (!task)
2553                 goto out_no_task;
2554
2555         /* A task may only write its own attributes. */
2556         length = -EACCES;
2557         if (current != task)
2558                 goto out;
2559
2560         if (count > PAGE_SIZE)
2561                 count = PAGE_SIZE;
2562
2563         /* No partial writes. */
2564         length = -EINVAL;
2565         if (*ppos != 0)
2566                 goto out;
2567
2568         page = memdup_user(buf, count);
2569         if (IS_ERR(page)) {
2570                 length = PTR_ERR(page);
2571                 goto out;
2572         }
2573
2574         /* Guard against adverse ptrace interaction */
2575         length = mutex_lock_interruptible(&current->signal->cred_guard_mutex);
2576         if (length < 0)
2577                 goto out_free;
2578
2579         length = security_setprocattr(file->f_path.dentry->d_name.name,
2580                                       page, count);
2581         mutex_unlock(&current->signal->cred_guard_mutex);
2582 out_free:
2583         kfree(page);
2584 out:
2585         put_task_struct(task);
2586 out_no_task:
2587         return length;
2588 }
2589
2590 static const struct file_operations proc_pid_attr_operations = {
2591         .read           = proc_pid_attr_read,
2592         .write          = proc_pid_attr_write,
2593         .llseek         = generic_file_llseek,
2594 };
2595
2596 static const struct pid_entry attr_dir_stuff[] = {
2597         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2598         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2599         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2600         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2601         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2602         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2603 };
2604
2605 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2606 {
2607         return proc_pident_readdir(file, ctx,
2608                                    attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2609 }
2610
2611 static const struct file_operations proc_attr_dir_operations = {
2612         .read           = generic_read_dir,
2613         .iterate_shared = proc_attr_dir_readdir,
2614         .llseek         = generic_file_llseek,
2615 };
2616
2617 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2618                                 struct dentry *dentry, unsigned int flags)
2619 {
2620         return proc_pident_lookup(dir, dentry,
2621                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2622 }
2623
2624 static const struct inode_operations proc_attr_dir_inode_operations = {
2625         .lookup         = proc_attr_dir_lookup,
2626         .getattr        = pid_getattr,
2627         .setattr        = proc_setattr,
2628 };
2629
2630 #endif
2631
2632 #ifdef CONFIG_ELF_CORE
2633 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2634                                          size_t count, loff_t *ppos)
2635 {
2636         struct task_struct *task = get_proc_task(file_inode(file));
2637         struct mm_struct *mm;
2638         char buffer[PROC_NUMBUF];
2639         size_t len;
2640         int ret;
2641
2642         if (!task)
2643                 return -ESRCH;
2644
2645         ret = 0;
2646         mm = get_task_mm(task);
2647         if (mm) {
2648                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2649                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2650                                 MMF_DUMP_FILTER_SHIFT));
2651                 mmput(mm);
2652                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2653         }
2654
2655         put_task_struct(task);
2656
2657         return ret;
2658 }
2659
2660 static ssize_t proc_coredump_filter_write(struct file *file,
2661                                           const char __user *buf,
2662                                           size_t count,
2663                                           loff_t *ppos)
2664 {
2665         struct task_struct *task;
2666         struct mm_struct *mm;
2667         unsigned int val;
2668         int ret;
2669         int i;
2670         unsigned long mask;
2671
2672         ret = kstrtouint_from_user(buf, count, 0, &val);
2673         if (ret < 0)
2674                 return ret;
2675
2676         ret = -ESRCH;
2677         task = get_proc_task(file_inode(file));
2678         if (!task)
2679                 goto out_no_task;
2680
2681         mm = get_task_mm(task);
2682         if (!mm)
2683                 goto out_no_mm;
2684         ret = 0;
2685
2686         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2687                 if (val & mask)
2688                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2689                 else
2690                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2691         }
2692
2693         mmput(mm);
2694  out_no_mm:
2695         put_task_struct(task);
2696  out_no_task:
2697         if (ret < 0)
2698                 return ret;
2699         return count;
2700 }
2701
2702 static const struct file_operations proc_coredump_filter_operations = {
2703         .read           = proc_coredump_filter_read,
2704         .write          = proc_coredump_filter_write,
2705         .llseek         = generic_file_llseek,
2706 };
2707 #endif
2708
2709 #ifdef CONFIG_TASK_IO_ACCOUNTING
2710 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2711 {
2712         struct task_io_accounting acct = task->ioac;
2713         unsigned long flags;
2714         int result;
2715
2716         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2717         if (result)
2718                 return result;
2719
2720         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2721                 result = -EACCES;
2722                 goto out_unlock;
2723         }
2724
2725         if (whole && lock_task_sighand(task, &flags)) {
2726                 struct task_struct *t = task;
2727
2728                 task_io_accounting_add(&acct, &task->signal->ioac);
2729                 while_each_thread(task, t)
2730                         task_io_accounting_add(&acct, &t->ioac);
2731
2732                 unlock_task_sighand(task, &flags);
2733         }
2734         seq_printf(m,
2735                    "rchar: %llu\n"
2736                    "wchar: %llu\n"
2737                    "syscr: %llu\n"
2738                    "syscw: %llu\n"
2739                    "read_bytes: %llu\n"
2740                    "write_bytes: %llu\n"
2741                    "cancelled_write_bytes: %llu\n",
2742                    (unsigned long long)acct.rchar,
2743                    (unsigned long long)acct.wchar,
2744                    (unsigned long long)acct.syscr,
2745                    (unsigned long long)acct.syscw,
2746                    (unsigned long long)acct.read_bytes,
2747                    (unsigned long long)acct.write_bytes,
2748                    (unsigned long long)acct.cancelled_write_bytes);
2749         result = 0;
2750
2751 out_unlock:
2752         mutex_unlock(&task->signal->cred_guard_mutex);
2753         return result;
2754 }
2755
2756 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2757                                   struct pid *pid, struct task_struct *task)
2758 {
2759         return do_io_accounting(task, m, 0);
2760 }
2761
2762 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2763                                    struct pid *pid, struct task_struct *task)
2764 {
2765         return do_io_accounting(task, m, 1);
2766 }
2767 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2768
2769 #ifdef CONFIG_USER_NS
2770 static int proc_id_map_open(struct inode *inode, struct file *file,
2771         const struct seq_operations *seq_ops)
2772 {
2773         struct user_namespace *ns = NULL;
2774         struct task_struct *task;
2775         struct seq_file *seq;
2776         int ret = -EINVAL;
2777
2778         task = get_proc_task(inode);
2779         if (task) {
2780                 rcu_read_lock();
2781                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2782                 rcu_read_unlock();
2783                 put_task_struct(task);
2784         }
2785         if (!ns)
2786                 goto err;
2787
2788         ret = seq_open(file, seq_ops);
2789         if (ret)
2790                 goto err_put_ns;
2791
2792         seq = file->private_data;
2793         seq->private = ns;
2794
2795         return 0;
2796 err_put_ns:
2797         put_user_ns(ns);
2798 err:
2799         return ret;
2800 }
2801
2802 static int proc_id_map_release(struct inode *inode, struct file *file)
2803 {
2804         struct seq_file *seq = file->private_data;
2805         struct user_namespace *ns = seq->private;
2806         put_user_ns(ns);
2807         return seq_release(inode, file);
2808 }
2809
2810 static int proc_uid_map_open(struct inode *inode, struct file *file)
2811 {
2812         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2813 }
2814
2815 static int proc_gid_map_open(struct inode *inode, struct file *file)
2816 {
2817         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2818 }
2819
2820 static int proc_projid_map_open(struct inode *inode, struct file *file)
2821 {
2822         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2823 }
2824
2825 static const struct file_operations proc_uid_map_operations = {
2826         .open           = proc_uid_map_open,
2827         .write          = proc_uid_map_write,
2828         .read           = seq_read,
2829         .llseek         = seq_lseek,
2830         .release        = proc_id_map_release,
2831 };
2832
2833 static const struct file_operations proc_gid_map_operations = {
2834         .open           = proc_gid_map_open,
2835         .write          = proc_gid_map_write,
2836         .read           = seq_read,
2837         .llseek         = seq_lseek,
2838         .release        = proc_id_map_release,
2839 };
2840
2841 static const struct file_operations proc_projid_map_operations = {
2842         .open           = proc_projid_map_open,
2843         .write          = proc_projid_map_write,
2844         .read           = seq_read,
2845         .llseek         = seq_lseek,
2846         .release        = proc_id_map_release,
2847 };
2848
2849 static int proc_setgroups_open(struct inode *inode, struct file *file)
2850 {
2851         struct user_namespace *ns = NULL;
2852         struct task_struct *task;
2853         int ret;
2854
2855         ret = -ESRCH;
2856         task = get_proc_task(inode);
2857         if (task) {
2858                 rcu_read_lock();
2859                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2860                 rcu_read_unlock();
2861                 put_task_struct(task);
2862         }
2863         if (!ns)
2864                 goto err;
2865
2866         if (file->f_mode & FMODE_WRITE) {
2867                 ret = -EACCES;
2868                 if (!ns_capable(ns, CAP_SYS_ADMIN))
2869                         goto err_put_ns;
2870         }
2871
2872         ret = single_open(file, &proc_setgroups_show, ns);
2873         if (ret)
2874                 goto err_put_ns;
2875
2876         return 0;
2877 err_put_ns:
2878         put_user_ns(ns);
2879 err:
2880         return ret;
2881 }
2882
2883 static int proc_setgroups_release(struct inode *inode, struct file *file)
2884 {
2885         struct seq_file *seq = file->private_data;
2886         struct user_namespace *ns = seq->private;
2887         int ret = single_release(inode, file);
2888         put_user_ns(ns);
2889         return ret;
2890 }
2891
2892 static const struct file_operations proc_setgroups_operations = {
2893         .open           = proc_setgroups_open,
2894         .write          = proc_setgroups_write,
2895         .read           = seq_read,
2896         .llseek         = seq_lseek,
2897         .release        = proc_setgroups_release,
2898 };
2899 #endif /* CONFIG_USER_NS */
2900
2901 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2902                                 struct pid *pid, struct task_struct *task)
2903 {
2904         int err = lock_trace(task);
2905         if (!err) {
2906                 seq_printf(m, "%08x\n", task->personality);
2907                 unlock_trace(task);
2908         }
2909         return err;
2910 }
2911
2912 #ifdef CONFIG_LIVEPATCH
2913 static int proc_pid_patch_state(struct seq_file *m, struct pid_namespace *ns,
2914                                 struct pid *pid, struct task_struct *task)
2915 {
2916         seq_printf(m, "%d\n", task->patch_state);
2917         return 0;
2918 }
2919 #endif /* CONFIG_LIVEPATCH */
2920
2921 /*
2922  * Thread groups
2923  */
2924 static const struct file_operations proc_task_operations;
2925 static const struct inode_operations proc_task_inode_operations;
2926
2927 static const struct pid_entry tgid_base_stuff[] = {
2928         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2929         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2930         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2931         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2932         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2933 #ifdef CONFIG_NET
2934         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2935 #endif
2936         REG("environ",    S_IRUSR, proc_environ_operations),
2937         REG("auxv",       S_IRUSR, proc_auxv_operations),
2938         ONE("status",     S_IRUGO, proc_pid_status),
2939         ONE("personality", S_IRUSR, proc_pid_personality),
2940         ONE("limits",     S_IRUGO, proc_pid_limits),
2941 #ifdef CONFIG_SCHED_DEBUG
2942         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2943 #endif
2944 #ifdef CONFIG_SCHED_AUTOGROUP
2945         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2946 #endif
2947         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2948 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2949         ONE("syscall",    S_IRUSR, proc_pid_syscall),
2950 #endif
2951         REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
2952         ONE("stat",       S_IRUGO, proc_tgid_stat),
2953         ONE("statm",      S_IRUGO, proc_pid_statm),
2954         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2955 #ifdef CONFIG_NUMA
2956         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2957 #endif
2958         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2959         LNK("cwd",        proc_cwd_link),
2960         LNK("root",       proc_root_link),
2961         LNK("exe",        proc_exe_link),
2962         REG("mounts",     S_IRUGO, proc_mounts_operations),
2963         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2964         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2965 #ifdef CONFIG_PROC_PAGE_MONITOR
2966         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2967         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2968         REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
2969         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2970 #endif
2971 #ifdef CONFIG_SECURITY
2972         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2973 #endif
2974 #ifdef CONFIG_KALLSYMS
2975         ONE("wchan",      S_IRUGO, proc_pid_wchan),
2976 #endif
2977 #ifdef CONFIG_STACKTRACE
2978         ONE("stack",      S_IRUSR, proc_pid_stack),
2979 #endif
2980 #ifdef CONFIG_SCHED_INFO
2981         ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
2982 #endif
2983 #ifdef CONFIG_LATENCYTOP
2984         REG("latency",  S_IRUGO, proc_lstats_operations),
2985 #endif
2986 #ifdef CONFIG_PROC_PID_CPUSET
2987         ONE("cpuset",     S_IRUGO, proc_cpuset_show),
2988 #endif
2989 #ifdef CONFIG_CGROUPS
2990         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
2991 #endif
2992         ONE("oom_score",  S_IRUGO, proc_oom_score),
2993         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2994         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2995 #ifdef CONFIG_AUDITSYSCALL
2996         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2997         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2998 #endif
2999 #ifdef CONFIG_FAULT_INJECTION
3000         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3001         REG("fail-nth", 0644, proc_fail_nth_operations),
3002 #endif
3003 #ifdef CONFIG_ELF_CORE
3004         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3005 #endif
3006 #ifdef CONFIG_TASK_IO_ACCOUNTING
3007         ONE("io",       S_IRUSR, proc_tgid_io_accounting),
3008 #endif
3009 #ifdef CONFIG_USER_NS
3010         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3011         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3012         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3013         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3014 #endif
3015 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
3016         REG("timers",     S_IRUGO, proc_timers_operations),
3017 #endif
3018         REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
3019 #ifdef CONFIG_LIVEPATCH
3020         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
3021 #endif
3022 };
3023
3024 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
3025 {
3026         return proc_pident_readdir(file, ctx,
3027                                    tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3028 }
3029
3030 static const struct file_operations proc_tgid_base_operations = {
3031         .read           = generic_read_dir,
3032         .iterate_shared = proc_tgid_base_readdir,
3033         .llseek         = generic_file_llseek,
3034 };
3035
3036 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3037 {
3038         return proc_pident_lookup(dir, dentry,
3039                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3040 }
3041
3042 static const struct inode_operations proc_tgid_base_inode_operations = {
3043         .lookup         = proc_tgid_base_lookup,
3044         .getattr        = pid_getattr,
3045         .setattr        = proc_setattr,
3046         .permission     = proc_pid_permission,
3047 };
3048
3049 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
3050 {
3051         struct dentry *dentry, *leader, *dir;
3052         char buf[10 + 1];
3053         struct qstr name;
3054
3055         name.name = buf;
3056         name.len = snprintf(buf, sizeof(buf), "%u", pid);
3057         /* no ->d_hash() rejects on procfs */
3058         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
3059         if (dentry) {
3060                 d_invalidate(dentry);
3061                 dput(dentry);
3062         }
3063
3064         if (pid == tgid)
3065                 return;
3066
3067         name.name = buf;
3068         name.len = snprintf(buf, sizeof(buf), "%u", tgid);
3069         leader = d_hash_and_lookup(mnt->mnt_root, &name);
3070         if (!leader)
3071                 goto out;
3072
3073         name.name = "task";
3074         name.len = strlen(name.name);
3075         dir = d_hash_and_lookup(leader, &name);
3076         if (!dir)
3077                 goto out_put_leader;
3078
3079         name.name = buf;
3080         name.len = snprintf(buf, sizeof(buf), "%u", pid);
3081         dentry = d_hash_and_lookup(dir, &name);
3082         if (dentry) {
3083                 d_invalidate(dentry);
3084                 dput(dentry);
3085         }
3086
3087         dput(dir);
3088 out_put_leader:
3089         dput(leader);
3090 out:
3091         return;
3092 }
3093
3094 /**
3095  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
3096  * @task: task that should be flushed.
3097  *
3098  * When flushing dentries from proc, one needs to flush them from global
3099  * proc (proc_mnt) and from all the namespaces' procs this task was seen
3100  * in. This call is supposed to do all of this job.
3101  *
3102  * Looks in the dcache for
3103  * /proc/@pid
3104  * /proc/@tgid/task/@pid
3105  * if either directory is present flushes it and all of it'ts children
3106  * from the dcache.
3107  *
3108  * It is safe and reasonable to cache /proc entries for a task until
3109  * that task exits.  After that they just clog up the dcache with
3110  * useless entries, possibly causing useful dcache entries to be
3111  * flushed instead.  This routine is proved to flush those useless
3112  * dcache entries at process exit time.
3113  *
3114  * NOTE: This routine is just an optimization so it does not guarantee
3115  *       that no dcache entries will exist at process exit time it
3116  *       just makes it very unlikely that any will persist.
3117  */
3118
3119 void proc_flush_task(struct task_struct *task)
3120 {
3121         int i;
3122         struct pid *pid, *tgid;
3123         struct upid *upid;
3124
3125         pid = task_pid(task);
3126         tgid = task_tgid(task);
3127
3128         for (i = 0; i <= pid->level; i++) {
3129                 upid = &pid->numbers[i];
3130                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3131                                         tgid->numbers[i].nr);
3132         }
3133 }
3134
3135 static int proc_pid_instantiate(struct inode *dir,
3136                                    struct dentry * dentry,
3137                                    struct task_struct *task, const void *ptr)
3138 {
3139         struct inode *inode;
3140
3141         inode = proc_pid_make_inode(dir->i_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3142         if (!inode)
3143                 goto out;
3144
3145         inode->i_op = &proc_tgid_base_inode_operations;
3146         inode->i_fop = &proc_tgid_base_operations;
3147         inode->i_flags|=S_IMMUTABLE;
3148
3149         set_nlink(inode, nlink_tgid);
3150
3151         d_set_d_op(dentry, &pid_dentry_operations);
3152
3153         d_add(dentry, inode);
3154         /* Close the race of the process dying before we return the dentry */
3155         if (pid_revalidate(dentry, 0))
3156                 return 0;
3157 out:
3158         return -ENOENT;
3159 }
3160
3161 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3162 {
3163         int result = -ENOENT;
3164         struct task_struct *task;
3165         unsigned tgid;
3166         struct pid_namespace *ns;
3167
3168         tgid = name_to_int(&dentry->d_name);
3169         if (tgid == ~0U)
3170                 goto out;
3171
3172         ns = dentry->d_sb->s_fs_info;
3173         rcu_read_lock();
3174         task = find_task_by_pid_ns(tgid, ns);
3175         if (task)
3176                 get_task_struct(task);
3177         rcu_read_unlock();
3178         if (!task)
3179                 goto out;
3180
3181         result = proc_pid_instantiate(dir, dentry, task, NULL);
3182         put_task_struct(task);
3183 out:
3184         return ERR_PTR(result);
3185 }
3186
3187 /*
3188  * Find the first task with tgid >= tgid
3189  *
3190  */
3191 struct tgid_iter {
3192         unsigned int tgid;
3193         struct task_struct *task;
3194 };
3195 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3196 {
3197         struct pid *pid;
3198
3199         if (iter.task)
3200                 put_task_struct(iter.task);
3201         rcu_read_lock();
3202 retry:
3203         iter.task = NULL;
3204         pid = find_ge_pid(iter.tgid, ns);
3205         if (pid) {
3206                 iter.tgid = pid_nr_ns(pid, ns);
3207                 iter.task = pid_task(pid, PIDTYPE_PID);
3208                 /* What we to know is if the pid we have find is the
3209                  * pid of a thread_group_leader.  Testing for task
3210                  * being a thread_group_leader is the obvious thing
3211                  * todo but there is a window when it fails, due to
3212                  * the pid transfer logic in de_thread.
3213                  *
3214                  * So we perform the straight forward test of seeing
3215                  * if the pid we have found is the pid of a thread
3216                  * group leader, and don't worry if the task we have
3217                  * found doesn't happen to be a thread group leader.
3218                  * As we don't care in the case of readdir.
3219                  */
3220                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3221                         iter.tgid += 1;
3222                         goto retry;
3223                 }
3224                 get_task_struct(iter.task);
3225         }
3226         rcu_read_unlock();
3227         return iter;
3228 }
3229
3230 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3231
3232 /* for the /proc/ directory itself, after non-process stuff has been done */
3233 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3234 {
3235         struct tgid_iter iter;
3236         struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
3237         loff_t pos = ctx->pos;
3238
3239         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3240                 return 0;
3241
3242         if (pos == TGID_OFFSET - 2) {
3243                 struct inode *inode = d_inode(ns->proc_self);
3244                 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3245                         return 0;
3246                 ctx->pos = pos = pos + 1;
3247         }
3248         if (pos == TGID_OFFSET - 1) {
3249                 struct inode *inode = d_inode(ns->proc_thread_self);
3250                 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3251                         return 0;
3252                 ctx->pos = pos = pos + 1;
3253         }
3254         iter.tgid = pos - TGID_OFFSET;
3255         iter.task = NULL;
3256         for (iter = next_tgid(ns, iter);
3257              iter.task;
3258              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3259                 char name[10 + 1];
3260                 int len;
3261
3262                 cond_resched();
3263                 if (!has_pid_permissions(ns, iter.task, HIDEPID_INVISIBLE))
3264                         continue;
3265
3266                 len = snprintf(name, sizeof(name), "%u", iter.tgid);
3267                 ctx->pos = iter.tgid + TGID_OFFSET;
3268                 if (!proc_fill_cache(file, ctx, name, len,
3269                                      proc_pid_instantiate, iter.task, NULL)) {
3270                         put_task_struct(iter.task);
3271                         return 0;
3272                 }
3273         }
3274         ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3275         return 0;
3276 }
3277
3278 /*
3279  * proc_tid_comm_permission is a special permission function exclusively
3280  * used for the node /proc/<pid>/task/<tid>/comm.
3281  * It bypasses generic permission checks in the case where a task of the same
3282  * task group attempts to access the node.
3283  * The rationale behind this is that glibc and bionic access this node for
3284  * cross thread naming (pthread_set/getname_np(!self)). However, if
3285  * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3286  * which locks out the cross thread naming implementation.
3287  * This function makes sure that the node is always accessible for members of
3288  * same thread group.
3289  */
3290 static int proc_tid_comm_permission(struct inode *inode, int mask)
3291 {
3292         bool is_same_tgroup;
3293         struct task_struct *task;
3294
3295         task = get_proc_task(inode);
3296         if (!task)
3297                 return -ESRCH;
3298         is_same_tgroup = same_thread_group(current, task);
3299         put_task_struct(task);
3300
3301         if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3302                 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3303                  * read or written by the members of the corresponding
3304                  * thread group.
3305                  */
3306                 return 0;
3307         }
3308
3309         return generic_permission(inode, mask);
3310 }
3311
3312 static const struct inode_operations proc_tid_comm_inode_operations = {
3313                 .permission = proc_tid_comm_permission,
3314 };
3315
3316 /*
3317  * Tasks
3318  */
3319 static const struct pid_entry tid_base_stuff[] = {
3320         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3321         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3322         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3323 #ifdef CONFIG_NET
3324         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3325 #endif
3326         REG("environ",   S_IRUSR, proc_environ_operations),
3327         REG("auxv",      S_IRUSR, proc_auxv_operations),
3328         ONE("status",    S_IRUGO, proc_pid_status),
3329         ONE("personality", S_IRUSR, proc_pid_personality),
3330         ONE("limits",    S_IRUGO, proc_pid_limits),
3331 #ifdef CONFIG_SCHED_DEBUG
3332         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3333 #endif
3334         NOD("comm",      S_IFREG|S_IRUGO|S_IWUSR,
3335                          &proc_tid_comm_inode_operations,
3336                          &proc_pid_set_comm_operations, {}),
3337 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3338         ONE("syscall",   S_IRUSR, proc_pid_syscall),
3339 #endif
3340         REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3341         ONE("stat",      S_IRUGO, proc_tid_stat),
3342         ONE("statm",     S_IRUGO, proc_pid_statm),
3343         REG("maps",      S_IRUGO, proc_tid_maps_operations),
3344 #ifdef CONFIG_PROC_CHILDREN
3345         REG("children",  S_IRUGO, proc_tid_children_operations),
3346 #endif
3347 #ifdef CONFIG_NUMA
3348         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3349 #endif
3350         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3351         LNK("cwd",       proc_cwd_link),
3352         LNK("root",      proc_root_link),
3353         LNK("exe",       proc_exe_link),
3354         REG("mounts",    S_IRUGO, proc_mounts_operations),
3355         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3356 #ifdef CONFIG_PROC_PAGE_MONITOR
3357         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3358         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3359         REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3360         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3361 #endif
3362 #ifdef CONFIG_SECURITY
3363         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3364 #endif
3365 #ifdef CONFIG_KALLSYMS
3366         ONE("wchan",     S_IRUGO, proc_pid_wchan),
3367 #endif
3368 #ifdef CONFIG_STACKTRACE
3369         ONE("stack",      S_IRUSR, proc_pid_stack),
3370 #endif
3371 #ifdef CONFIG_SCHED_INFO
3372         ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3373 #endif
3374 #ifdef CONFIG_LATENCYTOP
3375         REG("latency",  S_IRUGO, proc_lstats_operations),
3376 #endif
3377 #ifdef CONFIG_PROC_PID_CPUSET
3378         ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3379 #endif
3380 #ifdef CONFIG_CGROUPS
3381         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3382 #endif
3383         ONE("oom_score", S_IRUGO, proc_oom_score),
3384         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3385         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3386 #ifdef CONFIG_AUDITSYSCALL
3387         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3388         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3389 #endif
3390 #ifdef CONFIG_FAULT_INJECTION
3391         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3392         REG("fail-nth", 0644, proc_fail_nth_operations),
3393 #endif
3394 #ifdef CONFIG_TASK_IO_ACCOUNTING
3395         ONE("io",       S_IRUSR, proc_tid_io_accounting),
3396 #endif
3397 #ifdef CONFIG_USER_NS
3398         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3399         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3400         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3401         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3402 #endif
3403 #ifdef CONFIG_LIVEPATCH
3404         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
3405 #endif
3406 };
3407
3408 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3409 {
3410         return proc_pident_readdir(file, ctx,
3411                                    tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3412 }
3413
3414 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3415 {
3416         return proc_pident_lookup(dir, dentry,
3417                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3418 }
3419
3420 static const struct file_operations proc_tid_base_operations = {
3421         .read           = generic_read_dir,
3422         .iterate_shared = proc_tid_base_readdir,
3423         .llseek         = generic_file_llseek,
3424 };
3425
3426 static const struct inode_operations proc_tid_base_inode_operations = {
3427         .lookup         = proc_tid_base_lookup,
3428         .getattr        = pid_getattr,
3429         .setattr        = proc_setattr,
3430 };
3431
3432 static int proc_task_instantiate(struct inode *dir,
3433         struct dentry *dentry, struct task_struct *task, const void *ptr)
3434 {
3435         struct inode *inode;
3436         inode = proc_pid_make_inode(dir->i_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3437
3438         if (!inode)
3439                 goto out;
3440         inode->i_op = &proc_tid_base_inode_operations;
3441         inode->i_fop = &proc_tid_base_operations;
3442         inode->i_flags|=S_IMMUTABLE;
3443
3444         set_nlink(inode, nlink_tid);
3445
3446         d_set_d_op(dentry, &pid_dentry_operations);
3447
3448         d_add(dentry, inode);
3449         /* Close the race of the process dying before we return the dentry */
3450         if (pid_revalidate(dentry, 0))
3451                 return 0;
3452 out:
3453         return -ENOENT;
3454 }
3455
3456 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3457 {
3458         int result = -ENOENT;
3459         struct task_struct *task;
3460         struct task_struct *leader = get_proc_task(dir);
3461         unsigned tid;
3462         struct pid_namespace *ns;
3463
3464         if (!leader)
3465                 goto out_no_task;
3466
3467         tid = name_to_int(&dentry->d_name);
3468         if (tid == ~0U)
3469                 goto out;
3470
3471         ns = dentry->d_sb->s_fs_info;
3472         rcu_read_lock();
3473         task = find_task_by_pid_ns(tid, ns);
3474         if (task)
3475                 get_task_struct(task);
3476         rcu_read_unlock();
3477         if (!task)
3478                 goto out;
3479         if (!same_thread_group(leader, task))
3480                 goto out_drop_task;
3481
3482         result = proc_task_instantiate(dir, dentry, task, NULL);
3483 out_drop_task:
3484         put_task_struct(task);
3485 out:
3486         put_task_struct(leader);
3487 out_no_task:
3488         return ERR_PTR(result);
3489 }
3490
3491 /*
3492  * Find the first tid of a thread group to return to user space.
3493  *
3494  * Usually this is just the thread group leader, but if the users
3495  * buffer was too small or there was a seek into the middle of the
3496  * directory we have more work todo.
3497  *
3498  * In the case of a short read we start with find_task_by_pid.
3499  *
3500  * In the case of a seek we start with the leader and walk nr
3501  * threads past it.
3502  */
3503 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3504                                         struct pid_namespace *ns)
3505 {
3506         struct task_struct *pos, *task;
3507         unsigned long nr = f_pos;
3508
3509         if (nr != f_pos)        /* 32bit overflow? */
3510                 return NULL;
3511
3512         rcu_read_lock();
3513         task = pid_task(pid, PIDTYPE_PID);
3514         if (!task)
3515                 goto fail;
3516
3517         /* Attempt to start with the tid of a thread */
3518         if (tid && nr) {
3519                 pos = find_task_by_pid_ns(tid, ns);
3520                 if (pos && same_thread_group(pos, task))
3521                         goto found;
3522         }
3523
3524         /* If nr exceeds the number of threads there is nothing todo */
3525         if (nr >= get_nr_threads(task))
3526                 goto fail;
3527
3528         /* If we haven't found our starting place yet start
3529          * with the leader and walk nr threads forward.
3530          */
3531         pos = task = task->group_leader;
3532         do {
3533                 if (!nr--)
3534                         goto found;
3535         } while_each_thread(task, pos);
3536 fail:
3537         pos = NULL;
3538         goto out;
3539 found:
3540         get_task_struct(pos);
3541 out:
3542         rcu_read_unlock();
3543         return pos;
3544 }
3545
3546 /*
3547  * Find the next thread in the thread list.
3548  * Return NULL if there is an error or no next thread.
3549  *
3550  * The reference to the input task_struct is released.
3551  */
3552 static struct task_struct *next_tid(struct task_struct *start)
3553 {
3554         struct task_struct *pos = NULL;
3555         rcu_read_lock();
3556         if (pid_alive(start)) {
3557                 pos = next_thread(start);
3558                 if (thread_group_leader(pos))
3559                         pos = NULL;
3560                 else
3561                         get_task_struct(pos);
3562         }
3563         rcu_read_unlock();
3564         put_task_struct(start);
3565         return pos;
3566 }
3567
3568 /* for the /proc/TGID/task/ directories */
3569 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3570 {
3571         struct inode *inode = file_inode(file);
3572         struct task_struct *task;
3573         struct pid_namespace *ns;
3574         int tid;
3575
3576         if (proc_inode_is_dead(inode))
3577                 return -ENOENT;
3578
3579         if (!dir_emit_dots(file, ctx))
3580                 return 0;
3581
3582         /* f_version caches the tgid value that the last readdir call couldn't
3583          * return. lseek aka telldir automagically resets f_version to 0.
3584          */
3585         ns = inode->i_sb->s_fs_info;
3586         tid = (int)file->f_version;
3587         file->f_version = 0;
3588         for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3589              task;
3590              task = next_tid(task), ctx->pos++) {
3591                 char name[10 + 1];
3592                 int len;
3593                 tid = task_pid_nr_ns(task, ns);
3594                 len = snprintf(name, sizeof(name), "%u", tid);
3595                 if (!proc_fill_cache(file, ctx, name, len,
3596                                 proc_task_instantiate, task, NULL)) {
3597                         /* returning this tgid failed, save it as the first
3598                          * pid for the next readir call */
3599                         file->f_version = (u64)tid;
3600                         put_task_struct(task);
3601                         break;
3602                 }
3603         }
3604
3605         return 0;
3606 }
3607
3608 static int proc_task_getattr(const struct path *path, struct kstat *stat,
3609                              u32 request_mask, unsigned int query_flags)
3610 {
3611         struct inode *inode = d_inode(path->dentry);
3612         struct task_struct *p = get_proc_task(inode);
3613         generic_fillattr(inode, stat);
3614
3615         if (p) {
3616                 stat->nlink += get_nr_threads(p);
3617                 put_task_struct(p);
3618         }
3619
3620         return 0;
3621 }
3622
3623 static const struct inode_operations proc_task_inode_operations = {
3624         .lookup         = proc_task_lookup,
3625         .getattr        = proc_task_getattr,
3626         .setattr        = proc_setattr,
3627         .permission     = proc_pid_permission,
3628 };
3629
3630 static const struct file_operations proc_task_operations = {
3631         .read           = generic_read_dir,
3632         .iterate_shared = proc_task_readdir,
3633         .llseek         = generic_file_llseek,
3634 };
3635
3636 void __init set_proc_pid_nlink(void)
3637 {
3638         nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3639         nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3640 }