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/cgroup.h>
  79 #include <linux/cpuset.h>
  80 #include <linux/audit.h>
  81 #include <linux/poll.h>
  82 #include <linux/nsproxy.h>
  83 #include <linux/oom.h>
  84 #include <linux/elf.h>
  85 #include <linux/pid_namespace.h>
  86 #include <linux/user_namespace.h>
  87 #include <linux/fs_struct.h>
  88 #include <linux/slab.h>
  89 #include <linux/sched/autogroup.h>
  90 #include <linux/sched/mm.h>
  91 #include <linux/sched/coredump.h>
  92 #include <linux/sched/debug.h>
  93 #include <linux/sched/stat.h>
  94 #include <linux/flex_array.h>
  95 #include <linux/posix-timers.h>
  96 #ifdef CONFIG_HARDWALL
  97 #include <asm/hardwall.h>
  98 #endif
  99 #include <trace/events/oom.h>
 100 #include "internal.h"
 101 #include "fd.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;
 114 static u8 nlink_tgid;
 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         wchan = get_wchan(task);
 392
 393         if (wchan && ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)
 394                         && !lookup_symbol_name(wchan, symname))
 395                 seq_printf(m, "%s", symname);
 396         else
 397                 seq_putc(m, '0');
 398
 399         return 0;
 400 }
 401 #endif /* CONFIG_KALLSYMS */
 402
 403 static int lock_trace(struct task_struct *task)
 404 {
 405         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
 406         if (err)
 407                 return err;
 408         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
 409                 mutex_unlock(&task->signal->cred_guard_mutex);
 410                 return -EPERM;
 411         }
 412         return 0;
 413 }
 414
 415 static void unlock_trace(struct task_struct *task)
 416 {
 417         mutex_unlock(&task->signal->cred_guard_mutex);
 418 }
 419
 420 #ifdef CONFIG_STACKTRACE
 421
 422 #define MAX_STACK_TRACE_DEPTH   64
 423
 424 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 425                           struct pid *pid, struct task_struct *task)
 426 {
 427         struct stack_trace trace;
 428         unsigned long *entries;
 429         int err;
 430         int i;
 431
 432         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 433         if (!entries)
 434                 return -ENOMEM;
 435
 436         trace.nr_entries        = 0;
 437         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 438         trace.entries           = entries;
 439         trace.skip              = 0;
 440
 441         err = lock_trace(task);
 442         if (!err) {
 443                 save_stack_trace_tsk(task, &trace);
 444
 445                 for (i = 0; i < trace.nr_entries; i++) {
 446                         seq_printf(m, "[<%pK>] %pB\n",
 447                                    (void *)entries[i], (void *)entries[i]);
 448                 }
 449                 unlock_trace(task);
 450         }
 451         kfree(entries);
 452
 453         return err;
 454 }
 455 #endif
 456
 457 #ifdef CONFIG_SCHED_INFO
 458 /*
 459  * Provides /proc/PID/schedstat
 460  */
 461 static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
 462                               struct pid *pid, struct task_struct *task)
 463 {
 464         if (unlikely(!sched_info_on()))
 465                 seq_printf(m, "0 0 0\n");
 466         else
 467                 seq_printf(m, "%llu %llu %lu\n",
 468                    (unsigned long long)task->se.sum_exec_runtime,
 469                    (unsigned long long)task->sched_info.run_delay,
 470                    task->sched_info.pcount);
 471
 472         return 0;
 473 }
 474 #endif
 475
 476 #ifdef CONFIG_LATENCYTOP
 477 static int lstats_show_proc(struct seq_file *m, void *v)
 478 {
 479         int i;
 480         struct inode *inode = m->private;
 481         struct task_struct *task = get_proc_task(inode);
 482
 483         if (!task)
 484                 return -ESRCH;
 485         seq_puts(m, "Latency Top version : v0.1\n");
 486         for (i = 0; i < 32; i++) {
 487                 struct latency_record *lr = &task->latency_record[i];
 488                 if (lr->backtrace[0]) {
 489                         int q;
 490                         seq_printf(m, "%i %li %li",
 491                                    lr->count, lr->time, lr->max);
 492                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 493                                 unsigned long bt = lr->backtrace[q];
 494                                 if (!bt)
 495                                         break;
 496                                 if (bt == ULONG_MAX)
 497                                         break;
 498                                 seq_printf(m, " %ps", (void *)bt);
 499                         }
 500                         seq_putc(m, '\n');
 501                 }
 502
 503         }
 504         put_task_struct(task);
 505         return 0;
 506 }
 507
 508 static int lstats_open(struct inode *inode, struct file *file)
 509 {
 510         return single_open(file, lstats_show_proc, inode);
 511 }
 512
 513 static ssize_t lstats_write(struct file *file, const char __user *buf,
 514                             size_t count, loff_t *offs)
 515 {
 516         struct task_struct *task = get_proc_task(file_inode(file));
 517
 518         if (!task)
 519                 return -ESRCH;
 520         clear_all_latency_tracing(task);
 521         put_task_struct(task);
 522
 523         return count;
 524 }
 525
 526 static const struct file_operations proc_lstats_operations = {
 527         .open           = lstats_open,
 528         .read           = seq_read,
 529         .write          = lstats_write,
 530         .llseek         = seq_lseek,
 531         .release        = single_release,
 532 };
 533
 534 #endif
 535
 536 static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
 537                           struct pid *pid, struct task_struct *task)
 538 {
 539         unsigned long totalpages = totalram_pages + total_swap_pages;
 540         unsigned long points = 0;
 541
 542         points = oom_badness(task, NULL, NULL, totalpages) *
 543                                         1000 / totalpages;
 544         seq_printf(m, "%lu\n", points);
 545
 546         return 0;
 547 }
 548
 549 struct limit_names {
 550         const char *name;
 551         const char *unit;
 552 };
 553
 554 static const struct limit_names lnames[RLIM_NLIMITS] = {
 555         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 556         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 557         [RLIMIT_DATA] = {"Max data size", "bytes"},
 558         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 559         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 560         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 561         [RLIMIT_NPROC] = {"Max processes", "processes"},
 562         [RLIMIT_NOFILE] = {"Max open files", "files"},
 563         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 564         [RLIMIT_AS] = {"Max address space", "bytes"},
 565         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 566         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 567         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 568         [RLIMIT_NICE] = {"Max nice priority", NULL},
 569         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 570         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 571 };
 572
 573 /* Display limits for a process */
 574 static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
 575                            struct pid *pid, struct task_struct *task)
 576 {
 577         unsigned int i;
 578         unsigned long flags;
 579
 580         struct rlimit rlim[RLIM_NLIMITS];
 581
 582         if (!lock_task_sighand(task, &flags))
 583                 return 0;
 584         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 585         unlock_task_sighand(task, &flags);
 586
 587         /*
 588          * print the file header
 589          */
 590        seq_printf(m, "%-25s %-20s %-20s %-10s\n",
 591                   "Limit", "Soft Limit", "Hard Limit", "Units");
 592
 593         for (i = 0; i < RLIM_NLIMITS; i++) {
 594                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 595                         seq_printf(m, "%-25s %-20s ",
 596                                    lnames[i].name, "unlimited");
 597                 else
 598                         seq_printf(m, "%-25s %-20lu ",
 599                                    lnames[i].name, rlim[i].rlim_cur);
 600
 601                 if (rlim[i].rlim_max == RLIM_INFINITY)
 602                         seq_printf(m, "%-20s ", "unlimited");
 603                 else
 604                         seq_printf(m, "%-20lu ", rlim[i].rlim_max);
 605
 606                 if (lnames[i].unit)
 607                         seq_printf(m, "%-10s\n", lnames[i].unit);
 608                 else
 609                         seq_putc(m, '\n');
 610         }
 611
 612         return 0;
 613 }
 614
 615 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 616 static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
 617                             struct pid *pid, struct task_struct *task)
 618 {
 619         long nr;
 620         unsigned long args[6], sp, pc;
 621         int res;
 622
 623         res = lock_trace(task);
 624         if (res)
 625                 return res;
 626
 627         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 628                 seq_puts(m, "running\n");
 629         else if (nr < 0)
 630                 seq_printf(m, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 631         else
 632                 seq_printf(m,
 633                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 634                        nr,
 635                        args[0], args[1], args[2], args[3], args[4], args[5],
 636                        sp, pc);
 637         unlock_trace(task);
 638
 639         return 0;
 640 }
 641 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 642
 643 /************************************************************************/
 644 /*                       Here the fs part begins                        */
 645 /************************************************************************/
 646
 647 /* permission checks */
 648 static int proc_fd_access_allowed(struct inode *inode)
 649 {
 650         struct task_struct *task;
 651         int allowed = 0;
 652         /* Allow access to a task's file descriptors if it is us or we
 653          * may use ptrace attach to the process and find out that
 654          * information.
 655          */
 656         task = get_proc_task(inode);
 657         if (task) {
 658                 allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 659                 put_task_struct(task);
 660         }
 661         return allowed;
 662 }
 663
 664 int proc_setattr(struct dentry *dentry, struct iattr *attr)
 665 {
 666         int error;
 667         struct inode *inode = d_inode(dentry);
 668
 669         if (attr->ia_valid & ATTR_MODE)
 670                 return -EPERM;
 671
 672         error = setattr_prepare(dentry, attr);
 673         if (error)
 674                 return error;
 675
 676         setattr_copy(inode, attr);
 677         mark_inode_dirty(inode);
 678         return 0;
 679 }
 680
 681 /*
 682  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
 683  * or euid/egid (for hide_pid_min=2)?
 684  */
 685 static bool has_pid_permissions(struct pid_namespace *pid,
 686                                  struct task_struct *task,
 687                                  int hide_pid_min)
 688 {
 689         if (pid->hide_pid < hide_pid_min)
 690                 return true;
 691         if (in_group_p(pid->pid_gid))
 692                 return true;
 693         return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 694 }
 695
 696
 697 static int proc_pid_permission(struct inode *inode, int mask)
 698 {
 699         struct pid_namespace *pid = inode->i_sb->s_fs_info;
 700         struct task_struct *task;
 701         bool has_perms;
 702
 703         task = get_proc_task(inode);
 704         if (!task)
 705                 return -ESRCH;
 706         has_perms = has_pid_permissions(pid, task, HIDEPID_NO_ACCESS);
 707         put_task_struct(task);
 708
 709         if (!has_perms) {
 710                 if (pid->hide_pid == HIDEPID_INVISIBLE) {
 711                         /*
 712                          * Let's make getdents(), stat(), and open()
 713                          * consistent with each other.  If a process
 714                          * may not stat() a file, it shouldn't be seen
 715                          * in procfs at all.
 716                          */
 717                         return -ENOENT;
 718                 }
 719
 720                 return -EPERM;
 721         }
 722         return generic_permission(inode, mask);
 723 }
 724
 725
 726
 727 static const struct inode_operations proc_def_inode_operations = {
 728         .setattr        = proc_setattr,
 729 };
 730
 731 static int proc_single_show(struct seq_file *m, void *v)
 732 {
 733         struct inode *inode = m->private;
 734         struct pid_namespace *ns;
 735         struct pid *pid;
 736         struct task_struct *task;
 737         int ret;
 738
 739         ns = inode->i_sb->s_fs_info;
 740         pid = proc_pid(inode);
 741         task = get_pid_task(pid, PIDTYPE_PID);
 742         if (!task)
 743                 return -ESRCH;
 744
 745         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 746
 747         put_task_struct(task);
 748         return ret;
 749 }
 750
 751 static int proc_single_open(struct inode *inode, struct file *filp)
 752 {
 753         return single_open(filp, proc_single_show, inode);
 754 }
 755
 756 static const struct file_operations proc_single_file_operations = {
 757         .open           = proc_single_open,
 758         .read           = seq_read,
 759         .llseek         = seq_lseek,
 760         .release        = single_release,
 761 };
 762
 763
 764 struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
 765 {
 766         struct task_struct *task = get_proc_task(inode);
 767         struct mm_struct *mm = ERR_PTR(-ESRCH);
 768
 769         if (task) {
 770                 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
 771                 put_task_struct(task);
 772
 773                 if (!IS_ERR_OR_NULL(mm)) {
 774                         /* ensure this mm_struct can't be freed */
 775                         mmgrab(mm);
 776                         /* but do not pin its memory */
 777                         mmput(mm);
 778                 }
 779         }
 780
 781         return mm;
 782 }
 783
 784 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
 785 {
 786         struct mm_struct *mm = proc_mem_open(inode, mode);
 787
 788         if (IS_ERR(mm))
 789                 return PTR_ERR(mm);
 790
 791         file->private_data = mm;
 792         return 0;
 793 }
 794
 795 static int mem_open(struct inode *inode, struct file *file)
 796 {
 797         int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
 798
 799         /* OK to pass negative loff_t, we can catch out-of-range */
 800         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 801
 802         return ret;
 803 }
 804
 805 static ssize_t mem_rw(struct file *file, char __user *buf,
 806                         size_t count, loff_t *ppos, int write)
 807 {
 808         struct mm_struct *mm = file->private_data;
 809         unsigned long addr = *ppos;
 810         ssize_t copied;
 811         char *page;
 812         unsigned int flags;
 813
 814         if (!mm)
 815                 return 0;
 816
 817         page = (char *)__get_free_page(GFP_KERNEL);
 818         if (!page)
 819                 return -ENOMEM;
 820
 821         copied = 0;
 822         if (!mmget_not_zero(mm))
 823                 goto free;
 824
 825         flags = FOLL_FORCE | (write ? FOLL_WRITE : 0);
 826
 827         while (count > 0) {
 828                 int this_len = min_t(int, count, PAGE_SIZE);
 829
 830                 if (write && copy_from_user(page, buf, this_len)) {
 831                         copied = -EFAULT;
 832                         break;
 833                 }
 834
 835                 this_len = access_remote_vm(mm, addr, page, this_len, flags);
 836                 if (!this_len) {
 837                         if (!copied)
 838                                 copied = -EIO;
 839                         break;
 840                 }
 841
 842                 if (!write && copy_to_user(buf, page, this_len)) {
 843                         copied = -EFAULT;
 844                         break;
 845                 }
 846
 847                 buf += this_len;
 848                 addr += this_len;
 849                 copied += this_len;
 850                 count -= this_len;
 851         }
 852         *ppos = addr;
 853
 854         mmput(mm);
 855 free:
 856         free_page((unsigned long) page);
 857         return copied;
 858 }
 859
 860 static ssize_t mem_read(struct file *file, char __user *buf,
 861                         size_t count, loff_t *ppos)
 862 {
 863         return mem_rw(file, buf, count, ppos, 0);
 864 }
 865
 866 static ssize_t mem_write(struct file *file, const char __user *buf,
 867                          size_t count, loff_t *ppos)
 868 {
 869         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 870 }
 871
 872 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 873 {
 874         switch (orig) {
 875         case 0:
 876                 file->f_pos = offset;
 877                 break;
 878         case 1:
 879                 file->f_pos += offset;
 880                 break;
 881         default:
 882                 return -EINVAL;
 883         }
 884         force_successful_syscall_return();
 885         return file->f_pos;
 886 }
 887
 888 static int mem_release(struct inode *inode, struct file *file)
 889 {
 890         struct mm_struct *mm = file->private_data;
 891         if (mm)
 892                 mmdrop(mm);
 893         return 0;
 894 }
 895
 896 static const struct file_operations proc_mem_operations = {
 897         .llseek         = mem_lseek,
 898         .read           = mem_read,
 899         .write          = mem_write,
 900         .open           = mem_open,
 901         .release        = mem_release,
 902 };
 903
 904 static int environ_open(struct inode *inode, struct file *file)
 905 {
 906         return __mem_open(inode, file, PTRACE_MODE_READ);
 907 }
 908
 909 static ssize_t environ_read(struct file *file, char __user *buf,
 910                         size_t count, loff_t *ppos)
 911 {
 912         char *page;
 913         unsigned long src = *ppos;
 914         int ret = 0;
 915         struct mm_struct *mm = file->private_data;
 916         unsigned long env_start, env_end;
 917
 918         /* Ensure the process spawned far enough to have an environment. */
 919         if (!mm || !mm->env_end)
 920                 return 0;
 921
 922         page = (char *)__get_free_page(GFP_KERNEL);
 923         if (!page)
 924                 return -ENOMEM;
 925
 926         ret = 0;
 927         if (!mmget_not_zero(mm))
 928                 goto free;
 929
 930         down_read(&mm->mmap_sem);
 931         env_start = mm->env_start;
 932         env_end = mm->env_end;
 933         up_read(&mm->mmap_sem);
 934
 935         while (count > 0) {
 936                 size_t this_len, max_len;
 937                 int retval;
 938
 939                 if (src >= (env_end - env_start))
 940                         break;
 941
 942                 this_len = env_end - (env_start + src);
 943
 944                 max_len = min_t(size_t, PAGE_SIZE, count);
 945                 this_len = min(max_len, this_len);
 946
 947                 retval = access_remote_vm(mm, (env_start + src), page, this_len, 0);
 948
 949                 if (retval <= 0) {
 950                         ret = retval;
 951                         break;
 952                 }
 953
 954                 if (copy_to_user(buf, page, retval)) {
 955                         ret = -EFAULT;
 956                         break;
 957                 }
 958
 959                 ret += retval;
 960                 src += retval;
 961                 buf += retval;
 962                 count -= retval;
 963         }
 964         *ppos = src;
 965         mmput(mm);
 966
 967 free:
 968         free_page((unsigned long) page);
 969         return ret;
 970 }
 971
 972 static const struct file_operations proc_environ_operations = {
 973         .open           = environ_open,
 974         .read           = environ_read,
 975         .llseek         = generic_file_llseek,
 976         .release        = mem_release,
 977 };
 978
 979 static int auxv_open(struct inode *inode, struct file *file)
 980 {
 981         return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
 982 }
 983
 984 static ssize_t auxv_read(struct file *file, char __user *buf,
 985                         size_t count, loff_t *ppos)
 986 {
 987         struct mm_struct *mm = file->private_data;
 988         unsigned int nwords = 0;
 989
 990         if (!mm)
 991                 return 0;
 992         do {
 993                 nwords += 2;
 994         } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 995         return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
 996                                        nwords * sizeof(mm->saved_auxv[0]));
 997 }
 998
 999 static const struct file_operations proc_auxv_operations = {
1000         .open           = auxv_open,
1001         .read           = auxv_read,
1002         .llseek         = generic_file_llseek,
1003         .release        = mem_release,
1004 };
1005
1006 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1007                             loff_t *ppos)
1008 {
1009         struct task_struct *task = get_proc_task(file_inode(file));
1010         char buffer[PROC_NUMBUF];
1011         int oom_adj = OOM_ADJUST_MIN;
1012         size_t len;
1013
1014         if (!task)
1015                 return -ESRCH;
1016         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1017                 oom_adj = OOM_ADJUST_MAX;
1018         else
1019                 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1020                           OOM_SCORE_ADJ_MAX;
1021         put_task_struct(task);
1022         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1023         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1024 }
1025
1026 static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1027 {
1028         static DEFINE_MUTEX(oom_adj_mutex);
1029         struct mm_struct *mm = NULL;
1030         struct task_struct *task;
1031         int err = 0;
1032
1033         task = get_proc_task(file_inode(file));
1034         if (!task)
1035                 return -ESRCH;
1036
1037         mutex_lock(&oom_adj_mutex);
1038         if (legacy) {
1039                 if (oom_adj < task->signal->oom_score_adj &&
1040                                 !capable(CAP_SYS_RESOURCE)) {
1041                         err = -EACCES;
1042                         goto err_unlock;
1043                 }
1044                 /*
1045                  * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1046                  * /proc/pid/oom_score_adj instead.
1047                  */
1048                 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1049                           current->comm, task_pid_nr(current), task_pid_nr(task),
1050                           task_pid_nr(task));
1051         } else {
1052                 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1053                                 !capable(CAP_SYS_RESOURCE)) {
1054                         err = -EACCES;
1055                         goto err_unlock;
1056                 }
1057         }
1058
1059         /*
1060          * Make sure we will check other processes sharing the mm if this is
1061          * not vfrok which wants its own oom_score_adj.
1062          * pin the mm so it doesn't go away and get reused after task_unlock
1063          */
1064         if (!task->vfork_done) {
1065                 struct task_struct *p = find_lock_task_mm(task);
1066
1067                 if (p) {
1068                         if (atomic_read(&p->mm->mm_users) > 1) {
1069                                 mm = p->mm;
1070                                 mmgrab(mm);
1071                         }
1072                         task_unlock(p);
1073                 }
1074         }
1075
1076         task->signal->oom_score_adj = oom_adj;
1077         if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1078                 task->signal->oom_score_adj_min = (short)oom_adj;
1079         trace_oom_score_adj_update(task);
1080
1081         if (mm) {
1082                 struct task_struct *p;
1083
1084                 rcu_read_lock();
1085                 for_each_process(p) {
1086                         if (same_thread_group(task, p))
1087                                 continue;
1088
1089                         /* do not touch kernel threads or the global init */
1090                         if (p->flags & PF_KTHREAD || is_global_init(p))
1091                                 continue;
1092
1093                         task_lock(p);
1094                         if (!p->vfork_done && process_shares_mm(p, mm)) {
1095                                 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",
1096                                                 task_pid_nr(p), p->comm,
1097                                                 p->signal->oom_score_adj, oom_adj,
1098                                                 task_pid_nr(task), task->comm);
1099                                 p->signal->oom_score_adj = oom_adj;
1100                                 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1101                                         p->signal->oom_score_adj_min = (short)oom_adj;
1102                         }
1103                         task_unlock(p);
1104                 }
1105                 rcu_read_unlock();
1106                 mmdrop(mm);
1107         }
1108 err_unlock:
1109         mutex_unlock(&oom_adj_mutex);
1110         put_task_struct(task);
1111         return err;
1112 }
1113
1114 /*
1115  * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1116  * kernels.  The effective policy is defined by oom_score_adj, which has a
1117  * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1118  * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1119  * Processes that become oom disabled via oom_adj will still be oom disabled
1120  * with this implementation.
1121  *
1122  * oom_adj cannot be removed since existing userspace binaries use it.
1123  */
1124 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1125                              size_t count, loff_t *ppos)
1126 {
1127         char buffer[PROC_NUMBUF];
1128         int oom_adj;
1129         int err;
1130
1131         memset(buffer, 0, sizeof(buffer));
1132         if (count > sizeof(buffer) - 1)
1133                 count = sizeof(buffer) - 1;
1134         if (copy_from_user(buffer, buf, count)) {
1135                 err = -EFAULT;
1136                 goto out;
1137         }
1138
1139         err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1140         if (err)
1141                 goto out;
1142         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1143              oom_adj != OOM_DISABLE) {
1144                 err = -EINVAL;
1145                 goto out;
1146         }
1147
1148         /*
1149          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1150          * value is always attainable.
1151          */
1152         if (oom_adj == OOM_ADJUST_MAX)
1153                 oom_adj = OOM_SCORE_ADJ_MAX;
1154         else
1155                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1156
1157         err = __set_oom_adj(file, oom_adj, true);
1158 out:
1159         return err < 0 ? err : count;
1160 }
1161
1162 static const struct file_operations proc_oom_adj_operations = {
1163         .read           = oom_adj_read,
1164         .write          = oom_adj_write,
1165         .llseek         = generic_file_llseek,
1166 };
1167
1168 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1169                                         size_t count, loff_t *ppos)
1170 {
1171         struct task_struct *task = get_proc_task(file_inode(file));
1172         char buffer[PROC_NUMBUF];
1173         short oom_score_adj = OOM_SCORE_ADJ_MIN;
1174         size_t len;
1175
1176         if (!task)
1177                 return -ESRCH;
1178         oom_score_adj = task->signal->oom_score_adj;
1179         put_task_struct(task);
1180         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1181         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1182 }
1183
1184 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1185                                         size_t count, loff_t *ppos)
1186 {
1187         char buffer[PROC_NUMBUF];
1188         int oom_score_adj;
1189         int err;
1190
1191         memset(buffer, 0, sizeof(buffer));
1192         if (count > sizeof(buffer) - 1)
1193                 count = sizeof(buffer) - 1;
1194         if (copy_from_user(buffer, buf, count)) {
1195                 err = -EFAULT;
1196                 goto out;
1197         }
1198
1199         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1200         if (err)
1201                 goto out;
1202         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1203                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1204                 err = -EINVAL;
1205                 goto out;
1206         }
1207
1208         err = __set_oom_adj(file, oom_score_adj, false);
1209 out:
1210         return err < 0 ? err : count;
1211 }
1212
1213 static const struct file_operations proc_oom_score_adj_operations = {
1214         .read           = oom_score_adj_read,
1215         .write          = oom_score_adj_write,
1216         .llseek         = default_llseek,
1217 };
1218
1219 #ifdef CONFIG_AUDITSYSCALL
1220 #define TMPBUFLEN 11
1221 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1222                                   size_t count, loff_t *ppos)
1223 {
1224         struct inode * inode = file_inode(file);
1225         struct task_struct *task = get_proc_task(inode);
1226         ssize_t length;
1227         char tmpbuf[TMPBUFLEN];
1228
1229         if (!task)
1230                 return -ESRCH;
1231         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1232                            from_kuid(file->f_cred->user_ns,
1233                                      audit_get_loginuid(task)));
1234         put_task_struct(task);
1235         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1236 }
1237
1238 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1239                                    size_t count, loff_t *ppos)
1240 {
1241         struct inode * inode = file_inode(file);
1242         uid_t loginuid;
1243         kuid_t kloginuid;
1244         int rv;
1245
1246         rcu_read_lock();
1247         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1248                 rcu_read_unlock();
1249                 return -EPERM;
1250         }
1251         rcu_read_unlock();
1252
1253         if (*ppos != 0) {
1254                 /* No partial writes. */
1255                 return -EINVAL;
1256         }
1257
1258         rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1259         if (rv < 0)
1260                 return rv;
1261
1262         /* is userspace tring to explicitly UNSET the loginuid? */
1263         if (loginuid == AUDIT_UID_UNSET) {
1264                 kloginuid = INVALID_UID;
1265         } else {
1266                 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1267                 if (!uid_valid(kloginuid))
1268                         return -EINVAL;
1269         }
1270
1271         rv = audit_set_loginuid(kloginuid);
1272         if (rv < 0)
1273                 return rv;
1274         return count;
1275 }
1276
1277 static const struct file_operations proc_loginuid_operations = {
1278         .read           = proc_loginuid_read,
1279         .write          = proc_loginuid_write,
1280         .llseek         = generic_file_llseek,
1281 };
1282
1283 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1284                                   size_t count, loff_t *ppos)
1285 {
1286         struct inode * inode = file_inode(file);
1287         struct task_struct *task = get_proc_task(inode);
1288         ssize_t length;
1289         char tmpbuf[TMPBUFLEN];
1290
1291         if (!task)
1292                 return -ESRCH;
1293         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1294                                 audit_get_sessionid(task));
1295         put_task_struct(task);
1296         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1297 }
1298
1299 static const struct file_operations proc_sessionid_operations = {
1300         .read           = proc_sessionid_read,
1301         .llseek         = generic_file_llseek,
1302 };
1303 #endif
1304
1305 #ifdef CONFIG_FAULT_INJECTION
1306 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1307                                       size_t count, loff_t *ppos)
1308 {
1309         struct task_struct *task = get_proc_task(file_inode(file));
1310         char buffer[PROC_NUMBUF];
1311         size_t len;
1312         int make_it_fail;
1313
1314         if (!task)
1315                 return -ESRCH;
1316         make_it_fail = task->make_it_fail;
1317         put_task_struct(task);
1318
1319         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1320
1321         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1322 }
1323
1324 static ssize_t proc_fault_inject_write(struct file * file,
1325                         const char __user * buf, size_t count, loff_t *ppos)
1326 {
1327         struct task_struct *task;
1328         char buffer[PROC_NUMBUF];
1329         int make_it_fail;
1330         int rv;
1331
1332         if (!capable(CAP_SYS_RESOURCE))
1333                 return -EPERM;
1334         memset(buffer, 0, sizeof(buffer));
1335         if (count > sizeof(buffer) - 1)
1336                 count = sizeof(buffer) - 1;
1337         if (copy_from_user(buffer, buf, count))
1338                 return -EFAULT;
1339         rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1340         if (rv < 0)
1341                 return rv;
1342         if (make_it_fail < 0 || make_it_fail > 1)
1343                 return -EINVAL;
1344
1345         task = get_proc_task(file_inode(file));
1346         if (!task)
1347                 return -ESRCH;
1348         task->make_it_fail = make_it_fail;
1349         put_task_struct(task);
1350
1351         return count;
1352 }
1353
1354 static const struct file_operations proc_fault_inject_operations = {
1355         .read           = proc_fault_inject_read,
1356         .write          = proc_fault_inject_write,
1357         .llseek         = generic_file_llseek,
1358 };
1359
1360 static ssize_t proc_fail_nth_write(struct file *file, const char __user *buf,
1361                                    size_t count, loff_t *ppos)
1362 {
1363         struct task_struct *task;
1364         int err;
1365         unsigned int n;
1366
1367         err = kstrtouint_from_user(buf, count, 0, &n);
1368         if (err)
1369                 return err;
1370
1371         task = get_proc_task(file_inode(file));
1372         if (!task)
1373                 return -ESRCH;
1374         WRITE_ONCE(task->fail_nth, n);
1375         put_task_struct(task);
1376
1377         return count;
1378 }
1379
1380 static ssize_t proc_fail_nth_read(struct file *file, char __user *buf,
1381                                   size_t count, loff_t *ppos)
1382 {
1383         struct task_struct *task;
1384         char numbuf[PROC_NUMBUF];
1385         ssize_t len;
1386
1387         task = get_proc_task(file_inode(file));
1388         if (!task)
1389                 return -ESRCH;
1390         len = snprintf(numbuf, sizeof(numbuf), "%u\n",
1391                         READ_ONCE(task->fail_nth));
1392         len = simple_read_from_buffer(buf, count, ppos, numbuf, len);
1393         put_task_struct(task);
1394
1395         return len;
1396 }
1397
1398 static const struct file_operations proc_fail_nth_operations = {
1399         .read           = proc_fail_nth_read,
1400         .write          = proc_fail_nth_write,
1401 };
1402 #endif
1403
1404
1405 #ifdef CONFIG_SCHED_DEBUG
1406 /*
1407  * Print out various scheduling related per-task fields:
1408  */
1409 static int sched_show(struct seq_file *m, void *v)
1410 {
1411         struct inode *inode = m->private;
1412         struct pid_namespace *ns = inode->i_sb->s_fs_info;
1413         struct task_struct *p;
1414
1415         p = get_proc_task(inode);
1416         if (!p)
1417                 return -ESRCH;
1418         proc_sched_show_task(p, ns, m);
1419
1420         put_task_struct(p);
1421
1422         return 0;
1423 }
1424
1425 static ssize_t
1426 sched_write(struct file *file, const char __user *buf,
1427             size_t count, loff_t *offset)
1428 {
1429         struct inode *inode = file_inode(file);
1430         struct task_struct *p;
1431
1432         p = get_proc_task(inode);
1433         if (!p)
1434                 return -ESRCH;
1435         proc_sched_set_task(p);
1436
1437         put_task_struct(p);
1438
1439         return count;
1440 }
1441
1442 static int sched_open(struct inode *inode, struct file *filp)
1443 {
1444         return single_open(filp, sched_show, inode);
1445 }
1446
1447 static const struct file_operations proc_pid_sched_operations = {
1448         .open           = sched_open,
1449         .read           = seq_read,
1450         .write          = sched_write,
1451         .llseek         = seq_lseek,
1452         .release        = single_release,
1453 };
1454
1455 #endif
1456
1457 #ifdef CONFIG_SCHED_AUTOGROUP
1458 /*
1459  * Print out autogroup related information:
1460  */
1461 static int sched_autogroup_show(struct seq_file *m, void *v)
1462 {
1463         struct inode *inode = m->private;
1464         struct task_struct *p;
1465
1466         p = get_proc_task(inode);
1467         if (!p)
1468                 return -ESRCH;
1469         proc_sched_autogroup_show_task(p, m);
1470
1471         put_task_struct(p);
1472
1473         return 0;
1474 }
1475
1476 static ssize_t
1477 sched_autogroup_write(struct file *file, const char __user *buf,
1478             size_t count, loff_t *offset)
1479 {
1480         struct inode *inode = file_inode(file);
1481         struct task_struct *p;
1482         char buffer[PROC_NUMBUF];
1483         int nice;
1484         int err;
1485
1486         memset(buffer, 0, sizeof(buffer));
1487         if (count > sizeof(buffer) - 1)
1488                 count = sizeof(buffer) - 1;
1489         if (copy_from_user(buffer, buf, count))
1490                 return -EFAULT;
1491
1492         err = kstrtoint(strstrip(buffer), 0, &nice);
1493         if (err < 0)
1494                 return err;
1495
1496         p = get_proc_task(inode);
1497         if (!p)
1498                 return -ESRCH;
1499
1500         err = proc_sched_autogroup_set_nice(p, nice);
1501         if (err)
1502                 count = err;
1503
1504         put_task_struct(p);
1505
1506         return count;
1507 }
1508
1509 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1510 {
1511         int ret;
1512
1513         ret = single_open(filp, sched_autogroup_show, NULL);
1514         if (!ret) {
1515                 struct seq_file *m = filp->private_data;
1516
1517                 m->private = inode;
1518         }
1519         return ret;
1520 }
1521
1522 static const struct file_operations proc_pid_sched_autogroup_operations = {
1523         .open           = sched_autogroup_open,
1524         .read           = seq_read,
1525         .write          = sched_autogroup_write,
1526         .llseek         = seq_lseek,
1527         .release        = single_release,
1528 };
1529
1530 #endif /* CONFIG_SCHED_AUTOGROUP */
1531
1532 static ssize_t comm_write(struct file *file, const char __user *buf,
1533                                 size_t count, loff_t *offset)
1534 {
1535         struct inode *inode = file_inode(file);
1536         struct task_struct *p;
1537         char buffer[TASK_COMM_LEN];
1538         const size_t maxlen = sizeof(buffer) - 1;
1539
1540         memset(buffer, 0, sizeof(buffer));
1541         if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1542                 return -EFAULT;
1543
1544         p = get_proc_task(inode);
1545         if (!p)
1546                 return -ESRCH;
1547
1548         if (same_thread_group(current, p))
1549                 set_task_comm(p, buffer);
1550         else
1551                 count = -EINVAL;
1552
1553         put_task_struct(p);
1554
1555         return count;
1556 }
1557
1558 static int comm_show(struct seq_file *m, void *v)
1559 {
1560         struct inode *inode = m->private;
1561         struct task_struct *p;
1562
1563         p = get_proc_task(inode);
1564         if (!p)
1565                 return -ESRCH;
1566
1567         task_lock(p);
1568         seq_printf(m, "%s\n", p->comm);
1569         task_unlock(p);
1570
1571         put_task_struct(p);
1572
1573         return 0;
1574 }
1575
1576 static int comm_open(struct inode *inode, struct file *filp)
1577 {
1578         return single_open(filp, comm_show, inode);
1579 }
1580
1581 static const struct file_operations proc_pid_set_comm_operations = {
1582         .open           = comm_open,
1583         .read           = seq_read,
1584         .write          = comm_write,
1585         .llseek         = seq_lseek,
1586         .release        = single_release,
1587 };
1588
1589 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1590 {
1591         struct task_struct *task;
1592         struct file *exe_file;
1593
1594         task = get_proc_task(d_inode(dentry));
1595         if (!task)
1596                 return -ENOENT;
1597         exe_file = get_task_exe_file(task);
1598         put_task_struct(task);
1599         if (exe_file) {
1600                 *exe_path = exe_file->f_path;
1601                 path_get(&exe_file->f_path);
1602                 fput(exe_file);
1603                 return 0;
1604         } else
1605                 return -ENOENT;
1606 }
1607
1608 static const char *proc_pid_get_link(struct dentry *dentry,
1609                                      struct inode *inode,
1610                                      struct delayed_call *done)
1611 {
1612         struct path path;
1613         int error = -EACCES;
1614
1615         if (!dentry)
1616                 return ERR_PTR(-ECHILD);
1617
1618         /* Are we allowed to snoop on the tasks file descriptors? */
1619         if (!proc_fd_access_allowed(inode))
1620                 goto out;
1621
1622         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1623         if (error)
1624                 goto out;
1625
1626         nd_jump_link(&path);
1627         return NULL;
1628 out:
1629         return ERR_PTR(error);
1630 }
1631
1632 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1633 {
1634         char *tmp = (char *)__get_free_page(GFP_KERNEL);
1635         char *pathname;
1636         int len;
1637
1638         if (!tmp)
1639                 return -ENOMEM;
1640
1641         pathname = d_path(path, tmp, PAGE_SIZE);
1642         len = PTR_ERR(pathname);
1643         if (IS_ERR(pathname))
1644                 goto out;
1645         len = tmp + PAGE_SIZE - 1 - pathname;
1646
1647         if (len > buflen)
1648                 len = buflen;
1649         if (copy_to_user(buffer, pathname, len))
1650                 len = -EFAULT;
1651  out:
1652         free_page((unsigned long)tmp);
1653         return len;
1654 }
1655
1656 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1657 {
1658         int error = -EACCES;
1659         struct inode *inode = d_inode(dentry);
1660         struct path path;
1661
1662         /* Are we allowed to snoop on the tasks file descriptors? */
1663         if (!proc_fd_access_allowed(inode))
1664                 goto out;
1665
1666         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1667         if (error)
1668                 goto out;
1669
1670         error = do_proc_readlink(&path, buffer, buflen);
1671         path_put(&path);
1672 out:
1673         return error;
1674 }
1675
1676 const struct inode_operations proc_pid_link_inode_operations = {
1677         .readlink       = proc_pid_readlink,
1678         .get_link       = proc_pid_get_link,
1679         .setattr        = proc_setattr,
1680 };
1681
1682
1683 /* building an inode */
1684
1685 void task_dump_owner(struct task_struct *task, mode_t mode,
1686                      kuid_t *ruid, kgid_t *rgid)
1687 {
1688         /* Depending on the state of dumpable compute who should own a
1689          * proc file for a task.
1690          */
1691         const struct cred *cred;
1692         kuid_t uid;
1693         kgid_t gid;
1694
1695         /* Default to the tasks effective ownership */
1696         rcu_read_lock();
1697         cred = __task_cred(task);
1698         uid = cred->euid;
1699         gid = cred->egid;
1700         rcu_read_unlock();
1701
1702         /*
1703          * Before the /proc/pid/status file was created the only way to read
1704          * the effective uid of a /process was to stat /proc/pid.  Reading
1705          * /proc/pid/status is slow enough that procps and other packages
1706          * kept stating /proc/pid.  To keep the rules in /proc simple I have
1707          * made this apply to all per process world readable and executable
1708          * directories.
1709          */
1710         if (mode != (S_IFDIR|S_IRUGO|S_IXUGO)) {
1711                 struct mm_struct *mm;
1712                 task_lock(task);
1713                 mm = task->mm;
1714                 /* Make non-dumpable tasks owned by some root */
1715                 if (mm) {
1716                         if (get_dumpable(mm) != SUID_DUMP_USER) {
1717                                 struct user_namespace *user_ns = mm->user_ns;
1718
1719                                 uid = make_kuid(user_ns, 0);
1720                                 if (!uid_valid(uid))
1721                                         uid = GLOBAL_ROOT_UID;
1722
1723                                 gid = make_kgid(user_ns, 0);
1724                                 if (!gid_valid(gid))
1725                                         gid = GLOBAL_ROOT_GID;
1726                         }
1727                 } else {
1728                         uid = GLOBAL_ROOT_UID;
1729                         gid = GLOBAL_ROOT_GID;
1730                 }
1731                 task_unlock(task);
1732         }
1733         *ruid = uid;
1734         *rgid = gid;
1735 }
1736
1737 struct inode *proc_pid_make_inode(struct super_block * sb,
1738                                   struct task_struct *task, umode_t mode)
1739 {
1740         struct inode * inode;
1741         struct proc_inode *ei;
1742
1743         /* We need a new inode */
1744
1745         inode = new_inode(sb);
1746         if (!inode)
1747                 goto out;
1748
1749         /* Common stuff */
1750         ei = PROC_I(inode);
1751         inode->i_mode = mode;
1752         inode->i_ino = get_next_ino();
1753         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1754         inode->i_op = &proc_def_inode_operations;
1755
1756         /*
1757          * grab the reference to task.
1758          */
1759         ei->pid = get_task_pid(task, PIDTYPE_PID);
1760         if (!ei->pid)
1761                 goto out_unlock;
1762
1763         task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1764         security_task_to_inode(task, inode);
1765
1766 out:
1767         return inode;
1768
1769 out_unlock:
1770         iput(inode);
1771         return NULL;
1772 }
1773
1774 int pid_getattr(const struct path *path, struct kstat *stat,
1775                 u32 request_mask, unsigned int query_flags)
1776 {
1777         struct inode *inode = d_inode(path->dentry);
1778         struct task_struct *task;
1779         struct pid_namespace *pid = path->dentry->d_sb->s_fs_info;
1780
1781         generic_fillattr(inode, stat);
1782
1783         rcu_read_lock();
1784         stat->uid = GLOBAL_ROOT_UID;
1785         stat->gid = GLOBAL_ROOT_GID;
1786         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1787         if (task) {
1788                 if (!has_pid_permissions(pid, task, HIDEPID_INVISIBLE)) {
1789                         rcu_read_unlock();
1790                         /*
1791                          * This doesn't prevent learning whether PID exists,
1792                          * it only makes getattr() consistent with readdir().
1793                          */
1794                         return -ENOENT;
1795                 }
1796                 task_dump_owner(task, inode->i_mode, &stat->uid, &stat->gid);
1797         }
1798         rcu_read_unlock();
1799         return 0;
1800 }
1801
1802 /* dentry stuff */
1803
1804 /*
1805  *      Exceptional case: normally we are not allowed to unhash a busy
1806  * directory. In this case, however, we can do it - no aliasing problems
1807  * due to the way we treat inodes.
1808  *
1809  * Rewrite the inode's ownerships here because the owning task may have
1810  * performed a setuid(), etc.
1811  *
1812  */
1813 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1814 {
1815         struct inode *inode;
1816         struct task_struct *task;
1817
1818         if (flags & LOOKUP_RCU)
1819                 return -ECHILD;
1820
1821         inode = d_inode(dentry);
1822         task = get_proc_task(inode);
1823
1824         if (task) {
1825                 task_dump_owner(task, inode->i_mode, &inode->i_uid, &inode->i_gid);
1826
1827                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1828                 security_task_to_inode(task, inode);
1829                 put_task_struct(task);
1830                 return 1;
1831         }
1832         return 0;
1833 }
1834
1835 static inline bool proc_inode_is_dead(struct inode *inode)
1836 {
1837         return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1838 }
1839
1840 int pid_delete_dentry(const struct dentry *dentry)
1841 {
1842         /* Is the task we represent dead?
1843          * If so, then don't put the dentry on the lru list,
1844          * kill it immediately.
1845          */
1846         return proc_inode_is_dead(d_inode(dentry));
1847 }
1848
1849 const struct dentry_operations pid_dentry_operations =
1850 {
1851         .d_revalidate   = pid_revalidate,
1852         .d_delete       = pid_delete_dentry,
1853 };
1854
1855 /* Lookups */
1856
1857 /*
1858  * Fill a directory entry.
1859  *
1860  * If possible create the dcache entry and derive our inode number and
1861  * file type from dcache entry.
1862  *
1863  * Since all of the proc inode numbers are dynamically generated, the inode
1864  * numbers do not exist until the inode is cache.  This means creating the
1865  * the dcache entry in readdir is necessary to keep the inode numbers
1866  * reported by readdir in sync with the inode numbers reported
1867  * by stat.
1868  */
1869 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1870         const char *name, int len,
1871         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1872 {
1873         struct dentry *child, *dir = file->f_path.dentry;
1874         struct qstr qname = QSTR_INIT(name, len);
1875         struct inode *inode;
1876         unsigned type;
1877         ino_t ino;
1878
1879         child = d_hash_and_lookup(dir, &qname);
1880         if (!child) {
1881                 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1882                 child = d_alloc_parallel(dir, &qname, &wq);
1883                 if (IS_ERR(child))
1884                         goto end_instantiate;
1885                 if (d_in_lookup(child)) {
1886                         int err = instantiate(d_inode(dir), child, task, ptr);
1887                         d_lookup_done(child);
1888                         if (err < 0) {
1889                                 dput(child);
1890                                 goto end_instantiate;
1891                         }
1892                 }
1893         }
1894         inode = d_inode(child);
1895         ino = inode->i_ino;
1896         type = inode->i_mode >> 12;
1897         dput(child);
1898         return dir_emit(ctx, name, len, ino, type);
1899
1900 end_instantiate:
1901         return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1902 }
1903
1904 /*
1905  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1906  * which represent vma start and end addresses.
1907  */
1908 static int dname_to_vma_addr(struct dentry *dentry,
1909                              unsigned long *start, unsigned long *end)
1910 {
1911         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1912                 return -EINVAL;
1913
1914         return 0;
1915 }
1916
1917 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1918 {
1919         unsigned long vm_start, vm_end;
1920         bool exact_vma_exists = false;
1921         struct mm_struct *mm = NULL;
1922         struct task_struct *task;
1923         struct inode *inode;
1924         int status = 0;
1925
1926         if (flags & LOOKUP_RCU)
1927                 return -ECHILD;
1928
1929         inode = d_inode(dentry);
1930         task = get_proc_task(inode);
1931         if (!task)
1932                 goto out_notask;
1933
1934         mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1935         if (IS_ERR_OR_NULL(mm))
1936                 goto out;
1937
1938         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1939                 down_read(&mm->mmap_sem);
1940                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1941                 up_read(&mm->mmap_sem);
1942         }
1943
1944         mmput(mm);
1945
1946         if (exact_vma_exists) {
1947                 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1948
1949                 security_task_to_inode(task, inode);
1950                 status = 1;
1951         }
1952
1953 out:
1954         put_task_struct(task);
1955
1956 out_notask:
1957         return status;
1958 }
1959
1960 static const struct dentry_operations tid_map_files_dentry_operations = {
1961         .d_revalidate   = map_files_d_revalidate,
1962         .d_delete       = pid_delete_dentry,
1963 };
1964
1965 static int map_files_get_link(struct dentry *dentry, struct path *path)
1966 {
1967         unsigned long vm_start, vm_end;
1968         struct vm_area_struct *vma;
1969         struct task_struct *task;
1970         struct mm_struct *mm;
1971         int rc;
1972
1973         rc = -ENOENT;
1974         task = get_proc_task(d_inode(dentry));
1975         if (!task)
1976                 goto out;
1977
1978         mm = get_task_mm(task);
1979         put_task_struct(task);
1980         if (!mm)
1981                 goto out;
1982
1983         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1984         if (rc)
1985                 goto out_mmput;
1986
1987         rc = -ENOENT;
1988         down_read(&mm->mmap_sem);
1989         vma = find_exact_vma(mm, vm_start, vm_end);
1990         if (vma && vma->vm_file) {
1991                 *path = vma->vm_file->f_path;
1992                 path_get(path);
1993                 rc = 0;
1994         }
1995         up_read(&mm->mmap_sem);
1996
1997 out_mmput:
1998         mmput(mm);
1999 out:
2000         return rc;
2001 }
2002
2003 struct map_files_info {
2004         fmode_t         mode;
2005         unsigned int    len;
2006         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
2007 };
2008
2009 /*
2010  * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
2011  * symlinks may be used to bypass permissions on ancestor directories in the
2012  * path to the file in question.
2013  */
2014 static const char *
2015 proc_map_files_get_link(struct dentry *dentry,
2016                         struct inode *inode,
2017                         struct delayed_call *done)
2018 {
2019         if (!capable(CAP_SYS_ADMIN))
2020                 return ERR_PTR(-EPERM);
2021
2022         return proc_pid_get_link(dentry, inode, done);
2023 }
2024
2025 /*
2026  * Identical to proc_pid_link_inode_operations except for get_link()
2027  */
2028 static const struct inode_operations proc_map_files_link_inode_operations = {
2029         .readlink       = proc_pid_readlink,
2030         .get_link       = proc_map_files_get_link,
2031         .setattr        = proc_setattr,
2032 };
2033
2034 static int
2035 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
2036                            struct task_struct *task, const void *ptr)
2037 {
2038         fmode_t mode = (fmode_t)(unsigned long)ptr;
2039         struct proc_inode *ei;
2040         struct inode *inode;
2041
2042         inode = proc_pid_make_inode(dir->i_sb, task, S_IFLNK |
2043                                     ((mode & FMODE_READ ) ? S_IRUSR : 0) |
2044                                     ((mode & FMODE_WRITE) ? S_IWUSR : 0));
2045         if (!inode)
2046                 return -ENOENT;
2047
2048         ei = PROC_I(inode);
2049         ei->op.proc_get_link = map_files_get_link;
2050
2051         inode->i_op = &proc_map_files_link_inode_operations;
2052         inode->i_size = 64;
2053
2054         d_set_d_op(dentry, &tid_map_files_dentry_operations);
2055         d_add(dentry, inode);
2056
2057         return 0;
2058 }
2059
2060 static struct dentry *proc_map_files_lookup(struct inode *dir,
2061                 struct dentry *dentry, unsigned int flags)
2062 {
2063         unsigned long vm_start, vm_end;
2064         struct vm_area_struct *vma;
2065         struct task_struct *task;
2066         int result;
2067         struct mm_struct *mm;
2068
2069         result = -ENOENT;
2070         task = get_proc_task(dir);
2071         if (!task)
2072                 goto out;
2073
2074         result = -EACCES;
2075         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2076                 goto out_put_task;
2077
2078         result = -ENOENT;
2079         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2080                 goto out_put_task;
2081
2082         mm = get_task_mm(task);
2083         if (!mm)
2084                 goto out_put_task;
2085
2086         down_read(&mm->mmap_sem);
2087         vma = find_exact_vma(mm, vm_start, vm_end);
2088         if (!vma)
2089                 goto out_no_vma;
2090
2091         if (vma->vm_file)
2092                 result = proc_map_files_instantiate(dir, dentry, task,
2093                                 (void *)(unsigned long)vma->vm_file->f_mode);
2094
2095 out_no_vma:
2096         up_read(&mm->mmap_sem);
2097         mmput(mm);
2098 out_put_task:
2099         put_task_struct(task);
2100 out:
2101         return ERR_PTR(result);
2102 }
2103
2104 static const struct inode_operations proc_map_files_inode_operations = {
2105         .lookup         = proc_map_files_lookup,
2106         .permission     = proc_fd_permission,
2107         .setattr        = proc_setattr,
2108 };
2109
2110 static int
2111 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2112 {
2113         struct vm_area_struct *vma;
2114         struct task_struct *task;
2115         struct mm_struct *mm;
2116         unsigned long nr_files, pos, i;
2117         struct flex_array *fa = NULL;
2118         struct map_files_info info;
2119         struct map_files_info *p;
2120         int ret;
2121
2122         ret = -ENOENT;
2123         task = get_proc_task(file_inode(file));
2124         if (!task)
2125                 goto out;
2126
2127         ret = -EACCES;
2128         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2129                 goto out_put_task;
2130
2131         ret = 0;
2132         if (!dir_emit_dots(file, ctx))
2133                 goto out_put_task;
2134
2135         mm = get_task_mm(task);
2136         if (!mm)
2137                 goto out_put_task;
2138         down_read(&mm->mmap_sem);
2139
2140         nr_files = 0;
2141
2142         /*
2143          * We need two passes here:
2144          *
2145          *  1) Collect vmas of mapped files with mmap_sem taken
2146          *  2) Release mmap_sem and instantiate entries
2147          *
2148          * otherwise we get lockdep complained, since filldir()
2149          * routine might require mmap_sem taken in might_fault().
2150          */
2151
2152         for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2153                 if (vma->vm_file && ++pos > ctx->pos)
2154                         nr_files++;
2155         }
2156
2157         if (nr_files) {
2158                 fa = flex_array_alloc(sizeof(info), nr_files,
2159                                         GFP_KERNEL);
2160                 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2161                                                 GFP_KERNEL)) {
2162                         ret = -ENOMEM;
2163                         if (fa)
2164                                 flex_array_free(fa);
2165                         up_read(&mm->mmap_sem);
2166                         mmput(mm);
2167                         goto out_put_task;
2168                 }
2169                 for (i = 0, vma = mm->mmap, pos = 2; vma;
2170                                 vma = vma->vm_next) {
2171                         if (!vma->vm_file)
2172                                 continue;
2173                         if (++pos <= ctx->pos)
2174                                 continue;
2175
2176                         info.mode = vma->vm_file->f_mode;
2177                         info.len = snprintf(info.name,
2178                                         sizeof(info.name), "%lx-%lx",
2179                                         vma->vm_start, vma->vm_end);
2180                         if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2181                                 BUG();
2182                 }
2183         }
2184         up_read(&mm->mmap_sem);
2185
2186         for (i = 0; i < nr_files; i++) {
2187                 p = flex_array_get(fa, i);
2188                 if (!proc_fill_cache(file, ctx,
2189                                       p->name, p->len,
2190                                       proc_map_files_instantiate,
2191                                       task,
2192                                       (void *)(unsigned long)p->mode))
2193                         break;
2194                 ctx->pos++;
2195         }
2196         if (fa)
2197                 flex_array_free(fa);
2198         mmput(mm);
2199
2200 out_put_task:
2201         put_task_struct(task);
2202 out:
2203         return ret;
2204 }
2205
2206 static const struct file_operations proc_map_files_operations = {
2207         .read           = generic_read_dir,
2208         .iterate_shared = proc_map_files_readdir,
2209         .llseek         = generic_file_llseek,
2210 };
2211
2212 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2213 struct timers_private {
2214         struct pid *pid;
2215         struct task_struct *task;
2216         struct sighand_struct *sighand;
2217         struct pid_namespace *ns;
2218         unsigned long flags;
2219 };
2220
2221 static void *timers_start(struct seq_file *m, loff_t *pos)
2222 {
2223         struct timers_private *tp = m->private;
2224
2225         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2226         if (!tp->task)
2227                 return ERR_PTR(-ESRCH);
2228
2229         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2230         if (!tp->sighand)
2231                 return ERR_PTR(-ESRCH);
2232
2233         return seq_list_start(&tp->task->signal->posix_timers, *pos);
2234 }
2235
2236 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2237 {
2238         struct timers_private *tp = m->private;
2239         return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2240 }
2241
2242 static void timers_stop(struct seq_file *m, void *v)
2243 {
2244         struct timers_private *tp = m->private;
2245
2246         if (tp->sighand) {
2247                 unlock_task_sighand(tp->task, &tp->flags);
2248                 tp->sighand = NULL;
2249         }
2250
2251         if (tp->task) {
2252                 put_task_struct(tp->task);
2253                 tp->task = NULL;
2254         }
2255 }
2256
2257 static int show_timer(struct seq_file *m, void *v)
2258 {
2259         struct k_itimer *timer;
2260         struct timers_private *tp = m->private;
2261         int notify;
2262         static const char * const nstr[] = {
2263                 [SIGEV_SIGNAL] = "signal",
2264                 [SIGEV_NONE] = "none",
2265                 [SIGEV_THREAD] = "thread",
2266         };
2267
2268         timer = list_entry((struct list_head *)v, struct k_itimer, list);
2269         notify = timer->it_sigev_notify;
2270
2271         seq_printf(m, "ID: %d\n", timer->it_id);
2272         seq_printf(m, "signal: %d/%p\n",
2273                    timer->sigq->info.si_signo,
2274                    timer->sigq->info.si_value.sival_ptr);
2275         seq_printf(m, "notify: %s/%s.%d\n",
2276                    nstr[notify & ~SIGEV_THREAD_ID],
2277                    (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2278                    pid_nr_ns(timer->it_pid, tp->ns));
2279         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2280
2281         return 0;
2282 }
2283
2284 static const struct seq_operations proc_timers_seq_ops = {
2285         .start  = timers_start,
2286         .next   = timers_next,
2287         .stop   = timers_stop,
2288         .show   = show_timer,
2289 };
2290
2291 static int proc_timers_open(struct inode *inode, struct file *file)
2292 {
2293         struct timers_private *tp;
2294
2295         tp = __seq_open_private(file, &proc_timers_seq_ops,
2296                         sizeof(struct timers_private));
2297         if (!tp)
2298                 return -ENOMEM;
2299
2300         tp->pid = proc_pid(inode);
2301         tp->ns = inode->i_sb->s_fs_info;
2302         return 0;
2303 }
2304
2305 static const struct file_operations proc_timers_operations = {
2306         .open           = proc_timers_open,
2307         .read           = seq_read,
2308         .llseek         = seq_lseek,
2309         .release        = seq_release_private,
2310 };
2311 #endif
2312
2313 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2314                                         size_t count, loff_t *offset)
2315 {
2316         struct inode *inode = file_inode(file);
2317         struct task_struct *p;
2318         u64 slack_ns;
2319         int err;
2320
2321         err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2322         if (err < 0)
2323                 return err;
2324
2325         p = get_proc_task(inode);
2326         if (!p)
2327                 return -ESRCH;
2328
2329         if (p != current) {
2330                 if (!capable(CAP_SYS_NICE)) {
2331                         count = -EPERM;
2332                         goto out;
2333                 }
2334
2335                 err = security_task_setscheduler(p);
2336                 if (err) {
2337                         count = err;
2338                         goto out;
2339                 }
2340         }
2341
2342         task_lock(p);
2343         if (slack_ns == 0)
2344                 p->timer_slack_ns = p->default_timer_slack_ns;
2345         else
2346                 p->timer_slack_ns = slack_ns;
2347         task_unlock(p);
2348
2349 out:
2350         put_task_struct(p);
2351
2352         return count;
2353 }
2354
2355 static int timerslack_ns_show(struct seq_file *m, void *v)
2356 {
2357         struct inode *inode = m->private;
2358         struct task_struct *p;
2359         int err = 0;
2360
2361         p = get_proc_task(inode);
2362         if (!p)
2363                 return -ESRCH;
2364
2365         if (p != current) {
2366
2367                 if (!capable(CAP_SYS_NICE)) {
2368                         err = -EPERM;
2369                         goto out;
2370                 }
2371                 err = security_task_getscheduler(p);
2372                 if (err)
2373                         goto out;
2374         }
2375
2376         task_lock(p);
2377         seq_printf(m, "%llu\n", p->timer_slack_ns);
2378         task_unlock(p);
2379
2380 out:
2381         put_task_struct(p);
2382
2383         return err;
2384 }
2385
2386 static int timerslack_ns_open(struct inode *inode, struct file *filp)
2387 {
2388         return single_open(filp, timerslack_ns_show, inode);
2389 }
2390
2391 static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2392         .open           = timerslack_ns_open,
2393         .read           = seq_read,
2394         .write          = timerslack_ns_write,
2395         .llseek         = seq_lseek,
2396         .release        = single_release,
2397 };
2398
2399 static int proc_pident_instantiate(struct inode *dir,
2400         struct dentry *dentry, struct task_struct *task, const void *ptr)
2401 {
2402         const struct pid_entry *p = ptr;
2403         struct inode *inode;
2404         struct proc_inode *ei;
2405
2406         inode = proc_pid_make_inode(dir->i_sb, task, p->mode);
2407         if (!inode)
2408                 goto out;
2409
2410         ei = PROC_I(inode);
2411         if (S_ISDIR(inode->i_mode))
2412                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2413         if (p->iop)
2414                 inode->i_op = p->iop;
2415         if (p->fop)
2416                 inode->i_fop = p->fop;
2417         ei->op = p->op;
2418         d_set_d_op(dentry, &pid_dentry_operations);
2419         d_add(dentry, inode);
2420         /* Close the race of the process dying before we return the dentry */
2421         if (pid_revalidate(dentry, 0))
2422                 return 0;
2423 out:
2424         return -ENOENT;
2425 }
2426
2427 static struct dentry *proc_pident_lookup(struct inode *dir,
2428                                          struct dentry *dentry,
2429                                          const struct pid_entry *ents,
2430                                          unsigned int nents)
2431 {
2432         int error;
2433         struct task_struct *task = get_proc_task(dir);
2434         const struct pid_entry *p, *last;
2435
2436         error = -ENOENT;
2437
2438         if (!task)
2439                 goto out_no_task;
2440
2441         /*
2442          * Yes, it does not scale. And it should not. Don't add
2443          * new entries into /proc/<tgid>/ without very good reasons.
2444          */
2445         last = &ents[nents];
2446         for (p = ents; p < last; p++) {
2447                 if (p->len != dentry->d_name.len)
2448                         continue;
2449                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2450                         break;
2451         }
2452         if (p >= last)
2453                 goto out;
2454
2455         error = proc_pident_instantiate(dir, dentry, task, p);
2456 out:
2457         put_task_struct(task);
2458 out_no_task:
2459         return ERR_PTR(error);
2460 }
2461
2462 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2463                 const struct pid_entry *ents, unsigned int nents)
2464 {
2465         struct task_struct *task = get_proc_task(file_inode(file));
2466         const struct pid_entry *p;
2467
2468         if (!task)
2469                 return -ENOENT;
2470
2471         if (!dir_emit_dots(file, ctx))
2472                 goto out;
2473
2474         if (ctx->pos >= nents + 2)
2475                 goto out;
2476
2477         for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2478                 if (!proc_fill_cache(file, ctx, p->name, p->len,
2479                                 proc_pident_instantiate, task, p))
2480                         break;
2481                 ctx->pos++;
2482         }
2483 out:
2484         put_task_struct(task);
2485         return 0;
2486 }
2487
2488 #ifdef CONFIG_SECURITY
2489 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2490                                   size_t count, loff_t *ppos)
2491 {
2492         struct inode * inode = file_inode(file);
2493         char *p = NULL;
2494         ssize_t length;
2495         struct task_struct *task = get_proc_task(inode);
2496
2497         if (!task)
2498                 return -ESRCH;
2499
2500         length = security_getprocattr(task,
2501                                       (char*)file->f_path.dentry->d_name.name,
2502                                       &p);
2503         put_task_struct(task);
2504         if (length > 0)
2505                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2506         kfree(p);
2507         return length;
2508 }
2509
2510 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2511                                    size_t count, loff_t *ppos)
2512 {
2513         struct inode * inode = file_inode(file);
2514         void *page;
2515         ssize_t length;
2516         struct task_struct *task = get_proc_task(inode);
2517
2518         length = -ESRCH;
2519         if (!task)
2520                 goto out_no_task;
2521
2522         /* A task may only write its own attributes. */
2523         length = -EACCES;
2524         if (current != task)
2525                 goto out;
2526
2527         if (count > PAGE_SIZE)
2528                 count = PAGE_SIZE;
2529
2530         /* No partial writes. */
2531         length = -EINVAL;
2532         if (*ppos != 0)
2533                 goto out;
2534
2535         page = memdup_user(buf, count);
2536         if (IS_ERR(page)) {
2537                 length = PTR_ERR(page);
2538                 goto out;
2539         }
2540
2541         /* Guard against adverse ptrace interaction */
2542         length = mutex_lock_interruptible(&current->signal->cred_guard_mutex);
2543         if (length < 0)
2544                 goto out_free;
2545
2546         length = security_setprocattr(file->f_path.dentry->d_name.name,
2547                                       page, count);
2548         mutex_unlock(&current->signal->cred_guard_mutex);
2549 out_free:
2550         kfree(page);
2551 out:
2552         put_task_struct(task);
2553 out_no_task:
2554         return length;
2555 }
2556
2557 static const struct file_operations proc_pid_attr_operations = {
2558         .read           = proc_pid_attr_read,
2559         .write          = proc_pid_attr_write,
2560         .llseek         = generic_file_llseek,
2561 };
2562
2563 static const struct pid_entry attr_dir_stuff[] = {
2564         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2565         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2566         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2567         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2568         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2569         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2570 };
2571
2572 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2573 {
2574         return proc_pident_readdir(file, ctx,
2575                                    attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2576 }
2577
2578 static const struct file_operations proc_attr_dir_operations = {
2579         .read           = generic_read_dir,
2580         .iterate_shared = proc_attr_dir_readdir,
2581         .llseek         = generic_file_llseek,
2582 };
2583
2584 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2585                                 struct dentry *dentry, unsigned int flags)
2586 {
2587         return proc_pident_lookup(dir, dentry,
2588                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2589 }
2590
2591 static const struct inode_operations proc_attr_dir_inode_operations = {
2592         .lookup         = proc_attr_dir_lookup,
2593         .getattr        = pid_getattr,
2594         .setattr        = proc_setattr,
2595 };
2596
2597 #endif
2598
2599 #ifdef CONFIG_ELF_CORE
2600 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2601                                          size_t count, loff_t *ppos)
2602 {
2603         struct task_struct *task = get_proc_task(file_inode(file));
2604         struct mm_struct *mm;
2605         char buffer[PROC_NUMBUF];
2606         size_t len;
2607         int ret;
2608
2609         if (!task)
2610                 return -ESRCH;
2611
2612         ret = 0;
2613         mm = get_task_mm(task);
2614         if (mm) {
2615                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2616                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2617                                 MMF_DUMP_FILTER_SHIFT));
2618                 mmput(mm);
2619                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2620         }
2621
2622         put_task_struct(task);
2623
2624         return ret;
2625 }
2626
2627 static ssize_t proc_coredump_filter_write(struct file *file,
2628                                           const char __user *buf,
2629                                           size_t count,
2630                                           loff_t *ppos)
2631 {
2632         struct task_struct *task;
2633         struct mm_struct *mm;
2634         unsigned int val;
2635         int ret;
2636         int i;
2637         unsigned long mask;
2638
2639         ret = kstrtouint_from_user(buf, count, 0, &val);
2640         if (ret < 0)
2641                 return ret;
2642
2643         ret = -ESRCH;
2644         task = get_proc_task(file_inode(file));
2645         if (!task)
2646                 goto out_no_task;
2647
2648         mm = get_task_mm(task);
2649         if (!mm)
2650                 goto out_no_mm;
2651         ret = 0;
2652
2653         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2654                 if (val & mask)
2655                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2656                 else
2657                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2658         }
2659
2660         mmput(mm);
2661  out_no_mm:
2662         put_task_struct(task);
2663  out_no_task:
2664         if (ret < 0)
2665                 return ret;
2666         return count;
2667 }
2668
2669 static const struct file_operations proc_coredump_filter_operations = {
2670         .read           = proc_coredump_filter_read,
2671         .write          = proc_coredump_filter_write,
2672         .llseek         = generic_file_llseek,
2673 };
2674 #endif
2675
2676 #ifdef CONFIG_TASK_IO_ACCOUNTING
2677 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2678 {
2679         struct task_io_accounting acct = task->ioac;
2680         unsigned long flags;
2681         int result;
2682
2683         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2684         if (result)
2685                 return result;
2686
2687         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2688                 result = -EACCES;
2689                 goto out_unlock;
2690         }
2691
2692         if (whole && lock_task_sighand(task, &flags)) {
2693                 struct task_struct *t = task;
2694
2695                 task_io_accounting_add(&acct, &task->signal->ioac);
2696                 while_each_thread(task, t)
2697                         task_io_accounting_add(&acct, &t->ioac);
2698
2699                 unlock_task_sighand(task, &flags);
2700         }
2701         seq_printf(m,
2702                    "rchar: %llu\n"
2703                    "wchar: %llu\n"
2704                    "syscr: %llu\n"
2705                    "syscw: %llu\n"
2706                    "read_bytes: %llu\n"
2707                    "write_bytes: %llu\n"
2708                    "cancelled_write_bytes: %llu\n",
2709                    (unsigned long long)acct.rchar,
2710                    (unsigned long long)acct.wchar,
2711                    (unsigned long long)acct.syscr,
2712                    (unsigned long long)acct.syscw,
2713                    (unsigned long long)acct.read_bytes,
2714                    (unsigned long long)acct.write_bytes,
2715                    (unsigned long long)acct.cancelled_write_bytes);
2716         result = 0;
2717
2718 out_unlock:
2719         mutex_unlock(&task->signal->cred_guard_mutex);
2720         return result;
2721 }
2722
2723 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2724                                   struct pid *pid, struct task_struct *task)
2725 {
2726         return do_io_accounting(task, m, 0);
2727 }
2728
2729 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2730                                    struct pid *pid, struct task_struct *task)
2731 {
2732         return do_io_accounting(task, m, 1);
2733 }
2734 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2735
2736 #ifdef CONFIG_USER_NS
2737 static int proc_id_map_open(struct inode *inode, struct file *file,
2738         const struct seq_operations *seq_ops)
2739 {
2740         struct user_namespace *ns = NULL;
2741         struct task_struct *task;
2742         struct seq_file *seq;
2743         int ret = -EINVAL;
2744
2745         task = get_proc_task(inode);
2746         if (task) {
2747                 rcu_read_lock();
2748                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2749                 rcu_read_unlock();
2750                 put_task_struct(task);
2751         }
2752         if (!ns)
2753                 goto err;
2754
2755         ret = seq_open(file, seq_ops);
2756         if (ret)
2757                 goto err_put_ns;
2758
2759         seq = file->private_data;
2760         seq->private = ns;
2761
2762         return 0;
2763 err_put_ns:
2764         put_user_ns(ns);
2765 err:
2766         return ret;
2767 }
2768
2769 static int proc_id_map_release(struct inode *inode, struct file *file)
2770 {
2771         struct seq_file *seq = file->private_data;
2772         struct user_namespace *ns = seq->private;
2773         put_user_ns(ns);
2774         return seq_release(inode, file);
2775 }
2776
2777 static int proc_uid_map_open(struct inode *inode, struct file *file)
2778 {
2779         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2780 }
2781
2782 static int proc_gid_map_open(struct inode *inode, struct file *file)
2783 {
2784         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2785 }
2786
2787 static int proc_projid_map_open(struct inode *inode, struct file *file)
2788 {
2789         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2790 }
2791
2792 static const struct file_operations proc_uid_map_operations = {
2793         .open           = proc_uid_map_open,
2794         .write          = proc_uid_map_write,
2795         .read           = seq_read,
2796         .llseek         = seq_lseek,
2797         .release        = proc_id_map_release,
2798 };
2799
2800 static const struct file_operations proc_gid_map_operations = {
2801         .open           = proc_gid_map_open,
2802         .write          = proc_gid_map_write,
2803         .read           = seq_read,
2804         .llseek         = seq_lseek,
2805         .release        = proc_id_map_release,
2806 };
2807
2808 static const struct file_operations proc_projid_map_operations = {
2809         .open           = proc_projid_map_open,
2810         .write          = proc_projid_map_write,
2811         .read           = seq_read,
2812         .llseek         = seq_lseek,
2813         .release        = proc_id_map_release,
2814 };
2815
2816 static int proc_setgroups_open(struct inode *inode, struct file *file)
2817 {
2818         struct user_namespace *ns = NULL;
2819         struct task_struct *task;
2820         int ret;
2821
2822         ret = -ESRCH;
2823         task = get_proc_task(inode);
2824         if (task) {
2825                 rcu_read_lock();
2826                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2827                 rcu_read_unlock();
2828                 put_task_struct(task);
2829         }
2830         if (!ns)
2831                 goto err;
2832
2833         if (file->f_mode & FMODE_WRITE) {
2834                 ret = -EACCES;
2835                 if (!ns_capable(ns, CAP_SYS_ADMIN))
2836                         goto err_put_ns;
2837         }
2838
2839         ret = single_open(file, &proc_setgroups_show, ns);
2840         if (ret)
2841                 goto err_put_ns;
2842
2843         return 0;
2844 err_put_ns:
2845         put_user_ns(ns);
2846 err:
2847         return ret;
2848 }
2849
2850 static int proc_setgroups_release(struct inode *inode, struct file *file)
2851 {
2852         struct seq_file *seq = file->private_data;
2853         struct user_namespace *ns = seq->private;
2854         int ret = single_release(inode, file);
2855         put_user_ns(ns);
2856         return ret;
2857 }
2858
2859 static const struct file_operations proc_setgroups_operations = {
2860         .open           = proc_setgroups_open,
2861         .write          = proc_setgroups_write,
2862         .read           = seq_read,
2863         .llseek         = seq_lseek,
2864         .release        = proc_setgroups_release,
2865 };
2866 #endif /* CONFIG_USER_NS */
2867
2868 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2869                                 struct pid *pid, struct task_struct *task)
2870 {
2871         int err = lock_trace(task);
2872         if (!err) {
2873                 seq_printf(m, "%08x\n", task->personality);
2874                 unlock_trace(task);
2875         }
2876         return err;
2877 }
2878
2879 #ifdef CONFIG_LIVEPATCH
2880 static int proc_pid_patch_state(struct seq_file *m, struct pid_namespace *ns,
2881                                 struct pid *pid, struct task_struct *task)
2882 {
2883         seq_printf(m, "%d\n", task->patch_state);
2884         return 0;
2885 }
2886 #endif /* CONFIG_LIVEPATCH */
2887
2888 /*
2889  * Thread groups
2890  */
2891 static const struct file_operations proc_task_operations;
2892 static const struct inode_operations proc_task_inode_operations;
2893
2894 static const struct pid_entry tgid_base_stuff[] = {
2895         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2896         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2897         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2898         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2899         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2900 #ifdef CONFIG_NET
2901         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2902 #endif
2903         REG("environ",    S_IRUSR, proc_environ_operations),
2904         REG("auxv",       S_IRUSR, proc_auxv_operations),
2905         ONE("status",     S_IRUGO, proc_pid_status),
2906         ONE("personality", S_IRUSR, proc_pid_personality),
2907         ONE("limits",     S_IRUGO, proc_pid_limits),
2908 #ifdef CONFIG_SCHED_DEBUG
2909         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2910 #endif
2911 #ifdef CONFIG_SCHED_AUTOGROUP
2912         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2913 #endif
2914         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2915 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2916         ONE("syscall",    S_IRUSR, proc_pid_syscall),
2917 #endif
2918         REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
2919         ONE("stat",       S_IRUGO, proc_tgid_stat),
2920         ONE("statm",      S_IRUGO, proc_pid_statm),
2921         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2922 #ifdef CONFIG_NUMA
2923         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2924 #endif
2925         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2926         LNK("cwd",        proc_cwd_link),
2927         LNK("root",       proc_root_link),
2928         LNK("exe",        proc_exe_link),
2929         REG("mounts",     S_IRUGO, proc_mounts_operations),
2930         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2931         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2932 #ifdef CONFIG_PROC_PAGE_MONITOR
2933         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2934         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2935         REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
2936         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2937 #endif
2938 #ifdef CONFIG_SECURITY
2939         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2940 #endif
2941 #ifdef CONFIG_KALLSYMS
2942         ONE("wchan",      S_IRUGO, proc_pid_wchan),
2943 #endif
2944 #ifdef CONFIG_STACKTRACE
2945         ONE("stack",      S_IRUSR, proc_pid_stack),
2946 #endif
2947 #ifdef CONFIG_SCHED_INFO
2948         ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
2949 #endif
2950 #ifdef CONFIG_LATENCYTOP
2951         REG("latency",  S_IRUGO, proc_lstats_operations),
2952 #endif
2953 #ifdef CONFIG_PROC_PID_CPUSET
2954         ONE("cpuset",     S_IRUGO, proc_cpuset_show),
2955 #endif
2956 #ifdef CONFIG_CGROUPS
2957         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
2958 #endif
2959         ONE("oom_score",  S_IRUGO, proc_oom_score),
2960         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2961         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2962 #ifdef CONFIG_AUDITSYSCALL
2963         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2964         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2965 #endif
2966 #ifdef CONFIG_FAULT_INJECTION
2967         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2968         REG("fail-nth", 0644, proc_fail_nth_operations),
2969 #endif
2970 #ifdef CONFIG_ELF_CORE
2971         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2972 #endif
2973 #ifdef CONFIG_TASK_IO_ACCOUNTING
2974         ONE("io",       S_IRUSR, proc_tgid_io_accounting),
2975 #endif
2976 #ifdef CONFIG_HARDWALL
2977         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
2978 #endif
2979 #ifdef CONFIG_USER_NS
2980         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2981         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2982         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2983         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
2984 #endif
2985 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2986         REG("timers",     S_IRUGO, proc_timers_operations),
2987 #endif
2988         REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
2989 #ifdef CONFIG_LIVEPATCH
2990         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
2991 #endif
2992 };
2993
2994 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2995 {
2996         return proc_pident_readdir(file, ctx,
2997                                    tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2998 }
2999
3000 static const struct file_operations proc_tgid_base_operations = {
3001         .read           = generic_read_dir,
3002         .iterate_shared = proc_tgid_base_readdir,
3003         .llseek         = generic_file_llseek,
3004 };
3005
3006 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3007 {
3008         return proc_pident_lookup(dir, dentry,
3009                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3010 }
3011
3012 static const struct inode_operations proc_tgid_base_inode_operations = {
3013         .lookup         = proc_tgid_base_lookup,
3014         .getattr        = pid_getattr,
3015         .setattr        = proc_setattr,
3016         .permission     = proc_pid_permission,
3017 };
3018
3019 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
3020 {
3021         struct dentry *dentry, *leader, *dir;
3022         char buf[PROC_NUMBUF];
3023         struct qstr name;
3024
3025         name.name = buf;
3026         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3027         /* no ->d_hash() rejects on procfs */
3028         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
3029         if (dentry) {
3030                 d_invalidate(dentry);
3031                 dput(dentry);
3032         }
3033
3034         if (pid == tgid)
3035                 return;
3036
3037         name.name = buf;
3038         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
3039         leader = d_hash_and_lookup(mnt->mnt_root, &name);
3040         if (!leader)
3041                 goto out;
3042
3043         name.name = "task";
3044         name.len = strlen(name.name);
3045         dir = d_hash_and_lookup(leader, &name);
3046         if (!dir)
3047                 goto out_put_leader;
3048
3049         name.name = buf;
3050         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3051         dentry = d_hash_and_lookup(dir, &name);
3052         if (dentry) {
3053                 d_invalidate(dentry);
3054                 dput(dentry);
3055         }
3056
3057         dput(dir);
3058 out_put_leader:
3059         dput(leader);
3060 out:
3061         return;
3062 }
3063
3064 /**
3065  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
3066  * @task: task that should be flushed.
3067  *
3068  * When flushing dentries from proc, one needs to flush them from global
3069  * proc (proc_mnt) and from all the namespaces' procs this task was seen
3070  * in. This call is supposed to do all of this job.
3071  *
3072  * Looks in the dcache for
3073  * /proc/@pid
3074  * /proc/@tgid/task/@pid
3075  * if either directory is present flushes it and all of it'ts children
3076  * from the dcache.
3077  *
3078  * It is safe and reasonable to cache /proc entries for a task until
3079  * that task exits.  After that they just clog up the dcache with
3080  * useless entries, possibly causing useful dcache entries to be
3081  * flushed instead.  This routine is proved to flush those useless
3082  * dcache entries at process exit time.
3083  *
3084  * NOTE: This routine is just an optimization so it does not guarantee
3085  *       that no dcache entries will exist at process exit time it
3086  *       just makes it very unlikely that any will persist.
3087  */
3088
3089 void proc_flush_task(struct task_struct *task)
3090 {
3091         int i;
3092         struct pid *pid, *tgid;
3093         struct upid *upid;
3094
3095         pid = task_pid(task);
3096         tgid = task_tgid(task);
3097
3098         for (i = 0; i <= pid->level; i++) {
3099                 upid = &pid->numbers[i];
3100                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3101                                         tgid->numbers[i].nr);
3102         }
3103 }
3104
3105 static int proc_pid_instantiate(struct inode *dir,
3106                                    struct dentry * dentry,
3107                                    struct task_struct *task, const void *ptr)
3108 {
3109         struct inode *inode;
3110
3111         inode = proc_pid_make_inode(dir->i_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3112         if (!inode)
3113                 goto out;
3114
3115         inode->i_op = &proc_tgid_base_inode_operations;
3116         inode->i_fop = &proc_tgid_base_operations;
3117         inode->i_flags|=S_IMMUTABLE;
3118
3119         set_nlink(inode, nlink_tgid);
3120
3121         d_set_d_op(dentry, &pid_dentry_operations);
3122
3123         d_add(dentry, inode);
3124         /* Close the race of the process dying before we return the dentry */
3125         if (pid_revalidate(dentry, 0))
3126                 return 0;
3127 out:
3128         return -ENOENT;
3129 }
3130
3131 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3132 {
3133         int result = -ENOENT;
3134         struct task_struct *task;
3135         unsigned tgid;
3136         struct pid_namespace *ns;
3137
3138         tgid = name_to_int(&dentry->d_name);
3139         if (tgid == ~0U)
3140                 goto out;
3141
3142         ns = dentry->d_sb->s_fs_info;
3143         rcu_read_lock();
3144         task = find_task_by_pid_ns(tgid, ns);
3145         if (task)
3146                 get_task_struct(task);
3147         rcu_read_unlock();
3148         if (!task)
3149                 goto out;
3150
3151         result = proc_pid_instantiate(dir, dentry, task, NULL);
3152         put_task_struct(task);
3153 out:
3154         return ERR_PTR(result);
3155 }
3156
3157 /*
3158  * Find the first task with tgid >= tgid
3159  *
3160  */
3161 struct tgid_iter {
3162         unsigned int tgid;
3163         struct task_struct *task;
3164 };
3165 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3166 {
3167         struct pid *pid;
3168
3169         if (iter.task)
3170                 put_task_struct(iter.task);
3171         rcu_read_lock();
3172 retry:
3173         iter.task = NULL;
3174         pid = find_ge_pid(iter.tgid, ns);
3175         if (pid) {
3176                 iter.tgid = pid_nr_ns(pid, ns);
3177                 iter.task = pid_task(pid, PIDTYPE_PID);
3178                 /* What we to know is if the pid we have find is the
3179                  * pid of a thread_group_leader.  Testing for task
3180                  * being a thread_group_leader is the obvious thing
3181                  * todo but there is a window when it fails, due to
3182                  * the pid transfer logic in de_thread.
3183                  *
3184                  * So we perform the straight forward test of seeing
3185                  * if the pid we have found is the pid of a thread
3186                  * group leader, and don't worry if the task we have
3187                  * found doesn't happen to be a thread group leader.
3188                  * As we don't care in the case of readdir.
3189                  */
3190                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3191                         iter.tgid += 1;
3192                         goto retry;
3193                 }
3194                 get_task_struct(iter.task);
3195         }
3196         rcu_read_unlock();
3197         return iter;
3198 }
3199
3200 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3201
3202 /* for the /proc/ directory itself, after non-process stuff has been done */
3203 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3204 {
3205         struct tgid_iter iter;
3206         struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
3207         loff_t pos = ctx->pos;
3208
3209         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3210                 return 0;
3211
3212         if (pos == TGID_OFFSET - 2) {
3213                 struct inode *inode = d_inode(ns->proc_self);
3214                 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3215                         return 0;
3216                 ctx->pos = pos = pos + 1;
3217         }
3218         if (pos == TGID_OFFSET - 1) {
3219                 struct inode *inode = d_inode(ns->proc_thread_self);
3220                 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3221                         return 0;
3222                 ctx->pos = pos = pos + 1;
3223         }
3224         iter.tgid = pos - TGID_OFFSET;
3225         iter.task = NULL;
3226         for (iter = next_tgid(ns, iter);
3227              iter.task;
3228              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3229                 char name[PROC_NUMBUF];
3230                 int len;
3231
3232                 cond_resched();
3233                 if (!has_pid_permissions(ns, iter.task, HIDEPID_INVISIBLE))
3234                         continue;
3235
3236                 len = snprintf(name, sizeof(name), "%d", iter.tgid);
3237                 ctx->pos = iter.tgid + TGID_OFFSET;
3238                 if (!proc_fill_cache(file, ctx, name, len,
3239                                      proc_pid_instantiate, iter.task, NULL)) {
3240                         put_task_struct(iter.task);
3241                         return 0;
3242                 }
3243         }
3244         ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3245         return 0;
3246 }
3247
3248 /*
3249  * proc_tid_comm_permission is a special permission function exclusively
3250  * used for the node /proc/<pid>/task/<tid>/comm.
3251  * It bypasses generic permission checks in the case where a task of the same
3252  * task group attempts to access the node.
3253  * The rationale behind this is that glibc and bionic access this node for
3254  * cross thread naming (pthread_set/getname_np(!self)). However, if
3255  * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3256  * which locks out the cross thread naming implementation.
3257  * This function makes sure that the node is always accessible for members of
3258  * same thread group.
3259  */
3260 static int proc_tid_comm_permission(struct inode *inode, int mask)
3261 {
3262         bool is_same_tgroup;
3263         struct task_struct *task;
3264
3265         task = get_proc_task(inode);
3266         if (!task)
3267                 return -ESRCH;
3268         is_same_tgroup = same_thread_group(current, task);
3269         put_task_struct(task);
3270
3271         if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3272                 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3273                  * read or written by the members of the corresponding
3274                  * thread group.
3275                  */
3276                 return 0;
3277         }
3278
3279         return generic_permission(inode, mask);
3280 }
3281
3282 static const struct inode_operations proc_tid_comm_inode_operations = {
3283                 .permission = proc_tid_comm_permission,
3284 };
3285
3286 /*
3287  * Tasks
3288  */
3289 static const struct pid_entry tid_base_stuff[] = {
3290         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3291         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3292         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3293 #ifdef CONFIG_NET
3294         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3295 #endif
3296         REG("environ",   S_IRUSR, proc_environ_operations),
3297         REG("auxv",      S_IRUSR, proc_auxv_operations),
3298         ONE("status",    S_IRUGO, proc_pid_status),
3299         ONE("personality", S_IRUSR, proc_pid_personality),
3300         ONE("limits",    S_IRUGO, proc_pid_limits),
3301 #ifdef CONFIG_SCHED_DEBUG
3302         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3303 #endif
3304         NOD("comm",      S_IFREG|S_IRUGO|S_IWUSR,
3305                          &proc_tid_comm_inode_operations,
3306                          &proc_pid_set_comm_operations, {}),
3307 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3308         ONE("syscall",   S_IRUSR, proc_pid_syscall),
3309 #endif
3310         REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3311         ONE("stat",      S_IRUGO, proc_tid_stat),
3312         ONE("statm",     S_IRUGO, proc_pid_statm),
3313         REG("maps",      S_IRUGO, proc_tid_maps_operations),
3314 #ifdef CONFIG_PROC_CHILDREN
3315         REG("children",  S_IRUGO, proc_tid_children_operations),
3316 #endif
3317 #ifdef CONFIG_NUMA
3318         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3319 #endif
3320         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3321         LNK("cwd",       proc_cwd_link),
3322         LNK("root",      proc_root_link),
3323         LNK("exe",       proc_exe_link),
3324         REG("mounts",    S_IRUGO, proc_mounts_operations),
3325         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3326 #ifdef CONFIG_PROC_PAGE_MONITOR
3327         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3328         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3329         REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3330         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3331 #endif
3332 #ifdef CONFIG_SECURITY
3333         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3334 #endif
3335 #ifdef CONFIG_KALLSYMS
3336         ONE("wchan",     S_IRUGO, proc_pid_wchan),
3337 #endif
3338 #ifdef CONFIG_STACKTRACE
3339         ONE("stack",      S_IRUSR, proc_pid_stack),
3340 #endif
3341 #ifdef CONFIG_SCHED_INFO
3342         ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3343 #endif
3344 #ifdef CONFIG_LATENCYTOP
3345         REG("latency",  S_IRUGO, proc_lstats_operations),
3346 #endif
3347 #ifdef CONFIG_PROC_PID_CPUSET
3348         ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3349 #endif
3350 #ifdef CONFIG_CGROUPS
3351         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3352 #endif
3353         ONE("oom_score", S_IRUGO, proc_oom_score),
3354         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3355         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3356 #ifdef CONFIG_AUDITSYSCALL
3357         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3358         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3359 #endif
3360 #ifdef CONFIG_FAULT_INJECTION
3361         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3362         REG("fail-nth", 0644, proc_fail_nth_operations),
3363 #endif
3364 #ifdef CONFIG_TASK_IO_ACCOUNTING
3365         ONE("io",       S_IRUSR, proc_tid_io_accounting),
3366 #endif
3367 #ifdef CONFIG_HARDWALL
3368         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
3369 #endif
3370 #ifdef CONFIG_USER_NS
3371         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3372         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3373         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3374         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3375 #endif
3376 #ifdef CONFIG_LIVEPATCH
3377         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
3378 #endif
3379 };
3380
3381 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3382 {
3383         return proc_pident_readdir(file, ctx,
3384                                    tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3385 }
3386
3387 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3388 {
3389         return proc_pident_lookup(dir, dentry,
3390                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3391 }
3392
3393 static const struct file_operations proc_tid_base_operations = {
3394         .read           = generic_read_dir,
3395         .iterate_shared = proc_tid_base_readdir,
3396         .llseek         = generic_file_llseek,
3397 };
3398
3399 static const struct inode_operations proc_tid_base_inode_operations = {
3400         .lookup         = proc_tid_base_lookup,
3401         .getattr        = pid_getattr,
3402         .setattr        = proc_setattr,
3403 };
3404
3405 static int proc_task_instantiate(struct inode *dir,
3406         struct dentry *dentry, struct task_struct *task, const void *ptr)
3407 {
3408         struct inode *inode;
3409         inode = proc_pid_make_inode(dir->i_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3410
3411         if (!inode)
3412                 goto out;
3413         inode->i_op = &proc_tid_base_inode_operations;
3414         inode->i_fop = &proc_tid_base_operations;
3415         inode->i_flags|=S_IMMUTABLE;
3416
3417         set_nlink(inode, nlink_tid);
3418
3419         d_set_d_op(dentry, &pid_dentry_operations);
3420
3421         d_add(dentry, inode);
3422         /* Close the race of the process dying before we return the dentry */
3423         if (pid_revalidate(dentry, 0))
3424                 return 0;
3425 out:
3426         return -ENOENT;
3427 }
3428
3429 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3430 {
3431         int result = -ENOENT;
3432         struct task_struct *task;
3433         struct task_struct *leader = get_proc_task(dir);
3434         unsigned tid;
3435         struct pid_namespace *ns;
3436
3437         if (!leader)
3438                 goto out_no_task;
3439
3440         tid = name_to_int(&dentry->d_name);
3441         if (tid == ~0U)
3442                 goto out;
3443
3444         ns = dentry->d_sb->s_fs_info;
3445         rcu_read_lock();
3446         task = find_task_by_pid_ns(tid, ns);
3447         if (task)
3448                 get_task_struct(task);
3449         rcu_read_unlock();
3450         if (!task)
3451                 goto out;
3452         if (!same_thread_group(leader, task))
3453                 goto out_drop_task;
3454
3455         result = proc_task_instantiate(dir, dentry, task, NULL);
3456 out_drop_task:
3457         put_task_struct(task);
3458 out:
3459         put_task_struct(leader);
3460 out_no_task:
3461         return ERR_PTR(result);
3462 }
3463
3464 /*
3465  * Find the first tid of a thread group to return to user space.
3466  *
3467  * Usually this is just the thread group leader, but if the users
3468  * buffer was too small or there was a seek into the middle of the
3469  * directory we have more work todo.
3470  *
3471  * In the case of a short read we start with find_task_by_pid.
3472  *
3473  * In the case of a seek we start with the leader and walk nr
3474  * threads past it.
3475  */
3476 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3477                                         struct pid_namespace *ns)
3478 {
3479         struct task_struct *pos, *task;
3480         unsigned long nr = f_pos;
3481
3482         if (nr != f_pos)        /* 32bit overflow? */
3483                 return NULL;
3484
3485         rcu_read_lock();
3486         task = pid_task(pid, PIDTYPE_PID);
3487         if (!task)
3488                 goto fail;
3489
3490         /* Attempt to start with the tid of a thread */
3491         if (tid && nr) {
3492                 pos = find_task_by_pid_ns(tid, ns);
3493                 if (pos && same_thread_group(pos, task))
3494                         goto found;
3495         }
3496
3497         /* If nr exceeds the number of threads there is nothing todo */
3498         if (nr >= get_nr_threads(task))
3499                 goto fail;
3500
3501         /* If we haven't found our starting place yet start
3502          * with the leader and walk nr threads forward.
3503          */
3504         pos = task = task->group_leader;
3505         do {
3506                 if (!nr--)
3507                         goto found;
3508         } while_each_thread(task, pos);
3509 fail:
3510         pos = NULL;
3511         goto out;
3512 found:
3513         get_task_struct(pos);
3514 out:
3515         rcu_read_unlock();
3516         return pos;
3517 }
3518
3519 /*
3520  * Find the next thread in the thread list.
3521  * Return NULL if there is an error or no next thread.
3522  *
3523  * The reference to the input task_struct is released.
3524  */
3525 static struct task_struct *next_tid(struct task_struct *start)
3526 {
3527         struct task_struct *pos = NULL;
3528         rcu_read_lock();
3529         if (pid_alive(start)) {
3530                 pos = next_thread(start);
3531                 if (thread_group_leader(pos))
3532                         pos = NULL;
3533                 else
3534                         get_task_struct(pos);
3535         }
3536         rcu_read_unlock();
3537         put_task_struct(start);
3538         return pos;
3539 }
3540
3541 /* for the /proc/TGID/task/ directories */
3542 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3543 {
3544         struct inode *inode = file_inode(file);
3545         struct task_struct *task;
3546         struct pid_namespace *ns;
3547         int tid;
3548
3549         if (proc_inode_is_dead(inode))
3550                 return -ENOENT;
3551
3552         if (!dir_emit_dots(file, ctx))
3553                 return 0;
3554
3555         /* f_version caches the tgid value that the last readdir call couldn't
3556          * return. lseek aka telldir automagically resets f_version to 0.
3557          */
3558         ns = inode->i_sb->s_fs_info;
3559         tid = (int)file->f_version;
3560         file->f_version = 0;
3561         for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3562              task;
3563              task = next_tid(task), ctx->pos++) {
3564                 char name[PROC_NUMBUF];
3565                 int len;
3566                 tid = task_pid_nr_ns(task, ns);
3567                 len = snprintf(name, sizeof(name), "%d", tid);
3568                 if (!proc_fill_cache(file, ctx, name, len,
3569                                 proc_task_instantiate, task, NULL)) {
3570                         /* returning this tgid failed, save it as the first
3571                          * pid for the next readir call */
3572                         file->f_version = (u64)tid;
3573                         put_task_struct(task);
3574                         break;
3575                 }
3576         }
3577
3578         return 0;
3579 }
3580
3581 static int proc_task_getattr(const struct path *path, struct kstat *stat,
3582                              u32 request_mask, unsigned int query_flags)
3583 {
3584         struct inode *inode = d_inode(path->dentry);
3585         struct task_struct *p = get_proc_task(inode);
3586         generic_fillattr(inode, stat);
3587
3588         if (p) {
3589                 stat->nlink += get_nr_threads(p);
3590                 put_task_struct(p);
3591         }
3592
3593         return 0;
3594 }
3595
3596 static const struct inode_operations proc_task_inode_operations = {
3597         .lookup         = proc_task_lookup,
3598         .getattr        = proc_task_getattr,
3599         .setattr        = proc_setattr,
3600         .permission     = proc_pid_permission,
3601 };
3602
3603 static const struct file_operations proc_task_operations = {
3604         .read           = generic_read_dir,
3605         .iterate_shared = proc_task_readdir,
3606         .llseek         = generic_file_llseek,
3607 };
3608
3609 void __init set_proc_pid_nlink(void)
3610 {
3611         nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3612         nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3613 }