fs/proc/base.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/proc/base.c
   4  *
   5  *  Copyright (C) 1991, 1992 Linus Torvalds
   6  *
   7  *  proc base directory handling functions
   8  *
   9  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
  10  *  Instead of using magical inumbers to determine the kind of object
  11  *  we allocate and fill in-core inodes upon lookup. They don't even
  12  *  go into icache. We cache the reference to task_struct upon lookup too.
  13  *  Eventually it should become a filesystem in its own. We don't use the
  14  *  rest of procfs anymore.
  15  *
  16  *
  17  *  Changelog:
  18  *  17-Jan-2005
  19  *  Allan Bezerra
  20  *  Bruna Moreira <bruna.moreira@indt.org.br>
  21  *  Edjard Mota <edjard.mota@indt.org.br>
  22  *  Ilias Biris <ilias.biris@indt.org.br>
  23  *  Mauricio Lin <mauricio.lin@indt.org.br>
  24  *
  25  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  26  *
  27  *  A new process specific entry (smaps) included in /proc. It shows the
  28  *  size of rss for each memory area. The maps entry lacks information
  29  *  about physical memory size (rss) for each mapped file, i.e.,
  30  *  rss information for executables and library files.
  31  *  This additional information is useful for any tools that need to know
  32  *  about physical memory consumption for a process specific library.
  33  *
  34  *  Changelog:
  35  *  21-Feb-2005
  36  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  37  *  Pud inclusion in the page table walking.
  38  *
  39  *  ChangeLog:
  40  *  10-Mar-2005
  41  *  10LE Instituto Nokia de Tecnologia - INdT:
  42  *  A better way to walks through the page table as suggested by Hugh Dickins.
  43  *
  44  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
  45  *  Smaps information related to shared, private, clean and dirty pages.
  46  *
  47  *  Paul Mundt <paul.mundt@nokia.com>:
  48  *  Overall revision about smaps.
  49  */
  50
  51 #include <linux/uaccess.h>
  52
  53 #include <linux/errno.h>
  54 #include <linux/time.h>
  55 #include <linux/proc_fs.h>
  56 #include <linux/stat.h>
  57 #include <linux/task_io_accounting_ops.h>
  58 #include <linux/init.h>
  59 #include <linux/capability.h>
  60 #include <linux/file.h>
  61 #include <linux/fdtable.h>
  62 #include <linux/string.h>
  63 #include <linux/seq_file.h>
  64 #include <linux/namei.h>
  65 #include <linux/mnt_namespace.h>
  66 #include <linux/mm.h>
  67 #include <linux/swap.h>
  68 #include <linux/rcupdate.h>
  69 #include <linux/kallsyms.h>
  70 #include <linux/stacktrace.h>
  71 #include <linux/resource.h>
  72 #include <linux/module.h>
  73 #include <linux/mount.h>
  74 #include <linux/security.h>
  75 #include <linux/ptrace.h>
  76 #include <linux/tracehook.h>
  77 #include <linux/printk.h>
  78 #include <linux/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, "[<0>] %pB\n", (void *)entries[i]);
 447                 }
 448                 unlock_trace(task);
 449         }
 450         kfree(entries);
 451
 452         return err;
 453 }
 454 #endif
 455
 456 #ifdef CONFIG_SCHED_INFO
 457 /*
 458  * Provides /proc/PID/schedstat
 459  */
 460 static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
 461                               struct pid *pid, struct task_struct *task)
 462 {
 463         if (unlikely(!sched_info_on()))
 464                 seq_printf(m, "0 0 0\n");
 465         else
 466                 seq_printf(m, "%llu %llu %lu\n",
 467                    (unsigned long long)task->se.sum_exec_runtime,
 468                    (unsigned long long)task->sched_info.run_delay,
 469                    task->sched_info.pcount);
 470
 471         return 0;
 472 }
 473 #endif
 474
 475 #ifdef CONFIG_LATENCYTOP
 476 static int lstats_show_proc(struct seq_file *m, void *v)
 477 {
 478         int i;
 479         struct inode *inode = m->private;
 480         struct task_struct *task = get_proc_task(inode);
 481
 482         if (!task)
 483                 return -ESRCH;
 484         seq_puts(m, "Latency Top version : v0.1\n");
 485         for (i = 0; i < 32; i++) {
 486                 struct latency_record *lr = &task->latency_record[i];
 487                 if (lr->backtrace[0]) {
 488                         int q;
 489                         seq_printf(m, "%i %li %li",
 490                                    lr->count, lr->time, lr->max);
 491                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 492                                 unsigned long bt = lr->backtrace[q];
 493                                 if (!bt)
 494                                         break;
 495                                 if (bt == ULONG_MAX)
 496                                         break;
 497                                 seq_printf(m, " %ps", (void *)bt);
 498                         }
 499                         seq_putc(m, '\n');
 500                 }
 501
 502         }
 503         put_task_struct(task);
 504         return 0;
 505 }
 506
 507 static int lstats_open(struct inode *inode, struct file *file)
 508 {
 509         return single_open(file, lstats_show_proc, inode);
 510 }
 511
 512 static ssize_t lstats_write(struct file *file, const char __user *buf,
 513                             size_t count, loff_t *offs)
 514 {
 515         struct task_struct *task = get_proc_task(file_inode(file));
 516
 517         if (!task)
 518                 return -ESRCH;
 519         clear_all_latency_tracing(task);
 520         put_task_struct(task);
 521
 522         return count;
 523 }
 524
 525 static const struct file_operations proc_lstats_operations = {
 526         .open           = lstats_open,
 527         .read           = seq_read,
 528         .write          = lstats_write,
 529         .llseek         = seq_lseek,
 530         .release        = single_release,
 531 };
 532
 533 #endif
 534
 535 static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
 536                           struct pid *pid, struct task_struct *task)
 537 {
 538         unsigned long totalpages = totalram_pages + total_swap_pages;
 539         unsigned long points = 0;
 540
 541         points = oom_badness(task, NULL, NULL, totalpages) *
 542                                         1000 / totalpages;
 543         seq_printf(m, "%lu\n", points);
 544
 545         return 0;
 546 }
 547
 548 struct limit_names {
 549         const char *name;
 550         const char *unit;
 551 };
 552
 553 static const struct limit_names lnames[RLIM_NLIMITS] = {
 554         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 555         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 556         [RLIMIT_DATA] = {"Max data size", "bytes"},
 557         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 558         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 559         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 560         [RLIMIT_NPROC] = {"Max processes", "processes"},
 561         [RLIMIT_NOFILE] = {"Max open files", "files"},
 562         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 563         [RLIMIT_AS] = {"Max address space", "bytes"},
 564         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 565         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 566         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 567         [RLIMIT_NICE] = {"Max nice priority", NULL},
 568         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 569         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 570 };
 571
 572 /* Display limits for a process */
 573 static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
 574                            struct pid *pid, struct task_struct *task)
 575 {
 576         unsigned int i;
 577         unsigned long flags;
 578
 579         struct rlimit rlim[RLIM_NLIMITS];
 580
 581         if (!lock_task_sighand(task, &flags))
 582                 return 0;
 583         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 584         unlock_task_sighand(task, &flags);
 585
 586         /*
 587          * print the file header
 588          */
 589        seq_printf(m, "%-25s %-20s %-20s %-10s\n",
 590                   "Limit", "Soft Limit", "Hard Limit", "Units");
 591
 592         for (i = 0; i < RLIM_NLIMITS; i++) {
 593                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 594                         seq_printf(m, "%-25s %-20s ",
 595                                    lnames[i].name, "unlimited");
 596                 else
 597                         seq_printf(m, "%-25s %-20lu ",
 598                                    lnames[i].name, rlim[i].rlim_cur);
 599
 600                 if (rlim[i].rlim_max == RLIM_INFINITY)
 601                         seq_printf(m, "%-20s ", "unlimited");
 602                 else
 603                         seq_printf(m, "%-20lu ", rlim[i].rlim_max);
 604
 605                 if (lnames[i].unit)
 606                         seq_printf(m, "%-10s\n", lnames[i].unit);
 607                 else
 608                         seq_putc(m, '\n');
 609         }
 610
 611         return 0;
 612 }
 613
 614 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 615 static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
 616                             struct pid *pid, struct task_struct *task)
 617 {
 618         long nr;
 619         unsigned long args[6], sp, pc;
 620         int res;
 621
 622         res = lock_trace(task);
 623         if (res)
 624                 return res;
 625
 626         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 627                 seq_puts(m, "running\n");
 628         else if (nr < 0)
 629                 seq_printf(m, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 630         else
 631                 seq_printf(m,
 632                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 633                        nr,
 634                        args[0], args[1], args[2], args[3], args[4], args[5],
 635                        sp, pc);
 636         unlock_trace(task);
 637
 638         return 0;
 639 }
 640 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 641
 642 /************************************************************************/
 643 /*                       Here the fs part begins                        */
 644 /************************************************************************/
 645
 646 /* permission checks */
 647 static int proc_fd_access_allowed(struct inode *inode)
 648 {
 649         struct task_struct *task;
 650         int allowed = 0;
 651         /* Allow access to a task's file descriptors if it is us or we
 652          * may use ptrace attach to the process and find out that
 653          * information.
 654          */
 655         task = get_proc_task(inode);
 656         if (task) {
 657                 allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 658                 put_task_struct(task);
 659         }
 660         return allowed;
 661 }
 662
 663 int proc_setattr(struct dentry *dentry, struct iattr *attr)
 664 {
 665         int error;
 666         struct inode *inode = d_inode(dentry);
 667
 668         if (attr->ia_valid & ATTR_MODE)
 669                 return -EPERM;
 670
 671         error = setattr_prepare(dentry, attr);
 672         if (error)
 673                 return error;
 674
 675         setattr_copy(inode, attr);
 676         mark_inode_dirty(inode);
 677         return 0;
 678 }
 679
 680 /*
 681  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
 682  * or euid/egid (for hide_pid_min=2)?
 683  */
 684 static bool has_pid_permissions(struct pid_namespace *pid,
 685                                  struct task_struct *task,
 686                                  int hide_pid_min)
 687 {
 688         if (pid->hide_pid < hide_pid_min)
 689                 return true;
 690         if (in_group_p(pid->pid_gid))
 691                 return true;
 692         return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 693 }
 694
 695
 696 static int proc_pid_permission(struct inode *inode, int mask)
 697 {
 698         struct pid_namespace *pid = inode->i_sb->s_fs_info;
 699         struct task_struct *task;
 700         bool has_perms;
 701
 702         task = get_proc_task(inode);
 703         if (!task)
 704                 return -ESRCH;
 705         has_perms = has_pid_permissions(pid, task, HIDEPID_NO_ACCESS);
 706         put_task_struct(task);
 707
 708         if (!has_perms) {
 709                 if (pid->hide_pid == HIDEPID_INVISIBLE) {
 710                         /*
 711                          * Let's make getdents(), stat(), and open()
 712                          * consistent with each other.  If a process
 713                          * may not stat() a file, it shouldn't be seen
 714                          * in procfs at all.
 715                          */
 716                         return -ENOENT;
 717                 }
 718
 719                 return -EPERM;
 720         }
 721         return generic_permission(inode, mask);
 722 }
 723
 724
 725
 726 static const struct inode_operations proc_def_inode_operations = {
 727         .setattr        = proc_setattr,
 728 };
 729
 730 static int proc_single_show(struct seq_file *m, void *v)
 731 {
 732         struct inode *inode = m->private;
 733         struct pid_namespace *ns;
 734         struct pid *pid;
 735         struct task_struct *task;
 736         int ret;
 737
 738         ns = inode->i_sb->s_fs_info;
 739         pid = proc_pid(inode);
 740         task = get_pid_task(pid, PIDTYPE_PID);
 741         if (!task)
 742                 return -ESRCH;
 743
 744         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 745
 746         put_task_struct(task);
 747         return ret;
 748 }
 749
 750 static int proc_single_open(struct inode *inode, struct file *filp)
 751 {
 752         return single_open(filp, proc_single_show, inode);
 753 }
 754
 755 static const struct file_operations proc_single_file_operations = {
 756         .open           = proc_single_open,
 757         .read           = seq_read,
 758         .llseek         = seq_lseek,
 759         .release        = single_release,
 760 };
 761
 762
 763 struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
 764 {
 765         struct task_struct *task = get_proc_task(inode);
 766         struct mm_struct *mm = ERR_PTR(-ESRCH);
 767
 768         if (task) {
 769                 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
 770                 put_task_struct(task);
 771
 772                 if (!IS_ERR_OR_NULL(mm)) {
 773                         /* ensure this mm_struct can't be freed */
 774                         mmgrab(mm);
 775                         /* but do not pin its memory */
 776                         mmput(mm);
 777                 }
 778         }
 779
 780         return mm;
 781 }
 782
 783 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
 784 {
 785         struct mm_struct *mm = proc_mem_open(inode, mode);
 786
 787         if (IS_ERR(mm))
 788                 return PTR_ERR(mm);
 789
 790         file->private_data = mm;
 791         return 0;
 792 }
 793
 794 static int mem_open(struct inode *inode, struct file *file)
 795 {
 796         int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
 797
 798         /* OK to pass negative loff_t, we can catch out-of-range */
 799         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 800
 801         return ret;
 802 }
 803
 804 static ssize_t mem_rw(struct file *file, char __user *buf,
 805                         size_t count, loff_t *ppos, int write)
 806 {
 807         struct mm_struct *mm = file->private_data;
 808         unsigned long addr = *ppos;
 809         ssize_t copied;
 810         char *page;
 811         unsigned int flags;
 812
 813         if (!mm)
 814                 return 0;
 815
 816         page = (char *)__get_free_page(GFP_KERNEL);
 817         if (!page)
 818                 return -ENOMEM;
 819
 820         copied = 0;
 821         if (!mmget_not_zero(mm))
 822                 goto free;
 823
 824         flags = FOLL_FORCE | (write ? FOLL_WRITE : 0);
 825
 826         while (count > 0) {
 827                 int this_len = min_t(int, count, PAGE_SIZE);
 828
 829                 if (write && copy_from_user(page, buf, this_len)) {
 830                         copied = -EFAULT;
 831                         break;
 832                 }
 833
 834                 this_len = access_remote_vm(mm, addr, page, this_len, flags);
 835                 if (!this_len) {
 836                         if (!copied)
 837                                 copied = -EIO;
 838                         break;
 839                 }
 840
 841                 if (!write && copy_to_user(buf, page, this_len)) {
 842                         copied = -EFAULT;
 843                         break;
 844                 }
 845
 846                 buf += this_len;
 847                 addr += this_len;
 848                 copied += this_len;
 849                 count -= this_len;
 850         }
 851         *ppos = addr;
 852
 853         mmput(mm);
 854 free:
 855         free_page((unsigned long) page);
 856         return copied;
 857 }
 858
 859 static ssize_t mem_read(struct file *file, char __user *buf,
 860                         size_t count, loff_t *ppos)
 861 {
 862         return mem_rw(file, buf, count, ppos, 0);
 863 }
 864
 865 static ssize_t mem_write(struct file *file, const char __user *buf,
 866                          size_t count, loff_t *ppos)
 867 {
 868         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 869 }
 870
 871 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 872 {
 873         switch (orig) {
 874         case 0:
 875                 file->f_pos = offset;
 876                 break;
 877         case 1:
 878                 file->f_pos += offset;
 879                 break;
 880         default:
 881                 return -EINVAL;
 882         }
 883         force_successful_syscall_return();
 884         return file->f_pos;
 885 }
 886
 887 static int mem_release(struct inode *inode, struct file *file)
 888 {
 889         struct mm_struct *mm = file->private_data;
 890         if (mm)
 891                 mmdrop(mm);
 892         return 0;
 893 }
 894
 895 static const struct file_operations proc_mem_operations = {
 896         .llseek         = mem_lseek,
 897         .read           = mem_read,
 898         .write          = mem_write,
 899         .open           = mem_open,
 900         .release        = mem_release,
 901 };
 902
 903 static int environ_open(struct inode *inode, struct file *file)
 904 {
 905         return __mem_open(inode, file, PTRACE_MODE_READ);
 906 }
 907
 908 static ssize_t environ_read(struct file *file, char __user *buf,
 909                         size_t count, loff_t *ppos)
 910 {
 911         char *page;
 912         unsigned long src = *ppos;
 913         int ret = 0;
 914         struct mm_struct *mm = file->private_data;
 915         unsigned long env_start, env_end;
 916
 917         /* Ensure the process spawned far enough to have an environment. */
 918         if (!mm || !mm->env_end)
 919                 return 0;
 920
 921         page = (char *)__get_free_page(GFP_KERNEL);
 922         if (!page)
 923                 return -ENOMEM;
 924
 925         ret = 0;
 926         if (!mmget_not_zero(mm))
 927                 goto free;
 928
 929         down_read(&mm->mmap_sem);
 930         env_start = mm->env_start;
 931         env_end = mm->env_end;
 932         up_read(&mm->mmap_sem);
 933
 934         while (count > 0) {
 935                 size_t this_len, max_len;
 936                 int retval;
 937
 938                 if (src >= (env_end - env_start))
 939                         break;
 940
 941                 this_len = env_end - (env_start + src);
 942
 943                 max_len = min_t(size_t, PAGE_SIZE, count);
 944                 this_len = min(max_len, this_len);
 945
 946                 retval = access_remote_vm(mm, (env_start + src), page, this_len, 0);
 947
 948                 if (retval <= 0) {
 949                         ret = retval;
 950                         break;
 951                 }
 952
 953                 if (copy_to_user(buf, page, retval)) {
 954                         ret = -EFAULT;
 955                         break;
 956                 }
 957
 958                 ret += retval;
 959                 src += retval;
 960                 buf += retval;
 961                 count -= retval;
 962         }
 963         *ppos = src;
 964         mmput(mm);
 965
 966 free:
 967         free_page((unsigned long) page);
 968         return ret;
 969 }
 970
 971 static const struct file_operations proc_environ_operations = {
 972         .open           = environ_open,
 973         .read           = environ_read,
 974         .llseek         = generic_file_llseek,
 975         .release        = mem_release,
 976 };
 977
 978 static int auxv_open(struct inode *inode, struct file *file)
 979 {
 980         return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
 981 }
 982
 983 static ssize_t auxv_read(struct file *file, char __user *buf,
 984                         size_t count, loff_t *ppos)
 985 {
 986         struct mm_struct *mm = file->private_data;
 987         unsigned int nwords = 0;
 988
 989         if (!mm)
 990                 return 0;
 991         do {
 992                 nwords += 2;
 993         } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 994         return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
 995                                        nwords * sizeof(mm->saved_auxv[0]));
 996 }
 997
 998 static const struct file_operations proc_auxv_operations = {
 999         .open           = auxv_open,
1000         .read           = auxv_read,
1001         .llseek         = generic_file_llseek,
1002         .release        = mem_release,
1003 };
1004
1005 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1006                             loff_t *ppos)
1007 {
1008         struct task_struct *task = get_proc_task(file_inode(file));
1009         char buffer[PROC_NUMBUF];
1010         int oom_adj = OOM_ADJUST_MIN;
1011         size_t len;
1012
1013         if (!task)
1014                 return -ESRCH;
1015         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1016                 oom_adj = OOM_ADJUST_MAX;
1017         else
1018                 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1019                           OOM_SCORE_ADJ_MAX;
1020         put_task_struct(task);
1021         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1022         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1023 }
1024
1025 static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1026 {
1027         static DEFINE_MUTEX(oom_adj_mutex);
1028         struct mm_struct *mm = NULL;
1029         struct task_struct *task;
1030         int err = 0;
1031
1032         task = get_proc_task(file_inode(file));
1033         if (!task)
1034                 return -ESRCH;
1035
1036         mutex_lock(&oom_adj_mutex);
1037         if (legacy) {
1038                 if (oom_adj < task->signal->oom_score_adj &&
1039                                 !capable(CAP_SYS_RESOURCE)) {
1040                         err = -EACCES;
1041                         goto err_unlock;
1042                 }
1043                 /*
1044                  * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1045                  * /proc/pid/oom_score_adj instead.
1046                  */
1047                 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1048                           current->comm, task_pid_nr(current), task_pid_nr(task),
1049                           task_pid_nr(task));
1050         } else {
1051                 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1052                                 !capable(CAP_SYS_RESOURCE)) {
1053                         err = -EACCES;
1054                         goto err_unlock;
1055                 }
1056         }
1057
1058         /*
1059          * Make sure we will check other processes sharing the mm if this is
1060          * not vfrok which wants its own oom_score_adj.
1061          * pin the mm so it doesn't go away and get reused after task_unlock
1062          */
1063         if (!task->vfork_done) {
1064                 struct task_struct *p = find_lock_task_mm(task);
1065
1066                 if (p) {
1067                         if (atomic_read(&p->mm->mm_users) > 1) {
1068                                 mm = p->mm;
1069                                 mmgrab(mm);
1070                         }
1071                         task_unlock(p);
1072                 }
1073         }
1074
1075         task->signal->oom_score_adj = oom_adj;
1076         if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1077                 task->signal->oom_score_adj_min = (short)oom_adj;
1078         trace_oom_score_adj_update(task);
1079
1080         if (mm) {
1081                 struct task_struct *p;
1082
1083                 rcu_read_lock();
1084                 for_each_process(p) {
1085                         if (same_thread_group(task, p))
1086                                 continue;
1087
1088                         /* do not touch kernel threads or the global init */
1089                         if (p->flags & PF_KTHREAD || is_global_init(p))
1090                                 continue;
1091
1092                         task_lock(p);
1093                         if (!p->vfork_done && process_shares_mm(p, mm)) {
1094                                 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",
1095                                                 task_pid_nr(p), p->comm,
1096                                                 p->signal->oom_score_adj, oom_adj,
1097                                                 task_pid_nr(task), task->comm);
1098                                 p->signal->oom_score_adj = oom_adj;
1099                                 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1100                                         p->signal->oom_score_adj_min = (short)oom_adj;
1101                         }
1102                         task_unlock(p);
1103                 }
1104                 rcu_read_unlock();
1105                 mmdrop(mm);
1106         }
1107 err_unlock:
1108         mutex_unlock(&oom_adj_mutex);
1109         put_task_struct(task);
1110         return err;
1111 }
1112
1113 /*
1114  * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1115  * kernels.  The effective policy is defined by oom_score_adj, which has a
1116  * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1117  * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1118  * Processes that become oom disabled via oom_adj will still be oom disabled
1119  * with this implementation.
1120  *
1121  * oom_adj cannot be removed since existing userspace binaries use it.
1122  */
1123 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1124                              size_t count, loff_t *ppos)
1125 {
1126         char buffer[PROC_NUMBUF];
1127         int oom_adj;
1128         int err;
1129
1130         memset(buffer, 0, sizeof(buffer));
1131         if (count > sizeof(buffer) - 1)
1132                 count = sizeof(buffer) - 1;
1133         if (copy_from_user(buffer, buf, count)) {
1134                 err = -EFAULT;
1135                 goto out;
1136         }
1137
1138         err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1139         if (err)
1140                 goto out;
1141         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1142              oom_adj != OOM_DISABLE) {
1143                 err = -EINVAL;
1144                 goto out;
1145         }
1146
1147         /*
1148          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1149          * value is always attainable.
1150          */
1151         if (oom_adj == OOM_ADJUST_MAX)
1152                 oom_adj = OOM_SCORE_ADJ_MAX;
1153         else
1154                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1155
1156         err = __set_oom_adj(file, oom_adj, true);
1157 out:
1158         return err < 0 ? err : count;
1159 }
1160
1161 static const struct file_operations proc_oom_adj_operations = {
1162         .read           = oom_adj_read,
1163         .write          = oom_adj_write,
1164         .llseek         = generic_file_llseek,
1165 };
1166
1167 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1168                                         size_t count, loff_t *ppos)
1169 {
1170         struct task_struct *task = get_proc_task(file_inode(file));
1171         char buffer[PROC_NUMBUF];
1172         short oom_score_adj = OOM_SCORE_ADJ_MIN;
1173         size_t len;
1174
1175         if (!task)
1176                 return -ESRCH;
1177         oom_score_adj = task->signal->oom_score_adj;
1178         put_task_struct(task);
1179         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1180         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1181 }
1182
1183 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1184                                         size_t count, loff_t *ppos)
1185 {
1186         char buffer[PROC_NUMBUF];
1187         int oom_score_adj;
1188         int err;
1189
1190         memset(buffer, 0, sizeof(buffer));
1191         if (count > sizeof(buffer) - 1)
1192                 count = sizeof(buffer) - 1;
1193         if (copy_from_user(buffer, buf, count)) {
1194                 err = -EFAULT;
1195                 goto out;
1196         }
1197
1198         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1199         if (err)
1200                 goto out;
1201         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1202                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1203                 err = -EINVAL;
1204                 goto out;
1205         }
1206
1207         err = __set_oom_adj(file, oom_score_adj, false);
1208 out:
1209         return err < 0 ? err : count;
1210 }
1211
1212 static const struct file_operations proc_oom_score_adj_operations = {
1213         .read           = oom_score_adj_read,
1214         .write          = oom_score_adj_write,
1215         .llseek         = default_llseek,
1216 };
1217
1218 #ifdef CONFIG_AUDITSYSCALL
1219 #define TMPBUFLEN 11
1220 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1221                                   size_t count, loff_t *ppos)
1222 {
1223         struct inode * inode = file_inode(file);
1224         struct task_struct *task = get_proc_task(inode);
1225         ssize_t length;
1226         char tmpbuf[TMPBUFLEN];
1227
1228         if (!task)
1229                 return -ESRCH;
1230         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1231                            from_kuid(file->f_cred->user_ns,
1232                                      audit_get_loginuid(task)));
1233         put_task_struct(task);
1234         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1235 }
1236
1237 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1238                                    size_t count, loff_t *ppos)
1239 {
1240         struct inode * inode = file_inode(file);
1241         uid_t loginuid;
1242         kuid_t kloginuid;
1243         int rv;
1244
1245         rcu_read_lock();
1246         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1247                 rcu_read_unlock();
1248                 return -EPERM;
1249         }
1250         rcu_read_unlock();
1251
1252         if (*ppos != 0) {
1253                 /* No partial writes. */
1254                 return -EINVAL;
1255         }
1256
1257         rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1258         if (rv < 0)
1259                 return rv;
1260
1261         /* is userspace tring to explicitly UNSET the loginuid? */
1262         if (loginuid == AUDIT_UID_UNSET) {
1263                 kloginuid = INVALID_UID;
1264         } else {
1265                 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1266                 if (!uid_valid(kloginuid))
1267                         return -EINVAL;
1268         }
1269
1270         rv = audit_set_loginuid(kloginuid);
1271         if (rv < 0)
1272                 return rv;
1273         return count;
1274 }
1275
1276 static const struct file_operations proc_loginuid_operations = {
1277         .read           = proc_loginuid_read,
1278         .write          = proc_loginuid_write,
1279         .llseek         = generic_file_llseek,
1280 };
1281
1282 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1283                                   size_t count, loff_t *ppos)
1284 {
1285         struct inode * inode = file_inode(file);
1286         struct task_struct *task = get_proc_task(inode);
1287         ssize_t length;
1288         char tmpbuf[TMPBUFLEN];
1289
1290         if (!task)
1291                 return -ESRCH;
1292         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1293                                 audit_get_sessionid(task));
1294         put_task_struct(task);
1295         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1296 }
1297
1298 static const struct file_operations proc_sessionid_operations = {
1299         .read           = proc_sessionid_read,
1300         .llseek         = generic_file_llseek,
1301 };
1302 #endif
1303
1304 #ifdef CONFIG_FAULT_INJECTION
1305 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1306                                       size_t count, loff_t *ppos)
1307 {
1308         struct task_struct *task = get_proc_task(file_inode(file));
1309         char buffer[PROC_NUMBUF];
1310         size_t len;
1311         int make_it_fail;
1312
1313         if (!task)
1314                 return -ESRCH;
1315         make_it_fail = task->make_it_fail;
1316         put_task_struct(task);
1317
1318         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1319
1320         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1321 }
1322
1323 static ssize_t proc_fault_inject_write(struct file * file,
1324                         const char __user * buf, size_t count, loff_t *ppos)
1325 {
1326         struct task_struct *task;
1327         char buffer[PROC_NUMBUF];
1328         int make_it_fail;
1329         int rv;
1330
1331         if (!capable(CAP_SYS_RESOURCE))
1332                 return -EPERM;
1333         memset(buffer, 0, sizeof(buffer));
1334         if (count > sizeof(buffer) - 1)
1335                 count = sizeof(buffer) - 1;
1336         if (copy_from_user(buffer, buf, count))
1337                 return -EFAULT;
1338         rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1339         if (rv < 0)
1340                 return rv;
1341         if (make_it_fail < 0 || make_it_fail > 1)
1342                 return -EINVAL;
1343
1344         task = get_proc_task(file_inode(file));
1345         if (!task)
1346                 return -ESRCH;
1347         task->make_it_fail = make_it_fail;
1348         put_task_struct(task);
1349
1350         return count;
1351 }
1352
1353 static const struct file_operations proc_fault_inject_operations = {
1354         .read           = proc_fault_inject_read,
1355         .write          = proc_fault_inject_write,
1356         .llseek         = generic_file_llseek,
1357 };
1358
1359 static ssize_t proc_fail_nth_write(struct file *file, const char __user *buf,
1360                                    size_t count, loff_t *ppos)
1361 {
1362         struct task_struct *task;
1363         int err;
1364         unsigned int n;
1365
1366         err = kstrtouint_from_user(buf, count, 0, &n);
1367         if (err)
1368                 return err;
1369
1370         task = get_proc_task(file_inode(file));
1371         if (!task)
1372                 return -ESRCH;
1373         WRITE_ONCE(task->fail_nth, n);
1374         put_task_struct(task);
1375
1376         return count;
1377 }
1378
1379 static ssize_t proc_fail_nth_read(struct file *file, char __user *buf,
1380                                   size_t count, loff_t *ppos)
1381 {
1382         struct task_struct *task;
1383         char numbuf[PROC_NUMBUF];
1384         ssize_t len;
1385
1386         task = get_proc_task(file_inode(file));
1387         if (!task)
1388                 return -ESRCH;
1389         len = snprintf(numbuf, sizeof(numbuf), "%u\n",
1390                         READ_ONCE(task->fail_nth));
1391         len = simple_read_from_buffer(buf, count, ppos, numbuf, len);
1392         put_task_struct(task);
1393
1394         return len;
1395 }
1396
1397 static const struct file_operations proc_fail_nth_operations = {
1398         .read           = proc_fail_nth_read,
1399         .write          = proc_fail_nth_write,
1400 };
1401 #endif
1402
1403
1404 #ifdef CONFIG_SCHED_DEBUG
1405 /*
1406  * Print out various scheduling related per-task fields:
1407  */
1408 static int sched_show(struct seq_file *m, void *v)
1409 {
1410         struct inode *inode = m->private;
1411         struct pid_namespace *ns = inode->i_sb->s_fs_info;
1412         struct task_struct *p;
1413
1414         p = get_proc_task(inode);
1415         if (!p)
1416                 return -ESRCH;
1417         proc_sched_show_task(p, ns, m);
1418
1419         put_task_struct(p);
1420
1421         return 0;
1422 }
1423
1424 static ssize_t
1425 sched_write(struct file *file, const char __user *buf,
1426             size_t count, loff_t *offset)
1427 {
1428         struct inode *inode = file_inode(file);
1429         struct task_struct *p;
1430
1431         p = get_proc_task(inode);
1432         if (!p)
1433                 return -ESRCH;
1434         proc_sched_set_task(p);
1435
1436         put_task_struct(p);
1437
1438         return count;
1439 }
1440
1441 static int sched_open(struct inode *inode, struct file *filp)
1442 {
1443         return single_open(filp, sched_show, inode);
1444 }
1445
1446 static const struct file_operations proc_pid_sched_operations = {
1447         .open           = sched_open,
1448         .read           = seq_read,
1449         .write          = sched_write,
1450         .llseek         = seq_lseek,
1451         .release        = single_release,
1452 };
1453
1454 #endif
1455
1456 #ifdef CONFIG_SCHED_AUTOGROUP
1457 /*
1458  * Print out autogroup related information:
1459  */
1460 static int sched_autogroup_show(struct seq_file *m, void *v)
1461 {
1462         struct inode *inode = m->private;
1463         struct task_struct *p;
1464
1465         p = get_proc_task(inode);
1466         if (!p)
1467                 return -ESRCH;
1468         proc_sched_autogroup_show_task(p, m);
1469
1470         put_task_struct(p);
1471
1472         return 0;
1473 }
1474
1475 static ssize_t
1476 sched_autogroup_write(struct file *file, const char __user *buf,
1477             size_t count, loff_t *offset)
1478 {
1479         struct inode *inode = file_inode(file);
1480         struct task_struct *p;
1481         char buffer[PROC_NUMBUF];
1482         int nice;
1483         int err;
1484
1485         memset(buffer, 0, sizeof(buffer));
1486         if (count > sizeof(buffer) - 1)
1487                 count = sizeof(buffer) - 1;
1488         if (copy_from_user(buffer, buf, count))
1489                 return -EFAULT;
1490
1491         err = kstrtoint(strstrip(buffer), 0, &nice);
1492         if (err < 0)
1493                 return err;
1494
1495         p = get_proc_task(inode);
1496         if (!p)
1497                 return -ESRCH;
1498
1499         err = proc_sched_autogroup_set_nice(p, nice);
1500         if (err)
1501                 count = err;
1502
1503         put_task_struct(p);
1504
1505         return count;
1506 }
1507
1508 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1509 {
1510         int ret;
1511
1512         ret = single_open(filp, sched_autogroup_show, NULL);
1513         if (!ret) {
1514                 struct seq_file *m = filp->private_data;
1515
1516                 m->private = inode;
1517         }
1518         return ret;
1519 }
1520
1521 static const struct file_operations proc_pid_sched_autogroup_operations = {
1522         .open           = sched_autogroup_open,
1523         .read           = seq_read,
1524         .write          = sched_autogroup_write,
1525         .llseek         = seq_lseek,
1526         .release        = single_release,
1527 };
1528
1529 #endif /* CONFIG_SCHED_AUTOGROUP */
1530
1531 static ssize_t comm_write(struct file *file, const char __user *buf,
1532                                 size_t count, loff_t *offset)
1533 {
1534         struct inode *inode = file_inode(file);
1535         struct task_struct *p;
1536         char buffer[TASK_COMM_LEN];
1537         const size_t maxlen = sizeof(buffer) - 1;
1538
1539         memset(buffer, 0, sizeof(buffer));
1540         if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1541                 return -EFAULT;
1542
1543         p = get_proc_task(inode);
1544         if (!p)
1545                 return -ESRCH;
1546
1547         if (same_thread_group(current, p))
1548                 set_task_comm(p, buffer);
1549         else
1550                 count = -EINVAL;
1551
1552         put_task_struct(p);
1553
1554         return count;
1555 }
1556
1557 static int comm_show(struct seq_file *m, void *v)
1558 {
1559         struct inode *inode = m->private;
1560         struct task_struct *p;
1561
1562         p = get_proc_task(inode);
1563         if (!p)
1564                 return -ESRCH;
1565
1566         task_lock(p);
1567         seq_printf(m, "%s\n", p->comm);
1568         task_unlock(p);
1569
1570         put_task_struct(p);
1571
1572         return 0;
1573 }
1574
1575 static int comm_open(struct inode *inode, struct file *filp)
1576 {
1577         return single_open(filp, comm_show, inode);
1578 }
1579
1580 static const struct file_operations proc_pid_set_comm_operations = {
1581         .open           = comm_open,
1582         .read           = seq_read,
1583         .write          = comm_write,
1584         .llseek         = seq_lseek,
1585         .release        = single_release,
1586 };
1587
1588 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1589 {
1590         struct task_struct *task;
1591         struct file *exe_file;
1592
1593         task = get_proc_task(d_inode(dentry));
1594         if (!task)
1595                 return -ENOENT;
1596         exe_file = get_task_exe_file(task);
1597         put_task_struct(task);
1598         if (exe_file) {
1599                 *exe_path = exe_file->f_path;
1600                 path_get(&exe_file->f_path);
1601                 fput(exe_file);
1602                 return 0;
1603         } else
1604                 return -ENOENT;
1605 }
1606
1607 static const char *proc_pid_get_link(struct dentry *dentry,
1608                                      struct inode *inode,
1609                                      struct delayed_call *done)
1610 {
1611         struct path path;
1612         int error = -EACCES;
1613
1614         if (!dentry)
1615                 return ERR_PTR(-ECHILD);
1616
1617         /* Are we allowed to snoop on the tasks file descriptors? */
1618         if (!proc_fd_access_allowed(inode))
1619                 goto out;
1620
1621         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1622         if (error)
1623                 goto out;
1624
1625         nd_jump_link(&path);
1626         return NULL;
1627 out:
1628         return ERR_PTR(error);
1629 }
1630
1631 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1632 {
1633         char *tmp = (char *)__get_free_page(GFP_KERNEL);
1634         char *pathname;
1635         int len;
1636
1637         if (!tmp)
1638                 return -ENOMEM;
1639
1640         pathname = d_path(path, tmp, PAGE_SIZE);
1641         len = PTR_ERR(pathname);
1642         if (IS_ERR(pathname))
1643                 goto out;
1644         len = tmp + PAGE_SIZE - 1 - pathname;
1645
1646         if (len > buflen)
1647                 len = buflen;
1648         if (copy_to_user(buffer, pathname, len))
1649                 len = -EFAULT;
1650  out:
1651         free_page((unsigned long)tmp);
1652         return len;
1653 }
1654
1655 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1656 {
1657         int error = -EACCES;
1658         struct inode *inode = d_inode(dentry);
1659         struct path path;
1660
1661         /* Are we allowed to snoop on the tasks file descriptors? */
1662         if (!proc_fd_access_allowed(inode))
1663                 goto out;
1664
1665         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1666         if (error)
1667                 goto out;
1668
1669         error = do_proc_readlink(&path, buffer, buflen);
1670         path_put(&path);
1671 out:
1672         return error;
1673 }
1674
1675 const struct inode_operations proc_pid_link_inode_operations = {
1676         .readlink       = proc_pid_readlink,
1677         .get_link       = proc_pid_get_link,
1678         .setattr        = proc_setattr,
1679 };
1680
1681
1682 /* building an inode */
1683
1684 void task_dump_owner(struct task_struct *task, umode_t mode,
1685                      kuid_t *ruid, kgid_t *rgid)
1686 {
1687         /* Depending on the state of dumpable compute who should own a
1688          * proc file for a task.
1689          */
1690         const struct cred *cred;
1691         kuid_t uid;
1692         kgid_t gid;
1693
1694         /* Default to the tasks effective ownership */
1695         rcu_read_lock();
1696         cred = __task_cred(task);
1697         uid = cred->euid;
1698         gid = cred->egid;
1699         rcu_read_unlock();
1700
1701         /*
1702          * Before the /proc/pid/status file was created the only way to read
1703          * the effective uid of a /process was to stat /proc/pid.  Reading
1704          * /proc/pid/status is slow enough that procps and other packages
1705          * kept stating /proc/pid.  To keep the rules in /proc simple I have
1706          * made this apply to all per process world readable and executable
1707          * directories.
1708          */
1709         if (mode != (S_IFDIR|S_IRUGO|S_IXUGO)) {
1710                 struct mm_struct *mm;
1711                 task_lock(task);
1712                 mm = task->mm;
1713                 /* Make non-dumpable tasks owned by some root */
1714                 if (mm) {
1715                         if (get_dumpable(mm) != SUID_DUMP_USER) {
1716                                 struct user_namespace *user_ns = mm->user_ns;
1717
1718                                 uid = make_kuid(user_ns, 0);
1719                                 if (!uid_valid(uid))
1720                                         uid = GLOBAL_ROOT_UID;
1721
1722                                 gid = make_kgid(user_ns, 0);
1723                                 if (!gid_valid(gid))
1724                                         gid = GLOBAL_ROOT_GID;
1725                         }
1726                 } else {
1727                         uid = GLOBAL_ROOT_UID;
1728                         gid = GLOBAL_ROOT_GID;
1729                 }
1730                 task_unlock(task);
1731         }
1732         *ruid = uid;
1733         *rgid = gid;
1734 }
1735
1736 struct inode *proc_pid_make_inode(struct super_block * sb,
1737                                   struct task_struct *task, umode_t mode)
1738 {
1739         struct inode * inode;
1740         struct proc_inode *ei;
1741
1742         /* We need a new inode */
1743
1744         inode = new_inode(sb);
1745         if (!inode)
1746                 goto out;
1747
1748         /* Common stuff */
1749         ei = PROC_I(inode);
1750         inode->i_mode = mode;
1751         inode->i_ino = get_next_ino();
1752         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1753         inode->i_op = &proc_def_inode_operations;
1754
1755         /*
1756          * grab the reference to task.
1757          */
1758         ei->pid = get_task_pid(task, PIDTYPE_PID);
1759         if (!ei->pid)
1760                 goto out_unlock;
1761
1762         task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1763         security_task_to_inode(task, inode);
1764
1765 out:
1766         return inode;
1767
1768 out_unlock:
1769         iput(inode);
1770         return NULL;
1771 }
1772
1773 int pid_getattr(const struct path *path, struct kstat *stat,
1774                 u32 request_mask, unsigned int query_flags)
1775 {
1776         struct inode *inode = d_inode(path->dentry);
1777         struct task_struct *task;
1778         struct pid_namespace *pid = path->dentry->d_sb->s_fs_info;
1779
1780         generic_fillattr(inode, stat);
1781
1782         rcu_read_lock();
1783         stat->uid = GLOBAL_ROOT_UID;
1784         stat->gid = GLOBAL_ROOT_GID;
1785         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1786         if (task) {
1787                 if (!has_pid_permissions(pid, task, HIDEPID_INVISIBLE)) {
1788                         rcu_read_unlock();
1789                         /*
1790                          * This doesn't prevent learning whether PID exists,
1791                          * it only makes getattr() consistent with readdir().
1792                          */
1793                         return -ENOENT;
1794                 }
1795                 task_dump_owner(task, inode->i_mode, &stat->uid, &stat->gid);
1796         }
1797         rcu_read_unlock();
1798         return 0;
1799 }
1800
1801 /* dentry stuff */
1802
1803 /*
1804  *      Exceptional case: normally we are not allowed to unhash a busy
1805  * directory. In this case, however, we can do it - no aliasing problems
1806  * due to the way we treat inodes.
1807  *
1808  * Rewrite the inode's ownerships here because the owning task may have
1809  * performed a setuid(), etc.
1810  *
1811  */
1812 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1813 {
1814         struct inode *inode;
1815         struct task_struct *task;
1816
1817         if (flags & LOOKUP_RCU)
1818                 return -ECHILD;
1819
1820         inode = d_inode(dentry);
1821         task = get_proc_task(inode);
1822
1823         if (task) {
1824                 task_dump_owner(task, inode->i_mode, &inode->i_uid, &inode->i_gid);
1825
1826                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1827                 security_task_to_inode(task, inode);
1828                 put_task_struct(task);
1829                 return 1;
1830         }
1831         return 0;
1832 }
1833
1834 static inline bool proc_inode_is_dead(struct inode *inode)
1835 {
1836         return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1837 }
1838
1839 int pid_delete_dentry(const struct dentry *dentry)
1840 {
1841         /* Is the task we represent dead?
1842          * If so, then don't put the dentry on the lru list,
1843          * kill it immediately.
1844          */
1845         return proc_inode_is_dead(d_inode(dentry));
1846 }
1847
1848 const struct dentry_operations pid_dentry_operations =
1849 {
1850         .d_revalidate   = pid_revalidate,
1851         .d_delete       = pid_delete_dentry,
1852 };
1853
1854 /* Lookups */
1855
1856 /*
1857  * Fill a directory entry.
1858  *
1859  * If possible create the dcache entry and derive our inode number and
1860  * file type from dcache entry.
1861  *
1862  * Since all of the proc inode numbers are dynamically generated, the inode
1863  * numbers do not exist until the inode is cache.  This means creating the
1864  * the dcache entry in readdir is necessary to keep the inode numbers
1865  * reported by readdir in sync with the inode numbers reported
1866  * by stat.
1867  */
1868 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1869         const char *name, int len,
1870         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1871 {
1872         struct dentry *child, *dir = file->f_path.dentry;
1873         struct qstr qname = QSTR_INIT(name, len);
1874         struct inode *inode;
1875         unsigned type;
1876         ino_t ino;
1877
1878         child = d_hash_and_lookup(dir, &qname);
1879         if (!child) {
1880                 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1881                 child = d_alloc_parallel(dir, &qname, &wq);
1882                 if (IS_ERR(child))
1883                         goto end_instantiate;
1884                 if (d_in_lookup(child)) {
1885                         int err = instantiate(d_inode(dir), child, task, ptr);
1886                         d_lookup_done(child);
1887                         if (err < 0) {
1888                                 dput(child);
1889                                 goto end_instantiate;
1890                         }
1891                 }
1892         }
1893         inode = d_inode(child);
1894         ino = inode->i_ino;
1895         type = inode->i_mode >> 12;
1896         dput(child);
1897         return dir_emit(ctx, name, len, ino, type);
1898
1899 end_instantiate:
1900         return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1901 }
1902
1903 /*
1904  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1905  * which represent vma start and end addresses.
1906  */
1907 static int dname_to_vma_addr(struct dentry *dentry,
1908                              unsigned long *start, unsigned long *end)
1909 {
1910         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1911                 return -EINVAL;
1912
1913         return 0;
1914 }
1915
1916 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1917 {
1918         unsigned long vm_start, vm_end;
1919         bool exact_vma_exists = false;
1920         struct mm_struct *mm = NULL;
1921         struct task_struct *task;
1922         struct inode *inode;
1923         int status = 0;
1924
1925         if (flags & LOOKUP_RCU)
1926                 return -ECHILD;
1927
1928         inode = d_inode(dentry);
1929         task = get_proc_task(inode);
1930         if (!task)
1931                 goto out_notask;
1932
1933         mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1934         if (IS_ERR_OR_NULL(mm))
1935                 goto out;
1936
1937         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1938                 down_read(&mm->mmap_sem);
1939                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1940                 up_read(&mm->mmap_sem);
1941         }
1942
1943         mmput(mm);
1944
1945         if (exact_vma_exists) {
1946                 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1947
1948                 security_task_to_inode(task, inode);
1949                 status = 1;
1950         }
1951
1952 out:
1953         put_task_struct(task);
1954
1955 out_notask:
1956         return status;
1957 }
1958
1959 static const struct dentry_operations tid_map_files_dentry_operations = {
1960         .d_revalidate   = map_files_d_revalidate,
1961         .d_delete       = pid_delete_dentry,
1962 };
1963
1964 static int map_files_get_link(struct dentry *dentry, struct path *path)
1965 {
1966         unsigned long vm_start, vm_end;
1967         struct vm_area_struct *vma;
1968         struct task_struct *task;
1969         struct mm_struct *mm;
1970         int rc;
1971
1972         rc = -ENOENT;
1973         task = get_proc_task(d_inode(dentry));
1974         if (!task)
1975                 goto out;
1976
1977         mm = get_task_mm(task);
1978         put_task_struct(task);
1979         if (!mm)
1980                 goto out;
1981
1982         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1983         if (rc)
1984                 goto out_mmput;
1985
1986         rc = -ENOENT;
1987         down_read(&mm->mmap_sem);
1988         vma = find_exact_vma(mm, vm_start, vm_end);
1989         if (vma && vma->vm_file) {
1990                 *path = vma->vm_file->f_path;
1991                 path_get(path);
1992                 rc = 0;
1993         }
1994         up_read(&mm->mmap_sem);
1995
1996 out_mmput:
1997         mmput(mm);
1998 out:
1999         return rc;
2000 }
2001
2002 struct map_files_info {
2003         fmode_t         mode;
2004         unsigned int    len;
2005         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
2006 };
2007
2008 /*
2009  * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
2010  * symlinks may be used to bypass permissions on ancestor directories in the
2011  * path to the file in question.
2012  */
2013 static const char *
2014 proc_map_files_get_link(struct dentry *dentry,
2015                         struct inode *inode,
2016                         struct delayed_call *done)
2017 {
2018         if (!capable(CAP_SYS_ADMIN))
2019                 return ERR_PTR(-EPERM);
2020
2021         return proc_pid_get_link(dentry, inode, done);
2022 }
2023
2024 /*
2025  * Identical to proc_pid_link_inode_operations except for get_link()
2026  */
2027 static const struct inode_operations proc_map_files_link_inode_operations = {
2028         .readlink       = proc_pid_readlink,
2029         .get_link       = proc_map_files_get_link,
2030         .setattr        = proc_setattr,
2031 };
2032
2033 static int
2034 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
2035                            struct task_struct *task, const void *ptr)
2036 {
2037         fmode_t mode = (fmode_t)(unsigned long)ptr;
2038         struct proc_inode *ei;
2039         struct inode *inode;
2040
2041         inode = proc_pid_make_inode(dir->i_sb, task, S_IFLNK |
2042                                     ((mode & FMODE_READ ) ? S_IRUSR : 0) |
2043                                     ((mode & FMODE_WRITE) ? S_IWUSR : 0));
2044         if (!inode)
2045                 return -ENOENT;
2046
2047         ei = PROC_I(inode);
2048         ei->op.proc_get_link = map_files_get_link;
2049
2050         inode->i_op = &proc_map_files_link_inode_operations;
2051         inode->i_size = 64;
2052
2053         d_set_d_op(dentry, &tid_map_files_dentry_operations);
2054         d_add(dentry, inode);
2055
2056         return 0;
2057 }
2058
2059 static struct dentry *proc_map_files_lookup(struct inode *dir,
2060                 struct dentry *dentry, unsigned int flags)
2061 {
2062         unsigned long vm_start, vm_end;
2063         struct vm_area_struct *vma;
2064         struct task_struct *task;
2065         int result;
2066         struct mm_struct *mm;
2067
2068         result = -ENOENT;
2069         task = get_proc_task(dir);
2070         if (!task)
2071                 goto out;
2072
2073         result = -EACCES;
2074         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2075                 goto out_put_task;
2076
2077         result = -ENOENT;
2078         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2079                 goto out_put_task;
2080
2081         mm = get_task_mm(task);
2082         if (!mm)
2083                 goto out_put_task;
2084
2085         down_read(&mm->mmap_sem);
2086         vma = find_exact_vma(mm, vm_start, vm_end);
2087         if (!vma)
2088                 goto out_no_vma;
2089
2090         if (vma->vm_file)
2091                 result = proc_map_files_instantiate(dir, dentry, task,
2092                                 (void *)(unsigned long)vma->vm_file->f_mode);
2093
2094 out_no_vma:
2095         up_read(&mm->mmap_sem);
2096         mmput(mm);
2097 out_put_task:
2098         put_task_struct(task);
2099 out:
2100         return ERR_PTR(result);
2101 }
2102
2103 static const struct inode_operations proc_map_files_inode_operations = {
2104         .lookup         = proc_map_files_lookup,
2105         .permission     = proc_fd_permission,
2106         .setattr        = proc_setattr,
2107 };
2108
2109 static int
2110 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2111 {
2112         struct vm_area_struct *vma;
2113         struct task_struct *task;
2114         struct mm_struct *mm;
2115         unsigned long nr_files, pos, i;
2116         struct flex_array *fa = NULL;
2117         struct map_files_info info;
2118         struct map_files_info *p;
2119         int ret;
2120
2121         ret = -ENOENT;
2122         task = get_proc_task(file_inode(file));
2123         if (!task)
2124                 goto out;
2125
2126         ret = -EACCES;
2127         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2128                 goto out_put_task;
2129
2130         ret = 0;
2131         if (!dir_emit_dots(file, ctx))
2132                 goto out_put_task;
2133
2134         mm = get_task_mm(task);
2135         if (!mm)
2136                 goto out_put_task;
2137         down_read(&mm->mmap_sem);
2138
2139         nr_files = 0;
2140
2141         /*
2142          * We need two passes here:
2143          *
2144          *  1) Collect vmas of mapped files with mmap_sem taken
2145          *  2) Release mmap_sem and instantiate entries
2146          *
2147          * otherwise we get lockdep complained, since filldir()
2148          * routine might require mmap_sem taken in might_fault().
2149          */
2150
2151         for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2152                 if (vma->vm_file && ++pos > ctx->pos)
2153                         nr_files++;
2154         }
2155
2156         if (nr_files) {
2157                 fa = flex_array_alloc(sizeof(info), nr_files,
2158                                         GFP_KERNEL);
2159                 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2160                                                 GFP_KERNEL)) {
2161                         ret = -ENOMEM;
2162                         if (fa)
2163                                 flex_array_free(fa);
2164                         up_read(&mm->mmap_sem);
2165                         mmput(mm);
2166                         goto out_put_task;
2167                 }
2168                 for (i = 0, vma = mm->mmap, pos = 2; vma;
2169                                 vma = vma->vm_next) {
2170                         if (!vma->vm_file)
2171                                 continue;
2172                         if (++pos <= ctx->pos)
2173                                 continue;
2174
2175                         info.mode = vma->vm_file->f_mode;
2176                         info.len = snprintf(info.name,
2177                                         sizeof(info.name), "%lx-%lx",
2178                                         vma->vm_start, vma->vm_end);
2179                         if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2180                                 BUG();
2181                 }
2182         }
2183         up_read(&mm->mmap_sem);
2184
2185         for (i = 0; i < nr_files; i++) {
2186                 p = flex_array_get(fa, i);
2187                 if (!proc_fill_cache(file, ctx,
2188                                       p->name, p->len,
2189                                       proc_map_files_instantiate,
2190                                       task,
2191                                       (void *)(unsigned long)p->mode))
2192                         break;
2193                 ctx->pos++;
2194         }
2195         if (fa)
2196                 flex_array_free(fa);
2197         mmput(mm);
2198
2199 out_put_task:
2200         put_task_struct(task);
2201 out:
2202         return ret;
2203 }
2204
2205 static const struct file_operations proc_map_files_operations = {
2206         .read           = generic_read_dir,
2207         .iterate_shared = proc_map_files_readdir,
2208         .llseek         = generic_file_llseek,
2209 };
2210
2211 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2212 struct timers_private {
2213         struct pid *pid;
2214         struct task_struct *task;
2215         struct sighand_struct *sighand;
2216         struct pid_namespace *ns;
2217         unsigned long flags;
2218 };
2219
2220 static void *timers_start(struct seq_file *m, loff_t *pos)
2221 {
2222         struct timers_private *tp = m->private;
2223
2224         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2225         if (!tp->task)
2226                 return ERR_PTR(-ESRCH);
2227
2228         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2229         if (!tp->sighand)
2230                 return ERR_PTR(-ESRCH);
2231
2232         return seq_list_start(&tp->task->signal->posix_timers, *pos);
2233 }
2234
2235 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2236 {
2237         struct timers_private *tp = m->private;
2238         return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2239 }
2240
2241 static void timers_stop(struct seq_file *m, void *v)
2242 {
2243         struct timers_private *tp = m->private;
2244
2245         if (tp->sighand) {
2246                 unlock_task_sighand(tp->task, &tp->flags);
2247                 tp->sighand = NULL;
2248         }
2249
2250         if (tp->task) {
2251                 put_task_struct(tp->task);
2252                 tp->task = NULL;
2253         }
2254 }
2255
2256 static int show_timer(struct seq_file *m, void *v)
2257 {
2258         struct k_itimer *timer;
2259         struct timers_private *tp = m->private;
2260         int notify;
2261         static const char * const nstr[] = {
2262                 [SIGEV_SIGNAL] = "signal",
2263                 [SIGEV_NONE] = "none",
2264                 [SIGEV_THREAD] = "thread",
2265         };
2266
2267         timer = list_entry((struct list_head *)v, struct k_itimer, list);
2268         notify = timer->it_sigev_notify;
2269
2270         seq_printf(m, "ID: %d\n", timer->it_id);
2271         seq_printf(m, "signal: %d/%px\n",
2272                    timer->sigq->info.si_signo,
2273                    timer->sigq->info.si_value.sival_ptr);
2274         seq_printf(m, "notify: %s/%s.%d\n",
2275                    nstr[notify & ~SIGEV_THREAD_ID],
2276                    (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2277                    pid_nr_ns(timer->it_pid, tp->ns));
2278         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2279
2280         return 0;
2281 }
2282
2283 static const struct seq_operations proc_timers_seq_ops = {
2284         .start  = timers_start,
2285         .next   = timers_next,
2286         .stop   = timers_stop,
2287         .show   = show_timer,
2288 };
2289
2290 static int proc_timers_open(struct inode *inode, struct file *file)
2291 {
2292         struct timers_private *tp;
2293
2294         tp = __seq_open_private(file, &proc_timers_seq_ops,
2295                         sizeof(struct timers_private));
2296         if (!tp)
2297                 return -ENOMEM;
2298
2299         tp->pid = proc_pid(inode);
2300         tp->ns = inode->i_sb->s_fs_info;
2301         return 0;
2302 }
2303
2304 static const struct file_operations proc_timers_operations = {
2305         .open           = proc_timers_open,
2306         .read           = seq_read,
2307         .llseek         = seq_lseek,
2308         .release        = seq_release_private,
2309 };
2310 #endif
2311
2312 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2313                                         size_t count, loff_t *offset)
2314 {
2315         struct inode *inode = file_inode(file);
2316         struct task_struct *p;
2317         u64 slack_ns;
2318         int err;
2319
2320         err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2321         if (err < 0)
2322                 return err;
2323
2324         p = get_proc_task(inode);
2325         if (!p)
2326                 return -ESRCH;
2327
2328         if (p != current) {
2329                 if (!capable(CAP_SYS_NICE)) {
2330                         count = -EPERM;
2331                         goto out;
2332                 }
2333
2334                 err = security_task_setscheduler(p);
2335                 if (err) {
2336                         count = err;
2337                         goto out;
2338                 }
2339         }
2340
2341         task_lock(p);
2342         if (slack_ns == 0)
2343                 p->timer_slack_ns = p->default_timer_slack_ns;
2344         else
2345                 p->timer_slack_ns = slack_ns;
2346         task_unlock(p);
2347
2348 out:
2349         put_task_struct(p);
2350
2351         return count;
2352 }
2353
2354 static int timerslack_ns_show(struct seq_file *m, void *v)
2355 {
2356         struct inode *inode = m->private;
2357         struct task_struct *p;
2358         int err = 0;
2359
2360         p = get_proc_task(inode);
2361         if (!p)
2362                 return -ESRCH;
2363
2364         if (p != current) {
2365
2366                 if (!capable(CAP_SYS_NICE)) {
2367                         err = -EPERM;
2368                         goto out;
2369                 }
2370                 err = security_task_getscheduler(p);
2371                 if (err)
2372                         goto out;
2373         }
2374
2375         task_lock(p);
2376         seq_printf(m, "%llu\n", p->timer_slack_ns);
2377         task_unlock(p);
2378
2379 out:
2380         put_task_struct(p);
2381
2382         return err;
2383 }
2384
2385 static int timerslack_ns_open(struct inode *inode, struct file *filp)
2386 {
2387         return single_open(filp, timerslack_ns_show, inode);
2388 }
2389
2390 static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2391         .open           = timerslack_ns_open,
2392         .read           = seq_read,
2393         .write          = timerslack_ns_write,
2394         .llseek         = seq_lseek,
2395         .release        = single_release,
2396 };
2397
2398 static int proc_pident_instantiate(struct inode *dir,
2399         struct dentry *dentry, struct task_struct *task, const void *ptr)
2400 {
2401         const struct pid_entry *p = ptr;
2402         struct inode *inode;
2403         struct proc_inode *ei;
2404
2405         inode = proc_pid_make_inode(dir->i_sb, task, p->mode);
2406         if (!inode)
2407                 goto out;
2408
2409         ei = PROC_I(inode);
2410         if (S_ISDIR(inode->i_mode))
2411                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2412         if (p->iop)
2413                 inode->i_op = p->iop;
2414         if (p->fop)
2415                 inode->i_fop = p->fop;
2416         ei->op = p->op;
2417         d_set_d_op(dentry, &pid_dentry_operations);
2418         d_add(dentry, inode);
2419         /* Close the race of the process dying before we return the dentry */
2420         if (pid_revalidate(dentry, 0))
2421                 return 0;
2422 out:
2423         return -ENOENT;
2424 }
2425
2426 static struct dentry *proc_pident_lookup(struct inode *dir,
2427                                          struct dentry *dentry,
2428                                          const struct pid_entry *ents,
2429                                          unsigned int nents)
2430 {
2431         int error;
2432         struct task_struct *task = get_proc_task(dir);
2433         const struct pid_entry *p, *last;
2434
2435         error = -ENOENT;
2436
2437         if (!task)
2438                 goto out_no_task;
2439
2440         /*
2441          * Yes, it does not scale. And it should not. Don't add
2442          * new entries into /proc/<tgid>/ without very good reasons.
2443          */
2444         last = &ents[nents];
2445         for (p = ents; p < last; p++) {
2446                 if (p->len != dentry->d_name.len)
2447                         continue;
2448                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2449                         break;
2450         }
2451         if (p >= last)
2452                 goto out;
2453
2454         error = proc_pident_instantiate(dir, dentry, task, p);
2455 out:
2456         put_task_struct(task);
2457 out_no_task:
2458         return ERR_PTR(error);
2459 }
2460
2461 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2462                 const struct pid_entry *ents, unsigned int nents)
2463 {
2464         struct task_struct *task = get_proc_task(file_inode(file));
2465         const struct pid_entry *p;
2466
2467         if (!task)
2468                 return -ENOENT;
2469
2470         if (!dir_emit_dots(file, ctx))
2471                 goto out;
2472
2473         if (ctx->pos >= nents + 2)
2474                 goto out;
2475
2476         for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2477                 if (!proc_fill_cache(file, ctx, p->name, p->len,
2478                                 proc_pident_instantiate, task, p))
2479                         break;
2480                 ctx->pos++;
2481         }
2482 out:
2483         put_task_struct(task);
2484         return 0;
2485 }
2486
2487 #ifdef CONFIG_SECURITY
2488 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2489                                   size_t count, loff_t *ppos)
2490 {
2491         struct inode * inode = file_inode(file);
2492         char *p = NULL;
2493         ssize_t length;
2494         struct task_struct *task = get_proc_task(inode);
2495
2496         if (!task)
2497                 return -ESRCH;
2498
2499         length = security_getprocattr(task,
2500                                       (char*)file->f_path.dentry->d_name.name,
2501                                       &p);
2502         put_task_struct(task);
2503         if (length > 0)
2504                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2505         kfree(p);
2506         return length;
2507 }
2508
2509 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2510                                    size_t count, loff_t *ppos)
2511 {
2512         struct inode * inode = file_inode(file);
2513         void *page;
2514         ssize_t length;
2515         struct task_struct *task = get_proc_task(inode);
2516
2517         length = -ESRCH;
2518         if (!task)
2519                 goto out_no_task;
2520
2521         /* A task may only write its own attributes. */
2522         length = -EACCES;
2523         if (current != task)
2524                 goto out;
2525
2526         if (count > PAGE_SIZE)
2527                 count = PAGE_SIZE;
2528
2529         /* No partial writes. */
2530         length = -EINVAL;
2531         if (*ppos != 0)
2532                 goto out;
2533
2534         page = memdup_user(buf, count);
2535         if (IS_ERR(page)) {
2536                 length = PTR_ERR(page);
2537                 goto out;
2538         }
2539
2540         /* Guard against adverse ptrace interaction */
2541         length = mutex_lock_interruptible(&current->signal->cred_guard_mutex);
2542         if (length < 0)
2543                 goto out_free;
2544
2545         length = security_setprocattr(file->f_path.dentry->d_name.name,
2546                                       page, count);
2547         mutex_unlock(&current->signal->cred_guard_mutex);
2548 out_free:
2549         kfree(page);
2550 out:
2551         put_task_struct(task);
2552 out_no_task:
2553         return length;
2554 }
2555
2556 static const struct file_operations proc_pid_attr_operations = {
2557         .read           = proc_pid_attr_read,
2558         .write          = proc_pid_attr_write,
2559         .llseek         = generic_file_llseek,
2560 };
2561
2562 static const struct pid_entry attr_dir_stuff[] = {
2563         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2564         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2565         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2566         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2567         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2568         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2569 };
2570
2571 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2572 {
2573         return proc_pident_readdir(file, ctx,
2574                                    attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2575 }
2576
2577 static const struct file_operations proc_attr_dir_operations = {
2578         .read           = generic_read_dir,
2579         .iterate_shared = proc_attr_dir_readdir,
2580         .llseek         = generic_file_llseek,
2581 };
2582
2583 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2584                                 struct dentry *dentry, unsigned int flags)
2585 {
2586         return proc_pident_lookup(dir, dentry,
2587                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2588 }
2589
2590 static const struct inode_operations proc_attr_dir_inode_operations = {
2591         .lookup         = proc_attr_dir_lookup,
2592         .getattr        = pid_getattr,
2593         .setattr        = proc_setattr,
2594 };
2595
2596 #endif
2597
2598 #ifdef CONFIG_ELF_CORE
2599 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2600                                          size_t count, loff_t *ppos)
2601 {
2602         struct task_struct *task = get_proc_task(file_inode(file));
2603         struct mm_struct *mm;
2604         char buffer[PROC_NUMBUF];
2605         size_t len;
2606         int ret;
2607
2608         if (!task)
2609                 return -ESRCH;
2610
2611         ret = 0;
2612         mm = get_task_mm(task);
2613         if (mm) {
2614                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2615                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2616                                 MMF_DUMP_FILTER_SHIFT));
2617                 mmput(mm);
2618                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2619         }
2620
2621         put_task_struct(task);
2622
2623         return ret;
2624 }
2625
2626 static ssize_t proc_coredump_filter_write(struct file *file,
2627                                           const char __user *buf,
2628                                           size_t count,
2629                                           loff_t *ppos)
2630 {
2631         struct task_struct *task;
2632         struct mm_struct *mm;
2633         unsigned int val;
2634         int ret;
2635         int i;
2636         unsigned long mask;
2637
2638         ret = kstrtouint_from_user(buf, count, 0, &val);
2639         if (ret < 0)
2640                 return ret;
2641
2642         ret = -ESRCH;
2643         task = get_proc_task(file_inode(file));
2644         if (!task)
2645                 goto out_no_task;
2646
2647         mm = get_task_mm(task);
2648         if (!mm)
2649                 goto out_no_mm;
2650         ret = 0;
2651
2652         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2653                 if (val & mask)
2654                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2655                 else
2656                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2657         }
2658
2659         mmput(mm);
2660  out_no_mm:
2661         put_task_struct(task);
2662  out_no_task:
2663         if (ret < 0)
2664                 return ret;
2665         return count;
2666 }
2667
2668 static const struct file_operations proc_coredump_filter_operations = {
2669         .read           = proc_coredump_filter_read,
2670         .write          = proc_coredump_filter_write,
2671         .llseek         = generic_file_llseek,
2672 };
2673 #endif
2674
2675 #ifdef CONFIG_TASK_IO_ACCOUNTING
2676 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2677 {
2678         struct task_io_accounting acct = task->ioac;
2679         unsigned long flags;
2680         int result;
2681
2682         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2683         if (result)
2684                 return result;
2685
2686         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2687                 result = -EACCES;
2688                 goto out_unlock;
2689         }
2690
2691         if (whole && lock_task_sighand(task, &flags)) {
2692                 struct task_struct *t = task;
2693
2694                 task_io_accounting_add(&acct, &task->signal->ioac);
2695                 while_each_thread(task, t)
2696                         task_io_accounting_add(&acct, &t->ioac);
2697
2698                 unlock_task_sighand(task, &flags);
2699         }
2700         seq_printf(m,
2701                    "rchar: %llu\n"
2702                    "wchar: %llu\n"
2703                    "syscr: %llu\n"
2704                    "syscw: %llu\n"
2705                    "read_bytes: %llu\n"
2706                    "write_bytes: %llu\n"
2707                    "cancelled_write_bytes: %llu\n",
2708                    (unsigned long long)acct.rchar,
2709                    (unsigned long long)acct.wchar,
2710                    (unsigned long long)acct.syscr,
2711                    (unsigned long long)acct.syscw,
2712                    (unsigned long long)acct.read_bytes,
2713                    (unsigned long long)acct.write_bytes,
2714                    (unsigned long long)acct.cancelled_write_bytes);
2715         result = 0;
2716
2717 out_unlock:
2718         mutex_unlock(&task->signal->cred_guard_mutex);
2719         return result;
2720 }
2721
2722 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2723                                   struct pid *pid, struct task_struct *task)
2724 {
2725         return do_io_accounting(task, m, 0);
2726 }
2727
2728 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2729                                    struct pid *pid, struct task_struct *task)
2730 {
2731         return do_io_accounting(task, m, 1);
2732 }
2733 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2734
2735 #ifdef CONFIG_USER_NS
2736 static int proc_id_map_open(struct inode *inode, struct file *file,
2737         const struct seq_operations *seq_ops)
2738 {
2739         struct user_namespace *ns = NULL;
2740         struct task_struct *task;
2741         struct seq_file *seq;
2742         int ret = -EINVAL;
2743
2744         task = get_proc_task(inode);
2745         if (task) {
2746                 rcu_read_lock();
2747                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2748                 rcu_read_unlock();
2749                 put_task_struct(task);
2750         }
2751         if (!ns)
2752                 goto err;
2753
2754         ret = seq_open(file, seq_ops);
2755         if (ret)
2756                 goto err_put_ns;
2757
2758         seq = file->private_data;
2759         seq->private = ns;
2760
2761         return 0;
2762 err_put_ns:
2763         put_user_ns(ns);
2764 err:
2765         return ret;
2766 }
2767
2768 static int proc_id_map_release(struct inode *inode, struct file *file)
2769 {
2770         struct seq_file *seq = file->private_data;
2771         struct user_namespace *ns = seq->private;
2772         put_user_ns(ns);
2773         return seq_release(inode, file);
2774 }
2775
2776 static int proc_uid_map_open(struct inode *inode, struct file *file)
2777 {
2778         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2779 }
2780
2781 static int proc_gid_map_open(struct inode *inode, struct file *file)
2782 {
2783         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2784 }
2785
2786 static int proc_projid_map_open(struct inode *inode, struct file *file)
2787 {
2788         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2789 }
2790
2791 static const struct file_operations proc_uid_map_operations = {
2792         .open           = proc_uid_map_open,
2793         .write          = proc_uid_map_write,
2794         .read           = seq_read,
2795         .llseek         = seq_lseek,
2796         .release        = proc_id_map_release,
2797 };
2798
2799 static const struct file_operations proc_gid_map_operations = {
2800         .open           = proc_gid_map_open,
2801         .write          = proc_gid_map_write,
2802         .read           = seq_read,
2803         .llseek         = seq_lseek,
2804         .release        = proc_id_map_release,
2805 };
2806
2807 static const struct file_operations proc_projid_map_operations = {
2808         .open           = proc_projid_map_open,
2809         .write          = proc_projid_map_write,
2810         .read           = seq_read,
2811         .llseek         = seq_lseek,
2812         .release        = proc_id_map_release,
2813 };
2814
2815 static int proc_setgroups_open(struct inode *inode, struct file *file)
2816 {
2817         struct user_namespace *ns = NULL;
2818         struct task_struct *task;
2819         int ret;
2820
2821         ret = -ESRCH;
2822         task = get_proc_task(inode);
2823         if (task) {
2824                 rcu_read_lock();
2825                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2826                 rcu_read_unlock();
2827                 put_task_struct(task);
2828         }
2829         if (!ns)
2830                 goto err;
2831
2832         if (file->f_mode & FMODE_WRITE) {
2833                 ret = -EACCES;
2834                 if (!ns_capable(ns, CAP_SYS_ADMIN))
2835                         goto err_put_ns;
2836         }
2837
2838         ret = single_open(file, &proc_setgroups_show, ns);
2839         if (ret)
2840                 goto err_put_ns;
2841
2842         return 0;
2843 err_put_ns:
2844         put_user_ns(ns);
2845 err:
2846         return ret;
2847 }
2848
2849 static int proc_setgroups_release(struct inode *inode, struct file *file)
2850 {
2851         struct seq_file *seq = file->private_data;
2852         struct user_namespace *ns = seq->private;
2853         int ret = single_release(inode, file);
2854         put_user_ns(ns);
2855         return ret;
2856 }
2857
2858 static const struct file_operations proc_setgroups_operations = {
2859         .open           = proc_setgroups_open,
2860         .write          = proc_setgroups_write,
2861         .read           = seq_read,
2862         .llseek         = seq_lseek,
2863         .release        = proc_setgroups_release,
2864 };
2865 #endif /* CONFIG_USER_NS */
2866
2867 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2868                                 struct pid *pid, struct task_struct *task)
2869 {
2870         int err = lock_trace(task);
2871         if (!err) {
2872                 seq_printf(m, "%08x\n", task->personality);
2873                 unlock_trace(task);
2874         }
2875         return err;
2876 }
2877
2878 #ifdef CONFIG_LIVEPATCH
2879 static int proc_pid_patch_state(struct seq_file *m, struct pid_namespace *ns,
2880                                 struct pid *pid, struct task_struct *task)
2881 {
2882         seq_printf(m, "%d\n", task->patch_state);
2883         return 0;
2884 }
2885 #endif /* CONFIG_LIVEPATCH */
2886
2887 /*
2888  * Thread groups
2889  */
2890 static const struct file_operations proc_task_operations;
2891 static const struct inode_operations proc_task_inode_operations;
2892
2893 static const struct pid_entry tgid_base_stuff[] = {
2894         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2895         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2896         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2897         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2898         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2899 #ifdef CONFIG_NET
2900         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2901 #endif
2902         REG("environ",    S_IRUSR, proc_environ_operations),
2903         REG("auxv",       S_IRUSR, proc_auxv_operations),
2904         ONE("status",     S_IRUGO, proc_pid_status),
2905         ONE("personality", S_IRUSR, proc_pid_personality),
2906         ONE("limits",     S_IRUGO, proc_pid_limits),
2907 #ifdef CONFIG_SCHED_DEBUG
2908         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2909 #endif
2910 #ifdef CONFIG_SCHED_AUTOGROUP
2911         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2912 #endif
2913         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2914 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2915         ONE("syscall",    S_IRUSR, proc_pid_syscall),
2916 #endif
2917         REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
2918         ONE("stat",       S_IRUGO, proc_tgid_stat),
2919         ONE("statm",      S_IRUGO, proc_pid_statm),
2920         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2921 #ifdef CONFIG_NUMA
2922         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2923 #endif
2924         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2925         LNK("cwd",        proc_cwd_link),
2926         LNK("root",       proc_root_link),
2927         LNK("exe",        proc_exe_link),
2928         REG("mounts",     S_IRUGO, proc_mounts_operations),
2929         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2930         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2931 #ifdef CONFIG_PROC_PAGE_MONITOR
2932         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2933         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2934         REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
2935         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2936 #endif
2937 #ifdef CONFIG_SECURITY
2938         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2939 #endif
2940 #ifdef CONFIG_KALLSYMS
2941         ONE("wchan",      S_IRUGO, proc_pid_wchan),
2942 #endif
2943 #ifdef CONFIG_STACKTRACE
2944         ONE("stack",      S_IRUSR, proc_pid_stack),
2945 #endif
2946 #ifdef CONFIG_SCHED_INFO
2947         ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
2948 #endif
2949 #ifdef CONFIG_LATENCYTOP
2950         REG("latency",  S_IRUGO, proc_lstats_operations),
2951 #endif
2952 #ifdef CONFIG_PROC_PID_CPUSET
2953         ONE("cpuset",     S_IRUGO, proc_cpuset_show),
2954 #endif
2955 #ifdef CONFIG_CGROUPS
2956         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
2957 #endif
2958         ONE("oom_score",  S_IRUGO, proc_oom_score),
2959         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2960         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2961 #ifdef CONFIG_AUDITSYSCALL
2962         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2963         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2964 #endif
2965 #ifdef CONFIG_FAULT_INJECTION
2966         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2967         REG("fail-nth", 0644, proc_fail_nth_operations),
2968 #endif
2969 #ifdef CONFIG_ELF_CORE
2970         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2971 #endif
2972 #ifdef CONFIG_TASK_IO_ACCOUNTING
2973         ONE("io",       S_IRUSR, proc_tgid_io_accounting),
2974 #endif
2975 #ifdef CONFIG_HARDWALL
2976         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
2977 #endif
2978 #ifdef CONFIG_USER_NS
2979         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2980         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2981         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2982         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
2983 #endif
2984 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2985         REG("timers",     S_IRUGO, proc_timers_operations),
2986 #endif
2987         REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
2988 #ifdef CONFIG_LIVEPATCH
2989         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
2990 #endif
2991 };
2992
2993 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2994 {
2995         return proc_pident_readdir(file, ctx,
2996                                    tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2997 }
2998
2999 static const struct file_operations proc_tgid_base_operations = {
3000         .read           = generic_read_dir,
3001         .iterate_shared = proc_tgid_base_readdir,
3002         .llseek         = generic_file_llseek,
3003 };
3004
3005 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3006 {
3007         return proc_pident_lookup(dir, dentry,
3008                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3009 }
3010
3011 static const struct inode_operations proc_tgid_base_inode_operations = {
3012         .lookup         = proc_tgid_base_lookup,
3013         .getattr        = pid_getattr,
3014         .setattr        = proc_setattr,
3015         .permission     = proc_pid_permission,
3016 };
3017
3018 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
3019 {
3020         struct dentry *dentry, *leader, *dir;
3021         char buf[PROC_NUMBUF];
3022         struct qstr name;
3023
3024         name.name = buf;
3025         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3026         /* no ->d_hash() rejects on procfs */
3027         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
3028         if (dentry) {
3029                 d_invalidate(dentry);
3030                 dput(dentry);
3031         }
3032
3033         if (pid == tgid)
3034                 return;
3035
3036         name.name = buf;
3037         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
3038         leader = d_hash_and_lookup(mnt->mnt_root, &name);
3039         if (!leader)
3040                 goto out;
3041
3042         name.name = "task";
3043         name.len = strlen(name.name);
3044         dir = d_hash_and_lookup(leader, &name);
3045         if (!dir)
3046                 goto out_put_leader;
3047
3048         name.name = buf;
3049         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3050         dentry = d_hash_and_lookup(dir, &name);
3051         if (dentry) {
3052                 d_invalidate(dentry);
3053                 dput(dentry);
3054         }
3055
3056         dput(dir);
3057 out_put_leader:
3058         dput(leader);
3059 out:
3060         return;
3061 }
3062
3063 /**
3064  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
3065  * @task: task that should be flushed.
3066  *
3067  * When flushing dentries from proc, one needs to flush them from global
3068  * proc (proc_mnt) and from all the namespaces' procs this task was seen
3069  * in. This call is supposed to do all of this job.
3070  *
3071  * Looks in the dcache for
3072  * /proc/@pid
3073  * /proc/@tgid/task/@pid
3074  * if either directory is present flushes it and all of it'ts children
3075  * from the dcache.
3076  *
3077  * It is safe and reasonable to cache /proc entries for a task until
3078  * that task exits.  After that they just clog up the dcache with
3079  * useless entries, possibly causing useful dcache entries to be
3080  * flushed instead.  This routine is proved to flush those useless
3081  * dcache entries at process exit time.
3082  *
3083  * NOTE: This routine is just an optimization so it does not guarantee
3084  *       that no dcache entries will exist at process exit time it
3085  *       just makes it very unlikely that any will persist.
3086  */
3087
3088 void proc_flush_task(struct task_struct *task)
3089 {
3090         int i;
3091         struct pid *pid, *tgid;
3092         struct upid *upid;
3093
3094         pid = task_pid(task);
3095         tgid = task_tgid(task);
3096
3097         for (i = 0; i <= pid->level; i++) {
3098                 upid = &pid->numbers[i];
3099                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3100                                         tgid->numbers[i].nr);
3101         }
3102 }
3103
3104 static int proc_pid_instantiate(struct inode *dir,
3105                                    struct dentry * dentry,
3106                                    struct task_struct *task, const void *ptr)
3107 {
3108         struct inode *inode;
3109
3110         inode = proc_pid_make_inode(dir->i_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3111         if (!inode)
3112                 goto out;
3113
3114         inode->i_op = &proc_tgid_base_inode_operations;
3115         inode->i_fop = &proc_tgid_base_operations;
3116         inode->i_flags|=S_IMMUTABLE;
3117
3118         set_nlink(inode, nlink_tgid);
3119
3120         d_set_d_op(dentry, &pid_dentry_operations);
3121
3122         d_add(dentry, inode);
3123         /* Close the race of the process dying before we return the dentry */
3124         if (pid_revalidate(dentry, 0))
3125                 return 0;
3126 out:
3127         return -ENOENT;
3128 }
3129
3130 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3131 {
3132         int result = -ENOENT;
3133         struct task_struct *task;
3134         unsigned tgid;
3135         struct pid_namespace *ns;
3136
3137         tgid = name_to_int(&dentry->d_name);
3138         if (tgid == ~0U)
3139                 goto out;
3140
3141         ns = dentry->d_sb->s_fs_info;
3142         rcu_read_lock();
3143         task = find_task_by_pid_ns(tgid, ns);
3144         if (task)
3145                 get_task_struct(task);
3146         rcu_read_unlock();
3147         if (!task)
3148                 goto out;
3149
3150         result = proc_pid_instantiate(dir, dentry, task, NULL);
3151         put_task_struct(task);
3152 out:
3153         return ERR_PTR(result);
3154 }
3155
3156 /*
3157  * Find the first task with tgid >= tgid
3158  *
3159  */
3160 struct tgid_iter {
3161         unsigned int tgid;
3162         struct task_struct *task;
3163 };
3164 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3165 {
3166         struct pid *pid;
3167
3168         if (iter.task)
3169                 put_task_struct(iter.task);
3170         rcu_read_lock();
3171 retry:
3172         iter.task = NULL;
3173         pid = find_ge_pid(iter.tgid, ns);
3174         if (pid) {
3175                 iter.tgid = pid_nr_ns(pid, ns);
3176                 iter.task = pid_task(pid, PIDTYPE_PID);
3177                 /* What we to know is if the pid we have find is the
3178                  * pid of a thread_group_leader.  Testing for task
3179                  * being a thread_group_leader is the obvious thing
3180                  * todo but there is a window when it fails, due to
3181                  * the pid transfer logic in de_thread.
3182                  *
3183                  * So we perform the straight forward test of seeing
3184                  * if the pid we have found is the pid of a thread
3185                  * group leader, and don't worry if the task we have
3186                  * found doesn't happen to be a thread group leader.
3187                  * As we don't care in the case of readdir.
3188                  */
3189                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3190                         iter.tgid += 1;
3191                         goto retry;
3192                 }
3193                 get_task_struct(iter.task);
3194         }
3195         rcu_read_unlock();
3196         return iter;
3197 }
3198
3199 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3200
3201 /* for the /proc/ directory itself, after non-process stuff has been done */
3202 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3203 {
3204         struct tgid_iter iter;
3205         struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
3206         loff_t pos = ctx->pos;
3207
3208         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3209                 return 0;
3210
3211         if (pos == TGID_OFFSET - 2) {
3212                 struct inode *inode = d_inode(ns->proc_self);
3213                 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3214                         return 0;
3215                 ctx->pos = pos = pos + 1;
3216         }
3217         if (pos == TGID_OFFSET - 1) {
3218                 struct inode *inode = d_inode(ns->proc_thread_self);
3219                 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3220                         return 0;
3221                 ctx->pos = pos = pos + 1;
3222         }
3223         iter.tgid = pos - TGID_OFFSET;
3224         iter.task = NULL;
3225         for (iter = next_tgid(ns, iter);
3226              iter.task;
3227              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3228                 char name[PROC_NUMBUF];
3229                 int len;
3230
3231                 cond_resched();
3232                 if (!has_pid_permissions(ns, iter.task, HIDEPID_INVISIBLE))
3233                         continue;
3234
3235                 len = snprintf(name, sizeof(name), "%d", iter.tgid);
3236                 ctx->pos = iter.tgid + TGID_OFFSET;
3237                 if (!proc_fill_cache(file, ctx, name, len,
3238                                      proc_pid_instantiate, iter.task, NULL)) {
3239                         put_task_struct(iter.task);
3240                         return 0;
3241                 }
3242         }
3243         ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3244         return 0;
3245 }
3246
3247 /*
3248  * proc_tid_comm_permission is a special permission function exclusively
3249  * used for the node /proc/<pid>/task/<tid>/comm.
3250  * It bypasses generic permission checks in the case where a task of the same
3251  * task group attempts to access the node.
3252  * The rationale behind this is that glibc and bionic access this node for
3253  * cross thread naming (pthread_set/getname_np(!self)). However, if
3254  * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3255  * which locks out the cross thread naming implementation.
3256  * This function makes sure that the node is always accessible for members of
3257  * same thread group.
3258  */
3259 static int proc_tid_comm_permission(struct inode *inode, int mask)
3260 {
3261         bool is_same_tgroup;
3262         struct task_struct *task;
3263
3264         task = get_proc_task(inode);
3265         if (!task)
3266                 return -ESRCH;
3267         is_same_tgroup = same_thread_group(current, task);
3268         put_task_struct(task);
3269
3270         if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3271                 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3272                  * read or written by the members of the corresponding
3273                  * thread group.
3274                  */
3275                 return 0;
3276         }
3277
3278         return generic_permission(inode, mask);
3279 }
3280
3281 static const struct inode_operations proc_tid_comm_inode_operations = {
3282                 .permission = proc_tid_comm_permission,
3283 };
3284
3285 /*
3286  * Tasks
3287  */
3288 static const struct pid_entry tid_base_stuff[] = {
3289         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3290         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3291         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3292 #ifdef CONFIG_NET
3293         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3294 #endif
3295         REG("environ",   S_IRUSR, proc_environ_operations),
3296         REG("auxv",      S_IRUSR, proc_auxv_operations),
3297         ONE("status",    S_IRUGO, proc_pid_status),
3298         ONE("personality", S_IRUSR, proc_pid_personality),
3299         ONE("limits",    S_IRUGO, proc_pid_limits),
3300 #ifdef CONFIG_SCHED_DEBUG
3301         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3302 #endif
3303         NOD("comm",      S_IFREG|S_IRUGO|S_IWUSR,
3304                          &proc_tid_comm_inode_operations,
3305                          &proc_pid_set_comm_operations, {}),
3306 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3307         ONE("syscall",   S_IRUSR, proc_pid_syscall),
3308 #endif
3309         REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3310         ONE("stat",      S_IRUGO, proc_tid_stat),
3311         ONE("statm",     S_IRUGO, proc_pid_statm),
3312         REG("maps",      S_IRUGO, proc_tid_maps_operations),
3313 #ifdef CONFIG_PROC_CHILDREN
3314         REG("children",  S_IRUGO, proc_tid_children_operations),
3315 #endif
3316 #ifdef CONFIG_NUMA
3317         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3318 #endif
3319         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3320         LNK("cwd",       proc_cwd_link),
3321         LNK("root",      proc_root_link),
3322         LNK("exe",       proc_exe_link),
3323         REG("mounts",    S_IRUGO, proc_mounts_operations),
3324         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3325 #ifdef CONFIG_PROC_PAGE_MONITOR
3326         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3327         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3328         REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3329         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3330 #endif
3331 #ifdef CONFIG_SECURITY
3332         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3333 #endif
3334 #ifdef CONFIG_KALLSYMS
3335         ONE("wchan",     S_IRUGO, proc_pid_wchan),
3336 #endif
3337 #ifdef CONFIG_STACKTRACE
3338         ONE("stack",      S_IRUSR, proc_pid_stack),
3339 #endif
3340 #ifdef CONFIG_SCHED_INFO
3341         ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3342 #endif
3343 #ifdef CONFIG_LATENCYTOP
3344         REG("latency",  S_IRUGO, proc_lstats_operations),
3345 #endif
3346 #ifdef CONFIG_PROC_PID_CPUSET
3347         ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3348 #endif
3349 #ifdef CONFIG_CGROUPS
3350         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3351 #endif
3352         ONE("oom_score", S_IRUGO, proc_oom_score),
3353         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3354         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3355 #ifdef CONFIG_AUDITSYSCALL
3356         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3357         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3358 #endif
3359 #ifdef CONFIG_FAULT_INJECTION
3360         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3361         REG("fail-nth", 0644, proc_fail_nth_operations),
3362 #endif
3363 #ifdef CONFIG_TASK_IO_ACCOUNTING
3364         ONE("io",       S_IRUSR, proc_tid_io_accounting),
3365 #endif
3366 #ifdef CONFIG_HARDWALL
3367         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
3368 #endif
3369 #ifdef CONFIG_USER_NS
3370         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3371         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3372         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3373         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3374 #endif
3375 #ifdef CONFIG_LIVEPATCH
3376         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
3377 #endif
3378 };
3379
3380 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3381 {
3382         return proc_pident_readdir(file, ctx,
3383                                    tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3384 }
3385
3386 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3387 {
3388         return proc_pident_lookup(dir, dentry,
3389                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3390 }
3391
3392 static const struct file_operations proc_tid_base_operations = {
3393         .read           = generic_read_dir,
3394         .iterate_shared = proc_tid_base_readdir,
3395         .llseek         = generic_file_llseek,
3396 };
3397
3398 static const struct inode_operations proc_tid_base_inode_operations = {
3399         .lookup         = proc_tid_base_lookup,
3400         .getattr        = pid_getattr,
3401         .setattr        = proc_setattr,
3402 };
3403
3404 static int proc_task_instantiate(struct inode *dir,
3405         struct dentry *dentry, struct task_struct *task, const void *ptr)
3406 {
3407         struct inode *inode;
3408         inode = proc_pid_make_inode(dir->i_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3409
3410         if (!inode)
3411                 goto out;
3412         inode->i_op = &proc_tid_base_inode_operations;
3413         inode->i_fop = &proc_tid_base_operations;
3414         inode->i_flags|=S_IMMUTABLE;
3415
3416         set_nlink(inode, nlink_tid);
3417
3418         d_set_d_op(dentry, &pid_dentry_operations);
3419
3420         d_add(dentry, inode);
3421         /* Close the race of the process dying before we return the dentry */
3422         if (pid_revalidate(dentry, 0))
3423                 return 0;
3424 out:
3425         return -ENOENT;
3426 }
3427
3428 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3429 {
3430         int result = -ENOENT;
3431         struct task_struct *task;
3432         struct task_struct *leader = get_proc_task(dir);
3433         unsigned tid;
3434         struct pid_namespace *ns;
3435
3436         if (!leader)
3437                 goto out_no_task;
3438
3439         tid = name_to_int(&dentry->d_name);
3440         if (tid == ~0U)
3441                 goto out;
3442
3443         ns = dentry->d_sb->s_fs_info;
3444         rcu_read_lock();
3445         task = find_task_by_pid_ns(tid, ns);
3446         if (task)
3447                 get_task_struct(task);
3448         rcu_read_unlock();
3449         if (!task)
3450                 goto out;
3451         if (!same_thread_group(leader, task))
3452                 goto out_drop_task;
3453
3454         result = proc_task_instantiate(dir, dentry, task, NULL);
3455 out_drop_task:
3456         put_task_struct(task);
3457 out:
3458         put_task_struct(leader);
3459 out_no_task:
3460         return ERR_PTR(result);
3461 }
3462
3463 /*
3464  * Find the first tid of a thread group to return to user space.
3465  *
3466  * Usually this is just the thread group leader, but if the users
3467  * buffer was too small or there was a seek into the middle of the
3468  * directory we have more work todo.
3469  *
3470  * In the case of a short read we start with find_task_by_pid.
3471  *
3472  * In the case of a seek we start with the leader and walk nr
3473  * threads past it.
3474  */
3475 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3476                                         struct pid_namespace *ns)
3477 {
3478         struct task_struct *pos, *task;
3479         unsigned long nr = f_pos;
3480
3481         if (nr != f_pos)        /* 32bit overflow? */
3482                 return NULL;
3483
3484         rcu_read_lock();
3485         task = pid_task(pid, PIDTYPE_PID);
3486         if (!task)
3487                 goto fail;
3488
3489         /* Attempt to start with the tid of a thread */
3490         if (tid && nr) {
3491                 pos = find_task_by_pid_ns(tid, ns);
3492                 if (pos && same_thread_group(pos, task))
3493                         goto found;
3494         }
3495
3496         /* If nr exceeds the number of threads there is nothing todo */
3497         if (nr >= get_nr_threads(task))
3498                 goto fail;
3499
3500         /* If we haven't found our starting place yet start
3501          * with the leader and walk nr threads forward.
3502          */
3503         pos = task = task->group_leader;
3504         do {
3505                 if (!nr--)
3506                         goto found;
3507         } while_each_thread(task, pos);
3508 fail:
3509         pos = NULL;
3510         goto out;
3511 found:
3512         get_task_struct(pos);
3513 out:
3514         rcu_read_unlock();
3515         return pos;
3516 }
3517
3518 /*
3519  * Find the next thread in the thread list.
3520  * Return NULL if there is an error or no next thread.
3521  *
3522  * The reference to the input task_struct is released.
3523  */
3524 static struct task_struct *next_tid(struct task_struct *start)
3525 {
3526         struct task_struct *pos = NULL;
3527         rcu_read_lock();
3528         if (pid_alive(start)) {
3529                 pos = next_thread(start);
3530                 if (thread_group_leader(pos))
3531                         pos = NULL;
3532                 else
3533                         get_task_struct(pos);
3534         }
3535         rcu_read_unlock();
3536         put_task_struct(start);
3537         return pos;
3538 }
3539
3540 /* for the /proc/TGID/task/ directories */
3541 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3542 {
3543         struct inode *inode = file_inode(file);
3544         struct task_struct *task;
3545         struct pid_namespace *ns;
3546         int tid;
3547
3548         if (proc_inode_is_dead(inode))
3549                 return -ENOENT;
3550
3551         if (!dir_emit_dots(file, ctx))
3552                 return 0;
3553
3554         /* f_version caches the tgid value that the last readdir call couldn't
3555          * return. lseek aka telldir automagically resets f_version to 0.
3556          */
3557         ns = inode->i_sb->s_fs_info;
3558         tid = (int)file->f_version;
3559         file->f_version = 0;
3560         for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3561              task;
3562              task = next_tid(task), ctx->pos++) {
3563                 char name[PROC_NUMBUF];
3564                 int len;
3565                 tid = task_pid_nr_ns(task, ns);
3566                 len = snprintf(name, sizeof(name), "%d", tid);
3567                 if (!proc_fill_cache(file, ctx, name, len,
3568                                 proc_task_instantiate, task, NULL)) {
3569                         /* returning this tgid failed, save it as the first
3570                          * pid for the next readir call */
3571                         file->f_version = (u64)tid;
3572                         put_task_struct(task);
3573                         break;
3574                 }
3575         }
3576
3577         return 0;
3578 }
3579
3580 static int proc_task_getattr(const struct path *path, struct kstat *stat,
3581                              u32 request_mask, unsigned int query_flags)
3582 {
3583         struct inode *inode = d_inode(path->dentry);
3584         struct task_struct *p = get_proc_task(inode);
3585         generic_fillattr(inode, stat);
3586
3587         if (p) {
3588                 stat->nlink += get_nr_threads(p);
3589                 put_task_struct(p);
3590         }
3591
3592         return 0;
3593 }
3594
3595 static const struct inode_operations proc_task_inode_operations = {
3596         .lookup         = proc_task_lookup,
3597         .getattr        = proc_task_getattr,
3598         .setattr        = proc_setattr,
3599         .permission     = proc_pid_permission,
3600 };
3601
3602 static const struct file_operations proc_task_operations = {
3603         .read           = generic_read_dir,
3604         .iterate_shared = proc_task_readdir,
3605         .llseek         = generic_file_llseek,
3606 };
3607
3608 void __init set_proc_pid_nlink(void)
3609 {
3610         nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3611         nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3612 }