fs/proc/base.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/proc/base.c
   4  *
   5  *  Copyright (C) 1991, 1992 Linus Torvalds
   6  *
   7  *  proc base directory handling functions
   8  *
   9  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
  10  *  Instead of using magical inumbers to determine the kind of object
  11  *  we allocate and fill in-core inodes upon lookup. They don't even
  12  *  go into icache. We cache the reference to task_struct upon lookup too.
  13  *  Eventually it should become a filesystem in its own. We don't use the
  14  *  rest of procfs anymore.
  15  *
  16  *
  17  *  Changelog:
  18  *  17-Jan-2005
  19  *  Allan Bezerra
  20  *  Bruna Moreira <[email protected]>
  21  *  Edjard Mota <[email protected]>
  22  *  Ilias Biris <[email protected]>
  23  *  Mauricio Lin <[email protected]>
  24  *
  25  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  26  *
  27  *  A new process specific entry (smaps) included in /proc. It shows the
  28  *  size of rss for each memory area. The maps entry lacks information
  29  *  about physical memory size (rss) for each mapped file, i.e.,
  30  *  rss information for executables and library files.
  31  *  This additional information is useful for any tools that need to know
  32  *  about physical memory consumption for a process specific library.
  33  *
  34  *  Changelog:
  35  *  21-Feb-2005
  36  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  37  *  Pud inclusion in the page table walking.
  38  *
  39  *  ChangeLog:
  40  *  10-Mar-2005
  41  *  10LE Instituto Nokia de Tecnologia - INdT:
  42  *  A better way to walks through the page table as suggested by Hugh Dickins.
  43  *
  44  *  Simo Piiroinen <[email protected]>:
  45  *  Smaps information related to shared, private, clean and dirty pages.
  46  *
  47  *  Paul Mundt <[email protected]>:
  48  *  Overall revision about smaps.
  49  */
  50
  51 #include <linux/uaccess.h>
  52
  53 #include <linux/errno.h>
  54 #include <linux/time.h>
  55 #include <linux/proc_fs.h>
  56 #include <linux/stat.h>
  57 #include <linux/task_io_accounting_ops.h>
  58 #include <linux/init.h>
  59 #include <linux/capability.h>
  60 #include <linux/file.h>
  61 #include <linux/fdtable.h>
  62 #include <linux/string.h>
  63 #include <linux/seq_file.h>
  64 #include <linux/namei.h>
  65 #include <linux/mnt_namespace.h>
  66 #include <linux/mm.h>
  67 #include <linux/swap.h>
  68 #include <linux/rcupdate.h>
  69 #include <linux/kallsyms.h>
  70 #include <linux/stacktrace.h>
  71 #include <linux/resource.h>
  72 #include <linux/module.h>
  73 #include <linux/mount.h>
  74 #include <linux/security.h>
  75 #include <linux/ptrace.h>
  76 #include <linux/tracehook.h>
  77 #include <linux/printk.h>
  78 #include <linux/cache.h>
  79 #include <linux/cgroup.h>
  80 #include <linux/cpuset.h>
  81 #include <linux/audit.h>
  82 #include <linux/poll.h>
  83 #include <linux/nsproxy.h>
  84 #include <linux/oom.h>
  85 #include <linux/elf.h>
  86 #include <linux/pid_namespace.h>
  87 #include <linux/user_namespace.h>
  88 #include <linux/fs_struct.h>
  89 #include <linux/slab.h>
  90 #include <linux/sched/autogroup.h>
  91 #include <linux/sched/mm.h>
  92 #include <linux/sched/coredump.h>
  93 #include <linux/sched/debug.h>
  94 #include <linux/sched/stat.h>
  95 #include <linux/flex_array.h>
  96 #include <linux/posix-timers.h>
  97 #include <trace/events/oom.h>
  98 #include "internal.h"
  99 #include "fd.h"
 100
 101 #include "../../lib/kstrtox.h"
 102
 103 /* NOTE:
 104  *      Implementing inode permission operations in /proc is almost
 105  *      certainly an error.  Permission checks need to happen during
 106  *      each system call not at open time.  The reason is that most of
 107  *      what we wish to check for permissions in /proc varies at runtime.
 108  *
 109  *      The classic example of a problem is opening file descriptors
 110  *      in /proc for a task before it execs a suid executable.
 111  */
 112
 113 static u8 nlink_tid __ro_after_init;
 114 static u8 nlink_tgid __ro_after_init;
 115
 116 struct pid_entry {
 117         const char *name;
 118         unsigned int len;
 119         umode_t mode;
 120         const struct inode_operations *iop;
 121         const struct file_operations *fop;
 122         union proc_op op;
 123 };
 124
 125 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 126         .name = (NAME),                                 \
 127         .len  = sizeof(NAME) - 1,                       \
 128         .mode = MODE,                                   \
 129         .iop  = IOP,                                    \
 130         .fop  = FOP,                                    \
 131         .op   = OP,                                     \
 132 }
 133
 134 #define DIR(NAME, MODE, iops, fops)     \
 135         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 136 #define LNK(NAME, get_link)                                     \
 137         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 138                 &proc_pid_link_inode_operations, NULL,          \
 139                 { .proc_get_link = get_link } )
 140 #define REG(NAME, MODE, fops)                           \
 141         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 142 #define ONE(NAME, MODE, show)                           \
 143         NOD(NAME, (S_IFREG|(MODE)),                     \
 144                 NULL, &proc_single_file_operations,     \
 145                 { .proc_show = show } )
 146
 147 /*
 148  * Count the number of hardlinks for the pid_entry table, excluding the .
 149  * and .. links.
 150  */
 151 static unsigned int __init pid_entry_nlink(const struct pid_entry *entries,
 152         unsigned int n)
 153 {
 154         unsigned int i;
 155         unsigned int count;
 156
 157         count = 2;
 158         for (i = 0; i < n; ++i) {
 159                 if (S_ISDIR(entries[i].mode))
 160                         ++count;
 161         }
 162
 163         return count;
 164 }
 165
 166 static int get_task_root(struct task_struct *task, struct path *root)
 167 {
 168         int result = -ENOENT;
 169
 170         task_lock(task);
 171         if (task->fs) {
 172                 get_fs_root(task->fs, root);
 173                 result = 0;
 174         }
 175         task_unlock(task);
 176         return result;
 177 }
 178
 179 static int proc_cwd_link(struct dentry *dentry, struct path *path)
 180 {
 181         struct task_struct *task = get_proc_task(d_inode(dentry));
 182         int result = -ENOENT;
 183
 184         if (task) {
 185                 task_lock(task);
 186                 if (task->fs) {
 187                         get_fs_pwd(task->fs, path);
 188                         result = 0;
 189                 }
 190                 task_unlock(task);
 191                 put_task_struct(task);
 192         }
 193         return result;
 194 }
 195
 196 static int proc_root_link(struct dentry *dentry, struct path *path)
 197 {
 198         struct task_struct *task = get_proc_task(d_inode(dentry));
 199         int result = -ENOENT;
 200
 201         if (task) {
 202                 result = get_task_root(task, path);
 203                 put_task_struct(task);
 204         }
 205         return result;
 206 }
 207
 208 static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
 209                                      size_t _count, loff_t *pos)
 210 {
 211         struct task_struct *tsk;
 212         struct mm_struct *mm;
 213         char *page;
 214         unsigned long count = _count;
 215         unsigned long arg_start, arg_end, env_start, env_end;
 216         unsigned long len1, len2, len;
 217         unsigned long p;
 218         char c;
 219         ssize_t rv;
 220
 221         BUG_ON(*pos < 0);
 222
 223         tsk = get_proc_task(file_inode(file));
 224         if (!tsk)
 225                 return -ESRCH;
 226         mm = get_task_mm(tsk);
 227         put_task_struct(tsk);
 228         if (!mm)
 229                 return 0;
 230         /* Check if process spawned far enough to have cmdline. */
 231         if (!mm->env_end) {
 232                 rv = 0;
 233                 goto out_mmput;
 234         }
 235
 236         page = (char *)__get_free_page(GFP_KERNEL);
 237         if (!page) {
 238                 rv = -ENOMEM;
 239                 goto out_mmput;
 240         }
 241
 242         down_read(&mm->mmap_sem);
 243         arg_start = mm->arg_start;
 244         arg_end = mm->arg_end;
 245         env_start = mm->env_start;
 246         env_end = mm->env_end;
 247         up_read(&mm->mmap_sem);
 248
 249         BUG_ON(arg_start > arg_end);
 250         BUG_ON(env_start > env_end);
 251
 252         len1 = arg_end - arg_start;
 253         len2 = env_end - env_start;
 254
 255         /* Empty ARGV. */
 256         if (len1 == 0) {
 257                 rv = 0;
 258                 goto out_free_page;
 259         }
 260         /*
 261          * Inherently racy -- command line shares address space
 262          * with code and data.
 263          */
 264         rv = access_remote_vm(mm, arg_end - 1, &c, 1, FOLL_ANON);
 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, FOLL_ANON);
 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, FOLL_ANON);
 329                                 if (nr_read < 0)
 330                                         rv = nr_read;
 331                                 if (nr_read <= 0)
 332                                         goto out_free_page;
 333
 334                                 /*
 335                                  * Command line can be shorter than whole ARGV
 336                                  * even if last "marker" byte says it is not.
 337                                  */
 338                                 final = false;
 339                                 l = strnlen(page, nr_read);
 340                                 if (l < nr_read) {
 341                                         nr_read = l;
 342                                         final = true;
 343                                 }
 344
 345                                 if (copy_to_user(buf, page, nr_read)) {
 346                                         rv = -EFAULT;
 347                                         goto out_free_page;
 348                                 }
 349
 350                                 p       += nr_read;
 351                                 len     -= nr_read;
 352                                 buf     += nr_read;
 353                                 count   -= nr_read;
 354                                 rv      += nr_read;
 355
 356                                 if (final)
 357                                         goto out_free_page;
 358                         }
 359
 360                         /* Only first chunk can be read partially. */
 361                         pos1 = 0;
 362                         i++;
 363                 }
 364         }
 365
 366 out_free_page:
 367         free_page((unsigned long)page);
 368 out_mmput:
 369         mmput(mm);
 370         if (rv > 0)
 371                 *pos += rv;
 372         return rv;
 373 }
 374
 375 static const struct file_operations proc_pid_cmdline_ops = {
 376         .read   = proc_pid_cmdline_read,
 377         .llseek = generic_file_llseek,
 378 };
 379
 380 #ifdef CONFIG_KALLSYMS
 381 /*
 382  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 383  * Returns the resolved symbol.  If that fails, simply return the address.
 384  */
 385 static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
 386                           struct pid *pid, struct task_struct *task)
 387 {
 388         unsigned long wchan;
 389         char symname[KSYM_NAME_LEN];
 390
 391         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
 392                 goto print0;
 393
 394         wchan = get_wchan(task);
 395         if (wchan && !lookup_symbol_name(wchan, symname)) {
 396                 seq_puts(m, symname);
 397                 return 0;
 398         }
 399
 400 print0:
 401         seq_putc(m, '0');
 402         return 0;
 403 }
 404 #endif /* CONFIG_KALLSYMS */
 405
 406 static int lock_trace(struct task_struct *task)
 407 {
 408         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
 409         if (err)
 410                 return err;
 411         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
 412                 mutex_unlock(&task->signal->cred_guard_mutex);
 413                 return -EPERM;
 414         }
 415         return 0;
 416 }
 417
 418 static void unlock_trace(struct task_struct *task)
 419 {
 420         mutex_unlock(&task->signal->cred_guard_mutex);
 421 }
 422
 423 #ifdef CONFIG_STACKTRACE
 424
 425 #define MAX_STACK_TRACE_DEPTH   64
 426
 427 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 428                           struct pid *pid, struct task_struct *task)
 429 {
 430         struct stack_trace trace;
 431         unsigned long *entries;
 432         int err;
 433         int i;
 434
 435         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 436         if (!entries)
 437                 return -ENOMEM;
 438
 439         trace.nr_entries        = 0;
 440         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 441         trace.entries           = entries;
 442         trace.skip              = 0;
 443
 444         err = lock_trace(task);
 445         if (!err) {
 446                 save_stack_trace_tsk(task, &trace);
 447
 448                 for (i = 0; i < trace.nr_entries; i++) {
 449                         seq_printf(m, "[<0>] %pB\n", (void *)entries[i]);
 450                 }
 451                 unlock_trace(task);
 452         }
 453         kfree(entries);
 454
 455         return err;
 456 }
 457 #endif
 458
 459 #ifdef CONFIG_SCHED_INFO
 460 /*
 461  * Provides /proc/PID/schedstat
 462  */
 463 static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
 464                               struct pid *pid, struct task_struct *task)
 465 {
 466         if (unlikely(!sched_info_on()))
 467                 seq_printf(m, "0 0 0\n");
 468         else
 469                 seq_printf(m, "%llu %llu %lu\n",
 470                    (unsigned long long)task->se.sum_exec_runtime,
 471                    (unsigned long long)task->sched_info.run_delay,
 472                    task->sched_info.pcount);
 473
 474         return 0;
 475 }
 476 #endif
 477
 478 #ifdef CONFIG_LATENCYTOP
 479 static int lstats_show_proc(struct seq_file *m, void *v)
 480 {
 481         int i;
 482         struct inode *inode = m->private;
 483         struct task_struct *task = get_proc_task(inode);
 484
 485         if (!task)
 486                 return -ESRCH;
 487         seq_puts(m, "Latency Top version : v0.1\n");
 488         for (i = 0; i < 32; i++) {
 489                 struct latency_record *lr = &task->latency_record[i];
 490                 if (lr->backtrace[0]) {
 491                         int q;
 492                         seq_printf(m, "%i %li %li",
 493                                    lr->count, lr->time, lr->max);
 494                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 495                                 unsigned long bt = lr->backtrace[q];
 496                                 if (!bt)
 497                                         break;
 498                                 if (bt == ULONG_MAX)
 499                                         break;
 500                                 seq_printf(m, " %ps", (void *)bt);
 501                         }
 502                         seq_putc(m, '\n');
 503                 }
 504
 505         }
 506         put_task_struct(task);
 507         return 0;
 508 }
 509
 510 static int lstats_open(struct inode *inode, struct file *file)
 511 {
 512         return single_open(file, lstats_show_proc, inode);
 513 }
 514
 515 static ssize_t lstats_write(struct file *file, const char __user *buf,
 516                             size_t count, loff_t *offs)
 517 {
 518         struct task_struct *task = get_proc_task(file_inode(file));
 519
 520         if (!task)
 521                 return -ESRCH;
 522         clear_all_latency_tracing(task);
 523         put_task_struct(task);
 524
 525         return count;
 526 }
 527
 528 static const struct file_operations proc_lstats_operations = {
 529         .open           = lstats_open,
 530         .read           = seq_read,
 531         .write          = lstats_write,
 532         .llseek         = seq_lseek,
 533         .release        = single_release,
 534 };
 535
 536 #endif
 537
 538 static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
 539                           struct pid *pid, struct task_struct *task)
 540 {
 541         unsigned long totalpages = totalram_pages + total_swap_pages;
 542         unsigned long points = 0;
 543
 544         points = oom_badness(task, NULL, NULL, totalpages) *
 545                                         1000 / totalpages;
 546         seq_printf(m, "%lu\n", points);
 547
 548         return 0;
 549 }
 550
 551 struct limit_names {
 552         const char *name;
 553         const char *unit;
 554 };
 555
 556 static const struct limit_names lnames[RLIM_NLIMITS] = {
 557         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 558         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 559         [RLIMIT_DATA] = {"Max data size", "bytes"},
 560         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 561         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 562         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 563         [RLIMIT_NPROC] = {"Max processes", "processes"},
 564         [RLIMIT_NOFILE] = {"Max open files", "files"},
 565         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 566         [RLIMIT_AS] = {"Max address space", "bytes"},
 567         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 568         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 569         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 570         [RLIMIT_NICE] = {"Max nice priority", NULL},
 571         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 572         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 573 };
 574
 575 /* Display limits for a process */
 576 static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
 577                            struct pid *pid, struct task_struct *task)
 578 {
 579         unsigned int i;
 580         unsigned long flags;
 581
 582         struct rlimit rlim[RLIM_NLIMITS];
 583
 584         if (!lock_task_sighand(task, &flags))
 585                 return 0;
 586         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 587         unlock_task_sighand(task, &flags);
 588
 589         /*
 590          * print the file header
 591          */
 592        seq_printf(m, "%-25s %-20s %-20s %-10s\n",
 593                   "Limit", "Soft Limit", "Hard Limit", "Units");
 594
 595         for (i = 0; i < RLIM_NLIMITS; i++) {
 596                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 597                         seq_printf(m, "%-25s %-20s ",
 598                                    lnames[i].name, "unlimited");
 599                 else
 600                         seq_printf(m, "%-25s %-20lu ",
 601                                    lnames[i].name, rlim[i].rlim_cur);
 602
 603                 if (rlim[i].rlim_max == RLIM_INFINITY)
 604                         seq_printf(m, "%-20s ", "unlimited");
 605                 else
 606                         seq_printf(m, "%-20lu ", rlim[i].rlim_max);
 607
 608                 if (lnames[i].unit)
 609                         seq_printf(m, "%-10s\n", lnames[i].unit);
 610                 else
 611                         seq_putc(m, '\n');
 612         }
 613
 614         return 0;
 615 }
 616
 617 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 618 static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
 619                             struct pid *pid, struct task_struct *task)
 620 {
 621         long nr;
 622         unsigned long args[6], sp, pc;
 623         int res;
 624
 625         res = lock_trace(task);
 626         if (res)
 627                 return res;
 628
 629         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 630                 seq_puts(m, "running\n");
 631         else if (nr < 0)
 632                 seq_printf(m, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 633         else
 634                 seq_printf(m,
 635                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 636                        nr,
 637                        args[0], args[1], args[2], args[3], args[4], args[5],
 638                        sp, pc);
 639         unlock_trace(task);
 640
 641         return 0;
 642 }
 643 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 644
 645 /************************************************************************/
 646 /*                       Here the fs part begins                        */
 647 /************************************************************************/
 648
 649 /* permission checks */
 650 static int proc_fd_access_allowed(struct inode *inode)
 651 {
 652         struct task_struct *task;
 653         int allowed = 0;
 654         /* Allow access to a task's file descriptors if it is us or we
 655          * may use ptrace attach to the process and find out that
 656          * information.
 657          */
 658         task = get_proc_task(inode);
 659         if (task) {
 660                 allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 661                 put_task_struct(task);
 662         }
 663         return allowed;
 664 }
 665
 666 int proc_setattr(struct dentry *dentry, struct iattr *attr)
 667 {
 668         int error;
 669         struct inode *inode = d_inode(dentry);
 670
 671         if (attr->ia_valid & ATTR_MODE)
 672                 return -EPERM;
 673
 674         error = setattr_prepare(dentry, attr);
 675         if (error)
 676                 return error;
 677
 678         setattr_copy(inode, attr);
 679         mark_inode_dirty(inode);
 680         return 0;
 681 }
 682
 683 /*
 684  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
 685  * or euid/egid (for hide_pid_min=2)?
 686  */
 687 static bool has_pid_permissions(struct pid_namespace *pid,
 688                                  struct task_struct *task,
 689                                  int hide_pid_min)
 690 {
 691         if (pid->hide_pid < hide_pid_min)
 692                 return true;
 693         if (in_group_p(pid->pid_gid))
 694                 return true;
 695         return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 696 }
 697
 698
 699 static int proc_pid_permission(struct inode *inode, int mask)
 700 {
 701         struct pid_namespace *pid = inode->i_sb->s_fs_info;
 702         struct task_struct *task;
 703         bool has_perms;
 704
 705         task = get_proc_task(inode);
 706         if (!task)
 707                 return -ESRCH;
 708         has_perms = has_pid_permissions(pid, task, HIDEPID_NO_ACCESS);
 709         put_task_struct(task);
 710
 711         if (!has_perms) {
 712                 if (pid->hide_pid == HIDEPID_INVISIBLE) {
 713                         /*
 714                          * Let's make getdents(), stat(), and open()
 715                          * consistent with each other.  If a process
 716                          * may not stat() a file, it shouldn't be seen
 717                          * in procfs at all.
 718                          */
 719                         return -ENOENT;
 720                 }
 721
 722                 return -EPERM;
 723         }
 724         return generic_permission(inode, mask);
 725 }
 726
 727
 728
 729 static const struct inode_operations proc_def_inode_operations = {
 730         .setattr        = proc_setattr,
 731 };
 732
 733 static int proc_single_show(struct seq_file *m, void *v)
 734 {
 735         struct inode *inode = m->private;
 736         struct pid_namespace *ns;
 737         struct pid *pid;
 738         struct task_struct *task;
 739         int ret;
 740
 741         ns = inode->i_sb->s_fs_info;
 742         pid = proc_pid(inode);
 743         task = get_pid_task(pid, PIDTYPE_PID);
 744         if (!task)
 745                 return -ESRCH;
 746
 747         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 748
 749         put_task_struct(task);
 750         return ret;
 751 }
 752
 753 static int proc_single_open(struct inode *inode, struct file *filp)
 754 {
 755         return single_open(filp, proc_single_show, inode);
 756 }
 757
 758 static const struct file_operations proc_single_file_operations = {
 759         .open           = proc_single_open,
 760         .read           = seq_read,
 761         .llseek         = seq_lseek,
 762         .release        = single_release,
 763 };
 764
 765
 766 struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
 767 {
 768         struct task_struct *task = get_proc_task(inode);
 769         struct mm_struct *mm = ERR_PTR(-ESRCH);
 770
 771         if (task) {
 772                 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
 773                 put_task_struct(task);
 774
 775                 if (!IS_ERR_OR_NULL(mm)) {
 776                         /* ensure this mm_struct can't be freed */
 777                         mmgrab(mm);
 778                         /* but do not pin its memory */
 779                         mmput(mm);
 780                 }
 781         }
 782
 783         return mm;
 784 }
 785
 786 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
 787 {
 788         struct mm_struct *mm = proc_mem_open(inode, mode);
 789
 790         if (IS_ERR(mm))
 791                 return PTR_ERR(mm);
 792
 793         file->private_data = mm;
 794         return 0;
 795 }
 796
 797 static int mem_open(struct inode *inode, struct file *file)
 798 {
 799         int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
 800
 801         /* OK to pass negative loff_t, we can catch out-of-range */
 802         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 803
 804         return ret;
 805 }
 806
 807 static ssize_t mem_rw(struct file *file, char __user *buf,
 808                         size_t count, loff_t *ppos, int write)
 809 {
 810         struct mm_struct *mm = file->private_data;
 811         unsigned long addr = *ppos;
 812         ssize_t copied;
 813         char *page;
 814         unsigned int flags;
 815
 816         if (!mm)
 817                 return 0;
 818
 819         page = (char *)__get_free_page(GFP_KERNEL);
 820         if (!page)
 821                 return -ENOMEM;
 822
 823         copied = 0;
 824         if (!mmget_not_zero(mm))
 825                 goto free;
 826
 827         flags = FOLL_FORCE | (write ? FOLL_WRITE : 0);
 828
 829         while (count > 0) {
 830                 int this_len = min_t(int, count, PAGE_SIZE);
 831
 832                 if (write && copy_from_user(page, buf, this_len)) {
 833                         copied = -EFAULT;
 834                         break;
 835                 }
 836
 837                 this_len = access_remote_vm(mm, addr, page, this_len, flags);
 838                 if (!this_len) {
 839                         if (!copied)
 840                                 copied = -EIO;
 841                         break;
 842                 }
 843
 844                 if (!write && copy_to_user(buf, page, this_len)) {
 845                         copied = -EFAULT;
 846                         break;
 847                 }
 848
 849                 buf += this_len;
 850                 addr += this_len;
 851                 copied += this_len;
 852                 count -= this_len;
 853         }
 854         *ppos = addr;
 855
 856         mmput(mm);
 857 free:
 858         free_page((unsigned long) page);
 859         return copied;
 860 }
 861
 862 static ssize_t mem_read(struct file *file, char __user *buf,
 863                         size_t count, loff_t *ppos)
 864 {
 865         return mem_rw(file, buf, count, ppos, 0);
 866 }
 867
 868 static ssize_t mem_write(struct file *file, const char __user *buf,
 869                          size_t count, loff_t *ppos)
 870 {
 871         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 872 }
 873
 874 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 875 {
 876         switch (orig) {
 877         case 0:
 878                 file->f_pos = offset;
 879                 break;
 880         case 1:
 881                 file->f_pos += offset;
 882                 break;
 883         default:
 884                 return -EINVAL;
 885         }
 886         force_successful_syscall_return();
 887         return file->f_pos;
 888 }
 889
 890 static int mem_release(struct inode *inode, struct file *file)
 891 {
 892         struct mm_struct *mm = file->private_data;
 893         if (mm)
 894                 mmdrop(mm);
 895         return 0;
 896 }
 897
 898 static const struct file_operations proc_mem_operations = {
 899         .llseek         = mem_lseek,
 900         .read           = mem_read,
 901         .write          = mem_write,
 902         .open           = mem_open,
 903         .release        = mem_release,
 904 };
 905
 906 static int environ_open(struct inode *inode, struct file *file)
 907 {
 908         return __mem_open(inode, file, PTRACE_MODE_READ);
 909 }
 910
 911 static ssize_t environ_read(struct file *file, char __user *buf,
 912                         size_t count, loff_t *ppos)
 913 {
 914         char *page;
 915         unsigned long src = *ppos;
 916         int ret = 0;
 917         struct mm_struct *mm = file->private_data;
 918         unsigned long env_start, env_end;
 919
 920         /* Ensure the process spawned far enough to have an environment. */
 921         if (!mm || !mm->env_end)
 922                 return 0;
 923
 924         page = (char *)__get_free_page(GFP_KERNEL);
 925         if (!page)
 926                 return -ENOMEM;
 927
 928         ret = 0;
 929         if (!mmget_not_zero(mm))
 930                 goto free;
 931
 932         down_read(&mm->mmap_sem);
 933         env_start = mm->env_start;
 934         env_end = mm->env_end;
 935         up_read(&mm->mmap_sem);
 936
 937         while (count > 0) {
 938                 size_t this_len, max_len;
 939                 int retval;
 940
 941                 if (src >= (env_end - env_start))
 942                         break;
 943
 944                 this_len = env_end - (env_start + src);
 945
 946                 max_len = min_t(size_t, PAGE_SIZE, count);
 947                 this_len = min(max_len, this_len);
 948
 949                 retval = access_remote_vm(mm, (env_start + src), page, this_len, FOLL_ANON);
 950
 951                 if (retval <= 0) {
 952                         ret = retval;
 953                         break;
 954                 }
 955
 956                 if (copy_to_user(buf, page, retval)) {
 957                         ret = -EFAULT;
 958                         break;
 959                 }
 960
 961                 ret += retval;
 962                 src += retval;
 963                 buf += retval;
 964                 count -= retval;
 965         }
 966         *ppos = src;
 967         mmput(mm);
 968
 969 free:
 970         free_page((unsigned long) page);
 971         return ret;
 972 }
 973
 974 static const struct file_operations proc_environ_operations = {
 975         .open           = environ_open,
 976         .read           = environ_read,
 977         .llseek         = generic_file_llseek,
 978         .release        = mem_release,
 979 };
 980
 981 static int auxv_open(struct inode *inode, struct file *file)
 982 {
 983         return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
 984 }
 985
 986 static ssize_t auxv_read(struct file *file, char __user *buf,
 987                         size_t count, loff_t *ppos)
 988 {
 989         struct mm_struct *mm = file->private_data;
 990         unsigned int nwords = 0;
 991
 992         if (!mm)
 993                 return 0;
 994         do {
 995                 nwords += 2;
 996         } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 997         return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
 998                                        nwords * sizeof(mm->saved_auxv[0]));
 999 }
1000
1001 static const struct file_operations proc_auxv_operations = {
1002         .open           = auxv_open,
1003         .read           = auxv_read,
1004         .llseek         = generic_file_llseek,
1005         .release        = mem_release,
1006 };
1007
1008 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1009                             loff_t *ppos)
1010 {
1011         struct task_struct *task = get_proc_task(file_inode(file));
1012         char buffer[PROC_NUMBUF];
1013         int oom_adj = OOM_ADJUST_MIN;
1014         size_t len;
1015
1016         if (!task)
1017                 return -ESRCH;
1018         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1019                 oom_adj = OOM_ADJUST_MAX;
1020         else
1021                 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1022                           OOM_SCORE_ADJ_MAX;
1023         put_task_struct(task);
1024         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1025         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1026 }
1027
1028 static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1029 {
1030         static DEFINE_MUTEX(oom_adj_mutex);
1031         struct mm_struct *mm = NULL;
1032         struct task_struct *task;
1033         int err = 0;
1034
1035         task = get_proc_task(file_inode(file));
1036         if (!task)
1037                 return -ESRCH;
1038
1039         mutex_lock(&oom_adj_mutex);
1040         if (legacy) {
1041                 if (oom_adj < task->signal->oom_score_adj &&
1042                                 !capable(CAP_SYS_RESOURCE)) {
1043                         err = -EACCES;
1044                         goto err_unlock;
1045                 }
1046                 /*
1047                  * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1048                  * /proc/pid/oom_score_adj instead.
1049                  */
1050                 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1051                           current->comm, task_pid_nr(current), task_pid_nr(task),
1052                           task_pid_nr(task));
1053         } else {
1054                 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1055                                 !capable(CAP_SYS_RESOURCE)) {
1056                         err = -EACCES;
1057                         goto err_unlock;
1058                 }
1059         }
1060
1061         /*
1062          * Make sure we will check other processes sharing the mm if this is
1063          * not vfrok which wants its own oom_score_adj.
1064          * pin the mm so it doesn't go away and get reused after task_unlock
1065          */
1066         if (!task->vfork_done) {
1067                 struct task_struct *p = find_lock_task_mm(task);
1068
1069                 if (p) {
1070                         if (atomic_read(&p->mm->mm_users) > 1) {
1071                                 mm = p->mm;
1072                                 mmgrab(mm);
1073                         }
1074                         task_unlock(p);
1075                 }
1076         }
1077
1078         task->signal->oom_score_adj = oom_adj;
1079         if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1080                 task->signal->oom_score_adj_min = (short)oom_adj;
1081         trace_oom_score_adj_update(task);
1082
1083         if (mm) {
1084                 struct task_struct *p;
1085
1086                 rcu_read_lock();
1087                 for_each_process(p) {
1088                         if (same_thread_group(task, p))
1089                                 continue;
1090
1091                         /* do not touch kernel threads or the global init */
1092                         if (p->flags & PF_KTHREAD || is_global_init(p))
1093                                 continue;
1094
1095                         task_lock(p);
1096                         if (!p->vfork_done && process_shares_mm(p, mm)) {
1097                                 pr_info("updating oom_score_adj for %d (%s) from %d to %d because it shares mm with %d (%s). Report if this is unexpected.\n",
1098                                                 task_pid_nr(p), p->comm,
1099                                                 p->signal->oom_score_adj, oom_adj,
1100                                                 task_pid_nr(task), task->comm);
1101                                 p->signal->oom_score_adj = oom_adj;
1102                                 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1103                                         p->signal->oom_score_adj_min = (short)oom_adj;
1104                         }
1105                         task_unlock(p);
1106                 }
1107                 rcu_read_unlock();
1108                 mmdrop(mm);
1109         }
1110 err_unlock:
1111         mutex_unlock(&oom_adj_mutex);
1112         put_task_struct(task);
1113         return err;
1114 }
1115
1116 /*
1117  * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1118  * kernels.  The effective policy is defined by oom_score_adj, which has a
1119  * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1120  * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1121  * Processes that become oom disabled via oom_adj will still be oom disabled
1122  * with this implementation.
1123  *
1124  * oom_adj cannot be removed since existing userspace binaries use it.
1125  */
1126 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1127                              size_t count, loff_t *ppos)
1128 {
1129         char buffer[PROC_NUMBUF];
1130         int oom_adj;
1131         int err;
1132
1133         memset(buffer, 0, sizeof(buffer));
1134         if (count > sizeof(buffer) - 1)
1135                 count = sizeof(buffer) - 1;
1136         if (copy_from_user(buffer, buf, count)) {
1137                 err = -EFAULT;
1138                 goto out;
1139         }
1140
1141         err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1142         if (err)
1143                 goto out;
1144         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1145              oom_adj != OOM_DISABLE) {
1146                 err = -EINVAL;
1147                 goto out;
1148         }
1149
1150         /*
1151          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1152          * value is always attainable.
1153          */
1154         if (oom_adj == OOM_ADJUST_MAX)
1155                 oom_adj = OOM_SCORE_ADJ_MAX;
1156         else
1157                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1158
1159         err = __set_oom_adj(file, oom_adj, true);
1160 out:
1161         return err < 0 ? err : count;
1162 }
1163
1164 static const struct file_operations proc_oom_adj_operations = {
1165         .read           = oom_adj_read,
1166         .write          = oom_adj_write,
1167         .llseek         = generic_file_llseek,
1168 };
1169
1170 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1171                                         size_t count, loff_t *ppos)
1172 {
1173         struct task_struct *task = get_proc_task(file_inode(file));
1174         char buffer[PROC_NUMBUF];
1175         short oom_score_adj = OOM_SCORE_ADJ_MIN;
1176         size_t len;
1177
1178         if (!task)
1179                 return -ESRCH;
1180         oom_score_adj = task->signal->oom_score_adj;
1181         put_task_struct(task);
1182         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1183         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1184 }
1185
1186 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1187                                         size_t count, loff_t *ppos)
1188 {
1189         char buffer[PROC_NUMBUF];
1190         int oom_score_adj;
1191         int err;
1192
1193         memset(buffer, 0, sizeof(buffer));
1194         if (count > sizeof(buffer) - 1)
1195                 count = sizeof(buffer) - 1;
1196         if (copy_from_user(buffer, buf, count)) {
1197                 err = -EFAULT;
1198                 goto out;
1199         }
1200
1201         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1202         if (err)
1203                 goto out;
1204         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1205                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1206                 err = -EINVAL;
1207                 goto out;
1208         }
1209
1210         err = __set_oom_adj(file, oom_score_adj, false);
1211 out:
1212         return err < 0 ? err : count;
1213 }
1214
1215 static const struct file_operations proc_oom_score_adj_operations = {
1216         .read           = oom_score_adj_read,
1217         .write          = oom_score_adj_write,
1218         .llseek         = default_llseek,
1219 };
1220
1221 #ifdef CONFIG_AUDITSYSCALL
1222 #define TMPBUFLEN 11
1223 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1224                                   size_t count, loff_t *ppos)
1225 {
1226         struct inode * inode = file_inode(file);
1227         struct task_struct *task = get_proc_task(inode);
1228         ssize_t length;
1229         char tmpbuf[TMPBUFLEN];
1230
1231         if (!task)
1232                 return -ESRCH;
1233         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1234                            from_kuid(file->f_cred->user_ns,
1235                                      audit_get_loginuid(task)));
1236         put_task_struct(task);
1237         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1238 }
1239
1240 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1241                                    size_t count, loff_t *ppos)
1242 {
1243         struct inode * inode = file_inode(file);
1244         uid_t loginuid;
1245         kuid_t kloginuid;
1246         int rv;
1247
1248         rcu_read_lock();
1249         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1250                 rcu_read_unlock();
1251                 return -EPERM;
1252         }
1253         rcu_read_unlock();
1254
1255         if (*ppos != 0) {
1256                 /* No partial writes. */
1257                 return -EINVAL;
1258         }
1259
1260         rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1261         if (rv < 0)
1262                 return rv;
1263
1264         /* is userspace tring to explicitly UNSET the loginuid? */
1265         if (loginuid == AUDIT_UID_UNSET) {
1266                 kloginuid = INVALID_UID;
1267         } else {
1268                 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1269                 if (!uid_valid(kloginuid))
1270                         return -EINVAL;
1271         }
1272
1273         rv = audit_set_loginuid(kloginuid);
1274         if (rv < 0)
1275                 return rv;
1276         return count;
1277 }
1278
1279 static const struct file_operations proc_loginuid_operations = {
1280         .read           = proc_loginuid_read,
1281         .write          = proc_loginuid_write,
1282         .llseek         = generic_file_llseek,
1283 };
1284
1285 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1286                                   size_t count, loff_t *ppos)
1287 {
1288         struct inode * inode = file_inode(file);
1289         struct task_struct *task = get_proc_task(inode);
1290         ssize_t length;
1291         char tmpbuf[TMPBUFLEN];
1292
1293         if (!task)
1294                 return -ESRCH;
1295         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1296                                 audit_get_sessionid(task));
1297         put_task_struct(task);
1298         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1299 }
1300
1301 static const struct file_operations proc_sessionid_operations = {
1302         .read           = proc_sessionid_read,
1303         .llseek         = generic_file_llseek,
1304 };
1305 #endif
1306
1307 #ifdef CONFIG_FAULT_INJECTION
1308 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1309                                       size_t count, loff_t *ppos)
1310 {
1311         struct task_struct *task = get_proc_task(file_inode(file));
1312         char buffer[PROC_NUMBUF];
1313         size_t len;
1314         int make_it_fail;
1315
1316         if (!task)
1317                 return -ESRCH;
1318         make_it_fail = task->make_it_fail;
1319         put_task_struct(task);
1320
1321         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1322
1323         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1324 }
1325
1326 static ssize_t proc_fault_inject_write(struct file * file,
1327                         const char __user * buf, size_t count, loff_t *ppos)
1328 {
1329         struct task_struct *task;
1330         char buffer[PROC_NUMBUF];
1331         int make_it_fail;
1332         int rv;
1333
1334         if (!capable(CAP_SYS_RESOURCE))
1335                 return -EPERM;
1336         memset(buffer, 0, sizeof(buffer));
1337         if (count > sizeof(buffer) - 1)
1338                 count = sizeof(buffer) - 1;
1339         if (copy_from_user(buffer, buf, count))
1340                 return -EFAULT;
1341         rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1342         if (rv < 0)
1343                 return rv;
1344         if (make_it_fail < 0 || make_it_fail > 1)
1345                 return -EINVAL;
1346
1347         task = get_proc_task(file_inode(file));
1348         if (!task)
1349                 return -ESRCH;
1350         task->make_it_fail = make_it_fail;
1351         put_task_struct(task);
1352
1353         return count;
1354 }
1355
1356 static const struct file_operations proc_fault_inject_operations = {
1357         .read           = proc_fault_inject_read,
1358         .write          = proc_fault_inject_write,
1359         .llseek         = generic_file_llseek,
1360 };
1361
1362 static ssize_t proc_fail_nth_write(struct file *file, const char __user *buf,
1363                                    size_t count, loff_t *ppos)
1364 {
1365         struct task_struct *task;
1366         int err;
1367         unsigned int n;
1368
1369         err = kstrtouint_from_user(buf, count, 0, &n);
1370         if (err)
1371                 return err;
1372
1373         task = get_proc_task(file_inode(file));
1374         if (!task)
1375                 return -ESRCH;
1376         task->fail_nth = n;
1377         put_task_struct(task);
1378
1379         return count;
1380 }
1381
1382 static ssize_t proc_fail_nth_read(struct file *file, char __user *buf,
1383                                   size_t count, loff_t *ppos)
1384 {
1385         struct task_struct *task;
1386         char numbuf[PROC_NUMBUF];
1387         ssize_t len;
1388
1389         task = get_proc_task(file_inode(file));
1390         if (!task)
1391                 return -ESRCH;
1392         len = snprintf(numbuf, sizeof(numbuf), "%u\n", task->fail_nth);
1393         len = simple_read_from_buffer(buf, count, ppos, numbuf, len);
1394         put_task_struct(task);
1395
1396         return len;
1397 }
1398
1399 static const struct file_operations proc_fail_nth_operations = {
1400         .read           = proc_fail_nth_read,
1401         .write          = proc_fail_nth_write,
1402 };
1403 #endif
1404
1405
1406 #ifdef CONFIG_SCHED_DEBUG
1407 /*
1408  * Print out various scheduling related per-task fields:
1409  */
1410 static int sched_show(struct seq_file *m, void *v)
1411 {
1412         struct inode *inode = m->private;
1413         struct pid_namespace *ns = inode->i_sb->s_fs_info;
1414         struct task_struct *p;
1415
1416         p = get_proc_task(inode);
1417         if (!p)
1418                 return -ESRCH;
1419         proc_sched_show_task(p, ns, m);
1420
1421         put_task_struct(p);
1422
1423         return 0;
1424 }
1425
1426 static ssize_t
1427 sched_write(struct file *file, const char __user *buf,
1428             size_t count, loff_t *offset)
1429 {
1430         struct inode *inode = file_inode(file);
1431         struct task_struct *p;
1432
1433         p = get_proc_task(inode);
1434         if (!p)
1435                 return -ESRCH;
1436         proc_sched_set_task(p);
1437
1438         put_task_struct(p);
1439
1440         return count;
1441 }
1442
1443 static int sched_open(struct inode *inode, struct file *filp)
1444 {
1445         return single_open(filp, sched_show, inode);
1446 }
1447
1448 static const struct file_operations proc_pid_sched_operations = {
1449         .open           = sched_open,
1450         .read           = seq_read,
1451         .write          = sched_write,
1452         .llseek         = seq_lseek,
1453         .release        = single_release,
1454 };
1455
1456 #endif
1457
1458 #ifdef CONFIG_SCHED_AUTOGROUP
1459 /*
1460  * Print out autogroup related information:
1461  */
1462 static int sched_autogroup_show(struct seq_file *m, void *v)
1463 {
1464         struct inode *inode = m->private;
1465         struct task_struct *p;
1466
1467         p = get_proc_task(inode);
1468         if (!p)
1469                 return -ESRCH;
1470         proc_sched_autogroup_show_task(p, m);
1471
1472         put_task_struct(p);
1473
1474         return 0;
1475 }
1476
1477 static ssize_t
1478 sched_autogroup_write(struct file *file, const char __user *buf,
1479             size_t count, loff_t *offset)
1480 {
1481         struct inode *inode = file_inode(file);
1482         struct task_struct *p;
1483         char buffer[PROC_NUMBUF];
1484         int nice;
1485         int err;
1486
1487         memset(buffer, 0, sizeof(buffer));
1488         if (count > sizeof(buffer) - 1)
1489                 count = sizeof(buffer) - 1;
1490         if (copy_from_user(buffer, buf, count))
1491                 return -EFAULT;
1492
1493         err = kstrtoint(strstrip(buffer), 0, &nice);
1494         if (err < 0)
1495                 return err;
1496
1497         p = get_proc_task(inode);
1498         if (!p)
1499                 return -ESRCH;
1500
1501         err = proc_sched_autogroup_set_nice(p, nice);
1502         if (err)
1503                 count = err;
1504
1505         put_task_struct(p);
1506
1507         return count;
1508 }
1509
1510 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1511 {
1512         int ret;
1513
1514         ret = single_open(filp, sched_autogroup_show, NULL);
1515         if (!ret) {
1516                 struct seq_file *m = filp->private_data;
1517
1518                 m->private = inode;
1519         }
1520         return ret;
1521 }
1522
1523 static const struct file_operations proc_pid_sched_autogroup_operations = {
1524         .open           = sched_autogroup_open,
1525         .read           = seq_read,
1526         .write          = sched_autogroup_write,
1527         .llseek         = seq_lseek,
1528         .release        = single_release,
1529 };
1530
1531 #endif /* CONFIG_SCHED_AUTOGROUP */
1532
1533 static ssize_t comm_write(struct file *file, const char __user *buf,
1534                                 size_t count, loff_t *offset)
1535 {
1536         struct inode *inode = file_inode(file);
1537         struct task_struct *p;
1538         char buffer[TASK_COMM_LEN];
1539         const size_t maxlen = sizeof(buffer) - 1;
1540
1541         memset(buffer, 0, sizeof(buffer));
1542         if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1543                 return -EFAULT;
1544
1545         p = get_proc_task(inode);
1546         if (!p)
1547                 return -ESRCH;
1548
1549         if (same_thread_group(current, p))
1550                 set_task_comm(p, buffer);
1551         else
1552                 count = -EINVAL;
1553
1554         put_task_struct(p);
1555
1556         return count;
1557 }
1558
1559 static int comm_show(struct seq_file *m, void *v)
1560 {
1561         struct inode *inode = m->private;
1562         struct task_struct *p;
1563
1564         p = get_proc_task(inode);
1565         if (!p)
1566                 return -ESRCH;
1567
1568         task_lock(p);
1569         seq_printf(m, "%s\n", p->comm);
1570         task_unlock(p);
1571
1572         put_task_struct(p);
1573
1574         return 0;
1575 }
1576
1577 static int comm_open(struct inode *inode, struct file *filp)
1578 {
1579         return single_open(filp, comm_show, inode);
1580 }
1581
1582 static const struct file_operations proc_pid_set_comm_operations = {
1583         .open           = comm_open,
1584         .read           = seq_read,
1585         .write          = comm_write,
1586         .llseek         = seq_lseek,
1587         .release        = single_release,
1588 };
1589
1590 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1591 {
1592         struct task_struct *task;
1593         struct file *exe_file;
1594
1595         task = get_proc_task(d_inode(dentry));
1596         if (!task)
1597                 return -ENOENT;
1598         exe_file = get_task_exe_file(task);
1599         put_task_struct(task);
1600         if (exe_file) {
1601                 *exe_path = exe_file->f_path;
1602                 path_get(&exe_file->f_path);
1603                 fput(exe_file);
1604                 return 0;
1605         } else
1606                 return -ENOENT;
1607 }
1608
1609 static const char *proc_pid_get_link(struct dentry *dentry,
1610                                      struct inode *inode,
1611                                      struct delayed_call *done)
1612 {
1613         struct path path;
1614         int error = -EACCES;
1615
1616         if (!dentry)
1617                 return ERR_PTR(-ECHILD);
1618
1619         /* Are we allowed to snoop on the tasks file descriptors? */
1620         if (!proc_fd_access_allowed(inode))
1621                 goto out;
1622
1623         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1624         if (error)
1625                 goto out;
1626
1627         nd_jump_link(&path);
1628         return NULL;
1629 out:
1630         return ERR_PTR(error);
1631 }
1632
1633 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1634 {
1635         char *tmp = (char *)__get_free_page(GFP_KERNEL);
1636         char *pathname;
1637         int len;
1638
1639         if (!tmp)
1640                 return -ENOMEM;
1641
1642         pathname = d_path(path, tmp, PAGE_SIZE);
1643         len = PTR_ERR(pathname);
1644         if (IS_ERR(pathname))
1645                 goto out;
1646         len = tmp + PAGE_SIZE - 1 - pathname;
1647
1648         if (len > buflen)
1649                 len = buflen;
1650         if (copy_to_user(buffer, pathname, len))
1651                 len = -EFAULT;
1652  out:
1653         free_page((unsigned long)tmp);
1654         return len;
1655 }
1656
1657 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1658 {
1659         int error = -EACCES;
1660         struct inode *inode = d_inode(dentry);
1661         struct path path;
1662
1663         /* Are we allowed to snoop on the tasks file descriptors? */
1664         if (!proc_fd_access_allowed(inode))
1665                 goto out;
1666
1667         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1668         if (error)
1669                 goto out;
1670
1671         error = do_proc_readlink(&path, buffer, buflen);
1672         path_put(&path);
1673 out:
1674         return error;
1675 }
1676
1677 const struct inode_operations proc_pid_link_inode_operations = {
1678         .readlink       = proc_pid_readlink,
1679         .get_link       = proc_pid_get_link,
1680         .setattr        = proc_setattr,
1681 };
1682
1683
1684 /* building an inode */
1685
1686 void task_dump_owner(struct task_struct *task, umode_t mode,
1687                      kuid_t *ruid, kgid_t *rgid)
1688 {
1689         /* Depending on the state of dumpable compute who should own a
1690          * proc file for a task.
1691          */
1692         const struct cred *cred;
1693         kuid_t uid;
1694         kgid_t gid;
1695
1696         if (unlikely(task->flags & PF_KTHREAD)) {
1697                 *ruid = GLOBAL_ROOT_UID;
1698                 *rgid = GLOBAL_ROOT_GID;
1699                 return;
1700         }
1701
1702         /* Default to the tasks effective ownership */
1703         rcu_read_lock();
1704         cred = __task_cred(task);
1705         uid = cred->euid;
1706         gid = cred->egid;
1707         rcu_read_unlock();
1708
1709         /*
1710          * Before the /proc/pid/status file was created the only way to read
1711          * the effective uid of a /process was to stat /proc/pid.  Reading
1712          * /proc/pid/status is slow enough that procps and other packages
1713          * kept stating /proc/pid.  To keep the rules in /proc simple I have
1714          * made this apply to all per process world readable and executable
1715          * directories.
1716          */
1717         if (mode != (S_IFDIR|S_IRUGO|S_IXUGO)) {
1718                 struct mm_struct *mm;
1719                 task_lock(task);
1720                 mm = task->mm;
1721                 /* Make non-dumpable tasks owned by some root */
1722                 if (mm) {
1723                         if (get_dumpable(mm) != SUID_DUMP_USER) {
1724                                 struct user_namespace *user_ns = mm->user_ns;
1725
1726                                 uid = make_kuid(user_ns, 0);
1727                                 if (!uid_valid(uid))
1728                                         uid = GLOBAL_ROOT_UID;
1729
1730                                 gid = make_kgid(user_ns, 0);
1731                                 if (!gid_valid(gid))
1732                                         gid = GLOBAL_ROOT_GID;
1733                         }
1734                 } else {
1735                         uid = GLOBAL_ROOT_UID;
1736                         gid = GLOBAL_ROOT_GID;
1737                 }
1738                 task_unlock(task);
1739         }
1740         *ruid = uid;
1741         *rgid = gid;
1742 }
1743
1744 struct inode *proc_pid_make_inode(struct super_block * sb,
1745                                   struct task_struct *task, umode_t mode)
1746 {
1747         struct inode * inode;
1748         struct proc_inode *ei;
1749
1750         /* We need a new inode */
1751
1752         inode = new_inode(sb);
1753         if (!inode)
1754                 goto out;
1755
1756         /* Common stuff */
1757         ei = PROC_I(inode);
1758         inode->i_mode = mode;
1759         inode->i_ino = get_next_ino();
1760         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1761         inode->i_op = &proc_def_inode_operations;
1762
1763         /*
1764          * grab the reference to task.
1765          */
1766         ei->pid = get_task_pid(task, PIDTYPE_PID);
1767         if (!ei->pid)
1768                 goto out_unlock;
1769
1770         task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1771         security_task_to_inode(task, inode);
1772
1773 out:
1774         return inode;
1775
1776 out_unlock:
1777         iput(inode);
1778         return NULL;
1779 }
1780
1781 int pid_getattr(const struct path *path, struct kstat *stat,
1782                 u32 request_mask, unsigned int query_flags)
1783 {
1784         struct inode *inode = d_inode(path->dentry);
1785         struct task_struct *task;
1786         struct pid_namespace *pid = path->dentry->d_sb->s_fs_info;
1787
1788         generic_fillattr(inode, stat);
1789
1790         rcu_read_lock();
1791         stat->uid = GLOBAL_ROOT_UID;
1792         stat->gid = GLOBAL_ROOT_GID;
1793         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1794         if (task) {
1795                 if (!has_pid_permissions(pid, task, HIDEPID_INVISIBLE)) {
1796                         rcu_read_unlock();
1797                         /*
1798                          * This doesn't prevent learning whether PID exists,
1799                          * it only makes getattr() consistent with readdir().
1800                          */
1801                         return -ENOENT;
1802                 }
1803                 task_dump_owner(task, inode->i_mode, &stat->uid, &stat->gid);
1804         }
1805         rcu_read_unlock();
1806         return 0;
1807 }
1808
1809 /* dentry stuff */
1810
1811 /*
1812  *      Exceptional case: normally we are not allowed to unhash a busy
1813  * directory. In this case, however, we can do it - no aliasing problems
1814  * due to the way we treat inodes.
1815  *
1816  * Rewrite the inode's ownerships here because the owning task may have
1817  * performed a setuid(), etc.
1818  *
1819  */
1820 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1821 {
1822         struct inode *inode;
1823         struct task_struct *task;
1824
1825         if (flags & LOOKUP_RCU)
1826                 return -ECHILD;
1827
1828         inode = d_inode(dentry);
1829         task = get_proc_task(inode);
1830
1831         if (task) {
1832                 task_dump_owner(task, inode->i_mode, &inode->i_uid, &inode->i_gid);
1833
1834                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1835                 security_task_to_inode(task, inode);
1836                 put_task_struct(task);
1837                 return 1;
1838         }
1839         return 0;
1840 }
1841
1842 static inline bool proc_inode_is_dead(struct inode *inode)
1843 {
1844         return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1845 }
1846
1847 int pid_delete_dentry(const struct dentry *dentry)
1848 {
1849         /* Is the task we represent dead?
1850          * If so, then don't put the dentry on the lru list,
1851          * kill it immediately.
1852          */
1853         return proc_inode_is_dead(d_inode(dentry));
1854 }
1855
1856 const struct dentry_operations pid_dentry_operations =
1857 {
1858         .d_revalidate   = pid_revalidate,
1859         .d_delete       = pid_delete_dentry,
1860 };
1861
1862 /* Lookups */
1863
1864 /*
1865  * Fill a directory entry.
1866  *
1867  * If possible create the dcache entry and derive our inode number and
1868  * file type from dcache entry.
1869  *
1870  * Since all of the proc inode numbers are dynamically generated, the inode
1871  * numbers do not exist until the inode is cache.  This means creating the
1872  * the dcache entry in readdir is necessary to keep the inode numbers
1873  * reported by readdir in sync with the inode numbers reported
1874  * by stat.
1875  */
1876 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1877         const char *name, int len,
1878         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1879 {
1880         struct dentry *child, *dir = file->f_path.dentry;
1881         struct qstr qname = QSTR_INIT(name, len);
1882         struct inode *inode;
1883         unsigned type;
1884         ino_t ino;
1885
1886         child = d_hash_and_lookup(dir, &qname);
1887         if (!child) {
1888                 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1889                 child = d_alloc_parallel(dir, &qname, &wq);
1890                 if (IS_ERR(child))
1891                         goto end_instantiate;
1892                 if (d_in_lookup(child)) {
1893                         int err = instantiate(d_inode(dir), child, task, ptr);
1894                         d_lookup_done(child);
1895                         if (err < 0) {
1896                                 dput(child);
1897                                 goto end_instantiate;
1898                         }
1899                 }
1900         }
1901         inode = d_inode(child);
1902         ino = inode->i_ino;
1903         type = inode->i_mode >> 12;
1904         dput(child);
1905         return dir_emit(ctx, name, len, ino, type);
1906
1907 end_instantiate:
1908         return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1909 }
1910
1911 /*
1912  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1913  * which represent vma start and end addresses.
1914  */
1915 static int dname_to_vma_addr(struct dentry *dentry,
1916                              unsigned long *start, unsigned long *end)
1917 {
1918         const char *str = dentry->d_name.name;
1919         unsigned long long sval, eval;
1920         unsigned int len;
1921
1922         if (str[0] == '0' && str[1] != '-')
1923                 return -EINVAL;
1924         len = _parse_integer(str, 16, &sval);
1925         if (len & KSTRTOX_OVERFLOW)
1926                 return -EINVAL;
1927         if (sval != (unsigned long)sval)
1928                 return -EINVAL;
1929         str += len;
1930
1931         if (*str != '-')
1932                 return -EINVAL;
1933         str++;
1934
1935         if (str[0] == '0' && str[1])
1936                 return -EINVAL;
1937         len = _parse_integer(str, 16, &eval);
1938         if (len & KSTRTOX_OVERFLOW)
1939                 return -EINVAL;
1940         if (eval != (unsigned long)eval)
1941                 return -EINVAL;
1942         str += len;
1943
1944         if (*str != '\0')
1945                 return -EINVAL;
1946
1947         *start = sval;
1948         *end = eval;
1949
1950         return 0;
1951 }
1952
1953 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1954 {
1955         unsigned long vm_start, vm_end;
1956         bool exact_vma_exists = false;
1957         struct mm_struct *mm = NULL;
1958         struct task_struct *task;
1959         struct inode *inode;
1960         int status = 0;
1961
1962         if (flags & LOOKUP_RCU)
1963                 return -ECHILD;
1964
1965         inode = d_inode(dentry);
1966         task = get_proc_task(inode);
1967         if (!task)
1968                 goto out_notask;
1969
1970         mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1971         if (IS_ERR_OR_NULL(mm))
1972                 goto out;
1973
1974         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1975                 down_read(&mm->mmap_sem);
1976                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1977                 up_read(&mm->mmap_sem);
1978         }
1979
1980         mmput(mm);
1981
1982         if (exact_vma_exists) {
1983                 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1984
1985                 security_task_to_inode(task, inode);
1986                 status = 1;
1987         }
1988
1989 out:
1990         put_task_struct(task);
1991
1992 out_notask:
1993         return status;
1994 }
1995
1996 static const struct dentry_operations tid_map_files_dentry_operations = {
1997         .d_revalidate   = map_files_d_revalidate,
1998         .d_delete       = pid_delete_dentry,
1999 };
2000
2001 static int map_files_get_link(struct dentry *dentry, struct path *path)
2002 {
2003         unsigned long vm_start, vm_end;
2004         struct vm_area_struct *vma;
2005         struct task_struct *task;
2006         struct mm_struct *mm;
2007         int rc;
2008
2009         rc = -ENOENT;
2010         task = get_proc_task(d_inode(dentry));
2011         if (!task)
2012                 goto out;
2013
2014         mm = get_task_mm(task);
2015         put_task_struct(task);
2016         if (!mm)
2017                 goto out;
2018
2019         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2020         if (rc)
2021                 goto out_mmput;
2022
2023         rc = -ENOENT;
2024         down_read(&mm->mmap_sem);
2025         vma = find_exact_vma(mm, vm_start, vm_end);
2026         if (vma && vma->vm_file) {
2027                 *path = vma->vm_file->f_path;
2028                 path_get(path);
2029                 rc = 0;
2030         }
2031         up_read(&mm->mmap_sem);
2032
2033 out_mmput:
2034         mmput(mm);
2035 out:
2036         return rc;
2037 }
2038
2039 struct map_files_info {
2040         unsigned long   start;
2041         unsigned long   end;
2042         fmode_t         mode;
2043 };
2044
2045 /*
2046  * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
2047  * symlinks may be used to bypass permissions on ancestor directories in the
2048  * path to the file in question.
2049  */
2050 static const char *
2051 proc_map_files_get_link(struct dentry *dentry,
2052                         struct inode *inode,
2053                         struct delayed_call *done)
2054 {
2055         if (!capable(CAP_SYS_ADMIN))
2056                 return ERR_PTR(-EPERM);
2057
2058         return proc_pid_get_link(dentry, inode, done);
2059 }
2060
2061 /*
2062  * Identical to proc_pid_link_inode_operations except for get_link()
2063  */
2064 static const struct inode_operations proc_map_files_link_inode_operations = {
2065         .readlink       = proc_pid_readlink,
2066         .get_link       = proc_map_files_get_link,
2067         .setattr        = proc_setattr,
2068 };
2069
2070 static int
2071 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
2072                            struct task_struct *task, const void *ptr)
2073 {
2074         fmode_t mode = (fmode_t)(unsigned long)ptr;
2075         struct proc_inode *ei;
2076         struct inode *inode;
2077
2078         inode = proc_pid_make_inode(dir->i_sb, task, S_IFLNK |
2079                                     ((mode & FMODE_READ ) ? S_IRUSR : 0) |
2080                                     ((mode & FMODE_WRITE) ? S_IWUSR : 0));
2081         if (!inode)
2082                 return -ENOENT;
2083
2084         ei = PROC_I(inode);
2085         ei->op.proc_get_link = map_files_get_link;
2086
2087         inode->i_op = &proc_map_files_link_inode_operations;
2088         inode->i_size = 64;
2089
2090         d_set_d_op(dentry, &tid_map_files_dentry_operations);
2091         d_add(dentry, inode);
2092
2093         return 0;
2094 }
2095
2096 static struct dentry *proc_map_files_lookup(struct inode *dir,
2097                 struct dentry *dentry, unsigned int flags)
2098 {
2099         unsigned long vm_start, vm_end;
2100         struct vm_area_struct *vma;
2101         struct task_struct *task;
2102         int result;
2103         struct mm_struct *mm;
2104
2105         result = -ENOENT;
2106         task = get_proc_task(dir);
2107         if (!task)
2108                 goto out;
2109
2110         result = -EACCES;
2111         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2112                 goto out_put_task;
2113
2114         result = -ENOENT;
2115         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2116                 goto out_put_task;
2117
2118         mm = get_task_mm(task);
2119         if (!mm)
2120                 goto out_put_task;
2121
2122         down_read(&mm->mmap_sem);
2123         vma = find_exact_vma(mm, vm_start, vm_end);
2124         if (!vma)
2125                 goto out_no_vma;
2126
2127         if (vma->vm_file)
2128                 result = proc_map_files_instantiate(dir, dentry, task,
2129                                 (void *)(unsigned long)vma->vm_file->f_mode);
2130
2131 out_no_vma:
2132         up_read(&mm->mmap_sem);
2133         mmput(mm);
2134 out_put_task:
2135         put_task_struct(task);
2136 out:
2137         return ERR_PTR(result);
2138 }
2139
2140 static const struct inode_operations proc_map_files_inode_operations = {
2141         .lookup         = proc_map_files_lookup,
2142         .permission     = proc_fd_permission,
2143         .setattr        = proc_setattr,
2144 };
2145
2146 static int
2147 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2148 {
2149         struct vm_area_struct *vma;
2150         struct task_struct *task;
2151         struct mm_struct *mm;
2152         unsigned long nr_files, pos, i;
2153         struct flex_array *fa = NULL;
2154         struct map_files_info info;
2155         struct map_files_info *p;
2156         int ret;
2157
2158         ret = -ENOENT;
2159         task = get_proc_task(file_inode(file));
2160         if (!task)
2161                 goto out;
2162
2163         ret = -EACCES;
2164         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2165                 goto out_put_task;
2166
2167         ret = 0;
2168         if (!dir_emit_dots(file, ctx))
2169                 goto out_put_task;
2170
2171         mm = get_task_mm(task);
2172         if (!mm)
2173                 goto out_put_task;
2174         down_read(&mm->mmap_sem);
2175
2176         nr_files = 0;
2177
2178         /*
2179          * We need two passes here:
2180          *
2181          *  1) Collect vmas of mapped files with mmap_sem taken
2182          *  2) Release mmap_sem and instantiate entries
2183          *
2184          * otherwise we get lockdep complained, since filldir()
2185          * routine might require mmap_sem taken in might_fault().
2186          */
2187
2188         for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2189                 if (vma->vm_file && ++pos > ctx->pos)
2190                         nr_files++;
2191         }
2192
2193         if (nr_files) {
2194                 fa = flex_array_alloc(sizeof(info), nr_files,
2195                                         GFP_KERNEL);
2196                 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2197                                                 GFP_KERNEL)) {
2198                         ret = -ENOMEM;
2199                         if (fa)
2200                                 flex_array_free(fa);
2201                         up_read(&mm->mmap_sem);
2202                         mmput(mm);
2203                         goto out_put_task;
2204                 }
2205                 for (i = 0, vma = mm->mmap, pos = 2; vma;
2206                                 vma = vma->vm_next) {
2207                         if (!vma->vm_file)
2208                                 continue;
2209                         if (++pos <= ctx->pos)
2210                                 continue;
2211
2212                         info.start = vma->vm_start;
2213                         info.end = vma->vm_end;
2214                         info.mode = vma->vm_file->f_mode;
2215                         if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2216                                 BUG();
2217                 }
2218         }
2219         up_read(&mm->mmap_sem);
2220         mmput(mm);
2221
2222         for (i = 0; i < nr_files; i++) {
2223                 char buf[4 * sizeof(long) + 2]; /* max: %lx-%lx\0 */
2224                 unsigned int len;
2225
2226                 p = flex_array_get(fa, i);
2227                 len = snprintf(buf, sizeof(buf), "%lx-%lx", p->start, p->end);
2228                 if (!proc_fill_cache(file, ctx,
2229                                       buf, len,
2230                                       proc_map_files_instantiate,
2231                                       task,
2232                                       (void *)(unsigned long)p->mode))
2233                         break;
2234                 ctx->pos++;
2235         }
2236         if (fa)
2237                 flex_array_free(fa);
2238
2239 out_put_task:
2240         put_task_struct(task);
2241 out:
2242         return ret;
2243 }
2244
2245 static const struct file_operations proc_map_files_operations = {
2246         .read           = generic_read_dir,
2247         .iterate_shared = proc_map_files_readdir,
2248         .llseek         = generic_file_llseek,
2249 };
2250
2251 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2252 struct timers_private {
2253         struct pid *pid;
2254         struct task_struct *task;
2255         struct sighand_struct *sighand;
2256         struct pid_namespace *ns;
2257         unsigned long flags;
2258 };
2259
2260 static void *timers_start(struct seq_file *m, loff_t *pos)
2261 {
2262         struct timers_private *tp = m->private;
2263
2264         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2265         if (!tp->task)
2266                 return ERR_PTR(-ESRCH);
2267
2268         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2269         if (!tp->sighand)
2270                 return ERR_PTR(-ESRCH);
2271
2272         return seq_list_start(&tp->task->signal->posix_timers, *pos);
2273 }
2274
2275 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2276 {
2277         struct timers_private *tp = m->private;
2278         return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2279 }
2280
2281 static void timers_stop(struct seq_file *m, void *v)
2282 {
2283         struct timers_private *tp = m->private;
2284
2285         if (tp->sighand) {
2286                 unlock_task_sighand(tp->task, &tp->flags);
2287                 tp->sighand = NULL;
2288         }
2289
2290         if (tp->task) {
2291                 put_task_struct(tp->task);
2292                 tp->task = NULL;
2293         }
2294 }
2295
2296 static int show_timer(struct seq_file *m, void *v)
2297 {
2298         struct k_itimer *timer;
2299         struct timers_private *tp = m->private;
2300         int notify;
2301         static const char * const nstr[] = {
2302                 [SIGEV_SIGNAL] = "signal",
2303                 [SIGEV_NONE] = "none",
2304                 [SIGEV_THREAD] = "thread",
2305         };
2306
2307         timer = list_entry((struct list_head *)v, struct k_itimer, list);
2308         notify = timer->it_sigev_notify;
2309
2310         seq_printf(m, "ID: %d\n", timer->it_id);
2311         seq_printf(m, "signal: %d/%px\n",
2312                    timer->sigq->info.si_signo,
2313                    timer->sigq->info.si_value.sival_ptr);
2314         seq_printf(m, "notify: %s/%s.%d\n",
2315                    nstr[notify & ~SIGEV_THREAD_ID],
2316                    (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2317                    pid_nr_ns(timer->it_pid, tp->ns));
2318         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2319
2320         return 0;
2321 }
2322
2323 static const struct seq_operations proc_timers_seq_ops = {
2324         .start  = timers_start,
2325         .next   = timers_next,
2326         .stop   = timers_stop,
2327         .show   = show_timer,
2328 };
2329
2330 static int proc_timers_open(struct inode *inode, struct file *file)
2331 {
2332         struct timers_private *tp;
2333
2334         tp = __seq_open_private(file, &proc_timers_seq_ops,
2335                         sizeof(struct timers_private));
2336         if (!tp)
2337                 return -ENOMEM;
2338
2339         tp->pid = proc_pid(inode);
2340         tp->ns = inode->i_sb->s_fs_info;
2341         return 0;
2342 }
2343
2344 static const struct file_operations proc_timers_operations = {
2345         .open           = proc_timers_open,
2346         .read           = seq_read,
2347         .llseek         = seq_lseek,
2348         .release        = seq_release_private,
2349 };
2350 #endif
2351
2352 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2353                                         size_t count, loff_t *offset)
2354 {
2355         struct inode *inode = file_inode(file);
2356         struct task_struct *p;
2357         u64 slack_ns;
2358         int err;
2359
2360         err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2361         if (err < 0)
2362                 return err;
2363
2364         p = get_proc_task(inode);
2365         if (!p)
2366                 return -ESRCH;
2367
2368         if (p != current) {
2369                 if (!capable(CAP_SYS_NICE)) {
2370                         count = -EPERM;
2371                         goto out;
2372                 }
2373
2374                 err = security_task_setscheduler(p);
2375                 if (err) {
2376                         count = err;
2377                         goto out;
2378                 }
2379         }
2380
2381         task_lock(p);
2382         if (slack_ns == 0)
2383                 p->timer_slack_ns = p->default_timer_slack_ns;
2384         else
2385                 p->timer_slack_ns = slack_ns;
2386         task_unlock(p);
2387
2388 out:
2389         put_task_struct(p);
2390
2391         return count;
2392 }
2393
2394 static int timerslack_ns_show(struct seq_file *m, void *v)
2395 {
2396         struct inode *inode = m->private;
2397         struct task_struct *p;
2398         int err = 0;
2399
2400         p = get_proc_task(inode);
2401         if (!p)
2402                 return -ESRCH;
2403
2404         if (p != current) {
2405
2406                 if (!capable(CAP_SYS_NICE)) {
2407                         err = -EPERM;
2408                         goto out;
2409                 }
2410                 err = security_task_getscheduler(p);
2411                 if (err)
2412                         goto out;
2413         }
2414
2415         task_lock(p);
2416         seq_printf(m, "%llu\n", p->timer_slack_ns);
2417         task_unlock(p);
2418
2419 out:
2420         put_task_struct(p);
2421
2422         return err;
2423 }
2424
2425 static int timerslack_ns_open(struct inode *inode, struct file *filp)
2426 {
2427         return single_open(filp, timerslack_ns_show, inode);
2428 }
2429
2430 static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2431         .open           = timerslack_ns_open,
2432         .read           = seq_read,
2433         .write          = timerslack_ns_write,
2434         .llseek         = seq_lseek,
2435         .release        = single_release,
2436 };
2437
2438 static int proc_pident_instantiate(struct inode *dir,
2439         struct dentry *dentry, struct task_struct *task, const void *ptr)
2440 {
2441         const struct pid_entry *p = ptr;
2442         struct inode *inode;
2443         struct proc_inode *ei;
2444
2445         inode = proc_pid_make_inode(dir->i_sb, task, p->mode);
2446         if (!inode)
2447                 goto out;
2448
2449         ei = PROC_I(inode);
2450         if (S_ISDIR(inode->i_mode))
2451                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2452         if (p->iop)
2453                 inode->i_op = p->iop;
2454         if (p->fop)
2455                 inode->i_fop = p->fop;
2456         ei->op = p->op;
2457         d_set_d_op(dentry, &pid_dentry_operations);
2458         d_add(dentry, inode);
2459         /* Close the race of the process dying before we return the dentry */
2460         if (pid_revalidate(dentry, 0))
2461                 return 0;
2462 out:
2463         return -ENOENT;
2464 }
2465
2466 static struct dentry *proc_pident_lookup(struct inode *dir,
2467                                          struct dentry *dentry,
2468                                          const struct pid_entry *ents,
2469                                          unsigned int nents)
2470 {
2471         int error;
2472         struct task_struct *task = get_proc_task(dir);
2473         const struct pid_entry *p, *last;
2474
2475         error = -ENOENT;
2476
2477         if (!task)
2478                 goto out_no_task;
2479
2480         /*
2481          * Yes, it does not scale. And it should not. Don't add
2482          * new entries into /proc/<tgid>/ without very good reasons.
2483          */
2484         last = &ents[nents];
2485         for (p = ents; p < last; p++) {
2486                 if (p->len != dentry->d_name.len)
2487                         continue;
2488                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2489                         break;
2490         }
2491         if (p >= last)
2492                 goto out;
2493
2494         error = proc_pident_instantiate(dir, dentry, task, p);
2495 out:
2496         put_task_struct(task);
2497 out_no_task:
2498         return ERR_PTR(error);
2499 }
2500
2501 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2502                 const struct pid_entry *ents, unsigned int nents)
2503 {
2504         struct task_struct *task = get_proc_task(file_inode(file));
2505         const struct pid_entry *p;
2506
2507         if (!task)
2508                 return -ENOENT;
2509
2510         if (!dir_emit_dots(file, ctx))
2511                 goto out;
2512
2513         if (ctx->pos >= nents + 2)
2514                 goto out;
2515
2516         for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2517                 if (!proc_fill_cache(file, ctx, p->name, p->len,
2518                                 proc_pident_instantiate, task, p))
2519                         break;
2520                 ctx->pos++;
2521         }
2522 out:
2523         put_task_struct(task);
2524         return 0;
2525 }
2526
2527 #ifdef CONFIG_SECURITY
2528 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2529                                   size_t count, loff_t *ppos)
2530 {
2531         struct inode * inode = file_inode(file);
2532         char *p = NULL;
2533         ssize_t length;
2534         struct task_struct *task = get_proc_task(inode);
2535
2536         if (!task)
2537                 return -ESRCH;
2538
2539         length = security_getprocattr(task,
2540                                       (char*)file->f_path.dentry->d_name.name,
2541                                       &p);
2542         put_task_struct(task);
2543         if (length > 0)
2544                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2545         kfree(p);
2546         return length;
2547 }
2548
2549 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2550                                    size_t count, loff_t *ppos)
2551 {
2552         struct inode * inode = file_inode(file);
2553         void *page;
2554         ssize_t length;
2555         struct task_struct *task = get_proc_task(inode);
2556
2557         length = -ESRCH;
2558         if (!task)
2559                 goto out_no_task;
2560
2561         /* A task may only write its own attributes. */
2562         length = -EACCES;
2563         if (current != task)
2564                 goto out;
2565
2566         if (count > PAGE_SIZE)
2567                 count = PAGE_SIZE;
2568
2569         /* No partial writes. */
2570         length = -EINVAL;
2571         if (*ppos != 0)
2572                 goto out;
2573
2574         page = memdup_user(buf, count);
2575         if (IS_ERR(page)) {
2576                 length = PTR_ERR(page);
2577                 goto out;
2578         }
2579
2580         /* Guard against adverse ptrace interaction */
2581         length = mutex_lock_interruptible(&current->signal->cred_guard_mutex);
2582         if (length < 0)
2583                 goto out_free;
2584
2585         length = security_setprocattr(file->f_path.dentry->d_name.name,
2586                                       page, count);
2587         mutex_unlock(&current->signal->cred_guard_mutex);
2588 out_free:
2589         kfree(page);
2590 out:
2591         put_task_struct(task);
2592 out_no_task:
2593         return length;
2594 }
2595
2596 static const struct file_operations proc_pid_attr_operations = {
2597         .read           = proc_pid_attr_read,
2598         .write          = proc_pid_attr_write,
2599         .llseek         = generic_file_llseek,
2600 };
2601
2602 static const struct pid_entry attr_dir_stuff[] = {
2603         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2604         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2605         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2606         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2607         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2608         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2609 };
2610
2611 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2612 {
2613         return proc_pident_readdir(file, ctx,
2614                                    attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2615 }
2616
2617 static const struct file_operations proc_attr_dir_operations = {
2618         .read           = generic_read_dir,
2619         .iterate_shared = proc_attr_dir_readdir,
2620         .llseek         = generic_file_llseek,
2621 };
2622
2623 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2624                                 struct dentry *dentry, unsigned int flags)
2625 {
2626         return proc_pident_lookup(dir, dentry,
2627                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2628 }
2629
2630 static const struct inode_operations proc_attr_dir_inode_operations = {
2631         .lookup         = proc_attr_dir_lookup,
2632         .getattr        = pid_getattr,
2633         .setattr        = proc_setattr,
2634 };
2635
2636 #endif
2637
2638 #ifdef CONFIG_ELF_CORE
2639 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2640                                          size_t count, loff_t *ppos)
2641 {
2642         struct task_struct *task = get_proc_task(file_inode(file));
2643         struct mm_struct *mm;
2644         char buffer[PROC_NUMBUF];
2645         size_t len;
2646         int ret;
2647
2648         if (!task)
2649                 return -ESRCH;
2650
2651         ret = 0;
2652         mm = get_task_mm(task);
2653         if (mm) {
2654                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2655                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2656                                 MMF_DUMP_FILTER_SHIFT));
2657                 mmput(mm);
2658                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2659         }
2660
2661         put_task_struct(task);
2662
2663         return ret;
2664 }
2665
2666 static ssize_t proc_coredump_filter_write(struct file *file,
2667                                           const char __user *buf,
2668                                           size_t count,
2669                                           loff_t *ppos)
2670 {
2671         struct task_struct *task;
2672         struct mm_struct *mm;
2673         unsigned int val;
2674         int ret;
2675         int i;
2676         unsigned long mask;
2677
2678         ret = kstrtouint_from_user(buf, count, 0, &val);
2679         if (ret < 0)
2680                 return ret;
2681
2682         ret = -ESRCH;
2683         task = get_proc_task(file_inode(file));
2684         if (!task)
2685                 goto out_no_task;
2686
2687         mm = get_task_mm(task);
2688         if (!mm)
2689                 goto out_no_mm;
2690         ret = 0;
2691
2692         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2693                 if (val & mask)
2694                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2695                 else
2696                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2697         }
2698
2699         mmput(mm);
2700  out_no_mm:
2701         put_task_struct(task);
2702  out_no_task:
2703         if (ret < 0)
2704                 return ret;
2705         return count;
2706 }
2707
2708 static const struct file_operations proc_coredump_filter_operations = {
2709         .read           = proc_coredump_filter_read,
2710         .write          = proc_coredump_filter_write,
2711         .llseek         = generic_file_llseek,
2712 };
2713 #endif
2714
2715 #ifdef CONFIG_TASK_IO_ACCOUNTING
2716 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2717 {
2718         struct task_io_accounting acct = task->ioac;
2719         unsigned long flags;
2720         int result;
2721
2722         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2723         if (result)
2724                 return result;
2725
2726         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2727                 result = -EACCES;
2728                 goto out_unlock;
2729         }
2730
2731         if (whole && lock_task_sighand(task, &flags)) {
2732                 struct task_struct *t = task;
2733
2734                 task_io_accounting_add(&acct, &task->signal->ioac);
2735                 while_each_thread(task, t)
2736                         task_io_accounting_add(&acct, &t->ioac);
2737
2738                 unlock_task_sighand(task, &flags);
2739         }
2740         seq_printf(m,
2741                    "rchar: %llu\n"
2742                    "wchar: %llu\n"
2743                    "syscr: %llu\n"
2744                    "syscw: %llu\n"
2745                    "read_bytes: %llu\n"
2746                    "write_bytes: %llu\n"
2747                    "cancelled_write_bytes: %llu\n",
2748                    (unsigned long long)acct.rchar,
2749                    (unsigned long long)acct.wchar,
2750                    (unsigned long long)acct.syscr,
2751                    (unsigned long long)acct.syscw,
2752                    (unsigned long long)acct.read_bytes,
2753                    (unsigned long long)acct.write_bytes,
2754                    (unsigned long long)acct.cancelled_write_bytes);
2755         result = 0;
2756
2757 out_unlock:
2758         mutex_unlock(&task->signal->cred_guard_mutex);
2759         return result;
2760 }
2761
2762 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2763                                   struct pid *pid, struct task_struct *task)
2764 {
2765         return do_io_accounting(task, m, 0);
2766 }
2767
2768 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2769                                    struct pid *pid, struct task_struct *task)
2770 {
2771         return do_io_accounting(task, m, 1);
2772 }
2773 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2774
2775 #ifdef CONFIG_USER_NS
2776 static int proc_id_map_open(struct inode *inode, struct file *file,
2777         const struct seq_operations *seq_ops)
2778 {
2779         struct user_namespace *ns = NULL;
2780         struct task_struct *task;
2781         struct seq_file *seq;
2782         int ret = -EINVAL;
2783
2784         task = get_proc_task(inode);
2785         if (task) {
2786                 rcu_read_lock();
2787                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2788                 rcu_read_unlock();
2789                 put_task_struct(task);
2790         }
2791         if (!ns)
2792                 goto err;
2793
2794         ret = seq_open(file, seq_ops);
2795         if (ret)
2796                 goto err_put_ns;
2797
2798         seq = file->private_data;
2799         seq->private = ns;
2800
2801         return 0;
2802 err_put_ns:
2803         put_user_ns(ns);
2804 err:
2805         return ret;
2806 }
2807
2808 static int proc_id_map_release(struct inode *inode, struct file *file)
2809 {
2810         struct seq_file *seq = file->private_data;
2811         struct user_namespace *ns = seq->private;
2812         put_user_ns(ns);
2813         return seq_release(inode, file);
2814 }
2815
2816 static int proc_uid_map_open(struct inode *inode, struct file *file)
2817 {
2818         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2819 }
2820
2821 static int proc_gid_map_open(struct inode *inode, struct file *file)
2822 {
2823         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2824 }
2825
2826 static int proc_projid_map_open(struct inode *inode, struct file *file)
2827 {
2828         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2829 }
2830
2831 static const struct file_operations proc_uid_map_operations = {
2832         .open           = proc_uid_map_open,
2833         .write          = proc_uid_map_write,
2834         .read           = seq_read,
2835         .llseek         = seq_lseek,
2836         .release        = proc_id_map_release,
2837 };
2838
2839 static const struct file_operations proc_gid_map_operations = {
2840         .open           = proc_gid_map_open,
2841         .write          = proc_gid_map_write,
2842         .read           = seq_read,
2843         .llseek         = seq_lseek,
2844         .release        = proc_id_map_release,
2845 };
2846
2847 static const struct file_operations proc_projid_map_operations = {
2848         .open           = proc_projid_map_open,
2849         .write          = proc_projid_map_write,
2850         .read           = seq_read,
2851         .llseek         = seq_lseek,
2852         .release        = proc_id_map_release,
2853 };
2854
2855 static int proc_setgroups_open(struct inode *inode, struct file *file)
2856 {
2857         struct user_namespace *ns = NULL;
2858         struct task_struct *task;
2859         int ret;
2860
2861         ret = -ESRCH;
2862         task = get_proc_task(inode);
2863         if (task) {
2864                 rcu_read_lock();
2865                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2866                 rcu_read_unlock();
2867                 put_task_struct(task);
2868         }
2869         if (!ns)
2870                 goto err;
2871
2872         if (file->f_mode & FMODE_WRITE) {
2873                 ret = -EACCES;
2874                 if (!ns_capable(ns, CAP_SYS_ADMIN))
2875                         goto err_put_ns;
2876         }
2877
2878         ret = single_open(file, &proc_setgroups_show, ns);
2879         if (ret)
2880                 goto err_put_ns;
2881
2882         return 0;
2883 err_put_ns:
2884         put_user_ns(ns);
2885 err:
2886         return ret;
2887 }
2888
2889 static int proc_setgroups_release(struct inode *inode, struct file *file)
2890 {
2891         struct seq_file *seq = file->private_data;
2892         struct user_namespace *ns = seq->private;
2893         int ret = single_release(inode, file);
2894         put_user_ns(ns);
2895         return ret;
2896 }
2897
2898 static const struct file_operations proc_setgroups_operations = {
2899         .open           = proc_setgroups_open,
2900         .write          = proc_setgroups_write,
2901         .read           = seq_read,
2902         .llseek         = seq_lseek,
2903         .release        = proc_setgroups_release,
2904 };
2905 #endif /* CONFIG_USER_NS */
2906
2907 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2908                                 struct pid *pid, struct task_struct *task)
2909 {
2910         int err = lock_trace(task);
2911         if (!err) {
2912                 seq_printf(m, "%08x\n", task->personality);
2913                 unlock_trace(task);
2914         }
2915         return err;
2916 }
2917
2918 #ifdef CONFIG_LIVEPATCH
2919 static int proc_pid_patch_state(struct seq_file *m, struct pid_namespace *ns,
2920                                 struct pid *pid, struct task_struct *task)
2921 {
2922         seq_printf(m, "%d\n", task->patch_state);
2923         return 0;
2924 }
2925 #endif /* CONFIG_LIVEPATCH */
2926
2927 /*
2928  * Thread groups
2929  */
2930 static const struct file_operations proc_task_operations;
2931 static const struct inode_operations proc_task_inode_operations;
2932
2933 static const struct pid_entry tgid_base_stuff[] = {
2934         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2935         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2936         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2937         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2938         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2939 #ifdef CONFIG_NET
2940         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2941 #endif
2942         REG("environ",    S_IRUSR, proc_environ_operations),
2943         REG("auxv",       S_IRUSR, proc_auxv_operations),
2944         ONE("status",     S_IRUGO, proc_pid_status),
2945         ONE("personality", S_IRUSR, proc_pid_personality),
2946         ONE("limits",     S_IRUGO, proc_pid_limits),
2947 #ifdef CONFIG_SCHED_DEBUG
2948         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2949 #endif
2950 #ifdef CONFIG_SCHED_AUTOGROUP
2951         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2952 #endif
2953         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2954 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2955         ONE("syscall",    S_IRUSR, proc_pid_syscall),
2956 #endif
2957         REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
2958         ONE("stat",       S_IRUGO, proc_tgid_stat),
2959         ONE("statm",      S_IRUGO, proc_pid_statm),
2960         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2961 #ifdef CONFIG_NUMA
2962         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2963 #endif
2964         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2965         LNK("cwd",        proc_cwd_link),
2966         LNK("root",       proc_root_link),
2967         LNK("exe",        proc_exe_link),
2968         REG("mounts",     S_IRUGO, proc_mounts_operations),
2969         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2970         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2971 #ifdef CONFIG_PROC_PAGE_MONITOR
2972         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2973         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2974         REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
2975         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2976 #endif
2977 #ifdef CONFIG_SECURITY
2978         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2979 #endif
2980 #ifdef CONFIG_KALLSYMS
2981         ONE("wchan",      S_IRUGO, proc_pid_wchan),
2982 #endif
2983 #ifdef CONFIG_STACKTRACE
2984         ONE("stack",      S_IRUSR, proc_pid_stack),
2985 #endif
2986 #ifdef CONFIG_SCHED_INFO
2987         ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
2988 #endif
2989 #ifdef CONFIG_LATENCYTOP
2990         REG("latency",  S_IRUGO, proc_lstats_operations),
2991 #endif
2992 #ifdef CONFIG_PROC_PID_CPUSET
2993         ONE("cpuset",     S_IRUGO, proc_cpuset_show),
2994 #endif
2995 #ifdef CONFIG_CGROUPS
2996         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
2997 #endif
2998         ONE("oom_score",  S_IRUGO, proc_oom_score),
2999         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3000         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3001 #ifdef CONFIG_AUDITSYSCALL
3002         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
3003         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3004 #endif
3005 #ifdef CONFIG_FAULT_INJECTION
3006         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3007         REG("fail-nth", 0644, proc_fail_nth_operations),
3008 #endif
3009 #ifdef CONFIG_ELF_CORE
3010         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3011 #endif
3012 #ifdef CONFIG_TASK_IO_ACCOUNTING
3013         ONE("io",       S_IRUSR, proc_tgid_io_accounting),
3014 #endif
3015 #ifdef CONFIG_USER_NS
3016         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3017         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3018         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3019         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3020 #endif
3021 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
3022         REG("timers",     S_IRUGO, proc_timers_operations),
3023 #endif
3024         REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
3025 #ifdef CONFIG_LIVEPATCH
3026         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
3027 #endif
3028 };
3029
3030 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
3031 {
3032         return proc_pident_readdir(file, ctx,
3033                                    tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3034 }
3035
3036 static const struct file_operations proc_tgid_base_operations = {
3037         .read           = generic_read_dir,
3038         .iterate_shared = proc_tgid_base_readdir,
3039         .llseek         = generic_file_llseek,
3040 };
3041
3042 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3043 {
3044         return proc_pident_lookup(dir, dentry,
3045                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3046 }
3047
3048 static const struct inode_operations proc_tgid_base_inode_operations = {
3049         .lookup         = proc_tgid_base_lookup,
3050         .getattr        = pid_getattr,
3051         .setattr        = proc_setattr,
3052         .permission     = proc_pid_permission,
3053 };
3054
3055 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
3056 {
3057         struct dentry *dentry, *leader, *dir;
3058         char buf[10 + 1];
3059         struct qstr name;
3060
3061         name.name = buf;
3062         name.len = snprintf(buf, sizeof(buf), "%u", pid);
3063         /* no ->d_hash() rejects on procfs */
3064         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
3065         if (dentry) {
3066                 d_invalidate(dentry);
3067                 dput(dentry);
3068         }
3069
3070         if (pid == tgid)
3071                 return;
3072
3073         name.name = buf;
3074         name.len = snprintf(buf, sizeof(buf), "%u", tgid);
3075         leader = d_hash_and_lookup(mnt->mnt_root, &name);
3076         if (!leader)
3077                 goto out;
3078
3079         name.name = "task";
3080         name.len = strlen(name.name);
3081         dir = d_hash_and_lookup(leader, &name);
3082         if (!dir)
3083                 goto out_put_leader;
3084
3085         name.name = buf;
3086         name.len = snprintf(buf, sizeof(buf), "%u", pid);
3087         dentry = d_hash_and_lookup(dir, &name);
3088         if (dentry) {
3089                 d_invalidate(dentry);
3090                 dput(dentry);
3091         }
3092
3093         dput(dir);
3094 out_put_leader:
3095         dput(leader);
3096 out:
3097         return;
3098 }
3099
3100 /**
3101  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
3102  * @task: task that should be flushed.
3103  *
3104  * When flushing dentries from proc, one needs to flush them from global
3105  * proc (proc_mnt) and from all the namespaces' procs this task was seen
3106  * in. This call is supposed to do all of this job.
3107  *
3108  * Looks in the dcache for
3109  * /proc/@pid
3110  * /proc/@tgid/task/@pid
3111  * if either directory is present flushes it and all of it'ts children
3112  * from the dcache.
3113  *
3114  * It is safe and reasonable to cache /proc entries for a task until
3115  * that task exits.  After that they just clog up the dcache with
3116  * useless entries, possibly causing useful dcache entries to be
3117  * flushed instead.  This routine is proved to flush those useless
3118  * dcache entries at process exit time.
3119  *
3120  * NOTE: This routine is just an optimization so it does not guarantee
3121  *       that no dcache entries will exist at process exit time it
3122  *       just makes it very unlikely that any will persist.
3123  */
3124
3125 void proc_flush_task(struct task_struct *task)
3126 {
3127         int i;
3128         struct pid *pid, *tgid;
3129         struct upid *upid;
3130
3131         pid = task_pid(task);
3132         tgid = task_tgid(task);
3133
3134         for (i = 0; i <= pid->level; i++) {
3135                 upid = &pid->numbers[i];
3136                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3137                                         tgid->numbers[i].nr);
3138         }
3139 }
3140
3141 static int proc_pid_instantiate(struct inode *dir,
3142                                    struct dentry * dentry,
3143                                    struct task_struct *task, const void *ptr)
3144 {
3145         struct inode *inode;
3146
3147         inode = proc_pid_make_inode(dir->i_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3148         if (!inode)
3149                 goto out;
3150
3151         inode->i_op = &proc_tgid_base_inode_operations;
3152         inode->i_fop = &proc_tgid_base_operations;
3153         inode->i_flags|=S_IMMUTABLE;
3154
3155         set_nlink(inode, nlink_tgid);
3156
3157         d_set_d_op(dentry, &pid_dentry_operations);
3158
3159         d_add(dentry, inode);
3160         /* Close the race of the process dying before we return the dentry */
3161         if (pid_revalidate(dentry, 0))
3162                 return 0;
3163 out:
3164         return -ENOENT;
3165 }
3166
3167 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3168 {
3169         int result = -ENOENT;
3170         struct task_struct *task;
3171         unsigned tgid;
3172         struct pid_namespace *ns;
3173
3174         tgid = name_to_int(&dentry->d_name);
3175         if (tgid == ~0U)
3176                 goto out;
3177
3178         ns = dentry->d_sb->s_fs_info;
3179         rcu_read_lock();
3180         task = find_task_by_pid_ns(tgid, ns);
3181         if (task)
3182                 get_task_struct(task);
3183         rcu_read_unlock();
3184         if (!task)
3185                 goto out;
3186
3187         result = proc_pid_instantiate(dir, dentry, task, NULL);
3188         put_task_struct(task);
3189 out:
3190         return ERR_PTR(result);
3191 }
3192
3193 /*
3194  * Find the first task with tgid >= tgid
3195  *
3196  */
3197 struct tgid_iter {
3198         unsigned int tgid;
3199         struct task_struct *task;
3200 };
3201 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3202 {
3203         struct pid *pid;
3204
3205         if (iter.task)
3206                 put_task_struct(iter.task);
3207         rcu_read_lock();
3208 retry:
3209         iter.task = NULL;
3210         pid = find_ge_pid(iter.tgid, ns);
3211         if (pid) {
3212                 iter.tgid = pid_nr_ns(pid, ns);
3213                 iter.task = pid_task(pid, PIDTYPE_PID);
3214                 /* What we to know is if the pid we have find is the
3215                  * pid of a thread_group_leader.  Testing for task
3216                  * being a thread_group_leader is the obvious thing
3217                  * todo but there is a window when it fails, due to
3218                  * the pid transfer logic in de_thread.
3219                  *
3220                  * So we perform the straight forward test of seeing
3221                  * if the pid we have found is the pid of a thread
3222                  * group leader, and don't worry if the task we have
3223                  * found doesn't happen to be a thread group leader.
3224                  * As we don't care in the case of readdir.
3225                  */
3226                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3227                         iter.tgid += 1;
3228                         goto retry;
3229                 }
3230                 get_task_struct(iter.task);
3231         }
3232         rcu_read_unlock();
3233         return iter;
3234 }
3235
3236 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3237
3238 /* for the /proc/ directory itself, after non-process stuff has been done */
3239 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3240 {
3241         struct tgid_iter iter;
3242         struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
3243         loff_t pos = ctx->pos;
3244
3245         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3246                 return 0;
3247
3248         if (pos == TGID_OFFSET - 2) {
3249                 struct inode *inode = d_inode(ns->proc_self);
3250                 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3251                         return 0;
3252                 ctx->pos = pos = pos + 1;
3253         }
3254         if (pos == TGID_OFFSET - 1) {
3255                 struct inode *inode = d_inode(ns->proc_thread_self);
3256                 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3257                         return 0;
3258                 ctx->pos = pos = pos + 1;
3259         }
3260         iter.tgid = pos - TGID_OFFSET;
3261         iter.task = NULL;
3262         for (iter = next_tgid(ns, iter);
3263              iter.task;
3264              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3265                 char name[10 + 1];
3266                 int len;
3267
3268                 cond_resched();
3269                 if (!has_pid_permissions(ns, iter.task, HIDEPID_INVISIBLE))
3270                         continue;
3271
3272                 len = snprintf(name, sizeof(name), "%u", iter.tgid);
3273                 ctx->pos = iter.tgid + TGID_OFFSET;
3274                 if (!proc_fill_cache(file, ctx, name, len,
3275                                      proc_pid_instantiate, iter.task, NULL)) {
3276                         put_task_struct(iter.task);
3277                         return 0;
3278                 }
3279         }
3280         ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3281         return 0;
3282 }
3283
3284 /*
3285  * proc_tid_comm_permission is a special permission function exclusively
3286  * used for the node /proc/<pid>/task/<tid>/comm.
3287  * It bypasses generic permission checks in the case where a task of the same
3288  * task group attempts to access the node.
3289  * The rationale behind this is that glibc and bionic access this node for
3290  * cross thread naming (pthread_set/getname_np(!self)). However, if
3291  * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3292  * which locks out the cross thread naming implementation.
3293  * This function makes sure that the node is always accessible for members of
3294  * same thread group.
3295  */
3296 static int proc_tid_comm_permission(struct inode *inode, int mask)
3297 {
3298         bool is_same_tgroup;
3299         struct task_struct *task;
3300
3301         task = get_proc_task(inode);
3302         if (!task)
3303                 return -ESRCH;
3304         is_same_tgroup = same_thread_group(current, task);
3305         put_task_struct(task);
3306
3307         if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3308                 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3309                  * read or written by the members of the corresponding
3310                  * thread group.
3311                  */
3312                 return 0;
3313         }
3314
3315         return generic_permission(inode, mask);
3316 }
3317
3318 static const struct inode_operations proc_tid_comm_inode_operations = {
3319                 .permission = proc_tid_comm_permission,
3320 };
3321
3322 /*
3323  * Tasks
3324  */
3325 static const struct pid_entry tid_base_stuff[] = {
3326         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3327         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3328         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3329 #ifdef CONFIG_NET
3330         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3331 #endif
3332         REG("environ",   S_IRUSR, proc_environ_operations),
3333         REG("auxv",      S_IRUSR, proc_auxv_operations),
3334         ONE("status",    S_IRUGO, proc_pid_status),
3335         ONE("personality", S_IRUSR, proc_pid_personality),
3336         ONE("limits",    S_IRUGO, proc_pid_limits),
3337 #ifdef CONFIG_SCHED_DEBUG
3338         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3339 #endif
3340         NOD("comm",      S_IFREG|S_IRUGO|S_IWUSR,
3341                          &proc_tid_comm_inode_operations,
3342                          &proc_pid_set_comm_operations, {}),
3343 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3344         ONE("syscall",   S_IRUSR, proc_pid_syscall),
3345 #endif
3346         REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3347         ONE("stat",      S_IRUGO, proc_tid_stat),
3348         ONE("statm",     S_IRUGO, proc_pid_statm),
3349         REG("maps",      S_IRUGO, proc_tid_maps_operations),
3350 #ifdef CONFIG_PROC_CHILDREN
3351         REG("children",  S_IRUGO, proc_tid_children_operations),
3352 #endif
3353 #ifdef CONFIG_NUMA
3354         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3355 #endif
3356         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3357         LNK("cwd",       proc_cwd_link),
3358         LNK("root",      proc_root_link),
3359         LNK("exe",       proc_exe_link),
3360         REG("mounts",    S_IRUGO, proc_mounts_operations),
3361         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3362 #ifdef CONFIG_PROC_PAGE_MONITOR
3363         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3364         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3365         REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3366         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3367 #endif
3368 #ifdef CONFIG_SECURITY
3369         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3370 #endif
3371 #ifdef CONFIG_KALLSYMS
3372         ONE("wchan",     S_IRUGO, proc_pid_wchan),
3373 #endif
3374 #ifdef CONFIG_STACKTRACE
3375         ONE("stack",      S_IRUSR, proc_pid_stack),
3376 #endif
3377 #ifdef CONFIG_SCHED_INFO
3378         ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3379 #endif
3380 #ifdef CONFIG_LATENCYTOP
3381         REG("latency",  S_IRUGO, proc_lstats_operations),
3382 #endif
3383 #ifdef CONFIG_PROC_PID_CPUSET
3384         ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3385 #endif
3386 #ifdef CONFIG_CGROUPS
3387         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3388 #endif
3389         ONE("oom_score", S_IRUGO, proc_oom_score),
3390         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3391         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3392 #ifdef CONFIG_AUDITSYSCALL
3393         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3394         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3395 #endif
3396 #ifdef CONFIG_FAULT_INJECTION
3397         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3398         REG("fail-nth", 0644, proc_fail_nth_operations),
3399 #endif
3400 #ifdef CONFIG_TASK_IO_ACCOUNTING
3401         ONE("io",       S_IRUSR, proc_tid_io_accounting),
3402 #endif
3403 #ifdef CONFIG_USER_NS
3404         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3405         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3406         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3407         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3408 #endif
3409 #ifdef CONFIG_LIVEPATCH
3410         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
3411 #endif
3412 };
3413
3414 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3415 {
3416         return proc_pident_readdir(file, ctx,
3417                                    tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3418 }
3419
3420 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3421 {
3422         return proc_pident_lookup(dir, dentry,
3423                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3424 }
3425
3426 static const struct file_operations proc_tid_base_operations = {
3427         .read           = generic_read_dir,
3428         .iterate_shared = proc_tid_base_readdir,
3429         .llseek         = generic_file_llseek,
3430 };
3431
3432 static const struct inode_operations proc_tid_base_inode_operations = {
3433         .lookup         = proc_tid_base_lookup,
3434         .getattr        = pid_getattr,
3435         .setattr        = proc_setattr,
3436 };
3437
3438 static int proc_task_instantiate(struct inode *dir,
3439         struct dentry *dentry, struct task_struct *task, const void *ptr)
3440 {
3441         struct inode *inode;
3442         inode = proc_pid_make_inode(dir->i_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3443
3444         if (!inode)
3445                 goto out;
3446         inode->i_op = &proc_tid_base_inode_operations;
3447         inode->i_fop = &proc_tid_base_operations;
3448         inode->i_flags|=S_IMMUTABLE;
3449
3450         set_nlink(inode, nlink_tid);
3451
3452         d_set_d_op(dentry, &pid_dentry_operations);
3453
3454         d_add(dentry, inode);
3455         /* Close the race of the process dying before we return the dentry */
3456         if (pid_revalidate(dentry, 0))
3457                 return 0;
3458 out:
3459         return -ENOENT;
3460 }
3461
3462 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3463 {
3464         int result = -ENOENT;
3465         struct task_struct *task;
3466         struct task_struct *leader = get_proc_task(dir);
3467         unsigned tid;
3468         struct pid_namespace *ns;
3469
3470         if (!leader)
3471                 goto out_no_task;
3472
3473         tid = name_to_int(&dentry->d_name);
3474         if (tid == ~0U)
3475                 goto out;
3476
3477         ns = dentry->d_sb->s_fs_info;
3478         rcu_read_lock();
3479         task = find_task_by_pid_ns(tid, ns);
3480         if (task)
3481                 get_task_struct(task);
3482         rcu_read_unlock();
3483         if (!task)
3484                 goto out;
3485         if (!same_thread_group(leader, task))
3486                 goto out_drop_task;
3487
3488         result = proc_task_instantiate(dir, dentry, task, NULL);
3489 out_drop_task:
3490         put_task_struct(task);
3491 out:
3492         put_task_struct(leader);
3493 out_no_task:
3494         return ERR_PTR(result);
3495 }
3496
3497 /*
3498  * Find the first tid of a thread group to return to user space.
3499  *
3500  * Usually this is just the thread group leader, but if the users
3501  * buffer was too small or there was a seek into the middle of the
3502  * directory we have more work todo.
3503  *
3504  * In the case of a short read we start with find_task_by_pid.
3505  *
3506  * In the case of a seek we start with the leader and walk nr
3507  * threads past it.
3508  */
3509 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3510                                         struct pid_namespace *ns)
3511 {
3512         struct task_struct *pos, *task;
3513         unsigned long nr = f_pos;
3514
3515         if (nr != f_pos)        /* 32bit overflow? */
3516                 return NULL;
3517
3518         rcu_read_lock();
3519         task = pid_task(pid, PIDTYPE_PID);
3520         if (!task)
3521                 goto fail;
3522
3523         /* Attempt to start with the tid of a thread */
3524         if (tid && nr) {
3525                 pos = find_task_by_pid_ns(tid, ns);
3526                 if (pos && same_thread_group(pos, task))
3527                         goto found;
3528         }
3529
3530         /* If nr exceeds the number of threads there is nothing todo */
3531         if (nr >= get_nr_threads(task))
3532                 goto fail;
3533
3534         /* If we haven't found our starting place yet start
3535          * with the leader and walk nr threads forward.
3536          */
3537         pos = task = task->group_leader;
3538         do {
3539                 if (!nr--)
3540                         goto found;
3541         } while_each_thread(task, pos);
3542 fail:
3543         pos = NULL;
3544         goto out;
3545 found:
3546         get_task_struct(pos);
3547 out:
3548         rcu_read_unlock();
3549         return pos;
3550 }
3551
3552 /*
3553  * Find the next thread in the thread list.
3554  * Return NULL if there is an error or no next thread.
3555  *
3556  * The reference to the input task_struct is released.
3557  */
3558 static struct task_struct *next_tid(struct task_struct *start)
3559 {
3560         struct task_struct *pos = NULL;
3561         rcu_read_lock();
3562         if (pid_alive(start)) {
3563                 pos = next_thread(start);
3564                 if (thread_group_leader(pos))
3565                         pos = NULL;
3566                 else
3567                         get_task_struct(pos);
3568         }
3569         rcu_read_unlock();
3570         put_task_struct(start);
3571         return pos;
3572 }
3573
3574 /* for the /proc/TGID/task/ directories */
3575 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3576 {
3577         struct inode *inode = file_inode(file);
3578         struct task_struct *task;
3579         struct pid_namespace *ns;
3580         int tid;
3581
3582         if (proc_inode_is_dead(inode))
3583                 return -ENOENT;
3584
3585         if (!dir_emit_dots(file, ctx))
3586                 return 0;
3587
3588         /* f_version caches the tgid value that the last readdir call couldn't
3589          * return. lseek aka telldir automagically resets f_version to 0.
3590          */
3591         ns = inode->i_sb->s_fs_info;
3592         tid = (int)file->f_version;
3593         file->f_version = 0;
3594         for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3595              task;
3596              task = next_tid(task), ctx->pos++) {
3597                 char name[10 + 1];
3598                 int len;
3599                 tid = task_pid_nr_ns(task, ns);
3600                 len = snprintf(name, sizeof(name), "%u", tid);
3601                 if (!proc_fill_cache(file, ctx, name, len,
3602                                 proc_task_instantiate, task, NULL)) {
3603                         /* returning this tgid failed, save it as the first
3604                          * pid for the next readir call */
3605                         file->f_version = (u64)tid;
3606                         put_task_struct(task);
3607                         break;
3608                 }
3609         }
3610
3611         return 0;
3612 }
3613
3614 static int proc_task_getattr(const struct path *path, struct kstat *stat,
3615                              u32 request_mask, unsigned int query_flags)
3616 {
3617         struct inode *inode = d_inode(path->dentry);
3618         struct task_struct *p = get_proc_task(inode);
3619         generic_fillattr(inode, stat);
3620
3621         if (p) {
3622                 stat->nlink += get_nr_threads(p);
3623                 put_task_struct(p);
3624         }
3625
3626         return 0;
3627 }
3628
3629 static const struct inode_operations proc_task_inode_operations = {
3630         .lookup         = proc_task_lookup,
3631         .getattr        = proc_task_getattr,
3632         .setattr        = proc_setattr,
3633         .permission     = proc_pid_permission,
3634 };
3635
3636 static const struct file_operations proc_task_operations = {
3637         .read           = generic_read_dir,
3638         .iterate_shared = proc_task_readdir,
3639         .llseek         = generic_file_llseek,
3640 };
3641
3642 void __init set_proc_pid_nlink(void)
3643 {
3644         nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3645         nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3646 }