fs/proc/base.c

   1 /*
   2  *  linux/fs/proc/base.c
   3  *
   4  *  Copyright (C) 1991, 1992 Linus Torvalds
   5  *
   6  *  proc base directory handling functions
   7  *
   8  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
   9  *  Instead of using magical inumbers to determine the kind of object
  10  *  we allocate and fill in-core inodes upon lookup. They don't even
  11  *  go into icache. We cache the reference to task_struct upon lookup too.
  12  *  Eventually it should become a filesystem in its own. We don't use the
  13  *  rest of procfs anymore.
  14  *
  15  *
  16  *  Changelog:
  17  *  17-Jan-2005
  18  *  Allan Bezerra
  19  *  Bruna Moreira <bruna.moreira@indt.org.br>
  20  *  Edjard Mota <edjard.mota@indt.org.br>
  21  *  Ilias Biris <ilias.biris@indt.org.br>
  22  *  Mauricio Lin <mauricio.lin@indt.org.br>
  23  *
  24  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  25  *
  26  *  A new process specific entry (smaps) included in /proc. It shows the
  27  *  size of rss for each memory area. The maps entry lacks information
  28  *  about physical memory size (rss) for each mapped file, i.e.,
  29  *  rss information for executables and library files.
  30  *  This additional information is useful for any tools that need to know
  31  *  about physical memory consumption for a process specific library.
  32  *
  33  *  Changelog:
  34  *  21-Feb-2005
  35  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  36  *  Pud inclusion in the page table walking.
  37  *
  38  *  ChangeLog:
  39  *  10-Mar-2005
  40  *  10LE Instituto Nokia de Tecnologia - INdT:
  41  *  A better way to walks through the page table as suggested by Hugh Dickins.
  42  *
  43  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
  44  *  Smaps information related to shared, private, clean and dirty pages.
  45  *
  46  *  Paul Mundt <paul.mundt@nokia.com>:
  47  *  Overall revision about smaps.
  48  */
  49
  50 #include <asm/uaccess.h>
  51
  52 #include <linux/errno.h>
  53 #include <linux/time.h>
  54 #include <linux/proc_fs.h>
  55 #include <linux/stat.h>
  56 #include <linux/task_io_accounting_ops.h>
  57 #include <linux/init.h>
  58 #include <linux/capability.h>
  59 #include <linux/file.h>
  60 #include <linux/fdtable.h>
  61 #include <linux/string.h>
  62 #include <linux/seq_file.h>
  63 #include <linux/namei.h>
  64 #include <linux/mnt_namespace.h>
  65 #include <linux/mm.h>
  66 #include <linux/swap.h>
  67 #include <linux/rcupdate.h>
  68 #include <linux/kallsyms.h>
  69 #include <linux/stacktrace.h>
  70 #include <linux/resource.h>
  71 #include <linux/module.h>
  72 #include <linux/mount.h>
  73 #include <linux/security.h>
  74 #include <linux/ptrace.h>
  75 #include <linux/tracehook.h>
  76 #include <linux/printk.h>
  77 #include <linux/cgroup.h>
  78 #include <linux/cpuset.h>
  79 #include <linux/audit.h>
  80 #include <linux/poll.h>
  81 #include <linux/nsproxy.h>
  82 #include <linux/oom.h>
  83 #include <linux/elf.h>
  84 #include <linux/pid_namespace.h>
  85 #include <linux/user_namespace.h>
  86 #include <linux/fs_struct.h>
  87 #include <linux/slab.h>
  88 #include <linux/flex_array.h>
  89 #include <linux/posix-timers.h>
  90 #ifdef CONFIG_HARDWALL
  91 #include <asm/hardwall.h>
  92 #endif
  93 #include <trace/events/oom.h>
  94 #include "internal.h"
  95 #include "fd.h"
  96
  97 /* NOTE:
  98  *      Implementing inode permission operations in /proc is almost
  99  *      certainly an error.  Permission checks need to happen during
 100  *      each system call not at open time.  The reason is that most of
 101  *      what we wish to check for permissions in /proc varies at runtime.
 102  *
 103  *      The classic example of a problem is opening file descriptors
 104  *      in /proc for a task before it execs a suid executable.
 105  */
 106
 107 struct pid_entry {
 108         const char *name;
 109         int len;
 110         umode_t mode;
 111         const struct inode_operations *iop;
 112         const struct file_operations *fop;
 113         union proc_op op;
 114 };
 115
 116 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 117         .name = (NAME),                                 \
 118         .len  = sizeof(NAME) - 1,                       \
 119         .mode = MODE,                                   \
 120         .iop  = IOP,                                    \
 121         .fop  = FOP,                                    \
 122         .op   = OP,                                     \
 123 }
 124
 125 #define DIR(NAME, MODE, iops, fops)     \
 126         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 127 #define LNK(NAME, get_link)                                     \
 128         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 129                 &proc_pid_link_inode_operations, NULL,          \
 130                 { .proc_get_link = get_link } )
 131 #define REG(NAME, MODE, fops)                           \
 132         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 133 #define ONE(NAME, MODE, show)                           \
 134         NOD(NAME, (S_IFREG|(MODE)),                     \
 135                 NULL, &proc_single_file_operations,     \
 136                 { .proc_show = show } )
 137
 138 /*
 139  * Count the number of hardlinks for the pid_entry table, excluding the .
 140  * and .. links.
 141  */
 142 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
 143         unsigned int n)
 144 {
 145         unsigned int i;
 146         unsigned int count;
 147
 148         count = 0;
 149         for (i = 0; i < n; ++i) {
 150                 if (S_ISDIR(entries[i].mode))
 151                         ++count;
 152         }
 153
 154         return count;
 155 }
 156
 157 static int get_task_root(struct task_struct *task, struct path *root)
 158 {
 159         int result = -ENOENT;
 160
 161         task_lock(task);
 162         if (task->fs) {
 163                 get_fs_root(task->fs, root);
 164                 result = 0;
 165         }
 166         task_unlock(task);
 167         return result;
 168 }
 169
 170 static int proc_cwd_link(struct dentry *dentry, struct path *path)
 171 {
 172         struct task_struct *task = get_proc_task(d_inode(dentry));
 173         int result = -ENOENT;
 174
 175         if (task) {
 176                 task_lock(task);
 177                 if (task->fs) {
 178                         get_fs_pwd(task->fs, path);
 179                         result = 0;
 180                 }
 181                 task_unlock(task);
 182                 put_task_struct(task);
 183         }
 184         return result;
 185 }
 186
 187 static int proc_root_link(struct dentry *dentry, struct path *path)
 188 {
 189         struct task_struct *task = get_proc_task(d_inode(dentry));
 190         int result = -ENOENT;
 191
 192         if (task) {
 193                 result = get_task_root(task, path);
 194                 put_task_struct(task);
 195         }
 196         return result;
 197 }
 198
 199 static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
 200                                      size_t _count, loff_t *pos)
 201 {
 202         struct task_struct *tsk;
 203         struct mm_struct *mm;
 204         char *page;
 205         unsigned long count = _count;
 206         unsigned long arg_start, arg_end, env_start, env_end;
 207         unsigned long len1, len2, len;
 208         unsigned long p;
 209         char c;
 210         ssize_t rv;
 211
 212         BUG_ON(*pos < 0);
 213
 214         tsk = get_proc_task(file_inode(file));
 215         if (!tsk)
 216                 return -ESRCH;
 217         mm = get_task_mm(tsk);
 218         put_task_struct(tsk);
 219         if (!mm)
 220                 return 0;
 221         /* Check if process spawned far enough to have cmdline. */
 222         if (!mm->env_end) {
 223                 rv = 0;
 224                 goto out_mmput;
 225         }
 226
 227         page = (char *)__get_free_page(GFP_TEMPORARY);
 228         if (!page) {
 229                 rv = -ENOMEM;
 230                 goto out_mmput;
 231         }
 232
 233         down_read(&mm->mmap_sem);
 234         arg_start = mm->arg_start;
 235         arg_end = mm->arg_end;
 236         env_start = mm->env_start;
 237         env_end = mm->env_end;
 238         up_read(&mm->mmap_sem);
 239
 240         BUG_ON(arg_start > arg_end);
 241         BUG_ON(env_start > env_end);
 242
 243         len1 = arg_end - arg_start;
 244         len2 = env_end - env_start;
 245
 246         /* Empty ARGV. */
 247         if (len1 == 0) {
 248                 rv = 0;
 249                 goto out_free_page;
 250         }
 251         /*
 252          * Inherently racy -- command line shares address space
 253          * with code and data.
 254          */
 255         rv = access_remote_vm(mm, arg_end - 1, &c, 1, 0);
 256         if (rv <= 0)
 257                 goto out_free_page;
 258
 259         rv = 0;
 260
 261         if (c == '\0') {
 262                 /* Command line (set of strings) occupies whole ARGV. */
 263                 if (len1 <= *pos)
 264                         goto out_free_page;
 265
 266                 p = arg_start + *pos;
 267                 len = len1 - *pos;
 268                 while (count > 0 && len > 0) {
 269                         unsigned int _count;
 270                         int nr_read;
 271
 272                         _count = min3(count, len, PAGE_SIZE);
 273                         nr_read = access_remote_vm(mm, p, page, _count, 0);
 274                         if (nr_read < 0)
 275                                 rv = nr_read;
 276                         if (nr_read <= 0)
 277                                 goto out_free_page;
 278
 279                         if (copy_to_user(buf, page, nr_read)) {
 280                                 rv = -EFAULT;
 281                                 goto out_free_page;
 282                         }
 283
 284                         p       += nr_read;
 285                         len     -= nr_read;
 286                         buf     += nr_read;
 287                         count   -= nr_read;
 288                         rv      += nr_read;
 289                 }
 290         } else {
 291                 /*
 292                  * Command line (1 string) occupies ARGV and maybe
 293                  * extends into ENVP.
 294                  */
 295                 if (len1 + len2 <= *pos)
 296                         goto skip_argv_envp;
 297                 if (len1 <= *pos)
 298                         goto skip_argv;
 299
 300                 p = arg_start + *pos;
 301                 len = len1 - *pos;
 302                 while (count > 0 && len > 0) {
 303                         unsigned int _count, l;
 304                         int nr_read;
 305                         bool final;
 306
 307                         _count = min3(count, len, PAGE_SIZE);
 308                         nr_read = access_remote_vm(mm, p, page, _count, 0);
 309                         if (nr_read < 0)
 310                                 rv = nr_read;
 311                         if (nr_read <= 0)
 312                                 goto out_free_page;
 313
 314                         /*
 315                          * Command line can be shorter than whole ARGV
 316                          * even if last "marker" byte says it is not.
 317                          */
 318                         final = false;
 319                         l = strnlen(page, nr_read);
 320                         if (l < nr_read) {
 321                                 nr_read = l;
 322                                 final = true;
 323                         }
 324
 325                         if (copy_to_user(buf, page, nr_read)) {
 326                                 rv = -EFAULT;
 327                                 goto out_free_page;
 328                         }
 329
 330                         p       += nr_read;
 331                         len     -= nr_read;
 332                         buf     += nr_read;
 333                         count   -= nr_read;
 334                         rv      += nr_read;
 335
 336                         if (final)
 337                                 goto out_free_page;
 338                 }
 339 skip_argv:
 340                 /*
 341                  * Command line (1 string) occupies ARGV and
 342                  * extends into ENVP.
 343                  */
 344                 if (len1 <= *pos) {
 345                         p = env_start + *pos - len1;
 346                         len = len1 + len2 - *pos;
 347                 } else {
 348                         p = env_start;
 349                         len = len2;
 350                 }
 351                 while (count > 0 && len > 0) {
 352                         unsigned int _count, l;
 353                         int nr_read;
 354                         bool final;
 355
 356                         _count = min3(count, len, PAGE_SIZE);
 357                         nr_read = access_remote_vm(mm, p, page, _count, 0);
 358                         if (nr_read < 0)
 359                                 rv = nr_read;
 360                         if (nr_read <= 0)
 361                                 goto out_free_page;
 362
 363                         /* Find EOS. */
 364                         final = false;
 365                         l = strnlen(page, nr_read);
 366                         if (l < nr_read) {
 367                                 nr_read = l;
 368                                 final = true;
 369                         }
 370
 371                         if (copy_to_user(buf, page, nr_read)) {
 372                                 rv = -EFAULT;
 373                                 goto out_free_page;
 374                         }
 375
 376                         p       += nr_read;
 377                         len     -= nr_read;
 378                         buf     += nr_read;
 379                         count   -= nr_read;
 380                         rv      += nr_read;
 381
 382                         if (final)
 383                                 goto out_free_page;
 384                 }
 385 skip_argv_envp:
 386                 ;
 387         }
 388
 389 out_free_page:
 390         free_page((unsigned long)page);
 391 out_mmput:
 392         mmput(mm);
 393         if (rv > 0)
 394                 *pos += rv;
 395         return rv;
 396 }
 397
 398 static const struct file_operations proc_pid_cmdline_ops = {
 399         .read   = proc_pid_cmdline_read,
 400         .llseek = generic_file_llseek,
 401 };
 402
 403 static int proc_pid_auxv(struct seq_file *m, struct pid_namespace *ns,
 404                          struct pid *pid, struct task_struct *task)
 405 {
 406         struct mm_struct *mm = mm_access(task, PTRACE_MODE_READ);
 407         if (mm && !IS_ERR(mm)) {
 408                 unsigned int nwords = 0;
 409                 do {
 410                         nwords += 2;
 411                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 412                 seq_write(m, mm->saved_auxv, nwords * sizeof(mm->saved_auxv[0]));
 413                 mmput(mm);
 414                 return 0;
 415         } else
 416                 return PTR_ERR(mm);
 417 }
 418
 419
 420 #ifdef CONFIG_KALLSYMS
 421 /*
 422  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 423  * Returns the resolved symbol.  If that fails, simply return the address.
 424  */
 425 static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
 426                           struct pid *pid, struct task_struct *task)
 427 {
 428         unsigned long wchan;
 429         char symname[KSYM_NAME_LEN];
 430
 431         wchan = get_wchan(task);
 432
 433         if (lookup_symbol_name(wchan, symname) < 0) {
 434                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
 435                         return 0;
 436                 seq_printf(m, "%lu", wchan);
 437         } else {
 438                 seq_printf(m, "%s", symname);
 439         }
 440
 441         return 0;
 442 }
 443 #endif /* CONFIG_KALLSYMS */
 444
 445 static int lock_trace(struct task_struct *task)
 446 {
 447         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
 448         if (err)
 449                 return err;
 450         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH)) {
 451                 mutex_unlock(&task->signal->cred_guard_mutex);
 452                 return -EPERM;
 453         }
 454         return 0;
 455 }
 456
 457 static void unlock_trace(struct task_struct *task)
 458 {
 459         mutex_unlock(&task->signal->cred_guard_mutex);
 460 }
 461
 462 #ifdef CONFIG_STACKTRACE
 463
 464 #define MAX_STACK_TRACE_DEPTH   64
 465
 466 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 467                           struct pid *pid, struct task_struct *task)
 468 {
 469         struct stack_trace trace;
 470         unsigned long *entries;
 471         int err;
 472         int i;
 473
 474         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 475         if (!entries)
 476                 return -ENOMEM;
 477
 478         trace.nr_entries        = 0;
 479         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 480         trace.entries           = entries;
 481         trace.skip              = 0;
 482
 483         err = lock_trace(task);
 484         if (!err) {
 485                 save_stack_trace_tsk(task, &trace);
 486
 487                 for (i = 0; i < trace.nr_entries; i++) {
 488                         seq_printf(m, "[<%pK>] %pS\n",
 489                                    (void *)entries[i], (void *)entries[i]);
 490                 }
 491                 unlock_trace(task);
 492         }
 493         kfree(entries);
 494
 495         return err;
 496 }
 497 #endif
 498
 499 #ifdef CONFIG_SCHED_INFO
 500 /*
 501  * Provides /proc/PID/schedstat
 502  */
 503 static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
 504                               struct pid *pid, struct task_struct *task)
 505 {
 506         if (unlikely(!sched_info_on()))
 507                 seq_printf(m, "0 0 0\n");
 508         else
 509                 seq_printf(m, "%llu %llu %lu\n",
 510                    (unsigned long long)task->se.sum_exec_runtime,
 511                    (unsigned long long)task->sched_info.run_delay,
 512                    task->sched_info.pcount);
 513
 514         return 0;
 515 }
 516 #endif
 517
 518 #ifdef CONFIG_LATENCYTOP
 519 static int lstats_show_proc(struct seq_file *m, void *v)
 520 {
 521         int i;
 522         struct inode *inode = m->private;
 523         struct task_struct *task = get_proc_task(inode);
 524
 525         if (!task)
 526                 return -ESRCH;
 527         seq_puts(m, "Latency Top version : v0.1\n");
 528         for (i = 0; i < 32; i++) {
 529                 struct latency_record *lr = &task->latency_record[i];
 530                 if (lr->backtrace[0]) {
 531                         int q;
 532                         seq_printf(m, "%i %li %li",
 533                                    lr->count, lr->time, lr->max);
 534                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 535                                 unsigned long bt = lr->backtrace[q];
 536                                 if (!bt)
 537                                         break;
 538                                 if (bt == ULONG_MAX)
 539                                         break;
 540                                 seq_printf(m, " %ps", (void *)bt);
 541                         }
 542                         seq_putc(m, '\n');
 543                 }
 544
 545         }
 546         put_task_struct(task);
 547         return 0;
 548 }
 549
 550 static int lstats_open(struct inode *inode, struct file *file)
 551 {
 552         return single_open(file, lstats_show_proc, inode);
 553 }
 554
 555 static ssize_t lstats_write(struct file *file, const char __user *buf,
 556                             size_t count, loff_t *offs)
 557 {
 558         struct task_struct *task = get_proc_task(file_inode(file));
 559
 560         if (!task)
 561                 return -ESRCH;
 562         clear_all_latency_tracing(task);
 563         put_task_struct(task);
 564
 565         return count;
 566 }
 567
 568 static const struct file_operations proc_lstats_operations = {
 569         .open           = lstats_open,
 570         .read           = seq_read,
 571         .write          = lstats_write,
 572         .llseek         = seq_lseek,
 573         .release        = single_release,
 574 };
 575
 576 #endif
 577
 578 static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
 579                           struct pid *pid, struct task_struct *task)
 580 {
 581         unsigned long totalpages = totalram_pages + total_swap_pages;
 582         unsigned long points = 0;
 583
 584         read_lock(&tasklist_lock);
 585         if (pid_alive(task))
 586                 points = oom_badness(task, NULL, NULL, totalpages) *
 587                                                 1000 / totalpages;
 588         read_unlock(&tasklist_lock);
 589         seq_printf(m, "%lu\n", points);
 590
 591         return 0;
 592 }
 593
 594 struct limit_names {
 595         const char *name;
 596         const char *unit;
 597 };
 598
 599 static const struct limit_names lnames[RLIM_NLIMITS] = {
 600         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 601         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 602         [RLIMIT_DATA] = {"Max data size", "bytes"},
 603         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 604         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 605         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 606         [RLIMIT_NPROC] = {"Max processes", "processes"},
 607         [RLIMIT_NOFILE] = {"Max open files", "files"},
 608         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 609         [RLIMIT_AS] = {"Max address space", "bytes"},
 610         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 611         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 612         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 613         [RLIMIT_NICE] = {"Max nice priority", NULL},
 614         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 615         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 616 };
 617
 618 /* Display limits for a process */
 619 static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
 620                            struct pid *pid, struct task_struct *task)
 621 {
 622         unsigned int i;
 623         unsigned long flags;
 624
 625         struct rlimit rlim[RLIM_NLIMITS];
 626
 627         if (!lock_task_sighand(task, &flags))
 628                 return 0;
 629         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 630         unlock_task_sighand(task, &flags);
 631
 632         /*
 633          * print the file header
 634          */
 635        seq_printf(m, "%-25s %-20s %-20s %-10s\n",
 636                   "Limit", "Soft Limit", "Hard Limit", "Units");
 637
 638         for (i = 0; i < RLIM_NLIMITS; i++) {
 639                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 640                         seq_printf(m, "%-25s %-20s ",
 641                                    lnames[i].name, "unlimited");
 642                 else
 643                         seq_printf(m, "%-25s %-20lu ",
 644                                    lnames[i].name, rlim[i].rlim_cur);
 645
 646                 if (rlim[i].rlim_max == RLIM_INFINITY)
 647                         seq_printf(m, "%-20s ", "unlimited");
 648                 else
 649                         seq_printf(m, "%-20lu ", rlim[i].rlim_max);
 650
 651                 if (lnames[i].unit)
 652                         seq_printf(m, "%-10s\n", lnames[i].unit);
 653                 else
 654                         seq_putc(m, '\n');
 655         }
 656
 657         return 0;
 658 }
 659
 660 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 661 static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
 662                             struct pid *pid, struct task_struct *task)
 663 {
 664         long nr;
 665         unsigned long args[6], sp, pc;
 666         int res;
 667
 668         res = lock_trace(task);
 669         if (res)
 670                 return res;
 671
 672         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 673                 seq_puts(m, "running\n");
 674         else if (nr < 0)
 675                 seq_printf(m, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 676         else
 677                 seq_printf(m,
 678                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 679                        nr,
 680                        args[0], args[1], args[2], args[3], args[4], args[5],
 681                        sp, pc);
 682         unlock_trace(task);
 683
 684         return 0;
 685 }
 686 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 687
 688 /************************************************************************/
 689 /*                       Here the fs part begins                        */
 690 /************************************************************************/
 691
 692 /* permission checks */
 693 static int proc_fd_access_allowed(struct inode *inode)
 694 {
 695         struct task_struct *task;
 696         int allowed = 0;
 697         /* Allow access to a task's file descriptors if it is us or we
 698          * may use ptrace attach to the process and find out that
 699          * information.
 700          */
 701         task = get_proc_task(inode);
 702         if (task) {
 703                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
 704                 put_task_struct(task);
 705         }
 706         return allowed;
 707 }
 708
 709 int proc_setattr(struct dentry *dentry, struct iattr *attr)
 710 {
 711         int error;
 712         struct inode *inode = d_inode(dentry);
 713
 714         if (attr->ia_valid & ATTR_MODE)
 715                 return -EPERM;
 716
 717         error = inode_change_ok(inode, attr);
 718         if (error)
 719                 return error;
 720
 721         setattr_copy(inode, attr);
 722         mark_inode_dirty(inode);
 723         return 0;
 724 }
 725
 726 /*
 727  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
 728  * or euid/egid (for hide_pid_min=2)?
 729  */
 730 static bool has_pid_permissions(struct pid_namespace *pid,
 731                                  struct task_struct *task,
 732                                  int hide_pid_min)
 733 {
 734         if (pid->hide_pid < hide_pid_min)
 735                 return true;
 736         if (in_group_p(pid->pid_gid))
 737                 return true;
 738         return ptrace_may_access(task, PTRACE_MODE_READ);
 739 }
 740
 741
 742 static int proc_pid_permission(struct inode *inode, int mask)
 743 {
 744         struct pid_namespace *pid = inode->i_sb->s_fs_info;
 745         struct task_struct *task;
 746         bool has_perms;
 747
 748         task = get_proc_task(inode);
 749         if (!task)
 750                 return -ESRCH;
 751         has_perms = has_pid_permissions(pid, task, 1);
 752         put_task_struct(task);
 753
 754         if (!has_perms) {
 755                 if (pid->hide_pid == 2) {
 756                         /*
 757                          * Let's make getdents(), stat(), and open()
 758                          * consistent with each other.  If a process
 759                          * may not stat() a file, it shouldn't be seen
 760                          * in procfs at all.
 761                          */
 762                         return -ENOENT;
 763                 }
 764
 765                 return -EPERM;
 766         }
 767         return generic_permission(inode, mask);
 768 }
 769
 770
 771
 772 static const struct inode_operations proc_def_inode_operations = {
 773         .setattr        = proc_setattr,
 774 };
 775
 776 static int proc_single_show(struct seq_file *m, void *v)
 777 {
 778         struct inode *inode = m->private;
 779         struct pid_namespace *ns;
 780         struct pid *pid;
 781         struct task_struct *task;
 782         int ret;
 783
 784         ns = inode->i_sb->s_fs_info;
 785         pid = proc_pid(inode);
 786         task = get_pid_task(pid, PIDTYPE_PID);
 787         if (!task)
 788                 return -ESRCH;
 789
 790         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 791
 792         put_task_struct(task);
 793         return ret;
 794 }
 795
 796 static int proc_single_open(struct inode *inode, struct file *filp)
 797 {
 798         return single_open(filp, proc_single_show, inode);
 799 }
 800
 801 static const struct file_operations proc_single_file_operations = {
 802         .open           = proc_single_open,
 803         .read           = seq_read,
 804         .llseek         = seq_lseek,
 805         .release        = single_release,
 806 };
 807
 808
 809 struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
 810 {
 811         struct task_struct *task = get_proc_task(inode);
 812         struct mm_struct *mm = ERR_PTR(-ESRCH);
 813
 814         if (task) {
 815                 mm = mm_access(task, mode);
 816                 put_task_struct(task);
 817
 818                 if (!IS_ERR_OR_NULL(mm)) {
 819                         /* ensure this mm_struct can't be freed */
 820                         atomic_inc(&mm->mm_count);
 821                         /* but do not pin its memory */
 822                         mmput(mm);
 823                 }
 824         }
 825
 826         return mm;
 827 }
 828
 829 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
 830 {
 831         struct mm_struct *mm = proc_mem_open(inode, mode);
 832
 833         if (IS_ERR(mm))
 834                 return PTR_ERR(mm);
 835
 836         file->private_data = mm;
 837         return 0;
 838 }
 839
 840 static int mem_open(struct inode *inode, struct file *file)
 841 {
 842         int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
 843
 844         /* OK to pass negative loff_t, we can catch out-of-range */
 845         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 846
 847         return ret;
 848 }
 849
 850 static ssize_t mem_rw(struct file *file, char __user *buf,
 851                         size_t count, loff_t *ppos, int write)
 852 {
 853         struct mm_struct *mm = file->private_data;
 854         unsigned long addr = *ppos;
 855         ssize_t copied;
 856         char *page;
 857
 858         if (!mm)
 859                 return 0;
 860
 861         page = (char *)__get_free_page(GFP_TEMPORARY);
 862         if (!page)
 863                 return -ENOMEM;
 864
 865         copied = 0;
 866         if (!atomic_inc_not_zero(&mm->mm_users))
 867                 goto free;
 868
 869         while (count > 0) {
 870                 int this_len = min_t(int, count, PAGE_SIZE);
 871
 872                 if (write && copy_from_user(page, buf, this_len)) {
 873                         copied = -EFAULT;
 874                         break;
 875                 }
 876
 877                 this_len = access_remote_vm(mm, addr, page, this_len, write);
 878                 if (!this_len) {
 879                         if (!copied)
 880                                 copied = -EIO;
 881                         break;
 882                 }
 883
 884                 if (!write && copy_to_user(buf, page, this_len)) {
 885                         copied = -EFAULT;
 886                         break;
 887                 }
 888
 889                 buf += this_len;
 890                 addr += this_len;
 891                 copied += this_len;
 892                 count -= this_len;
 893         }
 894         *ppos = addr;
 895
 896         mmput(mm);
 897 free:
 898         free_page((unsigned long) page);
 899         return copied;
 900 }
 901
 902 static ssize_t mem_read(struct file *file, char __user *buf,
 903                         size_t count, loff_t *ppos)
 904 {
 905         return mem_rw(file, buf, count, ppos, 0);
 906 }
 907
 908 static ssize_t mem_write(struct file *file, const char __user *buf,
 909                          size_t count, loff_t *ppos)
 910 {
 911         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 912 }
 913
 914 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 915 {
 916         switch (orig) {
 917         case 0:
 918                 file->f_pos = offset;
 919                 break;
 920         case 1:
 921                 file->f_pos += offset;
 922                 break;
 923         default:
 924                 return -EINVAL;
 925         }
 926         force_successful_syscall_return();
 927         return file->f_pos;
 928 }
 929
 930 static int mem_release(struct inode *inode, struct file *file)
 931 {
 932         struct mm_struct *mm = file->private_data;
 933         if (mm)
 934                 mmdrop(mm);
 935         return 0;
 936 }
 937
 938 static const struct file_operations proc_mem_operations = {
 939         .llseek         = mem_lseek,
 940         .read           = mem_read,
 941         .write          = mem_write,
 942         .open           = mem_open,
 943         .release        = mem_release,
 944 };
 945
 946 static int environ_open(struct inode *inode, struct file *file)
 947 {
 948         return __mem_open(inode, file, PTRACE_MODE_READ);
 949 }
 950
 951 static ssize_t environ_read(struct file *file, char __user *buf,
 952                         size_t count, loff_t *ppos)
 953 {
 954         char *page;
 955         unsigned long src = *ppos;
 956         int ret = 0;
 957         struct mm_struct *mm = file->private_data;
 958
 959         if (!mm)
 960                 return 0;
 961
 962         page = (char *)__get_free_page(GFP_TEMPORARY);
 963         if (!page)
 964                 return -ENOMEM;
 965
 966         ret = 0;
 967         if (!atomic_inc_not_zero(&mm->mm_users))
 968                 goto free;
 969         while (count > 0) {
 970                 size_t this_len, max_len;
 971                 int retval;
 972
 973                 if (src >= (mm->env_end - mm->env_start))
 974                         break;
 975
 976                 this_len = mm->env_end - (mm->env_start + src);
 977
 978                 max_len = min_t(size_t, PAGE_SIZE, count);
 979                 this_len = min(max_len, this_len);
 980
 981                 retval = access_remote_vm(mm, (mm->env_start + src),
 982                         page, this_len, 0);
 983
 984                 if (retval <= 0) {
 985                         ret = retval;
 986                         break;
 987                 }
 988
 989                 if (copy_to_user(buf, page, retval)) {
 990                         ret = -EFAULT;
 991                         break;
 992                 }
 993
 994                 ret += retval;
 995                 src += retval;
 996                 buf += retval;
 997                 count -= retval;
 998         }
 999         *ppos = src;
1000         mmput(mm);
1001
1002 free:
1003         free_page((unsigned long) page);
1004         return ret;
1005 }
1006
1007 static const struct file_operations proc_environ_operations = {
1008         .open           = environ_open,
1009         .read           = environ_read,
1010         .llseek         = generic_file_llseek,
1011         .release        = mem_release,
1012 };
1013
1014 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1015                             loff_t *ppos)
1016 {
1017         struct task_struct *task = get_proc_task(file_inode(file));
1018         char buffer[PROC_NUMBUF];
1019         int oom_adj = OOM_ADJUST_MIN;
1020         size_t len;
1021         unsigned long flags;
1022
1023         if (!task)
1024                 return -ESRCH;
1025         if (lock_task_sighand(task, &flags)) {
1026                 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1027                         oom_adj = OOM_ADJUST_MAX;
1028                 else
1029                         oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1030                                   OOM_SCORE_ADJ_MAX;
1031                 unlock_task_sighand(task, &flags);
1032         }
1033         put_task_struct(task);
1034         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1035         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1036 }
1037
1038 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1039                              size_t count, loff_t *ppos)
1040 {
1041         struct task_struct *task;
1042         char buffer[PROC_NUMBUF];
1043         int oom_adj;
1044         unsigned long flags;
1045         int err;
1046
1047         memset(buffer, 0, sizeof(buffer));
1048         if (count > sizeof(buffer) - 1)
1049                 count = sizeof(buffer) - 1;
1050         if (copy_from_user(buffer, buf, count)) {
1051                 err = -EFAULT;
1052                 goto out;
1053         }
1054
1055         err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1056         if (err)
1057                 goto out;
1058         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1059              oom_adj != OOM_DISABLE) {
1060                 err = -EINVAL;
1061                 goto out;
1062         }
1063
1064         task = get_proc_task(file_inode(file));
1065         if (!task) {
1066                 err = -ESRCH;
1067                 goto out;
1068         }
1069
1070         task_lock(task);
1071         if (!task->mm) {
1072                 err = -EINVAL;
1073                 goto err_task_lock;
1074         }
1075
1076         if (!lock_task_sighand(task, &flags)) {
1077                 err = -ESRCH;
1078                 goto err_task_lock;
1079         }
1080
1081         /*
1082          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1083          * value is always attainable.
1084          */
1085         if (oom_adj == OOM_ADJUST_MAX)
1086                 oom_adj = OOM_SCORE_ADJ_MAX;
1087         else
1088                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1089
1090         if (oom_adj < task->signal->oom_score_adj &&
1091             !capable(CAP_SYS_RESOURCE)) {
1092                 err = -EACCES;
1093                 goto err_sighand;
1094         }
1095
1096         /*
1097          * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1098          * /proc/pid/oom_score_adj instead.
1099          */
1100         pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1101                   current->comm, task_pid_nr(current), task_pid_nr(task),
1102                   task_pid_nr(task));
1103
1104         task->signal->oom_score_adj = oom_adj;
1105         trace_oom_score_adj_update(task);
1106 err_sighand:
1107         unlock_task_sighand(task, &flags);
1108 err_task_lock:
1109         task_unlock(task);
1110         put_task_struct(task);
1111 out:
1112         return err < 0 ? err : count;
1113 }
1114
1115 static const struct file_operations proc_oom_adj_operations = {
1116         .read           = oom_adj_read,
1117         .write          = oom_adj_write,
1118         .llseek         = generic_file_llseek,
1119 };
1120
1121 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1122                                         size_t count, loff_t *ppos)
1123 {
1124         struct task_struct *task = get_proc_task(file_inode(file));
1125         char buffer[PROC_NUMBUF];
1126         short oom_score_adj = OOM_SCORE_ADJ_MIN;
1127         unsigned long flags;
1128         size_t len;
1129
1130         if (!task)
1131                 return -ESRCH;
1132         if (lock_task_sighand(task, &flags)) {
1133                 oom_score_adj = task->signal->oom_score_adj;
1134                 unlock_task_sighand(task, &flags);
1135         }
1136         put_task_struct(task);
1137         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1138         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1139 }
1140
1141 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1142                                         size_t count, loff_t *ppos)
1143 {
1144         struct task_struct *task;
1145         char buffer[PROC_NUMBUF];
1146         unsigned long flags;
1147         int oom_score_adj;
1148         int err;
1149
1150         memset(buffer, 0, sizeof(buffer));
1151         if (count > sizeof(buffer) - 1)
1152                 count = sizeof(buffer) - 1;
1153         if (copy_from_user(buffer, buf, count)) {
1154                 err = -EFAULT;
1155                 goto out;
1156         }
1157
1158         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1159         if (err)
1160                 goto out;
1161         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1162                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1163                 err = -EINVAL;
1164                 goto out;
1165         }
1166
1167         task = get_proc_task(file_inode(file));
1168         if (!task) {
1169                 err = -ESRCH;
1170                 goto out;
1171         }
1172
1173         task_lock(task);
1174         if (!task->mm) {
1175                 err = -EINVAL;
1176                 goto err_task_lock;
1177         }
1178
1179         if (!lock_task_sighand(task, &flags)) {
1180                 err = -ESRCH;
1181                 goto err_task_lock;
1182         }
1183
1184         if ((short)oom_score_adj < task->signal->oom_score_adj_min &&
1185                         !capable(CAP_SYS_RESOURCE)) {
1186                 err = -EACCES;
1187                 goto err_sighand;
1188         }
1189
1190         task->signal->oom_score_adj = (short)oom_score_adj;
1191         if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1192                 task->signal->oom_score_adj_min = (short)oom_score_adj;
1193         trace_oom_score_adj_update(task);
1194
1195 err_sighand:
1196         unlock_task_sighand(task, &flags);
1197 err_task_lock:
1198         task_unlock(task);
1199         put_task_struct(task);
1200 out:
1201         return err < 0 ? err : count;
1202 }
1203
1204 static const struct file_operations proc_oom_score_adj_operations = {
1205         .read           = oom_score_adj_read,
1206         .write          = oom_score_adj_write,
1207         .llseek         = default_llseek,
1208 };
1209
1210 #ifdef CONFIG_AUDITSYSCALL
1211 #define TMPBUFLEN 21
1212 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1213                                   size_t count, loff_t *ppos)
1214 {
1215         struct inode * inode = file_inode(file);
1216         struct task_struct *task = get_proc_task(inode);
1217         ssize_t length;
1218         char tmpbuf[TMPBUFLEN];
1219
1220         if (!task)
1221                 return -ESRCH;
1222         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1223                            from_kuid(file->f_cred->user_ns,
1224                                      audit_get_loginuid(task)));
1225         put_task_struct(task);
1226         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1227 }
1228
1229 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1230                                    size_t count, loff_t *ppos)
1231 {
1232         struct inode * inode = file_inode(file);
1233         uid_t loginuid;
1234         kuid_t kloginuid;
1235         int rv;
1236
1237         rcu_read_lock();
1238         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1239                 rcu_read_unlock();
1240                 return -EPERM;
1241         }
1242         rcu_read_unlock();
1243
1244         if (*ppos != 0) {
1245                 /* No partial writes. */
1246                 return -EINVAL;
1247         }
1248
1249         rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1250         if (rv < 0)
1251                 return rv;
1252
1253         /* is userspace tring to explicitly UNSET the loginuid? */
1254         if (loginuid == AUDIT_UID_UNSET) {
1255                 kloginuid = INVALID_UID;
1256         } else {
1257                 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1258                 if (!uid_valid(kloginuid))
1259                         return -EINVAL;
1260         }
1261
1262         rv = audit_set_loginuid(kloginuid);
1263         if (rv < 0)
1264                 return rv;
1265         return count;
1266 }
1267
1268 static const struct file_operations proc_loginuid_operations = {
1269         .read           = proc_loginuid_read,
1270         .write          = proc_loginuid_write,
1271         .llseek         = generic_file_llseek,
1272 };
1273
1274 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1275                                   size_t count, loff_t *ppos)
1276 {
1277         struct inode * inode = file_inode(file);
1278         struct task_struct *task = get_proc_task(inode);
1279         ssize_t length;
1280         char tmpbuf[TMPBUFLEN];
1281
1282         if (!task)
1283                 return -ESRCH;
1284         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1285                                 audit_get_sessionid(task));
1286         put_task_struct(task);
1287         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1288 }
1289
1290 static const struct file_operations proc_sessionid_operations = {
1291         .read           = proc_sessionid_read,
1292         .llseek         = generic_file_llseek,
1293 };
1294 #endif
1295
1296 #ifdef CONFIG_FAULT_INJECTION
1297 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1298                                       size_t count, loff_t *ppos)
1299 {
1300         struct task_struct *task = get_proc_task(file_inode(file));
1301         char buffer[PROC_NUMBUF];
1302         size_t len;
1303         int make_it_fail;
1304
1305         if (!task)
1306                 return -ESRCH;
1307         make_it_fail = task->make_it_fail;
1308         put_task_struct(task);
1309
1310         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1311
1312         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1313 }
1314
1315 static ssize_t proc_fault_inject_write(struct file * file,
1316                         const char __user * buf, size_t count, loff_t *ppos)
1317 {
1318         struct task_struct *task;
1319         char buffer[PROC_NUMBUF];
1320         int make_it_fail;
1321         int rv;
1322
1323         if (!capable(CAP_SYS_RESOURCE))
1324                 return -EPERM;
1325         memset(buffer, 0, sizeof(buffer));
1326         if (count > sizeof(buffer) - 1)
1327                 count = sizeof(buffer) - 1;
1328         if (copy_from_user(buffer, buf, count))
1329                 return -EFAULT;
1330         rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1331         if (rv < 0)
1332                 return rv;
1333         if (make_it_fail < 0 || make_it_fail > 1)
1334                 return -EINVAL;
1335
1336         task = get_proc_task(file_inode(file));
1337         if (!task)
1338                 return -ESRCH;
1339         task->make_it_fail = make_it_fail;
1340         put_task_struct(task);
1341
1342         return count;
1343 }
1344
1345 static const struct file_operations proc_fault_inject_operations = {
1346         .read           = proc_fault_inject_read,
1347         .write          = proc_fault_inject_write,
1348         .llseek         = generic_file_llseek,
1349 };
1350 #endif
1351
1352
1353 #ifdef CONFIG_SCHED_DEBUG
1354 /*
1355  * Print out various scheduling related per-task fields:
1356  */
1357 static int sched_show(struct seq_file *m, void *v)
1358 {
1359         struct inode *inode = m->private;
1360         struct task_struct *p;
1361
1362         p = get_proc_task(inode);
1363         if (!p)
1364                 return -ESRCH;
1365         proc_sched_show_task(p, m);
1366
1367         put_task_struct(p);
1368
1369         return 0;
1370 }
1371
1372 static ssize_t
1373 sched_write(struct file *file, const char __user *buf,
1374             size_t count, loff_t *offset)
1375 {
1376         struct inode *inode = file_inode(file);
1377         struct task_struct *p;
1378
1379         p = get_proc_task(inode);
1380         if (!p)
1381                 return -ESRCH;
1382         proc_sched_set_task(p);
1383
1384         put_task_struct(p);
1385
1386         return count;
1387 }
1388
1389 static int sched_open(struct inode *inode, struct file *filp)
1390 {
1391         return single_open(filp, sched_show, inode);
1392 }
1393
1394 static const struct file_operations proc_pid_sched_operations = {
1395         .open           = sched_open,
1396         .read           = seq_read,
1397         .write          = sched_write,
1398         .llseek         = seq_lseek,
1399         .release        = single_release,
1400 };
1401
1402 #endif
1403
1404 #ifdef CONFIG_SCHED_AUTOGROUP
1405 /*
1406  * Print out autogroup related information:
1407  */
1408 static int sched_autogroup_show(struct seq_file *m, void *v)
1409 {
1410         struct inode *inode = m->private;
1411         struct task_struct *p;
1412
1413         p = get_proc_task(inode);
1414         if (!p)
1415                 return -ESRCH;
1416         proc_sched_autogroup_show_task(p, m);
1417
1418         put_task_struct(p);
1419
1420         return 0;
1421 }
1422
1423 static ssize_t
1424 sched_autogroup_write(struct file *file, const char __user *buf,
1425             size_t count, loff_t *offset)
1426 {
1427         struct inode *inode = file_inode(file);
1428         struct task_struct *p;
1429         char buffer[PROC_NUMBUF];
1430         int nice;
1431         int err;
1432
1433         memset(buffer, 0, sizeof(buffer));
1434         if (count > sizeof(buffer) - 1)
1435                 count = sizeof(buffer) - 1;
1436         if (copy_from_user(buffer, buf, count))
1437                 return -EFAULT;
1438
1439         err = kstrtoint(strstrip(buffer), 0, &nice);
1440         if (err < 0)
1441                 return err;
1442
1443         p = get_proc_task(inode);
1444         if (!p)
1445                 return -ESRCH;
1446
1447         err = proc_sched_autogroup_set_nice(p, nice);
1448         if (err)
1449                 count = err;
1450
1451         put_task_struct(p);
1452
1453         return count;
1454 }
1455
1456 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1457 {
1458         int ret;
1459
1460         ret = single_open(filp, sched_autogroup_show, NULL);
1461         if (!ret) {
1462                 struct seq_file *m = filp->private_data;
1463
1464                 m->private = inode;
1465         }
1466         return ret;
1467 }
1468
1469 static const struct file_operations proc_pid_sched_autogroup_operations = {
1470         .open           = sched_autogroup_open,
1471         .read           = seq_read,
1472         .write          = sched_autogroup_write,
1473         .llseek         = seq_lseek,
1474         .release        = single_release,
1475 };
1476
1477 #endif /* CONFIG_SCHED_AUTOGROUP */
1478
1479 static ssize_t comm_write(struct file *file, const char __user *buf,
1480                                 size_t count, loff_t *offset)
1481 {
1482         struct inode *inode = file_inode(file);
1483         struct task_struct *p;
1484         char buffer[TASK_COMM_LEN];
1485         const size_t maxlen = sizeof(buffer) - 1;
1486
1487         memset(buffer, 0, sizeof(buffer));
1488         if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1489                 return -EFAULT;
1490
1491         p = get_proc_task(inode);
1492         if (!p)
1493                 return -ESRCH;
1494
1495         if (same_thread_group(current, p))
1496                 set_task_comm(p, buffer);
1497         else
1498                 count = -EINVAL;
1499
1500         put_task_struct(p);
1501
1502         return count;
1503 }
1504
1505 static int comm_show(struct seq_file *m, void *v)
1506 {
1507         struct inode *inode = m->private;
1508         struct task_struct *p;
1509
1510         p = get_proc_task(inode);
1511         if (!p)
1512                 return -ESRCH;
1513
1514         task_lock(p);
1515         seq_printf(m, "%s\n", p->comm);
1516         task_unlock(p);
1517
1518         put_task_struct(p);
1519
1520         return 0;
1521 }
1522
1523 static int comm_open(struct inode *inode, struct file *filp)
1524 {
1525         return single_open(filp, comm_show, inode);
1526 }
1527
1528 static const struct file_operations proc_pid_set_comm_operations = {
1529         .open           = comm_open,
1530         .read           = seq_read,
1531         .write          = comm_write,
1532         .llseek         = seq_lseek,
1533         .release        = single_release,
1534 };
1535
1536 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1537 {
1538         struct task_struct *task;
1539         struct mm_struct *mm;
1540         struct file *exe_file;
1541
1542         task = get_proc_task(d_inode(dentry));
1543         if (!task)
1544                 return -ENOENT;
1545         mm = get_task_mm(task);
1546         put_task_struct(task);
1547         if (!mm)
1548                 return -ENOENT;
1549         exe_file = get_mm_exe_file(mm);
1550         mmput(mm);
1551         if (exe_file) {
1552                 *exe_path = exe_file->f_path;
1553                 path_get(&exe_file->f_path);
1554                 fput(exe_file);
1555                 return 0;
1556         } else
1557                 return -ENOENT;
1558 }
1559
1560 static const char *proc_pid_follow_link(struct dentry *dentry, void **cookie)
1561 {
1562         struct inode *inode = d_inode(dentry);
1563         struct path path;
1564         int error = -EACCES;
1565
1566         /* Are we allowed to snoop on the tasks file descriptors? */
1567         if (!proc_fd_access_allowed(inode))
1568                 goto out;
1569
1570         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1571         if (error)
1572                 goto out;
1573
1574         nd_jump_link(&path);
1575         return NULL;
1576 out:
1577         return ERR_PTR(error);
1578 }
1579
1580 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1581 {
1582         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1583         char *pathname;
1584         int len;
1585
1586         if (!tmp)
1587                 return -ENOMEM;
1588
1589         pathname = d_path(path, tmp, PAGE_SIZE);
1590         len = PTR_ERR(pathname);
1591         if (IS_ERR(pathname))
1592                 goto out;
1593         len = tmp + PAGE_SIZE - 1 - pathname;
1594
1595         if (len > buflen)
1596                 len = buflen;
1597         if (copy_to_user(buffer, pathname, len))
1598                 len = -EFAULT;
1599  out:
1600         free_page((unsigned long)tmp);
1601         return len;
1602 }
1603
1604 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1605 {
1606         int error = -EACCES;
1607         struct inode *inode = d_inode(dentry);
1608         struct path path;
1609
1610         /* Are we allowed to snoop on the tasks file descriptors? */
1611         if (!proc_fd_access_allowed(inode))
1612                 goto out;
1613
1614         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1615         if (error)
1616                 goto out;
1617
1618         error = do_proc_readlink(&path, buffer, buflen);
1619         path_put(&path);
1620 out:
1621         return error;
1622 }
1623
1624 const struct inode_operations proc_pid_link_inode_operations = {
1625         .readlink       = proc_pid_readlink,
1626         .follow_link    = proc_pid_follow_link,
1627         .setattr        = proc_setattr,
1628 };
1629
1630
1631 /* building an inode */
1632
1633 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1634 {
1635         struct inode * inode;
1636         struct proc_inode *ei;
1637         const struct cred *cred;
1638
1639         /* We need a new inode */
1640
1641         inode = new_inode(sb);
1642         if (!inode)
1643                 goto out;
1644
1645         /* Common stuff */
1646         ei = PROC_I(inode);
1647         inode->i_ino = get_next_ino();
1648         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1649         inode->i_op = &proc_def_inode_operations;
1650
1651         /*
1652          * grab the reference to task.
1653          */
1654         ei->pid = get_task_pid(task, PIDTYPE_PID);
1655         if (!ei->pid)
1656                 goto out_unlock;
1657
1658         if (task_dumpable(task)) {
1659                 rcu_read_lock();
1660                 cred = __task_cred(task);
1661                 inode->i_uid = cred->euid;
1662                 inode->i_gid = cred->egid;
1663                 rcu_read_unlock();
1664         }
1665         security_task_to_inode(task, inode);
1666
1667 out:
1668         return inode;
1669
1670 out_unlock:
1671         iput(inode);
1672         return NULL;
1673 }
1674
1675 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1676 {
1677         struct inode *inode = d_inode(dentry);
1678         struct task_struct *task;
1679         const struct cred *cred;
1680         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1681
1682         generic_fillattr(inode, stat);
1683
1684         rcu_read_lock();
1685         stat->uid = GLOBAL_ROOT_UID;
1686         stat->gid = GLOBAL_ROOT_GID;
1687         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1688         if (task) {
1689                 if (!has_pid_permissions(pid, task, 2)) {
1690                         rcu_read_unlock();
1691                         /*
1692                          * This doesn't prevent learning whether PID exists,
1693                          * it only makes getattr() consistent with readdir().
1694                          */
1695                         return -ENOENT;
1696                 }
1697                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1698                     task_dumpable(task)) {
1699                         cred = __task_cred(task);
1700                         stat->uid = cred->euid;
1701                         stat->gid = cred->egid;
1702                 }
1703         }
1704         rcu_read_unlock();
1705         return 0;
1706 }
1707
1708 /* dentry stuff */
1709
1710 /*
1711  *      Exceptional case: normally we are not allowed to unhash a busy
1712  * directory. In this case, however, we can do it - no aliasing problems
1713  * due to the way we treat inodes.
1714  *
1715  * Rewrite the inode's ownerships here because the owning task may have
1716  * performed a setuid(), etc.
1717  *
1718  * Before the /proc/pid/status file was created the only way to read
1719  * the effective uid of a /process was to stat /proc/pid.  Reading
1720  * /proc/pid/status is slow enough that procps and other packages
1721  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1722  * made this apply to all per process world readable and executable
1723  * directories.
1724  */
1725 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1726 {
1727         struct inode *inode;
1728         struct task_struct *task;
1729         const struct cred *cred;
1730
1731         if (flags & LOOKUP_RCU)
1732                 return -ECHILD;
1733
1734         inode = d_inode(dentry);
1735         task = get_proc_task(inode);
1736
1737         if (task) {
1738                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1739                     task_dumpable(task)) {
1740                         rcu_read_lock();
1741                         cred = __task_cred(task);
1742                         inode->i_uid = cred->euid;
1743                         inode->i_gid = cred->egid;
1744                         rcu_read_unlock();
1745                 } else {
1746                         inode->i_uid = GLOBAL_ROOT_UID;
1747                         inode->i_gid = GLOBAL_ROOT_GID;
1748                 }
1749                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1750                 security_task_to_inode(task, inode);
1751                 put_task_struct(task);
1752                 return 1;
1753         }
1754         return 0;
1755 }
1756
1757 static inline bool proc_inode_is_dead(struct inode *inode)
1758 {
1759         return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1760 }
1761
1762 int pid_delete_dentry(const struct dentry *dentry)
1763 {
1764         /* Is the task we represent dead?
1765          * If so, then don't put the dentry on the lru list,
1766          * kill it immediately.
1767          */
1768         return proc_inode_is_dead(d_inode(dentry));
1769 }
1770
1771 const struct dentry_operations pid_dentry_operations =
1772 {
1773         .d_revalidate   = pid_revalidate,
1774         .d_delete       = pid_delete_dentry,
1775 };
1776
1777 /* Lookups */
1778
1779 /*
1780  * Fill a directory entry.
1781  *
1782  * If possible create the dcache entry and derive our inode number and
1783  * file type from dcache entry.
1784  *
1785  * Since all of the proc inode numbers are dynamically generated, the inode
1786  * numbers do not exist until the inode is cache.  This means creating the
1787  * the dcache entry in readdir is necessary to keep the inode numbers
1788  * reported by readdir in sync with the inode numbers reported
1789  * by stat.
1790  */
1791 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1792         const char *name, int len,
1793         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1794 {
1795         struct dentry *child, *dir = file->f_path.dentry;
1796         struct qstr qname = QSTR_INIT(name, len);
1797         struct inode *inode;
1798         unsigned type;
1799         ino_t ino;
1800
1801         child = d_hash_and_lookup(dir, &qname);
1802         if (!child) {
1803                 child = d_alloc(dir, &qname);
1804                 if (!child)
1805                         goto end_instantiate;
1806                 if (instantiate(d_inode(dir), child, task, ptr) < 0) {
1807                         dput(child);
1808                         goto end_instantiate;
1809                 }
1810         }
1811         inode = d_inode(child);
1812         ino = inode->i_ino;
1813         type = inode->i_mode >> 12;
1814         dput(child);
1815         return dir_emit(ctx, name, len, ino, type);
1816
1817 end_instantiate:
1818         return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1819 }
1820
1821 /*
1822  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1823  * which represent vma start and end addresses.
1824  */
1825 static int dname_to_vma_addr(struct dentry *dentry,
1826                              unsigned long *start, unsigned long *end)
1827 {
1828         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1829                 return -EINVAL;
1830
1831         return 0;
1832 }
1833
1834 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1835 {
1836         unsigned long vm_start, vm_end;
1837         bool exact_vma_exists = false;
1838         struct mm_struct *mm = NULL;
1839         struct task_struct *task;
1840         const struct cred *cred;
1841         struct inode *inode;
1842         int status = 0;
1843
1844         if (flags & LOOKUP_RCU)
1845                 return -ECHILD;
1846
1847         inode = d_inode(dentry);
1848         task = get_proc_task(inode);
1849         if (!task)
1850                 goto out_notask;
1851
1852         mm = mm_access(task, PTRACE_MODE_READ);
1853         if (IS_ERR_OR_NULL(mm))
1854                 goto out;
1855
1856         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1857                 down_read(&mm->mmap_sem);
1858                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1859                 up_read(&mm->mmap_sem);
1860         }
1861
1862         mmput(mm);
1863
1864         if (exact_vma_exists) {
1865                 if (task_dumpable(task)) {
1866                         rcu_read_lock();
1867                         cred = __task_cred(task);
1868                         inode->i_uid = cred->euid;
1869                         inode->i_gid = cred->egid;
1870                         rcu_read_unlock();
1871                 } else {
1872                         inode->i_uid = GLOBAL_ROOT_UID;
1873                         inode->i_gid = GLOBAL_ROOT_GID;
1874                 }
1875                 security_task_to_inode(task, inode);
1876                 status = 1;
1877         }
1878
1879 out:
1880         put_task_struct(task);
1881
1882 out_notask:
1883         return status;
1884 }
1885
1886 static const struct dentry_operations tid_map_files_dentry_operations = {
1887         .d_revalidate   = map_files_d_revalidate,
1888         .d_delete       = pid_delete_dentry,
1889 };
1890
1891 static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
1892 {
1893         unsigned long vm_start, vm_end;
1894         struct vm_area_struct *vma;
1895         struct task_struct *task;
1896         struct mm_struct *mm;
1897         int rc;
1898
1899         rc = -ENOENT;
1900         task = get_proc_task(d_inode(dentry));
1901         if (!task)
1902                 goto out;
1903
1904         mm = get_task_mm(task);
1905         put_task_struct(task);
1906         if (!mm)
1907                 goto out;
1908
1909         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1910         if (rc)
1911                 goto out_mmput;
1912
1913         rc = -ENOENT;
1914         down_read(&mm->mmap_sem);
1915         vma = find_exact_vma(mm, vm_start, vm_end);
1916         if (vma && vma->vm_file) {
1917                 *path = vma->vm_file->f_path;
1918                 path_get(path);
1919                 rc = 0;
1920         }
1921         up_read(&mm->mmap_sem);
1922
1923 out_mmput:
1924         mmput(mm);
1925 out:
1926         return rc;
1927 }
1928
1929 struct map_files_info {
1930         fmode_t         mode;
1931         unsigned long   len;
1932         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1933 };
1934
1935 /*
1936  * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
1937  * symlinks may be used to bypass permissions on ancestor directories in the
1938  * path to the file in question.
1939  */
1940 static const char *
1941 proc_map_files_follow_link(struct dentry *dentry, void **cookie)
1942 {
1943         if (!capable(CAP_SYS_ADMIN))
1944                 return ERR_PTR(-EPERM);
1945
1946         return proc_pid_follow_link(dentry, NULL);
1947 }
1948
1949 /*
1950  * Identical to proc_pid_link_inode_operations except for follow_link()
1951  */
1952 static const struct inode_operations proc_map_files_link_inode_operations = {
1953         .readlink       = proc_pid_readlink,
1954         .follow_link    = proc_map_files_follow_link,
1955         .setattr        = proc_setattr,
1956 };
1957
1958 static int
1959 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
1960                            struct task_struct *task, const void *ptr)
1961 {
1962         fmode_t mode = (fmode_t)(unsigned long)ptr;
1963         struct proc_inode *ei;
1964         struct inode *inode;
1965
1966         inode = proc_pid_make_inode(dir->i_sb, task);
1967         if (!inode)
1968                 return -ENOENT;
1969
1970         ei = PROC_I(inode);
1971         ei->op.proc_get_link = proc_map_files_get_link;
1972
1973         inode->i_op = &proc_map_files_link_inode_operations;
1974         inode->i_size = 64;
1975         inode->i_mode = S_IFLNK;
1976
1977         if (mode & FMODE_READ)
1978                 inode->i_mode |= S_IRUSR;
1979         if (mode & FMODE_WRITE)
1980                 inode->i_mode |= S_IWUSR;
1981
1982         d_set_d_op(dentry, &tid_map_files_dentry_operations);
1983         d_add(dentry, inode);
1984
1985         return 0;
1986 }
1987
1988 static struct dentry *proc_map_files_lookup(struct inode *dir,
1989                 struct dentry *dentry, unsigned int flags)
1990 {
1991         unsigned long vm_start, vm_end;
1992         struct vm_area_struct *vma;
1993         struct task_struct *task;
1994         int result;
1995         struct mm_struct *mm;
1996
1997         result = -ENOENT;
1998         task = get_proc_task(dir);
1999         if (!task)
2000                 goto out;
2001
2002         result = -EACCES;
2003         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2004                 goto out_put_task;
2005
2006         result = -ENOENT;
2007         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2008                 goto out_put_task;
2009
2010         mm = get_task_mm(task);
2011         if (!mm)
2012                 goto out_put_task;
2013
2014         down_read(&mm->mmap_sem);
2015         vma = find_exact_vma(mm, vm_start, vm_end);
2016         if (!vma)
2017                 goto out_no_vma;
2018
2019         if (vma->vm_file)
2020                 result = proc_map_files_instantiate(dir, dentry, task,
2021                                 (void *)(unsigned long)vma->vm_file->f_mode);
2022
2023 out_no_vma:
2024         up_read(&mm->mmap_sem);
2025         mmput(mm);
2026 out_put_task:
2027         put_task_struct(task);
2028 out:
2029         return ERR_PTR(result);
2030 }
2031
2032 static const struct inode_operations proc_map_files_inode_operations = {
2033         .lookup         = proc_map_files_lookup,
2034         .permission     = proc_fd_permission,
2035         .setattr        = proc_setattr,
2036 };
2037
2038 static int
2039 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2040 {
2041         struct vm_area_struct *vma;
2042         struct task_struct *task;
2043         struct mm_struct *mm;
2044         unsigned long nr_files, pos, i;
2045         struct flex_array *fa = NULL;
2046         struct map_files_info info;
2047         struct map_files_info *p;
2048         int ret;
2049
2050         ret = -ENOENT;
2051         task = get_proc_task(file_inode(file));
2052         if (!task)
2053                 goto out;
2054
2055         ret = -EACCES;
2056         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2057                 goto out_put_task;
2058
2059         ret = 0;
2060         if (!dir_emit_dots(file, ctx))
2061                 goto out_put_task;
2062
2063         mm = get_task_mm(task);
2064         if (!mm)
2065                 goto out_put_task;
2066         down_read(&mm->mmap_sem);
2067
2068         nr_files = 0;
2069
2070         /*
2071          * We need two passes here:
2072          *
2073          *  1) Collect vmas of mapped files with mmap_sem taken
2074          *  2) Release mmap_sem and instantiate entries
2075          *
2076          * otherwise we get lockdep complained, since filldir()
2077          * routine might require mmap_sem taken in might_fault().
2078          */
2079
2080         for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2081                 if (vma->vm_file && ++pos > ctx->pos)
2082                         nr_files++;
2083         }
2084
2085         if (nr_files) {
2086                 fa = flex_array_alloc(sizeof(info), nr_files,
2087                                         GFP_KERNEL);
2088                 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2089                                                 GFP_KERNEL)) {
2090                         ret = -ENOMEM;
2091                         if (fa)
2092                                 flex_array_free(fa);
2093                         up_read(&mm->mmap_sem);
2094                         mmput(mm);
2095                         goto out_put_task;
2096                 }
2097                 for (i = 0, vma = mm->mmap, pos = 2; vma;
2098                                 vma = vma->vm_next) {
2099                         if (!vma->vm_file)
2100                                 continue;
2101                         if (++pos <= ctx->pos)
2102                                 continue;
2103
2104                         info.mode = vma->vm_file->f_mode;
2105                         info.len = snprintf(info.name,
2106                                         sizeof(info.name), "%lx-%lx",
2107                                         vma->vm_start, vma->vm_end);
2108                         if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2109                                 BUG();
2110                 }
2111         }
2112         up_read(&mm->mmap_sem);
2113
2114         for (i = 0; i < nr_files; i++) {
2115                 p = flex_array_get(fa, i);
2116                 if (!proc_fill_cache(file, ctx,
2117                                       p->name, p->len,
2118                                       proc_map_files_instantiate,
2119                                       task,
2120                                       (void *)(unsigned long)p->mode))
2121                         break;
2122                 ctx->pos++;
2123         }
2124         if (fa)
2125                 flex_array_free(fa);
2126         mmput(mm);
2127
2128 out_put_task:
2129         put_task_struct(task);
2130 out:
2131         return ret;
2132 }
2133
2134 static const struct file_operations proc_map_files_operations = {
2135         .read           = generic_read_dir,
2136         .iterate        = proc_map_files_readdir,
2137         .llseek         = default_llseek,
2138 };
2139
2140 struct timers_private {
2141         struct pid *pid;
2142         struct task_struct *task;
2143         struct sighand_struct *sighand;
2144         struct pid_namespace *ns;
2145         unsigned long flags;
2146 };
2147
2148 static void *timers_start(struct seq_file *m, loff_t *pos)
2149 {
2150         struct timers_private *tp = m->private;
2151
2152         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2153         if (!tp->task)
2154                 return ERR_PTR(-ESRCH);
2155
2156         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2157         if (!tp->sighand)
2158                 return ERR_PTR(-ESRCH);
2159
2160         return seq_list_start(&tp->task->signal->posix_timers, *pos);
2161 }
2162
2163 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2164 {
2165         struct timers_private *tp = m->private;
2166         return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2167 }
2168
2169 static void timers_stop(struct seq_file *m, void *v)
2170 {
2171         struct timers_private *tp = m->private;
2172
2173         if (tp->sighand) {
2174                 unlock_task_sighand(tp->task, &tp->flags);
2175                 tp->sighand = NULL;
2176         }
2177
2178         if (tp->task) {
2179                 put_task_struct(tp->task);
2180                 tp->task = NULL;
2181         }
2182 }
2183
2184 static int show_timer(struct seq_file *m, void *v)
2185 {
2186         struct k_itimer *timer;
2187         struct timers_private *tp = m->private;
2188         int notify;
2189         static const char * const nstr[] = {
2190                 [SIGEV_SIGNAL] = "signal",
2191                 [SIGEV_NONE] = "none",
2192                 [SIGEV_THREAD] = "thread",
2193         };
2194
2195         timer = list_entry((struct list_head *)v, struct k_itimer, list);
2196         notify = timer->it_sigev_notify;
2197
2198         seq_printf(m, "ID: %d\n", timer->it_id);
2199         seq_printf(m, "signal: %d/%p\n",
2200                    timer->sigq->info.si_signo,
2201                    timer->sigq->info.si_value.sival_ptr);
2202         seq_printf(m, "notify: %s/%s.%d\n",
2203                    nstr[notify & ~SIGEV_THREAD_ID],
2204                    (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2205                    pid_nr_ns(timer->it_pid, tp->ns));
2206         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2207
2208         return 0;
2209 }
2210
2211 static const struct seq_operations proc_timers_seq_ops = {
2212         .start  = timers_start,
2213         .next   = timers_next,
2214         .stop   = timers_stop,
2215         .show   = show_timer,
2216 };
2217
2218 static int proc_timers_open(struct inode *inode, struct file *file)
2219 {
2220         struct timers_private *tp;
2221
2222         tp = __seq_open_private(file, &proc_timers_seq_ops,
2223                         sizeof(struct timers_private));
2224         if (!tp)
2225                 return -ENOMEM;
2226
2227         tp->pid = proc_pid(inode);
2228         tp->ns = inode->i_sb->s_fs_info;
2229         return 0;
2230 }
2231
2232 static const struct file_operations proc_timers_operations = {
2233         .open           = proc_timers_open,
2234         .read           = seq_read,
2235         .llseek         = seq_lseek,
2236         .release        = seq_release_private,
2237 };
2238
2239 static int proc_pident_instantiate(struct inode *dir,
2240         struct dentry *dentry, struct task_struct *task, const void *ptr)
2241 {
2242         const struct pid_entry *p = ptr;
2243         struct inode *inode;
2244         struct proc_inode *ei;
2245
2246         inode = proc_pid_make_inode(dir->i_sb, task);
2247         if (!inode)
2248                 goto out;
2249
2250         ei = PROC_I(inode);
2251         inode->i_mode = p->mode;
2252         if (S_ISDIR(inode->i_mode))
2253                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2254         if (p->iop)
2255                 inode->i_op = p->iop;
2256         if (p->fop)
2257                 inode->i_fop = p->fop;
2258         ei->op = p->op;
2259         d_set_d_op(dentry, &pid_dentry_operations);
2260         d_add(dentry, inode);
2261         /* Close the race of the process dying before we return the dentry */
2262         if (pid_revalidate(dentry, 0))
2263                 return 0;
2264 out:
2265         return -ENOENT;
2266 }
2267
2268 static struct dentry *proc_pident_lookup(struct inode *dir,
2269                                          struct dentry *dentry,
2270                                          const struct pid_entry *ents,
2271                                          unsigned int nents)
2272 {
2273         int error;
2274         struct task_struct *task = get_proc_task(dir);
2275         const struct pid_entry *p, *last;
2276
2277         error = -ENOENT;
2278
2279         if (!task)
2280                 goto out_no_task;
2281
2282         /*
2283          * Yes, it does not scale. And it should not. Don't add
2284          * new entries into /proc/<tgid>/ without very good reasons.
2285          */
2286         last = &ents[nents - 1];
2287         for (p = ents; p <= last; p++) {
2288                 if (p->len != dentry->d_name.len)
2289                         continue;
2290                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2291                         break;
2292         }
2293         if (p > last)
2294                 goto out;
2295
2296         error = proc_pident_instantiate(dir, dentry, task, p);
2297 out:
2298         put_task_struct(task);
2299 out_no_task:
2300         return ERR_PTR(error);
2301 }
2302
2303 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2304                 const struct pid_entry *ents, unsigned int nents)
2305 {
2306         struct task_struct *task = get_proc_task(file_inode(file));
2307         const struct pid_entry *p;
2308
2309         if (!task)
2310                 return -ENOENT;
2311
2312         if (!dir_emit_dots(file, ctx))
2313                 goto out;
2314
2315         if (ctx->pos >= nents + 2)
2316                 goto out;
2317
2318         for (p = ents + (ctx->pos - 2); p <= ents + nents - 1; p++) {
2319                 if (!proc_fill_cache(file, ctx, p->name, p->len,
2320                                 proc_pident_instantiate, task, p))
2321                         break;
2322                 ctx->pos++;
2323         }
2324 out:
2325         put_task_struct(task);
2326         return 0;
2327 }
2328
2329 #ifdef CONFIG_SECURITY
2330 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2331                                   size_t count, loff_t *ppos)
2332 {
2333         struct inode * inode = file_inode(file);
2334         char *p = NULL;
2335         ssize_t length;
2336         struct task_struct *task = get_proc_task(inode);
2337
2338         if (!task)
2339                 return -ESRCH;
2340
2341         length = security_getprocattr(task,
2342                                       (char*)file->f_path.dentry->d_name.name,
2343                                       &p);
2344         put_task_struct(task);
2345         if (length > 0)
2346                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2347         kfree(p);
2348         return length;
2349 }
2350
2351 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2352                                    size_t count, loff_t *ppos)
2353 {
2354         struct inode * inode = file_inode(file);
2355         char *page;
2356         ssize_t length;
2357         struct task_struct *task = get_proc_task(inode);
2358
2359         length = -ESRCH;
2360         if (!task)
2361                 goto out_no_task;
2362         if (count > PAGE_SIZE)
2363                 count = PAGE_SIZE;
2364
2365         /* No partial writes. */
2366         length = -EINVAL;
2367         if (*ppos != 0)
2368                 goto out;
2369
2370         length = -ENOMEM;
2371         page = (char*)__get_free_page(GFP_TEMPORARY);
2372         if (!page)
2373                 goto out;
2374
2375         length = -EFAULT;
2376         if (copy_from_user(page, buf, count))
2377                 goto out_free;
2378
2379         /* Guard against adverse ptrace interaction */
2380         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2381         if (length < 0)
2382                 goto out_free;
2383
2384         length = security_setprocattr(task,
2385                                       (char*)file->f_path.dentry->d_name.name,
2386                                       (void*)page, count);
2387         mutex_unlock(&task->signal->cred_guard_mutex);
2388 out_free:
2389         free_page((unsigned long) page);
2390 out:
2391         put_task_struct(task);
2392 out_no_task:
2393         return length;
2394 }
2395
2396 static const struct file_operations proc_pid_attr_operations = {
2397         .read           = proc_pid_attr_read,
2398         .write          = proc_pid_attr_write,
2399         .llseek         = generic_file_llseek,
2400 };
2401
2402 static const struct pid_entry attr_dir_stuff[] = {
2403         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2404         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2405         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2406         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2407         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2408         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2409 };
2410
2411 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2412 {
2413         return proc_pident_readdir(file, ctx,
2414                                    attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2415 }
2416
2417 static const struct file_operations proc_attr_dir_operations = {
2418         .read           = generic_read_dir,
2419         .iterate        = proc_attr_dir_readdir,
2420         .llseek         = default_llseek,
2421 };
2422
2423 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2424                                 struct dentry *dentry, unsigned int flags)
2425 {
2426         return proc_pident_lookup(dir, dentry,
2427                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2428 }
2429
2430 static const struct inode_operations proc_attr_dir_inode_operations = {
2431         .lookup         = proc_attr_dir_lookup,
2432         .getattr        = pid_getattr,
2433         .setattr        = proc_setattr,
2434 };
2435
2436 #endif
2437
2438 #ifdef CONFIG_ELF_CORE
2439 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2440                                          size_t count, loff_t *ppos)
2441 {
2442         struct task_struct *task = get_proc_task(file_inode(file));
2443         struct mm_struct *mm;
2444         char buffer[PROC_NUMBUF];
2445         size_t len;
2446         int ret;
2447
2448         if (!task)
2449                 return -ESRCH;
2450
2451         ret = 0;
2452         mm = get_task_mm(task);
2453         if (mm) {
2454                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2455                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2456                                 MMF_DUMP_FILTER_SHIFT));
2457                 mmput(mm);
2458                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2459         }
2460
2461         put_task_struct(task);
2462
2463         return ret;
2464 }
2465
2466 static ssize_t proc_coredump_filter_write(struct file *file,
2467                                           const char __user *buf,
2468                                           size_t count,
2469                                           loff_t *ppos)
2470 {
2471         struct task_struct *task;
2472         struct mm_struct *mm;
2473         unsigned int val;
2474         int ret;
2475         int i;
2476         unsigned long mask;
2477
2478         ret = kstrtouint_from_user(buf, count, 0, &val);
2479         if (ret < 0)
2480                 return ret;
2481
2482         ret = -ESRCH;
2483         task = get_proc_task(file_inode(file));
2484         if (!task)
2485                 goto out_no_task;
2486
2487         mm = get_task_mm(task);
2488         if (!mm)
2489                 goto out_no_mm;
2490
2491         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2492                 if (val & mask)
2493                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2494                 else
2495                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2496         }
2497
2498         mmput(mm);
2499  out_no_mm:
2500         put_task_struct(task);
2501  out_no_task:
2502         if (ret < 0)
2503                 return ret;
2504         return count;
2505 }
2506
2507 static const struct file_operations proc_coredump_filter_operations = {
2508         .read           = proc_coredump_filter_read,
2509         .write          = proc_coredump_filter_write,
2510         .llseek         = generic_file_llseek,
2511 };
2512 #endif
2513
2514 #ifdef CONFIG_TASK_IO_ACCOUNTING
2515 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2516 {
2517         struct task_io_accounting acct = task->ioac;
2518         unsigned long flags;
2519         int result;
2520
2521         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2522         if (result)
2523                 return result;
2524
2525         if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2526                 result = -EACCES;
2527                 goto out_unlock;
2528         }
2529
2530         if (whole && lock_task_sighand(task, &flags)) {
2531                 struct task_struct *t = task;
2532
2533                 task_io_accounting_add(&acct, &task->signal->ioac);
2534                 while_each_thread(task, t)
2535                         task_io_accounting_add(&acct, &t->ioac);
2536
2537                 unlock_task_sighand(task, &flags);
2538         }
2539         seq_printf(m,
2540                    "rchar: %llu\n"
2541                    "wchar: %llu\n"
2542                    "syscr: %llu\n"
2543                    "syscw: %llu\n"
2544                    "read_bytes: %llu\n"
2545                    "write_bytes: %llu\n"
2546                    "cancelled_write_bytes: %llu\n",
2547                    (unsigned long long)acct.rchar,
2548                    (unsigned long long)acct.wchar,
2549                    (unsigned long long)acct.syscr,
2550                    (unsigned long long)acct.syscw,
2551                    (unsigned long long)acct.read_bytes,
2552                    (unsigned long long)acct.write_bytes,
2553                    (unsigned long long)acct.cancelled_write_bytes);
2554         result = 0;
2555
2556 out_unlock:
2557         mutex_unlock(&task->signal->cred_guard_mutex);
2558         return result;
2559 }
2560
2561 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2562                                   struct pid *pid, struct task_struct *task)
2563 {
2564         return do_io_accounting(task, m, 0);
2565 }
2566
2567 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2568                                    struct pid *pid, struct task_struct *task)
2569 {
2570         return do_io_accounting(task, m, 1);
2571 }
2572 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2573
2574 #ifdef CONFIG_USER_NS
2575 static int proc_id_map_open(struct inode *inode, struct file *file,
2576         const struct seq_operations *seq_ops)
2577 {
2578         struct user_namespace *ns = NULL;
2579         struct task_struct *task;
2580         struct seq_file *seq;
2581         int ret = -EINVAL;
2582
2583         task = get_proc_task(inode);
2584         if (task) {
2585                 rcu_read_lock();
2586                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2587                 rcu_read_unlock();
2588                 put_task_struct(task);
2589         }
2590         if (!ns)
2591                 goto err;
2592
2593         ret = seq_open(file, seq_ops);
2594         if (ret)
2595                 goto err_put_ns;
2596
2597         seq = file->private_data;
2598         seq->private = ns;
2599
2600         return 0;
2601 err_put_ns:
2602         put_user_ns(ns);
2603 err:
2604         return ret;
2605 }
2606
2607 static int proc_id_map_release(struct inode *inode, struct file *file)
2608 {
2609         struct seq_file *seq = file->private_data;
2610         struct user_namespace *ns = seq->private;
2611         put_user_ns(ns);
2612         return seq_release(inode, file);
2613 }
2614
2615 static int proc_uid_map_open(struct inode *inode, struct file *file)
2616 {
2617         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2618 }
2619
2620 static int proc_gid_map_open(struct inode *inode, struct file *file)
2621 {
2622         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2623 }
2624
2625 static int proc_projid_map_open(struct inode *inode, struct file *file)
2626 {
2627         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2628 }
2629
2630 static const struct file_operations proc_uid_map_operations = {
2631         .open           = proc_uid_map_open,
2632         .write          = proc_uid_map_write,
2633         .read           = seq_read,
2634         .llseek         = seq_lseek,
2635         .release        = proc_id_map_release,
2636 };
2637
2638 static const struct file_operations proc_gid_map_operations = {
2639         .open           = proc_gid_map_open,
2640         .write          = proc_gid_map_write,
2641         .read           = seq_read,
2642         .llseek         = seq_lseek,
2643         .release        = proc_id_map_release,
2644 };
2645
2646 static const struct file_operations proc_projid_map_operations = {
2647         .open           = proc_projid_map_open,
2648         .write          = proc_projid_map_write,
2649         .read           = seq_read,
2650         .llseek         = seq_lseek,
2651         .release        = proc_id_map_release,
2652 };
2653
2654 static int proc_setgroups_open(struct inode *inode, struct file *file)
2655 {
2656         struct user_namespace *ns = NULL;
2657         struct task_struct *task;
2658         int ret;
2659
2660         ret = -ESRCH;
2661         task = get_proc_task(inode);
2662         if (task) {
2663                 rcu_read_lock();
2664                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2665                 rcu_read_unlock();
2666                 put_task_struct(task);
2667         }
2668         if (!ns)
2669                 goto err;
2670
2671         if (file->f_mode & FMODE_WRITE) {
2672                 ret = -EACCES;
2673                 if (!ns_capable(ns, CAP_SYS_ADMIN))
2674                         goto err_put_ns;
2675         }
2676
2677         ret = single_open(file, &proc_setgroups_show, ns);
2678         if (ret)
2679                 goto err_put_ns;
2680
2681         return 0;
2682 err_put_ns:
2683         put_user_ns(ns);
2684 err:
2685         return ret;
2686 }
2687
2688 static int proc_setgroups_release(struct inode *inode, struct file *file)
2689 {
2690         struct seq_file *seq = file->private_data;
2691         struct user_namespace *ns = seq->private;
2692         int ret = single_release(inode, file);
2693         put_user_ns(ns);
2694         return ret;
2695 }
2696
2697 static const struct file_operations proc_setgroups_operations = {
2698         .open           = proc_setgroups_open,
2699         .write          = proc_setgroups_write,
2700         .read           = seq_read,
2701         .llseek         = seq_lseek,
2702         .release        = proc_setgroups_release,
2703 };
2704 #endif /* CONFIG_USER_NS */
2705
2706 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2707                                 struct pid *pid, struct task_struct *task)
2708 {
2709         int err = lock_trace(task);
2710         if (!err) {
2711                 seq_printf(m, "%08x\n", task->personality);
2712                 unlock_trace(task);
2713         }
2714         return err;
2715 }
2716
2717 /*
2718  * Thread groups
2719  */
2720 static const struct file_operations proc_task_operations;
2721 static const struct inode_operations proc_task_inode_operations;
2722
2723 static const struct pid_entry tgid_base_stuff[] = {
2724         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2725         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2726         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2727         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2728         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2729 #ifdef CONFIG_NET
2730         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2731 #endif
2732         REG("environ",    S_IRUSR, proc_environ_operations),
2733         ONE("auxv",       S_IRUSR, proc_pid_auxv),
2734         ONE("status",     S_IRUGO, proc_pid_status),
2735         ONE("personality", S_IRUSR, proc_pid_personality),
2736         ONE("limits",     S_IRUGO, proc_pid_limits),
2737 #ifdef CONFIG_SCHED_DEBUG
2738         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2739 #endif
2740 #ifdef CONFIG_SCHED_AUTOGROUP
2741         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2742 #endif
2743         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2744 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2745         ONE("syscall",    S_IRUSR, proc_pid_syscall),
2746 #endif
2747         REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
2748         ONE("stat",       S_IRUGO, proc_tgid_stat),
2749         ONE("statm",      S_IRUGO, proc_pid_statm),
2750         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2751 #ifdef CONFIG_NUMA
2752         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2753 #endif
2754         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2755         LNK("cwd",        proc_cwd_link),
2756         LNK("root",       proc_root_link),
2757         LNK("exe",        proc_exe_link),
2758         REG("mounts",     S_IRUGO, proc_mounts_operations),
2759         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2760         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2761 #ifdef CONFIG_PROC_PAGE_MONITOR
2762         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2763         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2764         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2765 #endif
2766 #ifdef CONFIG_SECURITY
2767         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2768 #endif
2769 #ifdef CONFIG_KALLSYMS
2770         ONE("wchan",      S_IRUGO, proc_pid_wchan),
2771 #endif
2772 #ifdef CONFIG_STACKTRACE
2773         ONE("stack",      S_IRUSR, proc_pid_stack),
2774 #endif
2775 #ifdef CONFIG_SCHED_INFO
2776         ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
2777 #endif
2778 #ifdef CONFIG_LATENCYTOP
2779         REG("latency",  S_IRUGO, proc_lstats_operations),
2780 #endif
2781 #ifdef CONFIG_PROC_PID_CPUSET
2782         ONE("cpuset",     S_IRUGO, proc_cpuset_show),
2783 #endif
2784 #ifdef CONFIG_CGROUPS
2785         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
2786 #endif
2787         ONE("oom_score",  S_IRUGO, proc_oom_score),
2788         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2789         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2790 #ifdef CONFIG_AUDITSYSCALL
2791         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2792         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2793 #endif
2794 #ifdef CONFIG_FAULT_INJECTION
2795         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2796 #endif
2797 #ifdef CONFIG_ELF_CORE
2798         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2799 #endif
2800 #ifdef CONFIG_TASK_IO_ACCOUNTING
2801         ONE("io",       S_IRUSR, proc_tgid_io_accounting),
2802 #endif
2803 #ifdef CONFIG_HARDWALL
2804         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
2805 #endif
2806 #ifdef CONFIG_USER_NS
2807         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2808         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2809         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2810         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
2811 #endif
2812 #ifdef CONFIG_CHECKPOINT_RESTORE
2813         REG("timers",     S_IRUGO, proc_timers_operations),
2814 #endif
2815 };
2816
2817 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2818 {
2819         return proc_pident_readdir(file, ctx,
2820                                    tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2821 }
2822
2823 static const struct file_operations proc_tgid_base_operations = {
2824         .read           = generic_read_dir,
2825         .iterate        = proc_tgid_base_readdir,
2826         .llseek         = default_llseek,
2827 };
2828
2829 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2830 {
2831         return proc_pident_lookup(dir, dentry,
2832                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2833 }
2834
2835 static const struct inode_operations proc_tgid_base_inode_operations = {
2836         .lookup         = proc_tgid_base_lookup,
2837         .getattr        = pid_getattr,
2838         .setattr        = proc_setattr,
2839         .permission     = proc_pid_permission,
2840 };
2841
2842 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2843 {
2844         struct dentry *dentry, *leader, *dir;
2845         char buf[PROC_NUMBUF];
2846         struct qstr name;
2847
2848         name.name = buf;
2849         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2850         /* no ->d_hash() rejects on procfs */
2851         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2852         if (dentry) {
2853                 d_invalidate(dentry);
2854                 dput(dentry);
2855         }
2856
2857         if (pid == tgid)
2858                 return;
2859
2860         name.name = buf;
2861         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2862         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2863         if (!leader)
2864                 goto out;
2865
2866         name.name = "task";
2867         name.len = strlen(name.name);
2868         dir = d_hash_and_lookup(leader, &name);
2869         if (!dir)
2870                 goto out_put_leader;
2871
2872         name.name = buf;
2873         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2874         dentry = d_hash_and_lookup(dir, &name);
2875         if (dentry) {
2876                 d_invalidate(dentry);
2877                 dput(dentry);
2878         }
2879
2880         dput(dir);
2881 out_put_leader:
2882         dput(leader);
2883 out:
2884         return;
2885 }
2886
2887 /**
2888  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2889  * @task: task that should be flushed.
2890  *
2891  * When flushing dentries from proc, one needs to flush them from global
2892  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2893  * in. This call is supposed to do all of this job.
2894  *
2895  * Looks in the dcache for
2896  * /proc/@pid
2897  * /proc/@tgid/task/@pid
2898  * if either directory is present flushes it and all of it'ts children
2899  * from the dcache.
2900  *
2901  * It is safe and reasonable to cache /proc entries for a task until
2902  * that task exits.  After that they just clog up the dcache with
2903  * useless entries, possibly causing useful dcache entries to be
2904  * flushed instead.  This routine is proved to flush those useless
2905  * dcache entries at process exit time.
2906  *
2907  * NOTE: This routine is just an optimization so it does not guarantee
2908  *       that no dcache entries will exist at process exit time it
2909  *       just makes it very unlikely that any will persist.
2910  */
2911
2912 void proc_flush_task(struct task_struct *task)
2913 {
2914         int i;
2915         struct pid *pid, *tgid;
2916         struct upid *upid;
2917
2918         pid = task_pid(task);
2919         tgid = task_tgid(task);
2920
2921         for (i = 0; i <= pid->level; i++) {
2922                 upid = &pid->numbers[i];
2923                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2924                                         tgid->numbers[i].nr);
2925         }
2926 }
2927
2928 static int proc_pid_instantiate(struct inode *dir,
2929                                    struct dentry * dentry,
2930                                    struct task_struct *task, const void *ptr)
2931 {
2932         struct inode *inode;
2933
2934         inode = proc_pid_make_inode(dir->i_sb, task);
2935         if (!inode)
2936                 goto out;
2937
2938         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2939         inode->i_op = &proc_tgid_base_inode_operations;
2940         inode->i_fop = &proc_tgid_base_operations;
2941         inode->i_flags|=S_IMMUTABLE;
2942
2943         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
2944                                                   ARRAY_SIZE(tgid_base_stuff)));
2945
2946         d_set_d_op(dentry, &pid_dentry_operations);
2947
2948         d_add(dentry, inode);
2949         /* Close the race of the process dying before we return the dentry */
2950         if (pid_revalidate(dentry, 0))
2951                 return 0;
2952 out:
2953         return -ENOENT;
2954 }
2955
2956 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
2957 {
2958         int result = -ENOENT;
2959         struct task_struct *task;
2960         unsigned tgid;
2961         struct pid_namespace *ns;
2962
2963         tgid = name_to_int(&dentry->d_name);
2964         if (tgid == ~0U)
2965                 goto out;
2966
2967         ns = dentry->d_sb->s_fs_info;
2968         rcu_read_lock();
2969         task = find_task_by_pid_ns(tgid, ns);
2970         if (task)
2971                 get_task_struct(task);
2972         rcu_read_unlock();
2973         if (!task)
2974                 goto out;
2975
2976         result = proc_pid_instantiate(dir, dentry, task, NULL);
2977         put_task_struct(task);
2978 out:
2979         return ERR_PTR(result);
2980 }
2981
2982 /*
2983  * Find the first task with tgid >= tgid
2984  *
2985  */
2986 struct tgid_iter {
2987         unsigned int tgid;
2988         struct task_struct *task;
2989 };
2990 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2991 {
2992         struct pid *pid;
2993
2994         if (iter.task)
2995                 put_task_struct(iter.task);
2996         rcu_read_lock();
2997 retry:
2998         iter.task = NULL;
2999         pid = find_ge_pid(iter.tgid, ns);
3000         if (pid) {
3001                 iter.tgid = pid_nr_ns(pid, ns);
3002                 iter.task = pid_task(pid, PIDTYPE_PID);
3003                 /* What we to know is if the pid we have find is the
3004                  * pid of a thread_group_leader.  Testing for task
3005                  * being a thread_group_leader is the obvious thing
3006                  * todo but there is a window when it fails, due to
3007                  * the pid transfer logic in de_thread.
3008                  *
3009                  * So we perform the straight forward test of seeing
3010                  * if the pid we have found is the pid of a thread
3011                  * group leader, and don't worry if the task we have
3012                  * found doesn't happen to be a thread group leader.
3013                  * As we don't care in the case of readdir.
3014                  */
3015                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3016                         iter.tgid += 1;
3017                         goto retry;
3018                 }
3019                 get_task_struct(iter.task);
3020         }
3021         rcu_read_unlock();
3022         return iter;
3023 }
3024
3025 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3026
3027 /* for the /proc/ directory itself, after non-process stuff has been done */
3028 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3029 {
3030         struct tgid_iter iter;
3031         struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
3032         loff_t pos = ctx->pos;
3033
3034         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3035                 return 0;
3036
3037         if (pos == TGID_OFFSET - 2) {
3038                 struct inode *inode = d_inode(ns->proc_self);
3039                 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3040                         return 0;
3041                 ctx->pos = pos = pos + 1;
3042         }
3043         if (pos == TGID_OFFSET - 1) {
3044                 struct inode *inode = d_inode(ns->proc_thread_self);
3045                 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3046                         return 0;
3047                 ctx->pos = pos = pos + 1;
3048         }
3049         iter.tgid = pos - TGID_OFFSET;
3050         iter.task = NULL;
3051         for (iter = next_tgid(ns, iter);
3052              iter.task;
3053              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3054                 char name[PROC_NUMBUF];
3055                 int len;
3056                 if (!has_pid_permissions(ns, iter.task, 2))
3057                         continue;
3058
3059                 len = snprintf(name, sizeof(name), "%d", iter.tgid);
3060                 ctx->pos = iter.tgid + TGID_OFFSET;
3061                 if (!proc_fill_cache(file, ctx, name, len,
3062                                      proc_pid_instantiate, iter.task, NULL)) {
3063                         put_task_struct(iter.task);
3064                         return 0;
3065                 }
3066         }
3067         ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3068         return 0;
3069 }
3070
3071 /*
3072  * Tasks
3073  */
3074 static const struct pid_entry tid_base_stuff[] = {
3075         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3076         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3077         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3078 #ifdef CONFIG_NET
3079         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3080 #endif
3081         REG("environ",   S_IRUSR, proc_environ_operations),
3082         ONE("auxv",      S_IRUSR, proc_pid_auxv),
3083         ONE("status",    S_IRUGO, proc_pid_status),
3084         ONE("personality", S_IRUSR, proc_pid_personality),
3085         ONE("limits",    S_IRUGO, proc_pid_limits),
3086 #ifdef CONFIG_SCHED_DEBUG
3087         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3088 #endif
3089         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3090 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3091         ONE("syscall",   S_IRUSR, proc_pid_syscall),
3092 #endif
3093         REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3094         ONE("stat",      S_IRUGO, proc_tid_stat),
3095         ONE("statm",     S_IRUGO, proc_pid_statm),
3096         REG("maps",      S_IRUGO, proc_tid_maps_operations),
3097 #ifdef CONFIG_PROC_CHILDREN
3098         REG("children",  S_IRUGO, proc_tid_children_operations),
3099 #endif
3100 #ifdef CONFIG_NUMA
3101         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3102 #endif
3103         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3104         LNK("cwd",       proc_cwd_link),
3105         LNK("root",      proc_root_link),
3106         LNK("exe",       proc_exe_link),
3107         REG("mounts",    S_IRUGO, proc_mounts_operations),
3108         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3109 #ifdef CONFIG_PROC_PAGE_MONITOR
3110         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3111         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3112         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3113 #endif
3114 #ifdef CONFIG_SECURITY
3115         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3116 #endif
3117 #ifdef CONFIG_KALLSYMS
3118         ONE("wchan",     S_IRUGO, proc_pid_wchan),
3119 #endif
3120 #ifdef CONFIG_STACKTRACE
3121         ONE("stack",      S_IRUSR, proc_pid_stack),
3122 #endif
3123 #ifdef CONFIG_SCHED_INFO
3124         ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3125 #endif
3126 #ifdef CONFIG_LATENCYTOP
3127         REG("latency",  S_IRUGO, proc_lstats_operations),
3128 #endif
3129 #ifdef CONFIG_PROC_PID_CPUSET
3130         ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3131 #endif
3132 #ifdef CONFIG_CGROUPS
3133         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3134 #endif
3135         ONE("oom_score", S_IRUGO, proc_oom_score),
3136         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3137         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3138 #ifdef CONFIG_AUDITSYSCALL
3139         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3140         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3141 #endif
3142 #ifdef CONFIG_FAULT_INJECTION
3143         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3144 #endif
3145 #ifdef CONFIG_TASK_IO_ACCOUNTING
3146         ONE("io",       S_IRUSR, proc_tid_io_accounting),
3147 #endif
3148 #ifdef CONFIG_HARDWALL
3149         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
3150 #endif
3151 #ifdef CONFIG_USER_NS
3152         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3153         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3154         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3155         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3156 #endif
3157 };
3158
3159 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3160 {
3161         return proc_pident_readdir(file, ctx,
3162                                    tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3163 }
3164
3165 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3166 {
3167         return proc_pident_lookup(dir, dentry,
3168                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3169 }
3170
3171 static const struct file_operations proc_tid_base_operations = {
3172         .read           = generic_read_dir,
3173         .iterate        = proc_tid_base_readdir,
3174         .llseek         = default_llseek,
3175 };
3176
3177 static const struct inode_operations proc_tid_base_inode_operations = {
3178         .lookup         = proc_tid_base_lookup,
3179         .getattr        = pid_getattr,
3180         .setattr        = proc_setattr,
3181 };
3182
3183 static int proc_task_instantiate(struct inode *dir,
3184         struct dentry *dentry, struct task_struct *task, const void *ptr)
3185 {
3186         struct inode *inode;
3187         inode = proc_pid_make_inode(dir->i_sb, task);
3188
3189         if (!inode)
3190                 goto out;
3191         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3192         inode->i_op = &proc_tid_base_inode_operations;
3193         inode->i_fop = &proc_tid_base_operations;
3194         inode->i_flags|=S_IMMUTABLE;
3195
3196         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3197                                                   ARRAY_SIZE(tid_base_stuff)));
3198
3199         d_set_d_op(dentry, &pid_dentry_operations);
3200
3201         d_add(dentry, inode);
3202         /* Close the race of the process dying before we return the dentry */
3203         if (pid_revalidate(dentry, 0))
3204                 return 0;
3205 out:
3206         return -ENOENT;
3207 }
3208
3209 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3210 {
3211         int result = -ENOENT;
3212         struct task_struct *task;
3213         struct task_struct *leader = get_proc_task(dir);
3214         unsigned tid;
3215         struct pid_namespace *ns;
3216
3217         if (!leader)
3218                 goto out_no_task;
3219
3220         tid = name_to_int(&dentry->d_name);
3221         if (tid == ~0U)
3222                 goto out;
3223
3224         ns = dentry->d_sb->s_fs_info;
3225         rcu_read_lock();
3226         task = find_task_by_pid_ns(tid, ns);
3227         if (task)
3228                 get_task_struct(task);
3229         rcu_read_unlock();
3230         if (!task)
3231                 goto out;
3232         if (!same_thread_group(leader, task))
3233                 goto out_drop_task;
3234
3235         result = proc_task_instantiate(dir, dentry, task, NULL);
3236 out_drop_task:
3237         put_task_struct(task);
3238 out:
3239         put_task_struct(leader);
3240 out_no_task:
3241         return ERR_PTR(result);
3242 }
3243
3244 /*
3245  * Find the first tid of a thread group to return to user space.
3246  *
3247  * Usually this is just the thread group leader, but if the users
3248  * buffer was too small or there was a seek into the middle of the
3249  * directory we have more work todo.
3250  *
3251  * In the case of a short read we start with find_task_by_pid.
3252  *
3253  * In the case of a seek we start with the leader and walk nr
3254  * threads past it.
3255  */
3256 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3257                                         struct pid_namespace *ns)
3258 {
3259         struct task_struct *pos, *task;
3260         unsigned long nr = f_pos;
3261
3262         if (nr != f_pos)        /* 32bit overflow? */
3263                 return NULL;
3264
3265         rcu_read_lock();
3266         task = pid_task(pid, PIDTYPE_PID);
3267         if (!task)
3268                 goto fail;
3269
3270         /* Attempt to start with the tid of a thread */
3271         if (tid && nr) {
3272                 pos = find_task_by_pid_ns(tid, ns);
3273                 if (pos && same_thread_group(pos, task))
3274                         goto found;
3275         }
3276
3277         /* If nr exceeds the number of threads there is nothing todo */
3278         if (nr >= get_nr_threads(task))
3279                 goto fail;
3280
3281         /* If we haven't found our starting place yet start
3282          * with the leader and walk nr threads forward.
3283          */
3284         pos = task = task->group_leader;
3285         do {
3286                 if (!nr--)
3287                         goto found;
3288         } while_each_thread(task, pos);
3289 fail:
3290         pos = NULL;
3291         goto out;
3292 found:
3293         get_task_struct(pos);
3294 out:
3295         rcu_read_unlock();
3296         return pos;
3297 }
3298
3299 /*
3300  * Find the next thread in the thread list.
3301  * Return NULL if there is an error or no next thread.
3302  *
3303  * The reference to the input task_struct is released.
3304  */
3305 static struct task_struct *next_tid(struct task_struct *start)
3306 {
3307         struct task_struct *pos = NULL;
3308         rcu_read_lock();
3309         if (pid_alive(start)) {
3310                 pos = next_thread(start);
3311                 if (thread_group_leader(pos))
3312                         pos = NULL;
3313                 else
3314                         get_task_struct(pos);
3315         }
3316         rcu_read_unlock();
3317         put_task_struct(start);
3318         return pos;
3319 }
3320
3321 /* for the /proc/TGID/task/ directories */
3322 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3323 {
3324         struct inode *inode = file_inode(file);
3325         struct task_struct *task;
3326         struct pid_namespace *ns;
3327         int tid;
3328
3329         if (proc_inode_is_dead(inode))
3330                 return -ENOENT;
3331
3332         if (!dir_emit_dots(file, ctx))
3333                 return 0;
3334
3335         /* f_version caches the tgid value that the last readdir call couldn't
3336          * return. lseek aka telldir automagically resets f_version to 0.
3337          */
3338         ns = inode->i_sb->s_fs_info;
3339         tid = (int)file->f_version;
3340         file->f_version = 0;
3341         for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3342              task;
3343              task = next_tid(task), ctx->pos++) {
3344                 char name[PROC_NUMBUF];
3345                 int len;
3346                 tid = task_pid_nr_ns(task, ns);
3347                 len = snprintf(name, sizeof(name), "%d", tid);
3348                 if (!proc_fill_cache(file, ctx, name, len,
3349                                 proc_task_instantiate, task, NULL)) {
3350                         /* returning this tgid failed, save it as the first
3351                          * pid for the next readir call */
3352                         file->f_version = (u64)tid;
3353                         put_task_struct(task);
3354                         break;
3355                 }
3356         }
3357
3358         return 0;
3359 }
3360
3361 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3362 {
3363         struct inode *inode = d_inode(dentry);
3364         struct task_struct *p = get_proc_task(inode);
3365         generic_fillattr(inode, stat);
3366
3367         if (p) {
3368                 stat->nlink += get_nr_threads(p);
3369                 put_task_struct(p);
3370         }
3371
3372         return 0;
3373 }
3374
3375 static const struct inode_operations proc_task_inode_operations = {
3376         .lookup         = proc_task_lookup,
3377         .getattr        = proc_task_getattr,
3378         .setattr        = proc_setattr,
3379         .permission     = proc_pid_permission,
3380 };
3381
3382 static const struct file_operations proc_task_operations = {
3383         .read           = generic_read_dir,
3384         .iterate        = proc_task_readdir,
3385         .llseek         = default_llseek,
3386 };