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         char *page, *tmp;
1234         ssize_t length;
1235         uid_t loginuid;
1236         kuid_t kloginuid;
1237
1238         rcu_read_lock();
1239         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1240                 rcu_read_unlock();
1241                 return -EPERM;
1242         }
1243         rcu_read_unlock();
1244
1245         if (count >= PAGE_SIZE)
1246                 count = PAGE_SIZE - 1;
1247
1248         if (*ppos != 0) {
1249                 /* No partial writes. */
1250                 return -EINVAL;
1251         }
1252         page = (char*)__get_free_page(GFP_TEMPORARY);
1253         if (!page)
1254                 return -ENOMEM;
1255         length = -EFAULT;
1256         if (copy_from_user(page, buf, count))
1257                 goto out_free_page;
1258
1259         page[count] = '\0';
1260         loginuid = simple_strtoul(page, &tmp, 10);
1261         if (tmp == page) {
1262                 length = -EINVAL;
1263                 goto out_free_page;
1264
1265         }
1266
1267         /* is userspace tring to explicitly UNSET the loginuid? */
1268         if (loginuid == AUDIT_UID_UNSET) {
1269                 kloginuid = INVALID_UID;
1270         } else {
1271                 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1272                 if (!uid_valid(kloginuid)) {
1273                         length = -EINVAL;
1274                         goto out_free_page;
1275                 }
1276         }
1277
1278         length = audit_set_loginuid(kloginuid);
1279         if (likely(length == 0))
1280                 length = count;
1281
1282 out_free_page:
1283         free_page((unsigned long) page);
1284         return length;
1285 }
1286
1287 static const struct file_operations proc_loginuid_operations = {
1288         .read           = proc_loginuid_read,
1289         .write          = proc_loginuid_write,
1290         .llseek         = generic_file_llseek,
1291 };
1292
1293 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1294                                   size_t count, loff_t *ppos)
1295 {
1296         struct inode * inode = file_inode(file);
1297         struct task_struct *task = get_proc_task(inode);
1298         ssize_t length;
1299         char tmpbuf[TMPBUFLEN];
1300
1301         if (!task)
1302                 return -ESRCH;
1303         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1304                                 audit_get_sessionid(task));
1305         put_task_struct(task);
1306         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1307 }
1308
1309 static const struct file_operations proc_sessionid_operations = {
1310         .read           = proc_sessionid_read,
1311         .llseek         = generic_file_llseek,
1312 };
1313 #endif
1314
1315 #ifdef CONFIG_FAULT_INJECTION
1316 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1317                                       size_t count, loff_t *ppos)
1318 {
1319         struct task_struct *task = get_proc_task(file_inode(file));
1320         char buffer[PROC_NUMBUF];
1321         size_t len;
1322         int make_it_fail;
1323
1324         if (!task)
1325                 return -ESRCH;
1326         make_it_fail = task->make_it_fail;
1327         put_task_struct(task);
1328
1329         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1330
1331         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1332 }
1333
1334 static ssize_t proc_fault_inject_write(struct file * file,
1335                         const char __user * buf, size_t count, loff_t *ppos)
1336 {
1337         struct task_struct *task;
1338         char buffer[PROC_NUMBUF], *end;
1339         int make_it_fail;
1340
1341         if (!capable(CAP_SYS_RESOURCE))
1342                 return -EPERM;
1343         memset(buffer, 0, sizeof(buffer));
1344         if (count > sizeof(buffer) - 1)
1345                 count = sizeof(buffer) - 1;
1346         if (copy_from_user(buffer, buf, count))
1347                 return -EFAULT;
1348         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1349         if (*end)
1350                 return -EINVAL;
1351         if (make_it_fail < 0 || make_it_fail > 1)
1352                 return -EINVAL;
1353
1354         task = get_proc_task(file_inode(file));
1355         if (!task)
1356                 return -ESRCH;
1357         task->make_it_fail = make_it_fail;
1358         put_task_struct(task);
1359
1360         return count;
1361 }
1362
1363 static const struct file_operations proc_fault_inject_operations = {
1364         .read           = proc_fault_inject_read,
1365         .write          = proc_fault_inject_write,
1366         .llseek         = generic_file_llseek,
1367 };
1368 #endif
1369
1370
1371 #ifdef CONFIG_SCHED_DEBUG
1372 /*
1373  * Print out various scheduling related per-task fields:
1374  */
1375 static int sched_show(struct seq_file *m, void *v)
1376 {
1377         struct inode *inode = m->private;
1378         struct task_struct *p;
1379
1380         p = get_proc_task(inode);
1381         if (!p)
1382                 return -ESRCH;
1383         proc_sched_show_task(p, m);
1384
1385         put_task_struct(p);
1386
1387         return 0;
1388 }
1389
1390 static ssize_t
1391 sched_write(struct file *file, const char __user *buf,
1392             size_t count, loff_t *offset)
1393 {
1394         struct inode *inode = file_inode(file);
1395         struct task_struct *p;
1396
1397         p = get_proc_task(inode);
1398         if (!p)
1399                 return -ESRCH;
1400         proc_sched_set_task(p);
1401
1402         put_task_struct(p);
1403
1404         return count;
1405 }
1406
1407 static int sched_open(struct inode *inode, struct file *filp)
1408 {
1409         return single_open(filp, sched_show, inode);
1410 }
1411
1412 static const struct file_operations proc_pid_sched_operations = {
1413         .open           = sched_open,
1414         .read           = seq_read,
1415         .write          = sched_write,
1416         .llseek         = seq_lseek,
1417         .release        = single_release,
1418 };
1419
1420 #endif
1421
1422 #ifdef CONFIG_SCHED_AUTOGROUP
1423 /*
1424  * Print out autogroup related information:
1425  */
1426 static int sched_autogroup_show(struct seq_file *m, void *v)
1427 {
1428         struct inode *inode = m->private;
1429         struct task_struct *p;
1430
1431         p = get_proc_task(inode);
1432         if (!p)
1433                 return -ESRCH;
1434         proc_sched_autogroup_show_task(p, m);
1435
1436         put_task_struct(p);
1437
1438         return 0;
1439 }
1440
1441 static ssize_t
1442 sched_autogroup_write(struct file *file, const char __user *buf,
1443             size_t count, loff_t *offset)
1444 {
1445         struct inode *inode = file_inode(file);
1446         struct task_struct *p;
1447         char buffer[PROC_NUMBUF];
1448         int nice;
1449         int err;
1450
1451         memset(buffer, 0, sizeof(buffer));
1452         if (count > sizeof(buffer) - 1)
1453                 count = sizeof(buffer) - 1;
1454         if (copy_from_user(buffer, buf, count))
1455                 return -EFAULT;
1456
1457         err = kstrtoint(strstrip(buffer), 0, &nice);
1458         if (err < 0)
1459                 return err;
1460
1461         p = get_proc_task(inode);
1462         if (!p)
1463                 return -ESRCH;
1464
1465         err = proc_sched_autogroup_set_nice(p, nice);
1466         if (err)
1467                 count = err;
1468
1469         put_task_struct(p);
1470
1471         return count;
1472 }
1473
1474 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1475 {
1476         int ret;
1477
1478         ret = single_open(filp, sched_autogroup_show, NULL);
1479         if (!ret) {
1480                 struct seq_file *m = filp->private_data;
1481
1482                 m->private = inode;
1483         }
1484         return ret;
1485 }
1486
1487 static const struct file_operations proc_pid_sched_autogroup_operations = {
1488         .open           = sched_autogroup_open,
1489         .read           = seq_read,
1490         .write          = sched_autogroup_write,
1491         .llseek         = seq_lseek,
1492         .release        = single_release,
1493 };
1494
1495 #endif /* CONFIG_SCHED_AUTOGROUP */
1496
1497 static ssize_t comm_write(struct file *file, const char __user *buf,
1498                                 size_t count, loff_t *offset)
1499 {
1500         struct inode *inode = file_inode(file);
1501         struct task_struct *p;
1502         char buffer[TASK_COMM_LEN];
1503         const size_t maxlen = sizeof(buffer) - 1;
1504
1505         memset(buffer, 0, sizeof(buffer));
1506         if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1507                 return -EFAULT;
1508
1509         p = get_proc_task(inode);
1510         if (!p)
1511                 return -ESRCH;
1512
1513         if (same_thread_group(current, p))
1514                 set_task_comm(p, buffer);
1515         else
1516                 count = -EINVAL;
1517
1518         put_task_struct(p);
1519
1520         return count;
1521 }
1522
1523 static int comm_show(struct seq_file *m, void *v)
1524 {
1525         struct inode *inode = m->private;
1526         struct task_struct *p;
1527
1528         p = get_proc_task(inode);
1529         if (!p)
1530                 return -ESRCH;
1531
1532         task_lock(p);
1533         seq_printf(m, "%s\n", p->comm);
1534         task_unlock(p);
1535
1536         put_task_struct(p);
1537
1538         return 0;
1539 }
1540
1541 static int comm_open(struct inode *inode, struct file *filp)
1542 {
1543         return single_open(filp, comm_show, inode);
1544 }
1545
1546 static const struct file_operations proc_pid_set_comm_operations = {
1547         .open           = comm_open,
1548         .read           = seq_read,
1549         .write          = comm_write,
1550         .llseek         = seq_lseek,
1551         .release        = single_release,
1552 };
1553
1554 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1555 {
1556         struct task_struct *task;
1557         struct mm_struct *mm;
1558         struct file *exe_file;
1559
1560         task = get_proc_task(d_inode(dentry));
1561         if (!task)
1562                 return -ENOENT;
1563         mm = get_task_mm(task);
1564         put_task_struct(task);
1565         if (!mm)
1566                 return -ENOENT;
1567         exe_file = get_mm_exe_file(mm);
1568         mmput(mm);
1569         if (exe_file) {
1570                 *exe_path = exe_file->f_path;
1571                 path_get(&exe_file->f_path);
1572                 fput(exe_file);
1573                 return 0;
1574         } else
1575                 return -ENOENT;
1576 }
1577
1578 static const char *proc_pid_follow_link(struct dentry *dentry, void **cookie)
1579 {
1580         struct inode *inode = d_inode(dentry);
1581         struct path path;
1582         int error = -EACCES;
1583
1584         /* Are we allowed to snoop on the tasks file descriptors? */
1585         if (!proc_fd_access_allowed(inode))
1586                 goto out;
1587
1588         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1589         if (error)
1590                 goto out;
1591
1592         nd_jump_link(&path);
1593         return NULL;
1594 out:
1595         return ERR_PTR(error);
1596 }
1597
1598 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1599 {
1600         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1601         char *pathname;
1602         int len;
1603
1604         if (!tmp)
1605                 return -ENOMEM;
1606
1607         pathname = d_path(path, tmp, PAGE_SIZE);
1608         len = PTR_ERR(pathname);
1609         if (IS_ERR(pathname))
1610                 goto out;
1611         len = tmp + PAGE_SIZE - 1 - pathname;
1612
1613         if (len > buflen)
1614                 len = buflen;
1615         if (copy_to_user(buffer, pathname, len))
1616                 len = -EFAULT;
1617  out:
1618         free_page((unsigned long)tmp);
1619         return len;
1620 }
1621
1622 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1623 {
1624         int error = -EACCES;
1625         struct inode *inode = d_inode(dentry);
1626         struct path path;
1627
1628         /* Are we allowed to snoop on the tasks file descriptors? */
1629         if (!proc_fd_access_allowed(inode))
1630                 goto out;
1631
1632         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1633         if (error)
1634                 goto out;
1635
1636         error = do_proc_readlink(&path, buffer, buflen);
1637         path_put(&path);
1638 out:
1639         return error;
1640 }
1641
1642 const struct inode_operations proc_pid_link_inode_operations = {
1643         .readlink       = proc_pid_readlink,
1644         .follow_link    = proc_pid_follow_link,
1645         .setattr        = proc_setattr,
1646 };
1647
1648
1649 /* building an inode */
1650
1651 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1652 {
1653         struct inode * inode;
1654         struct proc_inode *ei;
1655         const struct cred *cred;
1656
1657         /* We need a new inode */
1658
1659         inode = new_inode(sb);
1660         if (!inode)
1661                 goto out;
1662
1663         /* Common stuff */
1664         ei = PROC_I(inode);
1665         inode->i_ino = get_next_ino();
1666         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1667         inode->i_op = &proc_def_inode_operations;
1668
1669         /*
1670          * grab the reference to task.
1671          */
1672         ei->pid = get_task_pid(task, PIDTYPE_PID);
1673         if (!ei->pid)
1674                 goto out_unlock;
1675
1676         if (task_dumpable(task)) {
1677                 rcu_read_lock();
1678                 cred = __task_cred(task);
1679                 inode->i_uid = cred->euid;
1680                 inode->i_gid = cred->egid;
1681                 rcu_read_unlock();
1682         }
1683         security_task_to_inode(task, inode);
1684
1685 out:
1686         return inode;
1687
1688 out_unlock:
1689         iput(inode);
1690         return NULL;
1691 }
1692
1693 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1694 {
1695         struct inode *inode = d_inode(dentry);
1696         struct task_struct *task;
1697         const struct cred *cred;
1698         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1699
1700         generic_fillattr(inode, stat);
1701
1702         rcu_read_lock();
1703         stat->uid = GLOBAL_ROOT_UID;
1704         stat->gid = GLOBAL_ROOT_GID;
1705         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1706         if (task) {
1707                 if (!has_pid_permissions(pid, task, 2)) {
1708                         rcu_read_unlock();
1709                         /*
1710                          * This doesn't prevent learning whether PID exists,
1711                          * it only makes getattr() consistent with readdir().
1712                          */
1713                         return -ENOENT;
1714                 }
1715                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1716                     task_dumpable(task)) {
1717                         cred = __task_cred(task);
1718                         stat->uid = cred->euid;
1719                         stat->gid = cred->egid;
1720                 }
1721         }
1722         rcu_read_unlock();
1723         return 0;
1724 }
1725
1726 /* dentry stuff */
1727
1728 /*
1729  *      Exceptional case: normally we are not allowed to unhash a busy
1730  * directory. In this case, however, we can do it - no aliasing problems
1731  * due to the way we treat inodes.
1732  *
1733  * Rewrite the inode's ownerships here because the owning task may have
1734  * performed a setuid(), etc.
1735  *
1736  * Before the /proc/pid/status file was created the only way to read
1737  * the effective uid of a /process was to stat /proc/pid.  Reading
1738  * /proc/pid/status is slow enough that procps and other packages
1739  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1740  * made this apply to all per process world readable and executable
1741  * directories.
1742  */
1743 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1744 {
1745         struct inode *inode;
1746         struct task_struct *task;
1747         const struct cred *cred;
1748
1749         if (flags & LOOKUP_RCU)
1750                 return -ECHILD;
1751
1752         inode = d_inode(dentry);
1753         task = get_proc_task(inode);
1754
1755         if (task) {
1756                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1757                     task_dumpable(task)) {
1758                         rcu_read_lock();
1759                         cred = __task_cred(task);
1760                         inode->i_uid = cred->euid;
1761                         inode->i_gid = cred->egid;
1762                         rcu_read_unlock();
1763                 } else {
1764                         inode->i_uid = GLOBAL_ROOT_UID;
1765                         inode->i_gid = GLOBAL_ROOT_GID;
1766                 }
1767                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1768                 security_task_to_inode(task, inode);
1769                 put_task_struct(task);
1770                 return 1;
1771         }
1772         return 0;
1773 }
1774
1775 static inline bool proc_inode_is_dead(struct inode *inode)
1776 {
1777         return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1778 }
1779
1780 int pid_delete_dentry(const struct dentry *dentry)
1781 {
1782         /* Is the task we represent dead?
1783          * If so, then don't put the dentry on the lru list,
1784          * kill it immediately.
1785          */
1786         return proc_inode_is_dead(d_inode(dentry));
1787 }
1788
1789 const struct dentry_operations pid_dentry_operations =
1790 {
1791         .d_revalidate   = pid_revalidate,
1792         .d_delete       = pid_delete_dentry,
1793 };
1794
1795 /* Lookups */
1796
1797 /*
1798  * Fill a directory entry.
1799  *
1800  * If possible create the dcache entry and derive our inode number and
1801  * file type from dcache entry.
1802  *
1803  * Since all of the proc inode numbers are dynamically generated, the inode
1804  * numbers do not exist until the inode is cache.  This means creating the
1805  * the dcache entry in readdir is necessary to keep the inode numbers
1806  * reported by readdir in sync with the inode numbers reported
1807  * by stat.
1808  */
1809 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1810         const char *name, int len,
1811         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1812 {
1813         struct dentry *child, *dir = file->f_path.dentry;
1814         struct qstr qname = QSTR_INIT(name, len);
1815         struct inode *inode;
1816         unsigned type;
1817         ino_t ino;
1818
1819         child = d_hash_and_lookup(dir, &qname);
1820         if (!child) {
1821                 child = d_alloc(dir, &qname);
1822                 if (!child)
1823                         goto end_instantiate;
1824                 if (instantiate(d_inode(dir), child, task, ptr) < 0) {
1825                         dput(child);
1826                         goto end_instantiate;
1827                 }
1828         }
1829         inode = d_inode(child);
1830         ino = inode->i_ino;
1831         type = inode->i_mode >> 12;
1832         dput(child);
1833         return dir_emit(ctx, name, len, ino, type);
1834
1835 end_instantiate:
1836         return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1837 }
1838
1839 #ifdef CONFIG_CHECKPOINT_RESTORE
1840
1841 /*
1842  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1843  * which represent vma start and end addresses.
1844  */
1845 static int dname_to_vma_addr(struct dentry *dentry,
1846                              unsigned long *start, unsigned long *end)
1847 {
1848         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1849                 return -EINVAL;
1850
1851         return 0;
1852 }
1853
1854 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1855 {
1856         unsigned long vm_start, vm_end;
1857         bool exact_vma_exists = false;
1858         struct mm_struct *mm = NULL;
1859         struct task_struct *task;
1860         const struct cred *cred;
1861         struct inode *inode;
1862         int status = 0;
1863
1864         if (flags & LOOKUP_RCU)
1865                 return -ECHILD;
1866
1867         if (!capable(CAP_SYS_ADMIN)) {
1868                 status = -EPERM;
1869                 goto out_notask;
1870         }
1871
1872         inode = d_inode(dentry);
1873         task = get_proc_task(inode);
1874         if (!task)
1875                 goto out_notask;
1876
1877         mm = mm_access(task, PTRACE_MODE_READ);
1878         if (IS_ERR_OR_NULL(mm))
1879                 goto out;
1880
1881         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1882                 down_read(&mm->mmap_sem);
1883                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1884                 up_read(&mm->mmap_sem);
1885         }
1886
1887         mmput(mm);
1888
1889         if (exact_vma_exists) {
1890                 if (task_dumpable(task)) {
1891                         rcu_read_lock();
1892                         cred = __task_cred(task);
1893                         inode->i_uid = cred->euid;
1894                         inode->i_gid = cred->egid;
1895                         rcu_read_unlock();
1896                 } else {
1897                         inode->i_uid = GLOBAL_ROOT_UID;
1898                         inode->i_gid = GLOBAL_ROOT_GID;
1899                 }
1900                 security_task_to_inode(task, inode);
1901                 status = 1;
1902         }
1903
1904 out:
1905         put_task_struct(task);
1906
1907 out_notask:
1908         return status;
1909 }
1910
1911 static const struct dentry_operations tid_map_files_dentry_operations = {
1912         .d_revalidate   = map_files_d_revalidate,
1913         .d_delete       = pid_delete_dentry,
1914 };
1915
1916 static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
1917 {
1918         unsigned long vm_start, vm_end;
1919         struct vm_area_struct *vma;
1920         struct task_struct *task;
1921         struct mm_struct *mm;
1922         int rc;
1923
1924         rc = -ENOENT;
1925         task = get_proc_task(d_inode(dentry));
1926         if (!task)
1927                 goto out;
1928
1929         mm = get_task_mm(task);
1930         put_task_struct(task);
1931         if (!mm)
1932                 goto out;
1933
1934         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1935         if (rc)
1936                 goto out_mmput;
1937
1938         rc = -ENOENT;
1939         down_read(&mm->mmap_sem);
1940         vma = find_exact_vma(mm, vm_start, vm_end);
1941         if (vma && vma->vm_file) {
1942                 *path = vma->vm_file->f_path;
1943                 path_get(path);
1944                 rc = 0;
1945         }
1946         up_read(&mm->mmap_sem);
1947
1948 out_mmput:
1949         mmput(mm);
1950 out:
1951         return rc;
1952 }
1953
1954 struct map_files_info {
1955         fmode_t         mode;
1956         unsigned long   len;
1957         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1958 };
1959
1960 static int
1961 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
1962                            struct task_struct *task, const void *ptr)
1963 {
1964         fmode_t mode = (fmode_t)(unsigned long)ptr;
1965         struct proc_inode *ei;
1966         struct inode *inode;
1967
1968         inode = proc_pid_make_inode(dir->i_sb, task);
1969         if (!inode)
1970                 return -ENOENT;
1971
1972         ei = PROC_I(inode);
1973         ei->op.proc_get_link = proc_map_files_get_link;
1974
1975         inode->i_op = &proc_pid_link_inode_operations;
1976         inode->i_size = 64;
1977         inode->i_mode = S_IFLNK;
1978
1979         if (mode & FMODE_READ)
1980                 inode->i_mode |= S_IRUSR;
1981         if (mode & FMODE_WRITE)
1982                 inode->i_mode |= S_IWUSR;
1983
1984         d_set_d_op(dentry, &tid_map_files_dentry_operations);
1985         d_add(dentry, inode);
1986
1987         return 0;
1988 }
1989
1990 static struct dentry *proc_map_files_lookup(struct inode *dir,
1991                 struct dentry *dentry, unsigned int flags)
1992 {
1993         unsigned long vm_start, vm_end;
1994         struct vm_area_struct *vma;
1995         struct task_struct *task;
1996         int result;
1997         struct mm_struct *mm;
1998
1999         result = -EPERM;
2000         if (!capable(CAP_SYS_ADMIN))
2001                 goto out;
2002
2003         result = -ENOENT;
2004         task = get_proc_task(dir);
2005         if (!task)
2006                 goto out;
2007
2008         result = -EACCES;
2009         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2010                 goto out_put_task;
2011
2012         result = -ENOENT;
2013         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2014                 goto out_put_task;
2015
2016         mm = get_task_mm(task);
2017         if (!mm)
2018                 goto out_put_task;
2019
2020         down_read(&mm->mmap_sem);
2021         vma = find_exact_vma(mm, vm_start, vm_end);
2022         if (!vma)
2023                 goto out_no_vma;
2024
2025         if (vma->vm_file)
2026                 result = proc_map_files_instantiate(dir, dentry, task,
2027                                 (void *)(unsigned long)vma->vm_file->f_mode);
2028
2029 out_no_vma:
2030         up_read(&mm->mmap_sem);
2031         mmput(mm);
2032 out_put_task:
2033         put_task_struct(task);
2034 out:
2035         return ERR_PTR(result);
2036 }
2037
2038 static const struct inode_operations proc_map_files_inode_operations = {
2039         .lookup         = proc_map_files_lookup,
2040         .permission     = proc_fd_permission,
2041         .setattr        = proc_setattr,
2042 };
2043
2044 static int
2045 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2046 {
2047         struct vm_area_struct *vma;
2048         struct task_struct *task;
2049         struct mm_struct *mm;
2050         unsigned long nr_files, pos, i;
2051         struct flex_array *fa = NULL;
2052         struct map_files_info info;
2053         struct map_files_info *p;
2054         int ret;
2055
2056         ret = -EPERM;
2057         if (!capable(CAP_SYS_ADMIN))
2058                 goto out;
2059
2060         ret = -ENOENT;
2061         task = get_proc_task(file_inode(file));
2062         if (!task)
2063                 goto out;
2064
2065         ret = -EACCES;
2066         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2067                 goto out_put_task;
2068
2069         ret = 0;
2070         if (!dir_emit_dots(file, ctx))
2071                 goto out_put_task;
2072
2073         mm = get_task_mm(task);
2074         if (!mm)
2075                 goto out_put_task;
2076         down_read(&mm->mmap_sem);
2077
2078         nr_files = 0;
2079
2080         /*
2081          * We need two passes here:
2082          *
2083          *  1) Collect vmas of mapped files with mmap_sem taken
2084          *  2) Release mmap_sem and instantiate entries
2085          *
2086          * otherwise we get lockdep complained, since filldir()
2087          * routine might require mmap_sem taken in might_fault().
2088          */
2089
2090         for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2091                 if (vma->vm_file && ++pos > ctx->pos)
2092                         nr_files++;
2093         }
2094
2095         if (nr_files) {
2096                 fa = flex_array_alloc(sizeof(info), nr_files,
2097                                         GFP_KERNEL);
2098                 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2099                                                 GFP_KERNEL)) {
2100                         ret = -ENOMEM;
2101                         if (fa)
2102                                 flex_array_free(fa);
2103                         up_read(&mm->mmap_sem);
2104                         mmput(mm);
2105                         goto out_put_task;
2106                 }
2107                 for (i = 0, vma = mm->mmap, pos = 2; vma;
2108                                 vma = vma->vm_next) {
2109                         if (!vma->vm_file)
2110                                 continue;
2111                         if (++pos <= ctx->pos)
2112                                 continue;
2113
2114                         info.mode = vma->vm_file->f_mode;
2115                         info.len = snprintf(info.name,
2116                                         sizeof(info.name), "%lx-%lx",
2117                                         vma->vm_start, vma->vm_end);
2118                         if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2119                                 BUG();
2120                 }
2121         }
2122         up_read(&mm->mmap_sem);
2123
2124         for (i = 0; i < nr_files; i++) {
2125                 p = flex_array_get(fa, i);
2126                 if (!proc_fill_cache(file, ctx,
2127                                       p->name, p->len,
2128                                       proc_map_files_instantiate,
2129                                       task,
2130                                       (void *)(unsigned long)p->mode))
2131                         break;
2132                 ctx->pos++;
2133         }
2134         if (fa)
2135                 flex_array_free(fa);
2136         mmput(mm);
2137
2138 out_put_task:
2139         put_task_struct(task);
2140 out:
2141         return ret;
2142 }
2143
2144 static const struct file_operations proc_map_files_operations = {
2145         .read           = generic_read_dir,
2146         .iterate        = proc_map_files_readdir,
2147         .llseek         = default_llseek,
2148 };
2149
2150 struct timers_private {
2151         struct pid *pid;
2152         struct task_struct *task;
2153         struct sighand_struct *sighand;
2154         struct pid_namespace *ns;
2155         unsigned long flags;
2156 };
2157
2158 static void *timers_start(struct seq_file *m, loff_t *pos)
2159 {
2160         struct timers_private *tp = m->private;
2161
2162         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2163         if (!tp->task)
2164                 return ERR_PTR(-ESRCH);
2165
2166         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2167         if (!tp->sighand)
2168                 return ERR_PTR(-ESRCH);
2169
2170         return seq_list_start(&tp->task->signal->posix_timers, *pos);
2171 }
2172
2173 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2174 {
2175         struct timers_private *tp = m->private;
2176         return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2177 }
2178
2179 static void timers_stop(struct seq_file *m, void *v)
2180 {
2181         struct timers_private *tp = m->private;
2182
2183         if (tp->sighand) {
2184                 unlock_task_sighand(tp->task, &tp->flags);
2185                 tp->sighand = NULL;
2186         }
2187
2188         if (tp->task) {
2189                 put_task_struct(tp->task);
2190                 tp->task = NULL;
2191         }
2192 }
2193
2194 static int show_timer(struct seq_file *m, void *v)
2195 {
2196         struct k_itimer *timer;
2197         struct timers_private *tp = m->private;
2198         int notify;
2199         static const char * const nstr[] = {
2200                 [SIGEV_SIGNAL] = "signal",
2201                 [SIGEV_NONE] = "none",
2202                 [SIGEV_THREAD] = "thread",
2203         };
2204
2205         timer = list_entry((struct list_head *)v, struct k_itimer, list);
2206         notify = timer->it_sigev_notify;
2207
2208         seq_printf(m, "ID: %d\n", timer->it_id);
2209         seq_printf(m, "signal: %d/%p\n",
2210                    timer->sigq->info.si_signo,
2211                    timer->sigq->info.si_value.sival_ptr);
2212         seq_printf(m, "notify: %s/%s.%d\n",
2213                    nstr[notify & ~SIGEV_THREAD_ID],
2214                    (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2215                    pid_nr_ns(timer->it_pid, tp->ns));
2216         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2217
2218         return 0;
2219 }
2220
2221 static const struct seq_operations proc_timers_seq_ops = {
2222         .start  = timers_start,
2223         .next   = timers_next,
2224         .stop   = timers_stop,
2225         .show   = show_timer,
2226 };
2227
2228 static int proc_timers_open(struct inode *inode, struct file *file)
2229 {
2230         struct timers_private *tp;
2231
2232         tp = __seq_open_private(file, &proc_timers_seq_ops,
2233                         sizeof(struct timers_private));
2234         if (!tp)
2235                 return -ENOMEM;
2236
2237         tp->pid = proc_pid(inode);
2238         tp->ns = inode->i_sb->s_fs_info;
2239         return 0;
2240 }
2241
2242 static const struct file_operations proc_timers_operations = {
2243         .open           = proc_timers_open,
2244         .read           = seq_read,
2245         .llseek         = seq_lseek,
2246         .release        = seq_release_private,
2247 };
2248 #endif /* CONFIG_CHECKPOINT_RESTORE */
2249
2250 static int proc_pident_instantiate(struct inode *dir,
2251         struct dentry *dentry, struct task_struct *task, const void *ptr)
2252 {
2253         const struct pid_entry *p = ptr;
2254         struct inode *inode;
2255         struct proc_inode *ei;
2256
2257         inode = proc_pid_make_inode(dir->i_sb, task);
2258         if (!inode)
2259                 goto out;
2260
2261         ei = PROC_I(inode);
2262         inode->i_mode = p->mode;
2263         if (S_ISDIR(inode->i_mode))
2264                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2265         if (p->iop)
2266                 inode->i_op = p->iop;
2267         if (p->fop)
2268                 inode->i_fop = p->fop;
2269         ei->op = p->op;
2270         d_set_d_op(dentry, &pid_dentry_operations);
2271         d_add(dentry, inode);
2272         /* Close the race of the process dying before we return the dentry */
2273         if (pid_revalidate(dentry, 0))
2274                 return 0;
2275 out:
2276         return -ENOENT;
2277 }
2278
2279 static struct dentry *proc_pident_lookup(struct inode *dir,
2280                                          struct dentry *dentry,
2281                                          const struct pid_entry *ents,
2282                                          unsigned int nents)
2283 {
2284         int error;
2285         struct task_struct *task = get_proc_task(dir);
2286         const struct pid_entry *p, *last;
2287
2288         error = -ENOENT;
2289
2290         if (!task)
2291                 goto out_no_task;
2292
2293         /*
2294          * Yes, it does not scale. And it should not. Don't add
2295          * new entries into /proc/<tgid>/ without very good reasons.
2296          */
2297         last = &ents[nents - 1];
2298         for (p = ents; p <= last; p++) {
2299                 if (p->len != dentry->d_name.len)
2300                         continue;
2301                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2302                         break;
2303         }
2304         if (p > last)
2305                 goto out;
2306
2307         error = proc_pident_instantiate(dir, dentry, task, p);
2308 out:
2309         put_task_struct(task);
2310 out_no_task:
2311         return ERR_PTR(error);
2312 }
2313
2314 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2315                 const struct pid_entry *ents, unsigned int nents)
2316 {
2317         struct task_struct *task = get_proc_task(file_inode(file));
2318         const struct pid_entry *p;
2319
2320         if (!task)
2321                 return -ENOENT;
2322
2323         if (!dir_emit_dots(file, ctx))
2324                 goto out;
2325
2326         if (ctx->pos >= nents + 2)
2327                 goto out;
2328
2329         for (p = ents + (ctx->pos - 2); p <= ents + nents - 1; p++) {
2330                 if (!proc_fill_cache(file, ctx, p->name, p->len,
2331                                 proc_pident_instantiate, task, p))
2332                         break;
2333                 ctx->pos++;
2334         }
2335 out:
2336         put_task_struct(task);
2337         return 0;
2338 }
2339
2340 #ifdef CONFIG_SECURITY
2341 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2342                                   size_t count, loff_t *ppos)
2343 {
2344         struct inode * inode = file_inode(file);
2345         char *p = NULL;
2346         ssize_t length;
2347         struct task_struct *task = get_proc_task(inode);
2348
2349         if (!task)
2350                 return -ESRCH;
2351
2352         length = security_getprocattr(task,
2353                                       (char*)file->f_path.dentry->d_name.name,
2354                                       &p);
2355         put_task_struct(task);
2356         if (length > 0)
2357                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2358         kfree(p);
2359         return length;
2360 }
2361
2362 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2363                                    size_t count, loff_t *ppos)
2364 {
2365         struct inode * inode = file_inode(file);
2366         char *page;
2367         ssize_t length;
2368         struct task_struct *task = get_proc_task(inode);
2369
2370         length = -ESRCH;
2371         if (!task)
2372                 goto out_no_task;
2373         if (count > PAGE_SIZE)
2374                 count = PAGE_SIZE;
2375
2376         /* No partial writes. */
2377         length = -EINVAL;
2378         if (*ppos != 0)
2379                 goto out;
2380
2381         length = -ENOMEM;
2382         page = (char*)__get_free_page(GFP_TEMPORARY);
2383         if (!page)
2384                 goto out;
2385
2386         length = -EFAULT;
2387         if (copy_from_user(page, buf, count))
2388                 goto out_free;
2389
2390         /* Guard against adverse ptrace interaction */
2391         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2392         if (length < 0)
2393                 goto out_free;
2394
2395         length = security_setprocattr(task,
2396                                       (char*)file->f_path.dentry->d_name.name,
2397                                       (void*)page, count);
2398         mutex_unlock(&task->signal->cred_guard_mutex);
2399 out_free:
2400         free_page((unsigned long) page);
2401 out:
2402         put_task_struct(task);
2403 out_no_task:
2404         return length;
2405 }
2406
2407 static const struct file_operations proc_pid_attr_operations = {
2408         .read           = proc_pid_attr_read,
2409         .write          = proc_pid_attr_write,
2410         .llseek         = generic_file_llseek,
2411 };
2412
2413 static const struct pid_entry attr_dir_stuff[] = {
2414         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2415         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2416         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2417         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2418         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2419         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2420 };
2421
2422 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2423 {
2424         return proc_pident_readdir(file, ctx,
2425                                    attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2426 }
2427
2428 static const struct file_operations proc_attr_dir_operations = {
2429         .read           = generic_read_dir,
2430         .iterate        = proc_attr_dir_readdir,
2431         .llseek         = default_llseek,
2432 };
2433
2434 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2435                                 struct dentry *dentry, unsigned int flags)
2436 {
2437         return proc_pident_lookup(dir, dentry,
2438                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2439 }
2440
2441 static const struct inode_operations proc_attr_dir_inode_operations = {
2442         .lookup         = proc_attr_dir_lookup,
2443         .getattr        = pid_getattr,
2444         .setattr        = proc_setattr,
2445 };
2446
2447 #endif
2448
2449 #ifdef CONFIG_ELF_CORE
2450 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2451                                          size_t count, loff_t *ppos)
2452 {
2453         struct task_struct *task = get_proc_task(file_inode(file));
2454         struct mm_struct *mm;
2455         char buffer[PROC_NUMBUF];
2456         size_t len;
2457         int ret;
2458
2459         if (!task)
2460                 return -ESRCH;
2461
2462         ret = 0;
2463         mm = get_task_mm(task);
2464         if (mm) {
2465                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2466                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2467                                 MMF_DUMP_FILTER_SHIFT));
2468                 mmput(mm);
2469                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2470         }
2471
2472         put_task_struct(task);
2473
2474         return ret;
2475 }
2476
2477 static ssize_t proc_coredump_filter_write(struct file *file,
2478                                           const char __user *buf,
2479                                           size_t count,
2480                                           loff_t *ppos)
2481 {
2482         struct task_struct *task;
2483         struct mm_struct *mm;
2484         char buffer[PROC_NUMBUF], *end;
2485         unsigned int val;
2486         int ret;
2487         int i;
2488         unsigned long mask;
2489
2490         ret = -EFAULT;
2491         memset(buffer, 0, sizeof(buffer));
2492         if (count > sizeof(buffer) - 1)
2493                 count = sizeof(buffer) - 1;
2494         if (copy_from_user(buffer, buf, count))
2495                 goto out_no_task;
2496
2497         ret = -EINVAL;
2498         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2499         if (*end == '\n')
2500                 end++;
2501         if (end - buffer == 0)
2502                 goto out_no_task;
2503
2504         ret = -ESRCH;
2505         task = get_proc_task(file_inode(file));
2506         if (!task)
2507                 goto out_no_task;
2508
2509         ret = end - buffer;
2510         mm = get_task_mm(task);
2511         if (!mm)
2512                 goto out_no_mm;
2513
2514         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2515                 if (val & mask)
2516                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2517                 else
2518                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2519         }
2520
2521         mmput(mm);
2522  out_no_mm:
2523         put_task_struct(task);
2524  out_no_task:
2525         return ret;
2526 }
2527
2528 static const struct file_operations proc_coredump_filter_operations = {
2529         .read           = proc_coredump_filter_read,
2530         .write          = proc_coredump_filter_write,
2531         .llseek         = generic_file_llseek,
2532 };
2533 #endif
2534
2535 #ifdef CONFIG_TASK_IO_ACCOUNTING
2536 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2537 {
2538         struct task_io_accounting acct = task->ioac;
2539         unsigned long flags;
2540         int result;
2541
2542         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2543         if (result)
2544                 return result;
2545
2546         if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2547                 result = -EACCES;
2548                 goto out_unlock;
2549         }
2550
2551         if (whole && lock_task_sighand(task, &flags)) {
2552                 struct task_struct *t = task;
2553
2554                 task_io_accounting_add(&acct, &task->signal->ioac);
2555                 while_each_thread(task, t)
2556                         task_io_accounting_add(&acct, &t->ioac);
2557
2558                 unlock_task_sighand(task, &flags);
2559         }
2560         seq_printf(m,
2561                    "rchar: %llu\n"
2562                    "wchar: %llu\n"
2563                    "syscr: %llu\n"
2564                    "syscw: %llu\n"
2565                    "read_bytes: %llu\n"
2566                    "write_bytes: %llu\n"
2567                    "cancelled_write_bytes: %llu\n",
2568                    (unsigned long long)acct.rchar,
2569                    (unsigned long long)acct.wchar,
2570                    (unsigned long long)acct.syscr,
2571                    (unsigned long long)acct.syscw,
2572                    (unsigned long long)acct.read_bytes,
2573                    (unsigned long long)acct.write_bytes,
2574                    (unsigned long long)acct.cancelled_write_bytes);
2575         result = 0;
2576
2577 out_unlock:
2578         mutex_unlock(&task->signal->cred_guard_mutex);
2579         return result;
2580 }
2581
2582 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2583                                   struct pid *pid, struct task_struct *task)
2584 {
2585         return do_io_accounting(task, m, 0);
2586 }
2587
2588 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2589                                    struct pid *pid, struct task_struct *task)
2590 {
2591         return do_io_accounting(task, m, 1);
2592 }
2593 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2594
2595 #ifdef CONFIG_USER_NS
2596 static int proc_id_map_open(struct inode *inode, struct file *file,
2597         const struct seq_operations *seq_ops)
2598 {
2599         struct user_namespace *ns = NULL;
2600         struct task_struct *task;
2601         struct seq_file *seq;
2602         int ret = -EINVAL;
2603
2604         task = get_proc_task(inode);
2605         if (task) {
2606                 rcu_read_lock();
2607                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2608                 rcu_read_unlock();
2609                 put_task_struct(task);
2610         }
2611         if (!ns)
2612                 goto err;
2613
2614         ret = seq_open(file, seq_ops);
2615         if (ret)
2616                 goto err_put_ns;
2617
2618         seq = file->private_data;
2619         seq->private = ns;
2620
2621         return 0;
2622 err_put_ns:
2623         put_user_ns(ns);
2624 err:
2625         return ret;
2626 }
2627
2628 static int proc_id_map_release(struct inode *inode, struct file *file)
2629 {
2630         struct seq_file *seq = file->private_data;
2631         struct user_namespace *ns = seq->private;
2632         put_user_ns(ns);
2633         return seq_release(inode, file);
2634 }
2635
2636 static int proc_uid_map_open(struct inode *inode, struct file *file)
2637 {
2638         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2639 }
2640
2641 static int proc_gid_map_open(struct inode *inode, struct file *file)
2642 {
2643         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2644 }
2645
2646 static int proc_projid_map_open(struct inode *inode, struct file *file)
2647 {
2648         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2649 }
2650
2651 static const struct file_operations proc_uid_map_operations = {
2652         .open           = proc_uid_map_open,
2653         .write          = proc_uid_map_write,
2654         .read           = seq_read,
2655         .llseek         = seq_lseek,
2656         .release        = proc_id_map_release,
2657 };
2658
2659 static const struct file_operations proc_gid_map_operations = {
2660         .open           = proc_gid_map_open,
2661         .write          = proc_gid_map_write,
2662         .read           = seq_read,
2663         .llseek         = seq_lseek,
2664         .release        = proc_id_map_release,
2665 };
2666
2667 static const struct file_operations proc_projid_map_operations = {
2668         .open           = proc_projid_map_open,
2669         .write          = proc_projid_map_write,
2670         .read           = seq_read,
2671         .llseek         = seq_lseek,
2672         .release        = proc_id_map_release,
2673 };
2674
2675 static int proc_setgroups_open(struct inode *inode, struct file *file)
2676 {
2677         struct user_namespace *ns = NULL;
2678         struct task_struct *task;
2679         int ret;
2680
2681         ret = -ESRCH;
2682         task = get_proc_task(inode);
2683         if (task) {
2684                 rcu_read_lock();
2685                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2686                 rcu_read_unlock();
2687                 put_task_struct(task);
2688         }
2689         if (!ns)
2690                 goto err;
2691
2692         if (file->f_mode & FMODE_WRITE) {
2693                 ret = -EACCES;
2694                 if (!ns_capable(ns, CAP_SYS_ADMIN))
2695                         goto err_put_ns;
2696         }
2697
2698         ret = single_open(file, &proc_setgroups_show, ns);
2699         if (ret)
2700                 goto err_put_ns;
2701
2702         return 0;
2703 err_put_ns:
2704         put_user_ns(ns);
2705 err:
2706         return ret;
2707 }
2708
2709 static int proc_setgroups_release(struct inode *inode, struct file *file)
2710 {
2711         struct seq_file *seq = file->private_data;
2712         struct user_namespace *ns = seq->private;
2713         int ret = single_release(inode, file);
2714         put_user_ns(ns);
2715         return ret;
2716 }
2717
2718 static const struct file_operations proc_setgroups_operations = {
2719         .open           = proc_setgroups_open,
2720         .write          = proc_setgroups_write,
2721         .read           = seq_read,
2722         .llseek         = seq_lseek,
2723         .release        = proc_setgroups_release,
2724 };
2725 #endif /* CONFIG_USER_NS */
2726
2727 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2728                                 struct pid *pid, struct task_struct *task)
2729 {
2730         int err = lock_trace(task);
2731         if (!err) {
2732                 seq_printf(m, "%08x\n", task->personality);
2733                 unlock_trace(task);
2734         }
2735         return err;
2736 }
2737
2738 /*
2739  * Thread groups
2740  */
2741 static const struct file_operations proc_task_operations;
2742 static const struct inode_operations proc_task_inode_operations;
2743
2744 static const struct pid_entry tgid_base_stuff[] = {
2745         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2746         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2747 #ifdef CONFIG_CHECKPOINT_RESTORE
2748         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2749 #endif
2750         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2751         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2752 #ifdef CONFIG_NET
2753         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2754 #endif
2755         REG("environ",    S_IRUSR, proc_environ_operations),
2756         ONE("auxv",       S_IRUSR, proc_pid_auxv),
2757         ONE("status",     S_IRUGO, proc_pid_status),
2758         ONE("personality", S_IRUSR, proc_pid_personality),
2759         ONE("limits",     S_IRUGO, proc_pid_limits),
2760 #ifdef CONFIG_SCHED_DEBUG
2761         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2762 #endif
2763 #ifdef CONFIG_SCHED_AUTOGROUP
2764         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2765 #endif
2766         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2767 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2768         ONE("syscall",    S_IRUSR, proc_pid_syscall),
2769 #endif
2770         REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
2771         ONE("stat",       S_IRUGO, proc_tgid_stat),
2772         ONE("statm",      S_IRUGO, proc_pid_statm),
2773         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2774 #ifdef CONFIG_NUMA
2775         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2776 #endif
2777         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2778         LNK("cwd",        proc_cwd_link),
2779         LNK("root",       proc_root_link),
2780         LNK("exe",        proc_exe_link),
2781         REG("mounts",     S_IRUGO, proc_mounts_operations),
2782         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2783         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2784 #ifdef CONFIG_PROC_PAGE_MONITOR
2785         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2786         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2787         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2788 #endif
2789 #ifdef CONFIG_SECURITY
2790         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2791 #endif
2792 #ifdef CONFIG_KALLSYMS
2793         ONE("wchan",      S_IRUGO, proc_pid_wchan),
2794 #endif
2795 #ifdef CONFIG_STACKTRACE
2796         ONE("stack",      S_IRUSR, proc_pid_stack),
2797 #endif
2798 #ifdef CONFIG_SCHED_INFO
2799         ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
2800 #endif
2801 #ifdef CONFIG_LATENCYTOP
2802         REG("latency",  S_IRUGO, proc_lstats_operations),
2803 #endif
2804 #ifdef CONFIG_PROC_PID_CPUSET
2805         ONE("cpuset",     S_IRUGO, proc_cpuset_show),
2806 #endif
2807 #ifdef CONFIG_CGROUPS
2808         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
2809 #endif
2810         ONE("oom_score",  S_IRUGO, proc_oom_score),
2811         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2812         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2813 #ifdef CONFIG_AUDITSYSCALL
2814         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2815         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2816 #endif
2817 #ifdef CONFIG_FAULT_INJECTION
2818         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2819 #endif
2820 #ifdef CONFIG_ELF_CORE
2821         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2822 #endif
2823 #ifdef CONFIG_TASK_IO_ACCOUNTING
2824         ONE("io",       S_IRUSR, proc_tgid_io_accounting),
2825 #endif
2826 #ifdef CONFIG_HARDWALL
2827         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
2828 #endif
2829 #ifdef CONFIG_USER_NS
2830         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2831         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2832         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2833         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
2834 #endif
2835 #ifdef CONFIG_CHECKPOINT_RESTORE
2836         REG("timers",     S_IRUGO, proc_timers_operations),
2837 #endif
2838 };
2839
2840 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2841 {
2842         return proc_pident_readdir(file, ctx,
2843                                    tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2844 }
2845
2846 static const struct file_operations proc_tgid_base_operations = {
2847         .read           = generic_read_dir,
2848         .iterate        = proc_tgid_base_readdir,
2849         .llseek         = default_llseek,
2850 };
2851
2852 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2853 {
2854         return proc_pident_lookup(dir, dentry,
2855                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2856 }
2857
2858 static const struct inode_operations proc_tgid_base_inode_operations = {
2859         .lookup         = proc_tgid_base_lookup,
2860         .getattr        = pid_getattr,
2861         .setattr        = proc_setattr,
2862         .permission     = proc_pid_permission,
2863 };
2864
2865 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2866 {
2867         struct dentry *dentry, *leader, *dir;
2868         char buf[PROC_NUMBUF];
2869         struct qstr name;
2870
2871         name.name = buf;
2872         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2873         /* no ->d_hash() rejects on procfs */
2874         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2875         if (dentry) {
2876                 d_invalidate(dentry);
2877                 dput(dentry);
2878         }
2879
2880         if (pid == tgid)
2881                 return;
2882
2883         name.name = buf;
2884         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2885         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2886         if (!leader)
2887                 goto out;
2888
2889         name.name = "task";
2890         name.len = strlen(name.name);
2891         dir = d_hash_and_lookup(leader, &name);
2892         if (!dir)
2893                 goto out_put_leader;
2894
2895         name.name = buf;
2896         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2897         dentry = d_hash_and_lookup(dir, &name);
2898         if (dentry) {
2899                 d_invalidate(dentry);
2900                 dput(dentry);
2901         }
2902
2903         dput(dir);
2904 out_put_leader:
2905         dput(leader);
2906 out:
2907         return;
2908 }
2909
2910 /**
2911  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2912  * @task: task that should be flushed.
2913  *
2914  * When flushing dentries from proc, one needs to flush them from global
2915  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2916  * in. This call is supposed to do all of this job.
2917  *
2918  * Looks in the dcache for
2919  * /proc/@pid
2920  * /proc/@tgid/task/@pid
2921  * if either directory is present flushes it and all of it'ts children
2922  * from the dcache.
2923  *
2924  * It is safe and reasonable to cache /proc entries for a task until
2925  * that task exits.  After that they just clog up the dcache with
2926  * useless entries, possibly causing useful dcache entries to be
2927  * flushed instead.  This routine is proved to flush those useless
2928  * dcache entries at process exit time.
2929  *
2930  * NOTE: This routine is just an optimization so it does not guarantee
2931  *       that no dcache entries will exist at process exit time it
2932  *       just makes it very unlikely that any will persist.
2933  */
2934
2935 void proc_flush_task(struct task_struct *task)
2936 {
2937         int i;
2938         struct pid *pid, *tgid;
2939         struct upid *upid;
2940
2941         pid = task_pid(task);
2942         tgid = task_tgid(task);
2943
2944         for (i = 0; i <= pid->level; i++) {
2945                 upid = &pid->numbers[i];
2946                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2947                                         tgid->numbers[i].nr);
2948         }
2949 }
2950
2951 static int proc_pid_instantiate(struct inode *dir,
2952                                    struct dentry * dentry,
2953                                    struct task_struct *task, const void *ptr)
2954 {
2955         struct inode *inode;
2956
2957         inode = proc_pid_make_inode(dir->i_sb, task);
2958         if (!inode)
2959                 goto out;
2960
2961         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2962         inode->i_op = &proc_tgid_base_inode_operations;
2963         inode->i_fop = &proc_tgid_base_operations;
2964         inode->i_flags|=S_IMMUTABLE;
2965
2966         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
2967                                                   ARRAY_SIZE(tgid_base_stuff)));
2968
2969         d_set_d_op(dentry, &pid_dentry_operations);
2970
2971         d_add(dentry, inode);
2972         /* Close the race of the process dying before we return the dentry */
2973         if (pid_revalidate(dentry, 0))
2974                 return 0;
2975 out:
2976         return -ENOENT;
2977 }
2978
2979 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
2980 {
2981         int result = -ENOENT;
2982         struct task_struct *task;
2983         unsigned tgid;
2984         struct pid_namespace *ns;
2985
2986         tgid = name_to_int(&dentry->d_name);
2987         if (tgid == ~0U)
2988                 goto out;
2989
2990         ns = dentry->d_sb->s_fs_info;
2991         rcu_read_lock();
2992         task = find_task_by_pid_ns(tgid, ns);
2993         if (task)
2994                 get_task_struct(task);
2995         rcu_read_unlock();
2996         if (!task)
2997                 goto out;
2998
2999         result = proc_pid_instantiate(dir, dentry, task, NULL);
3000         put_task_struct(task);
3001 out:
3002         return ERR_PTR(result);
3003 }
3004
3005 /*
3006  * Find the first task with tgid >= tgid
3007  *
3008  */
3009 struct tgid_iter {
3010         unsigned int tgid;
3011         struct task_struct *task;
3012 };
3013 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3014 {
3015         struct pid *pid;
3016
3017         if (iter.task)
3018                 put_task_struct(iter.task);
3019         rcu_read_lock();
3020 retry:
3021         iter.task = NULL;
3022         pid = find_ge_pid(iter.tgid, ns);
3023         if (pid) {
3024                 iter.tgid = pid_nr_ns(pid, ns);
3025                 iter.task = pid_task(pid, PIDTYPE_PID);
3026                 /* What we to know is if the pid we have find is the
3027                  * pid of a thread_group_leader.  Testing for task
3028                  * being a thread_group_leader is the obvious thing
3029                  * todo but there is a window when it fails, due to
3030                  * the pid transfer logic in de_thread.
3031                  *
3032                  * So we perform the straight forward test of seeing
3033                  * if the pid we have found is the pid of a thread
3034                  * group leader, and don't worry if the task we have
3035                  * found doesn't happen to be a thread group leader.
3036                  * As we don't care in the case of readdir.
3037                  */
3038                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3039                         iter.tgid += 1;
3040                         goto retry;
3041                 }
3042                 get_task_struct(iter.task);
3043         }
3044         rcu_read_unlock();
3045         return iter;
3046 }
3047
3048 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3049
3050 /* for the /proc/ directory itself, after non-process stuff has been done */
3051 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3052 {
3053         struct tgid_iter iter;
3054         struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
3055         loff_t pos = ctx->pos;
3056
3057         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3058                 return 0;
3059
3060         if (pos == TGID_OFFSET - 2) {
3061                 struct inode *inode = d_inode(ns->proc_self);
3062                 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3063                         return 0;
3064                 ctx->pos = pos = pos + 1;
3065         }
3066         if (pos == TGID_OFFSET - 1) {
3067                 struct inode *inode = d_inode(ns->proc_thread_self);
3068                 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3069                         return 0;
3070                 ctx->pos = pos = pos + 1;
3071         }
3072         iter.tgid = pos - TGID_OFFSET;
3073         iter.task = NULL;
3074         for (iter = next_tgid(ns, iter);
3075              iter.task;
3076              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3077                 char name[PROC_NUMBUF];
3078                 int len;
3079                 if (!has_pid_permissions(ns, iter.task, 2))
3080                         continue;
3081
3082                 len = snprintf(name, sizeof(name), "%d", iter.tgid);
3083                 ctx->pos = iter.tgid + TGID_OFFSET;
3084                 if (!proc_fill_cache(file, ctx, name, len,
3085                                      proc_pid_instantiate, iter.task, NULL)) {
3086                         put_task_struct(iter.task);
3087                         return 0;
3088                 }
3089         }
3090         ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3091         return 0;
3092 }
3093
3094 /*
3095  * Tasks
3096  */
3097 static const struct pid_entry tid_base_stuff[] = {
3098         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3099         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3100         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3101 #ifdef CONFIG_NET
3102         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3103 #endif
3104         REG("environ",   S_IRUSR, proc_environ_operations),
3105         ONE("auxv",      S_IRUSR, proc_pid_auxv),
3106         ONE("status",    S_IRUGO, proc_pid_status),
3107         ONE("personality", S_IRUSR, proc_pid_personality),
3108         ONE("limits",    S_IRUGO, proc_pid_limits),
3109 #ifdef CONFIG_SCHED_DEBUG
3110         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3111 #endif
3112         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3113 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3114         ONE("syscall",   S_IRUSR, proc_pid_syscall),
3115 #endif
3116         REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3117         ONE("stat",      S_IRUGO, proc_tid_stat),
3118         ONE("statm",     S_IRUGO, proc_pid_statm),
3119         REG("maps",      S_IRUGO, proc_tid_maps_operations),
3120 #ifdef CONFIG_PROC_CHILDREN
3121         REG("children",  S_IRUGO, proc_tid_children_operations),
3122 #endif
3123 #ifdef CONFIG_NUMA
3124         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3125 #endif
3126         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3127         LNK("cwd",       proc_cwd_link),
3128         LNK("root",      proc_root_link),
3129         LNK("exe",       proc_exe_link),
3130         REG("mounts",    S_IRUGO, proc_mounts_operations),
3131         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3132 #ifdef CONFIG_PROC_PAGE_MONITOR
3133         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3134         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3135         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3136 #endif
3137 #ifdef CONFIG_SECURITY
3138         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3139 #endif
3140 #ifdef CONFIG_KALLSYMS
3141         ONE("wchan",     S_IRUGO, proc_pid_wchan),
3142 #endif
3143 #ifdef CONFIG_STACKTRACE
3144         ONE("stack",      S_IRUSR, proc_pid_stack),
3145 #endif
3146 #ifdef CONFIG_SCHED_INFO
3147         ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3148 #endif
3149 #ifdef CONFIG_LATENCYTOP
3150         REG("latency",  S_IRUGO, proc_lstats_operations),
3151 #endif
3152 #ifdef CONFIG_PROC_PID_CPUSET
3153         ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3154 #endif
3155 #ifdef CONFIG_CGROUPS
3156         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3157 #endif
3158         ONE("oom_score", S_IRUGO, proc_oom_score),
3159         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3160         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3161 #ifdef CONFIG_AUDITSYSCALL
3162         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3163         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3164 #endif
3165 #ifdef CONFIG_FAULT_INJECTION
3166         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3167 #endif
3168 #ifdef CONFIG_TASK_IO_ACCOUNTING
3169         ONE("io",       S_IRUSR, proc_tid_io_accounting),
3170 #endif
3171 #ifdef CONFIG_HARDWALL
3172         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
3173 #endif
3174 #ifdef CONFIG_USER_NS
3175         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3176         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3177         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3178         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3179 #endif
3180 };
3181
3182 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3183 {
3184         return proc_pident_readdir(file, ctx,
3185                                    tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3186 }
3187
3188 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3189 {
3190         return proc_pident_lookup(dir, dentry,
3191                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3192 }
3193
3194 static const struct file_operations proc_tid_base_operations = {
3195         .read           = generic_read_dir,
3196         .iterate        = proc_tid_base_readdir,
3197         .llseek         = default_llseek,
3198 };
3199
3200 static const struct inode_operations proc_tid_base_inode_operations = {
3201         .lookup         = proc_tid_base_lookup,
3202         .getattr        = pid_getattr,
3203         .setattr        = proc_setattr,
3204 };
3205
3206 static int proc_task_instantiate(struct inode *dir,
3207         struct dentry *dentry, struct task_struct *task, const void *ptr)
3208 {
3209         struct inode *inode;
3210         inode = proc_pid_make_inode(dir->i_sb, task);
3211
3212         if (!inode)
3213                 goto out;
3214         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3215         inode->i_op = &proc_tid_base_inode_operations;
3216         inode->i_fop = &proc_tid_base_operations;
3217         inode->i_flags|=S_IMMUTABLE;
3218
3219         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3220                                                   ARRAY_SIZE(tid_base_stuff)));
3221
3222         d_set_d_op(dentry, &pid_dentry_operations);
3223
3224         d_add(dentry, inode);
3225         /* Close the race of the process dying before we return the dentry */
3226         if (pid_revalidate(dentry, 0))
3227                 return 0;
3228 out:
3229         return -ENOENT;
3230 }
3231
3232 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3233 {
3234         int result = -ENOENT;
3235         struct task_struct *task;
3236         struct task_struct *leader = get_proc_task(dir);
3237         unsigned tid;
3238         struct pid_namespace *ns;
3239
3240         if (!leader)
3241                 goto out_no_task;
3242
3243         tid = name_to_int(&dentry->d_name);
3244         if (tid == ~0U)
3245                 goto out;
3246
3247         ns = dentry->d_sb->s_fs_info;
3248         rcu_read_lock();
3249         task = find_task_by_pid_ns(tid, ns);
3250         if (task)
3251                 get_task_struct(task);
3252         rcu_read_unlock();
3253         if (!task)
3254                 goto out;
3255         if (!same_thread_group(leader, task))
3256                 goto out_drop_task;
3257
3258         result = proc_task_instantiate(dir, dentry, task, NULL);
3259 out_drop_task:
3260         put_task_struct(task);
3261 out:
3262         put_task_struct(leader);
3263 out_no_task:
3264         return ERR_PTR(result);
3265 }
3266
3267 /*
3268  * Find the first tid of a thread group to return to user space.
3269  *
3270  * Usually this is just the thread group leader, but if the users
3271  * buffer was too small or there was a seek into the middle of the
3272  * directory we have more work todo.
3273  *
3274  * In the case of a short read we start with find_task_by_pid.
3275  *
3276  * In the case of a seek we start with the leader and walk nr
3277  * threads past it.
3278  */
3279 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3280                                         struct pid_namespace *ns)
3281 {
3282         struct task_struct *pos, *task;
3283         unsigned long nr = f_pos;
3284
3285         if (nr != f_pos)        /* 32bit overflow? */
3286                 return NULL;
3287
3288         rcu_read_lock();
3289         task = pid_task(pid, PIDTYPE_PID);
3290         if (!task)
3291                 goto fail;
3292
3293         /* Attempt to start with the tid of a thread */
3294         if (tid && nr) {
3295                 pos = find_task_by_pid_ns(tid, ns);
3296                 if (pos && same_thread_group(pos, task))
3297                         goto found;
3298         }
3299
3300         /* If nr exceeds the number of threads there is nothing todo */
3301         if (nr >= get_nr_threads(task))
3302                 goto fail;
3303
3304         /* If we haven't found our starting place yet start
3305          * with the leader and walk nr threads forward.
3306          */
3307         pos = task = task->group_leader;
3308         do {
3309                 if (!nr--)
3310                         goto found;
3311         } while_each_thread(task, pos);
3312 fail:
3313         pos = NULL;
3314         goto out;
3315 found:
3316         get_task_struct(pos);
3317 out:
3318         rcu_read_unlock();
3319         return pos;
3320 }
3321
3322 /*
3323  * Find the next thread in the thread list.
3324  * Return NULL if there is an error or no next thread.
3325  *
3326  * The reference to the input task_struct is released.
3327  */
3328 static struct task_struct *next_tid(struct task_struct *start)
3329 {
3330         struct task_struct *pos = NULL;
3331         rcu_read_lock();
3332         if (pid_alive(start)) {
3333                 pos = next_thread(start);
3334                 if (thread_group_leader(pos))
3335                         pos = NULL;
3336                 else
3337                         get_task_struct(pos);
3338         }
3339         rcu_read_unlock();
3340         put_task_struct(start);
3341         return pos;
3342 }
3343
3344 /* for the /proc/TGID/task/ directories */
3345 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3346 {
3347         struct inode *inode = file_inode(file);
3348         struct task_struct *task;
3349         struct pid_namespace *ns;
3350         int tid;
3351
3352         if (proc_inode_is_dead(inode))
3353                 return -ENOENT;
3354
3355         if (!dir_emit_dots(file, ctx))
3356                 return 0;
3357
3358         /* f_version caches the tgid value that the last readdir call couldn't
3359          * return. lseek aka telldir automagically resets f_version to 0.
3360          */
3361         ns = inode->i_sb->s_fs_info;
3362         tid = (int)file->f_version;
3363         file->f_version = 0;
3364         for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3365              task;
3366              task = next_tid(task), ctx->pos++) {
3367                 char name[PROC_NUMBUF];
3368                 int len;
3369                 tid = task_pid_nr_ns(task, ns);
3370                 len = snprintf(name, sizeof(name), "%d", tid);
3371                 if (!proc_fill_cache(file, ctx, name, len,
3372                                 proc_task_instantiate, task, NULL)) {
3373                         /* returning this tgid failed, save it as the first
3374                          * pid for the next readir call */
3375                         file->f_version = (u64)tid;
3376                         put_task_struct(task);
3377                         break;
3378                 }
3379         }
3380
3381         return 0;
3382 }
3383
3384 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3385 {
3386         struct inode *inode = d_inode(dentry);
3387         struct task_struct *p = get_proc_task(inode);
3388         generic_fillattr(inode, stat);
3389
3390         if (p) {
3391                 stat->nlink += get_nr_threads(p);
3392                 put_task_struct(p);
3393         }
3394
3395         return 0;
3396 }
3397
3398 static const struct inode_operations proc_task_inode_operations = {
3399         .lookup         = proc_task_lookup,
3400         .getattr        = proc_task_getattr,
3401         .setattr        = proc_setattr,
3402         .permission     = proc_pid_permission,
3403 };
3404
3405 static const struct file_operations proc_task_operations = {
3406         .read           = generic_read_dir,
3407         .iterate        = proc_task_readdir,
3408         .llseek         = default_llseek,
3409 };