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