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         if (!mm)
 958                 return 0;
 959
 960         page = (char *)__get_free_page(GFP_TEMPORARY);
 961         if (!page)
 962                 return -ENOMEM;
 963
 964         ret = 0;
 965         if (!atomic_inc_not_zero(&mm->mm_users))
 966                 goto free;
 967         while (count > 0) {
 968                 size_t this_len, max_len;
 969                 int retval;
 970
 971                 if (src >= (mm->env_end - mm->env_start))
 972                         break;
 973
 974                 this_len = mm->env_end - (mm->env_start + src);
 975
 976                 max_len = min_t(size_t, PAGE_SIZE, count);
 977                 this_len = min(max_len, this_len);
 978
 979                 retval = access_remote_vm(mm, (mm->env_start + src),
 980                         page, this_len, 0);
 981
 982                 if (retval <= 0) {
 983                         ret = retval;
 984                         break;
 985                 }
 986
 987                 if (copy_to_user(buf, page, retval)) {
 988                         ret = -EFAULT;
 989                         break;
 990                 }
 991
 992                 ret += retval;
 993                 src += retval;
 994                 buf += retval;
 995                 count -= retval;
 996         }
 997         *ppos = src;
 998         mmput(mm);
 999
1000 free:
1001         free_page((unsigned long) page);
1002         return ret;
1003 }
1004
1005 static const struct file_operations proc_environ_operations = {
1006         .open           = environ_open,
1007         .read           = environ_read,
1008         .llseek         = generic_file_llseek,
1009         .release        = mem_release,
1010 };
1011
1012 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1013                             loff_t *ppos)
1014 {
1015         struct task_struct *task = get_proc_task(file_inode(file));
1016         char buffer[PROC_NUMBUF];
1017         int oom_adj = OOM_ADJUST_MIN;
1018         size_t len;
1019         unsigned long flags;
1020
1021         if (!task)
1022                 return -ESRCH;
1023         if (lock_task_sighand(task, &flags)) {
1024                 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1025                         oom_adj = OOM_ADJUST_MAX;
1026                 else
1027                         oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1028                                   OOM_SCORE_ADJ_MAX;
1029                 unlock_task_sighand(task, &flags);
1030         }
1031         put_task_struct(task);
1032         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1033         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1034 }
1035
1036 /*
1037  * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1038  * kernels.  The effective policy is defined by oom_score_adj, which has a
1039  * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1040  * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1041  * Processes that become oom disabled via oom_adj will still be oom disabled
1042  * with this implementation.
1043  *
1044  * oom_adj cannot be removed since existing userspace binaries use it.
1045  */
1046 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1047                              size_t count, loff_t *ppos)
1048 {
1049         struct task_struct *task;
1050         char buffer[PROC_NUMBUF];
1051         int oom_adj;
1052         unsigned long flags;
1053         int err;
1054
1055         memset(buffer, 0, sizeof(buffer));
1056         if (count > sizeof(buffer) - 1)
1057                 count = sizeof(buffer) - 1;
1058         if (copy_from_user(buffer, buf, count)) {
1059                 err = -EFAULT;
1060                 goto out;
1061         }
1062
1063         err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1064         if (err)
1065                 goto out;
1066         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1067              oom_adj != OOM_DISABLE) {
1068                 err = -EINVAL;
1069                 goto out;
1070         }
1071
1072         task = get_proc_task(file_inode(file));
1073         if (!task) {
1074                 err = -ESRCH;
1075                 goto out;
1076         }
1077
1078         task_lock(task);
1079         if (!task->mm) {
1080                 err = -EINVAL;
1081                 goto err_task_lock;
1082         }
1083
1084         if (!lock_task_sighand(task, &flags)) {
1085                 err = -ESRCH;
1086                 goto err_task_lock;
1087         }
1088
1089         /*
1090          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1091          * value is always attainable.
1092          */
1093         if (oom_adj == OOM_ADJUST_MAX)
1094                 oom_adj = OOM_SCORE_ADJ_MAX;
1095         else
1096                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1097
1098         if (oom_adj < task->signal->oom_score_adj &&
1099             !capable(CAP_SYS_RESOURCE)) {
1100                 err = -EACCES;
1101                 goto err_sighand;
1102         }
1103
1104         /*
1105          * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1106          * /proc/pid/oom_score_adj instead.
1107          */
1108         pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1109                   current->comm, task_pid_nr(current), task_pid_nr(task),
1110                   task_pid_nr(task));
1111
1112         task->signal->oom_score_adj = oom_adj;
1113         trace_oom_score_adj_update(task);
1114 err_sighand:
1115         unlock_task_sighand(task, &flags);
1116 err_task_lock:
1117         task_unlock(task);
1118         put_task_struct(task);
1119 out:
1120         return err < 0 ? err : count;
1121 }
1122
1123 static const struct file_operations proc_oom_adj_operations = {
1124         .read           = oom_adj_read,
1125         .write          = oom_adj_write,
1126         .llseek         = generic_file_llseek,
1127 };
1128
1129 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1130                                         size_t count, loff_t *ppos)
1131 {
1132         struct task_struct *task = get_proc_task(file_inode(file));
1133         char buffer[PROC_NUMBUF];
1134         short oom_score_adj = OOM_SCORE_ADJ_MIN;
1135         unsigned long flags;
1136         size_t len;
1137
1138         if (!task)
1139                 return -ESRCH;
1140         if (lock_task_sighand(task, &flags)) {
1141                 oom_score_adj = task->signal->oom_score_adj;
1142                 unlock_task_sighand(task, &flags);
1143         }
1144         put_task_struct(task);
1145         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1146         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1147 }
1148
1149 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1150                                         size_t count, loff_t *ppos)
1151 {
1152         struct task_struct *task;
1153         char buffer[PROC_NUMBUF];
1154         unsigned long flags;
1155         int oom_score_adj;
1156         int err;
1157
1158         memset(buffer, 0, sizeof(buffer));
1159         if (count > sizeof(buffer) - 1)
1160                 count = sizeof(buffer) - 1;
1161         if (copy_from_user(buffer, buf, count)) {
1162                 err = -EFAULT;
1163                 goto out;
1164         }
1165
1166         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1167         if (err)
1168                 goto out;
1169         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1170                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1171                 err = -EINVAL;
1172                 goto out;
1173         }
1174
1175         task = get_proc_task(file_inode(file));
1176         if (!task) {
1177                 err = -ESRCH;
1178                 goto out;
1179         }
1180
1181         task_lock(task);
1182         if (!task->mm) {
1183                 err = -EINVAL;
1184                 goto err_task_lock;
1185         }
1186
1187         if (!lock_task_sighand(task, &flags)) {
1188                 err = -ESRCH;
1189                 goto err_task_lock;
1190         }
1191
1192         if ((short)oom_score_adj < task->signal->oom_score_adj_min &&
1193                         !capable(CAP_SYS_RESOURCE)) {
1194                 err = -EACCES;
1195                 goto err_sighand;
1196         }
1197
1198         task->signal->oom_score_adj = (short)oom_score_adj;
1199         if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1200                 task->signal->oom_score_adj_min = (short)oom_score_adj;
1201         trace_oom_score_adj_update(task);
1202
1203 err_sighand:
1204         unlock_task_sighand(task, &flags);
1205 err_task_lock:
1206         task_unlock(task);
1207         put_task_struct(task);
1208 out:
1209         return err < 0 ? err : count;
1210 }
1211
1212 static const struct file_operations proc_oom_score_adj_operations = {
1213         .read           = oom_score_adj_read,
1214         .write          = oom_score_adj_write,
1215         .llseek         = default_llseek,
1216 };
1217
1218 #ifdef CONFIG_AUDITSYSCALL
1219 #define TMPBUFLEN 21
1220 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1221                                   size_t count, loff_t *ppos)
1222 {
1223         struct inode * inode = file_inode(file);
1224         struct task_struct *task = get_proc_task(inode);
1225         ssize_t length;
1226         char tmpbuf[TMPBUFLEN];
1227
1228         if (!task)
1229                 return -ESRCH;
1230         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1231                            from_kuid(file->f_cred->user_ns,
1232                                      audit_get_loginuid(task)));
1233         put_task_struct(task);
1234         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1235 }
1236
1237 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1238                                    size_t count, loff_t *ppos)
1239 {
1240         struct inode * inode = file_inode(file);
1241         uid_t loginuid;
1242         kuid_t kloginuid;
1243         int rv;
1244
1245         rcu_read_lock();
1246         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1247                 rcu_read_unlock();
1248                 return -EPERM;
1249         }
1250         rcu_read_unlock();
1251
1252         if (*ppos != 0) {
1253                 /* No partial writes. */
1254                 return -EINVAL;
1255         }
1256
1257         rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1258         if (rv < 0)
1259                 return rv;
1260
1261         /* is userspace tring to explicitly UNSET the loginuid? */
1262         if (loginuid == AUDIT_UID_UNSET) {
1263                 kloginuid = INVALID_UID;
1264         } else {
1265                 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1266                 if (!uid_valid(kloginuid))
1267                         return -EINVAL;
1268         }
1269
1270         rv = audit_set_loginuid(kloginuid);
1271         if (rv < 0)
1272                 return rv;
1273         return count;
1274 }
1275
1276 static const struct file_operations proc_loginuid_operations = {
1277         .read           = proc_loginuid_read,
1278         .write          = proc_loginuid_write,
1279         .llseek         = generic_file_llseek,
1280 };
1281
1282 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1283                                   size_t count, loff_t *ppos)
1284 {
1285         struct inode * inode = file_inode(file);
1286         struct task_struct *task = get_proc_task(inode);
1287         ssize_t length;
1288         char tmpbuf[TMPBUFLEN];
1289
1290         if (!task)
1291                 return -ESRCH;
1292         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1293                                 audit_get_sessionid(task));
1294         put_task_struct(task);
1295         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1296 }
1297
1298 static const struct file_operations proc_sessionid_operations = {
1299         .read           = proc_sessionid_read,
1300         .llseek         = generic_file_llseek,
1301 };
1302 #endif
1303
1304 #ifdef CONFIG_FAULT_INJECTION
1305 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1306                                       size_t count, loff_t *ppos)
1307 {
1308         struct task_struct *task = get_proc_task(file_inode(file));
1309         char buffer[PROC_NUMBUF];
1310         size_t len;
1311         int make_it_fail;
1312
1313         if (!task)
1314                 return -ESRCH;
1315         make_it_fail = task->make_it_fail;
1316         put_task_struct(task);
1317
1318         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1319
1320         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1321 }
1322
1323 static ssize_t proc_fault_inject_write(struct file * file,
1324                         const char __user * buf, size_t count, loff_t *ppos)
1325 {
1326         struct task_struct *task;
1327         char buffer[PROC_NUMBUF];
1328         int make_it_fail;
1329         int rv;
1330
1331         if (!capable(CAP_SYS_RESOURCE))
1332                 return -EPERM;
1333         memset(buffer, 0, sizeof(buffer));
1334         if (count > sizeof(buffer) - 1)
1335                 count = sizeof(buffer) - 1;
1336         if (copy_from_user(buffer, buf, count))
1337                 return -EFAULT;
1338         rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1339         if (rv < 0)
1340                 return rv;
1341         if (make_it_fail < 0 || make_it_fail > 1)
1342                 return -EINVAL;
1343
1344         task = get_proc_task(file_inode(file));
1345         if (!task)
1346                 return -ESRCH;
1347         task->make_it_fail = make_it_fail;
1348         put_task_struct(task);
1349
1350         return count;
1351 }
1352
1353 static const struct file_operations proc_fault_inject_operations = {
1354         .read           = proc_fault_inject_read,
1355         .write          = proc_fault_inject_write,
1356         .llseek         = generic_file_llseek,
1357 };
1358 #endif
1359
1360
1361 #ifdef CONFIG_SCHED_DEBUG
1362 /*
1363  * Print out various scheduling related per-task fields:
1364  */
1365 static int sched_show(struct seq_file *m, void *v)
1366 {
1367         struct inode *inode = m->private;
1368         struct task_struct *p;
1369
1370         p = get_proc_task(inode);
1371         if (!p)
1372                 return -ESRCH;
1373         proc_sched_show_task(p, m);
1374
1375         put_task_struct(p);
1376
1377         return 0;
1378 }
1379
1380 static ssize_t
1381 sched_write(struct file *file, const char __user *buf,
1382             size_t count, loff_t *offset)
1383 {
1384         struct inode *inode = file_inode(file);
1385         struct task_struct *p;
1386
1387         p = get_proc_task(inode);
1388         if (!p)
1389                 return -ESRCH;
1390         proc_sched_set_task(p);
1391
1392         put_task_struct(p);
1393
1394         return count;
1395 }
1396
1397 static int sched_open(struct inode *inode, struct file *filp)
1398 {
1399         return single_open(filp, sched_show, inode);
1400 }
1401
1402 static const struct file_operations proc_pid_sched_operations = {
1403         .open           = sched_open,
1404         .read           = seq_read,
1405         .write          = sched_write,
1406         .llseek         = seq_lseek,
1407         .release        = single_release,
1408 };
1409
1410 #endif
1411
1412 #ifdef CONFIG_SCHED_AUTOGROUP
1413 /*
1414  * Print out autogroup related information:
1415  */
1416 static int sched_autogroup_show(struct seq_file *m, void *v)
1417 {
1418         struct inode *inode = m->private;
1419         struct task_struct *p;
1420
1421         p = get_proc_task(inode);
1422         if (!p)
1423                 return -ESRCH;
1424         proc_sched_autogroup_show_task(p, m);
1425
1426         put_task_struct(p);
1427
1428         return 0;
1429 }
1430
1431 static ssize_t
1432 sched_autogroup_write(struct file *file, const char __user *buf,
1433             size_t count, loff_t *offset)
1434 {
1435         struct inode *inode = file_inode(file);
1436         struct task_struct *p;
1437         char buffer[PROC_NUMBUF];
1438         int nice;
1439         int err;
1440
1441         memset(buffer, 0, sizeof(buffer));
1442         if (count > sizeof(buffer) - 1)
1443                 count = sizeof(buffer) - 1;
1444         if (copy_from_user(buffer, buf, count))
1445                 return -EFAULT;
1446
1447         err = kstrtoint(strstrip(buffer), 0, &nice);
1448         if (err < 0)
1449                 return err;
1450
1451         p = get_proc_task(inode);
1452         if (!p)
1453                 return -ESRCH;
1454
1455         err = proc_sched_autogroup_set_nice(p, nice);
1456         if (err)
1457                 count = err;
1458
1459         put_task_struct(p);
1460
1461         return count;
1462 }
1463
1464 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1465 {
1466         int ret;
1467
1468         ret = single_open(filp, sched_autogroup_show, NULL);
1469         if (!ret) {
1470                 struct seq_file *m = filp->private_data;
1471
1472                 m->private = inode;
1473         }
1474         return ret;
1475 }
1476
1477 static const struct file_operations proc_pid_sched_autogroup_operations = {
1478         .open           = sched_autogroup_open,
1479         .read           = seq_read,
1480         .write          = sched_autogroup_write,
1481         .llseek         = seq_lseek,
1482         .release        = single_release,
1483 };
1484
1485 #endif /* CONFIG_SCHED_AUTOGROUP */
1486
1487 static ssize_t comm_write(struct file *file, const char __user *buf,
1488                                 size_t count, loff_t *offset)
1489 {
1490         struct inode *inode = file_inode(file);
1491         struct task_struct *p;
1492         char buffer[TASK_COMM_LEN];
1493         const size_t maxlen = sizeof(buffer) - 1;
1494
1495         memset(buffer, 0, sizeof(buffer));
1496         if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1497                 return -EFAULT;
1498
1499         p = get_proc_task(inode);
1500         if (!p)
1501                 return -ESRCH;
1502
1503         if (same_thread_group(current, p))
1504                 set_task_comm(p, buffer);
1505         else
1506                 count = -EINVAL;
1507
1508         put_task_struct(p);
1509
1510         return count;
1511 }
1512
1513 static int comm_show(struct seq_file *m, void *v)
1514 {
1515         struct inode *inode = m->private;
1516         struct task_struct *p;
1517
1518         p = get_proc_task(inode);
1519         if (!p)
1520                 return -ESRCH;
1521
1522         task_lock(p);
1523         seq_printf(m, "%s\n", p->comm);
1524         task_unlock(p);
1525
1526         put_task_struct(p);
1527
1528         return 0;
1529 }
1530
1531 static int comm_open(struct inode *inode, struct file *filp)
1532 {
1533         return single_open(filp, comm_show, inode);
1534 }
1535
1536 static const struct file_operations proc_pid_set_comm_operations = {
1537         .open           = comm_open,
1538         .read           = seq_read,
1539         .write          = comm_write,
1540         .llseek         = seq_lseek,
1541         .release        = single_release,
1542 };
1543
1544 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1545 {
1546         struct task_struct *task;
1547         struct mm_struct *mm;
1548         struct file *exe_file;
1549
1550         task = get_proc_task(d_inode(dentry));
1551         if (!task)
1552                 return -ENOENT;
1553         mm = get_task_mm(task);
1554         put_task_struct(task);
1555         if (!mm)
1556                 return -ENOENT;
1557         exe_file = get_mm_exe_file(mm);
1558         mmput(mm);
1559         if (exe_file) {
1560                 *exe_path = exe_file->f_path;
1561                 path_get(&exe_file->f_path);
1562                 fput(exe_file);
1563                 return 0;
1564         } else
1565                 return -ENOENT;
1566 }
1567
1568 static const char *proc_pid_follow_link(struct dentry *dentry, void **cookie)
1569 {
1570         struct inode *inode = d_inode(dentry);
1571         struct path path;
1572         int error = -EACCES;
1573
1574         /* Are we allowed to snoop on the tasks file descriptors? */
1575         if (!proc_fd_access_allowed(inode))
1576                 goto out;
1577
1578         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1579         if (error)
1580                 goto out;
1581
1582         nd_jump_link(&path);
1583         return NULL;
1584 out:
1585         return ERR_PTR(error);
1586 }
1587
1588 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1589 {
1590         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1591         char *pathname;
1592         int len;
1593
1594         if (!tmp)
1595                 return -ENOMEM;
1596
1597         pathname = d_path(path, tmp, PAGE_SIZE);
1598         len = PTR_ERR(pathname);
1599         if (IS_ERR(pathname))
1600                 goto out;
1601         len = tmp + PAGE_SIZE - 1 - pathname;
1602
1603         if (len > buflen)
1604                 len = buflen;
1605         if (copy_to_user(buffer, pathname, len))
1606                 len = -EFAULT;
1607  out:
1608         free_page((unsigned long)tmp);
1609         return len;
1610 }
1611
1612 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1613 {
1614         int error = -EACCES;
1615         struct inode *inode = d_inode(dentry);
1616         struct path path;
1617
1618         /* Are we allowed to snoop on the tasks file descriptors? */
1619         if (!proc_fd_access_allowed(inode))
1620                 goto out;
1621
1622         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1623         if (error)
1624                 goto out;
1625
1626         error = do_proc_readlink(&path, buffer, buflen);
1627         path_put(&path);
1628 out:
1629         return error;
1630 }
1631
1632 const struct inode_operations proc_pid_link_inode_operations = {
1633         .readlink       = proc_pid_readlink,
1634         .follow_link    = proc_pid_follow_link,
1635         .setattr        = proc_setattr,
1636 };
1637
1638
1639 /* building an inode */
1640
1641 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1642 {
1643         struct inode * inode;
1644         struct proc_inode *ei;
1645         const struct cred *cred;
1646
1647         /* We need a new inode */
1648
1649         inode = new_inode(sb);
1650         if (!inode)
1651                 goto out;
1652
1653         /* Common stuff */
1654         ei = PROC_I(inode);
1655         inode->i_ino = get_next_ino();
1656         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1657         inode->i_op = &proc_def_inode_operations;
1658
1659         /*
1660          * grab the reference to task.
1661          */
1662         ei->pid = get_task_pid(task, PIDTYPE_PID);
1663         if (!ei->pid)
1664                 goto out_unlock;
1665
1666         if (task_dumpable(task)) {
1667                 rcu_read_lock();
1668                 cred = __task_cred(task);
1669                 inode->i_uid = cred->euid;
1670                 inode->i_gid = cred->egid;
1671                 rcu_read_unlock();
1672         }
1673         security_task_to_inode(task, inode);
1674
1675 out:
1676         return inode;
1677
1678 out_unlock:
1679         iput(inode);
1680         return NULL;
1681 }
1682
1683 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1684 {
1685         struct inode *inode = d_inode(dentry);
1686         struct task_struct *task;
1687         const struct cred *cred;
1688         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1689
1690         generic_fillattr(inode, stat);
1691
1692         rcu_read_lock();
1693         stat->uid = GLOBAL_ROOT_UID;
1694         stat->gid = GLOBAL_ROOT_GID;
1695         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1696         if (task) {
1697                 if (!has_pid_permissions(pid, task, 2)) {
1698                         rcu_read_unlock();
1699                         /*
1700                          * This doesn't prevent learning whether PID exists,
1701                          * it only makes getattr() consistent with readdir().
1702                          */
1703                         return -ENOENT;
1704                 }
1705                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1706                     task_dumpable(task)) {
1707                         cred = __task_cred(task);
1708                         stat->uid = cred->euid;
1709                         stat->gid = cred->egid;
1710                 }
1711         }
1712         rcu_read_unlock();
1713         return 0;
1714 }
1715
1716 /* dentry stuff */
1717
1718 /*
1719  *      Exceptional case: normally we are not allowed to unhash a busy
1720  * directory. In this case, however, we can do it - no aliasing problems
1721  * due to the way we treat inodes.
1722  *
1723  * Rewrite the inode's ownerships here because the owning task may have
1724  * performed a setuid(), etc.
1725  *
1726  * Before the /proc/pid/status file was created the only way to read
1727  * the effective uid of a /process was to stat /proc/pid.  Reading
1728  * /proc/pid/status is slow enough that procps and other packages
1729  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1730  * made this apply to all per process world readable and executable
1731  * directories.
1732  */
1733 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1734 {
1735         struct inode *inode;
1736         struct task_struct *task;
1737         const struct cred *cred;
1738
1739         if (flags & LOOKUP_RCU)
1740                 return -ECHILD;
1741
1742         inode = d_inode(dentry);
1743         task = get_proc_task(inode);
1744
1745         if (task) {
1746                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1747                     task_dumpable(task)) {
1748                         rcu_read_lock();
1749                         cred = __task_cred(task);
1750                         inode->i_uid = cred->euid;
1751                         inode->i_gid = cred->egid;
1752                         rcu_read_unlock();
1753                 } else {
1754                         inode->i_uid = GLOBAL_ROOT_UID;
1755                         inode->i_gid = GLOBAL_ROOT_GID;
1756                 }
1757                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1758                 security_task_to_inode(task, inode);
1759                 put_task_struct(task);
1760                 return 1;
1761         }
1762         return 0;
1763 }
1764
1765 static inline bool proc_inode_is_dead(struct inode *inode)
1766 {
1767         return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1768 }
1769
1770 int pid_delete_dentry(const struct dentry *dentry)
1771 {
1772         /* Is the task we represent dead?
1773          * If so, then don't put the dentry on the lru list,
1774          * kill it immediately.
1775          */
1776         return proc_inode_is_dead(d_inode(dentry));
1777 }
1778
1779 const struct dentry_operations pid_dentry_operations =
1780 {
1781         .d_revalidate   = pid_revalidate,
1782         .d_delete       = pid_delete_dentry,
1783 };
1784
1785 /* Lookups */
1786
1787 /*
1788  * Fill a directory entry.
1789  *
1790  * If possible create the dcache entry and derive our inode number and
1791  * file type from dcache entry.
1792  *
1793  * Since all of the proc inode numbers are dynamically generated, the inode
1794  * numbers do not exist until the inode is cache.  This means creating the
1795  * the dcache entry in readdir is necessary to keep the inode numbers
1796  * reported by readdir in sync with the inode numbers reported
1797  * by stat.
1798  */
1799 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1800         const char *name, int len,
1801         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1802 {
1803         struct dentry *child, *dir = file->f_path.dentry;
1804         struct qstr qname = QSTR_INIT(name, len);
1805         struct inode *inode;
1806         unsigned type;
1807         ino_t ino;
1808
1809         child = d_hash_and_lookup(dir, &qname);
1810         if (!child) {
1811                 child = d_alloc(dir, &qname);
1812                 if (!child)
1813                         goto end_instantiate;
1814                 if (instantiate(d_inode(dir), child, task, ptr) < 0) {
1815                         dput(child);
1816                         goto end_instantiate;
1817                 }
1818         }
1819         inode = d_inode(child);
1820         ino = inode->i_ino;
1821         type = inode->i_mode >> 12;
1822         dput(child);
1823         return dir_emit(ctx, name, len, ino, type);
1824
1825 end_instantiate:
1826         return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1827 }
1828
1829 /*
1830  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1831  * which represent vma start and end addresses.
1832  */
1833 static int dname_to_vma_addr(struct dentry *dentry,
1834                              unsigned long *start, unsigned long *end)
1835 {
1836         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1837                 return -EINVAL;
1838
1839         return 0;
1840 }
1841
1842 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1843 {
1844         unsigned long vm_start, vm_end;
1845         bool exact_vma_exists = false;
1846         struct mm_struct *mm = NULL;
1847         struct task_struct *task;
1848         const struct cred *cred;
1849         struct inode *inode;
1850         int status = 0;
1851
1852         if (flags & LOOKUP_RCU)
1853                 return -ECHILD;
1854
1855         inode = d_inode(dentry);
1856         task = get_proc_task(inode);
1857         if (!task)
1858                 goto out_notask;
1859
1860         mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1861         if (IS_ERR_OR_NULL(mm))
1862                 goto out;
1863
1864         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1865                 down_read(&mm->mmap_sem);
1866                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1867                 up_read(&mm->mmap_sem);
1868         }
1869
1870         mmput(mm);
1871
1872         if (exact_vma_exists) {
1873                 if (task_dumpable(task)) {
1874                         rcu_read_lock();
1875                         cred = __task_cred(task);
1876                         inode->i_uid = cred->euid;
1877                         inode->i_gid = cred->egid;
1878                         rcu_read_unlock();
1879                 } else {
1880                         inode->i_uid = GLOBAL_ROOT_UID;
1881                         inode->i_gid = GLOBAL_ROOT_GID;
1882                 }
1883                 security_task_to_inode(task, inode);
1884                 status = 1;
1885         }
1886
1887 out:
1888         put_task_struct(task);
1889
1890 out_notask:
1891         return status;
1892 }
1893
1894 static const struct dentry_operations tid_map_files_dentry_operations = {
1895         .d_revalidate   = map_files_d_revalidate,
1896         .d_delete       = pid_delete_dentry,
1897 };
1898
1899 static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
1900 {
1901         unsigned long vm_start, vm_end;
1902         struct vm_area_struct *vma;
1903         struct task_struct *task;
1904         struct mm_struct *mm;
1905         int rc;
1906
1907         rc = -ENOENT;
1908         task = get_proc_task(d_inode(dentry));
1909         if (!task)
1910                 goto out;
1911
1912         mm = get_task_mm(task);
1913         put_task_struct(task);
1914         if (!mm)
1915                 goto out;
1916
1917         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1918         if (rc)
1919                 goto out_mmput;
1920
1921         rc = -ENOENT;
1922         down_read(&mm->mmap_sem);
1923         vma = find_exact_vma(mm, vm_start, vm_end);
1924         if (vma && vma->vm_file) {
1925                 *path = vma->vm_file->f_path;
1926                 path_get(path);
1927                 rc = 0;
1928         }
1929         up_read(&mm->mmap_sem);
1930
1931 out_mmput:
1932         mmput(mm);
1933 out:
1934         return rc;
1935 }
1936
1937 struct map_files_info {
1938         fmode_t         mode;
1939         unsigned long   len;
1940         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1941 };
1942
1943 /*
1944  * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
1945  * symlinks may be used to bypass permissions on ancestor directories in the
1946  * path to the file in question.
1947  */
1948 static const char *
1949 proc_map_files_follow_link(struct dentry *dentry, void **cookie)
1950 {
1951         if (!capable(CAP_SYS_ADMIN))
1952                 return ERR_PTR(-EPERM);
1953
1954         return proc_pid_follow_link(dentry, NULL);
1955 }
1956
1957 /*
1958  * Identical to proc_pid_link_inode_operations except for follow_link()
1959  */
1960 static const struct inode_operations proc_map_files_link_inode_operations = {
1961         .readlink       = proc_pid_readlink,
1962         .follow_link    = proc_map_files_follow_link,
1963         .setattr        = proc_setattr,
1964 };
1965
1966 static int
1967 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
1968                            struct task_struct *task, const void *ptr)
1969 {
1970         fmode_t mode = (fmode_t)(unsigned long)ptr;
1971         struct proc_inode *ei;
1972         struct inode *inode;
1973
1974         inode = proc_pid_make_inode(dir->i_sb, task);
1975         if (!inode)
1976                 return -ENOENT;
1977
1978         ei = PROC_I(inode);
1979         ei->op.proc_get_link = proc_map_files_get_link;
1980
1981         inode->i_op = &proc_map_files_link_inode_operations;
1982         inode->i_size = 64;
1983         inode->i_mode = S_IFLNK;
1984
1985         if (mode & FMODE_READ)
1986                 inode->i_mode |= S_IRUSR;
1987         if (mode & FMODE_WRITE)
1988                 inode->i_mode |= S_IWUSR;
1989
1990         d_set_d_op(dentry, &tid_map_files_dentry_operations);
1991         d_add(dentry, inode);
1992
1993         return 0;
1994 }
1995
1996 static struct dentry *proc_map_files_lookup(struct inode *dir,
1997                 struct dentry *dentry, unsigned int flags)
1998 {
1999         unsigned long vm_start, vm_end;
2000         struct vm_area_struct *vma;
2001         struct task_struct *task;
2002         int result;
2003         struct mm_struct *mm;
2004
2005         result = -ENOENT;
2006         task = get_proc_task(dir);
2007         if (!task)
2008                 goto out;
2009
2010         result = -EACCES;
2011         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2012                 goto out_put_task;
2013
2014         result = -ENOENT;
2015         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2016                 goto out_put_task;
2017
2018         mm = get_task_mm(task);
2019         if (!mm)
2020                 goto out_put_task;
2021
2022         down_read(&mm->mmap_sem);
2023         vma = find_exact_vma(mm, vm_start, vm_end);
2024         if (!vma)
2025                 goto out_no_vma;
2026
2027         if (vma->vm_file)
2028                 result = proc_map_files_instantiate(dir, dentry, task,
2029                                 (void *)(unsigned long)vma->vm_file->f_mode);
2030
2031 out_no_vma:
2032         up_read(&mm->mmap_sem);
2033         mmput(mm);
2034 out_put_task:
2035         put_task_struct(task);
2036 out:
2037         return ERR_PTR(result);
2038 }
2039
2040 static const struct inode_operations proc_map_files_inode_operations = {
2041         .lookup         = proc_map_files_lookup,
2042         .permission     = proc_fd_permission,
2043         .setattr        = proc_setattr,
2044 };
2045
2046 static int
2047 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2048 {
2049         struct vm_area_struct *vma;
2050         struct task_struct *task;
2051         struct mm_struct *mm;
2052         unsigned long nr_files, pos, i;
2053         struct flex_array *fa = NULL;
2054         struct map_files_info info;
2055         struct map_files_info *p;
2056         int ret;
2057
2058         ret = -ENOENT;
2059         task = get_proc_task(file_inode(file));
2060         if (!task)
2061                 goto out;
2062
2063         ret = -EACCES;
2064         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2065                 goto out_put_task;
2066
2067         ret = 0;
2068         if (!dir_emit_dots(file, ctx))
2069                 goto out_put_task;
2070
2071         mm = get_task_mm(task);
2072         if (!mm)
2073                 goto out_put_task;
2074         down_read(&mm->mmap_sem);
2075
2076         nr_files = 0;
2077
2078         /*
2079          * We need two passes here:
2080          *
2081          *  1) Collect vmas of mapped files with mmap_sem taken
2082          *  2) Release mmap_sem and instantiate entries
2083          *
2084          * otherwise we get lockdep complained, since filldir()
2085          * routine might require mmap_sem taken in might_fault().
2086          */
2087
2088         for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2089                 if (vma->vm_file && ++pos > ctx->pos)
2090                         nr_files++;
2091         }
2092
2093         if (nr_files) {
2094                 fa = flex_array_alloc(sizeof(info), nr_files,
2095                                         GFP_KERNEL);
2096                 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2097                                                 GFP_KERNEL)) {
2098                         ret = -ENOMEM;
2099                         if (fa)
2100                                 flex_array_free(fa);
2101                         up_read(&mm->mmap_sem);
2102                         mmput(mm);
2103                         goto out_put_task;
2104                 }
2105                 for (i = 0, vma = mm->mmap, pos = 2; vma;
2106                                 vma = vma->vm_next) {
2107                         if (!vma->vm_file)
2108                                 continue;
2109                         if (++pos <= ctx->pos)
2110                                 continue;
2111
2112                         info.mode = vma->vm_file->f_mode;
2113                         info.len = snprintf(info.name,
2114                                         sizeof(info.name), "%lx-%lx",
2115                                         vma->vm_start, vma->vm_end);
2116                         if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2117                                 BUG();
2118                 }
2119         }
2120         up_read(&mm->mmap_sem);
2121
2122         for (i = 0; i < nr_files; i++) {
2123                 p = flex_array_get(fa, i);
2124                 if (!proc_fill_cache(file, ctx,
2125                                       p->name, p->len,
2126                                       proc_map_files_instantiate,
2127                                       task,
2128                                       (void *)(unsigned long)p->mode))
2129                         break;
2130                 ctx->pos++;
2131         }
2132         if (fa)
2133                 flex_array_free(fa);
2134         mmput(mm);
2135
2136 out_put_task:
2137         put_task_struct(task);
2138 out:
2139         return ret;
2140 }
2141
2142 static const struct file_operations proc_map_files_operations = {
2143         .read           = generic_read_dir,
2144         .iterate        = proc_map_files_readdir,
2145         .llseek         = default_llseek,
2146 };
2147
2148 struct timers_private {
2149         struct pid *pid;
2150         struct task_struct *task;
2151         struct sighand_struct *sighand;
2152         struct pid_namespace *ns;
2153         unsigned long flags;
2154 };
2155
2156 static void *timers_start(struct seq_file *m, loff_t *pos)
2157 {
2158         struct timers_private *tp = m->private;
2159
2160         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2161         if (!tp->task)
2162                 return ERR_PTR(-ESRCH);
2163
2164         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2165         if (!tp->sighand)
2166                 return ERR_PTR(-ESRCH);
2167
2168         return seq_list_start(&tp->task->signal->posix_timers, *pos);
2169 }
2170
2171 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2172 {
2173         struct timers_private *tp = m->private;
2174         return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2175 }
2176
2177 static void timers_stop(struct seq_file *m, void *v)
2178 {
2179         struct timers_private *tp = m->private;
2180
2181         if (tp->sighand) {
2182                 unlock_task_sighand(tp->task, &tp->flags);
2183                 tp->sighand = NULL;
2184         }
2185
2186         if (tp->task) {
2187                 put_task_struct(tp->task);
2188                 tp->task = NULL;
2189         }
2190 }
2191
2192 static int show_timer(struct seq_file *m, void *v)
2193 {
2194         struct k_itimer *timer;
2195         struct timers_private *tp = m->private;
2196         int notify;
2197         static const char * const nstr[] = {
2198                 [SIGEV_SIGNAL] = "signal",
2199                 [SIGEV_NONE] = "none",
2200                 [SIGEV_THREAD] = "thread",
2201         };
2202
2203         timer = list_entry((struct list_head *)v, struct k_itimer, list);
2204         notify = timer->it_sigev_notify;
2205
2206         seq_printf(m, "ID: %d\n", timer->it_id);
2207         seq_printf(m, "signal: %d/%p\n",
2208                    timer->sigq->info.si_signo,
2209                    timer->sigq->info.si_value.sival_ptr);
2210         seq_printf(m, "notify: %s/%s.%d\n",
2211                    nstr[notify & ~SIGEV_THREAD_ID],
2212                    (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2213                    pid_nr_ns(timer->it_pid, tp->ns));
2214         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2215
2216         return 0;
2217 }
2218
2219 static const struct seq_operations proc_timers_seq_ops = {
2220         .start  = timers_start,
2221         .next   = timers_next,
2222         .stop   = timers_stop,
2223         .show   = show_timer,
2224 };
2225
2226 static int proc_timers_open(struct inode *inode, struct file *file)
2227 {
2228         struct timers_private *tp;
2229
2230         tp = __seq_open_private(file, &proc_timers_seq_ops,
2231                         sizeof(struct timers_private));
2232         if (!tp)
2233                 return -ENOMEM;
2234
2235         tp->pid = proc_pid(inode);
2236         tp->ns = inode->i_sb->s_fs_info;
2237         return 0;
2238 }
2239
2240 static const struct file_operations proc_timers_operations = {
2241         .open           = proc_timers_open,
2242         .read           = seq_read,
2243         .llseek         = seq_lseek,
2244         .release        = seq_release_private,
2245 };
2246
2247 static int proc_pident_instantiate(struct inode *dir,
2248         struct dentry *dentry, struct task_struct *task, const void *ptr)
2249 {
2250         const struct pid_entry *p = ptr;
2251         struct inode *inode;
2252         struct proc_inode *ei;
2253
2254         inode = proc_pid_make_inode(dir->i_sb, task);
2255         if (!inode)
2256                 goto out;
2257
2258         ei = PROC_I(inode);
2259         inode->i_mode = p->mode;
2260         if (S_ISDIR(inode->i_mode))
2261                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2262         if (p->iop)
2263                 inode->i_op = p->iop;
2264         if (p->fop)
2265                 inode->i_fop = p->fop;
2266         ei->op = p->op;
2267         d_set_d_op(dentry, &pid_dentry_operations);
2268         d_add(dentry, inode);
2269         /* Close the race of the process dying before we return the dentry */
2270         if (pid_revalidate(dentry, 0))
2271                 return 0;
2272 out:
2273         return -ENOENT;
2274 }
2275
2276 static struct dentry *proc_pident_lookup(struct inode *dir,
2277                                          struct dentry *dentry,
2278                                          const struct pid_entry *ents,
2279                                          unsigned int nents)
2280 {
2281         int error;
2282         struct task_struct *task = get_proc_task(dir);
2283         const struct pid_entry *p, *last;
2284
2285         error = -ENOENT;
2286
2287         if (!task)
2288                 goto out_no_task;
2289
2290         /*
2291          * Yes, it does not scale. And it should not. Don't add
2292          * new entries into /proc/<tgid>/ without very good reasons.
2293          */
2294         last = &ents[nents - 1];
2295         for (p = ents; p <= last; p++) {
2296                 if (p->len != dentry->d_name.len)
2297                         continue;
2298                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2299                         break;
2300         }
2301         if (p > last)
2302                 goto out;
2303
2304         error = proc_pident_instantiate(dir, dentry, task, p);
2305 out:
2306         put_task_struct(task);
2307 out_no_task:
2308         return ERR_PTR(error);
2309 }
2310
2311 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2312                 const struct pid_entry *ents, unsigned int nents)
2313 {
2314         struct task_struct *task = get_proc_task(file_inode(file));
2315         const struct pid_entry *p;
2316
2317         if (!task)
2318                 return -ENOENT;
2319
2320         if (!dir_emit_dots(file, ctx))
2321                 goto out;
2322
2323         if (ctx->pos >= nents + 2)
2324                 goto out;
2325
2326         for (p = ents + (ctx->pos - 2); p <= ents + nents - 1; p++) {
2327                 if (!proc_fill_cache(file, ctx, p->name, p->len,
2328                                 proc_pident_instantiate, task, p))
2329                         break;
2330                 ctx->pos++;
2331         }
2332 out:
2333         put_task_struct(task);
2334         return 0;
2335 }
2336
2337 #ifdef CONFIG_SECURITY
2338 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2339                                   size_t count, loff_t *ppos)
2340 {
2341         struct inode * inode = file_inode(file);
2342         char *p = NULL;
2343         ssize_t length;
2344         struct task_struct *task = get_proc_task(inode);
2345
2346         if (!task)
2347                 return -ESRCH;
2348
2349         length = security_getprocattr(task,
2350                                       (char*)file->f_path.dentry->d_name.name,
2351                                       &p);
2352         put_task_struct(task);
2353         if (length > 0)
2354                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2355         kfree(p);
2356         return length;
2357 }
2358
2359 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2360                                    size_t count, loff_t *ppos)
2361 {
2362         struct inode * inode = file_inode(file);
2363         char *page;
2364         ssize_t length;
2365         struct task_struct *task = get_proc_task(inode);
2366
2367         length = -ESRCH;
2368         if (!task)
2369                 goto out_no_task;
2370         if (count > PAGE_SIZE)
2371                 count = PAGE_SIZE;
2372
2373         /* No partial writes. */
2374         length = -EINVAL;
2375         if (*ppos != 0)
2376                 goto out;
2377
2378         length = -ENOMEM;
2379         page = (char*)__get_free_page(GFP_TEMPORARY);
2380         if (!page)
2381                 goto out;
2382
2383         length = -EFAULT;
2384         if (copy_from_user(page, buf, count))
2385                 goto out_free;
2386
2387         /* Guard against adverse ptrace interaction */
2388         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2389         if (length < 0)
2390                 goto out_free;
2391
2392         length = security_setprocattr(task,
2393                                       (char*)file->f_path.dentry->d_name.name,
2394                                       (void*)page, count);
2395         mutex_unlock(&task->signal->cred_guard_mutex);
2396 out_free:
2397         free_page((unsigned long) page);
2398 out:
2399         put_task_struct(task);
2400 out_no_task:
2401         return length;
2402 }
2403
2404 static const struct file_operations proc_pid_attr_operations = {
2405         .read           = proc_pid_attr_read,
2406         .write          = proc_pid_attr_write,
2407         .llseek         = generic_file_llseek,
2408 };
2409
2410 static const struct pid_entry attr_dir_stuff[] = {
2411         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2412         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2413         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2414         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2415         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2416         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2417 };
2418
2419 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2420 {
2421         return proc_pident_readdir(file, ctx,
2422                                    attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2423 }
2424
2425 static const struct file_operations proc_attr_dir_operations = {
2426         .read           = generic_read_dir,
2427         .iterate        = proc_attr_dir_readdir,
2428         .llseek         = default_llseek,
2429 };
2430
2431 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2432                                 struct dentry *dentry, unsigned int flags)
2433 {
2434         return proc_pident_lookup(dir, dentry,
2435                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2436 }
2437
2438 static const struct inode_operations proc_attr_dir_inode_operations = {
2439         .lookup         = proc_attr_dir_lookup,
2440         .getattr        = pid_getattr,
2441         .setattr        = proc_setattr,
2442 };
2443
2444 #endif
2445
2446 #ifdef CONFIG_ELF_CORE
2447 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2448                                          size_t count, loff_t *ppos)
2449 {
2450         struct task_struct *task = get_proc_task(file_inode(file));
2451         struct mm_struct *mm;
2452         char buffer[PROC_NUMBUF];
2453         size_t len;
2454         int ret;
2455
2456         if (!task)
2457                 return -ESRCH;
2458
2459         ret = 0;
2460         mm = get_task_mm(task);
2461         if (mm) {
2462                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2463                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2464                                 MMF_DUMP_FILTER_SHIFT));
2465                 mmput(mm);
2466                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2467         }
2468
2469         put_task_struct(task);
2470
2471         return ret;
2472 }
2473
2474 static ssize_t proc_coredump_filter_write(struct file *file,
2475                                           const char __user *buf,
2476                                           size_t count,
2477                                           loff_t *ppos)
2478 {
2479         struct task_struct *task;
2480         struct mm_struct *mm;
2481         unsigned int val;
2482         int ret;
2483         int i;
2484         unsigned long mask;
2485
2486         ret = kstrtouint_from_user(buf, count, 0, &val);
2487         if (ret < 0)
2488                 return ret;
2489
2490         ret = -ESRCH;
2491         task = get_proc_task(file_inode(file));
2492         if (!task)
2493                 goto out_no_task;
2494
2495         mm = get_task_mm(task);
2496         if (!mm)
2497                 goto out_no_mm;
2498         ret = 0;
2499
2500         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2501                 if (val & mask)
2502                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2503                 else
2504                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2505         }
2506
2507         mmput(mm);
2508  out_no_mm:
2509         put_task_struct(task);
2510  out_no_task:
2511         if (ret < 0)
2512                 return ret;
2513         return count;
2514 }
2515
2516 static const struct file_operations proc_coredump_filter_operations = {
2517         .read           = proc_coredump_filter_read,
2518         .write          = proc_coredump_filter_write,
2519         .llseek         = generic_file_llseek,
2520 };
2521 #endif
2522
2523 #ifdef CONFIG_TASK_IO_ACCOUNTING
2524 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2525 {
2526         struct task_io_accounting acct = task->ioac;
2527         unsigned long flags;
2528         int result;
2529
2530         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2531         if (result)
2532                 return result;
2533
2534         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2535                 result = -EACCES;
2536                 goto out_unlock;
2537         }
2538
2539         if (whole && lock_task_sighand(task, &flags)) {
2540                 struct task_struct *t = task;
2541
2542                 task_io_accounting_add(&acct, &task->signal->ioac);
2543                 while_each_thread(task, t)
2544                         task_io_accounting_add(&acct, &t->ioac);
2545
2546                 unlock_task_sighand(task, &flags);
2547         }
2548         seq_printf(m,
2549                    "rchar: %llu\n"
2550                    "wchar: %llu\n"
2551                    "syscr: %llu\n"
2552                    "syscw: %llu\n"
2553                    "read_bytes: %llu\n"
2554                    "write_bytes: %llu\n"
2555                    "cancelled_write_bytes: %llu\n",
2556                    (unsigned long long)acct.rchar,
2557                    (unsigned long long)acct.wchar,
2558                    (unsigned long long)acct.syscr,
2559                    (unsigned long long)acct.syscw,
2560                    (unsigned long long)acct.read_bytes,
2561                    (unsigned long long)acct.write_bytes,
2562                    (unsigned long long)acct.cancelled_write_bytes);
2563         result = 0;
2564
2565 out_unlock:
2566         mutex_unlock(&task->signal->cred_guard_mutex);
2567         return result;
2568 }
2569
2570 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2571                                   struct pid *pid, struct task_struct *task)
2572 {
2573         return do_io_accounting(task, m, 0);
2574 }
2575
2576 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2577                                    struct pid *pid, struct task_struct *task)
2578 {
2579         return do_io_accounting(task, m, 1);
2580 }
2581 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2582
2583 #ifdef CONFIG_USER_NS
2584 static int proc_id_map_open(struct inode *inode, struct file *file,
2585         const struct seq_operations *seq_ops)
2586 {
2587         struct user_namespace *ns = NULL;
2588         struct task_struct *task;
2589         struct seq_file *seq;
2590         int ret = -EINVAL;
2591
2592         task = get_proc_task(inode);
2593         if (task) {
2594                 rcu_read_lock();
2595                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2596                 rcu_read_unlock();
2597                 put_task_struct(task);
2598         }
2599         if (!ns)
2600                 goto err;
2601
2602         ret = seq_open(file, seq_ops);
2603         if (ret)
2604                 goto err_put_ns;
2605
2606         seq = file->private_data;
2607         seq->private = ns;
2608
2609         return 0;
2610 err_put_ns:
2611         put_user_ns(ns);
2612 err:
2613         return ret;
2614 }
2615
2616 static int proc_id_map_release(struct inode *inode, struct file *file)
2617 {
2618         struct seq_file *seq = file->private_data;
2619         struct user_namespace *ns = seq->private;
2620         put_user_ns(ns);
2621         return seq_release(inode, file);
2622 }
2623
2624 static int proc_uid_map_open(struct inode *inode, struct file *file)
2625 {
2626         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2627 }
2628
2629 static int proc_gid_map_open(struct inode *inode, struct file *file)
2630 {
2631         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2632 }
2633
2634 static int proc_projid_map_open(struct inode *inode, struct file *file)
2635 {
2636         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2637 }
2638
2639 static const struct file_operations proc_uid_map_operations = {
2640         .open           = proc_uid_map_open,
2641         .write          = proc_uid_map_write,
2642         .read           = seq_read,
2643         .llseek         = seq_lseek,
2644         .release        = proc_id_map_release,
2645 };
2646
2647 static const struct file_operations proc_gid_map_operations = {
2648         .open           = proc_gid_map_open,
2649         .write          = proc_gid_map_write,
2650         .read           = seq_read,
2651         .llseek         = seq_lseek,
2652         .release        = proc_id_map_release,
2653 };
2654
2655 static const struct file_operations proc_projid_map_operations = {
2656         .open           = proc_projid_map_open,
2657         .write          = proc_projid_map_write,
2658         .read           = seq_read,
2659         .llseek         = seq_lseek,
2660         .release        = proc_id_map_release,
2661 };
2662
2663 static int proc_setgroups_open(struct inode *inode, struct file *file)
2664 {
2665         struct user_namespace *ns = NULL;
2666         struct task_struct *task;
2667         int ret;
2668
2669         ret = -ESRCH;
2670         task = get_proc_task(inode);
2671         if (task) {
2672                 rcu_read_lock();
2673                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2674                 rcu_read_unlock();
2675                 put_task_struct(task);
2676         }
2677         if (!ns)
2678                 goto err;
2679
2680         if (file->f_mode & FMODE_WRITE) {
2681                 ret = -EACCES;
2682                 if (!ns_capable(ns, CAP_SYS_ADMIN))
2683                         goto err_put_ns;
2684         }
2685
2686         ret = single_open(file, &proc_setgroups_show, ns);
2687         if (ret)
2688                 goto err_put_ns;
2689
2690         return 0;
2691 err_put_ns:
2692         put_user_ns(ns);
2693 err:
2694         return ret;
2695 }
2696
2697 static int proc_setgroups_release(struct inode *inode, struct file *file)
2698 {
2699         struct seq_file *seq = file->private_data;
2700         struct user_namespace *ns = seq->private;
2701         int ret = single_release(inode, file);
2702         put_user_ns(ns);
2703         return ret;
2704 }
2705
2706 static const struct file_operations proc_setgroups_operations = {
2707         .open           = proc_setgroups_open,
2708         .write          = proc_setgroups_write,
2709         .read           = seq_read,
2710         .llseek         = seq_lseek,
2711         .release        = proc_setgroups_release,
2712 };
2713 #endif /* CONFIG_USER_NS */
2714
2715 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2716                                 struct pid *pid, struct task_struct *task)
2717 {
2718         int err = lock_trace(task);
2719         if (!err) {
2720                 seq_printf(m, "%08x\n", task->personality);
2721                 unlock_trace(task);
2722         }
2723         return err;
2724 }
2725
2726 /*
2727  * Thread groups
2728  */
2729 static const struct file_operations proc_task_operations;
2730 static const struct inode_operations proc_task_inode_operations;
2731
2732 static const struct pid_entry tgid_base_stuff[] = {
2733         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2734         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2735         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2736         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2737         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2738 #ifdef CONFIG_NET
2739         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2740 #endif
2741         REG("environ",    S_IRUSR, proc_environ_operations),
2742         ONE("auxv",       S_IRUSR, proc_pid_auxv),
2743         ONE("status",     S_IRUGO, proc_pid_status),
2744         ONE("personality", S_IRUSR, proc_pid_personality),
2745         ONE("limits",     S_IRUGO, proc_pid_limits),
2746 #ifdef CONFIG_SCHED_DEBUG
2747         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2748 #endif
2749 #ifdef CONFIG_SCHED_AUTOGROUP
2750         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2751 #endif
2752         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2753 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2754         ONE("syscall",    S_IRUSR, proc_pid_syscall),
2755 #endif
2756         REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
2757         ONE("stat",       S_IRUGO, proc_tgid_stat),
2758         ONE("statm",      S_IRUGO, proc_pid_statm),
2759         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2760 #ifdef CONFIG_NUMA
2761         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2762 #endif
2763         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2764         LNK("cwd",        proc_cwd_link),
2765         LNK("root",       proc_root_link),
2766         LNK("exe",        proc_exe_link),
2767         REG("mounts",     S_IRUGO, proc_mounts_operations),
2768         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2769         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2770 #ifdef CONFIG_PROC_PAGE_MONITOR
2771         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2772         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2773         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2774 #endif
2775 #ifdef CONFIG_SECURITY
2776         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2777 #endif
2778 #ifdef CONFIG_KALLSYMS
2779         ONE("wchan",      S_IRUGO, proc_pid_wchan),
2780 #endif
2781 #ifdef CONFIG_STACKTRACE
2782         ONE("stack",      S_IRUSR, proc_pid_stack),
2783 #endif
2784 #ifdef CONFIG_SCHED_INFO
2785         ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
2786 #endif
2787 #ifdef CONFIG_LATENCYTOP
2788         REG("latency",  S_IRUGO, proc_lstats_operations),
2789 #endif
2790 #ifdef CONFIG_PROC_PID_CPUSET
2791         ONE("cpuset",     S_IRUGO, proc_cpuset_show),
2792 #endif
2793 #ifdef CONFIG_CGROUPS
2794         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
2795 #endif
2796         ONE("oom_score",  S_IRUGO, proc_oom_score),
2797         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2798         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2799 #ifdef CONFIG_AUDITSYSCALL
2800         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2801         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2802 #endif
2803 #ifdef CONFIG_FAULT_INJECTION
2804         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2805 #endif
2806 #ifdef CONFIG_ELF_CORE
2807         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2808 #endif
2809 #ifdef CONFIG_TASK_IO_ACCOUNTING
2810         ONE("io",       S_IRUSR, proc_tgid_io_accounting),
2811 #endif
2812 #ifdef CONFIG_HARDWALL
2813         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
2814 #endif
2815 #ifdef CONFIG_USER_NS
2816         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2817         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2818         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2819         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
2820 #endif
2821 #ifdef CONFIG_CHECKPOINT_RESTORE
2822         REG("timers",     S_IRUGO, proc_timers_operations),
2823 #endif
2824 };
2825
2826 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2827 {
2828         return proc_pident_readdir(file, ctx,
2829                                    tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2830 }
2831
2832 static const struct file_operations proc_tgid_base_operations = {
2833         .read           = generic_read_dir,
2834         .iterate        = proc_tgid_base_readdir,
2835         .llseek         = default_llseek,
2836 };
2837
2838 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2839 {
2840         return proc_pident_lookup(dir, dentry,
2841                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2842 }
2843
2844 static const struct inode_operations proc_tgid_base_inode_operations = {
2845         .lookup         = proc_tgid_base_lookup,
2846         .getattr        = pid_getattr,
2847         .setattr        = proc_setattr,
2848         .permission     = proc_pid_permission,
2849 };
2850
2851 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2852 {
2853         struct dentry *dentry, *leader, *dir;
2854         char buf[PROC_NUMBUF];
2855         struct qstr name;
2856
2857         name.name = buf;
2858         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2859         /* no ->d_hash() rejects on procfs */
2860         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2861         if (dentry) {
2862                 d_invalidate(dentry);
2863                 dput(dentry);
2864         }
2865
2866         if (pid == tgid)
2867                 return;
2868
2869         name.name = buf;
2870         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2871         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2872         if (!leader)
2873                 goto out;
2874
2875         name.name = "task";
2876         name.len = strlen(name.name);
2877         dir = d_hash_and_lookup(leader, &name);
2878         if (!dir)
2879                 goto out_put_leader;
2880
2881         name.name = buf;
2882         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2883         dentry = d_hash_and_lookup(dir, &name);
2884         if (dentry) {
2885                 d_invalidate(dentry);
2886                 dput(dentry);
2887         }
2888
2889         dput(dir);
2890 out_put_leader:
2891         dput(leader);
2892 out:
2893         return;
2894 }
2895
2896 /**
2897  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2898  * @task: task that should be flushed.
2899  *
2900  * When flushing dentries from proc, one needs to flush them from global
2901  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2902  * in. This call is supposed to do all of this job.
2903  *
2904  * Looks in the dcache for
2905  * /proc/@pid
2906  * /proc/@tgid/task/@pid
2907  * if either directory is present flushes it and all of it'ts children
2908  * from the dcache.
2909  *
2910  * It is safe and reasonable to cache /proc entries for a task until
2911  * that task exits.  After that they just clog up the dcache with
2912  * useless entries, possibly causing useful dcache entries to be
2913  * flushed instead.  This routine is proved to flush those useless
2914  * dcache entries at process exit time.
2915  *
2916  * NOTE: This routine is just an optimization so it does not guarantee
2917  *       that no dcache entries will exist at process exit time it
2918  *       just makes it very unlikely that any will persist.
2919  */
2920
2921 void proc_flush_task(struct task_struct *task)
2922 {
2923         int i;
2924         struct pid *pid, *tgid;
2925         struct upid *upid;
2926
2927         pid = task_pid(task);
2928         tgid = task_tgid(task);
2929
2930         for (i = 0; i <= pid->level; i++) {
2931                 upid = &pid->numbers[i];
2932                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2933                                         tgid->numbers[i].nr);
2934         }
2935 }
2936
2937 static int proc_pid_instantiate(struct inode *dir,
2938                                    struct dentry * dentry,
2939                                    struct task_struct *task, const void *ptr)
2940 {
2941         struct inode *inode;
2942
2943         inode = proc_pid_make_inode(dir->i_sb, task);
2944         if (!inode)
2945                 goto out;
2946
2947         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2948         inode->i_op = &proc_tgid_base_inode_operations;
2949         inode->i_fop = &proc_tgid_base_operations;
2950         inode->i_flags|=S_IMMUTABLE;
2951
2952         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
2953                                                   ARRAY_SIZE(tgid_base_stuff)));
2954
2955         d_set_d_op(dentry, &pid_dentry_operations);
2956
2957         d_add(dentry, inode);
2958         /* Close the race of the process dying before we return the dentry */
2959         if (pid_revalidate(dentry, 0))
2960                 return 0;
2961 out:
2962         return -ENOENT;
2963 }
2964
2965 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
2966 {
2967         int result = -ENOENT;
2968         struct task_struct *task;
2969         unsigned tgid;
2970         struct pid_namespace *ns;
2971
2972         tgid = name_to_int(&dentry->d_name);
2973         if (tgid == ~0U)
2974                 goto out;
2975
2976         ns = dentry->d_sb->s_fs_info;
2977         rcu_read_lock();
2978         task = find_task_by_pid_ns(tgid, ns);
2979         if (task)
2980                 get_task_struct(task);
2981         rcu_read_unlock();
2982         if (!task)
2983                 goto out;
2984
2985         result = proc_pid_instantiate(dir, dentry, task, NULL);
2986         put_task_struct(task);
2987 out:
2988         return ERR_PTR(result);
2989 }
2990
2991 /*
2992  * Find the first task with tgid >= tgid
2993  *
2994  */
2995 struct tgid_iter {
2996         unsigned int tgid;
2997         struct task_struct *task;
2998 };
2999 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3000 {
3001         struct pid *pid;
3002
3003         if (iter.task)
3004                 put_task_struct(iter.task);
3005         rcu_read_lock();
3006 retry:
3007         iter.task = NULL;
3008         pid = find_ge_pid(iter.tgid, ns);
3009         if (pid) {
3010                 iter.tgid = pid_nr_ns(pid, ns);
3011                 iter.task = pid_task(pid, PIDTYPE_PID);
3012                 /* What we to know is if the pid we have find is the
3013                  * pid of a thread_group_leader.  Testing for task
3014                  * being a thread_group_leader is the obvious thing
3015                  * todo but there is a window when it fails, due to
3016                  * the pid transfer logic in de_thread.
3017                  *
3018                  * So we perform the straight forward test of seeing
3019                  * if the pid we have found is the pid of a thread
3020                  * group leader, and don't worry if the task we have
3021                  * found doesn't happen to be a thread group leader.
3022                  * As we don't care in the case of readdir.
3023                  */
3024                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3025                         iter.tgid += 1;
3026                         goto retry;
3027                 }
3028                 get_task_struct(iter.task);
3029         }
3030         rcu_read_unlock();
3031         return iter;
3032 }
3033
3034 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3035
3036 /* for the /proc/ directory itself, after non-process stuff has been done */
3037 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3038 {
3039         struct tgid_iter iter;
3040         struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
3041         loff_t pos = ctx->pos;
3042
3043         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3044                 return 0;
3045
3046         if (pos == TGID_OFFSET - 2) {
3047                 struct inode *inode = d_inode(ns->proc_self);
3048                 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3049                         return 0;
3050                 ctx->pos = pos = pos + 1;
3051         }
3052         if (pos == TGID_OFFSET - 1) {
3053                 struct inode *inode = d_inode(ns->proc_thread_self);
3054                 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3055                         return 0;
3056                 ctx->pos = pos = pos + 1;
3057         }
3058         iter.tgid = pos - TGID_OFFSET;
3059         iter.task = NULL;
3060         for (iter = next_tgid(ns, iter);
3061              iter.task;
3062              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3063                 char name[PROC_NUMBUF];
3064                 int len;
3065                 if (!has_pid_permissions(ns, iter.task, 2))
3066                         continue;
3067
3068                 len = snprintf(name, sizeof(name), "%d", iter.tgid);
3069                 ctx->pos = iter.tgid + TGID_OFFSET;
3070                 if (!proc_fill_cache(file, ctx, name, len,
3071                                      proc_pid_instantiate, iter.task, NULL)) {
3072                         put_task_struct(iter.task);
3073                         return 0;
3074                 }
3075         }
3076         ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3077         return 0;
3078 }
3079
3080 /*
3081  * Tasks
3082  */
3083 static const struct pid_entry tid_base_stuff[] = {
3084         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3085         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3086         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3087 #ifdef CONFIG_NET
3088         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3089 #endif
3090         REG("environ",   S_IRUSR, proc_environ_operations),
3091         ONE("auxv",      S_IRUSR, proc_pid_auxv),
3092         ONE("status",    S_IRUGO, proc_pid_status),
3093         ONE("personality", S_IRUSR, proc_pid_personality),
3094         ONE("limits",    S_IRUGO, proc_pid_limits),
3095 #ifdef CONFIG_SCHED_DEBUG
3096         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3097 #endif
3098         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3099 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3100         ONE("syscall",   S_IRUSR, proc_pid_syscall),
3101 #endif
3102         REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3103         ONE("stat",      S_IRUGO, proc_tid_stat),
3104         ONE("statm",     S_IRUGO, proc_pid_statm),
3105         REG("maps",      S_IRUGO, proc_tid_maps_operations),
3106 #ifdef CONFIG_PROC_CHILDREN
3107         REG("children",  S_IRUGO, proc_tid_children_operations),
3108 #endif
3109 #ifdef CONFIG_NUMA
3110         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3111 #endif
3112         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3113         LNK("cwd",       proc_cwd_link),
3114         LNK("root",      proc_root_link),
3115         LNK("exe",       proc_exe_link),
3116         REG("mounts",    S_IRUGO, proc_mounts_operations),
3117         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3118 #ifdef CONFIG_PROC_PAGE_MONITOR
3119         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3120         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3121         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3122 #endif
3123 #ifdef CONFIG_SECURITY
3124         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3125 #endif
3126 #ifdef CONFIG_KALLSYMS
3127         ONE("wchan",     S_IRUGO, proc_pid_wchan),
3128 #endif
3129 #ifdef CONFIG_STACKTRACE
3130         ONE("stack",      S_IRUSR, proc_pid_stack),
3131 #endif
3132 #ifdef CONFIG_SCHED_INFO
3133         ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3134 #endif
3135 #ifdef CONFIG_LATENCYTOP
3136         REG("latency",  S_IRUGO, proc_lstats_operations),
3137 #endif
3138 #ifdef CONFIG_PROC_PID_CPUSET
3139         ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3140 #endif
3141 #ifdef CONFIG_CGROUPS
3142         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3143 #endif
3144         ONE("oom_score", S_IRUGO, proc_oom_score),
3145         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3146         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3147 #ifdef CONFIG_AUDITSYSCALL
3148         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3149         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3150 #endif
3151 #ifdef CONFIG_FAULT_INJECTION
3152         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3153 #endif
3154 #ifdef CONFIG_TASK_IO_ACCOUNTING
3155         ONE("io",       S_IRUSR, proc_tid_io_accounting),
3156 #endif
3157 #ifdef CONFIG_HARDWALL
3158         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
3159 #endif
3160 #ifdef CONFIG_USER_NS
3161         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3162         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3163         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3164         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3165 #endif
3166 };
3167
3168 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3169 {
3170         return proc_pident_readdir(file, ctx,
3171                                    tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3172 }
3173
3174 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3175 {
3176         return proc_pident_lookup(dir, dentry,
3177                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3178 }
3179
3180 static const struct file_operations proc_tid_base_operations = {
3181         .read           = generic_read_dir,
3182         .iterate        = proc_tid_base_readdir,
3183         .llseek         = default_llseek,
3184 };
3185
3186 static const struct inode_operations proc_tid_base_inode_operations = {
3187         .lookup         = proc_tid_base_lookup,
3188         .getattr        = pid_getattr,
3189         .setattr        = proc_setattr,
3190 };
3191
3192 static int proc_task_instantiate(struct inode *dir,
3193         struct dentry *dentry, struct task_struct *task, const void *ptr)
3194 {
3195         struct inode *inode;
3196         inode = proc_pid_make_inode(dir->i_sb, task);
3197
3198         if (!inode)
3199                 goto out;
3200         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3201         inode->i_op = &proc_tid_base_inode_operations;
3202         inode->i_fop = &proc_tid_base_operations;
3203         inode->i_flags|=S_IMMUTABLE;
3204
3205         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3206                                                   ARRAY_SIZE(tid_base_stuff)));
3207
3208         d_set_d_op(dentry, &pid_dentry_operations);
3209
3210         d_add(dentry, inode);
3211         /* Close the race of the process dying before we return the dentry */
3212         if (pid_revalidate(dentry, 0))
3213                 return 0;
3214 out:
3215         return -ENOENT;
3216 }
3217
3218 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3219 {
3220         int result = -ENOENT;
3221         struct task_struct *task;
3222         struct task_struct *leader = get_proc_task(dir);
3223         unsigned tid;
3224         struct pid_namespace *ns;
3225
3226         if (!leader)
3227                 goto out_no_task;
3228
3229         tid = name_to_int(&dentry->d_name);
3230         if (tid == ~0U)
3231                 goto out;
3232
3233         ns = dentry->d_sb->s_fs_info;
3234         rcu_read_lock();
3235         task = find_task_by_pid_ns(tid, ns);
3236         if (task)
3237                 get_task_struct(task);
3238         rcu_read_unlock();
3239         if (!task)
3240                 goto out;
3241         if (!same_thread_group(leader, task))
3242                 goto out_drop_task;
3243
3244         result = proc_task_instantiate(dir, dentry, task, NULL);
3245 out_drop_task:
3246         put_task_struct(task);
3247 out:
3248         put_task_struct(leader);
3249 out_no_task:
3250         return ERR_PTR(result);
3251 }
3252
3253 /*
3254  * Find the first tid of a thread group to return to user space.
3255  *
3256  * Usually this is just the thread group leader, but if the users
3257  * buffer was too small or there was a seek into the middle of the
3258  * directory we have more work todo.
3259  *
3260  * In the case of a short read we start with find_task_by_pid.
3261  *
3262  * In the case of a seek we start with the leader and walk nr
3263  * threads past it.
3264  */
3265 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3266                                         struct pid_namespace *ns)
3267 {
3268         struct task_struct *pos, *task;
3269         unsigned long nr = f_pos;
3270
3271         if (nr != f_pos)        /* 32bit overflow? */
3272                 return NULL;
3273
3274         rcu_read_lock();
3275         task = pid_task(pid, PIDTYPE_PID);
3276         if (!task)
3277                 goto fail;
3278
3279         /* Attempt to start with the tid of a thread */
3280         if (tid && nr) {
3281                 pos = find_task_by_pid_ns(tid, ns);
3282                 if (pos && same_thread_group(pos, task))
3283                         goto found;
3284         }
3285
3286         /* If nr exceeds the number of threads there is nothing todo */
3287         if (nr >= get_nr_threads(task))
3288                 goto fail;
3289
3290         /* If we haven't found our starting place yet start
3291          * with the leader and walk nr threads forward.
3292          */
3293         pos = task = task->group_leader;
3294         do {
3295                 if (!nr--)
3296                         goto found;
3297         } while_each_thread(task, pos);
3298 fail:
3299         pos = NULL;
3300         goto out;
3301 found:
3302         get_task_struct(pos);
3303 out:
3304         rcu_read_unlock();
3305         return pos;
3306 }
3307
3308 /*
3309  * Find the next thread in the thread list.
3310  * Return NULL if there is an error or no next thread.
3311  *
3312  * The reference to the input task_struct is released.
3313  */
3314 static struct task_struct *next_tid(struct task_struct *start)
3315 {
3316         struct task_struct *pos = NULL;
3317         rcu_read_lock();
3318         if (pid_alive(start)) {
3319                 pos = next_thread(start);
3320                 if (thread_group_leader(pos))
3321                         pos = NULL;
3322                 else
3323                         get_task_struct(pos);
3324         }
3325         rcu_read_unlock();
3326         put_task_struct(start);
3327         return pos;
3328 }
3329
3330 /* for the /proc/TGID/task/ directories */
3331 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3332 {
3333         struct inode *inode = file_inode(file);
3334         struct task_struct *task;
3335         struct pid_namespace *ns;
3336         int tid;
3337
3338         if (proc_inode_is_dead(inode))
3339                 return -ENOENT;
3340
3341         if (!dir_emit_dots(file, ctx))
3342                 return 0;
3343
3344         /* f_version caches the tgid value that the last readdir call couldn't
3345          * return. lseek aka telldir automagically resets f_version to 0.
3346          */
3347         ns = inode->i_sb->s_fs_info;
3348         tid = (int)file->f_version;
3349         file->f_version = 0;
3350         for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3351              task;
3352              task = next_tid(task), ctx->pos++) {
3353                 char name[PROC_NUMBUF];
3354                 int len;
3355                 tid = task_pid_nr_ns(task, ns);
3356                 len = snprintf(name, sizeof(name), "%d", tid);
3357                 if (!proc_fill_cache(file, ctx, name, len,
3358                                 proc_task_instantiate, task, NULL)) {
3359                         /* returning this tgid failed, save it as the first
3360                          * pid for the next readir call */
3361                         file->f_version = (u64)tid;
3362                         put_task_struct(task);
3363                         break;
3364                 }
3365         }
3366
3367         return 0;
3368 }
3369
3370 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3371 {
3372         struct inode *inode = d_inode(dentry);
3373         struct task_struct *p = get_proc_task(inode);
3374         generic_fillattr(inode, stat);
3375
3376         if (p) {
3377                 stat->nlink += get_nr_threads(p);
3378                 put_task_struct(p);
3379         }
3380
3381         return 0;
3382 }
3383
3384 static const struct inode_operations proc_task_inode_operations = {
3385         .lookup         = proc_task_lookup,
3386         .getattr        = proc_task_getattr,
3387         .setattr        = proc_setattr,
3388         .permission     = proc_pid_permission,
3389 };
3390
3391 static const struct file_operations proc_task_operations = {
3392         .read           = generic_read_dir,
3393         .iterate        = proc_task_readdir,
3394         .llseek         = default_llseek,
3395 };