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_puts(m, "0\n");
 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         unsigned long env_start, env_end;
 957
 958         if (!mm)
 959                 return 0;
 960
 961         page = (char *)__get_free_page(GFP_TEMPORARY);
 962         if (!page)
 963                 return -ENOMEM;
 964
 965         ret = 0;
 966         if (!atomic_inc_not_zero(&mm->mm_users))
 967                 goto free;
 968
 969         down_read(&mm->mmap_sem);
 970         env_start = mm->env_start;
 971         env_end = mm->env_end;
 972         up_read(&mm->mmap_sem);
 973
 974         while (count > 0) {
 975                 size_t this_len, max_len;
 976                 int retval;
 977
 978                 if (src >= (env_end - env_start))
 979                         break;
 980
 981                 this_len = env_end - (env_start + src);
 982
 983                 max_len = min_t(size_t, PAGE_SIZE, count);
 984                 this_len = min(max_len, this_len);
 985
 986                 retval = access_remote_vm(mm, (env_start + src),
 987                         page, this_len, 0);
 988
 989                 if (retval <= 0) {
 990                         ret = retval;
 991                         break;
 992                 }
 993
 994                 if (copy_to_user(buf, page, retval)) {
 995                         ret = -EFAULT;
 996                         break;
 997                 }
 998
 999                 ret += retval;
1000                 src += retval;
1001                 buf += retval;
1002                 count -= retval;
1003         }
1004         *ppos = src;
1005         mmput(mm);
1006
1007 free:
1008         free_page((unsigned long) page);
1009         return ret;
1010 }
1011
1012 static const struct file_operations proc_environ_operations = {
1013         .open           = environ_open,
1014         .read           = environ_read,
1015         .llseek         = generic_file_llseek,
1016         .release        = mem_release,
1017 };
1018
1019 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1020                             loff_t *ppos)
1021 {
1022         struct task_struct *task = get_proc_task(file_inode(file));
1023         char buffer[PROC_NUMBUF];
1024         int oom_adj = OOM_ADJUST_MIN;
1025         size_t len;
1026         unsigned long flags;
1027
1028         if (!task)
1029                 return -ESRCH;
1030         if (lock_task_sighand(task, &flags)) {
1031                 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1032                         oom_adj = OOM_ADJUST_MAX;
1033                 else
1034                         oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1035                                   OOM_SCORE_ADJ_MAX;
1036                 unlock_task_sighand(task, &flags);
1037         }
1038         put_task_struct(task);
1039         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1040         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1041 }
1042
1043 /*
1044  * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1045  * kernels.  The effective policy is defined by oom_score_adj, which has a
1046  * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1047  * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1048  * Processes that become oom disabled via oom_adj will still be oom disabled
1049  * with this implementation.
1050  *
1051  * oom_adj cannot be removed since existing userspace binaries use it.
1052  */
1053 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1054                              size_t count, loff_t *ppos)
1055 {
1056         struct task_struct *task;
1057         char buffer[PROC_NUMBUF];
1058         int oom_adj;
1059         unsigned long flags;
1060         int err;
1061
1062         memset(buffer, 0, sizeof(buffer));
1063         if (count > sizeof(buffer) - 1)
1064                 count = sizeof(buffer) - 1;
1065         if (copy_from_user(buffer, buf, count)) {
1066                 err = -EFAULT;
1067                 goto out;
1068         }
1069
1070         err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1071         if (err)
1072                 goto out;
1073         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1074              oom_adj != OOM_DISABLE) {
1075                 err = -EINVAL;
1076                 goto out;
1077         }
1078
1079         task = get_proc_task(file_inode(file));
1080         if (!task) {
1081                 err = -ESRCH;
1082                 goto out;
1083         }
1084
1085         task_lock(task);
1086         if (!task->mm) {
1087                 err = -EINVAL;
1088                 goto err_task_lock;
1089         }
1090
1091         if (!lock_task_sighand(task, &flags)) {
1092                 err = -ESRCH;
1093                 goto err_task_lock;
1094         }
1095
1096         /*
1097          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1098          * value is always attainable.
1099          */
1100         if (oom_adj == OOM_ADJUST_MAX)
1101                 oom_adj = OOM_SCORE_ADJ_MAX;
1102         else
1103                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1104
1105         if (oom_adj < task->signal->oom_score_adj &&
1106             !capable(CAP_SYS_RESOURCE)) {
1107                 err = -EACCES;
1108                 goto err_sighand;
1109         }
1110
1111         /*
1112          * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1113          * /proc/pid/oom_score_adj instead.
1114          */
1115         pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1116                   current->comm, task_pid_nr(current), task_pid_nr(task),
1117                   task_pid_nr(task));
1118
1119         task->signal->oom_score_adj = oom_adj;
1120         trace_oom_score_adj_update(task);
1121 err_sighand:
1122         unlock_task_sighand(task, &flags);
1123 err_task_lock:
1124         task_unlock(task);
1125         put_task_struct(task);
1126 out:
1127         return err < 0 ? err : count;
1128 }
1129
1130 static const struct file_operations proc_oom_adj_operations = {
1131         .read           = oom_adj_read,
1132         .write          = oom_adj_write,
1133         .llseek         = generic_file_llseek,
1134 };
1135
1136 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1137                                         size_t count, loff_t *ppos)
1138 {
1139         struct task_struct *task = get_proc_task(file_inode(file));
1140         char buffer[PROC_NUMBUF];
1141         short oom_score_adj = OOM_SCORE_ADJ_MIN;
1142         unsigned long flags;
1143         size_t len;
1144
1145         if (!task)
1146                 return -ESRCH;
1147         if (lock_task_sighand(task, &flags)) {
1148                 oom_score_adj = task->signal->oom_score_adj;
1149                 unlock_task_sighand(task, &flags);
1150         }
1151         put_task_struct(task);
1152         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1153         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1154 }
1155
1156 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1157                                         size_t count, loff_t *ppos)
1158 {
1159         struct task_struct *task;
1160         char buffer[PROC_NUMBUF];
1161         unsigned long flags;
1162         int oom_score_adj;
1163         int err;
1164
1165         memset(buffer, 0, sizeof(buffer));
1166         if (count > sizeof(buffer) - 1)
1167                 count = sizeof(buffer) - 1;
1168         if (copy_from_user(buffer, buf, count)) {
1169                 err = -EFAULT;
1170                 goto out;
1171         }
1172
1173         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1174         if (err)
1175                 goto out;
1176         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1177                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1178                 err = -EINVAL;
1179                 goto out;
1180         }
1181
1182         task = get_proc_task(file_inode(file));
1183         if (!task) {
1184                 err = -ESRCH;
1185                 goto out;
1186         }
1187
1188         task_lock(task);
1189         if (!task->mm) {
1190                 err = -EINVAL;
1191                 goto err_task_lock;
1192         }
1193
1194         if (!lock_task_sighand(task, &flags)) {
1195                 err = -ESRCH;
1196                 goto err_task_lock;
1197         }
1198
1199         if ((short)oom_score_adj < task->signal->oom_score_adj_min &&
1200                         !capable(CAP_SYS_RESOURCE)) {
1201                 err = -EACCES;
1202                 goto err_sighand;
1203         }
1204
1205         task->signal->oom_score_adj = (short)oom_score_adj;
1206         if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1207                 task->signal->oom_score_adj_min = (short)oom_score_adj;
1208         trace_oom_score_adj_update(task);
1209
1210 err_sighand:
1211         unlock_task_sighand(task, &flags);
1212 err_task_lock:
1213         task_unlock(task);
1214         put_task_struct(task);
1215 out:
1216         return err < 0 ? err : count;
1217 }
1218
1219 static const struct file_operations proc_oom_score_adj_operations = {
1220         .read           = oom_score_adj_read,
1221         .write          = oom_score_adj_write,
1222         .llseek         = default_llseek,
1223 };
1224
1225 #ifdef CONFIG_AUDITSYSCALL
1226 #define TMPBUFLEN 21
1227 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1228                                   size_t count, loff_t *ppos)
1229 {
1230         struct inode * inode = file_inode(file);
1231         struct task_struct *task = get_proc_task(inode);
1232         ssize_t length;
1233         char tmpbuf[TMPBUFLEN];
1234
1235         if (!task)
1236                 return -ESRCH;
1237         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1238                            from_kuid(file->f_cred->user_ns,
1239                                      audit_get_loginuid(task)));
1240         put_task_struct(task);
1241         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1242 }
1243
1244 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1245                                    size_t count, loff_t *ppos)
1246 {
1247         struct inode * inode = file_inode(file);
1248         uid_t loginuid;
1249         kuid_t kloginuid;
1250         int rv;
1251
1252         rcu_read_lock();
1253         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1254                 rcu_read_unlock();
1255                 return -EPERM;
1256         }
1257         rcu_read_unlock();
1258
1259         if (*ppos != 0) {
1260                 /* No partial writes. */
1261                 return -EINVAL;
1262         }
1263
1264         rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1265         if (rv < 0)
1266                 return rv;
1267
1268         /* is userspace tring to explicitly UNSET the loginuid? */
1269         if (loginuid == AUDIT_UID_UNSET) {
1270                 kloginuid = INVALID_UID;
1271         } else {
1272                 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1273                 if (!uid_valid(kloginuid))
1274                         return -EINVAL;
1275         }
1276
1277         rv = audit_set_loginuid(kloginuid);
1278         if (rv < 0)
1279                 return rv;
1280         return count;
1281 }
1282
1283 static const struct file_operations proc_loginuid_operations = {
1284         .read           = proc_loginuid_read,
1285         .write          = proc_loginuid_write,
1286         .llseek         = generic_file_llseek,
1287 };
1288
1289 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1290                                   size_t count, loff_t *ppos)
1291 {
1292         struct inode * inode = file_inode(file);
1293         struct task_struct *task = get_proc_task(inode);
1294         ssize_t length;
1295         char tmpbuf[TMPBUFLEN];
1296
1297         if (!task)
1298                 return -ESRCH;
1299         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1300                                 audit_get_sessionid(task));
1301         put_task_struct(task);
1302         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1303 }
1304
1305 static const struct file_operations proc_sessionid_operations = {
1306         .read           = proc_sessionid_read,
1307         .llseek         = generic_file_llseek,
1308 };
1309 #endif
1310
1311 #ifdef CONFIG_FAULT_INJECTION
1312 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1313                                       size_t count, loff_t *ppos)
1314 {
1315         struct task_struct *task = get_proc_task(file_inode(file));
1316         char buffer[PROC_NUMBUF];
1317         size_t len;
1318         int make_it_fail;
1319
1320         if (!task)
1321                 return -ESRCH;
1322         make_it_fail = task->make_it_fail;
1323         put_task_struct(task);
1324
1325         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1326
1327         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1328 }
1329
1330 static ssize_t proc_fault_inject_write(struct file * file,
1331                         const char __user * buf, size_t count, loff_t *ppos)
1332 {
1333         struct task_struct *task;
1334         char buffer[PROC_NUMBUF];
1335         int make_it_fail;
1336         int rv;
1337
1338         if (!capable(CAP_SYS_RESOURCE))
1339                 return -EPERM;
1340         memset(buffer, 0, sizeof(buffer));
1341         if (count > sizeof(buffer) - 1)
1342                 count = sizeof(buffer) - 1;
1343         if (copy_from_user(buffer, buf, count))
1344                 return -EFAULT;
1345         rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1346         if (rv < 0)
1347                 return rv;
1348         if (make_it_fail < 0 || make_it_fail > 1)
1349                 return -EINVAL;
1350
1351         task = get_proc_task(file_inode(file));
1352         if (!task)
1353                 return -ESRCH;
1354         task->make_it_fail = make_it_fail;
1355         put_task_struct(task);
1356
1357         return count;
1358 }
1359
1360 static const struct file_operations proc_fault_inject_operations = {
1361         .read           = proc_fault_inject_read,
1362         .write          = proc_fault_inject_write,
1363         .llseek         = generic_file_llseek,
1364 };
1365 #endif
1366
1367
1368 #ifdef CONFIG_SCHED_DEBUG
1369 /*
1370  * Print out various scheduling related per-task fields:
1371  */
1372 static int sched_show(struct seq_file *m, void *v)
1373 {
1374         struct inode *inode = m->private;
1375         struct task_struct *p;
1376
1377         p = get_proc_task(inode);
1378         if (!p)
1379                 return -ESRCH;
1380         proc_sched_show_task(p, m);
1381
1382         put_task_struct(p);
1383
1384         return 0;
1385 }
1386
1387 static ssize_t
1388 sched_write(struct file *file, const char __user *buf,
1389             size_t count, loff_t *offset)
1390 {
1391         struct inode *inode = file_inode(file);
1392         struct task_struct *p;
1393
1394         p = get_proc_task(inode);
1395         if (!p)
1396                 return -ESRCH;
1397         proc_sched_set_task(p);
1398
1399         put_task_struct(p);
1400
1401         return count;
1402 }
1403
1404 static int sched_open(struct inode *inode, struct file *filp)
1405 {
1406         return single_open(filp, sched_show, inode);
1407 }
1408
1409 static const struct file_operations proc_pid_sched_operations = {
1410         .open           = sched_open,
1411         .read           = seq_read,
1412         .write          = sched_write,
1413         .llseek         = seq_lseek,
1414         .release        = single_release,
1415 };
1416
1417 #endif
1418
1419 #ifdef CONFIG_SCHED_AUTOGROUP
1420 /*
1421  * Print out autogroup related information:
1422  */
1423 static int sched_autogroup_show(struct seq_file *m, void *v)
1424 {
1425         struct inode *inode = m->private;
1426         struct task_struct *p;
1427
1428         p = get_proc_task(inode);
1429         if (!p)
1430                 return -ESRCH;
1431         proc_sched_autogroup_show_task(p, m);
1432
1433         put_task_struct(p);
1434
1435         return 0;
1436 }
1437
1438 static ssize_t
1439 sched_autogroup_write(struct file *file, const char __user *buf,
1440             size_t count, loff_t *offset)
1441 {
1442         struct inode *inode = file_inode(file);
1443         struct task_struct *p;
1444         char buffer[PROC_NUMBUF];
1445         int nice;
1446         int err;
1447
1448         memset(buffer, 0, sizeof(buffer));
1449         if (count > sizeof(buffer) - 1)
1450                 count = sizeof(buffer) - 1;
1451         if (copy_from_user(buffer, buf, count))
1452                 return -EFAULT;
1453
1454         err = kstrtoint(strstrip(buffer), 0, &nice);
1455         if (err < 0)
1456                 return err;
1457
1458         p = get_proc_task(inode);
1459         if (!p)
1460                 return -ESRCH;
1461
1462         err = proc_sched_autogroup_set_nice(p, nice);
1463         if (err)
1464                 count = err;
1465
1466         put_task_struct(p);
1467
1468         return count;
1469 }
1470
1471 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1472 {
1473         int ret;
1474
1475         ret = single_open(filp, sched_autogroup_show, NULL);
1476         if (!ret) {
1477                 struct seq_file *m = filp->private_data;
1478
1479                 m->private = inode;
1480         }
1481         return ret;
1482 }
1483
1484 static const struct file_operations proc_pid_sched_autogroup_operations = {
1485         .open           = sched_autogroup_open,
1486         .read           = seq_read,
1487         .write          = sched_autogroup_write,
1488         .llseek         = seq_lseek,
1489         .release        = single_release,
1490 };
1491
1492 #endif /* CONFIG_SCHED_AUTOGROUP */
1493
1494 static ssize_t comm_write(struct file *file, const char __user *buf,
1495                                 size_t count, loff_t *offset)
1496 {
1497         struct inode *inode = file_inode(file);
1498         struct task_struct *p;
1499         char buffer[TASK_COMM_LEN];
1500         const size_t maxlen = sizeof(buffer) - 1;
1501
1502         memset(buffer, 0, sizeof(buffer));
1503         if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1504                 return -EFAULT;
1505
1506         p = get_proc_task(inode);
1507         if (!p)
1508                 return -ESRCH;
1509
1510         if (same_thread_group(current, p))
1511                 set_task_comm(p, buffer);
1512         else
1513                 count = -EINVAL;
1514
1515         put_task_struct(p);
1516
1517         return count;
1518 }
1519
1520 static int comm_show(struct seq_file *m, void *v)
1521 {
1522         struct inode *inode = m->private;
1523         struct task_struct *p;
1524
1525         p = get_proc_task(inode);
1526         if (!p)
1527                 return -ESRCH;
1528
1529         task_lock(p);
1530         seq_printf(m, "%s\n", p->comm);
1531         task_unlock(p);
1532
1533         put_task_struct(p);
1534
1535         return 0;
1536 }
1537
1538 static int comm_open(struct inode *inode, struct file *filp)
1539 {
1540         return single_open(filp, comm_show, inode);
1541 }
1542
1543 static const struct file_operations proc_pid_set_comm_operations = {
1544         .open           = comm_open,
1545         .read           = seq_read,
1546         .write          = comm_write,
1547         .llseek         = seq_lseek,
1548         .release        = single_release,
1549 };
1550
1551 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1552 {
1553         struct task_struct *task;
1554         struct mm_struct *mm;
1555         struct file *exe_file;
1556
1557         task = get_proc_task(d_inode(dentry));
1558         if (!task)
1559                 return -ENOENT;
1560         mm = get_task_mm(task);
1561         put_task_struct(task);
1562         if (!mm)
1563                 return -ENOENT;
1564         exe_file = get_mm_exe_file(mm);
1565         mmput(mm);
1566         if (exe_file) {
1567                 *exe_path = exe_file->f_path;
1568                 path_get(&exe_file->f_path);
1569                 fput(exe_file);
1570                 return 0;
1571         } else
1572                 return -ENOENT;
1573 }
1574
1575 static const char *proc_pid_get_link(struct dentry *dentry,
1576                                      struct inode *inode,
1577                                      struct delayed_call *done)
1578 {
1579         struct path path;
1580         int error = -EACCES;
1581
1582         if (!dentry)
1583                 return ERR_PTR(-ECHILD);
1584
1585         /* Are we allowed to snoop on the tasks file descriptors? */
1586         if (!proc_fd_access_allowed(inode))
1587                 goto out;
1588
1589         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1590         if (error)
1591                 goto out;
1592
1593         nd_jump_link(&path);
1594         return NULL;
1595 out:
1596         return ERR_PTR(error);
1597 }
1598
1599 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1600 {
1601         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1602         char *pathname;
1603         int len;
1604
1605         if (!tmp)
1606                 return -ENOMEM;
1607
1608         pathname = d_path(path, tmp, PAGE_SIZE);
1609         len = PTR_ERR(pathname);
1610         if (IS_ERR(pathname))
1611                 goto out;
1612         len = tmp + PAGE_SIZE - 1 - pathname;
1613
1614         if (len > buflen)
1615                 len = buflen;
1616         if (copy_to_user(buffer, pathname, len))
1617                 len = -EFAULT;
1618  out:
1619         free_page((unsigned long)tmp);
1620         return len;
1621 }
1622
1623 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1624 {
1625         int error = -EACCES;
1626         struct inode *inode = d_inode(dentry);
1627         struct path path;
1628
1629         /* Are we allowed to snoop on the tasks file descriptors? */
1630         if (!proc_fd_access_allowed(inode))
1631                 goto out;
1632
1633         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1634         if (error)
1635                 goto out;
1636
1637         error = do_proc_readlink(&path, buffer, buflen);
1638         path_put(&path);
1639 out:
1640         return error;
1641 }
1642
1643 const struct inode_operations proc_pid_link_inode_operations = {
1644         .readlink       = proc_pid_readlink,
1645         .get_link       = proc_pid_get_link,
1646         .setattr        = proc_setattr,
1647 };
1648
1649
1650 /* building an inode */
1651
1652 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1653 {
1654         struct inode * inode;
1655         struct proc_inode *ei;
1656         const struct cred *cred;
1657
1658         /* We need a new inode */
1659
1660         inode = new_inode(sb);
1661         if (!inode)
1662                 goto out;
1663
1664         /* Common stuff */
1665         ei = PROC_I(inode);
1666         inode->i_ino = get_next_ino();
1667         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1668         inode->i_op = &proc_def_inode_operations;
1669
1670         /*
1671          * grab the reference to task.
1672          */
1673         ei->pid = get_task_pid(task, PIDTYPE_PID);
1674         if (!ei->pid)
1675                 goto out_unlock;
1676
1677         if (task_dumpable(task)) {
1678                 rcu_read_lock();
1679                 cred = __task_cred(task);
1680                 inode->i_uid = cred->euid;
1681                 inode->i_gid = cred->egid;
1682                 rcu_read_unlock();
1683         }
1684         security_task_to_inode(task, inode);
1685
1686 out:
1687         return inode;
1688
1689 out_unlock:
1690         iput(inode);
1691         return NULL;
1692 }
1693
1694 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1695 {
1696         struct inode *inode = d_inode(dentry);
1697         struct task_struct *task;
1698         const struct cred *cred;
1699         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1700
1701         generic_fillattr(inode, stat);
1702
1703         rcu_read_lock();
1704         stat->uid = GLOBAL_ROOT_UID;
1705         stat->gid = GLOBAL_ROOT_GID;
1706         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1707         if (task) {
1708                 if (!has_pid_permissions(pid, task, 2)) {
1709                         rcu_read_unlock();
1710                         /*
1711                          * This doesn't prevent learning whether PID exists,
1712                          * it only makes getattr() consistent with readdir().
1713                          */
1714                         return -ENOENT;
1715                 }
1716                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1717                     task_dumpable(task)) {
1718                         cred = __task_cred(task);
1719                         stat->uid = cred->euid;
1720                         stat->gid = cred->egid;
1721                 }
1722         }
1723         rcu_read_unlock();
1724         return 0;
1725 }
1726
1727 /* dentry stuff */
1728
1729 /*
1730  *      Exceptional case: normally we are not allowed to unhash a busy
1731  * directory. In this case, however, we can do it - no aliasing problems
1732  * due to the way we treat inodes.
1733  *
1734  * Rewrite the inode's ownerships here because the owning task may have
1735  * performed a setuid(), etc.
1736  *
1737  * Before the /proc/pid/status file was created the only way to read
1738  * the effective uid of a /process was to stat /proc/pid.  Reading
1739  * /proc/pid/status is slow enough that procps and other packages
1740  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1741  * made this apply to all per process world readable and executable
1742  * directories.
1743  */
1744 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1745 {
1746         struct inode *inode;
1747         struct task_struct *task;
1748         const struct cred *cred;
1749
1750         if (flags & LOOKUP_RCU)
1751                 return -ECHILD;
1752
1753         inode = d_inode(dentry);
1754         task = get_proc_task(inode);
1755
1756         if (task) {
1757                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1758                     task_dumpable(task)) {
1759                         rcu_read_lock();
1760                         cred = __task_cred(task);
1761                         inode->i_uid = cred->euid;
1762                         inode->i_gid = cred->egid;
1763                         rcu_read_unlock();
1764                 } else {
1765                         inode->i_uid = GLOBAL_ROOT_UID;
1766                         inode->i_gid = GLOBAL_ROOT_GID;
1767                 }
1768                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1769                 security_task_to_inode(task, inode);
1770                 put_task_struct(task);
1771                 return 1;
1772         }
1773         return 0;
1774 }
1775
1776 static inline bool proc_inode_is_dead(struct inode *inode)
1777 {
1778         return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1779 }
1780
1781 int pid_delete_dentry(const struct dentry *dentry)
1782 {
1783         /* Is the task we represent dead?
1784          * If so, then don't put the dentry on the lru list,
1785          * kill it immediately.
1786          */
1787         return proc_inode_is_dead(d_inode(dentry));
1788 }
1789
1790 const struct dentry_operations pid_dentry_operations =
1791 {
1792         .d_revalidate   = pid_revalidate,
1793         .d_delete       = pid_delete_dentry,
1794 };
1795
1796 /* Lookups */
1797
1798 /*
1799  * Fill a directory entry.
1800  *
1801  * If possible create the dcache entry and derive our inode number and
1802  * file type from dcache entry.
1803  *
1804  * Since all of the proc inode numbers are dynamically generated, the inode
1805  * numbers do not exist until the inode is cache.  This means creating the
1806  * the dcache entry in readdir is necessary to keep the inode numbers
1807  * reported by readdir in sync with the inode numbers reported
1808  * by stat.
1809  */
1810 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1811         const char *name, int len,
1812         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1813 {
1814         struct dentry *child, *dir = file->f_path.dentry;
1815         struct qstr qname = QSTR_INIT(name, len);
1816         struct inode *inode;
1817         unsigned type;
1818         ino_t ino;
1819
1820         child = d_hash_and_lookup(dir, &qname);
1821         if (!child) {
1822                 child = d_alloc(dir, &qname);
1823                 if (!child)
1824                         goto end_instantiate;
1825                 if (instantiate(d_inode(dir), child, task, ptr) < 0) {
1826                         dput(child);
1827                         goto end_instantiate;
1828                 }
1829         }
1830         inode = d_inode(child);
1831         ino = inode->i_ino;
1832         type = inode->i_mode >> 12;
1833         dput(child);
1834         return dir_emit(ctx, name, len, ino, type);
1835
1836 end_instantiate:
1837         return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1838 }
1839
1840 /*
1841  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1842  * which represent vma start and end addresses.
1843  */
1844 static int dname_to_vma_addr(struct dentry *dentry,
1845                              unsigned long *start, unsigned long *end)
1846 {
1847         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1848                 return -EINVAL;
1849
1850         return 0;
1851 }
1852
1853 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1854 {
1855         unsigned long vm_start, vm_end;
1856         bool exact_vma_exists = false;
1857         struct mm_struct *mm = NULL;
1858         struct task_struct *task;
1859         const struct cred *cred;
1860         struct inode *inode;
1861         int status = 0;
1862
1863         if (flags & LOOKUP_RCU)
1864                 return -ECHILD;
1865
1866         inode = d_inode(dentry);
1867         task = get_proc_task(inode);
1868         if (!task)
1869                 goto out_notask;
1870
1871         mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1872         if (IS_ERR_OR_NULL(mm))
1873                 goto out;
1874
1875         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1876                 down_read(&mm->mmap_sem);
1877                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1878                 up_read(&mm->mmap_sem);
1879         }
1880
1881         mmput(mm);
1882
1883         if (exact_vma_exists) {
1884                 if (task_dumpable(task)) {
1885                         rcu_read_lock();
1886                         cred = __task_cred(task);
1887                         inode->i_uid = cred->euid;
1888                         inode->i_gid = cred->egid;
1889                         rcu_read_unlock();
1890                 } else {
1891                         inode->i_uid = GLOBAL_ROOT_UID;
1892                         inode->i_gid = GLOBAL_ROOT_GID;
1893                 }
1894                 security_task_to_inode(task, inode);
1895                 status = 1;
1896         }
1897
1898 out:
1899         put_task_struct(task);
1900
1901 out_notask:
1902         return status;
1903 }
1904
1905 static const struct dentry_operations tid_map_files_dentry_operations = {
1906         .d_revalidate   = map_files_d_revalidate,
1907         .d_delete       = pid_delete_dentry,
1908 };
1909
1910 static int map_files_get_link(struct dentry *dentry, struct path *path)
1911 {
1912         unsigned long vm_start, vm_end;
1913         struct vm_area_struct *vma;
1914         struct task_struct *task;
1915         struct mm_struct *mm;
1916         int rc;
1917
1918         rc = -ENOENT;
1919         task = get_proc_task(d_inode(dentry));
1920         if (!task)
1921                 goto out;
1922
1923         mm = get_task_mm(task);
1924         put_task_struct(task);
1925         if (!mm)
1926                 goto out;
1927
1928         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1929         if (rc)
1930                 goto out_mmput;
1931
1932         rc = -ENOENT;
1933         down_read(&mm->mmap_sem);
1934         vma = find_exact_vma(mm, vm_start, vm_end);
1935         if (vma && vma->vm_file) {
1936                 *path = vma->vm_file->f_path;
1937                 path_get(path);
1938                 rc = 0;
1939         }
1940         up_read(&mm->mmap_sem);
1941
1942 out_mmput:
1943         mmput(mm);
1944 out:
1945         return rc;
1946 }
1947
1948 struct map_files_info {
1949         fmode_t         mode;
1950         unsigned long   len;
1951         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1952 };
1953
1954 /*
1955  * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
1956  * symlinks may be used to bypass permissions on ancestor directories in the
1957  * path to the file in question.
1958  */
1959 static const char *
1960 proc_map_files_get_link(struct dentry *dentry,
1961                         struct inode *inode,
1962                         struct delayed_call *done)
1963 {
1964         if (!capable(CAP_SYS_ADMIN))
1965                 return ERR_PTR(-EPERM);
1966
1967         return proc_pid_get_link(dentry, inode, done);
1968 }
1969
1970 /*
1971  * Identical to proc_pid_link_inode_operations except for get_link()
1972  */
1973 static const struct inode_operations proc_map_files_link_inode_operations = {
1974         .readlink       = proc_pid_readlink,
1975         .get_link       = proc_map_files_get_link,
1976         .setattr        = proc_setattr,
1977 };
1978
1979 static int
1980 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
1981                            struct task_struct *task, const void *ptr)
1982 {
1983         fmode_t mode = (fmode_t)(unsigned long)ptr;
1984         struct proc_inode *ei;
1985         struct inode *inode;
1986
1987         inode = proc_pid_make_inode(dir->i_sb, task);
1988         if (!inode)
1989                 return -ENOENT;
1990
1991         ei = PROC_I(inode);
1992         ei->op.proc_get_link = map_files_get_link;
1993
1994         inode->i_op = &proc_map_files_link_inode_operations;
1995         inode->i_size = 64;
1996         inode->i_mode = S_IFLNK;
1997
1998         if (mode & FMODE_READ)
1999                 inode->i_mode |= S_IRUSR;
2000         if (mode & FMODE_WRITE)
2001                 inode->i_mode |= S_IWUSR;
2002
2003         d_set_d_op(dentry, &tid_map_files_dentry_operations);
2004         d_add(dentry, inode);
2005
2006         return 0;
2007 }
2008
2009 static struct dentry *proc_map_files_lookup(struct inode *dir,
2010                 struct dentry *dentry, unsigned int flags)
2011 {
2012         unsigned long vm_start, vm_end;
2013         struct vm_area_struct *vma;
2014         struct task_struct *task;
2015         int result;
2016         struct mm_struct *mm;
2017
2018         result = -ENOENT;
2019         task = get_proc_task(dir);
2020         if (!task)
2021                 goto out;
2022
2023         result = -EACCES;
2024         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2025                 goto out_put_task;
2026
2027         result = -ENOENT;
2028         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2029                 goto out_put_task;
2030
2031         mm = get_task_mm(task);
2032         if (!mm)
2033                 goto out_put_task;
2034
2035         down_read(&mm->mmap_sem);
2036         vma = find_exact_vma(mm, vm_start, vm_end);
2037         if (!vma)
2038                 goto out_no_vma;
2039
2040         if (vma->vm_file)
2041                 result = proc_map_files_instantiate(dir, dentry, task,
2042                                 (void *)(unsigned long)vma->vm_file->f_mode);
2043
2044 out_no_vma:
2045         up_read(&mm->mmap_sem);
2046         mmput(mm);
2047 out_put_task:
2048         put_task_struct(task);
2049 out:
2050         return ERR_PTR(result);
2051 }
2052
2053 static const struct inode_operations proc_map_files_inode_operations = {
2054         .lookup         = proc_map_files_lookup,
2055         .permission     = proc_fd_permission,
2056         .setattr        = proc_setattr,
2057 };
2058
2059 static int
2060 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2061 {
2062         struct vm_area_struct *vma;
2063         struct task_struct *task;
2064         struct mm_struct *mm;
2065         unsigned long nr_files, pos, i;
2066         struct flex_array *fa = NULL;
2067         struct map_files_info info;
2068         struct map_files_info *p;
2069         int ret;
2070
2071         ret = -ENOENT;
2072         task = get_proc_task(file_inode(file));
2073         if (!task)
2074                 goto out;
2075
2076         ret = -EACCES;
2077         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2078                 goto out_put_task;
2079
2080         ret = 0;
2081         if (!dir_emit_dots(file, ctx))
2082                 goto out_put_task;
2083
2084         mm = get_task_mm(task);
2085         if (!mm)
2086                 goto out_put_task;
2087         down_read(&mm->mmap_sem);
2088
2089         nr_files = 0;
2090
2091         /*
2092          * We need two passes here:
2093          *
2094          *  1) Collect vmas of mapped files with mmap_sem taken
2095          *  2) Release mmap_sem and instantiate entries
2096          *
2097          * otherwise we get lockdep complained, since filldir()
2098          * routine might require mmap_sem taken in might_fault().
2099          */
2100
2101         for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2102                 if (vma->vm_file && ++pos > ctx->pos)
2103                         nr_files++;
2104         }
2105
2106         if (nr_files) {
2107                 fa = flex_array_alloc(sizeof(info), nr_files,
2108                                         GFP_KERNEL);
2109                 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2110                                                 GFP_KERNEL)) {
2111                         ret = -ENOMEM;
2112                         if (fa)
2113                                 flex_array_free(fa);
2114                         up_read(&mm->mmap_sem);
2115                         mmput(mm);
2116                         goto out_put_task;
2117                 }
2118                 for (i = 0, vma = mm->mmap, pos = 2; vma;
2119                                 vma = vma->vm_next) {
2120                         if (!vma->vm_file)
2121                                 continue;
2122                         if (++pos <= ctx->pos)
2123                                 continue;
2124
2125                         info.mode = vma->vm_file->f_mode;
2126                         info.len = snprintf(info.name,
2127                                         sizeof(info.name), "%lx-%lx",
2128                                         vma->vm_start, vma->vm_end);
2129                         if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2130                                 BUG();
2131                 }
2132         }
2133         up_read(&mm->mmap_sem);
2134
2135         for (i = 0; i < nr_files; i++) {
2136                 p = flex_array_get(fa, i);
2137                 if (!proc_fill_cache(file, ctx,
2138                                       p->name, p->len,
2139                                       proc_map_files_instantiate,
2140                                       task,
2141                                       (void *)(unsigned long)p->mode))
2142                         break;
2143                 ctx->pos++;
2144         }
2145         if (fa)
2146                 flex_array_free(fa);
2147         mmput(mm);
2148
2149 out_put_task:
2150         put_task_struct(task);
2151 out:
2152         return ret;
2153 }
2154
2155 static const struct file_operations proc_map_files_operations = {
2156         .read           = generic_read_dir,
2157         .iterate        = proc_map_files_readdir,
2158         .llseek         = default_llseek,
2159 };
2160
2161 #ifdef CONFIG_CHECKPOINT_RESTORE
2162 struct timers_private {
2163         struct pid *pid;
2164         struct task_struct *task;
2165         struct sighand_struct *sighand;
2166         struct pid_namespace *ns;
2167         unsigned long flags;
2168 };
2169
2170 static void *timers_start(struct seq_file *m, loff_t *pos)
2171 {
2172         struct timers_private *tp = m->private;
2173
2174         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2175         if (!tp->task)
2176                 return ERR_PTR(-ESRCH);
2177
2178         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2179         if (!tp->sighand)
2180                 return ERR_PTR(-ESRCH);
2181
2182         return seq_list_start(&tp->task->signal->posix_timers, *pos);
2183 }
2184
2185 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2186 {
2187         struct timers_private *tp = m->private;
2188         return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2189 }
2190
2191 static void timers_stop(struct seq_file *m, void *v)
2192 {
2193         struct timers_private *tp = m->private;
2194
2195         if (tp->sighand) {
2196                 unlock_task_sighand(tp->task, &tp->flags);
2197                 tp->sighand = NULL;
2198         }
2199
2200         if (tp->task) {
2201                 put_task_struct(tp->task);
2202                 tp->task = NULL;
2203         }
2204 }
2205
2206 static int show_timer(struct seq_file *m, void *v)
2207 {
2208         struct k_itimer *timer;
2209         struct timers_private *tp = m->private;
2210         int notify;
2211         static const char * const nstr[] = {
2212                 [SIGEV_SIGNAL] = "signal",
2213                 [SIGEV_NONE] = "none",
2214                 [SIGEV_THREAD] = "thread",
2215         };
2216
2217         timer = list_entry((struct list_head *)v, struct k_itimer, list);
2218         notify = timer->it_sigev_notify;
2219
2220         seq_printf(m, "ID: %d\n", timer->it_id);
2221         seq_printf(m, "signal: %d/%p\n",
2222                    timer->sigq->info.si_signo,
2223                    timer->sigq->info.si_value.sival_ptr);
2224         seq_printf(m, "notify: %s/%s.%d\n",
2225                    nstr[notify & ~SIGEV_THREAD_ID],
2226                    (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2227                    pid_nr_ns(timer->it_pid, tp->ns));
2228         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2229
2230         return 0;
2231 }
2232
2233 static const struct seq_operations proc_timers_seq_ops = {
2234         .start  = timers_start,
2235         .next   = timers_next,
2236         .stop   = timers_stop,
2237         .show   = show_timer,
2238 };
2239
2240 static int proc_timers_open(struct inode *inode, struct file *file)
2241 {
2242         struct timers_private *tp;
2243
2244         tp = __seq_open_private(file, &proc_timers_seq_ops,
2245                         sizeof(struct timers_private));
2246         if (!tp)
2247                 return -ENOMEM;
2248
2249         tp->pid = proc_pid(inode);
2250         tp->ns = inode->i_sb->s_fs_info;
2251         return 0;
2252 }
2253
2254 static const struct file_operations proc_timers_operations = {
2255         .open           = proc_timers_open,
2256         .read           = seq_read,
2257         .llseek         = seq_lseek,
2258         .release        = seq_release_private,
2259 };
2260 #endif
2261
2262 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2263                                         size_t count, loff_t *offset)
2264 {
2265         struct inode *inode = file_inode(file);
2266         struct task_struct *p;
2267         u64 slack_ns;
2268         int err;
2269
2270         err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2271         if (err < 0)
2272                 return err;
2273
2274         p = get_proc_task(inode);
2275         if (!p)
2276                 return -ESRCH;
2277
2278         if (ptrace_may_access(p, PTRACE_MODE_ATTACH_FSCREDS)) {
2279                 task_lock(p);
2280                 if (slack_ns == 0)
2281                         p->timer_slack_ns = p->default_timer_slack_ns;
2282                 else
2283                         p->timer_slack_ns = slack_ns;
2284                 task_unlock(p);
2285         } else
2286                 count = -EPERM;
2287
2288         put_task_struct(p);
2289
2290         return count;
2291 }
2292
2293 static int timerslack_ns_show(struct seq_file *m, void *v)
2294 {
2295         struct inode *inode = m->private;
2296         struct task_struct *p;
2297         int err =  0;
2298
2299         p = get_proc_task(inode);
2300         if (!p)
2301                 return -ESRCH;
2302
2303         if (ptrace_may_access(p, PTRACE_MODE_ATTACH_FSCREDS)) {
2304                 task_lock(p);
2305                 seq_printf(m, "%llu\n", p->timer_slack_ns);
2306                 task_unlock(p);
2307         } else
2308                 err = -EPERM;
2309
2310         put_task_struct(p);
2311
2312         return err;
2313 }
2314
2315 static int timerslack_ns_open(struct inode *inode, struct file *filp)
2316 {
2317         return single_open(filp, timerslack_ns_show, inode);
2318 }
2319
2320 static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2321         .open           = timerslack_ns_open,
2322         .read           = seq_read,
2323         .write          = timerslack_ns_write,
2324         .llseek         = seq_lseek,
2325         .release        = single_release,
2326 };
2327
2328 static int proc_pident_instantiate(struct inode *dir,
2329         struct dentry *dentry, struct task_struct *task, const void *ptr)
2330 {
2331         const struct pid_entry *p = ptr;
2332         struct inode *inode;
2333         struct proc_inode *ei;
2334
2335         inode = proc_pid_make_inode(dir->i_sb, task);
2336         if (!inode)
2337                 goto out;
2338
2339         ei = PROC_I(inode);
2340         inode->i_mode = p->mode;
2341         if (S_ISDIR(inode->i_mode))
2342                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2343         if (p->iop)
2344                 inode->i_op = p->iop;
2345         if (p->fop)
2346                 inode->i_fop = p->fop;
2347         ei->op = p->op;
2348         d_set_d_op(dentry, &pid_dentry_operations);
2349         d_add(dentry, inode);
2350         /* Close the race of the process dying before we return the dentry */
2351         if (pid_revalidate(dentry, 0))
2352                 return 0;
2353 out:
2354         return -ENOENT;
2355 }
2356
2357 static struct dentry *proc_pident_lookup(struct inode *dir,
2358                                          struct dentry *dentry,
2359                                          const struct pid_entry *ents,
2360                                          unsigned int nents)
2361 {
2362         int error;
2363         struct task_struct *task = get_proc_task(dir);
2364         const struct pid_entry *p, *last;
2365
2366         error = -ENOENT;
2367
2368         if (!task)
2369                 goto out_no_task;
2370
2371         /*
2372          * Yes, it does not scale. And it should not. Don't add
2373          * new entries into /proc/<tgid>/ without very good reasons.
2374          */
2375         last = &ents[nents - 1];
2376         for (p = ents; p <= last; p++) {
2377                 if (p->len != dentry->d_name.len)
2378                         continue;
2379                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2380                         break;
2381         }
2382         if (p > last)
2383                 goto out;
2384
2385         error = proc_pident_instantiate(dir, dentry, task, p);
2386 out:
2387         put_task_struct(task);
2388 out_no_task:
2389         return ERR_PTR(error);
2390 }
2391
2392 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2393                 const struct pid_entry *ents, unsigned int nents)
2394 {
2395         struct task_struct *task = get_proc_task(file_inode(file));
2396         const struct pid_entry *p;
2397
2398         if (!task)
2399                 return -ENOENT;
2400
2401         if (!dir_emit_dots(file, ctx))
2402                 goto out;
2403
2404         if (ctx->pos >= nents + 2)
2405                 goto out;
2406
2407         for (p = ents + (ctx->pos - 2); p <= ents + nents - 1; p++) {
2408                 if (!proc_fill_cache(file, ctx, p->name, p->len,
2409                                 proc_pident_instantiate, task, p))
2410                         break;
2411                 ctx->pos++;
2412         }
2413 out:
2414         put_task_struct(task);
2415         return 0;
2416 }
2417
2418 #ifdef CONFIG_SECURITY
2419 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2420                                   size_t count, loff_t *ppos)
2421 {
2422         struct inode * inode = file_inode(file);
2423         char *p = NULL;
2424         ssize_t length;
2425         struct task_struct *task = get_proc_task(inode);
2426
2427         if (!task)
2428                 return -ESRCH;
2429
2430         length = security_getprocattr(task,
2431                                       (char*)file->f_path.dentry->d_name.name,
2432                                       &p);
2433         put_task_struct(task);
2434         if (length > 0)
2435                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2436         kfree(p);
2437         return length;
2438 }
2439
2440 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2441                                    size_t count, loff_t *ppos)
2442 {
2443         struct inode * inode = file_inode(file);
2444         void *page;
2445         ssize_t length;
2446         struct task_struct *task = get_proc_task(inode);
2447
2448         length = -ESRCH;
2449         if (!task)
2450                 goto out_no_task;
2451         if (count > PAGE_SIZE)
2452                 count = PAGE_SIZE;
2453
2454         /* No partial writes. */
2455         length = -EINVAL;
2456         if (*ppos != 0)
2457                 goto out;
2458
2459         page = memdup_user(buf, count);
2460         if (IS_ERR(page)) {
2461                 length = PTR_ERR(page);
2462                 goto out;
2463         }
2464
2465         /* Guard against adverse ptrace interaction */
2466         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2467         if (length < 0)
2468                 goto out_free;
2469
2470         length = security_setprocattr(task,
2471                                       (char*)file->f_path.dentry->d_name.name,
2472                                       page, count);
2473         mutex_unlock(&task->signal->cred_guard_mutex);
2474 out_free:
2475         kfree(page);
2476 out:
2477         put_task_struct(task);
2478 out_no_task:
2479         return length;
2480 }
2481
2482 static const struct file_operations proc_pid_attr_operations = {
2483         .read           = proc_pid_attr_read,
2484         .write          = proc_pid_attr_write,
2485         .llseek         = generic_file_llseek,
2486 };
2487
2488 static const struct pid_entry attr_dir_stuff[] = {
2489         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2490         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2491         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2492         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2493         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2494         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2495 };
2496
2497 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2498 {
2499         return proc_pident_readdir(file, ctx,
2500                                    attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2501 }
2502
2503 static const struct file_operations proc_attr_dir_operations = {
2504         .read           = generic_read_dir,
2505         .iterate        = proc_attr_dir_readdir,
2506         .llseek         = default_llseek,
2507 };
2508
2509 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2510                                 struct dentry *dentry, unsigned int flags)
2511 {
2512         return proc_pident_lookup(dir, dentry,
2513                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2514 }
2515
2516 static const struct inode_operations proc_attr_dir_inode_operations = {
2517         .lookup         = proc_attr_dir_lookup,
2518         .getattr        = pid_getattr,
2519         .setattr        = proc_setattr,
2520 };
2521
2522 #endif
2523
2524 #ifdef CONFIG_ELF_CORE
2525 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2526                                          size_t count, loff_t *ppos)
2527 {
2528         struct task_struct *task = get_proc_task(file_inode(file));
2529         struct mm_struct *mm;
2530         char buffer[PROC_NUMBUF];
2531         size_t len;
2532         int ret;
2533
2534         if (!task)
2535                 return -ESRCH;
2536
2537         ret = 0;
2538         mm = get_task_mm(task);
2539         if (mm) {
2540                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2541                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2542                                 MMF_DUMP_FILTER_SHIFT));
2543                 mmput(mm);
2544                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2545         }
2546
2547         put_task_struct(task);
2548
2549         return ret;
2550 }
2551
2552 static ssize_t proc_coredump_filter_write(struct file *file,
2553                                           const char __user *buf,
2554                                           size_t count,
2555                                           loff_t *ppos)
2556 {
2557         struct task_struct *task;
2558         struct mm_struct *mm;
2559         unsigned int val;
2560         int ret;
2561         int i;
2562         unsigned long mask;
2563
2564         ret = kstrtouint_from_user(buf, count, 0, &val);
2565         if (ret < 0)
2566                 return ret;
2567
2568         ret = -ESRCH;
2569         task = get_proc_task(file_inode(file));
2570         if (!task)
2571                 goto out_no_task;
2572
2573         mm = get_task_mm(task);
2574         if (!mm)
2575                 goto out_no_mm;
2576         ret = 0;
2577
2578         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2579                 if (val & mask)
2580                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2581                 else
2582                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2583         }
2584
2585         mmput(mm);
2586  out_no_mm:
2587         put_task_struct(task);
2588  out_no_task:
2589         if (ret < 0)
2590                 return ret;
2591         return count;
2592 }
2593
2594 static const struct file_operations proc_coredump_filter_operations = {
2595         .read           = proc_coredump_filter_read,
2596         .write          = proc_coredump_filter_write,
2597         .llseek         = generic_file_llseek,
2598 };
2599 #endif
2600
2601 #ifdef CONFIG_TASK_IO_ACCOUNTING
2602 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2603 {
2604         struct task_io_accounting acct = task->ioac;
2605         unsigned long flags;
2606         int result;
2607
2608         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2609         if (result)
2610                 return result;
2611
2612         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2613                 result = -EACCES;
2614                 goto out_unlock;
2615         }
2616
2617         if (whole && lock_task_sighand(task, &flags)) {
2618                 struct task_struct *t = task;
2619
2620                 task_io_accounting_add(&acct, &task->signal->ioac);
2621                 while_each_thread(task, t)
2622                         task_io_accounting_add(&acct, &t->ioac);
2623
2624                 unlock_task_sighand(task, &flags);
2625         }
2626         seq_printf(m,
2627                    "rchar: %llu\n"
2628                    "wchar: %llu\n"
2629                    "syscr: %llu\n"
2630                    "syscw: %llu\n"
2631                    "read_bytes: %llu\n"
2632                    "write_bytes: %llu\n"
2633                    "cancelled_write_bytes: %llu\n",
2634                    (unsigned long long)acct.rchar,
2635                    (unsigned long long)acct.wchar,
2636                    (unsigned long long)acct.syscr,
2637                    (unsigned long long)acct.syscw,
2638                    (unsigned long long)acct.read_bytes,
2639                    (unsigned long long)acct.write_bytes,
2640                    (unsigned long long)acct.cancelled_write_bytes);
2641         result = 0;
2642
2643 out_unlock:
2644         mutex_unlock(&task->signal->cred_guard_mutex);
2645         return result;
2646 }
2647
2648 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2649                                   struct pid *pid, struct task_struct *task)
2650 {
2651         return do_io_accounting(task, m, 0);
2652 }
2653
2654 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2655                                    struct pid *pid, struct task_struct *task)
2656 {
2657         return do_io_accounting(task, m, 1);
2658 }
2659 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2660
2661 #ifdef CONFIG_USER_NS
2662 static int proc_id_map_open(struct inode *inode, struct file *file,
2663         const struct seq_operations *seq_ops)
2664 {
2665         struct user_namespace *ns = NULL;
2666         struct task_struct *task;
2667         struct seq_file *seq;
2668         int ret = -EINVAL;
2669
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         ret = seq_open(file, seq_ops);
2681         if (ret)
2682                 goto err_put_ns;
2683
2684         seq = file->private_data;
2685         seq->private = ns;
2686
2687         return 0;
2688 err_put_ns:
2689         put_user_ns(ns);
2690 err:
2691         return ret;
2692 }
2693
2694 static int proc_id_map_release(struct inode *inode, struct file *file)
2695 {
2696         struct seq_file *seq = file->private_data;
2697         struct user_namespace *ns = seq->private;
2698         put_user_ns(ns);
2699         return seq_release(inode, file);
2700 }
2701
2702 static int proc_uid_map_open(struct inode *inode, struct file *file)
2703 {
2704         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2705 }
2706
2707 static int proc_gid_map_open(struct inode *inode, struct file *file)
2708 {
2709         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2710 }
2711
2712 static int proc_projid_map_open(struct inode *inode, struct file *file)
2713 {
2714         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2715 }
2716
2717 static const struct file_operations proc_uid_map_operations = {
2718         .open           = proc_uid_map_open,
2719         .write          = proc_uid_map_write,
2720         .read           = seq_read,
2721         .llseek         = seq_lseek,
2722         .release        = proc_id_map_release,
2723 };
2724
2725 static const struct file_operations proc_gid_map_operations = {
2726         .open           = proc_gid_map_open,
2727         .write          = proc_gid_map_write,
2728         .read           = seq_read,
2729         .llseek         = seq_lseek,
2730         .release        = proc_id_map_release,
2731 };
2732
2733 static const struct file_operations proc_projid_map_operations = {
2734         .open           = proc_projid_map_open,
2735         .write          = proc_projid_map_write,
2736         .read           = seq_read,
2737         .llseek         = seq_lseek,
2738         .release        = proc_id_map_release,
2739 };
2740
2741 static int proc_setgroups_open(struct inode *inode, struct file *file)
2742 {
2743         struct user_namespace *ns = NULL;
2744         struct task_struct *task;
2745         int ret;
2746
2747         ret = -ESRCH;
2748         task = get_proc_task(inode);
2749         if (task) {
2750                 rcu_read_lock();
2751                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2752                 rcu_read_unlock();
2753                 put_task_struct(task);
2754         }
2755         if (!ns)
2756                 goto err;
2757
2758         if (file->f_mode & FMODE_WRITE) {
2759                 ret = -EACCES;
2760                 if (!ns_capable(ns, CAP_SYS_ADMIN))
2761                         goto err_put_ns;
2762         }
2763
2764         ret = single_open(file, &proc_setgroups_show, ns);
2765         if (ret)
2766                 goto err_put_ns;
2767
2768         return 0;
2769 err_put_ns:
2770         put_user_ns(ns);
2771 err:
2772         return ret;
2773 }
2774
2775 static int proc_setgroups_release(struct inode *inode, struct file *file)
2776 {
2777         struct seq_file *seq = file->private_data;
2778         struct user_namespace *ns = seq->private;
2779         int ret = single_release(inode, file);
2780         put_user_ns(ns);
2781         return ret;
2782 }
2783
2784 static const struct file_operations proc_setgroups_operations = {
2785         .open           = proc_setgroups_open,
2786         .write          = proc_setgroups_write,
2787         .read           = seq_read,
2788         .llseek         = seq_lseek,
2789         .release        = proc_setgroups_release,
2790 };
2791 #endif /* CONFIG_USER_NS */
2792
2793 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2794                                 struct pid *pid, struct task_struct *task)
2795 {
2796         int err = lock_trace(task);
2797         if (!err) {
2798                 seq_printf(m, "%08x\n", task->personality);
2799                 unlock_trace(task);
2800         }
2801         return err;
2802 }
2803
2804 /*
2805  * Thread groups
2806  */
2807 static const struct file_operations proc_task_operations;
2808 static const struct inode_operations proc_task_inode_operations;
2809
2810 static const struct pid_entry tgid_base_stuff[] = {
2811         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2812         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2813         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2814         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2815         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2816 #ifdef CONFIG_NET
2817         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2818 #endif
2819         REG("environ",    S_IRUSR, proc_environ_operations),
2820         ONE("auxv",       S_IRUSR, proc_pid_auxv),
2821         ONE("status",     S_IRUGO, proc_pid_status),
2822         ONE("personality", S_IRUSR, proc_pid_personality),
2823         ONE("limits",     S_IRUGO, proc_pid_limits),
2824 #ifdef CONFIG_SCHED_DEBUG
2825         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2826 #endif
2827 #ifdef CONFIG_SCHED_AUTOGROUP
2828         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2829 #endif
2830         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2831 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2832         ONE("syscall",    S_IRUSR, proc_pid_syscall),
2833 #endif
2834         REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
2835         ONE("stat",       S_IRUGO, proc_tgid_stat),
2836         ONE("statm",      S_IRUGO, proc_pid_statm),
2837         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2838 #ifdef CONFIG_NUMA
2839         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2840 #endif
2841         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2842         LNK("cwd",        proc_cwd_link),
2843         LNK("root",       proc_root_link),
2844         LNK("exe",        proc_exe_link),
2845         REG("mounts",     S_IRUGO, proc_mounts_operations),
2846         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2847         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2848 #ifdef CONFIG_PROC_PAGE_MONITOR
2849         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2850         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2851         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2852 #endif
2853 #ifdef CONFIG_SECURITY
2854         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2855 #endif
2856 #ifdef CONFIG_KALLSYMS
2857         ONE("wchan",      S_IRUGO, proc_pid_wchan),
2858 #endif
2859 #ifdef CONFIG_STACKTRACE
2860         ONE("stack",      S_IRUSR, proc_pid_stack),
2861 #endif
2862 #ifdef CONFIG_SCHED_INFO
2863         ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
2864 #endif
2865 #ifdef CONFIG_LATENCYTOP
2866         REG("latency",  S_IRUGO, proc_lstats_operations),
2867 #endif
2868 #ifdef CONFIG_PROC_PID_CPUSET
2869         ONE("cpuset",     S_IRUGO, proc_cpuset_show),
2870 #endif
2871 #ifdef CONFIG_CGROUPS
2872         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
2873 #endif
2874         ONE("oom_score",  S_IRUGO, proc_oom_score),
2875         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2876         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2877 #ifdef CONFIG_AUDITSYSCALL
2878         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2879         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2880 #endif
2881 #ifdef CONFIG_FAULT_INJECTION
2882         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2883 #endif
2884 #ifdef CONFIG_ELF_CORE
2885         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2886 #endif
2887 #ifdef CONFIG_TASK_IO_ACCOUNTING
2888         ONE("io",       S_IRUSR, proc_tgid_io_accounting),
2889 #endif
2890 #ifdef CONFIG_HARDWALL
2891         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
2892 #endif
2893 #ifdef CONFIG_USER_NS
2894         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2895         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2896         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2897         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
2898 #endif
2899 #ifdef CONFIG_CHECKPOINT_RESTORE
2900         REG("timers",     S_IRUGO, proc_timers_operations),
2901 #endif
2902         REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
2903 };
2904
2905 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2906 {
2907         return proc_pident_readdir(file, ctx,
2908                                    tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2909 }
2910
2911 static const struct file_operations proc_tgid_base_operations = {
2912         .read           = generic_read_dir,
2913         .iterate        = proc_tgid_base_readdir,
2914         .llseek         = default_llseek,
2915 };
2916
2917 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2918 {
2919         return proc_pident_lookup(dir, dentry,
2920                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2921 }
2922
2923 static const struct inode_operations proc_tgid_base_inode_operations = {
2924         .lookup         = proc_tgid_base_lookup,
2925         .getattr        = pid_getattr,
2926         .setattr        = proc_setattr,
2927         .permission     = proc_pid_permission,
2928 };
2929
2930 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2931 {
2932         struct dentry *dentry, *leader, *dir;
2933         char buf[PROC_NUMBUF];
2934         struct qstr name;
2935
2936         name.name = buf;
2937         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2938         /* no ->d_hash() rejects on procfs */
2939         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2940         if (dentry) {
2941                 d_invalidate(dentry);
2942                 dput(dentry);
2943         }
2944
2945         if (pid == tgid)
2946                 return;
2947
2948         name.name = buf;
2949         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2950         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2951         if (!leader)
2952                 goto out;
2953
2954         name.name = "task";
2955         name.len = strlen(name.name);
2956         dir = d_hash_and_lookup(leader, &name);
2957         if (!dir)
2958                 goto out_put_leader;
2959
2960         name.name = buf;
2961         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2962         dentry = d_hash_and_lookup(dir, &name);
2963         if (dentry) {
2964                 d_invalidate(dentry);
2965                 dput(dentry);
2966         }
2967
2968         dput(dir);
2969 out_put_leader:
2970         dput(leader);
2971 out:
2972         return;
2973 }
2974
2975 /**
2976  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2977  * @task: task that should be flushed.
2978  *
2979  * When flushing dentries from proc, one needs to flush them from global
2980  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2981  * in. This call is supposed to do all of this job.
2982  *
2983  * Looks in the dcache for
2984  * /proc/@pid
2985  * /proc/@tgid/task/@pid
2986  * if either directory is present flushes it and all of it'ts children
2987  * from the dcache.
2988  *
2989  * It is safe and reasonable to cache /proc entries for a task until
2990  * that task exits.  After that they just clog up the dcache with
2991  * useless entries, possibly causing useful dcache entries to be
2992  * flushed instead.  This routine is proved to flush those useless
2993  * dcache entries at process exit time.
2994  *
2995  * NOTE: This routine is just an optimization so it does not guarantee
2996  *       that no dcache entries will exist at process exit time it
2997  *       just makes it very unlikely that any will persist.
2998  */
2999
3000 void proc_flush_task(struct task_struct *task)
3001 {
3002         int i;
3003         struct pid *pid, *tgid;
3004         struct upid *upid;
3005
3006         pid = task_pid(task);
3007         tgid = task_tgid(task);
3008
3009         for (i = 0; i <= pid->level; i++) {
3010                 upid = &pid->numbers[i];
3011                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3012                                         tgid->numbers[i].nr);
3013         }
3014 }
3015
3016 static int proc_pid_instantiate(struct inode *dir,
3017                                    struct dentry * dentry,
3018                                    struct task_struct *task, const void *ptr)
3019 {
3020         struct inode *inode;
3021
3022         inode = proc_pid_make_inode(dir->i_sb, task);
3023         if (!inode)
3024                 goto out;
3025
3026         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3027         inode->i_op = &proc_tgid_base_inode_operations;
3028         inode->i_fop = &proc_tgid_base_operations;
3029         inode->i_flags|=S_IMMUTABLE;
3030
3031         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
3032                                                   ARRAY_SIZE(tgid_base_stuff)));
3033
3034         d_set_d_op(dentry, &pid_dentry_operations);
3035
3036         d_add(dentry, inode);
3037         /* Close the race of the process dying before we return the dentry */
3038         if (pid_revalidate(dentry, 0))
3039                 return 0;
3040 out:
3041         return -ENOENT;
3042 }
3043
3044 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3045 {
3046         int result = -ENOENT;
3047         struct task_struct *task;
3048         unsigned tgid;
3049         struct pid_namespace *ns;
3050
3051         tgid = name_to_int(&dentry->d_name);
3052         if (tgid == ~0U)
3053                 goto out;
3054
3055         ns = dentry->d_sb->s_fs_info;
3056         rcu_read_lock();
3057         task = find_task_by_pid_ns(tgid, ns);
3058         if (task)
3059                 get_task_struct(task);
3060         rcu_read_unlock();
3061         if (!task)
3062                 goto out;
3063
3064         result = proc_pid_instantiate(dir, dentry, task, NULL);
3065         put_task_struct(task);
3066 out:
3067         return ERR_PTR(result);
3068 }
3069
3070 /*
3071  * Find the first task with tgid >= tgid
3072  *
3073  */
3074 struct tgid_iter {
3075         unsigned int tgid;
3076         struct task_struct *task;
3077 };
3078 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3079 {
3080         struct pid *pid;
3081
3082         if (iter.task)
3083                 put_task_struct(iter.task);
3084         rcu_read_lock();
3085 retry:
3086         iter.task = NULL;
3087         pid = find_ge_pid(iter.tgid, ns);
3088         if (pid) {
3089                 iter.tgid = pid_nr_ns(pid, ns);
3090                 iter.task = pid_task(pid, PIDTYPE_PID);
3091                 /* What we to know is if the pid we have find is the
3092                  * pid of a thread_group_leader.  Testing for task
3093                  * being a thread_group_leader is the obvious thing
3094                  * todo but there is a window when it fails, due to
3095                  * the pid transfer logic in de_thread.
3096                  *
3097                  * So we perform the straight forward test of seeing
3098                  * if the pid we have found is the pid of a thread
3099                  * group leader, and don't worry if the task we have
3100                  * found doesn't happen to be a thread group leader.
3101                  * As we don't care in the case of readdir.
3102                  */
3103                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3104                         iter.tgid += 1;
3105                         goto retry;
3106                 }
3107                 get_task_struct(iter.task);
3108         }
3109         rcu_read_unlock();
3110         return iter;
3111 }
3112
3113 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3114
3115 /* for the /proc/ directory itself, after non-process stuff has been done */
3116 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3117 {
3118         struct tgid_iter iter;
3119         struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
3120         loff_t pos = ctx->pos;
3121
3122         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3123                 return 0;
3124
3125         if (pos == TGID_OFFSET - 2) {
3126                 struct inode *inode = d_inode(ns->proc_self);
3127                 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3128                         return 0;
3129                 ctx->pos = pos = pos + 1;
3130         }
3131         if (pos == TGID_OFFSET - 1) {
3132                 struct inode *inode = d_inode(ns->proc_thread_self);
3133                 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3134                         return 0;
3135                 ctx->pos = pos = pos + 1;
3136         }
3137         iter.tgid = pos - TGID_OFFSET;
3138         iter.task = NULL;
3139         for (iter = next_tgid(ns, iter);
3140              iter.task;
3141              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3142                 char name[PROC_NUMBUF];
3143                 int len;
3144                 if (!has_pid_permissions(ns, iter.task, 2))
3145                         continue;
3146
3147                 len = snprintf(name, sizeof(name), "%d", iter.tgid);
3148                 ctx->pos = iter.tgid + TGID_OFFSET;
3149                 if (!proc_fill_cache(file, ctx, name, len,
3150                                      proc_pid_instantiate, iter.task, NULL)) {
3151                         put_task_struct(iter.task);
3152                         return 0;
3153                 }
3154         }
3155         ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3156         return 0;
3157 }
3158
3159 /*
3160  * Tasks
3161  */
3162 static const struct pid_entry tid_base_stuff[] = {
3163         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3164         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3165         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3166 #ifdef CONFIG_NET
3167         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3168 #endif
3169         REG("environ",   S_IRUSR, proc_environ_operations),
3170         ONE("auxv",      S_IRUSR, proc_pid_auxv),
3171         ONE("status",    S_IRUGO, proc_pid_status),
3172         ONE("personality", S_IRUSR, proc_pid_personality),
3173         ONE("limits",    S_IRUGO, proc_pid_limits),
3174 #ifdef CONFIG_SCHED_DEBUG
3175         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3176 #endif
3177         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3178 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3179         ONE("syscall",   S_IRUSR, proc_pid_syscall),
3180 #endif
3181         REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3182         ONE("stat",      S_IRUGO, proc_tid_stat),
3183         ONE("statm",     S_IRUGO, proc_pid_statm),
3184         REG("maps",      S_IRUGO, proc_tid_maps_operations),
3185 #ifdef CONFIG_PROC_CHILDREN
3186         REG("children",  S_IRUGO, proc_tid_children_operations),
3187 #endif
3188 #ifdef CONFIG_NUMA
3189         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3190 #endif
3191         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3192         LNK("cwd",       proc_cwd_link),
3193         LNK("root",      proc_root_link),
3194         LNK("exe",       proc_exe_link),
3195         REG("mounts",    S_IRUGO, proc_mounts_operations),
3196         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3197 #ifdef CONFIG_PROC_PAGE_MONITOR
3198         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3199         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3200         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3201 #endif
3202 #ifdef CONFIG_SECURITY
3203         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3204 #endif
3205 #ifdef CONFIG_KALLSYMS
3206         ONE("wchan",     S_IRUGO, proc_pid_wchan),
3207 #endif
3208 #ifdef CONFIG_STACKTRACE
3209         ONE("stack",      S_IRUSR, proc_pid_stack),
3210 #endif
3211 #ifdef CONFIG_SCHED_INFO
3212         ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3213 #endif
3214 #ifdef CONFIG_LATENCYTOP
3215         REG("latency",  S_IRUGO, proc_lstats_operations),
3216 #endif
3217 #ifdef CONFIG_PROC_PID_CPUSET
3218         ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3219 #endif
3220 #ifdef CONFIG_CGROUPS
3221         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3222 #endif
3223         ONE("oom_score", S_IRUGO, proc_oom_score),
3224         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3225         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3226 #ifdef CONFIG_AUDITSYSCALL
3227         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3228         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3229 #endif
3230 #ifdef CONFIG_FAULT_INJECTION
3231         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3232 #endif
3233 #ifdef CONFIG_TASK_IO_ACCOUNTING
3234         ONE("io",       S_IRUSR, proc_tid_io_accounting),
3235 #endif
3236 #ifdef CONFIG_HARDWALL
3237         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
3238 #endif
3239 #ifdef CONFIG_USER_NS
3240         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3241         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3242         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3243         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3244 #endif
3245 };
3246
3247 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3248 {
3249         return proc_pident_readdir(file, ctx,
3250                                    tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3251 }
3252
3253 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3254 {
3255         return proc_pident_lookup(dir, dentry,
3256                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3257 }
3258
3259 static const struct file_operations proc_tid_base_operations = {
3260         .read           = generic_read_dir,
3261         .iterate        = proc_tid_base_readdir,
3262         .llseek         = default_llseek,
3263 };
3264
3265 static const struct inode_operations proc_tid_base_inode_operations = {
3266         .lookup         = proc_tid_base_lookup,
3267         .getattr        = pid_getattr,
3268         .setattr        = proc_setattr,
3269 };
3270
3271 static int proc_task_instantiate(struct inode *dir,
3272         struct dentry *dentry, struct task_struct *task, const void *ptr)
3273 {
3274         struct inode *inode;
3275         inode = proc_pid_make_inode(dir->i_sb, task);
3276
3277         if (!inode)
3278                 goto out;
3279         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3280         inode->i_op = &proc_tid_base_inode_operations;
3281         inode->i_fop = &proc_tid_base_operations;
3282         inode->i_flags|=S_IMMUTABLE;
3283
3284         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3285                                                   ARRAY_SIZE(tid_base_stuff)));
3286
3287         d_set_d_op(dentry, &pid_dentry_operations);
3288
3289         d_add(dentry, inode);
3290         /* Close the race of the process dying before we return the dentry */
3291         if (pid_revalidate(dentry, 0))
3292                 return 0;
3293 out:
3294         return -ENOENT;
3295 }
3296
3297 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3298 {
3299         int result = -ENOENT;
3300         struct task_struct *task;
3301         struct task_struct *leader = get_proc_task(dir);
3302         unsigned tid;
3303         struct pid_namespace *ns;
3304
3305         if (!leader)
3306                 goto out_no_task;
3307
3308         tid = name_to_int(&dentry->d_name);
3309         if (tid == ~0U)
3310                 goto out;
3311
3312         ns = dentry->d_sb->s_fs_info;
3313         rcu_read_lock();
3314         task = find_task_by_pid_ns(tid, ns);
3315         if (task)
3316                 get_task_struct(task);
3317         rcu_read_unlock();
3318         if (!task)
3319                 goto out;
3320         if (!same_thread_group(leader, task))
3321                 goto out_drop_task;
3322
3323         result = proc_task_instantiate(dir, dentry, task, NULL);
3324 out_drop_task:
3325         put_task_struct(task);
3326 out:
3327         put_task_struct(leader);
3328 out_no_task:
3329         return ERR_PTR(result);
3330 }
3331
3332 /*
3333  * Find the first tid of a thread group to return to user space.
3334  *
3335  * Usually this is just the thread group leader, but if the users
3336  * buffer was too small or there was a seek into the middle of the
3337  * directory we have more work todo.
3338  *
3339  * In the case of a short read we start with find_task_by_pid.
3340  *
3341  * In the case of a seek we start with the leader and walk nr
3342  * threads past it.
3343  */
3344 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3345                                         struct pid_namespace *ns)
3346 {
3347         struct task_struct *pos, *task;
3348         unsigned long nr = f_pos;
3349
3350         if (nr != f_pos)        /* 32bit overflow? */
3351                 return NULL;
3352
3353         rcu_read_lock();
3354         task = pid_task(pid, PIDTYPE_PID);
3355         if (!task)
3356                 goto fail;
3357
3358         /* Attempt to start with the tid of a thread */
3359         if (tid && nr) {
3360                 pos = find_task_by_pid_ns(tid, ns);
3361                 if (pos && same_thread_group(pos, task))
3362                         goto found;
3363         }
3364
3365         /* If nr exceeds the number of threads there is nothing todo */
3366         if (nr >= get_nr_threads(task))
3367                 goto fail;
3368
3369         /* If we haven't found our starting place yet start
3370          * with the leader and walk nr threads forward.
3371          */
3372         pos = task = task->group_leader;
3373         do {
3374                 if (!nr--)
3375                         goto found;
3376         } while_each_thread(task, pos);
3377 fail:
3378         pos = NULL;
3379         goto out;
3380 found:
3381         get_task_struct(pos);
3382 out:
3383         rcu_read_unlock();
3384         return pos;
3385 }
3386
3387 /*
3388  * Find the next thread in the thread list.
3389  * Return NULL if there is an error or no next thread.
3390  *
3391  * The reference to the input task_struct is released.
3392  */
3393 static struct task_struct *next_tid(struct task_struct *start)
3394 {
3395         struct task_struct *pos = NULL;
3396         rcu_read_lock();
3397         if (pid_alive(start)) {
3398                 pos = next_thread(start);
3399                 if (thread_group_leader(pos))
3400                         pos = NULL;
3401                 else
3402                         get_task_struct(pos);
3403         }
3404         rcu_read_unlock();
3405         put_task_struct(start);
3406         return pos;
3407 }
3408
3409 /* for the /proc/TGID/task/ directories */
3410 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3411 {
3412         struct inode *inode = file_inode(file);
3413         struct task_struct *task;
3414         struct pid_namespace *ns;
3415         int tid;
3416
3417         if (proc_inode_is_dead(inode))
3418                 return -ENOENT;
3419
3420         if (!dir_emit_dots(file, ctx))
3421                 return 0;
3422
3423         /* f_version caches the tgid value that the last readdir call couldn't
3424          * return. lseek aka telldir automagically resets f_version to 0.
3425          */
3426         ns = inode->i_sb->s_fs_info;
3427         tid = (int)file->f_version;
3428         file->f_version = 0;
3429         for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3430              task;
3431              task = next_tid(task), ctx->pos++) {
3432                 char name[PROC_NUMBUF];
3433                 int len;
3434                 tid = task_pid_nr_ns(task, ns);
3435                 len = snprintf(name, sizeof(name), "%d", tid);
3436                 if (!proc_fill_cache(file, ctx, name, len,
3437                                 proc_task_instantiate, task, NULL)) {
3438                         /* returning this tgid failed, save it as the first
3439                          * pid for the next readir call */
3440                         file->f_version = (u64)tid;
3441                         put_task_struct(task);
3442                         break;
3443                 }
3444         }
3445
3446         return 0;
3447 }
3448
3449 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3450 {
3451         struct inode *inode = d_inode(dentry);
3452         struct task_struct *p = get_proc_task(inode);
3453         generic_fillattr(inode, stat);
3454
3455         if (p) {
3456                 stat->nlink += get_nr_threads(p);
3457                 put_task_struct(p);
3458         }
3459
3460         return 0;
3461 }
3462
3463 static const struct inode_operations proc_task_inode_operations = {
3464         .lookup         = proc_task_lookup,
3465         .getattr        = proc_task_getattr,
3466         .setattr        = proc_setattr,
3467         .permission     = proc_pid_permission,
3468 };
3469
3470 static const struct file_operations proc_task_operations = {
3471         .read           = generic_read_dir,
3472         .iterate        = proc_task_readdir,
3473         .llseek         = default_llseek,
3474 };