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         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 INF(NAME, MODE, read)                           \
 134         NOD(NAME, (S_IFREG|(MODE)),                     \
 135                 NULL, &proc_info_file_operations,       \
 136                 { .proc_read = read } )
 137 #define ONE(NAME, MODE, show)                           \
 138         NOD(NAME, (S_IFREG|(MODE)),                     \
 139                 NULL, &proc_single_file_operations,     \
 140                 { .proc_show = show } )
 141
 142 /* ANDROID is for special files in /proc. */
 143 #define ANDROID(NAME, MODE, OTYPE)                      \
 144         NOD(NAME, (S_IFREG|(MODE)),                     \
 145                 &proc_##OTYPE##_inode_operations,       \
 146                 &proc_##OTYPE##_operations, {})
 147
 148 /*
 149  * Count the number of hardlinks for the pid_entry table, excluding the .
 150  * and .. links.
 151  */
 152 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
 153         unsigned int n)
 154 {
 155         unsigned int i;
 156         unsigned int count;
 157
 158         count = 0;
 159         for (i = 0; i < n; ++i) {
 160                 if (S_ISDIR(entries[i].mode))
 161                         ++count;
 162         }
 163
 164         return count;
 165 }
 166
 167 static int get_task_root(struct task_struct *task, struct path *root)
 168 {
 169         int result = -ENOENT;
 170
 171         task_lock(task);
 172         if (task->fs) {
 173                 get_fs_root(task->fs, root);
 174                 result = 0;
 175         }
 176         task_unlock(task);
 177         return result;
 178 }
 179
 180 static int proc_cwd_link(struct dentry *dentry, struct path *path)
 181 {
 182         struct task_struct *task = get_proc_task(dentry->d_inode);
 183         int result = -ENOENT;
 184
 185         if (task) {
 186                 task_lock(task);
 187                 if (task->fs) {
 188                         get_fs_pwd(task->fs, path);
 189                         result = 0;
 190                 }
 191                 task_unlock(task);
 192                 put_task_struct(task);
 193         }
 194         return result;
 195 }
 196
 197 static int proc_root_link(struct dentry *dentry, struct path *path)
 198 {
 199         struct task_struct *task = get_proc_task(dentry->d_inode);
 200         int result = -ENOENT;
 201
 202         if (task) {
 203                 result = get_task_root(task, path);
 204                 put_task_struct(task);
 205         }
 206         return result;
 207 }
 208
 209 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
 210 {
 211         int res = 0;
 212         unsigned int len;
 213         struct mm_struct *mm = get_task_mm(task);
 214         if (!mm)
 215                 goto out;
 216         if (!mm->arg_end)
 217                 goto out_mm;    /* Shh! No looking before we're done */
 218
 219         len = mm->arg_end - mm->arg_start;
 220
 221         if (len > PAGE_SIZE)
 222                 len = PAGE_SIZE;
 223
 224         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
 225
 226         // If the nul at the end of args has been overwritten, then
 227         // assume application is using setproctitle(3).
 228         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
 229                 len = strnlen(buffer, res);
 230                 if (len < res) {
 231                     res = len;
 232                 } else {
 233                         len = mm->env_end - mm->env_start;
 234                         if (len > PAGE_SIZE - res)
 235                                 len = PAGE_SIZE - res;
 236                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
 237                         res = strnlen(buffer, res);
 238                 }
 239         }
 240 out_mm:
 241         mmput(mm);
 242 out:
 243         return res;
 244 }
 245
 246 static int proc_pid_auxv(struct task_struct *task, char *buffer)
 247 {
 248         struct mm_struct *mm = mm_access(task, PTRACE_MODE_READ);
 249         int res = PTR_ERR(mm);
 250         if (mm && !IS_ERR(mm)) {
 251                 unsigned int nwords = 0;
 252                 do {
 253                         nwords += 2;
 254                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 255                 res = nwords * sizeof(mm->saved_auxv[0]);
 256                 if (res > PAGE_SIZE)
 257                         res = PAGE_SIZE;
 258                 memcpy(buffer, mm->saved_auxv, res);
 259                 mmput(mm);
 260         }
 261         return res;
 262 }
 263
 264
 265 #ifdef CONFIG_KALLSYMS
 266 /*
 267  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 268  * Returns the resolved symbol.  If that fails, simply return the address.
 269  */
 270 static int proc_pid_wchan(struct task_struct *task, char *buffer)
 271 {
 272         unsigned long wchan;
 273         char symname[KSYM_NAME_LEN];
 274
 275         wchan = get_wchan(task);
 276
 277         if (lookup_symbol_name(wchan, symname) < 0)
 278                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
 279                         return 0;
 280                 else
 281                         return sprintf(buffer, "%lu", wchan);
 282         else
 283                 return sprintf(buffer, "%s", symname);
 284 }
 285 #endif /* CONFIG_KALLSYMS */
 286
 287 static int lock_trace(struct task_struct *task)
 288 {
 289         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
 290         if (err)
 291                 return err;
 292         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH)) {
 293                 mutex_unlock(&task->signal->cred_guard_mutex);
 294                 return -EPERM;
 295         }
 296         return 0;
 297 }
 298
 299 static void unlock_trace(struct task_struct *task)
 300 {
 301         mutex_unlock(&task->signal->cred_guard_mutex);
 302 }
 303
 304 #ifdef CONFIG_STACKTRACE
 305
 306 #define MAX_STACK_TRACE_DEPTH   64
 307
 308 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 309                           struct pid *pid, struct task_struct *task)
 310 {
 311         struct stack_trace trace;
 312         unsigned long *entries;
 313         int err;
 314         int i;
 315
 316         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 317         if (!entries)
 318                 return -ENOMEM;
 319
 320         trace.nr_entries        = 0;
 321         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 322         trace.entries           = entries;
 323         trace.skip              = 0;
 324
 325         err = lock_trace(task);
 326         if (!err) {
 327                 save_stack_trace_tsk(task, &trace);
 328
 329                 for (i = 0; i < trace.nr_entries; i++) {
 330                         seq_printf(m, "[<%pK>] %pS\n",
 331                                    (void *)entries[i], (void *)entries[i]);
 332                 }
 333                 unlock_trace(task);
 334         }
 335         kfree(entries);
 336
 337         return err;
 338 }
 339 #endif
 340
 341 #ifdef CONFIG_SCHEDSTATS
 342 /*
 343  * Provides /proc/PID/schedstat
 344  */
 345 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
 346 {
 347         return sprintf(buffer, "%llu %llu %lu\n",
 348                         (unsigned long long)task->se.sum_exec_runtime,
 349                         (unsigned long long)task->sched_info.run_delay,
 350                         task->sched_info.pcount);
 351 }
 352 #endif
 353
 354 #ifdef CONFIG_LATENCYTOP
 355 static int lstats_show_proc(struct seq_file *m, void *v)
 356 {
 357         int i;
 358         struct inode *inode = m->private;
 359         struct task_struct *task = get_proc_task(inode);
 360
 361         if (!task)
 362                 return -ESRCH;
 363         seq_puts(m, "Latency Top version : v0.1\n");
 364         for (i = 0; i < 32; i++) {
 365                 struct latency_record *lr = &task->latency_record[i];
 366                 if (lr->backtrace[0]) {
 367                         int q;
 368                         seq_printf(m, "%i %li %li",
 369                                    lr->count, lr->time, lr->max);
 370                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 371                                 unsigned long bt = lr->backtrace[q];
 372                                 if (!bt)
 373                                         break;
 374                                 if (bt == ULONG_MAX)
 375                                         break;
 376                                 seq_printf(m, " %ps", (void *)bt);
 377                         }
 378                         seq_putc(m, '\n');
 379                 }
 380
 381         }
 382         put_task_struct(task);
 383         return 0;
 384 }
 385
 386 static int lstats_open(struct inode *inode, struct file *file)
 387 {
 388         return single_open(file, lstats_show_proc, inode);
 389 }
 390
 391 static ssize_t lstats_write(struct file *file, const char __user *buf,
 392                             size_t count, loff_t *offs)
 393 {
 394         struct task_struct *task = get_proc_task(file_inode(file));
 395
 396         if (!task)
 397                 return -ESRCH;
 398         clear_all_latency_tracing(task);
 399         put_task_struct(task);
 400
 401         return count;
 402 }
 403
 404 static const struct file_operations proc_lstats_operations = {
 405         .open           = lstats_open,
 406         .read           = seq_read,
 407         .write          = lstats_write,
 408         .llseek         = seq_lseek,
 409         .release        = single_release,
 410 };
 411
 412 #endif
 413
 414 #ifdef CONFIG_CGROUPS
 415 static int cgroup_open(struct inode *inode, struct file *file)
 416 {
 417         struct pid *pid = PROC_I(inode)->pid;
 418         return single_open(file, proc_cgroup_show, pid);
 419 }
 420
 421 static const struct file_operations proc_cgroup_operations = {
 422         .open           = cgroup_open,
 423         .read           = seq_read,
 424         .llseek         = seq_lseek,
 425         .release        = single_release,
 426 };
 427 #endif
 428
 429 #ifdef CONFIG_PROC_PID_CPUSET
 430
 431 static int cpuset_open(struct inode *inode, struct file *file)
 432 {
 433         struct pid *pid = PROC_I(inode)->pid;
 434         return single_open(file, proc_cpuset_show, pid);
 435 }
 436
 437 static const struct file_operations proc_cpuset_operations = {
 438         .open           = cpuset_open,
 439         .read           = seq_read,
 440         .llseek         = seq_lseek,
 441         .release        = single_release,
 442 };
 443 #endif
 444
 445 static int proc_oom_score(struct task_struct *task, char *buffer)
 446 {
 447         unsigned long totalpages = totalram_pages + total_swap_pages;
 448         unsigned long points = 0;
 449
 450         read_lock(&tasklist_lock);
 451         if (pid_alive(task))
 452                 points = oom_badness(task, NULL, NULL, totalpages) *
 453                                                 1000 / totalpages;
 454         read_unlock(&tasklist_lock);
 455         return sprintf(buffer, "%lu\n", points);
 456 }
 457
 458 struct limit_names {
 459         char *name;
 460         char *unit;
 461 };
 462
 463 static const struct limit_names lnames[RLIM_NLIMITS] = {
 464         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 465         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 466         [RLIMIT_DATA] = {"Max data size", "bytes"},
 467         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 468         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 469         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 470         [RLIMIT_NPROC] = {"Max processes", "processes"},
 471         [RLIMIT_NOFILE] = {"Max open files", "files"},
 472         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 473         [RLIMIT_AS] = {"Max address space", "bytes"},
 474         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 475         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 476         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 477         [RLIMIT_NICE] = {"Max nice priority", NULL},
 478         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 479         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 480 };
 481
 482 /* Display limits for a process */
 483 static int proc_pid_limits(struct task_struct *task, char *buffer)
 484 {
 485         unsigned int i;
 486         int count = 0;
 487         unsigned long flags;
 488         char *bufptr = buffer;
 489
 490         struct rlimit rlim[RLIM_NLIMITS];
 491
 492         if (!lock_task_sighand(task, &flags))
 493                 return 0;
 494         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 495         unlock_task_sighand(task, &flags);
 496
 497         /*
 498          * print the file header
 499          */
 500         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
 501                         "Limit", "Soft Limit", "Hard Limit", "Units");
 502
 503         for (i = 0; i < RLIM_NLIMITS; i++) {
 504                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 505                         count += sprintf(&bufptr[count], "%-25s %-20s ",
 506                                          lnames[i].name, "unlimited");
 507                 else
 508                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
 509                                          lnames[i].name, rlim[i].rlim_cur);
 510
 511                 if (rlim[i].rlim_max == RLIM_INFINITY)
 512                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
 513                 else
 514                         count += sprintf(&bufptr[count], "%-20lu ",
 515                                          rlim[i].rlim_max);
 516
 517                 if (lnames[i].unit)
 518                         count += sprintf(&bufptr[count], "%-10s\n",
 519                                          lnames[i].unit);
 520                 else
 521                         count += sprintf(&bufptr[count], "\n");
 522         }
 523
 524         return count;
 525 }
 526
 527 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 528 static int proc_pid_syscall(struct task_struct *task, char *buffer)
 529 {
 530         long nr;
 531         unsigned long args[6], sp, pc;
 532         int res = lock_trace(task);
 533         if (res)
 534                 return res;
 535
 536         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 537                 res = sprintf(buffer, "running\n");
 538         else if (nr < 0)
 539                 res = sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 540         else
 541                 res = sprintf(buffer,
 542                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 543                        nr,
 544                        args[0], args[1], args[2], args[3], args[4], args[5],
 545                        sp, pc);
 546         unlock_trace(task);
 547         return res;
 548 }
 549 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 550
 551 /************************************************************************/
 552 /*                       Here the fs part begins                        */
 553 /************************************************************************/
 554
 555 /* permission checks */
 556 static int proc_fd_access_allowed(struct inode *inode)
 557 {
 558         struct task_struct *task;
 559         int allowed = 0;
 560         /* Allow access to a task's file descriptors if it is us or we
 561          * may use ptrace attach to the process and find out that
 562          * information.
 563          */
 564         task = get_proc_task(inode);
 565         if (task) {
 566                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
 567                 put_task_struct(task);
 568         }
 569         return allowed;
 570 }
 571
 572 int proc_setattr(struct dentry *dentry, struct iattr *attr)
 573 {
 574         int error;
 575         struct inode *inode = dentry->d_inode;
 576
 577         if (attr->ia_valid & ATTR_MODE)
 578                 return -EPERM;
 579
 580         error = inode_change_ok(inode, attr);
 581         if (error)
 582                 return error;
 583
 584         setattr_copy(inode, attr);
 585         mark_inode_dirty(inode);
 586         return 0;
 587 }
 588
 589 /*
 590  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
 591  * or euid/egid (for hide_pid_min=2)?
 592  */
 593 static bool has_pid_permissions(struct pid_namespace *pid,
 594                                  struct task_struct *task,
 595                                  int hide_pid_min)
 596 {
 597         if (pid->hide_pid < hide_pid_min)
 598                 return true;
 599         if (in_group_p(pid->pid_gid))
 600                 return true;
 601         return ptrace_may_access(task, PTRACE_MODE_READ);
 602 }
 603
 604
 605 static int proc_pid_permission(struct inode *inode, int mask)
 606 {
 607         struct pid_namespace *pid = inode->i_sb->s_fs_info;
 608         struct task_struct *task;
 609         bool has_perms;
 610
 611         task = get_proc_task(inode);
 612         if (!task)
 613                 return -ESRCH;
 614         has_perms = has_pid_permissions(pid, task, 1);
 615         put_task_struct(task);
 616
 617         if (!has_perms) {
 618                 if (pid->hide_pid == 2) {
 619                         /*
 620                          * Let's make getdents(), stat(), and open()
 621                          * consistent with each other.  If a process
 622                          * may not stat() a file, it shouldn't be seen
 623                          * in procfs at all.
 624                          */
 625                         return -ENOENT;
 626                 }
 627
 628                 return -EPERM;
 629         }
 630         return generic_permission(inode, mask);
 631 }
 632
 633
 634
 635 static const struct inode_operations proc_def_inode_operations = {
 636         .setattr        = proc_setattr,
 637 };
 638
 639 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 640
 641 static ssize_t proc_info_read(struct file * file, char __user * buf,
 642                           size_t count, loff_t *ppos)
 643 {
 644         struct inode * inode = file_inode(file);
 645         unsigned long page;
 646         ssize_t length;
 647         struct task_struct *task = get_proc_task(inode);
 648
 649         length = -ESRCH;
 650         if (!task)
 651                 goto out_no_task;
 652
 653         if (count > PROC_BLOCK_SIZE)
 654                 count = PROC_BLOCK_SIZE;
 655
 656         length = -ENOMEM;
 657         if (!(page = __get_free_page(GFP_TEMPORARY)))
 658                 goto out;
 659
 660         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 661
 662         if (length >= 0)
 663                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 664         free_page(page);
 665 out:
 666         put_task_struct(task);
 667 out_no_task:
 668         return length;
 669 }
 670
 671 static const struct file_operations proc_info_file_operations = {
 672         .read           = proc_info_read,
 673         .llseek         = generic_file_llseek,
 674 };
 675
 676 static int proc_single_show(struct seq_file *m, void *v)
 677 {
 678         struct inode *inode = m->private;
 679         struct pid_namespace *ns;
 680         struct pid *pid;
 681         struct task_struct *task;
 682         int ret;
 683
 684         ns = inode->i_sb->s_fs_info;
 685         pid = proc_pid(inode);
 686         task = get_pid_task(pid, PIDTYPE_PID);
 687         if (!task)
 688                 return -ESRCH;
 689
 690         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 691
 692         put_task_struct(task);
 693         return ret;
 694 }
 695
 696 static int proc_single_open(struct inode *inode, struct file *filp)
 697 {
 698         return single_open(filp, proc_single_show, inode);
 699 }
 700
 701 static const struct file_operations proc_single_file_operations = {
 702         .open           = proc_single_open,
 703         .read           = seq_read,
 704         .llseek         = seq_lseek,
 705         .release        = single_release,
 706 };
 707
 708 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
 709 {
 710         struct task_struct *task = get_proc_task(file_inode(file));
 711         struct mm_struct *mm;
 712
 713         if (!task)
 714                 return -ESRCH;
 715
 716         mm = mm_access(task, mode);
 717         put_task_struct(task);
 718
 719         if (IS_ERR(mm))
 720                 return PTR_ERR(mm);
 721
 722         if (mm) {
 723                 /* ensure this mm_struct can't be freed */
 724                 atomic_inc(&mm->mm_count);
 725                 /* but do not pin its memory */
 726                 mmput(mm);
 727         }
 728
 729         file->private_data = mm;
 730
 731         return 0;
 732 }
 733
 734 static int mem_open(struct inode *inode, struct file *file)
 735 {
 736         int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
 737
 738         /* OK to pass negative loff_t, we can catch out-of-range */
 739         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 740
 741         return ret;
 742 }
 743
 744 static ssize_t mem_rw(struct file *file, char __user *buf,
 745                         size_t count, loff_t *ppos, int write)
 746 {
 747         struct mm_struct *mm = file->private_data;
 748         unsigned long addr = *ppos;
 749         ssize_t copied;
 750         char *page;
 751
 752         if (!mm)
 753                 return 0;
 754
 755         page = (char *)__get_free_page(GFP_TEMPORARY);
 756         if (!page)
 757                 return -ENOMEM;
 758
 759         copied = 0;
 760         if (!atomic_inc_not_zero(&mm->mm_users))
 761                 goto free;
 762
 763         while (count > 0) {
 764                 int this_len = min_t(int, count, PAGE_SIZE);
 765
 766                 if (write && copy_from_user(page, buf, this_len)) {
 767                         copied = -EFAULT;
 768                         break;
 769                 }
 770
 771                 this_len = access_remote_vm(mm, addr, page, this_len, write);
 772                 if (!this_len) {
 773                         if (!copied)
 774                                 copied = -EIO;
 775                         break;
 776                 }
 777
 778                 if (!write && copy_to_user(buf, page, this_len)) {
 779                         copied = -EFAULT;
 780                         break;
 781                 }
 782
 783                 buf += this_len;
 784                 addr += this_len;
 785                 copied += this_len;
 786                 count -= this_len;
 787         }
 788         *ppos = addr;
 789
 790         mmput(mm);
 791 free:
 792         free_page((unsigned long) page);
 793         return copied;
 794 }
 795
 796 static ssize_t mem_read(struct file *file, char __user *buf,
 797                         size_t count, loff_t *ppos)
 798 {
 799         return mem_rw(file, buf, count, ppos, 0);
 800 }
 801
 802 static ssize_t mem_write(struct file *file, const char __user *buf,
 803                          size_t count, loff_t *ppos)
 804 {
 805         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 806 }
 807
 808 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 809 {
 810         switch (orig) {
 811         case 0:
 812                 file->f_pos = offset;
 813                 break;
 814         case 1:
 815                 file->f_pos += offset;
 816                 break;
 817         default:
 818                 return -EINVAL;
 819         }
 820         force_successful_syscall_return();
 821         return file->f_pos;
 822 }
 823
 824 static int mem_release(struct inode *inode, struct file *file)
 825 {
 826         struct mm_struct *mm = file->private_data;
 827         if (mm)
 828                 mmdrop(mm);
 829         return 0;
 830 }
 831
 832 static const struct file_operations proc_mem_operations = {
 833         .llseek         = mem_lseek,
 834         .read           = mem_read,
 835         .write          = mem_write,
 836         .open           = mem_open,
 837         .release        = mem_release,
 838 };
 839
 840 static int environ_open(struct inode *inode, struct file *file)
 841 {
 842         return __mem_open(inode, file, PTRACE_MODE_READ);
 843 }
 844
 845 static ssize_t environ_read(struct file *file, char __user *buf,
 846                         size_t count, loff_t *ppos)
 847 {
 848         char *page;
 849         unsigned long src = *ppos;
 850         int ret = 0;
 851         struct mm_struct *mm = file->private_data;
 852
 853         if (!mm)
 854                 return 0;
 855
 856         page = (char *)__get_free_page(GFP_TEMPORARY);
 857         if (!page)
 858                 return -ENOMEM;
 859
 860         ret = 0;
 861         if (!atomic_inc_not_zero(&mm->mm_users))
 862                 goto free;
 863         while (count > 0) {
 864                 size_t this_len, max_len;
 865                 int retval;
 866
 867                 if (src >= (mm->env_end - mm->env_start))
 868                         break;
 869
 870                 this_len = mm->env_end - (mm->env_start + src);
 871
 872                 max_len = min_t(size_t, PAGE_SIZE, count);
 873                 this_len = min(max_len, this_len);
 874
 875                 retval = access_remote_vm(mm, (mm->env_start + src),
 876                         page, this_len, 0);
 877
 878                 if (retval <= 0) {
 879                         ret = retval;
 880                         break;
 881                 }
 882
 883                 if (copy_to_user(buf, page, retval)) {
 884                         ret = -EFAULT;
 885                         break;
 886                 }
 887
 888                 ret += retval;
 889                 src += retval;
 890                 buf += retval;
 891                 count -= retval;
 892         }
 893         *ppos = src;
 894         mmput(mm);
 895
 896 free:
 897         free_page((unsigned long) page);
 898         return ret;
 899 }
 900
 901 static const struct file_operations proc_environ_operations = {
 902         .open           = environ_open,
 903         .read           = environ_read,
 904         .llseek         = generic_file_llseek,
 905         .release        = mem_release,
 906 };
 907
 908 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
 909                             loff_t *ppos)
 910 {
 911         struct task_struct *task = get_proc_task(file_inode(file));
 912         char buffer[PROC_NUMBUF];
 913         int oom_adj = OOM_ADJUST_MIN;
 914         size_t len;
 915         unsigned long flags;
 916
 917         if (!task)
 918                 return -ESRCH;
 919         if (lock_task_sighand(task, &flags)) {
 920                 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
 921                         oom_adj = OOM_ADJUST_MAX;
 922                 else
 923                         oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
 924                                   OOM_SCORE_ADJ_MAX;
 925                 unlock_task_sighand(task, &flags);
 926         }
 927         put_task_struct(task);
 928         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
 929         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 930 }
 931
 932 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
 933                              size_t count, loff_t *ppos)
 934 {
 935         struct task_struct *task;
 936         char buffer[PROC_NUMBUF];
 937         int oom_adj;
 938         unsigned long flags;
 939         int err;
 940
 941         memset(buffer, 0, sizeof(buffer));
 942         if (count > sizeof(buffer) - 1)
 943                 count = sizeof(buffer) - 1;
 944         if (copy_from_user(buffer, buf, count)) {
 945                 err = -EFAULT;
 946                 goto out;
 947         }
 948
 949         err = kstrtoint(strstrip(buffer), 0, &oom_adj);
 950         if (err)
 951                 goto out;
 952         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
 953              oom_adj != OOM_DISABLE) {
 954                 err = -EINVAL;
 955                 goto out;
 956         }
 957
 958         task = get_proc_task(file_inode(file));
 959         if (!task) {
 960                 err = -ESRCH;
 961                 goto out;
 962         }
 963
 964         task_lock(task);
 965         if (!task->mm) {
 966                 err = -EINVAL;
 967                 goto err_task_lock;
 968         }
 969
 970         if (!lock_task_sighand(task, &flags)) {
 971                 err = -ESRCH;
 972                 goto err_task_lock;
 973         }
 974
 975         /*
 976          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
 977          * value is always attainable.
 978          */
 979         if (oom_adj == OOM_ADJUST_MAX)
 980                 oom_adj = OOM_SCORE_ADJ_MAX;
 981         else
 982                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
 983
 984         if (oom_adj < task->signal->oom_score_adj &&
 985             !capable(CAP_SYS_RESOURCE)) {
 986                 err = -EACCES;
 987                 goto err_sighand;
 988         }
 989
 990         /*
 991          * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
 992          * /proc/pid/oom_score_adj instead.
 993          */
 994         pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
 995                   current->comm, task_pid_nr(current), task_pid_nr(task),
 996                   task_pid_nr(task));
 997
 998         task->signal->oom_score_adj = oom_adj;
 999         trace_oom_score_adj_update(task);
1000 err_sighand:
1001         unlock_task_sighand(task, &flags);
1002 err_task_lock:
1003         task_unlock(task);
1004         put_task_struct(task);
1005 out:
1006         return err < 0 ? err : count;
1007 }
1008
1009 static int oom_adjust_permission(struct inode *inode, int mask)
1010 {
1011         uid_t uid;
1012         struct task_struct *p;
1013
1014         p = get_proc_task(inode);
1015         if(p) {
1016                 uid = task_uid(p);
1017                 put_task_struct(p);
1018         }
1019
1020         /*
1021          * System Server (uid == 1000) is granted access to oom_adj of all
1022          * android applications (uid > 10000) as and services (uid >= 1000)
1023          */
1024         if (p && (current_fsuid() == 1000) && (uid >= 1000)) {
1025                 if (inode->i_mode >> 6 & mask) {
1026                         return 0;
1027                 }
1028         }
1029
1030         /* Fall back to default. */
1031         return generic_permission(inode, mask);
1032 }
1033
1034 static const struct inode_operations proc_oom_adj_inode_operations = {
1035         .permission     = oom_adjust_permission,
1036 };
1037
1038 static const struct file_operations proc_oom_adj_operations = {
1039         .read           = oom_adj_read,
1040         .write          = oom_adj_write,
1041         .llseek         = generic_file_llseek,
1042 };
1043
1044 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1045                                         size_t count, loff_t *ppos)
1046 {
1047         struct task_struct *task = get_proc_task(file_inode(file));
1048         char buffer[PROC_NUMBUF];
1049         short oom_score_adj = OOM_SCORE_ADJ_MIN;
1050         unsigned long flags;
1051         size_t len;
1052
1053         if (!task)
1054                 return -ESRCH;
1055         if (lock_task_sighand(task, &flags)) {
1056                 oom_score_adj = task->signal->oom_score_adj;
1057                 unlock_task_sighand(task, &flags);
1058         }
1059         put_task_struct(task);
1060         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1061         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1062 }
1063
1064 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1065                                         size_t count, loff_t *ppos)
1066 {
1067         struct task_struct *task;
1068         char buffer[PROC_NUMBUF];
1069         unsigned long flags;
1070         int oom_score_adj;
1071         int err;
1072
1073         memset(buffer, 0, sizeof(buffer));
1074         if (count > sizeof(buffer) - 1)
1075                 count = sizeof(buffer) - 1;
1076         if (copy_from_user(buffer, buf, count)) {
1077                 err = -EFAULT;
1078                 goto out;
1079         }
1080
1081         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1082         if (err)
1083                 goto out;
1084         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1085                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1086                 err = -EINVAL;
1087                 goto out;
1088         }
1089
1090         task = get_proc_task(file_inode(file));
1091         if (!task) {
1092                 err = -ESRCH;
1093                 goto out;
1094         }
1095
1096         task_lock(task);
1097         if (!task->mm) {
1098                 err = -EINVAL;
1099                 goto err_task_lock;
1100         }
1101
1102         if (!lock_task_sighand(task, &flags)) {
1103                 err = -ESRCH;
1104                 goto err_task_lock;
1105         }
1106
1107         if ((short)oom_score_adj < task->signal->oom_score_adj_min &&
1108                         !capable(CAP_SYS_RESOURCE)) {
1109                 err = -EACCES;
1110                 goto err_sighand;
1111         }
1112
1113         task->signal->oom_score_adj = (short)oom_score_adj;
1114         if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1115                 task->signal->oom_score_adj_min = (short)oom_score_adj;
1116         trace_oom_score_adj_update(task);
1117
1118 err_sighand:
1119         unlock_task_sighand(task, &flags);
1120 err_task_lock:
1121         task_unlock(task);
1122         put_task_struct(task);
1123 out:
1124         return err < 0 ? err : count;
1125 }
1126
1127 static const struct file_operations proc_oom_score_adj_operations = {
1128         .read           = oom_score_adj_read,
1129         .write          = oom_score_adj_write,
1130         .llseek         = default_llseek,
1131 };
1132
1133 #ifdef CONFIG_AUDITSYSCALL
1134 #define TMPBUFLEN 21
1135 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1136                                   size_t count, loff_t *ppos)
1137 {
1138         struct inode * inode = file_inode(file);
1139         struct task_struct *task = get_proc_task(inode);
1140         ssize_t length;
1141         char tmpbuf[TMPBUFLEN];
1142
1143         if (!task)
1144                 return -ESRCH;
1145         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1146                            from_kuid(file->f_cred->user_ns,
1147                                      audit_get_loginuid(task)));
1148         put_task_struct(task);
1149         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1150 }
1151
1152 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1153                                    size_t count, loff_t *ppos)
1154 {
1155         struct inode * inode = file_inode(file);
1156         char *page, *tmp;
1157         ssize_t length;
1158         uid_t loginuid;
1159         kuid_t kloginuid;
1160
1161         rcu_read_lock();
1162         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1163                 rcu_read_unlock();
1164                 return -EPERM;
1165         }
1166         rcu_read_unlock();
1167
1168         if (count >= PAGE_SIZE)
1169                 count = PAGE_SIZE - 1;
1170
1171         if (*ppos != 0) {
1172                 /* No partial writes. */
1173                 return -EINVAL;
1174         }
1175         page = (char*)__get_free_page(GFP_TEMPORARY);
1176         if (!page)
1177                 return -ENOMEM;
1178         length = -EFAULT;
1179         if (copy_from_user(page, buf, count))
1180                 goto out_free_page;
1181
1182         page[count] = '\0';
1183         loginuid = simple_strtoul(page, &tmp, 10);
1184         if (tmp == page) {
1185                 length = -EINVAL;
1186                 goto out_free_page;
1187
1188         }
1189         kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1190         if (!uid_valid(kloginuid)) {
1191                 length = -EINVAL;
1192                 goto out_free_page;
1193         }
1194
1195         length = audit_set_loginuid(kloginuid);
1196         if (likely(length == 0))
1197                 length = count;
1198
1199 out_free_page:
1200         free_page((unsigned long) page);
1201         return length;
1202 }
1203
1204 static const struct file_operations proc_loginuid_operations = {
1205         .read           = proc_loginuid_read,
1206         .write          = proc_loginuid_write,
1207         .llseek         = generic_file_llseek,
1208 };
1209
1210 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1211                                   size_t count, loff_t *ppos)
1212 {
1213         struct inode * inode = file_inode(file);
1214         struct task_struct *task = get_proc_task(inode);
1215         ssize_t length;
1216         char tmpbuf[TMPBUFLEN];
1217
1218         if (!task)
1219                 return -ESRCH;
1220         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1221                                 audit_get_sessionid(task));
1222         put_task_struct(task);
1223         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1224 }
1225
1226 static const struct file_operations proc_sessionid_operations = {
1227         .read           = proc_sessionid_read,
1228         .llseek         = generic_file_llseek,
1229 };
1230 #endif
1231
1232 #ifdef CONFIG_FAULT_INJECTION
1233 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1234                                       size_t count, loff_t *ppos)
1235 {
1236         struct task_struct *task = get_proc_task(file_inode(file));
1237         char buffer[PROC_NUMBUF];
1238         size_t len;
1239         int make_it_fail;
1240
1241         if (!task)
1242                 return -ESRCH;
1243         make_it_fail = task->make_it_fail;
1244         put_task_struct(task);
1245
1246         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1247
1248         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1249 }
1250
1251 static ssize_t proc_fault_inject_write(struct file * file,
1252                         const char __user * buf, size_t count, loff_t *ppos)
1253 {
1254         struct task_struct *task;
1255         char buffer[PROC_NUMBUF], *end;
1256         int make_it_fail;
1257
1258         if (!capable(CAP_SYS_RESOURCE))
1259                 return -EPERM;
1260         memset(buffer, 0, sizeof(buffer));
1261         if (count > sizeof(buffer) - 1)
1262                 count = sizeof(buffer) - 1;
1263         if (copy_from_user(buffer, buf, count))
1264                 return -EFAULT;
1265         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1266         if (*end)
1267                 return -EINVAL;
1268         task = get_proc_task(file_inode(file));
1269         if (!task)
1270                 return -ESRCH;
1271         task->make_it_fail = make_it_fail;
1272         put_task_struct(task);
1273
1274         return count;
1275 }
1276
1277 static const struct file_operations proc_fault_inject_operations = {
1278         .read           = proc_fault_inject_read,
1279         .write          = proc_fault_inject_write,
1280         .llseek         = generic_file_llseek,
1281 };
1282 #endif
1283
1284
1285 #ifdef CONFIG_SCHED_DEBUG
1286 /*
1287  * Print out various scheduling related per-task fields:
1288  */
1289 static int sched_show(struct seq_file *m, void *v)
1290 {
1291         struct inode *inode = m->private;
1292         struct task_struct *p;
1293
1294         p = get_proc_task(inode);
1295         if (!p)
1296                 return -ESRCH;
1297         proc_sched_show_task(p, m);
1298
1299         put_task_struct(p);
1300
1301         return 0;
1302 }
1303
1304 static ssize_t
1305 sched_write(struct file *file, const char __user *buf,
1306             size_t count, loff_t *offset)
1307 {
1308         struct inode *inode = file_inode(file);
1309         struct task_struct *p;
1310
1311         p = get_proc_task(inode);
1312         if (!p)
1313                 return -ESRCH;
1314         proc_sched_set_task(p);
1315
1316         put_task_struct(p);
1317
1318         return count;
1319 }
1320
1321 static int sched_open(struct inode *inode, struct file *filp)
1322 {
1323         return single_open(filp, sched_show, inode);
1324 }
1325
1326 static const struct file_operations proc_pid_sched_operations = {
1327         .open           = sched_open,
1328         .read           = seq_read,
1329         .write          = sched_write,
1330         .llseek         = seq_lseek,
1331         .release        = single_release,
1332 };
1333
1334 #endif
1335
1336 #ifdef CONFIG_SCHED_AUTOGROUP
1337 /*
1338  * Print out autogroup related information:
1339  */
1340 static int sched_autogroup_show(struct seq_file *m, void *v)
1341 {
1342         struct inode *inode = m->private;
1343         struct task_struct *p;
1344
1345         p = get_proc_task(inode);
1346         if (!p)
1347                 return -ESRCH;
1348         proc_sched_autogroup_show_task(p, m);
1349
1350         put_task_struct(p);
1351
1352         return 0;
1353 }
1354
1355 static ssize_t
1356 sched_autogroup_write(struct file *file, const char __user *buf,
1357             size_t count, loff_t *offset)
1358 {
1359         struct inode *inode = file_inode(file);
1360         struct task_struct *p;
1361         char buffer[PROC_NUMBUF];
1362         int nice;
1363         int err;
1364
1365         memset(buffer, 0, sizeof(buffer));
1366         if (count > sizeof(buffer) - 1)
1367                 count = sizeof(buffer) - 1;
1368         if (copy_from_user(buffer, buf, count))
1369                 return -EFAULT;
1370
1371         err = kstrtoint(strstrip(buffer), 0, &nice);
1372         if (err < 0)
1373                 return err;
1374
1375         p = get_proc_task(inode);
1376         if (!p)
1377                 return -ESRCH;
1378
1379         err = proc_sched_autogroup_set_nice(p, nice);
1380         if (err)
1381                 count = err;
1382
1383         put_task_struct(p);
1384
1385         return count;
1386 }
1387
1388 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1389 {
1390         int ret;
1391
1392         ret = single_open(filp, sched_autogroup_show, NULL);
1393         if (!ret) {
1394                 struct seq_file *m = filp->private_data;
1395
1396                 m->private = inode;
1397         }
1398         return ret;
1399 }
1400
1401 static const struct file_operations proc_pid_sched_autogroup_operations = {
1402         .open           = sched_autogroup_open,
1403         .read           = seq_read,
1404         .write          = sched_autogroup_write,
1405         .llseek         = seq_lseek,
1406         .release        = single_release,
1407 };
1408
1409 #endif /* CONFIG_SCHED_AUTOGROUP */
1410
1411 static ssize_t comm_write(struct file *file, const char __user *buf,
1412                                 size_t count, loff_t *offset)
1413 {
1414         struct inode *inode = file_inode(file);
1415         struct task_struct *p;
1416         char buffer[TASK_COMM_LEN];
1417         const size_t maxlen = sizeof(buffer) - 1;
1418
1419         memset(buffer, 0, sizeof(buffer));
1420         if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1421                 return -EFAULT;
1422
1423         p = get_proc_task(inode);
1424         if (!p)
1425                 return -ESRCH;
1426
1427         if (same_thread_group(current, p))
1428                 set_task_comm(p, buffer);
1429         else
1430                 count = -EINVAL;
1431
1432         put_task_struct(p);
1433
1434         return count;
1435 }
1436
1437 static int comm_show(struct seq_file *m, void *v)
1438 {
1439         struct inode *inode = m->private;
1440         struct task_struct *p;
1441
1442         p = get_proc_task(inode);
1443         if (!p)
1444                 return -ESRCH;
1445
1446         task_lock(p);
1447         seq_printf(m, "%s\n", p->comm);
1448         task_unlock(p);
1449
1450         put_task_struct(p);
1451
1452         return 0;
1453 }
1454
1455 static int comm_open(struct inode *inode, struct file *filp)
1456 {
1457         return single_open(filp, comm_show, inode);
1458 }
1459
1460 static const struct file_operations proc_pid_set_comm_operations = {
1461         .open           = comm_open,
1462         .read           = seq_read,
1463         .write          = comm_write,
1464         .llseek         = seq_lseek,
1465         .release        = single_release,
1466 };
1467
1468 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1469 {
1470         struct task_struct *task;
1471         struct mm_struct *mm;
1472         struct file *exe_file;
1473
1474         task = get_proc_task(dentry->d_inode);
1475         if (!task)
1476                 return -ENOENT;
1477         mm = get_task_mm(task);
1478         put_task_struct(task);
1479         if (!mm)
1480                 return -ENOENT;
1481         exe_file = get_mm_exe_file(mm);
1482         mmput(mm);
1483         if (exe_file) {
1484                 *exe_path = exe_file->f_path;
1485                 path_get(&exe_file->f_path);
1486                 fput(exe_file);
1487                 return 0;
1488         } else
1489                 return -ENOENT;
1490 }
1491
1492 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1493 {
1494         struct inode *inode = dentry->d_inode;
1495         struct path path;
1496         int error = -EACCES;
1497
1498         /* Are we allowed to snoop on the tasks file descriptors? */
1499         if (!proc_fd_access_allowed(inode))
1500                 goto out;
1501
1502         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1503         if (error)
1504                 goto out;
1505
1506         nd_jump_link(nd, &path);
1507         return NULL;
1508 out:
1509         return ERR_PTR(error);
1510 }
1511
1512 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1513 {
1514         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1515         char *pathname;
1516         int len;
1517
1518         if (!tmp)
1519                 return -ENOMEM;
1520
1521         pathname = d_path(path, tmp, PAGE_SIZE);
1522         len = PTR_ERR(pathname);
1523         if (IS_ERR(pathname))
1524                 goto out;
1525         len = tmp + PAGE_SIZE - 1 - pathname;
1526
1527         if (len > buflen)
1528                 len = buflen;
1529         if (copy_to_user(buffer, pathname, len))
1530                 len = -EFAULT;
1531  out:
1532         free_page((unsigned long)tmp);
1533         return len;
1534 }
1535
1536 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1537 {
1538         int error = -EACCES;
1539         struct inode *inode = dentry->d_inode;
1540         struct path path;
1541
1542         /* Are we allowed to snoop on the tasks file descriptors? */
1543         if (!proc_fd_access_allowed(inode))
1544                 goto out;
1545
1546         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1547         if (error)
1548                 goto out;
1549
1550         error = do_proc_readlink(&path, buffer, buflen);
1551         path_put(&path);
1552 out:
1553         return error;
1554 }
1555
1556 const struct inode_operations proc_pid_link_inode_operations = {
1557         .readlink       = proc_pid_readlink,
1558         .follow_link    = proc_pid_follow_link,
1559         .setattr        = proc_setattr,
1560 };
1561
1562
1563 /* building an inode */
1564
1565 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1566 {
1567         struct inode * inode;
1568         struct proc_inode *ei;
1569         const struct cred *cred;
1570
1571         /* We need a new inode */
1572
1573         inode = new_inode(sb);
1574         if (!inode)
1575                 goto out;
1576
1577         /* Common stuff */
1578         ei = PROC_I(inode);
1579         inode->i_ino = get_next_ino();
1580         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1581         inode->i_op = &proc_def_inode_operations;
1582
1583         /*
1584          * grab the reference to task.
1585          */
1586         ei->pid = get_task_pid(task, PIDTYPE_PID);
1587         if (!ei->pid)
1588                 goto out_unlock;
1589
1590         if (task_dumpable(task)) {
1591                 rcu_read_lock();
1592                 cred = __task_cred(task);
1593                 inode->i_uid = cred->euid;
1594                 inode->i_gid = cred->egid;
1595                 rcu_read_unlock();
1596         }
1597         security_task_to_inode(task, inode);
1598
1599 out:
1600         return inode;
1601
1602 out_unlock:
1603         iput(inode);
1604         return NULL;
1605 }
1606
1607 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1608 {
1609         struct inode *inode = dentry->d_inode;
1610         struct task_struct *task;
1611         const struct cred *cred;
1612         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1613
1614         generic_fillattr(inode, stat);
1615
1616         rcu_read_lock();
1617         stat->uid = GLOBAL_ROOT_UID;
1618         stat->gid = GLOBAL_ROOT_GID;
1619         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1620         if (task) {
1621                 if (!has_pid_permissions(pid, task, 2)) {
1622                         rcu_read_unlock();
1623                         /*
1624                          * This doesn't prevent learning whether PID exists,
1625                          * it only makes getattr() consistent with readdir().
1626                          */
1627                         return -ENOENT;
1628                 }
1629                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1630                     task_dumpable(task)) {
1631                         cred = __task_cred(task);
1632                         stat->uid = cred->euid;
1633                         stat->gid = cred->egid;
1634                 }
1635         }
1636         rcu_read_unlock();
1637         return 0;
1638 }
1639
1640 /* dentry stuff */
1641
1642 /*
1643  *      Exceptional case: normally we are not allowed to unhash a busy
1644  * directory. In this case, however, we can do it - no aliasing problems
1645  * due to the way we treat inodes.
1646  *
1647  * Rewrite the inode's ownerships here because the owning task may have
1648  * performed a setuid(), etc.
1649  *
1650  * Before the /proc/pid/status file was created the only way to read
1651  * the effective uid of a /process was to stat /proc/pid.  Reading
1652  * /proc/pid/status is slow enough that procps and other packages
1653  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1654  * made this apply to all per process world readable and executable
1655  * directories.
1656  */
1657 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1658 {
1659         struct inode *inode;
1660         struct task_struct *task;
1661         const struct cred *cred;
1662
1663         if (flags & LOOKUP_RCU)
1664                 return -ECHILD;
1665
1666         inode = dentry->d_inode;
1667         task = get_proc_task(inode);
1668
1669         if (task) {
1670                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1671                     task_dumpable(task)) {
1672                         rcu_read_lock();
1673                         cred = __task_cred(task);
1674                         inode->i_uid = cred->euid;
1675                         inode->i_gid = cred->egid;
1676                         rcu_read_unlock();
1677                 } else {
1678                         inode->i_uid = GLOBAL_ROOT_UID;
1679                         inode->i_gid = GLOBAL_ROOT_GID;
1680                 }
1681                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1682                 security_task_to_inode(task, inode);
1683                 put_task_struct(task);
1684                 return 1;
1685         }
1686         d_drop(dentry);
1687         return 0;
1688 }
1689
1690 int pid_delete_dentry(const struct dentry *dentry)
1691 {
1692         /* Is the task we represent dead?
1693          * If so, then don't put the dentry on the lru list,
1694          * kill it immediately.
1695          */
1696         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1697 }
1698
1699 const struct dentry_operations pid_dentry_operations =
1700 {
1701         .d_revalidate   = pid_revalidate,
1702         .d_delete       = pid_delete_dentry,
1703 };
1704
1705 /* Lookups */
1706
1707 /*
1708  * Fill a directory entry.
1709  *
1710  * If possible create the dcache entry and derive our inode number and
1711  * file type from dcache entry.
1712  *
1713  * Since all of the proc inode numbers are dynamically generated, the inode
1714  * numbers do not exist until the inode is cache.  This means creating the
1715  * the dcache entry in readdir is necessary to keep the inode numbers
1716  * reported by readdir in sync with the inode numbers reported
1717  * by stat.
1718  */
1719 int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1720         const char *name, int len,
1721         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1722 {
1723         struct dentry *child, *dir = filp->f_path.dentry;
1724         struct inode *inode;
1725         struct qstr qname;
1726         ino_t ino = 0;
1727         unsigned type = DT_UNKNOWN;
1728
1729         qname.name = name;
1730         qname.len  = len;
1731         qname.hash = full_name_hash(name, len);
1732
1733         child = d_lookup(dir, &qname);
1734         if (!child) {
1735                 struct dentry *new;
1736                 new = d_alloc(dir, &qname);
1737                 if (new) {
1738                         child = instantiate(dir->d_inode, new, task, ptr);
1739                         if (child)
1740                                 dput(new);
1741                         else
1742                                 child = new;
1743                 }
1744         }
1745         if (!child || IS_ERR(child) || !child->d_inode)
1746                 goto end_instantiate;
1747         inode = child->d_inode;
1748         if (inode) {
1749                 ino = inode->i_ino;
1750                 type = inode->i_mode >> 12;
1751         }
1752         dput(child);
1753 end_instantiate:
1754         if (!ino)
1755                 ino = find_inode_number(dir, &qname);
1756         if (!ino)
1757                 ino = 1;
1758         return filldir(dirent, name, len, filp->f_pos, ino, type);
1759 }
1760
1761 #ifdef CONFIG_CHECKPOINT_RESTORE
1762
1763 /*
1764  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1765  * which represent vma start and end addresses.
1766  */
1767 static int dname_to_vma_addr(struct dentry *dentry,
1768                              unsigned long *start, unsigned long *end)
1769 {
1770         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1771                 return -EINVAL;
1772
1773         return 0;
1774 }
1775
1776 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1777 {
1778         unsigned long vm_start, vm_end;
1779         bool exact_vma_exists = false;
1780         struct mm_struct *mm = NULL;
1781         struct task_struct *task;
1782         const struct cred *cred;
1783         struct inode *inode;
1784         int status = 0;
1785
1786         if (flags & LOOKUP_RCU)
1787                 return -ECHILD;
1788
1789         if (!capable(CAP_SYS_ADMIN)) {
1790                 status = -EPERM;
1791                 goto out_notask;
1792         }
1793
1794         inode = dentry->d_inode;
1795         task = get_proc_task(inode);
1796         if (!task)
1797                 goto out_notask;
1798
1799         mm = mm_access(task, PTRACE_MODE_READ);
1800         if (IS_ERR_OR_NULL(mm))
1801                 goto out;
1802
1803         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1804                 down_read(&mm->mmap_sem);
1805                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1806                 up_read(&mm->mmap_sem);
1807         }
1808
1809         mmput(mm);
1810
1811         if (exact_vma_exists) {
1812                 if (task_dumpable(task)) {
1813                         rcu_read_lock();
1814                         cred = __task_cred(task);
1815                         inode->i_uid = cred->euid;
1816                         inode->i_gid = cred->egid;
1817                         rcu_read_unlock();
1818                 } else {
1819                         inode->i_uid = GLOBAL_ROOT_UID;
1820                         inode->i_gid = GLOBAL_ROOT_GID;
1821                 }
1822                 security_task_to_inode(task, inode);
1823                 status = 1;
1824         }
1825
1826 out:
1827         put_task_struct(task);
1828
1829 out_notask:
1830         if (status <= 0)
1831                 d_drop(dentry);
1832
1833         return status;
1834 }
1835
1836 static const struct dentry_operations tid_map_files_dentry_operations = {
1837         .d_revalidate   = map_files_d_revalidate,
1838         .d_delete       = pid_delete_dentry,
1839 };
1840
1841 static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
1842 {
1843         unsigned long vm_start, vm_end;
1844         struct vm_area_struct *vma;
1845         struct task_struct *task;
1846         struct mm_struct *mm;
1847         int rc;
1848
1849         rc = -ENOENT;
1850         task = get_proc_task(dentry->d_inode);
1851         if (!task)
1852                 goto out;
1853
1854         mm = get_task_mm(task);
1855         put_task_struct(task);
1856         if (!mm)
1857                 goto out;
1858
1859         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1860         if (rc)
1861                 goto out_mmput;
1862
1863         down_read(&mm->mmap_sem);
1864         vma = find_exact_vma(mm, vm_start, vm_end);
1865         if (vma && vma->vm_file) {
1866                 *path = vma->vm_file->f_path;
1867                 path_get(path);
1868                 rc = 0;
1869         }
1870         up_read(&mm->mmap_sem);
1871
1872 out_mmput:
1873         mmput(mm);
1874 out:
1875         return rc;
1876 }
1877
1878 struct map_files_info {
1879         fmode_t         mode;
1880         unsigned long   len;
1881         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1882 };
1883
1884 static struct dentry *
1885 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
1886                            struct task_struct *task, const void *ptr)
1887 {
1888         fmode_t mode = (fmode_t)(unsigned long)ptr;
1889         struct proc_inode *ei;
1890         struct inode *inode;
1891
1892         inode = proc_pid_make_inode(dir->i_sb, task);
1893         if (!inode)
1894                 return ERR_PTR(-ENOENT);
1895
1896         ei = PROC_I(inode);
1897         ei->op.proc_get_link = proc_map_files_get_link;
1898
1899         inode->i_op = &proc_pid_link_inode_operations;
1900         inode->i_size = 64;
1901         inode->i_mode = S_IFLNK;
1902
1903         if (mode & FMODE_READ)
1904                 inode->i_mode |= S_IRUSR;
1905         if (mode & FMODE_WRITE)
1906                 inode->i_mode |= S_IWUSR;
1907
1908         d_set_d_op(dentry, &tid_map_files_dentry_operations);
1909         d_add(dentry, inode);
1910
1911         return NULL;
1912 }
1913
1914 static struct dentry *proc_map_files_lookup(struct inode *dir,
1915                 struct dentry *dentry, unsigned int flags)
1916 {
1917         unsigned long vm_start, vm_end;
1918         struct vm_area_struct *vma;
1919         struct task_struct *task;
1920         struct dentry *result;
1921         struct mm_struct *mm;
1922
1923         result = ERR_PTR(-EPERM);
1924         if (!capable(CAP_SYS_ADMIN))
1925                 goto out;
1926
1927         result = ERR_PTR(-ENOENT);
1928         task = get_proc_task(dir);
1929         if (!task)
1930                 goto out;
1931
1932         result = ERR_PTR(-EACCES);
1933         if (!ptrace_may_access(task, PTRACE_MODE_READ))
1934                 goto out_put_task;
1935
1936         result = ERR_PTR(-ENOENT);
1937         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
1938                 goto out_put_task;
1939
1940         mm = get_task_mm(task);
1941         if (!mm)
1942                 goto out_put_task;
1943
1944         down_read(&mm->mmap_sem);
1945         vma = find_exact_vma(mm, vm_start, vm_end);
1946         if (!vma)
1947                 goto out_no_vma;
1948
1949         if (vma->vm_file)
1950                 result = proc_map_files_instantiate(dir, dentry, task,
1951                                 (void *)(unsigned long)vma->vm_file->f_mode);
1952
1953 out_no_vma:
1954         up_read(&mm->mmap_sem);
1955         mmput(mm);
1956 out_put_task:
1957         put_task_struct(task);
1958 out:
1959         return result;
1960 }
1961
1962 static const struct inode_operations proc_map_files_inode_operations = {
1963         .lookup         = proc_map_files_lookup,
1964         .permission     = proc_fd_permission,
1965         .setattr        = proc_setattr,
1966 };
1967
1968 static int
1969 proc_map_files_readdir(struct file *filp, void *dirent, filldir_t filldir)
1970 {
1971         struct dentry *dentry = filp->f_path.dentry;
1972         struct inode *inode = dentry->d_inode;
1973         struct vm_area_struct *vma;
1974         struct task_struct *task;
1975         struct mm_struct *mm;
1976         ino_t ino;
1977         int ret;
1978
1979         ret = -EPERM;
1980         if (!capable(CAP_SYS_ADMIN))
1981                 goto out;
1982
1983         ret = -ENOENT;
1984         task = get_proc_task(inode);
1985         if (!task)
1986                 goto out;
1987
1988         ret = -EACCES;
1989         if (!ptrace_may_access(task, PTRACE_MODE_READ))
1990                 goto out_put_task;
1991
1992         ret = 0;
1993         switch (filp->f_pos) {
1994         case 0:
1995                 ino = inode->i_ino;
1996                 if (filldir(dirent, ".", 1, 0, ino, DT_DIR) < 0)
1997                         goto out_put_task;
1998                 filp->f_pos++;
1999         case 1:
2000                 ino = parent_ino(dentry);
2001                 if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
2002                         goto out_put_task;
2003                 filp->f_pos++;
2004         default:
2005         {
2006                 unsigned long nr_files, pos, i;
2007                 struct flex_array *fa = NULL;
2008                 struct map_files_info info;
2009                 struct map_files_info *p;
2010
2011                 mm = get_task_mm(task);
2012                 if (!mm)
2013                         goto out_put_task;
2014                 down_read(&mm->mmap_sem);
2015
2016                 nr_files = 0;
2017
2018                 /*
2019                  * We need two passes here:
2020                  *
2021                  *  1) Collect vmas of mapped files with mmap_sem taken
2022                  *  2) Release mmap_sem and instantiate entries
2023                  *
2024                  * otherwise we get lockdep complained, since filldir()
2025                  * routine might require mmap_sem taken in might_fault().
2026                  */
2027
2028                 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2029                         if (vma->vm_file && ++pos > filp->f_pos)
2030                                 nr_files++;
2031                 }
2032
2033                 if (nr_files) {
2034                         fa = flex_array_alloc(sizeof(info), nr_files,
2035                                                 GFP_KERNEL);
2036                         if (!fa || flex_array_prealloc(fa, 0, nr_files,
2037                                                         GFP_KERNEL)) {
2038                                 ret = -ENOMEM;
2039                                 if (fa)
2040                                         flex_array_free(fa);
2041                                 up_read(&mm->mmap_sem);
2042                                 mmput(mm);
2043                                 goto out_put_task;
2044                         }
2045                         for (i = 0, vma = mm->mmap, pos = 2; vma;
2046                                         vma = vma->vm_next) {
2047                                 if (!vma->vm_file)
2048                                         continue;
2049                                 if (++pos <= filp->f_pos)
2050                                         continue;
2051
2052                                 info.mode = vma->vm_file->f_mode;
2053                                 info.len = snprintf(info.name,
2054                                                 sizeof(info.name), "%lx-%lx",
2055                                                 vma->vm_start, vma->vm_end);
2056                                 if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2057                                         BUG();
2058                         }
2059                 }
2060                 up_read(&mm->mmap_sem);
2061
2062                 for (i = 0; i < nr_files; i++) {
2063                         p = flex_array_get(fa, i);
2064                         ret = proc_fill_cache(filp, dirent, filldir,
2065                                               p->name, p->len,
2066                                               proc_map_files_instantiate,
2067                                               task,
2068                                               (void *)(unsigned long)p->mode);
2069                         if (ret)
2070                                 break;
2071                         filp->f_pos++;
2072                 }
2073                 if (fa)
2074                         flex_array_free(fa);
2075                 mmput(mm);
2076         }
2077         }
2078
2079 out_put_task:
2080         put_task_struct(task);
2081 out:
2082         return ret;
2083 }
2084
2085 static const struct file_operations proc_map_files_operations = {
2086         .read           = generic_read_dir,
2087         .readdir        = proc_map_files_readdir,
2088         .llseek         = default_llseek,
2089 };
2090
2091 struct timers_private {
2092         struct pid *pid;
2093         struct task_struct *task;
2094         struct sighand_struct *sighand;
2095         struct pid_namespace *ns;
2096         unsigned long flags;
2097 };
2098
2099 static void *timers_start(struct seq_file *m, loff_t *pos)
2100 {
2101         struct timers_private *tp = m->private;
2102
2103         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2104         if (!tp->task)
2105                 return ERR_PTR(-ESRCH);
2106
2107         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2108         if (!tp->sighand)
2109                 return ERR_PTR(-ESRCH);
2110
2111         return seq_list_start(&tp->task->signal->posix_timers, *pos);
2112 }
2113
2114 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2115 {
2116         struct timers_private *tp = m->private;
2117         return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2118 }
2119
2120 static void timers_stop(struct seq_file *m, void *v)
2121 {
2122         struct timers_private *tp = m->private;
2123
2124         if (tp->sighand) {
2125                 unlock_task_sighand(tp->task, &tp->flags);
2126                 tp->sighand = NULL;
2127         }
2128
2129         if (tp->task) {
2130                 put_task_struct(tp->task);
2131                 tp->task = NULL;
2132         }
2133 }
2134
2135 static int show_timer(struct seq_file *m, void *v)
2136 {
2137         struct k_itimer *timer;
2138         struct timers_private *tp = m->private;
2139         int notify;
2140         static char *nstr[] = {
2141                 [SIGEV_SIGNAL] = "signal",
2142                 [SIGEV_NONE] = "none",
2143                 [SIGEV_THREAD] = "thread",
2144         };
2145
2146         timer = list_entry((struct list_head *)v, struct k_itimer, list);
2147         notify = timer->it_sigev_notify;
2148
2149         seq_printf(m, "ID: %d\n", timer->it_id);
2150         seq_printf(m, "signal: %d/%p\n", timer->sigq->info.si_signo,
2151                         timer->sigq->info.si_value.sival_ptr);
2152         seq_printf(m, "notify: %s/%s.%d\n",
2153                 nstr[notify & ~SIGEV_THREAD_ID],
2154                 (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2155                 pid_nr_ns(timer->it_pid, tp->ns));
2156         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2157
2158         return 0;
2159 }
2160
2161 static const struct seq_operations proc_timers_seq_ops = {
2162         .start  = timers_start,
2163         .next   = timers_next,
2164         .stop   = timers_stop,
2165         .show   = show_timer,
2166 };
2167
2168 static int proc_timers_open(struct inode *inode, struct file *file)
2169 {
2170         struct timers_private *tp;
2171
2172         tp = __seq_open_private(file, &proc_timers_seq_ops,
2173                         sizeof(struct timers_private));
2174         if (!tp)
2175                 return -ENOMEM;
2176
2177         tp->pid = proc_pid(inode);
2178         tp->ns = inode->i_sb->s_fs_info;
2179         return 0;
2180 }
2181
2182 static const struct file_operations proc_timers_operations = {
2183         .open           = proc_timers_open,
2184         .read           = seq_read,
2185         .llseek         = seq_lseek,
2186         .release        = seq_release_private,
2187 };
2188 #endif /* CONFIG_CHECKPOINT_RESTORE */
2189
2190 static struct dentry *proc_pident_instantiate(struct inode *dir,
2191         struct dentry *dentry, struct task_struct *task, const void *ptr)
2192 {
2193         const struct pid_entry *p = ptr;
2194         struct inode *inode;
2195         struct proc_inode *ei;
2196         struct dentry *error = ERR_PTR(-ENOENT);
2197
2198         inode = proc_pid_make_inode(dir->i_sb, task);
2199         if (!inode)
2200                 goto out;
2201
2202         ei = PROC_I(inode);
2203         inode->i_mode = p->mode;
2204         if (S_ISDIR(inode->i_mode))
2205                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2206         if (p->iop)
2207                 inode->i_op = p->iop;
2208         if (p->fop)
2209                 inode->i_fop = p->fop;
2210         ei->op = p->op;
2211         d_set_d_op(dentry, &pid_dentry_operations);
2212         d_add(dentry, inode);
2213         /* Close the race of the process dying before we return the dentry */
2214         if (pid_revalidate(dentry, 0))
2215                 error = NULL;
2216 out:
2217         return error;
2218 }
2219
2220 static struct dentry *proc_pident_lookup(struct inode *dir,
2221                                          struct dentry *dentry,
2222                                          const struct pid_entry *ents,
2223                                          unsigned int nents)
2224 {
2225         struct dentry *error;
2226         struct task_struct *task = get_proc_task(dir);
2227         const struct pid_entry *p, *last;
2228
2229         error = ERR_PTR(-ENOENT);
2230
2231         if (!task)
2232                 goto out_no_task;
2233
2234         /*
2235          * Yes, it does not scale. And it should not. Don't add
2236          * new entries into /proc/<tgid>/ without very good reasons.
2237          */
2238         last = &ents[nents - 1];
2239         for (p = ents; p <= last; p++) {
2240                 if (p->len != dentry->d_name.len)
2241                         continue;
2242                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2243                         break;
2244         }
2245         if (p > last)
2246                 goto out;
2247
2248         error = proc_pident_instantiate(dir, dentry, task, p);
2249 out:
2250         put_task_struct(task);
2251 out_no_task:
2252         return error;
2253 }
2254
2255 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2256         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2257 {
2258         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2259                                 proc_pident_instantiate, task, p);
2260 }
2261
2262 static int proc_pident_readdir(struct file *filp,
2263                 void *dirent, filldir_t filldir,
2264                 const struct pid_entry *ents, unsigned int nents)
2265 {
2266         int i;
2267         struct dentry *dentry = filp->f_path.dentry;
2268         struct inode *inode = dentry->d_inode;
2269         struct task_struct *task = get_proc_task(inode);
2270         const struct pid_entry *p, *last;
2271         ino_t ino;
2272         int ret;
2273
2274         ret = -ENOENT;
2275         if (!task)
2276                 goto out_no_task;
2277
2278         ret = 0;
2279         i = filp->f_pos;
2280         switch (i) {
2281         case 0:
2282                 ino = inode->i_ino;
2283                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2284                         goto out;
2285                 i++;
2286                 filp->f_pos++;
2287                 /* fall through */
2288         case 1:
2289                 ino = parent_ino(dentry);
2290                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2291                         goto out;
2292                 i++;
2293                 filp->f_pos++;
2294                 /* fall through */
2295         default:
2296                 i -= 2;
2297                 if (i >= nents) {
2298                         ret = 1;
2299                         goto out;
2300                 }
2301                 p = ents + i;
2302                 last = &ents[nents - 1];
2303                 while (p <= last) {
2304                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2305                                 goto out;
2306                         filp->f_pos++;
2307                         p++;
2308                 }
2309         }
2310
2311         ret = 1;
2312 out:
2313         put_task_struct(task);
2314 out_no_task:
2315         return ret;
2316 }
2317
2318 #ifdef CONFIG_SECURITY
2319 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2320                                   size_t count, loff_t *ppos)
2321 {
2322         struct inode * inode = file_inode(file);
2323         char *p = NULL;
2324         ssize_t length;
2325         struct task_struct *task = get_proc_task(inode);
2326
2327         if (!task)
2328                 return -ESRCH;
2329
2330         length = security_getprocattr(task,
2331                                       (char*)file->f_path.dentry->d_name.name,
2332                                       &p);
2333         put_task_struct(task);
2334         if (length > 0)
2335                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2336         kfree(p);
2337         return length;
2338 }
2339
2340 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2341                                    size_t count, loff_t *ppos)
2342 {
2343         struct inode * inode = file_inode(file);
2344         char *page;
2345         ssize_t length;
2346         struct task_struct *task = get_proc_task(inode);
2347
2348         length = -ESRCH;
2349         if (!task)
2350                 goto out_no_task;
2351         if (count > PAGE_SIZE)
2352                 count = PAGE_SIZE;
2353
2354         /* No partial writes. */
2355         length = -EINVAL;
2356         if (*ppos != 0)
2357                 goto out;
2358
2359         length = -ENOMEM;
2360         page = (char*)__get_free_page(GFP_TEMPORARY);
2361         if (!page)
2362                 goto out;
2363
2364         length = -EFAULT;
2365         if (copy_from_user(page, buf, count))
2366                 goto out_free;
2367
2368         /* Guard against adverse ptrace interaction */
2369         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2370         if (length < 0)
2371                 goto out_free;
2372
2373         length = security_setprocattr(task,
2374                                       (char*)file->f_path.dentry->d_name.name,
2375                                       (void*)page, count);
2376         mutex_unlock(&task->signal->cred_guard_mutex);
2377 out_free:
2378         free_page((unsigned long) page);
2379 out:
2380         put_task_struct(task);
2381 out_no_task:
2382         return length;
2383 }
2384
2385 static const struct file_operations proc_pid_attr_operations = {
2386         .read           = proc_pid_attr_read,
2387         .write          = proc_pid_attr_write,
2388         .llseek         = generic_file_llseek,
2389 };
2390
2391 static const struct pid_entry attr_dir_stuff[] = {
2392         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2393         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2394         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2395         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2396         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2397         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2398 };
2399
2400 static int proc_attr_dir_readdir(struct file * filp,
2401                              void * dirent, filldir_t filldir)
2402 {
2403         return proc_pident_readdir(filp,dirent,filldir,
2404                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2405 }
2406
2407 static const struct file_operations proc_attr_dir_operations = {
2408         .read           = generic_read_dir,
2409         .readdir        = proc_attr_dir_readdir,
2410         .llseek         = default_llseek,
2411 };
2412
2413 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2414                                 struct dentry *dentry, unsigned int flags)
2415 {
2416         return proc_pident_lookup(dir, dentry,
2417                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2418 }
2419
2420 static const struct inode_operations proc_attr_dir_inode_operations = {
2421         .lookup         = proc_attr_dir_lookup,
2422         .getattr        = pid_getattr,
2423         .setattr        = proc_setattr,
2424 };
2425
2426 #endif
2427
2428 #ifdef CONFIG_ELF_CORE
2429 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2430                                          size_t count, loff_t *ppos)
2431 {
2432         struct task_struct *task = get_proc_task(file_inode(file));
2433         struct mm_struct *mm;
2434         char buffer[PROC_NUMBUF];
2435         size_t len;
2436         int ret;
2437
2438         if (!task)
2439                 return -ESRCH;
2440
2441         ret = 0;
2442         mm = get_task_mm(task);
2443         if (mm) {
2444                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2445                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2446                                 MMF_DUMP_FILTER_SHIFT));
2447                 mmput(mm);
2448                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2449         }
2450
2451         put_task_struct(task);
2452
2453         return ret;
2454 }
2455
2456 static ssize_t proc_coredump_filter_write(struct file *file,
2457                                           const char __user *buf,
2458                                           size_t count,
2459                                           loff_t *ppos)
2460 {
2461         struct task_struct *task;
2462         struct mm_struct *mm;
2463         char buffer[PROC_NUMBUF], *end;
2464         unsigned int val;
2465         int ret;
2466         int i;
2467         unsigned long mask;
2468
2469         ret = -EFAULT;
2470         memset(buffer, 0, sizeof(buffer));
2471         if (count > sizeof(buffer) - 1)
2472                 count = sizeof(buffer) - 1;
2473         if (copy_from_user(buffer, buf, count))
2474                 goto out_no_task;
2475
2476         ret = -EINVAL;
2477         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2478         if (*end == '\n')
2479                 end++;
2480         if (end - buffer == 0)
2481                 goto out_no_task;
2482
2483         ret = -ESRCH;
2484         task = get_proc_task(file_inode(file));
2485         if (!task)
2486                 goto out_no_task;
2487
2488         ret = end - buffer;
2489         mm = get_task_mm(task);
2490         if (!mm)
2491                 goto out_no_mm;
2492
2493         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2494                 if (val & mask)
2495                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2496                 else
2497                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2498         }
2499
2500         mmput(mm);
2501  out_no_mm:
2502         put_task_struct(task);
2503  out_no_task:
2504         return ret;
2505 }
2506
2507 static const struct file_operations proc_coredump_filter_operations = {
2508         .read           = proc_coredump_filter_read,
2509         .write          = proc_coredump_filter_write,
2510         .llseek         = generic_file_llseek,
2511 };
2512 #endif
2513
2514 #ifdef CONFIG_TASK_IO_ACCOUNTING
2515 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2516 {
2517         struct task_io_accounting acct = task->ioac;
2518         unsigned long flags;
2519         int result;
2520
2521         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2522         if (result)
2523                 return result;
2524
2525         if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2526                 result = -EACCES;
2527                 goto out_unlock;
2528         }
2529
2530         if (whole && lock_task_sighand(task, &flags)) {
2531                 struct task_struct *t = task;
2532
2533                 task_io_accounting_add(&acct, &task->signal->ioac);
2534                 while_each_thread(task, t)
2535                         task_io_accounting_add(&acct, &t->ioac);
2536
2537                 unlock_task_sighand(task, &flags);
2538         }
2539         result = sprintf(buffer,
2540                         "rchar: %llu\n"
2541                         "wchar: %llu\n"
2542                         "syscr: %llu\n"
2543                         "syscw: %llu\n"
2544                         "read_bytes: %llu\n"
2545                         "write_bytes: %llu\n"
2546                         "cancelled_write_bytes: %llu\n",
2547                         (unsigned long long)acct.rchar,
2548                         (unsigned long long)acct.wchar,
2549                         (unsigned long long)acct.syscr,
2550                         (unsigned long long)acct.syscw,
2551                         (unsigned long long)acct.read_bytes,
2552                         (unsigned long long)acct.write_bytes,
2553                         (unsigned long long)acct.cancelled_write_bytes);
2554 out_unlock:
2555         mutex_unlock(&task->signal->cred_guard_mutex);
2556         return result;
2557 }
2558
2559 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2560 {
2561         return do_io_accounting(task, buffer, 0);
2562 }
2563
2564 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2565 {
2566         return do_io_accounting(task, buffer, 1);
2567 }
2568 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2569
2570 #ifdef CONFIG_USER_NS
2571 static int proc_id_map_open(struct inode *inode, struct file *file,
2572         struct seq_operations *seq_ops)
2573 {
2574         struct user_namespace *ns = NULL;
2575         struct task_struct *task;
2576         struct seq_file *seq;
2577         int ret = -EINVAL;
2578
2579         task = get_proc_task(inode);
2580         if (task) {
2581                 rcu_read_lock();
2582                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2583                 rcu_read_unlock();
2584                 put_task_struct(task);
2585         }
2586         if (!ns)
2587                 goto err;
2588
2589         ret = seq_open(file, seq_ops);
2590         if (ret)
2591                 goto err_put_ns;
2592
2593         seq = file->private_data;
2594         seq->private = ns;
2595
2596         return 0;
2597 err_put_ns:
2598         put_user_ns(ns);
2599 err:
2600         return ret;
2601 }
2602
2603 static int proc_id_map_release(struct inode *inode, struct file *file)
2604 {
2605         struct seq_file *seq = file->private_data;
2606         struct user_namespace *ns = seq->private;
2607         put_user_ns(ns);
2608         return seq_release(inode, file);
2609 }
2610
2611 static int proc_uid_map_open(struct inode *inode, struct file *file)
2612 {
2613         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2614 }
2615
2616 static int proc_gid_map_open(struct inode *inode, struct file *file)
2617 {
2618         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2619 }
2620
2621 static int proc_projid_map_open(struct inode *inode, struct file *file)
2622 {
2623         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2624 }
2625
2626 static const struct file_operations proc_uid_map_operations = {
2627         .open           = proc_uid_map_open,
2628         .write          = proc_uid_map_write,
2629         .read           = seq_read,
2630         .llseek         = seq_lseek,
2631         .release        = proc_id_map_release,
2632 };
2633
2634 static const struct file_operations proc_gid_map_operations = {
2635         .open           = proc_gid_map_open,
2636         .write          = proc_gid_map_write,
2637         .read           = seq_read,
2638         .llseek         = seq_lseek,
2639         .release        = proc_id_map_release,
2640 };
2641
2642 static const struct file_operations proc_projid_map_operations = {
2643         .open           = proc_projid_map_open,
2644         .write          = proc_projid_map_write,
2645         .read           = seq_read,
2646         .llseek         = seq_lseek,
2647         .release        = proc_id_map_release,
2648 };
2649 #endif /* CONFIG_USER_NS */
2650
2651 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2652                                 struct pid *pid, struct task_struct *task)
2653 {
2654         int err = lock_trace(task);
2655         if (!err) {
2656                 seq_printf(m, "%08x\n", task->personality);
2657                 unlock_trace(task);
2658         }
2659         return err;
2660 }
2661
2662 /*
2663  * Thread groups
2664  */
2665 static const struct file_operations proc_task_operations;
2666 static const struct inode_operations proc_task_inode_operations;
2667
2668 static const struct pid_entry tgid_base_stuff[] = {
2669         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2670         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2671 #ifdef CONFIG_CHECKPOINT_RESTORE
2672         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2673 #endif
2674         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2675         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2676 #ifdef CONFIG_NET
2677         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2678 #endif
2679         REG("environ",    S_IRUSR, proc_environ_operations),
2680         INF("auxv",       S_IRUSR, proc_pid_auxv),
2681         ONE("status",     S_IRUGO, proc_pid_status),
2682         ONE("personality", S_IRUGO, proc_pid_personality),
2683         INF("limits",     S_IRUGO, proc_pid_limits),
2684 #ifdef CONFIG_SCHED_DEBUG
2685         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2686 #endif
2687 #ifdef CONFIG_SCHED_AUTOGROUP
2688         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2689 #endif
2690         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2691 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2692         INF("syscall",    S_IRUGO, proc_pid_syscall),
2693 #endif
2694         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2695         ONE("stat",       S_IRUGO, proc_tgid_stat),
2696         ONE("statm",      S_IRUGO, proc_pid_statm),
2697         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2698 #ifdef CONFIG_NUMA
2699         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2700 #endif
2701         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2702         LNK("cwd",        proc_cwd_link),
2703         LNK("root",       proc_root_link),
2704         LNK("exe",        proc_exe_link),
2705         REG("mounts",     S_IRUGO, proc_mounts_operations),
2706         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2707         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2708 #ifdef CONFIG_PROC_PAGE_MONITOR
2709         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2710         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2711         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
2712 #endif
2713 #ifdef CONFIG_SECURITY
2714         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2715 #endif
2716 #ifdef CONFIG_KALLSYMS
2717         INF("wchan",      S_IRUGO, proc_pid_wchan),
2718 #endif
2719 #ifdef CONFIG_STACKTRACE
2720         ONE("stack",      S_IRUGO, proc_pid_stack),
2721 #endif
2722 #ifdef CONFIG_SCHEDSTATS
2723         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
2724 #endif
2725 #ifdef CONFIG_LATENCYTOP
2726         REG("latency",  S_IRUGO, proc_lstats_operations),
2727 #endif
2728 #ifdef CONFIG_PROC_PID_CPUSET
2729         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2730 #endif
2731 #ifdef CONFIG_CGROUPS
2732         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2733 #endif
2734         INF("oom_score",  S_IRUGO, proc_oom_score),
2735         ANDROID("oom_adj", S_IRUGO|S_IWUSR, oom_adj),
2736         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2737 #ifdef CONFIG_AUDITSYSCALL
2738         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2739         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2740 #endif
2741 #ifdef CONFIG_FAULT_INJECTION
2742         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2743 #endif
2744 #ifdef CONFIG_ELF_CORE
2745         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2746 #endif
2747 #ifdef CONFIG_TASK_IO_ACCOUNTING
2748         INF("io",       S_IRUSR, proc_tgid_io_accounting),
2749 #endif
2750 #ifdef CONFIG_HARDWALL
2751         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
2752 #endif
2753 #ifdef CONFIG_USER_NS
2754         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2755         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2756         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2757 #endif
2758 #ifdef CONFIG_CHECKPOINT_RESTORE
2759         REG("timers",     S_IRUGO, proc_timers_operations),
2760 #endif
2761 };
2762
2763 static int proc_tgid_base_readdir(struct file * filp,
2764                              void * dirent, filldir_t filldir)
2765 {
2766         return proc_pident_readdir(filp,dirent,filldir,
2767                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2768 }
2769
2770 static const struct file_operations proc_tgid_base_operations = {
2771         .read           = generic_read_dir,
2772         .readdir        = proc_tgid_base_readdir,
2773         .llseek         = default_llseek,
2774 };
2775
2776 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2777 {
2778         return proc_pident_lookup(dir, dentry,
2779                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2780 }
2781
2782 static const struct inode_operations proc_tgid_base_inode_operations = {
2783         .lookup         = proc_tgid_base_lookup,
2784         .getattr        = pid_getattr,
2785         .setattr        = proc_setattr,
2786         .permission     = proc_pid_permission,
2787 };
2788
2789 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2790 {
2791         struct dentry *dentry, *leader, *dir;
2792         char buf[PROC_NUMBUF];
2793         struct qstr name;
2794
2795         name.name = buf;
2796         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2797         /* no ->d_hash() rejects on procfs */
2798         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2799         if (dentry) {
2800                 shrink_dcache_parent(dentry);
2801                 d_drop(dentry);
2802                 dput(dentry);
2803         }
2804
2805         name.name = buf;
2806         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2807         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2808         if (!leader)
2809                 goto out;
2810
2811         name.name = "task";
2812         name.len = strlen(name.name);
2813         dir = d_hash_and_lookup(leader, &name);
2814         if (!dir)
2815                 goto out_put_leader;
2816
2817         name.name = buf;
2818         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2819         dentry = d_hash_and_lookup(dir, &name);
2820         if (dentry) {
2821                 shrink_dcache_parent(dentry);
2822                 d_drop(dentry);
2823                 dput(dentry);
2824         }
2825
2826         dput(dir);
2827 out_put_leader:
2828         dput(leader);
2829 out:
2830         return;
2831 }
2832
2833 /**
2834  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2835  * @task: task that should be flushed.
2836  *
2837  * When flushing dentries from proc, one needs to flush them from global
2838  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2839  * in. This call is supposed to do all of this job.
2840  *
2841  * Looks in the dcache for
2842  * /proc/@pid
2843  * /proc/@tgid/task/@pid
2844  * if either directory is present flushes it and all of it'ts children
2845  * from the dcache.
2846  *
2847  * It is safe and reasonable to cache /proc entries for a task until
2848  * that task exits.  After that they just clog up the dcache with
2849  * useless entries, possibly causing useful dcache entries to be
2850  * flushed instead.  This routine is proved to flush those useless
2851  * dcache entries at process exit time.
2852  *
2853  * NOTE: This routine is just an optimization so it does not guarantee
2854  *       that no dcache entries will exist at process exit time it
2855  *       just makes it very unlikely that any will persist.
2856  */
2857
2858 void proc_flush_task(struct task_struct *task)
2859 {
2860         int i;
2861         struct pid *pid, *tgid;
2862         struct upid *upid;
2863
2864         pid = task_pid(task);
2865         tgid = task_tgid(task);
2866
2867         for (i = 0; i <= pid->level; i++) {
2868                 upid = &pid->numbers[i];
2869                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2870                                         tgid->numbers[i].nr);
2871         }
2872 }
2873
2874 static struct dentry *proc_pid_instantiate(struct inode *dir,
2875                                            struct dentry * dentry,
2876                                            struct task_struct *task, const void *ptr)
2877 {
2878         struct dentry *error = ERR_PTR(-ENOENT);
2879         struct inode *inode;
2880
2881         inode = proc_pid_make_inode(dir->i_sb, task);
2882         if (!inode)
2883                 goto out;
2884
2885         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2886         inode->i_op = &proc_tgid_base_inode_operations;
2887         inode->i_fop = &proc_tgid_base_operations;
2888         inode->i_flags|=S_IMMUTABLE;
2889
2890         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
2891                                                   ARRAY_SIZE(tgid_base_stuff)));
2892
2893         d_set_d_op(dentry, &pid_dentry_operations);
2894
2895         d_add(dentry, inode);
2896         /* Close the race of the process dying before we return the dentry */
2897         if (pid_revalidate(dentry, 0))
2898                 error = NULL;
2899 out:
2900         return error;
2901 }
2902
2903 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
2904 {
2905         struct dentry *result = NULL;
2906         struct task_struct *task;
2907         unsigned tgid;
2908         struct pid_namespace *ns;
2909
2910         tgid = name_to_int(dentry);
2911         if (tgid == ~0U)
2912                 goto out;
2913
2914         ns = dentry->d_sb->s_fs_info;
2915         rcu_read_lock();
2916         task = find_task_by_pid_ns(tgid, ns);
2917         if (task)
2918                 get_task_struct(task);
2919         rcu_read_unlock();
2920         if (!task)
2921                 goto out;
2922
2923         result = proc_pid_instantiate(dir, dentry, task, NULL);
2924         put_task_struct(task);
2925 out:
2926         return result;
2927 }
2928
2929 /*
2930  * Find the first task with tgid >= tgid
2931  *
2932  */
2933 struct tgid_iter {
2934         unsigned int tgid;
2935         struct task_struct *task;
2936 };
2937 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2938 {
2939         struct pid *pid;
2940
2941         if (iter.task)
2942                 put_task_struct(iter.task);
2943         rcu_read_lock();
2944 retry:
2945         iter.task = NULL;
2946         pid = find_ge_pid(iter.tgid, ns);
2947         if (pid) {
2948                 iter.tgid = pid_nr_ns(pid, ns);
2949                 iter.task = pid_task(pid, PIDTYPE_PID);
2950                 /* What we to know is if the pid we have find is the
2951                  * pid of a thread_group_leader.  Testing for task
2952                  * being a thread_group_leader is the obvious thing
2953                  * todo but there is a window when it fails, due to
2954                  * the pid transfer logic in de_thread.
2955                  *
2956                  * So we perform the straight forward test of seeing
2957                  * if the pid we have found is the pid of a thread
2958                  * group leader, and don't worry if the task we have
2959                  * found doesn't happen to be a thread group leader.
2960                  * As we don't care in the case of readdir.
2961                  */
2962                 if (!iter.task || !has_group_leader_pid(iter.task)) {
2963                         iter.tgid += 1;
2964                         goto retry;
2965                 }
2966                 get_task_struct(iter.task);
2967         }
2968         rcu_read_unlock();
2969         return iter;
2970 }
2971
2972 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 1)
2973
2974 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2975         struct tgid_iter iter)
2976 {
2977         char name[PROC_NUMBUF];
2978         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
2979         return proc_fill_cache(filp, dirent, filldir, name, len,
2980                                 proc_pid_instantiate, iter.task, NULL);
2981 }
2982
2983 static int fake_filldir(void *buf, const char *name, int namelen,
2984                         loff_t offset, u64 ino, unsigned d_type)
2985 {
2986         return 0;
2987 }
2988
2989 /* for the /proc/ directory itself, after non-process stuff has been done */
2990 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2991 {
2992         struct tgid_iter iter;
2993         struct pid_namespace *ns;
2994         filldir_t __filldir;
2995         loff_t pos = filp->f_pos;
2996
2997         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
2998                 goto out;
2999
3000         if (pos == TGID_OFFSET - 1) {
3001                 if (proc_fill_cache(filp, dirent, filldir, "self", 4,
3002                                         NULL, NULL, NULL) < 0)
3003                         goto out;
3004                 iter.tgid = 0;
3005         } else {
3006                 iter.tgid = pos - TGID_OFFSET;
3007         }
3008         iter.task = NULL;
3009         ns = filp->f_dentry->d_sb->s_fs_info;
3010         for (iter = next_tgid(ns, iter);
3011              iter.task;
3012              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3013                 if (has_pid_permissions(ns, iter.task, 2))
3014                         __filldir = filldir;
3015                 else
3016                         __filldir = fake_filldir;
3017
3018                 filp->f_pos = iter.tgid + TGID_OFFSET;
3019                 if (proc_pid_fill_cache(filp, dirent, __filldir, iter) < 0) {
3020                         put_task_struct(iter.task);
3021                         goto out;
3022                 }
3023         }
3024         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
3025 out:
3026         return 0;
3027 }
3028
3029 /*
3030  * Tasks
3031  */
3032 static const struct pid_entry tid_base_stuff[] = {
3033         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3034         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3035         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3036         REG("environ",   S_IRUSR, proc_environ_operations),
3037         INF("auxv",      S_IRUSR, proc_pid_auxv),
3038         ONE("status",    S_IRUGO, proc_pid_status),
3039         ONE("personality", S_IRUGO, proc_pid_personality),
3040         INF("limits",    S_IRUGO, proc_pid_limits),
3041 #ifdef CONFIG_SCHED_DEBUG
3042         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3043 #endif
3044         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3045 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3046         INF("syscall",   S_IRUGO, proc_pid_syscall),
3047 #endif
3048         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
3049         ONE("stat",      S_IRUGO, proc_tid_stat),
3050         ONE("statm",     S_IRUGO, proc_pid_statm),
3051         REG("maps",      S_IRUGO, proc_tid_maps_operations),
3052 #ifdef CONFIG_CHECKPOINT_RESTORE
3053         REG("children",  S_IRUGO, proc_tid_children_operations),
3054 #endif
3055 #ifdef CONFIG_NUMA
3056         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3057 #endif
3058         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3059         LNK("cwd",       proc_cwd_link),
3060         LNK("root",      proc_root_link),
3061         LNK("exe",       proc_exe_link),
3062         REG("mounts",    S_IRUGO, proc_mounts_operations),
3063         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3064 #ifdef CONFIG_PROC_PAGE_MONITOR
3065         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3066         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3067         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
3068 #endif
3069 #ifdef CONFIG_SECURITY
3070         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3071 #endif
3072 #ifdef CONFIG_KALLSYMS
3073         INF("wchan",     S_IRUGO, proc_pid_wchan),
3074 #endif
3075 #ifdef CONFIG_STACKTRACE
3076         ONE("stack",      S_IRUGO, proc_pid_stack),
3077 #endif
3078 #ifdef CONFIG_SCHEDSTATS
3079         INF("schedstat", S_IRUGO, proc_pid_schedstat),
3080 #endif
3081 #ifdef CONFIG_LATENCYTOP
3082         REG("latency",  S_IRUGO, proc_lstats_operations),
3083 #endif
3084 #ifdef CONFIG_PROC_PID_CPUSET
3085         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
3086 #endif
3087 #ifdef CONFIG_CGROUPS
3088         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3089 #endif
3090         INF("oom_score", S_IRUGO, proc_oom_score),
3091         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3092         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3093 #ifdef CONFIG_AUDITSYSCALL
3094         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3095         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3096 #endif
3097 #ifdef CONFIG_FAULT_INJECTION
3098         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3099 #endif
3100 #ifdef CONFIG_TASK_IO_ACCOUNTING
3101         INF("io",       S_IRUSR, proc_tid_io_accounting),
3102 #endif
3103 #ifdef CONFIG_HARDWALL
3104         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
3105 #endif
3106 #ifdef CONFIG_USER_NS
3107         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3108         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3109         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3110 #endif
3111 };
3112
3113 static int proc_tid_base_readdir(struct file * filp,
3114                              void * dirent, filldir_t filldir)
3115 {
3116         return proc_pident_readdir(filp,dirent,filldir,
3117                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
3118 }
3119
3120 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3121 {
3122         return proc_pident_lookup(dir, dentry,
3123                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3124 }
3125
3126 static const struct file_operations proc_tid_base_operations = {
3127         .read           = generic_read_dir,
3128         .readdir        = proc_tid_base_readdir,
3129         .llseek         = default_llseek,
3130 };
3131
3132 static const struct inode_operations proc_tid_base_inode_operations = {
3133         .lookup         = proc_tid_base_lookup,
3134         .getattr        = pid_getattr,
3135         .setattr        = proc_setattr,
3136 };
3137
3138 static struct dentry *proc_task_instantiate(struct inode *dir,
3139         struct dentry *dentry, struct task_struct *task, const void *ptr)
3140 {
3141         struct dentry *error = ERR_PTR(-ENOENT);
3142         struct inode *inode;
3143         inode = proc_pid_make_inode(dir->i_sb, task);
3144
3145         if (!inode)
3146                 goto out;
3147         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3148         inode->i_op = &proc_tid_base_inode_operations;
3149         inode->i_fop = &proc_tid_base_operations;
3150         inode->i_flags|=S_IMMUTABLE;
3151
3152         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3153                                                   ARRAY_SIZE(tid_base_stuff)));
3154
3155         d_set_d_op(dentry, &pid_dentry_operations);
3156
3157         d_add(dentry, inode);
3158         /* Close the race of the process dying before we return the dentry */
3159         if (pid_revalidate(dentry, 0))
3160                 error = NULL;
3161 out:
3162         return error;
3163 }
3164
3165 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3166 {
3167         struct dentry *result = ERR_PTR(-ENOENT);
3168         struct task_struct *task;
3169         struct task_struct *leader = get_proc_task(dir);
3170         unsigned tid;
3171         struct pid_namespace *ns;
3172
3173         if (!leader)
3174                 goto out_no_task;
3175
3176         tid = name_to_int(dentry);
3177         if (tid == ~0U)
3178                 goto out;
3179
3180         ns = dentry->d_sb->s_fs_info;
3181         rcu_read_lock();
3182         task = find_task_by_pid_ns(tid, ns);
3183         if (task)
3184                 get_task_struct(task);
3185         rcu_read_unlock();
3186         if (!task)
3187                 goto out;
3188         if (!same_thread_group(leader, task))
3189                 goto out_drop_task;
3190
3191         result = proc_task_instantiate(dir, dentry, task, NULL);
3192 out_drop_task:
3193         put_task_struct(task);
3194 out:
3195         put_task_struct(leader);
3196 out_no_task:
3197         return result;
3198 }
3199
3200 /*
3201  * Find the first tid of a thread group to return to user space.
3202  *
3203  * Usually this is just the thread group leader, but if the users
3204  * buffer was too small or there was a seek into the middle of the
3205  * directory we have more work todo.
3206  *
3207  * In the case of a short read we start with find_task_by_pid.
3208  *
3209  * In the case of a seek we start with the leader and walk nr
3210  * threads past it.
3211  */
3212 static struct task_struct *first_tid(struct task_struct *leader,
3213                 int tid, int nr, struct pid_namespace *ns)
3214 {
3215         struct task_struct *pos;
3216
3217         rcu_read_lock();
3218         /* Attempt to start with the pid of a thread */
3219         if (tid && (nr > 0)) {
3220                 pos = find_task_by_pid_ns(tid, ns);
3221                 if (pos && (pos->group_leader == leader))
3222                         goto found;
3223         }
3224
3225         /* If nr exceeds the number of threads there is nothing todo */
3226         pos = NULL;
3227         if (nr && nr >= get_nr_threads(leader))
3228                 goto out;
3229
3230         /* If we haven't found our starting place yet start
3231          * with the leader and walk nr threads forward.
3232          */
3233         for (pos = leader; nr > 0; --nr) {
3234                 pos = next_thread(pos);
3235                 if (pos == leader) {
3236                         pos = NULL;
3237                         goto out;
3238                 }
3239         }
3240 found:
3241         get_task_struct(pos);
3242 out:
3243         rcu_read_unlock();
3244         return pos;
3245 }
3246
3247 /*
3248  * Find the next thread in the thread list.
3249  * Return NULL if there is an error or no next thread.
3250  *
3251  * The reference to the input task_struct is released.
3252  */
3253 static struct task_struct *next_tid(struct task_struct *start)
3254 {
3255         struct task_struct *pos = NULL;
3256         rcu_read_lock();
3257         if (pid_alive(start)) {
3258                 pos = next_thread(start);
3259                 if (thread_group_leader(pos))
3260                         pos = NULL;
3261                 else
3262                         get_task_struct(pos);
3263         }
3264         rcu_read_unlock();
3265         put_task_struct(start);
3266         return pos;
3267 }
3268
3269 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3270         struct task_struct *task, int tid)
3271 {
3272         char name[PROC_NUMBUF];
3273         int len = snprintf(name, sizeof(name), "%d", tid);
3274         return proc_fill_cache(filp, dirent, filldir, name, len,
3275                                 proc_task_instantiate, task, NULL);
3276 }
3277
3278 /* for the /proc/TGID/task/ directories */
3279 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3280 {
3281         struct dentry *dentry = filp->f_path.dentry;
3282         struct inode *inode = dentry->d_inode;
3283         struct task_struct *leader = NULL;
3284         struct task_struct *task;
3285         int retval = -ENOENT;
3286         ino_t ino;
3287         int tid;
3288         struct pid_namespace *ns;
3289
3290         task = get_proc_task(inode);
3291         if (!task)
3292                 goto out_no_task;
3293         rcu_read_lock();
3294         if (pid_alive(task)) {
3295                 leader = task->group_leader;
3296                 get_task_struct(leader);
3297         }
3298         rcu_read_unlock();
3299         put_task_struct(task);
3300         if (!leader)
3301                 goto out_no_task;
3302         retval = 0;
3303
3304         switch ((unsigned long)filp->f_pos) {
3305         case 0:
3306                 ino = inode->i_ino;
3307                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3308                         goto out;
3309                 filp->f_pos++;
3310                 /* fall through */
3311         case 1:
3312                 ino = parent_ino(dentry);
3313                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3314                         goto out;
3315                 filp->f_pos++;
3316                 /* fall through */
3317         }
3318
3319         /* f_version caches the tgid value that the last readdir call couldn't
3320          * return. lseek aka telldir automagically resets f_version to 0.
3321          */
3322         ns = filp->f_dentry->d_sb->s_fs_info;
3323         tid = (int)filp->f_version;
3324         filp->f_version = 0;
3325         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3326              task;
3327              task = next_tid(task), filp->f_pos++) {
3328                 tid = task_pid_nr_ns(task, ns);
3329                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3330                         /* returning this tgid failed, save it as the first
3331                          * pid for the next readir call */
3332                         filp->f_version = (u64)tid;
3333                         put_task_struct(task);
3334                         break;
3335                 }
3336         }
3337 out:
3338         put_task_struct(leader);
3339 out_no_task:
3340         return retval;
3341 }
3342
3343 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3344 {
3345         struct inode *inode = dentry->d_inode;
3346         struct task_struct *p = get_proc_task(inode);
3347         generic_fillattr(inode, stat);
3348
3349         if (p) {
3350                 stat->nlink += get_nr_threads(p);
3351                 put_task_struct(p);
3352         }
3353
3354         return 0;
3355 }
3356
3357 static const struct inode_operations proc_task_inode_operations = {
3358         .lookup         = proc_task_lookup,
3359         .getattr        = proc_task_getattr,
3360         .setattr        = proc_setattr,
3361         .permission     = proc_pid_permission,
3362 };
3363
3364 static const struct file_operations proc_task_operations = {
3365         .read           = generic_read_dir,
3366         .readdir        = proc_task_readdir,
3367         .llseek         = default_llseek,
3368 };