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