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 <[email protected]>
  20  *  Edjard Mota <[email protected]>
  21  *  Ilias Biris <[email protected]>
  22  *  Mauricio Lin <[email protected]>
  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 <[email protected]>:
  44  *  Smaps information related to shared, private, clean and dirty pages.
  45  *
  46  *  Paul Mundt <[email protected]>:
  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
1155         /* is userspace tring to explicitly UNSET the loginuid? */
1156         if (loginuid == AUDIT_UID_UNSET) {
1157                 kloginuid = INVALID_UID;
1158         } else {
1159                 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1160                 if (!uid_valid(kloginuid)) {
1161                         length = -EINVAL;
1162                         goto out_free_page;
1163                 }
1164         }
1165
1166         length = audit_set_loginuid(kloginuid);
1167         if (likely(length == 0))
1168                 length = count;
1169
1170 out_free_page:
1171         free_page((unsigned long) page);
1172         return length;
1173 }
1174
1175 static const struct file_operations proc_loginuid_operations = {
1176         .read           = proc_loginuid_read,
1177         .write          = proc_loginuid_write,
1178         .llseek         = generic_file_llseek,
1179 };
1180
1181 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1182                                   size_t count, loff_t *ppos)
1183 {
1184         struct inode * inode = file_inode(file);
1185         struct task_struct *task = get_proc_task(inode);
1186         ssize_t length;
1187         char tmpbuf[TMPBUFLEN];
1188
1189         if (!task)
1190                 return -ESRCH;
1191         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1192                                 audit_get_sessionid(task));
1193         put_task_struct(task);
1194         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1195 }
1196
1197 static const struct file_operations proc_sessionid_operations = {
1198         .read           = proc_sessionid_read,
1199         .llseek         = generic_file_llseek,
1200 };
1201 #endif
1202
1203 #ifdef CONFIG_FAULT_INJECTION
1204 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1205                                       size_t count, loff_t *ppos)
1206 {
1207         struct task_struct *task = get_proc_task(file_inode(file));
1208         char buffer[PROC_NUMBUF];
1209         size_t len;
1210         int make_it_fail;
1211
1212         if (!task)
1213                 return -ESRCH;
1214         make_it_fail = task->make_it_fail;
1215         put_task_struct(task);
1216
1217         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1218
1219         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1220 }
1221
1222 static ssize_t proc_fault_inject_write(struct file * file,
1223                         const char __user * buf, size_t count, loff_t *ppos)
1224 {
1225         struct task_struct *task;
1226         char buffer[PROC_NUMBUF], *end;
1227         int make_it_fail;
1228
1229         if (!capable(CAP_SYS_RESOURCE))
1230                 return -EPERM;
1231         memset(buffer, 0, sizeof(buffer));
1232         if (count > sizeof(buffer) - 1)
1233                 count = sizeof(buffer) - 1;
1234         if (copy_from_user(buffer, buf, count))
1235                 return -EFAULT;
1236         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1237         if (*end)
1238                 return -EINVAL;
1239         task = get_proc_task(file_inode(file));
1240         if (!task)
1241                 return -ESRCH;
1242         task->make_it_fail = make_it_fail;
1243         put_task_struct(task);
1244
1245         return count;
1246 }
1247
1248 static const struct file_operations proc_fault_inject_operations = {
1249         .read           = proc_fault_inject_read,
1250         .write          = proc_fault_inject_write,
1251         .llseek         = generic_file_llseek,
1252 };
1253 #endif
1254
1255
1256 #ifdef CONFIG_SCHED_DEBUG
1257 /*
1258  * Print out various scheduling related per-task fields:
1259  */
1260 static int sched_show(struct seq_file *m, void *v)
1261 {
1262         struct inode *inode = m->private;
1263         struct task_struct *p;
1264
1265         p = get_proc_task(inode);
1266         if (!p)
1267                 return -ESRCH;
1268         proc_sched_show_task(p, m);
1269
1270         put_task_struct(p);
1271
1272         return 0;
1273 }
1274
1275 static ssize_t
1276 sched_write(struct file *file, const char __user *buf,
1277             size_t count, loff_t *offset)
1278 {
1279         struct inode *inode = file_inode(file);
1280         struct task_struct *p;
1281
1282         p = get_proc_task(inode);
1283         if (!p)
1284                 return -ESRCH;
1285         proc_sched_set_task(p);
1286
1287         put_task_struct(p);
1288
1289         return count;
1290 }
1291
1292 static int sched_open(struct inode *inode, struct file *filp)
1293 {
1294         return single_open(filp, sched_show, inode);
1295 }
1296
1297 static const struct file_operations proc_pid_sched_operations = {
1298         .open           = sched_open,
1299         .read           = seq_read,
1300         .write          = sched_write,
1301         .llseek         = seq_lseek,
1302         .release        = single_release,
1303 };
1304
1305 #endif
1306
1307 #ifdef CONFIG_SCHED_AUTOGROUP
1308 /*
1309  * Print out autogroup related information:
1310  */
1311 static int sched_autogroup_show(struct seq_file *m, void *v)
1312 {
1313         struct inode *inode = m->private;
1314         struct task_struct *p;
1315
1316         p = get_proc_task(inode);
1317         if (!p)
1318                 return -ESRCH;
1319         proc_sched_autogroup_show_task(p, m);
1320
1321         put_task_struct(p);
1322
1323         return 0;
1324 }
1325
1326 static ssize_t
1327 sched_autogroup_write(struct file *file, const char __user *buf,
1328             size_t count, loff_t *offset)
1329 {
1330         struct inode *inode = file_inode(file);
1331         struct task_struct *p;
1332         char buffer[PROC_NUMBUF];
1333         int nice;
1334         int err;
1335
1336         memset(buffer, 0, sizeof(buffer));
1337         if (count > sizeof(buffer) - 1)
1338                 count = sizeof(buffer) - 1;
1339         if (copy_from_user(buffer, buf, count))
1340                 return -EFAULT;
1341
1342         err = kstrtoint(strstrip(buffer), 0, &nice);
1343         if (err < 0)
1344                 return err;
1345
1346         p = get_proc_task(inode);
1347         if (!p)
1348                 return -ESRCH;
1349
1350         err = proc_sched_autogroup_set_nice(p, nice);
1351         if (err)
1352                 count = err;
1353
1354         put_task_struct(p);
1355
1356         return count;
1357 }
1358
1359 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1360 {
1361         int ret;
1362
1363         ret = single_open(filp, sched_autogroup_show, NULL);
1364         if (!ret) {
1365                 struct seq_file *m = filp->private_data;
1366
1367                 m->private = inode;
1368         }
1369         return ret;
1370 }
1371
1372 static const struct file_operations proc_pid_sched_autogroup_operations = {
1373         .open           = sched_autogroup_open,
1374         .read           = seq_read,
1375         .write          = sched_autogroup_write,
1376         .llseek         = seq_lseek,
1377         .release        = single_release,
1378 };
1379
1380 #endif /* CONFIG_SCHED_AUTOGROUP */
1381
1382 static ssize_t comm_write(struct file *file, const char __user *buf,
1383                                 size_t count, loff_t *offset)
1384 {
1385         struct inode *inode = file_inode(file);
1386         struct task_struct *p;
1387         char buffer[TASK_COMM_LEN];
1388         const size_t maxlen = sizeof(buffer) - 1;
1389
1390         memset(buffer, 0, sizeof(buffer));
1391         if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1392                 return -EFAULT;
1393
1394         p = get_proc_task(inode);
1395         if (!p)
1396                 return -ESRCH;
1397
1398         if (same_thread_group(current, p))
1399                 set_task_comm(p, buffer);
1400         else
1401                 count = -EINVAL;
1402
1403         put_task_struct(p);
1404
1405         return count;
1406 }
1407
1408 static int comm_show(struct seq_file *m, void *v)
1409 {
1410         struct inode *inode = m->private;
1411         struct task_struct *p;
1412
1413         p = get_proc_task(inode);
1414         if (!p)
1415                 return -ESRCH;
1416
1417         task_lock(p);
1418         seq_printf(m, "%s\n", p->comm);
1419         task_unlock(p);
1420
1421         put_task_struct(p);
1422
1423         return 0;
1424 }
1425
1426 static int comm_open(struct inode *inode, struct file *filp)
1427 {
1428         return single_open(filp, comm_show, inode);
1429 }
1430
1431 static const struct file_operations proc_pid_set_comm_operations = {
1432         .open           = comm_open,
1433         .read           = seq_read,
1434         .write          = comm_write,
1435         .llseek         = seq_lseek,
1436         .release        = single_release,
1437 };
1438
1439 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1440 {
1441         struct task_struct *task;
1442         struct mm_struct *mm;
1443         struct file *exe_file;
1444
1445         task = get_proc_task(dentry->d_inode);
1446         if (!task)
1447                 return -ENOENT;
1448         mm = get_task_mm(task);
1449         put_task_struct(task);
1450         if (!mm)
1451                 return -ENOENT;
1452         exe_file = get_mm_exe_file(mm);
1453         mmput(mm);
1454         if (exe_file) {
1455                 *exe_path = exe_file->f_path;
1456                 path_get(&exe_file->f_path);
1457                 fput(exe_file);
1458                 return 0;
1459         } else
1460                 return -ENOENT;
1461 }
1462
1463 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1464 {
1465         struct inode *inode = dentry->d_inode;
1466         struct path path;
1467         int error = -EACCES;
1468
1469         /* Are we allowed to snoop on the tasks file descriptors? */
1470         if (!proc_fd_access_allowed(inode))
1471                 goto out;
1472
1473         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1474         if (error)
1475                 goto out;
1476
1477         nd_jump_link(nd, &path);
1478         return NULL;
1479 out:
1480         return ERR_PTR(error);
1481 }
1482
1483 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1484 {
1485         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1486         char *pathname;
1487         int len;
1488
1489         if (!tmp)
1490                 return -ENOMEM;
1491
1492         pathname = d_path(path, tmp, PAGE_SIZE);
1493         len = PTR_ERR(pathname);
1494         if (IS_ERR(pathname))
1495                 goto out;
1496         len = tmp + PAGE_SIZE - 1 - pathname;
1497
1498         if (len > buflen)
1499                 len = buflen;
1500         if (copy_to_user(buffer, pathname, len))
1501                 len = -EFAULT;
1502  out:
1503         free_page((unsigned long)tmp);
1504         return len;
1505 }
1506
1507 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1508 {
1509         int error = -EACCES;
1510         struct inode *inode = dentry->d_inode;
1511         struct path path;
1512
1513         /* Are we allowed to snoop on the tasks file descriptors? */
1514         if (!proc_fd_access_allowed(inode))
1515                 goto out;
1516
1517         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1518         if (error)
1519                 goto out;
1520
1521         error = do_proc_readlink(&path, buffer, buflen);
1522         path_put(&path);
1523 out:
1524         return error;
1525 }
1526
1527 const struct inode_operations proc_pid_link_inode_operations = {
1528         .readlink       = proc_pid_readlink,
1529         .follow_link    = proc_pid_follow_link,
1530         .setattr        = proc_setattr,
1531 };
1532
1533
1534 /* building an inode */
1535
1536 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1537 {
1538         struct inode * inode;
1539         struct proc_inode *ei;
1540         const struct cred *cred;
1541
1542         /* We need a new inode */
1543
1544         inode = new_inode(sb);
1545         if (!inode)
1546                 goto out;
1547
1548         /* Common stuff */
1549         ei = PROC_I(inode);
1550         inode->i_ino = get_next_ino();
1551         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1552         inode->i_op = &proc_def_inode_operations;
1553
1554         /*
1555          * grab the reference to task.
1556          */
1557         ei->pid = get_task_pid(task, PIDTYPE_PID);
1558         if (!ei->pid)
1559                 goto out_unlock;
1560
1561         if (task_dumpable(task)) {
1562                 rcu_read_lock();
1563                 cred = __task_cred(task);
1564                 inode->i_uid = cred->euid;
1565                 inode->i_gid = cred->egid;
1566                 rcu_read_unlock();
1567         }
1568         security_task_to_inode(task, inode);
1569
1570 out:
1571         return inode;
1572
1573 out_unlock:
1574         iput(inode);
1575         return NULL;
1576 }
1577
1578 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1579 {
1580         struct inode *inode = dentry->d_inode;
1581         struct task_struct *task;
1582         const struct cred *cred;
1583         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1584
1585         generic_fillattr(inode, stat);
1586
1587         rcu_read_lock();
1588         stat->uid = GLOBAL_ROOT_UID;
1589         stat->gid = GLOBAL_ROOT_GID;
1590         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1591         if (task) {
1592                 if (!has_pid_permissions(pid, task, 2)) {
1593                         rcu_read_unlock();
1594                         /*
1595                          * This doesn't prevent learning whether PID exists,
1596                          * it only makes getattr() consistent with readdir().
1597                          */
1598                         return -ENOENT;
1599                 }
1600                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1601                     task_dumpable(task)) {
1602                         cred = __task_cred(task);
1603                         stat->uid = cred->euid;
1604                         stat->gid = cred->egid;
1605                 }
1606         }
1607         rcu_read_unlock();
1608         return 0;
1609 }
1610
1611 /* dentry stuff */
1612
1613 /*
1614  *      Exceptional case: normally we are not allowed to unhash a busy
1615  * directory. In this case, however, we can do it - no aliasing problems
1616  * due to the way we treat inodes.
1617  *
1618  * Rewrite the inode's ownerships here because the owning task may have
1619  * performed a setuid(), etc.
1620  *
1621  * Before the /proc/pid/status file was created the only way to read
1622  * the effective uid of a /process was to stat /proc/pid.  Reading
1623  * /proc/pid/status is slow enough that procps and other packages
1624  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1625  * made this apply to all per process world readable and executable
1626  * directories.
1627  */
1628 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1629 {
1630         struct inode *inode;
1631         struct task_struct *task;
1632         const struct cred *cred;
1633
1634         if (flags & LOOKUP_RCU)
1635                 return -ECHILD;
1636
1637         inode = dentry->d_inode;
1638         task = get_proc_task(inode);
1639
1640         if (task) {
1641                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1642                     task_dumpable(task)) {
1643                         rcu_read_lock();
1644                         cred = __task_cred(task);
1645                         inode->i_uid = cred->euid;
1646                         inode->i_gid = cred->egid;
1647                         rcu_read_unlock();
1648                 } else {
1649                         inode->i_uid = GLOBAL_ROOT_UID;
1650                         inode->i_gid = GLOBAL_ROOT_GID;
1651                 }
1652                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1653                 security_task_to_inode(task, inode);
1654                 put_task_struct(task);
1655                 return 1;
1656         }
1657         d_drop(dentry);
1658         return 0;
1659 }
1660
1661 static inline bool proc_inode_is_dead(struct inode *inode)
1662 {
1663         return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1664 }
1665
1666 int pid_delete_dentry(const struct dentry *dentry)
1667 {
1668         /* Is the task we represent dead?
1669          * If so, then don't put the dentry on the lru list,
1670          * kill it immediately.
1671          */
1672         return proc_inode_is_dead(dentry->d_inode);
1673 }
1674
1675 const struct dentry_operations pid_dentry_operations =
1676 {
1677         .d_revalidate   = pid_revalidate,
1678         .d_delete       = pid_delete_dentry,
1679 };
1680
1681 /* Lookups */
1682
1683 /*
1684  * Fill a directory entry.
1685  *
1686  * If possible create the dcache entry and derive our inode number and
1687  * file type from dcache entry.
1688  *
1689  * Since all of the proc inode numbers are dynamically generated, the inode
1690  * numbers do not exist until the inode is cache.  This means creating the
1691  * the dcache entry in readdir is necessary to keep the inode numbers
1692  * reported by readdir in sync with the inode numbers reported
1693  * by stat.
1694  */
1695 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1696         const char *name, int len,
1697         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1698 {
1699         struct dentry *child, *dir = file->f_path.dentry;
1700         struct qstr qname = QSTR_INIT(name, len);
1701         struct inode *inode;
1702         unsigned type;
1703         ino_t ino;
1704
1705         child = d_hash_and_lookup(dir, &qname);
1706         if (!child) {
1707                 child = d_alloc(dir, &qname);
1708                 if (!child)
1709                         goto end_instantiate;
1710                 if (instantiate(dir->d_inode, child, task, ptr) < 0) {
1711                         dput(child);
1712                         goto end_instantiate;
1713                 }
1714         }
1715         inode = child->d_inode;
1716         ino = inode->i_ino;
1717         type = inode->i_mode >> 12;
1718         dput(child);
1719         return dir_emit(ctx, name, len, ino, type);
1720
1721 end_instantiate:
1722         return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1723 }
1724
1725 #ifdef CONFIG_CHECKPOINT_RESTORE
1726
1727 /*
1728  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1729  * which represent vma start and end addresses.
1730  */
1731 static int dname_to_vma_addr(struct dentry *dentry,
1732                              unsigned long *start, unsigned long *end)
1733 {
1734         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1735                 return -EINVAL;
1736
1737         return 0;
1738 }
1739
1740 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1741 {
1742         unsigned long vm_start, vm_end;
1743         bool exact_vma_exists = false;
1744         struct mm_struct *mm = NULL;
1745         struct task_struct *task;
1746         const struct cred *cred;
1747         struct inode *inode;
1748         int status = 0;
1749
1750         if (flags & LOOKUP_RCU)
1751                 return -ECHILD;
1752
1753         if (!capable(CAP_SYS_ADMIN)) {
1754                 status = -EPERM;
1755                 goto out_notask;
1756         }
1757
1758         inode = dentry->d_inode;
1759         task = get_proc_task(inode);
1760         if (!task)
1761                 goto out_notask;
1762
1763         mm = mm_access(task, PTRACE_MODE_READ);
1764         if (IS_ERR_OR_NULL(mm))
1765                 goto out;
1766
1767         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1768                 down_read(&mm->mmap_sem);
1769                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1770                 up_read(&mm->mmap_sem);
1771         }
1772
1773         mmput(mm);
1774
1775         if (exact_vma_exists) {
1776                 if (task_dumpable(task)) {
1777                         rcu_read_lock();
1778                         cred = __task_cred(task);
1779                         inode->i_uid = cred->euid;
1780                         inode->i_gid = cred->egid;
1781                         rcu_read_unlock();
1782                 } else {
1783                         inode->i_uid = GLOBAL_ROOT_UID;
1784                         inode->i_gid = GLOBAL_ROOT_GID;
1785                 }
1786                 security_task_to_inode(task, inode);
1787                 status = 1;
1788         }
1789
1790 out:
1791         put_task_struct(task);
1792
1793 out_notask:
1794         if (status <= 0)
1795                 d_drop(dentry);
1796
1797         return status;
1798 }
1799
1800 static const struct dentry_operations tid_map_files_dentry_operations = {
1801         .d_revalidate   = map_files_d_revalidate,
1802         .d_delete       = pid_delete_dentry,
1803 };
1804
1805 static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
1806 {
1807         unsigned long vm_start, vm_end;
1808         struct vm_area_struct *vma;
1809         struct task_struct *task;
1810         struct mm_struct *mm;
1811         int rc;
1812
1813         rc = -ENOENT;
1814         task = get_proc_task(dentry->d_inode);
1815         if (!task)
1816                 goto out;
1817
1818         mm = get_task_mm(task);
1819         put_task_struct(task);
1820         if (!mm)
1821                 goto out;
1822
1823         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1824         if (rc)
1825                 goto out_mmput;
1826
1827         rc = -ENOENT;
1828         down_read(&mm->mmap_sem);
1829         vma = find_exact_vma(mm, vm_start, vm_end);
1830         if (vma && vma->vm_file) {
1831                 *path = vma->vm_file->f_path;
1832                 path_get(path);
1833                 rc = 0;
1834         }
1835         up_read(&mm->mmap_sem);
1836
1837 out_mmput:
1838         mmput(mm);
1839 out:
1840         return rc;
1841 }
1842
1843 struct map_files_info {
1844         fmode_t         mode;
1845         unsigned long   len;
1846         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1847 };
1848
1849 static int
1850 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
1851                            struct task_struct *task, const void *ptr)
1852 {
1853         fmode_t mode = (fmode_t)(unsigned long)ptr;
1854         struct proc_inode *ei;
1855         struct inode *inode;
1856
1857         inode = proc_pid_make_inode(dir->i_sb, task);
1858         if (!inode)
1859                 return -ENOENT;
1860
1861         ei = PROC_I(inode);
1862         ei->op.proc_get_link = proc_map_files_get_link;
1863
1864         inode->i_op = &proc_pid_link_inode_operations;
1865         inode->i_size = 64;
1866         inode->i_mode = S_IFLNK;
1867
1868         if (mode & FMODE_READ)
1869                 inode->i_mode |= S_IRUSR;
1870         if (mode & FMODE_WRITE)
1871                 inode->i_mode |= S_IWUSR;
1872
1873         d_set_d_op(dentry, &tid_map_files_dentry_operations);
1874         d_add(dentry, inode);
1875
1876         return 0;
1877 }
1878
1879 static struct dentry *proc_map_files_lookup(struct inode *dir,
1880                 struct dentry *dentry, unsigned int flags)
1881 {
1882         unsigned long vm_start, vm_end;
1883         struct vm_area_struct *vma;
1884         struct task_struct *task;
1885         int result;
1886         struct mm_struct *mm;
1887
1888         result = -EPERM;
1889         if (!capable(CAP_SYS_ADMIN))
1890                 goto out;
1891
1892         result = -ENOENT;
1893         task = get_proc_task(dir);
1894         if (!task)
1895                 goto out;
1896
1897         result = -EACCES;
1898         if (!ptrace_may_access(task, PTRACE_MODE_READ))
1899                 goto out_put_task;
1900
1901         result = -ENOENT;
1902         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
1903                 goto out_put_task;
1904
1905         mm = get_task_mm(task);
1906         if (!mm)
1907                 goto out_put_task;
1908
1909         down_read(&mm->mmap_sem);
1910         vma = find_exact_vma(mm, vm_start, vm_end);
1911         if (!vma)
1912                 goto out_no_vma;
1913
1914         if (vma->vm_file)
1915                 result = proc_map_files_instantiate(dir, dentry, task,
1916                                 (void *)(unsigned long)vma->vm_file->f_mode);
1917
1918 out_no_vma:
1919         up_read(&mm->mmap_sem);
1920         mmput(mm);
1921 out_put_task:
1922         put_task_struct(task);
1923 out:
1924         return ERR_PTR(result);
1925 }
1926
1927 static const struct inode_operations proc_map_files_inode_operations = {
1928         .lookup         = proc_map_files_lookup,
1929         .permission     = proc_fd_permission,
1930         .setattr        = proc_setattr,
1931 };
1932
1933 static int
1934 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
1935 {
1936         struct vm_area_struct *vma;
1937         struct task_struct *task;
1938         struct mm_struct *mm;
1939         unsigned long nr_files, pos, i;
1940         struct flex_array *fa = NULL;
1941         struct map_files_info info;
1942         struct map_files_info *p;
1943         int ret;
1944
1945         ret = -EPERM;
1946         if (!capable(CAP_SYS_ADMIN))
1947                 goto out;
1948
1949         ret = -ENOENT;
1950         task = get_proc_task(file_inode(file));
1951         if (!task)
1952                 goto out;
1953
1954         ret = -EACCES;
1955         if (!ptrace_may_access(task, PTRACE_MODE_READ))
1956                 goto out_put_task;
1957
1958         ret = 0;
1959         if (!dir_emit_dots(file, ctx))
1960                 goto out_put_task;
1961
1962         mm = get_task_mm(task);
1963         if (!mm)
1964                 goto out_put_task;
1965         down_read(&mm->mmap_sem);
1966
1967         nr_files = 0;
1968
1969         /*
1970          * We need two passes here:
1971          *
1972          *  1) Collect vmas of mapped files with mmap_sem taken
1973          *  2) Release mmap_sem and instantiate entries
1974          *
1975          * otherwise we get lockdep complained, since filldir()
1976          * routine might require mmap_sem taken in might_fault().
1977          */
1978
1979         for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
1980                 if (vma->vm_file && ++pos > ctx->pos)
1981                         nr_files++;
1982         }
1983
1984         if (nr_files) {
1985                 fa = flex_array_alloc(sizeof(info), nr_files,
1986                                         GFP_KERNEL);
1987                 if (!fa || flex_array_prealloc(fa, 0, nr_files,
1988                                                 GFP_KERNEL)) {
1989                         ret = -ENOMEM;
1990                         if (fa)
1991                                 flex_array_free(fa);
1992                         up_read(&mm->mmap_sem);
1993                         mmput(mm);
1994                         goto out_put_task;
1995                 }
1996                 for (i = 0, vma = mm->mmap, pos = 2; vma;
1997                                 vma = vma->vm_next) {
1998                         if (!vma->vm_file)
1999                                 continue;
2000                         if (++pos <= ctx->pos)
2001                                 continue;
2002
2003                         info.mode = vma->vm_file->f_mode;
2004                         info.len = snprintf(info.name,
2005                                         sizeof(info.name), "%lx-%lx",
2006                                         vma->vm_start, vma->vm_end);
2007                         if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2008                                 BUG();
2009                 }
2010         }
2011         up_read(&mm->mmap_sem);
2012
2013         for (i = 0; i < nr_files; i++) {
2014                 p = flex_array_get(fa, i);
2015                 if (!proc_fill_cache(file, ctx,
2016                                       p->name, p->len,
2017                                       proc_map_files_instantiate,
2018                                       task,
2019                                       (void *)(unsigned long)p->mode))
2020                         break;
2021                 ctx->pos++;
2022         }
2023         if (fa)
2024                 flex_array_free(fa);
2025         mmput(mm);
2026
2027 out_put_task:
2028         put_task_struct(task);
2029 out:
2030         return ret;
2031 }
2032
2033 static const struct file_operations proc_map_files_operations = {
2034         .read           = generic_read_dir,
2035         .iterate        = proc_map_files_readdir,
2036         .llseek         = default_llseek,
2037 };
2038
2039 struct timers_private {
2040         struct pid *pid;
2041         struct task_struct *task;
2042         struct sighand_struct *sighand;
2043         struct pid_namespace *ns;
2044         unsigned long flags;
2045 };
2046
2047 static void *timers_start(struct seq_file *m, loff_t *pos)
2048 {
2049         struct timers_private *tp = m->private;
2050
2051         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2052         if (!tp->task)
2053                 return ERR_PTR(-ESRCH);
2054
2055         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2056         if (!tp->sighand)
2057                 return ERR_PTR(-ESRCH);
2058
2059         return seq_list_start(&tp->task->signal->posix_timers, *pos);
2060 }
2061
2062 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2063 {
2064         struct timers_private *tp = m->private;
2065         return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2066 }
2067
2068 static void timers_stop(struct seq_file *m, void *v)
2069 {
2070         struct timers_private *tp = m->private;
2071
2072         if (tp->sighand) {
2073                 unlock_task_sighand(tp->task, &tp->flags);
2074                 tp->sighand = NULL;
2075         }
2076
2077         if (tp->task) {
2078                 put_task_struct(tp->task);
2079                 tp->task = NULL;
2080         }
2081 }
2082
2083 static int show_timer(struct seq_file *m, void *v)
2084 {
2085         struct k_itimer *timer;
2086         struct timers_private *tp = m->private;
2087         int notify;
2088         static char *nstr[] = {
2089                 [SIGEV_SIGNAL] = "signal",
2090                 [SIGEV_NONE] = "none",
2091                 [SIGEV_THREAD] = "thread",
2092         };
2093
2094         timer = list_entry((struct list_head *)v, struct k_itimer, list);
2095         notify = timer->it_sigev_notify;
2096
2097         seq_printf(m, "ID: %d\n", timer->it_id);
2098         seq_printf(m, "signal: %d/%p\n", timer->sigq->info.si_signo,
2099                         timer->sigq->info.si_value.sival_ptr);
2100         seq_printf(m, "notify: %s/%s.%d\n",
2101                 nstr[notify & ~SIGEV_THREAD_ID],
2102                 (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2103                 pid_nr_ns(timer->it_pid, tp->ns));
2104         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2105
2106         return 0;
2107 }
2108
2109 static const struct seq_operations proc_timers_seq_ops = {
2110         .start  = timers_start,
2111         .next   = timers_next,
2112         .stop   = timers_stop,
2113         .show   = show_timer,
2114 };
2115
2116 static int proc_timers_open(struct inode *inode, struct file *file)
2117 {
2118         struct timers_private *tp;
2119
2120         tp = __seq_open_private(file, &proc_timers_seq_ops,
2121                         sizeof(struct timers_private));
2122         if (!tp)
2123                 return -ENOMEM;
2124
2125         tp->pid = proc_pid(inode);
2126         tp->ns = inode->i_sb->s_fs_info;
2127         return 0;
2128 }
2129
2130 static const struct file_operations proc_timers_operations = {
2131         .open           = proc_timers_open,
2132         .read           = seq_read,
2133         .llseek         = seq_lseek,
2134         .release        = seq_release_private,
2135 };
2136 #endif /* CONFIG_CHECKPOINT_RESTORE */
2137
2138 static int proc_pident_instantiate(struct inode *dir,
2139         struct dentry *dentry, struct task_struct *task, const void *ptr)
2140 {
2141         const struct pid_entry *p = ptr;
2142         struct inode *inode;
2143         struct proc_inode *ei;
2144
2145         inode = proc_pid_make_inode(dir->i_sb, task);
2146         if (!inode)
2147                 goto out;
2148
2149         ei = PROC_I(inode);
2150         inode->i_mode = p->mode;
2151         if (S_ISDIR(inode->i_mode))
2152                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2153         if (p->iop)
2154                 inode->i_op = p->iop;
2155         if (p->fop)
2156                 inode->i_fop = p->fop;
2157         ei->op = p->op;
2158         d_set_d_op(dentry, &pid_dentry_operations);
2159         d_add(dentry, inode);
2160         /* Close the race of the process dying before we return the dentry */
2161         if (pid_revalidate(dentry, 0))
2162                 return 0;
2163 out:
2164         return -ENOENT;
2165 }
2166
2167 static struct dentry *proc_pident_lookup(struct inode *dir,
2168                                          struct dentry *dentry,
2169                                          const struct pid_entry *ents,
2170                                          unsigned int nents)
2171 {
2172         int error;
2173         struct task_struct *task = get_proc_task(dir);
2174         const struct pid_entry *p, *last;
2175
2176         error = -ENOENT;
2177
2178         if (!task)
2179                 goto out_no_task;
2180
2181         /*
2182          * Yes, it does not scale. And it should not. Don't add
2183          * new entries into /proc/<tgid>/ without very good reasons.
2184          */
2185         last = &ents[nents - 1];
2186         for (p = ents; p <= last; p++) {
2187                 if (p->len != dentry->d_name.len)
2188                         continue;
2189                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2190                         break;
2191         }
2192         if (p > last)
2193                 goto out;
2194
2195         error = proc_pident_instantiate(dir, dentry, task, p);
2196 out:
2197         put_task_struct(task);
2198 out_no_task:
2199         return ERR_PTR(error);
2200 }
2201
2202 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2203                 const struct pid_entry *ents, unsigned int nents)
2204 {
2205         struct task_struct *task = get_proc_task(file_inode(file));
2206         const struct pid_entry *p;
2207
2208         if (!task)
2209                 return -ENOENT;
2210
2211         if (!dir_emit_dots(file, ctx))
2212                 goto out;
2213
2214         if (ctx->pos >= nents + 2)
2215                 goto out;
2216
2217         for (p = ents + (ctx->pos - 2); p <= ents + nents - 1; p++) {
2218                 if (!proc_fill_cache(file, ctx, p->name, p->len,
2219                                 proc_pident_instantiate, task, p))
2220                         break;
2221                 ctx->pos++;
2222         }
2223 out:
2224         put_task_struct(task);
2225         return 0;
2226 }
2227
2228 #ifdef CONFIG_SECURITY
2229 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2230                                   size_t count, loff_t *ppos)
2231 {
2232         struct inode * inode = file_inode(file);
2233         char *p = NULL;
2234         ssize_t length;
2235         struct task_struct *task = get_proc_task(inode);
2236
2237         if (!task)
2238                 return -ESRCH;
2239
2240         length = security_getprocattr(task,
2241                                       (char*)file->f_path.dentry->d_name.name,
2242                                       &p);
2243         put_task_struct(task);
2244         if (length > 0)
2245                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2246         kfree(p);
2247         return length;
2248 }
2249
2250 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2251                                    size_t count, loff_t *ppos)
2252 {
2253         struct inode * inode = file_inode(file);
2254         char *page;
2255         ssize_t length;
2256         struct task_struct *task = get_proc_task(inode);
2257
2258         length = -ESRCH;
2259         if (!task)
2260                 goto out_no_task;
2261         if (count > PAGE_SIZE)
2262                 count = PAGE_SIZE;
2263
2264         /* No partial writes. */
2265         length = -EINVAL;
2266         if (*ppos != 0)
2267                 goto out;
2268
2269         length = -ENOMEM;
2270         page = (char*)__get_free_page(GFP_TEMPORARY);
2271         if (!page)
2272                 goto out;
2273
2274         length = -EFAULT;
2275         if (copy_from_user(page, buf, count))
2276                 goto out_free;
2277
2278         /* Guard against adverse ptrace interaction */
2279         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2280         if (length < 0)
2281                 goto out_free;
2282
2283         length = security_setprocattr(task,
2284                                       (char*)file->f_path.dentry->d_name.name,
2285                                       (void*)page, count);
2286         mutex_unlock(&task->signal->cred_guard_mutex);
2287 out_free:
2288         free_page((unsigned long) page);
2289 out:
2290         put_task_struct(task);
2291 out_no_task:
2292         return length;
2293 }
2294
2295 static const struct file_operations proc_pid_attr_operations = {
2296         .read           = proc_pid_attr_read,
2297         .write          = proc_pid_attr_write,
2298         .llseek         = generic_file_llseek,
2299 };
2300
2301 static const struct pid_entry attr_dir_stuff[] = {
2302         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2303         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2304         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2305         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2306         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2307         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2308 };
2309
2310 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2311 {
2312         return proc_pident_readdir(file, ctx,
2313                                    attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2314 }
2315
2316 static const struct file_operations proc_attr_dir_operations = {
2317         .read           = generic_read_dir,
2318         .iterate        = proc_attr_dir_readdir,
2319         .llseek         = default_llseek,
2320 };
2321
2322 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2323                                 struct dentry *dentry, unsigned int flags)
2324 {
2325         return proc_pident_lookup(dir, dentry,
2326                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2327 }
2328
2329 static const struct inode_operations proc_attr_dir_inode_operations = {
2330         .lookup         = proc_attr_dir_lookup,
2331         .getattr        = pid_getattr,
2332         .setattr        = proc_setattr,
2333 };
2334
2335 #endif
2336
2337 #ifdef CONFIG_ELF_CORE
2338 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2339                                          size_t count, loff_t *ppos)
2340 {
2341         struct task_struct *task = get_proc_task(file_inode(file));
2342         struct mm_struct *mm;
2343         char buffer[PROC_NUMBUF];
2344         size_t len;
2345         int ret;
2346
2347         if (!task)
2348                 return -ESRCH;
2349
2350         ret = 0;
2351         mm = get_task_mm(task);
2352         if (mm) {
2353                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2354                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2355                                 MMF_DUMP_FILTER_SHIFT));
2356                 mmput(mm);
2357                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2358         }
2359
2360         put_task_struct(task);
2361
2362         return ret;
2363 }
2364
2365 static ssize_t proc_coredump_filter_write(struct file *file,
2366                                           const char __user *buf,
2367                                           size_t count,
2368                                           loff_t *ppos)
2369 {
2370         struct task_struct *task;
2371         struct mm_struct *mm;
2372         char buffer[PROC_NUMBUF], *end;
2373         unsigned int val;
2374         int ret;
2375         int i;
2376         unsigned long mask;
2377
2378         ret = -EFAULT;
2379         memset(buffer, 0, sizeof(buffer));
2380         if (count > sizeof(buffer) - 1)
2381                 count = sizeof(buffer) - 1;
2382         if (copy_from_user(buffer, buf, count))
2383                 goto out_no_task;
2384
2385         ret = -EINVAL;
2386         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2387         if (*end == '\n')
2388                 end++;
2389         if (end - buffer == 0)
2390                 goto out_no_task;
2391
2392         ret = -ESRCH;
2393         task = get_proc_task(file_inode(file));
2394         if (!task)
2395                 goto out_no_task;
2396
2397         ret = end - buffer;
2398         mm = get_task_mm(task);
2399         if (!mm)
2400                 goto out_no_mm;
2401
2402         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2403                 if (val & mask)
2404                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2405                 else
2406                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2407         }
2408
2409         mmput(mm);
2410  out_no_mm:
2411         put_task_struct(task);
2412  out_no_task:
2413         return ret;
2414 }
2415
2416 static const struct file_operations proc_coredump_filter_operations = {
2417         .read           = proc_coredump_filter_read,
2418         .write          = proc_coredump_filter_write,
2419         .llseek         = generic_file_llseek,
2420 };
2421 #endif
2422
2423 #ifdef CONFIG_TASK_IO_ACCOUNTING
2424 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2425 {
2426         struct task_io_accounting acct = task->ioac;
2427         unsigned long flags;
2428         int result;
2429
2430         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2431         if (result)
2432                 return result;
2433
2434         if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2435                 result = -EACCES;
2436                 goto out_unlock;
2437         }
2438
2439         if (whole && lock_task_sighand(task, &flags)) {
2440                 struct task_struct *t = task;
2441
2442                 task_io_accounting_add(&acct, &task->signal->ioac);
2443                 while_each_thread(task, t)
2444                         task_io_accounting_add(&acct, &t->ioac);
2445
2446                 unlock_task_sighand(task, &flags);
2447         }
2448         result = sprintf(buffer,
2449                         "rchar: %llu\n"
2450                         "wchar: %llu\n"
2451                         "syscr: %llu\n"
2452                         "syscw: %llu\n"
2453                         "read_bytes: %llu\n"
2454                         "write_bytes: %llu\n"
2455                         "cancelled_write_bytes: %llu\n",
2456                         (unsigned long long)acct.rchar,
2457                         (unsigned long long)acct.wchar,
2458                         (unsigned long long)acct.syscr,
2459                         (unsigned long long)acct.syscw,
2460                         (unsigned long long)acct.read_bytes,
2461                         (unsigned long long)acct.write_bytes,
2462                         (unsigned long long)acct.cancelled_write_bytes);
2463 out_unlock:
2464         mutex_unlock(&task->signal->cred_guard_mutex);
2465         return result;
2466 }
2467
2468 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2469 {
2470         return do_io_accounting(task, buffer, 0);
2471 }
2472
2473 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2474 {
2475         return do_io_accounting(task, buffer, 1);
2476 }
2477 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2478
2479 #ifdef CONFIG_USER_NS
2480 static int proc_id_map_open(struct inode *inode, struct file *file,
2481         struct seq_operations *seq_ops)
2482 {
2483         struct user_namespace *ns = NULL;
2484         struct task_struct *task;
2485         struct seq_file *seq;
2486         int ret = -EINVAL;
2487
2488         task = get_proc_task(inode);
2489         if (task) {
2490                 rcu_read_lock();
2491                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2492                 rcu_read_unlock();
2493                 put_task_struct(task);
2494         }
2495         if (!ns)
2496                 goto err;
2497
2498         ret = seq_open(file, seq_ops);
2499         if (ret)
2500                 goto err_put_ns;
2501
2502         seq = file->private_data;
2503         seq->private = ns;
2504
2505         return 0;
2506 err_put_ns:
2507         put_user_ns(ns);
2508 err:
2509         return ret;
2510 }
2511
2512 static int proc_id_map_release(struct inode *inode, struct file *file)
2513 {
2514         struct seq_file *seq = file->private_data;
2515         struct user_namespace *ns = seq->private;
2516         put_user_ns(ns);
2517         return seq_release(inode, file);
2518 }
2519
2520 static int proc_uid_map_open(struct inode *inode, struct file *file)
2521 {
2522         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2523 }
2524
2525 static int proc_gid_map_open(struct inode *inode, struct file *file)
2526 {
2527         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2528 }
2529
2530 static int proc_projid_map_open(struct inode *inode, struct file *file)
2531 {
2532         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2533 }
2534
2535 static const struct file_operations proc_uid_map_operations = {
2536         .open           = proc_uid_map_open,
2537         .write          = proc_uid_map_write,
2538         .read           = seq_read,
2539         .llseek         = seq_lseek,
2540         .release        = proc_id_map_release,
2541 };
2542
2543 static const struct file_operations proc_gid_map_operations = {
2544         .open           = proc_gid_map_open,
2545         .write          = proc_gid_map_write,
2546         .read           = seq_read,
2547         .llseek         = seq_lseek,
2548         .release        = proc_id_map_release,
2549 };
2550
2551 static const struct file_operations proc_projid_map_operations = {
2552         .open           = proc_projid_map_open,
2553         .write          = proc_projid_map_write,
2554         .read           = seq_read,
2555         .llseek         = seq_lseek,
2556         .release        = proc_id_map_release,
2557 };
2558 #endif /* CONFIG_USER_NS */
2559
2560 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2561                                 struct pid *pid, struct task_struct *task)
2562 {
2563         int err = lock_trace(task);
2564         if (!err) {
2565                 seq_printf(m, "%08x\n", task->personality);
2566                 unlock_trace(task);
2567         }
2568         return err;
2569 }
2570
2571 /*
2572  * Thread groups
2573  */
2574 static const struct file_operations proc_task_operations;
2575 static const struct inode_operations proc_task_inode_operations;
2576
2577 static const struct pid_entry tgid_base_stuff[] = {
2578         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2579         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2580 #ifdef CONFIG_CHECKPOINT_RESTORE
2581         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2582 #endif
2583         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2584         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2585 #ifdef CONFIG_NET
2586         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2587 #endif
2588         REG("environ",    S_IRUSR, proc_environ_operations),
2589         INF("auxv",       S_IRUSR, proc_pid_auxv),
2590         ONE("status",     S_IRUGO, proc_pid_status),
2591         ONE("personality", S_IRUGO, proc_pid_personality),
2592         INF("limits",     S_IRUGO, proc_pid_limits),
2593 #ifdef CONFIG_SCHED_DEBUG
2594         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2595 #endif
2596 #ifdef CONFIG_SCHED_AUTOGROUP
2597         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2598 #endif
2599         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2600 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2601         INF("syscall",    S_IRUGO, proc_pid_syscall),
2602 #endif
2603         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2604         ONE("stat",       S_IRUGO, proc_tgid_stat),
2605         ONE("statm",      S_IRUGO, proc_pid_statm),
2606         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2607 #ifdef CONFIG_NUMA
2608         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2609 #endif
2610         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2611         LNK("cwd",        proc_cwd_link),
2612         LNK("root",       proc_root_link),
2613         LNK("exe",        proc_exe_link),
2614         REG("mounts",     S_IRUGO, proc_mounts_operations),
2615         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2616         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2617 #ifdef CONFIG_PROC_PAGE_MONITOR
2618         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2619         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2620         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
2621 #endif
2622 #ifdef CONFIG_SECURITY
2623         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2624 #endif
2625 #ifdef CONFIG_KALLSYMS
2626         INF("wchan",      S_IRUGO, proc_pid_wchan),
2627 #endif
2628 #ifdef CONFIG_STACKTRACE
2629         ONE("stack",      S_IRUGO, proc_pid_stack),
2630 #endif
2631 #ifdef CONFIG_SCHEDSTATS
2632         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
2633 #endif
2634 #ifdef CONFIG_LATENCYTOP
2635         REG("latency",  S_IRUGO, proc_lstats_operations),
2636 #endif
2637 #ifdef CONFIG_PROC_PID_CPUSET
2638         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2639 #endif
2640 #ifdef CONFIG_CGROUPS
2641         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2642 #endif
2643         INF("oom_score",  S_IRUGO, proc_oom_score),
2644         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2645         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2646 #ifdef CONFIG_AUDITSYSCALL
2647         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2648         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2649 #endif
2650 #ifdef CONFIG_FAULT_INJECTION
2651         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2652 #endif
2653 #ifdef CONFIG_ELF_CORE
2654         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2655 #endif
2656 #ifdef CONFIG_TASK_IO_ACCOUNTING
2657         INF("io",       S_IRUSR, proc_tgid_io_accounting),
2658 #endif
2659 #ifdef CONFIG_HARDWALL
2660         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
2661 #endif
2662 #ifdef CONFIG_USER_NS
2663         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2664         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2665         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2666 #endif
2667 #ifdef CONFIG_CHECKPOINT_RESTORE
2668         REG("timers",     S_IRUGO, proc_timers_operations),
2669 #endif
2670 };
2671
2672 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2673 {
2674         return proc_pident_readdir(file, ctx,
2675                                    tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2676 }
2677
2678 static const struct file_operations proc_tgid_base_operations = {
2679         .read           = generic_read_dir,
2680         .iterate        = proc_tgid_base_readdir,
2681         .llseek         = default_llseek,
2682 };
2683
2684 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2685 {
2686         return proc_pident_lookup(dir, dentry,
2687                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2688 }
2689
2690 static const struct inode_operations proc_tgid_base_inode_operations = {
2691         .lookup         = proc_tgid_base_lookup,
2692         .getattr        = pid_getattr,
2693         .setattr        = proc_setattr,
2694         .permission     = proc_pid_permission,
2695 };
2696
2697 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2698 {
2699         struct dentry *dentry, *leader, *dir;
2700         char buf[PROC_NUMBUF];
2701         struct qstr name;
2702
2703         name.name = buf;
2704         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2705         /* no ->d_hash() rejects on procfs */
2706         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2707         if (dentry) {
2708                 shrink_dcache_parent(dentry);
2709                 d_drop(dentry);
2710                 dput(dentry);
2711         }
2712
2713         name.name = buf;
2714         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2715         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2716         if (!leader)
2717                 goto out;
2718
2719         name.name = "task";
2720         name.len = strlen(name.name);
2721         dir = d_hash_and_lookup(leader, &name);
2722         if (!dir)
2723                 goto out_put_leader;
2724
2725         name.name = buf;
2726         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2727         dentry = d_hash_and_lookup(dir, &name);
2728         if (dentry) {
2729                 shrink_dcache_parent(dentry);
2730                 d_drop(dentry);
2731                 dput(dentry);
2732         }
2733
2734         dput(dir);
2735 out_put_leader:
2736         dput(leader);
2737 out:
2738         return;
2739 }
2740
2741 /**
2742  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2743  * @task: task that should be flushed.
2744  *
2745  * When flushing dentries from proc, one needs to flush them from global
2746  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2747  * in. This call is supposed to do all of this job.
2748  *
2749  * Looks in the dcache for
2750  * /proc/@pid
2751  * /proc/@tgid/task/@pid
2752  * if either directory is present flushes it and all of it'ts children
2753  * from the dcache.
2754  *
2755  * It is safe and reasonable to cache /proc entries for a task until
2756  * that task exits.  After that they just clog up the dcache with
2757  * useless entries, possibly causing useful dcache entries to be
2758  * flushed instead.  This routine is proved to flush those useless
2759  * dcache entries at process exit time.
2760  *
2761  * NOTE: This routine is just an optimization so it does not guarantee
2762  *       that no dcache entries will exist at process exit time it
2763  *       just makes it very unlikely that any will persist.
2764  */
2765
2766 void proc_flush_task(struct task_struct *task)
2767 {
2768         int i;
2769         struct pid *pid, *tgid;
2770         struct upid *upid;
2771
2772         pid = task_pid(task);
2773         tgid = task_tgid(task);
2774
2775         for (i = 0; i <= pid->level; i++) {
2776                 upid = &pid->numbers[i];
2777                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2778                                         tgid->numbers[i].nr);
2779         }
2780 }
2781
2782 static int proc_pid_instantiate(struct inode *dir,
2783                                    struct dentry * dentry,
2784                                    struct task_struct *task, const void *ptr)
2785 {
2786         struct inode *inode;
2787
2788         inode = proc_pid_make_inode(dir->i_sb, task);
2789         if (!inode)
2790                 goto out;
2791
2792         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2793         inode->i_op = &proc_tgid_base_inode_operations;
2794         inode->i_fop = &proc_tgid_base_operations;
2795         inode->i_flags|=S_IMMUTABLE;
2796
2797         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
2798                                                   ARRAY_SIZE(tgid_base_stuff)));
2799
2800         d_set_d_op(dentry, &pid_dentry_operations);
2801
2802         d_add(dentry, inode);
2803         /* Close the race of the process dying before we return the dentry */
2804         if (pid_revalidate(dentry, 0))
2805                 return 0;
2806 out:
2807         return -ENOENT;
2808 }
2809
2810 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
2811 {
2812         int result = 0;
2813         struct task_struct *task;
2814         unsigned tgid;
2815         struct pid_namespace *ns;
2816
2817         tgid = name_to_int(dentry);
2818         if (tgid == ~0U)
2819                 goto out;
2820
2821         ns = dentry->d_sb->s_fs_info;
2822         rcu_read_lock();
2823         task = find_task_by_pid_ns(tgid, ns);
2824         if (task)
2825                 get_task_struct(task);
2826         rcu_read_unlock();
2827         if (!task)
2828                 goto out;
2829
2830         result = proc_pid_instantiate(dir, dentry, task, NULL);
2831         put_task_struct(task);
2832 out:
2833         return ERR_PTR(result);
2834 }
2835
2836 /*
2837  * Find the first task with tgid >= tgid
2838  *
2839  */
2840 struct tgid_iter {
2841         unsigned int tgid;
2842         struct task_struct *task;
2843 };
2844 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2845 {
2846         struct pid *pid;
2847
2848         if (iter.task)
2849                 put_task_struct(iter.task);
2850         rcu_read_lock();
2851 retry:
2852         iter.task = NULL;
2853         pid = find_ge_pid(iter.tgid, ns);
2854         if (pid) {
2855                 iter.tgid = pid_nr_ns(pid, ns);
2856                 iter.task = pid_task(pid, PIDTYPE_PID);
2857                 /* What we to know is if the pid we have find is the
2858                  * pid of a thread_group_leader.  Testing for task
2859                  * being a thread_group_leader is the obvious thing
2860                  * todo but there is a window when it fails, due to
2861                  * the pid transfer logic in de_thread.
2862                  *
2863                  * So we perform the straight forward test of seeing
2864                  * if the pid we have found is the pid of a thread
2865                  * group leader, and don't worry if the task we have
2866                  * found doesn't happen to be a thread group leader.
2867                  * As we don't care in the case of readdir.
2868                  */
2869                 if (!iter.task || !has_group_leader_pid(iter.task)) {
2870                         iter.tgid += 1;
2871                         goto retry;
2872                 }
2873                 get_task_struct(iter.task);
2874         }
2875         rcu_read_unlock();
2876         return iter;
2877 }
2878
2879 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 1)
2880
2881 /* for the /proc/ directory itself, after non-process stuff has been done */
2882 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
2883 {
2884         struct tgid_iter iter;
2885         struct pid_namespace *ns = file->f_dentry->d_sb->s_fs_info;
2886         loff_t pos = ctx->pos;
2887
2888         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
2889                 return 0;
2890
2891         if (pos == TGID_OFFSET - 1) {
2892                 struct inode *inode = ns->proc_self->d_inode;
2893                 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
2894                         return 0;
2895                 iter.tgid = 0;
2896         } else {
2897                 iter.tgid = pos - TGID_OFFSET;
2898         }
2899         iter.task = NULL;
2900         for (iter = next_tgid(ns, iter);
2901              iter.task;
2902              iter.tgid += 1, iter = next_tgid(ns, iter)) {
2903                 char name[PROC_NUMBUF];
2904                 int len;
2905                 if (!has_pid_permissions(ns, iter.task, 2))
2906                         continue;
2907
2908                 len = snprintf(name, sizeof(name), "%d", iter.tgid);
2909                 ctx->pos = iter.tgid + TGID_OFFSET;
2910                 if (!proc_fill_cache(file, ctx, name, len,
2911                                      proc_pid_instantiate, iter.task, NULL)) {
2912                         put_task_struct(iter.task);
2913                         return 0;
2914                 }
2915         }
2916         ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
2917         return 0;
2918 }
2919
2920 /*
2921  * Tasks
2922  */
2923 static const struct pid_entry tid_base_stuff[] = {
2924         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2925         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2926         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2927         REG("environ",   S_IRUSR, proc_environ_operations),
2928         INF("auxv",      S_IRUSR, proc_pid_auxv),
2929         ONE("status",    S_IRUGO, proc_pid_status),
2930         ONE("personality", S_IRUGO, proc_pid_personality),
2931         INF("limits",    S_IRUGO, proc_pid_limits),
2932 #ifdef CONFIG_SCHED_DEBUG
2933         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2934 #endif
2935         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2936 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2937         INF("syscall",   S_IRUGO, proc_pid_syscall),
2938 #endif
2939         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
2940         ONE("stat",      S_IRUGO, proc_tid_stat),
2941         ONE("statm",     S_IRUGO, proc_pid_statm),
2942         REG("maps",      S_IRUGO, proc_tid_maps_operations),
2943 #ifdef CONFIG_CHECKPOINT_RESTORE
2944         REG("children",  S_IRUGO, proc_tid_children_operations),
2945 #endif
2946 #ifdef CONFIG_NUMA
2947         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
2948 #endif
2949         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
2950         LNK("cwd",       proc_cwd_link),
2951         LNK("root",      proc_root_link),
2952         LNK("exe",       proc_exe_link),
2953         REG("mounts",    S_IRUGO, proc_mounts_operations),
2954         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2955 #ifdef CONFIG_PROC_PAGE_MONITOR
2956         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2957         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
2958         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
2959 #endif
2960 #ifdef CONFIG_SECURITY
2961         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2962 #endif
2963 #ifdef CONFIG_KALLSYMS
2964         INF("wchan",     S_IRUGO, proc_pid_wchan),
2965 #endif
2966 #ifdef CONFIG_STACKTRACE
2967         ONE("stack",      S_IRUGO, proc_pid_stack),
2968 #endif
2969 #ifdef CONFIG_SCHEDSTATS
2970         INF("schedstat", S_IRUGO, proc_pid_schedstat),
2971 #endif
2972 #ifdef CONFIG_LATENCYTOP
2973         REG("latency",  S_IRUGO, proc_lstats_operations),
2974 #endif
2975 #ifdef CONFIG_PROC_PID_CPUSET
2976         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
2977 #endif
2978 #ifdef CONFIG_CGROUPS
2979         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2980 #endif
2981         INF("oom_score", S_IRUGO, proc_oom_score),
2982         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2983         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2984 #ifdef CONFIG_AUDITSYSCALL
2985         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
2986         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2987 #endif
2988 #ifdef CONFIG_FAULT_INJECTION
2989         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2990 #endif
2991 #ifdef CONFIG_TASK_IO_ACCOUNTING
2992         INF("io",       S_IRUSR, proc_tid_io_accounting),
2993 #endif
2994 #ifdef CONFIG_HARDWALL
2995         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
2996 #endif
2997 #ifdef CONFIG_USER_NS
2998         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2999         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3000         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3001 #endif
3002 };
3003
3004 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3005 {
3006         return proc_pident_readdir(file, ctx,
3007                                    tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3008 }
3009
3010 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3011 {
3012         return proc_pident_lookup(dir, dentry,
3013                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3014 }
3015
3016 static const struct file_operations proc_tid_base_operations = {
3017         .read           = generic_read_dir,
3018         .iterate        = proc_tid_base_readdir,
3019         .llseek         = default_llseek,
3020 };
3021
3022 static const struct inode_operations proc_tid_base_inode_operations = {
3023         .lookup         = proc_tid_base_lookup,
3024         .getattr        = pid_getattr,
3025         .setattr        = proc_setattr,
3026 };
3027
3028 static int proc_task_instantiate(struct inode *dir,
3029         struct dentry *dentry, struct task_struct *task, const void *ptr)
3030 {
3031         struct inode *inode;
3032         inode = proc_pid_make_inode(dir->i_sb, task);
3033
3034         if (!inode)
3035                 goto out;
3036         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3037         inode->i_op = &proc_tid_base_inode_operations;
3038         inode->i_fop = &proc_tid_base_operations;
3039         inode->i_flags|=S_IMMUTABLE;
3040
3041         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3042                                                   ARRAY_SIZE(tid_base_stuff)));
3043
3044         d_set_d_op(dentry, &pid_dentry_operations);
3045
3046         d_add(dentry, inode);
3047         /* Close the race of the process dying before we return the dentry */
3048         if (pid_revalidate(dentry, 0))
3049                 return 0;
3050 out:
3051         return -ENOENT;
3052 }
3053
3054 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3055 {
3056         int result = -ENOENT;
3057         struct task_struct *task;
3058         struct task_struct *leader = get_proc_task(dir);
3059         unsigned tid;
3060         struct pid_namespace *ns;
3061
3062         if (!leader)
3063                 goto out_no_task;
3064
3065         tid = name_to_int(dentry);
3066         if (tid == ~0U)
3067                 goto out;
3068
3069         ns = dentry->d_sb->s_fs_info;
3070         rcu_read_lock();
3071         task = find_task_by_pid_ns(tid, ns);
3072         if (task)
3073                 get_task_struct(task);
3074         rcu_read_unlock();
3075         if (!task)
3076                 goto out;
3077         if (!same_thread_group(leader, task))
3078                 goto out_drop_task;
3079
3080         result = proc_task_instantiate(dir, dentry, task, NULL);
3081 out_drop_task:
3082         put_task_struct(task);
3083 out:
3084         put_task_struct(leader);
3085 out_no_task:
3086         return ERR_PTR(result);
3087 }
3088
3089 /*
3090  * Find the first tid of a thread group to return to user space.
3091  *
3092  * Usually this is just the thread group leader, but if the users
3093  * buffer was too small or there was a seek into the middle of the
3094  * directory we have more work todo.
3095  *
3096  * In the case of a short read we start with find_task_by_pid.
3097  *
3098  * In the case of a seek we start with the leader and walk nr
3099  * threads past it.
3100  */
3101 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3102                                         struct pid_namespace *ns)
3103 {
3104         struct task_struct *pos, *task;
3105         unsigned long nr = f_pos;
3106
3107         if (nr != f_pos)        /* 32bit overflow? */
3108                 return NULL;
3109
3110         rcu_read_lock();
3111         task = pid_task(pid, PIDTYPE_PID);
3112         if (!task)
3113                 goto fail;
3114
3115         /* Attempt to start with the tid of a thread */
3116         if (tid && nr) {
3117                 pos = find_task_by_pid_ns(tid, ns);
3118                 if (pos && same_thread_group(pos, task))
3119                         goto found;
3120         }
3121
3122         /* If nr exceeds the number of threads there is nothing todo */
3123         if (nr >= get_nr_threads(task))
3124                 goto fail;
3125
3126         /* If we haven't found our starting place yet start
3127          * with the leader and walk nr threads forward.
3128          */
3129         pos = task = task->group_leader;
3130         do {
3131                 if (!nr--)
3132                         goto found;
3133         } while_each_thread(task, pos);
3134 fail:
3135         pos = NULL;
3136         goto out;
3137 found:
3138         get_task_struct(pos);
3139 out:
3140         rcu_read_unlock();
3141         return pos;
3142 }
3143
3144 /*
3145  * Find the next thread in the thread list.
3146  * Return NULL if there is an error or no next thread.
3147  *
3148  * The reference to the input task_struct is released.
3149  */
3150 static struct task_struct *next_tid(struct task_struct *start)
3151 {
3152         struct task_struct *pos = NULL;
3153         rcu_read_lock();
3154         if (pid_alive(start)) {
3155                 pos = next_thread(start);
3156                 if (thread_group_leader(pos))
3157                         pos = NULL;
3158                 else
3159                         get_task_struct(pos);
3160         }
3161         rcu_read_unlock();
3162         put_task_struct(start);
3163         return pos;
3164 }
3165
3166 /* for the /proc/TGID/task/ directories */
3167 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3168 {
3169         struct inode *inode = file_inode(file);
3170         struct task_struct *task;
3171         struct pid_namespace *ns;
3172         int tid;
3173
3174         if (proc_inode_is_dead(inode))
3175                 return -ENOENT;
3176
3177         if (!dir_emit_dots(file, ctx))
3178                 return 0;
3179
3180         /* f_version caches the tgid value that the last readdir call couldn't
3181          * return. lseek aka telldir automagically resets f_version to 0.
3182          */
3183         ns = file->f_dentry->d_sb->s_fs_info;
3184         tid = (int)file->f_version;
3185         file->f_version = 0;
3186         for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3187              task;
3188              task = next_tid(task), ctx->pos++) {
3189                 char name[PROC_NUMBUF];
3190                 int len;
3191                 tid = task_pid_nr_ns(task, ns);
3192                 len = snprintf(name, sizeof(name), "%d", tid);
3193                 if (!proc_fill_cache(file, ctx, name, len,
3194                                 proc_task_instantiate, task, NULL)) {
3195                         /* returning this tgid failed, save it as the first
3196                          * pid for the next readir call */
3197                         file->f_version = (u64)tid;
3198                         put_task_struct(task);
3199                         break;
3200                 }
3201         }
3202
3203         return 0;
3204 }
3205
3206 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3207 {
3208         struct inode *inode = dentry->d_inode;
3209         struct task_struct *p = get_proc_task(inode);
3210         generic_fillattr(inode, stat);
3211
3212         if (p) {
3213                 stat->nlink += get_nr_threads(p);
3214                 put_task_struct(p);
3215         }
3216
3217         return 0;
3218 }
3219
3220 static const struct inode_operations proc_task_inode_operations = {
3221         .lookup         = proc_task_lookup,
3222         .getattr        = proc_task_getattr,
3223         .setattr        = proc_setattr,
3224         .permission     = proc_pid_permission,
3225 };
3226
3227 static const struct file_operations proc_task_operations = {
3228         .read           = generic_read_dir,
3229         .iterate        = proc_task_readdir,
3230         .llseek         = default_llseek,
3231 };