fs/proc/base.c

   1 /*
   2  *  linux/fs/proc/base.c
   3  *
   4  *  Copyright (C) 1991, 1992 Linus Torvalds
   5  *
   6  *  proc base directory handling functions
   7  *
   8  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
   9  *  Instead of using magical inumbers to determine the kind of object
  10  *  we allocate and fill in-core inodes upon lookup. They don't even
  11  *  go into icache. We cache the reference to task_struct upon lookup too.
  12  *  Eventually it should become a filesystem in its own. We don't use the
  13  *  rest of procfs anymore.
  14  *
  15  *
  16  *  Changelog:
  17  *  17-Jan-2005
  18  *  Allan Bezerra
  19  *  Bruna Moreira <bruna.moreira@indt.org.br>
  20  *  Edjard Mota <edjard.mota@indt.org.br>
  21  *  Ilias Biris <ilias.biris@indt.org.br>
  22  *  Mauricio Lin <mauricio.lin@indt.org.br>
  23  *
  24  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  25  *
  26  *  A new process specific entry (smaps) included in /proc. It shows the
  27  *  size of rss for each memory area. The maps entry lacks information
  28  *  about physical memory size (rss) for each mapped file, i.e.,
  29  *  rss information for executables and library files.
  30  *  This additional information is useful for any tools that need to know
  31  *  about physical memory consumption for a process specific library.
  32  *
  33  *  Changelog:
  34  *  21-Feb-2005
  35  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  36  *  Pud inclusion in the page table walking.
  37  *
  38  *  ChangeLog:
  39  *  10-Mar-2005
  40  *  10LE Instituto Nokia de Tecnologia - INdT:
  41  *  A better way to walks through the page table as suggested by Hugh Dickins.
  42  *
  43  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
  44  *  Smaps information related to shared, private, clean and dirty pages.
  45  *
  46  *  Paul Mundt <paul.mundt@nokia.com>:
  47  *  Overall revision about smaps.
  48  */
  49
  50 #include <asm/uaccess.h>
  51
  52 #include <linux/errno.h>
  53 #include <linux/time.h>
  54 #include <linux/proc_fs.h>
  55 #include <linux/stat.h>
  56 #include <linux/task_io_accounting_ops.h>
  57 #include <linux/init.h>
  58 #include <linux/capability.h>
  59 #include <linux/file.h>
  60 #include <linux/fdtable.h>
  61 #include <linux/string.h>
  62 #include <linux/seq_file.h>
  63 #include <linux/namei.h>
  64 #include <linux/mnt_namespace.h>
  65 #include <linux/mm.h>
  66 #include <linux/swap.h>
  67 #include <linux/rcupdate.h>
  68 #include <linux/kallsyms.h>
  69 #include <linux/stacktrace.h>
  70 #include <linux/resource.h>
  71 #include <linux/module.h>
  72 #include <linux/mount.h>
  73 #include <linux/security.h>
  74 #include <linux/ptrace.h>
  75 #include <linux/tracehook.h>
  76 #include <linux/printk.h>
  77 #include <linux/cgroup.h>
  78 #include <linux/cpuset.h>
  79 #include <linux/audit.h>
  80 #include <linux/poll.h>
  81 #include <linux/nsproxy.h>
  82 #include <linux/oom.h>
  83 #include <linux/elf.h>
  84 #include <linux/pid_namespace.h>
  85 #include <linux/user_namespace.h>
  86 #include <linux/fs_struct.h>
  87 #include <linux/slab.h>
  88 #include <linux/flex_array.h>
  89 #include <linux/posix-timers.h>
  90 #ifdef CONFIG_HARDWALL
  91 #include <asm/hardwall.h>
  92 #endif
  93 #include <trace/events/oom.h>
  94 #include "internal.h"
  95 #include "fd.h"
  96
  97 /* NOTE:
  98  *      Implementing inode permission operations in /proc is almost
  99  *      certainly an error.  Permission checks need to happen during
 100  *      each system call not at open time.  The reason is that most of
 101  *      what we wish to check for permissions in /proc varies at runtime.
 102  *
 103  *      The classic example of a problem is opening file descriptors
 104  *      in /proc for a task before it execs a suid executable.
 105  */
 106
 107 struct pid_entry {
 108         char *name;
 109         int len;
 110         umode_t mode;
 111         const struct inode_operations *iop;
 112         const struct file_operations *fop;
 113         union proc_op op;
 114 };
 115
 116 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 117         .name = (NAME),                                 \
 118         .len  = sizeof(NAME) - 1,                       \
 119         .mode = MODE,                                   \
 120         .iop  = IOP,                                    \
 121         .fop  = FOP,                                    \
 122         .op   = OP,                                     \
 123 }
 124
 125 #define DIR(NAME, MODE, iops, fops)     \
 126         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 127 #define LNK(NAME, get_link)                                     \
 128         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 129                 &proc_pid_link_inode_operations, NULL,          \
 130                 { .proc_get_link = get_link } )
 131 #define REG(NAME, MODE, fops)                           \
 132         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 133 #define INF(NAME, MODE, read)                           \
 134         NOD(NAME, (S_IFREG|(MODE)),                     \
 135                 NULL, &proc_info_file_operations,       \
 136                 { .proc_read = read } )
 137 #define ONE(NAME, MODE, show)                           \
 138         NOD(NAME, (S_IFREG|(MODE)),                     \
 139                 NULL, &proc_single_file_operations,     \
 140                 { .proc_show = show } )
 141
 142 /*
 143  * Count the number of hardlinks for the pid_entry table, excluding the .
 144  * and .. links.
 145  */
 146 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
 147         unsigned int n)
 148 {
 149         unsigned int i;
 150         unsigned int count;
 151
 152         count = 0;
 153         for (i = 0; i < n; ++i) {
 154                 if (S_ISDIR(entries[i].mode))
 155                         ++count;
 156         }
 157
 158         return count;
 159 }
 160
 161 static int get_task_root(struct task_struct *task, struct path *root)
 162 {
 163         int result = -ENOENT;
 164
 165         task_lock(task);
 166         if (task->fs) {
 167                 get_fs_root(task->fs, root);
 168                 result = 0;
 169         }
 170         task_unlock(task);
 171         return result;
 172 }
 173
 174 static int proc_cwd_link(struct dentry *dentry, struct path *path)
 175 {
 176         struct task_struct *task = get_proc_task(dentry->d_inode);
 177         int result = -ENOENT;
 178
 179         if (task) {
 180                 task_lock(task);
 181                 if (task->fs) {
 182                         get_fs_pwd(task->fs, path);
 183                         result = 0;
 184                 }
 185                 task_unlock(task);
 186                 put_task_struct(task);
 187         }
 188         return result;
 189 }
 190
 191 static int proc_root_link(struct dentry *dentry, struct path *path)
 192 {
 193         struct task_struct *task = get_proc_task(dentry->d_inode);
 194         int result = -ENOENT;
 195
 196         if (task) {
 197                 result = get_task_root(task, path);
 198                 put_task_struct(task);
 199         }
 200         return result;
 201 }
 202
 203 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
 204 {
 205         int res = 0;
 206         unsigned int len;
 207         struct mm_struct *mm = get_task_mm(task);
 208         if (!mm)
 209                 goto out;
 210         if (!mm->arg_end)
 211                 goto out_mm;    /* Shh! No looking before we're done */
 212
 213         len = mm->arg_end - mm->arg_start;
 214
 215         if (len > PAGE_SIZE)
 216                 len = PAGE_SIZE;
 217
 218         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
 219
 220         // If the nul at the end of args has been overwritten, then
 221         // assume application is using setproctitle(3).
 222         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
 223                 len = strnlen(buffer, res);
 224                 if (len < res) {
 225                     res = len;
 226                 } else {
 227                         len = mm->env_end - mm->env_start;
 228                         if (len > PAGE_SIZE - res)
 229                                 len = PAGE_SIZE - res;
 230                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
 231                         res = strnlen(buffer, res);
 232                 }
 233         }
 234 out_mm:
 235         mmput(mm);
 236 out:
 237         return res;
 238 }
 239
 240 static int proc_pid_auxv(struct task_struct *task, char *buffer)
 241 {
 242         struct mm_struct *mm = mm_access(task, PTRACE_MODE_READ);
 243         int res = PTR_ERR(mm);
 244         if (mm && !IS_ERR(mm)) {
 245                 unsigned int nwords = 0;
 246                 do {
 247                         nwords += 2;
 248                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 249                 res = nwords * sizeof(mm->saved_auxv[0]);
 250                 if (res > PAGE_SIZE)
 251                         res = PAGE_SIZE;
 252                 memcpy(buffer, mm->saved_auxv, res);
 253                 mmput(mm);
 254         }
 255         return res;
 256 }
 257
 258
 259 #ifdef CONFIG_KALLSYMS
 260 /*
 261  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 262  * Returns the resolved symbol.  If that fails, simply return the address.
 263  */
 264 static int proc_pid_wchan(struct task_struct *task, char *buffer)
 265 {
 266         unsigned long wchan;
 267         char symname[KSYM_NAME_LEN];
 268
 269         wchan = get_wchan(task);
 270
 271         if (lookup_symbol_name(wchan, symname) < 0)
 272                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
 273                         return 0;
 274                 else
 275                         return sprintf(buffer, "%lu", wchan);
 276         else
 277                 return sprintf(buffer, "%s", symname);
 278 }
 279 #endif /* CONFIG_KALLSYMS */
 280
 281 static int lock_trace(struct task_struct *task)
 282 {
 283         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
 284         if (err)
 285                 return err;
 286         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH)) {
 287                 mutex_unlock(&task->signal->cred_guard_mutex);
 288                 return -EPERM;
 289         }
 290         return 0;
 291 }
 292
 293 static void unlock_trace(struct task_struct *task)
 294 {
 295         mutex_unlock(&task->signal->cred_guard_mutex);
 296 }
 297
 298 #ifdef CONFIG_STACKTRACE
 299
 300 #define MAX_STACK_TRACE_DEPTH   64
 301
 302 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 303                           struct pid *pid, struct task_struct *task)
 304 {
 305         struct stack_trace trace;
 306         unsigned long *entries;
 307         int err;
 308         int i;
 309
 310         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 311         if (!entries)
 312                 return -ENOMEM;
 313
 314         trace.nr_entries        = 0;
 315         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 316         trace.entries           = entries;
 317         trace.skip              = 0;
 318
 319         err = lock_trace(task);
 320         if (!err) {
 321                 save_stack_trace_tsk(task, &trace);
 322
 323                 for (i = 0; i < trace.nr_entries; i++) {
 324                         seq_printf(m, "[<%pK>] %pS\n",
 325                                    (void *)entries[i], (void *)entries[i]);
 326                 }
 327                 unlock_trace(task);
 328         }
 329         kfree(entries);
 330
 331         return err;
 332 }
 333 #endif
 334
 335 #ifdef CONFIG_SCHEDSTATS
 336 /*
 337  * Provides /proc/PID/schedstat
 338  */
 339 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
 340 {
 341         return sprintf(buffer, "%llu %llu %lu\n",
 342                         (unsigned long long)task->se.sum_exec_runtime,
 343                         (unsigned long long)task->sched_info.run_delay,
 344                         task->sched_info.pcount);
 345 }
 346 #endif
 347
 348 #ifdef CONFIG_LATENCYTOP
 349 static int lstats_show_proc(struct seq_file *m, void *v)
 350 {
 351         int i;
 352         struct inode *inode = m->private;
 353         struct task_struct *task = get_proc_task(inode);
 354
 355         if (!task)
 356                 return -ESRCH;
 357         seq_puts(m, "Latency Top version : v0.1\n");
 358         for (i = 0; i < 32; i++) {
 359                 struct latency_record *lr = &task->latency_record[i];
 360                 if (lr->backtrace[0]) {
 361                         int q;
 362                         seq_printf(m, "%i %li %li",
 363                                    lr->count, lr->time, lr->max);
 364                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 365                                 unsigned long bt = lr->backtrace[q];
 366                                 if (!bt)
 367                                         break;
 368                                 if (bt == ULONG_MAX)
 369                                         break;
 370                                 seq_printf(m, " %ps", (void *)bt);
 371                         }
 372                         seq_putc(m, '\n');
 373                 }
 374
 375         }
 376         put_task_struct(task);
 377         return 0;
 378 }
 379
 380 static int lstats_open(struct inode *inode, struct file *file)
 381 {
 382         return single_open(file, lstats_show_proc, inode);
 383 }
 384
 385 static ssize_t lstats_write(struct file *file, const char __user *buf,
 386                             size_t count, loff_t *offs)
 387 {
 388         struct task_struct *task = get_proc_task(file_inode(file));
 389
 390         if (!task)
 391                 return -ESRCH;
 392         clear_all_latency_tracing(task);
 393         put_task_struct(task);
 394
 395         return count;
 396 }
 397
 398 static const struct file_operations proc_lstats_operations = {
 399         .open           = lstats_open,
 400         .read           = seq_read,
 401         .write          = lstats_write,
 402         .llseek         = seq_lseek,
 403         .release        = single_release,
 404 };
 405
 406 #endif
 407
 408 static int proc_oom_score(struct task_struct *task, char *buffer)
 409 {
 410         unsigned long totalpages = totalram_pages + total_swap_pages;
 411         unsigned long points = 0;
 412
 413         read_lock(&tasklist_lock);
 414         if (pid_alive(task))
 415                 points = oom_badness(task, NULL, NULL, totalpages) *
 416                                                 1000 / totalpages;
 417         read_unlock(&tasklist_lock);
 418         return sprintf(buffer, "%lu\n", points);
 419 }
 420
 421 struct limit_names {
 422         char *name;
 423         char *unit;
 424 };
 425
 426 static const struct limit_names lnames[RLIM_NLIMITS] = {
 427         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 428         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 429         [RLIMIT_DATA] = {"Max data size", "bytes"},
 430         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 431         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 432         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 433         [RLIMIT_NPROC] = {"Max processes", "processes"},
 434         [RLIMIT_NOFILE] = {"Max open files", "files"},
 435         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 436         [RLIMIT_AS] = {"Max address space", "bytes"},
 437         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 438         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 439         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 440         [RLIMIT_NICE] = {"Max nice priority", NULL},
 441         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 442         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 443 };
 444
 445 /* Display limits for a process */
 446 static int proc_pid_limits(struct task_struct *task, char *buffer)
 447 {
 448         unsigned int i;
 449         int count = 0;
 450         unsigned long flags;
 451         char *bufptr = buffer;
 452
 453         struct rlimit rlim[RLIM_NLIMITS];
 454
 455         if (!lock_task_sighand(task, &flags))
 456                 return 0;
 457         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 458         unlock_task_sighand(task, &flags);
 459
 460         /*
 461          * print the file header
 462          */
 463         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
 464                         "Limit", "Soft Limit", "Hard Limit", "Units");
 465
 466         for (i = 0; i < RLIM_NLIMITS; i++) {
 467                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 468                         count += sprintf(&bufptr[count], "%-25s %-20s ",
 469                                          lnames[i].name, "unlimited");
 470                 else
 471                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
 472                                          lnames[i].name, rlim[i].rlim_cur);
 473
 474                 if (rlim[i].rlim_max == RLIM_INFINITY)
 475                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
 476                 else
 477                         count += sprintf(&bufptr[count], "%-20lu ",
 478                                          rlim[i].rlim_max);
 479
 480                 if (lnames[i].unit)
 481                         count += sprintf(&bufptr[count], "%-10s\n",
 482                                          lnames[i].unit);
 483                 else
 484                         count += sprintf(&bufptr[count], "\n");
 485         }
 486
 487         return count;
 488 }
 489
 490 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 491 static int proc_pid_syscall(struct task_struct *task, char *buffer)
 492 {
 493         long nr;
 494         unsigned long args[6], sp, pc;
 495         int res = lock_trace(task);
 496         if (res)
 497                 return res;
 498
 499         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 500                 res = sprintf(buffer, "running\n");
 501         else if (nr < 0)
 502                 res = sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 503         else
 504                 res = sprintf(buffer,
 505                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 506                        nr,
 507                        args[0], args[1], args[2], args[3], args[4], args[5],
 508                        sp, pc);
 509         unlock_trace(task);
 510         return res;
 511 }
 512 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 513
 514 /************************************************************************/
 515 /*                       Here the fs part begins                        */
 516 /************************************************************************/
 517
 518 /* permission checks */
 519 static int proc_fd_access_allowed(struct inode *inode)
 520 {
 521         struct task_struct *task;
 522         int allowed = 0;
 523         /* Allow access to a task's file descriptors if it is us or we
 524          * may use ptrace attach to the process and find out that
 525          * information.
 526          */
 527         task = get_proc_task(inode);
 528         if (task) {
 529                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
 530                 put_task_struct(task);
 531         }
 532         return allowed;
 533 }
 534
 535 int proc_setattr(struct dentry *dentry, struct iattr *attr)
 536 {
 537         int error;
 538         struct inode *inode = dentry->d_inode;
 539
 540         if (attr->ia_valid & ATTR_MODE)
 541                 return -EPERM;
 542
 543         error = inode_change_ok(inode, attr);
 544         if (error)
 545                 return error;
 546
 547         setattr_copy(inode, attr);
 548         mark_inode_dirty(inode);
 549         return 0;
 550 }
 551
 552 /*
 553  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
 554  * or euid/egid (for hide_pid_min=2)?
 555  */
 556 static bool has_pid_permissions(struct pid_namespace *pid,
 557                                  struct task_struct *task,
 558                                  int hide_pid_min)
 559 {
 560         if (pid->hide_pid < hide_pid_min)
 561                 return true;
 562         if (in_group_p(pid->pid_gid))
 563                 return true;
 564         return ptrace_may_access(task, PTRACE_MODE_READ);
 565 }
 566
 567
 568 static int proc_pid_permission(struct inode *inode, int mask)
 569 {
 570         struct pid_namespace *pid = inode->i_sb->s_fs_info;
 571         struct task_struct *task;
 572         bool has_perms;
 573
 574         task = get_proc_task(inode);
 575         if (!task)
 576                 return -ESRCH;
 577         has_perms = has_pid_permissions(pid, task, 1);
 578         put_task_struct(task);
 579
 580         if (!has_perms) {
 581                 if (pid->hide_pid == 2) {
 582                         /*
 583                          * Let's make getdents(), stat(), and open()
 584                          * consistent with each other.  If a process
 585                          * may not stat() a file, it shouldn't be seen
 586                          * in procfs at all.
 587                          */
 588                         return -ENOENT;
 589                 }
 590
 591                 return -EPERM;
 592         }
 593         return generic_permission(inode, mask);
 594 }
 595
 596
 597
 598 static const struct inode_operations proc_def_inode_operations = {
 599         .setattr        = proc_setattr,
 600 };
 601
 602 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 603
 604 static ssize_t proc_info_read(struct file * file, char __user * buf,
 605                           size_t count, loff_t *ppos)
 606 {
 607         struct inode * inode = file_inode(file);
 608         unsigned long page;
 609         ssize_t length;
 610         struct task_struct *task = get_proc_task(inode);
 611
 612         length = -ESRCH;
 613         if (!task)
 614                 goto out_no_task;
 615
 616         if (count > PROC_BLOCK_SIZE)
 617                 count = PROC_BLOCK_SIZE;
 618
 619         length = -ENOMEM;
 620         if (!(page = __get_free_page(GFP_TEMPORARY)))
 621                 goto out;
 622
 623         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 624
 625         if (length >= 0)
 626                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 627         free_page(page);
 628 out:
 629         put_task_struct(task);
 630 out_no_task:
 631         return length;
 632 }
 633
 634 static const struct file_operations proc_info_file_operations = {
 635         .read           = proc_info_read,
 636         .llseek         = generic_file_llseek,
 637 };
 638
 639 static int proc_single_show(struct seq_file *m, void *v)
 640 {
 641         struct inode *inode = m->private;
 642         struct pid_namespace *ns;
 643         struct pid *pid;
 644         struct task_struct *task;
 645         int ret;
 646
 647         ns = inode->i_sb->s_fs_info;
 648         pid = proc_pid(inode);
 649         task = get_pid_task(pid, PIDTYPE_PID);
 650         if (!task)
 651                 return -ESRCH;
 652
 653         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 654
 655         put_task_struct(task);
 656         return ret;
 657 }
 658
 659 static int proc_single_open(struct inode *inode, struct file *filp)
 660 {
 661         return single_open(filp, proc_single_show, inode);
 662 }
 663
 664 static const struct file_operations proc_single_file_operations = {
 665         .open           = proc_single_open,
 666         .read           = seq_read,
 667         .llseek         = seq_lseek,
 668         .release        = single_release,
 669 };
 670
 671 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
 672 {
 673         struct task_struct *task = get_proc_task(file_inode(file));
 674         struct mm_struct *mm;
 675
 676         if (!task)
 677                 return -ESRCH;
 678
 679         mm = mm_access(task, mode);
 680         put_task_struct(task);
 681
 682         if (IS_ERR(mm))
 683                 return PTR_ERR(mm);
 684
 685         if (mm) {
 686                 /* ensure this mm_struct can't be freed */
 687                 atomic_inc(&mm->mm_count);
 688                 /* but do not pin its memory */
 689                 mmput(mm);
 690         }
 691
 692         file->private_data = mm;
 693
 694         return 0;
 695 }
 696
 697 static int mem_open(struct inode *inode, struct file *file)
 698 {
 699         int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
 700
 701         /* OK to pass negative loff_t, we can catch out-of-range */
 702         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 703
 704         return ret;
 705 }
 706
 707 static ssize_t mem_rw(struct file *file, char __user *buf,
 708                         size_t count, loff_t *ppos, int write)
 709 {
 710         struct mm_struct *mm = file->private_data;
 711         unsigned long addr = *ppos;
 712         ssize_t copied;
 713         char *page;
 714
 715         if (!mm)
 716                 return 0;
 717
 718         page = (char *)__get_free_page(GFP_TEMPORARY);
 719         if (!page)
 720                 return -ENOMEM;
 721
 722         copied = 0;
 723         if (!atomic_inc_not_zero(&mm->mm_users))
 724                 goto free;
 725
 726         while (count > 0) {
 727                 int this_len = min_t(int, count, PAGE_SIZE);
 728
 729                 if (write && copy_from_user(page, buf, this_len)) {
 730                         copied = -EFAULT;
 731                         break;
 732                 }
 733
 734                 this_len = access_remote_vm(mm, addr, page, this_len, write);
 735                 if (!this_len) {
 736                         if (!copied)
 737                                 copied = -EIO;
 738                         break;
 739                 }
 740
 741                 if (!write && copy_to_user(buf, page, this_len)) {
 742                         copied = -EFAULT;
 743                         break;
 744                 }
 745
 746                 buf += this_len;
 747                 addr += this_len;
 748                 copied += this_len;
 749                 count -= this_len;
 750         }
 751         *ppos = addr;
 752
 753         mmput(mm);
 754 free:
 755         free_page((unsigned long) page);
 756         return copied;
 757 }
 758
 759 static ssize_t mem_read(struct file *file, char __user *buf,
 760                         size_t count, loff_t *ppos)
 761 {
 762         return mem_rw(file, buf, count, ppos, 0);
 763 }
 764
 765 static ssize_t mem_write(struct file *file, const char __user *buf,
 766                          size_t count, loff_t *ppos)
 767 {
 768         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 769 }
 770
 771 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 772 {
 773         switch (orig) {
 774         case 0:
 775                 file->f_pos = offset;
 776                 break;
 777         case 1:
 778                 file->f_pos += offset;
 779                 break;
 780         default:
 781                 return -EINVAL;
 782         }
 783         force_successful_syscall_return();
 784         return file->f_pos;
 785 }
 786
 787 static int mem_release(struct inode *inode, struct file *file)
 788 {
 789         struct mm_struct *mm = file->private_data;
 790         if (mm)
 791                 mmdrop(mm);
 792         return 0;
 793 }
 794
 795 static const struct file_operations proc_mem_operations = {
 796         .llseek         = mem_lseek,
 797         .read           = mem_read,
 798         .write          = mem_write,
 799         .open           = mem_open,
 800         .release        = mem_release,
 801 };
 802
 803 static int environ_open(struct inode *inode, struct file *file)
 804 {
 805         return __mem_open(inode, file, PTRACE_MODE_READ);
 806 }
 807
 808 static ssize_t environ_read(struct file *file, char __user *buf,
 809                         size_t count, loff_t *ppos)
 810 {
 811         char *page;
 812         unsigned long src = *ppos;
 813         int ret = 0;
 814         struct mm_struct *mm = file->private_data;
 815
 816         if (!mm)
 817                 return 0;
 818
 819         page = (char *)__get_free_page(GFP_TEMPORARY);
 820         if (!page)
 821                 return -ENOMEM;
 822
 823         ret = 0;
 824         if (!atomic_inc_not_zero(&mm->mm_users))
 825                 goto free;
 826         while (count > 0) {
 827                 size_t this_len, max_len;
 828                 int retval;
 829
 830                 if (src >= (mm->env_end - mm->env_start))
 831                         break;
 832
 833                 this_len = mm->env_end - (mm->env_start + src);
 834
 835                 max_len = min_t(size_t, PAGE_SIZE, count);
 836                 this_len = min(max_len, this_len);
 837
 838                 retval = access_remote_vm(mm, (mm->env_start + src),
 839                         page, this_len, 0);
 840
 841                 if (retval <= 0) {
 842                         ret = retval;
 843                         break;
 844                 }
 845
 846                 if (copy_to_user(buf, page, retval)) {
 847                         ret = -EFAULT;
 848                         break;
 849                 }
 850
 851                 ret += retval;
 852                 src += retval;
 853                 buf += retval;
 854                 count -= retval;
 855         }
 856         *ppos = src;
 857         mmput(mm);
 858
 859 free:
 860         free_page((unsigned long) page);
 861         return ret;
 862 }
 863
 864 static const struct file_operations proc_environ_operations = {
 865         .open           = environ_open,
 866         .read           = environ_read,
 867         .llseek         = generic_file_llseek,
 868         .release        = mem_release,
 869 };
 870
 871 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
 872                             loff_t *ppos)
 873 {
 874         struct task_struct *task = get_proc_task(file_inode(file));
 875         char buffer[PROC_NUMBUF];
 876         int oom_adj = OOM_ADJUST_MIN;
 877         size_t len;
 878         unsigned long flags;
 879
 880         if (!task)
 881                 return -ESRCH;
 882         if (lock_task_sighand(task, &flags)) {
 883                 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
 884                         oom_adj = OOM_ADJUST_MAX;
 885                 else
 886                         oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
 887                                   OOM_SCORE_ADJ_MAX;
 888                 unlock_task_sighand(task, &flags);
 889         }
 890         put_task_struct(task);
 891         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
 892         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 893 }
 894
 895 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
 896                              size_t count, loff_t *ppos)
 897 {
 898         struct task_struct *task;
 899         char buffer[PROC_NUMBUF];
 900         int oom_adj;
 901         unsigned long flags;
 902         int err;
 903
 904         memset(buffer, 0, sizeof(buffer));
 905         if (count > sizeof(buffer) - 1)
 906                 count = sizeof(buffer) - 1;
 907         if (copy_from_user(buffer, buf, count)) {
 908                 err = -EFAULT;
 909                 goto out;
 910         }
 911
 912         err = kstrtoint(strstrip(buffer), 0, &oom_adj);
 913         if (err)
 914                 goto out;
 915         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
 916              oom_adj != OOM_DISABLE) {
 917                 err = -EINVAL;
 918                 goto out;
 919         }
 920
 921         task = get_proc_task(file_inode(file));
 922         if (!task) {
 923                 err = -ESRCH;
 924                 goto out;
 925         }
 926
 927         task_lock(task);
 928         if (!task->mm) {
 929                 err = -EINVAL;
 930                 goto err_task_lock;
 931         }
 932
 933         if (!lock_task_sighand(task, &flags)) {
 934                 err = -ESRCH;
 935                 goto err_task_lock;
 936         }
 937
 938         /*
 939          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
 940          * value is always attainable.
 941          */
 942         if (oom_adj == OOM_ADJUST_MAX)
 943                 oom_adj = OOM_SCORE_ADJ_MAX;
 944         else
 945                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
 946
 947         if (oom_adj < task->signal->oom_score_adj &&
 948             !capable(CAP_SYS_RESOURCE)) {
 949                 err = -EACCES;
 950                 goto err_sighand;
 951         }
 952
 953         /*
 954          * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
 955          * /proc/pid/oom_score_adj instead.
 956          */
 957         pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
 958                   current->comm, task_pid_nr(current), task_pid_nr(task),
 959                   task_pid_nr(task));
 960
 961         task->signal->oom_score_adj = oom_adj;
 962         trace_oom_score_adj_update(task);
 963 err_sighand:
 964         unlock_task_sighand(task, &flags);
 965 err_task_lock:
 966         task_unlock(task);
 967         put_task_struct(task);
 968 out:
 969         return err < 0 ? err : count;
 970 }
 971
 972 static const struct file_operations proc_oom_adj_operations = {
 973         .read           = oom_adj_read,
 974         .write          = oom_adj_write,
 975         .llseek         = generic_file_llseek,
 976 };
 977
 978 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
 979                                         size_t count, loff_t *ppos)
 980 {
 981         struct task_struct *task = get_proc_task(file_inode(file));
 982         char buffer[PROC_NUMBUF];
 983         short oom_score_adj = OOM_SCORE_ADJ_MIN;
 984         unsigned long flags;
 985         size_t len;
 986
 987         if (!task)
 988                 return -ESRCH;
 989         if (lock_task_sighand(task, &flags)) {
 990                 oom_score_adj = task->signal->oom_score_adj;
 991                 unlock_task_sighand(task, &flags);
 992         }
 993         put_task_struct(task);
 994         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
 995         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 996 }
 997
 998 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
 999                                         size_t count, loff_t *ppos)
1000 {
1001         struct task_struct *task;
1002         char buffer[PROC_NUMBUF];
1003         unsigned long flags;
1004         int oom_score_adj;
1005         int err;
1006
1007         memset(buffer, 0, sizeof(buffer));
1008         if (count > sizeof(buffer) - 1)
1009                 count = sizeof(buffer) - 1;
1010         if (copy_from_user(buffer, buf, count)) {
1011                 err = -EFAULT;
1012                 goto out;
1013         }
1014
1015         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1016         if (err)
1017                 goto out;
1018         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1019                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1020                 err = -EINVAL;
1021                 goto out;
1022         }
1023
1024         task = get_proc_task(file_inode(file));
1025         if (!task) {
1026                 err = -ESRCH;
1027                 goto out;
1028         }
1029
1030         task_lock(task);
1031         if (!task->mm) {
1032                 err = -EINVAL;
1033                 goto err_task_lock;
1034         }
1035
1036         if (!lock_task_sighand(task, &flags)) {
1037                 err = -ESRCH;
1038                 goto err_task_lock;
1039         }
1040
1041         if ((short)oom_score_adj < task->signal->oom_score_adj_min &&
1042                         !capable(CAP_SYS_RESOURCE)) {
1043                 err = -EACCES;
1044                 goto err_sighand;
1045         }
1046
1047         task->signal->oom_score_adj = (short)oom_score_adj;
1048         if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1049                 task->signal->oom_score_adj_min = (short)oom_score_adj;
1050         trace_oom_score_adj_update(task);
1051
1052 err_sighand:
1053         unlock_task_sighand(task, &flags);
1054 err_task_lock:
1055         task_unlock(task);
1056         put_task_struct(task);
1057 out:
1058         return err < 0 ? err : count;
1059 }
1060
1061 static const struct file_operations proc_oom_score_adj_operations = {
1062         .read           = oom_score_adj_read,
1063         .write          = oom_score_adj_write,
1064         .llseek         = default_llseek,
1065 };
1066
1067 #ifdef CONFIG_AUDITSYSCALL
1068 #define TMPBUFLEN 21
1069 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1070                                   size_t count, loff_t *ppos)
1071 {
1072         struct inode * inode = file_inode(file);
1073         struct task_struct *task = get_proc_task(inode);
1074         ssize_t length;
1075         char tmpbuf[TMPBUFLEN];
1076
1077         if (!task)
1078                 return -ESRCH;
1079         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1080                            from_kuid(file->f_cred->user_ns,
1081                                      audit_get_loginuid(task)));
1082         put_task_struct(task);
1083         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1084 }
1085
1086 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1087                                    size_t count, loff_t *ppos)
1088 {
1089         struct inode * inode = file_inode(file);
1090         char *page, *tmp;
1091         ssize_t length;
1092         uid_t loginuid;
1093         kuid_t kloginuid;
1094
1095         rcu_read_lock();
1096         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1097                 rcu_read_unlock();
1098                 return -EPERM;
1099         }
1100         rcu_read_unlock();
1101
1102         if (count >= PAGE_SIZE)
1103                 count = PAGE_SIZE - 1;
1104
1105         if (*ppos != 0) {
1106                 /* No partial writes. */
1107                 return -EINVAL;
1108         }
1109         page = (char*)__get_free_page(GFP_TEMPORARY);
1110         if (!page)
1111                 return -ENOMEM;
1112         length = -EFAULT;
1113         if (copy_from_user(page, buf, count))
1114                 goto out_free_page;
1115
1116         page[count] = '\0';
1117         loginuid = simple_strtoul(page, &tmp, 10);
1118         if (tmp == page) {
1119                 length = -EINVAL;
1120                 goto out_free_page;
1121
1122         }
1123         kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1124         if (!uid_valid(kloginuid)) {
1125                 length = -EINVAL;
1126                 goto out_free_page;
1127         }
1128
1129         length = audit_set_loginuid(kloginuid);
1130         if (likely(length == 0))
1131                 length = count;
1132
1133 out_free_page:
1134         free_page((unsigned long) page);
1135         return length;
1136 }
1137
1138 static const struct file_operations proc_loginuid_operations = {
1139         .read           = proc_loginuid_read,
1140         .write          = proc_loginuid_write,
1141         .llseek         = generic_file_llseek,
1142 };
1143
1144 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1145                                   size_t count, loff_t *ppos)
1146 {
1147         struct inode * inode = file_inode(file);
1148         struct task_struct *task = get_proc_task(inode);
1149         ssize_t length;
1150         char tmpbuf[TMPBUFLEN];
1151
1152         if (!task)
1153                 return -ESRCH;
1154         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1155                                 audit_get_sessionid(task));
1156         put_task_struct(task);
1157         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1158 }
1159
1160 static const struct file_operations proc_sessionid_operations = {
1161         .read           = proc_sessionid_read,
1162         .llseek         = generic_file_llseek,
1163 };
1164 #endif
1165
1166 #ifdef CONFIG_FAULT_INJECTION
1167 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1168                                       size_t count, loff_t *ppos)
1169 {
1170         struct task_struct *task = get_proc_task(file_inode(file));
1171         char buffer[PROC_NUMBUF];
1172         size_t len;
1173         int make_it_fail;
1174
1175         if (!task)
1176                 return -ESRCH;
1177         make_it_fail = task->make_it_fail;
1178         put_task_struct(task);
1179
1180         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1181
1182         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1183 }
1184
1185 static ssize_t proc_fault_inject_write(struct file * file,
1186                         const char __user * buf, size_t count, loff_t *ppos)
1187 {
1188         struct task_struct *task;
1189         char buffer[PROC_NUMBUF], *end;
1190         int make_it_fail;
1191
1192         if (!capable(CAP_SYS_RESOURCE))
1193                 return -EPERM;
1194         memset(buffer, 0, sizeof(buffer));
1195         if (count > sizeof(buffer) - 1)
1196                 count = sizeof(buffer) - 1;
1197         if (copy_from_user(buffer, buf, count))
1198                 return -EFAULT;
1199         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1200         if (*end)
1201                 return -EINVAL;
1202         task = get_proc_task(file_inode(file));
1203         if (!task)
1204                 return -ESRCH;
1205         task->make_it_fail = make_it_fail;
1206         put_task_struct(task);
1207
1208         return count;
1209 }
1210
1211 static const struct file_operations proc_fault_inject_operations = {
1212         .read           = proc_fault_inject_read,
1213         .write          = proc_fault_inject_write,
1214         .llseek         = generic_file_llseek,
1215 };
1216 #endif
1217
1218
1219 #ifdef CONFIG_SCHED_DEBUG
1220 /*
1221  * Print out various scheduling related per-task fields:
1222  */
1223 static int sched_show(struct seq_file *m, void *v)
1224 {
1225         struct inode *inode = m->private;
1226         struct task_struct *p;
1227
1228         p = get_proc_task(inode);
1229         if (!p)
1230                 return -ESRCH;
1231         proc_sched_show_task(p, m);
1232
1233         put_task_struct(p);
1234
1235         return 0;
1236 }
1237
1238 static ssize_t
1239 sched_write(struct file *file, const char __user *buf,
1240             size_t count, loff_t *offset)
1241 {
1242         struct inode *inode = file_inode(file);
1243         struct task_struct *p;
1244
1245         p = get_proc_task(inode);
1246         if (!p)
1247                 return -ESRCH;
1248         proc_sched_set_task(p);
1249
1250         put_task_struct(p);
1251
1252         return count;
1253 }
1254
1255 static int sched_open(struct inode *inode, struct file *filp)
1256 {
1257         return single_open(filp, sched_show, inode);
1258 }
1259
1260 static const struct file_operations proc_pid_sched_operations = {
1261         .open           = sched_open,
1262         .read           = seq_read,
1263         .write          = sched_write,
1264         .llseek         = seq_lseek,
1265         .release        = single_release,
1266 };
1267
1268 #endif
1269
1270 #ifdef CONFIG_SCHED_AUTOGROUP
1271 /*
1272  * Print out autogroup related information:
1273  */
1274 static int sched_autogroup_show(struct seq_file *m, void *v)
1275 {
1276         struct inode *inode = m->private;
1277         struct task_struct *p;
1278
1279         p = get_proc_task(inode);
1280         if (!p)
1281                 return -ESRCH;
1282         proc_sched_autogroup_show_task(p, m);
1283
1284         put_task_struct(p);
1285
1286         return 0;
1287 }
1288
1289 static ssize_t
1290 sched_autogroup_write(struct file *file, const char __user *buf,
1291             size_t count, loff_t *offset)
1292 {
1293         struct inode *inode = file_inode(file);
1294         struct task_struct *p;
1295         char buffer[PROC_NUMBUF];
1296         int nice;
1297         int err;
1298
1299         memset(buffer, 0, sizeof(buffer));
1300         if (count > sizeof(buffer) - 1)
1301                 count = sizeof(buffer) - 1;
1302         if (copy_from_user(buffer, buf, count))
1303                 return -EFAULT;
1304
1305         err = kstrtoint(strstrip(buffer), 0, &nice);
1306         if (err < 0)
1307                 return err;
1308
1309         p = get_proc_task(inode);
1310         if (!p)
1311                 return -ESRCH;
1312
1313         err = proc_sched_autogroup_set_nice(p, nice);
1314         if (err)
1315                 count = err;
1316
1317         put_task_struct(p);
1318
1319         return count;
1320 }
1321
1322 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1323 {
1324         int ret;
1325
1326         ret = single_open(filp, sched_autogroup_show, NULL);
1327         if (!ret) {
1328                 struct seq_file *m = filp->private_data;
1329
1330                 m->private = inode;
1331         }
1332         return ret;
1333 }
1334
1335 static const struct file_operations proc_pid_sched_autogroup_operations = {
1336         .open           = sched_autogroup_open,
1337         .read           = seq_read,
1338         .write          = sched_autogroup_write,
1339         .llseek         = seq_lseek,
1340         .release        = single_release,
1341 };
1342
1343 #endif /* CONFIG_SCHED_AUTOGROUP */
1344
1345 static ssize_t comm_write(struct file *file, const char __user *buf,
1346                                 size_t count, loff_t *offset)
1347 {
1348         struct inode *inode = file_inode(file);
1349         struct task_struct *p;
1350         char buffer[TASK_COMM_LEN];
1351
1352         memset(buffer, 0, sizeof(buffer));
1353         if (count > sizeof(buffer) - 1)
1354                 count = sizeof(buffer) - 1;
1355         if (copy_from_user(buffer, buf, count))
1356                 return -EFAULT;
1357
1358         p = get_proc_task(inode);
1359         if (!p)
1360                 return -ESRCH;
1361
1362         if (same_thread_group(current, p))
1363                 set_task_comm(p, buffer);
1364         else
1365                 count = -EINVAL;
1366
1367         put_task_struct(p);
1368
1369         return count;
1370 }
1371
1372 static int comm_show(struct seq_file *m, void *v)
1373 {
1374         struct inode *inode = m->private;
1375         struct task_struct *p;
1376
1377         p = get_proc_task(inode);
1378         if (!p)
1379                 return -ESRCH;
1380
1381         task_lock(p);
1382         seq_printf(m, "%s\n", p->comm);
1383         task_unlock(p);
1384
1385         put_task_struct(p);
1386
1387         return 0;
1388 }
1389
1390 static int comm_open(struct inode *inode, struct file *filp)
1391 {
1392         return single_open(filp, comm_show, inode);
1393 }
1394
1395 static const struct file_operations proc_pid_set_comm_operations = {
1396         .open           = comm_open,
1397         .read           = seq_read,
1398         .write          = comm_write,
1399         .llseek         = seq_lseek,
1400         .release        = single_release,
1401 };
1402
1403 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1404 {
1405         struct task_struct *task;
1406         struct mm_struct *mm;
1407         struct file *exe_file;
1408
1409         task = get_proc_task(dentry->d_inode);
1410         if (!task)
1411                 return -ENOENT;
1412         mm = get_task_mm(task);
1413         put_task_struct(task);
1414         if (!mm)
1415                 return -ENOENT;
1416         exe_file = get_mm_exe_file(mm);
1417         mmput(mm);
1418         if (exe_file) {
1419                 *exe_path = exe_file->f_path;
1420                 path_get(&exe_file->f_path);
1421                 fput(exe_file);
1422                 return 0;
1423         } else
1424                 return -ENOENT;
1425 }
1426
1427 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1428 {
1429         struct inode *inode = dentry->d_inode;
1430         struct path path;
1431         int error = -EACCES;
1432
1433         /* Are we allowed to snoop on the tasks file descriptors? */
1434         if (!proc_fd_access_allowed(inode))
1435                 goto out;
1436
1437         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1438         if (error)
1439                 goto out;
1440
1441         nd_jump_link(nd, &path);
1442         return NULL;
1443 out:
1444         return ERR_PTR(error);
1445 }
1446
1447 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1448 {
1449         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1450         char *pathname;
1451         int len;
1452
1453         if (!tmp)
1454                 return -ENOMEM;
1455
1456         pathname = d_path(path, tmp, PAGE_SIZE);
1457         len = PTR_ERR(pathname);
1458         if (IS_ERR(pathname))
1459                 goto out;
1460         len = tmp + PAGE_SIZE - 1 - pathname;
1461
1462         if (len > buflen)
1463                 len = buflen;
1464         if (copy_to_user(buffer, pathname, len))
1465                 len = -EFAULT;
1466  out:
1467         free_page((unsigned long)tmp);
1468         return len;
1469 }
1470
1471 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1472 {
1473         int error = -EACCES;
1474         struct inode *inode = dentry->d_inode;
1475         struct path path;
1476
1477         /* Are we allowed to snoop on the tasks file descriptors? */
1478         if (!proc_fd_access_allowed(inode))
1479                 goto out;
1480
1481         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1482         if (error)
1483                 goto out;
1484
1485         error = do_proc_readlink(&path, buffer, buflen);
1486         path_put(&path);
1487 out:
1488         return error;
1489 }
1490
1491 const struct inode_operations proc_pid_link_inode_operations = {
1492         .readlink       = proc_pid_readlink,
1493         .follow_link    = proc_pid_follow_link,
1494         .setattr        = proc_setattr,
1495 };
1496
1497
1498 /* building an inode */
1499
1500 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1501 {
1502         struct inode * inode;
1503         struct proc_inode *ei;
1504         const struct cred *cred;
1505
1506         /* We need a new inode */
1507
1508         inode = new_inode(sb);
1509         if (!inode)
1510                 goto out;
1511
1512         /* Common stuff */
1513         ei = PROC_I(inode);
1514         inode->i_ino = get_next_ino();
1515         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1516         inode->i_op = &proc_def_inode_operations;
1517
1518         /*
1519          * grab the reference to task.
1520          */
1521         ei->pid = get_task_pid(task, PIDTYPE_PID);
1522         if (!ei->pid)
1523                 goto out_unlock;
1524
1525         if (task_dumpable(task)) {
1526                 rcu_read_lock();
1527                 cred = __task_cred(task);
1528                 inode->i_uid = cred->euid;
1529                 inode->i_gid = cred->egid;
1530                 rcu_read_unlock();
1531         }
1532         security_task_to_inode(task, inode);
1533
1534 out:
1535         return inode;
1536
1537 out_unlock:
1538         iput(inode);
1539         return NULL;
1540 }
1541
1542 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1543 {
1544         struct inode *inode = dentry->d_inode;
1545         struct task_struct *task;
1546         const struct cred *cred;
1547         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1548
1549         generic_fillattr(inode, stat);
1550
1551         rcu_read_lock();
1552         stat->uid = GLOBAL_ROOT_UID;
1553         stat->gid = GLOBAL_ROOT_GID;
1554         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1555         if (task) {
1556                 if (!has_pid_permissions(pid, task, 2)) {
1557                         rcu_read_unlock();
1558                         /*
1559                          * This doesn't prevent learning whether PID exists,
1560                          * it only makes getattr() consistent with readdir().
1561                          */
1562                         return -ENOENT;
1563                 }
1564                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1565                     task_dumpable(task)) {
1566                         cred = __task_cred(task);
1567                         stat->uid = cred->euid;
1568                         stat->gid = cred->egid;
1569                 }
1570         }
1571         rcu_read_unlock();
1572         return 0;
1573 }
1574
1575 /* dentry stuff */
1576
1577 /*
1578  *      Exceptional case: normally we are not allowed to unhash a busy
1579  * directory. In this case, however, we can do it - no aliasing problems
1580  * due to the way we treat inodes.
1581  *
1582  * Rewrite the inode's ownerships here because the owning task may have
1583  * performed a setuid(), etc.
1584  *
1585  * Before the /proc/pid/status file was created the only way to read
1586  * the effective uid of a /process was to stat /proc/pid.  Reading
1587  * /proc/pid/status is slow enough that procps and other packages
1588  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1589  * made this apply to all per process world readable and executable
1590  * directories.
1591  */
1592 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1593 {
1594         struct inode *inode;
1595         struct task_struct *task;
1596         const struct cred *cred;
1597
1598         if (flags & LOOKUP_RCU)
1599                 return -ECHILD;
1600
1601         inode = dentry->d_inode;
1602         task = get_proc_task(inode);
1603
1604         if (task) {
1605                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1606                     task_dumpable(task)) {
1607                         rcu_read_lock();
1608                         cred = __task_cred(task);
1609                         inode->i_uid = cred->euid;
1610                         inode->i_gid = cred->egid;
1611                         rcu_read_unlock();
1612                 } else {
1613                         inode->i_uid = GLOBAL_ROOT_UID;
1614                         inode->i_gid = GLOBAL_ROOT_GID;
1615                 }
1616                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1617                 security_task_to_inode(task, inode);
1618                 put_task_struct(task);
1619                 return 1;
1620         }
1621         d_drop(dentry);
1622         return 0;
1623 }
1624
1625 const struct dentry_operations pid_dentry_operations =
1626 {
1627         .d_revalidate   = pid_revalidate,
1628         .d_delete       = pid_delete_dentry,
1629 };
1630
1631 /* Lookups */
1632
1633 /*
1634  * Fill a directory entry.
1635  *
1636  * If possible create the dcache entry and derive our inode number and
1637  * file type from dcache entry.
1638  *
1639  * Since all of the proc inode numbers are dynamically generated, the inode
1640  * numbers do not exist until the inode is cache.  This means creating the
1641  * the dcache entry in readdir is necessary to keep the inode numbers
1642  * reported by readdir in sync with the inode numbers reported
1643  * by stat.
1644  */
1645 int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1646         const char *name, int len,
1647         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1648 {
1649         struct dentry *child, *dir = filp->f_path.dentry;
1650         struct inode *inode;
1651         struct qstr qname;
1652         ino_t ino = 0;
1653         unsigned type = DT_UNKNOWN;
1654
1655         qname.name = name;
1656         qname.len  = len;
1657         qname.hash = full_name_hash(name, len);
1658
1659         child = d_lookup(dir, &qname);
1660         if (!child) {
1661                 struct dentry *new;
1662                 new = d_alloc(dir, &qname);
1663                 if (new) {
1664                         child = instantiate(dir->d_inode, new, task, ptr);
1665                         if (child)
1666                                 dput(new);
1667                         else
1668                                 child = new;
1669                 }
1670         }
1671         if (!child || IS_ERR(child) || !child->d_inode)
1672                 goto end_instantiate;
1673         inode = child->d_inode;
1674         if (inode) {
1675                 ino = inode->i_ino;
1676                 type = inode->i_mode >> 12;
1677         }
1678         dput(child);
1679 end_instantiate:
1680         if (!ino)
1681                 ino = find_inode_number(dir, &qname);
1682         if (!ino)
1683                 ino = 1;
1684         return filldir(dirent, name, len, filp->f_pos, ino, type);
1685 }
1686
1687 #ifdef CONFIG_CHECKPOINT_RESTORE
1688
1689 /*
1690  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1691  * which represent vma start and end addresses.
1692  */
1693 static int dname_to_vma_addr(struct dentry *dentry,
1694                              unsigned long *start, unsigned long *end)
1695 {
1696         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1697                 return -EINVAL;
1698
1699         return 0;
1700 }
1701
1702 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1703 {
1704         unsigned long vm_start, vm_end;
1705         bool exact_vma_exists = false;
1706         struct mm_struct *mm = NULL;
1707         struct task_struct *task;
1708         const struct cred *cred;
1709         struct inode *inode;
1710         int status = 0;
1711
1712         if (flags & LOOKUP_RCU)
1713                 return -ECHILD;
1714
1715         if (!capable(CAP_SYS_ADMIN)) {
1716                 status = -EPERM;
1717                 goto out_notask;
1718         }
1719
1720         inode = dentry->d_inode;
1721         task = get_proc_task(inode);
1722         if (!task)
1723                 goto out_notask;
1724
1725         mm = mm_access(task, PTRACE_MODE_READ);
1726         if (IS_ERR_OR_NULL(mm))
1727                 goto out;
1728
1729         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1730                 down_read(&mm->mmap_sem);
1731                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1732                 up_read(&mm->mmap_sem);
1733         }
1734
1735         mmput(mm);
1736
1737         if (exact_vma_exists) {
1738                 if (task_dumpable(task)) {
1739                         rcu_read_lock();
1740                         cred = __task_cred(task);
1741                         inode->i_uid = cred->euid;
1742                         inode->i_gid = cred->egid;
1743                         rcu_read_unlock();
1744                 } else {
1745                         inode->i_uid = GLOBAL_ROOT_UID;
1746                         inode->i_gid = GLOBAL_ROOT_GID;
1747                 }
1748                 security_task_to_inode(task, inode);
1749                 status = 1;
1750         }
1751
1752 out:
1753         put_task_struct(task);
1754
1755 out_notask:
1756         if (status <= 0)
1757                 d_drop(dentry);
1758
1759         return status;
1760 }
1761
1762 static const struct dentry_operations tid_map_files_dentry_operations = {
1763         .d_revalidate   = map_files_d_revalidate,
1764         .d_delete       = pid_delete_dentry,
1765 };
1766
1767 static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
1768 {
1769         unsigned long vm_start, vm_end;
1770         struct vm_area_struct *vma;
1771         struct task_struct *task;
1772         struct mm_struct *mm;
1773         int rc;
1774
1775         rc = -ENOENT;
1776         task = get_proc_task(dentry->d_inode);
1777         if (!task)
1778                 goto out;
1779
1780         mm = get_task_mm(task);
1781         put_task_struct(task);
1782         if (!mm)
1783                 goto out;
1784
1785         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1786         if (rc)
1787                 goto out_mmput;
1788
1789         down_read(&mm->mmap_sem);
1790         vma = find_exact_vma(mm, vm_start, vm_end);
1791         if (vma && vma->vm_file) {
1792                 *path = vma->vm_file->f_path;
1793                 path_get(path);
1794                 rc = 0;
1795         }
1796         up_read(&mm->mmap_sem);
1797
1798 out_mmput:
1799         mmput(mm);
1800 out:
1801         return rc;
1802 }
1803
1804 struct map_files_info {
1805         fmode_t         mode;
1806         unsigned long   len;
1807         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1808 };
1809
1810 static struct dentry *
1811 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
1812                            struct task_struct *task, const void *ptr)
1813 {
1814         fmode_t mode = (fmode_t)(unsigned long)ptr;
1815         struct proc_inode *ei;
1816         struct inode *inode;
1817
1818         inode = proc_pid_make_inode(dir->i_sb, task);
1819         if (!inode)
1820                 return ERR_PTR(-ENOENT);
1821
1822         ei = PROC_I(inode);
1823         ei->op.proc_get_link = proc_map_files_get_link;
1824
1825         inode->i_op = &proc_pid_link_inode_operations;
1826         inode->i_size = 64;
1827         inode->i_mode = S_IFLNK;
1828
1829         if (mode & FMODE_READ)
1830                 inode->i_mode |= S_IRUSR;
1831         if (mode & FMODE_WRITE)
1832                 inode->i_mode |= S_IWUSR;
1833
1834         d_set_d_op(dentry, &tid_map_files_dentry_operations);
1835         d_add(dentry, inode);
1836
1837         return NULL;
1838 }
1839
1840 static struct dentry *proc_map_files_lookup(struct inode *dir,
1841                 struct dentry *dentry, unsigned int flags)
1842 {
1843         unsigned long vm_start, vm_end;
1844         struct vm_area_struct *vma;
1845         struct task_struct *task;
1846         struct dentry *result;
1847         struct mm_struct *mm;
1848
1849         result = ERR_PTR(-EPERM);
1850         if (!capable(CAP_SYS_ADMIN))
1851                 goto out;
1852
1853         result = ERR_PTR(-ENOENT);
1854         task = get_proc_task(dir);
1855         if (!task)
1856                 goto out;
1857
1858         result = ERR_PTR(-EACCES);
1859         if (!ptrace_may_access(task, PTRACE_MODE_READ))
1860                 goto out_put_task;
1861
1862         result = ERR_PTR(-ENOENT);
1863         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
1864                 goto out_put_task;
1865
1866         mm = get_task_mm(task);
1867         if (!mm)
1868                 goto out_put_task;
1869
1870         down_read(&mm->mmap_sem);
1871         vma = find_exact_vma(mm, vm_start, vm_end);
1872         if (!vma)
1873                 goto out_no_vma;
1874
1875         if (vma->vm_file)
1876                 result = proc_map_files_instantiate(dir, dentry, task,
1877                                 (void *)(unsigned long)vma->vm_file->f_mode);
1878
1879 out_no_vma:
1880         up_read(&mm->mmap_sem);
1881         mmput(mm);
1882 out_put_task:
1883         put_task_struct(task);
1884 out:
1885         return result;
1886 }
1887
1888 static const struct inode_operations proc_map_files_inode_operations = {
1889         .lookup         = proc_map_files_lookup,
1890         .permission     = proc_fd_permission,
1891         .setattr        = proc_setattr,
1892 };
1893
1894 static int
1895 proc_map_files_readdir(struct file *filp, void *dirent, filldir_t filldir)
1896 {
1897         struct dentry *dentry = filp->f_path.dentry;
1898         struct inode *inode = dentry->d_inode;
1899         struct vm_area_struct *vma;
1900         struct task_struct *task;
1901         struct mm_struct *mm;
1902         ino_t ino;
1903         int ret;
1904
1905         ret = -EPERM;
1906         if (!capable(CAP_SYS_ADMIN))
1907                 goto out;
1908
1909         ret = -ENOENT;
1910         task = get_proc_task(inode);
1911         if (!task)
1912                 goto out;
1913
1914         ret = -EACCES;
1915         if (!ptrace_may_access(task, PTRACE_MODE_READ))
1916                 goto out_put_task;
1917
1918         ret = 0;
1919         switch (filp->f_pos) {
1920         case 0:
1921                 ino = inode->i_ino;
1922                 if (filldir(dirent, ".", 1, 0, ino, DT_DIR) < 0)
1923                         goto out_put_task;
1924                 filp->f_pos++;
1925         case 1:
1926                 ino = parent_ino(dentry);
1927                 if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1928                         goto out_put_task;
1929                 filp->f_pos++;
1930         default:
1931         {
1932                 unsigned long nr_files, pos, i;
1933                 struct flex_array *fa = NULL;
1934                 struct map_files_info info;
1935                 struct map_files_info *p;
1936
1937                 mm = get_task_mm(task);
1938                 if (!mm)
1939                         goto out_put_task;
1940                 down_read(&mm->mmap_sem);
1941
1942                 nr_files = 0;
1943
1944                 /*
1945                  * We need two passes here:
1946                  *
1947                  *  1) Collect vmas of mapped files with mmap_sem taken
1948                  *  2) Release mmap_sem and instantiate entries
1949                  *
1950                  * otherwise we get lockdep complained, since filldir()
1951                  * routine might require mmap_sem taken in might_fault().
1952                  */
1953
1954                 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
1955                         if (vma->vm_file && ++pos > filp->f_pos)
1956                                 nr_files++;
1957                 }
1958
1959                 if (nr_files) {
1960                         fa = flex_array_alloc(sizeof(info), nr_files,
1961                                                 GFP_KERNEL);
1962                         if (!fa || flex_array_prealloc(fa, 0, nr_files,
1963                                                         GFP_KERNEL)) {
1964                                 ret = -ENOMEM;
1965                                 if (fa)
1966                                         flex_array_free(fa);
1967                                 up_read(&mm->mmap_sem);
1968                                 mmput(mm);
1969                                 goto out_put_task;
1970                         }
1971                         for (i = 0, vma = mm->mmap, pos = 2; vma;
1972                                         vma = vma->vm_next) {
1973                                 if (!vma->vm_file)
1974                                         continue;
1975                                 if (++pos <= filp->f_pos)
1976                                         continue;
1977
1978                                 info.mode = vma->vm_file->f_mode;
1979                                 info.len = snprintf(info.name,
1980                                                 sizeof(info.name), "%lx-%lx",
1981                                                 vma->vm_start, vma->vm_end);
1982                                 if (flex_array_put(fa, i++, &info, GFP_KERNEL))
1983                                         BUG();
1984                         }
1985                 }
1986                 up_read(&mm->mmap_sem);
1987
1988                 for (i = 0; i < nr_files; i++) {
1989                         p = flex_array_get(fa, i);
1990                         ret = proc_fill_cache(filp, dirent, filldir,
1991                                               p->name, p->len,
1992                                               proc_map_files_instantiate,
1993                                               task,
1994                                               (void *)(unsigned long)p->mode);
1995                         if (ret)
1996                                 break;
1997                         filp->f_pos++;
1998                 }
1999                 if (fa)
2000                         flex_array_free(fa);
2001                 mmput(mm);
2002         }
2003         }
2004
2005 out_put_task:
2006         put_task_struct(task);
2007 out:
2008         return ret;
2009 }
2010
2011 static const struct file_operations proc_map_files_operations = {
2012         .read           = generic_read_dir,
2013         .readdir        = proc_map_files_readdir,
2014         .llseek         = default_llseek,
2015 };
2016
2017 struct timers_private {
2018         struct pid *pid;
2019         struct task_struct *task;
2020         struct sighand_struct *sighand;
2021         unsigned long flags;
2022 };
2023
2024 static void *timers_start(struct seq_file *m, loff_t *pos)
2025 {
2026         struct timers_private *tp = m->private;
2027
2028         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2029         if (!tp->task)
2030                 return ERR_PTR(-ESRCH);
2031
2032         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2033         if (!tp->sighand)
2034                 return ERR_PTR(-ESRCH);
2035
2036         return seq_list_start(&tp->task->signal->posix_timers, *pos);
2037 }
2038
2039 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2040 {
2041         struct timers_private *tp = m->private;
2042         return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2043 }
2044
2045 static void timers_stop(struct seq_file *m, void *v)
2046 {
2047         struct timers_private *tp = m->private;
2048
2049         if (tp->sighand) {
2050                 unlock_task_sighand(tp->task, &tp->flags);
2051                 tp->sighand = NULL;
2052         }
2053
2054         if (tp->task) {
2055                 put_task_struct(tp->task);
2056                 tp->task = NULL;
2057         }
2058 }
2059
2060 static int show_timer(struct seq_file *m, void *v)
2061 {
2062         struct k_itimer *timer;
2063
2064         timer = list_entry((struct list_head *)v, struct k_itimer, list);
2065         seq_printf(m, "ID: %d\n", timer->it_id);
2066
2067         return 0;
2068 }
2069
2070 static const struct seq_operations proc_timers_seq_ops = {
2071         .start  = timers_start,
2072         .next   = timers_next,
2073         .stop   = timers_stop,
2074         .show   = show_timer,
2075 };
2076
2077 static int proc_timers_open(struct inode *inode, struct file *file)
2078 {
2079         struct timers_private *tp;
2080
2081         tp = __seq_open_private(file, &proc_timers_seq_ops,
2082                         sizeof(struct timers_private));
2083         if (!tp)
2084                 return -ENOMEM;
2085
2086         tp->pid = proc_pid(inode);
2087         return 0;
2088 }
2089
2090 static const struct file_operations proc_timers_operations = {
2091         .open           = proc_timers_open,
2092         .read           = seq_read,
2093         .llseek         = seq_lseek,
2094         .release        = seq_release_private,
2095 };
2096 #endif /* CONFIG_CHECKPOINT_RESTORE */
2097
2098 static struct dentry *proc_pident_instantiate(struct inode *dir,
2099         struct dentry *dentry, struct task_struct *task, const void *ptr)
2100 {
2101         const struct pid_entry *p = ptr;
2102         struct inode *inode;
2103         struct proc_inode *ei;
2104         struct dentry *error = ERR_PTR(-ENOENT);
2105
2106         inode = proc_pid_make_inode(dir->i_sb, task);
2107         if (!inode)
2108                 goto out;
2109
2110         ei = PROC_I(inode);
2111         inode->i_mode = p->mode;
2112         if (S_ISDIR(inode->i_mode))
2113                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2114         if (p->iop)
2115                 inode->i_op = p->iop;
2116         if (p->fop)
2117                 inode->i_fop = p->fop;
2118         ei->op = p->op;
2119         d_set_d_op(dentry, &pid_dentry_operations);
2120         d_add(dentry, inode);
2121         /* Close the race of the process dying before we return the dentry */
2122         if (pid_revalidate(dentry, 0))
2123                 error = NULL;
2124 out:
2125         return error;
2126 }
2127
2128 static struct dentry *proc_pident_lookup(struct inode *dir,
2129                                          struct dentry *dentry,
2130                                          const struct pid_entry *ents,
2131                                          unsigned int nents)
2132 {
2133         struct dentry *error;
2134         struct task_struct *task = get_proc_task(dir);
2135         const struct pid_entry *p, *last;
2136
2137         error = ERR_PTR(-ENOENT);
2138
2139         if (!task)
2140                 goto out_no_task;
2141
2142         /*
2143          * Yes, it does not scale. And it should not. Don't add
2144          * new entries into /proc/<tgid>/ without very good reasons.
2145          */
2146         last = &ents[nents - 1];
2147         for (p = ents; p <= last; p++) {
2148                 if (p->len != dentry->d_name.len)
2149                         continue;
2150                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2151                         break;
2152         }
2153         if (p > last)
2154                 goto out;
2155
2156         error = proc_pident_instantiate(dir, dentry, task, p);
2157 out:
2158         put_task_struct(task);
2159 out_no_task:
2160         return error;
2161 }
2162
2163 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2164         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2165 {
2166         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2167                                 proc_pident_instantiate, task, p);
2168 }
2169
2170 static int proc_pident_readdir(struct file *filp,
2171                 void *dirent, filldir_t filldir,
2172                 const struct pid_entry *ents, unsigned int nents)
2173 {
2174         int i;
2175         struct dentry *dentry = filp->f_path.dentry;
2176         struct inode *inode = dentry->d_inode;
2177         struct task_struct *task = get_proc_task(inode);
2178         const struct pid_entry *p, *last;
2179         ino_t ino;
2180         int ret;
2181
2182         ret = -ENOENT;
2183         if (!task)
2184                 goto out_no_task;
2185
2186         ret = 0;
2187         i = filp->f_pos;
2188         switch (i) {
2189         case 0:
2190                 ino = inode->i_ino;
2191                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2192                         goto out;
2193                 i++;
2194                 filp->f_pos++;
2195                 /* fall through */
2196         case 1:
2197                 ino = parent_ino(dentry);
2198                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2199                         goto out;
2200                 i++;
2201                 filp->f_pos++;
2202                 /* fall through */
2203         default:
2204                 i -= 2;
2205                 if (i >= nents) {
2206                         ret = 1;
2207                         goto out;
2208                 }
2209                 p = ents + i;
2210                 last = &ents[nents - 1];
2211                 while (p <= last) {
2212                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2213                                 goto out;
2214                         filp->f_pos++;
2215                         p++;
2216                 }
2217         }
2218
2219         ret = 1;
2220 out:
2221         put_task_struct(task);
2222 out_no_task:
2223         return ret;
2224 }
2225
2226 #ifdef CONFIG_SECURITY
2227 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2228                                   size_t count, loff_t *ppos)
2229 {
2230         struct inode * inode = file_inode(file);
2231         char *p = NULL;
2232         ssize_t length;
2233         struct task_struct *task = get_proc_task(inode);
2234
2235         if (!task)
2236                 return -ESRCH;
2237
2238         length = security_getprocattr(task,
2239                                       (char*)file->f_path.dentry->d_name.name,
2240                                       &p);
2241         put_task_struct(task);
2242         if (length > 0)
2243                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2244         kfree(p);
2245         return length;
2246 }
2247
2248 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2249                                    size_t count, loff_t *ppos)
2250 {
2251         struct inode * inode = file_inode(file);
2252         char *page;
2253         ssize_t length;
2254         struct task_struct *task = get_proc_task(inode);
2255
2256         length = -ESRCH;
2257         if (!task)
2258                 goto out_no_task;
2259         if (count > PAGE_SIZE)
2260                 count = PAGE_SIZE;
2261
2262         /* No partial writes. */
2263         length = -EINVAL;
2264         if (*ppos != 0)
2265                 goto out;
2266
2267         length = -ENOMEM;
2268         page = (char*)__get_free_page(GFP_TEMPORARY);
2269         if (!page)
2270                 goto out;
2271
2272         length = -EFAULT;
2273         if (copy_from_user(page, buf, count))
2274                 goto out_free;
2275
2276         /* Guard against adverse ptrace interaction */
2277         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2278         if (length < 0)
2279                 goto out_free;
2280
2281         length = security_setprocattr(task,
2282                                       (char*)file->f_path.dentry->d_name.name,
2283                                       (void*)page, count);
2284         mutex_unlock(&task->signal->cred_guard_mutex);
2285 out_free:
2286         free_page((unsigned long) page);
2287 out:
2288         put_task_struct(task);
2289 out_no_task:
2290         return length;
2291 }
2292
2293 static const struct file_operations proc_pid_attr_operations = {
2294         .read           = proc_pid_attr_read,
2295         .write          = proc_pid_attr_write,
2296         .llseek         = generic_file_llseek,
2297 };
2298
2299 static const struct pid_entry attr_dir_stuff[] = {
2300         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2301         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2302         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2303         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2304         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2305         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2306 };
2307
2308 static int proc_attr_dir_readdir(struct file * filp,
2309                              void * dirent, filldir_t filldir)
2310 {
2311         return proc_pident_readdir(filp,dirent,filldir,
2312                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2313 }
2314
2315 static const struct file_operations proc_attr_dir_operations = {
2316         .read           = generic_read_dir,
2317         .readdir        = proc_attr_dir_readdir,
2318         .llseek         = default_llseek,
2319 };
2320
2321 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2322                                 struct dentry *dentry, unsigned int flags)
2323 {
2324         return proc_pident_lookup(dir, dentry,
2325                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2326 }
2327
2328 static const struct inode_operations proc_attr_dir_inode_operations = {
2329         .lookup         = proc_attr_dir_lookup,
2330         .getattr        = pid_getattr,
2331         .setattr        = proc_setattr,
2332 };
2333
2334 #endif
2335
2336 #ifdef CONFIG_ELF_CORE
2337 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2338                                          size_t count, loff_t *ppos)
2339 {
2340         struct task_struct *task = get_proc_task(file_inode(file));
2341         struct mm_struct *mm;
2342         char buffer[PROC_NUMBUF];
2343         size_t len;
2344         int ret;
2345
2346         if (!task)
2347                 return -ESRCH;
2348
2349         ret = 0;
2350         mm = get_task_mm(task);
2351         if (mm) {
2352                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2353                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2354                                 MMF_DUMP_FILTER_SHIFT));
2355                 mmput(mm);
2356                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2357         }
2358
2359         put_task_struct(task);
2360
2361         return ret;
2362 }
2363
2364 static ssize_t proc_coredump_filter_write(struct file *file,
2365                                           const char __user *buf,
2366                                           size_t count,
2367                                           loff_t *ppos)
2368 {
2369         struct task_struct *task;
2370         struct mm_struct *mm;
2371         char buffer[PROC_NUMBUF], *end;
2372         unsigned int val;
2373         int ret;
2374         int i;
2375         unsigned long mask;
2376
2377         ret = -EFAULT;
2378         memset(buffer, 0, sizeof(buffer));
2379         if (count > sizeof(buffer) - 1)
2380                 count = sizeof(buffer) - 1;
2381         if (copy_from_user(buffer, buf, count))
2382                 goto out_no_task;
2383
2384         ret = -EINVAL;
2385         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2386         if (*end == '\n')
2387                 end++;
2388         if (end - buffer == 0)
2389                 goto out_no_task;
2390
2391         ret = -ESRCH;
2392         task = get_proc_task(file_inode(file));
2393         if (!task)
2394                 goto out_no_task;
2395
2396         ret = end - buffer;
2397         mm = get_task_mm(task);
2398         if (!mm)
2399                 goto out_no_mm;
2400
2401         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2402                 if (val & mask)
2403                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2404                 else
2405                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2406         }
2407
2408         mmput(mm);
2409  out_no_mm:
2410         put_task_struct(task);
2411  out_no_task:
2412         return ret;
2413 }
2414
2415 static const struct file_operations proc_coredump_filter_operations = {
2416         .read           = proc_coredump_filter_read,
2417         .write          = proc_coredump_filter_write,
2418         .llseek         = generic_file_llseek,
2419 };
2420 #endif
2421
2422 #ifdef CONFIG_TASK_IO_ACCOUNTING
2423 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2424 {
2425         struct task_io_accounting acct = task->ioac;
2426         unsigned long flags;
2427         int result;
2428
2429         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2430         if (result)
2431                 return result;
2432
2433         if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2434                 result = -EACCES;
2435                 goto out_unlock;
2436         }
2437
2438         if (whole && lock_task_sighand(task, &flags)) {
2439                 struct task_struct *t = task;
2440
2441                 task_io_accounting_add(&acct, &task->signal->ioac);
2442                 while_each_thread(task, t)
2443                         task_io_accounting_add(&acct, &t->ioac);
2444
2445                 unlock_task_sighand(task, &flags);
2446         }
2447         result = sprintf(buffer,
2448                         "rchar: %llu\n"
2449                         "wchar: %llu\n"
2450                         "syscr: %llu\n"
2451                         "syscw: %llu\n"
2452                         "read_bytes: %llu\n"
2453                         "write_bytes: %llu\n"
2454                         "cancelled_write_bytes: %llu\n",
2455                         (unsigned long long)acct.rchar,
2456                         (unsigned long long)acct.wchar,
2457                         (unsigned long long)acct.syscr,
2458                         (unsigned long long)acct.syscw,
2459                         (unsigned long long)acct.read_bytes,
2460                         (unsigned long long)acct.write_bytes,
2461                         (unsigned long long)acct.cancelled_write_bytes);
2462 out_unlock:
2463         mutex_unlock(&task->signal->cred_guard_mutex);
2464         return result;
2465 }
2466
2467 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2468 {
2469         return do_io_accounting(task, buffer, 0);
2470 }
2471
2472 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2473 {
2474         return do_io_accounting(task, buffer, 1);
2475 }
2476 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2477
2478 #ifdef CONFIG_USER_NS
2479 static int proc_id_map_open(struct inode *inode, struct file *file,
2480         struct seq_operations *seq_ops)
2481 {
2482         struct user_namespace *ns = NULL;
2483         struct task_struct *task;
2484         struct seq_file *seq;
2485         int ret = -EINVAL;
2486
2487         task = get_proc_task(inode);
2488         if (task) {
2489                 rcu_read_lock();
2490                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2491                 rcu_read_unlock();
2492                 put_task_struct(task);
2493         }
2494         if (!ns)
2495                 goto err;
2496
2497         ret = seq_open(file, seq_ops);
2498         if (ret)
2499                 goto err_put_ns;
2500
2501         seq = file->private_data;
2502         seq->private = ns;
2503
2504         return 0;
2505 err_put_ns:
2506         put_user_ns(ns);
2507 err:
2508         return ret;
2509 }
2510
2511 static int proc_id_map_release(struct inode *inode, struct file *file)
2512 {
2513         struct seq_file *seq = file->private_data;
2514         struct user_namespace *ns = seq->private;
2515         put_user_ns(ns);
2516         return seq_release(inode, file);
2517 }
2518
2519 static int proc_uid_map_open(struct inode *inode, struct file *file)
2520 {
2521         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2522 }
2523
2524 static int proc_gid_map_open(struct inode *inode, struct file *file)
2525 {
2526         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2527 }
2528
2529 static int proc_projid_map_open(struct inode *inode, struct file *file)
2530 {
2531         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2532 }
2533
2534 static const struct file_operations proc_uid_map_operations = {
2535         .open           = proc_uid_map_open,
2536         .write          = proc_uid_map_write,
2537         .read           = seq_read,
2538         .llseek         = seq_lseek,
2539         .release        = proc_id_map_release,
2540 };
2541
2542 static const struct file_operations proc_gid_map_operations = {
2543         .open           = proc_gid_map_open,
2544         .write          = proc_gid_map_write,
2545         .read           = seq_read,
2546         .llseek         = seq_lseek,
2547         .release        = proc_id_map_release,
2548 };
2549
2550 static const struct file_operations proc_projid_map_operations = {
2551         .open           = proc_projid_map_open,
2552         .write          = proc_projid_map_write,
2553         .read           = seq_read,
2554         .llseek         = seq_lseek,
2555         .release        = proc_id_map_release,
2556 };
2557 #endif /* CONFIG_USER_NS */
2558
2559 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2560                                 struct pid *pid, struct task_struct *task)
2561 {
2562         int err = lock_trace(task);
2563         if (!err) {
2564                 seq_printf(m, "%08x\n", task->personality);
2565                 unlock_trace(task);
2566         }
2567         return err;
2568 }
2569
2570 /*
2571  * Thread groups
2572  */
2573 static const struct file_operations proc_task_operations;
2574 static const struct inode_operations proc_task_inode_operations;
2575
2576 static const struct pid_entry tgid_base_stuff[] = {
2577         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2578         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2579 #ifdef CONFIG_CHECKPOINT_RESTORE
2580         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2581 #endif
2582         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2583         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2584 #ifdef CONFIG_NET
2585         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2586 #endif
2587         REG("environ",    S_IRUSR, proc_environ_operations),
2588         INF("auxv",       S_IRUSR, proc_pid_auxv),
2589         ONE("status",     S_IRUGO, proc_pid_status),
2590         ONE("personality", S_IRUGO, proc_pid_personality),
2591         INF("limits",     S_IRUGO, proc_pid_limits),
2592 #ifdef CONFIG_SCHED_DEBUG
2593         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2594 #endif
2595 #ifdef CONFIG_SCHED_AUTOGROUP
2596         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2597 #endif
2598         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2599 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2600         INF("syscall",    S_IRUGO, proc_pid_syscall),
2601 #endif
2602         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2603         ONE("stat",       S_IRUGO, proc_tgid_stat),
2604         ONE("statm",      S_IRUGO, proc_pid_statm),
2605         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2606 #ifdef CONFIG_NUMA
2607         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2608 #endif
2609         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2610         LNK("cwd",        proc_cwd_link),
2611         LNK("root",       proc_root_link),
2612         LNK("exe",        proc_exe_link),
2613         REG("mounts",     S_IRUGO, proc_mounts_operations),
2614         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2615         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2616 #ifdef CONFIG_PROC_PAGE_MONITOR
2617         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2618         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2619         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
2620 #endif
2621 #ifdef CONFIG_SECURITY
2622         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2623 #endif
2624 #ifdef CONFIG_KALLSYMS
2625         INF("wchan",      S_IRUGO, proc_pid_wchan),
2626 #endif
2627 #ifdef CONFIG_STACKTRACE
2628         ONE("stack",      S_IRUGO, proc_pid_stack),
2629 #endif
2630 #ifdef CONFIG_SCHEDSTATS
2631         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
2632 #endif
2633 #ifdef CONFIG_LATENCYTOP
2634         REG("latency",  S_IRUGO, proc_lstats_operations),
2635 #endif
2636 #ifdef CONFIG_PROC_PID_CPUSET
2637         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2638 #endif
2639 #ifdef CONFIG_CGROUPS
2640         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2641 #endif
2642         INF("oom_score",  S_IRUGO, proc_oom_score),
2643         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2644         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2645 #ifdef CONFIG_AUDITSYSCALL
2646         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2647         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2648 #endif
2649 #ifdef CONFIG_FAULT_INJECTION
2650         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2651 #endif
2652 #ifdef CONFIG_ELF_CORE
2653         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2654 #endif
2655 #ifdef CONFIG_TASK_IO_ACCOUNTING
2656         INF("io",       S_IRUSR, proc_tgid_io_accounting),
2657 #endif
2658 #ifdef CONFIG_HARDWALL
2659         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
2660 #endif
2661 #ifdef CONFIG_USER_NS
2662         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2663         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2664         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2665 #endif
2666 #ifdef CONFIG_CHECKPOINT_RESTORE
2667         REG("timers",     S_IRUGO, proc_timers_operations),
2668 #endif
2669 };
2670
2671 static int proc_tgid_base_readdir(struct file * filp,
2672                              void * dirent, filldir_t filldir)
2673 {
2674         return proc_pident_readdir(filp,dirent,filldir,
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         .readdir        = 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 struct dentry *proc_pid_instantiate(struct inode *dir,
2783                                            struct dentry * dentry,
2784                                            struct task_struct *task, const void *ptr)
2785 {
2786         struct dentry *error = ERR_PTR(-ENOENT);
2787         struct inode *inode;
2788
2789         inode = proc_pid_make_inode(dir->i_sb, task);
2790         if (!inode)
2791                 goto out;
2792
2793         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2794         inode->i_op = &proc_tgid_base_inode_operations;
2795         inode->i_fop = &proc_tgid_base_operations;
2796         inode->i_flags|=S_IMMUTABLE;
2797
2798         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
2799                                                   ARRAY_SIZE(tgid_base_stuff)));
2800
2801         d_set_d_op(dentry, &pid_dentry_operations);
2802
2803         d_add(dentry, inode);
2804         /* Close the race of the process dying before we return the dentry */
2805         if (pid_revalidate(dentry, 0))
2806                 error = NULL;
2807 out:
2808         return error;
2809 }
2810
2811 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
2812 {
2813         struct dentry *result = NULL;
2814         struct task_struct *task;
2815         unsigned tgid;
2816         struct pid_namespace *ns;
2817
2818         tgid = name_to_int(dentry);
2819         if (tgid == ~0U)
2820                 goto out;
2821
2822         ns = dentry->d_sb->s_fs_info;
2823         rcu_read_lock();
2824         task = find_task_by_pid_ns(tgid, ns);
2825         if (task)
2826                 get_task_struct(task);
2827         rcu_read_unlock();
2828         if (!task)
2829                 goto out;
2830
2831         result = proc_pid_instantiate(dir, dentry, task, NULL);
2832         put_task_struct(task);
2833 out:
2834         return result;
2835 }
2836
2837 /*
2838  * Find the first task with tgid >= tgid
2839  *
2840  */
2841 struct tgid_iter {
2842         unsigned int tgid;
2843         struct task_struct *task;
2844 };
2845 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2846 {
2847         struct pid *pid;
2848
2849         if (iter.task)
2850                 put_task_struct(iter.task);
2851         rcu_read_lock();
2852 retry:
2853         iter.task = NULL;
2854         pid = find_ge_pid(iter.tgid, ns);
2855         if (pid) {
2856                 iter.tgid = pid_nr_ns(pid, ns);
2857                 iter.task = pid_task(pid, PIDTYPE_PID);
2858                 /* What we to know is if the pid we have find is the
2859                  * pid of a thread_group_leader.  Testing for task
2860                  * being a thread_group_leader is the obvious thing
2861                  * todo but there is a window when it fails, due to
2862                  * the pid transfer logic in de_thread.
2863                  *
2864                  * So we perform the straight forward test of seeing
2865                  * if the pid we have found is the pid of a thread
2866                  * group leader, and don't worry if the task we have
2867                  * found doesn't happen to be a thread group leader.
2868                  * As we don't care in the case of readdir.
2869                  */
2870                 if (!iter.task || !has_group_leader_pid(iter.task)) {
2871                         iter.tgid += 1;
2872                         goto retry;
2873                 }
2874                 get_task_struct(iter.task);
2875         }
2876         rcu_read_unlock();
2877         return iter;
2878 }
2879
2880 #define TGID_OFFSET (FIRST_PROCESS_ENTRY)
2881
2882 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2883         struct tgid_iter iter)
2884 {
2885         char name[PROC_NUMBUF];
2886         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
2887         return proc_fill_cache(filp, dirent, filldir, name, len,
2888                                 proc_pid_instantiate, iter.task, NULL);
2889 }
2890
2891 static int fake_filldir(void *buf, const char *name, int namelen,
2892                         loff_t offset, u64 ino, unsigned d_type)
2893 {
2894         return 0;
2895 }
2896
2897 /* for the /proc/ directory itself, after non-process stuff has been done */
2898 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2899 {
2900         struct tgid_iter iter;
2901         struct pid_namespace *ns;
2902         filldir_t __filldir;
2903
2904         if (filp->f_pos >= PID_MAX_LIMIT + TGID_OFFSET)
2905                 goto out;
2906
2907         ns = filp->f_dentry->d_sb->s_fs_info;
2908         iter.task = NULL;
2909         iter.tgid = filp->f_pos - TGID_OFFSET;
2910         for (iter = next_tgid(ns, iter);
2911              iter.task;
2912              iter.tgid += 1, iter = next_tgid(ns, iter)) {
2913                 if (has_pid_permissions(ns, iter.task, 2))
2914                         __filldir = filldir;
2915                 else
2916                         __filldir = fake_filldir;
2917
2918                 filp->f_pos = iter.tgid + TGID_OFFSET;
2919                 if (proc_pid_fill_cache(filp, dirent, __filldir, iter) < 0) {
2920                         put_task_struct(iter.task);
2921                         goto out;
2922                 }
2923         }
2924         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
2925 out:
2926         return 0;
2927 }
2928
2929 /*
2930  * Tasks
2931  */
2932 static const struct pid_entry tid_base_stuff[] = {
2933         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2934         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2935         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2936         REG("environ",   S_IRUSR, proc_environ_operations),
2937         INF("auxv",      S_IRUSR, proc_pid_auxv),
2938         ONE("status",    S_IRUGO, proc_pid_status),
2939         ONE("personality", S_IRUGO, proc_pid_personality),
2940         INF("limits",    S_IRUGO, proc_pid_limits),
2941 #ifdef CONFIG_SCHED_DEBUG
2942         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2943 #endif
2944         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2945 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2946         INF("syscall",   S_IRUGO, proc_pid_syscall),
2947 #endif
2948         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
2949         ONE("stat",      S_IRUGO, proc_tid_stat),
2950         ONE("statm",     S_IRUGO, proc_pid_statm),
2951         REG("maps",      S_IRUGO, proc_tid_maps_operations),
2952 #ifdef CONFIG_CHECKPOINT_RESTORE
2953         REG("children",  S_IRUGO, proc_tid_children_operations),
2954 #endif
2955 #ifdef CONFIG_NUMA
2956         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
2957 #endif
2958         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
2959         LNK("cwd",       proc_cwd_link),
2960         LNK("root",      proc_root_link),
2961         LNK("exe",       proc_exe_link),
2962         REG("mounts",    S_IRUGO, proc_mounts_operations),
2963         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2964 #ifdef CONFIG_PROC_PAGE_MONITOR
2965         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2966         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
2967         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
2968 #endif
2969 #ifdef CONFIG_SECURITY
2970         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2971 #endif
2972 #ifdef CONFIG_KALLSYMS
2973         INF("wchan",     S_IRUGO, proc_pid_wchan),
2974 #endif
2975 #ifdef CONFIG_STACKTRACE
2976         ONE("stack",      S_IRUGO, proc_pid_stack),
2977 #endif
2978 #ifdef CONFIG_SCHEDSTATS
2979         INF("schedstat", S_IRUGO, proc_pid_schedstat),
2980 #endif
2981 #ifdef CONFIG_LATENCYTOP
2982         REG("latency",  S_IRUGO, proc_lstats_operations),
2983 #endif
2984 #ifdef CONFIG_PROC_PID_CPUSET
2985         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
2986 #endif
2987 #ifdef CONFIG_CGROUPS
2988         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2989 #endif
2990         INF("oom_score", S_IRUGO, proc_oom_score),
2991         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2992         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2993 #ifdef CONFIG_AUDITSYSCALL
2994         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
2995         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2996 #endif
2997 #ifdef CONFIG_FAULT_INJECTION
2998         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2999 #endif
3000 #ifdef CONFIG_TASK_IO_ACCOUNTING
3001         INF("io",       S_IRUSR, proc_tid_io_accounting),
3002 #endif
3003 #ifdef CONFIG_HARDWALL
3004         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
3005 #endif
3006 #ifdef CONFIG_USER_NS
3007         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3008         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3009         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3010 #endif
3011 };
3012
3013 static int proc_tid_base_readdir(struct file * filp,
3014                              void * dirent, filldir_t filldir)
3015 {
3016         return proc_pident_readdir(filp,dirent,filldir,
3017                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
3018 }
3019
3020 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3021 {
3022         return proc_pident_lookup(dir, dentry,
3023                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3024 }
3025
3026 static const struct file_operations proc_tid_base_operations = {
3027         .read           = generic_read_dir,
3028         .readdir        = proc_tid_base_readdir,
3029         .llseek         = default_llseek,
3030 };
3031
3032 static const struct inode_operations proc_tid_base_inode_operations = {
3033         .lookup         = proc_tid_base_lookup,
3034         .getattr        = pid_getattr,
3035         .setattr        = proc_setattr,
3036 };
3037
3038 static struct dentry *proc_task_instantiate(struct inode *dir,
3039         struct dentry *dentry, struct task_struct *task, const void *ptr)
3040 {
3041         struct dentry *error = ERR_PTR(-ENOENT);
3042         struct inode *inode;
3043         inode = proc_pid_make_inode(dir->i_sb, task);
3044
3045         if (!inode)
3046                 goto out;
3047         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3048         inode->i_op = &proc_tid_base_inode_operations;
3049         inode->i_fop = &proc_tid_base_operations;
3050         inode->i_flags|=S_IMMUTABLE;
3051
3052         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3053                                                   ARRAY_SIZE(tid_base_stuff)));
3054
3055         d_set_d_op(dentry, &pid_dentry_operations);
3056
3057         d_add(dentry, inode);
3058         /* Close the race of the process dying before we return the dentry */
3059         if (pid_revalidate(dentry, 0))
3060                 error = NULL;
3061 out:
3062         return error;
3063 }
3064
3065 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3066 {
3067         struct dentry *result = ERR_PTR(-ENOENT);
3068         struct task_struct *task;
3069         struct task_struct *leader = get_proc_task(dir);
3070         unsigned tid;
3071         struct pid_namespace *ns;
3072
3073         if (!leader)
3074                 goto out_no_task;
3075
3076         tid = name_to_int(dentry);
3077         if (tid == ~0U)
3078                 goto out;
3079
3080         ns = dentry->d_sb->s_fs_info;
3081         rcu_read_lock();
3082         task = find_task_by_pid_ns(tid, ns);
3083         if (task)
3084                 get_task_struct(task);
3085         rcu_read_unlock();
3086         if (!task)
3087                 goto out;
3088         if (!same_thread_group(leader, task))
3089                 goto out_drop_task;
3090
3091         result = proc_task_instantiate(dir, dentry, task, NULL);
3092 out_drop_task:
3093         put_task_struct(task);
3094 out:
3095         put_task_struct(leader);
3096 out_no_task:
3097         return result;
3098 }
3099
3100 /*
3101  * Find the first tid of a thread group to return to user space.
3102  *
3103  * Usually this is just the thread group leader, but if the users
3104  * buffer was too small or there was a seek into the middle of the
3105  * directory we have more work todo.
3106  *
3107  * In the case of a short read we start with find_task_by_pid.
3108  *
3109  * In the case of a seek we start with the leader and walk nr
3110  * threads past it.
3111  */
3112 static struct task_struct *first_tid(struct task_struct *leader,
3113                 int tid, int nr, struct pid_namespace *ns)
3114 {
3115         struct task_struct *pos;
3116
3117         rcu_read_lock();
3118         /* Attempt to start with the pid of a thread */
3119         if (tid && (nr > 0)) {
3120                 pos = find_task_by_pid_ns(tid, ns);
3121                 if (pos && (pos->group_leader == leader))
3122                         goto found;
3123         }
3124
3125         /* If nr exceeds the number of threads there is nothing todo */
3126         pos = NULL;
3127         if (nr && nr >= get_nr_threads(leader))
3128                 goto out;
3129
3130         /* If we haven't found our starting place yet start
3131          * with the leader and walk nr threads forward.
3132          */
3133         for (pos = leader; nr > 0; --nr) {
3134                 pos = next_thread(pos);
3135                 if (pos == leader) {
3136                         pos = NULL;
3137                         goto out;
3138                 }
3139         }
3140 found:
3141         get_task_struct(pos);
3142 out:
3143         rcu_read_unlock();
3144         return pos;
3145 }
3146
3147 /*
3148  * Find the next thread in the thread list.
3149  * Return NULL if there is an error or no next thread.
3150  *
3151  * The reference to the input task_struct is released.
3152  */
3153 static struct task_struct *next_tid(struct task_struct *start)
3154 {
3155         struct task_struct *pos = NULL;
3156         rcu_read_lock();
3157         if (pid_alive(start)) {
3158                 pos = next_thread(start);
3159                 if (thread_group_leader(pos))
3160                         pos = NULL;
3161                 else
3162                         get_task_struct(pos);
3163         }
3164         rcu_read_unlock();
3165         put_task_struct(start);
3166         return pos;
3167 }
3168
3169 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3170         struct task_struct *task, int tid)
3171 {
3172         char name[PROC_NUMBUF];
3173         int len = snprintf(name, sizeof(name), "%d", tid);
3174         return proc_fill_cache(filp, dirent, filldir, name, len,
3175                                 proc_task_instantiate, task, NULL);
3176 }
3177
3178 /* for the /proc/TGID/task/ directories */
3179 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3180 {
3181         struct dentry *dentry = filp->f_path.dentry;
3182         struct inode *inode = dentry->d_inode;
3183         struct task_struct *leader = NULL;
3184         struct task_struct *task;
3185         int retval = -ENOENT;
3186         ino_t ino;
3187         int tid;
3188         struct pid_namespace *ns;
3189
3190         task = get_proc_task(inode);
3191         if (!task)
3192                 goto out_no_task;
3193         rcu_read_lock();
3194         if (pid_alive(task)) {
3195                 leader = task->group_leader;
3196                 get_task_struct(leader);
3197         }
3198         rcu_read_unlock();
3199         put_task_struct(task);
3200         if (!leader)
3201                 goto out_no_task;
3202         retval = 0;
3203
3204         switch ((unsigned long)filp->f_pos) {
3205         case 0:
3206                 ino = inode->i_ino;
3207                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3208                         goto out;
3209                 filp->f_pos++;
3210                 /* fall through */
3211         case 1:
3212                 ino = parent_ino(dentry);
3213                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3214                         goto out;
3215                 filp->f_pos++;
3216                 /* fall through */
3217         }
3218
3219         /* f_version caches the tgid value that the last readdir call couldn't
3220          * return. lseek aka telldir automagically resets f_version to 0.
3221          */
3222         ns = filp->f_dentry->d_sb->s_fs_info;
3223         tid = (int)filp->f_version;
3224         filp->f_version = 0;
3225         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3226              task;
3227              task = next_tid(task), filp->f_pos++) {
3228                 tid = task_pid_nr_ns(task, ns);
3229                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3230                         /* returning this tgid failed, save it as the first
3231                          * pid for the next readir call */
3232                         filp->f_version = (u64)tid;
3233                         put_task_struct(task);
3234                         break;
3235                 }
3236         }
3237 out:
3238         put_task_struct(leader);
3239 out_no_task:
3240         return retval;
3241 }
3242
3243 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3244 {
3245         struct inode *inode = dentry->d_inode;
3246         struct task_struct *p = get_proc_task(inode);
3247         generic_fillattr(inode, stat);
3248
3249         if (p) {
3250                 stat->nlink += get_nr_threads(p);
3251                 put_task_struct(p);
3252         }
3253
3254         return 0;
3255 }
3256
3257 static const struct inode_operations proc_task_inode_operations = {
3258         .lookup         = proc_task_lookup,
3259         .getattr        = proc_task_getattr,
3260         .setattr        = proc_setattr,
3261         .permission     = proc_pid_permission,
3262 };
3263
3264 static const struct file_operations proc_task_operations = {
3265         .read           = generic_read_dir,
3266         .readdir        = proc_task_readdir,
3267         .llseek         = default_llseek,
3268 };