fs/proc/base.c

   1 /*
   2  *  linux/fs/proc/base.c
   3  *
   4  *  Copyright (C) 1991, 1992 Linus Torvalds
   5  *
   6  *  proc base directory handling functions
   7  *
   8  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
   9  *  Instead of using magical inumbers to determine the kind of object
  10  *  we allocate and fill in-core inodes upon lookup. They don't even
  11  *  go into icache. We cache the reference to task_struct upon lookup too.
  12  *  Eventually it should become a filesystem in its own. We don't use the
  13  *  rest of procfs anymore.
  14  *
  15  *
  16  *  Changelog:
  17  *  17-Jan-2005
  18  *  Allan Bezerra
  19  *  Bruna Moreira <bruna.moreira@indt.org.br>
  20  *  Edjard Mota <edjard.mota@indt.org.br>
  21  *  Ilias Biris <ilias.biris@indt.org.br>
  22  *  Mauricio Lin <mauricio.lin@indt.org.br>
  23  *
  24  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  25  *
  26  *  A new process specific entry (smaps) included in /proc. It shows the
  27  *  size of rss for each memory area. The maps entry lacks information
  28  *  about physical memory size (rss) for each mapped file, i.e.,
  29  *  rss information for executables and library files.
  30  *  This additional information is useful for any tools that need to know
  31  *  about physical memory consumption for a process specific library.
  32  *
  33  *  Changelog:
  34  *  21-Feb-2005
  35  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  36  *  Pud inclusion in the page table walking.
  37  *
  38  *  ChangeLog:
  39  *  10-Mar-2005
  40  *  10LE Instituto Nokia de Tecnologia - INdT:
  41  *  A better way to walks through the page table as suggested by Hugh Dickins.
  42  *
  43  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
  44  *  Smaps information related to shared, private, clean and dirty pages.
  45  *
  46  *  Paul Mundt <paul.mundt@nokia.com>:
  47  *  Overall revision about smaps.
  48  */
  49
  50 #include <asm/uaccess.h>
  51
  52 #include <linux/errno.h>
  53 #include <linux/time.h>
  54 #include <linux/proc_fs.h>
  55 #include <linux/stat.h>
  56 #include <linux/task_io_accounting_ops.h>
  57 #include <linux/init.h>
  58 #include <linux/capability.h>
  59 #include <linux/file.h>
  60 #include <linux/fdtable.h>
  61 #include <linux/string.h>
  62 #include <linux/seq_file.h>
  63 #include <linux/namei.h>
  64 #include <linux/mnt_namespace.h>
  65 #include <linux/mm.h>
  66 #include <linux/swap.h>
  67 #include <linux/rcupdate.h>
  68 #include <linux/kallsyms.h>
  69 #include <linux/stacktrace.h>
  70 #include <linux/resource.h>
  71 #include <linux/module.h>
  72 #include <linux/mount.h>
  73 #include <linux/security.h>
  74 #include <linux/ptrace.h>
  75 #include <linux/tracehook.h>
  76 #include <linux/printk.h>
  77 #include <linux/cgroup.h>
  78 #include <linux/cpuset.h>
  79 #include <linux/audit.h>
  80 #include <linux/poll.h>
  81 #include <linux/nsproxy.h>
  82 #include <linux/oom.h>
  83 #include <linux/elf.h>
  84 #include <linux/pid_namespace.h>
  85 #include <linux/user_namespace.h>
  86 #include <linux/fs_struct.h>
  87 #include <linux/slab.h>
  88 #include <linux/flex_array.h>
  89 #include <linux/posix-timers.h>
  90 #ifdef CONFIG_HARDWALL
  91 #include <asm/hardwall.h>
  92 #endif
  93 #include <trace/events/oom.h>
  94 #include "internal.h"
  95 #include "fd.h"
  96
  97 /* NOTE:
  98  *      Implementing inode permission operations in /proc is almost
  99  *      certainly an error.  Permission checks need to happen during
 100  *      each system call not at open time.  The reason is that most of
 101  *      what we wish to check for permissions in /proc varies at runtime.
 102  *
 103  *      The classic example of a problem is opening file descriptors
 104  *      in /proc for a task before it execs a suid executable.
 105  */
 106
 107 struct pid_entry {
 108         const char *name;
 109         int len;
 110         umode_t mode;
 111         const struct inode_operations *iop;
 112         const struct file_operations *fop;
 113         union proc_op op;
 114 };
 115
 116 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 117         .name = (NAME),                                 \
 118         .len  = sizeof(NAME) - 1,                       \
 119         .mode = MODE,                                   \
 120         .iop  = IOP,                                    \
 121         .fop  = FOP,                                    \
 122         .op   = OP,                                     \
 123 }
 124
 125 #define DIR(NAME, MODE, iops, fops)     \
 126         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 127 #define LNK(NAME, get_link)                                     \
 128         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 129                 &proc_pid_link_inode_operations, NULL,          \
 130                 { .proc_get_link = get_link } )
 131 #define REG(NAME, MODE, fops)                           \
 132         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 133 #define ONE(NAME, MODE, show)                           \
 134         NOD(NAME, (S_IFREG|(MODE)),                     \
 135                 NULL, &proc_single_file_operations,     \
 136                 { .proc_show = show } )
 137
 138 /*
 139  * Count the number of hardlinks for the pid_entry table, excluding the .
 140  * and .. links.
 141  */
 142 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
 143         unsigned int n)
 144 {
 145         unsigned int i;
 146         unsigned int count;
 147
 148         count = 0;
 149         for (i = 0; i < n; ++i) {
 150                 if (S_ISDIR(entries[i].mode))
 151                         ++count;
 152         }
 153
 154         return count;
 155 }
 156
 157 static int get_task_root(struct task_struct *task, struct path *root)
 158 {
 159         int result = -ENOENT;
 160
 161         task_lock(task);
 162         if (task->fs) {
 163                 get_fs_root(task->fs, root);
 164                 result = 0;
 165         }
 166         task_unlock(task);
 167         return result;
 168 }
 169
 170 static int proc_cwd_link(struct dentry *dentry, struct path *path)
 171 {
 172         struct task_struct *task = get_proc_task(dentry->d_inode);
 173         int result = -ENOENT;
 174
 175         if (task) {
 176                 task_lock(task);
 177                 if (task->fs) {
 178                         get_fs_pwd(task->fs, path);
 179                         result = 0;
 180                 }
 181                 task_unlock(task);
 182                 put_task_struct(task);
 183         }
 184         return result;
 185 }
 186
 187 static int proc_root_link(struct dentry *dentry, struct path *path)
 188 {
 189         struct task_struct *task = get_proc_task(dentry->d_inode);
 190         int result = -ENOENT;
 191
 192         if (task) {
 193                 result = get_task_root(task, path);
 194                 put_task_struct(task);
 195         }
 196         return result;
 197 }
 198
 199 static int proc_pid_cmdline(struct seq_file *m, struct pid_namespace *ns,
 200                             struct pid *pid, struct task_struct *task)
 201 {
 202         /*
 203          * Rely on struct seq_operations::show() being called once
 204          * per internal buffer allocation. See single_open(), traverse().
 205          */
 206         BUG_ON(m->size < PAGE_SIZE);
 207         m->count += get_cmdline(task, m->buf, PAGE_SIZE);
 208         return 0;
 209 }
 210
 211 static int proc_pid_auxv(struct seq_file *m, struct pid_namespace *ns,
 212                          struct pid *pid, struct task_struct *task)
 213 {
 214         struct mm_struct *mm = mm_access(task, PTRACE_MODE_READ);
 215         if (mm && !IS_ERR(mm)) {
 216                 unsigned int nwords = 0;
 217                 do {
 218                         nwords += 2;
 219                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 220                 seq_write(m, mm->saved_auxv, nwords * sizeof(mm->saved_auxv[0]));
 221                 mmput(mm);
 222                 return 0;
 223         } else
 224                 return PTR_ERR(mm);
 225 }
 226
 227
 228 #ifdef CONFIG_KALLSYMS
 229 /*
 230  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 231  * Returns the resolved symbol.  If that fails, simply return the address.
 232  */
 233 static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
 234                           struct pid *pid, struct task_struct *task)
 235 {
 236         unsigned long wchan;
 237         char symname[KSYM_NAME_LEN];
 238
 239         wchan = get_wchan(task);
 240
 241         if (lookup_symbol_name(wchan, symname) < 0)
 242                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
 243                         return 0;
 244                 else
 245                         return seq_printf(m, "%lu", wchan);
 246         else
 247                 return seq_printf(m, "%s", symname);
 248 }
 249 #endif /* CONFIG_KALLSYMS */
 250
 251 static int lock_trace(struct task_struct *task)
 252 {
 253         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
 254         if (err)
 255                 return err;
 256         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH)) {
 257                 mutex_unlock(&task->signal->cred_guard_mutex);
 258                 return -EPERM;
 259         }
 260         return 0;
 261 }
 262
 263 static void unlock_trace(struct task_struct *task)
 264 {
 265         mutex_unlock(&task->signal->cred_guard_mutex);
 266 }
 267
 268 #ifdef CONFIG_STACKTRACE
 269
 270 #define MAX_STACK_TRACE_DEPTH   64
 271
 272 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 273                           struct pid *pid, struct task_struct *task)
 274 {
 275         struct stack_trace trace;
 276         unsigned long *entries;
 277         int err;
 278         int i;
 279
 280         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 281         if (!entries)
 282                 return -ENOMEM;
 283
 284         trace.nr_entries        = 0;
 285         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 286         trace.entries           = entries;
 287         trace.skip              = 0;
 288
 289         err = lock_trace(task);
 290         if (!err) {
 291                 save_stack_trace_tsk(task, &trace);
 292
 293                 for (i = 0; i < trace.nr_entries; i++) {
 294                         seq_printf(m, "[<%pK>] %pS\n",
 295                                    (void *)entries[i], (void *)entries[i]);
 296                 }
 297                 unlock_trace(task);
 298         }
 299         kfree(entries);
 300
 301         return err;
 302 }
 303 #endif
 304
 305 #ifdef CONFIG_SCHEDSTATS
 306 /*
 307  * Provides /proc/PID/schedstat
 308  */
 309 static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
 310                               struct pid *pid, struct task_struct *task)
 311 {
 312         return seq_printf(m, "%llu %llu %lu\n",
 313                         (unsigned long long)task->se.sum_exec_runtime,
 314                         (unsigned long long)task->sched_info.run_delay,
 315                         task->sched_info.pcount);
 316 }
 317 #endif
 318
 319 #ifdef CONFIG_LATENCYTOP
 320 static int lstats_show_proc(struct seq_file *m, void *v)
 321 {
 322         int i;
 323         struct inode *inode = m->private;
 324         struct task_struct *task = get_proc_task(inode);
 325
 326         if (!task)
 327                 return -ESRCH;
 328         seq_puts(m, "Latency Top version : v0.1\n");
 329         for (i = 0; i < 32; i++) {
 330                 struct latency_record *lr = &task->latency_record[i];
 331                 if (lr->backtrace[0]) {
 332                         int q;
 333                         seq_printf(m, "%i %li %li",
 334                                    lr->count, lr->time, lr->max);
 335                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 336                                 unsigned long bt = lr->backtrace[q];
 337                                 if (!bt)
 338                                         break;
 339                                 if (bt == ULONG_MAX)
 340                                         break;
 341                                 seq_printf(m, " %ps", (void *)bt);
 342                         }
 343                         seq_putc(m, '\n');
 344                 }
 345
 346         }
 347         put_task_struct(task);
 348         return 0;
 349 }
 350
 351 static int lstats_open(struct inode *inode, struct file *file)
 352 {
 353         return single_open(file, lstats_show_proc, inode);
 354 }
 355
 356 static ssize_t lstats_write(struct file *file, const char __user *buf,
 357                             size_t count, loff_t *offs)
 358 {
 359         struct task_struct *task = get_proc_task(file_inode(file));
 360
 361         if (!task)
 362                 return -ESRCH;
 363         clear_all_latency_tracing(task);
 364         put_task_struct(task);
 365
 366         return count;
 367 }
 368
 369 static const struct file_operations proc_lstats_operations = {
 370         .open           = lstats_open,
 371         .read           = seq_read,
 372         .write          = lstats_write,
 373         .llseek         = seq_lseek,
 374         .release        = single_release,
 375 };
 376
 377 #endif
 378
 379 #ifdef CONFIG_CGROUPS
 380 static int cgroup_open(struct inode *inode, struct file *file)
 381 {
 382         struct pid *pid = PROC_I(inode)->pid;
 383         return single_open(file, proc_cgroup_show, pid);
 384 }
 385
 386 static const struct file_operations proc_cgroup_operations = {
 387         .open           = cgroup_open,
 388         .read           = seq_read,
 389         .llseek         = seq_lseek,
 390         .release        = single_release,
 391 };
 392 #endif
 393
 394 #ifdef CONFIG_PROC_PID_CPUSET
 395
 396 static int cpuset_open(struct inode *inode, struct file *file)
 397 {
 398         struct pid *pid = PROC_I(inode)->pid;
 399         return single_open(file, proc_cpuset_show, pid);
 400 }
 401
 402 static const struct file_operations proc_cpuset_operations = {
 403         .open           = cpuset_open,
 404         .read           = seq_read,
 405         .llseek         = seq_lseek,
 406         .release        = single_release,
 407 };
 408 #endif
 409
 410 static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
 411                           struct pid *pid, struct task_struct *task)
 412 {
 413         unsigned long totalpages = totalram_pages + total_swap_pages;
 414         unsigned long points = 0;
 415
 416         read_lock(&tasklist_lock);
 417         if (pid_alive(task))
 418                 points = oom_badness(task, NULL, NULL, totalpages) *
 419                                                 1000 / totalpages;
 420         read_unlock(&tasklist_lock);
 421         return seq_printf(m, "%lu\n", points);
 422 }
 423
 424 struct limit_names {
 425         const char *name;
 426         const char *unit;
 427 };
 428
 429 static const struct limit_names lnames[RLIM_NLIMITS] = {
 430         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 431         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 432         [RLIMIT_DATA] = {"Max data size", "bytes"},
 433         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 434         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 435         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 436         [RLIMIT_NPROC] = {"Max processes", "processes"},
 437         [RLIMIT_NOFILE] = {"Max open files", "files"},
 438         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 439         [RLIMIT_AS] = {"Max address space", "bytes"},
 440         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 441         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 442         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 443         [RLIMIT_NICE] = {"Max nice priority", NULL},
 444         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 445         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 446 };
 447
 448 /* Display limits for a process */
 449 static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
 450                            struct pid *pid, struct task_struct *task)
 451 {
 452         unsigned int i;
 453         unsigned long flags;
 454
 455         struct rlimit rlim[RLIM_NLIMITS];
 456
 457         if (!lock_task_sighand(task, &flags))
 458                 return 0;
 459         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 460         unlock_task_sighand(task, &flags);
 461
 462         /*
 463          * print the file header
 464          */
 465        seq_printf(m, "%-25s %-20s %-20s %-10s\n",
 466                         "Limit", "Soft Limit", "Hard Limit", "Units");
 467
 468         for (i = 0; i < RLIM_NLIMITS; i++) {
 469                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 470                         seq_printf(m, "%-25s %-20s ",
 471                                          lnames[i].name, "unlimited");
 472                 else
 473                         seq_printf(m, "%-25s %-20lu ",
 474                                          lnames[i].name, rlim[i].rlim_cur);
 475
 476                 if (rlim[i].rlim_max == RLIM_INFINITY)
 477                         seq_printf(m, "%-20s ", "unlimited");
 478                 else
 479                         seq_printf(m, "%-20lu ", rlim[i].rlim_max);
 480
 481                 if (lnames[i].unit)
 482                         seq_printf(m, "%-10s\n", lnames[i].unit);
 483                 else
 484                         seq_putc(m, '\n');
 485         }
 486
 487         return 0;
 488 }
 489
 490 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 491 static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
 492                             struct pid *pid, struct task_struct *task)
 493 {
 494         long nr;
 495         unsigned long args[6], sp, pc;
 496         int res = lock_trace(task);
 497         if (res)
 498                 return res;
 499
 500         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 501                 seq_puts(m, "running\n");
 502         else if (nr < 0)
 503                 seq_printf(m, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 504         else
 505                 seq_printf(m,
 506                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 507                        nr,
 508                        args[0], args[1], args[2], args[3], args[4], args[5],
 509                        sp, pc);
 510         unlock_trace(task);
 511         return res;
 512 }
 513 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 514
 515 /************************************************************************/
 516 /*                       Here the fs part begins                        */
 517 /************************************************************************/
 518
 519 /* permission checks */
 520 static int proc_fd_access_allowed(struct inode *inode)
 521 {
 522         struct task_struct *task;
 523         int allowed = 0;
 524         /* Allow access to a task's file descriptors if it is us or we
 525          * may use ptrace attach to the process and find out that
 526          * information.
 527          */
 528         task = get_proc_task(inode);
 529         if (task) {
 530                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
 531                 put_task_struct(task);
 532         }
 533         return allowed;
 534 }
 535
 536 int proc_setattr(struct dentry *dentry, struct iattr *attr)
 537 {
 538         int error;
 539         struct inode *inode = dentry->d_inode;
 540
 541         if (attr->ia_valid & ATTR_MODE)
 542                 return -EPERM;
 543
 544         error = inode_change_ok(inode, attr);
 545         if (error)
 546                 return error;
 547
 548         setattr_copy(inode, attr);
 549         mark_inode_dirty(inode);
 550         return 0;
 551 }
 552
 553 /*
 554  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
 555  * or euid/egid (for hide_pid_min=2)?
 556  */
 557 static bool has_pid_permissions(struct pid_namespace *pid,
 558                                  struct task_struct *task,
 559                                  int hide_pid_min)
 560 {
 561         if (pid->hide_pid < hide_pid_min)
 562                 return true;
 563         if (in_group_p(pid->pid_gid))
 564                 return true;
 565         return ptrace_may_access(task, PTRACE_MODE_READ);
 566 }
 567
 568
 569 static int proc_pid_permission(struct inode *inode, int mask)
 570 {
 571         struct pid_namespace *pid = inode->i_sb->s_fs_info;
 572         struct task_struct *task;
 573         bool has_perms;
 574
 575         task = get_proc_task(inode);
 576         if (!task)
 577                 return -ESRCH;
 578         has_perms = has_pid_permissions(pid, task, 1);
 579         put_task_struct(task);
 580
 581         if (!has_perms) {
 582                 if (pid->hide_pid == 2) {
 583                         /*
 584                          * Let's make getdents(), stat(), and open()
 585                          * consistent with each other.  If a process
 586                          * may not stat() a file, it shouldn't be seen
 587                          * in procfs at all.
 588                          */
 589                         return -ENOENT;
 590                 }
 591
 592                 return -EPERM;
 593         }
 594         return generic_permission(inode, mask);
 595 }
 596
 597
 598
 599 static const struct inode_operations proc_def_inode_operations = {
 600         .setattr        = proc_setattr,
 601 };
 602
 603 static int proc_single_show(struct seq_file *m, void *v)
 604 {
 605         struct inode *inode = m->private;
 606         struct pid_namespace *ns;
 607         struct pid *pid;
 608         struct task_struct *task;
 609         int ret;
 610
 611         ns = inode->i_sb->s_fs_info;
 612         pid = proc_pid(inode);
 613         task = get_pid_task(pid, PIDTYPE_PID);
 614         if (!task)
 615                 return -ESRCH;
 616
 617         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 618
 619         put_task_struct(task);
 620         return ret;
 621 }
 622
 623 static int proc_single_open(struct inode *inode, struct file *filp)
 624 {
 625         return single_open(filp, proc_single_show, inode);
 626 }
 627
 628 static const struct file_operations proc_single_file_operations = {
 629         .open           = proc_single_open,
 630         .read           = seq_read,
 631         .llseek         = seq_lseek,
 632         .release        = single_release,
 633 };
 634
 635 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
 636 {
 637         struct task_struct *task = get_proc_task(file_inode(file));
 638         struct mm_struct *mm;
 639
 640         if (!task)
 641                 return -ESRCH;
 642
 643         mm = mm_access(task, mode);
 644         put_task_struct(task);
 645
 646         if (IS_ERR(mm))
 647                 return PTR_ERR(mm);
 648
 649         if (mm) {
 650                 /* ensure this mm_struct can't be freed */
 651                 atomic_inc(&mm->mm_count);
 652                 /* but do not pin its memory */
 653                 mmput(mm);
 654         }
 655
 656         file->private_data = mm;
 657
 658         return 0;
 659 }
 660
 661 static int mem_open(struct inode *inode, struct file *file)
 662 {
 663         int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
 664
 665         /* OK to pass negative loff_t, we can catch out-of-range */
 666         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 667
 668         return ret;
 669 }
 670
 671 static ssize_t mem_rw(struct file *file, char __user *buf,
 672                         size_t count, loff_t *ppos, int write)
 673 {
 674         struct mm_struct *mm = file->private_data;
 675         unsigned long addr = *ppos;
 676         ssize_t copied;
 677         char *page;
 678
 679         if (!mm)
 680                 return 0;
 681
 682         page = (char *)__get_free_page(GFP_TEMPORARY);
 683         if (!page)
 684                 return -ENOMEM;
 685
 686         copied = 0;
 687         if (!atomic_inc_not_zero(&mm->mm_users))
 688                 goto free;
 689
 690         while (count > 0) {
 691                 int this_len = min_t(int, count, PAGE_SIZE);
 692
 693                 if (write && copy_from_user(page, buf, this_len)) {
 694                         copied = -EFAULT;
 695                         break;
 696                 }
 697
 698                 this_len = access_remote_vm(mm, addr, page, this_len, write);
 699                 if (!this_len) {
 700                         if (!copied)
 701                                 copied = -EIO;
 702                         break;
 703                 }
 704
 705                 if (!write && copy_to_user(buf, page, this_len)) {
 706                         copied = -EFAULT;
 707                         break;
 708                 }
 709
 710                 buf += this_len;
 711                 addr += this_len;
 712                 copied += this_len;
 713                 count -= this_len;
 714         }
 715         *ppos = addr;
 716
 717         mmput(mm);
 718 free:
 719         free_page((unsigned long) page);
 720         return copied;
 721 }
 722
 723 static ssize_t mem_read(struct file *file, char __user *buf,
 724                         size_t count, loff_t *ppos)
 725 {
 726         return mem_rw(file, buf, count, ppos, 0);
 727 }
 728
 729 static ssize_t mem_write(struct file *file, const char __user *buf,
 730                          size_t count, loff_t *ppos)
 731 {
 732         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 733 }
 734
 735 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 736 {
 737         switch (orig) {
 738         case 0:
 739                 file->f_pos = offset;
 740                 break;
 741         case 1:
 742                 file->f_pos += offset;
 743                 break;
 744         default:
 745                 return -EINVAL;
 746         }
 747         force_successful_syscall_return();
 748         return file->f_pos;
 749 }
 750
 751 static int mem_release(struct inode *inode, struct file *file)
 752 {
 753         struct mm_struct *mm = file->private_data;
 754         if (mm)
 755                 mmdrop(mm);
 756         return 0;
 757 }
 758
 759 static const struct file_operations proc_mem_operations = {
 760         .llseek         = mem_lseek,
 761         .read           = mem_read,
 762         .write          = mem_write,
 763         .open           = mem_open,
 764         .release        = mem_release,
 765 };
 766
 767 static int environ_open(struct inode *inode, struct file *file)
 768 {
 769         return __mem_open(inode, file, PTRACE_MODE_READ);
 770 }
 771
 772 static ssize_t environ_read(struct file *file, char __user *buf,
 773                         size_t count, loff_t *ppos)
 774 {
 775         char *page;
 776         unsigned long src = *ppos;
 777         int ret = 0;
 778         struct mm_struct *mm = file->private_data;
 779
 780         if (!mm)
 781                 return 0;
 782
 783         page = (char *)__get_free_page(GFP_TEMPORARY);
 784         if (!page)
 785                 return -ENOMEM;
 786
 787         ret = 0;
 788         if (!atomic_inc_not_zero(&mm->mm_users))
 789                 goto free;
 790         while (count > 0) {
 791                 size_t this_len, max_len;
 792                 int retval;
 793
 794                 if (src >= (mm->env_end - mm->env_start))
 795                         break;
 796
 797                 this_len = mm->env_end - (mm->env_start + src);
 798
 799                 max_len = min_t(size_t, PAGE_SIZE, count);
 800                 this_len = min(max_len, this_len);
 801
 802                 retval = access_remote_vm(mm, (mm->env_start + src),
 803                         page, this_len, 0);
 804
 805                 if (retval <= 0) {
 806                         ret = retval;
 807                         break;
 808                 }
 809
 810                 if (copy_to_user(buf, page, retval)) {
 811                         ret = -EFAULT;
 812                         break;
 813                 }
 814
 815                 ret += retval;
 816                 src += retval;
 817                 buf += retval;
 818                 count -= retval;
 819         }
 820         *ppos = src;
 821         mmput(mm);
 822
 823 free:
 824         free_page((unsigned long) page);
 825         return ret;
 826 }
 827
 828 static const struct file_operations proc_environ_operations = {
 829         .open           = environ_open,
 830         .read           = environ_read,
 831         .llseek         = generic_file_llseek,
 832         .release        = mem_release,
 833 };
 834
 835 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
 836                             loff_t *ppos)
 837 {
 838         struct task_struct *task = get_proc_task(file_inode(file));
 839         char buffer[PROC_NUMBUF];
 840         int oom_adj = OOM_ADJUST_MIN;
 841         size_t len;
 842         unsigned long flags;
 843
 844         if (!task)
 845                 return -ESRCH;
 846         if (lock_task_sighand(task, &flags)) {
 847                 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
 848                         oom_adj = OOM_ADJUST_MAX;
 849                 else
 850                         oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
 851                                   OOM_SCORE_ADJ_MAX;
 852                 unlock_task_sighand(task, &flags);
 853         }
 854         put_task_struct(task);
 855         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
 856         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 857 }
 858
 859 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
 860                              size_t count, loff_t *ppos)
 861 {
 862         struct task_struct *task;
 863         char buffer[PROC_NUMBUF];
 864         int oom_adj;
 865         unsigned long flags;
 866         int err;
 867
 868         memset(buffer, 0, sizeof(buffer));
 869         if (count > sizeof(buffer) - 1)
 870                 count = sizeof(buffer) - 1;
 871         if (copy_from_user(buffer, buf, count)) {
 872                 err = -EFAULT;
 873                 goto out;
 874         }
 875
 876         err = kstrtoint(strstrip(buffer), 0, &oom_adj);
 877         if (err)
 878                 goto out;
 879         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
 880              oom_adj != OOM_DISABLE) {
 881                 err = -EINVAL;
 882                 goto out;
 883         }
 884
 885         task = get_proc_task(file_inode(file));
 886         if (!task) {
 887                 err = -ESRCH;
 888                 goto out;
 889         }
 890
 891         task_lock(task);
 892         if (!task->mm) {
 893                 err = -EINVAL;
 894                 goto err_task_lock;
 895         }
 896
 897         if (!lock_task_sighand(task, &flags)) {
 898                 err = -ESRCH;
 899                 goto err_task_lock;
 900         }
 901
 902         /*
 903          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
 904          * value is always attainable.
 905          */
 906         if (oom_adj == OOM_ADJUST_MAX)
 907                 oom_adj = OOM_SCORE_ADJ_MAX;
 908         else
 909                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
 910
 911         if (oom_adj < task->signal->oom_score_adj &&
 912             !capable(CAP_SYS_RESOURCE)) {
 913                 err = -EACCES;
 914                 goto err_sighand;
 915         }
 916
 917         /*
 918          * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
 919          * /proc/pid/oom_score_adj instead.
 920          */
 921         pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
 922                   current->comm, task_pid_nr(current), task_pid_nr(task),
 923                   task_pid_nr(task));
 924
 925         task->signal->oom_score_adj = oom_adj;
 926         trace_oom_score_adj_update(task);
 927 err_sighand:
 928         unlock_task_sighand(task, &flags);
 929 err_task_lock:
 930         task_unlock(task);
 931         put_task_struct(task);
 932 out:
 933         return err < 0 ? err : count;
 934 }
 935
 936 static const struct file_operations proc_oom_adj_operations = {
 937         .read           = oom_adj_read,
 938         .write          = oom_adj_write,
 939         .llseek         = generic_file_llseek,
 940 };
 941
 942 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
 943                                         size_t count, loff_t *ppos)
 944 {
 945         struct task_struct *task = get_proc_task(file_inode(file));
 946         char buffer[PROC_NUMBUF];
 947         short oom_score_adj = OOM_SCORE_ADJ_MIN;
 948         unsigned long flags;
 949         size_t len;
 950
 951         if (!task)
 952                 return -ESRCH;
 953         if (lock_task_sighand(task, &flags)) {
 954                 oom_score_adj = task->signal->oom_score_adj;
 955                 unlock_task_sighand(task, &flags);
 956         }
 957         put_task_struct(task);
 958         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
 959         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 960 }
 961
 962 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
 963                                         size_t count, loff_t *ppos)
 964 {
 965         struct task_struct *task;
 966         char buffer[PROC_NUMBUF];
 967         unsigned long flags;
 968         int oom_score_adj;
 969         int err;
 970
 971         memset(buffer, 0, sizeof(buffer));
 972         if (count > sizeof(buffer) - 1)
 973                 count = sizeof(buffer) - 1;
 974         if (copy_from_user(buffer, buf, count)) {
 975                 err = -EFAULT;
 976                 goto out;
 977         }
 978
 979         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
 980         if (err)
 981                 goto out;
 982         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
 983                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
 984                 err = -EINVAL;
 985                 goto out;
 986         }
 987
 988         task = get_proc_task(file_inode(file));
 989         if (!task) {
 990                 err = -ESRCH;
 991                 goto out;
 992         }
 993
 994         task_lock(task);
 995         if (!task->mm) {
 996                 err = -EINVAL;
 997                 goto err_task_lock;
 998         }
 999
1000         if (!lock_task_sighand(task, &flags)) {
1001                 err = -ESRCH;
1002                 goto err_task_lock;
1003         }
1004
1005         if ((short)oom_score_adj < task->signal->oom_score_adj_min &&
1006                         !capable(CAP_SYS_RESOURCE)) {
1007                 err = -EACCES;
1008                 goto err_sighand;
1009         }
1010
1011         task->signal->oom_score_adj = (short)oom_score_adj;
1012         if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1013                 task->signal->oom_score_adj_min = (short)oom_score_adj;
1014         trace_oom_score_adj_update(task);
1015
1016 err_sighand:
1017         unlock_task_sighand(task, &flags);
1018 err_task_lock:
1019         task_unlock(task);
1020         put_task_struct(task);
1021 out:
1022         return err < 0 ? err : count;
1023 }
1024
1025 static const struct file_operations proc_oom_score_adj_operations = {
1026         .read           = oom_score_adj_read,
1027         .write          = oom_score_adj_write,
1028         .llseek         = default_llseek,
1029 };
1030
1031 #ifdef CONFIG_AUDITSYSCALL
1032 #define TMPBUFLEN 21
1033 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1034                                   size_t count, loff_t *ppos)
1035 {
1036         struct inode * inode = file_inode(file);
1037         struct task_struct *task = get_proc_task(inode);
1038         ssize_t length;
1039         char tmpbuf[TMPBUFLEN];
1040
1041         if (!task)
1042                 return -ESRCH;
1043         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1044                            from_kuid(file->f_cred->user_ns,
1045                                      audit_get_loginuid(task)));
1046         put_task_struct(task);
1047         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1048 }
1049
1050 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1051                                    size_t count, loff_t *ppos)
1052 {
1053         struct inode * inode = file_inode(file);
1054         char *page, *tmp;
1055         ssize_t length;
1056         uid_t loginuid;
1057         kuid_t kloginuid;
1058
1059         rcu_read_lock();
1060         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1061                 rcu_read_unlock();
1062                 return -EPERM;
1063         }
1064         rcu_read_unlock();
1065
1066         if (count >= PAGE_SIZE)
1067                 count = PAGE_SIZE - 1;
1068
1069         if (*ppos != 0) {
1070                 /* No partial writes. */
1071                 return -EINVAL;
1072         }
1073         page = (char*)__get_free_page(GFP_TEMPORARY);
1074         if (!page)
1075                 return -ENOMEM;
1076         length = -EFAULT;
1077         if (copy_from_user(page, buf, count))
1078                 goto out_free_page;
1079
1080         page[count] = '\0';
1081         loginuid = simple_strtoul(page, &tmp, 10);
1082         if (tmp == page) {
1083                 length = -EINVAL;
1084                 goto out_free_page;
1085
1086         }
1087
1088         /* is userspace tring to explicitly UNSET the loginuid? */
1089         if (loginuid == AUDIT_UID_UNSET) {
1090                 kloginuid = INVALID_UID;
1091         } else {
1092                 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1093                 if (!uid_valid(kloginuid)) {
1094                         length = -EINVAL;
1095                         goto out_free_page;
1096                 }
1097         }
1098
1099         length = audit_set_loginuid(kloginuid);
1100         if (likely(length == 0))
1101                 length = count;
1102
1103 out_free_page:
1104         free_page((unsigned long) page);
1105         return length;
1106 }
1107
1108 static const struct file_operations proc_loginuid_operations = {
1109         .read           = proc_loginuid_read,
1110         .write          = proc_loginuid_write,
1111         .llseek         = generic_file_llseek,
1112 };
1113
1114 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1115                                   size_t count, loff_t *ppos)
1116 {
1117         struct inode * inode = file_inode(file);
1118         struct task_struct *task = get_proc_task(inode);
1119         ssize_t length;
1120         char tmpbuf[TMPBUFLEN];
1121
1122         if (!task)
1123                 return -ESRCH;
1124         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1125                                 audit_get_sessionid(task));
1126         put_task_struct(task);
1127         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1128 }
1129
1130 static const struct file_operations proc_sessionid_operations = {
1131         .read           = proc_sessionid_read,
1132         .llseek         = generic_file_llseek,
1133 };
1134 #endif
1135
1136 #ifdef CONFIG_FAULT_INJECTION
1137 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1138                                       size_t count, loff_t *ppos)
1139 {
1140         struct task_struct *task = get_proc_task(file_inode(file));
1141         char buffer[PROC_NUMBUF];
1142         size_t len;
1143         int make_it_fail;
1144
1145         if (!task)
1146                 return -ESRCH;
1147         make_it_fail = task->make_it_fail;
1148         put_task_struct(task);
1149
1150         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1151
1152         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1153 }
1154
1155 static ssize_t proc_fault_inject_write(struct file * file,
1156                         const char __user * buf, size_t count, loff_t *ppos)
1157 {
1158         struct task_struct *task;
1159         char buffer[PROC_NUMBUF], *end;
1160         int make_it_fail;
1161
1162         if (!capable(CAP_SYS_RESOURCE))
1163                 return -EPERM;
1164         memset(buffer, 0, sizeof(buffer));
1165         if (count > sizeof(buffer) - 1)
1166                 count = sizeof(buffer) - 1;
1167         if (copy_from_user(buffer, buf, count))
1168                 return -EFAULT;
1169         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1170         if (*end)
1171                 return -EINVAL;
1172         if (make_it_fail < 0 || make_it_fail > 1)
1173                 return -EINVAL;
1174
1175         task = get_proc_task(file_inode(file));
1176         if (!task)
1177                 return -ESRCH;
1178         task->make_it_fail = make_it_fail;
1179         put_task_struct(task);
1180
1181         return count;
1182 }
1183
1184 static const struct file_operations proc_fault_inject_operations = {
1185         .read           = proc_fault_inject_read,
1186         .write          = proc_fault_inject_write,
1187         .llseek         = generic_file_llseek,
1188 };
1189 #endif
1190
1191
1192 #ifdef CONFIG_SCHED_DEBUG
1193 /*
1194  * Print out various scheduling related per-task fields:
1195  */
1196 static int sched_show(struct seq_file *m, void *v)
1197 {
1198         struct inode *inode = m->private;
1199         struct task_struct *p;
1200
1201         p = get_proc_task(inode);
1202         if (!p)
1203                 return -ESRCH;
1204         proc_sched_show_task(p, m);
1205
1206         put_task_struct(p);
1207
1208         return 0;
1209 }
1210
1211 static ssize_t
1212 sched_write(struct file *file, const char __user *buf,
1213             size_t count, loff_t *offset)
1214 {
1215         struct inode *inode = file_inode(file);
1216         struct task_struct *p;
1217
1218         p = get_proc_task(inode);
1219         if (!p)
1220                 return -ESRCH;
1221         proc_sched_set_task(p);
1222
1223         put_task_struct(p);
1224
1225         return count;
1226 }
1227
1228 static int sched_open(struct inode *inode, struct file *filp)
1229 {
1230         return single_open(filp, sched_show, inode);
1231 }
1232
1233 static const struct file_operations proc_pid_sched_operations = {
1234         .open           = sched_open,
1235         .read           = seq_read,
1236         .write          = sched_write,
1237         .llseek         = seq_lseek,
1238         .release        = single_release,
1239 };
1240
1241 #endif
1242
1243 #ifdef CONFIG_SCHED_AUTOGROUP
1244 /*
1245  * Print out autogroup related information:
1246  */
1247 static int sched_autogroup_show(struct seq_file *m, void *v)
1248 {
1249         struct inode *inode = m->private;
1250         struct task_struct *p;
1251
1252         p = get_proc_task(inode);
1253         if (!p)
1254                 return -ESRCH;
1255         proc_sched_autogroup_show_task(p, m);
1256
1257         put_task_struct(p);
1258
1259         return 0;
1260 }
1261
1262 static ssize_t
1263 sched_autogroup_write(struct file *file, const char __user *buf,
1264             size_t count, loff_t *offset)
1265 {
1266         struct inode *inode = file_inode(file);
1267         struct task_struct *p;
1268         char buffer[PROC_NUMBUF];
1269         int nice;
1270         int err;
1271
1272         memset(buffer, 0, sizeof(buffer));
1273         if (count > sizeof(buffer) - 1)
1274                 count = sizeof(buffer) - 1;
1275         if (copy_from_user(buffer, buf, count))
1276                 return -EFAULT;
1277
1278         err = kstrtoint(strstrip(buffer), 0, &nice);
1279         if (err < 0)
1280                 return err;
1281
1282         p = get_proc_task(inode);
1283         if (!p)
1284                 return -ESRCH;
1285
1286         err = proc_sched_autogroup_set_nice(p, nice);
1287         if (err)
1288                 count = err;
1289
1290         put_task_struct(p);
1291
1292         return count;
1293 }
1294
1295 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1296 {
1297         int ret;
1298
1299         ret = single_open(filp, sched_autogroup_show, NULL);
1300         if (!ret) {
1301                 struct seq_file *m = filp->private_data;
1302
1303                 m->private = inode;
1304         }
1305         return ret;
1306 }
1307
1308 static const struct file_operations proc_pid_sched_autogroup_operations = {
1309         .open           = sched_autogroup_open,
1310         .read           = seq_read,
1311         .write          = sched_autogroup_write,
1312         .llseek         = seq_lseek,
1313         .release        = single_release,
1314 };
1315
1316 #endif /* CONFIG_SCHED_AUTOGROUP */
1317
1318 static ssize_t comm_write(struct file *file, const char __user *buf,
1319                                 size_t count, loff_t *offset)
1320 {
1321         struct inode *inode = file_inode(file);
1322         struct task_struct *p;
1323         char buffer[TASK_COMM_LEN];
1324         const size_t maxlen = sizeof(buffer) - 1;
1325
1326         memset(buffer, 0, sizeof(buffer));
1327         if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1328                 return -EFAULT;
1329
1330         p = get_proc_task(inode);
1331         if (!p)
1332                 return -ESRCH;
1333
1334         if (same_thread_group(current, p))
1335                 set_task_comm(p, buffer);
1336         else
1337                 count = -EINVAL;
1338
1339         put_task_struct(p);
1340
1341         return count;
1342 }
1343
1344 static int comm_show(struct seq_file *m, void *v)
1345 {
1346         struct inode *inode = m->private;
1347         struct task_struct *p;
1348
1349         p = get_proc_task(inode);
1350         if (!p)
1351                 return -ESRCH;
1352
1353         task_lock(p);
1354         seq_printf(m, "%s\n", p->comm);
1355         task_unlock(p);
1356
1357         put_task_struct(p);
1358
1359         return 0;
1360 }
1361
1362 static int comm_open(struct inode *inode, struct file *filp)
1363 {
1364         return single_open(filp, comm_show, inode);
1365 }
1366
1367 static const struct file_operations proc_pid_set_comm_operations = {
1368         .open           = comm_open,
1369         .read           = seq_read,
1370         .write          = comm_write,
1371         .llseek         = seq_lseek,
1372         .release        = single_release,
1373 };
1374
1375 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1376 {
1377         struct task_struct *task;
1378         struct mm_struct *mm;
1379         struct file *exe_file;
1380
1381         task = get_proc_task(dentry->d_inode);
1382         if (!task)
1383                 return -ENOENT;
1384         mm = get_task_mm(task);
1385         put_task_struct(task);
1386         if (!mm)
1387                 return -ENOENT;
1388         exe_file = get_mm_exe_file(mm);
1389         mmput(mm);
1390         if (exe_file) {
1391                 *exe_path = exe_file->f_path;
1392                 path_get(&exe_file->f_path);
1393                 fput(exe_file);
1394                 return 0;
1395         } else
1396                 return -ENOENT;
1397 }
1398
1399 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1400 {
1401         struct inode *inode = dentry->d_inode;
1402         struct path path;
1403         int error = -EACCES;
1404
1405         /* Are we allowed to snoop on the tasks file descriptors? */
1406         if (!proc_fd_access_allowed(inode))
1407                 goto out;
1408
1409         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1410         if (error)
1411                 goto out;
1412
1413         nd_jump_link(nd, &path);
1414         return NULL;
1415 out:
1416         return ERR_PTR(error);
1417 }
1418
1419 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1420 {
1421         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1422         char *pathname;
1423         int len;
1424
1425         if (!tmp)
1426                 return -ENOMEM;
1427
1428         pathname = d_path(path, tmp, PAGE_SIZE);
1429         len = PTR_ERR(pathname);
1430         if (IS_ERR(pathname))
1431                 goto out;
1432         len = tmp + PAGE_SIZE - 1 - pathname;
1433
1434         if (len > buflen)
1435                 len = buflen;
1436         if (copy_to_user(buffer, pathname, len))
1437                 len = -EFAULT;
1438  out:
1439         free_page((unsigned long)tmp);
1440         return len;
1441 }
1442
1443 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1444 {
1445         int error = -EACCES;
1446         struct inode *inode = dentry->d_inode;
1447         struct path path;
1448
1449         /* Are we allowed to snoop on the tasks file descriptors? */
1450         if (!proc_fd_access_allowed(inode))
1451                 goto out;
1452
1453         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1454         if (error)
1455                 goto out;
1456
1457         error = do_proc_readlink(&path, buffer, buflen);
1458         path_put(&path);
1459 out:
1460         return error;
1461 }
1462
1463 const struct inode_operations proc_pid_link_inode_operations = {
1464         .readlink       = proc_pid_readlink,
1465         .follow_link    = proc_pid_follow_link,
1466         .setattr        = proc_setattr,
1467 };
1468
1469
1470 /* building an inode */
1471
1472 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1473 {
1474         struct inode * inode;
1475         struct proc_inode *ei;
1476         const struct cred *cred;
1477
1478         /* We need a new inode */
1479
1480         inode = new_inode(sb);
1481         if (!inode)
1482                 goto out;
1483
1484         /* Common stuff */
1485         ei = PROC_I(inode);
1486         inode->i_ino = get_next_ino();
1487         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1488         inode->i_op = &proc_def_inode_operations;
1489
1490         /*
1491          * grab the reference to task.
1492          */
1493         ei->pid = get_task_pid(task, PIDTYPE_PID);
1494         if (!ei->pid)
1495                 goto out_unlock;
1496
1497         if (task_dumpable(task)) {
1498                 rcu_read_lock();
1499                 cred = __task_cred(task);
1500                 inode->i_uid = cred->euid;
1501                 inode->i_gid = cred->egid;
1502                 rcu_read_unlock();
1503         }
1504         security_task_to_inode(task, inode);
1505
1506 out:
1507         return inode;
1508
1509 out_unlock:
1510         iput(inode);
1511         return NULL;
1512 }
1513
1514 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1515 {
1516         struct inode *inode = dentry->d_inode;
1517         struct task_struct *task;
1518         const struct cred *cred;
1519         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1520
1521         generic_fillattr(inode, stat);
1522
1523         rcu_read_lock();
1524         stat->uid = GLOBAL_ROOT_UID;
1525         stat->gid = GLOBAL_ROOT_GID;
1526         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1527         if (task) {
1528                 if (!has_pid_permissions(pid, task, 2)) {
1529                         rcu_read_unlock();
1530                         /*
1531                          * This doesn't prevent learning whether PID exists,
1532                          * it only makes getattr() consistent with readdir().
1533                          */
1534                         return -ENOENT;
1535                 }
1536                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1537                     task_dumpable(task)) {
1538                         cred = __task_cred(task);
1539                         stat->uid = cred->euid;
1540                         stat->gid = cred->egid;
1541                 }
1542         }
1543         rcu_read_unlock();
1544         return 0;
1545 }
1546
1547 /* dentry stuff */
1548
1549 /*
1550  *      Exceptional case: normally we are not allowed to unhash a busy
1551  * directory. In this case, however, we can do it - no aliasing problems
1552  * due to the way we treat inodes.
1553  *
1554  * Rewrite the inode's ownerships here because the owning task may have
1555  * performed a setuid(), etc.
1556  *
1557  * Before the /proc/pid/status file was created the only way to read
1558  * the effective uid of a /process was to stat /proc/pid.  Reading
1559  * /proc/pid/status is slow enough that procps and other packages
1560  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1561  * made this apply to all per process world readable and executable
1562  * directories.
1563  */
1564 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1565 {
1566         struct inode *inode;
1567         struct task_struct *task;
1568         const struct cred *cred;
1569
1570         if (flags & LOOKUP_RCU)
1571                 return -ECHILD;
1572
1573         inode = dentry->d_inode;
1574         task = get_proc_task(inode);
1575
1576         if (task) {
1577                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1578                     task_dumpable(task)) {
1579                         rcu_read_lock();
1580                         cred = __task_cred(task);
1581                         inode->i_uid = cred->euid;
1582                         inode->i_gid = cred->egid;
1583                         rcu_read_unlock();
1584                 } else {
1585                         inode->i_uid = GLOBAL_ROOT_UID;
1586                         inode->i_gid = GLOBAL_ROOT_GID;
1587                 }
1588                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1589                 security_task_to_inode(task, inode);
1590                 put_task_struct(task);
1591                 return 1;
1592         }
1593         d_drop(dentry);
1594         return 0;
1595 }
1596
1597 static inline bool proc_inode_is_dead(struct inode *inode)
1598 {
1599         return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1600 }
1601
1602 int pid_delete_dentry(const struct dentry *dentry)
1603 {
1604         /* Is the task we represent dead?
1605          * If so, then don't put the dentry on the lru list,
1606          * kill it immediately.
1607          */
1608         return proc_inode_is_dead(dentry->d_inode);
1609 }
1610
1611 const struct dentry_operations pid_dentry_operations =
1612 {
1613         .d_revalidate   = pid_revalidate,
1614         .d_delete       = pid_delete_dentry,
1615 };
1616
1617 /* Lookups */
1618
1619 /*
1620  * Fill a directory entry.
1621  *
1622  * If possible create the dcache entry and derive our inode number and
1623  * file type from dcache entry.
1624  *
1625  * Since all of the proc inode numbers are dynamically generated, the inode
1626  * numbers do not exist until the inode is cache.  This means creating the
1627  * the dcache entry in readdir is necessary to keep the inode numbers
1628  * reported by readdir in sync with the inode numbers reported
1629  * by stat.
1630  */
1631 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1632         const char *name, int len,
1633         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1634 {
1635         struct dentry *child, *dir = file->f_path.dentry;
1636         struct qstr qname = QSTR_INIT(name, len);
1637         struct inode *inode;
1638         unsigned type;
1639         ino_t ino;
1640
1641         child = d_hash_and_lookup(dir, &qname);
1642         if (!child) {
1643                 child = d_alloc(dir, &qname);
1644                 if (!child)
1645                         goto end_instantiate;
1646                 if (instantiate(dir->d_inode, child, task, ptr) < 0) {
1647                         dput(child);
1648                         goto end_instantiate;
1649                 }
1650         }
1651         inode = child->d_inode;
1652         ino = inode->i_ino;
1653         type = inode->i_mode >> 12;
1654         dput(child);
1655         return dir_emit(ctx, name, len, ino, type);
1656
1657 end_instantiate:
1658         return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1659 }
1660
1661 #ifdef CONFIG_CHECKPOINT_RESTORE
1662
1663 /*
1664  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1665  * which represent vma start and end addresses.
1666  */
1667 static int dname_to_vma_addr(struct dentry *dentry,
1668                              unsigned long *start, unsigned long *end)
1669 {
1670         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1671                 return -EINVAL;
1672
1673         return 0;
1674 }
1675
1676 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1677 {
1678         unsigned long vm_start, vm_end;
1679         bool exact_vma_exists = false;
1680         struct mm_struct *mm = NULL;
1681         struct task_struct *task;
1682         const struct cred *cred;
1683         struct inode *inode;
1684         int status = 0;
1685
1686         if (flags & LOOKUP_RCU)
1687                 return -ECHILD;
1688
1689         if (!capable(CAP_SYS_ADMIN)) {
1690                 status = -EPERM;
1691                 goto out_notask;
1692         }
1693
1694         inode = dentry->d_inode;
1695         task = get_proc_task(inode);
1696         if (!task)
1697                 goto out_notask;
1698
1699         mm = mm_access(task, PTRACE_MODE_READ);
1700         if (IS_ERR_OR_NULL(mm))
1701                 goto out;
1702
1703         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1704                 down_read(&mm->mmap_sem);
1705                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1706                 up_read(&mm->mmap_sem);
1707         }
1708
1709         mmput(mm);
1710
1711         if (exact_vma_exists) {
1712                 if (task_dumpable(task)) {
1713                         rcu_read_lock();
1714                         cred = __task_cred(task);
1715                         inode->i_uid = cred->euid;
1716                         inode->i_gid = cred->egid;
1717                         rcu_read_unlock();
1718                 } else {
1719                         inode->i_uid = GLOBAL_ROOT_UID;
1720                         inode->i_gid = GLOBAL_ROOT_GID;
1721                 }
1722                 security_task_to_inode(task, inode);
1723                 status = 1;
1724         }
1725
1726 out:
1727         put_task_struct(task);
1728
1729 out_notask:
1730         if (status <= 0)
1731                 d_drop(dentry);
1732
1733         return status;
1734 }
1735
1736 static const struct dentry_operations tid_map_files_dentry_operations = {
1737         .d_revalidate   = map_files_d_revalidate,
1738         .d_delete       = pid_delete_dentry,
1739 };
1740
1741 static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
1742 {
1743         unsigned long vm_start, vm_end;
1744         struct vm_area_struct *vma;
1745         struct task_struct *task;
1746         struct mm_struct *mm;
1747         int rc;
1748
1749         rc = -ENOENT;
1750         task = get_proc_task(dentry->d_inode);
1751         if (!task)
1752                 goto out;
1753
1754         mm = get_task_mm(task);
1755         put_task_struct(task);
1756         if (!mm)
1757                 goto out;
1758
1759         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1760         if (rc)
1761                 goto out_mmput;
1762
1763         rc = -ENOENT;
1764         down_read(&mm->mmap_sem);
1765         vma = find_exact_vma(mm, vm_start, vm_end);
1766         if (vma && vma->vm_file) {
1767                 *path = vma->vm_file->f_path;
1768                 path_get(path);
1769                 rc = 0;
1770         }
1771         up_read(&mm->mmap_sem);
1772
1773 out_mmput:
1774         mmput(mm);
1775 out:
1776         return rc;
1777 }
1778
1779 struct map_files_info {
1780         fmode_t         mode;
1781         unsigned long   len;
1782         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1783 };
1784
1785 static int
1786 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
1787                            struct task_struct *task, const void *ptr)
1788 {
1789         fmode_t mode = (fmode_t)(unsigned long)ptr;
1790         struct proc_inode *ei;
1791         struct inode *inode;
1792
1793         inode = proc_pid_make_inode(dir->i_sb, task);
1794         if (!inode)
1795                 return -ENOENT;
1796
1797         ei = PROC_I(inode);
1798         ei->op.proc_get_link = proc_map_files_get_link;
1799
1800         inode->i_op = &proc_pid_link_inode_operations;
1801         inode->i_size = 64;
1802         inode->i_mode = S_IFLNK;
1803
1804         if (mode & FMODE_READ)
1805                 inode->i_mode |= S_IRUSR;
1806         if (mode & FMODE_WRITE)
1807                 inode->i_mode |= S_IWUSR;
1808
1809         d_set_d_op(dentry, &tid_map_files_dentry_operations);
1810         d_add(dentry, inode);
1811
1812         return 0;
1813 }
1814
1815 static struct dentry *proc_map_files_lookup(struct inode *dir,
1816                 struct dentry *dentry, unsigned int flags)
1817 {
1818         unsigned long vm_start, vm_end;
1819         struct vm_area_struct *vma;
1820         struct task_struct *task;
1821         int result;
1822         struct mm_struct *mm;
1823
1824         result = -EPERM;
1825         if (!capable(CAP_SYS_ADMIN))
1826                 goto out;
1827
1828         result = -ENOENT;
1829         task = get_proc_task(dir);
1830         if (!task)
1831                 goto out;
1832
1833         result = -EACCES;
1834         if (!ptrace_may_access(task, PTRACE_MODE_READ))
1835                 goto out_put_task;
1836
1837         result = -ENOENT;
1838         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
1839                 goto out_put_task;
1840
1841         mm = get_task_mm(task);
1842         if (!mm)
1843                 goto out_put_task;
1844
1845         down_read(&mm->mmap_sem);
1846         vma = find_exact_vma(mm, vm_start, vm_end);
1847         if (!vma)
1848                 goto out_no_vma;
1849
1850         if (vma->vm_file)
1851                 result = proc_map_files_instantiate(dir, dentry, task,
1852                                 (void *)(unsigned long)vma->vm_file->f_mode);
1853
1854 out_no_vma:
1855         up_read(&mm->mmap_sem);
1856         mmput(mm);
1857 out_put_task:
1858         put_task_struct(task);
1859 out:
1860         return ERR_PTR(result);
1861 }
1862
1863 static const struct inode_operations proc_map_files_inode_operations = {
1864         .lookup         = proc_map_files_lookup,
1865         .permission     = proc_fd_permission,
1866         .setattr        = proc_setattr,
1867 };
1868
1869 static int
1870 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
1871 {
1872         struct vm_area_struct *vma;
1873         struct task_struct *task;
1874         struct mm_struct *mm;
1875         unsigned long nr_files, pos, i;
1876         struct flex_array *fa = NULL;
1877         struct map_files_info info;
1878         struct map_files_info *p;
1879         int ret;
1880
1881         ret = -EPERM;
1882         if (!capable(CAP_SYS_ADMIN))
1883                 goto out;
1884
1885         ret = -ENOENT;
1886         task = get_proc_task(file_inode(file));
1887         if (!task)
1888                 goto out;
1889
1890         ret = -EACCES;
1891         if (!ptrace_may_access(task, PTRACE_MODE_READ))
1892                 goto out_put_task;
1893
1894         ret = 0;
1895         if (!dir_emit_dots(file, ctx))
1896                 goto out_put_task;
1897
1898         mm = get_task_mm(task);
1899         if (!mm)
1900                 goto out_put_task;
1901         down_read(&mm->mmap_sem);
1902
1903         nr_files = 0;
1904
1905         /*
1906          * We need two passes here:
1907          *
1908          *  1) Collect vmas of mapped files with mmap_sem taken
1909          *  2) Release mmap_sem and instantiate entries
1910          *
1911          * otherwise we get lockdep complained, since filldir()
1912          * routine might require mmap_sem taken in might_fault().
1913          */
1914
1915         for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
1916                 if (vma->vm_file && ++pos > ctx->pos)
1917                         nr_files++;
1918         }
1919
1920         if (nr_files) {
1921                 fa = flex_array_alloc(sizeof(info), nr_files,
1922                                         GFP_KERNEL);
1923                 if (!fa || flex_array_prealloc(fa, 0, nr_files,
1924                                                 GFP_KERNEL)) {
1925                         ret = -ENOMEM;
1926                         if (fa)
1927                                 flex_array_free(fa);
1928                         up_read(&mm->mmap_sem);
1929                         mmput(mm);
1930                         goto out_put_task;
1931                 }
1932                 for (i = 0, vma = mm->mmap, pos = 2; vma;
1933                                 vma = vma->vm_next) {
1934                         if (!vma->vm_file)
1935                                 continue;
1936                         if (++pos <= ctx->pos)
1937                                 continue;
1938
1939                         info.mode = vma->vm_file->f_mode;
1940                         info.len = snprintf(info.name,
1941                                         sizeof(info.name), "%lx-%lx",
1942                                         vma->vm_start, vma->vm_end);
1943                         if (flex_array_put(fa, i++, &info, GFP_KERNEL))
1944                                 BUG();
1945                 }
1946         }
1947         up_read(&mm->mmap_sem);
1948
1949         for (i = 0; i < nr_files; i++) {
1950                 p = flex_array_get(fa, i);
1951                 if (!proc_fill_cache(file, ctx,
1952                                       p->name, p->len,
1953                                       proc_map_files_instantiate,
1954                                       task,
1955                                       (void *)(unsigned long)p->mode))
1956                         break;
1957                 ctx->pos++;
1958         }
1959         if (fa)
1960                 flex_array_free(fa);
1961         mmput(mm);
1962
1963 out_put_task:
1964         put_task_struct(task);
1965 out:
1966         return ret;
1967 }
1968
1969 static const struct file_operations proc_map_files_operations = {
1970         .read           = generic_read_dir,
1971         .iterate        = proc_map_files_readdir,
1972         .llseek         = default_llseek,
1973 };
1974
1975 struct timers_private {
1976         struct pid *pid;
1977         struct task_struct *task;
1978         struct sighand_struct *sighand;
1979         struct pid_namespace *ns;
1980         unsigned long flags;
1981 };
1982
1983 static void *timers_start(struct seq_file *m, loff_t *pos)
1984 {
1985         struct timers_private *tp = m->private;
1986
1987         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
1988         if (!tp->task)
1989                 return ERR_PTR(-ESRCH);
1990
1991         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
1992         if (!tp->sighand)
1993                 return ERR_PTR(-ESRCH);
1994
1995         return seq_list_start(&tp->task->signal->posix_timers, *pos);
1996 }
1997
1998 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
1999 {
2000         struct timers_private *tp = m->private;
2001         return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2002 }
2003
2004 static void timers_stop(struct seq_file *m, void *v)
2005 {
2006         struct timers_private *tp = m->private;
2007
2008         if (tp->sighand) {
2009                 unlock_task_sighand(tp->task, &tp->flags);
2010                 tp->sighand = NULL;
2011         }
2012
2013         if (tp->task) {
2014                 put_task_struct(tp->task);
2015                 tp->task = NULL;
2016         }
2017 }
2018
2019 static int show_timer(struct seq_file *m, void *v)
2020 {
2021         struct k_itimer *timer;
2022         struct timers_private *tp = m->private;
2023         int notify;
2024         static const char * const nstr[] = {
2025                 [SIGEV_SIGNAL] = "signal",
2026                 [SIGEV_NONE] = "none",
2027                 [SIGEV_THREAD] = "thread",
2028         };
2029
2030         timer = list_entry((struct list_head *)v, struct k_itimer, list);
2031         notify = timer->it_sigev_notify;
2032
2033         seq_printf(m, "ID: %d\n", timer->it_id);
2034         seq_printf(m, "signal: %d/%p\n", timer->sigq->info.si_signo,
2035                         timer->sigq->info.si_value.sival_ptr);
2036         seq_printf(m, "notify: %s/%s.%d\n",
2037                 nstr[notify & ~SIGEV_THREAD_ID],
2038                 (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2039                 pid_nr_ns(timer->it_pid, tp->ns));
2040         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2041
2042         return 0;
2043 }
2044
2045 static const struct seq_operations proc_timers_seq_ops = {
2046         .start  = timers_start,
2047         .next   = timers_next,
2048         .stop   = timers_stop,
2049         .show   = show_timer,
2050 };
2051
2052 static int proc_timers_open(struct inode *inode, struct file *file)
2053 {
2054         struct timers_private *tp;
2055
2056         tp = __seq_open_private(file, &proc_timers_seq_ops,
2057                         sizeof(struct timers_private));
2058         if (!tp)
2059                 return -ENOMEM;
2060
2061         tp->pid = proc_pid(inode);
2062         tp->ns = inode->i_sb->s_fs_info;
2063         return 0;
2064 }
2065
2066 static const struct file_operations proc_timers_operations = {
2067         .open           = proc_timers_open,
2068         .read           = seq_read,
2069         .llseek         = seq_lseek,
2070         .release        = seq_release_private,
2071 };
2072 #endif /* CONFIG_CHECKPOINT_RESTORE */
2073
2074 static int proc_pident_instantiate(struct inode *dir,
2075         struct dentry *dentry, struct task_struct *task, const void *ptr)
2076 {
2077         const struct pid_entry *p = ptr;
2078         struct inode *inode;
2079         struct proc_inode *ei;
2080
2081         inode = proc_pid_make_inode(dir->i_sb, task);
2082         if (!inode)
2083                 goto out;
2084
2085         ei = PROC_I(inode);
2086         inode->i_mode = p->mode;
2087         if (S_ISDIR(inode->i_mode))
2088                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2089         if (p->iop)
2090                 inode->i_op = p->iop;
2091         if (p->fop)
2092                 inode->i_fop = p->fop;
2093         ei->op = p->op;
2094         d_set_d_op(dentry, &pid_dentry_operations);
2095         d_add(dentry, inode);
2096         /* Close the race of the process dying before we return the dentry */
2097         if (pid_revalidate(dentry, 0))
2098                 return 0;
2099 out:
2100         return -ENOENT;
2101 }
2102
2103 static struct dentry *proc_pident_lookup(struct inode *dir,
2104                                          struct dentry *dentry,
2105                                          const struct pid_entry *ents,
2106                                          unsigned int nents)
2107 {
2108         int error;
2109         struct task_struct *task = get_proc_task(dir);
2110         const struct pid_entry *p, *last;
2111
2112         error = -ENOENT;
2113
2114         if (!task)
2115                 goto out_no_task;
2116
2117         /*
2118          * Yes, it does not scale. And it should not. Don't add
2119          * new entries into /proc/<tgid>/ without very good reasons.
2120          */
2121         last = &ents[nents - 1];
2122         for (p = ents; p <= last; p++) {
2123                 if (p->len != dentry->d_name.len)
2124                         continue;
2125                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2126                         break;
2127         }
2128         if (p > last)
2129                 goto out;
2130
2131         error = proc_pident_instantiate(dir, dentry, task, p);
2132 out:
2133         put_task_struct(task);
2134 out_no_task:
2135         return ERR_PTR(error);
2136 }
2137
2138 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2139                 const struct pid_entry *ents, unsigned int nents)
2140 {
2141         struct task_struct *task = get_proc_task(file_inode(file));
2142         const struct pid_entry *p;
2143
2144         if (!task)
2145                 return -ENOENT;
2146
2147         if (!dir_emit_dots(file, ctx))
2148                 goto out;
2149
2150         if (ctx->pos >= nents + 2)
2151                 goto out;
2152
2153         for (p = ents + (ctx->pos - 2); p <= ents + nents - 1; p++) {
2154                 if (!proc_fill_cache(file, ctx, p->name, p->len,
2155                                 proc_pident_instantiate, task, p))
2156                         break;
2157                 ctx->pos++;
2158         }
2159 out:
2160         put_task_struct(task);
2161         return 0;
2162 }
2163
2164 #ifdef CONFIG_SECURITY
2165 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2166                                   size_t count, loff_t *ppos)
2167 {
2168         struct inode * inode = file_inode(file);
2169         char *p = NULL;
2170         ssize_t length;
2171         struct task_struct *task = get_proc_task(inode);
2172
2173         if (!task)
2174                 return -ESRCH;
2175
2176         length = security_getprocattr(task,
2177                                       (char*)file->f_path.dentry->d_name.name,
2178                                       &p);
2179         put_task_struct(task);
2180         if (length > 0)
2181                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2182         kfree(p);
2183         return length;
2184 }
2185
2186 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2187                                    size_t count, loff_t *ppos)
2188 {
2189         struct inode * inode = file_inode(file);
2190         char *page;
2191         ssize_t length;
2192         struct task_struct *task = get_proc_task(inode);
2193
2194         length = -ESRCH;
2195         if (!task)
2196                 goto out_no_task;
2197         if (count > PAGE_SIZE)
2198                 count = PAGE_SIZE;
2199
2200         /* No partial writes. */
2201         length = -EINVAL;
2202         if (*ppos != 0)
2203                 goto out;
2204
2205         length = -ENOMEM;
2206         page = (char*)__get_free_page(GFP_TEMPORARY);
2207         if (!page)
2208                 goto out;
2209
2210         length = -EFAULT;
2211         if (copy_from_user(page, buf, count))
2212                 goto out_free;
2213
2214         /* Guard against adverse ptrace interaction */
2215         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2216         if (length < 0)
2217                 goto out_free;
2218
2219         length = security_setprocattr(task,
2220                                       (char*)file->f_path.dentry->d_name.name,
2221                                       (void*)page, count);
2222         mutex_unlock(&task->signal->cred_guard_mutex);
2223 out_free:
2224         free_page((unsigned long) page);
2225 out:
2226         put_task_struct(task);
2227 out_no_task:
2228         return length;
2229 }
2230
2231 static const struct file_operations proc_pid_attr_operations = {
2232         .read           = proc_pid_attr_read,
2233         .write          = proc_pid_attr_write,
2234         .llseek         = generic_file_llseek,
2235 };
2236
2237 static const struct pid_entry attr_dir_stuff[] = {
2238         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2239         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2240         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2241         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2242         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2243         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2244 };
2245
2246 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2247 {
2248         return proc_pident_readdir(file, ctx,
2249                                    attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2250 }
2251
2252 static const struct file_operations proc_attr_dir_operations = {
2253         .read           = generic_read_dir,
2254         .iterate        = proc_attr_dir_readdir,
2255         .llseek         = default_llseek,
2256 };
2257
2258 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2259                                 struct dentry *dentry, unsigned int flags)
2260 {
2261         return proc_pident_lookup(dir, dentry,
2262                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2263 }
2264
2265 static const struct inode_operations proc_attr_dir_inode_operations = {
2266         .lookup         = proc_attr_dir_lookup,
2267         .getattr        = pid_getattr,
2268         .setattr        = proc_setattr,
2269 };
2270
2271 #endif
2272
2273 #ifdef CONFIG_ELF_CORE
2274 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2275                                          size_t count, loff_t *ppos)
2276 {
2277         struct task_struct *task = get_proc_task(file_inode(file));
2278         struct mm_struct *mm;
2279         char buffer[PROC_NUMBUF];
2280         size_t len;
2281         int ret;
2282
2283         if (!task)
2284                 return -ESRCH;
2285
2286         ret = 0;
2287         mm = get_task_mm(task);
2288         if (mm) {
2289                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2290                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2291                                 MMF_DUMP_FILTER_SHIFT));
2292                 mmput(mm);
2293                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2294         }
2295
2296         put_task_struct(task);
2297
2298         return ret;
2299 }
2300
2301 static ssize_t proc_coredump_filter_write(struct file *file,
2302                                           const char __user *buf,
2303                                           size_t count,
2304                                           loff_t *ppos)
2305 {
2306         struct task_struct *task;
2307         struct mm_struct *mm;
2308         char buffer[PROC_NUMBUF], *end;
2309         unsigned int val;
2310         int ret;
2311         int i;
2312         unsigned long mask;
2313
2314         ret = -EFAULT;
2315         memset(buffer, 0, sizeof(buffer));
2316         if (count > sizeof(buffer) - 1)
2317                 count = sizeof(buffer) - 1;
2318         if (copy_from_user(buffer, buf, count))
2319                 goto out_no_task;
2320
2321         ret = -EINVAL;
2322         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2323         if (*end == '\n')
2324                 end++;
2325         if (end - buffer == 0)
2326                 goto out_no_task;
2327
2328         ret = -ESRCH;
2329         task = get_proc_task(file_inode(file));
2330         if (!task)
2331                 goto out_no_task;
2332
2333         ret = end - buffer;
2334         mm = get_task_mm(task);
2335         if (!mm)
2336                 goto out_no_mm;
2337
2338         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2339                 if (val & mask)
2340                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2341                 else
2342                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2343         }
2344
2345         mmput(mm);
2346  out_no_mm:
2347         put_task_struct(task);
2348  out_no_task:
2349         return ret;
2350 }
2351
2352 static const struct file_operations proc_coredump_filter_operations = {
2353         .read           = proc_coredump_filter_read,
2354         .write          = proc_coredump_filter_write,
2355         .llseek         = generic_file_llseek,
2356 };
2357 #endif
2358
2359 #ifdef CONFIG_TASK_IO_ACCOUNTING
2360 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2361 {
2362         struct task_io_accounting acct = task->ioac;
2363         unsigned long flags;
2364         int result;
2365
2366         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2367         if (result)
2368                 return result;
2369
2370         if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2371                 result = -EACCES;
2372                 goto out_unlock;
2373         }
2374
2375         if (whole && lock_task_sighand(task, &flags)) {
2376                 struct task_struct *t = task;
2377
2378                 task_io_accounting_add(&acct, &task->signal->ioac);
2379                 while_each_thread(task, t)
2380                         task_io_accounting_add(&acct, &t->ioac);
2381
2382                 unlock_task_sighand(task, &flags);
2383         }
2384         result = seq_printf(m,
2385                         "rchar: %llu\n"
2386                         "wchar: %llu\n"
2387                         "syscr: %llu\n"
2388                         "syscw: %llu\n"
2389                         "read_bytes: %llu\n"
2390                         "write_bytes: %llu\n"
2391                         "cancelled_write_bytes: %llu\n",
2392                         (unsigned long long)acct.rchar,
2393                         (unsigned long long)acct.wchar,
2394                         (unsigned long long)acct.syscr,
2395                         (unsigned long long)acct.syscw,
2396                         (unsigned long long)acct.read_bytes,
2397                         (unsigned long long)acct.write_bytes,
2398                         (unsigned long long)acct.cancelled_write_bytes);
2399 out_unlock:
2400         mutex_unlock(&task->signal->cred_guard_mutex);
2401         return result;
2402 }
2403
2404 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2405                                   struct pid *pid, struct task_struct *task)
2406 {
2407         return do_io_accounting(task, m, 0);
2408 }
2409
2410 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2411                                    struct pid *pid, struct task_struct *task)
2412 {
2413         return do_io_accounting(task, m, 1);
2414 }
2415 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2416
2417 #ifdef CONFIG_USER_NS
2418 static int proc_id_map_open(struct inode *inode, struct file *file,
2419         const struct seq_operations *seq_ops)
2420 {
2421         struct user_namespace *ns = NULL;
2422         struct task_struct *task;
2423         struct seq_file *seq;
2424         int ret = -EINVAL;
2425
2426         task = get_proc_task(inode);
2427         if (task) {
2428                 rcu_read_lock();
2429                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2430                 rcu_read_unlock();
2431                 put_task_struct(task);
2432         }
2433         if (!ns)
2434                 goto err;
2435
2436         ret = seq_open(file, seq_ops);
2437         if (ret)
2438                 goto err_put_ns;
2439
2440         seq = file->private_data;
2441         seq->private = ns;
2442
2443         return 0;
2444 err_put_ns:
2445         put_user_ns(ns);
2446 err:
2447         return ret;
2448 }
2449
2450 static int proc_id_map_release(struct inode *inode, struct file *file)
2451 {
2452         struct seq_file *seq = file->private_data;
2453         struct user_namespace *ns = seq->private;
2454         put_user_ns(ns);
2455         return seq_release(inode, file);
2456 }
2457
2458 static int proc_uid_map_open(struct inode *inode, struct file *file)
2459 {
2460         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2461 }
2462
2463 static int proc_gid_map_open(struct inode *inode, struct file *file)
2464 {
2465         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2466 }
2467
2468 static int proc_projid_map_open(struct inode *inode, struct file *file)
2469 {
2470         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2471 }
2472
2473 static const struct file_operations proc_uid_map_operations = {
2474         .open           = proc_uid_map_open,
2475         .write          = proc_uid_map_write,
2476         .read           = seq_read,
2477         .llseek         = seq_lseek,
2478         .release        = proc_id_map_release,
2479 };
2480
2481 static const struct file_operations proc_gid_map_operations = {
2482         .open           = proc_gid_map_open,
2483         .write          = proc_gid_map_write,
2484         .read           = seq_read,
2485         .llseek         = seq_lseek,
2486         .release        = proc_id_map_release,
2487 };
2488
2489 static const struct file_operations proc_projid_map_operations = {
2490         .open           = proc_projid_map_open,
2491         .write          = proc_projid_map_write,
2492         .read           = seq_read,
2493         .llseek         = seq_lseek,
2494         .release        = proc_id_map_release,
2495 };
2496 #endif /* CONFIG_USER_NS */
2497
2498 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2499                                 struct pid *pid, struct task_struct *task)
2500 {
2501         int err = lock_trace(task);
2502         if (!err) {
2503                 seq_printf(m, "%08x\n", task->personality);
2504                 unlock_trace(task);
2505         }
2506         return err;
2507 }
2508
2509 /*
2510  * Thread groups
2511  */
2512 static const struct file_operations proc_task_operations;
2513 static const struct inode_operations proc_task_inode_operations;
2514
2515 static const struct pid_entry tgid_base_stuff[] = {
2516         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2517         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2518 #ifdef CONFIG_CHECKPOINT_RESTORE
2519         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2520 #endif
2521         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2522         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2523 #ifdef CONFIG_NET
2524         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2525 #endif
2526         REG("environ",    S_IRUSR, proc_environ_operations),
2527         ONE("auxv",       S_IRUSR, proc_pid_auxv),
2528         ONE("status",     S_IRUGO, proc_pid_status),
2529         ONE("personality", S_IRUSR, proc_pid_personality),
2530         ONE("limits",     S_IRUGO, proc_pid_limits),
2531 #ifdef CONFIG_SCHED_DEBUG
2532         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2533 #endif
2534 #ifdef CONFIG_SCHED_AUTOGROUP
2535         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2536 #endif
2537         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2538 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2539         ONE("syscall",    S_IRUSR, proc_pid_syscall),
2540 #endif
2541         ONE("cmdline",    S_IRUGO, proc_pid_cmdline),
2542         ONE("stat",       S_IRUGO, proc_tgid_stat),
2543         ONE("statm",      S_IRUGO, proc_pid_statm),
2544         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2545 #ifdef CONFIG_NUMA
2546         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2547 #endif
2548         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2549         LNK("cwd",        proc_cwd_link),
2550         LNK("root",       proc_root_link),
2551         LNK("exe",        proc_exe_link),
2552         REG("mounts",     S_IRUGO, proc_mounts_operations),
2553         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2554         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2555 #ifdef CONFIG_PROC_PAGE_MONITOR
2556         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2557         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2558         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2559 #endif
2560 #ifdef CONFIG_SECURITY
2561         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2562 #endif
2563 #ifdef CONFIG_KALLSYMS
2564         ONE("wchan",      S_IRUGO, proc_pid_wchan),
2565 #endif
2566 #ifdef CONFIG_STACKTRACE
2567         ONE("stack",      S_IRUSR, proc_pid_stack),
2568 #endif
2569 #ifdef CONFIG_SCHEDSTATS
2570         ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
2571 #endif
2572 #ifdef CONFIG_LATENCYTOP
2573         REG("latency",  S_IRUGO, proc_lstats_operations),
2574 #endif
2575 #ifdef CONFIG_PROC_PID_CPUSET
2576         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2577 #endif
2578 #ifdef CONFIG_CGROUPS
2579         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2580 #endif
2581         ONE("oom_score",  S_IRUGO, proc_oom_score),
2582         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2583         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2584 #ifdef CONFIG_AUDITSYSCALL
2585         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2586         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2587 #endif
2588 #ifdef CONFIG_FAULT_INJECTION
2589         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2590 #endif
2591 #ifdef CONFIG_ELF_CORE
2592         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2593 #endif
2594 #ifdef CONFIG_TASK_IO_ACCOUNTING
2595         ONE("io",       S_IRUSR, proc_tgid_io_accounting),
2596 #endif
2597 #ifdef CONFIG_HARDWALL
2598         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
2599 #endif
2600 #ifdef CONFIG_USER_NS
2601         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2602         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2603         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2604 #endif
2605 #ifdef CONFIG_CHECKPOINT_RESTORE
2606         REG("timers",     S_IRUGO, proc_timers_operations),
2607 #endif
2608 };
2609
2610 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2611 {
2612         return proc_pident_readdir(file, ctx,
2613                                    tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2614 }
2615
2616 static const struct file_operations proc_tgid_base_operations = {
2617         .read           = generic_read_dir,
2618         .iterate        = proc_tgid_base_readdir,
2619         .llseek         = default_llseek,
2620 };
2621
2622 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2623 {
2624         return proc_pident_lookup(dir, dentry,
2625                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2626 }
2627
2628 static const struct inode_operations proc_tgid_base_inode_operations = {
2629         .lookup         = proc_tgid_base_lookup,
2630         .getattr        = pid_getattr,
2631         .setattr        = proc_setattr,
2632         .permission     = proc_pid_permission,
2633 };
2634
2635 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2636 {
2637         struct dentry *dentry, *leader, *dir;
2638         char buf[PROC_NUMBUF];
2639         struct qstr name;
2640
2641         name.name = buf;
2642         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2643         /* no ->d_hash() rejects on procfs */
2644         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2645         if (dentry) {
2646                 shrink_dcache_parent(dentry);
2647                 d_drop(dentry);
2648                 dput(dentry);
2649         }
2650
2651         name.name = buf;
2652         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2653         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2654         if (!leader)
2655                 goto out;
2656
2657         name.name = "task";
2658         name.len = strlen(name.name);
2659         dir = d_hash_and_lookup(leader, &name);
2660         if (!dir)
2661                 goto out_put_leader;
2662
2663         name.name = buf;
2664         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2665         dentry = d_hash_and_lookup(dir, &name);
2666         if (dentry) {
2667                 shrink_dcache_parent(dentry);
2668                 d_drop(dentry);
2669                 dput(dentry);
2670         }
2671
2672         dput(dir);
2673 out_put_leader:
2674         dput(leader);
2675 out:
2676         return;
2677 }
2678
2679 /**
2680  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2681  * @task: task that should be flushed.
2682  *
2683  * When flushing dentries from proc, one needs to flush them from global
2684  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2685  * in. This call is supposed to do all of this job.
2686  *
2687  * Looks in the dcache for
2688  * /proc/@pid
2689  * /proc/@tgid/task/@pid
2690  * if either directory is present flushes it and all of it'ts children
2691  * from the dcache.
2692  *
2693  * It is safe and reasonable to cache /proc entries for a task until
2694  * that task exits.  After that they just clog up the dcache with
2695  * useless entries, possibly causing useful dcache entries to be
2696  * flushed instead.  This routine is proved to flush those useless
2697  * dcache entries at process exit time.
2698  *
2699  * NOTE: This routine is just an optimization so it does not guarantee
2700  *       that no dcache entries will exist at process exit time it
2701  *       just makes it very unlikely that any will persist.
2702  */
2703
2704 void proc_flush_task(struct task_struct *task)
2705 {
2706         int i;
2707         struct pid *pid, *tgid;
2708         struct upid *upid;
2709
2710         pid = task_pid(task);
2711         tgid = task_tgid(task);
2712
2713         for (i = 0; i <= pid->level; i++) {
2714                 upid = &pid->numbers[i];
2715                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2716                                         tgid->numbers[i].nr);
2717         }
2718 }
2719
2720 static int proc_pid_instantiate(struct inode *dir,
2721                                    struct dentry * dentry,
2722                                    struct task_struct *task, const void *ptr)
2723 {
2724         struct inode *inode;
2725
2726         inode = proc_pid_make_inode(dir->i_sb, task);
2727         if (!inode)
2728                 goto out;
2729
2730         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2731         inode->i_op = &proc_tgid_base_inode_operations;
2732         inode->i_fop = &proc_tgid_base_operations;
2733         inode->i_flags|=S_IMMUTABLE;
2734
2735         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
2736                                                   ARRAY_SIZE(tgid_base_stuff)));
2737
2738         d_set_d_op(dentry, &pid_dentry_operations);
2739
2740         d_add(dentry, inode);
2741         /* Close the race of the process dying before we return the dentry */
2742         if (pid_revalidate(dentry, 0))
2743                 return 0;
2744 out:
2745         return -ENOENT;
2746 }
2747
2748 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
2749 {
2750         int result = -ENOENT;
2751         struct task_struct *task;
2752         unsigned tgid;
2753         struct pid_namespace *ns;
2754
2755         tgid = name_to_int(&dentry->d_name);
2756         if (tgid == ~0U)
2757                 goto out;
2758
2759         ns = dentry->d_sb->s_fs_info;
2760         rcu_read_lock();
2761         task = find_task_by_pid_ns(tgid, ns);
2762         if (task)
2763                 get_task_struct(task);
2764         rcu_read_unlock();
2765         if (!task)
2766                 goto out;
2767
2768         result = proc_pid_instantiate(dir, dentry, task, NULL);
2769         put_task_struct(task);
2770 out:
2771         return ERR_PTR(result);
2772 }
2773
2774 /*
2775  * Find the first task with tgid >= tgid
2776  *
2777  */
2778 struct tgid_iter {
2779         unsigned int tgid;
2780         struct task_struct *task;
2781 };
2782 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2783 {
2784         struct pid *pid;
2785
2786         if (iter.task)
2787                 put_task_struct(iter.task);
2788         rcu_read_lock();
2789 retry:
2790         iter.task = NULL;
2791         pid = find_ge_pid(iter.tgid, ns);
2792         if (pid) {
2793                 iter.tgid = pid_nr_ns(pid, ns);
2794                 iter.task = pid_task(pid, PIDTYPE_PID);
2795                 /* What we to know is if the pid we have find is the
2796                  * pid of a thread_group_leader.  Testing for task
2797                  * being a thread_group_leader is the obvious thing
2798                  * todo but there is a window when it fails, due to
2799                  * the pid transfer logic in de_thread.
2800                  *
2801                  * So we perform the straight forward test of seeing
2802                  * if the pid we have found is the pid of a thread
2803                  * group leader, and don't worry if the task we have
2804                  * found doesn't happen to be a thread group leader.
2805                  * As we don't care in the case of readdir.
2806                  */
2807                 if (!iter.task || !has_group_leader_pid(iter.task)) {
2808                         iter.tgid += 1;
2809                         goto retry;
2810                 }
2811                 get_task_struct(iter.task);
2812         }
2813         rcu_read_unlock();
2814         return iter;
2815 }
2816
2817 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
2818
2819 /* for the /proc/ directory itself, after non-process stuff has been done */
2820 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
2821 {
2822         struct tgid_iter iter;
2823         struct pid_namespace *ns = file->f_dentry->d_sb->s_fs_info;
2824         loff_t pos = ctx->pos;
2825
2826         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
2827                 return 0;
2828
2829         if (pos == TGID_OFFSET - 2) {
2830                 struct inode *inode = ns->proc_self->d_inode;
2831                 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
2832                         return 0;
2833                 ctx->pos = pos = pos + 1;
2834         }
2835         if (pos == TGID_OFFSET - 1) {
2836                 struct inode *inode = ns->proc_thread_self->d_inode;
2837                 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
2838                         return 0;
2839                 ctx->pos = pos = pos + 1;
2840         }
2841         iter.tgid = pos - TGID_OFFSET;
2842         iter.task = NULL;
2843         for (iter = next_tgid(ns, iter);
2844              iter.task;
2845              iter.tgid += 1, iter = next_tgid(ns, iter)) {
2846                 char name[PROC_NUMBUF];
2847                 int len;
2848                 if (!has_pid_permissions(ns, iter.task, 2))
2849                         continue;
2850
2851                 len = snprintf(name, sizeof(name), "%d", iter.tgid);
2852                 ctx->pos = iter.tgid + TGID_OFFSET;
2853                 if (!proc_fill_cache(file, ctx, name, len,
2854                                      proc_pid_instantiate, iter.task, NULL)) {
2855                         put_task_struct(iter.task);
2856                         return 0;
2857                 }
2858         }
2859         ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
2860         return 0;
2861 }
2862
2863 /*
2864  * Tasks
2865  */
2866 static const struct pid_entry tid_base_stuff[] = {
2867         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2868         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2869         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2870 #ifdef CONFIG_NET
2871         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2872 #endif
2873         REG("environ",   S_IRUSR, proc_environ_operations),
2874         ONE("auxv",      S_IRUSR, proc_pid_auxv),
2875         ONE("status",    S_IRUGO, proc_pid_status),
2876         ONE("personality", S_IRUSR, proc_pid_personality),
2877         ONE("limits",    S_IRUGO, proc_pid_limits),
2878 #ifdef CONFIG_SCHED_DEBUG
2879         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2880 #endif
2881         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2882 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2883         ONE("syscall",   S_IRUSR, proc_pid_syscall),
2884 #endif
2885         ONE("cmdline",   S_IRUGO, proc_pid_cmdline),
2886         ONE("stat",      S_IRUGO, proc_tid_stat),
2887         ONE("statm",     S_IRUGO, proc_pid_statm),
2888         REG("maps",      S_IRUGO, proc_tid_maps_operations),
2889 #ifdef CONFIG_CHECKPOINT_RESTORE
2890         REG("children",  S_IRUGO, proc_tid_children_operations),
2891 #endif
2892 #ifdef CONFIG_NUMA
2893         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
2894 #endif
2895         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
2896         LNK("cwd",       proc_cwd_link),
2897         LNK("root",      proc_root_link),
2898         LNK("exe",       proc_exe_link),
2899         REG("mounts",    S_IRUGO, proc_mounts_operations),
2900         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2901 #ifdef CONFIG_PROC_PAGE_MONITOR
2902         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2903         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
2904         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2905 #endif
2906 #ifdef CONFIG_SECURITY
2907         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2908 #endif
2909 #ifdef CONFIG_KALLSYMS
2910         ONE("wchan",     S_IRUGO, proc_pid_wchan),
2911 #endif
2912 #ifdef CONFIG_STACKTRACE
2913         ONE("stack",      S_IRUSR, proc_pid_stack),
2914 #endif
2915 #ifdef CONFIG_SCHEDSTATS
2916         ONE("schedstat", S_IRUGO, proc_pid_schedstat),
2917 #endif
2918 #ifdef CONFIG_LATENCYTOP
2919         REG("latency",  S_IRUGO, proc_lstats_operations),
2920 #endif
2921 #ifdef CONFIG_PROC_PID_CPUSET
2922         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
2923 #endif
2924 #ifdef CONFIG_CGROUPS
2925         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2926 #endif
2927         ONE("oom_score", S_IRUGO, proc_oom_score),
2928         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2929         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2930 #ifdef CONFIG_AUDITSYSCALL
2931         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
2932         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2933 #endif
2934 #ifdef CONFIG_FAULT_INJECTION
2935         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2936 #endif
2937 #ifdef CONFIG_TASK_IO_ACCOUNTING
2938         ONE("io",       S_IRUSR, proc_tid_io_accounting),
2939 #endif
2940 #ifdef CONFIG_HARDWALL
2941         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
2942 #endif
2943 #ifdef CONFIG_USER_NS
2944         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2945         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2946         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2947 #endif
2948 };
2949
2950 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
2951 {
2952         return proc_pident_readdir(file, ctx,
2953                                    tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
2954 }
2955
2956 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2957 {
2958         return proc_pident_lookup(dir, dentry,
2959                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
2960 }
2961
2962 static const struct file_operations proc_tid_base_operations = {
2963         .read           = generic_read_dir,
2964         .iterate        = proc_tid_base_readdir,
2965         .llseek         = default_llseek,
2966 };
2967
2968 static const struct inode_operations proc_tid_base_inode_operations = {
2969         .lookup         = proc_tid_base_lookup,
2970         .getattr        = pid_getattr,
2971         .setattr        = proc_setattr,
2972 };
2973
2974 static int proc_task_instantiate(struct inode *dir,
2975         struct dentry *dentry, struct task_struct *task, const void *ptr)
2976 {
2977         struct inode *inode;
2978         inode = proc_pid_make_inode(dir->i_sb, task);
2979
2980         if (!inode)
2981                 goto out;
2982         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2983         inode->i_op = &proc_tid_base_inode_operations;
2984         inode->i_fop = &proc_tid_base_operations;
2985         inode->i_flags|=S_IMMUTABLE;
2986
2987         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
2988                                                   ARRAY_SIZE(tid_base_stuff)));
2989
2990         d_set_d_op(dentry, &pid_dentry_operations);
2991
2992         d_add(dentry, inode);
2993         /* Close the race of the process dying before we return the dentry */
2994         if (pid_revalidate(dentry, 0))
2995                 return 0;
2996 out:
2997         return -ENOENT;
2998 }
2999
3000 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3001 {
3002         int result = -ENOENT;
3003         struct task_struct *task;
3004         struct task_struct *leader = get_proc_task(dir);
3005         unsigned tid;
3006         struct pid_namespace *ns;
3007
3008         if (!leader)
3009                 goto out_no_task;
3010
3011         tid = name_to_int(&dentry->d_name);
3012         if (tid == ~0U)
3013                 goto out;
3014
3015         ns = dentry->d_sb->s_fs_info;
3016         rcu_read_lock();
3017         task = find_task_by_pid_ns(tid, ns);
3018         if (task)
3019                 get_task_struct(task);
3020         rcu_read_unlock();
3021         if (!task)
3022                 goto out;
3023         if (!same_thread_group(leader, task))
3024                 goto out_drop_task;
3025
3026         result = proc_task_instantiate(dir, dentry, task, NULL);
3027 out_drop_task:
3028         put_task_struct(task);
3029 out:
3030         put_task_struct(leader);
3031 out_no_task:
3032         return ERR_PTR(result);
3033 }
3034
3035 /*
3036  * Find the first tid of a thread group to return to user space.
3037  *
3038  * Usually this is just the thread group leader, but if the users
3039  * buffer was too small or there was a seek into the middle of the
3040  * directory we have more work todo.
3041  *
3042  * In the case of a short read we start with find_task_by_pid.
3043  *
3044  * In the case of a seek we start with the leader and walk nr
3045  * threads past it.
3046  */
3047 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3048                                         struct pid_namespace *ns)
3049 {
3050         struct task_struct *pos, *task;
3051         unsigned long nr = f_pos;
3052
3053         if (nr != f_pos)        /* 32bit overflow? */
3054                 return NULL;
3055
3056         rcu_read_lock();
3057         task = pid_task(pid, PIDTYPE_PID);
3058         if (!task)
3059                 goto fail;
3060
3061         /* Attempt to start with the tid of a thread */
3062         if (tid && nr) {
3063                 pos = find_task_by_pid_ns(tid, ns);
3064                 if (pos && same_thread_group(pos, task))
3065                         goto found;
3066         }
3067
3068         /* If nr exceeds the number of threads there is nothing todo */
3069         if (nr >= get_nr_threads(task))
3070                 goto fail;
3071
3072         /* If we haven't found our starting place yet start
3073          * with the leader and walk nr threads forward.
3074          */
3075         pos = task = task->group_leader;
3076         do {
3077                 if (!nr--)
3078                         goto found;
3079         } while_each_thread(task, pos);
3080 fail:
3081         pos = NULL;
3082         goto out;
3083 found:
3084         get_task_struct(pos);
3085 out:
3086         rcu_read_unlock();
3087         return pos;
3088 }
3089
3090 /*
3091  * Find the next thread in the thread list.
3092  * Return NULL if there is an error or no next thread.
3093  *
3094  * The reference to the input task_struct is released.
3095  */
3096 static struct task_struct *next_tid(struct task_struct *start)
3097 {
3098         struct task_struct *pos = NULL;
3099         rcu_read_lock();
3100         if (pid_alive(start)) {
3101                 pos = next_thread(start);
3102                 if (thread_group_leader(pos))
3103                         pos = NULL;
3104                 else
3105                         get_task_struct(pos);
3106         }
3107         rcu_read_unlock();
3108         put_task_struct(start);
3109         return pos;
3110 }
3111
3112 /* for the /proc/TGID/task/ directories */
3113 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3114 {
3115         struct inode *inode = file_inode(file);
3116         struct task_struct *task;
3117         struct pid_namespace *ns;
3118         int tid;
3119
3120         if (proc_inode_is_dead(inode))
3121                 return -ENOENT;
3122
3123         if (!dir_emit_dots(file, ctx))
3124                 return 0;
3125
3126         /* f_version caches the tgid value that the last readdir call couldn't
3127          * return. lseek aka telldir automagically resets f_version to 0.
3128          */
3129         ns = file->f_dentry->d_sb->s_fs_info;
3130         tid = (int)file->f_version;
3131         file->f_version = 0;
3132         for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3133              task;
3134              task = next_tid(task), ctx->pos++) {
3135                 char name[PROC_NUMBUF];
3136                 int len;
3137                 tid = task_pid_nr_ns(task, ns);
3138                 len = snprintf(name, sizeof(name), "%d", tid);
3139                 if (!proc_fill_cache(file, ctx, name, len,
3140                                 proc_task_instantiate, task, NULL)) {
3141                         /* returning this tgid failed, save it as the first
3142                          * pid for the next readir call */
3143                         file->f_version = (u64)tid;
3144                         put_task_struct(task);
3145                         break;
3146                 }
3147         }
3148
3149         return 0;
3150 }
3151
3152 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3153 {
3154         struct inode *inode = dentry->d_inode;
3155         struct task_struct *p = get_proc_task(inode);
3156         generic_fillattr(inode, stat);
3157
3158         if (p) {
3159                 stat->nlink += get_nr_threads(p);
3160                 put_task_struct(p);
3161         }
3162
3163         return 0;
3164 }
3165
3166 static const struct inode_operations proc_task_inode_operations = {
3167         .lookup         = proc_task_lookup,
3168         .getattr        = proc_task_getattr,
3169         .setattr        = proc_setattr,
3170         .permission     = proc_pid_permission,
3171 };
3172
3173 static const struct file_operations proc_task_operations = {
3174         .read           = generic_read_dir,
3175         .iterate        = proc_task_readdir,
3176         .llseek         = default_llseek,
3177 };