fs/proc/base.c

   1 /*
   2  *  linux/fs/proc/base.c
   3  *
   4  *  Copyright (C) 1991, 1992 Linus Torvalds
   5  *
   6  *  proc base directory handling functions
   7  *
   8  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
   9  *  Instead of using magical inumbers to determine the kind of object
  10  *  we allocate and fill in-core inodes upon lookup. They don't even
  11  *  go into icache. We cache the reference to task_struct upon lookup too.
  12  *  Eventually it should become a filesystem in its own. We don't use the
  13  *  rest of procfs anymore.
  14  *
  15  *
  16  *  Changelog:
  17  *  17-Jan-2005
  18  *  Allan Bezerra
  19  *  Bruna Moreira <bruna.moreira@indt.org.br>
  20  *  Edjard Mota <edjard.mota@indt.org.br>
  21  *  Ilias Biris <ilias.biris@indt.org.br>
  22  *  Mauricio Lin <mauricio.lin@indt.org.br>
  23  *
  24  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  25  *
  26  *  A new process specific entry (smaps) included in /proc. It shows the
  27  *  size of rss for each memory area. The maps entry lacks information
  28  *  about physical memory size (rss) for each mapped file, i.e.,
  29  *  rss information for executables and library files.
  30  *  This additional information is useful for any tools that need to know
  31  *  about physical memory consumption for a process specific library.
  32  *
  33  *  Changelog:
  34  *  21-Feb-2005
  35  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  36  *  Pud inclusion in the page table walking.
  37  *
  38  *  ChangeLog:
  39  *  10-Mar-2005
  40  *  10LE Instituto Nokia de Tecnologia - INdT:
  41  *  A better way to walks through the page table as suggested by Hugh Dickins.
  42  *
  43  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
  44  *  Smaps information related to shared, private, clean and dirty pages.
  45  *
  46  *  Paul Mundt <paul.mundt@nokia.com>:
  47  *  Overall revision about smaps.
  48  */
  49
  50 #include <asm/uaccess.h>
  51
  52 #include <linux/errno.h>
  53 #include <linux/time.h>
  54 #include <linux/proc_fs.h>
  55 #include <linux/stat.h>
  56 #include <linux/task_io_accounting_ops.h>
  57 #include <linux/init.h>
  58 #include <linux/capability.h>
  59 #include <linux/file.h>
  60 #include <linux/fdtable.h>
  61 #include <linux/string.h>
  62 #include <linux/seq_file.h>
  63 #include <linux/namei.h>
  64 #include <linux/mnt_namespace.h>
  65 #include <linux/mm.h>
  66 #include <linux/swap.h>
  67 #include <linux/rcupdate.h>
  68 #include <linux/kallsyms.h>
  69 #include <linux/stacktrace.h>
  70 #include <linux/resource.h>
  71 #include <linux/module.h>
  72 #include <linux/mount.h>
  73 #include <linux/security.h>
  74 #include <linux/ptrace.h>
  75 #include <linux/tracehook.h>
  76 #include <linux/printk.h>
  77 #include <linux/cgroup.h>
  78 #include <linux/cpuset.h>
  79 #include <linux/audit.h>
  80 #include <linux/poll.h>
  81 #include <linux/nsproxy.h>
  82 #include <linux/oom.h>
  83 #include <linux/elf.h>
  84 #include <linux/pid_namespace.h>
  85 #include <linux/user_namespace.h>
  86 #include <linux/fs_struct.h>
  87 #include <linux/slab.h>
  88 #include <linux/flex_array.h>
  89 #include <linux/posix-timers.h>
  90 #ifdef CONFIG_HARDWALL
  91 #include <asm/hardwall.h>
  92 #endif
  93 #include <trace/events/oom.h>
  94 #include "internal.h"
  95 #include "fd.h"
  96
  97 /* NOTE:
  98  *      Implementing inode permission operations in /proc is almost
  99  *      certainly an error.  Permission checks need to happen during
 100  *      each system call not at open time.  The reason is that most of
 101  *      what we wish to check for permissions in /proc varies at runtime.
 102  *
 103  *      The classic example of a problem is opening file descriptors
 104  *      in /proc for a task before it execs a suid executable.
 105  */
 106
 107 struct pid_entry {
 108         char *name;
 109         int len;
 110         umode_t mode;
 111         const struct inode_operations *iop;
 112         const struct file_operations *fop;
 113         union proc_op op;
 114 };
 115
 116 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 117         .name = (NAME),                                 \
 118         .len  = sizeof(NAME) - 1,                       \
 119         .mode = MODE,                                   \
 120         .iop  = IOP,                                    \
 121         .fop  = FOP,                                    \
 122         .op   = OP,                                     \
 123 }
 124
 125 #define DIR(NAME, MODE, iops, fops)     \
 126         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 127 #define LNK(NAME, get_link)                                     \
 128         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 129                 &proc_pid_link_inode_operations, NULL,          \
 130                 { .proc_get_link = get_link } )
 131 #define REG(NAME, MODE, fops)                           \
 132         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 133 #define INF(NAME, MODE, read)                           \
 134         NOD(NAME, (S_IFREG|(MODE)),                     \
 135                 NULL, &proc_info_file_operations,       \
 136                 { .proc_read = read } )
 137 #define ONE(NAME, MODE, show)                           \
 138         NOD(NAME, (S_IFREG|(MODE)),                     \
 139                 NULL, &proc_single_file_operations,     \
 140                 { .proc_show = show } )
 141
 142 /*
 143  * Count the number of hardlinks for the pid_entry table, excluding the .
 144  * and .. links.
 145  */
 146 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
 147         unsigned int n)
 148 {
 149         unsigned int i;
 150         unsigned int count;
 151
 152         count = 0;
 153         for (i = 0; i < n; ++i) {
 154                 if (S_ISDIR(entries[i].mode))
 155                         ++count;
 156         }
 157
 158         return count;
 159 }
 160
 161 static int get_task_root(struct task_struct *task, struct path *root)
 162 {
 163         int result = -ENOENT;
 164
 165         task_lock(task);
 166         if (task->fs) {
 167                 get_fs_root(task->fs, root);
 168                 result = 0;
 169         }
 170         task_unlock(task);
 171         return result;
 172 }
 173
 174 static int proc_cwd_link(struct dentry *dentry, struct path *path)
 175 {
 176         struct task_struct *task = get_proc_task(dentry->d_inode);
 177         int result = -ENOENT;
 178
 179         if (task) {
 180                 task_lock(task);
 181                 if (task->fs) {
 182                         get_fs_pwd(task->fs, path);
 183                         result = 0;
 184                 }
 185                 task_unlock(task);
 186                 put_task_struct(task);
 187         }
 188         return result;
 189 }
 190
 191 static int proc_root_link(struct dentry *dentry, struct path *path)
 192 {
 193         struct task_struct *task = get_proc_task(dentry->d_inode);
 194         int result = -ENOENT;
 195
 196         if (task) {
 197                 result = get_task_root(task, path);
 198                 put_task_struct(task);
 199         }
 200         return result;
 201 }
 202
 203 static int proc_pid_cmdline(struct task_struct *task, char *buffer)
 204 {
 205         return get_cmdline(task, buffer, PAGE_SIZE);
 206 }
 207
 208 static int proc_pid_auxv(struct task_struct *task, char *buffer)
 209 {
 210         struct mm_struct *mm = mm_access(task, PTRACE_MODE_READ);
 211         int res = PTR_ERR(mm);
 212         if (mm && !IS_ERR(mm)) {
 213                 unsigned int nwords = 0;
 214                 do {
 215                         nwords += 2;
 216                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 217                 res = nwords * sizeof(mm->saved_auxv[0]);
 218                 if (res > PAGE_SIZE)
 219                         res = PAGE_SIZE;
 220                 memcpy(buffer, mm->saved_auxv, res);
 221                 mmput(mm);
 222         }
 223         return res;
 224 }
 225
 226
 227 #ifdef CONFIG_KALLSYMS
 228 /*
 229  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 230  * Returns the resolved symbol.  If that fails, simply return the address.
 231  */
 232 static int proc_pid_wchan(struct task_struct *task, char *buffer)
 233 {
 234         unsigned long wchan;
 235         char symname[KSYM_NAME_LEN];
 236
 237         wchan = get_wchan(task);
 238
 239         if (lookup_symbol_name(wchan, symname) < 0)
 240                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
 241                         return 0;
 242                 else
 243                         return sprintf(buffer, "%lu", wchan);
 244         else
 245                 return sprintf(buffer, "%s", symname);
 246 }
 247 #endif /* CONFIG_KALLSYMS */
 248
 249 static int lock_trace(struct task_struct *task)
 250 {
 251         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
 252         if (err)
 253                 return err;
 254         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH)) {
 255                 mutex_unlock(&task->signal->cred_guard_mutex);
 256                 return -EPERM;
 257         }
 258         return 0;
 259 }
 260
 261 static void unlock_trace(struct task_struct *task)
 262 {
 263         mutex_unlock(&task->signal->cred_guard_mutex);
 264 }
 265
 266 #ifdef CONFIG_STACKTRACE
 267
 268 #define MAX_STACK_TRACE_DEPTH   64
 269
 270 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 271                           struct pid *pid, struct task_struct *task)
 272 {
 273         struct stack_trace trace;
 274         unsigned long *entries;
 275         int err;
 276         int i;
 277
 278         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 279         if (!entries)
 280                 return -ENOMEM;
 281
 282         trace.nr_entries        = 0;
 283         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 284         trace.entries           = entries;
 285         trace.skip              = 0;
 286
 287         err = lock_trace(task);
 288         if (!err) {
 289                 save_stack_trace_tsk(task, &trace);
 290
 291                 for (i = 0; i < trace.nr_entries; i++) {
 292                         seq_printf(m, "[<%pK>] %pS\n",
 293                                    (void *)entries[i], (void *)entries[i]);
 294                 }
 295                 unlock_trace(task);
 296         }
 297         kfree(entries);
 298
 299         return err;
 300 }
 301 #endif
 302
 303 #ifdef CONFIG_SCHEDSTATS
 304 /*
 305  * Provides /proc/PID/schedstat
 306  */
 307 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
 308 {
 309         return sprintf(buffer, "%llu %llu %lu\n",
 310                         (unsigned long long)task->se.sum_exec_runtime,
 311                         (unsigned long long)task->sched_info.run_delay,
 312                         task->sched_info.pcount);
 313 }
 314 #endif
 315
 316 #ifdef CONFIG_LATENCYTOP
 317 static int lstats_show_proc(struct seq_file *m, void *v)
 318 {
 319         int i;
 320         struct inode *inode = m->private;
 321         struct task_struct *task = get_proc_task(inode);
 322
 323         if (!task)
 324                 return -ESRCH;
 325         seq_puts(m, "Latency Top version : v0.1\n");
 326         for (i = 0; i < 32; i++) {
 327                 struct latency_record *lr = &task->latency_record[i];
 328                 if (lr->backtrace[0]) {
 329                         int q;
 330                         seq_printf(m, "%i %li %li",
 331                                    lr->count, lr->time, lr->max);
 332                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 333                                 unsigned long bt = lr->backtrace[q];
 334                                 if (!bt)
 335                                         break;
 336                                 if (bt == ULONG_MAX)
 337                                         break;
 338                                 seq_printf(m, " %ps", (void *)bt);
 339                         }
 340                         seq_putc(m, '\n');
 341                 }
 342
 343         }
 344         put_task_struct(task);
 345         return 0;
 346 }
 347
 348 static int lstats_open(struct inode *inode, struct file *file)
 349 {
 350         return single_open(file, lstats_show_proc, inode);
 351 }
 352
 353 static ssize_t lstats_write(struct file *file, const char __user *buf,
 354                             size_t count, loff_t *offs)
 355 {
 356         struct task_struct *task = get_proc_task(file_inode(file));
 357
 358         if (!task)
 359                 return -ESRCH;
 360         clear_all_latency_tracing(task);
 361         put_task_struct(task);
 362
 363         return count;
 364 }
 365
 366 static const struct file_operations proc_lstats_operations = {
 367         .open           = lstats_open,
 368         .read           = seq_read,
 369         .write          = lstats_write,
 370         .llseek         = seq_lseek,
 371         .release        = single_release,
 372 };
 373
 374 #endif
 375
 376 #ifdef CONFIG_CGROUPS
 377 static int cgroup_open(struct inode *inode, struct file *file)
 378 {
 379         struct pid *pid = PROC_I(inode)->pid;
 380         return single_open(file, proc_cgroup_show, pid);
 381 }
 382
 383 static const struct file_operations proc_cgroup_operations = {
 384         .open           = cgroup_open,
 385         .read           = seq_read,
 386         .llseek         = seq_lseek,
 387         .release        = single_release,
 388 };
 389 #endif
 390
 391 #ifdef CONFIG_PROC_PID_CPUSET
 392
 393 static int cpuset_open(struct inode *inode, struct file *file)
 394 {
 395         struct pid *pid = PROC_I(inode)->pid;
 396         return single_open(file, proc_cpuset_show, pid);
 397 }
 398
 399 static const struct file_operations proc_cpuset_operations = {
 400         .open           = cpuset_open,
 401         .read           = seq_read,
 402         .llseek         = seq_lseek,
 403         .release        = single_release,
 404 };
 405 #endif
 406
 407 static int proc_oom_score(struct task_struct *task, char *buffer)
 408 {
 409         unsigned long totalpages = totalram_pages + total_swap_pages;
 410         unsigned long points = 0;
 411
 412         read_lock(&tasklist_lock);
 413         if (pid_alive(task))
 414                 points = oom_badness(task, NULL, NULL, totalpages) *
 415                                                 1000 / totalpages;
 416         read_unlock(&tasklist_lock);
 417         return sprintf(buffer, "%lu\n", points);
 418 }
 419
 420 struct limit_names {
 421         char *name;
 422         char *unit;
 423 };
 424
 425 static const struct limit_names lnames[RLIM_NLIMITS] = {
 426         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 427         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 428         [RLIMIT_DATA] = {"Max data size", "bytes"},
 429         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 430         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 431         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 432         [RLIMIT_NPROC] = {"Max processes", "processes"},
 433         [RLIMIT_NOFILE] = {"Max open files", "files"},
 434         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 435         [RLIMIT_AS] = {"Max address space", "bytes"},
 436         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 437         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 438         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 439         [RLIMIT_NICE] = {"Max nice priority", NULL},
 440         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 441         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 442 };
 443
 444 /* Display limits for a process */
 445 static int proc_pid_limits(struct task_struct *task, char *buffer)
 446 {
 447         unsigned int i;
 448         int count = 0;
 449         unsigned long flags;
 450         char *bufptr = buffer;
 451
 452         struct rlimit rlim[RLIM_NLIMITS];
 453
 454         if (!lock_task_sighand(task, &flags))
 455                 return 0;
 456         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 457         unlock_task_sighand(task, &flags);
 458
 459         /*
 460          * print the file header
 461          */
 462         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
 463                         "Limit", "Soft Limit", "Hard Limit", "Units");
 464
 465         for (i = 0; i < RLIM_NLIMITS; i++) {
 466                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 467                         count += sprintf(&bufptr[count], "%-25s %-20s ",
 468                                          lnames[i].name, "unlimited");
 469                 else
 470                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
 471                                          lnames[i].name, rlim[i].rlim_cur);
 472
 473                 if (rlim[i].rlim_max == RLIM_INFINITY)
 474                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
 475                 else
 476                         count += sprintf(&bufptr[count], "%-20lu ",
 477                                          rlim[i].rlim_max);
 478
 479                 if (lnames[i].unit)
 480                         count += sprintf(&bufptr[count], "%-10s\n",
 481                                          lnames[i].unit);
 482                 else
 483                         count += sprintf(&bufptr[count], "\n");
 484         }
 485
 486         return count;
 487 }
 488
 489 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 490 static int proc_pid_syscall(struct task_struct *task, char *buffer)
 491 {
 492         long nr;
 493         unsigned long args[6], sp, pc;
 494         int res = lock_trace(task);
 495         if (res)
 496                 return res;
 497
 498         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 499                 res = sprintf(buffer, "running\n");
 500         else if (nr < 0)
 501                 res = sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 502         else
 503                 res = sprintf(buffer,
 504                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 505                        nr,
 506                        args[0], args[1], args[2], args[3], args[4], args[5],
 507                        sp, pc);
 508         unlock_trace(task);
 509         return res;
 510 }
 511 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 512
 513 /************************************************************************/
 514 /*                       Here the fs part begins                        */
 515 /************************************************************************/
 516
 517 /* permission checks */
 518 static int proc_fd_access_allowed(struct inode *inode)
 519 {
 520         struct task_struct *task;
 521         int allowed = 0;
 522         /* Allow access to a task's file descriptors if it is us or we
 523          * may use ptrace attach to the process and find out that
 524          * information.
 525          */
 526         task = get_proc_task(inode);
 527         if (task) {
 528                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
 529                 put_task_struct(task);
 530         }
 531         return allowed;
 532 }
 533
 534 int proc_setattr(struct dentry *dentry, struct iattr *attr)
 535 {
 536         int error;
 537         struct inode *inode = dentry->d_inode;
 538
 539         if (attr->ia_valid & ATTR_MODE)
 540                 return -EPERM;
 541
 542         error = inode_change_ok(inode, attr);
 543         if (error)
 544                 return error;
 545
 546         setattr_copy(inode, attr);
 547         mark_inode_dirty(inode);
 548         return 0;
 549 }
 550
 551 /*
 552  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
 553  * or euid/egid (for hide_pid_min=2)?
 554  */
 555 static bool has_pid_permissions(struct pid_namespace *pid,
 556                                  struct task_struct *task,
 557                                  int hide_pid_min)
 558 {
 559         if (pid->hide_pid < hide_pid_min)
 560                 return true;
 561         if (in_group_p(pid->pid_gid))
 562                 return true;
 563         return ptrace_may_access(task, PTRACE_MODE_READ);
 564 }
 565
 566
 567 static int proc_pid_permission(struct inode *inode, int mask)
 568 {
 569         struct pid_namespace *pid = inode->i_sb->s_fs_info;
 570         struct task_struct *task;
 571         bool has_perms;
 572
 573         task = get_proc_task(inode);
 574         if (!task)
 575                 return -ESRCH;
 576         has_perms = has_pid_permissions(pid, task, 1);
 577         put_task_struct(task);
 578
 579         if (!has_perms) {
 580                 if (pid->hide_pid == 2) {
 581                         /*
 582                          * Let's make getdents(), stat(), and open()
 583                          * consistent with each other.  If a process
 584                          * may not stat() a file, it shouldn't be seen
 585                          * in procfs at all.
 586                          */
 587                         return -ENOENT;
 588                 }
 589
 590                 return -EPERM;
 591         }
 592         return generic_permission(inode, mask);
 593 }
 594
 595
 596
 597 static const struct inode_operations proc_def_inode_operations = {
 598         .setattr        = proc_setattr,
 599 };
 600
 601 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 602
 603 static ssize_t proc_info_read(struct file * file, char __user * buf,
 604                           size_t count, loff_t *ppos)
 605 {
 606         struct inode * inode = file_inode(file);
 607         unsigned long page;
 608         ssize_t length;
 609         struct task_struct *task = get_proc_task(inode);
 610
 611         length = -ESRCH;
 612         if (!task)
 613                 goto out_no_task;
 614
 615         if (count > PROC_BLOCK_SIZE)
 616                 count = PROC_BLOCK_SIZE;
 617
 618         length = -ENOMEM;
 619         if (!(page = __get_free_page(GFP_TEMPORARY)))
 620                 goto out;
 621
 622         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 623
 624         if (length >= 0)
 625                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 626         free_page(page);
 627 out:
 628         put_task_struct(task);
 629 out_no_task:
 630         return length;
 631 }
 632
 633 static const struct file_operations proc_info_file_operations = {
 634         .read           = proc_info_read,
 635         .llseek         = generic_file_llseek,
 636 };
 637
 638 static int proc_single_show(struct seq_file *m, void *v)
 639 {
 640         struct inode *inode = m->private;
 641         struct pid_namespace *ns;
 642         struct pid *pid;
 643         struct task_struct *task;
 644         int ret;
 645
 646         ns = inode->i_sb->s_fs_info;
 647         pid = proc_pid(inode);
 648         task = get_pid_task(pid, PIDTYPE_PID);
 649         if (!task)
 650                 return -ESRCH;
 651
 652         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 653
 654         put_task_struct(task);
 655         return ret;
 656 }
 657
 658 static int proc_single_open(struct inode *inode, struct file *filp)
 659 {
 660         return single_open(filp, proc_single_show, inode);
 661 }
 662
 663 static const struct file_operations proc_single_file_operations = {
 664         .open           = proc_single_open,
 665         .read           = seq_read,
 666         .llseek         = seq_lseek,
 667         .release        = single_release,
 668 };
 669
 670 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
 671 {
 672         struct task_struct *task = get_proc_task(file_inode(file));
 673         struct mm_struct *mm;
 674
 675         if (!task)
 676                 return -ESRCH;
 677
 678         mm = mm_access(task, mode);
 679         put_task_struct(task);
 680
 681         if (IS_ERR(mm))
 682                 return PTR_ERR(mm);
 683
 684         if (mm) {
 685                 /* ensure this mm_struct can't be freed */
 686                 atomic_inc(&mm->mm_count);
 687                 /* but do not pin its memory */
 688                 mmput(mm);
 689         }
 690
 691         file->private_data = mm;
 692
 693         return 0;
 694 }
 695
 696 static int mem_open(struct inode *inode, struct file *file)
 697 {
 698         int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
 699
 700         /* OK to pass negative loff_t, we can catch out-of-range */
 701         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 702
 703         return ret;
 704 }
 705
 706 static ssize_t mem_rw(struct file *file, char __user *buf,
 707                         size_t count, loff_t *ppos, int write)
 708 {
 709         struct mm_struct *mm = file->private_data;
 710         unsigned long addr = *ppos;
 711         ssize_t copied;
 712         char *page;
 713
 714         if (!mm)
 715                 return 0;
 716
 717         page = (char *)__get_free_page(GFP_TEMPORARY);
 718         if (!page)
 719                 return -ENOMEM;
 720
 721         copied = 0;
 722         if (!atomic_inc_not_zero(&mm->mm_users))
 723                 goto free;
 724
 725         while (count > 0) {
 726                 int this_len = min_t(int, count, PAGE_SIZE);
 727
 728                 if (write && copy_from_user(page, buf, this_len)) {
 729                         copied = -EFAULT;
 730                         break;
 731                 }
 732
 733                 this_len = access_remote_vm(mm, addr, page, this_len, write);
 734                 if (!this_len) {
 735                         if (!copied)
 736                                 copied = -EIO;
 737                         break;
 738                 }
 739
 740                 if (!write && copy_to_user(buf, page, this_len)) {
 741                         copied = -EFAULT;
 742                         break;
 743                 }
 744
 745                 buf += this_len;
 746                 addr += this_len;
 747                 copied += this_len;
 748                 count -= this_len;
 749         }
 750         *ppos = addr;
 751
 752         mmput(mm);
 753 free:
 754         free_page((unsigned long) page);
 755         return copied;
 756 }
 757
 758 static ssize_t mem_read(struct file *file, char __user *buf,
 759                         size_t count, loff_t *ppos)
 760 {
 761         return mem_rw(file, buf, count, ppos, 0);
 762 }
 763
 764 static ssize_t mem_write(struct file *file, const char __user *buf,
 765                          size_t count, loff_t *ppos)
 766 {
 767         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 768 }
 769
 770 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 771 {
 772         switch (orig) {
 773         case 0:
 774                 file->f_pos = offset;
 775                 break;
 776         case 1:
 777                 file->f_pos += offset;
 778                 break;
 779         default:
 780                 return -EINVAL;
 781         }
 782         force_successful_syscall_return();
 783         return file->f_pos;
 784 }
 785
 786 static int mem_release(struct inode *inode, struct file *file)
 787 {
 788         struct mm_struct *mm = file->private_data;
 789         if (mm)
 790                 mmdrop(mm);
 791         return 0;
 792 }
 793
 794 static const struct file_operations proc_mem_operations = {
 795         .llseek         = mem_lseek,
 796         .read           = mem_read,
 797         .write          = mem_write,
 798         .open           = mem_open,
 799         .release        = mem_release,
 800 };
 801
 802 static int environ_open(struct inode *inode, struct file *file)
 803 {
 804         return __mem_open(inode, file, PTRACE_MODE_READ);
 805 }
 806
 807 static ssize_t environ_read(struct file *file, char __user *buf,
 808                         size_t count, loff_t *ppos)
 809 {
 810         char *page;
 811         unsigned long src = *ppos;
 812         int ret = 0;
 813         struct mm_struct *mm = file->private_data;
 814
 815         if (!mm)
 816                 return 0;
 817
 818         page = (char *)__get_free_page(GFP_TEMPORARY);
 819         if (!page)
 820                 return -ENOMEM;
 821
 822         ret = 0;
 823         if (!atomic_inc_not_zero(&mm->mm_users))
 824                 goto free;
 825         while (count > 0) {
 826                 size_t this_len, max_len;
 827                 int retval;
 828
 829                 if (src >= (mm->env_end - mm->env_start))
 830                         break;
 831
 832                 this_len = mm->env_end - (mm->env_start + src);
 833
 834                 max_len = min_t(size_t, PAGE_SIZE, count);
 835                 this_len = min(max_len, this_len);
 836
 837                 retval = access_remote_vm(mm, (mm->env_start + src),
 838                         page, this_len, 0);
 839
 840                 if (retval <= 0) {
 841                         ret = retval;
 842                         break;
 843                 }
 844
 845                 if (copy_to_user(buf, page, retval)) {
 846                         ret = -EFAULT;
 847                         break;
 848                 }
 849
 850                 ret += retval;
 851                 src += retval;
 852                 buf += retval;
 853                 count -= retval;
 854         }
 855         *ppos = src;
 856         mmput(mm);
 857
 858 free:
 859         free_page((unsigned long) page);
 860         return ret;
 861 }
 862
 863 static const struct file_operations proc_environ_operations = {
 864         .open           = environ_open,
 865         .read           = environ_read,
 866         .llseek         = generic_file_llseek,
 867         .release        = mem_release,
 868 };
 869
 870 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
 871                             loff_t *ppos)
 872 {
 873         struct task_struct *task = get_proc_task(file_inode(file));
 874         char buffer[PROC_NUMBUF];
 875         int oom_adj = OOM_ADJUST_MIN;
 876         size_t len;
 877         unsigned long flags;
 878
 879         if (!task)
 880                 return -ESRCH;
 881         if (lock_task_sighand(task, &flags)) {
 882                 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
 883                         oom_adj = OOM_ADJUST_MAX;
 884                 else
 885                         oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
 886                                   OOM_SCORE_ADJ_MAX;
 887                 unlock_task_sighand(task, &flags);
 888         }
 889         put_task_struct(task);
 890         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
 891         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 892 }
 893
 894 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
 895                              size_t count, loff_t *ppos)
 896 {
 897         struct task_struct *task;
 898         char buffer[PROC_NUMBUF];
 899         int oom_adj;
 900         unsigned long flags;
 901         int err;
 902
 903         memset(buffer, 0, sizeof(buffer));
 904         if (count > sizeof(buffer) - 1)
 905                 count = sizeof(buffer) - 1;
 906         if (copy_from_user(buffer, buf, count)) {
 907                 err = -EFAULT;
 908                 goto out;
 909         }
 910
 911         err = kstrtoint(strstrip(buffer), 0, &oom_adj);
 912         if (err)
 913                 goto out;
 914         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
 915              oom_adj != OOM_DISABLE) {
 916                 err = -EINVAL;
 917                 goto out;
 918         }
 919
 920         task = get_proc_task(file_inode(file));
 921         if (!task) {
 922                 err = -ESRCH;
 923                 goto out;
 924         }
 925
 926         task_lock(task);
 927         if (!task->mm) {
 928                 err = -EINVAL;
 929                 goto err_task_lock;
 930         }
 931
 932         if (!lock_task_sighand(task, &flags)) {
 933                 err = -ESRCH;
 934                 goto err_task_lock;
 935         }
 936
 937         /*
 938          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
 939          * value is always attainable.
 940          */
 941         if (oom_adj == OOM_ADJUST_MAX)
 942                 oom_adj = OOM_SCORE_ADJ_MAX;
 943         else
 944                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
 945
 946         if (oom_adj < task->signal->oom_score_adj &&
 947             !capable(CAP_SYS_RESOURCE)) {
 948                 err = -EACCES;
 949                 goto err_sighand;
 950         }
 951
 952         /*
 953          * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
 954          * /proc/pid/oom_score_adj instead.
 955          */
 956         pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
 957                   current->comm, task_pid_nr(current), task_pid_nr(task),
 958                   task_pid_nr(task));
 959
 960         task->signal->oom_score_adj = oom_adj;
 961         trace_oom_score_adj_update(task);
 962 err_sighand:
 963         unlock_task_sighand(task, &flags);
 964 err_task_lock:
 965         task_unlock(task);
 966         put_task_struct(task);
 967 out:
 968         return err < 0 ? err : count;
 969 }
 970
 971 static const struct file_operations proc_oom_adj_operations = {
 972         .read           = oom_adj_read,
 973         .write          = oom_adj_write,
 974         .llseek         = generic_file_llseek,
 975 };
 976
 977 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
 978                                         size_t count, loff_t *ppos)
 979 {
 980         struct task_struct *task = get_proc_task(file_inode(file));
 981         char buffer[PROC_NUMBUF];
 982         short oom_score_adj = OOM_SCORE_ADJ_MIN;
 983         unsigned long flags;
 984         size_t len;
 985
 986         if (!task)
 987                 return -ESRCH;
 988         if (lock_task_sighand(task, &flags)) {
 989                 oom_score_adj = task->signal->oom_score_adj;
 990                 unlock_task_sighand(task, &flags);
 991         }
 992         put_task_struct(task);
 993         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
 994         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 995 }
 996
 997 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
 998                                         size_t count, loff_t *ppos)
 999 {
1000         struct task_struct *task;
1001         char buffer[PROC_NUMBUF];
1002         unsigned long flags;
1003         int oom_score_adj;
1004         int err;
1005
1006         memset(buffer, 0, sizeof(buffer));
1007         if (count > sizeof(buffer) - 1)
1008                 count = sizeof(buffer) - 1;
1009         if (copy_from_user(buffer, buf, count)) {
1010                 err = -EFAULT;
1011                 goto out;
1012         }
1013
1014         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1015         if (err)
1016                 goto out;
1017         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1018                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1019                 err = -EINVAL;
1020                 goto out;
1021         }
1022
1023         task = get_proc_task(file_inode(file));
1024         if (!task) {
1025                 err = -ESRCH;
1026                 goto out;
1027         }
1028
1029         task_lock(task);
1030         if (!task->mm) {
1031                 err = -EINVAL;
1032                 goto err_task_lock;
1033         }
1034
1035         if (!lock_task_sighand(task, &flags)) {
1036                 err = -ESRCH;
1037                 goto err_task_lock;
1038         }
1039
1040         if ((short)oom_score_adj < task->signal->oom_score_adj_min &&
1041                         !capable(CAP_SYS_RESOURCE)) {
1042                 err = -EACCES;
1043                 goto err_sighand;
1044         }
1045
1046         task->signal->oom_score_adj = (short)oom_score_adj;
1047         if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1048                 task->signal->oom_score_adj_min = (short)oom_score_adj;
1049         trace_oom_score_adj_update(task);
1050
1051 err_sighand:
1052         unlock_task_sighand(task, &flags);
1053 err_task_lock:
1054         task_unlock(task);
1055         put_task_struct(task);
1056 out:
1057         return err < 0 ? err : count;
1058 }
1059
1060 static const struct file_operations proc_oom_score_adj_operations = {
1061         .read           = oom_score_adj_read,
1062         .write          = oom_score_adj_write,
1063         .llseek         = default_llseek,
1064 };
1065
1066 #ifdef CONFIG_AUDITSYSCALL
1067 #define TMPBUFLEN 21
1068 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1069                                   size_t count, loff_t *ppos)
1070 {
1071         struct inode * inode = file_inode(file);
1072         struct task_struct *task = get_proc_task(inode);
1073         ssize_t length;
1074         char tmpbuf[TMPBUFLEN];
1075
1076         if (!task)
1077                 return -ESRCH;
1078         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1079                            from_kuid(file->f_cred->user_ns,
1080                                      audit_get_loginuid(task)));
1081         put_task_struct(task);
1082         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1083 }
1084
1085 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1086                                    size_t count, loff_t *ppos)
1087 {
1088         struct inode * inode = file_inode(file);
1089         char *page, *tmp;
1090         ssize_t length;
1091         uid_t loginuid;
1092         kuid_t kloginuid;
1093
1094         rcu_read_lock();
1095         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1096                 rcu_read_unlock();
1097                 return -EPERM;
1098         }
1099         rcu_read_unlock();
1100
1101         if (count >= PAGE_SIZE)
1102                 count = PAGE_SIZE - 1;
1103
1104         if (*ppos != 0) {
1105                 /* No partial writes. */
1106                 return -EINVAL;
1107         }
1108         page = (char*)__get_free_page(GFP_TEMPORARY);
1109         if (!page)
1110                 return -ENOMEM;
1111         length = -EFAULT;
1112         if (copy_from_user(page, buf, count))
1113                 goto out_free_page;
1114
1115         page[count] = '\0';
1116         loginuid = simple_strtoul(page, &tmp, 10);
1117         if (tmp == page) {
1118                 length = -EINVAL;
1119                 goto out_free_page;
1120
1121         }
1122
1123         /* is userspace tring to explicitly UNSET the loginuid? */
1124         if (loginuid == AUDIT_UID_UNSET) {
1125                 kloginuid = INVALID_UID;
1126         } else {
1127                 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1128                 if (!uid_valid(kloginuid)) {
1129                         length = -EINVAL;
1130                         goto out_free_page;
1131                 }
1132         }
1133
1134         length = audit_set_loginuid(kloginuid);
1135         if (likely(length == 0))
1136                 length = count;
1137
1138 out_free_page:
1139         free_page((unsigned long) page);
1140         return length;
1141 }
1142
1143 static const struct file_operations proc_loginuid_operations = {
1144         .read           = proc_loginuid_read,
1145         .write          = proc_loginuid_write,
1146         .llseek         = generic_file_llseek,
1147 };
1148
1149 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1150                                   size_t count, loff_t *ppos)
1151 {
1152         struct inode * inode = file_inode(file);
1153         struct task_struct *task = get_proc_task(inode);
1154         ssize_t length;
1155         char tmpbuf[TMPBUFLEN];
1156
1157         if (!task)
1158                 return -ESRCH;
1159         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1160                                 audit_get_sessionid(task));
1161         put_task_struct(task);
1162         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1163 }
1164
1165 static const struct file_operations proc_sessionid_operations = {
1166         .read           = proc_sessionid_read,
1167         .llseek         = generic_file_llseek,
1168 };
1169 #endif
1170
1171 #ifdef CONFIG_FAULT_INJECTION
1172 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1173                                       size_t count, loff_t *ppos)
1174 {
1175         struct task_struct *task = get_proc_task(file_inode(file));
1176         char buffer[PROC_NUMBUF];
1177         size_t len;
1178         int make_it_fail;
1179
1180         if (!task)
1181                 return -ESRCH;
1182         make_it_fail = task->make_it_fail;
1183         put_task_struct(task);
1184
1185         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1186
1187         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1188 }
1189
1190 static ssize_t proc_fault_inject_write(struct file * file,
1191                         const char __user * buf, size_t count, loff_t *ppos)
1192 {
1193         struct task_struct *task;
1194         char buffer[PROC_NUMBUF], *end;
1195         int make_it_fail;
1196
1197         if (!capable(CAP_SYS_RESOURCE))
1198                 return -EPERM;
1199         memset(buffer, 0, sizeof(buffer));
1200         if (count > sizeof(buffer) - 1)
1201                 count = sizeof(buffer) - 1;
1202         if (copy_from_user(buffer, buf, count))
1203                 return -EFAULT;
1204         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1205         if (*end)
1206                 return -EINVAL;
1207         task = get_proc_task(file_inode(file));
1208         if (!task)
1209                 return -ESRCH;
1210         task->make_it_fail = make_it_fail;
1211         put_task_struct(task);
1212
1213         return count;
1214 }
1215
1216 static const struct file_operations proc_fault_inject_operations = {
1217         .read           = proc_fault_inject_read,
1218         .write          = proc_fault_inject_write,
1219         .llseek         = generic_file_llseek,
1220 };
1221 #endif
1222
1223
1224 #ifdef CONFIG_SCHED_DEBUG
1225 /*
1226  * Print out various scheduling related per-task fields:
1227  */
1228 static int sched_show(struct seq_file *m, void *v)
1229 {
1230         struct inode *inode = m->private;
1231         struct task_struct *p;
1232
1233         p = get_proc_task(inode);
1234         if (!p)
1235                 return -ESRCH;
1236         proc_sched_show_task(p, m);
1237
1238         put_task_struct(p);
1239
1240         return 0;
1241 }
1242
1243 static ssize_t
1244 sched_write(struct file *file, const char __user *buf,
1245             size_t count, loff_t *offset)
1246 {
1247         struct inode *inode = file_inode(file);
1248         struct task_struct *p;
1249
1250         p = get_proc_task(inode);
1251         if (!p)
1252                 return -ESRCH;
1253         proc_sched_set_task(p);
1254
1255         put_task_struct(p);
1256
1257         return count;
1258 }
1259
1260 static int sched_open(struct inode *inode, struct file *filp)
1261 {
1262         return single_open(filp, sched_show, inode);
1263 }
1264
1265 static const struct file_operations proc_pid_sched_operations = {
1266         .open           = sched_open,
1267         .read           = seq_read,
1268         .write          = sched_write,
1269         .llseek         = seq_lseek,
1270         .release        = single_release,
1271 };
1272
1273 #endif
1274
1275 #ifdef CONFIG_SCHED_AUTOGROUP
1276 /*
1277  * Print out autogroup related information:
1278  */
1279 static int sched_autogroup_show(struct seq_file *m, void *v)
1280 {
1281         struct inode *inode = m->private;
1282         struct task_struct *p;
1283
1284         p = get_proc_task(inode);
1285         if (!p)
1286                 return -ESRCH;
1287         proc_sched_autogroup_show_task(p, m);
1288
1289         put_task_struct(p);
1290
1291         return 0;
1292 }
1293
1294 static ssize_t
1295 sched_autogroup_write(struct file *file, const char __user *buf,
1296             size_t count, loff_t *offset)
1297 {
1298         struct inode *inode = file_inode(file);
1299         struct task_struct *p;
1300         char buffer[PROC_NUMBUF];
1301         int nice;
1302         int err;
1303
1304         memset(buffer, 0, sizeof(buffer));
1305         if (count > sizeof(buffer) - 1)
1306                 count = sizeof(buffer) - 1;
1307         if (copy_from_user(buffer, buf, count))
1308                 return -EFAULT;
1309
1310         err = kstrtoint(strstrip(buffer), 0, &nice);
1311         if (err < 0)
1312                 return err;
1313
1314         p = get_proc_task(inode);
1315         if (!p)
1316                 return -ESRCH;
1317
1318         err = proc_sched_autogroup_set_nice(p, nice);
1319         if (err)
1320                 count = err;
1321
1322         put_task_struct(p);
1323
1324         return count;
1325 }
1326
1327 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1328 {
1329         int ret;
1330
1331         ret = single_open(filp, sched_autogroup_show, NULL);
1332         if (!ret) {
1333                 struct seq_file *m = filp->private_data;
1334
1335                 m->private = inode;
1336         }
1337         return ret;
1338 }
1339
1340 static const struct file_operations proc_pid_sched_autogroup_operations = {
1341         .open           = sched_autogroup_open,
1342         .read           = seq_read,
1343         .write          = sched_autogroup_write,
1344         .llseek         = seq_lseek,
1345         .release        = single_release,
1346 };
1347
1348 #endif /* CONFIG_SCHED_AUTOGROUP */
1349
1350 static ssize_t comm_write(struct file *file, const char __user *buf,
1351                                 size_t count, loff_t *offset)
1352 {
1353         struct inode *inode = file_inode(file);
1354         struct task_struct *p;
1355         char buffer[TASK_COMM_LEN];
1356         const size_t maxlen = sizeof(buffer) - 1;
1357
1358         memset(buffer, 0, sizeof(buffer));
1359         if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1360                 return -EFAULT;
1361
1362         p = get_proc_task(inode);
1363         if (!p)
1364                 return -ESRCH;
1365
1366         if (same_thread_group(current, p))
1367                 set_task_comm(p, buffer);
1368         else
1369                 count = -EINVAL;
1370
1371         put_task_struct(p);
1372
1373         return count;
1374 }
1375
1376 static int comm_show(struct seq_file *m, void *v)
1377 {
1378         struct inode *inode = m->private;
1379         struct task_struct *p;
1380
1381         p = get_proc_task(inode);
1382         if (!p)
1383                 return -ESRCH;
1384
1385         task_lock(p);
1386         seq_printf(m, "%s\n", p->comm);
1387         task_unlock(p);
1388
1389         put_task_struct(p);
1390
1391         return 0;
1392 }
1393
1394 static int comm_open(struct inode *inode, struct file *filp)
1395 {
1396         return single_open(filp, comm_show, inode);
1397 }
1398
1399 static const struct file_operations proc_pid_set_comm_operations = {
1400         .open           = comm_open,
1401         .read           = seq_read,
1402         .write          = comm_write,
1403         .llseek         = seq_lseek,
1404         .release        = single_release,
1405 };
1406
1407 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1408 {
1409         struct task_struct *task;
1410         struct mm_struct *mm;
1411         struct file *exe_file;
1412
1413         task = get_proc_task(dentry->d_inode);
1414         if (!task)
1415                 return -ENOENT;
1416         mm = get_task_mm(task);
1417         put_task_struct(task);
1418         if (!mm)
1419                 return -ENOENT;
1420         exe_file = get_mm_exe_file(mm);
1421         mmput(mm);
1422         if (exe_file) {
1423                 *exe_path = exe_file->f_path;
1424                 path_get(&exe_file->f_path);
1425                 fput(exe_file);
1426                 return 0;
1427         } else
1428                 return -ENOENT;
1429 }
1430
1431 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1432 {
1433         struct inode *inode = dentry->d_inode;
1434         struct path path;
1435         int error = -EACCES;
1436
1437         /* Are we allowed to snoop on the tasks file descriptors? */
1438         if (!proc_fd_access_allowed(inode))
1439                 goto out;
1440
1441         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1442         if (error)
1443                 goto out;
1444
1445         nd_jump_link(nd, &path);
1446         return NULL;
1447 out:
1448         return ERR_PTR(error);
1449 }
1450
1451 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1452 {
1453         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1454         char *pathname;
1455         int len;
1456
1457         if (!tmp)
1458                 return -ENOMEM;
1459
1460         pathname = d_path(path, tmp, PAGE_SIZE);
1461         len = PTR_ERR(pathname);
1462         if (IS_ERR(pathname))
1463                 goto out;
1464         len = tmp + PAGE_SIZE - 1 - pathname;
1465
1466         if (len > buflen)
1467                 len = buflen;
1468         if (copy_to_user(buffer, pathname, len))
1469                 len = -EFAULT;
1470  out:
1471         free_page((unsigned long)tmp);
1472         return len;
1473 }
1474
1475 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1476 {
1477         int error = -EACCES;
1478         struct inode *inode = dentry->d_inode;
1479         struct path path;
1480
1481         /* Are we allowed to snoop on the tasks file descriptors? */
1482         if (!proc_fd_access_allowed(inode))
1483                 goto out;
1484
1485         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1486         if (error)
1487                 goto out;
1488
1489         error = do_proc_readlink(&path, buffer, buflen);
1490         path_put(&path);
1491 out:
1492         return error;
1493 }
1494
1495 const struct inode_operations proc_pid_link_inode_operations = {
1496         .readlink       = proc_pid_readlink,
1497         .follow_link    = proc_pid_follow_link,
1498         .setattr        = proc_setattr,
1499 };
1500
1501
1502 /* building an inode */
1503
1504 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1505 {
1506         struct inode * inode;
1507         struct proc_inode *ei;
1508         const struct cred *cred;
1509
1510         /* We need a new inode */
1511
1512         inode = new_inode(sb);
1513         if (!inode)
1514                 goto out;
1515
1516         /* Common stuff */
1517         ei = PROC_I(inode);
1518         inode->i_ino = get_next_ino();
1519         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1520         inode->i_op = &proc_def_inode_operations;
1521
1522         /*
1523          * grab the reference to task.
1524          */
1525         ei->pid = get_task_pid(task, PIDTYPE_PID);
1526         if (!ei->pid)
1527                 goto out_unlock;
1528
1529         if (task_dumpable(task)) {
1530                 rcu_read_lock();
1531                 cred = __task_cred(task);
1532                 inode->i_uid = cred->euid;
1533                 inode->i_gid = cred->egid;
1534                 rcu_read_unlock();
1535         }
1536         security_task_to_inode(task, inode);
1537
1538 out:
1539         return inode;
1540
1541 out_unlock:
1542         iput(inode);
1543         return NULL;
1544 }
1545
1546 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1547 {
1548         struct inode *inode = dentry->d_inode;
1549         struct task_struct *task;
1550         const struct cred *cred;
1551         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1552
1553         generic_fillattr(inode, stat);
1554
1555         rcu_read_lock();
1556         stat->uid = GLOBAL_ROOT_UID;
1557         stat->gid = GLOBAL_ROOT_GID;
1558         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1559         if (task) {
1560                 if (!has_pid_permissions(pid, task, 2)) {
1561                         rcu_read_unlock();
1562                         /*
1563                          * This doesn't prevent learning whether PID exists,
1564                          * it only makes getattr() consistent with readdir().
1565                          */
1566                         return -ENOENT;
1567                 }
1568                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1569                     task_dumpable(task)) {
1570                         cred = __task_cred(task);
1571                         stat->uid = cred->euid;
1572                         stat->gid = cred->egid;
1573                 }
1574         }
1575         rcu_read_unlock();
1576         return 0;
1577 }
1578
1579 /* dentry stuff */
1580
1581 /*
1582  *      Exceptional case: normally we are not allowed to unhash a busy
1583  * directory. In this case, however, we can do it - no aliasing problems
1584  * due to the way we treat inodes.
1585  *
1586  * Rewrite the inode's ownerships here because the owning task may have
1587  * performed a setuid(), etc.
1588  *
1589  * Before the /proc/pid/status file was created the only way to read
1590  * the effective uid of a /process was to stat /proc/pid.  Reading
1591  * /proc/pid/status is slow enough that procps and other packages
1592  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1593  * made this apply to all per process world readable and executable
1594  * directories.
1595  */
1596 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1597 {
1598         struct inode *inode;
1599         struct task_struct *task;
1600         const struct cred *cred;
1601
1602         if (flags & LOOKUP_RCU)
1603                 return -ECHILD;
1604
1605         inode = dentry->d_inode;
1606         task = get_proc_task(inode);
1607
1608         if (task) {
1609                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1610                     task_dumpable(task)) {
1611                         rcu_read_lock();
1612                         cred = __task_cred(task);
1613                         inode->i_uid = cred->euid;
1614                         inode->i_gid = cred->egid;
1615                         rcu_read_unlock();
1616                 } else {
1617                         inode->i_uid = GLOBAL_ROOT_UID;
1618                         inode->i_gid = GLOBAL_ROOT_GID;
1619                 }
1620                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1621                 security_task_to_inode(task, inode);
1622                 put_task_struct(task);
1623                 return 1;
1624         }
1625         d_drop(dentry);
1626         return 0;
1627 }
1628
1629 static inline bool proc_inode_is_dead(struct inode *inode)
1630 {
1631         return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1632 }
1633
1634 int pid_delete_dentry(const struct dentry *dentry)
1635 {
1636         /* Is the task we represent dead?
1637          * If so, then don't put the dentry on the lru list,
1638          * kill it immediately.
1639          */
1640         return proc_inode_is_dead(dentry->d_inode);
1641 }
1642
1643 const struct dentry_operations pid_dentry_operations =
1644 {
1645         .d_revalidate   = pid_revalidate,
1646         .d_delete       = pid_delete_dentry,
1647 };
1648
1649 /* Lookups */
1650
1651 /*
1652  * Fill a directory entry.
1653  *
1654  * If possible create the dcache entry and derive our inode number and
1655  * file type from dcache entry.
1656  *
1657  * Since all of the proc inode numbers are dynamically generated, the inode
1658  * numbers do not exist until the inode is cache.  This means creating the
1659  * the dcache entry in readdir is necessary to keep the inode numbers
1660  * reported by readdir in sync with the inode numbers reported
1661  * by stat.
1662  */
1663 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1664         const char *name, int len,
1665         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1666 {
1667         struct dentry *child, *dir = file->f_path.dentry;
1668         struct qstr qname = QSTR_INIT(name, len);
1669         struct inode *inode;
1670         unsigned type;
1671         ino_t ino;
1672
1673         child = d_hash_and_lookup(dir, &qname);
1674         if (!child) {
1675                 child = d_alloc(dir, &qname);
1676                 if (!child)
1677                         goto end_instantiate;
1678                 if (instantiate(dir->d_inode, child, task, ptr) < 0) {
1679                         dput(child);
1680                         goto end_instantiate;
1681                 }
1682         }
1683         inode = child->d_inode;
1684         ino = inode->i_ino;
1685         type = inode->i_mode >> 12;
1686         dput(child);
1687         return dir_emit(ctx, name, len, ino, type);
1688
1689 end_instantiate:
1690         return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1691 }
1692
1693 #ifdef CONFIG_CHECKPOINT_RESTORE
1694
1695 /*
1696  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1697  * which represent vma start and end addresses.
1698  */
1699 static int dname_to_vma_addr(struct dentry *dentry,
1700                              unsigned long *start, unsigned long *end)
1701 {
1702         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1703                 return -EINVAL;
1704
1705         return 0;
1706 }
1707
1708 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1709 {
1710         unsigned long vm_start, vm_end;
1711         bool exact_vma_exists = false;
1712         struct mm_struct *mm = NULL;
1713         struct task_struct *task;
1714         const struct cred *cred;
1715         struct inode *inode;
1716         int status = 0;
1717
1718         if (flags & LOOKUP_RCU)
1719                 return -ECHILD;
1720
1721         if (!capable(CAP_SYS_ADMIN)) {
1722                 status = -EPERM;
1723                 goto out_notask;
1724         }
1725
1726         inode = dentry->d_inode;
1727         task = get_proc_task(inode);
1728         if (!task)
1729                 goto out_notask;
1730
1731         mm = mm_access(task, PTRACE_MODE_READ);
1732         if (IS_ERR_OR_NULL(mm))
1733                 goto out;
1734
1735         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1736                 down_read(&mm->mmap_sem);
1737                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1738                 up_read(&mm->mmap_sem);
1739         }
1740
1741         mmput(mm);
1742
1743         if (exact_vma_exists) {
1744                 if (task_dumpable(task)) {
1745                         rcu_read_lock();
1746                         cred = __task_cred(task);
1747                         inode->i_uid = cred->euid;
1748                         inode->i_gid = cred->egid;
1749                         rcu_read_unlock();
1750                 } else {
1751                         inode->i_uid = GLOBAL_ROOT_UID;
1752                         inode->i_gid = GLOBAL_ROOT_GID;
1753                 }
1754                 security_task_to_inode(task, inode);
1755                 status = 1;
1756         }
1757
1758 out:
1759         put_task_struct(task);
1760
1761 out_notask:
1762         if (status <= 0)
1763                 d_drop(dentry);
1764
1765         return status;
1766 }
1767
1768 static const struct dentry_operations tid_map_files_dentry_operations = {
1769         .d_revalidate   = map_files_d_revalidate,
1770         .d_delete       = pid_delete_dentry,
1771 };
1772
1773 static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
1774 {
1775         unsigned long vm_start, vm_end;
1776         struct vm_area_struct *vma;
1777         struct task_struct *task;
1778         struct mm_struct *mm;
1779         int rc;
1780
1781         rc = -ENOENT;
1782         task = get_proc_task(dentry->d_inode);
1783         if (!task)
1784                 goto out;
1785
1786         mm = get_task_mm(task);
1787         put_task_struct(task);
1788         if (!mm)
1789                 goto out;
1790
1791         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1792         if (rc)
1793                 goto out_mmput;
1794
1795         rc = -ENOENT;
1796         down_read(&mm->mmap_sem);
1797         vma = find_exact_vma(mm, vm_start, vm_end);
1798         if (vma && vma->vm_file) {
1799                 *path = vma->vm_file->f_path;
1800                 path_get(path);
1801                 rc = 0;
1802         }
1803         up_read(&mm->mmap_sem);
1804
1805 out_mmput:
1806         mmput(mm);
1807 out:
1808         return rc;
1809 }
1810
1811 struct map_files_info {
1812         fmode_t         mode;
1813         unsigned long   len;
1814         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1815 };
1816
1817 static int
1818 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
1819                            struct task_struct *task, const void *ptr)
1820 {
1821         fmode_t mode = (fmode_t)(unsigned long)ptr;
1822         struct proc_inode *ei;
1823         struct inode *inode;
1824
1825         inode = proc_pid_make_inode(dir->i_sb, task);
1826         if (!inode)
1827                 return -ENOENT;
1828
1829         ei = PROC_I(inode);
1830         ei->op.proc_get_link = proc_map_files_get_link;
1831
1832         inode->i_op = &proc_pid_link_inode_operations;
1833         inode->i_size = 64;
1834         inode->i_mode = S_IFLNK;
1835
1836         if (mode & FMODE_READ)
1837                 inode->i_mode |= S_IRUSR;
1838         if (mode & FMODE_WRITE)
1839                 inode->i_mode |= S_IWUSR;
1840
1841         d_set_d_op(dentry, &tid_map_files_dentry_operations);
1842         d_add(dentry, inode);
1843
1844         return 0;
1845 }
1846
1847 static struct dentry *proc_map_files_lookup(struct inode *dir,
1848                 struct dentry *dentry, unsigned int flags)
1849 {
1850         unsigned long vm_start, vm_end;
1851         struct vm_area_struct *vma;
1852         struct task_struct *task;
1853         int result;
1854         struct mm_struct *mm;
1855
1856         result = -EPERM;
1857         if (!capable(CAP_SYS_ADMIN))
1858                 goto out;
1859
1860         result = -ENOENT;
1861         task = get_proc_task(dir);
1862         if (!task)
1863                 goto out;
1864
1865         result = -EACCES;
1866         if (!ptrace_may_access(task, PTRACE_MODE_READ))
1867                 goto out_put_task;
1868
1869         result = -ENOENT;
1870         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
1871                 goto out_put_task;
1872
1873         mm = get_task_mm(task);
1874         if (!mm)
1875                 goto out_put_task;
1876
1877         down_read(&mm->mmap_sem);
1878         vma = find_exact_vma(mm, vm_start, vm_end);
1879         if (!vma)
1880                 goto out_no_vma;
1881
1882         if (vma->vm_file)
1883                 result = proc_map_files_instantiate(dir, dentry, task,
1884                                 (void *)(unsigned long)vma->vm_file->f_mode);
1885
1886 out_no_vma:
1887         up_read(&mm->mmap_sem);
1888         mmput(mm);
1889 out_put_task:
1890         put_task_struct(task);
1891 out:
1892         return ERR_PTR(result);
1893 }
1894
1895 static const struct inode_operations proc_map_files_inode_operations = {
1896         .lookup         = proc_map_files_lookup,
1897         .permission     = proc_fd_permission,
1898         .setattr        = proc_setattr,
1899 };
1900
1901 static int
1902 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
1903 {
1904         struct vm_area_struct *vma;
1905         struct task_struct *task;
1906         struct mm_struct *mm;
1907         unsigned long nr_files, pos, i;
1908         struct flex_array *fa = NULL;
1909         struct map_files_info info;
1910         struct map_files_info *p;
1911         int ret;
1912
1913         ret = -EPERM;
1914         if (!capable(CAP_SYS_ADMIN))
1915                 goto out;
1916
1917         ret = -ENOENT;
1918         task = get_proc_task(file_inode(file));
1919         if (!task)
1920                 goto out;
1921
1922         ret = -EACCES;
1923         if (!ptrace_may_access(task, PTRACE_MODE_READ))
1924                 goto out_put_task;
1925
1926         ret = 0;
1927         if (!dir_emit_dots(file, ctx))
1928                 goto out_put_task;
1929
1930         mm = get_task_mm(task);
1931         if (!mm)
1932                 goto out_put_task;
1933         down_read(&mm->mmap_sem);
1934
1935         nr_files = 0;
1936
1937         /*
1938          * We need two passes here:
1939          *
1940          *  1) Collect vmas of mapped files with mmap_sem taken
1941          *  2) Release mmap_sem and instantiate entries
1942          *
1943          * otherwise we get lockdep complained, since filldir()
1944          * routine might require mmap_sem taken in might_fault().
1945          */
1946
1947         for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
1948                 if (vma->vm_file && ++pos > ctx->pos)
1949                         nr_files++;
1950         }
1951
1952         if (nr_files) {
1953                 fa = flex_array_alloc(sizeof(info), nr_files,
1954                                         GFP_KERNEL);
1955                 if (!fa || flex_array_prealloc(fa, 0, nr_files,
1956                                                 GFP_KERNEL)) {
1957                         ret = -ENOMEM;
1958                         if (fa)
1959                                 flex_array_free(fa);
1960                         up_read(&mm->mmap_sem);
1961                         mmput(mm);
1962                         goto out_put_task;
1963                 }
1964                 for (i = 0, vma = mm->mmap, pos = 2; vma;
1965                                 vma = vma->vm_next) {
1966                         if (!vma->vm_file)
1967                                 continue;
1968                         if (++pos <= ctx->pos)
1969                                 continue;
1970
1971                         info.mode = vma->vm_file->f_mode;
1972                         info.len = snprintf(info.name,
1973                                         sizeof(info.name), "%lx-%lx",
1974                                         vma->vm_start, vma->vm_end);
1975                         if (flex_array_put(fa, i++, &info, GFP_KERNEL))
1976                                 BUG();
1977                 }
1978         }
1979         up_read(&mm->mmap_sem);
1980
1981         for (i = 0; i < nr_files; i++) {
1982                 p = flex_array_get(fa, i);
1983                 if (!proc_fill_cache(file, ctx,
1984                                       p->name, p->len,
1985                                       proc_map_files_instantiate,
1986                                       task,
1987                                       (void *)(unsigned long)p->mode))
1988                         break;
1989                 ctx->pos++;
1990         }
1991         if (fa)
1992                 flex_array_free(fa);
1993         mmput(mm);
1994
1995 out_put_task:
1996         put_task_struct(task);
1997 out:
1998         return ret;
1999 }
2000
2001 static const struct file_operations proc_map_files_operations = {
2002         .read           = generic_read_dir,
2003         .iterate        = proc_map_files_readdir,
2004         .llseek         = default_llseek,
2005 };
2006
2007 struct timers_private {
2008         struct pid *pid;
2009         struct task_struct *task;
2010         struct sighand_struct *sighand;
2011         struct pid_namespace *ns;
2012         unsigned long flags;
2013 };
2014
2015 static void *timers_start(struct seq_file *m, loff_t *pos)
2016 {
2017         struct timers_private *tp = m->private;
2018
2019         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2020         if (!tp->task)
2021                 return ERR_PTR(-ESRCH);
2022
2023         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2024         if (!tp->sighand)
2025                 return ERR_PTR(-ESRCH);
2026
2027         return seq_list_start(&tp->task->signal->posix_timers, *pos);
2028 }
2029
2030 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2031 {
2032         struct timers_private *tp = m->private;
2033         return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2034 }
2035
2036 static void timers_stop(struct seq_file *m, void *v)
2037 {
2038         struct timers_private *tp = m->private;
2039
2040         if (tp->sighand) {
2041                 unlock_task_sighand(tp->task, &tp->flags);
2042                 tp->sighand = NULL;
2043         }
2044
2045         if (tp->task) {
2046                 put_task_struct(tp->task);
2047                 tp->task = NULL;
2048         }
2049 }
2050
2051 static int show_timer(struct seq_file *m, void *v)
2052 {
2053         struct k_itimer *timer;
2054         struct timers_private *tp = m->private;
2055         int notify;
2056         static char *nstr[] = {
2057                 [SIGEV_SIGNAL] = "signal",
2058                 [SIGEV_NONE] = "none",
2059                 [SIGEV_THREAD] = "thread",
2060         };
2061
2062         timer = list_entry((struct list_head *)v, struct k_itimer, list);
2063         notify = timer->it_sigev_notify;
2064
2065         seq_printf(m, "ID: %d\n", timer->it_id);
2066         seq_printf(m, "signal: %d/%p\n", timer->sigq->info.si_signo,
2067                         timer->sigq->info.si_value.sival_ptr);
2068         seq_printf(m, "notify: %s/%s.%d\n",
2069                 nstr[notify & ~SIGEV_THREAD_ID],
2070                 (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2071                 pid_nr_ns(timer->it_pid, tp->ns));
2072         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2073
2074         return 0;
2075 }
2076
2077 static const struct seq_operations proc_timers_seq_ops = {
2078         .start  = timers_start,
2079         .next   = timers_next,
2080         .stop   = timers_stop,
2081         .show   = show_timer,
2082 };
2083
2084 static int proc_timers_open(struct inode *inode, struct file *file)
2085 {
2086         struct timers_private *tp;
2087
2088         tp = __seq_open_private(file, &proc_timers_seq_ops,
2089                         sizeof(struct timers_private));
2090         if (!tp)
2091                 return -ENOMEM;
2092
2093         tp->pid = proc_pid(inode);
2094         tp->ns = inode->i_sb->s_fs_info;
2095         return 0;
2096 }
2097
2098 static const struct file_operations proc_timers_operations = {
2099         .open           = proc_timers_open,
2100         .read           = seq_read,
2101         .llseek         = seq_lseek,
2102         .release        = seq_release_private,
2103 };
2104 #endif /* CONFIG_CHECKPOINT_RESTORE */
2105
2106 static int proc_pident_instantiate(struct inode *dir,
2107         struct dentry *dentry, struct task_struct *task, const void *ptr)
2108 {
2109         const struct pid_entry *p = ptr;
2110         struct inode *inode;
2111         struct proc_inode *ei;
2112
2113         inode = proc_pid_make_inode(dir->i_sb, task);
2114         if (!inode)
2115                 goto out;
2116
2117         ei = PROC_I(inode);
2118         inode->i_mode = p->mode;
2119         if (S_ISDIR(inode->i_mode))
2120                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2121         if (p->iop)
2122                 inode->i_op = p->iop;
2123         if (p->fop)
2124                 inode->i_fop = p->fop;
2125         ei->op = p->op;
2126         d_set_d_op(dentry, &pid_dentry_operations);
2127         d_add(dentry, inode);
2128         /* Close the race of the process dying before we return the dentry */
2129         if (pid_revalidate(dentry, 0))
2130                 return 0;
2131 out:
2132         return -ENOENT;
2133 }
2134
2135 static struct dentry *proc_pident_lookup(struct inode *dir,
2136                                          struct dentry *dentry,
2137                                          const struct pid_entry *ents,
2138                                          unsigned int nents)
2139 {
2140         int error;
2141         struct task_struct *task = get_proc_task(dir);
2142         const struct pid_entry *p, *last;
2143
2144         error = -ENOENT;
2145
2146         if (!task)
2147                 goto out_no_task;
2148
2149         /*
2150          * Yes, it does not scale. And it should not. Don't add
2151          * new entries into /proc/<tgid>/ without very good reasons.
2152          */
2153         last = &ents[nents - 1];
2154         for (p = ents; p <= last; p++) {
2155                 if (p->len != dentry->d_name.len)
2156                         continue;
2157                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2158                         break;
2159         }
2160         if (p > last)
2161                 goto out;
2162
2163         error = proc_pident_instantiate(dir, dentry, task, p);
2164 out:
2165         put_task_struct(task);
2166 out_no_task:
2167         return ERR_PTR(error);
2168 }
2169
2170 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2171                 const struct pid_entry *ents, unsigned int nents)
2172 {
2173         struct task_struct *task = get_proc_task(file_inode(file));
2174         const struct pid_entry *p;
2175
2176         if (!task)
2177                 return -ENOENT;
2178
2179         if (!dir_emit_dots(file, ctx))
2180                 goto out;
2181
2182         if (ctx->pos >= nents + 2)
2183                 goto out;
2184
2185         for (p = ents + (ctx->pos - 2); p <= ents + nents - 1; p++) {
2186                 if (!proc_fill_cache(file, ctx, p->name, p->len,
2187                                 proc_pident_instantiate, task, p))
2188                         break;
2189                 ctx->pos++;
2190         }
2191 out:
2192         put_task_struct(task);
2193         return 0;
2194 }
2195
2196 #ifdef CONFIG_SECURITY
2197 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2198                                   size_t count, loff_t *ppos)
2199 {
2200         struct inode * inode = file_inode(file);
2201         char *p = NULL;
2202         ssize_t length;
2203         struct task_struct *task = get_proc_task(inode);
2204
2205         if (!task)
2206                 return -ESRCH;
2207
2208         length = security_getprocattr(task,
2209                                       (char*)file->f_path.dentry->d_name.name,
2210                                       &p);
2211         put_task_struct(task);
2212         if (length > 0)
2213                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2214         kfree(p);
2215         return length;
2216 }
2217
2218 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2219                                    size_t count, loff_t *ppos)
2220 {
2221         struct inode * inode = file_inode(file);
2222         char *page;
2223         ssize_t length;
2224         struct task_struct *task = get_proc_task(inode);
2225
2226         length = -ESRCH;
2227         if (!task)
2228                 goto out_no_task;
2229         if (count > PAGE_SIZE)
2230                 count = PAGE_SIZE;
2231
2232         /* No partial writes. */
2233         length = -EINVAL;
2234         if (*ppos != 0)
2235                 goto out;
2236
2237         length = -ENOMEM;
2238         page = (char*)__get_free_page(GFP_TEMPORARY);
2239         if (!page)
2240                 goto out;
2241
2242         length = -EFAULT;
2243         if (copy_from_user(page, buf, count))
2244                 goto out_free;
2245
2246         /* Guard against adverse ptrace interaction */
2247         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2248         if (length < 0)
2249                 goto out_free;
2250
2251         length = security_setprocattr(task,
2252                                       (char*)file->f_path.dentry->d_name.name,
2253                                       (void*)page, count);
2254         mutex_unlock(&task->signal->cred_guard_mutex);
2255 out_free:
2256         free_page((unsigned long) page);
2257 out:
2258         put_task_struct(task);
2259 out_no_task:
2260         return length;
2261 }
2262
2263 static const struct file_operations proc_pid_attr_operations = {
2264         .read           = proc_pid_attr_read,
2265         .write          = proc_pid_attr_write,
2266         .llseek         = generic_file_llseek,
2267 };
2268
2269 static const struct pid_entry attr_dir_stuff[] = {
2270         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2271         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2272         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2273         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2274         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2275         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2276 };
2277
2278 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2279 {
2280         return proc_pident_readdir(file, ctx,
2281                                    attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2282 }
2283
2284 static const struct file_operations proc_attr_dir_operations = {
2285         .read           = generic_read_dir,
2286         .iterate        = proc_attr_dir_readdir,
2287         .llseek         = default_llseek,
2288 };
2289
2290 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2291                                 struct dentry *dentry, unsigned int flags)
2292 {
2293         return proc_pident_lookup(dir, dentry,
2294                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2295 }
2296
2297 static const struct inode_operations proc_attr_dir_inode_operations = {
2298         .lookup         = proc_attr_dir_lookup,
2299         .getattr        = pid_getattr,
2300         .setattr        = proc_setattr,
2301 };
2302
2303 #endif
2304
2305 #ifdef CONFIG_ELF_CORE
2306 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2307                                          size_t count, loff_t *ppos)
2308 {
2309         struct task_struct *task = get_proc_task(file_inode(file));
2310         struct mm_struct *mm;
2311         char buffer[PROC_NUMBUF];
2312         size_t len;
2313         int ret;
2314
2315         if (!task)
2316                 return -ESRCH;
2317
2318         ret = 0;
2319         mm = get_task_mm(task);
2320         if (mm) {
2321                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2322                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2323                                 MMF_DUMP_FILTER_SHIFT));
2324                 mmput(mm);
2325                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2326         }
2327
2328         put_task_struct(task);
2329
2330         return ret;
2331 }
2332
2333 static ssize_t proc_coredump_filter_write(struct file *file,
2334                                           const char __user *buf,
2335                                           size_t count,
2336                                           loff_t *ppos)
2337 {
2338         struct task_struct *task;
2339         struct mm_struct *mm;
2340         char buffer[PROC_NUMBUF], *end;
2341         unsigned int val;
2342         int ret;
2343         int i;
2344         unsigned long mask;
2345
2346         ret = -EFAULT;
2347         memset(buffer, 0, sizeof(buffer));
2348         if (count > sizeof(buffer) - 1)
2349                 count = sizeof(buffer) - 1;
2350         if (copy_from_user(buffer, buf, count))
2351                 goto out_no_task;
2352
2353         ret = -EINVAL;
2354         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2355         if (*end == '\n')
2356                 end++;
2357         if (end - buffer == 0)
2358                 goto out_no_task;
2359
2360         ret = -ESRCH;
2361         task = get_proc_task(file_inode(file));
2362         if (!task)
2363                 goto out_no_task;
2364
2365         ret = end - buffer;
2366         mm = get_task_mm(task);
2367         if (!mm)
2368                 goto out_no_mm;
2369
2370         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2371                 if (val & mask)
2372                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2373                 else
2374                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2375         }
2376
2377         mmput(mm);
2378  out_no_mm:
2379         put_task_struct(task);
2380  out_no_task:
2381         return ret;
2382 }
2383
2384 static const struct file_operations proc_coredump_filter_operations = {
2385         .read           = proc_coredump_filter_read,
2386         .write          = proc_coredump_filter_write,
2387         .llseek         = generic_file_llseek,
2388 };
2389 #endif
2390
2391 #ifdef CONFIG_TASK_IO_ACCOUNTING
2392 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2393 {
2394         struct task_io_accounting acct = task->ioac;
2395         unsigned long flags;
2396         int result;
2397
2398         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2399         if (result)
2400                 return result;
2401
2402         if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2403                 result = -EACCES;
2404                 goto out_unlock;
2405         }
2406
2407         if (whole && lock_task_sighand(task, &flags)) {
2408                 struct task_struct *t = task;
2409
2410                 task_io_accounting_add(&acct, &task->signal->ioac);
2411                 while_each_thread(task, t)
2412                         task_io_accounting_add(&acct, &t->ioac);
2413
2414                 unlock_task_sighand(task, &flags);
2415         }
2416         result = sprintf(buffer,
2417                         "rchar: %llu\n"
2418                         "wchar: %llu\n"
2419                         "syscr: %llu\n"
2420                         "syscw: %llu\n"
2421                         "read_bytes: %llu\n"
2422                         "write_bytes: %llu\n"
2423                         "cancelled_write_bytes: %llu\n",
2424                         (unsigned long long)acct.rchar,
2425                         (unsigned long long)acct.wchar,
2426                         (unsigned long long)acct.syscr,
2427                         (unsigned long long)acct.syscw,
2428                         (unsigned long long)acct.read_bytes,
2429                         (unsigned long long)acct.write_bytes,
2430                         (unsigned long long)acct.cancelled_write_bytes);
2431 out_unlock:
2432         mutex_unlock(&task->signal->cred_guard_mutex);
2433         return result;
2434 }
2435
2436 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2437 {
2438         return do_io_accounting(task, buffer, 0);
2439 }
2440
2441 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2442 {
2443         return do_io_accounting(task, buffer, 1);
2444 }
2445 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2446
2447 #ifdef CONFIG_USER_NS
2448 static int proc_id_map_open(struct inode *inode, struct file *file,
2449         struct seq_operations *seq_ops)
2450 {
2451         struct user_namespace *ns = NULL;
2452         struct task_struct *task;
2453         struct seq_file *seq;
2454         int ret = -EINVAL;
2455
2456         task = get_proc_task(inode);
2457         if (task) {
2458                 rcu_read_lock();
2459                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2460                 rcu_read_unlock();
2461                 put_task_struct(task);
2462         }
2463         if (!ns)
2464                 goto err;
2465
2466         ret = seq_open(file, seq_ops);
2467         if (ret)
2468                 goto err_put_ns;
2469
2470         seq = file->private_data;
2471         seq->private = ns;
2472
2473         return 0;
2474 err_put_ns:
2475         put_user_ns(ns);
2476 err:
2477         return ret;
2478 }
2479
2480 static int proc_id_map_release(struct inode *inode, struct file *file)
2481 {
2482         struct seq_file *seq = file->private_data;
2483         struct user_namespace *ns = seq->private;
2484         put_user_ns(ns);
2485         return seq_release(inode, file);
2486 }
2487
2488 static int proc_uid_map_open(struct inode *inode, struct file *file)
2489 {
2490         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2491 }
2492
2493 static int proc_gid_map_open(struct inode *inode, struct file *file)
2494 {
2495         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2496 }
2497
2498 static int proc_projid_map_open(struct inode *inode, struct file *file)
2499 {
2500         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2501 }
2502
2503 static const struct file_operations proc_uid_map_operations = {
2504         .open           = proc_uid_map_open,
2505         .write          = proc_uid_map_write,
2506         .read           = seq_read,
2507         .llseek         = seq_lseek,
2508         .release        = proc_id_map_release,
2509 };
2510
2511 static const struct file_operations proc_gid_map_operations = {
2512         .open           = proc_gid_map_open,
2513         .write          = proc_gid_map_write,
2514         .read           = seq_read,
2515         .llseek         = seq_lseek,
2516         .release        = proc_id_map_release,
2517 };
2518
2519 static const struct file_operations proc_projid_map_operations = {
2520         .open           = proc_projid_map_open,
2521         .write          = proc_projid_map_write,
2522         .read           = seq_read,
2523         .llseek         = seq_lseek,
2524         .release        = proc_id_map_release,
2525 };
2526 #endif /* CONFIG_USER_NS */
2527
2528 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2529                                 struct pid *pid, struct task_struct *task)
2530 {
2531         int err = lock_trace(task);
2532         if (!err) {
2533                 seq_printf(m, "%08x\n", task->personality);
2534                 unlock_trace(task);
2535         }
2536         return err;
2537 }
2538
2539 /*
2540  * Thread groups
2541  */
2542 static const struct file_operations proc_task_operations;
2543 static const struct inode_operations proc_task_inode_operations;
2544
2545 static const struct pid_entry tgid_base_stuff[] = {
2546         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2547         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2548 #ifdef CONFIG_CHECKPOINT_RESTORE
2549         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2550 #endif
2551         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2552         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2553 #ifdef CONFIG_NET
2554         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2555 #endif
2556         REG("environ",    S_IRUSR, proc_environ_operations),
2557         INF("auxv",       S_IRUSR, proc_pid_auxv),
2558         ONE("status",     S_IRUGO, proc_pid_status),
2559         ONE("personality", S_IRUGO, proc_pid_personality),
2560         INF("limits",     S_IRUGO, proc_pid_limits),
2561 #ifdef CONFIG_SCHED_DEBUG
2562         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2563 #endif
2564 #ifdef CONFIG_SCHED_AUTOGROUP
2565         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2566 #endif
2567         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2568 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2569         INF("syscall",    S_IRUGO, proc_pid_syscall),
2570 #endif
2571         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2572         ONE("stat",       S_IRUGO, proc_tgid_stat),
2573         ONE("statm",      S_IRUGO, proc_pid_statm),
2574         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2575 #ifdef CONFIG_NUMA
2576         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2577 #endif
2578         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2579         LNK("cwd",        proc_cwd_link),
2580         LNK("root",       proc_root_link),
2581         LNK("exe",        proc_exe_link),
2582         REG("mounts",     S_IRUGO, proc_mounts_operations),
2583         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2584         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2585 #ifdef CONFIG_PROC_PAGE_MONITOR
2586         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2587         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2588         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
2589 #endif
2590 #ifdef CONFIG_SECURITY
2591         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2592 #endif
2593 #ifdef CONFIG_KALLSYMS
2594         INF("wchan",      S_IRUGO, proc_pid_wchan),
2595 #endif
2596 #ifdef CONFIG_STACKTRACE
2597         ONE("stack",      S_IRUGO, proc_pid_stack),
2598 #endif
2599 #ifdef CONFIG_SCHEDSTATS
2600         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
2601 #endif
2602 #ifdef CONFIG_LATENCYTOP
2603         REG("latency",  S_IRUGO, proc_lstats_operations),
2604 #endif
2605 #ifdef CONFIG_PROC_PID_CPUSET
2606         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2607 #endif
2608 #ifdef CONFIG_CGROUPS
2609         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2610 #endif
2611         INF("oom_score",  S_IRUGO, proc_oom_score),
2612         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2613         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2614 #ifdef CONFIG_AUDITSYSCALL
2615         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2616         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2617 #endif
2618 #ifdef CONFIG_FAULT_INJECTION
2619         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2620 #endif
2621 #ifdef CONFIG_ELF_CORE
2622         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2623 #endif
2624 #ifdef CONFIG_TASK_IO_ACCOUNTING
2625         INF("io",       S_IRUSR, proc_tgid_io_accounting),
2626 #endif
2627 #ifdef CONFIG_HARDWALL
2628         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
2629 #endif
2630 #ifdef CONFIG_USER_NS
2631         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2632         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2633         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2634 #endif
2635 #ifdef CONFIG_CHECKPOINT_RESTORE
2636         REG("timers",     S_IRUGO, proc_timers_operations),
2637 #endif
2638 };
2639
2640 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2641 {
2642         return proc_pident_readdir(file, ctx,
2643                                    tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2644 }
2645
2646 static const struct file_operations proc_tgid_base_operations = {
2647         .read           = generic_read_dir,
2648         .iterate        = proc_tgid_base_readdir,
2649         .llseek         = default_llseek,
2650 };
2651
2652 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2653 {
2654         return proc_pident_lookup(dir, dentry,
2655                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2656 }
2657
2658 static const struct inode_operations proc_tgid_base_inode_operations = {
2659         .lookup         = proc_tgid_base_lookup,
2660         .getattr        = pid_getattr,
2661         .setattr        = proc_setattr,
2662         .permission     = proc_pid_permission,
2663 };
2664
2665 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2666 {
2667         struct dentry *dentry, *leader, *dir;
2668         char buf[PROC_NUMBUF];
2669         struct qstr name;
2670
2671         name.name = buf;
2672         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2673         /* no ->d_hash() rejects on procfs */
2674         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2675         if (dentry) {
2676                 shrink_dcache_parent(dentry);
2677                 d_drop(dentry);
2678                 dput(dentry);
2679         }
2680
2681         name.name = buf;
2682         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2683         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2684         if (!leader)
2685                 goto out;
2686
2687         name.name = "task";
2688         name.len = strlen(name.name);
2689         dir = d_hash_and_lookup(leader, &name);
2690         if (!dir)
2691                 goto out_put_leader;
2692
2693         name.name = buf;
2694         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2695         dentry = d_hash_and_lookup(dir, &name);
2696         if (dentry) {
2697                 shrink_dcache_parent(dentry);
2698                 d_drop(dentry);
2699                 dput(dentry);
2700         }
2701
2702         dput(dir);
2703 out_put_leader:
2704         dput(leader);
2705 out:
2706         return;
2707 }
2708
2709 /**
2710  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2711  * @task: task that should be flushed.
2712  *
2713  * When flushing dentries from proc, one needs to flush them from global
2714  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2715  * in. This call is supposed to do all of this job.
2716  *
2717  * Looks in the dcache for
2718  * /proc/@pid
2719  * /proc/@tgid/task/@pid
2720  * if either directory is present flushes it and all of it'ts children
2721  * from the dcache.
2722  *
2723  * It is safe and reasonable to cache /proc entries for a task until
2724  * that task exits.  After that they just clog up the dcache with
2725  * useless entries, possibly causing useful dcache entries to be
2726  * flushed instead.  This routine is proved to flush those useless
2727  * dcache entries at process exit time.
2728  *
2729  * NOTE: This routine is just an optimization so it does not guarantee
2730  *       that no dcache entries will exist at process exit time it
2731  *       just makes it very unlikely that any will persist.
2732  */
2733
2734 void proc_flush_task(struct task_struct *task)
2735 {
2736         int i;
2737         struct pid *pid, *tgid;
2738         struct upid *upid;
2739
2740         pid = task_pid(task);
2741         tgid = task_tgid(task);
2742
2743         for (i = 0; i <= pid->level; i++) {
2744                 upid = &pid->numbers[i];
2745                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2746                                         tgid->numbers[i].nr);
2747         }
2748 }
2749
2750 static int proc_pid_instantiate(struct inode *dir,
2751                                    struct dentry * dentry,
2752                                    struct task_struct *task, const void *ptr)
2753 {
2754         struct inode *inode;
2755
2756         inode = proc_pid_make_inode(dir->i_sb, task);
2757         if (!inode)
2758                 goto out;
2759
2760         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2761         inode->i_op = &proc_tgid_base_inode_operations;
2762         inode->i_fop = &proc_tgid_base_operations;
2763         inode->i_flags|=S_IMMUTABLE;
2764
2765         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
2766                                                   ARRAY_SIZE(tgid_base_stuff)));
2767
2768         d_set_d_op(dentry, &pid_dentry_operations);
2769
2770         d_add(dentry, inode);
2771         /* Close the race of the process dying before we return the dentry */
2772         if (pid_revalidate(dentry, 0))
2773                 return 0;
2774 out:
2775         return -ENOENT;
2776 }
2777
2778 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
2779 {
2780         int result = 0;
2781         struct task_struct *task;
2782         unsigned tgid;
2783         struct pid_namespace *ns;
2784
2785         tgid = name_to_int(dentry);
2786         if (tgid == ~0U)
2787                 goto out;
2788
2789         ns = dentry->d_sb->s_fs_info;
2790         rcu_read_lock();
2791         task = find_task_by_pid_ns(tgid, ns);
2792         if (task)
2793                 get_task_struct(task);
2794         rcu_read_unlock();
2795         if (!task)
2796                 goto out;
2797
2798         result = proc_pid_instantiate(dir, dentry, task, NULL);
2799         put_task_struct(task);
2800 out:
2801         return ERR_PTR(result);
2802 }
2803
2804 /*
2805  * Find the first task with tgid >= tgid
2806  *
2807  */
2808 struct tgid_iter {
2809         unsigned int tgid;
2810         struct task_struct *task;
2811 };
2812 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2813 {
2814         struct pid *pid;
2815
2816         if (iter.task)
2817                 put_task_struct(iter.task);
2818         rcu_read_lock();
2819 retry:
2820         iter.task = NULL;
2821         pid = find_ge_pid(iter.tgid, ns);
2822         if (pid) {
2823                 iter.tgid = pid_nr_ns(pid, ns);
2824                 iter.task = pid_task(pid, PIDTYPE_PID);
2825                 /* What we to know is if the pid we have find is the
2826                  * pid of a thread_group_leader.  Testing for task
2827                  * being a thread_group_leader is the obvious thing
2828                  * todo but there is a window when it fails, due to
2829                  * the pid transfer logic in de_thread.
2830                  *
2831                  * So we perform the straight forward test of seeing
2832                  * if the pid we have found is the pid of a thread
2833                  * group leader, and don't worry if the task we have
2834                  * found doesn't happen to be a thread group leader.
2835                  * As we don't care in the case of readdir.
2836                  */
2837                 if (!iter.task || !has_group_leader_pid(iter.task)) {
2838                         iter.tgid += 1;
2839                         goto retry;
2840                 }
2841                 get_task_struct(iter.task);
2842         }
2843         rcu_read_unlock();
2844         return iter;
2845 }
2846
2847 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 1)
2848
2849 /* for the /proc/ directory itself, after non-process stuff has been done */
2850 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
2851 {
2852         struct tgid_iter iter;
2853         struct pid_namespace *ns = file->f_dentry->d_sb->s_fs_info;
2854         loff_t pos = ctx->pos;
2855
2856         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
2857                 return 0;
2858
2859         if (pos == TGID_OFFSET - 1) {
2860                 struct inode *inode = ns->proc_self->d_inode;
2861                 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
2862                         return 0;
2863                 iter.tgid = 0;
2864         } else {
2865                 iter.tgid = pos - TGID_OFFSET;
2866         }
2867         iter.task = NULL;
2868         for (iter = next_tgid(ns, iter);
2869              iter.task;
2870              iter.tgid += 1, iter = next_tgid(ns, iter)) {
2871                 char name[PROC_NUMBUF];
2872                 int len;
2873                 if (!has_pid_permissions(ns, iter.task, 2))
2874                         continue;
2875
2876                 len = snprintf(name, sizeof(name), "%d", iter.tgid);
2877                 ctx->pos = iter.tgid + TGID_OFFSET;
2878                 if (!proc_fill_cache(file, ctx, name, len,
2879                                      proc_pid_instantiate, iter.task, NULL)) {
2880                         put_task_struct(iter.task);
2881                         return 0;
2882                 }
2883         }
2884         ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
2885         return 0;
2886 }
2887
2888 /*
2889  * Tasks
2890  */
2891 static const struct pid_entry tid_base_stuff[] = {
2892         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2893         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2894         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2895         REG("environ",   S_IRUSR, proc_environ_operations),
2896         INF("auxv",      S_IRUSR, proc_pid_auxv),
2897         ONE("status",    S_IRUGO, proc_pid_status),
2898         ONE("personality", S_IRUGO, proc_pid_personality),
2899         INF("limits",    S_IRUGO, proc_pid_limits),
2900 #ifdef CONFIG_SCHED_DEBUG
2901         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2902 #endif
2903         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2904 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2905         INF("syscall",   S_IRUGO, proc_pid_syscall),
2906 #endif
2907         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
2908         ONE("stat",      S_IRUGO, proc_tid_stat),
2909         ONE("statm",     S_IRUGO, proc_pid_statm),
2910         REG("maps",      S_IRUGO, proc_tid_maps_operations),
2911 #ifdef CONFIG_CHECKPOINT_RESTORE
2912         REG("children",  S_IRUGO, proc_tid_children_operations),
2913 #endif
2914 #ifdef CONFIG_NUMA
2915         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
2916 #endif
2917         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
2918         LNK("cwd",       proc_cwd_link),
2919         LNK("root",      proc_root_link),
2920         LNK("exe",       proc_exe_link),
2921         REG("mounts",    S_IRUGO, proc_mounts_operations),
2922         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2923 #ifdef CONFIG_PROC_PAGE_MONITOR
2924         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2925         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
2926         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
2927 #endif
2928 #ifdef CONFIG_SECURITY
2929         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2930 #endif
2931 #ifdef CONFIG_KALLSYMS
2932         INF("wchan",     S_IRUGO, proc_pid_wchan),
2933 #endif
2934 #ifdef CONFIG_STACKTRACE
2935         ONE("stack",      S_IRUGO, proc_pid_stack),
2936 #endif
2937 #ifdef CONFIG_SCHEDSTATS
2938         INF("schedstat", S_IRUGO, proc_pid_schedstat),
2939 #endif
2940 #ifdef CONFIG_LATENCYTOP
2941         REG("latency",  S_IRUGO, proc_lstats_operations),
2942 #endif
2943 #ifdef CONFIG_PROC_PID_CPUSET
2944         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
2945 #endif
2946 #ifdef CONFIG_CGROUPS
2947         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2948 #endif
2949         INF("oom_score", S_IRUGO, proc_oom_score),
2950         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2951         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2952 #ifdef CONFIG_AUDITSYSCALL
2953         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
2954         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2955 #endif
2956 #ifdef CONFIG_FAULT_INJECTION
2957         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2958 #endif
2959 #ifdef CONFIG_TASK_IO_ACCOUNTING
2960         INF("io",       S_IRUSR, proc_tid_io_accounting),
2961 #endif
2962 #ifdef CONFIG_HARDWALL
2963         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
2964 #endif
2965 #ifdef CONFIG_USER_NS
2966         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2967         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2968         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2969 #endif
2970 };
2971
2972 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
2973 {
2974         return proc_pident_readdir(file, ctx,
2975                                    tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
2976 }
2977
2978 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2979 {
2980         return proc_pident_lookup(dir, dentry,
2981                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
2982 }
2983
2984 static const struct file_operations proc_tid_base_operations = {
2985         .read           = generic_read_dir,
2986         .iterate        = proc_tid_base_readdir,
2987         .llseek         = default_llseek,
2988 };
2989
2990 static const struct inode_operations proc_tid_base_inode_operations = {
2991         .lookup         = proc_tid_base_lookup,
2992         .getattr        = pid_getattr,
2993         .setattr        = proc_setattr,
2994 };
2995
2996 static int proc_task_instantiate(struct inode *dir,
2997         struct dentry *dentry, struct task_struct *task, const void *ptr)
2998 {
2999         struct inode *inode;
3000         inode = proc_pid_make_inode(dir->i_sb, task);
3001
3002         if (!inode)
3003                 goto out;
3004         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3005         inode->i_op = &proc_tid_base_inode_operations;
3006         inode->i_fop = &proc_tid_base_operations;
3007         inode->i_flags|=S_IMMUTABLE;
3008
3009         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3010                                                   ARRAY_SIZE(tid_base_stuff)));
3011
3012         d_set_d_op(dentry, &pid_dentry_operations);
3013
3014         d_add(dentry, inode);
3015         /* Close the race of the process dying before we return the dentry */
3016         if (pid_revalidate(dentry, 0))
3017                 return 0;
3018 out:
3019         return -ENOENT;
3020 }
3021
3022 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3023 {
3024         int result = -ENOENT;
3025         struct task_struct *task;
3026         struct task_struct *leader = get_proc_task(dir);
3027         unsigned tid;
3028         struct pid_namespace *ns;
3029
3030         if (!leader)
3031                 goto out_no_task;
3032
3033         tid = name_to_int(dentry);
3034         if (tid == ~0U)
3035                 goto out;
3036
3037         ns = dentry->d_sb->s_fs_info;
3038         rcu_read_lock();
3039         task = find_task_by_pid_ns(tid, ns);
3040         if (task)
3041                 get_task_struct(task);
3042         rcu_read_unlock();
3043         if (!task)
3044                 goto out;
3045         if (!same_thread_group(leader, task))
3046                 goto out_drop_task;
3047
3048         result = proc_task_instantiate(dir, dentry, task, NULL);
3049 out_drop_task:
3050         put_task_struct(task);
3051 out:
3052         put_task_struct(leader);
3053 out_no_task:
3054         return ERR_PTR(result);
3055 }
3056
3057 /*
3058  * Find the first tid of a thread group to return to user space.
3059  *
3060  * Usually this is just the thread group leader, but if the users
3061  * buffer was too small or there was a seek into the middle of the
3062  * directory we have more work todo.
3063  *
3064  * In the case of a short read we start with find_task_by_pid.
3065  *
3066  * In the case of a seek we start with the leader and walk nr
3067  * threads past it.
3068  */
3069 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3070                                         struct pid_namespace *ns)
3071 {
3072         struct task_struct *pos, *task;
3073         unsigned long nr = f_pos;
3074
3075         if (nr != f_pos)        /* 32bit overflow? */
3076                 return NULL;
3077
3078         rcu_read_lock();
3079         task = pid_task(pid, PIDTYPE_PID);
3080         if (!task)
3081                 goto fail;
3082
3083         /* Attempt to start with the tid of a thread */
3084         if (tid && nr) {
3085                 pos = find_task_by_pid_ns(tid, ns);
3086                 if (pos && same_thread_group(pos, task))
3087                         goto found;
3088         }
3089
3090         /* If nr exceeds the number of threads there is nothing todo */
3091         if (nr >= get_nr_threads(task))
3092                 goto fail;
3093
3094         /* If we haven't found our starting place yet start
3095          * with the leader and walk nr threads forward.
3096          */
3097         pos = task = task->group_leader;
3098         do {
3099                 if (!nr--)
3100                         goto found;
3101         } while_each_thread(task, pos);
3102 fail:
3103         pos = NULL;
3104         goto out;
3105 found:
3106         get_task_struct(pos);
3107 out:
3108         rcu_read_unlock();
3109         return pos;
3110 }
3111
3112 /*
3113  * Find the next thread in the thread list.
3114  * Return NULL if there is an error or no next thread.
3115  *
3116  * The reference to the input task_struct is released.
3117  */
3118 static struct task_struct *next_tid(struct task_struct *start)
3119 {
3120         struct task_struct *pos = NULL;
3121         rcu_read_lock();
3122         if (pid_alive(start)) {
3123                 pos = next_thread(start);
3124                 if (thread_group_leader(pos))
3125                         pos = NULL;
3126                 else
3127                         get_task_struct(pos);
3128         }
3129         rcu_read_unlock();
3130         put_task_struct(start);
3131         return pos;
3132 }
3133
3134 /* for the /proc/TGID/task/ directories */
3135 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3136 {
3137         struct inode *inode = file_inode(file);
3138         struct task_struct *task;
3139         struct pid_namespace *ns;
3140         int tid;
3141
3142         if (proc_inode_is_dead(inode))
3143                 return -ENOENT;
3144
3145         if (!dir_emit_dots(file, ctx))
3146                 return 0;
3147
3148         /* f_version caches the tgid value that the last readdir call couldn't
3149          * return. lseek aka telldir automagically resets f_version to 0.
3150          */
3151         ns = file->f_dentry->d_sb->s_fs_info;
3152         tid = (int)file->f_version;
3153         file->f_version = 0;
3154         for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3155              task;
3156              task = next_tid(task), ctx->pos++) {
3157                 char name[PROC_NUMBUF];
3158                 int len;
3159                 tid = task_pid_nr_ns(task, ns);
3160                 len = snprintf(name, sizeof(name), "%d", tid);
3161                 if (!proc_fill_cache(file, ctx, name, len,
3162                                 proc_task_instantiate, task, NULL)) {
3163                         /* returning this tgid failed, save it as the first
3164                          * pid for the next readir call */
3165                         file->f_version = (u64)tid;
3166                         put_task_struct(task);
3167                         break;
3168                 }
3169         }
3170
3171         return 0;
3172 }
3173
3174 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3175 {
3176         struct inode *inode = dentry->d_inode;
3177         struct task_struct *p = get_proc_task(inode);
3178         generic_fillattr(inode, stat);
3179
3180         if (p) {
3181                 stat->nlink += get_nr_threads(p);
3182                 put_task_struct(p);
3183         }
3184
3185         return 0;
3186 }
3187
3188 static const struct inode_operations proc_task_inode_operations = {
3189         .lookup         = proc_task_lookup,
3190         .getattr        = proc_task_getattr,
3191         .setattr        = proc_setattr,
3192         .permission     = proc_pid_permission,
3193 };
3194
3195 static const struct file_operations proc_task_operations = {
3196         .read           = generic_read_dir,
3197         .iterate        = proc_task_readdir,
3198         .llseek         = default_llseek,
3199 };