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/cgroup.h>
  77 #include <linux/cpuset.h>
  78 #include <linux/audit.h>
  79 #include <linux/poll.h>
  80 #include <linux/nsproxy.h>
  81 #include <linux/oom.h>
  82 #include <linux/elf.h>
  83 #include <linux/pid_namespace.h>
  84 #include <linux/user_namespace.h>
  85 #include <linux/fs_struct.h>
  86 #include <linux/slab.h>
  87 #include <linux/flex_array.h>
  88 #ifdef CONFIG_HARDWALL
  89 #include <asm/hardwall.h>
  90 #endif
  91 #include <trace/events/oom.h>
  92 #include "internal.h"
  93 #include "fd.h"
  94
  95 /* NOTE:
  96  *      Implementing inode permission operations in /proc is almost
  97  *      certainly an error.  Permission checks need to happen during
  98  *      each system call not at open time.  The reason is that most of
  99  *      what we wish to check for permissions in /proc varies at runtime.
 100  *
 101  *      The classic example of a problem is opening file descriptors
 102  *      in /proc for a task before it execs a suid executable.
 103  */
 104
 105 struct pid_entry {
 106         char *name;
 107         int len;
 108         umode_t mode;
 109         const struct inode_operations *iop;
 110         const struct file_operations *fop;
 111         union proc_op op;
 112 };
 113
 114 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 115         .name = (NAME),                                 \
 116         .len  = sizeof(NAME) - 1,                       \
 117         .mode = MODE,                                   \
 118         .iop  = IOP,                                    \
 119         .fop  = FOP,                                    \
 120         .op   = OP,                                     \
 121 }
 122
 123 #define DIR(NAME, MODE, iops, fops)     \
 124         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 125 #define LNK(NAME, get_link)                                     \
 126         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 127                 &proc_pid_link_inode_operations, NULL,          \
 128                 { .proc_get_link = get_link } )
 129 #define REG(NAME, MODE, fops)                           \
 130         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 131 #define INF(NAME, MODE, read)                           \
 132         NOD(NAME, (S_IFREG|(MODE)),                     \
 133                 NULL, &proc_info_file_operations,       \
 134                 { .proc_read = read } )
 135 #define ONE(NAME, MODE, show)                           \
 136         NOD(NAME, (S_IFREG|(MODE)),                     \
 137                 NULL, &proc_single_file_operations,     \
 138                 { .proc_show = show } )
 139
 140 /*
 141  * Count the number of hardlinks for the pid_entry table, excluding the .
 142  * and .. links.
 143  */
 144 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
 145         unsigned int n)
 146 {
 147         unsigned int i;
 148         unsigned int count;
 149
 150         count = 0;
 151         for (i = 0; i < n; ++i) {
 152                 if (S_ISDIR(entries[i].mode))
 153                         ++count;
 154         }
 155
 156         return count;
 157 }
 158
 159 static int get_task_root(struct task_struct *task, struct path *root)
 160 {
 161         int result = -ENOENT;
 162
 163         task_lock(task);
 164         if (task->fs) {
 165                 get_fs_root(task->fs, root);
 166                 result = 0;
 167         }
 168         task_unlock(task);
 169         return result;
 170 }
 171
 172 static int proc_cwd_link(struct dentry *dentry, struct path *path)
 173 {
 174         struct task_struct *task = get_proc_task(dentry->d_inode);
 175         int result = -ENOENT;
 176
 177         if (task) {
 178                 task_lock(task);
 179                 if (task->fs) {
 180                         get_fs_pwd(task->fs, path);
 181                         result = 0;
 182                 }
 183                 task_unlock(task);
 184                 put_task_struct(task);
 185         }
 186         return result;
 187 }
 188
 189 static int proc_root_link(struct dentry *dentry, struct path *path)
 190 {
 191         struct task_struct *task = get_proc_task(dentry->d_inode);
 192         int result = -ENOENT;
 193
 194         if (task) {
 195                 result = get_task_root(task, path);
 196                 put_task_struct(task);
 197         }
 198         return result;
 199 }
 200
 201 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
 202 {
 203         int res = 0;
 204         unsigned int len;
 205         struct mm_struct *mm = get_task_mm(task);
 206         if (!mm)
 207                 goto out;
 208         if (!mm->arg_end)
 209                 goto out_mm;    /* Shh! No looking before we're done */
 210
 211         len = mm->arg_end - mm->arg_start;
 212
 213         if (len > PAGE_SIZE)
 214                 len = PAGE_SIZE;
 215
 216         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
 217
 218         // If the nul at the end of args has been overwritten, then
 219         // assume application is using setproctitle(3).
 220         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
 221                 len = strnlen(buffer, res);
 222                 if (len < res) {
 223                     res = len;
 224                 } else {
 225                         len = mm->env_end - mm->env_start;
 226                         if (len > PAGE_SIZE - res)
 227                                 len = PAGE_SIZE - res;
 228                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
 229                         res = strnlen(buffer, res);
 230                 }
 231         }
 232 out_mm:
 233         mmput(mm);
 234 out:
 235         return res;
 236 }
 237
 238 static int proc_pid_auxv(struct task_struct *task, char *buffer)
 239 {
 240         struct mm_struct *mm = mm_access(task, PTRACE_MODE_READ);
 241         int res = PTR_ERR(mm);
 242         if (mm && !IS_ERR(mm)) {
 243                 unsigned int nwords = 0;
 244                 do {
 245                         nwords += 2;
 246                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 247                 res = nwords * sizeof(mm->saved_auxv[0]);
 248                 if (res > PAGE_SIZE)
 249                         res = PAGE_SIZE;
 250                 memcpy(buffer, mm->saved_auxv, res);
 251                 mmput(mm);
 252         }
 253         return res;
 254 }
 255
 256
 257 #ifdef CONFIG_KALLSYMS
 258 /*
 259  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 260  * Returns the resolved symbol.  If that fails, simply return the address.
 261  */
 262 static int proc_pid_wchan(struct task_struct *task, char *buffer)
 263 {
 264         unsigned long wchan;
 265         char symname[KSYM_NAME_LEN];
 266
 267         wchan = get_wchan(task);
 268
 269         if (lookup_symbol_name(wchan, symname) < 0)
 270                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
 271                         return 0;
 272                 else
 273                         return sprintf(buffer, "%lu", wchan);
 274         else
 275                 return sprintf(buffer, "%s", symname);
 276 }
 277 #endif /* CONFIG_KALLSYMS */
 278
 279 static int lock_trace(struct task_struct *task)
 280 {
 281         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
 282         if (err)
 283                 return err;
 284         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH)) {
 285                 mutex_unlock(&task->signal->cred_guard_mutex);
 286                 return -EPERM;
 287         }
 288         return 0;
 289 }
 290
 291 static void unlock_trace(struct task_struct *task)
 292 {
 293         mutex_unlock(&task->signal->cred_guard_mutex);
 294 }
 295
 296 #ifdef CONFIG_STACKTRACE
 297
 298 #define MAX_STACK_TRACE_DEPTH   64
 299
 300 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 301                           struct pid *pid, struct task_struct *task)
 302 {
 303         struct stack_trace trace;
 304         unsigned long *entries;
 305         int err;
 306         int i;
 307
 308         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 309         if (!entries)
 310                 return -ENOMEM;
 311
 312         trace.nr_entries        = 0;
 313         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 314         trace.entries           = entries;
 315         trace.skip              = 0;
 316
 317         err = lock_trace(task);
 318         if (!err) {
 319                 save_stack_trace_tsk(task, &trace);
 320
 321                 for (i = 0; i < trace.nr_entries; i++) {
 322                         seq_printf(m, "[<%pK>] %pS\n",
 323                                    (void *)entries[i], (void *)entries[i]);
 324                 }
 325                 unlock_trace(task);
 326         }
 327         kfree(entries);
 328
 329         return err;
 330 }
 331 #endif
 332
 333 #ifdef CONFIG_SCHEDSTATS
 334 /*
 335  * Provides /proc/PID/schedstat
 336  */
 337 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
 338 {
 339         return sprintf(buffer, "%llu %llu %lu\n",
 340                         (unsigned long long)task->se.sum_exec_runtime,
 341                         (unsigned long long)task->sched_info.run_delay,
 342                         task->sched_info.pcount);
 343 }
 344 #endif
 345
 346 #ifdef CONFIG_LATENCYTOP
 347 static int lstats_show_proc(struct seq_file *m, void *v)
 348 {
 349         int i;
 350         struct inode *inode = m->private;
 351         struct task_struct *task = get_proc_task(inode);
 352
 353         if (!task)
 354                 return -ESRCH;
 355         seq_puts(m, "Latency Top version : v0.1\n");
 356         for (i = 0; i < 32; i++) {
 357                 struct latency_record *lr = &task->latency_record[i];
 358                 if (lr->backtrace[0]) {
 359                         int q;
 360                         seq_printf(m, "%i %li %li",
 361                                    lr->count, lr->time, lr->max);
 362                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 363                                 unsigned long bt = lr->backtrace[q];
 364                                 if (!bt)
 365                                         break;
 366                                 if (bt == ULONG_MAX)
 367                                         break;
 368                                 seq_printf(m, " %ps", (void *)bt);
 369                         }
 370                         seq_putc(m, '\n');
 371                 }
 372
 373         }
 374         put_task_struct(task);
 375         return 0;
 376 }
 377
 378 static int lstats_open(struct inode *inode, struct file *file)
 379 {
 380         return single_open(file, lstats_show_proc, inode);
 381 }
 382
 383 static ssize_t lstats_write(struct file *file, const char __user *buf,
 384                             size_t count, loff_t *offs)
 385 {
 386         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 387
 388         if (!task)
 389                 return -ESRCH;
 390         clear_all_latency_tracing(task);
 391         put_task_struct(task);
 392
 393         return count;
 394 }
 395
 396 static const struct file_operations proc_lstats_operations = {
 397         .open           = lstats_open,
 398         .read           = seq_read,
 399         .write          = lstats_write,
 400         .llseek         = seq_lseek,
 401         .release        = single_release,
 402 };
 403
 404 #endif
 405
 406 static int proc_oom_score(struct task_struct *task, char *buffer)
 407 {
 408         unsigned long totalpages = totalram_pages + total_swap_pages;
 409         unsigned long points = 0;
 410
 411         read_lock(&tasklist_lock);
 412         if (pid_alive(task))
 413                 points = oom_badness(task, NULL, NULL, totalpages) *
 414                                                 1000 / totalpages;
 415         read_unlock(&tasklist_lock);
 416         return sprintf(buffer, "%lu\n", points);
 417 }
 418
 419 struct limit_names {
 420         char *name;
 421         char *unit;
 422 };
 423
 424 static const struct limit_names lnames[RLIM_NLIMITS] = {
 425         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 426         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 427         [RLIMIT_DATA] = {"Max data size", "bytes"},
 428         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 429         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 430         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 431         [RLIMIT_NPROC] = {"Max processes", "processes"},
 432         [RLIMIT_NOFILE] = {"Max open files", "files"},
 433         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 434         [RLIMIT_AS] = {"Max address space", "bytes"},
 435         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 436         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 437         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 438         [RLIMIT_NICE] = {"Max nice priority", NULL},
 439         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 440         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 441 };
 442
 443 /* Display limits for a process */
 444 static int proc_pid_limits(struct task_struct *task, char *buffer)
 445 {
 446         unsigned int i;
 447         int count = 0;
 448         unsigned long flags;
 449         char *bufptr = buffer;
 450
 451         struct rlimit rlim[RLIM_NLIMITS];
 452
 453         if (!lock_task_sighand(task, &flags))
 454                 return 0;
 455         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 456         unlock_task_sighand(task, &flags);
 457
 458         /*
 459          * print the file header
 460          */
 461         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
 462                         "Limit", "Soft Limit", "Hard Limit", "Units");
 463
 464         for (i = 0; i < RLIM_NLIMITS; i++) {
 465                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 466                         count += sprintf(&bufptr[count], "%-25s %-20s ",
 467                                          lnames[i].name, "unlimited");
 468                 else
 469                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
 470                                          lnames[i].name, rlim[i].rlim_cur);
 471
 472                 if (rlim[i].rlim_max == RLIM_INFINITY)
 473                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
 474                 else
 475                         count += sprintf(&bufptr[count], "%-20lu ",
 476                                          rlim[i].rlim_max);
 477
 478                 if (lnames[i].unit)
 479                         count += sprintf(&bufptr[count], "%-10s\n",
 480                                          lnames[i].unit);
 481                 else
 482                         count += sprintf(&bufptr[count], "\n");
 483         }
 484
 485         return count;
 486 }
 487
 488 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 489 static int proc_pid_syscall(struct task_struct *task, char *buffer)
 490 {
 491         long nr;
 492         unsigned long args[6], sp, pc;
 493         int res = lock_trace(task);
 494         if (res)
 495                 return res;
 496
 497         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 498                 res = sprintf(buffer, "running\n");
 499         else if (nr < 0)
 500                 res = sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 501         else
 502                 res = sprintf(buffer,
 503                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 504                        nr,
 505                        args[0], args[1], args[2], args[3], args[4], args[5],
 506                        sp, pc);
 507         unlock_trace(task);
 508         return res;
 509 }
 510 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 511
 512 /************************************************************************/
 513 /*                       Here the fs part begins                        */
 514 /************************************************************************/
 515
 516 /* permission checks */
 517 static int proc_fd_access_allowed(struct inode *inode)
 518 {
 519         struct task_struct *task;
 520         int allowed = 0;
 521         /* Allow access to a task's file descriptors if it is us or we
 522          * may use ptrace attach to the process and find out that
 523          * information.
 524          */
 525         task = get_proc_task(inode);
 526         if (task) {
 527                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
 528                 put_task_struct(task);
 529         }
 530         return allowed;
 531 }
 532
 533 int proc_setattr(struct dentry *dentry, struct iattr *attr)
 534 {
 535         int error;
 536         struct inode *inode = dentry->d_inode;
 537
 538         if (attr->ia_valid & ATTR_MODE)
 539                 return -EPERM;
 540
 541         error = inode_change_ok(inode, attr);
 542         if (error)
 543                 return error;
 544
 545         if ((attr->ia_valid & ATTR_SIZE) &&
 546             attr->ia_size != i_size_read(inode)) {
 547                 error = vmtruncate(inode, attr->ia_size);
 548                 if (error)
 549                         return error;
 550         }
 551
 552         setattr_copy(inode, attr);
 553         mark_inode_dirty(inode);
 554         return 0;
 555 }
 556
 557 /*
 558  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
 559  * or euid/egid (for hide_pid_min=2)?
 560  */
 561 static bool has_pid_permissions(struct pid_namespace *pid,
 562                                  struct task_struct *task,
 563                                  int hide_pid_min)
 564 {
 565         if (pid->hide_pid < hide_pid_min)
 566                 return true;
 567         if (in_group_p(pid->pid_gid))
 568                 return true;
 569         return ptrace_may_access(task, PTRACE_MODE_READ);
 570 }
 571
 572
 573 static int proc_pid_permission(struct inode *inode, int mask)
 574 {
 575         struct pid_namespace *pid = inode->i_sb->s_fs_info;
 576         struct task_struct *task;
 577         bool has_perms;
 578
 579         task = get_proc_task(inode);
 580         if (!task)
 581                 return -ESRCH;
 582         has_perms = has_pid_permissions(pid, task, 1);
 583         put_task_struct(task);
 584
 585         if (!has_perms) {
 586                 if (pid->hide_pid == 2) {
 587                         /*
 588                          * Let's make getdents(), stat(), and open()
 589                          * consistent with each other.  If a process
 590                          * may not stat() a file, it shouldn't be seen
 591                          * in procfs at all.
 592                          */
 593                         return -ENOENT;
 594                 }
 595
 596                 return -EPERM;
 597         }
 598         return generic_permission(inode, mask);
 599 }
 600
 601
 602
 603 static const struct inode_operations proc_def_inode_operations = {
 604         .setattr        = proc_setattr,
 605 };
 606
 607 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 608
 609 static ssize_t proc_info_read(struct file * file, char __user * buf,
 610                           size_t count, loff_t *ppos)
 611 {
 612         struct inode * inode = file->f_path.dentry->d_inode;
 613         unsigned long page;
 614         ssize_t length;
 615         struct task_struct *task = get_proc_task(inode);
 616
 617         length = -ESRCH;
 618         if (!task)
 619                 goto out_no_task;
 620
 621         if (count > PROC_BLOCK_SIZE)
 622                 count = PROC_BLOCK_SIZE;
 623
 624         length = -ENOMEM;
 625         if (!(page = __get_free_page(GFP_TEMPORARY)))
 626                 goto out;
 627
 628         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 629
 630         if (length >= 0)
 631                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 632         free_page(page);
 633 out:
 634         put_task_struct(task);
 635 out_no_task:
 636         return length;
 637 }
 638
 639 static const struct file_operations proc_info_file_operations = {
 640         .read           = proc_info_read,
 641         .llseek         = generic_file_llseek,
 642 };
 643
 644 static int proc_single_show(struct seq_file *m, void *v)
 645 {
 646         struct inode *inode = m->private;
 647         struct pid_namespace *ns;
 648         struct pid *pid;
 649         struct task_struct *task;
 650         int ret;
 651
 652         ns = inode->i_sb->s_fs_info;
 653         pid = proc_pid(inode);
 654         task = get_pid_task(pid, PIDTYPE_PID);
 655         if (!task)
 656                 return -ESRCH;
 657
 658         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 659
 660         put_task_struct(task);
 661         return ret;
 662 }
 663
 664 static int proc_single_open(struct inode *inode, struct file *filp)
 665 {
 666         return single_open(filp, proc_single_show, inode);
 667 }
 668
 669 static const struct file_operations proc_single_file_operations = {
 670         .open           = proc_single_open,
 671         .read           = seq_read,
 672         .llseek         = seq_lseek,
 673         .release        = single_release,
 674 };
 675
 676 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
 677 {
 678         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 679         struct mm_struct *mm;
 680
 681         if (!task)
 682                 return -ESRCH;
 683
 684         mm = mm_access(task, mode);
 685         put_task_struct(task);
 686
 687         if (IS_ERR(mm))
 688                 return PTR_ERR(mm);
 689
 690         if (mm) {
 691                 /* ensure this mm_struct can't be freed */
 692                 atomic_inc(&mm->mm_count);
 693                 /* but do not pin its memory */
 694                 mmput(mm);
 695         }
 696
 697         file->private_data = mm;
 698
 699         return 0;
 700 }
 701
 702 static int mem_open(struct inode *inode, struct file *file)
 703 {
 704         int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
 705
 706         /* OK to pass negative loff_t, we can catch out-of-range */
 707         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 708
 709         return ret;
 710 }
 711
 712 static ssize_t mem_rw(struct file *file, char __user *buf,
 713                         size_t count, loff_t *ppos, int write)
 714 {
 715         struct mm_struct *mm = file->private_data;
 716         unsigned long addr = *ppos;
 717         ssize_t copied;
 718         char *page;
 719
 720         if (!mm)
 721                 return 0;
 722
 723         page = (char *)__get_free_page(GFP_TEMPORARY);
 724         if (!page)
 725                 return -ENOMEM;
 726
 727         copied = 0;
 728         if (!atomic_inc_not_zero(&mm->mm_users))
 729                 goto free;
 730
 731         while (count > 0) {
 732                 int this_len = min_t(int, count, PAGE_SIZE);
 733
 734                 if (write && copy_from_user(page, buf, this_len)) {
 735                         copied = -EFAULT;
 736                         break;
 737                 }
 738
 739                 this_len = access_remote_vm(mm, addr, page, this_len, write);
 740                 if (!this_len) {
 741                         if (!copied)
 742                                 copied = -EIO;
 743                         break;
 744                 }
 745
 746                 if (!write && copy_to_user(buf, page, this_len)) {
 747                         copied = -EFAULT;
 748                         break;
 749                 }
 750
 751                 buf += this_len;
 752                 addr += this_len;
 753                 copied += this_len;
 754                 count -= this_len;
 755         }
 756         *ppos = addr;
 757
 758         mmput(mm);
 759 free:
 760         free_page((unsigned long) page);
 761         return copied;
 762 }
 763
 764 static ssize_t mem_read(struct file *file, char __user *buf,
 765                         size_t count, loff_t *ppos)
 766 {
 767         return mem_rw(file, buf, count, ppos, 0);
 768 }
 769
 770 static ssize_t mem_write(struct file *file, const char __user *buf,
 771                          size_t count, loff_t *ppos)
 772 {
 773         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 774 }
 775
 776 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 777 {
 778         switch (orig) {
 779         case 0:
 780                 file->f_pos = offset;
 781                 break;
 782         case 1:
 783                 file->f_pos += offset;
 784                 break;
 785         default:
 786                 return -EINVAL;
 787         }
 788         force_successful_syscall_return();
 789         return file->f_pos;
 790 }
 791
 792 static int mem_release(struct inode *inode, struct file *file)
 793 {
 794         struct mm_struct *mm = file->private_data;
 795         if (mm)
 796                 mmdrop(mm);
 797         return 0;
 798 }
 799
 800 static const struct file_operations proc_mem_operations = {
 801         .llseek         = mem_lseek,
 802         .read           = mem_read,
 803         .write          = mem_write,
 804         .open           = mem_open,
 805         .release        = mem_release,
 806 };
 807
 808 static int environ_open(struct inode *inode, struct file *file)
 809 {
 810         return __mem_open(inode, file, PTRACE_MODE_READ);
 811 }
 812
 813 static ssize_t environ_read(struct file *file, char __user *buf,
 814                         size_t count, loff_t *ppos)
 815 {
 816         char *page;
 817         unsigned long src = *ppos;
 818         int ret = 0;
 819         struct mm_struct *mm = file->private_data;
 820
 821         if (!mm)
 822                 return 0;
 823
 824         page = (char *)__get_free_page(GFP_TEMPORARY);
 825         if (!page)
 826                 return -ENOMEM;
 827
 828         ret = 0;
 829         if (!atomic_inc_not_zero(&mm->mm_users))
 830                 goto free;
 831         while (count > 0) {
 832                 size_t this_len, max_len;
 833                 int retval;
 834
 835                 if (src >= (mm->env_end - mm->env_start))
 836                         break;
 837
 838                 this_len = mm->env_end - (mm->env_start + src);
 839
 840                 max_len = min_t(size_t, PAGE_SIZE, count);
 841                 this_len = min(max_len, this_len);
 842
 843                 retval = access_remote_vm(mm, (mm->env_start + src),
 844                         page, this_len, 0);
 845
 846                 if (retval <= 0) {
 847                         ret = retval;
 848                         break;
 849                 }
 850
 851                 if (copy_to_user(buf, page, retval)) {
 852                         ret = -EFAULT;
 853                         break;
 854                 }
 855
 856                 ret += retval;
 857                 src += retval;
 858                 buf += retval;
 859                 count -= retval;
 860         }
 861         *ppos = src;
 862         mmput(mm);
 863
 864 free:
 865         free_page((unsigned long) page);
 866         return ret;
 867 }
 868
 869 static const struct file_operations proc_environ_operations = {
 870         .open           = environ_open,
 871         .read           = environ_read,
 872         .llseek         = generic_file_llseek,
 873         .release        = mem_release,
 874 };
 875
 876 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
 877                                         size_t count, loff_t *ppos)
 878 {
 879         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 880         char buffer[PROC_NUMBUF];
 881         int oom_score_adj = OOM_SCORE_ADJ_MIN;
 882         unsigned long flags;
 883         size_t len;
 884
 885         if (!task)
 886                 return -ESRCH;
 887         if (lock_task_sighand(task, &flags)) {
 888                 oom_score_adj = task->signal->oom_score_adj;
 889                 unlock_task_sighand(task, &flags);
 890         }
 891         put_task_struct(task);
 892         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_score_adj);
 893         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 894 }
 895
 896 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
 897                                         size_t count, loff_t *ppos)
 898 {
 899         struct task_struct *task;
 900         char buffer[PROC_NUMBUF];
 901         unsigned long flags;
 902         int oom_score_adj;
 903         int err;
 904
 905         memset(buffer, 0, sizeof(buffer));
 906         if (count > sizeof(buffer) - 1)
 907                 count = sizeof(buffer) - 1;
 908         if (copy_from_user(buffer, buf, count)) {
 909                 err = -EFAULT;
 910                 goto out;
 911         }
 912
 913         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
 914         if (err)
 915                 goto out;
 916         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
 917                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
 918                 err = -EINVAL;
 919                 goto out;
 920         }
 921
 922         task = get_proc_task(file->f_path.dentry->d_inode);
 923         if (!task) {
 924                 err = -ESRCH;
 925                 goto out;
 926         }
 927
 928         task_lock(task);
 929         if (!task->mm) {
 930                 err = -EINVAL;
 931                 goto err_task_lock;
 932         }
 933
 934         if (!lock_task_sighand(task, &flags)) {
 935                 err = -ESRCH;
 936                 goto err_task_lock;
 937         }
 938
 939         if (oom_score_adj < task->signal->oom_score_adj_min &&
 940                         !capable(CAP_SYS_RESOURCE)) {
 941                 err = -EACCES;
 942                 goto err_sighand;
 943         }
 944
 945         task->signal->oom_score_adj = oom_score_adj;
 946         if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
 947                 task->signal->oom_score_adj_min = oom_score_adj;
 948         trace_oom_score_adj_update(task);
 949
 950 err_sighand:
 951         unlock_task_sighand(task, &flags);
 952 err_task_lock:
 953         task_unlock(task);
 954         put_task_struct(task);
 955 out:
 956         return err < 0 ? err : count;
 957 }
 958
 959 static const struct file_operations proc_oom_score_adj_operations = {
 960         .read           = oom_score_adj_read,
 961         .write          = oom_score_adj_write,
 962         .llseek         = default_llseek,
 963 };
 964
 965 #ifdef CONFIG_AUDITSYSCALL
 966 #define TMPBUFLEN 21
 967 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
 968                                   size_t count, loff_t *ppos)
 969 {
 970         struct inode * inode = file->f_path.dentry->d_inode;
 971         struct task_struct *task = get_proc_task(inode);
 972         ssize_t length;
 973         char tmpbuf[TMPBUFLEN];
 974
 975         if (!task)
 976                 return -ESRCH;
 977         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
 978                            from_kuid(file->f_cred->user_ns,
 979                                      audit_get_loginuid(task)));
 980         put_task_struct(task);
 981         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
 982 }
 983
 984 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
 985                                    size_t count, loff_t *ppos)
 986 {
 987         struct inode * inode = file->f_path.dentry->d_inode;
 988         char *page, *tmp;
 989         ssize_t length;
 990         uid_t loginuid;
 991         kuid_t kloginuid;
 992
 993         rcu_read_lock();
 994         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
 995                 rcu_read_unlock();
 996                 return -EPERM;
 997         }
 998         rcu_read_unlock();
 999
1000         if (count >= PAGE_SIZE)
1001                 count = PAGE_SIZE - 1;
1002
1003         if (*ppos != 0) {
1004                 /* No partial writes. */
1005                 return -EINVAL;
1006         }
1007         page = (char*)__get_free_page(GFP_TEMPORARY);
1008         if (!page)
1009                 return -ENOMEM;
1010         length = -EFAULT;
1011         if (copy_from_user(page, buf, count))
1012                 goto out_free_page;
1013
1014         page[count] = '\0';
1015         loginuid = simple_strtoul(page, &tmp, 10);
1016         if (tmp == page) {
1017                 length = -EINVAL;
1018                 goto out_free_page;
1019
1020         }
1021         kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1022         if (!uid_valid(kloginuid)) {
1023                 length = -EINVAL;
1024                 goto out_free_page;
1025         }
1026
1027         length = audit_set_loginuid(kloginuid);
1028         if (likely(length == 0))
1029                 length = count;
1030
1031 out_free_page:
1032         free_page((unsigned long) page);
1033         return length;
1034 }
1035
1036 static const struct file_operations proc_loginuid_operations = {
1037         .read           = proc_loginuid_read,
1038         .write          = proc_loginuid_write,
1039         .llseek         = generic_file_llseek,
1040 };
1041
1042 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1043                                   size_t count, loff_t *ppos)
1044 {
1045         struct inode * inode = file->f_path.dentry->d_inode;
1046         struct task_struct *task = get_proc_task(inode);
1047         ssize_t length;
1048         char tmpbuf[TMPBUFLEN];
1049
1050         if (!task)
1051                 return -ESRCH;
1052         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1053                                 audit_get_sessionid(task));
1054         put_task_struct(task);
1055         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1056 }
1057
1058 static const struct file_operations proc_sessionid_operations = {
1059         .read           = proc_sessionid_read,
1060         .llseek         = generic_file_llseek,
1061 };
1062 #endif
1063
1064 #ifdef CONFIG_FAULT_INJECTION
1065 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1066                                       size_t count, loff_t *ppos)
1067 {
1068         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1069         char buffer[PROC_NUMBUF];
1070         size_t len;
1071         int make_it_fail;
1072
1073         if (!task)
1074                 return -ESRCH;
1075         make_it_fail = task->make_it_fail;
1076         put_task_struct(task);
1077
1078         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1079
1080         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1081 }
1082
1083 static ssize_t proc_fault_inject_write(struct file * file,
1084                         const char __user * buf, size_t count, loff_t *ppos)
1085 {
1086         struct task_struct *task;
1087         char buffer[PROC_NUMBUF], *end;
1088         int make_it_fail;
1089
1090         if (!capable(CAP_SYS_RESOURCE))
1091                 return -EPERM;
1092         memset(buffer, 0, sizeof(buffer));
1093         if (count > sizeof(buffer) - 1)
1094                 count = sizeof(buffer) - 1;
1095         if (copy_from_user(buffer, buf, count))
1096                 return -EFAULT;
1097         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1098         if (*end)
1099                 return -EINVAL;
1100         task = get_proc_task(file->f_dentry->d_inode);
1101         if (!task)
1102                 return -ESRCH;
1103         task->make_it_fail = make_it_fail;
1104         put_task_struct(task);
1105
1106         return count;
1107 }
1108
1109 static const struct file_operations proc_fault_inject_operations = {
1110         .read           = proc_fault_inject_read,
1111         .write          = proc_fault_inject_write,
1112         .llseek         = generic_file_llseek,
1113 };
1114 #endif
1115
1116
1117 #ifdef CONFIG_SCHED_DEBUG
1118 /*
1119  * Print out various scheduling related per-task fields:
1120  */
1121 static int sched_show(struct seq_file *m, void *v)
1122 {
1123         struct inode *inode = m->private;
1124         struct task_struct *p;
1125
1126         p = get_proc_task(inode);
1127         if (!p)
1128                 return -ESRCH;
1129         proc_sched_show_task(p, m);
1130
1131         put_task_struct(p);
1132
1133         return 0;
1134 }
1135
1136 static ssize_t
1137 sched_write(struct file *file, const char __user *buf,
1138             size_t count, loff_t *offset)
1139 {
1140         struct inode *inode = file->f_path.dentry->d_inode;
1141         struct task_struct *p;
1142
1143         p = get_proc_task(inode);
1144         if (!p)
1145                 return -ESRCH;
1146         proc_sched_set_task(p);
1147
1148         put_task_struct(p);
1149
1150         return count;
1151 }
1152
1153 static int sched_open(struct inode *inode, struct file *filp)
1154 {
1155         return single_open(filp, sched_show, inode);
1156 }
1157
1158 static const struct file_operations proc_pid_sched_operations = {
1159         .open           = sched_open,
1160         .read           = seq_read,
1161         .write          = sched_write,
1162         .llseek         = seq_lseek,
1163         .release        = single_release,
1164 };
1165
1166 #endif
1167
1168 #ifdef CONFIG_SCHED_AUTOGROUP
1169 /*
1170  * Print out autogroup related information:
1171  */
1172 static int sched_autogroup_show(struct seq_file *m, void *v)
1173 {
1174         struct inode *inode = m->private;
1175         struct task_struct *p;
1176
1177         p = get_proc_task(inode);
1178         if (!p)
1179                 return -ESRCH;
1180         proc_sched_autogroup_show_task(p, m);
1181
1182         put_task_struct(p);
1183
1184         return 0;
1185 }
1186
1187 static ssize_t
1188 sched_autogroup_write(struct file *file, const char __user *buf,
1189             size_t count, loff_t *offset)
1190 {
1191         struct inode *inode = file->f_path.dentry->d_inode;
1192         struct task_struct *p;
1193         char buffer[PROC_NUMBUF];
1194         int nice;
1195         int err;
1196
1197         memset(buffer, 0, sizeof(buffer));
1198         if (count > sizeof(buffer) - 1)
1199                 count = sizeof(buffer) - 1;
1200         if (copy_from_user(buffer, buf, count))
1201                 return -EFAULT;
1202
1203         err = kstrtoint(strstrip(buffer), 0, &nice);
1204         if (err < 0)
1205                 return err;
1206
1207         p = get_proc_task(inode);
1208         if (!p)
1209                 return -ESRCH;
1210
1211         err = proc_sched_autogroup_set_nice(p, nice);
1212         if (err)
1213                 count = err;
1214
1215         put_task_struct(p);
1216
1217         return count;
1218 }
1219
1220 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1221 {
1222         int ret;
1223
1224         ret = single_open(filp, sched_autogroup_show, NULL);
1225         if (!ret) {
1226                 struct seq_file *m = filp->private_data;
1227
1228                 m->private = inode;
1229         }
1230         return ret;
1231 }
1232
1233 static const struct file_operations proc_pid_sched_autogroup_operations = {
1234         .open           = sched_autogroup_open,
1235         .read           = seq_read,
1236         .write          = sched_autogroup_write,
1237         .llseek         = seq_lseek,
1238         .release        = single_release,
1239 };
1240
1241 #endif /* CONFIG_SCHED_AUTOGROUP */
1242
1243 static ssize_t comm_write(struct file *file, const char __user *buf,
1244                                 size_t count, loff_t *offset)
1245 {
1246         struct inode *inode = file->f_path.dentry->d_inode;
1247         struct task_struct *p;
1248         char buffer[TASK_COMM_LEN];
1249
1250         memset(buffer, 0, sizeof(buffer));
1251         if (count > sizeof(buffer) - 1)
1252                 count = sizeof(buffer) - 1;
1253         if (copy_from_user(buffer, buf, count))
1254                 return -EFAULT;
1255
1256         p = get_proc_task(inode);
1257         if (!p)
1258                 return -ESRCH;
1259
1260         if (same_thread_group(current, p))
1261                 set_task_comm(p, buffer);
1262         else
1263                 count = -EINVAL;
1264
1265         put_task_struct(p);
1266
1267         return count;
1268 }
1269
1270 static int comm_show(struct seq_file *m, void *v)
1271 {
1272         struct inode *inode = m->private;
1273         struct task_struct *p;
1274
1275         p = get_proc_task(inode);
1276         if (!p)
1277                 return -ESRCH;
1278
1279         task_lock(p);
1280         seq_printf(m, "%s\n", p->comm);
1281         task_unlock(p);
1282
1283         put_task_struct(p);
1284
1285         return 0;
1286 }
1287
1288 static int comm_open(struct inode *inode, struct file *filp)
1289 {
1290         return single_open(filp, comm_show, inode);
1291 }
1292
1293 static const struct file_operations proc_pid_set_comm_operations = {
1294         .open           = comm_open,
1295         .read           = seq_read,
1296         .write          = comm_write,
1297         .llseek         = seq_lseek,
1298         .release        = single_release,
1299 };
1300
1301 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1302 {
1303         struct task_struct *task;
1304         struct mm_struct *mm;
1305         struct file *exe_file;
1306
1307         task = get_proc_task(dentry->d_inode);
1308         if (!task)
1309                 return -ENOENT;
1310         mm = get_task_mm(task);
1311         put_task_struct(task);
1312         if (!mm)
1313                 return -ENOENT;
1314         exe_file = get_mm_exe_file(mm);
1315         mmput(mm);
1316         if (exe_file) {
1317                 *exe_path = exe_file->f_path;
1318                 path_get(&exe_file->f_path);
1319                 fput(exe_file);
1320                 return 0;
1321         } else
1322                 return -ENOENT;
1323 }
1324
1325 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1326 {
1327         struct inode *inode = dentry->d_inode;
1328         struct path path;
1329         int error = -EACCES;
1330
1331         /* Are we allowed to snoop on the tasks file descriptors? */
1332         if (!proc_fd_access_allowed(inode))
1333                 goto out;
1334
1335         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1336         if (error)
1337                 goto out;
1338
1339         nd_jump_link(nd, &path);
1340         return NULL;
1341 out:
1342         return ERR_PTR(error);
1343 }
1344
1345 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1346 {
1347         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1348         char *pathname;
1349         int len;
1350
1351         if (!tmp)
1352                 return -ENOMEM;
1353
1354         pathname = d_path(path, tmp, PAGE_SIZE);
1355         len = PTR_ERR(pathname);
1356         if (IS_ERR(pathname))
1357                 goto out;
1358         len = tmp + PAGE_SIZE - 1 - pathname;
1359
1360         if (len > buflen)
1361                 len = buflen;
1362         if (copy_to_user(buffer, pathname, len))
1363                 len = -EFAULT;
1364  out:
1365         free_page((unsigned long)tmp);
1366         return len;
1367 }
1368
1369 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1370 {
1371         int error = -EACCES;
1372         struct inode *inode = dentry->d_inode;
1373         struct path path;
1374
1375         /* Are we allowed to snoop on the tasks file descriptors? */
1376         if (!proc_fd_access_allowed(inode))
1377                 goto out;
1378
1379         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1380         if (error)
1381                 goto out;
1382
1383         error = do_proc_readlink(&path, buffer, buflen);
1384         path_put(&path);
1385 out:
1386         return error;
1387 }
1388
1389 const struct inode_operations proc_pid_link_inode_operations = {
1390         .readlink       = proc_pid_readlink,
1391         .follow_link    = proc_pid_follow_link,
1392         .setattr        = proc_setattr,
1393 };
1394
1395
1396 /* building an inode */
1397
1398 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1399 {
1400         struct inode * inode;
1401         struct proc_inode *ei;
1402         const struct cred *cred;
1403
1404         /* We need a new inode */
1405
1406         inode = new_inode(sb);
1407         if (!inode)
1408                 goto out;
1409
1410         /* Common stuff */
1411         ei = PROC_I(inode);
1412         inode->i_ino = get_next_ino();
1413         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1414         inode->i_op = &proc_def_inode_operations;
1415
1416         /*
1417          * grab the reference to task.
1418          */
1419         ei->pid = get_task_pid(task, PIDTYPE_PID);
1420         if (!ei->pid)
1421                 goto out_unlock;
1422
1423         if (task_dumpable(task)) {
1424                 rcu_read_lock();
1425                 cred = __task_cred(task);
1426                 inode->i_uid = cred->euid;
1427                 inode->i_gid = cred->egid;
1428                 rcu_read_unlock();
1429         }
1430         security_task_to_inode(task, inode);
1431
1432 out:
1433         return inode;
1434
1435 out_unlock:
1436         iput(inode);
1437         return NULL;
1438 }
1439
1440 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1441 {
1442         struct inode *inode = dentry->d_inode;
1443         struct task_struct *task;
1444         const struct cred *cred;
1445         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1446
1447         generic_fillattr(inode, stat);
1448
1449         rcu_read_lock();
1450         stat->uid = GLOBAL_ROOT_UID;
1451         stat->gid = GLOBAL_ROOT_GID;
1452         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1453         if (task) {
1454                 if (!has_pid_permissions(pid, task, 2)) {
1455                         rcu_read_unlock();
1456                         /*
1457                          * This doesn't prevent learning whether PID exists,
1458                          * it only makes getattr() consistent with readdir().
1459                          */
1460                         return -ENOENT;
1461                 }
1462                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1463                     task_dumpable(task)) {
1464                         cred = __task_cred(task);
1465                         stat->uid = cred->euid;
1466                         stat->gid = cred->egid;
1467                 }
1468         }
1469         rcu_read_unlock();
1470         return 0;
1471 }
1472
1473 /* dentry stuff */
1474
1475 /*
1476  *      Exceptional case: normally we are not allowed to unhash a busy
1477  * directory. In this case, however, we can do it - no aliasing problems
1478  * due to the way we treat inodes.
1479  *
1480  * Rewrite the inode's ownerships here because the owning task may have
1481  * performed a setuid(), etc.
1482  *
1483  * Before the /proc/pid/status file was created the only way to read
1484  * the effective uid of a /process was to stat /proc/pid.  Reading
1485  * /proc/pid/status is slow enough that procps and other packages
1486  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1487  * made this apply to all per process world readable and executable
1488  * directories.
1489  */
1490 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1491 {
1492         struct inode *inode;
1493         struct task_struct *task;
1494         const struct cred *cred;
1495
1496         if (flags & LOOKUP_RCU)
1497                 return -ECHILD;
1498
1499         inode = dentry->d_inode;
1500         task = get_proc_task(inode);
1501
1502         if (task) {
1503                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1504                     task_dumpable(task)) {
1505                         rcu_read_lock();
1506                         cred = __task_cred(task);
1507                         inode->i_uid = cred->euid;
1508                         inode->i_gid = cred->egid;
1509                         rcu_read_unlock();
1510                 } else {
1511                         inode->i_uid = GLOBAL_ROOT_UID;
1512                         inode->i_gid = GLOBAL_ROOT_GID;
1513                 }
1514                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1515                 security_task_to_inode(task, inode);
1516                 put_task_struct(task);
1517                 return 1;
1518         }
1519         d_drop(dentry);
1520         return 0;
1521 }
1522
1523 const struct dentry_operations pid_dentry_operations =
1524 {
1525         .d_revalidate   = pid_revalidate,
1526         .d_delete       = pid_delete_dentry,
1527 };
1528
1529 /* Lookups */
1530
1531 /*
1532  * Fill a directory entry.
1533  *
1534  * If possible create the dcache entry and derive our inode number and
1535  * file type from dcache entry.
1536  *
1537  * Since all of the proc inode numbers are dynamically generated, the inode
1538  * numbers do not exist until the inode is cache.  This means creating the
1539  * the dcache entry in readdir is necessary to keep the inode numbers
1540  * reported by readdir in sync with the inode numbers reported
1541  * by stat.
1542  */
1543 int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1544         const char *name, int len,
1545         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1546 {
1547         struct dentry *child, *dir = filp->f_path.dentry;
1548         struct inode *inode;
1549         struct qstr qname;
1550         ino_t ino = 0;
1551         unsigned type = DT_UNKNOWN;
1552
1553         qname.name = name;
1554         qname.len  = len;
1555         qname.hash = full_name_hash(name, len);
1556
1557         child = d_lookup(dir, &qname);
1558         if (!child) {
1559                 struct dentry *new;
1560                 new = d_alloc(dir, &qname);
1561                 if (new) {
1562                         child = instantiate(dir->d_inode, new, task, ptr);
1563                         if (child)
1564                                 dput(new);
1565                         else
1566                                 child = new;
1567                 }
1568         }
1569         if (!child || IS_ERR(child) || !child->d_inode)
1570                 goto end_instantiate;
1571         inode = child->d_inode;
1572         if (inode) {
1573                 ino = inode->i_ino;
1574                 type = inode->i_mode >> 12;
1575         }
1576         dput(child);
1577 end_instantiate:
1578         if (!ino)
1579                 ino = find_inode_number(dir, &qname);
1580         if (!ino)
1581                 ino = 1;
1582         return filldir(dirent, name, len, filp->f_pos, ino, type);
1583 }
1584
1585 #ifdef CONFIG_CHECKPOINT_RESTORE
1586
1587 /*
1588  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1589  * which represent vma start and end addresses.
1590  */
1591 static int dname_to_vma_addr(struct dentry *dentry,
1592                              unsigned long *start, unsigned long *end)
1593 {
1594         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1595                 return -EINVAL;
1596
1597         return 0;
1598 }
1599
1600 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1601 {
1602         unsigned long vm_start, vm_end;
1603         bool exact_vma_exists = false;
1604         struct mm_struct *mm = NULL;
1605         struct task_struct *task;
1606         const struct cred *cred;
1607         struct inode *inode;
1608         int status = 0;
1609
1610         if (flags & LOOKUP_RCU)
1611                 return -ECHILD;
1612
1613         if (!capable(CAP_SYS_ADMIN)) {
1614                 status = -EACCES;
1615                 goto out_notask;
1616         }
1617
1618         inode = dentry->d_inode;
1619         task = get_proc_task(inode);
1620         if (!task)
1621                 goto out_notask;
1622
1623         mm = mm_access(task, PTRACE_MODE_READ);
1624         if (IS_ERR_OR_NULL(mm))
1625                 goto out;
1626
1627         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1628                 down_read(&mm->mmap_sem);
1629                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1630                 up_read(&mm->mmap_sem);
1631         }
1632
1633         mmput(mm);
1634
1635         if (exact_vma_exists) {
1636                 if (task_dumpable(task)) {
1637                         rcu_read_lock();
1638                         cred = __task_cred(task);
1639                         inode->i_uid = cred->euid;
1640                         inode->i_gid = cred->egid;
1641                         rcu_read_unlock();
1642                 } else {
1643                         inode->i_uid = GLOBAL_ROOT_UID;
1644                         inode->i_gid = GLOBAL_ROOT_GID;
1645                 }
1646                 security_task_to_inode(task, inode);
1647                 status = 1;
1648         }
1649
1650 out:
1651         put_task_struct(task);
1652
1653 out_notask:
1654         if (status <= 0)
1655                 d_drop(dentry);
1656
1657         return status;
1658 }
1659
1660 static const struct dentry_operations tid_map_files_dentry_operations = {
1661         .d_revalidate   = map_files_d_revalidate,
1662         .d_delete       = pid_delete_dentry,
1663 };
1664
1665 static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
1666 {
1667         unsigned long vm_start, vm_end;
1668         struct vm_area_struct *vma;
1669         struct task_struct *task;
1670         struct mm_struct *mm;
1671         int rc;
1672
1673         rc = -ENOENT;
1674         task = get_proc_task(dentry->d_inode);
1675         if (!task)
1676                 goto out;
1677
1678         mm = get_task_mm(task);
1679         put_task_struct(task);
1680         if (!mm)
1681                 goto out;
1682
1683         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1684         if (rc)
1685                 goto out_mmput;
1686
1687         down_read(&mm->mmap_sem);
1688         vma = find_exact_vma(mm, vm_start, vm_end);
1689         if (vma && vma->vm_file) {
1690                 *path = vma->vm_file->f_path;
1691                 path_get(path);
1692                 rc = 0;
1693         }
1694         up_read(&mm->mmap_sem);
1695
1696 out_mmput:
1697         mmput(mm);
1698 out:
1699         return rc;
1700 }
1701
1702 struct map_files_info {
1703         fmode_t         mode;
1704         unsigned long   len;
1705         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1706 };
1707
1708 static struct dentry *
1709 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
1710                            struct task_struct *task, const void *ptr)
1711 {
1712         fmode_t mode = (fmode_t)(unsigned long)ptr;
1713         struct proc_inode *ei;
1714         struct inode *inode;
1715
1716         inode = proc_pid_make_inode(dir->i_sb, task);
1717         if (!inode)
1718                 return ERR_PTR(-ENOENT);
1719
1720         ei = PROC_I(inode);
1721         ei->op.proc_get_link = proc_map_files_get_link;
1722
1723         inode->i_op = &proc_pid_link_inode_operations;
1724         inode->i_size = 64;
1725         inode->i_mode = S_IFLNK;
1726
1727         if (mode & FMODE_READ)
1728                 inode->i_mode |= S_IRUSR;
1729         if (mode & FMODE_WRITE)
1730                 inode->i_mode |= S_IWUSR;
1731
1732         d_set_d_op(dentry, &tid_map_files_dentry_operations);
1733         d_add(dentry, inode);
1734
1735         return NULL;
1736 }
1737
1738 static struct dentry *proc_map_files_lookup(struct inode *dir,
1739                 struct dentry *dentry, unsigned int flags)
1740 {
1741         unsigned long vm_start, vm_end;
1742         struct vm_area_struct *vma;
1743         struct task_struct *task;
1744         struct dentry *result;
1745         struct mm_struct *mm;
1746
1747         result = ERR_PTR(-EACCES);
1748         if (!capable(CAP_SYS_ADMIN))
1749                 goto out;
1750
1751         result = ERR_PTR(-ENOENT);
1752         task = get_proc_task(dir);
1753         if (!task)
1754                 goto out;
1755
1756         result = ERR_PTR(-EACCES);
1757         if (!ptrace_may_access(task, PTRACE_MODE_READ))
1758                 goto out_put_task;
1759
1760         result = ERR_PTR(-ENOENT);
1761         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
1762                 goto out_put_task;
1763
1764         mm = get_task_mm(task);
1765         if (!mm)
1766                 goto out_put_task;
1767
1768         down_read(&mm->mmap_sem);
1769         vma = find_exact_vma(mm, vm_start, vm_end);
1770         if (!vma)
1771                 goto out_no_vma;
1772
1773         result = proc_map_files_instantiate(dir, dentry, task,
1774                         (void *)(unsigned long)vma->vm_file->f_mode);
1775
1776 out_no_vma:
1777         up_read(&mm->mmap_sem);
1778         mmput(mm);
1779 out_put_task:
1780         put_task_struct(task);
1781 out:
1782         return result;
1783 }
1784
1785 static const struct inode_operations proc_map_files_inode_operations = {
1786         .lookup         = proc_map_files_lookup,
1787         .permission     = proc_fd_permission,
1788         .setattr        = proc_setattr,
1789 };
1790
1791 static int
1792 proc_map_files_readdir(struct file *filp, void *dirent, filldir_t filldir)
1793 {
1794         struct dentry *dentry = filp->f_path.dentry;
1795         struct inode *inode = dentry->d_inode;
1796         struct vm_area_struct *vma;
1797         struct task_struct *task;
1798         struct mm_struct *mm;
1799         ino_t ino;
1800         int ret;
1801
1802         ret = -EACCES;
1803         if (!capable(CAP_SYS_ADMIN))
1804                 goto out;
1805
1806         ret = -ENOENT;
1807         task = get_proc_task(inode);
1808         if (!task)
1809                 goto out;
1810
1811         ret = -EACCES;
1812         if (!ptrace_may_access(task, PTRACE_MODE_READ))
1813                 goto out_put_task;
1814
1815         ret = 0;
1816         switch (filp->f_pos) {
1817         case 0:
1818                 ino = inode->i_ino;
1819                 if (filldir(dirent, ".", 1, 0, ino, DT_DIR) < 0)
1820                         goto out_put_task;
1821                 filp->f_pos++;
1822         case 1:
1823                 ino = parent_ino(dentry);
1824                 if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1825                         goto out_put_task;
1826                 filp->f_pos++;
1827         default:
1828         {
1829                 unsigned long nr_files, pos, i;
1830                 struct flex_array *fa = NULL;
1831                 struct map_files_info info;
1832                 struct map_files_info *p;
1833
1834                 mm = get_task_mm(task);
1835                 if (!mm)
1836                         goto out_put_task;
1837                 down_read(&mm->mmap_sem);
1838
1839                 nr_files = 0;
1840
1841                 /*
1842                  * We need two passes here:
1843                  *
1844                  *  1) Collect vmas of mapped files with mmap_sem taken
1845                  *  2) Release mmap_sem and instantiate entries
1846                  *
1847                  * otherwise we get lockdep complained, since filldir()
1848                  * routine might require mmap_sem taken in might_fault().
1849                  */
1850
1851                 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
1852                         if (vma->vm_file && ++pos > filp->f_pos)
1853                                 nr_files++;
1854                 }
1855
1856                 if (nr_files) {
1857                         fa = flex_array_alloc(sizeof(info), nr_files,
1858                                                 GFP_KERNEL);
1859                         if (!fa || flex_array_prealloc(fa, 0, nr_files,
1860                                                         GFP_KERNEL)) {
1861                                 ret = -ENOMEM;
1862                                 if (fa)
1863                                         flex_array_free(fa);
1864                                 up_read(&mm->mmap_sem);
1865                                 mmput(mm);
1866                                 goto out_put_task;
1867                         }
1868                         for (i = 0, vma = mm->mmap, pos = 2; vma;
1869                                         vma = vma->vm_next) {
1870                                 if (!vma->vm_file)
1871                                         continue;
1872                                 if (++pos <= filp->f_pos)
1873                                         continue;
1874
1875                                 info.mode = vma->vm_file->f_mode;
1876                                 info.len = snprintf(info.name,
1877                                                 sizeof(info.name), "%lx-%lx",
1878                                                 vma->vm_start, vma->vm_end);
1879                                 if (flex_array_put(fa, i++, &info, GFP_KERNEL))
1880                                         BUG();
1881                         }
1882                 }
1883                 up_read(&mm->mmap_sem);
1884
1885                 for (i = 0; i < nr_files; i++) {
1886                         p = flex_array_get(fa, i);
1887                         ret = proc_fill_cache(filp, dirent, filldir,
1888                                               p->name, p->len,
1889                                               proc_map_files_instantiate,
1890                                               task,
1891                                               (void *)(unsigned long)p->mode);
1892                         if (ret)
1893                                 break;
1894                         filp->f_pos++;
1895                 }
1896                 if (fa)
1897                         flex_array_free(fa);
1898                 mmput(mm);
1899         }
1900         }
1901
1902 out_put_task:
1903         put_task_struct(task);
1904 out:
1905         return ret;
1906 }
1907
1908 static const struct file_operations proc_map_files_operations = {
1909         .read           = generic_read_dir,
1910         .readdir        = proc_map_files_readdir,
1911         .llseek         = default_llseek,
1912 };
1913
1914 #endif /* CONFIG_CHECKPOINT_RESTORE */
1915
1916 static struct dentry *proc_pident_instantiate(struct inode *dir,
1917         struct dentry *dentry, struct task_struct *task, const void *ptr)
1918 {
1919         const struct pid_entry *p = ptr;
1920         struct inode *inode;
1921         struct proc_inode *ei;
1922         struct dentry *error = ERR_PTR(-ENOENT);
1923
1924         inode = proc_pid_make_inode(dir->i_sb, task);
1925         if (!inode)
1926                 goto out;
1927
1928         ei = PROC_I(inode);
1929         inode->i_mode = p->mode;
1930         if (S_ISDIR(inode->i_mode))
1931                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
1932         if (p->iop)
1933                 inode->i_op = p->iop;
1934         if (p->fop)
1935                 inode->i_fop = p->fop;
1936         ei->op = p->op;
1937         d_set_d_op(dentry, &pid_dentry_operations);
1938         d_add(dentry, inode);
1939         /* Close the race of the process dying before we return the dentry */
1940         if (pid_revalidate(dentry, 0))
1941                 error = NULL;
1942 out:
1943         return error;
1944 }
1945
1946 static struct dentry *proc_pident_lookup(struct inode *dir,
1947                                          struct dentry *dentry,
1948                                          const struct pid_entry *ents,
1949                                          unsigned int nents)
1950 {
1951         struct dentry *error;
1952         struct task_struct *task = get_proc_task(dir);
1953         const struct pid_entry *p, *last;
1954
1955         error = ERR_PTR(-ENOENT);
1956
1957         if (!task)
1958                 goto out_no_task;
1959
1960         /*
1961          * Yes, it does not scale. And it should not. Don't add
1962          * new entries into /proc/<tgid>/ without very good reasons.
1963          */
1964         last = &ents[nents - 1];
1965         for (p = ents; p <= last; p++) {
1966                 if (p->len != dentry->d_name.len)
1967                         continue;
1968                 if (!memcmp(dentry->d_name.name, p->name, p->len))
1969                         break;
1970         }
1971         if (p > last)
1972                 goto out;
1973
1974         error = proc_pident_instantiate(dir, dentry, task, p);
1975 out:
1976         put_task_struct(task);
1977 out_no_task:
1978         return error;
1979 }
1980
1981 static int proc_pident_fill_cache(struct file *filp, void *dirent,
1982         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
1983 {
1984         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
1985                                 proc_pident_instantiate, task, p);
1986 }
1987
1988 static int proc_pident_readdir(struct file *filp,
1989                 void *dirent, filldir_t filldir,
1990                 const struct pid_entry *ents, unsigned int nents)
1991 {
1992         int i;
1993         struct dentry *dentry = filp->f_path.dentry;
1994         struct inode *inode = dentry->d_inode;
1995         struct task_struct *task = get_proc_task(inode);
1996         const struct pid_entry *p, *last;
1997         ino_t ino;
1998         int ret;
1999
2000         ret = -ENOENT;
2001         if (!task)
2002                 goto out_no_task;
2003
2004         ret = 0;
2005         i = filp->f_pos;
2006         switch (i) {
2007         case 0:
2008                 ino = inode->i_ino;
2009                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2010                         goto out;
2011                 i++;
2012                 filp->f_pos++;
2013                 /* fall through */
2014         case 1:
2015                 ino = parent_ino(dentry);
2016                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2017                         goto out;
2018                 i++;
2019                 filp->f_pos++;
2020                 /* fall through */
2021         default:
2022                 i -= 2;
2023                 if (i >= nents) {
2024                         ret = 1;
2025                         goto out;
2026                 }
2027                 p = ents + i;
2028                 last = &ents[nents - 1];
2029                 while (p <= last) {
2030                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2031                                 goto out;
2032                         filp->f_pos++;
2033                         p++;
2034                 }
2035         }
2036
2037         ret = 1;
2038 out:
2039         put_task_struct(task);
2040 out_no_task:
2041         return ret;
2042 }
2043
2044 #ifdef CONFIG_SECURITY
2045 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2046                                   size_t count, loff_t *ppos)
2047 {
2048         struct inode * inode = file->f_path.dentry->d_inode;
2049         char *p = NULL;
2050         ssize_t length;
2051         struct task_struct *task = get_proc_task(inode);
2052
2053         if (!task)
2054                 return -ESRCH;
2055
2056         length = security_getprocattr(task,
2057                                       (char*)file->f_path.dentry->d_name.name,
2058                                       &p);
2059         put_task_struct(task);
2060         if (length > 0)
2061                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2062         kfree(p);
2063         return length;
2064 }
2065
2066 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2067                                    size_t count, loff_t *ppos)
2068 {
2069         struct inode * inode = file->f_path.dentry->d_inode;
2070         char *page;
2071         ssize_t length;
2072         struct task_struct *task = get_proc_task(inode);
2073
2074         length = -ESRCH;
2075         if (!task)
2076                 goto out_no_task;
2077         if (count > PAGE_SIZE)
2078                 count = PAGE_SIZE;
2079
2080         /* No partial writes. */
2081         length = -EINVAL;
2082         if (*ppos != 0)
2083                 goto out;
2084
2085         length = -ENOMEM;
2086         page = (char*)__get_free_page(GFP_TEMPORARY);
2087         if (!page)
2088                 goto out;
2089
2090         length = -EFAULT;
2091         if (copy_from_user(page, buf, count))
2092                 goto out_free;
2093
2094         /* Guard against adverse ptrace interaction */
2095         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2096         if (length < 0)
2097                 goto out_free;
2098
2099         length = security_setprocattr(task,
2100                                       (char*)file->f_path.dentry->d_name.name,
2101                                       (void*)page, count);
2102         mutex_unlock(&task->signal->cred_guard_mutex);
2103 out_free:
2104         free_page((unsigned long) page);
2105 out:
2106         put_task_struct(task);
2107 out_no_task:
2108         return length;
2109 }
2110
2111 static const struct file_operations proc_pid_attr_operations = {
2112         .read           = proc_pid_attr_read,
2113         .write          = proc_pid_attr_write,
2114         .llseek         = generic_file_llseek,
2115 };
2116
2117 static const struct pid_entry attr_dir_stuff[] = {
2118         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2119         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2120         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2121         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2122         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2123         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2124 };
2125
2126 static int proc_attr_dir_readdir(struct file * filp,
2127                              void * dirent, filldir_t filldir)
2128 {
2129         return proc_pident_readdir(filp,dirent,filldir,
2130                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2131 }
2132
2133 static const struct file_operations proc_attr_dir_operations = {
2134         .read           = generic_read_dir,
2135         .readdir        = proc_attr_dir_readdir,
2136         .llseek         = default_llseek,
2137 };
2138
2139 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2140                                 struct dentry *dentry, unsigned int flags)
2141 {
2142         return proc_pident_lookup(dir, dentry,
2143                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2144 }
2145
2146 static const struct inode_operations proc_attr_dir_inode_operations = {
2147         .lookup         = proc_attr_dir_lookup,
2148         .getattr        = pid_getattr,
2149         .setattr        = proc_setattr,
2150 };
2151
2152 #endif
2153
2154 #ifdef CONFIG_ELF_CORE
2155 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2156                                          size_t count, loff_t *ppos)
2157 {
2158         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2159         struct mm_struct *mm;
2160         char buffer[PROC_NUMBUF];
2161         size_t len;
2162         int ret;
2163
2164         if (!task)
2165                 return -ESRCH;
2166
2167         ret = 0;
2168         mm = get_task_mm(task);
2169         if (mm) {
2170                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2171                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2172                                 MMF_DUMP_FILTER_SHIFT));
2173                 mmput(mm);
2174                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2175         }
2176
2177         put_task_struct(task);
2178
2179         return ret;
2180 }
2181
2182 static ssize_t proc_coredump_filter_write(struct file *file,
2183                                           const char __user *buf,
2184                                           size_t count,
2185                                           loff_t *ppos)
2186 {
2187         struct task_struct *task;
2188         struct mm_struct *mm;
2189         char buffer[PROC_NUMBUF], *end;
2190         unsigned int val;
2191         int ret;
2192         int i;
2193         unsigned long mask;
2194
2195         ret = -EFAULT;
2196         memset(buffer, 0, sizeof(buffer));
2197         if (count > sizeof(buffer) - 1)
2198                 count = sizeof(buffer) - 1;
2199         if (copy_from_user(buffer, buf, count))
2200                 goto out_no_task;
2201
2202         ret = -EINVAL;
2203         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2204         if (*end == '\n')
2205                 end++;
2206         if (end - buffer == 0)
2207                 goto out_no_task;
2208
2209         ret = -ESRCH;
2210         task = get_proc_task(file->f_dentry->d_inode);
2211         if (!task)
2212                 goto out_no_task;
2213
2214         ret = end - buffer;
2215         mm = get_task_mm(task);
2216         if (!mm)
2217                 goto out_no_mm;
2218
2219         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2220                 if (val & mask)
2221                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2222                 else
2223                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2224         }
2225
2226         mmput(mm);
2227  out_no_mm:
2228         put_task_struct(task);
2229  out_no_task:
2230         return ret;
2231 }
2232
2233 static const struct file_operations proc_coredump_filter_operations = {
2234         .read           = proc_coredump_filter_read,
2235         .write          = proc_coredump_filter_write,
2236         .llseek         = generic_file_llseek,
2237 };
2238 #endif
2239
2240 /*
2241  * /proc/self:
2242  */
2243 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2244                               int buflen)
2245 {
2246         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2247         pid_t tgid = task_tgid_nr_ns(current, ns);
2248         char tmp[PROC_NUMBUF];
2249         if (!tgid)
2250                 return -ENOENT;
2251         sprintf(tmp, "%d", tgid);
2252         return vfs_readlink(dentry,buffer,buflen,tmp);
2253 }
2254
2255 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2256 {
2257         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2258         pid_t tgid = task_tgid_nr_ns(current, ns);
2259         char *name = ERR_PTR(-ENOENT);
2260         if (tgid) {
2261                 name = __getname();
2262                 if (!name)
2263                         name = ERR_PTR(-ENOMEM);
2264                 else
2265                         sprintf(name, "%d", tgid);
2266         }
2267         nd_set_link(nd, name);
2268         return NULL;
2269 }
2270
2271 static void proc_self_put_link(struct dentry *dentry, struct nameidata *nd,
2272                                 void *cookie)
2273 {
2274         char *s = nd_get_link(nd);
2275         if (!IS_ERR(s))
2276                 __putname(s);
2277 }
2278
2279 static const struct inode_operations proc_self_inode_operations = {
2280         .readlink       = proc_self_readlink,
2281         .follow_link    = proc_self_follow_link,
2282         .put_link       = proc_self_put_link,
2283 };
2284
2285 /*
2286  * proc base
2287  *
2288  * These are the directory entries in the root directory of /proc
2289  * that properly belong to the /proc filesystem, as they describe
2290  * describe something that is process related.
2291  */
2292 static const struct pid_entry proc_base_stuff[] = {
2293         NOD("self", S_IFLNK|S_IRWXUGO,
2294                 &proc_self_inode_operations, NULL, {}),
2295 };
2296
2297 static struct dentry *proc_base_instantiate(struct inode *dir,
2298         struct dentry *dentry, struct task_struct *task, const void *ptr)
2299 {
2300         const struct pid_entry *p = ptr;
2301         struct inode *inode;
2302         struct proc_inode *ei;
2303         struct dentry *error;
2304
2305         /* Allocate the inode */
2306         error = ERR_PTR(-ENOMEM);
2307         inode = new_inode(dir->i_sb);
2308         if (!inode)
2309                 goto out;
2310
2311         /* Initialize the inode */
2312         ei = PROC_I(inode);
2313         inode->i_ino = get_next_ino();
2314         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2315
2316         /*
2317          * grab the reference to the task.
2318          */
2319         ei->pid = get_task_pid(task, PIDTYPE_PID);
2320         if (!ei->pid)
2321                 goto out_iput;
2322
2323         inode->i_mode = p->mode;
2324         if (S_ISDIR(inode->i_mode))
2325                 set_nlink(inode, 2);
2326         if (S_ISLNK(inode->i_mode))
2327                 inode->i_size = 64;
2328         if (p->iop)
2329                 inode->i_op = p->iop;
2330         if (p->fop)
2331                 inode->i_fop = p->fop;
2332         ei->op = p->op;
2333         d_add(dentry, inode);
2334         error = NULL;
2335 out:
2336         return error;
2337 out_iput:
2338         iput(inode);
2339         goto out;
2340 }
2341
2342 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2343 {
2344         struct dentry *error;
2345         struct task_struct *task = get_proc_task(dir);
2346         const struct pid_entry *p, *last;
2347
2348         error = ERR_PTR(-ENOENT);
2349
2350         if (!task)
2351                 goto out_no_task;
2352
2353         /* Lookup the directory entry */
2354         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2355         for (p = proc_base_stuff; p <= last; p++) {
2356                 if (p->len != dentry->d_name.len)
2357                         continue;
2358                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2359                         break;
2360         }
2361         if (p > last)
2362                 goto out;
2363
2364         error = proc_base_instantiate(dir, dentry, task, p);
2365
2366 out:
2367         put_task_struct(task);
2368 out_no_task:
2369         return error;
2370 }
2371
2372 static int proc_base_fill_cache(struct file *filp, void *dirent,
2373         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2374 {
2375         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2376                                 proc_base_instantiate, task, p);
2377 }
2378
2379 #ifdef CONFIG_TASK_IO_ACCOUNTING
2380 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2381 {
2382         struct task_io_accounting acct = task->ioac;
2383         unsigned long flags;
2384         int result;
2385
2386         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2387         if (result)
2388                 return result;
2389
2390         if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2391                 result = -EACCES;
2392                 goto out_unlock;
2393         }
2394
2395         if (whole && lock_task_sighand(task, &flags)) {
2396                 struct task_struct *t = task;
2397
2398                 task_io_accounting_add(&acct, &task->signal->ioac);
2399                 while_each_thread(task, t)
2400                         task_io_accounting_add(&acct, &t->ioac);
2401
2402                 unlock_task_sighand(task, &flags);
2403         }
2404         result = sprintf(buffer,
2405                         "rchar: %llu\n"
2406                         "wchar: %llu\n"
2407                         "syscr: %llu\n"
2408                         "syscw: %llu\n"
2409                         "read_bytes: %llu\n"
2410                         "write_bytes: %llu\n"
2411                         "cancelled_write_bytes: %llu\n",
2412                         (unsigned long long)acct.rchar,
2413                         (unsigned long long)acct.wchar,
2414                         (unsigned long long)acct.syscr,
2415                         (unsigned long long)acct.syscw,
2416                         (unsigned long long)acct.read_bytes,
2417                         (unsigned long long)acct.write_bytes,
2418                         (unsigned long long)acct.cancelled_write_bytes);
2419 out_unlock:
2420         mutex_unlock(&task->signal->cred_guard_mutex);
2421         return result;
2422 }
2423
2424 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2425 {
2426         return do_io_accounting(task, buffer, 0);
2427 }
2428
2429 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2430 {
2431         return do_io_accounting(task, buffer, 1);
2432 }
2433 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2434
2435 #ifdef CONFIG_USER_NS
2436 static int proc_id_map_open(struct inode *inode, struct file *file,
2437         struct seq_operations *seq_ops)
2438 {
2439         struct user_namespace *ns = NULL;
2440         struct task_struct *task;
2441         struct seq_file *seq;
2442         int ret = -EINVAL;
2443
2444         task = get_proc_task(inode);
2445         if (task) {
2446                 rcu_read_lock();
2447                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2448                 rcu_read_unlock();
2449                 put_task_struct(task);
2450         }
2451         if (!ns)
2452                 goto err;
2453
2454         ret = seq_open(file, seq_ops);
2455         if (ret)
2456                 goto err_put_ns;
2457
2458         seq = file->private_data;
2459         seq->private = ns;
2460
2461         return 0;
2462 err_put_ns:
2463         put_user_ns(ns);
2464 err:
2465         return ret;
2466 }
2467
2468 static int proc_id_map_release(struct inode *inode, struct file *file)
2469 {
2470         struct seq_file *seq = file->private_data;
2471         struct user_namespace *ns = seq->private;
2472         put_user_ns(ns);
2473         return seq_release(inode, file);
2474 }
2475
2476 static int proc_uid_map_open(struct inode *inode, struct file *file)
2477 {
2478         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2479 }
2480
2481 static int proc_gid_map_open(struct inode *inode, struct file *file)
2482 {
2483         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2484 }
2485
2486 static int proc_projid_map_open(struct inode *inode, struct file *file)
2487 {
2488         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2489 }
2490
2491 static const struct file_operations proc_uid_map_operations = {
2492         .open           = proc_uid_map_open,
2493         .write          = proc_uid_map_write,
2494         .read           = seq_read,
2495         .llseek         = seq_lseek,
2496         .release        = proc_id_map_release,
2497 };
2498
2499 static const struct file_operations proc_gid_map_operations = {
2500         .open           = proc_gid_map_open,
2501         .write          = proc_gid_map_write,
2502         .read           = seq_read,
2503         .llseek         = seq_lseek,
2504         .release        = proc_id_map_release,
2505 };
2506
2507 static const struct file_operations proc_projid_map_operations = {
2508         .open           = proc_projid_map_open,
2509         .write          = proc_projid_map_write,
2510         .read           = seq_read,
2511         .llseek         = seq_lseek,
2512         .release        = proc_id_map_release,
2513 };
2514 #endif /* CONFIG_USER_NS */
2515
2516 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2517                                 struct pid *pid, struct task_struct *task)
2518 {
2519         int err = lock_trace(task);
2520         if (!err) {
2521                 seq_printf(m, "%08x\n", task->personality);
2522                 unlock_trace(task);
2523         }
2524         return err;
2525 }
2526
2527 /*
2528  * Thread groups
2529  */
2530 static const struct file_operations proc_task_operations;
2531 static const struct inode_operations proc_task_inode_operations;
2532
2533 static const struct pid_entry tgid_base_stuff[] = {
2534         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2535         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2536 #ifdef CONFIG_CHECKPOINT_RESTORE
2537         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2538 #endif
2539         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2540         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2541 #ifdef CONFIG_NET
2542         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2543 #endif
2544         REG("environ",    S_IRUSR, proc_environ_operations),
2545         INF("auxv",       S_IRUSR, proc_pid_auxv),
2546         ONE("status",     S_IRUGO, proc_pid_status),
2547         ONE("personality", S_IRUGO, proc_pid_personality),
2548         INF("limits",     S_IRUGO, proc_pid_limits),
2549 #ifdef CONFIG_SCHED_DEBUG
2550         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2551 #endif
2552 #ifdef CONFIG_SCHED_AUTOGROUP
2553         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2554 #endif
2555         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2556 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2557         INF("syscall",    S_IRUGO, proc_pid_syscall),
2558 #endif
2559         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2560         ONE("stat",       S_IRUGO, proc_tgid_stat),
2561         ONE("statm",      S_IRUGO, proc_pid_statm),
2562         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2563 #ifdef CONFIG_NUMA
2564         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2565 #endif
2566         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2567         LNK("cwd",        proc_cwd_link),
2568         LNK("root",       proc_root_link),
2569         LNK("exe",        proc_exe_link),
2570         REG("mounts",     S_IRUGO, proc_mounts_operations),
2571         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2572         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2573 #ifdef CONFIG_PROC_PAGE_MONITOR
2574         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2575         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2576         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
2577 #endif
2578 #ifdef CONFIG_SECURITY
2579         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2580 #endif
2581 #ifdef CONFIG_KALLSYMS
2582         INF("wchan",      S_IRUGO, proc_pid_wchan),
2583 #endif
2584 #ifdef CONFIG_STACKTRACE
2585         ONE("stack",      S_IRUGO, proc_pid_stack),
2586 #endif
2587 #ifdef CONFIG_SCHEDSTATS
2588         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
2589 #endif
2590 #ifdef CONFIG_LATENCYTOP
2591         REG("latency",  S_IRUGO, proc_lstats_operations),
2592 #endif
2593 #ifdef CONFIG_PROC_PID_CPUSET
2594         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2595 #endif
2596 #ifdef CONFIG_CGROUPS
2597         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2598 #endif
2599         INF("oom_score",  S_IRUGO, proc_oom_score),
2600         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2601 #ifdef CONFIG_AUDITSYSCALL
2602         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2603         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2604 #endif
2605 #ifdef CONFIG_FAULT_INJECTION
2606         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2607 #endif
2608 #ifdef CONFIG_ELF_CORE
2609         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2610 #endif
2611 #ifdef CONFIG_TASK_IO_ACCOUNTING
2612         INF("io",       S_IRUSR, proc_tgid_io_accounting),
2613 #endif
2614 #ifdef CONFIG_HARDWALL
2615         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
2616 #endif
2617 #ifdef CONFIG_USER_NS
2618         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2619         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2620         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2621 #endif
2622 };
2623
2624 static int proc_tgid_base_readdir(struct file * filp,
2625                              void * dirent, filldir_t filldir)
2626 {
2627         return proc_pident_readdir(filp,dirent,filldir,
2628                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2629 }
2630
2631 static const struct file_operations proc_tgid_base_operations = {
2632         .read           = generic_read_dir,
2633         .readdir        = proc_tgid_base_readdir,
2634         .llseek         = default_llseek,
2635 };
2636
2637 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2638 {
2639         return proc_pident_lookup(dir, dentry,
2640                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2641 }
2642
2643 static const struct inode_operations proc_tgid_base_inode_operations = {
2644         .lookup         = proc_tgid_base_lookup,
2645         .getattr        = pid_getattr,
2646         .setattr        = proc_setattr,
2647         .permission     = proc_pid_permission,
2648 };
2649
2650 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2651 {
2652         struct dentry *dentry, *leader, *dir;
2653         char buf[PROC_NUMBUF];
2654         struct qstr name;
2655
2656         name.name = buf;
2657         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2658         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2659         if (dentry) {
2660                 shrink_dcache_parent(dentry);
2661                 d_drop(dentry);
2662                 dput(dentry);
2663         }
2664
2665         name.name = buf;
2666         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2667         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2668         if (!leader)
2669                 goto out;
2670
2671         name.name = "task";
2672         name.len = strlen(name.name);
2673         dir = d_hash_and_lookup(leader, &name);
2674         if (!dir)
2675                 goto out_put_leader;
2676
2677         name.name = buf;
2678         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2679         dentry = d_hash_and_lookup(dir, &name);
2680         if (dentry) {
2681                 shrink_dcache_parent(dentry);
2682                 d_drop(dentry);
2683                 dput(dentry);
2684         }
2685
2686         dput(dir);
2687 out_put_leader:
2688         dput(leader);
2689 out:
2690         return;
2691 }
2692
2693 /**
2694  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2695  * @task: task that should be flushed.
2696  *
2697  * When flushing dentries from proc, one needs to flush them from global
2698  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2699  * in. This call is supposed to do all of this job.
2700  *
2701  * Looks in the dcache for
2702  * /proc/@pid
2703  * /proc/@tgid/task/@pid
2704  * if either directory is present flushes it and all of it'ts children
2705  * from the dcache.
2706  *
2707  * It is safe and reasonable to cache /proc entries for a task until
2708  * that task exits.  After that they just clog up the dcache with
2709  * useless entries, possibly causing useful dcache entries to be
2710  * flushed instead.  This routine is proved to flush those useless
2711  * dcache entries at process exit time.
2712  *
2713  * NOTE: This routine is just an optimization so it does not guarantee
2714  *       that no dcache entries will exist at process exit time it
2715  *       just makes it very unlikely that any will persist.
2716  */
2717
2718 void proc_flush_task(struct task_struct *task)
2719 {
2720         int i;
2721         struct pid *pid, *tgid;
2722         struct upid *upid;
2723
2724         pid = task_pid(task);
2725         tgid = task_tgid(task);
2726
2727         for (i = 0; i <= pid->level; i++) {
2728                 upid = &pid->numbers[i];
2729                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2730                                         tgid->numbers[i].nr);
2731         }
2732
2733         upid = &pid->numbers[pid->level];
2734         if (upid->nr == 1)
2735                 pid_ns_release_proc(upid->ns);
2736 }
2737
2738 static struct dentry *proc_pid_instantiate(struct inode *dir,
2739                                            struct dentry * dentry,
2740                                            struct task_struct *task, const void *ptr)
2741 {
2742         struct dentry *error = ERR_PTR(-ENOENT);
2743         struct inode *inode;
2744
2745         inode = proc_pid_make_inode(dir->i_sb, task);
2746         if (!inode)
2747                 goto out;
2748
2749         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2750         inode->i_op = &proc_tgid_base_inode_operations;
2751         inode->i_fop = &proc_tgid_base_operations;
2752         inode->i_flags|=S_IMMUTABLE;
2753
2754         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
2755                                                   ARRAY_SIZE(tgid_base_stuff)));
2756
2757         d_set_d_op(dentry, &pid_dentry_operations);
2758
2759         d_add(dentry, inode);
2760         /* Close the race of the process dying before we return the dentry */
2761         if (pid_revalidate(dentry, 0))
2762                 error = NULL;
2763 out:
2764         return error;
2765 }
2766
2767 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
2768 {
2769         struct dentry *result;
2770         struct task_struct *task;
2771         unsigned tgid;
2772         struct pid_namespace *ns;
2773
2774         result = proc_base_lookup(dir, dentry);
2775         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
2776                 goto out;
2777
2778         tgid = name_to_int(dentry);
2779         if (tgid == ~0U)
2780                 goto out;
2781
2782         ns = dentry->d_sb->s_fs_info;
2783         rcu_read_lock();
2784         task = find_task_by_pid_ns(tgid, ns);
2785         if (task)
2786                 get_task_struct(task);
2787         rcu_read_unlock();
2788         if (!task)
2789                 goto out;
2790
2791         result = proc_pid_instantiate(dir, dentry, task, NULL);
2792         put_task_struct(task);
2793 out:
2794         return result;
2795 }
2796
2797 /*
2798  * Find the first task with tgid >= tgid
2799  *
2800  */
2801 struct tgid_iter {
2802         unsigned int tgid;
2803         struct task_struct *task;
2804 };
2805 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2806 {
2807         struct pid *pid;
2808
2809         if (iter.task)
2810                 put_task_struct(iter.task);
2811         rcu_read_lock();
2812 retry:
2813         iter.task = NULL;
2814         pid = find_ge_pid(iter.tgid, ns);
2815         if (pid) {
2816                 iter.tgid = pid_nr_ns(pid, ns);
2817                 iter.task = pid_task(pid, PIDTYPE_PID);
2818                 /* What we to know is if the pid we have find is the
2819                  * pid of a thread_group_leader.  Testing for task
2820                  * being a thread_group_leader is the obvious thing
2821                  * todo but there is a window when it fails, due to
2822                  * the pid transfer logic in de_thread.
2823                  *
2824                  * So we perform the straight forward test of seeing
2825                  * if the pid we have found is the pid of a thread
2826                  * group leader, and don't worry if the task we have
2827                  * found doesn't happen to be a thread group leader.
2828                  * As we don't care in the case of readdir.
2829                  */
2830                 if (!iter.task || !has_group_leader_pid(iter.task)) {
2831                         iter.tgid += 1;
2832                         goto retry;
2833                 }
2834                 get_task_struct(iter.task);
2835         }
2836         rcu_read_unlock();
2837         return iter;
2838 }
2839
2840 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
2841
2842 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2843         struct tgid_iter iter)
2844 {
2845         char name[PROC_NUMBUF];
2846         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
2847         return proc_fill_cache(filp, dirent, filldir, name, len,
2848                                 proc_pid_instantiate, iter.task, NULL);
2849 }
2850
2851 static int fake_filldir(void *buf, const char *name, int namelen,
2852                         loff_t offset, u64 ino, unsigned d_type)
2853 {
2854         return 0;
2855 }
2856
2857 /* for the /proc/ directory itself, after non-process stuff has been done */
2858 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2859 {
2860         unsigned int nr;
2861         struct task_struct *reaper;
2862         struct tgid_iter iter;
2863         struct pid_namespace *ns;
2864         filldir_t __filldir;
2865
2866         if (filp->f_pos >= PID_MAX_LIMIT + TGID_OFFSET)
2867                 goto out_no_task;
2868         nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2869
2870         reaper = get_proc_task(filp->f_path.dentry->d_inode);
2871         if (!reaper)
2872                 goto out_no_task;
2873
2874         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
2875                 const struct pid_entry *p = &proc_base_stuff[nr];
2876                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
2877                         goto out;
2878         }
2879
2880         ns = filp->f_dentry->d_sb->s_fs_info;
2881         iter.task = NULL;
2882         iter.tgid = filp->f_pos - TGID_OFFSET;
2883         for (iter = next_tgid(ns, iter);
2884              iter.task;
2885              iter.tgid += 1, iter = next_tgid(ns, iter)) {
2886                 if (has_pid_permissions(ns, iter.task, 2))
2887                         __filldir = filldir;
2888                 else
2889                         __filldir = fake_filldir;
2890
2891                 filp->f_pos = iter.tgid + TGID_OFFSET;
2892                 if (proc_pid_fill_cache(filp, dirent, __filldir, iter) < 0) {
2893                         put_task_struct(iter.task);
2894                         goto out;
2895                 }
2896         }
2897         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
2898 out:
2899         put_task_struct(reaper);
2900 out_no_task:
2901         return 0;
2902 }
2903
2904 /*
2905  * Tasks
2906  */
2907 static const struct pid_entry tid_base_stuff[] = {
2908         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2909         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2910         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2911         REG("environ",   S_IRUSR, proc_environ_operations),
2912         INF("auxv",      S_IRUSR, proc_pid_auxv),
2913         ONE("status",    S_IRUGO, proc_pid_status),
2914         ONE("personality", S_IRUGO, proc_pid_personality),
2915         INF("limits",    S_IRUGO, proc_pid_limits),
2916 #ifdef CONFIG_SCHED_DEBUG
2917         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2918 #endif
2919         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2920 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2921         INF("syscall",   S_IRUGO, proc_pid_syscall),
2922 #endif
2923         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
2924         ONE("stat",      S_IRUGO, proc_tid_stat),
2925         ONE("statm",     S_IRUGO, proc_pid_statm),
2926         REG("maps",      S_IRUGO, proc_tid_maps_operations),
2927 #ifdef CONFIG_CHECKPOINT_RESTORE
2928         REG("children",  S_IRUGO, proc_tid_children_operations),
2929 #endif
2930 #ifdef CONFIG_NUMA
2931         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
2932 #endif
2933         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
2934         LNK("cwd",       proc_cwd_link),
2935         LNK("root",      proc_root_link),
2936         LNK("exe",       proc_exe_link),
2937         REG("mounts",    S_IRUGO, proc_mounts_operations),
2938         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2939 #ifdef CONFIG_PROC_PAGE_MONITOR
2940         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2941         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
2942         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
2943 #endif
2944 #ifdef CONFIG_SECURITY
2945         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2946 #endif
2947 #ifdef CONFIG_KALLSYMS
2948         INF("wchan",     S_IRUGO, proc_pid_wchan),
2949 #endif
2950 #ifdef CONFIG_STACKTRACE
2951         ONE("stack",      S_IRUGO, proc_pid_stack),
2952 #endif
2953 #ifdef CONFIG_SCHEDSTATS
2954         INF("schedstat", S_IRUGO, proc_pid_schedstat),
2955 #endif
2956 #ifdef CONFIG_LATENCYTOP
2957         REG("latency",  S_IRUGO, proc_lstats_operations),
2958 #endif
2959 #ifdef CONFIG_PROC_PID_CPUSET
2960         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
2961 #endif
2962 #ifdef CONFIG_CGROUPS
2963         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2964 #endif
2965         INF("oom_score", S_IRUGO, proc_oom_score),
2966         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2967 #ifdef CONFIG_AUDITSYSCALL
2968         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
2969         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2970 #endif
2971 #ifdef CONFIG_FAULT_INJECTION
2972         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2973 #endif
2974 #ifdef CONFIG_TASK_IO_ACCOUNTING
2975         INF("io",       S_IRUSR, proc_tid_io_accounting),
2976 #endif
2977 #ifdef CONFIG_HARDWALL
2978         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
2979 #endif
2980 #ifdef CONFIG_USER_NS
2981         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2982         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2983         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2984 #endif
2985 };
2986
2987 static int proc_tid_base_readdir(struct file * filp,
2988                              void * dirent, filldir_t filldir)
2989 {
2990         return proc_pident_readdir(filp,dirent,filldir,
2991                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
2992 }
2993
2994 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2995 {
2996         return proc_pident_lookup(dir, dentry,
2997                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
2998 }
2999
3000 static const struct file_operations proc_tid_base_operations = {
3001         .read           = generic_read_dir,
3002         .readdir        = proc_tid_base_readdir,
3003         .llseek         = default_llseek,
3004 };
3005
3006 static const struct inode_operations proc_tid_base_inode_operations = {
3007         .lookup         = proc_tid_base_lookup,
3008         .getattr        = pid_getattr,
3009         .setattr        = proc_setattr,
3010 };
3011
3012 static struct dentry *proc_task_instantiate(struct inode *dir,
3013         struct dentry *dentry, struct task_struct *task, const void *ptr)
3014 {
3015         struct dentry *error = ERR_PTR(-ENOENT);
3016         struct inode *inode;
3017         inode = proc_pid_make_inode(dir->i_sb, task);
3018
3019         if (!inode)
3020                 goto out;
3021         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3022         inode->i_op = &proc_tid_base_inode_operations;
3023         inode->i_fop = &proc_tid_base_operations;
3024         inode->i_flags|=S_IMMUTABLE;
3025
3026         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3027                                                   ARRAY_SIZE(tid_base_stuff)));
3028
3029         d_set_d_op(dentry, &pid_dentry_operations);
3030
3031         d_add(dentry, inode);
3032         /* Close the race of the process dying before we return the dentry */
3033         if (pid_revalidate(dentry, 0))
3034                 error = NULL;
3035 out:
3036         return error;
3037 }
3038
3039 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3040 {
3041         struct dentry *result = ERR_PTR(-ENOENT);
3042         struct task_struct *task;
3043         struct task_struct *leader = get_proc_task(dir);
3044         unsigned tid;
3045         struct pid_namespace *ns;
3046
3047         if (!leader)
3048                 goto out_no_task;
3049
3050         tid = name_to_int(dentry);
3051         if (tid == ~0U)
3052                 goto out;
3053
3054         ns = dentry->d_sb->s_fs_info;
3055         rcu_read_lock();
3056         task = find_task_by_pid_ns(tid, ns);
3057         if (task)
3058                 get_task_struct(task);
3059         rcu_read_unlock();
3060         if (!task)
3061                 goto out;
3062         if (!same_thread_group(leader, task))
3063                 goto out_drop_task;
3064
3065         result = proc_task_instantiate(dir, dentry, task, NULL);
3066 out_drop_task:
3067         put_task_struct(task);
3068 out:
3069         put_task_struct(leader);
3070 out_no_task:
3071         return result;
3072 }
3073
3074 /*
3075  * Find the first tid of a thread group to return to user space.
3076  *
3077  * Usually this is just the thread group leader, but if the users
3078  * buffer was too small or there was a seek into the middle of the
3079  * directory we have more work todo.
3080  *
3081  * In the case of a short read we start with find_task_by_pid.
3082  *
3083  * In the case of a seek we start with the leader and walk nr
3084  * threads past it.
3085  */
3086 static struct task_struct *first_tid(struct task_struct *leader,
3087                 int tid, int nr, struct pid_namespace *ns)
3088 {
3089         struct task_struct *pos;
3090
3091         rcu_read_lock();
3092         /* Attempt to start with the pid of a thread */
3093         if (tid && (nr > 0)) {
3094                 pos = find_task_by_pid_ns(tid, ns);
3095                 if (pos && (pos->group_leader == leader))
3096                         goto found;
3097         }
3098
3099         /* If nr exceeds the number of threads there is nothing todo */
3100         pos = NULL;
3101         if (nr && nr >= get_nr_threads(leader))
3102                 goto out;
3103
3104         /* If we haven't found our starting place yet start
3105          * with the leader and walk nr threads forward.
3106          */
3107         for (pos = leader; nr > 0; --nr) {
3108                 pos = next_thread(pos);
3109                 if (pos == leader) {
3110                         pos = NULL;
3111                         goto out;
3112                 }
3113         }
3114 found:
3115         get_task_struct(pos);
3116 out:
3117         rcu_read_unlock();
3118         return pos;
3119 }
3120
3121 /*
3122  * Find the next thread in the thread list.
3123  * Return NULL if there is an error or no next thread.
3124  *
3125  * The reference to the input task_struct is released.
3126  */
3127 static struct task_struct *next_tid(struct task_struct *start)
3128 {
3129         struct task_struct *pos = NULL;
3130         rcu_read_lock();
3131         if (pid_alive(start)) {
3132                 pos = next_thread(start);
3133                 if (thread_group_leader(pos))
3134                         pos = NULL;
3135                 else
3136                         get_task_struct(pos);
3137         }
3138         rcu_read_unlock();
3139         put_task_struct(start);
3140         return pos;
3141 }
3142
3143 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3144         struct task_struct *task, int tid)
3145 {
3146         char name[PROC_NUMBUF];
3147         int len = snprintf(name, sizeof(name), "%d", tid);
3148         return proc_fill_cache(filp, dirent, filldir, name, len,
3149                                 proc_task_instantiate, task, NULL);
3150 }
3151
3152 /* for the /proc/TGID/task/ directories */
3153 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3154 {
3155         struct dentry *dentry = filp->f_path.dentry;
3156         struct inode *inode = dentry->d_inode;
3157         struct task_struct *leader = NULL;
3158         struct task_struct *task;
3159         int retval = -ENOENT;
3160         ino_t ino;
3161         int tid;
3162         struct pid_namespace *ns;
3163
3164         task = get_proc_task(inode);
3165         if (!task)
3166                 goto out_no_task;
3167         rcu_read_lock();
3168         if (pid_alive(task)) {
3169                 leader = task->group_leader;
3170                 get_task_struct(leader);
3171         }
3172         rcu_read_unlock();
3173         put_task_struct(task);
3174         if (!leader)
3175                 goto out_no_task;
3176         retval = 0;
3177
3178         switch ((unsigned long)filp->f_pos) {
3179         case 0:
3180                 ino = inode->i_ino;
3181                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3182                         goto out;
3183                 filp->f_pos++;
3184                 /* fall through */
3185         case 1:
3186                 ino = parent_ino(dentry);
3187                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3188                         goto out;
3189                 filp->f_pos++;
3190                 /* fall through */
3191         }
3192
3193         /* f_version caches the tgid value that the last readdir call couldn't
3194          * return. lseek aka telldir automagically resets f_version to 0.
3195          */
3196         ns = filp->f_dentry->d_sb->s_fs_info;
3197         tid = (int)filp->f_version;
3198         filp->f_version = 0;
3199         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3200              task;
3201              task = next_tid(task), filp->f_pos++) {
3202                 tid = task_pid_nr_ns(task, ns);
3203                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3204                         /* returning this tgid failed, save it as the first
3205                          * pid for the next readir call */
3206                         filp->f_version = (u64)tid;
3207                         put_task_struct(task);
3208                         break;
3209                 }
3210         }
3211 out:
3212         put_task_struct(leader);
3213 out_no_task:
3214         return retval;
3215 }
3216
3217 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3218 {
3219         struct inode *inode = dentry->d_inode;
3220         struct task_struct *p = get_proc_task(inode);
3221         generic_fillattr(inode, stat);
3222
3223         if (p) {
3224                 stat->nlink += get_nr_threads(p);
3225                 put_task_struct(p);
3226         }
3227
3228         return 0;
3229 }
3230
3231 static const struct inode_operations proc_task_inode_operations = {
3232         .lookup         = proc_task_lookup,
3233         .getattr        = proc_task_getattr,
3234         .setattr        = proc_setattr,
3235         .permission     = proc_pid_permission,
3236 };
3237
3238 static const struct file_operations proc_task_operations = {
3239         .read           = generic_read_dir,
3240         .readdir        = proc_task_readdir,
3241         .llseek         = default_llseek,
3242 };