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_adjust_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         size_t len;
 882         int oom_adjust = OOM_DISABLE;
 883         unsigned long flags;
 884
 885         if (!task)
 886                 return -ESRCH;
 887
 888         if (lock_task_sighand(task, &flags)) {
 889                 oom_adjust = task->signal->oom_adj;
 890                 unlock_task_sighand(task, &flags);
 891         }
 892
 893         put_task_struct(task);
 894
 895         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
 896
 897         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 898 }
 899
 900 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
 901                                 size_t count, loff_t *ppos)
 902 {
 903         struct task_struct *task;
 904         char buffer[PROC_NUMBUF];
 905         int oom_adjust;
 906         unsigned long flags;
 907         int err;
 908
 909         memset(buffer, 0, sizeof(buffer));
 910         if (count > sizeof(buffer) - 1)
 911                 count = sizeof(buffer) - 1;
 912         if (copy_from_user(buffer, buf, count)) {
 913                 err = -EFAULT;
 914                 goto out;
 915         }
 916
 917         err = kstrtoint(strstrip(buffer), 0, &oom_adjust);
 918         if (err)
 919                 goto out;
 920         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
 921              oom_adjust != OOM_DISABLE) {
 922                 err = -EINVAL;
 923                 goto out;
 924         }
 925
 926         task = get_proc_task(file->f_path.dentry->d_inode);
 927         if (!task) {
 928                 err = -ESRCH;
 929                 goto out;
 930         }
 931
 932         task_lock(task);
 933         if (!task->mm) {
 934                 err = -EINVAL;
 935                 goto err_task_lock;
 936         }
 937
 938         if (!lock_task_sighand(task, &flags)) {
 939                 err = -ESRCH;
 940                 goto err_task_lock;
 941         }
 942
 943         if (oom_adjust < task->signal->oom_adj && !capable(CAP_SYS_RESOURCE)) {
 944                 err = -EACCES;
 945                 goto err_sighand;
 946         }
 947
 948         /*
 949          * Warn that /proc/pid/oom_adj is deprecated, see
 950          * Documentation/feature-removal-schedule.txt.
 951          */
 952         printk_once(KERN_WARNING "%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
 953                   current->comm, task_pid_nr(current), task_pid_nr(task),
 954                   task_pid_nr(task));
 955         task->signal->oom_adj = oom_adjust;
 956         /*
 957          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
 958          * value is always attainable.
 959          */
 960         if (task->signal->oom_adj == OOM_ADJUST_MAX)
 961                 task->signal->oom_score_adj = OOM_SCORE_ADJ_MAX;
 962         else
 963                 task->signal->oom_score_adj = (oom_adjust * OOM_SCORE_ADJ_MAX) /
 964                                                                 -OOM_DISABLE;
 965         trace_oom_score_adj_update(task);
 966 err_sighand:
 967         unlock_task_sighand(task, &flags);
 968 err_task_lock:
 969         task_unlock(task);
 970         put_task_struct(task);
 971 out:
 972         return err < 0 ? err : count;
 973 }
 974
 975 static const struct file_operations proc_oom_adjust_operations = {
 976         .read           = oom_adjust_read,
 977         .write          = oom_adjust_write,
 978         .llseek         = generic_file_llseek,
 979 };
 980
 981 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
 982                                         size_t count, loff_t *ppos)
 983 {
 984         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 985         char buffer[PROC_NUMBUF];
 986         int oom_score_adj = OOM_SCORE_ADJ_MIN;
 987         unsigned long flags;
 988         size_t len;
 989
 990         if (!task)
 991                 return -ESRCH;
 992         if (lock_task_sighand(task, &flags)) {
 993                 oom_score_adj = task->signal->oom_score_adj;
 994                 unlock_task_sighand(task, &flags);
 995         }
 996         put_task_struct(task);
 997         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_score_adj);
 998         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 999 }
1000
1001 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1002                                         size_t count, loff_t *ppos)
1003 {
1004         struct task_struct *task;
1005         char buffer[PROC_NUMBUF];
1006         unsigned long flags;
1007         int oom_score_adj;
1008         int err;
1009
1010         memset(buffer, 0, sizeof(buffer));
1011         if (count > sizeof(buffer) - 1)
1012                 count = sizeof(buffer) - 1;
1013         if (copy_from_user(buffer, buf, count)) {
1014                 err = -EFAULT;
1015                 goto out;
1016         }
1017
1018         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1019         if (err)
1020                 goto out;
1021         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1022                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1023                 err = -EINVAL;
1024                 goto out;
1025         }
1026
1027         task = get_proc_task(file->f_path.dentry->d_inode);
1028         if (!task) {
1029                 err = -ESRCH;
1030                 goto out;
1031         }
1032
1033         task_lock(task);
1034         if (!task->mm) {
1035                 err = -EINVAL;
1036                 goto err_task_lock;
1037         }
1038
1039         if (!lock_task_sighand(task, &flags)) {
1040                 err = -ESRCH;
1041                 goto err_task_lock;
1042         }
1043
1044         if (oom_score_adj < task->signal->oom_score_adj_min &&
1045                         !capable(CAP_SYS_RESOURCE)) {
1046                 err = -EACCES;
1047                 goto err_sighand;
1048         }
1049
1050         task->signal->oom_score_adj = oom_score_adj;
1051         if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1052                 task->signal->oom_score_adj_min = oom_score_adj;
1053         trace_oom_score_adj_update(task);
1054         /*
1055          * Scale /proc/pid/oom_adj appropriately ensuring that OOM_DISABLE is
1056          * always attainable.
1057          */
1058         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1059                 task->signal->oom_adj = OOM_DISABLE;
1060         else
1061                 task->signal->oom_adj = (oom_score_adj * OOM_ADJUST_MAX) /
1062                                                         OOM_SCORE_ADJ_MAX;
1063 err_sighand:
1064         unlock_task_sighand(task, &flags);
1065 err_task_lock:
1066         task_unlock(task);
1067         put_task_struct(task);
1068 out:
1069         return err < 0 ? err : count;
1070 }
1071
1072 static const struct file_operations proc_oom_score_adj_operations = {
1073         .read           = oom_score_adj_read,
1074         .write          = oom_score_adj_write,
1075         .llseek         = default_llseek,
1076 };
1077
1078 #ifdef CONFIG_AUDITSYSCALL
1079 #define TMPBUFLEN 21
1080 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1081                                   size_t count, loff_t *ppos)
1082 {
1083         struct inode * inode = file->f_path.dentry->d_inode;
1084         struct task_struct *task = get_proc_task(inode);
1085         ssize_t length;
1086         char tmpbuf[TMPBUFLEN];
1087
1088         if (!task)
1089                 return -ESRCH;
1090         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1091                            from_kuid(file->f_cred->user_ns,
1092                                      audit_get_loginuid(task)));
1093         put_task_struct(task);
1094         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1095 }
1096
1097 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1098                                    size_t count, loff_t *ppos)
1099 {
1100         struct inode * inode = file->f_path.dentry->d_inode;
1101         char *page, *tmp;
1102         ssize_t length;
1103         uid_t loginuid;
1104         kuid_t kloginuid;
1105
1106         rcu_read_lock();
1107         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1108                 rcu_read_unlock();
1109                 return -EPERM;
1110         }
1111         rcu_read_unlock();
1112
1113         if (count >= PAGE_SIZE)
1114                 count = PAGE_SIZE - 1;
1115
1116         if (*ppos != 0) {
1117                 /* No partial writes. */
1118                 return -EINVAL;
1119         }
1120         page = (char*)__get_free_page(GFP_TEMPORARY);
1121         if (!page)
1122                 return -ENOMEM;
1123         length = -EFAULT;
1124         if (copy_from_user(page, buf, count))
1125                 goto out_free_page;
1126
1127         page[count] = '\0';
1128         loginuid = simple_strtoul(page, &tmp, 10);
1129         if (tmp == page) {
1130                 length = -EINVAL;
1131                 goto out_free_page;
1132
1133         }
1134         kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1135         if (!uid_valid(kloginuid)) {
1136                 length = -EINVAL;
1137                 goto out_free_page;
1138         }
1139
1140         length = audit_set_loginuid(kloginuid);
1141         if (likely(length == 0))
1142                 length = count;
1143
1144 out_free_page:
1145         free_page((unsigned long) page);
1146         return length;
1147 }
1148
1149 static const struct file_operations proc_loginuid_operations = {
1150         .read           = proc_loginuid_read,
1151         .write          = proc_loginuid_write,
1152         .llseek         = generic_file_llseek,
1153 };
1154
1155 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1156                                   size_t count, loff_t *ppos)
1157 {
1158         struct inode * inode = file->f_path.dentry->d_inode;
1159         struct task_struct *task = get_proc_task(inode);
1160         ssize_t length;
1161         char tmpbuf[TMPBUFLEN];
1162
1163         if (!task)
1164                 return -ESRCH;
1165         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1166                                 audit_get_sessionid(task));
1167         put_task_struct(task);
1168         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1169 }
1170
1171 static const struct file_operations proc_sessionid_operations = {
1172         .read           = proc_sessionid_read,
1173         .llseek         = generic_file_llseek,
1174 };
1175 #endif
1176
1177 #ifdef CONFIG_FAULT_INJECTION
1178 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1179                                       size_t count, loff_t *ppos)
1180 {
1181         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1182         char buffer[PROC_NUMBUF];
1183         size_t len;
1184         int make_it_fail;
1185
1186         if (!task)
1187                 return -ESRCH;
1188         make_it_fail = task->make_it_fail;
1189         put_task_struct(task);
1190
1191         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1192
1193         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1194 }
1195
1196 static ssize_t proc_fault_inject_write(struct file * file,
1197                         const char __user * buf, size_t count, loff_t *ppos)
1198 {
1199         struct task_struct *task;
1200         char buffer[PROC_NUMBUF], *end;
1201         int make_it_fail;
1202
1203         if (!capable(CAP_SYS_RESOURCE))
1204                 return -EPERM;
1205         memset(buffer, 0, sizeof(buffer));
1206         if (count > sizeof(buffer) - 1)
1207                 count = sizeof(buffer) - 1;
1208         if (copy_from_user(buffer, buf, count))
1209                 return -EFAULT;
1210         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1211         if (*end)
1212                 return -EINVAL;
1213         task = get_proc_task(file->f_dentry->d_inode);
1214         if (!task)
1215                 return -ESRCH;
1216         task->make_it_fail = make_it_fail;
1217         put_task_struct(task);
1218
1219         return count;
1220 }
1221
1222 static const struct file_operations proc_fault_inject_operations = {
1223         .read           = proc_fault_inject_read,
1224         .write          = proc_fault_inject_write,
1225         .llseek         = generic_file_llseek,
1226 };
1227 #endif
1228
1229
1230 #ifdef CONFIG_SCHED_DEBUG
1231 /*
1232  * Print out various scheduling related per-task fields:
1233  */
1234 static int sched_show(struct seq_file *m, void *v)
1235 {
1236         struct inode *inode = m->private;
1237         struct task_struct *p;
1238
1239         p = get_proc_task(inode);
1240         if (!p)
1241                 return -ESRCH;
1242         proc_sched_show_task(p, m);
1243
1244         put_task_struct(p);
1245
1246         return 0;
1247 }
1248
1249 static ssize_t
1250 sched_write(struct file *file, const char __user *buf,
1251             size_t count, loff_t *offset)
1252 {
1253         struct inode *inode = file->f_path.dentry->d_inode;
1254         struct task_struct *p;
1255
1256         p = get_proc_task(inode);
1257         if (!p)
1258                 return -ESRCH;
1259         proc_sched_set_task(p);
1260
1261         put_task_struct(p);
1262
1263         return count;
1264 }
1265
1266 static int sched_open(struct inode *inode, struct file *filp)
1267 {
1268         return single_open(filp, sched_show, inode);
1269 }
1270
1271 static const struct file_operations proc_pid_sched_operations = {
1272         .open           = sched_open,
1273         .read           = seq_read,
1274         .write          = sched_write,
1275         .llseek         = seq_lseek,
1276         .release        = single_release,
1277 };
1278
1279 #endif
1280
1281 #ifdef CONFIG_SCHED_AUTOGROUP
1282 /*
1283  * Print out autogroup related information:
1284  */
1285 static int sched_autogroup_show(struct seq_file *m, void *v)
1286 {
1287         struct inode *inode = m->private;
1288         struct task_struct *p;
1289
1290         p = get_proc_task(inode);
1291         if (!p)
1292                 return -ESRCH;
1293         proc_sched_autogroup_show_task(p, m);
1294
1295         put_task_struct(p);
1296
1297         return 0;
1298 }
1299
1300 static ssize_t
1301 sched_autogroup_write(struct file *file, const char __user *buf,
1302             size_t count, loff_t *offset)
1303 {
1304         struct inode *inode = file->f_path.dentry->d_inode;
1305         struct task_struct *p;
1306         char buffer[PROC_NUMBUF];
1307         int nice;
1308         int err;
1309
1310         memset(buffer, 0, sizeof(buffer));
1311         if (count > sizeof(buffer) - 1)
1312                 count = sizeof(buffer) - 1;
1313         if (copy_from_user(buffer, buf, count))
1314                 return -EFAULT;
1315
1316         err = kstrtoint(strstrip(buffer), 0, &nice);
1317         if (err < 0)
1318                 return err;
1319
1320         p = get_proc_task(inode);
1321         if (!p)
1322                 return -ESRCH;
1323
1324         err = proc_sched_autogroup_set_nice(p, nice);
1325         if (err)
1326                 count = err;
1327
1328         put_task_struct(p);
1329
1330         return count;
1331 }
1332
1333 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1334 {
1335         int ret;
1336
1337         ret = single_open(filp, sched_autogroup_show, NULL);
1338         if (!ret) {
1339                 struct seq_file *m = filp->private_data;
1340
1341                 m->private = inode;
1342         }
1343         return ret;
1344 }
1345
1346 static const struct file_operations proc_pid_sched_autogroup_operations = {
1347         .open           = sched_autogroup_open,
1348         .read           = seq_read,
1349         .write          = sched_autogroup_write,
1350         .llseek         = seq_lseek,
1351         .release        = single_release,
1352 };
1353
1354 #endif /* CONFIG_SCHED_AUTOGROUP */
1355
1356 static ssize_t comm_write(struct file *file, const char __user *buf,
1357                                 size_t count, loff_t *offset)
1358 {
1359         struct inode *inode = file->f_path.dentry->d_inode;
1360         struct task_struct *p;
1361         char buffer[TASK_COMM_LEN];
1362
1363         memset(buffer, 0, sizeof(buffer));
1364         if (count > sizeof(buffer) - 1)
1365                 count = sizeof(buffer) - 1;
1366         if (copy_from_user(buffer, buf, count))
1367                 return -EFAULT;
1368
1369         p = get_proc_task(inode);
1370         if (!p)
1371                 return -ESRCH;
1372
1373         if (same_thread_group(current, p))
1374                 set_task_comm(p, buffer);
1375         else
1376                 count = -EINVAL;
1377
1378         put_task_struct(p);
1379
1380         return count;
1381 }
1382
1383 static int comm_show(struct seq_file *m, void *v)
1384 {
1385         struct inode *inode = m->private;
1386         struct task_struct *p;
1387
1388         p = get_proc_task(inode);
1389         if (!p)
1390                 return -ESRCH;
1391
1392         task_lock(p);
1393         seq_printf(m, "%s\n", p->comm);
1394         task_unlock(p);
1395
1396         put_task_struct(p);
1397
1398         return 0;
1399 }
1400
1401 static int comm_open(struct inode *inode, struct file *filp)
1402 {
1403         return single_open(filp, comm_show, inode);
1404 }
1405
1406 static const struct file_operations proc_pid_set_comm_operations = {
1407         .open           = comm_open,
1408         .read           = seq_read,
1409         .write          = comm_write,
1410         .llseek         = seq_lseek,
1411         .release        = single_release,
1412 };
1413
1414 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1415 {
1416         struct task_struct *task;
1417         struct mm_struct *mm;
1418         struct file *exe_file;
1419
1420         task = get_proc_task(dentry->d_inode);
1421         if (!task)
1422                 return -ENOENT;
1423         mm = get_task_mm(task);
1424         put_task_struct(task);
1425         if (!mm)
1426                 return -ENOENT;
1427         exe_file = get_mm_exe_file(mm);
1428         mmput(mm);
1429         if (exe_file) {
1430                 *exe_path = exe_file->f_path;
1431                 path_get(&exe_file->f_path);
1432                 fput(exe_file);
1433                 return 0;
1434         } else
1435                 return -ENOENT;
1436 }
1437
1438 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1439 {
1440         struct inode *inode = dentry->d_inode;
1441         struct path path;
1442         int error = -EACCES;
1443
1444         /* Are we allowed to snoop on the tasks file descriptors? */
1445         if (!proc_fd_access_allowed(inode))
1446                 goto out;
1447
1448         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1449         if (error)
1450                 goto out;
1451
1452         nd_jump_link(nd, &path);
1453         return NULL;
1454 out:
1455         return ERR_PTR(error);
1456 }
1457
1458 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1459 {
1460         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1461         char *pathname;
1462         int len;
1463
1464         if (!tmp)
1465                 return -ENOMEM;
1466
1467         pathname = d_path(path, tmp, PAGE_SIZE);
1468         len = PTR_ERR(pathname);
1469         if (IS_ERR(pathname))
1470                 goto out;
1471         len = tmp + PAGE_SIZE - 1 - pathname;
1472
1473         if (len > buflen)
1474                 len = buflen;
1475         if (copy_to_user(buffer, pathname, len))
1476                 len = -EFAULT;
1477  out:
1478         free_page((unsigned long)tmp);
1479         return len;
1480 }
1481
1482 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1483 {
1484         int error = -EACCES;
1485         struct inode *inode = dentry->d_inode;
1486         struct path path;
1487
1488         /* Are we allowed to snoop on the tasks file descriptors? */
1489         if (!proc_fd_access_allowed(inode))
1490                 goto out;
1491
1492         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1493         if (error)
1494                 goto out;
1495
1496         error = do_proc_readlink(&path, buffer, buflen);
1497         path_put(&path);
1498 out:
1499         return error;
1500 }
1501
1502 const struct inode_operations proc_pid_link_inode_operations = {
1503         .readlink       = proc_pid_readlink,
1504         .follow_link    = proc_pid_follow_link,
1505         .setattr        = proc_setattr,
1506 };
1507
1508
1509 /* building an inode */
1510
1511 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1512 {
1513         struct inode * inode;
1514         struct proc_inode *ei;
1515         const struct cred *cred;
1516
1517         /* We need a new inode */
1518
1519         inode = new_inode(sb);
1520         if (!inode)
1521                 goto out;
1522
1523         /* Common stuff */
1524         ei = PROC_I(inode);
1525         inode->i_ino = get_next_ino();
1526         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1527         inode->i_op = &proc_def_inode_operations;
1528
1529         /*
1530          * grab the reference to task.
1531          */
1532         ei->pid = get_task_pid(task, PIDTYPE_PID);
1533         if (!ei->pid)
1534                 goto out_unlock;
1535
1536         if (task_dumpable(task)) {
1537                 rcu_read_lock();
1538                 cred = __task_cred(task);
1539                 inode->i_uid = cred->euid;
1540                 inode->i_gid = cred->egid;
1541                 rcu_read_unlock();
1542         }
1543         security_task_to_inode(task, inode);
1544
1545 out:
1546         return inode;
1547
1548 out_unlock:
1549         iput(inode);
1550         return NULL;
1551 }
1552
1553 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1554 {
1555         struct inode *inode = dentry->d_inode;
1556         struct task_struct *task;
1557         const struct cred *cred;
1558         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1559
1560         generic_fillattr(inode, stat);
1561
1562         rcu_read_lock();
1563         stat->uid = GLOBAL_ROOT_UID;
1564         stat->gid = GLOBAL_ROOT_GID;
1565         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1566         if (task) {
1567                 if (!has_pid_permissions(pid, task, 2)) {
1568                         rcu_read_unlock();
1569                         /*
1570                          * This doesn't prevent learning whether PID exists,
1571                          * it only makes getattr() consistent with readdir().
1572                          */
1573                         return -ENOENT;
1574                 }
1575                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1576                     task_dumpable(task)) {
1577                         cred = __task_cred(task);
1578                         stat->uid = cred->euid;
1579                         stat->gid = cred->egid;
1580                 }
1581         }
1582         rcu_read_unlock();
1583         return 0;
1584 }
1585
1586 /* dentry stuff */
1587
1588 /*
1589  *      Exceptional case: normally we are not allowed to unhash a busy
1590  * directory. In this case, however, we can do it - no aliasing problems
1591  * due to the way we treat inodes.
1592  *
1593  * Rewrite the inode's ownerships here because the owning task may have
1594  * performed a setuid(), etc.
1595  *
1596  * Before the /proc/pid/status file was created the only way to read
1597  * the effective uid of a /process was to stat /proc/pid.  Reading
1598  * /proc/pid/status is slow enough that procps and other packages
1599  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1600  * made this apply to all per process world readable and executable
1601  * directories.
1602  */
1603 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1604 {
1605         struct inode *inode;
1606         struct task_struct *task;
1607         const struct cred *cred;
1608
1609         if (flags & LOOKUP_RCU)
1610                 return -ECHILD;
1611
1612         inode = dentry->d_inode;
1613         task = get_proc_task(inode);
1614
1615         if (task) {
1616                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1617                     task_dumpable(task)) {
1618                         rcu_read_lock();
1619                         cred = __task_cred(task);
1620                         inode->i_uid = cred->euid;
1621                         inode->i_gid = cred->egid;
1622                         rcu_read_unlock();
1623                 } else {
1624                         inode->i_uid = GLOBAL_ROOT_UID;
1625                         inode->i_gid = GLOBAL_ROOT_GID;
1626                 }
1627                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1628                 security_task_to_inode(task, inode);
1629                 put_task_struct(task);
1630                 return 1;
1631         }
1632         d_drop(dentry);
1633         return 0;
1634 }
1635
1636 const struct dentry_operations pid_dentry_operations =
1637 {
1638         .d_revalidate   = pid_revalidate,
1639         .d_delete       = pid_delete_dentry,
1640 };
1641
1642 /* Lookups */
1643
1644 /*
1645  * Fill a directory entry.
1646  *
1647  * If possible create the dcache entry and derive our inode number and
1648  * file type from dcache entry.
1649  *
1650  * Since all of the proc inode numbers are dynamically generated, the inode
1651  * numbers do not exist until the inode is cache.  This means creating the
1652  * the dcache entry in readdir is necessary to keep the inode numbers
1653  * reported by readdir in sync with the inode numbers reported
1654  * by stat.
1655  */
1656 int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1657         const char *name, int len,
1658         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1659 {
1660         struct dentry *child, *dir = filp->f_path.dentry;
1661         struct inode *inode;
1662         struct qstr qname;
1663         ino_t ino = 0;
1664         unsigned type = DT_UNKNOWN;
1665
1666         qname.name = name;
1667         qname.len  = len;
1668         qname.hash = full_name_hash(name, len);
1669
1670         child = d_lookup(dir, &qname);
1671         if (!child) {
1672                 struct dentry *new;
1673                 new = d_alloc(dir, &qname);
1674                 if (new) {
1675                         child = instantiate(dir->d_inode, new, task, ptr);
1676                         if (child)
1677                                 dput(new);
1678                         else
1679                                 child = new;
1680                 }
1681         }
1682         if (!child || IS_ERR(child) || !child->d_inode)
1683                 goto end_instantiate;
1684         inode = child->d_inode;
1685         if (inode) {
1686                 ino = inode->i_ino;
1687                 type = inode->i_mode >> 12;
1688         }
1689         dput(child);
1690 end_instantiate:
1691         if (!ino)
1692                 ino = find_inode_number(dir, &qname);
1693         if (!ino)
1694                 ino = 1;
1695         return filldir(dirent, name, len, filp->f_pos, ino, type);
1696 }
1697
1698 #ifdef CONFIG_CHECKPOINT_RESTORE
1699
1700 /*
1701  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1702  * which represent vma start and end addresses.
1703  */
1704 static int dname_to_vma_addr(struct dentry *dentry,
1705                              unsigned long *start, unsigned long *end)
1706 {
1707         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1708                 return -EINVAL;
1709
1710         return 0;
1711 }
1712
1713 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1714 {
1715         unsigned long vm_start, vm_end;
1716         bool exact_vma_exists = false;
1717         struct mm_struct *mm = NULL;
1718         struct task_struct *task;
1719         const struct cred *cred;
1720         struct inode *inode;
1721         int status = 0;
1722
1723         if (flags & LOOKUP_RCU)
1724                 return -ECHILD;
1725
1726         if (!capable(CAP_SYS_ADMIN)) {
1727                 status = -EACCES;
1728                 goto out_notask;
1729         }
1730
1731         inode = dentry->d_inode;
1732         task = get_proc_task(inode);
1733         if (!task)
1734                 goto out_notask;
1735
1736         mm = mm_access(task, PTRACE_MODE_READ);
1737         if (IS_ERR_OR_NULL(mm))
1738                 goto out;
1739
1740         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1741                 down_read(&mm->mmap_sem);
1742                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1743                 up_read(&mm->mmap_sem);
1744         }
1745
1746         mmput(mm);
1747
1748         if (exact_vma_exists) {
1749                 if (task_dumpable(task)) {
1750                         rcu_read_lock();
1751                         cred = __task_cred(task);
1752                         inode->i_uid = cred->euid;
1753                         inode->i_gid = cred->egid;
1754                         rcu_read_unlock();
1755                 } else {
1756                         inode->i_uid = GLOBAL_ROOT_UID;
1757                         inode->i_gid = GLOBAL_ROOT_GID;
1758                 }
1759                 security_task_to_inode(task, inode);
1760                 status = 1;
1761         }
1762
1763 out:
1764         put_task_struct(task);
1765
1766 out_notask:
1767         if (status <= 0)
1768                 d_drop(dentry);
1769
1770         return status;
1771 }
1772
1773 static const struct dentry_operations tid_map_files_dentry_operations = {
1774         .d_revalidate   = map_files_d_revalidate,
1775         .d_delete       = pid_delete_dentry,
1776 };
1777
1778 static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
1779 {
1780         unsigned long vm_start, vm_end;
1781         struct vm_area_struct *vma;
1782         struct task_struct *task;
1783         struct mm_struct *mm;
1784         int rc;
1785
1786         rc = -ENOENT;
1787         task = get_proc_task(dentry->d_inode);
1788         if (!task)
1789                 goto out;
1790
1791         mm = get_task_mm(task);
1792         put_task_struct(task);
1793         if (!mm)
1794                 goto out;
1795
1796         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1797         if (rc)
1798                 goto out_mmput;
1799
1800         down_read(&mm->mmap_sem);
1801         vma = find_exact_vma(mm, vm_start, vm_end);
1802         if (vma && vma->vm_file) {
1803                 *path = vma->vm_file->f_path;
1804                 path_get(path);
1805                 rc = 0;
1806         }
1807         up_read(&mm->mmap_sem);
1808
1809 out_mmput:
1810         mmput(mm);
1811 out:
1812         return rc;
1813 }
1814
1815 struct map_files_info {
1816         fmode_t         mode;
1817         unsigned long   len;
1818         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1819 };
1820
1821 static struct dentry *
1822 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
1823                            struct task_struct *task, const void *ptr)
1824 {
1825         fmode_t mode = (fmode_t)(unsigned long)ptr;
1826         struct proc_inode *ei;
1827         struct inode *inode;
1828
1829         inode = proc_pid_make_inode(dir->i_sb, task);
1830         if (!inode)
1831                 return ERR_PTR(-ENOENT);
1832
1833         ei = PROC_I(inode);
1834         ei->op.proc_get_link = proc_map_files_get_link;
1835
1836         inode->i_op = &proc_pid_link_inode_operations;
1837         inode->i_size = 64;
1838         inode->i_mode = S_IFLNK;
1839
1840         if (mode & FMODE_READ)
1841                 inode->i_mode |= S_IRUSR;
1842         if (mode & FMODE_WRITE)
1843                 inode->i_mode |= S_IWUSR;
1844
1845         d_set_d_op(dentry, &tid_map_files_dentry_operations);
1846         d_add(dentry, inode);
1847
1848         return NULL;
1849 }
1850
1851 static struct dentry *proc_map_files_lookup(struct inode *dir,
1852                 struct dentry *dentry, unsigned int flags)
1853 {
1854         unsigned long vm_start, vm_end;
1855         struct vm_area_struct *vma;
1856         struct task_struct *task;
1857         struct dentry *result;
1858         struct mm_struct *mm;
1859
1860         result = ERR_PTR(-EACCES);
1861         if (!capable(CAP_SYS_ADMIN))
1862                 goto out;
1863
1864         result = ERR_PTR(-ENOENT);
1865         task = get_proc_task(dir);
1866         if (!task)
1867                 goto out;
1868
1869         result = ERR_PTR(-EACCES);
1870         if (!ptrace_may_access(task, PTRACE_MODE_READ))
1871                 goto out_put_task;
1872
1873         result = ERR_PTR(-ENOENT);
1874         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
1875                 goto out_put_task;
1876
1877         mm = get_task_mm(task);
1878         if (!mm)
1879                 goto out_put_task;
1880
1881         down_read(&mm->mmap_sem);
1882         vma = find_exact_vma(mm, vm_start, vm_end);
1883         if (!vma)
1884                 goto out_no_vma;
1885
1886         result = proc_map_files_instantiate(dir, dentry, task,
1887                         (void *)(unsigned long)vma->vm_file->f_mode);
1888
1889 out_no_vma:
1890         up_read(&mm->mmap_sem);
1891         mmput(mm);
1892 out_put_task:
1893         put_task_struct(task);
1894 out:
1895         return result;
1896 }
1897
1898 static const struct inode_operations proc_map_files_inode_operations = {
1899         .lookup         = proc_map_files_lookup,
1900         .permission     = proc_fd_permission,
1901         .setattr        = proc_setattr,
1902 };
1903
1904 static int
1905 proc_map_files_readdir(struct file *filp, void *dirent, filldir_t filldir)
1906 {
1907         struct dentry *dentry = filp->f_path.dentry;
1908         struct inode *inode = dentry->d_inode;
1909         struct vm_area_struct *vma;
1910         struct task_struct *task;
1911         struct mm_struct *mm;
1912         ino_t ino;
1913         int ret;
1914
1915         ret = -EACCES;
1916         if (!capable(CAP_SYS_ADMIN))
1917                 goto out;
1918
1919         ret = -ENOENT;
1920         task = get_proc_task(inode);
1921         if (!task)
1922                 goto out;
1923
1924         ret = -EACCES;
1925         if (!ptrace_may_access(task, PTRACE_MODE_READ))
1926                 goto out_put_task;
1927
1928         ret = 0;
1929         switch (filp->f_pos) {
1930         case 0:
1931                 ino = inode->i_ino;
1932                 if (filldir(dirent, ".", 1, 0, ino, DT_DIR) < 0)
1933                         goto out_put_task;
1934                 filp->f_pos++;
1935         case 1:
1936                 ino = parent_ino(dentry);
1937                 if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1938                         goto out_put_task;
1939                 filp->f_pos++;
1940         default:
1941         {
1942                 unsigned long nr_files, pos, i;
1943                 struct flex_array *fa = NULL;
1944                 struct map_files_info info;
1945                 struct map_files_info *p;
1946
1947                 mm = get_task_mm(task);
1948                 if (!mm)
1949                         goto out_put_task;
1950                 down_read(&mm->mmap_sem);
1951
1952                 nr_files = 0;
1953
1954                 /*
1955                  * We need two passes here:
1956                  *
1957                  *  1) Collect vmas of mapped files with mmap_sem taken
1958                  *  2) Release mmap_sem and instantiate entries
1959                  *
1960                  * otherwise we get lockdep complained, since filldir()
1961                  * routine might require mmap_sem taken in might_fault().
1962                  */
1963
1964                 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
1965                         if (vma->vm_file && ++pos > filp->f_pos)
1966                                 nr_files++;
1967                 }
1968
1969                 if (nr_files) {
1970                         fa = flex_array_alloc(sizeof(info), nr_files,
1971                                                 GFP_KERNEL);
1972                         if (!fa || flex_array_prealloc(fa, 0, nr_files,
1973                                                         GFP_KERNEL)) {
1974                                 ret = -ENOMEM;
1975                                 if (fa)
1976                                         flex_array_free(fa);
1977                                 up_read(&mm->mmap_sem);
1978                                 mmput(mm);
1979                                 goto out_put_task;
1980                         }
1981                         for (i = 0, vma = mm->mmap, pos = 2; vma;
1982                                         vma = vma->vm_next) {
1983                                 if (!vma->vm_file)
1984                                         continue;
1985                                 if (++pos <= filp->f_pos)
1986                                         continue;
1987
1988                                 info.mode = vma->vm_file->f_mode;
1989                                 info.len = snprintf(info.name,
1990                                                 sizeof(info.name), "%lx-%lx",
1991                                                 vma->vm_start, vma->vm_end);
1992                                 if (flex_array_put(fa, i++, &info, GFP_KERNEL))
1993                                         BUG();
1994                         }
1995                 }
1996                 up_read(&mm->mmap_sem);
1997
1998                 for (i = 0; i < nr_files; i++) {
1999                         p = flex_array_get(fa, i);
2000                         ret = proc_fill_cache(filp, dirent, filldir,
2001                                               p->name, p->len,
2002                                               proc_map_files_instantiate,
2003                                               task,
2004                                               (void *)(unsigned long)p->mode);
2005                         if (ret)
2006                                 break;
2007                         filp->f_pos++;
2008                 }
2009                 if (fa)
2010                         flex_array_free(fa);
2011                 mmput(mm);
2012         }
2013         }
2014
2015 out_put_task:
2016         put_task_struct(task);
2017 out:
2018         return ret;
2019 }
2020
2021 static const struct file_operations proc_map_files_operations = {
2022         .read           = generic_read_dir,
2023         .readdir        = proc_map_files_readdir,
2024         .llseek         = default_llseek,
2025 };
2026
2027 #endif /* CONFIG_CHECKPOINT_RESTORE */
2028
2029 static struct dentry *proc_pident_instantiate(struct inode *dir,
2030         struct dentry *dentry, struct task_struct *task, const void *ptr)
2031 {
2032         const struct pid_entry *p = ptr;
2033         struct inode *inode;
2034         struct proc_inode *ei;
2035         struct dentry *error = ERR_PTR(-ENOENT);
2036
2037         inode = proc_pid_make_inode(dir->i_sb, task);
2038         if (!inode)
2039                 goto out;
2040
2041         ei = PROC_I(inode);
2042         inode->i_mode = p->mode;
2043         if (S_ISDIR(inode->i_mode))
2044                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2045         if (p->iop)
2046                 inode->i_op = p->iop;
2047         if (p->fop)
2048                 inode->i_fop = p->fop;
2049         ei->op = p->op;
2050         d_set_d_op(dentry, &pid_dentry_operations);
2051         d_add(dentry, inode);
2052         /* Close the race of the process dying before we return the dentry */
2053         if (pid_revalidate(dentry, 0))
2054                 error = NULL;
2055 out:
2056         return error;
2057 }
2058
2059 static struct dentry *proc_pident_lookup(struct inode *dir,
2060                                          struct dentry *dentry,
2061                                          const struct pid_entry *ents,
2062                                          unsigned int nents)
2063 {
2064         struct dentry *error;
2065         struct task_struct *task = get_proc_task(dir);
2066         const struct pid_entry *p, *last;
2067
2068         error = ERR_PTR(-ENOENT);
2069
2070         if (!task)
2071                 goto out_no_task;
2072
2073         /*
2074          * Yes, it does not scale. And it should not. Don't add
2075          * new entries into /proc/<tgid>/ without very good reasons.
2076          */
2077         last = &ents[nents - 1];
2078         for (p = ents; p <= last; p++) {
2079                 if (p->len != dentry->d_name.len)
2080                         continue;
2081                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2082                         break;
2083         }
2084         if (p > last)
2085                 goto out;
2086
2087         error = proc_pident_instantiate(dir, dentry, task, p);
2088 out:
2089         put_task_struct(task);
2090 out_no_task:
2091         return error;
2092 }
2093
2094 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2095         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2096 {
2097         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2098                                 proc_pident_instantiate, task, p);
2099 }
2100
2101 static int proc_pident_readdir(struct file *filp,
2102                 void *dirent, filldir_t filldir,
2103                 const struct pid_entry *ents, unsigned int nents)
2104 {
2105         int i;
2106         struct dentry *dentry = filp->f_path.dentry;
2107         struct inode *inode = dentry->d_inode;
2108         struct task_struct *task = get_proc_task(inode);
2109         const struct pid_entry *p, *last;
2110         ino_t ino;
2111         int ret;
2112
2113         ret = -ENOENT;
2114         if (!task)
2115                 goto out_no_task;
2116
2117         ret = 0;
2118         i = filp->f_pos;
2119         switch (i) {
2120         case 0:
2121                 ino = inode->i_ino;
2122                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2123                         goto out;
2124                 i++;
2125                 filp->f_pos++;
2126                 /* fall through */
2127         case 1:
2128                 ino = parent_ino(dentry);
2129                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2130                         goto out;
2131                 i++;
2132                 filp->f_pos++;
2133                 /* fall through */
2134         default:
2135                 i -= 2;
2136                 if (i >= nents) {
2137                         ret = 1;
2138                         goto out;
2139                 }
2140                 p = ents + i;
2141                 last = &ents[nents - 1];
2142                 while (p <= last) {
2143                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2144                                 goto out;
2145                         filp->f_pos++;
2146                         p++;
2147                 }
2148         }
2149
2150         ret = 1;
2151 out:
2152         put_task_struct(task);
2153 out_no_task:
2154         return ret;
2155 }
2156
2157 #ifdef CONFIG_SECURITY
2158 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2159                                   size_t count, loff_t *ppos)
2160 {
2161         struct inode * inode = file->f_path.dentry->d_inode;
2162         char *p = NULL;
2163         ssize_t length;
2164         struct task_struct *task = get_proc_task(inode);
2165
2166         if (!task)
2167                 return -ESRCH;
2168
2169         length = security_getprocattr(task,
2170                                       (char*)file->f_path.dentry->d_name.name,
2171                                       &p);
2172         put_task_struct(task);
2173         if (length > 0)
2174                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2175         kfree(p);
2176         return length;
2177 }
2178
2179 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2180                                    size_t count, loff_t *ppos)
2181 {
2182         struct inode * inode = file->f_path.dentry->d_inode;
2183         char *page;
2184         ssize_t length;
2185         struct task_struct *task = get_proc_task(inode);
2186
2187         length = -ESRCH;
2188         if (!task)
2189                 goto out_no_task;
2190         if (count > PAGE_SIZE)
2191                 count = PAGE_SIZE;
2192
2193         /* No partial writes. */
2194         length = -EINVAL;
2195         if (*ppos != 0)
2196                 goto out;
2197
2198         length = -ENOMEM;
2199         page = (char*)__get_free_page(GFP_TEMPORARY);
2200         if (!page)
2201                 goto out;
2202
2203         length = -EFAULT;
2204         if (copy_from_user(page, buf, count))
2205                 goto out_free;
2206
2207         /* Guard against adverse ptrace interaction */
2208         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2209         if (length < 0)
2210                 goto out_free;
2211
2212         length = security_setprocattr(task,
2213                                       (char*)file->f_path.dentry->d_name.name,
2214                                       (void*)page, count);
2215         mutex_unlock(&task->signal->cred_guard_mutex);
2216 out_free:
2217         free_page((unsigned long) page);
2218 out:
2219         put_task_struct(task);
2220 out_no_task:
2221         return length;
2222 }
2223
2224 static const struct file_operations proc_pid_attr_operations = {
2225         .read           = proc_pid_attr_read,
2226         .write          = proc_pid_attr_write,
2227         .llseek         = generic_file_llseek,
2228 };
2229
2230 static const struct pid_entry attr_dir_stuff[] = {
2231         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2232         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2233         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2234         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2235         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2236         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2237 };
2238
2239 static int proc_attr_dir_readdir(struct file * filp,
2240                              void * dirent, filldir_t filldir)
2241 {
2242         return proc_pident_readdir(filp,dirent,filldir,
2243                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2244 }
2245
2246 static const struct file_operations proc_attr_dir_operations = {
2247         .read           = generic_read_dir,
2248         .readdir        = proc_attr_dir_readdir,
2249         .llseek         = default_llseek,
2250 };
2251
2252 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2253                                 struct dentry *dentry, unsigned int flags)
2254 {
2255         return proc_pident_lookup(dir, dentry,
2256                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2257 }
2258
2259 static const struct inode_operations proc_attr_dir_inode_operations = {
2260         .lookup         = proc_attr_dir_lookup,
2261         .getattr        = pid_getattr,
2262         .setattr        = proc_setattr,
2263 };
2264
2265 #endif
2266
2267 #ifdef CONFIG_ELF_CORE
2268 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2269                                          size_t count, loff_t *ppos)
2270 {
2271         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2272         struct mm_struct *mm;
2273         char buffer[PROC_NUMBUF];
2274         size_t len;
2275         int ret;
2276
2277         if (!task)
2278                 return -ESRCH;
2279
2280         ret = 0;
2281         mm = get_task_mm(task);
2282         if (mm) {
2283                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2284                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2285                                 MMF_DUMP_FILTER_SHIFT));
2286                 mmput(mm);
2287                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2288         }
2289
2290         put_task_struct(task);
2291
2292         return ret;
2293 }
2294
2295 static ssize_t proc_coredump_filter_write(struct file *file,
2296                                           const char __user *buf,
2297                                           size_t count,
2298                                           loff_t *ppos)
2299 {
2300         struct task_struct *task;
2301         struct mm_struct *mm;
2302         char buffer[PROC_NUMBUF], *end;
2303         unsigned int val;
2304         int ret;
2305         int i;
2306         unsigned long mask;
2307
2308         ret = -EFAULT;
2309         memset(buffer, 0, sizeof(buffer));
2310         if (count > sizeof(buffer) - 1)
2311                 count = sizeof(buffer) - 1;
2312         if (copy_from_user(buffer, buf, count))
2313                 goto out_no_task;
2314
2315         ret = -EINVAL;
2316         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2317         if (*end == '\n')
2318                 end++;
2319         if (end - buffer == 0)
2320                 goto out_no_task;
2321
2322         ret = -ESRCH;
2323         task = get_proc_task(file->f_dentry->d_inode);
2324         if (!task)
2325                 goto out_no_task;
2326
2327         ret = end - buffer;
2328         mm = get_task_mm(task);
2329         if (!mm)
2330                 goto out_no_mm;
2331
2332         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2333                 if (val & mask)
2334                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2335                 else
2336                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2337         }
2338
2339         mmput(mm);
2340  out_no_mm:
2341         put_task_struct(task);
2342  out_no_task:
2343         return ret;
2344 }
2345
2346 static const struct file_operations proc_coredump_filter_operations = {
2347         .read           = proc_coredump_filter_read,
2348         .write          = proc_coredump_filter_write,
2349         .llseek         = generic_file_llseek,
2350 };
2351 #endif
2352
2353 /*
2354  * /proc/self:
2355  */
2356 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2357                               int buflen)
2358 {
2359         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2360         pid_t tgid = task_tgid_nr_ns(current, ns);
2361         char tmp[PROC_NUMBUF];
2362         if (!tgid)
2363                 return -ENOENT;
2364         sprintf(tmp, "%d", tgid);
2365         return vfs_readlink(dentry,buffer,buflen,tmp);
2366 }
2367
2368 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2369 {
2370         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2371         pid_t tgid = task_tgid_nr_ns(current, ns);
2372         char *name = ERR_PTR(-ENOENT);
2373         if (tgid) {
2374                 name = __getname();
2375                 if (!name)
2376                         name = ERR_PTR(-ENOMEM);
2377                 else
2378                         sprintf(name, "%d", tgid);
2379         }
2380         nd_set_link(nd, name);
2381         return NULL;
2382 }
2383
2384 static void proc_self_put_link(struct dentry *dentry, struct nameidata *nd,
2385                                 void *cookie)
2386 {
2387         char *s = nd_get_link(nd);
2388         if (!IS_ERR(s))
2389                 __putname(s);
2390 }
2391
2392 static const struct inode_operations proc_self_inode_operations = {
2393         .readlink       = proc_self_readlink,
2394         .follow_link    = proc_self_follow_link,
2395         .put_link       = proc_self_put_link,
2396 };
2397
2398 /*
2399  * proc base
2400  *
2401  * These are the directory entries in the root directory of /proc
2402  * that properly belong to the /proc filesystem, as they describe
2403  * describe something that is process related.
2404  */
2405 static const struct pid_entry proc_base_stuff[] = {
2406         NOD("self", S_IFLNK|S_IRWXUGO,
2407                 &proc_self_inode_operations, NULL, {}),
2408 };
2409
2410 static struct dentry *proc_base_instantiate(struct inode *dir,
2411         struct dentry *dentry, struct task_struct *task, const void *ptr)
2412 {
2413         const struct pid_entry *p = ptr;
2414         struct inode *inode;
2415         struct proc_inode *ei;
2416         struct dentry *error;
2417
2418         /* Allocate the inode */
2419         error = ERR_PTR(-ENOMEM);
2420         inode = new_inode(dir->i_sb);
2421         if (!inode)
2422                 goto out;
2423
2424         /* Initialize the inode */
2425         ei = PROC_I(inode);
2426         inode->i_ino = get_next_ino();
2427         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2428
2429         /*
2430          * grab the reference to the task.
2431          */
2432         ei->pid = get_task_pid(task, PIDTYPE_PID);
2433         if (!ei->pid)
2434                 goto out_iput;
2435
2436         inode->i_mode = p->mode;
2437         if (S_ISDIR(inode->i_mode))
2438                 set_nlink(inode, 2);
2439         if (S_ISLNK(inode->i_mode))
2440                 inode->i_size = 64;
2441         if (p->iop)
2442                 inode->i_op = p->iop;
2443         if (p->fop)
2444                 inode->i_fop = p->fop;
2445         ei->op = p->op;
2446         d_add(dentry, inode);
2447         error = NULL;
2448 out:
2449         return error;
2450 out_iput:
2451         iput(inode);
2452         goto out;
2453 }
2454
2455 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2456 {
2457         struct dentry *error;
2458         struct task_struct *task = get_proc_task(dir);
2459         const struct pid_entry *p, *last;
2460
2461         error = ERR_PTR(-ENOENT);
2462
2463         if (!task)
2464                 goto out_no_task;
2465
2466         /* Lookup the directory entry */
2467         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2468         for (p = proc_base_stuff; p <= last; p++) {
2469                 if (p->len != dentry->d_name.len)
2470                         continue;
2471                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2472                         break;
2473         }
2474         if (p > last)
2475                 goto out;
2476
2477         error = proc_base_instantiate(dir, dentry, task, p);
2478
2479 out:
2480         put_task_struct(task);
2481 out_no_task:
2482         return error;
2483 }
2484
2485 static int proc_base_fill_cache(struct file *filp, void *dirent,
2486         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2487 {
2488         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2489                                 proc_base_instantiate, task, p);
2490 }
2491
2492 #ifdef CONFIG_TASK_IO_ACCOUNTING
2493 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2494 {
2495         struct task_io_accounting acct = task->ioac;
2496         unsigned long flags;
2497         int result;
2498
2499         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2500         if (result)
2501                 return result;
2502
2503         if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2504                 result = -EACCES;
2505                 goto out_unlock;
2506         }
2507
2508         if (whole && lock_task_sighand(task, &flags)) {
2509                 struct task_struct *t = task;
2510
2511                 task_io_accounting_add(&acct, &task->signal->ioac);
2512                 while_each_thread(task, t)
2513                         task_io_accounting_add(&acct, &t->ioac);
2514
2515                 unlock_task_sighand(task, &flags);
2516         }
2517         result = sprintf(buffer,
2518                         "rchar: %llu\n"
2519                         "wchar: %llu\n"
2520                         "syscr: %llu\n"
2521                         "syscw: %llu\n"
2522                         "read_bytes: %llu\n"
2523                         "write_bytes: %llu\n"
2524                         "cancelled_write_bytes: %llu\n",
2525                         (unsigned long long)acct.rchar,
2526                         (unsigned long long)acct.wchar,
2527                         (unsigned long long)acct.syscr,
2528                         (unsigned long long)acct.syscw,
2529                         (unsigned long long)acct.read_bytes,
2530                         (unsigned long long)acct.write_bytes,
2531                         (unsigned long long)acct.cancelled_write_bytes);
2532 out_unlock:
2533         mutex_unlock(&task->signal->cred_guard_mutex);
2534         return result;
2535 }
2536
2537 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2538 {
2539         return do_io_accounting(task, buffer, 0);
2540 }
2541
2542 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2543 {
2544         return do_io_accounting(task, buffer, 1);
2545 }
2546 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2547
2548 #ifdef CONFIG_USER_NS
2549 static int proc_id_map_open(struct inode *inode, struct file *file,
2550         struct seq_operations *seq_ops)
2551 {
2552         struct user_namespace *ns = NULL;
2553         struct task_struct *task;
2554         struct seq_file *seq;
2555         int ret = -EINVAL;
2556
2557         task = get_proc_task(inode);
2558         if (task) {
2559                 rcu_read_lock();
2560                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2561                 rcu_read_unlock();
2562                 put_task_struct(task);
2563         }
2564         if (!ns)
2565                 goto err;
2566
2567         ret = seq_open(file, seq_ops);
2568         if (ret)
2569                 goto err_put_ns;
2570
2571         seq = file->private_data;
2572         seq->private = ns;
2573
2574         return 0;
2575 err_put_ns:
2576         put_user_ns(ns);
2577 err:
2578         return ret;
2579 }
2580
2581 static int proc_id_map_release(struct inode *inode, struct file *file)
2582 {
2583         struct seq_file *seq = file->private_data;
2584         struct user_namespace *ns = seq->private;
2585         put_user_ns(ns);
2586         return seq_release(inode, file);
2587 }
2588
2589 static int proc_uid_map_open(struct inode *inode, struct file *file)
2590 {
2591         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2592 }
2593
2594 static int proc_gid_map_open(struct inode *inode, struct file *file)
2595 {
2596         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2597 }
2598
2599 static int proc_projid_map_open(struct inode *inode, struct file *file)
2600 {
2601         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2602 }
2603
2604 static const struct file_operations proc_uid_map_operations = {
2605         .open           = proc_uid_map_open,
2606         .write          = proc_uid_map_write,
2607         .read           = seq_read,
2608         .llseek         = seq_lseek,
2609         .release        = proc_id_map_release,
2610 };
2611
2612 static const struct file_operations proc_gid_map_operations = {
2613         .open           = proc_gid_map_open,
2614         .write          = proc_gid_map_write,
2615         .read           = seq_read,
2616         .llseek         = seq_lseek,
2617         .release        = proc_id_map_release,
2618 };
2619
2620 static const struct file_operations proc_projid_map_operations = {
2621         .open           = proc_projid_map_open,
2622         .write          = proc_projid_map_write,
2623         .read           = seq_read,
2624         .llseek         = seq_lseek,
2625         .release        = proc_id_map_release,
2626 };
2627 #endif /* CONFIG_USER_NS */
2628
2629 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2630                                 struct pid *pid, struct task_struct *task)
2631 {
2632         int err = lock_trace(task);
2633         if (!err) {
2634                 seq_printf(m, "%08x\n", task->personality);
2635                 unlock_trace(task);
2636         }
2637         return err;
2638 }
2639
2640 /*
2641  * Thread groups
2642  */
2643 static const struct file_operations proc_task_operations;
2644 static const struct inode_operations proc_task_inode_operations;
2645
2646 static const struct pid_entry tgid_base_stuff[] = {
2647         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2648         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2649 #ifdef CONFIG_CHECKPOINT_RESTORE
2650         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2651 #endif
2652         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2653         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2654 #ifdef CONFIG_NET
2655         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2656 #endif
2657         REG("environ",    S_IRUSR, proc_environ_operations),
2658         INF("auxv",       S_IRUSR, proc_pid_auxv),
2659         ONE("status",     S_IRUGO, proc_pid_status),
2660         ONE("personality", S_IRUGO, proc_pid_personality),
2661         INF("limits",     S_IRUGO, proc_pid_limits),
2662 #ifdef CONFIG_SCHED_DEBUG
2663         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2664 #endif
2665 #ifdef CONFIG_SCHED_AUTOGROUP
2666         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2667 #endif
2668         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2669 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2670         INF("syscall",    S_IRUGO, proc_pid_syscall),
2671 #endif
2672         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2673         ONE("stat",       S_IRUGO, proc_tgid_stat),
2674         ONE("statm",      S_IRUGO, proc_pid_statm),
2675         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2676 #ifdef CONFIG_NUMA
2677         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2678 #endif
2679         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2680         LNK("cwd",        proc_cwd_link),
2681         LNK("root",       proc_root_link),
2682         LNK("exe",        proc_exe_link),
2683         REG("mounts",     S_IRUGO, proc_mounts_operations),
2684         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2685         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2686 #ifdef CONFIG_PROC_PAGE_MONITOR
2687         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2688         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2689         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
2690 #endif
2691 #ifdef CONFIG_SECURITY
2692         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2693 #endif
2694 #ifdef CONFIG_KALLSYMS
2695         INF("wchan",      S_IRUGO, proc_pid_wchan),
2696 #endif
2697 #ifdef CONFIG_STACKTRACE
2698         ONE("stack",      S_IRUGO, proc_pid_stack),
2699 #endif
2700 #ifdef CONFIG_SCHEDSTATS
2701         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
2702 #endif
2703 #ifdef CONFIG_LATENCYTOP
2704         REG("latency",  S_IRUGO, proc_lstats_operations),
2705 #endif
2706 #ifdef CONFIG_PROC_PID_CPUSET
2707         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2708 #endif
2709 #ifdef CONFIG_CGROUPS
2710         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2711 #endif
2712         INF("oom_score",  S_IRUGO, proc_oom_score),
2713         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2714         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2715 #ifdef CONFIG_AUDITSYSCALL
2716         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2717         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2718 #endif
2719 #ifdef CONFIG_FAULT_INJECTION
2720         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2721 #endif
2722 #ifdef CONFIG_ELF_CORE
2723         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2724 #endif
2725 #ifdef CONFIG_TASK_IO_ACCOUNTING
2726         INF("io",       S_IRUSR, proc_tgid_io_accounting),
2727 #endif
2728 #ifdef CONFIG_HARDWALL
2729         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
2730 #endif
2731 #ifdef CONFIG_USER_NS
2732         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2733         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2734         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2735 #endif
2736 };
2737
2738 static int proc_tgid_base_readdir(struct file * filp,
2739                              void * dirent, filldir_t filldir)
2740 {
2741         return proc_pident_readdir(filp,dirent,filldir,
2742                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2743 }
2744
2745 static const struct file_operations proc_tgid_base_operations = {
2746         .read           = generic_read_dir,
2747         .readdir        = proc_tgid_base_readdir,
2748         .llseek         = default_llseek,
2749 };
2750
2751 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2752 {
2753         return proc_pident_lookup(dir, dentry,
2754                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2755 }
2756
2757 static const struct inode_operations proc_tgid_base_inode_operations = {
2758         .lookup         = proc_tgid_base_lookup,
2759         .getattr        = pid_getattr,
2760         .setattr        = proc_setattr,
2761         .permission     = proc_pid_permission,
2762 };
2763
2764 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2765 {
2766         struct dentry *dentry, *leader, *dir;
2767         char buf[PROC_NUMBUF];
2768         struct qstr name;
2769
2770         name.name = buf;
2771         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2772         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2773         if (dentry) {
2774                 shrink_dcache_parent(dentry);
2775                 d_drop(dentry);
2776                 dput(dentry);
2777         }
2778
2779         name.name = buf;
2780         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2781         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2782         if (!leader)
2783                 goto out;
2784
2785         name.name = "task";
2786         name.len = strlen(name.name);
2787         dir = d_hash_and_lookup(leader, &name);
2788         if (!dir)
2789                 goto out_put_leader;
2790
2791         name.name = buf;
2792         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2793         dentry = d_hash_and_lookup(dir, &name);
2794         if (dentry) {
2795                 shrink_dcache_parent(dentry);
2796                 d_drop(dentry);
2797                 dput(dentry);
2798         }
2799
2800         dput(dir);
2801 out_put_leader:
2802         dput(leader);
2803 out:
2804         return;
2805 }
2806
2807 /**
2808  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2809  * @task: task that should be flushed.
2810  *
2811  * When flushing dentries from proc, one needs to flush them from global
2812  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2813  * in. This call is supposed to do all of this job.
2814  *
2815  * Looks in the dcache for
2816  * /proc/@pid
2817  * /proc/@tgid/task/@pid
2818  * if either directory is present flushes it and all of it'ts children
2819  * from the dcache.
2820  *
2821  * It is safe and reasonable to cache /proc entries for a task until
2822  * that task exits.  After that they just clog up the dcache with
2823  * useless entries, possibly causing useful dcache entries to be
2824  * flushed instead.  This routine is proved to flush those useless
2825  * dcache entries at process exit time.
2826  *
2827  * NOTE: This routine is just an optimization so it does not guarantee
2828  *       that no dcache entries will exist at process exit time it
2829  *       just makes it very unlikely that any will persist.
2830  */
2831
2832 void proc_flush_task(struct task_struct *task)
2833 {
2834         int i;
2835         struct pid *pid, *tgid;
2836         struct upid *upid;
2837
2838         pid = task_pid(task);
2839         tgid = task_tgid(task);
2840
2841         for (i = 0; i <= pid->level; i++) {
2842                 upid = &pid->numbers[i];
2843                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2844                                         tgid->numbers[i].nr);
2845         }
2846
2847         upid = &pid->numbers[pid->level];
2848         if (upid->nr == 1)
2849                 pid_ns_release_proc(upid->ns);
2850 }
2851
2852 static struct dentry *proc_pid_instantiate(struct inode *dir,
2853                                            struct dentry * dentry,
2854                                            struct task_struct *task, const void *ptr)
2855 {
2856         struct dentry *error = ERR_PTR(-ENOENT);
2857         struct inode *inode;
2858
2859         inode = proc_pid_make_inode(dir->i_sb, task);
2860         if (!inode)
2861                 goto out;
2862
2863         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2864         inode->i_op = &proc_tgid_base_inode_operations;
2865         inode->i_fop = &proc_tgid_base_operations;
2866         inode->i_flags|=S_IMMUTABLE;
2867
2868         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
2869                                                   ARRAY_SIZE(tgid_base_stuff)));
2870
2871         d_set_d_op(dentry, &pid_dentry_operations);
2872
2873         d_add(dentry, inode);
2874         /* Close the race of the process dying before we return the dentry */
2875         if (pid_revalidate(dentry, 0))
2876                 error = NULL;
2877 out:
2878         return error;
2879 }
2880
2881 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
2882 {
2883         struct dentry *result;
2884         struct task_struct *task;
2885         unsigned tgid;
2886         struct pid_namespace *ns;
2887
2888         result = proc_base_lookup(dir, dentry);
2889         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
2890                 goto out;
2891
2892         tgid = name_to_int(dentry);
2893         if (tgid == ~0U)
2894                 goto out;
2895
2896         ns = dentry->d_sb->s_fs_info;
2897         rcu_read_lock();
2898         task = find_task_by_pid_ns(tgid, ns);
2899         if (task)
2900                 get_task_struct(task);
2901         rcu_read_unlock();
2902         if (!task)
2903                 goto out;
2904
2905         result = proc_pid_instantiate(dir, dentry, task, NULL);
2906         put_task_struct(task);
2907 out:
2908         return result;
2909 }
2910
2911 /*
2912  * Find the first task with tgid >= tgid
2913  *
2914  */
2915 struct tgid_iter {
2916         unsigned int tgid;
2917         struct task_struct *task;
2918 };
2919 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2920 {
2921         struct pid *pid;
2922
2923         if (iter.task)
2924                 put_task_struct(iter.task);
2925         rcu_read_lock();
2926 retry:
2927         iter.task = NULL;
2928         pid = find_ge_pid(iter.tgid, ns);
2929         if (pid) {
2930                 iter.tgid = pid_nr_ns(pid, ns);
2931                 iter.task = pid_task(pid, PIDTYPE_PID);
2932                 /* What we to know is if the pid we have find is the
2933                  * pid of a thread_group_leader.  Testing for task
2934                  * being a thread_group_leader is the obvious thing
2935                  * todo but there is a window when it fails, due to
2936                  * the pid transfer logic in de_thread.
2937                  *
2938                  * So we perform the straight forward test of seeing
2939                  * if the pid we have found is the pid of a thread
2940                  * group leader, and don't worry if the task we have
2941                  * found doesn't happen to be a thread group leader.
2942                  * As we don't care in the case of readdir.
2943                  */
2944                 if (!iter.task || !has_group_leader_pid(iter.task)) {
2945                         iter.tgid += 1;
2946                         goto retry;
2947                 }
2948                 get_task_struct(iter.task);
2949         }
2950         rcu_read_unlock();
2951         return iter;
2952 }
2953
2954 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
2955
2956 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2957         struct tgid_iter iter)
2958 {
2959         char name[PROC_NUMBUF];
2960         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
2961         return proc_fill_cache(filp, dirent, filldir, name, len,
2962                                 proc_pid_instantiate, iter.task, NULL);
2963 }
2964
2965 static int fake_filldir(void *buf, const char *name, int namelen,
2966                         loff_t offset, u64 ino, unsigned d_type)
2967 {
2968         return 0;
2969 }
2970
2971 /* for the /proc/ directory itself, after non-process stuff has been done */
2972 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2973 {
2974         unsigned int nr;
2975         struct task_struct *reaper;
2976         struct tgid_iter iter;
2977         struct pid_namespace *ns;
2978         filldir_t __filldir;
2979
2980         if (filp->f_pos >= PID_MAX_LIMIT + TGID_OFFSET)
2981                 goto out_no_task;
2982         nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2983
2984         reaper = get_proc_task(filp->f_path.dentry->d_inode);
2985         if (!reaper)
2986                 goto out_no_task;
2987
2988         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
2989                 const struct pid_entry *p = &proc_base_stuff[nr];
2990                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
2991                         goto out;
2992         }
2993
2994         ns = filp->f_dentry->d_sb->s_fs_info;
2995         iter.task = NULL;
2996         iter.tgid = filp->f_pos - TGID_OFFSET;
2997         for (iter = next_tgid(ns, iter);
2998              iter.task;
2999              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3000                 if (has_pid_permissions(ns, iter.task, 2))
3001                         __filldir = filldir;
3002                 else
3003                         __filldir = fake_filldir;
3004
3005                 filp->f_pos = iter.tgid + TGID_OFFSET;
3006                 if (proc_pid_fill_cache(filp, dirent, __filldir, iter) < 0) {
3007                         put_task_struct(iter.task);
3008                         goto out;
3009                 }
3010         }
3011         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
3012 out:
3013         put_task_struct(reaper);
3014 out_no_task:
3015         return 0;
3016 }
3017
3018 /*
3019  * Tasks
3020  */
3021 static const struct pid_entry tid_base_stuff[] = {
3022         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3023         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3024         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3025         REG("environ",   S_IRUSR, proc_environ_operations),
3026         INF("auxv",      S_IRUSR, proc_pid_auxv),
3027         ONE("status",    S_IRUGO, proc_pid_status),
3028         ONE("personality", S_IRUGO, proc_pid_personality),
3029         INF("limits",    S_IRUGO, proc_pid_limits),
3030 #ifdef CONFIG_SCHED_DEBUG
3031         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3032 #endif
3033         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3034 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3035         INF("syscall",   S_IRUGO, proc_pid_syscall),
3036 #endif
3037         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
3038         ONE("stat",      S_IRUGO, proc_tid_stat),
3039         ONE("statm",     S_IRUGO, proc_pid_statm),
3040         REG("maps",      S_IRUGO, proc_tid_maps_operations),
3041 #ifdef CONFIG_CHECKPOINT_RESTORE
3042         REG("children",  S_IRUGO, proc_tid_children_operations),
3043 #endif
3044 #ifdef CONFIG_NUMA
3045         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3046 #endif
3047         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3048         LNK("cwd",       proc_cwd_link),
3049         LNK("root",      proc_root_link),
3050         LNK("exe",       proc_exe_link),
3051         REG("mounts",    S_IRUGO, proc_mounts_operations),
3052         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3053 #ifdef CONFIG_PROC_PAGE_MONITOR
3054         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3055         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3056         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
3057 #endif
3058 #ifdef CONFIG_SECURITY
3059         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3060 #endif
3061 #ifdef CONFIG_KALLSYMS
3062         INF("wchan",     S_IRUGO, proc_pid_wchan),
3063 #endif
3064 #ifdef CONFIG_STACKTRACE
3065         ONE("stack",      S_IRUGO, proc_pid_stack),
3066 #endif
3067 #ifdef CONFIG_SCHEDSTATS
3068         INF("schedstat", S_IRUGO, proc_pid_schedstat),
3069 #endif
3070 #ifdef CONFIG_LATENCYTOP
3071         REG("latency",  S_IRUGO, proc_lstats_operations),
3072 #endif
3073 #ifdef CONFIG_PROC_PID_CPUSET
3074         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
3075 #endif
3076 #ifdef CONFIG_CGROUPS
3077         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3078 #endif
3079         INF("oom_score", S_IRUGO, proc_oom_score),
3080         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3081         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3082 #ifdef CONFIG_AUDITSYSCALL
3083         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3084         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3085 #endif
3086 #ifdef CONFIG_FAULT_INJECTION
3087         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3088 #endif
3089 #ifdef CONFIG_TASK_IO_ACCOUNTING
3090         INF("io",       S_IRUSR, proc_tid_io_accounting),
3091 #endif
3092 #ifdef CONFIG_HARDWALL
3093         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
3094 #endif
3095 #ifdef CONFIG_USER_NS
3096         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3097         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3098         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3099 #endif
3100 };
3101
3102 static int proc_tid_base_readdir(struct file * filp,
3103                              void * dirent, filldir_t filldir)
3104 {
3105         return proc_pident_readdir(filp,dirent,filldir,
3106                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
3107 }
3108
3109 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3110 {
3111         return proc_pident_lookup(dir, dentry,
3112                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3113 }
3114
3115 static const struct file_operations proc_tid_base_operations = {
3116         .read           = generic_read_dir,
3117         .readdir        = proc_tid_base_readdir,
3118         .llseek         = default_llseek,
3119 };
3120
3121 static const struct inode_operations proc_tid_base_inode_operations = {
3122         .lookup         = proc_tid_base_lookup,
3123         .getattr        = pid_getattr,
3124         .setattr        = proc_setattr,
3125 };
3126
3127 static struct dentry *proc_task_instantiate(struct inode *dir,
3128         struct dentry *dentry, struct task_struct *task, const void *ptr)
3129 {
3130         struct dentry *error = ERR_PTR(-ENOENT);
3131         struct inode *inode;
3132         inode = proc_pid_make_inode(dir->i_sb, task);
3133
3134         if (!inode)
3135                 goto out;
3136         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3137         inode->i_op = &proc_tid_base_inode_operations;
3138         inode->i_fop = &proc_tid_base_operations;
3139         inode->i_flags|=S_IMMUTABLE;
3140
3141         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3142                                                   ARRAY_SIZE(tid_base_stuff)));
3143
3144         d_set_d_op(dentry, &pid_dentry_operations);
3145
3146         d_add(dentry, inode);
3147         /* Close the race of the process dying before we return the dentry */
3148         if (pid_revalidate(dentry, 0))
3149                 error = NULL;
3150 out:
3151         return error;
3152 }
3153
3154 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3155 {
3156         struct dentry *result = ERR_PTR(-ENOENT);
3157         struct task_struct *task;
3158         struct task_struct *leader = get_proc_task(dir);
3159         unsigned tid;
3160         struct pid_namespace *ns;
3161
3162         if (!leader)
3163                 goto out_no_task;
3164
3165         tid = name_to_int(dentry);
3166         if (tid == ~0U)
3167                 goto out;
3168
3169         ns = dentry->d_sb->s_fs_info;
3170         rcu_read_lock();
3171         task = find_task_by_pid_ns(tid, ns);
3172         if (task)
3173                 get_task_struct(task);
3174         rcu_read_unlock();
3175         if (!task)
3176                 goto out;
3177         if (!same_thread_group(leader, task))
3178                 goto out_drop_task;
3179
3180         result = proc_task_instantiate(dir, dentry, task, NULL);
3181 out_drop_task:
3182         put_task_struct(task);
3183 out:
3184         put_task_struct(leader);
3185 out_no_task:
3186         return result;
3187 }
3188
3189 /*
3190  * Find the first tid of a thread group to return to user space.
3191  *
3192  * Usually this is just the thread group leader, but if the users
3193  * buffer was too small or there was a seek into the middle of the
3194  * directory we have more work todo.
3195  *
3196  * In the case of a short read we start with find_task_by_pid.
3197  *
3198  * In the case of a seek we start with the leader and walk nr
3199  * threads past it.
3200  */
3201 static struct task_struct *first_tid(struct task_struct *leader,
3202                 int tid, int nr, struct pid_namespace *ns)
3203 {
3204         struct task_struct *pos;
3205
3206         rcu_read_lock();
3207         /* Attempt to start with the pid of a thread */
3208         if (tid && (nr > 0)) {
3209                 pos = find_task_by_pid_ns(tid, ns);
3210                 if (pos && (pos->group_leader == leader))
3211                         goto found;
3212         }
3213
3214         /* If nr exceeds the number of threads there is nothing todo */
3215         pos = NULL;
3216         if (nr && nr >= get_nr_threads(leader))
3217                 goto out;
3218
3219         /* If we haven't found our starting place yet start
3220          * with the leader and walk nr threads forward.
3221          */
3222         for (pos = leader; nr > 0; --nr) {
3223                 pos = next_thread(pos);
3224                 if (pos == leader) {
3225                         pos = NULL;
3226                         goto out;
3227                 }
3228         }
3229 found:
3230         get_task_struct(pos);
3231 out:
3232         rcu_read_unlock();
3233         return pos;
3234 }
3235
3236 /*
3237  * Find the next thread in the thread list.
3238  * Return NULL if there is an error or no next thread.
3239  *
3240  * The reference to the input task_struct is released.
3241  */
3242 static struct task_struct *next_tid(struct task_struct *start)
3243 {
3244         struct task_struct *pos = NULL;
3245         rcu_read_lock();
3246         if (pid_alive(start)) {
3247                 pos = next_thread(start);
3248                 if (thread_group_leader(pos))
3249                         pos = NULL;
3250                 else
3251                         get_task_struct(pos);
3252         }
3253         rcu_read_unlock();
3254         put_task_struct(start);
3255         return pos;
3256 }
3257
3258 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3259         struct task_struct *task, int tid)
3260 {
3261         char name[PROC_NUMBUF];
3262         int len = snprintf(name, sizeof(name), "%d", tid);
3263         return proc_fill_cache(filp, dirent, filldir, name, len,
3264                                 proc_task_instantiate, task, NULL);
3265 }
3266
3267 /* for the /proc/TGID/task/ directories */
3268 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3269 {
3270         struct dentry *dentry = filp->f_path.dentry;
3271         struct inode *inode = dentry->d_inode;
3272         struct task_struct *leader = NULL;
3273         struct task_struct *task;
3274         int retval = -ENOENT;
3275         ino_t ino;
3276         int tid;
3277         struct pid_namespace *ns;
3278
3279         task = get_proc_task(inode);
3280         if (!task)
3281                 goto out_no_task;
3282         rcu_read_lock();
3283         if (pid_alive(task)) {
3284                 leader = task->group_leader;
3285                 get_task_struct(leader);
3286         }
3287         rcu_read_unlock();
3288         put_task_struct(task);
3289         if (!leader)
3290                 goto out_no_task;
3291         retval = 0;
3292
3293         switch ((unsigned long)filp->f_pos) {
3294         case 0:
3295                 ino = inode->i_ino;
3296                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3297                         goto out;
3298                 filp->f_pos++;
3299                 /* fall through */
3300         case 1:
3301                 ino = parent_ino(dentry);
3302                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3303                         goto out;
3304                 filp->f_pos++;
3305                 /* fall through */
3306         }
3307
3308         /* f_version caches the tgid value that the last readdir call couldn't
3309          * return. lseek aka telldir automagically resets f_version to 0.
3310          */
3311         ns = filp->f_dentry->d_sb->s_fs_info;
3312         tid = (int)filp->f_version;
3313         filp->f_version = 0;
3314         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3315              task;
3316              task = next_tid(task), filp->f_pos++) {
3317                 tid = task_pid_nr_ns(task, ns);
3318                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3319                         /* returning this tgid failed, save it as the first
3320                          * pid for the next readir call */
3321                         filp->f_version = (u64)tid;
3322                         put_task_struct(task);
3323                         break;
3324                 }
3325         }
3326 out:
3327         put_task_struct(leader);
3328 out_no_task:
3329         return retval;
3330 }
3331
3332 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3333 {
3334         struct inode *inode = dentry->d_inode;
3335         struct task_struct *p = get_proc_task(inode);
3336         generic_fillattr(inode, stat);
3337
3338         if (p) {
3339                 stat->nlink += get_nr_threads(p);
3340                 put_task_struct(p);
3341         }
3342
3343         return 0;
3344 }
3345
3346 static const struct inode_operations proc_task_inode_operations = {
3347         .lookup         = proc_task_lookup,
3348         .getattr        = proc_task_getattr,
3349         .setattr        = proc_setattr,
3350         .permission     = proc_pid_permission,
3351 };
3352
3353 static const struct file_operations proc_task_operations = {
3354         .read           = generic_read_dir,
3355         .readdir        = proc_task_readdir,
3356         .llseek         = default_llseek,
3357 };