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_adj_read(struct file *file, char __user *buf, size_t count,
 877                             loff_t *ppos)
 878 {
 879         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 880         char buffer[PROC_NUMBUF];
 881         int oom_adj = OOM_ADJUST_MIN;
 882         size_t len;
 883         unsigned long flags;
 884
 885         if (!task)
 886                 return -ESRCH;
 887         if (lock_task_sighand(task, &flags)) {
 888                 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
 889                         oom_adj = OOM_ADJUST_MAX;
 890                 else
 891                         oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
 892                                   OOM_SCORE_ADJ_MAX;
 893                 unlock_task_sighand(task, &flags);
 894         }
 895         put_task_struct(task);
 896         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
 897         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 898 }
 899
 900 static ssize_t oom_adj_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_adj;
 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_adj);
 918         if (err)
 919                 goto out;
 920         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
 921              oom_adj != 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         /*
 944          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
 945          * value is always attainable.
 946          */
 947         if (oom_adj == OOM_ADJUST_MAX)
 948                 oom_adj = OOM_SCORE_ADJ_MAX;
 949         else
 950                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
 951
 952         if (oom_adj < task->signal->oom_score_adj &&
 953             !capable(CAP_SYS_RESOURCE)) {
 954                 err = -EACCES;
 955                 goto err_sighand;
 956         }
 957
 958         /*
 959          * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
 960          * /proc/pid/oom_score_adj instead.
 961          */
 962         printk_once(KERN_WARNING "%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
 963                   current->comm, task_pid_nr(current), task_pid_nr(task),
 964                   task_pid_nr(task));
 965
 966         task->signal->oom_score_adj = oom_adj;
 967         trace_oom_score_adj_update(task);
 968 err_sighand:
 969         unlock_task_sighand(task, &flags);
 970 err_task_lock:
 971         task_unlock(task);
 972         put_task_struct(task);
 973 out:
 974         return err < 0 ? err : count;
 975 }
 976
 977 static const struct file_operations proc_oom_adj_operations = {
 978         .read           = oom_adj_read,
 979         .write          = oom_adj_write,
 980         .llseek         = generic_file_llseek,
 981 };
 982
 983 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
 984                                         size_t count, loff_t *ppos)
 985 {
 986         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 987         char buffer[PROC_NUMBUF];
 988         int oom_score_adj = OOM_SCORE_ADJ_MIN;
 989         unsigned long flags;
 990         size_t len;
 991
 992         if (!task)
 993                 return -ESRCH;
 994         if (lock_task_sighand(task, &flags)) {
 995                 oom_score_adj = task->signal->oom_score_adj;
 996                 unlock_task_sighand(task, &flags);
 997         }
 998         put_task_struct(task);
 999         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_score_adj);
1000         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1001 }
1002
1003 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1004                                         size_t count, loff_t *ppos)
1005 {
1006         struct task_struct *task;
1007         char buffer[PROC_NUMBUF];
1008         unsigned long flags;
1009         int oom_score_adj;
1010         int err;
1011
1012         memset(buffer, 0, sizeof(buffer));
1013         if (count > sizeof(buffer) - 1)
1014                 count = sizeof(buffer) - 1;
1015         if (copy_from_user(buffer, buf, count)) {
1016                 err = -EFAULT;
1017                 goto out;
1018         }
1019
1020         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1021         if (err)
1022                 goto out;
1023         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1024                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1025                 err = -EINVAL;
1026                 goto out;
1027         }
1028
1029         task = get_proc_task(file->f_path.dentry->d_inode);
1030         if (!task) {
1031                 err = -ESRCH;
1032                 goto out;
1033         }
1034
1035         task_lock(task);
1036         if (!task->mm) {
1037                 err = -EINVAL;
1038                 goto err_task_lock;
1039         }
1040
1041         if (!lock_task_sighand(task, &flags)) {
1042                 err = -ESRCH;
1043                 goto err_task_lock;
1044         }
1045
1046         if (oom_score_adj < task->signal->oom_score_adj_min &&
1047                         !capable(CAP_SYS_RESOURCE)) {
1048                 err = -EACCES;
1049                 goto err_sighand;
1050         }
1051
1052         task->signal->oom_score_adj = oom_score_adj;
1053         if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1054                 task->signal->oom_score_adj_min = oom_score_adj;
1055         trace_oom_score_adj_update(task);
1056
1057 err_sighand:
1058         unlock_task_sighand(task, &flags);
1059 err_task_lock:
1060         task_unlock(task);
1061         put_task_struct(task);
1062 out:
1063         return err < 0 ? err : count;
1064 }
1065
1066 static const struct file_operations proc_oom_score_adj_operations = {
1067         .read           = oom_score_adj_read,
1068         .write          = oom_score_adj_write,
1069         .llseek         = default_llseek,
1070 };
1071
1072 #ifdef CONFIG_AUDITSYSCALL
1073 #define TMPBUFLEN 21
1074 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1075                                   size_t count, loff_t *ppos)
1076 {
1077         struct inode * inode = file->f_path.dentry->d_inode;
1078         struct task_struct *task = get_proc_task(inode);
1079         ssize_t length;
1080         char tmpbuf[TMPBUFLEN];
1081
1082         if (!task)
1083                 return -ESRCH;
1084         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1085                            from_kuid(file->f_cred->user_ns,
1086                                      audit_get_loginuid(task)));
1087         put_task_struct(task);
1088         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1089 }
1090
1091 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1092                                    size_t count, loff_t *ppos)
1093 {
1094         struct inode * inode = file->f_path.dentry->d_inode;
1095         char *page, *tmp;
1096         ssize_t length;
1097         uid_t loginuid;
1098         kuid_t kloginuid;
1099
1100         rcu_read_lock();
1101         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1102                 rcu_read_unlock();
1103                 return -EPERM;
1104         }
1105         rcu_read_unlock();
1106
1107         if (count >= PAGE_SIZE)
1108                 count = PAGE_SIZE - 1;
1109
1110         if (*ppos != 0) {
1111                 /* No partial writes. */
1112                 return -EINVAL;
1113         }
1114         page = (char*)__get_free_page(GFP_TEMPORARY);
1115         if (!page)
1116                 return -ENOMEM;
1117         length = -EFAULT;
1118         if (copy_from_user(page, buf, count))
1119                 goto out_free_page;
1120
1121         page[count] = '\0';
1122         loginuid = simple_strtoul(page, &tmp, 10);
1123         if (tmp == page) {
1124                 length = -EINVAL;
1125                 goto out_free_page;
1126
1127         }
1128         kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1129         if (!uid_valid(kloginuid)) {
1130                 length = -EINVAL;
1131                 goto out_free_page;
1132         }
1133
1134         length = audit_set_loginuid(kloginuid);
1135         if (likely(length == 0))
1136                 length = count;
1137
1138 out_free_page:
1139         free_page((unsigned long) page);
1140         return length;
1141 }
1142
1143 static const struct file_operations proc_loginuid_operations = {
1144         .read           = proc_loginuid_read,
1145         .write          = proc_loginuid_write,
1146         .llseek         = generic_file_llseek,
1147 };
1148
1149 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1150                                   size_t count, loff_t *ppos)
1151 {
1152         struct inode * inode = file->f_path.dentry->d_inode;
1153         struct task_struct *task = get_proc_task(inode);
1154         ssize_t length;
1155         char tmpbuf[TMPBUFLEN];
1156
1157         if (!task)
1158                 return -ESRCH;
1159         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1160                                 audit_get_sessionid(task));
1161         put_task_struct(task);
1162         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1163 }
1164
1165 static const struct file_operations proc_sessionid_operations = {
1166         .read           = proc_sessionid_read,
1167         .llseek         = generic_file_llseek,
1168 };
1169 #endif
1170
1171 #ifdef CONFIG_FAULT_INJECTION
1172 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1173                                       size_t count, loff_t *ppos)
1174 {
1175         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1176         char buffer[PROC_NUMBUF];
1177         size_t len;
1178         int make_it_fail;
1179
1180         if (!task)
1181                 return -ESRCH;
1182         make_it_fail = task->make_it_fail;
1183         put_task_struct(task);
1184
1185         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1186
1187         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1188 }
1189
1190 static ssize_t proc_fault_inject_write(struct file * file,
1191                         const char __user * buf, size_t count, loff_t *ppos)
1192 {
1193         struct task_struct *task;
1194         char buffer[PROC_NUMBUF], *end;
1195         int make_it_fail;
1196
1197         if (!capable(CAP_SYS_RESOURCE))
1198                 return -EPERM;
1199         memset(buffer, 0, sizeof(buffer));
1200         if (count > sizeof(buffer) - 1)
1201                 count = sizeof(buffer) - 1;
1202         if (copy_from_user(buffer, buf, count))
1203                 return -EFAULT;
1204         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1205         if (*end)
1206                 return -EINVAL;
1207         task = get_proc_task(file->f_dentry->d_inode);
1208         if (!task)
1209                 return -ESRCH;
1210         task->make_it_fail = make_it_fail;
1211         put_task_struct(task);
1212
1213         return count;
1214 }
1215
1216 static const struct file_operations proc_fault_inject_operations = {
1217         .read           = proc_fault_inject_read,
1218         .write          = proc_fault_inject_write,
1219         .llseek         = generic_file_llseek,
1220 };
1221 #endif
1222
1223
1224 #ifdef CONFIG_SCHED_DEBUG
1225 /*
1226  * Print out various scheduling related per-task fields:
1227  */
1228 static int sched_show(struct seq_file *m, void *v)
1229 {
1230         struct inode *inode = m->private;
1231         struct task_struct *p;
1232
1233         p = get_proc_task(inode);
1234         if (!p)
1235                 return -ESRCH;
1236         proc_sched_show_task(p, m);
1237
1238         put_task_struct(p);
1239
1240         return 0;
1241 }
1242
1243 static ssize_t
1244 sched_write(struct file *file, const char __user *buf,
1245             size_t count, loff_t *offset)
1246 {
1247         struct inode *inode = file->f_path.dentry->d_inode;
1248         struct task_struct *p;
1249
1250         p = get_proc_task(inode);
1251         if (!p)
1252                 return -ESRCH;
1253         proc_sched_set_task(p);
1254
1255         put_task_struct(p);
1256
1257         return count;
1258 }
1259
1260 static int sched_open(struct inode *inode, struct file *filp)
1261 {
1262         return single_open(filp, sched_show, inode);
1263 }
1264
1265 static const struct file_operations proc_pid_sched_operations = {
1266         .open           = sched_open,
1267         .read           = seq_read,
1268         .write          = sched_write,
1269         .llseek         = seq_lseek,
1270         .release        = single_release,
1271 };
1272
1273 #endif
1274
1275 #ifdef CONFIG_SCHED_AUTOGROUP
1276 /*
1277  * Print out autogroup related information:
1278  */
1279 static int sched_autogroup_show(struct seq_file *m, void *v)
1280 {
1281         struct inode *inode = m->private;
1282         struct task_struct *p;
1283
1284         p = get_proc_task(inode);
1285         if (!p)
1286                 return -ESRCH;
1287         proc_sched_autogroup_show_task(p, m);
1288
1289         put_task_struct(p);
1290
1291         return 0;
1292 }
1293
1294 static ssize_t
1295 sched_autogroup_write(struct file *file, const char __user *buf,
1296             size_t count, loff_t *offset)
1297 {
1298         struct inode *inode = file->f_path.dentry->d_inode;
1299         struct task_struct *p;
1300         char buffer[PROC_NUMBUF];
1301         int nice;
1302         int err;
1303
1304         memset(buffer, 0, sizeof(buffer));
1305         if (count > sizeof(buffer) - 1)
1306                 count = sizeof(buffer) - 1;
1307         if (copy_from_user(buffer, buf, count))
1308                 return -EFAULT;
1309
1310         err = kstrtoint(strstrip(buffer), 0, &nice);
1311         if (err < 0)
1312                 return err;
1313
1314         p = get_proc_task(inode);
1315         if (!p)
1316                 return -ESRCH;
1317
1318         err = proc_sched_autogroup_set_nice(p, nice);
1319         if (err)
1320                 count = err;
1321
1322         put_task_struct(p);
1323
1324         return count;
1325 }
1326
1327 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1328 {
1329         int ret;
1330
1331         ret = single_open(filp, sched_autogroup_show, NULL);
1332         if (!ret) {
1333                 struct seq_file *m = filp->private_data;
1334
1335                 m->private = inode;
1336         }
1337         return ret;
1338 }
1339
1340 static const struct file_operations proc_pid_sched_autogroup_operations = {
1341         .open           = sched_autogroup_open,
1342         .read           = seq_read,
1343         .write          = sched_autogroup_write,
1344         .llseek         = seq_lseek,
1345         .release        = single_release,
1346 };
1347
1348 #endif /* CONFIG_SCHED_AUTOGROUP */
1349
1350 static ssize_t comm_write(struct file *file, const char __user *buf,
1351                                 size_t count, loff_t *offset)
1352 {
1353         struct inode *inode = file->f_path.dentry->d_inode;
1354         struct task_struct *p;
1355         char buffer[TASK_COMM_LEN];
1356
1357         memset(buffer, 0, sizeof(buffer));
1358         if (count > sizeof(buffer) - 1)
1359                 count = sizeof(buffer) - 1;
1360         if (copy_from_user(buffer, buf, count))
1361                 return -EFAULT;
1362
1363         p = get_proc_task(inode);
1364         if (!p)
1365                 return -ESRCH;
1366
1367         if (same_thread_group(current, p))
1368                 set_task_comm(p, buffer);
1369         else
1370                 count = -EINVAL;
1371
1372         put_task_struct(p);
1373
1374         return count;
1375 }
1376
1377 static int comm_show(struct seq_file *m, void *v)
1378 {
1379         struct inode *inode = m->private;
1380         struct task_struct *p;
1381
1382         p = get_proc_task(inode);
1383         if (!p)
1384                 return -ESRCH;
1385
1386         task_lock(p);
1387         seq_printf(m, "%s\n", p->comm);
1388         task_unlock(p);
1389
1390         put_task_struct(p);
1391
1392         return 0;
1393 }
1394
1395 static int comm_open(struct inode *inode, struct file *filp)
1396 {
1397         return single_open(filp, comm_show, inode);
1398 }
1399
1400 static const struct file_operations proc_pid_set_comm_operations = {
1401         .open           = comm_open,
1402         .read           = seq_read,
1403         .write          = comm_write,
1404         .llseek         = seq_lseek,
1405         .release        = single_release,
1406 };
1407
1408 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1409 {
1410         struct task_struct *task;
1411         struct mm_struct *mm;
1412         struct file *exe_file;
1413
1414         task = get_proc_task(dentry->d_inode);
1415         if (!task)
1416                 return -ENOENT;
1417         mm = get_task_mm(task);
1418         put_task_struct(task);
1419         if (!mm)
1420                 return -ENOENT;
1421         exe_file = get_mm_exe_file(mm);
1422         mmput(mm);
1423         if (exe_file) {
1424                 *exe_path = exe_file->f_path;
1425                 path_get(&exe_file->f_path);
1426                 fput(exe_file);
1427                 return 0;
1428         } else
1429                 return -ENOENT;
1430 }
1431
1432 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1433 {
1434         struct inode *inode = dentry->d_inode;
1435         struct path path;
1436         int error = -EACCES;
1437
1438         /* Are we allowed to snoop on the tasks file descriptors? */
1439         if (!proc_fd_access_allowed(inode))
1440                 goto out;
1441
1442         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1443         if (error)
1444                 goto out;
1445
1446         nd_jump_link(nd, &path);
1447         return NULL;
1448 out:
1449         return ERR_PTR(error);
1450 }
1451
1452 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1453 {
1454         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1455         char *pathname;
1456         int len;
1457
1458         if (!tmp)
1459                 return -ENOMEM;
1460
1461         pathname = d_path(path, tmp, PAGE_SIZE);
1462         len = PTR_ERR(pathname);
1463         if (IS_ERR(pathname))
1464                 goto out;
1465         len = tmp + PAGE_SIZE - 1 - pathname;
1466
1467         if (len > buflen)
1468                 len = buflen;
1469         if (copy_to_user(buffer, pathname, len))
1470                 len = -EFAULT;
1471  out:
1472         free_page((unsigned long)tmp);
1473         return len;
1474 }
1475
1476 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1477 {
1478         int error = -EACCES;
1479         struct inode *inode = dentry->d_inode;
1480         struct path path;
1481
1482         /* Are we allowed to snoop on the tasks file descriptors? */
1483         if (!proc_fd_access_allowed(inode))
1484                 goto out;
1485
1486         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1487         if (error)
1488                 goto out;
1489
1490         error = do_proc_readlink(&path, buffer, buflen);
1491         path_put(&path);
1492 out:
1493         return error;
1494 }
1495
1496 const struct inode_operations proc_pid_link_inode_operations = {
1497         .readlink       = proc_pid_readlink,
1498         .follow_link    = proc_pid_follow_link,
1499         .setattr        = proc_setattr,
1500 };
1501
1502
1503 /* building an inode */
1504
1505 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1506 {
1507         struct inode * inode;
1508         struct proc_inode *ei;
1509         const struct cred *cred;
1510
1511         /* We need a new inode */
1512
1513         inode = new_inode(sb);
1514         if (!inode)
1515                 goto out;
1516
1517         /* Common stuff */
1518         ei = PROC_I(inode);
1519         inode->i_ino = get_next_ino();
1520         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1521         inode->i_op = &proc_def_inode_operations;
1522
1523         /*
1524          * grab the reference to task.
1525          */
1526         ei->pid = get_task_pid(task, PIDTYPE_PID);
1527         if (!ei->pid)
1528                 goto out_unlock;
1529
1530         if (task_dumpable(task)) {
1531                 rcu_read_lock();
1532                 cred = __task_cred(task);
1533                 inode->i_uid = cred->euid;
1534                 inode->i_gid = cred->egid;
1535                 rcu_read_unlock();
1536         }
1537         security_task_to_inode(task, inode);
1538
1539 out:
1540         return inode;
1541
1542 out_unlock:
1543         iput(inode);
1544         return NULL;
1545 }
1546
1547 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1548 {
1549         struct inode *inode = dentry->d_inode;
1550         struct task_struct *task;
1551         const struct cred *cred;
1552         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1553
1554         generic_fillattr(inode, stat);
1555
1556         rcu_read_lock();
1557         stat->uid = GLOBAL_ROOT_UID;
1558         stat->gid = GLOBAL_ROOT_GID;
1559         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1560         if (task) {
1561                 if (!has_pid_permissions(pid, task, 2)) {
1562                         rcu_read_unlock();
1563                         /*
1564                          * This doesn't prevent learning whether PID exists,
1565                          * it only makes getattr() consistent with readdir().
1566                          */
1567                         return -ENOENT;
1568                 }
1569                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1570                     task_dumpable(task)) {
1571                         cred = __task_cred(task);
1572                         stat->uid = cred->euid;
1573                         stat->gid = cred->egid;
1574                 }
1575         }
1576         rcu_read_unlock();
1577         return 0;
1578 }
1579
1580 /* dentry stuff */
1581
1582 /*
1583  *      Exceptional case: normally we are not allowed to unhash a busy
1584  * directory. In this case, however, we can do it - no aliasing problems
1585  * due to the way we treat inodes.
1586  *
1587  * Rewrite the inode's ownerships here because the owning task may have
1588  * performed a setuid(), etc.
1589  *
1590  * Before the /proc/pid/status file was created the only way to read
1591  * the effective uid of a /process was to stat /proc/pid.  Reading
1592  * /proc/pid/status is slow enough that procps and other packages
1593  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1594  * made this apply to all per process world readable and executable
1595  * directories.
1596  */
1597 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1598 {
1599         struct inode *inode;
1600         struct task_struct *task;
1601         const struct cred *cred;
1602
1603         if (flags & LOOKUP_RCU)
1604                 return -ECHILD;
1605
1606         inode = dentry->d_inode;
1607         task = get_proc_task(inode);
1608
1609         if (task) {
1610                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1611                     task_dumpable(task)) {
1612                         rcu_read_lock();
1613                         cred = __task_cred(task);
1614                         inode->i_uid = cred->euid;
1615                         inode->i_gid = cred->egid;
1616                         rcu_read_unlock();
1617                 } else {
1618                         inode->i_uid = GLOBAL_ROOT_UID;
1619                         inode->i_gid = GLOBAL_ROOT_GID;
1620                 }
1621                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1622                 security_task_to_inode(task, inode);
1623                 put_task_struct(task);
1624                 return 1;
1625         }
1626         d_drop(dentry);
1627         return 0;
1628 }
1629
1630 const struct dentry_operations pid_dentry_operations =
1631 {
1632         .d_revalidate   = pid_revalidate,
1633         .d_delete       = pid_delete_dentry,
1634 };
1635
1636 /* Lookups */
1637
1638 /*
1639  * Fill a directory entry.
1640  *
1641  * If possible create the dcache entry and derive our inode number and
1642  * file type from dcache entry.
1643  *
1644  * Since all of the proc inode numbers are dynamically generated, the inode
1645  * numbers do not exist until the inode is cache.  This means creating the
1646  * the dcache entry in readdir is necessary to keep the inode numbers
1647  * reported by readdir in sync with the inode numbers reported
1648  * by stat.
1649  */
1650 int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1651         const char *name, int len,
1652         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1653 {
1654         struct dentry *child, *dir = filp->f_path.dentry;
1655         struct inode *inode;
1656         struct qstr qname;
1657         ino_t ino = 0;
1658         unsigned type = DT_UNKNOWN;
1659
1660         qname.name = name;
1661         qname.len  = len;
1662         qname.hash = full_name_hash(name, len);
1663
1664         child = d_lookup(dir, &qname);
1665         if (!child) {
1666                 struct dentry *new;
1667                 new = d_alloc(dir, &qname);
1668                 if (new) {
1669                         child = instantiate(dir->d_inode, new, task, ptr);
1670                         if (child)
1671                                 dput(new);
1672                         else
1673                                 child = new;
1674                 }
1675         }
1676         if (!child || IS_ERR(child) || !child->d_inode)
1677                 goto end_instantiate;
1678         inode = child->d_inode;
1679         if (inode) {
1680                 ino = inode->i_ino;
1681                 type = inode->i_mode >> 12;
1682         }
1683         dput(child);
1684 end_instantiate:
1685         if (!ino)
1686                 ino = find_inode_number(dir, &qname);
1687         if (!ino)
1688                 ino = 1;
1689         return filldir(dirent, name, len, filp->f_pos, ino, type);
1690 }
1691
1692 #ifdef CONFIG_CHECKPOINT_RESTORE
1693
1694 /*
1695  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1696  * which represent vma start and end addresses.
1697  */
1698 static int dname_to_vma_addr(struct dentry *dentry,
1699                              unsigned long *start, unsigned long *end)
1700 {
1701         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1702                 return -EINVAL;
1703
1704         return 0;
1705 }
1706
1707 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1708 {
1709         unsigned long vm_start, vm_end;
1710         bool exact_vma_exists = false;
1711         struct mm_struct *mm = NULL;
1712         struct task_struct *task;
1713         const struct cred *cred;
1714         struct inode *inode;
1715         int status = 0;
1716
1717         if (flags & LOOKUP_RCU)
1718                 return -ECHILD;
1719
1720         if (!capable(CAP_SYS_ADMIN)) {
1721                 status = -EACCES;
1722                 goto out_notask;
1723         }
1724
1725         inode = dentry->d_inode;
1726         task = get_proc_task(inode);
1727         if (!task)
1728                 goto out_notask;
1729
1730         mm = mm_access(task, PTRACE_MODE_READ);
1731         if (IS_ERR_OR_NULL(mm))
1732                 goto out;
1733
1734         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1735                 down_read(&mm->mmap_sem);
1736                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1737                 up_read(&mm->mmap_sem);
1738         }
1739
1740         mmput(mm);
1741
1742         if (exact_vma_exists) {
1743                 if (task_dumpable(task)) {
1744                         rcu_read_lock();
1745                         cred = __task_cred(task);
1746                         inode->i_uid = cred->euid;
1747                         inode->i_gid = cred->egid;
1748                         rcu_read_unlock();
1749                 } else {
1750                         inode->i_uid = GLOBAL_ROOT_UID;
1751                         inode->i_gid = GLOBAL_ROOT_GID;
1752                 }
1753                 security_task_to_inode(task, inode);
1754                 status = 1;
1755         }
1756
1757 out:
1758         put_task_struct(task);
1759
1760 out_notask:
1761         if (status <= 0)
1762                 d_drop(dentry);
1763
1764         return status;
1765 }
1766
1767 static const struct dentry_operations tid_map_files_dentry_operations = {
1768         .d_revalidate   = map_files_d_revalidate,
1769         .d_delete       = pid_delete_dentry,
1770 };
1771
1772 static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
1773 {
1774         unsigned long vm_start, vm_end;
1775         struct vm_area_struct *vma;
1776         struct task_struct *task;
1777         struct mm_struct *mm;
1778         int rc;
1779
1780         rc = -ENOENT;
1781         task = get_proc_task(dentry->d_inode);
1782         if (!task)
1783                 goto out;
1784
1785         mm = get_task_mm(task);
1786         put_task_struct(task);
1787         if (!mm)
1788                 goto out;
1789
1790         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1791         if (rc)
1792                 goto out_mmput;
1793
1794         down_read(&mm->mmap_sem);
1795         vma = find_exact_vma(mm, vm_start, vm_end);
1796         if (vma && vma->vm_file) {
1797                 *path = vma->vm_file->f_path;
1798                 path_get(path);
1799                 rc = 0;
1800         }
1801         up_read(&mm->mmap_sem);
1802
1803 out_mmput:
1804         mmput(mm);
1805 out:
1806         return rc;
1807 }
1808
1809 struct map_files_info {
1810         fmode_t         mode;
1811         unsigned long   len;
1812         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1813 };
1814
1815 static struct dentry *
1816 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
1817                            struct task_struct *task, const void *ptr)
1818 {
1819         fmode_t mode = (fmode_t)(unsigned long)ptr;
1820         struct proc_inode *ei;
1821         struct inode *inode;
1822
1823         inode = proc_pid_make_inode(dir->i_sb, task);
1824         if (!inode)
1825                 return ERR_PTR(-ENOENT);
1826
1827         ei = PROC_I(inode);
1828         ei->op.proc_get_link = proc_map_files_get_link;
1829
1830         inode->i_op = &proc_pid_link_inode_operations;
1831         inode->i_size = 64;
1832         inode->i_mode = S_IFLNK;
1833
1834         if (mode & FMODE_READ)
1835                 inode->i_mode |= S_IRUSR;
1836         if (mode & FMODE_WRITE)
1837                 inode->i_mode |= S_IWUSR;
1838
1839         d_set_d_op(dentry, &tid_map_files_dentry_operations);
1840         d_add(dentry, inode);
1841
1842         return NULL;
1843 }
1844
1845 static struct dentry *proc_map_files_lookup(struct inode *dir,
1846                 struct dentry *dentry, unsigned int flags)
1847 {
1848         unsigned long vm_start, vm_end;
1849         struct vm_area_struct *vma;
1850         struct task_struct *task;
1851         struct dentry *result;
1852         struct mm_struct *mm;
1853
1854         result = ERR_PTR(-EACCES);
1855         if (!capable(CAP_SYS_ADMIN))
1856                 goto out;
1857
1858         result = ERR_PTR(-ENOENT);
1859         task = get_proc_task(dir);
1860         if (!task)
1861                 goto out;
1862
1863         result = ERR_PTR(-EACCES);
1864         if (!ptrace_may_access(task, PTRACE_MODE_READ))
1865                 goto out_put_task;
1866
1867         result = ERR_PTR(-ENOENT);
1868         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
1869                 goto out_put_task;
1870
1871         mm = get_task_mm(task);
1872         if (!mm)
1873                 goto out_put_task;
1874
1875         down_read(&mm->mmap_sem);
1876         vma = find_exact_vma(mm, vm_start, vm_end);
1877         if (!vma)
1878                 goto out_no_vma;
1879
1880         if (vma->vm_file)
1881                 result = proc_map_files_instantiate(dir, dentry, task,
1882                                 (void *)(unsigned long)vma->vm_file->f_mode);
1883
1884 out_no_vma:
1885         up_read(&mm->mmap_sem);
1886         mmput(mm);
1887 out_put_task:
1888         put_task_struct(task);
1889 out:
1890         return result;
1891 }
1892
1893 static const struct inode_operations proc_map_files_inode_operations = {
1894         .lookup         = proc_map_files_lookup,
1895         .permission     = proc_fd_permission,
1896         .setattr        = proc_setattr,
1897 };
1898
1899 static int
1900 proc_map_files_readdir(struct file *filp, void *dirent, filldir_t filldir)
1901 {
1902         struct dentry *dentry = filp->f_path.dentry;
1903         struct inode *inode = dentry->d_inode;
1904         struct vm_area_struct *vma;
1905         struct task_struct *task;
1906         struct mm_struct *mm;
1907         ino_t ino;
1908         int ret;
1909
1910         ret = -EACCES;
1911         if (!capable(CAP_SYS_ADMIN))
1912                 goto out;
1913
1914         ret = -ENOENT;
1915         task = get_proc_task(inode);
1916         if (!task)
1917                 goto out;
1918
1919         ret = -EACCES;
1920         if (!ptrace_may_access(task, PTRACE_MODE_READ))
1921                 goto out_put_task;
1922
1923         ret = 0;
1924         switch (filp->f_pos) {
1925         case 0:
1926                 ino = inode->i_ino;
1927                 if (filldir(dirent, ".", 1, 0, ino, DT_DIR) < 0)
1928                         goto out_put_task;
1929                 filp->f_pos++;
1930         case 1:
1931                 ino = parent_ino(dentry);
1932                 if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1933                         goto out_put_task;
1934                 filp->f_pos++;
1935         default:
1936         {
1937                 unsigned long nr_files, pos, i;
1938                 struct flex_array *fa = NULL;
1939                 struct map_files_info info;
1940                 struct map_files_info *p;
1941
1942                 mm = get_task_mm(task);
1943                 if (!mm)
1944                         goto out_put_task;
1945                 down_read(&mm->mmap_sem);
1946
1947                 nr_files = 0;
1948
1949                 /*
1950                  * We need two passes here:
1951                  *
1952                  *  1) Collect vmas of mapped files with mmap_sem taken
1953                  *  2) Release mmap_sem and instantiate entries
1954                  *
1955                  * otherwise we get lockdep complained, since filldir()
1956                  * routine might require mmap_sem taken in might_fault().
1957                  */
1958
1959                 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
1960                         if (vma->vm_file && ++pos > filp->f_pos)
1961                                 nr_files++;
1962                 }
1963
1964                 if (nr_files) {
1965                         fa = flex_array_alloc(sizeof(info), nr_files,
1966                                                 GFP_KERNEL);
1967                         if (!fa || flex_array_prealloc(fa, 0, nr_files,
1968                                                         GFP_KERNEL)) {
1969                                 ret = -ENOMEM;
1970                                 if (fa)
1971                                         flex_array_free(fa);
1972                                 up_read(&mm->mmap_sem);
1973                                 mmput(mm);
1974                                 goto out_put_task;
1975                         }
1976                         for (i = 0, vma = mm->mmap, pos = 2; vma;
1977                                         vma = vma->vm_next) {
1978                                 if (!vma->vm_file)
1979                                         continue;
1980                                 if (++pos <= filp->f_pos)
1981                                         continue;
1982
1983                                 info.mode = vma->vm_file->f_mode;
1984                                 info.len = snprintf(info.name,
1985                                                 sizeof(info.name), "%lx-%lx",
1986                                                 vma->vm_start, vma->vm_end);
1987                                 if (flex_array_put(fa, i++, &info, GFP_KERNEL))
1988                                         BUG();
1989                         }
1990                 }
1991                 up_read(&mm->mmap_sem);
1992
1993                 for (i = 0; i < nr_files; i++) {
1994                         p = flex_array_get(fa, i);
1995                         ret = proc_fill_cache(filp, dirent, filldir,
1996                                               p->name, p->len,
1997                                               proc_map_files_instantiate,
1998                                               task,
1999                                               (void *)(unsigned long)p->mode);
2000                         if (ret)
2001                                 break;
2002                         filp->f_pos++;
2003                 }
2004                 if (fa)
2005                         flex_array_free(fa);
2006                 mmput(mm);
2007         }
2008         }
2009
2010 out_put_task:
2011         put_task_struct(task);
2012 out:
2013         return ret;
2014 }
2015
2016 static const struct file_operations proc_map_files_operations = {
2017         .read           = generic_read_dir,
2018         .readdir        = proc_map_files_readdir,
2019         .llseek         = default_llseek,
2020 };
2021
2022 #endif /* CONFIG_CHECKPOINT_RESTORE */
2023
2024 static struct dentry *proc_pident_instantiate(struct inode *dir,
2025         struct dentry *dentry, struct task_struct *task, const void *ptr)
2026 {
2027         const struct pid_entry *p = ptr;
2028         struct inode *inode;
2029         struct proc_inode *ei;
2030         struct dentry *error = ERR_PTR(-ENOENT);
2031
2032         inode = proc_pid_make_inode(dir->i_sb, task);
2033         if (!inode)
2034                 goto out;
2035
2036         ei = PROC_I(inode);
2037         inode->i_mode = p->mode;
2038         if (S_ISDIR(inode->i_mode))
2039                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2040         if (p->iop)
2041                 inode->i_op = p->iop;
2042         if (p->fop)
2043                 inode->i_fop = p->fop;
2044         ei->op = p->op;
2045         d_set_d_op(dentry, &pid_dentry_operations);
2046         d_add(dentry, inode);
2047         /* Close the race of the process dying before we return the dentry */
2048         if (pid_revalidate(dentry, 0))
2049                 error = NULL;
2050 out:
2051         return error;
2052 }
2053
2054 static struct dentry *proc_pident_lookup(struct inode *dir,
2055                                          struct dentry *dentry,
2056                                          const struct pid_entry *ents,
2057                                          unsigned int nents)
2058 {
2059         struct dentry *error;
2060         struct task_struct *task = get_proc_task(dir);
2061         const struct pid_entry *p, *last;
2062
2063         error = ERR_PTR(-ENOENT);
2064
2065         if (!task)
2066                 goto out_no_task;
2067
2068         /*
2069          * Yes, it does not scale. And it should not. Don't add
2070          * new entries into /proc/<tgid>/ without very good reasons.
2071          */
2072         last = &ents[nents - 1];
2073         for (p = ents; p <= last; p++) {
2074                 if (p->len != dentry->d_name.len)
2075                         continue;
2076                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2077                         break;
2078         }
2079         if (p > last)
2080                 goto out;
2081
2082         error = proc_pident_instantiate(dir, dentry, task, p);
2083 out:
2084         put_task_struct(task);
2085 out_no_task:
2086         return error;
2087 }
2088
2089 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2090         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2091 {
2092         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2093                                 proc_pident_instantiate, task, p);
2094 }
2095
2096 static int proc_pident_readdir(struct file *filp,
2097                 void *dirent, filldir_t filldir,
2098                 const struct pid_entry *ents, unsigned int nents)
2099 {
2100         int i;
2101         struct dentry *dentry = filp->f_path.dentry;
2102         struct inode *inode = dentry->d_inode;
2103         struct task_struct *task = get_proc_task(inode);
2104         const struct pid_entry *p, *last;
2105         ino_t ino;
2106         int ret;
2107
2108         ret = -ENOENT;
2109         if (!task)
2110                 goto out_no_task;
2111
2112         ret = 0;
2113         i = filp->f_pos;
2114         switch (i) {
2115         case 0:
2116                 ino = inode->i_ino;
2117                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2118                         goto out;
2119                 i++;
2120                 filp->f_pos++;
2121                 /* fall through */
2122         case 1:
2123                 ino = parent_ino(dentry);
2124                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2125                         goto out;
2126                 i++;
2127                 filp->f_pos++;
2128                 /* fall through */
2129         default:
2130                 i -= 2;
2131                 if (i >= nents) {
2132                         ret = 1;
2133                         goto out;
2134                 }
2135                 p = ents + i;
2136                 last = &ents[nents - 1];
2137                 while (p <= last) {
2138                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2139                                 goto out;
2140                         filp->f_pos++;
2141                         p++;
2142                 }
2143         }
2144
2145         ret = 1;
2146 out:
2147         put_task_struct(task);
2148 out_no_task:
2149         return ret;
2150 }
2151
2152 #ifdef CONFIG_SECURITY
2153 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2154                                   size_t count, loff_t *ppos)
2155 {
2156         struct inode * inode = file->f_path.dentry->d_inode;
2157         char *p = NULL;
2158         ssize_t length;
2159         struct task_struct *task = get_proc_task(inode);
2160
2161         if (!task)
2162                 return -ESRCH;
2163
2164         length = security_getprocattr(task,
2165                                       (char*)file->f_path.dentry->d_name.name,
2166                                       &p);
2167         put_task_struct(task);
2168         if (length > 0)
2169                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2170         kfree(p);
2171         return length;
2172 }
2173
2174 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2175                                    size_t count, loff_t *ppos)
2176 {
2177         struct inode * inode = file->f_path.dentry->d_inode;
2178         char *page;
2179         ssize_t length;
2180         struct task_struct *task = get_proc_task(inode);
2181
2182         length = -ESRCH;
2183         if (!task)
2184                 goto out_no_task;
2185         if (count > PAGE_SIZE)
2186                 count = PAGE_SIZE;
2187
2188         /* No partial writes. */
2189         length = -EINVAL;
2190         if (*ppos != 0)
2191                 goto out;
2192
2193         length = -ENOMEM;
2194         page = (char*)__get_free_page(GFP_TEMPORARY);
2195         if (!page)
2196                 goto out;
2197
2198         length = -EFAULT;
2199         if (copy_from_user(page, buf, count))
2200                 goto out_free;
2201
2202         /* Guard against adverse ptrace interaction */
2203         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2204         if (length < 0)
2205                 goto out_free;
2206
2207         length = security_setprocattr(task,
2208                                       (char*)file->f_path.dentry->d_name.name,
2209                                       (void*)page, count);
2210         mutex_unlock(&task->signal->cred_guard_mutex);
2211 out_free:
2212         free_page((unsigned long) page);
2213 out:
2214         put_task_struct(task);
2215 out_no_task:
2216         return length;
2217 }
2218
2219 static const struct file_operations proc_pid_attr_operations = {
2220         .read           = proc_pid_attr_read,
2221         .write          = proc_pid_attr_write,
2222         .llseek         = generic_file_llseek,
2223 };
2224
2225 static const struct pid_entry attr_dir_stuff[] = {
2226         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2227         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2228         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2229         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2230         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2231         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2232 };
2233
2234 static int proc_attr_dir_readdir(struct file * filp,
2235                              void * dirent, filldir_t filldir)
2236 {
2237         return proc_pident_readdir(filp,dirent,filldir,
2238                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2239 }
2240
2241 static const struct file_operations proc_attr_dir_operations = {
2242         .read           = generic_read_dir,
2243         .readdir        = proc_attr_dir_readdir,
2244         .llseek         = default_llseek,
2245 };
2246
2247 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2248                                 struct dentry *dentry, unsigned int flags)
2249 {
2250         return proc_pident_lookup(dir, dentry,
2251                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2252 }
2253
2254 static const struct inode_operations proc_attr_dir_inode_operations = {
2255         .lookup         = proc_attr_dir_lookup,
2256         .getattr        = pid_getattr,
2257         .setattr        = proc_setattr,
2258 };
2259
2260 #endif
2261
2262 #ifdef CONFIG_ELF_CORE
2263 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2264                                          size_t count, loff_t *ppos)
2265 {
2266         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2267         struct mm_struct *mm;
2268         char buffer[PROC_NUMBUF];
2269         size_t len;
2270         int ret;
2271
2272         if (!task)
2273                 return -ESRCH;
2274
2275         ret = 0;
2276         mm = get_task_mm(task);
2277         if (mm) {
2278                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2279                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2280                                 MMF_DUMP_FILTER_SHIFT));
2281                 mmput(mm);
2282                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2283         }
2284
2285         put_task_struct(task);
2286
2287         return ret;
2288 }
2289
2290 static ssize_t proc_coredump_filter_write(struct file *file,
2291                                           const char __user *buf,
2292                                           size_t count,
2293                                           loff_t *ppos)
2294 {
2295         struct task_struct *task;
2296         struct mm_struct *mm;
2297         char buffer[PROC_NUMBUF], *end;
2298         unsigned int val;
2299         int ret;
2300         int i;
2301         unsigned long mask;
2302
2303         ret = -EFAULT;
2304         memset(buffer, 0, sizeof(buffer));
2305         if (count > sizeof(buffer) - 1)
2306                 count = sizeof(buffer) - 1;
2307         if (copy_from_user(buffer, buf, count))
2308                 goto out_no_task;
2309
2310         ret = -EINVAL;
2311         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2312         if (*end == '\n')
2313                 end++;
2314         if (end - buffer == 0)
2315                 goto out_no_task;
2316
2317         ret = -ESRCH;
2318         task = get_proc_task(file->f_dentry->d_inode);
2319         if (!task)
2320                 goto out_no_task;
2321
2322         ret = end - buffer;
2323         mm = get_task_mm(task);
2324         if (!mm)
2325                 goto out_no_mm;
2326
2327         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2328                 if (val & mask)
2329                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2330                 else
2331                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2332         }
2333
2334         mmput(mm);
2335  out_no_mm:
2336         put_task_struct(task);
2337  out_no_task:
2338         return ret;
2339 }
2340
2341 static const struct file_operations proc_coredump_filter_operations = {
2342         .read           = proc_coredump_filter_read,
2343         .write          = proc_coredump_filter_write,
2344         .llseek         = generic_file_llseek,
2345 };
2346 #endif
2347
2348 /*
2349  * /proc/self:
2350  */
2351 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2352                               int buflen)
2353 {
2354         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2355         pid_t tgid = task_tgid_nr_ns(current, ns);
2356         char tmp[PROC_NUMBUF];
2357         if (!tgid)
2358                 return -ENOENT;
2359         sprintf(tmp, "%d", tgid);
2360         return vfs_readlink(dentry,buffer,buflen,tmp);
2361 }
2362
2363 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2364 {
2365         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2366         pid_t tgid = task_tgid_nr_ns(current, ns);
2367         char *name = ERR_PTR(-ENOENT);
2368         if (tgid) {
2369                 /* 11 for max length of signed int in decimal + NULL term */
2370                 name = kmalloc(12, GFP_KERNEL);
2371                 if (!name)
2372                         name = ERR_PTR(-ENOMEM);
2373                 else
2374                         sprintf(name, "%d", tgid);
2375         }
2376         nd_set_link(nd, name);
2377         return NULL;
2378 }
2379
2380 static void proc_self_put_link(struct dentry *dentry, struct nameidata *nd,
2381                                 void *cookie)
2382 {
2383         char *s = nd_get_link(nd);
2384         if (!IS_ERR(s))
2385                 kfree(s);
2386 }
2387
2388 static const struct inode_operations proc_self_inode_operations = {
2389         .readlink       = proc_self_readlink,
2390         .follow_link    = proc_self_follow_link,
2391         .put_link       = proc_self_put_link,
2392 };
2393
2394 /*
2395  * proc base
2396  *
2397  * These are the directory entries in the root directory of /proc
2398  * that properly belong to the /proc filesystem, as they describe
2399  * describe something that is process related.
2400  */
2401 static const struct pid_entry proc_base_stuff[] = {
2402         NOD("self", S_IFLNK|S_IRWXUGO,
2403                 &proc_self_inode_operations, NULL, {}),
2404 };
2405
2406 static struct dentry *proc_base_instantiate(struct inode *dir,
2407         struct dentry *dentry, struct task_struct *task, const void *ptr)
2408 {
2409         const struct pid_entry *p = ptr;
2410         struct inode *inode;
2411         struct proc_inode *ei;
2412         struct dentry *error;
2413
2414         /* Allocate the inode */
2415         error = ERR_PTR(-ENOMEM);
2416         inode = new_inode(dir->i_sb);
2417         if (!inode)
2418                 goto out;
2419
2420         /* Initialize the inode */
2421         ei = PROC_I(inode);
2422         inode->i_ino = get_next_ino();
2423         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2424
2425         /*
2426          * grab the reference to the task.
2427          */
2428         ei->pid = get_task_pid(task, PIDTYPE_PID);
2429         if (!ei->pid)
2430                 goto out_iput;
2431
2432         inode->i_mode = p->mode;
2433         if (S_ISDIR(inode->i_mode))
2434                 set_nlink(inode, 2);
2435         if (S_ISLNK(inode->i_mode))
2436                 inode->i_size = 64;
2437         if (p->iop)
2438                 inode->i_op = p->iop;
2439         if (p->fop)
2440                 inode->i_fop = p->fop;
2441         ei->op = p->op;
2442         d_add(dentry, inode);
2443         error = NULL;
2444 out:
2445         return error;
2446 out_iput:
2447         iput(inode);
2448         goto out;
2449 }
2450
2451 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2452 {
2453         struct dentry *error;
2454         struct task_struct *task = get_proc_task(dir);
2455         const struct pid_entry *p, *last;
2456
2457         error = ERR_PTR(-ENOENT);
2458
2459         if (!task)
2460                 goto out_no_task;
2461
2462         /* Lookup the directory entry */
2463         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2464         for (p = proc_base_stuff; p <= last; p++) {
2465                 if (p->len != dentry->d_name.len)
2466                         continue;
2467                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2468                         break;
2469         }
2470         if (p > last)
2471                 goto out;
2472
2473         error = proc_base_instantiate(dir, dentry, task, p);
2474
2475 out:
2476         put_task_struct(task);
2477 out_no_task:
2478         return error;
2479 }
2480
2481 static int proc_base_fill_cache(struct file *filp, void *dirent,
2482         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2483 {
2484         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2485                                 proc_base_instantiate, task, p);
2486 }
2487
2488 #ifdef CONFIG_TASK_IO_ACCOUNTING
2489 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2490 {
2491         struct task_io_accounting acct = task->ioac;
2492         unsigned long flags;
2493         int result;
2494
2495         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2496         if (result)
2497                 return result;
2498
2499         if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2500                 result = -EACCES;
2501                 goto out_unlock;
2502         }
2503
2504         if (whole && lock_task_sighand(task, &flags)) {
2505                 struct task_struct *t = task;
2506
2507                 task_io_accounting_add(&acct, &task->signal->ioac);
2508                 while_each_thread(task, t)
2509                         task_io_accounting_add(&acct, &t->ioac);
2510
2511                 unlock_task_sighand(task, &flags);
2512         }
2513         result = sprintf(buffer,
2514                         "rchar: %llu\n"
2515                         "wchar: %llu\n"
2516                         "syscr: %llu\n"
2517                         "syscw: %llu\n"
2518                         "read_bytes: %llu\n"
2519                         "write_bytes: %llu\n"
2520                         "cancelled_write_bytes: %llu\n",
2521                         (unsigned long long)acct.rchar,
2522                         (unsigned long long)acct.wchar,
2523                         (unsigned long long)acct.syscr,
2524                         (unsigned long long)acct.syscw,
2525                         (unsigned long long)acct.read_bytes,
2526                         (unsigned long long)acct.write_bytes,
2527                         (unsigned long long)acct.cancelled_write_bytes);
2528 out_unlock:
2529         mutex_unlock(&task->signal->cred_guard_mutex);
2530         return result;
2531 }
2532
2533 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2534 {
2535         return do_io_accounting(task, buffer, 0);
2536 }
2537
2538 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2539 {
2540         return do_io_accounting(task, buffer, 1);
2541 }
2542 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2543
2544 #ifdef CONFIG_USER_NS
2545 static int proc_id_map_open(struct inode *inode, struct file *file,
2546         struct seq_operations *seq_ops)
2547 {
2548         struct user_namespace *ns = NULL;
2549         struct task_struct *task;
2550         struct seq_file *seq;
2551         int ret = -EINVAL;
2552
2553         task = get_proc_task(inode);
2554         if (task) {
2555                 rcu_read_lock();
2556                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2557                 rcu_read_unlock();
2558                 put_task_struct(task);
2559         }
2560         if (!ns)
2561                 goto err;
2562
2563         ret = seq_open(file, seq_ops);
2564         if (ret)
2565                 goto err_put_ns;
2566
2567         seq = file->private_data;
2568         seq->private = ns;
2569
2570         return 0;
2571 err_put_ns:
2572         put_user_ns(ns);
2573 err:
2574         return ret;
2575 }
2576
2577 static int proc_id_map_release(struct inode *inode, struct file *file)
2578 {
2579         struct seq_file *seq = file->private_data;
2580         struct user_namespace *ns = seq->private;
2581         put_user_ns(ns);
2582         return seq_release(inode, file);
2583 }
2584
2585 static int proc_uid_map_open(struct inode *inode, struct file *file)
2586 {
2587         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2588 }
2589
2590 static int proc_gid_map_open(struct inode *inode, struct file *file)
2591 {
2592         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2593 }
2594
2595 static int proc_projid_map_open(struct inode *inode, struct file *file)
2596 {
2597         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2598 }
2599
2600 static const struct file_operations proc_uid_map_operations = {
2601         .open           = proc_uid_map_open,
2602         .write          = proc_uid_map_write,
2603         .read           = seq_read,
2604         .llseek         = seq_lseek,
2605         .release        = proc_id_map_release,
2606 };
2607
2608 static const struct file_operations proc_gid_map_operations = {
2609         .open           = proc_gid_map_open,
2610         .write          = proc_gid_map_write,
2611         .read           = seq_read,
2612         .llseek         = seq_lseek,
2613         .release        = proc_id_map_release,
2614 };
2615
2616 static const struct file_operations proc_projid_map_operations = {
2617         .open           = proc_projid_map_open,
2618         .write          = proc_projid_map_write,
2619         .read           = seq_read,
2620         .llseek         = seq_lseek,
2621         .release        = proc_id_map_release,
2622 };
2623 #endif /* CONFIG_USER_NS */
2624
2625 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2626                                 struct pid *pid, struct task_struct *task)
2627 {
2628         int err = lock_trace(task);
2629         if (!err) {
2630                 seq_printf(m, "%08x\n", task->personality);
2631                 unlock_trace(task);
2632         }
2633         return err;
2634 }
2635
2636 /*
2637  * Thread groups
2638  */
2639 static const struct file_operations proc_task_operations;
2640 static const struct inode_operations proc_task_inode_operations;
2641
2642 static const struct pid_entry tgid_base_stuff[] = {
2643         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2644         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2645 #ifdef CONFIG_CHECKPOINT_RESTORE
2646         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2647 #endif
2648         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2649         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2650 #ifdef CONFIG_NET
2651         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2652 #endif
2653         REG("environ",    S_IRUSR, proc_environ_operations),
2654         INF("auxv",       S_IRUSR, proc_pid_auxv),
2655         ONE("status",     S_IRUGO, proc_pid_status),
2656         ONE("personality", S_IRUGO, proc_pid_personality),
2657         INF("limits",     S_IRUGO, proc_pid_limits),
2658 #ifdef CONFIG_SCHED_DEBUG
2659         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2660 #endif
2661 #ifdef CONFIG_SCHED_AUTOGROUP
2662         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2663 #endif
2664         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2665 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2666         INF("syscall",    S_IRUGO, proc_pid_syscall),
2667 #endif
2668         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2669         ONE("stat",       S_IRUGO, proc_tgid_stat),
2670         ONE("statm",      S_IRUGO, proc_pid_statm),
2671         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2672 #ifdef CONFIG_NUMA
2673         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2674 #endif
2675         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2676         LNK("cwd",        proc_cwd_link),
2677         LNK("root",       proc_root_link),
2678         LNK("exe",        proc_exe_link),
2679         REG("mounts",     S_IRUGO, proc_mounts_operations),
2680         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2681         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2682 #ifdef CONFIG_PROC_PAGE_MONITOR
2683         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2684         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2685         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
2686 #endif
2687 #ifdef CONFIG_SECURITY
2688         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2689 #endif
2690 #ifdef CONFIG_KALLSYMS
2691         INF("wchan",      S_IRUGO, proc_pid_wchan),
2692 #endif
2693 #ifdef CONFIG_STACKTRACE
2694         ONE("stack",      S_IRUGO, proc_pid_stack),
2695 #endif
2696 #ifdef CONFIG_SCHEDSTATS
2697         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
2698 #endif
2699 #ifdef CONFIG_LATENCYTOP
2700         REG("latency",  S_IRUGO, proc_lstats_operations),
2701 #endif
2702 #ifdef CONFIG_PROC_PID_CPUSET
2703         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2704 #endif
2705 #ifdef CONFIG_CGROUPS
2706         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2707 #endif
2708         INF("oom_score",  S_IRUGO, proc_oom_score),
2709         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2710         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2711 #ifdef CONFIG_AUDITSYSCALL
2712         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2713         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2714 #endif
2715 #ifdef CONFIG_FAULT_INJECTION
2716         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2717 #endif
2718 #ifdef CONFIG_ELF_CORE
2719         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2720 #endif
2721 #ifdef CONFIG_TASK_IO_ACCOUNTING
2722         INF("io",       S_IRUSR, proc_tgid_io_accounting),
2723 #endif
2724 #ifdef CONFIG_HARDWALL
2725         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
2726 #endif
2727 #ifdef CONFIG_USER_NS
2728         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2729         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2730         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2731 #endif
2732 };
2733
2734 static int proc_tgid_base_readdir(struct file * filp,
2735                              void * dirent, filldir_t filldir)
2736 {
2737         return proc_pident_readdir(filp,dirent,filldir,
2738                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2739 }
2740
2741 static const struct file_operations proc_tgid_base_operations = {
2742         .read           = generic_read_dir,
2743         .readdir        = proc_tgid_base_readdir,
2744         .llseek         = default_llseek,
2745 };
2746
2747 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2748 {
2749         return proc_pident_lookup(dir, dentry,
2750                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2751 }
2752
2753 static const struct inode_operations proc_tgid_base_inode_operations = {
2754         .lookup         = proc_tgid_base_lookup,
2755         .getattr        = pid_getattr,
2756         .setattr        = proc_setattr,
2757         .permission     = proc_pid_permission,
2758 };
2759
2760 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2761 {
2762         struct dentry *dentry, *leader, *dir;
2763         char buf[PROC_NUMBUF];
2764         struct qstr name;
2765
2766         name.name = buf;
2767         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2768         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2769         if (dentry) {
2770                 shrink_dcache_parent(dentry);
2771                 d_drop(dentry);
2772                 dput(dentry);
2773         }
2774
2775         name.name = buf;
2776         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2777         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2778         if (!leader)
2779                 goto out;
2780
2781         name.name = "task";
2782         name.len = strlen(name.name);
2783         dir = d_hash_and_lookup(leader, &name);
2784         if (!dir)
2785                 goto out_put_leader;
2786
2787         name.name = buf;
2788         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2789         dentry = d_hash_and_lookup(dir, &name);
2790         if (dentry) {
2791                 shrink_dcache_parent(dentry);
2792                 d_drop(dentry);
2793                 dput(dentry);
2794         }
2795
2796         dput(dir);
2797 out_put_leader:
2798         dput(leader);
2799 out:
2800         return;
2801 }
2802
2803 /**
2804  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2805  * @task: task that should be flushed.
2806  *
2807  * When flushing dentries from proc, one needs to flush them from global
2808  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2809  * in. This call is supposed to do all of this job.
2810  *
2811  * Looks in the dcache for
2812  * /proc/@pid
2813  * /proc/@tgid/task/@pid
2814  * if either directory is present flushes it and all of it'ts children
2815  * from the dcache.
2816  *
2817  * It is safe and reasonable to cache /proc entries for a task until
2818  * that task exits.  After that they just clog up the dcache with
2819  * useless entries, possibly causing useful dcache entries to be
2820  * flushed instead.  This routine is proved to flush those useless
2821  * dcache entries at process exit time.
2822  *
2823  * NOTE: This routine is just an optimization so it does not guarantee
2824  *       that no dcache entries will exist at process exit time it
2825  *       just makes it very unlikely that any will persist.
2826  */
2827
2828 void proc_flush_task(struct task_struct *task)
2829 {
2830         int i;
2831         struct pid *pid, *tgid;
2832         struct upid *upid;
2833
2834         pid = task_pid(task);
2835         tgid = task_tgid(task);
2836
2837         for (i = 0; i <= pid->level; i++) {
2838                 upid = &pid->numbers[i];
2839                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2840                                         tgid->numbers[i].nr);
2841         }
2842
2843         upid = &pid->numbers[pid->level];
2844         if (upid->nr == 1)
2845                 pid_ns_release_proc(upid->ns);
2846 }
2847
2848 static struct dentry *proc_pid_instantiate(struct inode *dir,
2849                                            struct dentry * dentry,
2850                                            struct task_struct *task, const void *ptr)
2851 {
2852         struct dentry *error = ERR_PTR(-ENOENT);
2853         struct inode *inode;
2854
2855         inode = proc_pid_make_inode(dir->i_sb, task);
2856         if (!inode)
2857                 goto out;
2858
2859         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2860         inode->i_op = &proc_tgid_base_inode_operations;
2861         inode->i_fop = &proc_tgid_base_operations;
2862         inode->i_flags|=S_IMMUTABLE;
2863
2864         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
2865                                                   ARRAY_SIZE(tgid_base_stuff)));
2866
2867         d_set_d_op(dentry, &pid_dentry_operations);
2868
2869         d_add(dentry, inode);
2870         /* Close the race of the process dying before we return the dentry */
2871         if (pid_revalidate(dentry, 0))
2872                 error = NULL;
2873 out:
2874         return error;
2875 }
2876
2877 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
2878 {
2879         struct dentry *result;
2880         struct task_struct *task;
2881         unsigned tgid;
2882         struct pid_namespace *ns;
2883
2884         result = proc_base_lookup(dir, dentry);
2885         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
2886                 goto out;
2887
2888         tgid = name_to_int(dentry);
2889         if (tgid == ~0U)
2890                 goto out;
2891
2892         ns = dentry->d_sb->s_fs_info;
2893         rcu_read_lock();
2894         task = find_task_by_pid_ns(tgid, ns);
2895         if (task)
2896                 get_task_struct(task);
2897         rcu_read_unlock();
2898         if (!task)
2899                 goto out;
2900
2901         result = proc_pid_instantiate(dir, dentry, task, NULL);
2902         put_task_struct(task);
2903 out:
2904         return result;
2905 }
2906
2907 /*
2908  * Find the first task with tgid >= tgid
2909  *
2910  */
2911 struct tgid_iter {
2912         unsigned int tgid;
2913         struct task_struct *task;
2914 };
2915 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2916 {
2917         struct pid *pid;
2918
2919         if (iter.task)
2920                 put_task_struct(iter.task);
2921         rcu_read_lock();
2922 retry:
2923         iter.task = NULL;
2924         pid = find_ge_pid(iter.tgid, ns);
2925         if (pid) {
2926                 iter.tgid = pid_nr_ns(pid, ns);
2927                 iter.task = pid_task(pid, PIDTYPE_PID);
2928                 /* What we to know is if the pid we have find is the
2929                  * pid of a thread_group_leader.  Testing for task
2930                  * being a thread_group_leader is the obvious thing
2931                  * todo but there is a window when it fails, due to
2932                  * the pid transfer logic in de_thread.
2933                  *
2934                  * So we perform the straight forward test of seeing
2935                  * if the pid we have found is the pid of a thread
2936                  * group leader, and don't worry if the task we have
2937                  * found doesn't happen to be a thread group leader.
2938                  * As we don't care in the case of readdir.
2939                  */
2940                 if (!iter.task || !has_group_leader_pid(iter.task)) {
2941                         iter.tgid += 1;
2942                         goto retry;
2943                 }
2944                 get_task_struct(iter.task);
2945         }
2946         rcu_read_unlock();
2947         return iter;
2948 }
2949
2950 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
2951
2952 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2953         struct tgid_iter iter)
2954 {
2955         char name[PROC_NUMBUF];
2956         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
2957         return proc_fill_cache(filp, dirent, filldir, name, len,
2958                                 proc_pid_instantiate, iter.task, NULL);
2959 }
2960
2961 static int fake_filldir(void *buf, const char *name, int namelen,
2962                         loff_t offset, u64 ino, unsigned d_type)
2963 {
2964         return 0;
2965 }
2966
2967 /* for the /proc/ directory itself, after non-process stuff has been done */
2968 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2969 {
2970         unsigned int nr;
2971         struct task_struct *reaper;
2972         struct tgid_iter iter;
2973         struct pid_namespace *ns;
2974         filldir_t __filldir;
2975
2976         if (filp->f_pos >= PID_MAX_LIMIT + TGID_OFFSET)
2977                 goto out_no_task;
2978         nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2979
2980         reaper = get_proc_task(filp->f_path.dentry->d_inode);
2981         if (!reaper)
2982                 goto out_no_task;
2983
2984         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
2985                 const struct pid_entry *p = &proc_base_stuff[nr];
2986                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
2987                         goto out;
2988         }
2989
2990         ns = filp->f_dentry->d_sb->s_fs_info;
2991         iter.task = NULL;
2992         iter.tgid = filp->f_pos - TGID_OFFSET;
2993         for (iter = next_tgid(ns, iter);
2994              iter.task;
2995              iter.tgid += 1, iter = next_tgid(ns, iter)) {
2996                 if (has_pid_permissions(ns, iter.task, 2))
2997                         __filldir = filldir;
2998                 else
2999                         __filldir = fake_filldir;
3000
3001                 filp->f_pos = iter.tgid + TGID_OFFSET;
3002                 if (proc_pid_fill_cache(filp, dirent, __filldir, iter) < 0) {
3003                         put_task_struct(iter.task);
3004                         goto out;
3005                 }
3006         }
3007         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
3008 out:
3009         put_task_struct(reaper);
3010 out_no_task:
3011         return 0;
3012 }
3013
3014 /*
3015  * Tasks
3016  */
3017 static const struct pid_entry tid_base_stuff[] = {
3018         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3019         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3020         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3021         REG("environ",   S_IRUSR, proc_environ_operations),
3022         INF("auxv",      S_IRUSR, proc_pid_auxv),
3023         ONE("status",    S_IRUGO, proc_pid_status),
3024         ONE("personality", S_IRUGO, proc_pid_personality),
3025         INF("limits",    S_IRUGO, proc_pid_limits),
3026 #ifdef CONFIG_SCHED_DEBUG
3027         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3028 #endif
3029         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3030 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3031         INF("syscall",   S_IRUGO, proc_pid_syscall),
3032 #endif
3033         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
3034         ONE("stat",      S_IRUGO, proc_tid_stat),
3035         ONE("statm",     S_IRUGO, proc_pid_statm),
3036         REG("maps",      S_IRUGO, proc_tid_maps_operations),
3037 #ifdef CONFIG_CHECKPOINT_RESTORE
3038         REG("children",  S_IRUGO, proc_tid_children_operations),
3039 #endif
3040 #ifdef CONFIG_NUMA
3041         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3042 #endif
3043         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3044         LNK("cwd",       proc_cwd_link),
3045         LNK("root",      proc_root_link),
3046         LNK("exe",       proc_exe_link),
3047         REG("mounts",    S_IRUGO, proc_mounts_operations),
3048         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3049 #ifdef CONFIG_PROC_PAGE_MONITOR
3050         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3051         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3052         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
3053 #endif
3054 #ifdef CONFIG_SECURITY
3055         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3056 #endif
3057 #ifdef CONFIG_KALLSYMS
3058         INF("wchan",     S_IRUGO, proc_pid_wchan),
3059 #endif
3060 #ifdef CONFIG_STACKTRACE
3061         ONE("stack",      S_IRUGO, proc_pid_stack),
3062 #endif
3063 #ifdef CONFIG_SCHEDSTATS
3064         INF("schedstat", S_IRUGO, proc_pid_schedstat),
3065 #endif
3066 #ifdef CONFIG_LATENCYTOP
3067         REG("latency",  S_IRUGO, proc_lstats_operations),
3068 #endif
3069 #ifdef CONFIG_PROC_PID_CPUSET
3070         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
3071 #endif
3072 #ifdef CONFIG_CGROUPS
3073         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3074 #endif
3075         INF("oom_score", S_IRUGO, proc_oom_score),
3076         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3077         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3078 #ifdef CONFIG_AUDITSYSCALL
3079         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3080         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3081 #endif
3082 #ifdef CONFIG_FAULT_INJECTION
3083         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3084 #endif
3085 #ifdef CONFIG_TASK_IO_ACCOUNTING
3086         INF("io",       S_IRUSR, proc_tid_io_accounting),
3087 #endif
3088 #ifdef CONFIG_HARDWALL
3089         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
3090 #endif
3091 #ifdef CONFIG_USER_NS
3092         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3093         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3094         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3095 #endif
3096 };
3097
3098 static int proc_tid_base_readdir(struct file * filp,
3099                              void * dirent, filldir_t filldir)
3100 {
3101         return proc_pident_readdir(filp,dirent,filldir,
3102                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
3103 }
3104
3105 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3106 {
3107         return proc_pident_lookup(dir, dentry,
3108                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3109 }
3110
3111 static const struct file_operations proc_tid_base_operations = {
3112         .read           = generic_read_dir,
3113         .readdir        = proc_tid_base_readdir,
3114         .llseek         = default_llseek,
3115 };
3116
3117 static const struct inode_operations proc_tid_base_inode_operations = {
3118         .lookup         = proc_tid_base_lookup,
3119         .getattr        = pid_getattr,
3120         .setattr        = proc_setattr,
3121 };
3122
3123 static struct dentry *proc_task_instantiate(struct inode *dir,
3124         struct dentry *dentry, struct task_struct *task, const void *ptr)
3125 {
3126         struct dentry *error = ERR_PTR(-ENOENT);
3127         struct inode *inode;
3128         inode = proc_pid_make_inode(dir->i_sb, task);
3129
3130         if (!inode)
3131                 goto out;
3132         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3133         inode->i_op = &proc_tid_base_inode_operations;
3134         inode->i_fop = &proc_tid_base_operations;
3135         inode->i_flags|=S_IMMUTABLE;
3136
3137         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3138                                                   ARRAY_SIZE(tid_base_stuff)));
3139
3140         d_set_d_op(dentry, &pid_dentry_operations);
3141
3142         d_add(dentry, inode);
3143         /* Close the race of the process dying before we return the dentry */
3144         if (pid_revalidate(dentry, 0))
3145                 error = NULL;
3146 out:
3147         return error;
3148 }
3149
3150 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3151 {
3152         struct dentry *result = ERR_PTR(-ENOENT);
3153         struct task_struct *task;
3154         struct task_struct *leader = get_proc_task(dir);
3155         unsigned tid;
3156         struct pid_namespace *ns;
3157
3158         if (!leader)
3159                 goto out_no_task;
3160
3161         tid = name_to_int(dentry);
3162         if (tid == ~0U)
3163                 goto out;
3164
3165         ns = dentry->d_sb->s_fs_info;
3166         rcu_read_lock();
3167         task = find_task_by_pid_ns(tid, ns);
3168         if (task)
3169                 get_task_struct(task);
3170         rcu_read_unlock();
3171         if (!task)
3172                 goto out;
3173         if (!same_thread_group(leader, task))
3174                 goto out_drop_task;
3175
3176         result = proc_task_instantiate(dir, dentry, task, NULL);
3177 out_drop_task:
3178         put_task_struct(task);
3179 out:
3180         put_task_struct(leader);
3181 out_no_task:
3182         return result;
3183 }
3184
3185 /*
3186  * Find the first tid of a thread group to return to user space.
3187  *
3188  * Usually this is just the thread group leader, but if the users
3189  * buffer was too small or there was a seek into the middle of the
3190  * directory we have more work todo.
3191  *
3192  * In the case of a short read we start with find_task_by_pid.
3193  *
3194  * In the case of a seek we start with the leader and walk nr
3195  * threads past it.
3196  */
3197 static struct task_struct *first_tid(struct task_struct *leader,
3198                 int tid, int nr, struct pid_namespace *ns)
3199 {
3200         struct task_struct *pos;
3201
3202         rcu_read_lock();
3203         /* Attempt to start with the pid of a thread */
3204         if (tid && (nr > 0)) {
3205                 pos = find_task_by_pid_ns(tid, ns);
3206                 if (pos && (pos->group_leader == leader))
3207                         goto found;
3208         }
3209
3210         /* If nr exceeds the number of threads there is nothing todo */
3211         pos = NULL;
3212         if (nr && nr >= get_nr_threads(leader))
3213                 goto out;
3214
3215         /* If we haven't found our starting place yet start
3216          * with the leader and walk nr threads forward.
3217          */
3218         for (pos = leader; nr > 0; --nr) {
3219                 pos = next_thread(pos);
3220                 if (pos == leader) {
3221                         pos = NULL;
3222                         goto out;
3223                 }
3224         }
3225 found:
3226         get_task_struct(pos);
3227 out:
3228         rcu_read_unlock();
3229         return pos;
3230 }
3231
3232 /*
3233  * Find the next thread in the thread list.
3234  * Return NULL if there is an error or no next thread.
3235  *
3236  * The reference to the input task_struct is released.
3237  */
3238 static struct task_struct *next_tid(struct task_struct *start)
3239 {
3240         struct task_struct *pos = NULL;
3241         rcu_read_lock();
3242         if (pid_alive(start)) {
3243                 pos = next_thread(start);
3244                 if (thread_group_leader(pos))
3245                         pos = NULL;
3246                 else
3247                         get_task_struct(pos);
3248         }
3249         rcu_read_unlock();
3250         put_task_struct(start);
3251         return pos;
3252 }
3253
3254 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3255         struct task_struct *task, int tid)
3256 {
3257         char name[PROC_NUMBUF];
3258         int len = snprintf(name, sizeof(name), "%d", tid);
3259         return proc_fill_cache(filp, dirent, filldir, name, len,
3260                                 proc_task_instantiate, task, NULL);
3261 }
3262
3263 /* for the /proc/TGID/task/ directories */
3264 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3265 {
3266         struct dentry *dentry = filp->f_path.dentry;
3267         struct inode *inode = dentry->d_inode;
3268         struct task_struct *leader = NULL;
3269         struct task_struct *task;
3270         int retval = -ENOENT;
3271         ino_t ino;
3272         int tid;
3273         struct pid_namespace *ns;
3274
3275         task = get_proc_task(inode);
3276         if (!task)
3277                 goto out_no_task;
3278         rcu_read_lock();
3279         if (pid_alive(task)) {
3280                 leader = task->group_leader;
3281                 get_task_struct(leader);
3282         }
3283         rcu_read_unlock();
3284         put_task_struct(task);
3285         if (!leader)
3286                 goto out_no_task;
3287         retval = 0;
3288
3289         switch ((unsigned long)filp->f_pos) {
3290         case 0:
3291                 ino = inode->i_ino;
3292                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3293                         goto out;
3294                 filp->f_pos++;
3295                 /* fall through */
3296         case 1:
3297                 ino = parent_ino(dentry);
3298                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3299                         goto out;
3300                 filp->f_pos++;
3301                 /* fall through */
3302         }
3303
3304         /* f_version caches the tgid value that the last readdir call couldn't
3305          * return. lseek aka telldir automagically resets f_version to 0.
3306          */
3307         ns = filp->f_dentry->d_sb->s_fs_info;
3308         tid = (int)filp->f_version;
3309         filp->f_version = 0;
3310         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3311              task;
3312              task = next_tid(task), filp->f_pos++) {
3313                 tid = task_pid_nr_ns(task, ns);
3314                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3315                         /* returning this tgid failed, save it as the first
3316                          * pid for the next readir call */
3317                         filp->f_version = (u64)tid;
3318                         put_task_struct(task);
3319                         break;
3320                 }
3321         }
3322 out:
3323         put_task_struct(leader);
3324 out_no_task:
3325         return retval;
3326 }
3327
3328 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3329 {
3330         struct inode *inode = dentry->d_inode;
3331         struct task_struct *p = get_proc_task(inode);
3332         generic_fillattr(inode, stat);
3333
3334         if (p) {
3335                 stat->nlink += get_nr_threads(p);
3336                 put_task_struct(p);
3337         }
3338
3339         return 0;
3340 }
3341
3342 static const struct inode_operations proc_task_inode_operations = {
3343         .lookup         = proc_task_lookup,
3344         .getattr        = proc_task_getattr,
3345         .setattr        = proc_setattr,
3346         .permission     = proc_pid_permission,
3347 };
3348
3349 static const struct file_operations proc_task_operations = {
3350         .read           = generic_read_dir,
3351         .readdir        = proc_task_readdir,
3352         .llseek         = default_llseek,
3353 };