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
  94 /* NOTE:
  95  *      Implementing inode permission operations in /proc is almost
  96  *      certainly an error.  Permission checks need to happen during
  97  *      each system call not at open time.  The reason is that most of
  98  *      what we wish to check for permissions in /proc varies at runtime.
  99  *
 100  *      The classic example of a problem is opening file descriptors
 101  *      in /proc for a task before it execs a suid executable.
 102  */
 103
 104 struct pid_entry {
 105         char *name;
 106         int len;
 107         umode_t mode;
 108         const struct inode_operations *iop;
 109         const struct file_operations *fop;
 110         union proc_op op;
 111 };
 112
 113 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 114         .name = (NAME),                                 \
 115         .len  = sizeof(NAME) - 1,                       \
 116         .mode = MODE,                                   \
 117         .iop  = IOP,                                    \
 118         .fop  = FOP,                                    \
 119         .op   = OP,                                     \
 120 }
 121
 122 #define DIR(NAME, MODE, iops, fops)     \
 123         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 124 #define LNK(NAME, get_link)                                     \
 125         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 126                 &proc_pid_link_inode_operations, NULL,          \
 127                 { .proc_get_link = get_link } )
 128 #define REG(NAME, MODE, fops)                           \
 129         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 130 #define INF(NAME, MODE, read)                           \
 131         NOD(NAME, (S_IFREG|(MODE)),                     \
 132                 NULL, &proc_info_file_operations,       \
 133                 { .proc_read = read } )
 134 #define ONE(NAME, MODE, show)                           \
 135         NOD(NAME, (S_IFREG|(MODE)),                     \
 136                 NULL, &proc_single_file_operations,     \
 137                 { .proc_show = show } )
 138
 139 static int proc_fd_permission(struct inode *inode, int mask);
 140
 141 /*
 142  * Count the number of hardlinks for the pid_entry table, excluding the .
 143  * and .. links.
 144  */
 145 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
 146         unsigned int n)
 147 {
 148         unsigned int i;
 149         unsigned int count;
 150
 151         count = 0;
 152         for (i = 0; i < n; ++i) {
 153                 if (S_ISDIR(entries[i].mode))
 154                         ++count;
 155         }
 156
 157         return count;
 158 }
 159
 160 static int get_task_root(struct task_struct *task, struct path *root)
 161 {
 162         int result = -ENOENT;
 163
 164         task_lock(task);
 165         if (task->fs) {
 166                 get_fs_root(task->fs, root);
 167                 result = 0;
 168         }
 169         task_unlock(task);
 170         return result;
 171 }
 172
 173 static int proc_cwd_link(struct dentry *dentry, struct path *path)
 174 {
 175         struct task_struct *task = get_proc_task(dentry->d_inode);
 176         int result = -ENOENT;
 177
 178         if (task) {
 179                 task_lock(task);
 180                 if (task->fs) {
 181                         get_fs_pwd(task->fs, path);
 182                         result = 0;
 183                 }
 184                 task_unlock(task);
 185                 put_task_struct(task);
 186         }
 187         return result;
 188 }
 189
 190 static int proc_root_link(struct dentry *dentry, struct path *path)
 191 {
 192         struct task_struct *task = get_proc_task(dentry->d_inode);
 193         int result = -ENOENT;
 194
 195         if (task) {
 196                 result = get_task_root(task, path);
 197                 put_task_struct(task);
 198         }
 199         return result;
 200 }
 201
 202 struct mm_struct *mm_for_maps(struct task_struct *task)
 203 {
 204         return mm_access(task, PTRACE_MODE_READ);
 205 }
 206
 207 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
 208 {
 209         int res = 0;
 210         unsigned int len;
 211         struct mm_struct *mm = get_task_mm(task);
 212         if (!mm)
 213                 goto out;
 214         if (!mm->arg_end)
 215                 goto out_mm;    /* Shh! No looking before we're done */
 216
 217         len = mm->arg_end - mm->arg_start;
 218
 219         if (len > PAGE_SIZE)
 220                 len = PAGE_SIZE;
 221
 222         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
 223
 224         // If the nul at the end of args has been overwritten, then
 225         // assume application is using setproctitle(3).
 226         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
 227                 len = strnlen(buffer, res);
 228                 if (len < res) {
 229                     res = len;
 230                 } else {
 231                         len = mm->env_end - mm->env_start;
 232                         if (len > PAGE_SIZE - res)
 233                                 len = PAGE_SIZE - res;
 234                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
 235                         res = strnlen(buffer, res);
 236                 }
 237         }
 238 out_mm:
 239         mmput(mm);
 240 out:
 241         return res;
 242 }
 243
 244 static int proc_pid_auxv(struct task_struct *task, char *buffer)
 245 {
 246         struct mm_struct *mm = mm_for_maps(task);
 247         int res = PTR_ERR(mm);
 248         if (mm && !IS_ERR(mm)) {
 249                 unsigned int nwords = 0;
 250                 do {
 251                         nwords += 2;
 252                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 253                 res = nwords * sizeof(mm->saved_auxv[0]);
 254                 if (res > PAGE_SIZE)
 255                         res = PAGE_SIZE;
 256                 memcpy(buffer, mm->saved_auxv, res);
 257                 mmput(mm);
 258         }
 259         return res;
 260 }
 261
 262
 263 #ifdef CONFIG_KALLSYMS
 264 /*
 265  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 266  * Returns the resolved symbol.  If that fails, simply return the address.
 267  */
 268 static int proc_pid_wchan(struct task_struct *task, char *buffer)
 269 {
 270         unsigned long wchan;
 271         char symname[KSYM_NAME_LEN];
 272
 273         wchan = get_wchan(task);
 274
 275         if (lookup_symbol_name(wchan, symname) < 0)
 276                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
 277                         return 0;
 278                 else
 279                         return sprintf(buffer, "%lu", wchan);
 280         else
 281                 return sprintf(buffer, "%s", symname);
 282 }
 283 #endif /* CONFIG_KALLSYMS */
 284
 285 static int lock_trace(struct task_struct *task)
 286 {
 287         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
 288         if (err)
 289                 return err;
 290         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH)) {
 291                 mutex_unlock(&task->signal->cred_guard_mutex);
 292                 return -EPERM;
 293         }
 294         return 0;
 295 }
 296
 297 static void unlock_trace(struct task_struct *task)
 298 {
 299         mutex_unlock(&task->signal->cred_guard_mutex);
 300 }
 301
 302 #ifdef CONFIG_STACKTRACE
 303
 304 #define MAX_STACK_TRACE_DEPTH   64
 305
 306 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 307                           struct pid *pid, struct task_struct *task)
 308 {
 309         struct stack_trace trace;
 310         unsigned long *entries;
 311         int err;
 312         int i;
 313
 314         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 315         if (!entries)
 316                 return -ENOMEM;
 317
 318         trace.nr_entries        = 0;
 319         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 320         trace.entries           = entries;
 321         trace.skip              = 0;
 322
 323         err = lock_trace(task);
 324         if (!err) {
 325                 save_stack_trace_tsk(task, &trace);
 326
 327                 for (i = 0; i < trace.nr_entries; i++) {
 328                         seq_printf(m, "[<%pK>] %pS\n",
 329                                    (void *)entries[i], (void *)entries[i]);
 330                 }
 331                 unlock_trace(task);
 332         }
 333         kfree(entries);
 334
 335         return err;
 336 }
 337 #endif
 338
 339 #ifdef CONFIG_SCHEDSTATS
 340 /*
 341  * Provides /proc/PID/schedstat
 342  */
 343 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
 344 {
 345         return sprintf(buffer, "%llu %llu %lu\n",
 346                         (unsigned long long)task->se.sum_exec_runtime,
 347                         (unsigned long long)task->sched_info.run_delay,
 348                         task->sched_info.pcount);
 349 }
 350 #endif
 351
 352 #ifdef CONFIG_LATENCYTOP
 353 static int lstats_show_proc(struct seq_file *m, void *v)
 354 {
 355         int i;
 356         struct inode *inode = m->private;
 357         struct task_struct *task = get_proc_task(inode);
 358
 359         if (!task)
 360                 return -ESRCH;
 361         seq_puts(m, "Latency Top version : v0.1\n");
 362         for (i = 0; i < 32; i++) {
 363                 struct latency_record *lr = &task->latency_record[i];
 364                 if (lr->backtrace[0]) {
 365                         int q;
 366                         seq_printf(m, "%i %li %li",
 367                                    lr->count, lr->time, lr->max);
 368                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 369                                 unsigned long bt = lr->backtrace[q];
 370                                 if (!bt)
 371                                         break;
 372                                 if (bt == ULONG_MAX)
 373                                         break;
 374                                 seq_printf(m, " %ps", (void *)bt);
 375                         }
 376                         seq_putc(m, '\n');
 377                 }
 378
 379         }
 380         put_task_struct(task);
 381         return 0;
 382 }
 383
 384 static int lstats_open(struct inode *inode, struct file *file)
 385 {
 386         return single_open(file, lstats_show_proc, inode);
 387 }
 388
 389 static ssize_t lstats_write(struct file *file, const char __user *buf,
 390                             size_t count, loff_t *offs)
 391 {
 392         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 393
 394         if (!task)
 395                 return -ESRCH;
 396         clear_all_latency_tracing(task);
 397         put_task_struct(task);
 398
 399         return count;
 400 }
 401
 402 static const struct file_operations proc_lstats_operations = {
 403         .open           = lstats_open,
 404         .read           = seq_read,
 405         .write          = lstats_write,
 406         .llseek         = seq_lseek,
 407         .release        = single_release,
 408 };
 409
 410 #endif
 411
 412 static int proc_oom_score(struct task_struct *task, char *buffer)
 413 {
 414         unsigned long totalpages = totalram_pages + total_swap_pages;
 415         unsigned long points = 0;
 416
 417         read_lock(&tasklist_lock);
 418         if (pid_alive(task))
 419                 points = oom_badness(task, NULL, NULL, totalpages) *
 420                                                 1000 / totalpages;
 421         read_unlock(&tasklist_lock);
 422         return sprintf(buffer, "%lu\n", points);
 423 }
 424
 425 struct limit_names {
 426         char *name;
 427         char *unit;
 428 };
 429
 430 static const struct limit_names lnames[RLIM_NLIMITS] = {
 431         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 432         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 433         [RLIMIT_DATA] = {"Max data size", "bytes"},
 434         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 435         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 436         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 437         [RLIMIT_NPROC] = {"Max processes", "processes"},
 438         [RLIMIT_NOFILE] = {"Max open files", "files"},
 439         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 440         [RLIMIT_AS] = {"Max address space", "bytes"},
 441         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 442         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 443         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 444         [RLIMIT_NICE] = {"Max nice priority", NULL},
 445         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 446         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 447 };
 448
 449 /* Display limits for a process */
 450 static int proc_pid_limits(struct task_struct *task, char *buffer)
 451 {
 452         unsigned int i;
 453         int count = 0;
 454         unsigned long flags;
 455         char *bufptr = buffer;
 456
 457         struct rlimit rlim[RLIM_NLIMITS];
 458
 459         if (!lock_task_sighand(task, &flags))
 460                 return 0;
 461         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 462         unlock_task_sighand(task, &flags);
 463
 464         /*
 465          * print the file header
 466          */
 467         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
 468                         "Limit", "Soft Limit", "Hard Limit", "Units");
 469
 470         for (i = 0; i < RLIM_NLIMITS; i++) {
 471                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 472                         count += sprintf(&bufptr[count], "%-25s %-20s ",
 473                                          lnames[i].name, "unlimited");
 474                 else
 475                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
 476                                          lnames[i].name, rlim[i].rlim_cur);
 477
 478                 if (rlim[i].rlim_max == RLIM_INFINITY)
 479                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
 480                 else
 481                         count += sprintf(&bufptr[count], "%-20lu ",
 482                                          rlim[i].rlim_max);
 483
 484                 if (lnames[i].unit)
 485                         count += sprintf(&bufptr[count], "%-10s\n",
 486                                          lnames[i].unit);
 487                 else
 488                         count += sprintf(&bufptr[count], "\n");
 489         }
 490
 491         return count;
 492 }
 493
 494 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 495 static int proc_pid_syscall(struct task_struct *task, char *buffer)
 496 {
 497         long nr;
 498         unsigned long args[6], sp, pc;
 499         int res = lock_trace(task);
 500         if (res)
 501                 return res;
 502
 503         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 504                 res = sprintf(buffer, "running\n");
 505         else if (nr < 0)
 506                 res = sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 507         else
 508                 res = sprintf(buffer,
 509                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 510                        nr,
 511                        args[0], args[1], args[2], args[3], args[4], args[5],
 512                        sp, pc);
 513         unlock_trace(task);
 514         return res;
 515 }
 516 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 517
 518 /************************************************************************/
 519 /*                       Here the fs part begins                        */
 520 /************************************************************************/
 521
 522 /* permission checks */
 523 static int proc_fd_access_allowed(struct inode *inode)
 524 {
 525         struct task_struct *task;
 526         int allowed = 0;
 527         /* Allow access to a task's file descriptors if it is us or we
 528          * may use ptrace attach to the process and find out that
 529          * information.
 530          */
 531         task = get_proc_task(inode);
 532         if (task) {
 533                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
 534                 put_task_struct(task);
 535         }
 536         return allowed;
 537 }
 538
 539 int proc_setattr(struct dentry *dentry, struct iattr *attr)
 540 {
 541         int error;
 542         struct inode *inode = dentry->d_inode;
 543
 544         if (attr->ia_valid & ATTR_MODE)
 545                 return -EPERM;
 546
 547         error = inode_change_ok(inode, attr);
 548         if (error)
 549                 return error;
 550
 551         if ((attr->ia_valid & ATTR_SIZE) &&
 552             attr->ia_size != i_size_read(inode)) {
 553                 error = vmtruncate(inode, attr->ia_size);
 554                 if (error)
 555                         return error;
 556         }
 557
 558         setattr_copy(inode, attr);
 559         mark_inode_dirty(inode);
 560         return 0;
 561 }
 562
 563 /*
 564  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
 565  * or euid/egid (for hide_pid_min=2)?
 566  */
 567 static bool has_pid_permissions(struct pid_namespace *pid,
 568                                  struct task_struct *task,
 569                                  int hide_pid_min)
 570 {
 571         if (pid->hide_pid < hide_pid_min)
 572                 return true;
 573         if (in_group_p(pid->pid_gid))
 574                 return true;
 575         return ptrace_may_access(task, PTRACE_MODE_READ);
 576 }
 577
 578
 579 static int proc_pid_permission(struct inode *inode, int mask)
 580 {
 581         struct pid_namespace *pid = inode->i_sb->s_fs_info;
 582         struct task_struct *task;
 583         bool has_perms;
 584
 585         task = get_proc_task(inode);
 586         if (!task)
 587                 return -ESRCH;
 588         has_perms = has_pid_permissions(pid, task, 1);
 589         put_task_struct(task);
 590
 591         if (!has_perms) {
 592                 if (pid->hide_pid == 2) {
 593                         /*
 594                          * Let's make getdents(), stat(), and open()
 595                          * consistent with each other.  If a process
 596                          * may not stat() a file, it shouldn't be seen
 597                          * in procfs at all.
 598                          */
 599                         return -ENOENT;
 600                 }
 601
 602                 return -EPERM;
 603         }
 604         return generic_permission(inode, mask);
 605 }
 606
 607
 608
 609 static const struct inode_operations proc_def_inode_operations = {
 610         .setattr        = proc_setattr,
 611 };
 612
 613 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 614
 615 static ssize_t proc_info_read(struct file * file, char __user * buf,
 616                           size_t count, loff_t *ppos)
 617 {
 618         struct inode * inode = file->f_path.dentry->d_inode;
 619         unsigned long page;
 620         ssize_t length;
 621         struct task_struct *task = get_proc_task(inode);
 622
 623         length = -ESRCH;
 624         if (!task)
 625                 goto out_no_task;
 626
 627         if (count > PROC_BLOCK_SIZE)
 628                 count = PROC_BLOCK_SIZE;
 629
 630         length = -ENOMEM;
 631         if (!(page = __get_free_page(GFP_TEMPORARY)))
 632                 goto out;
 633
 634         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 635
 636         if (length >= 0)
 637                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 638         free_page(page);
 639 out:
 640         put_task_struct(task);
 641 out_no_task:
 642         return length;
 643 }
 644
 645 static const struct file_operations proc_info_file_operations = {
 646         .read           = proc_info_read,
 647         .llseek         = generic_file_llseek,
 648 };
 649
 650 static int proc_single_show(struct seq_file *m, void *v)
 651 {
 652         struct inode *inode = m->private;
 653         struct pid_namespace *ns;
 654         struct pid *pid;
 655         struct task_struct *task;
 656         int ret;
 657
 658         ns = inode->i_sb->s_fs_info;
 659         pid = proc_pid(inode);
 660         task = get_pid_task(pid, PIDTYPE_PID);
 661         if (!task)
 662                 return -ESRCH;
 663
 664         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 665
 666         put_task_struct(task);
 667         return ret;
 668 }
 669
 670 static int proc_single_open(struct inode *inode, struct file *filp)
 671 {
 672         return single_open(filp, proc_single_show, inode);
 673 }
 674
 675 static const struct file_operations proc_single_file_operations = {
 676         .open           = proc_single_open,
 677         .read           = seq_read,
 678         .llseek         = seq_lseek,
 679         .release        = single_release,
 680 };
 681
 682 static int mem_open(struct inode* inode, struct file* file)
 683 {
 684         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 685         struct mm_struct *mm;
 686
 687         if (!task)
 688                 return -ESRCH;
 689
 690         mm = mm_access(task, PTRACE_MODE_ATTACH);
 691         put_task_struct(task);
 692
 693         if (IS_ERR(mm))
 694                 return PTR_ERR(mm);
 695
 696         if (mm) {
 697                 /* ensure this mm_struct can't be freed */
 698                 atomic_inc(&mm->mm_count);
 699                 /* but do not pin its memory */
 700                 mmput(mm);
 701         }
 702
 703         /* OK to pass negative loff_t, we can catch out-of-range */
 704         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 705         file->private_data = mm;
 706
 707         return 0;
 708 }
 709
 710 static ssize_t mem_rw(struct file *file, char __user *buf,
 711                         size_t count, loff_t *ppos, int write)
 712 {
 713         struct mm_struct *mm = file->private_data;
 714         unsigned long addr = *ppos;
 715         ssize_t copied;
 716         char *page;
 717
 718         if (!mm)
 719                 return 0;
 720
 721         page = (char *)__get_free_page(GFP_TEMPORARY);
 722         if (!page)
 723                 return -ENOMEM;
 724
 725         copied = 0;
 726         if (!atomic_inc_not_zero(&mm->mm_users))
 727                 goto free;
 728
 729         while (count > 0) {
 730                 int this_len = min_t(int, count, PAGE_SIZE);
 731
 732                 if (write && copy_from_user(page, buf, this_len)) {
 733                         copied = -EFAULT;
 734                         break;
 735                 }
 736
 737                 this_len = access_remote_vm(mm, addr, page, this_len, write);
 738                 if (!this_len) {
 739                         if (!copied)
 740                                 copied = -EIO;
 741                         break;
 742                 }
 743
 744                 if (!write && copy_to_user(buf, page, this_len)) {
 745                         copied = -EFAULT;
 746                         break;
 747                 }
 748
 749                 buf += this_len;
 750                 addr += this_len;
 751                 copied += this_len;
 752                 count -= this_len;
 753         }
 754         *ppos = addr;
 755
 756         mmput(mm);
 757 free:
 758         free_page((unsigned long) page);
 759         return copied;
 760 }
 761
 762 static ssize_t mem_read(struct file *file, char __user *buf,
 763                         size_t count, loff_t *ppos)
 764 {
 765         return mem_rw(file, buf, count, ppos, 0);
 766 }
 767
 768 static ssize_t mem_write(struct file *file, const char __user *buf,
 769                          size_t count, loff_t *ppos)
 770 {
 771         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 772 }
 773
 774 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 775 {
 776         switch (orig) {
 777         case 0:
 778                 file->f_pos = offset;
 779                 break;
 780         case 1:
 781                 file->f_pos += offset;
 782                 break;
 783         default:
 784                 return -EINVAL;
 785         }
 786         force_successful_syscall_return();
 787         return file->f_pos;
 788 }
 789
 790 static int mem_release(struct inode *inode, struct file *file)
 791 {
 792         struct mm_struct *mm = file->private_data;
 793         if (mm)
 794                 mmdrop(mm);
 795         return 0;
 796 }
 797
 798 static const struct file_operations proc_mem_operations = {
 799         .llseek         = mem_lseek,
 800         .read           = mem_read,
 801         .write          = mem_write,
 802         .open           = mem_open,
 803         .release        = mem_release,
 804 };
 805
 806 static ssize_t environ_read(struct file *file, char __user *buf,
 807                         size_t count, loff_t *ppos)
 808 {
 809         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 810         char *page;
 811         unsigned long src = *ppos;
 812         int ret = -ESRCH;
 813         struct mm_struct *mm;
 814
 815         if (!task)
 816                 goto out_no_task;
 817
 818         ret = -ENOMEM;
 819         page = (char *)__get_free_page(GFP_TEMPORARY);
 820         if (!page)
 821                 goto out;
 822
 823
 824         mm = mm_for_maps(task);
 825         ret = PTR_ERR(mm);
 826         if (!mm || IS_ERR(mm))
 827                 goto out_free;
 828
 829         ret = 0;
 830         while (count > 0) {
 831                 int this_len, retval, max_len;
 832
 833                 this_len = mm->env_end - (mm->env_start + src);
 834
 835                 if (this_len <= 0)
 836                         break;
 837
 838                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 839                 this_len = (this_len > max_len) ? max_len : this_len;
 840
 841                 retval = access_process_vm(task, (mm->env_start + src),
 842                         page, this_len, 0);
 843
 844                 if (retval <= 0) {
 845                         ret = retval;
 846                         break;
 847                 }
 848
 849                 if (copy_to_user(buf, page, retval)) {
 850                         ret = -EFAULT;
 851                         break;
 852                 }
 853
 854                 ret += retval;
 855                 src += retval;
 856                 buf += retval;
 857                 count -= retval;
 858         }
 859         *ppos = src;
 860
 861         mmput(mm);
 862 out_free:
 863         free_page((unsigned long) page);
 864 out:
 865         put_task_struct(task);
 866 out_no_task:
 867         return ret;
 868 }
 869
 870 static const struct file_operations proc_environ_operations = {
 871         .read           = environ_read,
 872         .llseek         = generic_file_llseek,
 873 };
 874
 875 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
 876                                 size_t count, loff_t *ppos)
 877 {
 878         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 879         char buffer[PROC_NUMBUF];
 880         size_t len;
 881         int oom_adjust = OOM_DISABLE;
 882         unsigned long flags;
 883
 884         if (!task)
 885                 return -ESRCH;
 886
 887         if (lock_task_sighand(task, &flags)) {
 888                 oom_adjust = task->signal->oom_adj;
 889                 unlock_task_sighand(task, &flags);
 890         }
 891
 892         put_task_struct(task);
 893
 894         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
 895
 896         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 897 }
 898
 899 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
 900                                 size_t count, loff_t *ppos)
 901 {
 902         struct task_struct *task;
 903         char buffer[PROC_NUMBUF];
 904         int oom_adjust;
 905         unsigned long flags;
 906         int err;
 907
 908         memset(buffer, 0, sizeof(buffer));
 909         if (count > sizeof(buffer) - 1)
 910                 count = sizeof(buffer) - 1;
 911         if (copy_from_user(buffer, buf, count)) {
 912                 err = -EFAULT;
 913                 goto out;
 914         }
 915
 916         err = kstrtoint(strstrip(buffer), 0, &oom_adjust);
 917         if (err)
 918                 goto out;
 919         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
 920              oom_adjust != OOM_DISABLE) {
 921                 err = -EINVAL;
 922                 goto out;
 923         }
 924
 925         task = get_proc_task(file->f_path.dentry->d_inode);
 926         if (!task) {
 927                 err = -ESRCH;
 928                 goto out;
 929         }
 930
 931         task_lock(task);
 932         if (!task->mm) {
 933                 err = -EINVAL;
 934                 goto err_task_lock;
 935         }
 936
 937         if (!lock_task_sighand(task, &flags)) {
 938                 err = -ESRCH;
 939                 goto err_task_lock;
 940         }
 941
 942         if (oom_adjust < task->signal->oom_adj && !capable(CAP_SYS_RESOURCE)) {
 943                 err = -EACCES;
 944                 goto err_sighand;
 945         }
 946
 947         /*
 948          * Warn that /proc/pid/oom_adj is deprecated, see
 949          * Documentation/feature-removal-schedule.txt.
 950          */
 951         printk_once(KERN_WARNING "%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
 952                   current->comm, task_pid_nr(current), task_pid_nr(task),
 953                   task_pid_nr(task));
 954         task->signal->oom_adj = oom_adjust;
 955         /*
 956          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
 957          * value is always attainable.
 958          */
 959         if (task->signal->oom_adj == OOM_ADJUST_MAX)
 960                 task->signal->oom_score_adj = OOM_SCORE_ADJ_MAX;
 961         else
 962                 task->signal->oom_score_adj = (oom_adjust * OOM_SCORE_ADJ_MAX) /
 963                                                                 -OOM_DISABLE;
 964         trace_oom_score_adj_update(task);
 965 err_sighand:
 966         unlock_task_sighand(task, &flags);
 967 err_task_lock:
 968         task_unlock(task);
 969         put_task_struct(task);
 970 out:
 971         return err < 0 ? err : count;
 972 }
 973
 974 static const struct file_operations proc_oom_adjust_operations = {
 975         .read           = oom_adjust_read,
 976         .write          = oom_adjust_write,
 977         .llseek         = generic_file_llseek,
 978 };
 979
 980 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
 981                                         size_t count, loff_t *ppos)
 982 {
 983         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 984         char buffer[PROC_NUMBUF];
 985         int oom_score_adj = OOM_SCORE_ADJ_MIN;
 986         unsigned long flags;
 987         size_t len;
 988
 989         if (!task)
 990                 return -ESRCH;
 991         if (lock_task_sighand(task, &flags)) {
 992                 oom_score_adj = task->signal->oom_score_adj;
 993                 unlock_task_sighand(task, &flags);
 994         }
 995         put_task_struct(task);
 996         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_score_adj);
 997         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 998 }
 999
1000 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1001                                         size_t count, loff_t *ppos)
1002 {
1003         struct task_struct *task;
1004         char buffer[PROC_NUMBUF];
1005         unsigned long flags;
1006         int oom_score_adj;
1007         int err;
1008
1009         memset(buffer, 0, sizeof(buffer));
1010         if (count > sizeof(buffer) - 1)
1011                 count = sizeof(buffer) - 1;
1012         if (copy_from_user(buffer, buf, count)) {
1013                 err = -EFAULT;
1014                 goto out;
1015         }
1016
1017         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1018         if (err)
1019                 goto out;
1020         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1021                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1022                 err = -EINVAL;
1023                 goto out;
1024         }
1025
1026         task = get_proc_task(file->f_path.dentry->d_inode);
1027         if (!task) {
1028                 err = -ESRCH;
1029                 goto out;
1030         }
1031
1032         task_lock(task);
1033         if (!task->mm) {
1034                 err = -EINVAL;
1035                 goto err_task_lock;
1036         }
1037
1038         if (!lock_task_sighand(task, &flags)) {
1039                 err = -ESRCH;
1040                 goto err_task_lock;
1041         }
1042
1043         if (oom_score_adj < task->signal->oom_score_adj_min &&
1044                         !capable(CAP_SYS_RESOURCE)) {
1045                 err = -EACCES;
1046                 goto err_sighand;
1047         }
1048
1049         task->signal->oom_score_adj = oom_score_adj;
1050         if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1051                 task->signal->oom_score_adj_min = oom_score_adj;
1052         trace_oom_score_adj_update(task);
1053         /*
1054          * Scale /proc/pid/oom_adj appropriately ensuring that OOM_DISABLE is
1055          * always attainable.
1056          */
1057         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1058                 task->signal->oom_adj = OOM_DISABLE;
1059         else
1060                 task->signal->oom_adj = (oom_score_adj * OOM_ADJUST_MAX) /
1061                                                         OOM_SCORE_ADJ_MAX;
1062 err_sighand:
1063         unlock_task_sighand(task, &flags);
1064 err_task_lock:
1065         task_unlock(task);
1066         put_task_struct(task);
1067 out:
1068         return err < 0 ? err : count;
1069 }
1070
1071 static const struct file_operations proc_oom_score_adj_operations = {
1072         .read           = oom_score_adj_read,
1073         .write          = oom_score_adj_write,
1074         .llseek         = default_llseek,
1075 };
1076
1077 #ifdef CONFIG_AUDITSYSCALL
1078 #define TMPBUFLEN 21
1079 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1080                                   size_t count, loff_t *ppos)
1081 {
1082         struct inode * inode = file->f_path.dentry->d_inode;
1083         struct task_struct *task = get_proc_task(inode);
1084         ssize_t length;
1085         char tmpbuf[TMPBUFLEN];
1086
1087         if (!task)
1088                 return -ESRCH;
1089         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1090                                 audit_get_loginuid(task));
1091         put_task_struct(task);
1092         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1093 }
1094
1095 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1096                                    size_t count, loff_t *ppos)
1097 {
1098         struct inode * inode = file->f_path.dentry->d_inode;
1099         char *page, *tmp;
1100         ssize_t length;
1101         uid_t loginuid;
1102
1103         rcu_read_lock();
1104         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1105                 rcu_read_unlock();
1106                 return -EPERM;
1107         }
1108         rcu_read_unlock();
1109
1110         if (count >= PAGE_SIZE)
1111                 count = PAGE_SIZE - 1;
1112
1113         if (*ppos != 0) {
1114                 /* No partial writes. */
1115                 return -EINVAL;
1116         }
1117         page = (char*)__get_free_page(GFP_TEMPORARY);
1118         if (!page)
1119                 return -ENOMEM;
1120         length = -EFAULT;
1121         if (copy_from_user(page, buf, count))
1122                 goto out_free_page;
1123
1124         page[count] = '\0';
1125         loginuid = simple_strtoul(page, &tmp, 10);
1126         if (tmp == page) {
1127                 length = -EINVAL;
1128                 goto out_free_page;
1129
1130         }
1131         length = audit_set_loginuid(loginuid);
1132         if (likely(length == 0))
1133                 length = count;
1134
1135 out_free_page:
1136         free_page((unsigned long) page);
1137         return length;
1138 }
1139
1140 static const struct file_operations proc_loginuid_operations = {
1141         .read           = proc_loginuid_read,
1142         .write          = proc_loginuid_write,
1143         .llseek         = generic_file_llseek,
1144 };
1145
1146 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1147                                   size_t count, loff_t *ppos)
1148 {
1149         struct inode * inode = file->f_path.dentry->d_inode;
1150         struct task_struct *task = get_proc_task(inode);
1151         ssize_t length;
1152         char tmpbuf[TMPBUFLEN];
1153
1154         if (!task)
1155                 return -ESRCH;
1156         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1157                                 audit_get_sessionid(task));
1158         put_task_struct(task);
1159         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1160 }
1161
1162 static const struct file_operations proc_sessionid_operations = {
1163         .read           = proc_sessionid_read,
1164         .llseek         = generic_file_llseek,
1165 };
1166 #endif
1167
1168 #ifdef CONFIG_FAULT_INJECTION
1169 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1170                                       size_t count, loff_t *ppos)
1171 {
1172         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1173         char buffer[PROC_NUMBUF];
1174         size_t len;
1175         int make_it_fail;
1176
1177         if (!task)
1178                 return -ESRCH;
1179         make_it_fail = task->make_it_fail;
1180         put_task_struct(task);
1181
1182         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1183
1184         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1185 }
1186
1187 static ssize_t proc_fault_inject_write(struct file * file,
1188                         const char __user * buf, size_t count, loff_t *ppos)
1189 {
1190         struct task_struct *task;
1191         char buffer[PROC_NUMBUF], *end;
1192         int make_it_fail;
1193
1194         if (!capable(CAP_SYS_RESOURCE))
1195                 return -EPERM;
1196         memset(buffer, 0, sizeof(buffer));
1197         if (count > sizeof(buffer) - 1)
1198                 count = sizeof(buffer) - 1;
1199         if (copy_from_user(buffer, buf, count))
1200                 return -EFAULT;
1201         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1202         if (*end)
1203                 return -EINVAL;
1204         task = get_proc_task(file->f_dentry->d_inode);
1205         if (!task)
1206                 return -ESRCH;
1207         task->make_it_fail = make_it_fail;
1208         put_task_struct(task);
1209
1210         return count;
1211 }
1212
1213 static const struct file_operations proc_fault_inject_operations = {
1214         .read           = proc_fault_inject_read,
1215         .write          = proc_fault_inject_write,
1216         .llseek         = generic_file_llseek,
1217 };
1218 #endif
1219
1220
1221 #ifdef CONFIG_SCHED_DEBUG
1222 /*
1223  * Print out various scheduling related per-task fields:
1224  */
1225 static int sched_show(struct seq_file *m, void *v)
1226 {
1227         struct inode *inode = m->private;
1228         struct task_struct *p;
1229
1230         p = get_proc_task(inode);
1231         if (!p)
1232                 return -ESRCH;
1233         proc_sched_show_task(p, m);
1234
1235         put_task_struct(p);
1236
1237         return 0;
1238 }
1239
1240 static ssize_t
1241 sched_write(struct file *file, const char __user *buf,
1242             size_t count, loff_t *offset)
1243 {
1244         struct inode *inode = file->f_path.dentry->d_inode;
1245         struct task_struct *p;
1246
1247         p = get_proc_task(inode);
1248         if (!p)
1249                 return -ESRCH;
1250         proc_sched_set_task(p);
1251
1252         put_task_struct(p);
1253
1254         return count;
1255 }
1256
1257 static int sched_open(struct inode *inode, struct file *filp)
1258 {
1259         return single_open(filp, sched_show, inode);
1260 }
1261
1262 static const struct file_operations proc_pid_sched_operations = {
1263         .open           = sched_open,
1264         .read           = seq_read,
1265         .write          = sched_write,
1266         .llseek         = seq_lseek,
1267         .release        = single_release,
1268 };
1269
1270 #endif
1271
1272 #ifdef CONFIG_SCHED_AUTOGROUP
1273 /*
1274  * Print out autogroup related information:
1275  */
1276 static int sched_autogroup_show(struct seq_file *m, void *v)
1277 {
1278         struct inode *inode = m->private;
1279         struct task_struct *p;
1280
1281         p = get_proc_task(inode);
1282         if (!p)
1283                 return -ESRCH;
1284         proc_sched_autogroup_show_task(p, m);
1285
1286         put_task_struct(p);
1287
1288         return 0;
1289 }
1290
1291 static ssize_t
1292 sched_autogroup_write(struct file *file, const char __user *buf,
1293             size_t count, loff_t *offset)
1294 {
1295         struct inode *inode = file->f_path.dentry->d_inode;
1296         struct task_struct *p;
1297         char buffer[PROC_NUMBUF];
1298         int nice;
1299         int err;
1300
1301         memset(buffer, 0, sizeof(buffer));
1302         if (count > sizeof(buffer) - 1)
1303                 count = sizeof(buffer) - 1;
1304         if (copy_from_user(buffer, buf, count))
1305                 return -EFAULT;
1306
1307         err = kstrtoint(strstrip(buffer), 0, &nice);
1308         if (err < 0)
1309                 return err;
1310
1311         p = get_proc_task(inode);
1312         if (!p)
1313                 return -ESRCH;
1314
1315         err = proc_sched_autogroup_set_nice(p, nice);
1316         if (err)
1317                 count = err;
1318
1319         put_task_struct(p);
1320
1321         return count;
1322 }
1323
1324 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1325 {
1326         int ret;
1327
1328         ret = single_open(filp, sched_autogroup_show, NULL);
1329         if (!ret) {
1330                 struct seq_file *m = filp->private_data;
1331
1332                 m->private = inode;
1333         }
1334         return ret;
1335 }
1336
1337 static const struct file_operations proc_pid_sched_autogroup_operations = {
1338         .open           = sched_autogroup_open,
1339         .read           = seq_read,
1340         .write          = sched_autogroup_write,
1341         .llseek         = seq_lseek,
1342         .release        = single_release,
1343 };
1344
1345 #endif /* CONFIG_SCHED_AUTOGROUP */
1346
1347 static ssize_t comm_write(struct file *file, const char __user *buf,
1348                                 size_t count, loff_t *offset)
1349 {
1350         struct inode *inode = file->f_path.dentry->d_inode;
1351         struct task_struct *p;
1352         char buffer[TASK_COMM_LEN];
1353
1354         memset(buffer, 0, sizeof(buffer));
1355         if (count > sizeof(buffer) - 1)
1356                 count = sizeof(buffer) - 1;
1357         if (copy_from_user(buffer, buf, count))
1358                 return -EFAULT;
1359
1360         p = get_proc_task(inode);
1361         if (!p)
1362                 return -ESRCH;
1363
1364         if (same_thread_group(current, p))
1365                 set_task_comm(p, buffer);
1366         else
1367                 count = -EINVAL;
1368
1369         put_task_struct(p);
1370
1371         return count;
1372 }
1373
1374 static int comm_show(struct seq_file *m, void *v)
1375 {
1376         struct inode *inode = m->private;
1377         struct task_struct *p;
1378
1379         p = get_proc_task(inode);
1380         if (!p)
1381                 return -ESRCH;
1382
1383         task_lock(p);
1384         seq_printf(m, "%s\n", p->comm);
1385         task_unlock(p);
1386
1387         put_task_struct(p);
1388
1389         return 0;
1390 }
1391
1392 static int comm_open(struct inode *inode, struct file *filp)
1393 {
1394         return single_open(filp, comm_show, inode);
1395 }
1396
1397 static const struct file_operations proc_pid_set_comm_operations = {
1398         .open           = comm_open,
1399         .read           = seq_read,
1400         .write          = comm_write,
1401         .llseek         = seq_lseek,
1402         .release        = single_release,
1403 };
1404
1405 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1406 {
1407         struct task_struct *task;
1408         struct mm_struct *mm;
1409         struct file *exe_file;
1410
1411         task = get_proc_task(dentry->d_inode);
1412         if (!task)
1413                 return -ENOENT;
1414         mm = get_task_mm(task);
1415         put_task_struct(task);
1416         if (!mm)
1417                 return -ENOENT;
1418         exe_file = get_mm_exe_file(mm);
1419         mmput(mm);
1420         if (exe_file) {
1421                 *exe_path = exe_file->f_path;
1422                 path_get(&exe_file->f_path);
1423                 fput(exe_file);
1424                 return 0;
1425         } else
1426                 return -ENOENT;
1427 }
1428
1429 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1430 {
1431         struct inode *inode = dentry->d_inode;
1432         int error = -EACCES;
1433
1434         /* We don't need a base pointer in the /proc filesystem */
1435         path_put(&nd->path);
1436
1437         /* Are we allowed to snoop on the tasks file descriptors? */
1438         if (!proc_fd_access_allowed(inode))
1439                 goto out;
1440
1441         error = PROC_I(inode)->op.proc_get_link(dentry, &nd->path);
1442 out:
1443         return ERR_PTR(error);
1444 }
1445
1446 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1447 {
1448         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1449         char *pathname;
1450         int len;
1451
1452         if (!tmp)
1453                 return -ENOMEM;
1454
1455         pathname = d_path(path, tmp, PAGE_SIZE);
1456         len = PTR_ERR(pathname);
1457         if (IS_ERR(pathname))
1458                 goto out;
1459         len = tmp + PAGE_SIZE - 1 - pathname;
1460
1461         if (len > buflen)
1462                 len = buflen;
1463         if (copy_to_user(buffer, pathname, len))
1464                 len = -EFAULT;
1465  out:
1466         free_page((unsigned long)tmp);
1467         return len;
1468 }
1469
1470 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1471 {
1472         int error = -EACCES;
1473         struct inode *inode = dentry->d_inode;
1474         struct path path;
1475
1476         /* Are we allowed to snoop on the tasks file descriptors? */
1477         if (!proc_fd_access_allowed(inode))
1478                 goto out;
1479
1480         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1481         if (error)
1482                 goto out;
1483
1484         error = do_proc_readlink(&path, buffer, buflen);
1485         path_put(&path);
1486 out:
1487         return error;
1488 }
1489
1490 static const struct inode_operations proc_pid_link_inode_operations = {
1491         .readlink       = proc_pid_readlink,
1492         .follow_link    = proc_pid_follow_link,
1493         .setattr        = proc_setattr,
1494 };
1495
1496
1497 /* building an inode */
1498
1499 static int task_dumpable(struct task_struct *task)
1500 {
1501         int dumpable = 0;
1502         struct mm_struct *mm;
1503
1504         task_lock(task);
1505         mm = task->mm;
1506         if (mm)
1507                 dumpable = get_dumpable(mm);
1508         task_unlock(task);
1509         if(dumpable == 1)
1510                 return 1;
1511         return 0;
1512 }
1513
1514 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1515 {
1516         struct inode * inode;
1517         struct proc_inode *ei;
1518         const struct cred *cred;
1519
1520         /* We need a new inode */
1521
1522         inode = new_inode(sb);
1523         if (!inode)
1524                 goto out;
1525
1526         /* Common stuff */
1527         ei = PROC_I(inode);
1528         inode->i_ino = get_next_ino();
1529         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1530         inode->i_op = &proc_def_inode_operations;
1531
1532         /*
1533          * grab the reference to task.
1534          */
1535         ei->pid = get_task_pid(task, PIDTYPE_PID);
1536         if (!ei->pid)
1537                 goto out_unlock;
1538
1539         if (task_dumpable(task)) {
1540                 rcu_read_lock();
1541                 cred = __task_cred(task);
1542                 inode->i_uid = cred->euid;
1543                 inode->i_gid = cred->egid;
1544                 rcu_read_unlock();
1545         }
1546         security_task_to_inode(task, inode);
1547
1548 out:
1549         return inode;
1550
1551 out_unlock:
1552         iput(inode);
1553         return NULL;
1554 }
1555
1556 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1557 {
1558         struct inode *inode = dentry->d_inode;
1559         struct task_struct *task;
1560         const struct cred *cred;
1561         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1562
1563         generic_fillattr(inode, stat);
1564
1565         rcu_read_lock();
1566         stat->uid = GLOBAL_ROOT_UID;
1567         stat->gid = GLOBAL_ROOT_GID;
1568         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1569         if (task) {
1570                 if (!has_pid_permissions(pid, task, 2)) {
1571                         rcu_read_unlock();
1572                         /*
1573                          * This doesn't prevent learning whether PID exists,
1574                          * it only makes getattr() consistent with readdir().
1575                          */
1576                         return -ENOENT;
1577                 }
1578                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1579                     task_dumpable(task)) {
1580                         cred = __task_cred(task);
1581                         stat->uid = cred->euid;
1582                         stat->gid = cred->egid;
1583                 }
1584         }
1585         rcu_read_unlock();
1586         return 0;
1587 }
1588
1589 /* dentry stuff */
1590
1591 /*
1592  *      Exceptional case: normally we are not allowed to unhash a busy
1593  * directory. In this case, however, we can do it - no aliasing problems
1594  * due to the way we treat inodes.
1595  *
1596  * Rewrite the inode's ownerships here because the owning task may have
1597  * performed a setuid(), etc.
1598  *
1599  * Before the /proc/pid/status file was created the only way to read
1600  * the effective uid of a /process was to stat /proc/pid.  Reading
1601  * /proc/pid/status is slow enough that procps and other packages
1602  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1603  * made this apply to all per process world readable and executable
1604  * directories.
1605  */
1606 int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1607 {
1608         struct inode *inode;
1609         struct task_struct *task;
1610         const struct cred *cred;
1611
1612         if (nd && nd->flags & LOOKUP_RCU)
1613                 return -ECHILD;
1614
1615         inode = dentry->d_inode;
1616         task = get_proc_task(inode);
1617
1618         if (task) {
1619                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1620                     task_dumpable(task)) {
1621                         rcu_read_lock();
1622                         cred = __task_cred(task);
1623                         inode->i_uid = cred->euid;
1624                         inode->i_gid = cred->egid;
1625                         rcu_read_unlock();
1626                 } else {
1627                         inode->i_uid = GLOBAL_ROOT_UID;
1628                         inode->i_gid = GLOBAL_ROOT_GID;
1629                 }
1630                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1631                 security_task_to_inode(task, inode);
1632                 put_task_struct(task);
1633                 return 1;
1634         }
1635         d_drop(dentry);
1636         return 0;
1637 }
1638
1639 static int pid_delete_dentry(const struct dentry * dentry)
1640 {
1641         /* Is the task we represent dead?
1642          * If so, then don't put the dentry on the lru list,
1643          * kill it immediately.
1644          */
1645         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1646 }
1647
1648 const struct dentry_operations pid_dentry_operations =
1649 {
1650         .d_revalidate   = pid_revalidate,
1651         .d_delete       = pid_delete_dentry,
1652 };
1653
1654 /* Lookups */
1655
1656 /*
1657  * Fill a directory entry.
1658  *
1659  * If possible create the dcache entry and derive our inode number and
1660  * file type from dcache entry.
1661  *
1662  * Since all of the proc inode numbers are dynamically generated, the inode
1663  * numbers do not exist until the inode is cache.  This means creating the
1664  * the dcache entry in readdir is necessary to keep the inode numbers
1665  * reported by readdir in sync with the inode numbers reported
1666  * by stat.
1667  */
1668 int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1669         const char *name, int len,
1670         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1671 {
1672         struct dentry *child, *dir = filp->f_path.dentry;
1673         struct inode *inode;
1674         struct qstr qname;
1675         ino_t ino = 0;
1676         unsigned type = DT_UNKNOWN;
1677
1678         qname.name = name;
1679         qname.len  = len;
1680         qname.hash = full_name_hash(name, len);
1681
1682         child = d_lookup(dir, &qname);
1683         if (!child) {
1684                 struct dentry *new;
1685                 new = d_alloc(dir, &qname);
1686                 if (new) {
1687                         child = instantiate(dir->d_inode, new, task, ptr);
1688                         if (child)
1689                                 dput(new);
1690                         else
1691                                 child = new;
1692                 }
1693         }
1694         if (!child || IS_ERR(child) || !child->d_inode)
1695                 goto end_instantiate;
1696         inode = child->d_inode;
1697         if (inode) {
1698                 ino = inode->i_ino;
1699                 type = inode->i_mode >> 12;
1700         }
1701         dput(child);
1702 end_instantiate:
1703         if (!ino)
1704                 ino = find_inode_number(dir, &qname);
1705         if (!ino)
1706                 ino = 1;
1707         return filldir(dirent, name, len, filp->f_pos, ino, type);
1708 }
1709
1710 static unsigned name_to_int(struct dentry *dentry)
1711 {
1712         const char *name = dentry->d_name.name;
1713         int len = dentry->d_name.len;
1714         unsigned n = 0;
1715
1716         if (len > 1 && *name == '0')
1717                 goto out;
1718         while (len-- > 0) {
1719                 unsigned c = *name++ - '0';
1720                 if (c > 9)
1721                         goto out;
1722                 if (n >= (~0U-9)/10)
1723                         goto out;
1724                 n *= 10;
1725                 n += c;
1726         }
1727         return n;
1728 out:
1729         return ~0U;
1730 }
1731
1732 #define PROC_FDINFO_MAX 64
1733
1734 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1735 {
1736         struct task_struct *task = get_proc_task(inode);
1737         struct files_struct *files = NULL;
1738         struct file *file;
1739         int fd = proc_fd(inode);
1740
1741         if (task) {
1742                 files = get_files_struct(task);
1743                 put_task_struct(task);
1744         }
1745         if (files) {
1746                 /*
1747                  * We are not taking a ref to the file structure, so we must
1748                  * hold ->file_lock.
1749                  */
1750                 spin_lock(&files->file_lock);
1751                 file = fcheck_files(files, fd);
1752                 if (file) {
1753                         unsigned int f_flags;
1754                         struct fdtable *fdt;
1755
1756                         fdt = files_fdtable(files);
1757                         f_flags = file->f_flags & ~O_CLOEXEC;
1758                         if (close_on_exec(fd, fdt))
1759                                 f_flags |= O_CLOEXEC;
1760
1761                         if (path) {
1762                                 *path = file->f_path;
1763                                 path_get(&file->f_path);
1764                         }
1765                         if (info)
1766                                 snprintf(info, PROC_FDINFO_MAX,
1767                                          "pos:\t%lli\n"
1768                                          "flags:\t0%o\n",
1769                                          (long long) file->f_pos,
1770                                          f_flags);
1771                         spin_unlock(&files->file_lock);
1772                         put_files_struct(files);
1773                         return 0;
1774                 }
1775                 spin_unlock(&files->file_lock);
1776                 put_files_struct(files);
1777         }
1778         return -ENOENT;
1779 }
1780
1781 static int proc_fd_link(struct dentry *dentry, struct path *path)
1782 {
1783         return proc_fd_info(dentry->d_inode, path, NULL);
1784 }
1785
1786 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1787 {
1788         struct inode *inode;
1789         struct task_struct *task;
1790         int fd;
1791         struct files_struct *files;
1792         const struct cred *cred;
1793
1794         if (nd && nd->flags & LOOKUP_RCU)
1795                 return -ECHILD;
1796
1797         inode = dentry->d_inode;
1798         task = get_proc_task(inode);
1799         fd = proc_fd(inode);
1800
1801         if (task) {
1802                 files = get_files_struct(task);
1803                 if (files) {
1804                         struct file *file;
1805                         rcu_read_lock();
1806                         file = fcheck_files(files, fd);
1807                         if (file) {
1808                                 unsigned i_mode, f_mode = file->f_mode;
1809
1810                                 rcu_read_unlock();
1811                                 put_files_struct(files);
1812
1813                                 if (task_dumpable(task)) {
1814                                         rcu_read_lock();
1815                                         cred = __task_cred(task);
1816                                         inode->i_uid = cred->euid;
1817                                         inode->i_gid = cred->egid;
1818                                         rcu_read_unlock();
1819                                 } else {
1820                                         inode->i_uid = GLOBAL_ROOT_UID;
1821                                         inode->i_gid = GLOBAL_ROOT_GID;
1822                                 }
1823
1824                                 i_mode = S_IFLNK;
1825                                 if (f_mode & FMODE_READ)
1826                                         i_mode |= S_IRUSR | S_IXUSR;
1827                                 if (f_mode & FMODE_WRITE)
1828                                         i_mode |= S_IWUSR | S_IXUSR;
1829                                 inode->i_mode = i_mode;
1830
1831                                 security_task_to_inode(task, inode);
1832                                 put_task_struct(task);
1833                                 return 1;
1834                         }
1835                         rcu_read_unlock();
1836                         put_files_struct(files);
1837                 }
1838                 put_task_struct(task);
1839         }
1840         d_drop(dentry);
1841         return 0;
1842 }
1843
1844 static const struct dentry_operations tid_fd_dentry_operations =
1845 {
1846         .d_revalidate   = tid_fd_revalidate,
1847         .d_delete       = pid_delete_dentry,
1848 };
1849
1850 static struct dentry *proc_fd_instantiate(struct inode *dir,
1851         struct dentry *dentry, struct task_struct *task, const void *ptr)
1852 {
1853         unsigned fd = *(const unsigned *)ptr;
1854         struct inode *inode;
1855         struct proc_inode *ei;
1856         struct dentry *error = ERR_PTR(-ENOENT);
1857
1858         inode = proc_pid_make_inode(dir->i_sb, task);
1859         if (!inode)
1860                 goto out;
1861         ei = PROC_I(inode);
1862         ei->fd = fd;
1863
1864         inode->i_op = &proc_pid_link_inode_operations;
1865         inode->i_size = 64;
1866         ei->op.proc_get_link = proc_fd_link;
1867         d_set_d_op(dentry, &tid_fd_dentry_operations);
1868         d_add(dentry, inode);
1869         /* Close the race of the process dying before we return the dentry */
1870         if (tid_fd_revalidate(dentry, NULL))
1871                 error = NULL;
1872
1873  out:
1874         return error;
1875 }
1876
1877 static struct dentry *proc_lookupfd_common(struct inode *dir,
1878                                            struct dentry *dentry,
1879                                            instantiate_t instantiate)
1880 {
1881         struct task_struct *task = get_proc_task(dir);
1882         unsigned fd = name_to_int(dentry);
1883         struct dentry *result = ERR_PTR(-ENOENT);
1884
1885         if (!task)
1886                 goto out_no_task;
1887         if (fd == ~0U)
1888                 goto out;
1889
1890         result = instantiate(dir, dentry, task, &fd);
1891 out:
1892         put_task_struct(task);
1893 out_no_task:
1894         return result;
1895 }
1896
1897 static int proc_readfd_common(struct file * filp, void * dirent,
1898                               filldir_t filldir, instantiate_t instantiate)
1899 {
1900         struct dentry *dentry = filp->f_path.dentry;
1901         struct inode *inode = dentry->d_inode;
1902         struct task_struct *p = get_proc_task(inode);
1903         unsigned int fd, ino;
1904         int retval;
1905         struct files_struct * files;
1906
1907         retval = -ENOENT;
1908         if (!p)
1909                 goto out_no_task;
1910         retval = 0;
1911
1912         fd = filp->f_pos;
1913         switch (fd) {
1914                 case 0:
1915                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1916                                 goto out;
1917                         filp->f_pos++;
1918                 case 1:
1919                         ino = parent_ino(dentry);
1920                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1921                                 goto out;
1922                         filp->f_pos++;
1923                 default:
1924                         files = get_files_struct(p);
1925                         if (!files)
1926                                 goto out;
1927                         rcu_read_lock();
1928                         for (fd = filp->f_pos-2;
1929                              fd < files_fdtable(files)->max_fds;
1930                              fd++, filp->f_pos++) {
1931                                 char name[PROC_NUMBUF];
1932                                 int len;
1933
1934                                 if (!fcheck_files(files, fd))
1935                                         continue;
1936                                 rcu_read_unlock();
1937
1938                                 len = snprintf(name, sizeof(name), "%d", fd);
1939                                 if (proc_fill_cache(filp, dirent, filldir,
1940                                                     name, len, instantiate,
1941                                                     p, &fd) < 0) {
1942                                         rcu_read_lock();
1943                                         break;
1944                                 }
1945                                 rcu_read_lock();
1946                         }
1947                         rcu_read_unlock();
1948                         put_files_struct(files);
1949         }
1950 out:
1951         put_task_struct(p);
1952 out_no_task:
1953         return retval;
1954 }
1955
1956 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1957                                     struct nameidata *nd)
1958 {
1959         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1960 }
1961
1962 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1963 {
1964         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1965 }
1966
1967 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1968                                       size_t len, loff_t *ppos)
1969 {
1970         char tmp[PROC_FDINFO_MAX];
1971         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
1972         if (!err)
1973                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1974         return err;
1975 }
1976
1977 static const struct file_operations proc_fdinfo_file_operations = {
1978         .open           = nonseekable_open,
1979         .read           = proc_fdinfo_read,
1980         .llseek         = no_llseek,
1981 };
1982
1983 static const struct file_operations proc_fd_operations = {
1984         .read           = generic_read_dir,
1985         .readdir        = proc_readfd,
1986         .llseek         = default_llseek,
1987 };
1988
1989 #ifdef CONFIG_CHECKPOINT_RESTORE
1990
1991 /*
1992  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1993  * which represent vma start and end addresses.
1994  */
1995 static int dname_to_vma_addr(struct dentry *dentry,
1996                              unsigned long *start, unsigned long *end)
1997 {
1998         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1999                 return -EINVAL;
2000
2001         return 0;
2002 }
2003
2004 static int map_files_d_revalidate(struct dentry *dentry, struct nameidata *nd)
2005 {
2006         unsigned long vm_start, vm_end;
2007         bool exact_vma_exists = false;
2008         struct mm_struct *mm = NULL;
2009         struct task_struct *task;
2010         const struct cred *cred;
2011         struct inode *inode;
2012         int status = 0;
2013
2014         if (nd && nd->flags & LOOKUP_RCU)
2015                 return -ECHILD;
2016
2017         if (!capable(CAP_SYS_ADMIN)) {
2018                 status = -EACCES;
2019                 goto out_notask;
2020         }
2021
2022         inode = dentry->d_inode;
2023         task = get_proc_task(inode);
2024         if (!task)
2025                 goto out_notask;
2026
2027         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2028                 goto out;
2029
2030         mm = get_task_mm(task);
2031         if (!mm)
2032                 goto out;
2033
2034         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
2035                 down_read(&mm->mmap_sem);
2036                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
2037                 up_read(&mm->mmap_sem);
2038         }
2039
2040         mmput(mm);
2041
2042         if (exact_vma_exists) {
2043                 if (task_dumpable(task)) {
2044                         rcu_read_lock();
2045                         cred = __task_cred(task);
2046                         inode->i_uid = cred->euid;
2047                         inode->i_gid = cred->egid;
2048                         rcu_read_unlock();
2049                 } else {
2050                         inode->i_uid = GLOBAL_ROOT_UID;
2051                         inode->i_gid = GLOBAL_ROOT_GID;
2052                 }
2053                 security_task_to_inode(task, inode);
2054                 status = 1;
2055         }
2056
2057 out:
2058         put_task_struct(task);
2059
2060 out_notask:
2061         if (status <= 0)
2062                 d_drop(dentry);
2063
2064         return status;
2065 }
2066
2067 static const struct dentry_operations tid_map_files_dentry_operations = {
2068         .d_revalidate   = map_files_d_revalidate,
2069         .d_delete       = pid_delete_dentry,
2070 };
2071
2072 static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
2073 {
2074         unsigned long vm_start, vm_end;
2075         struct vm_area_struct *vma;
2076         struct task_struct *task;
2077         struct mm_struct *mm;
2078         int rc;
2079
2080         rc = -ENOENT;
2081         task = get_proc_task(dentry->d_inode);
2082         if (!task)
2083                 goto out;
2084
2085         mm = get_task_mm(task);
2086         put_task_struct(task);
2087         if (!mm)
2088                 goto out;
2089
2090         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2091         if (rc)
2092                 goto out_mmput;
2093
2094         down_read(&mm->mmap_sem);
2095         vma = find_exact_vma(mm, vm_start, vm_end);
2096         if (vma && vma->vm_file) {
2097                 *path = vma->vm_file->f_path;
2098                 path_get(path);
2099                 rc = 0;
2100         }
2101         up_read(&mm->mmap_sem);
2102
2103 out_mmput:
2104         mmput(mm);
2105 out:
2106         return rc;
2107 }
2108
2109 struct map_files_info {
2110         struct file     *file;
2111         unsigned long   len;
2112         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
2113 };
2114
2115 static struct dentry *
2116 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
2117                            struct task_struct *task, const void *ptr)
2118 {
2119         const struct file *file = ptr;
2120         struct proc_inode *ei;
2121         struct inode *inode;
2122
2123         if (!file)
2124                 return ERR_PTR(-ENOENT);
2125
2126         inode = proc_pid_make_inode(dir->i_sb, task);
2127         if (!inode)
2128                 return ERR_PTR(-ENOENT);
2129
2130         ei = PROC_I(inode);
2131         ei->op.proc_get_link = proc_map_files_get_link;
2132
2133         inode->i_op = &proc_pid_link_inode_operations;
2134         inode->i_size = 64;
2135         inode->i_mode = S_IFLNK;
2136
2137         if (file->f_mode & FMODE_READ)
2138                 inode->i_mode |= S_IRUSR;
2139         if (file->f_mode & FMODE_WRITE)
2140                 inode->i_mode |= S_IWUSR;
2141
2142         d_set_d_op(dentry, &tid_map_files_dentry_operations);
2143         d_add(dentry, inode);
2144
2145         return NULL;
2146 }
2147
2148 static struct dentry *proc_map_files_lookup(struct inode *dir,
2149                 struct dentry *dentry, struct nameidata *nd)
2150 {
2151         unsigned long vm_start, vm_end;
2152         struct vm_area_struct *vma;
2153         struct task_struct *task;
2154         struct dentry *result;
2155         struct mm_struct *mm;
2156
2157         result = ERR_PTR(-EACCES);
2158         if (!capable(CAP_SYS_ADMIN))
2159                 goto out;
2160
2161         result = ERR_PTR(-ENOENT);
2162         task = get_proc_task(dir);
2163         if (!task)
2164                 goto out;
2165
2166         result = ERR_PTR(-EACCES);
2167         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2168                 goto out_put_task;
2169
2170         result = ERR_PTR(-ENOENT);
2171         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2172                 goto out_put_task;
2173
2174         mm = get_task_mm(task);
2175         if (!mm)
2176                 goto out_put_task;
2177
2178         down_read(&mm->mmap_sem);
2179         vma = find_exact_vma(mm, vm_start, vm_end);
2180         if (!vma)
2181                 goto out_no_vma;
2182
2183         result = proc_map_files_instantiate(dir, dentry, task, vma->vm_file);
2184
2185 out_no_vma:
2186         up_read(&mm->mmap_sem);
2187         mmput(mm);
2188 out_put_task:
2189         put_task_struct(task);
2190 out:
2191         return result;
2192 }
2193
2194 static const struct inode_operations proc_map_files_inode_operations = {
2195         .lookup         = proc_map_files_lookup,
2196         .permission     = proc_fd_permission,
2197         .setattr        = proc_setattr,
2198 };
2199
2200 static int
2201 proc_map_files_readdir(struct file *filp, void *dirent, filldir_t filldir)
2202 {
2203         struct dentry *dentry = filp->f_path.dentry;
2204         struct inode *inode = dentry->d_inode;
2205         struct vm_area_struct *vma;
2206         struct task_struct *task;
2207         struct mm_struct *mm;
2208         ino_t ino;
2209         int ret;
2210
2211         ret = -EACCES;
2212         if (!capable(CAP_SYS_ADMIN))
2213                 goto out;
2214
2215         ret = -ENOENT;
2216         task = get_proc_task(inode);
2217         if (!task)
2218                 goto out;
2219
2220         ret = -EACCES;
2221         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2222                 goto out_put_task;
2223
2224         ret = 0;
2225         switch (filp->f_pos) {
2226         case 0:
2227                 ino = inode->i_ino;
2228                 if (filldir(dirent, ".", 1, 0, ino, DT_DIR) < 0)
2229                         goto out_put_task;
2230                 filp->f_pos++;
2231         case 1:
2232                 ino = parent_ino(dentry);
2233                 if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
2234                         goto out_put_task;
2235                 filp->f_pos++;
2236         default:
2237         {
2238                 unsigned long nr_files, pos, i;
2239                 struct flex_array *fa = NULL;
2240                 struct map_files_info info;
2241                 struct map_files_info *p;
2242
2243                 mm = get_task_mm(task);
2244                 if (!mm)
2245                         goto out_put_task;
2246                 down_read(&mm->mmap_sem);
2247
2248                 nr_files = 0;
2249
2250                 /*
2251                  * We need two passes here:
2252                  *
2253                  *  1) Collect vmas of mapped files with mmap_sem taken
2254                  *  2) Release mmap_sem and instantiate entries
2255                  *
2256                  * otherwise we get lockdep complained, since filldir()
2257                  * routine might require mmap_sem taken in might_fault().
2258                  */
2259
2260                 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2261                         if (vma->vm_file && ++pos > filp->f_pos)
2262                                 nr_files++;
2263                 }
2264
2265                 if (nr_files) {
2266                         fa = flex_array_alloc(sizeof(info), nr_files,
2267                                                 GFP_KERNEL);
2268                         if (!fa || flex_array_prealloc(fa, 0, nr_files,
2269                                                         GFP_KERNEL)) {
2270                                 ret = -ENOMEM;
2271                                 if (fa)
2272                                         flex_array_free(fa);
2273                                 up_read(&mm->mmap_sem);
2274                                 mmput(mm);
2275                                 goto out_put_task;
2276                         }
2277                         for (i = 0, vma = mm->mmap, pos = 2; vma;
2278                                         vma = vma->vm_next) {
2279                                 if (!vma->vm_file)
2280                                         continue;
2281                                 if (++pos <= filp->f_pos)
2282                                         continue;
2283
2284                                 get_file(vma->vm_file);
2285                                 info.file = vma->vm_file;
2286                                 info.len = snprintf(info.name,
2287                                                 sizeof(info.name), "%lx-%lx",
2288                                                 vma->vm_start, vma->vm_end);
2289                                 if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2290                                         BUG();
2291                         }
2292                 }
2293                 up_read(&mm->mmap_sem);
2294
2295                 for (i = 0; i < nr_files; i++) {
2296                         p = flex_array_get(fa, i);
2297                         ret = proc_fill_cache(filp, dirent, filldir,
2298                                               p->name, p->len,
2299                                               proc_map_files_instantiate,
2300                                               task, p->file);
2301                         if (ret)
2302                                 break;
2303                         filp->f_pos++;
2304                         fput(p->file);
2305                 }
2306                 for (; i < nr_files; i++) {
2307                         /*
2308                          * In case of error don't forget
2309                          * to put rest of file refs.
2310                          */
2311                         p = flex_array_get(fa, i);
2312                         fput(p->file);
2313                 }
2314                 if (fa)
2315                         flex_array_free(fa);
2316                 mmput(mm);
2317         }
2318         }
2319
2320 out_put_task:
2321         put_task_struct(task);
2322 out:
2323         return ret;
2324 }
2325
2326 static const struct file_operations proc_map_files_operations = {
2327         .read           = generic_read_dir,
2328         .readdir        = proc_map_files_readdir,
2329         .llseek         = default_llseek,
2330 };
2331
2332 #endif /* CONFIG_CHECKPOINT_RESTORE */
2333
2334 /*
2335  * /proc/pid/fd needs a special permission handler so that a process can still
2336  * access /proc/self/fd after it has executed a setuid().
2337  */
2338 static int proc_fd_permission(struct inode *inode, int mask)
2339 {
2340         int rv = generic_permission(inode, mask);
2341         if (rv == 0)
2342                 return 0;
2343         if (task_pid(current) == proc_pid(inode))
2344                 rv = 0;
2345         return rv;
2346 }
2347
2348 /*
2349  * proc directories can do almost nothing..
2350  */
2351 static const struct inode_operations proc_fd_inode_operations = {
2352         .lookup         = proc_lookupfd,
2353         .permission     = proc_fd_permission,
2354         .setattr        = proc_setattr,
2355 };
2356
2357 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
2358         struct dentry *dentry, struct task_struct *task, const void *ptr)
2359 {
2360         unsigned fd = *(unsigned *)ptr;
2361         struct inode *inode;
2362         struct proc_inode *ei;
2363         struct dentry *error = ERR_PTR(-ENOENT);
2364
2365         inode = proc_pid_make_inode(dir->i_sb, task);
2366         if (!inode)
2367                 goto out;
2368         ei = PROC_I(inode);
2369         ei->fd = fd;
2370         inode->i_mode = S_IFREG | S_IRUSR;
2371         inode->i_fop = &proc_fdinfo_file_operations;
2372         d_set_d_op(dentry, &tid_fd_dentry_operations);
2373         d_add(dentry, inode);
2374         /* Close the race of the process dying before we return the dentry */
2375         if (tid_fd_revalidate(dentry, NULL))
2376                 error = NULL;
2377
2378  out:
2379         return error;
2380 }
2381
2382 static struct dentry *proc_lookupfdinfo(struct inode *dir,
2383                                         struct dentry *dentry,
2384                                         struct nameidata *nd)
2385 {
2386         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
2387 }
2388
2389 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
2390 {
2391         return proc_readfd_common(filp, dirent, filldir,
2392                                   proc_fdinfo_instantiate);
2393 }
2394
2395 static const struct file_operations proc_fdinfo_operations = {
2396         .read           = generic_read_dir,
2397         .readdir        = proc_readfdinfo,
2398         .llseek         = default_llseek,
2399 };
2400
2401 /*
2402  * proc directories can do almost nothing..
2403  */
2404 static const struct inode_operations proc_fdinfo_inode_operations = {
2405         .lookup         = proc_lookupfdinfo,
2406         .setattr        = proc_setattr,
2407 };
2408
2409
2410 static struct dentry *proc_pident_instantiate(struct inode *dir,
2411         struct dentry *dentry, struct task_struct *task, const void *ptr)
2412 {
2413         const struct pid_entry *p = ptr;
2414         struct inode *inode;
2415         struct proc_inode *ei;
2416         struct dentry *error = ERR_PTR(-ENOENT);
2417
2418         inode = proc_pid_make_inode(dir->i_sb, task);
2419         if (!inode)
2420                 goto out;
2421
2422         ei = PROC_I(inode);
2423         inode->i_mode = p->mode;
2424         if (S_ISDIR(inode->i_mode))
2425                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2426         if (p->iop)
2427                 inode->i_op = p->iop;
2428         if (p->fop)
2429                 inode->i_fop = p->fop;
2430         ei->op = p->op;
2431         d_set_d_op(dentry, &pid_dentry_operations);
2432         d_add(dentry, inode);
2433         /* Close the race of the process dying before we return the dentry */
2434         if (pid_revalidate(dentry, NULL))
2435                 error = NULL;
2436 out:
2437         return error;
2438 }
2439
2440 static struct dentry *proc_pident_lookup(struct inode *dir,
2441                                          struct dentry *dentry,
2442                                          const struct pid_entry *ents,
2443                                          unsigned int nents)
2444 {
2445         struct dentry *error;
2446         struct task_struct *task = get_proc_task(dir);
2447         const struct pid_entry *p, *last;
2448
2449         error = ERR_PTR(-ENOENT);
2450
2451         if (!task)
2452                 goto out_no_task;
2453
2454         /*
2455          * Yes, it does not scale. And it should not. Don't add
2456          * new entries into /proc/<tgid>/ without very good reasons.
2457          */
2458         last = &ents[nents - 1];
2459         for (p = ents; p <= last; p++) {
2460                 if (p->len != dentry->d_name.len)
2461                         continue;
2462                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2463                         break;
2464         }
2465         if (p > last)
2466                 goto out;
2467
2468         error = proc_pident_instantiate(dir, dentry, task, p);
2469 out:
2470         put_task_struct(task);
2471 out_no_task:
2472         return error;
2473 }
2474
2475 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2476         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2477 {
2478         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2479                                 proc_pident_instantiate, task, p);
2480 }
2481
2482 static int proc_pident_readdir(struct file *filp,
2483                 void *dirent, filldir_t filldir,
2484                 const struct pid_entry *ents, unsigned int nents)
2485 {
2486         int i;
2487         struct dentry *dentry = filp->f_path.dentry;
2488         struct inode *inode = dentry->d_inode;
2489         struct task_struct *task = get_proc_task(inode);
2490         const struct pid_entry *p, *last;
2491         ino_t ino;
2492         int ret;
2493
2494         ret = -ENOENT;
2495         if (!task)
2496                 goto out_no_task;
2497
2498         ret = 0;
2499         i = filp->f_pos;
2500         switch (i) {
2501         case 0:
2502                 ino = inode->i_ino;
2503                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2504                         goto out;
2505                 i++;
2506                 filp->f_pos++;
2507                 /* fall through */
2508         case 1:
2509                 ino = parent_ino(dentry);
2510                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2511                         goto out;
2512                 i++;
2513                 filp->f_pos++;
2514                 /* fall through */
2515         default:
2516                 i -= 2;
2517                 if (i >= nents) {
2518                         ret = 1;
2519                         goto out;
2520                 }
2521                 p = ents + i;
2522                 last = &ents[nents - 1];
2523                 while (p <= last) {
2524                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2525                                 goto out;
2526                         filp->f_pos++;
2527                         p++;
2528                 }
2529         }
2530
2531         ret = 1;
2532 out:
2533         put_task_struct(task);
2534 out_no_task:
2535         return ret;
2536 }
2537
2538 #ifdef CONFIG_SECURITY
2539 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2540                                   size_t count, loff_t *ppos)
2541 {
2542         struct inode * inode = file->f_path.dentry->d_inode;
2543         char *p = NULL;
2544         ssize_t length;
2545         struct task_struct *task = get_proc_task(inode);
2546
2547         if (!task)
2548                 return -ESRCH;
2549
2550         length = security_getprocattr(task,
2551                                       (char*)file->f_path.dentry->d_name.name,
2552                                       &p);
2553         put_task_struct(task);
2554         if (length > 0)
2555                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2556         kfree(p);
2557         return length;
2558 }
2559
2560 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2561                                    size_t count, loff_t *ppos)
2562 {
2563         struct inode * inode = file->f_path.dentry->d_inode;
2564         char *page;
2565         ssize_t length;
2566         struct task_struct *task = get_proc_task(inode);
2567
2568         length = -ESRCH;
2569         if (!task)
2570                 goto out_no_task;
2571         if (count > PAGE_SIZE)
2572                 count = PAGE_SIZE;
2573
2574         /* No partial writes. */
2575         length = -EINVAL;
2576         if (*ppos != 0)
2577                 goto out;
2578
2579         length = -ENOMEM;
2580         page = (char*)__get_free_page(GFP_TEMPORARY);
2581         if (!page)
2582                 goto out;
2583
2584         length = -EFAULT;
2585         if (copy_from_user(page, buf, count))
2586                 goto out_free;
2587
2588         /* Guard against adverse ptrace interaction */
2589         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2590         if (length < 0)
2591                 goto out_free;
2592
2593         length = security_setprocattr(task,
2594                                       (char*)file->f_path.dentry->d_name.name,
2595                                       (void*)page, count);
2596         mutex_unlock(&task->signal->cred_guard_mutex);
2597 out_free:
2598         free_page((unsigned long) page);
2599 out:
2600         put_task_struct(task);
2601 out_no_task:
2602         return length;
2603 }
2604
2605 static const struct file_operations proc_pid_attr_operations = {
2606         .read           = proc_pid_attr_read,
2607         .write          = proc_pid_attr_write,
2608         .llseek         = generic_file_llseek,
2609 };
2610
2611 static const struct pid_entry attr_dir_stuff[] = {
2612         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2613         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2614         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2615         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2616         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2617         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2618 };
2619
2620 static int proc_attr_dir_readdir(struct file * filp,
2621                              void * dirent, filldir_t filldir)
2622 {
2623         return proc_pident_readdir(filp,dirent,filldir,
2624                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2625 }
2626
2627 static const struct file_operations proc_attr_dir_operations = {
2628         .read           = generic_read_dir,
2629         .readdir        = proc_attr_dir_readdir,
2630         .llseek         = default_llseek,
2631 };
2632
2633 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2634                                 struct dentry *dentry, struct nameidata *nd)
2635 {
2636         return proc_pident_lookup(dir, dentry,
2637                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2638 }
2639
2640 static const struct inode_operations proc_attr_dir_inode_operations = {
2641         .lookup         = proc_attr_dir_lookup,
2642         .getattr        = pid_getattr,
2643         .setattr        = proc_setattr,
2644 };
2645
2646 #endif
2647
2648 #ifdef CONFIG_ELF_CORE
2649 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2650                                          size_t count, loff_t *ppos)
2651 {
2652         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2653         struct mm_struct *mm;
2654         char buffer[PROC_NUMBUF];
2655         size_t len;
2656         int ret;
2657
2658         if (!task)
2659                 return -ESRCH;
2660
2661         ret = 0;
2662         mm = get_task_mm(task);
2663         if (mm) {
2664                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2665                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2666                                 MMF_DUMP_FILTER_SHIFT));
2667                 mmput(mm);
2668                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2669         }
2670
2671         put_task_struct(task);
2672
2673         return ret;
2674 }
2675
2676 static ssize_t proc_coredump_filter_write(struct file *file,
2677                                           const char __user *buf,
2678                                           size_t count,
2679                                           loff_t *ppos)
2680 {
2681         struct task_struct *task;
2682         struct mm_struct *mm;
2683         char buffer[PROC_NUMBUF], *end;
2684         unsigned int val;
2685         int ret;
2686         int i;
2687         unsigned long mask;
2688
2689         ret = -EFAULT;
2690         memset(buffer, 0, sizeof(buffer));
2691         if (count > sizeof(buffer) - 1)
2692                 count = sizeof(buffer) - 1;
2693         if (copy_from_user(buffer, buf, count))
2694                 goto out_no_task;
2695
2696         ret = -EINVAL;
2697         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2698         if (*end == '\n')
2699                 end++;
2700         if (end - buffer == 0)
2701                 goto out_no_task;
2702
2703         ret = -ESRCH;
2704         task = get_proc_task(file->f_dentry->d_inode);
2705         if (!task)
2706                 goto out_no_task;
2707
2708         ret = end - buffer;
2709         mm = get_task_mm(task);
2710         if (!mm)
2711                 goto out_no_mm;
2712
2713         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2714                 if (val & mask)
2715                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2716                 else
2717                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2718         }
2719
2720         mmput(mm);
2721  out_no_mm:
2722         put_task_struct(task);
2723  out_no_task:
2724         return ret;
2725 }
2726
2727 static const struct file_operations proc_coredump_filter_operations = {
2728         .read           = proc_coredump_filter_read,
2729         .write          = proc_coredump_filter_write,
2730         .llseek         = generic_file_llseek,
2731 };
2732 #endif
2733
2734 /*
2735  * /proc/self:
2736  */
2737 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2738                               int buflen)
2739 {
2740         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2741         pid_t tgid = task_tgid_nr_ns(current, ns);
2742         char tmp[PROC_NUMBUF];
2743         if (!tgid)
2744                 return -ENOENT;
2745         sprintf(tmp, "%d", tgid);
2746         return vfs_readlink(dentry,buffer,buflen,tmp);
2747 }
2748
2749 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2750 {
2751         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2752         pid_t tgid = task_tgid_nr_ns(current, ns);
2753         char *name = ERR_PTR(-ENOENT);
2754         if (tgid) {
2755                 name = __getname();
2756                 if (!name)
2757                         name = ERR_PTR(-ENOMEM);
2758                 else
2759                         sprintf(name, "%d", tgid);
2760         }
2761         nd_set_link(nd, name);
2762         return NULL;
2763 }
2764
2765 static void proc_self_put_link(struct dentry *dentry, struct nameidata *nd,
2766                                 void *cookie)
2767 {
2768         char *s = nd_get_link(nd);
2769         if (!IS_ERR(s))
2770                 __putname(s);
2771 }
2772
2773 static const struct inode_operations proc_self_inode_operations = {
2774         .readlink       = proc_self_readlink,
2775         .follow_link    = proc_self_follow_link,
2776         .put_link       = proc_self_put_link,
2777 };
2778
2779 /*
2780  * proc base
2781  *
2782  * These are the directory entries in the root directory of /proc
2783  * that properly belong to the /proc filesystem, as they describe
2784  * describe something that is process related.
2785  */
2786 static const struct pid_entry proc_base_stuff[] = {
2787         NOD("self", S_IFLNK|S_IRWXUGO,
2788                 &proc_self_inode_operations, NULL, {}),
2789 };
2790
2791 static struct dentry *proc_base_instantiate(struct inode *dir,
2792         struct dentry *dentry, struct task_struct *task, const void *ptr)
2793 {
2794         const struct pid_entry *p = ptr;
2795         struct inode *inode;
2796         struct proc_inode *ei;
2797         struct dentry *error;
2798
2799         /* Allocate the inode */
2800         error = ERR_PTR(-ENOMEM);
2801         inode = new_inode(dir->i_sb);
2802         if (!inode)
2803                 goto out;
2804
2805         /* Initialize the inode */
2806         ei = PROC_I(inode);
2807         inode->i_ino = get_next_ino();
2808         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2809
2810         /*
2811          * grab the reference to the task.
2812          */
2813         ei->pid = get_task_pid(task, PIDTYPE_PID);
2814         if (!ei->pid)
2815                 goto out_iput;
2816
2817         inode->i_mode = p->mode;
2818         if (S_ISDIR(inode->i_mode))
2819                 set_nlink(inode, 2);
2820         if (S_ISLNK(inode->i_mode))
2821                 inode->i_size = 64;
2822         if (p->iop)
2823                 inode->i_op = p->iop;
2824         if (p->fop)
2825                 inode->i_fop = p->fop;
2826         ei->op = p->op;
2827         d_add(dentry, inode);
2828         error = NULL;
2829 out:
2830         return error;
2831 out_iput:
2832         iput(inode);
2833         goto out;
2834 }
2835
2836 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2837 {
2838         struct dentry *error;
2839         struct task_struct *task = get_proc_task(dir);
2840         const struct pid_entry *p, *last;
2841
2842         error = ERR_PTR(-ENOENT);
2843
2844         if (!task)
2845                 goto out_no_task;
2846
2847         /* Lookup the directory entry */
2848         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2849         for (p = proc_base_stuff; p <= last; p++) {
2850                 if (p->len != dentry->d_name.len)
2851                         continue;
2852                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2853                         break;
2854         }
2855         if (p > last)
2856                 goto out;
2857
2858         error = proc_base_instantiate(dir, dentry, task, p);
2859
2860 out:
2861         put_task_struct(task);
2862 out_no_task:
2863         return error;
2864 }
2865
2866 static int proc_base_fill_cache(struct file *filp, void *dirent,
2867         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2868 {
2869         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2870                                 proc_base_instantiate, task, p);
2871 }
2872
2873 #ifdef CONFIG_TASK_IO_ACCOUNTING
2874 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2875 {
2876         struct task_io_accounting acct = task->ioac;
2877         unsigned long flags;
2878         int result;
2879
2880         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2881         if (result)
2882                 return result;
2883
2884         if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2885                 result = -EACCES;
2886                 goto out_unlock;
2887         }
2888
2889         if (whole && lock_task_sighand(task, &flags)) {
2890                 struct task_struct *t = task;
2891
2892                 task_io_accounting_add(&acct, &task->signal->ioac);
2893                 while_each_thread(task, t)
2894                         task_io_accounting_add(&acct, &t->ioac);
2895
2896                 unlock_task_sighand(task, &flags);
2897         }
2898         result = sprintf(buffer,
2899                         "rchar: %llu\n"
2900                         "wchar: %llu\n"
2901                         "syscr: %llu\n"
2902                         "syscw: %llu\n"
2903                         "read_bytes: %llu\n"
2904                         "write_bytes: %llu\n"
2905                         "cancelled_write_bytes: %llu\n",
2906                         (unsigned long long)acct.rchar,
2907                         (unsigned long long)acct.wchar,
2908                         (unsigned long long)acct.syscr,
2909                         (unsigned long long)acct.syscw,
2910                         (unsigned long long)acct.read_bytes,
2911                         (unsigned long long)acct.write_bytes,
2912                         (unsigned long long)acct.cancelled_write_bytes);
2913 out_unlock:
2914         mutex_unlock(&task->signal->cred_guard_mutex);
2915         return result;
2916 }
2917
2918 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2919 {
2920         return do_io_accounting(task, buffer, 0);
2921 }
2922
2923 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2924 {
2925         return do_io_accounting(task, buffer, 1);
2926 }
2927 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2928
2929 #ifdef CONFIG_USER_NS
2930 static int proc_id_map_open(struct inode *inode, struct file *file,
2931         struct seq_operations *seq_ops)
2932 {
2933         struct user_namespace *ns = NULL;
2934         struct task_struct *task;
2935         struct seq_file *seq;
2936         int ret = -EINVAL;
2937
2938         task = get_proc_task(inode);
2939         if (task) {
2940                 rcu_read_lock();
2941                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2942                 rcu_read_unlock();
2943                 put_task_struct(task);
2944         }
2945         if (!ns)
2946                 goto err;
2947
2948         ret = seq_open(file, seq_ops);
2949         if (ret)
2950                 goto err_put_ns;
2951
2952         seq = file->private_data;
2953         seq->private = ns;
2954
2955         return 0;
2956 err_put_ns:
2957         put_user_ns(ns);
2958 err:
2959         return ret;
2960 }
2961
2962 static int proc_id_map_release(struct inode *inode, struct file *file)
2963 {
2964         struct seq_file *seq = file->private_data;
2965         struct user_namespace *ns = seq->private;
2966         put_user_ns(ns);
2967         return seq_release(inode, file);
2968 }
2969
2970 static int proc_uid_map_open(struct inode *inode, struct file *file)
2971 {
2972         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2973 }
2974
2975 static int proc_gid_map_open(struct inode *inode, struct file *file)
2976 {
2977         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2978 }
2979
2980 static const struct file_operations proc_uid_map_operations = {
2981         .open           = proc_uid_map_open,
2982         .write          = proc_uid_map_write,
2983         .read           = seq_read,
2984         .llseek         = seq_lseek,
2985         .release        = proc_id_map_release,
2986 };
2987
2988 static const struct file_operations proc_gid_map_operations = {
2989         .open           = proc_gid_map_open,
2990         .write          = proc_gid_map_write,
2991         .read           = seq_read,
2992         .llseek         = seq_lseek,
2993         .release        = proc_id_map_release,
2994 };
2995 #endif /* CONFIG_USER_NS */
2996
2997 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2998                                 struct pid *pid, struct task_struct *task)
2999 {
3000         int err = lock_trace(task);
3001         if (!err) {
3002                 seq_printf(m, "%08x\n", task->personality);
3003                 unlock_trace(task);
3004         }
3005         return err;
3006 }
3007
3008 /*
3009  * Thread groups
3010  */
3011 static const struct file_operations proc_task_operations;
3012 static const struct inode_operations proc_task_inode_operations;
3013
3014 static const struct pid_entry tgid_base_stuff[] = {
3015         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
3016         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3017 #ifdef CONFIG_CHECKPOINT_RESTORE
3018         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
3019 #endif
3020         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3021         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3022 #ifdef CONFIG_NET
3023         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3024 #endif
3025         REG("environ",    S_IRUSR, proc_environ_operations),
3026         INF("auxv",       S_IRUSR, proc_pid_auxv),
3027         ONE("status",     S_IRUGO, proc_pid_status),
3028         ONE("personality", S_IRUGO, proc_pid_personality),
3029         INF("limits",     S_IRUGO, proc_pid_limits),
3030 #ifdef CONFIG_SCHED_DEBUG
3031         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3032 #endif
3033 #ifdef CONFIG_SCHED_AUTOGROUP
3034         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
3035 #endif
3036         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3037 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3038         INF("syscall",    S_IRUGO, proc_pid_syscall),
3039 #endif
3040         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
3041         ONE("stat",       S_IRUGO, proc_tgid_stat),
3042         ONE("statm",      S_IRUGO, proc_pid_statm),
3043         REG("maps",       S_IRUGO, proc_pid_maps_operations),
3044 #ifdef CONFIG_NUMA
3045         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
3046 #endif
3047         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
3048         LNK("cwd",        proc_cwd_link),
3049         LNK("root",       proc_root_link),
3050         LNK("exe",        proc_exe_link),
3051         REG("mounts",     S_IRUGO, proc_mounts_operations),
3052         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3053         REG("mountstats", S_IRUSR, proc_mountstats_operations),
3054 #ifdef CONFIG_PROC_PAGE_MONITOR
3055         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3056         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
3057         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
3058 #endif
3059 #ifdef CONFIG_SECURITY
3060         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3061 #endif
3062 #ifdef CONFIG_KALLSYMS
3063         INF("wchan",      S_IRUGO, proc_pid_wchan),
3064 #endif
3065 #ifdef CONFIG_STACKTRACE
3066         ONE("stack",      S_IRUGO, proc_pid_stack),
3067 #endif
3068 #ifdef CONFIG_SCHEDSTATS
3069         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
3070 #endif
3071 #ifdef CONFIG_LATENCYTOP
3072         REG("latency",  S_IRUGO, proc_lstats_operations),
3073 #endif
3074 #ifdef CONFIG_PROC_PID_CPUSET
3075         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
3076 #endif
3077 #ifdef CONFIG_CGROUPS
3078         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3079 #endif
3080         INF("oom_score",  S_IRUGO, proc_oom_score),
3081         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3082         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3083 #ifdef CONFIG_AUDITSYSCALL
3084         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
3085         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3086 #endif
3087 #ifdef CONFIG_FAULT_INJECTION
3088         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3089 #endif
3090 #ifdef CONFIG_ELF_CORE
3091         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3092 #endif
3093 #ifdef CONFIG_TASK_IO_ACCOUNTING
3094         INF("io",       S_IRUSR, proc_tgid_io_accounting),
3095 #endif
3096 #ifdef CONFIG_HARDWALL
3097         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
3098 #endif
3099 #ifdef CONFIG_USER_NS
3100         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3101         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3102 #endif
3103 };
3104
3105 static int proc_tgid_base_readdir(struct file * filp,
3106                              void * dirent, filldir_t filldir)
3107 {
3108         return proc_pident_readdir(filp,dirent,filldir,
3109                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
3110 }
3111
3112 static const struct file_operations proc_tgid_base_operations = {
3113         .read           = generic_read_dir,
3114         .readdir        = proc_tgid_base_readdir,
3115         .llseek         = default_llseek,
3116 };
3117
3118 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3119         return proc_pident_lookup(dir, dentry,
3120                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3121 }
3122
3123 static const struct inode_operations proc_tgid_base_inode_operations = {
3124         .lookup         = proc_tgid_base_lookup,
3125         .getattr        = pid_getattr,
3126         .setattr        = proc_setattr,
3127         .permission     = proc_pid_permission,
3128 };
3129
3130 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
3131 {
3132         struct dentry *dentry, *leader, *dir;
3133         char buf[PROC_NUMBUF];
3134         struct qstr name;
3135
3136         name.name = buf;
3137         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3138         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
3139         if (dentry) {
3140                 shrink_dcache_parent(dentry);
3141                 d_drop(dentry);
3142                 dput(dentry);
3143         }
3144
3145         name.name = buf;
3146         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
3147         leader = d_hash_and_lookup(mnt->mnt_root, &name);
3148         if (!leader)
3149                 goto out;
3150
3151         name.name = "task";
3152         name.len = strlen(name.name);
3153         dir = d_hash_and_lookup(leader, &name);
3154         if (!dir)
3155                 goto out_put_leader;
3156
3157         name.name = buf;
3158         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3159         dentry = d_hash_and_lookup(dir, &name);
3160         if (dentry) {
3161                 shrink_dcache_parent(dentry);
3162                 d_drop(dentry);
3163                 dput(dentry);
3164         }
3165
3166         dput(dir);
3167 out_put_leader:
3168         dput(leader);
3169 out:
3170         return;
3171 }
3172
3173 /**
3174  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
3175  * @task: task that should be flushed.
3176  *
3177  * When flushing dentries from proc, one needs to flush them from global
3178  * proc (proc_mnt) and from all the namespaces' procs this task was seen
3179  * in. This call is supposed to do all of this job.
3180  *
3181  * Looks in the dcache for
3182  * /proc/@pid
3183  * /proc/@tgid/task/@pid
3184  * if either directory is present flushes it and all of it'ts children
3185  * from the dcache.
3186  *
3187  * It is safe and reasonable to cache /proc entries for a task until
3188  * that task exits.  After that they just clog up the dcache with
3189  * useless entries, possibly causing useful dcache entries to be
3190  * flushed instead.  This routine is proved to flush those useless
3191  * dcache entries at process exit time.
3192  *
3193  * NOTE: This routine is just an optimization so it does not guarantee
3194  *       that no dcache entries will exist at process exit time it
3195  *       just makes it very unlikely that any will persist.
3196  */
3197
3198 void proc_flush_task(struct task_struct *task)
3199 {
3200         int i;
3201         struct pid *pid, *tgid;
3202         struct upid *upid;
3203
3204         pid = task_pid(task);
3205         tgid = task_tgid(task);
3206
3207         for (i = 0; i <= pid->level; i++) {
3208                 upid = &pid->numbers[i];
3209                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3210                                         tgid->numbers[i].nr);
3211         }
3212
3213         upid = &pid->numbers[pid->level];
3214         if (upid->nr == 1)
3215                 pid_ns_release_proc(upid->ns);
3216 }
3217
3218 static struct dentry *proc_pid_instantiate(struct inode *dir,
3219                                            struct dentry * dentry,
3220                                            struct task_struct *task, const void *ptr)
3221 {
3222         struct dentry *error = ERR_PTR(-ENOENT);
3223         struct inode *inode;
3224
3225         inode = proc_pid_make_inode(dir->i_sb, task);
3226         if (!inode)
3227                 goto out;
3228
3229         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3230         inode->i_op = &proc_tgid_base_inode_operations;
3231         inode->i_fop = &proc_tgid_base_operations;
3232         inode->i_flags|=S_IMMUTABLE;
3233
3234         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
3235                                                   ARRAY_SIZE(tgid_base_stuff)));
3236
3237         d_set_d_op(dentry, &pid_dentry_operations);
3238
3239         d_add(dentry, inode);
3240         /* Close the race of the process dying before we return the dentry */
3241         if (pid_revalidate(dentry, NULL))
3242                 error = NULL;
3243 out:
3244         return error;
3245 }
3246
3247 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3248 {
3249         struct dentry *result;
3250         struct task_struct *task;
3251         unsigned tgid;
3252         struct pid_namespace *ns;
3253
3254         result = proc_base_lookup(dir, dentry);
3255         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
3256                 goto out;
3257
3258         tgid = name_to_int(dentry);
3259         if (tgid == ~0U)
3260                 goto out;
3261
3262         ns = dentry->d_sb->s_fs_info;
3263         rcu_read_lock();
3264         task = find_task_by_pid_ns(tgid, ns);
3265         if (task)
3266                 get_task_struct(task);
3267         rcu_read_unlock();
3268         if (!task)
3269                 goto out;
3270
3271         result = proc_pid_instantiate(dir, dentry, task, NULL);
3272         put_task_struct(task);
3273 out:
3274         return result;
3275 }
3276
3277 /*
3278  * Find the first task with tgid >= tgid
3279  *
3280  */
3281 struct tgid_iter {
3282         unsigned int tgid;
3283         struct task_struct *task;
3284 };
3285 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3286 {
3287         struct pid *pid;
3288
3289         if (iter.task)
3290                 put_task_struct(iter.task);
3291         rcu_read_lock();
3292 retry:
3293         iter.task = NULL;
3294         pid = find_ge_pid(iter.tgid, ns);
3295         if (pid) {
3296                 iter.tgid = pid_nr_ns(pid, ns);
3297                 iter.task = pid_task(pid, PIDTYPE_PID);
3298                 /* What we to know is if the pid we have find is the
3299                  * pid of a thread_group_leader.  Testing for task
3300                  * being a thread_group_leader is the obvious thing
3301                  * todo but there is a window when it fails, due to
3302                  * the pid transfer logic in de_thread.
3303                  *
3304                  * So we perform the straight forward test of seeing
3305                  * if the pid we have found is the pid of a thread
3306                  * group leader, and don't worry if the task we have
3307                  * found doesn't happen to be a thread group leader.
3308                  * As we don't care in the case of readdir.
3309                  */
3310                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3311                         iter.tgid += 1;
3312                         goto retry;
3313                 }
3314                 get_task_struct(iter.task);
3315         }
3316         rcu_read_unlock();
3317         return iter;
3318 }
3319
3320 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
3321
3322 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3323         struct tgid_iter iter)
3324 {
3325         char name[PROC_NUMBUF];
3326         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
3327         return proc_fill_cache(filp, dirent, filldir, name, len,
3328                                 proc_pid_instantiate, iter.task, NULL);
3329 }
3330
3331 static int fake_filldir(void *buf, const char *name, int namelen,
3332                         loff_t offset, u64 ino, unsigned d_type)
3333 {
3334         return 0;
3335 }
3336
3337 /* for the /proc/ directory itself, after non-process stuff has been done */
3338 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
3339 {
3340         unsigned int nr;
3341         struct task_struct *reaper;
3342         struct tgid_iter iter;
3343         struct pid_namespace *ns;
3344         filldir_t __filldir;
3345
3346         if (filp->f_pos >= PID_MAX_LIMIT + TGID_OFFSET)
3347                 goto out_no_task;
3348         nr = filp->f_pos - FIRST_PROCESS_ENTRY;
3349
3350         reaper = get_proc_task(filp->f_path.dentry->d_inode);
3351         if (!reaper)
3352                 goto out_no_task;
3353
3354         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
3355                 const struct pid_entry *p = &proc_base_stuff[nr];
3356                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
3357                         goto out;
3358         }
3359
3360         ns = filp->f_dentry->d_sb->s_fs_info;
3361         iter.task = NULL;
3362         iter.tgid = filp->f_pos - TGID_OFFSET;
3363         for (iter = next_tgid(ns, iter);
3364              iter.task;
3365              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3366                 if (has_pid_permissions(ns, iter.task, 2))
3367                         __filldir = filldir;
3368                 else
3369                         __filldir = fake_filldir;
3370
3371                 filp->f_pos = iter.tgid + TGID_OFFSET;
3372                 if (proc_pid_fill_cache(filp, dirent, __filldir, iter) < 0) {
3373                         put_task_struct(iter.task);
3374                         goto out;
3375                 }
3376         }
3377         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
3378 out:
3379         put_task_struct(reaper);
3380 out_no_task:
3381         return 0;
3382 }
3383
3384 /*
3385  * Tasks
3386  */
3387 static const struct pid_entry tid_base_stuff[] = {
3388         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3389         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3390         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3391         REG("environ",   S_IRUSR, proc_environ_operations),
3392         INF("auxv",      S_IRUSR, proc_pid_auxv),
3393         ONE("status",    S_IRUGO, proc_pid_status),
3394         ONE("personality", S_IRUGO, proc_pid_personality),
3395         INF("limits",    S_IRUGO, proc_pid_limits),
3396 #ifdef CONFIG_SCHED_DEBUG
3397         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3398 #endif
3399         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3400 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3401         INF("syscall",   S_IRUGO, proc_pid_syscall),
3402 #endif
3403         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
3404         ONE("stat",      S_IRUGO, proc_tid_stat),
3405         ONE("statm",     S_IRUGO, proc_pid_statm),
3406         REG("maps",      S_IRUGO, proc_tid_maps_operations),
3407 #ifdef CONFIG_NUMA
3408         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3409 #endif
3410         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3411         LNK("cwd",       proc_cwd_link),
3412         LNK("root",      proc_root_link),
3413         LNK("exe",       proc_exe_link),
3414         REG("mounts",    S_IRUGO, proc_mounts_operations),
3415         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3416 #ifdef CONFIG_PROC_PAGE_MONITOR
3417         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3418         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3419         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
3420 #endif
3421 #ifdef CONFIG_SECURITY
3422         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3423 #endif
3424 #ifdef CONFIG_KALLSYMS
3425         INF("wchan",     S_IRUGO, proc_pid_wchan),
3426 #endif
3427 #ifdef CONFIG_STACKTRACE
3428         ONE("stack",      S_IRUGO, proc_pid_stack),
3429 #endif
3430 #ifdef CONFIG_SCHEDSTATS
3431         INF("schedstat", S_IRUGO, proc_pid_schedstat),
3432 #endif
3433 #ifdef CONFIG_LATENCYTOP
3434         REG("latency",  S_IRUGO, proc_lstats_operations),
3435 #endif
3436 #ifdef CONFIG_PROC_PID_CPUSET
3437         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
3438 #endif
3439 #ifdef CONFIG_CGROUPS
3440         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3441 #endif
3442         INF("oom_score", S_IRUGO, proc_oom_score),
3443         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3444         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3445 #ifdef CONFIG_AUDITSYSCALL
3446         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3447         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3448 #endif
3449 #ifdef CONFIG_FAULT_INJECTION
3450         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3451 #endif
3452 #ifdef CONFIG_TASK_IO_ACCOUNTING
3453         INF("io",       S_IRUSR, proc_tid_io_accounting),
3454 #endif
3455 #ifdef CONFIG_HARDWALL
3456         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
3457 #endif
3458 #ifdef CONFIG_USER_NS
3459         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3460         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3461 #endif
3462 };
3463
3464 static int proc_tid_base_readdir(struct file * filp,
3465                              void * dirent, filldir_t filldir)
3466 {
3467         return proc_pident_readdir(filp,dirent,filldir,
3468                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
3469 }
3470
3471 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3472         return proc_pident_lookup(dir, dentry,
3473                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3474 }
3475
3476 static const struct file_operations proc_tid_base_operations = {
3477         .read           = generic_read_dir,
3478         .readdir        = proc_tid_base_readdir,
3479         .llseek         = default_llseek,
3480 };
3481
3482 static const struct inode_operations proc_tid_base_inode_operations = {
3483         .lookup         = proc_tid_base_lookup,
3484         .getattr        = pid_getattr,
3485         .setattr        = proc_setattr,
3486 };
3487
3488 static struct dentry *proc_task_instantiate(struct inode *dir,
3489         struct dentry *dentry, struct task_struct *task, const void *ptr)
3490 {
3491         struct dentry *error = ERR_PTR(-ENOENT);
3492         struct inode *inode;
3493         inode = proc_pid_make_inode(dir->i_sb, task);
3494
3495         if (!inode)
3496                 goto out;
3497         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3498         inode->i_op = &proc_tid_base_inode_operations;
3499         inode->i_fop = &proc_tid_base_operations;
3500         inode->i_flags|=S_IMMUTABLE;
3501
3502         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3503                                                   ARRAY_SIZE(tid_base_stuff)));
3504
3505         d_set_d_op(dentry, &pid_dentry_operations);
3506
3507         d_add(dentry, inode);
3508         /* Close the race of the process dying before we return the dentry */
3509         if (pid_revalidate(dentry, NULL))
3510                 error = NULL;
3511 out:
3512         return error;
3513 }
3514
3515 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3516 {
3517         struct dentry *result = ERR_PTR(-ENOENT);
3518         struct task_struct *task;
3519         struct task_struct *leader = get_proc_task(dir);
3520         unsigned tid;
3521         struct pid_namespace *ns;
3522
3523         if (!leader)
3524                 goto out_no_task;
3525
3526         tid = name_to_int(dentry);
3527         if (tid == ~0U)
3528                 goto out;
3529
3530         ns = dentry->d_sb->s_fs_info;
3531         rcu_read_lock();
3532         task = find_task_by_pid_ns(tid, ns);
3533         if (task)
3534                 get_task_struct(task);
3535         rcu_read_unlock();
3536         if (!task)
3537                 goto out;
3538         if (!same_thread_group(leader, task))
3539                 goto out_drop_task;
3540
3541         result = proc_task_instantiate(dir, dentry, task, NULL);
3542 out_drop_task:
3543         put_task_struct(task);
3544 out:
3545         put_task_struct(leader);
3546 out_no_task:
3547         return result;
3548 }
3549
3550 /*
3551  * Find the first tid of a thread group to return to user space.
3552  *
3553  * Usually this is just the thread group leader, but if the users
3554  * buffer was too small or there was a seek into the middle of the
3555  * directory we have more work todo.
3556  *
3557  * In the case of a short read we start with find_task_by_pid.
3558  *
3559  * In the case of a seek we start with the leader and walk nr
3560  * threads past it.
3561  */
3562 static struct task_struct *first_tid(struct task_struct *leader,
3563                 int tid, int nr, struct pid_namespace *ns)
3564 {
3565         struct task_struct *pos;
3566
3567         rcu_read_lock();
3568         /* Attempt to start with the pid of a thread */
3569         if (tid && (nr > 0)) {
3570                 pos = find_task_by_pid_ns(tid, ns);
3571                 if (pos && (pos->group_leader == leader))
3572                         goto found;
3573         }
3574
3575         /* If nr exceeds the number of threads there is nothing todo */
3576         pos = NULL;
3577         if (nr && nr >= get_nr_threads(leader))
3578                 goto out;
3579
3580         /* If we haven't found our starting place yet start
3581          * with the leader and walk nr threads forward.
3582          */
3583         for (pos = leader; nr > 0; --nr) {
3584                 pos = next_thread(pos);
3585                 if (pos == leader) {
3586                         pos = NULL;
3587                         goto out;
3588                 }
3589         }
3590 found:
3591         get_task_struct(pos);
3592 out:
3593         rcu_read_unlock();
3594         return pos;
3595 }
3596
3597 /*
3598  * Find the next thread in the thread list.
3599  * Return NULL if there is an error or no next thread.
3600  *
3601  * The reference to the input task_struct is released.
3602  */
3603 static struct task_struct *next_tid(struct task_struct *start)
3604 {
3605         struct task_struct *pos = NULL;
3606         rcu_read_lock();
3607         if (pid_alive(start)) {
3608                 pos = next_thread(start);
3609                 if (thread_group_leader(pos))
3610                         pos = NULL;
3611                 else
3612                         get_task_struct(pos);
3613         }
3614         rcu_read_unlock();
3615         put_task_struct(start);
3616         return pos;
3617 }
3618
3619 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3620         struct task_struct *task, int tid)
3621 {
3622         char name[PROC_NUMBUF];
3623         int len = snprintf(name, sizeof(name), "%d", tid);
3624         return proc_fill_cache(filp, dirent, filldir, name, len,
3625                                 proc_task_instantiate, task, NULL);
3626 }
3627
3628 /* for the /proc/TGID/task/ directories */
3629 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3630 {
3631         struct dentry *dentry = filp->f_path.dentry;
3632         struct inode *inode = dentry->d_inode;
3633         struct task_struct *leader = NULL;
3634         struct task_struct *task;
3635         int retval = -ENOENT;
3636         ino_t ino;
3637         int tid;
3638         struct pid_namespace *ns;
3639
3640         task = get_proc_task(inode);
3641         if (!task)
3642                 goto out_no_task;
3643         rcu_read_lock();
3644         if (pid_alive(task)) {
3645                 leader = task->group_leader;
3646                 get_task_struct(leader);
3647         }
3648         rcu_read_unlock();
3649         put_task_struct(task);
3650         if (!leader)
3651                 goto out_no_task;
3652         retval = 0;
3653
3654         switch ((unsigned long)filp->f_pos) {
3655         case 0:
3656                 ino = inode->i_ino;
3657                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3658                         goto out;
3659                 filp->f_pos++;
3660                 /* fall through */
3661         case 1:
3662                 ino = parent_ino(dentry);
3663                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3664                         goto out;
3665                 filp->f_pos++;
3666                 /* fall through */
3667         }
3668
3669         /* f_version caches the tgid value that the last readdir call couldn't
3670          * return. lseek aka telldir automagically resets f_version to 0.
3671          */
3672         ns = filp->f_dentry->d_sb->s_fs_info;
3673         tid = (int)filp->f_version;
3674         filp->f_version = 0;
3675         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3676              task;
3677              task = next_tid(task), filp->f_pos++) {
3678                 tid = task_pid_nr_ns(task, ns);
3679                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3680                         /* returning this tgid failed, save it as the first
3681                          * pid for the next readir call */
3682                         filp->f_version = (u64)tid;
3683                         put_task_struct(task);
3684                         break;
3685                 }
3686         }
3687 out:
3688         put_task_struct(leader);
3689 out_no_task:
3690         return retval;
3691 }
3692
3693 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3694 {
3695         struct inode *inode = dentry->d_inode;
3696         struct task_struct *p = get_proc_task(inode);
3697         generic_fillattr(inode, stat);
3698
3699         if (p) {
3700                 stat->nlink += get_nr_threads(p);
3701                 put_task_struct(p);
3702         }
3703
3704         return 0;
3705 }
3706
3707 static const struct inode_operations proc_task_inode_operations = {
3708         .lookup         = proc_task_lookup,
3709         .getattr        = proc_task_getattr,
3710         .setattr        = proc_setattr,
3711         .permission     = proc_pid_permission,
3712 };
3713
3714 static const struct file_operations proc_task_operations = {
3715         .read           = generic_read_dir,
3716         .readdir        = proc_task_readdir,
3717         .llseek         = default_llseek,
3718 };