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 points = 0;
 415
 416         read_lock(&tasklist_lock);
 417         if (pid_alive(task))
 418                 points = oom_badness(task, NULL, NULL,
 419                                         totalram_pages + total_swap_pages);
 420         read_unlock(&tasklist_lock);
 421         return sprintf(buffer, "%lu\n", points);
 422 }
 423
 424 struct limit_names {
 425         char *name;
 426         char *unit;
 427 };
 428
 429 static const struct limit_names lnames[RLIM_NLIMITS] = {
 430         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 431         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 432         [RLIMIT_DATA] = {"Max data size", "bytes"},
 433         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 434         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 435         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 436         [RLIMIT_NPROC] = {"Max processes", "processes"},
 437         [RLIMIT_NOFILE] = {"Max open files", "files"},
 438         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 439         [RLIMIT_AS] = {"Max address space", "bytes"},
 440         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 441         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 442         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 443         [RLIMIT_NICE] = {"Max nice priority", NULL},
 444         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 445         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 446 };
 447
 448 /* Display limits for a process */
 449 static int proc_pid_limits(struct task_struct *task, char *buffer)
 450 {
 451         unsigned int i;
 452         int count = 0;
 453         unsigned long flags;
 454         char *bufptr = buffer;
 455
 456         struct rlimit rlim[RLIM_NLIMITS];
 457
 458         if (!lock_task_sighand(task, &flags))
 459                 return 0;
 460         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 461         unlock_task_sighand(task, &flags);
 462
 463         /*
 464          * print the file header
 465          */
 466         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
 467                         "Limit", "Soft Limit", "Hard Limit", "Units");
 468
 469         for (i = 0; i < RLIM_NLIMITS; i++) {
 470                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 471                         count += sprintf(&bufptr[count], "%-25s %-20s ",
 472                                          lnames[i].name, "unlimited");
 473                 else
 474                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
 475                                          lnames[i].name, rlim[i].rlim_cur);
 476
 477                 if (rlim[i].rlim_max == RLIM_INFINITY)
 478                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
 479                 else
 480                         count += sprintf(&bufptr[count], "%-20lu ",
 481                                          rlim[i].rlim_max);
 482
 483                 if (lnames[i].unit)
 484                         count += sprintf(&bufptr[count], "%-10s\n",
 485                                          lnames[i].unit);
 486                 else
 487                         count += sprintf(&bufptr[count], "\n");
 488         }
 489
 490         return count;
 491 }
 492
 493 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 494 static int proc_pid_syscall(struct task_struct *task, char *buffer)
 495 {
 496         long nr;
 497         unsigned long args[6], sp, pc;
 498         int res = lock_trace(task);
 499         if (res)
 500                 return res;
 501
 502         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 503                 res = sprintf(buffer, "running\n");
 504         else if (nr < 0)
 505                 res = sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 506         else
 507                 res = sprintf(buffer,
 508                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 509                        nr,
 510                        args[0], args[1], args[2], args[3], args[4], args[5],
 511                        sp, pc);
 512         unlock_trace(task);
 513         return res;
 514 }
 515 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 516
 517 /************************************************************************/
 518 /*                       Here the fs part begins                        */
 519 /************************************************************************/
 520
 521 /* permission checks */
 522 static int proc_fd_access_allowed(struct inode *inode)
 523 {
 524         struct task_struct *task;
 525         int allowed = 0;
 526         /* Allow access to a task's file descriptors if it is us or we
 527          * may use ptrace attach to the process and find out that
 528          * information.
 529          */
 530         task = get_proc_task(inode);
 531         if (task) {
 532                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
 533                 put_task_struct(task);
 534         }
 535         return allowed;
 536 }
 537
 538 int proc_setattr(struct dentry *dentry, struct iattr *attr)
 539 {
 540         int error;
 541         struct inode *inode = dentry->d_inode;
 542
 543         if (attr->ia_valid & ATTR_MODE)
 544                 return -EPERM;
 545
 546         error = inode_change_ok(inode, attr);
 547         if (error)
 548                 return error;
 549
 550         if ((attr->ia_valid & ATTR_SIZE) &&
 551             attr->ia_size != i_size_read(inode)) {
 552                 error = vmtruncate(inode, attr->ia_size);
 553                 if (error)
 554                         return error;
 555         }
 556
 557         setattr_copy(inode, attr);
 558         mark_inode_dirty(inode);
 559         return 0;
 560 }
 561
 562 /*
 563  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
 564  * or euid/egid (for hide_pid_min=2)?
 565  */
 566 static bool has_pid_permissions(struct pid_namespace *pid,
 567                                  struct task_struct *task,
 568                                  int hide_pid_min)
 569 {
 570         if (pid->hide_pid < hide_pid_min)
 571                 return true;
 572         if (in_group_p(pid->pid_gid))
 573                 return true;
 574         return ptrace_may_access(task, PTRACE_MODE_READ);
 575 }
 576
 577
 578 static int proc_pid_permission(struct inode *inode, int mask)
 579 {
 580         struct pid_namespace *pid = inode->i_sb->s_fs_info;
 581         struct task_struct *task;
 582         bool has_perms;
 583
 584         task = get_proc_task(inode);
 585         if (!task)
 586                 return -ESRCH;
 587         has_perms = has_pid_permissions(pid, task, 1);
 588         put_task_struct(task);
 589
 590         if (!has_perms) {
 591                 if (pid->hide_pid == 2) {
 592                         /*
 593                          * Let's make getdents(), stat(), and open()
 594                          * consistent with each other.  If a process
 595                          * may not stat() a file, it shouldn't be seen
 596                          * in procfs at all.
 597                          */
 598                         return -ENOENT;
 599                 }
 600
 601                 return -EPERM;
 602         }
 603         return generic_permission(inode, mask);
 604 }
 605
 606
 607
 608 static const struct inode_operations proc_def_inode_operations = {
 609         .setattr        = proc_setattr,
 610 };
 611
 612 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 613
 614 static ssize_t proc_info_read(struct file * file, char __user * buf,
 615                           size_t count, loff_t *ppos)
 616 {
 617         struct inode * inode = file->f_path.dentry->d_inode;
 618         unsigned long page;
 619         ssize_t length;
 620         struct task_struct *task = get_proc_task(inode);
 621
 622         length = -ESRCH;
 623         if (!task)
 624                 goto out_no_task;
 625
 626         if (count > PROC_BLOCK_SIZE)
 627                 count = PROC_BLOCK_SIZE;
 628
 629         length = -ENOMEM;
 630         if (!(page = __get_free_page(GFP_TEMPORARY)))
 631                 goto out;
 632
 633         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 634
 635         if (length >= 0)
 636                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 637         free_page(page);
 638 out:
 639         put_task_struct(task);
 640 out_no_task:
 641         return length;
 642 }
 643
 644 static const struct file_operations proc_info_file_operations = {
 645         .read           = proc_info_read,
 646         .llseek         = generic_file_llseek,
 647 };
 648
 649 static int proc_single_show(struct seq_file *m, void *v)
 650 {
 651         struct inode *inode = m->private;
 652         struct pid_namespace *ns;
 653         struct pid *pid;
 654         struct task_struct *task;
 655         int ret;
 656
 657         ns = inode->i_sb->s_fs_info;
 658         pid = proc_pid(inode);
 659         task = get_pid_task(pid, PIDTYPE_PID);
 660         if (!task)
 661                 return -ESRCH;
 662
 663         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 664
 665         put_task_struct(task);
 666         return ret;
 667 }
 668
 669 static int proc_single_open(struct inode *inode, struct file *filp)
 670 {
 671         return single_open(filp, proc_single_show, inode);
 672 }
 673
 674 static const struct file_operations proc_single_file_operations = {
 675         .open           = proc_single_open,
 676         .read           = seq_read,
 677         .llseek         = seq_lseek,
 678         .release        = single_release,
 679 };
 680
 681 static int mem_open(struct inode* inode, struct file* file)
 682 {
 683         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 684         struct mm_struct *mm;
 685
 686         if (!task)
 687                 return -ESRCH;
 688
 689         mm = mm_access(task, PTRACE_MODE_ATTACH);
 690         put_task_struct(task);
 691
 692         if (IS_ERR(mm))
 693                 return PTR_ERR(mm);
 694
 695         if (mm) {
 696                 /* ensure this mm_struct can't be freed */
 697                 atomic_inc(&mm->mm_count);
 698                 /* but do not pin its memory */
 699                 mmput(mm);
 700         }
 701
 702         /* OK to pass negative loff_t, we can catch out-of-range */
 703         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 704         file->private_data = mm;
 705
 706         return 0;
 707 }
 708
 709 static ssize_t mem_rw(struct file *file, char __user *buf,
 710                         size_t count, loff_t *ppos, int write)
 711 {
 712         struct mm_struct *mm = file->private_data;
 713         unsigned long addr = *ppos;
 714         ssize_t copied;
 715         char *page;
 716
 717         if (!mm)
 718                 return 0;
 719
 720         page = (char *)__get_free_page(GFP_TEMPORARY);
 721         if (!page)
 722                 return -ENOMEM;
 723
 724         copied = 0;
 725         if (!atomic_inc_not_zero(&mm->mm_users))
 726                 goto free;
 727
 728         while (count > 0) {
 729                 int this_len = min_t(int, count, PAGE_SIZE);
 730
 731                 if (write && copy_from_user(page, buf, this_len)) {
 732                         copied = -EFAULT;
 733                         break;
 734                 }
 735
 736                 this_len = access_remote_vm(mm, addr, page, this_len, write);
 737                 if (!this_len) {
 738                         if (!copied)
 739                                 copied = -EIO;
 740                         break;
 741                 }
 742
 743                 if (!write && copy_to_user(buf, page, this_len)) {
 744                         copied = -EFAULT;
 745                         break;
 746                 }
 747
 748                 buf += this_len;
 749                 addr += this_len;
 750                 copied += this_len;
 751                 count -= this_len;
 752         }
 753         *ppos = addr;
 754
 755         mmput(mm);
 756 free:
 757         free_page((unsigned long) page);
 758         return copied;
 759 }
 760
 761 static ssize_t mem_read(struct file *file, char __user *buf,
 762                         size_t count, loff_t *ppos)
 763 {
 764         return mem_rw(file, buf, count, ppos, 0);
 765 }
 766
 767 static ssize_t mem_write(struct file *file, const char __user *buf,
 768                          size_t count, loff_t *ppos)
 769 {
 770         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 771 }
 772
 773 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 774 {
 775         switch (orig) {
 776         case 0:
 777                 file->f_pos = offset;
 778                 break;
 779         case 1:
 780                 file->f_pos += offset;
 781                 break;
 782         default:
 783                 return -EINVAL;
 784         }
 785         force_successful_syscall_return();
 786         return file->f_pos;
 787 }
 788
 789 static int mem_release(struct inode *inode, struct file *file)
 790 {
 791         struct mm_struct *mm = file->private_data;
 792         if (mm)
 793                 mmdrop(mm);
 794         return 0;
 795 }
 796
 797 static const struct file_operations proc_mem_operations = {
 798         .llseek         = mem_lseek,
 799         .read           = mem_read,
 800         .write          = mem_write,
 801         .open           = mem_open,
 802         .release        = mem_release,
 803 };
 804
 805 static ssize_t environ_read(struct file *file, char __user *buf,
 806                         size_t count, loff_t *ppos)
 807 {
 808         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 809         char *page;
 810         unsigned long src = *ppos;
 811         int ret = -ESRCH;
 812         struct mm_struct *mm;
 813
 814         if (!task)
 815                 goto out_no_task;
 816
 817         ret = -ENOMEM;
 818         page = (char *)__get_free_page(GFP_TEMPORARY);
 819         if (!page)
 820                 goto out;
 821
 822
 823         mm = mm_for_maps(task);
 824         ret = PTR_ERR(mm);
 825         if (!mm || IS_ERR(mm))
 826                 goto out_free;
 827
 828         ret = 0;
 829         while (count > 0) {
 830                 int this_len, retval, max_len;
 831
 832                 this_len = mm->env_end - (mm->env_start + src);
 833
 834                 if (this_len <= 0)
 835                         break;
 836
 837                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 838                 this_len = (this_len > max_len) ? max_len : this_len;
 839
 840                 retval = access_process_vm(task, (mm->env_start + src),
 841                         page, this_len, 0);
 842
 843                 if (retval <= 0) {
 844                         ret = retval;
 845                         break;
 846                 }
 847
 848                 if (copy_to_user(buf, page, retval)) {
 849                         ret = -EFAULT;
 850                         break;
 851                 }
 852
 853                 ret += retval;
 854                 src += retval;
 855                 buf += retval;
 856                 count -= retval;
 857         }
 858         *ppos = src;
 859
 860         mmput(mm);
 861 out_free:
 862         free_page((unsigned long) page);
 863 out:
 864         put_task_struct(task);
 865 out_no_task:
 866         return ret;
 867 }
 868
 869 static const struct file_operations proc_environ_operations = {
 870         .read           = environ_read,
 871         .llseek         = generic_file_llseek,
 872 };
 873
 874 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
 875                                 size_t count, loff_t *ppos)
 876 {
 877         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 878         char buffer[PROC_NUMBUF];
 879         size_t len;
 880         int oom_adjust = OOM_DISABLE;
 881         unsigned long flags;
 882
 883         if (!task)
 884                 return -ESRCH;
 885
 886         if (lock_task_sighand(task, &flags)) {
 887                 oom_adjust = task->signal->oom_adj;
 888                 unlock_task_sighand(task, &flags);
 889         }
 890
 891         put_task_struct(task);
 892
 893         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
 894
 895         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 896 }
 897
 898 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
 899                                 size_t count, loff_t *ppos)
 900 {
 901         struct task_struct *task;
 902         char buffer[PROC_NUMBUF];
 903         int oom_adjust;
 904         unsigned long flags;
 905         int err;
 906
 907         memset(buffer, 0, sizeof(buffer));
 908         if (count > sizeof(buffer) - 1)
 909                 count = sizeof(buffer) - 1;
 910         if (copy_from_user(buffer, buf, count)) {
 911                 err = -EFAULT;
 912                 goto out;
 913         }
 914
 915         err = kstrtoint(strstrip(buffer), 0, &oom_adjust);
 916         if (err)
 917                 goto out;
 918         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
 919              oom_adjust != OOM_DISABLE) {
 920                 err = -EINVAL;
 921                 goto out;
 922         }
 923
 924         task = get_proc_task(file->f_path.dentry->d_inode);
 925         if (!task) {
 926                 err = -ESRCH;
 927                 goto out;
 928         }
 929
 930         task_lock(task);
 931         if (!task->mm) {
 932                 err = -EINVAL;
 933                 goto err_task_lock;
 934         }
 935
 936         if (!lock_task_sighand(task, &flags)) {
 937                 err = -ESRCH;
 938                 goto err_task_lock;
 939         }
 940
 941         if (oom_adjust < task->signal->oom_adj && !capable(CAP_SYS_RESOURCE)) {
 942                 err = -EACCES;
 943                 goto err_sighand;
 944         }
 945
 946         /*
 947          * Warn that /proc/pid/oom_adj is deprecated, see
 948          * Documentation/feature-removal-schedule.txt.
 949          */
 950         printk_once(KERN_WARNING "%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
 951                   current->comm, task_pid_nr(current), task_pid_nr(task),
 952                   task_pid_nr(task));
 953         task->signal->oom_adj = oom_adjust;
 954         /*
 955          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
 956          * value is always attainable.
 957          */
 958         if (task->signal->oom_adj == OOM_ADJUST_MAX)
 959                 task->signal->oom_score_adj = OOM_SCORE_ADJ_MAX;
 960         else
 961                 task->signal->oom_score_adj = (oom_adjust * OOM_SCORE_ADJ_MAX) /
 962                                                                 -OOM_DISABLE;
 963         trace_oom_score_adj_update(task);
 964 err_sighand:
 965         unlock_task_sighand(task, &flags);
 966 err_task_lock:
 967         task_unlock(task);
 968         put_task_struct(task);
 969 out:
 970         return err < 0 ? err : count;
 971 }
 972
 973 static const struct file_operations proc_oom_adjust_operations = {
 974         .read           = oom_adjust_read,
 975         .write          = oom_adjust_write,
 976         .llseek         = generic_file_llseek,
 977 };
 978
 979 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
 980                                         size_t count, loff_t *ppos)
 981 {
 982         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 983         char buffer[PROC_NUMBUF];
 984         int oom_score_adj = OOM_SCORE_ADJ_MIN;
 985         unsigned long flags;
 986         size_t len;
 987
 988         if (!task)
 989                 return -ESRCH;
 990         if (lock_task_sighand(task, &flags)) {
 991                 oom_score_adj = task->signal->oom_score_adj;
 992                 unlock_task_sighand(task, &flags);
 993         }
 994         put_task_struct(task);
 995         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_score_adj);
 996         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 997 }
 998
 999 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1000                                         size_t count, loff_t *ppos)
1001 {
1002         struct task_struct *task;
1003         char buffer[PROC_NUMBUF];
1004         unsigned long flags;
1005         int oom_score_adj;
1006         int err;
1007
1008         memset(buffer, 0, sizeof(buffer));
1009         if (count > sizeof(buffer) - 1)
1010                 count = sizeof(buffer) - 1;
1011         if (copy_from_user(buffer, buf, count)) {
1012                 err = -EFAULT;
1013                 goto out;
1014         }
1015
1016         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1017         if (err)
1018                 goto out;
1019         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1020                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1021                 err = -EINVAL;
1022                 goto out;
1023         }
1024
1025         task = get_proc_task(file->f_path.dentry->d_inode);
1026         if (!task) {
1027                 err = -ESRCH;
1028                 goto out;
1029         }
1030
1031         task_lock(task);
1032         if (!task->mm) {
1033                 err = -EINVAL;
1034                 goto err_task_lock;
1035         }
1036
1037         if (!lock_task_sighand(task, &flags)) {
1038                 err = -ESRCH;
1039                 goto err_task_lock;
1040         }
1041
1042         if (oom_score_adj < task->signal->oom_score_adj_min &&
1043                         !capable(CAP_SYS_RESOURCE)) {
1044                 err = -EACCES;
1045                 goto err_sighand;
1046         }
1047
1048         task->signal->oom_score_adj = oom_score_adj;
1049         if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1050                 task->signal->oom_score_adj_min = oom_score_adj;
1051         trace_oom_score_adj_update(task);
1052         /*
1053          * Scale /proc/pid/oom_adj appropriately ensuring that OOM_DISABLE is
1054          * always attainable.
1055          */
1056         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1057                 task->signal->oom_adj = OOM_DISABLE;
1058         else
1059                 task->signal->oom_adj = (oom_score_adj * OOM_ADJUST_MAX) /
1060                                                         OOM_SCORE_ADJ_MAX;
1061 err_sighand:
1062         unlock_task_sighand(task, &flags);
1063 err_task_lock:
1064         task_unlock(task);
1065         put_task_struct(task);
1066 out:
1067         return err < 0 ? err : count;
1068 }
1069
1070 static const struct file_operations proc_oom_score_adj_operations = {
1071         .read           = oom_score_adj_read,
1072         .write          = oom_score_adj_write,
1073         .llseek         = default_llseek,
1074 };
1075
1076 #ifdef CONFIG_AUDITSYSCALL
1077 #define TMPBUFLEN 21
1078 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1079                                   size_t count, loff_t *ppos)
1080 {
1081         struct inode * inode = file->f_path.dentry->d_inode;
1082         struct task_struct *task = get_proc_task(inode);
1083         ssize_t length;
1084         char tmpbuf[TMPBUFLEN];
1085
1086         if (!task)
1087                 return -ESRCH;
1088         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1089                                 audit_get_loginuid(task));
1090         put_task_struct(task);
1091         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1092 }
1093
1094 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1095                                    size_t count, loff_t *ppos)
1096 {
1097         struct inode * inode = file->f_path.dentry->d_inode;
1098         char *page, *tmp;
1099         ssize_t length;
1100         uid_t loginuid;
1101
1102         rcu_read_lock();
1103         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1104                 rcu_read_unlock();
1105                 return -EPERM;
1106         }
1107         rcu_read_unlock();
1108
1109         if (count >= PAGE_SIZE)
1110                 count = PAGE_SIZE - 1;
1111
1112         if (*ppos != 0) {
1113                 /* No partial writes. */
1114                 return -EINVAL;
1115         }
1116         page = (char*)__get_free_page(GFP_TEMPORARY);
1117         if (!page)
1118                 return -ENOMEM;
1119         length = -EFAULT;
1120         if (copy_from_user(page, buf, count))
1121                 goto out_free_page;
1122
1123         page[count] = '\0';
1124         loginuid = simple_strtoul(page, &tmp, 10);
1125         if (tmp == page) {
1126                 length = -EINVAL;
1127                 goto out_free_page;
1128
1129         }
1130         length = audit_set_loginuid(loginuid);
1131         if (likely(length == 0))
1132                 length = count;
1133
1134 out_free_page:
1135         free_page((unsigned long) page);
1136         return length;
1137 }
1138
1139 static const struct file_operations proc_loginuid_operations = {
1140         .read           = proc_loginuid_read,
1141         .write          = proc_loginuid_write,
1142         .llseek         = generic_file_llseek,
1143 };
1144
1145 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1146                                   size_t count, loff_t *ppos)
1147 {
1148         struct inode * inode = file->f_path.dentry->d_inode;
1149         struct task_struct *task = get_proc_task(inode);
1150         ssize_t length;
1151         char tmpbuf[TMPBUFLEN];
1152
1153         if (!task)
1154                 return -ESRCH;
1155         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1156                                 audit_get_sessionid(task));
1157         put_task_struct(task);
1158         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1159 }
1160
1161 static const struct file_operations proc_sessionid_operations = {
1162         .read           = proc_sessionid_read,
1163         .llseek         = generic_file_llseek,
1164 };
1165 #endif
1166
1167 #ifdef CONFIG_FAULT_INJECTION
1168 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1169                                       size_t count, loff_t *ppos)
1170 {
1171         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1172         char buffer[PROC_NUMBUF];
1173         size_t len;
1174         int make_it_fail;
1175
1176         if (!task)
1177                 return -ESRCH;
1178         make_it_fail = task->make_it_fail;
1179         put_task_struct(task);
1180
1181         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1182
1183         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1184 }
1185
1186 static ssize_t proc_fault_inject_write(struct file * file,
1187                         const char __user * buf, size_t count, loff_t *ppos)
1188 {
1189         struct task_struct *task;
1190         char buffer[PROC_NUMBUF], *end;
1191         int make_it_fail;
1192
1193         if (!capable(CAP_SYS_RESOURCE))
1194                 return -EPERM;
1195         memset(buffer, 0, sizeof(buffer));
1196         if (count > sizeof(buffer) - 1)
1197                 count = sizeof(buffer) - 1;
1198         if (copy_from_user(buffer, buf, count))
1199                 return -EFAULT;
1200         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1201         if (*end)
1202                 return -EINVAL;
1203         task = get_proc_task(file->f_dentry->d_inode);
1204         if (!task)
1205                 return -ESRCH;
1206         task->make_it_fail = make_it_fail;
1207         put_task_struct(task);
1208
1209         return count;
1210 }
1211
1212 static const struct file_operations proc_fault_inject_operations = {
1213         .read           = proc_fault_inject_read,
1214         .write          = proc_fault_inject_write,
1215         .llseek         = generic_file_llseek,
1216 };
1217 #endif
1218
1219
1220 #ifdef CONFIG_SCHED_DEBUG
1221 /*
1222  * Print out various scheduling related per-task fields:
1223  */
1224 static int sched_show(struct seq_file *m, void *v)
1225 {
1226         struct inode *inode = m->private;
1227         struct task_struct *p;
1228
1229         p = get_proc_task(inode);
1230         if (!p)
1231                 return -ESRCH;
1232         proc_sched_show_task(p, m);
1233
1234         put_task_struct(p);
1235
1236         return 0;
1237 }
1238
1239 static ssize_t
1240 sched_write(struct file *file, const char __user *buf,
1241             size_t count, loff_t *offset)
1242 {
1243         struct inode *inode = file->f_path.dentry->d_inode;
1244         struct task_struct *p;
1245
1246         p = get_proc_task(inode);
1247         if (!p)
1248                 return -ESRCH;
1249         proc_sched_set_task(p);
1250
1251         put_task_struct(p);
1252
1253         return count;
1254 }
1255
1256 static int sched_open(struct inode *inode, struct file *filp)
1257 {
1258         return single_open(filp, sched_show, inode);
1259 }
1260
1261 static const struct file_operations proc_pid_sched_operations = {
1262         .open           = sched_open,
1263         .read           = seq_read,
1264         .write          = sched_write,
1265         .llseek         = seq_lseek,
1266         .release        = single_release,
1267 };
1268
1269 #endif
1270
1271 #ifdef CONFIG_SCHED_AUTOGROUP
1272 /*
1273  * Print out autogroup related information:
1274  */
1275 static int sched_autogroup_show(struct seq_file *m, void *v)
1276 {
1277         struct inode *inode = m->private;
1278         struct task_struct *p;
1279
1280         p = get_proc_task(inode);
1281         if (!p)
1282                 return -ESRCH;
1283         proc_sched_autogroup_show_task(p, m);
1284
1285         put_task_struct(p);
1286
1287         return 0;
1288 }
1289
1290 static ssize_t
1291 sched_autogroup_write(struct file *file, const char __user *buf,
1292             size_t count, loff_t *offset)
1293 {
1294         struct inode *inode = file->f_path.dentry->d_inode;
1295         struct task_struct *p;
1296         char buffer[PROC_NUMBUF];
1297         int nice;
1298         int err;
1299
1300         memset(buffer, 0, sizeof(buffer));
1301         if (count > sizeof(buffer) - 1)
1302                 count = sizeof(buffer) - 1;
1303         if (copy_from_user(buffer, buf, count))
1304                 return -EFAULT;
1305
1306         err = kstrtoint(strstrip(buffer), 0, &nice);
1307         if (err < 0)
1308                 return err;
1309
1310         p = get_proc_task(inode);
1311         if (!p)
1312                 return -ESRCH;
1313
1314         err = proc_sched_autogroup_set_nice(p, nice);
1315         if (err)
1316                 count = err;
1317
1318         put_task_struct(p);
1319
1320         return count;
1321 }
1322
1323 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1324 {
1325         int ret;
1326
1327         ret = single_open(filp, sched_autogroup_show, NULL);
1328         if (!ret) {
1329                 struct seq_file *m = filp->private_data;
1330
1331                 m->private = inode;
1332         }
1333         return ret;
1334 }
1335
1336 static const struct file_operations proc_pid_sched_autogroup_operations = {
1337         .open           = sched_autogroup_open,
1338         .read           = seq_read,
1339         .write          = sched_autogroup_write,
1340         .llseek         = seq_lseek,
1341         .release        = single_release,
1342 };
1343
1344 #endif /* CONFIG_SCHED_AUTOGROUP */
1345
1346 static ssize_t comm_write(struct file *file, const char __user *buf,
1347                                 size_t count, loff_t *offset)
1348 {
1349         struct inode *inode = file->f_path.dentry->d_inode;
1350         struct task_struct *p;
1351         char buffer[TASK_COMM_LEN];
1352
1353         memset(buffer, 0, sizeof(buffer));
1354         if (count > sizeof(buffer) - 1)
1355                 count = sizeof(buffer) - 1;
1356         if (copy_from_user(buffer, buf, count))
1357                 return -EFAULT;
1358
1359         p = get_proc_task(inode);
1360         if (!p)
1361                 return -ESRCH;
1362
1363         if (same_thread_group(current, p))
1364                 set_task_comm(p, buffer);
1365         else
1366                 count = -EINVAL;
1367
1368         put_task_struct(p);
1369
1370         return count;
1371 }
1372
1373 static int comm_show(struct seq_file *m, void *v)
1374 {
1375         struct inode *inode = m->private;
1376         struct task_struct *p;
1377
1378         p = get_proc_task(inode);
1379         if (!p)
1380                 return -ESRCH;
1381
1382         task_lock(p);
1383         seq_printf(m, "%s\n", p->comm);
1384         task_unlock(p);
1385
1386         put_task_struct(p);
1387
1388         return 0;
1389 }
1390
1391 static int comm_open(struct inode *inode, struct file *filp)
1392 {
1393         return single_open(filp, comm_show, inode);
1394 }
1395
1396 static const struct file_operations proc_pid_set_comm_operations = {
1397         .open           = comm_open,
1398         .read           = seq_read,
1399         .write          = comm_write,
1400         .llseek         = seq_lseek,
1401         .release        = single_release,
1402 };
1403
1404 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1405 {
1406         struct task_struct *task;
1407         struct mm_struct *mm;
1408         struct file *exe_file;
1409
1410         task = get_proc_task(dentry->d_inode);
1411         if (!task)
1412                 return -ENOENT;
1413         mm = get_task_mm(task);
1414         put_task_struct(task);
1415         if (!mm)
1416                 return -ENOENT;
1417         exe_file = get_mm_exe_file(mm);
1418         mmput(mm);
1419         if (exe_file) {
1420                 *exe_path = exe_file->f_path;
1421                 path_get(&exe_file->f_path);
1422                 fput(exe_file);
1423                 return 0;
1424         } else
1425                 return -ENOENT;
1426 }
1427
1428 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1429 {
1430         struct inode *inode = dentry->d_inode;
1431         int error = -EACCES;
1432
1433         /* We don't need a base pointer in the /proc filesystem */
1434         path_put(&nd->path);
1435
1436         /* Are we allowed to snoop on the tasks file descriptors? */
1437         if (!proc_fd_access_allowed(inode))
1438                 goto out;
1439
1440         error = PROC_I(inode)->op.proc_get_link(dentry, &nd->path);
1441 out:
1442         return ERR_PTR(error);
1443 }
1444
1445 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1446 {
1447         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1448         char *pathname;
1449         int len;
1450
1451         if (!tmp)
1452                 return -ENOMEM;
1453
1454         pathname = d_path(path, tmp, PAGE_SIZE);
1455         len = PTR_ERR(pathname);
1456         if (IS_ERR(pathname))
1457                 goto out;
1458         len = tmp + PAGE_SIZE - 1 - pathname;
1459
1460         if (len > buflen)
1461                 len = buflen;
1462         if (copy_to_user(buffer, pathname, len))
1463                 len = -EFAULT;
1464  out:
1465         free_page((unsigned long)tmp);
1466         return len;
1467 }
1468
1469 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1470 {
1471         int error = -EACCES;
1472         struct inode *inode = dentry->d_inode;
1473         struct path path;
1474
1475         /* Are we allowed to snoop on the tasks file descriptors? */
1476         if (!proc_fd_access_allowed(inode))
1477                 goto out;
1478
1479         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1480         if (error)
1481                 goto out;
1482
1483         error = do_proc_readlink(&path, buffer, buflen);
1484         path_put(&path);
1485 out:
1486         return error;
1487 }
1488
1489 static const struct inode_operations proc_pid_link_inode_operations = {
1490         .readlink       = proc_pid_readlink,
1491         .follow_link    = proc_pid_follow_link,
1492         .setattr        = proc_setattr,
1493 };
1494
1495
1496 /* building an inode */
1497
1498 static int task_dumpable(struct task_struct *task)
1499 {
1500         int dumpable = 0;
1501         struct mm_struct *mm;
1502
1503         task_lock(task);
1504         mm = task->mm;
1505         if (mm)
1506                 dumpable = get_dumpable(mm);
1507         task_unlock(task);
1508         if(dumpable == 1)
1509                 return 1;
1510         return 0;
1511 }
1512
1513 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1514 {
1515         struct inode * inode;
1516         struct proc_inode *ei;
1517         const struct cred *cred;
1518
1519         /* We need a new inode */
1520
1521         inode = new_inode(sb);
1522         if (!inode)
1523                 goto out;
1524
1525         /* Common stuff */
1526         ei = PROC_I(inode);
1527         inode->i_ino = get_next_ino();
1528         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1529         inode->i_op = &proc_def_inode_operations;
1530
1531         /*
1532          * grab the reference to task.
1533          */
1534         ei->pid = get_task_pid(task, PIDTYPE_PID);
1535         if (!ei->pid)
1536                 goto out_unlock;
1537
1538         if (task_dumpable(task)) {
1539                 rcu_read_lock();
1540                 cred = __task_cred(task);
1541                 inode->i_uid = cred->euid;
1542                 inode->i_gid = cred->egid;
1543                 rcu_read_unlock();
1544         }
1545         security_task_to_inode(task, inode);
1546
1547 out:
1548         return inode;
1549
1550 out_unlock:
1551         iput(inode);
1552         return NULL;
1553 }
1554
1555 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1556 {
1557         struct inode *inode = dentry->d_inode;
1558         struct task_struct *task;
1559         const struct cred *cred;
1560         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1561
1562         generic_fillattr(inode, stat);
1563
1564         rcu_read_lock();
1565         stat->uid = GLOBAL_ROOT_UID;
1566         stat->gid = GLOBAL_ROOT_GID;
1567         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1568         if (task) {
1569                 if (!has_pid_permissions(pid, task, 2)) {
1570                         rcu_read_unlock();
1571                         /*
1572                          * This doesn't prevent learning whether PID exists,
1573                          * it only makes getattr() consistent with readdir().
1574                          */
1575                         return -ENOENT;
1576                 }
1577                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1578                     task_dumpable(task)) {
1579                         cred = __task_cred(task);
1580                         stat->uid = cred->euid;
1581                         stat->gid = cred->egid;
1582                 }
1583         }
1584         rcu_read_unlock();
1585         return 0;
1586 }
1587
1588 /* dentry stuff */
1589
1590 /*
1591  *      Exceptional case: normally we are not allowed to unhash a busy
1592  * directory. In this case, however, we can do it - no aliasing problems
1593  * due to the way we treat inodes.
1594  *
1595  * Rewrite the inode's ownerships here because the owning task may have
1596  * performed a setuid(), etc.
1597  *
1598  * Before the /proc/pid/status file was created the only way to read
1599  * the effective uid of a /process was to stat /proc/pid.  Reading
1600  * /proc/pid/status is slow enough that procps and other packages
1601  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1602  * made this apply to all per process world readable and executable
1603  * directories.
1604  */
1605 int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1606 {
1607         struct inode *inode;
1608         struct task_struct *task;
1609         const struct cred *cred;
1610
1611         if (nd && nd->flags & LOOKUP_RCU)
1612                 return -ECHILD;
1613
1614         inode = dentry->d_inode;
1615         task = get_proc_task(inode);
1616
1617         if (task) {
1618                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1619                     task_dumpable(task)) {
1620                         rcu_read_lock();
1621                         cred = __task_cred(task);
1622                         inode->i_uid = cred->euid;
1623                         inode->i_gid = cred->egid;
1624                         rcu_read_unlock();
1625                 } else {
1626                         inode->i_uid = GLOBAL_ROOT_UID;
1627                         inode->i_gid = GLOBAL_ROOT_GID;
1628                 }
1629                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1630                 security_task_to_inode(task, inode);
1631                 put_task_struct(task);
1632                 return 1;
1633         }
1634         d_drop(dentry);
1635         return 0;
1636 }
1637
1638 static int pid_delete_dentry(const struct dentry * dentry)
1639 {
1640         /* Is the task we represent dead?
1641          * If so, then don't put the dentry on the lru list,
1642          * kill it immediately.
1643          */
1644         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1645 }
1646
1647 const struct dentry_operations pid_dentry_operations =
1648 {
1649         .d_revalidate   = pid_revalidate,
1650         .d_delete       = pid_delete_dentry,
1651 };
1652
1653 /* Lookups */
1654
1655 /*
1656  * Fill a directory entry.
1657  *
1658  * If possible create the dcache entry and derive our inode number and
1659  * file type from dcache entry.
1660  *
1661  * Since all of the proc inode numbers are dynamically generated, the inode
1662  * numbers do not exist until the inode is cache.  This means creating the
1663  * the dcache entry in readdir is necessary to keep the inode numbers
1664  * reported by readdir in sync with the inode numbers reported
1665  * by stat.
1666  */
1667 int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1668         const char *name, int len,
1669         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1670 {
1671         struct dentry *child, *dir = filp->f_path.dentry;
1672         struct inode *inode;
1673         struct qstr qname;
1674         ino_t ino = 0;
1675         unsigned type = DT_UNKNOWN;
1676
1677         qname.name = name;
1678         qname.len  = len;
1679         qname.hash = full_name_hash(name, len);
1680
1681         child = d_lookup(dir, &qname);
1682         if (!child) {
1683                 struct dentry *new;
1684                 new = d_alloc(dir, &qname);
1685                 if (new) {
1686                         child = instantiate(dir->d_inode, new, task, ptr);
1687                         if (child)
1688                                 dput(new);
1689                         else
1690                                 child = new;
1691                 }
1692         }
1693         if (!child || IS_ERR(child) || !child->d_inode)
1694                 goto end_instantiate;
1695         inode = child->d_inode;
1696         if (inode) {
1697                 ino = inode->i_ino;
1698                 type = inode->i_mode >> 12;
1699         }
1700         dput(child);
1701 end_instantiate:
1702         if (!ino)
1703                 ino = find_inode_number(dir, &qname);
1704         if (!ino)
1705                 ino = 1;
1706         return filldir(dirent, name, len, filp->f_pos, ino, type);
1707 }
1708
1709 static unsigned name_to_int(struct dentry *dentry)
1710 {
1711         const char *name = dentry->d_name.name;
1712         int len = dentry->d_name.len;
1713         unsigned n = 0;
1714
1715         if (len > 1 && *name == '0')
1716                 goto out;
1717         while (len-- > 0) {
1718                 unsigned c = *name++ - '0';
1719                 if (c > 9)
1720                         goto out;
1721                 if (n >= (~0U-9)/10)
1722                         goto out;
1723                 n *= 10;
1724                 n += c;
1725         }
1726         return n;
1727 out:
1728         return ~0U;
1729 }
1730
1731 #define PROC_FDINFO_MAX 64
1732
1733 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1734 {
1735         struct task_struct *task = get_proc_task(inode);
1736         struct files_struct *files = NULL;
1737         struct file *file;
1738         int fd = proc_fd(inode);
1739
1740         if (task) {
1741                 files = get_files_struct(task);
1742                 put_task_struct(task);
1743         }
1744         if (files) {
1745                 /*
1746                  * We are not taking a ref to the file structure, so we must
1747                  * hold ->file_lock.
1748                  */
1749                 spin_lock(&files->file_lock);
1750                 file = fcheck_files(files, fd);
1751                 if (file) {
1752                         unsigned int f_flags;
1753                         struct fdtable *fdt;
1754
1755                         fdt = files_fdtable(files);
1756                         f_flags = file->f_flags & ~O_CLOEXEC;
1757                         if (close_on_exec(fd, fdt))
1758                                 f_flags |= O_CLOEXEC;
1759
1760                         if (path) {
1761                                 *path = file->f_path;
1762                                 path_get(&file->f_path);
1763                         }
1764                         if (info)
1765                                 snprintf(info, PROC_FDINFO_MAX,
1766                                          "pos:\t%lli\n"
1767                                          "flags:\t0%o\n",
1768                                          (long long) file->f_pos,
1769                                          f_flags);
1770                         spin_unlock(&files->file_lock);
1771                         put_files_struct(files);
1772                         return 0;
1773                 }
1774                 spin_unlock(&files->file_lock);
1775                 put_files_struct(files);
1776         }
1777         return -ENOENT;
1778 }
1779
1780 static int proc_fd_link(struct dentry *dentry, struct path *path)
1781 {
1782         return proc_fd_info(dentry->d_inode, path, NULL);
1783 }
1784
1785 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1786 {
1787         struct inode *inode;
1788         struct task_struct *task;
1789         int fd;
1790         struct files_struct *files;
1791         const struct cred *cred;
1792
1793         if (nd && nd->flags & LOOKUP_RCU)
1794                 return -ECHILD;
1795
1796         inode = dentry->d_inode;
1797         task = get_proc_task(inode);
1798         fd = proc_fd(inode);
1799
1800         if (task) {
1801                 files = get_files_struct(task);
1802                 if (files) {
1803                         rcu_read_lock();
1804                         if (fcheck_files(files, fd)) {
1805                                 rcu_read_unlock();
1806                                 put_files_struct(files);
1807                                 if (task_dumpable(task)) {
1808                                         rcu_read_lock();
1809                                         cred = __task_cred(task);
1810                                         inode->i_uid = cred->euid;
1811                                         inode->i_gid = cred->egid;
1812                                         rcu_read_unlock();
1813                                 } else {
1814                                         inode->i_uid = GLOBAL_ROOT_UID;
1815                                         inode->i_gid = GLOBAL_ROOT_GID;
1816                                 }
1817                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1818                                 security_task_to_inode(task, inode);
1819                                 put_task_struct(task);
1820                                 return 1;
1821                         }
1822                         rcu_read_unlock();
1823                         put_files_struct(files);
1824                 }
1825                 put_task_struct(task);
1826         }
1827         d_drop(dentry);
1828         return 0;
1829 }
1830
1831 static const struct dentry_operations tid_fd_dentry_operations =
1832 {
1833         .d_revalidate   = tid_fd_revalidate,
1834         .d_delete       = pid_delete_dentry,
1835 };
1836
1837 static struct dentry *proc_fd_instantiate(struct inode *dir,
1838         struct dentry *dentry, struct task_struct *task, const void *ptr)
1839 {
1840         unsigned fd = *(const unsigned *)ptr;
1841         struct file *file;
1842         struct files_struct *files;
1843         struct inode *inode;
1844         struct proc_inode *ei;
1845         struct dentry *error = ERR_PTR(-ENOENT);
1846
1847         inode = proc_pid_make_inode(dir->i_sb, task);
1848         if (!inode)
1849                 goto out;
1850         ei = PROC_I(inode);
1851         ei->fd = fd;
1852         files = get_files_struct(task);
1853         if (!files)
1854                 goto out_iput;
1855         inode->i_mode = S_IFLNK;
1856
1857         /*
1858          * We are not taking a ref to the file structure, so we must
1859          * hold ->file_lock.
1860          */
1861         spin_lock(&files->file_lock);
1862         file = fcheck_files(files, fd);
1863         if (!file)
1864                 goto out_unlock;
1865         if (file->f_mode & FMODE_READ)
1866                 inode->i_mode |= S_IRUSR | S_IXUSR;
1867         if (file->f_mode & FMODE_WRITE)
1868                 inode->i_mode |= S_IWUSR | S_IXUSR;
1869         spin_unlock(&files->file_lock);
1870         put_files_struct(files);
1871
1872         inode->i_op = &proc_pid_link_inode_operations;
1873         inode->i_size = 64;
1874         ei->op.proc_get_link = proc_fd_link;
1875         d_set_d_op(dentry, &tid_fd_dentry_operations);
1876         d_add(dentry, inode);
1877         /* Close the race of the process dying before we return the dentry */
1878         if (tid_fd_revalidate(dentry, NULL))
1879                 error = NULL;
1880
1881  out:
1882         return error;
1883 out_unlock:
1884         spin_unlock(&files->file_lock);
1885         put_files_struct(files);
1886 out_iput:
1887         iput(inode);
1888         goto out;
1889 }
1890
1891 static struct dentry *proc_lookupfd_common(struct inode *dir,
1892                                            struct dentry *dentry,
1893                                            instantiate_t instantiate)
1894 {
1895         struct task_struct *task = get_proc_task(dir);
1896         unsigned fd = name_to_int(dentry);
1897         struct dentry *result = ERR_PTR(-ENOENT);
1898
1899         if (!task)
1900                 goto out_no_task;
1901         if (fd == ~0U)
1902                 goto out;
1903
1904         result = instantiate(dir, dentry, task, &fd);
1905 out:
1906         put_task_struct(task);
1907 out_no_task:
1908         return result;
1909 }
1910
1911 static int proc_readfd_common(struct file * filp, void * dirent,
1912                               filldir_t filldir, instantiate_t instantiate)
1913 {
1914         struct dentry *dentry = filp->f_path.dentry;
1915         struct inode *inode = dentry->d_inode;
1916         struct task_struct *p = get_proc_task(inode);
1917         unsigned int fd, ino;
1918         int retval;
1919         struct files_struct * files;
1920
1921         retval = -ENOENT;
1922         if (!p)
1923                 goto out_no_task;
1924         retval = 0;
1925
1926         fd = filp->f_pos;
1927         switch (fd) {
1928                 case 0:
1929                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1930                                 goto out;
1931                         filp->f_pos++;
1932                 case 1:
1933                         ino = parent_ino(dentry);
1934                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1935                                 goto out;
1936                         filp->f_pos++;
1937                 default:
1938                         files = get_files_struct(p);
1939                         if (!files)
1940                                 goto out;
1941                         rcu_read_lock();
1942                         for (fd = filp->f_pos-2;
1943                              fd < files_fdtable(files)->max_fds;
1944                              fd++, filp->f_pos++) {
1945                                 char name[PROC_NUMBUF];
1946                                 int len;
1947
1948                                 if (!fcheck_files(files, fd))
1949                                         continue;
1950                                 rcu_read_unlock();
1951
1952                                 len = snprintf(name, sizeof(name), "%d", fd);
1953                                 if (proc_fill_cache(filp, dirent, filldir,
1954                                                     name, len, instantiate,
1955                                                     p, &fd) < 0) {
1956                                         rcu_read_lock();
1957                                         break;
1958                                 }
1959                                 rcu_read_lock();
1960                         }
1961                         rcu_read_unlock();
1962                         put_files_struct(files);
1963         }
1964 out:
1965         put_task_struct(p);
1966 out_no_task:
1967         return retval;
1968 }
1969
1970 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1971                                     struct nameidata *nd)
1972 {
1973         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1974 }
1975
1976 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1977 {
1978         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1979 }
1980
1981 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1982                                       size_t len, loff_t *ppos)
1983 {
1984         char tmp[PROC_FDINFO_MAX];
1985         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
1986         if (!err)
1987                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1988         return err;
1989 }
1990
1991 static const struct file_operations proc_fdinfo_file_operations = {
1992         .open           = nonseekable_open,
1993         .read           = proc_fdinfo_read,
1994         .llseek         = no_llseek,
1995 };
1996
1997 static const struct file_operations proc_fd_operations = {
1998         .read           = generic_read_dir,
1999         .readdir        = proc_readfd,
2000         .llseek         = default_llseek,
2001 };
2002
2003 #ifdef CONFIG_CHECKPOINT_RESTORE
2004
2005 /*
2006  * dname_to_vma_addr - maps a dentry name into two unsigned longs
2007  * which represent vma start and end addresses.
2008  */
2009 static int dname_to_vma_addr(struct dentry *dentry,
2010                              unsigned long *start, unsigned long *end)
2011 {
2012         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
2013                 return -EINVAL;
2014
2015         return 0;
2016 }
2017
2018 static int map_files_d_revalidate(struct dentry *dentry, struct nameidata *nd)
2019 {
2020         unsigned long vm_start, vm_end;
2021         bool exact_vma_exists = false;
2022         struct mm_struct *mm = NULL;
2023         struct task_struct *task;
2024         const struct cred *cred;
2025         struct inode *inode;
2026         int status = 0;
2027
2028         if (nd && nd->flags & LOOKUP_RCU)
2029                 return -ECHILD;
2030
2031         if (!capable(CAP_SYS_ADMIN)) {
2032                 status = -EACCES;
2033                 goto out_notask;
2034         }
2035
2036         inode = dentry->d_inode;
2037         task = get_proc_task(inode);
2038         if (!task)
2039                 goto out_notask;
2040
2041         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2042                 goto out;
2043
2044         mm = get_task_mm(task);
2045         if (!mm)
2046                 goto out;
2047
2048         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
2049                 down_read(&mm->mmap_sem);
2050                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
2051                 up_read(&mm->mmap_sem);
2052         }
2053
2054         mmput(mm);
2055
2056         if (exact_vma_exists) {
2057                 if (task_dumpable(task)) {
2058                         rcu_read_lock();
2059                         cred = __task_cred(task);
2060                         inode->i_uid = cred->euid;
2061                         inode->i_gid = cred->egid;
2062                         rcu_read_unlock();
2063                 } else {
2064                         inode->i_uid = GLOBAL_ROOT_UID;
2065                         inode->i_gid = GLOBAL_ROOT_GID;
2066                 }
2067                 security_task_to_inode(task, inode);
2068                 status = 1;
2069         }
2070
2071 out:
2072         put_task_struct(task);
2073
2074 out_notask:
2075         if (status <= 0)
2076                 d_drop(dentry);
2077
2078         return status;
2079 }
2080
2081 static const struct dentry_operations tid_map_files_dentry_operations = {
2082         .d_revalidate   = map_files_d_revalidate,
2083         .d_delete       = pid_delete_dentry,
2084 };
2085
2086 static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
2087 {
2088         unsigned long vm_start, vm_end;
2089         struct vm_area_struct *vma;
2090         struct task_struct *task;
2091         struct mm_struct *mm;
2092         int rc;
2093
2094         rc = -ENOENT;
2095         task = get_proc_task(dentry->d_inode);
2096         if (!task)
2097                 goto out;
2098
2099         mm = get_task_mm(task);
2100         put_task_struct(task);
2101         if (!mm)
2102                 goto out;
2103
2104         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2105         if (rc)
2106                 goto out_mmput;
2107
2108         down_read(&mm->mmap_sem);
2109         vma = find_exact_vma(mm, vm_start, vm_end);
2110         if (vma && vma->vm_file) {
2111                 *path = vma->vm_file->f_path;
2112                 path_get(path);
2113                 rc = 0;
2114         }
2115         up_read(&mm->mmap_sem);
2116
2117 out_mmput:
2118         mmput(mm);
2119 out:
2120         return rc;
2121 }
2122
2123 struct map_files_info {
2124         struct file     *file;
2125         unsigned long   len;
2126         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
2127 };
2128
2129 static struct dentry *
2130 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
2131                            struct task_struct *task, const void *ptr)
2132 {
2133         const struct file *file = ptr;
2134         struct proc_inode *ei;
2135         struct inode *inode;
2136
2137         if (!file)
2138                 return ERR_PTR(-ENOENT);
2139
2140         inode = proc_pid_make_inode(dir->i_sb, task);
2141         if (!inode)
2142                 return ERR_PTR(-ENOENT);
2143
2144         ei = PROC_I(inode);
2145         ei->op.proc_get_link = proc_map_files_get_link;
2146
2147         inode->i_op = &proc_pid_link_inode_operations;
2148         inode->i_size = 64;
2149         inode->i_mode = S_IFLNK;
2150
2151         if (file->f_mode & FMODE_READ)
2152                 inode->i_mode |= S_IRUSR;
2153         if (file->f_mode & FMODE_WRITE)
2154                 inode->i_mode |= S_IWUSR;
2155
2156         d_set_d_op(dentry, &tid_map_files_dentry_operations);
2157         d_add(dentry, inode);
2158
2159         return NULL;
2160 }
2161
2162 static struct dentry *proc_map_files_lookup(struct inode *dir,
2163                 struct dentry *dentry, struct nameidata *nd)
2164 {
2165         unsigned long vm_start, vm_end;
2166         struct vm_area_struct *vma;
2167         struct task_struct *task;
2168         struct dentry *result;
2169         struct mm_struct *mm;
2170
2171         result = ERR_PTR(-EACCES);
2172         if (!capable(CAP_SYS_ADMIN))
2173                 goto out;
2174
2175         result = ERR_PTR(-ENOENT);
2176         task = get_proc_task(dir);
2177         if (!task)
2178                 goto out;
2179
2180         result = ERR_PTR(-EACCES);
2181         if (lock_trace(task))
2182                 goto out_put_task;
2183
2184         result = ERR_PTR(-ENOENT);
2185         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2186                 goto out_unlock;
2187
2188         mm = get_task_mm(task);
2189         if (!mm)
2190                 goto out_unlock;
2191
2192         down_read(&mm->mmap_sem);
2193         vma = find_exact_vma(mm, vm_start, vm_end);
2194         if (!vma)
2195                 goto out_no_vma;
2196
2197         result = proc_map_files_instantiate(dir, dentry, task, vma->vm_file);
2198
2199 out_no_vma:
2200         up_read(&mm->mmap_sem);
2201         mmput(mm);
2202 out_unlock:
2203         unlock_trace(task);
2204 out_put_task:
2205         put_task_struct(task);
2206 out:
2207         return result;
2208 }
2209
2210 static const struct inode_operations proc_map_files_inode_operations = {
2211         .lookup         = proc_map_files_lookup,
2212         .permission     = proc_fd_permission,
2213         .setattr        = proc_setattr,
2214 };
2215
2216 static int
2217 proc_map_files_readdir(struct file *filp, void *dirent, filldir_t filldir)
2218 {
2219         struct dentry *dentry = filp->f_path.dentry;
2220         struct inode *inode = dentry->d_inode;
2221         struct vm_area_struct *vma;
2222         struct task_struct *task;
2223         struct mm_struct *mm;
2224         ino_t ino;
2225         int ret;
2226
2227         ret = -EACCES;
2228         if (!capable(CAP_SYS_ADMIN))
2229                 goto out;
2230
2231         ret = -ENOENT;
2232         task = get_proc_task(inode);
2233         if (!task)
2234                 goto out;
2235
2236         ret = -EACCES;
2237         if (lock_trace(task))
2238                 goto out_put_task;
2239
2240         ret = 0;
2241         switch (filp->f_pos) {
2242         case 0:
2243                 ino = inode->i_ino;
2244                 if (filldir(dirent, ".", 1, 0, ino, DT_DIR) < 0)
2245                         goto out_unlock;
2246                 filp->f_pos++;
2247         case 1:
2248                 ino = parent_ino(dentry);
2249                 if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
2250                         goto out_unlock;
2251                 filp->f_pos++;
2252         default:
2253         {
2254                 unsigned long nr_files, pos, i;
2255                 struct flex_array *fa = NULL;
2256                 struct map_files_info info;
2257                 struct map_files_info *p;
2258
2259                 mm = get_task_mm(task);
2260                 if (!mm)
2261                         goto out_unlock;
2262                 down_read(&mm->mmap_sem);
2263
2264                 nr_files = 0;
2265
2266                 /*
2267                  * We need two passes here:
2268                  *
2269                  *  1) Collect vmas of mapped files with mmap_sem taken
2270                  *  2) Release mmap_sem and instantiate entries
2271                  *
2272                  * otherwise we get lockdep complained, since filldir()
2273                  * routine might require mmap_sem taken in might_fault().
2274                  */
2275
2276                 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2277                         if (vma->vm_file && ++pos > filp->f_pos)
2278                                 nr_files++;
2279                 }
2280
2281                 if (nr_files) {
2282                         fa = flex_array_alloc(sizeof(info), nr_files,
2283                                                 GFP_KERNEL);
2284                         if (!fa || flex_array_prealloc(fa, 0, nr_files,
2285                                                         GFP_KERNEL)) {
2286                                 ret = -ENOMEM;
2287                                 if (fa)
2288                                         flex_array_free(fa);
2289                                 up_read(&mm->mmap_sem);
2290                                 mmput(mm);
2291                                 goto out_unlock;
2292                         }
2293                         for (i = 0, vma = mm->mmap, pos = 2; vma;
2294                                         vma = vma->vm_next) {
2295                                 if (!vma->vm_file)
2296                                         continue;
2297                                 if (++pos <= filp->f_pos)
2298                                         continue;
2299
2300                                 get_file(vma->vm_file);
2301                                 info.file = vma->vm_file;
2302                                 info.len = snprintf(info.name,
2303                                                 sizeof(info.name), "%lx-%lx",
2304                                                 vma->vm_start, vma->vm_end);
2305                                 if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2306                                         BUG();
2307                         }
2308                 }
2309                 up_read(&mm->mmap_sem);
2310
2311                 for (i = 0; i < nr_files; i++) {
2312                         p = flex_array_get(fa, i);
2313                         ret = proc_fill_cache(filp, dirent, filldir,
2314                                               p->name, p->len,
2315                                               proc_map_files_instantiate,
2316                                               task, p->file);
2317                         if (ret)
2318                                 break;
2319                         filp->f_pos++;
2320                         fput(p->file);
2321                 }
2322                 for (; i < nr_files; i++) {
2323                         /*
2324                          * In case of error don't forget
2325                          * to put rest of file refs.
2326                          */
2327                         p = flex_array_get(fa, i);
2328                         fput(p->file);
2329                 }
2330                 if (fa)
2331                         flex_array_free(fa);
2332                 mmput(mm);
2333         }
2334         }
2335
2336 out_unlock:
2337         unlock_trace(task);
2338 out_put_task:
2339         put_task_struct(task);
2340 out:
2341         return ret;
2342 }
2343
2344 static const struct file_operations proc_map_files_operations = {
2345         .read           = generic_read_dir,
2346         .readdir        = proc_map_files_readdir,
2347         .llseek         = default_llseek,
2348 };
2349
2350 #endif /* CONFIG_CHECKPOINT_RESTORE */
2351
2352 /*
2353  * /proc/pid/fd needs a special permission handler so that a process can still
2354  * access /proc/self/fd after it has executed a setuid().
2355  */
2356 static int proc_fd_permission(struct inode *inode, int mask)
2357 {
2358         int rv = generic_permission(inode, mask);
2359         if (rv == 0)
2360                 return 0;
2361         if (task_pid(current) == proc_pid(inode))
2362                 rv = 0;
2363         return rv;
2364 }
2365
2366 /*
2367  * proc directories can do almost nothing..
2368  */
2369 static const struct inode_operations proc_fd_inode_operations = {
2370         .lookup         = proc_lookupfd,
2371         .permission     = proc_fd_permission,
2372         .setattr        = proc_setattr,
2373 };
2374
2375 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
2376         struct dentry *dentry, struct task_struct *task, const void *ptr)
2377 {
2378         unsigned fd = *(unsigned *)ptr;
2379         struct inode *inode;
2380         struct proc_inode *ei;
2381         struct dentry *error = ERR_PTR(-ENOENT);
2382
2383         inode = proc_pid_make_inode(dir->i_sb, task);
2384         if (!inode)
2385                 goto out;
2386         ei = PROC_I(inode);
2387         ei->fd = fd;
2388         inode->i_mode = S_IFREG | S_IRUSR;
2389         inode->i_fop = &proc_fdinfo_file_operations;
2390         d_set_d_op(dentry, &tid_fd_dentry_operations);
2391         d_add(dentry, inode);
2392         /* Close the race of the process dying before we return the dentry */
2393         if (tid_fd_revalidate(dentry, NULL))
2394                 error = NULL;
2395
2396  out:
2397         return error;
2398 }
2399
2400 static struct dentry *proc_lookupfdinfo(struct inode *dir,
2401                                         struct dentry *dentry,
2402                                         struct nameidata *nd)
2403 {
2404         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
2405 }
2406
2407 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
2408 {
2409         return proc_readfd_common(filp, dirent, filldir,
2410                                   proc_fdinfo_instantiate);
2411 }
2412
2413 static const struct file_operations proc_fdinfo_operations = {
2414         .read           = generic_read_dir,
2415         .readdir        = proc_readfdinfo,
2416         .llseek         = default_llseek,
2417 };
2418
2419 /*
2420  * proc directories can do almost nothing..
2421  */
2422 static const struct inode_operations proc_fdinfo_inode_operations = {
2423         .lookup         = proc_lookupfdinfo,
2424         .setattr        = proc_setattr,
2425 };
2426
2427
2428 static struct dentry *proc_pident_instantiate(struct inode *dir,
2429         struct dentry *dentry, struct task_struct *task, const void *ptr)
2430 {
2431         const struct pid_entry *p = ptr;
2432         struct inode *inode;
2433         struct proc_inode *ei;
2434         struct dentry *error = ERR_PTR(-ENOENT);
2435
2436         inode = proc_pid_make_inode(dir->i_sb, task);
2437         if (!inode)
2438                 goto out;
2439
2440         ei = PROC_I(inode);
2441         inode->i_mode = p->mode;
2442         if (S_ISDIR(inode->i_mode))
2443                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2444         if (p->iop)
2445                 inode->i_op = p->iop;
2446         if (p->fop)
2447                 inode->i_fop = p->fop;
2448         ei->op = p->op;
2449         d_set_d_op(dentry, &pid_dentry_operations);
2450         d_add(dentry, inode);
2451         /* Close the race of the process dying before we return the dentry */
2452         if (pid_revalidate(dentry, NULL))
2453                 error = NULL;
2454 out:
2455         return error;
2456 }
2457
2458 static struct dentry *proc_pident_lookup(struct inode *dir,
2459                                          struct dentry *dentry,
2460                                          const struct pid_entry *ents,
2461                                          unsigned int nents)
2462 {
2463         struct dentry *error;
2464         struct task_struct *task = get_proc_task(dir);
2465         const struct pid_entry *p, *last;
2466
2467         error = ERR_PTR(-ENOENT);
2468
2469         if (!task)
2470                 goto out_no_task;
2471
2472         /*
2473          * Yes, it does not scale. And it should not. Don't add
2474          * new entries into /proc/<tgid>/ without very good reasons.
2475          */
2476         last = &ents[nents - 1];
2477         for (p = ents; p <= last; p++) {
2478                 if (p->len != dentry->d_name.len)
2479                         continue;
2480                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2481                         break;
2482         }
2483         if (p > last)
2484                 goto out;
2485
2486         error = proc_pident_instantiate(dir, dentry, task, p);
2487 out:
2488         put_task_struct(task);
2489 out_no_task:
2490         return error;
2491 }
2492
2493 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2494         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2495 {
2496         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2497                                 proc_pident_instantiate, task, p);
2498 }
2499
2500 static int proc_pident_readdir(struct file *filp,
2501                 void *dirent, filldir_t filldir,
2502                 const struct pid_entry *ents, unsigned int nents)
2503 {
2504         int i;
2505         struct dentry *dentry = filp->f_path.dentry;
2506         struct inode *inode = dentry->d_inode;
2507         struct task_struct *task = get_proc_task(inode);
2508         const struct pid_entry *p, *last;
2509         ino_t ino;
2510         int ret;
2511
2512         ret = -ENOENT;
2513         if (!task)
2514                 goto out_no_task;
2515
2516         ret = 0;
2517         i = filp->f_pos;
2518         switch (i) {
2519         case 0:
2520                 ino = inode->i_ino;
2521                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2522                         goto out;
2523                 i++;
2524                 filp->f_pos++;
2525                 /* fall through */
2526         case 1:
2527                 ino = parent_ino(dentry);
2528                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2529                         goto out;
2530                 i++;
2531                 filp->f_pos++;
2532                 /* fall through */
2533         default:
2534                 i -= 2;
2535                 if (i >= nents) {
2536                         ret = 1;
2537                         goto out;
2538                 }
2539                 p = ents + i;
2540                 last = &ents[nents - 1];
2541                 while (p <= last) {
2542                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2543                                 goto out;
2544                         filp->f_pos++;
2545                         p++;
2546                 }
2547         }
2548
2549         ret = 1;
2550 out:
2551         put_task_struct(task);
2552 out_no_task:
2553         return ret;
2554 }
2555
2556 #ifdef CONFIG_SECURITY
2557 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2558                                   size_t count, loff_t *ppos)
2559 {
2560         struct inode * inode = file->f_path.dentry->d_inode;
2561         char *p = NULL;
2562         ssize_t length;
2563         struct task_struct *task = get_proc_task(inode);
2564
2565         if (!task)
2566                 return -ESRCH;
2567
2568         length = security_getprocattr(task,
2569                                       (char*)file->f_path.dentry->d_name.name,
2570                                       &p);
2571         put_task_struct(task);
2572         if (length > 0)
2573                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2574         kfree(p);
2575         return length;
2576 }
2577
2578 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2579                                    size_t count, loff_t *ppos)
2580 {
2581         struct inode * inode = file->f_path.dentry->d_inode;
2582         char *page;
2583         ssize_t length;
2584         struct task_struct *task = get_proc_task(inode);
2585
2586         length = -ESRCH;
2587         if (!task)
2588                 goto out_no_task;
2589         if (count > PAGE_SIZE)
2590                 count = PAGE_SIZE;
2591
2592         /* No partial writes. */
2593         length = -EINVAL;
2594         if (*ppos != 0)
2595                 goto out;
2596
2597         length = -ENOMEM;
2598         page = (char*)__get_free_page(GFP_TEMPORARY);
2599         if (!page)
2600                 goto out;
2601
2602         length = -EFAULT;
2603         if (copy_from_user(page, buf, count))
2604                 goto out_free;
2605
2606         /* Guard against adverse ptrace interaction */
2607         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2608         if (length < 0)
2609                 goto out_free;
2610
2611         length = security_setprocattr(task,
2612                                       (char*)file->f_path.dentry->d_name.name,
2613                                       (void*)page, count);
2614         mutex_unlock(&task->signal->cred_guard_mutex);
2615 out_free:
2616         free_page((unsigned long) page);
2617 out:
2618         put_task_struct(task);
2619 out_no_task:
2620         return length;
2621 }
2622
2623 static const struct file_operations proc_pid_attr_operations = {
2624         .read           = proc_pid_attr_read,
2625         .write          = proc_pid_attr_write,
2626         .llseek         = generic_file_llseek,
2627 };
2628
2629 static const struct pid_entry attr_dir_stuff[] = {
2630         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2631         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2632         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2633         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2634         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2635         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2636 };
2637
2638 static int proc_attr_dir_readdir(struct file * filp,
2639                              void * dirent, filldir_t filldir)
2640 {
2641         return proc_pident_readdir(filp,dirent,filldir,
2642                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2643 }
2644
2645 static const struct file_operations proc_attr_dir_operations = {
2646         .read           = generic_read_dir,
2647         .readdir        = proc_attr_dir_readdir,
2648         .llseek         = default_llseek,
2649 };
2650
2651 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2652                                 struct dentry *dentry, struct nameidata *nd)
2653 {
2654         return proc_pident_lookup(dir, dentry,
2655                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2656 }
2657
2658 static const struct inode_operations proc_attr_dir_inode_operations = {
2659         .lookup         = proc_attr_dir_lookup,
2660         .getattr        = pid_getattr,
2661         .setattr        = proc_setattr,
2662 };
2663
2664 #endif
2665
2666 #ifdef CONFIG_ELF_CORE
2667 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2668                                          size_t count, loff_t *ppos)
2669 {
2670         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2671         struct mm_struct *mm;
2672         char buffer[PROC_NUMBUF];
2673         size_t len;
2674         int ret;
2675
2676         if (!task)
2677                 return -ESRCH;
2678
2679         ret = 0;
2680         mm = get_task_mm(task);
2681         if (mm) {
2682                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2683                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2684                                 MMF_DUMP_FILTER_SHIFT));
2685                 mmput(mm);
2686                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2687         }
2688
2689         put_task_struct(task);
2690
2691         return ret;
2692 }
2693
2694 static ssize_t proc_coredump_filter_write(struct file *file,
2695                                           const char __user *buf,
2696                                           size_t count,
2697                                           loff_t *ppos)
2698 {
2699         struct task_struct *task;
2700         struct mm_struct *mm;
2701         char buffer[PROC_NUMBUF], *end;
2702         unsigned int val;
2703         int ret;
2704         int i;
2705         unsigned long mask;
2706
2707         ret = -EFAULT;
2708         memset(buffer, 0, sizeof(buffer));
2709         if (count > sizeof(buffer) - 1)
2710                 count = sizeof(buffer) - 1;
2711         if (copy_from_user(buffer, buf, count))
2712                 goto out_no_task;
2713
2714         ret = -EINVAL;
2715         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2716         if (*end == '\n')
2717                 end++;
2718         if (end - buffer == 0)
2719                 goto out_no_task;
2720
2721         ret = -ESRCH;
2722         task = get_proc_task(file->f_dentry->d_inode);
2723         if (!task)
2724                 goto out_no_task;
2725
2726         ret = end - buffer;
2727         mm = get_task_mm(task);
2728         if (!mm)
2729                 goto out_no_mm;
2730
2731         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2732                 if (val & mask)
2733                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2734                 else
2735                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2736         }
2737
2738         mmput(mm);
2739  out_no_mm:
2740         put_task_struct(task);
2741  out_no_task:
2742         return ret;
2743 }
2744
2745 static const struct file_operations proc_coredump_filter_operations = {
2746         .read           = proc_coredump_filter_read,
2747         .write          = proc_coredump_filter_write,
2748         .llseek         = generic_file_llseek,
2749 };
2750 #endif
2751
2752 /*
2753  * /proc/self:
2754  */
2755 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2756                               int buflen)
2757 {
2758         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2759         pid_t tgid = task_tgid_nr_ns(current, ns);
2760         char tmp[PROC_NUMBUF];
2761         if (!tgid)
2762                 return -ENOENT;
2763         sprintf(tmp, "%d", tgid);
2764         return vfs_readlink(dentry,buffer,buflen,tmp);
2765 }
2766
2767 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2768 {
2769         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2770         pid_t tgid = task_tgid_nr_ns(current, ns);
2771         char *name = ERR_PTR(-ENOENT);
2772         if (tgid) {
2773                 name = __getname();
2774                 if (!name)
2775                         name = ERR_PTR(-ENOMEM);
2776                 else
2777                         sprintf(name, "%d", tgid);
2778         }
2779         nd_set_link(nd, name);
2780         return NULL;
2781 }
2782
2783 static void proc_self_put_link(struct dentry *dentry, struct nameidata *nd,
2784                                 void *cookie)
2785 {
2786         char *s = nd_get_link(nd);
2787         if (!IS_ERR(s))
2788                 __putname(s);
2789 }
2790
2791 static const struct inode_operations proc_self_inode_operations = {
2792         .readlink       = proc_self_readlink,
2793         .follow_link    = proc_self_follow_link,
2794         .put_link       = proc_self_put_link,
2795 };
2796
2797 /*
2798  * proc base
2799  *
2800  * These are the directory entries in the root directory of /proc
2801  * that properly belong to the /proc filesystem, as they describe
2802  * describe something that is process related.
2803  */
2804 static const struct pid_entry proc_base_stuff[] = {
2805         NOD("self", S_IFLNK|S_IRWXUGO,
2806                 &proc_self_inode_operations, NULL, {}),
2807 };
2808
2809 static struct dentry *proc_base_instantiate(struct inode *dir,
2810         struct dentry *dentry, struct task_struct *task, const void *ptr)
2811 {
2812         const struct pid_entry *p = ptr;
2813         struct inode *inode;
2814         struct proc_inode *ei;
2815         struct dentry *error;
2816
2817         /* Allocate the inode */
2818         error = ERR_PTR(-ENOMEM);
2819         inode = new_inode(dir->i_sb);
2820         if (!inode)
2821                 goto out;
2822
2823         /* Initialize the inode */
2824         ei = PROC_I(inode);
2825         inode->i_ino = get_next_ino();
2826         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2827
2828         /*
2829          * grab the reference to the task.
2830          */
2831         ei->pid = get_task_pid(task, PIDTYPE_PID);
2832         if (!ei->pid)
2833                 goto out_iput;
2834
2835         inode->i_mode = p->mode;
2836         if (S_ISDIR(inode->i_mode))
2837                 set_nlink(inode, 2);
2838         if (S_ISLNK(inode->i_mode))
2839                 inode->i_size = 64;
2840         if (p->iop)
2841                 inode->i_op = p->iop;
2842         if (p->fop)
2843                 inode->i_fop = p->fop;
2844         ei->op = p->op;
2845         d_add(dentry, inode);
2846         error = NULL;
2847 out:
2848         return error;
2849 out_iput:
2850         iput(inode);
2851         goto out;
2852 }
2853
2854 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2855 {
2856         struct dentry *error;
2857         struct task_struct *task = get_proc_task(dir);
2858         const struct pid_entry *p, *last;
2859
2860         error = ERR_PTR(-ENOENT);
2861
2862         if (!task)
2863                 goto out_no_task;
2864
2865         /* Lookup the directory entry */
2866         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2867         for (p = proc_base_stuff; p <= last; p++) {
2868                 if (p->len != dentry->d_name.len)
2869                         continue;
2870                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2871                         break;
2872         }
2873         if (p > last)
2874                 goto out;
2875
2876         error = proc_base_instantiate(dir, dentry, task, p);
2877
2878 out:
2879         put_task_struct(task);
2880 out_no_task:
2881         return error;
2882 }
2883
2884 static int proc_base_fill_cache(struct file *filp, void *dirent,
2885         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2886 {
2887         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2888                                 proc_base_instantiate, task, p);
2889 }
2890
2891 #ifdef CONFIG_TASK_IO_ACCOUNTING
2892 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2893 {
2894         struct task_io_accounting acct = task->ioac;
2895         unsigned long flags;
2896         int result;
2897
2898         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2899         if (result)
2900                 return result;
2901
2902         if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2903                 result = -EACCES;
2904                 goto out_unlock;
2905         }
2906
2907         if (whole && lock_task_sighand(task, &flags)) {
2908                 struct task_struct *t = task;
2909
2910                 task_io_accounting_add(&acct, &task->signal->ioac);
2911                 while_each_thread(task, t)
2912                         task_io_accounting_add(&acct, &t->ioac);
2913
2914                 unlock_task_sighand(task, &flags);
2915         }
2916         result = sprintf(buffer,
2917                         "rchar: %llu\n"
2918                         "wchar: %llu\n"
2919                         "syscr: %llu\n"
2920                         "syscw: %llu\n"
2921                         "read_bytes: %llu\n"
2922                         "write_bytes: %llu\n"
2923                         "cancelled_write_bytes: %llu\n",
2924                         (unsigned long long)acct.rchar,
2925                         (unsigned long long)acct.wchar,
2926                         (unsigned long long)acct.syscr,
2927                         (unsigned long long)acct.syscw,
2928                         (unsigned long long)acct.read_bytes,
2929                         (unsigned long long)acct.write_bytes,
2930                         (unsigned long long)acct.cancelled_write_bytes);
2931 out_unlock:
2932         mutex_unlock(&task->signal->cred_guard_mutex);
2933         return result;
2934 }
2935
2936 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2937 {
2938         return do_io_accounting(task, buffer, 0);
2939 }
2940
2941 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2942 {
2943         return do_io_accounting(task, buffer, 1);
2944 }
2945 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2946
2947 #ifdef CONFIG_USER_NS
2948 static int proc_id_map_open(struct inode *inode, struct file *file,
2949         struct seq_operations *seq_ops)
2950 {
2951         struct user_namespace *ns = NULL;
2952         struct task_struct *task;
2953         struct seq_file *seq;
2954         int ret = -EINVAL;
2955
2956         task = get_proc_task(inode);
2957         if (task) {
2958                 rcu_read_lock();
2959                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2960                 rcu_read_unlock();
2961                 put_task_struct(task);
2962         }
2963         if (!ns)
2964                 goto err;
2965
2966         ret = seq_open(file, seq_ops);
2967         if (ret)
2968                 goto err_put_ns;
2969
2970         seq = file->private_data;
2971         seq->private = ns;
2972
2973         return 0;
2974 err_put_ns:
2975         put_user_ns(ns);
2976 err:
2977         return ret;
2978 }
2979
2980 static int proc_id_map_release(struct inode *inode, struct file *file)
2981 {
2982         struct seq_file *seq = file->private_data;
2983         struct user_namespace *ns = seq->private;
2984         put_user_ns(ns);
2985         return seq_release(inode, file);
2986 }
2987
2988 static int proc_uid_map_open(struct inode *inode, struct file *file)
2989 {
2990         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2991 }
2992
2993 static int proc_gid_map_open(struct inode *inode, struct file *file)
2994 {
2995         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2996 }
2997
2998 static const struct file_operations proc_uid_map_operations = {
2999         .open           = proc_uid_map_open,
3000         .write          = proc_uid_map_write,
3001         .read           = seq_read,
3002         .llseek         = seq_lseek,
3003         .release        = proc_id_map_release,
3004 };
3005
3006 static const struct file_operations proc_gid_map_operations = {
3007         .open           = proc_gid_map_open,
3008         .write          = proc_gid_map_write,
3009         .read           = seq_read,
3010         .llseek         = seq_lseek,
3011         .release        = proc_id_map_release,
3012 };
3013 #endif /* CONFIG_USER_NS */
3014
3015 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
3016                                 struct pid *pid, struct task_struct *task)
3017 {
3018         int err = lock_trace(task);
3019         if (!err) {
3020                 seq_printf(m, "%08x\n", task->personality);
3021                 unlock_trace(task);
3022         }
3023         return err;
3024 }
3025
3026 /*
3027  * Thread groups
3028  */
3029 static const struct file_operations proc_task_operations;
3030 static const struct inode_operations proc_task_inode_operations;
3031
3032 static const struct pid_entry tgid_base_stuff[] = {
3033         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
3034         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3035 #ifdef CONFIG_CHECKPOINT_RESTORE
3036         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
3037 #endif
3038         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3039         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3040 #ifdef CONFIG_NET
3041         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3042 #endif
3043         REG("environ",    S_IRUSR, proc_environ_operations),
3044         INF("auxv",       S_IRUSR, proc_pid_auxv),
3045         ONE("status",     S_IRUGO, proc_pid_status),
3046         ONE("personality", S_IRUGO, proc_pid_personality),
3047         INF("limits",     S_IRUGO, proc_pid_limits),
3048 #ifdef CONFIG_SCHED_DEBUG
3049         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3050 #endif
3051 #ifdef CONFIG_SCHED_AUTOGROUP
3052         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
3053 #endif
3054         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3055 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3056         INF("syscall",    S_IRUGO, proc_pid_syscall),
3057 #endif
3058         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
3059         ONE("stat",       S_IRUGO, proc_tgid_stat),
3060         ONE("statm",      S_IRUGO, proc_pid_statm),
3061         REG("maps",       S_IRUGO, proc_pid_maps_operations),
3062 #ifdef CONFIG_NUMA
3063         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
3064 #endif
3065         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
3066         LNK("cwd",        proc_cwd_link),
3067         LNK("root",       proc_root_link),
3068         LNK("exe",        proc_exe_link),
3069         REG("mounts",     S_IRUGO, proc_mounts_operations),
3070         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3071         REG("mountstats", S_IRUSR, proc_mountstats_operations),
3072 #ifdef CONFIG_PROC_PAGE_MONITOR
3073         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3074         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
3075         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
3076 #endif
3077 #ifdef CONFIG_SECURITY
3078         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3079 #endif
3080 #ifdef CONFIG_KALLSYMS
3081         INF("wchan",      S_IRUGO, proc_pid_wchan),
3082 #endif
3083 #ifdef CONFIG_STACKTRACE
3084         ONE("stack",      S_IRUGO, proc_pid_stack),
3085 #endif
3086 #ifdef CONFIG_SCHEDSTATS
3087         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
3088 #endif
3089 #ifdef CONFIG_LATENCYTOP
3090         REG("latency",  S_IRUGO, proc_lstats_operations),
3091 #endif
3092 #ifdef CONFIG_PROC_PID_CPUSET
3093         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
3094 #endif
3095 #ifdef CONFIG_CGROUPS
3096         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3097 #endif
3098         INF("oom_score",  S_IRUGO, proc_oom_score),
3099         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3100         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3101 #ifdef CONFIG_AUDITSYSCALL
3102         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
3103         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3104 #endif
3105 #ifdef CONFIG_FAULT_INJECTION
3106         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3107 #endif
3108 #ifdef CONFIG_ELF_CORE
3109         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3110 #endif
3111 #ifdef CONFIG_TASK_IO_ACCOUNTING
3112         INF("io",       S_IRUSR, proc_tgid_io_accounting),
3113 #endif
3114 #ifdef CONFIG_HARDWALL
3115         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
3116 #endif
3117 #ifdef CONFIG_USER_NS
3118         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3119         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3120 #endif
3121 };
3122
3123 static int proc_tgid_base_readdir(struct file * filp,
3124                              void * dirent, filldir_t filldir)
3125 {
3126         return proc_pident_readdir(filp,dirent,filldir,
3127                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
3128 }
3129
3130 static const struct file_operations proc_tgid_base_operations = {
3131         .read           = generic_read_dir,
3132         .readdir        = proc_tgid_base_readdir,
3133         .llseek         = default_llseek,
3134 };
3135
3136 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3137         return proc_pident_lookup(dir, dentry,
3138                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3139 }
3140
3141 static const struct inode_operations proc_tgid_base_inode_operations = {
3142         .lookup         = proc_tgid_base_lookup,
3143         .getattr        = pid_getattr,
3144         .setattr        = proc_setattr,
3145         .permission     = proc_pid_permission,
3146 };
3147
3148 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
3149 {
3150         struct dentry *dentry, *leader, *dir;
3151         char buf[PROC_NUMBUF];
3152         struct qstr name;
3153
3154         name.name = buf;
3155         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3156         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
3157         if (dentry) {
3158                 shrink_dcache_parent(dentry);
3159                 d_drop(dentry);
3160                 dput(dentry);
3161         }
3162
3163         name.name = buf;
3164         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
3165         leader = d_hash_and_lookup(mnt->mnt_root, &name);
3166         if (!leader)
3167                 goto out;
3168
3169         name.name = "task";
3170         name.len = strlen(name.name);
3171         dir = d_hash_and_lookup(leader, &name);
3172         if (!dir)
3173                 goto out_put_leader;
3174
3175         name.name = buf;
3176         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3177         dentry = d_hash_and_lookup(dir, &name);
3178         if (dentry) {
3179                 shrink_dcache_parent(dentry);
3180                 d_drop(dentry);
3181                 dput(dentry);
3182         }
3183
3184         dput(dir);
3185 out_put_leader:
3186         dput(leader);
3187 out:
3188         return;
3189 }
3190
3191 /**
3192  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
3193  * @task: task that should be flushed.
3194  *
3195  * When flushing dentries from proc, one needs to flush them from global
3196  * proc (proc_mnt) and from all the namespaces' procs this task was seen
3197  * in. This call is supposed to do all of this job.
3198  *
3199  * Looks in the dcache for
3200  * /proc/@pid
3201  * /proc/@tgid/task/@pid
3202  * if either directory is present flushes it and all of it'ts children
3203  * from the dcache.
3204  *
3205  * It is safe and reasonable to cache /proc entries for a task until
3206  * that task exits.  After that they just clog up the dcache with
3207  * useless entries, possibly causing useful dcache entries to be
3208  * flushed instead.  This routine is proved to flush those useless
3209  * dcache entries at process exit time.
3210  *
3211  * NOTE: This routine is just an optimization so it does not guarantee
3212  *       that no dcache entries will exist at process exit time it
3213  *       just makes it very unlikely that any will persist.
3214  */
3215
3216 void proc_flush_task(struct task_struct *task)
3217 {
3218         int i;
3219         struct pid *pid, *tgid;
3220         struct upid *upid;
3221
3222         pid = task_pid(task);
3223         tgid = task_tgid(task);
3224
3225         for (i = 0; i <= pid->level; i++) {
3226                 upid = &pid->numbers[i];
3227                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3228                                         tgid->numbers[i].nr);
3229         }
3230
3231         upid = &pid->numbers[pid->level];
3232         if (upid->nr == 1)
3233                 pid_ns_release_proc(upid->ns);
3234 }
3235
3236 static struct dentry *proc_pid_instantiate(struct inode *dir,
3237                                            struct dentry * dentry,
3238                                            struct task_struct *task, const void *ptr)
3239 {
3240         struct dentry *error = ERR_PTR(-ENOENT);
3241         struct inode *inode;
3242
3243         inode = proc_pid_make_inode(dir->i_sb, task);
3244         if (!inode)
3245                 goto out;
3246
3247         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3248         inode->i_op = &proc_tgid_base_inode_operations;
3249         inode->i_fop = &proc_tgid_base_operations;
3250         inode->i_flags|=S_IMMUTABLE;
3251
3252         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
3253                                                   ARRAY_SIZE(tgid_base_stuff)));
3254
3255         d_set_d_op(dentry, &pid_dentry_operations);
3256
3257         d_add(dentry, inode);
3258         /* Close the race of the process dying before we return the dentry */
3259         if (pid_revalidate(dentry, NULL))
3260                 error = NULL;
3261 out:
3262         return error;
3263 }
3264
3265 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3266 {
3267         struct dentry *result;
3268         struct task_struct *task;
3269         unsigned tgid;
3270         struct pid_namespace *ns;
3271
3272         result = proc_base_lookup(dir, dentry);
3273         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
3274                 goto out;
3275
3276         tgid = name_to_int(dentry);
3277         if (tgid == ~0U)
3278                 goto out;
3279
3280         ns = dentry->d_sb->s_fs_info;
3281         rcu_read_lock();
3282         task = find_task_by_pid_ns(tgid, ns);
3283         if (task)
3284                 get_task_struct(task);
3285         rcu_read_unlock();
3286         if (!task)
3287                 goto out;
3288
3289         result = proc_pid_instantiate(dir, dentry, task, NULL);
3290         put_task_struct(task);
3291 out:
3292         return result;
3293 }
3294
3295 /*
3296  * Find the first task with tgid >= tgid
3297  *
3298  */
3299 struct tgid_iter {
3300         unsigned int tgid;
3301         struct task_struct *task;
3302 };
3303 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3304 {
3305         struct pid *pid;
3306
3307         if (iter.task)
3308                 put_task_struct(iter.task);
3309         rcu_read_lock();
3310 retry:
3311         iter.task = NULL;
3312         pid = find_ge_pid(iter.tgid, ns);
3313         if (pid) {
3314                 iter.tgid = pid_nr_ns(pid, ns);
3315                 iter.task = pid_task(pid, PIDTYPE_PID);
3316                 /* What we to know is if the pid we have find is the
3317                  * pid of a thread_group_leader.  Testing for task
3318                  * being a thread_group_leader is the obvious thing
3319                  * todo but there is a window when it fails, due to
3320                  * the pid transfer logic in de_thread.
3321                  *
3322                  * So we perform the straight forward test of seeing
3323                  * if the pid we have found is the pid of a thread
3324                  * group leader, and don't worry if the task we have
3325                  * found doesn't happen to be a thread group leader.
3326                  * As we don't care in the case of readdir.
3327                  */
3328                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3329                         iter.tgid += 1;
3330                         goto retry;
3331                 }
3332                 get_task_struct(iter.task);
3333         }
3334         rcu_read_unlock();
3335         return iter;
3336 }
3337
3338 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
3339
3340 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3341         struct tgid_iter iter)
3342 {
3343         char name[PROC_NUMBUF];
3344         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
3345         return proc_fill_cache(filp, dirent, filldir, name, len,
3346                                 proc_pid_instantiate, iter.task, NULL);
3347 }
3348
3349 static int fake_filldir(void *buf, const char *name, int namelen,
3350                         loff_t offset, u64 ino, unsigned d_type)
3351 {
3352         return 0;
3353 }
3354
3355 /* for the /proc/ directory itself, after non-process stuff has been done */
3356 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
3357 {
3358         unsigned int nr;
3359         struct task_struct *reaper;
3360         struct tgid_iter iter;
3361         struct pid_namespace *ns;
3362         filldir_t __filldir;
3363
3364         if (filp->f_pos >= PID_MAX_LIMIT + TGID_OFFSET)
3365                 goto out_no_task;
3366         nr = filp->f_pos - FIRST_PROCESS_ENTRY;
3367
3368         reaper = get_proc_task(filp->f_path.dentry->d_inode);
3369         if (!reaper)
3370                 goto out_no_task;
3371
3372         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
3373                 const struct pid_entry *p = &proc_base_stuff[nr];
3374                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
3375                         goto out;
3376         }
3377
3378         ns = filp->f_dentry->d_sb->s_fs_info;
3379         iter.task = NULL;
3380         iter.tgid = filp->f_pos - TGID_OFFSET;
3381         for (iter = next_tgid(ns, iter);
3382              iter.task;
3383              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3384                 if (has_pid_permissions(ns, iter.task, 2))
3385                         __filldir = filldir;
3386                 else
3387                         __filldir = fake_filldir;
3388
3389                 filp->f_pos = iter.tgid + TGID_OFFSET;
3390                 if (proc_pid_fill_cache(filp, dirent, __filldir, iter) < 0) {
3391                         put_task_struct(iter.task);
3392                         goto out;
3393                 }
3394         }
3395         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
3396 out:
3397         put_task_struct(reaper);
3398 out_no_task:
3399         return 0;
3400 }
3401
3402 /*
3403  * Tasks
3404  */
3405 static const struct pid_entry tid_base_stuff[] = {
3406         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3407         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3408         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3409         REG("environ",   S_IRUSR, proc_environ_operations),
3410         INF("auxv",      S_IRUSR, proc_pid_auxv),
3411         ONE("status",    S_IRUGO, proc_pid_status),
3412         ONE("personality", S_IRUGO, proc_pid_personality),
3413         INF("limits",    S_IRUGO, proc_pid_limits),
3414 #ifdef CONFIG_SCHED_DEBUG
3415         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3416 #endif
3417         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3418 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3419         INF("syscall",   S_IRUGO, proc_pid_syscall),
3420 #endif
3421         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
3422         ONE("stat",      S_IRUGO, proc_tid_stat),
3423         ONE("statm",     S_IRUGO, proc_pid_statm),
3424         REG("maps",      S_IRUGO, proc_tid_maps_operations),
3425 #ifdef CONFIG_NUMA
3426         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3427 #endif
3428         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3429         LNK("cwd",       proc_cwd_link),
3430         LNK("root",      proc_root_link),
3431         LNK("exe",       proc_exe_link),
3432         REG("mounts",    S_IRUGO, proc_mounts_operations),
3433         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3434 #ifdef CONFIG_PROC_PAGE_MONITOR
3435         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3436         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3437         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
3438 #endif
3439 #ifdef CONFIG_SECURITY
3440         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3441 #endif
3442 #ifdef CONFIG_KALLSYMS
3443         INF("wchan",     S_IRUGO, proc_pid_wchan),
3444 #endif
3445 #ifdef CONFIG_STACKTRACE
3446         ONE("stack",      S_IRUGO, proc_pid_stack),
3447 #endif
3448 #ifdef CONFIG_SCHEDSTATS
3449         INF("schedstat", S_IRUGO, proc_pid_schedstat),
3450 #endif
3451 #ifdef CONFIG_LATENCYTOP
3452         REG("latency",  S_IRUGO, proc_lstats_operations),
3453 #endif
3454 #ifdef CONFIG_PROC_PID_CPUSET
3455         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
3456 #endif
3457 #ifdef CONFIG_CGROUPS
3458         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3459 #endif
3460         INF("oom_score", S_IRUGO, proc_oom_score),
3461         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3462         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3463 #ifdef CONFIG_AUDITSYSCALL
3464         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3465         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3466 #endif
3467 #ifdef CONFIG_FAULT_INJECTION
3468         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3469 #endif
3470 #ifdef CONFIG_TASK_IO_ACCOUNTING
3471         INF("io",       S_IRUSR, proc_tid_io_accounting),
3472 #endif
3473 #ifdef CONFIG_HARDWALL
3474         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
3475 #endif
3476 #ifdef CONFIG_USER_NS
3477         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3478         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3479 #endif
3480 };
3481
3482 static int proc_tid_base_readdir(struct file * filp,
3483                              void * dirent, filldir_t filldir)
3484 {
3485         return proc_pident_readdir(filp,dirent,filldir,
3486                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
3487 }
3488
3489 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3490         return proc_pident_lookup(dir, dentry,
3491                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3492 }
3493
3494 static const struct file_operations proc_tid_base_operations = {
3495         .read           = generic_read_dir,
3496         .readdir        = proc_tid_base_readdir,
3497         .llseek         = default_llseek,
3498 };
3499
3500 static const struct inode_operations proc_tid_base_inode_operations = {
3501         .lookup         = proc_tid_base_lookup,
3502         .getattr        = pid_getattr,
3503         .setattr        = proc_setattr,
3504 };
3505
3506 static struct dentry *proc_task_instantiate(struct inode *dir,
3507         struct dentry *dentry, struct task_struct *task, const void *ptr)
3508 {
3509         struct dentry *error = ERR_PTR(-ENOENT);
3510         struct inode *inode;
3511         inode = proc_pid_make_inode(dir->i_sb, task);
3512
3513         if (!inode)
3514                 goto out;
3515         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3516         inode->i_op = &proc_tid_base_inode_operations;
3517         inode->i_fop = &proc_tid_base_operations;
3518         inode->i_flags|=S_IMMUTABLE;
3519
3520         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3521                                                   ARRAY_SIZE(tid_base_stuff)));
3522
3523         d_set_d_op(dentry, &pid_dentry_operations);
3524
3525         d_add(dentry, inode);
3526         /* Close the race of the process dying before we return the dentry */
3527         if (pid_revalidate(dentry, NULL))
3528                 error = NULL;
3529 out:
3530         return error;
3531 }
3532
3533 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3534 {
3535         struct dentry *result = ERR_PTR(-ENOENT);
3536         struct task_struct *task;
3537         struct task_struct *leader = get_proc_task(dir);
3538         unsigned tid;
3539         struct pid_namespace *ns;
3540
3541         if (!leader)
3542                 goto out_no_task;
3543
3544         tid = name_to_int(dentry);
3545         if (tid == ~0U)
3546                 goto out;
3547
3548         ns = dentry->d_sb->s_fs_info;
3549         rcu_read_lock();
3550         task = find_task_by_pid_ns(tid, ns);
3551         if (task)
3552                 get_task_struct(task);
3553         rcu_read_unlock();
3554         if (!task)
3555                 goto out;
3556         if (!same_thread_group(leader, task))
3557                 goto out_drop_task;
3558
3559         result = proc_task_instantiate(dir, dentry, task, NULL);
3560 out_drop_task:
3561         put_task_struct(task);
3562 out:
3563         put_task_struct(leader);
3564 out_no_task:
3565         return result;
3566 }
3567
3568 /*
3569  * Find the first tid of a thread group to return to user space.
3570  *
3571  * Usually this is just the thread group leader, but if the users
3572  * buffer was too small or there was a seek into the middle of the
3573  * directory we have more work todo.
3574  *
3575  * In the case of a short read we start with find_task_by_pid.
3576  *
3577  * In the case of a seek we start with the leader and walk nr
3578  * threads past it.
3579  */
3580 static struct task_struct *first_tid(struct task_struct *leader,
3581                 int tid, int nr, struct pid_namespace *ns)
3582 {
3583         struct task_struct *pos;
3584
3585         rcu_read_lock();
3586         /* Attempt to start with the pid of a thread */
3587         if (tid && (nr > 0)) {
3588                 pos = find_task_by_pid_ns(tid, ns);
3589                 if (pos && (pos->group_leader == leader))
3590                         goto found;
3591         }
3592
3593         /* If nr exceeds the number of threads there is nothing todo */
3594         pos = NULL;
3595         if (nr && nr >= get_nr_threads(leader))
3596                 goto out;
3597
3598         /* If we haven't found our starting place yet start
3599          * with the leader and walk nr threads forward.
3600          */
3601         for (pos = leader; nr > 0; --nr) {
3602                 pos = next_thread(pos);
3603                 if (pos == leader) {
3604                         pos = NULL;
3605                         goto out;
3606                 }
3607         }
3608 found:
3609         get_task_struct(pos);
3610 out:
3611         rcu_read_unlock();
3612         return pos;
3613 }
3614
3615 /*
3616  * Find the next thread in the thread list.
3617  * Return NULL if there is an error or no next thread.
3618  *
3619  * The reference to the input task_struct is released.
3620  */
3621 static struct task_struct *next_tid(struct task_struct *start)
3622 {
3623         struct task_struct *pos = NULL;
3624         rcu_read_lock();
3625         if (pid_alive(start)) {
3626                 pos = next_thread(start);
3627                 if (thread_group_leader(pos))
3628                         pos = NULL;
3629                 else
3630                         get_task_struct(pos);
3631         }
3632         rcu_read_unlock();
3633         put_task_struct(start);
3634         return pos;
3635 }
3636
3637 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3638         struct task_struct *task, int tid)
3639 {
3640         char name[PROC_NUMBUF];
3641         int len = snprintf(name, sizeof(name), "%d", tid);
3642         return proc_fill_cache(filp, dirent, filldir, name, len,
3643                                 proc_task_instantiate, task, NULL);
3644 }
3645
3646 /* for the /proc/TGID/task/ directories */
3647 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3648 {
3649         struct dentry *dentry = filp->f_path.dentry;
3650         struct inode *inode = dentry->d_inode;
3651         struct task_struct *leader = NULL;
3652         struct task_struct *task;
3653         int retval = -ENOENT;
3654         ino_t ino;
3655         int tid;
3656         struct pid_namespace *ns;
3657
3658         task = get_proc_task(inode);
3659         if (!task)
3660                 goto out_no_task;
3661         rcu_read_lock();
3662         if (pid_alive(task)) {
3663                 leader = task->group_leader;
3664                 get_task_struct(leader);
3665         }
3666         rcu_read_unlock();
3667         put_task_struct(task);
3668         if (!leader)
3669                 goto out_no_task;
3670         retval = 0;
3671
3672         switch ((unsigned long)filp->f_pos) {
3673         case 0:
3674                 ino = inode->i_ino;
3675                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3676                         goto out;
3677                 filp->f_pos++;
3678                 /* fall through */
3679         case 1:
3680                 ino = parent_ino(dentry);
3681                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3682                         goto out;
3683                 filp->f_pos++;
3684                 /* fall through */
3685         }
3686
3687         /* f_version caches the tgid value that the last readdir call couldn't
3688          * return. lseek aka telldir automagically resets f_version to 0.
3689          */
3690         ns = filp->f_dentry->d_sb->s_fs_info;
3691         tid = (int)filp->f_version;
3692         filp->f_version = 0;
3693         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3694              task;
3695              task = next_tid(task), filp->f_pos++) {
3696                 tid = task_pid_nr_ns(task, ns);
3697                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3698                         /* returning this tgid failed, save it as the first
3699                          * pid for the next readir call */
3700                         filp->f_version = (u64)tid;
3701                         put_task_struct(task);
3702                         break;
3703                 }
3704         }
3705 out:
3706         put_task_struct(leader);
3707 out_no_task:
3708         return retval;
3709 }
3710
3711 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3712 {
3713         struct inode *inode = dentry->d_inode;
3714         struct task_struct *p = get_proc_task(inode);
3715         generic_fillattr(inode, stat);
3716
3717         if (p) {
3718                 stat->nlink += get_nr_threads(p);
3719                 put_task_struct(p);
3720         }
3721
3722         return 0;
3723 }
3724
3725 static const struct inode_operations proc_task_inode_operations = {
3726         .lookup         = proc_task_lookup,
3727         .getattr        = proc_task_getattr,
3728         .setattr        = proc_setattr,
3729         .permission     = proc_pid_permission,
3730 };
3731
3732 static const struct file_operations proc_task_operations = {
3733         .read           = generic_read_dir,
3734         .readdir        = proc_task_readdir,
3735         .llseek         = default_llseek,
3736 };