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/fs_struct.h>
  85 #include <linux/slab.h>
  86 #include "internal.h"
  87
  88 /* NOTE:
  89  *      Implementing inode permission operations in /proc is almost
  90  *      certainly an error.  Permission checks need to happen during
  91  *      each system call not at open time.  The reason is that most of
  92  *      what we wish to check for permissions in /proc varies at runtime.
  93  *
  94  *      The classic example of a problem is opening file descriptors
  95  *      in /proc for a task before it execs a suid executable.
  96  */
  97
  98 struct pid_entry {
  99         char *name;
 100         int len;
 101         mode_t mode;
 102         const struct inode_operations *iop;
 103         const struct file_operations *fop;
 104         union proc_op op;
 105 };
 106
 107 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 108         .name = (NAME),                                 \
 109         .len  = sizeof(NAME) - 1,                       \
 110         .mode = MODE,                                   \
 111         .iop  = IOP,                                    \
 112         .fop  = FOP,                                    \
 113         .op   = OP,                                     \
 114 }
 115
 116 #define DIR(NAME, MODE, iops, fops)     \
 117         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 118 #define LNK(NAME, get_link)                                     \
 119         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 120                 &proc_pid_link_inode_operations, NULL,          \
 121                 { .proc_get_link = get_link } )
 122 #define REG(NAME, MODE, fops)                           \
 123         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 124 #define INF(NAME, MODE, read)                           \
 125         NOD(NAME, (S_IFREG|(MODE)),                     \
 126                 NULL, &proc_info_file_operations,       \
 127                 { .proc_read = read } )
 128 #define ONE(NAME, MODE, show)                           \
 129         NOD(NAME, (S_IFREG|(MODE)),                     \
 130                 NULL, &proc_single_file_operations,     \
 131                 { .proc_show = show } )
 132
 133 /*
 134  * Count the number of hardlinks for the pid_entry table, excluding the .
 135  * and .. links.
 136  */
 137 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
 138         unsigned int n)
 139 {
 140         unsigned int i;
 141         unsigned int count;
 142
 143         count = 0;
 144         for (i = 0; i < n; ++i) {
 145                 if (S_ISDIR(entries[i].mode))
 146                         ++count;
 147         }
 148
 149         return count;
 150 }
 151
 152 static int get_task_root(struct task_struct *task, struct path *root)
 153 {
 154         int result = -ENOENT;
 155
 156         task_lock(task);
 157         if (task->fs) {
 158                 get_fs_root(task->fs, root);
 159                 result = 0;
 160         }
 161         task_unlock(task);
 162         return result;
 163 }
 164
 165 static int proc_cwd_link(struct inode *inode, struct path *path)
 166 {
 167         struct task_struct *task = get_proc_task(inode);
 168         int result = -ENOENT;
 169
 170         if (task) {
 171                 task_lock(task);
 172                 if (task->fs) {
 173                         get_fs_pwd(task->fs, path);
 174                         result = 0;
 175                 }
 176                 task_unlock(task);
 177                 put_task_struct(task);
 178         }
 179         return result;
 180 }
 181
 182 static int proc_root_link(struct inode *inode, struct path *path)
 183 {
 184         struct task_struct *task = get_proc_task(inode);
 185         int result = -ENOENT;
 186
 187         if (task) {
 188                 result = get_task_root(task, path);
 189                 put_task_struct(task);
 190         }
 191         return result;
 192 }
 193
 194 /*
 195  * Return zero if current may access user memory in @task, -error if not.
 196  */
 197 static int check_mem_permission(struct task_struct *task)
 198 {
 199         /*
 200          * A task can always look at itself, in case it chooses
 201          * to use system calls instead of load instructions.
 202          */
 203         if (task == current)
 204                 return 0;
 205
 206         /*
 207          * If current is actively ptrace'ing, and would also be
 208          * permitted to freshly attach with ptrace now, permit it.
 209          */
 210         if (task_is_stopped_or_traced(task)) {
 211                 int match;
 212                 rcu_read_lock();
 213                 match = (tracehook_tracer_task(task) == current);
 214                 rcu_read_unlock();
 215                 if (match && ptrace_may_access(task, PTRACE_MODE_ATTACH))
 216                         return 0;
 217         }
 218
 219         /*
 220          * Noone else is allowed.
 221          */
 222         return -EPERM;
 223 }
 224
 225 struct mm_struct *mm_for_maps(struct task_struct *task)
 226 {
 227         struct mm_struct *mm;
 228
 229         if (mutex_lock_killable(&task->signal->cred_guard_mutex))
 230                 return NULL;
 231
 232         mm = get_task_mm(task);
 233         if (mm && mm != current->mm &&
 234                         !ptrace_may_access(task, PTRACE_MODE_READ)) {
 235                 mmput(mm);
 236                 mm = NULL;
 237         }
 238         mutex_unlock(&task->signal->cred_guard_mutex);
 239
 240         return mm;
 241 }
 242
 243 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
 244 {
 245         int res = 0;
 246         unsigned int len;
 247         struct mm_struct *mm = get_task_mm(task);
 248         if (!mm)
 249                 goto out;
 250         if (!mm->arg_end)
 251                 goto out_mm;    /* Shh! No looking before we're done */
 252
 253         len = mm->arg_end - mm->arg_start;
 254
 255         if (len > PAGE_SIZE)
 256                 len = PAGE_SIZE;
 257
 258         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
 259
 260         // If the nul at the end of args has been overwritten, then
 261         // assume application is using setproctitle(3).
 262         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
 263                 len = strnlen(buffer, res);
 264                 if (len < res) {
 265                     res = len;
 266                 } else {
 267                         len = mm->env_end - mm->env_start;
 268                         if (len > PAGE_SIZE - res)
 269                                 len = PAGE_SIZE - res;
 270                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
 271                         res = strnlen(buffer, res);
 272                 }
 273         }
 274 out_mm:
 275         mmput(mm);
 276 out:
 277         return res;
 278 }
 279
 280 static int proc_pid_auxv(struct task_struct *task, char *buffer)
 281 {
 282         int res = 0;
 283         struct mm_struct *mm = get_task_mm(task);
 284         if (mm) {
 285                 unsigned int nwords = 0;
 286                 do {
 287                         nwords += 2;
 288                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 289                 res = nwords * sizeof(mm->saved_auxv[0]);
 290                 if (res > PAGE_SIZE)
 291                         res = PAGE_SIZE;
 292                 memcpy(buffer, mm->saved_auxv, res);
 293                 mmput(mm);
 294         }
 295         return res;
 296 }
 297
 298
 299 #ifdef CONFIG_KALLSYMS
 300 /*
 301  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 302  * Returns the resolved symbol.  If that fails, simply return the address.
 303  */
 304 static int proc_pid_wchan(struct task_struct *task, char *buffer)
 305 {
 306         unsigned long wchan;
 307         char symname[KSYM_NAME_LEN];
 308
 309         wchan = get_wchan(task);
 310
 311         if (lookup_symbol_name(wchan, symname) < 0)
 312                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
 313                         return 0;
 314                 else
 315                         return sprintf(buffer, "%lu", wchan);
 316         else
 317                 return sprintf(buffer, "%s", symname);
 318 }
 319 #endif /* CONFIG_KALLSYMS */
 320
 321 #ifdef CONFIG_STACKTRACE
 322
 323 #define MAX_STACK_TRACE_DEPTH   64
 324
 325 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 326                           struct pid *pid, struct task_struct *task)
 327 {
 328         struct stack_trace trace;
 329         unsigned long *entries;
 330         int i;
 331
 332         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 333         if (!entries)
 334                 return -ENOMEM;
 335
 336         trace.nr_entries        = 0;
 337         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 338         trace.entries           = entries;
 339         trace.skip              = 0;
 340         save_stack_trace_tsk(task, &trace);
 341
 342         for (i = 0; i < trace.nr_entries; i++) {
 343                 seq_printf(m, "[<%p>] %pS\n",
 344                            (void *)entries[i], (void *)entries[i]);
 345         }
 346         kfree(entries);
 347
 348         return 0;
 349 }
 350 #endif
 351
 352 #ifdef CONFIG_SCHEDSTATS
 353 /*
 354  * Provides /proc/PID/schedstat
 355  */
 356 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
 357 {
 358         return sprintf(buffer, "%llu %llu %lu\n",
 359                         (unsigned long long)task->se.sum_exec_runtime,
 360                         (unsigned long long)task->sched_info.run_delay,
 361                         task->sched_info.pcount);
 362 }
 363 #endif
 364
 365 #ifdef CONFIG_LATENCYTOP
 366 static int lstats_show_proc(struct seq_file *m, void *v)
 367 {
 368         int i;
 369         struct inode *inode = m->private;
 370         struct task_struct *task = get_proc_task(inode);
 371
 372         if (!task)
 373                 return -ESRCH;
 374         seq_puts(m, "Latency Top version : v0.1\n");
 375         for (i = 0; i < 32; i++) {
 376                 if (task->latency_record[i].backtrace[0]) {
 377                         int q;
 378                         seq_printf(m, "%i %li %li ",
 379                                 task->latency_record[i].count,
 380                                 task->latency_record[i].time,
 381                                 task->latency_record[i].max);
 382                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 383                                 char sym[KSYM_SYMBOL_LEN];
 384                                 char *c;
 385                                 if (!task->latency_record[i].backtrace[q])
 386                                         break;
 387                                 if (task->latency_record[i].backtrace[q] == ULONG_MAX)
 388                                         break;
 389                                 sprint_symbol(sym, task->latency_record[i].backtrace[q]);
 390                                 c = strchr(sym, '+');
 391                                 if (c)
 392                                         *c = 0;
 393                                 seq_printf(m, "%s ", sym);
 394                         }
 395                         seq_printf(m, "\n");
 396                 }
 397
 398         }
 399         put_task_struct(task);
 400         return 0;
 401 }
 402
 403 static int lstats_open(struct inode *inode, struct file *file)
 404 {
 405         return single_open(file, lstats_show_proc, inode);
 406 }
 407
 408 static ssize_t lstats_write(struct file *file, const char __user *buf,
 409                             size_t count, loff_t *offs)
 410 {
 411         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 412
 413         if (!task)
 414                 return -ESRCH;
 415         clear_all_latency_tracing(task);
 416         put_task_struct(task);
 417
 418         return count;
 419 }
 420
 421 static const struct file_operations proc_lstats_operations = {
 422         .open           = lstats_open,
 423         .read           = seq_read,
 424         .write          = lstats_write,
 425         .llseek         = seq_lseek,
 426         .release        = single_release,
 427 };
 428
 429 #endif
 430
 431 static int proc_oom_score(struct task_struct *task, char *buffer)
 432 {
 433         unsigned long points = 0;
 434
 435         read_lock(&tasklist_lock);
 436         if (pid_alive(task))
 437                 points = oom_badness(task, NULL, NULL,
 438                                         totalram_pages + total_swap_pages);
 439         read_unlock(&tasklist_lock);
 440         return sprintf(buffer, "%lu\n", points);
 441 }
 442
 443 struct limit_names {
 444         char *name;
 445         char *unit;
 446 };
 447
 448 static const struct limit_names lnames[RLIM_NLIMITS] = {
 449         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 450         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 451         [RLIMIT_DATA] = {"Max data size", "bytes"},
 452         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 453         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 454         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 455         [RLIMIT_NPROC] = {"Max processes", "processes"},
 456         [RLIMIT_NOFILE] = {"Max open files", "files"},
 457         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 458         [RLIMIT_AS] = {"Max address space", "bytes"},
 459         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 460         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 461         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 462         [RLIMIT_NICE] = {"Max nice priority", NULL},
 463         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 464         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 465 };
 466
 467 /* Display limits for a process */
 468 static int proc_pid_limits(struct task_struct *task, char *buffer)
 469 {
 470         unsigned int i;
 471         int count = 0;
 472         unsigned long flags;
 473         char *bufptr = buffer;
 474
 475         struct rlimit rlim[RLIM_NLIMITS];
 476
 477         if (!lock_task_sighand(task, &flags))
 478                 return 0;
 479         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 480         unlock_task_sighand(task, &flags);
 481
 482         /*
 483          * print the file header
 484          */
 485         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
 486                         "Limit", "Soft Limit", "Hard Limit", "Units");
 487
 488         for (i = 0; i < RLIM_NLIMITS; i++) {
 489                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 490                         count += sprintf(&bufptr[count], "%-25s %-20s ",
 491                                          lnames[i].name, "unlimited");
 492                 else
 493                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
 494                                          lnames[i].name, rlim[i].rlim_cur);
 495
 496                 if (rlim[i].rlim_max == RLIM_INFINITY)
 497                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
 498                 else
 499                         count += sprintf(&bufptr[count], "%-20lu ",
 500                                          rlim[i].rlim_max);
 501
 502                 if (lnames[i].unit)
 503                         count += sprintf(&bufptr[count], "%-10s\n",
 504                                          lnames[i].unit);
 505                 else
 506                         count += sprintf(&bufptr[count], "\n");
 507         }
 508
 509         return count;
 510 }
 511
 512 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 513 static int proc_pid_syscall(struct task_struct *task, char *buffer)
 514 {
 515         long nr;
 516         unsigned long args[6], sp, pc;
 517
 518         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 519                 return sprintf(buffer, "running\n");
 520
 521         if (nr < 0)
 522                 return sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 523
 524         return sprintf(buffer,
 525                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 526                        nr,
 527                        args[0], args[1], args[2], args[3], args[4], args[5],
 528                        sp, pc);
 529 }
 530 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 531
 532 /************************************************************************/
 533 /*                       Here the fs part begins                        */
 534 /************************************************************************/
 535
 536 /* permission checks */
 537 static int proc_fd_access_allowed(struct inode *inode)
 538 {
 539         struct task_struct *task;
 540         int allowed = 0;
 541         /* Allow access to a task's file descriptors if it is us or we
 542          * may use ptrace attach to the process and find out that
 543          * information.
 544          */
 545         task = get_proc_task(inode);
 546         if (task) {
 547                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
 548                 put_task_struct(task);
 549         }
 550         return allowed;
 551 }
 552
 553 static int proc_setattr(struct dentry *dentry, struct iattr *attr)
 554 {
 555         int error;
 556         struct inode *inode = dentry->d_inode;
 557
 558         if (attr->ia_valid & ATTR_MODE)
 559                 return -EPERM;
 560
 561         error = inode_change_ok(inode, attr);
 562         if (error)
 563                 return error;
 564
 565         if ((attr->ia_valid & ATTR_SIZE) &&
 566             attr->ia_size != i_size_read(inode)) {
 567                 error = vmtruncate(inode, attr->ia_size);
 568                 if (error)
 569                         return error;
 570         }
 571
 572         setattr_copy(inode, attr);
 573         mark_inode_dirty(inode);
 574         return 0;
 575 }
 576
 577 static const struct inode_operations proc_def_inode_operations = {
 578         .setattr        = proc_setattr,
 579 };
 580
 581 static int mounts_open_common(struct inode *inode, struct file *file,
 582                               const struct seq_operations *op)
 583 {
 584         struct task_struct *task = get_proc_task(inode);
 585         struct nsproxy *nsp;
 586         struct mnt_namespace *ns = NULL;
 587         struct path root;
 588         struct proc_mounts *p;
 589         int ret = -EINVAL;
 590
 591         if (task) {
 592                 rcu_read_lock();
 593                 nsp = task_nsproxy(task);
 594                 if (nsp) {
 595                         ns = nsp->mnt_ns;
 596                         if (ns)
 597                                 get_mnt_ns(ns);
 598                 }
 599                 rcu_read_unlock();
 600                 if (ns && get_task_root(task, &root) == 0)
 601                         ret = 0;
 602                 put_task_struct(task);
 603         }
 604
 605         if (!ns)
 606                 goto err;
 607         if (ret)
 608                 goto err_put_ns;
 609
 610         ret = -ENOMEM;
 611         p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
 612         if (!p)
 613                 goto err_put_path;
 614
 615         file->private_data = &p->m;
 616         ret = seq_open(file, op);
 617         if (ret)
 618                 goto err_free;
 619
 620         p->m.private = p;
 621         p->ns = ns;
 622         p->root = root;
 623         p->event = ns->event;
 624
 625         return 0;
 626
 627  err_free:
 628         kfree(p);
 629  err_put_path:
 630         path_put(&root);
 631  err_put_ns:
 632         put_mnt_ns(ns);
 633  err:
 634         return ret;
 635 }
 636
 637 static int mounts_release(struct inode *inode, struct file *file)
 638 {
 639         struct proc_mounts *p = file->private_data;
 640         path_put(&p->root);
 641         put_mnt_ns(p->ns);
 642         return seq_release(inode, file);
 643 }
 644
 645 static unsigned mounts_poll(struct file *file, poll_table *wait)
 646 {
 647         struct proc_mounts *p = file->private_data;
 648         unsigned res = POLLIN | POLLRDNORM;
 649
 650         poll_wait(file, &p->ns->poll, wait);
 651         if (mnt_had_events(p))
 652                 res |= POLLERR | POLLPRI;
 653
 654         return res;
 655 }
 656
 657 static int mounts_open(struct inode *inode, struct file *file)
 658 {
 659         return mounts_open_common(inode, file, &mounts_op);
 660 }
 661
 662 static const struct file_operations proc_mounts_operations = {
 663         .open           = mounts_open,
 664         .read           = seq_read,
 665         .llseek         = seq_lseek,
 666         .release        = mounts_release,
 667         .poll           = mounts_poll,
 668 };
 669
 670 static int mountinfo_open(struct inode *inode, struct file *file)
 671 {
 672         return mounts_open_common(inode, file, &mountinfo_op);
 673 }
 674
 675 static const struct file_operations proc_mountinfo_operations = {
 676         .open           = mountinfo_open,
 677         .read           = seq_read,
 678         .llseek         = seq_lseek,
 679         .release        = mounts_release,
 680         .poll           = mounts_poll,
 681 };
 682
 683 static int mountstats_open(struct inode *inode, struct file *file)
 684 {
 685         return mounts_open_common(inode, file, &mountstats_op);
 686 }
 687
 688 static const struct file_operations proc_mountstats_operations = {
 689         .open           = mountstats_open,
 690         .read           = seq_read,
 691         .llseek         = seq_lseek,
 692         .release        = mounts_release,
 693 };
 694
 695 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 696
 697 static ssize_t proc_info_read(struct file * file, char __user * buf,
 698                           size_t count, loff_t *ppos)
 699 {
 700         struct inode * inode = file->f_path.dentry->d_inode;
 701         unsigned long page;
 702         ssize_t length;
 703         struct task_struct *task = get_proc_task(inode);
 704
 705         length = -ESRCH;
 706         if (!task)
 707                 goto out_no_task;
 708
 709         if (count > PROC_BLOCK_SIZE)
 710                 count = PROC_BLOCK_SIZE;
 711
 712         length = -ENOMEM;
 713         if (!(page = __get_free_page(GFP_TEMPORARY)))
 714                 goto out;
 715
 716         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 717
 718         if (length >= 0)
 719                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 720         free_page(page);
 721 out:
 722         put_task_struct(task);
 723 out_no_task:
 724         return length;
 725 }
 726
 727 static const struct file_operations proc_info_file_operations = {
 728         .read           = proc_info_read,
 729         .llseek         = generic_file_llseek,
 730 };
 731
 732 static int proc_single_show(struct seq_file *m, void *v)
 733 {
 734         struct inode *inode = m->private;
 735         struct pid_namespace *ns;
 736         struct pid *pid;
 737         struct task_struct *task;
 738         int ret;
 739
 740         ns = inode->i_sb->s_fs_info;
 741         pid = proc_pid(inode);
 742         task = get_pid_task(pid, PIDTYPE_PID);
 743         if (!task)
 744                 return -ESRCH;
 745
 746         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 747
 748         put_task_struct(task);
 749         return ret;
 750 }
 751
 752 static int proc_single_open(struct inode *inode, struct file *filp)
 753 {
 754         int ret;
 755         ret = single_open(filp, proc_single_show, NULL);
 756         if (!ret) {
 757                 struct seq_file *m = filp->private_data;
 758
 759                 m->private = inode;
 760         }
 761         return ret;
 762 }
 763
 764 static const struct file_operations proc_single_file_operations = {
 765         .open           = proc_single_open,
 766         .read           = seq_read,
 767         .llseek         = seq_lseek,
 768         .release        = single_release,
 769 };
 770
 771 static int mem_open(struct inode* inode, struct file* file)
 772 {
 773         file->private_data = (void*)((long)current->self_exec_id);
 774         /* OK to pass negative loff_t, we can catch out-of-range */
 775         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 776         return 0;
 777 }
 778
 779 static ssize_t mem_read(struct file * file, char __user * buf,
 780                         size_t count, loff_t *ppos)
 781 {
 782         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 783         char *page;
 784         unsigned long src = *ppos;
 785         int ret = -ESRCH;
 786         struct mm_struct *mm;
 787
 788         if (!task)
 789                 goto out_no_task;
 790
 791         if (check_mem_permission(task))
 792                 goto out;
 793
 794         ret = -ENOMEM;
 795         page = (char *)__get_free_page(GFP_TEMPORARY);
 796         if (!page)
 797                 goto out;
 798
 799         ret = 0;
 800
 801         mm = get_task_mm(task);
 802         if (!mm)
 803                 goto out_free;
 804
 805         ret = -EIO;
 806
 807         if (file->private_data != (void*)((long)current->self_exec_id))
 808                 goto out_put;
 809
 810         ret = 0;
 811
 812         while (count > 0) {
 813                 int this_len, retval;
 814
 815                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 816                 retval = access_process_vm(task, src, page, this_len, 0);
 817                 if (!retval || check_mem_permission(task)) {
 818                         if (!ret)
 819                                 ret = -EIO;
 820                         break;
 821                 }
 822
 823                 if (copy_to_user(buf, page, retval)) {
 824                         ret = -EFAULT;
 825                         break;
 826                 }
 827
 828                 ret += retval;
 829                 src += retval;
 830                 buf += retval;
 831                 count -= retval;
 832         }
 833         *ppos = src;
 834
 835 out_put:
 836         mmput(mm);
 837 out_free:
 838         free_page((unsigned long) page);
 839 out:
 840         put_task_struct(task);
 841 out_no_task:
 842         return ret;
 843 }
 844
 845 #define mem_write NULL
 846
 847 #ifndef mem_write
 848 /* This is a security hazard */
 849 static ssize_t mem_write(struct file * file, const char __user *buf,
 850                          size_t count, loff_t *ppos)
 851 {
 852         int copied;
 853         char *page;
 854         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 855         unsigned long dst = *ppos;
 856
 857         copied = -ESRCH;
 858         if (!task)
 859                 goto out_no_task;
 860
 861         if (check_mem_permission(task))
 862                 goto out;
 863
 864         copied = -ENOMEM;
 865         page = (char *)__get_free_page(GFP_TEMPORARY);
 866         if (!page)
 867                 goto out;
 868
 869         copied = 0;
 870         while (count > 0) {
 871                 int this_len, retval;
 872
 873                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 874                 if (copy_from_user(page, buf, this_len)) {
 875                         copied = -EFAULT;
 876                         break;
 877                 }
 878                 retval = access_process_vm(task, dst, page, this_len, 1);
 879                 if (!retval) {
 880                         if (!copied)
 881                                 copied = -EIO;
 882                         break;
 883                 }
 884                 copied += retval;
 885                 buf += retval;
 886                 dst += retval;
 887                 count -= retval;
 888         }
 889         *ppos = dst;
 890         free_page((unsigned long) page);
 891 out:
 892         put_task_struct(task);
 893 out_no_task:
 894         return copied;
 895 }
 896 #endif
 897
 898 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 899 {
 900         switch (orig) {
 901         case 0:
 902                 file->f_pos = offset;
 903                 break;
 904         case 1:
 905                 file->f_pos += offset;
 906                 break;
 907         default:
 908                 return -EINVAL;
 909         }
 910         force_successful_syscall_return();
 911         return file->f_pos;
 912 }
 913
 914 static const struct file_operations proc_mem_operations = {
 915         .llseek         = mem_lseek,
 916         .read           = mem_read,
 917         .write          = mem_write,
 918         .open           = mem_open,
 919 };
 920
 921 static ssize_t environ_read(struct file *file, char __user *buf,
 922                         size_t count, loff_t *ppos)
 923 {
 924         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 925         char *page;
 926         unsigned long src = *ppos;
 927         int ret = -ESRCH;
 928         struct mm_struct *mm;
 929
 930         if (!task)
 931                 goto out_no_task;
 932
 933         if (!ptrace_may_access(task, PTRACE_MODE_READ))
 934                 goto out;
 935
 936         ret = -ENOMEM;
 937         page = (char *)__get_free_page(GFP_TEMPORARY);
 938         if (!page)
 939                 goto out;
 940
 941         ret = 0;
 942
 943         mm = get_task_mm(task);
 944         if (!mm)
 945                 goto out_free;
 946
 947         while (count > 0) {
 948                 int this_len, retval, max_len;
 949
 950                 this_len = mm->env_end - (mm->env_start + src);
 951
 952                 if (this_len <= 0)
 953                         break;
 954
 955                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 956                 this_len = (this_len > max_len) ? max_len : this_len;
 957
 958                 retval = access_process_vm(task, (mm->env_start + src),
 959                         page, this_len, 0);
 960
 961                 if (retval <= 0) {
 962                         ret = retval;
 963                         break;
 964                 }
 965
 966                 if (copy_to_user(buf, page, retval)) {
 967                         ret = -EFAULT;
 968                         break;
 969                 }
 970
 971                 ret += retval;
 972                 src += retval;
 973                 buf += retval;
 974                 count -= retval;
 975         }
 976         *ppos = src;
 977
 978         mmput(mm);
 979 out_free:
 980         free_page((unsigned long) page);
 981 out:
 982         put_task_struct(task);
 983 out_no_task:
 984         return ret;
 985 }
 986
 987 static const struct file_operations proc_environ_operations = {
 988         .read           = environ_read,
 989         .llseek         = generic_file_llseek,
 990 };
 991
 992 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
 993                                 size_t count, loff_t *ppos)
 994 {
 995         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 996         char buffer[PROC_NUMBUF];
 997         size_t len;
 998         int oom_adjust = OOM_DISABLE;
 999         unsigned long flags;
1000
1001         if (!task)
1002                 return -ESRCH;
1003
1004         if (lock_task_sighand(task, &flags)) {
1005                 oom_adjust = task->signal->oom_adj;
1006                 unlock_task_sighand(task, &flags);
1007         }
1008
1009         put_task_struct(task);
1010
1011         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
1012
1013         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1014 }
1015
1016 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
1017                                 size_t count, loff_t *ppos)
1018 {
1019         struct task_struct *task;
1020         char buffer[PROC_NUMBUF];
1021         long oom_adjust;
1022         unsigned long flags;
1023         int err;
1024
1025         memset(buffer, 0, sizeof(buffer));
1026         if (count > sizeof(buffer) - 1)
1027                 count = sizeof(buffer) - 1;
1028         if (copy_from_user(buffer, buf, count)) {
1029                 err = -EFAULT;
1030                 goto out;
1031         }
1032
1033         err = strict_strtol(strstrip(buffer), 0, &oom_adjust);
1034         if (err)
1035                 goto out;
1036         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
1037              oom_adjust != OOM_DISABLE) {
1038                 err = -EINVAL;
1039                 goto out;
1040         }
1041
1042         task = get_proc_task(file->f_path.dentry->d_inode);
1043         if (!task) {
1044                 err = -ESRCH;
1045                 goto out;
1046         }
1047
1048         task_lock(task);
1049         if (!task->mm) {
1050                 err = -EINVAL;
1051                 goto err_task_lock;
1052         }
1053
1054         if (!lock_task_sighand(task, &flags)) {
1055                 err = -ESRCH;
1056                 goto err_task_lock;
1057         }
1058
1059         if (oom_adjust < task->signal->oom_adj && !capable(CAP_SYS_RESOURCE)) {
1060                 err = -EACCES;
1061                 goto err_sighand;
1062         }
1063
1064         if (oom_adjust != task->signal->oom_adj) {
1065                 if (oom_adjust == OOM_DISABLE)
1066                         atomic_inc(&task->mm->oom_disable_count);
1067                 if (task->signal->oom_adj == OOM_DISABLE)
1068                         atomic_dec(&task->mm->oom_disable_count);
1069         }
1070
1071         /*
1072          * Warn that /proc/pid/oom_adj is deprecated, see
1073          * Documentation/feature-removal-schedule.txt.
1074          */
1075         printk_once(KERN_WARNING "%s (%d): /proc/%d/oom_adj is deprecated, "
1076                         "please use /proc/%d/oom_score_adj instead.\n",
1077                         current->comm, task_pid_nr(current),
1078                         task_pid_nr(task), task_pid_nr(task));
1079         task->signal->oom_adj = oom_adjust;
1080         /*
1081          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1082          * value is always attainable.
1083          */
1084         if (task->signal->oom_adj == OOM_ADJUST_MAX)
1085                 task->signal->oom_score_adj = OOM_SCORE_ADJ_MAX;
1086         else
1087                 task->signal->oom_score_adj = (oom_adjust * OOM_SCORE_ADJ_MAX) /
1088                                                                 -OOM_DISABLE;
1089 err_sighand:
1090         unlock_task_sighand(task, &flags);
1091 err_task_lock:
1092         task_unlock(task);
1093         put_task_struct(task);
1094 out:
1095         return err < 0 ? err : count;
1096 }
1097
1098 static const struct file_operations proc_oom_adjust_operations = {
1099         .read           = oom_adjust_read,
1100         .write          = oom_adjust_write,
1101         .llseek         = generic_file_llseek,
1102 };
1103
1104 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1105                                         size_t count, loff_t *ppos)
1106 {
1107         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
1108         char buffer[PROC_NUMBUF];
1109         int oom_score_adj = OOM_SCORE_ADJ_MIN;
1110         unsigned long flags;
1111         size_t len;
1112
1113         if (!task)
1114                 return -ESRCH;
1115         if (lock_task_sighand(task, &flags)) {
1116                 oom_score_adj = task->signal->oom_score_adj;
1117                 unlock_task_sighand(task, &flags);
1118         }
1119         put_task_struct(task);
1120         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_score_adj);
1121         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1122 }
1123
1124 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1125                                         size_t count, loff_t *ppos)
1126 {
1127         struct task_struct *task;
1128         char buffer[PROC_NUMBUF];
1129         unsigned long flags;
1130         long oom_score_adj;
1131         int err;
1132
1133         memset(buffer, 0, sizeof(buffer));
1134         if (count > sizeof(buffer) - 1)
1135                 count = sizeof(buffer) - 1;
1136         if (copy_from_user(buffer, buf, count)) {
1137                 err = -EFAULT;
1138                 goto out;
1139         }
1140
1141         err = strict_strtol(strstrip(buffer), 0, &oom_score_adj);
1142         if (err)
1143                 goto out;
1144         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1145                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1146                 err = -EINVAL;
1147                 goto out;
1148         }
1149
1150         task = get_proc_task(file->f_path.dentry->d_inode);
1151         if (!task) {
1152                 err = -ESRCH;
1153                 goto out;
1154         }
1155
1156         task_lock(task);
1157         if (!task->mm) {
1158                 err = -EINVAL;
1159                 goto err_task_lock;
1160         }
1161
1162         if (!lock_task_sighand(task, &flags)) {
1163                 err = -ESRCH;
1164                 goto err_task_lock;
1165         }
1166
1167         if (oom_score_adj < task->signal->oom_score_adj &&
1168                         !capable(CAP_SYS_RESOURCE)) {
1169                 err = -EACCES;
1170                 goto err_sighand;
1171         }
1172
1173         if (oom_score_adj != task->signal->oom_score_adj) {
1174                 if (oom_score_adj == OOM_SCORE_ADJ_MIN)
1175                         atomic_inc(&task->mm->oom_disable_count);
1176                 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1177                         atomic_dec(&task->mm->oom_disable_count);
1178         }
1179         task->signal->oom_score_adj = oom_score_adj;
1180         /*
1181          * Scale /proc/pid/oom_adj appropriately ensuring that OOM_DISABLE is
1182          * always attainable.
1183          */
1184         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1185                 task->signal->oom_adj = OOM_DISABLE;
1186         else
1187                 task->signal->oom_adj = (oom_score_adj * OOM_ADJUST_MAX) /
1188                                                         OOM_SCORE_ADJ_MAX;
1189 err_sighand:
1190         unlock_task_sighand(task, &flags);
1191 err_task_lock:
1192         task_unlock(task);
1193         put_task_struct(task);
1194 out:
1195         return err < 0 ? err : count;
1196 }
1197
1198 static const struct file_operations proc_oom_score_adj_operations = {
1199         .read           = oom_score_adj_read,
1200         .write          = oom_score_adj_write,
1201         .llseek         = default_llseek,
1202 };
1203
1204 #ifdef CONFIG_AUDITSYSCALL
1205 #define TMPBUFLEN 21
1206 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1207                                   size_t count, loff_t *ppos)
1208 {
1209         struct inode * inode = file->f_path.dentry->d_inode;
1210         struct task_struct *task = get_proc_task(inode);
1211         ssize_t length;
1212         char tmpbuf[TMPBUFLEN];
1213
1214         if (!task)
1215                 return -ESRCH;
1216         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1217                                 audit_get_loginuid(task));
1218         put_task_struct(task);
1219         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1220 }
1221
1222 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1223                                    size_t count, loff_t *ppos)
1224 {
1225         struct inode * inode = file->f_path.dentry->d_inode;
1226         char *page, *tmp;
1227         ssize_t length;
1228         uid_t loginuid;
1229
1230         if (!capable(CAP_AUDIT_CONTROL))
1231                 return -EPERM;
1232
1233         rcu_read_lock();
1234         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1235                 rcu_read_unlock();
1236                 return -EPERM;
1237         }
1238         rcu_read_unlock();
1239
1240         if (count >= PAGE_SIZE)
1241                 count = PAGE_SIZE - 1;
1242
1243         if (*ppos != 0) {
1244                 /* No partial writes. */
1245                 return -EINVAL;
1246         }
1247         page = (char*)__get_free_page(GFP_TEMPORARY);
1248         if (!page)
1249                 return -ENOMEM;
1250         length = -EFAULT;
1251         if (copy_from_user(page, buf, count))
1252                 goto out_free_page;
1253
1254         page[count] = '\0';
1255         loginuid = simple_strtoul(page, &tmp, 10);
1256         if (tmp == page) {
1257                 length = -EINVAL;
1258                 goto out_free_page;
1259
1260         }
1261         length = audit_set_loginuid(current, loginuid);
1262         if (likely(length == 0))
1263                 length = count;
1264
1265 out_free_page:
1266         free_page((unsigned long) page);
1267         return length;
1268 }
1269
1270 static const struct file_operations proc_loginuid_operations = {
1271         .read           = proc_loginuid_read,
1272         .write          = proc_loginuid_write,
1273         .llseek         = generic_file_llseek,
1274 };
1275
1276 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1277                                   size_t count, loff_t *ppos)
1278 {
1279         struct inode * inode = file->f_path.dentry->d_inode;
1280         struct task_struct *task = get_proc_task(inode);
1281         ssize_t length;
1282         char tmpbuf[TMPBUFLEN];
1283
1284         if (!task)
1285                 return -ESRCH;
1286         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1287                                 audit_get_sessionid(task));
1288         put_task_struct(task);
1289         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1290 }
1291
1292 static const struct file_operations proc_sessionid_operations = {
1293         .read           = proc_sessionid_read,
1294         .llseek         = generic_file_llseek,
1295 };
1296 #endif
1297
1298 #ifdef CONFIG_FAULT_INJECTION
1299 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1300                                       size_t count, loff_t *ppos)
1301 {
1302         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1303         char buffer[PROC_NUMBUF];
1304         size_t len;
1305         int make_it_fail;
1306
1307         if (!task)
1308                 return -ESRCH;
1309         make_it_fail = task->make_it_fail;
1310         put_task_struct(task);
1311
1312         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1313
1314         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1315 }
1316
1317 static ssize_t proc_fault_inject_write(struct file * file,
1318                         const char __user * buf, size_t count, loff_t *ppos)
1319 {
1320         struct task_struct *task;
1321         char buffer[PROC_NUMBUF], *end;
1322         int make_it_fail;
1323
1324         if (!capable(CAP_SYS_RESOURCE))
1325                 return -EPERM;
1326         memset(buffer, 0, sizeof(buffer));
1327         if (count > sizeof(buffer) - 1)
1328                 count = sizeof(buffer) - 1;
1329         if (copy_from_user(buffer, buf, count))
1330                 return -EFAULT;
1331         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1332         if (*end)
1333                 return -EINVAL;
1334         task = get_proc_task(file->f_dentry->d_inode);
1335         if (!task)
1336                 return -ESRCH;
1337         task->make_it_fail = make_it_fail;
1338         put_task_struct(task);
1339
1340         return count;
1341 }
1342
1343 static const struct file_operations proc_fault_inject_operations = {
1344         .read           = proc_fault_inject_read,
1345         .write          = proc_fault_inject_write,
1346         .llseek         = generic_file_llseek,
1347 };
1348 #endif
1349
1350
1351 #ifdef CONFIG_SCHED_DEBUG
1352 /*
1353  * Print out various scheduling related per-task fields:
1354  */
1355 static int sched_show(struct seq_file *m, void *v)
1356 {
1357         struct inode *inode = m->private;
1358         struct task_struct *p;
1359
1360         p = get_proc_task(inode);
1361         if (!p)
1362                 return -ESRCH;
1363         proc_sched_show_task(p, m);
1364
1365         put_task_struct(p);
1366
1367         return 0;
1368 }
1369
1370 static ssize_t
1371 sched_write(struct file *file, const char __user *buf,
1372             size_t count, loff_t *offset)
1373 {
1374         struct inode *inode = file->f_path.dentry->d_inode;
1375         struct task_struct *p;
1376
1377         p = get_proc_task(inode);
1378         if (!p)
1379                 return -ESRCH;
1380         proc_sched_set_task(p);
1381
1382         put_task_struct(p);
1383
1384         return count;
1385 }
1386
1387 static int sched_open(struct inode *inode, struct file *filp)
1388 {
1389         int ret;
1390
1391         ret = single_open(filp, sched_show, NULL);
1392         if (!ret) {
1393                 struct seq_file *m = filp->private_data;
1394
1395                 m->private = inode;
1396         }
1397         return ret;
1398 }
1399
1400 static const struct file_operations proc_pid_sched_operations = {
1401         .open           = sched_open,
1402         .read           = seq_read,
1403         .write          = sched_write,
1404         .llseek         = seq_lseek,
1405         .release        = single_release,
1406 };
1407
1408 #endif
1409
1410 #ifdef CONFIG_SCHED_AUTOGROUP
1411 /*
1412  * Print out autogroup related information:
1413  */
1414 static int sched_autogroup_show(struct seq_file *m, void *v)
1415 {
1416         struct inode *inode = m->private;
1417         struct task_struct *p;
1418
1419         p = get_proc_task(inode);
1420         if (!p)
1421                 return -ESRCH;
1422         proc_sched_autogroup_show_task(p, m);
1423
1424         put_task_struct(p);
1425
1426         return 0;
1427 }
1428
1429 static ssize_t
1430 sched_autogroup_write(struct file *file, const char __user *buf,
1431             size_t count, loff_t *offset)
1432 {
1433         struct inode *inode = file->f_path.dentry->d_inode;
1434         struct task_struct *p;
1435         char buffer[PROC_NUMBUF];
1436         long nice;
1437         int err;
1438
1439         memset(buffer, 0, sizeof(buffer));
1440         if (count > sizeof(buffer) - 1)
1441                 count = sizeof(buffer) - 1;
1442         if (copy_from_user(buffer, buf, count))
1443                 return -EFAULT;
1444
1445         err = strict_strtol(strstrip(buffer), 0, &nice);
1446         if (err)
1447                 return -EINVAL;
1448
1449         p = get_proc_task(inode);
1450         if (!p)
1451                 return -ESRCH;
1452
1453         err = nice;
1454         err = proc_sched_autogroup_set_nice(p, &err);
1455         if (err)
1456                 count = err;
1457
1458         put_task_struct(p);
1459
1460         return count;
1461 }
1462
1463 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1464 {
1465         int ret;
1466
1467         ret = single_open(filp, sched_autogroup_show, NULL);
1468         if (!ret) {
1469                 struct seq_file *m = filp->private_data;
1470
1471                 m->private = inode;
1472         }
1473         return ret;
1474 }
1475
1476 static const struct file_operations proc_pid_sched_autogroup_operations = {
1477         .open           = sched_autogroup_open,
1478         .read           = seq_read,
1479         .write          = sched_autogroup_write,
1480         .llseek         = seq_lseek,
1481         .release        = single_release,
1482 };
1483
1484 #endif /* CONFIG_SCHED_AUTOGROUP */
1485
1486 static ssize_t comm_write(struct file *file, const char __user *buf,
1487                                 size_t count, loff_t *offset)
1488 {
1489         struct inode *inode = file->f_path.dentry->d_inode;
1490         struct task_struct *p;
1491         char buffer[TASK_COMM_LEN];
1492
1493         memset(buffer, 0, sizeof(buffer));
1494         if (count > sizeof(buffer) - 1)
1495                 count = sizeof(buffer) - 1;
1496         if (copy_from_user(buffer, buf, count))
1497                 return -EFAULT;
1498
1499         p = get_proc_task(inode);
1500         if (!p)
1501                 return -ESRCH;
1502
1503         if (same_thread_group(current, p))
1504                 set_task_comm(p, buffer);
1505         else
1506                 count = -EINVAL;
1507
1508         put_task_struct(p);
1509
1510         return count;
1511 }
1512
1513 static int comm_show(struct seq_file *m, void *v)
1514 {
1515         struct inode *inode = m->private;
1516         struct task_struct *p;
1517
1518         p = get_proc_task(inode);
1519         if (!p)
1520                 return -ESRCH;
1521
1522         task_lock(p);
1523         seq_printf(m, "%s\n", p->comm);
1524         task_unlock(p);
1525
1526         put_task_struct(p);
1527
1528         return 0;
1529 }
1530
1531 static int comm_open(struct inode *inode, struct file *filp)
1532 {
1533         int ret;
1534
1535         ret = single_open(filp, comm_show, NULL);
1536         if (!ret) {
1537                 struct seq_file *m = filp->private_data;
1538
1539                 m->private = inode;
1540         }
1541         return ret;
1542 }
1543
1544 static const struct file_operations proc_pid_set_comm_operations = {
1545         .open           = comm_open,
1546         .read           = seq_read,
1547         .write          = comm_write,
1548         .llseek         = seq_lseek,
1549         .release        = single_release,
1550 };
1551
1552 /*
1553  * We added or removed a vma mapping the executable. The vmas are only mapped
1554  * during exec and are not mapped with the mmap system call.
1555  * Callers must hold down_write() on the mm's mmap_sem for these
1556  */
1557 void added_exe_file_vma(struct mm_struct *mm)
1558 {
1559         mm->num_exe_file_vmas++;
1560 }
1561
1562 void removed_exe_file_vma(struct mm_struct *mm)
1563 {
1564         mm->num_exe_file_vmas--;
1565         if ((mm->num_exe_file_vmas == 0) && mm->exe_file){
1566                 fput(mm->exe_file);
1567                 mm->exe_file = NULL;
1568         }
1569
1570 }
1571
1572 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1573 {
1574         if (new_exe_file)
1575                 get_file(new_exe_file);
1576         if (mm->exe_file)
1577                 fput(mm->exe_file);
1578         mm->exe_file = new_exe_file;
1579         mm->num_exe_file_vmas = 0;
1580 }
1581
1582 struct file *get_mm_exe_file(struct mm_struct *mm)
1583 {
1584         struct file *exe_file;
1585
1586         /* We need mmap_sem to protect against races with removal of
1587          * VM_EXECUTABLE vmas */
1588         down_read(&mm->mmap_sem);
1589         exe_file = mm->exe_file;
1590         if (exe_file)
1591                 get_file(exe_file);
1592         up_read(&mm->mmap_sem);
1593         return exe_file;
1594 }
1595
1596 void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
1597 {
1598         /* It's safe to write the exe_file pointer without exe_file_lock because
1599          * this is called during fork when the task is not yet in /proc */
1600         newmm->exe_file = get_mm_exe_file(oldmm);
1601 }
1602
1603 static int proc_exe_link(struct inode *inode, struct path *exe_path)
1604 {
1605         struct task_struct *task;
1606         struct mm_struct *mm;
1607         struct file *exe_file;
1608
1609         task = get_proc_task(inode);
1610         if (!task)
1611                 return -ENOENT;
1612         mm = get_task_mm(task);
1613         put_task_struct(task);
1614         if (!mm)
1615                 return -ENOENT;
1616         exe_file = get_mm_exe_file(mm);
1617         mmput(mm);
1618         if (exe_file) {
1619                 *exe_path = exe_file->f_path;
1620                 path_get(&exe_file->f_path);
1621                 fput(exe_file);
1622                 return 0;
1623         } else
1624                 return -ENOENT;
1625 }
1626
1627 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1628 {
1629         struct inode *inode = dentry->d_inode;
1630         int error = -EACCES;
1631
1632         /* We don't need a base pointer in the /proc filesystem */
1633         path_put(&nd->path);
1634
1635         /* Are we allowed to snoop on the tasks file descriptors? */
1636         if (!proc_fd_access_allowed(inode))
1637                 goto out;
1638
1639         error = PROC_I(inode)->op.proc_get_link(inode, &nd->path);
1640 out:
1641         return ERR_PTR(error);
1642 }
1643
1644 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1645 {
1646         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1647         char *pathname;
1648         int len;
1649
1650         if (!tmp)
1651                 return -ENOMEM;
1652
1653         pathname = d_path(path, tmp, PAGE_SIZE);
1654         len = PTR_ERR(pathname);
1655         if (IS_ERR(pathname))
1656                 goto out;
1657         len = tmp + PAGE_SIZE - 1 - pathname;
1658
1659         if (len > buflen)
1660                 len = buflen;
1661         if (copy_to_user(buffer, pathname, len))
1662                 len = -EFAULT;
1663  out:
1664         free_page((unsigned long)tmp);
1665         return len;
1666 }
1667
1668 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1669 {
1670         int error = -EACCES;
1671         struct inode *inode = dentry->d_inode;
1672         struct path path;
1673
1674         /* Are we allowed to snoop on the tasks file descriptors? */
1675         if (!proc_fd_access_allowed(inode))
1676                 goto out;
1677
1678         error = PROC_I(inode)->op.proc_get_link(inode, &path);
1679         if (error)
1680                 goto out;
1681
1682         error = do_proc_readlink(&path, buffer, buflen);
1683         path_put(&path);
1684 out:
1685         return error;
1686 }
1687
1688 static const struct inode_operations proc_pid_link_inode_operations = {
1689         .readlink       = proc_pid_readlink,
1690         .follow_link    = proc_pid_follow_link,
1691         .setattr        = proc_setattr,
1692 };
1693
1694
1695 /* building an inode */
1696
1697 static int task_dumpable(struct task_struct *task)
1698 {
1699         int dumpable = 0;
1700         struct mm_struct *mm;
1701
1702         task_lock(task);
1703         mm = task->mm;
1704         if (mm)
1705                 dumpable = get_dumpable(mm);
1706         task_unlock(task);
1707         if(dumpable == 1)
1708                 return 1;
1709         return 0;
1710 }
1711
1712
1713 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1714 {
1715         struct inode * inode;
1716         struct proc_inode *ei;
1717         const struct cred *cred;
1718
1719         /* We need a new inode */
1720
1721         inode = new_inode(sb);
1722         if (!inode)
1723                 goto out;
1724
1725         /* Common stuff */
1726         ei = PROC_I(inode);
1727         inode->i_ino = get_next_ino();
1728         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1729         inode->i_op = &proc_def_inode_operations;
1730
1731         /*
1732          * grab the reference to task.
1733          */
1734         ei->pid = get_task_pid(task, PIDTYPE_PID);
1735         if (!ei->pid)
1736                 goto out_unlock;
1737
1738         if (task_dumpable(task)) {
1739                 rcu_read_lock();
1740                 cred = __task_cred(task);
1741                 inode->i_uid = cred->euid;
1742                 inode->i_gid = cred->egid;
1743                 rcu_read_unlock();
1744         }
1745         security_task_to_inode(task, inode);
1746
1747 out:
1748         return inode;
1749
1750 out_unlock:
1751         iput(inode);
1752         return NULL;
1753 }
1754
1755 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1756 {
1757         struct inode *inode = dentry->d_inode;
1758         struct task_struct *task;
1759         const struct cred *cred;
1760
1761         generic_fillattr(inode, stat);
1762
1763         rcu_read_lock();
1764         stat->uid = 0;
1765         stat->gid = 0;
1766         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1767         if (task) {
1768                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1769                     task_dumpable(task)) {
1770                         cred = __task_cred(task);
1771                         stat->uid = cred->euid;
1772                         stat->gid = cred->egid;
1773                 }
1774         }
1775         rcu_read_unlock();
1776         return 0;
1777 }
1778
1779 /* dentry stuff */
1780
1781 /*
1782  *      Exceptional case: normally we are not allowed to unhash a busy
1783  * directory. In this case, however, we can do it - no aliasing problems
1784  * due to the way we treat inodes.
1785  *
1786  * Rewrite the inode's ownerships here because the owning task may have
1787  * performed a setuid(), etc.
1788  *
1789  * Before the /proc/pid/status file was created the only way to read
1790  * the effective uid of a /process was to stat /proc/pid.  Reading
1791  * /proc/pid/status is slow enough that procps and other packages
1792  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1793  * made this apply to all per process world readable and executable
1794  * directories.
1795  */
1796 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1797 {
1798         struct inode *inode;
1799         struct task_struct *task;
1800         const struct cred *cred;
1801
1802         if (nd && nd->flags & LOOKUP_RCU)
1803                 return -ECHILD;
1804
1805         inode = dentry->d_inode;
1806         task = get_proc_task(inode);
1807
1808         if (task) {
1809                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1810                     task_dumpable(task)) {
1811                         rcu_read_lock();
1812                         cred = __task_cred(task);
1813                         inode->i_uid = cred->euid;
1814                         inode->i_gid = cred->egid;
1815                         rcu_read_unlock();
1816                 } else {
1817                         inode->i_uid = 0;
1818                         inode->i_gid = 0;
1819                 }
1820                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1821                 security_task_to_inode(task, inode);
1822                 put_task_struct(task);
1823                 return 1;
1824         }
1825         d_drop(dentry);
1826         return 0;
1827 }
1828
1829 static int pid_delete_dentry(const struct dentry * dentry)
1830 {
1831         /* Is the task we represent dead?
1832          * If so, then don't put the dentry on the lru list,
1833          * kill it immediately.
1834          */
1835         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1836 }
1837
1838 static const struct dentry_operations pid_dentry_operations =
1839 {
1840         .d_revalidate   = pid_revalidate,
1841         .d_delete       = pid_delete_dentry,
1842 };
1843
1844 /* Lookups */
1845
1846 typedef struct dentry *instantiate_t(struct inode *, struct dentry *,
1847                                 struct task_struct *, const void *);
1848
1849 /*
1850  * Fill a directory entry.
1851  *
1852  * If possible create the dcache entry and derive our inode number and
1853  * file type from dcache entry.
1854  *
1855  * Since all of the proc inode numbers are dynamically generated, the inode
1856  * numbers do not exist until the inode is cache.  This means creating the
1857  * the dcache entry in readdir is necessary to keep the inode numbers
1858  * reported by readdir in sync with the inode numbers reported
1859  * by stat.
1860  */
1861 static int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1862         char *name, int len,
1863         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1864 {
1865         struct dentry *child, *dir = filp->f_path.dentry;
1866         struct inode *inode;
1867         struct qstr qname;
1868         ino_t ino = 0;
1869         unsigned type = DT_UNKNOWN;
1870
1871         qname.name = name;
1872         qname.len  = len;
1873         qname.hash = full_name_hash(name, len);
1874
1875         child = d_lookup(dir, &qname);
1876         if (!child) {
1877                 struct dentry *new;
1878                 new = d_alloc(dir, &qname);
1879                 if (new) {
1880                         child = instantiate(dir->d_inode, new, task, ptr);
1881                         if (child)
1882                                 dput(new);
1883                         else
1884                                 child = new;
1885                 }
1886         }
1887         if (!child || IS_ERR(child) || !child->d_inode)
1888                 goto end_instantiate;
1889         inode = child->d_inode;
1890         if (inode) {
1891                 ino = inode->i_ino;
1892                 type = inode->i_mode >> 12;
1893         }
1894         dput(child);
1895 end_instantiate:
1896         if (!ino)
1897                 ino = find_inode_number(dir, &qname);
1898         if (!ino)
1899                 ino = 1;
1900         return filldir(dirent, name, len, filp->f_pos, ino, type);
1901 }
1902
1903 static unsigned name_to_int(struct dentry *dentry)
1904 {
1905         const char *name = dentry->d_name.name;
1906         int len = dentry->d_name.len;
1907         unsigned n = 0;
1908
1909         if (len > 1 && *name == '0')
1910                 goto out;
1911         while (len-- > 0) {
1912                 unsigned c = *name++ - '0';
1913                 if (c > 9)
1914                         goto out;
1915                 if (n >= (~0U-9)/10)
1916                         goto out;
1917                 n *= 10;
1918                 n += c;
1919         }
1920         return n;
1921 out:
1922         return ~0U;
1923 }
1924
1925 #define PROC_FDINFO_MAX 64
1926
1927 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1928 {
1929         struct task_struct *task = get_proc_task(inode);
1930         struct files_struct *files = NULL;
1931         struct file *file;
1932         int fd = proc_fd(inode);
1933
1934         if (task) {
1935                 files = get_files_struct(task);
1936                 put_task_struct(task);
1937         }
1938         if (files) {
1939                 /*
1940                  * We are not taking a ref to the file structure, so we must
1941                  * hold ->file_lock.
1942                  */
1943                 spin_lock(&files->file_lock);
1944                 file = fcheck_files(files, fd);
1945                 if (file) {
1946                         if (path) {
1947                                 *path = file->f_path;
1948                                 path_get(&file->f_path);
1949                         }
1950                         if (info)
1951                                 snprintf(info, PROC_FDINFO_MAX,
1952                                          "pos:\t%lli\n"
1953                                          "flags:\t0%o\n",
1954                                          (long long) file->f_pos,
1955                                          file->f_flags);
1956                         spin_unlock(&files->file_lock);
1957                         put_files_struct(files);
1958                         return 0;
1959                 }
1960                 spin_unlock(&files->file_lock);
1961                 put_files_struct(files);
1962         }
1963         return -ENOENT;
1964 }
1965
1966 static int proc_fd_link(struct inode *inode, struct path *path)
1967 {
1968         return proc_fd_info(inode, path, NULL);
1969 }
1970
1971 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1972 {
1973         struct inode *inode;
1974         struct task_struct *task;
1975         int fd;
1976         struct files_struct *files;
1977         const struct cred *cred;
1978
1979         if (nd && nd->flags & LOOKUP_RCU)
1980                 return -ECHILD;
1981
1982         inode = dentry->d_inode;
1983         task = get_proc_task(inode);
1984         fd = proc_fd(inode);
1985
1986         if (task) {
1987                 files = get_files_struct(task);
1988                 if (files) {
1989                         rcu_read_lock();
1990                         if (fcheck_files(files, fd)) {
1991                                 rcu_read_unlock();
1992                                 put_files_struct(files);
1993                                 if (task_dumpable(task)) {
1994                                         rcu_read_lock();
1995                                         cred = __task_cred(task);
1996                                         inode->i_uid = cred->euid;
1997                                         inode->i_gid = cred->egid;
1998                                         rcu_read_unlock();
1999                                 } else {
2000                                         inode->i_uid = 0;
2001                                         inode->i_gid = 0;
2002                                 }
2003                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
2004                                 security_task_to_inode(task, inode);
2005                                 put_task_struct(task);
2006                                 return 1;
2007                         }
2008                         rcu_read_unlock();
2009                         put_files_struct(files);
2010                 }
2011                 put_task_struct(task);
2012         }
2013         d_drop(dentry);
2014         return 0;
2015 }
2016
2017 static const struct dentry_operations tid_fd_dentry_operations =
2018 {
2019         .d_revalidate   = tid_fd_revalidate,
2020         .d_delete       = pid_delete_dentry,
2021 };
2022
2023 static struct dentry *proc_fd_instantiate(struct inode *dir,
2024         struct dentry *dentry, struct task_struct *task, const void *ptr)
2025 {
2026         unsigned fd = *(const unsigned *)ptr;
2027         struct file *file;
2028         struct files_struct *files;
2029         struct inode *inode;
2030         struct proc_inode *ei;
2031         struct dentry *error = ERR_PTR(-ENOENT);
2032
2033         inode = proc_pid_make_inode(dir->i_sb, task);
2034         if (!inode)
2035                 goto out;
2036         ei = PROC_I(inode);
2037         ei->fd = fd;
2038         files = get_files_struct(task);
2039         if (!files)
2040                 goto out_iput;
2041         inode->i_mode = S_IFLNK;
2042
2043         /*
2044          * We are not taking a ref to the file structure, so we must
2045          * hold ->file_lock.
2046          */
2047         spin_lock(&files->file_lock);
2048         file = fcheck_files(files, fd);
2049         if (!file)
2050                 goto out_unlock;
2051         if (file->f_mode & FMODE_READ)
2052                 inode->i_mode |= S_IRUSR | S_IXUSR;
2053         if (file->f_mode & FMODE_WRITE)
2054                 inode->i_mode |= S_IWUSR | S_IXUSR;
2055         spin_unlock(&files->file_lock);
2056         put_files_struct(files);
2057
2058         inode->i_op = &proc_pid_link_inode_operations;
2059         inode->i_size = 64;
2060         ei->op.proc_get_link = proc_fd_link;
2061         d_set_d_op(dentry, &tid_fd_dentry_operations);
2062         d_add(dentry, inode);
2063         /* Close the race of the process dying before we return the dentry */
2064         if (tid_fd_revalidate(dentry, NULL))
2065                 error = NULL;
2066
2067  out:
2068         return error;
2069 out_unlock:
2070         spin_unlock(&files->file_lock);
2071         put_files_struct(files);
2072 out_iput:
2073         iput(inode);
2074         goto out;
2075 }
2076
2077 static struct dentry *proc_lookupfd_common(struct inode *dir,
2078                                            struct dentry *dentry,
2079                                            instantiate_t instantiate)
2080 {
2081         struct task_struct *task = get_proc_task(dir);
2082         unsigned fd = name_to_int(dentry);
2083         struct dentry *result = ERR_PTR(-ENOENT);
2084
2085         if (!task)
2086                 goto out_no_task;
2087         if (fd == ~0U)
2088                 goto out;
2089
2090         result = instantiate(dir, dentry, task, &fd);
2091 out:
2092         put_task_struct(task);
2093 out_no_task:
2094         return result;
2095 }
2096
2097 static int proc_readfd_common(struct file * filp, void * dirent,
2098                               filldir_t filldir, instantiate_t instantiate)
2099 {
2100         struct dentry *dentry = filp->f_path.dentry;
2101         struct inode *inode = dentry->d_inode;
2102         struct task_struct *p = get_proc_task(inode);
2103         unsigned int fd, ino;
2104         int retval;
2105         struct files_struct * files;
2106
2107         retval = -ENOENT;
2108         if (!p)
2109                 goto out_no_task;
2110         retval = 0;
2111
2112         fd = filp->f_pos;
2113         switch (fd) {
2114                 case 0:
2115                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
2116                                 goto out;
2117                         filp->f_pos++;
2118                 case 1:
2119                         ino = parent_ino(dentry);
2120                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
2121                                 goto out;
2122                         filp->f_pos++;
2123                 default:
2124                         files = get_files_struct(p);
2125                         if (!files)
2126                                 goto out;
2127                         rcu_read_lock();
2128                         for (fd = filp->f_pos-2;
2129                              fd < files_fdtable(files)->max_fds;
2130                              fd++, filp->f_pos++) {
2131                                 char name[PROC_NUMBUF];
2132                                 int len;
2133
2134                                 if (!fcheck_files(files, fd))
2135                                         continue;
2136                                 rcu_read_unlock();
2137
2138                                 len = snprintf(name, sizeof(name), "%d", fd);
2139                                 if (proc_fill_cache(filp, dirent, filldir,
2140                                                     name, len, instantiate,
2141                                                     p, &fd) < 0) {
2142                                         rcu_read_lock();
2143                                         break;
2144                                 }
2145                                 rcu_read_lock();
2146                         }
2147                         rcu_read_unlock();
2148                         put_files_struct(files);
2149         }
2150 out:
2151         put_task_struct(p);
2152 out_no_task:
2153         return retval;
2154 }
2155
2156 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
2157                                     struct nameidata *nd)
2158 {
2159         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
2160 }
2161
2162 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
2163 {
2164         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
2165 }
2166
2167 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
2168                                       size_t len, loff_t *ppos)
2169 {
2170         char tmp[PROC_FDINFO_MAX];
2171         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
2172         if (!err)
2173                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
2174         return err;
2175 }
2176
2177 static const struct file_operations proc_fdinfo_file_operations = {
2178         .open           = nonseekable_open,
2179         .read           = proc_fdinfo_read,
2180         .llseek         = no_llseek,
2181 };
2182
2183 static const struct file_operations proc_fd_operations = {
2184         .read           = generic_read_dir,
2185         .readdir        = proc_readfd,
2186         .llseek         = default_llseek,
2187 };
2188
2189 /*
2190  * /proc/pid/fd needs a special permission handler so that a process can still
2191  * access /proc/self/fd after it has executed a setuid().
2192  */
2193 static int proc_fd_permission(struct inode *inode, int mask, unsigned int flags)
2194 {
2195         int rv;
2196
2197         if (flags & IPERM_FLAG_RCU)
2198                 return -ECHILD;
2199         rv = generic_permission(inode, mask, flags, NULL);
2200         if (rv == 0)
2201                 return 0;
2202         if (task_pid(current) == proc_pid(inode))
2203                 rv = 0;
2204         return rv;
2205 }
2206
2207 /*
2208  * proc directories can do almost nothing..
2209  */
2210 static const struct inode_operations proc_fd_inode_operations = {
2211         .lookup         = proc_lookupfd,
2212         .permission     = proc_fd_permission,
2213         .setattr        = proc_setattr,
2214 };
2215
2216 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
2217         struct dentry *dentry, struct task_struct *task, const void *ptr)
2218 {
2219         unsigned fd = *(unsigned *)ptr;
2220         struct inode *inode;
2221         struct proc_inode *ei;
2222         struct dentry *error = ERR_PTR(-ENOENT);
2223
2224         inode = proc_pid_make_inode(dir->i_sb, task);
2225         if (!inode)
2226                 goto out;
2227         ei = PROC_I(inode);
2228         ei->fd = fd;
2229         inode->i_mode = S_IFREG | S_IRUSR;
2230         inode->i_fop = &proc_fdinfo_file_operations;
2231         d_set_d_op(dentry, &tid_fd_dentry_operations);
2232         d_add(dentry, inode);
2233         /* Close the race of the process dying before we return the dentry */
2234         if (tid_fd_revalidate(dentry, NULL))
2235                 error = NULL;
2236
2237  out:
2238         return error;
2239 }
2240
2241 static struct dentry *proc_lookupfdinfo(struct inode *dir,
2242                                         struct dentry *dentry,
2243                                         struct nameidata *nd)
2244 {
2245         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
2246 }
2247
2248 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
2249 {
2250         return proc_readfd_common(filp, dirent, filldir,
2251                                   proc_fdinfo_instantiate);
2252 }
2253
2254 static const struct file_operations proc_fdinfo_operations = {
2255         .read           = generic_read_dir,
2256         .readdir        = proc_readfdinfo,
2257         .llseek         = default_llseek,
2258 };
2259
2260 /*
2261  * proc directories can do almost nothing..
2262  */
2263 static const struct inode_operations proc_fdinfo_inode_operations = {
2264         .lookup         = proc_lookupfdinfo,
2265         .setattr        = proc_setattr,
2266 };
2267
2268
2269 static struct dentry *proc_pident_instantiate(struct inode *dir,
2270         struct dentry *dentry, struct task_struct *task, const void *ptr)
2271 {
2272         const struct pid_entry *p = ptr;
2273         struct inode *inode;
2274         struct proc_inode *ei;
2275         struct dentry *error = ERR_PTR(-ENOENT);
2276
2277         inode = proc_pid_make_inode(dir->i_sb, task);
2278         if (!inode)
2279                 goto out;
2280
2281         ei = PROC_I(inode);
2282         inode->i_mode = p->mode;
2283         if (S_ISDIR(inode->i_mode))
2284                 inode->i_nlink = 2;     /* Use getattr to fix if necessary */
2285         if (p->iop)
2286                 inode->i_op = p->iop;
2287         if (p->fop)
2288                 inode->i_fop = p->fop;
2289         ei->op = p->op;
2290         d_set_d_op(dentry, &pid_dentry_operations);
2291         d_add(dentry, inode);
2292         /* Close the race of the process dying before we return the dentry */
2293         if (pid_revalidate(dentry, NULL))
2294                 error = NULL;
2295 out:
2296         return error;
2297 }
2298
2299 static struct dentry *proc_pident_lookup(struct inode *dir,
2300                                          struct dentry *dentry,
2301                                          const struct pid_entry *ents,
2302                                          unsigned int nents)
2303 {
2304         struct dentry *error;
2305         struct task_struct *task = get_proc_task(dir);
2306         const struct pid_entry *p, *last;
2307
2308         error = ERR_PTR(-ENOENT);
2309
2310         if (!task)
2311                 goto out_no_task;
2312
2313         /*
2314          * Yes, it does not scale. And it should not. Don't add
2315          * new entries into /proc/<tgid>/ without very good reasons.
2316          */
2317         last = &ents[nents - 1];
2318         for (p = ents; p <= last; p++) {
2319                 if (p->len != dentry->d_name.len)
2320                         continue;
2321                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2322                         break;
2323         }
2324         if (p > last)
2325                 goto out;
2326
2327         error = proc_pident_instantiate(dir, dentry, task, p);
2328 out:
2329         put_task_struct(task);
2330 out_no_task:
2331         return error;
2332 }
2333
2334 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2335         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2336 {
2337         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2338                                 proc_pident_instantiate, task, p);
2339 }
2340
2341 static int proc_pident_readdir(struct file *filp,
2342                 void *dirent, filldir_t filldir,
2343                 const struct pid_entry *ents, unsigned int nents)
2344 {
2345         int i;
2346         struct dentry *dentry = filp->f_path.dentry;
2347         struct inode *inode = dentry->d_inode;
2348         struct task_struct *task = get_proc_task(inode);
2349         const struct pid_entry *p, *last;
2350         ino_t ino;
2351         int ret;
2352
2353         ret = -ENOENT;
2354         if (!task)
2355                 goto out_no_task;
2356
2357         ret = 0;
2358         i = filp->f_pos;
2359         switch (i) {
2360         case 0:
2361                 ino = inode->i_ino;
2362                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2363                         goto out;
2364                 i++;
2365                 filp->f_pos++;
2366                 /* fall through */
2367         case 1:
2368                 ino = parent_ino(dentry);
2369                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2370                         goto out;
2371                 i++;
2372                 filp->f_pos++;
2373                 /* fall through */
2374         default:
2375                 i -= 2;
2376                 if (i >= nents) {
2377                         ret = 1;
2378                         goto out;
2379                 }
2380                 p = ents + i;
2381                 last = &ents[nents - 1];
2382                 while (p <= last) {
2383                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2384                                 goto out;
2385                         filp->f_pos++;
2386                         p++;
2387                 }
2388         }
2389
2390         ret = 1;
2391 out:
2392         put_task_struct(task);
2393 out_no_task:
2394         return ret;
2395 }
2396
2397 #ifdef CONFIG_SECURITY
2398 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2399                                   size_t count, loff_t *ppos)
2400 {
2401         struct inode * inode = file->f_path.dentry->d_inode;
2402         char *p = NULL;
2403         ssize_t length;
2404         struct task_struct *task = get_proc_task(inode);
2405
2406         if (!task)
2407                 return -ESRCH;
2408
2409         length = security_getprocattr(task,
2410                                       (char*)file->f_path.dentry->d_name.name,
2411                                       &p);
2412         put_task_struct(task);
2413         if (length > 0)
2414                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2415         kfree(p);
2416         return length;
2417 }
2418
2419 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2420                                    size_t count, loff_t *ppos)
2421 {
2422         struct inode * inode = file->f_path.dentry->d_inode;
2423         char *page;
2424         ssize_t length;
2425         struct task_struct *task = get_proc_task(inode);
2426
2427         length = -ESRCH;
2428         if (!task)
2429                 goto out_no_task;
2430         if (count > PAGE_SIZE)
2431                 count = PAGE_SIZE;
2432
2433         /* No partial writes. */
2434         length = -EINVAL;
2435         if (*ppos != 0)
2436                 goto out;
2437
2438         length = -ENOMEM;
2439         page = (char*)__get_free_page(GFP_TEMPORARY);
2440         if (!page)
2441                 goto out;
2442
2443         length = -EFAULT;
2444         if (copy_from_user(page, buf, count))
2445                 goto out_free;
2446
2447         /* Guard against adverse ptrace interaction */
2448         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2449         if (length < 0)
2450                 goto out_free;
2451
2452         length = security_setprocattr(task,
2453                                       (char*)file->f_path.dentry->d_name.name,
2454                                       (void*)page, count);
2455         mutex_unlock(&task->signal->cred_guard_mutex);
2456 out_free:
2457         free_page((unsigned long) page);
2458 out:
2459         put_task_struct(task);
2460 out_no_task:
2461         return length;
2462 }
2463
2464 static const struct file_operations proc_pid_attr_operations = {
2465         .read           = proc_pid_attr_read,
2466         .write          = proc_pid_attr_write,
2467         .llseek         = generic_file_llseek,
2468 };
2469
2470 static const struct pid_entry attr_dir_stuff[] = {
2471         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2472         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2473         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2474         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2475         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2476         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2477 };
2478
2479 static int proc_attr_dir_readdir(struct file * filp,
2480                              void * dirent, filldir_t filldir)
2481 {
2482         return proc_pident_readdir(filp,dirent,filldir,
2483                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2484 }
2485
2486 static const struct file_operations proc_attr_dir_operations = {
2487         .read           = generic_read_dir,
2488         .readdir        = proc_attr_dir_readdir,
2489         .llseek         = default_llseek,
2490 };
2491
2492 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2493                                 struct dentry *dentry, struct nameidata *nd)
2494 {
2495         return proc_pident_lookup(dir, dentry,
2496                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2497 }
2498
2499 static const struct inode_operations proc_attr_dir_inode_operations = {
2500         .lookup         = proc_attr_dir_lookup,
2501         .getattr        = pid_getattr,
2502         .setattr        = proc_setattr,
2503 };
2504
2505 #endif
2506
2507 #ifdef CONFIG_ELF_CORE
2508 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2509                                          size_t count, loff_t *ppos)
2510 {
2511         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2512         struct mm_struct *mm;
2513         char buffer[PROC_NUMBUF];
2514         size_t len;
2515         int ret;
2516
2517         if (!task)
2518                 return -ESRCH;
2519
2520         ret = 0;
2521         mm = get_task_mm(task);
2522         if (mm) {
2523                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2524                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2525                                 MMF_DUMP_FILTER_SHIFT));
2526                 mmput(mm);
2527                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2528         }
2529
2530         put_task_struct(task);
2531
2532         return ret;
2533 }
2534
2535 static ssize_t proc_coredump_filter_write(struct file *file,
2536                                           const char __user *buf,
2537                                           size_t count,
2538                                           loff_t *ppos)
2539 {
2540         struct task_struct *task;
2541         struct mm_struct *mm;
2542         char buffer[PROC_NUMBUF], *end;
2543         unsigned int val;
2544         int ret;
2545         int i;
2546         unsigned long mask;
2547
2548         ret = -EFAULT;
2549         memset(buffer, 0, sizeof(buffer));
2550         if (count > sizeof(buffer) - 1)
2551                 count = sizeof(buffer) - 1;
2552         if (copy_from_user(buffer, buf, count))
2553                 goto out_no_task;
2554
2555         ret = -EINVAL;
2556         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2557         if (*end == '\n')
2558                 end++;
2559         if (end - buffer == 0)
2560                 goto out_no_task;
2561
2562         ret = -ESRCH;
2563         task = get_proc_task(file->f_dentry->d_inode);
2564         if (!task)
2565                 goto out_no_task;
2566
2567         ret = end - buffer;
2568         mm = get_task_mm(task);
2569         if (!mm)
2570                 goto out_no_mm;
2571
2572         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2573                 if (val & mask)
2574                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2575                 else
2576                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2577         }
2578
2579         mmput(mm);
2580  out_no_mm:
2581         put_task_struct(task);
2582  out_no_task:
2583         return ret;
2584 }
2585
2586 static const struct file_operations proc_coredump_filter_operations = {
2587         .read           = proc_coredump_filter_read,
2588         .write          = proc_coredump_filter_write,
2589         .llseek         = generic_file_llseek,
2590 };
2591 #endif
2592
2593 /*
2594  * /proc/self:
2595  */
2596 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2597                               int buflen)
2598 {
2599         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2600         pid_t tgid = task_tgid_nr_ns(current, ns);
2601         char tmp[PROC_NUMBUF];
2602         if (!tgid)
2603                 return -ENOENT;
2604         sprintf(tmp, "%d", tgid);
2605         return vfs_readlink(dentry,buffer,buflen,tmp);
2606 }
2607
2608 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2609 {
2610         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2611         pid_t tgid = task_tgid_nr_ns(current, ns);
2612         char *name = ERR_PTR(-ENOENT);
2613         if (tgid) {
2614                 name = __getname();
2615                 if (!name)
2616                         name = ERR_PTR(-ENOMEM);
2617                 else
2618                         sprintf(name, "%d", tgid);
2619         }
2620         nd_set_link(nd, name);
2621         return NULL;
2622 }
2623
2624 static void proc_self_put_link(struct dentry *dentry, struct nameidata *nd,
2625                                 void *cookie)
2626 {
2627         char *s = nd_get_link(nd);
2628         if (!IS_ERR(s))
2629                 __putname(s);
2630 }
2631
2632 static const struct inode_operations proc_self_inode_operations = {
2633         .readlink       = proc_self_readlink,
2634         .follow_link    = proc_self_follow_link,
2635         .put_link       = proc_self_put_link,
2636 };
2637
2638 /*
2639  * proc base
2640  *
2641  * These are the directory entries in the root directory of /proc
2642  * that properly belong to the /proc filesystem, as they describe
2643  * describe something that is process related.
2644  */
2645 static const struct pid_entry proc_base_stuff[] = {
2646         NOD("self", S_IFLNK|S_IRWXUGO,
2647                 &proc_self_inode_operations, NULL, {}),
2648 };
2649
2650 /*
2651  *      Exceptional case: normally we are not allowed to unhash a busy
2652  * directory. In this case, however, we can do it - no aliasing problems
2653  * due to the way we treat inodes.
2654  */
2655 static int proc_base_revalidate(struct dentry *dentry, struct nameidata *nd)
2656 {
2657         struct inode *inode;
2658         struct task_struct *task;
2659
2660         if (nd->flags & LOOKUP_RCU)
2661                 return -ECHILD;
2662
2663         inode = dentry->d_inode;
2664         task = get_proc_task(inode);
2665         if (task) {
2666                 put_task_struct(task);
2667                 return 1;
2668         }
2669         d_drop(dentry);
2670         return 0;
2671 }
2672
2673 static const struct dentry_operations proc_base_dentry_operations =
2674 {
2675         .d_revalidate   = proc_base_revalidate,
2676         .d_delete       = pid_delete_dentry,
2677 };
2678
2679 static struct dentry *proc_base_instantiate(struct inode *dir,
2680         struct dentry *dentry, struct task_struct *task, const void *ptr)
2681 {
2682         const struct pid_entry *p = ptr;
2683         struct inode *inode;
2684         struct proc_inode *ei;
2685         struct dentry *error;
2686
2687         /* Allocate the inode */
2688         error = ERR_PTR(-ENOMEM);
2689         inode = new_inode(dir->i_sb);
2690         if (!inode)
2691                 goto out;
2692
2693         /* Initialize the inode */
2694         ei = PROC_I(inode);
2695         inode->i_ino = get_next_ino();
2696         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2697
2698         /*
2699          * grab the reference to the task.
2700          */
2701         ei->pid = get_task_pid(task, PIDTYPE_PID);
2702         if (!ei->pid)
2703                 goto out_iput;
2704
2705         inode->i_mode = p->mode;
2706         if (S_ISDIR(inode->i_mode))
2707                 inode->i_nlink = 2;
2708         if (S_ISLNK(inode->i_mode))
2709                 inode->i_size = 64;
2710         if (p->iop)
2711                 inode->i_op = p->iop;
2712         if (p->fop)
2713                 inode->i_fop = p->fop;
2714         ei->op = p->op;
2715         d_set_d_op(dentry, &proc_base_dentry_operations);
2716         d_add(dentry, inode);
2717         error = NULL;
2718 out:
2719         return error;
2720 out_iput:
2721         iput(inode);
2722         goto out;
2723 }
2724
2725 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2726 {
2727         struct dentry *error;
2728         struct task_struct *task = get_proc_task(dir);
2729         const struct pid_entry *p, *last;
2730
2731         error = ERR_PTR(-ENOENT);
2732
2733         if (!task)
2734                 goto out_no_task;
2735
2736         /* Lookup the directory entry */
2737         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2738         for (p = proc_base_stuff; p <= last; p++) {
2739                 if (p->len != dentry->d_name.len)
2740                         continue;
2741                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2742                         break;
2743         }
2744         if (p > last)
2745                 goto out;
2746
2747         error = proc_base_instantiate(dir, dentry, task, p);
2748
2749 out:
2750         put_task_struct(task);
2751 out_no_task:
2752         return error;
2753 }
2754
2755 static int proc_base_fill_cache(struct file *filp, void *dirent,
2756         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2757 {
2758         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2759                                 proc_base_instantiate, task, p);
2760 }
2761
2762 #ifdef CONFIG_TASK_IO_ACCOUNTING
2763 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2764 {
2765         struct task_io_accounting acct = task->ioac;
2766         unsigned long flags;
2767
2768         if (whole && lock_task_sighand(task, &flags)) {
2769                 struct task_struct *t = task;
2770
2771                 task_io_accounting_add(&acct, &task->signal->ioac);
2772                 while_each_thread(task, t)
2773                         task_io_accounting_add(&acct, &t->ioac);
2774
2775                 unlock_task_sighand(task, &flags);
2776         }
2777         return sprintf(buffer,
2778                         "rchar: %llu\n"
2779                         "wchar: %llu\n"
2780                         "syscr: %llu\n"
2781                         "syscw: %llu\n"
2782                         "read_bytes: %llu\n"
2783                         "write_bytes: %llu\n"
2784                         "cancelled_write_bytes: %llu\n",
2785                         (unsigned long long)acct.rchar,
2786                         (unsigned long long)acct.wchar,
2787                         (unsigned long long)acct.syscr,
2788                         (unsigned long long)acct.syscw,
2789                         (unsigned long long)acct.read_bytes,
2790                         (unsigned long long)acct.write_bytes,
2791                         (unsigned long long)acct.cancelled_write_bytes);
2792 }
2793
2794 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2795 {
2796         return do_io_accounting(task, buffer, 0);
2797 }
2798
2799 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2800 {
2801         return do_io_accounting(task, buffer, 1);
2802 }
2803 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2804
2805 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2806                                 struct pid *pid, struct task_struct *task)
2807 {
2808         seq_printf(m, "%08x\n", task->personality);
2809         return 0;
2810 }
2811
2812 /*
2813  * Thread groups
2814  */
2815 static const struct file_operations proc_task_operations;
2816 static const struct inode_operations proc_task_inode_operations;
2817
2818 static const struct pid_entry tgid_base_stuff[] = {
2819         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2820         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2821         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2822 #ifdef CONFIG_NET
2823         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2824 #endif
2825         REG("environ",    S_IRUSR, proc_environ_operations),
2826         INF("auxv",       S_IRUSR, proc_pid_auxv),
2827         ONE("status",     S_IRUGO, proc_pid_status),
2828         ONE("personality", S_IRUSR, proc_pid_personality),
2829         INF("limits",     S_IRUGO, proc_pid_limits),
2830 #ifdef CONFIG_SCHED_DEBUG
2831         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2832 #endif
2833 #ifdef CONFIG_SCHED_AUTOGROUP
2834         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2835 #endif
2836         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2837 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2838         INF("syscall",    S_IRUSR, proc_pid_syscall),
2839 #endif
2840         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2841         ONE("stat",       S_IRUGO, proc_tgid_stat),
2842         ONE("statm",      S_IRUGO, proc_pid_statm),
2843         REG("maps",       S_IRUGO, proc_maps_operations),
2844 #ifdef CONFIG_NUMA
2845         REG("numa_maps",  S_IRUGO, proc_numa_maps_operations),
2846 #endif
2847         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2848         LNK("cwd",        proc_cwd_link),
2849         LNK("root",       proc_root_link),
2850         LNK("exe",        proc_exe_link),
2851         REG("mounts",     S_IRUGO, proc_mounts_operations),
2852         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2853         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2854 #ifdef CONFIG_PROC_PAGE_MONITOR
2855         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2856         REG("smaps",      S_IRUGO, proc_smaps_operations),
2857         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2858 #endif
2859 #ifdef CONFIG_SECURITY
2860         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2861 #endif
2862 #ifdef CONFIG_KALLSYMS
2863         INF("wchan",      S_IRUGO, proc_pid_wchan),
2864 #endif
2865 #ifdef CONFIG_STACKTRACE
2866         ONE("stack",      S_IRUSR, proc_pid_stack),
2867 #endif
2868 #ifdef CONFIG_SCHEDSTATS
2869         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
2870 #endif
2871 #ifdef CONFIG_LATENCYTOP
2872         REG("latency",  S_IRUGO, proc_lstats_operations),
2873 #endif
2874 #ifdef CONFIG_PROC_PID_CPUSET
2875         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2876 #endif
2877 #ifdef CONFIG_CGROUPS
2878         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2879 #endif
2880         INF("oom_score",  S_IRUGO, proc_oom_score),
2881         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2882         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2883 #ifdef CONFIG_AUDITSYSCALL
2884         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2885         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2886 #endif
2887 #ifdef CONFIG_FAULT_INJECTION
2888         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2889 #endif
2890 #ifdef CONFIG_ELF_CORE
2891         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2892 #endif
2893 #ifdef CONFIG_TASK_IO_ACCOUNTING
2894         INF("io",       S_IRUGO, proc_tgid_io_accounting),
2895 #endif
2896 };
2897
2898 static int proc_tgid_base_readdir(struct file * filp,
2899                              void * dirent, filldir_t filldir)
2900 {
2901         return proc_pident_readdir(filp,dirent,filldir,
2902                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2903 }
2904
2905 static const struct file_operations proc_tgid_base_operations = {
2906         .read           = generic_read_dir,
2907         .readdir        = proc_tgid_base_readdir,
2908         .llseek         = default_llseek,
2909 };
2910
2911 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2912         return proc_pident_lookup(dir, dentry,
2913                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2914 }
2915
2916 static const struct inode_operations proc_tgid_base_inode_operations = {
2917         .lookup         = proc_tgid_base_lookup,
2918         .getattr        = pid_getattr,
2919         .setattr        = proc_setattr,
2920 };
2921
2922 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2923 {
2924         struct dentry *dentry, *leader, *dir;
2925         char buf[PROC_NUMBUF];
2926         struct qstr name;
2927
2928         name.name = buf;
2929         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2930         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2931         if (dentry) {
2932                 shrink_dcache_parent(dentry);
2933                 d_drop(dentry);
2934                 dput(dentry);
2935         }
2936
2937         name.name = buf;
2938         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2939         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2940         if (!leader)
2941                 goto out;
2942
2943         name.name = "task";
2944         name.len = strlen(name.name);
2945         dir = d_hash_and_lookup(leader, &name);
2946         if (!dir)
2947                 goto out_put_leader;
2948
2949         name.name = buf;
2950         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2951         dentry = d_hash_and_lookup(dir, &name);
2952         if (dentry) {
2953                 shrink_dcache_parent(dentry);
2954                 d_drop(dentry);
2955                 dput(dentry);
2956         }
2957
2958         dput(dir);
2959 out_put_leader:
2960         dput(leader);
2961 out:
2962         return;
2963 }
2964
2965 /**
2966  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2967  * @task: task that should be flushed.
2968  *
2969  * When flushing dentries from proc, one needs to flush them from global
2970  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2971  * in. This call is supposed to do all of this job.
2972  *
2973  * Looks in the dcache for
2974  * /proc/@pid
2975  * /proc/@tgid/task/@pid
2976  * if either directory is present flushes it and all of it'ts children
2977  * from the dcache.
2978  *
2979  * It is safe and reasonable to cache /proc entries for a task until
2980  * that task exits.  After that they just clog up the dcache with
2981  * useless entries, possibly causing useful dcache entries to be
2982  * flushed instead.  This routine is proved to flush those useless
2983  * dcache entries at process exit time.
2984  *
2985  * NOTE: This routine is just an optimization so it does not guarantee
2986  *       that no dcache entries will exist at process exit time it
2987  *       just makes it very unlikely that any will persist.
2988  */
2989
2990 void proc_flush_task(struct task_struct *task)
2991 {
2992         int i;
2993         struct pid *pid, *tgid;
2994         struct upid *upid;
2995
2996         pid = task_pid(task);
2997         tgid = task_tgid(task);
2998
2999         for (i = 0; i <= pid->level; i++) {
3000                 upid = &pid->numbers[i];
3001                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3002                                         tgid->numbers[i].nr);
3003         }
3004
3005         upid = &pid->numbers[pid->level];
3006         if (upid->nr == 1)
3007                 pid_ns_release_proc(upid->ns);
3008 }
3009
3010 static struct dentry *proc_pid_instantiate(struct inode *dir,
3011                                            struct dentry * dentry,
3012                                            struct task_struct *task, const void *ptr)
3013 {
3014         struct dentry *error = ERR_PTR(-ENOENT);
3015         struct inode *inode;
3016
3017         inode = proc_pid_make_inode(dir->i_sb, task);
3018         if (!inode)
3019                 goto out;
3020
3021         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3022         inode->i_op = &proc_tgid_base_inode_operations;
3023         inode->i_fop = &proc_tgid_base_operations;
3024         inode->i_flags|=S_IMMUTABLE;
3025
3026         inode->i_nlink = 2 + pid_entry_count_dirs(tgid_base_stuff,
3027                 ARRAY_SIZE(tgid_base_stuff));
3028
3029         d_set_d_op(dentry, &pid_dentry_operations);
3030
3031         d_add(dentry, inode);
3032         /* Close the race of the process dying before we return the dentry */
3033         if (pid_revalidate(dentry, NULL))
3034                 error = NULL;
3035 out:
3036         return error;
3037 }
3038
3039 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3040 {
3041         struct dentry *result;
3042         struct task_struct *task;
3043         unsigned tgid;
3044         struct pid_namespace *ns;
3045
3046         result = proc_base_lookup(dir, dentry);
3047         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
3048                 goto out;
3049
3050         tgid = name_to_int(dentry);
3051         if (tgid == ~0U)
3052                 goto out;
3053
3054         ns = dentry->d_sb->s_fs_info;
3055         rcu_read_lock();
3056         task = find_task_by_pid_ns(tgid, ns);
3057         if (task)
3058                 get_task_struct(task);
3059         rcu_read_unlock();
3060         if (!task)
3061                 goto out;
3062
3063         result = proc_pid_instantiate(dir, dentry, task, NULL);
3064         put_task_struct(task);
3065 out:
3066         return result;
3067 }
3068
3069 /*
3070  * Find the first task with tgid >= tgid
3071  *
3072  */
3073 struct tgid_iter {
3074         unsigned int tgid;
3075         struct task_struct *task;
3076 };
3077 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3078 {
3079         struct pid *pid;
3080
3081         if (iter.task)
3082                 put_task_struct(iter.task);
3083         rcu_read_lock();
3084 retry:
3085         iter.task = NULL;
3086         pid = find_ge_pid(iter.tgid, ns);
3087         if (pid) {
3088                 iter.tgid = pid_nr_ns(pid, ns);
3089                 iter.task = pid_task(pid, PIDTYPE_PID);
3090                 /* What we to know is if the pid we have find is the
3091                  * pid of a thread_group_leader.  Testing for task
3092                  * being a thread_group_leader is the obvious thing
3093                  * todo but there is a window when it fails, due to
3094                  * the pid transfer logic in de_thread.
3095                  *
3096                  * So we perform the straight forward test of seeing
3097                  * if the pid we have found is the pid of a thread
3098                  * group leader, and don't worry if the task we have
3099                  * found doesn't happen to be a thread group leader.
3100                  * As we don't care in the case of readdir.
3101                  */
3102                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3103                         iter.tgid += 1;
3104                         goto retry;
3105                 }
3106                 get_task_struct(iter.task);
3107         }
3108         rcu_read_unlock();
3109         return iter;
3110 }
3111
3112 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
3113
3114 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3115         struct tgid_iter iter)
3116 {
3117         char name[PROC_NUMBUF];
3118         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
3119         return proc_fill_cache(filp, dirent, filldir, name, len,
3120                                 proc_pid_instantiate, iter.task, NULL);
3121 }
3122
3123 /* for the /proc/ directory itself, after non-process stuff has been done */
3124 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
3125 {
3126         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
3127         struct task_struct *reaper = get_proc_task(filp->f_path.dentry->d_inode);
3128         struct tgid_iter iter;
3129         struct pid_namespace *ns;
3130
3131         if (!reaper)
3132                 goto out_no_task;
3133
3134         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
3135                 const struct pid_entry *p = &proc_base_stuff[nr];
3136                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
3137                         goto out;
3138         }
3139
3140         ns = filp->f_dentry->d_sb->s_fs_info;
3141         iter.task = NULL;
3142         iter.tgid = filp->f_pos - TGID_OFFSET;
3143         for (iter = next_tgid(ns, iter);
3144              iter.task;
3145              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3146                 filp->f_pos = iter.tgid + TGID_OFFSET;
3147                 if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
3148                         put_task_struct(iter.task);
3149                         goto out;
3150                 }
3151         }
3152         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
3153 out:
3154         put_task_struct(reaper);
3155 out_no_task:
3156         return 0;
3157 }
3158
3159 /*
3160  * Tasks
3161  */
3162 static const struct pid_entry tid_base_stuff[] = {
3163         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3164         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3165         REG("environ",   S_IRUSR, proc_environ_operations),
3166         INF("auxv",      S_IRUSR, proc_pid_auxv),
3167         ONE("status",    S_IRUGO, proc_pid_status),
3168         ONE("personality", S_IRUSR, proc_pid_personality),
3169         INF("limits",    S_IRUGO, proc_pid_limits),
3170 #ifdef CONFIG_SCHED_DEBUG
3171         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3172 #endif
3173         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3174 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3175         INF("syscall",   S_IRUSR, proc_pid_syscall),
3176 #endif
3177         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
3178         ONE("stat",      S_IRUGO, proc_tid_stat),
3179         ONE("statm",     S_IRUGO, proc_pid_statm),
3180         REG("maps",      S_IRUGO, proc_maps_operations),
3181 #ifdef CONFIG_NUMA
3182         REG("numa_maps", S_IRUGO, proc_numa_maps_operations),
3183 #endif
3184         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3185         LNK("cwd",       proc_cwd_link),
3186         LNK("root",      proc_root_link),
3187         LNK("exe",       proc_exe_link),
3188         REG("mounts",    S_IRUGO, proc_mounts_operations),
3189         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3190 #ifdef CONFIG_PROC_PAGE_MONITOR
3191         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3192         REG("smaps",     S_IRUGO, proc_smaps_operations),
3193         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3194 #endif
3195 #ifdef CONFIG_SECURITY
3196         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3197 #endif
3198 #ifdef CONFIG_KALLSYMS
3199         INF("wchan",     S_IRUGO, proc_pid_wchan),
3200 #endif
3201 #ifdef CONFIG_STACKTRACE
3202         ONE("stack",      S_IRUSR, proc_pid_stack),
3203 #endif
3204 #ifdef CONFIG_SCHEDSTATS
3205         INF("schedstat", S_IRUGO, proc_pid_schedstat),
3206 #endif
3207 #ifdef CONFIG_LATENCYTOP
3208         REG("latency",  S_IRUGO, proc_lstats_operations),
3209 #endif
3210 #ifdef CONFIG_PROC_PID_CPUSET
3211         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
3212 #endif
3213 #ifdef CONFIG_CGROUPS
3214         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3215 #endif
3216         INF("oom_score", S_IRUGO, proc_oom_score),
3217         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3218         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3219 #ifdef CONFIG_AUDITSYSCALL
3220         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3221         REG("sessionid",  S_IRUSR, proc_sessionid_operations),
3222 #endif
3223 #ifdef CONFIG_FAULT_INJECTION
3224         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3225 #endif
3226 #ifdef CONFIG_TASK_IO_ACCOUNTING
3227         INF("io",       S_IRUGO, proc_tid_io_accounting),
3228 #endif
3229 };
3230
3231 static int proc_tid_base_readdir(struct file * filp,
3232                              void * dirent, filldir_t filldir)
3233 {
3234         return proc_pident_readdir(filp,dirent,filldir,
3235                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
3236 }
3237
3238 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3239         return proc_pident_lookup(dir, dentry,
3240                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3241 }
3242
3243 static const struct file_operations proc_tid_base_operations = {
3244         .read           = generic_read_dir,
3245         .readdir        = proc_tid_base_readdir,
3246         .llseek         = default_llseek,
3247 };
3248
3249 static const struct inode_operations proc_tid_base_inode_operations = {
3250         .lookup         = proc_tid_base_lookup,
3251         .getattr        = pid_getattr,
3252         .setattr        = proc_setattr,
3253 };
3254
3255 static struct dentry *proc_task_instantiate(struct inode *dir,
3256         struct dentry *dentry, struct task_struct *task, const void *ptr)
3257 {
3258         struct dentry *error = ERR_PTR(-ENOENT);
3259         struct inode *inode;
3260         inode = proc_pid_make_inode(dir->i_sb, task);
3261
3262         if (!inode)
3263                 goto out;
3264         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3265         inode->i_op = &proc_tid_base_inode_operations;
3266         inode->i_fop = &proc_tid_base_operations;
3267         inode->i_flags|=S_IMMUTABLE;
3268
3269         inode->i_nlink = 2 + pid_entry_count_dirs(tid_base_stuff,
3270                 ARRAY_SIZE(tid_base_stuff));
3271
3272         d_set_d_op(dentry, &pid_dentry_operations);
3273
3274         d_add(dentry, inode);
3275         /* Close the race of the process dying before we return the dentry */
3276         if (pid_revalidate(dentry, NULL))
3277                 error = NULL;
3278 out:
3279         return error;
3280 }
3281
3282 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3283 {
3284         struct dentry *result = ERR_PTR(-ENOENT);
3285         struct task_struct *task;
3286         struct task_struct *leader = get_proc_task(dir);
3287         unsigned tid;
3288         struct pid_namespace *ns;
3289
3290         if (!leader)
3291                 goto out_no_task;
3292
3293         tid = name_to_int(dentry);
3294         if (tid == ~0U)
3295                 goto out;
3296
3297         ns = dentry->d_sb->s_fs_info;
3298         rcu_read_lock();
3299         task = find_task_by_pid_ns(tid, ns);
3300         if (task)
3301                 get_task_struct(task);
3302         rcu_read_unlock();
3303         if (!task)
3304                 goto out;
3305         if (!same_thread_group(leader, task))
3306                 goto out_drop_task;
3307
3308         result = proc_task_instantiate(dir, dentry, task, NULL);
3309 out_drop_task:
3310         put_task_struct(task);
3311 out:
3312         put_task_struct(leader);
3313 out_no_task:
3314         return result;
3315 }
3316
3317 /*
3318  * Find the first tid of a thread group to return to user space.
3319  *
3320  * Usually this is just the thread group leader, but if the users
3321  * buffer was too small or there was a seek into the middle of the
3322  * directory we have more work todo.
3323  *
3324  * In the case of a short read we start with find_task_by_pid.
3325  *
3326  * In the case of a seek we start with the leader and walk nr
3327  * threads past it.
3328  */
3329 static struct task_struct *first_tid(struct task_struct *leader,
3330                 int tid, int nr, struct pid_namespace *ns)
3331 {
3332         struct task_struct *pos;
3333
3334         rcu_read_lock();
3335         /* Attempt to start with the pid of a thread */
3336         if (tid && (nr > 0)) {
3337                 pos = find_task_by_pid_ns(tid, ns);
3338                 if (pos && (pos->group_leader == leader))
3339                         goto found;
3340         }
3341
3342         /* If nr exceeds the number of threads there is nothing todo */
3343         pos = NULL;
3344         if (nr && nr >= get_nr_threads(leader))
3345                 goto out;
3346
3347         /* If we haven't found our starting place yet start
3348          * with the leader and walk nr threads forward.
3349          */
3350         for (pos = leader; nr > 0; --nr) {
3351                 pos = next_thread(pos);
3352                 if (pos == leader) {
3353                         pos = NULL;
3354                         goto out;
3355                 }
3356         }
3357 found:
3358         get_task_struct(pos);
3359 out:
3360         rcu_read_unlock();
3361         return pos;
3362 }
3363
3364 /*
3365  * Find the next thread in the thread list.
3366  * Return NULL if there is an error or no next thread.
3367  *
3368  * The reference to the input task_struct is released.
3369  */
3370 static struct task_struct *next_tid(struct task_struct *start)
3371 {
3372         struct task_struct *pos = NULL;
3373         rcu_read_lock();
3374         if (pid_alive(start)) {
3375                 pos = next_thread(start);
3376                 if (thread_group_leader(pos))
3377                         pos = NULL;
3378                 else
3379                         get_task_struct(pos);
3380         }
3381         rcu_read_unlock();
3382         put_task_struct(start);
3383         return pos;
3384 }
3385
3386 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3387         struct task_struct *task, int tid)
3388 {
3389         char name[PROC_NUMBUF];
3390         int len = snprintf(name, sizeof(name), "%d", tid);
3391         return proc_fill_cache(filp, dirent, filldir, name, len,
3392                                 proc_task_instantiate, task, NULL);
3393 }
3394
3395 /* for the /proc/TGID/task/ directories */
3396 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3397 {
3398         struct dentry *dentry = filp->f_path.dentry;
3399         struct inode *inode = dentry->d_inode;
3400         struct task_struct *leader = NULL;
3401         struct task_struct *task;
3402         int retval = -ENOENT;
3403         ino_t ino;
3404         int tid;
3405         struct pid_namespace *ns;
3406
3407         task = get_proc_task(inode);
3408         if (!task)
3409                 goto out_no_task;
3410         rcu_read_lock();
3411         if (pid_alive(task)) {
3412                 leader = task->group_leader;
3413                 get_task_struct(leader);
3414         }
3415         rcu_read_unlock();
3416         put_task_struct(task);
3417         if (!leader)
3418                 goto out_no_task;
3419         retval = 0;
3420
3421         switch ((unsigned long)filp->f_pos) {
3422         case 0:
3423                 ino = inode->i_ino;
3424                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3425                         goto out;
3426                 filp->f_pos++;
3427                 /* fall through */
3428         case 1:
3429                 ino = parent_ino(dentry);
3430                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3431                         goto out;
3432                 filp->f_pos++;
3433                 /* fall through */
3434         }
3435
3436         /* f_version caches the tgid value that the last readdir call couldn't
3437          * return. lseek aka telldir automagically resets f_version to 0.
3438          */
3439         ns = filp->f_dentry->d_sb->s_fs_info;
3440         tid = (int)filp->f_version;
3441         filp->f_version = 0;
3442         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3443              task;
3444              task = next_tid(task), filp->f_pos++) {
3445                 tid = task_pid_nr_ns(task, ns);
3446                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3447                         /* returning this tgid failed, save it as the first
3448                          * pid for the next readir call */
3449                         filp->f_version = (u64)tid;
3450                         put_task_struct(task);
3451                         break;
3452                 }
3453         }
3454 out:
3455         put_task_struct(leader);
3456 out_no_task:
3457         return retval;
3458 }
3459
3460 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3461 {
3462         struct inode *inode = dentry->d_inode;
3463         struct task_struct *p = get_proc_task(inode);
3464         generic_fillattr(inode, stat);
3465
3466         if (p) {
3467                 stat->nlink += get_nr_threads(p);
3468                 put_task_struct(p);
3469         }
3470
3471         return 0;
3472 }
3473
3474 static const struct inode_operations proc_task_inode_operations = {
3475         .lookup         = proc_task_lookup,
3476         .getattr        = proc_task_getattr,
3477         .setattr        = proc_setattr,
3478 };
3479
3480 static const struct file_operations proc_task_operations = {
3481         .read           = generic_read_dir,
3482         .readdir        = proc_task_readdir,
3483         .llseek         = default_llseek,
3484 };