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