fs/proc/base.c

   1 /*
   2  *  linux/fs/proc/base.c
   3  *
   4  *  Copyright (C) 1991, 1992 Linus Torvalds
   5  *
   6  *  proc base directory handling functions
   7  *
   8  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
   9  *  Instead of using magical inumbers to determine the kind of object
  10  *  we allocate and fill in-core inodes upon lookup. They don't even
  11  *  go into icache. We cache the reference to task_struct upon lookup too.
  12  *  Eventually it should become a filesystem in its own. We don't use the
  13  *  rest of procfs anymore.
  14  *
  15  *
  16  *  Changelog:
  17  *  17-Jan-2005
  18  *  Allan Bezerra
  19  *  Bruna Moreira <bruna.moreira@indt.org.br>
  20  *  Edjard Mota <edjard.mota@indt.org.br>
  21  *  Ilias Biris <ilias.biris@indt.org.br>
  22  *  Mauricio Lin <mauricio.lin@indt.org.br>
  23  *
  24  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  25  *
  26  *  A new process specific entry (smaps) included in /proc. It shows the
  27  *  size of rss for each memory area. The maps entry lacks information
  28  *  about physical memory size (rss) for each mapped file, i.e.,
  29  *  rss information for executables and library files.
  30  *  This additional information is useful for any tools that need to know
  31  *  about physical memory consumption for a process specific library.
  32  *
  33  *  Changelog:
  34  *  21-Feb-2005
  35  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  36  *  Pud inclusion in the page table walking.
  37  *
  38  *  ChangeLog:
  39  *  10-Mar-2005
  40  *  10LE Instituto Nokia de Tecnologia - INdT:
  41  *  A better way to walks through the page table as suggested by Hugh Dickins.
  42  *
  43  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
  44  *  Smaps information related to shared, private, clean and dirty pages.
  45  *
  46  *  Paul Mundt <paul.mundt@nokia.com>:
  47  *  Overall revision about smaps.
  48  */
  49
  50 #include <asm/uaccess.h>
  51
  52 #include <linux/errno.h>
  53 #include <linux/time.h>
  54 #include <linux/proc_fs.h>
  55 #include <linux/stat.h>
  56 #include <linux/task_io_accounting_ops.h>
  57 #include <linux/init.h>
  58 #include <linux/capability.h>
  59 #include <linux/file.h>
  60 #include <linux/fdtable.h>
  61 #include <linux/string.h>
  62 #include <linux/seq_file.h>
  63 #include <linux/namei.h>
  64 #include <linux/mnt_namespace.h>
  65 #include <linux/mm.h>
  66 #include <linux/swap.h>
  67 #include <linux/rcupdate.h>
  68 #include <linux/kallsyms.h>
  69 #include <linux/stacktrace.h>
  70 #include <linux/resource.h>
  71 #include <linux/module.h>
  72 #include <linux/mount.h>
  73 #include <linux/security.h>
  74 #include <linux/ptrace.h>
  75 #include <linux/tracehook.h>
  76 #include <linux/cgroup.h>
  77 #include <linux/cpuset.h>
  78 #include <linux/audit.h>
  79 #include <linux/poll.h>
  80 #include <linux/nsproxy.h>
  81 #include <linux/oom.h>
  82 #include <linux/elf.h>
  83 #include <linux/pid_namespace.h>
  84 #include <linux/user_namespace.h>
  85 #include <linux/fs_struct.h>
  86 #include <linux/slab.h>
  87 #include <linux/flex_array.h>
  88 #ifdef CONFIG_HARDWALL
  89 #include <asm/hardwall.h>
  90 #endif
  91 #include <trace/events/oom.h>
  92 #include "internal.h"
  93 #include "fd.h"
  94
  95 /* NOTE:
  96  *      Implementing inode permission operations in /proc is almost
  97  *      certainly an error.  Permission checks need to happen during
  98  *      each system call not at open time.  The reason is that most of
  99  *      what we wish to check for permissions in /proc varies at runtime.
 100  *
 101  *      The classic example of a problem is opening file descriptors
 102  *      in /proc for a task before it execs a suid executable.
 103  */
 104
 105 struct pid_entry {
 106         char *name;
 107         int len;
 108         umode_t mode;
 109         const struct inode_operations *iop;
 110         const struct file_operations *fop;
 111         union proc_op op;
 112 };
 113
 114 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 115         .name = (NAME),                                 \
 116         .len  = sizeof(NAME) - 1,                       \
 117         .mode = MODE,                                   \
 118         .iop  = IOP,                                    \
 119         .fop  = FOP,                                    \
 120         .op   = OP,                                     \
 121 }
 122
 123 #define DIR(NAME, MODE, iops, fops)     \
 124         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 125 #define LNK(NAME, get_link)                                     \
 126         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 127                 &proc_pid_link_inode_operations, NULL,          \
 128                 { .proc_get_link = get_link } )
 129 #define REG(NAME, MODE, fops)                           \
 130         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 131 #define INF(NAME, MODE, read)                           \
 132         NOD(NAME, (S_IFREG|(MODE)),                     \
 133                 NULL, &proc_info_file_operations,       \
 134                 { .proc_read = read } )
 135 #define ONE(NAME, MODE, show)                           \
 136         NOD(NAME, (S_IFREG|(MODE)),                     \
 137                 NULL, &proc_single_file_operations,     \
 138                 { .proc_show = show } )
 139
 140 /*
 141  * Count the number of hardlinks for the pid_entry table, excluding the .
 142  * and .. links.
 143  */
 144 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
 145         unsigned int n)
 146 {
 147         unsigned int i;
 148         unsigned int count;
 149
 150         count = 0;
 151         for (i = 0; i < n; ++i) {
 152                 if (S_ISDIR(entries[i].mode))
 153                         ++count;
 154         }
 155
 156         return count;
 157 }
 158
 159 static int get_task_root(struct task_struct *task, struct path *root)
 160 {
 161         int result = -ENOENT;
 162
 163         task_lock(task);
 164         if (task->fs) {
 165                 get_fs_root(task->fs, root);
 166                 result = 0;
 167         }
 168         task_unlock(task);
 169         return result;
 170 }
 171
 172 static int proc_cwd_link(struct dentry *dentry, struct path *path)
 173 {
 174         struct task_struct *task = get_proc_task(dentry->d_inode);
 175         int result = -ENOENT;
 176
 177         if (task) {
 178                 task_lock(task);
 179                 if (task->fs) {
 180                         get_fs_pwd(task->fs, path);
 181                         result = 0;
 182                 }
 183                 task_unlock(task);
 184                 put_task_struct(task);
 185         }
 186         return result;
 187 }
 188
 189 static int proc_root_link(struct dentry *dentry, struct path *path)
 190 {
 191         struct task_struct *task = get_proc_task(dentry->d_inode);
 192         int result = -ENOENT;
 193
 194         if (task) {
 195                 result = get_task_root(task, path);
 196                 put_task_struct(task);
 197         }
 198         return result;
 199 }
 200
 201 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
 202 {
 203         int res = 0;
 204         unsigned int len;
 205         struct mm_struct *mm = get_task_mm(task);
 206         if (!mm)
 207                 goto out;
 208         if (!mm->arg_end)
 209                 goto out_mm;    /* Shh! No looking before we're done */
 210
 211         len = mm->arg_end - mm->arg_start;
 212
 213         if (len > PAGE_SIZE)
 214                 len = PAGE_SIZE;
 215
 216         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
 217
 218         // If the nul at the end of args has been overwritten, then
 219         // assume application is using setproctitle(3).
 220         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
 221                 len = strnlen(buffer, res);
 222                 if (len < res) {
 223                     res = len;
 224                 } else {
 225                         len = mm->env_end - mm->env_start;
 226                         if (len > PAGE_SIZE - res)
 227                                 len = PAGE_SIZE - res;
 228                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
 229                         res = strnlen(buffer, res);
 230                 }
 231         }
 232 out_mm:
 233         mmput(mm);
 234 out:
 235         return res;
 236 }
 237
 238 static int proc_pid_auxv(struct task_struct *task, char *buffer)
 239 {
 240         struct mm_struct *mm = mm_access(task, PTRACE_MODE_READ);
 241         int res = PTR_ERR(mm);
 242         if (mm && !IS_ERR(mm)) {
 243                 unsigned int nwords = 0;
 244                 do {
 245                         nwords += 2;
 246                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 247                 res = nwords * sizeof(mm->saved_auxv[0]);
 248                 if (res > PAGE_SIZE)
 249                         res = PAGE_SIZE;
 250                 memcpy(buffer, mm->saved_auxv, res);
 251                 mmput(mm);
 252         }
 253         return res;
 254 }
 255
 256
 257 #ifdef CONFIG_KALLSYMS
 258 /*
 259  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 260  * Returns the resolved symbol.  If that fails, simply return the address.
 261  */
 262 static int proc_pid_wchan(struct task_struct *task, char *buffer)
 263 {
 264         unsigned long wchan;
 265         char symname[KSYM_NAME_LEN];
 266
 267         wchan = get_wchan(task);
 268
 269         if (lookup_symbol_name(wchan, symname) < 0)
 270                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
 271                         return 0;
 272                 else
 273                         return sprintf(buffer, "%lu", wchan);
 274         else
 275                 return sprintf(buffer, "%s", symname);
 276 }
 277 #endif /* CONFIG_KALLSYMS */
 278
 279 static int lock_trace(struct task_struct *task)
 280 {
 281         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
 282         if (err)
 283                 return err;
 284         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH)) {
 285                 mutex_unlock(&task->signal->cred_guard_mutex);
 286                 return -EPERM;
 287         }
 288         return 0;
 289 }
 290
 291 static void unlock_trace(struct task_struct *task)
 292 {
 293         mutex_unlock(&task->signal->cred_guard_mutex);
 294 }
 295
 296 #ifdef CONFIG_STACKTRACE
 297
 298 #define MAX_STACK_TRACE_DEPTH   64
 299
 300 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 301                           struct pid *pid, struct task_struct *task)
 302 {
 303         struct stack_trace trace;
 304         unsigned long *entries;
 305         int err;
 306         int i;
 307
 308         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 309         if (!entries)
 310                 return -ENOMEM;
 311
 312         trace.nr_entries        = 0;
 313         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 314         trace.entries           = entries;
 315         trace.skip              = 0;
 316
 317         err = lock_trace(task);
 318         if (!err) {
 319                 save_stack_trace_tsk(task, &trace);
 320
 321                 for (i = 0; i < trace.nr_entries; i++) {
 322                         seq_printf(m, "[<%pK>] %pS\n",
 323                                    (void *)entries[i], (void *)entries[i]);
 324                 }
 325                 unlock_trace(task);
 326         }
 327         kfree(entries);
 328
 329         return err;
 330 }
 331 #endif
 332
 333 #ifdef CONFIG_SCHEDSTATS
 334 /*
 335  * Provides /proc/PID/schedstat
 336  */
 337 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
 338 {
 339         return sprintf(buffer, "%llu %llu %lu\n",
 340                         (unsigned long long)task->se.sum_exec_runtime,
 341                         (unsigned long long)task->sched_info.run_delay,
 342                         task->sched_info.pcount);
 343 }
 344 #endif
 345
 346 #ifdef CONFIG_LATENCYTOP
 347 static int lstats_show_proc(struct seq_file *m, void *v)
 348 {
 349         int i;
 350         struct inode *inode = m->private;
 351         struct task_struct *task = get_proc_task(inode);
 352
 353         if (!task)
 354                 return -ESRCH;
 355         seq_puts(m, "Latency Top version : v0.1\n");
 356         for (i = 0; i < 32; i++) {
 357                 struct latency_record *lr = &task->latency_record[i];
 358                 if (lr->backtrace[0]) {
 359                         int q;
 360                         seq_printf(m, "%i %li %li",
 361                                    lr->count, lr->time, lr->max);
 362                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 363                                 unsigned long bt = lr->backtrace[q];
 364                                 if (!bt)
 365                                         break;
 366                                 if (bt == ULONG_MAX)
 367                                         break;
 368                                 seq_printf(m, " %ps", (void *)bt);
 369                         }
 370                         seq_putc(m, '\n');
 371                 }
 372
 373         }
 374         put_task_struct(task);
 375         return 0;
 376 }
 377
 378 static int lstats_open(struct inode *inode, struct file *file)
 379 {
 380         return single_open(file, lstats_show_proc, inode);
 381 }
 382
 383 static ssize_t lstats_write(struct file *file, const char __user *buf,
 384                             size_t count, loff_t *offs)
 385 {
 386         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 387
 388         if (!task)
 389                 return -ESRCH;
 390         clear_all_latency_tracing(task);
 391         put_task_struct(task);
 392
 393         return count;
 394 }
 395
 396 static const struct file_operations proc_lstats_operations = {
 397         .open           = lstats_open,
 398         .read           = seq_read,
 399         .write          = lstats_write,
 400         .llseek         = seq_lseek,
 401         .release        = single_release,
 402 };
 403
 404 #endif
 405
 406 static int proc_oom_score(struct task_struct *task, char *buffer)
 407 {
 408         unsigned long totalpages = totalram_pages + total_swap_pages;
 409         unsigned long points = 0;
 410
 411         read_lock(&tasklist_lock);
 412         if (pid_alive(task))
 413                 points = oom_badness(task, NULL, NULL, totalpages) *
 414                                                 1000 / totalpages;
 415         read_unlock(&tasklist_lock);
 416         return sprintf(buffer, "%lu\n", points);
 417 }
 418
 419 struct limit_names {
 420         char *name;
 421         char *unit;
 422 };
 423
 424 static const struct limit_names lnames[RLIM_NLIMITS] = {
 425         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 426         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 427         [RLIMIT_DATA] = {"Max data size", "bytes"},
 428         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 429         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 430         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 431         [RLIMIT_NPROC] = {"Max processes", "processes"},
 432         [RLIMIT_NOFILE] = {"Max open files", "files"},
 433         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 434         [RLIMIT_AS] = {"Max address space", "bytes"},
 435         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 436         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 437         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 438         [RLIMIT_NICE] = {"Max nice priority", NULL},
 439         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 440         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 441 };
 442
 443 /* Display limits for a process */
 444 static int proc_pid_limits(struct task_struct *task, char *buffer)
 445 {
 446         unsigned int i;
 447         int count = 0;
 448         unsigned long flags;
 449         char *bufptr = buffer;
 450
 451         struct rlimit rlim[RLIM_NLIMITS];
 452
 453         if (!lock_task_sighand(task, &flags))
 454                 return 0;
 455         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 456         unlock_task_sighand(task, &flags);
 457
 458         /*
 459          * print the file header
 460          */
 461         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
 462                         "Limit", "Soft Limit", "Hard Limit", "Units");
 463
 464         for (i = 0; i < RLIM_NLIMITS; i++) {
 465                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 466                         count += sprintf(&bufptr[count], "%-25s %-20s ",
 467                                          lnames[i].name, "unlimited");
 468                 else
 469                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
 470                                          lnames[i].name, rlim[i].rlim_cur);
 471
 472                 if (rlim[i].rlim_max == RLIM_INFINITY)
 473                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
 474                 else
 475                         count += sprintf(&bufptr[count], "%-20lu ",
 476                                          rlim[i].rlim_max);
 477
 478                 if (lnames[i].unit)
 479                         count += sprintf(&bufptr[count], "%-10s\n",
 480                                          lnames[i].unit);
 481                 else
 482                         count += sprintf(&bufptr[count], "\n");
 483         }
 484
 485         return count;
 486 }
 487
 488 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 489 static int proc_pid_syscall(struct task_struct *task, char *buffer)
 490 {
 491         long nr;
 492         unsigned long args[6], sp, pc;
 493         int res = lock_trace(task);
 494         if (res)
 495                 return res;
 496
 497         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 498                 res = sprintf(buffer, "running\n");
 499         else if (nr < 0)
 500                 res = sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 501         else
 502                 res = sprintf(buffer,
 503                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 504                        nr,
 505                        args[0], args[1], args[2], args[3], args[4], args[5],
 506                        sp, pc);
 507         unlock_trace(task);
 508         return res;
 509 }
 510 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 511
 512 /************************************************************************/
 513 /*                       Here the fs part begins                        */
 514 /************************************************************************/
 515
 516 /* permission checks */
 517 static int proc_fd_access_allowed(struct inode *inode)
 518 {
 519         struct task_struct *task;
 520         int allowed = 0;
 521         /* Allow access to a task's file descriptors if it is us or we
 522          * may use ptrace attach to the process and find out that
 523          * information.
 524          */
 525         task = get_proc_task(inode);
 526         if (task) {
 527                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
 528                 put_task_struct(task);
 529         }
 530         return allowed;
 531 }
 532
 533 int proc_setattr(struct dentry *dentry, struct iattr *attr)
 534 {
 535         int error;
 536         struct inode *inode = dentry->d_inode;
 537
 538         if (attr->ia_valid & ATTR_MODE)
 539                 return -EPERM;
 540
 541         error = inode_change_ok(inode, attr);
 542         if (error)
 543                 return error;
 544
 545         setattr_copy(inode, attr);
 546         mark_inode_dirty(inode);
 547         return 0;
 548 }
 549
 550 /*
 551  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
 552  * or euid/egid (for hide_pid_min=2)?
 553  */
 554 static bool has_pid_permissions(struct pid_namespace *pid,
 555                                  struct task_struct *task,
 556                                  int hide_pid_min)
 557 {
 558         if (pid->hide_pid < hide_pid_min)
 559                 return true;
 560         if (in_group_p(pid->pid_gid))
 561                 return true;
 562         return ptrace_may_access(task, PTRACE_MODE_READ);
 563 }
 564
 565
 566 static int proc_pid_permission(struct inode *inode, int mask)
 567 {
 568         struct pid_namespace *pid = inode->i_sb->s_fs_info;
 569         struct task_struct *task;
 570         bool has_perms;
 571
 572         task = get_proc_task(inode);
 573         if (!task)
 574                 return -ESRCH;
 575         has_perms = has_pid_permissions(pid, task, 1);
 576         put_task_struct(task);
 577
 578         if (!has_perms) {
 579                 if (pid->hide_pid == 2) {
 580                         /*
 581                          * Let's make getdents(), stat(), and open()
 582                          * consistent with each other.  If a process
 583                          * may not stat() a file, it shouldn't be seen
 584                          * in procfs at all.
 585                          */
 586                         return -ENOENT;
 587                 }
 588
 589                 return -EPERM;
 590         }
 591         return generic_permission(inode, mask);
 592 }
 593
 594
 595
 596 static const struct inode_operations proc_def_inode_operations = {
 597         .setattr        = proc_setattr,
 598 };
 599
 600 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 601
 602 static ssize_t proc_info_read(struct file * file, char __user * buf,
 603                           size_t count, loff_t *ppos)
 604 {
 605         struct inode * inode = file->f_path.dentry->d_inode;
 606         unsigned long page;
 607         ssize_t length;
 608         struct task_struct *task = get_proc_task(inode);
 609
 610         length = -ESRCH;
 611         if (!task)
 612                 goto out_no_task;
 613
 614         if (count > PROC_BLOCK_SIZE)
 615                 count = PROC_BLOCK_SIZE;
 616
 617         length = -ENOMEM;
 618         if (!(page = __get_free_page(GFP_TEMPORARY)))
 619                 goto out;
 620
 621         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 622
 623         if (length >= 0)
 624                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 625         free_page(page);
 626 out:
 627         put_task_struct(task);
 628 out_no_task:
 629         return length;
 630 }
 631
 632 static const struct file_operations proc_info_file_operations = {
 633         .read           = proc_info_read,
 634         .llseek         = generic_file_llseek,
 635 };
 636
 637 static int proc_single_show(struct seq_file *m, void *v)
 638 {
 639         struct inode *inode = m->private;
 640         struct pid_namespace *ns;
 641         struct pid *pid;
 642         struct task_struct *task;
 643         int ret;
 644
 645         ns = inode->i_sb->s_fs_info;
 646         pid = proc_pid(inode);
 647         task = get_pid_task(pid, PIDTYPE_PID);
 648         if (!task)
 649                 return -ESRCH;
 650
 651         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 652
 653         put_task_struct(task);
 654         return ret;
 655 }
 656
 657 static int proc_single_open(struct inode *inode, struct file *filp)
 658 {
 659         return single_open(filp, proc_single_show, inode);
 660 }
 661
 662 static const struct file_operations proc_single_file_operations = {
 663         .open           = proc_single_open,
 664         .read           = seq_read,
 665         .llseek         = seq_lseek,
 666         .release        = single_release,
 667 };
 668
 669 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
 670 {
 671         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 672         struct mm_struct *mm;
 673
 674         if (!task)
 675                 return -ESRCH;
 676
 677         mm = mm_access(task, mode);
 678         put_task_struct(task);
 679
 680         if (IS_ERR(mm))
 681                 return PTR_ERR(mm);
 682
 683         if (mm) {
 684                 /* ensure this mm_struct can't be freed */
 685                 atomic_inc(&mm->mm_count);
 686                 /* but do not pin its memory */
 687                 mmput(mm);
 688         }
 689
 690         file->private_data = mm;
 691
 692         return 0;
 693 }
 694
 695 static int mem_open(struct inode *inode, struct file *file)
 696 {
 697         int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
 698
 699         /* OK to pass negative loff_t, we can catch out-of-range */
 700         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 701
 702         return ret;
 703 }
 704
 705 static ssize_t mem_rw(struct file *file, char __user *buf,
 706                         size_t count, loff_t *ppos, int write)
 707 {
 708         struct mm_struct *mm = file->private_data;
 709         unsigned long addr = *ppos;
 710         ssize_t copied;
 711         char *page;
 712
 713         if (!mm)
 714                 return 0;
 715
 716         page = (char *)__get_free_page(GFP_TEMPORARY);
 717         if (!page)
 718                 return -ENOMEM;
 719
 720         copied = 0;
 721         if (!atomic_inc_not_zero(&mm->mm_users))
 722                 goto free;
 723
 724         while (count > 0) {
 725                 int this_len = min_t(int, count, PAGE_SIZE);
 726
 727                 if (write && copy_from_user(page, buf, this_len)) {
 728                         copied = -EFAULT;
 729                         break;
 730                 }
 731
 732                 this_len = access_remote_vm(mm, addr, page, this_len, write);
 733                 if (!this_len) {
 734                         if (!copied)
 735                                 copied = -EIO;
 736                         break;
 737                 }
 738
 739                 if (!write && copy_to_user(buf, page, this_len)) {
 740                         copied = -EFAULT;
 741                         break;
 742                 }
 743
 744                 buf += this_len;
 745                 addr += this_len;
 746                 copied += this_len;
 747                 count -= this_len;
 748         }
 749         *ppos = addr;
 750
 751         mmput(mm);
 752 free:
 753         free_page((unsigned long) page);
 754         return copied;
 755 }
 756
 757 static ssize_t mem_read(struct file *file, char __user *buf,
 758                         size_t count, loff_t *ppos)
 759 {
 760         return mem_rw(file, buf, count, ppos, 0);
 761 }
 762
 763 static ssize_t mem_write(struct file *file, const char __user *buf,
 764                          size_t count, loff_t *ppos)
 765 {
 766         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 767 }
 768
 769 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 770 {
 771         switch (orig) {
 772         case 0:
 773                 file->f_pos = offset;
 774                 break;
 775         case 1:
 776                 file->f_pos += offset;
 777                 break;
 778         default:
 779                 return -EINVAL;
 780         }
 781         force_successful_syscall_return();
 782         return file->f_pos;
 783 }
 784
 785 static int mem_release(struct inode *inode, struct file *file)
 786 {
 787         struct mm_struct *mm = file->private_data;
 788         if (mm)
 789                 mmdrop(mm);
 790         return 0;
 791 }
 792
 793 static const struct file_operations proc_mem_operations = {
 794         .llseek         = mem_lseek,
 795         .read           = mem_read,
 796         .write          = mem_write,
 797         .open           = mem_open,
 798         .release        = mem_release,
 799 };
 800
 801 static int environ_open(struct inode *inode, struct file *file)
 802 {
 803         return __mem_open(inode, file, PTRACE_MODE_READ);
 804 }
 805
 806 static ssize_t environ_read(struct file *file, char __user *buf,
 807                         size_t count, loff_t *ppos)
 808 {
 809         char *page;
 810         unsigned long src = *ppos;
 811         int ret = 0;
 812         struct mm_struct *mm = file->private_data;
 813
 814         if (!mm)
 815                 return 0;
 816
 817         page = (char *)__get_free_page(GFP_TEMPORARY);
 818         if (!page)
 819                 return -ENOMEM;
 820
 821         ret = 0;
 822         if (!atomic_inc_not_zero(&mm->mm_users))
 823                 goto free;
 824         while (count > 0) {
 825                 size_t this_len, max_len;
 826                 int retval;
 827
 828                 if (src >= (mm->env_end - mm->env_start))
 829                         break;
 830
 831                 this_len = mm->env_end - (mm->env_start + src);
 832
 833                 max_len = min_t(size_t, PAGE_SIZE, count);
 834                 this_len = min(max_len, this_len);
 835
 836                 retval = access_remote_vm(mm, (mm->env_start + src),
 837                         page, this_len, 0);
 838
 839                 if (retval <= 0) {
 840                         ret = retval;
 841                         break;
 842                 }
 843
 844                 if (copy_to_user(buf, page, retval)) {
 845                         ret = -EFAULT;
 846                         break;
 847                 }
 848
 849                 ret += retval;
 850                 src += retval;
 851                 buf += retval;
 852                 count -= retval;
 853         }
 854         *ppos = src;
 855         mmput(mm);
 856
 857 free:
 858         free_page((unsigned long) page);
 859         return ret;
 860 }
 861
 862 static const struct file_operations proc_environ_operations = {
 863         .open           = environ_open,
 864         .read           = environ_read,
 865         .llseek         = generic_file_llseek,
 866         .release        = mem_release,
 867 };
 868
 869 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
 870                             loff_t *ppos)
 871 {
 872         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 873         char buffer[PROC_NUMBUF];
 874         int oom_adj = OOM_ADJUST_MIN;
 875         size_t len;
 876         unsigned long flags;
 877
 878         if (!task)
 879                 return -ESRCH;
 880         if (lock_task_sighand(task, &flags)) {
 881                 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
 882                         oom_adj = OOM_ADJUST_MAX;
 883                 else
 884                         oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
 885                                   OOM_SCORE_ADJ_MAX;
 886                 unlock_task_sighand(task, &flags);
 887         }
 888         put_task_struct(task);
 889         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
 890         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 891 }
 892
 893 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
 894                              size_t count, loff_t *ppos)
 895 {
 896         struct task_struct *task;
 897         char buffer[PROC_NUMBUF];
 898         int oom_adj;
 899         unsigned long flags;
 900         int err;
 901
 902         memset(buffer, 0, sizeof(buffer));
 903         if (count > sizeof(buffer) - 1)
 904                 count = sizeof(buffer) - 1;
 905         if (copy_from_user(buffer, buf, count)) {
 906                 err = -EFAULT;
 907                 goto out;
 908         }
 909
 910         err = kstrtoint(strstrip(buffer), 0, &oom_adj);
 911         if (err)
 912                 goto out;
 913         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
 914              oom_adj != OOM_DISABLE) {
 915                 err = -EINVAL;
 916                 goto out;
 917         }
 918
 919         task = get_proc_task(file->f_path.dentry->d_inode);
 920         if (!task) {
 921                 err = -ESRCH;
 922                 goto out;
 923         }
 924
 925         task_lock(task);
 926         if (!task->mm) {
 927                 err = -EINVAL;
 928                 goto err_task_lock;
 929         }
 930
 931         if (!lock_task_sighand(task, &flags)) {
 932                 err = -ESRCH;
 933                 goto err_task_lock;
 934         }
 935
 936         /*
 937          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
 938          * value is always attainable.
 939          */
 940         if (oom_adj == OOM_ADJUST_MAX)
 941                 oom_adj = OOM_SCORE_ADJ_MAX;
 942         else
 943                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
 944
 945         if (oom_adj < task->signal->oom_score_adj &&
 946             !capable(CAP_SYS_RESOURCE)) {
 947                 err = -EACCES;
 948                 goto err_sighand;
 949         }
 950
 951         /*
 952          * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
 953          * /proc/pid/oom_score_adj instead.
 954          */
 955         printk_once(KERN_WARNING "%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
 956                   current->comm, task_pid_nr(current), task_pid_nr(task),
 957                   task_pid_nr(task));
 958
 959         task->signal->oom_score_adj = oom_adj;
 960         trace_oom_score_adj_update(task);
 961 err_sighand:
 962         unlock_task_sighand(task, &flags);
 963 err_task_lock:
 964         task_unlock(task);
 965         put_task_struct(task);
 966 out:
 967         return err < 0 ? err : count;
 968 }
 969
 970 static const struct file_operations proc_oom_adj_operations = {
 971         .read           = oom_adj_read,
 972         .write          = oom_adj_write,
 973         .llseek         = generic_file_llseek,
 974 };
 975
 976 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
 977                                         size_t count, loff_t *ppos)
 978 {
 979         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 980         char buffer[PROC_NUMBUF];
 981         short oom_score_adj = OOM_SCORE_ADJ_MIN;
 982         unsigned long flags;
 983         size_t len;
 984
 985         if (!task)
 986                 return -ESRCH;
 987         if (lock_task_sighand(task, &flags)) {
 988                 oom_score_adj = task->signal->oom_score_adj;
 989                 unlock_task_sighand(task, &flags);
 990         }
 991         put_task_struct(task);
 992         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
 993         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 994 }
 995
 996 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
 997                                         size_t count, loff_t *ppos)
 998 {
 999         struct task_struct *task;
1000         char buffer[PROC_NUMBUF];
1001         unsigned long flags;
1002         int oom_score_adj;
1003         int err;
1004
1005         memset(buffer, 0, sizeof(buffer));
1006         if (count > sizeof(buffer) - 1)
1007                 count = sizeof(buffer) - 1;
1008         if (copy_from_user(buffer, buf, count)) {
1009                 err = -EFAULT;
1010                 goto out;
1011         }
1012
1013         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1014         if (err)
1015                 goto out;
1016         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1017                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1018                 err = -EINVAL;
1019                 goto out;
1020         }
1021
1022         task = get_proc_task(file->f_path.dentry->d_inode);
1023         if (!task) {
1024                 err = -ESRCH;
1025                 goto out;
1026         }
1027
1028         task_lock(task);
1029         if (!task->mm) {
1030                 err = -EINVAL;
1031                 goto err_task_lock;
1032         }
1033
1034         if (!lock_task_sighand(task, &flags)) {
1035                 err = -ESRCH;
1036                 goto err_task_lock;
1037         }
1038
1039         if ((short)oom_score_adj < task->signal->oom_score_adj_min &&
1040                         !capable(CAP_SYS_RESOURCE)) {
1041                 err = -EACCES;
1042                 goto err_sighand;
1043         }
1044
1045         task->signal->oom_score_adj = (short)oom_score_adj;
1046         if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1047                 task->signal->oom_score_adj_min = (short)oom_score_adj;
1048         trace_oom_score_adj_update(task);
1049
1050 err_sighand:
1051         unlock_task_sighand(task, &flags);
1052 err_task_lock:
1053         task_unlock(task);
1054         put_task_struct(task);
1055 out:
1056         return err < 0 ? err : count;
1057 }
1058
1059 static const struct file_operations proc_oom_score_adj_operations = {
1060         .read           = oom_score_adj_read,
1061         .write          = oom_score_adj_write,
1062         .llseek         = default_llseek,
1063 };
1064
1065 #ifdef CONFIG_AUDITSYSCALL
1066 #define TMPBUFLEN 21
1067 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1068                                   size_t count, loff_t *ppos)
1069 {
1070         struct inode * inode = file->f_path.dentry->d_inode;
1071         struct task_struct *task = get_proc_task(inode);
1072         ssize_t length;
1073         char tmpbuf[TMPBUFLEN];
1074
1075         if (!task)
1076                 return -ESRCH;
1077         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1078                            from_kuid(file->f_cred->user_ns,
1079                                      audit_get_loginuid(task)));
1080         put_task_struct(task);
1081         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1082 }
1083
1084 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1085                                    size_t count, loff_t *ppos)
1086 {
1087         struct inode * inode = file->f_path.dentry->d_inode;
1088         char *page, *tmp;
1089         ssize_t length;
1090         uid_t loginuid;
1091         kuid_t kloginuid;
1092
1093         rcu_read_lock();
1094         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1095                 rcu_read_unlock();
1096                 return -EPERM;
1097         }
1098         rcu_read_unlock();
1099
1100         if (count >= PAGE_SIZE)
1101                 count = PAGE_SIZE - 1;
1102
1103         if (*ppos != 0) {
1104                 /* No partial writes. */
1105                 return -EINVAL;
1106         }
1107         page = (char*)__get_free_page(GFP_TEMPORARY);
1108         if (!page)
1109                 return -ENOMEM;
1110         length = -EFAULT;
1111         if (copy_from_user(page, buf, count))
1112                 goto out_free_page;
1113
1114         page[count] = '\0';
1115         loginuid = simple_strtoul(page, &tmp, 10);
1116         if (tmp == page) {
1117                 length = -EINVAL;
1118                 goto out_free_page;
1119
1120         }
1121         kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1122         if (!uid_valid(kloginuid)) {
1123                 length = -EINVAL;
1124                 goto out_free_page;
1125         }
1126
1127         length = audit_set_loginuid(kloginuid);
1128         if (likely(length == 0))
1129                 length = count;
1130
1131 out_free_page:
1132         free_page((unsigned long) page);
1133         return length;
1134 }
1135
1136 static const struct file_operations proc_loginuid_operations = {
1137         .read           = proc_loginuid_read,
1138         .write          = proc_loginuid_write,
1139         .llseek         = generic_file_llseek,
1140 };
1141
1142 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1143                                   size_t count, loff_t *ppos)
1144 {
1145         struct inode * inode = file->f_path.dentry->d_inode;
1146         struct task_struct *task = get_proc_task(inode);
1147         ssize_t length;
1148         char tmpbuf[TMPBUFLEN];
1149
1150         if (!task)
1151                 return -ESRCH;
1152         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1153                                 audit_get_sessionid(task));
1154         put_task_struct(task);
1155         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1156 }
1157
1158 static const struct file_operations proc_sessionid_operations = {
1159         .read           = proc_sessionid_read,
1160         .llseek         = generic_file_llseek,
1161 };
1162 #endif
1163
1164 #ifdef CONFIG_FAULT_INJECTION
1165 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1166                                       size_t count, loff_t *ppos)
1167 {
1168         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1169         char buffer[PROC_NUMBUF];
1170         size_t len;
1171         int make_it_fail;
1172
1173         if (!task)
1174                 return -ESRCH;
1175         make_it_fail = task->make_it_fail;
1176         put_task_struct(task);
1177
1178         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1179
1180         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1181 }
1182
1183 static ssize_t proc_fault_inject_write(struct file * file,
1184                         const char __user * buf, size_t count, loff_t *ppos)
1185 {
1186         struct task_struct *task;
1187         char buffer[PROC_NUMBUF], *end;
1188         int make_it_fail;
1189
1190         if (!capable(CAP_SYS_RESOURCE))
1191                 return -EPERM;
1192         memset(buffer, 0, sizeof(buffer));
1193         if (count > sizeof(buffer) - 1)
1194                 count = sizeof(buffer) - 1;
1195         if (copy_from_user(buffer, buf, count))
1196                 return -EFAULT;
1197         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1198         if (*end)
1199                 return -EINVAL;
1200         task = get_proc_task(file->f_dentry->d_inode);
1201         if (!task)
1202                 return -ESRCH;
1203         task->make_it_fail = make_it_fail;
1204         put_task_struct(task);
1205
1206         return count;
1207 }
1208
1209 static const struct file_operations proc_fault_inject_operations = {
1210         .read           = proc_fault_inject_read,
1211         .write          = proc_fault_inject_write,
1212         .llseek         = generic_file_llseek,
1213 };
1214 #endif
1215
1216
1217 #ifdef CONFIG_SCHED_DEBUG
1218 /*
1219  * Print out various scheduling related per-task fields:
1220  */
1221 static int sched_show(struct seq_file *m, void *v)
1222 {
1223         struct inode *inode = m->private;
1224         struct task_struct *p;
1225
1226         p = get_proc_task(inode);
1227         if (!p)
1228                 return -ESRCH;
1229         proc_sched_show_task(p, m);
1230
1231         put_task_struct(p);
1232
1233         return 0;
1234 }
1235
1236 static ssize_t
1237 sched_write(struct file *file, const char __user *buf,
1238             size_t count, loff_t *offset)
1239 {
1240         struct inode *inode = file->f_path.dentry->d_inode;
1241         struct task_struct *p;
1242
1243         p = get_proc_task(inode);
1244         if (!p)
1245                 return -ESRCH;
1246         proc_sched_set_task(p);
1247
1248         put_task_struct(p);
1249
1250         return count;
1251 }
1252
1253 static int sched_open(struct inode *inode, struct file *filp)
1254 {
1255         return single_open(filp, sched_show, inode);
1256 }
1257
1258 static const struct file_operations proc_pid_sched_operations = {
1259         .open           = sched_open,
1260         .read           = seq_read,
1261         .write          = sched_write,
1262         .llseek         = seq_lseek,
1263         .release        = single_release,
1264 };
1265
1266 #endif
1267
1268 #ifdef CONFIG_SCHED_AUTOGROUP
1269 /*
1270  * Print out autogroup related information:
1271  */
1272 static int sched_autogroup_show(struct seq_file *m, void *v)
1273 {
1274         struct inode *inode = m->private;
1275         struct task_struct *p;
1276
1277         p = get_proc_task(inode);
1278         if (!p)
1279                 return -ESRCH;
1280         proc_sched_autogroup_show_task(p, m);
1281
1282         put_task_struct(p);
1283
1284         return 0;
1285 }
1286
1287 static ssize_t
1288 sched_autogroup_write(struct file *file, const char __user *buf,
1289             size_t count, loff_t *offset)
1290 {
1291         struct inode *inode = file->f_path.dentry->d_inode;
1292         struct task_struct *p;
1293         char buffer[PROC_NUMBUF];
1294         int nice;
1295         int err;
1296
1297         memset(buffer, 0, sizeof(buffer));
1298         if (count > sizeof(buffer) - 1)
1299                 count = sizeof(buffer) - 1;
1300         if (copy_from_user(buffer, buf, count))
1301                 return -EFAULT;
1302
1303         err = kstrtoint(strstrip(buffer), 0, &nice);
1304         if (err < 0)
1305                 return err;
1306
1307         p = get_proc_task(inode);
1308         if (!p)
1309                 return -ESRCH;
1310
1311         err = proc_sched_autogroup_set_nice(p, nice);
1312         if (err)
1313                 count = err;
1314
1315         put_task_struct(p);
1316
1317         return count;
1318 }
1319
1320 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1321 {
1322         int ret;
1323
1324         ret = single_open(filp, sched_autogroup_show, NULL);
1325         if (!ret) {
1326                 struct seq_file *m = filp->private_data;
1327
1328                 m->private = inode;
1329         }
1330         return ret;
1331 }
1332
1333 static const struct file_operations proc_pid_sched_autogroup_operations = {
1334         .open           = sched_autogroup_open,
1335         .read           = seq_read,
1336         .write          = sched_autogroup_write,
1337         .llseek         = seq_lseek,
1338         .release        = single_release,
1339 };
1340
1341 #endif /* CONFIG_SCHED_AUTOGROUP */
1342
1343 static ssize_t comm_write(struct file *file, const char __user *buf,
1344                                 size_t count, loff_t *offset)
1345 {
1346         struct inode *inode = file->f_path.dentry->d_inode;
1347         struct task_struct *p;
1348         char buffer[TASK_COMM_LEN];
1349
1350         memset(buffer, 0, sizeof(buffer));
1351         if (count > sizeof(buffer) - 1)
1352                 count = sizeof(buffer) - 1;
1353         if (copy_from_user(buffer, buf, count))
1354                 return -EFAULT;
1355
1356         p = get_proc_task(inode);
1357         if (!p)
1358                 return -ESRCH;
1359
1360         if (same_thread_group(current, p))
1361                 set_task_comm(p, buffer);
1362         else
1363                 count = -EINVAL;
1364
1365         put_task_struct(p);
1366
1367         return count;
1368 }
1369
1370 static int comm_show(struct seq_file *m, void *v)
1371 {
1372         struct inode *inode = m->private;
1373         struct task_struct *p;
1374
1375         p = get_proc_task(inode);
1376         if (!p)
1377                 return -ESRCH;
1378
1379         task_lock(p);
1380         seq_printf(m, "%s\n", p->comm);
1381         task_unlock(p);
1382
1383         put_task_struct(p);
1384
1385         return 0;
1386 }
1387
1388 static int comm_open(struct inode *inode, struct file *filp)
1389 {
1390         return single_open(filp, comm_show, inode);
1391 }
1392
1393 static const struct file_operations proc_pid_set_comm_operations = {
1394         .open           = comm_open,
1395         .read           = seq_read,
1396         .write          = comm_write,
1397         .llseek         = seq_lseek,
1398         .release        = single_release,
1399 };
1400
1401 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1402 {
1403         struct task_struct *task;
1404         struct mm_struct *mm;
1405         struct file *exe_file;
1406
1407         task = get_proc_task(dentry->d_inode);
1408         if (!task)
1409                 return -ENOENT;
1410         mm = get_task_mm(task);
1411         put_task_struct(task);
1412         if (!mm)
1413                 return -ENOENT;
1414         exe_file = get_mm_exe_file(mm);
1415         mmput(mm);
1416         if (exe_file) {
1417                 *exe_path = exe_file->f_path;
1418                 path_get(&exe_file->f_path);
1419                 fput(exe_file);
1420                 return 0;
1421         } else
1422                 return -ENOENT;
1423 }
1424
1425 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1426 {
1427         struct inode *inode = dentry->d_inode;
1428         struct path path;
1429         int error = -EACCES;
1430
1431         /* Are we allowed to snoop on the tasks file descriptors? */
1432         if (!proc_fd_access_allowed(inode))
1433                 goto out;
1434
1435         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1436         if (error)
1437                 goto out;
1438
1439         nd_jump_link(nd, &path);
1440         return NULL;
1441 out:
1442         return ERR_PTR(error);
1443 }
1444
1445 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1446 {
1447         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1448         char *pathname;
1449         int len;
1450
1451         if (!tmp)
1452                 return -ENOMEM;
1453
1454         pathname = d_path(path, tmp, PAGE_SIZE);
1455         len = PTR_ERR(pathname);
1456         if (IS_ERR(pathname))
1457                 goto out;
1458         len = tmp + PAGE_SIZE - 1 - pathname;
1459
1460         if (len > buflen)
1461                 len = buflen;
1462         if (copy_to_user(buffer, pathname, len))
1463                 len = -EFAULT;
1464  out:
1465         free_page((unsigned long)tmp);
1466         return len;
1467 }
1468
1469 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1470 {
1471         int error = -EACCES;
1472         struct inode *inode = dentry->d_inode;
1473         struct path path;
1474
1475         /* Are we allowed to snoop on the tasks file descriptors? */
1476         if (!proc_fd_access_allowed(inode))
1477                 goto out;
1478
1479         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1480         if (error)
1481                 goto out;
1482
1483         error = do_proc_readlink(&path, buffer, buflen);
1484         path_put(&path);
1485 out:
1486         return error;
1487 }
1488
1489 const struct inode_operations proc_pid_link_inode_operations = {
1490         .readlink       = proc_pid_readlink,
1491         .follow_link    = proc_pid_follow_link,
1492         .setattr        = proc_setattr,
1493 };
1494
1495
1496 /* building an inode */
1497
1498 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1499 {
1500         struct inode * inode;
1501         struct proc_inode *ei;
1502         const struct cred *cred;
1503
1504         /* We need a new inode */
1505
1506         inode = new_inode(sb);
1507         if (!inode)
1508                 goto out;
1509
1510         /* Common stuff */
1511         ei = PROC_I(inode);
1512         inode->i_ino = get_next_ino();
1513         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1514         inode->i_op = &proc_def_inode_operations;
1515
1516         /*
1517          * grab the reference to task.
1518          */
1519         ei->pid = get_task_pid(task, PIDTYPE_PID);
1520         if (!ei->pid)
1521                 goto out_unlock;
1522
1523         if (task_dumpable(task)) {
1524                 rcu_read_lock();
1525                 cred = __task_cred(task);
1526                 inode->i_uid = cred->euid;
1527                 inode->i_gid = cred->egid;
1528                 rcu_read_unlock();
1529         }
1530         security_task_to_inode(task, inode);
1531
1532 out:
1533         return inode;
1534
1535 out_unlock:
1536         iput(inode);
1537         return NULL;
1538 }
1539
1540 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1541 {
1542         struct inode *inode = dentry->d_inode;
1543         struct task_struct *task;
1544         const struct cred *cred;
1545         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1546
1547         generic_fillattr(inode, stat);
1548
1549         rcu_read_lock();
1550         stat->uid = GLOBAL_ROOT_UID;
1551         stat->gid = GLOBAL_ROOT_GID;
1552         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1553         if (task) {
1554                 if (!has_pid_permissions(pid, task, 2)) {
1555                         rcu_read_unlock();
1556                         /*
1557                          * This doesn't prevent learning whether PID exists,
1558                          * it only makes getattr() consistent with readdir().
1559                          */
1560                         return -ENOENT;
1561                 }
1562                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1563                     task_dumpable(task)) {
1564                         cred = __task_cred(task);
1565                         stat->uid = cred->euid;
1566                         stat->gid = cred->egid;
1567                 }
1568         }
1569         rcu_read_unlock();
1570         return 0;
1571 }
1572
1573 /* dentry stuff */
1574
1575 /*
1576  *      Exceptional case: normally we are not allowed to unhash a busy
1577  * directory. In this case, however, we can do it - no aliasing problems
1578  * due to the way we treat inodes.
1579  *
1580  * Rewrite the inode's ownerships here because the owning task may have
1581  * performed a setuid(), etc.
1582  *
1583  * Before the /proc/pid/status file was created the only way to read
1584  * the effective uid of a /process was to stat /proc/pid.  Reading
1585  * /proc/pid/status is slow enough that procps and other packages
1586  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1587  * made this apply to all per process world readable and executable
1588  * directories.
1589  */
1590 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1591 {
1592         struct inode *inode;
1593         struct task_struct *task;
1594         const struct cred *cred;
1595
1596         if (flags & LOOKUP_RCU)
1597                 return -ECHILD;
1598
1599         inode = dentry->d_inode;
1600         task = get_proc_task(inode);
1601
1602         if (task) {
1603                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1604                     task_dumpable(task)) {
1605                         rcu_read_lock();
1606                         cred = __task_cred(task);
1607                         inode->i_uid = cred->euid;
1608                         inode->i_gid = cred->egid;
1609                         rcu_read_unlock();
1610                 } else {
1611                         inode->i_uid = GLOBAL_ROOT_UID;
1612                         inode->i_gid = GLOBAL_ROOT_GID;
1613                 }
1614                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1615                 security_task_to_inode(task, inode);
1616                 put_task_struct(task);
1617                 return 1;
1618         }
1619         d_drop(dentry);
1620         return 0;
1621 }
1622
1623 const struct dentry_operations pid_dentry_operations =
1624 {
1625         .d_revalidate   = pid_revalidate,
1626         .d_delete       = pid_delete_dentry,
1627 };
1628
1629 /* Lookups */
1630
1631 /*
1632  * Fill a directory entry.
1633  *
1634  * If possible create the dcache entry and derive our inode number and
1635  * file type from dcache entry.
1636  *
1637  * Since all of the proc inode numbers are dynamically generated, the inode
1638  * numbers do not exist until the inode is cache.  This means creating the
1639  * the dcache entry in readdir is necessary to keep the inode numbers
1640  * reported by readdir in sync with the inode numbers reported
1641  * by stat.
1642  */
1643 int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1644         const char *name, int len,
1645         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1646 {
1647         struct dentry *child, *dir = filp->f_path.dentry;
1648         struct inode *inode;
1649         struct qstr qname;
1650         ino_t ino = 0;
1651         unsigned type = DT_UNKNOWN;
1652
1653         qname.name = name;
1654         qname.len  = len;
1655         qname.hash = full_name_hash(name, len);
1656
1657         child = d_lookup(dir, &qname);
1658         if (!child) {
1659                 struct dentry *new;
1660                 new = d_alloc(dir, &qname);
1661                 if (new) {
1662                         child = instantiate(dir->d_inode, new, task, ptr);
1663                         if (child)
1664                                 dput(new);
1665                         else
1666                                 child = new;
1667                 }
1668         }
1669         if (!child || IS_ERR(child) || !child->d_inode)
1670                 goto end_instantiate;
1671         inode = child->d_inode;
1672         if (inode) {
1673                 ino = inode->i_ino;
1674                 type = inode->i_mode >> 12;
1675         }
1676         dput(child);
1677 end_instantiate:
1678         if (!ino)
1679                 ino = find_inode_number(dir, &qname);
1680         if (!ino)
1681                 ino = 1;
1682         return filldir(dirent, name, len, filp->f_pos, ino, type);
1683 }
1684
1685 #ifdef CONFIG_CHECKPOINT_RESTORE
1686
1687 /*
1688  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1689  * which represent vma start and end addresses.
1690  */
1691 static int dname_to_vma_addr(struct dentry *dentry,
1692                              unsigned long *start, unsigned long *end)
1693 {
1694         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1695                 return -EINVAL;
1696
1697         return 0;
1698 }
1699
1700 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1701 {
1702         unsigned long vm_start, vm_end;
1703         bool exact_vma_exists = false;
1704         struct mm_struct *mm = NULL;
1705         struct task_struct *task;
1706         const struct cred *cred;
1707         struct inode *inode;
1708         int status = 0;
1709
1710         if (flags & LOOKUP_RCU)
1711                 return -ECHILD;
1712
1713         if (!capable(CAP_SYS_ADMIN)) {
1714                 status = -EACCES;
1715                 goto out_notask;
1716         }
1717
1718         inode = dentry->d_inode;
1719         task = get_proc_task(inode);
1720         if (!task)
1721                 goto out_notask;
1722
1723         mm = mm_access(task, PTRACE_MODE_READ);
1724         if (IS_ERR_OR_NULL(mm))
1725                 goto out;
1726
1727         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1728                 down_read(&mm->mmap_sem);
1729                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1730                 up_read(&mm->mmap_sem);
1731         }
1732
1733         mmput(mm);
1734
1735         if (exact_vma_exists) {
1736                 if (task_dumpable(task)) {
1737                         rcu_read_lock();
1738                         cred = __task_cred(task);
1739                         inode->i_uid = cred->euid;
1740                         inode->i_gid = cred->egid;
1741                         rcu_read_unlock();
1742                 } else {
1743                         inode->i_uid = GLOBAL_ROOT_UID;
1744                         inode->i_gid = GLOBAL_ROOT_GID;
1745                 }
1746                 security_task_to_inode(task, inode);
1747                 status = 1;
1748         }
1749
1750 out:
1751         put_task_struct(task);
1752
1753 out_notask:
1754         if (status <= 0)
1755                 d_drop(dentry);
1756
1757         return status;
1758 }
1759
1760 static const struct dentry_operations tid_map_files_dentry_operations = {
1761         .d_revalidate   = map_files_d_revalidate,
1762         .d_delete       = pid_delete_dentry,
1763 };
1764
1765 static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
1766 {
1767         unsigned long vm_start, vm_end;
1768         struct vm_area_struct *vma;
1769         struct task_struct *task;
1770         struct mm_struct *mm;
1771         int rc;
1772
1773         rc = -ENOENT;
1774         task = get_proc_task(dentry->d_inode);
1775         if (!task)
1776                 goto out;
1777
1778         mm = get_task_mm(task);
1779         put_task_struct(task);
1780         if (!mm)
1781                 goto out;
1782
1783         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1784         if (rc)
1785                 goto out_mmput;
1786
1787         down_read(&mm->mmap_sem);
1788         vma = find_exact_vma(mm, vm_start, vm_end);
1789         if (vma && vma->vm_file) {
1790                 *path = vma->vm_file->f_path;
1791                 path_get(path);
1792                 rc = 0;
1793         }
1794         up_read(&mm->mmap_sem);
1795
1796 out_mmput:
1797         mmput(mm);
1798 out:
1799         return rc;
1800 }
1801
1802 struct map_files_info {
1803         fmode_t         mode;
1804         unsigned long   len;
1805         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1806 };
1807
1808 static struct dentry *
1809 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
1810                            struct task_struct *task, const void *ptr)
1811 {
1812         fmode_t mode = (fmode_t)(unsigned long)ptr;
1813         struct proc_inode *ei;
1814         struct inode *inode;
1815
1816         inode = proc_pid_make_inode(dir->i_sb, task);
1817         if (!inode)
1818                 return ERR_PTR(-ENOENT);
1819
1820         ei = PROC_I(inode);
1821         ei->op.proc_get_link = proc_map_files_get_link;
1822
1823         inode->i_op = &proc_pid_link_inode_operations;
1824         inode->i_size = 64;
1825         inode->i_mode = S_IFLNK;
1826
1827         if (mode & FMODE_READ)
1828                 inode->i_mode |= S_IRUSR;
1829         if (mode & FMODE_WRITE)
1830                 inode->i_mode |= S_IWUSR;
1831
1832         d_set_d_op(dentry, &tid_map_files_dentry_operations);
1833         d_add(dentry, inode);
1834
1835         return NULL;
1836 }
1837
1838 static struct dentry *proc_map_files_lookup(struct inode *dir,
1839                 struct dentry *dentry, unsigned int flags)
1840 {
1841         unsigned long vm_start, vm_end;
1842         struct vm_area_struct *vma;
1843         struct task_struct *task;
1844         struct dentry *result;
1845         struct mm_struct *mm;
1846
1847         result = ERR_PTR(-EACCES);
1848         if (!capable(CAP_SYS_ADMIN))
1849                 goto out;
1850
1851         result = ERR_PTR(-ENOENT);
1852         task = get_proc_task(dir);
1853         if (!task)
1854                 goto out;
1855
1856         result = ERR_PTR(-EACCES);
1857         if (!ptrace_may_access(task, PTRACE_MODE_READ))
1858                 goto out_put_task;
1859
1860         result = ERR_PTR(-ENOENT);
1861         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
1862                 goto out_put_task;
1863
1864         mm = get_task_mm(task);
1865         if (!mm)
1866                 goto out_put_task;
1867
1868         down_read(&mm->mmap_sem);
1869         vma = find_exact_vma(mm, vm_start, vm_end);
1870         if (!vma)
1871                 goto out_no_vma;
1872
1873         if (vma->vm_file)
1874                 result = proc_map_files_instantiate(dir, dentry, task,
1875                                 (void *)(unsigned long)vma->vm_file->f_mode);
1876
1877 out_no_vma:
1878         up_read(&mm->mmap_sem);
1879         mmput(mm);
1880 out_put_task:
1881         put_task_struct(task);
1882 out:
1883         return result;
1884 }
1885
1886 static const struct inode_operations proc_map_files_inode_operations = {
1887         .lookup         = proc_map_files_lookup,
1888         .permission     = proc_fd_permission,
1889         .setattr        = proc_setattr,
1890 };
1891
1892 static int
1893 proc_map_files_readdir(struct file *filp, void *dirent, filldir_t filldir)
1894 {
1895         struct dentry *dentry = filp->f_path.dentry;
1896         struct inode *inode = dentry->d_inode;
1897         struct vm_area_struct *vma;
1898         struct task_struct *task;
1899         struct mm_struct *mm;
1900         ino_t ino;
1901         int ret;
1902
1903         ret = -EACCES;
1904         if (!capable(CAP_SYS_ADMIN))
1905                 goto out;
1906
1907         ret = -ENOENT;
1908         task = get_proc_task(inode);
1909         if (!task)
1910                 goto out;
1911
1912         ret = -EACCES;
1913         if (!ptrace_may_access(task, PTRACE_MODE_READ))
1914                 goto out_put_task;
1915
1916         ret = 0;
1917         switch (filp->f_pos) {
1918         case 0:
1919                 ino = inode->i_ino;
1920                 if (filldir(dirent, ".", 1, 0, ino, DT_DIR) < 0)
1921                         goto out_put_task;
1922                 filp->f_pos++;
1923         case 1:
1924                 ino = parent_ino(dentry);
1925                 if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1926                         goto out_put_task;
1927                 filp->f_pos++;
1928         default:
1929         {
1930                 unsigned long nr_files, pos, i;
1931                 struct flex_array *fa = NULL;
1932                 struct map_files_info info;
1933                 struct map_files_info *p;
1934
1935                 mm = get_task_mm(task);
1936                 if (!mm)
1937                         goto out_put_task;
1938                 down_read(&mm->mmap_sem);
1939
1940                 nr_files = 0;
1941
1942                 /*
1943                  * We need two passes here:
1944                  *
1945                  *  1) Collect vmas of mapped files with mmap_sem taken
1946                  *  2) Release mmap_sem and instantiate entries
1947                  *
1948                  * otherwise we get lockdep complained, since filldir()
1949                  * routine might require mmap_sem taken in might_fault().
1950                  */
1951
1952                 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
1953                         if (vma->vm_file && ++pos > filp->f_pos)
1954                                 nr_files++;
1955                 }
1956
1957                 if (nr_files) {
1958                         fa = flex_array_alloc(sizeof(info), nr_files,
1959                                                 GFP_KERNEL);
1960                         if (!fa || flex_array_prealloc(fa, 0, nr_files,
1961                                                         GFP_KERNEL)) {
1962                                 ret = -ENOMEM;
1963                                 if (fa)
1964                                         flex_array_free(fa);
1965                                 up_read(&mm->mmap_sem);
1966                                 mmput(mm);
1967                                 goto out_put_task;
1968                         }
1969                         for (i = 0, vma = mm->mmap, pos = 2; vma;
1970                                         vma = vma->vm_next) {
1971                                 if (!vma->vm_file)
1972                                         continue;
1973                                 if (++pos <= filp->f_pos)
1974                                         continue;
1975
1976                                 info.mode = vma->vm_file->f_mode;
1977                                 info.len = snprintf(info.name,
1978                                                 sizeof(info.name), "%lx-%lx",
1979                                                 vma->vm_start, vma->vm_end);
1980                                 if (flex_array_put(fa, i++, &info, GFP_KERNEL))
1981                                         BUG();
1982                         }
1983                 }
1984                 up_read(&mm->mmap_sem);
1985
1986                 for (i = 0; i < nr_files; i++) {
1987                         p = flex_array_get(fa, i);
1988                         ret = proc_fill_cache(filp, dirent, filldir,
1989                                               p->name, p->len,
1990                                               proc_map_files_instantiate,
1991                                               task,
1992                                               (void *)(unsigned long)p->mode);
1993                         if (ret)
1994                                 break;
1995                         filp->f_pos++;
1996                 }
1997                 if (fa)
1998                         flex_array_free(fa);
1999                 mmput(mm);
2000         }
2001         }
2002
2003 out_put_task:
2004         put_task_struct(task);
2005 out:
2006         return ret;
2007 }
2008
2009 static const struct file_operations proc_map_files_operations = {
2010         .read           = generic_read_dir,
2011         .readdir        = proc_map_files_readdir,
2012         .llseek         = default_llseek,
2013 };
2014
2015 #endif /* CONFIG_CHECKPOINT_RESTORE */
2016
2017 static struct dentry *proc_pident_instantiate(struct inode *dir,
2018         struct dentry *dentry, struct task_struct *task, const void *ptr)
2019 {
2020         const struct pid_entry *p = ptr;
2021         struct inode *inode;
2022         struct proc_inode *ei;
2023         struct dentry *error = ERR_PTR(-ENOENT);
2024
2025         inode = proc_pid_make_inode(dir->i_sb, task);
2026         if (!inode)
2027                 goto out;
2028
2029         ei = PROC_I(inode);
2030         inode->i_mode = p->mode;
2031         if (S_ISDIR(inode->i_mode))
2032                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2033         if (p->iop)
2034                 inode->i_op = p->iop;
2035         if (p->fop)
2036                 inode->i_fop = p->fop;
2037         ei->op = p->op;
2038         d_set_d_op(dentry, &pid_dentry_operations);
2039         d_add(dentry, inode);
2040         /* Close the race of the process dying before we return the dentry */
2041         if (pid_revalidate(dentry, 0))
2042                 error = NULL;
2043 out:
2044         return error;
2045 }
2046
2047 static struct dentry *proc_pident_lookup(struct inode *dir,
2048                                          struct dentry *dentry,
2049                                          const struct pid_entry *ents,
2050                                          unsigned int nents)
2051 {
2052         struct dentry *error;
2053         struct task_struct *task = get_proc_task(dir);
2054         const struct pid_entry *p, *last;
2055
2056         error = ERR_PTR(-ENOENT);
2057
2058         if (!task)
2059                 goto out_no_task;
2060
2061         /*
2062          * Yes, it does not scale. And it should not. Don't add
2063          * new entries into /proc/<tgid>/ without very good reasons.
2064          */
2065         last = &ents[nents - 1];
2066         for (p = ents; p <= last; p++) {
2067                 if (p->len != dentry->d_name.len)
2068                         continue;
2069                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2070                         break;
2071         }
2072         if (p > last)
2073                 goto out;
2074
2075         error = proc_pident_instantiate(dir, dentry, task, p);
2076 out:
2077         put_task_struct(task);
2078 out_no_task:
2079         return error;
2080 }
2081
2082 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2083         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2084 {
2085         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2086                                 proc_pident_instantiate, task, p);
2087 }
2088
2089 static int proc_pident_readdir(struct file *filp,
2090                 void *dirent, filldir_t filldir,
2091                 const struct pid_entry *ents, unsigned int nents)
2092 {
2093         int i;
2094         struct dentry *dentry = filp->f_path.dentry;
2095         struct inode *inode = dentry->d_inode;
2096         struct task_struct *task = get_proc_task(inode);
2097         const struct pid_entry *p, *last;
2098         ino_t ino;
2099         int ret;
2100
2101         ret = -ENOENT;
2102         if (!task)
2103                 goto out_no_task;
2104
2105         ret = 0;
2106         i = filp->f_pos;
2107         switch (i) {
2108         case 0:
2109                 ino = inode->i_ino;
2110                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2111                         goto out;
2112                 i++;
2113                 filp->f_pos++;
2114                 /* fall through */
2115         case 1:
2116                 ino = parent_ino(dentry);
2117                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2118                         goto out;
2119                 i++;
2120                 filp->f_pos++;
2121                 /* fall through */
2122         default:
2123                 i -= 2;
2124                 if (i >= nents) {
2125                         ret = 1;
2126                         goto out;
2127                 }
2128                 p = ents + i;
2129                 last = &ents[nents - 1];
2130                 while (p <= last) {
2131                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2132                                 goto out;
2133                         filp->f_pos++;
2134                         p++;
2135                 }
2136         }
2137
2138         ret = 1;
2139 out:
2140         put_task_struct(task);
2141 out_no_task:
2142         return ret;
2143 }
2144
2145 #ifdef CONFIG_SECURITY
2146 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2147                                   size_t count, loff_t *ppos)
2148 {
2149         struct inode * inode = file->f_path.dentry->d_inode;
2150         char *p = NULL;
2151         ssize_t length;
2152         struct task_struct *task = get_proc_task(inode);
2153
2154         if (!task)
2155                 return -ESRCH;
2156
2157         length = security_getprocattr(task,
2158                                       (char*)file->f_path.dentry->d_name.name,
2159                                       &p);
2160         put_task_struct(task);
2161         if (length > 0)
2162                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2163         kfree(p);
2164         return length;
2165 }
2166
2167 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2168                                    size_t count, loff_t *ppos)
2169 {
2170         struct inode * inode = file->f_path.dentry->d_inode;
2171         char *page;
2172         ssize_t length;
2173         struct task_struct *task = get_proc_task(inode);
2174
2175         length = -ESRCH;
2176         if (!task)
2177                 goto out_no_task;
2178         if (count > PAGE_SIZE)
2179                 count = PAGE_SIZE;
2180
2181         /* No partial writes. */
2182         length = -EINVAL;
2183         if (*ppos != 0)
2184                 goto out;
2185
2186         length = -ENOMEM;
2187         page = (char*)__get_free_page(GFP_TEMPORARY);
2188         if (!page)
2189                 goto out;
2190
2191         length = -EFAULT;
2192         if (copy_from_user(page, buf, count))
2193                 goto out_free;
2194
2195         /* Guard against adverse ptrace interaction */
2196         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2197         if (length < 0)
2198                 goto out_free;
2199
2200         length = security_setprocattr(task,
2201                                       (char*)file->f_path.dentry->d_name.name,
2202                                       (void*)page, count);
2203         mutex_unlock(&task->signal->cred_guard_mutex);
2204 out_free:
2205         free_page((unsigned long) page);
2206 out:
2207         put_task_struct(task);
2208 out_no_task:
2209         return length;
2210 }
2211
2212 static const struct file_operations proc_pid_attr_operations = {
2213         .read           = proc_pid_attr_read,
2214         .write          = proc_pid_attr_write,
2215         .llseek         = generic_file_llseek,
2216 };
2217
2218 static const struct pid_entry attr_dir_stuff[] = {
2219         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2220         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2221         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2222         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2223         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2224         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2225 };
2226
2227 static int proc_attr_dir_readdir(struct file * filp,
2228                              void * dirent, filldir_t filldir)
2229 {
2230         return proc_pident_readdir(filp,dirent,filldir,
2231                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2232 }
2233
2234 static const struct file_operations proc_attr_dir_operations = {
2235         .read           = generic_read_dir,
2236         .readdir        = proc_attr_dir_readdir,
2237         .llseek         = default_llseek,
2238 };
2239
2240 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2241                                 struct dentry *dentry, unsigned int flags)
2242 {
2243         return proc_pident_lookup(dir, dentry,
2244                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2245 }
2246
2247 static const struct inode_operations proc_attr_dir_inode_operations = {
2248         .lookup         = proc_attr_dir_lookup,
2249         .getattr        = pid_getattr,
2250         .setattr        = proc_setattr,
2251 };
2252
2253 #endif
2254
2255 #ifdef CONFIG_ELF_CORE
2256 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2257                                          size_t count, loff_t *ppos)
2258 {
2259         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2260         struct mm_struct *mm;
2261         char buffer[PROC_NUMBUF];
2262         size_t len;
2263         int ret;
2264
2265         if (!task)
2266                 return -ESRCH;
2267
2268         ret = 0;
2269         mm = get_task_mm(task);
2270         if (mm) {
2271                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2272                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2273                                 MMF_DUMP_FILTER_SHIFT));
2274                 mmput(mm);
2275                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2276         }
2277
2278         put_task_struct(task);
2279
2280         return ret;
2281 }
2282
2283 static ssize_t proc_coredump_filter_write(struct file *file,
2284                                           const char __user *buf,
2285                                           size_t count,
2286                                           loff_t *ppos)
2287 {
2288         struct task_struct *task;
2289         struct mm_struct *mm;
2290         char buffer[PROC_NUMBUF], *end;
2291         unsigned int val;
2292         int ret;
2293         int i;
2294         unsigned long mask;
2295
2296         ret = -EFAULT;
2297         memset(buffer, 0, sizeof(buffer));
2298         if (count > sizeof(buffer) - 1)
2299                 count = sizeof(buffer) - 1;
2300         if (copy_from_user(buffer, buf, count))
2301                 goto out_no_task;
2302
2303         ret = -EINVAL;
2304         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2305         if (*end == '\n')
2306                 end++;
2307         if (end - buffer == 0)
2308                 goto out_no_task;
2309
2310         ret = -ESRCH;
2311         task = get_proc_task(file->f_dentry->d_inode);
2312         if (!task)
2313                 goto out_no_task;
2314
2315         ret = end - buffer;
2316         mm = get_task_mm(task);
2317         if (!mm)
2318                 goto out_no_mm;
2319
2320         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2321                 if (val & mask)
2322                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2323                 else
2324                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2325         }
2326
2327         mmput(mm);
2328  out_no_mm:
2329         put_task_struct(task);
2330  out_no_task:
2331         return ret;
2332 }
2333
2334 static const struct file_operations proc_coredump_filter_operations = {
2335         .read           = proc_coredump_filter_read,
2336         .write          = proc_coredump_filter_write,
2337         .llseek         = generic_file_llseek,
2338 };
2339 #endif
2340
2341 #ifdef CONFIG_TASK_IO_ACCOUNTING
2342 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2343 {
2344         struct task_io_accounting acct = task->ioac;
2345         unsigned long flags;
2346         int result;
2347
2348         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2349         if (result)
2350                 return result;
2351
2352         if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2353                 result = -EACCES;
2354                 goto out_unlock;
2355         }
2356
2357         if (whole && lock_task_sighand(task, &flags)) {
2358                 struct task_struct *t = task;
2359
2360                 task_io_accounting_add(&acct, &task->signal->ioac);
2361                 while_each_thread(task, t)
2362                         task_io_accounting_add(&acct, &t->ioac);
2363
2364                 unlock_task_sighand(task, &flags);
2365         }
2366         result = sprintf(buffer,
2367                         "rchar: %llu\n"
2368                         "wchar: %llu\n"
2369                         "syscr: %llu\n"
2370                         "syscw: %llu\n"
2371                         "read_bytes: %llu\n"
2372                         "write_bytes: %llu\n"
2373                         "cancelled_write_bytes: %llu\n",
2374                         (unsigned long long)acct.rchar,
2375                         (unsigned long long)acct.wchar,
2376                         (unsigned long long)acct.syscr,
2377                         (unsigned long long)acct.syscw,
2378                         (unsigned long long)acct.read_bytes,
2379                         (unsigned long long)acct.write_bytes,
2380                         (unsigned long long)acct.cancelled_write_bytes);
2381 out_unlock:
2382         mutex_unlock(&task->signal->cred_guard_mutex);
2383         return result;
2384 }
2385
2386 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2387 {
2388         return do_io_accounting(task, buffer, 0);
2389 }
2390
2391 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2392 {
2393         return do_io_accounting(task, buffer, 1);
2394 }
2395 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2396
2397 #ifdef CONFIG_USER_NS
2398 static int proc_id_map_open(struct inode *inode, struct file *file,
2399         struct seq_operations *seq_ops)
2400 {
2401         struct user_namespace *ns = NULL;
2402         struct task_struct *task;
2403         struct seq_file *seq;
2404         int ret = -EINVAL;
2405
2406         task = get_proc_task(inode);
2407         if (task) {
2408                 rcu_read_lock();
2409                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2410                 rcu_read_unlock();
2411                 put_task_struct(task);
2412         }
2413         if (!ns)
2414                 goto err;
2415
2416         ret = seq_open(file, seq_ops);
2417         if (ret)
2418                 goto err_put_ns;
2419
2420         seq = file->private_data;
2421         seq->private = ns;
2422
2423         return 0;
2424 err_put_ns:
2425         put_user_ns(ns);
2426 err:
2427         return ret;
2428 }
2429
2430 static int proc_id_map_release(struct inode *inode, struct file *file)
2431 {
2432         struct seq_file *seq = file->private_data;
2433         struct user_namespace *ns = seq->private;
2434         put_user_ns(ns);
2435         return seq_release(inode, file);
2436 }
2437
2438 static int proc_uid_map_open(struct inode *inode, struct file *file)
2439 {
2440         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2441 }
2442
2443 static int proc_gid_map_open(struct inode *inode, struct file *file)
2444 {
2445         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2446 }
2447
2448 static int proc_projid_map_open(struct inode *inode, struct file *file)
2449 {
2450         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2451 }
2452
2453 static const struct file_operations proc_uid_map_operations = {
2454         .open           = proc_uid_map_open,
2455         .write          = proc_uid_map_write,
2456         .read           = seq_read,
2457         .llseek         = seq_lseek,
2458         .release        = proc_id_map_release,
2459 };
2460
2461 static const struct file_operations proc_gid_map_operations = {
2462         .open           = proc_gid_map_open,
2463         .write          = proc_gid_map_write,
2464         .read           = seq_read,
2465         .llseek         = seq_lseek,
2466         .release        = proc_id_map_release,
2467 };
2468
2469 static const struct file_operations proc_projid_map_operations = {
2470         .open           = proc_projid_map_open,
2471         .write          = proc_projid_map_write,
2472         .read           = seq_read,
2473         .llseek         = seq_lseek,
2474         .release        = proc_id_map_release,
2475 };
2476 #endif /* CONFIG_USER_NS */
2477
2478 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2479                                 struct pid *pid, struct task_struct *task)
2480 {
2481         int err = lock_trace(task);
2482         if (!err) {
2483                 seq_printf(m, "%08x\n", task->personality);
2484                 unlock_trace(task);
2485         }
2486         return err;
2487 }
2488
2489 /*
2490  * Thread groups
2491  */
2492 static const struct file_operations proc_task_operations;
2493 static const struct inode_operations proc_task_inode_operations;
2494
2495 static const struct pid_entry tgid_base_stuff[] = {
2496         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2497         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2498 #ifdef CONFIG_CHECKPOINT_RESTORE
2499         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2500 #endif
2501         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2502         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2503 #ifdef CONFIG_NET
2504         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2505 #endif
2506         REG("environ",    S_IRUSR, proc_environ_operations),
2507         INF("auxv",       S_IRUSR, proc_pid_auxv),
2508         ONE("status",     S_IRUGO, proc_pid_status),
2509         ONE("personality", S_IRUGO, proc_pid_personality),
2510         INF("limits",     S_IRUGO, proc_pid_limits),
2511 #ifdef CONFIG_SCHED_DEBUG
2512         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2513 #endif
2514 #ifdef CONFIG_SCHED_AUTOGROUP
2515         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2516 #endif
2517         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2518 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2519         INF("syscall",    S_IRUGO, proc_pid_syscall),
2520 #endif
2521         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2522         ONE("stat",       S_IRUGO, proc_tgid_stat),
2523         ONE("statm",      S_IRUGO, proc_pid_statm),
2524         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2525 #ifdef CONFIG_NUMA
2526         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2527 #endif
2528         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2529         LNK("cwd",        proc_cwd_link),
2530         LNK("root",       proc_root_link),
2531         LNK("exe",        proc_exe_link),
2532         REG("mounts",     S_IRUGO, proc_mounts_operations),
2533         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2534         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2535 #ifdef CONFIG_PROC_PAGE_MONITOR
2536         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2537         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2538         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
2539 #endif
2540 #ifdef CONFIG_SECURITY
2541         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2542 #endif
2543 #ifdef CONFIG_KALLSYMS
2544         INF("wchan",      S_IRUGO, proc_pid_wchan),
2545 #endif
2546 #ifdef CONFIG_STACKTRACE
2547         ONE("stack",      S_IRUGO, proc_pid_stack),
2548 #endif
2549 #ifdef CONFIG_SCHEDSTATS
2550         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
2551 #endif
2552 #ifdef CONFIG_LATENCYTOP
2553         REG("latency",  S_IRUGO, proc_lstats_operations),
2554 #endif
2555 #ifdef CONFIG_PROC_PID_CPUSET
2556         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2557 #endif
2558 #ifdef CONFIG_CGROUPS
2559         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2560 #endif
2561         INF("oom_score",  S_IRUGO, proc_oom_score),
2562         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2563         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2564 #ifdef CONFIG_AUDITSYSCALL
2565         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2566         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2567 #endif
2568 #ifdef CONFIG_FAULT_INJECTION
2569         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2570 #endif
2571 #ifdef CONFIG_ELF_CORE
2572         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2573 #endif
2574 #ifdef CONFIG_TASK_IO_ACCOUNTING
2575         INF("io",       S_IRUSR, proc_tgid_io_accounting),
2576 #endif
2577 #ifdef CONFIG_HARDWALL
2578         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
2579 #endif
2580 #ifdef CONFIG_USER_NS
2581         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2582         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2583         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2584 #endif
2585 };
2586
2587 static int proc_tgid_base_readdir(struct file * filp,
2588                              void * dirent, filldir_t filldir)
2589 {
2590         return proc_pident_readdir(filp,dirent,filldir,
2591                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2592 }
2593
2594 static const struct file_operations proc_tgid_base_operations = {
2595         .read           = generic_read_dir,
2596         .readdir        = proc_tgid_base_readdir,
2597         .llseek         = default_llseek,
2598 };
2599
2600 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2601 {
2602         return proc_pident_lookup(dir, dentry,
2603                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2604 }
2605
2606 static const struct inode_operations proc_tgid_base_inode_operations = {
2607         .lookup         = proc_tgid_base_lookup,
2608         .getattr        = pid_getattr,
2609         .setattr        = proc_setattr,
2610         .permission     = proc_pid_permission,
2611 };
2612
2613 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2614 {
2615         struct dentry *dentry, *leader, *dir;
2616         char buf[PROC_NUMBUF];
2617         struct qstr name;
2618
2619         name.name = buf;
2620         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2621         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2622         if (dentry) {
2623                 shrink_dcache_parent(dentry);
2624                 d_drop(dentry);
2625                 dput(dentry);
2626         }
2627
2628         name.name = buf;
2629         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2630         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2631         if (!leader)
2632                 goto out;
2633
2634         name.name = "task";
2635         name.len = strlen(name.name);
2636         dir = d_hash_and_lookup(leader, &name);
2637         if (!dir)
2638                 goto out_put_leader;
2639
2640         name.name = buf;
2641         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2642         dentry = d_hash_and_lookup(dir, &name);
2643         if (dentry) {
2644                 shrink_dcache_parent(dentry);
2645                 d_drop(dentry);
2646                 dput(dentry);
2647         }
2648
2649         dput(dir);
2650 out_put_leader:
2651         dput(leader);
2652 out:
2653         return;
2654 }
2655
2656 /**
2657  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2658  * @task: task that should be flushed.
2659  *
2660  * When flushing dentries from proc, one needs to flush them from global
2661  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2662  * in. This call is supposed to do all of this job.
2663  *
2664  * Looks in the dcache for
2665  * /proc/@pid
2666  * /proc/@tgid/task/@pid
2667  * if either directory is present flushes it and all of it'ts children
2668  * from the dcache.
2669  *
2670  * It is safe and reasonable to cache /proc entries for a task until
2671  * that task exits.  After that they just clog up the dcache with
2672  * useless entries, possibly causing useful dcache entries to be
2673  * flushed instead.  This routine is proved to flush those useless
2674  * dcache entries at process exit time.
2675  *
2676  * NOTE: This routine is just an optimization so it does not guarantee
2677  *       that no dcache entries will exist at process exit time it
2678  *       just makes it very unlikely that any will persist.
2679  */
2680
2681 void proc_flush_task(struct task_struct *task)
2682 {
2683         int i;
2684         struct pid *pid, *tgid;
2685         struct upid *upid;
2686
2687         pid = task_pid(task);
2688         tgid = task_tgid(task);
2689
2690         for (i = 0; i <= pid->level; i++) {
2691                 upid = &pid->numbers[i];
2692                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2693                                         tgid->numbers[i].nr);
2694         }
2695 }
2696
2697 static struct dentry *proc_pid_instantiate(struct inode *dir,
2698                                            struct dentry * dentry,
2699                                            struct task_struct *task, const void *ptr)
2700 {
2701         struct dentry *error = ERR_PTR(-ENOENT);
2702         struct inode *inode;
2703
2704         inode = proc_pid_make_inode(dir->i_sb, task);
2705         if (!inode)
2706                 goto out;
2707
2708         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2709         inode->i_op = &proc_tgid_base_inode_operations;
2710         inode->i_fop = &proc_tgid_base_operations;
2711         inode->i_flags|=S_IMMUTABLE;
2712
2713         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
2714                                                   ARRAY_SIZE(tgid_base_stuff)));
2715
2716         d_set_d_op(dentry, &pid_dentry_operations);
2717
2718         d_add(dentry, inode);
2719         /* Close the race of the process dying before we return the dentry */
2720         if (pid_revalidate(dentry, 0))
2721                 error = NULL;
2722 out:
2723         return error;
2724 }
2725
2726 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
2727 {
2728         struct dentry *result = NULL;
2729         struct task_struct *task;
2730         unsigned tgid;
2731         struct pid_namespace *ns;
2732
2733         tgid = name_to_int(dentry);
2734         if (tgid == ~0U)
2735                 goto out;
2736
2737         ns = dentry->d_sb->s_fs_info;
2738         rcu_read_lock();
2739         task = find_task_by_pid_ns(tgid, ns);
2740         if (task)
2741                 get_task_struct(task);
2742         rcu_read_unlock();
2743         if (!task)
2744                 goto out;
2745
2746         result = proc_pid_instantiate(dir, dentry, task, NULL);
2747         put_task_struct(task);
2748 out:
2749         return result;
2750 }
2751
2752 /*
2753  * Find the first task with tgid >= tgid
2754  *
2755  */
2756 struct tgid_iter {
2757         unsigned int tgid;
2758         struct task_struct *task;
2759 };
2760 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2761 {
2762         struct pid *pid;
2763
2764         if (iter.task)
2765                 put_task_struct(iter.task);
2766         rcu_read_lock();
2767 retry:
2768         iter.task = NULL;
2769         pid = find_ge_pid(iter.tgid, ns);
2770         if (pid) {
2771                 iter.tgid = pid_nr_ns(pid, ns);
2772                 iter.task = pid_task(pid, PIDTYPE_PID);
2773                 /* What we to know is if the pid we have find is the
2774                  * pid of a thread_group_leader.  Testing for task
2775                  * being a thread_group_leader is the obvious thing
2776                  * todo but there is a window when it fails, due to
2777                  * the pid transfer logic in de_thread.
2778                  *
2779                  * So we perform the straight forward test of seeing
2780                  * if the pid we have found is the pid of a thread
2781                  * group leader, and don't worry if the task we have
2782                  * found doesn't happen to be a thread group leader.
2783                  * As we don't care in the case of readdir.
2784                  */
2785                 if (!iter.task || !has_group_leader_pid(iter.task)) {
2786                         iter.tgid += 1;
2787                         goto retry;
2788                 }
2789                 get_task_struct(iter.task);
2790         }
2791         rcu_read_unlock();
2792         return iter;
2793 }
2794
2795 #define TGID_OFFSET (FIRST_PROCESS_ENTRY)
2796
2797 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2798         struct tgid_iter iter)
2799 {
2800         char name[PROC_NUMBUF];
2801         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
2802         return proc_fill_cache(filp, dirent, filldir, name, len,
2803                                 proc_pid_instantiate, iter.task, NULL);
2804 }
2805
2806 static int fake_filldir(void *buf, const char *name, int namelen,
2807                         loff_t offset, u64 ino, unsigned d_type)
2808 {
2809         return 0;
2810 }
2811
2812 /* for the /proc/ directory itself, after non-process stuff has been done */
2813 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2814 {
2815         struct tgid_iter iter;
2816         struct pid_namespace *ns;
2817         filldir_t __filldir;
2818
2819         if (filp->f_pos >= PID_MAX_LIMIT + TGID_OFFSET)
2820                 goto out;
2821
2822         ns = filp->f_dentry->d_sb->s_fs_info;
2823         iter.task = NULL;
2824         iter.tgid = filp->f_pos - TGID_OFFSET;
2825         for (iter = next_tgid(ns, iter);
2826              iter.task;
2827              iter.tgid += 1, iter = next_tgid(ns, iter)) {
2828                 if (has_pid_permissions(ns, iter.task, 2))
2829                         __filldir = filldir;
2830                 else
2831                         __filldir = fake_filldir;
2832
2833                 filp->f_pos = iter.tgid + TGID_OFFSET;
2834                 if (proc_pid_fill_cache(filp, dirent, __filldir, iter) < 0) {
2835                         put_task_struct(iter.task);
2836                         goto out;
2837                 }
2838         }
2839         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
2840 out:
2841         return 0;
2842 }
2843
2844 /*
2845  * Tasks
2846  */
2847 static const struct pid_entry tid_base_stuff[] = {
2848         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2849         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2850         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2851         REG("environ",   S_IRUSR, proc_environ_operations),
2852         INF("auxv",      S_IRUSR, proc_pid_auxv),
2853         ONE("status",    S_IRUGO, proc_pid_status),
2854         ONE("personality", S_IRUGO, proc_pid_personality),
2855         INF("limits",    S_IRUGO, proc_pid_limits),
2856 #ifdef CONFIG_SCHED_DEBUG
2857         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2858 #endif
2859         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2860 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2861         INF("syscall",   S_IRUGO, proc_pid_syscall),
2862 #endif
2863         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
2864         ONE("stat",      S_IRUGO, proc_tid_stat),
2865         ONE("statm",     S_IRUGO, proc_pid_statm),
2866         REG("maps",      S_IRUGO, proc_tid_maps_operations),
2867 #ifdef CONFIG_CHECKPOINT_RESTORE
2868         REG("children",  S_IRUGO, proc_tid_children_operations),
2869 #endif
2870 #ifdef CONFIG_NUMA
2871         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
2872 #endif
2873         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
2874         LNK("cwd",       proc_cwd_link),
2875         LNK("root",      proc_root_link),
2876         LNK("exe",       proc_exe_link),
2877         REG("mounts",    S_IRUGO, proc_mounts_operations),
2878         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2879 #ifdef CONFIG_PROC_PAGE_MONITOR
2880         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2881         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
2882         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
2883 #endif
2884 #ifdef CONFIG_SECURITY
2885         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2886 #endif
2887 #ifdef CONFIG_KALLSYMS
2888         INF("wchan",     S_IRUGO, proc_pid_wchan),
2889 #endif
2890 #ifdef CONFIG_STACKTRACE
2891         ONE("stack",      S_IRUGO, proc_pid_stack),
2892 #endif
2893 #ifdef CONFIG_SCHEDSTATS
2894         INF("schedstat", S_IRUGO, proc_pid_schedstat),
2895 #endif
2896 #ifdef CONFIG_LATENCYTOP
2897         REG("latency",  S_IRUGO, proc_lstats_operations),
2898 #endif
2899 #ifdef CONFIG_PROC_PID_CPUSET
2900         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
2901 #endif
2902 #ifdef CONFIG_CGROUPS
2903         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2904 #endif
2905         INF("oom_score", S_IRUGO, proc_oom_score),
2906         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2907         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2908 #ifdef CONFIG_AUDITSYSCALL
2909         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
2910         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2911 #endif
2912 #ifdef CONFIG_FAULT_INJECTION
2913         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2914 #endif
2915 #ifdef CONFIG_TASK_IO_ACCOUNTING
2916         INF("io",       S_IRUSR, proc_tid_io_accounting),
2917 #endif
2918 #ifdef CONFIG_HARDWALL
2919         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
2920 #endif
2921 #ifdef CONFIG_USER_NS
2922         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2923         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2924         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2925 #endif
2926 };
2927
2928 static int proc_tid_base_readdir(struct file * filp,
2929                              void * dirent, filldir_t filldir)
2930 {
2931         return proc_pident_readdir(filp,dirent,filldir,
2932                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
2933 }
2934
2935 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2936 {
2937         return proc_pident_lookup(dir, dentry,
2938                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
2939 }
2940
2941 static const struct file_operations proc_tid_base_operations = {
2942         .read           = generic_read_dir,
2943         .readdir        = proc_tid_base_readdir,
2944         .llseek         = default_llseek,
2945 };
2946
2947 static const struct inode_operations proc_tid_base_inode_operations = {
2948         .lookup         = proc_tid_base_lookup,
2949         .getattr        = pid_getattr,
2950         .setattr        = proc_setattr,
2951 };
2952
2953 static struct dentry *proc_task_instantiate(struct inode *dir,
2954         struct dentry *dentry, struct task_struct *task, const void *ptr)
2955 {
2956         struct dentry *error = ERR_PTR(-ENOENT);
2957         struct inode *inode;
2958         inode = proc_pid_make_inode(dir->i_sb, task);
2959
2960         if (!inode)
2961                 goto out;
2962         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2963         inode->i_op = &proc_tid_base_inode_operations;
2964         inode->i_fop = &proc_tid_base_operations;
2965         inode->i_flags|=S_IMMUTABLE;
2966
2967         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
2968                                                   ARRAY_SIZE(tid_base_stuff)));
2969
2970         d_set_d_op(dentry, &pid_dentry_operations);
2971
2972         d_add(dentry, inode);
2973         /* Close the race of the process dying before we return the dentry */
2974         if (pid_revalidate(dentry, 0))
2975                 error = NULL;
2976 out:
2977         return error;
2978 }
2979
2980 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
2981 {
2982         struct dentry *result = ERR_PTR(-ENOENT);
2983         struct task_struct *task;
2984         struct task_struct *leader = get_proc_task(dir);
2985         unsigned tid;
2986         struct pid_namespace *ns;
2987
2988         if (!leader)
2989                 goto out_no_task;
2990
2991         tid = name_to_int(dentry);
2992         if (tid == ~0U)
2993                 goto out;
2994
2995         ns = dentry->d_sb->s_fs_info;
2996         rcu_read_lock();
2997         task = find_task_by_pid_ns(tid, ns);
2998         if (task)
2999                 get_task_struct(task);
3000         rcu_read_unlock();
3001         if (!task)
3002                 goto out;
3003         if (!same_thread_group(leader, task))
3004                 goto out_drop_task;
3005
3006         result = proc_task_instantiate(dir, dentry, task, NULL);
3007 out_drop_task:
3008         put_task_struct(task);
3009 out:
3010         put_task_struct(leader);
3011 out_no_task:
3012         return result;
3013 }
3014
3015 /*
3016  * Find the first tid of a thread group to return to user space.
3017  *
3018  * Usually this is just the thread group leader, but if the users
3019  * buffer was too small or there was a seek into the middle of the
3020  * directory we have more work todo.
3021  *
3022  * In the case of a short read we start with find_task_by_pid.
3023  *
3024  * In the case of a seek we start with the leader and walk nr
3025  * threads past it.
3026  */
3027 static struct task_struct *first_tid(struct task_struct *leader,
3028                 int tid, int nr, struct pid_namespace *ns)
3029 {
3030         struct task_struct *pos;
3031
3032         rcu_read_lock();
3033         /* Attempt to start with the pid of a thread */
3034         if (tid && (nr > 0)) {
3035                 pos = find_task_by_pid_ns(tid, ns);
3036                 if (pos && (pos->group_leader == leader))
3037                         goto found;
3038         }
3039
3040         /* If nr exceeds the number of threads there is nothing todo */
3041         pos = NULL;
3042         if (nr && nr >= get_nr_threads(leader))
3043                 goto out;
3044
3045         /* If we haven't found our starting place yet start
3046          * with the leader and walk nr threads forward.
3047          */
3048         for (pos = leader; nr > 0; --nr) {
3049                 pos = next_thread(pos);
3050                 if (pos == leader) {
3051                         pos = NULL;
3052                         goto out;
3053                 }
3054         }
3055 found:
3056         get_task_struct(pos);
3057 out:
3058         rcu_read_unlock();
3059         return pos;
3060 }
3061
3062 /*
3063  * Find the next thread in the thread list.
3064  * Return NULL if there is an error or no next thread.
3065  *
3066  * The reference to the input task_struct is released.
3067  */
3068 static struct task_struct *next_tid(struct task_struct *start)
3069 {
3070         struct task_struct *pos = NULL;
3071         rcu_read_lock();
3072         if (pid_alive(start)) {
3073                 pos = next_thread(start);
3074                 if (thread_group_leader(pos))
3075                         pos = NULL;
3076                 else
3077                         get_task_struct(pos);
3078         }
3079         rcu_read_unlock();
3080         put_task_struct(start);
3081         return pos;
3082 }
3083
3084 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3085         struct task_struct *task, int tid)
3086 {
3087         char name[PROC_NUMBUF];
3088         int len = snprintf(name, sizeof(name), "%d", tid);
3089         return proc_fill_cache(filp, dirent, filldir, name, len,
3090                                 proc_task_instantiate, task, NULL);
3091 }
3092
3093 /* for the /proc/TGID/task/ directories */
3094 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3095 {
3096         struct dentry *dentry = filp->f_path.dentry;
3097         struct inode *inode = dentry->d_inode;
3098         struct task_struct *leader = NULL;
3099         struct task_struct *task;
3100         int retval = -ENOENT;
3101         ino_t ino;
3102         int tid;
3103         struct pid_namespace *ns;
3104
3105         task = get_proc_task(inode);
3106         if (!task)
3107                 goto out_no_task;
3108         rcu_read_lock();
3109         if (pid_alive(task)) {
3110                 leader = task->group_leader;
3111                 get_task_struct(leader);
3112         }
3113         rcu_read_unlock();
3114         put_task_struct(task);
3115         if (!leader)
3116                 goto out_no_task;
3117         retval = 0;
3118
3119         switch ((unsigned long)filp->f_pos) {
3120         case 0:
3121                 ino = inode->i_ino;
3122                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3123                         goto out;
3124                 filp->f_pos++;
3125                 /* fall through */
3126         case 1:
3127                 ino = parent_ino(dentry);
3128                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3129                         goto out;
3130                 filp->f_pos++;
3131                 /* fall through */
3132         }
3133
3134         /* f_version caches the tgid value that the last readdir call couldn't
3135          * return. lseek aka telldir automagically resets f_version to 0.
3136          */
3137         ns = filp->f_dentry->d_sb->s_fs_info;
3138         tid = (int)filp->f_version;
3139         filp->f_version = 0;
3140         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3141              task;
3142              task = next_tid(task), filp->f_pos++) {
3143                 tid = task_pid_nr_ns(task, ns);
3144                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3145                         /* returning this tgid failed, save it as the first
3146                          * pid for the next readir call */
3147                         filp->f_version = (u64)tid;
3148                         put_task_struct(task);
3149                         break;
3150                 }
3151         }
3152 out:
3153         put_task_struct(leader);
3154 out_no_task:
3155         return retval;
3156 }
3157
3158 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3159 {
3160         struct inode *inode = dentry->d_inode;
3161         struct task_struct *p = get_proc_task(inode);
3162         generic_fillattr(inode, stat);
3163
3164         if (p) {
3165                 stat->nlink += get_nr_threads(p);
3166                 put_task_struct(p);
3167         }
3168
3169         return 0;
3170 }
3171
3172 static const struct inode_operations proc_task_inode_operations = {
3173         .lookup         = proc_task_lookup,
3174         .getattr        = proc_task_getattr,
3175         .setattr        = proc_setattr,
3176         .permission     = proc_pid_permission,
3177 };
3178
3179 static const struct file_operations proc_task_operations = {
3180         .read           = generic_read_dir,
3181         .readdir        = proc_task_readdir,
3182         .llseek         = default_llseek,
3183 };