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 <[email protected]>
  20  *  Edjard Mota <[email protected]>
  21  *  Ilias Biris <[email protected]>
  22  *  Mauricio Lin <[email protected]>
  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 <[email protected]>:
  44  *  Smaps information related to shared, private, clean and dirty pages.
  45  *
  46  *  Paul Mundt <[email protected]>:
  47  *  Overall revision about smaps.
  48  */
  49
  50 #include <asm/uaccess.h>
  51
  52 #include <linux/errno.h>
  53 #include <linux/time.h>
  54 #include <linux/proc_fs.h>
  55 #include <linux/stat.h>
  56 #include <linux/task_io_accounting_ops.h>
  57 #include <linux/init.h>
  58 #include <linux/capability.h>
  59 #include <linux/file.h>
  60 #include <linux/fdtable.h>
  61 #include <linux/string.h>
  62 #include <linux/seq_file.h>
  63 #include <linux/namei.h>
  64 #include <linux/mnt_namespace.h>
  65 #include <linux/mm.h>
  66 #include <linux/swap.h>
  67 #include <linux/rcupdate.h>
  68 #include <linux/kallsyms.h>
  69 #include <linux/stacktrace.h>
  70 #include <linux/resource.h>
  71 #include <linux/module.h>
  72 #include <linux/mount.h>
  73 #include <linux/security.h>
  74 #include <linux/ptrace.h>
  75 #include <linux/tracehook.h>
  76 #include <linux/printk.h>
  77 #include <linux/cgroup.h>
  78 #include <linux/cpuset.h>
  79 #include <linux/audit.h>
  80 #include <linux/poll.h>
  81 #include <linux/nsproxy.h>
  82 #include <linux/oom.h>
  83 #include <linux/elf.h>
  84 #include <linux/pid_namespace.h>
  85 #include <linux/user_namespace.h>
  86 #include <linux/fs_struct.h>
  87 #include <linux/slab.h>
  88 #include <linux/flex_array.h>
  89 #include <linux/posix-timers.h>
  90 #ifdef CONFIG_HARDWALL
  91 #include <asm/hardwall.h>
  92 #endif
  93 #include <trace/events/oom.h>
  94 #include "internal.h"
  95 #include "fd.h"
  96
  97 /* NOTE:
  98  *      Implementing inode permission operations in /proc is almost
  99  *      certainly an error.  Permission checks need to happen during
 100  *      each system call not at open time.  The reason is that most of
 101  *      what we wish to check for permissions in /proc varies at runtime.
 102  *
 103  *      The classic example of a problem is opening file descriptors
 104  *      in /proc for a task before it execs a suid executable.
 105  */
 106
 107 struct pid_entry {
 108         char *name;
 109         int len;
 110         umode_t mode;
 111         const struct inode_operations *iop;
 112         const struct file_operations *fop;
 113         union proc_op op;
 114 };
 115
 116 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 117         .name = (NAME),                                 \
 118         .len  = sizeof(NAME) - 1,                       \
 119         .mode = MODE,                                   \
 120         .iop  = IOP,                                    \
 121         .fop  = FOP,                                    \
 122         .op   = OP,                                     \
 123 }
 124
 125 #define DIR(NAME, MODE, iops, fops)     \
 126         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 127 #define LNK(NAME, get_link)                                     \
 128         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 129                 &proc_pid_link_inode_operations, NULL,          \
 130                 { .proc_get_link = get_link } )
 131 #define REG(NAME, MODE, fops)                           \
 132         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 133 #define INF(NAME, MODE, read)                           \
 134         NOD(NAME, (S_IFREG|(MODE)),                     \
 135                 NULL, &proc_info_file_operations,       \
 136                 { .proc_read = read } )
 137 #define ONE(NAME, MODE, show)                           \
 138         NOD(NAME, (S_IFREG|(MODE)),                     \
 139                 NULL, &proc_single_file_operations,     \
 140                 { .proc_show = show } )
 141
 142 /*
 143  * Count the number of hardlinks for the pid_entry table, excluding the .
 144  * and .. links.
 145  */
 146 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
 147         unsigned int n)
 148 {
 149         unsigned int i;
 150         unsigned int count;
 151
 152         count = 0;
 153         for (i = 0; i < n; ++i) {
 154                 if (S_ISDIR(entries[i].mode))
 155                         ++count;
 156         }
 157
 158         return count;
 159 }
 160
 161 static int get_task_root(struct task_struct *task, struct path *root)
 162 {
 163         int result = -ENOENT;
 164
 165         task_lock(task);
 166         if (task->fs) {
 167                 get_fs_root(task->fs, root);
 168                 result = 0;
 169         }
 170         task_unlock(task);
 171         return result;
 172 }
 173
 174 static int proc_cwd_link(struct dentry *dentry, struct path *path)
 175 {
 176         struct task_struct *task = get_proc_task(dentry->d_inode);
 177         int result = -ENOENT;
 178
 179         if (task) {
 180                 task_lock(task);
 181                 if (task->fs) {
 182                         get_fs_pwd(task->fs, path);
 183                         result = 0;
 184                 }
 185                 task_unlock(task);
 186                 put_task_struct(task);
 187         }
 188         return result;
 189 }
 190
 191 static int proc_root_link(struct dentry *dentry, struct path *path)
 192 {
 193         struct task_struct *task = get_proc_task(dentry->d_inode);
 194         int result = -ENOENT;
 195
 196         if (task) {
 197                 result = get_task_root(task, path);
 198                 put_task_struct(task);
 199         }
 200         return result;
 201 }
 202
 203 static int proc_pid_cmdline(struct task_struct *task, char *buffer)
 204 {
 205         return get_cmdline(task, buffer, PAGE_SIZE);
 206 }
 207
 208 static int proc_pid_auxv(struct task_struct *task, char *buffer)
 209 {
 210         struct mm_struct *mm = mm_access(task, PTRACE_MODE_READ);
 211         int res = PTR_ERR(mm);
 212         if (mm && !IS_ERR(mm)) {
 213                 unsigned int nwords = 0;
 214                 do {
 215                         nwords += 2;
 216                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 217                 res = nwords * sizeof(mm->saved_auxv[0]);
 218                 if (res > PAGE_SIZE)
 219                         res = PAGE_SIZE;
 220                 memcpy(buffer, mm->saved_auxv, res);
 221                 mmput(mm);
 222         }
 223         return res;
 224 }
 225
 226
 227 #ifdef CONFIG_KALLSYMS
 228 /*
 229  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 230  * Returns the resolved symbol.  If that fails, simply return the address.
 231  */
 232 static int proc_pid_wchan(struct task_struct *task, char *buffer)
 233 {
 234         unsigned long wchan;
 235         char symname[KSYM_NAME_LEN];
 236
 237         wchan = get_wchan(task);
 238
 239         if (lookup_symbol_name(wchan, symname) < 0)
 240                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
 241                         return 0;
 242                 else
 243                         return sprintf(buffer, "%lu", wchan);
 244         else
 245                 return sprintf(buffer, "%s", symname);
 246 }
 247 #endif /* CONFIG_KALLSYMS */
 248
 249 static int lock_trace(struct task_struct *task)
 250 {
 251         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
 252         if (err)
 253                 return err;
 254         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH)) {
 255                 mutex_unlock(&task->signal->cred_guard_mutex);
 256                 return -EPERM;
 257         }
 258         return 0;
 259 }
 260
 261 static void unlock_trace(struct task_struct *task)
 262 {
 263         mutex_unlock(&task->signal->cred_guard_mutex);
 264 }
 265
 266 #ifdef CONFIG_STACKTRACE
 267
 268 #define MAX_STACK_TRACE_DEPTH   64
 269
 270 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 271                           struct pid *pid, struct task_struct *task)
 272 {
 273         struct stack_trace trace;
 274         unsigned long *entries;
 275         int err;
 276         int i;
 277
 278         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 279         if (!entries)
 280                 return -ENOMEM;
 281
 282         trace.nr_entries        = 0;
 283         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 284         trace.entries           = entries;
 285         trace.skip              = 0;
 286
 287         err = lock_trace(task);
 288         if (!err) {
 289                 save_stack_trace_tsk(task, &trace);
 290
 291                 for (i = 0; i < trace.nr_entries; i++) {
 292                         seq_printf(m, "[<%pK>] %pS\n",
 293                                    (void *)entries[i], (void *)entries[i]);
 294                 }
 295                 unlock_trace(task);
 296         }
 297         kfree(entries);
 298
 299         return err;
 300 }
 301 #endif
 302
 303 #ifdef CONFIG_SCHEDSTATS
 304 /*
 305  * Provides /proc/PID/schedstat
 306  */
 307 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
 308 {
 309         return sprintf(buffer, "%llu %llu %lu\n",
 310                         (unsigned long long)task->se.sum_exec_runtime,
 311                         (unsigned long long)task->sched_info.run_delay,
 312                         task->sched_info.pcount);
 313 }
 314 #endif
 315
 316 #ifdef CONFIG_LATENCYTOP
 317 static int lstats_show_proc(struct seq_file *m, void *v)
 318 {
 319         int i;
 320         struct inode *inode = m->private;
 321         struct task_struct *task = get_proc_task(inode);
 322
 323         if (!task)
 324                 return -ESRCH;
 325         seq_puts(m, "Latency Top version : v0.1\n");
 326         for (i = 0; i < 32; i++) {
 327                 struct latency_record *lr = &task->latency_record[i];
 328                 if (lr->backtrace[0]) {
 329                         int q;
 330                         seq_printf(m, "%i %li %li",
 331                                    lr->count, lr->time, lr->max);
 332                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 333                                 unsigned long bt = lr->backtrace[q];
 334                                 if (!bt)
 335                                         break;
 336                                 if (bt == ULONG_MAX)
 337                                         break;
 338                                 seq_printf(m, " %ps", (void *)bt);
 339                         }
 340                         seq_putc(m, '\n');
 341                 }
 342
 343         }
 344         put_task_struct(task);
 345         return 0;
 346 }
 347
 348 static int lstats_open(struct inode *inode, struct file *file)
 349 {
 350         return single_open(file, lstats_show_proc, inode);
 351 }
 352
 353 static ssize_t lstats_write(struct file *file, const char __user *buf,
 354                             size_t count, loff_t *offs)
 355 {
 356         struct task_struct *task = get_proc_task(file_inode(file));
 357
 358         if (!task)
 359                 return -ESRCH;
 360         clear_all_latency_tracing(task);
 361         put_task_struct(task);
 362
 363         return count;
 364 }
 365
 366 static const struct file_operations proc_lstats_operations = {
 367         .open           = lstats_open,
 368         .read           = seq_read,
 369         .write          = lstats_write,
 370         .llseek         = seq_lseek,
 371         .release        = single_release,
 372 };
 373
 374 #endif
 375
 376 #ifdef CONFIG_CGROUPS
 377 static int cgroup_open(struct inode *inode, struct file *file)
 378 {
 379         struct pid *pid = PROC_I(inode)->pid;
 380         return single_open(file, proc_cgroup_show, pid);
 381 }
 382
 383 static const struct file_operations proc_cgroup_operations = {
 384         .open           = cgroup_open,
 385         .read           = seq_read,
 386         .llseek         = seq_lseek,
 387         .release        = single_release,
 388 };
 389 #endif
 390
 391 #ifdef CONFIG_PROC_PID_CPUSET
 392
 393 static int cpuset_open(struct inode *inode, struct file *file)
 394 {
 395         struct pid *pid = PROC_I(inode)->pid;
 396         return single_open(file, proc_cpuset_show, pid);
 397 }
 398
 399 static const struct file_operations proc_cpuset_operations = {
 400         .open           = cpuset_open,
 401         .read           = seq_read,
 402         .llseek         = seq_lseek,
 403         .release        = single_release,
 404 };
 405 #endif
 406
 407 static int proc_oom_score(struct task_struct *task, char *buffer)
 408 {
 409         unsigned long totalpages = totalram_pages + total_swap_pages;
 410         unsigned long points = 0;
 411
 412         read_lock(&tasklist_lock);
 413         if (pid_alive(task))
 414                 points = oom_badness(task, NULL, NULL, totalpages) *
 415                                                 1000 / totalpages;
 416         read_unlock(&tasklist_lock);
 417         return sprintf(buffer, "%lu\n", points);
 418 }
 419
 420 struct limit_names {
 421         char *name;
 422         char *unit;
 423 };
 424
 425 static const struct limit_names lnames[RLIM_NLIMITS] = {
 426         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 427         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 428         [RLIMIT_DATA] = {"Max data size", "bytes"},
 429         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 430         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 431         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 432         [RLIMIT_NPROC] = {"Max processes", "processes"},
 433         [RLIMIT_NOFILE] = {"Max open files", "files"},
 434         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 435         [RLIMIT_AS] = {"Max address space", "bytes"},
 436         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 437         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 438         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 439         [RLIMIT_NICE] = {"Max nice priority", NULL},
 440         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 441         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 442 };
 443
 444 /* Display limits for a process */
 445 static int proc_pid_limits(struct task_struct *task, char *buffer)
 446 {
 447         unsigned int i;
 448         int count = 0;
 449         unsigned long flags;
 450         char *bufptr = buffer;
 451
 452         struct rlimit rlim[RLIM_NLIMITS];
 453
 454         if (!lock_task_sighand(task, &flags))
 455                 return 0;
 456         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 457         unlock_task_sighand(task, &flags);
 458
 459         /*
 460          * print the file header
 461          */
 462         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
 463                         "Limit", "Soft Limit", "Hard Limit", "Units");
 464
 465         for (i = 0; i < RLIM_NLIMITS; i++) {
 466                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 467                         count += sprintf(&bufptr[count], "%-25s %-20s ",
 468                                          lnames[i].name, "unlimited");
 469                 else
 470                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
 471                                          lnames[i].name, rlim[i].rlim_cur);
 472
 473                 if (rlim[i].rlim_max == RLIM_INFINITY)
 474                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
 475                 else
 476                         count += sprintf(&bufptr[count], "%-20lu ",
 477                                          rlim[i].rlim_max);
 478
 479                 if (lnames[i].unit)
 480                         count += sprintf(&bufptr[count], "%-10s\n",
 481                                          lnames[i].unit);
 482                 else
 483                         count += sprintf(&bufptr[count], "\n");
 484         }
 485
 486         return count;
 487 }
 488
 489 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 490 static int proc_pid_syscall(struct task_struct *task, char *buffer)
 491 {
 492         long nr;
 493         unsigned long args[6], sp, pc;
 494         int res = lock_trace(task);
 495         if (res)
 496                 return res;
 497
 498         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 499                 res = sprintf(buffer, "running\n");
 500         else if (nr < 0)
 501                 res = sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 502         else
 503                 res = sprintf(buffer,
 504                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 505                        nr,
 506                        args[0], args[1], args[2], args[3], args[4], args[5],
 507                        sp, pc);
 508         unlock_trace(task);
 509         return res;
 510 }
 511 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 512
 513 /************************************************************************/
 514 /*                       Here the fs part begins                        */
 515 /************************************************************************/
 516
 517 /* permission checks */
 518 static int proc_fd_access_allowed(struct inode *inode)
 519 {
 520         struct task_struct *task;
 521         int allowed = 0;
 522         /* Allow access to a task's file descriptors if it is us or we
 523          * may use ptrace attach to the process and find out that
 524          * information.
 525          */
 526         task = get_proc_task(inode);
 527         if (task) {
 528                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
 529                 put_task_struct(task);
 530         }
 531         return allowed;
 532 }
 533
 534 int proc_setattr(struct dentry *dentry, struct iattr *attr)
 535 {
 536         int error;
 537         struct inode *inode = dentry->d_inode;
 538
 539         if (attr->ia_valid & ATTR_MODE)
 540                 return -EPERM;
 541
 542         error = inode_change_ok(inode, attr);
 543         if (error)
 544                 return error;
 545
 546         setattr_copy(inode, attr);
 547         mark_inode_dirty(inode);
 548         return 0;
 549 }
 550
 551 /*
 552  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
 553  * or euid/egid (for hide_pid_min=2)?
 554  */
 555 static bool has_pid_permissions(struct pid_namespace *pid,
 556                                  struct task_struct *task,
 557                                  int hide_pid_min)
 558 {
 559         if (pid->hide_pid < hide_pid_min)
 560                 return true;
 561         if (in_group_p(pid->pid_gid))
 562                 return true;
 563         return ptrace_may_access(task, PTRACE_MODE_READ);
 564 }
 565
 566
 567 static int proc_pid_permission(struct inode *inode, int mask)
 568 {
 569         struct pid_namespace *pid = inode->i_sb->s_fs_info;
 570         struct task_struct *task;
 571         bool has_perms;
 572
 573         task = get_proc_task(inode);
 574         if (!task)
 575                 return -ESRCH;
 576         has_perms = has_pid_permissions(pid, task, 1);
 577         put_task_struct(task);
 578
 579         if (!has_perms) {
 580                 if (pid->hide_pid == 2) {
 581                         /*
 582                          * Let's make getdents(), stat(), and open()
 583                          * consistent with each other.  If a process
 584                          * may not stat() a file, it shouldn't be seen
 585                          * in procfs at all.
 586                          */
 587                         return -ENOENT;
 588                 }
 589
 590                 return -EPERM;
 591         }
 592         return generic_permission(inode, mask);
 593 }
 594
 595
 596
 597 static const struct inode_operations proc_def_inode_operations = {
 598         .setattr        = proc_setattr,
 599 };
 600
 601 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 602
 603 static ssize_t proc_info_read(struct file * file, char __user * buf,
 604                           size_t count, loff_t *ppos)
 605 {
 606         struct inode * inode = file_inode(file);
 607         unsigned long page;
 608         ssize_t length;
 609         struct task_struct *task = get_proc_task(inode);
 610
 611         length = -ESRCH;
 612         if (!task)
 613                 goto out_no_task;
 614
 615         if (count > PROC_BLOCK_SIZE)
 616                 count = PROC_BLOCK_SIZE;
 617
 618         length = -ENOMEM;
 619         if (!(page = __get_free_page(GFP_TEMPORARY)))
 620                 goto out;
 621
 622         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 623
 624         if (length >= 0)
 625                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 626         free_page(page);
 627 out:
 628         put_task_struct(task);
 629 out_no_task:
 630         return length;
 631 }
 632
 633 static const struct file_operations proc_info_file_operations = {
 634         .read           = proc_info_read,
 635         .llseek         = generic_file_llseek,
 636 };
 637
 638 static int proc_single_show(struct seq_file *m, void *v)
 639 {
 640         struct inode *inode = m->private;
 641         struct pid_namespace *ns;
 642         struct pid *pid;
 643         struct task_struct *task;
 644         int ret;
 645
 646         ns = inode->i_sb->s_fs_info;
 647         pid = proc_pid(inode);
 648         task = get_pid_task(pid, PIDTYPE_PID);
 649         if (!task)
 650                 return -ESRCH;
 651
 652         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 653
 654         put_task_struct(task);
 655         return ret;
 656 }
 657
 658 static int proc_single_open(struct inode *inode, struct file *filp)
 659 {
 660         return single_open(filp, proc_single_show, inode);
 661 }
 662
 663 static const struct file_operations proc_single_file_operations = {
 664         .open           = proc_single_open,
 665         .read           = seq_read,
 666         .llseek         = seq_lseek,
 667         .release        = single_release,
 668 };
 669
 670 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
 671 {
 672         struct task_struct *task = get_proc_task(file_inode(file));
 673         struct mm_struct *mm;
 674
 675         if (!task)
 676                 return -ESRCH;
 677
 678         mm = mm_access(task, mode);
 679         put_task_struct(task);
 680
 681         if (IS_ERR(mm))
 682                 return PTR_ERR(mm);
 683
 684         if (mm) {
 685                 /* ensure this mm_struct can't be freed */
 686                 atomic_inc(&mm->mm_count);
 687                 /* but do not pin its memory */
 688                 mmput(mm);
 689         }
 690
 691         file->private_data = mm;
 692
 693         return 0;
 694 }
 695
 696 static int mem_open(struct inode *inode, struct file *file)
 697 {
 698         int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
 699
 700         /* OK to pass negative loff_t, we can catch out-of-range */
 701         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 702
 703         return ret;
 704 }
 705
 706 static ssize_t mem_rw(struct file *file, char __user *buf,
 707                         size_t count, loff_t *ppos, int write)
 708 {
 709         struct mm_struct *mm = file->private_data;
 710         unsigned long addr = *ppos;
 711         ssize_t copied;
 712         char *page;
 713
 714         if (!mm)
 715                 return 0;
 716
 717         page = (char *)__get_free_page(GFP_TEMPORARY);
 718         if (!page)
 719                 return -ENOMEM;
 720
 721         copied = 0;
 722         if (!atomic_inc_not_zero(&mm->mm_users))
 723                 goto free;
 724
 725         while (count > 0) {
 726                 int this_len = min_t(int, count, PAGE_SIZE);
 727
 728                 if (write && copy_from_user(page, buf, this_len)) {
 729                         copied = -EFAULT;
 730                         break;
 731                 }
 732
 733                 this_len = access_remote_vm(mm, addr, page, this_len, write);
 734                 if (!this_len) {
 735                         if (!copied)
 736                                 copied = -EIO;
 737                         break;
 738                 }
 739
 740                 if (!write && copy_to_user(buf, page, this_len)) {
 741                         copied = -EFAULT;
 742                         break;
 743                 }
 744
 745                 buf += this_len;
 746                 addr += this_len;
 747                 copied += this_len;
 748                 count -= this_len;
 749         }
 750         *ppos = addr;
 751
 752         mmput(mm);
 753 free:
 754         free_page((unsigned long) page);
 755         return copied;
 756 }
 757
 758 static ssize_t mem_read(struct file *file, char __user *buf,
 759                         size_t count, loff_t *ppos)
 760 {
 761         return mem_rw(file, buf, count, ppos, 0);
 762 }
 763
 764 static ssize_t mem_write(struct file *file, const char __user *buf,
 765                          size_t count, loff_t *ppos)
 766 {
 767         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 768 }
 769
 770 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 771 {
 772         switch (orig) {
 773         case 0:
 774                 file->f_pos = offset;
 775                 break;
 776         case 1:
 777                 file->f_pos += offset;
 778                 break;
 779         default:
 780                 return -EINVAL;
 781         }
 782         force_successful_syscall_return();
 783         return file->f_pos;
 784 }
 785
 786 static int mem_release(struct inode *inode, struct file *file)
 787 {
 788         struct mm_struct *mm = file->private_data;
 789         if (mm)
 790                 mmdrop(mm);
 791         return 0;
 792 }
 793
 794 static const struct file_operations proc_mem_operations = {
 795         .llseek         = mem_lseek,
 796         .read           = mem_read,
 797         .write          = mem_write,
 798         .open           = mem_open,
 799         .release        = mem_release,
 800 };
 801
 802 static int environ_open(struct inode *inode, struct file *file)
 803 {
 804         return __mem_open(inode, file, PTRACE_MODE_READ);
 805 }
 806
 807 static ssize_t environ_read(struct file *file, char __user *buf,
 808                         size_t count, loff_t *ppos)
 809 {
 810         char *page;
 811         unsigned long src = *ppos;
 812         int ret = 0;
 813         struct mm_struct *mm = file->private_data;
 814
 815         if (!mm)
 816                 return 0;
 817
 818         page = (char *)__get_free_page(GFP_TEMPORARY);
 819         if (!page)
 820                 return -ENOMEM;
 821
 822         ret = 0;
 823         if (!atomic_inc_not_zero(&mm->mm_users))
 824                 goto free;
 825         while (count > 0) {
 826                 size_t this_len, max_len;
 827                 int retval;
 828
 829                 if (src >= (mm->env_end - mm->env_start))
 830                         break;
 831
 832                 this_len = mm->env_end - (mm->env_start + src);
 833
 834                 max_len = min_t(size_t, PAGE_SIZE, count);
 835                 this_len = min(max_len, this_len);
 836
 837                 retval = access_remote_vm(mm, (mm->env_start + src),
 838                         page, this_len, 0);
 839
 840                 if (retval <= 0) {
 841                         ret = retval;
 842                         break;
 843                 }
 844
 845                 if (copy_to_user(buf, page, retval)) {
 846                         ret = -EFAULT;
 847                         break;
 848                 }
 849
 850                 ret += retval;
 851                 src += retval;
 852                 buf += retval;
 853                 count -= retval;
 854         }
 855         *ppos = src;
 856         mmput(mm);
 857
 858 free:
 859         free_page((unsigned long) page);
 860         return ret;
 861 }
 862
 863 static const struct file_operations proc_environ_operations = {
 864         .open           = environ_open,
 865         .read           = environ_read,
 866         .llseek         = generic_file_llseek,
 867         .release        = mem_release,
 868 };
 869
 870 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
 871                             loff_t *ppos)
 872 {
 873         struct task_struct *task = get_proc_task(file_inode(file));
 874         char buffer[PROC_NUMBUF];
 875         int oom_adj = OOM_ADJUST_MIN;
 876         size_t len;
 877         unsigned long flags;
 878
 879         if (!task)
 880                 return -ESRCH;
 881         if (lock_task_sighand(task, &flags)) {
 882                 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
 883                         oom_adj = OOM_ADJUST_MAX;
 884                 else
 885                         oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
 886                                   OOM_SCORE_ADJ_MAX;
 887                 unlock_task_sighand(task, &flags);
 888         }
 889         put_task_struct(task);
 890         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
 891         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 892 }
 893
 894 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
 895                              size_t count, loff_t *ppos)
 896 {
 897         struct task_struct *task;
 898         char buffer[PROC_NUMBUF];
 899         int oom_adj;
 900         unsigned long flags;
 901         int err;
 902
 903         memset(buffer, 0, sizeof(buffer));
 904         if (count > sizeof(buffer) - 1)
 905                 count = sizeof(buffer) - 1;
 906         if (copy_from_user(buffer, buf, count)) {
 907                 err = -EFAULT;
 908                 goto out;
 909         }
 910
 911         err = kstrtoint(strstrip(buffer), 0, &oom_adj);
 912         if (err)
 913                 goto out;
 914         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
 915              oom_adj != OOM_DISABLE) {
 916                 err = -EINVAL;
 917                 goto out;
 918         }
 919
 920         task = get_proc_task(file_inode(file));
 921         if (!task) {
 922                 err = -ESRCH;
 923                 goto out;
 924         }
 925
 926         task_lock(task);
 927         if (!task->mm) {
 928                 err = -EINVAL;
 929                 goto err_task_lock;
 930         }
 931
 932         if (!lock_task_sighand(task, &flags)) {
 933                 err = -ESRCH;
 934                 goto err_task_lock;
 935         }
 936
 937         /*
 938          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
 939          * value is always attainable.
 940          */
 941         if (oom_adj == OOM_ADJUST_MAX)
 942                 oom_adj = OOM_SCORE_ADJ_MAX;
 943         else
 944                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
 945
 946         if (oom_adj < task->signal->oom_score_adj &&
 947             !capable(CAP_SYS_RESOURCE)) {
 948                 err = -EACCES;
 949                 goto err_sighand;
 950         }
 951
 952         /*
 953          * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
 954          * /proc/pid/oom_score_adj instead.
 955          */
 956         pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
 957                   current->comm, task_pid_nr(current), task_pid_nr(task),
 958                   task_pid_nr(task));
 959
 960         task->signal->oom_score_adj = oom_adj;
 961         trace_oom_score_adj_update(task);
 962 err_sighand:
 963         unlock_task_sighand(task, &flags);
 964 err_task_lock:
 965         task_unlock(task);
 966         put_task_struct(task);
 967 out:
 968         return err < 0 ? err : count;
 969 }
 970
 971 static const struct file_operations proc_oom_adj_operations = {
 972         .read           = oom_adj_read,
 973         .write          = oom_adj_write,
 974         .llseek         = generic_file_llseek,
 975 };
 976
 977 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
 978                                         size_t count, loff_t *ppos)
 979 {
 980         struct task_struct *task = get_proc_task(file_inode(file));
 981         char buffer[PROC_NUMBUF];
 982         short oom_score_adj = OOM_SCORE_ADJ_MIN;
 983         unsigned long flags;
 984         size_t len;
 985
 986         if (!task)
 987                 return -ESRCH;
 988         if (lock_task_sighand(task, &flags)) {
 989                 oom_score_adj = task->signal->oom_score_adj;
 990                 unlock_task_sighand(task, &flags);
 991         }
 992         put_task_struct(task);
 993         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
 994         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 995 }
 996
 997 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
 998                                         size_t count, loff_t *ppos)
 999 {
1000         struct task_struct *task;
1001         char buffer[PROC_NUMBUF];
1002         unsigned long flags;
1003         int oom_score_adj;
1004         int err;
1005
1006         memset(buffer, 0, sizeof(buffer));
1007         if (count > sizeof(buffer) - 1)
1008                 count = sizeof(buffer) - 1;
1009         if (copy_from_user(buffer, buf, count)) {
1010                 err = -EFAULT;
1011                 goto out;
1012         }
1013
1014         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1015         if (err)
1016                 goto out;
1017         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1018                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1019                 err = -EINVAL;
1020                 goto out;
1021         }
1022
1023         task = get_proc_task(file_inode(file));
1024         if (!task) {
1025                 err = -ESRCH;
1026                 goto out;
1027         }
1028
1029         task_lock(task);
1030         if (!task->mm) {
1031                 err = -EINVAL;
1032                 goto err_task_lock;
1033         }
1034
1035         if (!lock_task_sighand(task, &flags)) {
1036                 err = -ESRCH;
1037                 goto err_task_lock;
1038         }
1039
1040         if ((short)oom_score_adj < task->signal->oom_score_adj_min &&
1041                         !capable(CAP_SYS_RESOURCE)) {
1042                 err = -EACCES;
1043                 goto err_sighand;
1044         }
1045
1046         task->signal->oom_score_adj = (short)oom_score_adj;
1047         if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1048                 task->signal->oom_score_adj_min = (short)oom_score_adj;
1049         trace_oom_score_adj_update(task);
1050
1051 err_sighand:
1052         unlock_task_sighand(task, &flags);
1053 err_task_lock:
1054         task_unlock(task);
1055         put_task_struct(task);
1056 out:
1057         return err < 0 ? err : count;
1058 }
1059
1060 static const struct file_operations proc_oom_score_adj_operations = {
1061         .read           = oom_score_adj_read,
1062         .write          = oom_score_adj_write,
1063         .llseek         = default_llseek,
1064 };
1065
1066 #ifdef CONFIG_AUDITSYSCALL
1067 #define TMPBUFLEN 21
1068 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1069                                   size_t count, loff_t *ppos)
1070 {
1071         struct inode * inode = file_inode(file);
1072         struct task_struct *task = get_proc_task(inode);
1073         ssize_t length;
1074         char tmpbuf[TMPBUFLEN];
1075
1076         if (!task)
1077                 return -ESRCH;
1078         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1079                            from_kuid(file->f_cred->user_ns,
1080                                      audit_get_loginuid(task)));
1081         put_task_struct(task);
1082         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1083 }
1084
1085 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1086                                    size_t count, loff_t *ppos)
1087 {
1088         struct inode * inode = file_inode(file);
1089         char *page, *tmp;
1090         ssize_t length;
1091         uid_t loginuid;
1092         kuid_t kloginuid;
1093
1094         rcu_read_lock();
1095         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1096                 rcu_read_unlock();
1097                 return -EPERM;
1098         }
1099         rcu_read_unlock();
1100
1101         if (count >= PAGE_SIZE)
1102                 count = PAGE_SIZE - 1;
1103
1104         if (*ppos != 0) {
1105                 /* No partial writes. */
1106                 return -EINVAL;
1107         }
1108         page = (char*)__get_free_page(GFP_TEMPORARY);
1109         if (!page)
1110                 return -ENOMEM;
1111         length = -EFAULT;
1112         if (copy_from_user(page, buf, count))
1113                 goto out_free_page;
1114
1115         page[count] = '\0';
1116         loginuid = simple_strtoul(page, &tmp, 10);
1117         if (tmp == page) {
1118                 length = -EINVAL;
1119                 goto out_free_page;
1120
1121         }
1122
1123         /* is userspace tring to explicitly UNSET the loginuid? */
1124         if (loginuid == AUDIT_UID_UNSET) {
1125                 kloginuid = INVALID_UID;
1126         } else {
1127                 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1128                 if (!uid_valid(kloginuid)) {
1129                         length = -EINVAL;
1130                         goto out_free_page;
1131                 }
1132         }
1133
1134         length = audit_set_loginuid(kloginuid);
1135         if (likely(length == 0))
1136                 length = count;
1137
1138 out_free_page:
1139         free_page((unsigned long) page);
1140         return length;
1141 }
1142
1143 static const struct file_operations proc_loginuid_operations = {
1144         .read           = proc_loginuid_read,
1145         .write          = proc_loginuid_write,
1146         .llseek         = generic_file_llseek,
1147 };
1148
1149 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1150                                   size_t count, loff_t *ppos)
1151 {
1152         struct inode * inode = file_inode(file);
1153         struct task_struct *task = get_proc_task(inode);
1154         ssize_t length;
1155         char tmpbuf[TMPBUFLEN];
1156
1157         if (!task)
1158                 return -ESRCH;
1159         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1160                                 audit_get_sessionid(task));
1161         put_task_struct(task);
1162         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1163 }
1164
1165 static const struct file_operations proc_sessionid_operations = {
1166         .read           = proc_sessionid_read,
1167         .llseek         = generic_file_llseek,
1168 };
1169 #endif
1170
1171 #ifdef CONFIG_FAULT_INJECTION
1172 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1173                                       size_t count, loff_t *ppos)
1174 {
1175         struct task_struct *task = get_proc_task(file_inode(file));
1176         char buffer[PROC_NUMBUF];
1177         size_t len;
1178         int make_it_fail;
1179
1180         if (!task)
1181                 return -ESRCH;
1182         make_it_fail = task->make_it_fail;
1183         put_task_struct(task);
1184
1185         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1186
1187         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1188 }
1189
1190 static ssize_t proc_fault_inject_write(struct file * file,
1191                         const char __user * buf, size_t count, loff_t *ppos)
1192 {
1193         struct task_struct *task;
1194         char buffer[PROC_NUMBUF], *end;
1195         int make_it_fail;
1196
1197         if (!capable(CAP_SYS_RESOURCE))
1198                 return -EPERM;
1199         memset(buffer, 0, sizeof(buffer));
1200         if (count > sizeof(buffer) - 1)
1201                 count = sizeof(buffer) - 1;
1202         if (copy_from_user(buffer, buf, count))
1203                 return -EFAULT;
1204         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1205         if (*end)
1206                 return -EINVAL;
1207         if (make_it_fail < 0 || make_it_fail > 1)
1208                 return -EINVAL;
1209
1210         task = get_proc_task(file_inode(file));
1211         if (!task)
1212                 return -ESRCH;
1213         task->make_it_fail = make_it_fail;
1214         put_task_struct(task);
1215
1216         return count;
1217 }
1218
1219 static const struct file_operations proc_fault_inject_operations = {
1220         .read           = proc_fault_inject_read,
1221         .write          = proc_fault_inject_write,
1222         .llseek         = generic_file_llseek,
1223 };
1224 #endif
1225
1226
1227 #ifdef CONFIG_SCHED_DEBUG
1228 /*
1229  * Print out various scheduling related per-task fields:
1230  */
1231 static int sched_show(struct seq_file *m, void *v)
1232 {
1233         struct inode *inode = m->private;
1234         struct task_struct *p;
1235
1236         p = get_proc_task(inode);
1237         if (!p)
1238                 return -ESRCH;
1239         proc_sched_show_task(p, m);
1240
1241         put_task_struct(p);
1242
1243         return 0;
1244 }
1245
1246 static ssize_t
1247 sched_write(struct file *file, const char __user *buf,
1248             size_t count, loff_t *offset)
1249 {
1250         struct inode *inode = file_inode(file);
1251         struct task_struct *p;
1252
1253         p = get_proc_task(inode);
1254         if (!p)
1255                 return -ESRCH;
1256         proc_sched_set_task(p);
1257
1258         put_task_struct(p);
1259
1260         return count;
1261 }
1262
1263 static int sched_open(struct inode *inode, struct file *filp)
1264 {
1265         return single_open(filp, sched_show, inode);
1266 }
1267
1268 static const struct file_operations proc_pid_sched_operations = {
1269         .open           = sched_open,
1270         .read           = seq_read,
1271         .write          = sched_write,
1272         .llseek         = seq_lseek,
1273         .release        = single_release,
1274 };
1275
1276 #endif
1277
1278 #ifdef CONFIG_SCHED_AUTOGROUP
1279 /*
1280  * Print out autogroup related information:
1281  */
1282 static int sched_autogroup_show(struct seq_file *m, void *v)
1283 {
1284         struct inode *inode = m->private;
1285         struct task_struct *p;
1286
1287         p = get_proc_task(inode);
1288         if (!p)
1289                 return -ESRCH;
1290         proc_sched_autogroup_show_task(p, m);
1291
1292         put_task_struct(p);
1293
1294         return 0;
1295 }
1296
1297 static ssize_t
1298 sched_autogroup_write(struct file *file, const char __user *buf,
1299             size_t count, loff_t *offset)
1300 {
1301         struct inode *inode = file_inode(file);
1302         struct task_struct *p;
1303         char buffer[PROC_NUMBUF];
1304         int nice;
1305         int err;
1306
1307         memset(buffer, 0, sizeof(buffer));
1308         if (count > sizeof(buffer) - 1)
1309                 count = sizeof(buffer) - 1;
1310         if (copy_from_user(buffer, buf, count))
1311                 return -EFAULT;
1312
1313         err = kstrtoint(strstrip(buffer), 0, &nice);
1314         if (err < 0)
1315                 return err;
1316
1317         p = get_proc_task(inode);
1318         if (!p)
1319                 return -ESRCH;
1320
1321         err = proc_sched_autogroup_set_nice(p, nice);
1322         if (err)
1323                 count = err;
1324
1325         put_task_struct(p);
1326
1327         return count;
1328 }
1329
1330 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1331 {
1332         int ret;
1333
1334         ret = single_open(filp, sched_autogroup_show, NULL);
1335         if (!ret) {
1336                 struct seq_file *m = filp->private_data;
1337
1338                 m->private = inode;
1339         }
1340         return ret;
1341 }
1342
1343 static const struct file_operations proc_pid_sched_autogroup_operations = {
1344         .open           = sched_autogroup_open,
1345         .read           = seq_read,
1346         .write          = sched_autogroup_write,
1347         .llseek         = seq_lseek,
1348         .release        = single_release,
1349 };
1350
1351 #endif /* CONFIG_SCHED_AUTOGROUP */
1352
1353 static ssize_t comm_write(struct file *file, const char __user *buf,
1354                                 size_t count, loff_t *offset)
1355 {
1356         struct inode *inode = file_inode(file);
1357         struct task_struct *p;
1358         char buffer[TASK_COMM_LEN];
1359         const size_t maxlen = sizeof(buffer) - 1;
1360
1361         memset(buffer, 0, sizeof(buffer));
1362         if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1363                 return -EFAULT;
1364
1365         p = get_proc_task(inode);
1366         if (!p)
1367                 return -ESRCH;
1368
1369         if (same_thread_group(current, p))
1370                 set_task_comm(p, buffer);
1371         else
1372                 count = -EINVAL;
1373
1374         put_task_struct(p);
1375
1376         return count;
1377 }
1378
1379 static int comm_show(struct seq_file *m, void *v)
1380 {
1381         struct inode *inode = m->private;
1382         struct task_struct *p;
1383
1384         p = get_proc_task(inode);
1385         if (!p)
1386                 return -ESRCH;
1387
1388         task_lock(p);
1389         seq_printf(m, "%s\n", p->comm);
1390         task_unlock(p);
1391
1392         put_task_struct(p);
1393
1394         return 0;
1395 }
1396
1397 static int comm_open(struct inode *inode, struct file *filp)
1398 {
1399         return single_open(filp, comm_show, inode);
1400 }
1401
1402 static const struct file_operations proc_pid_set_comm_operations = {
1403         .open           = comm_open,
1404         .read           = seq_read,
1405         .write          = comm_write,
1406         .llseek         = seq_lseek,
1407         .release        = single_release,
1408 };
1409
1410 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1411 {
1412         struct task_struct *task;
1413         struct mm_struct *mm;
1414         struct file *exe_file;
1415
1416         task = get_proc_task(dentry->d_inode);
1417         if (!task)
1418                 return -ENOENT;
1419         mm = get_task_mm(task);
1420         put_task_struct(task);
1421         if (!mm)
1422                 return -ENOENT;
1423         exe_file = get_mm_exe_file(mm);
1424         mmput(mm);
1425         if (exe_file) {
1426                 *exe_path = exe_file->f_path;
1427                 path_get(&exe_file->f_path);
1428                 fput(exe_file);
1429                 return 0;
1430         } else
1431                 return -ENOENT;
1432 }
1433
1434 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1435 {
1436         struct inode *inode = dentry->d_inode;
1437         struct path path;
1438         int error = -EACCES;
1439
1440         /* Are we allowed to snoop on the tasks file descriptors? */
1441         if (!proc_fd_access_allowed(inode))
1442                 goto out;
1443
1444         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1445         if (error)
1446                 goto out;
1447
1448         nd_jump_link(nd, &path);
1449         return NULL;
1450 out:
1451         return ERR_PTR(error);
1452 }
1453
1454 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1455 {
1456         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1457         char *pathname;
1458         int len;
1459
1460         if (!tmp)
1461                 return -ENOMEM;
1462
1463         pathname = d_path(path, tmp, PAGE_SIZE);
1464         len = PTR_ERR(pathname);
1465         if (IS_ERR(pathname))
1466                 goto out;
1467         len = tmp + PAGE_SIZE - 1 - pathname;
1468
1469         if (len > buflen)
1470                 len = buflen;
1471         if (copy_to_user(buffer, pathname, len))
1472                 len = -EFAULT;
1473  out:
1474         free_page((unsigned long)tmp);
1475         return len;
1476 }
1477
1478 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1479 {
1480         int error = -EACCES;
1481         struct inode *inode = dentry->d_inode;
1482         struct path path;
1483
1484         /* Are we allowed to snoop on the tasks file descriptors? */
1485         if (!proc_fd_access_allowed(inode))
1486                 goto out;
1487
1488         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1489         if (error)
1490                 goto out;
1491
1492         error = do_proc_readlink(&path, buffer, buflen);
1493         path_put(&path);
1494 out:
1495         return error;
1496 }
1497
1498 const struct inode_operations proc_pid_link_inode_operations = {
1499         .readlink       = proc_pid_readlink,
1500         .follow_link    = proc_pid_follow_link,
1501         .setattr        = proc_setattr,
1502 };
1503
1504
1505 /* building an inode */
1506
1507 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1508 {
1509         struct inode * inode;
1510         struct proc_inode *ei;
1511         const struct cred *cred;
1512
1513         /* We need a new inode */
1514
1515         inode = new_inode(sb);
1516         if (!inode)
1517                 goto out;
1518
1519         /* Common stuff */
1520         ei = PROC_I(inode);
1521         inode->i_ino = get_next_ino();
1522         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1523         inode->i_op = &proc_def_inode_operations;
1524
1525         /*
1526          * grab the reference to task.
1527          */
1528         ei->pid = get_task_pid(task, PIDTYPE_PID);
1529         if (!ei->pid)
1530                 goto out_unlock;
1531
1532         if (task_dumpable(task)) {
1533                 rcu_read_lock();
1534                 cred = __task_cred(task);
1535                 inode->i_uid = cred->euid;
1536                 inode->i_gid = cred->egid;
1537                 rcu_read_unlock();
1538         }
1539         security_task_to_inode(task, inode);
1540
1541 out:
1542         return inode;
1543
1544 out_unlock:
1545         iput(inode);
1546         return NULL;
1547 }
1548
1549 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1550 {
1551         struct inode *inode = dentry->d_inode;
1552         struct task_struct *task;
1553         const struct cred *cred;
1554         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1555
1556         generic_fillattr(inode, stat);
1557
1558         rcu_read_lock();
1559         stat->uid = GLOBAL_ROOT_UID;
1560         stat->gid = GLOBAL_ROOT_GID;
1561         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1562         if (task) {
1563                 if (!has_pid_permissions(pid, task, 2)) {
1564                         rcu_read_unlock();
1565                         /*
1566                          * This doesn't prevent learning whether PID exists,
1567                          * it only makes getattr() consistent with readdir().
1568                          */
1569                         return -ENOENT;
1570                 }
1571                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1572                     task_dumpable(task)) {
1573                         cred = __task_cred(task);
1574                         stat->uid = cred->euid;
1575                         stat->gid = cred->egid;
1576                 }
1577         }
1578         rcu_read_unlock();
1579         return 0;
1580 }
1581
1582 /* dentry stuff */
1583
1584 /*
1585  *      Exceptional case: normally we are not allowed to unhash a busy
1586  * directory. In this case, however, we can do it - no aliasing problems
1587  * due to the way we treat inodes.
1588  *
1589  * Rewrite the inode's ownerships here because the owning task may have
1590  * performed a setuid(), etc.
1591  *
1592  * Before the /proc/pid/status file was created the only way to read
1593  * the effective uid of a /process was to stat /proc/pid.  Reading
1594  * /proc/pid/status is slow enough that procps and other packages
1595  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1596  * made this apply to all per process world readable and executable
1597  * directories.
1598  */
1599 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1600 {
1601         struct inode *inode;
1602         struct task_struct *task;
1603         const struct cred *cred;
1604
1605         if (flags & LOOKUP_RCU)
1606                 return -ECHILD;
1607
1608         inode = dentry->d_inode;
1609         task = get_proc_task(inode);
1610
1611         if (task) {
1612                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1613                     task_dumpable(task)) {
1614                         rcu_read_lock();
1615                         cred = __task_cred(task);
1616                         inode->i_uid = cred->euid;
1617                         inode->i_gid = cred->egid;
1618                         rcu_read_unlock();
1619                 } else {
1620                         inode->i_uid = GLOBAL_ROOT_UID;
1621                         inode->i_gid = GLOBAL_ROOT_GID;
1622                 }
1623                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1624                 security_task_to_inode(task, inode);
1625                 put_task_struct(task);
1626                 return 1;
1627         }
1628         d_drop(dentry);
1629         return 0;
1630 }
1631
1632 static inline bool proc_inode_is_dead(struct inode *inode)
1633 {
1634         return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1635 }
1636
1637 int pid_delete_dentry(const struct dentry *dentry)
1638 {
1639         /* Is the task we represent dead?
1640          * If so, then don't put the dentry on the lru list,
1641          * kill it immediately.
1642          */
1643         return proc_inode_is_dead(dentry->d_inode);
1644 }
1645
1646 const struct dentry_operations pid_dentry_operations =
1647 {
1648         .d_revalidate   = pid_revalidate,
1649         .d_delete       = pid_delete_dentry,
1650 };
1651
1652 /* Lookups */
1653
1654 /*
1655  * Fill a directory entry.
1656  *
1657  * If possible create the dcache entry and derive our inode number and
1658  * file type from dcache entry.
1659  *
1660  * Since all of the proc inode numbers are dynamically generated, the inode
1661  * numbers do not exist until the inode is cache.  This means creating the
1662  * the dcache entry in readdir is necessary to keep the inode numbers
1663  * reported by readdir in sync with the inode numbers reported
1664  * by stat.
1665  */
1666 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1667         const char *name, int len,
1668         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1669 {
1670         struct dentry *child, *dir = file->f_path.dentry;
1671         struct qstr qname = QSTR_INIT(name, len);
1672         struct inode *inode;
1673         unsigned type;
1674         ino_t ino;
1675
1676         child = d_hash_and_lookup(dir, &qname);
1677         if (!child) {
1678                 child = d_alloc(dir, &qname);
1679                 if (!child)
1680                         goto end_instantiate;
1681                 if (instantiate(dir->d_inode, child, task, ptr) < 0) {
1682                         dput(child);
1683                         goto end_instantiate;
1684                 }
1685         }
1686         inode = child->d_inode;
1687         ino = inode->i_ino;
1688         type = inode->i_mode >> 12;
1689         dput(child);
1690         return dir_emit(ctx, name, len, ino, type);
1691
1692 end_instantiate:
1693         return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1694 }
1695
1696 #ifdef CONFIG_CHECKPOINT_RESTORE
1697
1698 /*
1699  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1700  * which represent vma start and end addresses.
1701  */
1702 static int dname_to_vma_addr(struct dentry *dentry,
1703                              unsigned long *start, unsigned long *end)
1704 {
1705         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1706                 return -EINVAL;
1707
1708         return 0;
1709 }
1710
1711 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1712 {
1713         unsigned long vm_start, vm_end;
1714         bool exact_vma_exists = false;
1715         struct mm_struct *mm = NULL;
1716         struct task_struct *task;
1717         const struct cred *cred;
1718         struct inode *inode;
1719         int status = 0;
1720
1721         if (flags & LOOKUP_RCU)
1722                 return -ECHILD;
1723
1724         if (!capable(CAP_SYS_ADMIN)) {
1725                 status = -EPERM;
1726                 goto out_notask;
1727         }
1728
1729         inode = dentry->d_inode;
1730         task = get_proc_task(inode);
1731         if (!task)
1732                 goto out_notask;
1733
1734         mm = mm_access(task, PTRACE_MODE_READ);
1735         if (IS_ERR_OR_NULL(mm))
1736                 goto out;
1737
1738         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1739                 down_read(&mm->mmap_sem);
1740                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1741                 up_read(&mm->mmap_sem);
1742         }
1743
1744         mmput(mm);
1745
1746         if (exact_vma_exists) {
1747                 if (task_dumpable(task)) {
1748                         rcu_read_lock();
1749                         cred = __task_cred(task);
1750                         inode->i_uid = cred->euid;
1751                         inode->i_gid = cred->egid;
1752                         rcu_read_unlock();
1753                 } else {
1754                         inode->i_uid = GLOBAL_ROOT_UID;
1755                         inode->i_gid = GLOBAL_ROOT_GID;
1756                 }
1757                 security_task_to_inode(task, inode);
1758                 status = 1;
1759         }
1760
1761 out:
1762         put_task_struct(task);
1763
1764 out_notask:
1765         if (status <= 0)
1766                 d_drop(dentry);
1767
1768         return status;
1769 }
1770
1771 static const struct dentry_operations tid_map_files_dentry_operations = {
1772         .d_revalidate   = map_files_d_revalidate,
1773         .d_delete       = pid_delete_dentry,
1774 };
1775
1776 static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
1777 {
1778         unsigned long vm_start, vm_end;
1779         struct vm_area_struct *vma;
1780         struct task_struct *task;
1781         struct mm_struct *mm;
1782         int rc;
1783
1784         rc = -ENOENT;
1785         task = get_proc_task(dentry->d_inode);
1786         if (!task)
1787                 goto out;
1788
1789         mm = get_task_mm(task);
1790         put_task_struct(task);
1791         if (!mm)
1792                 goto out;
1793
1794         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1795         if (rc)
1796                 goto out_mmput;
1797
1798         rc = -ENOENT;
1799         down_read(&mm->mmap_sem);
1800         vma = find_exact_vma(mm, vm_start, vm_end);
1801         if (vma && vma->vm_file) {
1802                 *path = vma->vm_file->f_path;
1803                 path_get(path);
1804                 rc = 0;
1805         }
1806         up_read(&mm->mmap_sem);
1807
1808 out_mmput:
1809         mmput(mm);
1810 out:
1811         return rc;
1812 }
1813
1814 struct map_files_info {
1815         fmode_t         mode;
1816         unsigned long   len;
1817         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1818 };
1819
1820 static int
1821 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
1822                            struct task_struct *task, const void *ptr)
1823 {
1824         fmode_t mode = (fmode_t)(unsigned long)ptr;
1825         struct proc_inode *ei;
1826         struct inode *inode;
1827
1828         inode = proc_pid_make_inode(dir->i_sb, task);
1829         if (!inode)
1830                 return -ENOENT;
1831
1832         ei = PROC_I(inode);
1833         ei->op.proc_get_link = proc_map_files_get_link;
1834
1835         inode->i_op = &proc_pid_link_inode_operations;
1836         inode->i_size = 64;
1837         inode->i_mode = S_IFLNK;
1838
1839         if (mode & FMODE_READ)
1840                 inode->i_mode |= S_IRUSR;
1841         if (mode & FMODE_WRITE)
1842                 inode->i_mode |= S_IWUSR;
1843
1844         d_set_d_op(dentry, &tid_map_files_dentry_operations);
1845         d_add(dentry, inode);
1846
1847         return 0;
1848 }
1849
1850 static struct dentry *proc_map_files_lookup(struct inode *dir,
1851                 struct dentry *dentry, unsigned int flags)
1852 {
1853         unsigned long vm_start, vm_end;
1854         struct vm_area_struct *vma;
1855         struct task_struct *task;
1856         int result;
1857         struct mm_struct *mm;
1858
1859         result = -EPERM;
1860         if (!capable(CAP_SYS_ADMIN))
1861                 goto out;
1862
1863         result = -ENOENT;
1864         task = get_proc_task(dir);
1865         if (!task)
1866                 goto out;
1867
1868         result = -EACCES;
1869         if (!ptrace_may_access(task, PTRACE_MODE_READ))
1870                 goto out_put_task;
1871
1872         result = -ENOENT;
1873         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
1874                 goto out_put_task;
1875
1876         mm = get_task_mm(task);
1877         if (!mm)
1878                 goto out_put_task;
1879
1880         down_read(&mm->mmap_sem);
1881         vma = find_exact_vma(mm, vm_start, vm_end);
1882         if (!vma)
1883                 goto out_no_vma;
1884
1885         if (vma->vm_file)
1886                 result = proc_map_files_instantiate(dir, dentry, task,
1887                                 (void *)(unsigned long)vma->vm_file->f_mode);
1888
1889 out_no_vma:
1890         up_read(&mm->mmap_sem);
1891         mmput(mm);
1892 out_put_task:
1893         put_task_struct(task);
1894 out:
1895         return ERR_PTR(result);
1896 }
1897
1898 static const struct inode_operations proc_map_files_inode_operations = {
1899         .lookup         = proc_map_files_lookup,
1900         .permission     = proc_fd_permission,
1901         .setattr        = proc_setattr,
1902 };
1903
1904 static int
1905 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
1906 {
1907         struct vm_area_struct *vma;
1908         struct task_struct *task;
1909         struct mm_struct *mm;
1910         unsigned long nr_files, pos, i;
1911         struct flex_array *fa = NULL;
1912         struct map_files_info info;
1913         struct map_files_info *p;
1914         int ret;
1915
1916         ret = -EPERM;
1917         if (!capable(CAP_SYS_ADMIN))
1918                 goto out;
1919
1920         ret = -ENOENT;
1921         task = get_proc_task(file_inode(file));
1922         if (!task)
1923                 goto out;
1924
1925         ret = -EACCES;
1926         if (!ptrace_may_access(task, PTRACE_MODE_READ))
1927                 goto out_put_task;
1928
1929         ret = 0;
1930         if (!dir_emit_dots(file, ctx))
1931                 goto out_put_task;
1932
1933         mm = get_task_mm(task);
1934         if (!mm)
1935                 goto out_put_task;
1936         down_read(&mm->mmap_sem);
1937
1938         nr_files = 0;
1939
1940         /*
1941          * We need two passes here:
1942          *
1943          *  1) Collect vmas of mapped files with mmap_sem taken
1944          *  2) Release mmap_sem and instantiate entries
1945          *
1946          * otherwise we get lockdep complained, since filldir()
1947          * routine might require mmap_sem taken in might_fault().
1948          */
1949
1950         for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
1951                 if (vma->vm_file && ++pos > ctx->pos)
1952                         nr_files++;
1953         }
1954
1955         if (nr_files) {
1956                 fa = flex_array_alloc(sizeof(info), nr_files,
1957                                         GFP_KERNEL);
1958                 if (!fa || flex_array_prealloc(fa, 0, nr_files,
1959                                                 GFP_KERNEL)) {
1960                         ret = -ENOMEM;
1961                         if (fa)
1962                                 flex_array_free(fa);
1963                         up_read(&mm->mmap_sem);
1964                         mmput(mm);
1965                         goto out_put_task;
1966                 }
1967                 for (i = 0, vma = mm->mmap, pos = 2; vma;
1968                                 vma = vma->vm_next) {
1969                         if (!vma->vm_file)
1970                                 continue;
1971                         if (++pos <= ctx->pos)
1972                                 continue;
1973
1974                         info.mode = vma->vm_file->f_mode;
1975                         info.len = snprintf(info.name,
1976                                         sizeof(info.name), "%lx-%lx",
1977                                         vma->vm_start, vma->vm_end);
1978                         if (flex_array_put(fa, i++, &info, GFP_KERNEL))
1979                                 BUG();
1980                 }
1981         }
1982         up_read(&mm->mmap_sem);
1983
1984         for (i = 0; i < nr_files; i++) {
1985                 p = flex_array_get(fa, i);
1986                 if (!proc_fill_cache(file, ctx,
1987                                       p->name, p->len,
1988                                       proc_map_files_instantiate,
1989                                       task,
1990                                       (void *)(unsigned long)p->mode))
1991                         break;
1992                 ctx->pos++;
1993         }
1994         if (fa)
1995                 flex_array_free(fa);
1996         mmput(mm);
1997
1998 out_put_task:
1999         put_task_struct(task);
2000 out:
2001         return ret;
2002 }
2003
2004 static const struct file_operations proc_map_files_operations = {
2005         .read           = generic_read_dir,
2006         .iterate        = proc_map_files_readdir,
2007         .llseek         = default_llseek,
2008 };
2009
2010 struct timers_private {
2011         struct pid *pid;
2012         struct task_struct *task;
2013         struct sighand_struct *sighand;
2014         struct pid_namespace *ns;
2015         unsigned long flags;
2016 };
2017
2018 static void *timers_start(struct seq_file *m, loff_t *pos)
2019 {
2020         struct timers_private *tp = m->private;
2021
2022         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2023         if (!tp->task)
2024                 return ERR_PTR(-ESRCH);
2025
2026         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2027         if (!tp->sighand)
2028                 return ERR_PTR(-ESRCH);
2029
2030         return seq_list_start(&tp->task->signal->posix_timers, *pos);
2031 }
2032
2033 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2034 {
2035         struct timers_private *tp = m->private;
2036         return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2037 }
2038
2039 static void timers_stop(struct seq_file *m, void *v)
2040 {
2041         struct timers_private *tp = m->private;
2042
2043         if (tp->sighand) {
2044                 unlock_task_sighand(tp->task, &tp->flags);
2045                 tp->sighand = NULL;
2046         }
2047
2048         if (tp->task) {
2049                 put_task_struct(tp->task);
2050                 tp->task = NULL;
2051         }
2052 }
2053
2054 static int show_timer(struct seq_file *m, void *v)
2055 {
2056         struct k_itimer *timer;
2057         struct timers_private *tp = m->private;
2058         int notify;
2059         static char *nstr[] = {
2060                 [SIGEV_SIGNAL] = "signal",
2061                 [SIGEV_NONE] = "none",
2062                 [SIGEV_THREAD] = "thread",
2063         };
2064
2065         timer = list_entry((struct list_head *)v, struct k_itimer, list);
2066         notify = timer->it_sigev_notify;
2067
2068         seq_printf(m, "ID: %d\n", timer->it_id);
2069         seq_printf(m, "signal: %d/%p\n", timer->sigq->info.si_signo,
2070                         timer->sigq->info.si_value.sival_ptr);
2071         seq_printf(m, "notify: %s/%s.%d\n",
2072                 nstr[notify & ~SIGEV_THREAD_ID],
2073                 (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2074                 pid_nr_ns(timer->it_pid, tp->ns));
2075         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2076
2077         return 0;
2078 }
2079
2080 static const struct seq_operations proc_timers_seq_ops = {
2081         .start  = timers_start,
2082         .next   = timers_next,
2083         .stop   = timers_stop,
2084         .show   = show_timer,
2085 };
2086
2087 static int proc_timers_open(struct inode *inode, struct file *file)
2088 {
2089         struct timers_private *tp;
2090
2091         tp = __seq_open_private(file, &proc_timers_seq_ops,
2092                         sizeof(struct timers_private));
2093         if (!tp)
2094                 return -ENOMEM;
2095
2096         tp->pid = proc_pid(inode);
2097         tp->ns = inode->i_sb->s_fs_info;
2098         return 0;
2099 }
2100
2101 static const struct file_operations proc_timers_operations = {
2102         .open           = proc_timers_open,
2103         .read           = seq_read,
2104         .llseek         = seq_lseek,
2105         .release        = seq_release_private,
2106 };
2107 #endif /* CONFIG_CHECKPOINT_RESTORE */
2108
2109 static int proc_pident_instantiate(struct inode *dir,
2110         struct dentry *dentry, struct task_struct *task, const void *ptr)
2111 {
2112         const struct pid_entry *p = ptr;
2113         struct inode *inode;
2114         struct proc_inode *ei;
2115
2116         inode = proc_pid_make_inode(dir->i_sb, task);
2117         if (!inode)
2118                 goto out;
2119
2120         ei = PROC_I(inode);
2121         inode->i_mode = p->mode;
2122         if (S_ISDIR(inode->i_mode))
2123                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2124         if (p->iop)
2125                 inode->i_op = p->iop;
2126         if (p->fop)
2127                 inode->i_fop = p->fop;
2128         ei->op = p->op;
2129         d_set_d_op(dentry, &pid_dentry_operations);
2130         d_add(dentry, inode);
2131         /* Close the race of the process dying before we return the dentry */
2132         if (pid_revalidate(dentry, 0))
2133                 return 0;
2134 out:
2135         return -ENOENT;
2136 }
2137
2138 static struct dentry *proc_pident_lookup(struct inode *dir,
2139                                          struct dentry *dentry,
2140                                          const struct pid_entry *ents,
2141                                          unsigned int nents)
2142 {
2143         int error;
2144         struct task_struct *task = get_proc_task(dir);
2145         const struct pid_entry *p, *last;
2146
2147         error = -ENOENT;
2148
2149         if (!task)
2150                 goto out_no_task;
2151
2152         /*
2153          * Yes, it does not scale. And it should not. Don't add
2154          * new entries into /proc/<tgid>/ without very good reasons.
2155          */
2156         last = &ents[nents - 1];
2157         for (p = ents; p <= last; p++) {
2158                 if (p->len != dentry->d_name.len)
2159                         continue;
2160                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2161                         break;
2162         }
2163         if (p > last)
2164                 goto out;
2165
2166         error = proc_pident_instantiate(dir, dentry, task, p);
2167 out:
2168         put_task_struct(task);
2169 out_no_task:
2170         return ERR_PTR(error);
2171 }
2172
2173 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2174                 const struct pid_entry *ents, unsigned int nents)
2175 {
2176         struct task_struct *task = get_proc_task(file_inode(file));
2177         const struct pid_entry *p;
2178
2179         if (!task)
2180                 return -ENOENT;
2181
2182         if (!dir_emit_dots(file, ctx))
2183                 goto out;
2184
2185         if (ctx->pos >= nents + 2)
2186                 goto out;
2187
2188         for (p = ents + (ctx->pos - 2); p <= ents + nents - 1; p++) {
2189                 if (!proc_fill_cache(file, ctx, p->name, p->len,
2190                                 proc_pident_instantiate, task, p))
2191                         break;
2192                 ctx->pos++;
2193         }
2194 out:
2195         put_task_struct(task);
2196         return 0;
2197 }
2198
2199 #ifdef CONFIG_SECURITY
2200 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2201                                   size_t count, loff_t *ppos)
2202 {
2203         struct inode * inode = file_inode(file);
2204         char *p = NULL;
2205         ssize_t length;
2206         struct task_struct *task = get_proc_task(inode);
2207
2208         if (!task)
2209                 return -ESRCH;
2210
2211         length = security_getprocattr(task,
2212                                       (char*)file->f_path.dentry->d_name.name,
2213                                       &p);
2214         put_task_struct(task);
2215         if (length > 0)
2216                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2217         kfree(p);
2218         return length;
2219 }
2220
2221 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2222                                    size_t count, loff_t *ppos)
2223 {
2224         struct inode * inode = file_inode(file);
2225         char *page;
2226         ssize_t length;
2227         struct task_struct *task = get_proc_task(inode);
2228
2229         length = -ESRCH;
2230         if (!task)
2231                 goto out_no_task;
2232         if (count > PAGE_SIZE)
2233                 count = PAGE_SIZE;
2234
2235         /* No partial writes. */
2236         length = -EINVAL;
2237         if (*ppos != 0)
2238                 goto out;
2239
2240         length = -ENOMEM;
2241         page = (char*)__get_free_page(GFP_TEMPORARY);
2242         if (!page)
2243                 goto out;
2244
2245         length = -EFAULT;
2246         if (copy_from_user(page, buf, count))
2247                 goto out_free;
2248
2249         /* Guard against adverse ptrace interaction */
2250         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2251         if (length < 0)
2252                 goto out_free;
2253
2254         length = security_setprocattr(task,
2255                                       (char*)file->f_path.dentry->d_name.name,
2256                                       (void*)page, count);
2257         mutex_unlock(&task->signal->cred_guard_mutex);
2258 out_free:
2259         free_page((unsigned long) page);
2260 out:
2261         put_task_struct(task);
2262 out_no_task:
2263         return length;
2264 }
2265
2266 static const struct file_operations proc_pid_attr_operations = {
2267         .read           = proc_pid_attr_read,
2268         .write          = proc_pid_attr_write,
2269         .llseek         = generic_file_llseek,
2270 };
2271
2272 static const struct pid_entry attr_dir_stuff[] = {
2273         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2274         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2275         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2276         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2277         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2278         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2279 };
2280
2281 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2282 {
2283         return proc_pident_readdir(file, ctx,
2284                                    attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2285 }
2286
2287 static const struct file_operations proc_attr_dir_operations = {
2288         .read           = generic_read_dir,
2289         .iterate        = proc_attr_dir_readdir,
2290         .llseek         = default_llseek,
2291 };
2292
2293 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2294                                 struct dentry *dentry, unsigned int flags)
2295 {
2296         return proc_pident_lookup(dir, dentry,
2297                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2298 }
2299
2300 static const struct inode_operations proc_attr_dir_inode_operations = {
2301         .lookup         = proc_attr_dir_lookup,
2302         .getattr        = pid_getattr,
2303         .setattr        = proc_setattr,
2304 };
2305
2306 #endif
2307
2308 #ifdef CONFIG_ELF_CORE
2309 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2310                                          size_t count, loff_t *ppos)
2311 {
2312         struct task_struct *task = get_proc_task(file_inode(file));
2313         struct mm_struct *mm;
2314         char buffer[PROC_NUMBUF];
2315         size_t len;
2316         int ret;
2317
2318         if (!task)
2319                 return -ESRCH;
2320
2321         ret = 0;
2322         mm = get_task_mm(task);
2323         if (mm) {
2324                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2325                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2326                                 MMF_DUMP_FILTER_SHIFT));
2327                 mmput(mm);
2328                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2329         }
2330
2331         put_task_struct(task);
2332
2333         return ret;
2334 }
2335
2336 static ssize_t proc_coredump_filter_write(struct file *file,
2337                                           const char __user *buf,
2338                                           size_t count,
2339                                           loff_t *ppos)
2340 {
2341         struct task_struct *task;
2342         struct mm_struct *mm;
2343         char buffer[PROC_NUMBUF], *end;
2344         unsigned int val;
2345         int ret;
2346         int i;
2347         unsigned long mask;
2348
2349         ret = -EFAULT;
2350         memset(buffer, 0, sizeof(buffer));
2351         if (count > sizeof(buffer) - 1)
2352                 count = sizeof(buffer) - 1;
2353         if (copy_from_user(buffer, buf, count))
2354                 goto out_no_task;
2355
2356         ret = -EINVAL;
2357         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2358         if (*end == '\n')
2359                 end++;
2360         if (end - buffer == 0)
2361                 goto out_no_task;
2362
2363         ret = -ESRCH;
2364         task = get_proc_task(file_inode(file));
2365         if (!task)
2366                 goto out_no_task;
2367
2368         ret = end - buffer;
2369         mm = get_task_mm(task);
2370         if (!mm)
2371                 goto out_no_mm;
2372
2373         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2374                 if (val & mask)
2375                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2376                 else
2377                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2378         }
2379
2380         mmput(mm);
2381  out_no_mm:
2382         put_task_struct(task);
2383  out_no_task:
2384         return ret;
2385 }
2386
2387 static const struct file_operations proc_coredump_filter_operations = {
2388         .read           = proc_coredump_filter_read,
2389         .write          = proc_coredump_filter_write,
2390         .llseek         = generic_file_llseek,
2391 };
2392 #endif
2393
2394 #ifdef CONFIG_TASK_IO_ACCOUNTING
2395 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2396 {
2397         struct task_io_accounting acct = task->ioac;
2398         unsigned long flags;
2399         int result;
2400
2401         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2402         if (result)
2403                 return result;
2404
2405         if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2406                 result = -EACCES;
2407                 goto out_unlock;
2408         }
2409
2410         if (whole && lock_task_sighand(task, &flags)) {
2411                 struct task_struct *t = task;
2412
2413                 task_io_accounting_add(&acct, &task->signal->ioac);
2414                 while_each_thread(task, t)
2415                         task_io_accounting_add(&acct, &t->ioac);
2416
2417                 unlock_task_sighand(task, &flags);
2418         }
2419         result = sprintf(buffer,
2420                         "rchar: %llu\n"
2421                         "wchar: %llu\n"
2422                         "syscr: %llu\n"
2423                         "syscw: %llu\n"
2424                         "read_bytes: %llu\n"
2425                         "write_bytes: %llu\n"
2426                         "cancelled_write_bytes: %llu\n",
2427                         (unsigned long long)acct.rchar,
2428                         (unsigned long long)acct.wchar,
2429                         (unsigned long long)acct.syscr,
2430                         (unsigned long long)acct.syscw,
2431                         (unsigned long long)acct.read_bytes,
2432                         (unsigned long long)acct.write_bytes,
2433                         (unsigned long long)acct.cancelled_write_bytes);
2434 out_unlock:
2435         mutex_unlock(&task->signal->cred_guard_mutex);
2436         return result;
2437 }
2438
2439 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2440 {
2441         return do_io_accounting(task, buffer, 0);
2442 }
2443
2444 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2445 {
2446         return do_io_accounting(task, buffer, 1);
2447 }
2448 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2449
2450 #ifdef CONFIG_USER_NS
2451 static int proc_id_map_open(struct inode *inode, struct file *file,
2452         struct seq_operations *seq_ops)
2453 {
2454         struct user_namespace *ns = NULL;
2455         struct task_struct *task;
2456         struct seq_file *seq;
2457         int ret = -EINVAL;
2458
2459         task = get_proc_task(inode);
2460         if (task) {
2461                 rcu_read_lock();
2462                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2463                 rcu_read_unlock();
2464                 put_task_struct(task);
2465         }
2466         if (!ns)
2467                 goto err;
2468
2469         ret = seq_open(file, seq_ops);
2470         if (ret)
2471                 goto err_put_ns;
2472
2473         seq = file->private_data;
2474         seq->private = ns;
2475
2476         return 0;
2477 err_put_ns:
2478         put_user_ns(ns);
2479 err:
2480         return ret;
2481 }
2482
2483 static int proc_id_map_release(struct inode *inode, struct file *file)
2484 {
2485         struct seq_file *seq = file->private_data;
2486         struct user_namespace *ns = seq->private;
2487         put_user_ns(ns);
2488         return seq_release(inode, file);
2489 }
2490
2491 static int proc_uid_map_open(struct inode *inode, struct file *file)
2492 {
2493         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2494 }
2495
2496 static int proc_gid_map_open(struct inode *inode, struct file *file)
2497 {
2498         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2499 }
2500
2501 static int proc_projid_map_open(struct inode *inode, struct file *file)
2502 {
2503         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2504 }
2505
2506 static const struct file_operations proc_uid_map_operations = {
2507         .open           = proc_uid_map_open,
2508         .write          = proc_uid_map_write,
2509         .read           = seq_read,
2510         .llseek         = seq_lseek,
2511         .release        = proc_id_map_release,
2512 };
2513
2514 static const struct file_operations proc_gid_map_operations = {
2515         .open           = proc_gid_map_open,
2516         .write          = proc_gid_map_write,
2517         .read           = seq_read,
2518         .llseek         = seq_lseek,
2519         .release        = proc_id_map_release,
2520 };
2521
2522 static const struct file_operations proc_projid_map_operations = {
2523         .open           = proc_projid_map_open,
2524         .write          = proc_projid_map_write,
2525         .read           = seq_read,
2526         .llseek         = seq_lseek,
2527         .release        = proc_id_map_release,
2528 };
2529 #endif /* CONFIG_USER_NS */
2530
2531 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2532                                 struct pid *pid, struct task_struct *task)
2533 {
2534         int err = lock_trace(task);
2535         if (!err) {
2536                 seq_printf(m, "%08x\n", task->personality);
2537                 unlock_trace(task);
2538         }
2539         return err;
2540 }
2541
2542 /*
2543  * Thread groups
2544  */
2545 static const struct file_operations proc_task_operations;
2546 static const struct inode_operations proc_task_inode_operations;
2547
2548 static const struct pid_entry tgid_base_stuff[] = {
2549         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2550         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2551 #ifdef CONFIG_CHECKPOINT_RESTORE
2552         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2553 #endif
2554         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2555         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2556 #ifdef CONFIG_NET
2557         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2558 #endif
2559         REG("environ",    S_IRUSR, proc_environ_operations),
2560         INF("auxv",       S_IRUSR, proc_pid_auxv),
2561         ONE("status",     S_IRUGO, proc_pid_status),
2562         ONE("personality", S_IRUSR, proc_pid_personality),
2563         INF("limits",     S_IRUGO, proc_pid_limits),
2564 #ifdef CONFIG_SCHED_DEBUG
2565         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2566 #endif
2567 #ifdef CONFIG_SCHED_AUTOGROUP
2568         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2569 #endif
2570         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2571 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2572         INF("syscall",    S_IRUSR, proc_pid_syscall),
2573 #endif
2574         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2575         ONE("stat",       S_IRUGO, proc_tgid_stat),
2576         ONE("statm",      S_IRUGO, proc_pid_statm),
2577         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2578 #ifdef CONFIG_NUMA
2579         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2580 #endif
2581         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2582         LNK("cwd",        proc_cwd_link),
2583         LNK("root",       proc_root_link),
2584         LNK("exe",        proc_exe_link),
2585         REG("mounts",     S_IRUGO, proc_mounts_operations),
2586         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2587         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2588 #ifdef CONFIG_PROC_PAGE_MONITOR
2589         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2590         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2591         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2592 #endif
2593 #ifdef CONFIG_SECURITY
2594         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2595 #endif
2596 #ifdef CONFIG_KALLSYMS
2597         INF("wchan",      S_IRUGO, proc_pid_wchan),
2598 #endif
2599 #ifdef CONFIG_STACKTRACE
2600         ONE("stack",      S_IRUSR, proc_pid_stack),
2601 #endif
2602 #ifdef CONFIG_SCHEDSTATS
2603         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
2604 #endif
2605 #ifdef CONFIG_LATENCYTOP
2606         REG("latency",  S_IRUGO, proc_lstats_operations),
2607 #endif
2608 #ifdef CONFIG_PROC_PID_CPUSET
2609         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2610 #endif
2611 #ifdef CONFIG_CGROUPS
2612         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2613 #endif
2614         INF("oom_score",  S_IRUGO, proc_oom_score),
2615         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2616         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2617 #ifdef CONFIG_AUDITSYSCALL
2618         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2619         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2620 #endif
2621 #ifdef CONFIG_FAULT_INJECTION
2622         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2623 #endif
2624 #ifdef CONFIG_ELF_CORE
2625         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2626 #endif
2627 #ifdef CONFIG_TASK_IO_ACCOUNTING
2628         INF("io",       S_IRUSR, proc_tgid_io_accounting),
2629 #endif
2630 #ifdef CONFIG_HARDWALL
2631         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
2632 #endif
2633 #ifdef CONFIG_USER_NS
2634         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2635         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2636         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2637 #endif
2638 #ifdef CONFIG_CHECKPOINT_RESTORE
2639         REG("timers",     S_IRUGO, proc_timers_operations),
2640 #endif
2641 };
2642
2643 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2644 {
2645         return proc_pident_readdir(file, ctx,
2646                                    tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2647 }
2648
2649 static const struct file_operations proc_tgid_base_operations = {
2650         .read           = generic_read_dir,
2651         .iterate        = proc_tgid_base_readdir,
2652         .llseek         = default_llseek,
2653 };
2654
2655 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2656 {
2657         return proc_pident_lookup(dir, dentry,
2658                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2659 }
2660
2661 static const struct inode_operations proc_tgid_base_inode_operations = {
2662         .lookup         = proc_tgid_base_lookup,
2663         .getattr        = pid_getattr,
2664         .setattr        = proc_setattr,
2665         .permission     = proc_pid_permission,
2666 };
2667
2668 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2669 {
2670         struct dentry *dentry, *leader, *dir;
2671         char buf[PROC_NUMBUF];
2672         struct qstr name;
2673
2674         name.name = buf;
2675         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2676         /* no ->d_hash() rejects on procfs */
2677         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2678         if (dentry) {
2679                 shrink_dcache_parent(dentry);
2680                 d_drop(dentry);
2681                 dput(dentry);
2682         }
2683
2684         name.name = buf;
2685         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2686         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2687         if (!leader)
2688                 goto out;
2689
2690         name.name = "task";
2691         name.len = strlen(name.name);
2692         dir = d_hash_and_lookup(leader, &name);
2693         if (!dir)
2694                 goto out_put_leader;
2695
2696         name.name = buf;
2697         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2698         dentry = d_hash_and_lookup(dir, &name);
2699         if (dentry) {
2700                 shrink_dcache_parent(dentry);
2701                 d_drop(dentry);
2702                 dput(dentry);
2703         }
2704
2705         dput(dir);
2706 out_put_leader:
2707         dput(leader);
2708 out:
2709         return;
2710 }
2711
2712 /**
2713  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2714  * @task: task that should be flushed.
2715  *
2716  * When flushing dentries from proc, one needs to flush them from global
2717  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2718  * in. This call is supposed to do all of this job.
2719  *
2720  * Looks in the dcache for
2721  * /proc/@pid
2722  * /proc/@tgid/task/@pid
2723  * if either directory is present flushes it and all of it'ts children
2724  * from the dcache.
2725  *
2726  * It is safe and reasonable to cache /proc entries for a task until
2727  * that task exits.  After that they just clog up the dcache with
2728  * useless entries, possibly causing useful dcache entries to be
2729  * flushed instead.  This routine is proved to flush those useless
2730  * dcache entries at process exit time.
2731  *
2732  * NOTE: This routine is just an optimization so it does not guarantee
2733  *       that no dcache entries will exist at process exit time it
2734  *       just makes it very unlikely that any will persist.
2735  */
2736
2737 void proc_flush_task(struct task_struct *task)
2738 {
2739         int i;
2740         struct pid *pid, *tgid;
2741         struct upid *upid;
2742
2743         pid = task_pid(task);
2744         tgid = task_tgid(task);
2745
2746         for (i = 0; i <= pid->level; i++) {
2747                 upid = &pid->numbers[i];
2748                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2749                                         tgid->numbers[i].nr);
2750         }
2751 }
2752
2753 static int proc_pid_instantiate(struct inode *dir,
2754                                    struct dentry * dentry,
2755                                    struct task_struct *task, const void *ptr)
2756 {
2757         struct inode *inode;
2758
2759         inode = proc_pid_make_inode(dir->i_sb, task);
2760         if (!inode)
2761                 goto out;
2762
2763         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2764         inode->i_op = &proc_tgid_base_inode_operations;
2765         inode->i_fop = &proc_tgid_base_operations;
2766         inode->i_flags|=S_IMMUTABLE;
2767
2768         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
2769                                                   ARRAY_SIZE(tgid_base_stuff)));
2770
2771         d_set_d_op(dentry, &pid_dentry_operations);
2772
2773         d_add(dentry, inode);
2774         /* Close the race of the process dying before we return the dentry */
2775         if (pid_revalidate(dentry, 0))
2776                 return 0;
2777 out:
2778         return -ENOENT;
2779 }
2780
2781 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
2782 {
2783         int result = 0;
2784         struct task_struct *task;
2785         unsigned tgid;
2786         struct pid_namespace *ns;
2787
2788         tgid = name_to_int(dentry);
2789         if (tgid == ~0U)
2790                 goto out;
2791
2792         ns = dentry->d_sb->s_fs_info;
2793         rcu_read_lock();
2794         task = find_task_by_pid_ns(tgid, ns);
2795         if (task)
2796                 get_task_struct(task);
2797         rcu_read_unlock();
2798         if (!task)
2799                 goto out;
2800
2801         result = proc_pid_instantiate(dir, dentry, task, NULL);
2802         put_task_struct(task);
2803 out:
2804         return ERR_PTR(result);
2805 }
2806
2807 /*
2808  * Find the first task with tgid >= tgid
2809  *
2810  */
2811 struct tgid_iter {
2812         unsigned int tgid;
2813         struct task_struct *task;
2814 };
2815 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2816 {
2817         struct pid *pid;
2818
2819         if (iter.task)
2820                 put_task_struct(iter.task);
2821         rcu_read_lock();
2822 retry:
2823         iter.task = NULL;
2824         pid = find_ge_pid(iter.tgid, ns);
2825         if (pid) {
2826                 iter.tgid = pid_nr_ns(pid, ns);
2827                 iter.task = pid_task(pid, PIDTYPE_PID);
2828                 /* What we to know is if the pid we have find is the
2829                  * pid of a thread_group_leader.  Testing for task
2830                  * being a thread_group_leader is the obvious thing
2831                  * todo but there is a window when it fails, due to
2832                  * the pid transfer logic in de_thread.
2833                  *
2834                  * So we perform the straight forward test of seeing
2835                  * if the pid we have found is the pid of a thread
2836                  * group leader, and don't worry if the task we have
2837                  * found doesn't happen to be a thread group leader.
2838                  * As we don't care in the case of readdir.
2839                  */
2840                 if (!iter.task || !has_group_leader_pid(iter.task)) {
2841                         iter.tgid += 1;
2842                         goto retry;
2843                 }
2844                 get_task_struct(iter.task);
2845         }
2846         rcu_read_unlock();
2847         return iter;
2848 }
2849
2850 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 1)
2851
2852 /* for the /proc/ directory itself, after non-process stuff has been done */
2853 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
2854 {
2855         struct tgid_iter iter;
2856         struct pid_namespace *ns = file->f_dentry->d_sb->s_fs_info;
2857         loff_t pos = ctx->pos;
2858
2859         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
2860                 return 0;
2861
2862         if (pos == TGID_OFFSET - 1) {
2863                 struct inode *inode = ns->proc_self->d_inode;
2864                 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
2865                         return 0;
2866                 iter.tgid = 0;
2867         } else {
2868                 iter.tgid = pos - TGID_OFFSET;
2869         }
2870         iter.task = NULL;
2871         for (iter = next_tgid(ns, iter);
2872              iter.task;
2873              iter.tgid += 1, iter = next_tgid(ns, iter)) {
2874                 char name[PROC_NUMBUF];
2875                 int len;
2876                 if (!has_pid_permissions(ns, iter.task, 2))
2877                         continue;
2878
2879                 len = snprintf(name, sizeof(name), "%d", iter.tgid);
2880                 ctx->pos = iter.tgid + TGID_OFFSET;
2881                 if (!proc_fill_cache(file, ctx, name, len,
2882                                      proc_pid_instantiate, iter.task, NULL)) {
2883                         put_task_struct(iter.task);
2884                         return 0;
2885                 }
2886         }
2887         ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
2888         return 0;
2889 }
2890
2891 /*
2892  * Tasks
2893  */
2894 static const struct pid_entry tid_base_stuff[] = {
2895         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2896         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2897         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2898         REG("environ",   S_IRUSR, proc_environ_operations),
2899         INF("auxv",      S_IRUSR, proc_pid_auxv),
2900         ONE("status",    S_IRUGO, proc_pid_status),
2901         ONE("personality", S_IRUSR, proc_pid_personality),
2902         INF("limits",    S_IRUGO, proc_pid_limits),
2903 #ifdef CONFIG_SCHED_DEBUG
2904         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2905 #endif
2906         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2907 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2908         INF("syscall",   S_IRUSR, proc_pid_syscall),
2909 #endif
2910         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
2911         ONE("stat",      S_IRUGO, proc_tid_stat),
2912         ONE("statm",     S_IRUGO, proc_pid_statm),
2913         REG("maps",      S_IRUGO, proc_tid_maps_operations),
2914 #ifdef CONFIG_CHECKPOINT_RESTORE
2915         REG("children",  S_IRUGO, proc_tid_children_operations),
2916 #endif
2917 #ifdef CONFIG_NUMA
2918         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
2919 #endif
2920         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
2921         LNK("cwd",       proc_cwd_link),
2922         LNK("root",      proc_root_link),
2923         LNK("exe",       proc_exe_link),
2924         REG("mounts",    S_IRUGO, proc_mounts_operations),
2925         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2926 #ifdef CONFIG_PROC_PAGE_MONITOR
2927         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2928         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
2929         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2930 #endif
2931 #ifdef CONFIG_SECURITY
2932         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2933 #endif
2934 #ifdef CONFIG_KALLSYMS
2935         INF("wchan",     S_IRUGO, proc_pid_wchan),
2936 #endif
2937 #ifdef CONFIG_STACKTRACE
2938         ONE("stack",      S_IRUSR, proc_pid_stack),
2939 #endif
2940 #ifdef CONFIG_SCHEDSTATS
2941         INF("schedstat", S_IRUGO, proc_pid_schedstat),
2942 #endif
2943 #ifdef CONFIG_LATENCYTOP
2944         REG("latency",  S_IRUGO, proc_lstats_operations),
2945 #endif
2946 #ifdef CONFIG_PROC_PID_CPUSET
2947         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
2948 #endif
2949 #ifdef CONFIG_CGROUPS
2950         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2951 #endif
2952         INF("oom_score", S_IRUGO, proc_oom_score),
2953         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2954         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2955 #ifdef CONFIG_AUDITSYSCALL
2956         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
2957         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2958 #endif
2959 #ifdef CONFIG_FAULT_INJECTION
2960         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2961 #endif
2962 #ifdef CONFIG_TASK_IO_ACCOUNTING
2963         INF("io",       S_IRUSR, proc_tid_io_accounting),
2964 #endif
2965 #ifdef CONFIG_HARDWALL
2966         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
2967 #endif
2968 #ifdef CONFIG_USER_NS
2969         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2970         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2971         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2972 #endif
2973 };
2974
2975 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
2976 {
2977         return proc_pident_readdir(file, ctx,
2978                                    tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
2979 }
2980
2981 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2982 {
2983         return proc_pident_lookup(dir, dentry,
2984                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
2985 }
2986
2987 static const struct file_operations proc_tid_base_operations = {
2988         .read           = generic_read_dir,
2989         .iterate        = proc_tid_base_readdir,
2990         .llseek         = default_llseek,
2991 };
2992
2993 static const struct inode_operations proc_tid_base_inode_operations = {
2994         .lookup         = proc_tid_base_lookup,
2995         .getattr        = pid_getattr,
2996         .setattr        = proc_setattr,
2997 };
2998
2999 static int proc_task_instantiate(struct inode *dir,
3000         struct dentry *dentry, struct task_struct *task, const void *ptr)
3001 {
3002         struct inode *inode;
3003         inode = proc_pid_make_inode(dir->i_sb, task);
3004
3005         if (!inode)
3006                 goto out;
3007         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3008         inode->i_op = &proc_tid_base_inode_operations;
3009         inode->i_fop = &proc_tid_base_operations;
3010         inode->i_flags|=S_IMMUTABLE;
3011
3012         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3013                                                   ARRAY_SIZE(tid_base_stuff)));
3014
3015         d_set_d_op(dentry, &pid_dentry_operations);
3016
3017         d_add(dentry, inode);
3018         /* Close the race of the process dying before we return the dentry */
3019         if (pid_revalidate(dentry, 0))
3020                 return 0;
3021 out:
3022         return -ENOENT;
3023 }
3024
3025 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3026 {
3027         int result = -ENOENT;
3028         struct task_struct *task;
3029         struct task_struct *leader = get_proc_task(dir);
3030         unsigned tid;
3031         struct pid_namespace *ns;
3032
3033         if (!leader)
3034                 goto out_no_task;
3035
3036         tid = name_to_int(dentry);
3037         if (tid == ~0U)
3038                 goto out;
3039
3040         ns = dentry->d_sb->s_fs_info;
3041         rcu_read_lock();
3042         task = find_task_by_pid_ns(tid, ns);
3043         if (task)
3044                 get_task_struct(task);
3045         rcu_read_unlock();
3046         if (!task)
3047                 goto out;
3048         if (!same_thread_group(leader, task))
3049                 goto out_drop_task;
3050
3051         result = proc_task_instantiate(dir, dentry, task, NULL);
3052 out_drop_task:
3053         put_task_struct(task);
3054 out:
3055         put_task_struct(leader);
3056 out_no_task:
3057         return ERR_PTR(result);
3058 }
3059
3060 /*
3061  * Find the first tid of a thread group to return to user space.
3062  *
3063  * Usually this is just the thread group leader, but if the users
3064  * buffer was too small or there was a seek into the middle of the
3065  * directory we have more work todo.
3066  *
3067  * In the case of a short read we start with find_task_by_pid.
3068  *
3069  * In the case of a seek we start with the leader and walk nr
3070  * threads past it.
3071  */
3072 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3073                                         struct pid_namespace *ns)
3074 {
3075         struct task_struct *pos, *task;
3076         unsigned long nr = f_pos;
3077
3078         if (nr != f_pos)        /* 32bit overflow? */
3079                 return NULL;
3080
3081         rcu_read_lock();
3082         task = pid_task(pid, PIDTYPE_PID);
3083         if (!task)
3084                 goto fail;
3085
3086         /* Attempt to start with the tid of a thread */
3087         if (tid && nr) {
3088                 pos = find_task_by_pid_ns(tid, ns);
3089                 if (pos && same_thread_group(pos, task))
3090                         goto found;
3091         }
3092
3093         /* If nr exceeds the number of threads there is nothing todo */
3094         if (nr >= get_nr_threads(task))
3095                 goto fail;
3096
3097         /* If we haven't found our starting place yet start
3098          * with the leader and walk nr threads forward.
3099          */
3100         pos = task = task->group_leader;
3101         do {
3102                 if (!nr--)
3103                         goto found;
3104         } while_each_thread(task, pos);
3105 fail:
3106         pos = NULL;
3107         goto out;
3108 found:
3109         get_task_struct(pos);
3110 out:
3111         rcu_read_unlock();
3112         return pos;
3113 }
3114
3115 /*
3116  * Find the next thread in the thread list.
3117  * Return NULL if there is an error or no next thread.
3118  *
3119  * The reference to the input task_struct is released.
3120  */
3121 static struct task_struct *next_tid(struct task_struct *start)
3122 {
3123         struct task_struct *pos = NULL;
3124         rcu_read_lock();
3125         if (pid_alive(start)) {
3126                 pos = next_thread(start);
3127                 if (thread_group_leader(pos))
3128                         pos = NULL;
3129                 else
3130                         get_task_struct(pos);
3131         }
3132         rcu_read_unlock();
3133         put_task_struct(start);
3134         return pos;
3135 }
3136
3137 /* for the /proc/TGID/task/ directories */
3138 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3139 {
3140         struct inode *inode = file_inode(file);
3141         struct task_struct *task;
3142         struct pid_namespace *ns;
3143         int tid;
3144
3145         if (proc_inode_is_dead(inode))
3146                 return -ENOENT;
3147
3148         if (!dir_emit_dots(file, ctx))
3149                 return 0;
3150
3151         /* f_version caches the tgid value that the last readdir call couldn't
3152          * return. lseek aka telldir automagically resets f_version to 0.
3153          */
3154         ns = file->f_dentry->d_sb->s_fs_info;
3155         tid = (int)file->f_version;
3156         file->f_version = 0;
3157         for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3158              task;
3159              task = next_tid(task), ctx->pos++) {
3160                 char name[PROC_NUMBUF];
3161                 int len;
3162                 tid = task_pid_nr_ns(task, ns);
3163                 len = snprintf(name, sizeof(name), "%d", tid);
3164                 if (!proc_fill_cache(file, ctx, name, len,
3165                                 proc_task_instantiate, task, NULL)) {
3166                         /* returning this tgid failed, save it as the first
3167                          * pid for the next readir call */
3168                         file->f_version = (u64)tid;
3169                         put_task_struct(task);
3170                         break;
3171                 }
3172         }
3173
3174         return 0;
3175 }
3176
3177 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3178 {
3179         struct inode *inode = dentry->d_inode;
3180         struct task_struct *p = get_proc_task(inode);
3181         generic_fillattr(inode, stat);
3182
3183         if (p) {
3184                 stat->nlink += get_nr_threads(p);
3185                 put_task_struct(p);
3186         }
3187
3188         return 0;
3189 }
3190
3191 static const struct inode_operations proc_task_inode_operations = {
3192         .lookup         = proc_task_lookup,
3193         .getattr        = proc_task_getattr,
3194         .setattr        = proc_setattr,
3195         .permission     = proc_pid_permission,
3196 };
3197
3198 static const struct file_operations proc_task_operations = {
3199         .read           = generic_read_dir,
3200         .iterate        = proc_task_readdir,
3201         .llseek         = default_llseek,
3202 };