fs/proc/base.c

   1 /*
   2  *  linux/fs/proc/base.c
   3  *
   4  *  Copyright (C) 1991, 1992 Linus Torvalds
   5  *
   6  *  proc base directory handling functions
   7  *
   8  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
   9  *  Instead of using magical inumbers to determine the kind of object
  10  *  we allocate and fill in-core inodes upon lookup. They don't even
  11  *  go into icache. We cache the reference to task_struct upon lookup too.
  12  *  Eventually it should become a filesystem in its own. We don't use the
  13  *  rest of procfs anymore.
  14  *
  15  *
  16  *  Changelog:
  17  *  17-Jan-2005
  18  *  Allan Bezerra
  19  *  Bruna Moreira <bruna.moreira@indt.org.br>
  20  *  Edjard Mota <edjard.mota@indt.org.br>
  21  *  Ilias Biris <ilias.biris@indt.org.br>
  22  *  Mauricio Lin <mauricio.lin@indt.org.br>
  23  *
  24  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  25  *
  26  *  A new process specific entry (smaps) included in /proc. It shows the
  27  *  size of rss for each memory area. The maps entry lacks information
  28  *  about physical memory size (rss) for each mapped file, i.e.,
  29  *  rss information for executables and library files.
  30  *  This additional information is useful for any tools that need to know
  31  *  about physical memory consumption for a process specific library.
  32  *
  33  *  Changelog:
  34  *  21-Feb-2005
  35  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  36  *  Pud inclusion in the page table walking.
  37  *
  38  *  ChangeLog:
  39  *  10-Mar-2005
  40  *  10LE Instituto Nokia de Tecnologia - INdT:
  41  *  A better way to walks through the page table as suggested by Hugh Dickins.
  42  *
  43  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
  44  *  Smaps information related to shared, private, clean and dirty pages.
  45  *
  46  *  Paul Mundt <paul.mundt@nokia.com>:
  47  *  Overall revision about smaps.
  48  */
  49
  50 #include <asm/uaccess.h>
  51
  52 #include <linux/errno.h>
  53 #include <linux/time.h>
  54 #include <linux/proc_fs.h>
  55 #include <linux/stat.h>
  56 #include <linux/task_io_accounting_ops.h>
  57 #include <linux/init.h>
  58 #include <linux/capability.h>
  59 #include <linux/file.h>
  60 #include <linux/fdtable.h>
  61 #include <linux/string.h>
  62 #include <linux/seq_file.h>
  63 #include <linux/namei.h>
  64 #include <linux/mnt_namespace.h>
  65 #include <linux/mm.h>
  66 #include <linux/swap.h>
  67 #include <linux/rcupdate.h>
  68 #include <linux/kallsyms.h>
  69 #include <linux/stacktrace.h>
  70 #include <linux/resource.h>
  71 #include <linux/module.h>
  72 #include <linux/mount.h>
  73 #include <linux/security.h>
  74 #include <linux/ptrace.h>
  75 #include <linux/tracehook.h>
  76 #include <linux/printk.h>
  77 #include <linux/cgroup.h>
  78 #include <linux/cpuset.h>
  79 #include <linux/audit.h>
  80 #include <linux/poll.h>
  81 #include <linux/nsproxy.h>
  82 #include <linux/oom.h>
  83 #include <linux/elf.h>
  84 #include <linux/pid_namespace.h>
  85 #include <linux/user_namespace.h>
  86 #include <linux/fs_struct.h>
  87 #include <linux/slab.h>
  88 #include <linux/flex_array.h>
  89 #include <linux/posix-timers.h>
  90 #ifdef CONFIG_HARDWALL
  91 #include <asm/hardwall.h>
  92 #endif
  93 #include <trace/events/oom.h>
  94 #include "internal.h"
  95 #include "fd.h"
  96
  97 /* NOTE:
  98  *      Implementing inode permission operations in /proc is almost
  99  *      certainly an error.  Permission checks need to happen during
 100  *      each system call not at open time.  The reason is that most of
 101  *      what we wish to check for permissions in /proc varies at runtime.
 102  *
 103  *      The classic example of a problem is opening file descriptors
 104  *      in /proc for a task before it execs a suid executable.
 105  */
 106
 107 struct pid_entry {
 108         const 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 seq_file *m, struct pid_namespace *ns,
 204                             struct pid *pid, struct task_struct *task)
 205 {
 206         /*
 207          * Rely on struct seq_operations::show() being called once
 208          * per internal buffer allocation. See single_open(), traverse().
 209          */
 210         BUG_ON(m->size < PAGE_SIZE);
 211         m->count += get_cmdline(task, m->buf, PAGE_SIZE);
 212         return 0;
 213 }
 214
 215 static int proc_pid_auxv(struct seq_file *m, struct pid_namespace *ns,
 216                          struct pid *pid, struct task_struct *task)
 217 {
 218         struct mm_struct *mm = mm_access(task, PTRACE_MODE_READ);
 219         if (mm && !IS_ERR(mm)) {
 220                 unsigned int nwords = 0;
 221                 do {
 222                         nwords += 2;
 223                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 224                 seq_write(m, mm->saved_auxv, nwords * sizeof(mm->saved_auxv[0]));
 225                 mmput(mm);
 226                 return 0;
 227         } else
 228                 return PTR_ERR(mm);
 229 }
 230
 231
 232 #ifdef CONFIG_KALLSYMS
 233 /*
 234  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 235  * Returns the resolved symbol.  If that fails, simply return the address.
 236  */
 237 static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
 238                           struct pid *pid, struct task_struct *task)
 239 {
 240         unsigned long wchan;
 241         char symname[KSYM_NAME_LEN];
 242
 243         wchan = get_wchan(task);
 244
 245         if (lookup_symbol_name(wchan, symname) < 0)
 246                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
 247                         return 0;
 248                 else
 249                         return seq_printf(m, "%lu", wchan);
 250         else
 251                 return seq_printf(m, "%s", symname);
 252 }
 253 #endif /* CONFIG_KALLSYMS */
 254
 255 static int lock_trace(struct task_struct *task)
 256 {
 257         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
 258         if (err)
 259                 return err;
 260         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH)) {
 261                 mutex_unlock(&task->signal->cred_guard_mutex);
 262                 return -EPERM;
 263         }
 264         return 0;
 265 }
 266
 267 static void unlock_trace(struct task_struct *task)
 268 {
 269         mutex_unlock(&task->signal->cred_guard_mutex);
 270 }
 271
 272 #ifdef CONFIG_STACKTRACE
 273
 274 #define MAX_STACK_TRACE_DEPTH   64
 275
 276 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 277                           struct pid *pid, struct task_struct *task)
 278 {
 279         struct stack_trace trace;
 280         unsigned long *entries;
 281         int err;
 282         int i;
 283
 284         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 285         if (!entries)
 286                 return -ENOMEM;
 287
 288         trace.nr_entries        = 0;
 289         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 290         trace.entries           = entries;
 291         trace.skip              = 0;
 292
 293         err = lock_trace(task);
 294         if (!err) {
 295                 save_stack_trace_tsk(task, &trace);
 296
 297                 for (i = 0; i < trace.nr_entries; i++) {
 298                         seq_printf(m, "[<%pK>] %pS\n",
 299                                    (void *)entries[i], (void *)entries[i]);
 300                 }
 301                 unlock_trace(task);
 302         }
 303         kfree(entries);
 304
 305         return err;
 306 }
 307 #endif
 308
 309 #ifdef CONFIG_SCHEDSTATS
 310 /*
 311  * Provides /proc/PID/schedstat
 312  */
 313 static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
 314                               struct pid *pid, struct task_struct *task)
 315 {
 316         return seq_printf(m, "%llu %llu %lu\n",
 317                         (unsigned long long)task->se.sum_exec_runtime,
 318                         (unsigned long long)task->sched_info.run_delay,
 319                         task->sched_info.pcount);
 320 }
 321 #endif
 322
 323 #ifdef CONFIG_LATENCYTOP
 324 static int lstats_show_proc(struct seq_file *m, void *v)
 325 {
 326         int i;
 327         struct inode *inode = m->private;
 328         struct task_struct *task = get_proc_task(inode);
 329
 330         if (!task)
 331                 return -ESRCH;
 332         seq_puts(m, "Latency Top version : v0.1\n");
 333         for (i = 0; i < 32; i++) {
 334                 struct latency_record *lr = &task->latency_record[i];
 335                 if (lr->backtrace[0]) {
 336                         int q;
 337                         seq_printf(m, "%i %li %li",
 338                                    lr->count, lr->time, lr->max);
 339                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 340                                 unsigned long bt = lr->backtrace[q];
 341                                 if (!bt)
 342                                         break;
 343                                 if (bt == ULONG_MAX)
 344                                         break;
 345                                 seq_printf(m, " %ps", (void *)bt);
 346                         }
 347                         seq_putc(m, '\n');
 348                 }
 349
 350         }
 351         put_task_struct(task);
 352         return 0;
 353 }
 354
 355 static int lstats_open(struct inode *inode, struct file *file)
 356 {
 357         return single_open(file, lstats_show_proc, inode);
 358 }
 359
 360 static ssize_t lstats_write(struct file *file, const char __user *buf,
 361                             size_t count, loff_t *offs)
 362 {
 363         struct task_struct *task = get_proc_task(file_inode(file));
 364
 365         if (!task)
 366                 return -ESRCH;
 367         clear_all_latency_tracing(task);
 368         put_task_struct(task);
 369
 370         return count;
 371 }
 372
 373 static const struct file_operations proc_lstats_operations = {
 374         .open           = lstats_open,
 375         .read           = seq_read,
 376         .write          = lstats_write,
 377         .llseek         = seq_lseek,
 378         .release        = single_release,
 379 };
 380
 381 #endif
 382
 383 #ifdef CONFIG_CGROUPS
 384 static int cgroup_open(struct inode *inode, struct file *file)
 385 {
 386         struct pid *pid = PROC_I(inode)->pid;
 387         return single_open(file, proc_cgroup_show, pid);
 388 }
 389
 390 static const struct file_operations proc_cgroup_operations = {
 391         .open           = cgroup_open,
 392         .read           = seq_read,
 393         .llseek         = seq_lseek,
 394         .release        = single_release,
 395 };
 396 #endif
 397
 398 #ifdef CONFIG_PROC_PID_CPUSET
 399
 400 static int cpuset_open(struct inode *inode, struct file *file)
 401 {
 402         struct pid *pid = PROC_I(inode)->pid;
 403         return single_open(file, proc_cpuset_show, pid);
 404 }
 405
 406 static const struct file_operations proc_cpuset_operations = {
 407         .open           = cpuset_open,
 408         .read           = seq_read,
 409         .llseek         = seq_lseek,
 410         .release        = single_release,
 411 };
 412 #endif
 413
 414 static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
 415                           struct pid *pid, struct task_struct *task)
 416 {
 417         unsigned long totalpages = totalram_pages + total_swap_pages;
 418         unsigned long points = 0;
 419
 420         read_lock(&tasklist_lock);
 421         if (pid_alive(task))
 422                 points = oom_badness(task, NULL, NULL, totalpages) *
 423                                                 1000 / totalpages;
 424         read_unlock(&tasklist_lock);
 425         return seq_printf(m, "%lu\n", points);
 426 }
 427
 428 struct limit_names {
 429         const char *name;
 430         const char *unit;
 431 };
 432
 433 static const struct limit_names lnames[RLIM_NLIMITS] = {
 434         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 435         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 436         [RLIMIT_DATA] = {"Max data size", "bytes"},
 437         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 438         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 439         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 440         [RLIMIT_NPROC] = {"Max processes", "processes"},
 441         [RLIMIT_NOFILE] = {"Max open files", "files"},
 442         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 443         [RLIMIT_AS] = {"Max address space", "bytes"},
 444         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 445         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 446         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 447         [RLIMIT_NICE] = {"Max nice priority", NULL},
 448         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 449         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 450 };
 451
 452 /* Display limits for a process */
 453 static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
 454                            struct pid *pid, struct task_struct *task)
 455 {
 456         unsigned int i;
 457         unsigned long flags;
 458
 459         struct rlimit rlim[RLIM_NLIMITS];
 460
 461         if (!lock_task_sighand(task, &flags))
 462                 return 0;
 463         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 464         unlock_task_sighand(task, &flags);
 465
 466         /*
 467          * print the file header
 468          */
 469        seq_printf(m, "%-25s %-20s %-20s %-10s\n",
 470                         "Limit", "Soft Limit", "Hard Limit", "Units");
 471
 472         for (i = 0; i < RLIM_NLIMITS; i++) {
 473                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 474                         seq_printf(m, "%-25s %-20s ",
 475                                          lnames[i].name, "unlimited");
 476                 else
 477                         seq_printf(m, "%-25s %-20lu ",
 478                                          lnames[i].name, rlim[i].rlim_cur);
 479
 480                 if (rlim[i].rlim_max == RLIM_INFINITY)
 481                         seq_printf(m, "%-20s ", "unlimited");
 482                 else
 483                         seq_printf(m, "%-20lu ", rlim[i].rlim_max);
 484
 485                 if (lnames[i].unit)
 486                         seq_printf(m, "%-10s\n", lnames[i].unit);
 487                 else
 488                         seq_putc(m, '\n');
 489         }
 490
 491         return 0;
 492 }
 493
 494 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 495 static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
 496                             struct pid *pid, struct task_struct *task)
 497 {
 498         long nr;
 499         unsigned long args[6], sp, pc;
 500         int res = lock_trace(task);
 501         if (res)
 502                 return res;
 503
 504         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 505                 seq_puts(m, "running\n");
 506         else if (nr < 0)
 507                 seq_printf(m, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 508         else
 509                 seq_printf(m,
 510                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 511                        nr,
 512                        args[0], args[1], args[2], args[3], args[4], args[5],
 513                        sp, pc);
 514         unlock_trace(task);
 515         return res;
 516 }
 517 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 518
 519 /************************************************************************/
 520 /*                       Here the fs part begins                        */
 521 /************************************************************************/
 522
 523 /* permission checks */
 524 static int proc_fd_access_allowed(struct inode *inode)
 525 {
 526         struct task_struct *task;
 527         int allowed = 0;
 528         /* Allow access to a task's file descriptors if it is us or we
 529          * may use ptrace attach to the process and find out that
 530          * information.
 531          */
 532         task = get_proc_task(inode);
 533         if (task) {
 534                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
 535                 put_task_struct(task);
 536         }
 537         return allowed;
 538 }
 539
 540 int proc_setattr(struct dentry *dentry, struct iattr *attr)
 541 {
 542         int error;
 543         struct inode *inode = dentry->d_inode;
 544
 545         if (attr->ia_valid & ATTR_MODE)
 546                 return -EPERM;
 547
 548         error = inode_change_ok(inode, attr);
 549         if (error)
 550                 return error;
 551
 552         setattr_copy(inode, attr);
 553         mark_inode_dirty(inode);
 554         return 0;
 555 }
 556
 557 /*
 558  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
 559  * or euid/egid (for hide_pid_min=2)?
 560  */
 561 static bool has_pid_permissions(struct pid_namespace *pid,
 562                                  struct task_struct *task,
 563                                  int hide_pid_min)
 564 {
 565         if (pid->hide_pid < hide_pid_min)
 566                 return true;
 567         if (in_group_p(pid->pid_gid))
 568                 return true;
 569         return ptrace_may_access(task, PTRACE_MODE_READ);
 570 }
 571
 572
 573 static int proc_pid_permission(struct inode *inode, int mask)
 574 {
 575         struct pid_namespace *pid = inode->i_sb->s_fs_info;
 576         struct task_struct *task;
 577         bool has_perms;
 578
 579         task = get_proc_task(inode);
 580         if (!task)
 581                 return -ESRCH;
 582         has_perms = has_pid_permissions(pid, task, 1);
 583         put_task_struct(task);
 584
 585         if (!has_perms) {
 586                 if (pid->hide_pid == 2) {
 587                         /*
 588                          * Let's make getdents(), stat(), and open()
 589                          * consistent with each other.  If a process
 590                          * may not stat() a file, it shouldn't be seen
 591                          * in procfs at all.
 592                          */
 593                         return -ENOENT;
 594                 }
 595
 596                 return -EPERM;
 597         }
 598         return generic_permission(inode, mask);
 599 }
 600
 601
 602
 603 static const struct inode_operations proc_def_inode_operations = {
 604         .setattr        = proc_setattr,
 605 };
 606
 607 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 608
 609 static ssize_t proc_info_read(struct file * file, char __user * buf,
 610                           size_t count, loff_t *ppos)
 611 {
 612         struct inode * inode = file_inode(file);
 613         unsigned long page;
 614         ssize_t length;
 615         struct task_struct *task = get_proc_task(inode);
 616
 617         length = -ESRCH;
 618         if (!task)
 619                 goto out_no_task;
 620
 621         if (count > PROC_BLOCK_SIZE)
 622                 count = PROC_BLOCK_SIZE;
 623
 624         length = -ENOMEM;
 625         if (!(page = __get_free_page(GFP_TEMPORARY)))
 626                 goto out;
 627
 628         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 629
 630         if (length >= 0)
 631                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 632         free_page(page);
 633 out:
 634         put_task_struct(task);
 635 out_no_task:
 636         return length;
 637 }
 638
 639 static const struct file_operations proc_info_file_operations = {
 640         .read           = proc_info_read,
 641         .llseek         = generic_file_llseek,
 642 };
 643
 644 static int proc_single_show(struct seq_file *m, void *v)
 645 {
 646         struct inode *inode = m->private;
 647         struct pid_namespace *ns;
 648         struct pid *pid;
 649         struct task_struct *task;
 650         int ret;
 651
 652         ns = inode->i_sb->s_fs_info;
 653         pid = proc_pid(inode);
 654         task = get_pid_task(pid, PIDTYPE_PID);
 655         if (!task)
 656                 return -ESRCH;
 657
 658         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 659
 660         put_task_struct(task);
 661         return ret;
 662 }
 663
 664 static int proc_single_open(struct inode *inode, struct file *filp)
 665 {
 666         return single_open(filp, proc_single_show, inode);
 667 }
 668
 669 static const struct file_operations proc_single_file_operations = {
 670         .open           = proc_single_open,
 671         .read           = seq_read,
 672         .llseek         = seq_lseek,
 673         .release        = single_release,
 674 };
 675
 676 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
 677 {
 678         struct task_struct *task = get_proc_task(file_inode(file));
 679         struct mm_struct *mm;
 680
 681         if (!task)
 682                 return -ESRCH;
 683
 684         mm = mm_access(task, mode);
 685         put_task_struct(task);
 686
 687         if (IS_ERR(mm))
 688                 return PTR_ERR(mm);
 689
 690         if (mm) {
 691                 /* ensure this mm_struct can't be freed */
 692                 atomic_inc(&mm->mm_count);
 693                 /* but do not pin its memory */
 694                 mmput(mm);
 695         }
 696
 697         file->private_data = mm;
 698
 699         return 0;
 700 }
 701
 702 static int mem_open(struct inode *inode, struct file *file)
 703 {
 704         int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
 705
 706         /* OK to pass negative loff_t, we can catch out-of-range */
 707         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 708
 709         return ret;
 710 }
 711
 712 static ssize_t mem_rw(struct file *file, char __user *buf,
 713                         size_t count, loff_t *ppos, int write)
 714 {
 715         struct mm_struct *mm = file->private_data;
 716         unsigned long addr = *ppos;
 717         ssize_t copied;
 718         char *page;
 719
 720         if (!mm)
 721                 return 0;
 722
 723         page = (char *)__get_free_page(GFP_TEMPORARY);
 724         if (!page)
 725                 return -ENOMEM;
 726
 727         copied = 0;
 728         if (!atomic_inc_not_zero(&mm->mm_users))
 729                 goto free;
 730
 731         while (count > 0) {
 732                 int this_len = min_t(int, count, PAGE_SIZE);
 733
 734                 if (write && copy_from_user(page, buf, this_len)) {
 735                         copied = -EFAULT;
 736                         break;
 737                 }
 738
 739                 this_len = access_remote_vm(mm, addr, page, this_len, write);
 740                 if (!this_len) {
 741                         if (!copied)
 742                                 copied = -EIO;
 743                         break;
 744                 }
 745
 746                 if (!write && copy_to_user(buf, page, this_len)) {
 747                         copied = -EFAULT;
 748                         break;
 749                 }
 750
 751                 buf += this_len;
 752                 addr += this_len;
 753                 copied += this_len;
 754                 count -= this_len;
 755         }
 756         *ppos = addr;
 757
 758         mmput(mm);
 759 free:
 760         free_page((unsigned long) page);
 761         return copied;
 762 }
 763
 764 static ssize_t mem_read(struct file *file, char __user *buf,
 765                         size_t count, loff_t *ppos)
 766 {
 767         return mem_rw(file, buf, count, ppos, 0);
 768 }
 769
 770 static ssize_t mem_write(struct file *file, const char __user *buf,
 771                          size_t count, loff_t *ppos)
 772 {
 773         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 774 }
 775
 776 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 777 {
 778         switch (orig) {
 779         case 0:
 780                 file->f_pos = offset;
 781                 break;
 782         case 1:
 783                 file->f_pos += offset;
 784                 break;
 785         default:
 786                 return -EINVAL;
 787         }
 788         force_successful_syscall_return();
 789         return file->f_pos;
 790 }
 791
 792 static int mem_release(struct inode *inode, struct file *file)
 793 {
 794         struct mm_struct *mm = file->private_data;
 795         if (mm)
 796                 mmdrop(mm);
 797         return 0;
 798 }
 799
 800 static const struct file_operations proc_mem_operations = {
 801         .llseek         = mem_lseek,
 802         .read           = mem_read,
 803         .write          = mem_write,
 804         .open           = mem_open,
 805         .release        = mem_release,
 806 };
 807
 808 static int environ_open(struct inode *inode, struct file *file)
 809 {
 810         return __mem_open(inode, file, PTRACE_MODE_READ);
 811 }
 812
 813 static ssize_t environ_read(struct file *file, char __user *buf,
 814                         size_t count, loff_t *ppos)
 815 {
 816         char *page;
 817         unsigned long src = *ppos;
 818         int ret = 0;
 819         struct mm_struct *mm = file->private_data;
 820
 821         if (!mm)
 822                 return 0;
 823
 824         page = (char *)__get_free_page(GFP_TEMPORARY);
 825         if (!page)
 826                 return -ENOMEM;
 827
 828         ret = 0;
 829         if (!atomic_inc_not_zero(&mm->mm_users))
 830                 goto free;
 831         while (count > 0) {
 832                 size_t this_len, max_len;
 833                 int retval;
 834
 835                 if (src >= (mm->env_end - mm->env_start))
 836                         break;
 837
 838                 this_len = mm->env_end - (mm->env_start + src);
 839
 840                 max_len = min_t(size_t, PAGE_SIZE, count);
 841                 this_len = min(max_len, this_len);
 842
 843                 retval = access_remote_vm(mm, (mm->env_start + src),
 844                         page, this_len, 0);
 845
 846                 if (retval <= 0) {
 847                         ret = retval;
 848                         break;
 849                 }
 850
 851                 if (copy_to_user(buf, page, retval)) {
 852                         ret = -EFAULT;
 853                         break;
 854                 }
 855
 856                 ret += retval;
 857                 src += retval;
 858                 buf += retval;
 859                 count -= retval;
 860         }
 861         *ppos = src;
 862         mmput(mm);
 863
 864 free:
 865         free_page((unsigned long) page);
 866         return ret;
 867 }
 868
 869 static const struct file_operations proc_environ_operations = {
 870         .open           = environ_open,
 871         .read           = environ_read,
 872         .llseek         = generic_file_llseek,
 873         .release        = mem_release,
 874 };
 875
 876 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
 877                             loff_t *ppos)
 878 {
 879         struct task_struct *task = get_proc_task(file_inode(file));
 880         char buffer[PROC_NUMBUF];
 881         int oom_adj = OOM_ADJUST_MIN;
 882         size_t len;
 883         unsigned long flags;
 884
 885         if (!task)
 886                 return -ESRCH;
 887         if (lock_task_sighand(task, &flags)) {
 888                 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
 889                         oom_adj = OOM_ADJUST_MAX;
 890                 else
 891                         oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
 892                                   OOM_SCORE_ADJ_MAX;
 893                 unlock_task_sighand(task, &flags);
 894         }
 895         put_task_struct(task);
 896         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
 897         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 898 }
 899
 900 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
 901                              size_t count, loff_t *ppos)
 902 {
 903         struct task_struct *task;
 904         char buffer[PROC_NUMBUF];
 905         int oom_adj;
 906         unsigned long flags;
 907         int err;
 908
 909         memset(buffer, 0, sizeof(buffer));
 910         if (count > sizeof(buffer) - 1)
 911                 count = sizeof(buffer) - 1;
 912         if (copy_from_user(buffer, buf, count)) {
 913                 err = -EFAULT;
 914                 goto out;
 915         }
 916
 917         err = kstrtoint(strstrip(buffer), 0, &oom_adj);
 918         if (err)
 919                 goto out;
 920         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
 921              oom_adj != OOM_DISABLE) {
 922                 err = -EINVAL;
 923                 goto out;
 924         }
 925
 926         task = get_proc_task(file_inode(file));
 927         if (!task) {
 928                 err = -ESRCH;
 929                 goto out;
 930         }
 931
 932         task_lock(task);
 933         if (!task->mm) {
 934                 err = -EINVAL;
 935                 goto err_task_lock;
 936         }
 937
 938         if (!lock_task_sighand(task, &flags)) {
 939                 err = -ESRCH;
 940                 goto err_task_lock;
 941         }
 942
 943         /*
 944          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
 945          * value is always attainable.
 946          */
 947         if (oom_adj == OOM_ADJUST_MAX)
 948                 oom_adj = OOM_SCORE_ADJ_MAX;
 949         else
 950                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
 951
 952         if (oom_adj < task->signal->oom_score_adj &&
 953             !capable(CAP_SYS_RESOURCE)) {
 954                 err = -EACCES;
 955                 goto err_sighand;
 956         }
 957
 958         /*
 959          * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
 960          * /proc/pid/oom_score_adj instead.
 961          */
 962         pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
 963                   current->comm, task_pid_nr(current), task_pid_nr(task),
 964                   task_pid_nr(task));
 965
 966         task->signal->oom_score_adj = oom_adj;
 967         trace_oom_score_adj_update(task);
 968 err_sighand:
 969         unlock_task_sighand(task, &flags);
 970 err_task_lock:
 971         task_unlock(task);
 972         put_task_struct(task);
 973 out:
 974         return err < 0 ? err : count;
 975 }
 976
 977 static const struct file_operations proc_oom_adj_operations = {
 978         .read           = oom_adj_read,
 979         .write          = oom_adj_write,
 980         .llseek         = generic_file_llseek,
 981 };
 982
 983 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
 984                                         size_t count, loff_t *ppos)
 985 {
 986         struct task_struct *task = get_proc_task(file_inode(file));
 987         char buffer[PROC_NUMBUF];
 988         short oom_score_adj = OOM_SCORE_ADJ_MIN;
 989         unsigned long flags;
 990         size_t len;
 991
 992         if (!task)
 993                 return -ESRCH;
 994         if (lock_task_sighand(task, &flags)) {
 995                 oom_score_adj = task->signal->oom_score_adj;
 996                 unlock_task_sighand(task, &flags);
 997         }
 998         put_task_struct(task);
 999         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1000         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1001 }
1002
1003 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1004                                         size_t count, loff_t *ppos)
1005 {
1006         struct task_struct *task;
1007         char buffer[PROC_NUMBUF];
1008         unsigned long flags;
1009         int oom_score_adj;
1010         int err;
1011
1012         memset(buffer, 0, sizeof(buffer));
1013         if (count > sizeof(buffer) - 1)
1014                 count = sizeof(buffer) - 1;
1015         if (copy_from_user(buffer, buf, count)) {
1016                 err = -EFAULT;
1017                 goto out;
1018         }
1019
1020         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1021         if (err)
1022                 goto out;
1023         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1024                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1025                 err = -EINVAL;
1026                 goto out;
1027         }
1028
1029         task = get_proc_task(file_inode(file));
1030         if (!task) {
1031                 err = -ESRCH;
1032                 goto out;
1033         }
1034
1035         task_lock(task);
1036         if (!task->mm) {
1037                 err = -EINVAL;
1038                 goto err_task_lock;
1039         }
1040
1041         if (!lock_task_sighand(task, &flags)) {
1042                 err = -ESRCH;
1043                 goto err_task_lock;
1044         }
1045
1046         if ((short)oom_score_adj < task->signal->oom_score_adj_min &&
1047                         !capable(CAP_SYS_RESOURCE)) {
1048                 err = -EACCES;
1049                 goto err_sighand;
1050         }
1051
1052         task->signal->oom_score_adj = (short)oom_score_adj;
1053         if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1054                 task->signal->oom_score_adj_min = (short)oom_score_adj;
1055         trace_oom_score_adj_update(task);
1056
1057 err_sighand:
1058         unlock_task_sighand(task, &flags);
1059 err_task_lock:
1060         task_unlock(task);
1061         put_task_struct(task);
1062 out:
1063         return err < 0 ? err : count;
1064 }
1065
1066 static const struct file_operations proc_oom_score_adj_operations = {
1067         .read           = oom_score_adj_read,
1068         .write          = oom_score_adj_write,
1069         .llseek         = default_llseek,
1070 };
1071
1072 #ifdef CONFIG_AUDITSYSCALL
1073 #define TMPBUFLEN 21
1074 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1075                                   size_t count, loff_t *ppos)
1076 {
1077         struct inode * inode = file_inode(file);
1078         struct task_struct *task = get_proc_task(inode);
1079         ssize_t length;
1080         char tmpbuf[TMPBUFLEN];
1081
1082         if (!task)
1083                 return -ESRCH;
1084         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1085                            from_kuid(file->f_cred->user_ns,
1086                                      audit_get_loginuid(task)));
1087         put_task_struct(task);
1088         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1089 }
1090
1091 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1092                                    size_t count, loff_t *ppos)
1093 {
1094         struct inode * inode = file_inode(file);
1095         char *page, *tmp;
1096         ssize_t length;
1097         uid_t loginuid;
1098         kuid_t kloginuid;
1099
1100         rcu_read_lock();
1101         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1102                 rcu_read_unlock();
1103                 return -EPERM;
1104         }
1105         rcu_read_unlock();
1106
1107         if (count >= PAGE_SIZE)
1108                 count = PAGE_SIZE - 1;
1109
1110         if (*ppos != 0) {
1111                 /* No partial writes. */
1112                 return -EINVAL;
1113         }
1114         page = (char*)__get_free_page(GFP_TEMPORARY);
1115         if (!page)
1116                 return -ENOMEM;
1117         length = -EFAULT;
1118         if (copy_from_user(page, buf, count))
1119                 goto out_free_page;
1120
1121         page[count] = '\0';
1122         loginuid = simple_strtoul(page, &tmp, 10);
1123         if (tmp == page) {
1124                 length = -EINVAL;
1125                 goto out_free_page;
1126
1127         }
1128
1129         /* is userspace tring to explicitly UNSET the loginuid? */
1130         if (loginuid == AUDIT_UID_UNSET) {
1131                 kloginuid = INVALID_UID;
1132         } else {
1133                 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1134                 if (!uid_valid(kloginuid)) {
1135                         length = -EINVAL;
1136                         goto out_free_page;
1137                 }
1138         }
1139
1140         length = audit_set_loginuid(kloginuid);
1141         if (likely(length == 0))
1142                 length = count;
1143
1144 out_free_page:
1145         free_page((unsigned long) page);
1146         return length;
1147 }
1148
1149 static const struct file_operations proc_loginuid_operations = {
1150         .read           = proc_loginuid_read,
1151         .write          = proc_loginuid_write,
1152         .llseek         = generic_file_llseek,
1153 };
1154
1155 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1156                                   size_t count, loff_t *ppos)
1157 {
1158         struct inode * inode = file_inode(file);
1159         struct task_struct *task = get_proc_task(inode);
1160         ssize_t length;
1161         char tmpbuf[TMPBUFLEN];
1162
1163         if (!task)
1164                 return -ESRCH;
1165         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1166                                 audit_get_sessionid(task));
1167         put_task_struct(task);
1168         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1169 }
1170
1171 static const struct file_operations proc_sessionid_operations = {
1172         .read           = proc_sessionid_read,
1173         .llseek         = generic_file_llseek,
1174 };
1175 #endif
1176
1177 #ifdef CONFIG_FAULT_INJECTION
1178 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1179                                       size_t count, loff_t *ppos)
1180 {
1181         struct task_struct *task = get_proc_task(file_inode(file));
1182         char buffer[PROC_NUMBUF];
1183         size_t len;
1184         int make_it_fail;
1185
1186         if (!task)
1187                 return -ESRCH;
1188         make_it_fail = task->make_it_fail;
1189         put_task_struct(task);
1190
1191         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1192
1193         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1194 }
1195
1196 static ssize_t proc_fault_inject_write(struct file * file,
1197                         const char __user * buf, size_t count, loff_t *ppos)
1198 {
1199         struct task_struct *task;
1200         char buffer[PROC_NUMBUF], *end;
1201         int make_it_fail;
1202
1203         if (!capable(CAP_SYS_RESOURCE))
1204                 return -EPERM;
1205         memset(buffer, 0, sizeof(buffer));
1206         if (count > sizeof(buffer) - 1)
1207                 count = sizeof(buffer) - 1;
1208         if (copy_from_user(buffer, buf, count))
1209                 return -EFAULT;
1210         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1211         if (*end)
1212                 return -EINVAL;
1213         if (make_it_fail < 0 || make_it_fail > 1)
1214                 return -EINVAL;
1215
1216         task = get_proc_task(file_inode(file));
1217         if (!task)
1218                 return -ESRCH;
1219         task->make_it_fail = make_it_fail;
1220         put_task_struct(task);
1221
1222         return count;
1223 }
1224
1225 static const struct file_operations proc_fault_inject_operations = {
1226         .read           = proc_fault_inject_read,
1227         .write          = proc_fault_inject_write,
1228         .llseek         = generic_file_llseek,
1229 };
1230 #endif
1231
1232
1233 #ifdef CONFIG_SCHED_DEBUG
1234 /*
1235  * Print out various scheduling related per-task fields:
1236  */
1237 static int sched_show(struct seq_file *m, void *v)
1238 {
1239         struct inode *inode = m->private;
1240         struct task_struct *p;
1241
1242         p = get_proc_task(inode);
1243         if (!p)
1244                 return -ESRCH;
1245         proc_sched_show_task(p, m);
1246
1247         put_task_struct(p);
1248
1249         return 0;
1250 }
1251
1252 static ssize_t
1253 sched_write(struct file *file, const char __user *buf,
1254             size_t count, loff_t *offset)
1255 {
1256         struct inode *inode = file_inode(file);
1257         struct task_struct *p;
1258
1259         p = get_proc_task(inode);
1260         if (!p)
1261                 return -ESRCH;
1262         proc_sched_set_task(p);
1263
1264         put_task_struct(p);
1265
1266         return count;
1267 }
1268
1269 static int sched_open(struct inode *inode, struct file *filp)
1270 {
1271         return single_open(filp, sched_show, inode);
1272 }
1273
1274 static const struct file_operations proc_pid_sched_operations = {
1275         .open           = sched_open,
1276         .read           = seq_read,
1277         .write          = sched_write,
1278         .llseek         = seq_lseek,
1279         .release        = single_release,
1280 };
1281
1282 #endif
1283
1284 #ifdef CONFIG_SCHED_AUTOGROUP
1285 /*
1286  * Print out autogroup related information:
1287  */
1288 static int sched_autogroup_show(struct seq_file *m, void *v)
1289 {
1290         struct inode *inode = m->private;
1291         struct task_struct *p;
1292
1293         p = get_proc_task(inode);
1294         if (!p)
1295                 return -ESRCH;
1296         proc_sched_autogroup_show_task(p, m);
1297
1298         put_task_struct(p);
1299
1300         return 0;
1301 }
1302
1303 static ssize_t
1304 sched_autogroup_write(struct file *file, const char __user *buf,
1305             size_t count, loff_t *offset)
1306 {
1307         struct inode *inode = file_inode(file);
1308         struct task_struct *p;
1309         char buffer[PROC_NUMBUF];
1310         int nice;
1311         int err;
1312
1313         memset(buffer, 0, sizeof(buffer));
1314         if (count > sizeof(buffer) - 1)
1315                 count = sizeof(buffer) - 1;
1316         if (copy_from_user(buffer, buf, count))
1317                 return -EFAULT;
1318
1319         err = kstrtoint(strstrip(buffer), 0, &nice);
1320         if (err < 0)
1321                 return err;
1322
1323         p = get_proc_task(inode);
1324         if (!p)
1325                 return -ESRCH;
1326
1327         err = proc_sched_autogroup_set_nice(p, nice);
1328         if (err)
1329                 count = err;
1330
1331         put_task_struct(p);
1332
1333         return count;
1334 }
1335
1336 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1337 {
1338         int ret;
1339
1340         ret = single_open(filp, sched_autogroup_show, NULL);
1341         if (!ret) {
1342                 struct seq_file *m = filp->private_data;
1343
1344                 m->private = inode;
1345         }
1346         return ret;
1347 }
1348
1349 static const struct file_operations proc_pid_sched_autogroup_operations = {
1350         .open           = sched_autogroup_open,
1351         .read           = seq_read,
1352         .write          = sched_autogroup_write,
1353         .llseek         = seq_lseek,
1354         .release        = single_release,
1355 };
1356
1357 #endif /* CONFIG_SCHED_AUTOGROUP */
1358
1359 static ssize_t comm_write(struct file *file, const char __user *buf,
1360                                 size_t count, loff_t *offset)
1361 {
1362         struct inode *inode = file_inode(file);
1363         struct task_struct *p;
1364         char buffer[TASK_COMM_LEN];
1365         const size_t maxlen = sizeof(buffer) - 1;
1366
1367         memset(buffer, 0, sizeof(buffer));
1368         if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1369                 return -EFAULT;
1370
1371         p = get_proc_task(inode);
1372         if (!p)
1373                 return -ESRCH;
1374
1375         if (same_thread_group(current, p))
1376                 set_task_comm(p, buffer);
1377         else
1378                 count = -EINVAL;
1379
1380         put_task_struct(p);
1381
1382         return count;
1383 }
1384
1385 static int comm_show(struct seq_file *m, void *v)
1386 {
1387         struct inode *inode = m->private;
1388         struct task_struct *p;
1389
1390         p = get_proc_task(inode);
1391         if (!p)
1392                 return -ESRCH;
1393
1394         task_lock(p);
1395         seq_printf(m, "%s\n", p->comm);
1396         task_unlock(p);
1397
1398         put_task_struct(p);
1399
1400         return 0;
1401 }
1402
1403 static int comm_open(struct inode *inode, struct file *filp)
1404 {
1405         return single_open(filp, comm_show, inode);
1406 }
1407
1408 static const struct file_operations proc_pid_set_comm_operations = {
1409         .open           = comm_open,
1410         .read           = seq_read,
1411         .write          = comm_write,
1412         .llseek         = seq_lseek,
1413         .release        = single_release,
1414 };
1415
1416 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1417 {
1418         struct task_struct *task;
1419         struct mm_struct *mm;
1420         struct file *exe_file;
1421
1422         task = get_proc_task(dentry->d_inode);
1423         if (!task)
1424                 return -ENOENT;
1425         mm = get_task_mm(task);
1426         put_task_struct(task);
1427         if (!mm)
1428                 return -ENOENT;
1429         exe_file = get_mm_exe_file(mm);
1430         mmput(mm);
1431         if (exe_file) {
1432                 *exe_path = exe_file->f_path;
1433                 path_get(&exe_file->f_path);
1434                 fput(exe_file);
1435                 return 0;
1436         } else
1437                 return -ENOENT;
1438 }
1439
1440 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1441 {
1442         struct inode *inode = dentry->d_inode;
1443         struct path path;
1444         int error = -EACCES;
1445
1446         /* Are we allowed to snoop on the tasks file descriptors? */
1447         if (!proc_fd_access_allowed(inode))
1448                 goto out;
1449
1450         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1451         if (error)
1452                 goto out;
1453
1454         nd_jump_link(nd, &path);
1455         return NULL;
1456 out:
1457         return ERR_PTR(error);
1458 }
1459
1460 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1461 {
1462         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1463         char *pathname;
1464         int len;
1465
1466         if (!tmp)
1467                 return -ENOMEM;
1468
1469         pathname = d_path(path, tmp, PAGE_SIZE);
1470         len = PTR_ERR(pathname);
1471         if (IS_ERR(pathname))
1472                 goto out;
1473         len = tmp + PAGE_SIZE - 1 - pathname;
1474
1475         if (len > buflen)
1476                 len = buflen;
1477         if (copy_to_user(buffer, pathname, len))
1478                 len = -EFAULT;
1479  out:
1480         free_page((unsigned long)tmp);
1481         return len;
1482 }
1483
1484 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1485 {
1486         int error = -EACCES;
1487         struct inode *inode = dentry->d_inode;
1488         struct path path;
1489
1490         /* Are we allowed to snoop on the tasks file descriptors? */
1491         if (!proc_fd_access_allowed(inode))
1492                 goto out;
1493
1494         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1495         if (error)
1496                 goto out;
1497
1498         error = do_proc_readlink(&path, buffer, buflen);
1499         path_put(&path);
1500 out:
1501         return error;
1502 }
1503
1504 const struct inode_operations proc_pid_link_inode_operations = {
1505         .readlink       = proc_pid_readlink,
1506         .follow_link    = proc_pid_follow_link,
1507         .setattr        = proc_setattr,
1508 };
1509
1510
1511 /* building an inode */
1512
1513 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1514 {
1515         struct inode * inode;
1516         struct proc_inode *ei;
1517         const struct cred *cred;
1518
1519         /* We need a new inode */
1520
1521         inode = new_inode(sb);
1522         if (!inode)
1523                 goto out;
1524
1525         /* Common stuff */
1526         ei = PROC_I(inode);
1527         inode->i_ino = get_next_ino();
1528         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1529         inode->i_op = &proc_def_inode_operations;
1530
1531         /*
1532          * grab the reference to task.
1533          */
1534         ei->pid = get_task_pid(task, PIDTYPE_PID);
1535         if (!ei->pid)
1536                 goto out_unlock;
1537
1538         if (task_dumpable(task)) {
1539                 rcu_read_lock();
1540                 cred = __task_cred(task);
1541                 inode->i_uid = cred->euid;
1542                 inode->i_gid = cred->egid;
1543                 rcu_read_unlock();
1544         }
1545         security_task_to_inode(task, inode);
1546
1547 out:
1548         return inode;
1549
1550 out_unlock:
1551         iput(inode);
1552         return NULL;
1553 }
1554
1555 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1556 {
1557         struct inode *inode = dentry->d_inode;
1558         struct task_struct *task;
1559         const struct cred *cred;
1560         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1561
1562         generic_fillattr(inode, stat);
1563
1564         rcu_read_lock();
1565         stat->uid = GLOBAL_ROOT_UID;
1566         stat->gid = GLOBAL_ROOT_GID;
1567         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1568         if (task) {
1569                 if (!has_pid_permissions(pid, task, 2)) {
1570                         rcu_read_unlock();
1571                         /*
1572                          * This doesn't prevent learning whether PID exists,
1573                          * it only makes getattr() consistent with readdir().
1574                          */
1575                         return -ENOENT;
1576                 }
1577                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1578                     task_dumpable(task)) {
1579                         cred = __task_cred(task);
1580                         stat->uid = cred->euid;
1581                         stat->gid = cred->egid;
1582                 }
1583         }
1584         rcu_read_unlock();
1585         return 0;
1586 }
1587
1588 /* dentry stuff */
1589
1590 /*
1591  *      Exceptional case: normally we are not allowed to unhash a busy
1592  * directory. In this case, however, we can do it - no aliasing problems
1593  * due to the way we treat inodes.
1594  *
1595  * Rewrite the inode's ownerships here because the owning task may have
1596  * performed a setuid(), etc.
1597  *
1598  * Before the /proc/pid/status file was created the only way to read
1599  * the effective uid of a /process was to stat /proc/pid.  Reading
1600  * /proc/pid/status is slow enough that procps and other packages
1601  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1602  * made this apply to all per process world readable and executable
1603  * directories.
1604  */
1605 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1606 {
1607         struct inode *inode;
1608         struct task_struct *task;
1609         const struct cred *cred;
1610
1611         if (flags & LOOKUP_RCU)
1612                 return -ECHILD;
1613
1614         inode = dentry->d_inode;
1615         task = get_proc_task(inode);
1616
1617         if (task) {
1618                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1619                     task_dumpable(task)) {
1620                         rcu_read_lock();
1621                         cred = __task_cred(task);
1622                         inode->i_uid = cred->euid;
1623                         inode->i_gid = cred->egid;
1624                         rcu_read_unlock();
1625                 } else {
1626                         inode->i_uid = GLOBAL_ROOT_UID;
1627                         inode->i_gid = GLOBAL_ROOT_GID;
1628                 }
1629                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1630                 security_task_to_inode(task, inode);
1631                 put_task_struct(task);
1632                 return 1;
1633         }
1634         d_drop(dentry);
1635         return 0;
1636 }
1637
1638 static inline bool proc_inode_is_dead(struct inode *inode)
1639 {
1640         return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1641 }
1642
1643 int pid_delete_dentry(const struct dentry *dentry)
1644 {
1645         /* Is the task we represent dead?
1646          * If so, then don't put the dentry on the lru list,
1647          * kill it immediately.
1648          */
1649         return proc_inode_is_dead(dentry->d_inode);
1650 }
1651
1652 const struct dentry_operations pid_dentry_operations =
1653 {
1654         .d_revalidate   = pid_revalidate,
1655         .d_delete       = pid_delete_dentry,
1656 };
1657
1658 /* Lookups */
1659
1660 /*
1661  * Fill a directory entry.
1662  *
1663  * If possible create the dcache entry and derive our inode number and
1664  * file type from dcache entry.
1665  *
1666  * Since all of the proc inode numbers are dynamically generated, the inode
1667  * numbers do not exist until the inode is cache.  This means creating the
1668  * the dcache entry in readdir is necessary to keep the inode numbers
1669  * reported by readdir in sync with the inode numbers reported
1670  * by stat.
1671  */
1672 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1673         const char *name, int len,
1674         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1675 {
1676         struct dentry *child, *dir = file->f_path.dentry;
1677         struct qstr qname = QSTR_INIT(name, len);
1678         struct inode *inode;
1679         unsigned type;
1680         ino_t ino;
1681
1682         child = d_hash_and_lookup(dir, &qname);
1683         if (!child) {
1684                 child = d_alloc(dir, &qname);
1685                 if (!child)
1686                         goto end_instantiate;
1687                 if (instantiate(dir->d_inode, child, task, ptr) < 0) {
1688                         dput(child);
1689                         goto end_instantiate;
1690                 }
1691         }
1692         inode = child->d_inode;
1693         ino = inode->i_ino;
1694         type = inode->i_mode >> 12;
1695         dput(child);
1696         return dir_emit(ctx, name, len, ino, type);
1697
1698 end_instantiate:
1699         return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1700 }
1701
1702 #ifdef CONFIG_CHECKPOINT_RESTORE
1703
1704 /*
1705  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1706  * which represent vma start and end addresses.
1707  */
1708 static int dname_to_vma_addr(struct dentry *dentry,
1709                              unsigned long *start, unsigned long *end)
1710 {
1711         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1712                 return -EINVAL;
1713
1714         return 0;
1715 }
1716
1717 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1718 {
1719         unsigned long vm_start, vm_end;
1720         bool exact_vma_exists = false;
1721         struct mm_struct *mm = NULL;
1722         struct task_struct *task;
1723         const struct cred *cred;
1724         struct inode *inode;
1725         int status = 0;
1726
1727         if (flags & LOOKUP_RCU)
1728                 return -ECHILD;
1729
1730         if (!capable(CAP_SYS_ADMIN)) {
1731                 status = -EPERM;
1732                 goto out_notask;
1733         }
1734
1735         inode = dentry->d_inode;
1736         task = get_proc_task(inode);
1737         if (!task)
1738                 goto out_notask;
1739
1740         mm = mm_access(task, PTRACE_MODE_READ);
1741         if (IS_ERR_OR_NULL(mm))
1742                 goto out;
1743
1744         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1745                 down_read(&mm->mmap_sem);
1746                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1747                 up_read(&mm->mmap_sem);
1748         }
1749
1750         mmput(mm);
1751
1752         if (exact_vma_exists) {
1753                 if (task_dumpable(task)) {
1754                         rcu_read_lock();
1755                         cred = __task_cred(task);
1756                         inode->i_uid = cred->euid;
1757                         inode->i_gid = cred->egid;
1758                         rcu_read_unlock();
1759                 } else {
1760                         inode->i_uid = GLOBAL_ROOT_UID;
1761                         inode->i_gid = GLOBAL_ROOT_GID;
1762                 }
1763                 security_task_to_inode(task, inode);
1764                 status = 1;
1765         }
1766
1767 out:
1768         put_task_struct(task);
1769
1770 out_notask:
1771         if (status <= 0)
1772                 d_drop(dentry);
1773
1774         return status;
1775 }
1776
1777 static const struct dentry_operations tid_map_files_dentry_operations = {
1778         .d_revalidate   = map_files_d_revalidate,
1779         .d_delete       = pid_delete_dentry,
1780 };
1781
1782 static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
1783 {
1784         unsigned long vm_start, vm_end;
1785         struct vm_area_struct *vma;
1786         struct task_struct *task;
1787         struct mm_struct *mm;
1788         int rc;
1789
1790         rc = -ENOENT;
1791         task = get_proc_task(dentry->d_inode);
1792         if (!task)
1793                 goto out;
1794
1795         mm = get_task_mm(task);
1796         put_task_struct(task);
1797         if (!mm)
1798                 goto out;
1799
1800         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1801         if (rc)
1802                 goto out_mmput;
1803
1804         rc = -ENOENT;
1805         down_read(&mm->mmap_sem);
1806         vma = find_exact_vma(mm, vm_start, vm_end);
1807         if (vma && vma->vm_file) {
1808                 *path = vma->vm_file->f_path;
1809                 path_get(path);
1810                 rc = 0;
1811         }
1812         up_read(&mm->mmap_sem);
1813
1814 out_mmput:
1815         mmput(mm);
1816 out:
1817         return rc;
1818 }
1819
1820 struct map_files_info {
1821         fmode_t         mode;
1822         unsigned long   len;
1823         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1824 };
1825
1826 static int
1827 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
1828                            struct task_struct *task, const void *ptr)
1829 {
1830         fmode_t mode = (fmode_t)(unsigned long)ptr;
1831         struct proc_inode *ei;
1832         struct inode *inode;
1833
1834         inode = proc_pid_make_inode(dir->i_sb, task);
1835         if (!inode)
1836                 return -ENOENT;
1837
1838         ei = PROC_I(inode);
1839         ei->op.proc_get_link = proc_map_files_get_link;
1840
1841         inode->i_op = &proc_pid_link_inode_operations;
1842         inode->i_size = 64;
1843         inode->i_mode = S_IFLNK;
1844
1845         if (mode & FMODE_READ)
1846                 inode->i_mode |= S_IRUSR;
1847         if (mode & FMODE_WRITE)
1848                 inode->i_mode |= S_IWUSR;
1849
1850         d_set_d_op(dentry, &tid_map_files_dentry_operations);
1851         d_add(dentry, inode);
1852
1853         return 0;
1854 }
1855
1856 static struct dentry *proc_map_files_lookup(struct inode *dir,
1857                 struct dentry *dentry, unsigned int flags)
1858 {
1859         unsigned long vm_start, vm_end;
1860         struct vm_area_struct *vma;
1861         struct task_struct *task;
1862         int result;
1863         struct mm_struct *mm;
1864
1865         result = -EPERM;
1866         if (!capable(CAP_SYS_ADMIN))
1867                 goto out;
1868
1869         result = -ENOENT;
1870         task = get_proc_task(dir);
1871         if (!task)
1872                 goto out;
1873
1874         result = -EACCES;
1875         if (!ptrace_may_access(task, PTRACE_MODE_READ))
1876                 goto out_put_task;
1877
1878         result = -ENOENT;
1879         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
1880                 goto out_put_task;
1881
1882         mm = get_task_mm(task);
1883         if (!mm)
1884                 goto out_put_task;
1885
1886         down_read(&mm->mmap_sem);
1887         vma = find_exact_vma(mm, vm_start, vm_end);
1888         if (!vma)
1889                 goto out_no_vma;
1890
1891         if (vma->vm_file)
1892                 result = proc_map_files_instantiate(dir, dentry, task,
1893                                 (void *)(unsigned long)vma->vm_file->f_mode);
1894
1895 out_no_vma:
1896         up_read(&mm->mmap_sem);
1897         mmput(mm);
1898 out_put_task:
1899         put_task_struct(task);
1900 out:
1901         return ERR_PTR(result);
1902 }
1903
1904 static const struct inode_operations proc_map_files_inode_operations = {
1905         .lookup         = proc_map_files_lookup,
1906         .permission     = proc_fd_permission,
1907         .setattr        = proc_setattr,
1908 };
1909
1910 static int
1911 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
1912 {
1913         struct vm_area_struct *vma;
1914         struct task_struct *task;
1915         struct mm_struct *mm;
1916         unsigned long nr_files, pos, i;
1917         struct flex_array *fa = NULL;
1918         struct map_files_info info;
1919         struct map_files_info *p;
1920         int ret;
1921
1922         ret = -EPERM;
1923         if (!capable(CAP_SYS_ADMIN))
1924                 goto out;
1925
1926         ret = -ENOENT;
1927         task = get_proc_task(file_inode(file));
1928         if (!task)
1929                 goto out;
1930
1931         ret = -EACCES;
1932         if (!ptrace_may_access(task, PTRACE_MODE_READ))
1933                 goto out_put_task;
1934
1935         ret = 0;
1936         if (!dir_emit_dots(file, ctx))
1937                 goto out_put_task;
1938
1939         mm = get_task_mm(task);
1940         if (!mm)
1941                 goto out_put_task;
1942         down_read(&mm->mmap_sem);
1943
1944         nr_files = 0;
1945
1946         /*
1947          * We need two passes here:
1948          *
1949          *  1) Collect vmas of mapped files with mmap_sem taken
1950          *  2) Release mmap_sem and instantiate entries
1951          *
1952          * otherwise we get lockdep complained, since filldir()
1953          * routine might require mmap_sem taken in might_fault().
1954          */
1955
1956         for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
1957                 if (vma->vm_file && ++pos > ctx->pos)
1958                         nr_files++;
1959         }
1960
1961         if (nr_files) {
1962                 fa = flex_array_alloc(sizeof(info), nr_files,
1963                                         GFP_KERNEL);
1964                 if (!fa || flex_array_prealloc(fa, 0, nr_files,
1965                                                 GFP_KERNEL)) {
1966                         ret = -ENOMEM;
1967                         if (fa)
1968                                 flex_array_free(fa);
1969                         up_read(&mm->mmap_sem);
1970                         mmput(mm);
1971                         goto out_put_task;
1972                 }
1973                 for (i = 0, vma = mm->mmap, pos = 2; vma;
1974                                 vma = vma->vm_next) {
1975                         if (!vma->vm_file)
1976                                 continue;
1977                         if (++pos <= ctx->pos)
1978                                 continue;
1979
1980                         info.mode = vma->vm_file->f_mode;
1981                         info.len = snprintf(info.name,
1982                                         sizeof(info.name), "%lx-%lx",
1983                                         vma->vm_start, vma->vm_end);
1984                         if (flex_array_put(fa, i++, &info, GFP_KERNEL))
1985                                 BUG();
1986                 }
1987         }
1988         up_read(&mm->mmap_sem);
1989
1990         for (i = 0; i < nr_files; i++) {
1991                 p = flex_array_get(fa, i);
1992                 if (!proc_fill_cache(file, ctx,
1993                                       p->name, p->len,
1994                                       proc_map_files_instantiate,
1995                                       task,
1996                                       (void *)(unsigned long)p->mode))
1997                         break;
1998                 ctx->pos++;
1999         }
2000         if (fa)
2001                 flex_array_free(fa);
2002         mmput(mm);
2003
2004 out_put_task:
2005         put_task_struct(task);
2006 out:
2007         return ret;
2008 }
2009
2010 static const struct file_operations proc_map_files_operations = {
2011         .read           = generic_read_dir,
2012         .iterate        = proc_map_files_readdir,
2013         .llseek         = default_llseek,
2014 };
2015
2016 struct timers_private {
2017         struct pid *pid;
2018         struct task_struct *task;
2019         struct sighand_struct *sighand;
2020         struct pid_namespace *ns;
2021         unsigned long flags;
2022 };
2023
2024 static void *timers_start(struct seq_file *m, loff_t *pos)
2025 {
2026         struct timers_private *tp = m->private;
2027
2028         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2029         if (!tp->task)
2030                 return ERR_PTR(-ESRCH);
2031
2032         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2033         if (!tp->sighand)
2034                 return ERR_PTR(-ESRCH);
2035
2036         return seq_list_start(&tp->task->signal->posix_timers, *pos);
2037 }
2038
2039 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2040 {
2041         struct timers_private *tp = m->private;
2042         return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2043 }
2044
2045 static void timers_stop(struct seq_file *m, void *v)
2046 {
2047         struct timers_private *tp = m->private;
2048
2049         if (tp->sighand) {
2050                 unlock_task_sighand(tp->task, &tp->flags);
2051                 tp->sighand = NULL;
2052         }
2053
2054         if (tp->task) {
2055                 put_task_struct(tp->task);
2056                 tp->task = NULL;
2057         }
2058 }
2059
2060 static int show_timer(struct seq_file *m, void *v)
2061 {
2062         struct k_itimer *timer;
2063         struct timers_private *tp = m->private;
2064         int notify;
2065         static const char * const nstr[] = {
2066                 [SIGEV_SIGNAL] = "signal",
2067                 [SIGEV_NONE] = "none",
2068                 [SIGEV_THREAD] = "thread",
2069         };
2070
2071         timer = list_entry((struct list_head *)v, struct k_itimer, list);
2072         notify = timer->it_sigev_notify;
2073
2074         seq_printf(m, "ID: %d\n", timer->it_id);
2075         seq_printf(m, "signal: %d/%p\n", timer->sigq->info.si_signo,
2076                         timer->sigq->info.si_value.sival_ptr);
2077         seq_printf(m, "notify: %s/%s.%d\n",
2078                 nstr[notify & ~SIGEV_THREAD_ID],
2079                 (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2080                 pid_nr_ns(timer->it_pid, tp->ns));
2081         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2082
2083         return 0;
2084 }
2085
2086 static const struct seq_operations proc_timers_seq_ops = {
2087         .start  = timers_start,
2088         .next   = timers_next,
2089         .stop   = timers_stop,
2090         .show   = show_timer,
2091 };
2092
2093 static int proc_timers_open(struct inode *inode, struct file *file)
2094 {
2095         struct timers_private *tp;
2096
2097         tp = __seq_open_private(file, &proc_timers_seq_ops,
2098                         sizeof(struct timers_private));
2099         if (!tp)
2100                 return -ENOMEM;
2101
2102         tp->pid = proc_pid(inode);
2103         tp->ns = inode->i_sb->s_fs_info;
2104         return 0;
2105 }
2106
2107 static const struct file_operations proc_timers_operations = {
2108         .open           = proc_timers_open,
2109         .read           = seq_read,
2110         .llseek         = seq_lseek,
2111         .release        = seq_release_private,
2112 };
2113 #endif /* CONFIG_CHECKPOINT_RESTORE */
2114
2115 static int proc_pident_instantiate(struct inode *dir,
2116         struct dentry *dentry, struct task_struct *task, const void *ptr)
2117 {
2118         const struct pid_entry *p = ptr;
2119         struct inode *inode;
2120         struct proc_inode *ei;
2121
2122         inode = proc_pid_make_inode(dir->i_sb, task);
2123         if (!inode)
2124                 goto out;
2125
2126         ei = PROC_I(inode);
2127         inode->i_mode = p->mode;
2128         if (S_ISDIR(inode->i_mode))
2129                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2130         if (p->iop)
2131                 inode->i_op = p->iop;
2132         if (p->fop)
2133                 inode->i_fop = p->fop;
2134         ei->op = p->op;
2135         d_set_d_op(dentry, &pid_dentry_operations);
2136         d_add(dentry, inode);
2137         /* Close the race of the process dying before we return the dentry */
2138         if (pid_revalidate(dentry, 0))
2139                 return 0;
2140 out:
2141         return -ENOENT;
2142 }
2143
2144 static struct dentry *proc_pident_lookup(struct inode *dir,
2145                                          struct dentry *dentry,
2146                                          const struct pid_entry *ents,
2147                                          unsigned int nents)
2148 {
2149         int error;
2150         struct task_struct *task = get_proc_task(dir);
2151         const struct pid_entry *p, *last;
2152
2153         error = -ENOENT;
2154
2155         if (!task)
2156                 goto out_no_task;
2157
2158         /*
2159          * Yes, it does not scale. And it should not. Don't add
2160          * new entries into /proc/<tgid>/ without very good reasons.
2161          */
2162         last = &ents[nents - 1];
2163         for (p = ents; p <= last; p++) {
2164                 if (p->len != dentry->d_name.len)
2165                         continue;
2166                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2167                         break;
2168         }
2169         if (p > last)
2170                 goto out;
2171
2172         error = proc_pident_instantiate(dir, dentry, task, p);
2173 out:
2174         put_task_struct(task);
2175 out_no_task:
2176         return ERR_PTR(error);
2177 }
2178
2179 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2180                 const struct pid_entry *ents, unsigned int nents)
2181 {
2182         struct task_struct *task = get_proc_task(file_inode(file));
2183         const struct pid_entry *p;
2184
2185         if (!task)
2186                 return -ENOENT;
2187
2188         if (!dir_emit_dots(file, ctx))
2189                 goto out;
2190
2191         if (ctx->pos >= nents + 2)
2192                 goto out;
2193
2194         for (p = ents + (ctx->pos - 2); p <= ents + nents - 1; p++) {
2195                 if (!proc_fill_cache(file, ctx, p->name, p->len,
2196                                 proc_pident_instantiate, task, p))
2197                         break;
2198                 ctx->pos++;
2199         }
2200 out:
2201         put_task_struct(task);
2202         return 0;
2203 }
2204
2205 #ifdef CONFIG_SECURITY
2206 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2207                                   size_t count, loff_t *ppos)
2208 {
2209         struct inode * inode = file_inode(file);
2210         char *p = NULL;
2211         ssize_t length;
2212         struct task_struct *task = get_proc_task(inode);
2213
2214         if (!task)
2215                 return -ESRCH;
2216
2217         length = security_getprocattr(task,
2218                                       (char*)file->f_path.dentry->d_name.name,
2219                                       &p);
2220         put_task_struct(task);
2221         if (length > 0)
2222                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2223         kfree(p);
2224         return length;
2225 }
2226
2227 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2228                                    size_t count, loff_t *ppos)
2229 {
2230         struct inode * inode = file_inode(file);
2231         char *page;
2232         ssize_t length;
2233         struct task_struct *task = get_proc_task(inode);
2234
2235         length = -ESRCH;
2236         if (!task)
2237                 goto out_no_task;
2238         if (count > PAGE_SIZE)
2239                 count = PAGE_SIZE;
2240
2241         /* No partial writes. */
2242         length = -EINVAL;
2243         if (*ppos != 0)
2244                 goto out;
2245
2246         length = -ENOMEM;
2247         page = (char*)__get_free_page(GFP_TEMPORARY);
2248         if (!page)
2249                 goto out;
2250
2251         length = -EFAULT;
2252         if (copy_from_user(page, buf, count))
2253                 goto out_free;
2254
2255         /* Guard against adverse ptrace interaction */
2256         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2257         if (length < 0)
2258                 goto out_free;
2259
2260         length = security_setprocattr(task,
2261                                       (char*)file->f_path.dentry->d_name.name,
2262                                       (void*)page, count);
2263         mutex_unlock(&task->signal->cred_guard_mutex);
2264 out_free:
2265         free_page((unsigned long) page);
2266 out:
2267         put_task_struct(task);
2268 out_no_task:
2269         return length;
2270 }
2271
2272 static const struct file_operations proc_pid_attr_operations = {
2273         .read           = proc_pid_attr_read,
2274         .write          = proc_pid_attr_write,
2275         .llseek         = generic_file_llseek,
2276 };
2277
2278 static const struct pid_entry attr_dir_stuff[] = {
2279         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2280         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2281         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2282         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2283         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2284         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2285 };
2286
2287 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2288 {
2289         return proc_pident_readdir(file, ctx,
2290                                    attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2291 }
2292
2293 static const struct file_operations proc_attr_dir_operations = {
2294         .read           = generic_read_dir,
2295         .iterate        = proc_attr_dir_readdir,
2296         .llseek         = default_llseek,
2297 };
2298
2299 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2300                                 struct dentry *dentry, unsigned int flags)
2301 {
2302         return proc_pident_lookup(dir, dentry,
2303                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2304 }
2305
2306 static const struct inode_operations proc_attr_dir_inode_operations = {
2307         .lookup         = proc_attr_dir_lookup,
2308         .getattr        = pid_getattr,
2309         .setattr        = proc_setattr,
2310 };
2311
2312 #endif
2313
2314 #ifdef CONFIG_ELF_CORE
2315 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2316                                          size_t count, loff_t *ppos)
2317 {
2318         struct task_struct *task = get_proc_task(file_inode(file));
2319         struct mm_struct *mm;
2320         char buffer[PROC_NUMBUF];
2321         size_t len;
2322         int ret;
2323
2324         if (!task)
2325                 return -ESRCH;
2326
2327         ret = 0;
2328         mm = get_task_mm(task);
2329         if (mm) {
2330                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2331                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2332                                 MMF_DUMP_FILTER_SHIFT));
2333                 mmput(mm);
2334                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2335         }
2336
2337         put_task_struct(task);
2338
2339         return ret;
2340 }
2341
2342 static ssize_t proc_coredump_filter_write(struct file *file,
2343                                           const char __user *buf,
2344                                           size_t count,
2345                                           loff_t *ppos)
2346 {
2347         struct task_struct *task;
2348         struct mm_struct *mm;
2349         char buffer[PROC_NUMBUF], *end;
2350         unsigned int val;
2351         int ret;
2352         int i;
2353         unsigned long mask;
2354
2355         ret = -EFAULT;
2356         memset(buffer, 0, sizeof(buffer));
2357         if (count > sizeof(buffer) - 1)
2358                 count = sizeof(buffer) - 1;
2359         if (copy_from_user(buffer, buf, count))
2360                 goto out_no_task;
2361
2362         ret = -EINVAL;
2363         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2364         if (*end == '\n')
2365                 end++;
2366         if (end - buffer == 0)
2367                 goto out_no_task;
2368
2369         ret = -ESRCH;
2370         task = get_proc_task(file_inode(file));
2371         if (!task)
2372                 goto out_no_task;
2373
2374         ret = end - buffer;
2375         mm = get_task_mm(task);
2376         if (!mm)
2377                 goto out_no_mm;
2378
2379         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2380                 if (val & mask)
2381                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2382                 else
2383                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2384         }
2385
2386         mmput(mm);
2387  out_no_mm:
2388         put_task_struct(task);
2389  out_no_task:
2390         return ret;
2391 }
2392
2393 static const struct file_operations proc_coredump_filter_operations = {
2394         .read           = proc_coredump_filter_read,
2395         .write          = proc_coredump_filter_write,
2396         .llseek         = generic_file_llseek,
2397 };
2398 #endif
2399
2400 #ifdef CONFIG_TASK_IO_ACCOUNTING
2401 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2402 {
2403         struct task_io_accounting acct = task->ioac;
2404         unsigned long flags;
2405         int result;
2406
2407         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2408         if (result)
2409                 return result;
2410
2411         if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2412                 result = -EACCES;
2413                 goto out_unlock;
2414         }
2415
2416         if (whole && lock_task_sighand(task, &flags)) {
2417                 struct task_struct *t = task;
2418
2419                 task_io_accounting_add(&acct, &task->signal->ioac);
2420                 while_each_thread(task, t)
2421                         task_io_accounting_add(&acct, &t->ioac);
2422
2423                 unlock_task_sighand(task, &flags);
2424         }
2425         result = sprintf(buffer,
2426                         "rchar: %llu\n"
2427                         "wchar: %llu\n"
2428                         "syscr: %llu\n"
2429                         "syscw: %llu\n"
2430                         "read_bytes: %llu\n"
2431                         "write_bytes: %llu\n"
2432                         "cancelled_write_bytes: %llu\n",
2433                         (unsigned long long)acct.rchar,
2434                         (unsigned long long)acct.wchar,
2435                         (unsigned long long)acct.syscr,
2436                         (unsigned long long)acct.syscw,
2437                         (unsigned long long)acct.read_bytes,
2438                         (unsigned long long)acct.write_bytes,
2439                         (unsigned long long)acct.cancelled_write_bytes);
2440 out_unlock:
2441         mutex_unlock(&task->signal->cred_guard_mutex);
2442         return result;
2443 }
2444
2445 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2446 {
2447         return do_io_accounting(task, buffer, 0);
2448 }
2449
2450 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2451 {
2452         return do_io_accounting(task, buffer, 1);
2453 }
2454 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2455
2456 #ifdef CONFIG_USER_NS
2457 static int proc_id_map_open(struct inode *inode, struct file *file,
2458         const struct seq_operations *seq_ops)
2459 {
2460         struct user_namespace *ns = NULL;
2461         struct task_struct *task;
2462         struct seq_file *seq;
2463         int ret = -EINVAL;
2464
2465         task = get_proc_task(inode);
2466         if (task) {
2467                 rcu_read_lock();
2468                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2469                 rcu_read_unlock();
2470                 put_task_struct(task);
2471         }
2472         if (!ns)
2473                 goto err;
2474
2475         ret = seq_open(file, seq_ops);
2476         if (ret)
2477                 goto err_put_ns;
2478
2479         seq = file->private_data;
2480         seq->private = ns;
2481
2482         return 0;
2483 err_put_ns:
2484         put_user_ns(ns);
2485 err:
2486         return ret;
2487 }
2488
2489 static int proc_id_map_release(struct inode *inode, struct file *file)
2490 {
2491         struct seq_file *seq = file->private_data;
2492         struct user_namespace *ns = seq->private;
2493         put_user_ns(ns);
2494         return seq_release(inode, file);
2495 }
2496
2497 static int proc_uid_map_open(struct inode *inode, struct file *file)
2498 {
2499         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2500 }
2501
2502 static int proc_gid_map_open(struct inode *inode, struct file *file)
2503 {
2504         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2505 }
2506
2507 static int proc_projid_map_open(struct inode *inode, struct file *file)
2508 {
2509         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2510 }
2511
2512 static const struct file_operations proc_uid_map_operations = {
2513         .open           = proc_uid_map_open,
2514         .write          = proc_uid_map_write,
2515         .read           = seq_read,
2516         .llseek         = seq_lseek,
2517         .release        = proc_id_map_release,
2518 };
2519
2520 static const struct file_operations proc_gid_map_operations = {
2521         .open           = proc_gid_map_open,
2522         .write          = proc_gid_map_write,
2523         .read           = seq_read,
2524         .llseek         = seq_lseek,
2525         .release        = proc_id_map_release,
2526 };
2527
2528 static const struct file_operations proc_projid_map_operations = {
2529         .open           = proc_projid_map_open,
2530         .write          = proc_projid_map_write,
2531         .read           = seq_read,
2532         .llseek         = seq_lseek,
2533         .release        = proc_id_map_release,
2534 };
2535 #endif /* CONFIG_USER_NS */
2536
2537 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2538                                 struct pid *pid, struct task_struct *task)
2539 {
2540         int err = lock_trace(task);
2541         if (!err) {
2542                 seq_printf(m, "%08x\n", task->personality);
2543                 unlock_trace(task);
2544         }
2545         return err;
2546 }
2547
2548 /*
2549  * Thread groups
2550  */
2551 static const struct file_operations proc_task_operations;
2552 static const struct inode_operations proc_task_inode_operations;
2553
2554 static const struct pid_entry tgid_base_stuff[] = {
2555         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2556         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2557 #ifdef CONFIG_CHECKPOINT_RESTORE
2558         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2559 #endif
2560         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2561         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2562 #ifdef CONFIG_NET
2563         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2564 #endif
2565         REG("environ",    S_IRUSR, proc_environ_operations),
2566         ONE("auxv",       S_IRUSR, proc_pid_auxv),
2567         ONE("status",     S_IRUGO, proc_pid_status),
2568         ONE("personality", S_IRUSR, proc_pid_personality),
2569         ONE("limits",     S_IRUGO, proc_pid_limits),
2570 #ifdef CONFIG_SCHED_DEBUG
2571         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2572 #endif
2573 #ifdef CONFIG_SCHED_AUTOGROUP
2574         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2575 #endif
2576         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2577 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2578         ONE("syscall",    S_IRUSR, proc_pid_syscall),
2579 #endif
2580         ONE("cmdline",    S_IRUGO, proc_pid_cmdline),
2581         ONE("stat",       S_IRUGO, proc_tgid_stat),
2582         ONE("statm",      S_IRUGO, proc_pid_statm),
2583         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2584 #ifdef CONFIG_NUMA
2585         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2586 #endif
2587         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2588         LNK("cwd",        proc_cwd_link),
2589         LNK("root",       proc_root_link),
2590         LNK("exe",        proc_exe_link),
2591         REG("mounts",     S_IRUGO, proc_mounts_operations),
2592         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2593         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2594 #ifdef CONFIG_PROC_PAGE_MONITOR
2595         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2596         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2597         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2598 #endif
2599 #ifdef CONFIG_SECURITY
2600         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2601 #endif
2602 #ifdef CONFIG_KALLSYMS
2603         ONE("wchan",      S_IRUGO, proc_pid_wchan),
2604 #endif
2605 #ifdef CONFIG_STACKTRACE
2606         ONE("stack",      S_IRUSR, proc_pid_stack),
2607 #endif
2608 #ifdef CONFIG_SCHEDSTATS
2609         ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
2610 #endif
2611 #ifdef CONFIG_LATENCYTOP
2612         REG("latency",  S_IRUGO, proc_lstats_operations),
2613 #endif
2614 #ifdef CONFIG_PROC_PID_CPUSET
2615         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2616 #endif
2617 #ifdef CONFIG_CGROUPS
2618         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2619 #endif
2620         ONE("oom_score",  S_IRUGO, proc_oom_score),
2621         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2622         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2623 #ifdef CONFIG_AUDITSYSCALL
2624         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2625         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2626 #endif
2627 #ifdef CONFIG_FAULT_INJECTION
2628         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2629 #endif
2630 #ifdef CONFIG_ELF_CORE
2631         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2632 #endif
2633 #ifdef CONFIG_TASK_IO_ACCOUNTING
2634         INF("io",       S_IRUSR, proc_tgid_io_accounting),
2635 #endif
2636 #ifdef CONFIG_HARDWALL
2637         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
2638 #endif
2639 #ifdef CONFIG_USER_NS
2640         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2641         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2642         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2643 #endif
2644 #ifdef CONFIG_CHECKPOINT_RESTORE
2645         REG("timers",     S_IRUGO, proc_timers_operations),
2646 #endif
2647 };
2648
2649 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2650 {
2651         return proc_pident_readdir(file, ctx,
2652                                    tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2653 }
2654
2655 static const struct file_operations proc_tgid_base_operations = {
2656         .read           = generic_read_dir,
2657         .iterate        = proc_tgid_base_readdir,
2658         .llseek         = default_llseek,
2659 };
2660
2661 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2662 {
2663         return proc_pident_lookup(dir, dentry,
2664                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2665 }
2666
2667 static const struct inode_operations proc_tgid_base_inode_operations = {
2668         .lookup         = proc_tgid_base_lookup,
2669         .getattr        = pid_getattr,
2670         .setattr        = proc_setattr,
2671         .permission     = proc_pid_permission,
2672 };
2673
2674 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2675 {
2676         struct dentry *dentry, *leader, *dir;
2677         char buf[PROC_NUMBUF];
2678         struct qstr name;
2679
2680         name.name = buf;
2681         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2682         /* no ->d_hash() rejects on procfs */
2683         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2684         if (dentry) {
2685                 shrink_dcache_parent(dentry);
2686                 d_drop(dentry);
2687                 dput(dentry);
2688         }
2689
2690         name.name = buf;
2691         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2692         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2693         if (!leader)
2694                 goto out;
2695
2696         name.name = "task";
2697         name.len = strlen(name.name);
2698         dir = d_hash_and_lookup(leader, &name);
2699         if (!dir)
2700                 goto out_put_leader;
2701
2702         name.name = buf;
2703         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2704         dentry = d_hash_and_lookup(dir, &name);
2705         if (dentry) {
2706                 shrink_dcache_parent(dentry);
2707                 d_drop(dentry);
2708                 dput(dentry);
2709         }
2710
2711         dput(dir);
2712 out_put_leader:
2713         dput(leader);
2714 out:
2715         return;
2716 }
2717
2718 /**
2719  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2720  * @task: task that should be flushed.
2721  *
2722  * When flushing dentries from proc, one needs to flush them from global
2723  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2724  * in. This call is supposed to do all of this job.
2725  *
2726  * Looks in the dcache for
2727  * /proc/@pid
2728  * /proc/@tgid/task/@pid
2729  * if either directory is present flushes it and all of it'ts children
2730  * from the dcache.
2731  *
2732  * It is safe and reasonable to cache /proc entries for a task until
2733  * that task exits.  After that they just clog up the dcache with
2734  * useless entries, possibly causing useful dcache entries to be
2735  * flushed instead.  This routine is proved to flush those useless
2736  * dcache entries at process exit time.
2737  *
2738  * NOTE: This routine is just an optimization so it does not guarantee
2739  *       that no dcache entries will exist at process exit time it
2740  *       just makes it very unlikely that any will persist.
2741  */
2742
2743 void proc_flush_task(struct task_struct *task)
2744 {
2745         int i;
2746         struct pid *pid, *tgid;
2747         struct upid *upid;
2748
2749         pid = task_pid(task);
2750         tgid = task_tgid(task);
2751
2752         for (i = 0; i <= pid->level; i++) {
2753                 upid = &pid->numbers[i];
2754                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2755                                         tgid->numbers[i].nr);
2756         }
2757 }
2758
2759 static int proc_pid_instantiate(struct inode *dir,
2760                                    struct dentry * dentry,
2761                                    struct task_struct *task, const void *ptr)
2762 {
2763         struct inode *inode;
2764
2765         inode = proc_pid_make_inode(dir->i_sb, task);
2766         if (!inode)
2767                 goto out;
2768
2769         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2770         inode->i_op = &proc_tgid_base_inode_operations;
2771         inode->i_fop = &proc_tgid_base_operations;
2772         inode->i_flags|=S_IMMUTABLE;
2773
2774         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
2775                                                   ARRAY_SIZE(tgid_base_stuff)));
2776
2777         d_set_d_op(dentry, &pid_dentry_operations);
2778
2779         d_add(dentry, inode);
2780         /* Close the race of the process dying before we return the dentry */
2781         if (pid_revalidate(dentry, 0))
2782                 return 0;
2783 out:
2784         return -ENOENT;
2785 }
2786
2787 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
2788 {
2789         int result = -ENOENT;
2790         struct task_struct *task;
2791         unsigned tgid;
2792         struct pid_namespace *ns;
2793
2794         tgid = name_to_int(&dentry->d_name);
2795         if (tgid == ~0U)
2796                 goto out;
2797
2798         ns = dentry->d_sb->s_fs_info;
2799         rcu_read_lock();
2800         task = find_task_by_pid_ns(tgid, ns);
2801         if (task)
2802                 get_task_struct(task);
2803         rcu_read_unlock();
2804         if (!task)
2805                 goto out;
2806
2807         result = proc_pid_instantiate(dir, dentry, task, NULL);
2808         put_task_struct(task);
2809 out:
2810         return ERR_PTR(result);
2811 }
2812
2813 /*
2814  * Find the first task with tgid >= tgid
2815  *
2816  */
2817 struct tgid_iter {
2818         unsigned int tgid;
2819         struct task_struct *task;
2820 };
2821 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2822 {
2823         struct pid *pid;
2824
2825         if (iter.task)
2826                 put_task_struct(iter.task);
2827         rcu_read_lock();
2828 retry:
2829         iter.task = NULL;
2830         pid = find_ge_pid(iter.tgid, ns);
2831         if (pid) {
2832                 iter.tgid = pid_nr_ns(pid, ns);
2833                 iter.task = pid_task(pid, PIDTYPE_PID);
2834                 /* What we to know is if the pid we have find is the
2835                  * pid of a thread_group_leader.  Testing for task
2836                  * being a thread_group_leader is the obvious thing
2837                  * todo but there is a window when it fails, due to
2838                  * the pid transfer logic in de_thread.
2839                  *
2840                  * So we perform the straight forward test of seeing
2841                  * if the pid we have found is the pid of a thread
2842                  * group leader, and don't worry if the task we have
2843                  * found doesn't happen to be a thread group leader.
2844                  * As we don't care in the case of readdir.
2845                  */
2846                 if (!iter.task || !has_group_leader_pid(iter.task)) {
2847                         iter.tgid += 1;
2848                         goto retry;
2849                 }
2850                 get_task_struct(iter.task);
2851         }
2852         rcu_read_unlock();
2853         return iter;
2854 }
2855
2856 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 1)
2857
2858 /* for the /proc/ directory itself, after non-process stuff has been done */
2859 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
2860 {
2861         struct tgid_iter iter;
2862         struct pid_namespace *ns = file->f_dentry->d_sb->s_fs_info;
2863         loff_t pos = ctx->pos;
2864
2865         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
2866                 return 0;
2867
2868         if (pos == TGID_OFFSET - 1) {
2869                 struct inode *inode = ns->proc_self->d_inode;
2870                 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
2871                         return 0;
2872                 iter.tgid = 0;
2873         } else {
2874                 iter.tgid = pos - TGID_OFFSET;
2875         }
2876         iter.task = NULL;
2877         for (iter = next_tgid(ns, iter);
2878              iter.task;
2879              iter.tgid += 1, iter = next_tgid(ns, iter)) {
2880                 char name[PROC_NUMBUF];
2881                 int len;
2882                 if (!has_pid_permissions(ns, iter.task, 2))
2883                         continue;
2884
2885                 len = snprintf(name, sizeof(name), "%d", iter.tgid);
2886                 ctx->pos = iter.tgid + TGID_OFFSET;
2887                 if (!proc_fill_cache(file, ctx, name, len,
2888                                      proc_pid_instantiate, iter.task, NULL)) {
2889                         put_task_struct(iter.task);
2890                         return 0;
2891                 }
2892         }
2893         ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
2894         return 0;
2895 }
2896
2897 /*
2898  * Tasks
2899  */
2900 static const struct pid_entry tid_base_stuff[] = {
2901         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2902         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2903         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2904         REG("environ",   S_IRUSR, proc_environ_operations),
2905         ONE("auxv",      S_IRUSR, proc_pid_auxv),
2906         ONE("status",    S_IRUGO, proc_pid_status),
2907         ONE("personality", S_IRUSR, proc_pid_personality),
2908         ONE("limits",    S_IRUGO, proc_pid_limits),
2909 #ifdef CONFIG_SCHED_DEBUG
2910         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2911 #endif
2912         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2913 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2914         ONE("syscall",   S_IRUSR, proc_pid_syscall),
2915 #endif
2916         ONE("cmdline",   S_IRUGO, proc_pid_cmdline),
2917         ONE("stat",      S_IRUGO, proc_tid_stat),
2918         ONE("statm",     S_IRUGO, proc_pid_statm),
2919         REG("maps",      S_IRUGO, proc_tid_maps_operations),
2920 #ifdef CONFIG_CHECKPOINT_RESTORE
2921         REG("children",  S_IRUGO, proc_tid_children_operations),
2922 #endif
2923 #ifdef CONFIG_NUMA
2924         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
2925 #endif
2926         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
2927         LNK("cwd",       proc_cwd_link),
2928         LNK("root",      proc_root_link),
2929         LNK("exe",       proc_exe_link),
2930         REG("mounts",    S_IRUGO, proc_mounts_operations),
2931         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2932 #ifdef CONFIG_PROC_PAGE_MONITOR
2933         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2934         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
2935         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2936 #endif
2937 #ifdef CONFIG_SECURITY
2938         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2939 #endif
2940 #ifdef CONFIG_KALLSYMS
2941         ONE("wchan",     S_IRUGO, proc_pid_wchan),
2942 #endif
2943 #ifdef CONFIG_STACKTRACE
2944         ONE("stack",      S_IRUSR, proc_pid_stack),
2945 #endif
2946 #ifdef CONFIG_SCHEDSTATS
2947         ONE("schedstat", S_IRUGO, proc_pid_schedstat),
2948 #endif
2949 #ifdef CONFIG_LATENCYTOP
2950         REG("latency",  S_IRUGO, proc_lstats_operations),
2951 #endif
2952 #ifdef CONFIG_PROC_PID_CPUSET
2953         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
2954 #endif
2955 #ifdef CONFIG_CGROUPS
2956         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2957 #endif
2958         ONE("oom_score", S_IRUGO, proc_oom_score),
2959         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2960         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2961 #ifdef CONFIG_AUDITSYSCALL
2962         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
2963         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2964 #endif
2965 #ifdef CONFIG_FAULT_INJECTION
2966         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2967 #endif
2968 #ifdef CONFIG_TASK_IO_ACCOUNTING
2969         INF("io",       S_IRUSR, proc_tid_io_accounting),
2970 #endif
2971 #ifdef CONFIG_HARDWALL
2972         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
2973 #endif
2974 #ifdef CONFIG_USER_NS
2975         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2976         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2977         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2978 #endif
2979 };
2980
2981 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
2982 {
2983         return proc_pident_readdir(file, ctx,
2984                                    tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
2985 }
2986
2987 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2988 {
2989         return proc_pident_lookup(dir, dentry,
2990                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
2991 }
2992
2993 static const struct file_operations proc_tid_base_operations = {
2994         .read           = generic_read_dir,
2995         .iterate        = proc_tid_base_readdir,
2996         .llseek         = default_llseek,
2997 };
2998
2999 static const struct inode_operations proc_tid_base_inode_operations = {
3000         .lookup         = proc_tid_base_lookup,
3001         .getattr        = pid_getattr,
3002         .setattr        = proc_setattr,
3003 };
3004
3005 static int proc_task_instantiate(struct inode *dir,
3006         struct dentry *dentry, struct task_struct *task, const void *ptr)
3007 {
3008         struct inode *inode;
3009         inode = proc_pid_make_inode(dir->i_sb, task);
3010
3011         if (!inode)
3012                 goto out;
3013         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3014         inode->i_op = &proc_tid_base_inode_operations;
3015         inode->i_fop = &proc_tid_base_operations;
3016         inode->i_flags|=S_IMMUTABLE;
3017
3018         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3019                                                   ARRAY_SIZE(tid_base_stuff)));
3020
3021         d_set_d_op(dentry, &pid_dentry_operations);
3022
3023         d_add(dentry, inode);
3024         /* Close the race of the process dying before we return the dentry */
3025         if (pid_revalidate(dentry, 0))
3026                 return 0;
3027 out:
3028         return -ENOENT;
3029 }
3030
3031 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3032 {
3033         int result = -ENOENT;
3034         struct task_struct *task;
3035         struct task_struct *leader = get_proc_task(dir);
3036         unsigned tid;
3037         struct pid_namespace *ns;
3038
3039         if (!leader)
3040                 goto out_no_task;
3041
3042         tid = name_to_int(&dentry->d_name);
3043         if (tid == ~0U)
3044                 goto out;
3045
3046         ns = dentry->d_sb->s_fs_info;
3047         rcu_read_lock();
3048         task = find_task_by_pid_ns(tid, ns);
3049         if (task)
3050                 get_task_struct(task);
3051         rcu_read_unlock();
3052         if (!task)
3053                 goto out;
3054         if (!same_thread_group(leader, task))
3055                 goto out_drop_task;
3056
3057         result = proc_task_instantiate(dir, dentry, task, NULL);
3058 out_drop_task:
3059         put_task_struct(task);
3060 out:
3061         put_task_struct(leader);
3062 out_no_task:
3063         return ERR_PTR(result);
3064 }
3065
3066 /*
3067  * Find the first tid of a thread group to return to user space.
3068  *
3069  * Usually this is just the thread group leader, but if the users
3070  * buffer was too small or there was a seek into the middle of the
3071  * directory we have more work todo.
3072  *
3073  * In the case of a short read we start with find_task_by_pid.
3074  *
3075  * In the case of a seek we start with the leader and walk nr
3076  * threads past it.
3077  */
3078 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3079                                         struct pid_namespace *ns)
3080 {
3081         struct task_struct *pos, *task;
3082         unsigned long nr = f_pos;
3083
3084         if (nr != f_pos)        /* 32bit overflow? */
3085                 return NULL;
3086
3087         rcu_read_lock();
3088         task = pid_task(pid, PIDTYPE_PID);
3089         if (!task)
3090                 goto fail;
3091
3092         /* Attempt to start with the tid of a thread */
3093         if (tid && nr) {
3094                 pos = find_task_by_pid_ns(tid, ns);
3095                 if (pos && same_thread_group(pos, task))
3096                         goto found;
3097         }
3098
3099         /* If nr exceeds the number of threads there is nothing todo */
3100         if (nr >= get_nr_threads(task))
3101                 goto fail;
3102
3103         /* If we haven't found our starting place yet start
3104          * with the leader and walk nr threads forward.
3105          */
3106         pos = task = task->group_leader;
3107         do {
3108                 if (!nr--)
3109                         goto found;
3110         } while_each_thread(task, pos);
3111 fail:
3112         pos = NULL;
3113         goto out;
3114 found:
3115         get_task_struct(pos);
3116 out:
3117         rcu_read_unlock();
3118         return pos;
3119 }
3120
3121 /*
3122  * Find the next thread in the thread list.
3123  * Return NULL if there is an error or no next thread.
3124  *
3125  * The reference to the input task_struct is released.
3126  */
3127 static struct task_struct *next_tid(struct task_struct *start)
3128 {
3129         struct task_struct *pos = NULL;
3130         rcu_read_lock();
3131         if (pid_alive(start)) {
3132                 pos = next_thread(start);
3133                 if (thread_group_leader(pos))
3134                         pos = NULL;
3135                 else
3136                         get_task_struct(pos);
3137         }
3138         rcu_read_unlock();
3139         put_task_struct(start);
3140         return pos;
3141 }
3142
3143 /* for the /proc/TGID/task/ directories */
3144 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3145 {
3146         struct inode *inode = file_inode(file);
3147         struct task_struct *task;
3148         struct pid_namespace *ns;
3149         int tid;
3150
3151         if (proc_inode_is_dead(inode))
3152                 return -ENOENT;
3153
3154         if (!dir_emit_dots(file, ctx))
3155                 return 0;
3156
3157         /* f_version caches the tgid value that the last readdir call couldn't
3158          * return. lseek aka telldir automagically resets f_version to 0.
3159          */
3160         ns = file->f_dentry->d_sb->s_fs_info;
3161         tid = (int)file->f_version;
3162         file->f_version = 0;
3163         for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3164              task;
3165              task = next_tid(task), ctx->pos++) {
3166                 char name[PROC_NUMBUF];
3167                 int len;
3168                 tid = task_pid_nr_ns(task, ns);
3169                 len = snprintf(name, sizeof(name), "%d", tid);
3170                 if (!proc_fill_cache(file, ctx, name, len,
3171                                 proc_task_instantiate, task, NULL)) {
3172                         /* returning this tgid failed, save it as the first
3173                          * pid for the next readir call */
3174                         file->f_version = (u64)tid;
3175                         put_task_struct(task);
3176                         break;
3177                 }
3178         }
3179
3180         return 0;
3181 }
3182
3183 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3184 {
3185         struct inode *inode = dentry->d_inode;
3186         struct task_struct *p = get_proc_task(inode);
3187         generic_fillattr(inode, stat);
3188
3189         if (p) {
3190                 stat->nlink += get_nr_threads(p);
3191                 put_task_struct(p);
3192         }
3193
3194         return 0;
3195 }
3196
3197 static const struct inode_operations proc_task_inode_operations = {
3198         .lookup         = proc_task_lookup,
3199         .getattr        = proc_task_getattr,
3200         .setattr        = proc_setattr,
3201         .permission     = proc_pid_permission,
3202 };
3203
3204 static const struct file_operations proc_task_operations = {
3205         .read           = generic_read_dir,
3206         .iterate        = proc_task_readdir,
3207         .llseek         = default_llseek,
3208 };