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