linux/fs/proc/base.c
<<
>>
Prefs
   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
 107struct 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 ONE(NAME, MODE, show)                           \
 134        NOD(NAME, (S_IFREG|(MODE)),                     \
 135                NULL, &proc_single_file_operations,     \
 136                { .proc_show = show } )
 137
 138/*
 139 * Count the number of hardlinks for the pid_entry table, excluding the .
 140 * and .. links.
 141 */
 142static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
 143        unsigned int n)
 144{
 145        unsigned int i;
 146        unsigned int count;
 147
 148        count = 0;
 149        for (i = 0; i < n; ++i) {
 150                if (S_ISDIR(entries[i].mode))
 151                        ++count;
 152        }
 153
 154        return count;
 155}
 156
 157static int get_task_root(struct task_struct *task, struct path *root)
 158{
 159        int result = -ENOENT;
 160
 161        task_lock(task);
 162        if (task->fs) {
 163                get_fs_root(task->fs, root);
 164                result = 0;
 165        }
 166        task_unlock(task);
 167        return result;
 168}
 169
 170static int proc_cwd_link(struct dentry *dentry, struct path *path)
 171{
 172        struct task_struct *task = get_proc_task(d_inode(dentry));
 173        int result = -ENOENT;
 174
 175        if (task) {
 176                task_lock(task);
 177                if (task->fs) {
 178                        get_fs_pwd(task->fs, path);
 179                        result = 0;
 180                }
 181                task_unlock(task);
 182                put_task_struct(task);
 183        }
 184        return result;
 185}
 186
 187static int proc_root_link(struct dentry *dentry, struct path *path)
 188{
 189        struct task_struct *task = get_proc_task(d_inode(dentry));
 190        int result = -ENOENT;
 191
 192        if (task) {
 193                result = get_task_root(task, path);
 194                put_task_struct(task);
 195        }
 196        return result;
 197}
 198
 199static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
 200                                     size_t _count, loff_t *pos)
 201{
 202        struct task_struct *tsk;
 203        struct mm_struct *mm;
 204        char *page;
 205        unsigned long count = _count;
 206        unsigned long arg_start, arg_end, env_start, env_end;
 207        unsigned long len1, len2, len;
 208        unsigned long p;
 209        char c;
 210        ssize_t rv;
 211
 212        BUG_ON(*pos < 0);
 213
 214        tsk = get_proc_task(file_inode(file));
 215        if (!tsk)
 216                return -ESRCH;
 217        mm = get_task_mm(tsk);
 218        put_task_struct(tsk);
 219        if (!mm)
 220                return 0;
 221        /* Check if process spawned far enough to have cmdline. */
 222        if (!mm->env_end) {
 223                rv = 0;
 224                goto out_mmput;
 225        }
 226
 227        page = (char *)__get_free_page(GFP_TEMPORARY);
 228        if (!page) {
 229                rv = -ENOMEM;
 230                goto out_mmput;
 231        }
 232
 233        down_read(&mm->mmap_sem);
 234        arg_start = mm->arg_start;
 235        arg_end = mm->arg_end;
 236        env_start = mm->env_start;
 237        env_end = mm->env_end;
 238        up_read(&mm->mmap_sem);
 239
 240        BUG_ON(arg_start > arg_end);
 241        BUG_ON(env_start > env_end);
 242
 243        len1 = arg_end - arg_start;
 244        len2 = env_end - env_start;
 245
 246        /* Empty ARGV. */
 247        if (len1 == 0) {
 248                rv = 0;
 249                goto out_free_page;
 250        }
 251        /*
 252         * Inherently racy -- command line shares address space
 253         * with code and data.
 254         */
 255        rv = access_remote_vm(mm, arg_end - 1, &c, 1, 0);
 256        if (rv <= 0)
 257                goto out_free_page;
 258
 259        rv = 0;
 260
 261        if (c == '\0') {
 262                /* Command line (set of strings) occupies whole ARGV. */
 263                if (len1 <= *pos)
 264                        goto out_free_page;
 265
 266                p = arg_start + *pos;
 267                len = len1 - *pos;
 268                while (count > 0 && len > 0) {
 269                        unsigned int _count;
 270                        int nr_read;
 271
 272                        _count = min3(count, len, PAGE_SIZE);
 273                        nr_read = access_remote_vm(mm, p, page, _count, 0);
 274                        if (nr_read < 0)
 275                                rv = nr_read;
 276                        if (nr_read <= 0)
 277                                goto out_free_page;
 278
 279                        if (copy_to_user(buf, page, nr_read)) {
 280                                rv = -EFAULT;
 281                                goto out_free_page;
 282                        }
 283
 284                        p       += nr_read;
 285                        len     -= nr_read;
 286                        buf     += nr_read;
 287                        count   -= nr_read;
 288                        rv      += nr_read;
 289                }
 290        } else {
 291                /*
 292                 * Command line (1 string) occupies ARGV and maybe
 293                 * extends into ENVP.
 294                 */
 295                if (len1 + len2 <= *pos)
 296                        goto skip_argv_envp;
 297                if (len1 <= *pos)
 298                        goto skip_argv;
 299
 300                p = arg_start + *pos;
 301                len = len1 - *pos;
 302                while (count > 0 && len > 0) {
 303                        unsigned int _count, l;
 304                        int nr_read;
 305                        bool final;
 306
 307                        _count = min3(count, len, PAGE_SIZE);
 308                        nr_read = access_remote_vm(mm, p, page, _count, 0);
 309                        if (nr_read < 0)
 310                                rv = nr_read;
 311                        if (nr_read <= 0)
 312                                goto out_free_page;
 313
 314                        /*
 315                         * Command line can be shorter than whole ARGV
 316                         * even if last "marker" byte says it is not.
 317                         */
 318                        final = false;
 319                        l = strnlen(page, nr_read);
 320                        if (l < nr_read) {
 321                                nr_read = l;
 322                                final = true;
 323                        }
 324
 325                        if (copy_to_user(buf, page, nr_read)) {
 326                                rv = -EFAULT;
 327                                goto out_free_page;
 328                        }
 329
 330                        p       += nr_read;
 331                        len     -= nr_read;
 332                        buf     += nr_read;
 333                        count   -= nr_read;
 334                        rv      += nr_read;
 335
 336                        if (final)
 337                                goto out_free_page;
 338                }
 339skip_argv:
 340                /*
 341                 * Command line (1 string) occupies ARGV and
 342                 * extends into ENVP.
 343                 */
 344                if (len1 <= *pos) {
 345                        p = env_start + *pos - len1;
 346                        len = len1 + len2 - *pos;
 347                } else {
 348                        p = env_start;
 349                        len = len2;
 350                }
 351                while (count > 0 && len > 0) {
 352                        unsigned int _count, l;
 353                        int nr_read;
 354                        bool final;
 355
 356                        _count = min3(count, len, PAGE_SIZE);
 357                        nr_read = access_remote_vm(mm, p, page, _count, 0);
 358                        if (nr_read < 0)
 359                                rv = nr_read;
 360                        if (nr_read <= 0)
 361                                goto out_free_page;
 362
 363                        /* Find EOS. */
 364                        final = false;
 365                        l = strnlen(page, nr_read);
 366                        if (l < nr_read) {
 367                                nr_read = l;
 368                                final = true;
 369                        }
 370
 371                        if (copy_to_user(buf, page, nr_read)) {
 372                                rv = -EFAULT;
 373                                goto out_free_page;
 374                        }
 375
 376                        p       += nr_read;
 377                        len     -= nr_read;
 378                        buf     += nr_read;
 379                        count   -= nr_read;
 380                        rv      += nr_read;
 381
 382                        if (final)
 383                                goto out_free_page;
 384                }
 385skip_argv_envp:
 386                ;
 387        }
 388
 389out_free_page:
 390        free_page((unsigned long)page);
 391out_mmput:
 392        mmput(mm);
 393        if (rv > 0)
 394                *pos += rv;
 395        return rv;
 396}
 397
 398static const struct file_operations proc_pid_cmdline_ops = {
 399        .read   = proc_pid_cmdline_read,
 400        .llseek = generic_file_llseek,
 401};
 402
 403static int proc_pid_auxv(struct seq_file *m, struct pid_namespace *ns,
 404                         struct pid *pid, struct task_struct *task)
 405{
 406        struct mm_struct *mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
 407        if (mm && !IS_ERR(mm)) {
 408                unsigned int nwords = 0;
 409                do {
 410                        nwords += 2;
 411                } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 412                seq_write(m, mm->saved_auxv, nwords * sizeof(mm->saved_auxv[0]));
 413                mmput(mm);
 414                return 0;
 415        } else
 416                return PTR_ERR(mm);
 417}
 418
 419
 420#ifdef CONFIG_KALLSYMS
 421/*
 422 * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 423 * Returns the resolved symbol.  If that fails, simply return the address.
 424 */
 425static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
 426                          struct pid *pid, struct task_struct *task)
 427{
 428        unsigned long wchan;
 429        char symname[KSYM_NAME_LEN];
 430
 431        wchan = get_wchan(task);
 432
 433        if (wchan && ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)
 434                        && !lookup_symbol_name(wchan, symname))
 435                seq_printf(m, "%s", symname);
 436        else
 437                seq_putc(m, '0');
 438
 439        return 0;
 440}
 441#endif /* CONFIG_KALLSYMS */
 442
 443static int lock_trace(struct task_struct *task)
 444{
 445        int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
 446        if (err)
 447                return err;
 448        if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
 449                mutex_unlock(&task->signal->cred_guard_mutex);
 450                return -EPERM;
 451        }
 452        return 0;
 453}
 454
 455static void unlock_trace(struct task_struct *task)
 456{
 457        mutex_unlock(&task->signal->cred_guard_mutex);
 458}
 459
 460#ifdef CONFIG_STACKTRACE
 461
 462#define MAX_STACK_TRACE_DEPTH   64
 463
 464static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 465                          struct pid *pid, struct task_struct *task)
 466{
 467        struct stack_trace trace;
 468        unsigned long *entries;
 469        int err;
 470        int i;
 471
 472        entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 473        if (!entries)
 474                return -ENOMEM;
 475
 476        trace.nr_entries        = 0;
 477        trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 478        trace.entries           = entries;
 479        trace.skip              = 0;
 480
 481        err = lock_trace(task);
 482        if (!err) {
 483                save_stack_trace_tsk(task, &trace);
 484
 485                for (i = 0; i < trace.nr_entries; i++) {
 486                        seq_printf(m, "[<%pK>] %pS\n",
 487                                   (void *)entries[i], (void *)entries[i]);
 488                }
 489                unlock_trace(task);
 490        }
 491        kfree(entries);
 492
 493        return err;
 494}
 495#endif
 496
 497#ifdef CONFIG_SCHED_INFO
 498/*
 499 * Provides /proc/PID/schedstat
 500 */
 501static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
 502                              struct pid *pid, struct task_struct *task)
 503{
 504        if (unlikely(!sched_info_on()))
 505                seq_printf(m, "0 0 0\n");
 506        else
 507                seq_printf(m, "%llu %llu %lu\n",
 508                   (unsigned long long)task->se.sum_exec_runtime,
 509                   (unsigned long long)task->sched_info.run_delay,
 510                   task->sched_info.pcount);
 511
 512        return 0;
 513}
 514#endif
 515
 516#ifdef CONFIG_LATENCYTOP
 517static int lstats_show_proc(struct seq_file *m, void *v)
 518{
 519        int i;
 520        struct inode *inode = m->private;
 521        struct task_struct *task = get_proc_task(inode);
 522
 523        if (!task)
 524                return -ESRCH;
 525        seq_puts(m, "Latency Top version : v0.1\n");
 526        for (i = 0; i < 32; i++) {
 527                struct latency_record *lr = &task->latency_record[i];
 528                if (lr->backtrace[0]) {
 529                        int q;
 530                        seq_printf(m, "%i %li %li",
 531                                   lr->count, lr->time, lr->max);
 532                        for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 533                                unsigned long bt = lr->backtrace[q];
 534                                if (!bt)
 535                                        break;
 536                                if (bt == ULONG_MAX)
 537                                        break;
 538                                seq_printf(m, " %ps", (void *)bt);
 539                        }
 540                        seq_putc(m, '\n');
 541                }
 542
 543        }
 544        put_task_struct(task);
 545        return 0;
 546}
 547
 548static int lstats_open(struct inode *inode, struct file *file)
 549{
 550        return single_open(file, lstats_show_proc, inode);
 551}
 552
 553static ssize_t lstats_write(struct file *file, const char __user *buf,
 554                            size_t count, loff_t *offs)
 555{
 556        struct task_struct *task = get_proc_task(file_inode(file));
 557
 558        if (!task)
 559                return -ESRCH;
 560        clear_all_latency_tracing(task);
 561        put_task_struct(task);
 562
 563        return count;
 564}
 565
 566static const struct file_operations proc_lstats_operations = {
 567        .open           = lstats_open,
 568        .read           = seq_read,
 569        .write          = lstats_write,
 570        .llseek         = seq_lseek,
 571        .release        = single_release,
 572};
 573
 574#endif
 575
 576static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
 577                          struct pid *pid, struct task_struct *task)
 578{
 579        unsigned long totalpages = totalram_pages + total_swap_pages;
 580        unsigned long points = 0;
 581
 582        read_lock(&tasklist_lock);
 583        if (pid_alive(task))
 584                points = oom_badness(task, NULL, NULL, totalpages) *
 585                                                1000 / totalpages;
 586        read_unlock(&tasklist_lock);
 587        seq_printf(m, "%lu\n", points);
 588
 589        return 0;
 590}
 591
 592struct limit_names {
 593        const char *name;
 594        const char *unit;
 595};
 596
 597static const struct limit_names lnames[RLIM_NLIMITS] = {
 598        [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 599        [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 600        [RLIMIT_DATA] = {"Max data size", "bytes"},
 601        [RLIMIT_STACK] = {"Max stack size", "bytes"},
 602        [RLIMIT_CORE] = {"Max core file size", "bytes"},
 603        [RLIMIT_RSS] = {"Max resident set", "bytes"},
 604        [RLIMIT_NPROC] = {"Max processes", "processes"},
 605        [RLIMIT_NOFILE] = {"Max open files", "files"},
 606        [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 607        [RLIMIT_AS] = {"Max address space", "bytes"},
 608        [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 609        [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 610        [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 611        [RLIMIT_NICE] = {"Max nice priority", NULL},
 612        [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 613        [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 614};
 615
 616/* Display limits for a process */
 617static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
 618                           struct pid *pid, struct task_struct *task)
 619{
 620        unsigned int i;
 621        unsigned long flags;
 622
 623        struct rlimit rlim[RLIM_NLIMITS];
 624
 625        if (!lock_task_sighand(task, &flags))
 626                return 0;
 627        memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 628        unlock_task_sighand(task, &flags);
 629
 630        /*
 631         * print the file header
 632         */
 633       seq_printf(m, "%-25s %-20s %-20s %-10s\n",
 634                  "Limit", "Soft Limit", "Hard Limit", "Units");
 635
 636        for (i = 0; i < RLIM_NLIMITS; i++) {
 637                if (rlim[i].rlim_cur == RLIM_INFINITY)
 638                        seq_printf(m, "%-25s %-20s ",
 639                                   lnames[i].name, "unlimited");
 640                else
 641                        seq_printf(m, "%-25s %-20lu ",
 642                                   lnames[i].name, rlim[i].rlim_cur);
 643
 644                if (rlim[i].rlim_max == RLIM_INFINITY)
 645                        seq_printf(m, "%-20s ", "unlimited");
 646                else
 647                        seq_printf(m, "%-20lu ", rlim[i].rlim_max);
 648
 649                if (lnames[i].unit)
 650                        seq_printf(m, "%-10s\n", lnames[i].unit);
 651                else
 652                        seq_putc(m, '\n');
 653        }
 654
 655        return 0;
 656}
 657
 658#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 659static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
 660                            struct pid *pid, struct task_struct *task)
 661{
 662        long nr;
 663        unsigned long args[6], sp, pc;
 664        int res;
 665
 666        res = lock_trace(task);
 667        if (res)
 668                return res;
 669
 670        if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 671                seq_puts(m, "running\n");
 672        else if (nr < 0)
 673                seq_printf(m, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 674        else
 675                seq_printf(m,
 676                       "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 677                       nr,
 678                       args[0], args[1], args[2], args[3], args[4], args[5],
 679                       sp, pc);
 680        unlock_trace(task);
 681
 682        return 0;
 683}
 684#endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 685
 686/************************************************************************/
 687/*                       Here the fs part begins                        */
 688/************************************************************************/
 689
 690/* permission checks */
 691static int proc_fd_access_allowed(struct inode *inode)
 692{
 693        struct task_struct *task;
 694        int allowed = 0;
 695        /* Allow access to a task's file descriptors if it is us or we
 696         * may use ptrace attach to the process and find out that
 697         * information.
 698         */
 699        task = get_proc_task(inode);
 700        if (task) {
 701                allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 702                put_task_struct(task);
 703        }
 704        return allowed;
 705}
 706
 707int proc_setattr(struct dentry *dentry, struct iattr *attr)
 708{
 709        int error;
 710        struct inode *inode = d_inode(dentry);
 711
 712        if (attr->ia_valid & ATTR_MODE)
 713                return -EPERM;
 714
 715        error = inode_change_ok(inode, attr);
 716        if (error)
 717                return error;
 718
 719        setattr_copy(inode, attr);
 720        mark_inode_dirty(inode);
 721        return 0;
 722}
 723
 724/*
 725 * May current process learn task's sched/cmdline info (for hide_pid_min=1)
 726 * or euid/egid (for hide_pid_min=2)?
 727 */
 728static bool has_pid_permissions(struct pid_namespace *pid,
 729                                 struct task_struct *task,
 730                                 int hide_pid_min)
 731{
 732        if (pid->hide_pid < hide_pid_min)
 733                return true;
 734        if (in_group_p(pid->pid_gid))
 735                return true;
 736        return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 737}
 738
 739
 740static int proc_pid_permission(struct inode *inode, int mask)
 741{
 742        struct pid_namespace *pid = inode->i_sb->s_fs_info;
 743        struct task_struct *task;
 744        bool has_perms;
 745
 746        task = get_proc_task(inode);
 747        if (!task)
 748                return -ESRCH;
 749        has_perms = has_pid_permissions(pid, task, 1);
 750        put_task_struct(task);
 751
 752        if (!has_perms) {
 753                if (pid->hide_pid == 2) {
 754                        /*
 755                         * Let's make getdents(), stat(), and open()
 756                         * consistent with each other.  If a process
 757                         * may not stat() a file, it shouldn't be seen
 758                         * in procfs at all.
 759                         */
 760                        return -ENOENT;
 761                }
 762
 763                return -EPERM;
 764        }
 765        return generic_permission(inode, mask);
 766}
 767
 768
 769
 770static const struct inode_operations proc_def_inode_operations = {
 771        .setattr        = proc_setattr,
 772};
 773
 774static int proc_single_show(struct seq_file *m, void *v)
 775{
 776        struct inode *inode = m->private;
 777        struct pid_namespace *ns;
 778        struct pid *pid;
 779        struct task_struct *task;
 780        int ret;
 781
 782        ns = inode->i_sb->s_fs_info;
 783        pid = proc_pid(inode);
 784        task = get_pid_task(pid, PIDTYPE_PID);
 785        if (!task)
 786                return -ESRCH;
 787
 788        ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 789
 790        put_task_struct(task);
 791        return ret;
 792}
 793
 794static int proc_single_open(struct inode *inode, struct file *filp)
 795{
 796        return single_open(filp, proc_single_show, inode);
 797}
 798
 799static const struct file_operations proc_single_file_operations = {
 800        .open           = proc_single_open,
 801        .read           = seq_read,
 802        .llseek         = seq_lseek,
 803        .release        = single_release,
 804};
 805
 806
 807struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
 808{
 809        struct task_struct *task = get_proc_task(inode);
 810        struct mm_struct *mm = ERR_PTR(-ESRCH);
 811
 812        if (task) {
 813                mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
 814                put_task_struct(task);
 815
 816                if (!IS_ERR_OR_NULL(mm)) {
 817                        /* ensure this mm_struct can't be freed */
 818                        atomic_inc(&mm->mm_count);
 819                        /* but do not pin its memory */
 820                        mmput(mm);
 821                }
 822        }
 823
 824        return mm;
 825}
 826
 827static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
 828{
 829        struct mm_struct *mm = proc_mem_open(inode, mode);
 830
 831        if (IS_ERR(mm))
 832                return PTR_ERR(mm);
 833
 834        file->private_data = mm;
 835        return 0;
 836}
 837
 838static int mem_open(struct inode *inode, struct file *file)
 839{
 840        int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
 841
 842        /* OK to pass negative loff_t, we can catch out-of-range */
 843        file->f_mode |= FMODE_UNSIGNED_OFFSET;
 844
 845        return ret;
 846}
 847
 848static ssize_t mem_rw(struct file *file, char __user *buf,
 849                        size_t count, loff_t *ppos, int write)
 850{
 851        struct mm_struct *mm = file->private_data;
 852        unsigned long addr = *ppos;
 853        ssize_t copied;
 854        char *page;
 855
 856        if (!mm)
 857                return 0;
 858
 859        page = (char *)__get_free_page(GFP_TEMPORARY);
 860        if (!page)
 861                return -ENOMEM;
 862
 863        copied = 0;
 864        if (!atomic_inc_not_zero(&mm->mm_users))
 865                goto free;
 866
 867        while (count > 0) {
 868                int this_len = min_t(int, count, PAGE_SIZE);
 869
 870                if (write && copy_from_user(page, buf, this_len)) {
 871                        copied = -EFAULT;
 872                        break;
 873                }
 874
 875                this_len = access_remote_vm(mm, addr, page, this_len, write);
 876                if (!this_len) {
 877                        if (!copied)
 878                                copied = -EIO;
 879                        break;
 880                }
 881
 882                if (!write && copy_to_user(buf, page, this_len)) {
 883                        copied = -EFAULT;
 884                        break;
 885                }
 886
 887                buf += this_len;
 888                addr += this_len;
 889                copied += this_len;
 890                count -= this_len;
 891        }
 892        *ppos = addr;
 893
 894        mmput(mm);
 895free:
 896        free_page((unsigned long) page);
 897        return copied;
 898}
 899
 900static ssize_t mem_read(struct file *file, char __user *buf,
 901                        size_t count, loff_t *ppos)
 902{
 903        return mem_rw(file, buf, count, ppos, 0);
 904}
 905
 906static ssize_t mem_write(struct file *file, const char __user *buf,
 907                         size_t count, loff_t *ppos)
 908{
 909        return mem_rw(file, (char __user*)buf, count, ppos, 1);
 910}
 911
 912loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 913{
 914        switch (orig) {
 915        case 0:
 916                file->f_pos = offset;
 917                break;
 918        case 1:
 919                file->f_pos += offset;
 920                break;
 921        default:
 922                return -EINVAL;
 923        }
 924        force_successful_syscall_return();
 925        return file->f_pos;
 926}
 927
 928static int mem_release(struct inode *inode, struct file *file)
 929{
 930        struct mm_struct *mm = file->private_data;
 931        if (mm)
 932                mmdrop(mm);
 933        return 0;
 934}
 935
 936static const struct file_operations proc_mem_operations = {
 937        .llseek         = mem_lseek,
 938        .read           = mem_read,
 939        .write          = mem_write,
 940        .open           = mem_open,
 941        .release        = mem_release,
 942};
 943
 944static int environ_open(struct inode *inode, struct file *file)
 945{
 946        return __mem_open(inode, file, PTRACE_MODE_READ);
 947}
 948
 949static ssize_t environ_read(struct file *file, char __user *buf,
 950                        size_t count, loff_t *ppos)
 951{
 952        char *page;
 953        unsigned long src = *ppos;
 954        int ret = 0;
 955        struct mm_struct *mm = file->private_data;
 956        unsigned long env_start, env_end;
 957
 958        /* Ensure the process spawned far enough to have an environment. */
 959        if (!mm || !mm->env_end)
 960                return 0;
 961
 962        page = (char *)__get_free_page(GFP_TEMPORARY);
 963        if (!page)
 964                return -ENOMEM;
 965
 966        ret = 0;
 967        if (!atomic_inc_not_zero(&mm->mm_users))
 968                goto free;
 969
 970        down_read(&mm->mmap_sem);
 971        env_start = mm->env_start;
 972        env_end = mm->env_end;
 973        up_read(&mm->mmap_sem);
 974
 975        while (count > 0) {
 976                size_t this_len, max_len;
 977                int retval;
 978
 979                if (src >= (env_end - env_start))
 980                        break;
 981
 982                this_len = env_end - (env_start + src);
 983
 984                max_len = min_t(size_t, PAGE_SIZE, count);
 985                this_len = min(max_len, this_len);
 986
 987                retval = access_remote_vm(mm, (env_start + src),
 988                        page, this_len, 0);
 989
 990                if (retval <= 0) {
 991                        ret = retval;
 992                        break;
 993                }
 994
 995                if (copy_to_user(buf, page, retval)) {
 996                        ret = -EFAULT;
 997                        break;
 998                }
 999
1000                ret += retval;
1001                src += retval;
1002                buf += retval;
1003                count -= retval;
1004        }
1005        *ppos = src;
1006        mmput(mm);
1007
1008free:
1009        free_page((unsigned long) page);
1010        return ret;
1011}
1012
1013static const struct file_operations proc_environ_operations = {
1014        .open           = environ_open,
1015        .read           = environ_read,
1016        .llseek         = generic_file_llseek,
1017        .release        = mem_release,
1018};
1019
1020static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1021                            loff_t *ppos)
1022{
1023        struct task_struct *task = get_proc_task(file_inode(file));
1024        char buffer[PROC_NUMBUF];
1025        int oom_adj = OOM_ADJUST_MIN;
1026        size_t len;
1027        unsigned long flags;
1028
1029        if (!task)
1030                return -ESRCH;
1031        if (lock_task_sighand(task, &flags)) {
1032                if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1033                        oom_adj = OOM_ADJUST_MAX;
1034                else
1035                        oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1036                                  OOM_SCORE_ADJ_MAX;
1037                unlock_task_sighand(task, &flags);
1038        }
1039        put_task_struct(task);
1040        len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1041        return simple_read_from_buffer(buf, count, ppos, buffer, len);
1042}
1043
1044/*
1045 * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1046 * kernels.  The effective policy is defined by oom_score_adj, which has a
1047 * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1048 * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1049 * Processes that become oom disabled via oom_adj will still be oom disabled
1050 * with this implementation.
1051 *
1052 * oom_adj cannot be removed since existing userspace binaries use it.
1053 */
1054static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1055                             size_t count, loff_t *ppos)
1056{
1057        struct task_struct *task;
1058        char buffer[PROC_NUMBUF];
1059        int oom_adj;
1060        unsigned long flags;
1061        int err;
1062
1063        memset(buffer, 0, sizeof(buffer));
1064        if (count > sizeof(buffer) - 1)
1065                count = sizeof(buffer) - 1;
1066        if (copy_from_user(buffer, buf, count)) {
1067                err = -EFAULT;
1068                goto out;
1069        }
1070
1071        err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1072        if (err)
1073                goto out;
1074        if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1075             oom_adj != OOM_DISABLE) {
1076                err = -EINVAL;
1077                goto out;
1078        }
1079
1080        task = get_proc_task(file_inode(file));
1081        if (!task) {
1082                err = -ESRCH;
1083                goto out;
1084        }
1085
1086        task_lock(task);
1087        if (!task->mm) {
1088                err = -EINVAL;
1089                goto err_task_lock;
1090        }
1091
1092        if (!lock_task_sighand(task, &flags)) {
1093                err = -ESRCH;
1094                goto err_task_lock;
1095        }
1096
1097        /*
1098         * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1099         * value is always attainable.
1100         */
1101        if (oom_adj == OOM_ADJUST_MAX)
1102                oom_adj = OOM_SCORE_ADJ_MAX;
1103        else
1104                oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1105
1106        if (oom_adj < task->signal->oom_score_adj &&
1107            !capable(CAP_SYS_RESOURCE)) {
1108                err = -EACCES;
1109                goto err_sighand;
1110        }
1111
1112        /*
1113         * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1114         * /proc/pid/oom_score_adj instead.
1115         */
1116        pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1117                  current->comm, task_pid_nr(current), task_pid_nr(task),
1118                  task_pid_nr(task));
1119
1120        task->signal->oom_score_adj = oom_adj;
1121        trace_oom_score_adj_update(task);
1122err_sighand:
1123        unlock_task_sighand(task, &flags);
1124err_task_lock:
1125        task_unlock(task);
1126        put_task_struct(task);
1127out:
1128        return err < 0 ? err : count;
1129}
1130
1131static const struct file_operations proc_oom_adj_operations = {
1132        .read           = oom_adj_read,
1133        .write          = oom_adj_write,
1134        .llseek         = generic_file_llseek,
1135};
1136
1137static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1138                                        size_t count, loff_t *ppos)
1139{
1140        struct task_struct *task = get_proc_task(file_inode(file));
1141        char buffer[PROC_NUMBUF];
1142        short oom_score_adj = OOM_SCORE_ADJ_MIN;
1143        unsigned long flags;
1144        size_t len;
1145
1146        if (!task)
1147                return -ESRCH;
1148        if (lock_task_sighand(task, &flags)) {
1149                oom_score_adj = task->signal->oom_score_adj;
1150                unlock_task_sighand(task, &flags);
1151        }
1152        put_task_struct(task);
1153        len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1154        return simple_read_from_buffer(buf, count, ppos, buffer, len);
1155}
1156
1157static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1158                                        size_t count, loff_t *ppos)
1159{
1160        struct task_struct *task;
1161        char buffer[PROC_NUMBUF];
1162        unsigned long flags;
1163        int oom_score_adj;
1164        int err;
1165
1166        memset(buffer, 0, sizeof(buffer));
1167        if (count > sizeof(buffer) - 1)
1168                count = sizeof(buffer) - 1;
1169        if (copy_from_user(buffer, buf, count)) {
1170                err = -EFAULT;
1171                goto out;
1172        }
1173
1174        err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1175        if (err)
1176                goto out;
1177        if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1178                        oom_score_adj > OOM_SCORE_ADJ_MAX) {
1179                err = -EINVAL;
1180                goto out;
1181        }
1182
1183        task = get_proc_task(file_inode(file));
1184        if (!task) {
1185                err = -ESRCH;
1186                goto out;
1187        }
1188
1189        task_lock(task);
1190        if (!task->mm) {
1191                err = -EINVAL;
1192                goto err_task_lock;
1193        }
1194
1195        if (!lock_task_sighand(task, &flags)) {
1196                err = -ESRCH;
1197                goto err_task_lock;
1198        }
1199
1200        if ((short)oom_score_adj < task->signal->oom_score_adj_min &&
1201                        !capable(CAP_SYS_RESOURCE)) {
1202                err = -EACCES;
1203                goto err_sighand;
1204        }
1205
1206        task->signal->oom_score_adj = (short)oom_score_adj;
1207        if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1208                task->signal->oom_score_adj_min = (short)oom_score_adj;
1209        trace_oom_score_adj_update(task);
1210
1211err_sighand:
1212        unlock_task_sighand(task, &flags);
1213err_task_lock:
1214        task_unlock(task);
1215        put_task_struct(task);
1216out:
1217        return err < 0 ? err : count;
1218}
1219
1220static const struct file_operations proc_oom_score_adj_operations = {
1221        .read           = oom_score_adj_read,
1222        .write          = oom_score_adj_write,
1223        .llseek         = default_llseek,
1224};
1225
1226#ifdef CONFIG_AUDITSYSCALL
1227#define TMPBUFLEN 21
1228static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1229                                  size_t count, loff_t *ppos)
1230{
1231        struct inode * inode = file_inode(file);
1232        struct task_struct *task = get_proc_task(inode);
1233        ssize_t length;
1234        char tmpbuf[TMPBUFLEN];
1235
1236        if (!task)
1237                return -ESRCH;
1238        length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1239                           from_kuid(file->f_cred->user_ns,
1240                                     audit_get_loginuid(task)));
1241        put_task_struct(task);
1242        return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1243}
1244
1245static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1246                                   size_t count, loff_t *ppos)
1247{
1248        struct inode * inode = file_inode(file);
1249        uid_t loginuid;
1250        kuid_t kloginuid;
1251        int rv;
1252
1253        rcu_read_lock();
1254        if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1255                rcu_read_unlock();
1256                return -EPERM;
1257        }
1258        rcu_read_unlock();
1259
1260        if (*ppos != 0) {
1261                /* No partial writes. */
1262                return -EINVAL;
1263        }
1264
1265        rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1266        if (rv < 0)
1267                return rv;
1268
1269        /* is userspace tring to explicitly UNSET the loginuid? */
1270        if (loginuid == AUDIT_UID_UNSET) {
1271                kloginuid = INVALID_UID;
1272        } else {
1273                kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1274                if (!uid_valid(kloginuid))
1275                        return -EINVAL;
1276        }
1277
1278        rv = audit_set_loginuid(kloginuid);
1279        if (rv < 0)
1280                return rv;
1281        return count;
1282}
1283
1284static const struct file_operations proc_loginuid_operations = {
1285        .read           = proc_loginuid_read,
1286        .write          = proc_loginuid_write,
1287        .llseek         = generic_file_llseek,
1288};
1289
1290static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1291                                  size_t count, loff_t *ppos)
1292{
1293        struct inode * inode = file_inode(file);
1294        struct task_struct *task = get_proc_task(inode);
1295        ssize_t length;
1296        char tmpbuf[TMPBUFLEN];
1297
1298        if (!task)
1299                return -ESRCH;
1300        length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1301                                audit_get_sessionid(task));
1302        put_task_struct(task);
1303        return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1304}
1305
1306static const struct file_operations proc_sessionid_operations = {
1307        .read           = proc_sessionid_read,
1308        .llseek         = generic_file_llseek,
1309};
1310#endif
1311
1312#ifdef CONFIG_FAULT_INJECTION
1313static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1314                                      size_t count, loff_t *ppos)
1315{
1316        struct task_struct *task = get_proc_task(file_inode(file));
1317        char buffer[PROC_NUMBUF];
1318        size_t len;
1319        int make_it_fail;
1320
1321        if (!task)
1322                return -ESRCH;
1323        make_it_fail = task->make_it_fail;
1324        put_task_struct(task);
1325
1326        len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1327
1328        return simple_read_from_buffer(buf, count, ppos, buffer, len);
1329}
1330
1331static ssize_t proc_fault_inject_write(struct file * file,
1332                        const char __user * buf, size_t count, loff_t *ppos)
1333{
1334        struct task_struct *task;
1335        char buffer[PROC_NUMBUF];
1336        int make_it_fail;
1337        int rv;
1338
1339        if (!capable(CAP_SYS_RESOURCE))
1340                return -EPERM;
1341        memset(buffer, 0, sizeof(buffer));
1342        if (count > sizeof(buffer) - 1)
1343                count = sizeof(buffer) - 1;
1344        if (copy_from_user(buffer, buf, count))
1345                return -EFAULT;
1346        rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1347        if (rv < 0)
1348                return rv;
1349        if (make_it_fail < 0 || make_it_fail > 1)
1350                return -EINVAL;
1351
1352        task = get_proc_task(file_inode(file));
1353        if (!task)
1354                return -ESRCH;
1355        task->make_it_fail = make_it_fail;
1356        put_task_struct(task);
1357
1358        return count;
1359}
1360
1361static const struct file_operations proc_fault_inject_operations = {
1362        .read           = proc_fault_inject_read,
1363        .write          = proc_fault_inject_write,
1364        .llseek         = generic_file_llseek,
1365};
1366#endif
1367
1368
1369#ifdef CONFIG_SCHED_DEBUG
1370/*
1371 * Print out various scheduling related per-task fields:
1372 */
1373static int sched_show(struct seq_file *m, void *v)
1374{
1375        struct inode *inode = m->private;
1376        struct task_struct *p;
1377
1378        p = get_proc_task(inode);
1379        if (!p)
1380                return -ESRCH;
1381        proc_sched_show_task(p, m);
1382
1383        put_task_struct(p);
1384
1385        return 0;
1386}
1387
1388static ssize_t
1389sched_write(struct file *file, const char __user *buf,
1390            size_t count, loff_t *offset)
1391{
1392        struct inode *inode = file_inode(file);
1393        struct task_struct *p;
1394
1395        p = get_proc_task(inode);
1396        if (!p)
1397                return -ESRCH;
1398        proc_sched_set_task(p);
1399
1400        put_task_struct(p);
1401
1402        return count;
1403}
1404
1405static int sched_open(struct inode *inode, struct file *filp)
1406{
1407        return single_open(filp, sched_show, inode);
1408}
1409
1410static const struct file_operations proc_pid_sched_operations = {
1411        .open           = sched_open,
1412        .read           = seq_read,
1413        .write          = sched_write,
1414        .llseek         = seq_lseek,
1415        .release        = single_release,
1416};
1417
1418#endif
1419
1420#ifdef CONFIG_SCHED_AUTOGROUP
1421/*
1422 * Print out autogroup related information:
1423 */
1424static int sched_autogroup_show(struct seq_file *m, void *v)
1425{
1426        struct inode *inode = m->private;
1427        struct task_struct *p;
1428
1429        p = get_proc_task(inode);
1430        if (!p)
1431                return -ESRCH;
1432        proc_sched_autogroup_show_task(p, m);
1433
1434        put_task_struct(p);
1435
1436        return 0;
1437}
1438
1439static ssize_t
1440sched_autogroup_write(struct file *file, const char __user *buf,
1441            size_t count, loff_t *offset)
1442{
1443        struct inode *inode = file_inode(file);
1444        struct task_struct *p;
1445        char buffer[PROC_NUMBUF];
1446        int nice;
1447        int err;
1448
1449        memset(buffer, 0, sizeof(buffer));
1450        if (count > sizeof(buffer) - 1)
1451                count = sizeof(buffer) - 1;
1452        if (copy_from_user(buffer, buf, count))
1453                return -EFAULT;
1454
1455        err = kstrtoint(strstrip(buffer), 0, &nice);
1456        if (err < 0)
1457                return err;
1458
1459        p = get_proc_task(inode);
1460        if (!p)
1461                return -ESRCH;
1462
1463        err = proc_sched_autogroup_set_nice(p, nice);
1464        if (err)
1465                count = err;
1466
1467        put_task_struct(p);
1468
1469        return count;
1470}
1471
1472static int sched_autogroup_open(struct inode *inode, struct file *filp)
1473{
1474        int ret;
1475
1476        ret = single_open(filp, sched_autogroup_show, NULL);
1477        if (!ret) {
1478                struct seq_file *m = filp->private_data;
1479
1480                m->private = inode;
1481        }
1482        return ret;
1483}
1484
1485static const struct file_operations proc_pid_sched_autogroup_operations = {
1486        .open           = sched_autogroup_open,
1487        .read           = seq_read,
1488        .write          = sched_autogroup_write,
1489        .llseek         = seq_lseek,
1490        .release        = single_release,
1491};
1492
1493#endif /* CONFIG_SCHED_AUTOGROUP */
1494
1495static ssize_t comm_write(struct file *file, const char __user *buf,
1496                                size_t count, loff_t *offset)
1497{
1498        struct inode *inode = file_inode(file);
1499        struct task_struct *p;
1500        char buffer[TASK_COMM_LEN];
1501        const size_t maxlen = sizeof(buffer) - 1;
1502
1503        memset(buffer, 0, sizeof(buffer));
1504        if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1505                return -EFAULT;
1506
1507        p = get_proc_task(inode);
1508        if (!p)
1509                return -ESRCH;
1510
1511        if (same_thread_group(current, p))
1512                set_task_comm(p, buffer);
1513        else
1514                count = -EINVAL;
1515
1516        put_task_struct(p);
1517
1518        return count;
1519}
1520
1521static int comm_show(struct seq_file *m, void *v)
1522{
1523        struct inode *inode = m->private;
1524        struct task_struct *p;
1525
1526        p = get_proc_task(inode);
1527        if (!p)
1528                return -ESRCH;
1529
1530        task_lock(p);
1531        seq_printf(m, "%s\n", p->comm);
1532        task_unlock(p);
1533
1534        put_task_struct(p);
1535
1536        return 0;
1537}
1538
1539static int comm_open(struct inode *inode, struct file *filp)
1540{
1541        return single_open(filp, comm_show, inode);
1542}
1543
1544static const struct file_operations proc_pid_set_comm_operations = {
1545        .open           = comm_open,
1546        .read           = seq_read,
1547        .write          = comm_write,
1548        .llseek         = seq_lseek,
1549        .release        = single_release,
1550};
1551
1552static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1553{
1554        struct task_struct *task;
1555        struct file *exe_file;
1556
1557        task = get_proc_task(d_inode(dentry));
1558        if (!task)
1559                return -ENOENT;
1560        exe_file = get_task_exe_file(task);
1561        put_task_struct(task);
1562        if (exe_file) {
1563                *exe_path = exe_file->f_path;
1564                path_get(&exe_file->f_path);
1565                fput(exe_file);
1566                return 0;
1567        } else
1568                return -ENOENT;
1569}
1570
1571static const char *proc_pid_get_link(struct dentry *dentry,
1572                                     struct inode *inode,
1573                                     struct delayed_call *done)
1574{
1575        struct path path;
1576        int error = -EACCES;
1577
1578        if (!dentry)
1579                return ERR_PTR(-ECHILD);
1580
1581        /* Are we allowed to snoop on the tasks file descriptors? */
1582        if (!proc_fd_access_allowed(inode))
1583                goto out;
1584
1585        error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1586        if (error)
1587                goto out;
1588
1589        nd_jump_link(&path);
1590        return NULL;
1591out:
1592        return ERR_PTR(error);
1593}
1594
1595static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1596{
1597        char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1598        char *pathname;
1599        int len;
1600
1601        if (!tmp)
1602                return -ENOMEM;
1603
1604        pathname = d_path(path, tmp, PAGE_SIZE);
1605        len = PTR_ERR(pathname);
1606        if (IS_ERR(pathname))
1607                goto out;
1608        len = tmp + PAGE_SIZE - 1 - pathname;
1609
1610        if (len > buflen)
1611                len = buflen;
1612        if (copy_to_user(buffer, pathname, len))
1613                len = -EFAULT;
1614 out:
1615        free_page((unsigned long)tmp);
1616        return len;
1617}
1618
1619static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1620{
1621        int error = -EACCES;
1622        struct inode *inode = d_inode(dentry);
1623        struct path path;
1624
1625        /* Are we allowed to snoop on the tasks file descriptors? */
1626        if (!proc_fd_access_allowed(inode))
1627                goto out;
1628
1629        error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1630        if (error)
1631                goto out;
1632
1633        error = do_proc_readlink(&path, buffer, buflen);
1634        path_put(&path);
1635out:
1636        return error;
1637}
1638
1639const struct inode_operations proc_pid_link_inode_operations = {
1640        .readlink       = proc_pid_readlink,
1641        .get_link       = proc_pid_get_link,
1642        .setattr        = proc_setattr,
1643};
1644
1645
1646/* building an inode */
1647
1648struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1649{
1650        struct inode * inode;
1651        struct proc_inode *ei;
1652        const struct cred *cred;
1653
1654        /* We need a new inode */
1655
1656        inode = new_inode(sb);
1657        if (!inode)
1658                goto out;
1659
1660        /* Common stuff */
1661        ei = PROC_I(inode);
1662        inode->i_ino = get_next_ino();
1663        inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1664        inode->i_op = &proc_def_inode_operations;
1665
1666        /*
1667         * grab the reference to task.
1668         */
1669        ei->pid = get_task_pid(task, PIDTYPE_PID);
1670        if (!ei->pid)
1671                goto out_unlock;
1672
1673        if (task_dumpable(task)) {
1674                rcu_read_lock();
1675                cred = __task_cred(task);
1676                inode->i_uid = cred->euid;
1677                inode->i_gid = cred->egid;
1678                rcu_read_unlock();
1679        }
1680        security_task_to_inode(task, inode);
1681
1682out:
1683        return inode;
1684
1685out_unlock:
1686        iput(inode);
1687        return NULL;
1688}
1689
1690int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1691{
1692        struct inode *inode = d_inode(dentry);
1693        struct task_struct *task;
1694        const struct cred *cred;
1695        struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1696
1697        generic_fillattr(inode, stat);
1698
1699        rcu_read_lock();
1700        stat->uid = GLOBAL_ROOT_UID;
1701        stat->gid = GLOBAL_ROOT_GID;
1702        task = pid_task(proc_pid(inode), PIDTYPE_PID);
1703        if (task) {
1704                if (!has_pid_permissions(pid, task, 2)) {
1705                        rcu_read_unlock();
1706                        /*
1707                         * This doesn't prevent learning whether PID exists,
1708                         * it only makes getattr() consistent with readdir().
1709                         */
1710                        return -ENOENT;
1711                }
1712                if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1713                    task_dumpable(task)) {
1714                        cred = __task_cred(task);
1715                        stat->uid = cred->euid;
1716                        stat->gid = cred->egid;
1717                }
1718        }
1719        rcu_read_unlock();
1720        return 0;
1721}
1722
1723/* dentry stuff */
1724
1725/*
1726 *      Exceptional case: normally we are not allowed to unhash a busy
1727 * directory. In this case, however, we can do it - no aliasing problems
1728 * due to the way we treat inodes.
1729 *
1730 * Rewrite the inode's ownerships here because the owning task may have
1731 * performed a setuid(), etc.
1732 *
1733 * Before the /proc/pid/status file was created the only way to read
1734 * the effective uid of a /process was to stat /proc/pid.  Reading
1735 * /proc/pid/status is slow enough that procps and other packages
1736 * kept stating /proc/pid.  To keep the rules in /proc simple I have
1737 * made this apply to all per process world readable and executable
1738 * directories.
1739 */
1740int pid_revalidate(struct dentry *dentry, unsigned int flags)
1741{
1742        struct inode *inode;
1743        struct task_struct *task;
1744        const struct cred *cred;
1745
1746        if (flags & LOOKUP_RCU)
1747                return -ECHILD;
1748
1749        inode = d_inode(dentry);
1750        task = get_proc_task(inode);
1751
1752        if (task) {
1753                if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1754                    task_dumpable(task)) {
1755                        rcu_read_lock();
1756                        cred = __task_cred(task);
1757                        inode->i_uid = cred->euid;
1758                        inode->i_gid = cred->egid;
1759                        rcu_read_unlock();
1760                } else {
1761                        inode->i_uid = GLOBAL_ROOT_UID;
1762                        inode->i_gid = GLOBAL_ROOT_GID;
1763                }
1764                inode->i_mode &= ~(S_ISUID | S_ISGID);
1765                security_task_to_inode(task, inode);
1766                put_task_struct(task);
1767                return 1;
1768        }
1769        return 0;
1770}
1771
1772static inline bool proc_inode_is_dead(struct inode *inode)
1773{
1774        return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1775}
1776
1777int pid_delete_dentry(const struct dentry *dentry)
1778{
1779        /* Is the task we represent dead?
1780         * If so, then don't put the dentry on the lru list,
1781         * kill it immediately.
1782         */
1783        return proc_inode_is_dead(d_inode(dentry));
1784}
1785
1786const struct dentry_operations pid_dentry_operations =
1787{
1788        .d_revalidate   = pid_revalidate,
1789        .d_delete       = pid_delete_dentry,
1790};
1791
1792/* Lookups */
1793
1794/*
1795 * Fill a directory entry.
1796 *
1797 * If possible create the dcache entry and derive our inode number and
1798 * file type from dcache entry.
1799 *
1800 * Since all of the proc inode numbers are dynamically generated, the inode
1801 * numbers do not exist until the inode is cache.  This means creating the
1802 * the dcache entry in readdir is necessary to keep the inode numbers
1803 * reported by readdir in sync with the inode numbers reported
1804 * by stat.
1805 */
1806bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1807        const char *name, int len,
1808        instantiate_t instantiate, struct task_struct *task, const void *ptr)
1809{
1810        struct dentry *child, *dir = file->f_path.dentry;
1811        struct qstr qname = QSTR_INIT(name, len);
1812        struct inode *inode;
1813        unsigned type;
1814        ino_t ino;
1815
1816        child = d_hash_and_lookup(dir, &qname);
1817        if (!child) {
1818                DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1819                child = d_alloc_parallel(dir, &qname, &wq);
1820                if (IS_ERR(child))
1821                        goto end_instantiate;
1822                if (d_in_lookup(child)) {
1823                        int err = instantiate(d_inode(dir), child, task, ptr);
1824                        d_lookup_done(child);
1825                        if (err < 0) {
1826                                dput(child);
1827                                goto end_instantiate;
1828                        }
1829                }
1830        }
1831        inode = d_inode(child);
1832        ino = inode->i_ino;
1833        type = inode->i_mode >> 12;
1834        dput(child);
1835        return dir_emit(ctx, name, len, ino, type);
1836
1837end_instantiate:
1838        return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1839}
1840
1841/*
1842 * dname_to_vma_addr - maps a dentry name into two unsigned longs
1843 * which represent vma start and end addresses.
1844 */
1845static int dname_to_vma_addr(struct dentry *dentry,
1846                             unsigned long *start, unsigned long *end)
1847{
1848        if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1849                return -EINVAL;
1850
1851        return 0;
1852}
1853
1854static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1855{
1856        unsigned long vm_start, vm_end;
1857        bool exact_vma_exists = false;
1858        struct mm_struct *mm = NULL;
1859        struct task_struct *task;
1860        const struct cred *cred;
1861        struct inode *inode;
1862        int status = 0;
1863
1864        if (flags & LOOKUP_RCU)
1865                return -ECHILD;
1866
1867        inode = d_inode(dentry);
1868        task = get_proc_task(inode);
1869        if (!task)
1870                goto out_notask;
1871
1872        mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1873        if (IS_ERR_OR_NULL(mm))
1874                goto out;
1875
1876        if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1877                down_read(&mm->mmap_sem);
1878                exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1879                up_read(&mm->mmap_sem);
1880        }
1881
1882        mmput(mm);
1883
1884        if (exact_vma_exists) {
1885                if (task_dumpable(task)) {
1886                        rcu_read_lock();
1887                        cred = __task_cred(task);
1888                        inode->i_uid = cred->euid;
1889                        inode->i_gid = cred->egid;
1890                        rcu_read_unlock();
1891                } else {
1892                        inode->i_uid = GLOBAL_ROOT_UID;
1893                        inode->i_gid = GLOBAL_ROOT_GID;
1894                }
1895                security_task_to_inode(task, inode);
1896                status = 1;
1897        }
1898
1899out:
1900        put_task_struct(task);
1901
1902out_notask:
1903        return status;
1904}
1905
1906static const struct dentry_operations tid_map_files_dentry_operations = {
1907        .d_revalidate   = map_files_d_revalidate,
1908        .d_delete       = pid_delete_dentry,
1909};
1910
1911static int map_files_get_link(struct dentry *dentry, struct path *path)
1912{
1913        unsigned long vm_start, vm_end;
1914        struct vm_area_struct *vma;
1915        struct task_struct *task;
1916        struct mm_struct *mm;
1917        int rc;
1918
1919        rc = -ENOENT;
1920        task = get_proc_task(d_inode(dentry));
1921        if (!task)
1922                goto out;
1923
1924        mm = get_task_mm(task);
1925        put_task_struct(task);
1926        if (!mm)
1927                goto out;
1928
1929        rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1930        if (rc)
1931                goto out_mmput;
1932
1933        rc = -ENOENT;
1934        down_read(&mm->mmap_sem);
1935        vma = find_exact_vma(mm, vm_start, vm_end);
1936        if (vma && vma->vm_file) {
1937                *path = vma->vm_file->f_path;
1938                path_get(path);
1939                rc = 0;
1940        }
1941        up_read(&mm->mmap_sem);
1942
1943out_mmput:
1944        mmput(mm);
1945out:
1946        return rc;
1947}
1948
1949struct map_files_info {
1950        fmode_t         mode;
1951        unsigned long   len;
1952        unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1953};
1954
1955/*
1956 * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
1957 * symlinks may be used to bypass permissions on ancestor directories in the
1958 * path to the file in question.
1959 */
1960static const char *
1961proc_map_files_get_link(struct dentry *dentry,
1962                        struct inode *inode,
1963                        struct delayed_call *done)
1964{
1965        if (!capable(CAP_SYS_ADMIN))
1966                return ERR_PTR(-EPERM);
1967
1968        return proc_pid_get_link(dentry, inode, done);
1969}
1970
1971/*
1972 * Identical to proc_pid_link_inode_operations except for get_link()
1973 */
1974static const struct inode_operations proc_map_files_link_inode_operations = {
1975        .readlink       = proc_pid_readlink,
1976        .get_link       = proc_map_files_get_link,
1977        .setattr        = proc_setattr,
1978};
1979
1980static int
1981proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
1982                           struct task_struct *task, const void *ptr)
1983{
1984        fmode_t mode = (fmode_t)(unsigned long)ptr;
1985        struct proc_inode *ei;
1986        struct inode *inode;
1987
1988        inode = proc_pid_make_inode(dir->i_sb, task);
1989        if (!inode)
1990                return -ENOENT;
1991
1992        ei = PROC_I(inode);
1993        ei->op.proc_get_link = map_files_get_link;
1994
1995        inode->i_op = &proc_map_files_link_inode_operations;
1996        inode->i_size = 64;
1997        inode->i_mode = S_IFLNK;
1998
1999        if (mode & FMODE_READ)
2000                inode->i_mode |= S_IRUSR;
2001        if (mode & FMODE_WRITE)
2002                inode->i_mode |= S_IWUSR;
2003
2004        d_set_d_op(dentry, &tid_map_files_dentry_operations);
2005        d_add(dentry, inode);
2006
2007        return 0;
2008}
2009
2010static struct dentry *proc_map_files_lookup(struct inode *dir,
2011                struct dentry *dentry, unsigned int flags)
2012{
2013        unsigned long vm_start, vm_end;
2014        struct vm_area_struct *vma;
2015        struct task_struct *task;
2016        int result;
2017        struct mm_struct *mm;
2018
2019        result = -ENOENT;
2020        task = get_proc_task(dir);
2021        if (!task)
2022                goto out;
2023
2024        result = -EACCES;
2025        if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2026                goto out_put_task;
2027
2028        result = -ENOENT;
2029        if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2030                goto out_put_task;
2031
2032        mm = get_task_mm(task);
2033        if (!mm)
2034                goto out_put_task;
2035
2036        down_read(&mm->mmap_sem);
2037        vma = find_exact_vma(mm, vm_start, vm_end);
2038        if (!vma)
2039                goto out_no_vma;
2040
2041        if (vma->vm_file)
2042                result = proc_map_files_instantiate(dir, dentry, task,
2043                                (void *)(unsigned long)vma->vm_file->f_mode);
2044
2045out_no_vma:
2046        up_read(&mm->mmap_sem);
2047        mmput(mm);
2048out_put_task:
2049        put_task_struct(task);
2050out:
2051        return ERR_PTR(result);
2052}
2053
2054static const struct inode_operations proc_map_files_inode_operations = {
2055        .lookup         = proc_map_files_lookup,
2056        .permission     = proc_fd_permission,
2057        .setattr        = proc_setattr,
2058};
2059
2060static int
2061proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2062{
2063        struct vm_area_struct *vma;
2064        struct task_struct *task;
2065        struct mm_struct *mm;
2066        unsigned long nr_files, pos, i;
2067        struct flex_array *fa = NULL;
2068        struct map_files_info info;
2069        struct map_files_info *p;
2070        int ret;
2071
2072        ret = -ENOENT;
2073        task = get_proc_task(file_inode(file));
2074        if (!task)
2075                goto out;
2076
2077        ret = -EACCES;
2078        if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2079                goto out_put_task;
2080
2081        ret = 0;
2082        if (!dir_emit_dots(file, ctx))
2083                goto out_put_task;
2084
2085        mm = get_task_mm(task);
2086        if (!mm)
2087                goto out_put_task;
2088        down_read(&mm->mmap_sem);
2089
2090        nr_files = 0;
2091
2092        /*
2093         * We need two passes here:
2094         *
2095         *  1) Collect vmas of mapped files with mmap_sem taken
2096         *  2) Release mmap_sem and instantiate entries
2097         *
2098         * otherwise we get lockdep complained, since filldir()
2099         * routine might require mmap_sem taken in might_fault().
2100         */
2101
2102        for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2103                if (vma->vm_file && ++pos > ctx->pos)
2104                        nr_files++;
2105        }
2106
2107        if (nr_files) {
2108                fa = flex_array_alloc(sizeof(info), nr_files,
2109                                        GFP_KERNEL);
2110                if (!fa || flex_array_prealloc(fa, 0, nr_files,
2111                                                GFP_KERNEL)) {
2112                        ret = -ENOMEM;
2113                        if (fa)
2114                                flex_array_free(fa);
2115                        up_read(&mm->mmap_sem);
2116                        mmput(mm);
2117                        goto out_put_task;
2118                }
2119                for (i = 0, vma = mm->mmap, pos = 2; vma;
2120                                vma = vma->vm_next) {
2121                        if (!vma->vm_file)
2122                                continue;
2123                        if (++pos <= ctx->pos)
2124                                continue;
2125
2126                        info.mode = vma->vm_file->f_mode;
2127                        info.len = snprintf(info.name,
2128                                        sizeof(info.name), "%lx-%lx",
2129                                        vma->vm_start, vma->vm_end);
2130                        if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2131                                BUG();
2132                }
2133        }
2134        up_read(&mm->mmap_sem);
2135
2136        for (i = 0; i < nr_files; i++) {
2137                p = flex_array_get(fa, i);
2138                if (!proc_fill_cache(file, ctx,
2139                                      p->name, p->len,
2140                                      proc_map_files_instantiate,
2141                                      task,
2142                                      (void *)(unsigned long)p->mode))
2143                        break;
2144                ctx->pos++;
2145        }
2146        if (fa)
2147                flex_array_free(fa);
2148        mmput(mm);
2149
2150out_put_task:
2151        put_task_struct(task);
2152out:
2153        return ret;
2154}
2155
2156static const struct file_operations proc_map_files_operations = {
2157        .read           = generic_read_dir,
2158        .iterate_shared = proc_map_files_readdir,
2159        .llseek         = generic_file_llseek,
2160};
2161
2162#ifdef CONFIG_CHECKPOINT_RESTORE
2163struct timers_private {
2164        struct pid *pid;
2165        struct task_struct *task;
2166        struct sighand_struct *sighand;
2167        struct pid_namespace *ns;
2168        unsigned long flags;
2169};
2170
2171static void *timers_start(struct seq_file *m, loff_t *pos)
2172{
2173        struct timers_private *tp = m->private;
2174
2175        tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2176        if (!tp->task)
2177                return ERR_PTR(-ESRCH);
2178
2179        tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2180        if (!tp->sighand)
2181                return ERR_PTR(-ESRCH);
2182
2183        return seq_list_start(&tp->task->signal->posix_timers, *pos);
2184}
2185
2186static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2187{
2188        struct timers_private *tp = m->private;
2189        return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2190}
2191
2192static void timers_stop(struct seq_file *m, void *v)
2193{
2194        struct timers_private *tp = m->private;
2195
2196        if (tp->sighand) {
2197                unlock_task_sighand(tp->task, &tp->flags);
2198                tp->sighand = NULL;
2199        }
2200
2201        if (tp->task) {
2202                put_task_struct(tp->task);
2203                tp->task = NULL;
2204        }
2205}
2206
2207static int show_timer(struct seq_file *m, void *v)
2208{
2209        struct k_itimer *timer;
2210        struct timers_private *tp = m->private;
2211        int notify;
2212        static const char * const nstr[] = {
2213                [SIGEV_SIGNAL] = "signal",
2214                [SIGEV_NONE] = "none",
2215                [SIGEV_THREAD] = "thread",
2216        };
2217
2218        timer = list_entry((struct list_head *)v, struct k_itimer, list);
2219        notify = timer->it_sigev_notify;
2220
2221        seq_printf(m, "ID: %d\n", timer->it_id);
2222        seq_printf(m, "signal: %d/%p\n",
2223                   timer->sigq->info.si_signo,
2224                   timer->sigq->info.si_value.sival_ptr);
2225        seq_printf(m, "notify: %s/%s.%d\n",
2226                   nstr[notify & ~SIGEV_THREAD_ID],
2227                   (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2228                   pid_nr_ns(timer->it_pid, tp->ns));
2229        seq_printf(m, "ClockID: %d\n", timer->it_clock);
2230
2231        return 0;
2232}
2233
2234static const struct seq_operations proc_timers_seq_ops = {
2235        .start  = timers_start,
2236        .next   = timers_next,
2237        .stop   = timers_stop,
2238        .show   = show_timer,
2239};
2240
2241static int proc_timers_open(struct inode *inode, struct file *file)
2242{
2243        struct timers_private *tp;
2244
2245        tp = __seq_open_private(file, &proc_timers_seq_ops,
2246                        sizeof(struct timers_private));
2247        if (!tp)
2248                return -ENOMEM;
2249
2250        tp->pid = proc_pid(inode);
2251        tp->ns = inode->i_sb->s_fs_info;
2252        return 0;
2253}
2254
2255static const struct file_operations proc_timers_operations = {
2256        .open           = proc_timers_open,
2257        .read           = seq_read,
2258        .llseek         = seq_lseek,
2259        .release        = seq_release_private,
2260};
2261#endif
2262
2263static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2264                                        size_t count, loff_t *offset)
2265{
2266        struct inode *inode = file_inode(file);
2267        struct task_struct *p;
2268        u64 slack_ns;
2269        int err;
2270
2271        err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2272        if (err < 0)
2273                return err;
2274
2275        p = get_proc_task(inode);
2276        if (!p)
2277                return -ESRCH;
2278
2279        if (ptrace_may_access(p, PTRACE_MODE_ATTACH_FSCREDS)) {
2280                task_lock(p);
2281                if (slack_ns == 0)
2282                        p->timer_slack_ns = p->default_timer_slack_ns;
2283                else
2284                        p->timer_slack_ns = slack_ns;
2285                task_unlock(p);
2286        } else
2287                count = -EPERM;
2288
2289        put_task_struct(p);
2290
2291        return count;
2292}
2293
2294static int timerslack_ns_show(struct seq_file *m, void *v)
2295{
2296        struct inode *inode = m->private;
2297        struct task_struct *p;
2298        int err =  0;
2299
2300        p = get_proc_task(inode);
2301        if (!p)
2302                return -ESRCH;
2303
2304        if (ptrace_may_access(p, PTRACE_MODE_ATTACH_FSCREDS)) {
2305                task_lock(p);
2306                seq_printf(m, "%llu\n", p->timer_slack_ns);
2307                task_unlock(p);
2308        } else
2309                err = -EPERM;
2310
2311        put_task_struct(p);
2312
2313        return err;
2314}
2315
2316static int timerslack_ns_open(struct inode *inode, struct file *filp)
2317{
2318        return single_open(filp, timerslack_ns_show, inode);
2319}
2320
2321static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2322        .open           = timerslack_ns_open,
2323        .read           = seq_read,
2324        .write          = timerslack_ns_write,
2325        .llseek         = seq_lseek,
2326        .release        = single_release,
2327};
2328
2329static int proc_pident_instantiate(struct inode *dir,
2330        struct dentry *dentry, struct task_struct *task, const void *ptr)
2331{
2332        const struct pid_entry *p = ptr;
2333        struct inode *inode;
2334        struct proc_inode *ei;
2335
2336        inode = proc_pid_make_inode(dir->i_sb, task);
2337        if (!inode)
2338                goto out;
2339
2340        ei = PROC_I(inode);
2341        inode->i_mode = p->mode;
2342        if (S_ISDIR(inode->i_mode))
2343                set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2344        if (p->iop)
2345                inode->i_op = p->iop;
2346        if (p->fop)
2347                inode->i_fop = p->fop;
2348        ei->op = p->op;
2349        d_set_d_op(dentry, &pid_dentry_operations);
2350        d_add(dentry, inode);
2351        /* Close the race of the process dying before we return the dentry */
2352        if (pid_revalidate(dentry, 0))
2353                return 0;
2354out:
2355        return -ENOENT;
2356}
2357
2358static struct dentry *proc_pident_lookup(struct inode *dir, 
2359                                         struct dentry *dentry,
2360                                         const struct pid_entry *ents,
2361                                         unsigned int nents)
2362{
2363        int error;
2364        struct task_struct *task = get_proc_task(dir);
2365        const struct pid_entry *p, *last;
2366
2367        error = -ENOENT;
2368
2369        if (!task)
2370                goto out_no_task;
2371
2372        /*
2373         * Yes, it does not scale. And it should not. Don't add
2374         * new entries into /proc/<tgid>/ without very good reasons.
2375         */
2376        last = &ents[nents - 1];
2377        for (p = ents; p <= last; p++) {
2378                if (p->len != dentry->d_name.len)
2379                        continue;
2380                if (!memcmp(dentry->d_name.name, p->name, p->len))
2381                        break;
2382        }
2383        if (p > last)
2384                goto out;
2385
2386        error = proc_pident_instantiate(dir, dentry, task, p);
2387out:
2388        put_task_struct(task);
2389out_no_task:
2390        return ERR_PTR(error);
2391}
2392
2393static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2394                const struct pid_entry *ents, unsigned int nents)
2395{
2396        struct task_struct *task = get_proc_task(file_inode(file));
2397        const struct pid_entry *p;
2398
2399        if (!task)
2400                return -ENOENT;
2401
2402        if (!dir_emit_dots(file, ctx))
2403                goto out;
2404
2405        if (ctx->pos >= nents + 2)
2406                goto out;
2407
2408        for (p = ents + (ctx->pos - 2); p <= ents + nents - 1; p++) {
2409                if (!proc_fill_cache(file, ctx, p->name, p->len,
2410                                proc_pident_instantiate, task, p))
2411                        break;
2412                ctx->pos++;
2413        }
2414out:
2415        put_task_struct(task);
2416        return 0;
2417}
2418
2419#ifdef CONFIG_SECURITY
2420static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2421                                  size_t count, loff_t *ppos)
2422{
2423        struct inode * inode = file_inode(file);
2424        char *p = NULL;
2425        ssize_t length;
2426        struct task_struct *task = get_proc_task(inode);
2427
2428        if (!task)
2429                return -ESRCH;
2430
2431        length = security_getprocattr(task,
2432                                      (char*)file->f_path.dentry->d_name.name,
2433                                      &p);
2434        put_task_struct(task);
2435        if (length > 0)
2436                length = simple_read_from_buffer(buf, count, ppos, p, length);
2437        kfree(p);
2438        return length;
2439}
2440
2441static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2442                                   size_t count, loff_t *ppos)
2443{
2444        struct inode * inode = file_inode(file);
2445        void *page;
2446        ssize_t length;
2447        struct task_struct *task = get_proc_task(inode);
2448
2449        length = -ESRCH;
2450        if (!task)
2451                goto out_no_task;
2452        if (count > PAGE_SIZE)
2453                count = PAGE_SIZE;
2454
2455        /* No partial writes. */
2456        length = -EINVAL;
2457        if (*ppos != 0)
2458                goto out;
2459
2460        page = memdup_user(buf, count);
2461        if (IS_ERR(page)) {
2462                length = PTR_ERR(page);
2463                goto out;
2464        }
2465
2466        /* Guard against adverse ptrace interaction */
2467        length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2468        if (length < 0)
2469                goto out_free;
2470
2471        length = security_setprocattr(task,
2472                                      (char*)file->f_path.dentry->d_name.name,
2473                                      page, count);
2474        mutex_unlock(&task->signal->cred_guard_mutex);
2475out_free:
2476        kfree(page);
2477out:
2478        put_task_struct(task);
2479out_no_task:
2480        return length;
2481}
2482
2483static const struct file_operations proc_pid_attr_operations = {
2484        .read           = proc_pid_attr_read,
2485        .write          = proc_pid_attr_write,
2486        .llseek         = generic_file_llseek,
2487};
2488
2489static const struct pid_entry attr_dir_stuff[] = {
2490        REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2491        REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2492        REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2493        REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2494        REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2495        REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2496};
2497
2498static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2499{
2500        return proc_pident_readdir(file, ctx, 
2501                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2502}
2503
2504static const struct file_operations proc_attr_dir_operations = {
2505        .read           = generic_read_dir,
2506        .iterate_shared = proc_attr_dir_readdir,
2507        .llseek         = generic_file_llseek,
2508};
2509
2510static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2511                                struct dentry *dentry, unsigned int flags)
2512{
2513        return proc_pident_lookup(dir, dentry,
2514                                  attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2515}
2516
2517static const struct inode_operations proc_attr_dir_inode_operations = {
2518        .lookup         = proc_attr_dir_lookup,
2519        .getattr        = pid_getattr,
2520        .setattr        = proc_setattr,
2521};
2522
2523#endif
2524
2525#ifdef CONFIG_ELF_CORE
2526static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2527                                         size_t count, loff_t *ppos)
2528{
2529        struct task_struct *task = get_proc_task(file_inode(file));
2530        struct mm_struct *mm;
2531        char buffer[PROC_NUMBUF];
2532        size_t len;
2533        int ret;
2534
2535        if (!task)
2536                return -ESRCH;
2537
2538        ret = 0;
2539        mm = get_task_mm(task);
2540        if (mm) {
2541                len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2542                               ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2543                                MMF_DUMP_FILTER_SHIFT));
2544                mmput(mm);
2545                ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2546        }
2547
2548        put_task_struct(task);
2549
2550        return ret;
2551}
2552
2553static ssize_t proc_coredump_filter_write(struct file *file,
2554                                          const char __user *buf,
2555                                          size_t count,
2556                                          loff_t *ppos)
2557{
2558        struct task_struct *task;
2559        struct mm_struct *mm;
2560        unsigned int val;
2561        int ret;
2562        int i;
2563        unsigned long mask;
2564
2565        ret = kstrtouint_from_user(buf, count, 0, &val);
2566        if (ret < 0)
2567                return ret;
2568
2569        ret = -ESRCH;
2570        task = get_proc_task(file_inode(file));
2571        if (!task)
2572                goto out_no_task;
2573
2574        mm = get_task_mm(task);
2575        if (!mm)
2576                goto out_no_mm;
2577        ret = 0;
2578
2579        for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2580                if (val & mask)
2581                        set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2582                else
2583                        clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2584        }
2585
2586        mmput(mm);
2587 out_no_mm:
2588        put_task_struct(task);
2589 out_no_task:
2590        if (ret < 0)
2591                return ret;
2592        return count;
2593}
2594
2595static const struct file_operations proc_coredump_filter_operations = {
2596        .read           = proc_coredump_filter_read,
2597        .write          = proc_coredump_filter_write,
2598        .llseek         = generic_file_llseek,
2599};
2600#endif
2601
2602#ifdef CONFIG_TASK_IO_ACCOUNTING
2603static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2604{
2605        struct task_io_accounting acct = task->ioac;
2606        unsigned long flags;
2607        int result;
2608
2609        result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2610        if (result)
2611                return result;
2612
2613        if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2614                result = -EACCES;
2615                goto out_unlock;
2616        }
2617
2618        if (whole && lock_task_sighand(task, &flags)) {
2619                struct task_struct *t = task;
2620
2621                task_io_accounting_add(&acct, &task->signal->ioac);
2622                while_each_thread(task, t)
2623                        task_io_accounting_add(&acct, &t->ioac);
2624
2625                unlock_task_sighand(task, &flags);
2626        }
2627        seq_printf(m,
2628                   "rchar: %llu\n"
2629                   "wchar: %llu\n"
2630                   "syscr: %llu\n"
2631                   "syscw: %llu\n"
2632                   "read_bytes: %llu\n"
2633                   "write_bytes: %llu\n"
2634                   "cancelled_write_bytes: %llu\n",
2635                   (unsigned long long)acct.rchar,
2636                   (unsigned long long)acct.wchar,
2637                   (unsigned long long)acct.syscr,
2638                   (unsigned long long)acct.syscw,
2639                   (unsigned long long)acct.read_bytes,
2640                   (unsigned long long)acct.write_bytes,
2641                   (unsigned long long)acct.cancelled_write_bytes);
2642        result = 0;
2643
2644out_unlock:
2645        mutex_unlock(&task->signal->cred_guard_mutex);
2646        return result;
2647}
2648
2649static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2650                                  struct pid *pid, struct task_struct *task)
2651{
2652        return do_io_accounting(task, m, 0);
2653}
2654
2655static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2656                                   struct pid *pid, struct task_struct *task)
2657{
2658        return do_io_accounting(task, m, 1);
2659}
2660#endif /* CONFIG_TASK_IO_ACCOUNTING */
2661
2662#ifdef CONFIG_USER_NS
2663static int proc_id_map_open(struct inode *inode, struct file *file,
2664        const struct seq_operations *seq_ops)
2665{
2666        struct user_namespace *ns = NULL;
2667        struct task_struct *task;
2668        struct seq_file *seq;
2669        int ret = -EINVAL;
2670
2671        task = get_proc_task(inode);
2672        if (task) {
2673                rcu_read_lock();
2674                ns = get_user_ns(task_cred_xxx(task, user_ns));
2675                rcu_read_unlock();
2676                put_task_struct(task);
2677        }
2678        if (!ns)
2679                goto err;
2680
2681        ret = seq_open(file, seq_ops);
2682        if (ret)
2683                goto err_put_ns;
2684
2685        seq = file->private_data;
2686        seq->private = ns;
2687
2688        return 0;
2689err_put_ns:
2690        put_user_ns(ns);
2691err:
2692        return ret;
2693}
2694
2695static int proc_id_map_release(struct inode *inode, struct file *file)
2696{
2697        struct seq_file *seq = file->private_data;
2698        struct user_namespace *ns = seq->private;
2699        put_user_ns(ns);
2700        return seq_release(inode, file);
2701}
2702
2703static int proc_uid_map_open(struct inode *inode, struct file *file)
2704{
2705        return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2706}
2707
2708static int proc_gid_map_open(struct inode *inode, struct file *file)
2709{
2710        return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2711}
2712
2713static int proc_projid_map_open(struct inode *inode, struct file *file)
2714{
2715        return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2716}
2717
2718static const struct file_operations proc_uid_map_operations = {
2719        .open           = proc_uid_map_open,
2720        .write          = proc_uid_map_write,
2721        .read           = seq_read,
2722        .llseek         = seq_lseek,
2723        .release        = proc_id_map_release,
2724};
2725
2726static const struct file_operations proc_gid_map_operations = {
2727        .open           = proc_gid_map_open,
2728        .write          = proc_gid_map_write,
2729        .read           = seq_read,
2730        .llseek         = seq_lseek,
2731        .release        = proc_id_map_release,
2732};
2733
2734static const struct file_operations proc_projid_map_operations = {
2735        .open           = proc_projid_map_open,
2736        .write          = proc_projid_map_write,
2737        .read           = seq_read,
2738        .llseek         = seq_lseek,
2739        .release        = proc_id_map_release,
2740};
2741
2742static int proc_setgroups_open(struct inode *inode, struct file *file)
2743{
2744        struct user_namespace *ns = NULL;
2745        struct task_struct *task;
2746        int ret;
2747
2748        ret = -ESRCH;
2749        task = get_proc_task(inode);
2750        if (task) {
2751                rcu_read_lock();
2752                ns = get_user_ns(task_cred_xxx(task, user_ns));
2753                rcu_read_unlock();
2754                put_task_struct(task);
2755        }
2756        if (!ns)
2757                goto err;
2758
2759        if (file->f_mode & FMODE_WRITE) {
2760                ret = -EACCES;
2761                if (!ns_capable(ns, CAP_SYS_ADMIN))
2762                        goto err_put_ns;
2763        }
2764
2765        ret = single_open(file, &proc_setgroups_show, ns);
2766        if (ret)
2767                goto err_put_ns;
2768
2769        return 0;
2770err_put_ns:
2771        put_user_ns(ns);
2772err:
2773        return ret;
2774}
2775
2776static int proc_setgroups_release(struct inode *inode, struct file *file)
2777{
2778        struct seq_file *seq = file->private_data;
2779        struct user_namespace *ns = seq->private;
2780        int ret = single_release(inode, file);
2781        put_user_ns(ns);
2782        return ret;
2783}
2784
2785static const struct file_operations proc_setgroups_operations = {
2786        .open           = proc_setgroups_open,
2787        .write          = proc_setgroups_write,
2788        .read           = seq_read,
2789        .llseek         = seq_lseek,
2790        .release        = proc_setgroups_release,
2791};
2792#endif /* CONFIG_USER_NS */
2793
2794static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2795                                struct pid *pid, struct task_struct *task)
2796{
2797        int err = lock_trace(task);
2798        if (!err) {
2799                seq_printf(m, "%08x\n", task->personality);
2800                unlock_trace(task);
2801        }
2802        return err;
2803}
2804
2805/*
2806 * Thread groups
2807 */
2808static const struct file_operations proc_task_operations;
2809static const struct inode_operations proc_task_inode_operations;
2810
2811static const struct pid_entry tgid_base_stuff[] = {
2812        DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2813        DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2814        DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2815        DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2816        DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2817#ifdef CONFIG_NET
2818        DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2819#endif
2820        REG("environ",    S_IRUSR, proc_environ_operations),
2821        ONE("auxv",       S_IRUSR, proc_pid_auxv),
2822        ONE("status",     S_IRUGO, proc_pid_status),
2823        ONE("personality", S_IRUSR, proc_pid_personality),
2824        ONE("limits",     S_IRUGO, proc_pid_limits),
2825#ifdef CONFIG_SCHED_DEBUG
2826        REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2827#endif
2828#ifdef CONFIG_SCHED_AUTOGROUP
2829        REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2830#endif
2831        REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2832#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2833        ONE("syscall",    S_IRUSR, proc_pid_syscall),
2834#endif
2835        REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
2836        ONE("stat",       S_IRUGO, proc_tgid_stat),
2837        ONE("statm",      S_IRUGO, proc_pid_statm),
2838        REG("maps",       S_IRUGO, proc_pid_maps_operations),
2839#ifdef CONFIG_NUMA
2840        REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2841#endif
2842        REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2843        LNK("cwd",        proc_cwd_link),
2844        LNK("root",       proc_root_link),
2845        LNK("exe",        proc_exe_link),
2846        REG("mounts",     S_IRUGO, proc_mounts_operations),
2847        REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2848        REG("mountstats", S_IRUSR, proc_mountstats_operations),
2849#ifdef CONFIG_PROC_PAGE_MONITOR
2850        REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2851        REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2852        REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2853#endif
2854#ifdef CONFIG_SECURITY
2855        DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2856#endif
2857#ifdef CONFIG_KALLSYMS
2858        ONE("wchan",      S_IRUGO, proc_pid_wchan),
2859#endif
2860#ifdef CONFIG_STACKTRACE
2861        ONE("stack",      S_IRUSR, proc_pid_stack),
2862#endif
2863#ifdef CONFIG_SCHED_INFO
2864        ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
2865#endif
2866#ifdef CONFIG_LATENCYTOP
2867        REG("latency",  S_IRUGO, proc_lstats_operations),
2868#endif
2869#ifdef CONFIG_PROC_PID_CPUSET
2870        ONE("cpuset",     S_IRUGO, proc_cpuset_show),
2871#endif
2872#ifdef CONFIG_CGROUPS
2873        ONE("cgroup",  S_IRUGO, proc_cgroup_show),
2874#endif
2875        ONE("oom_score",  S_IRUGO, proc_oom_score),
2876        REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2877        REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2878#ifdef CONFIG_AUDITSYSCALL
2879        REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2880        REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2881#endif
2882#ifdef CONFIG_FAULT_INJECTION
2883        REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2884#endif
2885#ifdef CONFIG_ELF_CORE
2886        REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2887#endif
2888#ifdef CONFIG_TASK_IO_ACCOUNTING
2889        ONE("io",       S_IRUSR, proc_tgid_io_accounting),
2890#endif
2891#ifdef CONFIG_HARDWALL
2892        ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
2893#endif
2894#ifdef CONFIG_USER_NS
2895        REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2896        REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2897        REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2898        REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
2899#endif
2900#ifdef CONFIG_CHECKPOINT_RESTORE
2901        REG("timers",     S_IRUGO, proc_timers_operations),
2902#endif
2903        REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
2904};
2905
2906static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2907{
2908        return proc_pident_readdir(file, ctx,
2909                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2910}
2911
2912static const struct file_operations proc_tgid_base_operations = {
2913        .read           = generic_read_dir,
2914        .iterate_shared = proc_tgid_base_readdir,
2915        .llseek         = generic_file_llseek,
2916};
2917
2918static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2919{
2920        return proc_pident_lookup(dir, dentry,
2921                                  tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2922}
2923
2924static const struct inode_operations proc_tgid_base_inode_operations = {
2925        .lookup         = proc_tgid_base_lookup,
2926        .getattr        = pid_getattr,
2927        .setattr        = proc_setattr,
2928        .permission     = proc_pid_permission,
2929};
2930
2931static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2932{
2933        struct dentry *dentry, *leader, *dir;
2934        char buf[PROC_NUMBUF];
2935        struct qstr name;
2936
2937        name.name = buf;
2938        name.len = snprintf(buf, sizeof(buf), "%d", pid);
2939        /* no ->d_hash() rejects on procfs */
2940        dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2941        if (dentry) {
2942                d_invalidate(dentry);
2943                dput(dentry);
2944        }
2945
2946        if (pid == tgid)
2947                return;
2948
2949        name.name = buf;
2950        name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2951        leader = d_hash_and_lookup(mnt->mnt_root, &name);
2952        if (!leader)
2953                goto out;
2954
2955        name.name = "task";
2956        name.len = strlen(name.name);
2957        dir = d_hash_and_lookup(leader, &name);
2958        if (!dir)
2959                goto out_put_leader;
2960
2961        name.name = buf;
2962        name.len = snprintf(buf, sizeof(buf), "%d", pid);
2963        dentry = d_hash_and_lookup(dir, &name);
2964        if (dentry) {
2965                d_invalidate(dentry);
2966                dput(dentry);
2967        }
2968
2969        dput(dir);
2970out_put_leader:
2971        dput(leader);
2972out:
2973        return;
2974}
2975
2976/**
2977 * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2978 * @task: task that should be flushed.
2979 *
2980 * When flushing dentries from proc, one needs to flush them from global
2981 * proc (proc_mnt) and from all the namespaces' procs this task was seen
2982 * in. This call is supposed to do all of this job.
2983 *
2984 * Looks in the dcache for
2985 * /proc/@pid
2986 * /proc/@tgid/task/@pid
2987 * if either directory is present flushes it and all of it'ts children
2988 * from the dcache.
2989 *
2990 * It is safe and reasonable to cache /proc entries for a task until
2991 * that task exits.  After that they just clog up the dcache with
2992 * useless entries, possibly causing useful dcache entries to be
2993 * flushed instead.  This routine is proved to flush those useless
2994 * dcache entries at process exit time.
2995 *
2996 * NOTE: This routine is just an optimization so it does not guarantee
2997 *       that no dcache entries will exist at process exit time it
2998 *       just makes it very unlikely that any will persist.
2999 */
3000
3001void proc_flush_task(struct task_struct *task)
3002{
3003        int i;
3004        struct pid *pid, *tgid;
3005        struct upid *upid;
3006
3007        pid = task_pid(task);
3008        tgid = task_tgid(task);
3009
3010        for (i = 0; i <= pid->level; i++) {
3011                upid = &pid->numbers[i];
3012                proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3013                                        tgid->numbers[i].nr);
3014        }
3015}
3016
3017static int proc_pid_instantiate(struct inode *dir,
3018                                   struct dentry * dentry,
3019                                   struct task_struct *task, const void *ptr)
3020{
3021        struct inode *inode;
3022
3023        inode = proc_pid_make_inode(dir->i_sb, task);
3024        if (!inode)
3025                goto out;
3026
3027        inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3028        inode->i_op = &proc_tgid_base_inode_operations;
3029        inode->i_fop = &proc_tgid_base_operations;
3030        inode->i_flags|=S_IMMUTABLE;
3031
3032        set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
3033                                                  ARRAY_SIZE(tgid_base_stuff)));
3034
3035        d_set_d_op(dentry, &pid_dentry_operations);
3036
3037        d_add(dentry, inode);
3038        /* Close the race of the process dying before we return the dentry */
3039        if (pid_revalidate(dentry, 0))
3040                return 0;
3041out:
3042        return -ENOENT;
3043}
3044
3045struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3046{
3047        int result = -ENOENT;
3048        struct task_struct *task;
3049        unsigned tgid;
3050        struct pid_namespace *ns;
3051
3052        tgid = name_to_int(&dentry->d_name);
3053        if (tgid == ~0U)
3054                goto out;
3055
3056        ns = dentry->d_sb->s_fs_info;
3057        rcu_read_lock();
3058        task = find_task_by_pid_ns(tgid, ns);
3059        if (task)
3060                get_task_struct(task);
3061        rcu_read_unlock();
3062        if (!task)
3063                goto out;
3064
3065        result = proc_pid_instantiate(dir, dentry, task, NULL);
3066        put_task_struct(task);
3067out:
3068        return ERR_PTR(result);
3069}
3070
3071/*
3072 * Find the first task with tgid >= tgid
3073 *
3074 */
3075struct tgid_iter {
3076        unsigned int tgid;
3077        struct task_struct *task;
3078};
3079static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3080{
3081        struct pid *pid;
3082
3083        if (iter.task)
3084                put_task_struct(iter.task);
3085        rcu_read_lock();
3086retry:
3087        iter.task = NULL;
3088        pid = find_ge_pid(iter.tgid, ns);
3089        if (pid) {
3090                iter.tgid = pid_nr_ns(pid, ns);
3091                iter.task = pid_task(pid, PIDTYPE_PID);
3092                /* What we to know is if the pid we have find is the
3093                 * pid of a thread_group_leader.  Testing for task
3094                 * being a thread_group_leader is the obvious thing
3095                 * todo but there is a window when it fails, due to
3096                 * the pid transfer logic in de_thread.
3097                 *
3098                 * So we perform the straight forward test of seeing
3099                 * if the pid we have found is the pid of a thread
3100                 * group leader, and don't worry if the task we have
3101                 * found doesn't happen to be a thread group leader.
3102                 * As we don't care in the case of readdir.
3103                 */
3104                if (!iter.task || !has_group_leader_pid(iter.task)) {
3105                        iter.tgid += 1;
3106                        goto retry;
3107                }
3108                get_task_struct(iter.task);
3109        }
3110        rcu_read_unlock();
3111        return iter;
3112}
3113
3114#define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3115
3116/* for the /proc/ directory itself, after non-process stuff has been done */
3117int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3118{
3119        struct tgid_iter iter;
3120        struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
3121        loff_t pos = ctx->pos;
3122
3123        if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3124                return 0;
3125
3126        if (pos == TGID_OFFSET - 2) {
3127                struct inode *inode = d_inode(ns->proc_self);
3128                if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3129                        return 0;
3130                ctx->pos = pos = pos + 1;
3131        }
3132        if (pos == TGID_OFFSET - 1) {
3133                struct inode *inode = d_inode(ns->proc_thread_self);
3134                if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3135                        return 0;
3136                ctx->pos = pos = pos + 1;
3137        }
3138        iter.tgid = pos - TGID_OFFSET;
3139        iter.task = NULL;
3140        for (iter = next_tgid(ns, iter);
3141             iter.task;
3142             iter.tgid += 1, iter = next_tgid(ns, iter)) {
3143                char name[PROC_NUMBUF];
3144                int len;
3145                if (!has_pid_permissions(ns, iter.task, 2))
3146                        continue;
3147
3148                len = snprintf(name, sizeof(name), "%d", iter.tgid);
3149                ctx->pos = iter.tgid + TGID_OFFSET;
3150                if (!proc_fill_cache(file, ctx, name, len,
3151                                     proc_pid_instantiate, iter.task, NULL)) {
3152                        put_task_struct(iter.task);
3153                        return 0;
3154                }
3155        }
3156        ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3157        return 0;
3158}
3159
3160/*
3161 * proc_tid_comm_permission is a special permission function exclusively
3162 * used for the node /proc/<pid>/task/<tid>/comm.
3163 * It bypasses generic permission checks in the case where a task of the same
3164 * task group attempts to access the node.
3165 * The rationale behind this is that glibc and bionic access this node for
3166 * cross thread naming (pthread_set/getname_np(!self)). However, if
3167 * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3168 * which locks out the cross thread naming implementation.
3169 * This function makes sure that the node is always accessible for members of
3170 * same thread group.
3171 */
3172static int proc_tid_comm_permission(struct inode *inode, int mask)
3173{
3174        bool is_same_tgroup;
3175        struct task_struct *task;
3176
3177        task = get_proc_task(inode);
3178        if (!task)
3179                return -ESRCH;
3180        is_same_tgroup = same_thread_group(current, task);
3181        put_task_struct(task);
3182
3183        if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3184                /* This file (/proc/<pid>/task/<tid>/comm) can always be
3185                 * read or written by the members of the corresponding
3186                 * thread group.
3187                 */
3188                return 0;
3189        }
3190
3191        return generic_permission(inode, mask);
3192}
3193
3194static const struct inode_operations proc_tid_comm_inode_operations = {
3195                .permission = proc_tid_comm_permission,
3196};
3197
3198/*
3199 * Tasks
3200 */
3201static const struct pid_entry tid_base_stuff[] = {
3202        DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3203        DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3204        DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3205#ifdef CONFIG_NET
3206        DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3207#endif
3208        REG("environ",   S_IRUSR, proc_environ_operations),
3209        ONE("auxv",      S_IRUSR, proc_pid_auxv),
3210        ONE("status",    S_IRUGO, proc_pid_status),
3211        ONE("personality", S_IRUSR, proc_pid_personality),
3212        ONE("limits",    S_IRUGO, proc_pid_limits),
3213#ifdef CONFIG_SCHED_DEBUG
3214        REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3215#endif
3216        NOD("comm",      S_IFREG|S_IRUGO|S_IWUSR,
3217                         &proc_tid_comm_inode_operations,
3218                         &proc_pid_set_comm_operations, {}),
3219#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3220        ONE("syscall",   S_IRUSR, proc_pid_syscall),
3221#endif
3222        REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3223        ONE("stat",      S_IRUGO, proc_tid_stat),
3224        ONE("statm",     S_IRUGO, proc_pid_statm),
3225        REG("maps",      S_IRUGO, proc_tid_maps_operations),
3226#ifdef CONFIG_PROC_CHILDREN
3227        REG("children",  S_IRUGO, proc_tid_children_operations),
3228#endif
3229#ifdef CONFIG_NUMA
3230        REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3231#endif
3232        REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3233        LNK("cwd",       proc_cwd_link),
3234        LNK("root",      proc_root_link),
3235        LNK("exe",       proc_exe_link),
3236        REG("mounts",    S_IRUGO, proc_mounts_operations),
3237        REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3238#ifdef CONFIG_PROC_PAGE_MONITOR
3239        REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3240        REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3241        REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3242#endif
3243#ifdef CONFIG_SECURITY
3244        DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3245#endif
3246#ifdef CONFIG_KALLSYMS
3247        ONE("wchan",     S_IRUGO, proc_pid_wchan),
3248#endif
3249#ifdef CONFIG_STACKTRACE
3250        ONE("stack",      S_IRUSR, proc_pid_stack),
3251#endif
3252#ifdef CONFIG_SCHED_INFO
3253        ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3254#endif
3255#ifdef CONFIG_LATENCYTOP
3256        REG("latency",  S_IRUGO, proc_lstats_operations),
3257#endif
3258#ifdef CONFIG_PROC_PID_CPUSET
3259        ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3260#endif
3261#ifdef CONFIG_CGROUPS
3262        ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3263#endif
3264        ONE("oom_score", S_IRUGO, proc_oom_score),
3265        REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3266        REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3267#ifdef CONFIG_AUDITSYSCALL
3268        REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3269        REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3270#endif
3271#ifdef CONFIG_FAULT_INJECTION
3272        REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3273#endif
3274#ifdef CONFIG_TASK_IO_ACCOUNTING
3275        ONE("io",       S_IRUSR, proc_tid_io_accounting),
3276#endif
3277#ifdef CONFIG_HARDWALL
3278        ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
3279#endif
3280#ifdef CONFIG_USER_NS
3281        REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3282        REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3283        REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3284        REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3285#endif
3286};
3287
3288static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3289{
3290        return proc_pident_readdir(file, ctx,
3291                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3292}
3293
3294static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3295{
3296        return proc_pident_lookup(dir, dentry,
3297                                  tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3298}
3299
3300static const struct file_operations proc_tid_base_operations = {
3301        .read           = generic_read_dir,
3302        .iterate_shared = proc_tid_base_readdir,
3303        .llseek         = generic_file_llseek,
3304};
3305
3306static const struct inode_operations proc_tid_base_inode_operations = {
3307        .lookup         = proc_tid_base_lookup,
3308        .getattr        = pid_getattr,
3309        .setattr        = proc_setattr,
3310};
3311
3312static int proc_task_instantiate(struct inode *dir,
3313        struct dentry *dentry, struct task_struct *task, const void *ptr)
3314{
3315        struct inode *inode;
3316        inode = proc_pid_make_inode(dir->i_sb, task);
3317
3318        if (!inode)
3319                goto out;
3320        inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3321        inode->i_op = &proc_tid_base_inode_operations;
3322        inode->i_fop = &proc_tid_base_operations;
3323        inode->i_flags|=S_IMMUTABLE;
3324
3325        set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3326                                                  ARRAY_SIZE(tid_base_stuff)));
3327
3328        d_set_d_op(dentry, &pid_dentry_operations);
3329
3330        d_add(dentry, inode);
3331        /* Close the race of the process dying before we return the dentry */
3332        if (pid_revalidate(dentry, 0))
3333                return 0;
3334out:
3335        return -ENOENT;
3336}
3337
3338static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3339{
3340        int result = -ENOENT;
3341        struct task_struct *task;
3342        struct task_struct *leader = get_proc_task(dir);
3343        unsigned tid;
3344        struct pid_namespace *ns;
3345
3346        if (!leader)
3347                goto out_no_task;
3348
3349        tid = name_to_int(&dentry->d_name);
3350        if (tid == ~0U)
3351                goto out;
3352
3353        ns = dentry->d_sb->s_fs_info;
3354        rcu_read_lock();
3355        task = find_task_by_pid_ns(tid, ns);
3356        if (task)
3357                get_task_struct(task);
3358        rcu_read_unlock();
3359        if (!task)
3360                goto out;
3361        if (!same_thread_group(leader, task))
3362                goto out_drop_task;
3363
3364        result = proc_task_instantiate(dir, dentry, task, NULL);
3365out_drop_task:
3366        put_task_struct(task);
3367out:
3368        put_task_struct(leader);
3369out_no_task:
3370        return ERR_PTR(result);
3371}
3372
3373/*
3374 * Find the first tid of a thread group to return to user space.
3375 *
3376 * Usually this is just the thread group leader, but if the users
3377 * buffer was too small or there was a seek into the middle of the
3378 * directory we have more work todo.
3379 *
3380 * In the case of a short read we start with find_task_by_pid.
3381 *
3382 * In the case of a seek we start with the leader and walk nr
3383 * threads past it.
3384 */
3385static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3386                                        struct pid_namespace *ns)
3387{
3388        struct task_struct *pos, *task;
3389        unsigned long nr = f_pos;
3390
3391        if (nr != f_pos)        /* 32bit overflow? */
3392                return NULL;
3393
3394        rcu_read_lock();
3395        task = pid_task(pid, PIDTYPE_PID);
3396        if (!task)
3397                goto fail;
3398
3399        /* Attempt to start with the tid of a thread */
3400        if (tid && nr) {
3401                pos = find_task_by_pid_ns(tid, ns);
3402                if (pos && same_thread_group(pos, task))
3403                        goto found;
3404        }
3405
3406        /* If nr exceeds the number of threads there is nothing todo */
3407        if (nr >= get_nr_threads(task))
3408                goto fail;
3409
3410        /* If we haven't found our starting place yet start
3411         * with the leader and walk nr threads forward.
3412         */
3413        pos = task = task->group_leader;
3414        do {
3415                if (!nr--)
3416                        goto found;
3417        } while_each_thread(task, pos);
3418fail:
3419        pos = NULL;
3420        goto out;
3421found:
3422        get_task_struct(pos);
3423out:
3424        rcu_read_unlock();
3425        return pos;
3426}
3427
3428/*
3429 * Find the next thread in the thread list.
3430 * Return NULL if there is an error or no next thread.
3431 *
3432 * The reference to the input task_struct is released.
3433 */
3434static struct task_struct *next_tid(struct task_struct *start)
3435{
3436        struct task_struct *pos = NULL;
3437        rcu_read_lock();
3438        if (pid_alive(start)) {
3439                pos = next_thread(start);
3440                if (thread_group_leader(pos))
3441                        pos = NULL;
3442                else
3443                        get_task_struct(pos);
3444        }
3445        rcu_read_unlock();
3446        put_task_struct(start);
3447        return pos;
3448}
3449
3450/* for the /proc/TGID/task/ directories */
3451static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3452{
3453        struct inode *inode = file_inode(file);
3454        struct task_struct *task;
3455        struct pid_namespace *ns;
3456        int tid;
3457
3458        if (proc_inode_is_dead(inode))
3459                return -ENOENT;
3460
3461        if (!dir_emit_dots(file, ctx))
3462                return 0;
3463
3464        /* f_version caches the tgid value that the last readdir call couldn't
3465         * return. lseek aka telldir automagically resets f_version to 0.
3466         */
3467        ns = inode->i_sb->s_fs_info;
3468        tid = (int)file->f_version;
3469        file->f_version = 0;
3470        for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3471             task;
3472             task = next_tid(task), ctx->pos++) {
3473                char name[PROC_NUMBUF];
3474                int len;
3475                tid = task_pid_nr_ns(task, ns);
3476                len = snprintf(name, sizeof(name), "%d", tid);
3477                if (!proc_fill_cache(file, ctx, name, len,
3478                                proc_task_instantiate, task, NULL)) {
3479                        /* returning this tgid failed, save it as the first
3480                         * pid for the next readir call */
3481                        file->f_version = (u64)tid;
3482                        put_task_struct(task);
3483                        break;
3484                }
3485        }
3486
3487        return 0;
3488}
3489
3490static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3491{
3492        struct inode *inode = d_inode(dentry);
3493        struct task_struct *p = get_proc_task(inode);
3494        generic_fillattr(inode, stat);
3495
3496        if (p) {
3497                stat->nlink += get_nr_threads(p);
3498                put_task_struct(p);
3499        }
3500
3501        return 0;
3502}
3503
3504static const struct inode_operations proc_task_inode_operations = {
3505        .lookup         = proc_task_lookup,
3506        .getattr        = proc_task_getattr,
3507        .setattr        = proc_setattr,
3508        .permission     = proc_pid_permission,
3509};
3510
3511static const struct file_operations proc_task_operations = {
3512        .read           = generic_read_dir,
3513        .iterate_shared = proc_task_readdir,
3514        .llseek         = generic_file_llseek,
3515};
3516
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.