LXR linux/mm/process_vm

   1/*
   2 * linux/mm/process_vm_access.c
   3 *
   4 * Copyright (C) 2010-2011 Christopher Yeoh <cyeoh@au1.ibm.com>, IBM Corp.
   5 *
   6 * This program is free software; you can redistribute it and/or
   7 * modify it under the terms of the GNU General Public License
   8 * as published by the Free Software Foundation; either version
   9 * 2 of the License, or (at your option) any later version.
  10 */
  11
  12#include <linux/mm.h>
  13#include <linux/uio.h>
  14#include <linux/sched.h>
  15#include <linux/sched/mm.h>
  16#include <linux/highmem.h>
  17#include <linux/ptrace.h>
  18#include <linux/slab.h>
  19#include <linux/syscalls.h>
  20
  21#ifdef CONFIG_COMPAT
  22#include <linux/compat.h>
  23#endif
  24
  25/**
  26 * process_vm_rw_pages - read/write pages from task specified
  27 * @pages: array of pointers to pages we want to copy
  28 * @start_offset: offset in page to start copying from/to
  29 * @len: number of bytes to copy
  30 * @iter: where to copy to/from locally
  31 * @vm_write: 0 means copy from, 1 means copy to
  32 * Returns 0 on success, error code otherwise
  33 */
  34static int process_vm_rw_pages(struct page **pages,
  35                               unsigned offset,
  36                               size_t len,
  37                               struct iov_iter *iter,
  38                               int vm_write)
  39{
  40        /* Do the copy for each page */
  41        while (len && iov_iter_count(iter)) {
  42                struct page *page = *pages++;
  43                size_t copy = PAGE_SIZE - offset;
  44                size_t copied;
  45
  46                if (copy > len)
  47                        copy = len;
  48
  49                if (vm_write) {
  50                        copied = copy_page_from_iter(page, offset, copy, iter);
  51                        set_page_dirty_lock(page);
  52                } else {
  53                        copied = copy_page_to_iter(page, offset, copy, iter);
  54                }
  55                len -= copied;
  56                if (copied < copy && iov_iter_count(iter))
  57                        return -EFAULT;
  58                offset = 0;
  59        }
  60        return 0;
  61}
  62
  63/* Maximum number of pages kmalloc'd to hold struct page's during copy */
  64#define PVM_MAX_KMALLOC_PAGES (PAGE_SIZE * 2)
  65
  66/**
  67 * process_vm_rw_single_vec - read/write pages from task specified
  68 * @addr: start memory address of target process
  69 * @len: size of area to copy to/from
  70 * @iter: where to copy to/from locally
  71 * @process_pages: struct pages area that can store at least
  72 *  nr_pages_to_copy struct page pointers
  73 * @mm: mm for task
  74 * @task: task to read/write from
  75 * @vm_write: 0 means copy from, 1 means copy to
  76 * Returns 0 on success or on failure error code
  77 */
  78static int process_vm_rw_single_vec(unsigned long addr,
  79                                    unsigned long len,
  80                                    struct iov_iter *iter,
  81                                    struct page **process_pages,
  82                                    struct mm_struct *mm,
  83                                    struct task_struct *task,
  84                                    int vm_write)
  85{
  86        unsigned long pa = addr & PAGE_MASK;
  87        unsigned long start_offset = addr - pa;
  88        unsigned long nr_pages;
  89        ssize_t rc = 0;
  90        unsigned long max_pages_per_loop = PVM_MAX_KMALLOC_PAGES
  91                / sizeof(struct pages *);
  92        unsigned int flags = 0;
  93
  94        /* Work out address and page range required */
  95        if (len == 0)
  96                return 0;
  97        nr_pages = (addr + len - 1) / PAGE_SIZE - addr / PAGE_SIZE + 1;
  98
  99        if (vm_write)
 100                flags |= FOLL_WRITE;
 101
 102        while (!rc && nr_pages && iov_iter_count(iter)) {
 103                int pages = min(nr_pages, max_pages_per_loop);
 104                int locked = 1;
 105                size_t bytes;
 106
 107                /*
 108                 * Get the pages we're interested in.  We must
 109                 * access remotely because task/mm might not
 110                 * current/current->mm
 111                 */
 112                down_read(&mm->mmap_sem);
 113                pages = get_user_pages_remote(task, mm, pa, pages, flags,
 114                                              process_pages, NULL, &locked);
 115                if (locked)
 116                        up_read(&mm->mmap_sem);
 117                if (pages <= 0)
 118                        return -EFAULT;
 119
 120                bytes = pages * PAGE_SIZE - start_offset;
 121                if (bytes > len)
 122                        bytes = len;
 123
 124                rc = process_vm_rw_pages(process_pages,
 125                                         start_offset, bytes, iter,
 126                                         vm_write);
 127                len -= bytes;
 128                start_offset = 0;
 129                nr_pages -= pages;
 130                pa += pages * PAGE_SIZE;
 131                while (pages)
 132                        put_page(process_pages[--pages]);
 133        }
 134
 135        return rc;
 136}
 137
 138/* Maximum number of entries for process pages array
 139   which lives on stack */
 140#define PVM_MAX_PP_ARRAY_COUNT 16
 141
 142/**
 143 * process_vm_rw_core - core of reading/writing pages from task specified
 144 * @pid: PID of process to read/write from/to
 145 * @iter: where to copy to/from locally
 146 * @rvec: iovec array specifying where to copy to/from in the other process
 147 * @riovcnt: size of rvec array
 148 * @flags: currently unused
 149 * @vm_write: 0 if reading from other process, 1 if writing to other process
 150 * Returns the number of bytes read/written or error code. May
 151 *  return less bytes than expected if an error occurs during the copying
 152 *  process.
 153 */
 154static ssize_t process_vm_rw_core(pid_t pid, struct iov_iter *iter,
 155                                  const struct iovec *rvec,
 156                                  unsigned long riovcnt,
 157                                  unsigned long flags, int vm_write)
 158{
 159        struct task_struct *task;
 160        struct page *pp_stack[PVM_MAX_PP_ARRAY_COUNT];
 161        struct page **process_pages = pp_stack;
 162        struct mm_struct *mm;
 163        unsigned long i;
 164        ssize_t rc = 0;
 165        unsigned long nr_pages = 0;
 166        unsigned long nr_pages_iov;
 167        ssize_t iov_len;
 168        size_t total_len = iov_iter_count(iter);
 169
 170        /*
 171         * Work out how many pages of struct pages we're going to need
 172         * when eventually calling get_user_pages
 173         */
 174        for (i = 0; i < riovcnt; i++) {
 175                iov_len = rvec[i].iov_len;
 176                if (iov_len > 0) {
 177                        nr_pages_iov = ((unsigned long)rvec[i].iov_base
 178                                        + iov_len)
 179                                / PAGE_SIZE - (unsigned long)rvec[i].iov_base
 180                                / PAGE_SIZE + 1;
 181                        nr_pages = max(nr_pages, nr_pages_iov);
 182                }
 183        }
 184
 185        if (nr_pages == 0)
 186                return 0;
 187
 188        if (nr_pages > PVM_MAX_PP_ARRAY_COUNT) {
 189                /* For reliability don't try to kmalloc more than
 190                   2 pages worth */
 191                process_pages = kmalloc(min_t(size_t, PVM_MAX_KMALLOC_PAGES,
 192                                              sizeof(struct pages *)*nr_pages),
 193                                        GFP_KERNEL);
 194
 195                if (!process_pages)
 196                        return -ENOMEM;
 197        }
 198
 199        /* Get process information */
 200        rcu_read_lock();
 201        task = find_task_by_vpid(pid);
 202        if (task)
 203                get_task_struct(task);
 204        rcu_read_unlock();
 205        if (!task) {
 206                rc = -ESRCH;
 207                goto free_proc_pages;
 208        }
 209
 210        mm = mm_access(task, PTRACE_MODE_ATTACH_REALCREDS);
 211        if (!mm || IS_ERR(mm)) {
 212                rc = IS_ERR(mm) ? PTR_ERR(mm) : -ESRCH;
 213                /*
 214                 * Explicitly map EACCES to EPERM as EPERM is a more a
 215                 * appropriate error code for process_vw_readv/writev
 216                 */
 217                if (rc == -EACCES)
 218                        rc = -EPERM;
 219                goto put_task_struct;
 220        }
 221
 222        for (i = 0; i < riovcnt && iov_iter_count(iter) && !rc; i++)
 223                rc = process_vm_rw_single_vec(
 224                        (unsigned long)rvec[i].iov_base, rvec[i].iov_len,
 225                        iter, process_pages, mm, task, vm_write);
 226
 227        /* copied = space before - space after */
 228        total_len -= iov_iter_count(iter);
 229
 230        /* If we have managed to copy any data at all then
 231           we return the number of bytes copied. Otherwise
 232           we return the error code */
 233        if (total_len)
 234                rc = total_len;
 235
 236        mmput(mm);
 237
 238put_task_struct:
 239        put_task_struct(task);
 240
 241free_proc_pages:
 242        if (process_pages != pp_stack)
 243                kfree(process_pages);
 244        return rc;
 245}
 246
 247/**
 248 * process_vm_rw - check iovecs before calling core routine
 249 * @pid: PID of process to read/write from/to
 250 * @lvec: iovec array specifying where to copy to/from locally
 251 * @liovcnt: size of lvec array
 252 * @rvec: iovec array specifying where to copy to/from in the other process
 253 * @riovcnt: size of rvec array
 254 * @flags: currently unused
 255 * @vm_write: 0 if reading from other process, 1 if writing to other process
 256 * Returns the number of bytes read/written or error code. May
 257 *  return less bytes than expected if an error occurs during the copying
 258 *  process.
 259 */
 260static ssize_t process_vm_rw(pid_t pid,
 261                             const struct iovec __user *lvec,
 262                             unsigned long liovcnt,
 263                             const struct iovec __user *rvec,
 264                             unsigned long riovcnt,
 265                             unsigned long flags, int vm_write)
 266{
 267        struct iovec iovstack_l[UIO_FASTIOV];
 268        struct iovec iovstack_r[UIO_FASTIOV];
 269        struct iovec *iov_l = iovstack_l;
 270        struct iovec *iov_r = iovstack_r;
 271        struct iov_iter iter;
 272        ssize_t rc;
 273        int dir = vm_write ? WRITE : READ;
 274
 275        if (flags != 0)
 276                return -EINVAL;
 277
 278        /* Check iovecs */
 279        rc = import_iovec(dir, lvec, liovcnt, UIO_FASTIOV, &iov_l, &iter);
 280        if (rc < 0)
 281                return rc;
 282        if (!iov_iter_count(&iter))
 283                goto free_iovecs;
 284
 285        rc = rw_copy_check_uvector(CHECK_IOVEC_ONLY, rvec, riovcnt, UIO_FASTIOV,
 286                                   iovstack_r, &iov_r);
 287        if (rc <= 0)
 288                goto free_iovecs;
 289
 290        rc = process_vm_rw_core(pid, &iter, iov_r, riovcnt, flags, vm_write);
 291
 292free_iovecs:
 293        if (iov_r != iovstack_r)
 294                kfree(iov_r);
 295        kfree(iov_l);
 296
 297        return rc;
 298}
 299
 300SYSCALL_DEFINE6(process_vm_readv, pid_t, pid, const struct iovec __user *, lvec,
 301                unsigned long, liovcnt, const struct iovec __user *, rvec,
 302                unsigned long, riovcnt, unsigned long, flags)
 303{
 304        return process_vm_rw(pid, lvec, liovcnt, rvec, riovcnt, flags, 0);
 305}
 306
 307SYSCALL_DEFINE6(process_vm_writev, pid_t, pid,
 308                const struct iovec __user *, lvec,
 309                unsigned long, liovcnt, const struct iovec __user *, rvec,
 310                unsigned long, riovcnt, unsigned long, flags)
 311{
 312        return process_vm_rw(pid, lvec, liovcnt, rvec, riovcnt, flags, 1);
 313}
 314
 315#ifdef CONFIG_COMPAT
 316
 317static ssize_t
 318compat_process_vm_rw(compat_pid_t pid,
 319                     const struct compat_iovec __user *lvec,
 320                     unsigned long liovcnt,
 321                     const struct compat_iovec __user *rvec,
 322                     unsigned long riovcnt,
 323                     unsigned long flags, int vm_write)
 324{
 325        struct iovec iovstack_l[UIO_FASTIOV];
 326        struct iovec iovstack_r[UIO_FASTIOV];
 327        struct iovec *iov_l = iovstack_l;
 328        struct iovec *iov_r = iovstack_r;
 329        struct iov_iter iter;
 330        ssize_t rc = -EFAULT;
 331        int dir = vm_write ? WRITE : READ;
 332
 333        if (flags != 0)
 334                return -EINVAL;
 335
 336        rc = compat_import_iovec(dir, lvec, liovcnt, UIO_FASTIOV, &iov_l, &iter);
 337        if (rc < 0)
 338                return rc;
 339        if (!iov_iter_count(&iter))
 340                goto free_iovecs;
 341        rc = compat_rw_copy_check_uvector(CHECK_IOVEC_ONLY, rvec, riovcnt,
 342                                          UIO_FASTIOV, iovstack_r,
 343                                          &iov_r);
 344        if (rc <= 0)
 345                goto free_iovecs;
 346
 347        rc = process_vm_rw_core(pid, &iter, iov_r, riovcnt, flags, vm_write);
 348
 349free_iovecs:
 350        if (iov_r != iovstack_r)
 351                kfree(iov_r);
 352        kfree(iov_l);
 353        return rc;
 354}
 355
 356COMPAT_SYSCALL_DEFINE6(process_vm_readv, compat_pid_t, pid,
 357                       const struct compat_iovec __user *, lvec,
 358                       compat_ulong_t, liovcnt,
 359                       const struct compat_iovec __user *, rvec,
 360                       compat_ulong_t, riovcnt,
 361                       compat_ulong_t, flags)
 362{
 363        return compat_process_vm_rw(pid, lvec, liovcnt, rvec,
 364                                    riovcnt, flags, 0);
 365}
 366
 367COMPAT_SYSCALL_DEFINE6(process_vm_writev, compat_pid_t, pid,
 368                       const struct compat_iovec __user *, lvec,
 369                       compat_ulong_t, liovcnt,
 370                       const struct compat_iovec __user *, rvec,
 371                       compat_ulong_t, riovcnt,
 372                       compat_ulong_t, flags)
 373{
 374        return compat_process_vm_rw(pid, lvec, liovcnt, rvec,
 375                                    riovcnt, flags, 1);
 376}
 377
 378#endif
 379