mm/process_vm_access.c

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