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                 /* Get the pages we're interested in */
 102                 down_read(&mm->mmap_sem);
 103                 pages = get_user_pages(task, mm, pa, pages,
 104                                       vm_write, 0, process_pages, NULL);
 105                 up_read(&mm->mmap_sem);
 106
 107                 if (pages <= 0)
 108                         return -EFAULT;
 109
 110                 bytes = pages * PAGE_SIZE - start_offset;
 111                 if (bytes > len)
 112                         bytes = len;
 113
 114                 rc = process_vm_rw_pages(process_pages,
 115                                          start_offset, bytes, iter,
 116                                          vm_write);
 117                 len -= bytes;
 118                 start_offset = 0;
 119                 nr_pages -= pages;
 120                 pa += pages * PAGE_SIZE;
 121                 while (pages)
 122                         put_page(process_pages[--pages]);
 123         }
 124
 125         return rc;
 126 }
 127
 128 /* Maximum number of entries for process pages array
 129    which lives on stack */
 130 #define PVM_MAX_PP_ARRAY_COUNT 16
 131
 132 /**
 133  * process_vm_rw_core - core of reading/writing pages from task specified
 134  * @pid: PID of process to read/write from/to
 135  * @iter: where to copy to/from locally
 136  * @rvec: iovec array specifying where to copy to/from in the other process
 137  * @riovcnt: size of rvec array
 138  * @flags: currently unused
 139  * @vm_write: 0 if reading from other process, 1 if writing to other process
 140  * Returns the number of bytes read/written or error code. May
 141  *  return less bytes than expected if an error occurs during the copying
 142  *  process.
 143  */
 144 static ssize_t process_vm_rw_core(pid_t pid, struct iov_iter *iter,
 145                                   const struct iovec *rvec,
 146                                   unsigned long riovcnt,
 147                                   unsigned long flags, int vm_write)
 148 {
 149         struct task_struct *task;
 150         struct page *pp_stack[PVM_MAX_PP_ARRAY_COUNT];
 151         struct page **process_pages = pp_stack;
 152         struct mm_struct *mm;
 153         unsigned long i;
 154         ssize_t rc = 0;
 155         unsigned long nr_pages = 0;
 156         unsigned long nr_pages_iov;
 157         ssize_t iov_len;
 158         size_t total_len = iov_iter_count(iter);
 159
 160         /*
 161          * Work out how many pages of struct pages we're going to need
 162          * when eventually calling get_user_pages
 163          */
 164         for (i = 0; i < riovcnt; i++) {
 165                 iov_len = rvec[i].iov_len;
 166                 if (iov_len > 0) {
 167                         nr_pages_iov = ((unsigned long)rvec[i].iov_base
 168                                         + iov_len)
 169                                 / PAGE_SIZE - (unsigned long)rvec[i].iov_base
 170                                 / PAGE_SIZE + 1;
 171                         nr_pages = max(nr_pages, nr_pages_iov);
 172                 }
 173         }
 174
 175         if (nr_pages == 0)
 176                 return 0;
 177
 178         if (nr_pages > PVM_MAX_PP_ARRAY_COUNT) {
 179                 /* For reliability don't try to kmalloc more than
 180                    2 pages worth */
 181                 process_pages = kmalloc(min_t(size_t, PVM_MAX_KMALLOC_PAGES,
 182                                               sizeof(struct pages *)*nr_pages),
 183                                         GFP_KERNEL);
 184
 185                 if (!process_pages)
 186                         return -ENOMEM;
 187         }
 188
 189         /* Get process information */
 190         rcu_read_lock();
 191         task = find_task_by_vpid(pid);
 192         if (task)
 193                 get_task_struct(task);
 194         rcu_read_unlock();
 195         if (!task) {
 196                 rc = -ESRCH;
 197                 goto free_proc_pages;
 198         }
 199
 200         mm = mm_access(task, PTRACE_MODE_ATTACH_REALCREDS);
 201         if (!mm || IS_ERR(mm)) {
 202                 rc = IS_ERR(mm) ? PTR_ERR(mm) : -ESRCH;
 203                 /*
 204                  * Explicitly map EACCES to EPERM as EPERM is a more a
 205                  * appropriate error code for process_vw_readv/writev
 206                  */
 207                 if (rc == -EACCES)
 208                         rc = -EPERM;
 209                 goto put_task_struct;
 210         }
 211
 212         for (i = 0; i < riovcnt && iov_iter_count(iter) && !rc; i++)
 213                 rc = process_vm_rw_single_vec(
 214                         (unsigned long)rvec[i].iov_base, rvec[i].iov_len,
 215                         iter, process_pages, mm, task, vm_write);
 216
 217         /* copied = space before - space after */
 218         total_len -= iov_iter_count(iter);
 219
 220         /* If we have managed to copy any data at all then
 221            we return the number of bytes copied. Otherwise
 222            we return the error code */
 223         if (total_len)
 224                 rc = total_len;
 225
 226         mmput(mm);
 227
 228 put_task_struct:
 229         put_task_struct(task);
 230
 231 free_proc_pages:
 232         if (process_pages != pp_stack)
 233                 kfree(process_pages);
 234         return rc;
 235 }
 236
 237 /**
 238  * process_vm_rw - check iovecs before calling core routine
 239  * @pid: PID of process to read/write from/to
 240  * @lvec: iovec array specifying where to copy to/from locally
 241  * @liovcnt: size of lvec array
 242  * @rvec: iovec array specifying where to copy to/from in the other process
 243  * @riovcnt: size of rvec array
 244  * @flags: currently unused
 245  * @vm_write: 0 if reading from other process, 1 if writing to other process
 246  * Returns the number of bytes read/written or error code. May
 247  *  return less bytes than expected if an error occurs during the copying
 248  *  process.
 249  */
 250 static ssize_t process_vm_rw(pid_t pid,
 251                              const struct iovec __user *lvec,
 252                              unsigned long liovcnt,
 253                              const struct iovec __user *rvec,
 254                              unsigned long riovcnt,
 255                              unsigned long flags, int vm_write)
 256 {
 257         struct iovec iovstack_l[UIO_FASTIOV];
 258         struct iovec iovstack_r[UIO_FASTIOV];
 259         struct iovec *iov_l = iovstack_l;
 260         struct iovec *iov_r = iovstack_r;
 261         struct iov_iter iter;
 262         ssize_t rc;
 263
 264         if (flags != 0)
 265                 return -EINVAL;
 266
 267         /* Check iovecs */
 268         if (vm_write)
 269                 rc = rw_copy_check_uvector(WRITE, lvec, liovcnt, UIO_FASTIOV,
 270                                            iovstack_l, &iov_l);
 271         else
 272                 rc = rw_copy_check_uvector(READ, lvec, liovcnt, UIO_FASTIOV,
 273                                            iovstack_l, &iov_l);
 274         if (rc <= 0)
 275                 goto free_iovecs;
 276
 277         iov_iter_init(&iter, vm_write ? WRITE : READ, iov_l, liovcnt, rc);
 278
 279         rc = rw_copy_check_uvector(CHECK_IOVEC_ONLY, rvec, riovcnt, UIO_FASTIOV,
 280                                    iovstack_r, &iov_r);
 281         if (rc <= 0)
 282                 goto free_iovecs;
 283
 284         rc = process_vm_rw_core(pid, &iter, iov_r, riovcnt, flags, vm_write);
 285
 286 free_iovecs:
 287         if (iov_r != iovstack_r)
 288                 kfree(iov_r);
 289         if (iov_l != iovstack_l)
 290                 kfree(iov_l);
 291
 292         return rc;
 293 }
 294
 295 SYSCALL_DEFINE6(process_vm_readv, pid_t, pid, const struct iovec __user *, lvec,
 296                 unsigned long, liovcnt, const struct iovec __user *, rvec,
 297                 unsigned long, riovcnt, unsigned long, flags)
 298 {
 299         return process_vm_rw(pid, lvec, liovcnt, rvec, riovcnt, flags, 0);
 300 }
 301
 302 SYSCALL_DEFINE6(process_vm_writev, pid_t, pid,
 303                 const struct iovec __user *, lvec,
 304                 unsigned long, liovcnt, const struct iovec __user *, rvec,
 305                 unsigned long, riovcnt, unsigned long, flags)
 306 {
 307         return process_vm_rw(pid, lvec, liovcnt, rvec, riovcnt, flags, 1);
 308 }
 309
 310 #ifdef CONFIG_COMPAT
 311
 312 static ssize_t
 313 compat_process_vm_rw(compat_pid_t pid,
 314                      const struct compat_iovec __user *lvec,
 315                      unsigned long liovcnt,
 316                      const struct compat_iovec __user *rvec,
 317                      unsigned long riovcnt,
 318                      unsigned long flags, int vm_write)
 319 {
 320         struct iovec iovstack_l[UIO_FASTIOV];
 321         struct iovec iovstack_r[UIO_FASTIOV];
 322         struct iovec *iov_l = iovstack_l;
 323         struct iovec *iov_r = iovstack_r;
 324         struct iov_iter iter;
 325         ssize_t rc = -EFAULT;
 326
 327         if (flags != 0)
 328                 return -EINVAL;
 329
 330         if (vm_write)
 331                 rc = compat_rw_copy_check_uvector(WRITE, lvec, liovcnt,
 332                                                   UIO_FASTIOV, iovstack_l,
 333                                                   &iov_l);
 334         else
 335                 rc = compat_rw_copy_check_uvector(READ, lvec, liovcnt,
 336                                                   UIO_FASTIOV, iovstack_l,
 337                                                   &iov_l);
 338         if (rc <= 0)
 339                 goto free_iovecs;
 340         iov_iter_init(&iter, vm_write ? WRITE : READ, iov_l, liovcnt, rc);
 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
 349 free_iovecs:
 350         if (iov_r != iovstack_r)
 351                 kfree(iov_r);
 352         if (iov_l != iovstack_l)
 353                 kfree(iov_l);
 354         return rc;
 355 }
 356
 357 COMPAT_SYSCALL_DEFINE6(process_vm_readv, compat_pid_t, pid,
 358                        const struct compat_iovec __user *, lvec,
 359                        compat_ulong_t, liovcnt,
 360                        const struct compat_iovec __user *, rvec,
 361                        compat_ulong_t, riovcnt,
 362                        compat_ulong_t, flags)
 363 {
 364         return compat_process_vm_rw(pid, lvec, liovcnt, rvec,
 365                                     riovcnt, flags, 0);
 366 }
 367
 368 COMPAT_SYSCALL_DEFINE6(process_vm_writev, compat_pid_t, pid,
 369                        const struct compat_iovec __user *, lvec,
 370                        compat_ulong_t, liovcnt,
 371                        const struct compat_iovec __user *, rvec,
 372                        compat_ulong_t, riovcnt,
 373                        compat_ulong_t, flags)
 374 {
 375         return compat_process_vm_rw(pid, lvec, liovcnt, rvec,
 376                                     riovcnt, flags, 1);
 377 }
 378
 379 #endif