lib/iov_iter.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 #include <crypto/hash.h>
   3 #include <linux/export.h>
   4 #include <linux/bvec.h>
   5 #include <linux/fault-inject-usercopy.h>
   6 #include <linux/uio.h>
   7 #include <linux/pagemap.h>
   8 #include <linux/slab.h>
   9 #include <linux/vmalloc.h>
  10 #include <linux/splice.h>
  11 #include <linux/compat.h>
  12 #include <net/checksum.h>
  13 #include <linux/scatterlist.h>
  14 #include <linux/instrumented.h>
  15
  16 #define PIPE_PARANOIA /* for now */
  17
  18 #define iterate_iovec(i, n, __v, __p, skip, STEP) {     \
  19         size_t left;                                    \
  20         size_t wanted = n;                              \
  21         __p = i->iov;                                   \
  22         __v.iov_len = min(n, __p->iov_len - skip);      \
  23         if (likely(__v.iov_len)) {                      \
  24                 __v.iov_base = __p->iov_base + skip;    \
  25                 left = (STEP);                          \
  26                 __v.iov_len -= left;                    \
  27                 skip += __v.iov_len;                    \
  28                 n -= __v.iov_len;                       \
  29         } else {                                        \
  30                 left = 0;                               \
  31         }                                               \
  32         while (unlikely(!left && n)) {                  \
  33                 __p++;                                  \
  34                 __v.iov_len = min(n, __p->iov_len);     \
  35                 if (unlikely(!__v.iov_len))             \
  36                         continue;                       \
  37                 __v.iov_base = __p->iov_base;           \
  38                 left = (STEP);                          \
  39                 __v.iov_len -= left;                    \
  40                 skip = __v.iov_len;                     \
  41                 n -= __v.iov_len;                       \
  42         }                                               \
  43         n = wanted - n;                                 \
  44 }
  45
  46 #define iterate_kvec(i, n, __v, __p, skip, STEP) {      \
  47         size_t wanted = n;                              \
  48         __p = i->kvec;                                  \
  49         __v.iov_len = min(n, __p->iov_len - skip);      \
  50         if (likely(__v.iov_len)) {                      \
  51                 __v.iov_base = __p->iov_base + skip;    \
  52                 (void)(STEP);                           \
  53                 skip += __v.iov_len;                    \
  54                 n -= __v.iov_len;                       \
  55         }                                               \
  56         while (unlikely(n)) {                           \
  57                 __p++;                                  \
  58                 __v.iov_len = min(n, __p->iov_len);     \
  59                 if (unlikely(!__v.iov_len))             \
  60                         continue;                       \
  61                 __v.iov_base = __p->iov_base;           \
  62                 (void)(STEP);                           \
  63                 skip = __v.iov_len;                     \
  64                 n -= __v.iov_len;                       \
  65         }                                               \
  66         n = wanted;                                     \
  67 }
  68
  69 #define iterate_bvec(i, n, __v, __bi, skip, STEP) {     \
  70         struct bvec_iter __start;                       \
  71         __start.bi_size = n;                            \
  72         __start.bi_bvec_done = skip;                    \
  73         __start.bi_idx = 0;                             \
  74         for_each_bvec(__v, i->bvec, __bi, __start) {    \
  75                 (void)(STEP);                           \
  76         }                                               \
  77 }
  78
  79 #define iterate_all_kinds(i, n, v, I, B, K) {                   \
  80         if (likely(n)) {                                        \
  81                 size_t skip = i->iov_offset;                    \
  82                 if (unlikely(i->type & ITER_BVEC)) {            \
  83                         struct bio_vec v;                       \
  84                         struct bvec_iter __bi;                  \
  85                         iterate_bvec(i, n, v, __bi, skip, (B))  \
  86                 } else if (unlikely(i->type & ITER_KVEC)) {     \
  87                         const struct kvec *kvec;                \
  88                         struct kvec v;                          \
  89                         iterate_kvec(i, n, v, kvec, skip, (K))  \
  90                 } else if (unlikely(i->type & ITER_DISCARD)) {  \
  91                 } else {                                        \
  92                         const struct iovec *iov;                \
  93                         struct iovec v;                         \
  94                         iterate_iovec(i, n, v, iov, skip, (I))  \
  95                 }                                               \
  96         }                                                       \
  97 }
  98
  99 #define iterate_and_advance(i, n, v, I, B, K) {                 \
 100         if (unlikely(i->count < n))                             \
 101                 n = i->count;                                   \
 102         if (i->count) {                                         \
 103                 size_t skip = i->iov_offset;                    \
 104                 if (unlikely(i->type & ITER_BVEC)) {            \
 105                         const struct bio_vec *bvec = i->bvec;   \
 106                         struct bio_vec v;                       \
 107                         struct bvec_iter __bi;                  \
 108                         iterate_bvec(i, n, v, __bi, skip, (B))  \
 109                         i->bvec = __bvec_iter_bvec(i->bvec, __bi);      \
 110                         i->nr_segs -= i->bvec - bvec;           \
 111                         skip = __bi.bi_bvec_done;               \
 112                 } else if (unlikely(i->type & ITER_KVEC)) {     \
 113                         const struct kvec *kvec;                \
 114                         struct kvec v;                          \
 115                         iterate_kvec(i, n, v, kvec, skip, (K))  \
 116                         if (skip == kvec->iov_len) {            \
 117                                 kvec++;                         \
 118                                 skip = 0;                       \
 119                         }                                       \
 120                         i->nr_segs -= kvec - i->kvec;           \
 121                         i->kvec = kvec;                         \
 122                 } else if (unlikely(i->type & ITER_DISCARD)) {  \
 123                         skip += n;                              \
 124                 } else {                                        \
 125                         const struct iovec *iov;                \
 126                         struct iovec v;                         \
 127                         iterate_iovec(i, n, v, iov, skip, (I))  \
 128                         if (skip == iov->iov_len) {             \
 129                                 iov++;                          \
 130                                 skip = 0;                       \
 131                         }                                       \
 132                         i->nr_segs -= iov - i->iov;             \
 133                         i->iov = iov;                           \
 134                 }                                               \
 135                 i->count -= n;                                  \
 136                 i->iov_offset = skip;                           \
 137         }                                                       \
 138 }
 139
 140 static int copyout(void __user *to, const void *from, size_t n)
 141 {
 142         if (should_fail_usercopy())
 143                 return n;
 144         if (access_ok(to, n)) {
 145                 instrument_copy_to_user(to, from, n);
 146                 n = raw_copy_to_user(to, from, n);
 147         }
 148         return n;
 149 }
 150
 151 static int copyin(void *to, const void __user *from, size_t n)
 152 {
 153         if (should_fail_usercopy())
 154                 return n;
 155         if (access_ok(from, n)) {
 156                 instrument_copy_from_user(to, from, n);
 157                 n = raw_copy_from_user(to, from, n);
 158         }
 159         return n;
 160 }
 161
 162 static size_t copy_page_to_iter_iovec(struct page *page, size_t offset, size_t bytes,
 163                          struct iov_iter *i)
 164 {
 165         size_t skip, copy, left, wanted;
 166         const struct iovec *iov;
 167         char __user *buf;
 168         void *kaddr, *from;
 169
 170         if (unlikely(bytes > i->count))
 171                 bytes = i->count;
 172
 173         if (unlikely(!bytes))
 174                 return 0;
 175
 176         might_fault();
 177         wanted = bytes;
 178         iov = i->iov;
 179         skip = i->iov_offset;
 180         buf = iov->iov_base + skip;
 181         copy = min(bytes, iov->iov_len - skip);
 182
 183         if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_pages_writeable(buf, copy)) {
 184                 kaddr = kmap_atomic(page);
 185                 from = kaddr + offset;
 186
 187                 /* first chunk, usually the only one */
 188                 left = copyout(buf, from, copy);
 189                 copy -= left;
 190                 skip += copy;
 191                 from += copy;
 192                 bytes -= copy;
 193
 194                 while (unlikely(!left && bytes)) {
 195                         iov++;
 196                         buf = iov->iov_base;
 197                         copy = min(bytes, iov->iov_len);
 198                         left = copyout(buf, from, copy);
 199                         copy -= left;
 200                         skip = copy;
 201                         from += copy;
 202                         bytes -= copy;
 203                 }
 204                 if (likely(!bytes)) {
 205                         kunmap_atomic(kaddr);
 206                         goto done;
 207                 }
 208                 offset = from - kaddr;
 209                 buf += copy;
 210                 kunmap_atomic(kaddr);
 211                 copy = min(bytes, iov->iov_len - skip);
 212         }
 213         /* Too bad - revert to non-atomic kmap */
 214
 215         kaddr = kmap(page);
 216         from = kaddr + offset;
 217         left = copyout(buf, from, copy);
 218         copy -= left;
 219         skip += copy;
 220         from += copy;
 221         bytes -= copy;
 222         while (unlikely(!left && bytes)) {
 223                 iov++;
 224                 buf = iov->iov_base;
 225                 copy = min(bytes, iov->iov_len);
 226                 left = copyout(buf, from, copy);
 227                 copy -= left;
 228                 skip = copy;
 229                 from += copy;
 230                 bytes -= copy;
 231         }
 232         kunmap(page);
 233
 234 done:
 235         if (skip == iov->iov_len) {
 236                 iov++;
 237                 skip = 0;
 238         }
 239         i->count -= wanted - bytes;
 240         i->nr_segs -= iov - i->iov;
 241         i->iov = iov;
 242         i->iov_offset = skip;
 243         return wanted - bytes;
 244 }
 245
 246 static size_t copy_page_from_iter_iovec(struct page *page, size_t offset, size_t bytes,
 247                          struct iov_iter *i)
 248 {
 249         size_t skip, copy, left, wanted;
 250         const struct iovec *iov;
 251         char __user *buf;
 252         void *kaddr, *to;
 253
 254         if (unlikely(bytes > i->count))
 255                 bytes = i->count;
 256
 257         if (unlikely(!bytes))
 258                 return 0;
 259
 260         might_fault();
 261         wanted = bytes;
 262         iov = i->iov;
 263         skip = i->iov_offset;
 264         buf = iov->iov_base + skip;
 265         copy = min(bytes, iov->iov_len - skip);
 266
 267         if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_pages_readable(buf, copy)) {
 268                 kaddr = kmap_atomic(page);
 269                 to = kaddr + offset;
 270
 271                 /* first chunk, usually the only one */
 272                 left = copyin(to, buf, copy);
 273                 copy -= left;
 274                 skip += copy;
 275                 to += copy;
 276                 bytes -= copy;
 277
 278                 while (unlikely(!left && bytes)) {
 279                         iov++;
 280                         buf = iov->iov_base;
 281                         copy = min(bytes, iov->iov_len);
 282                         left = copyin(to, buf, copy);
 283                         copy -= left;
 284                         skip = copy;
 285                         to += copy;
 286                         bytes -= copy;
 287                 }
 288                 if (likely(!bytes)) {
 289                         kunmap_atomic(kaddr);
 290                         goto done;
 291                 }
 292                 offset = to - kaddr;
 293                 buf += copy;
 294                 kunmap_atomic(kaddr);
 295                 copy = min(bytes, iov->iov_len - skip);
 296         }
 297         /* Too bad - revert to non-atomic kmap */
 298
 299         kaddr = kmap(page);
 300         to = kaddr + offset;
 301         left = copyin(to, buf, copy);
 302         copy -= left;
 303         skip += copy;
 304         to += copy;
 305         bytes -= copy;
 306         while (unlikely(!left && bytes)) {
 307                 iov++;
 308                 buf = iov->iov_base;
 309                 copy = min(bytes, iov->iov_len);
 310                 left = copyin(to, buf, copy);
 311                 copy -= left;
 312                 skip = copy;
 313                 to += copy;
 314                 bytes -= copy;
 315         }
 316         kunmap(page);
 317
 318 done:
 319         if (skip == iov->iov_len) {
 320                 iov++;
 321                 skip = 0;
 322         }
 323         i->count -= wanted - bytes;
 324         i->nr_segs -= iov - i->iov;
 325         i->iov = iov;
 326         i->iov_offset = skip;
 327         return wanted - bytes;
 328 }
 329
 330 #ifdef PIPE_PARANOIA
 331 static bool sanity(const struct iov_iter *i)
 332 {
 333         struct pipe_inode_info *pipe = i->pipe;
 334         unsigned int p_head = pipe->head;
 335         unsigned int p_tail = pipe->tail;
 336         unsigned int p_mask = pipe->ring_size - 1;
 337         unsigned int p_occupancy = pipe_occupancy(p_head, p_tail);
 338         unsigned int i_head = i->head;
 339         unsigned int idx;
 340
 341         if (i->iov_offset) {
 342                 struct pipe_buffer *p;
 343                 if (unlikely(p_occupancy == 0))
 344                         goto Bad;       // pipe must be non-empty
 345                 if (unlikely(i_head != p_head - 1))
 346                         goto Bad;       // must be at the last buffer...
 347
 348                 p = &pipe->bufs[i_head & p_mask];
 349                 if (unlikely(p->offset + p->len != i->iov_offset))
 350                         goto Bad;       // ... at the end of segment
 351         } else {
 352                 if (i_head != p_head)
 353                         goto Bad;       // must be right after the last buffer
 354         }
 355         return true;
 356 Bad:
 357         printk(KERN_ERR "idx = %d, offset = %zd\n", i_head, i->iov_offset);
 358         printk(KERN_ERR "head = %d, tail = %d, buffers = %d\n",
 359                         p_head, p_tail, pipe->ring_size);
 360         for (idx = 0; idx < pipe->ring_size; idx++)
 361                 printk(KERN_ERR "[%p %p %d %d]\n",
 362                         pipe->bufs[idx].ops,
 363                         pipe->bufs[idx].page,
 364                         pipe->bufs[idx].offset,
 365                         pipe->bufs[idx].len);
 366         WARN_ON(1);
 367         return false;
 368 }
 369 #else
 370 #define sanity(i) true
 371 #endif
 372
 373 static size_t copy_page_to_iter_pipe(struct page *page, size_t offset, size_t bytes,
 374                          struct iov_iter *i)
 375 {
 376         struct pipe_inode_info *pipe = i->pipe;
 377         struct pipe_buffer *buf;
 378         unsigned int p_tail = pipe->tail;
 379         unsigned int p_mask = pipe->ring_size - 1;
 380         unsigned int i_head = i->head;
 381         size_t off;
 382
 383         if (unlikely(bytes > i->count))
 384                 bytes = i->count;
 385
 386         if (unlikely(!bytes))
 387                 return 0;
 388
 389         if (!sanity(i))
 390                 return 0;
 391
 392         off = i->iov_offset;
 393         buf = &pipe->bufs[i_head & p_mask];
 394         if (off) {
 395                 if (offset == off && buf->page == page) {
 396                         /* merge with the last one */
 397                         buf->len += bytes;
 398                         i->iov_offset += bytes;
 399                         goto out;
 400                 }
 401                 i_head++;
 402                 buf = &pipe->bufs[i_head & p_mask];
 403         }
 404         if (pipe_full(i_head, p_tail, pipe->max_usage))
 405                 return 0;
 406
 407         buf->ops = &page_cache_pipe_buf_ops;
 408         get_page(page);
 409         buf->page = page;
 410         buf->offset = offset;
 411         buf->len = bytes;
 412
 413         pipe->head = i_head + 1;
 414         i->iov_offset = offset + bytes;
 415         i->head = i_head;
 416 out:
 417         i->count -= bytes;
 418         return bytes;
 419 }
 420
 421 /*
 422  * Fault in one or more iovecs of the given iov_iter, to a maximum length of
 423  * bytes.  For each iovec, fault in each page that constitutes the iovec.
 424  *
 425  * Return 0 on success, or non-zero if the memory could not be accessed (i.e.
 426  * because it is an invalid address).
 427  */
 428 int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
 429 {
 430         size_t skip = i->iov_offset;
 431         const struct iovec *iov;
 432         int err;
 433         struct iovec v;
 434
 435         if (!(i->type & (ITER_BVEC|ITER_KVEC))) {
 436                 iterate_iovec(i, bytes, v, iov, skip, ({
 437                         err = fault_in_pages_readable(v.iov_base, v.iov_len);
 438                         if (unlikely(err))
 439                         return err;
 440                 0;}))
 441         }
 442         return 0;
 443 }
 444 EXPORT_SYMBOL(iov_iter_fault_in_readable);
 445
 446 void iov_iter_init(struct iov_iter *i, unsigned int direction,
 447                         const struct iovec *iov, unsigned long nr_segs,
 448                         size_t count)
 449 {
 450         WARN_ON(direction & ~(READ | WRITE));
 451         direction &= READ | WRITE;
 452
 453         /* It will get better.  Eventually... */
 454         if (uaccess_kernel()) {
 455                 i->type = ITER_KVEC | direction;
 456                 i->kvec = (struct kvec *)iov;
 457         } else {
 458                 i->type = ITER_IOVEC | direction;
 459                 i->iov = iov;
 460         }
 461         i->nr_segs = nr_segs;
 462         i->iov_offset = 0;
 463         i->count = count;
 464 }
 465 EXPORT_SYMBOL(iov_iter_init);
 466
 467 static void memcpy_from_page(char *to, struct page *page, size_t offset, size_t len)
 468 {
 469         char *from = kmap_atomic(page);
 470         memcpy(to, from + offset, len);
 471         kunmap_atomic(from);
 472 }
 473
 474 static void memcpy_to_page(struct page *page, size_t offset, const char *from, size_t len)
 475 {
 476         char *to = kmap_atomic(page);
 477         memcpy(to + offset, from, len);
 478         kunmap_atomic(to);
 479 }
 480
 481 static void memzero_page(struct page *page, size_t offset, size_t len)
 482 {
 483         char *addr = kmap_atomic(page);
 484         memset(addr + offset, 0, len);
 485         kunmap_atomic(addr);
 486 }
 487
 488 static inline bool allocated(struct pipe_buffer *buf)
 489 {
 490         return buf->ops == &default_pipe_buf_ops;
 491 }
 492
 493 static inline void data_start(const struct iov_iter *i,
 494                               unsigned int *iter_headp, size_t *offp)
 495 {
 496         unsigned int p_mask = i->pipe->ring_size - 1;
 497         unsigned int iter_head = i->head;
 498         size_t off = i->iov_offset;
 499
 500         if (off && (!allocated(&i->pipe->bufs[iter_head & p_mask]) ||
 501                     off == PAGE_SIZE)) {
 502                 iter_head++;
 503                 off = 0;
 504         }
 505         *iter_headp = iter_head;
 506         *offp = off;
 507 }
 508
 509 static size_t push_pipe(struct iov_iter *i, size_t size,
 510                         int *iter_headp, size_t *offp)
 511 {
 512         struct pipe_inode_info *pipe = i->pipe;
 513         unsigned int p_tail = pipe->tail;
 514         unsigned int p_mask = pipe->ring_size - 1;
 515         unsigned int iter_head;
 516         size_t off;
 517         ssize_t left;
 518
 519         if (unlikely(size > i->count))
 520                 size = i->count;
 521         if (unlikely(!size))
 522                 return 0;
 523
 524         left = size;
 525         data_start(i, &iter_head, &off);
 526         *iter_headp = iter_head;
 527         *offp = off;
 528         if (off) {
 529                 left -= PAGE_SIZE - off;
 530                 if (left <= 0) {
 531                         pipe->bufs[iter_head & p_mask].len += size;
 532                         return size;
 533                 }
 534                 pipe->bufs[iter_head & p_mask].len = PAGE_SIZE;
 535                 iter_head++;
 536         }
 537         while (!pipe_full(iter_head, p_tail, pipe->max_usage)) {
 538                 struct pipe_buffer *buf = &pipe->bufs[iter_head & p_mask];
 539                 struct page *page = alloc_page(GFP_USER);
 540                 if (!page)
 541                         break;
 542
 543                 buf->ops = &default_pipe_buf_ops;
 544                 buf->page = page;
 545                 buf->offset = 0;
 546                 buf->len = min_t(ssize_t, left, PAGE_SIZE);
 547                 left -= buf->len;
 548                 iter_head++;
 549                 pipe->head = iter_head;
 550
 551                 if (left == 0)
 552                         return size;
 553         }
 554         return size - left;
 555 }
 556
 557 static size_t copy_pipe_to_iter(const void *addr, size_t bytes,
 558                                 struct iov_iter *i)
 559 {
 560         struct pipe_inode_info *pipe = i->pipe;
 561         unsigned int p_mask = pipe->ring_size - 1;
 562         unsigned int i_head;
 563         size_t n, off;
 564
 565         if (!sanity(i))
 566                 return 0;
 567
 568         bytes = n = push_pipe(i, bytes, &i_head, &off);
 569         if (unlikely(!n))
 570                 return 0;
 571         do {
 572                 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
 573                 memcpy_to_page(pipe->bufs[i_head & p_mask].page, off, addr, chunk);
 574                 i->head = i_head;
 575                 i->iov_offset = off + chunk;
 576                 n -= chunk;
 577                 addr += chunk;
 578                 off = 0;
 579                 i_head++;
 580         } while (n);
 581         i->count -= bytes;
 582         return bytes;
 583 }
 584
 585 static __wsum csum_and_memcpy(void *to, const void *from, size_t len,
 586                               __wsum sum, size_t off)
 587 {
 588         __wsum next = csum_partial_copy_nocheck(from, to, len);
 589         return csum_block_add(sum, next, off);
 590 }
 591
 592 static size_t csum_and_copy_to_pipe_iter(const void *addr, size_t bytes,
 593                                          struct csum_state *csstate,
 594                                          struct iov_iter *i)
 595 {
 596         struct pipe_inode_info *pipe = i->pipe;
 597         unsigned int p_mask = pipe->ring_size - 1;
 598         __wsum sum = csstate->csum;
 599         size_t off = csstate->off;
 600         unsigned int i_head;
 601         size_t n, r;
 602
 603         if (!sanity(i))
 604                 return 0;
 605
 606         bytes = n = push_pipe(i, bytes, &i_head, &r);
 607         if (unlikely(!n))
 608                 return 0;
 609         do {
 610                 size_t chunk = min_t(size_t, n, PAGE_SIZE - r);
 611                 char *p = kmap_atomic(pipe->bufs[i_head & p_mask].page);
 612                 sum = csum_and_memcpy(p + r, addr, chunk, sum, off);
 613                 kunmap_atomic(p);
 614                 i->head = i_head;
 615                 i->iov_offset = r + chunk;
 616                 n -= chunk;
 617                 off += chunk;
 618                 addr += chunk;
 619                 r = 0;
 620                 i_head++;
 621         } while (n);
 622         i->count -= bytes;
 623         csstate->csum = sum;
 624         csstate->off = off;
 625         return bytes;
 626 }
 627
 628 size_t _copy_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
 629 {
 630         const char *from = addr;
 631         if (unlikely(iov_iter_is_pipe(i)))
 632                 return copy_pipe_to_iter(addr, bytes, i);
 633         if (iter_is_iovec(i))
 634                 might_fault();
 635         iterate_and_advance(i, bytes, v,
 636                 copyout(v.iov_base, (from += v.iov_len) - v.iov_len, v.iov_len),
 637                 memcpy_to_page(v.bv_page, v.bv_offset,
 638                                (from += v.bv_len) - v.bv_len, v.bv_len),
 639                 memcpy(v.iov_base, (from += v.iov_len) - v.iov_len, v.iov_len)
 640         )
 641
 642         return bytes;
 643 }
 644 EXPORT_SYMBOL(_copy_to_iter);
 645
 646 #ifdef CONFIG_ARCH_HAS_COPY_MC
 647 static int copyout_mc(void __user *to, const void *from, size_t n)
 648 {
 649         if (access_ok(to, n)) {
 650                 instrument_copy_to_user(to, from, n);
 651                 n = copy_mc_to_user((__force void *) to, from, n);
 652         }
 653         return n;
 654 }
 655
 656 static unsigned long copy_mc_to_page(struct page *page, size_t offset,
 657                 const char *from, size_t len)
 658 {
 659         unsigned long ret;
 660         char *to;
 661
 662         to = kmap_atomic(page);
 663         ret = copy_mc_to_kernel(to + offset, from, len);
 664         kunmap_atomic(to);
 665
 666         return ret;
 667 }
 668
 669 static size_t copy_mc_pipe_to_iter(const void *addr, size_t bytes,
 670                                 struct iov_iter *i)
 671 {
 672         struct pipe_inode_info *pipe = i->pipe;
 673         unsigned int p_mask = pipe->ring_size - 1;
 674         unsigned int i_head;
 675         size_t n, off, xfer = 0;
 676
 677         if (!sanity(i))
 678                 return 0;
 679
 680         bytes = n = push_pipe(i, bytes, &i_head, &off);
 681         if (unlikely(!n))
 682                 return 0;
 683         do {
 684                 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
 685                 unsigned long rem;
 686
 687                 rem = copy_mc_to_page(pipe->bufs[i_head & p_mask].page,
 688                                             off, addr, chunk);
 689                 i->head = i_head;
 690                 i->iov_offset = off + chunk - rem;
 691                 xfer += chunk - rem;
 692                 if (rem)
 693                         break;
 694                 n -= chunk;
 695                 addr += chunk;
 696                 off = 0;
 697                 i_head++;
 698         } while (n);
 699         i->count -= xfer;
 700         return xfer;
 701 }
 702
 703 /**
 704  * _copy_mc_to_iter - copy to iter with source memory error exception handling
 705  * @addr: source kernel address
 706  * @bytes: total transfer length
 707  * @iter: destination iterator
 708  *
 709  * The pmem driver deploys this for the dax operation
 710  * (dax_copy_to_iter()) for dax reads (bypass page-cache and the
 711  * block-layer). Upon #MC read(2) aborts and returns EIO or the bytes
 712  * successfully copied.
 713  *
 714  * The main differences between this and typical _copy_to_iter().
 715  *
 716  * * Typical tail/residue handling after a fault retries the copy
 717  *   byte-by-byte until the fault happens again. Re-triggering machine
 718  *   checks is potentially fatal so the implementation uses source
 719  *   alignment and poison alignment assumptions to avoid re-triggering
 720  *   hardware exceptions.
 721  *
 722  * * ITER_KVEC, ITER_PIPE, and ITER_BVEC can return short copies.
 723  *   Compare to copy_to_iter() where only ITER_IOVEC attempts might return
 724  *   a short copy.
 725  */
 726 size_t _copy_mc_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
 727 {
 728         const char *from = addr;
 729         unsigned long rem, curr_addr, s_addr = (unsigned long) addr;
 730
 731         if (unlikely(iov_iter_is_pipe(i)))
 732                 return copy_mc_pipe_to_iter(addr, bytes, i);
 733         if (iter_is_iovec(i))
 734                 might_fault();
 735         iterate_and_advance(i, bytes, v,
 736                 copyout_mc(v.iov_base, (from += v.iov_len) - v.iov_len,
 737                            v.iov_len),
 738                 ({
 739                 rem = copy_mc_to_page(v.bv_page, v.bv_offset,
 740                                       (from += v.bv_len) - v.bv_len, v.bv_len);
 741                 if (rem) {
 742                         curr_addr = (unsigned long) from;
 743                         bytes = curr_addr - s_addr - rem;
 744                         return bytes;
 745                 }
 746                 }),
 747                 ({
 748                 rem = copy_mc_to_kernel(v.iov_base, (from += v.iov_len)
 749                                         - v.iov_len, v.iov_len);
 750                 if (rem) {
 751                         curr_addr = (unsigned long) from;
 752                         bytes = curr_addr - s_addr - rem;
 753                         return bytes;
 754                 }
 755                 })
 756         )
 757
 758         return bytes;
 759 }
 760 EXPORT_SYMBOL_GPL(_copy_mc_to_iter);
 761 #endif /* CONFIG_ARCH_HAS_COPY_MC */
 762
 763 size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i)
 764 {
 765         char *to = addr;
 766         if (unlikely(iov_iter_is_pipe(i))) {
 767                 WARN_ON(1);
 768                 return 0;
 769         }
 770         if (iter_is_iovec(i))
 771                 might_fault();
 772         iterate_and_advance(i, bytes, v,
 773                 copyin((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len),
 774                 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
 775                                  v.bv_offset, v.bv_len),
 776                 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
 777         )
 778
 779         return bytes;
 780 }
 781 EXPORT_SYMBOL(_copy_from_iter);
 782
 783 bool _copy_from_iter_full(void *addr, size_t bytes, struct iov_iter *i)
 784 {
 785         char *to = addr;
 786         if (unlikely(iov_iter_is_pipe(i))) {
 787                 WARN_ON(1);
 788                 return false;
 789         }
 790         if (unlikely(i->count < bytes))
 791                 return false;
 792
 793         if (iter_is_iovec(i))
 794                 might_fault();
 795         iterate_all_kinds(i, bytes, v, ({
 796                 if (copyin((to += v.iov_len) - v.iov_len,
 797                                       v.iov_base, v.iov_len))
 798                         return false;
 799                 0;}),
 800                 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
 801                                  v.bv_offset, v.bv_len),
 802                 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
 803         )
 804
 805         iov_iter_advance(i, bytes);
 806         return true;
 807 }
 808 EXPORT_SYMBOL(_copy_from_iter_full);
 809
 810 size_t _copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i)
 811 {
 812         char *to = addr;
 813         if (unlikely(iov_iter_is_pipe(i))) {
 814                 WARN_ON(1);
 815                 return 0;
 816         }
 817         iterate_and_advance(i, bytes, v,
 818                 __copy_from_user_inatomic_nocache((to += v.iov_len) - v.iov_len,
 819                                          v.iov_base, v.iov_len),
 820                 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
 821                                  v.bv_offset, v.bv_len),
 822                 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
 823         )
 824
 825         return bytes;
 826 }
 827 EXPORT_SYMBOL(_copy_from_iter_nocache);
 828
 829 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
 830 /**
 831  * _copy_from_iter_flushcache - write destination through cpu cache
 832  * @addr: destination kernel address
 833  * @bytes: total transfer length
 834  * @iter: source iterator
 835  *
 836  * The pmem driver arranges for filesystem-dax to use this facility via
 837  * dax_copy_from_iter() for ensuring that writes to persistent memory
 838  * are flushed through the CPU cache. It is differentiated from
 839  * _copy_from_iter_nocache() in that guarantees all data is flushed for
 840  * all iterator types. The _copy_from_iter_nocache() only attempts to
 841  * bypass the cache for the ITER_IOVEC case, and on some archs may use
 842  * instructions that strand dirty-data in the cache.
 843  */
 844 size_t _copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i)
 845 {
 846         char *to = addr;
 847         if (unlikely(iov_iter_is_pipe(i))) {
 848                 WARN_ON(1);
 849                 return 0;
 850         }
 851         iterate_and_advance(i, bytes, v,
 852                 __copy_from_user_flushcache((to += v.iov_len) - v.iov_len,
 853                                          v.iov_base, v.iov_len),
 854                 memcpy_page_flushcache((to += v.bv_len) - v.bv_len, v.bv_page,
 855                                  v.bv_offset, v.bv_len),
 856                 memcpy_flushcache((to += v.iov_len) - v.iov_len, v.iov_base,
 857                         v.iov_len)
 858         )
 859
 860         return bytes;
 861 }
 862 EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache);
 863 #endif
 864
 865 bool _copy_from_iter_full_nocache(void *addr, size_t bytes, struct iov_iter *i)
 866 {
 867         char *to = addr;
 868         if (unlikely(iov_iter_is_pipe(i))) {
 869                 WARN_ON(1);
 870                 return false;
 871         }
 872         if (unlikely(i->count < bytes))
 873                 return false;
 874         iterate_all_kinds(i, bytes, v, ({
 875                 if (__copy_from_user_inatomic_nocache((to += v.iov_len) - v.iov_len,
 876                                              v.iov_base, v.iov_len))
 877                         return false;
 878                 0;}),
 879                 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
 880                                  v.bv_offset, v.bv_len),
 881                 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
 882         )
 883
 884         iov_iter_advance(i, bytes);
 885         return true;
 886 }
 887 EXPORT_SYMBOL(_copy_from_iter_full_nocache);
 888
 889 static inline bool page_copy_sane(struct page *page, size_t offset, size_t n)
 890 {
 891         struct page *head;
 892         size_t v = n + offset;
 893
 894         /*
 895          * The general case needs to access the page order in order
 896          * to compute the page size.
 897          * However, we mostly deal with order-0 pages and thus can
 898          * avoid a possible cache line miss for requests that fit all
 899          * page orders.
 900          */
 901         if (n <= v && v <= PAGE_SIZE)
 902                 return true;
 903
 904         head = compound_head(page);
 905         v += (page - head) << PAGE_SHIFT;
 906
 907         if (likely(n <= v && v <= (page_size(head))))
 908                 return true;
 909         WARN_ON(1);
 910         return false;
 911 }
 912
 913 size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
 914                          struct iov_iter *i)
 915 {
 916         if (unlikely(!page_copy_sane(page, offset, bytes)))
 917                 return 0;
 918         if (i->type & (ITER_BVEC|ITER_KVEC)) {
 919                 void *kaddr = kmap_atomic(page);
 920                 size_t wanted = copy_to_iter(kaddr + offset, bytes, i);
 921                 kunmap_atomic(kaddr);
 922                 return wanted;
 923         } else if (unlikely(iov_iter_is_discard(i)))
 924                 return bytes;
 925         else if (likely(!iov_iter_is_pipe(i)))
 926                 return copy_page_to_iter_iovec(page, offset, bytes, i);
 927         else
 928                 return copy_page_to_iter_pipe(page, offset, bytes, i);
 929 }
 930 EXPORT_SYMBOL(copy_page_to_iter);
 931
 932 size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes,
 933                          struct iov_iter *i)
 934 {
 935         if (unlikely(!page_copy_sane(page, offset, bytes)))
 936                 return 0;
 937         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
 938                 WARN_ON(1);
 939                 return 0;
 940         }
 941         if (i->type & (ITER_BVEC|ITER_KVEC)) {
 942                 void *kaddr = kmap_atomic(page);
 943                 size_t wanted = _copy_from_iter(kaddr + offset, bytes, i);
 944                 kunmap_atomic(kaddr);
 945                 return wanted;
 946         } else
 947                 return copy_page_from_iter_iovec(page, offset, bytes, i);
 948 }
 949 EXPORT_SYMBOL(copy_page_from_iter);
 950
 951 static size_t pipe_zero(size_t bytes, struct iov_iter *i)
 952 {
 953         struct pipe_inode_info *pipe = i->pipe;
 954         unsigned int p_mask = pipe->ring_size - 1;
 955         unsigned int i_head;
 956         size_t n, off;
 957
 958         if (!sanity(i))
 959                 return 0;
 960
 961         bytes = n = push_pipe(i, bytes, &i_head, &off);
 962         if (unlikely(!n))
 963                 return 0;
 964
 965         do {
 966                 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
 967                 memzero_page(pipe->bufs[i_head & p_mask].page, off, chunk);
 968                 i->head = i_head;
 969                 i->iov_offset = off + chunk;
 970                 n -= chunk;
 971                 off = 0;
 972                 i_head++;
 973         } while (n);
 974         i->count -= bytes;
 975         return bytes;
 976 }
 977
 978 size_t iov_iter_zero(size_t bytes, struct iov_iter *i)
 979 {
 980         if (unlikely(iov_iter_is_pipe(i)))
 981                 return pipe_zero(bytes, i);
 982         iterate_and_advance(i, bytes, v,
 983                 clear_user(v.iov_base, v.iov_len),
 984                 memzero_page(v.bv_page, v.bv_offset, v.bv_len),
 985                 memset(v.iov_base, 0, v.iov_len)
 986         )
 987
 988         return bytes;
 989 }
 990 EXPORT_SYMBOL(iov_iter_zero);
 991
 992 size_t iov_iter_copy_from_user_atomic(struct page *page,
 993                 struct iov_iter *i, unsigned long offset, size_t bytes)
 994 {
 995         char *kaddr = kmap_atomic(page), *p = kaddr + offset;
 996         if (unlikely(!page_copy_sane(page, offset, bytes))) {
 997                 kunmap_atomic(kaddr);
 998                 return 0;
 999         }
1000         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1001                 kunmap_atomic(kaddr);
1002                 WARN_ON(1);
1003                 return 0;
1004         }
1005         iterate_all_kinds(i, bytes, v,
1006                 copyin((p += v.iov_len) - v.iov_len, v.iov_base, v.iov_len),
1007                 memcpy_from_page((p += v.bv_len) - v.bv_len, v.bv_page,
1008                                  v.bv_offset, v.bv_len),
1009                 memcpy((p += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
1010         )
1011         kunmap_atomic(kaddr);
1012         return bytes;
1013 }
1014 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
1015
1016 static inline void pipe_truncate(struct iov_iter *i)
1017 {
1018         struct pipe_inode_info *pipe = i->pipe;
1019         unsigned int p_tail = pipe->tail;
1020         unsigned int p_head = pipe->head;
1021         unsigned int p_mask = pipe->ring_size - 1;
1022
1023         if (!pipe_empty(p_head, p_tail)) {
1024                 struct pipe_buffer *buf;
1025                 unsigned int i_head = i->head;
1026                 size_t off = i->iov_offset;
1027
1028                 if (off) {
1029                         buf = &pipe->bufs[i_head & p_mask];
1030                         buf->len = off - buf->offset;
1031                         i_head++;
1032                 }
1033                 while (p_head != i_head) {
1034                         p_head--;
1035                         pipe_buf_release(pipe, &pipe->bufs[p_head & p_mask]);
1036                 }
1037
1038                 pipe->head = p_head;
1039         }
1040 }
1041
1042 static void pipe_advance(struct iov_iter *i, size_t size)
1043 {
1044         struct pipe_inode_info *pipe = i->pipe;
1045         if (unlikely(i->count < size))
1046                 size = i->count;
1047         if (size) {
1048                 struct pipe_buffer *buf;
1049                 unsigned int p_mask = pipe->ring_size - 1;
1050                 unsigned int i_head = i->head;
1051                 size_t off = i->iov_offset, left = size;
1052
1053                 if (off) /* make it relative to the beginning of buffer */
1054                         left += off - pipe->bufs[i_head & p_mask].offset;
1055                 while (1) {
1056                         buf = &pipe->bufs[i_head & p_mask];
1057                         if (left <= buf->len)
1058                                 break;
1059                         left -= buf->len;
1060                         i_head++;
1061                 }
1062                 i->head = i_head;
1063                 i->iov_offset = buf->offset + left;
1064         }
1065         i->count -= size;
1066         /* ... and discard everything past that point */
1067         pipe_truncate(i);
1068 }
1069
1070 static void iov_iter_bvec_advance(struct iov_iter *i, size_t size)
1071 {
1072         struct bvec_iter bi;
1073
1074         bi.bi_size = i->count;
1075         bi.bi_bvec_done = i->iov_offset;
1076         bi.bi_idx = 0;
1077         bvec_iter_advance(i->bvec, &bi, size);
1078
1079         i->bvec += bi.bi_idx;
1080         i->nr_segs -= bi.bi_idx;
1081         i->count = bi.bi_size;
1082         i->iov_offset = bi.bi_bvec_done;
1083 }
1084
1085 void iov_iter_advance(struct iov_iter *i, size_t size)
1086 {
1087         if (unlikely(iov_iter_is_pipe(i))) {
1088                 pipe_advance(i, size);
1089                 return;
1090         }
1091         if (unlikely(iov_iter_is_discard(i))) {
1092                 i->count -= size;
1093                 return;
1094         }
1095         if (iov_iter_is_bvec(i)) {
1096                 iov_iter_bvec_advance(i, size);
1097                 return;
1098         }
1099         iterate_and_advance(i, size, v, 0, 0, 0)
1100 }
1101 EXPORT_SYMBOL(iov_iter_advance);
1102
1103 void iov_iter_revert(struct iov_iter *i, size_t unroll)
1104 {
1105         if (!unroll)
1106                 return;
1107         if (WARN_ON(unroll > MAX_RW_COUNT))
1108                 return;
1109         i->count += unroll;
1110         if (unlikely(iov_iter_is_pipe(i))) {
1111                 struct pipe_inode_info *pipe = i->pipe;
1112                 unsigned int p_mask = pipe->ring_size - 1;
1113                 unsigned int i_head = i->head;
1114                 size_t off = i->iov_offset;
1115                 while (1) {
1116                         struct pipe_buffer *b = &pipe->bufs[i_head & p_mask];
1117                         size_t n = off - b->offset;
1118                         if (unroll < n) {
1119                                 off -= unroll;
1120                                 break;
1121                         }
1122                         unroll -= n;
1123                         if (!unroll && i_head == i->start_head) {
1124                                 off = 0;
1125                                 break;
1126                         }
1127                         i_head--;
1128                         b = &pipe->bufs[i_head & p_mask];
1129                         off = b->offset + b->len;
1130                 }
1131                 i->iov_offset = off;
1132                 i->head = i_head;
1133                 pipe_truncate(i);
1134                 return;
1135         }
1136         if (unlikely(iov_iter_is_discard(i)))
1137                 return;
1138         if (unroll <= i->iov_offset) {
1139                 i->iov_offset -= unroll;
1140                 return;
1141         }
1142         unroll -= i->iov_offset;
1143         if (iov_iter_is_bvec(i)) {
1144                 const struct bio_vec *bvec = i->bvec;
1145                 while (1) {
1146                         size_t n = (--bvec)->bv_len;
1147                         i->nr_segs++;
1148                         if (unroll <= n) {
1149                                 i->bvec = bvec;
1150                                 i->iov_offset = n - unroll;
1151                                 return;
1152                         }
1153                         unroll -= n;
1154                 }
1155         } else { /* same logics for iovec and kvec */
1156                 const struct iovec *iov = i->iov;
1157                 while (1) {
1158                         size_t n = (--iov)->iov_len;
1159                         i->nr_segs++;
1160                         if (unroll <= n) {
1161                                 i->iov = iov;
1162                                 i->iov_offset = n - unroll;
1163                                 return;
1164                         }
1165                         unroll -= n;
1166                 }
1167         }
1168 }
1169 EXPORT_SYMBOL(iov_iter_revert);
1170
1171 /*
1172  * Return the count of just the current iov_iter segment.
1173  */
1174 size_t iov_iter_single_seg_count(const struct iov_iter *i)
1175 {
1176         if (unlikely(iov_iter_is_pipe(i)))
1177                 return i->count;        // it is a silly place, anyway
1178         if (i->nr_segs == 1)
1179                 return i->count;
1180         if (unlikely(iov_iter_is_discard(i)))
1181                 return i->count;
1182         else if (iov_iter_is_bvec(i))
1183                 return min(i->count, i->bvec->bv_len - i->iov_offset);
1184         else
1185                 return min(i->count, i->iov->iov_len - i->iov_offset);
1186 }
1187 EXPORT_SYMBOL(iov_iter_single_seg_count);
1188
1189 void iov_iter_kvec(struct iov_iter *i, unsigned int direction,
1190                         const struct kvec *kvec, unsigned long nr_segs,
1191                         size_t count)
1192 {
1193         WARN_ON(direction & ~(READ | WRITE));
1194         i->type = ITER_KVEC | (direction & (READ | WRITE));
1195         i->kvec = kvec;
1196         i->nr_segs = nr_segs;
1197         i->iov_offset = 0;
1198         i->count = count;
1199 }
1200 EXPORT_SYMBOL(iov_iter_kvec);
1201
1202 void iov_iter_bvec(struct iov_iter *i, unsigned int direction,
1203                         const struct bio_vec *bvec, unsigned long nr_segs,
1204                         size_t count)
1205 {
1206         WARN_ON(direction & ~(READ | WRITE));
1207         i->type = ITER_BVEC | (direction & (READ | WRITE));
1208         i->bvec = bvec;
1209         i->nr_segs = nr_segs;
1210         i->iov_offset = 0;
1211         i->count = count;
1212 }
1213 EXPORT_SYMBOL(iov_iter_bvec);
1214
1215 void iov_iter_pipe(struct iov_iter *i, unsigned int direction,
1216                         struct pipe_inode_info *pipe,
1217                         size_t count)
1218 {
1219         BUG_ON(direction != READ);
1220         WARN_ON(pipe_full(pipe->head, pipe->tail, pipe->ring_size));
1221         i->type = ITER_PIPE | READ;
1222         i->pipe = pipe;
1223         i->head = pipe->head;
1224         i->iov_offset = 0;
1225         i->count = count;
1226         i->start_head = i->head;
1227 }
1228 EXPORT_SYMBOL(iov_iter_pipe);
1229
1230 /**
1231  * iov_iter_discard - Initialise an I/O iterator that discards data
1232  * @i: The iterator to initialise.
1233  * @direction: The direction of the transfer.
1234  * @count: The size of the I/O buffer in bytes.
1235  *
1236  * Set up an I/O iterator that just discards everything that's written to it.
1237  * It's only available as a READ iterator.
1238  */
1239 void iov_iter_discard(struct iov_iter *i, unsigned int direction, size_t count)
1240 {
1241         BUG_ON(direction != READ);
1242         i->type = ITER_DISCARD | READ;
1243         i->count = count;
1244         i->iov_offset = 0;
1245 }
1246 EXPORT_SYMBOL(iov_iter_discard);
1247
1248 unsigned long iov_iter_alignment(const struct iov_iter *i)
1249 {
1250         unsigned long res = 0;
1251         size_t size = i->count;
1252
1253         if (unlikely(iov_iter_is_pipe(i))) {
1254                 unsigned int p_mask = i->pipe->ring_size - 1;
1255
1256                 if (size && i->iov_offset && allocated(&i->pipe->bufs[i->head & p_mask]))
1257                         return size | i->iov_offset;
1258                 return size;
1259         }
1260         iterate_all_kinds(i, size, v,
1261                 (res |= (unsigned long)v.iov_base | v.iov_len, 0),
1262                 res |= v.bv_offset | v.bv_len,
1263                 res |= (unsigned long)v.iov_base | v.iov_len
1264         )
1265         return res;
1266 }
1267 EXPORT_SYMBOL(iov_iter_alignment);
1268
1269 unsigned long iov_iter_gap_alignment(const struct iov_iter *i)
1270 {
1271         unsigned long res = 0;
1272         size_t size = i->count;
1273
1274         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1275                 WARN_ON(1);
1276                 return ~0U;
1277         }
1278
1279         iterate_all_kinds(i, size, v,
1280                 (res |= (!res ? 0 : (unsigned long)v.iov_base) |
1281                         (size != v.iov_len ? size : 0), 0),
1282                 (res |= (!res ? 0 : (unsigned long)v.bv_offset) |
1283                         (size != v.bv_len ? size : 0)),
1284                 (res |= (!res ? 0 : (unsigned long)v.iov_base) |
1285                         (size != v.iov_len ? size : 0))
1286                 );
1287         return res;
1288 }
1289 EXPORT_SYMBOL(iov_iter_gap_alignment);
1290
1291 static inline ssize_t __pipe_get_pages(struct iov_iter *i,
1292                                 size_t maxsize,
1293                                 struct page **pages,
1294                                 int iter_head,
1295                                 size_t *start)
1296 {
1297         struct pipe_inode_info *pipe = i->pipe;
1298         unsigned int p_mask = pipe->ring_size - 1;
1299         ssize_t n = push_pipe(i, maxsize, &iter_head, start);
1300         if (!n)
1301                 return -EFAULT;
1302
1303         maxsize = n;
1304         n += *start;
1305         while (n > 0) {
1306                 get_page(*pages++ = pipe->bufs[iter_head & p_mask].page);
1307                 iter_head++;
1308                 n -= PAGE_SIZE;
1309         }
1310
1311         return maxsize;
1312 }
1313
1314 static ssize_t pipe_get_pages(struct iov_iter *i,
1315                    struct page **pages, size_t maxsize, unsigned maxpages,
1316                    size_t *start)
1317 {
1318         unsigned int iter_head, npages;
1319         size_t capacity;
1320
1321         if (!maxsize)
1322                 return 0;
1323
1324         if (!sanity(i))
1325                 return -EFAULT;
1326
1327         data_start(i, &iter_head, start);
1328         /* Amount of free space: some of this one + all after this one */
1329         npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1330         capacity = min(npages, maxpages) * PAGE_SIZE - *start;
1331
1332         return __pipe_get_pages(i, min(maxsize, capacity), pages, iter_head, start);
1333 }
1334
1335 ssize_t iov_iter_get_pages(struct iov_iter *i,
1336                    struct page **pages, size_t maxsize, unsigned maxpages,
1337                    size_t *start)
1338 {
1339         if (maxsize > i->count)
1340                 maxsize = i->count;
1341
1342         if (unlikely(iov_iter_is_pipe(i)))
1343                 return pipe_get_pages(i, pages, maxsize, maxpages, start);
1344         if (unlikely(iov_iter_is_discard(i)))
1345                 return -EFAULT;
1346
1347         iterate_all_kinds(i, maxsize, v, ({
1348                 unsigned long addr = (unsigned long)v.iov_base;
1349                 size_t len = v.iov_len + (*start = addr & (PAGE_SIZE - 1));
1350                 int n;
1351                 int res;
1352
1353                 if (len > maxpages * PAGE_SIZE)
1354                         len = maxpages * PAGE_SIZE;
1355                 addr &= ~(PAGE_SIZE - 1);
1356                 n = DIV_ROUND_UP(len, PAGE_SIZE);
1357                 res = get_user_pages_fast(addr, n,
1358                                 iov_iter_rw(i) != WRITE ?  FOLL_WRITE : 0,
1359                                 pages);
1360                 if (unlikely(res < 0))
1361                         return res;
1362                 return (res == n ? len : res * PAGE_SIZE) - *start;
1363         0;}),({
1364                 /* can't be more than PAGE_SIZE */
1365                 *start = v.bv_offset;
1366                 get_page(*pages = v.bv_page);
1367                 return v.bv_len;
1368         }),({
1369                 return -EFAULT;
1370         })
1371         )
1372         return 0;
1373 }
1374 EXPORT_SYMBOL(iov_iter_get_pages);
1375
1376 static struct page **get_pages_array(size_t n)
1377 {
1378         return kvmalloc_array(n, sizeof(struct page *), GFP_KERNEL);
1379 }
1380
1381 static ssize_t pipe_get_pages_alloc(struct iov_iter *i,
1382                    struct page ***pages, size_t maxsize,
1383                    size_t *start)
1384 {
1385         struct page **p;
1386         unsigned int iter_head, npages;
1387         ssize_t n;
1388
1389         if (!maxsize)
1390                 return 0;
1391
1392         if (!sanity(i))
1393                 return -EFAULT;
1394
1395         data_start(i, &iter_head, start);
1396         /* Amount of free space: some of this one + all after this one */
1397         npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe);
1398         n = npages * PAGE_SIZE - *start;
1399         if (maxsize > n)
1400                 maxsize = n;
1401         else
1402                 npages = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE);
1403         p = get_pages_array(npages);
1404         if (!p)
1405                 return -ENOMEM;
1406         n = __pipe_get_pages(i, maxsize, p, iter_head, start);
1407         if (n > 0)
1408                 *pages = p;
1409         else
1410                 kvfree(p);
1411         return n;
1412 }
1413
1414 ssize_t iov_iter_get_pages_alloc(struct iov_iter *i,
1415                    struct page ***pages, size_t maxsize,
1416                    size_t *start)
1417 {
1418         struct page **p;
1419
1420         if (maxsize > i->count)
1421                 maxsize = i->count;
1422
1423         if (unlikely(iov_iter_is_pipe(i)))
1424                 return pipe_get_pages_alloc(i, pages, maxsize, start);
1425         if (unlikely(iov_iter_is_discard(i)))
1426                 return -EFAULT;
1427
1428         iterate_all_kinds(i, maxsize, v, ({
1429                 unsigned long addr = (unsigned long)v.iov_base;
1430                 size_t len = v.iov_len + (*start = addr & (PAGE_SIZE - 1));
1431                 int n;
1432                 int res;
1433
1434                 addr &= ~(PAGE_SIZE - 1);
1435                 n = DIV_ROUND_UP(len, PAGE_SIZE);
1436                 p = get_pages_array(n);
1437                 if (!p)
1438                         return -ENOMEM;
1439                 res = get_user_pages_fast(addr, n,
1440                                 iov_iter_rw(i) != WRITE ?  FOLL_WRITE : 0, p);
1441                 if (unlikely(res < 0)) {
1442                         kvfree(p);
1443                         return res;
1444                 }
1445                 *pages = p;
1446                 return (res == n ? len : res * PAGE_SIZE) - *start;
1447         0;}),({
1448                 /* can't be more than PAGE_SIZE */
1449                 *start = v.bv_offset;
1450                 *pages = p = get_pages_array(1);
1451                 if (!p)
1452                         return -ENOMEM;
1453                 get_page(*p = v.bv_page);
1454                 return v.bv_len;
1455         }),({
1456                 return -EFAULT;
1457         })
1458         )
1459         return 0;
1460 }
1461 EXPORT_SYMBOL(iov_iter_get_pages_alloc);
1462
1463 size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum,
1464                                struct iov_iter *i)
1465 {
1466         char *to = addr;
1467         __wsum sum, next;
1468         size_t off = 0;
1469         sum = *csum;
1470         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1471                 WARN_ON(1);
1472                 return 0;
1473         }
1474         iterate_and_advance(i, bytes, v, ({
1475                 next = csum_and_copy_from_user(v.iov_base,
1476                                                (to += v.iov_len) - v.iov_len,
1477                                                v.iov_len);
1478                 if (next) {
1479                         sum = csum_block_add(sum, next, off);
1480                         off += v.iov_len;
1481                 }
1482                 next ? 0 : v.iov_len;
1483         }), ({
1484                 char *p = kmap_atomic(v.bv_page);
1485                 sum = csum_and_memcpy((to += v.bv_len) - v.bv_len,
1486                                       p + v.bv_offset, v.bv_len,
1487                                       sum, off);
1488                 kunmap_atomic(p);
1489                 off += v.bv_len;
1490         }),({
1491                 sum = csum_and_memcpy((to += v.iov_len) - v.iov_len,
1492                                       v.iov_base, v.iov_len,
1493                                       sum, off);
1494                 off += v.iov_len;
1495         })
1496         )
1497         *csum = sum;
1498         return bytes;
1499 }
1500 EXPORT_SYMBOL(csum_and_copy_from_iter);
1501
1502 bool csum_and_copy_from_iter_full(void *addr, size_t bytes, __wsum *csum,
1503                                struct iov_iter *i)
1504 {
1505         char *to = addr;
1506         __wsum sum, next;
1507         size_t off = 0;
1508         sum = *csum;
1509         if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) {
1510                 WARN_ON(1);
1511                 return false;
1512         }
1513         if (unlikely(i->count < bytes))
1514                 return false;
1515         iterate_all_kinds(i, bytes, v, ({
1516                 next = csum_and_copy_from_user(v.iov_base,
1517                                                (to += v.iov_len) - v.iov_len,
1518                                                v.iov_len);
1519                 if (!next)
1520                         return false;
1521                 sum = csum_block_add(sum, next, off);
1522                 off += v.iov_len;
1523                 0;
1524         }), ({
1525                 char *p = kmap_atomic(v.bv_page);
1526                 sum = csum_and_memcpy((to += v.bv_len) - v.bv_len,
1527                                       p + v.bv_offset, v.bv_len,
1528                                       sum, off);
1529                 kunmap_atomic(p);
1530                 off += v.bv_len;
1531         }),({
1532                 sum = csum_and_memcpy((to += v.iov_len) - v.iov_len,
1533                                       v.iov_base, v.iov_len,
1534                                       sum, off);
1535                 off += v.iov_len;
1536         })
1537         )
1538         *csum = sum;
1539         iov_iter_advance(i, bytes);
1540         return true;
1541 }
1542 EXPORT_SYMBOL(csum_and_copy_from_iter_full);
1543
1544 size_t csum_and_copy_to_iter(const void *addr, size_t bytes, void *_csstate,
1545                              struct iov_iter *i)
1546 {
1547         struct csum_state *csstate = _csstate;
1548         const char *from = addr;
1549         __wsum sum, next;
1550         size_t off;
1551
1552         if (unlikely(iov_iter_is_pipe(i)))
1553                 return csum_and_copy_to_pipe_iter(addr, bytes, _csstate, i);
1554
1555         sum = csstate->csum;
1556         off = csstate->off;
1557         if (unlikely(iov_iter_is_discard(i))) {
1558                 WARN_ON(1);     /* for now */
1559                 return 0;
1560         }
1561         iterate_and_advance(i, bytes, v, ({
1562                 next = csum_and_copy_to_user((from += v.iov_len) - v.iov_len,
1563                                              v.iov_base,
1564                                              v.iov_len);
1565                 if (next) {
1566                         sum = csum_block_add(sum, next, off);
1567                         off += v.iov_len;
1568                 }
1569                 next ? 0 : v.iov_len;
1570         }), ({
1571                 char *p = kmap_atomic(v.bv_page);
1572                 sum = csum_and_memcpy(p + v.bv_offset,
1573                                       (from += v.bv_len) - v.bv_len,
1574                                       v.bv_len, sum, off);
1575                 kunmap_atomic(p);
1576                 off += v.bv_len;
1577         }),({
1578                 sum = csum_and_memcpy(v.iov_base,
1579                                      (from += v.iov_len) - v.iov_len,
1580                                      v.iov_len, sum, off);
1581                 off += v.iov_len;
1582         })
1583         )
1584         csstate->csum = sum;
1585         csstate->off = off;
1586         return bytes;
1587 }
1588 EXPORT_SYMBOL(csum_and_copy_to_iter);
1589
1590 size_t hash_and_copy_to_iter(const void *addr, size_t bytes, void *hashp,
1591                 struct iov_iter *i)
1592 {
1593 #ifdef CONFIG_CRYPTO_HASH
1594         struct ahash_request *hash = hashp;
1595         struct scatterlist sg;
1596         size_t copied;
1597
1598         copied = copy_to_iter(addr, bytes, i);
1599         sg_init_one(&sg, addr, copied);
1600         ahash_request_set_crypt(hash, &sg, NULL, copied);
1601         crypto_ahash_update(hash);
1602         return copied;
1603 #else
1604         return 0;
1605 #endif
1606 }
1607 EXPORT_SYMBOL(hash_and_copy_to_iter);
1608
1609 int iov_iter_npages(const struct iov_iter *i, int maxpages)
1610 {
1611         size_t size = i->count;
1612         int npages = 0;
1613
1614         if (!size)
1615                 return 0;
1616         if (unlikely(iov_iter_is_discard(i)))
1617                 return 0;
1618
1619         if (unlikely(iov_iter_is_pipe(i))) {
1620                 struct pipe_inode_info *pipe = i->pipe;
1621                 unsigned int iter_head;
1622                 size_t off;
1623
1624                 if (!sanity(i))
1625                         return 0;
1626
1627                 data_start(i, &iter_head, &off);
1628                 /* some of this one + all after this one */
1629                 npages = pipe_space_for_user(iter_head, pipe->tail, pipe);
1630                 if (npages >= maxpages)
1631                         return maxpages;
1632         } else iterate_all_kinds(i, size, v, ({
1633                 unsigned long p = (unsigned long)v.iov_base;
1634                 npages += DIV_ROUND_UP(p + v.iov_len, PAGE_SIZE)
1635                         - p / PAGE_SIZE;
1636                 if (npages >= maxpages)
1637                         return maxpages;
1638         0;}),({
1639                 npages++;
1640                 if (npages >= maxpages)
1641                         return maxpages;
1642         }),({
1643                 unsigned long p = (unsigned long)v.iov_base;
1644                 npages += DIV_ROUND_UP(p + v.iov_len, PAGE_SIZE)
1645                         - p / PAGE_SIZE;
1646                 if (npages >= maxpages)
1647                         return maxpages;
1648         })
1649         )
1650         return npages;
1651 }
1652 EXPORT_SYMBOL(iov_iter_npages);
1653
1654 const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags)
1655 {
1656         *new = *old;
1657         if (unlikely(iov_iter_is_pipe(new))) {
1658                 WARN_ON(1);
1659                 return NULL;
1660         }
1661         if (unlikely(iov_iter_is_discard(new)))
1662                 return NULL;
1663         if (iov_iter_is_bvec(new))
1664                 return new->bvec = kmemdup(new->bvec,
1665                                     new->nr_segs * sizeof(struct bio_vec),
1666                                     flags);
1667         else
1668                 /* iovec and kvec have identical layout */
1669                 return new->iov = kmemdup(new->iov,
1670                                    new->nr_segs * sizeof(struct iovec),
1671                                    flags);
1672 }
1673 EXPORT_SYMBOL(dup_iter);
1674
1675 static int copy_compat_iovec_from_user(struct iovec *iov,
1676                 const struct iovec __user *uvec, unsigned long nr_segs)
1677 {
1678         const struct compat_iovec __user *uiov =
1679                 (const struct compat_iovec __user *)uvec;
1680         int ret = -EFAULT, i;
1681
1682         if (!user_access_begin(uiov, nr_segs * sizeof(*uiov)))
1683                 return -EFAULT;
1684
1685         for (i = 0; i < nr_segs; i++) {
1686                 compat_uptr_t buf;
1687                 compat_ssize_t len;
1688
1689                 unsafe_get_user(len, &uiov[i].iov_len, uaccess_end);
1690                 unsafe_get_user(buf, &uiov[i].iov_base, uaccess_end);
1691
1692                 /* check for compat_size_t not fitting in compat_ssize_t .. */
1693                 if (len < 0) {
1694                         ret = -EINVAL;
1695                         goto uaccess_end;
1696                 }
1697                 iov[i].iov_base = compat_ptr(buf);
1698                 iov[i].iov_len = len;
1699         }
1700
1701         ret = 0;
1702 uaccess_end:
1703         user_access_end();
1704         return ret;
1705 }
1706
1707 static int copy_iovec_from_user(struct iovec *iov,
1708                 const struct iovec __user *uvec, unsigned long nr_segs)
1709 {
1710         unsigned long seg;
1711
1712         if (copy_from_user(iov, uvec, nr_segs * sizeof(*uvec)))
1713                 return -EFAULT;
1714         for (seg = 0; seg < nr_segs; seg++) {
1715                 if ((ssize_t)iov[seg].iov_len < 0)
1716                         return -EINVAL;
1717         }
1718
1719         return 0;
1720 }
1721
1722 struct iovec *iovec_from_user(const struct iovec __user *uvec,
1723                 unsigned long nr_segs, unsigned long fast_segs,
1724                 struct iovec *fast_iov, bool compat)
1725 {
1726         struct iovec *iov = fast_iov;
1727         int ret;
1728
1729         /*
1730          * SuS says "The readv() function *may* fail if the iovcnt argument was
1731          * less than or equal to 0, or greater than {IOV_MAX}.  Linux has
1732          * traditionally returned zero for zero segments, so...
1733          */
1734         if (nr_segs == 0)
1735                 return iov;
1736         if (nr_segs > UIO_MAXIOV)
1737                 return ERR_PTR(-EINVAL);
1738         if (nr_segs > fast_segs) {
1739                 iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
1740                 if (!iov)
1741                         return ERR_PTR(-ENOMEM);
1742         }
1743
1744         if (compat)
1745                 ret = copy_compat_iovec_from_user(iov, uvec, nr_segs);
1746         else
1747                 ret = copy_iovec_from_user(iov, uvec, nr_segs);
1748         if (ret) {
1749                 if (iov != fast_iov)
1750                         kfree(iov);
1751                 return ERR_PTR(ret);
1752         }
1753
1754         return iov;
1755 }
1756
1757 ssize_t __import_iovec(int type, const struct iovec __user *uvec,
1758                  unsigned nr_segs, unsigned fast_segs, struct iovec **iovp,
1759                  struct iov_iter *i, bool compat)
1760 {
1761         ssize_t total_len = 0;
1762         unsigned long seg;
1763         struct iovec *iov;
1764
1765         iov = iovec_from_user(uvec, nr_segs, fast_segs, *iovp, compat);
1766         if (IS_ERR(iov)) {
1767                 *iovp = NULL;
1768                 return PTR_ERR(iov);
1769         }
1770
1771         /*
1772          * According to the Single Unix Specification we should return EINVAL if
1773          * an element length is < 0 when cast to ssize_t or if the total length
1774          * would overflow the ssize_t return value of the system call.
1775          *
1776          * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
1777          * overflow case.
1778          */
1779         for (seg = 0; seg < nr_segs; seg++) {
1780                 ssize_t len = (ssize_t)iov[seg].iov_len;
1781
1782                 if (!access_ok(iov[seg].iov_base, len)) {
1783                         if (iov != *iovp)
1784                                 kfree(iov);
1785                         *iovp = NULL;
1786                         return -EFAULT;
1787                 }
1788
1789                 if (len > MAX_RW_COUNT - total_len) {
1790                         len = MAX_RW_COUNT - total_len;
1791                         iov[seg].iov_len = len;
1792                 }
1793                 total_len += len;
1794         }
1795
1796         iov_iter_init(i, type, iov, nr_segs, total_len);
1797         if (iov == *iovp)
1798                 *iovp = NULL;
1799         else
1800                 *iovp = iov;
1801         return total_len;
1802 }
1803
1804 /**
1805  * import_iovec() - Copy an array of &struct iovec from userspace
1806  *     into the kernel, check that it is valid, and initialize a new
1807  *     &struct iov_iter iterator to access it.
1808  *
1809  * @type: One of %READ or %WRITE.
1810  * @uvec: Pointer to the userspace array.
1811  * @nr_segs: Number of elements in userspace array.
1812  * @fast_segs: Number of elements in @iov.
1813  * @iovp: (input and output parameter) Pointer to pointer to (usually small
1814  *     on-stack) kernel array.
1815  * @i: Pointer to iterator that will be initialized on success.
1816  *
1817  * If the array pointed to by *@iov is large enough to hold all @nr_segs,
1818  * then this function places %NULL in *@iov on return. Otherwise, a new
1819  * array will be allocated and the result placed in *@iov. This means that
1820  * the caller may call kfree() on *@iov regardless of whether the small
1821  * on-stack array was used or not (and regardless of whether this function
1822  * returns an error or not).
1823  *
1824  * Return: Negative error code on error, bytes imported on success
1825  */
1826 ssize_t import_iovec(int type, const struct iovec __user *uvec,
1827                  unsigned nr_segs, unsigned fast_segs,
1828                  struct iovec **iovp, struct iov_iter *i)
1829 {
1830         return __import_iovec(type, uvec, nr_segs, fast_segs, iovp, i,
1831                               in_compat_syscall());
1832 }
1833 EXPORT_SYMBOL(import_iovec);
1834
1835 int import_single_range(int rw, void __user *buf, size_t len,
1836                  struct iovec *iov, struct iov_iter *i)
1837 {
1838         if (len > MAX_RW_COUNT)
1839                 len = MAX_RW_COUNT;
1840         if (unlikely(!access_ok(buf, len)))
1841                 return -EFAULT;
1842
1843         iov->iov_base = buf;
1844         iov->iov_len = len;
1845         iov_iter_init(i, rw, iov, 1, len);
1846         return 0;
1847 }
1848 EXPORT_SYMBOL(import_single_range);
1849
1850 int iov_iter_for_each_range(struct iov_iter *i, size_t bytes,
1851                             int (*f)(struct kvec *vec, void *context),
1852                             void *context)
1853 {
1854         struct kvec w;
1855         int err = -EINVAL;
1856         if (!bytes)
1857                 return 0;
1858
1859         iterate_all_kinds(i, bytes, v, -EINVAL, ({
1860                 w.iov_base = kmap(v.bv_page) + v.bv_offset;
1861                 w.iov_len = v.bv_len;
1862                 err = f(&w, context);
1863                 kunmap(v.bv_page);
1864                 err;}), ({
1865                 w = v;
1866                 err = f(&w, context);})
1867         )
1868         return err;
1869 }
1870 EXPORT_SYMBOL(iov_iter_for_each_range);