lib/iov_iter.c

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