]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - lib/iov_iter.c
projects / mirror_ubuntu-artful-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge tag 'openrisc-for-linus' of git://github.com/openrisc/linux
[mirror_ubuntu-artful-kernel.git] / lib / iov_iter.c
1 #include <linux/export.h>
2 #include <linux/bvec.h>
3 #include <linux/uio.h>
4 #include <linux/pagemap.h>
5 #include <linux/slab.h>
6 #include <linux/vmalloc.h>
7 #include <linux/splice.h>
8 #include <net/checksum.h>
9
10 #define PIPE_PARANOIA /* for now */
11
12 #define iterate_iovec(i, n, __v, __p, skip, STEP) { \
13 size_t left; \
14 size_t wanted = n; \
15 __p = i->iov; \
16 __v.iov_len = min(n, __p->iov_len - skip); \
17 if (likely(__v.iov_len)) { \
18 __v.iov_base = __p->iov_base + skip; \
19 left = (STEP); \
20 __v.iov_len -= left; \
21 skip += __v.iov_len; \
22 n -= __v.iov_len; \
23 } else { \
24 left = 0; \
25 } \
26 while (unlikely(!left && n)) { \
27 __p++; \
28 __v.iov_len = min(n, __p->iov_len); \
29 if (unlikely(!__v.iov_len)) \
30 continue; \
31 __v.iov_base = __p->iov_base; \
32 left = (STEP); \
33 __v.iov_len -= left; \
34 skip = __v.iov_len; \
35 n -= __v.iov_len; \
36 } \
37 n = wanted - n; \
38 }
39
40 #define iterate_kvec(i, n, __v, __p, skip, STEP) { \
41 size_t wanted = n; \
42 __p = i->kvec; \
43 __v.iov_len = min(n, __p->iov_len - skip); \
44 if (likely(__v.iov_len)) { \
45 __v.iov_base = __p->iov_base + skip; \
46 (void)(STEP); \
47 skip += __v.iov_len; \
48 n -= __v.iov_len; \
49 } \
50 while (unlikely(n)) { \
51 __p++; \
52 __v.iov_len = min(n, __p->iov_len); \
53 if (unlikely(!__v.iov_len)) \
54 continue; \
55 __v.iov_base = __p->iov_base; \
56 (void)(STEP); \
57 skip = __v.iov_len; \
58 n -= __v.iov_len; \
59 } \
60 n = wanted; \
61 }
62
63 #define iterate_bvec(i, n, __v, __bi, skip, STEP) { \
64 struct bvec_iter __start; \
65 __start.bi_size = n; \
66 __start.bi_bvec_done = skip; \
67 __start.bi_idx = 0; \
68 for_each_bvec(__v, i->bvec, __bi, __start) { \
69 if (!__v.bv_len) \
70 continue; \
71 (void)(STEP); \
72 } \
73 }
74
75 #define iterate_all_kinds(i, n, v, I, B, K) { \
76 if (likely(n)) { \
77 size_t skip = i->iov_offset; \
78 if (unlikely(i->type & ITER_BVEC)) { \
79 struct bio_vec v; \
80 struct bvec_iter __bi; \
81 iterate_bvec(i, n, v, __bi, skip, (B)) \
82 } else if (unlikely(i->type & ITER_KVEC)) { \
83 const struct kvec *kvec; \
84 struct kvec v; \
85 iterate_kvec(i, n, v, kvec, skip, (K)) \
86 } else { \
87 const struct iovec *iov; \
88 struct iovec v; \
89 iterate_iovec(i, n, v, iov, skip, (I)) \
90 } \
91 } \
92 }
93
94 #define iterate_and_advance(i, n, v, I, B, K) { \
95 if (unlikely(i->count < n)) \
96 n = i->count; \
97 if (i->count) { \
98 size_t skip = i->iov_offset; \
99 if (unlikely(i->type & ITER_BVEC)) { \
100 const struct bio_vec *bvec = i->bvec; \
101 struct bio_vec v; \
102 struct bvec_iter __bi; \
103 iterate_bvec(i, n, v, __bi, skip, (B)) \
104 i->bvec = __bvec_iter_bvec(i->bvec, __bi); \
105 i->nr_segs -= i->bvec - bvec; \
106 skip = __bi.bi_bvec_done; \
107 } else if (unlikely(i->type & ITER_KVEC)) { \
108 const struct kvec *kvec; \
109 struct kvec v; \
110 iterate_kvec(i, n, v, kvec, skip, (K)) \
111 if (skip == kvec->iov_len) { \
112 kvec++; \
113 skip = 0; \
114 } \
115 i->nr_segs -= kvec - i->kvec; \
116 i->kvec = kvec; \
117 } else { \
118 const struct iovec *iov; \
119 struct iovec v; \
120 iterate_iovec(i, n, v, iov, skip, (I)) \
121 if (skip == iov->iov_len) { \
122 iov++; \
123 skip = 0; \
124 } \
125 i->nr_segs -= iov - i->iov; \
126 i->iov = iov; \
127 } \
128 i->count -= n; \
129 i->iov_offset = skip; \
130 } \
131 }
132
133 static size_t copy_page_to_iter_iovec(struct page *page, size_t offset, size_t bytes,
134 struct iov_iter *i)
135 {
136 size_t skip, copy, left, wanted;
137 const struct iovec *iov;
138 char __user *buf;
139 void *kaddr, *from;
140
141 if (unlikely(bytes > i->count))
142 bytes = i->count;
143
144 if (unlikely(!bytes))
145 return 0;
146
147 wanted = bytes;
148 iov = i->iov;
149 skip = i->iov_offset;
150 buf = iov->iov_base + skip;
151 copy = min(bytes, iov->iov_len - skip);
152
153 if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_pages_writeable(buf, copy)) {
154 kaddr = kmap_atomic(page);
155 from = kaddr + offset;
156
157 /* first chunk, usually the only one */
158 left = __copy_to_user_inatomic(buf, from, copy);
159 copy -= left;
160 skip += copy;
161 from += copy;
162 bytes -= copy;
163
164 while (unlikely(!left && bytes)) {
165 iov++;
166 buf = iov->iov_base;
167 copy = min(bytes, iov->iov_len);
168 left = __copy_to_user_inatomic(buf, from, copy);
169 copy -= left;
170 skip = copy;
171 from += copy;
172 bytes -= copy;
173 }
174 if (likely(!bytes)) {
175 kunmap_atomic(kaddr);
176 goto done;
177 }
178 offset = from - kaddr;
179 buf += copy;
180 kunmap_atomic(kaddr);
181 copy = min(bytes, iov->iov_len - skip);
182 }
183 /* Too bad - revert to non-atomic kmap */
184
185 kaddr = kmap(page);
186 from = kaddr + offset;
187 left = __copy_to_user(buf, from, copy);
188 copy -= left;
189 skip += copy;
190 from += copy;
191 bytes -= copy;
192 while (unlikely(!left && bytes)) {
193 iov++;
194 buf = iov->iov_base;
195 copy = min(bytes, iov->iov_len);
196 left = __copy_to_user(buf, from, copy);
197 copy -= left;
198 skip = copy;
199 from += copy;
200 bytes -= copy;
201 }
202 kunmap(page);
203
204 done:
205 if (skip == iov->iov_len) {
206 iov++;
207 skip = 0;
208 }
209 i->count -= wanted - bytes;
210 i->nr_segs -= iov - i->iov;
211 i->iov = iov;
212 i->iov_offset = skip;
213 return wanted - bytes;
214 }
215
216 static size_t copy_page_from_iter_iovec(struct page *page, size_t offset, size_t bytes,
217 struct iov_iter *i)
218 {
219 size_t skip, copy, left, wanted;
220 const struct iovec *iov;
221 char __user *buf;
222 void *kaddr, *to;
223
224 if (unlikely(bytes > i->count))
225 bytes = i->count;
226
227 if (unlikely(!bytes))
228 return 0;
229
230 wanted = bytes;
231 iov = i->iov;
232 skip = i->iov_offset;
233 buf = iov->iov_base + skip;
234 copy = min(bytes, iov->iov_len - skip);
235
236 if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_pages_readable(buf, copy)) {
237 kaddr = kmap_atomic(page);
238 to = kaddr + offset;
239
240 /* first chunk, usually the only one */
241 left = __copy_from_user_inatomic(to, buf, copy);
242 copy -= left;
243 skip += copy;
244 to += copy;
245 bytes -= copy;
246
247 while (unlikely(!left && bytes)) {
248 iov++;
249 buf = iov->iov_base;
250 copy = min(bytes, iov->iov_len);
251 left = __copy_from_user_inatomic(to, buf, copy);
252 copy -= left;
253 skip = copy;
254 to += copy;
255 bytes -= copy;
256 }
257 if (likely(!bytes)) {
258 kunmap_atomic(kaddr);
259 goto done;
260 }
261 offset = to - kaddr;
262 buf += copy;
263 kunmap_atomic(kaddr);
264 copy = min(bytes, iov->iov_len - skip);
265 }
266 /* Too bad - revert to non-atomic kmap */
267
268 kaddr = kmap(page);
269 to = kaddr + offset;
270 left = __copy_from_user(to, buf, copy);
271 copy -= left;
272 skip += copy;
273 to += copy;
274 bytes -= copy;
275 while (unlikely(!left && bytes)) {
276 iov++;
277 buf = iov->iov_base;
278 copy = min(bytes, iov->iov_len);
279 left = __copy_from_user(to, buf, copy);
280 copy -= left;
281 skip = copy;
282 to += copy;
283 bytes -= copy;
284 }
285 kunmap(page);
286
287 done:
288 if (skip == iov->iov_len) {
289 iov++;
290 skip = 0;
291 }
292 i->count -= wanted - bytes;
293 i->nr_segs -= iov - i->iov;
294 i->iov = iov;
295 i->iov_offset = skip;
296 return wanted - bytes;
297 }
298
299 #ifdef PIPE_PARANOIA
300 static bool sanity(const struct iov_iter *i)
301 {
302 struct pipe_inode_info *pipe = i->pipe;
303 int idx = i->idx;
304 int next = pipe->curbuf + pipe->nrbufs;
305 if (i->iov_offset) {
306 struct pipe_buffer *p;
307 if (unlikely(!pipe->nrbufs))
308 goto Bad; // pipe must be non-empty
309 if (unlikely(idx != ((next - 1) & (pipe->buffers - 1))))
310 goto Bad; // must be at the last buffer...
311
312 p = &pipe->bufs[idx];
313 if (unlikely(p->offset + p->len != i->iov_offset))
314 goto Bad; // ... at the end of segment
315 } else {
316 if (idx != (next & (pipe->buffers - 1)))
317 goto Bad; // must be right after the last buffer
318 }
319 return true;
320 Bad:
321 printk(KERN_ERR "idx = %d, offset = %zd\n", i->idx, i->iov_offset);
322 printk(KERN_ERR "curbuf = %d, nrbufs = %d, buffers = %d\n",
323 pipe->curbuf, pipe->nrbufs, pipe->buffers);
324 for (idx = 0; idx < pipe->buffers; idx++)
325 printk(KERN_ERR "[%p %p %d %d]\n",
326 pipe->bufs[idx].ops,
327 pipe->bufs[idx].page,
328 pipe->bufs[idx].offset,
329 pipe->bufs[idx].len);
330 WARN_ON(1);
331 return false;
332 }
333 #else
334 #define sanity(i) true
335 #endif
336
337 static inline int next_idx(int idx, struct pipe_inode_info *pipe)
338 {
339 return (idx + 1) & (pipe->buffers - 1);
340 }
341
342 static size_t copy_page_to_iter_pipe(struct page *page, size_t offset, size_t bytes,
343 struct iov_iter *i)
344 {
345 struct pipe_inode_info *pipe = i->pipe;
346 struct pipe_buffer *buf;
347 size_t off;
348 int idx;
349
350 if (unlikely(bytes > i->count))
351 bytes = i->count;
352
353 if (unlikely(!bytes))
354 return 0;
355
356 if (!sanity(i))
357 return 0;
358
359 off = i->iov_offset;
360 idx = i->idx;
361 buf = &pipe->bufs[idx];
362 if (off) {
363 if (offset == off && buf->page == page) {
364 /* merge with the last one */
365 buf->len += bytes;
366 i->iov_offset += bytes;
367 goto out;
368 }
369 idx = next_idx(idx, pipe);
370 buf = &pipe->bufs[idx];
371 }
372 if (idx == pipe->curbuf && pipe->nrbufs)
373 return 0;
374 pipe->nrbufs++;
375 buf->ops = &page_cache_pipe_buf_ops;
376 get_page(buf->page = page);
377 buf->offset = offset;
378 buf->len = bytes;
379 i->iov_offset = offset + bytes;
380 i->idx = idx;
381 out:
382 i->count -= bytes;
383 return bytes;
384 }
385
386 /*
387 * Fault in one or more iovecs of the given iov_iter, to a maximum length of
388 * bytes. For each iovec, fault in each page that constitutes the iovec.
389 *
390 * Return 0 on success, or non-zero if the memory could not be accessed (i.e.
391 * because it is an invalid address).
392 */
393 int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
394 {
395 size_t skip = i->iov_offset;
396 const struct iovec *iov;
397 int err;
398 struct iovec v;
399
400 if (!(i->type & (ITER_BVEC|ITER_KVEC))) {
401 iterate_iovec(i, bytes, v, iov, skip, ({
402 err = fault_in_pages_readable(v.iov_base, v.iov_len);
403 if (unlikely(err))
404 return err;
405 0;}))
406 }
407 return 0;
408 }
409 EXPORT_SYMBOL(iov_iter_fault_in_readable);
410
411 void iov_iter_init(struct iov_iter *i, int direction,
412 const struct iovec *iov, unsigned long nr_segs,
413 size_t count)
414 {
415 /* It will get better. Eventually... */
416 if (uaccess_kernel()) {
417 direction |= ITER_KVEC;
418 i->type = direction;
419 i->kvec = (struct kvec *)iov;
420 } else {
421 i->type = direction;
422 i->iov = iov;
423 }
424 i->nr_segs = nr_segs;
425 i->iov_offset = 0;
426 i->count = count;
427 }
428 EXPORT_SYMBOL(iov_iter_init);
429
430 static void memcpy_from_page(char *to, struct page *page, size_t offset, size_t len)
431 {
432 char *from = kmap_atomic(page);
433 memcpy(to, from + offset, len);
434 kunmap_atomic(from);
435 }
436
437 static void memcpy_to_page(struct page *page, size_t offset, const char *from, size_t len)
438 {
439 char *to = kmap_atomic(page);
440 memcpy(to + offset, from, len);
441 kunmap_atomic(to);
442 }
443
444 static void memzero_page(struct page *page, size_t offset, size_t len)
445 {
446 char *addr = kmap_atomic(page);
447 memset(addr + offset, 0, len);
448 kunmap_atomic(addr);
449 }
450
451 static inline bool allocated(struct pipe_buffer *buf)
452 {
453 return buf->ops == &default_pipe_buf_ops;
454 }
455
456 static inline void data_start(const struct iov_iter *i, int *idxp, size_t *offp)
457 {
458 size_t off = i->iov_offset;
459 int idx = i->idx;
460 if (off && (!allocated(&i->pipe->bufs[idx]) || off == PAGE_SIZE)) {
461 idx = next_idx(idx, i->pipe);
462 off = 0;
463 }
464 *idxp = idx;
465 *offp = off;
466 }
467
468 static size_t push_pipe(struct iov_iter *i, size_t size,
469 int *idxp, size_t *offp)
470 {
471 struct pipe_inode_info *pipe = i->pipe;
472 size_t off;
473 int idx;
474 ssize_t left;
475
476 if (unlikely(size > i->count))
477 size = i->count;
478 if (unlikely(!size))
479 return 0;
480
481 left = size;
482 data_start(i, &idx, &off);
483 *idxp = idx;
484 *offp = off;
485 if (off) {
486 left -= PAGE_SIZE - off;
487 if (left <= 0) {
488 pipe->bufs[idx].len += size;
489 return size;
490 }
491 pipe->bufs[idx].len = PAGE_SIZE;
492 idx = next_idx(idx, pipe);
493 }
494 while (idx != pipe->curbuf || !pipe->nrbufs) {
495 struct page *page = alloc_page(GFP_USER);
496 if (!page)
497 break;
498 pipe->nrbufs++;
499 pipe->bufs[idx].ops = &default_pipe_buf_ops;
500 pipe->bufs[idx].page = page;
501 pipe->bufs[idx].offset = 0;
502 if (left <= PAGE_SIZE) {
503 pipe->bufs[idx].len = left;
504 return size;
505 }
506 pipe->bufs[idx].len = PAGE_SIZE;
507 left -= PAGE_SIZE;
508 idx = next_idx(idx, pipe);
509 }
510 return size - left;
511 }
512
513 static size_t copy_pipe_to_iter(const void *addr, size_t bytes,
514 struct iov_iter *i)
515 {
516 struct pipe_inode_info *pipe = i->pipe;
517 size_t n, off;
518 int idx;
519
520 if (!sanity(i))
521 return 0;
522
523 bytes = n = push_pipe(i, bytes, &idx, &off);
524 if (unlikely(!n))
525 return 0;
526 for ( ; n; idx = next_idx(idx, pipe), off = 0) {
527 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
528 memcpy_to_page(pipe->bufs[idx].page, off, addr, chunk);
529 i->idx = idx;
530 i->iov_offset = off + chunk;
531 n -= chunk;
532 addr += chunk;
533 }
534 i->count -= bytes;
535 return bytes;
536 }
537
538 size_t copy_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
539 {
540 const char *from = addr;
541 if (unlikely(i->type & ITER_PIPE))
542 return copy_pipe_to_iter(addr, bytes, i);
543 iterate_and_advance(i, bytes, v,
544 __copy_to_user(v.iov_base, (from += v.iov_len) - v.iov_len,
545 v.iov_len),
546 memcpy_to_page(v.bv_page, v.bv_offset,
547 (from += v.bv_len) - v.bv_len, v.bv_len),
548 memcpy(v.iov_base, (from += v.iov_len) - v.iov_len, v.iov_len)
549 )
550
551 return bytes;
552 }
553 EXPORT_SYMBOL(copy_to_iter);
554
555 size_t copy_from_iter(void *addr, size_t bytes, struct iov_iter *i)
556 {
557 char *to = addr;
558 if (unlikely(i->type & ITER_PIPE)) {
559 WARN_ON(1);
560 return 0;
561 }
562 iterate_and_advance(i, bytes, v,
563 __copy_from_user((to += v.iov_len) - v.iov_len, v.iov_base,
564 v.iov_len),
565 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
566 v.bv_offset, v.bv_len),
567 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
568 )
569
570 return bytes;
571 }
572 EXPORT_SYMBOL(copy_from_iter);
573
574 bool copy_from_iter_full(void *addr, size_t bytes, struct iov_iter *i)
575 {
576 char *to = addr;
577 if (unlikely(i->type & ITER_PIPE)) {
578 WARN_ON(1);
579 return false;
580 }
581 if (unlikely(i->count < bytes))
582 return false;
583
584 iterate_all_kinds(i, bytes, v, ({
585 if (__copy_from_user((to += v.iov_len) - v.iov_len,
586 v.iov_base, v.iov_len))
587 return false;
588 0;}),
589 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
590 v.bv_offset, v.bv_len),
591 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
592 )
593
594 iov_iter_advance(i, bytes);
595 return true;
596 }
597 EXPORT_SYMBOL(copy_from_iter_full);
598
599 size_t copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i)
600 {
601 char *to = addr;
602 if (unlikely(i->type & ITER_PIPE)) {
603 WARN_ON(1);
604 return 0;
605 }
606 iterate_and_advance(i, bytes, v,
607 __copy_from_user_inatomic_nocache((to += v.iov_len) - v.iov_len,
608 v.iov_base, v.iov_len),
609 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
610 v.bv_offset, v.bv_len),
611 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
612 )
613
614 return bytes;
615 }
616 EXPORT_SYMBOL(copy_from_iter_nocache);
617
618 #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
619 size_t copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i)
620 {
621 char *to = addr;
622 if (unlikely(i->type & ITER_PIPE)) {
623 WARN_ON(1);
624 return 0;
625 }
626 iterate_and_advance(i, bytes, v,
627 __copy_from_user_flushcache((to += v.iov_len) - v.iov_len,
628 v.iov_base, v.iov_len),
629 memcpy_page_flushcache((to += v.bv_len) - v.bv_len, v.bv_page,
630 v.bv_offset, v.bv_len),
631 memcpy_flushcache((to += v.iov_len) - v.iov_len, v.iov_base,
632 v.iov_len)
633 )
634
635 return bytes;
636 }
637 EXPORT_SYMBOL_GPL(copy_from_iter_flushcache);
638 #endif
639
640 bool copy_from_iter_full_nocache(void *addr, size_t bytes, struct iov_iter *i)
641 {
642 char *to = addr;
643 if (unlikely(i->type & ITER_PIPE)) {
644 WARN_ON(1);
645 return false;
646 }
647 if (unlikely(i->count < bytes))
648 return false;
649 iterate_all_kinds(i, bytes, v, ({
650 if (__copy_from_user_inatomic_nocache((to += v.iov_len) - v.iov_len,
651 v.iov_base, v.iov_len))
652 return false;
653 0;}),
654 memcpy_from_page((to += v.bv_len) - v.bv_len, v.bv_page,
655 v.bv_offset, v.bv_len),
656 memcpy((to += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
657 )
658
659 iov_iter_advance(i, bytes);
660 return true;
661 }
662 EXPORT_SYMBOL(copy_from_iter_full_nocache);
663
664 size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes,
665 struct iov_iter *i)
666 {
667 if (i->type & (ITER_BVEC|ITER_KVEC)) {
668 void *kaddr = kmap_atomic(page);
669 size_t wanted = copy_to_iter(kaddr + offset, bytes, i);
670 kunmap_atomic(kaddr);
671 return wanted;
672 } else if (likely(!(i->type & ITER_PIPE)))
673 return copy_page_to_iter_iovec(page, offset, bytes, i);
674 else
675 return copy_page_to_iter_pipe(page, offset, bytes, i);
676 }
677 EXPORT_SYMBOL(copy_page_to_iter);
678
679 size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes,
680 struct iov_iter *i)
681 {
682 if (unlikely(i->type & ITER_PIPE)) {
683 WARN_ON(1);
684 return 0;
685 }
686 if (i->type & (ITER_BVEC|ITER_KVEC)) {
687 void *kaddr = kmap_atomic(page);
688 size_t wanted = copy_from_iter(kaddr + offset, bytes, i);
689 kunmap_atomic(kaddr);
690 return wanted;
691 } else
692 return copy_page_from_iter_iovec(page, offset, bytes, i);
693 }
694 EXPORT_SYMBOL(copy_page_from_iter);
695
696 static size_t pipe_zero(size_t bytes, struct iov_iter *i)
697 {
698 struct pipe_inode_info *pipe = i->pipe;
699 size_t n, off;
700 int idx;
701
702 if (!sanity(i))
703 return 0;
704
705 bytes = n = push_pipe(i, bytes, &idx, &off);
706 if (unlikely(!n))
707 return 0;
708
709 for ( ; n; idx = next_idx(idx, pipe), off = 0) {
710 size_t chunk = min_t(size_t, n, PAGE_SIZE - off);
711 memzero_page(pipe->bufs[idx].page, off, chunk);
712 i->idx = idx;
713 i->iov_offset = off + chunk;
714 n -= chunk;
715 }
716 i->count -= bytes;
717 return bytes;
718 }
719
720 size_t iov_iter_zero(size_t bytes, struct iov_iter *i)
721 {
722 if (unlikely(i->type & ITER_PIPE))
723 return pipe_zero(bytes, i);
724 iterate_and_advance(i, bytes, v,
725 __clear_user(v.iov_base, v.iov_len),
726 memzero_page(v.bv_page, v.bv_offset, v.bv_len),
727 memset(v.iov_base, 0, v.iov_len)
728 )
729
730 return bytes;
731 }
732 EXPORT_SYMBOL(iov_iter_zero);
733
734 size_t iov_iter_copy_from_user_atomic(struct page *page,
735 struct iov_iter *i, unsigned long offset, size_t bytes)
736 {
737 char *kaddr = kmap_atomic(page), *p = kaddr + offset;
738 if (unlikely(i->type & ITER_PIPE)) {
739 kunmap_atomic(kaddr);
740 WARN_ON(1);
741 return 0;
742 }
743 iterate_all_kinds(i, bytes, v,
744 __copy_from_user_inatomic((p += v.iov_len) - v.iov_len,
745 v.iov_base, v.iov_len),
746 memcpy_from_page((p += v.bv_len) - v.bv_len, v.bv_page,
747 v.bv_offset, v.bv_len),
748 memcpy((p += v.iov_len) - v.iov_len, v.iov_base, v.iov_len)
749 )
750 kunmap_atomic(kaddr);
751 return bytes;
752 }
753 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
754
755 static inline void pipe_truncate(struct iov_iter *i)
756 {
757 struct pipe_inode_info *pipe = i->pipe;
758 if (pipe->nrbufs) {
759 size_t off = i->iov_offset;
760 int idx = i->idx;
761 int nrbufs = (idx - pipe->curbuf) & (pipe->buffers - 1);
762 if (off) {
763 pipe->bufs[idx].len = off - pipe->bufs[idx].offset;
764 idx = next_idx(idx, pipe);
765 nrbufs++;
766 }
767 while (pipe->nrbufs > nrbufs) {
768 pipe_buf_release(pipe, &pipe->bufs[idx]);
769 idx = next_idx(idx, pipe);
770 pipe->nrbufs--;
771 }
772 }
773 }
774
775 static void pipe_advance(struct iov_iter *i, size_t size)
776 {
777 struct pipe_inode_info *pipe = i->pipe;
778 if (unlikely(i->count < size))
779 size = i->count;
780 if (size) {
781 struct pipe_buffer *buf;
782 size_t off = i->iov_offset, left = size;
783 int idx = i->idx;
784 if (off) /* make it relative to the beginning of buffer */
785 left += off - pipe->bufs[idx].offset;
786 while (1) {
787 buf = &pipe->bufs[idx];
788 if (left <= buf->len)
789 break;
790 left -= buf->len;
791 idx = next_idx(idx, pipe);
792 }
793 i->idx = idx;
794 i->iov_offset = buf->offset + left;
795 }
796 i->count -= size;
797 /* ... and discard everything past that point */
798 pipe_truncate(i);
799 }
800
801 void iov_iter_advance(struct iov_iter *i, size_t size)
802 {
803 if (unlikely(i->type & ITER_PIPE)) {
804 pipe_advance(i, size);
805 return;
806 }
807 iterate_and_advance(i, size, v, 0, 0, 0)
808 }
809 EXPORT_SYMBOL(iov_iter_advance);
810
811 void iov_iter_revert(struct iov_iter *i, size_t unroll)
812 {
813 if (!unroll)
814 return;
815 if (WARN_ON(unroll > MAX_RW_COUNT))
816 return;
817 i->count += unroll;
818 if (unlikely(i->type & ITER_PIPE)) {
819 struct pipe_inode_info *pipe = i->pipe;
820 int idx = i->idx;
821 size_t off = i->iov_offset;
822 while (1) {
823 size_t n = off - pipe->bufs[idx].offset;
824 if (unroll < n) {
825 off -= unroll;
826 break;
827 }
828 unroll -= n;
829 if (!unroll && idx == i->start_idx) {
830 off = 0;
831 break;
832 }
833 if (!idx--)
834 idx = pipe->buffers - 1;
835 off = pipe->bufs[idx].offset + pipe->bufs[idx].len;
836 }
837 i->iov_offset = off;
838 i->idx = idx;
839 pipe_truncate(i);
840 return;
841 }
842 if (unroll <= i->iov_offset) {
843 i->iov_offset -= unroll;
844 return;
845 }
846 unroll -= i->iov_offset;
847 if (i->type & ITER_BVEC) {
848 const struct bio_vec *bvec = i->bvec;
849 while (1) {
850 size_t n = (--bvec)->bv_len;
851 i->nr_segs++;
852 if (unroll <= n) {
853 i->bvec = bvec;
854 i->iov_offset = n - unroll;
855 return;
856 }
857 unroll -= n;
858 }
859 } else { /* same logics for iovec and kvec */
860 const struct iovec *iov = i->iov;
861 while (1) {
862 size_t n = (--iov)->iov_len;
863 i->nr_segs++;
864 if (unroll <= n) {
865 i->iov = iov;
866 i->iov_offset = n - unroll;
867 return;
868 }
869 unroll -= n;
870 }
871 }
872 }
873 EXPORT_SYMBOL(iov_iter_revert);
874
875 /*
876 * Return the count of just the current iov_iter segment.
877 */
878 size_t iov_iter_single_seg_count(const struct iov_iter *i)
879 {
880 if (unlikely(i->type & ITER_PIPE))
881 return i->count; // it is a silly place, anyway
882 if (i->nr_segs == 1)
883 return i->count;
884 else if (i->type & ITER_BVEC)
885 return min(i->count, i->bvec->bv_len - i->iov_offset);
886 else
887 return min(i->count, i->iov->iov_len - i->iov_offset);
888 }
889 EXPORT_SYMBOL(iov_iter_single_seg_count);
890
891 void iov_iter_kvec(struct iov_iter *i, int direction,
892 const struct kvec *kvec, unsigned long nr_segs,
893 size_t count)
894 {
895 BUG_ON(!(direction & ITER_KVEC));
896 i->type = direction;
897 i->kvec = kvec;
898 i->nr_segs = nr_segs;
899 i->iov_offset = 0;
900 i->count = count;
901 }
902 EXPORT_SYMBOL(iov_iter_kvec);
903
904 void iov_iter_bvec(struct iov_iter *i, int direction,
905 const struct bio_vec *bvec, unsigned long nr_segs,
906 size_t count)
907 {
908 BUG_ON(!(direction & ITER_BVEC));
909 i->type = direction;
910 i->bvec = bvec;
911 i->nr_segs = nr_segs;
912 i->iov_offset = 0;
913 i->count = count;
914 }
915 EXPORT_SYMBOL(iov_iter_bvec);
916
917 void iov_iter_pipe(struct iov_iter *i, int direction,
918 struct pipe_inode_info *pipe,
919 size_t count)
920 {
921 BUG_ON(direction != ITER_PIPE);
922 WARN_ON(pipe->nrbufs == pipe->buffers);
923 i->type = direction;
924 i->pipe = pipe;
925 i->idx = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
926 i->iov_offset = 0;
927 i->count = count;
928 i->start_idx = i->idx;
929 }
930 EXPORT_SYMBOL(iov_iter_pipe);
931
932 unsigned long iov_iter_alignment(const struct iov_iter *i)
933 {
934 unsigned long res = 0;
935 size_t size = i->count;
936
937 if (unlikely(i->type & ITER_PIPE)) {
938 if (size && i->iov_offset && allocated(&i->pipe->bufs[i->idx]))
939 return size | i->iov_offset;
940 return size;
941 }
942 iterate_all_kinds(i, size, v,
943 (res |= (unsigned long)v.iov_base | v.iov_len, 0),
944 res |= v.bv_offset | v.bv_len,
945 res |= (unsigned long)v.iov_base | v.iov_len
946 )
947 return res;
948 }
949 EXPORT_SYMBOL(iov_iter_alignment);
950
951 unsigned long iov_iter_gap_alignment(const struct iov_iter *i)
952 {
953 unsigned long res = 0;
954 size_t size = i->count;
955
956 if (unlikely(i->type & ITER_PIPE)) {
957 WARN_ON(1);
958 return ~0U;
959 }
960
961 iterate_all_kinds(i, size, v,
962 (res |= (!res ? 0 : (unsigned long)v.iov_base) |
963 (size != v.iov_len ? size : 0), 0),
964 (res |= (!res ? 0 : (unsigned long)v.bv_offset) |
965 (size != v.bv_len ? size : 0)),
966 (res |= (!res ? 0 : (unsigned long)v.iov_base) |
967 (size != v.iov_len ? size : 0))
968 );
969 return res;
970 }
971 EXPORT_SYMBOL(iov_iter_gap_alignment);
972
973 static inline size_t __pipe_get_pages(struct iov_iter *i,
974 size_t maxsize,
975 struct page **pages,
976 int idx,
977 size_t *start)
978 {
979 struct pipe_inode_info *pipe = i->pipe;
980 ssize_t n = push_pipe(i, maxsize, &idx, start);
981 if (!n)
982 return -EFAULT;
983
984 maxsize = n;
985 n += *start;
986 while (n > 0) {
987 get_page(*pages++ = pipe->bufs[idx].page);
988 idx = next_idx(idx, pipe);
989 n -= PAGE_SIZE;
990 }
991
992 return maxsize;
993 }
994
995 static ssize_t pipe_get_pages(struct iov_iter *i,
996 struct page **pages, size_t maxsize, unsigned maxpages,
997 size_t *start)
998 {
999 unsigned npages;
1000 size_t capacity;
1001 int idx;
1002
1003 if (!maxsize)
1004 return 0;
1005
1006 if (!sanity(i))
1007 return -EFAULT;
1008
1009 data_start(i, &idx, start);
1010 /* some of this one + all after this one */
1011 npages = ((i->pipe->curbuf - idx - 1) & (i->pipe->buffers - 1)) + 1;
1012 capacity = min(npages,maxpages) * PAGE_SIZE - *start;
1013
1014 return __pipe_get_pages(i, min(maxsize, capacity), pages, idx, start);
1015 }
1016
1017 ssize_t iov_iter_get_pages(struct iov_iter *i,
1018 struct page **pages, size_t maxsize, unsigned maxpages,
1019 size_t *start)
1020 {
1021 if (maxsize > i->count)
1022 maxsize = i->count;
1023
1024 if (unlikely(i->type & ITER_PIPE))
1025 return pipe_get_pages(i, pages, maxsize, maxpages, start);
1026 iterate_all_kinds(i, maxsize, v, ({
1027 unsigned long addr = (unsigned long)v.iov_base;
1028 size_t len = v.iov_len + (*start = addr & (PAGE_SIZE - 1));
1029 int n;
1030 int res;
1031
1032 if (len > maxpages * PAGE_SIZE)
1033 len = maxpages * PAGE_SIZE;
1034 addr &= ~(PAGE_SIZE - 1);
1035 n = DIV_ROUND_UP(len, PAGE_SIZE);
1036 res = get_user_pages_fast(addr, n, (i->type & WRITE) != WRITE, pages);
1037 if (unlikely(res < 0))
1038 return res;
1039 return (res == n ? len : res * PAGE_SIZE) - *start;
1040 0;}),({
1041 /* can't be more than PAGE_SIZE */
1042 *start = v.bv_offset;
1043 get_page(*pages = v.bv_page);
1044 return v.bv_len;
1045 }),({
1046 return -EFAULT;
1047 })
1048 )
1049 return 0;
1050 }
1051 EXPORT_SYMBOL(iov_iter_get_pages);
1052
1053 static struct page **get_pages_array(size_t n)
1054 {
1055 return kvmalloc_array(n, sizeof(struct page *), GFP_KERNEL);
1056 }
1057
1058 static ssize_t pipe_get_pages_alloc(struct iov_iter *i,
1059 struct page ***pages, size_t maxsize,
1060 size_t *start)
1061 {
1062 struct page **p;
1063 size_t n;
1064 int idx;
1065 int npages;
1066
1067 if (!maxsize)
1068 return 0;
1069
1070 if (!sanity(i))
1071 return -EFAULT;
1072
1073 data_start(i, &idx, start);
1074 /* some of this one + all after this one */
1075 npages = ((i->pipe->curbuf - idx - 1) & (i->pipe->buffers - 1)) + 1;
1076 n = npages * PAGE_SIZE - *start;
1077 if (maxsize > n)
1078 maxsize = n;
1079 else
1080 npages = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE);
1081 p = get_pages_array(npages);
1082 if (!p)
1083 return -ENOMEM;
1084 n = __pipe_get_pages(i, maxsize, p, idx, start);
1085 if (n > 0)
1086 *pages = p;
1087 else
1088 kvfree(p);
1089 return n;
1090 }
1091
1092 ssize_t iov_iter_get_pages_alloc(struct iov_iter *i,
1093 struct page ***pages, size_t maxsize,
1094 size_t *start)
1095 {
1096 struct page **p;
1097
1098 if (maxsize > i->count)
1099 maxsize = i->count;
1100
1101 if (unlikely(i->type & ITER_PIPE))
1102 return pipe_get_pages_alloc(i, pages, maxsize, start);
1103 iterate_all_kinds(i, maxsize, v, ({
1104 unsigned long addr = (unsigned long)v.iov_base;
1105 size_t len = v.iov_len + (*start = addr & (PAGE_SIZE - 1));
1106 int n;
1107 int res;
1108
1109 addr &= ~(PAGE_SIZE - 1);
1110 n = DIV_ROUND_UP(len, PAGE_SIZE);
1111 p = get_pages_array(n);
1112 if (!p)
1113 return -ENOMEM;
1114 res = get_user_pages_fast(addr, n, (i->type & WRITE) != WRITE, p);
1115 if (unlikely(res < 0)) {
1116 kvfree(p);
1117 return res;
1118 }
1119 *pages = p;
1120 return (res == n ? len : res * PAGE_SIZE) - *start;
1121 0;}),({
1122 /* can't be more than PAGE_SIZE */
1123 *start = v.bv_offset;
1124 *pages = p = get_pages_array(1);
1125 if (!p)
1126 return -ENOMEM;
1127 get_page(*p = v.bv_page);
1128 return v.bv_len;
1129 }),({
1130 return -EFAULT;
1131 })
1132 )
1133 return 0;
1134 }
1135 EXPORT_SYMBOL(iov_iter_get_pages_alloc);
1136
1137 size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum,
1138 struct iov_iter *i)
1139 {
1140 char *to = addr;
1141 __wsum sum, next;
1142 size_t off = 0;
1143 sum = *csum;
1144 if (unlikely(i->type & ITER_PIPE)) {
1145 WARN_ON(1);
1146 return 0;
1147 }
1148 iterate_and_advance(i, bytes, v, ({
1149 int err = 0;
1150 next = csum_and_copy_from_user(v.iov_base,
1151 (to += v.iov_len) - v.iov_len,
1152 v.iov_len, 0, &err);
1153 if (!err) {
1154 sum = csum_block_add(sum, next, off);
1155 off += v.iov_len;
1156 }
1157 err ? v.iov_len : 0;
1158 }), ({
1159 char *p = kmap_atomic(v.bv_page);
1160 next = csum_partial_copy_nocheck(p + v.bv_offset,
1161 (to += v.bv_len) - v.bv_len,
1162 v.bv_len, 0);
1163 kunmap_atomic(p);
1164 sum = csum_block_add(sum, next, off);
1165 off += v.bv_len;
1166 }),({
1167 next = csum_partial_copy_nocheck(v.iov_base,
1168 (to += v.iov_len) - v.iov_len,
1169 v.iov_len, 0);
1170 sum = csum_block_add(sum, next, off);
1171 off += v.iov_len;
1172 })
1173 )
1174 *csum = sum;
1175 return bytes;
1176 }
1177 EXPORT_SYMBOL(csum_and_copy_from_iter);
1178
1179 bool csum_and_copy_from_iter_full(void *addr, size_t bytes, __wsum *csum,
1180 struct iov_iter *i)
1181 {
1182 char *to = addr;
1183 __wsum sum, next;
1184 size_t off = 0;
1185 sum = *csum;
1186 if (unlikely(i->type & ITER_PIPE)) {
1187 WARN_ON(1);
1188 return false;
1189 }
1190 if (unlikely(i->count < bytes))
1191 return false;
1192 iterate_all_kinds(i, bytes, v, ({
1193 int err = 0;
1194 next = csum_and_copy_from_user(v.iov_base,
1195 (to += v.iov_len) - v.iov_len,
1196 v.iov_len, 0, &err);
1197 if (err)
1198 return false;
1199 sum = csum_block_add(sum, next, off);
1200 off += v.iov_len;
1201 0;
1202 }), ({
1203 char *p = kmap_atomic(v.bv_page);
1204 next = csum_partial_copy_nocheck(p + v.bv_offset,
1205 (to += v.bv_len) - v.bv_len,
1206 v.bv_len, 0);
1207 kunmap_atomic(p);
1208 sum = csum_block_add(sum, next, off);
1209 off += v.bv_len;
1210 }),({
1211 next = csum_partial_copy_nocheck(v.iov_base,
1212 (to += v.iov_len) - v.iov_len,
1213 v.iov_len, 0);
1214 sum = csum_block_add(sum, next, off);
1215 off += v.iov_len;
1216 })
1217 )
1218 *csum = sum;
1219 iov_iter_advance(i, bytes);
1220 return true;
1221 }
1222 EXPORT_SYMBOL(csum_and_copy_from_iter_full);
1223
1224 size_t csum_and_copy_to_iter(const void *addr, size_t bytes, __wsum *csum,
1225 struct iov_iter *i)
1226 {
1227 const char *from = addr;
1228 __wsum sum, next;
1229 size_t off = 0;
1230 sum = *csum;
1231 if (unlikely(i->type & ITER_PIPE)) {
1232 WARN_ON(1); /* for now */
1233 return 0;
1234 }
1235 iterate_and_advance(i, bytes, v, ({
1236 int err = 0;
1237 next = csum_and_copy_to_user((from += v.iov_len) - v.iov_len,
1238 v.iov_base,
1239 v.iov_len, 0, &err);
1240 if (!err) {
1241 sum = csum_block_add(sum, next, off);
1242 off += v.iov_len;
1243 }
1244 err ? v.iov_len : 0;
1245 }), ({
1246 char *p = kmap_atomic(v.bv_page);
1247 next = csum_partial_copy_nocheck((from += v.bv_len) - v.bv_len,
1248 p + v.bv_offset,
1249 v.bv_len, 0);
1250 kunmap_atomic(p);
1251 sum = csum_block_add(sum, next, off);
1252 off += v.bv_len;
1253 }),({
1254 next = csum_partial_copy_nocheck((from += v.iov_len) - v.iov_len,
1255 v.iov_base,
1256 v.iov_len, 0);
1257 sum = csum_block_add(sum, next, off);
1258 off += v.iov_len;
1259 })
1260 )
1261 *csum = sum;
1262 return bytes;
1263 }
1264 EXPORT_SYMBOL(csum_and_copy_to_iter);
1265
1266 int iov_iter_npages(const struct iov_iter *i, int maxpages)
1267 {
1268 size_t size = i->count;
1269 int npages = 0;
1270
1271 if (!size)
1272 return 0;
1273
1274 if (unlikely(i->type & ITER_PIPE)) {
1275 struct pipe_inode_info *pipe = i->pipe;
1276 size_t off;
1277 int idx;
1278
1279 if (!sanity(i))
1280 return 0;
1281
1282 data_start(i, &idx, &off);
1283 /* some of this one + all after this one */
1284 npages = ((pipe->curbuf - idx - 1) & (pipe->buffers - 1)) + 1;
1285 if (npages >= maxpages)
1286 return maxpages;
1287 } else iterate_all_kinds(i, size, v, ({
1288 unsigned long p = (unsigned long)v.iov_base;
1289 npages += DIV_ROUND_UP(p + v.iov_len, PAGE_SIZE)
1290 - p / PAGE_SIZE;
1291 if (npages >= maxpages)
1292 return maxpages;
1293 0;}),({
1294 npages++;
1295 if (npages >= maxpages)
1296 return maxpages;
1297 }),({
1298 unsigned long p = (unsigned long)v.iov_base;
1299 npages += DIV_ROUND_UP(p + v.iov_len, PAGE_SIZE)
1300 - p / PAGE_SIZE;
1301 if (npages >= maxpages)
1302 return maxpages;
1303 })
1304 )
1305 return npages;
1306 }
1307 EXPORT_SYMBOL(iov_iter_npages);
1308
1309 const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags)
1310 {
1311 *new = *old;
1312 if (unlikely(new->type & ITER_PIPE)) {
1313 WARN_ON(1);
1314 return NULL;
1315 }
1316 if (new->type & ITER_BVEC)
1317 return new->bvec = kmemdup(new->bvec,
1318 new->nr_segs * sizeof(struct bio_vec),
1319 flags);
1320 else
1321 /* iovec and kvec have identical layout */
1322 return new->iov = kmemdup(new->iov,
1323 new->nr_segs * sizeof(struct iovec),
1324 flags);
1325 }
1326 EXPORT_SYMBOL(dup_iter);
1327
1328 /**
1329 * import_iovec() - Copy an array of &struct iovec from userspace
1330 * into the kernel, check that it is valid, and initialize a new
1331 * &struct iov_iter iterator to access it.
1332 *
1333 * @type: One of %READ or %WRITE.
1334 * @uvector: Pointer to the userspace array.
1335 * @nr_segs: Number of elements in userspace array.
1336 * @fast_segs: Number of elements in @iov.
1337 * @iov: (input and output parameter) Pointer to pointer to (usually small
1338 * on-stack) kernel array.
1339 * @i: Pointer to iterator that will be initialized on success.
1340 *
1341 * If the array pointed to by *@iov is large enough to hold all @nr_segs,
1342 * then this function places %NULL in *@iov on return. Otherwise, a new
1343 * array will be allocated and the result placed in *@iov. This means that
1344 * the caller may call kfree() on *@iov regardless of whether the small
1345 * on-stack array was used or not (and regardless of whether this function
1346 * returns an error or not).
1347 *
1348 * Return: 0 on success or negative error code on error.
1349 */
1350 int import_iovec(int type, const struct iovec __user * uvector,
1351 unsigned nr_segs, unsigned fast_segs,
1352 struct iovec **iov, struct iov_iter *i)
1353 {
1354 ssize_t n;
1355 struct iovec *p;
1356 n = rw_copy_check_uvector(type, uvector, nr_segs, fast_segs,
1357 *iov, &p);
1358 if (n < 0) {
1359 if (p != *iov)
1360 kfree(p);
1361 *iov = NULL;
1362 return n;
1363 }
1364 iov_iter_init(i, type, p, nr_segs, n);
1365 *iov = p == *iov ? NULL : p;
1366 return 0;
1367 }
1368 EXPORT_SYMBOL(import_iovec);
1369
1370 #ifdef CONFIG_COMPAT
1371 #include <linux/compat.h>
1372
1373 int compat_import_iovec(int type, const struct compat_iovec __user * uvector,
1374 unsigned nr_segs, unsigned fast_segs,
1375 struct iovec **iov, struct iov_iter *i)
1376 {
1377 ssize_t n;
1378 struct iovec *p;
1379 n = compat_rw_copy_check_uvector(type, uvector, nr_segs, fast_segs,
1380 *iov, &p);
1381 if (n < 0) {
1382 if (p != *iov)
1383 kfree(p);
1384 *iov = NULL;
1385 return n;
1386 }
1387 iov_iter_init(i, type, p, nr_segs, n);
1388 *iov = p == *iov ? NULL : p;
1389 return 0;
1390 }
1391 #endif
1392
1393 int import_single_range(int rw, void __user *buf, size_t len,
1394 struct iovec *iov, struct iov_iter *i)
1395 {
1396 if (len > MAX_RW_COUNT)
1397 len = MAX_RW_COUNT;
1398 if (unlikely(!access_ok(!rw, buf, len)))
1399 return -EFAULT;
1400
1401 iov->iov_base = buf;
1402 iov->iov_len = len;
1403 iov_iter_init(i, rw, iov, 1, len);
1404 return 0;
1405 }
1406 EXPORT_SYMBOL(import_single_range);
Ubuntu Artful kernel mirror