Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
7999096f | 2 | #include <crypto/hash.h> |
4f18cd31 | 3 | #include <linux/export.h> |
2f8b5444 | 4 | #include <linux/bvec.h> |
4d0e9df5 | 5 | #include <linux/fault-inject-usercopy.h> |
4f18cd31 AV |
6 | #include <linux/uio.h> |
7 | #include <linux/pagemap.h> | |
28961998 | 8 | #include <linux/highmem.h> |
91f79c43 AV |
9 | #include <linux/slab.h> |
10 | #include <linux/vmalloc.h> | |
241699cd | 11 | #include <linux/splice.h> |
bfdc5970 | 12 | #include <linux/compat.h> |
a604ec7e | 13 | #include <net/checksum.h> |
d05f4435 | 14 | #include <linux/scatterlist.h> |
d0ef4c36 | 15 | #include <linux/instrumented.h> |
4f18cd31 | 16 | |
241699cd AV |
17 | #define PIPE_PARANOIA /* for now */ |
18 | ||
fcb14cb1 AV |
19 | /* covers ubuf and kbuf alike */ |
20 | #define iterate_buf(i, n, base, len, off, __p, STEP) { \ | |
21 | size_t __maybe_unused off = 0; \ | |
22 | len = n; \ | |
23 | base = __p + i->iov_offset; \ | |
24 | len -= (STEP); \ | |
25 | i->iov_offset += len; \ | |
26 | n = len; \ | |
27 | } | |
28 | ||
5c67aa90 | 29 | /* covers iovec and kvec alike */ |
a6e4ec7b | 30 | #define iterate_iovec(i, n, base, len, off, __p, STEP) { \ |
7baa5099 | 31 | size_t off = 0; \ |
a6e4ec7b | 32 | size_t skip = i->iov_offset; \ |
7a1bcb5d | 33 | do { \ |
7baa5099 AV |
34 | len = min(n, __p->iov_len - skip); \ |
35 | if (likely(len)) { \ | |
36 | base = __p->iov_base + skip; \ | |
37 | len -= (STEP); \ | |
38 | off += len; \ | |
39 | skip += len; \ | |
40 | n -= len; \ | |
7a1bcb5d AV |
41 | if (skip < __p->iov_len) \ |
42 | break; \ | |
43 | } \ | |
44 | __p++; \ | |
45 | skip = 0; \ | |
46 | } while (n); \ | |
a6e4ec7b | 47 | i->iov_offset = skip; \ |
7baa5099 | 48 | n = off; \ |
04a31165 AV |
49 | } |
50 | ||
a6e4ec7b | 51 | #define iterate_bvec(i, n, base, len, off, p, STEP) { \ |
7baa5099 | 52 | size_t off = 0; \ |
a6e4ec7b | 53 | unsigned skip = i->iov_offset; \ |
7491a2bf AV |
54 | while (n) { \ |
55 | unsigned offset = p->bv_offset + skip; \ | |
1b4fb5ff | 56 | unsigned left; \ |
21b56c84 AV |
57 | void *kaddr = kmap_local_page(p->bv_page + \ |
58 | offset / PAGE_SIZE); \ | |
7baa5099 | 59 | base = kaddr + offset % PAGE_SIZE; \ |
a6e4ec7b | 60 | len = min(min(n, (size_t)(p->bv_len - skip)), \ |
7491a2bf | 61 | (size_t)(PAGE_SIZE - offset % PAGE_SIZE)); \ |
1b4fb5ff | 62 | left = (STEP); \ |
21b56c84 | 63 | kunmap_local(kaddr); \ |
7baa5099 AV |
64 | len -= left; \ |
65 | off += len; \ | |
66 | skip += len; \ | |
7491a2bf AV |
67 | if (skip == p->bv_len) { \ |
68 | skip = 0; \ | |
69 | p++; \ | |
70 | } \ | |
7baa5099 | 71 | n -= len; \ |
1b4fb5ff AV |
72 | if (left) \ |
73 | break; \ | |
7491a2bf | 74 | } \ |
a6e4ec7b | 75 | i->iov_offset = skip; \ |
7baa5099 | 76 | n = off; \ |
04a31165 AV |
77 | } |
78 | ||
a6e4ec7b | 79 | #define iterate_xarray(i, n, base, len, __off, STEP) { \ |
1b4fb5ff | 80 | __label__ __out; \ |
622838f3 | 81 | size_t __off = 0; \ |
821979f5 | 82 | struct folio *folio; \ |
a6e4ec7b | 83 | loff_t start = i->xarray_start + i->iov_offset; \ |
4b179e9a | 84 | pgoff_t index = start / PAGE_SIZE; \ |
7ff50620 DH |
85 | XA_STATE(xas, i->xarray, index); \ |
86 | \ | |
821979f5 | 87 | len = PAGE_SIZE - offset_in_page(start); \ |
7baa5099 | 88 | rcu_read_lock(); \ |
821979f5 | 89 | xas_for_each(&xas, folio, ULONG_MAX) { \ |
7baa5099 | 90 | unsigned left; \ |
821979f5 MWO |
91 | size_t offset; \ |
92 | if (xas_retry(&xas, folio)) \ | |
7baa5099 | 93 | continue; \ |
821979f5 | 94 | if (WARN_ON(xa_is_value(folio))) \ |
7baa5099 | 95 | break; \ |
821979f5 | 96 | if (WARN_ON(folio_test_hugetlb(folio))) \ |
7baa5099 | 97 | break; \ |
821979f5 MWO |
98 | offset = offset_in_folio(folio, start + __off); \ |
99 | while (offset < folio_size(folio)) { \ | |
100 | base = kmap_local_folio(folio, offset); \ | |
7baa5099 AV |
101 | len = min(n, len); \ |
102 | left = (STEP); \ | |
821979f5 | 103 | kunmap_local(base); \ |
7baa5099 AV |
104 | len -= left; \ |
105 | __off += len; \ | |
106 | n -= len; \ | |
107 | if (left || n == 0) \ | |
108 | goto __out; \ | |
821979f5 MWO |
109 | offset += len; \ |
110 | len = PAGE_SIZE; \ | |
7baa5099 | 111 | } \ |
7ff50620 | 112 | } \ |
1b4fb5ff | 113 | __out: \ |
7ff50620 | 114 | rcu_read_unlock(); \ |
821979f5 | 115 | i->iov_offset += __off; \ |
622838f3 | 116 | n = __off; \ |
7ff50620 DH |
117 | } |
118 | ||
7baa5099 | 119 | #define __iterate_and_advance(i, n, base, len, off, I, K) { \ |
dd254f5a AV |
120 | if (unlikely(i->count < n)) \ |
121 | n = i->count; \ | |
f5da8354 | 122 | if (likely(n)) { \ |
fcb14cb1 AV |
123 | if (likely(iter_is_ubuf(i))) { \ |
124 | void __user *base; \ | |
125 | size_t len; \ | |
126 | iterate_buf(i, n, base, len, off, \ | |
127 | i->ubuf, (I)) \ | |
128 | } else if (likely(iter_is_iovec(i))) { \ | |
5c67aa90 | 129 | const struct iovec *iov = i->iov; \ |
7baa5099 AV |
130 | void __user *base; \ |
131 | size_t len; \ | |
132 | iterate_iovec(i, n, base, len, off, \ | |
a6e4ec7b | 133 | iov, (I)) \ |
28f38db7 AV |
134 | i->nr_segs -= iov - i->iov; \ |
135 | i->iov = iov; \ | |
136 | } else if (iov_iter_is_bvec(i)) { \ | |
1bdc76ae | 137 | const struct bio_vec *bvec = i->bvec; \ |
7baa5099 AV |
138 | void *base; \ |
139 | size_t len; \ | |
140 | iterate_bvec(i, n, base, len, off, \ | |
a6e4ec7b | 141 | bvec, (K)) \ |
7491a2bf AV |
142 | i->nr_segs -= bvec - i->bvec; \ |
143 | i->bvec = bvec; \ | |
28f38db7 | 144 | } else if (iov_iter_is_kvec(i)) { \ |
5c67aa90 | 145 | const struct kvec *kvec = i->kvec; \ |
7baa5099 AV |
146 | void *base; \ |
147 | size_t len; \ | |
148 | iterate_iovec(i, n, base, len, off, \ | |
a6e4ec7b | 149 | kvec, (K)) \ |
dd254f5a AV |
150 | i->nr_segs -= kvec - i->kvec; \ |
151 | i->kvec = kvec; \ | |
28f38db7 | 152 | } else if (iov_iter_is_xarray(i)) { \ |
7baa5099 AV |
153 | void *base; \ |
154 | size_t len; \ | |
155 | iterate_xarray(i, n, base, len, off, \ | |
a6e4ec7b | 156 | (K)) \ |
7ce2a91e | 157 | } \ |
dd254f5a | 158 | i->count -= n; \ |
7ce2a91e | 159 | } \ |
7ce2a91e | 160 | } |
7baa5099 AV |
161 | #define iterate_and_advance(i, n, base, len, off, I, K) \ |
162 | __iterate_and_advance(i, n, base, len, off, I, ((void)(K),0)) | |
7ce2a91e | 163 | |
09fc68dc AV |
164 | static int copyout(void __user *to, const void *from, size_t n) |
165 | { | |
4d0e9df5 AL |
166 | if (should_fail_usercopy()) |
167 | return n; | |
96d4f267 | 168 | if (access_ok(to, n)) { |
d0ef4c36 | 169 | instrument_copy_to_user(to, from, n); |
09fc68dc AV |
170 | n = raw_copy_to_user(to, from, n); |
171 | } | |
172 | return n; | |
173 | } | |
174 | ||
175 | static int copyin(void *to, const void __user *from, size_t n) | |
176 | { | |
4d0e9df5 AL |
177 | if (should_fail_usercopy()) |
178 | return n; | |
96d4f267 | 179 | if (access_ok(from, n)) { |
d0ef4c36 | 180 | instrument_copy_from_user(to, from, n); |
09fc68dc AV |
181 | n = raw_copy_from_user(to, from, n); |
182 | } | |
183 | return n; | |
184 | } | |
185 | ||
241699cd AV |
186 | #ifdef PIPE_PARANOIA |
187 | static bool sanity(const struct iov_iter *i) | |
188 | { | |
189 | struct pipe_inode_info *pipe = i->pipe; | |
8cefc107 DH |
190 | unsigned int p_head = pipe->head; |
191 | unsigned int p_tail = pipe->tail; | |
192 | unsigned int p_mask = pipe->ring_size - 1; | |
193 | unsigned int p_occupancy = pipe_occupancy(p_head, p_tail); | |
194 | unsigned int i_head = i->head; | |
195 | unsigned int idx; | |
196 | ||
241699cd AV |
197 | if (i->iov_offset) { |
198 | struct pipe_buffer *p; | |
8cefc107 | 199 | if (unlikely(p_occupancy == 0)) |
241699cd | 200 | goto Bad; // pipe must be non-empty |
8cefc107 | 201 | if (unlikely(i_head != p_head - 1)) |
241699cd AV |
202 | goto Bad; // must be at the last buffer... |
203 | ||
8cefc107 | 204 | p = &pipe->bufs[i_head & p_mask]; |
241699cd AV |
205 | if (unlikely(p->offset + p->len != i->iov_offset)) |
206 | goto Bad; // ... at the end of segment | |
207 | } else { | |
8cefc107 | 208 | if (i_head != p_head) |
241699cd AV |
209 | goto Bad; // must be right after the last buffer |
210 | } | |
211 | return true; | |
212 | Bad: | |
8cefc107 DH |
213 | printk(KERN_ERR "idx = %d, offset = %zd\n", i_head, i->iov_offset); |
214 | printk(KERN_ERR "head = %d, tail = %d, buffers = %d\n", | |
215 | p_head, p_tail, pipe->ring_size); | |
216 | for (idx = 0; idx < pipe->ring_size; idx++) | |
241699cd AV |
217 | printk(KERN_ERR "[%p %p %d %d]\n", |
218 | pipe->bufs[idx].ops, | |
219 | pipe->bufs[idx].page, | |
220 | pipe->bufs[idx].offset, | |
221 | pipe->bufs[idx].len); | |
222 | WARN_ON(1); | |
223 | return false; | |
224 | } | |
225 | #else | |
226 | #define sanity(i) true | |
227 | #endif | |
228 | ||
241699cd AV |
229 | static size_t copy_page_to_iter_pipe(struct page *page, size_t offset, size_t bytes, |
230 | struct iov_iter *i) | |
231 | { | |
232 | struct pipe_inode_info *pipe = i->pipe; | |
233 | struct pipe_buffer *buf; | |
8cefc107 DH |
234 | unsigned int p_tail = pipe->tail; |
235 | unsigned int p_mask = pipe->ring_size - 1; | |
236 | unsigned int i_head = i->head; | |
241699cd | 237 | size_t off; |
241699cd AV |
238 | |
239 | if (unlikely(bytes > i->count)) | |
240 | bytes = i->count; | |
241 | ||
242 | if (unlikely(!bytes)) | |
243 | return 0; | |
244 | ||
245 | if (!sanity(i)) | |
246 | return 0; | |
247 | ||
248 | off = i->iov_offset; | |
8cefc107 | 249 | buf = &pipe->bufs[i_head & p_mask]; |
241699cd AV |
250 | if (off) { |
251 | if (offset == off && buf->page == page) { | |
252 | /* merge with the last one */ | |
253 | buf->len += bytes; | |
254 | i->iov_offset += bytes; | |
255 | goto out; | |
256 | } | |
8cefc107 DH |
257 | i_head++; |
258 | buf = &pipe->bufs[i_head & p_mask]; | |
241699cd | 259 | } |
6718b6f8 | 260 | if (pipe_full(i_head, p_tail, pipe->max_usage)) |
241699cd | 261 | return 0; |
8cefc107 | 262 | |
241699cd | 263 | buf->ops = &page_cache_pipe_buf_ops; |
9d2231c5 | 264 | buf->flags = 0; |
8cefc107 DH |
265 | get_page(page); |
266 | buf->page = page; | |
241699cd AV |
267 | buf->offset = offset; |
268 | buf->len = bytes; | |
8cefc107 DH |
269 | |
270 | pipe->head = i_head + 1; | |
241699cd | 271 | i->iov_offset = offset + bytes; |
8cefc107 | 272 | i->head = i_head; |
241699cd AV |
273 | out: |
274 | i->count -= bytes; | |
275 | return bytes; | |
276 | } | |
277 | ||
171a0203 | 278 | /* |
a6294593 AG |
279 | * fault_in_iov_iter_readable - fault in iov iterator for reading |
280 | * @i: iterator | |
281 | * @size: maximum length | |
282 | * | |
171a0203 | 283 | * Fault in one or more iovecs of the given iov_iter, to a maximum length of |
a6294593 AG |
284 | * @size. For each iovec, fault in each page that constitutes the iovec. |
285 | * | |
286 | * Returns the number of bytes not faulted in (like copy_to_user() and | |
287 | * copy_from_user()). | |
171a0203 | 288 | * |
a6294593 | 289 | * Always returns 0 for non-userspace iterators. |
171a0203 | 290 | */ |
a6294593 | 291 | size_t fault_in_iov_iter_readable(const struct iov_iter *i, size_t size) |
171a0203 | 292 | { |
fcb14cb1 AV |
293 | if (iter_is_ubuf(i)) { |
294 | size_t n = min(size, iov_iter_count(i)); | |
295 | n -= fault_in_readable(i->ubuf + i->iov_offset, n); | |
296 | return size - n; | |
297 | } else if (iter_is_iovec(i)) { | |
a6294593 | 298 | size_t count = min(size, iov_iter_count(i)); |
8409a0d2 AV |
299 | const struct iovec *p; |
300 | size_t skip; | |
301 | ||
a6294593 AG |
302 | size -= count; |
303 | for (p = i->iov, skip = i->iov_offset; count; p++, skip = 0) { | |
304 | size_t len = min(count, p->iov_len - skip); | |
305 | size_t ret; | |
8409a0d2 AV |
306 | |
307 | if (unlikely(!len)) | |
308 | continue; | |
a6294593 AG |
309 | ret = fault_in_readable(p->iov_base + skip, len); |
310 | count -= len - ret; | |
311 | if (ret) | |
312 | break; | |
8409a0d2 | 313 | } |
a6294593 | 314 | return count + size; |
171a0203 AA |
315 | } |
316 | return 0; | |
317 | } | |
a6294593 | 318 | EXPORT_SYMBOL(fault_in_iov_iter_readable); |
171a0203 | 319 | |
cdd591fc AG |
320 | /* |
321 | * fault_in_iov_iter_writeable - fault in iov iterator for writing | |
322 | * @i: iterator | |
323 | * @size: maximum length | |
324 | * | |
325 | * Faults in the iterator using get_user_pages(), i.e., without triggering | |
326 | * hardware page faults. This is primarily useful when we already know that | |
327 | * some or all of the pages in @i aren't in memory. | |
328 | * | |
329 | * Returns the number of bytes not faulted in, like copy_to_user() and | |
330 | * copy_from_user(). | |
331 | * | |
332 | * Always returns 0 for non-user-space iterators. | |
333 | */ | |
334 | size_t fault_in_iov_iter_writeable(const struct iov_iter *i, size_t size) | |
335 | { | |
fcb14cb1 AV |
336 | if (iter_is_ubuf(i)) { |
337 | size_t n = min(size, iov_iter_count(i)); | |
338 | n -= fault_in_safe_writeable(i->ubuf + i->iov_offset, n); | |
339 | return size - n; | |
340 | } else if (iter_is_iovec(i)) { | |
cdd591fc AG |
341 | size_t count = min(size, iov_iter_count(i)); |
342 | const struct iovec *p; | |
343 | size_t skip; | |
344 | ||
345 | size -= count; | |
346 | for (p = i->iov, skip = i->iov_offset; count; p++, skip = 0) { | |
347 | size_t len = min(count, p->iov_len - skip); | |
348 | size_t ret; | |
349 | ||
350 | if (unlikely(!len)) | |
351 | continue; | |
352 | ret = fault_in_safe_writeable(p->iov_base + skip, len); | |
353 | count -= len - ret; | |
354 | if (ret) | |
355 | break; | |
356 | } | |
357 | return count + size; | |
358 | } | |
359 | return 0; | |
360 | } | |
361 | EXPORT_SYMBOL(fault_in_iov_iter_writeable); | |
362 | ||
aa563d7b | 363 | void iov_iter_init(struct iov_iter *i, unsigned int direction, |
71d8e532 AV |
364 | const struct iovec *iov, unsigned long nr_segs, |
365 | size_t count) | |
366 | { | |
aa563d7b | 367 | WARN_ON(direction & ~(READ | WRITE)); |
8cd54c1c AV |
368 | *i = (struct iov_iter) { |
369 | .iter_type = ITER_IOVEC, | |
3337ab08 | 370 | .nofault = false, |
fcb14cb1 | 371 | .user_backed = true, |
8cd54c1c AV |
372 | .data_source = direction, |
373 | .iov = iov, | |
374 | .nr_segs = nr_segs, | |
375 | .iov_offset = 0, | |
376 | .count = count | |
377 | }; | |
71d8e532 AV |
378 | } |
379 | EXPORT_SYMBOL(iov_iter_init); | |
7b2c99d1 | 380 | |
241699cd AV |
381 | static inline bool allocated(struct pipe_buffer *buf) |
382 | { | |
383 | return buf->ops == &default_pipe_buf_ops; | |
384 | } | |
385 | ||
8cefc107 DH |
386 | static inline void data_start(const struct iov_iter *i, |
387 | unsigned int *iter_headp, size_t *offp) | |
241699cd | 388 | { |
8cefc107 DH |
389 | unsigned int p_mask = i->pipe->ring_size - 1; |
390 | unsigned int iter_head = i->head; | |
241699cd | 391 | size_t off = i->iov_offset; |
8cefc107 DH |
392 | |
393 | if (off && (!allocated(&i->pipe->bufs[iter_head & p_mask]) || | |
394 | off == PAGE_SIZE)) { | |
395 | iter_head++; | |
241699cd AV |
396 | off = 0; |
397 | } | |
8cefc107 | 398 | *iter_headp = iter_head; |
241699cd AV |
399 | *offp = off; |
400 | } | |
401 | ||
402 | static size_t push_pipe(struct iov_iter *i, size_t size, | |
8cefc107 | 403 | int *iter_headp, size_t *offp) |
241699cd AV |
404 | { |
405 | struct pipe_inode_info *pipe = i->pipe; | |
8cefc107 DH |
406 | unsigned int p_tail = pipe->tail; |
407 | unsigned int p_mask = pipe->ring_size - 1; | |
408 | unsigned int iter_head; | |
241699cd | 409 | size_t off; |
241699cd AV |
410 | ssize_t left; |
411 | ||
412 | if (unlikely(size > i->count)) | |
413 | size = i->count; | |
414 | if (unlikely(!size)) | |
415 | return 0; | |
416 | ||
417 | left = size; | |
8cefc107 DH |
418 | data_start(i, &iter_head, &off); |
419 | *iter_headp = iter_head; | |
241699cd AV |
420 | *offp = off; |
421 | if (off) { | |
422 | left -= PAGE_SIZE - off; | |
423 | if (left <= 0) { | |
8cefc107 | 424 | pipe->bufs[iter_head & p_mask].len += size; |
241699cd AV |
425 | return size; |
426 | } | |
8cefc107 DH |
427 | pipe->bufs[iter_head & p_mask].len = PAGE_SIZE; |
428 | iter_head++; | |
241699cd | 429 | } |
6718b6f8 | 430 | while (!pipe_full(iter_head, p_tail, pipe->max_usage)) { |
8cefc107 | 431 | struct pipe_buffer *buf = &pipe->bufs[iter_head & p_mask]; |
241699cd AV |
432 | struct page *page = alloc_page(GFP_USER); |
433 | if (!page) | |
434 | break; | |
8cefc107 DH |
435 | |
436 | buf->ops = &default_pipe_buf_ops; | |
9d2231c5 | 437 | buf->flags = 0; |
8cefc107 DH |
438 | buf->page = page; |
439 | buf->offset = 0; | |
440 | buf->len = min_t(ssize_t, left, PAGE_SIZE); | |
441 | left -= buf->len; | |
442 | iter_head++; | |
443 | pipe->head = iter_head; | |
444 | ||
445 | if (left == 0) | |
241699cd | 446 | return size; |
241699cd AV |
447 | } |
448 | return size - left; | |
449 | } | |
450 | ||
451 | static size_t copy_pipe_to_iter(const void *addr, size_t bytes, | |
452 | struct iov_iter *i) | |
453 | { | |
454 | struct pipe_inode_info *pipe = i->pipe; | |
8cefc107 DH |
455 | unsigned int p_mask = pipe->ring_size - 1; |
456 | unsigned int i_head; | |
241699cd | 457 | size_t n, off; |
241699cd AV |
458 | |
459 | if (!sanity(i)) | |
460 | return 0; | |
461 | ||
8cefc107 | 462 | bytes = n = push_pipe(i, bytes, &i_head, &off); |
241699cd AV |
463 | if (unlikely(!n)) |
464 | return 0; | |
8cefc107 | 465 | do { |
241699cd | 466 | size_t chunk = min_t(size_t, n, PAGE_SIZE - off); |
8cefc107 DH |
467 | memcpy_to_page(pipe->bufs[i_head & p_mask].page, off, addr, chunk); |
468 | i->head = i_head; | |
241699cd AV |
469 | i->iov_offset = off + chunk; |
470 | n -= chunk; | |
471 | addr += chunk; | |
8cefc107 DH |
472 | off = 0; |
473 | i_head++; | |
474 | } while (n); | |
241699cd AV |
475 | i->count -= bytes; |
476 | return bytes; | |
477 | } | |
478 | ||
f9152895 AV |
479 | static __wsum csum_and_memcpy(void *to, const void *from, size_t len, |
480 | __wsum sum, size_t off) | |
481 | { | |
cc44c17b | 482 | __wsum next = csum_partial_copy_nocheck(from, to, len); |
f9152895 AV |
483 | return csum_block_add(sum, next, off); |
484 | } | |
485 | ||
78e1f386 | 486 | static size_t csum_and_copy_to_pipe_iter(const void *addr, size_t bytes, |
6852df12 | 487 | struct iov_iter *i, __wsum *sump) |
78e1f386 AV |
488 | { |
489 | struct pipe_inode_info *pipe = i->pipe; | |
8cefc107 | 490 | unsigned int p_mask = pipe->ring_size - 1; |
6852df12 AV |
491 | __wsum sum = *sump; |
492 | size_t off = 0; | |
8cefc107 | 493 | unsigned int i_head; |
6852df12 | 494 | size_t r; |
78e1f386 AV |
495 | |
496 | if (!sanity(i)) | |
497 | return 0; | |
498 | ||
6852df12 AV |
499 | bytes = push_pipe(i, bytes, &i_head, &r); |
500 | while (bytes) { | |
501 | size_t chunk = min_t(size_t, bytes, PAGE_SIZE - r); | |
2495bdcc | 502 | char *p = kmap_local_page(pipe->bufs[i_head & p_mask].page); |
6852df12 | 503 | sum = csum_and_memcpy(p + r, addr + off, chunk, sum, off); |
2495bdcc | 504 | kunmap_local(p); |
8cefc107 | 505 | i->head = i_head; |
78e1f386 | 506 | i->iov_offset = r + chunk; |
6852df12 | 507 | bytes -= chunk; |
78e1f386 | 508 | off += chunk; |
8cefc107 DH |
509 | r = 0; |
510 | i_head++; | |
6852df12 AV |
511 | } |
512 | *sump = sum; | |
513 | i->count -= off; | |
514 | return off; | |
78e1f386 AV |
515 | } |
516 | ||
aa28de27 | 517 | size_t _copy_to_iter(const void *addr, size_t bytes, struct iov_iter *i) |
62a8067a | 518 | { |
00e23707 | 519 | if (unlikely(iov_iter_is_pipe(i))) |
241699cd | 520 | return copy_pipe_to_iter(addr, bytes, i); |
fcb14cb1 | 521 | if (user_backed_iter(i)) |
09fc68dc | 522 | might_fault(); |
7baa5099 AV |
523 | iterate_and_advance(i, bytes, base, len, off, |
524 | copyout(base, addr + off, len), | |
525 | memcpy(base, addr + off, len) | |
3d4d3e48 | 526 | ) |
62a8067a | 527 | |
3d4d3e48 | 528 | return bytes; |
c35e0248 | 529 | } |
aa28de27 | 530 | EXPORT_SYMBOL(_copy_to_iter); |
c35e0248 | 531 | |
ec6347bb DW |
532 | #ifdef CONFIG_ARCH_HAS_COPY_MC |
533 | static int copyout_mc(void __user *to, const void *from, size_t n) | |
8780356e | 534 | { |
96d4f267 | 535 | if (access_ok(to, n)) { |
d0ef4c36 | 536 | instrument_copy_to_user(to, from, n); |
ec6347bb | 537 | n = copy_mc_to_user((__force void *) to, from, n); |
8780356e DW |
538 | } |
539 | return n; | |
540 | } | |
541 | ||
ec6347bb | 542 | static size_t copy_mc_pipe_to_iter(const void *addr, size_t bytes, |
ca146f6f DW |
543 | struct iov_iter *i) |
544 | { | |
545 | struct pipe_inode_info *pipe = i->pipe; | |
8cefc107 DH |
546 | unsigned int p_mask = pipe->ring_size - 1; |
547 | unsigned int i_head; | |
c3497fd0 | 548 | unsigned int valid = pipe->head; |
ca146f6f | 549 | size_t n, off, xfer = 0; |
ca146f6f DW |
550 | |
551 | if (!sanity(i)) | |
552 | return 0; | |
553 | ||
2a510a74 AV |
554 | n = push_pipe(i, bytes, &i_head, &off); |
555 | while (n) { | |
ca146f6f | 556 | size_t chunk = min_t(size_t, n, PAGE_SIZE - off); |
2a510a74 | 557 | char *p = kmap_local_page(pipe->bufs[i_head & p_mask].page); |
ca146f6f | 558 | unsigned long rem; |
2a510a74 AV |
559 | rem = copy_mc_to_kernel(p + off, addr + xfer, chunk); |
560 | chunk -= rem; | |
561 | kunmap_local(p); | |
c3497fd0 AV |
562 | if (chunk) { |
563 | i->head = i_head; | |
564 | i->iov_offset = off + chunk; | |
565 | xfer += chunk; | |
566 | valid = i_head + 1; | |
567 | } | |
568 | if (rem) { | |
569 | pipe->bufs[i_head & p_mask].len -= rem; | |
570 | pipe_discard_from(pipe, valid); | |
ca146f6f | 571 | break; |
c3497fd0 | 572 | } |
ca146f6f | 573 | n -= chunk; |
8cefc107 DH |
574 | off = 0; |
575 | i_head++; | |
2a510a74 | 576 | } |
ca146f6f DW |
577 | i->count -= xfer; |
578 | return xfer; | |
579 | } | |
580 | ||
bf3eeb9b | 581 | /** |
ec6347bb | 582 | * _copy_mc_to_iter - copy to iter with source memory error exception handling |
bf3eeb9b DW |
583 | * @addr: source kernel address |
584 | * @bytes: total transfer length | |
44e55997 | 585 | * @i: destination iterator |
bf3eeb9b | 586 | * |
ec6347bb DW |
587 | * The pmem driver deploys this for the dax operation |
588 | * (dax_copy_to_iter()) for dax reads (bypass page-cache and the | |
589 | * block-layer). Upon #MC read(2) aborts and returns EIO or the bytes | |
590 | * successfully copied. | |
bf3eeb9b | 591 | * |
ec6347bb | 592 | * The main differences between this and typical _copy_to_iter(). |
bf3eeb9b DW |
593 | * |
594 | * * Typical tail/residue handling after a fault retries the copy | |
595 | * byte-by-byte until the fault happens again. Re-triggering machine | |
596 | * checks is potentially fatal so the implementation uses source | |
597 | * alignment and poison alignment assumptions to avoid re-triggering | |
598 | * hardware exceptions. | |
599 | * | |
600 | * * ITER_KVEC, ITER_PIPE, and ITER_BVEC can return short copies. | |
601 | * Compare to copy_to_iter() where only ITER_IOVEC attempts might return | |
602 | * a short copy. | |
44e55997 RD |
603 | * |
604 | * Return: number of bytes copied (may be %0) | |
bf3eeb9b | 605 | */ |
ec6347bb | 606 | size_t _copy_mc_to_iter(const void *addr, size_t bytes, struct iov_iter *i) |
8780356e | 607 | { |
00e23707 | 608 | if (unlikely(iov_iter_is_pipe(i))) |
ec6347bb | 609 | return copy_mc_pipe_to_iter(addr, bytes, i); |
fcb14cb1 | 610 | if (user_backed_iter(i)) |
8780356e | 611 | might_fault(); |
7baa5099 AV |
612 | __iterate_and_advance(i, bytes, base, len, off, |
613 | copyout_mc(base, addr + off, len), | |
614 | copy_mc_to_kernel(base, addr + off, len) | |
8780356e DW |
615 | ) |
616 | ||
617 | return bytes; | |
618 | } | |
ec6347bb DW |
619 | EXPORT_SYMBOL_GPL(_copy_mc_to_iter); |
620 | #endif /* CONFIG_ARCH_HAS_COPY_MC */ | |
8780356e | 621 | |
aa28de27 | 622 | size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i) |
c35e0248 | 623 | { |
00e23707 | 624 | if (unlikely(iov_iter_is_pipe(i))) { |
241699cd AV |
625 | WARN_ON(1); |
626 | return 0; | |
627 | } | |
fcb14cb1 | 628 | if (user_backed_iter(i)) |
09fc68dc | 629 | might_fault(); |
7baa5099 AV |
630 | iterate_and_advance(i, bytes, base, len, off, |
631 | copyin(addr + off, base, len), | |
632 | memcpy(addr + off, base, len) | |
0dbca9a4 AV |
633 | ) |
634 | ||
635 | return bytes; | |
c35e0248 | 636 | } |
aa28de27 | 637 | EXPORT_SYMBOL(_copy_from_iter); |
c35e0248 | 638 | |
aa28de27 | 639 | size_t _copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i) |
aa583096 | 640 | { |
00e23707 | 641 | if (unlikely(iov_iter_is_pipe(i))) { |
241699cd AV |
642 | WARN_ON(1); |
643 | return 0; | |
644 | } | |
7baa5099 AV |
645 | iterate_and_advance(i, bytes, base, len, off, |
646 | __copy_from_user_inatomic_nocache(addr + off, base, len), | |
647 | memcpy(addr + off, base, len) | |
aa583096 AV |
648 | ) |
649 | ||
650 | return bytes; | |
651 | } | |
aa28de27 | 652 | EXPORT_SYMBOL(_copy_from_iter_nocache); |
aa583096 | 653 | |
0aed55af | 654 | #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE |
abd08d7d DW |
655 | /** |
656 | * _copy_from_iter_flushcache - write destination through cpu cache | |
657 | * @addr: destination kernel address | |
658 | * @bytes: total transfer length | |
44e55997 | 659 | * @i: source iterator |
abd08d7d DW |
660 | * |
661 | * The pmem driver arranges for filesystem-dax to use this facility via | |
662 | * dax_copy_from_iter() for ensuring that writes to persistent memory | |
663 | * are flushed through the CPU cache. It is differentiated from | |
664 | * _copy_from_iter_nocache() in that guarantees all data is flushed for | |
665 | * all iterator types. The _copy_from_iter_nocache() only attempts to | |
666 | * bypass the cache for the ITER_IOVEC case, and on some archs may use | |
667 | * instructions that strand dirty-data in the cache. | |
44e55997 RD |
668 | * |
669 | * Return: number of bytes copied (may be %0) | |
abd08d7d | 670 | */ |
6a37e940 | 671 | size_t _copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i) |
0aed55af | 672 | { |
00e23707 | 673 | if (unlikely(iov_iter_is_pipe(i))) { |
0aed55af DW |
674 | WARN_ON(1); |
675 | return 0; | |
676 | } | |
7baa5099 AV |
677 | iterate_and_advance(i, bytes, base, len, off, |
678 | __copy_from_user_flushcache(addr + off, base, len), | |
679 | memcpy_flushcache(addr + off, base, len) | |
0aed55af DW |
680 | ) |
681 | ||
682 | return bytes; | |
683 | } | |
6a37e940 | 684 | EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache); |
0aed55af DW |
685 | #endif |
686 | ||
72e809ed AV |
687 | static inline bool page_copy_sane(struct page *page, size_t offset, size_t n) |
688 | { | |
6daef95b ED |
689 | struct page *head; |
690 | size_t v = n + offset; | |
691 | ||
692 | /* | |
693 | * The general case needs to access the page order in order | |
694 | * to compute the page size. | |
695 | * However, we mostly deal with order-0 pages and thus can | |
696 | * avoid a possible cache line miss for requests that fit all | |
697 | * page orders. | |
698 | */ | |
699 | if (n <= v && v <= PAGE_SIZE) | |
700 | return true; | |
701 | ||
702 | head = compound_head(page); | |
703 | v += (page - head) << PAGE_SHIFT; | |
a90bcb86 | 704 | |
a50b854e | 705 | if (likely(n <= v && v <= (page_size(head)))) |
72e809ed AV |
706 | return true; |
707 | WARN_ON(1); | |
708 | return false; | |
709 | } | |
cbbd26b8 | 710 | |
08aa6479 | 711 | static size_t __copy_page_to_iter(struct page *page, size_t offset, size_t bytes, |
62a8067a AV |
712 | struct iov_iter *i) |
713 | { | |
59bb69c6 AV |
714 | if (unlikely(iov_iter_is_pipe(i))) { |
715 | return copy_page_to_iter_pipe(page, offset, bytes, i); | |
716 | } else { | |
c1d4d6a9 AV |
717 | void *kaddr = kmap_local_page(page); |
718 | size_t wanted = _copy_to_iter(kaddr + offset, bytes, i); | |
719 | kunmap_local(kaddr); | |
d271524a | 720 | return wanted; |
28f38db7 | 721 | } |
62a8067a | 722 | } |
08aa6479 AV |
723 | |
724 | size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes, | |
725 | struct iov_iter *i) | |
726 | { | |
727 | size_t res = 0; | |
728 | if (unlikely(!page_copy_sane(page, offset, bytes))) | |
729 | return 0; | |
730 | page += offset / PAGE_SIZE; // first subpage | |
731 | offset %= PAGE_SIZE; | |
732 | while (1) { | |
733 | size_t n = __copy_page_to_iter(page, offset, | |
734 | min(bytes, (size_t)PAGE_SIZE - offset), i); | |
735 | res += n; | |
736 | bytes -= n; | |
737 | if (!bytes || !n) | |
738 | break; | |
739 | offset += n; | |
740 | if (offset == PAGE_SIZE) { | |
741 | page++; | |
742 | offset = 0; | |
743 | } | |
744 | } | |
745 | return res; | |
746 | } | |
62a8067a AV |
747 | EXPORT_SYMBOL(copy_page_to_iter); |
748 | ||
749 | size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes, | |
750 | struct iov_iter *i) | |
751 | { | |
59bb69c6 | 752 | if (page_copy_sane(page, offset, bytes)) { |
55ca375c | 753 | void *kaddr = kmap_local_page(page); |
aa28de27 | 754 | size_t wanted = _copy_from_iter(kaddr + offset, bytes, i); |
55ca375c | 755 | kunmap_local(kaddr); |
d271524a | 756 | return wanted; |
28f38db7 | 757 | } |
28f38db7 | 758 | return 0; |
62a8067a AV |
759 | } |
760 | EXPORT_SYMBOL(copy_page_from_iter); | |
761 | ||
241699cd AV |
762 | static size_t pipe_zero(size_t bytes, struct iov_iter *i) |
763 | { | |
764 | struct pipe_inode_info *pipe = i->pipe; | |
8cefc107 DH |
765 | unsigned int p_mask = pipe->ring_size - 1; |
766 | unsigned int i_head; | |
241699cd | 767 | size_t n, off; |
241699cd AV |
768 | |
769 | if (!sanity(i)) | |
770 | return 0; | |
771 | ||
8cefc107 | 772 | bytes = n = push_pipe(i, bytes, &i_head, &off); |
241699cd AV |
773 | if (unlikely(!n)) |
774 | return 0; | |
775 | ||
8cefc107 | 776 | do { |
241699cd | 777 | size_t chunk = min_t(size_t, n, PAGE_SIZE - off); |
893839fd AV |
778 | char *p = kmap_local_page(pipe->bufs[i_head & p_mask].page); |
779 | memset(p + off, 0, chunk); | |
780 | kunmap_local(p); | |
8cefc107 | 781 | i->head = i_head; |
241699cd AV |
782 | i->iov_offset = off + chunk; |
783 | n -= chunk; | |
8cefc107 DH |
784 | off = 0; |
785 | i_head++; | |
786 | } while (n); | |
241699cd AV |
787 | i->count -= bytes; |
788 | return bytes; | |
789 | } | |
790 | ||
c35e0248 MW |
791 | size_t iov_iter_zero(size_t bytes, struct iov_iter *i) |
792 | { | |
00e23707 | 793 | if (unlikely(iov_iter_is_pipe(i))) |
241699cd | 794 | return pipe_zero(bytes, i); |
7baa5099 AV |
795 | iterate_and_advance(i, bytes, base, len, count, |
796 | clear_user(base, len), | |
797 | memset(base, 0, len) | |
8442fa46 AV |
798 | ) |
799 | ||
800 | return bytes; | |
c35e0248 MW |
801 | } |
802 | EXPORT_SYMBOL(iov_iter_zero); | |
803 | ||
f0b65f39 AV |
804 | size_t copy_page_from_iter_atomic(struct page *page, unsigned offset, size_t bytes, |
805 | struct iov_iter *i) | |
62a8067a | 806 | { |
04a31165 | 807 | char *kaddr = kmap_atomic(page), *p = kaddr + offset; |
72e809ed AV |
808 | if (unlikely(!page_copy_sane(page, offset, bytes))) { |
809 | kunmap_atomic(kaddr); | |
810 | return 0; | |
811 | } | |
9ea9ce04 | 812 | if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) { |
241699cd AV |
813 | kunmap_atomic(kaddr); |
814 | WARN_ON(1); | |
815 | return 0; | |
816 | } | |
7baa5099 AV |
817 | iterate_and_advance(i, bytes, base, len, off, |
818 | copyin(p + off, base, len), | |
819 | memcpy(p + off, base, len) | |
04a31165 AV |
820 | ) |
821 | kunmap_atomic(kaddr); | |
822 | return bytes; | |
62a8067a | 823 | } |
f0b65f39 | 824 | EXPORT_SYMBOL(copy_page_from_iter_atomic); |
62a8067a | 825 | |
b9dc6f65 AV |
826 | static inline void pipe_truncate(struct iov_iter *i) |
827 | { | |
828 | struct pipe_inode_info *pipe = i->pipe; | |
8cefc107 DH |
829 | unsigned int p_tail = pipe->tail; |
830 | unsigned int p_head = pipe->head; | |
831 | unsigned int p_mask = pipe->ring_size - 1; | |
832 | ||
833 | if (!pipe_empty(p_head, p_tail)) { | |
834 | struct pipe_buffer *buf; | |
835 | unsigned int i_head = i->head; | |
b9dc6f65 | 836 | size_t off = i->iov_offset; |
8cefc107 | 837 | |
b9dc6f65 | 838 | if (off) { |
8cefc107 DH |
839 | buf = &pipe->bufs[i_head & p_mask]; |
840 | buf->len = off - buf->offset; | |
841 | i_head++; | |
b9dc6f65 | 842 | } |
8cefc107 DH |
843 | while (p_head != i_head) { |
844 | p_head--; | |
845 | pipe_buf_release(pipe, &pipe->bufs[p_head & p_mask]); | |
b9dc6f65 | 846 | } |
8cefc107 DH |
847 | |
848 | pipe->head = p_head; | |
b9dc6f65 AV |
849 | } |
850 | } | |
851 | ||
241699cd AV |
852 | static void pipe_advance(struct iov_iter *i, size_t size) |
853 | { | |
854 | struct pipe_inode_info *pipe = i->pipe; | |
241699cd | 855 | if (size) { |
b9dc6f65 | 856 | struct pipe_buffer *buf; |
8cefc107 DH |
857 | unsigned int p_mask = pipe->ring_size - 1; |
858 | unsigned int i_head = i->head; | |
b9dc6f65 | 859 | size_t off = i->iov_offset, left = size; |
8cefc107 | 860 | |
241699cd | 861 | if (off) /* make it relative to the beginning of buffer */ |
8cefc107 | 862 | left += off - pipe->bufs[i_head & p_mask].offset; |
241699cd | 863 | while (1) { |
8cefc107 | 864 | buf = &pipe->bufs[i_head & p_mask]; |
b9dc6f65 | 865 | if (left <= buf->len) |
241699cd | 866 | break; |
b9dc6f65 | 867 | left -= buf->len; |
8cefc107 | 868 | i_head++; |
241699cd | 869 | } |
8cefc107 | 870 | i->head = i_head; |
b9dc6f65 | 871 | i->iov_offset = buf->offset + left; |
241699cd | 872 | } |
b9dc6f65 AV |
873 | i->count -= size; |
874 | /* ... and discard everything past that point */ | |
875 | pipe_truncate(i); | |
241699cd AV |
876 | } |
877 | ||
54c8195b PB |
878 | static void iov_iter_bvec_advance(struct iov_iter *i, size_t size) |
879 | { | |
18fa9af7 | 880 | const struct bio_vec *bvec, *end; |
54c8195b | 881 | |
18fa9af7 AV |
882 | if (!i->count) |
883 | return; | |
884 | i->count -= size; | |
885 | ||
886 | size += i->iov_offset; | |
54c8195b | 887 | |
18fa9af7 AV |
888 | for (bvec = i->bvec, end = bvec + i->nr_segs; bvec < end; bvec++) { |
889 | if (likely(size < bvec->bv_len)) | |
890 | break; | |
891 | size -= bvec->bv_len; | |
892 | } | |
893 | i->iov_offset = size; | |
894 | i->nr_segs -= bvec - i->bvec; | |
895 | i->bvec = bvec; | |
54c8195b PB |
896 | } |
897 | ||
185ac4d4 AV |
898 | static void iov_iter_iovec_advance(struct iov_iter *i, size_t size) |
899 | { | |
900 | const struct iovec *iov, *end; | |
901 | ||
902 | if (!i->count) | |
903 | return; | |
904 | i->count -= size; | |
905 | ||
906 | size += i->iov_offset; // from beginning of current segment | |
907 | for (iov = i->iov, end = iov + i->nr_segs; iov < end; iov++) { | |
908 | if (likely(size < iov->iov_len)) | |
909 | break; | |
910 | size -= iov->iov_len; | |
911 | } | |
912 | i->iov_offset = size; | |
913 | i->nr_segs -= iov - i->iov; | |
914 | i->iov = iov; | |
915 | } | |
916 | ||
62a8067a AV |
917 | void iov_iter_advance(struct iov_iter *i, size_t size) |
918 | { | |
3b3fc051 AV |
919 | if (unlikely(i->count < size)) |
920 | size = i->count; | |
fcb14cb1 AV |
921 | if (likely(iter_is_ubuf(i)) || unlikely(iov_iter_is_xarray(i))) { |
922 | i->iov_offset += size; | |
923 | i->count -= size; | |
924 | } else if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) { | |
185ac4d4 AV |
925 | /* iovec and kvec have identical layouts */ |
926 | iov_iter_iovec_advance(i, size); | |
927 | } else if (iov_iter_is_bvec(i)) { | |
928 | iov_iter_bvec_advance(i, size); | |
929 | } else if (iov_iter_is_pipe(i)) { | |
241699cd | 930 | pipe_advance(i, size); |
185ac4d4 AV |
931 | } else if (iov_iter_is_discard(i)) { |
932 | i->count -= size; | |
54c8195b | 933 | } |
62a8067a AV |
934 | } |
935 | EXPORT_SYMBOL(iov_iter_advance); | |
936 | ||
27c0e374 AV |
937 | void iov_iter_revert(struct iov_iter *i, size_t unroll) |
938 | { | |
939 | if (!unroll) | |
940 | return; | |
5b47d59a AV |
941 | if (WARN_ON(unroll > MAX_RW_COUNT)) |
942 | return; | |
27c0e374 | 943 | i->count += unroll; |
00e23707 | 944 | if (unlikely(iov_iter_is_pipe(i))) { |
27c0e374 | 945 | struct pipe_inode_info *pipe = i->pipe; |
8cefc107 DH |
946 | unsigned int p_mask = pipe->ring_size - 1; |
947 | unsigned int i_head = i->head; | |
27c0e374 AV |
948 | size_t off = i->iov_offset; |
949 | while (1) { | |
8cefc107 DH |
950 | struct pipe_buffer *b = &pipe->bufs[i_head & p_mask]; |
951 | size_t n = off - b->offset; | |
27c0e374 | 952 | if (unroll < n) { |
4fa55cef | 953 | off -= unroll; |
27c0e374 AV |
954 | break; |
955 | } | |
956 | unroll -= n; | |
8cefc107 | 957 | if (!unroll && i_head == i->start_head) { |
27c0e374 AV |
958 | off = 0; |
959 | break; | |
960 | } | |
8cefc107 DH |
961 | i_head--; |
962 | b = &pipe->bufs[i_head & p_mask]; | |
963 | off = b->offset + b->len; | |
27c0e374 AV |
964 | } |
965 | i->iov_offset = off; | |
8cefc107 | 966 | i->head = i_head; |
27c0e374 AV |
967 | pipe_truncate(i); |
968 | return; | |
969 | } | |
9ea9ce04 DH |
970 | if (unlikely(iov_iter_is_discard(i))) |
971 | return; | |
27c0e374 AV |
972 | if (unroll <= i->iov_offset) { |
973 | i->iov_offset -= unroll; | |
974 | return; | |
975 | } | |
976 | unroll -= i->iov_offset; | |
fcb14cb1 | 977 | if (iov_iter_is_xarray(i) || iter_is_ubuf(i)) { |
7ff50620 DH |
978 | BUG(); /* We should never go beyond the start of the specified |
979 | * range since we might then be straying into pages that | |
980 | * aren't pinned. | |
981 | */ | |
982 | } else if (iov_iter_is_bvec(i)) { | |
27c0e374 AV |
983 | const struct bio_vec *bvec = i->bvec; |
984 | while (1) { | |
985 | size_t n = (--bvec)->bv_len; | |
986 | i->nr_segs++; | |
987 | if (unroll <= n) { | |
988 | i->bvec = bvec; | |
989 | i->iov_offset = n - unroll; | |
990 | return; | |
991 | } | |
992 | unroll -= n; | |
993 | } | |
994 | } else { /* same logics for iovec and kvec */ | |
995 | const struct iovec *iov = i->iov; | |
996 | while (1) { | |
997 | size_t n = (--iov)->iov_len; | |
998 | i->nr_segs++; | |
999 | if (unroll <= n) { | |
1000 | i->iov = iov; | |
1001 | i->iov_offset = n - unroll; | |
1002 | return; | |
1003 | } | |
1004 | unroll -= n; | |
1005 | } | |
1006 | } | |
1007 | } | |
1008 | EXPORT_SYMBOL(iov_iter_revert); | |
1009 | ||
62a8067a AV |
1010 | /* |
1011 | * Return the count of just the current iov_iter segment. | |
1012 | */ | |
1013 | size_t iov_iter_single_seg_count(const struct iov_iter *i) | |
1014 | { | |
28f38db7 AV |
1015 | if (i->nr_segs > 1) { |
1016 | if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) | |
1017 | return min(i->count, i->iov->iov_len - i->iov_offset); | |
1018 | if (iov_iter_is_bvec(i)) | |
1019 | return min(i->count, i->bvec->bv_len - i->iov_offset); | |
1020 | } | |
1021 | return i->count; | |
62a8067a AV |
1022 | } |
1023 | EXPORT_SYMBOL(iov_iter_single_seg_count); | |
1024 | ||
aa563d7b | 1025 | void iov_iter_kvec(struct iov_iter *i, unsigned int direction, |
05afcb77 | 1026 | const struct kvec *kvec, unsigned long nr_segs, |
abb78f87 AV |
1027 | size_t count) |
1028 | { | |
aa563d7b | 1029 | WARN_ON(direction & ~(READ | WRITE)); |
8cd54c1c AV |
1030 | *i = (struct iov_iter){ |
1031 | .iter_type = ITER_KVEC, | |
1032 | .data_source = direction, | |
1033 | .kvec = kvec, | |
1034 | .nr_segs = nr_segs, | |
1035 | .iov_offset = 0, | |
1036 | .count = count | |
1037 | }; | |
abb78f87 AV |
1038 | } |
1039 | EXPORT_SYMBOL(iov_iter_kvec); | |
1040 | ||
aa563d7b | 1041 | void iov_iter_bvec(struct iov_iter *i, unsigned int direction, |
05afcb77 AV |
1042 | const struct bio_vec *bvec, unsigned long nr_segs, |
1043 | size_t count) | |
1044 | { | |
aa563d7b | 1045 | WARN_ON(direction & ~(READ | WRITE)); |
8cd54c1c AV |
1046 | *i = (struct iov_iter){ |
1047 | .iter_type = ITER_BVEC, | |
1048 | .data_source = direction, | |
1049 | .bvec = bvec, | |
1050 | .nr_segs = nr_segs, | |
1051 | .iov_offset = 0, | |
1052 | .count = count | |
1053 | }; | |
05afcb77 AV |
1054 | } |
1055 | EXPORT_SYMBOL(iov_iter_bvec); | |
1056 | ||
aa563d7b | 1057 | void iov_iter_pipe(struct iov_iter *i, unsigned int direction, |
241699cd AV |
1058 | struct pipe_inode_info *pipe, |
1059 | size_t count) | |
1060 | { | |
aa563d7b | 1061 | BUG_ON(direction != READ); |
8cefc107 | 1062 | WARN_ON(pipe_full(pipe->head, pipe->tail, pipe->ring_size)); |
8cd54c1c AV |
1063 | *i = (struct iov_iter){ |
1064 | .iter_type = ITER_PIPE, | |
1065 | .data_source = false, | |
1066 | .pipe = pipe, | |
1067 | .head = pipe->head, | |
1068 | .start_head = pipe->head, | |
1069 | .iov_offset = 0, | |
1070 | .count = count | |
1071 | }; | |
241699cd AV |
1072 | } |
1073 | EXPORT_SYMBOL(iov_iter_pipe); | |
1074 | ||
7ff50620 DH |
1075 | /** |
1076 | * iov_iter_xarray - Initialise an I/O iterator to use the pages in an xarray | |
1077 | * @i: The iterator to initialise. | |
1078 | * @direction: The direction of the transfer. | |
1079 | * @xarray: The xarray to access. | |
1080 | * @start: The start file position. | |
1081 | * @count: The size of the I/O buffer in bytes. | |
1082 | * | |
1083 | * Set up an I/O iterator to either draw data out of the pages attached to an | |
1084 | * inode or to inject data into those pages. The pages *must* be prevented | |
1085 | * from evaporation, either by taking a ref on them or locking them by the | |
1086 | * caller. | |
1087 | */ | |
1088 | void iov_iter_xarray(struct iov_iter *i, unsigned int direction, | |
1089 | struct xarray *xarray, loff_t start, size_t count) | |
1090 | { | |
1091 | BUG_ON(direction & ~1); | |
8cd54c1c AV |
1092 | *i = (struct iov_iter) { |
1093 | .iter_type = ITER_XARRAY, | |
1094 | .data_source = direction, | |
1095 | .xarray = xarray, | |
1096 | .xarray_start = start, | |
1097 | .count = count, | |
1098 | .iov_offset = 0 | |
1099 | }; | |
7ff50620 DH |
1100 | } |
1101 | EXPORT_SYMBOL(iov_iter_xarray); | |
1102 | ||
9ea9ce04 DH |
1103 | /** |
1104 | * iov_iter_discard - Initialise an I/O iterator that discards data | |
1105 | * @i: The iterator to initialise. | |
1106 | * @direction: The direction of the transfer. | |
1107 | * @count: The size of the I/O buffer in bytes. | |
1108 | * | |
1109 | * Set up an I/O iterator that just discards everything that's written to it. | |
1110 | * It's only available as a READ iterator. | |
1111 | */ | |
1112 | void iov_iter_discard(struct iov_iter *i, unsigned int direction, size_t count) | |
1113 | { | |
1114 | BUG_ON(direction != READ); | |
8cd54c1c AV |
1115 | *i = (struct iov_iter){ |
1116 | .iter_type = ITER_DISCARD, | |
1117 | .data_source = false, | |
1118 | .count = count, | |
1119 | .iov_offset = 0 | |
1120 | }; | |
9ea9ce04 DH |
1121 | } |
1122 | EXPORT_SYMBOL(iov_iter_discard); | |
1123 | ||
cfa320f7 KB |
1124 | static bool iov_iter_aligned_iovec(const struct iov_iter *i, unsigned addr_mask, |
1125 | unsigned len_mask) | |
1126 | { | |
1127 | size_t size = i->count; | |
1128 | size_t skip = i->iov_offset; | |
1129 | unsigned k; | |
1130 | ||
1131 | for (k = 0; k < i->nr_segs; k++, skip = 0) { | |
1132 | size_t len = i->iov[k].iov_len - skip; | |
1133 | ||
1134 | if (len > size) | |
1135 | len = size; | |
1136 | if (len & len_mask) | |
1137 | return false; | |
1138 | if ((unsigned long)(i->iov[k].iov_base + skip) & addr_mask) | |
1139 | return false; | |
1140 | ||
1141 | size -= len; | |
1142 | if (!size) | |
1143 | break; | |
1144 | } | |
1145 | return true; | |
1146 | } | |
1147 | ||
1148 | static bool iov_iter_aligned_bvec(const struct iov_iter *i, unsigned addr_mask, | |
1149 | unsigned len_mask) | |
1150 | { | |
1151 | size_t size = i->count; | |
1152 | unsigned skip = i->iov_offset; | |
1153 | unsigned k; | |
1154 | ||
1155 | for (k = 0; k < i->nr_segs; k++, skip = 0) { | |
1156 | size_t len = i->bvec[k].bv_len - skip; | |
1157 | ||
1158 | if (len > size) | |
1159 | len = size; | |
1160 | if (len & len_mask) | |
1161 | return false; | |
1162 | if ((unsigned long)(i->bvec[k].bv_offset + skip) & addr_mask) | |
1163 | return false; | |
1164 | ||
1165 | size -= len; | |
1166 | if (!size) | |
1167 | break; | |
1168 | } | |
1169 | return true; | |
1170 | } | |
1171 | ||
1172 | /** | |
1173 | * iov_iter_is_aligned() - Check if the addresses and lengths of each segments | |
1174 | * are aligned to the parameters. | |
1175 | * | |
1176 | * @i: &struct iov_iter to restore | |
1177 | * @addr_mask: bit mask to check against the iov element's addresses | |
1178 | * @len_mask: bit mask to check against the iov element's lengths | |
1179 | * | |
1180 | * Return: false if any addresses or lengths intersect with the provided masks | |
1181 | */ | |
1182 | bool iov_iter_is_aligned(const struct iov_iter *i, unsigned addr_mask, | |
1183 | unsigned len_mask) | |
1184 | { | |
fcb14cb1 AV |
1185 | if (likely(iter_is_ubuf(i))) { |
1186 | if (i->count & len_mask) | |
1187 | return false; | |
1188 | if ((unsigned long)(i->ubuf + i->iov_offset) & addr_mask) | |
1189 | return false; | |
1190 | return true; | |
1191 | } | |
1192 | ||
cfa320f7 KB |
1193 | if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) |
1194 | return iov_iter_aligned_iovec(i, addr_mask, len_mask); | |
1195 | ||
1196 | if (iov_iter_is_bvec(i)) | |
1197 | return iov_iter_aligned_bvec(i, addr_mask, len_mask); | |
1198 | ||
1199 | if (iov_iter_is_pipe(i)) { | |
1200 | unsigned int p_mask = i->pipe->ring_size - 1; | |
1201 | size_t size = i->count; | |
1202 | ||
1203 | if (size & len_mask) | |
1204 | return false; | |
1205 | if (size && allocated(&i->pipe->bufs[i->head & p_mask])) { | |
1206 | if (i->iov_offset & addr_mask) | |
1207 | return false; | |
1208 | } | |
1209 | ||
1210 | return true; | |
1211 | } | |
1212 | ||
1213 | if (iov_iter_is_xarray(i)) { | |
1214 | if (i->count & len_mask) | |
1215 | return false; | |
1216 | if ((i->xarray_start + i->iov_offset) & addr_mask) | |
1217 | return false; | |
1218 | } | |
1219 | ||
1220 | return true; | |
1221 | } | |
1222 | EXPORT_SYMBOL_GPL(iov_iter_is_aligned); | |
1223 | ||
9221d2e3 | 1224 | static unsigned long iov_iter_alignment_iovec(const struct iov_iter *i) |
62a8067a | 1225 | { |
04a31165 AV |
1226 | unsigned long res = 0; |
1227 | size_t size = i->count; | |
9221d2e3 AV |
1228 | size_t skip = i->iov_offset; |
1229 | unsigned k; | |
1230 | ||
1231 | for (k = 0; k < i->nr_segs; k++, skip = 0) { | |
1232 | size_t len = i->iov[k].iov_len - skip; | |
1233 | if (len) { | |
1234 | res |= (unsigned long)i->iov[k].iov_base + skip; | |
1235 | if (len > size) | |
1236 | len = size; | |
1237 | res |= len; | |
1238 | size -= len; | |
1239 | if (!size) | |
1240 | break; | |
1241 | } | |
1242 | } | |
1243 | return res; | |
1244 | } | |
04a31165 | 1245 | |
9221d2e3 AV |
1246 | static unsigned long iov_iter_alignment_bvec(const struct iov_iter *i) |
1247 | { | |
1248 | unsigned res = 0; | |
1249 | size_t size = i->count; | |
1250 | unsigned skip = i->iov_offset; | |
1251 | unsigned k; | |
1252 | ||
1253 | for (k = 0; k < i->nr_segs; k++, skip = 0) { | |
1254 | size_t len = i->bvec[k].bv_len - skip; | |
1255 | res |= (unsigned long)i->bvec[k].bv_offset + skip; | |
1256 | if (len > size) | |
1257 | len = size; | |
1258 | res |= len; | |
1259 | size -= len; | |
1260 | if (!size) | |
1261 | break; | |
1262 | } | |
1263 | return res; | |
1264 | } | |
1265 | ||
1266 | unsigned long iov_iter_alignment(const struct iov_iter *i) | |
1267 | { | |
fcb14cb1 AV |
1268 | if (likely(iter_is_ubuf(i))) { |
1269 | size_t size = i->count; | |
1270 | if (size) | |
1271 | return ((unsigned long)i->ubuf + i->iov_offset) | size; | |
1272 | return 0; | |
1273 | } | |
1274 | ||
9221d2e3 AV |
1275 | /* iovec and kvec have identical layouts */ |
1276 | if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) | |
1277 | return iov_iter_alignment_iovec(i); | |
1278 | ||
1279 | if (iov_iter_is_bvec(i)) | |
1280 | return iov_iter_alignment_bvec(i); | |
1281 | ||
1282 | if (iov_iter_is_pipe(i)) { | |
e0ff126e | 1283 | unsigned int p_mask = i->pipe->ring_size - 1; |
9221d2e3 | 1284 | size_t size = i->count; |
e0ff126e | 1285 | |
8cefc107 | 1286 | if (size && i->iov_offset && allocated(&i->pipe->bufs[i->head & p_mask])) |
241699cd AV |
1287 | return size | i->iov_offset; |
1288 | return size; | |
1289 | } | |
9221d2e3 AV |
1290 | |
1291 | if (iov_iter_is_xarray(i)) | |
3d14ec1f | 1292 | return (i->xarray_start + i->iov_offset) | i->count; |
9221d2e3 AV |
1293 | |
1294 | return 0; | |
62a8067a AV |
1295 | } |
1296 | EXPORT_SYMBOL(iov_iter_alignment); | |
1297 | ||
357f435d AV |
1298 | unsigned long iov_iter_gap_alignment(const struct iov_iter *i) |
1299 | { | |
33844e66 | 1300 | unsigned long res = 0; |
610c7a71 | 1301 | unsigned long v = 0; |
357f435d | 1302 | size_t size = i->count; |
610c7a71 | 1303 | unsigned k; |
357f435d | 1304 | |
fcb14cb1 AV |
1305 | if (iter_is_ubuf(i)) |
1306 | return 0; | |
1307 | ||
610c7a71 | 1308 | if (WARN_ON(!iter_is_iovec(i))) |
241699cd | 1309 | return ~0U; |
241699cd | 1310 | |
610c7a71 AV |
1311 | for (k = 0; k < i->nr_segs; k++) { |
1312 | if (i->iov[k].iov_len) { | |
1313 | unsigned long base = (unsigned long)i->iov[k].iov_base; | |
1314 | if (v) // if not the first one | |
1315 | res |= base | v; // this start | previous end | |
1316 | v = base + i->iov[k].iov_len; | |
1317 | if (size <= i->iov[k].iov_len) | |
1318 | break; | |
1319 | size -= i->iov[k].iov_len; | |
1320 | } | |
1321 | } | |
33844e66 | 1322 | return res; |
357f435d AV |
1323 | } |
1324 | EXPORT_SYMBOL(iov_iter_gap_alignment); | |
1325 | ||
e76b6312 | 1326 | static inline ssize_t __pipe_get_pages(struct iov_iter *i, |
241699cd AV |
1327 | size_t maxsize, |
1328 | struct page **pages, | |
8cefc107 | 1329 | int iter_head, |
241699cd AV |
1330 | size_t *start) |
1331 | { | |
1332 | struct pipe_inode_info *pipe = i->pipe; | |
8cefc107 DH |
1333 | unsigned int p_mask = pipe->ring_size - 1; |
1334 | ssize_t n = push_pipe(i, maxsize, &iter_head, start); | |
241699cd AV |
1335 | if (!n) |
1336 | return -EFAULT; | |
1337 | ||
1338 | maxsize = n; | |
1339 | n += *start; | |
1689c73a | 1340 | while (n > 0) { |
8cefc107 DH |
1341 | get_page(*pages++ = pipe->bufs[iter_head & p_mask].page); |
1342 | iter_head++; | |
241699cd AV |
1343 | n -= PAGE_SIZE; |
1344 | } | |
1345 | ||
1346 | return maxsize; | |
1347 | } | |
1348 | ||
1349 | static ssize_t pipe_get_pages(struct iov_iter *i, | |
1350 | struct page **pages, size_t maxsize, unsigned maxpages, | |
1351 | size_t *start) | |
1352 | { | |
8cefc107 | 1353 | unsigned int iter_head, npages; |
241699cd | 1354 | size_t capacity; |
241699cd AV |
1355 | |
1356 | if (!sanity(i)) | |
1357 | return -EFAULT; | |
1358 | ||
8cefc107 DH |
1359 | data_start(i, &iter_head, start); |
1360 | /* Amount of free space: some of this one + all after this one */ | |
1361 | npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe); | |
1362 | capacity = min(npages, maxpages) * PAGE_SIZE - *start; | |
241699cd | 1363 | |
8cefc107 | 1364 | return __pipe_get_pages(i, min(maxsize, capacity), pages, iter_head, start); |
241699cd AV |
1365 | } |
1366 | ||
7ff50620 DH |
1367 | static ssize_t iter_xarray_populate_pages(struct page **pages, struct xarray *xa, |
1368 | pgoff_t index, unsigned int nr_pages) | |
1369 | { | |
1370 | XA_STATE(xas, xa, index); | |
1371 | struct page *page; | |
1372 | unsigned int ret = 0; | |
1373 | ||
1374 | rcu_read_lock(); | |
1375 | for (page = xas_load(&xas); page; page = xas_next(&xas)) { | |
1376 | if (xas_retry(&xas, page)) | |
1377 | continue; | |
1378 | ||
1379 | /* Has the page moved or been split? */ | |
1380 | if (unlikely(page != xas_reload(&xas))) { | |
1381 | xas_reset(&xas); | |
1382 | continue; | |
1383 | } | |
1384 | ||
1385 | pages[ret] = find_subpage(page, xas.xa_index); | |
1386 | get_page(pages[ret]); | |
1387 | if (++ret == nr_pages) | |
1388 | break; | |
1389 | } | |
1390 | rcu_read_unlock(); | |
1391 | return ret; | |
1392 | } | |
1393 | ||
1394 | static ssize_t iter_xarray_get_pages(struct iov_iter *i, | |
1395 | struct page **pages, size_t maxsize, | |
1396 | unsigned maxpages, size_t *_start_offset) | |
1397 | { | |
1398 | unsigned nr, offset; | |
1399 | pgoff_t index, count; | |
6c776766 | 1400 | size_t size = maxsize; |
7ff50620 DH |
1401 | loff_t pos; |
1402 | ||
1403 | if (!size || !maxpages) | |
1404 | return 0; | |
1405 | ||
1406 | pos = i->xarray_start + i->iov_offset; | |
1407 | index = pos >> PAGE_SHIFT; | |
1408 | offset = pos & ~PAGE_MASK; | |
1409 | *_start_offset = offset; | |
1410 | ||
1411 | count = 1; | |
1412 | if (size > PAGE_SIZE - offset) { | |
1413 | size -= PAGE_SIZE - offset; | |
1414 | count += size >> PAGE_SHIFT; | |
1415 | size &= ~PAGE_MASK; | |
1416 | if (size) | |
1417 | count++; | |
1418 | } | |
1419 | ||
1420 | if (count > maxpages) | |
1421 | count = maxpages; | |
1422 | ||
1423 | nr = iter_xarray_populate_pages(pages, i->xarray, index, count); | |
1424 | if (nr == 0) | |
1425 | return 0; | |
1426 | ||
1c27f1fc | 1427 | return min_t(size_t, nr * PAGE_SIZE - offset, maxsize); |
7ff50620 DH |
1428 | } |
1429 | ||
fcb14cb1 | 1430 | /* must be done on non-empty ITER_UBUF or ITER_IOVEC one */ |
dd45ab9d | 1431 | static unsigned long first_iovec_segment(const struct iov_iter *i, size_t *size) |
3d671ca6 AV |
1432 | { |
1433 | size_t skip; | |
1434 | long k; | |
1435 | ||
fcb14cb1 AV |
1436 | if (iter_is_ubuf(i)) |
1437 | return (unsigned long)i->ubuf + i->iov_offset; | |
1438 | ||
3d671ca6 | 1439 | for (k = 0, skip = i->iov_offset; k < i->nr_segs; k++, skip = 0) { |
3d671ca6 AV |
1440 | size_t len = i->iov[k].iov_len - skip; |
1441 | ||
1442 | if (unlikely(!len)) | |
1443 | continue; | |
59dbd7d0 AV |
1444 | if (*size > len) |
1445 | *size = len; | |
dd45ab9d | 1446 | return (unsigned long)i->iov[k].iov_base + skip; |
3d671ca6 AV |
1447 | } |
1448 | BUG(); // if it had been empty, we wouldn't get called | |
1449 | } | |
1450 | ||
1451 | /* must be done on non-empty ITER_BVEC one */ | |
1452 | static struct page *first_bvec_segment(const struct iov_iter *i, | |
59dbd7d0 | 1453 | size_t *size, size_t *start) |
3d671ca6 AV |
1454 | { |
1455 | struct page *page; | |
1456 | size_t skip = i->iov_offset, len; | |
1457 | ||
1458 | len = i->bvec->bv_len - skip; | |
59dbd7d0 AV |
1459 | if (*size > len) |
1460 | *size = len; | |
3d671ca6 AV |
1461 | skip += i->bvec->bv_offset; |
1462 | page = i->bvec->bv_page + skip / PAGE_SIZE; | |
dda8e5d1 | 1463 | *start = skip % PAGE_SIZE; |
3d671ca6 AV |
1464 | return page; |
1465 | } | |
1466 | ||
62a8067a | 1467 | ssize_t iov_iter_get_pages(struct iov_iter *i, |
2c80929c | 1468 | struct page **pages, size_t maxsize, unsigned maxpages, |
62a8067a AV |
1469 | size_t *start) |
1470 | { | |
3d671ca6 AV |
1471 | int n, res; |
1472 | ||
e5393fae AV |
1473 | if (maxsize > i->count) |
1474 | maxsize = i->count; | |
3d671ca6 AV |
1475 | if (!maxsize) |
1476 | return 0; | |
7392ed17 AV |
1477 | if (maxsize > MAX_RW_COUNT) |
1478 | maxsize = MAX_RW_COUNT; | |
e5393fae | 1479 | |
fcb14cb1 | 1480 | if (likely(user_backed_iter(i))) { |
3337ab08 | 1481 | unsigned int gup_flags = 0; |
3d671ca6 | 1482 | unsigned long addr; |
e5393fae | 1483 | |
3337ab08 AG |
1484 | if (iov_iter_rw(i) != WRITE) |
1485 | gup_flags |= FOLL_WRITE; | |
1486 | if (i->nofault) | |
1487 | gup_flags |= FOLL_NOFAULT; | |
1488 | ||
dd45ab9d AV |
1489 | addr = first_iovec_segment(i, &maxsize); |
1490 | *start = addr % PAGE_SIZE; | |
1491 | addr &= PAGE_MASK; | |
59dbd7d0 | 1492 | n = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE); |
dda8e5d1 AV |
1493 | if (n > maxpages) |
1494 | n = maxpages; | |
3337ab08 | 1495 | res = get_user_pages_fast(addr, n, gup_flags, pages); |
814a6674 | 1496 | if (unlikely(res <= 0)) |
e5393fae | 1497 | return res; |
59dbd7d0 | 1498 | return min_t(size_t, maxsize, res * PAGE_SIZE - *start); |
3d671ca6 AV |
1499 | } |
1500 | if (iov_iter_is_bvec(i)) { | |
1501 | struct page *page; | |
1502 | ||
59dbd7d0 AV |
1503 | page = first_bvec_segment(i, &maxsize, start); |
1504 | n = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE); | |
dda8e5d1 AV |
1505 | if (n > maxpages) |
1506 | n = maxpages; | |
1507 | for (int k = 0; k < n; k++) | |
3d671ca6 | 1508 | get_page(*pages++ = page++); |
59dbd7d0 | 1509 | return min_t(size_t, maxsize, n * PAGE_SIZE - *start); |
3d671ca6 AV |
1510 | } |
1511 | if (iov_iter_is_pipe(i)) | |
1512 | return pipe_get_pages(i, pages, maxsize, maxpages, start); | |
1513 | if (iov_iter_is_xarray(i)) | |
1514 | return iter_xarray_get_pages(i, pages, maxsize, maxpages, start); | |
1515 | return -EFAULT; | |
62a8067a AV |
1516 | } |
1517 | EXPORT_SYMBOL(iov_iter_get_pages); | |
1518 | ||
1b17f1f2 AV |
1519 | static struct page **get_pages_array(size_t n) |
1520 | { | |
752ade68 | 1521 | return kvmalloc_array(n, sizeof(struct page *), GFP_KERNEL); |
1b17f1f2 AV |
1522 | } |
1523 | ||
241699cd AV |
1524 | static ssize_t pipe_get_pages_alloc(struct iov_iter *i, |
1525 | struct page ***pages, size_t maxsize, | |
1526 | size_t *start) | |
1527 | { | |
1528 | struct page **p; | |
8cefc107 | 1529 | unsigned int iter_head, npages; |
d7760d63 | 1530 | ssize_t n; |
241699cd AV |
1531 | |
1532 | if (!sanity(i)) | |
1533 | return -EFAULT; | |
1534 | ||
8cefc107 DH |
1535 | data_start(i, &iter_head, start); |
1536 | /* Amount of free space: some of this one + all after this one */ | |
1537 | npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe); | |
241699cd AV |
1538 | n = npages * PAGE_SIZE - *start; |
1539 | if (maxsize > n) | |
1540 | maxsize = n; | |
1541 | else | |
1542 | npages = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE); | |
1543 | p = get_pages_array(npages); | |
1544 | if (!p) | |
1545 | return -ENOMEM; | |
8cefc107 | 1546 | n = __pipe_get_pages(i, maxsize, p, iter_head, start); |
241699cd AV |
1547 | if (n > 0) |
1548 | *pages = p; | |
1549 | else | |
1550 | kvfree(p); | |
1551 | return n; | |
1552 | } | |
1553 | ||
7ff50620 DH |
1554 | static ssize_t iter_xarray_get_pages_alloc(struct iov_iter *i, |
1555 | struct page ***pages, size_t maxsize, | |
1556 | size_t *_start_offset) | |
1557 | { | |
1558 | struct page **p; | |
1559 | unsigned nr, offset; | |
1560 | pgoff_t index, count; | |
6c776766 | 1561 | size_t size = maxsize; |
7ff50620 DH |
1562 | loff_t pos; |
1563 | ||
1564 | if (!size) | |
1565 | return 0; | |
1566 | ||
1567 | pos = i->xarray_start + i->iov_offset; | |
1568 | index = pos >> PAGE_SHIFT; | |
1569 | offset = pos & ~PAGE_MASK; | |
1570 | *_start_offset = offset; | |
1571 | ||
1572 | count = 1; | |
1573 | if (size > PAGE_SIZE - offset) { | |
1574 | size -= PAGE_SIZE - offset; | |
1575 | count += size >> PAGE_SHIFT; | |
1576 | size &= ~PAGE_MASK; | |
1577 | if (size) | |
1578 | count++; | |
1579 | } | |
1580 | ||
1581 | p = get_pages_array(count); | |
1582 | if (!p) | |
1583 | return -ENOMEM; | |
1584 | *pages = p; | |
1585 | ||
1586 | nr = iter_xarray_populate_pages(p, i->xarray, index, count); | |
1587 | if (nr == 0) | |
1588 | return 0; | |
1589 | ||
1c27f1fc | 1590 | return min_t(size_t, nr * PAGE_SIZE - offset, maxsize); |
7ff50620 DH |
1591 | } |
1592 | ||
62a8067a AV |
1593 | ssize_t iov_iter_get_pages_alloc(struct iov_iter *i, |
1594 | struct page ***pages, size_t maxsize, | |
1595 | size_t *start) | |
1596 | { | |
1b17f1f2 | 1597 | struct page **p; |
3d671ca6 | 1598 | int n, res; |
1b17f1f2 AV |
1599 | |
1600 | if (maxsize > i->count) | |
1601 | maxsize = i->count; | |
3d671ca6 AV |
1602 | if (!maxsize) |
1603 | return 0; | |
7392ed17 AV |
1604 | if (maxsize > MAX_RW_COUNT) |
1605 | maxsize = MAX_RW_COUNT; | |
1b17f1f2 | 1606 | |
fcb14cb1 | 1607 | if (likely(user_backed_iter(i))) { |
3337ab08 | 1608 | unsigned int gup_flags = 0; |
3d671ca6 | 1609 | unsigned long addr; |
1b17f1f2 | 1610 | |
3337ab08 AG |
1611 | if (iov_iter_rw(i) != WRITE) |
1612 | gup_flags |= FOLL_WRITE; | |
1613 | if (i->nofault) | |
1614 | gup_flags |= FOLL_NOFAULT; | |
1615 | ||
dd45ab9d AV |
1616 | addr = first_iovec_segment(i, &maxsize); |
1617 | *start = addr % PAGE_SIZE; | |
1618 | addr &= PAGE_MASK; | |
59dbd7d0 | 1619 | n = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE); |
1b17f1f2 AV |
1620 | p = get_pages_array(n); |
1621 | if (!p) | |
1622 | return -ENOMEM; | |
3337ab08 | 1623 | res = get_user_pages_fast(addr, n, gup_flags, p); |
814a6674 | 1624 | if (unlikely(res <= 0)) { |
1b17f1f2 | 1625 | kvfree(p); |
814a6674 | 1626 | *pages = NULL; |
1b17f1f2 AV |
1627 | return res; |
1628 | } | |
1629 | *pages = p; | |
59dbd7d0 | 1630 | return min_t(size_t, maxsize, res * PAGE_SIZE - *start); |
3d671ca6 AV |
1631 | } |
1632 | if (iov_iter_is_bvec(i)) { | |
1633 | struct page *page; | |
1634 | ||
59dbd7d0 AV |
1635 | page = first_bvec_segment(i, &maxsize, start); |
1636 | n = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE); | |
3d671ca6 | 1637 | *pages = p = get_pages_array(n); |
1b17f1f2 AV |
1638 | if (!p) |
1639 | return -ENOMEM; | |
dda8e5d1 | 1640 | for (int k = 0; k < n; k++) |
3d671ca6 | 1641 | get_page(*p++ = page++); |
59dbd7d0 | 1642 | return min_t(size_t, maxsize, n * PAGE_SIZE - *start); |
3d671ca6 AV |
1643 | } |
1644 | if (iov_iter_is_pipe(i)) | |
1645 | return pipe_get_pages_alloc(i, pages, maxsize, start); | |
1646 | if (iov_iter_is_xarray(i)) | |
1647 | return iter_xarray_get_pages_alloc(i, pages, maxsize, start); | |
1648 | return -EFAULT; | |
62a8067a AV |
1649 | } |
1650 | EXPORT_SYMBOL(iov_iter_get_pages_alloc); | |
1651 | ||
a604ec7e AV |
1652 | size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum, |
1653 | struct iov_iter *i) | |
1654 | { | |
a604ec7e | 1655 | __wsum sum, next; |
a604ec7e | 1656 | sum = *csum; |
9ea9ce04 | 1657 | if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) { |
241699cd AV |
1658 | WARN_ON(1); |
1659 | return 0; | |
1660 | } | |
7baa5099 AV |
1661 | iterate_and_advance(i, bytes, base, len, off, ({ |
1662 | next = csum_and_copy_from_user(base, addr + off, len); | |
2495bdcc | 1663 | sum = csum_block_add(sum, next, off); |
7baa5099 | 1664 | next ? 0 : len; |
a604ec7e | 1665 | }), ({ |
7baa5099 | 1666 | sum = csum_and_memcpy(addr + off, base, len, sum, off); |
a604ec7e AV |
1667 | }) |
1668 | ) | |
1669 | *csum = sum; | |
1670 | return bytes; | |
1671 | } | |
1672 | EXPORT_SYMBOL(csum_and_copy_from_iter); | |
1673 | ||
52cbd23a | 1674 | size_t csum_and_copy_to_iter(const void *addr, size_t bytes, void *_csstate, |
a604ec7e AV |
1675 | struct iov_iter *i) |
1676 | { | |
52cbd23a | 1677 | struct csum_state *csstate = _csstate; |
a604ec7e | 1678 | __wsum sum, next; |
78e1f386 | 1679 | |
78e1f386 | 1680 | if (unlikely(iov_iter_is_discard(i))) { |
241699cd AV |
1681 | WARN_ON(1); /* for now */ |
1682 | return 0; | |
1683 | } | |
6852df12 AV |
1684 | |
1685 | sum = csum_shift(csstate->csum, csstate->off); | |
1686 | if (unlikely(iov_iter_is_pipe(i))) | |
1687 | bytes = csum_and_copy_to_pipe_iter(addr, bytes, i, &sum); | |
1688 | else iterate_and_advance(i, bytes, base, len, off, ({ | |
7baa5099 | 1689 | next = csum_and_copy_to_user(addr + off, base, len); |
2495bdcc | 1690 | sum = csum_block_add(sum, next, off); |
7baa5099 | 1691 | next ? 0 : len; |
a604ec7e | 1692 | }), ({ |
7baa5099 | 1693 | sum = csum_and_memcpy(base, addr + off, len, sum, off); |
a604ec7e AV |
1694 | }) |
1695 | ) | |
594e450b AV |
1696 | csstate->csum = csum_shift(sum, csstate->off); |
1697 | csstate->off += bytes; | |
a604ec7e AV |
1698 | return bytes; |
1699 | } | |
1700 | EXPORT_SYMBOL(csum_and_copy_to_iter); | |
1701 | ||
d05f4435 SG |
1702 | size_t hash_and_copy_to_iter(const void *addr, size_t bytes, void *hashp, |
1703 | struct iov_iter *i) | |
1704 | { | |
7999096f | 1705 | #ifdef CONFIG_CRYPTO_HASH |
d05f4435 SG |
1706 | struct ahash_request *hash = hashp; |
1707 | struct scatterlist sg; | |
1708 | size_t copied; | |
1709 | ||
1710 | copied = copy_to_iter(addr, bytes, i); | |
1711 | sg_init_one(&sg, addr, copied); | |
1712 | ahash_request_set_crypt(hash, &sg, NULL, copied); | |
1713 | crypto_ahash_update(hash); | |
1714 | return copied; | |
27fad74a Y |
1715 | #else |
1716 | return 0; | |
1717 | #endif | |
d05f4435 SG |
1718 | } |
1719 | EXPORT_SYMBOL(hash_and_copy_to_iter); | |
1720 | ||
66531c65 | 1721 | static int iov_npages(const struct iov_iter *i, int maxpages) |
62a8067a | 1722 | { |
66531c65 AV |
1723 | size_t skip = i->iov_offset, size = i->count; |
1724 | const struct iovec *p; | |
e0f2dc40 AV |
1725 | int npages = 0; |
1726 | ||
66531c65 AV |
1727 | for (p = i->iov; size; skip = 0, p++) { |
1728 | unsigned offs = offset_in_page(p->iov_base + skip); | |
1729 | size_t len = min(p->iov_len - skip, size); | |
e0f2dc40 | 1730 | |
66531c65 AV |
1731 | if (len) { |
1732 | size -= len; | |
1733 | npages += DIV_ROUND_UP(offs + len, PAGE_SIZE); | |
1734 | if (unlikely(npages > maxpages)) | |
1735 | return maxpages; | |
1736 | } | |
1737 | } | |
1738 | return npages; | |
1739 | } | |
1740 | ||
1741 | static int bvec_npages(const struct iov_iter *i, int maxpages) | |
1742 | { | |
1743 | size_t skip = i->iov_offset, size = i->count; | |
1744 | const struct bio_vec *p; | |
1745 | int npages = 0; | |
1746 | ||
1747 | for (p = i->bvec; size; skip = 0, p++) { | |
1748 | unsigned offs = (p->bv_offset + skip) % PAGE_SIZE; | |
1749 | size_t len = min(p->bv_len - skip, size); | |
1750 | ||
1751 | size -= len; | |
1752 | npages += DIV_ROUND_UP(offs + len, PAGE_SIZE); | |
1753 | if (unlikely(npages > maxpages)) | |
1754 | return maxpages; | |
1755 | } | |
1756 | return npages; | |
1757 | } | |
1758 | ||
1759 | int iov_iter_npages(const struct iov_iter *i, int maxpages) | |
1760 | { | |
1761 | if (unlikely(!i->count)) | |
1762 | return 0; | |
fcb14cb1 AV |
1763 | if (likely(iter_is_ubuf(i))) { |
1764 | unsigned offs = offset_in_page(i->ubuf + i->iov_offset); | |
1765 | int npages = DIV_ROUND_UP(offs + i->count, PAGE_SIZE); | |
1766 | return min(npages, maxpages); | |
1767 | } | |
66531c65 AV |
1768 | /* iovec and kvec have identical layouts */ |
1769 | if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) | |
1770 | return iov_npages(i, maxpages); | |
1771 | if (iov_iter_is_bvec(i)) | |
1772 | return bvec_npages(i, maxpages); | |
1773 | if (iov_iter_is_pipe(i)) { | |
8cefc107 | 1774 | unsigned int iter_head; |
66531c65 | 1775 | int npages; |
241699cd | 1776 | size_t off; |
241699cd AV |
1777 | |
1778 | if (!sanity(i)) | |
1779 | return 0; | |
1780 | ||
8cefc107 | 1781 | data_start(i, &iter_head, &off); |
241699cd | 1782 | /* some of this one + all after this one */ |
66531c65 AV |
1783 | npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe); |
1784 | return min(npages, maxpages); | |
1785 | } | |
1786 | if (iov_iter_is_xarray(i)) { | |
e4f8df86 AV |
1787 | unsigned offset = (i->xarray_start + i->iov_offset) % PAGE_SIZE; |
1788 | int npages = DIV_ROUND_UP(offset + i->count, PAGE_SIZE); | |
66531c65 AV |
1789 | return min(npages, maxpages); |
1790 | } | |
1791 | return 0; | |
62a8067a | 1792 | } |
f67da30c | 1793 | EXPORT_SYMBOL(iov_iter_npages); |
4b8164b9 AV |
1794 | |
1795 | const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags) | |
1796 | { | |
1797 | *new = *old; | |
00e23707 | 1798 | if (unlikely(iov_iter_is_pipe(new))) { |
241699cd AV |
1799 | WARN_ON(1); |
1800 | return NULL; | |
1801 | } | |
00e23707 | 1802 | if (iov_iter_is_bvec(new)) |
4b8164b9 AV |
1803 | return new->bvec = kmemdup(new->bvec, |
1804 | new->nr_segs * sizeof(struct bio_vec), | |
1805 | flags); | |
fcb14cb1 | 1806 | else if (iov_iter_is_kvec(new) || iter_is_iovec(new)) |
4b8164b9 AV |
1807 | /* iovec and kvec have identical layout */ |
1808 | return new->iov = kmemdup(new->iov, | |
1809 | new->nr_segs * sizeof(struct iovec), | |
1810 | flags); | |
fcb14cb1 | 1811 | return NULL; |
4b8164b9 AV |
1812 | } |
1813 | EXPORT_SYMBOL(dup_iter); | |
bc917be8 | 1814 | |
bfdc5970 CH |
1815 | static int copy_compat_iovec_from_user(struct iovec *iov, |
1816 | const struct iovec __user *uvec, unsigned long nr_segs) | |
1817 | { | |
1818 | const struct compat_iovec __user *uiov = | |
1819 | (const struct compat_iovec __user *)uvec; | |
1820 | int ret = -EFAULT, i; | |
1821 | ||
a959a978 | 1822 | if (!user_access_begin(uiov, nr_segs * sizeof(*uiov))) |
bfdc5970 CH |
1823 | return -EFAULT; |
1824 | ||
1825 | for (i = 0; i < nr_segs; i++) { | |
1826 | compat_uptr_t buf; | |
1827 | compat_ssize_t len; | |
1828 | ||
1829 | unsafe_get_user(len, &uiov[i].iov_len, uaccess_end); | |
1830 | unsafe_get_user(buf, &uiov[i].iov_base, uaccess_end); | |
1831 | ||
1832 | /* check for compat_size_t not fitting in compat_ssize_t .. */ | |
1833 | if (len < 0) { | |
1834 | ret = -EINVAL; | |
1835 | goto uaccess_end; | |
1836 | } | |
1837 | iov[i].iov_base = compat_ptr(buf); | |
1838 | iov[i].iov_len = len; | |
1839 | } | |
1840 | ||
1841 | ret = 0; | |
1842 | uaccess_end: | |
1843 | user_access_end(); | |
1844 | return ret; | |
1845 | } | |
1846 | ||
1847 | static int copy_iovec_from_user(struct iovec *iov, | |
1848 | const struct iovec __user *uvec, unsigned long nr_segs) | |
fb041b59 DL |
1849 | { |
1850 | unsigned long seg; | |
fb041b59 | 1851 | |
bfdc5970 CH |
1852 | if (copy_from_user(iov, uvec, nr_segs * sizeof(*uvec))) |
1853 | return -EFAULT; | |
1854 | for (seg = 0; seg < nr_segs; seg++) { | |
1855 | if ((ssize_t)iov[seg].iov_len < 0) | |
1856 | return -EINVAL; | |
fb041b59 DL |
1857 | } |
1858 | ||
bfdc5970 CH |
1859 | return 0; |
1860 | } | |
1861 | ||
1862 | struct iovec *iovec_from_user(const struct iovec __user *uvec, | |
1863 | unsigned long nr_segs, unsigned long fast_segs, | |
1864 | struct iovec *fast_iov, bool compat) | |
1865 | { | |
1866 | struct iovec *iov = fast_iov; | |
1867 | int ret; | |
1868 | ||
fb041b59 | 1869 | /* |
bfdc5970 CH |
1870 | * SuS says "The readv() function *may* fail if the iovcnt argument was |
1871 | * less than or equal to 0, or greater than {IOV_MAX}. Linux has | |
1872 | * traditionally returned zero for zero segments, so... | |
fb041b59 | 1873 | */ |
bfdc5970 CH |
1874 | if (nr_segs == 0) |
1875 | return iov; | |
1876 | if (nr_segs > UIO_MAXIOV) | |
1877 | return ERR_PTR(-EINVAL); | |
fb041b59 DL |
1878 | if (nr_segs > fast_segs) { |
1879 | iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL); | |
bfdc5970 CH |
1880 | if (!iov) |
1881 | return ERR_PTR(-ENOMEM); | |
fb041b59 | 1882 | } |
bfdc5970 CH |
1883 | |
1884 | if (compat) | |
1885 | ret = copy_compat_iovec_from_user(iov, uvec, nr_segs); | |
1886 | else | |
1887 | ret = copy_iovec_from_user(iov, uvec, nr_segs); | |
1888 | if (ret) { | |
1889 | if (iov != fast_iov) | |
1890 | kfree(iov); | |
1891 | return ERR_PTR(ret); | |
1892 | } | |
1893 | ||
1894 | return iov; | |
1895 | } | |
1896 | ||
1897 | ssize_t __import_iovec(int type, const struct iovec __user *uvec, | |
1898 | unsigned nr_segs, unsigned fast_segs, struct iovec **iovp, | |
1899 | struct iov_iter *i, bool compat) | |
1900 | { | |
1901 | ssize_t total_len = 0; | |
1902 | unsigned long seg; | |
1903 | struct iovec *iov; | |
1904 | ||
1905 | iov = iovec_from_user(uvec, nr_segs, fast_segs, *iovp, compat); | |
1906 | if (IS_ERR(iov)) { | |
1907 | *iovp = NULL; | |
1908 | return PTR_ERR(iov); | |
fb041b59 DL |
1909 | } |
1910 | ||
1911 | /* | |
bfdc5970 CH |
1912 | * According to the Single Unix Specification we should return EINVAL if |
1913 | * an element length is < 0 when cast to ssize_t or if the total length | |
1914 | * would overflow the ssize_t return value of the system call. | |
fb041b59 DL |
1915 | * |
1916 | * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the | |
1917 | * overflow case. | |
1918 | */ | |
fb041b59 | 1919 | for (seg = 0; seg < nr_segs; seg++) { |
fb041b59 DL |
1920 | ssize_t len = (ssize_t)iov[seg].iov_len; |
1921 | ||
bfdc5970 CH |
1922 | if (!access_ok(iov[seg].iov_base, len)) { |
1923 | if (iov != *iovp) | |
1924 | kfree(iov); | |
1925 | *iovp = NULL; | |
1926 | return -EFAULT; | |
fb041b59 | 1927 | } |
bfdc5970 CH |
1928 | |
1929 | if (len > MAX_RW_COUNT - total_len) { | |
1930 | len = MAX_RW_COUNT - total_len; | |
fb041b59 DL |
1931 | iov[seg].iov_len = len; |
1932 | } | |
bfdc5970 | 1933 | total_len += len; |
fb041b59 | 1934 | } |
bfdc5970 CH |
1935 | |
1936 | iov_iter_init(i, type, iov, nr_segs, total_len); | |
1937 | if (iov == *iovp) | |
1938 | *iovp = NULL; | |
1939 | else | |
1940 | *iovp = iov; | |
1941 | return total_len; | |
fb041b59 DL |
1942 | } |
1943 | ||
ffecee4f VN |
1944 | /** |
1945 | * import_iovec() - Copy an array of &struct iovec from userspace | |
1946 | * into the kernel, check that it is valid, and initialize a new | |
1947 | * &struct iov_iter iterator to access it. | |
1948 | * | |
1949 | * @type: One of %READ or %WRITE. | |
bfdc5970 | 1950 | * @uvec: Pointer to the userspace array. |
ffecee4f VN |
1951 | * @nr_segs: Number of elements in userspace array. |
1952 | * @fast_segs: Number of elements in @iov. | |
bfdc5970 | 1953 | * @iovp: (input and output parameter) Pointer to pointer to (usually small |
ffecee4f VN |
1954 | * on-stack) kernel array. |
1955 | * @i: Pointer to iterator that will be initialized on success. | |
1956 | * | |
1957 | * If the array pointed to by *@iov is large enough to hold all @nr_segs, | |
1958 | * then this function places %NULL in *@iov on return. Otherwise, a new | |
1959 | * array will be allocated and the result placed in *@iov. This means that | |
1960 | * the caller may call kfree() on *@iov regardless of whether the small | |
1961 | * on-stack array was used or not (and regardless of whether this function | |
1962 | * returns an error or not). | |
1963 | * | |
87e5e6da | 1964 | * Return: Negative error code on error, bytes imported on success |
ffecee4f | 1965 | */ |
bfdc5970 | 1966 | ssize_t import_iovec(int type, const struct iovec __user *uvec, |
bc917be8 | 1967 | unsigned nr_segs, unsigned fast_segs, |
bfdc5970 | 1968 | struct iovec **iovp, struct iov_iter *i) |
bc917be8 | 1969 | { |
89cd35c5 CH |
1970 | return __import_iovec(type, uvec, nr_segs, fast_segs, iovp, i, |
1971 | in_compat_syscall()); | |
bc917be8 AV |
1972 | } |
1973 | EXPORT_SYMBOL(import_iovec); | |
1974 | ||
bc917be8 AV |
1975 | int import_single_range(int rw, void __user *buf, size_t len, |
1976 | struct iovec *iov, struct iov_iter *i) | |
1977 | { | |
1978 | if (len > MAX_RW_COUNT) | |
1979 | len = MAX_RW_COUNT; | |
96d4f267 | 1980 | if (unlikely(!access_ok(buf, len))) |
bc917be8 AV |
1981 | return -EFAULT; |
1982 | ||
1983 | iov->iov_base = buf; | |
1984 | iov->iov_len = len; | |
1985 | iov_iter_init(i, rw, iov, 1, len); | |
1986 | return 0; | |
1987 | } | |
e1267585 | 1988 | EXPORT_SYMBOL(import_single_range); |
8fb0f47a JA |
1989 | |
1990 | /** | |
1991 | * iov_iter_restore() - Restore a &struct iov_iter to the same state as when | |
1992 | * iov_iter_save_state() was called. | |
1993 | * | |
1994 | * @i: &struct iov_iter to restore | |
1995 | * @state: state to restore from | |
1996 | * | |
1997 | * Used after iov_iter_save_state() to bring restore @i, if operations may | |
1998 | * have advanced it. | |
1999 | * | |
2000 | * Note: only works on ITER_IOVEC, ITER_BVEC, and ITER_KVEC | |
2001 | */ | |
2002 | void iov_iter_restore(struct iov_iter *i, struct iov_iter_state *state) | |
2003 | { | |
2004 | if (WARN_ON_ONCE(!iov_iter_is_bvec(i) && !iter_is_iovec(i)) && | |
fcb14cb1 | 2005 | !iov_iter_is_kvec(i) && !iter_is_ubuf(i)) |
8fb0f47a JA |
2006 | return; |
2007 | i->iov_offset = state->iov_offset; | |
2008 | i->count = state->count; | |
fcb14cb1 AV |
2009 | if (iter_is_ubuf(i)) |
2010 | return; | |
8fb0f47a JA |
2011 | /* |
2012 | * For the *vec iters, nr_segs + iov is constant - if we increment | |
2013 | * the vec, then we also decrement the nr_segs count. Hence we don't | |
2014 | * need to track both of these, just one is enough and we can deduct | |
2015 | * the other from that. ITER_KVEC and ITER_IOVEC are the same struct | |
2016 | * size, so we can just increment the iov pointer as they are unionzed. | |
2017 | * ITER_BVEC _may_ be the same size on some archs, but on others it is | |
2018 | * not. Be safe and handle it separately. | |
2019 | */ | |
2020 | BUILD_BUG_ON(sizeof(struct iovec) != sizeof(struct kvec)); | |
2021 | if (iov_iter_is_bvec(i)) | |
2022 | i->bvec -= state->nr_segs - i->nr_segs; | |
2023 | else | |
2024 | i->iov -= state->nr_segs - i->nr_segs; | |
2025 | i->nr_segs = state->nr_segs; | |
2026 | } |