]>
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 | ||
5c67aa90 | 19 | /* covers iovec and kvec alike */ |
a6e4ec7b | 20 | #define iterate_iovec(i, n, base, len, off, __p, STEP) { \ |
7baa5099 | 21 | size_t off = 0; \ |
a6e4ec7b | 22 | size_t skip = i->iov_offset; \ |
7a1bcb5d | 23 | do { \ |
7baa5099 AV |
24 | len = min(n, __p->iov_len - skip); \ |
25 | if (likely(len)) { \ | |
26 | base = __p->iov_base + skip; \ | |
27 | len -= (STEP); \ | |
28 | off += len; \ | |
29 | skip += len; \ | |
30 | n -= len; \ | |
7a1bcb5d AV |
31 | if (skip < __p->iov_len) \ |
32 | break; \ | |
33 | } \ | |
34 | __p++; \ | |
35 | skip = 0; \ | |
36 | } while (n); \ | |
a6e4ec7b | 37 | i->iov_offset = skip; \ |
7baa5099 | 38 | n = off; \ |
04a31165 AV |
39 | } |
40 | ||
a6e4ec7b | 41 | #define iterate_bvec(i, n, base, len, off, p, STEP) { \ |
7baa5099 | 42 | size_t off = 0; \ |
a6e4ec7b | 43 | unsigned skip = i->iov_offset; \ |
7491a2bf AV |
44 | while (n) { \ |
45 | unsigned offset = p->bv_offset + skip; \ | |
1b4fb5ff | 46 | unsigned left; \ |
21b56c84 AV |
47 | void *kaddr = kmap_local_page(p->bv_page + \ |
48 | offset / PAGE_SIZE); \ | |
7baa5099 | 49 | base = kaddr + offset % PAGE_SIZE; \ |
a6e4ec7b | 50 | len = min(min(n, (size_t)(p->bv_len - skip)), \ |
7491a2bf | 51 | (size_t)(PAGE_SIZE - offset % PAGE_SIZE)); \ |
1b4fb5ff | 52 | left = (STEP); \ |
21b56c84 | 53 | kunmap_local(kaddr); \ |
7baa5099 AV |
54 | len -= left; \ |
55 | off += len; \ | |
56 | skip += len; \ | |
7491a2bf AV |
57 | if (skip == p->bv_len) { \ |
58 | skip = 0; \ | |
59 | p++; \ | |
60 | } \ | |
7baa5099 | 61 | n -= len; \ |
1b4fb5ff AV |
62 | if (left) \ |
63 | break; \ | |
7491a2bf | 64 | } \ |
a6e4ec7b | 65 | i->iov_offset = skip; \ |
7baa5099 | 66 | n = off; \ |
04a31165 AV |
67 | } |
68 | ||
a6e4ec7b | 69 | #define iterate_xarray(i, n, base, len, __off, STEP) { \ |
1b4fb5ff | 70 | __label__ __out; \ |
622838f3 | 71 | size_t __off = 0; \ |
7ff50620 | 72 | struct page *head = NULL; \ |
a6e4ec7b | 73 | loff_t start = i->xarray_start + i->iov_offset; \ |
4b179e9a AV |
74 | unsigned offset = start % PAGE_SIZE; \ |
75 | pgoff_t index = start / PAGE_SIZE; \ | |
7ff50620 DH |
76 | int j; \ |
77 | \ | |
78 | XA_STATE(xas, i->xarray, index); \ | |
79 | \ | |
7baa5099 AV |
80 | rcu_read_lock(); \ |
81 | xas_for_each(&xas, head, ULONG_MAX) { \ | |
82 | unsigned left; \ | |
83 | if (xas_retry(&xas, head)) \ | |
84 | continue; \ | |
85 | if (WARN_ON(xa_is_value(head))) \ | |
86 | break; \ | |
87 | if (WARN_ON(PageHuge(head))) \ | |
88 | break; \ | |
7ff50620 | 89 | for (j = (head->index < index) ? index - head->index : 0; \ |
7baa5099 | 90 | j < thp_nr_pages(head); j++) { \ |
21b56c84 | 91 | void *kaddr = kmap_local_page(head + j); \ |
7baa5099 AV |
92 | base = kaddr + offset; \ |
93 | len = PAGE_SIZE - offset; \ | |
94 | len = min(n, len); \ | |
95 | left = (STEP); \ | |
96 | kunmap_local(kaddr); \ | |
97 | len -= left; \ | |
98 | __off += len; \ | |
99 | n -= len; \ | |
100 | if (left || n == 0) \ | |
101 | goto __out; \ | |
4b179e9a | 102 | offset = 0; \ |
7baa5099 | 103 | } \ |
7ff50620 | 104 | } \ |
1b4fb5ff | 105 | __out: \ |
7ff50620 | 106 | rcu_read_unlock(); \ |
a6e4ec7b | 107 | i->iov_offset += __off; \ |
622838f3 | 108 | n = __off; \ |
7ff50620 DH |
109 | } |
110 | ||
7baa5099 | 111 | #define __iterate_and_advance(i, n, base, len, off, I, K) { \ |
dd254f5a AV |
112 | if (unlikely(i->count < n)) \ |
113 | n = i->count; \ | |
f5da8354 | 114 | if (likely(n)) { \ |
28f38db7 | 115 | if (likely(iter_is_iovec(i))) { \ |
5c67aa90 | 116 | const struct iovec *iov = i->iov; \ |
7baa5099 AV |
117 | void __user *base; \ |
118 | size_t len; \ | |
119 | iterate_iovec(i, n, base, len, off, \ | |
a6e4ec7b | 120 | iov, (I)) \ |
28f38db7 AV |
121 | i->nr_segs -= iov - i->iov; \ |
122 | i->iov = iov; \ | |
123 | } else if (iov_iter_is_bvec(i)) { \ | |
1bdc76ae | 124 | const struct bio_vec *bvec = i->bvec; \ |
7baa5099 AV |
125 | void *base; \ |
126 | size_t len; \ | |
127 | iterate_bvec(i, n, base, len, off, \ | |
a6e4ec7b | 128 | bvec, (K)) \ |
7491a2bf AV |
129 | i->nr_segs -= bvec - i->bvec; \ |
130 | i->bvec = bvec; \ | |
28f38db7 | 131 | } else if (iov_iter_is_kvec(i)) { \ |
5c67aa90 | 132 | const struct kvec *kvec = i->kvec; \ |
7baa5099 AV |
133 | void *base; \ |
134 | size_t len; \ | |
135 | iterate_iovec(i, n, base, len, off, \ | |
a6e4ec7b | 136 | kvec, (K)) \ |
dd254f5a AV |
137 | i->nr_segs -= kvec - i->kvec; \ |
138 | i->kvec = kvec; \ | |
28f38db7 | 139 | } else if (iov_iter_is_xarray(i)) { \ |
7baa5099 AV |
140 | void *base; \ |
141 | size_t len; \ | |
142 | iterate_xarray(i, n, base, len, off, \ | |
a6e4ec7b | 143 | (K)) \ |
7ce2a91e | 144 | } \ |
dd254f5a | 145 | i->count -= n; \ |
7ce2a91e | 146 | } \ |
7ce2a91e | 147 | } |
7baa5099 AV |
148 | #define iterate_and_advance(i, n, base, len, off, I, K) \ |
149 | __iterate_and_advance(i, n, base, len, off, I, ((void)(K),0)) | |
7ce2a91e | 150 | |
09fc68dc AV |
151 | static int copyout(void __user *to, const void *from, size_t n) |
152 | { | |
4d0e9df5 AL |
153 | if (should_fail_usercopy()) |
154 | return n; | |
96d4f267 | 155 | if (access_ok(to, n)) { |
d0ef4c36 | 156 | instrument_copy_to_user(to, from, n); |
09fc68dc AV |
157 | n = raw_copy_to_user(to, from, n); |
158 | } | |
159 | return n; | |
160 | } | |
161 | ||
162 | static int copyin(void *to, const void __user *from, size_t n) | |
163 | { | |
4d0e9df5 AL |
164 | if (should_fail_usercopy()) |
165 | return n; | |
96d4f267 | 166 | if (access_ok(from, n)) { |
d0ef4c36 | 167 | instrument_copy_from_user(to, from, n); |
09fc68dc AV |
168 | n = raw_copy_from_user(to, from, n); |
169 | } | |
170 | return n; | |
171 | } | |
172 | ||
62a8067a | 173 | static size_t copy_page_to_iter_iovec(struct page *page, size_t offset, size_t bytes, |
4f18cd31 AV |
174 | struct iov_iter *i) |
175 | { | |
176 | size_t skip, copy, left, wanted; | |
177 | const struct iovec *iov; | |
178 | char __user *buf; | |
179 | void *kaddr, *from; | |
180 | ||
181 | if (unlikely(bytes > i->count)) | |
182 | bytes = i->count; | |
183 | ||
184 | if (unlikely(!bytes)) | |
185 | return 0; | |
186 | ||
09fc68dc | 187 | might_fault(); |
4f18cd31 AV |
188 | wanted = bytes; |
189 | iov = i->iov; | |
190 | skip = i->iov_offset; | |
191 | buf = iov->iov_base + skip; | |
192 | copy = min(bytes, iov->iov_len - skip); | |
193 | ||
bb523b40 | 194 | if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_writeable(buf, copy)) { |
4f18cd31 AV |
195 | kaddr = kmap_atomic(page); |
196 | from = kaddr + offset; | |
197 | ||
198 | /* first chunk, usually the only one */ | |
09fc68dc | 199 | left = copyout(buf, from, copy); |
4f18cd31 AV |
200 | copy -= left; |
201 | skip += copy; | |
202 | from += copy; | |
203 | bytes -= copy; | |
204 | ||
205 | while (unlikely(!left && bytes)) { | |
206 | iov++; | |
207 | buf = iov->iov_base; | |
208 | copy = min(bytes, iov->iov_len); | |
09fc68dc | 209 | left = copyout(buf, from, copy); |
4f18cd31 AV |
210 | copy -= left; |
211 | skip = copy; | |
212 | from += copy; | |
213 | bytes -= copy; | |
214 | } | |
215 | if (likely(!bytes)) { | |
216 | kunmap_atomic(kaddr); | |
217 | goto done; | |
218 | } | |
219 | offset = from - kaddr; | |
220 | buf += copy; | |
221 | kunmap_atomic(kaddr); | |
222 | copy = min(bytes, iov->iov_len - skip); | |
223 | } | |
224 | /* Too bad - revert to non-atomic kmap */ | |
3fa6c507 | 225 | |
4f18cd31 AV |
226 | kaddr = kmap(page); |
227 | from = kaddr + offset; | |
09fc68dc | 228 | left = copyout(buf, from, copy); |
4f18cd31 AV |
229 | copy -= left; |
230 | skip += copy; | |
231 | from += copy; | |
232 | bytes -= copy; | |
233 | while (unlikely(!left && bytes)) { | |
234 | iov++; | |
235 | buf = iov->iov_base; | |
236 | copy = min(bytes, iov->iov_len); | |
09fc68dc | 237 | left = copyout(buf, from, copy); |
4f18cd31 AV |
238 | copy -= left; |
239 | skip = copy; | |
240 | from += copy; | |
241 | bytes -= copy; | |
242 | } | |
243 | kunmap(page); | |
3fa6c507 | 244 | |
4f18cd31 | 245 | done: |
81055e58 AV |
246 | if (skip == iov->iov_len) { |
247 | iov++; | |
248 | skip = 0; | |
249 | } | |
4f18cd31 AV |
250 | i->count -= wanted - bytes; |
251 | i->nr_segs -= iov - i->iov; | |
252 | i->iov = iov; | |
253 | i->iov_offset = skip; | |
254 | return wanted - bytes; | |
255 | } | |
4f18cd31 | 256 | |
62a8067a | 257 | static size_t copy_page_from_iter_iovec(struct page *page, size_t offset, size_t bytes, |
f0d1bec9 AV |
258 | struct iov_iter *i) |
259 | { | |
260 | size_t skip, copy, left, wanted; | |
261 | const struct iovec *iov; | |
262 | char __user *buf; | |
263 | void *kaddr, *to; | |
264 | ||
265 | if (unlikely(bytes > i->count)) | |
266 | bytes = i->count; | |
267 | ||
268 | if (unlikely(!bytes)) | |
269 | return 0; | |
270 | ||
09fc68dc | 271 | might_fault(); |
f0d1bec9 AV |
272 | wanted = bytes; |
273 | iov = i->iov; | |
274 | skip = i->iov_offset; | |
275 | buf = iov->iov_base + skip; | |
276 | copy = min(bytes, iov->iov_len - skip); | |
277 | ||
bb523b40 | 278 | if (IS_ENABLED(CONFIG_HIGHMEM) && !fault_in_readable(buf, copy)) { |
f0d1bec9 AV |
279 | kaddr = kmap_atomic(page); |
280 | to = kaddr + offset; | |
281 | ||
282 | /* first chunk, usually the only one */ | |
09fc68dc | 283 | left = copyin(to, buf, copy); |
f0d1bec9 AV |
284 | copy -= left; |
285 | skip += copy; | |
286 | to += copy; | |
287 | bytes -= copy; | |
288 | ||
289 | while (unlikely(!left && bytes)) { | |
290 | iov++; | |
291 | buf = iov->iov_base; | |
292 | copy = min(bytes, iov->iov_len); | |
09fc68dc | 293 | left = copyin(to, buf, copy); |
f0d1bec9 AV |
294 | copy -= left; |
295 | skip = copy; | |
296 | to += copy; | |
297 | bytes -= copy; | |
298 | } | |
299 | if (likely(!bytes)) { | |
300 | kunmap_atomic(kaddr); | |
301 | goto done; | |
302 | } | |
303 | offset = to - kaddr; | |
304 | buf += copy; | |
305 | kunmap_atomic(kaddr); | |
306 | copy = min(bytes, iov->iov_len - skip); | |
307 | } | |
308 | /* Too bad - revert to non-atomic kmap */ | |
3fa6c507 | 309 | |
f0d1bec9 AV |
310 | kaddr = kmap(page); |
311 | to = kaddr + offset; | |
09fc68dc | 312 | left = copyin(to, buf, copy); |
f0d1bec9 AV |
313 | copy -= left; |
314 | skip += copy; | |
315 | to += copy; | |
316 | bytes -= copy; | |
317 | while (unlikely(!left && bytes)) { | |
318 | iov++; | |
319 | buf = iov->iov_base; | |
320 | copy = min(bytes, iov->iov_len); | |
09fc68dc | 321 | left = copyin(to, buf, copy); |
f0d1bec9 AV |
322 | copy -= left; |
323 | skip = copy; | |
324 | to += copy; | |
325 | bytes -= copy; | |
326 | } | |
327 | kunmap(page); | |
3fa6c507 | 328 | |
f0d1bec9 | 329 | done: |
81055e58 AV |
330 | if (skip == iov->iov_len) { |
331 | iov++; | |
332 | skip = 0; | |
333 | } | |
f0d1bec9 AV |
334 | i->count -= wanted - bytes; |
335 | i->nr_segs -= iov - i->iov; | |
336 | i->iov = iov; | |
337 | i->iov_offset = skip; | |
338 | return wanted - bytes; | |
339 | } | |
f0d1bec9 | 340 | |
241699cd AV |
341 | #ifdef PIPE_PARANOIA |
342 | static bool sanity(const struct iov_iter *i) | |
343 | { | |
344 | struct pipe_inode_info *pipe = i->pipe; | |
8cefc107 DH |
345 | unsigned int p_head = pipe->head; |
346 | unsigned int p_tail = pipe->tail; | |
347 | unsigned int p_mask = pipe->ring_size - 1; | |
348 | unsigned int p_occupancy = pipe_occupancy(p_head, p_tail); | |
349 | unsigned int i_head = i->head; | |
350 | unsigned int idx; | |
351 | ||
241699cd AV |
352 | if (i->iov_offset) { |
353 | struct pipe_buffer *p; | |
8cefc107 | 354 | if (unlikely(p_occupancy == 0)) |
241699cd | 355 | goto Bad; // pipe must be non-empty |
8cefc107 | 356 | if (unlikely(i_head != p_head - 1)) |
241699cd AV |
357 | goto Bad; // must be at the last buffer... |
358 | ||
8cefc107 | 359 | p = &pipe->bufs[i_head & p_mask]; |
241699cd AV |
360 | if (unlikely(p->offset + p->len != i->iov_offset)) |
361 | goto Bad; // ... at the end of segment | |
362 | } else { | |
8cefc107 | 363 | if (i_head != p_head) |
241699cd AV |
364 | goto Bad; // must be right after the last buffer |
365 | } | |
366 | return true; | |
367 | Bad: | |
8cefc107 DH |
368 | printk(KERN_ERR "idx = %d, offset = %zd\n", i_head, i->iov_offset); |
369 | printk(KERN_ERR "head = %d, tail = %d, buffers = %d\n", | |
370 | p_head, p_tail, pipe->ring_size); | |
371 | for (idx = 0; idx < pipe->ring_size; idx++) | |
241699cd AV |
372 | printk(KERN_ERR "[%p %p %d %d]\n", |
373 | pipe->bufs[idx].ops, | |
374 | pipe->bufs[idx].page, | |
375 | pipe->bufs[idx].offset, | |
376 | pipe->bufs[idx].len); | |
377 | WARN_ON(1); | |
378 | return false; | |
379 | } | |
380 | #else | |
381 | #define sanity(i) true | |
382 | #endif | |
383 | ||
241699cd AV |
384 | static size_t copy_page_to_iter_pipe(struct page *page, size_t offset, size_t bytes, |
385 | struct iov_iter *i) | |
386 | { | |
387 | struct pipe_inode_info *pipe = i->pipe; | |
388 | struct pipe_buffer *buf; | |
8cefc107 DH |
389 | unsigned int p_tail = pipe->tail; |
390 | unsigned int p_mask = pipe->ring_size - 1; | |
391 | unsigned int i_head = i->head; | |
241699cd | 392 | size_t off; |
241699cd AV |
393 | |
394 | if (unlikely(bytes > i->count)) | |
395 | bytes = i->count; | |
396 | ||
397 | if (unlikely(!bytes)) | |
398 | return 0; | |
399 | ||
400 | if (!sanity(i)) | |
401 | return 0; | |
402 | ||
403 | off = i->iov_offset; | |
8cefc107 | 404 | buf = &pipe->bufs[i_head & p_mask]; |
241699cd AV |
405 | if (off) { |
406 | if (offset == off && buf->page == page) { | |
407 | /* merge with the last one */ | |
408 | buf->len += bytes; | |
409 | i->iov_offset += bytes; | |
410 | goto out; | |
411 | } | |
8cefc107 DH |
412 | i_head++; |
413 | buf = &pipe->bufs[i_head & p_mask]; | |
241699cd | 414 | } |
6718b6f8 | 415 | if (pipe_full(i_head, p_tail, pipe->max_usage)) |
241699cd | 416 | return 0; |
8cefc107 | 417 | |
241699cd | 418 | buf->ops = &page_cache_pipe_buf_ops; |
8cefc107 DH |
419 | get_page(page); |
420 | buf->page = page; | |
241699cd AV |
421 | buf->offset = offset; |
422 | buf->len = bytes; | |
8cefc107 DH |
423 | |
424 | pipe->head = i_head + 1; | |
241699cd | 425 | i->iov_offset = offset + bytes; |
8cefc107 | 426 | i->head = i_head; |
241699cd AV |
427 | out: |
428 | i->count -= bytes; | |
429 | return bytes; | |
430 | } | |
431 | ||
171a0203 | 432 | /* |
a6294593 AG |
433 | * fault_in_iov_iter_readable - fault in iov iterator for reading |
434 | * @i: iterator | |
435 | * @size: maximum length | |
436 | * | |
171a0203 | 437 | * Fault in one or more iovecs of the given iov_iter, to a maximum length of |
a6294593 AG |
438 | * @size. For each iovec, fault in each page that constitutes the iovec. |
439 | * | |
440 | * Returns the number of bytes not faulted in (like copy_to_user() and | |
441 | * copy_from_user()). | |
171a0203 | 442 | * |
a6294593 | 443 | * Always returns 0 for non-userspace iterators. |
171a0203 | 444 | */ |
a6294593 | 445 | size_t fault_in_iov_iter_readable(const struct iov_iter *i, size_t size) |
171a0203 | 446 | { |
0e8f0d67 | 447 | if (iter_is_iovec(i)) { |
a6294593 | 448 | size_t count = min(size, iov_iter_count(i)); |
8409a0d2 AV |
449 | const struct iovec *p; |
450 | size_t skip; | |
451 | ||
a6294593 AG |
452 | size -= count; |
453 | for (p = i->iov, skip = i->iov_offset; count; p++, skip = 0) { | |
454 | size_t len = min(count, p->iov_len - skip); | |
455 | size_t ret; | |
8409a0d2 AV |
456 | |
457 | if (unlikely(!len)) | |
458 | continue; | |
a6294593 AG |
459 | ret = fault_in_readable(p->iov_base + skip, len); |
460 | count -= len - ret; | |
461 | if (ret) | |
462 | break; | |
8409a0d2 | 463 | } |
a6294593 | 464 | return count + size; |
171a0203 AA |
465 | } |
466 | return 0; | |
467 | } | |
a6294593 | 468 | EXPORT_SYMBOL(fault_in_iov_iter_readable); |
171a0203 | 469 | |
cdd591fc AG |
470 | /* |
471 | * fault_in_iov_iter_writeable - fault in iov iterator for writing | |
472 | * @i: iterator | |
473 | * @size: maximum length | |
474 | * | |
475 | * Faults in the iterator using get_user_pages(), i.e., without triggering | |
476 | * hardware page faults. This is primarily useful when we already know that | |
477 | * some or all of the pages in @i aren't in memory. | |
478 | * | |
479 | * Returns the number of bytes not faulted in, like copy_to_user() and | |
480 | * copy_from_user(). | |
481 | * | |
482 | * Always returns 0 for non-user-space iterators. | |
483 | */ | |
484 | size_t fault_in_iov_iter_writeable(const struct iov_iter *i, size_t size) | |
485 | { | |
486 | if (iter_is_iovec(i)) { | |
487 | size_t count = min(size, iov_iter_count(i)); | |
488 | const struct iovec *p; | |
489 | size_t skip; | |
490 | ||
491 | size -= count; | |
492 | for (p = i->iov, skip = i->iov_offset; count; p++, skip = 0) { | |
493 | size_t len = min(count, p->iov_len - skip); | |
494 | size_t ret; | |
495 | ||
496 | if (unlikely(!len)) | |
497 | continue; | |
498 | ret = fault_in_safe_writeable(p->iov_base + skip, len); | |
499 | count -= len - ret; | |
500 | if (ret) | |
501 | break; | |
502 | } | |
503 | return count + size; | |
504 | } | |
505 | return 0; | |
506 | } | |
507 | EXPORT_SYMBOL(fault_in_iov_iter_writeable); | |
508 | ||
aa563d7b | 509 | void iov_iter_init(struct iov_iter *i, unsigned int direction, |
71d8e532 AV |
510 | const struct iovec *iov, unsigned long nr_segs, |
511 | size_t count) | |
512 | { | |
aa563d7b | 513 | WARN_ON(direction & ~(READ | WRITE)); |
8cd54c1c AV |
514 | *i = (struct iov_iter) { |
515 | .iter_type = ITER_IOVEC, | |
3337ab08 | 516 | .nofault = false, |
8cd54c1c AV |
517 | .data_source = direction, |
518 | .iov = iov, | |
519 | .nr_segs = nr_segs, | |
520 | .iov_offset = 0, | |
521 | .count = count | |
522 | }; | |
71d8e532 AV |
523 | } |
524 | EXPORT_SYMBOL(iov_iter_init); | |
7b2c99d1 | 525 | |
241699cd AV |
526 | static inline bool allocated(struct pipe_buffer *buf) |
527 | { | |
528 | return buf->ops == &default_pipe_buf_ops; | |
529 | } | |
530 | ||
8cefc107 DH |
531 | static inline void data_start(const struct iov_iter *i, |
532 | unsigned int *iter_headp, size_t *offp) | |
241699cd | 533 | { |
8cefc107 DH |
534 | unsigned int p_mask = i->pipe->ring_size - 1; |
535 | unsigned int iter_head = i->head; | |
241699cd | 536 | size_t off = i->iov_offset; |
8cefc107 DH |
537 | |
538 | if (off && (!allocated(&i->pipe->bufs[iter_head & p_mask]) || | |
539 | off == PAGE_SIZE)) { | |
540 | iter_head++; | |
241699cd AV |
541 | off = 0; |
542 | } | |
8cefc107 | 543 | *iter_headp = iter_head; |
241699cd AV |
544 | *offp = off; |
545 | } | |
546 | ||
547 | static size_t push_pipe(struct iov_iter *i, size_t size, | |
8cefc107 | 548 | int *iter_headp, size_t *offp) |
241699cd AV |
549 | { |
550 | struct pipe_inode_info *pipe = i->pipe; | |
8cefc107 DH |
551 | unsigned int p_tail = pipe->tail; |
552 | unsigned int p_mask = pipe->ring_size - 1; | |
553 | unsigned int iter_head; | |
241699cd | 554 | size_t off; |
241699cd AV |
555 | ssize_t left; |
556 | ||
557 | if (unlikely(size > i->count)) | |
558 | size = i->count; | |
559 | if (unlikely(!size)) | |
560 | return 0; | |
561 | ||
562 | left = size; | |
8cefc107 DH |
563 | data_start(i, &iter_head, &off); |
564 | *iter_headp = iter_head; | |
241699cd AV |
565 | *offp = off; |
566 | if (off) { | |
567 | left -= PAGE_SIZE - off; | |
568 | if (left <= 0) { | |
8cefc107 | 569 | pipe->bufs[iter_head & p_mask].len += size; |
241699cd AV |
570 | return size; |
571 | } | |
8cefc107 DH |
572 | pipe->bufs[iter_head & p_mask].len = PAGE_SIZE; |
573 | iter_head++; | |
241699cd | 574 | } |
6718b6f8 | 575 | while (!pipe_full(iter_head, p_tail, pipe->max_usage)) { |
8cefc107 | 576 | struct pipe_buffer *buf = &pipe->bufs[iter_head & p_mask]; |
241699cd AV |
577 | struct page *page = alloc_page(GFP_USER); |
578 | if (!page) | |
579 | break; | |
8cefc107 DH |
580 | |
581 | buf->ops = &default_pipe_buf_ops; | |
582 | buf->page = page; | |
583 | buf->offset = 0; | |
584 | buf->len = min_t(ssize_t, left, PAGE_SIZE); | |
585 | left -= buf->len; | |
586 | iter_head++; | |
587 | pipe->head = iter_head; | |
588 | ||
589 | if (left == 0) | |
241699cd | 590 | return size; |
241699cd AV |
591 | } |
592 | return size - left; | |
593 | } | |
594 | ||
595 | static size_t copy_pipe_to_iter(const void *addr, size_t bytes, | |
596 | struct iov_iter *i) | |
597 | { | |
598 | struct pipe_inode_info *pipe = i->pipe; | |
8cefc107 DH |
599 | unsigned int p_mask = pipe->ring_size - 1; |
600 | unsigned int i_head; | |
241699cd | 601 | size_t n, off; |
241699cd AV |
602 | |
603 | if (!sanity(i)) | |
604 | return 0; | |
605 | ||
8cefc107 | 606 | bytes = n = push_pipe(i, bytes, &i_head, &off); |
241699cd AV |
607 | if (unlikely(!n)) |
608 | return 0; | |
8cefc107 | 609 | do { |
241699cd | 610 | size_t chunk = min_t(size_t, n, PAGE_SIZE - off); |
8cefc107 DH |
611 | memcpy_to_page(pipe->bufs[i_head & p_mask].page, off, addr, chunk); |
612 | i->head = i_head; | |
241699cd AV |
613 | i->iov_offset = off + chunk; |
614 | n -= chunk; | |
615 | addr += chunk; | |
8cefc107 DH |
616 | off = 0; |
617 | i_head++; | |
618 | } while (n); | |
241699cd AV |
619 | i->count -= bytes; |
620 | return bytes; | |
621 | } | |
622 | ||
f9152895 AV |
623 | static __wsum csum_and_memcpy(void *to, const void *from, size_t len, |
624 | __wsum sum, size_t off) | |
625 | { | |
cc44c17b | 626 | __wsum next = csum_partial_copy_nocheck(from, to, len); |
f9152895 AV |
627 | return csum_block_add(sum, next, off); |
628 | } | |
629 | ||
78e1f386 | 630 | static size_t csum_and_copy_to_pipe_iter(const void *addr, size_t bytes, |
6852df12 | 631 | struct iov_iter *i, __wsum *sump) |
78e1f386 AV |
632 | { |
633 | struct pipe_inode_info *pipe = i->pipe; | |
8cefc107 | 634 | unsigned int p_mask = pipe->ring_size - 1; |
6852df12 AV |
635 | __wsum sum = *sump; |
636 | size_t off = 0; | |
8cefc107 | 637 | unsigned int i_head; |
6852df12 | 638 | size_t r; |
78e1f386 AV |
639 | |
640 | if (!sanity(i)) | |
641 | return 0; | |
642 | ||
6852df12 AV |
643 | bytes = push_pipe(i, bytes, &i_head, &r); |
644 | while (bytes) { | |
645 | size_t chunk = min_t(size_t, bytes, PAGE_SIZE - r); | |
2495bdcc | 646 | char *p = kmap_local_page(pipe->bufs[i_head & p_mask].page); |
6852df12 | 647 | sum = csum_and_memcpy(p + r, addr + off, chunk, sum, off); |
2495bdcc | 648 | kunmap_local(p); |
8cefc107 | 649 | i->head = i_head; |
78e1f386 | 650 | i->iov_offset = r + chunk; |
6852df12 | 651 | bytes -= chunk; |
78e1f386 | 652 | off += chunk; |
8cefc107 DH |
653 | r = 0; |
654 | i_head++; | |
6852df12 AV |
655 | } |
656 | *sump = sum; | |
657 | i->count -= off; | |
658 | return off; | |
78e1f386 AV |
659 | } |
660 | ||
aa28de27 | 661 | size_t _copy_to_iter(const void *addr, size_t bytes, struct iov_iter *i) |
62a8067a | 662 | { |
00e23707 | 663 | if (unlikely(iov_iter_is_pipe(i))) |
241699cd | 664 | return copy_pipe_to_iter(addr, bytes, i); |
09fc68dc AV |
665 | if (iter_is_iovec(i)) |
666 | might_fault(); | |
7baa5099 AV |
667 | iterate_and_advance(i, bytes, base, len, off, |
668 | copyout(base, addr + off, len), | |
669 | memcpy(base, addr + off, len) | |
3d4d3e48 | 670 | ) |
62a8067a | 671 | |
3d4d3e48 | 672 | return bytes; |
c35e0248 | 673 | } |
aa28de27 | 674 | EXPORT_SYMBOL(_copy_to_iter); |
c35e0248 | 675 | |
ec6347bb DW |
676 | #ifdef CONFIG_ARCH_HAS_COPY_MC |
677 | static int copyout_mc(void __user *to, const void *from, size_t n) | |
8780356e | 678 | { |
96d4f267 | 679 | if (access_ok(to, n)) { |
d0ef4c36 | 680 | instrument_copy_to_user(to, from, n); |
ec6347bb | 681 | n = copy_mc_to_user((__force void *) to, from, n); |
8780356e DW |
682 | } |
683 | return n; | |
684 | } | |
685 | ||
ec6347bb | 686 | static size_t copy_mc_pipe_to_iter(const void *addr, size_t bytes, |
ca146f6f DW |
687 | struct iov_iter *i) |
688 | { | |
689 | struct pipe_inode_info *pipe = i->pipe; | |
8cefc107 DH |
690 | unsigned int p_mask = pipe->ring_size - 1; |
691 | unsigned int i_head; | |
ca146f6f | 692 | size_t n, off, xfer = 0; |
ca146f6f DW |
693 | |
694 | if (!sanity(i)) | |
695 | return 0; | |
696 | ||
2a510a74 AV |
697 | n = push_pipe(i, bytes, &i_head, &off); |
698 | while (n) { | |
ca146f6f | 699 | size_t chunk = min_t(size_t, n, PAGE_SIZE - off); |
2a510a74 | 700 | char *p = kmap_local_page(pipe->bufs[i_head & p_mask].page); |
ca146f6f | 701 | unsigned long rem; |
2a510a74 AV |
702 | rem = copy_mc_to_kernel(p + off, addr + xfer, chunk); |
703 | chunk -= rem; | |
704 | kunmap_local(p); | |
8cefc107 | 705 | i->head = i_head; |
2a510a74 AV |
706 | i->iov_offset = off + chunk; |
707 | xfer += chunk; | |
ca146f6f DW |
708 | if (rem) |
709 | break; | |
710 | n -= chunk; | |
8cefc107 DH |
711 | off = 0; |
712 | i_head++; | |
2a510a74 | 713 | } |
ca146f6f DW |
714 | i->count -= xfer; |
715 | return xfer; | |
716 | } | |
717 | ||
bf3eeb9b | 718 | /** |
ec6347bb | 719 | * _copy_mc_to_iter - copy to iter with source memory error exception handling |
bf3eeb9b DW |
720 | * @addr: source kernel address |
721 | * @bytes: total transfer length | |
44e55997 | 722 | * @i: destination iterator |
bf3eeb9b | 723 | * |
ec6347bb DW |
724 | * The pmem driver deploys this for the dax operation |
725 | * (dax_copy_to_iter()) for dax reads (bypass page-cache and the | |
726 | * block-layer). Upon #MC read(2) aborts and returns EIO or the bytes | |
727 | * successfully copied. | |
bf3eeb9b | 728 | * |
ec6347bb | 729 | * The main differences between this and typical _copy_to_iter(). |
bf3eeb9b DW |
730 | * |
731 | * * Typical tail/residue handling after a fault retries the copy | |
732 | * byte-by-byte until the fault happens again. Re-triggering machine | |
733 | * checks is potentially fatal so the implementation uses source | |
734 | * alignment and poison alignment assumptions to avoid re-triggering | |
735 | * hardware exceptions. | |
736 | * | |
737 | * * ITER_KVEC, ITER_PIPE, and ITER_BVEC can return short copies. | |
738 | * Compare to copy_to_iter() where only ITER_IOVEC attempts might return | |
739 | * a short copy. | |
44e55997 RD |
740 | * |
741 | * Return: number of bytes copied (may be %0) | |
bf3eeb9b | 742 | */ |
ec6347bb | 743 | size_t _copy_mc_to_iter(const void *addr, size_t bytes, struct iov_iter *i) |
8780356e | 744 | { |
00e23707 | 745 | if (unlikely(iov_iter_is_pipe(i))) |
ec6347bb | 746 | return copy_mc_pipe_to_iter(addr, bytes, i); |
8780356e DW |
747 | if (iter_is_iovec(i)) |
748 | might_fault(); | |
7baa5099 AV |
749 | __iterate_and_advance(i, bytes, base, len, off, |
750 | copyout_mc(base, addr + off, len), | |
751 | copy_mc_to_kernel(base, addr + off, len) | |
8780356e DW |
752 | ) |
753 | ||
754 | return bytes; | |
755 | } | |
ec6347bb DW |
756 | EXPORT_SYMBOL_GPL(_copy_mc_to_iter); |
757 | #endif /* CONFIG_ARCH_HAS_COPY_MC */ | |
8780356e | 758 | |
aa28de27 | 759 | size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i) |
c35e0248 | 760 | { |
00e23707 | 761 | if (unlikely(iov_iter_is_pipe(i))) { |
241699cd AV |
762 | WARN_ON(1); |
763 | return 0; | |
764 | } | |
09fc68dc AV |
765 | if (iter_is_iovec(i)) |
766 | might_fault(); | |
7baa5099 AV |
767 | iterate_and_advance(i, bytes, base, len, off, |
768 | copyin(addr + off, base, len), | |
769 | memcpy(addr + off, base, len) | |
0dbca9a4 AV |
770 | ) |
771 | ||
772 | return bytes; | |
c35e0248 | 773 | } |
aa28de27 | 774 | EXPORT_SYMBOL(_copy_from_iter); |
c35e0248 | 775 | |
aa28de27 | 776 | size_t _copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i) |
aa583096 | 777 | { |
00e23707 | 778 | if (unlikely(iov_iter_is_pipe(i))) { |
241699cd AV |
779 | WARN_ON(1); |
780 | return 0; | |
781 | } | |
7baa5099 AV |
782 | iterate_and_advance(i, bytes, base, len, off, |
783 | __copy_from_user_inatomic_nocache(addr + off, base, len), | |
784 | memcpy(addr + off, base, len) | |
aa583096 AV |
785 | ) |
786 | ||
787 | return bytes; | |
788 | } | |
aa28de27 | 789 | EXPORT_SYMBOL(_copy_from_iter_nocache); |
aa583096 | 790 | |
0aed55af | 791 | #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE |
abd08d7d DW |
792 | /** |
793 | * _copy_from_iter_flushcache - write destination through cpu cache | |
794 | * @addr: destination kernel address | |
795 | * @bytes: total transfer length | |
44e55997 | 796 | * @i: source iterator |
abd08d7d DW |
797 | * |
798 | * The pmem driver arranges for filesystem-dax to use this facility via | |
799 | * dax_copy_from_iter() for ensuring that writes to persistent memory | |
800 | * are flushed through the CPU cache. It is differentiated from | |
801 | * _copy_from_iter_nocache() in that guarantees all data is flushed for | |
802 | * all iterator types. The _copy_from_iter_nocache() only attempts to | |
803 | * bypass the cache for the ITER_IOVEC case, and on some archs may use | |
804 | * instructions that strand dirty-data in the cache. | |
44e55997 RD |
805 | * |
806 | * Return: number of bytes copied (may be %0) | |
abd08d7d | 807 | */ |
6a37e940 | 808 | size_t _copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i) |
0aed55af | 809 | { |
00e23707 | 810 | if (unlikely(iov_iter_is_pipe(i))) { |
0aed55af DW |
811 | WARN_ON(1); |
812 | return 0; | |
813 | } | |
7baa5099 AV |
814 | iterate_and_advance(i, bytes, base, len, off, |
815 | __copy_from_user_flushcache(addr + off, base, len), | |
816 | memcpy_flushcache(addr + off, base, len) | |
0aed55af DW |
817 | ) |
818 | ||
819 | return bytes; | |
820 | } | |
6a37e940 | 821 | EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache); |
0aed55af DW |
822 | #endif |
823 | ||
72e809ed AV |
824 | static inline bool page_copy_sane(struct page *page, size_t offset, size_t n) |
825 | { | |
6daef95b ED |
826 | struct page *head; |
827 | size_t v = n + offset; | |
828 | ||
829 | /* | |
830 | * The general case needs to access the page order in order | |
831 | * to compute the page size. | |
832 | * However, we mostly deal with order-0 pages and thus can | |
833 | * avoid a possible cache line miss for requests that fit all | |
834 | * page orders. | |
835 | */ | |
836 | if (n <= v && v <= PAGE_SIZE) | |
837 | return true; | |
838 | ||
839 | head = compound_head(page); | |
840 | v += (page - head) << PAGE_SHIFT; | |
a90bcb86 | 841 | |
a50b854e | 842 | if (likely(n <= v && v <= (page_size(head)))) |
72e809ed AV |
843 | return true; |
844 | WARN_ON(1); | |
845 | return false; | |
846 | } | |
cbbd26b8 | 847 | |
08aa6479 | 848 | static size_t __copy_page_to_iter(struct page *page, size_t offset, size_t bytes, |
62a8067a AV |
849 | struct iov_iter *i) |
850 | { | |
28f38db7 AV |
851 | if (likely(iter_is_iovec(i))) |
852 | return copy_page_to_iter_iovec(page, offset, bytes, i); | |
853 | if (iov_iter_is_bvec(i) || iov_iter_is_kvec(i) || iov_iter_is_xarray(i)) { | |
c1d4d6a9 AV |
854 | void *kaddr = kmap_local_page(page); |
855 | size_t wanted = _copy_to_iter(kaddr + offset, bytes, i); | |
856 | kunmap_local(kaddr); | |
d271524a | 857 | return wanted; |
28f38db7 AV |
858 | } |
859 | if (iov_iter_is_pipe(i)) | |
860 | return copy_page_to_iter_pipe(page, offset, bytes, i); | |
861 | if (unlikely(iov_iter_is_discard(i))) { | |
a506abc7 AV |
862 | if (unlikely(i->count < bytes)) |
863 | bytes = i->count; | |
864 | i->count -= bytes; | |
9ea9ce04 | 865 | return bytes; |
28f38db7 AV |
866 | } |
867 | WARN_ON(1); | |
868 | return 0; | |
62a8067a | 869 | } |
08aa6479 AV |
870 | |
871 | size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes, | |
872 | struct iov_iter *i) | |
873 | { | |
874 | size_t res = 0; | |
875 | if (unlikely(!page_copy_sane(page, offset, bytes))) | |
876 | return 0; | |
877 | page += offset / PAGE_SIZE; // first subpage | |
878 | offset %= PAGE_SIZE; | |
879 | while (1) { | |
880 | size_t n = __copy_page_to_iter(page, offset, | |
881 | min(bytes, (size_t)PAGE_SIZE - offset), i); | |
882 | res += n; | |
883 | bytes -= n; | |
884 | if (!bytes || !n) | |
885 | break; | |
886 | offset += n; | |
887 | if (offset == PAGE_SIZE) { | |
888 | page++; | |
889 | offset = 0; | |
890 | } | |
891 | } | |
892 | return res; | |
893 | } | |
62a8067a AV |
894 | EXPORT_SYMBOL(copy_page_to_iter); |
895 | ||
896 | size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes, | |
897 | struct iov_iter *i) | |
898 | { | |
72e809ed AV |
899 | if (unlikely(!page_copy_sane(page, offset, bytes))) |
900 | return 0; | |
28f38db7 AV |
901 | if (likely(iter_is_iovec(i))) |
902 | return copy_page_from_iter_iovec(page, offset, bytes, i); | |
903 | if (iov_iter_is_bvec(i) || iov_iter_is_kvec(i) || iov_iter_is_xarray(i)) { | |
55ca375c | 904 | void *kaddr = kmap_local_page(page); |
aa28de27 | 905 | size_t wanted = _copy_from_iter(kaddr + offset, bytes, i); |
55ca375c | 906 | kunmap_local(kaddr); |
d271524a | 907 | return wanted; |
28f38db7 AV |
908 | } |
909 | WARN_ON(1); | |
910 | return 0; | |
62a8067a AV |
911 | } |
912 | EXPORT_SYMBOL(copy_page_from_iter); | |
913 | ||
241699cd AV |
914 | static size_t pipe_zero(size_t bytes, struct iov_iter *i) |
915 | { | |
916 | struct pipe_inode_info *pipe = i->pipe; | |
8cefc107 DH |
917 | unsigned int p_mask = pipe->ring_size - 1; |
918 | unsigned int i_head; | |
241699cd | 919 | size_t n, off; |
241699cd AV |
920 | |
921 | if (!sanity(i)) | |
922 | return 0; | |
923 | ||
8cefc107 | 924 | bytes = n = push_pipe(i, bytes, &i_head, &off); |
241699cd AV |
925 | if (unlikely(!n)) |
926 | return 0; | |
927 | ||
8cefc107 | 928 | do { |
241699cd | 929 | size_t chunk = min_t(size_t, n, PAGE_SIZE - off); |
893839fd AV |
930 | char *p = kmap_local_page(pipe->bufs[i_head & p_mask].page); |
931 | memset(p + off, 0, chunk); | |
932 | kunmap_local(p); | |
8cefc107 | 933 | i->head = i_head; |
241699cd AV |
934 | i->iov_offset = off + chunk; |
935 | n -= chunk; | |
8cefc107 DH |
936 | off = 0; |
937 | i_head++; | |
938 | } while (n); | |
241699cd AV |
939 | i->count -= bytes; |
940 | return bytes; | |
941 | } | |
942 | ||
c35e0248 MW |
943 | size_t iov_iter_zero(size_t bytes, struct iov_iter *i) |
944 | { | |
00e23707 | 945 | if (unlikely(iov_iter_is_pipe(i))) |
241699cd | 946 | return pipe_zero(bytes, i); |
7baa5099 AV |
947 | iterate_and_advance(i, bytes, base, len, count, |
948 | clear_user(base, len), | |
949 | memset(base, 0, len) | |
8442fa46 AV |
950 | ) |
951 | ||
952 | return bytes; | |
c35e0248 MW |
953 | } |
954 | EXPORT_SYMBOL(iov_iter_zero); | |
955 | ||
f0b65f39 AV |
956 | size_t copy_page_from_iter_atomic(struct page *page, unsigned offset, size_t bytes, |
957 | struct iov_iter *i) | |
62a8067a | 958 | { |
04a31165 | 959 | char *kaddr = kmap_atomic(page), *p = kaddr + offset; |
72e809ed AV |
960 | if (unlikely(!page_copy_sane(page, offset, bytes))) { |
961 | kunmap_atomic(kaddr); | |
962 | return 0; | |
963 | } | |
9ea9ce04 | 964 | if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) { |
241699cd AV |
965 | kunmap_atomic(kaddr); |
966 | WARN_ON(1); | |
967 | return 0; | |
968 | } | |
7baa5099 AV |
969 | iterate_and_advance(i, bytes, base, len, off, |
970 | copyin(p + off, base, len), | |
971 | memcpy(p + off, base, len) | |
04a31165 AV |
972 | ) |
973 | kunmap_atomic(kaddr); | |
974 | return bytes; | |
62a8067a | 975 | } |
f0b65f39 | 976 | EXPORT_SYMBOL(copy_page_from_iter_atomic); |
62a8067a | 977 | |
b9dc6f65 AV |
978 | static inline void pipe_truncate(struct iov_iter *i) |
979 | { | |
980 | struct pipe_inode_info *pipe = i->pipe; | |
8cefc107 DH |
981 | unsigned int p_tail = pipe->tail; |
982 | unsigned int p_head = pipe->head; | |
983 | unsigned int p_mask = pipe->ring_size - 1; | |
984 | ||
985 | if (!pipe_empty(p_head, p_tail)) { | |
986 | struct pipe_buffer *buf; | |
987 | unsigned int i_head = i->head; | |
b9dc6f65 | 988 | size_t off = i->iov_offset; |
8cefc107 | 989 | |
b9dc6f65 | 990 | if (off) { |
8cefc107 DH |
991 | buf = &pipe->bufs[i_head & p_mask]; |
992 | buf->len = off - buf->offset; | |
993 | i_head++; | |
b9dc6f65 | 994 | } |
8cefc107 DH |
995 | while (p_head != i_head) { |
996 | p_head--; | |
997 | pipe_buf_release(pipe, &pipe->bufs[p_head & p_mask]); | |
b9dc6f65 | 998 | } |
8cefc107 DH |
999 | |
1000 | pipe->head = p_head; | |
b9dc6f65 AV |
1001 | } |
1002 | } | |
1003 | ||
241699cd AV |
1004 | static void pipe_advance(struct iov_iter *i, size_t size) |
1005 | { | |
1006 | struct pipe_inode_info *pipe = i->pipe; | |
241699cd | 1007 | if (size) { |
b9dc6f65 | 1008 | struct pipe_buffer *buf; |
8cefc107 DH |
1009 | unsigned int p_mask = pipe->ring_size - 1; |
1010 | unsigned int i_head = i->head; | |
b9dc6f65 | 1011 | size_t off = i->iov_offset, left = size; |
8cefc107 | 1012 | |
241699cd | 1013 | if (off) /* make it relative to the beginning of buffer */ |
8cefc107 | 1014 | left += off - pipe->bufs[i_head & p_mask].offset; |
241699cd | 1015 | while (1) { |
8cefc107 | 1016 | buf = &pipe->bufs[i_head & p_mask]; |
b9dc6f65 | 1017 | if (left <= buf->len) |
241699cd | 1018 | break; |
b9dc6f65 | 1019 | left -= buf->len; |
8cefc107 | 1020 | i_head++; |
241699cd | 1021 | } |
8cefc107 | 1022 | i->head = i_head; |
b9dc6f65 | 1023 | i->iov_offset = buf->offset + left; |
241699cd | 1024 | } |
b9dc6f65 AV |
1025 | i->count -= size; |
1026 | /* ... and discard everything past that point */ | |
1027 | pipe_truncate(i); | |
241699cd AV |
1028 | } |
1029 | ||
54c8195b PB |
1030 | static void iov_iter_bvec_advance(struct iov_iter *i, size_t size) |
1031 | { | |
1032 | struct bvec_iter bi; | |
1033 | ||
1034 | bi.bi_size = i->count; | |
1035 | bi.bi_bvec_done = i->iov_offset; | |
1036 | bi.bi_idx = 0; | |
1037 | bvec_iter_advance(i->bvec, &bi, size); | |
1038 | ||
1039 | i->bvec += bi.bi_idx; | |
1040 | i->nr_segs -= bi.bi_idx; | |
1041 | i->count = bi.bi_size; | |
1042 | i->iov_offset = bi.bi_bvec_done; | |
1043 | } | |
1044 | ||
185ac4d4 AV |
1045 | static void iov_iter_iovec_advance(struct iov_iter *i, size_t size) |
1046 | { | |
1047 | const struct iovec *iov, *end; | |
1048 | ||
1049 | if (!i->count) | |
1050 | return; | |
1051 | i->count -= size; | |
1052 | ||
1053 | size += i->iov_offset; // from beginning of current segment | |
1054 | for (iov = i->iov, end = iov + i->nr_segs; iov < end; iov++) { | |
1055 | if (likely(size < iov->iov_len)) | |
1056 | break; | |
1057 | size -= iov->iov_len; | |
1058 | } | |
1059 | i->iov_offset = size; | |
1060 | i->nr_segs -= iov - i->iov; | |
1061 | i->iov = iov; | |
1062 | } | |
1063 | ||
62a8067a AV |
1064 | void iov_iter_advance(struct iov_iter *i, size_t size) |
1065 | { | |
3b3fc051 AV |
1066 | if (unlikely(i->count < size)) |
1067 | size = i->count; | |
185ac4d4 AV |
1068 | if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) { |
1069 | /* iovec and kvec have identical layouts */ | |
1070 | iov_iter_iovec_advance(i, size); | |
1071 | } else if (iov_iter_is_bvec(i)) { | |
1072 | iov_iter_bvec_advance(i, size); | |
1073 | } else if (iov_iter_is_pipe(i)) { | |
241699cd | 1074 | pipe_advance(i, size); |
185ac4d4 | 1075 | } else if (unlikely(iov_iter_is_xarray(i))) { |
7ff50620 DH |
1076 | i->iov_offset += size; |
1077 | i->count -= size; | |
185ac4d4 AV |
1078 | } else if (iov_iter_is_discard(i)) { |
1079 | i->count -= size; | |
54c8195b | 1080 | } |
62a8067a AV |
1081 | } |
1082 | EXPORT_SYMBOL(iov_iter_advance); | |
1083 | ||
27c0e374 AV |
1084 | void iov_iter_revert(struct iov_iter *i, size_t unroll) |
1085 | { | |
1086 | if (!unroll) | |
1087 | return; | |
5b47d59a AV |
1088 | if (WARN_ON(unroll > MAX_RW_COUNT)) |
1089 | return; | |
27c0e374 | 1090 | i->count += unroll; |
00e23707 | 1091 | if (unlikely(iov_iter_is_pipe(i))) { |
27c0e374 | 1092 | struct pipe_inode_info *pipe = i->pipe; |
8cefc107 DH |
1093 | unsigned int p_mask = pipe->ring_size - 1; |
1094 | unsigned int i_head = i->head; | |
27c0e374 AV |
1095 | size_t off = i->iov_offset; |
1096 | while (1) { | |
8cefc107 DH |
1097 | struct pipe_buffer *b = &pipe->bufs[i_head & p_mask]; |
1098 | size_t n = off - b->offset; | |
27c0e374 | 1099 | if (unroll < n) { |
4fa55cef | 1100 | off -= unroll; |
27c0e374 AV |
1101 | break; |
1102 | } | |
1103 | unroll -= n; | |
8cefc107 | 1104 | if (!unroll && i_head == i->start_head) { |
27c0e374 AV |
1105 | off = 0; |
1106 | break; | |
1107 | } | |
8cefc107 DH |
1108 | i_head--; |
1109 | b = &pipe->bufs[i_head & p_mask]; | |
1110 | off = b->offset + b->len; | |
27c0e374 AV |
1111 | } |
1112 | i->iov_offset = off; | |
8cefc107 | 1113 | i->head = i_head; |
27c0e374 AV |
1114 | pipe_truncate(i); |
1115 | return; | |
1116 | } | |
9ea9ce04 DH |
1117 | if (unlikely(iov_iter_is_discard(i))) |
1118 | return; | |
27c0e374 AV |
1119 | if (unroll <= i->iov_offset) { |
1120 | i->iov_offset -= unroll; | |
1121 | return; | |
1122 | } | |
1123 | unroll -= i->iov_offset; | |
7ff50620 DH |
1124 | if (iov_iter_is_xarray(i)) { |
1125 | BUG(); /* We should never go beyond the start of the specified | |
1126 | * range since we might then be straying into pages that | |
1127 | * aren't pinned. | |
1128 | */ | |
1129 | } else if (iov_iter_is_bvec(i)) { | |
27c0e374 AV |
1130 | const struct bio_vec *bvec = i->bvec; |
1131 | while (1) { | |
1132 | size_t n = (--bvec)->bv_len; | |
1133 | i->nr_segs++; | |
1134 | if (unroll <= n) { | |
1135 | i->bvec = bvec; | |
1136 | i->iov_offset = n - unroll; | |
1137 | return; | |
1138 | } | |
1139 | unroll -= n; | |
1140 | } | |
1141 | } else { /* same logics for iovec and kvec */ | |
1142 | const struct iovec *iov = i->iov; | |
1143 | while (1) { | |
1144 | size_t n = (--iov)->iov_len; | |
1145 | i->nr_segs++; | |
1146 | if (unroll <= n) { | |
1147 | i->iov = iov; | |
1148 | i->iov_offset = n - unroll; | |
1149 | return; | |
1150 | } | |
1151 | unroll -= n; | |
1152 | } | |
1153 | } | |
1154 | } | |
1155 | EXPORT_SYMBOL(iov_iter_revert); | |
1156 | ||
62a8067a AV |
1157 | /* |
1158 | * Return the count of just the current iov_iter segment. | |
1159 | */ | |
1160 | size_t iov_iter_single_seg_count(const struct iov_iter *i) | |
1161 | { | |
28f38db7 AV |
1162 | if (i->nr_segs > 1) { |
1163 | if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) | |
1164 | return min(i->count, i->iov->iov_len - i->iov_offset); | |
1165 | if (iov_iter_is_bvec(i)) | |
1166 | return min(i->count, i->bvec->bv_len - i->iov_offset); | |
1167 | } | |
1168 | return i->count; | |
62a8067a AV |
1169 | } |
1170 | EXPORT_SYMBOL(iov_iter_single_seg_count); | |
1171 | ||
aa563d7b | 1172 | void iov_iter_kvec(struct iov_iter *i, unsigned int direction, |
05afcb77 | 1173 | const struct kvec *kvec, unsigned long nr_segs, |
abb78f87 AV |
1174 | size_t count) |
1175 | { | |
aa563d7b | 1176 | WARN_ON(direction & ~(READ | WRITE)); |
8cd54c1c AV |
1177 | *i = (struct iov_iter){ |
1178 | .iter_type = ITER_KVEC, | |
1179 | .data_source = direction, | |
1180 | .kvec = kvec, | |
1181 | .nr_segs = nr_segs, | |
1182 | .iov_offset = 0, | |
1183 | .count = count | |
1184 | }; | |
abb78f87 AV |
1185 | } |
1186 | EXPORT_SYMBOL(iov_iter_kvec); | |
1187 | ||
aa563d7b | 1188 | void iov_iter_bvec(struct iov_iter *i, unsigned int direction, |
05afcb77 AV |
1189 | const struct bio_vec *bvec, unsigned long nr_segs, |
1190 | size_t count) | |
1191 | { | |
aa563d7b | 1192 | WARN_ON(direction & ~(READ | WRITE)); |
8cd54c1c AV |
1193 | *i = (struct iov_iter){ |
1194 | .iter_type = ITER_BVEC, | |
1195 | .data_source = direction, | |
1196 | .bvec = bvec, | |
1197 | .nr_segs = nr_segs, | |
1198 | .iov_offset = 0, | |
1199 | .count = count | |
1200 | }; | |
05afcb77 AV |
1201 | } |
1202 | EXPORT_SYMBOL(iov_iter_bvec); | |
1203 | ||
aa563d7b | 1204 | void iov_iter_pipe(struct iov_iter *i, unsigned int direction, |
241699cd AV |
1205 | struct pipe_inode_info *pipe, |
1206 | size_t count) | |
1207 | { | |
aa563d7b | 1208 | BUG_ON(direction != READ); |
8cefc107 | 1209 | WARN_ON(pipe_full(pipe->head, pipe->tail, pipe->ring_size)); |
8cd54c1c AV |
1210 | *i = (struct iov_iter){ |
1211 | .iter_type = ITER_PIPE, | |
1212 | .data_source = false, | |
1213 | .pipe = pipe, | |
1214 | .head = pipe->head, | |
1215 | .start_head = pipe->head, | |
1216 | .iov_offset = 0, | |
1217 | .count = count | |
1218 | }; | |
241699cd AV |
1219 | } |
1220 | EXPORT_SYMBOL(iov_iter_pipe); | |
1221 | ||
7ff50620 DH |
1222 | /** |
1223 | * iov_iter_xarray - Initialise an I/O iterator to use the pages in an xarray | |
1224 | * @i: The iterator to initialise. | |
1225 | * @direction: The direction of the transfer. | |
1226 | * @xarray: The xarray to access. | |
1227 | * @start: The start file position. | |
1228 | * @count: The size of the I/O buffer in bytes. | |
1229 | * | |
1230 | * Set up an I/O iterator to either draw data out of the pages attached to an | |
1231 | * inode or to inject data into those pages. The pages *must* be prevented | |
1232 | * from evaporation, either by taking a ref on them or locking them by the | |
1233 | * caller. | |
1234 | */ | |
1235 | void iov_iter_xarray(struct iov_iter *i, unsigned int direction, | |
1236 | struct xarray *xarray, loff_t start, size_t count) | |
1237 | { | |
1238 | BUG_ON(direction & ~1); | |
8cd54c1c AV |
1239 | *i = (struct iov_iter) { |
1240 | .iter_type = ITER_XARRAY, | |
1241 | .data_source = direction, | |
1242 | .xarray = xarray, | |
1243 | .xarray_start = start, | |
1244 | .count = count, | |
1245 | .iov_offset = 0 | |
1246 | }; | |
7ff50620 DH |
1247 | } |
1248 | EXPORT_SYMBOL(iov_iter_xarray); | |
1249 | ||
9ea9ce04 DH |
1250 | /** |
1251 | * iov_iter_discard - Initialise an I/O iterator that discards data | |
1252 | * @i: The iterator to initialise. | |
1253 | * @direction: The direction of the transfer. | |
1254 | * @count: The size of the I/O buffer in bytes. | |
1255 | * | |
1256 | * Set up an I/O iterator that just discards everything that's written to it. | |
1257 | * It's only available as a READ iterator. | |
1258 | */ | |
1259 | void iov_iter_discard(struct iov_iter *i, unsigned int direction, size_t count) | |
1260 | { | |
1261 | BUG_ON(direction != READ); | |
8cd54c1c AV |
1262 | *i = (struct iov_iter){ |
1263 | .iter_type = ITER_DISCARD, | |
1264 | .data_source = false, | |
1265 | .count = count, | |
1266 | .iov_offset = 0 | |
1267 | }; | |
9ea9ce04 DH |
1268 | } |
1269 | EXPORT_SYMBOL(iov_iter_discard); | |
1270 | ||
9221d2e3 | 1271 | static unsigned long iov_iter_alignment_iovec(const struct iov_iter *i) |
62a8067a | 1272 | { |
04a31165 AV |
1273 | unsigned long res = 0; |
1274 | size_t size = i->count; | |
9221d2e3 AV |
1275 | size_t skip = i->iov_offset; |
1276 | unsigned k; | |
1277 | ||
1278 | for (k = 0; k < i->nr_segs; k++, skip = 0) { | |
1279 | size_t len = i->iov[k].iov_len - skip; | |
1280 | if (len) { | |
1281 | res |= (unsigned long)i->iov[k].iov_base + skip; | |
1282 | if (len > size) | |
1283 | len = size; | |
1284 | res |= len; | |
1285 | size -= len; | |
1286 | if (!size) | |
1287 | break; | |
1288 | } | |
1289 | } | |
1290 | return res; | |
1291 | } | |
04a31165 | 1292 | |
9221d2e3 AV |
1293 | static unsigned long iov_iter_alignment_bvec(const struct iov_iter *i) |
1294 | { | |
1295 | unsigned res = 0; | |
1296 | size_t size = i->count; | |
1297 | unsigned skip = i->iov_offset; | |
1298 | unsigned k; | |
1299 | ||
1300 | for (k = 0; k < i->nr_segs; k++, skip = 0) { | |
1301 | size_t len = i->bvec[k].bv_len - skip; | |
1302 | res |= (unsigned long)i->bvec[k].bv_offset + skip; | |
1303 | if (len > size) | |
1304 | len = size; | |
1305 | res |= len; | |
1306 | size -= len; | |
1307 | if (!size) | |
1308 | break; | |
1309 | } | |
1310 | return res; | |
1311 | } | |
1312 | ||
1313 | unsigned long iov_iter_alignment(const struct iov_iter *i) | |
1314 | { | |
1315 | /* iovec and kvec have identical layouts */ | |
1316 | if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) | |
1317 | return iov_iter_alignment_iovec(i); | |
1318 | ||
1319 | if (iov_iter_is_bvec(i)) | |
1320 | return iov_iter_alignment_bvec(i); | |
1321 | ||
1322 | if (iov_iter_is_pipe(i)) { | |
e0ff126e | 1323 | unsigned int p_mask = i->pipe->ring_size - 1; |
9221d2e3 | 1324 | size_t size = i->count; |
e0ff126e | 1325 | |
8cefc107 | 1326 | if (size && i->iov_offset && allocated(&i->pipe->bufs[i->head & p_mask])) |
241699cd AV |
1327 | return size | i->iov_offset; |
1328 | return size; | |
1329 | } | |
9221d2e3 AV |
1330 | |
1331 | if (iov_iter_is_xarray(i)) | |
3d14ec1f | 1332 | return (i->xarray_start + i->iov_offset) | i->count; |
9221d2e3 AV |
1333 | |
1334 | return 0; | |
62a8067a AV |
1335 | } |
1336 | EXPORT_SYMBOL(iov_iter_alignment); | |
1337 | ||
357f435d AV |
1338 | unsigned long iov_iter_gap_alignment(const struct iov_iter *i) |
1339 | { | |
33844e66 | 1340 | unsigned long res = 0; |
610c7a71 | 1341 | unsigned long v = 0; |
357f435d | 1342 | size_t size = i->count; |
610c7a71 | 1343 | unsigned k; |
357f435d | 1344 | |
610c7a71 | 1345 | if (WARN_ON(!iter_is_iovec(i))) |
241699cd | 1346 | return ~0U; |
241699cd | 1347 | |
610c7a71 AV |
1348 | for (k = 0; k < i->nr_segs; k++) { |
1349 | if (i->iov[k].iov_len) { | |
1350 | unsigned long base = (unsigned long)i->iov[k].iov_base; | |
1351 | if (v) // if not the first one | |
1352 | res |= base | v; // this start | previous end | |
1353 | v = base + i->iov[k].iov_len; | |
1354 | if (size <= i->iov[k].iov_len) | |
1355 | break; | |
1356 | size -= i->iov[k].iov_len; | |
1357 | } | |
1358 | } | |
33844e66 | 1359 | return res; |
357f435d AV |
1360 | } |
1361 | EXPORT_SYMBOL(iov_iter_gap_alignment); | |
1362 | ||
e76b6312 | 1363 | static inline ssize_t __pipe_get_pages(struct iov_iter *i, |
241699cd AV |
1364 | size_t maxsize, |
1365 | struct page **pages, | |
8cefc107 | 1366 | int iter_head, |
241699cd AV |
1367 | size_t *start) |
1368 | { | |
1369 | struct pipe_inode_info *pipe = i->pipe; | |
8cefc107 DH |
1370 | unsigned int p_mask = pipe->ring_size - 1; |
1371 | ssize_t n = push_pipe(i, maxsize, &iter_head, start); | |
241699cd AV |
1372 | if (!n) |
1373 | return -EFAULT; | |
1374 | ||
1375 | maxsize = n; | |
1376 | n += *start; | |
1689c73a | 1377 | while (n > 0) { |
8cefc107 DH |
1378 | get_page(*pages++ = pipe->bufs[iter_head & p_mask].page); |
1379 | iter_head++; | |
241699cd AV |
1380 | n -= PAGE_SIZE; |
1381 | } | |
1382 | ||
1383 | return maxsize; | |
1384 | } | |
1385 | ||
1386 | static ssize_t pipe_get_pages(struct iov_iter *i, | |
1387 | struct page **pages, size_t maxsize, unsigned maxpages, | |
1388 | size_t *start) | |
1389 | { | |
8cefc107 | 1390 | unsigned int iter_head, npages; |
241699cd | 1391 | size_t capacity; |
241699cd AV |
1392 | |
1393 | if (!sanity(i)) | |
1394 | return -EFAULT; | |
1395 | ||
8cefc107 DH |
1396 | data_start(i, &iter_head, start); |
1397 | /* Amount of free space: some of this one + all after this one */ | |
1398 | npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe); | |
1399 | capacity = min(npages, maxpages) * PAGE_SIZE - *start; | |
241699cd | 1400 | |
8cefc107 | 1401 | return __pipe_get_pages(i, min(maxsize, capacity), pages, iter_head, start); |
241699cd AV |
1402 | } |
1403 | ||
7ff50620 DH |
1404 | static ssize_t iter_xarray_populate_pages(struct page **pages, struct xarray *xa, |
1405 | pgoff_t index, unsigned int nr_pages) | |
1406 | { | |
1407 | XA_STATE(xas, xa, index); | |
1408 | struct page *page; | |
1409 | unsigned int ret = 0; | |
1410 | ||
1411 | rcu_read_lock(); | |
1412 | for (page = xas_load(&xas); page; page = xas_next(&xas)) { | |
1413 | if (xas_retry(&xas, page)) | |
1414 | continue; | |
1415 | ||
1416 | /* Has the page moved or been split? */ | |
1417 | if (unlikely(page != xas_reload(&xas))) { | |
1418 | xas_reset(&xas); | |
1419 | continue; | |
1420 | } | |
1421 | ||
1422 | pages[ret] = find_subpage(page, xas.xa_index); | |
1423 | get_page(pages[ret]); | |
1424 | if (++ret == nr_pages) | |
1425 | break; | |
1426 | } | |
1427 | rcu_read_unlock(); | |
1428 | return ret; | |
1429 | } | |
1430 | ||
1431 | static ssize_t iter_xarray_get_pages(struct iov_iter *i, | |
1432 | struct page **pages, size_t maxsize, | |
1433 | unsigned maxpages, size_t *_start_offset) | |
1434 | { | |
1435 | unsigned nr, offset; | |
1436 | pgoff_t index, count; | |
1437 | size_t size = maxsize, actual; | |
1438 | loff_t pos; | |
1439 | ||
1440 | if (!size || !maxpages) | |
1441 | return 0; | |
1442 | ||
1443 | pos = i->xarray_start + i->iov_offset; | |
1444 | index = pos >> PAGE_SHIFT; | |
1445 | offset = pos & ~PAGE_MASK; | |
1446 | *_start_offset = offset; | |
1447 | ||
1448 | count = 1; | |
1449 | if (size > PAGE_SIZE - offset) { | |
1450 | size -= PAGE_SIZE - offset; | |
1451 | count += size >> PAGE_SHIFT; | |
1452 | size &= ~PAGE_MASK; | |
1453 | if (size) | |
1454 | count++; | |
1455 | } | |
1456 | ||
1457 | if (count > maxpages) | |
1458 | count = maxpages; | |
1459 | ||
1460 | nr = iter_xarray_populate_pages(pages, i->xarray, index, count); | |
1461 | if (nr == 0) | |
1462 | return 0; | |
1463 | ||
1464 | actual = PAGE_SIZE * nr; | |
1465 | actual -= offset; | |
1466 | if (nr == count && size > 0) { | |
1467 | unsigned last_offset = (nr > 1) ? 0 : offset; | |
1468 | actual -= PAGE_SIZE - (last_offset + size); | |
1469 | } | |
1470 | return actual; | |
1471 | } | |
1472 | ||
3d671ca6 AV |
1473 | /* must be done on non-empty ITER_IOVEC one */ |
1474 | static unsigned long first_iovec_segment(const struct iov_iter *i, | |
1475 | size_t *size, size_t *start, | |
1476 | size_t maxsize, unsigned maxpages) | |
1477 | { | |
1478 | size_t skip; | |
1479 | long k; | |
1480 | ||
1481 | for (k = 0, skip = i->iov_offset; k < i->nr_segs; k++, skip = 0) { | |
1482 | unsigned long addr = (unsigned long)i->iov[k].iov_base + skip; | |
1483 | size_t len = i->iov[k].iov_len - skip; | |
1484 | ||
1485 | if (unlikely(!len)) | |
1486 | continue; | |
1487 | if (len > maxsize) | |
1488 | len = maxsize; | |
1489 | len += (*start = addr % PAGE_SIZE); | |
1490 | if (len > maxpages * PAGE_SIZE) | |
1491 | len = maxpages * PAGE_SIZE; | |
1492 | *size = len; | |
1493 | return addr & PAGE_MASK; | |
1494 | } | |
1495 | BUG(); // if it had been empty, we wouldn't get called | |
1496 | } | |
1497 | ||
1498 | /* must be done on non-empty ITER_BVEC one */ | |
1499 | static struct page *first_bvec_segment(const struct iov_iter *i, | |
1500 | size_t *size, size_t *start, | |
1501 | size_t maxsize, unsigned maxpages) | |
1502 | { | |
1503 | struct page *page; | |
1504 | size_t skip = i->iov_offset, len; | |
1505 | ||
1506 | len = i->bvec->bv_len - skip; | |
1507 | if (len > maxsize) | |
1508 | len = maxsize; | |
1509 | skip += i->bvec->bv_offset; | |
1510 | page = i->bvec->bv_page + skip / PAGE_SIZE; | |
1511 | len += (*start = skip % PAGE_SIZE); | |
1512 | if (len > maxpages * PAGE_SIZE) | |
1513 | len = maxpages * PAGE_SIZE; | |
1514 | *size = len; | |
1515 | return page; | |
1516 | } | |
1517 | ||
62a8067a | 1518 | ssize_t iov_iter_get_pages(struct iov_iter *i, |
2c80929c | 1519 | struct page **pages, size_t maxsize, unsigned maxpages, |
62a8067a AV |
1520 | size_t *start) |
1521 | { | |
3d671ca6 AV |
1522 | size_t len; |
1523 | int n, res; | |
1524 | ||
e5393fae AV |
1525 | if (maxsize > i->count) |
1526 | maxsize = i->count; | |
3d671ca6 AV |
1527 | if (!maxsize) |
1528 | return 0; | |
e5393fae | 1529 | |
3d671ca6 | 1530 | if (likely(iter_is_iovec(i))) { |
3337ab08 | 1531 | unsigned int gup_flags = 0; |
3d671ca6 | 1532 | unsigned long addr; |
e5393fae | 1533 | |
3337ab08 AG |
1534 | if (iov_iter_rw(i) != WRITE) |
1535 | gup_flags |= FOLL_WRITE; | |
1536 | if (i->nofault) | |
1537 | gup_flags |= FOLL_NOFAULT; | |
1538 | ||
3d671ca6 | 1539 | addr = first_iovec_segment(i, &len, start, maxsize, maxpages); |
e5393fae | 1540 | n = DIV_ROUND_UP(len, PAGE_SIZE); |
3337ab08 | 1541 | res = get_user_pages_fast(addr, n, gup_flags, pages); |
814a6674 | 1542 | if (unlikely(res <= 0)) |
e5393fae AV |
1543 | return res; |
1544 | return (res == n ? len : res * PAGE_SIZE) - *start; | |
3d671ca6 AV |
1545 | } |
1546 | if (iov_iter_is_bvec(i)) { | |
1547 | struct page *page; | |
1548 | ||
1549 | page = first_bvec_segment(i, &len, start, maxsize, maxpages); | |
1550 | n = DIV_ROUND_UP(len, PAGE_SIZE); | |
1551 | while (n--) | |
1552 | get_page(*pages++ = page++); | |
1553 | return len - *start; | |
1554 | } | |
1555 | if (iov_iter_is_pipe(i)) | |
1556 | return pipe_get_pages(i, pages, maxsize, maxpages, start); | |
1557 | if (iov_iter_is_xarray(i)) | |
1558 | return iter_xarray_get_pages(i, pages, maxsize, maxpages, start); | |
1559 | return -EFAULT; | |
62a8067a AV |
1560 | } |
1561 | EXPORT_SYMBOL(iov_iter_get_pages); | |
1562 | ||
1b17f1f2 AV |
1563 | static struct page **get_pages_array(size_t n) |
1564 | { | |
752ade68 | 1565 | return kvmalloc_array(n, sizeof(struct page *), GFP_KERNEL); |
1b17f1f2 AV |
1566 | } |
1567 | ||
241699cd AV |
1568 | static ssize_t pipe_get_pages_alloc(struct iov_iter *i, |
1569 | struct page ***pages, size_t maxsize, | |
1570 | size_t *start) | |
1571 | { | |
1572 | struct page **p; | |
8cefc107 | 1573 | unsigned int iter_head, npages; |
d7760d63 | 1574 | ssize_t n; |
241699cd AV |
1575 | |
1576 | if (!sanity(i)) | |
1577 | return -EFAULT; | |
1578 | ||
8cefc107 DH |
1579 | data_start(i, &iter_head, start); |
1580 | /* Amount of free space: some of this one + all after this one */ | |
1581 | npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe); | |
241699cd AV |
1582 | n = npages * PAGE_SIZE - *start; |
1583 | if (maxsize > n) | |
1584 | maxsize = n; | |
1585 | else | |
1586 | npages = DIV_ROUND_UP(maxsize + *start, PAGE_SIZE); | |
1587 | p = get_pages_array(npages); | |
1588 | if (!p) | |
1589 | return -ENOMEM; | |
8cefc107 | 1590 | n = __pipe_get_pages(i, maxsize, p, iter_head, start); |
241699cd AV |
1591 | if (n > 0) |
1592 | *pages = p; | |
1593 | else | |
1594 | kvfree(p); | |
1595 | return n; | |
1596 | } | |
1597 | ||
7ff50620 DH |
1598 | static ssize_t iter_xarray_get_pages_alloc(struct iov_iter *i, |
1599 | struct page ***pages, size_t maxsize, | |
1600 | size_t *_start_offset) | |
1601 | { | |
1602 | struct page **p; | |
1603 | unsigned nr, offset; | |
1604 | pgoff_t index, count; | |
1605 | size_t size = maxsize, actual; | |
1606 | loff_t pos; | |
1607 | ||
1608 | if (!size) | |
1609 | return 0; | |
1610 | ||
1611 | pos = i->xarray_start + i->iov_offset; | |
1612 | index = pos >> PAGE_SHIFT; | |
1613 | offset = pos & ~PAGE_MASK; | |
1614 | *_start_offset = offset; | |
1615 | ||
1616 | count = 1; | |
1617 | if (size > PAGE_SIZE - offset) { | |
1618 | size -= PAGE_SIZE - offset; | |
1619 | count += size >> PAGE_SHIFT; | |
1620 | size &= ~PAGE_MASK; | |
1621 | if (size) | |
1622 | count++; | |
1623 | } | |
1624 | ||
1625 | p = get_pages_array(count); | |
1626 | if (!p) | |
1627 | return -ENOMEM; | |
1628 | *pages = p; | |
1629 | ||
1630 | nr = iter_xarray_populate_pages(p, i->xarray, index, count); | |
1631 | if (nr == 0) | |
1632 | return 0; | |
1633 | ||
1634 | actual = PAGE_SIZE * nr; | |
1635 | actual -= offset; | |
1636 | if (nr == count && size > 0) { | |
1637 | unsigned last_offset = (nr > 1) ? 0 : offset; | |
1638 | actual -= PAGE_SIZE - (last_offset + size); | |
1639 | } | |
1640 | return actual; | |
1641 | } | |
1642 | ||
62a8067a AV |
1643 | ssize_t iov_iter_get_pages_alloc(struct iov_iter *i, |
1644 | struct page ***pages, size_t maxsize, | |
1645 | size_t *start) | |
1646 | { | |
1b17f1f2 | 1647 | struct page **p; |
3d671ca6 AV |
1648 | size_t len; |
1649 | int n, res; | |
1b17f1f2 AV |
1650 | |
1651 | if (maxsize > i->count) | |
1652 | maxsize = i->count; | |
3d671ca6 AV |
1653 | if (!maxsize) |
1654 | return 0; | |
1b17f1f2 | 1655 | |
3d671ca6 | 1656 | if (likely(iter_is_iovec(i))) { |
3337ab08 | 1657 | unsigned int gup_flags = 0; |
3d671ca6 | 1658 | unsigned long addr; |
1b17f1f2 | 1659 | |
3337ab08 AG |
1660 | if (iov_iter_rw(i) != WRITE) |
1661 | gup_flags |= FOLL_WRITE; | |
1662 | if (i->nofault) | |
1663 | gup_flags |= FOLL_NOFAULT; | |
1664 | ||
3d671ca6 | 1665 | addr = first_iovec_segment(i, &len, start, maxsize, ~0U); |
1b17f1f2 AV |
1666 | n = DIV_ROUND_UP(len, PAGE_SIZE); |
1667 | p = get_pages_array(n); | |
1668 | if (!p) | |
1669 | return -ENOMEM; | |
3337ab08 | 1670 | res = get_user_pages_fast(addr, n, gup_flags, p); |
814a6674 | 1671 | if (unlikely(res <= 0)) { |
1b17f1f2 | 1672 | kvfree(p); |
814a6674 | 1673 | *pages = NULL; |
1b17f1f2 AV |
1674 | return res; |
1675 | } | |
1676 | *pages = p; | |
1677 | return (res == n ? len : res * PAGE_SIZE) - *start; | |
3d671ca6 AV |
1678 | } |
1679 | if (iov_iter_is_bvec(i)) { | |
1680 | struct page *page; | |
1681 | ||
1682 | page = first_bvec_segment(i, &len, start, maxsize, ~0U); | |
1683 | n = DIV_ROUND_UP(len, PAGE_SIZE); | |
1684 | *pages = p = get_pages_array(n); | |
1b17f1f2 AV |
1685 | if (!p) |
1686 | return -ENOMEM; | |
3d671ca6 AV |
1687 | while (n--) |
1688 | get_page(*p++ = page++); | |
1689 | return len - *start; | |
1690 | } | |
1691 | if (iov_iter_is_pipe(i)) | |
1692 | return pipe_get_pages_alloc(i, pages, maxsize, start); | |
1693 | if (iov_iter_is_xarray(i)) | |
1694 | return iter_xarray_get_pages_alloc(i, pages, maxsize, start); | |
1695 | return -EFAULT; | |
62a8067a AV |
1696 | } |
1697 | EXPORT_SYMBOL(iov_iter_get_pages_alloc); | |
1698 | ||
a604ec7e AV |
1699 | size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum, |
1700 | struct iov_iter *i) | |
1701 | { | |
a604ec7e | 1702 | __wsum sum, next; |
a604ec7e | 1703 | sum = *csum; |
9ea9ce04 | 1704 | if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) { |
241699cd AV |
1705 | WARN_ON(1); |
1706 | return 0; | |
1707 | } | |
7baa5099 AV |
1708 | iterate_and_advance(i, bytes, base, len, off, ({ |
1709 | next = csum_and_copy_from_user(base, addr + off, len); | |
2495bdcc | 1710 | sum = csum_block_add(sum, next, off); |
7baa5099 | 1711 | next ? 0 : len; |
a604ec7e | 1712 | }), ({ |
7baa5099 | 1713 | sum = csum_and_memcpy(addr + off, base, len, sum, off); |
a604ec7e AV |
1714 | }) |
1715 | ) | |
1716 | *csum = sum; | |
1717 | return bytes; | |
1718 | } | |
1719 | EXPORT_SYMBOL(csum_and_copy_from_iter); | |
1720 | ||
52cbd23a | 1721 | size_t csum_and_copy_to_iter(const void *addr, size_t bytes, void *_csstate, |
a604ec7e AV |
1722 | struct iov_iter *i) |
1723 | { | |
52cbd23a | 1724 | struct csum_state *csstate = _csstate; |
a604ec7e | 1725 | __wsum sum, next; |
78e1f386 | 1726 | |
78e1f386 | 1727 | if (unlikely(iov_iter_is_discard(i))) { |
241699cd AV |
1728 | WARN_ON(1); /* for now */ |
1729 | return 0; | |
1730 | } | |
6852df12 AV |
1731 | |
1732 | sum = csum_shift(csstate->csum, csstate->off); | |
1733 | if (unlikely(iov_iter_is_pipe(i))) | |
1734 | bytes = csum_and_copy_to_pipe_iter(addr, bytes, i, &sum); | |
1735 | else iterate_and_advance(i, bytes, base, len, off, ({ | |
7baa5099 | 1736 | next = csum_and_copy_to_user(addr + off, base, len); |
2495bdcc | 1737 | sum = csum_block_add(sum, next, off); |
7baa5099 | 1738 | next ? 0 : len; |
a604ec7e | 1739 | }), ({ |
7baa5099 | 1740 | sum = csum_and_memcpy(base, addr + off, len, sum, off); |
a604ec7e AV |
1741 | }) |
1742 | ) | |
594e450b AV |
1743 | csstate->csum = csum_shift(sum, csstate->off); |
1744 | csstate->off += bytes; | |
a604ec7e AV |
1745 | return bytes; |
1746 | } | |
1747 | EXPORT_SYMBOL(csum_and_copy_to_iter); | |
1748 | ||
d05f4435 SG |
1749 | size_t hash_and_copy_to_iter(const void *addr, size_t bytes, void *hashp, |
1750 | struct iov_iter *i) | |
1751 | { | |
7999096f | 1752 | #ifdef CONFIG_CRYPTO_HASH |
d05f4435 SG |
1753 | struct ahash_request *hash = hashp; |
1754 | struct scatterlist sg; | |
1755 | size_t copied; | |
1756 | ||
1757 | copied = copy_to_iter(addr, bytes, i); | |
1758 | sg_init_one(&sg, addr, copied); | |
1759 | ahash_request_set_crypt(hash, &sg, NULL, copied); | |
1760 | crypto_ahash_update(hash); | |
1761 | return copied; | |
27fad74a Y |
1762 | #else |
1763 | return 0; | |
1764 | #endif | |
d05f4435 SG |
1765 | } |
1766 | EXPORT_SYMBOL(hash_and_copy_to_iter); | |
1767 | ||
66531c65 | 1768 | static int iov_npages(const struct iov_iter *i, int maxpages) |
62a8067a | 1769 | { |
66531c65 AV |
1770 | size_t skip = i->iov_offset, size = i->count; |
1771 | const struct iovec *p; | |
e0f2dc40 AV |
1772 | int npages = 0; |
1773 | ||
66531c65 AV |
1774 | for (p = i->iov; size; skip = 0, p++) { |
1775 | unsigned offs = offset_in_page(p->iov_base + skip); | |
1776 | size_t len = min(p->iov_len - skip, size); | |
e0f2dc40 | 1777 | |
66531c65 AV |
1778 | if (len) { |
1779 | size -= len; | |
1780 | npages += DIV_ROUND_UP(offs + len, PAGE_SIZE); | |
1781 | if (unlikely(npages > maxpages)) | |
1782 | return maxpages; | |
1783 | } | |
1784 | } | |
1785 | return npages; | |
1786 | } | |
1787 | ||
1788 | static int bvec_npages(const struct iov_iter *i, int maxpages) | |
1789 | { | |
1790 | size_t skip = i->iov_offset, size = i->count; | |
1791 | const struct bio_vec *p; | |
1792 | int npages = 0; | |
1793 | ||
1794 | for (p = i->bvec; size; skip = 0, p++) { | |
1795 | unsigned offs = (p->bv_offset + skip) % PAGE_SIZE; | |
1796 | size_t len = min(p->bv_len - skip, size); | |
1797 | ||
1798 | size -= len; | |
1799 | npages += DIV_ROUND_UP(offs + len, PAGE_SIZE); | |
1800 | if (unlikely(npages > maxpages)) | |
1801 | return maxpages; | |
1802 | } | |
1803 | return npages; | |
1804 | } | |
1805 | ||
1806 | int iov_iter_npages(const struct iov_iter *i, int maxpages) | |
1807 | { | |
1808 | if (unlikely(!i->count)) | |
1809 | return 0; | |
1810 | /* iovec and kvec have identical layouts */ | |
1811 | if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) | |
1812 | return iov_npages(i, maxpages); | |
1813 | if (iov_iter_is_bvec(i)) | |
1814 | return bvec_npages(i, maxpages); | |
1815 | if (iov_iter_is_pipe(i)) { | |
8cefc107 | 1816 | unsigned int iter_head; |
66531c65 | 1817 | int npages; |
241699cd | 1818 | size_t off; |
241699cd AV |
1819 | |
1820 | if (!sanity(i)) | |
1821 | return 0; | |
1822 | ||
8cefc107 | 1823 | data_start(i, &iter_head, &off); |
241699cd | 1824 | /* some of this one + all after this one */ |
66531c65 AV |
1825 | npages = pipe_space_for_user(iter_head, i->pipe->tail, i->pipe); |
1826 | return min(npages, maxpages); | |
1827 | } | |
1828 | if (iov_iter_is_xarray(i)) { | |
e4f8df86 AV |
1829 | unsigned offset = (i->xarray_start + i->iov_offset) % PAGE_SIZE; |
1830 | int npages = DIV_ROUND_UP(offset + i->count, PAGE_SIZE); | |
66531c65 AV |
1831 | return min(npages, maxpages); |
1832 | } | |
1833 | return 0; | |
62a8067a | 1834 | } |
f67da30c | 1835 | EXPORT_SYMBOL(iov_iter_npages); |
4b8164b9 AV |
1836 | |
1837 | const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags) | |
1838 | { | |
1839 | *new = *old; | |
00e23707 | 1840 | if (unlikely(iov_iter_is_pipe(new))) { |
241699cd AV |
1841 | WARN_ON(1); |
1842 | return NULL; | |
1843 | } | |
7ff50620 | 1844 | if (unlikely(iov_iter_is_discard(new) || iov_iter_is_xarray(new))) |
9ea9ce04 | 1845 | return NULL; |
00e23707 | 1846 | if (iov_iter_is_bvec(new)) |
4b8164b9 AV |
1847 | return new->bvec = kmemdup(new->bvec, |
1848 | new->nr_segs * sizeof(struct bio_vec), | |
1849 | flags); | |
1850 | else | |
1851 | /* iovec and kvec have identical layout */ | |
1852 | return new->iov = kmemdup(new->iov, | |
1853 | new->nr_segs * sizeof(struct iovec), | |
1854 | flags); | |
1855 | } | |
1856 | EXPORT_SYMBOL(dup_iter); | |
bc917be8 | 1857 | |
bfdc5970 CH |
1858 | static int copy_compat_iovec_from_user(struct iovec *iov, |
1859 | const struct iovec __user *uvec, unsigned long nr_segs) | |
1860 | { | |
1861 | const struct compat_iovec __user *uiov = | |
1862 | (const struct compat_iovec __user *)uvec; | |
1863 | int ret = -EFAULT, i; | |
1864 | ||
a959a978 | 1865 | if (!user_access_begin(uiov, nr_segs * sizeof(*uiov))) |
bfdc5970 CH |
1866 | return -EFAULT; |
1867 | ||
1868 | for (i = 0; i < nr_segs; i++) { | |
1869 | compat_uptr_t buf; | |
1870 | compat_ssize_t len; | |
1871 | ||
1872 | unsafe_get_user(len, &uiov[i].iov_len, uaccess_end); | |
1873 | unsafe_get_user(buf, &uiov[i].iov_base, uaccess_end); | |
1874 | ||
1875 | /* check for compat_size_t not fitting in compat_ssize_t .. */ | |
1876 | if (len < 0) { | |
1877 | ret = -EINVAL; | |
1878 | goto uaccess_end; | |
1879 | } | |
1880 | iov[i].iov_base = compat_ptr(buf); | |
1881 | iov[i].iov_len = len; | |
1882 | } | |
1883 | ||
1884 | ret = 0; | |
1885 | uaccess_end: | |
1886 | user_access_end(); | |
1887 | return ret; | |
1888 | } | |
1889 | ||
1890 | static int copy_iovec_from_user(struct iovec *iov, | |
1891 | const struct iovec __user *uvec, unsigned long nr_segs) | |
fb041b59 DL |
1892 | { |
1893 | unsigned long seg; | |
fb041b59 | 1894 | |
bfdc5970 CH |
1895 | if (copy_from_user(iov, uvec, nr_segs * sizeof(*uvec))) |
1896 | return -EFAULT; | |
1897 | for (seg = 0; seg < nr_segs; seg++) { | |
1898 | if ((ssize_t)iov[seg].iov_len < 0) | |
1899 | return -EINVAL; | |
fb041b59 DL |
1900 | } |
1901 | ||
bfdc5970 CH |
1902 | return 0; |
1903 | } | |
1904 | ||
1905 | struct iovec *iovec_from_user(const struct iovec __user *uvec, | |
1906 | unsigned long nr_segs, unsigned long fast_segs, | |
1907 | struct iovec *fast_iov, bool compat) | |
1908 | { | |
1909 | struct iovec *iov = fast_iov; | |
1910 | int ret; | |
1911 | ||
fb041b59 | 1912 | /* |
bfdc5970 CH |
1913 | * SuS says "The readv() function *may* fail if the iovcnt argument was |
1914 | * less than or equal to 0, or greater than {IOV_MAX}. Linux has | |
1915 | * traditionally returned zero for zero segments, so... | |
fb041b59 | 1916 | */ |
bfdc5970 CH |
1917 | if (nr_segs == 0) |
1918 | return iov; | |
1919 | if (nr_segs > UIO_MAXIOV) | |
1920 | return ERR_PTR(-EINVAL); | |
fb041b59 DL |
1921 | if (nr_segs > fast_segs) { |
1922 | iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL); | |
bfdc5970 CH |
1923 | if (!iov) |
1924 | return ERR_PTR(-ENOMEM); | |
fb041b59 | 1925 | } |
bfdc5970 CH |
1926 | |
1927 | if (compat) | |
1928 | ret = copy_compat_iovec_from_user(iov, uvec, nr_segs); | |
1929 | else | |
1930 | ret = copy_iovec_from_user(iov, uvec, nr_segs); | |
1931 | if (ret) { | |
1932 | if (iov != fast_iov) | |
1933 | kfree(iov); | |
1934 | return ERR_PTR(ret); | |
1935 | } | |
1936 | ||
1937 | return iov; | |
1938 | } | |
1939 | ||
1940 | ssize_t __import_iovec(int type, const struct iovec __user *uvec, | |
1941 | unsigned nr_segs, unsigned fast_segs, struct iovec **iovp, | |
1942 | struct iov_iter *i, bool compat) | |
1943 | { | |
1944 | ssize_t total_len = 0; | |
1945 | unsigned long seg; | |
1946 | struct iovec *iov; | |
1947 | ||
1948 | iov = iovec_from_user(uvec, nr_segs, fast_segs, *iovp, compat); | |
1949 | if (IS_ERR(iov)) { | |
1950 | *iovp = NULL; | |
1951 | return PTR_ERR(iov); | |
fb041b59 DL |
1952 | } |
1953 | ||
1954 | /* | |
bfdc5970 CH |
1955 | * According to the Single Unix Specification we should return EINVAL if |
1956 | * an element length is < 0 when cast to ssize_t or if the total length | |
1957 | * would overflow the ssize_t return value of the system call. | |
fb041b59 DL |
1958 | * |
1959 | * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the | |
1960 | * overflow case. | |
1961 | */ | |
fb041b59 | 1962 | for (seg = 0; seg < nr_segs; seg++) { |
fb041b59 DL |
1963 | ssize_t len = (ssize_t)iov[seg].iov_len; |
1964 | ||
bfdc5970 CH |
1965 | if (!access_ok(iov[seg].iov_base, len)) { |
1966 | if (iov != *iovp) | |
1967 | kfree(iov); | |
1968 | *iovp = NULL; | |
1969 | return -EFAULT; | |
fb041b59 | 1970 | } |
bfdc5970 CH |
1971 | |
1972 | if (len > MAX_RW_COUNT - total_len) { | |
1973 | len = MAX_RW_COUNT - total_len; | |
fb041b59 DL |
1974 | iov[seg].iov_len = len; |
1975 | } | |
bfdc5970 | 1976 | total_len += len; |
fb041b59 | 1977 | } |
bfdc5970 CH |
1978 | |
1979 | iov_iter_init(i, type, iov, nr_segs, total_len); | |
1980 | if (iov == *iovp) | |
1981 | *iovp = NULL; | |
1982 | else | |
1983 | *iovp = iov; | |
1984 | return total_len; | |
fb041b59 DL |
1985 | } |
1986 | ||
ffecee4f VN |
1987 | /** |
1988 | * import_iovec() - Copy an array of &struct iovec from userspace | |
1989 | * into the kernel, check that it is valid, and initialize a new | |
1990 | * &struct iov_iter iterator to access it. | |
1991 | * | |
1992 | * @type: One of %READ or %WRITE. | |
bfdc5970 | 1993 | * @uvec: Pointer to the userspace array. |
ffecee4f VN |
1994 | * @nr_segs: Number of elements in userspace array. |
1995 | * @fast_segs: Number of elements in @iov. | |
bfdc5970 | 1996 | * @iovp: (input and output parameter) Pointer to pointer to (usually small |
ffecee4f VN |
1997 | * on-stack) kernel array. |
1998 | * @i: Pointer to iterator that will be initialized on success. | |
1999 | * | |
2000 | * If the array pointed to by *@iov is large enough to hold all @nr_segs, | |
2001 | * then this function places %NULL in *@iov on return. Otherwise, a new | |
2002 | * array will be allocated and the result placed in *@iov. This means that | |
2003 | * the caller may call kfree() on *@iov regardless of whether the small | |
2004 | * on-stack array was used or not (and regardless of whether this function | |
2005 | * returns an error or not). | |
2006 | * | |
87e5e6da | 2007 | * Return: Negative error code on error, bytes imported on success |
ffecee4f | 2008 | */ |
bfdc5970 | 2009 | ssize_t import_iovec(int type, const struct iovec __user *uvec, |
bc917be8 | 2010 | unsigned nr_segs, unsigned fast_segs, |
bfdc5970 | 2011 | struct iovec **iovp, struct iov_iter *i) |
bc917be8 | 2012 | { |
89cd35c5 CH |
2013 | return __import_iovec(type, uvec, nr_segs, fast_segs, iovp, i, |
2014 | in_compat_syscall()); | |
bc917be8 AV |
2015 | } |
2016 | EXPORT_SYMBOL(import_iovec); | |
2017 | ||
bc917be8 AV |
2018 | int import_single_range(int rw, void __user *buf, size_t len, |
2019 | struct iovec *iov, struct iov_iter *i) | |
2020 | { | |
2021 | if (len > MAX_RW_COUNT) | |
2022 | len = MAX_RW_COUNT; | |
96d4f267 | 2023 | if (unlikely(!access_ok(buf, len))) |
bc917be8 AV |
2024 | return -EFAULT; |
2025 | ||
2026 | iov->iov_base = buf; | |
2027 | iov->iov_len = len; | |
2028 | iov_iter_init(i, rw, iov, 1, len); | |
2029 | return 0; | |
2030 | } | |
e1267585 | 2031 | EXPORT_SYMBOL(import_single_range); |
8fb0f47a JA |
2032 | |
2033 | /** | |
2034 | * iov_iter_restore() - Restore a &struct iov_iter to the same state as when | |
2035 | * iov_iter_save_state() was called. | |
2036 | * | |
2037 | * @i: &struct iov_iter to restore | |
2038 | * @state: state to restore from | |
2039 | * | |
2040 | * Used after iov_iter_save_state() to bring restore @i, if operations may | |
2041 | * have advanced it. | |
2042 | * | |
2043 | * Note: only works on ITER_IOVEC, ITER_BVEC, and ITER_KVEC | |
2044 | */ | |
2045 | void iov_iter_restore(struct iov_iter *i, struct iov_iter_state *state) | |
2046 | { | |
2047 | if (WARN_ON_ONCE(!iov_iter_is_bvec(i) && !iter_is_iovec(i)) && | |
2048 | !iov_iter_is_kvec(i)) | |
2049 | return; | |
2050 | i->iov_offset = state->iov_offset; | |
2051 | i->count = state->count; | |
2052 | /* | |
2053 | * For the *vec iters, nr_segs + iov is constant - if we increment | |
2054 | * the vec, then we also decrement the nr_segs count. Hence we don't | |
2055 | * need to track both of these, just one is enough and we can deduct | |
2056 | * the other from that. ITER_KVEC and ITER_IOVEC are the same struct | |
2057 | * size, so we can just increment the iov pointer as they are unionzed. | |
2058 | * ITER_BVEC _may_ be the same size on some archs, but on others it is | |
2059 | * not. Be safe and handle it separately. | |
2060 | */ | |
2061 | BUILD_BUG_ON(sizeof(struct iovec) != sizeof(struct kvec)); | |
2062 | if (iov_iter_is_bvec(i)) | |
2063 | i->bvec -= state->nr_segs - i->nr_segs; | |
2064 | else | |
2065 | i->iov -= state->nr_segs - i->nr_segs; | |
2066 | i->nr_segs = state->nr_segs; | |
2067 | } |