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))) { \ | |
de4f5fed | 129 | const struct iovec *iov = iter_iov(i); \ |
7baa5099 AV |
130 | void __user *base; \ |
131 | size_t len; \ | |
132 | iterate_iovec(i, n, base, len, off, \ | |
a6e4ec7b | 133 | iov, (I)) \ |
de4f5fed JA |
134 | i->nr_segs -= iov - iter_iov(i); \ |
135 | i->__iov = iov; \ | |
28f38db7 | 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 | ||
4f80818b LS |
175 | static int copyout_nofault(void __user *to, const void *from, size_t n) |
176 | { | |
177 | long res; | |
178 | ||
179 | if (should_fail_usercopy()) | |
180 | return n; | |
181 | ||
182 | res = copy_to_user_nofault(to, from, n); | |
183 | ||
184 | return res < 0 ? n : res; | |
185 | } | |
186 | ||
09fc68dc AV |
187 | static int copyin(void *to, const void __user *from, size_t n) |
188 | { | |
33b75c1d AP |
189 | size_t res = n; |
190 | ||
4d0e9df5 AL |
191 | if (should_fail_usercopy()) |
192 | return n; | |
96d4f267 | 193 | if (access_ok(from, n)) { |
33b75c1d AP |
194 | instrument_copy_from_user_before(to, from, n); |
195 | res = raw_copy_from_user(to, from, n); | |
196 | instrument_copy_from_user_after(to, from, n, res); | |
09fc68dc | 197 | } |
33b75c1d | 198 | return res; |
09fc68dc AV |
199 | } |
200 | ||
241699cd AV |
201 | #ifdef PIPE_PARANOIA |
202 | static bool sanity(const struct iov_iter *i) | |
203 | { | |
204 | struct pipe_inode_info *pipe = i->pipe; | |
8cefc107 DH |
205 | unsigned int p_head = pipe->head; |
206 | unsigned int p_tail = pipe->tail; | |
8cefc107 DH |
207 | unsigned int p_occupancy = pipe_occupancy(p_head, p_tail); |
208 | unsigned int i_head = i->head; | |
209 | unsigned int idx; | |
210 | ||
10f525a8 | 211 | if (i->last_offset) { |
241699cd | 212 | struct pipe_buffer *p; |
8cefc107 | 213 | if (unlikely(p_occupancy == 0)) |
241699cd | 214 | goto Bad; // pipe must be non-empty |
8cefc107 | 215 | if (unlikely(i_head != p_head - 1)) |
241699cd AV |
216 | goto Bad; // must be at the last buffer... |
217 | ||
2dcedb2a | 218 | p = pipe_buf(pipe, i_head); |
10f525a8 | 219 | if (unlikely(p->offset + p->len != abs(i->last_offset))) |
241699cd AV |
220 | goto Bad; // ... at the end of segment |
221 | } else { | |
8cefc107 | 222 | if (i_head != p_head) |
241699cd AV |
223 | goto Bad; // must be right after the last buffer |
224 | } | |
225 | return true; | |
226 | Bad: | |
10f525a8 | 227 | printk(KERN_ERR "idx = %d, offset = %d\n", i_head, i->last_offset); |
8cefc107 DH |
228 | printk(KERN_ERR "head = %d, tail = %d, buffers = %d\n", |
229 | p_head, p_tail, pipe->ring_size); | |
230 | for (idx = 0; idx < pipe->ring_size; idx++) | |
241699cd AV |
231 | printk(KERN_ERR "[%p %p %d %d]\n", |
232 | pipe->bufs[idx].ops, | |
233 | pipe->bufs[idx].page, | |
234 | pipe->bufs[idx].offset, | |
235 | pipe->bufs[idx].len); | |
236 | WARN_ON(1); | |
237 | return false; | |
238 | } | |
239 | #else | |
240 | #define sanity(i) true | |
241 | #endif | |
242 | ||
47b7fcae AV |
243 | static struct page *push_anon(struct pipe_inode_info *pipe, unsigned size) |
244 | { | |
245 | struct page *page = alloc_page(GFP_USER); | |
246 | if (page) { | |
247 | struct pipe_buffer *buf = pipe_buf(pipe, pipe->head++); | |
248 | *buf = (struct pipe_buffer) { | |
249 | .ops = &default_pipe_buf_ops, | |
250 | .page = page, | |
251 | .offset = 0, | |
252 | .len = size | |
253 | }; | |
254 | } | |
255 | return page; | |
256 | } | |
257 | ||
258 | static void push_page(struct pipe_inode_info *pipe, struct page *page, | |
259 | unsigned int offset, unsigned int size) | |
260 | { | |
261 | struct pipe_buffer *buf = pipe_buf(pipe, pipe->head++); | |
262 | *buf = (struct pipe_buffer) { | |
263 | .ops = &page_cache_pipe_buf_ops, | |
264 | .page = page, | |
265 | .offset = offset, | |
266 | .len = size | |
267 | }; | |
268 | get_page(page); | |
269 | } | |
270 | ||
10f525a8 | 271 | static inline int last_offset(const struct pipe_buffer *buf) |
8fad7767 | 272 | { |
10f525a8 AV |
273 | if (buf->ops == &default_pipe_buf_ops) |
274 | return buf->len; // buf->offset is 0 for those | |
275 | else | |
276 | return -(buf->offset + buf->len); | |
8fad7767 AV |
277 | } |
278 | ||
279 | static struct page *append_pipe(struct iov_iter *i, size_t size, | |
280 | unsigned int *off) | |
281 | { | |
282 | struct pipe_inode_info *pipe = i->pipe; | |
10f525a8 | 283 | int offset = i->last_offset; |
8fad7767 AV |
284 | struct pipe_buffer *buf; |
285 | struct page *page; | |
286 | ||
10f525a8 AV |
287 | if (offset > 0 && offset < PAGE_SIZE) { |
288 | // some space in the last buffer; add to it | |
8fad7767 | 289 | buf = pipe_buf(pipe, pipe->head - 1); |
10f525a8 AV |
290 | size = min_t(size_t, size, PAGE_SIZE - offset); |
291 | buf->len += size; | |
292 | i->last_offset += size; | |
293 | i->count -= size; | |
294 | *off = offset; | |
295 | return buf->page; | |
8fad7767 AV |
296 | } |
297 | // OK, we need a new buffer | |
298 | *off = 0; | |
299 | size = min_t(size_t, size, PAGE_SIZE); | |
300 | if (pipe_full(pipe->head, pipe->tail, pipe->max_usage)) | |
301 | return NULL; | |
302 | page = push_anon(pipe, size); | |
303 | if (!page) | |
304 | return NULL; | |
305 | i->head = pipe->head - 1; | |
10f525a8 | 306 | i->last_offset = size; |
8fad7767 AV |
307 | i->count -= size; |
308 | return page; | |
309 | } | |
310 | ||
241699cd AV |
311 | static size_t copy_page_to_iter_pipe(struct page *page, size_t offset, size_t bytes, |
312 | struct iov_iter *i) | |
313 | { | |
314 | struct pipe_inode_info *pipe = i->pipe; | |
47b7fcae | 315 | unsigned int head = pipe->head; |
241699cd AV |
316 | |
317 | if (unlikely(bytes > i->count)) | |
318 | bytes = i->count; | |
319 | ||
320 | if (unlikely(!bytes)) | |
321 | return 0; | |
322 | ||
323 | if (!sanity(i)) | |
324 | return 0; | |
325 | ||
10f525a8 | 326 | if (offset && i->last_offset == -offset) { // could we merge it? |
47b7fcae AV |
327 | struct pipe_buffer *buf = pipe_buf(pipe, head - 1); |
328 | if (buf->page == page) { | |
241699cd | 329 | buf->len += bytes; |
10f525a8 | 330 | i->last_offset -= bytes; |
47b7fcae AV |
331 | i->count -= bytes; |
332 | return bytes; | |
241699cd | 333 | } |
241699cd | 334 | } |
47b7fcae | 335 | if (pipe_full(pipe->head, pipe->tail, pipe->max_usage)) |
241699cd | 336 | return 0; |
8cefc107 | 337 | |
47b7fcae | 338 | push_page(pipe, page, offset, bytes); |
10f525a8 | 339 | i->last_offset = -(offset + bytes); |
47b7fcae | 340 | i->head = head; |
241699cd AV |
341 | i->count -= bytes; |
342 | return bytes; | |
343 | } | |
344 | ||
171a0203 | 345 | /* |
a6294593 AG |
346 | * fault_in_iov_iter_readable - fault in iov iterator for reading |
347 | * @i: iterator | |
348 | * @size: maximum length | |
349 | * | |
171a0203 | 350 | * Fault in one or more iovecs of the given iov_iter, to a maximum length of |
a6294593 AG |
351 | * @size. For each iovec, fault in each page that constitutes the iovec. |
352 | * | |
353 | * Returns the number of bytes not faulted in (like copy_to_user() and | |
354 | * copy_from_user()). | |
171a0203 | 355 | * |
a6294593 | 356 | * Always returns 0 for non-userspace iterators. |
171a0203 | 357 | */ |
a6294593 | 358 | size_t fault_in_iov_iter_readable(const struct iov_iter *i, size_t size) |
171a0203 | 359 | { |
fcb14cb1 AV |
360 | if (iter_is_ubuf(i)) { |
361 | size_t n = min(size, iov_iter_count(i)); | |
362 | n -= fault_in_readable(i->ubuf + i->iov_offset, n); | |
363 | return size - n; | |
364 | } else if (iter_is_iovec(i)) { | |
a6294593 | 365 | size_t count = min(size, iov_iter_count(i)); |
8409a0d2 AV |
366 | const struct iovec *p; |
367 | size_t skip; | |
368 | ||
a6294593 | 369 | size -= count; |
de4f5fed | 370 | for (p = iter_iov(i), skip = i->iov_offset; count; p++, skip = 0) { |
a6294593 AG |
371 | size_t len = min(count, p->iov_len - skip); |
372 | size_t ret; | |
8409a0d2 AV |
373 | |
374 | if (unlikely(!len)) | |
375 | continue; | |
a6294593 AG |
376 | ret = fault_in_readable(p->iov_base + skip, len); |
377 | count -= len - ret; | |
378 | if (ret) | |
379 | break; | |
8409a0d2 | 380 | } |
a6294593 | 381 | return count + size; |
171a0203 AA |
382 | } |
383 | return 0; | |
384 | } | |
a6294593 | 385 | EXPORT_SYMBOL(fault_in_iov_iter_readable); |
171a0203 | 386 | |
cdd591fc AG |
387 | /* |
388 | * fault_in_iov_iter_writeable - fault in iov iterator for writing | |
389 | * @i: iterator | |
390 | * @size: maximum length | |
391 | * | |
392 | * Faults in the iterator using get_user_pages(), i.e., without triggering | |
393 | * hardware page faults. This is primarily useful when we already know that | |
394 | * some or all of the pages in @i aren't in memory. | |
395 | * | |
396 | * Returns the number of bytes not faulted in, like copy_to_user() and | |
397 | * copy_from_user(). | |
398 | * | |
399 | * Always returns 0 for non-user-space iterators. | |
400 | */ | |
401 | size_t fault_in_iov_iter_writeable(const struct iov_iter *i, size_t size) | |
402 | { | |
fcb14cb1 AV |
403 | if (iter_is_ubuf(i)) { |
404 | size_t n = min(size, iov_iter_count(i)); | |
405 | n -= fault_in_safe_writeable(i->ubuf + i->iov_offset, n); | |
406 | return size - n; | |
407 | } else if (iter_is_iovec(i)) { | |
cdd591fc AG |
408 | size_t count = min(size, iov_iter_count(i)); |
409 | const struct iovec *p; | |
410 | size_t skip; | |
411 | ||
412 | size -= count; | |
de4f5fed | 413 | for (p = iter_iov(i), skip = i->iov_offset; count; p++, skip = 0) { |
cdd591fc AG |
414 | size_t len = min(count, p->iov_len - skip); |
415 | size_t ret; | |
416 | ||
417 | if (unlikely(!len)) | |
418 | continue; | |
419 | ret = fault_in_safe_writeable(p->iov_base + skip, len); | |
420 | count -= len - ret; | |
421 | if (ret) | |
422 | break; | |
423 | } | |
424 | return count + size; | |
425 | } | |
426 | return 0; | |
427 | } | |
428 | EXPORT_SYMBOL(fault_in_iov_iter_writeable); | |
429 | ||
aa563d7b | 430 | void iov_iter_init(struct iov_iter *i, unsigned int direction, |
71d8e532 AV |
431 | const struct iovec *iov, unsigned long nr_segs, |
432 | size_t count) | |
433 | { | |
aa563d7b | 434 | WARN_ON(direction & ~(READ | WRITE)); |
8cd54c1c AV |
435 | *i = (struct iov_iter) { |
436 | .iter_type = ITER_IOVEC, | |
3337ab08 | 437 | .nofault = false, |
fcb14cb1 | 438 | .user_backed = true, |
8cd54c1c | 439 | .data_source = direction, |
de4f5fed | 440 | .__iov = iov, |
8cd54c1c AV |
441 | .nr_segs = nr_segs, |
442 | .iov_offset = 0, | |
443 | .count = count | |
444 | }; | |
71d8e532 AV |
445 | } |
446 | EXPORT_SYMBOL(iov_iter_init); | |
7b2c99d1 | 447 | |
12d426ab AV |
448 | // returns the offset in partial buffer (if any) |
449 | static inline unsigned int pipe_npages(const struct iov_iter *i, int *npages) | |
241699cd | 450 | { |
12d426ab AV |
451 | struct pipe_inode_info *pipe = i->pipe; |
452 | int used = pipe->head - pipe->tail; | |
10f525a8 | 453 | int off = i->last_offset; |
8cefc107 | 454 | |
12d426ab AV |
455 | *npages = max((int)pipe->max_usage - used, 0); |
456 | ||
10f525a8 | 457 | if (off > 0 && off < PAGE_SIZE) { // anon and not full |
12d426ab AV |
458 | (*npages)++; |
459 | return off; | |
241699cd | 460 | } |
12d426ab | 461 | return 0; |
241699cd AV |
462 | } |
463 | ||
241699cd AV |
464 | static size_t copy_pipe_to_iter(const void *addr, size_t bytes, |
465 | struct iov_iter *i) | |
466 | { | |
8fad7767 | 467 | unsigned int off, chunk; |
241699cd | 468 | |
8fad7767 AV |
469 | if (unlikely(bytes > i->count)) |
470 | bytes = i->count; | |
471 | if (unlikely(!bytes)) | |
241699cd AV |
472 | return 0; |
473 | ||
8fad7767 | 474 | if (!sanity(i)) |
241699cd | 475 | return 0; |
8fad7767 AV |
476 | |
477 | for (size_t n = bytes; n; n -= chunk) { | |
478 | struct page *page = append_pipe(i, n, &off); | |
479 | chunk = min_t(size_t, n, PAGE_SIZE - off); | |
480 | if (!page) | |
481 | return bytes - n; | |
482 | memcpy_to_page(page, off, addr, chunk); | |
241699cd | 483 | addr += chunk; |
8fad7767 | 484 | } |
241699cd AV |
485 | return bytes; |
486 | } | |
487 | ||
f9152895 AV |
488 | static __wsum csum_and_memcpy(void *to, const void *from, size_t len, |
489 | __wsum sum, size_t off) | |
490 | { | |
cc44c17b | 491 | __wsum next = csum_partial_copy_nocheck(from, to, len); |
f9152895 AV |
492 | return csum_block_add(sum, next, off); |
493 | } | |
494 | ||
78e1f386 | 495 | static size_t csum_and_copy_to_pipe_iter(const void *addr, size_t bytes, |
6852df12 | 496 | struct iov_iter *i, __wsum *sump) |
78e1f386 | 497 | { |
6852df12 AV |
498 | __wsum sum = *sump; |
499 | size_t off = 0; | |
8fad7767 AV |
500 | unsigned int chunk, r; |
501 | ||
502 | if (unlikely(bytes > i->count)) | |
503 | bytes = i->count; | |
504 | if (unlikely(!bytes)) | |
505 | return 0; | |
78e1f386 AV |
506 | |
507 | if (!sanity(i)) | |
508 | return 0; | |
509 | ||
6852df12 | 510 | while (bytes) { |
8fad7767 AV |
511 | struct page *page = append_pipe(i, bytes, &r); |
512 | char *p; | |
513 | ||
514 | if (!page) | |
515 | break; | |
516 | chunk = min_t(size_t, bytes, PAGE_SIZE - r); | |
517 | p = kmap_local_page(page); | |
6852df12 | 518 | sum = csum_and_memcpy(p + r, addr + off, chunk, sum, off); |
2495bdcc | 519 | kunmap_local(p); |
78e1f386 | 520 | off += chunk; |
8fad7767 | 521 | bytes -= chunk; |
6852df12 AV |
522 | } |
523 | *sump = sum; | |
6852df12 | 524 | return off; |
78e1f386 AV |
525 | } |
526 | ||
aa28de27 | 527 | size_t _copy_to_iter(const void *addr, size_t bytes, struct iov_iter *i) |
62a8067a | 528 | { |
a41dad90 AV |
529 | if (WARN_ON_ONCE(i->data_source)) |
530 | return 0; | |
00e23707 | 531 | if (unlikely(iov_iter_is_pipe(i))) |
241699cd | 532 | return copy_pipe_to_iter(addr, bytes, i); |
fcb14cb1 | 533 | if (user_backed_iter(i)) |
09fc68dc | 534 | might_fault(); |
7baa5099 AV |
535 | iterate_and_advance(i, bytes, base, len, off, |
536 | copyout(base, addr + off, len), | |
537 | memcpy(base, addr + off, len) | |
3d4d3e48 | 538 | ) |
62a8067a | 539 | |
3d4d3e48 | 540 | return bytes; |
c35e0248 | 541 | } |
aa28de27 | 542 | EXPORT_SYMBOL(_copy_to_iter); |
c35e0248 | 543 | |
ec6347bb DW |
544 | #ifdef CONFIG_ARCH_HAS_COPY_MC |
545 | static int copyout_mc(void __user *to, const void *from, size_t n) | |
8780356e | 546 | { |
96d4f267 | 547 | if (access_ok(to, n)) { |
d0ef4c36 | 548 | instrument_copy_to_user(to, from, n); |
ec6347bb | 549 | n = copy_mc_to_user((__force void *) to, from, n); |
8780356e DW |
550 | } |
551 | return n; | |
552 | } | |
553 | ||
ec6347bb | 554 | static size_t copy_mc_pipe_to_iter(const void *addr, size_t bytes, |
ca146f6f DW |
555 | struct iov_iter *i) |
556 | { | |
8fad7767 AV |
557 | size_t xfer = 0; |
558 | unsigned int off, chunk; | |
559 | ||
560 | if (unlikely(bytes > i->count)) | |
561 | bytes = i->count; | |
562 | if (unlikely(!bytes)) | |
563 | return 0; | |
ca146f6f DW |
564 | |
565 | if (!sanity(i)) | |
566 | return 0; | |
567 | ||
8fad7767 AV |
568 | while (bytes) { |
569 | struct page *page = append_pipe(i, bytes, &off); | |
ca146f6f | 570 | unsigned long rem; |
8fad7767 AV |
571 | char *p; |
572 | ||
573 | if (!page) | |
574 | break; | |
575 | chunk = min_t(size_t, bytes, PAGE_SIZE - off); | |
576 | p = kmap_local_page(page); | |
2a510a74 AV |
577 | rem = copy_mc_to_kernel(p + off, addr + xfer, chunk); |
578 | chunk -= rem; | |
579 | kunmap_local(p); | |
8fad7767 AV |
580 | xfer += chunk; |
581 | bytes -= chunk; | |
c3497fd0 | 582 | if (rem) { |
8fad7767 | 583 | iov_iter_revert(i, rem); |
ca146f6f | 584 | break; |
c3497fd0 | 585 | } |
2a510a74 | 586 | } |
ca146f6f DW |
587 | return xfer; |
588 | } | |
589 | ||
bf3eeb9b | 590 | /** |
ec6347bb | 591 | * _copy_mc_to_iter - copy to iter with source memory error exception handling |
bf3eeb9b DW |
592 | * @addr: source kernel address |
593 | * @bytes: total transfer length | |
44e55997 | 594 | * @i: destination iterator |
bf3eeb9b | 595 | * |
ec6347bb DW |
596 | * The pmem driver deploys this for the dax operation |
597 | * (dax_copy_to_iter()) for dax reads (bypass page-cache and the | |
598 | * block-layer). Upon #MC read(2) aborts and returns EIO or the bytes | |
599 | * successfully copied. | |
bf3eeb9b | 600 | * |
ec6347bb | 601 | * The main differences between this and typical _copy_to_iter(). |
bf3eeb9b DW |
602 | * |
603 | * * Typical tail/residue handling after a fault retries the copy | |
604 | * byte-by-byte until the fault happens again. Re-triggering machine | |
605 | * checks is potentially fatal so the implementation uses source | |
606 | * alignment and poison alignment assumptions to avoid re-triggering | |
607 | * hardware exceptions. | |
608 | * | |
609 | * * ITER_KVEC, ITER_PIPE, and ITER_BVEC can return short copies. | |
610 | * Compare to copy_to_iter() where only ITER_IOVEC attempts might return | |
611 | * a short copy. | |
44e55997 RD |
612 | * |
613 | * Return: number of bytes copied (may be %0) | |
bf3eeb9b | 614 | */ |
ec6347bb | 615 | size_t _copy_mc_to_iter(const void *addr, size_t bytes, struct iov_iter *i) |
8780356e | 616 | { |
a41dad90 AV |
617 | if (WARN_ON_ONCE(i->data_source)) |
618 | return 0; | |
00e23707 | 619 | if (unlikely(iov_iter_is_pipe(i))) |
ec6347bb | 620 | return copy_mc_pipe_to_iter(addr, bytes, i); |
fcb14cb1 | 621 | if (user_backed_iter(i)) |
8780356e | 622 | might_fault(); |
7baa5099 AV |
623 | __iterate_and_advance(i, bytes, base, len, off, |
624 | copyout_mc(base, addr + off, len), | |
625 | copy_mc_to_kernel(base, addr + off, len) | |
8780356e DW |
626 | ) |
627 | ||
628 | return bytes; | |
629 | } | |
ec6347bb DW |
630 | EXPORT_SYMBOL_GPL(_copy_mc_to_iter); |
631 | #endif /* CONFIG_ARCH_HAS_COPY_MC */ | |
8780356e | 632 | |
aa28de27 | 633 | size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i) |
c35e0248 | 634 | { |
a41dad90 | 635 | if (WARN_ON_ONCE(!i->data_source)) |
241699cd | 636 | return 0; |
a41dad90 | 637 | |
fcb14cb1 | 638 | if (user_backed_iter(i)) |
09fc68dc | 639 | might_fault(); |
7baa5099 AV |
640 | iterate_and_advance(i, bytes, base, len, off, |
641 | copyin(addr + off, base, len), | |
642 | memcpy(addr + off, base, len) | |
0dbca9a4 AV |
643 | ) |
644 | ||
645 | return bytes; | |
c35e0248 | 646 | } |
aa28de27 | 647 | EXPORT_SYMBOL(_copy_from_iter); |
c35e0248 | 648 | |
aa28de27 | 649 | size_t _copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i) |
aa583096 | 650 | { |
a41dad90 | 651 | if (WARN_ON_ONCE(!i->data_source)) |
241699cd | 652 | return 0; |
a41dad90 | 653 | |
7baa5099 AV |
654 | iterate_and_advance(i, bytes, base, len, off, |
655 | __copy_from_user_inatomic_nocache(addr + off, base, len), | |
656 | memcpy(addr + off, base, len) | |
aa583096 AV |
657 | ) |
658 | ||
659 | return bytes; | |
660 | } | |
aa28de27 | 661 | EXPORT_SYMBOL(_copy_from_iter_nocache); |
aa583096 | 662 | |
0aed55af | 663 | #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE |
abd08d7d DW |
664 | /** |
665 | * _copy_from_iter_flushcache - write destination through cpu cache | |
666 | * @addr: destination kernel address | |
667 | * @bytes: total transfer length | |
44e55997 | 668 | * @i: source iterator |
abd08d7d DW |
669 | * |
670 | * The pmem driver arranges for filesystem-dax to use this facility via | |
671 | * dax_copy_from_iter() for ensuring that writes to persistent memory | |
672 | * are flushed through the CPU cache. It is differentiated from | |
673 | * _copy_from_iter_nocache() in that guarantees all data is flushed for | |
674 | * all iterator types. The _copy_from_iter_nocache() only attempts to | |
675 | * bypass the cache for the ITER_IOVEC case, and on some archs may use | |
676 | * instructions that strand dirty-data in the cache. | |
44e55997 RD |
677 | * |
678 | * Return: number of bytes copied (may be %0) | |
abd08d7d | 679 | */ |
6a37e940 | 680 | size_t _copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i) |
0aed55af | 681 | { |
a41dad90 | 682 | if (WARN_ON_ONCE(!i->data_source)) |
0aed55af | 683 | return 0; |
a41dad90 | 684 | |
7baa5099 AV |
685 | iterate_and_advance(i, bytes, base, len, off, |
686 | __copy_from_user_flushcache(addr + off, base, len), | |
687 | memcpy_flushcache(addr + off, base, len) | |
0aed55af DW |
688 | ) |
689 | ||
690 | return bytes; | |
691 | } | |
6a37e940 | 692 | EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache); |
0aed55af DW |
693 | #endif |
694 | ||
72e809ed AV |
695 | static inline bool page_copy_sane(struct page *page, size_t offset, size_t n) |
696 | { | |
6daef95b ED |
697 | struct page *head; |
698 | size_t v = n + offset; | |
699 | ||
700 | /* | |
701 | * The general case needs to access the page order in order | |
702 | * to compute the page size. | |
703 | * However, we mostly deal with order-0 pages and thus can | |
704 | * avoid a possible cache line miss for requests that fit all | |
705 | * page orders. | |
706 | */ | |
707 | if (n <= v && v <= PAGE_SIZE) | |
708 | return true; | |
709 | ||
710 | head = compound_head(page); | |
711 | v += (page - head) << PAGE_SHIFT; | |
a90bcb86 | 712 | |
40a86061 AV |
713 | if (WARN_ON(n > v || v > page_size(head))) |
714 | return false; | |
715 | return true; | |
72e809ed | 716 | } |
cbbd26b8 | 717 | |
08aa6479 AV |
718 | size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes, |
719 | struct iov_iter *i) | |
720 | { | |
721 | size_t res = 0; | |
40a86061 | 722 | if (!page_copy_sane(page, offset, bytes)) |
08aa6479 | 723 | return 0; |
a41dad90 AV |
724 | if (WARN_ON_ONCE(i->data_source)) |
725 | return 0; | |
f0f6b614 AV |
726 | if (unlikely(iov_iter_is_pipe(i))) |
727 | return copy_page_to_iter_pipe(page, offset, bytes, i); | |
08aa6479 AV |
728 | page += offset / PAGE_SIZE; // first subpage |
729 | offset %= PAGE_SIZE; | |
730 | while (1) { | |
f0f6b614 AV |
731 | void *kaddr = kmap_local_page(page); |
732 | size_t n = min(bytes, (size_t)PAGE_SIZE - offset); | |
733 | n = _copy_to_iter(kaddr + offset, n, i); | |
734 | kunmap_local(kaddr); | |
08aa6479 AV |
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 | ||
4f80818b LS |
749 | size_t copy_page_to_iter_nofault(struct page *page, unsigned offset, size_t bytes, |
750 | struct iov_iter *i) | |
751 | { | |
752 | size_t res = 0; | |
753 | ||
754 | if (!page_copy_sane(page, offset, bytes)) | |
755 | return 0; | |
756 | if (WARN_ON_ONCE(i->data_source)) | |
757 | return 0; | |
758 | if (unlikely(iov_iter_is_pipe(i))) | |
759 | return copy_page_to_iter_pipe(page, offset, bytes, i); | |
760 | page += offset / PAGE_SIZE; // first subpage | |
761 | offset %= PAGE_SIZE; | |
762 | while (1) { | |
763 | void *kaddr = kmap_local_page(page); | |
764 | size_t n = min(bytes, (size_t)PAGE_SIZE - offset); | |
765 | ||
766 | iterate_and_advance(i, n, base, len, off, | |
767 | copyout_nofault(base, kaddr + offset + off, len), | |
768 | memcpy(base, kaddr + offset + off, len) | |
769 | ) | |
770 | kunmap_local(kaddr); | |
771 | res += n; | |
772 | bytes -= n; | |
773 | if (!bytes || !n) | |
774 | break; | |
775 | offset += n; | |
776 | if (offset == PAGE_SIZE) { | |
777 | page++; | |
778 | offset = 0; | |
779 | } | |
780 | } | |
781 | return res; | |
782 | } | |
783 | EXPORT_SYMBOL(copy_page_to_iter_nofault); | |
784 | ||
62a8067a AV |
785 | size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes, |
786 | struct iov_iter *i) | |
787 | { | |
c03f05f1 AV |
788 | size_t res = 0; |
789 | if (!page_copy_sane(page, offset, bytes)) | |
790 | return 0; | |
791 | page += offset / PAGE_SIZE; // first subpage | |
792 | offset %= PAGE_SIZE; | |
793 | while (1) { | |
55ca375c | 794 | void *kaddr = kmap_local_page(page); |
c03f05f1 AV |
795 | size_t n = min(bytes, (size_t)PAGE_SIZE - offset); |
796 | n = _copy_from_iter(kaddr + offset, n, i); | |
55ca375c | 797 | kunmap_local(kaddr); |
c03f05f1 AV |
798 | res += n; |
799 | bytes -= n; | |
800 | if (!bytes || !n) | |
801 | break; | |
802 | offset += n; | |
803 | if (offset == PAGE_SIZE) { | |
804 | page++; | |
805 | offset = 0; | |
806 | } | |
28f38db7 | 807 | } |
c03f05f1 | 808 | return res; |
62a8067a AV |
809 | } |
810 | EXPORT_SYMBOL(copy_page_from_iter); | |
811 | ||
241699cd AV |
812 | static size_t pipe_zero(size_t bytes, struct iov_iter *i) |
813 | { | |
8fad7767 | 814 | unsigned int chunk, off; |
241699cd | 815 | |
8fad7767 AV |
816 | if (unlikely(bytes > i->count)) |
817 | bytes = i->count; | |
818 | if (unlikely(!bytes)) | |
241699cd AV |
819 | return 0; |
820 | ||
8fad7767 | 821 | if (!sanity(i)) |
241699cd AV |
822 | return 0; |
823 | ||
8fad7767 AV |
824 | for (size_t n = bytes; n; n -= chunk) { |
825 | struct page *page = append_pipe(i, n, &off); | |
826 | char *p; | |
827 | ||
828 | if (!page) | |
829 | return bytes - n; | |
830 | chunk = min_t(size_t, n, PAGE_SIZE - off); | |
831 | p = kmap_local_page(page); | |
893839fd AV |
832 | memset(p + off, 0, chunk); |
833 | kunmap_local(p); | |
8fad7767 | 834 | } |
241699cd AV |
835 | return bytes; |
836 | } | |
837 | ||
c35e0248 MW |
838 | size_t iov_iter_zero(size_t bytes, struct iov_iter *i) |
839 | { | |
00e23707 | 840 | if (unlikely(iov_iter_is_pipe(i))) |
241699cd | 841 | return pipe_zero(bytes, i); |
7baa5099 AV |
842 | iterate_and_advance(i, bytes, base, len, count, |
843 | clear_user(base, len), | |
844 | memset(base, 0, len) | |
8442fa46 AV |
845 | ) |
846 | ||
847 | return bytes; | |
c35e0248 MW |
848 | } |
849 | EXPORT_SYMBOL(iov_iter_zero); | |
850 | ||
f0b65f39 AV |
851 | size_t copy_page_from_iter_atomic(struct page *page, unsigned offset, size_t bytes, |
852 | struct iov_iter *i) | |
62a8067a | 853 | { |
04a31165 | 854 | char *kaddr = kmap_atomic(page), *p = kaddr + offset; |
40a86061 | 855 | if (!page_copy_sane(page, offset, bytes)) { |
72e809ed AV |
856 | kunmap_atomic(kaddr); |
857 | return 0; | |
858 | } | |
a41dad90 | 859 | if (WARN_ON_ONCE(!i->data_source)) { |
241699cd | 860 | kunmap_atomic(kaddr); |
241699cd AV |
861 | return 0; |
862 | } | |
7baa5099 AV |
863 | iterate_and_advance(i, bytes, base, len, off, |
864 | copyin(p + off, base, len), | |
865 | memcpy(p + off, base, len) | |
04a31165 AV |
866 | ) |
867 | kunmap_atomic(kaddr); | |
868 | return bytes; | |
62a8067a | 869 | } |
f0b65f39 | 870 | EXPORT_SYMBOL(copy_page_from_iter_atomic); |
62a8067a | 871 | |
241699cd AV |
872 | static void pipe_advance(struct iov_iter *i, size_t size) |
873 | { | |
874 | struct pipe_inode_info *pipe = i->pipe; | |
10f525a8 | 875 | int off = i->last_offset; |
8cefc107 | 876 | |
2c855de9 AV |
877 | if (!off && !size) { |
878 | pipe_discard_from(pipe, i->start_head); // discard everything | |
879 | return; | |
880 | } | |
881 | i->count -= size; | |
882 | while (1) { | |
883 | struct pipe_buffer *buf = pipe_buf(pipe, i->head); | |
241699cd | 884 | if (off) /* make it relative to the beginning of buffer */ |
10f525a8 | 885 | size += abs(off) - buf->offset; |
2c855de9 AV |
886 | if (size <= buf->len) { |
887 | buf->len = size; | |
10f525a8 | 888 | i->last_offset = last_offset(buf); |
2c855de9 | 889 | break; |
241699cd | 890 | } |
2c855de9 AV |
891 | size -= buf->len; |
892 | i->head++; | |
893 | off = 0; | |
241699cd | 894 | } |
2c855de9 | 895 | pipe_discard_from(pipe, i->head + 1); // discard everything past this one |
241699cd AV |
896 | } |
897 | ||
54c8195b PB |
898 | static void iov_iter_bvec_advance(struct iov_iter *i, size_t size) |
899 | { | |
18fa9af7 | 900 | const struct bio_vec *bvec, *end; |
54c8195b | 901 | |
18fa9af7 AV |
902 | if (!i->count) |
903 | return; | |
904 | i->count -= size; | |
905 | ||
906 | size += i->iov_offset; | |
54c8195b | 907 | |
18fa9af7 AV |
908 | for (bvec = i->bvec, end = bvec + i->nr_segs; bvec < end; bvec++) { |
909 | if (likely(size < bvec->bv_len)) | |
910 | break; | |
911 | size -= bvec->bv_len; | |
912 | } | |
913 | i->iov_offset = size; | |
914 | i->nr_segs -= bvec - i->bvec; | |
915 | i->bvec = bvec; | |
54c8195b PB |
916 | } |
917 | ||
185ac4d4 AV |
918 | static void iov_iter_iovec_advance(struct iov_iter *i, size_t size) |
919 | { | |
920 | const struct iovec *iov, *end; | |
921 | ||
922 | if (!i->count) | |
923 | return; | |
924 | i->count -= size; | |
925 | ||
926 | size += i->iov_offset; // from beginning of current segment | |
de4f5fed | 927 | for (iov = iter_iov(i), end = iov + i->nr_segs; iov < end; iov++) { |
185ac4d4 AV |
928 | if (likely(size < iov->iov_len)) |
929 | break; | |
930 | size -= iov->iov_len; | |
931 | } | |
932 | i->iov_offset = size; | |
de4f5fed JA |
933 | i->nr_segs -= iov - iter_iov(i); |
934 | i->__iov = iov; | |
185ac4d4 AV |
935 | } |
936 | ||
62a8067a AV |
937 | void iov_iter_advance(struct iov_iter *i, size_t size) |
938 | { | |
3b3fc051 AV |
939 | if (unlikely(i->count < size)) |
940 | size = i->count; | |
fcb14cb1 AV |
941 | if (likely(iter_is_ubuf(i)) || unlikely(iov_iter_is_xarray(i))) { |
942 | i->iov_offset += size; | |
943 | i->count -= size; | |
944 | } else if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) { | |
185ac4d4 AV |
945 | /* iovec and kvec have identical layouts */ |
946 | iov_iter_iovec_advance(i, size); | |
947 | } else if (iov_iter_is_bvec(i)) { | |
948 | iov_iter_bvec_advance(i, size); | |
949 | } else if (iov_iter_is_pipe(i)) { | |
241699cd | 950 | pipe_advance(i, size); |
185ac4d4 AV |
951 | } else if (iov_iter_is_discard(i)) { |
952 | i->count -= size; | |
54c8195b | 953 | } |
62a8067a AV |
954 | } |
955 | EXPORT_SYMBOL(iov_iter_advance); | |
956 | ||
27c0e374 AV |
957 | void iov_iter_revert(struct iov_iter *i, size_t unroll) |
958 | { | |
959 | if (!unroll) | |
960 | return; | |
5b47d59a AV |
961 | if (WARN_ON(unroll > MAX_RW_COUNT)) |
962 | return; | |
27c0e374 | 963 | i->count += unroll; |
00e23707 | 964 | if (unlikely(iov_iter_is_pipe(i))) { |
27c0e374 | 965 | struct pipe_inode_info *pipe = i->pipe; |
92acdc4f AV |
966 | unsigned int head = pipe->head; |
967 | ||
968 | while (head > i->start_head) { | |
969 | struct pipe_buffer *b = pipe_buf(pipe, --head); | |
970 | if (unroll < b->len) { | |
971 | b->len -= unroll; | |
10f525a8 | 972 | i->last_offset = last_offset(b); |
92acdc4f AV |
973 | i->head = head; |
974 | return; | |
27c0e374 | 975 | } |
92acdc4f AV |
976 | unroll -= b->len; |
977 | pipe_buf_release(pipe, b); | |
978 | pipe->head--; | |
27c0e374 | 979 | } |
10f525a8 | 980 | i->last_offset = 0; |
92acdc4f | 981 | i->head = head; |
27c0e374 AV |
982 | return; |
983 | } | |
9ea9ce04 DH |
984 | if (unlikely(iov_iter_is_discard(i))) |
985 | return; | |
27c0e374 AV |
986 | if (unroll <= i->iov_offset) { |
987 | i->iov_offset -= unroll; | |
988 | return; | |
989 | } | |
990 | unroll -= i->iov_offset; | |
fcb14cb1 | 991 | if (iov_iter_is_xarray(i) || iter_is_ubuf(i)) { |
7ff50620 DH |
992 | BUG(); /* We should never go beyond the start of the specified |
993 | * range since we might then be straying into pages that | |
994 | * aren't pinned. | |
995 | */ | |
996 | } else if (iov_iter_is_bvec(i)) { | |
27c0e374 AV |
997 | const struct bio_vec *bvec = i->bvec; |
998 | while (1) { | |
999 | size_t n = (--bvec)->bv_len; | |
1000 | i->nr_segs++; | |
1001 | if (unroll <= n) { | |
1002 | i->bvec = bvec; | |
1003 | i->iov_offset = n - unroll; | |
1004 | return; | |
1005 | } | |
1006 | unroll -= n; | |
1007 | } | |
1008 | } else { /* same logics for iovec and kvec */ | |
de4f5fed | 1009 | const struct iovec *iov = iter_iov(i); |
27c0e374 AV |
1010 | while (1) { |
1011 | size_t n = (--iov)->iov_len; | |
1012 | i->nr_segs++; | |
1013 | if (unroll <= n) { | |
de4f5fed | 1014 | i->__iov = iov; |
27c0e374 AV |
1015 | i->iov_offset = n - unroll; |
1016 | return; | |
1017 | } | |
1018 | unroll -= n; | |
1019 | } | |
1020 | } | |
1021 | } | |
1022 | EXPORT_SYMBOL(iov_iter_revert); | |
1023 | ||
62a8067a AV |
1024 | /* |
1025 | * Return the count of just the current iov_iter segment. | |
1026 | */ | |
1027 | size_t iov_iter_single_seg_count(const struct iov_iter *i) | |
1028 | { | |
28f38db7 AV |
1029 | if (i->nr_segs > 1) { |
1030 | if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) | |
de4f5fed | 1031 | return min(i->count, iter_iov(i)->iov_len - i->iov_offset); |
28f38db7 AV |
1032 | if (iov_iter_is_bvec(i)) |
1033 | return min(i->count, i->bvec->bv_len - i->iov_offset); | |
1034 | } | |
1035 | return i->count; | |
62a8067a AV |
1036 | } |
1037 | EXPORT_SYMBOL(iov_iter_single_seg_count); | |
1038 | ||
aa563d7b | 1039 | void iov_iter_kvec(struct iov_iter *i, unsigned int direction, |
05afcb77 | 1040 | const struct kvec *kvec, unsigned long nr_segs, |
abb78f87 AV |
1041 | size_t count) |
1042 | { | |
aa563d7b | 1043 | WARN_ON(direction & ~(READ | WRITE)); |
8cd54c1c AV |
1044 | *i = (struct iov_iter){ |
1045 | .iter_type = ITER_KVEC, | |
1046 | .data_source = direction, | |
1047 | .kvec = kvec, | |
1048 | .nr_segs = nr_segs, | |
1049 | .iov_offset = 0, | |
1050 | .count = count | |
1051 | }; | |
abb78f87 AV |
1052 | } |
1053 | EXPORT_SYMBOL(iov_iter_kvec); | |
1054 | ||
aa563d7b | 1055 | void iov_iter_bvec(struct iov_iter *i, unsigned int direction, |
05afcb77 AV |
1056 | const struct bio_vec *bvec, unsigned long nr_segs, |
1057 | size_t count) | |
1058 | { | |
aa563d7b | 1059 | WARN_ON(direction & ~(READ | WRITE)); |
8cd54c1c AV |
1060 | *i = (struct iov_iter){ |
1061 | .iter_type = ITER_BVEC, | |
1062 | .data_source = direction, | |
1063 | .bvec = bvec, | |
1064 | .nr_segs = nr_segs, | |
1065 | .iov_offset = 0, | |
1066 | .count = count | |
1067 | }; | |
05afcb77 AV |
1068 | } |
1069 | EXPORT_SYMBOL(iov_iter_bvec); | |
1070 | ||
aa563d7b | 1071 | void iov_iter_pipe(struct iov_iter *i, unsigned int direction, |
241699cd AV |
1072 | struct pipe_inode_info *pipe, |
1073 | size_t count) | |
1074 | { | |
aa563d7b | 1075 | BUG_ON(direction != READ); |
8cefc107 | 1076 | WARN_ON(pipe_full(pipe->head, pipe->tail, pipe->ring_size)); |
8cd54c1c AV |
1077 | *i = (struct iov_iter){ |
1078 | .iter_type = ITER_PIPE, | |
1079 | .data_source = false, | |
1080 | .pipe = pipe, | |
1081 | .head = pipe->head, | |
1082 | .start_head = pipe->head, | |
10f525a8 | 1083 | .last_offset = 0, |
8cd54c1c AV |
1084 | .count = count |
1085 | }; | |
241699cd AV |
1086 | } |
1087 | EXPORT_SYMBOL(iov_iter_pipe); | |
1088 | ||
7ff50620 DH |
1089 | /** |
1090 | * iov_iter_xarray - Initialise an I/O iterator to use the pages in an xarray | |
1091 | * @i: The iterator to initialise. | |
1092 | * @direction: The direction of the transfer. | |
1093 | * @xarray: The xarray to access. | |
1094 | * @start: The start file position. | |
1095 | * @count: The size of the I/O buffer in bytes. | |
1096 | * | |
1097 | * Set up an I/O iterator to either draw data out of the pages attached to an | |
1098 | * inode or to inject data into those pages. The pages *must* be prevented | |
1099 | * from evaporation, either by taking a ref on them or locking them by the | |
1100 | * caller. | |
1101 | */ | |
1102 | void iov_iter_xarray(struct iov_iter *i, unsigned int direction, | |
1103 | struct xarray *xarray, loff_t start, size_t count) | |
1104 | { | |
1105 | BUG_ON(direction & ~1); | |
8cd54c1c AV |
1106 | *i = (struct iov_iter) { |
1107 | .iter_type = ITER_XARRAY, | |
1108 | .data_source = direction, | |
1109 | .xarray = xarray, | |
1110 | .xarray_start = start, | |
1111 | .count = count, | |
1112 | .iov_offset = 0 | |
1113 | }; | |
7ff50620 DH |
1114 | } |
1115 | EXPORT_SYMBOL(iov_iter_xarray); | |
1116 | ||
9ea9ce04 DH |
1117 | /** |
1118 | * iov_iter_discard - Initialise an I/O iterator that discards data | |
1119 | * @i: The iterator to initialise. | |
1120 | * @direction: The direction of the transfer. | |
1121 | * @count: The size of the I/O buffer in bytes. | |
1122 | * | |
1123 | * Set up an I/O iterator that just discards everything that's written to it. | |
1124 | * It's only available as a READ iterator. | |
1125 | */ | |
1126 | void iov_iter_discard(struct iov_iter *i, unsigned int direction, size_t count) | |
1127 | { | |
1128 | BUG_ON(direction != READ); | |
8cd54c1c AV |
1129 | *i = (struct iov_iter){ |
1130 | .iter_type = ITER_DISCARD, | |
1131 | .data_source = false, | |
1132 | .count = count, | |
1133 | .iov_offset = 0 | |
1134 | }; | |
9ea9ce04 DH |
1135 | } |
1136 | EXPORT_SYMBOL(iov_iter_discard); | |
1137 | ||
cfa320f7 KB |
1138 | static bool iov_iter_aligned_iovec(const struct iov_iter *i, unsigned addr_mask, |
1139 | unsigned len_mask) | |
1140 | { | |
1141 | size_t size = i->count; | |
1142 | size_t skip = i->iov_offset; | |
1143 | unsigned k; | |
1144 | ||
1145 | for (k = 0; k < i->nr_segs; k++, skip = 0) { | |
de4f5fed JA |
1146 | const struct iovec *iov = iter_iov(i) + k; |
1147 | size_t len = iov->iov_len - skip; | |
cfa320f7 KB |
1148 | |
1149 | if (len > size) | |
1150 | len = size; | |
1151 | if (len & len_mask) | |
1152 | return false; | |
de4f5fed | 1153 | if ((unsigned long)(iov->iov_base + skip) & addr_mask) |
cfa320f7 KB |
1154 | return false; |
1155 | ||
1156 | size -= len; | |
1157 | if (!size) | |
1158 | break; | |
1159 | } | |
1160 | return true; | |
1161 | } | |
1162 | ||
1163 | static bool iov_iter_aligned_bvec(const struct iov_iter *i, unsigned addr_mask, | |
1164 | unsigned len_mask) | |
1165 | { | |
1166 | size_t size = i->count; | |
1167 | unsigned skip = i->iov_offset; | |
1168 | unsigned k; | |
1169 | ||
1170 | for (k = 0; k < i->nr_segs; k++, skip = 0) { | |
1171 | size_t len = i->bvec[k].bv_len - skip; | |
1172 | ||
1173 | if (len > size) | |
1174 | len = size; | |
1175 | if (len & len_mask) | |
1176 | return false; | |
1177 | if ((unsigned long)(i->bvec[k].bv_offset + skip) & addr_mask) | |
1178 | return false; | |
1179 | ||
1180 | size -= len; | |
1181 | if (!size) | |
1182 | break; | |
1183 | } | |
1184 | return true; | |
1185 | } | |
1186 | ||
1187 | /** | |
1188 | * iov_iter_is_aligned() - Check if the addresses and lengths of each segments | |
1189 | * are aligned to the parameters. | |
1190 | * | |
1191 | * @i: &struct iov_iter to restore | |
1192 | * @addr_mask: bit mask to check against the iov element's addresses | |
1193 | * @len_mask: bit mask to check against the iov element's lengths | |
1194 | * | |
1195 | * Return: false if any addresses or lengths intersect with the provided masks | |
1196 | */ | |
1197 | bool iov_iter_is_aligned(const struct iov_iter *i, unsigned addr_mask, | |
1198 | unsigned len_mask) | |
1199 | { | |
fcb14cb1 AV |
1200 | if (likely(iter_is_ubuf(i))) { |
1201 | if (i->count & len_mask) | |
1202 | return false; | |
1203 | if ((unsigned long)(i->ubuf + i->iov_offset) & addr_mask) | |
1204 | return false; | |
1205 | return true; | |
1206 | } | |
1207 | ||
cfa320f7 KB |
1208 | if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) |
1209 | return iov_iter_aligned_iovec(i, addr_mask, len_mask); | |
1210 | ||
1211 | if (iov_iter_is_bvec(i)) | |
1212 | return iov_iter_aligned_bvec(i, addr_mask, len_mask); | |
1213 | ||
1214 | if (iov_iter_is_pipe(i)) { | |
cfa320f7 KB |
1215 | size_t size = i->count; |
1216 | ||
1217 | if (size & len_mask) | |
1218 | return false; | |
10f525a8 AV |
1219 | if (size && i->last_offset > 0) { |
1220 | if (i->last_offset & addr_mask) | |
cfa320f7 KB |
1221 | return false; |
1222 | } | |
1223 | ||
1224 | return true; | |
1225 | } | |
1226 | ||
1227 | if (iov_iter_is_xarray(i)) { | |
1228 | if (i->count & len_mask) | |
1229 | return false; | |
1230 | if ((i->xarray_start + i->iov_offset) & addr_mask) | |
1231 | return false; | |
1232 | } | |
1233 | ||
1234 | return true; | |
1235 | } | |
1236 | EXPORT_SYMBOL_GPL(iov_iter_is_aligned); | |
1237 | ||
9221d2e3 | 1238 | static unsigned long iov_iter_alignment_iovec(const struct iov_iter *i) |
62a8067a | 1239 | { |
04a31165 AV |
1240 | unsigned long res = 0; |
1241 | size_t size = i->count; | |
9221d2e3 AV |
1242 | size_t skip = i->iov_offset; |
1243 | unsigned k; | |
1244 | ||
1245 | for (k = 0; k < i->nr_segs; k++, skip = 0) { | |
de4f5fed JA |
1246 | const struct iovec *iov = iter_iov(i) + k; |
1247 | size_t len = iov->iov_len - skip; | |
9221d2e3 | 1248 | if (len) { |
de4f5fed | 1249 | res |= (unsigned long)iov->iov_base + skip; |
9221d2e3 AV |
1250 | if (len > size) |
1251 | len = size; | |
1252 | res |= len; | |
1253 | size -= len; | |
1254 | if (!size) | |
1255 | break; | |
1256 | } | |
1257 | } | |
1258 | return res; | |
1259 | } | |
04a31165 | 1260 | |
9221d2e3 AV |
1261 | static unsigned long iov_iter_alignment_bvec(const struct iov_iter *i) |
1262 | { | |
1263 | unsigned res = 0; | |
1264 | size_t size = i->count; | |
1265 | unsigned skip = i->iov_offset; | |
1266 | unsigned k; | |
1267 | ||
1268 | for (k = 0; k < i->nr_segs; k++, skip = 0) { | |
1269 | size_t len = i->bvec[k].bv_len - skip; | |
1270 | res |= (unsigned long)i->bvec[k].bv_offset + skip; | |
1271 | if (len > size) | |
1272 | len = size; | |
1273 | res |= len; | |
1274 | size -= len; | |
1275 | if (!size) | |
1276 | break; | |
1277 | } | |
1278 | return res; | |
1279 | } | |
1280 | ||
1281 | unsigned long iov_iter_alignment(const struct iov_iter *i) | |
1282 | { | |
fcb14cb1 AV |
1283 | if (likely(iter_is_ubuf(i))) { |
1284 | size_t size = i->count; | |
1285 | if (size) | |
1286 | return ((unsigned long)i->ubuf + i->iov_offset) | size; | |
1287 | return 0; | |
1288 | } | |
1289 | ||
9221d2e3 AV |
1290 | /* iovec and kvec have identical layouts */ |
1291 | if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) | |
1292 | return iov_iter_alignment_iovec(i); | |
1293 | ||
1294 | if (iov_iter_is_bvec(i)) | |
1295 | return iov_iter_alignment_bvec(i); | |
1296 | ||
1297 | if (iov_iter_is_pipe(i)) { | |
9221d2e3 | 1298 | size_t size = i->count; |
e0ff126e | 1299 | |
10f525a8 AV |
1300 | if (size && i->last_offset > 0) |
1301 | return size | i->last_offset; | |
241699cd AV |
1302 | return size; |
1303 | } | |
9221d2e3 AV |
1304 | |
1305 | if (iov_iter_is_xarray(i)) | |
3d14ec1f | 1306 | return (i->xarray_start + i->iov_offset) | i->count; |
9221d2e3 AV |
1307 | |
1308 | return 0; | |
62a8067a AV |
1309 | } |
1310 | EXPORT_SYMBOL(iov_iter_alignment); | |
1311 | ||
357f435d AV |
1312 | unsigned long iov_iter_gap_alignment(const struct iov_iter *i) |
1313 | { | |
33844e66 | 1314 | unsigned long res = 0; |
610c7a71 | 1315 | unsigned long v = 0; |
357f435d | 1316 | size_t size = i->count; |
610c7a71 | 1317 | unsigned k; |
357f435d | 1318 | |
fcb14cb1 AV |
1319 | if (iter_is_ubuf(i)) |
1320 | return 0; | |
1321 | ||
610c7a71 | 1322 | if (WARN_ON(!iter_is_iovec(i))) |
241699cd | 1323 | return ~0U; |
241699cd | 1324 | |
610c7a71 | 1325 | for (k = 0; k < i->nr_segs; k++) { |
de4f5fed JA |
1326 | const struct iovec *iov = iter_iov(i) + k; |
1327 | if (iov->iov_len) { | |
1328 | unsigned long base = (unsigned long)iov->iov_base; | |
610c7a71 AV |
1329 | if (v) // if not the first one |
1330 | res |= base | v; // this start | previous end | |
de4f5fed JA |
1331 | v = base + iov->iov_len; |
1332 | if (size <= iov->iov_len) | |
610c7a71 | 1333 | break; |
de4f5fed | 1334 | size -= iov->iov_len; |
610c7a71 AV |
1335 | } |
1336 | } | |
33844e66 | 1337 | return res; |
357f435d AV |
1338 | } |
1339 | EXPORT_SYMBOL(iov_iter_gap_alignment); | |
1340 | ||
3cf42da3 AV |
1341 | static int want_pages_array(struct page ***res, size_t size, |
1342 | size_t start, unsigned int maxpages) | |
acbdeb83 | 1343 | { |
3cf42da3 AV |
1344 | unsigned int count = DIV_ROUND_UP(size + start, PAGE_SIZE); |
1345 | ||
1346 | if (count > maxpages) | |
1347 | count = maxpages; | |
1348 | WARN_ON(!count); // caller should've prevented that | |
1349 | if (!*res) { | |
1350 | *res = kvmalloc_array(count, sizeof(struct page *), GFP_KERNEL); | |
1351 | if (!*res) | |
1352 | return 0; | |
1353 | } | |
1354 | return count; | |
acbdeb83 AV |
1355 | } |
1356 | ||
241699cd | 1357 | static ssize_t pipe_get_pages(struct iov_iter *i, |
acbdeb83 | 1358 | struct page ***pages, size_t maxsize, unsigned maxpages, |
241699cd AV |
1359 | size_t *start) |
1360 | { | |
746de1f8 | 1361 | unsigned int npages, count, off, chunk; |
acbdeb83 | 1362 | struct page **p; |
746de1f8 | 1363 | size_t left; |
241699cd AV |
1364 | |
1365 | if (!sanity(i)) | |
1366 | return -EFAULT; | |
1367 | ||
12d426ab | 1368 | *start = off = pipe_npages(i, &npages); |
3cf42da3 AV |
1369 | if (!npages) |
1370 | return -EFAULT; | |
1371 | count = want_pages_array(pages, maxsize, off, min(npages, maxpages)); | |
1372 | if (!count) | |
1373 | return -ENOMEM; | |
acbdeb83 | 1374 | p = *pages; |
746de1f8 AV |
1375 | for (npages = 0, left = maxsize ; npages < count; npages++, left -= chunk) { |
1376 | struct page *page = append_pipe(i, left, &off); | |
85200084 AV |
1377 | if (!page) |
1378 | break; | |
746de1f8 | 1379 | chunk = min_t(size_t, left, PAGE_SIZE - off); |
85200084 | 1380 | get_page(*p++ = page); |
85200084 AV |
1381 | } |
1382 | if (!npages) | |
1383 | return -EFAULT; | |
746de1f8 | 1384 | return maxsize - left; |
241699cd AV |
1385 | } |
1386 | ||
7ff50620 DH |
1387 | static ssize_t iter_xarray_populate_pages(struct page **pages, struct xarray *xa, |
1388 | pgoff_t index, unsigned int nr_pages) | |
1389 | { | |
1390 | XA_STATE(xas, xa, index); | |
1391 | struct page *page; | |
1392 | unsigned int ret = 0; | |
1393 | ||
1394 | rcu_read_lock(); | |
1395 | for (page = xas_load(&xas); page; page = xas_next(&xas)) { | |
1396 | if (xas_retry(&xas, page)) | |
1397 | continue; | |
1398 | ||
1399 | /* Has the page moved or been split? */ | |
1400 | if (unlikely(page != xas_reload(&xas))) { | |
1401 | xas_reset(&xas); | |
1402 | continue; | |
1403 | } | |
1404 | ||
1405 | pages[ret] = find_subpage(page, xas.xa_index); | |
1406 | get_page(pages[ret]); | |
1407 | if (++ret == nr_pages) | |
1408 | break; | |
1409 | } | |
1410 | rcu_read_unlock(); | |
1411 | return ret; | |
1412 | } | |
1413 | ||
1414 | static ssize_t iter_xarray_get_pages(struct iov_iter *i, | |
68fe506f | 1415 | struct page ***pages, size_t maxsize, |
7ff50620 DH |
1416 | unsigned maxpages, size_t *_start_offset) |
1417 | { | |
3cf42da3 AV |
1418 | unsigned nr, offset, count; |
1419 | pgoff_t index; | |
7ff50620 DH |
1420 | loff_t pos; |
1421 | ||
7ff50620 DH |
1422 | pos = i->xarray_start + i->iov_offset; |
1423 | index = pos >> PAGE_SHIFT; | |
1424 | offset = pos & ~PAGE_MASK; | |
1425 | *_start_offset = offset; | |
1426 | ||
3cf42da3 AV |
1427 | count = want_pages_array(pages, maxsize, offset, maxpages); |
1428 | if (!count) | |
1429 | return -ENOMEM; | |
68fe506f | 1430 | nr = iter_xarray_populate_pages(*pages, i->xarray, index, count); |
7ff50620 DH |
1431 | if (nr == 0) |
1432 | return 0; | |
1433 | ||
eba2d3d7 | 1434 | maxsize = min_t(size_t, nr * PAGE_SIZE - offset, maxsize); |
310d9d5a AV |
1435 | i->iov_offset += maxsize; |
1436 | i->count -= maxsize; | |
eba2d3d7 | 1437 | return maxsize; |
7ff50620 DH |
1438 | } |
1439 | ||
fcb14cb1 | 1440 | /* must be done on non-empty ITER_UBUF or ITER_IOVEC one */ |
dd45ab9d | 1441 | static unsigned long first_iovec_segment(const struct iov_iter *i, size_t *size) |
3d671ca6 AV |
1442 | { |
1443 | size_t skip; | |
1444 | long k; | |
1445 | ||
fcb14cb1 AV |
1446 | if (iter_is_ubuf(i)) |
1447 | return (unsigned long)i->ubuf + i->iov_offset; | |
1448 | ||
3d671ca6 | 1449 | for (k = 0, skip = i->iov_offset; k < i->nr_segs; k++, skip = 0) { |
de4f5fed JA |
1450 | const struct iovec *iov = iter_iov(i) + k; |
1451 | size_t len = iov->iov_len - skip; | |
3d671ca6 AV |
1452 | |
1453 | if (unlikely(!len)) | |
1454 | continue; | |
59dbd7d0 AV |
1455 | if (*size > len) |
1456 | *size = len; | |
de4f5fed | 1457 | return (unsigned long)iov->iov_base + skip; |
3d671ca6 AV |
1458 | } |
1459 | BUG(); // if it had been empty, we wouldn't get called | |
1460 | } | |
1461 | ||
1462 | /* must be done on non-empty ITER_BVEC one */ | |
1463 | static struct page *first_bvec_segment(const struct iov_iter *i, | |
59dbd7d0 | 1464 | size_t *size, size_t *start) |
3d671ca6 AV |
1465 | { |
1466 | struct page *page; | |
1467 | size_t skip = i->iov_offset, len; | |
1468 | ||
1469 | len = i->bvec->bv_len - skip; | |
59dbd7d0 AV |
1470 | if (*size > len) |
1471 | *size = len; | |
3d671ca6 AV |
1472 | skip += i->bvec->bv_offset; |
1473 | page = i->bvec->bv_page + skip / PAGE_SIZE; | |
dda8e5d1 | 1474 | *start = skip % PAGE_SIZE; |
3d671ca6 AV |
1475 | return page; |
1476 | } | |
1477 | ||
451c0ba9 AV |
1478 | static ssize_t __iov_iter_get_pages_alloc(struct iov_iter *i, |
1479 | struct page ***pages, size_t maxsize, | |
d8207640 | 1480 | unsigned int maxpages, size_t *start, |
f62e52d1 | 1481 | iov_iter_extraction_t extraction_flags) |
62a8067a | 1482 | { |
f62e52d1 | 1483 | unsigned int n, gup_flags = 0; |
3d671ca6 | 1484 | |
e5393fae AV |
1485 | if (maxsize > i->count) |
1486 | maxsize = i->count; | |
451c0ba9 | 1487 | if (!maxsize) |
3d671ca6 | 1488 | return 0; |
7392ed17 AV |
1489 | if (maxsize > MAX_RW_COUNT) |
1490 | maxsize = MAX_RW_COUNT; | |
f62e52d1 DH |
1491 | if (extraction_flags & ITER_ALLOW_P2PDMA) |
1492 | gup_flags |= FOLL_PCI_P2PDMA; | |
e5393fae | 1493 | |
fcb14cb1 | 1494 | if (likely(user_backed_iter(i))) { |
3d671ca6 | 1495 | unsigned long addr; |
3cf42da3 | 1496 | int res; |
e5393fae | 1497 | |
3337ab08 AG |
1498 | if (iov_iter_rw(i) != WRITE) |
1499 | gup_flags |= FOLL_WRITE; | |
1500 | if (i->nofault) | |
1501 | gup_flags |= FOLL_NOFAULT; | |
1502 | ||
dd45ab9d AV |
1503 | addr = first_iovec_segment(i, &maxsize); |
1504 | *start = addr % PAGE_SIZE; | |
1505 | addr &= PAGE_MASK; | |
3cf42da3 AV |
1506 | n = want_pages_array(pages, maxsize, *start, maxpages); |
1507 | if (!n) | |
1508 | return -ENOMEM; | |
451c0ba9 | 1509 | res = get_user_pages_fast(addr, n, gup_flags, *pages); |
814a6674 | 1510 | if (unlikely(res <= 0)) |
e5393fae | 1511 | return res; |
eba2d3d7 AV |
1512 | maxsize = min_t(size_t, maxsize, res * PAGE_SIZE - *start); |
1513 | iov_iter_advance(i, maxsize); | |
1514 | return maxsize; | |
3d671ca6 AV |
1515 | } |
1516 | if (iov_iter_is_bvec(i)) { | |
451c0ba9 | 1517 | struct page **p; |
3d671ca6 AV |
1518 | struct page *page; |
1519 | ||
59dbd7d0 | 1520 | page = first_bvec_segment(i, &maxsize, start); |
3cf42da3 AV |
1521 | n = want_pages_array(pages, maxsize, *start, maxpages); |
1522 | if (!n) | |
1523 | return -ENOMEM; | |
451c0ba9 | 1524 | p = *pages; |
dda8e5d1 | 1525 | for (int k = 0; k < n; k++) |
eba2d3d7 AV |
1526 | get_page(p[k] = page + k); |
1527 | maxsize = min_t(size_t, maxsize, n * PAGE_SIZE - *start); | |
310d9d5a AV |
1528 | i->count -= maxsize; |
1529 | i->iov_offset += maxsize; | |
1530 | if (i->iov_offset == i->bvec->bv_len) { | |
1531 | i->iov_offset = 0; | |
1532 | i->bvec++; | |
1533 | i->nr_segs--; | |
1534 | } | |
eba2d3d7 | 1535 | return maxsize; |
3d671ca6 AV |
1536 | } |
1537 | if (iov_iter_is_pipe(i)) | |
451c0ba9 | 1538 | return pipe_get_pages(i, pages, maxsize, maxpages, start); |
3d671ca6 | 1539 | if (iov_iter_is_xarray(i)) |
451c0ba9 | 1540 | return iter_xarray_get_pages(i, pages, maxsize, maxpages, start); |
3d671ca6 | 1541 | return -EFAULT; |
62a8067a | 1542 | } |
62a8067a | 1543 | |
d8207640 | 1544 | ssize_t iov_iter_get_pages(struct iov_iter *i, |
451c0ba9 | 1545 | struct page **pages, size_t maxsize, unsigned maxpages, |
f62e52d1 | 1546 | size_t *start, iov_iter_extraction_t extraction_flags) |
62a8067a | 1547 | { |
451c0ba9 | 1548 | if (!maxpages) |
3d671ca6 | 1549 | return 0; |
451c0ba9 | 1550 | BUG_ON(!pages); |
3d671ca6 | 1551 | |
d8207640 | 1552 | return __iov_iter_get_pages_alloc(i, &pages, maxsize, maxpages, |
f62e52d1 | 1553 | start, extraction_flags); |
d8207640 LG |
1554 | } |
1555 | EXPORT_SYMBOL_GPL(iov_iter_get_pages); | |
1556 | ||
1557 | ssize_t iov_iter_get_pages2(struct iov_iter *i, struct page **pages, | |
1558 | size_t maxsize, unsigned maxpages, size_t *start) | |
1559 | { | |
1560 | return iov_iter_get_pages(i, pages, maxsize, maxpages, start, 0); | |
62a8067a | 1561 | } |
eba2d3d7 | 1562 | EXPORT_SYMBOL(iov_iter_get_pages2); |
91329559 | 1563 | |
d8207640 | 1564 | ssize_t iov_iter_get_pages_alloc(struct iov_iter *i, |
91329559 | 1565 | struct page ***pages, size_t maxsize, |
f62e52d1 | 1566 | size_t *start, iov_iter_extraction_t extraction_flags) |
91329559 AV |
1567 | { |
1568 | ssize_t len; | |
1569 | ||
1570 | *pages = NULL; | |
1571 | ||
d8207640 | 1572 | len = __iov_iter_get_pages_alloc(i, pages, maxsize, ~0U, start, |
f62e52d1 | 1573 | extraction_flags); |
91329559 AV |
1574 | if (len <= 0) { |
1575 | kvfree(*pages); | |
1576 | *pages = NULL; | |
1577 | } | |
1578 | return len; | |
1579 | } | |
d8207640 LG |
1580 | EXPORT_SYMBOL_GPL(iov_iter_get_pages_alloc); |
1581 | ||
1582 | ssize_t iov_iter_get_pages_alloc2(struct iov_iter *i, | |
1583 | struct page ***pages, size_t maxsize, size_t *start) | |
1584 | { | |
1585 | return iov_iter_get_pages_alloc(i, pages, maxsize, start, 0); | |
91329559 | 1586 | } |
eba2d3d7 | 1587 | EXPORT_SYMBOL(iov_iter_get_pages_alloc2); |
62a8067a | 1588 | |
a604ec7e AV |
1589 | size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum, |
1590 | struct iov_iter *i) | |
1591 | { | |
a604ec7e | 1592 | __wsum sum, next; |
a604ec7e | 1593 | sum = *csum; |
a41dad90 | 1594 | if (WARN_ON_ONCE(!i->data_source)) |
241699cd | 1595 | return 0; |
a41dad90 | 1596 | |
7baa5099 AV |
1597 | iterate_and_advance(i, bytes, base, len, off, ({ |
1598 | next = csum_and_copy_from_user(base, addr + off, len); | |
2495bdcc | 1599 | sum = csum_block_add(sum, next, off); |
7baa5099 | 1600 | next ? 0 : len; |
a604ec7e | 1601 | }), ({ |
7baa5099 | 1602 | sum = csum_and_memcpy(addr + off, base, len, sum, off); |
a604ec7e AV |
1603 | }) |
1604 | ) | |
1605 | *csum = sum; | |
1606 | return bytes; | |
1607 | } | |
1608 | EXPORT_SYMBOL(csum_and_copy_from_iter); | |
1609 | ||
52cbd23a | 1610 | size_t csum_and_copy_to_iter(const void *addr, size_t bytes, void *_csstate, |
a604ec7e AV |
1611 | struct iov_iter *i) |
1612 | { | |
52cbd23a | 1613 | struct csum_state *csstate = _csstate; |
a604ec7e | 1614 | __wsum sum, next; |
78e1f386 | 1615 | |
a41dad90 AV |
1616 | if (WARN_ON_ONCE(i->data_source)) |
1617 | return 0; | |
78e1f386 | 1618 | if (unlikely(iov_iter_is_discard(i))) { |
c67f1fd2 AV |
1619 | // can't use csum_memcpy() for that one - data is not copied |
1620 | csstate->csum = csum_block_add(csstate->csum, | |
1621 | csum_partial(addr, bytes, 0), | |
1622 | csstate->off); | |
1623 | csstate->off += bytes; | |
1624 | return bytes; | |
241699cd | 1625 | } |
6852df12 AV |
1626 | |
1627 | sum = csum_shift(csstate->csum, csstate->off); | |
1628 | if (unlikely(iov_iter_is_pipe(i))) | |
1629 | bytes = csum_and_copy_to_pipe_iter(addr, bytes, i, &sum); | |
1630 | else iterate_and_advance(i, bytes, base, len, off, ({ | |
7baa5099 | 1631 | next = csum_and_copy_to_user(addr + off, base, len); |
2495bdcc | 1632 | sum = csum_block_add(sum, next, off); |
7baa5099 | 1633 | next ? 0 : len; |
a604ec7e | 1634 | }), ({ |
7baa5099 | 1635 | sum = csum_and_memcpy(base, addr + off, len, sum, off); |
a604ec7e AV |
1636 | }) |
1637 | ) | |
594e450b AV |
1638 | csstate->csum = csum_shift(sum, csstate->off); |
1639 | csstate->off += bytes; | |
a604ec7e AV |
1640 | return bytes; |
1641 | } | |
1642 | EXPORT_SYMBOL(csum_and_copy_to_iter); | |
1643 | ||
d05f4435 SG |
1644 | size_t hash_and_copy_to_iter(const void *addr, size_t bytes, void *hashp, |
1645 | struct iov_iter *i) | |
1646 | { | |
7999096f | 1647 | #ifdef CONFIG_CRYPTO_HASH |
d05f4435 SG |
1648 | struct ahash_request *hash = hashp; |
1649 | struct scatterlist sg; | |
1650 | size_t copied; | |
1651 | ||
1652 | copied = copy_to_iter(addr, bytes, i); | |
1653 | sg_init_one(&sg, addr, copied); | |
1654 | ahash_request_set_crypt(hash, &sg, NULL, copied); | |
1655 | crypto_ahash_update(hash); | |
1656 | return copied; | |
27fad74a Y |
1657 | #else |
1658 | return 0; | |
1659 | #endif | |
d05f4435 SG |
1660 | } |
1661 | EXPORT_SYMBOL(hash_and_copy_to_iter); | |
1662 | ||
66531c65 | 1663 | static int iov_npages(const struct iov_iter *i, int maxpages) |
62a8067a | 1664 | { |
66531c65 AV |
1665 | size_t skip = i->iov_offset, size = i->count; |
1666 | const struct iovec *p; | |
e0f2dc40 AV |
1667 | int npages = 0; |
1668 | ||
de4f5fed | 1669 | for (p = iter_iov(i); size; skip = 0, p++) { |
66531c65 AV |
1670 | unsigned offs = offset_in_page(p->iov_base + skip); |
1671 | size_t len = min(p->iov_len - skip, size); | |
e0f2dc40 | 1672 | |
66531c65 AV |
1673 | if (len) { |
1674 | size -= len; | |
1675 | npages += DIV_ROUND_UP(offs + len, PAGE_SIZE); | |
1676 | if (unlikely(npages > maxpages)) | |
1677 | return maxpages; | |
1678 | } | |
1679 | } | |
1680 | return npages; | |
1681 | } | |
1682 | ||
1683 | static int bvec_npages(const struct iov_iter *i, int maxpages) | |
1684 | { | |
1685 | size_t skip = i->iov_offset, size = i->count; | |
1686 | const struct bio_vec *p; | |
1687 | int npages = 0; | |
1688 | ||
1689 | for (p = i->bvec; size; skip = 0, p++) { | |
1690 | unsigned offs = (p->bv_offset + skip) % PAGE_SIZE; | |
1691 | size_t len = min(p->bv_len - skip, size); | |
1692 | ||
1693 | size -= len; | |
1694 | npages += DIV_ROUND_UP(offs + len, PAGE_SIZE); | |
1695 | if (unlikely(npages > maxpages)) | |
1696 | return maxpages; | |
1697 | } | |
1698 | return npages; | |
1699 | } | |
1700 | ||
1701 | int iov_iter_npages(const struct iov_iter *i, int maxpages) | |
1702 | { | |
1703 | if (unlikely(!i->count)) | |
1704 | return 0; | |
fcb14cb1 AV |
1705 | if (likely(iter_is_ubuf(i))) { |
1706 | unsigned offs = offset_in_page(i->ubuf + i->iov_offset); | |
1707 | int npages = DIV_ROUND_UP(offs + i->count, PAGE_SIZE); | |
1708 | return min(npages, maxpages); | |
1709 | } | |
66531c65 AV |
1710 | /* iovec and kvec have identical layouts */ |
1711 | if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) | |
1712 | return iov_npages(i, maxpages); | |
1713 | if (iov_iter_is_bvec(i)) | |
1714 | return bvec_npages(i, maxpages); | |
1715 | if (iov_iter_is_pipe(i)) { | |
66531c65 | 1716 | int npages; |
241699cd AV |
1717 | |
1718 | if (!sanity(i)) | |
1719 | return 0; | |
1720 | ||
12d426ab | 1721 | pipe_npages(i, &npages); |
66531c65 AV |
1722 | return min(npages, maxpages); |
1723 | } | |
1724 | if (iov_iter_is_xarray(i)) { | |
e4f8df86 AV |
1725 | unsigned offset = (i->xarray_start + i->iov_offset) % PAGE_SIZE; |
1726 | int npages = DIV_ROUND_UP(offset + i->count, PAGE_SIZE); | |
66531c65 AV |
1727 | return min(npages, maxpages); |
1728 | } | |
1729 | return 0; | |
62a8067a | 1730 | } |
f67da30c | 1731 | EXPORT_SYMBOL(iov_iter_npages); |
4b8164b9 AV |
1732 | |
1733 | const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags) | |
1734 | { | |
1735 | *new = *old; | |
00e23707 | 1736 | if (unlikely(iov_iter_is_pipe(new))) { |
241699cd AV |
1737 | WARN_ON(1); |
1738 | return NULL; | |
1739 | } | |
00e23707 | 1740 | if (iov_iter_is_bvec(new)) |
4b8164b9 AV |
1741 | return new->bvec = kmemdup(new->bvec, |
1742 | new->nr_segs * sizeof(struct bio_vec), | |
1743 | flags); | |
fcb14cb1 | 1744 | else if (iov_iter_is_kvec(new) || iter_is_iovec(new)) |
4b8164b9 | 1745 | /* iovec and kvec have identical layout */ |
de4f5fed | 1746 | return new->__iov = kmemdup(new->__iov, |
4b8164b9 AV |
1747 | new->nr_segs * sizeof(struct iovec), |
1748 | flags); | |
fcb14cb1 | 1749 | return NULL; |
4b8164b9 AV |
1750 | } |
1751 | EXPORT_SYMBOL(dup_iter); | |
bc917be8 | 1752 | |
50f9a76e | 1753 | static __noclone int copy_compat_iovec_from_user(struct iovec *iov, |
bfdc5970 CH |
1754 | const struct iovec __user *uvec, unsigned long nr_segs) |
1755 | { | |
1756 | const struct compat_iovec __user *uiov = | |
1757 | (const struct compat_iovec __user *)uvec; | |
1758 | int ret = -EFAULT, i; | |
1759 | ||
a959a978 | 1760 | if (!user_access_begin(uiov, nr_segs * sizeof(*uiov))) |
bfdc5970 CH |
1761 | return -EFAULT; |
1762 | ||
1763 | for (i = 0; i < nr_segs; i++) { | |
1764 | compat_uptr_t buf; | |
1765 | compat_ssize_t len; | |
1766 | ||
1767 | unsafe_get_user(len, &uiov[i].iov_len, uaccess_end); | |
1768 | unsafe_get_user(buf, &uiov[i].iov_base, uaccess_end); | |
1769 | ||
1770 | /* check for compat_size_t not fitting in compat_ssize_t .. */ | |
1771 | if (len < 0) { | |
1772 | ret = -EINVAL; | |
1773 | goto uaccess_end; | |
1774 | } | |
1775 | iov[i].iov_base = compat_ptr(buf); | |
1776 | iov[i].iov_len = len; | |
1777 | } | |
1778 | ||
1779 | ret = 0; | |
1780 | uaccess_end: | |
1781 | user_access_end(); | |
1782 | return ret; | |
1783 | } | |
1784 | ||
1785 | static int copy_iovec_from_user(struct iovec *iov, | |
487c20b0 | 1786 | const struct iovec __user *uiov, unsigned long nr_segs) |
fb041b59 | 1787 | { |
487c20b0 | 1788 | int ret = -EFAULT; |
fb041b59 | 1789 | |
487c20b0 | 1790 | if (!user_access_begin(uiov, nr_segs * sizeof(*uiov))) |
bfdc5970 | 1791 | return -EFAULT; |
fb041b59 | 1792 | |
487c20b0 LT |
1793 | do { |
1794 | void __user *buf; | |
1795 | ssize_t len; | |
1796 | ||
1797 | unsafe_get_user(len, &uiov->iov_len, uaccess_end); | |
1798 | unsafe_get_user(buf, &uiov->iov_base, uaccess_end); | |
1799 | ||
1800 | /* check for size_t not fitting in ssize_t .. */ | |
1801 | if (unlikely(len < 0)) { | |
1802 | ret = -EINVAL; | |
1803 | goto uaccess_end; | |
1804 | } | |
1805 | iov->iov_base = buf; | |
1806 | iov->iov_len = len; | |
1807 | ||
1808 | uiov++; iov++; | |
1809 | } while (--nr_segs); | |
1810 | ||
1811 | ret = 0; | |
1812 | uaccess_end: | |
1813 | user_access_end(); | |
1814 | return ret; | |
bfdc5970 CH |
1815 | } |
1816 | ||
1817 | struct iovec *iovec_from_user(const struct iovec __user *uvec, | |
1818 | unsigned long nr_segs, unsigned long fast_segs, | |
1819 | struct iovec *fast_iov, bool compat) | |
1820 | { | |
1821 | struct iovec *iov = fast_iov; | |
1822 | int ret; | |
1823 | ||
fb041b59 | 1824 | /* |
bfdc5970 CH |
1825 | * SuS says "The readv() function *may* fail if the iovcnt argument was |
1826 | * less than or equal to 0, or greater than {IOV_MAX}. Linux has | |
1827 | * traditionally returned zero for zero segments, so... | |
fb041b59 | 1828 | */ |
bfdc5970 CH |
1829 | if (nr_segs == 0) |
1830 | return iov; | |
1831 | if (nr_segs > UIO_MAXIOV) | |
1832 | return ERR_PTR(-EINVAL); | |
fb041b59 DL |
1833 | if (nr_segs > fast_segs) { |
1834 | iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL); | |
bfdc5970 CH |
1835 | if (!iov) |
1836 | return ERR_PTR(-ENOMEM); | |
fb041b59 | 1837 | } |
bfdc5970 | 1838 | |
487c20b0 | 1839 | if (unlikely(compat)) |
bfdc5970 CH |
1840 | ret = copy_compat_iovec_from_user(iov, uvec, nr_segs); |
1841 | else | |
1842 | ret = copy_iovec_from_user(iov, uvec, nr_segs); | |
1843 | if (ret) { | |
1844 | if (iov != fast_iov) | |
1845 | kfree(iov); | |
1846 | return ERR_PTR(ret); | |
1847 | } | |
1848 | ||
1849 | return iov; | |
1850 | } | |
1851 | ||
3b2deb0e JA |
1852 | /* |
1853 | * Single segment iovec supplied by the user, import it as ITER_UBUF. | |
1854 | */ | |
1855 | static ssize_t __import_iovec_ubuf(int type, const struct iovec __user *uvec, | |
1856 | struct iovec **iovp, struct iov_iter *i, | |
1857 | bool compat) | |
1858 | { | |
1859 | struct iovec *iov = *iovp; | |
1860 | ssize_t ret; | |
1861 | ||
1862 | if (compat) | |
1863 | ret = copy_compat_iovec_from_user(iov, uvec, 1); | |
1864 | else | |
1865 | ret = copy_iovec_from_user(iov, uvec, 1); | |
1866 | if (unlikely(ret)) | |
1867 | return ret; | |
1868 | ||
1869 | ret = import_ubuf(type, iov->iov_base, iov->iov_len, i); | |
1870 | if (unlikely(ret)) | |
1871 | return ret; | |
1872 | *iovp = NULL; | |
1873 | return i->count; | |
1874 | } | |
1875 | ||
bfdc5970 CH |
1876 | ssize_t __import_iovec(int type, const struct iovec __user *uvec, |
1877 | unsigned nr_segs, unsigned fast_segs, struct iovec **iovp, | |
1878 | struct iov_iter *i, bool compat) | |
1879 | { | |
1880 | ssize_t total_len = 0; | |
1881 | unsigned long seg; | |
1882 | struct iovec *iov; | |
1883 | ||
3b2deb0e JA |
1884 | if (nr_segs == 1) |
1885 | return __import_iovec_ubuf(type, uvec, iovp, i, compat); | |
1886 | ||
bfdc5970 CH |
1887 | iov = iovec_from_user(uvec, nr_segs, fast_segs, *iovp, compat); |
1888 | if (IS_ERR(iov)) { | |
1889 | *iovp = NULL; | |
1890 | return PTR_ERR(iov); | |
fb041b59 DL |
1891 | } |
1892 | ||
1893 | /* | |
bfdc5970 CH |
1894 | * According to the Single Unix Specification we should return EINVAL if |
1895 | * an element length is < 0 when cast to ssize_t or if the total length | |
1896 | * would overflow the ssize_t return value of the system call. | |
fb041b59 DL |
1897 | * |
1898 | * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the | |
1899 | * overflow case. | |
1900 | */ | |
fb041b59 | 1901 | for (seg = 0; seg < nr_segs; seg++) { |
fb041b59 DL |
1902 | ssize_t len = (ssize_t)iov[seg].iov_len; |
1903 | ||
bfdc5970 CH |
1904 | if (!access_ok(iov[seg].iov_base, len)) { |
1905 | if (iov != *iovp) | |
1906 | kfree(iov); | |
1907 | *iovp = NULL; | |
1908 | return -EFAULT; | |
fb041b59 | 1909 | } |
bfdc5970 CH |
1910 | |
1911 | if (len > MAX_RW_COUNT - total_len) { | |
1912 | len = MAX_RW_COUNT - total_len; | |
fb041b59 DL |
1913 | iov[seg].iov_len = len; |
1914 | } | |
bfdc5970 | 1915 | total_len += len; |
fb041b59 | 1916 | } |
bfdc5970 CH |
1917 | |
1918 | iov_iter_init(i, type, iov, nr_segs, total_len); | |
1919 | if (iov == *iovp) | |
1920 | *iovp = NULL; | |
1921 | else | |
1922 | *iovp = iov; | |
1923 | return total_len; | |
fb041b59 DL |
1924 | } |
1925 | ||
ffecee4f VN |
1926 | /** |
1927 | * import_iovec() - Copy an array of &struct iovec from userspace | |
1928 | * into the kernel, check that it is valid, and initialize a new | |
1929 | * &struct iov_iter iterator to access it. | |
1930 | * | |
1931 | * @type: One of %READ or %WRITE. | |
bfdc5970 | 1932 | * @uvec: Pointer to the userspace array. |
ffecee4f VN |
1933 | * @nr_segs: Number of elements in userspace array. |
1934 | * @fast_segs: Number of elements in @iov. | |
bfdc5970 | 1935 | * @iovp: (input and output parameter) Pointer to pointer to (usually small |
ffecee4f VN |
1936 | * on-stack) kernel array. |
1937 | * @i: Pointer to iterator that will be initialized on success. | |
1938 | * | |
1939 | * If the array pointed to by *@iov is large enough to hold all @nr_segs, | |
1940 | * then this function places %NULL in *@iov on return. Otherwise, a new | |
1941 | * array will be allocated and the result placed in *@iov. This means that | |
1942 | * the caller may call kfree() on *@iov regardless of whether the small | |
1943 | * on-stack array was used or not (and regardless of whether this function | |
1944 | * returns an error or not). | |
1945 | * | |
87e5e6da | 1946 | * Return: Negative error code on error, bytes imported on success |
ffecee4f | 1947 | */ |
bfdc5970 | 1948 | ssize_t import_iovec(int type, const struct iovec __user *uvec, |
bc917be8 | 1949 | unsigned nr_segs, unsigned fast_segs, |
bfdc5970 | 1950 | struct iovec **iovp, struct iov_iter *i) |
bc917be8 | 1951 | { |
89cd35c5 CH |
1952 | return __import_iovec(type, uvec, nr_segs, fast_segs, iovp, i, |
1953 | in_compat_syscall()); | |
bc917be8 AV |
1954 | } |
1955 | EXPORT_SYMBOL(import_iovec); | |
1956 | ||
bc917be8 AV |
1957 | int import_single_range(int rw, void __user *buf, size_t len, |
1958 | struct iovec *iov, struct iov_iter *i) | |
1959 | { | |
1960 | if (len > MAX_RW_COUNT) | |
1961 | len = MAX_RW_COUNT; | |
96d4f267 | 1962 | if (unlikely(!access_ok(buf, len))) |
bc917be8 AV |
1963 | return -EFAULT; |
1964 | ||
e03ad4ee | 1965 | iov_iter_ubuf(i, rw, buf, len); |
bc917be8 AV |
1966 | return 0; |
1967 | } | |
e1267585 | 1968 | EXPORT_SYMBOL(import_single_range); |
8fb0f47a | 1969 | |
2ad9bd83 JA |
1970 | int import_ubuf(int rw, void __user *buf, size_t len, struct iov_iter *i) |
1971 | { | |
1972 | if (len > MAX_RW_COUNT) | |
1973 | len = MAX_RW_COUNT; | |
1974 | if (unlikely(!access_ok(buf, len))) | |
1975 | return -EFAULT; | |
1976 | ||
1977 | iov_iter_ubuf(i, rw, buf, len); | |
1978 | return 0; | |
1979 | } | |
1980 | ||
8fb0f47a JA |
1981 | /** |
1982 | * iov_iter_restore() - Restore a &struct iov_iter to the same state as when | |
1983 | * iov_iter_save_state() was called. | |
1984 | * | |
1985 | * @i: &struct iov_iter to restore | |
1986 | * @state: state to restore from | |
1987 | * | |
1988 | * Used after iov_iter_save_state() to bring restore @i, if operations may | |
1989 | * have advanced it. | |
1990 | * | |
1991 | * Note: only works on ITER_IOVEC, ITER_BVEC, and ITER_KVEC | |
1992 | */ | |
1993 | void iov_iter_restore(struct iov_iter *i, struct iov_iter_state *state) | |
1994 | { | |
4397a17c KB |
1995 | if (WARN_ON_ONCE(!iov_iter_is_bvec(i) && !iter_is_iovec(i) && |
1996 | !iter_is_ubuf(i)) && !iov_iter_is_kvec(i)) | |
8fb0f47a JA |
1997 | return; |
1998 | i->iov_offset = state->iov_offset; | |
1999 | i->count = state->count; | |
fcb14cb1 AV |
2000 | if (iter_is_ubuf(i)) |
2001 | return; | |
8fb0f47a JA |
2002 | /* |
2003 | * For the *vec iters, nr_segs + iov is constant - if we increment | |
2004 | * the vec, then we also decrement the nr_segs count. Hence we don't | |
2005 | * need to track both of these, just one is enough and we can deduct | |
2006 | * the other from that. ITER_KVEC and ITER_IOVEC are the same struct | |
2007 | * size, so we can just increment the iov pointer as they are unionzed. | |
2008 | * ITER_BVEC _may_ be the same size on some archs, but on others it is | |
2009 | * not. Be safe and handle it separately. | |
2010 | */ | |
2011 | BUILD_BUG_ON(sizeof(struct iovec) != sizeof(struct kvec)); | |
2012 | if (iov_iter_is_bvec(i)) | |
2013 | i->bvec -= state->nr_segs - i->nr_segs; | |
2014 | else | |
de4f5fed | 2015 | i->__iov -= state->nr_segs - i->nr_segs; |
8fb0f47a JA |
2016 | i->nr_segs = state->nr_segs; |
2017 | } | |
7d58fe73 DH |
2018 | |
2019 | /* | |
2020 | * Extract a list of contiguous pages from an ITER_XARRAY iterator. This does not | |
2021 | * get references on the pages, nor does it get a pin on them. | |
2022 | */ | |
2023 | static ssize_t iov_iter_extract_xarray_pages(struct iov_iter *i, | |
2024 | struct page ***pages, size_t maxsize, | |
2025 | unsigned int maxpages, | |
2026 | iov_iter_extraction_t extraction_flags, | |
2027 | size_t *offset0) | |
2028 | { | |
2029 | struct page *page, **p; | |
2030 | unsigned int nr = 0, offset; | |
2031 | loff_t pos = i->xarray_start + i->iov_offset; | |
2032 | pgoff_t index = pos >> PAGE_SHIFT; | |
2033 | XA_STATE(xas, i->xarray, index); | |
2034 | ||
2035 | offset = pos & ~PAGE_MASK; | |
2036 | *offset0 = offset; | |
2037 | ||
2038 | maxpages = want_pages_array(pages, maxsize, offset, maxpages); | |
2039 | if (!maxpages) | |
2040 | return -ENOMEM; | |
2041 | p = *pages; | |
2042 | ||
2043 | rcu_read_lock(); | |
2044 | for (page = xas_load(&xas); page; page = xas_next(&xas)) { | |
2045 | if (xas_retry(&xas, page)) | |
2046 | continue; | |
2047 | ||
2048 | /* Has the page moved or been split? */ | |
2049 | if (unlikely(page != xas_reload(&xas))) { | |
2050 | xas_reset(&xas); | |
2051 | continue; | |
2052 | } | |
2053 | ||
2054 | p[nr++] = find_subpage(page, xas.xa_index); | |
2055 | if (nr == maxpages) | |
2056 | break; | |
2057 | } | |
2058 | rcu_read_unlock(); | |
2059 | ||
2060 | maxsize = min_t(size_t, nr * PAGE_SIZE - offset, maxsize); | |
2061 | iov_iter_advance(i, maxsize); | |
2062 | return maxsize; | |
2063 | } | |
2064 | ||
2065 | /* | |
2066 | * Extract a list of contiguous pages from an ITER_BVEC iterator. This does | |
2067 | * not get references on the pages, nor does it get a pin on them. | |
2068 | */ | |
2069 | static ssize_t iov_iter_extract_bvec_pages(struct iov_iter *i, | |
2070 | struct page ***pages, size_t maxsize, | |
2071 | unsigned int maxpages, | |
2072 | iov_iter_extraction_t extraction_flags, | |
2073 | size_t *offset0) | |
2074 | { | |
2075 | struct page **p, *page; | |
2076 | size_t skip = i->iov_offset, offset; | |
2077 | int k; | |
2078 | ||
2079 | for (;;) { | |
2080 | if (i->nr_segs == 0) | |
2081 | return 0; | |
2082 | maxsize = min(maxsize, i->bvec->bv_len - skip); | |
2083 | if (maxsize) | |
2084 | break; | |
2085 | i->iov_offset = 0; | |
2086 | i->nr_segs--; | |
2087 | i->bvec++; | |
2088 | skip = 0; | |
2089 | } | |
2090 | ||
2091 | skip += i->bvec->bv_offset; | |
2092 | page = i->bvec->bv_page + skip / PAGE_SIZE; | |
2093 | offset = skip % PAGE_SIZE; | |
2094 | *offset0 = offset; | |
2095 | ||
2096 | maxpages = want_pages_array(pages, maxsize, offset, maxpages); | |
2097 | if (!maxpages) | |
2098 | return -ENOMEM; | |
2099 | p = *pages; | |
2100 | for (k = 0; k < maxpages; k++) | |
2101 | p[k] = page + k; | |
2102 | ||
2103 | maxsize = min_t(size_t, maxsize, maxpages * PAGE_SIZE - offset); | |
2104 | iov_iter_advance(i, maxsize); | |
2105 | return maxsize; | |
2106 | } | |
2107 | ||
2108 | /* | |
2109 | * Extract a list of virtually contiguous pages from an ITER_KVEC iterator. | |
2110 | * This does not get references on the pages, nor does it get a pin on them. | |
2111 | */ | |
2112 | static ssize_t iov_iter_extract_kvec_pages(struct iov_iter *i, | |
2113 | struct page ***pages, size_t maxsize, | |
2114 | unsigned int maxpages, | |
2115 | iov_iter_extraction_t extraction_flags, | |
2116 | size_t *offset0) | |
2117 | { | |
2118 | struct page **p, *page; | |
2119 | const void *kaddr; | |
2120 | size_t skip = i->iov_offset, offset, len; | |
2121 | int k; | |
2122 | ||
2123 | for (;;) { | |
2124 | if (i->nr_segs == 0) | |
2125 | return 0; | |
2126 | maxsize = min(maxsize, i->kvec->iov_len - skip); | |
2127 | if (maxsize) | |
2128 | break; | |
2129 | i->iov_offset = 0; | |
2130 | i->nr_segs--; | |
2131 | i->kvec++; | |
2132 | skip = 0; | |
2133 | } | |
2134 | ||
2135 | kaddr = i->kvec->iov_base + skip; | |
2136 | offset = (unsigned long)kaddr & ~PAGE_MASK; | |
2137 | *offset0 = offset; | |
2138 | ||
2139 | maxpages = want_pages_array(pages, maxsize, offset, maxpages); | |
2140 | if (!maxpages) | |
2141 | return -ENOMEM; | |
2142 | p = *pages; | |
2143 | ||
2144 | kaddr -= offset; | |
2145 | len = offset + maxsize; | |
2146 | for (k = 0; k < maxpages; k++) { | |
2147 | size_t seg = min_t(size_t, len, PAGE_SIZE); | |
2148 | ||
2149 | if (is_vmalloc_or_module_addr(kaddr)) | |
2150 | page = vmalloc_to_page(kaddr); | |
2151 | else | |
2152 | page = virt_to_page(kaddr); | |
2153 | ||
2154 | p[k] = page; | |
2155 | len -= seg; | |
2156 | kaddr += PAGE_SIZE; | |
2157 | } | |
2158 | ||
2159 | maxsize = min_t(size_t, maxsize, maxpages * PAGE_SIZE - offset); | |
2160 | iov_iter_advance(i, maxsize); | |
2161 | return maxsize; | |
2162 | } | |
2163 | ||
2164 | /* | |
2165 | * Extract a list of contiguous pages from a user iterator and get a pin on | |
2166 | * each of them. This should only be used if the iterator is user-backed | |
2167 | * (IOBUF/UBUF). | |
2168 | * | |
2169 | * It does not get refs on the pages, but the pages must be unpinned by the | |
2170 | * caller once the transfer is complete. | |
2171 | * | |
2172 | * This is safe to be used where background IO/DMA *is* going to be modifying | |
2173 | * the buffer; using a pin rather than a ref makes forces fork() to give the | |
2174 | * child a copy of the page. | |
2175 | */ | |
2176 | static ssize_t iov_iter_extract_user_pages(struct iov_iter *i, | |
2177 | struct page ***pages, | |
2178 | size_t maxsize, | |
2179 | unsigned int maxpages, | |
2180 | iov_iter_extraction_t extraction_flags, | |
2181 | size_t *offset0) | |
2182 | { | |
2183 | unsigned long addr; | |
2184 | unsigned int gup_flags = 0; | |
2185 | size_t offset; | |
2186 | int res; | |
2187 | ||
2188 | if (i->data_source == ITER_DEST) | |
2189 | gup_flags |= FOLL_WRITE; | |
2190 | if (extraction_flags & ITER_ALLOW_P2PDMA) | |
2191 | gup_flags |= FOLL_PCI_P2PDMA; | |
2192 | if (i->nofault) | |
2193 | gup_flags |= FOLL_NOFAULT; | |
2194 | ||
2195 | addr = first_iovec_segment(i, &maxsize); | |
2196 | *offset0 = offset = addr % PAGE_SIZE; | |
2197 | addr &= PAGE_MASK; | |
2198 | maxpages = want_pages_array(pages, maxsize, offset, maxpages); | |
2199 | if (!maxpages) | |
2200 | return -ENOMEM; | |
2201 | res = pin_user_pages_fast(addr, maxpages, gup_flags, *pages); | |
2202 | if (unlikely(res <= 0)) | |
2203 | return res; | |
2204 | maxsize = min_t(size_t, maxsize, res * PAGE_SIZE - offset); | |
2205 | iov_iter_advance(i, maxsize); | |
2206 | return maxsize; | |
2207 | } | |
2208 | ||
2209 | /** | |
2210 | * iov_iter_extract_pages - Extract a list of contiguous pages from an iterator | |
2211 | * @i: The iterator to extract from | |
2212 | * @pages: Where to return the list of pages | |
2213 | * @maxsize: The maximum amount of iterator to extract | |
2214 | * @maxpages: The maximum size of the list of pages | |
2215 | * @extraction_flags: Flags to qualify request | |
2216 | * @offset0: Where to return the starting offset into (*@pages)[0] | |
2217 | * | |
2218 | * Extract a list of contiguous pages from the current point of the iterator, | |
2219 | * advancing the iterator. The maximum number of pages and the maximum amount | |
2220 | * of page contents can be set. | |
2221 | * | |
2222 | * If *@pages is NULL, a page list will be allocated to the required size and | |
2223 | * *@pages will be set to its base. If *@pages is not NULL, it will be assumed | |
2224 | * that the caller allocated a page list at least @maxpages in size and this | |
2225 | * will be filled in. | |
2226 | * | |
2227 | * @extraction_flags can have ITER_ALLOW_P2PDMA set to request peer-to-peer DMA | |
2228 | * be allowed on the pages extracted. | |
2229 | * | |
2230 | * The iov_iter_extract_will_pin() function can be used to query how cleanup | |
2231 | * should be performed. | |
2232 | * | |
2233 | * Extra refs or pins on the pages may be obtained as follows: | |
2234 | * | |
2235 | * (*) If the iterator is user-backed (ITER_IOVEC/ITER_UBUF), pins will be | |
2236 | * added to the pages, but refs will not be taken. | |
2237 | * iov_iter_extract_will_pin() will return true. | |
2238 | * | |
2239 | * (*) If the iterator is ITER_KVEC, ITER_BVEC or ITER_XARRAY, the pages are | |
2240 | * merely listed; no extra refs or pins are obtained. | |
2241 | * iov_iter_extract_will_pin() will return 0. | |
2242 | * | |
2243 | * Note also: | |
2244 | * | |
2245 | * (*) Use with ITER_DISCARD is not supported as that has no content. | |
2246 | * | |
2247 | * On success, the function sets *@pages to the new pagelist, if allocated, and | |
2248 | * sets *offset0 to the offset into the first page. | |
2249 | * | |
2250 | * It may also return -ENOMEM and -EFAULT. | |
2251 | */ | |
2252 | ssize_t iov_iter_extract_pages(struct iov_iter *i, | |
2253 | struct page ***pages, | |
2254 | size_t maxsize, | |
2255 | unsigned int maxpages, | |
2256 | iov_iter_extraction_t extraction_flags, | |
2257 | size_t *offset0) | |
2258 | { | |
2259 | maxsize = min_t(size_t, min_t(size_t, maxsize, i->count), MAX_RW_COUNT); | |
2260 | if (!maxsize) | |
2261 | return 0; | |
2262 | ||
2263 | if (likely(user_backed_iter(i))) | |
2264 | return iov_iter_extract_user_pages(i, pages, maxsize, | |
2265 | maxpages, extraction_flags, | |
2266 | offset0); | |
2267 | if (iov_iter_is_kvec(i)) | |
2268 | return iov_iter_extract_kvec_pages(i, pages, maxsize, | |
2269 | maxpages, extraction_flags, | |
2270 | offset0); | |
2271 | if (iov_iter_is_bvec(i)) | |
2272 | return iov_iter_extract_bvec_pages(i, pages, maxsize, | |
2273 | maxpages, extraction_flags, | |
2274 | offset0); | |
2275 | if (iov_iter_is_xarray(i)) | |
2276 | return iov_iter_extract_xarray_pages(i, pages, maxsize, | |
2277 | maxpages, extraction_flags, | |
2278 | offset0); | |
2279 | return -EFAULT; | |
2280 | } | |
2281 | EXPORT_SYMBOL_GPL(iov_iter_extract_pages); |