Commit | Line | Data |
---|---|---|
4f18cd31 AV |
1 | #include <linux/export.h> |
2 | #include <linux/uio.h> | |
3 | #include <linux/pagemap.h> | |
91f79c43 AV |
4 | #include <linux/slab.h> |
5 | #include <linux/vmalloc.h> | |
4f18cd31 AV |
6 | |
7 | size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes, | |
8 | struct iov_iter *i) | |
9 | { | |
10 | size_t skip, copy, left, wanted; | |
11 | const struct iovec *iov; | |
12 | char __user *buf; | |
13 | void *kaddr, *from; | |
14 | ||
15 | if (unlikely(bytes > i->count)) | |
16 | bytes = i->count; | |
17 | ||
18 | if (unlikely(!bytes)) | |
19 | return 0; | |
20 | ||
21 | wanted = bytes; | |
22 | iov = i->iov; | |
23 | skip = i->iov_offset; | |
24 | buf = iov->iov_base + skip; | |
25 | copy = min(bytes, iov->iov_len - skip); | |
26 | ||
27 | if (!fault_in_pages_writeable(buf, copy)) { | |
28 | kaddr = kmap_atomic(page); | |
29 | from = kaddr + offset; | |
30 | ||
31 | /* first chunk, usually the only one */ | |
32 | left = __copy_to_user_inatomic(buf, from, copy); | |
33 | copy -= left; | |
34 | skip += copy; | |
35 | from += copy; | |
36 | bytes -= copy; | |
37 | ||
38 | while (unlikely(!left && bytes)) { | |
39 | iov++; | |
40 | buf = iov->iov_base; | |
41 | copy = min(bytes, iov->iov_len); | |
42 | left = __copy_to_user_inatomic(buf, from, copy); | |
43 | copy -= left; | |
44 | skip = copy; | |
45 | from += copy; | |
46 | bytes -= copy; | |
47 | } | |
48 | if (likely(!bytes)) { | |
49 | kunmap_atomic(kaddr); | |
50 | goto done; | |
51 | } | |
52 | offset = from - kaddr; | |
53 | buf += copy; | |
54 | kunmap_atomic(kaddr); | |
55 | copy = min(bytes, iov->iov_len - skip); | |
56 | } | |
57 | /* Too bad - revert to non-atomic kmap */ | |
58 | kaddr = kmap(page); | |
59 | from = kaddr + offset; | |
60 | left = __copy_to_user(buf, from, copy); | |
61 | copy -= left; | |
62 | skip += copy; | |
63 | from += copy; | |
64 | bytes -= copy; | |
65 | while (unlikely(!left && bytes)) { | |
66 | iov++; | |
67 | buf = iov->iov_base; | |
68 | copy = min(bytes, iov->iov_len); | |
69 | left = __copy_to_user(buf, from, copy); | |
70 | copy -= left; | |
71 | skip = copy; | |
72 | from += copy; | |
73 | bytes -= copy; | |
74 | } | |
75 | kunmap(page); | |
76 | done: | |
81055e58 AV |
77 | if (skip == iov->iov_len) { |
78 | iov++; | |
79 | skip = 0; | |
80 | } | |
4f18cd31 AV |
81 | i->count -= wanted - bytes; |
82 | i->nr_segs -= iov - i->iov; | |
83 | i->iov = iov; | |
84 | i->iov_offset = skip; | |
85 | return wanted - bytes; | |
86 | } | |
87 | EXPORT_SYMBOL(copy_page_to_iter); | |
88 | ||
f0d1bec9 AV |
89 | size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes, |
90 | struct iov_iter *i) | |
91 | { | |
92 | size_t skip, copy, left, wanted; | |
93 | const struct iovec *iov; | |
94 | char __user *buf; | |
95 | void *kaddr, *to; | |
96 | ||
97 | if (unlikely(bytes > i->count)) | |
98 | bytes = i->count; | |
99 | ||
100 | if (unlikely(!bytes)) | |
101 | return 0; | |
102 | ||
103 | wanted = bytes; | |
104 | iov = i->iov; | |
105 | skip = i->iov_offset; | |
106 | buf = iov->iov_base + skip; | |
107 | copy = min(bytes, iov->iov_len - skip); | |
108 | ||
109 | if (!fault_in_pages_readable(buf, copy)) { | |
110 | kaddr = kmap_atomic(page); | |
111 | to = kaddr + offset; | |
112 | ||
113 | /* first chunk, usually the only one */ | |
114 | left = __copy_from_user_inatomic(to, buf, copy); | |
115 | copy -= left; | |
116 | skip += copy; | |
117 | to += copy; | |
118 | bytes -= copy; | |
119 | ||
120 | while (unlikely(!left && bytes)) { | |
121 | iov++; | |
122 | buf = iov->iov_base; | |
123 | copy = min(bytes, iov->iov_len); | |
124 | left = __copy_from_user_inatomic(to, buf, copy); | |
125 | copy -= left; | |
126 | skip = copy; | |
127 | to += copy; | |
128 | bytes -= copy; | |
129 | } | |
130 | if (likely(!bytes)) { | |
131 | kunmap_atomic(kaddr); | |
132 | goto done; | |
133 | } | |
134 | offset = to - kaddr; | |
135 | buf += copy; | |
136 | kunmap_atomic(kaddr); | |
137 | copy = min(bytes, iov->iov_len - skip); | |
138 | } | |
139 | /* Too bad - revert to non-atomic kmap */ | |
140 | kaddr = kmap(page); | |
141 | to = kaddr + offset; | |
142 | left = __copy_from_user(to, buf, copy); | |
143 | copy -= left; | |
144 | skip += copy; | |
145 | to += copy; | |
146 | bytes -= copy; | |
147 | while (unlikely(!left && bytes)) { | |
148 | iov++; | |
149 | buf = iov->iov_base; | |
150 | copy = min(bytes, iov->iov_len); | |
151 | left = __copy_from_user(to, buf, copy); | |
152 | copy -= left; | |
153 | skip = copy; | |
154 | to += copy; | |
155 | bytes -= copy; | |
156 | } | |
157 | kunmap(page); | |
158 | done: | |
81055e58 AV |
159 | if (skip == iov->iov_len) { |
160 | iov++; | |
161 | skip = 0; | |
162 | } | |
f0d1bec9 AV |
163 | i->count -= wanted - bytes; |
164 | i->nr_segs -= iov - i->iov; | |
165 | i->iov = iov; | |
166 | i->iov_offset = skip; | |
167 | return wanted - bytes; | |
168 | } | |
169 | EXPORT_SYMBOL(copy_page_from_iter); | |
170 | ||
4f18cd31 AV |
171 | static size_t __iovec_copy_from_user_inatomic(char *vaddr, |
172 | const struct iovec *iov, size_t base, size_t bytes) | |
173 | { | |
174 | size_t copied = 0, left = 0; | |
175 | ||
176 | while (bytes) { | |
177 | char __user *buf = iov->iov_base + base; | |
178 | int copy = min(bytes, iov->iov_len - base); | |
179 | ||
180 | base = 0; | |
181 | left = __copy_from_user_inatomic(vaddr, buf, copy); | |
182 | copied += copy; | |
183 | bytes -= copy; | |
184 | vaddr += copy; | |
185 | iov++; | |
186 | ||
187 | if (unlikely(left)) | |
188 | break; | |
189 | } | |
190 | return copied - left; | |
191 | } | |
192 | ||
193 | /* | |
194 | * Copy as much as we can into the page and return the number of bytes which | |
195 | * were successfully copied. If a fault is encountered then return the number of | |
196 | * bytes which were copied. | |
197 | */ | |
198 | size_t iov_iter_copy_from_user_atomic(struct page *page, | |
199 | struct iov_iter *i, unsigned long offset, size_t bytes) | |
200 | { | |
201 | char *kaddr; | |
202 | size_t copied; | |
203 | ||
204 | kaddr = kmap_atomic(page); | |
205 | if (likely(i->nr_segs == 1)) { | |
206 | int left; | |
207 | char __user *buf = i->iov->iov_base + i->iov_offset; | |
208 | left = __copy_from_user_inatomic(kaddr + offset, buf, bytes); | |
209 | copied = bytes - left; | |
210 | } else { | |
211 | copied = __iovec_copy_from_user_inatomic(kaddr + offset, | |
212 | i->iov, i->iov_offset, bytes); | |
213 | } | |
214 | kunmap_atomic(kaddr); | |
215 | ||
216 | return copied; | |
217 | } | |
218 | EXPORT_SYMBOL(iov_iter_copy_from_user_atomic); | |
219 | ||
4f18cd31 AV |
220 | void iov_iter_advance(struct iov_iter *i, size_t bytes) |
221 | { | |
222 | BUG_ON(i->count < bytes); | |
223 | ||
224 | if (likely(i->nr_segs == 1)) { | |
225 | i->iov_offset += bytes; | |
226 | i->count -= bytes; | |
227 | } else { | |
228 | const struct iovec *iov = i->iov; | |
229 | size_t base = i->iov_offset; | |
230 | unsigned long nr_segs = i->nr_segs; | |
231 | ||
232 | /* | |
233 | * The !iov->iov_len check ensures we skip over unlikely | |
234 | * zero-length segments (without overruning the iovec). | |
235 | */ | |
236 | while (bytes || unlikely(i->count && !iov->iov_len)) { | |
237 | int copy; | |
238 | ||
239 | copy = min(bytes, iov->iov_len - base); | |
240 | BUG_ON(!i->count || i->count < copy); | |
241 | i->count -= copy; | |
242 | bytes -= copy; | |
243 | base += copy; | |
244 | if (iov->iov_len == base) { | |
245 | iov++; | |
246 | nr_segs--; | |
247 | base = 0; | |
248 | } | |
249 | } | |
250 | i->iov = iov; | |
251 | i->iov_offset = base; | |
252 | i->nr_segs = nr_segs; | |
253 | } | |
254 | } | |
255 | EXPORT_SYMBOL(iov_iter_advance); | |
256 | ||
257 | /* | |
258 | * Fault in the first iovec of the given iov_iter, to a maximum length | |
259 | * of bytes. Returns 0 on success, or non-zero if the memory could not be | |
260 | * accessed (ie. because it is an invalid address). | |
261 | * | |
262 | * writev-intensive code may want this to prefault several iovecs -- that | |
263 | * would be possible (callers must not rely on the fact that _only_ the | |
264 | * first iovec will be faulted with the current implementation). | |
265 | */ | |
266 | int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes) | |
267 | { | |
268 | char __user *buf = i->iov->iov_base + i->iov_offset; | |
269 | bytes = min(bytes, i->iov->iov_len - i->iov_offset); | |
270 | return fault_in_pages_readable(buf, bytes); | |
271 | } | |
272 | EXPORT_SYMBOL(iov_iter_fault_in_readable); | |
273 | ||
274 | /* | |
275 | * Return the count of just the current iov_iter segment. | |
276 | */ | |
277 | size_t iov_iter_single_seg_count(const struct iov_iter *i) | |
278 | { | |
279 | const struct iovec *iov = i->iov; | |
280 | if (i->nr_segs == 1) | |
281 | return i->count; | |
282 | else | |
283 | return min(i->count, iov->iov_len - i->iov_offset); | |
284 | } | |
285 | EXPORT_SYMBOL(iov_iter_single_seg_count); | |
886a3911 AV |
286 | |
287 | unsigned long iov_iter_alignment(const struct iov_iter *i) | |
288 | { | |
289 | const struct iovec *iov = i->iov; | |
290 | unsigned long res; | |
291 | size_t size = i->count; | |
292 | size_t n; | |
293 | ||
294 | if (!size) | |
295 | return 0; | |
296 | ||
297 | res = (unsigned long)iov->iov_base + i->iov_offset; | |
298 | n = iov->iov_len - i->iov_offset; | |
299 | if (n >= size) | |
300 | return res | size; | |
301 | size -= n; | |
302 | res |= n; | |
303 | while (size > (++iov)->iov_len) { | |
304 | res |= (unsigned long)iov->iov_base | iov->iov_len; | |
305 | size -= iov->iov_len; | |
306 | } | |
307 | res |= (unsigned long)iov->iov_base | size; | |
308 | return res; | |
309 | } | |
310 | EXPORT_SYMBOL(iov_iter_alignment); | |
71d8e532 AV |
311 | |
312 | void iov_iter_init(struct iov_iter *i, int direction, | |
313 | const struct iovec *iov, unsigned long nr_segs, | |
314 | size_t count) | |
315 | { | |
316 | /* It will get better. Eventually... */ | |
317 | if (segment_eq(get_fs(), KERNEL_DS)) | |
318 | direction |= REQ_KERNEL; | |
319 | i->type = direction; | |
320 | i->iov = iov; | |
321 | i->nr_segs = nr_segs; | |
322 | i->iov_offset = 0; | |
323 | i->count = count; | |
324 | } | |
325 | EXPORT_SYMBOL(iov_iter_init); | |
7b2c99d1 AV |
326 | |
327 | ssize_t iov_iter_get_pages(struct iov_iter *i, | |
328 | struct page **pages, size_t maxsize, | |
329 | size_t *start) | |
330 | { | |
331 | size_t offset = i->iov_offset; | |
332 | const struct iovec *iov = i->iov; | |
333 | size_t len; | |
334 | unsigned long addr; | |
335 | int n; | |
336 | int res; | |
337 | ||
338 | len = iov->iov_len - offset; | |
339 | if (len > i->count) | |
340 | len = i->count; | |
341 | if (len > maxsize) | |
342 | len = maxsize; | |
343 | addr = (unsigned long)iov->iov_base + offset; | |
344 | len += *start = addr & (PAGE_SIZE - 1); | |
345 | addr &= ~(PAGE_SIZE - 1); | |
346 | n = (len + PAGE_SIZE - 1) / PAGE_SIZE; | |
347 | res = get_user_pages_fast(addr, n, (i->type & WRITE) != WRITE, pages); | |
348 | if (unlikely(res < 0)) | |
349 | return res; | |
350 | return (res == n ? len : res * PAGE_SIZE) - *start; | |
351 | } | |
352 | EXPORT_SYMBOL(iov_iter_get_pages); | |
f67da30c | 353 | |
91f79c43 AV |
354 | ssize_t iov_iter_get_pages_alloc(struct iov_iter *i, |
355 | struct page ***pages, size_t maxsize, | |
356 | size_t *start) | |
357 | { | |
358 | size_t offset = i->iov_offset; | |
359 | const struct iovec *iov = i->iov; | |
360 | size_t len; | |
361 | unsigned long addr; | |
362 | void *p; | |
363 | int n; | |
364 | int res; | |
365 | ||
366 | len = iov->iov_len - offset; | |
367 | if (len > i->count) | |
368 | len = i->count; | |
369 | if (len > maxsize) | |
370 | len = maxsize; | |
371 | addr = (unsigned long)iov->iov_base + offset; | |
372 | len += *start = addr & (PAGE_SIZE - 1); | |
373 | addr &= ~(PAGE_SIZE - 1); | |
374 | n = (len + PAGE_SIZE - 1) / PAGE_SIZE; | |
375 | ||
376 | p = kmalloc(n * sizeof(struct page *), GFP_KERNEL); | |
377 | if (!p) | |
378 | p = vmalloc(n * sizeof(struct page *)); | |
379 | if (!p) | |
380 | return -ENOMEM; | |
381 | ||
382 | res = get_user_pages_fast(addr, n, (i->type & WRITE) != WRITE, p); | |
383 | if (unlikely(res < 0)) { | |
384 | kvfree(p); | |
385 | return res; | |
386 | } | |
387 | *pages = p; | |
388 | return (res == n ? len : res * PAGE_SIZE) - *start; | |
389 | } | |
390 | EXPORT_SYMBOL(iov_iter_get_pages_alloc); | |
391 | ||
f67da30c AV |
392 | int iov_iter_npages(const struct iov_iter *i, int maxpages) |
393 | { | |
394 | size_t offset = i->iov_offset; | |
395 | size_t size = i->count; | |
396 | const struct iovec *iov = i->iov; | |
397 | int npages = 0; | |
398 | int n; | |
399 | ||
400 | for (n = 0; size && n < i->nr_segs; n++, iov++) { | |
401 | unsigned long addr = (unsigned long)iov->iov_base + offset; | |
402 | size_t len = iov->iov_len - offset; | |
403 | offset = 0; | |
404 | if (unlikely(!len)) /* empty segment */ | |
405 | continue; | |
406 | if (len > size) | |
407 | len = size; | |
408 | npages += (addr + len + PAGE_SIZE - 1) / PAGE_SIZE | |
409 | - addr / PAGE_SIZE; | |
410 | if (npages >= maxpages) /* don't bother going further */ | |
411 | return maxpages; | |
412 | size -= len; | |
413 | offset = 0; | |
414 | } | |
415 | return min(npages, maxpages); | |
416 | } | |
417 | EXPORT_SYMBOL(iov_iter_npages); |