Al Viro | 4f18cd3 | 2014-02-05 19:11:33 -0500 | [diff] [blame] | 1 | #include <linux/export.h> |
| 2 | #include <linux/uio.h> |
| 3 | #include <linux/pagemap.h> |
Al Viro | 91f79c4 | 2014-03-21 04:58:33 -0400 | [diff] [blame] | 4 | #include <linux/slab.h> |
| 5 | #include <linux/vmalloc.h> |
Al Viro | 4f18cd3 | 2014-02-05 19:11:33 -0500 | [diff] [blame] | 6 | |
Matthew Wilcox | c35e024 | 2014-08-01 09:27:22 -0400 | [diff] [blame] | 7 | static size_t copy_to_iter_iovec(void *from, size_t bytes, struct iov_iter *i) |
| 8 | { |
| 9 | size_t skip, copy, left, wanted; |
| 10 | const struct iovec *iov; |
| 11 | char __user *buf; |
| 12 | |
| 13 | if (unlikely(bytes > i->count)) |
| 14 | bytes = i->count; |
| 15 | |
| 16 | if (unlikely(!bytes)) |
| 17 | return 0; |
| 18 | |
| 19 | wanted = bytes; |
| 20 | iov = i->iov; |
| 21 | skip = i->iov_offset; |
| 22 | buf = iov->iov_base + skip; |
| 23 | copy = min(bytes, iov->iov_len - skip); |
| 24 | |
| 25 | left = __copy_to_user(buf, from, copy); |
| 26 | copy -= left; |
| 27 | skip += copy; |
| 28 | from += copy; |
| 29 | bytes -= copy; |
| 30 | while (unlikely(!left && bytes)) { |
| 31 | iov++; |
| 32 | buf = iov->iov_base; |
| 33 | copy = min(bytes, iov->iov_len); |
| 34 | left = __copy_to_user(buf, from, copy); |
| 35 | copy -= left; |
| 36 | skip = copy; |
| 37 | from += copy; |
| 38 | bytes -= copy; |
| 39 | } |
| 40 | |
| 41 | if (skip == iov->iov_len) { |
| 42 | iov++; |
| 43 | skip = 0; |
| 44 | } |
| 45 | i->count -= wanted - bytes; |
| 46 | i->nr_segs -= iov - i->iov; |
| 47 | i->iov = iov; |
| 48 | i->iov_offset = skip; |
| 49 | return wanted - bytes; |
| 50 | } |
| 51 | |
| 52 | static size_t copy_from_iter_iovec(void *to, size_t bytes, struct iov_iter *i) |
| 53 | { |
| 54 | size_t skip, copy, left, wanted; |
| 55 | const struct iovec *iov; |
| 56 | char __user *buf; |
| 57 | |
| 58 | if (unlikely(bytes > i->count)) |
| 59 | bytes = i->count; |
| 60 | |
| 61 | if (unlikely(!bytes)) |
| 62 | return 0; |
| 63 | |
| 64 | wanted = bytes; |
| 65 | iov = i->iov; |
| 66 | skip = i->iov_offset; |
| 67 | buf = iov->iov_base + skip; |
| 68 | copy = min(bytes, iov->iov_len - skip); |
| 69 | |
| 70 | left = __copy_from_user(to, buf, copy); |
| 71 | copy -= left; |
| 72 | skip += copy; |
| 73 | to += copy; |
| 74 | bytes -= copy; |
| 75 | while (unlikely(!left && bytes)) { |
| 76 | iov++; |
| 77 | buf = iov->iov_base; |
| 78 | copy = min(bytes, iov->iov_len); |
| 79 | left = __copy_from_user(to, buf, copy); |
| 80 | copy -= left; |
| 81 | skip = copy; |
| 82 | to += copy; |
| 83 | bytes -= copy; |
| 84 | } |
| 85 | |
| 86 | if (skip == iov->iov_len) { |
| 87 | iov++; |
| 88 | skip = 0; |
| 89 | } |
| 90 | i->count -= wanted - bytes; |
| 91 | i->nr_segs -= iov - i->iov; |
| 92 | i->iov = iov; |
| 93 | i->iov_offset = skip; |
| 94 | return wanted - bytes; |
| 95 | } |
| 96 | |
Al Viro | 62a8067 | 2014-04-04 23:12:29 -0400 | [diff] [blame] | 97 | static size_t copy_page_to_iter_iovec(struct page *page, size_t offset, size_t bytes, |
Al Viro | 4f18cd3 | 2014-02-05 19:11:33 -0500 | [diff] [blame] | 98 | struct iov_iter *i) |
| 99 | { |
| 100 | size_t skip, copy, left, wanted; |
| 101 | const struct iovec *iov; |
| 102 | char __user *buf; |
| 103 | void *kaddr, *from; |
| 104 | |
| 105 | if (unlikely(bytes > i->count)) |
| 106 | bytes = i->count; |
| 107 | |
| 108 | if (unlikely(!bytes)) |
| 109 | return 0; |
| 110 | |
| 111 | wanted = bytes; |
| 112 | iov = i->iov; |
| 113 | skip = i->iov_offset; |
| 114 | buf = iov->iov_base + skip; |
| 115 | copy = min(bytes, iov->iov_len - skip); |
| 116 | |
| 117 | if (!fault_in_pages_writeable(buf, copy)) { |
| 118 | kaddr = kmap_atomic(page); |
| 119 | from = kaddr + offset; |
| 120 | |
| 121 | /* first chunk, usually the only one */ |
| 122 | left = __copy_to_user_inatomic(buf, from, copy); |
| 123 | copy -= left; |
| 124 | skip += copy; |
| 125 | from += copy; |
| 126 | bytes -= copy; |
| 127 | |
| 128 | while (unlikely(!left && bytes)) { |
| 129 | iov++; |
| 130 | buf = iov->iov_base; |
| 131 | copy = min(bytes, iov->iov_len); |
| 132 | left = __copy_to_user_inatomic(buf, from, copy); |
| 133 | copy -= left; |
| 134 | skip = copy; |
| 135 | from += copy; |
| 136 | bytes -= copy; |
| 137 | } |
| 138 | if (likely(!bytes)) { |
| 139 | kunmap_atomic(kaddr); |
| 140 | goto done; |
| 141 | } |
| 142 | offset = from - kaddr; |
| 143 | buf += copy; |
| 144 | kunmap_atomic(kaddr); |
| 145 | copy = min(bytes, iov->iov_len - skip); |
| 146 | } |
| 147 | /* Too bad - revert to non-atomic kmap */ |
| 148 | kaddr = kmap(page); |
| 149 | from = kaddr + offset; |
| 150 | left = __copy_to_user(buf, from, copy); |
| 151 | copy -= left; |
| 152 | skip += copy; |
| 153 | from += copy; |
| 154 | bytes -= copy; |
| 155 | while (unlikely(!left && bytes)) { |
| 156 | iov++; |
| 157 | buf = iov->iov_base; |
| 158 | copy = min(bytes, iov->iov_len); |
| 159 | left = __copy_to_user(buf, from, copy); |
| 160 | copy -= left; |
| 161 | skip = copy; |
| 162 | from += copy; |
| 163 | bytes -= copy; |
| 164 | } |
| 165 | kunmap(page); |
| 166 | done: |
Al Viro | 81055e5 | 2014-04-04 19:23:46 -0400 | [diff] [blame] | 167 | if (skip == iov->iov_len) { |
| 168 | iov++; |
| 169 | skip = 0; |
| 170 | } |
Al Viro | 4f18cd3 | 2014-02-05 19:11:33 -0500 | [diff] [blame] | 171 | i->count -= wanted - bytes; |
| 172 | i->nr_segs -= iov - i->iov; |
| 173 | i->iov = iov; |
| 174 | i->iov_offset = skip; |
| 175 | return wanted - bytes; |
| 176 | } |
Al Viro | 4f18cd3 | 2014-02-05 19:11:33 -0500 | [diff] [blame] | 177 | |
Al Viro | 62a8067 | 2014-04-04 23:12:29 -0400 | [diff] [blame] | 178 | static size_t copy_page_from_iter_iovec(struct page *page, size_t offset, size_t bytes, |
Al Viro | f0d1bec | 2014-04-03 15:05:18 -0400 | [diff] [blame] | 179 | struct iov_iter *i) |
| 180 | { |
| 181 | size_t skip, copy, left, wanted; |
| 182 | const struct iovec *iov; |
| 183 | char __user *buf; |
| 184 | void *kaddr, *to; |
| 185 | |
| 186 | if (unlikely(bytes > i->count)) |
| 187 | bytes = i->count; |
| 188 | |
| 189 | if (unlikely(!bytes)) |
| 190 | return 0; |
| 191 | |
| 192 | wanted = bytes; |
| 193 | iov = i->iov; |
| 194 | skip = i->iov_offset; |
| 195 | buf = iov->iov_base + skip; |
| 196 | copy = min(bytes, iov->iov_len - skip); |
| 197 | |
| 198 | if (!fault_in_pages_readable(buf, copy)) { |
| 199 | kaddr = kmap_atomic(page); |
| 200 | to = kaddr + offset; |
| 201 | |
| 202 | /* first chunk, usually the only one */ |
| 203 | left = __copy_from_user_inatomic(to, buf, copy); |
| 204 | copy -= left; |
| 205 | skip += copy; |
| 206 | to += copy; |
| 207 | bytes -= copy; |
| 208 | |
| 209 | while (unlikely(!left && bytes)) { |
| 210 | iov++; |
| 211 | buf = iov->iov_base; |
| 212 | copy = min(bytes, iov->iov_len); |
| 213 | left = __copy_from_user_inatomic(to, buf, copy); |
| 214 | copy -= left; |
| 215 | skip = copy; |
| 216 | to += copy; |
| 217 | bytes -= copy; |
| 218 | } |
| 219 | if (likely(!bytes)) { |
| 220 | kunmap_atomic(kaddr); |
| 221 | goto done; |
| 222 | } |
| 223 | offset = to - kaddr; |
| 224 | buf += copy; |
| 225 | kunmap_atomic(kaddr); |
| 226 | copy = min(bytes, iov->iov_len - skip); |
| 227 | } |
| 228 | /* Too bad - revert to non-atomic kmap */ |
| 229 | kaddr = kmap(page); |
| 230 | to = kaddr + offset; |
| 231 | left = __copy_from_user(to, buf, copy); |
| 232 | copy -= left; |
| 233 | skip += copy; |
| 234 | to += copy; |
| 235 | bytes -= copy; |
| 236 | while (unlikely(!left && bytes)) { |
| 237 | iov++; |
| 238 | buf = iov->iov_base; |
| 239 | copy = min(bytes, iov->iov_len); |
| 240 | left = __copy_from_user(to, buf, copy); |
| 241 | copy -= left; |
| 242 | skip = copy; |
| 243 | to += copy; |
| 244 | bytes -= copy; |
| 245 | } |
| 246 | kunmap(page); |
| 247 | done: |
Al Viro | 81055e5 | 2014-04-04 19:23:46 -0400 | [diff] [blame] | 248 | if (skip == iov->iov_len) { |
| 249 | iov++; |
| 250 | skip = 0; |
| 251 | } |
Al Viro | f0d1bec | 2014-04-03 15:05:18 -0400 | [diff] [blame] | 252 | i->count -= wanted - bytes; |
| 253 | i->nr_segs -= iov - i->iov; |
| 254 | i->iov = iov; |
| 255 | i->iov_offset = skip; |
| 256 | return wanted - bytes; |
| 257 | } |
Al Viro | f0d1bec | 2014-04-03 15:05:18 -0400 | [diff] [blame] | 258 | |
Matthew Wilcox | c35e024 | 2014-08-01 09:27:22 -0400 | [diff] [blame] | 259 | static size_t zero_iovec(size_t bytes, struct iov_iter *i) |
| 260 | { |
| 261 | size_t skip, copy, left, wanted; |
| 262 | const struct iovec *iov; |
| 263 | char __user *buf; |
| 264 | |
| 265 | if (unlikely(bytes > i->count)) |
| 266 | bytes = i->count; |
| 267 | |
| 268 | if (unlikely(!bytes)) |
| 269 | return 0; |
| 270 | |
| 271 | wanted = bytes; |
| 272 | iov = i->iov; |
| 273 | skip = i->iov_offset; |
| 274 | buf = iov->iov_base + skip; |
| 275 | copy = min(bytes, iov->iov_len - skip); |
| 276 | |
| 277 | left = __clear_user(buf, copy); |
| 278 | copy -= left; |
| 279 | skip += copy; |
| 280 | bytes -= copy; |
| 281 | |
| 282 | while (unlikely(!left && bytes)) { |
| 283 | iov++; |
| 284 | buf = iov->iov_base; |
| 285 | copy = min(bytes, iov->iov_len); |
| 286 | left = __clear_user(buf, copy); |
| 287 | copy -= left; |
| 288 | skip = copy; |
| 289 | bytes -= copy; |
| 290 | } |
| 291 | |
| 292 | if (skip == iov->iov_len) { |
| 293 | iov++; |
| 294 | skip = 0; |
| 295 | } |
| 296 | i->count -= wanted - bytes; |
| 297 | i->nr_segs -= iov - i->iov; |
| 298 | i->iov = iov; |
| 299 | i->iov_offset = skip; |
| 300 | return wanted - bytes; |
| 301 | } |
| 302 | |
Al Viro | 4f18cd3 | 2014-02-05 19:11:33 -0500 | [diff] [blame] | 303 | static size_t __iovec_copy_from_user_inatomic(char *vaddr, |
| 304 | const struct iovec *iov, size_t base, size_t bytes) |
| 305 | { |
| 306 | size_t copied = 0, left = 0; |
| 307 | |
| 308 | while (bytes) { |
| 309 | char __user *buf = iov->iov_base + base; |
| 310 | int copy = min(bytes, iov->iov_len - base); |
| 311 | |
| 312 | base = 0; |
| 313 | left = __copy_from_user_inatomic(vaddr, buf, copy); |
| 314 | copied += copy; |
| 315 | bytes -= copy; |
| 316 | vaddr += copy; |
| 317 | iov++; |
| 318 | |
| 319 | if (unlikely(left)) |
| 320 | break; |
| 321 | } |
| 322 | return copied - left; |
| 323 | } |
| 324 | |
| 325 | /* |
| 326 | * Copy as much as we can into the page and return the number of bytes which |
| 327 | * were successfully copied. If a fault is encountered then return the number of |
| 328 | * bytes which were copied. |
| 329 | */ |
Al Viro | 62a8067 | 2014-04-04 23:12:29 -0400 | [diff] [blame] | 330 | static size_t copy_from_user_atomic_iovec(struct page *page, |
Al Viro | 4f18cd3 | 2014-02-05 19:11:33 -0500 | [diff] [blame] | 331 | struct iov_iter *i, unsigned long offset, size_t bytes) |
| 332 | { |
| 333 | char *kaddr; |
| 334 | size_t copied; |
| 335 | |
| 336 | kaddr = kmap_atomic(page); |
| 337 | if (likely(i->nr_segs == 1)) { |
| 338 | int left; |
| 339 | char __user *buf = i->iov->iov_base + i->iov_offset; |
| 340 | left = __copy_from_user_inatomic(kaddr + offset, buf, bytes); |
| 341 | copied = bytes - left; |
| 342 | } else { |
| 343 | copied = __iovec_copy_from_user_inatomic(kaddr + offset, |
| 344 | i->iov, i->iov_offset, bytes); |
| 345 | } |
| 346 | kunmap_atomic(kaddr); |
| 347 | |
| 348 | return copied; |
| 349 | } |
Al Viro | 4f18cd3 | 2014-02-05 19:11:33 -0500 | [diff] [blame] | 350 | |
Al Viro | 62a8067 | 2014-04-04 23:12:29 -0400 | [diff] [blame] | 351 | static void advance_iovec(struct iov_iter *i, size_t bytes) |
Al Viro | 4f18cd3 | 2014-02-05 19:11:33 -0500 | [diff] [blame] | 352 | { |
| 353 | BUG_ON(i->count < bytes); |
| 354 | |
| 355 | if (likely(i->nr_segs == 1)) { |
| 356 | i->iov_offset += bytes; |
| 357 | i->count -= bytes; |
| 358 | } else { |
| 359 | const struct iovec *iov = i->iov; |
| 360 | size_t base = i->iov_offset; |
| 361 | unsigned long nr_segs = i->nr_segs; |
| 362 | |
| 363 | /* |
| 364 | * The !iov->iov_len check ensures we skip over unlikely |
| 365 | * zero-length segments (without overruning the iovec). |
| 366 | */ |
| 367 | while (bytes || unlikely(i->count && !iov->iov_len)) { |
| 368 | int copy; |
| 369 | |
| 370 | copy = min(bytes, iov->iov_len - base); |
| 371 | BUG_ON(!i->count || i->count < copy); |
| 372 | i->count -= copy; |
| 373 | bytes -= copy; |
| 374 | base += copy; |
| 375 | if (iov->iov_len == base) { |
| 376 | iov++; |
| 377 | nr_segs--; |
| 378 | base = 0; |
| 379 | } |
| 380 | } |
| 381 | i->iov = iov; |
| 382 | i->iov_offset = base; |
| 383 | i->nr_segs = nr_segs; |
| 384 | } |
| 385 | } |
Al Viro | 4f18cd3 | 2014-02-05 19:11:33 -0500 | [diff] [blame] | 386 | |
| 387 | /* |
| 388 | * Fault in the first iovec of the given iov_iter, to a maximum length |
| 389 | * of bytes. Returns 0 on success, or non-zero if the memory could not be |
| 390 | * accessed (ie. because it is an invalid address). |
| 391 | * |
| 392 | * writev-intensive code may want this to prefault several iovecs -- that |
| 393 | * would be possible (callers must not rely on the fact that _only_ the |
| 394 | * first iovec will be faulted with the current implementation). |
| 395 | */ |
| 396 | int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes) |
| 397 | { |
Al Viro | 62a8067 | 2014-04-04 23:12:29 -0400 | [diff] [blame] | 398 | if (!(i->type & ITER_BVEC)) { |
| 399 | char __user *buf = i->iov->iov_base + i->iov_offset; |
| 400 | bytes = min(bytes, i->iov->iov_len - i->iov_offset); |
| 401 | return fault_in_pages_readable(buf, bytes); |
| 402 | } |
| 403 | return 0; |
Al Viro | 4f18cd3 | 2014-02-05 19:11:33 -0500 | [diff] [blame] | 404 | } |
| 405 | EXPORT_SYMBOL(iov_iter_fault_in_readable); |
| 406 | |
Al Viro | 62a8067 | 2014-04-04 23:12:29 -0400 | [diff] [blame] | 407 | static unsigned long alignment_iovec(const struct iov_iter *i) |
Al Viro | 886a391 | 2014-03-05 13:50:45 -0500 | [diff] [blame] | 408 | { |
| 409 | const struct iovec *iov = i->iov; |
| 410 | unsigned long res; |
| 411 | size_t size = i->count; |
| 412 | size_t n; |
| 413 | |
| 414 | if (!size) |
| 415 | return 0; |
| 416 | |
| 417 | res = (unsigned long)iov->iov_base + i->iov_offset; |
| 418 | n = iov->iov_len - i->iov_offset; |
| 419 | if (n >= size) |
| 420 | return res | size; |
| 421 | size -= n; |
| 422 | res |= n; |
| 423 | while (size > (++iov)->iov_len) { |
| 424 | res |= (unsigned long)iov->iov_base | iov->iov_len; |
| 425 | size -= iov->iov_len; |
| 426 | } |
| 427 | res |= (unsigned long)iov->iov_base | size; |
| 428 | return res; |
| 429 | } |
Al Viro | 71d8e53 | 2014-03-05 19:28:09 -0500 | [diff] [blame] | 430 | |
| 431 | void iov_iter_init(struct iov_iter *i, int direction, |
| 432 | const struct iovec *iov, unsigned long nr_segs, |
| 433 | size_t count) |
| 434 | { |
| 435 | /* It will get better. Eventually... */ |
| 436 | if (segment_eq(get_fs(), KERNEL_DS)) |
Al Viro | 62a8067 | 2014-04-04 23:12:29 -0400 | [diff] [blame] | 437 | direction |= ITER_KVEC; |
Al Viro | 71d8e53 | 2014-03-05 19:28:09 -0500 | [diff] [blame] | 438 | i->type = direction; |
| 439 | i->iov = iov; |
| 440 | i->nr_segs = nr_segs; |
| 441 | i->iov_offset = 0; |
| 442 | i->count = count; |
| 443 | } |
| 444 | EXPORT_SYMBOL(iov_iter_init); |
Al Viro | 7b2c99d | 2014-03-15 04:05:57 -0400 | [diff] [blame] | 445 | |
Al Viro | 62a8067 | 2014-04-04 23:12:29 -0400 | [diff] [blame] | 446 | static ssize_t get_pages_iovec(struct iov_iter *i, |
Miklos Szeredi | 2c80929 | 2014-09-24 17:09:11 +0200 | [diff] [blame] | 447 | struct page **pages, size_t maxsize, unsigned maxpages, |
Al Viro | 7b2c99d | 2014-03-15 04:05:57 -0400 | [diff] [blame] | 448 | size_t *start) |
| 449 | { |
| 450 | size_t offset = i->iov_offset; |
| 451 | const struct iovec *iov = i->iov; |
| 452 | size_t len; |
| 453 | unsigned long addr; |
| 454 | int n; |
| 455 | int res; |
| 456 | |
| 457 | len = iov->iov_len - offset; |
| 458 | if (len > i->count) |
| 459 | len = i->count; |
Miklos Szeredi | 2c80929 | 2014-09-24 17:09:11 +0200 | [diff] [blame] | 460 | if (len > maxsize) |
| 461 | len = maxsize; |
Al Viro | 7b2c99d | 2014-03-15 04:05:57 -0400 | [diff] [blame] | 462 | addr = (unsigned long)iov->iov_base + offset; |
| 463 | len += *start = addr & (PAGE_SIZE - 1); |
Al Viro | c7f3888 | 2014-06-18 20:34:33 -0400 | [diff] [blame] | 464 | if (len > maxpages * PAGE_SIZE) |
| 465 | len = maxpages * PAGE_SIZE; |
Al Viro | 7b2c99d | 2014-03-15 04:05:57 -0400 | [diff] [blame] | 466 | addr &= ~(PAGE_SIZE - 1); |
| 467 | n = (len + PAGE_SIZE - 1) / PAGE_SIZE; |
| 468 | res = get_user_pages_fast(addr, n, (i->type & WRITE) != WRITE, pages); |
| 469 | if (unlikely(res < 0)) |
| 470 | return res; |
| 471 | return (res == n ? len : res * PAGE_SIZE) - *start; |
| 472 | } |
Al Viro | f67da30 | 2014-03-19 01:16:16 -0400 | [diff] [blame] | 473 | |
Al Viro | 62a8067 | 2014-04-04 23:12:29 -0400 | [diff] [blame] | 474 | static ssize_t get_pages_alloc_iovec(struct iov_iter *i, |
Al Viro | 91f79c4 | 2014-03-21 04:58:33 -0400 | [diff] [blame] | 475 | struct page ***pages, size_t maxsize, |
| 476 | size_t *start) |
| 477 | { |
| 478 | size_t offset = i->iov_offset; |
| 479 | const struct iovec *iov = i->iov; |
| 480 | size_t len; |
| 481 | unsigned long addr; |
| 482 | void *p; |
| 483 | int n; |
| 484 | int res; |
| 485 | |
| 486 | len = iov->iov_len - offset; |
| 487 | if (len > i->count) |
| 488 | len = i->count; |
| 489 | if (len > maxsize) |
| 490 | len = maxsize; |
| 491 | addr = (unsigned long)iov->iov_base + offset; |
| 492 | len += *start = addr & (PAGE_SIZE - 1); |
| 493 | addr &= ~(PAGE_SIZE - 1); |
| 494 | n = (len + PAGE_SIZE - 1) / PAGE_SIZE; |
| 495 | |
| 496 | p = kmalloc(n * sizeof(struct page *), GFP_KERNEL); |
| 497 | if (!p) |
| 498 | p = vmalloc(n * sizeof(struct page *)); |
| 499 | if (!p) |
| 500 | return -ENOMEM; |
| 501 | |
| 502 | res = get_user_pages_fast(addr, n, (i->type & WRITE) != WRITE, p); |
| 503 | if (unlikely(res < 0)) { |
| 504 | kvfree(p); |
| 505 | return res; |
| 506 | } |
| 507 | *pages = p; |
| 508 | return (res == n ? len : res * PAGE_SIZE) - *start; |
| 509 | } |
Al Viro | 91f79c4 | 2014-03-21 04:58:33 -0400 | [diff] [blame] | 510 | |
Al Viro | 62a8067 | 2014-04-04 23:12:29 -0400 | [diff] [blame] | 511 | static int iov_iter_npages_iovec(const struct iov_iter *i, int maxpages) |
Al Viro | f67da30 | 2014-03-19 01:16:16 -0400 | [diff] [blame] | 512 | { |
| 513 | size_t offset = i->iov_offset; |
| 514 | size_t size = i->count; |
| 515 | const struct iovec *iov = i->iov; |
| 516 | int npages = 0; |
| 517 | int n; |
| 518 | |
| 519 | for (n = 0; size && n < i->nr_segs; n++, iov++) { |
| 520 | unsigned long addr = (unsigned long)iov->iov_base + offset; |
| 521 | size_t len = iov->iov_len - offset; |
| 522 | offset = 0; |
| 523 | if (unlikely(!len)) /* empty segment */ |
| 524 | continue; |
| 525 | if (len > size) |
| 526 | len = size; |
| 527 | npages += (addr + len + PAGE_SIZE - 1) / PAGE_SIZE |
| 528 | - addr / PAGE_SIZE; |
| 529 | if (npages >= maxpages) /* don't bother going further */ |
| 530 | return maxpages; |
| 531 | size -= len; |
| 532 | offset = 0; |
| 533 | } |
| 534 | return min(npages, maxpages); |
| 535 | } |
Al Viro | 62a8067 | 2014-04-04 23:12:29 -0400 | [diff] [blame] | 536 | |
| 537 | static void memcpy_from_page(char *to, struct page *page, size_t offset, size_t len) |
| 538 | { |
| 539 | char *from = kmap_atomic(page); |
| 540 | memcpy(to, from + offset, len); |
| 541 | kunmap_atomic(from); |
| 542 | } |
| 543 | |
| 544 | static void memcpy_to_page(struct page *page, size_t offset, char *from, size_t len) |
| 545 | { |
| 546 | char *to = kmap_atomic(page); |
| 547 | memcpy(to + offset, from, len); |
| 548 | kunmap_atomic(to); |
| 549 | } |
| 550 | |
Matthew Wilcox | c35e024 | 2014-08-01 09:27:22 -0400 | [diff] [blame] | 551 | static void memzero_page(struct page *page, size_t offset, size_t len) |
| 552 | { |
| 553 | char *addr = kmap_atomic(page); |
| 554 | memset(addr + offset, 0, len); |
| 555 | kunmap_atomic(addr); |
| 556 | } |
| 557 | |
| 558 | static size_t copy_to_iter_bvec(void *from, size_t bytes, struct iov_iter *i) |
Al Viro | 62a8067 | 2014-04-04 23:12:29 -0400 | [diff] [blame] | 559 | { |
| 560 | size_t skip, copy, wanted; |
| 561 | const struct bio_vec *bvec; |
Al Viro | 62a8067 | 2014-04-04 23:12:29 -0400 | [diff] [blame] | 562 | |
| 563 | if (unlikely(bytes > i->count)) |
| 564 | bytes = i->count; |
| 565 | |
| 566 | if (unlikely(!bytes)) |
| 567 | return 0; |
| 568 | |
| 569 | wanted = bytes; |
| 570 | bvec = i->bvec; |
| 571 | skip = i->iov_offset; |
| 572 | copy = min_t(size_t, bytes, bvec->bv_len - skip); |
| 573 | |
Al Viro | 62a8067 | 2014-04-04 23:12:29 -0400 | [diff] [blame] | 574 | memcpy_to_page(bvec->bv_page, skip + bvec->bv_offset, from, copy); |
| 575 | skip += copy; |
| 576 | from += copy; |
| 577 | bytes -= copy; |
| 578 | while (bytes) { |
| 579 | bvec++; |
| 580 | copy = min(bytes, (size_t)bvec->bv_len); |
| 581 | memcpy_to_page(bvec->bv_page, bvec->bv_offset, from, copy); |
| 582 | skip = copy; |
| 583 | from += copy; |
| 584 | bytes -= copy; |
| 585 | } |
Al Viro | 62a8067 | 2014-04-04 23:12:29 -0400 | [diff] [blame] | 586 | if (skip == bvec->bv_len) { |
| 587 | bvec++; |
| 588 | skip = 0; |
| 589 | } |
| 590 | i->count -= wanted - bytes; |
| 591 | i->nr_segs -= bvec - i->bvec; |
| 592 | i->bvec = bvec; |
| 593 | i->iov_offset = skip; |
| 594 | return wanted - bytes; |
| 595 | } |
| 596 | |
Matthew Wilcox | c35e024 | 2014-08-01 09:27:22 -0400 | [diff] [blame] | 597 | static size_t copy_from_iter_bvec(void *to, size_t bytes, struct iov_iter *i) |
Al Viro | 62a8067 | 2014-04-04 23:12:29 -0400 | [diff] [blame] | 598 | { |
| 599 | size_t skip, copy, wanted; |
| 600 | const struct bio_vec *bvec; |
Al Viro | 62a8067 | 2014-04-04 23:12:29 -0400 | [diff] [blame] | 601 | |
| 602 | if (unlikely(bytes > i->count)) |
| 603 | bytes = i->count; |
| 604 | |
| 605 | if (unlikely(!bytes)) |
| 606 | return 0; |
| 607 | |
| 608 | wanted = bytes; |
| 609 | bvec = i->bvec; |
| 610 | skip = i->iov_offset; |
| 611 | |
Al Viro | 62a8067 | 2014-04-04 23:12:29 -0400 | [diff] [blame] | 612 | copy = min(bytes, bvec->bv_len - skip); |
| 613 | |
| 614 | memcpy_from_page(to, bvec->bv_page, bvec->bv_offset + skip, copy); |
| 615 | |
| 616 | to += copy; |
| 617 | skip += copy; |
| 618 | bytes -= copy; |
| 619 | |
| 620 | while (bytes) { |
| 621 | bvec++; |
| 622 | copy = min(bytes, (size_t)bvec->bv_len); |
| 623 | memcpy_from_page(to, bvec->bv_page, bvec->bv_offset, copy); |
| 624 | skip = copy; |
| 625 | to += copy; |
| 626 | bytes -= copy; |
| 627 | } |
Al Viro | 62a8067 | 2014-04-04 23:12:29 -0400 | [diff] [blame] | 628 | if (skip == bvec->bv_len) { |
| 629 | bvec++; |
| 630 | skip = 0; |
| 631 | } |
| 632 | i->count -= wanted; |
| 633 | i->nr_segs -= bvec - i->bvec; |
| 634 | i->bvec = bvec; |
| 635 | i->iov_offset = skip; |
| 636 | return wanted; |
| 637 | } |
| 638 | |
Matthew Wilcox | c35e024 | 2014-08-01 09:27:22 -0400 | [diff] [blame] | 639 | static size_t copy_page_to_iter_bvec(struct page *page, size_t offset, |
| 640 | size_t bytes, struct iov_iter *i) |
| 641 | { |
| 642 | void *kaddr = kmap_atomic(page); |
| 643 | size_t wanted = copy_to_iter_bvec(kaddr + offset, bytes, i); |
| 644 | kunmap_atomic(kaddr); |
| 645 | return wanted; |
| 646 | } |
| 647 | |
| 648 | static size_t copy_page_from_iter_bvec(struct page *page, size_t offset, |
| 649 | size_t bytes, struct iov_iter *i) |
| 650 | { |
| 651 | void *kaddr = kmap_atomic(page); |
| 652 | size_t wanted = copy_from_iter_bvec(kaddr + offset, bytes, i); |
| 653 | kunmap_atomic(kaddr); |
| 654 | return wanted; |
| 655 | } |
| 656 | |
| 657 | static size_t zero_bvec(size_t bytes, struct iov_iter *i) |
| 658 | { |
| 659 | size_t skip, copy, wanted; |
| 660 | const struct bio_vec *bvec; |
| 661 | |
| 662 | if (unlikely(bytes > i->count)) |
| 663 | bytes = i->count; |
| 664 | |
| 665 | if (unlikely(!bytes)) |
| 666 | return 0; |
| 667 | |
| 668 | wanted = bytes; |
| 669 | bvec = i->bvec; |
| 670 | skip = i->iov_offset; |
| 671 | copy = min_t(size_t, bytes, bvec->bv_len - skip); |
| 672 | |
| 673 | memzero_page(bvec->bv_page, skip + bvec->bv_offset, copy); |
| 674 | skip += copy; |
| 675 | bytes -= copy; |
| 676 | while (bytes) { |
| 677 | bvec++; |
| 678 | copy = min(bytes, (size_t)bvec->bv_len); |
| 679 | memzero_page(bvec->bv_page, bvec->bv_offset, copy); |
| 680 | skip = copy; |
| 681 | bytes -= copy; |
| 682 | } |
| 683 | if (skip == bvec->bv_len) { |
| 684 | bvec++; |
| 685 | skip = 0; |
| 686 | } |
| 687 | i->count -= wanted - bytes; |
| 688 | i->nr_segs -= bvec - i->bvec; |
| 689 | i->bvec = bvec; |
| 690 | i->iov_offset = skip; |
| 691 | return wanted - bytes; |
| 692 | } |
| 693 | |
Al Viro | 62a8067 | 2014-04-04 23:12:29 -0400 | [diff] [blame] | 694 | static size_t copy_from_user_bvec(struct page *page, |
| 695 | struct iov_iter *i, unsigned long offset, size_t bytes) |
| 696 | { |
| 697 | char *kaddr; |
| 698 | size_t left; |
| 699 | const struct bio_vec *bvec; |
| 700 | size_t base = i->iov_offset; |
| 701 | |
| 702 | kaddr = kmap_atomic(page); |
| 703 | for (left = bytes, bvec = i->bvec; left; bvec++, base = 0) { |
| 704 | size_t copy = min(left, bvec->bv_len - base); |
| 705 | if (!bvec->bv_len) |
| 706 | continue; |
| 707 | memcpy_from_page(kaddr + offset, bvec->bv_page, |
| 708 | bvec->bv_offset + base, copy); |
| 709 | offset += copy; |
| 710 | left -= copy; |
| 711 | } |
| 712 | kunmap_atomic(kaddr); |
| 713 | return bytes; |
| 714 | } |
| 715 | |
| 716 | static void advance_bvec(struct iov_iter *i, size_t bytes) |
| 717 | { |
| 718 | BUG_ON(i->count < bytes); |
| 719 | |
| 720 | if (likely(i->nr_segs == 1)) { |
| 721 | i->iov_offset += bytes; |
| 722 | i->count -= bytes; |
| 723 | } else { |
| 724 | const struct bio_vec *bvec = i->bvec; |
| 725 | size_t base = i->iov_offset; |
| 726 | unsigned long nr_segs = i->nr_segs; |
| 727 | |
| 728 | /* |
| 729 | * The !iov->iov_len check ensures we skip over unlikely |
| 730 | * zero-length segments (without overruning the iovec). |
| 731 | */ |
| 732 | while (bytes || unlikely(i->count && !bvec->bv_len)) { |
| 733 | int copy; |
| 734 | |
| 735 | copy = min(bytes, bvec->bv_len - base); |
| 736 | BUG_ON(!i->count || i->count < copy); |
| 737 | i->count -= copy; |
| 738 | bytes -= copy; |
| 739 | base += copy; |
| 740 | if (bvec->bv_len == base) { |
| 741 | bvec++; |
| 742 | nr_segs--; |
| 743 | base = 0; |
| 744 | } |
| 745 | } |
| 746 | i->bvec = bvec; |
| 747 | i->iov_offset = base; |
| 748 | i->nr_segs = nr_segs; |
| 749 | } |
| 750 | } |
| 751 | |
| 752 | static unsigned long alignment_bvec(const struct iov_iter *i) |
| 753 | { |
| 754 | const struct bio_vec *bvec = i->bvec; |
| 755 | unsigned long res; |
| 756 | size_t size = i->count; |
| 757 | size_t n; |
| 758 | |
| 759 | if (!size) |
| 760 | return 0; |
| 761 | |
| 762 | res = bvec->bv_offset + i->iov_offset; |
| 763 | n = bvec->bv_len - i->iov_offset; |
| 764 | if (n >= size) |
| 765 | return res | size; |
| 766 | size -= n; |
| 767 | res |= n; |
| 768 | while (size > (++bvec)->bv_len) { |
| 769 | res |= bvec->bv_offset | bvec->bv_len; |
| 770 | size -= bvec->bv_len; |
| 771 | } |
| 772 | res |= bvec->bv_offset | size; |
| 773 | return res; |
| 774 | } |
| 775 | |
| 776 | static ssize_t get_pages_bvec(struct iov_iter *i, |
Miklos Szeredi | 2c80929 | 2014-09-24 17:09:11 +0200 | [diff] [blame] | 777 | struct page **pages, size_t maxsize, unsigned maxpages, |
Al Viro | 62a8067 | 2014-04-04 23:12:29 -0400 | [diff] [blame] | 778 | size_t *start) |
| 779 | { |
| 780 | const struct bio_vec *bvec = i->bvec; |
| 781 | size_t len = bvec->bv_len - i->iov_offset; |
| 782 | if (len > i->count) |
| 783 | len = i->count; |
Miklos Szeredi | 2c80929 | 2014-09-24 17:09:11 +0200 | [diff] [blame] | 784 | if (len > maxsize) |
| 785 | len = maxsize; |
Al Viro | c7f3888 | 2014-06-18 20:34:33 -0400 | [diff] [blame] | 786 | /* can't be more than PAGE_SIZE */ |
Al Viro | 62a8067 | 2014-04-04 23:12:29 -0400 | [diff] [blame] | 787 | *start = bvec->bv_offset + i->iov_offset; |
| 788 | |
| 789 | get_page(*pages = bvec->bv_page); |
| 790 | |
| 791 | return len; |
| 792 | } |
| 793 | |
| 794 | static ssize_t get_pages_alloc_bvec(struct iov_iter *i, |
| 795 | struct page ***pages, size_t maxsize, |
| 796 | size_t *start) |
| 797 | { |
| 798 | const struct bio_vec *bvec = i->bvec; |
| 799 | size_t len = bvec->bv_len - i->iov_offset; |
| 800 | if (len > i->count) |
| 801 | len = i->count; |
| 802 | if (len > maxsize) |
| 803 | len = maxsize; |
| 804 | *start = bvec->bv_offset + i->iov_offset; |
| 805 | |
| 806 | *pages = kmalloc(sizeof(struct page *), GFP_KERNEL); |
| 807 | if (!*pages) |
| 808 | return -ENOMEM; |
| 809 | |
| 810 | get_page(**pages = bvec->bv_page); |
| 811 | |
| 812 | return len; |
| 813 | } |
| 814 | |
| 815 | static int iov_iter_npages_bvec(const struct iov_iter *i, int maxpages) |
| 816 | { |
| 817 | size_t offset = i->iov_offset; |
| 818 | size_t size = i->count; |
| 819 | const struct bio_vec *bvec = i->bvec; |
| 820 | int npages = 0; |
| 821 | int n; |
| 822 | |
| 823 | for (n = 0; size && n < i->nr_segs; n++, bvec++) { |
| 824 | size_t len = bvec->bv_len - offset; |
| 825 | offset = 0; |
| 826 | if (unlikely(!len)) /* empty segment */ |
| 827 | continue; |
| 828 | if (len > size) |
| 829 | len = size; |
| 830 | npages++; |
| 831 | if (npages >= maxpages) /* don't bother going further */ |
| 832 | return maxpages; |
| 833 | size -= len; |
| 834 | offset = 0; |
| 835 | } |
| 836 | return min(npages, maxpages); |
| 837 | } |
| 838 | |
| 839 | size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes, |
| 840 | struct iov_iter *i) |
| 841 | { |
| 842 | if (i->type & ITER_BVEC) |
| 843 | return copy_page_to_iter_bvec(page, offset, bytes, i); |
| 844 | else |
| 845 | return copy_page_to_iter_iovec(page, offset, bytes, i); |
| 846 | } |
| 847 | EXPORT_SYMBOL(copy_page_to_iter); |
| 848 | |
| 849 | size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes, |
| 850 | struct iov_iter *i) |
| 851 | { |
| 852 | if (i->type & ITER_BVEC) |
| 853 | return copy_page_from_iter_bvec(page, offset, bytes, i); |
| 854 | else |
| 855 | return copy_page_from_iter_iovec(page, offset, bytes, i); |
| 856 | } |
| 857 | EXPORT_SYMBOL(copy_page_from_iter); |
| 858 | |
Matthew Wilcox | c35e024 | 2014-08-01 09:27:22 -0400 | [diff] [blame] | 859 | size_t copy_to_iter(void *addr, size_t bytes, struct iov_iter *i) |
| 860 | { |
| 861 | if (i->type & ITER_BVEC) |
| 862 | return copy_to_iter_bvec(addr, bytes, i); |
| 863 | else |
| 864 | return copy_to_iter_iovec(addr, bytes, i); |
| 865 | } |
| 866 | EXPORT_SYMBOL(copy_to_iter); |
| 867 | |
| 868 | size_t copy_from_iter(void *addr, size_t bytes, struct iov_iter *i) |
| 869 | { |
| 870 | if (i->type & ITER_BVEC) |
| 871 | return copy_from_iter_bvec(addr, bytes, i); |
| 872 | else |
| 873 | return copy_from_iter_iovec(addr, bytes, i); |
| 874 | } |
| 875 | EXPORT_SYMBOL(copy_from_iter); |
| 876 | |
| 877 | size_t iov_iter_zero(size_t bytes, struct iov_iter *i) |
| 878 | { |
| 879 | if (i->type & ITER_BVEC) { |
| 880 | return zero_bvec(bytes, i); |
| 881 | } else { |
| 882 | return zero_iovec(bytes, i); |
| 883 | } |
| 884 | } |
| 885 | EXPORT_SYMBOL(iov_iter_zero); |
| 886 | |
Al Viro | 62a8067 | 2014-04-04 23:12:29 -0400 | [diff] [blame] | 887 | size_t iov_iter_copy_from_user_atomic(struct page *page, |
| 888 | struct iov_iter *i, unsigned long offset, size_t bytes) |
| 889 | { |
| 890 | if (i->type & ITER_BVEC) |
| 891 | return copy_from_user_bvec(page, i, offset, bytes); |
| 892 | else |
| 893 | return copy_from_user_atomic_iovec(page, i, offset, bytes); |
| 894 | } |
| 895 | EXPORT_SYMBOL(iov_iter_copy_from_user_atomic); |
| 896 | |
| 897 | void iov_iter_advance(struct iov_iter *i, size_t size) |
| 898 | { |
| 899 | if (i->type & ITER_BVEC) |
| 900 | advance_bvec(i, size); |
| 901 | else |
| 902 | advance_iovec(i, size); |
| 903 | } |
| 904 | EXPORT_SYMBOL(iov_iter_advance); |
| 905 | |
| 906 | /* |
| 907 | * Return the count of just the current iov_iter segment. |
| 908 | */ |
| 909 | size_t iov_iter_single_seg_count(const struct iov_iter *i) |
| 910 | { |
| 911 | if (i->nr_segs == 1) |
| 912 | return i->count; |
| 913 | else if (i->type & ITER_BVEC) |
Al Viro | 62a8067 | 2014-04-04 23:12:29 -0400 | [diff] [blame] | 914 | return min(i->count, i->bvec->bv_len - i->iov_offset); |
Paul Mackerras | ad0eab9 | 2014-11-13 20:15:23 +1100 | [diff] [blame^] | 915 | else |
| 916 | return min(i->count, i->iov->iov_len - i->iov_offset); |
Al Viro | 62a8067 | 2014-04-04 23:12:29 -0400 | [diff] [blame] | 917 | } |
| 918 | EXPORT_SYMBOL(iov_iter_single_seg_count); |
| 919 | |
| 920 | unsigned long iov_iter_alignment(const struct iov_iter *i) |
| 921 | { |
| 922 | if (i->type & ITER_BVEC) |
| 923 | return alignment_bvec(i); |
| 924 | else |
| 925 | return alignment_iovec(i); |
| 926 | } |
| 927 | EXPORT_SYMBOL(iov_iter_alignment); |
| 928 | |
| 929 | ssize_t iov_iter_get_pages(struct iov_iter *i, |
Miklos Szeredi | 2c80929 | 2014-09-24 17:09:11 +0200 | [diff] [blame] | 930 | struct page **pages, size_t maxsize, unsigned maxpages, |
Al Viro | 62a8067 | 2014-04-04 23:12:29 -0400 | [diff] [blame] | 931 | size_t *start) |
| 932 | { |
| 933 | if (i->type & ITER_BVEC) |
Miklos Szeredi | 2c80929 | 2014-09-24 17:09:11 +0200 | [diff] [blame] | 934 | return get_pages_bvec(i, pages, maxsize, maxpages, start); |
Al Viro | 62a8067 | 2014-04-04 23:12:29 -0400 | [diff] [blame] | 935 | else |
Miklos Szeredi | 2c80929 | 2014-09-24 17:09:11 +0200 | [diff] [blame] | 936 | return get_pages_iovec(i, pages, maxsize, maxpages, start); |
Al Viro | 62a8067 | 2014-04-04 23:12:29 -0400 | [diff] [blame] | 937 | } |
| 938 | EXPORT_SYMBOL(iov_iter_get_pages); |
| 939 | |
| 940 | ssize_t iov_iter_get_pages_alloc(struct iov_iter *i, |
| 941 | struct page ***pages, size_t maxsize, |
| 942 | size_t *start) |
| 943 | { |
| 944 | if (i->type & ITER_BVEC) |
| 945 | return get_pages_alloc_bvec(i, pages, maxsize, start); |
| 946 | else |
| 947 | return get_pages_alloc_iovec(i, pages, maxsize, start); |
| 948 | } |
| 949 | EXPORT_SYMBOL(iov_iter_get_pages_alloc); |
| 950 | |
| 951 | int iov_iter_npages(const struct iov_iter *i, int maxpages) |
| 952 | { |
| 953 | if (i->type & ITER_BVEC) |
| 954 | return iov_iter_npages_bvec(i, maxpages); |
| 955 | else |
| 956 | return iov_iter_npages_iovec(i, maxpages); |
| 957 | } |
Al Viro | f67da30 | 2014-03-19 01:16:16 -0400 | [diff] [blame] | 958 | EXPORT_SYMBOL(iov_iter_npages); |