1457c8996SThomas Gleixner // SPDX-License-Identifier: GPL-2.0-only 27999096fSHerbert Xu #include <crypto/hash.h> 3d879cb83SAl Viro #include <linux/export.h> 42f8b5444SChristoph Hellwig #include <linux/bvec.h> 54d0e9df5SAlbert van der Linde #include <linux/fault-inject-usercopy.h> 6d879cb83SAl Viro #include <linux/uio.h> 7d879cb83SAl Viro #include <linux/pagemap.h> 828961998SIra Weiny #include <linux/highmem.h> 9d879cb83SAl Viro #include <linux/slab.h> 10d879cb83SAl Viro #include <linux/vmalloc.h> 11241699cdSAl Viro #include <linux/splice.h> 12bfdc5970SChristoph Hellwig #include <linux/compat.h> 13d879cb83SAl Viro #include <net/checksum.h> 14d05f4435SSagi Grimberg #include <linux/scatterlist.h> 15d0ef4c36SMarco Elver #include <linux/instrumented.h> 16d879cb83SAl Viro 17fcb14cb1SAl Viro /* covers ubuf and kbuf alike */ 18fcb14cb1SAl Viro #define iterate_buf(i, n, base, len, off, __p, STEP) { \ 19fcb14cb1SAl Viro size_t __maybe_unused off = 0; \ 20fcb14cb1SAl Viro len = n; \ 21fcb14cb1SAl Viro base = __p + i->iov_offset; \ 22fcb14cb1SAl Viro len -= (STEP); \ 23fcb14cb1SAl Viro i->iov_offset += len; \ 24fcb14cb1SAl Viro n = len; \ 25fcb14cb1SAl Viro } 26fcb14cb1SAl Viro 275c67aa90SAl Viro /* covers iovec and kvec alike */ 28a6e4ec7bSAl Viro #define iterate_iovec(i, n, base, len, off, __p, STEP) { \ 297baa5099SAl Viro size_t off = 0; \ 30a6e4ec7bSAl Viro size_t skip = i->iov_offset; \ 317a1bcb5dSAl Viro do { \ 327baa5099SAl Viro len = min(n, __p->iov_len - skip); \ 337baa5099SAl Viro if (likely(len)) { \ 347baa5099SAl Viro base = __p->iov_base + skip; \ 357baa5099SAl Viro len -= (STEP); \ 367baa5099SAl Viro off += len; \ 377baa5099SAl Viro skip += len; \ 387baa5099SAl Viro n -= len; \ 397a1bcb5dSAl Viro if (skip < __p->iov_len) \ 407a1bcb5dSAl Viro break; \ 41d879cb83SAl Viro } \ 42d879cb83SAl Viro __p++; \ 437a1bcb5dSAl Viro skip = 0; \ 447a1bcb5dSAl Viro } while (n); \ 45a6e4ec7bSAl Viro i->iov_offset = skip; \ 467baa5099SAl Viro n = off; \ 47d879cb83SAl Viro } 48d879cb83SAl Viro 49a6e4ec7bSAl Viro #define iterate_bvec(i, n, base, len, off, p, STEP) { \ 507baa5099SAl Viro size_t off = 0; \ 51a6e4ec7bSAl Viro unsigned skip = i->iov_offset; \ 527491a2bfSAl Viro while (n) { \ 537491a2bfSAl Viro unsigned offset = p->bv_offset + skip; \ 541b4fb5ffSAl Viro unsigned left; \ 5521b56c84SAl Viro void *kaddr = kmap_local_page(p->bv_page + \ 5621b56c84SAl Viro offset / PAGE_SIZE); \ 577baa5099SAl Viro base = kaddr + offset % PAGE_SIZE; \ 58a6e4ec7bSAl Viro len = min(min(n, (size_t)(p->bv_len - skip)), \ 597491a2bfSAl Viro (size_t)(PAGE_SIZE - offset % PAGE_SIZE)); \ 601b4fb5ffSAl Viro left = (STEP); \ 6121b56c84SAl Viro kunmap_local(kaddr); \ 627baa5099SAl Viro len -= left; \ 637baa5099SAl Viro off += len; \ 647baa5099SAl Viro skip += len; \ 657491a2bfSAl Viro if (skip == p->bv_len) { \ 667491a2bfSAl Viro skip = 0; \ 677491a2bfSAl Viro p++; \ 68d879cb83SAl Viro } \ 697baa5099SAl Viro n -= len; \ 701b4fb5ffSAl Viro if (left) \ 711b4fb5ffSAl Viro break; \ 727491a2bfSAl Viro } \ 73a6e4ec7bSAl Viro i->iov_offset = skip; \ 747baa5099SAl Viro n = off; \ 75d879cb83SAl Viro } 76d879cb83SAl Viro 77a6e4ec7bSAl Viro #define iterate_xarray(i, n, base, len, __off, STEP) { \ 781b4fb5ffSAl Viro __label__ __out; \ 79622838f3SAl Viro size_t __off = 0; \ 80821979f5SMatthew Wilcox (Oracle) struct folio *folio; \ 81a6e4ec7bSAl Viro loff_t start = i->xarray_start + i->iov_offset; \ 824b179e9aSAl Viro pgoff_t index = start / PAGE_SIZE; \ 837ff50620SDavid Howells XA_STATE(xas, i->xarray, index); \ 847ff50620SDavid Howells \ 85821979f5SMatthew Wilcox (Oracle) len = PAGE_SIZE - offset_in_page(start); \ 867ff50620SDavid Howells rcu_read_lock(); \ 87821979f5SMatthew Wilcox (Oracle) xas_for_each(&xas, folio, ULONG_MAX) { \ 881b4fb5ffSAl Viro unsigned left; \ 89821979f5SMatthew Wilcox (Oracle) size_t offset; \ 90821979f5SMatthew Wilcox (Oracle) if (xas_retry(&xas, folio)) \ 917ff50620SDavid Howells continue; \ 92821979f5SMatthew Wilcox (Oracle) if (WARN_ON(xa_is_value(folio))) \ 937ff50620SDavid Howells break; \ 94821979f5SMatthew Wilcox (Oracle) if (WARN_ON(folio_test_hugetlb(folio))) \ 957ff50620SDavid Howells break; \ 96821979f5SMatthew Wilcox (Oracle) offset = offset_in_folio(folio, start + __off); \ 97821979f5SMatthew Wilcox (Oracle) while (offset < folio_size(folio)) { \ 98821979f5SMatthew Wilcox (Oracle) base = kmap_local_folio(folio, offset); \ 997baa5099SAl Viro len = min(n, len); \ 1001b4fb5ffSAl Viro left = (STEP); \ 101821979f5SMatthew Wilcox (Oracle) kunmap_local(base); \ 1027baa5099SAl Viro len -= left; \ 1037baa5099SAl Viro __off += len; \ 1047baa5099SAl Viro n -= len; \ 1051b4fb5ffSAl Viro if (left || n == 0) \ 1061b4fb5ffSAl Viro goto __out; \ 107821979f5SMatthew Wilcox (Oracle) offset += len; \ 108821979f5SMatthew Wilcox (Oracle) len = PAGE_SIZE; \ 1097ff50620SDavid Howells } \ 1107ff50620SDavid Howells } \ 1111b4fb5ffSAl Viro __out: \ 1127ff50620SDavid Howells rcu_read_unlock(); \ 113a6e4ec7bSAl Viro i->iov_offset += __off; \ 114622838f3SAl Viro n = __off; \ 1157ff50620SDavid Howells } 1167ff50620SDavid Howells 1177baa5099SAl Viro #define __iterate_and_advance(i, n, base, len, off, I, K) { \ 118dd254f5aSAl Viro if (unlikely(i->count < n)) \ 119dd254f5aSAl Viro n = i->count; \ 120f5da8354SAl Viro if (likely(n)) { \ 121fcb14cb1SAl Viro if (likely(iter_is_ubuf(i))) { \ 122fcb14cb1SAl Viro void __user *base; \ 123fcb14cb1SAl Viro size_t len; \ 124fcb14cb1SAl Viro iterate_buf(i, n, base, len, off, \ 125fcb14cb1SAl Viro i->ubuf, (I)) \ 126fcb14cb1SAl Viro } else if (likely(iter_is_iovec(i))) { \ 127de4f5fedSJens Axboe const struct iovec *iov = iter_iov(i); \ 1287baa5099SAl Viro void __user *base; \ 1297baa5099SAl Viro size_t len; \ 1307baa5099SAl Viro iterate_iovec(i, n, base, len, off, \ 131a6e4ec7bSAl Viro iov, (I)) \ 132de4f5fedSJens Axboe i->nr_segs -= iov - iter_iov(i); \ 133de4f5fedSJens Axboe i->__iov = iov; \ 13428f38db7SAl Viro } else if (iov_iter_is_bvec(i)) { \ 13528f38db7SAl Viro const struct bio_vec *bvec = i->bvec; \ 1367baa5099SAl Viro void *base; \ 1377baa5099SAl Viro size_t len; \ 1387baa5099SAl Viro iterate_bvec(i, n, base, len, off, \ 139a6e4ec7bSAl Viro bvec, (K)) \ 1407491a2bfSAl Viro i->nr_segs -= bvec - i->bvec; \ 1417491a2bfSAl Viro i->bvec = bvec; \ 14228f38db7SAl Viro } else if (iov_iter_is_kvec(i)) { \ 1435c67aa90SAl Viro const struct kvec *kvec = i->kvec; \ 1447baa5099SAl Viro void *base; \ 1457baa5099SAl Viro size_t len; \ 1467baa5099SAl Viro iterate_iovec(i, n, base, len, off, \ 147a6e4ec7bSAl Viro kvec, (K)) \ 14828f38db7SAl Viro i->nr_segs -= kvec - i->kvec; \ 14928f38db7SAl Viro i->kvec = kvec; \ 15028f38db7SAl Viro } else if (iov_iter_is_xarray(i)) { \ 1517baa5099SAl Viro void *base; \ 1527baa5099SAl Viro size_t len; \ 1537baa5099SAl Viro iterate_xarray(i, n, base, len, off, \ 154a6e4ec7bSAl Viro (K)) \ 155d879cb83SAl Viro } \ 156d879cb83SAl Viro i->count -= n; \ 157dd254f5aSAl Viro } \ 158d879cb83SAl Viro } 1597baa5099SAl Viro #define iterate_and_advance(i, n, base, len, off, I, K) \ 1607baa5099SAl Viro __iterate_and_advance(i, n, base, len, off, I, ((void)(K),0)) 161d879cb83SAl Viro 16209fc68dcSAl Viro static int copyout(void __user *to, const void *from, size_t n) 16309fc68dcSAl Viro { 1644d0e9df5SAlbert van der Linde if (should_fail_usercopy()) 1654d0e9df5SAlbert van der Linde return n; 16696d4f267SLinus Torvalds if (access_ok(to, n)) { 167d0ef4c36SMarco Elver instrument_copy_to_user(to, from, n); 16809fc68dcSAl Viro n = raw_copy_to_user(to, from, n); 16909fc68dcSAl Viro } 17009fc68dcSAl Viro return n; 17109fc68dcSAl Viro } 17209fc68dcSAl Viro 1734f80818bSLorenzo Stoakes static int copyout_nofault(void __user *to, const void *from, size_t n) 1744f80818bSLorenzo Stoakes { 1754f80818bSLorenzo Stoakes long res; 1764f80818bSLorenzo Stoakes 1774f80818bSLorenzo Stoakes if (should_fail_usercopy()) 1784f80818bSLorenzo Stoakes return n; 1794f80818bSLorenzo Stoakes 1804f80818bSLorenzo Stoakes res = copy_to_user_nofault(to, from, n); 1814f80818bSLorenzo Stoakes 1824f80818bSLorenzo Stoakes return res < 0 ? n : res; 1834f80818bSLorenzo Stoakes } 1844f80818bSLorenzo Stoakes 18509fc68dcSAl Viro static int copyin(void *to, const void __user *from, size_t n) 18609fc68dcSAl Viro { 18733b75c1dSAlexander Potapenko size_t res = n; 18833b75c1dSAlexander Potapenko 1894d0e9df5SAlbert van der Linde if (should_fail_usercopy()) 1904d0e9df5SAlbert van der Linde return n; 19196d4f267SLinus Torvalds if (access_ok(from, n)) { 19233b75c1dSAlexander Potapenko instrument_copy_from_user_before(to, from, n); 19333b75c1dSAlexander Potapenko res = raw_copy_from_user(to, from, n); 19433b75c1dSAlexander Potapenko instrument_copy_from_user_after(to, from, n, res); 19509fc68dcSAl Viro } 19633b75c1dSAlexander Potapenko return res; 19709fc68dcSAl Viro } 19809fc68dcSAl Viro 199d879cb83SAl Viro /* 200a6294593SAndreas Gruenbacher * fault_in_iov_iter_readable - fault in iov iterator for reading 201a6294593SAndreas Gruenbacher * @i: iterator 202a6294593SAndreas Gruenbacher * @size: maximum length 203171a0203SAnton Altaparmakov * 204a6294593SAndreas Gruenbacher * Fault in one or more iovecs of the given iov_iter, to a maximum length of 205a6294593SAndreas Gruenbacher * @size. For each iovec, fault in each page that constitutes the iovec. 206a6294593SAndreas Gruenbacher * 207a6294593SAndreas Gruenbacher * Returns the number of bytes not faulted in (like copy_to_user() and 208a6294593SAndreas Gruenbacher * copy_from_user()). 209a6294593SAndreas Gruenbacher * 210a6294593SAndreas Gruenbacher * Always returns 0 for non-userspace iterators. 211171a0203SAnton Altaparmakov */ 212a6294593SAndreas Gruenbacher size_t fault_in_iov_iter_readable(const struct iov_iter *i, size_t size) 213171a0203SAnton Altaparmakov { 214fcb14cb1SAl Viro if (iter_is_ubuf(i)) { 215fcb14cb1SAl Viro size_t n = min(size, iov_iter_count(i)); 216fcb14cb1SAl Viro n -= fault_in_readable(i->ubuf + i->iov_offset, n); 217fcb14cb1SAl Viro return size - n; 218fcb14cb1SAl Viro } else if (iter_is_iovec(i)) { 219a6294593SAndreas Gruenbacher size_t count = min(size, iov_iter_count(i)); 2208409a0d2SAl Viro const struct iovec *p; 2218409a0d2SAl Viro size_t skip; 2228409a0d2SAl Viro 223a6294593SAndreas Gruenbacher size -= count; 224de4f5fedSJens Axboe for (p = iter_iov(i), skip = i->iov_offset; count; p++, skip = 0) { 225a6294593SAndreas Gruenbacher size_t len = min(count, p->iov_len - skip); 226a6294593SAndreas Gruenbacher size_t ret; 2278409a0d2SAl Viro 2288409a0d2SAl Viro if (unlikely(!len)) 2298409a0d2SAl Viro continue; 230a6294593SAndreas Gruenbacher ret = fault_in_readable(p->iov_base + skip, len); 231a6294593SAndreas Gruenbacher count -= len - ret; 232a6294593SAndreas Gruenbacher if (ret) 233a6294593SAndreas Gruenbacher break; 2348409a0d2SAl Viro } 235a6294593SAndreas Gruenbacher return count + size; 236171a0203SAnton Altaparmakov } 237171a0203SAnton Altaparmakov return 0; 238171a0203SAnton Altaparmakov } 239a6294593SAndreas Gruenbacher EXPORT_SYMBOL(fault_in_iov_iter_readable); 240171a0203SAnton Altaparmakov 241cdd591fcSAndreas Gruenbacher /* 242cdd591fcSAndreas Gruenbacher * fault_in_iov_iter_writeable - fault in iov iterator for writing 243cdd591fcSAndreas Gruenbacher * @i: iterator 244cdd591fcSAndreas Gruenbacher * @size: maximum length 245cdd591fcSAndreas Gruenbacher * 246cdd591fcSAndreas Gruenbacher * Faults in the iterator using get_user_pages(), i.e., without triggering 247cdd591fcSAndreas Gruenbacher * hardware page faults. This is primarily useful when we already know that 248cdd591fcSAndreas Gruenbacher * some or all of the pages in @i aren't in memory. 249cdd591fcSAndreas Gruenbacher * 250cdd591fcSAndreas Gruenbacher * Returns the number of bytes not faulted in, like copy_to_user() and 251cdd591fcSAndreas Gruenbacher * copy_from_user(). 252cdd591fcSAndreas Gruenbacher * 253cdd591fcSAndreas Gruenbacher * Always returns 0 for non-user-space iterators. 254cdd591fcSAndreas Gruenbacher */ 255cdd591fcSAndreas Gruenbacher size_t fault_in_iov_iter_writeable(const struct iov_iter *i, size_t size) 256cdd591fcSAndreas Gruenbacher { 257fcb14cb1SAl Viro if (iter_is_ubuf(i)) { 258fcb14cb1SAl Viro size_t n = min(size, iov_iter_count(i)); 259fcb14cb1SAl Viro n -= fault_in_safe_writeable(i->ubuf + i->iov_offset, n); 260fcb14cb1SAl Viro return size - n; 261fcb14cb1SAl Viro } else if (iter_is_iovec(i)) { 262cdd591fcSAndreas Gruenbacher size_t count = min(size, iov_iter_count(i)); 263cdd591fcSAndreas Gruenbacher const struct iovec *p; 264cdd591fcSAndreas Gruenbacher size_t skip; 265cdd591fcSAndreas Gruenbacher 266cdd591fcSAndreas Gruenbacher size -= count; 267de4f5fedSJens Axboe for (p = iter_iov(i), skip = i->iov_offset; count; p++, skip = 0) { 268cdd591fcSAndreas Gruenbacher size_t len = min(count, p->iov_len - skip); 269cdd591fcSAndreas Gruenbacher size_t ret; 270cdd591fcSAndreas Gruenbacher 271cdd591fcSAndreas Gruenbacher if (unlikely(!len)) 272cdd591fcSAndreas Gruenbacher continue; 273cdd591fcSAndreas Gruenbacher ret = fault_in_safe_writeable(p->iov_base + skip, len); 274cdd591fcSAndreas Gruenbacher count -= len - ret; 275cdd591fcSAndreas Gruenbacher if (ret) 276cdd591fcSAndreas Gruenbacher break; 277cdd591fcSAndreas Gruenbacher } 278cdd591fcSAndreas Gruenbacher return count + size; 279cdd591fcSAndreas Gruenbacher } 280cdd591fcSAndreas Gruenbacher return 0; 281cdd591fcSAndreas Gruenbacher } 282cdd591fcSAndreas Gruenbacher EXPORT_SYMBOL(fault_in_iov_iter_writeable); 283cdd591fcSAndreas Gruenbacher 284aa563d7bSDavid Howells void iov_iter_init(struct iov_iter *i, unsigned int direction, 285d879cb83SAl Viro const struct iovec *iov, unsigned long nr_segs, 286d879cb83SAl Viro size_t count) 287d879cb83SAl Viro { 288aa563d7bSDavid Howells WARN_ON(direction & ~(READ | WRITE)); 2898cd54c1cSAl Viro *i = (struct iov_iter) { 2908cd54c1cSAl Viro .iter_type = ITER_IOVEC, 291245f0922SKefeng Wang .copy_mc = false, 2923337ab08SAndreas Gruenbacher .nofault = false, 293fcb14cb1SAl Viro .user_backed = true, 2948cd54c1cSAl Viro .data_source = direction, 295de4f5fedSJens Axboe .__iov = iov, 2968cd54c1cSAl Viro .nr_segs = nr_segs, 2978cd54c1cSAl Viro .iov_offset = 0, 2988cd54c1cSAl Viro .count = count 2998cd54c1cSAl Viro }; 300d879cb83SAl Viro } 301d879cb83SAl Viro EXPORT_SYMBOL(iov_iter_init); 302d879cb83SAl Viro 303f9152895SAl Viro static __wsum csum_and_memcpy(void *to, const void *from, size_t len, 304f9152895SAl Viro __wsum sum, size_t off) 305f9152895SAl Viro { 306cc44c17bSAl Viro __wsum next = csum_partial_copy_nocheck(from, to, len); 307f9152895SAl Viro return csum_block_add(sum, next, off); 308f9152895SAl Viro } 309f9152895SAl Viro 310aa28de27SAl Viro size_t _copy_to_iter(const void *addr, size_t bytes, struct iov_iter *i) 311d879cb83SAl Viro { 312a41dad90SAl Viro if (WARN_ON_ONCE(i->data_source)) 313a41dad90SAl Viro return 0; 314fcb14cb1SAl Viro if (user_backed_iter(i)) 31509fc68dcSAl Viro might_fault(); 3167baa5099SAl Viro iterate_and_advance(i, bytes, base, len, off, 3177baa5099SAl Viro copyout(base, addr + off, len), 3187baa5099SAl Viro memcpy(base, addr + off, len) 319d879cb83SAl Viro ) 320d879cb83SAl Viro 321d879cb83SAl Viro return bytes; 322d879cb83SAl Viro } 323aa28de27SAl Viro EXPORT_SYMBOL(_copy_to_iter); 324d879cb83SAl Viro 325ec6347bbSDan Williams #ifdef CONFIG_ARCH_HAS_COPY_MC 326ec6347bbSDan Williams static int copyout_mc(void __user *to, const void *from, size_t n) 3278780356eSDan Williams { 32896d4f267SLinus Torvalds if (access_ok(to, n)) { 329d0ef4c36SMarco Elver instrument_copy_to_user(to, from, n); 330ec6347bbSDan Williams n = copy_mc_to_user((__force void *) to, from, n); 3318780356eSDan Williams } 3328780356eSDan Williams return n; 3338780356eSDan Williams } 3348780356eSDan Williams 335bf3eeb9bSDan Williams /** 336ec6347bbSDan Williams * _copy_mc_to_iter - copy to iter with source memory error exception handling 337bf3eeb9bSDan Williams * @addr: source kernel address 338bf3eeb9bSDan Williams * @bytes: total transfer length 33944e55997SRandy Dunlap * @i: destination iterator 340bf3eeb9bSDan Williams * 341ec6347bbSDan Williams * The pmem driver deploys this for the dax operation 342ec6347bbSDan Williams * (dax_copy_to_iter()) for dax reads (bypass page-cache and the 343ec6347bbSDan Williams * block-layer). Upon #MC read(2) aborts and returns EIO or the bytes 344ec6347bbSDan Williams * successfully copied. 345bf3eeb9bSDan Williams * 346ec6347bbSDan Williams * The main differences between this and typical _copy_to_iter(). 347bf3eeb9bSDan Williams * 348bf3eeb9bSDan Williams * * Typical tail/residue handling after a fault retries the copy 349bf3eeb9bSDan Williams * byte-by-byte until the fault happens again. Re-triggering machine 350bf3eeb9bSDan Williams * checks is potentially fatal so the implementation uses source 351bf3eeb9bSDan Williams * alignment and poison alignment assumptions to avoid re-triggering 352bf3eeb9bSDan Williams * hardware exceptions. 353bf3eeb9bSDan Williams * 3543fc40265SDavid Howells * * ITER_KVEC and ITER_BVEC can return short copies. Compare to 3553fc40265SDavid Howells * copy_to_iter() where only ITER_IOVEC attempts might return a short copy. 35644e55997SRandy Dunlap * 35744e55997SRandy Dunlap * Return: number of bytes copied (may be %0) 358bf3eeb9bSDan Williams */ 359ec6347bbSDan Williams size_t _copy_mc_to_iter(const void *addr, size_t bytes, struct iov_iter *i) 3608780356eSDan Williams { 361a41dad90SAl Viro if (WARN_ON_ONCE(i->data_source)) 362a41dad90SAl Viro return 0; 363fcb14cb1SAl Viro if (user_backed_iter(i)) 3648780356eSDan Williams might_fault(); 3657baa5099SAl Viro __iterate_and_advance(i, bytes, base, len, off, 3667baa5099SAl Viro copyout_mc(base, addr + off, len), 3677baa5099SAl Viro copy_mc_to_kernel(base, addr + off, len) 3688780356eSDan Williams ) 3698780356eSDan Williams 3708780356eSDan Williams return bytes; 3718780356eSDan Williams } 372ec6347bbSDan Williams EXPORT_SYMBOL_GPL(_copy_mc_to_iter); 373ec6347bbSDan Williams #endif /* CONFIG_ARCH_HAS_COPY_MC */ 3748780356eSDan Williams 375245f0922SKefeng Wang static void *memcpy_from_iter(struct iov_iter *i, void *to, const void *from, 376245f0922SKefeng Wang size_t size) 377245f0922SKefeng Wang { 378245f0922SKefeng Wang if (iov_iter_is_copy_mc(i)) 379245f0922SKefeng Wang return (void *)copy_mc_to_kernel(to, from, size); 380245f0922SKefeng Wang return memcpy(to, from, size); 381245f0922SKefeng Wang } 382245f0922SKefeng Wang 383aa28de27SAl Viro size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i) 384d879cb83SAl Viro { 385a41dad90SAl Viro if (WARN_ON_ONCE(!i->data_source)) 386241699cdSAl Viro return 0; 387a41dad90SAl Viro 388fcb14cb1SAl Viro if (user_backed_iter(i)) 38909fc68dcSAl Viro might_fault(); 3907baa5099SAl Viro iterate_and_advance(i, bytes, base, len, off, 3917baa5099SAl Viro copyin(addr + off, base, len), 392245f0922SKefeng Wang memcpy_from_iter(i, addr + off, base, len) 393d879cb83SAl Viro ) 394d879cb83SAl Viro 395d879cb83SAl Viro return bytes; 396d879cb83SAl Viro } 397aa28de27SAl Viro EXPORT_SYMBOL(_copy_from_iter); 398d879cb83SAl Viro 399aa28de27SAl Viro size_t _copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i) 400d879cb83SAl Viro { 401a41dad90SAl Viro if (WARN_ON_ONCE(!i->data_source)) 402241699cdSAl Viro return 0; 403a41dad90SAl Viro 4047baa5099SAl Viro iterate_and_advance(i, bytes, base, len, off, 4057baa5099SAl Viro __copy_from_user_inatomic_nocache(addr + off, base, len), 4067baa5099SAl Viro memcpy(addr + off, base, len) 407d879cb83SAl Viro ) 408d879cb83SAl Viro 409d879cb83SAl Viro return bytes; 410d879cb83SAl Viro } 411aa28de27SAl Viro EXPORT_SYMBOL(_copy_from_iter_nocache); 412d879cb83SAl Viro 4130aed55afSDan Williams #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE 414abd08d7dSDan Williams /** 415abd08d7dSDan Williams * _copy_from_iter_flushcache - write destination through cpu cache 416abd08d7dSDan Williams * @addr: destination kernel address 417abd08d7dSDan Williams * @bytes: total transfer length 41844e55997SRandy Dunlap * @i: source iterator 419abd08d7dSDan Williams * 420abd08d7dSDan Williams * The pmem driver arranges for filesystem-dax to use this facility via 421abd08d7dSDan Williams * dax_copy_from_iter() for ensuring that writes to persistent memory 422abd08d7dSDan Williams * are flushed through the CPU cache. It is differentiated from 423abd08d7dSDan Williams * _copy_from_iter_nocache() in that guarantees all data is flushed for 424abd08d7dSDan Williams * all iterator types. The _copy_from_iter_nocache() only attempts to 425abd08d7dSDan Williams * bypass the cache for the ITER_IOVEC case, and on some archs may use 426abd08d7dSDan Williams * instructions that strand dirty-data in the cache. 42744e55997SRandy Dunlap * 42844e55997SRandy Dunlap * Return: number of bytes copied (may be %0) 429abd08d7dSDan Williams */ 4306a37e940SLinus Torvalds size_t _copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i) 4310aed55afSDan Williams { 432a41dad90SAl Viro if (WARN_ON_ONCE(!i->data_source)) 4330aed55afSDan Williams return 0; 434a41dad90SAl Viro 4357baa5099SAl Viro iterate_and_advance(i, bytes, base, len, off, 4367baa5099SAl Viro __copy_from_user_flushcache(addr + off, base, len), 4377baa5099SAl Viro memcpy_flushcache(addr + off, base, len) 4380aed55afSDan Williams ) 4390aed55afSDan Williams 4400aed55afSDan Williams return bytes; 4410aed55afSDan Williams } 4426a37e940SLinus Torvalds EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache); 4430aed55afSDan Williams #endif 4440aed55afSDan Williams 44572e809edSAl Viro static inline bool page_copy_sane(struct page *page, size_t offset, size_t n) 44672e809edSAl Viro { 4476daef95bSEric Dumazet struct page *head; 4486daef95bSEric Dumazet size_t v = n + offset; 4496daef95bSEric Dumazet 4506daef95bSEric Dumazet /* 4516daef95bSEric Dumazet * The general case needs to access the page order in order 4526daef95bSEric Dumazet * to compute the page size. 4536daef95bSEric Dumazet * However, we mostly deal with order-0 pages and thus can 4546daef95bSEric Dumazet * avoid a possible cache line miss for requests that fit all 4556daef95bSEric Dumazet * page orders. 4566daef95bSEric Dumazet */ 4576daef95bSEric Dumazet if (n <= v && v <= PAGE_SIZE) 4586daef95bSEric Dumazet return true; 4596daef95bSEric Dumazet 4606daef95bSEric Dumazet head = compound_head(page); 4616daef95bSEric Dumazet v += (page - head) << PAGE_SHIFT; 462a90bcb86SPetar Penkov 46340a86061SAl Viro if (WARN_ON(n > v || v > page_size(head))) 46472e809edSAl Viro return false; 46540a86061SAl Viro return true; 46672e809edSAl Viro } 467cbbd26b8SAl Viro 46808aa6479SAl Viro size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes, 46908aa6479SAl Viro struct iov_iter *i) 47008aa6479SAl Viro { 47108aa6479SAl Viro size_t res = 0; 47240a86061SAl Viro if (!page_copy_sane(page, offset, bytes)) 47308aa6479SAl Viro return 0; 474a41dad90SAl Viro if (WARN_ON_ONCE(i->data_source)) 475a41dad90SAl Viro return 0; 47608aa6479SAl Viro page += offset / PAGE_SIZE; // first subpage 47708aa6479SAl Viro offset %= PAGE_SIZE; 47808aa6479SAl Viro while (1) { 479f0f6b614SAl Viro void *kaddr = kmap_local_page(page); 480f0f6b614SAl Viro size_t n = min(bytes, (size_t)PAGE_SIZE - offset); 481f0f6b614SAl Viro n = _copy_to_iter(kaddr + offset, n, i); 482f0f6b614SAl Viro kunmap_local(kaddr); 48308aa6479SAl Viro res += n; 48408aa6479SAl Viro bytes -= n; 48508aa6479SAl Viro if (!bytes || !n) 48608aa6479SAl Viro break; 48708aa6479SAl Viro offset += n; 48808aa6479SAl Viro if (offset == PAGE_SIZE) { 48908aa6479SAl Viro page++; 49008aa6479SAl Viro offset = 0; 49108aa6479SAl Viro } 49208aa6479SAl Viro } 49308aa6479SAl Viro return res; 49408aa6479SAl Viro } 495d879cb83SAl Viro EXPORT_SYMBOL(copy_page_to_iter); 496d879cb83SAl Viro 4974f80818bSLorenzo Stoakes size_t copy_page_to_iter_nofault(struct page *page, unsigned offset, size_t bytes, 4984f80818bSLorenzo Stoakes struct iov_iter *i) 4994f80818bSLorenzo Stoakes { 5004f80818bSLorenzo Stoakes size_t res = 0; 5014f80818bSLorenzo Stoakes 5024f80818bSLorenzo Stoakes if (!page_copy_sane(page, offset, bytes)) 5034f80818bSLorenzo Stoakes return 0; 5044f80818bSLorenzo Stoakes if (WARN_ON_ONCE(i->data_source)) 5054f80818bSLorenzo Stoakes return 0; 5064f80818bSLorenzo Stoakes page += offset / PAGE_SIZE; // first subpage 5074f80818bSLorenzo Stoakes offset %= PAGE_SIZE; 5084f80818bSLorenzo Stoakes while (1) { 5094f80818bSLorenzo Stoakes void *kaddr = kmap_local_page(page); 5104f80818bSLorenzo Stoakes size_t n = min(bytes, (size_t)PAGE_SIZE - offset); 5114f80818bSLorenzo Stoakes 5124f80818bSLorenzo Stoakes iterate_and_advance(i, n, base, len, off, 5134f80818bSLorenzo Stoakes copyout_nofault(base, kaddr + offset + off, len), 5144f80818bSLorenzo Stoakes memcpy(base, kaddr + offset + off, len) 5154f80818bSLorenzo Stoakes ) 5164f80818bSLorenzo Stoakes kunmap_local(kaddr); 5174f80818bSLorenzo Stoakes res += n; 5184f80818bSLorenzo Stoakes bytes -= n; 5194f80818bSLorenzo Stoakes if (!bytes || !n) 5204f80818bSLorenzo Stoakes break; 5214f80818bSLorenzo Stoakes offset += n; 5224f80818bSLorenzo Stoakes if (offset == PAGE_SIZE) { 5234f80818bSLorenzo Stoakes page++; 5244f80818bSLorenzo Stoakes offset = 0; 5254f80818bSLorenzo Stoakes } 5264f80818bSLorenzo Stoakes } 5274f80818bSLorenzo Stoakes return res; 5284f80818bSLorenzo Stoakes } 5294f80818bSLorenzo Stoakes EXPORT_SYMBOL(copy_page_to_iter_nofault); 5304f80818bSLorenzo Stoakes 531d879cb83SAl Viro size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes, 532d879cb83SAl Viro struct iov_iter *i) 533d879cb83SAl Viro { 534c03f05f1SAl Viro size_t res = 0; 535c03f05f1SAl Viro if (!page_copy_sane(page, offset, bytes)) 53628f38db7SAl Viro return 0; 537c03f05f1SAl Viro page += offset / PAGE_SIZE; // first subpage 538c03f05f1SAl Viro offset %= PAGE_SIZE; 539c03f05f1SAl Viro while (1) { 540c03f05f1SAl Viro void *kaddr = kmap_local_page(page); 541c03f05f1SAl Viro size_t n = min(bytes, (size_t)PAGE_SIZE - offset); 542c03f05f1SAl Viro n = _copy_from_iter(kaddr + offset, n, i); 543c03f05f1SAl Viro kunmap_local(kaddr); 544c03f05f1SAl Viro res += n; 545c03f05f1SAl Viro bytes -= n; 546c03f05f1SAl Viro if (!bytes || !n) 547c03f05f1SAl Viro break; 548c03f05f1SAl Viro offset += n; 549c03f05f1SAl Viro if (offset == PAGE_SIZE) { 550c03f05f1SAl Viro page++; 551c03f05f1SAl Viro offset = 0; 552c03f05f1SAl Viro } 553c03f05f1SAl Viro } 554c03f05f1SAl Viro return res; 555d879cb83SAl Viro } 556d879cb83SAl Viro EXPORT_SYMBOL(copy_page_from_iter); 557d879cb83SAl Viro 558d879cb83SAl Viro size_t iov_iter_zero(size_t bytes, struct iov_iter *i) 559d879cb83SAl Viro { 5607baa5099SAl Viro iterate_and_advance(i, bytes, base, len, count, 5617baa5099SAl Viro clear_user(base, len), 5627baa5099SAl Viro memset(base, 0, len) 563d879cb83SAl Viro ) 564d879cb83SAl Viro 565d879cb83SAl Viro return bytes; 566d879cb83SAl Viro } 567d879cb83SAl Viro EXPORT_SYMBOL(iov_iter_zero); 568d879cb83SAl Viro 569*908a1ad8SMatthew Wilcox (Oracle) size_t copy_page_from_iter_atomic(struct page *page, unsigned offset, 570*908a1ad8SMatthew Wilcox (Oracle) size_t bytes, struct iov_iter *i) 571d879cb83SAl Viro { 572*908a1ad8SMatthew Wilcox (Oracle) size_t n, copied = 0; 573f7f9a0c8SMatthew Wilcox (Oracle) 574f7f9a0c8SMatthew Wilcox (Oracle) if (!page_copy_sane(page, offset, bytes)) 57572e809edSAl Viro return 0; 576f7f9a0c8SMatthew Wilcox (Oracle) if (WARN_ON_ONCE(!i->data_source)) 577241699cdSAl Viro return 0; 578f7f9a0c8SMatthew Wilcox (Oracle) 579*908a1ad8SMatthew Wilcox (Oracle) do { 580*908a1ad8SMatthew Wilcox (Oracle) char *p; 581*908a1ad8SMatthew Wilcox (Oracle) 582*908a1ad8SMatthew Wilcox (Oracle) n = bytes - copied; 583*908a1ad8SMatthew Wilcox (Oracle) if (PageHighMem(page)) { 584*908a1ad8SMatthew Wilcox (Oracle) page += offset / PAGE_SIZE; 585*908a1ad8SMatthew Wilcox (Oracle) offset %= PAGE_SIZE; 586*908a1ad8SMatthew Wilcox (Oracle) n = min_t(size_t, n, PAGE_SIZE - offset); 587*908a1ad8SMatthew Wilcox (Oracle) } 588*908a1ad8SMatthew Wilcox (Oracle) 589f7f9a0c8SMatthew Wilcox (Oracle) p = kmap_atomic(page) + offset; 590*908a1ad8SMatthew Wilcox (Oracle) iterate_and_advance(i, n, base, len, off, 5917baa5099SAl Viro copyin(p + off, base, len), 592245f0922SKefeng Wang memcpy_from_iter(i, p + off, base, len) 593d879cb83SAl Viro ) 594f7f9a0c8SMatthew Wilcox (Oracle) kunmap_atomic(p); 595*908a1ad8SMatthew Wilcox (Oracle) copied += n; 596*908a1ad8SMatthew Wilcox (Oracle) offset += n; 597*908a1ad8SMatthew Wilcox (Oracle) } while (PageHighMem(page) && copied != bytes && n > 0); 598f7f9a0c8SMatthew Wilcox (Oracle) 599*908a1ad8SMatthew Wilcox (Oracle) return copied; 600d879cb83SAl Viro } 601f0b65f39SAl Viro EXPORT_SYMBOL(copy_page_from_iter_atomic); 602d879cb83SAl Viro 60354c8195bSPavel Begunkov static void iov_iter_bvec_advance(struct iov_iter *i, size_t size) 60454c8195bSPavel Begunkov { 60518fa9af7SAl Viro const struct bio_vec *bvec, *end; 60654c8195bSPavel Begunkov 60718fa9af7SAl Viro if (!i->count) 60818fa9af7SAl Viro return; 60918fa9af7SAl Viro i->count -= size; 61054c8195bSPavel Begunkov 61118fa9af7SAl Viro size += i->iov_offset; 61218fa9af7SAl Viro 61318fa9af7SAl Viro for (bvec = i->bvec, end = bvec + i->nr_segs; bvec < end; bvec++) { 61418fa9af7SAl Viro if (likely(size < bvec->bv_len)) 61518fa9af7SAl Viro break; 61618fa9af7SAl Viro size -= bvec->bv_len; 61718fa9af7SAl Viro } 61818fa9af7SAl Viro i->iov_offset = size; 61918fa9af7SAl Viro i->nr_segs -= bvec - i->bvec; 62018fa9af7SAl Viro i->bvec = bvec; 62154c8195bSPavel Begunkov } 62254c8195bSPavel Begunkov 623185ac4d4SAl Viro static void iov_iter_iovec_advance(struct iov_iter *i, size_t size) 624185ac4d4SAl Viro { 625185ac4d4SAl Viro const struct iovec *iov, *end; 626185ac4d4SAl Viro 627185ac4d4SAl Viro if (!i->count) 628185ac4d4SAl Viro return; 629185ac4d4SAl Viro i->count -= size; 630185ac4d4SAl Viro 631185ac4d4SAl Viro size += i->iov_offset; // from beginning of current segment 632de4f5fedSJens Axboe for (iov = iter_iov(i), end = iov + i->nr_segs; iov < end; iov++) { 633185ac4d4SAl Viro if (likely(size < iov->iov_len)) 634185ac4d4SAl Viro break; 635185ac4d4SAl Viro size -= iov->iov_len; 636185ac4d4SAl Viro } 637185ac4d4SAl Viro i->iov_offset = size; 638de4f5fedSJens Axboe i->nr_segs -= iov - iter_iov(i); 639de4f5fedSJens Axboe i->__iov = iov; 640185ac4d4SAl Viro } 641185ac4d4SAl Viro 642d879cb83SAl Viro void iov_iter_advance(struct iov_iter *i, size_t size) 643d879cb83SAl Viro { 6443b3fc051SAl Viro if (unlikely(i->count < size)) 6453b3fc051SAl Viro size = i->count; 646fcb14cb1SAl Viro if (likely(iter_is_ubuf(i)) || unlikely(iov_iter_is_xarray(i))) { 647fcb14cb1SAl Viro i->iov_offset += size; 648fcb14cb1SAl Viro i->count -= size; 649fcb14cb1SAl Viro } else if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) { 650185ac4d4SAl Viro /* iovec and kvec have identical layouts */ 651185ac4d4SAl Viro iov_iter_iovec_advance(i, size); 652185ac4d4SAl Viro } else if (iov_iter_is_bvec(i)) { 653185ac4d4SAl Viro iov_iter_bvec_advance(i, size); 654185ac4d4SAl Viro } else if (iov_iter_is_discard(i)) { 655185ac4d4SAl Viro i->count -= size; 6567ff50620SDavid Howells } 657d879cb83SAl Viro } 658d879cb83SAl Viro EXPORT_SYMBOL(iov_iter_advance); 659d879cb83SAl Viro 66027c0e374SAl Viro void iov_iter_revert(struct iov_iter *i, size_t unroll) 66127c0e374SAl Viro { 66227c0e374SAl Viro if (!unroll) 66327c0e374SAl Viro return; 6645b47d59aSAl Viro if (WARN_ON(unroll > MAX_RW_COUNT)) 6655b47d59aSAl Viro return; 66627c0e374SAl Viro i->count += unroll; 6679ea9ce04SDavid Howells if (unlikely(iov_iter_is_discard(i))) 6689ea9ce04SDavid Howells return; 66927c0e374SAl Viro if (unroll <= i->iov_offset) { 67027c0e374SAl Viro i->iov_offset -= unroll; 67127c0e374SAl Viro return; 67227c0e374SAl Viro } 67327c0e374SAl Viro unroll -= i->iov_offset; 674fcb14cb1SAl Viro if (iov_iter_is_xarray(i) || iter_is_ubuf(i)) { 6757ff50620SDavid Howells BUG(); /* We should never go beyond the start of the specified 6767ff50620SDavid Howells * range since we might then be straying into pages that 6777ff50620SDavid Howells * aren't pinned. 6787ff50620SDavid Howells */ 6797ff50620SDavid Howells } else if (iov_iter_is_bvec(i)) { 68027c0e374SAl Viro const struct bio_vec *bvec = i->bvec; 68127c0e374SAl Viro while (1) { 68227c0e374SAl Viro size_t n = (--bvec)->bv_len; 68327c0e374SAl Viro i->nr_segs++; 68427c0e374SAl Viro if (unroll <= n) { 68527c0e374SAl Viro i->bvec = bvec; 68627c0e374SAl Viro i->iov_offset = n - unroll; 68727c0e374SAl Viro return; 68827c0e374SAl Viro } 68927c0e374SAl Viro unroll -= n; 69027c0e374SAl Viro } 69127c0e374SAl Viro } else { /* same logics for iovec and kvec */ 692de4f5fedSJens Axboe const struct iovec *iov = iter_iov(i); 69327c0e374SAl Viro while (1) { 69427c0e374SAl Viro size_t n = (--iov)->iov_len; 69527c0e374SAl Viro i->nr_segs++; 69627c0e374SAl Viro if (unroll <= n) { 697de4f5fedSJens Axboe i->__iov = iov; 69827c0e374SAl Viro i->iov_offset = n - unroll; 69927c0e374SAl Viro return; 70027c0e374SAl Viro } 70127c0e374SAl Viro unroll -= n; 70227c0e374SAl Viro } 70327c0e374SAl Viro } 70427c0e374SAl Viro } 70527c0e374SAl Viro EXPORT_SYMBOL(iov_iter_revert); 70627c0e374SAl Viro 707d879cb83SAl Viro /* 708d879cb83SAl Viro * Return the count of just the current iov_iter segment. 709d879cb83SAl Viro */ 710d879cb83SAl Viro size_t iov_iter_single_seg_count(const struct iov_iter *i) 711d879cb83SAl Viro { 71228f38db7SAl Viro if (i->nr_segs > 1) { 71328f38db7SAl Viro if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) 714de4f5fedSJens Axboe return min(i->count, iter_iov(i)->iov_len - i->iov_offset); 7157ff50620SDavid Howells if (iov_iter_is_bvec(i)) 716d879cb83SAl Viro return min(i->count, i->bvec->bv_len - i->iov_offset); 71728f38db7SAl Viro } 71828f38db7SAl Viro return i->count; 719d879cb83SAl Viro } 720d879cb83SAl Viro EXPORT_SYMBOL(iov_iter_single_seg_count); 721d879cb83SAl Viro 722aa563d7bSDavid Howells void iov_iter_kvec(struct iov_iter *i, unsigned int direction, 723d879cb83SAl Viro const struct kvec *kvec, unsigned long nr_segs, 724d879cb83SAl Viro size_t count) 725d879cb83SAl Viro { 726aa563d7bSDavid Howells WARN_ON(direction & ~(READ | WRITE)); 7278cd54c1cSAl Viro *i = (struct iov_iter){ 7288cd54c1cSAl Viro .iter_type = ITER_KVEC, 729245f0922SKefeng Wang .copy_mc = false, 7308cd54c1cSAl Viro .data_source = direction, 7318cd54c1cSAl Viro .kvec = kvec, 7328cd54c1cSAl Viro .nr_segs = nr_segs, 7338cd54c1cSAl Viro .iov_offset = 0, 7348cd54c1cSAl Viro .count = count 7358cd54c1cSAl Viro }; 736d879cb83SAl Viro } 737d879cb83SAl Viro EXPORT_SYMBOL(iov_iter_kvec); 738d879cb83SAl Viro 739aa563d7bSDavid Howells void iov_iter_bvec(struct iov_iter *i, unsigned int direction, 740d879cb83SAl Viro const struct bio_vec *bvec, unsigned long nr_segs, 741d879cb83SAl Viro size_t count) 742d879cb83SAl Viro { 743aa563d7bSDavid Howells WARN_ON(direction & ~(READ | WRITE)); 7448cd54c1cSAl Viro *i = (struct iov_iter){ 7458cd54c1cSAl Viro .iter_type = ITER_BVEC, 746245f0922SKefeng Wang .copy_mc = false, 7478cd54c1cSAl Viro .data_source = direction, 7488cd54c1cSAl Viro .bvec = bvec, 7498cd54c1cSAl Viro .nr_segs = nr_segs, 7508cd54c1cSAl Viro .iov_offset = 0, 7518cd54c1cSAl Viro .count = count 7528cd54c1cSAl Viro }; 753d879cb83SAl Viro } 754d879cb83SAl Viro EXPORT_SYMBOL(iov_iter_bvec); 755d879cb83SAl Viro 7569ea9ce04SDavid Howells /** 7577ff50620SDavid Howells * iov_iter_xarray - Initialise an I/O iterator to use the pages in an xarray 7587ff50620SDavid Howells * @i: The iterator to initialise. 7597ff50620SDavid Howells * @direction: The direction of the transfer. 7607ff50620SDavid Howells * @xarray: The xarray to access. 7617ff50620SDavid Howells * @start: The start file position. 7627ff50620SDavid Howells * @count: The size of the I/O buffer in bytes. 7637ff50620SDavid Howells * 7647ff50620SDavid Howells * Set up an I/O iterator to either draw data out of the pages attached to an 7657ff50620SDavid Howells * inode or to inject data into those pages. The pages *must* be prevented 7667ff50620SDavid Howells * from evaporation, either by taking a ref on them or locking them by the 7677ff50620SDavid Howells * caller. 7687ff50620SDavid Howells */ 7697ff50620SDavid Howells void iov_iter_xarray(struct iov_iter *i, unsigned int direction, 7707ff50620SDavid Howells struct xarray *xarray, loff_t start, size_t count) 7717ff50620SDavid Howells { 7727ff50620SDavid Howells BUG_ON(direction & ~1); 7738cd54c1cSAl Viro *i = (struct iov_iter) { 7748cd54c1cSAl Viro .iter_type = ITER_XARRAY, 775245f0922SKefeng Wang .copy_mc = false, 7768cd54c1cSAl Viro .data_source = direction, 7778cd54c1cSAl Viro .xarray = xarray, 7788cd54c1cSAl Viro .xarray_start = start, 7798cd54c1cSAl Viro .count = count, 7808cd54c1cSAl Viro .iov_offset = 0 7818cd54c1cSAl Viro }; 7827ff50620SDavid Howells } 7837ff50620SDavid Howells EXPORT_SYMBOL(iov_iter_xarray); 7847ff50620SDavid Howells 7857ff50620SDavid Howells /** 7869ea9ce04SDavid Howells * iov_iter_discard - Initialise an I/O iterator that discards data 7879ea9ce04SDavid Howells * @i: The iterator to initialise. 7889ea9ce04SDavid Howells * @direction: The direction of the transfer. 7899ea9ce04SDavid Howells * @count: The size of the I/O buffer in bytes. 7909ea9ce04SDavid Howells * 7919ea9ce04SDavid Howells * Set up an I/O iterator that just discards everything that's written to it. 7929ea9ce04SDavid Howells * It's only available as a READ iterator. 7939ea9ce04SDavid Howells */ 7949ea9ce04SDavid Howells void iov_iter_discard(struct iov_iter *i, unsigned int direction, size_t count) 7959ea9ce04SDavid Howells { 7969ea9ce04SDavid Howells BUG_ON(direction != READ); 7978cd54c1cSAl Viro *i = (struct iov_iter){ 7988cd54c1cSAl Viro .iter_type = ITER_DISCARD, 799245f0922SKefeng Wang .copy_mc = false, 8008cd54c1cSAl Viro .data_source = false, 8018cd54c1cSAl Viro .count = count, 8028cd54c1cSAl Viro .iov_offset = 0 8038cd54c1cSAl Viro }; 8049ea9ce04SDavid Howells } 8059ea9ce04SDavid Howells EXPORT_SYMBOL(iov_iter_discard); 8069ea9ce04SDavid Howells 807cfa320f7SKeith Busch static bool iov_iter_aligned_iovec(const struct iov_iter *i, unsigned addr_mask, 808cfa320f7SKeith Busch unsigned len_mask) 809cfa320f7SKeith Busch { 810cfa320f7SKeith Busch size_t size = i->count; 811cfa320f7SKeith Busch size_t skip = i->iov_offset; 812cfa320f7SKeith Busch unsigned k; 813cfa320f7SKeith Busch 814cfa320f7SKeith Busch for (k = 0; k < i->nr_segs; k++, skip = 0) { 815de4f5fedSJens Axboe const struct iovec *iov = iter_iov(i) + k; 816de4f5fedSJens Axboe size_t len = iov->iov_len - skip; 817cfa320f7SKeith Busch 818cfa320f7SKeith Busch if (len > size) 819cfa320f7SKeith Busch len = size; 820cfa320f7SKeith Busch if (len & len_mask) 821cfa320f7SKeith Busch return false; 822de4f5fedSJens Axboe if ((unsigned long)(iov->iov_base + skip) & addr_mask) 823cfa320f7SKeith Busch return false; 824cfa320f7SKeith Busch 825cfa320f7SKeith Busch size -= len; 826cfa320f7SKeith Busch if (!size) 827cfa320f7SKeith Busch break; 828cfa320f7SKeith Busch } 829cfa320f7SKeith Busch return true; 830cfa320f7SKeith Busch } 831cfa320f7SKeith Busch 832cfa320f7SKeith Busch static bool iov_iter_aligned_bvec(const struct iov_iter *i, unsigned addr_mask, 833cfa320f7SKeith Busch unsigned len_mask) 834cfa320f7SKeith Busch { 835cfa320f7SKeith Busch size_t size = i->count; 836cfa320f7SKeith Busch unsigned skip = i->iov_offset; 837cfa320f7SKeith Busch unsigned k; 838cfa320f7SKeith Busch 839cfa320f7SKeith Busch for (k = 0; k < i->nr_segs; k++, skip = 0) { 840cfa320f7SKeith Busch size_t len = i->bvec[k].bv_len - skip; 841cfa320f7SKeith Busch 842cfa320f7SKeith Busch if (len > size) 843cfa320f7SKeith Busch len = size; 844cfa320f7SKeith Busch if (len & len_mask) 845cfa320f7SKeith Busch return false; 846cfa320f7SKeith Busch if ((unsigned long)(i->bvec[k].bv_offset + skip) & addr_mask) 847cfa320f7SKeith Busch return false; 848cfa320f7SKeith Busch 849cfa320f7SKeith Busch size -= len; 850cfa320f7SKeith Busch if (!size) 851cfa320f7SKeith Busch break; 852cfa320f7SKeith Busch } 853cfa320f7SKeith Busch return true; 854cfa320f7SKeith Busch } 855cfa320f7SKeith Busch 856cfa320f7SKeith Busch /** 857cfa320f7SKeith Busch * iov_iter_is_aligned() - Check if the addresses and lengths of each segments 858cfa320f7SKeith Busch * are aligned to the parameters. 859cfa320f7SKeith Busch * 860cfa320f7SKeith Busch * @i: &struct iov_iter to restore 861cfa320f7SKeith Busch * @addr_mask: bit mask to check against the iov element's addresses 862cfa320f7SKeith Busch * @len_mask: bit mask to check against the iov element's lengths 863cfa320f7SKeith Busch * 864cfa320f7SKeith Busch * Return: false if any addresses or lengths intersect with the provided masks 865cfa320f7SKeith Busch */ 866cfa320f7SKeith Busch bool iov_iter_is_aligned(const struct iov_iter *i, unsigned addr_mask, 867cfa320f7SKeith Busch unsigned len_mask) 868cfa320f7SKeith Busch { 869fcb14cb1SAl Viro if (likely(iter_is_ubuf(i))) { 870fcb14cb1SAl Viro if (i->count & len_mask) 871fcb14cb1SAl Viro return false; 872fcb14cb1SAl Viro if ((unsigned long)(i->ubuf + i->iov_offset) & addr_mask) 873fcb14cb1SAl Viro return false; 874fcb14cb1SAl Viro return true; 875fcb14cb1SAl Viro } 876fcb14cb1SAl Viro 877cfa320f7SKeith Busch if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) 878cfa320f7SKeith Busch return iov_iter_aligned_iovec(i, addr_mask, len_mask); 879cfa320f7SKeith Busch 880cfa320f7SKeith Busch if (iov_iter_is_bvec(i)) 881cfa320f7SKeith Busch return iov_iter_aligned_bvec(i, addr_mask, len_mask); 882cfa320f7SKeith Busch 883cfa320f7SKeith Busch if (iov_iter_is_xarray(i)) { 884cfa320f7SKeith Busch if (i->count & len_mask) 885cfa320f7SKeith Busch return false; 886cfa320f7SKeith Busch if ((i->xarray_start + i->iov_offset) & addr_mask) 887cfa320f7SKeith Busch return false; 888cfa320f7SKeith Busch } 889cfa320f7SKeith Busch 890cfa320f7SKeith Busch return true; 891cfa320f7SKeith Busch } 892cfa320f7SKeith Busch EXPORT_SYMBOL_GPL(iov_iter_is_aligned); 893cfa320f7SKeith Busch 8949221d2e3SAl Viro static unsigned long iov_iter_alignment_iovec(const struct iov_iter *i) 895d879cb83SAl Viro { 896d879cb83SAl Viro unsigned long res = 0; 897d879cb83SAl Viro size_t size = i->count; 8989221d2e3SAl Viro size_t skip = i->iov_offset; 8999221d2e3SAl Viro unsigned k; 900d879cb83SAl Viro 9019221d2e3SAl Viro for (k = 0; k < i->nr_segs; k++, skip = 0) { 902de4f5fedSJens Axboe const struct iovec *iov = iter_iov(i) + k; 903de4f5fedSJens Axboe size_t len = iov->iov_len - skip; 9049221d2e3SAl Viro if (len) { 905de4f5fedSJens Axboe res |= (unsigned long)iov->iov_base + skip; 9069221d2e3SAl Viro if (len > size) 9079221d2e3SAl Viro len = size; 9089221d2e3SAl Viro res |= len; 9099221d2e3SAl Viro size -= len; 9109221d2e3SAl Viro if (!size) 9119221d2e3SAl Viro break; 9129221d2e3SAl Viro } 9139221d2e3SAl Viro } 9149221d2e3SAl Viro return res; 9159221d2e3SAl Viro } 9169221d2e3SAl Viro 9179221d2e3SAl Viro static unsigned long iov_iter_alignment_bvec(const struct iov_iter *i) 9189221d2e3SAl Viro { 9199221d2e3SAl Viro unsigned res = 0; 9209221d2e3SAl Viro size_t size = i->count; 9219221d2e3SAl Viro unsigned skip = i->iov_offset; 9229221d2e3SAl Viro unsigned k; 9239221d2e3SAl Viro 9249221d2e3SAl Viro for (k = 0; k < i->nr_segs; k++, skip = 0) { 9259221d2e3SAl Viro size_t len = i->bvec[k].bv_len - skip; 9269221d2e3SAl Viro res |= (unsigned long)i->bvec[k].bv_offset + skip; 9279221d2e3SAl Viro if (len > size) 9289221d2e3SAl Viro len = size; 9299221d2e3SAl Viro res |= len; 9309221d2e3SAl Viro size -= len; 9319221d2e3SAl Viro if (!size) 9329221d2e3SAl Viro break; 9339221d2e3SAl Viro } 9349221d2e3SAl Viro return res; 9359221d2e3SAl Viro } 9369221d2e3SAl Viro 9379221d2e3SAl Viro unsigned long iov_iter_alignment(const struct iov_iter *i) 9389221d2e3SAl Viro { 939fcb14cb1SAl Viro if (likely(iter_is_ubuf(i))) { 940fcb14cb1SAl Viro size_t size = i->count; 941fcb14cb1SAl Viro if (size) 942fcb14cb1SAl Viro return ((unsigned long)i->ubuf + i->iov_offset) | size; 943fcb14cb1SAl Viro return 0; 944fcb14cb1SAl Viro } 945fcb14cb1SAl Viro 9469221d2e3SAl Viro /* iovec and kvec have identical layouts */ 9479221d2e3SAl Viro if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) 9489221d2e3SAl Viro return iov_iter_alignment_iovec(i); 9499221d2e3SAl Viro 9509221d2e3SAl Viro if (iov_iter_is_bvec(i)) 9519221d2e3SAl Viro return iov_iter_alignment_bvec(i); 9529221d2e3SAl Viro 9539221d2e3SAl Viro if (iov_iter_is_xarray(i)) 9543d14ec1fSDavid Howells return (i->xarray_start + i->iov_offset) | i->count; 9559221d2e3SAl Viro 9569221d2e3SAl Viro return 0; 957d879cb83SAl Viro } 958d879cb83SAl Viro EXPORT_SYMBOL(iov_iter_alignment); 959d879cb83SAl Viro 960357f435dSAl Viro unsigned long iov_iter_gap_alignment(const struct iov_iter *i) 961357f435dSAl Viro { 962357f435dSAl Viro unsigned long res = 0; 963610c7a71SAl Viro unsigned long v = 0; 964357f435dSAl Viro size_t size = i->count; 965610c7a71SAl Viro unsigned k; 966357f435dSAl Viro 967fcb14cb1SAl Viro if (iter_is_ubuf(i)) 968fcb14cb1SAl Viro return 0; 969fcb14cb1SAl Viro 970610c7a71SAl Viro if (WARN_ON(!iter_is_iovec(i))) 971241699cdSAl Viro return ~0U; 972241699cdSAl Viro 973610c7a71SAl Viro for (k = 0; k < i->nr_segs; k++) { 974de4f5fedSJens Axboe const struct iovec *iov = iter_iov(i) + k; 975de4f5fedSJens Axboe if (iov->iov_len) { 976de4f5fedSJens Axboe unsigned long base = (unsigned long)iov->iov_base; 977610c7a71SAl Viro if (v) // if not the first one 978610c7a71SAl Viro res |= base | v; // this start | previous end 979de4f5fedSJens Axboe v = base + iov->iov_len; 980de4f5fedSJens Axboe if (size <= iov->iov_len) 981610c7a71SAl Viro break; 982de4f5fedSJens Axboe size -= iov->iov_len; 983610c7a71SAl Viro } 984610c7a71SAl Viro } 985357f435dSAl Viro return res; 986357f435dSAl Viro } 987357f435dSAl Viro EXPORT_SYMBOL(iov_iter_gap_alignment); 988357f435dSAl Viro 9893cf42da3SAl Viro static int want_pages_array(struct page ***res, size_t size, 9903cf42da3SAl Viro size_t start, unsigned int maxpages) 991acbdeb83SAl Viro { 9923cf42da3SAl Viro unsigned int count = DIV_ROUND_UP(size + start, PAGE_SIZE); 9933cf42da3SAl Viro 9943cf42da3SAl Viro if (count > maxpages) 9953cf42da3SAl Viro count = maxpages; 9963cf42da3SAl Viro WARN_ON(!count); // caller should've prevented that 9973cf42da3SAl Viro if (!*res) { 9983cf42da3SAl Viro *res = kvmalloc_array(count, sizeof(struct page *), GFP_KERNEL); 9993cf42da3SAl Viro if (!*res) 10003cf42da3SAl Viro return 0; 10013cf42da3SAl Viro } 10023cf42da3SAl Viro return count; 1003acbdeb83SAl Viro } 1004acbdeb83SAl Viro 10057ff50620SDavid Howells static ssize_t iter_xarray_populate_pages(struct page **pages, struct xarray *xa, 10067ff50620SDavid Howells pgoff_t index, unsigned int nr_pages) 10077ff50620SDavid Howells { 10087ff50620SDavid Howells XA_STATE(xas, xa, index); 10097ff50620SDavid Howells struct page *page; 10107ff50620SDavid Howells unsigned int ret = 0; 10117ff50620SDavid Howells 10127ff50620SDavid Howells rcu_read_lock(); 10137ff50620SDavid Howells for (page = xas_load(&xas); page; page = xas_next(&xas)) { 10147ff50620SDavid Howells if (xas_retry(&xas, page)) 10157ff50620SDavid Howells continue; 10167ff50620SDavid Howells 10177ff50620SDavid Howells /* Has the page moved or been split? */ 10187ff50620SDavid Howells if (unlikely(page != xas_reload(&xas))) { 10197ff50620SDavid Howells xas_reset(&xas); 10207ff50620SDavid Howells continue; 10217ff50620SDavid Howells } 10227ff50620SDavid Howells 10237ff50620SDavid Howells pages[ret] = find_subpage(page, xas.xa_index); 10247ff50620SDavid Howells get_page(pages[ret]); 10257ff50620SDavid Howells if (++ret == nr_pages) 10267ff50620SDavid Howells break; 10277ff50620SDavid Howells } 10287ff50620SDavid Howells rcu_read_unlock(); 10297ff50620SDavid Howells return ret; 10307ff50620SDavid Howells } 10317ff50620SDavid Howells 10327ff50620SDavid Howells static ssize_t iter_xarray_get_pages(struct iov_iter *i, 103368fe506fSAl Viro struct page ***pages, size_t maxsize, 10347ff50620SDavid Howells unsigned maxpages, size_t *_start_offset) 10357ff50620SDavid Howells { 10363cf42da3SAl Viro unsigned nr, offset, count; 10373cf42da3SAl Viro pgoff_t index; 10387ff50620SDavid Howells loff_t pos; 10397ff50620SDavid Howells 10407ff50620SDavid Howells pos = i->xarray_start + i->iov_offset; 10417ff50620SDavid Howells index = pos >> PAGE_SHIFT; 10427ff50620SDavid Howells offset = pos & ~PAGE_MASK; 10437ff50620SDavid Howells *_start_offset = offset; 10447ff50620SDavid Howells 10453cf42da3SAl Viro count = want_pages_array(pages, maxsize, offset, maxpages); 10463cf42da3SAl Viro if (!count) 104768fe506fSAl Viro return -ENOMEM; 104868fe506fSAl Viro nr = iter_xarray_populate_pages(*pages, i->xarray, index, count); 10497ff50620SDavid Howells if (nr == 0) 10507ff50620SDavid Howells return 0; 10517ff50620SDavid Howells 1052eba2d3d7SAl Viro maxsize = min_t(size_t, nr * PAGE_SIZE - offset, maxsize); 1053310d9d5aSAl Viro i->iov_offset += maxsize; 1054310d9d5aSAl Viro i->count -= maxsize; 1055eba2d3d7SAl Viro return maxsize; 10567ff50620SDavid Howells } 10577ff50620SDavid Howells 1058fcb14cb1SAl Viro /* must be done on non-empty ITER_UBUF or ITER_IOVEC one */ 1059dd45ab9dSAl Viro static unsigned long first_iovec_segment(const struct iov_iter *i, size_t *size) 10603d671ca6SAl Viro { 10613d671ca6SAl Viro size_t skip; 10623d671ca6SAl Viro long k; 10633d671ca6SAl Viro 1064fcb14cb1SAl Viro if (iter_is_ubuf(i)) 1065fcb14cb1SAl Viro return (unsigned long)i->ubuf + i->iov_offset; 1066fcb14cb1SAl Viro 10673d671ca6SAl Viro for (k = 0, skip = i->iov_offset; k < i->nr_segs; k++, skip = 0) { 1068de4f5fedSJens Axboe const struct iovec *iov = iter_iov(i) + k; 1069de4f5fedSJens Axboe size_t len = iov->iov_len - skip; 10703d671ca6SAl Viro 10713d671ca6SAl Viro if (unlikely(!len)) 10723d671ca6SAl Viro continue; 107359dbd7d0SAl Viro if (*size > len) 10743d671ca6SAl Viro *size = len; 1075de4f5fedSJens Axboe return (unsigned long)iov->iov_base + skip; 10763d671ca6SAl Viro } 10773d671ca6SAl Viro BUG(); // if it had been empty, we wouldn't get called 10783d671ca6SAl Viro } 10793d671ca6SAl Viro 10803d671ca6SAl Viro /* must be done on non-empty ITER_BVEC one */ 10813d671ca6SAl Viro static struct page *first_bvec_segment(const struct iov_iter *i, 108259dbd7d0SAl Viro size_t *size, size_t *start) 10833d671ca6SAl Viro { 10843d671ca6SAl Viro struct page *page; 10853d671ca6SAl Viro size_t skip = i->iov_offset, len; 10863d671ca6SAl Viro 10873d671ca6SAl Viro len = i->bvec->bv_len - skip; 108859dbd7d0SAl Viro if (*size > len) 108959dbd7d0SAl Viro *size = len; 10903d671ca6SAl Viro skip += i->bvec->bv_offset; 10913d671ca6SAl Viro page = i->bvec->bv_page + skip / PAGE_SIZE; 1092dda8e5d1SAl Viro *start = skip % PAGE_SIZE; 10933d671ca6SAl Viro return page; 10943d671ca6SAl Viro } 10953d671ca6SAl Viro 109691329559SAl Viro static ssize_t __iov_iter_get_pages_alloc(struct iov_iter *i, 1097d879cb83SAl Viro struct page ***pages, size_t maxsize, 109884bd06c6SChristoph Hellwig unsigned int maxpages, size_t *start) 1099d879cb83SAl Viro { 1100f62e52d1SDavid Howells unsigned int n, gup_flags = 0; 1101d879cb83SAl Viro 1102d879cb83SAl Viro if (maxsize > i->count) 1103d879cb83SAl Viro maxsize = i->count; 11043d671ca6SAl Viro if (!maxsize) 11053d671ca6SAl Viro return 0; 11067392ed17SAl Viro if (maxsize > MAX_RW_COUNT) 11077392ed17SAl Viro maxsize = MAX_RW_COUNT; 1108d879cb83SAl Viro 1109fcb14cb1SAl Viro if (likely(user_backed_iter(i))) { 11103d671ca6SAl Viro unsigned long addr; 11113cf42da3SAl Viro int res; 11129ea9ce04SDavid Howells 11133337ab08SAndreas Gruenbacher if (iov_iter_rw(i) != WRITE) 11143337ab08SAndreas Gruenbacher gup_flags |= FOLL_WRITE; 11153337ab08SAndreas Gruenbacher if (i->nofault) 11163337ab08SAndreas Gruenbacher gup_flags |= FOLL_NOFAULT; 11173337ab08SAndreas Gruenbacher 1118dd45ab9dSAl Viro addr = first_iovec_segment(i, &maxsize); 1119dd45ab9dSAl Viro *start = addr % PAGE_SIZE; 1120dd45ab9dSAl Viro addr &= PAGE_MASK; 11213cf42da3SAl Viro n = want_pages_array(pages, maxsize, *start, maxpages); 11223cf42da3SAl Viro if (!n) 1123d879cb83SAl Viro return -ENOMEM; 1124451c0ba9SAl Viro res = get_user_pages_fast(addr, n, gup_flags, *pages); 112591329559SAl Viro if (unlikely(res <= 0)) 1126d879cb83SAl Viro return res; 1127eba2d3d7SAl Viro maxsize = min_t(size_t, maxsize, res * PAGE_SIZE - *start); 1128eba2d3d7SAl Viro iov_iter_advance(i, maxsize); 1129eba2d3d7SAl Viro return maxsize; 11303d671ca6SAl Viro } 11313d671ca6SAl Viro if (iov_iter_is_bvec(i)) { 1132451c0ba9SAl Viro struct page **p; 11333d671ca6SAl Viro struct page *page; 11343d671ca6SAl Viro 113559dbd7d0SAl Viro page = first_bvec_segment(i, &maxsize, start); 11363cf42da3SAl Viro n = want_pages_array(pages, maxsize, *start, maxpages); 11373cf42da3SAl Viro if (!n) 1138d879cb83SAl Viro return -ENOMEM; 11393cf42da3SAl Viro p = *pages; 1140dda8e5d1SAl Viro for (int k = 0; k < n; k++) 1141eba2d3d7SAl Viro get_page(p[k] = page + k); 1142eba2d3d7SAl Viro maxsize = min_t(size_t, maxsize, n * PAGE_SIZE - *start); 1143310d9d5aSAl Viro i->count -= maxsize; 1144310d9d5aSAl Viro i->iov_offset += maxsize; 1145310d9d5aSAl Viro if (i->iov_offset == i->bvec->bv_len) { 1146310d9d5aSAl Viro i->iov_offset = 0; 1147310d9d5aSAl Viro i->bvec++; 1148310d9d5aSAl Viro i->nr_segs--; 1149310d9d5aSAl Viro } 1150eba2d3d7SAl Viro return maxsize; 11513d671ca6SAl Viro } 11523d671ca6SAl Viro if (iov_iter_is_xarray(i)) 1153451c0ba9SAl Viro return iter_xarray_get_pages(i, pages, maxsize, maxpages, start); 1154d879cb83SAl Viro return -EFAULT; 1155d879cb83SAl Viro } 115691329559SAl Viro 115784bd06c6SChristoph Hellwig ssize_t iov_iter_get_pages2(struct iov_iter *i, struct page **pages, 115884bd06c6SChristoph Hellwig size_t maxsize, unsigned maxpages, size_t *start) 1159451c0ba9SAl Viro { 1160451c0ba9SAl Viro if (!maxpages) 1161451c0ba9SAl Viro return 0; 1162451c0ba9SAl Viro BUG_ON(!pages); 1163451c0ba9SAl Viro 116484bd06c6SChristoph Hellwig return __iov_iter_get_pages_alloc(i, &pages, maxsize, maxpages, start); 1165451c0ba9SAl Viro } 1166eba2d3d7SAl Viro EXPORT_SYMBOL(iov_iter_get_pages2); 1167451c0ba9SAl Viro 116884bd06c6SChristoph Hellwig ssize_t iov_iter_get_pages_alloc2(struct iov_iter *i, 116984bd06c6SChristoph Hellwig struct page ***pages, size_t maxsize, size_t *start) 117091329559SAl Viro { 117191329559SAl Viro ssize_t len; 117291329559SAl Viro 117391329559SAl Viro *pages = NULL; 117491329559SAl Viro 117584bd06c6SChristoph Hellwig len = __iov_iter_get_pages_alloc(i, pages, maxsize, ~0U, start); 117691329559SAl Viro if (len <= 0) { 117791329559SAl Viro kvfree(*pages); 117891329559SAl Viro *pages = NULL; 117991329559SAl Viro } 118091329559SAl Viro return len; 118191329559SAl Viro } 1182eba2d3d7SAl Viro EXPORT_SYMBOL(iov_iter_get_pages_alloc2); 1183d879cb83SAl Viro 1184d879cb83SAl Viro size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum, 1185d879cb83SAl Viro struct iov_iter *i) 1186d879cb83SAl Viro { 1187d879cb83SAl Viro __wsum sum, next; 1188d879cb83SAl Viro sum = *csum; 1189a41dad90SAl Viro if (WARN_ON_ONCE(!i->data_source)) 1190241699cdSAl Viro return 0; 1191a41dad90SAl Viro 11927baa5099SAl Viro iterate_and_advance(i, bytes, base, len, off, ({ 11937baa5099SAl Viro next = csum_and_copy_from_user(base, addr + off, len); 1194d879cb83SAl Viro sum = csum_block_add(sum, next, off); 11957baa5099SAl Viro next ? 0 : len; 1196d879cb83SAl Viro }), ({ 11977baa5099SAl Viro sum = csum_and_memcpy(addr + off, base, len, sum, off); 1198d879cb83SAl Viro }) 1199d879cb83SAl Viro ) 1200d879cb83SAl Viro *csum = sum; 1201d879cb83SAl Viro return bytes; 1202d879cb83SAl Viro } 1203d879cb83SAl Viro EXPORT_SYMBOL(csum_and_copy_from_iter); 1204d879cb83SAl Viro 120552cbd23aSWillem de Bruijn size_t csum_and_copy_to_iter(const void *addr, size_t bytes, void *_csstate, 1206d879cb83SAl Viro struct iov_iter *i) 1207d879cb83SAl Viro { 120852cbd23aSWillem de Bruijn struct csum_state *csstate = _csstate; 1209d879cb83SAl Viro __wsum sum, next; 121078e1f386SAl Viro 1211a41dad90SAl Viro if (WARN_ON_ONCE(i->data_source)) 1212a41dad90SAl Viro return 0; 121378e1f386SAl Viro if (unlikely(iov_iter_is_discard(i))) { 1214c67f1fd2SAl Viro // can't use csum_memcpy() for that one - data is not copied 1215c67f1fd2SAl Viro csstate->csum = csum_block_add(csstate->csum, 1216c67f1fd2SAl Viro csum_partial(addr, bytes, 0), 1217c67f1fd2SAl Viro csstate->off); 1218c67f1fd2SAl Viro csstate->off += bytes; 1219c67f1fd2SAl Viro return bytes; 1220241699cdSAl Viro } 12216852df12SAl Viro 12226852df12SAl Viro sum = csum_shift(csstate->csum, csstate->off); 12233fc40265SDavid Howells iterate_and_advance(i, bytes, base, len, off, ({ 12247baa5099SAl Viro next = csum_and_copy_to_user(addr + off, base, len); 1225d879cb83SAl Viro sum = csum_block_add(sum, next, off); 12267baa5099SAl Viro next ? 0 : len; 1227d879cb83SAl Viro }), ({ 12287baa5099SAl Viro sum = csum_and_memcpy(base, addr + off, len, sum, off); 1229d879cb83SAl Viro }) 1230d879cb83SAl Viro ) 1231594e450bSAl Viro csstate->csum = csum_shift(sum, csstate->off); 1232594e450bSAl Viro csstate->off += bytes; 1233d879cb83SAl Viro return bytes; 1234d879cb83SAl Viro } 1235d879cb83SAl Viro EXPORT_SYMBOL(csum_and_copy_to_iter); 1236d879cb83SAl Viro 1237d05f4435SSagi Grimberg size_t hash_and_copy_to_iter(const void *addr, size_t bytes, void *hashp, 1238d05f4435SSagi Grimberg struct iov_iter *i) 1239d05f4435SSagi Grimberg { 12407999096fSHerbert Xu #ifdef CONFIG_CRYPTO_HASH 1241d05f4435SSagi Grimberg struct ahash_request *hash = hashp; 1242d05f4435SSagi Grimberg struct scatterlist sg; 1243d05f4435SSagi Grimberg size_t copied; 1244d05f4435SSagi Grimberg 1245d05f4435SSagi Grimberg copied = copy_to_iter(addr, bytes, i); 1246d05f4435SSagi Grimberg sg_init_one(&sg, addr, copied); 1247d05f4435SSagi Grimberg ahash_request_set_crypt(hash, &sg, NULL, copied); 1248d05f4435SSagi Grimberg crypto_ahash_update(hash); 1249d05f4435SSagi Grimberg return copied; 125027fad74aSYueHaibing #else 125127fad74aSYueHaibing return 0; 125227fad74aSYueHaibing #endif 1253d05f4435SSagi Grimberg } 1254d05f4435SSagi Grimberg EXPORT_SYMBOL(hash_and_copy_to_iter); 1255d05f4435SSagi Grimberg 125666531c65SAl Viro static int iov_npages(const struct iov_iter *i, int maxpages) 1257d879cb83SAl Viro { 125866531c65SAl Viro size_t skip = i->iov_offset, size = i->count; 125966531c65SAl Viro const struct iovec *p; 1260d879cb83SAl Viro int npages = 0; 1261d879cb83SAl Viro 1262de4f5fedSJens Axboe for (p = iter_iov(i); size; skip = 0, p++) { 126366531c65SAl Viro unsigned offs = offset_in_page(p->iov_base + skip); 126466531c65SAl Viro size_t len = min(p->iov_len - skip, size); 1265d879cb83SAl Viro 126666531c65SAl Viro if (len) { 126766531c65SAl Viro size -= len; 126866531c65SAl Viro npages += DIV_ROUND_UP(offs + len, PAGE_SIZE); 126966531c65SAl Viro if (unlikely(npages > maxpages)) 127066531c65SAl Viro return maxpages; 127166531c65SAl Viro } 127266531c65SAl Viro } 127366531c65SAl Viro return npages; 127466531c65SAl Viro } 127566531c65SAl Viro 127666531c65SAl Viro static int bvec_npages(const struct iov_iter *i, int maxpages) 127766531c65SAl Viro { 127866531c65SAl Viro size_t skip = i->iov_offset, size = i->count; 127966531c65SAl Viro const struct bio_vec *p; 128066531c65SAl Viro int npages = 0; 128166531c65SAl Viro 128266531c65SAl Viro for (p = i->bvec; size; skip = 0, p++) { 128366531c65SAl Viro unsigned offs = (p->bv_offset + skip) % PAGE_SIZE; 128466531c65SAl Viro size_t len = min(p->bv_len - skip, size); 128566531c65SAl Viro 128666531c65SAl Viro size -= len; 128766531c65SAl Viro npages += DIV_ROUND_UP(offs + len, PAGE_SIZE); 128866531c65SAl Viro if (unlikely(npages > maxpages)) 128966531c65SAl Viro return maxpages; 129066531c65SAl Viro } 129166531c65SAl Viro return npages; 129266531c65SAl Viro } 129366531c65SAl Viro 129466531c65SAl Viro int iov_iter_npages(const struct iov_iter *i, int maxpages) 129566531c65SAl Viro { 129666531c65SAl Viro if (unlikely(!i->count)) 129766531c65SAl Viro return 0; 1298fcb14cb1SAl Viro if (likely(iter_is_ubuf(i))) { 1299fcb14cb1SAl Viro unsigned offs = offset_in_page(i->ubuf + i->iov_offset); 1300fcb14cb1SAl Viro int npages = DIV_ROUND_UP(offs + i->count, PAGE_SIZE); 1301fcb14cb1SAl Viro return min(npages, maxpages); 1302fcb14cb1SAl Viro } 130366531c65SAl Viro /* iovec and kvec have identical layouts */ 130466531c65SAl Viro if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) 130566531c65SAl Viro return iov_npages(i, maxpages); 130666531c65SAl Viro if (iov_iter_is_bvec(i)) 130766531c65SAl Viro return bvec_npages(i, maxpages); 130866531c65SAl Viro if (iov_iter_is_xarray(i)) { 1309e4f8df86SAl Viro unsigned offset = (i->xarray_start + i->iov_offset) % PAGE_SIZE; 1310e4f8df86SAl Viro int npages = DIV_ROUND_UP(offset + i->count, PAGE_SIZE); 131166531c65SAl Viro return min(npages, maxpages); 131266531c65SAl Viro } 131366531c65SAl Viro return 0; 1314d879cb83SAl Viro } 1315d879cb83SAl Viro EXPORT_SYMBOL(iov_iter_npages); 1316d879cb83SAl Viro 1317d879cb83SAl Viro const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags) 1318d879cb83SAl Viro { 1319d879cb83SAl Viro *new = *old; 132000e23707SDavid Howells if (iov_iter_is_bvec(new)) 1321d879cb83SAl Viro return new->bvec = kmemdup(new->bvec, 1322d879cb83SAl Viro new->nr_segs * sizeof(struct bio_vec), 1323d879cb83SAl Viro flags); 1324fcb14cb1SAl Viro else if (iov_iter_is_kvec(new) || iter_is_iovec(new)) 1325d879cb83SAl Viro /* iovec and kvec have identical layout */ 1326de4f5fedSJens Axboe return new->__iov = kmemdup(new->__iov, 1327d879cb83SAl Viro new->nr_segs * sizeof(struct iovec), 1328d879cb83SAl Viro flags); 1329fcb14cb1SAl Viro return NULL; 1330d879cb83SAl Viro } 1331d879cb83SAl Viro EXPORT_SYMBOL(dup_iter); 1332bc917be8SAl Viro 133350f9a76eSJosh Poimboeuf static __noclone int copy_compat_iovec_from_user(struct iovec *iov, 1334bfdc5970SChristoph Hellwig const struct iovec __user *uvec, unsigned long nr_segs) 1335bfdc5970SChristoph Hellwig { 1336bfdc5970SChristoph Hellwig const struct compat_iovec __user *uiov = 1337bfdc5970SChristoph Hellwig (const struct compat_iovec __user *)uvec; 1338bfdc5970SChristoph Hellwig int ret = -EFAULT, i; 1339bfdc5970SChristoph Hellwig 1340a959a978SChristoph Hellwig if (!user_access_begin(uiov, nr_segs * sizeof(*uiov))) 1341bfdc5970SChristoph Hellwig return -EFAULT; 1342bfdc5970SChristoph Hellwig 1343bfdc5970SChristoph Hellwig for (i = 0; i < nr_segs; i++) { 1344bfdc5970SChristoph Hellwig compat_uptr_t buf; 1345bfdc5970SChristoph Hellwig compat_ssize_t len; 1346bfdc5970SChristoph Hellwig 1347bfdc5970SChristoph Hellwig unsafe_get_user(len, &uiov[i].iov_len, uaccess_end); 1348bfdc5970SChristoph Hellwig unsafe_get_user(buf, &uiov[i].iov_base, uaccess_end); 1349bfdc5970SChristoph Hellwig 1350bfdc5970SChristoph Hellwig /* check for compat_size_t not fitting in compat_ssize_t .. */ 1351bfdc5970SChristoph Hellwig if (len < 0) { 1352bfdc5970SChristoph Hellwig ret = -EINVAL; 1353bfdc5970SChristoph Hellwig goto uaccess_end; 1354bfdc5970SChristoph Hellwig } 1355bfdc5970SChristoph Hellwig iov[i].iov_base = compat_ptr(buf); 1356bfdc5970SChristoph Hellwig iov[i].iov_len = len; 1357bfdc5970SChristoph Hellwig } 1358bfdc5970SChristoph Hellwig 1359bfdc5970SChristoph Hellwig ret = 0; 1360bfdc5970SChristoph Hellwig uaccess_end: 1361bfdc5970SChristoph Hellwig user_access_end(); 1362bfdc5970SChristoph Hellwig return ret; 1363bfdc5970SChristoph Hellwig } 1364bfdc5970SChristoph Hellwig 1365719a937bSPeter Zijlstra static __noclone int copy_iovec_from_user(struct iovec *iov, 1366487c20b0SLinus Torvalds const struct iovec __user *uiov, unsigned long nr_segs) 1367fb041b59SDavid Laight { 1368487c20b0SLinus Torvalds int ret = -EFAULT; 1369bfdc5970SChristoph Hellwig 1370487c20b0SLinus Torvalds if (!user_access_begin(uiov, nr_segs * sizeof(*uiov))) 1371bfdc5970SChristoph Hellwig return -EFAULT; 1372bfdc5970SChristoph Hellwig 1373487c20b0SLinus Torvalds do { 1374487c20b0SLinus Torvalds void __user *buf; 1375487c20b0SLinus Torvalds ssize_t len; 1376487c20b0SLinus Torvalds 1377487c20b0SLinus Torvalds unsafe_get_user(len, &uiov->iov_len, uaccess_end); 1378487c20b0SLinus Torvalds unsafe_get_user(buf, &uiov->iov_base, uaccess_end); 1379487c20b0SLinus Torvalds 1380487c20b0SLinus Torvalds /* check for size_t not fitting in ssize_t .. */ 1381487c20b0SLinus Torvalds if (unlikely(len < 0)) { 1382487c20b0SLinus Torvalds ret = -EINVAL; 1383487c20b0SLinus Torvalds goto uaccess_end; 1384487c20b0SLinus Torvalds } 1385487c20b0SLinus Torvalds iov->iov_base = buf; 1386487c20b0SLinus Torvalds iov->iov_len = len; 1387487c20b0SLinus Torvalds 1388487c20b0SLinus Torvalds uiov++; iov++; 1389487c20b0SLinus Torvalds } while (--nr_segs); 1390487c20b0SLinus Torvalds 1391487c20b0SLinus Torvalds ret = 0; 1392487c20b0SLinus Torvalds uaccess_end: 1393487c20b0SLinus Torvalds user_access_end(); 1394487c20b0SLinus Torvalds return ret; 1395bfdc5970SChristoph Hellwig } 1396bfdc5970SChristoph Hellwig 1397bfdc5970SChristoph Hellwig struct iovec *iovec_from_user(const struct iovec __user *uvec, 1398bfdc5970SChristoph Hellwig unsigned long nr_segs, unsigned long fast_segs, 1399bfdc5970SChristoph Hellwig struct iovec *fast_iov, bool compat) 1400bfdc5970SChristoph Hellwig { 1401bfdc5970SChristoph Hellwig struct iovec *iov = fast_iov; 1402bfdc5970SChristoph Hellwig int ret; 1403fb041b59SDavid Laight 1404fb041b59SDavid Laight /* 1405bfdc5970SChristoph Hellwig * SuS says "The readv() function *may* fail if the iovcnt argument was 1406bfdc5970SChristoph Hellwig * less than or equal to 0, or greater than {IOV_MAX}. Linux has 1407fb041b59SDavid Laight * traditionally returned zero for zero segments, so... 1408fb041b59SDavid Laight */ 1409bfdc5970SChristoph Hellwig if (nr_segs == 0) 1410bfdc5970SChristoph Hellwig return iov; 1411bfdc5970SChristoph Hellwig if (nr_segs > UIO_MAXIOV) 1412bfdc5970SChristoph Hellwig return ERR_PTR(-EINVAL); 1413fb041b59SDavid Laight if (nr_segs > fast_segs) { 1414fb041b59SDavid Laight iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL); 1415bfdc5970SChristoph Hellwig if (!iov) 1416bfdc5970SChristoph Hellwig return ERR_PTR(-ENOMEM); 1417fb041b59SDavid Laight } 1418bfdc5970SChristoph Hellwig 1419487c20b0SLinus Torvalds if (unlikely(compat)) 1420bfdc5970SChristoph Hellwig ret = copy_compat_iovec_from_user(iov, uvec, nr_segs); 1421bfdc5970SChristoph Hellwig else 1422bfdc5970SChristoph Hellwig ret = copy_iovec_from_user(iov, uvec, nr_segs); 1423bfdc5970SChristoph Hellwig if (ret) { 1424bfdc5970SChristoph Hellwig if (iov != fast_iov) 1425bfdc5970SChristoph Hellwig kfree(iov); 1426bfdc5970SChristoph Hellwig return ERR_PTR(ret); 1427fb041b59SDavid Laight } 1428bfdc5970SChristoph Hellwig 1429bfdc5970SChristoph Hellwig return iov; 1430bfdc5970SChristoph Hellwig } 1431bfdc5970SChristoph Hellwig 14323b2deb0eSJens Axboe /* 14333b2deb0eSJens Axboe * Single segment iovec supplied by the user, import it as ITER_UBUF. 14343b2deb0eSJens Axboe */ 14353b2deb0eSJens Axboe static ssize_t __import_iovec_ubuf(int type, const struct iovec __user *uvec, 14363b2deb0eSJens Axboe struct iovec **iovp, struct iov_iter *i, 14373b2deb0eSJens Axboe bool compat) 14383b2deb0eSJens Axboe { 14393b2deb0eSJens Axboe struct iovec *iov = *iovp; 14403b2deb0eSJens Axboe ssize_t ret; 14413b2deb0eSJens Axboe 14423b2deb0eSJens Axboe if (compat) 14433b2deb0eSJens Axboe ret = copy_compat_iovec_from_user(iov, uvec, 1); 14443b2deb0eSJens Axboe else 14453b2deb0eSJens Axboe ret = copy_iovec_from_user(iov, uvec, 1); 14463b2deb0eSJens Axboe if (unlikely(ret)) 14473b2deb0eSJens Axboe return ret; 14483b2deb0eSJens Axboe 14493b2deb0eSJens Axboe ret = import_ubuf(type, iov->iov_base, iov->iov_len, i); 14503b2deb0eSJens Axboe if (unlikely(ret)) 14513b2deb0eSJens Axboe return ret; 14523b2deb0eSJens Axboe *iovp = NULL; 14533b2deb0eSJens Axboe return i->count; 14543b2deb0eSJens Axboe } 14553b2deb0eSJens Axboe 1456bfdc5970SChristoph Hellwig ssize_t __import_iovec(int type, const struct iovec __user *uvec, 1457bfdc5970SChristoph Hellwig unsigned nr_segs, unsigned fast_segs, struct iovec **iovp, 1458bfdc5970SChristoph Hellwig struct iov_iter *i, bool compat) 1459bfdc5970SChristoph Hellwig { 1460bfdc5970SChristoph Hellwig ssize_t total_len = 0; 1461bfdc5970SChristoph Hellwig unsigned long seg; 1462bfdc5970SChristoph Hellwig struct iovec *iov; 1463bfdc5970SChristoph Hellwig 14643b2deb0eSJens Axboe if (nr_segs == 1) 14653b2deb0eSJens Axboe return __import_iovec_ubuf(type, uvec, iovp, i, compat); 14663b2deb0eSJens Axboe 1467bfdc5970SChristoph Hellwig iov = iovec_from_user(uvec, nr_segs, fast_segs, *iovp, compat); 1468bfdc5970SChristoph Hellwig if (IS_ERR(iov)) { 1469bfdc5970SChristoph Hellwig *iovp = NULL; 1470bfdc5970SChristoph Hellwig return PTR_ERR(iov); 1471fb041b59SDavid Laight } 1472fb041b59SDavid Laight 1473fb041b59SDavid Laight /* 1474bfdc5970SChristoph Hellwig * According to the Single Unix Specification we should return EINVAL if 1475bfdc5970SChristoph Hellwig * an element length is < 0 when cast to ssize_t or if the total length 1476bfdc5970SChristoph Hellwig * would overflow the ssize_t return value of the system call. 1477fb041b59SDavid Laight * 1478fb041b59SDavid Laight * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the 1479fb041b59SDavid Laight * overflow case. 1480fb041b59SDavid Laight */ 1481fb041b59SDavid Laight for (seg = 0; seg < nr_segs; seg++) { 1482fb041b59SDavid Laight ssize_t len = (ssize_t)iov[seg].iov_len; 1483fb041b59SDavid Laight 1484bfdc5970SChristoph Hellwig if (!access_ok(iov[seg].iov_base, len)) { 1485bfdc5970SChristoph Hellwig if (iov != *iovp) 1486bfdc5970SChristoph Hellwig kfree(iov); 1487bfdc5970SChristoph Hellwig *iovp = NULL; 1488bfdc5970SChristoph Hellwig return -EFAULT; 1489fb041b59SDavid Laight } 1490bfdc5970SChristoph Hellwig 1491bfdc5970SChristoph Hellwig if (len > MAX_RW_COUNT - total_len) { 1492bfdc5970SChristoph Hellwig len = MAX_RW_COUNT - total_len; 1493fb041b59SDavid Laight iov[seg].iov_len = len; 1494fb041b59SDavid Laight } 1495bfdc5970SChristoph Hellwig total_len += len; 1496fb041b59SDavid Laight } 1497bfdc5970SChristoph Hellwig 1498bfdc5970SChristoph Hellwig iov_iter_init(i, type, iov, nr_segs, total_len); 1499bfdc5970SChristoph Hellwig if (iov == *iovp) 1500bfdc5970SChristoph Hellwig *iovp = NULL; 1501bfdc5970SChristoph Hellwig else 1502bfdc5970SChristoph Hellwig *iovp = iov; 1503bfdc5970SChristoph Hellwig return total_len; 1504fb041b59SDavid Laight } 1505fb041b59SDavid Laight 1506ffecee4fSVegard Nossum /** 1507ffecee4fSVegard Nossum * import_iovec() - Copy an array of &struct iovec from userspace 1508ffecee4fSVegard Nossum * into the kernel, check that it is valid, and initialize a new 1509ffecee4fSVegard Nossum * &struct iov_iter iterator to access it. 1510ffecee4fSVegard Nossum * 1511ffecee4fSVegard Nossum * @type: One of %READ or %WRITE. 1512bfdc5970SChristoph Hellwig * @uvec: Pointer to the userspace array. 1513ffecee4fSVegard Nossum * @nr_segs: Number of elements in userspace array. 1514ffecee4fSVegard Nossum * @fast_segs: Number of elements in @iov. 1515bfdc5970SChristoph Hellwig * @iovp: (input and output parameter) Pointer to pointer to (usually small 1516ffecee4fSVegard Nossum * on-stack) kernel array. 1517ffecee4fSVegard Nossum * @i: Pointer to iterator that will be initialized on success. 1518ffecee4fSVegard Nossum * 1519ffecee4fSVegard Nossum * If the array pointed to by *@iov is large enough to hold all @nr_segs, 1520ffecee4fSVegard Nossum * then this function places %NULL in *@iov on return. Otherwise, a new 1521ffecee4fSVegard Nossum * array will be allocated and the result placed in *@iov. This means that 1522ffecee4fSVegard Nossum * the caller may call kfree() on *@iov regardless of whether the small 1523ffecee4fSVegard Nossum * on-stack array was used or not (and regardless of whether this function 1524ffecee4fSVegard Nossum * returns an error or not). 1525ffecee4fSVegard Nossum * 152687e5e6daSJens Axboe * Return: Negative error code on error, bytes imported on success 1527ffecee4fSVegard Nossum */ 1528bfdc5970SChristoph Hellwig ssize_t import_iovec(int type, const struct iovec __user *uvec, 1529bc917be8SAl Viro unsigned nr_segs, unsigned fast_segs, 1530bfdc5970SChristoph Hellwig struct iovec **iovp, struct iov_iter *i) 1531bc917be8SAl Viro { 153289cd35c5SChristoph Hellwig return __import_iovec(type, uvec, nr_segs, fast_segs, iovp, i, 153389cd35c5SChristoph Hellwig in_compat_syscall()); 1534bc917be8SAl Viro } 1535bc917be8SAl Viro EXPORT_SYMBOL(import_iovec); 1536bc917be8SAl Viro 1537bc917be8SAl Viro int import_single_range(int rw, void __user *buf, size_t len, 1538bc917be8SAl Viro struct iovec *iov, struct iov_iter *i) 1539bc917be8SAl Viro { 1540bc917be8SAl Viro if (len > MAX_RW_COUNT) 1541bc917be8SAl Viro len = MAX_RW_COUNT; 154296d4f267SLinus Torvalds if (unlikely(!access_ok(buf, len))) 1543bc917be8SAl Viro return -EFAULT; 1544bc917be8SAl Viro 1545e03ad4eeSJens Axboe iov_iter_ubuf(i, rw, buf, len); 1546bc917be8SAl Viro return 0; 1547bc917be8SAl Viro } 1548e1267585SAl Viro EXPORT_SYMBOL(import_single_range); 15498fb0f47aSJens Axboe 15502ad9bd83SJens Axboe int import_ubuf(int rw, void __user *buf, size_t len, struct iov_iter *i) 15512ad9bd83SJens Axboe { 15522ad9bd83SJens Axboe if (len > MAX_RW_COUNT) 15532ad9bd83SJens Axboe len = MAX_RW_COUNT; 15542ad9bd83SJens Axboe if (unlikely(!access_ok(buf, len))) 15552ad9bd83SJens Axboe return -EFAULT; 15562ad9bd83SJens Axboe 15572ad9bd83SJens Axboe iov_iter_ubuf(i, rw, buf, len); 15582ad9bd83SJens Axboe return 0; 15592ad9bd83SJens Axboe } 15602ad9bd83SJens Axboe 15618fb0f47aSJens Axboe /** 15628fb0f47aSJens Axboe * iov_iter_restore() - Restore a &struct iov_iter to the same state as when 15638fb0f47aSJens Axboe * iov_iter_save_state() was called. 15648fb0f47aSJens Axboe * 15658fb0f47aSJens Axboe * @i: &struct iov_iter to restore 15668fb0f47aSJens Axboe * @state: state to restore from 15678fb0f47aSJens Axboe * 15688fb0f47aSJens Axboe * Used after iov_iter_save_state() to bring restore @i, if operations may 15698fb0f47aSJens Axboe * have advanced it. 15708fb0f47aSJens Axboe * 15718fb0f47aSJens Axboe * Note: only works on ITER_IOVEC, ITER_BVEC, and ITER_KVEC 15728fb0f47aSJens Axboe */ 15738fb0f47aSJens Axboe void iov_iter_restore(struct iov_iter *i, struct iov_iter_state *state) 15748fb0f47aSJens Axboe { 15754397a17cSKeith Busch if (WARN_ON_ONCE(!iov_iter_is_bvec(i) && !iter_is_iovec(i) && 15764397a17cSKeith Busch !iter_is_ubuf(i)) && !iov_iter_is_kvec(i)) 15778fb0f47aSJens Axboe return; 15788fb0f47aSJens Axboe i->iov_offset = state->iov_offset; 15798fb0f47aSJens Axboe i->count = state->count; 1580fcb14cb1SAl Viro if (iter_is_ubuf(i)) 1581fcb14cb1SAl Viro return; 15828fb0f47aSJens Axboe /* 15838fb0f47aSJens Axboe * For the *vec iters, nr_segs + iov is constant - if we increment 15848fb0f47aSJens Axboe * the vec, then we also decrement the nr_segs count. Hence we don't 15858fb0f47aSJens Axboe * need to track both of these, just one is enough and we can deduct 15868fb0f47aSJens Axboe * the other from that. ITER_KVEC and ITER_IOVEC are the same struct 15878fb0f47aSJens Axboe * size, so we can just increment the iov pointer as they are unionzed. 15888fb0f47aSJens Axboe * ITER_BVEC _may_ be the same size on some archs, but on others it is 15898fb0f47aSJens Axboe * not. Be safe and handle it separately. 15908fb0f47aSJens Axboe */ 15918fb0f47aSJens Axboe BUILD_BUG_ON(sizeof(struct iovec) != sizeof(struct kvec)); 15928fb0f47aSJens Axboe if (iov_iter_is_bvec(i)) 15938fb0f47aSJens Axboe i->bvec -= state->nr_segs - i->nr_segs; 15948fb0f47aSJens Axboe else 1595de4f5fedSJens Axboe i->__iov -= state->nr_segs - i->nr_segs; 15968fb0f47aSJens Axboe i->nr_segs = state->nr_segs; 15978fb0f47aSJens Axboe } 15987d58fe73SDavid Howells 15997d58fe73SDavid Howells /* 16007d58fe73SDavid Howells * Extract a list of contiguous pages from an ITER_XARRAY iterator. This does not 16017d58fe73SDavid Howells * get references on the pages, nor does it get a pin on them. 16027d58fe73SDavid Howells */ 16037d58fe73SDavid Howells static ssize_t iov_iter_extract_xarray_pages(struct iov_iter *i, 16047d58fe73SDavid Howells struct page ***pages, size_t maxsize, 16057d58fe73SDavid Howells unsigned int maxpages, 16067d58fe73SDavid Howells iov_iter_extraction_t extraction_flags, 16077d58fe73SDavid Howells size_t *offset0) 16087d58fe73SDavid Howells { 16097d58fe73SDavid Howells struct page *page, **p; 16107d58fe73SDavid Howells unsigned int nr = 0, offset; 16117d58fe73SDavid Howells loff_t pos = i->xarray_start + i->iov_offset; 16127d58fe73SDavid Howells pgoff_t index = pos >> PAGE_SHIFT; 16137d58fe73SDavid Howells XA_STATE(xas, i->xarray, index); 16147d58fe73SDavid Howells 16157d58fe73SDavid Howells offset = pos & ~PAGE_MASK; 16167d58fe73SDavid Howells *offset0 = offset; 16177d58fe73SDavid Howells 16187d58fe73SDavid Howells maxpages = want_pages_array(pages, maxsize, offset, maxpages); 16197d58fe73SDavid Howells if (!maxpages) 16207d58fe73SDavid Howells return -ENOMEM; 16217d58fe73SDavid Howells p = *pages; 16227d58fe73SDavid Howells 16237d58fe73SDavid Howells rcu_read_lock(); 16247d58fe73SDavid Howells for (page = xas_load(&xas); page; page = xas_next(&xas)) { 16257d58fe73SDavid Howells if (xas_retry(&xas, page)) 16267d58fe73SDavid Howells continue; 16277d58fe73SDavid Howells 16287d58fe73SDavid Howells /* Has the page moved or been split? */ 16297d58fe73SDavid Howells if (unlikely(page != xas_reload(&xas))) { 16307d58fe73SDavid Howells xas_reset(&xas); 16317d58fe73SDavid Howells continue; 16327d58fe73SDavid Howells } 16337d58fe73SDavid Howells 16347d58fe73SDavid Howells p[nr++] = find_subpage(page, xas.xa_index); 16357d58fe73SDavid Howells if (nr == maxpages) 16367d58fe73SDavid Howells break; 16377d58fe73SDavid Howells } 16387d58fe73SDavid Howells rcu_read_unlock(); 16397d58fe73SDavid Howells 16407d58fe73SDavid Howells maxsize = min_t(size_t, nr * PAGE_SIZE - offset, maxsize); 16417d58fe73SDavid Howells iov_iter_advance(i, maxsize); 16427d58fe73SDavid Howells return maxsize; 16437d58fe73SDavid Howells } 16447d58fe73SDavid Howells 16457d58fe73SDavid Howells /* 16467d58fe73SDavid Howells * Extract a list of contiguous pages from an ITER_BVEC iterator. This does 16477d58fe73SDavid Howells * not get references on the pages, nor does it get a pin on them. 16487d58fe73SDavid Howells */ 16497d58fe73SDavid Howells static ssize_t iov_iter_extract_bvec_pages(struct iov_iter *i, 16507d58fe73SDavid Howells struct page ***pages, size_t maxsize, 16517d58fe73SDavid Howells unsigned int maxpages, 16527d58fe73SDavid Howells iov_iter_extraction_t extraction_flags, 16537d58fe73SDavid Howells size_t *offset0) 16547d58fe73SDavid Howells { 16557d58fe73SDavid Howells struct page **p, *page; 16567d58fe73SDavid Howells size_t skip = i->iov_offset, offset; 16577d58fe73SDavid Howells int k; 16587d58fe73SDavid Howells 16597d58fe73SDavid Howells for (;;) { 16607d58fe73SDavid Howells if (i->nr_segs == 0) 16617d58fe73SDavid Howells return 0; 16627d58fe73SDavid Howells maxsize = min(maxsize, i->bvec->bv_len - skip); 16637d58fe73SDavid Howells if (maxsize) 16647d58fe73SDavid Howells break; 16657d58fe73SDavid Howells i->iov_offset = 0; 16667d58fe73SDavid Howells i->nr_segs--; 16677d58fe73SDavid Howells i->bvec++; 16687d58fe73SDavid Howells skip = 0; 16697d58fe73SDavid Howells } 16707d58fe73SDavid Howells 16717d58fe73SDavid Howells skip += i->bvec->bv_offset; 16727d58fe73SDavid Howells page = i->bvec->bv_page + skip / PAGE_SIZE; 16737d58fe73SDavid Howells offset = skip % PAGE_SIZE; 16747d58fe73SDavid Howells *offset0 = offset; 16757d58fe73SDavid Howells 16767d58fe73SDavid Howells maxpages = want_pages_array(pages, maxsize, offset, maxpages); 16777d58fe73SDavid Howells if (!maxpages) 16787d58fe73SDavid Howells return -ENOMEM; 16797d58fe73SDavid Howells p = *pages; 16807d58fe73SDavid Howells for (k = 0; k < maxpages; k++) 16817d58fe73SDavid Howells p[k] = page + k; 16827d58fe73SDavid Howells 16837d58fe73SDavid Howells maxsize = min_t(size_t, maxsize, maxpages * PAGE_SIZE - offset); 16847d58fe73SDavid Howells iov_iter_advance(i, maxsize); 16857d58fe73SDavid Howells return maxsize; 16867d58fe73SDavid Howells } 16877d58fe73SDavid Howells 16887d58fe73SDavid Howells /* 16897d58fe73SDavid Howells * Extract a list of virtually contiguous pages from an ITER_KVEC iterator. 16907d58fe73SDavid Howells * This does not get references on the pages, nor does it get a pin on them. 16917d58fe73SDavid Howells */ 16927d58fe73SDavid Howells static ssize_t iov_iter_extract_kvec_pages(struct iov_iter *i, 16937d58fe73SDavid Howells struct page ***pages, size_t maxsize, 16947d58fe73SDavid Howells unsigned int maxpages, 16957d58fe73SDavid Howells iov_iter_extraction_t extraction_flags, 16967d58fe73SDavid Howells size_t *offset0) 16977d58fe73SDavid Howells { 16987d58fe73SDavid Howells struct page **p, *page; 16997d58fe73SDavid Howells const void *kaddr; 17007d58fe73SDavid Howells size_t skip = i->iov_offset, offset, len; 17017d58fe73SDavid Howells int k; 17027d58fe73SDavid Howells 17037d58fe73SDavid Howells for (;;) { 17047d58fe73SDavid Howells if (i->nr_segs == 0) 17057d58fe73SDavid Howells return 0; 17067d58fe73SDavid Howells maxsize = min(maxsize, i->kvec->iov_len - skip); 17077d58fe73SDavid Howells if (maxsize) 17087d58fe73SDavid Howells break; 17097d58fe73SDavid Howells i->iov_offset = 0; 17107d58fe73SDavid Howells i->nr_segs--; 17117d58fe73SDavid Howells i->kvec++; 17127d58fe73SDavid Howells skip = 0; 17137d58fe73SDavid Howells } 17147d58fe73SDavid Howells 17157d58fe73SDavid Howells kaddr = i->kvec->iov_base + skip; 17167d58fe73SDavid Howells offset = (unsigned long)kaddr & ~PAGE_MASK; 17177d58fe73SDavid Howells *offset0 = offset; 17187d58fe73SDavid Howells 17197d58fe73SDavid Howells maxpages = want_pages_array(pages, maxsize, offset, maxpages); 17207d58fe73SDavid Howells if (!maxpages) 17217d58fe73SDavid Howells return -ENOMEM; 17227d58fe73SDavid Howells p = *pages; 17237d58fe73SDavid Howells 17247d58fe73SDavid Howells kaddr -= offset; 17257d58fe73SDavid Howells len = offset + maxsize; 17267d58fe73SDavid Howells for (k = 0; k < maxpages; k++) { 17277d58fe73SDavid Howells size_t seg = min_t(size_t, len, PAGE_SIZE); 17287d58fe73SDavid Howells 17297d58fe73SDavid Howells if (is_vmalloc_or_module_addr(kaddr)) 17307d58fe73SDavid Howells page = vmalloc_to_page(kaddr); 17317d58fe73SDavid Howells else 17327d58fe73SDavid Howells page = virt_to_page(kaddr); 17337d58fe73SDavid Howells 17347d58fe73SDavid Howells p[k] = page; 17357d58fe73SDavid Howells len -= seg; 17367d58fe73SDavid Howells kaddr += PAGE_SIZE; 17377d58fe73SDavid Howells } 17387d58fe73SDavid Howells 17397d58fe73SDavid Howells maxsize = min_t(size_t, maxsize, maxpages * PAGE_SIZE - offset); 17407d58fe73SDavid Howells iov_iter_advance(i, maxsize); 17417d58fe73SDavid Howells return maxsize; 17427d58fe73SDavid Howells } 17437d58fe73SDavid Howells 17447d58fe73SDavid Howells /* 17457d58fe73SDavid Howells * Extract a list of contiguous pages from a user iterator and get a pin on 17467d58fe73SDavid Howells * each of them. This should only be used if the iterator is user-backed 17477d58fe73SDavid Howells * (IOBUF/UBUF). 17487d58fe73SDavid Howells * 17497d58fe73SDavid Howells * It does not get refs on the pages, but the pages must be unpinned by the 17507d58fe73SDavid Howells * caller once the transfer is complete. 17517d58fe73SDavid Howells * 17527d58fe73SDavid Howells * This is safe to be used where background IO/DMA *is* going to be modifying 17537d58fe73SDavid Howells * the buffer; using a pin rather than a ref makes forces fork() to give the 17547d58fe73SDavid Howells * child a copy of the page. 17557d58fe73SDavid Howells */ 17567d58fe73SDavid Howells static ssize_t iov_iter_extract_user_pages(struct iov_iter *i, 17577d58fe73SDavid Howells struct page ***pages, 17587d58fe73SDavid Howells size_t maxsize, 17597d58fe73SDavid Howells unsigned int maxpages, 17607d58fe73SDavid Howells iov_iter_extraction_t extraction_flags, 17617d58fe73SDavid Howells size_t *offset0) 17627d58fe73SDavid Howells { 17637d58fe73SDavid Howells unsigned long addr; 17647d58fe73SDavid Howells unsigned int gup_flags = 0; 17657d58fe73SDavid Howells size_t offset; 17667d58fe73SDavid Howells int res; 17677d58fe73SDavid Howells 17687d58fe73SDavid Howells if (i->data_source == ITER_DEST) 17697d58fe73SDavid Howells gup_flags |= FOLL_WRITE; 17707d58fe73SDavid Howells if (extraction_flags & ITER_ALLOW_P2PDMA) 17717d58fe73SDavid Howells gup_flags |= FOLL_PCI_P2PDMA; 17727d58fe73SDavid Howells if (i->nofault) 17737d58fe73SDavid Howells gup_flags |= FOLL_NOFAULT; 17747d58fe73SDavid Howells 17757d58fe73SDavid Howells addr = first_iovec_segment(i, &maxsize); 17767d58fe73SDavid Howells *offset0 = offset = addr % PAGE_SIZE; 17777d58fe73SDavid Howells addr &= PAGE_MASK; 17787d58fe73SDavid Howells maxpages = want_pages_array(pages, maxsize, offset, maxpages); 17797d58fe73SDavid Howells if (!maxpages) 17807d58fe73SDavid Howells return -ENOMEM; 17817d58fe73SDavid Howells res = pin_user_pages_fast(addr, maxpages, gup_flags, *pages); 17827d58fe73SDavid Howells if (unlikely(res <= 0)) 17837d58fe73SDavid Howells return res; 17847d58fe73SDavid Howells maxsize = min_t(size_t, maxsize, res * PAGE_SIZE - offset); 17857d58fe73SDavid Howells iov_iter_advance(i, maxsize); 17867d58fe73SDavid Howells return maxsize; 17877d58fe73SDavid Howells } 17887d58fe73SDavid Howells 17897d58fe73SDavid Howells /** 17907d58fe73SDavid Howells * iov_iter_extract_pages - Extract a list of contiguous pages from an iterator 17917d58fe73SDavid Howells * @i: The iterator to extract from 17927d58fe73SDavid Howells * @pages: Where to return the list of pages 17937d58fe73SDavid Howells * @maxsize: The maximum amount of iterator to extract 17947d58fe73SDavid Howells * @maxpages: The maximum size of the list of pages 17957d58fe73SDavid Howells * @extraction_flags: Flags to qualify request 17967d58fe73SDavid Howells * @offset0: Where to return the starting offset into (*@pages)[0] 17977d58fe73SDavid Howells * 17987d58fe73SDavid Howells * Extract a list of contiguous pages from the current point of the iterator, 17997d58fe73SDavid Howells * advancing the iterator. The maximum number of pages and the maximum amount 18007d58fe73SDavid Howells * of page contents can be set. 18017d58fe73SDavid Howells * 18027d58fe73SDavid Howells * If *@pages is NULL, a page list will be allocated to the required size and 18037d58fe73SDavid Howells * *@pages will be set to its base. If *@pages is not NULL, it will be assumed 18047d58fe73SDavid Howells * that the caller allocated a page list at least @maxpages in size and this 18057d58fe73SDavid Howells * will be filled in. 18067d58fe73SDavid Howells * 18077d58fe73SDavid Howells * @extraction_flags can have ITER_ALLOW_P2PDMA set to request peer-to-peer DMA 18087d58fe73SDavid Howells * be allowed on the pages extracted. 18097d58fe73SDavid Howells * 18107d58fe73SDavid Howells * The iov_iter_extract_will_pin() function can be used to query how cleanup 18117d58fe73SDavid Howells * should be performed. 18127d58fe73SDavid Howells * 18137d58fe73SDavid Howells * Extra refs or pins on the pages may be obtained as follows: 18147d58fe73SDavid Howells * 18157d58fe73SDavid Howells * (*) If the iterator is user-backed (ITER_IOVEC/ITER_UBUF), pins will be 18167d58fe73SDavid Howells * added to the pages, but refs will not be taken. 18177d58fe73SDavid Howells * iov_iter_extract_will_pin() will return true. 18187d58fe73SDavid Howells * 18197d58fe73SDavid Howells * (*) If the iterator is ITER_KVEC, ITER_BVEC or ITER_XARRAY, the pages are 18207d58fe73SDavid Howells * merely listed; no extra refs or pins are obtained. 18217d58fe73SDavid Howells * iov_iter_extract_will_pin() will return 0. 18227d58fe73SDavid Howells * 18237d58fe73SDavid Howells * Note also: 18247d58fe73SDavid Howells * 18257d58fe73SDavid Howells * (*) Use with ITER_DISCARD is not supported as that has no content. 18267d58fe73SDavid Howells * 18277d58fe73SDavid Howells * On success, the function sets *@pages to the new pagelist, if allocated, and 18287d58fe73SDavid Howells * sets *offset0 to the offset into the first page. 18297d58fe73SDavid Howells * 18307d58fe73SDavid Howells * It may also return -ENOMEM and -EFAULT. 18317d58fe73SDavid Howells */ 18327d58fe73SDavid Howells ssize_t iov_iter_extract_pages(struct iov_iter *i, 18337d58fe73SDavid Howells struct page ***pages, 18347d58fe73SDavid Howells size_t maxsize, 18357d58fe73SDavid Howells unsigned int maxpages, 18367d58fe73SDavid Howells iov_iter_extraction_t extraction_flags, 18377d58fe73SDavid Howells size_t *offset0) 18387d58fe73SDavid Howells { 18397d58fe73SDavid Howells maxsize = min_t(size_t, min_t(size_t, maxsize, i->count), MAX_RW_COUNT); 18407d58fe73SDavid Howells if (!maxsize) 18417d58fe73SDavid Howells return 0; 18427d58fe73SDavid Howells 18437d58fe73SDavid Howells if (likely(user_backed_iter(i))) 18447d58fe73SDavid Howells return iov_iter_extract_user_pages(i, pages, maxsize, 18457d58fe73SDavid Howells maxpages, extraction_flags, 18467d58fe73SDavid Howells offset0); 18477d58fe73SDavid Howells if (iov_iter_is_kvec(i)) 18487d58fe73SDavid Howells return iov_iter_extract_kvec_pages(i, pages, maxsize, 18497d58fe73SDavid Howells maxpages, extraction_flags, 18507d58fe73SDavid Howells offset0); 18517d58fe73SDavid Howells if (iov_iter_is_bvec(i)) 18527d58fe73SDavid Howells return iov_iter_extract_bvec_pages(i, pages, maxsize, 18537d58fe73SDavid Howells maxpages, extraction_flags, 18547d58fe73SDavid Howells offset0); 18557d58fe73SDavid Howells if (iov_iter_is_xarray(i)) 18567d58fe73SDavid Howells return iov_iter_extract_xarray_pages(i, pages, maxsize, 18577d58fe73SDavid Howells maxpages, extraction_flags, 18587d58fe73SDavid Howells offset0); 18597d58fe73SDavid Howells return -EFAULT; 18607d58fe73SDavid Howells } 18617d58fe73SDavid Howells EXPORT_SYMBOL_GPL(iov_iter_extract_pages); 1862