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 17241699cdSAl Viro #define PIPE_PARANOIA /* for now */ 18241699cdSAl Viro 19fcb14cb1SAl Viro /* covers ubuf and kbuf alike */ 20fcb14cb1SAl Viro #define iterate_buf(i, n, base, len, off, __p, STEP) { \ 21fcb14cb1SAl Viro size_t __maybe_unused off = 0; \ 22fcb14cb1SAl Viro len = n; \ 23fcb14cb1SAl Viro base = __p + i->iov_offset; \ 24fcb14cb1SAl Viro len -= (STEP); \ 25fcb14cb1SAl Viro i->iov_offset += len; \ 26fcb14cb1SAl Viro n = len; \ 27fcb14cb1SAl Viro } 28fcb14cb1SAl Viro 295c67aa90SAl Viro /* covers iovec and kvec alike */ 30a6e4ec7bSAl Viro #define iterate_iovec(i, n, base, len, off, __p, STEP) { \ 317baa5099SAl Viro size_t off = 0; \ 32a6e4ec7bSAl Viro size_t skip = i->iov_offset; \ 337a1bcb5dSAl Viro do { \ 347baa5099SAl Viro len = min(n, __p->iov_len - skip); \ 357baa5099SAl Viro if (likely(len)) { \ 367baa5099SAl Viro base = __p->iov_base + skip; \ 377baa5099SAl Viro len -= (STEP); \ 387baa5099SAl Viro off += len; \ 397baa5099SAl Viro skip += len; \ 407baa5099SAl Viro n -= len; \ 417a1bcb5dSAl Viro if (skip < __p->iov_len) \ 427a1bcb5dSAl Viro break; \ 43d879cb83SAl Viro } \ 44d879cb83SAl Viro __p++; \ 457a1bcb5dSAl Viro skip = 0; \ 467a1bcb5dSAl Viro } while (n); \ 47a6e4ec7bSAl Viro i->iov_offset = skip; \ 487baa5099SAl Viro n = off; \ 49d879cb83SAl Viro } 50d879cb83SAl Viro 51a6e4ec7bSAl Viro #define iterate_bvec(i, n, base, len, off, p, STEP) { \ 527baa5099SAl Viro size_t off = 0; \ 53a6e4ec7bSAl Viro unsigned skip = i->iov_offset; \ 547491a2bfSAl Viro while (n) { \ 557491a2bfSAl Viro unsigned offset = p->bv_offset + skip; \ 561b4fb5ffSAl Viro unsigned left; \ 5721b56c84SAl Viro void *kaddr = kmap_local_page(p->bv_page + \ 5821b56c84SAl Viro offset / PAGE_SIZE); \ 597baa5099SAl Viro base = kaddr + offset % PAGE_SIZE; \ 60a6e4ec7bSAl Viro len = min(min(n, (size_t)(p->bv_len - skip)), \ 617491a2bfSAl Viro (size_t)(PAGE_SIZE - offset % PAGE_SIZE)); \ 621b4fb5ffSAl Viro left = (STEP); \ 6321b56c84SAl Viro kunmap_local(kaddr); \ 647baa5099SAl Viro len -= left; \ 657baa5099SAl Viro off += len; \ 667baa5099SAl Viro skip += len; \ 677491a2bfSAl Viro if (skip == p->bv_len) { \ 687491a2bfSAl Viro skip = 0; \ 697491a2bfSAl Viro p++; \ 70d879cb83SAl Viro } \ 717baa5099SAl Viro n -= len; \ 721b4fb5ffSAl Viro if (left) \ 731b4fb5ffSAl Viro break; \ 747491a2bfSAl Viro } \ 75a6e4ec7bSAl Viro i->iov_offset = skip; \ 767baa5099SAl Viro n = off; \ 77d879cb83SAl Viro } 78d879cb83SAl Viro 79a6e4ec7bSAl Viro #define iterate_xarray(i, n, base, len, __off, STEP) { \ 801b4fb5ffSAl Viro __label__ __out; \ 81622838f3SAl Viro size_t __off = 0; \ 82821979f5SMatthew Wilcox (Oracle) struct folio *folio; \ 83a6e4ec7bSAl Viro loff_t start = i->xarray_start + i->iov_offset; \ 844b179e9aSAl Viro pgoff_t index = start / PAGE_SIZE; \ 857ff50620SDavid Howells XA_STATE(xas, i->xarray, index); \ 867ff50620SDavid Howells \ 87821979f5SMatthew Wilcox (Oracle) len = PAGE_SIZE - offset_in_page(start); \ 887ff50620SDavid Howells rcu_read_lock(); \ 89821979f5SMatthew Wilcox (Oracle) xas_for_each(&xas, folio, ULONG_MAX) { \ 901b4fb5ffSAl Viro unsigned left; \ 91821979f5SMatthew Wilcox (Oracle) size_t offset; \ 92821979f5SMatthew Wilcox (Oracle) if (xas_retry(&xas, folio)) \ 937ff50620SDavid Howells continue; \ 94821979f5SMatthew Wilcox (Oracle) if (WARN_ON(xa_is_value(folio))) \ 957ff50620SDavid Howells break; \ 96821979f5SMatthew Wilcox (Oracle) if (WARN_ON(folio_test_hugetlb(folio))) \ 977ff50620SDavid Howells break; \ 98821979f5SMatthew Wilcox (Oracle) offset = offset_in_folio(folio, start + __off); \ 99821979f5SMatthew Wilcox (Oracle) while (offset < folio_size(folio)) { \ 100821979f5SMatthew Wilcox (Oracle) base = kmap_local_folio(folio, offset); \ 1017baa5099SAl Viro len = min(n, len); \ 1021b4fb5ffSAl Viro left = (STEP); \ 103821979f5SMatthew Wilcox (Oracle) kunmap_local(base); \ 1047baa5099SAl Viro len -= left; \ 1057baa5099SAl Viro __off += len; \ 1067baa5099SAl Viro n -= len; \ 1071b4fb5ffSAl Viro if (left || n == 0) \ 1081b4fb5ffSAl Viro goto __out; \ 109821979f5SMatthew Wilcox (Oracle) offset += len; \ 110821979f5SMatthew Wilcox (Oracle) len = PAGE_SIZE; \ 1117ff50620SDavid Howells } \ 1127ff50620SDavid Howells } \ 1131b4fb5ffSAl Viro __out: \ 1147ff50620SDavid Howells rcu_read_unlock(); \ 115a6e4ec7bSAl Viro i->iov_offset += __off; \ 116622838f3SAl Viro n = __off; \ 1177ff50620SDavid Howells } 1187ff50620SDavid Howells 1197baa5099SAl Viro #define __iterate_and_advance(i, n, base, len, off, I, K) { \ 120dd254f5aSAl Viro if (unlikely(i->count < n)) \ 121dd254f5aSAl Viro n = i->count; \ 122f5da8354SAl Viro if (likely(n)) { \ 123fcb14cb1SAl Viro if (likely(iter_is_ubuf(i))) { \ 124fcb14cb1SAl Viro void __user *base; \ 125fcb14cb1SAl Viro size_t len; \ 126fcb14cb1SAl Viro iterate_buf(i, n, base, len, off, \ 127fcb14cb1SAl Viro i->ubuf, (I)) \ 128fcb14cb1SAl Viro } else if (likely(iter_is_iovec(i))) { \ 129de4f5fedSJens Axboe const struct iovec *iov = iter_iov(i); \ 1307baa5099SAl Viro void __user *base; \ 1317baa5099SAl Viro size_t len; \ 1327baa5099SAl Viro iterate_iovec(i, n, base, len, off, \ 133a6e4ec7bSAl Viro iov, (I)) \ 134de4f5fedSJens Axboe i->nr_segs -= iov - iter_iov(i); \ 135de4f5fedSJens Axboe i->__iov = iov; \ 13628f38db7SAl Viro } else if (iov_iter_is_bvec(i)) { \ 13728f38db7SAl Viro const struct bio_vec *bvec = i->bvec; \ 1387baa5099SAl Viro void *base; \ 1397baa5099SAl Viro size_t len; \ 1407baa5099SAl Viro iterate_bvec(i, n, base, len, off, \ 141a6e4ec7bSAl Viro bvec, (K)) \ 1427491a2bfSAl Viro i->nr_segs -= bvec - i->bvec; \ 1437491a2bfSAl Viro i->bvec = bvec; \ 14428f38db7SAl Viro } else if (iov_iter_is_kvec(i)) { \ 1455c67aa90SAl Viro const struct kvec *kvec = i->kvec; \ 1467baa5099SAl Viro void *base; \ 1477baa5099SAl Viro size_t len; \ 1487baa5099SAl Viro iterate_iovec(i, n, base, len, off, \ 149a6e4ec7bSAl Viro kvec, (K)) \ 15028f38db7SAl Viro i->nr_segs -= kvec - i->kvec; \ 15128f38db7SAl Viro i->kvec = kvec; \ 15228f38db7SAl Viro } else if (iov_iter_is_xarray(i)) { \ 1537baa5099SAl Viro void *base; \ 1547baa5099SAl Viro size_t len; \ 1557baa5099SAl Viro iterate_xarray(i, n, base, len, off, \ 156a6e4ec7bSAl Viro (K)) \ 157d879cb83SAl Viro } \ 158d879cb83SAl Viro i->count -= n; \ 159dd254f5aSAl Viro } \ 160d879cb83SAl Viro } 1617baa5099SAl Viro #define iterate_and_advance(i, n, base, len, off, I, K) \ 1627baa5099SAl Viro __iterate_and_advance(i, n, base, len, off, I, ((void)(K),0)) 163d879cb83SAl Viro 16409fc68dcSAl Viro static int copyout(void __user *to, const void *from, size_t n) 16509fc68dcSAl Viro { 1664d0e9df5SAlbert van der Linde if (should_fail_usercopy()) 1674d0e9df5SAlbert van der Linde return n; 16896d4f267SLinus Torvalds if (access_ok(to, n)) { 169d0ef4c36SMarco Elver instrument_copy_to_user(to, from, n); 17009fc68dcSAl Viro n = raw_copy_to_user(to, from, n); 17109fc68dcSAl Viro } 17209fc68dcSAl Viro return n; 17309fc68dcSAl Viro } 17409fc68dcSAl Viro 17509fc68dcSAl Viro static int copyin(void *to, const void __user *from, size_t n) 17609fc68dcSAl Viro { 17733b75c1dSAlexander Potapenko size_t res = n; 17833b75c1dSAlexander Potapenko 1794d0e9df5SAlbert van der Linde if (should_fail_usercopy()) 1804d0e9df5SAlbert van der Linde return n; 18196d4f267SLinus Torvalds if (access_ok(from, n)) { 18233b75c1dSAlexander Potapenko instrument_copy_from_user_before(to, from, n); 18333b75c1dSAlexander Potapenko res = raw_copy_from_user(to, from, n); 18433b75c1dSAlexander Potapenko instrument_copy_from_user_after(to, from, n, res); 18509fc68dcSAl Viro } 18633b75c1dSAlexander Potapenko return res; 18709fc68dcSAl Viro } 18809fc68dcSAl Viro 189241699cdSAl Viro #ifdef PIPE_PARANOIA 190241699cdSAl Viro static bool sanity(const struct iov_iter *i) 191241699cdSAl Viro { 192241699cdSAl Viro struct pipe_inode_info *pipe = i->pipe; 1938cefc107SDavid Howells unsigned int p_head = pipe->head; 1948cefc107SDavid Howells unsigned int p_tail = pipe->tail; 1958cefc107SDavid Howells unsigned int p_occupancy = pipe_occupancy(p_head, p_tail); 1968cefc107SDavid Howells unsigned int i_head = i->head; 1978cefc107SDavid Howells unsigned int idx; 1988cefc107SDavid Howells 19910f525a8SAl Viro if (i->last_offset) { 200241699cdSAl Viro struct pipe_buffer *p; 2018cefc107SDavid Howells if (unlikely(p_occupancy == 0)) 202241699cdSAl Viro goto Bad; // pipe must be non-empty 2038cefc107SDavid Howells if (unlikely(i_head != p_head - 1)) 204241699cdSAl Viro goto Bad; // must be at the last buffer... 205241699cdSAl Viro 2062dcedb2aSAl Viro p = pipe_buf(pipe, i_head); 20710f525a8SAl Viro if (unlikely(p->offset + p->len != abs(i->last_offset))) 208241699cdSAl Viro goto Bad; // ... at the end of segment 209241699cdSAl Viro } else { 2108cefc107SDavid Howells if (i_head != p_head) 211241699cdSAl Viro goto Bad; // must be right after the last buffer 212241699cdSAl Viro } 213241699cdSAl Viro return true; 214241699cdSAl Viro Bad: 21510f525a8SAl Viro printk(KERN_ERR "idx = %d, offset = %d\n", i_head, i->last_offset); 2168cefc107SDavid Howells printk(KERN_ERR "head = %d, tail = %d, buffers = %d\n", 2178cefc107SDavid Howells p_head, p_tail, pipe->ring_size); 2188cefc107SDavid Howells for (idx = 0; idx < pipe->ring_size; idx++) 219241699cdSAl Viro printk(KERN_ERR "[%p %p %d %d]\n", 220241699cdSAl Viro pipe->bufs[idx].ops, 221241699cdSAl Viro pipe->bufs[idx].page, 222241699cdSAl Viro pipe->bufs[idx].offset, 223241699cdSAl Viro pipe->bufs[idx].len); 224241699cdSAl Viro WARN_ON(1); 225241699cdSAl Viro return false; 226241699cdSAl Viro } 227241699cdSAl Viro #else 228241699cdSAl Viro #define sanity(i) true 229241699cdSAl Viro #endif 230241699cdSAl Viro 23147b7fcaeSAl Viro static struct page *push_anon(struct pipe_inode_info *pipe, unsigned size) 23247b7fcaeSAl Viro { 23347b7fcaeSAl Viro struct page *page = alloc_page(GFP_USER); 23447b7fcaeSAl Viro if (page) { 23547b7fcaeSAl Viro struct pipe_buffer *buf = pipe_buf(pipe, pipe->head++); 23647b7fcaeSAl Viro *buf = (struct pipe_buffer) { 23747b7fcaeSAl Viro .ops = &default_pipe_buf_ops, 23847b7fcaeSAl Viro .page = page, 23947b7fcaeSAl Viro .offset = 0, 24047b7fcaeSAl Viro .len = size 24147b7fcaeSAl Viro }; 24247b7fcaeSAl Viro } 24347b7fcaeSAl Viro return page; 24447b7fcaeSAl Viro } 24547b7fcaeSAl Viro 24647b7fcaeSAl Viro static void push_page(struct pipe_inode_info *pipe, struct page *page, 24747b7fcaeSAl Viro unsigned int offset, unsigned int size) 24847b7fcaeSAl Viro { 24947b7fcaeSAl Viro struct pipe_buffer *buf = pipe_buf(pipe, pipe->head++); 25047b7fcaeSAl Viro *buf = (struct pipe_buffer) { 25147b7fcaeSAl Viro .ops = &page_cache_pipe_buf_ops, 25247b7fcaeSAl Viro .page = page, 25347b7fcaeSAl Viro .offset = offset, 25447b7fcaeSAl Viro .len = size 25547b7fcaeSAl Viro }; 25647b7fcaeSAl Viro get_page(page); 25747b7fcaeSAl Viro } 25847b7fcaeSAl Viro 25910f525a8SAl Viro static inline int last_offset(const struct pipe_buffer *buf) 2608fad7767SAl Viro { 26110f525a8SAl Viro if (buf->ops == &default_pipe_buf_ops) 26210f525a8SAl Viro return buf->len; // buf->offset is 0 for those 26310f525a8SAl Viro else 26410f525a8SAl Viro return -(buf->offset + buf->len); 2658fad7767SAl Viro } 2668fad7767SAl Viro 2678fad7767SAl Viro static struct page *append_pipe(struct iov_iter *i, size_t size, 2688fad7767SAl Viro unsigned int *off) 2698fad7767SAl Viro { 2708fad7767SAl Viro struct pipe_inode_info *pipe = i->pipe; 27110f525a8SAl Viro int offset = i->last_offset; 2728fad7767SAl Viro struct pipe_buffer *buf; 2738fad7767SAl Viro struct page *page; 2748fad7767SAl Viro 27510f525a8SAl Viro if (offset > 0 && offset < PAGE_SIZE) { 27610f525a8SAl Viro // some space in the last buffer; add to it 2778fad7767SAl Viro buf = pipe_buf(pipe, pipe->head - 1); 2788fad7767SAl Viro size = min_t(size_t, size, PAGE_SIZE - offset); 2798fad7767SAl Viro buf->len += size; 28010f525a8SAl Viro i->last_offset += size; 2818fad7767SAl Viro i->count -= size; 2828fad7767SAl Viro *off = offset; 2838fad7767SAl Viro return buf->page; 2848fad7767SAl Viro } 2858fad7767SAl Viro // OK, we need a new buffer 2868fad7767SAl Viro *off = 0; 2878fad7767SAl Viro size = min_t(size_t, size, PAGE_SIZE); 2888fad7767SAl Viro if (pipe_full(pipe->head, pipe->tail, pipe->max_usage)) 2898fad7767SAl Viro return NULL; 2908fad7767SAl Viro page = push_anon(pipe, size); 2918fad7767SAl Viro if (!page) 2928fad7767SAl Viro return NULL; 2938fad7767SAl Viro i->head = pipe->head - 1; 29410f525a8SAl Viro i->last_offset = size; 2958fad7767SAl Viro i->count -= size; 2968fad7767SAl Viro return page; 2978fad7767SAl Viro } 2988fad7767SAl Viro 299241699cdSAl Viro static size_t copy_page_to_iter_pipe(struct page *page, size_t offset, size_t bytes, 300241699cdSAl Viro struct iov_iter *i) 301241699cdSAl Viro { 302241699cdSAl Viro struct pipe_inode_info *pipe = i->pipe; 30347b7fcaeSAl Viro unsigned int head = pipe->head; 304241699cdSAl Viro 305241699cdSAl Viro if (unlikely(bytes > i->count)) 306241699cdSAl Viro bytes = i->count; 307241699cdSAl Viro 308241699cdSAl Viro if (unlikely(!bytes)) 309241699cdSAl Viro return 0; 310241699cdSAl Viro 311241699cdSAl Viro if (!sanity(i)) 312241699cdSAl Viro return 0; 313241699cdSAl Viro 31410f525a8SAl Viro if (offset && i->last_offset == -offset) { // could we merge it? 31547b7fcaeSAl Viro struct pipe_buffer *buf = pipe_buf(pipe, head - 1); 31647b7fcaeSAl Viro if (buf->page == page) { 317241699cdSAl Viro buf->len += bytes; 31810f525a8SAl Viro i->last_offset -= bytes; 31947b7fcaeSAl Viro i->count -= bytes; 32047b7fcaeSAl Viro return bytes; 321241699cdSAl Viro } 322241699cdSAl Viro } 32347b7fcaeSAl Viro if (pipe_full(pipe->head, pipe->tail, pipe->max_usage)) 324241699cdSAl Viro return 0; 3258cefc107SDavid Howells 32647b7fcaeSAl Viro push_page(pipe, page, offset, bytes); 32710f525a8SAl Viro i->last_offset = -(offset + bytes); 32847b7fcaeSAl Viro i->head = head; 329241699cdSAl Viro i->count -= bytes; 330241699cdSAl Viro return bytes; 331241699cdSAl Viro } 332241699cdSAl Viro 333d879cb83SAl Viro /* 334a6294593SAndreas Gruenbacher * fault_in_iov_iter_readable - fault in iov iterator for reading 335a6294593SAndreas Gruenbacher * @i: iterator 336a6294593SAndreas Gruenbacher * @size: maximum length 337171a0203SAnton Altaparmakov * 338a6294593SAndreas Gruenbacher * Fault in one or more iovecs of the given iov_iter, to a maximum length of 339a6294593SAndreas Gruenbacher * @size. For each iovec, fault in each page that constitutes the iovec. 340a6294593SAndreas Gruenbacher * 341a6294593SAndreas Gruenbacher * Returns the number of bytes not faulted in (like copy_to_user() and 342a6294593SAndreas Gruenbacher * copy_from_user()). 343a6294593SAndreas Gruenbacher * 344a6294593SAndreas Gruenbacher * Always returns 0 for non-userspace iterators. 345171a0203SAnton Altaparmakov */ 346a6294593SAndreas Gruenbacher size_t fault_in_iov_iter_readable(const struct iov_iter *i, size_t size) 347171a0203SAnton Altaparmakov { 348fcb14cb1SAl Viro if (iter_is_ubuf(i)) { 349fcb14cb1SAl Viro size_t n = min(size, iov_iter_count(i)); 350fcb14cb1SAl Viro n -= fault_in_readable(i->ubuf + i->iov_offset, n); 351fcb14cb1SAl Viro return size - n; 352fcb14cb1SAl Viro } else if (iter_is_iovec(i)) { 353a6294593SAndreas Gruenbacher size_t count = min(size, iov_iter_count(i)); 3548409a0d2SAl Viro const struct iovec *p; 3558409a0d2SAl Viro size_t skip; 3568409a0d2SAl Viro 357a6294593SAndreas Gruenbacher size -= count; 358de4f5fedSJens Axboe for (p = iter_iov(i), skip = i->iov_offset; count; p++, skip = 0) { 359a6294593SAndreas Gruenbacher size_t len = min(count, p->iov_len - skip); 360a6294593SAndreas Gruenbacher size_t ret; 3618409a0d2SAl Viro 3628409a0d2SAl Viro if (unlikely(!len)) 3638409a0d2SAl Viro continue; 364a6294593SAndreas Gruenbacher ret = fault_in_readable(p->iov_base + skip, len); 365a6294593SAndreas Gruenbacher count -= len - ret; 366a6294593SAndreas Gruenbacher if (ret) 367a6294593SAndreas Gruenbacher break; 3688409a0d2SAl Viro } 369a6294593SAndreas Gruenbacher return count + size; 370171a0203SAnton Altaparmakov } 371171a0203SAnton Altaparmakov return 0; 372171a0203SAnton Altaparmakov } 373a6294593SAndreas Gruenbacher EXPORT_SYMBOL(fault_in_iov_iter_readable); 374171a0203SAnton Altaparmakov 375cdd591fcSAndreas Gruenbacher /* 376cdd591fcSAndreas Gruenbacher * fault_in_iov_iter_writeable - fault in iov iterator for writing 377cdd591fcSAndreas Gruenbacher * @i: iterator 378cdd591fcSAndreas Gruenbacher * @size: maximum length 379cdd591fcSAndreas Gruenbacher * 380cdd591fcSAndreas Gruenbacher * Faults in the iterator using get_user_pages(), i.e., without triggering 381cdd591fcSAndreas Gruenbacher * hardware page faults. This is primarily useful when we already know that 382cdd591fcSAndreas Gruenbacher * some or all of the pages in @i aren't in memory. 383cdd591fcSAndreas Gruenbacher * 384cdd591fcSAndreas Gruenbacher * Returns the number of bytes not faulted in, like copy_to_user() and 385cdd591fcSAndreas Gruenbacher * copy_from_user(). 386cdd591fcSAndreas Gruenbacher * 387cdd591fcSAndreas Gruenbacher * Always returns 0 for non-user-space iterators. 388cdd591fcSAndreas Gruenbacher */ 389cdd591fcSAndreas Gruenbacher size_t fault_in_iov_iter_writeable(const struct iov_iter *i, size_t size) 390cdd591fcSAndreas Gruenbacher { 391fcb14cb1SAl Viro if (iter_is_ubuf(i)) { 392fcb14cb1SAl Viro size_t n = min(size, iov_iter_count(i)); 393fcb14cb1SAl Viro n -= fault_in_safe_writeable(i->ubuf + i->iov_offset, n); 394fcb14cb1SAl Viro return size - n; 395fcb14cb1SAl Viro } else if (iter_is_iovec(i)) { 396cdd591fcSAndreas Gruenbacher size_t count = min(size, iov_iter_count(i)); 397cdd591fcSAndreas Gruenbacher const struct iovec *p; 398cdd591fcSAndreas Gruenbacher size_t skip; 399cdd591fcSAndreas Gruenbacher 400cdd591fcSAndreas Gruenbacher size -= count; 401de4f5fedSJens Axboe for (p = iter_iov(i), skip = i->iov_offset; count; p++, skip = 0) { 402cdd591fcSAndreas Gruenbacher size_t len = min(count, p->iov_len - skip); 403cdd591fcSAndreas Gruenbacher size_t ret; 404cdd591fcSAndreas Gruenbacher 405cdd591fcSAndreas Gruenbacher if (unlikely(!len)) 406cdd591fcSAndreas Gruenbacher continue; 407cdd591fcSAndreas Gruenbacher ret = fault_in_safe_writeable(p->iov_base + skip, len); 408cdd591fcSAndreas Gruenbacher count -= len - ret; 409cdd591fcSAndreas Gruenbacher if (ret) 410cdd591fcSAndreas Gruenbacher break; 411cdd591fcSAndreas Gruenbacher } 412cdd591fcSAndreas Gruenbacher return count + size; 413cdd591fcSAndreas Gruenbacher } 414cdd591fcSAndreas Gruenbacher return 0; 415cdd591fcSAndreas Gruenbacher } 416cdd591fcSAndreas Gruenbacher EXPORT_SYMBOL(fault_in_iov_iter_writeable); 417cdd591fcSAndreas Gruenbacher 418aa563d7bSDavid Howells void iov_iter_init(struct iov_iter *i, unsigned int direction, 419d879cb83SAl Viro const struct iovec *iov, unsigned long nr_segs, 420d879cb83SAl Viro size_t count) 421d879cb83SAl Viro { 422aa563d7bSDavid Howells WARN_ON(direction & ~(READ | WRITE)); 4238cd54c1cSAl Viro *i = (struct iov_iter) { 4248cd54c1cSAl Viro .iter_type = ITER_IOVEC, 4253337ab08SAndreas Gruenbacher .nofault = false, 426fcb14cb1SAl Viro .user_backed = true, 4278cd54c1cSAl Viro .data_source = direction, 428de4f5fedSJens Axboe .__iov = iov, 4298cd54c1cSAl Viro .nr_segs = nr_segs, 4308cd54c1cSAl Viro .iov_offset = 0, 4318cd54c1cSAl Viro .count = count 4328cd54c1cSAl Viro }; 433d879cb83SAl Viro } 434d879cb83SAl Viro EXPORT_SYMBOL(iov_iter_init); 435d879cb83SAl Viro 43612d426abSAl Viro // returns the offset in partial buffer (if any) 43712d426abSAl Viro static inline unsigned int pipe_npages(const struct iov_iter *i, int *npages) 438241699cdSAl Viro { 43912d426abSAl Viro struct pipe_inode_info *pipe = i->pipe; 44012d426abSAl Viro int used = pipe->head - pipe->tail; 44110f525a8SAl Viro int off = i->last_offset; 4428cefc107SDavid Howells 44312d426abSAl Viro *npages = max((int)pipe->max_usage - used, 0); 44412d426abSAl Viro 44510f525a8SAl Viro if (off > 0 && off < PAGE_SIZE) { // anon and not full 44612d426abSAl Viro (*npages)++; 44712d426abSAl Viro return off; 44810f525a8SAl Viro } 44912d426abSAl Viro return 0; 450241699cdSAl Viro } 451241699cdSAl Viro 452241699cdSAl Viro static size_t copy_pipe_to_iter(const void *addr, size_t bytes, 453241699cdSAl Viro struct iov_iter *i) 454241699cdSAl Viro { 4558fad7767SAl Viro unsigned int off, chunk; 4568fad7767SAl Viro 4578fad7767SAl Viro if (unlikely(bytes > i->count)) 4588fad7767SAl Viro bytes = i->count; 4598fad7767SAl Viro if (unlikely(!bytes)) 4608fad7767SAl Viro return 0; 461241699cdSAl Viro 462241699cdSAl Viro if (!sanity(i)) 463241699cdSAl Viro return 0; 464241699cdSAl Viro 4658fad7767SAl Viro for (size_t n = bytes; n; n -= chunk) { 4668fad7767SAl Viro struct page *page = append_pipe(i, n, &off); 4678fad7767SAl Viro chunk = min_t(size_t, n, PAGE_SIZE - off); 4688fad7767SAl Viro if (!page) 4698fad7767SAl Viro return bytes - n; 4708fad7767SAl Viro memcpy_to_page(page, off, addr, chunk); 471241699cdSAl Viro addr += chunk; 4728fad7767SAl Viro } 473241699cdSAl Viro return bytes; 474241699cdSAl Viro } 475241699cdSAl Viro 476f9152895SAl Viro static __wsum csum_and_memcpy(void *to, const void *from, size_t len, 477f9152895SAl Viro __wsum sum, size_t off) 478f9152895SAl Viro { 479cc44c17bSAl Viro __wsum next = csum_partial_copy_nocheck(from, to, len); 480f9152895SAl Viro return csum_block_add(sum, next, off); 481f9152895SAl Viro } 482f9152895SAl Viro 48378e1f386SAl Viro static size_t csum_and_copy_to_pipe_iter(const void *addr, size_t bytes, 4846852df12SAl Viro struct iov_iter *i, __wsum *sump) 48578e1f386SAl Viro { 4866852df12SAl Viro __wsum sum = *sump; 4876852df12SAl Viro size_t off = 0; 4888fad7767SAl Viro unsigned int chunk, r; 4898fad7767SAl Viro 4908fad7767SAl Viro if (unlikely(bytes > i->count)) 4918fad7767SAl Viro bytes = i->count; 4928fad7767SAl Viro if (unlikely(!bytes)) 4938fad7767SAl Viro return 0; 49478e1f386SAl Viro 49578e1f386SAl Viro if (!sanity(i)) 49678e1f386SAl Viro return 0; 49778e1f386SAl Viro 4986852df12SAl Viro while (bytes) { 4998fad7767SAl Viro struct page *page = append_pipe(i, bytes, &r); 5008fad7767SAl Viro char *p; 5018fad7767SAl Viro 5028fad7767SAl Viro if (!page) 5038fad7767SAl Viro break; 5048fad7767SAl Viro chunk = min_t(size_t, bytes, PAGE_SIZE - r); 5058fad7767SAl Viro p = kmap_local_page(page); 5066852df12SAl Viro sum = csum_and_memcpy(p + r, addr + off, chunk, sum, off); 5072495bdccSAl Viro kunmap_local(p); 50878e1f386SAl Viro off += chunk; 5098fad7767SAl Viro bytes -= chunk; 5106852df12SAl Viro } 5116852df12SAl Viro *sump = sum; 5126852df12SAl Viro return off; 51378e1f386SAl Viro } 51478e1f386SAl Viro 515aa28de27SAl Viro size_t _copy_to_iter(const void *addr, size_t bytes, struct iov_iter *i) 516d879cb83SAl Viro { 517a41dad90SAl Viro if (WARN_ON_ONCE(i->data_source)) 518a41dad90SAl Viro return 0; 51900e23707SDavid Howells if (unlikely(iov_iter_is_pipe(i))) 520241699cdSAl Viro return copy_pipe_to_iter(addr, bytes, i); 521fcb14cb1SAl Viro if (user_backed_iter(i)) 52209fc68dcSAl Viro might_fault(); 5237baa5099SAl Viro iterate_and_advance(i, bytes, base, len, off, 5247baa5099SAl Viro copyout(base, addr + off, len), 5257baa5099SAl Viro memcpy(base, addr + off, len) 526d879cb83SAl Viro ) 527d879cb83SAl Viro 528d879cb83SAl Viro return bytes; 529d879cb83SAl Viro } 530aa28de27SAl Viro EXPORT_SYMBOL(_copy_to_iter); 531d879cb83SAl Viro 532ec6347bbSDan Williams #ifdef CONFIG_ARCH_HAS_COPY_MC 533ec6347bbSDan Williams static int copyout_mc(void __user *to, const void *from, size_t n) 5348780356eSDan Williams { 53596d4f267SLinus Torvalds if (access_ok(to, n)) { 536d0ef4c36SMarco Elver instrument_copy_to_user(to, from, n); 537ec6347bbSDan Williams n = copy_mc_to_user((__force void *) to, from, n); 5388780356eSDan Williams } 5398780356eSDan Williams return n; 5408780356eSDan Williams } 5418780356eSDan Williams 542ec6347bbSDan Williams static size_t copy_mc_pipe_to_iter(const void *addr, size_t bytes, 543ca146f6fSDan Williams struct iov_iter *i) 544ca146f6fSDan Williams { 5458fad7767SAl Viro size_t xfer = 0; 5468fad7767SAl Viro unsigned int off, chunk; 5478fad7767SAl Viro 5488fad7767SAl Viro if (unlikely(bytes > i->count)) 5498fad7767SAl Viro bytes = i->count; 5508fad7767SAl Viro if (unlikely(!bytes)) 5518fad7767SAl Viro return 0; 552ca146f6fSDan Williams 553ca146f6fSDan Williams if (!sanity(i)) 554ca146f6fSDan Williams return 0; 555ca146f6fSDan Williams 5568fad7767SAl Viro while (bytes) { 5578fad7767SAl Viro struct page *page = append_pipe(i, bytes, &off); 558ca146f6fSDan Williams unsigned long rem; 5598fad7767SAl Viro char *p; 5608fad7767SAl Viro 5618fad7767SAl Viro if (!page) 5628fad7767SAl Viro break; 5638fad7767SAl Viro chunk = min_t(size_t, bytes, PAGE_SIZE - off); 5648fad7767SAl Viro p = kmap_local_page(page); 5652a510a74SAl Viro rem = copy_mc_to_kernel(p + off, addr + xfer, chunk); 5662a510a74SAl Viro chunk -= rem; 5672a510a74SAl Viro kunmap_local(p); 5682a510a74SAl Viro xfer += chunk; 5698fad7767SAl Viro bytes -= chunk; 570c3497fd0SAl Viro if (rem) { 5718fad7767SAl Viro iov_iter_revert(i, rem); 572ca146f6fSDan Williams break; 573c3497fd0SAl Viro } 5742a510a74SAl Viro } 575ca146f6fSDan Williams return xfer; 576ca146f6fSDan Williams } 577ca146f6fSDan Williams 578bf3eeb9bSDan Williams /** 579ec6347bbSDan Williams * _copy_mc_to_iter - copy to iter with source memory error exception handling 580bf3eeb9bSDan Williams * @addr: source kernel address 581bf3eeb9bSDan Williams * @bytes: total transfer length 58244e55997SRandy Dunlap * @i: destination iterator 583bf3eeb9bSDan Williams * 584ec6347bbSDan Williams * The pmem driver deploys this for the dax operation 585ec6347bbSDan Williams * (dax_copy_to_iter()) for dax reads (bypass page-cache and the 586ec6347bbSDan Williams * block-layer). Upon #MC read(2) aborts and returns EIO or the bytes 587ec6347bbSDan Williams * successfully copied. 588bf3eeb9bSDan Williams * 589ec6347bbSDan Williams * The main differences between this and typical _copy_to_iter(). 590bf3eeb9bSDan Williams * 591bf3eeb9bSDan Williams * * Typical tail/residue handling after a fault retries the copy 592bf3eeb9bSDan Williams * byte-by-byte until the fault happens again. Re-triggering machine 593bf3eeb9bSDan Williams * checks is potentially fatal so the implementation uses source 594bf3eeb9bSDan Williams * alignment and poison alignment assumptions to avoid re-triggering 595bf3eeb9bSDan Williams * hardware exceptions. 596bf3eeb9bSDan Williams * 597bf3eeb9bSDan Williams * * ITER_KVEC, ITER_PIPE, and ITER_BVEC can return short copies. 598bf3eeb9bSDan Williams * Compare to copy_to_iter() where only ITER_IOVEC attempts might return 599bf3eeb9bSDan Williams * a short copy. 60044e55997SRandy Dunlap * 60144e55997SRandy Dunlap * Return: number of bytes copied (may be %0) 602bf3eeb9bSDan Williams */ 603ec6347bbSDan Williams size_t _copy_mc_to_iter(const void *addr, size_t bytes, struct iov_iter *i) 6048780356eSDan Williams { 605a41dad90SAl Viro if (WARN_ON_ONCE(i->data_source)) 606a41dad90SAl Viro return 0; 60700e23707SDavid Howells if (unlikely(iov_iter_is_pipe(i))) 608ec6347bbSDan Williams return copy_mc_pipe_to_iter(addr, bytes, i); 609fcb14cb1SAl Viro if (user_backed_iter(i)) 6108780356eSDan Williams might_fault(); 6117baa5099SAl Viro __iterate_and_advance(i, bytes, base, len, off, 6127baa5099SAl Viro copyout_mc(base, addr + off, len), 6137baa5099SAl Viro copy_mc_to_kernel(base, addr + off, len) 6148780356eSDan Williams ) 6158780356eSDan Williams 6168780356eSDan Williams return bytes; 6178780356eSDan Williams } 618ec6347bbSDan Williams EXPORT_SYMBOL_GPL(_copy_mc_to_iter); 619ec6347bbSDan Williams #endif /* CONFIG_ARCH_HAS_COPY_MC */ 6208780356eSDan Williams 621aa28de27SAl Viro size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i) 622d879cb83SAl Viro { 623a41dad90SAl Viro if (WARN_ON_ONCE(!i->data_source)) 624241699cdSAl Viro return 0; 625a41dad90SAl Viro 626fcb14cb1SAl Viro if (user_backed_iter(i)) 62709fc68dcSAl Viro might_fault(); 6287baa5099SAl Viro iterate_and_advance(i, bytes, base, len, off, 6297baa5099SAl Viro copyin(addr + off, base, len), 6307baa5099SAl Viro memcpy(addr + off, base, len) 631d879cb83SAl Viro ) 632d879cb83SAl Viro 633d879cb83SAl Viro return bytes; 634d879cb83SAl Viro } 635aa28de27SAl Viro EXPORT_SYMBOL(_copy_from_iter); 636d879cb83SAl Viro 637aa28de27SAl Viro size_t _copy_from_iter_nocache(void *addr, size_t bytes, struct iov_iter *i) 638d879cb83SAl Viro { 639a41dad90SAl Viro if (WARN_ON_ONCE(!i->data_source)) 640241699cdSAl Viro return 0; 641a41dad90SAl Viro 6427baa5099SAl Viro iterate_and_advance(i, bytes, base, len, off, 6437baa5099SAl Viro __copy_from_user_inatomic_nocache(addr + off, base, len), 6447baa5099SAl Viro memcpy(addr + off, base, len) 645d879cb83SAl Viro ) 646d879cb83SAl Viro 647d879cb83SAl Viro return bytes; 648d879cb83SAl Viro } 649aa28de27SAl Viro EXPORT_SYMBOL(_copy_from_iter_nocache); 650d879cb83SAl Viro 6510aed55afSDan Williams #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE 652abd08d7dSDan Williams /** 653abd08d7dSDan Williams * _copy_from_iter_flushcache - write destination through cpu cache 654abd08d7dSDan Williams * @addr: destination kernel address 655abd08d7dSDan Williams * @bytes: total transfer length 65644e55997SRandy Dunlap * @i: source iterator 657abd08d7dSDan Williams * 658abd08d7dSDan Williams * The pmem driver arranges for filesystem-dax to use this facility via 659abd08d7dSDan Williams * dax_copy_from_iter() for ensuring that writes to persistent memory 660abd08d7dSDan Williams * are flushed through the CPU cache. It is differentiated from 661abd08d7dSDan Williams * _copy_from_iter_nocache() in that guarantees all data is flushed for 662abd08d7dSDan Williams * all iterator types. The _copy_from_iter_nocache() only attempts to 663abd08d7dSDan Williams * bypass the cache for the ITER_IOVEC case, and on some archs may use 664abd08d7dSDan Williams * instructions that strand dirty-data in the cache. 66544e55997SRandy Dunlap * 66644e55997SRandy Dunlap * Return: number of bytes copied (may be %0) 667abd08d7dSDan Williams */ 6686a37e940SLinus Torvalds size_t _copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i) 6690aed55afSDan Williams { 670a41dad90SAl Viro if (WARN_ON_ONCE(!i->data_source)) 6710aed55afSDan Williams return 0; 672a41dad90SAl Viro 6737baa5099SAl Viro iterate_and_advance(i, bytes, base, len, off, 6747baa5099SAl Viro __copy_from_user_flushcache(addr + off, base, len), 6757baa5099SAl Viro memcpy_flushcache(addr + off, base, len) 6760aed55afSDan Williams ) 6770aed55afSDan Williams 6780aed55afSDan Williams return bytes; 6790aed55afSDan Williams } 6806a37e940SLinus Torvalds EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache); 6810aed55afSDan Williams #endif 6820aed55afSDan Williams 68372e809edSAl Viro static inline bool page_copy_sane(struct page *page, size_t offset, size_t n) 68472e809edSAl Viro { 6856daef95bSEric Dumazet struct page *head; 6866daef95bSEric Dumazet size_t v = n + offset; 6876daef95bSEric Dumazet 6886daef95bSEric Dumazet /* 6896daef95bSEric Dumazet * The general case needs to access the page order in order 6906daef95bSEric Dumazet * to compute the page size. 6916daef95bSEric Dumazet * However, we mostly deal with order-0 pages and thus can 6926daef95bSEric Dumazet * avoid a possible cache line miss for requests that fit all 6936daef95bSEric Dumazet * page orders. 6946daef95bSEric Dumazet */ 6956daef95bSEric Dumazet if (n <= v && v <= PAGE_SIZE) 6966daef95bSEric Dumazet return true; 6976daef95bSEric Dumazet 6986daef95bSEric Dumazet head = compound_head(page); 6996daef95bSEric Dumazet v += (page - head) << PAGE_SHIFT; 700a90bcb86SPetar Penkov 70140a86061SAl Viro if (WARN_ON(n > v || v > page_size(head))) 70272e809edSAl Viro return false; 70340a86061SAl Viro return true; 70472e809edSAl Viro } 705cbbd26b8SAl Viro 70608aa6479SAl Viro size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes, 70708aa6479SAl Viro struct iov_iter *i) 70808aa6479SAl Viro { 70908aa6479SAl Viro size_t res = 0; 71040a86061SAl Viro if (!page_copy_sane(page, offset, bytes)) 71108aa6479SAl Viro return 0; 712a41dad90SAl Viro if (WARN_ON_ONCE(i->data_source)) 713a41dad90SAl Viro return 0; 714f0f6b614SAl Viro if (unlikely(iov_iter_is_pipe(i))) 715f0f6b614SAl Viro return copy_page_to_iter_pipe(page, offset, bytes, i); 71608aa6479SAl Viro page += offset / PAGE_SIZE; // first subpage 71708aa6479SAl Viro offset %= PAGE_SIZE; 71808aa6479SAl Viro while (1) { 719f0f6b614SAl Viro void *kaddr = kmap_local_page(page); 720f0f6b614SAl Viro size_t n = min(bytes, (size_t)PAGE_SIZE - offset); 721f0f6b614SAl Viro n = _copy_to_iter(kaddr + offset, n, i); 722f0f6b614SAl Viro kunmap_local(kaddr); 72308aa6479SAl Viro res += n; 72408aa6479SAl Viro bytes -= n; 72508aa6479SAl Viro if (!bytes || !n) 72608aa6479SAl Viro break; 72708aa6479SAl Viro offset += n; 72808aa6479SAl Viro if (offset == PAGE_SIZE) { 72908aa6479SAl Viro page++; 73008aa6479SAl Viro offset = 0; 73108aa6479SAl Viro } 73208aa6479SAl Viro } 73308aa6479SAl Viro return res; 73408aa6479SAl Viro } 735d879cb83SAl Viro EXPORT_SYMBOL(copy_page_to_iter); 736d879cb83SAl Viro 737d879cb83SAl Viro size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes, 738d879cb83SAl Viro struct iov_iter *i) 739d879cb83SAl Viro { 740c03f05f1SAl Viro size_t res = 0; 741c03f05f1SAl Viro if (!page_copy_sane(page, offset, bytes)) 74228f38db7SAl Viro return 0; 743c03f05f1SAl Viro page += offset / PAGE_SIZE; // first subpage 744c03f05f1SAl Viro offset %= PAGE_SIZE; 745c03f05f1SAl Viro while (1) { 746c03f05f1SAl Viro void *kaddr = kmap_local_page(page); 747c03f05f1SAl Viro size_t n = min(bytes, (size_t)PAGE_SIZE - offset); 748c03f05f1SAl Viro n = _copy_from_iter(kaddr + offset, n, i); 749c03f05f1SAl Viro kunmap_local(kaddr); 750c03f05f1SAl Viro res += n; 751c03f05f1SAl Viro bytes -= n; 752c03f05f1SAl Viro if (!bytes || !n) 753c03f05f1SAl Viro break; 754c03f05f1SAl Viro offset += n; 755c03f05f1SAl Viro if (offset == PAGE_SIZE) { 756c03f05f1SAl Viro page++; 757c03f05f1SAl Viro offset = 0; 758c03f05f1SAl Viro } 759c03f05f1SAl Viro } 760c03f05f1SAl Viro return res; 761d879cb83SAl Viro } 762d879cb83SAl Viro EXPORT_SYMBOL(copy_page_from_iter); 763d879cb83SAl Viro 764241699cdSAl Viro static size_t pipe_zero(size_t bytes, struct iov_iter *i) 765241699cdSAl Viro { 7668fad7767SAl Viro unsigned int chunk, off; 7678fad7767SAl Viro 7688fad7767SAl Viro if (unlikely(bytes > i->count)) 7698fad7767SAl Viro bytes = i->count; 7708fad7767SAl Viro if (unlikely(!bytes)) 7718fad7767SAl Viro return 0; 772241699cdSAl Viro 773241699cdSAl Viro if (!sanity(i)) 774241699cdSAl Viro return 0; 775241699cdSAl Viro 7768fad7767SAl Viro for (size_t n = bytes; n; n -= chunk) { 7778fad7767SAl Viro struct page *page = append_pipe(i, n, &off); 7788fad7767SAl Viro char *p; 779241699cdSAl Viro 7808fad7767SAl Viro if (!page) 7818fad7767SAl Viro return bytes - n; 7828fad7767SAl Viro chunk = min_t(size_t, n, PAGE_SIZE - off); 7838fad7767SAl Viro p = kmap_local_page(page); 784893839fdSAl Viro memset(p + off, 0, chunk); 785893839fdSAl Viro kunmap_local(p); 7868fad7767SAl Viro } 787241699cdSAl Viro return bytes; 788241699cdSAl Viro } 789241699cdSAl Viro 790d879cb83SAl Viro size_t iov_iter_zero(size_t bytes, struct iov_iter *i) 791d879cb83SAl Viro { 79200e23707SDavid Howells if (unlikely(iov_iter_is_pipe(i))) 793241699cdSAl Viro return pipe_zero(bytes, i); 7947baa5099SAl Viro iterate_and_advance(i, bytes, base, len, count, 7957baa5099SAl Viro clear_user(base, len), 7967baa5099SAl Viro memset(base, 0, len) 797d879cb83SAl Viro ) 798d879cb83SAl Viro 799d879cb83SAl Viro return bytes; 800d879cb83SAl Viro } 801d879cb83SAl Viro EXPORT_SYMBOL(iov_iter_zero); 802d879cb83SAl Viro 803f0b65f39SAl Viro size_t copy_page_from_iter_atomic(struct page *page, unsigned offset, size_t bytes, 804f0b65f39SAl Viro struct iov_iter *i) 805d879cb83SAl Viro { 806d879cb83SAl Viro char *kaddr = kmap_atomic(page), *p = kaddr + offset; 80740a86061SAl Viro if (!page_copy_sane(page, offset, bytes)) { 80872e809edSAl Viro kunmap_atomic(kaddr); 80972e809edSAl Viro return 0; 81072e809edSAl Viro } 811a41dad90SAl Viro if (WARN_ON_ONCE(!i->data_source)) { 812241699cdSAl Viro kunmap_atomic(kaddr); 813241699cdSAl Viro return 0; 814241699cdSAl Viro } 8157baa5099SAl Viro iterate_and_advance(i, bytes, base, len, off, 8167baa5099SAl Viro copyin(p + off, base, len), 8177baa5099SAl Viro memcpy(p + off, base, len) 818d879cb83SAl Viro ) 819d879cb83SAl Viro kunmap_atomic(kaddr); 820d879cb83SAl Viro return bytes; 821d879cb83SAl Viro } 822f0b65f39SAl Viro EXPORT_SYMBOL(copy_page_from_iter_atomic); 823d879cb83SAl Viro 824b9dc6f65SAl Viro static void pipe_advance(struct iov_iter *i, size_t size) 825b9dc6f65SAl Viro { 826b9dc6f65SAl Viro struct pipe_inode_info *pipe = i->pipe; 82710f525a8SAl Viro int off = i->last_offset; 8288cefc107SDavid Howells 8292c855de9SAl Viro if (!off && !size) { 8302c855de9SAl Viro pipe_discard_from(pipe, i->start_head); // discard everything 8312c855de9SAl Viro return; 832b9dc6f65SAl Viro } 833b9dc6f65SAl Viro i->count -= size; 8342c855de9SAl Viro while (1) { 8352c855de9SAl Viro struct pipe_buffer *buf = pipe_buf(pipe, i->head); 8362c855de9SAl Viro if (off) /* make it relative to the beginning of buffer */ 83710f525a8SAl Viro size += abs(off) - buf->offset; 8382c855de9SAl Viro if (size <= buf->len) { 8392c855de9SAl Viro buf->len = size; 84010f525a8SAl Viro i->last_offset = last_offset(buf); 8412c855de9SAl Viro break; 8422c855de9SAl Viro } 8432c855de9SAl Viro size -= buf->len; 8442c855de9SAl Viro i->head++; 8452c855de9SAl Viro off = 0; 8462c855de9SAl Viro } 8472c855de9SAl Viro pipe_discard_from(pipe, i->head + 1); // discard everything past this one 848241699cdSAl Viro } 849241699cdSAl Viro 85054c8195bSPavel Begunkov static void iov_iter_bvec_advance(struct iov_iter *i, size_t size) 85154c8195bSPavel Begunkov { 85218fa9af7SAl Viro const struct bio_vec *bvec, *end; 85354c8195bSPavel Begunkov 85418fa9af7SAl Viro if (!i->count) 85518fa9af7SAl Viro return; 85618fa9af7SAl Viro i->count -= size; 85754c8195bSPavel Begunkov 85818fa9af7SAl Viro size += i->iov_offset; 85918fa9af7SAl Viro 86018fa9af7SAl Viro for (bvec = i->bvec, end = bvec + i->nr_segs; bvec < end; bvec++) { 86118fa9af7SAl Viro if (likely(size < bvec->bv_len)) 86218fa9af7SAl Viro break; 86318fa9af7SAl Viro size -= bvec->bv_len; 86418fa9af7SAl Viro } 86518fa9af7SAl Viro i->iov_offset = size; 86618fa9af7SAl Viro i->nr_segs -= bvec - i->bvec; 86718fa9af7SAl Viro i->bvec = bvec; 86854c8195bSPavel Begunkov } 86954c8195bSPavel Begunkov 870185ac4d4SAl Viro static void iov_iter_iovec_advance(struct iov_iter *i, size_t size) 871185ac4d4SAl Viro { 872185ac4d4SAl Viro const struct iovec *iov, *end; 873185ac4d4SAl Viro 874185ac4d4SAl Viro if (!i->count) 875185ac4d4SAl Viro return; 876185ac4d4SAl Viro i->count -= size; 877185ac4d4SAl Viro 878185ac4d4SAl Viro size += i->iov_offset; // from beginning of current segment 879de4f5fedSJens Axboe for (iov = iter_iov(i), end = iov + i->nr_segs; iov < end; iov++) { 880185ac4d4SAl Viro if (likely(size < iov->iov_len)) 881185ac4d4SAl Viro break; 882185ac4d4SAl Viro size -= iov->iov_len; 883185ac4d4SAl Viro } 884185ac4d4SAl Viro i->iov_offset = size; 885de4f5fedSJens Axboe i->nr_segs -= iov - iter_iov(i); 886de4f5fedSJens Axboe i->__iov = iov; 887185ac4d4SAl Viro } 888185ac4d4SAl Viro 889d879cb83SAl Viro void iov_iter_advance(struct iov_iter *i, size_t size) 890d879cb83SAl Viro { 8913b3fc051SAl Viro if (unlikely(i->count < size)) 8923b3fc051SAl Viro size = i->count; 893fcb14cb1SAl Viro if (likely(iter_is_ubuf(i)) || unlikely(iov_iter_is_xarray(i))) { 894fcb14cb1SAl Viro i->iov_offset += size; 895fcb14cb1SAl Viro i->count -= size; 896fcb14cb1SAl Viro } else if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) { 897185ac4d4SAl Viro /* iovec and kvec have identical layouts */ 898185ac4d4SAl Viro iov_iter_iovec_advance(i, size); 899185ac4d4SAl Viro } else if (iov_iter_is_bvec(i)) { 900185ac4d4SAl Viro iov_iter_bvec_advance(i, size); 901185ac4d4SAl Viro } else if (iov_iter_is_pipe(i)) { 902241699cdSAl Viro pipe_advance(i, size); 903185ac4d4SAl Viro } else if (iov_iter_is_discard(i)) { 904185ac4d4SAl Viro i->count -= size; 9057ff50620SDavid Howells } 906d879cb83SAl Viro } 907d879cb83SAl Viro EXPORT_SYMBOL(iov_iter_advance); 908d879cb83SAl Viro 90927c0e374SAl Viro void iov_iter_revert(struct iov_iter *i, size_t unroll) 91027c0e374SAl Viro { 91127c0e374SAl Viro if (!unroll) 91227c0e374SAl Viro return; 9135b47d59aSAl Viro if (WARN_ON(unroll > MAX_RW_COUNT)) 9145b47d59aSAl Viro return; 91527c0e374SAl Viro i->count += unroll; 91600e23707SDavid Howells if (unlikely(iov_iter_is_pipe(i))) { 91727c0e374SAl Viro struct pipe_inode_info *pipe = i->pipe; 91892acdc4fSAl Viro unsigned int head = pipe->head; 91992acdc4fSAl Viro 92092acdc4fSAl Viro while (head > i->start_head) { 92192acdc4fSAl Viro struct pipe_buffer *b = pipe_buf(pipe, --head); 92292acdc4fSAl Viro if (unroll < b->len) { 92392acdc4fSAl Viro b->len -= unroll; 92410f525a8SAl Viro i->last_offset = last_offset(b); 92592acdc4fSAl Viro i->head = head; 92692acdc4fSAl Viro return; 92727c0e374SAl Viro } 92892acdc4fSAl Viro unroll -= b->len; 92992acdc4fSAl Viro pipe_buf_release(pipe, b); 93092acdc4fSAl Viro pipe->head--; 93127c0e374SAl Viro } 93210f525a8SAl Viro i->last_offset = 0; 93392acdc4fSAl Viro i->head = head; 93427c0e374SAl Viro return; 93527c0e374SAl Viro } 9369ea9ce04SDavid Howells if (unlikely(iov_iter_is_discard(i))) 9379ea9ce04SDavid Howells return; 93827c0e374SAl Viro if (unroll <= i->iov_offset) { 93927c0e374SAl Viro i->iov_offset -= unroll; 94027c0e374SAl Viro return; 94127c0e374SAl Viro } 94227c0e374SAl Viro unroll -= i->iov_offset; 943fcb14cb1SAl Viro if (iov_iter_is_xarray(i) || iter_is_ubuf(i)) { 9447ff50620SDavid Howells BUG(); /* We should never go beyond the start of the specified 9457ff50620SDavid Howells * range since we might then be straying into pages that 9467ff50620SDavid Howells * aren't pinned. 9477ff50620SDavid Howells */ 9487ff50620SDavid Howells } else if (iov_iter_is_bvec(i)) { 94927c0e374SAl Viro const struct bio_vec *bvec = i->bvec; 95027c0e374SAl Viro while (1) { 95127c0e374SAl Viro size_t n = (--bvec)->bv_len; 95227c0e374SAl Viro i->nr_segs++; 95327c0e374SAl Viro if (unroll <= n) { 95427c0e374SAl Viro i->bvec = bvec; 95527c0e374SAl Viro i->iov_offset = n - unroll; 95627c0e374SAl Viro return; 95727c0e374SAl Viro } 95827c0e374SAl Viro unroll -= n; 95927c0e374SAl Viro } 96027c0e374SAl Viro } else { /* same logics for iovec and kvec */ 961de4f5fedSJens Axboe const struct iovec *iov = iter_iov(i); 96227c0e374SAl Viro while (1) { 96327c0e374SAl Viro size_t n = (--iov)->iov_len; 96427c0e374SAl Viro i->nr_segs++; 96527c0e374SAl Viro if (unroll <= n) { 966de4f5fedSJens Axboe i->__iov = iov; 96727c0e374SAl Viro i->iov_offset = n - unroll; 96827c0e374SAl Viro return; 96927c0e374SAl Viro } 97027c0e374SAl Viro unroll -= n; 97127c0e374SAl Viro } 97227c0e374SAl Viro } 97327c0e374SAl Viro } 97427c0e374SAl Viro EXPORT_SYMBOL(iov_iter_revert); 97527c0e374SAl Viro 976d879cb83SAl Viro /* 977d879cb83SAl Viro * Return the count of just the current iov_iter segment. 978d879cb83SAl Viro */ 979d879cb83SAl Viro size_t iov_iter_single_seg_count(const struct iov_iter *i) 980d879cb83SAl Viro { 98128f38db7SAl Viro if (i->nr_segs > 1) { 98228f38db7SAl Viro if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) 983de4f5fedSJens Axboe return min(i->count, iter_iov(i)->iov_len - i->iov_offset); 9847ff50620SDavid Howells if (iov_iter_is_bvec(i)) 985d879cb83SAl Viro return min(i->count, i->bvec->bv_len - i->iov_offset); 98628f38db7SAl Viro } 98728f38db7SAl Viro return i->count; 988d879cb83SAl Viro } 989d879cb83SAl Viro EXPORT_SYMBOL(iov_iter_single_seg_count); 990d879cb83SAl Viro 991aa563d7bSDavid Howells void iov_iter_kvec(struct iov_iter *i, unsigned int direction, 992d879cb83SAl Viro const struct kvec *kvec, unsigned long nr_segs, 993d879cb83SAl Viro size_t count) 994d879cb83SAl Viro { 995aa563d7bSDavid Howells WARN_ON(direction & ~(READ | WRITE)); 9968cd54c1cSAl Viro *i = (struct iov_iter){ 9978cd54c1cSAl Viro .iter_type = ITER_KVEC, 9988cd54c1cSAl Viro .data_source = direction, 9998cd54c1cSAl Viro .kvec = kvec, 10008cd54c1cSAl Viro .nr_segs = nr_segs, 10018cd54c1cSAl Viro .iov_offset = 0, 10028cd54c1cSAl Viro .count = count 10038cd54c1cSAl Viro }; 1004d879cb83SAl Viro } 1005d879cb83SAl Viro EXPORT_SYMBOL(iov_iter_kvec); 1006d879cb83SAl Viro 1007aa563d7bSDavid Howells void iov_iter_bvec(struct iov_iter *i, unsigned int direction, 1008d879cb83SAl Viro const struct bio_vec *bvec, unsigned long nr_segs, 1009d879cb83SAl Viro size_t count) 1010d879cb83SAl Viro { 1011aa563d7bSDavid Howells WARN_ON(direction & ~(READ | WRITE)); 10128cd54c1cSAl Viro *i = (struct iov_iter){ 10138cd54c1cSAl Viro .iter_type = ITER_BVEC, 10148cd54c1cSAl Viro .data_source = direction, 10158cd54c1cSAl Viro .bvec = bvec, 10168cd54c1cSAl Viro .nr_segs = nr_segs, 10178cd54c1cSAl Viro .iov_offset = 0, 10188cd54c1cSAl Viro .count = count 10198cd54c1cSAl Viro }; 1020d879cb83SAl Viro } 1021d879cb83SAl Viro EXPORT_SYMBOL(iov_iter_bvec); 1022d879cb83SAl Viro 1023aa563d7bSDavid Howells void iov_iter_pipe(struct iov_iter *i, unsigned int direction, 1024241699cdSAl Viro struct pipe_inode_info *pipe, 1025241699cdSAl Viro size_t count) 1026241699cdSAl Viro { 1027aa563d7bSDavid Howells BUG_ON(direction != READ); 10288cefc107SDavid Howells WARN_ON(pipe_full(pipe->head, pipe->tail, pipe->ring_size)); 10298cd54c1cSAl Viro *i = (struct iov_iter){ 10308cd54c1cSAl Viro .iter_type = ITER_PIPE, 10318cd54c1cSAl Viro .data_source = false, 10328cd54c1cSAl Viro .pipe = pipe, 10338cd54c1cSAl Viro .head = pipe->head, 10348cd54c1cSAl Viro .start_head = pipe->head, 103510f525a8SAl Viro .last_offset = 0, 10368cd54c1cSAl Viro .count = count 10378cd54c1cSAl Viro }; 1038241699cdSAl Viro } 1039241699cdSAl Viro EXPORT_SYMBOL(iov_iter_pipe); 1040241699cdSAl Viro 10419ea9ce04SDavid Howells /** 10427ff50620SDavid Howells * iov_iter_xarray - Initialise an I/O iterator to use the pages in an xarray 10437ff50620SDavid Howells * @i: The iterator to initialise. 10447ff50620SDavid Howells * @direction: The direction of the transfer. 10457ff50620SDavid Howells * @xarray: The xarray to access. 10467ff50620SDavid Howells * @start: The start file position. 10477ff50620SDavid Howells * @count: The size of the I/O buffer in bytes. 10487ff50620SDavid Howells * 10497ff50620SDavid Howells * Set up an I/O iterator to either draw data out of the pages attached to an 10507ff50620SDavid Howells * inode or to inject data into those pages. The pages *must* be prevented 10517ff50620SDavid Howells * from evaporation, either by taking a ref on them or locking them by the 10527ff50620SDavid Howells * caller. 10537ff50620SDavid Howells */ 10547ff50620SDavid Howells void iov_iter_xarray(struct iov_iter *i, unsigned int direction, 10557ff50620SDavid Howells struct xarray *xarray, loff_t start, size_t count) 10567ff50620SDavid Howells { 10577ff50620SDavid Howells BUG_ON(direction & ~1); 10588cd54c1cSAl Viro *i = (struct iov_iter) { 10598cd54c1cSAl Viro .iter_type = ITER_XARRAY, 10608cd54c1cSAl Viro .data_source = direction, 10618cd54c1cSAl Viro .xarray = xarray, 10628cd54c1cSAl Viro .xarray_start = start, 10638cd54c1cSAl Viro .count = count, 10648cd54c1cSAl Viro .iov_offset = 0 10658cd54c1cSAl Viro }; 10667ff50620SDavid Howells } 10677ff50620SDavid Howells EXPORT_SYMBOL(iov_iter_xarray); 10687ff50620SDavid Howells 10697ff50620SDavid Howells /** 10709ea9ce04SDavid Howells * iov_iter_discard - Initialise an I/O iterator that discards data 10719ea9ce04SDavid Howells * @i: The iterator to initialise. 10729ea9ce04SDavid Howells * @direction: The direction of the transfer. 10739ea9ce04SDavid Howells * @count: The size of the I/O buffer in bytes. 10749ea9ce04SDavid Howells * 10759ea9ce04SDavid Howells * Set up an I/O iterator that just discards everything that's written to it. 10769ea9ce04SDavid Howells * It's only available as a READ iterator. 10779ea9ce04SDavid Howells */ 10789ea9ce04SDavid Howells void iov_iter_discard(struct iov_iter *i, unsigned int direction, size_t count) 10799ea9ce04SDavid Howells { 10809ea9ce04SDavid Howells BUG_ON(direction != READ); 10818cd54c1cSAl Viro *i = (struct iov_iter){ 10828cd54c1cSAl Viro .iter_type = ITER_DISCARD, 10838cd54c1cSAl Viro .data_source = false, 10848cd54c1cSAl Viro .count = count, 10858cd54c1cSAl Viro .iov_offset = 0 10868cd54c1cSAl Viro }; 10879ea9ce04SDavid Howells } 10889ea9ce04SDavid Howells EXPORT_SYMBOL(iov_iter_discard); 10899ea9ce04SDavid Howells 1090cfa320f7SKeith Busch static bool iov_iter_aligned_iovec(const struct iov_iter *i, unsigned addr_mask, 1091cfa320f7SKeith Busch unsigned len_mask) 1092cfa320f7SKeith Busch { 1093cfa320f7SKeith Busch size_t size = i->count; 1094cfa320f7SKeith Busch size_t skip = i->iov_offset; 1095cfa320f7SKeith Busch unsigned k; 1096cfa320f7SKeith Busch 1097cfa320f7SKeith Busch for (k = 0; k < i->nr_segs; k++, skip = 0) { 1098de4f5fedSJens Axboe const struct iovec *iov = iter_iov(i) + k; 1099de4f5fedSJens Axboe size_t len = iov->iov_len - skip; 1100cfa320f7SKeith Busch 1101cfa320f7SKeith Busch if (len > size) 1102cfa320f7SKeith Busch len = size; 1103cfa320f7SKeith Busch if (len & len_mask) 1104cfa320f7SKeith Busch return false; 1105de4f5fedSJens Axboe if ((unsigned long)(iov->iov_base + skip) & addr_mask) 1106cfa320f7SKeith Busch return false; 1107cfa320f7SKeith Busch 1108cfa320f7SKeith Busch size -= len; 1109cfa320f7SKeith Busch if (!size) 1110cfa320f7SKeith Busch break; 1111cfa320f7SKeith Busch } 1112cfa320f7SKeith Busch return true; 1113cfa320f7SKeith Busch } 1114cfa320f7SKeith Busch 1115cfa320f7SKeith Busch static bool iov_iter_aligned_bvec(const struct iov_iter *i, unsigned addr_mask, 1116cfa320f7SKeith Busch unsigned len_mask) 1117cfa320f7SKeith Busch { 1118cfa320f7SKeith Busch size_t size = i->count; 1119cfa320f7SKeith Busch unsigned skip = i->iov_offset; 1120cfa320f7SKeith Busch unsigned k; 1121cfa320f7SKeith Busch 1122cfa320f7SKeith Busch for (k = 0; k < i->nr_segs; k++, skip = 0) { 1123cfa320f7SKeith Busch size_t len = i->bvec[k].bv_len - skip; 1124cfa320f7SKeith Busch 1125cfa320f7SKeith Busch if (len > size) 1126cfa320f7SKeith Busch len = size; 1127cfa320f7SKeith Busch if (len & len_mask) 1128cfa320f7SKeith Busch return false; 1129cfa320f7SKeith Busch if ((unsigned long)(i->bvec[k].bv_offset + skip) & addr_mask) 1130cfa320f7SKeith Busch return false; 1131cfa320f7SKeith Busch 1132cfa320f7SKeith Busch size -= len; 1133cfa320f7SKeith Busch if (!size) 1134cfa320f7SKeith Busch break; 1135cfa320f7SKeith Busch } 1136cfa320f7SKeith Busch return true; 1137cfa320f7SKeith Busch } 1138cfa320f7SKeith Busch 1139cfa320f7SKeith Busch /** 1140cfa320f7SKeith Busch * iov_iter_is_aligned() - Check if the addresses and lengths of each segments 1141cfa320f7SKeith Busch * are aligned to the parameters. 1142cfa320f7SKeith Busch * 1143cfa320f7SKeith Busch * @i: &struct iov_iter to restore 1144cfa320f7SKeith Busch * @addr_mask: bit mask to check against the iov element's addresses 1145cfa320f7SKeith Busch * @len_mask: bit mask to check against the iov element's lengths 1146cfa320f7SKeith Busch * 1147cfa320f7SKeith Busch * Return: false if any addresses or lengths intersect with the provided masks 1148cfa320f7SKeith Busch */ 1149cfa320f7SKeith Busch bool iov_iter_is_aligned(const struct iov_iter *i, unsigned addr_mask, 1150cfa320f7SKeith Busch unsigned len_mask) 1151cfa320f7SKeith Busch { 1152fcb14cb1SAl Viro if (likely(iter_is_ubuf(i))) { 1153fcb14cb1SAl Viro if (i->count & len_mask) 1154fcb14cb1SAl Viro return false; 1155fcb14cb1SAl Viro if ((unsigned long)(i->ubuf + i->iov_offset) & addr_mask) 1156fcb14cb1SAl Viro return false; 1157fcb14cb1SAl Viro return true; 1158fcb14cb1SAl Viro } 1159fcb14cb1SAl Viro 1160cfa320f7SKeith Busch if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) 1161cfa320f7SKeith Busch return iov_iter_aligned_iovec(i, addr_mask, len_mask); 1162cfa320f7SKeith Busch 1163cfa320f7SKeith Busch if (iov_iter_is_bvec(i)) 1164cfa320f7SKeith Busch return iov_iter_aligned_bvec(i, addr_mask, len_mask); 1165cfa320f7SKeith Busch 1166cfa320f7SKeith Busch if (iov_iter_is_pipe(i)) { 1167cfa320f7SKeith Busch size_t size = i->count; 1168cfa320f7SKeith Busch 1169cfa320f7SKeith Busch if (size & len_mask) 1170cfa320f7SKeith Busch return false; 117110f525a8SAl Viro if (size && i->last_offset > 0) { 117210f525a8SAl Viro if (i->last_offset & addr_mask) 1173cfa320f7SKeith Busch return false; 1174cfa320f7SKeith Busch } 1175cfa320f7SKeith Busch 1176cfa320f7SKeith Busch return true; 1177cfa320f7SKeith Busch } 1178cfa320f7SKeith Busch 1179cfa320f7SKeith Busch if (iov_iter_is_xarray(i)) { 1180cfa320f7SKeith Busch if (i->count & len_mask) 1181cfa320f7SKeith Busch return false; 1182cfa320f7SKeith Busch if ((i->xarray_start + i->iov_offset) & addr_mask) 1183cfa320f7SKeith Busch return false; 1184cfa320f7SKeith Busch } 1185cfa320f7SKeith Busch 1186cfa320f7SKeith Busch return true; 1187cfa320f7SKeith Busch } 1188cfa320f7SKeith Busch EXPORT_SYMBOL_GPL(iov_iter_is_aligned); 1189cfa320f7SKeith Busch 11909221d2e3SAl Viro static unsigned long iov_iter_alignment_iovec(const struct iov_iter *i) 1191d879cb83SAl Viro { 1192d879cb83SAl Viro unsigned long res = 0; 1193d879cb83SAl Viro size_t size = i->count; 11949221d2e3SAl Viro size_t skip = i->iov_offset; 11959221d2e3SAl Viro unsigned k; 1196d879cb83SAl Viro 11979221d2e3SAl Viro for (k = 0; k < i->nr_segs; k++, skip = 0) { 1198de4f5fedSJens Axboe const struct iovec *iov = iter_iov(i) + k; 1199de4f5fedSJens Axboe size_t len = iov->iov_len - skip; 12009221d2e3SAl Viro if (len) { 1201de4f5fedSJens Axboe res |= (unsigned long)iov->iov_base + skip; 12029221d2e3SAl Viro if (len > size) 12039221d2e3SAl Viro len = size; 12049221d2e3SAl Viro res |= len; 12059221d2e3SAl Viro size -= len; 12069221d2e3SAl Viro if (!size) 12079221d2e3SAl Viro break; 12089221d2e3SAl Viro } 12099221d2e3SAl Viro } 12109221d2e3SAl Viro return res; 12119221d2e3SAl Viro } 12129221d2e3SAl Viro 12139221d2e3SAl Viro static unsigned long iov_iter_alignment_bvec(const struct iov_iter *i) 12149221d2e3SAl Viro { 12159221d2e3SAl Viro unsigned res = 0; 12169221d2e3SAl Viro size_t size = i->count; 12179221d2e3SAl Viro unsigned skip = i->iov_offset; 12189221d2e3SAl Viro unsigned k; 12199221d2e3SAl Viro 12209221d2e3SAl Viro for (k = 0; k < i->nr_segs; k++, skip = 0) { 12219221d2e3SAl Viro size_t len = i->bvec[k].bv_len - skip; 12229221d2e3SAl Viro res |= (unsigned long)i->bvec[k].bv_offset + skip; 12239221d2e3SAl Viro if (len > size) 12249221d2e3SAl Viro len = size; 12259221d2e3SAl Viro res |= len; 12269221d2e3SAl Viro size -= len; 12279221d2e3SAl Viro if (!size) 12289221d2e3SAl Viro break; 12299221d2e3SAl Viro } 12309221d2e3SAl Viro return res; 12319221d2e3SAl Viro } 12329221d2e3SAl Viro 12339221d2e3SAl Viro unsigned long iov_iter_alignment(const struct iov_iter *i) 12349221d2e3SAl Viro { 1235fcb14cb1SAl Viro if (likely(iter_is_ubuf(i))) { 1236fcb14cb1SAl Viro size_t size = i->count; 1237fcb14cb1SAl Viro if (size) 1238fcb14cb1SAl Viro return ((unsigned long)i->ubuf + i->iov_offset) | size; 1239fcb14cb1SAl Viro return 0; 1240fcb14cb1SAl Viro } 1241fcb14cb1SAl Viro 12429221d2e3SAl Viro /* iovec and kvec have identical layouts */ 12439221d2e3SAl Viro if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) 12449221d2e3SAl Viro return iov_iter_alignment_iovec(i); 12459221d2e3SAl Viro 12469221d2e3SAl Viro if (iov_iter_is_bvec(i)) 12479221d2e3SAl Viro return iov_iter_alignment_bvec(i); 12489221d2e3SAl Viro 12499221d2e3SAl Viro if (iov_iter_is_pipe(i)) { 12509221d2e3SAl Viro size_t size = i->count; 1251e0ff126eSJan Kara 125210f525a8SAl Viro if (size && i->last_offset > 0) 125310f525a8SAl Viro return size | i->last_offset; 1254241699cdSAl Viro return size; 1255241699cdSAl Viro } 12569221d2e3SAl Viro 12579221d2e3SAl Viro if (iov_iter_is_xarray(i)) 12583d14ec1fSDavid Howells return (i->xarray_start + i->iov_offset) | i->count; 12599221d2e3SAl Viro 12609221d2e3SAl Viro return 0; 1261d879cb83SAl Viro } 1262d879cb83SAl Viro EXPORT_SYMBOL(iov_iter_alignment); 1263d879cb83SAl Viro 1264357f435dSAl Viro unsigned long iov_iter_gap_alignment(const struct iov_iter *i) 1265357f435dSAl Viro { 1266357f435dSAl Viro unsigned long res = 0; 1267610c7a71SAl Viro unsigned long v = 0; 1268357f435dSAl Viro size_t size = i->count; 1269610c7a71SAl Viro unsigned k; 1270357f435dSAl Viro 1271fcb14cb1SAl Viro if (iter_is_ubuf(i)) 1272fcb14cb1SAl Viro return 0; 1273fcb14cb1SAl Viro 1274610c7a71SAl Viro if (WARN_ON(!iter_is_iovec(i))) 1275241699cdSAl Viro return ~0U; 1276241699cdSAl Viro 1277610c7a71SAl Viro for (k = 0; k < i->nr_segs; k++) { 1278de4f5fedSJens Axboe const struct iovec *iov = iter_iov(i) + k; 1279de4f5fedSJens Axboe if (iov->iov_len) { 1280de4f5fedSJens Axboe unsigned long base = (unsigned long)iov->iov_base; 1281610c7a71SAl Viro if (v) // if not the first one 1282610c7a71SAl Viro res |= base | v; // this start | previous end 1283de4f5fedSJens Axboe v = base + iov->iov_len; 1284de4f5fedSJens Axboe if (size <= iov->iov_len) 1285610c7a71SAl Viro break; 1286de4f5fedSJens Axboe size -= iov->iov_len; 1287610c7a71SAl Viro } 1288610c7a71SAl Viro } 1289357f435dSAl Viro return res; 1290357f435dSAl Viro } 1291357f435dSAl Viro EXPORT_SYMBOL(iov_iter_gap_alignment); 1292357f435dSAl Viro 12933cf42da3SAl Viro static int want_pages_array(struct page ***res, size_t size, 12943cf42da3SAl Viro size_t start, unsigned int maxpages) 1295acbdeb83SAl Viro { 12963cf42da3SAl Viro unsigned int count = DIV_ROUND_UP(size + start, PAGE_SIZE); 12973cf42da3SAl Viro 12983cf42da3SAl Viro if (count > maxpages) 12993cf42da3SAl Viro count = maxpages; 13003cf42da3SAl Viro WARN_ON(!count); // caller should've prevented that 13013cf42da3SAl Viro if (!*res) { 13023cf42da3SAl Viro *res = kvmalloc_array(count, sizeof(struct page *), GFP_KERNEL); 13033cf42da3SAl Viro if (!*res) 13043cf42da3SAl Viro return 0; 13053cf42da3SAl Viro } 13063cf42da3SAl Viro return count; 1307acbdeb83SAl Viro } 1308acbdeb83SAl Viro 130985200084SAl Viro static ssize_t pipe_get_pages(struct iov_iter *i, 131085200084SAl Viro struct page ***pages, size_t maxsize, unsigned maxpages, 131185200084SAl Viro size_t *start) 1312241699cdSAl Viro { 1313746de1f8SAl Viro unsigned int npages, count, off, chunk; 131485200084SAl Viro struct page **p; 1315746de1f8SAl Viro size_t left; 1316241699cdSAl Viro 131785200084SAl Viro if (!sanity(i)) 131885200084SAl Viro return -EFAULT; 131985200084SAl Viro 132085200084SAl Viro *start = off = pipe_npages(i, &npages); 13213cf42da3SAl Viro if (!npages) 13223cf42da3SAl Viro return -EFAULT; 13233cf42da3SAl Viro count = want_pages_array(pages, maxsize, off, min(npages, maxpages)); 13243cf42da3SAl Viro if (!count) 132585200084SAl Viro return -ENOMEM; 13263cf42da3SAl Viro p = *pages; 1327746de1f8SAl Viro for (npages = 0, left = maxsize ; npages < count; npages++, left -= chunk) { 1328746de1f8SAl Viro struct page *page = append_pipe(i, left, &off); 1329e3b42964SAl Viro if (!page) 1330e3b42964SAl Viro break; 1331746de1f8SAl Viro chunk = min_t(size_t, left, PAGE_SIZE - off); 133285200084SAl Viro get_page(*p++ = page); 1333e3b42964SAl Viro } 133485200084SAl Viro if (!npages) 1335241699cdSAl Viro return -EFAULT; 1336746de1f8SAl Viro return maxsize - left; 1337241699cdSAl Viro } 1338241699cdSAl Viro 13397ff50620SDavid Howells static ssize_t iter_xarray_populate_pages(struct page **pages, struct xarray *xa, 13407ff50620SDavid Howells pgoff_t index, unsigned int nr_pages) 13417ff50620SDavid Howells { 13427ff50620SDavid Howells XA_STATE(xas, xa, index); 13437ff50620SDavid Howells struct page *page; 13447ff50620SDavid Howells unsigned int ret = 0; 13457ff50620SDavid Howells 13467ff50620SDavid Howells rcu_read_lock(); 13477ff50620SDavid Howells for (page = xas_load(&xas); page; page = xas_next(&xas)) { 13487ff50620SDavid Howells if (xas_retry(&xas, page)) 13497ff50620SDavid Howells continue; 13507ff50620SDavid Howells 13517ff50620SDavid Howells /* Has the page moved or been split? */ 13527ff50620SDavid Howells if (unlikely(page != xas_reload(&xas))) { 13537ff50620SDavid Howells xas_reset(&xas); 13547ff50620SDavid Howells continue; 13557ff50620SDavid Howells } 13567ff50620SDavid Howells 13577ff50620SDavid Howells pages[ret] = find_subpage(page, xas.xa_index); 13587ff50620SDavid Howells get_page(pages[ret]); 13597ff50620SDavid Howells if (++ret == nr_pages) 13607ff50620SDavid Howells break; 13617ff50620SDavid Howells } 13627ff50620SDavid Howells rcu_read_unlock(); 13637ff50620SDavid Howells return ret; 13647ff50620SDavid Howells } 13657ff50620SDavid Howells 13667ff50620SDavid Howells static ssize_t iter_xarray_get_pages(struct iov_iter *i, 136768fe506fSAl Viro struct page ***pages, size_t maxsize, 13687ff50620SDavid Howells unsigned maxpages, size_t *_start_offset) 13697ff50620SDavid Howells { 13703cf42da3SAl Viro unsigned nr, offset, count; 13713cf42da3SAl Viro pgoff_t index; 13727ff50620SDavid Howells loff_t pos; 13737ff50620SDavid Howells 13747ff50620SDavid Howells pos = i->xarray_start + i->iov_offset; 13757ff50620SDavid Howells index = pos >> PAGE_SHIFT; 13767ff50620SDavid Howells offset = pos & ~PAGE_MASK; 13777ff50620SDavid Howells *_start_offset = offset; 13787ff50620SDavid Howells 13793cf42da3SAl Viro count = want_pages_array(pages, maxsize, offset, maxpages); 13803cf42da3SAl Viro if (!count) 138168fe506fSAl Viro return -ENOMEM; 138268fe506fSAl Viro nr = iter_xarray_populate_pages(*pages, i->xarray, index, count); 13837ff50620SDavid Howells if (nr == 0) 13847ff50620SDavid Howells return 0; 13857ff50620SDavid Howells 1386eba2d3d7SAl Viro maxsize = min_t(size_t, nr * PAGE_SIZE - offset, maxsize); 1387310d9d5aSAl Viro i->iov_offset += maxsize; 1388310d9d5aSAl Viro i->count -= maxsize; 1389eba2d3d7SAl Viro return maxsize; 13907ff50620SDavid Howells } 13917ff50620SDavid Howells 1392fcb14cb1SAl Viro /* must be done on non-empty ITER_UBUF or ITER_IOVEC one */ 1393dd45ab9dSAl Viro static unsigned long first_iovec_segment(const struct iov_iter *i, size_t *size) 13943d671ca6SAl Viro { 13953d671ca6SAl Viro size_t skip; 13963d671ca6SAl Viro long k; 13973d671ca6SAl Viro 1398fcb14cb1SAl Viro if (iter_is_ubuf(i)) 1399fcb14cb1SAl Viro return (unsigned long)i->ubuf + i->iov_offset; 1400fcb14cb1SAl Viro 14013d671ca6SAl Viro for (k = 0, skip = i->iov_offset; k < i->nr_segs; k++, skip = 0) { 1402de4f5fedSJens Axboe const struct iovec *iov = iter_iov(i) + k; 1403de4f5fedSJens Axboe size_t len = iov->iov_len - skip; 14043d671ca6SAl Viro 14053d671ca6SAl Viro if (unlikely(!len)) 14063d671ca6SAl Viro continue; 140759dbd7d0SAl Viro if (*size > len) 14083d671ca6SAl Viro *size = len; 1409de4f5fedSJens Axboe return (unsigned long)iov->iov_base + skip; 14103d671ca6SAl Viro } 14113d671ca6SAl Viro BUG(); // if it had been empty, we wouldn't get called 14123d671ca6SAl Viro } 14133d671ca6SAl Viro 14143d671ca6SAl Viro /* must be done on non-empty ITER_BVEC one */ 14153d671ca6SAl Viro static struct page *first_bvec_segment(const struct iov_iter *i, 141659dbd7d0SAl Viro size_t *size, size_t *start) 14173d671ca6SAl Viro { 14183d671ca6SAl Viro struct page *page; 14193d671ca6SAl Viro size_t skip = i->iov_offset, len; 14203d671ca6SAl Viro 14213d671ca6SAl Viro len = i->bvec->bv_len - skip; 142259dbd7d0SAl Viro if (*size > len) 142359dbd7d0SAl Viro *size = len; 14243d671ca6SAl Viro skip += i->bvec->bv_offset; 14253d671ca6SAl Viro page = i->bvec->bv_page + skip / PAGE_SIZE; 1426dda8e5d1SAl Viro *start = skip % PAGE_SIZE; 14273d671ca6SAl Viro return page; 14283d671ca6SAl Viro } 14293d671ca6SAl Viro 143091329559SAl Viro static ssize_t __iov_iter_get_pages_alloc(struct iov_iter *i, 1431d879cb83SAl Viro struct page ***pages, size_t maxsize, 1432d8207640SLogan Gunthorpe unsigned int maxpages, size_t *start, 1433f62e52d1SDavid Howells iov_iter_extraction_t extraction_flags) 1434d879cb83SAl Viro { 1435f62e52d1SDavid Howells unsigned int n, gup_flags = 0; 1436d879cb83SAl Viro 1437d879cb83SAl Viro if (maxsize > i->count) 1438d879cb83SAl Viro maxsize = i->count; 14393d671ca6SAl Viro if (!maxsize) 14403d671ca6SAl Viro return 0; 14417392ed17SAl Viro if (maxsize > MAX_RW_COUNT) 14427392ed17SAl Viro maxsize = MAX_RW_COUNT; 1443f62e52d1SDavid Howells if (extraction_flags & ITER_ALLOW_P2PDMA) 1444f62e52d1SDavid Howells gup_flags |= FOLL_PCI_P2PDMA; 1445d879cb83SAl Viro 1446fcb14cb1SAl Viro if (likely(user_backed_iter(i))) { 14473d671ca6SAl Viro unsigned long addr; 14483cf42da3SAl Viro int res; 14499ea9ce04SDavid Howells 14503337ab08SAndreas Gruenbacher if (iov_iter_rw(i) != WRITE) 14513337ab08SAndreas Gruenbacher gup_flags |= FOLL_WRITE; 14523337ab08SAndreas Gruenbacher if (i->nofault) 14533337ab08SAndreas Gruenbacher gup_flags |= FOLL_NOFAULT; 14543337ab08SAndreas Gruenbacher 1455dd45ab9dSAl Viro addr = first_iovec_segment(i, &maxsize); 1456dd45ab9dSAl Viro *start = addr % PAGE_SIZE; 1457dd45ab9dSAl Viro addr &= PAGE_MASK; 14583cf42da3SAl Viro n = want_pages_array(pages, maxsize, *start, maxpages); 14593cf42da3SAl Viro if (!n) 1460d879cb83SAl Viro return -ENOMEM; 1461451c0ba9SAl Viro res = get_user_pages_fast(addr, n, gup_flags, *pages); 146291329559SAl Viro if (unlikely(res <= 0)) 1463d879cb83SAl Viro return res; 1464eba2d3d7SAl Viro maxsize = min_t(size_t, maxsize, res * PAGE_SIZE - *start); 1465eba2d3d7SAl Viro iov_iter_advance(i, maxsize); 1466eba2d3d7SAl Viro return maxsize; 14673d671ca6SAl Viro } 14683d671ca6SAl Viro if (iov_iter_is_bvec(i)) { 1469451c0ba9SAl Viro struct page **p; 14703d671ca6SAl Viro struct page *page; 14713d671ca6SAl Viro 147259dbd7d0SAl Viro page = first_bvec_segment(i, &maxsize, start); 14733cf42da3SAl Viro n = want_pages_array(pages, maxsize, *start, maxpages); 14743cf42da3SAl Viro if (!n) 1475d879cb83SAl Viro return -ENOMEM; 14763cf42da3SAl Viro p = *pages; 1477dda8e5d1SAl Viro for (int k = 0; k < n; k++) 1478eba2d3d7SAl Viro get_page(p[k] = page + k); 1479eba2d3d7SAl Viro maxsize = min_t(size_t, maxsize, n * PAGE_SIZE - *start); 1480310d9d5aSAl Viro i->count -= maxsize; 1481310d9d5aSAl Viro i->iov_offset += maxsize; 1482310d9d5aSAl Viro if (i->iov_offset == i->bvec->bv_len) { 1483310d9d5aSAl Viro i->iov_offset = 0; 1484310d9d5aSAl Viro i->bvec++; 1485310d9d5aSAl Viro i->nr_segs--; 1486310d9d5aSAl Viro } 1487eba2d3d7SAl Viro return maxsize; 14883d671ca6SAl Viro } 14893d671ca6SAl Viro if (iov_iter_is_pipe(i)) 1490451c0ba9SAl Viro return pipe_get_pages(i, pages, maxsize, maxpages, start); 14913d671ca6SAl Viro if (iov_iter_is_xarray(i)) 1492451c0ba9SAl Viro return iter_xarray_get_pages(i, pages, maxsize, maxpages, start); 1493d879cb83SAl Viro return -EFAULT; 1494d879cb83SAl Viro } 149591329559SAl Viro 1496d8207640SLogan Gunthorpe ssize_t iov_iter_get_pages(struct iov_iter *i, 1497451c0ba9SAl Viro struct page **pages, size_t maxsize, unsigned maxpages, 1498f62e52d1SDavid Howells size_t *start, iov_iter_extraction_t extraction_flags) 1499451c0ba9SAl Viro { 1500451c0ba9SAl Viro if (!maxpages) 1501451c0ba9SAl Viro return 0; 1502451c0ba9SAl Viro BUG_ON(!pages); 1503451c0ba9SAl Viro 1504d8207640SLogan Gunthorpe return __iov_iter_get_pages_alloc(i, &pages, maxsize, maxpages, 1505f62e52d1SDavid Howells start, extraction_flags); 1506d8207640SLogan Gunthorpe } 1507d8207640SLogan Gunthorpe EXPORT_SYMBOL_GPL(iov_iter_get_pages); 1508d8207640SLogan Gunthorpe 1509d8207640SLogan Gunthorpe ssize_t iov_iter_get_pages2(struct iov_iter *i, struct page **pages, 1510d8207640SLogan Gunthorpe size_t maxsize, unsigned maxpages, size_t *start) 1511d8207640SLogan Gunthorpe { 1512d8207640SLogan Gunthorpe return iov_iter_get_pages(i, pages, maxsize, maxpages, start, 0); 1513451c0ba9SAl Viro } 1514eba2d3d7SAl Viro EXPORT_SYMBOL(iov_iter_get_pages2); 1515451c0ba9SAl Viro 1516d8207640SLogan Gunthorpe ssize_t iov_iter_get_pages_alloc(struct iov_iter *i, 151791329559SAl Viro struct page ***pages, size_t maxsize, 1518f62e52d1SDavid Howells size_t *start, iov_iter_extraction_t extraction_flags) 151991329559SAl Viro { 152091329559SAl Viro ssize_t len; 152191329559SAl Viro 152291329559SAl Viro *pages = NULL; 152391329559SAl Viro 1524d8207640SLogan Gunthorpe len = __iov_iter_get_pages_alloc(i, pages, maxsize, ~0U, start, 1525f62e52d1SDavid Howells extraction_flags); 152691329559SAl Viro if (len <= 0) { 152791329559SAl Viro kvfree(*pages); 152891329559SAl Viro *pages = NULL; 152991329559SAl Viro } 153091329559SAl Viro return len; 153191329559SAl Viro } 1532d8207640SLogan Gunthorpe EXPORT_SYMBOL_GPL(iov_iter_get_pages_alloc); 1533d8207640SLogan Gunthorpe 1534d8207640SLogan Gunthorpe ssize_t iov_iter_get_pages_alloc2(struct iov_iter *i, 1535d8207640SLogan Gunthorpe struct page ***pages, size_t maxsize, size_t *start) 1536d8207640SLogan Gunthorpe { 1537d8207640SLogan Gunthorpe return iov_iter_get_pages_alloc(i, pages, maxsize, start, 0); 1538d8207640SLogan Gunthorpe } 1539eba2d3d7SAl Viro EXPORT_SYMBOL(iov_iter_get_pages_alloc2); 1540d879cb83SAl Viro 1541d879cb83SAl Viro size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum, 1542d879cb83SAl Viro struct iov_iter *i) 1543d879cb83SAl Viro { 1544d879cb83SAl Viro __wsum sum, next; 1545d879cb83SAl Viro sum = *csum; 1546a41dad90SAl Viro if (WARN_ON_ONCE(!i->data_source)) 1547241699cdSAl Viro return 0; 1548a41dad90SAl Viro 15497baa5099SAl Viro iterate_and_advance(i, bytes, base, len, off, ({ 15507baa5099SAl Viro next = csum_and_copy_from_user(base, addr + off, len); 1551d879cb83SAl Viro sum = csum_block_add(sum, next, off); 15527baa5099SAl Viro next ? 0 : len; 1553d879cb83SAl Viro }), ({ 15547baa5099SAl Viro sum = csum_and_memcpy(addr + off, base, len, sum, off); 1555d879cb83SAl Viro }) 1556d879cb83SAl Viro ) 1557d879cb83SAl Viro *csum = sum; 1558d879cb83SAl Viro return bytes; 1559d879cb83SAl Viro } 1560d879cb83SAl Viro EXPORT_SYMBOL(csum_and_copy_from_iter); 1561d879cb83SAl Viro 156252cbd23aSWillem de Bruijn size_t csum_and_copy_to_iter(const void *addr, size_t bytes, void *_csstate, 1563d879cb83SAl Viro struct iov_iter *i) 1564d879cb83SAl Viro { 156552cbd23aSWillem de Bruijn struct csum_state *csstate = _csstate; 1566d879cb83SAl Viro __wsum sum, next; 156778e1f386SAl Viro 1568a41dad90SAl Viro if (WARN_ON_ONCE(i->data_source)) 1569a41dad90SAl Viro return 0; 157078e1f386SAl Viro if (unlikely(iov_iter_is_discard(i))) { 1571c67f1fd2SAl Viro // can't use csum_memcpy() for that one - data is not copied 1572c67f1fd2SAl Viro csstate->csum = csum_block_add(csstate->csum, 1573c67f1fd2SAl Viro csum_partial(addr, bytes, 0), 1574c67f1fd2SAl Viro csstate->off); 1575c67f1fd2SAl Viro csstate->off += bytes; 1576c67f1fd2SAl Viro return bytes; 1577241699cdSAl Viro } 15786852df12SAl Viro 15796852df12SAl Viro sum = csum_shift(csstate->csum, csstate->off); 15806852df12SAl Viro if (unlikely(iov_iter_is_pipe(i))) 15816852df12SAl Viro bytes = csum_and_copy_to_pipe_iter(addr, bytes, i, &sum); 15826852df12SAl Viro else iterate_and_advance(i, bytes, base, len, off, ({ 15837baa5099SAl Viro next = csum_and_copy_to_user(addr + off, base, len); 1584d879cb83SAl Viro sum = csum_block_add(sum, next, off); 15857baa5099SAl Viro next ? 0 : len; 1586d879cb83SAl Viro }), ({ 15877baa5099SAl Viro sum = csum_and_memcpy(base, addr + off, len, sum, off); 1588d879cb83SAl Viro }) 1589d879cb83SAl Viro ) 1590594e450bSAl Viro csstate->csum = csum_shift(sum, csstate->off); 1591594e450bSAl Viro csstate->off += bytes; 1592d879cb83SAl Viro return bytes; 1593d879cb83SAl Viro } 1594d879cb83SAl Viro EXPORT_SYMBOL(csum_and_copy_to_iter); 1595d879cb83SAl Viro 1596d05f4435SSagi Grimberg size_t hash_and_copy_to_iter(const void *addr, size_t bytes, void *hashp, 1597d05f4435SSagi Grimberg struct iov_iter *i) 1598d05f4435SSagi Grimberg { 15997999096fSHerbert Xu #ifdef CONFIG_CRYPTO_HASH 1600d05f4435SSagi Grimberg struct ahash_request *hash = hashp; 1601d05f4435SSagi Grimberg struct scatterlist sg; 1602d05f4435SSagi Grimberg size_t copied; 1603d05f4435SSagi Grimberg 1604d05f4435SSagi Grimberg copied = copy_to_iter(addr, bytes, i); 1605d05f4435SSagi Grimberg sg_init_one(&sg, addr, copied); 1606d05f4435SSagi Grimberg ahash_request_set_crypt(hash, &sg, NULL, copied); 1607d05f4435SSagi Grimberg crypto_ahash_update(hash); 1608d05f4435SSagi Grimberg return copied; 160927fad74aSYueHaibing #else 161027fad74aSYueHaibing return 0; 161127fad74aSYueHaibing #endif 1612d05f4435SSagi Grimberg } 1613d05f4435SSagi Grimberg EXPORT_SYMBOL(hash_and_copy_to_iter); 1614d05f4435SSagi Grimberg 161566531c65SAl Viro static int iov_npages(const struct iov_iter *i, int maxpages) 1616d879cb83SAl Viro { 161766531c65SAl Viro size_t skip = i->iov_offset, size = i->count; 161866531c65SAl Viro const struct iovec *p; 1619d879cb83SAl Viro int npages = 0; 1620d879cb83SAl Viro 1621de4f5fedSJens Axboe for (p = iter_iov(i); size; skip = 0, p++) { 162266531c65SAl Viro unsigned offs = offset_in_page(p->iov_base + skip); 162366531c65SAl Viro size_t len = min(p->iov_len - skip, size); 1624d879cb83SAl Viro 162566531c65SAl Viro if (len) { 162666531c65SAl Viro size -= len; 162766531c65SAl Viro npages += DIV_ROUND_UP(offs + len, PAGE_SIZE); 162866531c65SAl Viro if (unlikely(npages > maxpages)) 162966531c65SAl Viro return maxpages; 163066531c65SAl Viro } 163166531c65SAl Viro } 163266531c65SAl Viro return npages; 163366531c65SAl Viro } 163466531c65SAl Viro 163566531c65SAl Viro static int bvec_npages(const struct iov_iter *i, int maxpages) 163666531c65SAl Viro { 163766531c65SAl Viro size_t skip = i->iov_offset, size = i->count; 163866531c65SAl Viro const struct bio_vec *p; 163966531c65SAl Viro int npages = 0; 164066531c65SAl Viro 164166531c65SAl Viro for (p = i->bvec; size; skip = 0, p++) { 164266531c65SAl Viro unsigned offs = (p->bv_offset + skip) % PAGE_SIZE; 164366531c65SAl Viro size_t len = min(p->bv_len - skip, size); 164466531c65SAl Viro 164566531c65SAl Viro size -= len; 164666531c65SAl Viro npages += DIV_ROUND_UP(offs + len, PAGE_SIZE); 164766531c65SAl Viro if (unlikely(npages > maxpages)) 164866531c65SAl Viro return maxpages; 164966531c65SAl Viro } 165066531c65SAl Viro return npages; 165166531c65SAl Viro } 165266531c65SAl Viro 165366531c65SAl Viro int iov_iter_npages(const struct iov_iter *i, int maxpages) 165466531c65SAl Viro { 165566531c65SAl Viro if (unlikely(!i->count)) 165666531c65SAl Viro return 0; 1657fcb14cb1SAl Viro if (likely(iter_is_ubuf(i))) { 1658fcb14cb1SAl Viro unsigned offs = offset_in_page(i->ubuf + i->iov_offset); 1659fcb14cb1SAl Viro int npages = DIV_ROUND_UP(offs + i->count, PAGE_SIZE); 1660fcb14cb1SAl Viro return min(npages, maxpages); 1661fcb14cb1SAl Viro } 166266531c65SAl Viro /* iovec and kvec have identical layouts */ 166366531c65SAl Viro if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) 166466531c65SAl Viro return iov_npages(i, maxpages); 166566531c65SAl Viro if (iov_iter_is_bvec(i)) 166666531c65SAl Viro return bvec_npages(i, maxpages); 166766531c65SAl Viro if (iov_iter_is_pipe(i)) { 166866531c65SAl Viro int npages; 1669241699cdSAl Viro 1670241699cdSAl Viro if (!sanity(i)) 1671241699cdSAl Viro return 0; 1672241699cdSAl Viro 167312d426abSAl Viro pipe_npages(i, &npages); 167466531c65SAl Viro return min(npages, maxpages); 167566531c65SAl Viro } 167666531c65SAl Viro if (iov_iter_is_xarray(i)) { 1677e4f8df86SAl Viro unsigned offset = (i->xarray_start + i->iov_offset) % PAGE_SIZE; 1678e4f8df86SAl Viro int npages = DIV_ROUND_UP(offset + i->count, PAGE_SIZE); 167966531c65SAl Viro return min(npages, maxpages); 168066531c65SAl Viro } 168166531c65SAl Viro return 0; 1682d879cb83SAl Viro } 1683d879cb83SAl Viro EXPORT_SYMBOL(iov_iter_npages); 1684d879cb83SAl Viro 1685d879cb83SAl Viro const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags) 1686d879cb83SAl Viro { 1687d879cb83SAl Viro *new = *old; 168800e23707SDavid Howells if (unlikely(iov_iter_is_pipe(new))) { 1689241699cdSAl Viro WARN_ON(1); 1690241699cdSAl Viro return NULL; 1691241699cdSAl Viro } 169200e23707SDavid Howells if (iov_iter_is_bvec(new)) 1693d879cb83SAl Viro return new->bvec = kmemdup(new->bvec, 1694d879cb83SAl Viro new->nr_segs * sizeof(struct bio_vec), 1695d879cb83SAl Viro flags); 1696fcb14cb1SAl Viro else if (iov_iter_is_kvec(new) || iter_is_iovec(new)) 1697d879cb83SAl Viro /* iovec and kvec have identical layout */ 1698de4f5fedSJens Axboe return new->__iov = kmemdup(new->__iov, 1699d879cb83SAl Viro new->nr_segs * sizeof(struct iovec), 1700d879cb83SAl Viro flags); 1701fcb14cb1SAl Viro return NULL; 1702d879cb83SAl Viro } 1703d879cb83SAl Viro EXPORT_SYMBOL(dup_iter); 1704bc917be8SAl Viro 170550f9a76eSJosh Poimboeuf static __noclone int copy_compat_iovec_from_user(struct iovec *iov, 1706bfdc5970SChristoph Hellwig const struct iovec __user *uvec, unsigned long nr_segs) 1707bfdc5970SChristoph Hellwig { 1708bfdc5970SChristoph Hellwig const struct compat_iovec __user *uiov = 1709bfdc5970SChristoph Hellwig (const struct compat_iovec __user *)uvec; 1710bfdc5970SChristoph Hellwig int ret = -EFAULT, i; 1711bfdc5970SChristoph Hellwig 1712a959a978SChristoph Hellwig if (!user_access_begin(uiov, nr_segs * sizeof(*uiov))) 1713bfdc5970SChristoph Hellwig return -EFAULT; 1714bfdc5970SChristoph Hellwig 1715bfdc5970SChristoph Hellwig for (i = 0; i < nr_segs; i++) { 1716bfdc5970SChristoph Hellwig compat_uptr_t buf; 1717bfdc5970SChristoph Hellwig compat_ssize_t len; 1718bfdc5970SChristoph Hellwig 1719bfdc5970SChristoph Hellwig unsafe_get_user(len, &uiov[i].iov_len, uaccess_end); 1720bfdc5970SChristoph Hellwig unsafe_get_user(buf, &uiov[i].iov_base, uaccess_end); 1721bfdc5970SChristoph Hellwig 1722bfdc5970SChristoph Hellwig /* check for compat_size_t not fitting in compat_ssize_t .. */ 1723bfdc5970SChristoph Hellwig if (len < 0) { 1724bfdc5970SChristoph Hellwig ret = -EINVAL; 1725bfdc5970SChristoph Hellwig goto uaccess_end; 1726bfdc5970SChristoph Hellwig } 1727bfdc5970SChristoph Hellwig iov[i].iov_base = compat_ptr(buf); 1728bfdc5970SChristoph Hellwig iov[i].iov_len = len; 1729bfdc5970SChristoph Hellwig } 1730bfdc5970SChristoph Hellwig 1731bfdc5970SChristoph Hellwig ret = 0; 1732bfdc5970SChristoph Hellwig uaccess_end: 1733bfdc5970SChristoph Hellwig user_access_end(); 1734bfdc5970SChristoph Hellwig return ret; 1735bfdc5970SChristoph Hellwig } 1736bfdc5970SChristoph Hellwig 1737bfdc5970SChristoph Hellwig static int copy_iovec_from_user(struct iovec *iov, 1738*487c20b0SLinus Torvalds const struct iovec __user *uiov, unsigned long nr_segs) 1739fb041b59SDavid Laight { 1740*487c20b0SLinus Torvalds int ret = -EFAULT; 1741bfdc5970SChristoph Hellwig 1742*487c20b0SLinus Torvalds if (!user_access_begin(uiov, nr_segs * sizeof(*uiov))) 1743bfdc5970SChristoph Hellwig return -EFAULT; 1744bfdc5970SChristoph Hellwig 1745*487c20b0SLinus Torvalds do { 1746*487c20b0SLinus Torvalds void __user *buf; 1747*487c20b0SLinus Torvalds ssize_t len; 1748*487c20b0SLinus Torvalds 1749*487c20b0SLinus Torvalds unsafe_get_user(len, &uiov->iov_len, uaccess_end); 1750*487c20b0SLinus Torvalds unsafe_get_user(buf, &uiov->iov_base, uaccess_end); 1751*487c20b0SLinus Torvalds 1752*487c20b0SLinus Torvalds /* check for size_t not fitting in ssize_t .. */ 1753*487c20b0SLinus Torvalds if (unlikely(len < 0)) { 1754*487c20b0SLinus Torvalds ret = -EINVAL; 1755*487c20b0SLinus Torvalds goto uaccess_end; 1756*487c20b0SLinus Torvalds } 1757*487c20b0SLinus Torvalds iov->iov_base = buf; 1758*487c20b0SLinus Torvalds iov->iov_len = len; 1759*487c20b0SLinus Torvalds 1760*487c20b0SLinus Torvalds uiov++; iov++; 1761*487c20b0SLinus Torvalds } while (--nr_segs); 1762*487c20b0SLinus Torvalds 1763*487c20b0SLinus Torvalds ret = 0; 1764*487c20b0SLinus Torvalds uaccess_end: 1765*487c20b0SLinus Torvalds user_access_end(); 1766*487c20b0SLinus Torvalds return ret; 1767bfdc5970SChristoph Hellwig } 1768bfdc5970SChristoph Hellwig 1769bfdc5970SChristoph Hellwig struct iovec *iovec_from_user(const struct iovec __user *uvec, 1770bfdc5970SChristoph Hellwig unsigned long nr_segs, unsigned long fast_segs, 1771bfdc5970SChristoph Hellwig struct iovec *fast_iov, bool compat) 1772bfdc5970SChristoph Hellwig { 1773bfdc5970SChristoph Hellwig struct iovec *iov = fast_iov; 1774bfdc5970SChristoph Hellwig int ret; 1775fb041b59SDavid Laight 1776fb041b59SDavid Laight /* 1777bfdc5970SChristoph Hellwig * SuS says "The readv() function *may* fail if the iovcnt argument was 1778bfdc5970SChristoph Hellwig * less than or equal to 0, or greater than {IOV_MAX}. Linux has 1779fb041b59SDavid Laight * traditionally returned zero for zero segments, so... 1780fb041b59SDavid Laight */ 1781bfdc5970SChristoph Hellwig if (nr_segs == 0) 1782bfdc5970SChristoph Hellwig return iov; 1783bfdc5970SChristoph Hellwig if (nr_segs > UIO_MAXIOV) 1784bfdc5970SChristoph Hellwig return ERR_PTR(-EINVAL); 1785fb041b59SDavid Laight if (nr_segs > fast_segs) { 1786fb041b59SDavid Laight iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL); 1787bfdc5970SChristoph Hellwig if (!iov) 1788bfdc5970SChristoph Hellwig return ERR_PTR(-ENOMEM); 1789fb041b59SDavid Laight } 1790bfdc5970SChristoph Hellwig 1791*487c20b0SLinus Torvalds if (unlikely(compat)) 1792bfdc5970SChristoph Hellwig ret = copy_compat_iovec_from_user(iov, uvec, nr_segs); 1793bfdc5970SChristoph Hellwig else 1794bfdc5970SChristoph Hellwig ret = copy_iovec_from_user(iov, uvec, nr_segs); 1795bfdc5970SChristoph Hellwig if (ret) { 1796bfdc5970SChristoph Hellwig if (iov != fast_iov) 1797bfdc5970SChristoph Hellwig kfree(iov); 1798bfdc5970SChristoph Hellwig return ERR_PTR(ret); 1799fb041b59SDavid Laight } 1800bfdc5970SChristoph Hellwig 1801bfdc5970SChristoph Hellwig return iov; 1802bfdc5970SChristoph Hellwig } 1803bfdc5970SChristoph Hellwig 18043b2deb0eSJens Axboe /* 18053b2deb0eSJens Axboe * Single segment iovec supplied by the user, import it as ITER_UBUF. 18063b2deb0eSJens Axboe */ 18073b2deb0eSJens Axboe static ssize_t __import_iovec_ubuf(int type, const struct iovec __user *uvec, 18083b2deb0eSJens Axboe struct iovec **iovp, struct iov_iter *i, 18093b2deb0eSJens Axboe bool compat) 18103b2deb0eSJens Axboe { 18113b2deb0eSJens Axboe struct iovec *iov = *iovp; 18123b2deb0eSJens Axboe ssize_t ret; 18133b2deb0eSJens Axboe 18143b2deb0eSJens Axboe if (compat) 18153b2deb0eSJens Axboe ret = copy_compat_iovec_from_user(iov, uvec, 1); 18163b2deb0eSJens Axboe else 18173b2deb0eSJens Axboe ret = copy_iovec_from_user(iov, uvec, 1); 18183b2deb0eSJens Axboe if (unlikely(ret)) 18193b2deb0eSJens Axboe return ret; 18203b2deb0eSJens Axboe 18213b2deb0eSJens Axboe ret = import_ubuf(type, iov->iov_base, iov->iov_len, i); 18223b2deb0eSJens Axboe if (unlikely(ret)) 18233b2deb0eSJens Axboe return ret; 18243b2deb0eSJens Axboe *iovp = NULL; 18253b2deb0eSJens Axboe return i->count; 18263b2deb0eSJens Axboe } 18273b2deb0eSJens Axboe 1828bfdc5970SChristoph Hellwig ssize_t __import_iovec(int type, const struct iovec __user *uvec, 1829bfdc5970SChristoph Hellwig unsigned nr_segs, unsigned fast_segs, struct iovec **iovp, 1830bfdc5970SChristoph Hellwig struct iov_iter *i, bool compat) 1831bfdc5970SChristoph Hellwig { 1832bfdc5970SChristoph Hellwig ssize_t total_len = 0; 1833bfdc5970SChristoph Hellwig unsigned long seg; 1834bfdc5970SChristoph Hellwig struct iovec *iov; 1835bfdc5970SChristoph Hellwig 18363b2deb0eSJens Axboe if (nr_segs == 1) 18373b2deb0eSJens Axboe return __import_iovec_ubuf(type, uvec, iovp, i, compat); 18383b2deb0eSJens Axboe 1839bfdc5970SChristoph Hellwig iov = iovec_from_user(uvec, nr_segs, fast_segs, *iovp, compat); 1840bfdc5970SChristoph Hellwig if (IS_ERR(iov)) { 1841bfdc5970SChristoph Hellwig *iovp = NULL; 1842bfdc5970SChristoph Hellwig return PTR_ERR(iov); 1843fb041b59SDavid Laight } 1844fb041b59SDavid Laight 1845fb041b59SDavid Laight /* 1846bfdc5970SChristoph Hellwig * According to the Single Unix Specification we should return EINVAL if 1847bfdc5970SChristoph Hellwig * an element length is < 0 when cast to ssize_t or if the total length 1848bfdc5970SChristoph Hellwig * would overflow the ssize_t return value of the system call. 1849fb041b59SDavid Laight * 1850fb041b59SDavid Laight * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the 1851fb041b59SDavid Laight * overflow case. 1852fb041b59SDavid Laight */ 1853fb041b59SDavid Laight for (seg = 0; seg < nr_segs; seg++) { 1854fb041b59SDavid Laight ssize_t len = (ssize_t)iov[seg].iov_len; 1855fb041b59SDavid Laight 1856bfdc5970SChristoph Hellwig if (!access_ok(iov[seg].iov_base, len)) { 1857bfdc5970SChristoph Hellwig if (iov != *iovp) 1858bfdc5970SChristoph Hellwig kfree(iov); 1859bfdc5970SChristoph Hellwig *iovp = NULL; 1860bfdc5970SChristoph Hellwig return -EFAULT; 1861fb041b59SDavid Laight } 1862bfdc5970SChristoph Hellwig 1863bfdc5970SChristoph Hellwig if (len > MAX_RW_COUNT - total_len) { 1864bfdc5970SChristoph Hellwig len = MAX_RW_COUNT - total_len; 1865fb041b59SDavid Laight iov[seg].iov_len = len; 1866fb041b59SDavid Laight } 1867bfdc5970SChristoph Hellwig total_len += len; 1868fb041b59SDavid Laight } 1869bfdc5970SChristoph Hellwig 1870bfdc5970SChristoph Hellwig iov_iter_init(i, type, iov, nr_segs, total_len); 1871bfdc5970SChristoph Hellwig if (iov == *iovp) 1872bfdc5970SChristoph Hellwig *iovp = NULL; 1873bfdc5970SChristoph Hellwig else 1874bfdc5970SChristoph Hellwig *iovp = iov; 1875bfdc5970SChristoph Hellwig return total_len; 1876fb041b59SDavid Laight } 1877fb041b59SDavid Laight 1878ffecee4fSVegard Nossum /** 1879ffecee4fSVegard Nossum * import_iovec() - Copy an array of &struct iovec from userspace 1880ffecee4fSVegard Nossum * into the kernel, check that it is valid, and initialize a new 1881ffecee4fSVegard Nossum * &struct iov_iter iterator to access it. 1882ffecee4fSVegard Nossum * 1883ffecee4fSVegard Nossum * @type: One of %READ or %WRITE. 1884bfdc5970SChristoph Hellwig * @uvec: Pointer to the userspace array. 1885ffecee4fSVegard Nossum * @nr_segs: Number of elements in userspace array. 1886ffecee4fSVegard Nossum * @fast_segs: Number of elements in @iov. 1887bfdc5970SChristoph Hellwig * @iovp: (input and output parameter) Pointer to pointer to (usually small 1888ffecee4fSVegard Nossum * on-stack) kernel array. 1889ffecee4fSVegard Nossum * @i: Pointer to iterator that will be initialized on success. 1890ffecee4fSVegard Nossum * 1891ffecee4fSVegard Nossum * If the array pointed to by *@iov is large enough to hold all @nr_segs, 1892ffecee4fSVegard Nossum * then this function places %NULL in *@iov on return. Otherwise, a new 1893ffecee4fSVegard Nossum * array will be allocated and the result placed in *@iov. This means that 1894ffecee4fSVegard Nossum * the caller may call kfree() on *@iov regardless of whether the small 1895ffecee4fSVegard Nossum * on-stack array was used or not (and regardless of whether this function 1896ffecee4fSVegard Nossum * returns an error or not). 1897ffecee4fSVegard Nossum * 189887e5e6daSJens Axboe * Return: Negative error code on error, bytes imported on success 1899ffecee4fSVegard Nossum */ 1900bfdc5970SChristoph Hellwig ssize_t import_iovec(int type, const struct iovec __user *uvec, 1901bc917be8SAl Viro unsigned nr_segs, unsigned fast_segs, 1902bfdc5970SChristoph Hellwig struct iovec **iovp, struct iov_iter *i) 1903bc917be8SAl Viro { 190489cd35c5SChristoph Hellwig return __import_iovec(type, uvec, nr_segs, fast_segs, iovp, i, 190589cd35c5SChristoph Hellwig in_compat_syscall()); 1906bc917be8SAl Viro } 1907bc917be8SAl Viro EXPORT_SYMBOL(import_iovec); 1908bc917be8SAl Viro 1909bc917be8SAl Viro int import_single_range(int rw, void __user *buf, size_t len, 1910bc917be8SAl Viro struct iovec *iov, struct iov_iter *i) 1911bc917be8SAl Viro { 1912bc917be8SAl Viro if (len > MAX_RW_COUNT) 1913bc917be8SAl Viro len = MAX_RW_COUNT; 191496d4f267SLinus Torvalds if (unlikely(!access_ok(buf, len))) 1915bc917be8SAl Viro return -EFAULT; 1916bc917be8SAl Viro 1917e03ad4eeSJens Axboe iov_iter_ubuf(i, rw, buf, len); 1918bc917be8SAl Viro return 0; 1919bc917be8SAl Viro } 1920e1267585SAl Viro EXPORT_SYMBOL(import_single_range); 19218fb0f47aSJens Axboe 19222ad9bd83SJens Axboe int import_ubuf(int rw, void __user *buf, size_t len, struct iov_iter *i) 19232ad9bd83SJens Axboe { 19242ad9bd83SJens Axboe if (len > MAX_RW_COUNT) 19252ad9bd83SJens Axboe len = MAX_RW_COUNT; 19262ad9bd83SJens Axboe if (unlikely(!access_ok(buf, len))) 19272ad9bd83SJens Axboe return -EFAULT; 19282ad9bd83SJens Axboe 19292ad9bd83SJens Axboe iov_iter_ubuf(i, rw, buf, len); 19302ad9bd83SJens Axboe return 0; 19312ad9bd83SJens Axboe } 19322ad9bd83SJens Axboe 19338fb0f47aSJens Axboe /** 19348fb0f47aSJens Axboe * iov_iter_restore() - Restore a &struct iov_iter to the same state as when 19358fb0f47aSJens Axboe * iov_iter_save_state() was called. 19368fb0f47aSJens Axboe * 19378fb0f47aSJens Axboe * @i: &struct iov_iter to restore 19388fb0f47aSJens Axboe * @state: state to restore from 19398fb0f47aSJens Axboe * 19408fb0f47aSJens Axboe * Used after iov_iter_save_state() to bring restore @i, if operations may 19418fb0f47aSJens Axboe * have advanced it. 19428fb0f47aSJens Axboe * 19438fb0f47aSJens Axboe * Note: only works on ITER_IOVEC, ITER_BVEC, and ITER_KVEC 19448fb0f47aSJens Axboe */ 19458fb0f47aSJens Axboe void iov_iter_restore(struct iov_iter *i, struct iov_iter_state *state) 19468fb0f47aSJens Axboe { 19474397a17cSKeith Busch if (WARN_ON_ONCE(!iov_iter_is_bvec(i) && !iter_is_iovec(i) && 19484397a17cSKeith Busch !iter_is_ubuf(i)) && !iov_iter_is_kvec(i)) 19498fb0f47aSJens Axboe return; 19508fb0f47aSJens Axboe i->iov_offset = state->iov_offset; 19518fb0f47aSJens Axboe i->count = state->count; 1952fcb14cb1SAl Viro if (iter_is_ubuf(i)) 1953fcb14cb1SAl Viro return; 19548fb0f47aSJens Axboe /* 19558fb0f47aSJens Axboe * For the *vec iters, nr_segs + iov is constant - if we increment 19568fb0f47aSJens Axboe * the vec, then we also decrement the nr_segs count. Hence we don't 19578fb0f47aSJens Axboe * need to track both of these, just one is enough and we can deduct 19588fb0f47aSJens Axboe * the other from that. ITER_KVEC and ITER_IOVEC are the same struct 19598fb0f47aSJens Axboe * size, so we can just increment the iov pointer as they are unionzed. 19608fb0f47aSJens Axboe * ITER_BVEC _may_ be the same size on some archs, but on others it is 19618fb0f47aSJens Axboe * not. Be safe and handle it separately. 19628fb0f47aSJens Axboe */ 19638fb0f47aSJens Axboe BUILD_BUG_ON(sizeof(struct iovec) != sizeof(struct kvec)); 19648fb0f47aSJens Axboe if (iov_iter_is_bvec(i)) 19658fb0f47aSJens Axboe i->bvec -= state->nr_segs - i->nr_segs; 19668fb0f47aSJens Axboe else 1967de4f5fedSJens Axboe i->__iov -= state->nr_segs - i->nr_segs; 19688fb0f47aSJens Axboe i->nr_segs = state->nr_segs; 19698fb0f47aSJens Axboe } 19707d58fe73SDavid Howells 19717d58fe73SDavid Howells /* 19727d58fe73SDavid Howells * Extract a list of contiguous pages from an ITER_XARRAY iterator. This does not 19737d58fe73SDavid Howells * get references on the pages, nor does it get a pin on them. 19747d58fe73SDavid Howells */ 19757d58fe73SDavid Howells static ssize_t iov_iter_extract_xarray_pages(struct iov_iter *i, 19767d58fe73SDavid Howells struct page ***pages, size_t maxsize, 19777d58fe73SDavid Howells unsigned int maxpages, 19787d58fe73SDavid Howells iov_iter_extraction_t extraction_flags, 19797d58fe73SDavid Howells size_t *offset0) 19807d58fe73SDavid Howells { 19817d58fe73SDavid Howells struct page *page, **p; 19827d58fe73SDavid Howells unsigned int nr = 0, offset; 19837d58fe73SDavid Howells loff_t pos = i->xarray_start + i->iov_offset; 19847d58fe73SDavid Howells pgoff_t index = pos >> PAGE_SHIFT; 19857d58fe73SDavid Howells XA_STATE(xas, i->xarray, index); 19867d58fe73SDavid Howells 19877d58fe73SDavid Howells offset = pos & ~PAGE_MASK; 19887d58fe73SDavid Howells *offset0 = offset; 19897d58fe73SDavid Howells 19907d58fe73SDavid Howells maxpages = want_pages_array(pages, maxsize, offset, maxpages); 19917d58fe73SDavid Howells if (!maxpages) 19927d58fe73SDavid Howells return -ENOMEM; 19937d58fe73SDavid Howells p = *pages; 19947d58fe73SDavid Howells 19957d58fe73SDavid Howells rcu_read_lock(); 19967d58fe73SDavid Howells for (page = xas_load(&xas); page; page = xas_next(&xas)) { 19977d58fe73SDavid Howells if (xas_retry(&xas, page)) 19987d58fe73SDavid Howells continue; 19997d58fe73SDavid Howells 20007d58fe73SDavid Howells /* Has the page moved or been split? */ 20017d58fe73SDavid Howells if (unlikely(page != xas_reload(&xas))) { 20027d58fe73SDavid Howells xas_reset(&xas); 20037d58fe73SDavid Howells continue; 20047d58fe73SDavid Howells } 20057d58fe73SDavid Howells 20067d58fe73SDavid Howells p[nr++] = find_subpage(page, xas.xa_index); 20077d58fe73SDavid Howells if (nr == maxpages) 20087d58fe73SDavid Howells break; 20097d58fe73SDavid Howells } 20107d58fe73SDavid Howells rcu_read_unlock(); 20117d58fe73SDavid Howells 20127d58fe73SDavid Howells maxsize = min_t(size_t, nr * PAGE_SIZE - offset, maxsize); 20137d58fe73SDavid Howells iov_iter_advance(i, maxsize); 20147d58fe73SDavid Howells return maxsize; 20157d58fe73SDavid Howells } 20167d58fe73SDavid Howells 20177d58fe73SDavid Howells /* 20187d58fe73SDavid Howells * Extract a list of contiguous pages from an ITER_BVEC iterator. This does 20197d58fe73SDavid Howells * not get references on the pages, nor does it get a pin on them. 20207d58fe73SDavid Howells */ 20217d58fe73SDavid Howells static ssize_t iov_iter_extract_bvec_pages(struct iov_iter *i, 20227d58fe73SDavid Howells struct page ***pages, size_t maxsize, 20237d58fe73SDavid Howells unsigned int maxpages, 20247d58fe73SDavid Howells iov_iter_extraction_t extraction_flags, 20257d58fe73SDavid Howells size_t *offset0) 20267d58fe73SDavid Howells { 20277d58fe73SDavid Howells struct page **p, *page; 20287d58fe73SDavid Howells size_t skip = i->iov_offset, offset; 20297d58fe73SDavid Howells int k; 20307d58fe73SDavid Howells 20317d58fe73SDavid Howells for (;;) { 20327d58fe73SDavid Howells if (i->nr_segs == 0) 20337d58fe73SDavid Howells return 0; 20347d58fe73SDavid Howells maxsize = min(maxsize, i->bvec->bv_len - skip); 20357d58fe73SDavid Howells if (maxsize) 20367d58fe73SDavid Howells break; 20377d58fe73SDavid Howells i->iov_offset = 0; 20387d58fe73SDavid Howells i->nr_segs--; 20397d58fe73SDavid Howells i->bvec++; 20407d58fe73SDavid Howells skip = 0; 20417d58fe73SDavid Howells } 20427d58fe73SDavid Howells 20437d58fe73SDavid Howells skip += i->bvec->bv_offset; 20447d58fe73SDavid Howells page = i->bvec->bv_page + skip / PAGE_SIZE; 20457d58fe73SDavid Howells offset = skip % PAGE_SIZE; 20467d58fe73SDavid Howells *offset0 = offset; 20477d58fe73SDavid Howells 20487d58fe73SDavid Howells maxpages = want_pages_array(pages, maxsize, offset, maxpages); 20497d58fe73SDavid Howells if (!maxpages) 20507d58fe73SDavid Howells return -ENOMEM; 20517d58fe73SDavid Howells p = *pages; 20527d58fe73SDavid Howells for (k = 0; k < maxpages; k++) 20537d58fe73SDavid Howells p[k] = page + k; 20547d58fe73SDavid Howells 20557d58fe73SDavid Howells maxsize = min_t(size_t, maxsize, maxpages * PAGE_SIZE - offset); 20567d58fe73SDavid Howells iov_iter_advance(i, maxsize); 20577d58fe73SDavid Howells return maxsize; 20587d58fe73SDavid Howells } 20597d58fe73SDavid Howells 20607d58fe73SDavid Howells /* 20617d58fe73SDavid Howells * Extract a list of virtually contiguous pages from an ITER_KVEC iterator. 20627d58fe73SDavid Howells * This does not get references on the pages, nor does it get a pin on them. 20637d58fe73SDavid Howells */ 20647d58fe73SDavid Howells static ssize_t iov_iter_extract_kvec_pages(struct iov_iter *i, 20657d58fe73SDavid Howells struct page ***pages, size_t maxsize, 20667d58fe73SDavid Howells unsigned int maxpages, 20677d58fe73SDavid Howells iov_iter_extraction_t extraction_flags, 20687d58fe73SDavid Howells size_t *offset0) 20697d58fe73SDavid Howells { 20707d58fe73SDavid Howells struct page **p, *page; 20717d58fe73SDavid Howells const void *kaddr; 20727d58fe73SDavid Howells size_t skip = i->iov_offset, offset, len; 20737d58fe73SDavid Howells int k; 20747d58fe73SDavid Howells 20757d58fe73SDavid Howells for (;;) { 20767d58fe73SDavid Howells if (i->nr_segs == 0) 20777d58fe73SDavid Howells return 0; 20787d58fe73SDavid Howells maxsize = min(maxsize, i->kvec->iov_len - skip); 20797d58fe73SDavid Howells if (maxsize) 20807d58fe73SDavid Howells break; 20817d58fe73SDavid Howells i->iov_offset = 0; 20827d58fe73SDavid Howells i->nr_segs--; 20837d58fe73SDavid Howells i->kvec++; 20847d58fe73SDavid Howells skip = 0; 20857d58fe73SDavid Howells } 20867d58fe73SDavid Howells 20877d58fe73SDavid Howells kaddr = i->kvec->iov_base + skip; 20887d58fe73SDavid Howells offset = (unsigned long)kaddr & ~PAGE_MASK; 20897d58fe73SDavid Howells *offset0 = offset; 20907d58fe73SDavid Howells 20917d58fe73SDavid Howells maxpages = want_pages_array(pages, maxsize, offset, maxpages); 20927d58fe73SDavid Howells if (!maxpages) 20937d58fe73SDavid Howells return -ENOMEM; 20947d58fe73SDavid Howells p = *pages; 20957d58fe73SDavid Howells 20967d58fe73SDavid Howells kaddr -= offset; 20977d58fe73SDavid Howells len = offset + maxsize; 20987d58fe73SDavid Howells for (k = 0; k < maxpages; k++) { 20997d58fe73SDavid Howells size_t seg = min_t(size_t, len, PAGE_SIZE); 21007d58fe73SDavid Howells 21017d58fe73SDavid Howells if (is_vmalloc_or_module_addr(kaddr)) 21027d58fe73SDavid Howells page = vmalloc_to_page(kaddr); 21037d58fe73SDavid Howells else 21047d58fe73SDavid Howells page = virt_to_page(kaddr); 21057d58fe73SDavid Howells 21067d58fe73SDavid Howells p[k] = page; 21077d58fe73SDavid Howells len -= seg; 21087d58fe73SDavid Howells kaddr += PAGE_SIZE; 21097d58fe73SDavid Howells } 21107d58fe73SDavid Howells 21117d58fe73SDavid Howells maxsize = min_t(size_t, maxsize, maxpages * PAGE_SIZE - offset); 21127d58fe73SDavid Howells iov_iter_advance(i, maxsize); 21137d58fe73SDavid Howells return maxsize; 21147d58fe73SDavid Howells } 21157d58fe73SDavid Howells 21167d58fe73SDavid Howells /* 21177d58fe73SDavid Howells * Extract a list of contiguous pages from a user iterator and get a pin on 21187d58fe73SDavid Howells * each of them. This should only be used if the iterator is user-backed 21197d58fe73SDavid Howells * (IOBUF/UBUF). 21207d58fe73SDavid Howells * 21217d58fe73SDavid Howells * It does not get refs on the pages, but the pages must be unpinned by the 21227d58fe73SDavid Howells * caller once the transfer is complete. 21237d58fe73SDavid Howells * 21247d58fe73SDavid Howells * This is safe to be used where background IO/DMA *is* going to be modifying 21257d58fe73SDavid Howells * the buffer; using a pin rather than a ref makes forces fork() to give the 21267d58fe73SDavid Howells * child a copy of the page. 21277d58fe73SDavid Howells */ 21287d58fe73SDavid Howells static ssize_t iov_iter_extract_user_pages(struct iov_iter *i, 21297d58fe73SDavid Howells struct page ***pages, 21307d58fe73SDavid Howells size_t maxsize, 21317d58fe73SDavid Howells unsigned int maxpages, 21327d58fe73SDavid Howells iov_iter_extraction_t extraction_flags, 21337d58fe73SDavid Howells size_t *offset0) 21347d58fe73SDavid Howells { 21357d58fe73SDavid Howells unsigned long addr; 21367d58fe73SDavid Howells unsigned int gup_flags = 0; 21377d58fe73SDavid Howells size_t offset; 21387d58fe73SDavid Howells int res; 21397d58fe73SDavid Howells 21407d58fe73SDavid Howells if (i->data_source == ITER_DEST) 21417d58fe73SDavid Howells gup_flags |= FOLL_WRITE; 21427d58fe73SDavid Howells if (extraction_flags & ITER_ALLOW_P2PDMA) 21437d58fe73SDavid Howells gup_flags |= FOLL_PCI_P2PDMA; 21447d58fe73SDavid Howells if (i->nofault) 21457d58fe73SDavid Howells gup_flags |= FOLL_NOFAULT; 21467d58fe73SDavid Howells 21477d58fe73SDavid Howells addr = first_iovec_segment(i, &maxsize); 21487d58fe73SDavid Howells *offset0 = offset = addr % PAGE_SIZE; 21497d58fe73SDavid Howells addr &= PAGE_MASK; 21507d58fe73SDavid Howells maxpages = want_pages_array(pages, maxsize, offset, maxpages); 21517d58fe73SDavid Howells if (!maxpages) 21527d58fe73SDavid Howells return -ENOMEM; 21537d58fe73SDavid Howells res = pin_user_pages_fast(addr, maxpages, gup_flags, *pages); 21547d58fe73SDavid Howells if (unlikely(res <= 0)) 21557d58fe73SDavid Howells return res; 21567d58fe73SDavid Howells maxsize = min_t(size_t, maxsize, res * PAGE_SIZE - offset); 21577d58fe73SDavid Howells iov_iter_advance(i, maxsize); 21587d58fe73SDavid Howells return maxsize; 21597d58fe73SDavid Howells } 21607d58fe73SDavid Howells 21617d58fe73SDavid Howells /** 21627d58fe73SDavid Howells * iov_iter_extract_pages - Extract a list of contiguous pages from an iterator 21637d58fe73SDavid Howells * @i: The iterator to extract from 21647d58fe73SDavid Howells * @pages: Where to return the list of pages 21657d58fe73SDavid Howells * @maxsize: The maximum amount of iterator to extract 21667d58fe73SDavid Howells * @maxpages: The maximum size of the list of pages 21677d58fe73SDavid Howells * @extraction_flags: Flags to qualify request 21687d58fe73SDavid Howells * @offset0: Where to return the starting offset into (*@pages)[0] 21697d58fe73SDavid Howells * 21707d58fe73SDavid Howells * Extract a list of contiguous pages from the current point of the iterator, 21717d58fe73SDavid Howells * advancing the iterator. The maximum number of pages and the maximum amount 21727d58fe73SDavid Howells * of page contents can be set. 21737d58fe73SDavid Howells * 21747d58fe73SDavid Howells * If *@pages is NULL, a page list will be allocated to the required size and 21757d58fe73SDavid Howells * *@pages will be set to its base. If *@pages is not NULL, it will be assumed 21767d58fe73SDavid Howells * that the caller allocated a page list at least @maxpages in size and this 21777d58fe73SDavid Howells * will be filled in. 21787d58fe73SDavid Howells * 21797d58fe73SDavid Howells * @extraction_flags can have ITER_ALLOW_P2PDMA set to request peer-to-peer DMA 21807d58fe73SDavid Howells * be allowed on the pages extracted. 21817d58fe73SDavid Howells * 21827d58fe73SDavid Howells * The iov_iter_extract_will_pin() function can be used to query how cleanup 21837d58fe73SDavid Howells * should be performed. 21847d58fe73SDavid Howells * 21857d58fe73SDavid Howells * Extra refs or pins on the pages may be obtained as follows: 21867d58fe73SDavid Howells * 21877d58fe73SDavid Howells * (*) If the iterator is user-backed (ITER_IOVEC/ITER_UBUF), pins will be 21887d58fe73SDavid Howells * added to the pages, but refs will not be taken. 21897d58fe73SDavid Howells * iov_iter_extract_will_pin() will return true. 21907d58fe73SDavid Howells * 21917d58fe73SDavid Howells * (*) If the iterator is ITER_KVEC, ITER_BVEC or ITER_XARRAY, the pages are 21927d58fe73SDavid Howells * merely listed; no extra refs or pins are obtained. 21937d58fe73SDavid Howells * iov_iter_extract_will_pin() will return 0. 21947d58fe73SDavid Howells * 21957d58fe73SDavid Howells * Note also: 21967d58fe73SDavid Howells * 21977d58fe73SDavid Howells * (*) Use with ITER_DISCARD is not supported as that has no content. 21987d58fe73SDavid Howells * 21997d58fe73SDavid Howells * On success, the function sets *@pages to the new pagelist, if allocated, and 22007d58fe73SDavid Howells * sets *offset0 to the offset into the first page. 22017d58fe73SDavid Howells * 22027d58fe73SDavid Howells * It may also return -ENOMEM and -EFAULT. 22037d58fe73SDavid Howells */ 22047d58fe73SDavid Howells ssize_t iov_iter_extract_pages(struct iov_iter *i, 22057d58fe73SDavid Howells struct page ***pages, 22067d58fe73SDavid Howells size_t maxsize, 22077d58fe73SDavid Howells unsigned int maxpages, 22087d58fe73SDavid Howells iov_iter_extraction_t extraction_flags, 22097d58fe73SDavid Howells size_t *offset0) 22107d58fe73SDavid Howells { 22117d58fe73SDavid Howells maxsize = min_t(size_t, min_t(size_t, maxsize, i->count), MAX_RW_COUNT); 22127d58fe73SDavid Howells if (!maxsize) 22137d58fe73SDavid Howells return 0; 22147d58fe73SDavid Howells 22157d58fe73SDavid Howells if (likely(user_backed_iter(i))) 22167d58fe73SDavid Howells return iov_iter_extract_user_pages(i, pages, maxsize, 22177d58fe73SDavid Howells maxpages, extraction_flags, 22187d58fe73SDavid Howells offset0); 22197d58fe73SDavid Howells if (iov_iter_is_kvec(i)) 22207d58fe73SDavid Howells return iov_iter_extract_kvec_pages(i, pages, maxsize, 22217d58fe73SDavid Howells maxpages, extraction_flags, 22227d58fe73SDavid Howells offset0); 22237d58fe73SDavid Howells if (iov_iter_is_bvec(i)) 22247d58fe73SDavid Howells return iov_iter_extract_bvec_pages(i, pages, maxsize, 22257d58fe73SDavid Howells maxpages, extraction_flags, 22267d58fe73SDavid Howells offset0); 22277d58fe73SDavid Howells if (iov_iter_is_xarray(i)) 22287d58fe73SDavid Howells return iov_iter_extract_xarray_pages(i, pages, maxsize, 22297d58fe73SDavid Howells maxpages, extraction_flags, 22307d58fe73SDavid Howells offset0); 22317d58fe73SDavid Howells return -EFAULT; 22327d58fe73SDavid Howells } 22337d58fe73SDavid Howells EXPORT_SYMBOL_GPL(iov_iter_extract_pages); 2234