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))) { \ 1295c67aa90SAl Viro const struct iovec *iov = i->iov; \ 1307baa5099SAl Viro void __user *base; \ 1317baa5099SAl Viro size_t len; \ 1327baa5099SAl Viro iterate_iovec(i, n, base, len, off, \ 133a6e4ec7bSAl Viro iov, (I)) \ 134d879cb83SAl Viro i->nr_segs -= iov - i->iov; \ 135d879cb83SAl Viro 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 { 1774d0e9df5SAlbert van der Linde if (should_fail_usercopy()) 1784d0e9df5SAlbert van der Linde return n; 17996d4f267SLinus Torvalds if (access_ok(from, n)) { 180d0ef4c36SMarco Elver instrument_copy_from_user(to, from, n); 18109fc68dcSAl Viro n = raw_copy_from_user(to, from, n); 18209fc68dcSAl Viro } 18309fc68dcSAl Viro return n; 18409fc68dcSAl Viro } 18509fc68dcSAl Viro 1862dcedb2aSAl Viro static inline struct pipe_buffer *pipe_buf(const struct pipe_inode_info *pipe, 1872dcedb2aSAl Viro unsigned int slot) 1882dcedb2aSAl Viro { 1892dcedb2aSAl Viro return &pipe->bufs[slot & (pipe->ring_size - 1)]; 1902dcedb2aSAl Viro } 1912dcedb2aSAl Viro 192241699cdSAl Viro #ifdef PIPE_PARANOIA 193241699cdSAl Viro static bool sanity(const struct iov_iter *i) 194241699cdSAl Viro { 195241699cdSAl Viro struct pipe_inode_info *pipe = i->pipe; 1968cefc107SDavid Howells unsigned int p_head = pipe->head; 1978cefc107SDavid Howells unsigned int p_tail = pipe->tail; 1988cefc107SDavid Howells unsigned int p_occupancy = pipe_occupancy(p_head, p_tail); 1998cefc107SDavid Howells unsigned int i_head = i->head; 2008cefc107SDavid Howells unsigned int idx; 2018cefc107SDavid Howells 20210f525a8SAl Viro if (i->last_offset) { 203241699cdSAl Viro struct pipe_buffer *p; 2048cefc107SDavid Howells if (unlikely(p_occupancy == 0)) 205241699cdSAl Viro goto Bad; // pipe must be non-empty 2068cefc107SDavid Howells if (unlikely(i_head != p_head - 1)) 207241699cdSAl Viro goto Bad; // must be at the last buffer... 208241699cdSAl Viro 2092dcedb2aSAl Viro p = pipe_buf(pipe, i_head); 21010f525a8SAl Viro if (unlikely(p->offset + p->len != abs(i->last_offset))) 211241699cdSAl Viro goto Bad; // ... at the end of segment 212241699cdSAl Viro } else { 2138cefc107SDavid Howells if (i_head != p_head) 214241699cdSAl Viro goto Bad; // must be right after the last buffer 215241699cdSAl Viro } 216241699cdSAl Viro return true; 217241699cdSAl Viro Bad: 21810f525a8SAl Viro printk(KERN_ERR "idx = %d, offset = %d\n", i_head, i->last_offset); 2198cefc107SDavid Howells printk(KERN_ERR "head = %d, tail = %d, buffers = %d\n", 2208cefc107SDavid Howells p_head, p_tail, pipe->ring_size); 2218cefc107SDavid Howells for (idx = 0; idx < pipe->ring_size; idx++) 222241699cdSAl Viro printk(KERN_ERR "[%p %p %d %d]\n", 223241699cdSAl Viro pipe->bufs[idx].ops, 224241699cdSAl Viro pipe->bufs[idx].page, 225241699cdSAl Viro pipe->bufs[idx].offset, 226241699cdSAl Viro pipe->bufs[idx].len); 227241699cdSAl Viro WARN_ON(1); 228241699cdSAl Viro return false; 229241699cdSAl Viro } 230241699cdSAl Viro #else 231241699cdSAl Viro #define sanity(i) true 232241699cdSAl Viro #endif 233241699cdSAl Viro 23447b7fcaeSAl Viro static struct page *push_anon(struct pipe_inode_info *pipe, unsigned size) 23547b7fcaeSAl Viro { 23647b7fcaeSAl Viro struct page *page = alloc_page(GFP_USER); 23747b7fcaeSAl Viro if (page) { 23847b7fcaeSAl Viro struct pipe_buffer *buf = pipe_buf(pipe, pipe->head++); 23947b7fcaeSAl Viro *buf = (struct pipe_buffer) { 24047b7fcaeSAl Viro .ops = &default_pipe_buf_ops, 24147b7fcaeSAl Viro .page = page, 24247b7fcaeSAl Viro .offset = 0, 24347b7fcaeSAl Viro .len = size 24447b7fcaeSAl Viro }; 24547b7fcaeSAl Viro } 24647b7fcaeSAl Viro return page; 24747b7fcaeSAl Viro } 24847b7fcaeSAl Viro 24947b7fcaeSAl Viro static void push_page(struct pipe_inode_info *pipe, struct page *page, 25047b7fcaeSAl Viro unsigned int offset, unsigned int size) 25147b7fcaeSAl Viro { 25247b7fcaeSAl Viro struct pipe_buffer *buf = pipe_buf(pipe, pipe->head++); 25347b7fcaeSAl Viro *buf = (struct pipe_buffer) { 25447b7fcaeSAl Viro .ops = &page_cache_pipe_buf_ops, 25547b7fcaeSAl Viro .page = page, 25647b7fcaeSAl Viro .offset = offset, 25747b7fcaeSAl Viro .len = size 25847b7fcaeSAl Viro }; 25947b7fcaeSAl Viro get_page(page); 26047b7fcaeSAl Viro } 26147b7fcaeSAl Viro 26210f525a8SAl Viro static inline int last_offset(const struct pipe_buffer *buf) 2638fad7767SAl Viro { 26410f525a8SAl Viro if (buf->ops == &default_pipe_buf_ops) 26510f525a8SAl Viro return buf->len; // buf->offset is 0 for those 26610f525a8SAl Viro else 26710f525a8SAl Viro return -(buf->offset + buf->len); 2688fad7767SAl Viro } 2698fad7767SAl Viro 2708fad7767SAl Viro static struct page *append_pipe(struct iov_iter *i, size_t size, 2718fad7767SAl Viro unsigned int *off) 2728fad7767SAl Viro { 2738fad7767SAl Viro struct pipe_inode_info *pipe = i->pipe; 27410f525a8SAl Viro int offset = i->last_offset; 2758fad7767SAl Viro struct pipe_buffer *buf; 2768fad7767SAl Viro struct page *page; 2778fad7767SAl Viro 27810f525a8SAl Viro if (offset > 0 && offset < PAGE_SIZE) { 27910f525a8SAl Viro // some space in the last buffer; add to it 2808fad7767SAl Viro buf = pipe_buf(pipe, pipe->head - 1); 2818fad7767SAl Viro size = min_t(size_t, size, PAGE_SIZE - offset); 2828fad7767SAl Viro buf->len += size; 28310f525a8SAl Viro i->last_offset += size; 2848fad7767SAl Viro i->count -= size; 2858fad7767SAl Viro *off = offset; 2868fad7767SAl Viro return buf->page; 2878fad7767SAl Viro } 2888fad7767SAl Viro // OK, we need a new buffer 2898fad7767SAl Viro *off = 0; 2908fad7767SAl Viro size = min_t(size_t, size, PAGE_SIZE); 2918fad7767SAl Viro if (pipe_full(pipe->head, pipe->tail, pipe->max_usage)) 2928fad7767SAl Viro return NULL; 2938fad7767SAl Viro page = push_anon(pipe, size); 2948fad7767SAl Viro if (!page) 2958fad7767SAl Viro return NULL; 2968fad7767SAl Viro i->head = pipe->head - 1; 29710f525a8SAl Viro i->last_offset = size; 2988fad7767SAl Viro i->count -= size; 2998fad7767SAl Viro return page; 3008fad7767SAl Viro } 3018fad7767SAl Viro 302241699cdSAl Viro static size_t copy_page_to_iter_pipe(struct page *page, size_t offset, size_t bytes, 303241699cdSAl Viro struct iov_iter *i) 304241699cdSAl Viro { 305241699cdSAl Viro struct pipe_inode_info *pipe = i->pipe; 30647b7fcaeSAl Viro unsigned int head = pipe->head; 307241699cdSAl Viro 308241699cdSAl Viro if (unlikely(bytes > i->count)) 309241699cdSAl Viro bytes = i->count; 310241699cdSAl Viro 311241699cdSAl Viro if (unlikely(!bytes)) 312241699cdSAl Viro return 0; 313241699cdSAl Viro 314241699cdSAl Viro if (!sanity(i)) 315241699cdSAl Viro return 0; 316241699cdSAl Viro 31710f525a8SAl Viro if (offset && i->last_offset == -offset) { // could we merge it? 31847b7fcaeSAl Viro struct pipe_buffer *buf = pipe_buf(pipe, head - 1); 31947b7fcaeSAl Viro if (buf->page == page) { 320241699cdSAl Viro buf->len += bytes; 32110f525a8SAl Viro i->last_offset -= bytes; 32247b7fcaeSAl Viro i->count -= bytes; 32347b7fcaeSAl Viro return bytes; 324241699cdSAl Viro } 325241699cdSAl Viro } 32647b7fcaeSAl Viro if (pipe_full(pipe->head, pipe->tail, pipe->max_usage)) 327241699cdSAl Viro return 0; 3288cefc107SDavid Howells 32947b7fcaeSAl Viro push_page(pipe, page, offset, bytes); 33010f525a8SAl Viro i->last_offset = -(offset + bytes); 33147b7fcaeSAl Viro i->head = head; 332241699cdSAl Viro i->count -= bytes; 333241699cdSAl Viro return bytes; 334241699cdSAl Viro } 335241699cdSAl Viro 336d879cb83SAl Viro /* 337a6294593SAndreas Gruenbacher * fault_in_iov_iter_readable - fault in iov iterator for reading 338a6294593SAndreas Gruenbacher * @i: iterator 339a6294593SAndreas Gruenbacher * @size: maximum length 340171a0203SAnton Altaparmakov * 341a6294593SAndreas Gruenbacher * Fault in one or more iovecs of the given iov_iter, to a maximum length of 342a6294593SAndreas Gruenbacher * @size. For each iovec, fault in each page that constitutes the iovec. 343a6294593SAndreas Gruenbacher * 344a6294593SAndreas Gruenbacher * Returns the number of bytes not faulted in (like copy_to_user() and 345a6294593SAndreas Gruenbacher * copy_from_user()). 346a6294593SAndreas Gruenbacher * 347a6294593SAndreas Gruenbacher * Always returns 0 for non-userspace iterators. 348171a0203SAnton Altaparmakov */ 349a6294593SAndreas Gruenbacher size_t fault_in_iov_iter_readable(const struct iov_iter *i, size_t size) 350171a0203SAnton Altaparmakov { 351fcb14cb1SAl Viro if (iter_is_ubuf(i)) { 352fcb14cb1SAl Viro size_t n = min(size, iov_iter_count(i)); 353fcb14cb1SAl Viro n -= fault_in_readable(i->ubuf + i->iov_offset, n); 354fcb14cb1SAl Viro return size - n; 355fcb14cb1SAl Viro } else if (iter_is_iovec(i)) { 356a6294593SAndreas Gruenbacher size_t count = min(size, iov_iter_count(i)); 3578409a0d2SAl Viro const struct iovec *p; 3588409a0d2SAl Viro size_t skip; 3598409a0d2SAl Viro 360a6294593SAndreas Gruenbacher size -= count; 361a6294593SAndreas Gruenbacher for (p = i->iov, skip = i->iov_offset; count; p++, skip = 0) { 362a6294593SAndreas Gruenbacher size_t len = min(count, p->iov_len - skip); 363a6294593SAndreas Gruenbacher size_t ret; 3648409a0d2SAl Viro 3658409a0d2SAl Viro if (unlikely(!len)) 3668409a0d2SAl Viro continue; 367a6294593SAndreas Gruenbacher ret = fault_in_readable(p->iov_base + skip, len); 368a6294593SAndreas Gruenbacher count -= len - ret; 369a6294593SAndreas Gruenbacher if (ret) 370a6294593SAndreas Gruenbacher break; 3718409a0d2SAl Viro } 372a6294593SAndreas Gruenbacher return count + size; 373171a0203SAnton Altaparmakov } 374171a0203SAnton Altaparmakov return 0; 375171a0203SAnton Altaparmakov } 376a6294593SAndreas Gruenbacher EXPORT_SYMBOL(fault_in_iov_iter_readable); 377171a0203SAnton Altaparmakov 378cdd591fcSAndreas Gruenbacher /* 379cdd591fcSAndreas Gruenbacher * fault_in_iov_iter_writeable - fault in iov iterator for writing 380cdd591fcSAndreas Gruenbacher * @i: iterator 381cdd591fcSAndreas Gruenbacher * @size: maximum length 382cdd591fcSAndreas Gruenbacher * 383cdd591fcSAndreas Gruenbacher * Faults in the iterator using get_user_pages(), i.e., without triggering 384cdd591fcSAndreas Gruenbacher * hardware page faults. This is primarily useful when we already know that 385cdd591fcSAndreas Gruenbacher * some or all of the pages in @i aren't in memory. 386cdd591fcSAndreas Gruenbacher * 387cdd591fcSAndreas Gruenbacher * Returns the number of bytes not faulted in, like copy_to_user() and 388cdd591fcSAndreas Gruenbacher * copy_from_user(). 389cdd591fcSAndreas Gruenbacher * 390cdd591fcSAndreas Gruenbacher * Always returns 0 for non-user-space iterators. 391cdd591fcSAndreas Gruenbacher */ 392cdd591fcSAndreas Gruenbacher size_t fault_in_iov_iter_writeable(const struct iov_iter *i, size_t size) 393cdd591fcSAndreas Gruenbacher { 394fcb14cb1SAl Viro if (iter_is_ubuf(i)) { 395fcb14cb1SAl Viro size_t n = min(size, iov_iter_count(i)); 396fcb14cb1SAl Viro n -= fault_in_safe_writeable(i->ubuf + i->iov_offset, n); 397fcb14cb1SAl Viro return size - n; 398fcb14cb1SAl Viro } else if (iter_is_iovec(i)) { 399cdd591fcSAndreas Gruenbacher size_t count = min(size, iov_iter_count(i)); 400cdd591fcSAndreas Gruenbacher const struct iovec *p; 401cdd591fcSAndreas Gruenbacher size_t skip; 402cdd591fcSAndreas Gruenbacher 403cdd591fcSAndreas Gruenbacher size -= count; 404cdd591fcSAndreas Gruenbacher for (p = i->iov, skip = i->iov_offset; count; p++, skip = 0) { 405cdd591fcSAndreas Gruenbacher size_t len = min(count, p->iov_len - skip); 406cdd591fcSAndreas Gruenbacher size_t ret; 407cdd591fcSAndreas Gruenbacher 408cdd591fcSAndreas Gruenbacher if (unlikely(!len)) 409cdd591fcSAndreas Gruenbacher continue; 410cdd591fcSAndreas Gruenbacher ret = fault_in_safe_writeable(p->iov_base + skip, len); 411cdd591fcSAndreas Gruenbacher count -= len - ret; 412cdd591fcSAndreas Gruenbacher if (ret) 413cdd591fcSAndreas Gruenbacher break; 414cdd591fcSAndreas Gruenbacher } 415cdd591fcSAndreas Gruenbacher return count + size; 416cdd591fcSAndreas Gruenbacher } 417cdd591fcSAndreas Gruenbacher return 0; 418cdd591fcSAndreas Gruenbacher } 419cdd591fcSAndreas Gruenbacher EXPORT_SYMBOL(fault_in_iov_iter_writeable); 420cdd591fcSAndreas Gruenbacher 421aa563d7bSDavid Howells void iov_iter_init(struct iov_iter *i, unsigned int direction, 422d879cb83SAl Viro const struct iovec *iov, unsigned long nr_segs, 423d879cb83SAl Viro size_t count) 424d879cb83SAl Viro { 425aa563d7bSDavid Howells WARN_ON(direction & ~(READ | WRITE)); 4268cd54c1cSAl Viro *i = (struct iov_iter) { 4278cd54c1cSAl Viro .iter_type = ITER_IOVEC, 4283337ab08SAndreas Gruenbacher .nofault = false, 429fcb14cb1SAl Viro .user_backed = true, 4308cd54c1cSAl Viro .data_source = direction, 4318cd54c1cSAl Viro .iov = iov, 4328cd54c1cSAl Viro .nr_segs = nr_segs, 4338cd54c1cSAl Viro .iov_offset = 0, 4348cd54c1cSAl Viro .count = count 4358cd54c1cSAl Viro }; 436d879cb83SAl Viro } 437d879cb83SAl Viro EXPORT_SYMBOL(iov_iter_init); 438d879cb83SAl Viro 43912d426abSAl Viro // returns the offset in partial buffer (if any) 44012d426abSAl Viro static inline unsigned int pipe_npages(const struct iov_iter *i, int *npages) 441241699cdSAl Viro { 44212d426abSAl Viro struct pipe_inode_info *pipe = i->pipe; 44312d426abSAl Viro int used = pipe->head - pipe->tail; 44410f525a8SAl Viro int off = i->last_offset; 4458cefc107SDavid Howells 44612d426abSAl Viro *npages = max((int)pipe->max_usage - used, 0); 44712d426abSAl Viro 44810f525a8SAl Viro if (off > 0 && off < PAGE_SIZE) { // anon and not full 44912d426abSAl Viro (*npages)++; 45012d426abSAl Viro return off; 45110f525a8SAl Viro } 45212d426abSAl Viro return 0; 453241699cdSAl Viro } 454241699cdSAl Viro 455241699cdSAl Viro static size_t copy_pipe_to_iter(const void *addr, size_t bytes, 456241699cdSAl Viro struct iov_iter *i) 457241699cdSAl Viro { 4588fad7767SAl Viro unsigned int off, chunk; 4598fad7767SAl Viro 4608fad7767SAl Viro if (unlikely(bytes > i->count)) 4618fad7767SAl Viro bytes = i->count; 4628fad7767SAl Viro if (unlikely(!bytes)) 4638fad7767SAl Viro return 0; 464241699cdSAl Viro 465241699cdSAl Viro if (!sanity(i)) 466241699cdSAl Viro return 0; 467241699cdSAl Viro 4688fad7767SAl Viro for (size_t n = bytes; n; n -= chunk) { 4698fad7767SAl Viro struct page *page = append_pipe(i, n, &off); 4708fad7767SAl Viro chunk = min_t(size_t, n, PAGE_SIZE - off); 4718fad7767SAl Viro if (!page) 4728fad7767SAl Viro return bytes - n; 4738fad7767SAl Viro memcpy_to_page(page, off, addr, chunk); 474241699cdSAl Viro addr += chunk; 4758fad7767SAl Viro } 476241699cdSAl Viro return bytes; 477241699cdSAl Viro } 478241699cdSAl Viro 479f9152895SAl Viro static __wsum csum_and_memcpy(void *to, const void *from, size_t len, 480f9152895SAl Viro __wsum sum, size_t off) 481f9152895SAl Viro { 482cc44c17bSAl Viro __wsum next = csum_partial_copy_nocheck(from, to, len); 483f9152895SAl Viro return csum_block_add(sum, next, off); 484f9152895SAl Viro } 485f9152895SAl Viro 48678e1f386SAl Viro static size_t csum_and_copy_to_pipe_iter(const void *addr, size_t bytes, 4876852df12SAl Viro struct iov_iter *i, __wsum *sump) 48878e1f386SAl Viro { 4896852df12SAl Viro __wsum sum = *sump; 4906852df12SAl Viro size_t off = 0; 4918fad7767SAl Viro unsigned int chunk, r; 4928fad7767SAl Viro 4938fad7767SAl Viro if (unlikely(bytes > i->count)) 4948fad7767SAl Viro bytes = i->count; 4958fad7767SAl Viro if (unlikely(!bytes)) 4968fad7767SAl Viro return 0; 49778e1f386SAl Viro 49878e1f386SAl Viro if (!sanity(i)) 49978e1f386SAl Viro return 0; 50078e1f386SAl Viro 5016852df12SAl Viro while (bytes) { 5028fad7767SAl Viro struct page *page = append_pipe(i, bytes, &r); 5038fad7767SAl Viro char *p; 5048fad7767SAl Viro 5058fad7767SAl Viro if (!page) 5068fad7767SAl Viro break; 5078fad7767SAl Viro chunk = min_t(size_t, bytes, PAGE_SIZE - r); 5088fad7767SAl Viro p = kmap_local_page(page); 5096852df12SAl Viro sum = csum_and_memcpy(p + r, addr + off, chunk, sum, off); 5102495bdccSAl Viro kunmap_local(p); 51178e1f386SAl Viro off += chunk; 5128fad7767SAl Viro bytes -= chunk; 5136852df12SAl Viro } 5146852df12SAl Viro *sump = sum; 5156852df12SAl Viro return off; 51678e1f386SAl Viro } 51778e1f386SAl Viro 518aa28de27SAl Viro size_t _copy_to_iter(const void *addr, size_t bytes, struct iov_iter *i) 519d879cb83SAl Viro { 52000e23707SDavid Howells if (unlikely(iov_iter_is_pipe(i))) 521241699cdSAl Viro return copy_pipe_to_iter(addr, bytes, i); 522fcb14cb1SAl Viro if (user_backed_iter(i)) 52309fc68dcSAl Viro might_fault(); 5247baa5099SAl Viro iterate_and_advance(i, bytes, base, len, off, 5257baa5099SAl Viro copyout(base, addr + off, len), 5267baa5099SAl Viro memcpy(base, addr + off, len) 527d879cb83SAl Viro ) 528d879cb83SAl Viro 529d879cb83SAl Viro return bytes; 530d879cb83SAl Viro } 531aa28de27SAl Viro EXPORT_SYMBOL(_copy_to_iter); 532d879cb83SAl Viro 533ec6347bbSDan Williams #ifdef CONFIG_ARCH_HAS_COPY_MC 534ec6347bbSDan Williams static int copyout_mc(void __user *to, const void *from, size_t n) 5358780356eSDan Williams { 53696d4f267SLinus Torvalds if (access_ok(to, n)) { 537d0ef4c36SMarco Elver instrument_copy_to_user(to, from, n); 538ec6347bbSDan Williams n = copy_mc_to_user((__force void *) to, from, n); 5398780356eSDan Williams } 5408780356eSDan Williams return n; 5418780356eSDan Williams } 5428780356eSDan Williams 543ec6347bbSDan Williams static size_t copy_mc_pipe_to_iter(const void *addr, size_t bytes, 544ca146f6fSDan Williams struct iov_iter *i) 545ca146f6fSDan Williams { 5468fad7767SAl Viro size_t xfer = 0; 5478fad7767SAl Viro unsigned int off, chunk; 5488fad7767SAl Viro 5498fad7767SAl Viro if (unlikely(bytes > i->count)) 5508fad7767SAl Viro bytes = i->count; 5518fad7767SAl Viro if (unlikely(!bytes)) 5528fad7767SAl Viro return 0; 553ca146f6fSDan Williams 554ca146f6fSDan Williams if (!sanity(i)) 555ca146f6fSDan Williams return 0; 556ca146f6fSDan Williams 5578fad7767SAl Viro while (bytes) { 5588fad7767SAl Viro struct page *page = append_pipe(i, bytes, &off); 559ca146f6fSDan Williams unsigned long rem; 5608fad7767SAl Viro char *p; 5618fad7767SAl Viro 5628fad7767SAl Viro if (!page) 5638fad7767SAl Viro break; 5648fad7767SAl Viro chunk = min_t(size_t, bytes, PAGE_SIZE - off); 5658fad7767SAl Viro p = kmap_local_page(page); 5662a510a74SAl Viro rem = copy_mc_to_kernel(p + off, addr + xfer, chunk); 5672a510a74SAl Viro chunk -= rem; 5682a510a74SAl Viro kunmap_local(p); 5692a510a74SAl Viro xfer += chunk; 5708fad7767SAl Viro bytes -= chunk; 571c3497fd0SAl Viro if (rem) { 5728fad7767SAl Viro iov_iter_revert(i, rem); 573ca146f6fSDan Williams break; 574c3497fd0SAl Viro } 5752a510a74SAl Viro } 576ca146f6fSDan Williams return xfer; 577ca146f6fSDan Williams } 578ca146f6fSDan Williams 579bf3eeb9bSDan Williams /** 580ec6347bbSDan Williams * _copy_mc_to_iter - copy to iter with source memory error exception handling 581bf3eeb9bSDan Williams * @addr: source kernel address 582bf3eeb9bSDan Williams * @bytes: total transfer length 58344e55997SRandy Dunlap * @i: destination iterator 584bf3eeb9bSDan Williams * 585ec6347bbSDan Williams * The pmem driver deploys this for the dax operation 586ec6347bbSDan Williams * (dax_copy_to_iter()) for dax reads (bypass page-cache and the 587ec6347bbSDan Williams * block-layer). Upon #MC read(2) aborts and returns EIO or the bytes 588ec6347bbSDan Williams * successfully copied. 589bf3eeb9bSDan Williams * 590ec6347bbSDan Williams * The main differences between this and typical _copy_to_iter(). 591bf3eeb9bSDan Williams * 592bf3eeb9bSDan Williams * * Typical tail/residue handling after a fault retries the copy 593bf3eeb9bSDan Williams * byte-by-byte until the fault happens again. Re-triggering machine 594bf3eeb9bSDan Williams * checks is potentially fatal so the implementation uses source 595bf3eeb9bSDan Williams * alignment and poison alignment assumptions to avoid re-triggering 596bf3eeb9bSDan Williams * hardware exceptions. 597bf3eeb9bSDan Williams * 598bf3eeb9bSDan Williams * * ITER_KVEC, ITER_PIPE, and ITER_BVEC can return short copies. 599bf3eeb9bSDan Williams * Compare to copy_to_iter() where only ITER_IOVEC attempts might return 600bf3eeb9bSDan Williams * a short copy. 60144e55997SRandy Dunlap * 60244e55997SRandy Dunlap * Return: number of bytes copied (may be %0) 603bf3eeb9bSDan Williams */ 604ec6347bbSDan Williams size_t _copy_mc_to_iter(const void *addr, size_t bytes, struct iov_iter *i) 6058780356eSDan Williams { 60600e23707SDavid Howells if (unlikely(iov_iter_is_pipe(i))) 607ec6347bbSDan Williams return copy_mc_pipe_to_iter(addr, bytes, i); 608fcb14cb1SAl Viro if (user_backed_iter(i)) 6098780356eSDan Williams might_fault(); 6107baa5099SAl Viro __iterate_and_advance(i, bytes, base, len, off, 6117baa5099SAl Viro copyout_mc(base, addr + off, len), 6127baa5099SAl Viro copy_mc_to_kernel(base, addr + off, len) 6138780356eSDan Williams ) 6148780356eSDan Williams 6158780356eSDan Williams return bytes; 6168780356eSDan Williams } 617ec6347bbSDan Williams EXPORT_SYMBOL_GPL(_copy_mc_to_iter); 618ec6347bbSDan Williams #endif /* CONFIG_ARCH_HAS_COPY_MC */ 6198780356eSDan Williams 620aa28de27SAl Viro size_t _copy_from_iter(void *addr, size_t bytes, struct iov_iter *i) 621d879cb83SAl Viro { 62200e23707SDavid Howells if (unlikely(iov_iter_is_pipe(i))) { 623241699cdSAl Viro WARN_ON(1); 624241699cdSAl Viro return 0; 625241699cdSAl 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 { 63900e23707SDavid Howells if (unlikely(iov_iter_is_pipe(i))) { 640241699cdSAl Viro WARN_ON(1); 641241699cdSAl Viro return 0; 642241699cdSAl Viro } 6437baa5099SAl Viro iterate_and_advance(i, bytes, base, len, off, 6447baa5099SAl Viro __copy_from_user_inatomic_nocache(addr + off, base, len), 6457baa5099SAl Viro memcpy(addr + off, base, len) 646d879cb83SAl Viro ) 647d879cb83SAl Viro 648d879cb83SAl Viro return bytes; 649d879cb83SAl Viro } 650aa28de27SAl Viro EXPORT_SYMBOL(_copy_from_iter_nocache); 651d879cb83SAl Viro 6520aed55afSDan Williams #ifdef CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE 653abd08d7dSDan Williams /** 654abd08d7dSDan Williams * _copy_from_iter_flushcache - write destination through cpu cache 655abd08d7dSDan Williams * @addr: destination kernel address 656abd08d7dSDan Williams * @bytes: total transfer length 65744e55997SRandy Dunlap * @i: source iterator 658abd08d7dSDan Williams * 659abd08d7dSDan Williams * The pmem driver arranges for filesystem-dax to use this facility via 660abd08d7dSDan Williams * dax_copy_from_iter() for ensuring that writes to persistent memory 661abd08d7dSDan Williams * are flushed through the CPU cache. It is differentiated from 662abd08d7dSDan Williams * _copy_from_iter_nocache() in that guarantees all data is flushed for 663abd08d7dSDan Williams * all iterator types. The _copy_from_iter_nocache() only attempts to 664abd08d7dSDan Williams * bypass the cache for the ITER_IOVEC case, and on some archs may use 665abd08d7dSDan Williams * instructions that strand dirty-data in the cache. 66644e55997SRandy Dunlap * 66744e55997SRandy Dunlap * Return: number of bytes copied (may be %0) 668abd08d7dSDan Williams */ 6696a37e940SLinus Torvalds size_t _copy_from_iter_flushcache(void *addr, size_t bytes, struct iov_iter *i) 6700aed55afSDan Williams { 67100e23707SDavid Howells if (unlikely(iov_iter_is_pipe(i))) { 6720aed55afSDan Williams WARN_ON(1); 6730aed55afSDan Williams return 0; 6740aed55afSDan Williams } 6757baa5099SAl Viro iterate_and_advance(i, bytes, base, len, off, 6767baa5099SAl Viro __copy_from_user_flushcache(addr + off, base, len), 6777baa5099SAl Viro memcpy_flushcache(addr + off, base, len) 6780aed55afSDan Williams ) 6790aed55afSDan Williams 6800aed55afSDan Williams return bytes; 6810aed55afSDan Williams } 6826a37e940SLinus Torvalds EXPORT_SYMBOL_GPL(_copy_from_iter_flushcache); 6830aed55afSDan Williams #endif 6840aed55afSDan Williams 68572e809edSAl Viro static inline bool page_copy_sane(struct page *page, size_t offset, size_t n) 68672e809edSAl Viro { 6876daef95bSEric Dumazet struct page *head; 6886daef95bSEric Dumazet size_t v = n + offset; 6896daef95bSEric Dumazet 6906daef95bSEric Dumazet /* 6916daef95bSEric Dumazet * The general case needs to access the page order in order 6926daef95bSEric Dumazet * to compute the page size. 6936daef95bSEric Dumazet * However, we mostly deal with order-0 pages and thus can 6946daef95bSEric Dumazet * avoid a possible cache line miss for requests that fit all 6956daef95bSEric Dumazet * page orders. 6966daef95bSEric Dumazet */ 6976daef95bSEric Dumazet if (n <= v && v <= PAGE_SIZE) 6986daef95bSEric Dumazet return true; 6996daef95bSEric Dumazet 7006daef95bSEric Dumazet head = compound_head(page); 7016daef95bSEric Dumazet v += (page - head) << PAGE_SHIFT; 702a90bcb86SPetar Penkov 703a50b854eSMatthew Wilcox (Oracle) if (likely(n <= v && v <= (page_size(head)))) 70472e809edSAl Viro return true; 70572e809edSAl Viro WARN_ON(1); 70672e809edSAl Viro return false; 70772e809edSAl Viro } 708cbbd26b8SAl Viro 70908aa6479SAl Viro static size_t __copy_page_to_iter(struct page *page, size_t offset, size_t bytes, 710d879cb83SAl Viro struct iov_iter *i) 711d879cb83SAl Viro { 71259bb69c6SAl Viro if (unlikely(iov_iter_is_pipe(i))) { 71359bb69c6SAl Viro return copy_page_to_iter_pipe(page, offset, bytes, i); 71459bb69c6SAl Viro } else { 715c1d4d6a9SAl Viro void *kaddr = kmap_local_page(page); 716c1d4d6a9SAl Viro size_t wanted = _copy_to_iter(kaddr + offset, bytes, i); 717c1d4d6a9SAl Viro kunmap_local(kaddr); 718d879cb83SAl Viro return wanted; 71928f38db7SAl Viro } 720d879cb83SAl Viro } 72108aa6479SAl Viro 72208aa6479SAl Viro size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes, 72308aa6479SAl Viro struct iov_iter *i) 72408aa6479SAl Viro { 72508aa6479SAl Viro size_t res = 0; 72608aa6479SAl Viro if (unlikely(!page_copy_sane(page, offset, bytes))) 72708aa6479SAl Viro return 0; 72808aa6479SAl Viro page += offset / PAGE_SIZE; // first subpage 72908aa6479SAl Viro offset %= PAGE_SIZE; 73008aa6479SAl Viro while (1) { 73108aa6479SAl Viro size_t n = __copy_page_to_iter(page, offset, 73208aa6479SAl Viro min(bytes, (size_t)PAGE_SIZE - offset), i); 73308aa6479SAl Viro res += n; 73408aa6479SAl Viro bytes -= n; 73508aa6479SAl Viro if (!bytes || !n) 73608aa6479SAl Viro break; 73708aa6479SAl Viro offset += n; 73808aa6479SAl Viro if (offset == PAGE_SIZE) { 73908aa6479SAl Viro page++; 74008aa6479SAl Viro offset = 0; 74108aa6479SAl Viro } 74208aa6479SAl Viro } 74308aa6479SAl Viro return res; 74408aa6479SAl Viro } 745d879cb83SAl Viro EXPORT_SYMBOL(copy_page_to_iter); 746d879cb83SAl Viro 747d879cb83SAl Viro size_t copy_page_from_iter(struct page *page, size_t offset, size_t bytes, 748d879cb83SAl Viro struct iov_iter *i) 749d879cb83SAl Viro { 75059bb69c6SAl Viro if (page_copy_sane(page, offset, bytes)) { 75155ca375cSAl Viro void *kaddr = kmap_local_page(page); 752aa28de27SAl Viro size_t wanted = _copy_from_iter(kaddr + offset, bytes, i); 75355ca375cSAl Viro kunmap_local(kaddr); 754d879cb83SAl Viro return wanted; 75528f38db7SAl Viro } 75628f38db7SAl Viro return 0; 757d879cb83SAl Viro } 758d879cb83SAl Viro EXPORT_SYMBOL(copy_page_from_iter); 759d879cb83SAl Viro 760241699cdSAl Viro static size_t pipe_zero(size_t bytes, struct iov_iter *i) 761241699cdSAl Viro { 7628fad7767SAl Viro unsigned int chunk, off; 7638fad7767SAl Viro 7648fad7767SAl Viro if (unlikely(bytes > i->count)) 7658fad7767SAl Viro bytes = i->count; 7668fad7767SAl Viro if (unlikely(!bytes)) 7678fad7767SAl Viro return 0; 768241699cdSAl Viro 769241699cdSAl Viro if (!sanity(i)) 770241699cdSAl Viro return 0; 771241699cdSAl Viro 7728fad7767SAl Viro for (size_t n = bytes; n; n -= chunk) { 7738fad7767SAl Viro struct page *page = append_pipe(i, n, &off); 7748fad7767SAl Viro char *p; 775241699cdSAl Viro 7768fad7767SAl Viro if (!page) 7778fad7767SAl Viro return bytes - n; 7788fad7767SAl Viro chunk = min_t(size_t, n, PAGE_SIZE - off); 7798fad7767SAl Viro p = kmap_local_page(page); 780893839fdSAl Viro memset(p + off, 0, chunk); 781893839fdSAl Viro kunmap_local(p); 7828fad7767SAl Viro } 783241699cdSAl Viro return bytes; 784241699cdSAl Viro } 785241699cdSAl Viro 786d879cb83SAl Viro size_t iov_iter_zero(size_t bytes, struct iov_iter *i) 787d879cb83SAl Viro { 78800e23707SDavid Howells if (unlikely(iov_iter_is_pipe(i))) 789241699cdSAl Viro return pipe_zero(bytes, i); 7907baa5099SAl Viro iterate_and_advance(i, bytes, base, len, count, 7917baa5099SAl Viro clear_user(base, len), 7927baa5099SAl Viro memset(base, 0, len) 793d879cb83SAl Viro ) 794d879cb83SAl Viro 795d879cb83SAl Viro return bytes; 796d879cb83SAl Viro } 797d879cb83SAl Viro EXPORT_SYMBOL(iov_iter_zero); 798d879cb83SAl Viro 799f0b65f39SAl Viro size_t copy_page_from_iter_atomic(struct page *page, unsigned offset, size_t bytes, 800f0b65f39SAl Viro struct iov_iter *i) 801d879cb83SAl Viro { 802d879cb83SAl Viro char *kaddr = kmap_atomic(page), *p = kaddr + offset; 80372e809edSAl Viro if (unlikely(!page_copy_sane(page, offset, bytes))) { 80472e809edSAl Viro kunmap_atomic(kaddr); 80572e809edSAl Viro return 0; 80672e809edSAl Viro } 8079ea9ce04SDavid Howells if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) { 808241699cdSAl Viro kunmap_atomic(kaddr); 809241699cdSAl Viro WARN_ON(1); 810241699cdSAl Viro return 0; 811241699cdSAl Viro } 8127baa5099SAl Viro iterate_and_advance(i, bytes, base, len, off, 8137baa5099SAl Viro copyin(p + off, base, len), 8147baa5099SAl Viro memcpy(p + off, base, len) 815d879cb83SAl Viro ) 816d879cb83SAl Viro kunmap_atomic(kaddr); 817d879cb83SAl Viro return bytes; 818d879cb83SAl Viro } 819f0b65f39SAl Viro EXPORT_SYMBOL(copy_page_from_iter_atomic); 820d879cb83SAl Viro 821b9dc6f65SAl Viro static void pipe_advance(struct iov_iter *i, size_t size) 822b9dc6f65SAl Viro { 823b9dc6f65SAl Viro struct pipe_inode_info *pipe = i->pipe; 82410f525a8SAl Viro int off = i->last_offset; 8258cefc107SDavid Howells 8262c855de9SAl Viro if (!off && !size) { 8272c855de9SAl Viro pipe_discard_from(pipe, i->start_head); // discard everything 8282c855de9SAl Viro return; 829b9dc6f65SAl Viro } 830b9dc6f65SAl Viro i->count -= size; 8312c855de9SAl Viro while (1) { 8322c855de9SAl Viro struct pipe_buffer *buf = pipe_buf(pipe, i->head); 8332c855de9SAl Viro if (off) /* make it relative to the beginning of buffer */ 83410f525a8SAl Viro size += abs(off) - buf->offset; 8352c855de9SAl Viro if (size <= buf->len) { 8362c855de9SAl Viro buf->len = size; 83710f525a8SAl Viro i->last_offset = last_offset(buf); 8382c855de9SAl Viro break; 8392c855de9SAl Viro } 8402c855de9SAl Viro size -= buf->len; 8412c855de9SAl Viro i->head++; 8422c855de9SAl Viro off = 0; 8432c855de9SAl Viro } 8442c855de9SAl Viro pipe_discard_from(pipe, i->head + 1); // discard everything past this one 845241699cdSAl Viro } 846241699cdSAl Viro 84754c8195bSPavel Begunkov static void iov_iter_bvec_advance(struct iov_iter *i, size_t size) 84854c8195bSPavel Begunkov { 84918fa9af7SAl Viro const struct bio_vec *bvec, *end; 85054c8195bSPavel Begunkov 85118fa9af7SAl Viro if (!i->count) 85218fa9af7SAl Viro return; 85318fa9af7SAl Viro i->count -= size; 85454c8195bSPavel Begunkov 85518fa9af7SAl Viro size += i->iov_offset; 85618fa9af7SAl Viro 85718fa9af7SAl Viro for (bvec = i->bvec, end = bvec + i->nr_segs; bvec < end; bvec++) { 85818fa9af7SAl Viro if (likely(size < bvec->bv_len)) 85918fa9af7SAl Viro break; 86018fa9af7SAl Viro size -= bvec->bv_len; 86118fa9af7SAl Viro } 86218fa9af7SAl Viro i->iov_offset = size; 86318fa9af7SAl Viro i->nr_segs -= bvec - i->bvec; 86418fa9af7SAl Viro i->bvec = bvec; 86554c8195bSPavel Begunkov } 86654c8195bSPavel Begunkov 867185ac4d4SAl Viro static void iov_iter_iovec_advance(struct iov_iter *i, size_t size) 868185ac4d4SAl Viro { 869185ac4d4SAl Viro const struct iovec *iov, *end; 870185ac4d4SAl Viro 871185ac4d4SAl Viro if (!i->count) 872185ac4d4SAl Viro return; 873185ac4d4SAl Viro i->count -= size; 874185ac4d4SAl Viro 875185ac4d4SAl Viro size += i->iov_offset; // from beginning of current segment 876185ac4d4SAl Viro for (iov = i->iov, end = iov + i->nr_segs; iov < end; iov++) { 877185ac4d4SAl Viro if (likely(size < iov->iov_len)) 878185ac4d4SAl Viro break; 879185ac4d4SAl Viro size -= iov->iov_len; 880185ac4d4SAl Viro } 881185ac4d4SAl Viro i->iov_offset = size; 882185ac4d4SAl Viro i->nr_segs -= iov - i->iov; 883185ac4d4SAl Viro i->iov = iov; 884185ac4d4SAl Viro } 885185ac4d4SAl Viro 886d879cb83SAl Viro void iov_iter_advance(struct iov_iter *i, size_t size) 887d879cb83SAl Viro { 8883b3fc051SAl Viro if (unlikely(i->count < size)) 8893b3fc051SAl Viro size = i->count; 890fcb14cb1SAl Viro if (likely(iter_is_ubuf(i)) || unlikely(iov_iter_is_xarray(i))) { 891fcb14cb1SAl Viro i->iov_offset += size; 892fcb14cb1SAl Viro i->count -= size; 893fcb14cb1SAl Viro } else if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) { 894185ac4d4SAl Viro /* iovec and kvec have identical layouts */ 895185ac4d4SAl Viro iov_iter_iovec_advance(i, size); 896185ac4d4SAl Viro } else if (iov_iter_is_bvec(i)) { 897185ac4d4SAl Viro iov_iter_bvec_advance(i, size); 898185ac4d4SAl Viro } else if (iov_iter_is_pipe(i)) { 899241699cdSAl Viro pipe_advance(i, size); 900185ac4d4SAl Viro } else if (iov_iter_is_discard(i)) { 901185ac4d4SAl Viro i->count -= size; 9027ff50620SDavid Howells } 903d879cb83SAl Viro } 904d879cb83SAl Viro EXPORT_SYMBOL(iov_iter_advance); 905d879cb83SAl Viro 90627c0e374SAl Viro void iov_iter_revert(struct iov_iter *i, size_t unroll) 90727c0e374SAl Viro { 90827c0e374SAl Viro if (!unroll) 90927c0e374SAl Viro return; 9105b47d59aSAl Viro if (WARN_ON(unroll > MAX_RW_COUNT)) 9115b47d59aSAl Viro return; 91227c0e374SAl Viro i->count += unroll; 91300e23707SDavid Howells if (unlikely(iov_iter_is_pipe(i))) { 91427c0e374SAl Viro struct pipe_inode_info *pipe = i->pipe; 91592acdc4fSAl Viro unsigned int head = pipe->head; 91692acdc4fSAl Viro 91792acdc4fSAl Viro while (head > i->start_head) { 91892acdc4fSAl Viro struct pipe_buffer *b = pipe_buf(pipe, --head); 91992acdc4fSAl Viro if (unroll < b->len) { 92092acdc4fSAl Viro b->len -= unroll; 92110f525a8SAl Viro i->last_offset = last_offset(b); 92292acdc4fSAl Viro i->head = head; 92392acdc4fSAl Viro return; 92427c0e374SAl Viro } 92592acdc4fSAl Viro unroll -= b->len; 92692acdc4fSAl Viro pipe_buf_release(pipe, b); 92792acdc4fSAl Viro pipe->head--; 92827c0e374SAl Viro } 92910f525a8SAl Viro i->last_offset = 0; 93092acdc4fSAl Viro i->head = head; 93127c0e374SAl Viro return; 93227c0e374SAl Viro } 9339ea9ce04SDavid Howells if (unlikely(iov_iter_is_discard(i))) 9349ea9ce04SDavid Howells return; 93527c0e374SAl Viro if (unroll <= i->iov_offset) { 93627c0e374SAl Viro i->iov_offset -= unroll; 93727c0e374SAl Viro return; 93827c0e374SAl Viro } 93927c0e374SAl Viro unroll -= i->iov_offset; 940fcb14cb1SAl Viro if (iov_iter_is_xarray(i) || iter_is_ubuf(i)) { 9417ff50620SDavid Howells BUG(); /* We should never go beyond the start of the specified 9427ff50620SDavid Howells * range since we might then be straying into pages that 9437ff50620SDavid Howells * aren't pinned. 9447ff50620SDavid Howells */ 9457ff50620SDavid Howells } else if (iov_iter_is_bvec(i)) { 94627c0e374SAl Viro const struct bio_vec *bvec = i->bvec; 94727c0e374SAl Viro while (1) { 94827c0e374SAl Viro size_t n = (--bvec)->bv_len; 94927c0e374SAl Viro i->nr_segs++; 95027c0e374SAl Viro if (unroll <= n) { 95127c0e374SAl Viro i->bvec = bvec; 95227c0e374SAl Viro i->iov_offset = n - unroll; 95327c0e374SAl Viro return; 95427c0e374SAl Viro } 95527c0e374SAl Viro unroll -= n; 95627c0e374SAl Viro } 95727c0e374SAl Viro } else { /* same logics for iovec and kvec */ 95827c0e374SAl Viro const struct iovec *iov = i->iov; 95927c0e374SAl Viro while (1) { 96027c0e374SAl Viro size_t n = (--iov)->iov_len; 96127c0e374SAl Viro i->nr_segs++; 96227c0e374SAl Viro if (unroll <= n) { 96327c0e374SAl Viro i->iov = iov; 96427c0e374SAl Viro i->iov_offset = n - unroll; 96527c0e374SAl Viro return; 96627c0e374SAl Viro } 96727c0e374SAl Viro unroll -= n; 96827c0e374SAl Viro } 96927c0e374SAl Viro } 97027c0e374SAl Viro } 97127c0e374SAl Viro EXPORT_SYMBOL(iov_iter_revert); 97227c0e374SAl Viro 973d879cb83SAl Viro /* 974d879cb83SAl Viro * Return the count of just the current iov_iter segment. 975d879cb83SAl Viro */ 976d879cb83SAl Viro size_t iov_iter_single_seg_count(const struct iov_iter *i) 977d879cb83SAl Viro { 97828f38db7SAl Viro if (i->nr_segs > 1) { 97928f38db7SAl Viro if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) 98028f38db7SAl Viro return min(i->count, i->iov->iov_len - i->iov_offset); 9817ff50620SDavid Howells if (iov_iter_is_bvec(i)) 982d879cb83SAl Viro return min(i->count, i->bvec->bv_len - i->iov_offset); 98328f38db7SAl Viro } 98428f38db7SAl Viro return i->count; 985d879cb83SAl Viro } 986d879cb83SAl Viro EXPORT_SYMBOL(iov_iter_single_seg_count); 987d879cb83SAl Viro 988aa563d7bSDavid Howells void iov_iter_kvec(struct iov_iter *i, unsigned int direction, 989d879cb83SAl Viro const struct kvec *kvec, unsigned long nr_segs, 990d879cb83SAl Viro size_t count) 991d879cb83SAl Viro { 992aa563d7bSDavid Howells WARN_ON(direction & ~(READ | WRITE)); 9938cd54c1cSAl Viro *i = (struct iov_iter){ 9948cd54c1cSAl Viro .iter_type = ITER_KVEC, 9958cd54c1cSAl Viro .data_source = direction, 9968cd54c1cSAl Viro .kvec = kvec, 9978cd54c1cSAl Viro .nr_segs = nr_segs, 9988cd54c1cSAl Viro .iov_offset = 0, 9998cd54c1cSAl Viro .count = count 10008cd54c1cSAl Viro }; 1001d879cb83SAl Viro } 1002d879cb83SAl Viro EXPORT_SYMBOL(iov_iter_kvec); 1003d879cb83SAl Viro 1004aa563d7bSDavid Howells void iov_iter_bvec(struct iov_iter *i, unsigned int direction, 1005d879cb83SAl Viro const struct bio_vec *bvec, unsigned long nr_segs, 1006d879cb83SAl Viro size_t count) 1007d879cb83SAl Viro { 1008aa563d7bSDavid Howells WARN_ON(direction & ~(READ | WRITE)); 10098cd54c1cSAl Viro *i = (struct iov_iter){ 10108cd54c1cSAl Viro .iter_type = ITER_BVEC, 10118cd54c1cSAl Viro .data_source = direction, 10128cd54c1cSAl Viro .bvec = bvec, 10138cd54c1cSAl Viro .nr_segs = nr_segs, 10148cd54c1cSAl Viro .iov_offset = 0, 10158cd54c1cSAl Viro .count = count 10168cd54c1cSAl Viro }; 1017d879cb83SAl Viro } 1018d879cb83SAl Viro EXPORT_SYMBOL(iov_iter_bvec); 1019d879cb83SAl Viro 1020aa563d7bSDavid Howells void iov_iter_pipe(struct iov_iter *i, unsigned int direction, 1021241699cdSAl Viro struct pipe_inode_info *pipe, 1022241699cdSAl Viro size_t count) 1023241699cdSAl Viro { 1024aa563d7bSDavid Howells BUG_ON(direction != READ); 10258cefc107SDavid Howells WARN_ON(pipe_full(pipe->head, pipe->tail, pipe->ring_size)); 10268cd54c1cSAl Viro *i = (struct iov_iter){ 10278cd54c1cSAl Viro .iter_type = ITER_PIPE, 10288cd54c1cSAl Viro .data_source = false, 10298cd54c1cSAl Viro .pipe = pipe, 10308cd54c1cSAl Viro .head = pipe->head, 10318cd54c1cSAl Viro .start_head = pipe->head, 103210f525a8SAl Viro .last_offset = 0, 10338cd54c1cSAl Viro .count = count 10348cd54c1cSAl Viro }; 1035241699cdSAl Viro } 1036241699cdSAl Viro EXPORT_SYMBOL(iov_iter_pipe); 1037241699cdSAl Viro 10389ea9ce04SDavid Howells /** 10397ff50620SDavid Howells * iov_iter_xarray - Initialise an I/O iterator to use the pages in an xarray 10407ff50620SDavid Howells * @i: The iterator to initialise. 10417ff50620SDavid Howells * @direction: The direction of the transfer. 10427ff50620SDavid Howells * @xarray: The xarray to access. 10437ff50620SDavid Howells * @start: The start file position. 10447ff50620SDavid Howells * @count: The size of the I/O buffer in bytes. 10457ff50620SDavid Howells * 10467ff50620SDavid Howells * Set up an I/O iterator to either draw data out of the pages attached to an 10477ff50620SDavid Howells * inode or to inject data into those pages. The pages *must* be prevented 10487ff50620SDavid Howells * from evaporation, either by taking a ref on them or locking them by the 10497ff50620SDavid Howells * caller. 10507ff50620SDavid Howells */ 10517ff50620SDavid Howells void iov_iter_xarray(struct iov_iter *i, unsigned int direction, 10527ff50620SDavid Howells struct xarray *xarray, loff_t start, size_t count) 10537ff50620SDavid Howells { 10547ff50620SDavid Howells BUG_ON(direction & ~1); 10558cd54c1cSAl Viro *i = (struct iov_iter) { 10568cd54c1cSAl Viro .iter_type = ITER_XARRAY, 10578cd54c1cSAl Viro .data_source = direction, 10588cd54c1cSAl Viro .xarray = xarray, 10598cd54c1cSAl Viro .xarray_start = start, 10608cd54c1cSAl Viro .count = count, 10618cd54c1cSAl Viro .iov_offset = 0 10628cd54c1cSAl Viro }; 10637ff50620SDavid Howells } 10647ff50620SDavid Howells EXPORT_SYMBOL(iov_iter_xarray); 10657ff50620SDavid Howells 10667ff50620SDavid Howells /** 10679ea9ce04SDavid Howells * iov_iter_discard - Initialise an I/O iterator that discards data 10689ea9ce04SDavid Howells * @i: The iterator to initialise. 10699ea9ce04SDavid Howells * @direction: The direction of the transfer. 10709ea9ce04SDavid Howells * @count: The size of the I/O buffer in bytes. 10719ea9ce04SDavid Howells * 10729ea9ce04SDavid Howells * Set up an I/O iterator that just discards everything that's written to it. 10739ea9ce04SDavid Howells * It's only available as a READ iterator. 10749ea9ce04SDavid Howells */ 10759ea9ce04SDavid Howells void iov_iter_discard(struct iov_iter *i, unsigned int direction, size_t count) 10769ea9ce04SDavid Howells { 10779ea9ce04SDavid Howells BUG_ON(direction != READ); 10788cd54c1cSAl Viro *i = (struct iov_iter){ 10798cd54c1cSAl Viro .iter_type = ITER_DISCARD, 10808cd54c1cSAl Viro .data_source = false, 10818cd54c1cSAl Viro .count = count, 10828cd54c1cSAl Viro .iov_offset = 0 10838cd54c1cSAl Viro }; 10849ea9ce04SDavid Howells } 10859ea9ce04SDavid Howells EXPORT_SYMBOL(iov_iter_discard); 10869ea9ce04SDavid Howells 1087cfa320f7SKeith Busch static bool iov_iter_aligned_iovec(const struct iov_iter *i, unsigned addr_mask, 1088cfa320f7SKeith Busch unsigned len_mask) 1089cfa320f7SKeith Busch { 1090cfa320f7SKeith Busch size_t size = i->count; 1091cfa320f7SKeith Busch size_t skip = i->iov_offset; 1092cfa320f7SKeith Busch unsigned k; 1093cfa320f7SKeith Busch 1094cfa320f7SKeith Busch for (k = 0; k < i->nr_segs; k++, skip = 0) { 1095cfa320f7SKeith Busch size_t len = i->iov[k].iov_len - skip; 1096cfa320f7SKeith Busch 1097cfa320f7SKeith Busch if (len > size) 1098cfa320f7SKeith Busch len = size; 1099cfa320f7SKeith Busch if (len & len_mask) 1100cfa320f7SKeith Busch return false; 1101cfa320f7SKeith Busch if ((unsigned long)(i->iov[k].iov_base + skip) & addr_mask) 1102cfa320f7SKeith Busch return false; 1103cfa320f7SKeith Busch 1104cfa320f7SKeith Busch size -= len; 1105cfa320f7SKeith Busch if (!size) 1106cfa320f7SKeith Busch break; 1107cfa320f7SKeith Busch } 1108cfa320f7SKeith Busch return true; 1109cfa320f7SKeith Busch } 1110cfa320f7SKeith Busch 1111cfa320f7SKeith Busch static bool iov_iter_aligned_bvec(const struct iov_iter *i, unsigned addr_mask, 1112cfa320f7SKeith Busch unsigned len_mask) 1113cfa320f7SKeith Busch { 1114cfa320f7SKeith Busch size_t size = i->count; 1115cfa320f7SKeith Busch unsigned skip = i->iov_offset; 1116cfa320f7SKeith Busch unsigned k; 1117cfa320f7SKeith Busch 1118cfa320f7SKeith Busch for (k = 0; k < i->nr_segs; k++, skip = 0) { 1119cfa320f7SKeith Busch size_t len = i->bvec[k].bv_len - skip; 1120cfa320f7SKeith Busch 1121cfa320f7SKeith Busch if (len > size) 1122cfa320f7SKeith Busch len = size; 1123cfa320f7SKeith Busch if (len & len_mask) 1124cfa320f7SKeith Busch return false; 1125cfa320f7SKeith Busch if ((unsigned long)(i->bvec[k].bv_offset + skip) & addr_mask) 1126cfa320f7SKeith Busch return false; 1127cfa320f7SKeith Busch 1128cfa320f7SKeith Busch size -= len; 1129cfa320f7SKeith Busch if (!size) 1130cfa320f7SKeith Busch break; 1131cfa320f7SKeith Busch } 1132cfa320f7SKeith Busch return true; 1133cfa320f7SKeith Busch } 1134cfa320f7SKeith Busch 1135cfa320f7SKeith Busch /** 1136cfa320f7SKeith Busch * iov_iter_is_aligned() - Check if the addresses and lengths of each segments 1137cfa320f7SKeith Busch * are aligned to the parameters. 1138cfa320f7SKeith Busch * 1139cfa320f7SKeith Busch * @i: &struct iov_iter to restore 1140cfa320f7SKeith Busch * @addr_mask: bit mask to check against the iov element's addresses 1141cfa320f7SKeith Busch * @len_mask: bit mask to check against the iov element's lengths 1142cfa320f7SKeith Busch * 1143cfa320f7SKeith Busch * Return: false if any addresses or lengths intersect with the provided masks 1144cfa320f7SKeith Busch */ 1145cfa320f7SKeith Busch bool iov_iter_is_aligned(const struct iov_iter *i, unsigned addr_mask, 1146cfa320f7SKeith Busch unsigned len_mask) 1147cfa320f7SKeith Busch { 1148fcb14cb1SAl Viro if (likely(iter_is_ubuf(i))) { 1149fcb14cb1SAl Viro if (i->count & len_mask) 1150fcb14cb1SAl Viro return false; 1151fcb14cb1SAl Viro if ((unsigned long)(i->ubuf + i->iov_offset) & addr_mask) 1152fcb14cb1SAl Viro return false; 1153fcb14cb1SAl Viro return true; 1154fcb14cb1SAl Viro } 1155fcb14cb1SAl Viro 1156cfa320f7SKeith Busch if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) 1157cfa320f7SKeith Busch return iov_iter_aligned_iovec(i, addr_mask, len_mask); 1158cfa320f7SKeith Busch 1159cfa320f7SKeith Busch if (iov_iter_is_bvec(i)) 1160cfa320f7SKeith Busch return iov_iter_aligned_bvec(i, addr_mask, len_mask); 1161cfa320f7SKeith Busch 1162cfa320f7SKeith Busch if (iov_iter_is_pipe(i)) { 1163cfa320f7SKeith Busch size_t size = i->count; 1164cfa320f7SKeith Busch 1165cfa320f7SKeith Busch if (size & len_mask) 1166cfa320f7SKeith Busch return false; 116710f525a8SAl Viro if (size && i->last_offset > 0) { 116810f525a8SAl Viro if (i->last_offset & addr_mask) 1169cfa320f7SKeith Busch return false; 1170cfa320f7SKeith Busch } 1171cfa320f7SKeith Busch 1172cfa320f7SKeith Busch return true; 1173cfa320f7SKeith Busch } 1174cfa320f7SKeith Busch 1175cfa320f7SKeith Busch if (iov_iter_is_xarray(i)) { 1176cfa320f7SKeith Busch if (i->count & len_mask) 1177cfa320f7SKeith Busch return false; 1178cfa320f7SKeith Busch if ((i->xarray_start + i->iov_offset) & addr_mask) 1179cfa320f7SKeith Busch return false; 1180cfa320f7SKeith Busch } 1181cfa320f7SKeith Busch 1182cfa320f7SKeith Busch return true; 1183cfa320f7SKeith Busch } 1184cfa320f7SKeith Busch EXPORT_SYMBOL_GPL(iov_iter_is_aligned); 1185cfa320f7SKeith Busch 11869221d2e3SAl Viro static unsigned long iov_iter_alignment_iovec(const struct iov_iter *i) 1187d879cb83SAl Viro { 1188d879cb83SAl Viro unsigned long res = 0; 1189d879cb83SAl Viro size_t size = i->count; 11909221d2e3SAl Viro size_t skip = i->iov_offset; 11919221d2e3SAl Viro unsigned k; 1192d879cb83SAl Viro 11939221d2e3SAl Viro for (k = 0; k < i->nr_segs; k++, skip = 0) { 11949221d2e3SAl Viro size_t len = i->iov[k].iov_len - skip; 11959221d2e3SAl Viro if (len) { 11969221d2e3SAl Viro res |= (unsigned long)i->iov[k].iov_base + skip; 11979221d2e3SAl Viro if (len > size) 11989221d2e3SAl Viro len = size; 11999221d2e3SAl Viro res |= len; 12009221d2e3SAl Viro size -= len; 12019221d2e3SAl Viro if (!size) 12029221d2e3SAl Viro break; 12039221d2e3SAl Viro } 12049221d2e3SAl Viro } 12059221d2e3SAl Viro return res; 12069221d2e3SAl Viro } 12079221d2e3SAl Viro 12089221d2e3SAl Viro static unsigned long iov_iter_alignment_bvec(const struct iov_iter *i) 12099221d2e3SAl Viro { 12109221d2e3SAl Viro unsigned res = 0; 12119221d2e3SAl Viro size_t size = i->count; 12129221d2e3SAl Viro unsigned skip = i->iov_offset; 12139221d2e3SAl Viro unsigned k; 12149221d2e3SAl Viro 12159221d2e3SAl Viro for (k = 0; k < i->nr_segs; k++, skip = 0) { 12169221d2e3SAl Viro size_t len = i->bvec[k].bv_len - skip; 12179221d2e3SAl Viro res |= (unsigned long)i->bvec[k].bv_offset + skip; 12189221d2e3SAl Viro if (len > size) 12199221d2e3SAl Viro len = size; 12209221d2e3SAl Viro res |= len; 12219221d2e3SAl Viro size -= len; 12229221d2e3SAl Viro if (!size) 12239221d2e3SAl Viro break; 12249221d2e3SAl Viro } 12259221d2e3SAl Viro return res; 12269221d2e3SAl Viro } 12279221d2e3SAl Viro 12289221d2e3SAl Viro unsigned long iov_iter_alignment(const struct iov_iter *i) 12299221d2e3SAl Viro { 1230fcb14cb1SAl Viro if (likely(iter_is_ubuf(i))) { 1231fcb14cb1SAl Viro size_t size = i->count; 1232fcb14cb1SAl Viro if (size) 1233fcb14cb1SAl Viro return ((unsigned long)i->ubuf + i->iov_offset) | size; 1234fcb14cb1SAl Viro return 0; 1235fcb14cb1SAl Viro } 1236fcb14cb1SAl Viro 12379221d2e3SAl Viro /* iovec and kvec have identical layouts */ 12389221d2e3SAl Viro if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) 12399221d2e3SAl Viro return iov_iter_alignment_iovec(i); 12409221d2e3SAl Viro 12419221d2e3SAl Viro if (iov_iter_is_bvec(i)) 12429221d2e3SAl Viro return iov_iter_alignment_bvec(i); 12439221d2e3SAl Viro 12449221d2e3SAl Viro if (iov_iter_is_pipe(i)) { 12459221d2e3SAl Viro size_t size = i->count; 1246e0ff126eSJan Kara 124710f525a8SAl Viro if (size && i->last_offset > 0) 124810f525a8SAl Viro return size | i->last_offset; 1249241699cdSAl Viro return size; 1250241699cdSAl Viro } 12519221d2e3SAl Viro 12529221d2e3SAl Viro if (iov_iter_is_xarray(i)) 12533d14ec1fSDavid Howells return (i->xarray_start + i->iov_offset) | i->count; 12549221d2e3SAl Viro 12559221d2e3SAl Viro return 0; 1256d879cb83SAl Viro } 1257d879cb83SAl Viro EXPORT_SYMBOL(iov_iter_alignment); 1258d879cb83SAl Viro 1259357f435dSAl Viro unsigned long iov_iter_gap_alignment(const struct iov_iter *i) 1260357f435dSAl Viro { 1261357f435dSAl Viro unsigned long res = 0; 1262610c7a71SAl Viro unsigned long v = 0; 1263357f435dSAl Viro size_t size = i->count; 1264610c7a71SAl Viro unsigned k; 1265357f435dSAl Viro 1266fcb14cb1SAl Viro if (iter_is_ubuf(i)) 1267fcb14cb1SAl Viro return 0; 1268fcb14cb1SAl Viro 1269610c7a71SAl Viro if (WARN_ON(!iter_is_iovec(i))) 1270241699cdSAl Viro return ~0U; 1271241699cdSAl Viro 1272610c7a71SAl Viro for (k = 0; k < i->nr_segs; k++) { 1273610c7a71SAl Viro if (i->iov[k].iov_len) { 1274610c7a71SAl Viro unsigned long base = (unsigned long)i->iov[k].iov_base; 1275610c7a71SAl Viro if (v) // if not the first one 1276610c7a71SAl Viro res |= base | v; // this start | previous end 1277610c7a71SAl Viro v = base + i->iov[k].iov_len; 1278610c7a71SAl Viro if (size <= i->iov[k].iov_len) 1279610c7a71SAl Viro break; 1280610c7a71SAl Viro size -= i->iov[k].iov_len; 1281610c7a71SAl Viro } 1282610c7a71SAl Viro } 1283357f435dSAl Viro return res; 1284357f435dSAl Viro } 1285357f435dSAl Viro EXPORT_SYMBOL(iov_iter_gap_alignment); 1286357f435dSAl Viro 1287*3cf42da3SAl Viro static int want_pages_array(struct page ***res, size_t size, 1288*3cf42da3SAl Viro size_t start, unsigned int maxpages) 1289acbdeb83SAl Viro { 1290*3cf42da3SAl Viro unsigned int count = DIV_ROUND_UP(size + start, PAGE_SIZE); 1291*3cf42da3SAl Viro 1292*3cf42da3SAl Viro if (count > maxpages) 1293*3cf42da3SAl Viro count = maxpages; 1294*3cf42da3SAl Viro WARN_ON(!count); // caller should've prevented that 1295*3cf42da3SAl Viro if (!*res) { 1296*3cf42da3SAl Viro *res = kvmalloc_array(count, sizeof(struct page *), GFP_KERNEL); 1297*3cf42da3SAl Viro if (!*res) 1298*3cf42da3SAl Viro return 0; 1299*3cf42da3SAl Viro } 1300*3cf42da3SAl Viro return count; 1301acbdeb83SAl Viro } 1302acbdeb83SAl Viro 130385200084SAl Viro static ssize_t pipe_get_pages(struct iov_iter *i, 130485200084SAl Viro struct page ***pages, size_t maxsize, unsigned maxpages, 130585200084SAl Viro size_t *start) 1306241699cdSAl Viro { 1307241699cdSAl Viro struct pipe_inode_info *pipe = i->pipe; 1308*3cf42da3SAl Viro unsigned int npages, off, count; 130985200084SAl Viro struct page **p; 131085200084SAl Viro ssize_t left; 1311241699cdSAl Viro 131285200084SAl Viro if (!sanity(i)) 131385200084SAl Viro return -EFAULT; 131485200084SAl Viro 131585200084SAl Viro *start = off = pipe_npages(i, &npages); 1316*3cf42da3SAl Viro if (!npages) 1317*3cf42da3SAl Viro return -EFAULT; 1318*3cf42da3SAl Viro count = want_pages_array(pages, maxsize, off, min(npages, maxpages)); 1319*3cf42da3SAl Viro if (!count) 132085200084SAl Viro return -ENOMEM; 1321*3cf42da3SAl Viro p = *pages; 132285200084SAl Viro left = maxsize; 132385200084SAl Viro npages = 0; 1324e3b42964SAl Viro if (off) { 1325ca591967SAl Viro struct pipe_buffer *buf = pipe_buf(pipe, pipe->head - 1); 1326241699cdSAl Viro 132785200084SAl Viro get_page(*p++ = buf->page); 1328e3b42964SAl Viro left -= PAGE_SIZE - off; 1329e3b42964SAl Viro if (left <= 0) { 1330e3b42964SAl Viro buf->len += maxsize; 1331241699cdSAl Viro return maxsize; 1332241699cdSAl Viro } 1333e3b42964SAl Viro buf->len = PAGE_SIZE; 133485200084SAl Viro npages = 1; 1335e3b42964SAl Viro } 133685200084SAl Viro for ( ; npages < count; npages++) { 133785200084SAl Viro struct page *page; 133885200084SAl Viro unsigned int size = min_t(ssize_t, left, PAGE_SIZE); 133985200084SAl Viro 134085200084SAl Viro if (pipe_full(pipe->head, pipe->tail, pipe->max_usage)) 134185200084SAl Viro break; 134285200084SAl Viro page = push_anon(pipe, size); 1343e3b42964SAl Viro if (!page) 1344e3b42964SAl Viro break; 134585200084SAl Viro get_page(*p++ = page); 134685200084SAl Viro left -= size; 1347e3b42964SAl Viro } 134885200084SAl Viro if (!npages) 1349241699cdSAl Viro return -EFAULT; 135085200084SAl Viro return maxsize - left; 1351241699cdSAl Viro } 1352241699cdSAl Viro 13537ff50620SDavid Howells static ssize_t iter_xarray_populate_pages(struct page **pages, struct xarray *xa, 13547ff50620SDavid Howells pgoff_t index, unsigned int nr_pages) 13557ff50620SDavid Howells { 13567ff50620SDavid Howells XA_STATE(xas, xa, index); 13577ff50620SDavid Howells struct page *page; 13587ff50620SDavid Howells unsigned int ret = 0; 13597ff50620SDavid Howells 13607ff50620SDavid Howells rcu_read_lock(); 13617ff50620SDavid Howells for (page = xas_load(&xas); page; page = xas_next(&xas)) { 13627ff50620SDavid Howells if (xas_retry(&xas, page)) 13637ff50620SDavid Howells continue; 13647ff50620SDavid Howells 13657ff50620SDavid Howells /* Has the page moved or been split? */ 13667ff50620SDavid Howells if (unlikely(page != xas_reload(&xas))) { 13677ff50620SDavid Howells xas_reset(&xas); 13687ff50620SDavid Howells continue; 13697ff50620SDavid Howells } 13707ff50620SDavid Howells 13717ff50620SDavid Howells pages[ret] = find_subpage(page, xas.xa_index); 13727ff50620SDavid Howells get_page(pages[ret]); 13737ff50620SDavid Howells if (++ret == nr_pages) 13747ff50620SDavid Howells break; 13757ff50620SDavid Howells } 13767ff50620SDavid Howells rcu_read_unlock(); 13777ff50620SDavid Howells return ret; 13787ff50620SDavid Howells } 13797ff50620SDavid Howells 13807ff50620SDavid Howells static ssize_t iter_xarray_get_pages(struct iov_iter *i, 138168fe506fSAl Viro struct page ***pages, size_t maxsize, 13827ff50620SDavid Howells unsigned maxpages, size_t *_start_offset) 13837ff50620SDavid Howells { 1384*3cf42da3SAl Viro unsigned nr, offset, count; 1385*3cf42da3SAl Viro pgoff_t index; 13867ff50620SDavid Howells loff_t pos; 13877ff50620SDavid Howells 13887ff50620SDavid Howells pos = i->xarray_start + i->iov_offset; 13897ff50620SDavid Howells index = pos >> PAGE_SHIFT; 13907ff50620SDavid Howells offset = pos & ~PAGE_MASK; 13917ff50620SDavid Howells *_start_offset = offset; 13927ff50620SDavid Howells 1393*3cf42da3SAl Viro count = want_pages_array(pages, maxsize, offset, maxpages); 1394*3cf42da3SAl Viro if (!count) 139568fe506fSAl Viro return -ENOMEM; 139668fe506fSAl Viro nr = iter_xarray_populate_pages(*pages, i->xarray, index, count); 13977ff50620SDavid Howells if (nr == 0) 13987ff50620SDavid Howells return 0; 13997ff50620SDavid Howells 14001c27f1fcSLinus Torvalds return min_t(size_t, nr * PAGE_SIZE - offset, maxsize); 14017ff50620SDavid Howells } 14027ff50620SDavid Howells 1403fcb14cb1SAl Viro /* must be done on non-empty ITER_UBUF or ITER_IOVEC one */ 1404dd45ab9dSAl Viro static unsigned long first_iovec_segment(const struct iov_iter *i, size_t *size) 14053d671ca6SAl Viro { 14063d671ca6SAl Viro size_t skip; 14073d671ca6SAl Viro long k; 14083d671ca6SAl Viro 1409fcb14cb1SAl Viro if (iter_is_ubuf(i)) 1410fcb14cb1SAl Viro return (unsigned long)i->ubuf + i->iov_offset; 1411fcb14cb1SAl Viro 14123d671ca6SAl Viro for (k = 0, skip = i->iov_offset; k < i->nr_segs; k++, skip = 0) { 14133d671ca6SAl Viro size_t len = i->iov[k].iov_len - skip; 14143d671ca6SAl Viro 14153d671ca6SAl Viro if (unlikely(!len)) 14163d671ca6SAl Viro continue; 141759dbd7d0SAl Viro if (*size > len) 14183d671ca6SAl Viro *size = len; 1419dd45ab9dSAl Viro return (unsigned long)i->iov[k].iov_base + skip; 14203d671ca6SAl Viro } 14213d671ca6SAl Viro BUG(); // if it had been empty, we wouldn't get called 14223d671ca6SAl Viro } 14233d671ca6SAl Viro 14243d671ca6SAl Viro /* must be done on non-empty ITER_BVEC one */ 14253d671ca6SAl Viro static struct page *first_bvec_segment(const struct iov_iter *i, 142659dbd7d0SAl Viro size_t *size, size_t *start) 14273d671ca6SAl Viro { 14283d671ca6SAl Viro struct page *page; 14293d671ca6SAl Viro size_t skip = i->iov_offset, len; 14303d671ca6SAl Viro 14313d671ca6SAl Viro len = i->bvec->bv_len - skip; 143259dbd7d0SAl Viro if (*size > len) 143359dbd7d0SAl Viro *size = len; 14343d671ca6SAl Viro skip += i->bvec->bv_offset; 14353d671ca6SAl Viro page = i->bvec->bv_page + skip / PAGE_SIZE; 1436dda8e5d1SAl Viro *start = skip % PAGE_SIZE; 14373d671ca6SAl Viro return page; 14383d671ca6SAl Viro } 14393d671ca6SAl Viro 144091329559SAl Viro static ssize_t __iov_iter_get_pages_alloc(struct iov_iter *i, 1441d879cb83SAl Viro struct page ***pages, size_t maxsize, 1442451c0ba9SAl Viro unsigned int maxpages, size_t *start) 1443d879cb83SAl Viro { 1444*3cf42da3SAl Viro unsigned int n; 1445d879cb83SAl Viro 1446d879cb83SAl Viro if (maxsize > i->count) 1447d879cb83SAl Viro maxsize = i->count; 14483d671ca6SAl Viro if (!maxsize) 14493d671ca6SAl Viro return 0; 14507392ed17SAl Viro if (maxsize > MAX_RW_COUNT) 14517392ed17SAl Viro maxsize = MAX_RW_COUNT; 1452d879cb83SAl Viro 1453fcb14cb1SAl Viro if (likely(user_backed_iter(i))) { 14543337ab08SAndreas Gruenbacher unsigned int gup_flags = 0; 14553d671ca6SAl Viro unsigned long addr; 1456*3cf42da3SAl Viro int res; 14579ea9ce04SDavid Howells 14583337ab08SAndreas Gruenbacher if (iov_iter_rw(i) != WRITE) 14593337ab08SAndreas Gruenbacher gup_flags |= FOLL_WRITE; 14603337ab08SAndreas Gruenbacher if (i->nofault) 14613337ab08SAndreas Gruenbacher gup_flags |= FOLL_NOFAULT; 14623337ab08SAndreas Gruenbacher 1463dd45ab9dSAl Viro addr = first_iovec_segment(i, &maxsize); 1464dd45ab9dSAl Viro *start = addr % PAGE_SIZE; 1465dd45ab9dSAl Viro addr &= PAGE_MASK; 1466*3cf42da3SAl Viro n = want_pages_array(pages, maxsize, *start, maxpages); 1467*3cf42da3SAl Viro if (!n) 1468d879cb83SAl Viro return -ENOMEM; 1469451c0ba9SAl Viro res = get_user_pages_fast(addr, n, gup_flags, *pages); 147091329559SAl Viro if (unlikely(res <= 0)) 1471d879cb83SAl Viro return res; 147259dbd7d0SAl Viro return min_t(size_t, maxsize, res * PAGE_SIZE - *start); 14733d671ca6SAl Viro } 14743d671ca6SAl Viro if (iov_iter_is_bvec(i)) { 1475451c0ba9SAl Viro struct page **p; 14763d671ca6SAl Viro struct page *page; 14773d671ca6SAl Viro 147859dbd7d0SAl Viro page = first_bvec_segment(i, &maxsize, start); 1479*3cf42da3SAl Viro n = want_pages_array(pages, maxsize, *start, maxpages); 1480*3cf42da3SAl Viro if (!n) 1481d879cb83SAl Viro return -ENOMEM; 1482*3cf42da3SAl Viro p = *pages; 1483dda8e5d1SAl Viro for (int k = 0; k < n; k++) 14843d671ca6SAl Viro get_page(*p++ = page++); 148559dbd7d0SAl Viro return min_t(size_t, maxsize, n * PAGE_SIZE - *start); 14863d671ca6SAl Viro } 14873d671ca6SAl Viro if (iov_iter_is_pipe(i)) 1488451c0ba9SAl Viro return pipe_get_pages(i, pages, maxsize, maxpages, start); 14893d671ca6SAl Viro if (iov_iter_is_xarray(i)) 1490451c0ba9SAl Viro return iter_xarray_get_pages(i, pages, maxsize, maxpages, start); 1491d879cb83SAl Viro return -EFAULT; 1492d879cb83SAl Viro } 149391329559SAl Viro 1494451c0ba9SAl Viro ssize_t iov_iter_get_pages(struct iov_iter *i, 1495451c0ba9SAl Viro struct page **pages, size_t maxsize, unsigned maxpages, 1496451c0ba9SAl Viro size_t *start) 1497451c0ba9SAl Viro { 1498451c0ba9SAl Viro if (!maxpages) 1499451c0ba9SAl Viro return 0; 1500451c0ba9SAl Viro BUG_ON(!pages); 1501451c0ba9SAl Viro 1502451c0ba9SAl Viro return __iov_iter_get_pages_alloc(i, &pages, maxsize, maxpages, start); 1503451c0ba9SAl Viro } 1504451c0ba9SAl Viro EXPORT_SYMBOL(iov_iter_get_pages); 1505451c0ba9SAl Viro 150691329559SAl Viro ssize_t iov_iter_get_pages_alloc(struct iov_iter *i, 150791329559SAl Viro struct page ***pages, size_t maxsize, 150891329559SAl Viro size_t *start) 150991329559SAl Viro { 151091329559SAl Viro ssize_t len; 151191329559SAl Viro 151291329559SAl Viro *pages = NULL; 151391329559SAl Viro 1514451c0ba9SAl Viro len = __iov_iter_get_pages_alloc(i, pages, maxsize, ~0U, start); 151591329559SAl Viro if (len <= 0) { 151691329559SAl Viro kvfree(*pages); 151791329559SAl Viro *pages = NULL; 151891329559SAl Viro } 151991329559SAl Viro return len; 152091329559SAl Viro } 1521d879cb83SAl Viro EXPORT_SYMBOL(iov_iter_get_pages_alloc); 1522d879cb83SAl Viro 1523d879cb83SAl Viro size_t csum_and_copy_from_iter(void *addr, size_t bytes, __wsum *csum, 1524d879cb83SAl Viro struct iov_iter *i) 1525d879cb83SAl Viro { 1526d879cb83SAl Viro __wsum sum, next; 1527d879cb83SAl Viro sum = *csum; 15289ea9ce04SDavid Howells if (unlikely(iov_iter_is_pipe(i) || iov_iter_is_discard(i))) { 1529241699cdSAl Viro WARN_ON(1); 1530241699cdSAl Viro return 0; 1531241699cdSAl Viro } 15327baa5099SAl Viro iterate_and_advance(i, bytes, base, len, off, ({ 15337baa5099SAl Viro next = csum_and_copy_from_user(base, addr + off, len); 1534d879cb83SAl Viro sum = csum_block_add(sum, next, off); 15357baa5099SAl Viro next ? 0 : len; 1536d879cb83SAl Viro }), ({ 15377baa5099SAl Viro sum = csum_and_memcpy(addr + off, base, len, sum, off); 1538d879cb83SAl Viro }) 1539d879cb83SAl Viro ) 1540d879cb83SAl Viro *csum = sum; 1541d879cb83SAl Viro return bytes; 1542d879cb83SAl Viro } 1543d879cb83SAl Viro EXPORT_SYMBOL(csum_and_copy_from_iter); 1544d879cb83SAl Viro 154552cbd23aSWillem de Bruijn size_t csum_and_copy_to_iter(const void *addr, size_t bytes, void *_csstate, 1546d879cb83SAl Viro struct iov_iter *i) 1547d879cb83SAl Viro { 154852cbd23aSWillem de Bruijn struct csum_state *csstate = _csstate; 1549d879cb83SAl Viro __wsum sum, next; 155078e1f386SAl Viro 155178e1f386SAl Viro if (unlikely(iov_iter_is_discard(i))) { 1552241699cdSAl Viro WARN_ON(1); /* for now */ 1553241699cdSAl Viro return 0; 1554241699cdSAl Viro } 15556852df12SAl Viro 15566852df12SAl Viro sum = csum_shift(csstate->csum, csstate->off); 15576852df12SAl Viro if (unlikely(iov_iter_is_pipe(i))) 15586852df12SAl Viro bytes = csum_and_copy_to_pipe_iter(addr, bytes, i, &sum); 15596852df12SAl Viro else iterate_and_advance(i, bytes, base, len, off, ({ 15607baa5099SAl Viro next = csum_and_copy_to_user(addr + off, base, len); 1561d879cb83SAl Viro sum = csum_block_add(sum, next, off); 15627baa5099SAl Viro next ? 0 : len; 1563d879cb83SAl Viro }), ({ 15647baa5099SAl Viro sum = csum_and_memcpy(base, addr + off, len, sum, off); 1565d879cb83SAl Viro }) 1566d879cb83SAl Viro ) 1567594e450bSAl Viro csstate->csum = csum_shift(sum, csstate->off); 1568594e450bSAl Viro csstate->off += bytes; 1569d879cb83SAl Viro return bytes; 1570d879cb83SAl Viro } 1571d879cb83SAl Viro EXPORT_SYMBOL(csum_and_copy_to_iter); 1572d879cb83SAl Viro 1573d05f4435SSagi Grimberg size_t hash_and_copy_to_iter(const void *addr, size_t bytes, void *hashp, 1574d05f4435SSagi Grimberg struct iov_iter *i) 1575d05f4435SSagi Grimberg { 15767999096fSHerbert Xu #ifdef CONFIG_CRYPTO_HASH 1577d05f4435SSagi Grimberg struct ahash_request *hash = hashp; 1578d05f4435SSagi Grimberg struct scatterlist sg; 1579d05f4435SSagi Grimberg size_t copied; 1580d05f4435SSagi Grimberg 1581d05f4435SSagi Grimberg copied = copy_to_iter(addr, bytes, i); 1582d05f4435SSagi Grimberg sg_init_one(&sg, addr, copied); 1583d05f4435SSagi Grimberg ahash_request_set_crypt(hash, &sg, NULL, copied); 1584d05f4435SSagi Grimberg crypto_ahash_update(hash); 1585d05f4435SSagi Grimberg return copied; 158627fad74aSYueHaibing #else 158727fad74aSYueHaibing return 0; 158827fad74aSYueHaibing #endif 1589d05f4435SSagi Grimberg } 1590d05f4435SSagi Grimberg EXPORT_SYMBOL(hash_and_copy_to_iter); 1591d05f4435SSagi Grimberg 159266531c65SAl Viro static int iov_npages(const struct iov_iter *i, int maxpages) 1593d879cb83SAl Viro { 159466531c65SAl Viro size_t skip = i->iov_offset, size = i->count; 159566531c65SAl Viro const struct iovec *p; 1596d879cb83SAl Viro int npages = 0; 1597d879cb83SAl Viro 159866531c65SAl Viro for (p = i->iov; size; skip = 0, p++) { 159966531c65SAl Viro unsigned offs = offset_in_page(p->iov_base + skip); 160066531c65SAl Viro size_t len = min(p->iov_len - skip, size); 1601d879cb83SAl Viro 160266531c65SAl Viro if (len) { 160366531c65SAl Viro size -= len; 160466531c65SAl Viro npages += DIV_ROUND_UP(offs + len, PAGE_SIZE); 160566531c65SAl Viro if (unlikely(npages > maxpages)) 160666531c65SAl Viro return maxpages; 160766531c65SAl Viro } 160866531c65SAl Viro } 160966531c65SAl Viro return npages; 161066531c65SAl Viro } 161166531c65SAl Viro 161266531c65SAl Viro static int bvec_npages(const struct iov_iter *i, int maxpages) 161366531c65SAl Viro { 161466531c65SAl Viro size_t skip = i->iov_offset, size = i->count; 161566531c65SAl Viro const struct bio_vec *p; 161666531c65SAl Viro int npages = 0; 161766531c65SAl Viro 161866531c65SAl Viro for (p = i->bvec; size; skip = 0, p++) { 161966531c65SAl Viro unsigned offs = (p->bv_offset + skip) % PAGE_SIZE; 162066531c65SAl Viro size_t len = min(p->bv_len - skip, size); 162166531c65SAl Viro 162266531c65SAl Viro size -= len; 162366531c65SAl Viro npages += DIV_ROUND_UP(offs + len, PAGE_SIZE); 162466531c65SAl Viro if (unlikely(npages > maxpages)) 162566531c65SAl Viro return maxpages; 162666531c65SAl Viro } 162766531c65SAl Viro return npages; 162866531c65SAl Viro } 162966531c65SAl Viro 163066531c65SAl Viro int iov_iter_npages(const struct iov_iter *i, int maxpages) 163166531c65SAl Viro { 163266531c65SAl Viro if (unlikely(!i->count)) 163366531c65SAl Viro return 0; 1634fcb14cb1SAl Viro if (likely(iter_is_ubuf(i))) { 1635fcb14cb1SAl Viro unsigned offs = offset_in_page(i->ubuf + i->iov_offset); 1636fcb14cb1SAl Viro int npages = DIV_ROUND_UP(offs + i->count, PAGE_SIZE); 1637fcb14cb1SAl Viro return min(npages, maxpages); 1638fcb14cb1SAl Viro } 163966531c65SAl Viro /* iovec and kvec have identical layouts */ 164066531c65SAl Viro if (likely(iter_is_iovec(i) || iov_iter_is_kvec(i))) 164166531c65SAl Viro return iov_npages(i, maxpages); 164266531c65SAl Viro if (iov_iter_is_bvec(i)) 164366531c65SAl Viro return bvec_npages(i, maxpages); 164466531c65SAl Viro if (iov_iter_is_pipe(i)) { 164566531c65SAl Viro int npages; 1646241699cdSAl Viro 1647241699cdSAl Viro if (!sanity(i)) 1648241699cdSAl Viro return 0; 1649241699cdSAl Viro 165012d426abSAl Viro pipe_npages(i, &npages); 165166531c65SAl Viro return min(npages, maxpages); 165266531c65SAl Viro } 165366531c65SAl Viro if (iov_iter_is_xarray(i)) { 1654e4f8df86SAl Viro unsigned offset = (i->xarray_start + i->iov_offset) % PAGE_SIZE; 1655e4f8df86SAl Viro int npages = DIV_ROUND_UP(offset + i->count, PAGE_SIZE); 165666531c65SAl Viro return min(npages, maxpages); 165766531c65SAl Viro } 165866531c65SAl Viro return 0; 1659d879cb83SAl Viro } 1660d879cb83SAl Viro EXPORT_SYMBOL(iov_iter_npages); 1661d879cb83SAl Viro 1662d879cb83SAl Viro const void *dup_iter(struct iov_iter *new, struct iov_iter *old, gfp_t flags) 1663d879cb83SAl Viro { 1664d879cb83SAl Viro *new = *old; 166500e23707SDavid Howells if (unlikely(iov_iter_is_pipe(new))) { 1666241699cdSAl Viro WARN_ON(1); 1667241699cdSAl Viro return NULL; 1668241699cdSAl Viro } 166900e23707SDavid Howells if (iov_iter_is_bvec(new)) 1670d879cb83SAl Viro return new->bvec = kmemdup(new->bvec, 1671d879cb83SAl Viro new->nr_segs * sizeof(struct bio_vec), 1672d879cb83SAl Viro flags); 1673fcb14cb1SAl Viro else if (iov_iter_is_kvec(new) || iter_is_iovec(new)) 1674d879cb83SAl Viro /* iovec and kvec have identical layout */ 1675d879cb83SAl Viro return new->iov = kmemdup(new->iov, 1676d879cb83SAl Viro new->nr_segs * sizeof(struct iovec), 1677d879cb83SAl Viro flags); 1678fcb14cb1SAl Viro return NULL; 1679d879cb83SAl Viro } 1680d879cb83SAl Viro EXPORT_SYMBOL(dup_iter); 1681bc917be8SAl Viro 1682bfdc5970SChristoph Hellwig static int copy_compat_iovec_from_user(struct iovec *iov, 1683bfdc5970SChristoph Hellwig const struct iovec __user *uvec, unsigned long nr_segs) 1684bfdc5970SChristoph Hellwig { 1685bfdc5970SChristoph Hellwig const struct compat_iovec __user *uiov = 1686bfdc5970SChristoph Hellwig (const struct compat_iovec __user *)uvec; 1687bfdc5970SChristoph Hellwig int ret = -EFAULT, i; 1688bfdc5970SChristoph Hellwig 1689a959a978SChristoph Hellwig if (!user_access_begin(uiov, nr_segs * sizeof(*uiov))) 1690bfdc5970SChristoph Hellwig return -EFAULT; 1691bfdc5970SChristoph Hellwig 1692bfdc5970SChristoph Hellwig for (i = 0; i < nr_segs; i++) { 1693bfdc5970SChristoph Hellwig compat_uptr_t buf; 1694bfdc5970SChristoph Hellwig compat_ssize_t len; 1695bfdc5970SChristoph Hellwig 1696bfdc5970SChristoph Hellwig unsafe_get_user(len, &uiov[i].iov_len, uaccess_end); 1697bfdc5970SChristoph Hellwig unsafe_get_user(buf, &uiov[i].iov_base, uaccess_end); 1698bfdc5970SChristoph Hellwig 1699bfdc5970SChristoph Hellwig /* check for compat_size_t not fitting in compat_ssize_t .. */ 1700bfdc5970SChristoph Hellwig if (len < 0) { 1701bfdc5970SChristoph Hellwig ret = -EINVAL; 1702bfdc5970SChristoph Hellwig goto uaccess_end; 1703bfdc5970SChristoph Hellwig } 1704bfdc5970SChristoph Hellwig iov[i].iov_base = compat_ptr(buf); 1705bfdc5970SChristoph Hellwig iov[i].iov_len = len; 1706bfdc5970SChristoph Hellwig } 1707bfdc5970SChristoph Hellwig 1708bfdc5970SChristoph Hellwig ret = 0; 1709bfdc5970SChristoph Hellwig uaccess_end: 1710bfdc5970SChristoph Hellwig user_access_end(); 1711bfdc5970SChristoph Hellwig return ret; 1712bfdc5970SChristoph Hellwig } 1713bfdc5970SChristoph Hellwig 1714bfdc5970SChristoph Hellwig static int copy_iovec_from_user(struct iovec *iov, 1715bfdc5970SChristoph Hellwig const struct iovec __user *uvec, unsigned long nr_segs) 1716fb041b59SDavid Laight { 1717fb041b59SDavid Laight unsigned long seg; 1718bfdc5970SChristoph Hellwig 1719bfdc5970SChristoph Hellwig if (copy_from_user(iov, uvec, nr_segs * sizeof(*uvec))) 1720bfdc5970SChristoph Hellwig return -EFAULT; 1721bfdc5970SChristoph Hellwig for (seg = 0; seg < nr_segs; seg++) { 1722bfdc5970SChristoph Hellwig if ((ssize_t)iov[seg].iov_len < 0) 1723bfdc5970SChristoph Hellwig return -EINVAL; 1724bfdc5970SChristoph Hellwig } 1725bfdc5970SChristoph Hellwig 1726bfdc5970SChristoph Hellwig return 0; 1727bfdc5970SChristoph Hellwig } 1728bfdc5970SChristoph Hellwig 1729bfdc5970SChristoph Hellwig struct iovec *iovec_from_user(const struct iovec __user *uvec, 1730bfdc5970SChristoph Hellwig unsigned long nr_segs, unsigned long fast_segs, 1731bfdc5970SChristoph Hellwig struct iovec *fast_iov, bool compat) 1732bfdc5970SChristoph Hellwig { 1733bfdc5970SChristoph Hellwig struct iovec *iov = fast_iov; 1734bfdc5970SChristoph Hellwig int ret; 1735fb041b59SDavid Laight 1736fb041b59SDavid Laight /* 1737bfdc5970SChristoph Hellwig * SuS says "The readv() function *may* fail if the iovcnt argument was 1738bfdc5970SChristoph Hellwig * less than or equal to 0, or greater than {IOV_MAX}. Linux has 1739fb041b59SDavid Laight * traditionally returned zero for zero segments, so... 1740fb041b59SDavid Laight */ 1741bfdc5970SChristoph Hellwig if (nr_segs == 0) 1742bfdc5970SChristoph Hellwig return iov; 1743bfdc5970SChristoph Hellwig if (nr_segs > UIO_MAXIOV) 1744bfdc5970SChristoph Hellwig return ERR_PTR(-EINVAL); 1745fb041b59SDavid Laight if (nr_segs > fast_segs) { 1746fb041b59SDavid Laight iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL); 1747bfdc5970SChristoph Hellwig if (!iov) 1748bfdc5970SChristoph Hellwig return ERR_PTR(-ENOMEM); 1749fb041b59SDavid Laight } 1750bfdc5970SChristoph Hellwig 1751bfdc5970SChristoph Hellwig if (compat) 1752bfdc5970SChristoph Hellwig ret = copy_compat_iovec_from_user(iov, uvec, nr_segs); 1753bfdc5970SChristoph Hellwig else 1754bfdc5970SChristoph Hellwig ret = copy_iovec_from_user(iov, uvec, nr_segs); 1755bfdc5970SChristoph Hellwig if (ret) { 1756bfdc5970SChristoph Hellwig if (iov != fast_iov) 1757bfdc5970SChristoph Hellwig kfree(iov); 1758bfdc5970SChristoph Hellwig return ERR_PTR(ret); 1759fb041b59SDavid Laight } 1760bfdc5970SChristoph Hellwig 1761bfdc5970SChristoph Hellwig return iov; 1762bfdc5970SChristoph Hellwig } 1763bfdc5970SChristoph Hellwig 1764bfdc5970SChristoph Hellwig ssize_t __import_iovec(int type, const struct iovec __user *uvec, 1765bfdc5970SChristoph Hellwig unsigned nr_segs, unsigned fast_segs, struct iovec **iovp, 1766bfdc5970SChristoph Hellwig struct iov_iter *i, bool compat) 1767bfdc5970SChristoph Hellwig { 1768bfdc5970SChristoph Hellwig ssize_t total_len = 0; 1769bfdc5970SChristoph Hellwig unsigned long seg; 1770bfdc5970SChristoph Hellwig struct iovec *iov; 1771bfdc5970SChristoph Hellwig 1772bfdc5970SChristoph Hellwig iov = iovec_from_user(uvec, nr_segs, fast_segs, *iovp, compat); 1773bfdc5970SChristoph Hellwig if (IS_ERR(iov)) { 1774bfdc5970SChristoph Hellwig *iovp = NULL; 1775bfdc5970SChristoph Hellwig return PTR_ERR(iov); 1776fb041b59SDavid Laight } 1777fb041b59SDavid Laight 1778fb041b59SDavid Laight /* 1779bfdc5970SChristoph Hellwig * According to the Single Unix Specification we should return EINVAL if 1780bfdc5970SChristoph Hellwig * an element length is < 0 when cast to ssize_t or if the total length 1781bfdc5970SChristoph Hellwig * would overflow the ssize_t return value of the system call. 1782fb041b59SDavid Laight * 1783fb041b59SDavid Laight * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the 1784fb041b59SDavid Laight * overflow case. 1785fb041b59SDavid Laight */ 1786fb041b59SDavid Laight for (seg = 0; seg < nr_segs; seg++) { 1787fb041b59SDavid Laight ssize_t len = (ssize_t)iov[seg].iov_len; 1788fb041b59SDavid Laight 1789bfdc5970SChristoph Hellwig if (!access_ok(iov[seg].iov_base, len)) { 1790bfdc5970SChristoph Hellwig if (iov != *iovp) 1791bfdc5970SChristoph Hellwig kfree(iov); 1792bfdc5970SChristoph Hellwig *iovp = NULL; 1793bfdc5970SChristoph Hellwig return -EFAULT; 1794fb041b59SDavid Laight } 1795bfdc5970SChristoph Hellwig 1796bfdc5970SChristoph Hellwig if (len > MAX_RW_COUNT - total_len) { 1797bfdc5970SChristoph Hellwig len = MAX_RW_COUNT - total_len; 1798fb041b59SDavid Laight iov[seg].iov_len = len; 1799fb041b59SDavid Laight } 1800bfdc5970SChristoph Hellwig total_len += len; 1801fb041b59SDavid Laight } 1802bfdc5970SChristoph Hellwig 1803bfdc5970SChristoph Hellwig iov_iter_init(i, type, iov, nr_segs, total_len); 1804bfdc5970SChristoph Hellwig if (iov == *iovp) 1805bfdc5970SChristoph Hellwig *iovp = NULL; 1806bfdc5970SChristoph Hellwig else 1807bfdc5970SChristoph Hellwig *iovp = iov; 1808bfdc5970SChristoph Hellwig return total_len; 1809fb041b59SDavid Laight } 1810fb041b59SDavid Laight 1811ffecee4fSVegard Nossum /** 1812ffecee4fSVegard Nossum * import_iovec() - Copy an array of &struct iovec from userspace 1813ffecee4fSVegard Nossum * into the kernel, check that it is valid, and initialize a new 1814ffecee4fSVegard Nossum * &struct iov_iter iterator to access it. 1815ffecee4fSVegard Nossum * 1816ffecee4fSVegard Nossum * @type: One of %READ or %WRITE. 1817bfdc5970SChristoph Hellwig * @uvec: Pointer to the userspace array. 1818ffecee4fSVegard Nossum * @nr_segs: Number of elements in userspace array. 1819ffecee4fSVegard Nossum * @fast_segs: Number of elements in @iov. 1820bfdc5970SChristoph Hellwig * @iovp: (input and output parameter) Pointer to pointer to (usually small 1821ffecee4fSVegard Nossum * on-stack) kernel array. 1822ffecee4fSVegard Nossum * @i: Pointer to iterator that will be initialized on success. 1823ffecee4fSVegard Nossum * 1824ffecee4fSVegard Nossum * If the array pointed to by *@iov is large enough to hold all @nr_segs, 1825ffecee4fSVegard Nossum * then this function places %NULL in *@iov on return. Otherwise, a new 1826ffecee4fSVegard Nossum * array will be allocated and the result placed in *@iov. This means that 1827ffecee4fSVegard Nossum * the caller may call kfree() on *@iov regardless of whether the small 1828ffecee4fSVegard Nossum * on-stack array was used or not (and regardless of whether this function 1829ffecee4fSVegard Nossum * returns an error or not). 1830ffecee4fSVegard Nossum * 183187e5e6daSJens Axboe * Return: Negative error code on error, bytes imported on success 1832ffecee4fSVegard Nossum */ 1833bfdc5970SChristoph Hellwig ssize_t import_iovec(int type, const struct iovec __user *uvec, 1834bc917be8SAl Viro unsigned nr_segs, unsigned fast_segs, 1835bfdc5970SChristoph Hellwig struct iovec **iovp, struct iov_iter *i) 1836bc917be8SAl Viro { 183789cd35c5SChristoph Hellwig return __import_iovec(type, uvec, nr_segs, fast_segs, iovp, i, 183889cd35c5SChristoph Hellwig in_compat_syscall()); 1839bc917be8SAl Viro } 1840bc917be8SAl Viro EXPORT_SYMBOL(import_iovec); 1841bc917be8SAl Viro 1842bc917be8SAl Viro int import_single_range(int rw, void __user *buf, size_t len, 1843bc917be8SAl Viro struct iovec *iov, struct iov_iter *i) 1844bc917be8SAl Viro { 1845bc917be8SAl Viro if (len > MAX_RW_COUNT) 1846bc917be8SAl Viro len = MAX_RW_COUNT; 184796d4f267SLinus Torvalds if (unlikely(!access_ok(buf, len))) 1848bc917be8SAl Viro return -EFAULT; 1849bc917be8SAl Viro 1850bc917be8SAl Viro iov->iov_base = buf; 1851bc917be8SAl Viro iov->iov_len = len; 1852bc917be8SAl Viro iov_iter_init(i, rw, iov, 1, len); 1853bc917be8SAl Viro return 0; 1854bc917be8SAl Viro } 1855e1267585SAl Viro EXPORT_SYMBOL(import_single_range); 18568fb0f47aSJens Axboe 18578fb0f47aSJens Axboe /** 18588fb0f47aSJens Axboe * iov_iter_restore() - Restore a &struct iov_iter to the same state as when 18598fb0f47aSJens Axboe * iov_iter_save_state() was called. 18608fb0f47aSJens Axboe * 18618fb0f47aSJens Axboe * @i: &struct iov_iter to restore 18628fb0f47aSJens Axboe * @state: state to restore from 18638fb0f47aSJens Axboe * 18648fb0f47aSJens Axboe * Used after iov_iter_save_state() to bring restore @i, if operations may 18658fb0f47aSJens Axboe * have advanced it. 18668fb0f47aSJens Axboe * 18678fb0f47aSJens Axboe * Note: only works on ITER_IOVEC, ITER_BVEC, and ITER_KVEC 18688fb0f47aSJens Axboe */ 18698fb0f47aSJens Axboe void iov_iter_restore(struct iov_iter *i, struct iov_iter_state *state) 18708fb0f47aSJens Axboe { 18718fb0f47aSJens Axboe if (WARN_ON_ONCE(!iov_iter_is_bvec(i) && !iter_is_iovec(i)) && 1872fcb14cb1SAl Viro !iov_iter_is_kvec(i) && !iter_is_ubuf(i)) 18738fb0f47aSJens Axboe return; 18748fb0f47aSJens Axboe i->iov_offset = state->iov_offset; 18758fb0f47aSJens Axboe i->count = state->count; 1876fcb14cb1SAl Viro if (iter_is_ubuf(i)) 1877fcb14cb1SAl Viro return; 18788fb0f47aSJens Axboe /* 18798fb0f47aSJens Axboe * For the *vec iters, nr_segs + iov is constant - if we increment 18808fb0f47aSJens Axboe * the vec, then we also decrement the nr_segs count. Hence we don't 18818fb0f47aSJens Axboe * need to track both of these, just one is enough and we can deduct 18828fb0f47aSJens Axboe * the other from that. ITER_KVEC and ITER_IOVEC are the same struct 18838fb0f47aSJens Axboe * size, so we can just increment the iov pointer as they are unionzed. 18848fb0f47aSJens Axboe * ITER_BVEC _may_ be the same size on some archs, but on others it is 18858fb0f47aSJens Axboe * not. Be safe and handle it separately. 18868fb0f47aSJens Axboe */ 18878fb0f47aSJens Axboe BUILD_BUG_ON(sizeof(struct iovec) != sizeof(struct kvec)); 18888fb0f47aSJens Axboe if (iov_iter_is_bvec(i)) 18898fb0f47aSJens Axboe i->bvec -= state->nr_segs - i->nr_segs; 18908fb0f47aSJens Axboe else 18918fb0f47aSJens Axboe i->iov -= state->nr_segs - i->nr_segs; 18928fb0f47aSJens Axboe i->nr_segs = state->nr_segs; 18938fb0f47aSJens Axboe } 1894