Al Viro | 4f18cd3 | 2014-02-05 19:11:33 -0500 | [diff] [blame] | 1 | #include <linux/export.h> |
| 2 | #include <linux/uio.h> |
| 3 | #include <linux/pagemap.h> |
| 4 | |
| 5 | size_t copy_page_to_iter(struct page *page, size_t offset, size_t bytes, |
| 6 | struct iov_iter *i) |
| 7 | { |
| 8 | size_t skip, copy, left, wanted; |
| 9 | const struct iovec *iov; |
| 10 | char __user *buf; |
| 11 | void *kaddr, *from; |
| 12 | |
| 13 | if (unlikely(bytes > i->count)) |
| 14 | bytes = i->count; |
| 15 | |
| 16 | if (unlikely(!bytes)) |
| 17 | return 0; |
| 18 | |
| 19 | wanted = bytes; |
| 20 | iov = i->iov; |
| 21 | skip = i->iov_offset; |
| 22 | buf = iov->iov_base + skip; |
| 23 | copy = min(bytes, iov->iov_len - skip); |
| 24 | |
| 25 | if (!fault_in_pages_writeable(buf, copy)) { |
| 26 | kaddr = kmap_atomic(page); |
| 27 | from = kaddr + offset; |
| 28 | |
| 29 | /* first chunk, usually the only one */ |
| 30 | left = __copy_to_user_inatomic(buf, from, copy); |
| 31 | copy -= left; |
| 32 | skip += copy; |
| 33 | from += copy; |
| 34 | bytes -= copy; |
| 35 | |
| 36 | while (unlikely(!left && bytes)) { |
| 37 | iov++; |
| 38 | buf = iov->iov_base; |
| 39 | copy = min(bytes, iov->iov_len); |
| 40 | left = __copy_to_user_inatomic(buf, from, copy); |
| 41 | copy -= left; |
| 42 | skip = copy; |
| 43 | from += copy; |
| 44 | bytes -= copy; |
| 45 | } |
| 46 | if (likely(!bytes)) { |
| 47 | kunmap_atomic(kaddr); |
| 48 | goto done; |
| 49 | } |
| 50 | offset = from - kaddr; |
| 51 | buf += copy; |
| 52 | kunmap_atomic(kaddr); |
| 53 | copy = min(bytes, iov->iov_len - skip); |
| 54 | } |
| 55 | /* Too bad - revert to non-atomic kmap */ |
| 56 | kaddr = kmap(page); |
| 57 | from = kaddr + offset; |
| 58 | left = __copy_to_user(buf, from, copy); |
| 59 | copy -= left; |
| 60 | skip += copy; |
| 61 | from += copy; |
| 62 | bytes -= copy; |
| 63 | while (unlikely(!left && bytes)) { |
| 64 | iov++; |
| 65 | buf = iov->iov_base; |
| 66 | copy = min(bytes, iov->iov_len); |
| 67 | left = __copy_to_user(buf, from, copy); |
| 68 | copy -= left; |
| 69 | skip = copy; |
| 70 | from += copy; |
| 71 | bytes -= copy; |
| 72 | } |
| 73 | kunmap(page); |
| 74 | done: |
| 75 | i->count -= wanted - bytes; |
| 76 | i->nr_segs -= iov - i->iov; |
| 77 | i->iov = iov; |
| 78 | i->iov_offset = skip; |
| 79 | return wanted - bytes; |
| 80 | } |
| 81 | EXPORT_SYMBOL(copy_page_to_iter); |
| 82 | |
| 83 | static size_t __iovec_copy_from_user_inatomic(char *vaddr, |
| 84 | const struct iovec *iov, size_t base, size_t bytes) |
| 85 | { |
| 86 | size_t copied = 0, left = 0; |
| 87 | |
| 88 | while (bytes) { |
| 89 | char __user *buf = iov->iov_base + base; |
| 90 | int copy = min(bytes, iov->iov_len - base); |
| 91 | |
| 92 | base = 0; |
| 93 | left = __copy_from_user_inatomic(vaddr, buf, copy); |
| 94 | copied += copy; |
| 95 | bytes -= copy; |
| 96 | vaddr += copy; |
| 97 | iov++; |
| 98 | |
| 99 | if (unlikely(left)) |
| 100 | break; |
| 101 | } |
| 102 | return copied - left; |
| 103 | } |
| 104 | |
| 105 | /* |
| 106 | * Copy as much as we can into the page and return the number of bytes which |
| 107 | * were successfully copied. If a fault is encountered then return the number of |
| 108 | * bytes which were copied. |
| 109 | */ |
| 110 | size_t iov_iter_copy_from_user_atomic(struct page *page, |
| 111 | struct iov_iter *i, unsigned long offset, size_t bytes) |
| 112 | { |
| 113 | char *kaddr; |
| 114 | size_t copied; |
| 115 | |
| 116 | kaddr = kmap_atomic(page); |
| 117 | if (likely(i->nr_segs == 1)) { |
| 118 | int left; |
| 119 | char __user *buf = i->iov->iov_base + i->iov_offset; |
| 120 | left = __copy_from_user_inatomic(kaddr + offset, buf, bytes); |
| 121 | copied = bytes - left; |
| 122 | } else { |
| 123 | copied = __iovec_copy_from_user_inatomic(kaddr + offset, |
| 124 | i->iov, i->iov_offset, bytes); |
| 125 | } |
| 126 | kunmap_atomic(kaddr); |
| 127 | |
| 128 | return copied; |
| 129 | } |
| 130 | EXPORT_SYMBOL(iov_iter_copy_from_user_atomic); |
| 131 | |
| 132 | /* |
| 133 | * This has the same sideeffects and return value as |
| 134 | * iov_iter_copy_from_user_atomic(). |
| 135 | * The difference is that it attempts to resolve faults. |
| 136 | * Page must not be locked. |
| 137 | */ |
| 138 | size_t iov_iter_copy_from_user(struct page *page, |
| 139 | struct iov_iter *i, unsigned long offset, size_t bytes) |
| 140 | { |
| 141 | char *kaddr; |
| 142 | size_t copied; |
| 143 | |
| 144 | kaddr = kmap(page); |
| 145 | if (likely(i->nr_segs == 1)) { |
| 146 | int left; |
| 147 | char __user *buf = i->iov->iov_base + i->iov_offset; |
| 148 | left = __copy_from_user(kaddr + offset, buf, bytes); |
| 149 | copied = bytes - left; |
| 150 | } else { |
| 151 | copied = __iovec_copy_from_user_inatomic(kaddr + offset, |
| 152 | i->iov, i->iov_offset, bytes); |
| 153 | } |
| 154 | kunmap(page); |
| 155 | return copied; |
| 156 | } |
| 157 | EXPORT_SYMBOL(iov_iter_copy_from_user); |
| 158 | |
| 159 | void iov_iter_advance(struct iov_iter *i, size_t bytes) |
| 160 | { |
| 161 | BUG_ON(i->count < bytes); |
| 162 | |
| 163 | if (likely(i->nr_segs == 1)) { |
| 164 | i->iov_offset += bytes; |
| 165 | i->count -= bytes; |
| 166 | } else { |
| 167 | const struct iovec *iov = i->iov; |
| 168 | size_t base = i->iov_offset; |
| 169 | unsigned long nr_segs = i->nr_segs; |
| 170 | |
| 171 | /* |
| 172 | * The !iov->iov_len check ensures we skip over unlikely |
| 173 | * zero-length segments (without overruning the iovec). |
| 174 | */ |
| 175 | while (bytes || unlikely(i->count && !iov->iov_len)) { |
| 176 | int copy; |
| 177 | |
| 178 | copy = min(bytes, iov->iov_len - base); |
| 179 | BUG_ON(!i->count || i->count < copy); |
| 180 | i->count -= copy; |
| 181 | bytes -= copy; |
| 182 | base += copy; |
| 183 | if (iov->iov_len == base) { |
| 184 | iov++; |
| 185 | nr_segs--; |
| 186 | base = 0; |
| 187 | } |
| 188 | } |
| 189 | i->iov = iov; |
| 190 | i->iov_offset = base; |
| 191 | i->nr_segs = nr_segs; |
| 192 | } |
| 193 | } |
| 194 | EXPORT_SYMBOL(iov_iter_advance); |
| 195 | |
| 196 | /* |
| 197 | * Fault in the first iovec of the given iov_iter, to a maximum length |
| 198 | * of bytes. Returns 0 on success, or non-zero if the memory could not be |
| 199 | * accessed (ie. because it is an invalid address). |
| 200 | * |
| 201 | * writev-intensive code may want this to prefault several iovecs -- that |
| 202 | * would be possible (callers must not rely on the fact that _only_ the |
| 203 | * first iovec will be faulted with the current implementation). |
| 204 | */ |
| 205 | int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes) |
| 206 | { |
| 207 | char __user *buf = i->iov->iov_base + i->iov_offset; |
| 208 | bytes = min(bytes, i->iov->iov_len - i->iov_offset); |
| 209 | return fault_in_pages_readable(buf, bytes); |
| 210 | } |
| 211 | EXPORT_SYMBOL(iov_iter_fault_in_readable); |
| 212 | |
| 213 | /* |
| 214 | * Return the count of just the current iov_iter segment. |
| 215 | */ |
| 216 | size_t iov_iter_single_seg_count(const struct iov_iter *i) |
| 217 | { |
| 218 | const struct iovec *iov = i->iov; |
| 219 | if (i->nr_segs == 1) |
| 220 | return i->count; |
| 221 | else |
| 222 | return min(i->count, iov->iov_len - i->iov_offset); |
| 223 | } |
| 224 | EXPORT_SYMBOL(iov_iter_single_seg_count); |