| 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | /* |
| 3 | * mm/fadvise.c |
| 4 | * |
| 5 | * Copyright (C) 2002, Linus Torvalds |
| 6 | * |
| 7 | * 11Jan2003 Andrew Morton |
| 8 | * Initial version. |
| 9 | */ |
| 10 | |
| 11 | #include <linux/kernel.h> |
| 12 | #include <linux/file.h> |
| 13 | #include <linux/fs.h> |
| 14 | #include <linux/mm.h> |
| 15 | #include <linux/pagemap.h> |
| 16 | #include <linux/backing-dev.h> |
| 17 | #include <linux/pagevec.h> |
| 18 | #include <linux/fadvise.h> |
| 19 | #include <linux/writeback.h> |
| 20 | #include <linux/syscalls.h> |
| 21 | #include <linux/swap.h> |
| 22 | |
| 23 | #include <asm/unistd.h> |
| 24 | |
| 25 | /* |
| 26 | * POSIX_FADV_WILLNEED could set PG_Referenced, and POSIX_FADV_NOREUSE could |
| 27 | * deactivate the pages and clear PG_Referenced. |
| 28 | */ |
| 29 | SYSCALL_DEFINE4(fadvise64_64, int, fd, loff_t, offset, loff_t, len, int, advice) |
| 30 | { |
| 31 | struct fd f = fdget(fd); |
| 32 | struct inode *inode; |
| 33 | struct address_space *mapping; |
| 34 | struct backing_dev_info *bdi; |
| 35 | loff_t endbyte; /* inclusive */ |
| 36 | pgoff_t start_index; |
| 37 | pgoff_t end_index; |
| 38 | unsigned long nrpages; |
| 39 | int ret = 0; |
| 40 | |
| 41 | if (!f.file) |
| 42 | return -EBADF; |
| 43 | |
| 44 | inode = file_inode(f.file); |
| 45 | if (S_ISFIFO(inode->i_mode)) { |
| 46 | ret = -ESPIPE; |
| 47 | goto out; |
| 48 | } |
| 49 | |
| 50 | mapping = f.file->f_mapping; |
| 51 | if (!mapping || len < 0) { |
| 52 | ret = -EINVAL; |
| 53 | goto out; |
| 54 | } |
| 55 | |
| 56 | bdi = inode_to_bdi(mapping->host); |
| 57 | |
| 58 | if (IS_DAX(inode) || (bdi == &noop_backing_dev_info)) { |
| 59 | switch (advice) { |
| 60 | case POSIX_FADV_NORMAL: |
| 61 | case POSIX_FADV_RANDOM: |
| 62 | case POSIX_FADV_SEQUENTIAL: |
| 63 | case POSIX_FADV_WILLNEED: |
| 64 | case POSIX_FADV_NOREUSE: |
| 65 | case POSIX_FADV_DONTNEED: |
| 66 | /* no bad return value, but ignore advice */ |
| 67 | break; |
| 68 | default: |
| 69 | ret = -EINVAL; |
| 70 | } |
| 71 | goto out; |
| 72 | } |
| 73 | |
| 74 | /* |
| 75 | * Careful about overflows. Len == 0 means "as much as possible". Use |
| 76 | * unsigned math because signed overflows are undefined and UBSan |
| 77 | * complains. |
| 78 | */ |
| 79 | endbyte = (u64)offset + (u64)len; |
| 80 | if (!len || endbyte < len) |
| 81 | endbyte = -1; |
| 82 | else |
| 83 | endbyte--; /* inclusive */ |
| 84 | |
| 85 | switch (advice) { |
| 86 | case POSIX_FADV_NORMAL: |
| 87 | f.file->f_ra.ra_pages = bdi->ra_pages; |
| 88 | spin_lock(&f.file->f_lock); |
| 89 | f.file->f_mode &= ~FMODE_RANDOM; |
| 90 | spin_unlock(&f.file->f_lock); |
| 91 | break; |
| 92 | case POSIX_FADV_RANDOM: |
| 93 | spin_lock(&f.file->f_lock); |
| 94 | f.file->f_mode |= FMODE_RANDOM; |
| 95 | spin_unlock(&f.file->f_lock); |
| 96 | break; |
| 97 | case POSIX_FADV_SEQUENTIAL: |
| 98 | f.file->f_ra.ra_pages = bdi->ra_pages * 2; |
| 99 | spin_lock(&f.file->f_lock); |
| 100 | f.file->f_mode &= ~FMODE_RANDOM; |
| 101 | spin_unlock(&f.file->f_lock); |
| 102 | break; |
| 103 | case POSIX_FADV_WILLNEED: |
| 104 | /* First and last PARTIAL page! */ |
| 105 | start_index = offset >> PAGE_SHIFT; |
| 106 | end_index = endbyte >> PAGE_SHIFT; |
| 107 | |
| 108 | /* Careful about overflow on the "+1" */ |
| 109 | nrpages = end_index - start_index + 1; |
| 110 | if (!nrpages) |
| 111 | nrpages = ~0UL; |
| 112 | |
| 113 | /* |
| 114 | * Ignore return value because fadvise() shall return |
| 115 | * success even if filesystem can't retrieve a hint, |
| 116 | */ |
| 117 | force_page_cache_readahead(mapping, f.file, start_index, |
| 118 | nrpages); |
| 119 | break; |
| 120 | case POSIX_FADV_NOREUSE: |
| 121 | break; |
| 122 | case POSIX_FADV_DONTNEED: |
| 123 | if (!inode_write_congested(mapping->host)) |
| 124 | __filemap_fdatawrite_range(mapping, offset, endbyte, |
| 125 | WB_SYNC_NONE); |
| 126 | |
| 127 | /* |
| 128 | * First and last FULL page! Partial pages are deliberately |
| 129 | * preserved on the expectation that it is better to preserve |
| 130 | * needed memory than to discard unneeded memory. |
| 131 | */ |
| 132 | start_index = (offset+(PAGE_SIZE-1)) >> PAGE_SHIFT; |
| 133 | end_index = (endbyte >> PAGE_SHIFT); |
| 134 | /* |
| 135 | * The page at end_index will be inclusively discarded according |
| 136 | * by invalidate_mapping_pages(), so subtracting 1 from |
| 137 | * end_index means we will skip the last page. But if endbyte |
| 138 | * is page aligned or is at the end of file, we should not skip |
| 139 | * that page - discarding the last page is safe enough. |
| 140 | */ |
| 141 | if ((endbyte & ~PAGE_MASK) != ~PAGE_MASK && |
| 142 | endbyte != inode->i_size - 1) { |
| 143 | /* First page is tricky as 0 - 1 = -1, but pgoff_t |
| 144 | * is unsigned, so the end_index >= start_index |
| 145 | * check below would be true and we'll discard the whole |
| 146 | * file cache which is not what was asked. |
| 147 | */ |
| 148 | if (end_index == 0) |
| 149 | break; |
| 150 | |
| 151 | end_index--; |
| 152 | } |
| 153 | |
| 154 | if (end_index >= start_index) { |
| 155 | unsigned long count; |
| 156 | |
| 157 | /* |
| 158 | * It's common to FADV_DONTNEED right after |
| 159 | * the read or write that instantiates the |
| 160 | * pages, in which case there will be some |
| 161 | * sitting on the local LRU cache. Try to |
| 162 | * avoid the expensive remote drain and the |
| 163 | * second cache tree walk below by flushing |
| 164 | * them out right away. |
| 165 | */ |
| 166 | lru_add_drain(); |
| 167 | |
| 168 | count = invalidate_mapping_pages(mapping, |
| 169 | start_index, end_index); |
| 170 | |
| 171 | /* |
| 172 | * If fewer pages were invalidated than expected then |
| 173 | * it is possible that some of the pages were on |
| 174 | * a per-cpu pagevec for a remote CPU. Drain all |
| 175 | * pagevecs and try again. |
| 176 | */ |
| 177 | if (count < (end_index - start_index + 1)) { |
| 178 | lru_add_drain_all(); |
| 179 | invalidate_mapping_pages(mapping, start_index, |
| 180 | end_index); |
| 181 | } |
| 182 | } |
| 183 | break; |
| 184 | default: |
| 185 | ret = -EINVAL; |
| 186 | } |
| 187 | out: |
| 188 | fdput(f); |
| 189 | return ret; |
| 190 | } |
| 191 | |
| 192 | #ifdef __ARCH_WANT_SYS_FADVISE64 |
| 193 | |
| 194 | SYSCALL_DEFINE4(fadvise64, int, fd, loff_t, offset, size_t, len, int, advice) |
| 195 | { |
| 196 | return sys_fadvise64_64(fd, offset, len, advice); |
| 197 | } |
| 198 | |
| 199 | #endif |