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 30 int ksys_fadvise64_64(int fd, loff_t offset, loff_t len, int advice) 31 { 32 struct fd f = fdget(fd); 33 struct inode *inode; 34 struct address_space *mapping; 35 struct backing_dev_info *bdi; 36 loff_t endbyte; /* inclusive */ 37 pgoff_t start_index; 38 pgoff_t end_index; 39 unsigned long nrpages; 40 int ret = 0; 41 42 if (!f.file) 43 return -EBADF; 44 45 inode = file_inode(f.file); 46 if (S_ISFIFO(inode->i_mode)) { 47 ret = -ESPIPE; 48 goto out; 49 } 50 51 mapping = f.file->f_mapping; 52 if (!mapping || len < 0) { 53 ret = -EINVAL; 54 goto out; 55 } 56 57 bdi = inode_to_bdi(mapping->host); 58 59 if (IS_DAX(inode) || (bdi == &noop_backing_dev_info)) { 60 switch (advice) { 61 case POSIX_FADV_NORMAL: 62 case POSIX_FADV_RANDOM: 63 case POSIX_FADV_SEQUENTIAL: 64 case POSIX_FADV_WILLNEED: 65 case POSIX_FADV_NOREUSE: 66 case POSIX_FADV_DONTNEED: 67 /* no bad return value, but ignore advice */ 68 break; 69 default: 70 ret = -EINVAL; 71 } 72 goto out; 73 } 74 75 /* Careful about overflows. Len == 0 means "as much as possible" */ 76 endbyte = offset + len; 77 if (!len || endbyte < len) 78 endbyte = -1; 79 else 80 endbyte--; /* inclusive */ 81 82 switch (advice) { 83 case POSIX_FADV_NORMAL: 84 f.file->f_ra.ra_pages = bdi->ra_pages; 85 spin_lock(&f.file->f_lock); 86 f.file->f_mode &= ~FMODE_RANDOM; 87 spin_unlock(&f.file->f_lock); 88 break; 89 case POSIX_FADV_RANDOM: 90 spin_lock(&f.file->f_lock); 91 f.file->f_mode |= FMODE_RANDOM; 92 spin_unlock(&f.file->f_lock); 93 break; 94 case POSIX_FADV_SEQUENTIAL: 95 f.file->f_ra.ra_pages = bdi->ra_pages * 2; 96 spin_lock(&f.file->f_lock); 97 f.file->f_mode &= ~FMODE_RANDOM; 98 spin_unlock(&f.file->f_lock); 99 break; 100 case POSIX_FADV_WILLNEED: 101 /* First and last PARTIAL page! */ 102 start_index = offset >> PAGE_SHIFT; 103 end_index = endbyte >> PAGE_SHIFT; 104 105 /* Careful about overflow on the "+1" */ 106 nrpages = end_index - start_index + 1; 107 if (!nrpages) 108 nrpages = ~0UL; 109 110 /* 111 * Ignore return value because fadvise() shall return 112 * success even if filesystem can't retrieve a hint, 113 */ 114 force_page_cache_readahead(mapping, f.file, start_index, 115 nrpages); 116 break; 117 case POSIX_FADV_NOREUSE: 118 break; 119 case POSIX_FADV_DONTNEED: 120 if (!inode_write_congested(mapping->host)) 121 __filemap_fdatawrite_range(mapping, offset, endbyte, 122 WB_SYNC_NONE); 123 124 /* 125 * First and last FULL page! Partial pages are deliberately 126 * preserved on the expectation that it is better to preserve 127 * needed memory than to discard unneeded memory. 128 */ 129 start_index = (offset+(PAGE_SIZE-1)) >> PAGE_SHIFT; 130 end_index = (endbyte >> PAGE_SHIFT); 131 /* 132 * The page at end_index will be inclusively discarded according 133 * by invalidate_mapping_pages(), so subtracting 1 from 134 * end_index means we will skip the last page. But if endbyte 135 * is page aligned or is at the end of file, we should not skip 136 * that page - discarding the last page is safe enough. 137 */ 138 if ((endbyte & ~PAGE_MASK) != ~PAGE_MASK && 139 endbyte != inode->i_size - 1) { 140 /* First page is tricky as 0 - 1 = -1, but pgoff_t 141 * is unsigned, so the end_index >= start_index 142 * check below would be true and we'll discard the whole 143 * file cache which is not what was asked. 144 */ 145 if (end_index == 0) 146 break; 147 148 end_index--; 149 } 150 151 if (end_index >= start_index) { 152 unsigned long count; 153 154 /* 155 * It's common to FADV_DONTNEED right after 156 * the read or write that instantiates the 157 * pages, in which case there will be some 158 * sitting on the local LRU cache. Try to 159 * avoid the expensive remote drain and the 160 * second cache tree walk below by flushing 161 * them out right away. 162 */ 163 lru_add_drain(); 164 165 count = invalidate_mapping_pages(mapping, 166 start_index, end_index); 167 168 /* 169 * If fewer pages were invalidated than expected then 170 * it is possible that some of the pages were on 171 * a per-cpu pagevec for a remote CPU. Drain all 172 * pagevecs and try again. 173 */ 174 if (count < (end_index - start_index + 1)) { 175 lru_add_drain_all(); 176 invalidate_mapping_pages(mapping, start_index, 177 end_index); 178 } 179 } 180 break; 181 default: 182 ret = -EINVAL; 183 } 184 out: 185 fdput(f); 186 return ret; 187 } 188 189 SYSCALL_DEFINE4(fadvise64_64, int, fd, loff_t, offset, loff_t, len, int, advice) 190 { 191 return ksys_fadvise64_64(fd, offset, len, advice); 192 } 193 194 #ifdef __ARCH_WANT_SYS_FADVISE64 195 196 SYSCALL_DEFINE4(fadvise64, int, fd, loff_t, offset, size_t, len, int, advice) 197 { 198 return ksys_fadvise64_64(fd, offset, len, advice); 199 } 200 201 #endif 202