| 1 | /* |
| 2 | * linux/mm/page_io.c |
| 3 | * |
| 4 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds |
| 5 | * |
| 6 | * Swap reorganised 29.12.95, |
| 7 | * Asynchronous swapping added 30.12.95. Stephen Tweedie |
| 8 | * Removed race in async swapping. 14.4.1996. Bruno Haible |
| 9 | * Add swap of shared pages through the page cache. 20.2.1998. Stephen Tweedie |
| 10 | * Always use brw_page, life becomes simpler. 12 May 1998 Eric Biederman |
| 11 | */ |
| 12 | |
| 13 | #include <linux/mm.h> |
| 14 | #include <linux/kernel_stat.h> |
| 15 | #include <linux/gfp.h> |
| 16 | #include <linux/pagemap.h> |
| 17 | #include <linux/swap.h> |
| 18 | #include <linux/bio.h> |
| 19 | #include <linux/swapops.h> |
| 20 | #include <linux/buffer_head.h> |
| 21 | #include <linux/writeback.h> |
| 22 | #include <linux/frontswap.h> |
| 23 | #include <linux/aio.h> |
| 24 | #include <linux/blkdev.h> |
| 25 | #include <asm/pgtable.h> |
| 26 | |
| 27 | static struct bio *get_swap_bio(gfp_t gfp_flags, |
| 28 | struct page *page, bio_end_io_t end_io) |
| 29 | { |
| 30 | struct bio *bio; |
| 31 | |
| 32 | bio = bio_alloc(gfp_flags, 1); |
| 33 | if (bio) { |
| 34 | bio->bi_sector = map_swap_page(page, &bio->bi_bdev); |
| 35 | bio->bi_sector <<= PAGE_SHIFT - 9; |
| 36 | bio->bi_io_vec[0].bv_page = page; |
| 37 | bio->bi_io_vec[0].bv_len = PAGE_SIZE; |
| 38 | bio->bi_io_vec[0].bv_offset = 0; |
| 39 | bio->bi_vcnt = 1; |
| 40 | bio->bi_size = PAGE_SIZE; |
| 41 | bio->bi_end_io = end_io; |
| 42 | } |
| 43 | return bio; |
| 44 | } |
| 45 | |
| 46 | void end_swap_bio_write(struct bio *bio, int err) |
| 47 | { |
| 48 | const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); |
| 49 | struct page *page = bio->bi_io_vec[0].bv_page; |
| 50 | |
| 51 | if (!uptodate) { |
| 52 | if (!task_in_mtkpasr(current)) { |
| 53 | SetPageError(page); |
| 54 | } |
| 55 | /* |
| 56 | * We failed to write the page out to swap-space. |
| 57 | * Re-dirty the page in order to avoid it being reclaimed. |
| 58 | * Also print a dire warning that things will go BAD (tm) |
| 59 | * very quickly. |
| 60 | * |
| 61 | * Also clear PG_reclaim to avoid rotate_reclaimable_page() |
| 62 | */ |
| 63 | set_page_dirty(page); |
| 64 | if (!task_in_mtkpasr(current)) { |
| 65 | printk(KERN_ALERT "Write-error on swap-device (%u:%u:%Lu)\n", |
| 66 | imajor(bio->bi_bdev->bd_inode), |
| 67 | iminor(bio->bi_bdev->bd_inode), |
| 68 | (unsigned long long)bio->bi_sector); |
| 69 | } |
| 70 | ClearPageReclaim(page); |
| 71 | } |
| 72 | end_page_writeback(page); |
| 73 | bio_put(bio); |
| 74 | } |
| 75 | |
| 76 | void end_swap_bio_read(struct bio *bio, int err) |
| 77 | { |
| 78 | const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); |
| 79 | struct page *page = bio->bi_io_vec[0].bv_page; |
| 80 | |
| 81 | if (!uptodate) { |
| 82 | SetPageError(page); |
| 83 | ClearPageUptodate(page); |
| 84 | printk(KERN_ALERT "Read-error on swap-device (%u:%u:%Lu)\n", |
| 85 | imajor(bio->bi_bdev->bd_inode), |
| 86 | iminor(bio->bi_bdev->bd_inode), |
| 87 | (unsigned long long)bio->bi_sector); |
| 88 | goto out; |
| 89 | } |
| 90 | |
| 91 | SetPageUptodate(page); |
| 92 | |
| 93 | /* |
| 94 | * There is no guarantee that the page is in swap cache - the software |
| 95 | * suspend code (at least) uses end_swap_bio_read() against a non- |
| 96 | * swapcache page. So we must check PG_swapcache before proceeding with |
| 97 | * this optimization. |
| 98 | */ |
| 99 | if (likely(PageSwapCache(page))) { |
| 100 | struct swap_info_struct *sis; |
| 101 | |
| 102 | sis = page_swap_info(page); |
| 103 | if (sis->flags & SWP_BLKDEV) { |
| 104 | /* |
| 105 | * The swap subsystem performs lazy swap slot freeing, |
| 106 | * expecting that the page will be swapped out again. |
| 107 | * So we can avoid an unnecessary write if the page |
| 108 | * isn't redirtied. |
| 109 | * This is good for real swap storage because we can |
| 110 | * reduce unnecessary I/O and enhance wear-leveling |
| 111 | * if an SSD is used as the as swap device. |
| 112 | * But if in-memory swap device (eg zram) is used, |
| 113 | * this causes a duplicated copy between uncompressed |
| 114 | * data in VM-owned memory and compressed data in |
| 115 | * zram-owned memory. So let's free zram-owned memory |
| 116 | * and make the VM-owned decompressed page *dirty*, |
| 117 | * so the page should be swapped out somewhere again if |
| 118 | * we again wish to reclaim it. |
| 119 | */ |
| 120 | struct gendisk *disk = sis->bdev->bd_disk; |
| 121 | if (disk->fops->swap_slot_free_notify) { |
| 122 | swp_entry_t entry; |
| 123 | unsigned long offset; |
| 124 | |
| 125 | entry.val = page_private(page); |
| 126 | offset = swp_offset(entry); |
| 127 | |
| 128 | SetPageDirty(page); |
| 129 | disk->fops->swap_slot_free_notify(sis->bdev, |
| 130 | offset); |
| 131 | } |
| 132 | } |
| 133 | } |
| 134 | |
| 135 | out: |
| 136 | unlock_page(page); |
| 137 | bio_put(bio); |
| 138 | } |
| 139 | |
| 140 | int generic_swapfile_activate(struct swap_info_struct *sis, |
| 141 | struct file *swap_file, |
| 142 | sector_t *span) |
| 143 | { |
| 144 | struct address_space *mapping = swap_file->f_mapping; |
| 145 | struct inode *inode = mapping->host; |
| 146 | unsigned blocks_per_page; |
| 147 | unsigned long page_no; |
| 148 | unsigned blkbits; |
| 149 | sector_t probe_block; |
| 150 | sector_t last_block; |
| 151 | sector_t lowest_block = -1; |
| 152 | sector_t highest_block = 0; |
| 153 | int nr_extents = 0; |
| 154 | int ret; |
| 155 | |
| 156 | blkbits = inode->i_blkbits; |
| 157 | blocks_per_page = PAGE_SIZE >> blkbits; |
| 158 | |
| 159 | /* |
| 160 | * Map all the blocks into the extent list. This code doesn't try |
| 161 | * to be very smart. |
| 162 | */ |
| 163 | probe_block = 0; |
| 164 | page_no = 0; |
| 165 | last_block = i_size_read(inode) >> blkbits; |
| 166 | while ((probe_block + blocks_per_page) <= last_block && |
| 167 | page_no < sis->max) { |
| 168 | unsigned block_in_page; |
| 169 | sector_t first_block; |
| 170 | |
| 171 | first_block = bmap(inode, probe_block); |
| 172 | if (first_block == 0) |
| 173 | goto bad_bmap; |
| 174 | |
| 175 | /* |
| 176 | * It must be PAGE_SIZE aligned on-disk |
| 177 | */ |
| 178 | if (first_block & (blocks_per_page - 1)) { |
| 179 | probe_block++; |
| 180 | goto reprobe; |
| 181 | } |
| 182 | |
| 183 | for (block_in_page = 1; block_in_page < blocks_per_page; |
| 184 | block_in_page++) { |
| 185 | sector_t block; |
| 186 | |
| 187 | block = bmap(inode, probe_block + block_in_page); |
| 188 | if (block == 0) |
| 189 | goto bad_bmap; |
| 190 | if (block != first_block + block_in_page) { |
| 191 | /* Discontiguity */ |
| 192 | probe_block++; |
| 193 | goto reprobe; |
| 194 | } |
| 195 | } |
| 196 | |
| 197 | first_block >>= (PAGE_SHIFT - blkbits); |
| 198 | if (page_no) { /* exclude the header page */ |
| 199 | if (first_block < lowest_block) |
| 200 | lowest_block = first_block; |
| 201 | if (first_block > highest_block) |
| 202 | highest_block = first_block; |
| 203 | } |
| 204 | |
| 205 | /* |
| 206 | * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks |
| 207 | */ |
| 208 | ret = add_swap_extent(sis, page_no, 1, first_block); |
| 209 | if (ret < 0) |
| 210 | goto out; |
| 211 | nr_extents += ret; |
| 212 | page_no++; |
| 213 | probe_block += blocks_per_page; |
| 214 | reprobe: |
| 215 | continue; |
| 216 | } |
| 217 | ret = nr_extents; |
| 218 | *span = 1 + highest_block - lowest_block; |
| 219 | if (page_no == 0) |
| 220 | page_no = 1; /* force Empty message */ |
| 221 | sis->max = page_no; |
| 222 | sis->pages = page_no - 1; |
| 223 | sis->highest_bit = page_no - 1; |
| 224 | out: |
| 225 | return ret; |
| 226 | bad_bmap: |
| 227 | printk(KERN_ERR "swapon: swapfile has holes\n"); |
| 228 | ret = -EINVAL; |
| 229 | goto out; |
| 230 | } |
| 231 | |
| 232 | /* |
| 233 | * We may have stale swap cache pages in memory: notice |
| 234 | * them here and get rid of the unnecessary final write. |
| 235 | */ |
| 236 | int swap_writepage(struct page *page, struct writeback_control *wbc) |
| 237 | { |
| 238 | int ret = 0; |
| 239 | |
| 240 | if (try_to_free_swap(page)) { |
| 241 | unlock_page(page); |
| 242 | goto out; |
| 243 | } |
| 244 | if (frontswap_store(page) == 0) { |
| 245 | set_page_writeback(page); |
| 246 | unlock_page(page); |
| 247 | end_page_writeback(page); |
| 248 | goto out; |
| 249 | } |
| 250 | ret = __swap_writepage(page, wbc, end_swap_bio_write); |
| 251 | out: |
| 252 | return ret; |
| 253 | } |
| 254 | |
| 255 | int __swap_writepage(struct page *page, struct writeback_control *wbc, |
| 256 | void (*end_write_func)(struct bio *, int)) |
| 257 | { |
| 258 | struct bio *bio; |
| 259 | int ret = 0, rw = WRITE; |
| 260 | struct swap_info_struct *sis = page_swap_info(page); |
| 261 | |
| 262 | if (sis->flags & SWP_FILE) { |
| 263 | struct kiocb kiocb; |
| 264 | struct file *swap_file = sis->swap_file; |
| 265 | struct address_space *mapping = swap_file->f_mapping; |
| 266 | struct iovec iov = { |
| 267 | .iov_base = kmap(page), |
| 268 | .iov_len = PAGE_SIZE, |
| 269 | }; |
| 270 | |
| 271 | init_sync_kiocb(&kiocb, swap_file); |
| 272 | kiocb.ki_pos = page_file_offset(page); |
| 273 | kiocb.ki_left = PAGE_SIZE; |
| 274 | kiocb.ki_nbytes = PAGE_SIZE; |
| 275 | |
| 276 | set_page_writeback(page); |
| 277 | unlock_page(page); |
| 278 | ret = mapping->a_ops->direct_IO(KERNEL_WRITE, |
| 279 | &kiocb, &iov, |
| 280 | kiocb.ki_pos, 1); |
| 281 | kunmap(page); |
| 282 | if (ret == PAGE_SIZE) { |
| 283 | count_vm_event(PSWPOUT); |
| 284 | ret = 0; |
| 285 | } else { |
| 286 | /* |
| 287 | * In the case of swap-over-nfs, this can be a |
| 288 | * temporary failure if the system has limited |
| 289 | * memory for allocating transmit buffers. |
| 290 | * Mark the page dirty and avoid |
| 291 | * rotate_reclaimable_page but rate-limit the |
| 292 | * messages but do not flag PageError like |
| 293 | * the normal direct-to-bio case as it could |
| 294 | * be temporary. |
| 295 | */ |
| 296 | set_page_dirty(page); |
| 297 | ClearPageReclaim(page); |
| 298 | pr_err_ratelimited("Write error on dio swapfile (%Lu)\n", |
| 299 | page_file_offset(page)); |
| 300 | } |
| 301 | end_page_writeback(page); |
| 302 | return ret; |
| 303 | } |
| 304 | |
| 305 | bio = get_swap_bio(GFP_NOIO, page, end_write_func); |
| 306 | if (bio == NULL) { |
| 307 | set_page_dirty(page); |
| 308 | unlock_page(page); |
| 309 | ret = -ENOMEM; |
| 310 | goto out; |
| 311 | } |
| 312 | if (wbc->sync_mode == WB_SYNC_ALL) |
| 313 | rw |= REQ_SYNC; |
| 314 | |
| 315 | #ifdef CONFIG_ZRAM |
| 316 | current->swap_out++; |
| 317 | #endif |
| 318 | count_vm_event(PSWPOUT); |
| 319 | set_page_writeback(page); |
| 320 | unlock_page(page); |
| 321 | submit_bio(rw, bio); |
| 322 | out: |
| 323 | return ret; |
| 324 | } |
| 325 | |
| 326 | int swap_readpage(struct page *page) |
| 327 | { |
| 328 | struct bio *bio; |
| 329 | int ret = 0; |
| 330 | struct swap_info_struct *sis = page_swap_info(page); |
| 331 | |
| 332 | VM_BUG_ON(!PageLocked(page)); |
| 333 | VM_BUG_ON(PageUptodate(page)); |
| 334 | if (frontswap_load(page) == 0) { |
| 335 | SetPageUptodate(page); |
| 336 | unlock_page(page); |
| 337 | goto out; |
| 338 | } |
| 339 | |
| 340 | if (sis->flags & SWP_FILE) { |
| 341 | struct file *swap_file = sis->swap_file; |
| 342 | struct address_space *mapping = swap_file->f_mapping; |
| 343 | |
| 344 | ret = mapping->a_ops->readpage(swap_file, page); |
| 345 | if (!ret) { |
| 346 | #ifdef CONFIG_ZRAM |
| 347 | current->swap_in++; |
| 348 | #endif |
| 349 | count_vm_event(PSWPIN); |
| 350 | } |
| 351 | return ret; |
| 352 | } |
| 353 | |
| 354 | bio = get_swap_bio(GFP_KERNEL, page, end_swap_bio_read); |
| 355 | if (bio == NULL) { |
| 356 | unlock_page(page); |
| 357 | ret = -ENOMEM; |
| 358 | goto out; |
| 359 | } |
| 360 | |
| 361 | #ifdef CONFIG_ZRAM |
| 362 | current->swap_in++; |
| 363 | #endif |
| 364 | count_vm_event(PSWPIN); |
| 365 | submit_bio(READ, bio); |
| 366 | out: |
| 367 | return ret; |
| 368 | } |
| 369 | |
| 370 | int swap_set_page_dirty(struct page *page) |
| 371 | { |
| 372 | struct swap_info_struct *sis = page_swap_info(page); |
| 373 | |
| 374 | if (sis->flags & SWP_FILE) { |
| 375 | struct address_space *mapping = sis->swap_file->f_mapping; |
| 376 | return mapping->a_ops->set_page_dirty(page); |
| 377 | } else { |
| 378 | return __set_page_dirty_no_writeback(page); |
| 379 | } |
| 380 | } |