4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
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
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>
27 static struct bio
*get_swap_bio(gfp_t gfp_flags
,
28 struct page
*page
, bio_end_io_t end_io
)
32 bio
= bio_alloc(gfp_flags
, 1);
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;
40 bio
->bi_size
= PAGE_SIZE
;
41 bio
->bi_end_io
= end_io
;
46 void end_swap_bio_write(struct bio
*bio
, int err
)
48 const int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
49 struct page
*page
= bio
->bi_io_vec
[0].bv_page
;
52 if (!task_in_mtkpasr(current
)) {
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)
61 * Also clear PG_reclaim to avoid rotate_reclaimable_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
);
70 ClearPageReclaim(page
);
72 end_page_writeback(page
);
76 void end_swap_bio_read(struct bio
*bio
, int err
)
78 const int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
79 struct page
*page
= bio
->bi_io_vec
[0].bv_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
);
91 SetPageUptodate(page
);
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
99 if (likely(PageSwapCache(page
))) {
100 struct swap_info_struct
*sis
;
102 sis
= page_swap_info(page
);
103 if (sis
->flags
& SWP_BLKDEV
) {
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
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.
120 struct gendisk
*disk
= sis
->bdev
->bd_disk
;
121 if (disk
->fops
->swap_slot_free_notify
) {
123 unsigned long offset
;
125 entry
.val
= page_private(page
);
126 offset
= swp_offset(entry
);
129 disk
->fops
->swap_slot_free_notify(sis
->bdev
,
140 int generic_swapfile_activate(struct swap_info_struct
*sis
,
141 struct file
*swap_file
,
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
;
149 sector_t probe_block
;
151 sector_t lowest_block
= -1;
152 sector_t highest_block
= 0;
156 blkbits
= inode
->i_blkbits
;
157 blocks_per_page
= PAGE_SIZE
>> blkbits
;
160 * Map all the blocks into the extent list. This code doesn't try
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
;
171 first_block
= bmap(inode
, probe_block
);
172 if (first_block
== 0)
176 * It must be PAGE_SIZE aligned on-disk
178 if (first_block
& (blocks_per_page
- 1)) {
183 for (block_in_page
= 1; block_in_page
< blocks_per_page
;
187 block
= bmap(inode
, probe_block
+ block_in_page
);
190 if (block
!= first_block
+ block_in_page
) {
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
;
206 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
208 ret
= add_swap_extent(sis
, page_no
, 1, first_block
);
213 probe_block
+= blocks_per_page
;
218 *span
= 1 + highest_block
- lowest_block
;
220 page_no
= 1; /* force Empty message */
222 sis
->pages
= page_no
- 1;
223 sis
->highest_bit
= page_no
- 1;
227 printk(KERN_ERR
"swapon: swapfile has holes\n");
233 * We may have stale swap cache pages in memory: notice
234 * them here and get rid of the unnecessary final write.
236 int swap_writepage(struct page
*page
, struct writeback_control
*wbc
)
240 if (try_to_free_swap(page
)) {
244 if (frontswap_store(page
) == 0) {
245 set_page_writeback(page
);
247 end_page_writeback(page
);
250 ret
= __swap_writepage(page
, wbc
, end_swap_bio_write
);
255 int __swap_writepage(struct page
*page
, struct writeback_control
*wbc
,
256 void (*end_write_func
)(struct bio
*, int))
259 int ret
= 0, rw
= WRITE
;
260 struct swap_info_struct
*sis
= page_swap_info(page
);
262 if (sis
->flags
& SWP_FILE
) {
264 struct file
*swap_file
= sis
->swap_file
;
265 struct address_space
*mapping
= swap_file
->f_mapping
;
267 .iov_base
= kmap(page
),
268 .iov_len
= PAGE_SIZE
,
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
;
276 set_page_writeback(page
);
278 ret
= mapping
->a_ops
->direct_IO(KERNEL_WRITE
,
282 if (ret
== PAGE_SIZE
) {
283 count_vm_event(PSWPOUT
);
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
296 set_page_dirty(page
);
297 ClearPageReclaim(page
);
298 pr_err_ratelimited("Write error on dio swapfile (%Lu)\n",
299 page_file_offset(page
));
301 end_page_writeback(page
);
305 bio
= get_swap_bio(GFP_NOIO
, page
, end_write_func
);
307 set_page_dirty(page
);
312 if (wbc
->sync_mode
== WB_SYNC_ALL
)
318 count_vm_event(PSWPOUT
);
319 set_page_writeback(page
);
326 int swap_readpage(struct page
*page
)
330 struct swap_info_struct
*sis
= page_swap_info(page
);
332 VM_BUG_ON(!PageLocked(page
));
333 VM_BUG_ON(PageUptodate(page
));
334 if (frontswap_load(page
) == 0) {
335 SetPageUptodate(page
);
340 if (sis
->flags
& SWP_FILE
) {
341 struct file
*swap_file
= sis
->swap_file
;
342 struct address_space
*mapping
= swap_file
->f_mapping
;
344 ret
= mapping
->a_ops
->readpage(swap_file
, page
);
349 count_vm_event(PSWPIN
);
354 bio
= get_swap_bio(GFP_KERNEL
, page
, end_swap_bio_read
);
364 count_vm_event(PSWPIN
);
365 submit_bio(READ
, bio
);
370 int swap_set_page_dirty(struct page
*page
)
372 struct swap_info_struct
*sis
= page_swap_info(page
);
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
);
378 return __set_page_dirty_no_writeback(page
);