1 // SPDX-License-Identifier: GPL-2.0 1 // SPDX-License-Identifier: GPL-2.0
2 /* 2 /*
3 * fs/mpage.c 3 * fs/mpage.c
4 * 4 *
5 * Copyright (C) 2002, Linus Torvalds. 5 * Copyright (C) 2002, Linus Torvalds.
6 * 6 *
7 * Contains functions related to preparing and 7 * Contains functions related to preparing and submitting BIOs which contain
8 * multiple pagecache pages. 8 * multiple pagecache pages.
9 * 9 *
10 * 15May2002 Andrew Morton 10 * 15May2002 Andrew Morton
11 * Initial version 11 * Initial version
12 * 27Jun2002 axboe@suse.de 12 * 27Jun2002 axboe@suse.de
13 * use bio_add_page() to build bi 13 * use bio_add_page() to build bio's just the right size
14 */ 14 */
15 15
16 #include <linux/kernel.h> 16 #include <linux/kernel.h>
17 #include <linux/export.h> 17 #include <linux/export.h>
18 #include <linux/mm.h> 18 #include <linux/mm.h>
19 #include <linux/kdev_t.h> 19 #include <linux/kdev_t.h>
20 #include <linux/gfp.h> 20 #include <linux/gfp.h>
21 #include <linux/bio.h> 21 #include <linux/bio.h>
22 #include <linux/fs.h> 22 #include <linux/fs.h>
23 #include <linux/buffer_head.h> 23 #include <linux/buffer_head.h>
24 #include <linux/blkdev.h> 24 #include <linux/blkdev.h>
25 #include <linux/highmem.h> 25 #include <linux/highmem.h>
26 #include <linux/prefetch.h> 26 #include <linux/prefetch.h>
27 #include <linux/mpage.h> 27 #include <linux/mpage.h>
28 #include <linux/mm_inline.h> 28 #include <linux/mm_inline.h>
29 #include <linux/writeback.h> 29 #include <linux/writeback.h>
30 #include <linux/backing-dev.h> 30 #include <linux/backing-dev.h>
31 #include <linux/pagevec.h> 31 #include <linux/pagevec.h>
>> 32 #include <linux/cleancache.h>
32 #include "internal.h" 33 #include "internal.h"
33 34
34 /* 35 /*
35 * I/O completion handler for multipage BIOs. 36 * I/O completion handler for multipage BIOs.
36 * 37 *
37 * The mpage code never puts partial pages int 38 * The mpage code never puts partial pages into a BIO (except for end-of-file).
38 * If a page does not map to a contiguous run 39 * If a page does not map to a contiguous run of blocks then it simply falls
39 * back to block_read_full_folio(). !! 40 * back to block_read_full_page().
40 * 41 *
41 * Why is this? If a page's completion depend 42 * Why is this? If a page's completion depends on a number of different BIOs
42 * which can complete in any order (or at the 43 * which can complete in any order (or at the same time) then determining the
43 * status of that page is hard. See end_buffe 44 * status of that page is hard. See end_buffer_async_read() for the details.
44 * There is no point in duplicating all that c 45 * There is no point in duplicating all that complexity.
45 */ 46 */
46 static void mpage_read_end_io(struct bio *bio) !! 47 static void mpage_end_io(struct bio *bio)
47 { 48 {
48 struct folio_iter fi; !! 49 struct bio_vec *bv;
49 int err = blk_status_to_errno(bio->bi_ !! 50 int i;
50 51
51 bio_for_each_folio_all(fi, bio) !! 52 bio_for_each_segment_all(bv, bio, i) {
52 folio_end_read(fi.folio, err = !! 53 struct page *page = bv->bv_page;
53 !! 54 page_endio(page, bio_op(bio),
54 bio_put(bio); !! 55 blk_status_to_errno(bio->bi_status));
55 } <<
56 <<
57 static void mpage_write_end_io(struct bio *bio <<
58 { <<
59 struct folio_iter fi; <<
60 int err = blk_status_to_errno(bio->bi_ <<
61 <<
62 bio_for_each_folio_all(fi, bio) { <<
63 if (err) <<
64 mapping_set_error(fi.f <<
65 folio_end_writeback(fi.folio); <<
66 } 56 }
67 57
68 bio_put(bio); 58 bio_put(bio);
69 } 59 }
70 60
71 static struct bio *mpage_bio_submit_read(struc !! 61 static struct bio *mpage_bio_submit(int op, int op_flags, struct bio *bio)
72 { 62 {
73 bio->bi_end_io = mpage_read_end_io; !! 63 bio->bi_end_io = mpage_end_io;
74 guard_bio_eod(bio); !! 64 bio_set_op_attrs(bio, op, op_flags);
>> 65 guard_bio_eod(op, bio);
75 submit_bio(bio); 66 submit_bio(bio);
76 return NULL; 67 return NULL;
77 } 68 }
78 69
79 static struct bio *mpage_bio_submit_write(stru !! 70 static struct bio *
>> 71 mpage_alloc(struct block_device *bdev,
>> 72 sector_t first_sector, int nr_vecs,
>> 73 gfp_t gfp_flags)
80 { 74 {
81 bio->bi_end_io = mpage_write_end_io; !! 75 struct bio *bio;
82 guard_bio_eod(bio); !! 76
83 submit_bio(bio); !! 77 /* Restrict the given (page cache) mask for slab allocations */
84 return NULL; !! 78 gfp_flags &= GFP_KERNEL;
>> 79 bio = bio_alloc(gfp_flags, nr_vecs);
>> 80
>> 81 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
>> 82 while (!bio && (nr_vecs /= 2))
>> 83 bio = bio_alloc(gfp_flags, nr_vecs);
>> 84 }
>> 85
>> 86 if (bio) {
>> 87 bio_set_dev(bio, bdev);
>> 88 bio->bi_iter.bi_sector = first_sector;
>> 89 }
>> 90 return bio;
85 } 91 }
86 92
87 /* 93 /*
88 * support function for mpage_readahead. The !! 94 * support function for mpage_readpages. The fs supplied get_block might
89 * return an up to date buffer. This is used 95 * return an up to date buffer. This is used to map that buffer into
90 * the page, which allows read_folio to avoid !! 96 * the page, which allows readpage to avoid triggering a duplicate call
91 * to get_block. 97 * to get_block.
92 * 98 *
93 * The idea is to avoid adding buffers to page 99 * The idea is to avoid adding buffers to pages that don't already have
94 * them. So when the buffer is up to date and 100 * them. So when the buffer is up to date and the page size == block size,
95 * this marks the page up to date instead of a 101 * this marks the page up to date instead of adding new buffers.
96 */ 102 */
97 static void map_buffer_to_folio(struct folio * !! 103 static void
98 int page_block) !! 104 map_buffer_to_page(struct page *page, struct buffer_head *bh, int page_block)
99 { 105 {
100 struct inode *inode = folio->mapping-> !! 106 struct inode *inode = page->mapping->host;
101 struct buffer_head *page_bh, *head; 107 struct buffer_head *page_bh, *head;
102 int block = 0; 108 int block = 0;
103 109
104 head = folio_buffers(folio); !! 110 if (!page_has_buffers(page)) {
105 if (!head) { <<
106 /* 111 /*
107 * don't make any buffers if t 112 * don't make any buffers if there is only one buffer on
108 * the folio and the folio jus !! 113 * the page and the page just needs to be set up to date
109 */ 114 */
110 if (inode->i_blkbits == PAGE_S 115 if (inode->i_blkbits == PAGE_SHIFT &&
111 buffer_uptodate(bh)) { 116 buffer_uptodate(bh)) {
112 folio_mark_uptodate(fo !! 117 SetPageUptodate(page);
113 return; 118 return;
114 } 119 }
115 head = create_empty_buffers(fo !! 120 create_empty_buffers(page, i_blocksize(inode), 0);
116 } 121 }
117 !! 122 head = page_buffers(page);
118 page_bh = head; 123 page_bh = head;
119 do { 124 do {
120 if (block == page_block) { 125 if (block == page_block) {
121 page_bh->b_state = bh- 126 page_bh->b_state = bh->b_state;
122 page_bh->b_bdev = bh-> 127 page_bh->b_bdev = bh->b_bdev;
123 page_bh->b_blocknr = b 128 page_bh->b_blocknr = bh->b_blocknr;
124 break; 129 break;
125 } 130 }
126 page_bh = page_bh->b_this_page 131 page_bh = page_bh->b_this_page;
127 block++; 132 block++;
128 } while (page_bh != head); 133 } while (page_bh != head);
129 } 134 }
130 135
131 struct mpage_readpage_args { 136 struct mpage_readpage_args {
132 struct bio *bio; 137 struct bio *bio;
133 struct folio *folio; !! 138 struct page *page;
134 unsigned int nr_pages; 139 unsigned int nr_pages;
135 bool is_readahead; 140 bool is_readahead;
136 sector_t last_block_in_bio; 141 sector_t last_block_in_bio;
137 struct buffer_head map_bh; 142 struct buffer_head map_bh;
138 unsigned long first_logical_block; 143 unsigned long first_logical_block;
139 get_block_t *get_block; 144 get_block_t *get_block;
140 }; 145 };
141 146
142 /* 147 /*
143 * This is the worker routine which does all t 148 * This is the worker routine which does all the work of mapping the disk
144 * blocks and constructs largest possible bios 149 * blocks and constructs largest possible bios, submits them for IO if the
145 * blocks are not contiguous on the disk. 150 * blocks are not contiguous on the disk.
146 * 151 *
147 * We pass a buffer_head back and forth and us 152 * We pass a buffer_head back and forth and use its buffer_mapped() flag to
148 * represent the validity of its disk mapping 153 * represent the validity of its disk mapping and to decide when to do the next
149 * get_block() call. 154 * get_block() call.
150 */ 155 */
151 static struct bio *do_mpage_readpage(struct mp 156 static struct bio *do_mpage_readpage(struct mpage_readpage_args *args)
152 { 157 {
153 struct folio *folio = args->folio; !! 158 struct page *page = args->page;
154 struct inode *inode = folio->mapping-> !! 159 struct inode *inode = page->mapping->host;
155 const unsigned blkbits = inode->i_blkb 160 const unsigned blkbits = inode->i_blkbits;
156 const unsigned blocks_per_page = PAGE_ 161 const unsigned blocks_per_page = PAGE_SIZE >> blkbits;
157 const unsigned blocksize = 1 << blkbit 162 const unsigned blocksize = 1 << blkbits;
158 struct buffer_head *map_bh = &args->ma 163 struct buffer_head *map_bh = &args->map_bh;
159 sector_t block_in_file; 164 sector_t block_in_file;
160 sector_t last_block; 165 sector_t last_block;
161 sector_t last_block_in_file; 166 sector_t last_block_in_file;
162 sector_t first_block; !! 167 sector_t blocks[MAX_BUF_PER_PAGE];
163 unsigned page_block; 168 unsigned page_block;
164 unsigned first_hole = blocks_per_page; 169 unsigned first_hole = blocks_per_page;
165 struct block_device *bdev = NULL; 170 struct block_device *bdev = NULL;
166 int length; 171 int length;
167 int fully_mapped = 1; 172 int fully_mapped = 1;
168 blk_opf_t opf = REQ_OP_READ; !! 173 int op_flags;
169 unsigned nblocks; 174 unsigned nblocks;
170 unsigned relative_block; 175 unsigned relative_block;
171 gfp_t gfp = mapping_gfp_constraint(fol !! 176 gfp_t gfp;
172 <<
173 /* MAX_BUF_PER_PAGE, for example */ <<
174 VM_BUG_ON_FOLIO(folio_test_large(folio <<
175 177
176 if (args->is_readahead) { 178 if (args->is_readahead) {
177 opf |= REQ_RAHEAD; !! 179 op_flags = REQ_RAHEAD;
178 gfp |= __GFP_NORETRY | __GFP_N !! 180 gfp = readahead_gfp_mask(page->mapping);
>> 181 } else {
>> 182 op_flags = 0;
>> 183 gfp = mapping_gfp_constraint(page->mapping, GFP_KERNEL);
179 } 184 }
180 185
181 if (folio_buffers(folio)) !! 186 if (page_has_buffers(page))
182 goto confused; 187 goto confused;
183 188
184 block_in_file = (sector_t)folio->index !! 189 block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits);
185 last_block = block_in_file + args->nr_ 190 last_block = block_in_file + args->nr_pages * blocks_per_page;
186 last_block_in_file = (i_size_read(inod 191 last_block_in_file = (i_size_read(inode) + blocksize - 1) >> blkbits;
187 if (last_block > last_block_in_file) 192 if (last_block > last_block_in_file)
188 last_block = last_block_in_fil 193 last_block = last_block_in_file;
189 page_block = 0; 194 page_block = 0;
190 195
191 /* 196 /*
192 * Map blocks using the result from th 197 * Map blocks using the result from the previous get_blocks call first.
193 */ 198 */
194 nblocks = map_bh->b_size >> blkbits; 199 nblocks = map_bh->b_size >> blkbits;
195 if (buffer_mapped(map_bh) && 200 if (buffer_mapped(map_bh) &&
196 block_in_file > args-> 201 block_in_file > args->first_logical_block &&
197 block_in_file < (args- 202 block_in_file < (args->first_logical_block + nblocks)) {
198 unsigned map_offset = block_in 203 unsigned map_offset = block_in_file - args->first_logical_block;
199 unsigned last = nblocks - map_ 204 unsigned last = nblocks - map_offset;
200 205
201 first_block = map_bh->b_blockn <<
202 for (relative_block = 0; ; rel 206 for (relative_block = 0; ; relative_block++) {
203 if (relative_block == 207 if (relative_block == last) {
204 clear_buffer_m 208 clear_buffer_mapped(map_bh);
205 break; 209 break;
206 } 210 }
207 if (page_block == bloc 211 if (page_block == blocks_per_page)
208 break; 212 break;
>> 213 blocks[page_block] = map_bh->b_blocknr + map_offset +
>> 214 relative_block;
209 page_block++; 215 page_block++;
210 block_in_file++; 216 block_in_file++;
211 } 217 }
212 bdev = map_bh->b_bdev; 218 bdev = map_bh->b_bdev;
213 } 219 }
214 220
215 /* 221 /*
216 * Then do more get_blocks calls until !! 222 * Then do more get_blocks calls until we are done with this page.
217 */ 223 */
218 map_bh->b_folio = folio; !! 224 map_bh->b_page = page;
219 while (page_block < blocks_per_page) { 225 while (page_block < blocks_per_page) {
220 map_bh->b_state = 0; 226 map_bh->b_state = 0;
221 map_bh->b_size = 0; 227 map_bh->b_size = 0;
222 228
223 if (block_in_file < last_block 229 if (block_in_file < last_block) {
224 map_bh->b_size = (last 230 map_bh->b_size = (last_block-block_in_file) << blkbits;
225 if (args->get_block(in 231 if (args->get_block(inode, block_in_file, map_bh, 0))
226 goto confused; 232 goto confused;
227 args->first_logical_bl 233 args->first_logical_block = block_in_file;
228 } 234 }
229 235
230 if (!buffer_mapped(map_bh)) { 236 if (!buffer_mapped(map_bh)) {
231 fully_mapped = 0; 237 fully_mapped = 0;
232 if (first_hole == bloc 238 if (first_hole == blocks_per_page)
233 first_hole = p 239 first_hole = page_block;
234 page_block++; 240 page_block++;
235 block_in_file++; 241 block_in_file++;
236 continue; 242 continue;
237 } 243 }
238 244
239 /* some filesystems will copy 245 /* some filesystems will copy data into the page during
240 * the get_block call, in whic 246 * the get_block call, in which case we don't want to
241 * read it again. map_buffer_ !! 247 * read it again. map_buffer_to_page copies the data
242 * we just collected from get_ !! 248 * we just collected from get_block into the page's buffers
243 * so read_folio doesn't have !! 249 * so readpage doesn't have to repeat the get_block call
244 */ 250 */
245 if (buffer_uptodate(map_bh)) { 251 if (buffer_uptodate(map_bh)) {
246 map_buffer_to_folio(fo !! 252 map_buffer_to_page(page, map_bh, page_block);
247 goto confused; 253 goto confused;
248 } 254 }
249 255
250 if (first_hole != blocks_per_p 256 if (first_hole != blocks_per_page)
251 goto confused; 257 goto confused; /* hole -> non-hole */
252 258
253 /* Contiguous blocks? */ 259 /* Contiguous blocks? */
254 if (!page_block) !! 260 if (page_block && blocks[page_block-1] != map_bh->b_blocknr-1)
255 first_block = map_bh-> <<
256 else if (first_block + page_bl <<
257 goto confused; 261 goto confused;
258 nblocks = map_bh->b_size >> bl 262 nblocks = map_bh->b_size >> blkbits;
259 for (relative_block = 0; ; rel 263 for (relative_block = 0; ; relative_block++) {
260 if (relative_block == 264 if (relative_block == nblocks) {
261 clear_buffer_m 265 clear_buffer_mapped(map_bh);
262 break; 266 break;
263 } else if (page_block 267 } else if (page_block == blocks_per_page)
264 break; 268 break;
>> 269 blocks[page_block] = map_bh->b_blocknr+relative_block;
265 page_block++; 270 page_block++;
266 block_in_file++; 271 block_in_file++;
267 } 272 }
268 bdev = map_bh->b_bdev; 273 bdev = map_bh->b_bdev;
269 } 274 }
270 275
271 if (first_hole != blocks_per_page) { 276 if (first_hole != blocks_per_page) {
272 folio_zero_segment(folio, firs !! 277 zero_user_segment(page, first_hole << blkbits, PAGE_SIZE);
273 if (first_hole == 0) { 278 if (first_hole == 0) {
274 folio_mark_uptodate(fo !! 279 SetPageUptodate(page);
275 folio_unlock(folio); !! 280 unlock_page(page);
276 goto out; 281 goto out;
277 } 282 }
278 } else if (fully_mapped) { 283 } else if (fully_mapped) {
279 folio_set_mappedtodisk(folio); !! 284 SetPageMappedToDisk(page);
>> 285 }
>> 286
>> 287 if (fully_mapped && blocks_per_page == 1 && !PageUptodate(page) &&
>> 288 cleancache_get_page(page) == 0) {
>> 289 SetPageUptodate(page);
>> 290 goto confused;
280 } 291 }
281 292
282 /* 293 /*
283 * This folio will go to BIO. Do we n !! 294 * This page will go to BIO. Do we need to send this BIO off first?
284 */ 295 */
285 if (args->bio && (args->last_block_in_ !! 296 if (args->bio && (args->last_block_in_bio != blocks[0] - 1))
286 args->bio = mpage_bio_submit_r !! 297 args->bio = mpage_bio_submit(REQ_OP_READ, op_flags, args->bio);
287 298
288 alloc_new: 299 alloc_new:
289 if (args->bio == NULL) { 300 if (args->bio == NULL) {
290 args->bio = bio_alloc(bdev, bi !! 301 if (first_hole == blocks_per_page) {
291 gfp); !! 302 if (!bdev_read_page(bdev, blocks[0] << (blkbits - 9),
>> 303 page))
>> 304 goto out;
>> 305 }
>> 306 args->bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
>> 307 min_t(int, args->nr_pages,
>> 308 BIO_MAX_PAGES),
>> 309 gfp);
292 if (args->bio == NULL) 310 if (args->bio == NULL)
293 goto confused; 311 goto confused;
294 args->bio->bi_iter.bi_sector = <<
295 } 312 }
296 313
297 length = first_hole << blkbits; 314 length = first_hole << blkbits;
298 if (!bio_add_folio(args->bio, folio, l !! 315 if (bio_add_page(args->bio, page, length, 0) < length) {
299 args->bio = mpage_bio_submit_r !! 316 args->bio = mpage_bio_submit(REQ_OP_READ, op_flags, args->bio);
300 goto alloc_new; 317 goto alloc_new;
301 } 318 }
302 319
303 relative_block = block_in_file - args- 320 relative_block = block_in_file - args->first_logical_block;
304 nblocks = map_bh->b_size >> blkbits; 321 nblocks = map_bh->b_size >> blkbits;
305 if ((buffer_boundary(map_bh) && relati 322 if ((buffer_boundary(map_bh) && relative_block == nblocks) ||
306 (first_hole != blocks_per_page)) 323 (first_hole != blocks_per_page))
307 args->bio = mpage_bio_submit_r !! 324 args->bio = mpage_bio_submit(REQ_OP_READ, op_flags, args->bio);
308 else 325 else
309 args->last_block_in_bio = firs !! 326 args->last_block_in_bio = blocks[blocks_per_page - 1];
310 out: 327 out:
311 return args->bio; 328 return args->bio;
312 329
313 confused: 330 confused:
314 if (args->bio) 331 if (args->bio)
315 args->bio = mpage_bio_submit_r !! 332 args->bio = mpage_bio_submit(REQ_OP_READ, op_flags, args->bio);
316 if (!folio_test_uptodate(folio)) !! 333 if (!PageUptodate(page))
317 block_read_full_folio(folio, a !! 334 block_read_full_page(page, args->get_block);
318 else 335 else
319 folio_unlock(folio); !! 336 unlock_page(page);
320 goto out; 337 goto out;
321 } 338 }
322 339
323 /** 340 /**
324 * mpage_readahead - start reads against pages !! 341 * mpage_readpages - populate an address space with some pages & start reads against them
325 * @rac: Describes which pages to read. !! 342 * @mapping: the address_space
>> 343 * @pages: The address of a list_head which contains the target pages. These
>> 344 * pages have their ->index populated and are otherwise uninitialised.
>> 345 * The page at @pages->prev has the lowest file offset, and reads should be
>> 346 * issued in @pages->prev to @pages->next order.
>> 347 * @nr_pages: The number of pages at *@pages
326 * @get_block: The filesystem's block mapper f 348 * @get_block: The filesystem's block mapper function.
327 * 349 *
328 * This function walks the pages and the block 350 * This function walks the pages and the blocks within each page, building and
329 * emitting large BIOs. 351 * emitting large BIOs.
330 * 352 *
331 * If anything unusual happens, such as: 353 * If anything unusual happens, such as:
332 * 354 *
333 * - encountering a page which has buffers 355 * - encountering a page which has buffers
334 * - encountering a page which has a non-hole 356 * - encountering a page which has a non-hole after a hole
335 * - encountering a page with non-contiguous b 357 * - encountering a page with non-contiguous blocks
336 * 358 *
337 * then this code just gives up and calls the 359 * then this code just gives up and calls the buffer_head-based read function.
338 * It does handle a page which has holes at th 360 * It does handle a page which has holes at the end - that is a common case:
339 * the end-of-file on blocksize < PAGE_SIZE se 361 * the end-of-file on blocksize < PAGE_SIZE setups.
340 * 362 *
341 * BH_Boundary explanation: 363 * BH_Boundary explanation:
342 * 364 *
343 * There is a problem. The mpage read code as 365 * There is a problem. The mpage read code assembles several pages, gets all
344 * their disk mappings, and then submits them 366 * their disk mappings, and then submits them all. That's fine, but obtaining
345 * the disk mappings may require I/O. Reads o 367 * the disk mappings may require I/O. Reads of indirect blocks, for example.
346 * 368 *
347 * So an mpage read of the first 16 blocks of 369 * So an mpage read of the first 16 blocks of an ext2 file will cause I/O to be
348 * submitted in the following order: 370 * submitted in the following order:
349 * 371 *
350 * 12 0 1 2 3 4 5 6 7 8 9 10 11 13 14 15 372 * 12 0 1 2 3 4 5 6 7 8 9 10 11 13 14 15 16
351 * 373 *
352 * because the indirect block has to be read t 374 * because the indirect block has to be read to get the mappings of blocks
353 * 13,14,15,16. Obviously, this impacts perfo 375 * 13,14,15,16. Obviously, this impacts performance.
354 * 376 *
355 * So what we do it to allow the filesystem's 377 * So what we do it to allow the filesystem's get_block() function to set
356 * BH_Boundary when it maps block 11. BH_Boun 378 * BH_Boundary when it maps block 11. BH_Boundary says: mapping of the block
357 * after this one will require I/O against a b 379 * after this one will require I/O against a block which is probably close to
358 * this one. So you should push what I/O you 380 * this one. So you should push what I/O you have currently accumulated.
359 * 381 *
360 * This all causes the disk requests to be iss 382 * This all causes the disk requests to be issued in the correct order.
361 */ 383 */
362 void mpage_readahead(struct readahead_control !! 384 int
>> 385 mpage_readpages(struct address_space *mapping, struct list_head *pages,
>> 386 unsigned nr_pages, get_block_t get_block)
363 { 387 {
364 struct folio *folio; <<
365 struct mpage_readpage_args args = { 388 struct mpage_readpage_args args = {
366 .get_block = get_block, 389 .get_block = get_block,
367 .is_readahead = true, 390 .is_readahead = true,
368 }; 391 };
>> 392 unsigned page_idx;
369 393
370 while ((folio = readahead_folio(rac))) !! 394 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
371 prefetchw(&folio->flags); !! 395 struct page *page = lru_to_page(pages);
372 args.folio = folio; !! 396
373 args.nr_pages = readahead_coun !! 397 prefetchw(&page->flags);
374 args.bio = do_mpage_readpage(& !! 398 list_del(&page->lru);
>> 399 if (!add_to_page_cache_lru(page, mapping,
>> 400 page->index,
>> 401 readahead_gfp_mask(mapping))) {
>> 402 args.page = page;
>> 403 args.nr_pages = nr_pages - page_idx;
>> 404 args.bio = do_mpage_readpage(&args);
>> 405 }
>> 406 put_page(page);
375 } 407 }
>> 408 BUG_ON(!list_empty(pages));
376 if (args.bio) 409 if (args.bio)
377 mpage_bio_submit_read(args.bio !! 410 mpage_bio_submit(REQ_OP_READ, REQ_RAHEAD, args.bio);
>> 411 return 0;
378 } 412 }
379 EXPORT_SYMBOL(mpage_readahead); !! 413 EXPORT_SYMBOL(mpage_readpages);
380 414
381 /* 415 /*
382 * This isn't called much at all 416 * This isn't called much at all
383 */ 417 */
384 int mpage_read_folio(struct folio *folio, get_ !! 418 int mpage_readpage(struct page *page, get_block_t get_block)
385 { 419 {
386 struct mpage_readpage_args args = { 420 struct mpage_readpage_args args = {
387 .folio = folio, !! 421 .page = page,
388 .nr_pages = 1, 422 .nr_pages = 1,
389 .get_block = get_block, 423 .get_block = get_block,
390 }; 424 };
391 425
392 args.bio = do_mpage_readpage(&args); 426 args.bio = do_mpage_readpage(&args);
393 if (args.bio) 427 if (args.bio)
394 mpage_bio_submit_read(args.bio !! 428 mpage_bio_submit(REQ_OP_READ, 0, args.bio);
395 return 0; 429 return 0;
396 } 430 }
397 EXPORT_SYMBOL(mpage_read_folio); !! 431 EXPORT_SYMBOL(mpage_readpage);
398 432
399 /* 433 /*
400 * Writing is not so simple. 434 * Writing is not so simple.
401 * 435 *
402 * If the page has buffers then they will be u 436 * If the page has buffers then they will be used for obtaining the disk
403 * mapping. We only support pages which are f 437 * mapping. We only support pages which are fully mapped-and-dirty, with a
404 * special case for pages which are unmapped a 438 * special case for pages which are unmapped at the end: end-of-file.
405 * 439 *
406 * If the page has no buffers (preferred) then 440 * If the page has no buffers (preferred) then the page is mapped here.
407 * 441 *
408 * If all blocks are found to be contiguous th 442 * If all blocks are found to be contiguous then the page can go into the
409 * BIO. Otherwise fall back to the mapping's 443 * BIO. Otherwise fall back to the mapping's writepage().
410 * 444 *
411 * FIXME: This code wants an estimate of how m 445 * FIXME: This code wants an estimate of how many pages are still to be
412 * written, so it can intelligently allocate a 446 * written, so it can intelligently allocate a suitably-sized BIO. For now,
413 * just allocate full-size (16-page) BIOs. 447 * just allocate full-size (16-page) BIOs.
414 */ 448 */
415 449
416 struct mpage_data { 450 struct mpage_data {
417 struct bio *bio; 451 struct bio *bio;
418 sector_t last_block_in_bio; 452 sector_t last_block_in_bio;
419 get_block_t *get_block; 453 get_block_t *get_block;
>> 454 unsigned use_writepage;
420 }; 455 };
421 456
422 /* 457 /*
423 * We have our BIO, so we can now mark the buf 458 * We have our BIO, so we can now mark the buffers clean. Make
424 * sure to only clean buffers which we know we 459 * sure to only clean buffers which we know we'll be writing.
425 */ 460 */
426 static void clean_buffers(struct folio *folio, !! 461 static void clean_buffers(struct page *page, unsigned first_unmapped)
427 { 462 {
428 unsigned buffer_counter = 0; 463 unsigned buffer_counter = 0;
429 struct buffer_head *bh, *head = folio_ !! 464 struct buffer_head *bh, *head;
430 !! 465 if (!page_has_buffers(page))
431 if (!head) <<
432 return; 466 return;
>> 467 head = page_buffers(page);
433 bh = head; 468 bh = head;
434 469
435 do { 470 do {
436 if (buffer_counter++ == first_ 471 if (buffer_counter++ == first_unmapped)
437 break; 472 break;
438 clear_buffer_dirty(bh); 473 clear_buffer_dirty(bh);
439 bh = bh->b_this_page; 474 bh = bh->b_this_page;
440 } while (bh != head); 475 } while (bh != head);
441 476
442 /* 477 /*
443 * we cannot drop the bh if the page i 478 * we cannot drop the bh if the page is not uptodate or a concurrent
444 * read_folio would fail to serialize !! 479 * readpage would fail to serialize with the bh and it would read from
445 * disk before we reach the platter. 480 * disk before we reach the platter.
446 */ 481 */
447 if (buffer_heads_over_limit && folio_t !! 482 if (buffer_heads_over_limit && PageUptodate(page))
448 try_to_free_buffers(folio); !! 483 try_to_free_buffers(page);
449 } 484 }
450 485
451 static int __mpage_writepage(struct folio *fol !! 486 /*
>> 487 * For situations where we want to clean all buffers attached to a page.
>> 488 * We don't need to calculate how many buffers are attached to the page,
>> 489 * we just need to specify a number larger than the maximum number of buffers.
>> 490 */
>> 491 void clean_page_buffers(struct page *page)
>> 492 {
>> 493 clean_buffers(page, ~0U);
>> 494 }
>> 495
>> 496 static int __mpage_writepage(struct page *page, struct writeback_control *wbc,
452 void *data) 497 void *data)
453 { 498 {
454 struct mpage_data *mpd = data; 499 struct mpage_data *mpd = data;
455 struct bio *bio = mpd->bio; 500 struct bio *bio = mpd->bio;
456 struct address_space *mapping = folio- !! 501 struct address_space *mapping = page->mapping;
457 struct inode *inode = mapping->host; !! 502 struct inode *inode = page->mapping->host;
458 const unsigned blkbits = inode->i_blkb 503 const unsigned blkbits = inode->i_blkbits;
>> 504 unsigned long end_index;
459 const unsigned blocks_per_page = PAGE_ 505 const unsigned blocks_per_page = PAGE_SIZE >> blkbits;
460 sector_t last_block; 506 sector_t last_block;
461 sector_t block_in_file; 507 sector_t block_in_file;
462 sector_t first_block; !! 508 sector_t blocks[MAX_BUF_PER_PAGE];
463 unsigned page_block; 509 unsigned page_block;
464 unsigned first_unmapped = blocks_per_p 510 unsigned first_unmapped = blocks_per_page;
465 struct block_device *bdev = NULL; 511 struct block_device *bdev = NULL;
466 int boundary = 0; 512 int boundary = 0;
467 sector_t boundary_block = 0; 513 sector_t boundary_block = 0;
468 struct block_device *boundary_bdev = N 514 struct block_device *boundary_bdev = NULL;
469 size_t length; !! 515 int length;
470 struct buffer_head map_bh; 516 struct buffer_head map_bh;
471 loff_t i_size = i_size_read(inode); 517 loff_t i_size = i_size_read(inode);
472 int ret = 0; 518 int ret = 0;
473 struct buffer_head *head = folio_buffe !! 519 int op_flags = wbc_to_write_flags(wbc);
474 520
475 if (head) { !! 521 if (page_has_buffers(page)) {
>> 522 struct buffer_head *head = page_buffers(page);
476 struct buffer_head *bh = head; 523 struct buffer_head *bh = head;
477 524
478 /* If they're all mapped and d 525 /* If they're all mapped and dirty, do it */
479 page_block = 0; 526 page_block = 0;
480 do { 527 do {
481 BUG_ON(buffer_locked(b 528 BUG_ON(buffer_locked(bh));
482 if (!buffer_mapped(bh) 529 if (!buffer_mapped(bh)) {
483 /* 530 /*
484 * unmapped di 531 * unmapped dirty buffers are created by
485 * block_dirty !! 532 * __set_page_dirty_buffers -> mmapped data
486 */ 533 */
487 if (buffer_dir 534 if (buffer_dirty(bh))
488 goto c 535 goto confused;
489 if (first_unma 536 if (first_unmapped == blocks_per_page)
490 first_ 537 first_unmapped = page_block;
491 continue; 538 continue;
492 } 539 }
493 540
494 if (first_unmapped != 541 if (first_unmapped != blocks_per_page)
495 goto confused; 542 goto confused; /* hole -> non-hole */
496 543
497 if (!buffer_dirty(bh) 544 if (!buffer_dirty(bh) || !buffer_uptodate(bh))
498 goto confused; 545 goto confused;
499 if (page_block) { 546 if (page_block) {
500 if (bh->b_bloc !! 547 if (bh->b_blocknr != blocks[page_block-1] + 1)
501 goto c 548 goto confused;
502 } else { <<
503 first_block = <<
504 } 549 }
505 page_block++; !! 550 blocks[page_block++] = bh->b_blocknr;
506 boundary = buffer_boun 551 boundary = buffer_boundary(bh);
507 if (boundary) { 552 if (boundary) {
508 boundary_block 553 boundary_block = bh->b_blocknr;
509 boundary_bdev 554 boundary_bdev = bh->b_bdev;
510 } 555 }
511 bdev = bh->b_bdev; 556 bdev = bh->b_bdev;
512 } while ((bh = bh->b_this_page 557 } while ((bh = bh->b_this_page) != head);
513 558
514 if (first_unmapped) 559 if (first_unmapped)
515 goto page_is_mapped; 560 goto page_is_mapped;
516 561
517 /* 562 /*
518 * Page has buffers, but they 563 * Page has buffers, but they are all unmapped. The page was
519 * created by pagein or read o 564 * created by pagein or read over a hole which was handled by
520 * block_read_full_folio(). I !! 565 * block_read_full_page(). If this address_space is also
521 * using mpage_readahead then !! 566 * using mpage_readpages then this can rarely happen.
522 */ 567 */
523 goto confused; 568 goto confused;
524 } 569 }
525 570
526 /* 571 /*
527 * The page has no buffers: map it to 572 * The page has no buffers: map it to disk
528 */ 573 */
529 BUG_ON(!folio_test_uptodate(folio)); !! 574 BUG_ON(!PageUptodate(page));
530 block_in_file = (sector_t)folio->index !! 575 block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits);
531 /* <<
532 * Whole page beyond EOF? Skip allocat <<
533 * space. <<
534 */ <<
535 if (block_in_file >= (i_size + (1 << b <<
536 goto page_is_mapped; <<
537 last_block = (i_size - 1) >> blkbits; 576 last_block = (i_size - 1) >> blkbits;
538 map_bh.b_folio = folio; !! 577 map_bh.b_page = page;
539 for (page_block = 0; page_block < bloc 578 for (page_block = 0; page_block < blocks_per_page; ) {
540 579
541 map_bh.b_state = 0; 580 map_bh.b_state = 0;
542 map_bh.b_size = 1 << blkbits; 581 map_bh.b_size = 1 << blkbits;
543 if (mpd->get_block(inode, bloc 582 if (mpd->get_block(inode, block_in_file, &map_bh, 1))
544 goto confused; 583 goto confused;
545 if (!buffer_mapped(&map_bh)) <<
546 goto confused; <<
547 if (buffer_new(&map_bh)) 584 if (buffer_new(&map_bh))
548 clean_bdev_bh_alias(&m 585 clean_bdev_bh_alias(&map_bh);
549 if (buffer_boundary(&map_bh)) 586 if (buffer_boundary(&map_bh)) {
550 boundary_block = map_b 587 boundary_block = map_bh.b_blocknr;
551 boundary_bdev = map_bh 588 boundary_bdev = map_bh.b_bdev;
552 } 589 }
553 if (page_block) { 590 if (page_block) {
554 if (map_bh.b_blocknr ! !! 591 if (map_bh.b_blocknr != blocks[page_block-1] + 1)
555 goto confused; 592 goto confused;
556 } else { <<
557 first_block = map_bh.b <<
558 } 593 }
559 page_block++; !! 594 blocks[page_block++] = map_bh.b_blocknr;
560 boundary = buffer_boundary(&ma 595 boundary = buffer_boundary(&map_bh);
561 bdev = map_bh.b_bdev; 596 bdev = map_bh.b_bdev;
562 if (block_in_file == last_bloc 597 if (block_in_file == last_block)
563 break; 598 break;
564 block_in_file++; 599 block_in_file++;
565 } 600 }
566 BUG_ON(page_block == 0); 601 BUG_ON(page_block == 0);
567 602
568 first_unmapped = page_block; 603 first_unmapped = page_block;
569 604
570 page_is_mapped: 605 page_is_mapped:
571 /* Don't bother writing beyond EOF, tr !! 606 end_index = i_size >> PAGE_SHIFT;
572 if (folio_pos(folio) >= i_size) !! 607 if (page->index >= end_index) {
573 goto confused; <<
574 length = folio_size(folio); <<
575 if (folio_pos(folio) + length > i_size <<
576 /* 608 /*
577 * The page straddles i_size. 609 * The page straddles i_size. It must be zeroed out on each
578 * and every writepage invocat 610 * and every writepage invocation because it may be mmapped.
579 * "A file is mapped in multip 611 * "A file is mapped in multiples of the page size. For a file
580 * that is not a multiple of t 612 * that is not a multiple of the page size, the remaining memory
581 * is zeroed when mapped, and 613 * is zeroed when mapped, and writes to that region are not
582 * written out to the file." 614 * written out to the file."
583 */ 615 */
584 length = i_size - folio_pos(fo !! 616 unsigned offset = i_size & (PAGE_SIZE - 1);
585 folio_zero_segment(folio, leng !! 617
>> 618 if (page->index > end_index || !offset)
>> 619 goto confused;
>> 620 zero_user_segment(page, offset, PAGE_SIZE);
586 } 621 }
587 622
588 /* 623 /*
589 * This page will go to BIO. Do we ne 624 * This page will go to BIO. Do we need to send this BIO off first?
590 */ 625 */
591 if (bio && mpd->last_block_in_bio != f !! 626 if (bio && mpd->last_block_in_bio != blocks[0] - 1)
592 bio = mpage_bio_submit_write(b !! 627 bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio);
593 628
594 alloc_new: 629 alloc_new:
595 if (bio == NULL) { 630 if (bio == NULL) {
596 bio = bio_alloc(bdev, BIO_MAX_ !! 631 if (first_unmapped == blocks_per_page) {
597 REQ_OP_WRITE | !! 632 if (!bdev_write_page(bdev, blocks[0] << (blkbits - 9),
598 GFP_NOFS); !! 633 page, wbc))
599 bio->bi_iter.bi_sector = first !! 634 goto out;
>> 635 }
>> 636 bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
>> 637 BIO_MAX_PAGES, GFP_NOFS|__GFP_HIGH);
>> 638 if (bio == NULL)
>> 639 goto confused;
>> 640
600 wbc_init_bio(wbc, bio); 641 wbc_init_bio(wbc, bio);
601 bio->bi_write_hint = inode->i_ 642 bio->bi_write_hint = inode->i_write_hint;
602 } 643 }
603 644
604 /* 645 /*
605 * Must try to add the page before mar 646 * Must try to add the page before marking the buffer clean or
606 * the confused fail path above (OOM) 647 * the confused fail path above (OOM) will be very confused when
607 * it finds all bh marked clean (i.e. 648 * it finds all bh marked clean (i.e. it will not write anything)
608 */ 649 */
609 wbc_account_cgroup_owner(wbc, &folio-> !! 650 wbc_account_io(wbc, page, PAGE_SIZE);
610 length = first_unmapped << blkbits; 651 length = first_unmapped << blkbits;
611 if (!bio_add_folio(bio, folio, length, !! 652 if (bio_add_page(bio, page, length, 0) < length) {
612 bio = mpage_bio_submit_write(b !! 653 bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio);
613 goto alloc_new; 654 goto alloc_new;
614 } 655 }
615 656
616 clean_buffers(folio, first_unmapped); !! 657 clean_buffers(page, first_unmapped);
617 658
618 BUG_ON(folio_test_writeback(folio)); !! 659 BUG_ON(PageWriteback(page));
619 folio_start_writeback(folio); !! 660 set_page_writeback(page);
620 folio_unlock(folio); !! 661 unlock_page(page);
621 if (boundary || (first_unmapped != blo 662 if (boundary || (first_unmapped != blocks_per_page)) {
622 bio = mpage_bio_submit_write(b !! 663 bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio);
623 if (boundary_block) { 664 if (boundary_block) {
624 write_boundary_block(b 665 write_boundary_block(boundary_bdev,
625 bounda 666 boundary_block, 1 << blkbits);
626 } 667 }
627 } else { 668 } else {
628 mpd->last_block_in_bio = first !! 669 mpd->last_block_in_bio = blocks[blocks_per_page - 1];
629 } 670 }
630 goto out; 671 goto out;
631 672
632 confused: 673 confused:
633 if (bio) 674 if (bio)
634 bio = mpage_bio_submit_write(b !! 675 bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio);
635 676
>> 677 if (mpd->use_writepage) {
>> 678 ret = mapping->a_ops->writepage(page, wbc);
>> 679 } else {
>> 680 ret = -EAGAIN;
>> 681 goto out;
>> 682 }
636 /* 683 /*
637 * The caller has a ref on the inode, 684 * The caller has a ref on the inode, so *mapping is stable
638 */ 685 */
639 ret = block_write_full_folio(folio, wb <<
640 mapping_set_error(mapping, ret); 686 mapping_set_error(mapping, ret);
641 out: 687 out:
642 mpd->bio = bio; 688 mpd->bio = bio;
643 return ret; 689 return ret;
644 } 690 }
645 691
646 /** 692 /**
647 * mpage_writepages - walk the list of dirty p 693 * mpage_writepages - walk the list of dirty pages of the given address space & writepage() all of them
648 * @mapping: address space structure to write 694 * @mapping: address space structure to write
649 * @wbc: subtract the number of written pages 695 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
650 * @get_block: the filesystem's block mapper f 696 * @get_block: the filesystem's block mapper function.
>> 697 * If this is NULL then use a_ops->writepage. Otherwise, go
>> 698 * direct-to-BIO.
651 * 699 *
652 * This is a library function, which implement 700 * This is a library function, which implements the writepages()
653 * address_space_operation. 701 * address_space_operation.
>> 702 *
>> 703 * If a page is already under I/O, generic_writepages() skips it, even
>> 704 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
>> 705 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
>> 706 * and msync() need to guarantee that all the data which was dirty at the time
>> 707 * the call was made get new I/O started against them. If wbc->sync_mode is
>> 708 * WB_SYNC_ALL then we were called for data integrity and we must wait for
>> 709 * existing IO to complete.
654 */ 710 */
655 int 711 int
656 mpage_writepages(struct address_space *mapping 712 mpage_writepages(struct address_space *mapping,
657 struct writeback_control *wbc, 713 struct writeback_control *wbc, get_block_t get_block)
658 { 714 {
659 struct mpage_data mpd = { <<
660 .get_block = get_block, <<
661 }; <<
662 struct blk_plug plug; 715 struct blk_plug plug;
663 int ret; 716 int ret;
664 717
665 blk_start_plug(&plug); 718 blk_start_plug(&plug);
666 ret = write_cache_pages(mapping, wbc, !! 719
667 if (mpd.bio) !! 720 if (!get_block)
668 mpage_bio_submit_write(mpd.bio !! 721 ret = generic_writepages(mapping, wbc);
>> 722 else {
>> 723 struct mpage_data mpd = {
>> 724 .bio = NULL,
>> 725 .last_block_in_bio = 0,
>> 726 .get_block = get_block,
>> 727 .use_writepage = 1,
>> 728 };
>> 729
>> 730 ret = write_cache_pages(mapping, wbc, __mpage_writepage, &mpd);
>> 731 if (mpd.bio) {
>> 732 int op_flags = (wbc->sync_mode == WB_SYNC_ALL ?
>> 733 REQ_SYNC : 0);
>> 734 mpage_bio_submit(REQ_OP_WRITE, op_flags, mpd.bio);
>> 735 }
>> 736 }
669 blk_finish_plug(&plug); 737 blk_finish_plug(&plug);
670 return ret; 738 return ret;
671 } 739 }
672 EXPORT_SYMBOL(mpage_writepages); 740 EXPORT_SYMBOL(mpage_writepages);
>> 741
>> 742 int mpage_writepage(struct page *page, get_block_t get_block,
>> 743 struct writeback_control *wbc)
>> 744 {
>> 745 struct mpage_data mpd = {
>> 746 .bio = NULL,
>> 747 .last_block_in_bio = 0,
>> 748 .get_block = get_block,
>> 749 .use_writepage = 0,
>> 750 };
>> 751 int ret = __mpage_writepage(page, wbc, &mpd);
>> 752 if (mpd.bio) {
>> 753 int op_flags = (wbc->sync_mode == WB_SYNC_ALL ?
>> 754 REQ_SYNC : 0);
>> 755 mpage_bio_submit(REQ_OP_WRITE, op_flags, mpd.bio);
>> 756 }
>> 757 return ret;
>> 758 }
>> 759 EXPORT_SYMBOL(mpage_writepage);
673 760
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.