xref: /openbmc/linux/fs/ext4/readpage.c (revision f9834f18)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * linux/fs/ext4/readpage.c
4  *
5  * Copyright (C) 2002, Linus Torvalds.
6  * Copyright (C) 2015, Google, Inc.
7  *
8  * This was originally taken from fs/mpage.c
9  *
10  * The intent is the ext4_mpage_readpages() function here is intended
11  * to replace mpage_readpages() in the general case, not just for
12  * encrypted files.  It has some limitations (see below), where it
13  * will fall back to read_block_full_page(), but these limitations
14  * should only be hit when page_size != block_size.
15  *
16  * This will allow us to attach a callback function to support ext4
17  * encryption.
18  *
19  * If anything unusual happens, such as:
20  *
21  * - encountering a page which has buffers
22  * - encountering a page which has a non-hole after a hole
23  * - encountering a page with non-contiguous blocks
24  *
25  * then this code just gives up and calls the buffer_head-based read function.
26  * It does handle a page which has holes at the end - that is a common case:
27  * the end-of-file on blocksize < PAGE_SIZE setups.
28  *
29  */
30 
31 #include <linux/kernel.h>
32 #include <linux/export.h>
33 #include <linux/mm.h>
34 #include <linux/kdev_t.h>
35 #include <linux/gfp.h>
36 #include <linux/bio.h>
37 #include <linux/fs.h>
38 #include <linux/buffer_head.h>
39 #include <linux/blkdev.h>
40 #include <linux/highmem.h>
41 #include <linux/prefetch.h>
42 #include <linux/mpage.h>
43 #include <linux/writeback.h>
44 #include <linux/backing-dev.h>
45 #include <linux/pagevec.h>
46 #include <linux/cleancache.h>
47 
48 #include "ext4.h"
49 
50 #define NUM_PREALLOC_POST_READ_CTXS	128
51 
52 static struct kmem_cache *bio_post_read_ctx_cache;
53 static mempool_t *bio_post_read_ctx_pool;
54 
55 /* postprocessing steps for read bios */
56 enum bio_post_read_step {
57 	STEP_INITIAL = 0,
58 	STEP_DECRYPT,
59 	STEP_VERITY,
60 	STEP_MAX,
61 };
62 
63 struct bio_post_read_ctx {
64 	struct bio *bio;
65 	struct work_struct work;
66 	unsigned int cur_step;
67 	unsigned int enabled_steps;
68 };
69 
70 static void __read_end_io(struct bio *bio)
71 {
72 	struct page *page;
73 	struct bio_vec *bv;
74 	struct bvec_iter_all iter_all;
75 
76 	bio_for_each_segment_all(bv, bio, iter_all) {
77 		page = bv->bv_page;
78 
79 		/* PG_error was set if any post_read step failed */
80 		if (bio->bi_status || PageError(page)) {
81 			ClearPageUptodate(page);
82 			/* will re-read again later */
83 			ClearPageError(page);
84 		} else {
85 			SetPageUptodate(page);
86 		}
87 		unlock_page(page);
88 	}
89 	if (bio->bi_private)
90 		mempool_free(bio->bi_private, bio_post_read_ctx_pool);
91 	bio_put(bio);
92 }
93 
94 static void bio_post_read_processing(struct bio_post_read_ctx *ctx);
95 
96 static void decrypt_work(struct work_struct *work)
97 {
98 	struct bio_post_read_ctx *ctx =
99 		container_of(work, struct bio_post_read_ctx, work);
100 
101 	fscrypt_decrypt_bio(ctx->bio);
102 
103 	bio_post_read_processing(ctx);
104 }
105 
106 static void verity_work(struct work_struct *work)
107 {
108 	struct bio_post_read_ctx *ctx =
109 		container_of(work, struct bio_post_read_ctx, work);
110 	struct bio *bio = ctx->bio;
111 
112 	/*
113 	 * fsverity_verify_bio() may call readpages() again, and although verity
114 	 * will be disabled for that, decryption may still be needed, causing
115 	 * another bio_post_read_ctx to be allocated.  So to guarantee that
116 	 * mempool_alloc() never deadlocks we must free the current ctx first.
117 	 * This is safe because verity is the last post-read step.
118 	 */
119 	BUILD_BUG_ON(STEP_VERITY + 1 != STEP_MAX);
120 	mempool_free(ctx, bio_post_read_ctx_pool);
121 	bio->bi_private = NULL;
122 
123 	fsverity_verify_bio(bio);
124 
125 	__read_end_io(bio);
126 }
127 
128 static void bio_post_read_processing(struct bio_post_read_ctx *ctx)
129 {
130 	/*
131 	 * We use different work queues for decryption and for verity because
132 	 * verity may require reading metadata pages that need decryption, and
133 	 * we shouldn't recurse to the same workqueue.
134 	 */
135 	switch (++ctx->cur_step) {
136 	case STEP_DECRYPT:
137 		if (ctx->enabled_steps & (1 << STEP_DECRYPT)) {
138 			INIT_WORK(&ctx->work, decrypt_work);
139 			fscrypt_enqueue_decrypt_work(&ctx->work);
140 			return;
141 		}
142 		ctx->cur_step++;
143 		/* fall-through */
144 	case STEP_VERITY:
145 		if (ctx->enabled_steps & (1 << STEP_VERITY)) {
146 			INIT_WORK(&ctx->work, verity_work);
147 			fsverity_enqueue_verify_work(&ctx->work);
148 			return;
149 		}
150 		ctx->cur_step++;
151 		/* fall-through */
152 	default:
153 		__read_end_io(ctx->bio);
154 	}
155 }
156 
157 static bool bio_post_read_required(struct bio *bio)
158 {
159 	return bio->bi_private && !bio->bi_status;
160 }
161 
162 /*
163  * I/O completion handler for multipage BIOs.
164  *
165  * The mpage code never puts partial pages into a BIO (except for end-of-file).
166  * If a page does not map to a contiguous run of blocks then it simply falls
167  * back to block_read_full_page().
168  *
169  * Why is this?  If a page's completion depends on a number of different BIOs
170  * which can complete in any order (or at the same time) then determining the
171  * status of that page is hard.  See end_buffer_async_read() for the details.
172  * There is no point in duplicating all that complexity.
173  */
174 static void mpage_end_io(struct bio *bio)
175 {
176 	if (bio_post_read_required(bio)) {
177 		struct bio_post_read_ctx *ctx = bio->bi_private;
178 
179 		ctx->cur_step = STEP_INITIAL;
180 		bio_post_read_processing(ctx);
181 		return;
182 	}
183 	__read_end_io(bio);
184 }
185 
186 static inline bool ext4_need_verity(const struct inode *inode, pgoff_t idx)
187 {
188 	return fsverity_active(inode) &&
189 	       idx < DIV_ROUND_UP(inode->i_size, PAGE_SIZE);
190 }
191 
192 static void ext4_set_bio_post_read_ctx(struct bio *bio,
193 				       const struct inode *inode,
194 				       pgoff_t first_idx)
195 {
196 	unsigned int post_read_steps = 0;
197 
198 	if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode))
199 		post_read_steps |= 1 << STEP_DECRYPT;
200 
201 	if (ext4_need_verity(inode, first_idx))
202 		post_read_steps |= 1 << STEP_VERITY;
203 
204 	if (post_read_steps) {
205 		/* Due to the mempool, this never fails. */
206 		struct bio_post_read_ctx *ctx =
207 			mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
208 
209 		ctx->bio = bio;
210 		ctx->enabled_steps = post_read_steps;
211 		bio->bi_private = ctx;
212 	}
213 }
214 
215 static inline loff_t ext4_readpage_limit(struct inode *inode)
216 {
217 	if (IS_ENABLED(CONFIG_FS_VERITY) &&
218 	    (IS_VERITY(inode) || ext4_verity_in_progress(inode)))
219 		return inode->i_sb->s_maxbytes;
220 
221 	return i_size_read(inode);
222 }
223 
224 int ext4_mpage_readpages(struct address_space *mapping,
225 			 struct list_head *pages, struct page *page,
226 			 unsigned nr_pages, bool is_readahead)
227 {
228 	struct bio *bio = NULL;
229 	sector_t last_block_in_bio = 0;
230 
231 	struct inode *inode = mapping->host;
232 	const unsigned blkbits = inode->i_blkbits;
233 	const unsigned blocks_per_page = PAGE_SIZE >> blkbits;
234 	const unsigned blocksize = 1 << blkbits;
235 	sector_t block_in_file;
236 	sector_t last_block;
237 	sector_t last_block_in_file;
238 	sector_t blocks[MAX_BUF_PER_PAGE];
239 	unsigned page_block;
240 	struct block_device *bdev = inode->i_sb->s_bdev;
241 	int length;
242 	unsigned relative_block = 0;
243 	struct ext4_map_blocks map;
244 
245 	map.m_pblk = 0;
246 	map.m_lblk = 0;
247 	map.m_len = 0;
248 	map.m_flags = 0;
249 
250 	for (; nr_pages; nr_pages--) {
251 		int fully_mapped = 1;
252 		unsigned first_hole = blocks_per_page;
253 
254 		if (pages) {
255 			page = lru_to_page(pages);
256 
257 			prefetchw(&page->flags);
258 			list_del(&page->lru);
259 			if (add_to_page_cache_lru(page, mapping, page->index,
260 				  readahead_gfp_mask(mapping)))
261 				goto next_page;
262 		}
263 
264 		if (page_has_buffers(page))
265 			goto confused;
266 
267 		block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits);
268 		last_block = block_in_file + nr_pages * blocks_per_page;
269 		last_block_in_file = (ext4_readpage_limit(inode) +
270 				      blocksize - 1) >> blkbits;
271 		if (last_block > last_block_in_file)
272 			last_block = last_block_in_file;
273 		page_block = 0;
274 
275 		/*
276 		 * Map blocks using the previous result first.
277 		 */
278 		if ((map.m_flags & EXT4_MAP_MAPPED) &&
279 		    block_in_file > map.m_lblk &&
280 		    block_in_file < (map.m_lblk + map.m_len)) {
281 			unsigned map_offset = block_in_file - map.m_lblk;
282 			unsigned last = map.m_len - map_offset;
283 
284 			for (relative_block = 0; ; relative_block++) {
285 				if (relative_block == last) {
286 					/* needed? */
287 					map.m_flags &= ~EXT4_MAP_MAPPED;
288 					break;
289 				}
290 				if (page_block == blocks_per_page)
291 					break;
292 				blocks[page_block] = map.m_pblk + map_offset +
293 					relative_block;
294 				page_block++;
295 				block_in_file++;
296 			}
297 		}
298 
299 		/*
300 		 * Then do more ext4_map_blocks() calls until we are
301 		 * done with this page.
302 		 */
303 		while (page_block < blocks_per_page) {
304 			if (block_in_file < last_block) {
305 				map.m_lblk = block_in_file;
306 				map.m_len = last_block - block_in_file;
307 
308 				if (ext4_map_blocks(NULL, inode, &map, 0) < 0) {
309 				set_error_page:
310 					SetPageError(page);
311 					zero_user_segment(page, 0,
312 							  PAGE_SIZE);
313 					unlock_page(page);
314 					goto next_page;
315 				}
316 			}
317 			if ((map.m_flags & EXT4_MAP_MAPPED) == 0) {
318 				fully_mapped = 0;
319 				if (first_hole == blocks_per_page)
320 					first_hole = page_block;
321 				page_block++;
322 				block_in_file++;
323 				continue;
324 			}
325 			if (first_hole != blocks_per_page)
326 				goto confused;		/* hole -> non-hole */
327 
328 			/* Contiguous blocks? */
329 			if (page_block && blocks[page_block-1] != map.m_pblk-1)
330 				goto confused;
331 			for (relative_block = 0; ; relative_block++) {
332 				if (relative_block == map.m_len) {
333 					/* needed? */
334 					map.m_flags &= ~EXT4_MAP_MAPPED;
335 					break;
336 				} else if (page_block == blocks_per_page)
337 					break;
338 				blocks[page_block] = map.m_pblk+relative_block;
339 				page_block++;
340 				block_in_file++;
341 			}
342 		}
343 		if (first_hole != blocks_per_page) {
344 			zero_user_segment(page, first_hole << blkbits,
345 					  PAGE_SIZE);
346 			if (first_hole == 0) {
347 				if (ext4_need_verity(inode, page->index) &&
348 				    !fsverity_verify_page(page))
349 					goto set_error_page;
350 				SetPageUptodate(page);
351 				unlock_page(page);
352 				goto next_page;
353 			}
354 		} else if (fully_mapped) {
355 			SetPageMappedToDisk(page);
356 		}
357 		if (fully_mapped && blocks_per_page == 1 &&
358 		    !PageUptodate(page) && cleancache_get_page(page) == 0) {
359 			SetPageUptodate(page);
360 			goto confused;
361 		}
362 
363 		/*
364 		 * This page will go to BIO.  Do we need to send this
365 		 * BIO off first?
366 		 */
367 		if (bio && (last_block_in_bio != blocks[0] - 1)) {
368 		submit_and_realloc:
369 			submit_bio(bio);
370 			bio = NULL;
371 		}
372 		if (bio == NULL) {
373 			/*
374 			 * bio_alloc will _always_ be able to allocate a bio if
375 			 * __GFP_DIRECT_RECLAIM is set, see bio_alloc_bioset().
376 			 */
377 			bio = bio_alloc(GFP_KERNEL,
378 				min_t(int, nr_pages, BIO_MAX_PAGES));
379 			ext4_set_bio_post_read_ctx(bio, inode, page->index);
380 			bio_set_dev(bio, bdev);
381 			bio->bi_iter.bi_sector = blocks[0] << (blkbits - 9);
382 			bio->bi_end_io = mpage_end_io;
383 			bio_set_op_attrs(bio, REQ_OP_READ,
384 						is_readahead ? REQ_RAHEAD : 0);
385 		}
386 
387 		length = first_hole << blkbits;
388 		if (bio_add_page(bio, page, length, 0) < length)
389 			goto submit_and_realloc;
390 
391 		if (((map.m_flags & EXT4_MAP_BOUNDARY) &&
392 		     (relative_block == map.m_len)) ||
393 		    (first_hole != blocks_per_page)) {
394 			submit_bio(bio);
395 			bio = NULL;
396 		} else
397 			last_block_in_bio = blocks[blocks_per_page - 1];
398 		goto next_page;
399 	confused:
400 		if (bio) {
401 			submit_bio(bio);
402 			bio = NULL;
403 		}
404 		if (!PageUptodate(page))
405 			block_read_full_page(page, ext4_get_block);
406 		else
407 			unlock_page(page);
408 	next_page:
409 		if (pages)
410 			put_page(page);
411 	}
412 	BUG_ON(pages && !list_empty(pages));
413 	if (bio)
414 		submit_bio(bio);
415 	return 0;
416 }
417 
418 int __init ext4_init_post_read_processing(void)
419 {
420 	bio_post_read_ctx_cache =
421 		kmem_cache_create("ext4_bio_post_read_ctx",
422 				  sizeof(struct bio_post_read_ctx), 0, 0, NULL);
423 	if (!bio_post_read_ctx_cache)
424 		goto fail;
425 	bio_post_read_ctx_pool =
426 		mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
427 					 bio_post_read_ctx_cache);
428 	if (!bio_post_read_ctx_pool)
429 		goto fail_free_cache;
430 	return 0;
431 
432 fail_free_cache:
433 	kmem_cache_destroy(bio_post_read_ctx_cache);
434 fail:
435 	return -ENOMEM;
436 }
437 
438 void ext4_exit_post_read_processing(void)
439 {
440 	mempool_destroy(bio_post_read_ctx_pool);
441 	kmem_cache_destroy(bio_post_read_ctx_cache);
442 }
443