readpage.c

  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 ext4_mpage_readpages() function here is intended to
 11 * replace mpage_readahead() 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
 47#include "ext4.h"
 48
 49#define NUM_PREALLOC_POST_READ_CTXS	128
 50
 51static struct kmem_cache *bio_post_read_ctx_cache;
 52static mempool_t *bio_post_read_ctx_pool;
 53
 54/* postprocessing steps for read bios */
 55enum bio_post_read_step {
 56	STEP_INITIAL = 0,
 57	STEP_DECRYPT,
 58	STEP_VERITY,
 59	STEP_MAX,
 60};
 61
 62struct bio_post_read_ctx {
 63	struct bio *bio;
 64	struct work_struct work;
 65	unsigned int cur_step;
 66	unsigned int enabled_steps;
 67};
 68
 69static void __read_end_io(struct bio *bio)
 70{
 71	struct folio_iter fi;
 72
 73	bio_for_each_folio_all(fi, bio)
 74		folio_end_read(fi.folio, bio->bi_status == 0);
 75	if (bio->bi_private)
 76		mempool_free(bio->bi_private, bio_post_read_ctx_pool);
 77	bio_put(bio);
 78}
 79
 80static void bio_post_read_processing(struct bio_post_read_ctx *ctx);
 81
 82static void decrypt_work(struct work_struct *work)
 83{
 84	struct bio_post_read_ctx *ctx =
 85		container_of(work, struct bio_post_read_ctx, work);
 86	struct bio *bio = ctx->bio;
 87
 88	if (fscrypt_decrypt_bio(bio))
 89		bio_post_read_processing(ctx);
 90	else
 91		__read_end_io(bio);
 92}
 93
 94static void verity_work(struct work_struct *work)
 95{
 96	struct bio_post_read_ctx *ctx =
 97		container_of(work, struct bio_post_read_ctx, work);
 98	struct bio *bio = ctx->bio;
 99
100	/*
101	 * fsverity_verify_bio() may call readahead() again, and although verity
102	 * will be disabled for that, decryption may still be needed, causing
103	 * another bio_post_read_ctx to be allocated.  So to guarantee that
104	 * mempool_alloc() never deadlocks we must free the current ctx first.
105	 * This is safe because verity is the last post-read step.
106	 */
107	BUILD_BUG_ON(STEP_VERITY + 1 != STEP_MAX);
108	mempool_free(ctx, bio_post_read_ctx_pool);
109	bio->bi_private = NULL;
110
111	fsverity_verify_bio(bio);
112
113	__read_end_io(bio);
114}
115
116static void bio_post_read_processing(struct bio_post_read_ctx *ctx)
117{
118	/*
119	 * We use different work queues for decryption and for verity because
120	 * verity may require reading metadata pages that need decryption, and
121	 * we shouldn't recurse to the same workqueue.
122	 */
123	switch (++ctx->cur_step) {
124	case STEP_DECRYPT:
125		if (ctx->enabled_steps & (1 << STEP_DECRYPT)) {
126			INIT_WORK(&ctx->work, decrypt_work);
127			fscrypt_enqueue_decrypt_work(&ctx->work);
128			return;
129		}
130		ctx->cur_step++;
131		fallthrough;
132	case STEP_VERITY:
133		if (ctx->enabled_steps & (1 << STEP_VERITY)) {
134			INIT_WORK(&ctx->work, verity_work);
135			fsverity_enqueue_verify_work(&ctx->work);
136			return;
137		}
138		ctx->cur_step++;
139		fallthrough;
140	default:
141		__read_end_io(ctx->bio);
142	}
143}
144
145static bool bio_post_read_required(struct bio *bio)
146{
147	return bio->bi_private && !bio->bi_status;
148}
149
150/*
151 * I/O completion handler for multipage BIOs.
152 *
153 * The mpage code never puts partial pages into a BIO (except for end-of-file).
154 * If a page does not map to a contiguous run of blocks then it simply falls
155 * back to block_read_full_folio().
156 *
157 * Why is this?  If a page's completion depends on a number of different BIOs
158 * which can complete in any order (or at the same time) then determining the
159 * status of that page is hard.  See end_buffer_async_read() for the details.
160 * There is no point in duplicating all that complexity.
161 */
162static void mpage_end_io(struct bio *bio)
163{
164	if (bio_post_read_required(bio)) {
165		struct bio_post_read_ctx *ctx = bio->bi_private;
166
167		ctx->cur_step = STEP_INITIAL;
168		bio_post_read_processing(ctx);
169		return;
170	}
171	__read_end_io(bio);
172}
173
174static inline bool ext4_need_verity(const struct inode *inode, pgoff_t idx)
175{
176	return fsverity_active(inode) &&
177	       idx < DIV_ROUND_UP(inode->i_size, PAGE_SIZE);
178}
179
180static void ext4_set_bio_post_read_ctx(struct bio *bio,
181				       const struct inode *inode,
182				       pgoff_t first_idx)
183{
184	unsigned int post_read_steps = 0;
185
186	if (fscrypt_inode_uses_fs_layer_crypto(inode))
187		post_read_steps |= 1 << STEP_DECRYPT;
188
189	if (ext4_need_verity(inode, first_idx))
190		post_read_steps |= 1 << STEP_VERITY;
191
192	if (post_read_steps) {
193		/* Due to the mempool, this never fails. */
194		struct bio_post_read_ctx *ctx =
195			mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
196
197		ctx->bio = bio;
198		ctx->enabled_steps = post_read_steps;
199		bio->bi_private = ctx;
200	}
201}
202
203static inline loff_t ext4_readpage_limit(struct inode *inode)
204{
205	if (IS_ENABLED(CONFIG_FS_VERITY) && IS_VERITY(inode))
206		return inode->i_sb->s_maxbytes;
207
208	return i_size_read(inode);
209}
210
211int ext4_mpage_readpages(struct inode *inode,
212		struct readahead_control *rac, struct folio *folio)
213{
214	struct bio *bio = NULL;
215	sector_t last_block_in_bio = 0;
216	const unsigned blkbits = inode->i_blkbits;
217	const unsigned blocksize = 1 << blkbits;
218	sector_t next_block;
219	sector_t block_in_file;
220	sector_t last_block;
221	sector_t last_block_in_file;
222	sector_t first_block;
223	unsigned page_block;
224	struct block_device *bdev = inode->i_sb->s_bdev;
225	int length;
226	unsigned relative_block = 0;
227	struct ext4_map_blocks map;
228	unsigned int nr_pages, folio_pages;
229
230	map.m_pblk = 0;
231	map.m_lblk = 0;
232	map.m_len = 0;
233	map.m_flags = 0;
234
235	nr_pages = rac ? readahead_count(rac) : folio_nr_pages(folio);
236	for (; nr_pages; nr_pages -= folio_pages) {
237		int fully_mapped = 1;
238		unsigned int first_hole;
239		unsigned int blocks_per_folio;
240
241		if (rac)
242			folio = readahead_folio(rac);
243
244		folio_pages = folio_nr_pages(folio);
245		prefetchw(&folio->flags);
246
247		if (folio_buffers(folio))
248			goto confused;
249
250		blocks_per_folio = folio_size(folio) >> blkbits;
251		first_hole = blocks_per_folio;
252		block_in_file = next_block = EXT4_PG_TO_LBLK(inode, folio->index);
253		last_block = EXT4_PG_TO_LBLK(inode, folio->index + nr_pages);
254		last_block_in_file = (ext4_readpage_limit(inode) +
255				      blocksize - 1) >> blkbits;
256		if (last_block > last_block_in_file)
257			last_block = last_block_in_file;
258		page_block = 0;
259
260		/*
261		 * Map blocks using the previous result first.
262		 */
263		if ((map.m_flags & EXT4_MAP_MAPPED) &&
264		    block_in_file > map.m_lblk &&
265		    block_in_file < (map.m_lblk + map.m_len)) {
266			unsigned map_offset = block_in_file - map.m_lblk;
267			unsigned last = map.m_len - map_offset;
268
269			first_block = map.m_pblk + map_offset;
270			for (relative_block = 0; ; relative_block++) {
271				if (relative_block == last) {
272					/* needed? */
273					map.m_flags &= ~EXT4_MAP_MAPPED;
274					break;
275				}
276				if (page_block == blocks_per_folio)
277					break;
278				page_block++;
279				block_in_file++;
280			}
281		}
282
283		/*
284		 * Then do more ext4_map_blocks() calls until we are
285		 * done with this folio.
286		 */
287		while (page_block < blocks_per_folio) {
288			if (block_in_file < last_block) {
289				map.m_lblk = block_in_file;
290				map.m_len = last_block - block_in_file;
291
292				if (ext4_map_blocks(NULL, inode, &map, 0) < 0) {
293				set_error_page:
294					folio_zero_segment(folio, 0,
295							  folio_size(folio));
296					folio_unlock(folio);
297					goto next_page;
298				}
299			}
300			if ((map.m_flags & EXT4_MAP_MAPPED) == 0) {
301				fully_mapped = 0;
302				if (first_hole == blocks_per_folio)
303					first_hole = page_block;
304				page_block++;
305				block_in_file++;
306				continue;
307			}
308			if (first_hole != blocks_per_folio)
309				goto confused;		/* hole -> non-hole */
310
311			/* Contiguous blocks? */
312			if (!page_block)
313				first_block = map.m_pblk;
314			else if (first_block + page_block != map.m_pblk)
315				goto confused;
316			for (relative_block = 0; ; relative_block++) {
317				if (relative_block == map.m_len) {
318					/* needed? */
319					map.m_flags &= ~EXT4_MAP_MAPPED;
320					break;
321				} else if (page_block == blocks_per_folio)
322					break;
323				page_block++;
324				block_in_file++;
325			}
326		}
327		if (first_hole != blocks_per_folio) {
328			folio_zero_segment(folio, first_hole << blkbits,
329					  folio_size(folio));
330			if (first_hole == 0) {
331				if (ext4_need_verity(inode, folio->index) &&
332				    !fsverity_verify_folio(folio))
333					goto set_error_page;
334				folio_end_read(folio, true);
335				continue;
336			}
337		} else if (fully_mapped) {
338			folio_set_mappedtodisk(folio);
339		}
340
341		/*
342		 * This folio will go to BIO.  Do we need to send this
343		 * BIO off first?
344		 */
345		if (bio && (last_block_in_bio != first_block - 1 ||
346			    !fscrypt_mergeable_bio(bio, inode, next_block))) {
347		submit_and_realloc:
348			submit_bio(bio);
349			bio = NULL;
350		}
351		if (bio == NULL) {
352			/*
353			 * bio_alloc will _always_ be able to allocate a bio if
354			 * __GFP_DIRECT_RECLAIM is set, see bio_alloc_bioset().
355			 */
356			bio = bio_alloc(bdev, bio_max_segs(nr_pages),
357					REQ_OP_READ, GFP_KERNEL);
358			fscrypt_set_bio_crypt_ctx(bio, inode, next_block,
359						  GFP_KERNEL);
360			ext4_set_bio_post_read_ctx(bio, inode, folio->index);
361			bio->bi_iter.bi_sector = first_block << (blkbits - 9);
362			bio->bi_end_io = mpage_end_io;
363			if (rac)
364				bio->bi_opf |= REQ_RAHEAD;
365		}
366
367		length = first_hole << blkbits;
368		if (!bio_add_folio(bio, folio, length, 0))
369			goto submit_and_realloc;
370
371		if (((map.m_flags & EXT4_MAP_BOUNDARY) &&
372		     (relative_block == map.m_len)) ||
373		    (first_hole != blocks_per_folio)) {
374			submit_bio(bio);
375			bio = NULL;
376		} else
377			last_block_in_bio = first_block + blocks_per_folio - 1;
378		continue;
379	confused:
380		if (bio) {
381			submit_bio(bio);
382			bio = NULL;
383		}
384		if (!folio_test_uptodate(folio))
385			block_read_full_folio(folio, ext4_get_block);
386		else
387			folio_unlock(folio);
388next_page:
389		; /* A label shall be followed by a statement until C23 */
390	}
391	if (bio)
392		submit_bio(bio);
393	return 0;
394}
395
396int __init ext4_init_post_read_processing(void)
397{
398	bio_post_read_ctx_cache = KMEM_CACHE(bio_post_read_ctx, SLAB_RECLAIM_ACCOUNT);
399
400	if (!bio_post_read_ctx_cache)
401		goto fail;
402	bio_post_read_ctx_pool =
403		mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
404					 bio_post_read_ctx_cache);
405	if (!bio_post_read_ctx_pool)
406		goto fail_free_cache;
407	return 0;
408
409fail_free_cache:
410	kmem_cache_destroy(bio_post_read_ctx_cache);
411fail:
412	return -ENOMEM;
413}
414
415void ext4_exit_post_read_processing(void)
416{
417	mempool_destroy(bio_post_read_ctx_pool);
418	kmem_cache_destroy(bio_post_read_ctx_cache);
419}