1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Copyright (C) 2022, Alibaba Cloud
4 * Copyright (C) 2022, Bytedance Inc. All rights reserved.
5 */
6#include <linux/pseudo_fs.h>
7#include <linux/fscache.h>
8#include "internal.h"
9
10static DEFINE_MUTEX(erofs_domain_list_lock);
11static DEFINE_MUTEX(erofs_domain_cookies_lock);
12static LIST_HEAD(erofs_domain_list);
13static LIST_HEAD(erofs_domain_cookies_list);
14static struct vfsmount *erofs_pseudo_mnt;
15
16static int erofs_anon_init_fs_context(struct fs_context *fc)
17{
18	return init_pseudo(fc, EROFS_SUPER_MAGIC) ? 0 : -ENOMEM;
19}
20
21static struct file_system_type erofs_anon_fs_type = {
22	.owner		= THIS_MODULE,
23	.name           = "pseudo_erofs",
24	.init_fs_context = erofs_anon_init_fs_context,
25	.kill_sb        = kill_anon_super,
26};
27
28struct erofs_fscache_io {
29	struct netfs_cache_resources cres;
30	struct iov_iter		iter;
31	netfs_io_terminated_t	end_io;
32	void			*private;
33	refcount_t		ref;
34};
35
36struct erofs_fscache_rq {
37	struct address_space	*mapping;	/* The mapping being accessed */
38	loff_t			start;		/* Start position */
39	size_t			len;		/* Length of the request */
40	size_t			submitted;	/* Length of submitted */
41	short			error;		/* 0 or error that occurred */
42	refcount_t		ref;
43};
44
45static bool erofs_fscache_io_put(struct erofs_fscache_io *io)
46{
47	if (!refcount_dec_and_test(&io->ref))
48		return false;
49	if (io->cres.ops)
50		io->cres.ops->end_operation(&io->cres);
51	kfree(io);
52	return true;
53}
54
55static void erofs_fscache_req_complete(struct erofs_fscache_rq *req)
56{
57	struct folio *folio;
58	bool failed = req->error;
59	pgoff_t start_page = req->start / PAGE_SIZE;
60	pgoff_t last_page = ((req->start + req->len) / PAGE_SIZE) - 1;
61
62	XA_STATE(xas, &req->mapping->i_pages, start_page);
63
64	rcu_read_lock();
65	xas_for_each(&xas, folio, last_page) {
66		if (xas_retry(&xas, folio))
67			continue;
68		if (!failed)
69			folio_mark_uptodate(folio);
70		folio_unlock(folio);
71	}
72	rcu_read_unlock();
73}
74
75static void erofs_fscache_req_put(struct erofs_fscache_rq *req)
76{
77	if (!refcount_dec_and_test(&req->ref))
78		return;
79	erofs_fscache_req_complete(req);
80	kfree(req);
81}
82
83static struct erofs_fscache_rq *erofs_fscache_req_alloc(struct address_space *mapping,
84						loff_t start, size_t len)
85{
86	struct erofs_fscache_rq *req = kzalloc(sizeof(*req), GFP_KERNEL);
87
88	if (!req)
89		return NULL;
90	req->mapping = mapping;
91	req->start = start;
92	req->len = len;
93	refcount_set(&req->ref, 1);
94	return req;
95}
96
97static void erofs_fscache_req_io_put(struct erofs_fscache_io *io)
98{
99	struct erofs_fscache_rq *req = io->private;
100
101	if (erofs_fscache_io_put(io))
102		erofs_fscache_req_put(req);
103}
104
105static void erofs_fscache_req_end_io(void *priv,
106		ssize_t transferred_or_error, bool was_async)
107{
108	struct erofs_fscache_io *io = priv;
109	struct erofs_fscache_rq *req = io->private;
110
111	if (IS_ERR_VALUE(transferred_or_error))
112		req->error = transferred_or_error;
113	erofs_fscache_req_io_put(io);
114}
115
116static struct erofs_fscache_io *erofs_fscache_req_io_alloc(struct erofs_fscache_rq *req)
117{
118	struct erofs_fscache_io *io = kzalloc(sizeof(*io), GFP_KERNEL);
119
120	if (!io)
121		return NULL;
122	io->end_io = erofs_fscache_req_end_io;
123	io->private = req;
124	refcount_inc(&req->ref);
125	refcount_set(&io->ref, 1);
126	return io;
127}
128
129/*
130 * Read data from fscache described by cookie at pstart physical address
131 * offset, and fill the read data into buffer described by io->iter.
132 */
133static int erofs_fscache_read_io_async(struct fscache_cookie *cookie,
134		loff_t pstart, struct erofs_fscache_io *io)
135{
136	enum netfs_io_source source;
137	struct netfs_cache_resources *cres = &io->cres;
138	struct iov_iter *iter = &io->iter;
139	int ret;
140
141	ret = fscache_begin_read_operation(cres, cookie);
142	if (ret)
143		return ret;
144
145	while (iov_iter_count(iter)) {
146		size_t orig_count = iov_iter_count(iter), len = orig_count;
147		unsigned long flags = 1 << NETFS_SREQ_ONDEMAND;
148
149		source = cres->ops->prepare_ondemand_read(cres,
150				pstart, &len, LLONG_MAX, &flags, 0);
151		if (WARN_ON(len == 0))
152			source = NETFS_INVALID_READ;
153		if (source != NETFS_READ_FROM_CACHE) {
154			erofs_err(NULL, "prepare_read failed (source %d)", source);
155			return -EIO;
156		}
157
158		iov_iter_truncate(iter, len);
159		refcount_inc(&io->ref);
160		ret = fscache_read(cres, pstart, iter, NETFS_READ_HOLE_FAIL,
161				   io->end_io, io);
162		if (ret == -EIOCBQUEUED)
163			ret = 0;
164		if (ret) {
165			erofs_err(NULL, "fscache_read failed (ret %d)", ret);
166			return ret;
167		}
168		if (WARN_ON(iov_iter_count(iter)))
169			return -EIO;
170
171		iov_iter_reexpand(iter, orig_count - len);
172		pstart += len;
173	}
174	return 0;
175}
176
177struct erofs_fscache_bio {
178	struct erofs_fscache_io io;
179	struct bio bio;		/* w/o bdev to share bio_add_page/endio() */
180	struct bio_vec bvecs[BIO_MAX_VECS];
181};
182
183static void erofs_fscache_bio_endio(void *priv,
184		ssize_t transferred_or_error, bool was_async)
185{
186	struct erofs_fscache_bio *io = priv;
187
188	if (IS_ERR_VALUE(transferred_or_error))
189		io->bio.bi_status = errno_to_blk_status(transferred_or_error);
190	io->bio.bi_end_io(&io->bio);
191	BUILD_BUG_ON(offsetof(struct erofs_fscache_bio, io) != 0);
192	erofs_fscache_io_put(&io->io);
193}
194
195struct bio *erofs_fscache_bio_alloc(struct erofs_map_dev *mdev)
196{
197	struct erofs_fscache_bio *io;
198
199	io = kmalloc(sizeof(*io), GFP_KERNEL | __GFP_NOFAIL);
200	bio_init(&io->bio, NULL, io->bvecs, BIO_MAX_VECS, REQ_OP_READ);
201	io->io.private = mdev->m_fscache->cookie;
202	io->io.end_io = erofs_fscache_bio_endio;
203	refcount_set(&io->io.ref, 1);
204	return &io->bio;
205}
206
207void erofs_fscache_submit_bio(struct bio *bio)
208{
209	struct erofs_fscache_bio *io = container_of(bio,
210			struct erofs_fscache_bio, bio);
211	int ret;
212
213	iov_iter_bvec(&io->io.iter, ITER_DEST, io->bvecs, bio->bi_vcnt,
214		      bio->bi_iter.bi_size);
215	ret = erofs_fscache_read_io_async(io->io.private,
216				bio->bi_iter.bi_sector << 9, &io->io);
217	erofs_fscache_io_put(&io->io);
218	if (!ret)
219		return;
220	bio->bi_status = errno_to_blk_status(ret);
221	bio->bi_end_io(bio);
222}
223
224static int erofs_fscache_meta_read_folio(struct file *data, struct folio *folio)
225{
226	struct erofs_fscache *ctx = folio->mapping->host->i_private;
227	int ret = -ENOMEM;
228	struct erofs_fscache_rq *req;
229	struct erofs_fscache_io *io;
230
231	req = erofs_fscache_req_alloc(folio->mapping,
232				folio_pos(folio), folio_size(folio));
233	if (!req) {
234		folio_unlock(folio);
235		return ret;
236	}
237
238	io = erofs_fscache_req_io_alloc(req);
239	if (!io) {
240		req->error = ret;
241		goto out;
242	}
243	iov_iter_xarray(&io->iter, ITER_DEST, &folio->mapping->i_pages,
244			folio_pos(folio), folio_size(folio));
245
246	ret = erofs_fscache_read_io_async(ctx->cookie, folio_pos(folio), io);
247	if (ret)
248		req->error = ret;
249
250	erofs_fscache_req_io_put(io);
251out:
252	erofs_fscache_req_put(req);
253	return ret;
254}
255
256static int erofs_fscache_data_read_slice(struct erofs_fscache_rq *req)
257{
258	struct address_space *mapping = req->mapping;
259	struct inode *inode = mapping->host;
260	struct super_block *sb = inode->i_sb;
261	struct erofs_fscache_io *io;
262	struct erofs_map_blocks map;
263	struct erofs_map_dev mdev;
264	loff_t pos = req->start + req->submitted;
265	size_t count;
266	int ret;
267
268	map.m_la = pos;
269	ret = erofs_map_blocks(inode, &map);
270	if (ret)
271		return ret;
272
273	if (map.m_flags & EROFS_MAP_META) {
274		struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
275		struct iov_iter iter;
276		erofs_blk_t blknr;
277		size_t offset, size;
278		void *src;
279
280		/* For tail packing layout, the offset may be non-zero. */
281		offset = erofs_blkoff(sb, map.m_pa);
282		blknr = erofs_blknr(sb, map.m_pa);
283		size = map.m_llen;
284
285		src = erofs_read_metabuf(&buf, sb, blknr, EROFS_KMAP);
286		if (IS_ERR(src))
287			return PTR_ERR(src);
288
289		iov_iter_xarray(&iter, ITER_DEST, &mapping->i_pages, pos, PAGE_SIZE);
290		if (copy_to_iter(src + offset, size, &iter) != size) {
291			erofs_put_metabuf(&buf);
292			return -EFAULT;
293		}
294		iov_iter_zero(PAGE_SIZE - size, &iter);
295		erofs_put_metabuf(&buf);
296		req->submitted += PAGE_SIZE;
297		return 0;
298	}
299
300	count = req->len - req->submitted;
301	if (!(map.m_flags & EROFS_MAP_MAPPED)) {
302		struct iov_iter iter;
303
304		iov_iter_xarray(&iter, ITER_DEST, &mapping->i_pages, pos, count);
305		iov_iter_zero(count, &iter);
306		req->submitted += count;
307		return 0;
308	}
309
310	count = min_t(size_t, map.m_llen - (pos - map.m_la), count);
311	DBG_BUGON(!count || count % PAGE_SIZE);
312
313	mdev = (struct erofs_map_dev) {
314		.m_deviceid = map.m_deviceid,
315		.m_pa = map.m_pa,
316	};
317	ret = erofs_map_dev(sb, &mdev);
318	if (ret)
319		return ret;
320
321	io = erofs_fscache_req_io_alloc(req);
322	if (!io)
323		return -ENOMEM;
324	iov_iter_xarray(&io->iter, ITER_DEST, &mapping->i_pages, pos, count);
325	ret = erofs_fscache_read_io_async(mdev.m_fscache->cookie,
326			mdev.m_pa + (pos - map.m_la), io);
327	erofs_fscache_req_io_put(io);
328
329	req->submitted += count;
330	return ret;
331}
332
333static int erofs_fscache_data_read(struct erofs_fscache_rq *req)
334{
335	int ret;
336
337	do {
338		ret = erofs_fscache_data_read_slice(req);
339		if (ret)
340			req->error = ret;
341	} while (!ret && req->submitted < req->len);
342	return ret;
343}
344
345static int erofs_fscache_read_folio(struct file *file, struct folio *folio)
346{
347	struct erofs_fscache_rq *req;
348	int ret;
349
350	req = erofs_fscache_req_alloc(folio->mapping,
351			folio_pos(folio), folio_size(folio));
352	if (!req) {
353		folio_unlock(folio);
354		return -ENOMEM;
355	}
356
357	ret = erofs_fscache_data_read(req);
358	erofs_fscache_req_put(req);
359	return ret;
360}
361
362static void erofs_fscache_readahead(struct readahead_control *rac)
363{
364	struct erofs_fscache_rq *req;
365
366	if (!readahead_count(rac))
367		return;
368
369	req = erofs_fscache_req_alloc(rac->mapping,
370			readahead_pos(rac), readahead_length(rac));
371	if (!req)
372		return;
373
374	/* The request completion will drop refs on the folios. */
375	while (readahead_folio(rac))
376		;
377
378	erofs_fscache_data_read(req);
379	erofs_fscache_req_put(req);
380}
381
382static const struct address_space_operations erofs_fscache_meta_aops = {
383	.read_folio = erofs_fscache_meta_read_folio,
384};
385
386const struct address_space_operations erofs_fscache_access_aops = {
387	.read_folio = erofs_fscache_read_folio,
388	.readahead = erofs_fscache_readahead,
389};
390
391static void erofs_fscache_domain_put(struct erofs_domain *domain)
392{
393	mutex_lock(&erofs_domain_list_lock);
394	if (refcount_dec_and_test(&domain->ref)) {
395		list_del(&domain->list);
396		if (list_empty(&erofs_domain_list)) {
397			kern_unmount(erofs_pseudo_mnt);
398			erofs_pseudo_mnt = NULL;
399		}
400		fscache_relinquish_volume(domain->volume, NULL, false);
401		mutex_unlock(&erofs_domain_list_lock);
402		kfree(domain->domain_id);
403		kfree(domain);
404		return;
405	}
406	mutex_unlock(&erofs_domain_list_lock);
407}
408
409static int erofs_fscache_register_volume(struct super_block *sb)
410{
411	struct erofs_sb_info *sbi = EROFS_SB(sb);
412	char *domain_id = sbi->domain_id;
413	struct fscache_volume *volume;
414	char *name;
415	int ret = 0;
416
417	name = kasprintf(GFP_KERNEL, "erofs,%s",
418			 domain_id ? domain_id : sbi->fsid);
419	if (!name)
420		return -ENOMEM;
421
422	volume = fscache_acquire_volume(name, NULL, NULL, 0);
423	if (IS_ERR_OR_NULL(volume)) {
424		erofs_err(sb, "failed to register volume for %s", name);
425		ret = volume ? PTR_ERR(volume) : -EOPNOTSUPP;
426		volume = NULL;
427	}
428
429	sbi->volume = volume;
430	kfree(name);
431	return ret;
432}
433
434static int erofs_fscache_init_domain(struct super_block *sb)
435{
436	int err;
437	struct erofs_domain *domain;
438	struct erofs_sb_info *sbi = EROFS_SB(sb);
439
440	domain = kzalloc(sizeof(struct erofs_domain), GFP_KERNEL);
441	if (!domain)
442		return -ENOMEM;
443
444	domain->domain_id = kstrdup(sbi->domain_id, GFP_KERNEL);
445	if (!domain->domain_id) {
446		kfree(domain);
447		return -ENOMEM;
448	}
449
450	err = erofs_fscache_register_volume(sb);
451	if (err)
452		goto out;
453
454	if (!erofs_pseudo_mnt) {
455		struct vfsmount *mnt = kern_mount(&erofs_anon_fs_type);
456		if (IS_ERR(mnt)) {
457			err = PTR_ERR(mnt);
458			goto out;
459		}
460		erofs_pseudo_mnt = mnt;
461	}
462
463	domain->volume = sbi->volume;
464	refcount_set(&domain->ref, 1);
465	list_add(&domain->list, &erofs_domain_list);
466	sbi->domain = domain;
467	return 0;
468out:
469	kfree(domain->domain_id);
470	kfree(domain);
471	return err;
472}
473
474static int erofs_fscache_register_domain(struct super_block *sb)
475{
476	int err;
477	struct erofs_domain *domain;
478	struct erofs_sb_info *sbi = EROFS_SB(sb);
479
480	mutex_lock(&erofs_domain_list_lock);
481	list_for_each_entry(domain, &erofs_domain_list, list) {
482		if (!strcmp(domain->domain_id, sbi->domain_id)) {
483			sbi->domain = domain;
484			sbi->volume = domain->volume;
485			refcount_inc(&domain->ref);
486			mutex_unlock(&erofs_domain_list_lock);
487			return 0;
488		}
489	}
490	err = erofs_fscache_init_domain(sb);
491	mutex_unlock(&erofs_domain_list_lock);
492	return err;
493}
494
495static struct erofs_fscache *erofs_fscache_acquire_cookie(struct super_block *sb,
496						char *name, unsigned int flags)
497{
498	struct fscache_volume *volume = EROFS_SB(sb)->volume;
499	struct erofs_fscache *ctx;
500	struct fscache_cookie *cookie;
501	struct super_block *isb;
502	struct inode *inode;
503	int ret;
504
505	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
506	if (!ctx)
507		return ERR_PTR(-ENOMEM);
508	INIT_LIST_HEAD(&ctx->node);
509	refcount_set(&ctx->ref, 1);
510
511	cookie = fscache_acquire_cookie(volume, FSCACHE_ADV_WANT_CACHE_SIZE,
512					name, strlen(name), NULL, 0, 0);
513	if (!cookie) {
514		erofs_err(sb, "failed to get cookie for %s", name);
515		ret = -EINVAL;
516		goto err;
517	}
518	fscache_use_cookie(cookie, false);
519
520	/*
521	 * Allocate anonymous inode in global pseudo mount for shareable blobs,
522	 * so that they are accessible among erofs fs instances.
523	 */
524	isb = flags & EROFS_REG_COOKIE_SHARE ? erofs_pseudo_mnt->mnt_sb : sb;
525	inode = new_inode(isb);
526	if (!inode) {
527		erofs_err(sb, "failed to get anon inode for %s", name);
528		ret = -ENOMEM;
529		goto err_cookie;
530	}
531
532	inode->i_size = OFFSET_MAX;
533	inode->i_mapping->a_ops = &erofs_fscache_meta_aops;
534	mapping_set_gfp_mask(inode->i_mapping, GFP_KERNEL);
535	inode->i_blkbits = EROFS_SB(sb)->blkszbits;
536	inode->i_private = ctx;
537
538	ctx->cookie = cookie;
539	ctx->inode = inode;
540	return ctx;
541
542err_cookie:
543	fscache_unuse_cookie(cookie, NULL, NULL);
544	fscache_relinquish_cookie(cookie, false);
545err:
546	kfree(ctx);
547	return ERR_PTR(ret);
548}
549
550static void erofs_fscache_relinquish_cookie(struct erofs_fscache *ctx)
551{
552	fscache_unuse_cookie(ctx->cookie, NULL, NULL);
553	fscache_relinquish_cookie(ctx->cookie, false);
554	iput(ctx->inode);
555	kfree(ctx->name);
556	kfree(ctx);
557}
558
559static struct erofs_fscache *erofs_domain_init_cookie(struct super_block *sb,
560						char *name, unsigned int flags)
561{
562	struct erofs_fscache *ctx;
563	struct erofs_domain *domain = EROFS_SB(sb)->domain;
564
565	ctx = erofs_fscache_acquire_cookie(sb, name, flags);
566	if (IS_ERR(ctx))
567		return ctx;
568
569	ctx->name = kstrdup(name, GFP_KERNEL);
570	if (!ctx->name) {
571		erofs_fscache_relinquish_cookie(ctx);
572		return ERR_PTR(-ENOMEM);
573	}
574
575	refcount_inc(&domain->ref);
576	ctx->domain = domain;
577	list_add(&ctx->node, &erofs_domain_cookies_list);
578	return ctx;
579}
580
581static struct erofs_fscache *erofs_domain_register_cookie(struct super_block *sb,
582						char *name, unsigned int flags)
583{
584	struct erofs_fscache *ctx;
585	struct erofs_domain *domain = EROFS_SB(sb)->domain;
586
587	flags |= EROFS_REG_COOKIE_SHARE;
588	mutex_lock(&erofs_domain_cookies_lock);
589	list_for_each_entry(ctx, &erofs_domain_cookies_list, node) {
590		if (ctx->domain != domain || strcmp(ctx->name, name))
591			continue;
592		if (!(flags & EROFS_REG_COOKIE_NEED_NOEXIST)) {
593			refcount_inc(&ctx->ref);
594		} else {
595			erofs_err(sb, "%s already exists in domain %s", name,
596				  domain->domain_id);
597			ctx = ERR_PTR(-EEXIST);
598		}
599		mutex_unlock(&erofs_domain_cookies_lock);
600		return ctx;
601	}
602	ctx = erofs_domain_init_cookie(sb, name, flags);
603	mutex_unlock(&erofs_domain_cookies_lock);
604	return ctx;
605}
606
607struct erofs_fscache *erofs_fscache_register_cookie(struct super_block *sb,
608						    char *name,
609						    unsigned int flags)
610{
611	if (EROFS_SB(sb)->domain_id)
612		return erofs_domain_register_cookie(sb, name, flags);
613	return erofs_fscache_acquire_cookie(sb, name, flags);
614}
615
616void erofs_fscache_unregister_cookie(struct erofs_fscache *ctx)
617{
618	struct erofs_domain *domain = NULL;
619
620	if (!ctx)
621		return;
622	if (!ctx->domain)
623		return erofs_fscache_relinquish_cookie(ctx);
624
625	mutex_lock(&erofs_domain_cookies_lock);
626	if (refcount_dec_and_test(&ctx->ref)) {
627		domain = ctx->domain;
628		list_del(&ctx->node);
629		erofs_fscache_relinquish_cookie(ctx);
630	}
631	mutex_unlock(&erofs_domain_cookies_lock);
632	if (domain)
633		erofs_fscache_domain_put(domain);
634}
635
636int erofs_fscache_register_fs(struct super_block *sb)
637{
638	int ret;
639	struct erofs_sb_info *sbi = EROFS_SB(sb);
640	struct erofs_fscache *fscache;
641	unsigned int flags = 0;
642
643	if (sbi->domain_id)
644		ret = erofs_fscache_register_domain(sb);
645	else
646		ret = erofs_fscache_register_volume(sb);
647	if (ret)
648		return ret;
649
650	/*
651	 * When shared domain is enabled, using NEED_NOEXIST to guarantee
652	 * the primary data blob (aka fsid) is unique in the shared domain.
653	 *
654	 * For non-shared-domain case, fscache_acquire_volume() invoked by
655	 * erofs_fscache_register_volume() has already guaranteed
656	 * the uniqueness of primary data blob.
657	 *
658	 * Acquired domain/volume will be relinquished in kill_sb() on error.
659	 */
660	if (sbi->domain_id)
661		flags |= EROFS_REG_COOKIE_NEED_NOEXIST;
662	fscache = erofs_fscache_register_cookie(sb, sbi->fsid, flags);
663	if (IS_ERR(fscache))
664		return PTR_ERR(fscache);
665
666	sbi->s_fscache = fscache;
667	return 0;
668}
669
670void erofs_fscache_unregister_fs(struct super_block *sb)
671{
672	struct erofs_sb_info *sbi = EROFS_SB(sb);
673
674	erofs_fscache_unregister_cookie(sbi->s_fscache);
675
676	if (sbi->domain)
677		erofs_fscache_domain_put(sbi->domain);
678	else
679		fscache_relinquish_volume(sbi->volume, NULL, false);
680
681	sbi->s_fscache = NULL;
682	sbi->volume = NULL;
683	sbi->domain = NULL;
684}
685