Skip to content

Commit 7c35de4

Browse files
hsiangkaohsiangkao
committed
erofs: Zstandard compression support
Add Zstandard compression as the 4th supported algorithm since it becomes more popular now and some end users have asked this for quite a while [1][2]. Each EROFS physical cluster contains only one valid standard Zstandard frame as described in [3] so that decompression can be performed on a per-pcluster basis independently. Currently, it just leverages multi-call stream decompression APIs with internal sliding window buffers. One-shot or bufferless decompression could be implemented later for even better performance if needed. [1] erofs/erofs-utils#6 [2] https://lore.kernel.org/r/[email protected] [3] https://www.rfc-editor.org/rfc/rfc8478.txt Acked-by: Chao Yu <[email protected]> Signed-off-by: Gao Xiang <[email protected]> Link: https://lore.kernel.org/r/[email protected]
1 parent d691894 commit 7c35de4

File tree

9 files changed

+333
-1
lines changed

9 files changed

+333
-1
lines changed

‎fs/erofs/Kconfig

Lines changed: 15 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -112,6 +112,21 @@ config EROFS_FS_ZIP_DEFLATE
112112

113113
If unsure, say N.
114114

115+
config EROFS_FS_ZIP_ZSTD
116+
bool "EROFS Zstandard compressed data support"
117+
depends on EROFS_FS_ZIP
118+
select ZSTD_DECOMPRESS
119+
help
120+
Saying Y here includes support for reading EROFS file systems
121+
containing Zstandard compressed data. It gives better compression
122+
ratios than the default LZ4 format, while it costs more CPU
123+
overhead.
124+
125+
Zstandard support is an experimental feature for now and so most
126+
file systems will be readable without selecting this option.
127+
128+
If unsure, say N.
129+
115130
config EROFS_FS_ONDEMAND
116131
bool "EROFS fscache-based on-demand read support"
117132
depends on EROFS_FS

‎fs/erofs/Makefile

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -6,4 +6,5 @@ erofs-$(CONFIG_EROFS_FS_XATTR) += xattr.o
66
erofs-$(CONFIG_EROFS_FS_ZIP) += decompressor.o zmap.o zdata.o zutil.o
77
erofs-$(CONFIG_EROFS_FS_ZIP_LZMA) += decompressor_lzma.o
88
erofs-$(CONFIG_EROFS_FS_ZIP_DEFLATE) += decompressor_deflate.o
9+
erofs-$(CONFIG_EROFS_FS_ZIP_ZSTD) += decompressor_zstd.o
910
erofs-$(CONFIG_EROFS_FS_ONDEMAND) += fscache.o

‎fs/erofs/compress.h

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -90,8 +90,12 @@ int z_erofs_load_lzma_config(struct super_block *sb,
9090
struct erofs_super_block *dsb, void *data, int size);
9191
int z_erofs_load_deflate_config(struct super_block *sb,
9292
struct erofs_super_block *dsb, void *data, int size);
93+
int z_erofs_load_zstd_config(struct super_block *sb,
94+
struct erofs_super_block *dsb, void *data, int size);
9395
int z_erofs_lzma_decompress(struct z_erofs_decompress_req *rq,
9496
struct page **pagepool);
9597
int z_erofs_deflate_decompress(struct z_erofs_decompress_req *rq,
9698
struct page **pagepool);
99+
int z_erofs_zstd_decompress(struct z_erofs_decompress_req *rq,
100+
struct page **pgpl);
97101
#endif

‎fs/erofs/decompressor.c

Lines changed: 7 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -399,6 +399,13 @@ const struct z_erofs_decompressor erofs_decompressors[] = {
399399
.name = "deflate"
400400
},
401401
#endif
402+
#ifdef CONFIG_EROFS_FS_ZIP_ZSTD
403+
[Z_EROFS_COMPRESSION_ZSTD] = {
404+
.config = z_erofs_load_zstd_config,
405+
.decompress = z_erofs_zstd_decompress,
406+
.name = "zstd"
407+
},
408+
#endif
402409
};
403410

404411
int z_erofs_parse_cfgs(struct super_block *sb, struct erofs_super_block *dsb)

‎fs/erofs/decompressor_zstd.c

Lines changed: 279 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,279 @@
1+
// SPDX-License-Identifier: GPL-2.0-or-later
2+
#include <linux/zstd.h>
3+
#include "compress.h"
4+
5+
struct z_erofs_zstd {
6+
struct z_erofs_zstd *next;
7+
u8 bounce[PAGE_SIZE];
8+
void *wksp;
9+
unsigned int wkspsz;
10+
};
11+
12+
static DEFINE_SPINLOCK(z_erofs_zstd_lock);
13+
static unsigned int z_erofs_zstd_max_dictsize;
14+
static unsigned int z_erofs_zstd_nstrms, z_erofs_zstd_avail_strms;
15+
static struct z_erofs_zstd *z_erofs_zstd_head;
16+
static DECLARE_WAIT_QUEUE_HEAD(z_erofs_zstd_wq);
17+
18+
module_param_named(zstd_streams, z_erofs_zstd_nstrms, uint, 0444);
19+
20+
static struct z_erofs_zstd *z_erofs_isolate_strms(bool all)
21+
{
22+
struct z_erofs_zstd *strm;
23+
24+
again:
25+
spin_lock(&z_erofs_zstd_lock);
26+
strm = z_erofs_zstd_head;
27+
if (!strm) {
28+
spin_unlock(&z_erofs_zstd_lock);
29+
wait_event(z_erofs_zstd_wq, READ_ONCE(z_erofs_zstd_head));
30+
goto again;
31+
}
32+
z_erofs_zstd_head = all ? NULL : strm->next;
33+
spin_unlock(&z_erofs_zstd_lock);
34+
return strm;
35+
}
36+
37+
void z_erofs_zstd_exit(void)
38+
{
39+
while (z_erofs_zstd_avail_strms) {
40+
struct z_erofs_zstd *strm, *n;
41+
42+
for (strm = z_erofs_isolate_strms(true); strm; strm = n) {
43+
n = strm->next;
44+
45+
kvfree(strm->wksp);
46+
kfree(strm);
47+
--z_erofs_zstd_avail_strms;
48+
}
49+
}
50+
}
51+
52+
int __init z_erofs_zstd_init(void)
53+
{
54+
/* by default, use # of possible CPUs instead */
55+
if (!z_erofs_zstd_nstrms)
56+
z_erofs_zstd_nstrms = num_possible_cpus();
57+
58+
for (; z_erofs_zstd_avail_strms < z_erofs_zstd_nstrms;
59+
++z_erofs_zstd_avail_strms) {
60+
struct z_erofs_zstd *strm;
61+
62+
strm = kzalloc(sizeof(*strm), GFP_KERNEL);
63+
if (!strm) {
64+
z_erofs_zstd_exit();
65+
return -ENOMEM;
66+
}
67+
spin_lock(&z_erofs_zstd_lock);
68+
strm->next = z_erofs_zstd_head;
69+
z_erofs_zstd_head = strm;
70+
spin_unlock(&z_erofs_zstd_lock);
71+
}
72+
return 0;
73+
}
74+
75+
int z_erofs_load_zstd_config(struct super_block *sb,
76+
struct erofs_super_block *dsb, void *data, int size)
77+
{
78+
static DEFINE_MUTEX(zstd_resize_mutex);
79+
struct z_erofs_zstd_cfgs *zstd = data;
80+
unsigned int dict_size, wkspsz;
81+
struct z_erofs_zstd *strm, *head = NULL;
82+
void *wksp;
83+
84+
if (!zstd || size < sizeof(struct z_erofs_zstd_cfgs) || zstd->format) {
85+
erofs_err(sb, "unsupported zstd format, size=%u", size);
86+
return -EINVAL;
87+
}
88+
89+
if (zstd->windowlog > ilog2(Z_EROFS_ZSTD_MAX_DICT_SIZE) - 10) {
90+
erofs_err(sb, "unsupported zstd window log %u", zstd->windowlog);
91+
return -EINVAL;
92+
}
93+
dict_size = 1U << (zstd->windowlog + 10);
94+
95+
/* in case 2 z_erofs_load_zstd_config() race to avoid deadlock */
96+
mutex_lock(&zstd_resize_mutex);
97+
if (z_erofs_zstd_max_dictsize >= dict_size) {
98+
mutex_unlock(&zstd_resize_mutex);
99+
return 0;
100+
}
101+
102+
/* 1. collect/isolate all streams for the following check */
103+
while (z_erofs_zstd_avail_strms) {
104+
struct z_erofs_zstd *n;
105+
106+
for (strm = z_erofs_isolate_strms(true); strm; strm = n) {
107+
n = strm->next;
108+
strm->next = head;
109+
head = strm;
110+
--z_erofs_zstd_avail_strms;
111+
}
112+
}
113+
114+
/* 2. walk each isolated stream and grow max dict_size if needed */
115+
wkspsz = zstd_dstream_workspace_bound(dict_size);
116+
for (strm = head; strm; strm = strm->next) {
117+
wksp = kvmalloc(wkspsz, GFP_KERNEL);
118+
if (!wksp)
119+
break;
120+
kvfree(strm->wksp);
121+
strm->wksp = wksp;
122+
strm->wkspsz = wkspsz;
123+
}
124+
125+
/* 3. push back all to the global list and update max dict_size */
126+
spin_lock(&z_erofs_zstd_lock);
127+
DBG_BUGON(z_erofs_zstd_head);
128+
z_erofs_zstd_head = head;
129+
spin_unlock(&z_erofs_zstd_lock);
130+
z_erofs_zstd_avail_strms = z_erofs_zstd_nstrms;
131+
wake_up_all(&z_erofs_zstd_wq);
132+
if (!strm)
133+
z_erofs_zstd_max_dictsize = dict_size;
134+
mutex_unlock(&zstd_resize_mutex);
135+
return strm ? -ENOMEM : 0;
136+
}
137+
138+
int z_erofs_zstd_decompress(struct z_erofs_decompress_req *rq,
139+
struct page **pgpl)
140+
{
141+
const unsigned int nrpages_out =
142+
PAGE_ALIGN(rq->pageofs_out + rq->outputsize) >> PAGE_SHIFT;
143+
const unsigned int nrpages_in =
144+
PAGE_ALIGN(rq->inputsize) >> PAGE_SHIFT;
145+
zstd_dstream *stream;
146+
struct super_block *sb = rq->sb;
147+
unsigned int insz, outsz, pofs;
148+
struct z_erofs_zstd *strm;
149+
zstd_in_buffer in_buf = { NULL, 0, 0 };
150+
zstd_out_buffer out_buf = { NULL, 0, 0 };
151+
u8 *kin, *kout = NULL;
152+
bool bounced = false;
153+
int no = -1, ni = 0, j = 0, zerr, err;
154+
155+
/* 1. get the exact compressed size */
156+
kin = kmap_local_page(*rq->in);
157+
err = z_erofs_fixup_insize(rq, kin + rq->pageofs_in,
158+
min_t(unsigned int, rq->inputsize,
159+
sb->s_blocksize - rq->pageofs_in));
160+
if (err) {
161+
kunmap_local(kin);
162+
return err;
163+
}
164+
165+
/* 2. get an available ZSTD context */
166+
strm = z_erofs_isolate_strms(false);
167+
168+
/* 3. multi-call decompress */
169+
insz = rq->inputsize;
170+
outsz = rq->outputsize;
171+
stream = zstd_init_dstream(z_erofs_zstd_max_dictsize, strm->wksp, strm->wkspsz);
172+
if (!stream) {
173+
err = -EIO;
174+
goto failed_zinit;
175+
}
176+
177+
pofs = rq->pageofs_out;
178+
in_buf.size = min_t(u32, insz, PAGE_SIZE - rq->pageofs_in);
179+
insz -= in_buf.size;
180+
in_buf.src = kin + rq->pageofs_in;
181+
do {
182+
if (out_buf.size == out_buf.pos) {
183+
if (++no >= nrpages_out || !outsz) {
184+
erofs_err(sb, "insufficient space for decompressed data");
185+
err = -EFSCORRUPTED;
186+
break;
187+
}
188+
189+
if (kout)
190+
kunmap_local(kout);
191+
out_buf.size = min_t(u32, outsz, PAGE_SIZE - pofs);
192+
outsz -= out_buf.size;
193+
if (!rq->out[no]) {
194+
rq->out[no] = erofs_allocpage(pgpl, rq->gfp);
195+
if (!rq->out[no]) {
196+
kout = NULL;
197+
err = -ENOMEM;
198+
break;
199+
}
200+
set_page_private(rq->out[no],
201+
Z_EROFS_SHORTLIVED_PAGE);
202+
}
203+
kout = kmap_local_page(rq->out[no]);
204+
out_buf.dst = kout + pofs;
205+
out_buf.pos = 0;
206+
pofs = 0;
207+
}
208+
209+
if (in_buf.size == in_buf.pos && insz) {
210+
if (++ni >= nrpages_in) {
211+
erofs_err(sb, "invalid compressed data");
212+
err = -EFSCORRUPTED;
213+
break;
214+
}
215+
216+
if (kout) /* unlike kmap(), take care of the orders */
217+
kunmap_local(kout);
218+
kunmap_local(kin);
219+
in_buf.size = min_t(u32, insz, PAGE_SIZE);
220+
insz -= in_buf.size;
221+
kin = kmap_local_page(rq->in[ni]);
222+
in_buf.src = kin;
223+
in_buf.pos = 0;
224+
bounced = false;
225+
if (kout) {
226+
j = (u8 *)out_buf.dst - kout;
227+
kout = kmap_local_page(rq->out[no]);
228+
out_buf.dst = kout + j;
229+
}
230+
}
231+
232+
/*
233+
* Handle overlapping: Use bounced buffer if the compressed
234+
* data is under processing; Or use short-lived pages from the
235+
* on-stack pagepool where pages share among the same request
236+
* and not _all_ inplace I/O pages are needed to be doubled.
237+
*/
238+
if (!bounced && rq->out[no] == rq->in[ni]) {
239+
memcpy(strm->bounce, in_buf.src, in_buf.size);
240+
in_buf.src = strm->bounce;
241+
bounced = true;
242+
}
243+
244+
for (j = ni + 1; j < nrpages_in; ++j) {
245+
struct page *tmppage;
246+
247+
if (rq->out[no] != rq->in[j])
248+
continue;
249+
tmppage = erofs_allocpage(pgpl, rq->gfp);
250+
if (!tmppage) {
251+
err = -ENOMEM;
252+
goto failed;
253+
}
254+
set_page_private(tmppage, Z_EROFS_SHORTLIVED_PAGE);
255+
copy_highpage(tmppage, rq->in[j]);
256+
rq->in[j] = tmppage;
257+
}
258+
zerr = zstd_decompress_stream(stream, &out_buf, &in_buf);
259+
if (zstd_is_error(zerr) || (!zerr && outsz)) {
260+
erofs_err(sb, "failed to decompress in[%u] out[%u]: %s",
261+
rq->inputsize, rq->outputsize,
262+
zerr ? zstd_get_error_name(zerr) : "unexpected end of stream");
263+
err = -EFSCORRUPTED;
264+
break;
265+
}
266+
} while (outsz || out_buf.pos < out_buf.size);
267+
failed:
268+
if (kout)
269+
kunmap_local(kout);
270+
failed_zinit:
271+
kunmap_local(kin);
272+
/* 4. push back ZSTD stream context to the global list */
273+
spin_lock(&z_erofs_zstd_lock);
274+
strm->next = z_erofs_zstd_head;
275+
z_erofs_zstd_head = strm;
276+
spin_unlock(&z_erofs_zstd_lock);
277+
wake_up(&z_erofs_zstd_wq);
278+
return err;
279+
}

‎fs/erofs/erofs_fs.h

Lines changed: 10 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -296,6 +296,7 @@ enum {
296296
Z_EROFS_COMPRESSION_LZ4 = 0,
297297
Z_EROFS_COMPRESSION_LZMA = 1,
298298
Z_EROFS_COMPRESSION_DEFLATE = 2,
299+
Z_EROFS_COMPRESSION_ZSTD = 3,
299300
Z_EROFS_COMPRESSION_MAX
300301
};
301302
#define Z_EROFS_ALL_COMPR_ALGS ((1 << Z_EROFS_COMPRESSION_MAX) - 1)
@@ -322,6 +323,15 @@ struct z_erofs_deflate_cfgs {
322323
u8 reserved[5];
323324
} __packed;
324325

326+
/* 6 bytes (+ length field = 8 bytes) */
327+
struct z_erofs_zstd_cfgs {
328+
u8 format;
329+
u8 windowlog; /* windowLog - ZSTD_WINDOWLOG_ABSOLUTEMIN(10) */
330+
u8 reserved[4];
331+
} __packed;
332+
333+
#define Z_EROFS_ZSTD_MAX_DICT_SIZE Z_EROFS_PCLUSTER_MAX_SIZE
334+
325335
/*
326336
* bit 0 : COMPACTED_2B indexes (0 - off; 1 - on)
327337
* e.g. for 4k logical cluster size, 4B if compacted 2B is off;

‎fs/erofs/internal.h

Lines changed: 8 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -502,6 +502,14 @@ static inline int z_erofs_deflate_init(void) { return 0; }
502502
static inline int z_erofs_deflate_exit(void) { return 0; }
503503
#endif /* !CONFIG_EROFS_FS_ZIP_DEFLATE */
504504

505+
#ifdef CONFIG_EROFS_FS_ZIP_ZSTD
506+
int __init z_erofs_zstd_init(void);
507+
void z_erofs_zstd_exit(void);
508+
#else
509+
static inline int z_erofs_zstd_init(void) { return 0; }
510+
static inline int z_erofs_zstd_exit(void) { return 0; }
511+
#endif /* !CONFIG_EROFS_FS_ZIP_ZSTD */
512+
505513
#ifdef CONFIG_EROFS_FS_ONDEMAND
506514
int erofs_fscache_register_fs(struct super_block *sb);
507515
void erofs_fscache_unregister_fs(struct super_block *sb);

0 commit comments

Comments
 (0)