xref: /linux/fs/erofs/decompressor_deflate.c (revision 001821b0e79716c4e17c71d8e053a23599a7a508)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 #include <linux/zlib.h>
3 #include "compress.h"
4 
5 struct z_erofs_deflate {
6 	struct z_erofs_deflate *next;
7 	struct z_stream_s z;
8 	u8 bounce[PAGE_SIZE];
9 };
10 
11 static DEFINE_SPINLOCK(z_erofs_deflate_lock);
12 static unsigned int z_erofs_deflate_nstrms, z_erofs_deflate_avail_strms;
13 static struct z_erofs_deflate *z_erofs_deflate_head;
14 static DECLARE_WAIT_QUEUE_HEAD(z_erofs_deflate_wq);
15 
16 module_param_named(deflate_streams, z_erofs_deflate_nstrms, uint, 0444);
17 
18 void z_erofs_deflate_exit(void)
19 {
20 	/* there should be no running fs instance */
21 	while (z_erofs_deflate_avail_strms) {
22 		struct z_erofs_deflate *strm;
23 
24 		spin_lock(&z_erofs_deflate_lock);
25 		strm = z_erofs_deflate_head;
26 		if (!strm) {
27 			spin_unlock(&z_erofs_deflate_lock);
28 			continue;
29 		}
30 		z_erofs_deflate_head = NULL;
31 		spin_unlock(&z_erofs_deflate_lock);
32 
33 		while (strm) {
34 			struct z_erofs_deflate *n = strm->next;
35 
36 			vfree(strm->z.workspace);
37 			kfree(strm);
38 			--z_erofs_deflate_avail_strms;
39 			strm = n;
40 		}
41 	}
42 }
43 
44 int __init z_erofs_deflate_init(void)
45 {
46 	/* by default, use # of possible CPUs instead */
47 	if (!z_erofs_deflate_nstrms)
48 		z_erofs_deflate_nstrms = num_possible_cpus();
49 	return 0;
50 }
51 
52 int z_erofs_load_deflate_config(struct super_block *sb,
53 			struct erofs_super_block *dsb, void *data, int size)
54 {
55 	struct z_erofs_deflate_cfgs *dfl = data;
56 	static DEFINE_MUTEX(deflate_resize_mutex);
57 	static bool inited;
58 
59 	if (!dfl || size < sizeof(struct z_erofs_deflate_cfgs)) {
60 		erofs_err(sb, "invalid deflate cfgs, size=%u", size);
61 		return -EINVAL;
62 	}
63 
64 	if (dfl->windowbits > MAX_WBITS) {
65 		erofs_err(sb, "unsupported windowbits %u", dfl->windowbits);
66 		return -EOPNOTSUPP;
67 	}
68 	mutex_lock(&deflate_resize_mutex);
69 	if (!inited) {
70 		for (; z_erofs_deflate_avail_strms < z_erofs_deflate_nstrms;
71 		     ++z_erofs_deflate_avail_strms) {
72 			struct z_erofs_deflate *strm;
73 
74 			strm = kzalloc(sizeof(*strm), GFP_KERNEL);
75 			if (!strm)
76 				goto failed;
77 			/* XXX: in-kernel zlib cannot customize windowbits */
78 			strm->z.workspace = vmalloc(zlib_inflate_workspacesize());
79 			if (!strm->z.workspace) {
80 				kfree(strm);
81 				goto failed;
82 			}
83 
84 			spin_lock(&z_erofs_deflate_lock);
85 			strm->next = z_erofs_deflate_head;
86 			z_erofs_deflate_head = strm;
87 			spin_unlock(&z_erofs_deflate_lock);
88 		}
89 		inited = true;
90 	}
91 	mutex_unlock(&deflate_resize_mutex);
92 	erofs_info(sb, "EXPERIMENTAL DEFLATE feature in use. Use at your own risk!");
93 	return 0;
94 failed:
95 	mutex_unlock(&deflate_resize_mutex);
96 	z_erofs_deflate_exit();
97 	return -ENOMEM;
98 }
99 
100 int z_erofs_deflate_decompress(struct z_erofs_decompress_req *rq,
101 			       struct page **pgpl)
102 {
103 	const unsigned int nrpages_out =
104 		PAGE_ALIGN(rq->pageofs_out + rq->outputsize) >> PAGE_SHIFT;
105 	const unsigned int nrpages_in =
106 		PAGE_ALIGN(rq->inputsize) >> PAGE_SHIFT;
107 	struct super_block *sb = rq->sb;
108 	unsigned int insz, outsz, pofs;
109 	struct z_erofs_deflate *strm;
110 	u8 *kin, *kout = NULL;
111 	bool bounced = false;
112 	int no = -1, ni = 0, j = 0, zerr, err;
113 
114 	/* 1. get the exact DEFLATE compressed size */
115 	kin = kmap_local_page(*rq->in);
116 	err = z_erofs_fixup_insize(rq, kin + rq->pageofs_in,
117 			min_t(unsigned int, rq->inputsize,
118 			      sb->s_blocksize - rq->pageofs_in));
119 	if (err) {
120 		kunmap_local(kin);
121 		return err;
122 	}
123 
124 	/* 2. get an available DEFLATE context */
125 again:
126 	spin_lock(&z_erofs_deflate_lock);
127 	strm = z_erofs_deflate_head;
128 	if (!strm) {
129 		spin_unlock(&z_erofs_deflate_lock);
130 		wait_event(z_erofs_deflate_wq, READ_ONCE(z_erofs_deflate_head));
131 		goto again;
132 	}
133 	z_erofs_deflate_head = strm->next;
134 	spin_unlock(&z_erofs_deflate_lock);
135 
136 	/* 3. multi-call decompress */
137 	insz = rq->inputsize;
138 	outsz = rq->outputsize;
139 	zerr = zlib_inflateInit2(&strm->z, -MAX_WBITS);
140 	if (zerr != Z_OK) {
141 		err = -EIO;
142 		goto failed_zinit;
143 	}
144 
145 	pofs = rq->pageofs_out;
146 	strm->z.avail_in = min_t(u32, insz, PAGE_SIZE - rq->pageofs_in);
147 	insz -= strm->z.avail_in;
148 	strm->z.next_in = kin + rq->pageofs_in;
149 	strm->z.avail_out = 0;
150 
151 	while (1) {
152 		if (!strm->z.avail_out) {
153 			if (++no >= nrpages_out || !outsz) {
154 				erofs_err(sb, "insufficient space for decompressed data");
155 				err = -EFSCORRUPTED;
156 				break;
157 			}
158 
159 			if (kout)
160 				kunmap_local(kout);
161 			strm->z.avail_out = min_t(u32, outsz, PAGE_SIZE - pofs);
162 			outsz -= strm->z.avail_out;
163 			if (!rq->out[no]) {
164 				rq->out[no] = erofs_allocpage(pgpl, rq->gfp);
165 				if (!rq->out[no]) {
166 					kout = NULL;
167 					err = -ENOMEM;
168 					break;
169 				}
170 				set_page_private(rq->out[no],
171 						 Z_EROFS_SHORTLIVED_PAGE);
172 			}
173 			kout = kmap_local_page(rq->out[no]);
174 			strm->z.next_out = kout + pofs;
175 			pofs = 0;
176 		}
177 
178 		if (!strm->z.avail_in && insz) {
179 			if (++ni >= nrpages_in) {
180 				erofs_err(sb, "invalid compressed data");
181 				err = -EFSCORRUPTED;
182 				break;
183 			}
184 
185 			if (kout) { /* unlike kmap(), take care of the orders */
186 				j = strm->z.next_out - kout;
187 				kunmap_local(kout);
188 			}
189 			kunmap_local(kin);
190 			strm->z.avail_in = min_t(u32, insz, PAGE_SIZE);
191 			insz -= strm->z.avail_in;
192 			kin = kmap_local_page(rq->in[ni]);
193 			strm->z.next_in = kin;
194 			bounced = false;
195 			if (kout) {
196 				kout = kmap_local_page(rq->out[no]);
197 				strm->z.next_out = kout + j;
198 			}
199 		}
200 
201 		/*
202 		 * Handle overlapping: Use bounced buffer if the compressed
203 		 * data is under processing; Or use short-lived pages from the
204 		 * on-stack pagepool where pages share among the same request
205 		 * and not _all_ inplace I/O pages are needed to be doubled.
206 		 */
207 		if (!bounced && rq->out[no] == rq->in[ni]) {
208 			memcpy(strm->bounce, strm->z.next_in, strm->z.avail_in);
209 			strm->z.next_in = strm->bounce;
210 			bounced = true;
211 		}
212 
213 		for (j = ni + 1; j < nrpages_in; ++j) {
214 			struct page *tmppage;
215 
216 			if (rq->out[no] != rq->in[j])
217 				continue;
218 			tmppage = erofs_allocpage(pgpl, rq->gfp);
219 			if (!tmppage) {
220 				err = -ENOMEM;
221 				goto failed;
222 			}
223 			set_page_private(tmppage, Z_EROFS_SHORTLIVED_PAGE);
224 			copy_highpage(tmppage, rq->in[j]);
225 			rq->in[j] = tmppage;
226 		}
227 
228 		zerr = zlib_inflate(&strm->z, Z_SYNC_FLUSH);
229 		if (zerr != Z_OK || !(outsz + strm->z.avail_out)) {
230 			if (zerr == Z_OK && rq->partial_decoding)
231 				break;
232 			if (zerr == Z_STREAM_END && !outsz)
233 				break;
234 			erofs_err(sb, "failed to decompress %d in[%u] out[%u]",
235 				  zerr, rq->inputsize, rq->outputsize);
236 			err = -EFSCORRUPTED;
237 			break;
238 		}
239 	}
240 failed:
241 	if (zlib_inflateEnd(&strm->z) != Z_OK && !err)
242 		err = -EIO;
243 	if (kout)
244 		kunmap_local(kout);
245 failed_zinit:
246 	kunmap_local(kin);
247 	/* 4. push back DEFLATE stream context to the global list */
248 	spin_lock(&z_erofs_deflate_lock);
249 	strm->next = z_erofs_deflate_head;
250 	z_erofs_deflate_head = strm;
251 	spin_unlock(&z_erofs_deflate_lock);
252 	wake_up(&z_erofs_deflate_wq);
253 	return err;
254 }
255