xref: /openbmc/linux/fs/erofs/inode.c (revision 5d7800d9)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2017-2018 HUAWEI, Inc.
4  *             https://www.huawei.com/
5  * Copyright (C) 2021, Alibaba Cloud
6  */
7 #include "xattr.h"
8 
9 #include <trace/events/erofs.h>
10 
11 static void *erofs_read_inode(struct erofs_buf *buf,
12 			      struct inode *inode, unsigned int *ofs)
13 {
14 	struct super_block *sb = inode->i_sb;
15 	struct erofs_sb_info *sbi = EROFS_SB(sb);
16 	struct erofs_inode *vi = EROFS_I(inode);
17 	const erofs_off_t inode_loc = erofs_iloc(inode);
18 
19 	erofs_blk_t blkaddr, nblks = 0;
20 	void *kaddr;
21 	struct erofs_inode_compact *dic;
22 	struct erofs_inode_extended *die, *copied = NULL;
23 	unsigned int ifmt;
24 	int err;
25 
26 	blkaddr = erofs_blknr(sb, inode_loc);
27 	*ofs = erofs_blkoff(sb, inode_loc);
28 
29 	kaddr = erofs_read_metabuf(buf, sb, blkaddr, EROFS_KMAP);
30 	if (IS_ERR(kaddr)) {
31 		erofs_err(sb, "failed to get inode (nid: %llu) page, err %ld",
32 			  vi->nid, PTR_ERR(kaddr));
33 		return kaddr;
34 	}
35 
36 	dic = kaddr + *ofs;
37 	ifmt = le16_to_cpu(dic->i_format);
38 
39 	if (ifmt & ~EROFS_I_ALL) {
40 		erofs_err(inode->i_sb, "unsupported i_format %u of nid %llu",
41 			  ifmt, vi->nid);
42 		err = -EOPNOTSUPP;
43 		goto err_out;
44 	}
45 
46 	vi->datalayout = erofs_inode_datalayout(ifmt);
47 	if (vi->datalayout >= EROFS_INODE_DATALAYOUT_MAX) {
48 		erofs_err(inode->i_sb, "unsupported datalayout %u of nid %llu",
49 			  vi->datalayout, vi->nid);
50 		err = -EOPNOTSUPP;
51 		goto err_out;
52 	}
53 
54 	switch (erofs_inode_version(ifmt)) {
55 	case EROFS_INODE_LAYOUT_EXTENDED:
56 		vi->inode_isize = sizeof(struct erofs_inode_extended);
57 		/* check if the extended inode acrosses block boundary */
58 		if (*ofs + vi->inode_isize <= sb->s_blocksize) {
59 			*ofs += vi->inode_isize;
60 			die = (struct erofs_inode_extended *)dic;
61 		} else {
62 			const unsigned int gotten = sb->s_blocksize - *ofs;
63 
64 			copied = kmalloc(vi->inode_isize, GFP_NOFS);
65 			if (!copied) {
66 				err = -ENOMEM;
67 				goto err_out;
68 			}
69 			memcpy(copied, dic, gotten);
70 			kaddr = erofs_read_metabuf(buf, sb, blkaddr + 1,
71 						   EROFS_KMAP);
72 			if (IS_ERR(kaddr)) {
73 				erofs_err(sb, "failed to get inode payload block (nid: %llu), err %ld",
74 					  vi->nid, PTR_ERR(kaddr));
75 				kfree(copied);
76 				return kaddr;
77 			}
78 			*ofs = vi->inode_isize - gotten;
79 			memcpy((u8 *)copied + gotten, kaddr, *ofs);
80 			die = copied;
81 		}
82 		vi->xattr_isize = erofs_xattr_ibody_size(die->i_xattr_icount);
83 
84 		inode->i_mode = le16_to_cpu(die->i_mode);
85 		switch (inode->i_mode & S_IFMT) {
86 		case S_IFREG:
87 		case S_IFDIR:
88 		case S_IFLNK:
89 			vi->raw_blkaddr = le32_to_cpu(die->i_u.raw_blkaddr);
90 			break;
91 		case S_IFCHR:
92 		case S_IFBLK:
93 			inode->i_rdev =
94 				new_decode_dev(le32_to_cpu(die->i_u.rdev));
95 			break;
96 		case S_IFIFO:
97 		case S_IFSOCK:
98 			inode->i_rdev = 0;
99 			break;
100 		default:
101 			goto bogusimode;
102 		}
103 		i_uid_write(inode, le32_to_cpu(die->i_uid));
104 		i_gid_write(inode, le32_to_cpu(die->i_gid));
105 		set_nlink(inode, le32_to_cpu(die->i_nlink));
106 
107 		/* extended inode has its own timestamp */
108 		inode->i_ctime.tv_sec = le64_to_cpu(die->i_mtime);
109 		inode->i_ctime.tv_nsec = le32_to_cpu(die->i_mtime_nsec);
110 
111 		inode->i_size = le64_to_cpu(die->i_size);
112 
113 		/* total blocks for compressed files */
114 		if (erofs_inode_is_data_compressed(vi->datalayout))
115 			nblks = le32_to_cpu(die->i_u.compressed_blocks);
116 		else if (vi->datalayout == EROFS_INODE_CHUNK_BASED)
117 			/* fill chunked inode summary info */
118 			vi->chunkformat = le16_to_cpu(die->i_u.c.format);
119 		kfree(copied);
120 		copied = NULL;
121 		break;
122 	case EROFS_INODE_LAYOUT_COMPACT:
123 		vi->inode_isize = sizeof(struct erofs_inode_compact);
124 		*ofs += vi->inode_isize;
125 		vi->xattr_isize = erofs_xattr_ibody_size(dic->i_xattr_icount);
126 
127 		inode->i_mode = le16_to_cpu(dic->i_mode);
128 		switch (inode->i_mode & S_IFMT) {
129 		case S_IFREG:
130 		case S_IFDIR:
131 		case S_IFLNK:
132 			vi->raw_blkaddr = le32_to_cpu(dic->i_u.raw_blkaddr);
133 			break;
134 		case S_IFCHR:
135 		case S_IFBLK:
136 			inode->i_rdev =
137 				new_decode_dev(le32_to_cpu(dic->i_u.rdev));
138 			break;
139 		case S_IFIFO:
140 		case S_IFSOCK:
141 			inode->i_rdev = 0;
142 			break;
143 		default:
144 			goto bogusimode;
145 		}
146 		i_uid_write(inode, le16_to_cpu(dic->i_uid));
147 		i_gid_write(inode, le16_to_cpu(dic->i_gid));
148 		set_nlink(inode, le16_to_cpu(dic->i_nlink));
149 
150 		/* use build time for compact inodes */
151 		inode->i_ctime.tv_sec = sbi->build_time;
152 		inode->i_ctime.tv_nsec = sbi->build_time_nsec;
153 
154 		inode->i_size = le32_to_cpu(dic->i_size);
155 		if (erofs_inode_is_data_compressed(vi->datalayout))
156 			nblks = le32_to_cpu(dic->i_u.compressed_blocks);
157 		else if (vi->datalayout == EROFS_INODE_CHUNK_BASED)
158 			vi->chunkformat = le16_to_cpu(dic->i_u.c.format);
159 		break;
160 	default:
161 		erofs_err(inode->i_sb,
162 			  "unsupported on-disk inode version %u of nid %llu",
163 			  erofs_inode_version(ifmt), vi->nid);
164 		err = -EOPNOTSUPP;
165 		goto err_out;
166 	}
167 
168 	if (vi->datalayout == EROFS_INODE_CHUNK_BASED) {
169 		if (vi->chunkformat & ~EROFS_CHUNK_FORMAT_ALL) {
170 			erofs_err(inode->i_sb,
171 				  "unsupported chunk format %x of nid %llu",
172 				  vi->chunkformat, vi->nid);
173 			err = -EOPNOTSUPP;
174 			goto err_out;
175 		}
176 		vi->chunkbits = sb->s_blocksize_bits +
177 			(vi->chunkformat & EROFS_CHUNK_FORMAT_BLKBITS_MASK);
178 	}
179 	inode->i_mtime.tv_sec = inode->i_ctime.tv_sec;
180 	inode->i_atime.tv_sec = inode->i_ctime.tv_sec;
181 	inode->i_mtime.tv_nsec = inode->i_ctime.tv_nsec;
182 	inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec;
183 
184 	inode->i_flags &= ~S_DAX;
185 	if (test_opt(&sbi->opt, DAX_ALWAYS) && S_ISREG(inode->i_mode) &&
186 	    (vi->datalayout == EROFS_INODE_FLAT_PLAIN ||
187 	     vi->datalayout == EROFS_INODE_CHUNK_BASED))
188 		inode->i_flags |= S_DAX;
189 
190 	if (!nblks)
191 		/* measure inode.i_blocks as generic filesystems */
192 		inode->i_blocks = round_up(inode->i_size, sb->s_blocksize) >> 9;
193 	else
194 		inode->i_blocks = nblks << (sb->s_blocksize_bits - 9);
195 	return kaddr;
196 
197 bogusimode:
198 	erofs_err(inode->i_sb, "bogus i_mode (%o) @ nid %llu",
199 		  inode->i_mode, vi->nid);
200 	err = -EFSCORRUPTED;
201 err_out:
202 	DBG_BUGON(1);
203 	kfree(copied);
204 	erofs_put_metabuf(buf);
205 	return ERR_PTR(err);
206 }
207 
208 static int erofs_fill_symlink(struct inode *inode, void *kaddr,
209 			      unsigned int m_pofs)
210 {
211 	struct erofs_inode *vi = EROFS_I(inode);
212 	unsigned int bsz = i_blocksize(inode);
213 	char *lnk;
214 
215 	/* if it cannot be handled with fast symlink scheme */
216 	if (vi->datalayout != EROFS_INODE_FLAT_INLINE ||
217 	    inode->i_size >= bsz || inode->i_size < 0) {
218 		inode->i_op = &erofs_symlink_iops;
219 		return 0;
220 	}
221 
222 	lnk = kmalloc(inode->i_size + 1, GFP_KERNEL);
223 	if (!lnk)
224 		return -ENOMEM;
225 
226 	m_pofs += vi->xattr_isize;
227 	/* inline symlink data shouldn't cross block boundary */
228 	if (m_pofs + inode->i_size > bsz) {
229 		kfree(lnk);
230 		erofs_err(inode->i_sb,
231 			  "inline data cross block boundary @ nid %llu",
232 			  vi->nid);
233 		DBG_BUGON(1);
234 		return -EFSCORRUPTED;
235 	}
236 	memcpy(lnk, kaddr + m_pofs, inode->i_size);
237 	lnk[inode->i_size] = '\0';
238 
239 	inode->i_link = lnk;
240 	inode->i_op = &erofs_fast_symlink_iops;
241 	return 0;
242 }
243 
244 static int erofs_fill_inode(struct inode *inode)
245 {
246 	struct erofs_inode *vi = EROFS_I(inode);
247 	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
248 	void *kaddr;
249 	unsigned int ofs;
250 	int err = 0;
251 
252 	trace_erofs_fill_inode(inode);
253 
254 	/* read inode base data from disk */
255 	kaddr = erofs_read_inode(&buf, inode, &ofs);
256 	if (IS_ERR(kaddr))
257 		return PTR_ERR(kaddr);
258 
259 	/* setup the new inode */
260 	switch (inode->i_mode & S_IFMT) {
261 	case S_IFREG:
262 		inode->i_op = &erofs_generic_iops;
263 		if (erofs_inode_is_data_compressed(vi->datalayout))
264 			inode->i_fop = &generic_ro_fops;
265 		else
266 			inode->i_fop = &erofs_file_fops;
267 		break;
268 	case S_IFDIR:
269 		inode->i_op = &erofs_dir_iops;
270 		inode->i_fop = &erofs_dir_fops;
271 		inode_nohighmem(inode);
272 		break;
273 	case S_IFLNK:
274 		err = erofs_fill_symlink(inode, kaddr, ofs);
275 		if (err)
276 			goto out_unlock;
277 		inode_nohighmem(inode);
278 		break;
279 	case S_IFCHR:
280 	case S_IFBLK:
281 	case S_IFIFO:
282 	case S_IFSOCK:
283 		inode->i_op = &erofs_generic_iops;
284 		init_special_inode(inode, inode->i_mode, inode->i_rdev);
285 		goto out_unlock;
286 	default:
287 		err = -EFSCORRUPTED;
288 		goto out_unlock;
289 	}
290 
291 	if (erofs_inode_is_data_compressed(vi->datalayout)) {
292 #ifdef CONFIG_EROFS_FS_ZIP
293 		if (!erofs_is_fscache_mode(inode->i_sb) &&
294 		    inode->i_sb->s_blocksize_bits == PAGE_SHIFT) {
295 			inode->i_mapping->a_ops = &z_erofs_aops;
296 			err = 0;
297 			goto out_unlock;
298 		}
299 #endif
300 		err = -EOPNOTSUPP;
301 		goto out_unlock;
302 	}
303 	inode->i_mapping->a_ops = &erofs_raw_access_aops;
304 	mapping_set_large_folios(inode->i_mapping);
305 #ifdef CONFIG_EROFS_FS_ONDEMAND
306 	if (erofs_is_fscache_mode(inode->i_sb))
307 		inode->i_mapping->a_ops = &erofs_fscache_access_aops;
308 #endif
309 
310 out_unlock:
311 	erofs_put_metabuf(&buf);
312 	return err;
313 }
314 
315 /*
316  * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
317  * so that it will fit.
318  */
319 static ino_t erofs_squash_ino(erofs_nid_t nid)
320 {
321 	ino_t ino = (ino_t)nid;
322 
323 	if (sizeof(ino_t) < sizeof(erofs_nid_t))
324 		ino ^= nid >> (sizeof(erofs_nid_t) - sizeof(ino_t)) * 8;
325 	return ino;
326 }
327 
328 static int erofs_iget5_eq(struct inode *inode, void *opaque)
329 {
330 	return EROFS_I(inode)->nid == *(erofs_nid_t *)opaque;
331 }
332 
333 static int erofs_iget5_set(struct inode *inode, void *opaque)
334 {
335 	const erofs_nid_t nid = *(erofs_nid_t *)opaque;
336 
337 	inode->i_ino = erofs_squash_ino(nid);
338 	EROFS_I(inode)->nid = nid;
339 	return 0;
340 }
341 
342 struct inode *erofs_iget(struct super_block *sb, erofs_nid_t nid)
343 {
344 	struct inode *inode;
345 
346 	inode = iget5_locked(sb, erofs_squash_ino(nid), erofs_iget5_eq,
347 			     erofs_iget5_set, &nid);
348 	if (!inode)
349 		return ERR_PTR(-ENOMEM);
350 
351 	if (inode->i_state & I_NEW) {
352 		int err = erofs_fill_inode(inode);
353 
354 		if (err) {
355 			iget_failed(inode);
356 			return ERR_PTR(err);
357 		}
358 		unlock_new_inode(inode);
359 	}
360 	return inode;
361 }
362 
363 int erofs_getattr(struct mnt_idmap *idmap, const struct path *path,
364 		  struct kstat *stat, u32 request_mask,
365 		  unsigned int query_flags)
366 {
367 	struct inode *const inode = d_inode(path->dentry);
368 
369 	if (erofs_inode_is_data_compressed(EROFS_I(inode)->datalayout))
370 		stat->attributes |= STATX_ATTR_COMPRESSED;
371 
372 	stat->attributes |= STATX_ATTR_IMMUTABLE;
373 	stat->attributes_mask |= (STATX_ATTR_COMPRESSED |
374 				  STATX_ATTR_IMMUTABLE);
375 
376 	generic_fillattr(idmap, inode, stat);
377 	return 0;
378 }
379 
380 const struct inode_operations erofs_generic_iops = {
381 	.getattr = erofs_getattr,
382 	.listxattr = erofs_listxattr,
383 	.get_inode_acl = erofs_get_acl,
384 	.fiemap = erofs_fiemap,
385 };
386 
387 const struct inode_operations erofs_symlink_iops = {
388 	.get_link = page_get_link,
389 	.getattr = erofs_getattr,
390 	.listxattr = erofs_listxattr,
391 	.get_inode_acl = erofs_get_acl,
392 };
393 
394 const struct inode_operations erofs_fast_symlink_iops = {
395 	.get_link = simple_get_link,
396 	.getattr = erofs_getattr,
397 	.listxattr = erofs_listxattr,
398 	.get_inode_acl = erofs_get_acl,
399 };
400