xref: /openbmc/linux/fs/erofs/inode.c (revision c9933d49)
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 /*
12  * if inode is successfully read, return its inode page (or sometimes
13  * the inode payload page if it's an extended inode) in order to fill
14  * inline data if possible.
15  */
16 static void *erofs_read_inode(struct erofs_buf *buf,
17 			      struct inode *inode, unsigned int *ofs)
18 {
19 	struct super_block *sb = inode->i_sb;
20 	struct erofs_sb_info *sbi = EROFS_SB(sb);
21 	struct erofs_inode *vi = EROFS_I(inode);
22 	const erofs_off_t inode_loc = iloc(sbi, vi->nid);
23 
24 	erofs_blk_t blkaddr, nblks = 0;
25 	void *kaddr;
26 	struct erofs_inode_compact *dic;
27 	struct erofs_inode_extended *die, *copied = NULL;
28 	unsigned int ifmt;
29 	int err;
30 
31 	blkaddr = erofs_blknr(inode_loc);
32 	*ofs = erofs_blkoff(inode_loc);
33 
34 	erofs_dbg("%s, reading inode nid %llu at %u of blkaddr %u",
35 		  __func__, vi->nid, *ofs, blkaddr);
36 
37 	kaddr = erofs_read_metabuf(buf, sb, blkaddr, EROFS_KMAP);
38 	if (IS_ERR(kaddr)) {
39 		erofs_err(sb, "failed to get inode (nid: %llu) page, err %ld",
40 			  vi->nid, PTR_ERR(kaddr));
41 		return kaddr;
42 	}
43 
44 	dic = kaddr + *ofs;
45 	ifmt = le16_to_cpu(dic->i_format);
46 
47 	if (ifmt & ~EROFS_I_ALL) {
48 		erofs_err(inode->i_sb, "unsupported i_format %u of nid %llu",
49 			  ifmt, vi->nid);
50 		err = -EOPNOTSUPP;
51 		goto err_out;
52 	}
53 
54 	vi->datalayout = erofs_inode_datalayout(ifmt);
55 	if (vi->datalayout >= EROFS_INODE_DATALAYOUT_MAX) {
56 		erofs_err(inode->i_sb, "unsupported datalayout %u of nid %llu",
57 			  vi->datalayout, vi->nid);
58 		err = -EOPNOTSUPP;
59 		goto err_out;
60 	}
61 
62 	switch (erofs_inode_version(ifmt)) {
63 	case EROFS_INODE_LAYOUT_EXTENDED:
64 		vi->inode_isize = sizeof(struct erofs_inode_extended);
65 		/* check if the extended inode acrosses block boundary */
66 		if (*ofs + vi->inode_isize <= EROFS_BLKSIZ) {
67 			*ofs += vi->inode_isize;
68 			die = (struct erofs_inode_extended *)dic;
69 		} else {
70 			const unsigned int gotten = EROFS_BLKSIZ - *ofs;
71 
72 			copied = kmalloc(vi->inode_isize, GFP_NOFS);
73 			if (!copied) {
74 				err = -ENOMEM;
75 				goto err_out;
76 			}
77 			memcpy(copied, dic, gotten);
78 			kaddr = erofs_read_metabuf(buf, sb, blkaddr + 1,
79 						   EROFS_KMAP);
80 			if (IS_ERR(kaddr)) {
81 				erofs_err(sb, "failed to get inode payload block (nid: %llu), err %ld",
82 					  vi->nid, PTR_ERR(kaddr));
83 				kfree(copied);
84 				return kaddr;
85 			}
86 			*ofs = vi->inode_isize - gotten;
87 			memcpy((u8 *)copied + gotten, kaddr, *ofs);
88 			die = copied;
89 		}
90 		vi->xattr_isize = erofs_xattr_ibody_size(die->i_xattr_icount);
91 
92 		inode->i_mode = le16_to_cpu(die->i_mode);
93 		switch (inode->i_mode & S_IFMT) {
94 		case S_IFREG:
95 		case S_IFDIR:
96 		case S_IFLNK:
97 			vi->raw_blkaddr = le32_to_cpu(die->i_u.raw_blkaddr);
98 			break;
99 		case S_IFCHR:
100 		case S_IFBLK:
101 			inode->i_rdev =
102 				new_decode_dev(le32_to_cpu(die->i_u.rdev));
103 			break;
104 		case S_IFIFO:
105 		case S_IFSOCK:
106 			inode->i_rdev = 0;
107 			break;
108 		default:
109 			goto bogusimode;
110 		}
111 		i_uid_write(inode, le32_to_cpu(die->i_uid));
112 		i_gid_write(inode, le32_to_cpu(die->i_gid));
113 		set_nlink(inode, le32_to_cpu(die->i_nlink));
114 
115 		/* extended inode has its own timestamp */
116 		inode->i_ctime.tv_sec = le64_to_cpu(die->i_mtime);
117 		inode->i_ctime.tv_nsec = le32_to_cpu(die->i_mtime_nsec);
118 
119 		inode->i_size = le64_to_cpu(die->i_size);
120 
121 		/* total blocks for compressed files */
122 		if (erofs_inode_is_data_compressed(vi->datalayout))
123 			nblks = le32_to_cpu(die->i_u.compressed_blocks);
124 		else if (vi->datalayout == EROFS_INODE_CHUNK_BASED)
125 			/* fill chunked inode summary info */
126 			vi->chunkformat = le16_to_cpu(die->i_u.c.format);
127 		kfree(copied);
128 		copied = NULL;
129 		break;
130 	case EROFS_INODE_LAYOUT_COMPACT:
131 		vi->inode_isize = sizeof(struct erofs_inode_compact);
132 		*ofs += vi->inode_isize;
133 		vi->xattr_isize = erofs_xattr_ibody_size(dic->i_xattr_icount);
134 
135 		inode->i_mode = le16_to_cpu(dic->i_mode);
136 		switch (inode->i_mode & S_IFMT) {
137 		case S_IFREG:
138 		case S_IFDIR:
139 		case S_IFLNK:
140 			vi->raw_blkaddr = le32_to_cpu(dic->i_u.raw_blkaddr);
141 			break;
142 		case S_IFCHR:
143 		case S_IFBLK:
144 			inode->i_rdev =
145 				new_decode_dev(le32_to_cpu(dic->i_u.rdev));
146 			break;
147 		case S_IFIFO:
148 		case S_IFSOCK:
149 			inode->i_rdev = 0;
150 			break;
151 		default:
152 			goto bogusimode;
153 		}
154 		i_uid_write(inode, le16_to_cpu(dic->i_uid));
155 		i_gid_write(inode, le16_to_cpu(dic->i_gid));
156 		set_nlink(inode, le16_to_cpu(dic->i_nlink));
157 
158 		/* use build time for compact inodes */
159 		inode->i_ctime.tv_sec = sbi->build_time;
160 		inode->i_ctime.tv_nsec = sbi->build_time_nsec;
161 
162 		inode->i_size = le32_to_cpu(dic->i_size);
163 		if (erofs_inode_is_data_compressed(vi->datalayout))
164 			nblks = le32_to_cpu(dic->i_u.compressed_blocks);
165 		else if (vi->datalayout == EROFS_INODE_CHUNK_BASED)
166 			vi->chunkformat = le16_to_cpu(dic->i_u.c.format);
167 		break;
168 	default:
169 		erofs_err(inode->i_sb,
170 			  "unsupported on-disk inode version %u of nid %llu",
171 			  erofs_inode_version(ifmt), vi->nid);
172 		err = -EOPNOTSUPP;
173 		goto err_out;
174 	}
175 
176 	if (vi->datalayout == EROFS_INODE_CHUNK_BASED) {
177 		if (vi->chunkformat & ~EROFS_CHUNK_FORMAT_ALL) {
178 			erofs_err(inode->i_sb,
179 				  "unsupported chunk format %x of nid %llu",
180 				  vi->chunkformat, vi->nid);
181 			err = -EOPNOTSUPP;
182 			goto err_out;
183 		}
184 		vi->chunkbits = LOG_BLOCK_SIZE +
185 			(vi->chunkformat & EROFS_CHUNK_FORMAT_BLKBITS_MASK);
186 	}
187 	inode->i_mtime.tv_sec = inode->i_ctime.tv_sec;
188 	inode->i_atime.tv_sec = inode->i_ctime.tv_sec;
189 	inode->i_mtime.tv_nsec = inode->i_ctime.tv_nsec;
190 	inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec;
191 
192 	inode->i_flags &= ~S_DAX;
193 	if (test_opt(&sbi->opt, DAX_ALWAYS) && S_ISREG(inode->i_mode) &&
194 	    vi->datalayout == EROFS_INODE_FLAT_PLAIN)
195 		inode->i_flags |= S_DAX;
196 	if (!nblks)
197 		/* measure inode.i_blocks as generic filesystems */
198 		inode->i_blocks = roundup(inode->i_size, EROFS_BLKSIZ) >> 9;
199 	else
200 		inode->i_blocks = nblks << LOG_SECTORS_PER_BLOCK;
201 	return kaddr;
202 
203 bogusimode:
204 	erofs_err(inode->i_sb, "bogus i_mode (%o) @ nid %llu",
205 		  inode->i_mode, vi->nid);
206 	err = -EFSCORRUPTED;
207 err_out:
208 	DBG_BUGON(1);
209 	kfree(copied);
210 	erofs_put_metabuf(buf);
211 	return ERR_PTR(err);
212 }
213 
214 static int erofs_fill_symlink(struct inode *inode, void *kaddr,
215 			      unsigned int m_pofs)
216 {
217 	struct erofs_inode *vi = EROFS_I(inode);
218 	char *lnk;
219 
220 	/* if it cannot be handled with fast symlink scheme */
221 	if (vi->datalayout != EROFS_INODE_FLAT_INLINE ||
222 	    inode->i_size >= EROFS_BLKSIZ) {
223 		inode->i_op = &erofs_symlink_iops;
224 		return 0;
225 	}
226 
227 	lnk = kmalloc(inode->i_size + 1, GFP_KERNEL);
228 	if (!lnk)
229 		return -ENOMEM;
230 
231 	m_pofs += vi->xattr_isize;
232 	/* inline symlink data shouldn't cross block boundary */
233 	if (m_pofs + inode->i_size > EROFS_BLKSIZ) {
234 		kfree(lnk);
235 		erofs_err(inode->i_sb,
236 			  "inline data cross block boundary @ nid %llu",
237 			  vi->nid);
238 		DBG_BUGON(1);
239 		return -EFSCORRUPTED;
240 	}
241 	memcpy(lnk, kaddr + m_pofs, inode->i_size);
242 	lnk[inode->i_size] = '\0';
243 
244 	inode->i_link = lnk;
245 	inode->i_op = &erofs_fast_symlink_iops;
246 	return 0;
247 }
248 
249 static int erofs_fill_inode(struct inode *inode, int isdir)
250 {
251 	struct erofs_inode *vi = EROFS_I(inode);
252 	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
253 	void *kaddr;
254 	unsigned int ofs;
255 	int err = 0;
256 
257 	trace_erofs_fill_inode(inode, isdir);
258 
259 	/* read inode base data from disk */
260 	kaddr = erofs_read_inode(&buf, inode, &ofs);
261 	if (IS_ERR(kaddr))
262 		return PTR_ERR(kaddr);
263 
264 	/* setup the new inode */
265 	switch (inode->i_mode & S_IFMT) {
266 	case S_IFREG:
267 		inode->i_op = &erofs_generic_iops;
268 		if (erofs_inode_is_data_compressed(vi->datalayout))
269 			inode->i_fop = &generic_ro_fops;
270 		else
271 			inode->i_fop = &erofs_file_fops;
272 		break;
273 	case S_IFDIR:
274 		inode->i_op = &erofs_dir_iops;
275 		inode->i_fop = &erofs_dir_fops;
276 		break;
277 	case S_IFLNK:
278 		err = erofs_fill_symlink(inode, kaddr, ofs);
279 		if (err)
280 			goto out_unlock;
281 		inode_nohighmem(inode);
282 		break;
283 	case S_IFCHR:
284 	case S_IFBLK:
285 	case S_IFIFO:
286 	case S_IFSOCK:
287 		inode->i_op = &erofs_generic_iops;
288 		init_special_inode(inode, inode->i_mode, inode->i_rdev);
289 		goto out_unlock;
290 	default:
291 		err = -EFSCORRUPTED;
292 		goto out_unlock;
293 	}
294 
295 	if (erofs_inode_is_data_compressed(vi->datalayout)) {
296 		err = z_erofs_fill_inode(inode);
297 		goto out_unlock;
298 	}
299 	inode->i_mapping->a_ops = &erofs_raw_access_aops;
300 
301 out_unlock:
302 	erofs_put_metabuf(&buf);
303 	return err;
304 }
305 
306 /*
307  * erofs nid is 64bits, but i_ino is 'unsigned long', therefore
308  * we should do more for 32-bit platform to find the right inode.
309  */
310 static int erofs_ilookup_test_actor(struct inode *inode, void *opaque)
311 {
312 	const erofs_nid_t nid = *(erofs_nid_t *)opaque;
313 
314 	return EROFS_I(inode)->nid == nid;
315 }
316 
317 static int erofs_iget_set_actor(struct inode *inode, void *opaque)
318 {
319 	const erofs_nid_t nid = *(erofs_nid_t *)opaque;
320 
321 	inode->i_ino = erofs_inode_hash(nid);
322 	return 0;
323 }
324 
325 static inline struct inode *erofs_iget_locked(struct super_block *sb,
326 					      erofs_nid_t nid)
327 {
328 	const unsigned long hashval = erofs_inode_hash(nid);
329 
330 	return iget5_locked(sb, hashval, erofs_ilookup_test_actor,
331 		erofs_iget_set_actor, &nid);
332 }
333 
334 struct inode *erofs_iget(struct super_block *sb,
335 			 erofs_nid_t nid,
336 			 bool isdir)
337 {
338 	struct inode *inode = erofs_iget_locked(sb, nid);
339 
340 	if (!inode)
341 		return ERR_PTR(-ENOMEM);
342 
343 	if (inode->i_state & I_NEW) {
344 		int err;
345 		struct erofs_inode *vi = EROFS_I(inode);
346 
347 		vi->nid = nid;
348 
349 		err = erofs_fill_inode(inode, isdir);
350 		if (!err)
351 			unlock_new_inode(inode);
352 		else {
353 			iget_failed(inode);
354 			inode = ERR_PTR(err);
355 		}
356 	}
357 	return inode;
358 }
359 
360 int erofs_getattr(struct user_namespace *mnt_userns, const struct path *path,
361 		  struct kstat *stat, u32 request_mask,
362 		  unsigned int query_flags)
363 {
364 	struct inode *const inode = d_inode(path->dentry);
365 
366 	if (erofs_inode_is_data_compressed(EROFS_I(inode)->datalayout))
367 		stat->attributes |= STATX_ATTR_COMPRESSED;
368 
369 	stat->attributes |= STATX_ATTR_IMMUTABLE;
370 	stat->attributes_mask |= (STATX_ATTR_COMPRESSED |
371 				  STATX_ATTR_IMMUTABLE);
372 
373 	generic_fillattr(&init_user_ns, inode, stat);
374 	return 0;
375 }
376 
377 const struct inode_operations erofs_generic_iops = {
378 	.getattr = erofs_getattr,
379 	.listxattr = erofs_listxattr,
380 	.get_acl = erofs_get_acl,
381 	.fiemap = erofs_fiemap,
382 };
383 
384 const struct inode_operations erofs_symlink_iops = {
385 	.get_link = page_get_link,
386 	.getattr = erofs_getattr,
387 	.listxattr = erofs_listxattr,
388 	.get_acl = erofs_get_acl,
389 };
390 
391 const struct inode_operations erofs_fast_symlink_iops = {
392 	.get_link = simple_get_link,
393 	.getattr = erofs_getattr,
394 	.listxattr = erofs_listxattr,
395 	.get_acl = erofs_get_acl,
396 };
397