xref: /linux/fs/erofs/inode.c (revision 76a9701325d39d8602695b19c49a9d0828c897ca)
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 #include <trace/events/erofs.h>
9 
erofs_fill_symlink(struct inode * inode,void * kaddr,unsigned int m_pofs)10 static int erofs_fill_symlink(struct inode *inode, void *kaddr,
11 			      unsigned int m_pofs)
12 {
13 	struct erofs_inode *vi = EROFS_I(inode);
14 	loff_t off;
15 
16 	m_pofs += vi->xattr_isize;
17 	/* check if it cannot be handled with fast symlink scheme */
18 	if (vi->datalayout != EROFS_INODE_FLAT_INLINE ||
19 	    check_add_overflow(m_pofs, inode->i_size, &off) ||
20 	    off > i_blocksize(inode))
21 		return 0;
22 
23 	inode->i_link = kmemdup_nul(kaddr + m_pofs, inode->i_size, GFP_KERNEL);
24 	return inode->i_link ? 0 : -ENOMEM;
25 }
26 
erofs_read_inode(struct inode * inode)27 static int erofs_read_inode(struct inode *inode)
28 {
29 	struct super_block *sb = inode->i_sb;
30 	erofs_blk_t blkaddr = erofs_blknr(sb, erofs_iloc(inode));
31 	unsigned int ofs = erofs_blkoff(sb, erofs_iloc(inode));
32 	bool in_mbox = erofs_inode_in_metabox(inode);
33 	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
34 	struct erofs_sb_info *sbi = EROFS_SB(sb);
35 	erofs_blk_t addrmask = BIT_ULL(48) - 1;
36 	struct erofs_inode *vi = EROFS_I(inode);
37 	struct erofs_inode_extended *die, copied;
38 	struct erofs_inode_compact *dic;
39 	unsigned int ifmt;
40 	void *ptr;
41 	int err = 0;
42 
43 	ptr = erofs_read_metabuf(&buf, sb, erofs_pos(sb, blkaddr), in_mbox);
44 	if (IS_ERR(ptr)) {
45 		err = PTR_ERR(ptr);
46 		erofs_err(sb, "failed to read inode meta block (nid: %llu): %d",
47 			  vi->nid, err);
48 		goto err_out;
49 	}
50 
51 	dic = ptr + ofs;
52 	ifmt = le16_to_cpu(dic->i_format);
53 	if (ifmt & ~EROFS_I_ALL) {
54 		erofs_err(sb, "unsupported i_format %u of nid %llu",
55 			  ifmt, vi->nid);
56 		err = -EOPNOTSUPP;
57 		goto err_out;
58 	}
59 
60 	vi->datalayout = erofs_inode_datalayout(ifmt);
61 	if (vi->datalayout >= EROFS_INODE_DATALAYOUT_MAX) {
62 		erofs_err(sb, "unsupported datalayout %u of nid %llu",
63 			  vi->datalayout, vi->nid);
64 		err = -EOPNOTSUPP;
65 		goto err_out;
66 	}
67 
68 	switch (erofs_inode_version(ifmt)) {
69 	case EROFS_INODE_LAYOUT_EXTENDED:
70 		vi->inode_isize = sizeof(struct erofs_inode_extended);
71 		/* check if the extended inode acrosses block boundary */
72 		if (ofs + vi->inode_isize <= sb->s_blocksize) {
73 			ofs += vi->inode_isize;
74 			die = (struct erofs_inode_extended *)dic;
75 			copied.i_u = die->i_u;
76 			copied.i_nb = die->i_nb;
77 		} else {
78 			const unsigned int gotten = sb->s_blocksize - ofs;
79 
80 			memcpy(&copied, dic, gotten);
81 			ptr = erofs_read_metabuf(&buf, sb,
82 					erofs_pos(sb, blkaddr + 1), in_mbox);
83 			if (IS_ERR(ptr)) {
84 				err = PTR_ERR(ptr);
85 				erofs_err(sb, "failed to read inode payload block (nid: %llu): %d",
86 					  vi->nid, err);
87 				goto err_out;
88 			}
89 			ofs = vi->inode_isize - gotten;
90 			memcpy((u8 *)&copied + gotten, ptr, ofs);
91 			die = &copied;
92 		}
93 		vi->xattr_isize = erofs_xattr_ibody_size(die->i_xattr_icount);
94 
95 		inode->i_mode = le16_to_cpu(die->i_mode);
96 		i_uid_write(inode, le32_to_cpu(die->i_uid));
97 		i_gid_write(inode, le32_to_cpu(die->i_gid));
98 		set_nlink(inode, le32_to_cpu(die->i_nlink));
99 		inode_set_mtime(inode, le64_to_cpu(die->i_mtime),
100 				le32_to_cpu(die->i_mtime_nsec));
101 
102 		inode->i_size = le64_to_cpu(die->i_size);
103 		break;
104 	case EROFS_INODE_LAYOUT_COMPACT:
105 		vi->inode_isize = sizeof(struct erofs_inode_compact);
106 		ofs += vi->inode_isize;
107 		vi->xattr_isize = erofs_xattr_ibody_size(dic->i_xattr_icount);
108 
109 		inode->i_mode = le16_to_cpu(dic->i_mode);
110 		copied.i_u = dic->i_u;
111 		i_uid_write(inode, le16_to_cpu(dic->i_uid));
112 		i_gid_write(inode, le16_to_cpu(dic->i_gid));
113 		if (!S_ISDIR(inode->i_mode) &&
114 		    ((ifmt >> EROFS_I_NLINK_1_BIT) & 1)) {
115 			set_nlink(inode, 1);
116 			copied.i_nb = dic->i_nb;
117 		} else {
118 			set_nlink(inode, le16_to_cpu(dic->i_nb.nlink));
119 			copied.i_nb.startblk_hi = 0;
120 			addrmask = BIT_ULL(32) - 1;
121 		}
122 		inode_set_mtime(inode, sbi->epoch + le32_to_cpu(dic->i_mtime),
123 				sbi->fixed_nsec);
124 
125 		inode->i_size = le32_to_cpu(dic->i_size);
126 		break;
127 	default:
128 		erofs_err(sb, "unsupported on-disk inode version %u of nid %llu",
129 			  erofs_inode_version(ifmt), vi->nid);
130 		err = -EOPNOTSUPP;
131 		goto err_out;
132 	}
133 
134 	if (unlikely(inode->i_size < 0)) {
135 		erofs_err(sb, "negative i_size @ nid %llu", vi->nid);
136 		err = -EFSCORRUPTED;
137 		goto err_out;
138 	}
139 	switch (inode->i_mode & S_IFMT) {
140 	case S_IFDIR:
141 		vi->dot_omitted = (ifmt >> EROFS_I_DOT_OMITTED_BIT) & 1;
142 		fallthrough;
143 	case S_IFREG:
144 	case S_IFLNK:
145 		vi->startblk = le32_to_cpu(copied.i_u.startblk_lo) |
146 			((u64)le16_to_cpu(copied.i_nb.startblk_hi) << 32);
147 		if (vi->datalayout == EROFS_INODE_FLAT_PLAIN &&
148 		    !((vi->startblk ^ EROFS_NULL_ADDR) & addrmask))
149 			vi->startblk = EROFS_NULL_ADDR;
150 
151 		if(S_ISLNK(inode->i_mode)) {
152 			err = erofs_fill_symlink(inode, ptr, ofs);
153 			if (err)
154 				goto err_out;
155 		}
156 		break;
157 	case S_IFCHR:
158 	case S_IFBLK:
159 		inode->i_rdev = new_decode_dev(le32_to_cpu(copied.i_u.rdev));
160 		break;
161 	case S_IFIFO:
162 	case S_IFSOCK:
163 		inode->i_rdev = 0;
164 		break;
165 	default:
166 		erofs_err(sb, "bogus i_mode (%o) @ nid %llu", inode->i_mode,
167 			  vi->nid);
168 		err = -EFSCORRUPTED;
169 		goto err_out;
170 	}
171 
172 	if (erofs_inode_is_data_compressed(vi->datalayout))
173 		inode->i_blocks = le32_to_cpu(copied.i_u.blocks_lo) <<
174 					(sb->s_blocksize_bits - 9);
175 	else
176 		inode->i_blocks = round_up(inode->i_size, sb->s_blocksize) >> 9;
177 
178 	if (vi->datalayout == EROFS_INODE_CHUNK_BASED) {
179 		/* fill chunked inode summary info */
180 		vi->chunkformat = le16_to_cpu(copied.i_u.c.format);
181 		if (vi->chunkformat & ~EROFS_CHUNK_FORMAT_ALL) {
182 			erofs_err(sb, "unsupported chunk format %x of nid %llu",
183 				  vi->chunkformat, vi->nid);
184 			err = -EOPNOTSUPP;
185 			goto err_out;
186 		}
187 		vi->chunkbits = sb->s_blocksize_bits +
188 			(vi->chunkformat & EROFS_CHUNK_FORMAT_BLKBITS_MASK);
189 	}
190 	inode_set_atime_to_ts(inode,
191 			      inode_set_ctime_to_ts(inode, inode_get_mtime(inode)));
192 
193 	inode->i_flags &= ~S_DAX;
194 	if (test_opt(&sbi->opt, DAX_ALWAYS) && S_ISREG(inode->i_mode) &&
195 	    (vi->datalayout == EROFS_INODE_FLAT_PLAIN ||
196 	     vi->datalayout == EROFS_INODE_CHUNK_BASED))
197 		inode->i_flags |= S_DAX;
198 err_out:
199 	erofs_put_metabuf(&buf);
200 	return err;
201 }
202 
erofs_fill_inode(struct inode * inode)203 static int erofs_fill_inode(struct inode *inode)
204 {
205 	struct erofs_inode *vi = EROFS_I(inode);
206 	int err;
207 
208 	trace_erofs_fill_inode(inode);
209 	err = erofs_read_inode(inode);
210 	if (err)
211 		return err;
212 
213 	switch (inode->i_mode & S_IFMT) {
214 	case S_IFREG:
215 		inode->i_op = &erofs_generic_iops;
216 		if (erofs_inode_is_data_compressed(vi->datalayout))
217 			inode->i_fop = &generic_ro_fops;
218 		else
219 			inode->i_fop = &erofs_file_fops;
220 		break;
221 	case S_IFDIR:
222 		inode->i_op = &erofs_dir_iops;
223 		inode->i_fop = &erofs_dir_fops;
224 		inode_nohighmem(inode);
225 		break;
226 	case S_IFLNK:
227 		if (inode->i_link)
228 			inode->i_op = &erofs_fast_symlink_iops;
229 		else
230 			inode->i_op = &erofs_symlink_iops;
231 		inode_nohighmem(inode);
232 		break;
233 	default:
234 		inode->i_op = &erofs_generic_iops;
235 		init_special_inode(inode, inode->i_mode, inode->i_rdev);
236 		return 0;
237 	}
238 
239 	mapping_set_large_folios(inode->i_mapping);
240 	if (erofs_inode_is_data_compressed(vi->datalayout)) {
241 #ifdef CONFIG_EROFS_FS_ZIP
242 		DO_ONCE_LITE_IF(inode->i_blkbits != PAGE_SHIFT,
243 			  erofs_info, inode->i_sb,
244 			  "EXPERIMENTAL EROFS subpage compressed block support in use. Use at your own risk!");
245 		inode->i_mapping->a_ops = &z_erofs_aops;
246 #else
247 		err = -EOPNOTSUPP;
248 #endif
249 	} else {
250 		inode->i_mapping->a_ops = &erofs_aops;
251 #ifdef CONFIG_EROFS_FS_ONDEMAND
252 		if (erofs_is_fscache_mode(inode->i_sb))
253 			inode->i_mapping->a_ops = &erofs_fscache_access_aops;
254 #endif
255 #ifdef CONFIG_EROFS_FS_BACKED_BY_FILE
256 		if (erofs_is_fileio_mode(EROFS_SB(inode->i_sb)))
257 			inode->i_mapping->a_ops = &erofs_fileio_aops;
258 #endif
259 	}
260 
261 	return err;
262 }
263 
264 /*
265  * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
266  * so that it will fit.
267  */
erofs_squash_ino(struct super_block * sb,erofs_nid_t nid)268 static ino_t erofs_squash_ino(struct super_block *sb, erofs_nid_t nid)
269 {
270 	u64 ino64 = erofs_nid_to_ino64(EROFS_SB(sb), nid);
271 
272 	if (sizeof(ino_t) < sizeof(erofs_nid_t))
273 		ino64 ^= ino64 >> (sizeof(erofs_nid_t) - sizeof(ino_t)) * 8;
274 	return (ino_t)ino64;
275 }
276 
erofs_iget5_eq(struct inode * inode,void * opaque)277 static int erofs_iget5_eq(struct inode *inode, void *opaque)
278 {
279 	return EROFS_I(inode)->nid == *(erofs_nid_t *)opaque;
280 }
281 
erofs_iget5_set(struct inode * inode,void * opaque)282 static int erofs_iget5_set(struct inode *inode, void *opaque)
283 {
284 	const erofs_nid_t nid = *(erofs_nid_t *)opaque;
285 
286 	inode->i_ino = erofs_squash_ino(inode->i_sb, nid);
287 	EROFS_I(inode)->nid = nid;
288 	return 0;
289 }
290 
erofs_iget(struct super_block * sb,erofs_nid_t nid)291 struct inode *erofs_iget(struct super_block *sb, erofs_nid_t nid)
292 {
293 	struct inode *inode;
294 
295 	inode = iget5_locked(sb, erofs_squash_ino(sb, nid), erofs_iget5_eq,
296 			     erofs_iget5_set, &nid);
297 	if (!inode)
298 		return ERR_PTR(-ENOMEM);
299 
300 	if (inode->i_state & I_NEW) {
301 		int err = erofs_fill_inode(inode);
302 
303 		if (err) {
304 			iget_failed(inode);
305 			return ERR_PTR(err);
306 		}
307 		unlock_new_inode(inode);
308 	}
309 	return inode;
310 }
311 
erofs_getattr(struct mnt_idmap * idmap,const struct path * path,struct kstat * stat,u32 request_mask,unsigned int query_flags)312 int erofs_getattr(struct mnt_idmap *idmap, const struct path *path,
313 		  struct kstat *stat, u32 request_mask,
314 		  unsigned int query_flags)
315 {
316 	struct inode *const inode = d_inode(path->dentry);
317 	struct block_device *bdev = inode->i_sb->s_bdev;
318 	bool compressed =
319 		erofs_inode_is_data_compressed(EROFS_I(inode)->datalayout);
320 
321 	if (compressed)
322 		stat->attributes |= STATX_ATTR_COMPRESSED;
323 	stat->attributes |= STATX_ATTR_IMMUTABLE;
324 	stat->attributes_mask |= (STATX_ATTR_COMPRESSED |
325 				  STATX_ATTR_IMMUTABLE);
326 
327 	/*
328 	 * Return the DIO alignment restrictions if requested.
329 	 *
330 	 * In EROFS, STATX_DIOALIGN is only supported in bdev-based mode
331 	 * and uncompressed inodes, otherwise we report no DIO support.
332 	 */
333 	if ((request_mask & STATX_DIOALIGN) && S_ISREG(inode->i_mode)) {
334 		stat->result_mask |= STATX_DIOALIGN;
335 		if (bdev && !compressed) {
336 			stat->dio_mem_align = bdev_dma_alignment(bdev) + 1;
337 			stat->dio_offset_align = bdev_logical_block_size(bdev);
338 		}
339 	}
340 	generic_fillattr(idmap, request_mask, inode, stat);
341 	return 0;
342 }
343 
344 const struct inode_operations erofs_generic_iops = {
345 	.getattr = erofs_getattr,
346 	.listxattr = erofs_listxattr,
347 	.get_inode_acl = erofs_get_acl,
348 	.fiemap = erofs_fiemap,
349 };
350 
351 const struct inode_operations erofs_symlink_iops = {
352 	.get_link = page_get_link,
353 	.getattr = erofs_getattr,
354 	.listxattr = erofs_listxattr,
355 	.get_inode_acl = erofs_get_acl,
356 };
357 
358 const struct inode_operations erofs_fast_symlink_iops = {
359 	.get_link = simple_get_link,
360 	.getattr = erofs_getattr,
361 	.listxattr = erofs_listxattr,
362 	.get_inode_acl = erofs_get_acl,
363 };
364