inode.c

  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 <linux/compat.h>
  9#include <trace/events/erofs.h>
 10
 11static int erofs_fill_symlink(struct inode *inode, void *kaddr,
 12			      unsigned int m_pofs)
 13{
 14	struct erofs_inode *vi = EROFS_I(inode);
 15	loff_t off;
 16
 17	m_pofs += vi->xattr_isize;
 18	/* check if it cannot be handled with fast symlink scheme */
 19	if (vi->datalayout != EROFS_INODE_FLAT_INLINE ||
 20	    check_add_overflow(m_pofs, inode->i_size, &off) ||
 21	    off > i_blocksize(inode))
 22		return 0;
 23
 24	inode->i_link = kmemdup_nul(kaddr + m_pofs, inode->i_size, GFP_KERNEL);
 25	return inode->i_link ? 0 : -ENOMEM;
 26}
 27
 28static int erofs_read_inode(struct inode *inode)
 29{
 30	struct super_block *sb = inode->i_sb;
 31	erofs_blk_t blkaddr = erofs_blknr(sb, erofs_iloc(inode));
 32	unsigned int ofs = erofs_blkoff(sb, erofs_iloc(inode));
 33	bool in_mbox = erofs_inode_in_metabox(inode);
 34	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
 35	struct erofs_sb_info *sbi = EROFS_SB(sb);
 36	erofs_blk_t addrmask = BIT_ULL(48) - 1;
 37	struct erofs_inode *vi = EROFS_I(inode);
 38	struct erofs_inode_extended *die, copied;
 39	struct erofs_inode_compact *dic;
 40	unsigned int ifmt;
 41	void *ptr;
 42	int err = 0;
 43
 44	ptr = erofs_read_metabuf(&buf, sb, erofs_pos(sb, blkaddr), in_mbox);
 45	if (IS_ERR(ptr)) {
 46		err = PTR_ERR(ptr);
 47		erofs_err(sb, "failed to read inode meta block (nid: %llu): %d",
 48			  vi->nid, err);
 49		goto err_out;
 50	}
 51
 52	dic = ptr + ofs;
 53	ifmt = le16_to_cpu(dic->i_format);
 54	if (ifmt & ~EROFS_I_ALL) {
 55		erofs_err(sb, "unsupported i_format %u of nid %llu",
 56			  ifmt, vi->nid);
 57		err = -EOPNOTSUPP;
 58		goto err_out;
 59	}
 60
 61	vi->datalayout = erofs_inode_datalayout(ifmt);
 62	if (vi->datalayout >= EROFS_INODE_DATALAYOUT_MAX) {
 63		erofs_err(sb, "unsupported datalayout %u of nid %llu",
 64			  vi->datalayout, vi->nid);
 65		err = -EOPNOTSUPP;
 66		goto err_out;
 67	}
 68
 69	switch (erofs_inode_version(ifmt)) {
 70	case EROFS_INODE_LAYOUT_EXTENDED:
 71		vi->inode_isize = sizeof(struct erofs_inode_extended);
 72		/* check if the extended inode acrosses block boundary */
 73		if (ofs + vi->inode_isize <= sb->s_blocksize) {
 74			ofs += vi->inode_isize;
 75			die = (struct erofs_inode_extended *)dic;
 76			copied.i_u = die->i_u;
 77			copied.i_nb = die->i_nb;
 78		} else {
 79			const unsigned int gotten = sb->s_blocksize - ofs;
 80
 81			memcpy(&copied, dic, gotten);
 82			ptr = erofs_read_metabuf(&buf, sb,
 83					erofs_pos(sb, blkaddr + 1), in_mbox);
 84			if (IS_ERR(ptr)) {
 85				err = PTR_ERR(ptr);
 86				erofs_err(sb, "failed to read inode payload block (nid: %llu): %d",
 87					  vi->nid, err);
 88				goto err_out;
 89			}
 90			ofs = vi->inode_isize - gotten;
 91			memcpy((u8 *)&copied + gotten, ptr, ofs);
 92			die = &copied;
 93		}
 94		vi->xattr_isize = erofs_xattr_ibody_size(die->i_xattr_icount);
 95
 96		inode->i_mode = le16_to_cpu(die->i_mode);
 97		i_uid_write(inode, le32_to_cpu(die->i_uid));
 98		i_gid_write(inode, le32_to_cpu(die->i_gid));
 99		set_nlink(inode, le32_to_cpu(die->i_nlink));
100		inode_set_mtime(inode, le64_to_cpu(die->i_mtime),
101				le32_to_cpu(die->i_mtime_nsec));
102
103		inode->i_size = le64_to_cpu(die->i_size);
104		break;
105	case EROFS_INODE_LAYOUT_COMPACT:
106		vi->inode_isize = sizeof(struct erofs_inode_compact);
107		ofs += vi->inode_isize;
108		vi->xattr_isize = erofs_xattr_ibody_size(dic->i_xattr_icount);
109
110		inode->i_mode = le16_to_cpu(dic->i_mode);
111		copied.i_u = dic->i_u;
112		i_uid_write(inode, le16_to_cpu(dic->i_uid));
113		i_gid_write(inode, le16_to_cpu(dic->i_gid));
114		if (!S_ISDIR(inode->i_mode) &&
115		    ((ifmt >> EROFS_I_NLINK_1_BIT) & 1)) {
116			set_nlink(inode, 1);
117			copied.i_nb = dic->i_nb;
118		} else {
119			set_nlink(inode, le16_to_cpu(dic->i_nb.nlink));
120			copied.i_nb.startblk_hi = 0;
121			addrmask = BIT_ULL(32) - 1;
122		}
123		inode_set_mtime(inode, sbi->epoch + le32_to_cpu(dic->i_mtime),
124				sbi->fixed_nsec);
125
126		inode->i_size = le32_to_cpu(dic->i_size);
127		break;
128	default:
129		erofs_err(sb, "unsupported on-disk inode version %u of nid %llu",
130			  erofs_inode_version(ifmt), vi->nid);
131		err = -EOPNOTSUPP;
132		goto err_out;
133	}
134
135	if (unlikely(inode->i_size < 0)) {
136		erofs_err(sb, "negative i_size @ nid %llu", vi->nid);
137		err = -EFSCORRUPTED;
138		goto err_out;
139	}
140	switch (inode->i_mode & S_IFMT) {
141	case S_IFDIR:
142		vi->dot_omitted = (ifmt >> EROFS_I_DOT_OMITTED_BIT) & 1;
143		fallthrough;
144	case S_IFREG:
145	case S_IFLNK:
146		vi->startblk = le32_to_cpu(copied.i_u.startblk_lo) |
147			((u64)le16_to_cpu(copied.i_nb.startblk_hi) << 32);
148		if (vi->datalayout == EROFS_INODE_FLAT_PLAIN &&
149		    !((vi->startblk ^ EROFS_NULL_ADDR) & addrmask))
150			vi->startblk = EROFS_NULL_ADDR;
151
152		if(S_ISLNK(inode->i_mode)) {
153			err = erofs_fill_symlink(inode, ptr, ofs);
154			if (err)
155				goto err_out;
156		}
157		break;
158	case S_IFCHR:
159	case S_IFBLK:
160		inode->i_rdev = new_decode_dev(le32_to_cpu(copied.i_u.rdev));
161		break;
162	case S_IFIFO:
163	case S_IFSOCK:
164		inode->i_rdev = 0;
165		break;
166	default:
167		erofs_err(sb, "bogus i_mode (%o) @ nid %llu", inode->i_mode,
168			  vi->nid);
169		err = -EFSCORRUPTED;
170		goto err_out;
171	}
172
173	if (erofs_inode_is_data_compressed(vi->datalayout))
174		inode->i_blocks = le32_to_cpu(copied.i_u.blocks_lo) <<
175					(sb->s_blocksize_bits - 9);
176	else
177		inode->i_blocks = round_up(inode->i_size, sb->s_blocksize) >> 9;
178
179	if (vi->datalayout == EROFS_INODE_CHUNK_BASED) {
180		/* fill chunked inode summary info */
181		vi->chunkformat = le16_to_cpu(copied.i_u.c.format);
182		if (vi->chunkformat & ~EROFS_CHUNK_FORMAT_ALL) {
183			erofs_err(sb, "unsupported chunk format %x of nid %llu",
184				  vi->chunkformat, vi->nid);
185			err = -EOPNOTSUPP;
186			goto err_out;
187		}
188		vi->chunkbits = sb->s_blocksize_bits +
189			(vi->chunkformat & EROFS_CHUNK_FORMAT_BLKBITS_MASK);
190	}
191	inode_set_atime_to_ts(inode,
192			      inode_set_ctime_to_ts(inode, inode_get_mtime(inode)));
193
194	inode->i_flags &= ~S_DAX;
195	if (test_opt(&sbi->opt, DAX_ALWAYS) && S_ISREG(inode->i_mode) &&
196	    (vi->datalayout == EROFS_INODE_FLAT_PLAIN ||
197	     vi->datalayout == EROFS_INODE_CHUNK_BASED))
198		inode->i_flags |= S_DAX;
199err_out:
200	erofs_put_metabuf(&buf);
201	return err;
202}
203
204static int erofs_fill_inode(struct inode *inode)
205{
206	struct erofs_inode *vi = EROFS_I(inode);
207	int err;
208
209	trace_erofs_fill_inode(inode);
210	err = erofs_read_inode(inode);
211	if (err)
212		return err;
213
214	switch (inode->i_mode & S_IFMT) {
215	case S_IFREG:
216		inode->i_op = &erofs_generic_iops;
217		inode->i_fop = &erofs_file_fops;
218		break;
219	case S_IFDIR:
220		inode->i_op = &erofs_dir_iops;
221		inode->i_fop = &erofs_dir_fops;
222		inode_nohighmem(inode);
223		break;
224	case S_IFLNK:
225		if (inode->i_link)
226			inode->i_op = &erofs_fast_symlink_iops;
227		else
228			inode->i_op = &erofs_symlink_iops;
229		inode_nohighmem(inode);
230		break;
231	default:
232		inode->i_op = &erofs_generic_iops;
233		init_special_inode(inode, inode->i_mode, inode->i_rdev);
234		return 0;
235	}
236
237	mapping_set_large_folios(inode->i_mapping);
238	if (erofs_inode_is_data_compressed(vi->datalayout)) {
239#ifdef CONFIG_EROFS_FS_ZIP
240		DO_ONCE_LITE_IF(inode->i_blkbits != PAGE_SHIFT,
241			  erofs_info, inode->i_sb,
242			  "EXPERIMENTAL EROFS subpage compressed block support in use. Use at your own risk!");
243		inode->i_mapping->a_ops = &z_erofs_aops;
244#else
245		err = -EOPNOTSUPP;
246#endif
247	} else {
248		inode->i_mapping->a_ops = &erofs_aops;
249#ifdef CONFIG_EROFS_FS_ONDEMAND
250		if (erofs_is_fscache_mode(inode->i_sb))
251			inode->i_mapping->a_ops = &erofs_fscache_access_aops;
252#endif
253#ifdef CONFIG_EROFS_FS_BACKED_BY_FILE
254		if (erofs_is_fileio_mode(EROFS_SB(inode->i_sb)))
255			inode->i_mapping->a_ops = &erofs_fileio_aops;
256#endif
257	}
258
259	return err;
260}
261
262/*
263 * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
264 * so that it will fit.
265 */
266static ino_t erofs_squash_ino(struct super_block *sb, erofs_nid_t nid)
267{
268	u64 ino64 = erofs_nid_to_ino64(EROFS_SB(sb), nid);
269
270	if (sizeof(ino_t) < sizeof(erofs_nid_t))
271		ino64 ^= ino64 >> (sizeof(erofs_nid_t) - sizeof(ino_t)) * 8;
272	return (ino_t)ino64;
273}
274
275static int erofs_iget5_eq(struct inode *inode, void *opaque)
276{
277	return EROFS_I(inode)->nid == *(erofs_nid_t *)opaque;
278}
279
280static int erofs_iget5_set(struct inode *inode, void *opaque)
281{
282	const erofs_nid_t nid = *(erofs_nid_t *)opaque;
283
284	inode->i_ino = erofs_squash_ino(inode->i_sb, nid);
285	EROFS_I(inode)->nid = nid;
286	return 0;
287}
288
289struct inode *erofs_iget(struct super_block *sb, erofs_nid_t nid)
290{
291	struct inode *inode;
292
293	inode = iget5_locked(sb, erofs_squash_ino(sb, nid), erofs_iget5_eq,
294			     erofs_iget5_set, &nid);
295	if (!inode)
296		return ERR_PTR(-ENOMEM);
297
298	if (inode_state_read_once(inode) & I_NEW) {
299		int err = erofs_fill_inode(inode);
300
301		if (err) {
302			iget_failed(inode);
303			return ERR_PTR(err);
304		}
305		unlock_new_inode(inode);
306	}
307	return inode;
308}
309
310int erofs_getattr(struct mnt_idmap *idmap, const struct path *path,
311		  struct kstat *stat, u32 request_mask,
312		  unsigned int query_flags)
313{
314	struct inode *const inode = d_inode(path->dentry);
315	struct block_device *bdev = inode->i_sb->s_bdev;
316	bool compressed =
317		erofs_inode_is_data_compressed(EROFS_I(inode)->datalayout);
318
319	if (compressed)
320		stat->attributes |= STATX_ATTR_COMPRESSED;
321	stat->attributes |= STATX_ATTR_IMMUTABLE;
322	stat->attributes_mask |= (STATX_ATTR_COMPRESSED |
323				  STATX_ATTR_IMMUTABLE);
324
325	/*
326	 * Return the DIO alignment restrictions if requested.
327	 *
328	 * In EROFS, STATX_DIOALIGN is only supported in bdev-based mode
329	 * and uncompressed inodes, otherwise we report no DIO support.
330	 */
331	if ((request_mask & STATX_DIOALIGN) && S_ISREG(inode->i_mode)) {
332		stat->result_mask |= STATX_DIOALIGN;
333		if (bdev && !compressed) {
334			stat->dio_mem_align = bdev_dma_alignment(bdev) + 1;
335			stat->dio_offset_align = bdev_logical_block_size(bdev);
336		}
337	}
338	generic_fillattr(idmap, request_mask, inode, stat);
339	return 0;
340}
341
342static int erofs_ioctl_get_volume_label(struct inode *inode, void __user *arg)
343{
344	struct erofs_sb_info *sbi = EROFS_I_SB(inode);
345	int ret;
346
347	if (!sbi->volume_name)
348		ret = clear_user(arg, 1);
349	else
350		ret = copy_to_user(arg, sbi->volume_name,
351				   strlen(sbi->volume_name));
352	return ret ? -EFAULT : 0;
353}
354
355long erofs_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
356{
357	struct inode *inode = file_inode(filp);
358	void __user *argp = (void __user *)arg;
359
360	switch (cmd) {
361	case FS_IOC_GETFSLABEL:
362		return erofs_ioctl_get_volume_label(inode, argp);
363	default:
364		return -ENOTTY;
365	}
366}
367
368#ifdef CONFIG_COMPAT
369long erofs_compat_ioctl(struct file *filp, unsigned int cmd,
370			unsigned long arg)
371{
372	return erofs_ioctl(filp, cmd, (unsigned long)compat_ptr(arg));
373}
374#endif
375
376const struct inode_operations erofs_generic_iops = {
377	.getattr = erofs_getattr,
378	.listxattr = erofs_listxattr,
379	.get_inode_acl = erofs_get_acl,
380	.fiemap = erofs_fiemap,
381};
382
383const struct inode_operations erofs_symlink_iops = {
384	.get_link = page_get_link,
385	.getattr = erofs_getattr,
386	.listxattr = erofs_listxattr,
387	.get_inode_acl = erofs_get_acl,
388};
389
390const struct inode_operations erofs_fast_symlink_iops = {
391	.get_link = simple_get_link,
392	.getattr = erofs_getattr,
393	.listxattr = erofs_listxattr,
394	.get_inode_acl = erofs_get_acl,
395};