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 "internal.h"
8 #include <linux/sched/mm.h>
9 #include <trace/events/erofs.h>
10
erofs_unmap_metabuf(struct erofs_buf * buf)11 void erofs_unmap_metabuf(struct erofs_buf *buf)
12 {
13 if (buf->kmap_type == EROFS_KMAP)
14 kunmap_local(buf->base);
15 buf->base = NULL;
16 buf->kmap_type = EROFS_NO_KMAP;
17 }
18
erofs_put_metabuf(struct erofs_buf * buf)19 void erofs_put_metabuf(struct erofs_buf *buf)
20 {
21 if (!buf->page)
22 return;
23 erofs_unmap_metabuf(buf);
24 folio_put(page_folio(buf->page));
25 buf->page = NULL;
26 }
27
erofs_bread(struct erofs_buf * buf,erofs_off_t offset,enum erofs_kmap_type type)28 void *erofs_bread(struct erofs_buf *buf, erofs_off_t offset,
29 enum erofs_kmap_type type)
30 {
31 pgoff_t index = (buf->off + offset) >> PAGE_SHIFT;
32 struct folio *folio = NULL;
33
34 if (buf->page) {
35 folio = page_folio(buf->page);
36 if (folio_file_page(folio, index) != buf->page)
37 erofs_unmap_metabuf(buf);
38 }
39 if (!folio || !folio_contains(folio, index)) {
40 erofs_put_metabuf(buf);
41 folio = read_mapping_folio(buf->mapping, index, buf->file);
42 if (IS_ERR(folio))
43 return folio;
44 }
45 buf->page = folio_file_page(folio, index);
46
47 if (buf->kmap_type == EROFS_NO_KMAP) {
48 if (type == EROFS_KMAP)
49 buf->base = kmap_local_page(buf->page);
50 buf->kmap_type = type;
51 } else if (buf->kmap_type != type) {
52 DBG_BUGON(1);
53 return ERR_PTR(-EFAULT);
54 }
55 if (type == EROFS_NO_KMAP)
56 return NULL;
57 return buf->base + (offset & ~PAGE_MASK);
58 }
59
erofs_init_metabuf(struct erofs_buf * buf,struct super_block * sb)60 void erofs_init_metabuf(struct erofs_buf *buf, struct super_block *sb)
61 {
62 struct erofs_sb_info *sbi = EROFS_SB(sb);
63
64 buf->file = NULL;
65 buf->off = sbi->dif0.fsoff;
66 if (erofs_is_fileio_mode(sbi)) {
67 buf->file = sbi->dif0.file; /* some fs like FUSE needs it */
68 buf->mapping = buf->file->f_mapping;
69 } else if (erofs_is_fscache_mode(sb))
70 buf->mapping = sbi->dif0.fscache->inode->i_mapping;
71 else
72 buf->mapping = sb->s_bdev->bd_mapping;
73 }
74
erofs_read_metabuf(struct erofs_buf * buf,struct super_block * sb,erofs_off_t offset,enum erofs_kmap_type type)75 void *erofs_read_metabuf(struct erofs_buf *buf, struct super_block *sb,
76 erofs_off_t offset, enum erofs_kmap_type type)
77 {
78 erofs_init_metabuf(buf, sb);
79 return erofs_bread(buf, offset, type);
80 }
81
erofs_map_blocks_flatmode(struct inode * inode,struct erofs_map_blocks * map)82 static int erofs_map_blocks_flatmode(struct inode *inode,
83 struct erofs_map_blocks *map)
84 {
85 struct erofs_inode *vi = EROFS_I(inode);
86 struct super_block *sb = inode->i_sb;
87 bool tailendpacking = (vi->datalayout == EROFS_INODE_FLAT_INLINE);
88 erofs_blk_t lastblk = erofs_iblks(inode) - tailendpacking;
89
90 map->m_flags = EROFS_MAP_MAPPED; /* no hole in flat inodes */
91 if (map->m_la < erofs_pos(sb, lastblk)) {
92 map->m_pa = erofs_pos(sb, vi->raw_blkaddr) + map->m_la;
93 map->m_plen = erofs_pos(sb, lastblk) - map->m_la;
94 } else {
95 DBG_BUGON(!tailendpacking);
96 map->m_pa = erofs_iloc(inode) + vi->inode_isize +
97 vi->xattr_isize + erofs_blkoff(sb, map->m_la);
98 map->m_plen = inode->i_size - map->m_la;
99
100 /* inline data should be located in the same meta block */
101 if (erofs_blkoff(sb, map->m_pa) + map->m_plen > sb->s_blocksize) {
102 erofs_err(sb, "inline data across blocks @ nid %llu", vi->nid);
103 DBG_BUGON(1);
104 return -EFSCORRUPTED;
105 }
106 map->m_flags |= EROFS_MAP_META;
107 }
108 return 0;
109 }
110
erofs_map_blocks(struct inode * inode,struct erofs_map_blocks * map)111 int erofs_map_blocks(struct inode *inode, struct erofs_map_blocks *map)
112 {
113 struct super_block *sb = inode->i_sb;
114 struct erofs_inode *vi = EROFS_I(inode);
115 struct erofs_inode_chunk_index *idx;
116 struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
117 u64 chunknr;
118 unsigned int unit;
119 erofs_off_t pos;
120 void *kaddr;
121 int err = 0;
122
123 trace_erofs_map_blocks_enter(inode, map, 0);
124 map->m_deviceid = 0;
125 if (map->m_la >= inode->i_size) {
126 /* leave out-of-bound access unmapped */
127 map->m_flags = 0;
128 map->m_plen = map->m_llen;
129 goto out;
130 }
131
132 if (vi->datalayout != EROFS_INODE_CHUNK_BASED) {
133 err = erofs_map_blocks_flatmode(inode, map);
134 goto out;
135 }
136
137 if (vi->chunkformat & EROFS_CHUNK_FORMAT_INDEXES)
138 unit = sizeof(*idx); /* chunk index */
139 else
140 unit = EROFS_BLOCK_MAP_ENTRY_SIZE; /* block map */
141
142 chunknr = map->m_la >> vi->chunkbits;
143 pos = ALIGN(erofs_iloc(inode) + vi->inode_isize +
144 vi->xattr_isize, unit) + unit * chunknr;
145
146 kaddr = erofs_read_metabuf(&buf, sb, pos, EROFS_KMAP);
147 if (IS_ERR(kaddr)) {
148 err = PTR_ERR(kaddr);
149 goto out;
150 }
151 map->m_la = chunknr << vi->chunkbits;
152 map->m_plen = min_t(erofs_off_t, 1UL << vi->chunkbits,
153 round_up(inode->i_size - map->m_la, sb->s_blocksize));
154
155 /* handle block map */
156 if (!(vi->chunkformat & EROFS_CHUNK_FORMAT_INDEXES)) {
157 __le32 *blkaddr = kaddr;
158
159 if (le32_to_cpu(*blkaddr) == EROFS_NULL_ADDR) {
160 map->m_flags = 0;
161 } else {
162 map->m_pa = erofs_pos(sb, le32_to_cpu(*blkaddr));
163 map->m_flags = EROFS_MAP_MAPPED;
164 }
165 goto out_unlock;
166 }
167 /* parse chunk indexes */
168 idx = kaddr;
169 switch (le32_to_cpu(idx->blkaddr)) {
170 case EROFS_NULL_ADDR:
171 map->m_flags = 0;
172 break;
173 default:
174 map->m_deviceid = le16_to_cpu(idx->device_id) &
175 EROFS_SB(sb)->device_id_mask;
176 map->m_pa = erofs_pos(sb, le32_to_cpu(idx->blkaddr));
177 map->m_flags = EROFS_MAP_MAPPED;
178 break;
179 }
180 out_unlock:
181 erofs_put_metabuf(&buf);
182 out:
183 if (!err)
184 map->m_llen = map->m_plen;
185 trace_erofs_map_blocks_exit(inode, map, 0, err);
186 return err;
187 }
188
erofs_fill_from_devinfo(struct erofs_map_dev * map,struct super_block * sb,struct erofs_device_info * dif)189 static void erofs_fill_from_devinfo(struct erofs_map_dev *map,
190 struct super_block *sb, struct erofs_device_info *dif)
191 {
192 map->m_sb = sb;
193 map->m_dif = dif;
194 map->m_bdev = NULL;
195 if (dif->file && S_ISBLK(file_inode(dif->file)->i_mode))
196 map->m_bdev = file_bdev(dif->file);
197 }
198
erofs_map_dev(struct super_block * sb,struct erofs_map_dev * map)199 int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *map)
200 {
201 struct erofs_dev_context *devs = EROFS_SB(sb)->devs;
202 struct erofs_device_info *dif;
203 erofs_off_t startoff, length;
204 int id;
205
206 erofs_fill_from_devinfo(map, sb, &EROFS_SB(sb)->dif0);
207 map->m_bdev = sb->s_bdev; /* use s_bdev for the primary device */
208 if (map->m_deviceid) {
209 down_read(&devs->rwsem);
210 dif = idr_find(&devs->tree, map->m_deviceid - 1);
211 if (!dif) {
212 up_read(&devs->rwsem);
213 return -ENODEV;
214 }
215 if (devs->flatdev) {
216 map->m_pa += erofs_pos(sb, dif->mapped_blkaddr);
217 up_read(&devs->rwsem);
218 return 0;
219 }
220 erofs_fill_from_devinfo(map, sb, dif);
221 up_read(&devs->rwsem);
222 } else if (devs->extra_devices && !devs->flatdev) {
223 down_read(&devs->rwsem);
224 idr_for_each_entry(&devs->tree, dif, id) {
225 if (!dif->mapped_blkaddr)
226 continue;
227
228 startoff = erofs_pos(sb, dif->mapped_blkaddr);
229 length = erofs_pos(sb, dif->blocks);
230 if (map->m_pa >= startoff &&
231 map->m_pa < startoff + length) {
232 map->m_pa -= startoff;
233 erofs_fill_from_devinfo(map, sb, dif);
234 break;
235 }
236 }
237 up_read(&devs->rwsem);
238 }
239 return 0;
240 }
241
242 /*
243 * bit 30: I/O error occurred on this folio
244 * bit 29: CPU has dirty data in D-cache (needs aliasing handling);
245 * bit 0 - 29: remaining parts to complete this folio
246 */
247 #define EROFS_ONLINEFOLIO_EIO 30
248 #define EROFS_ONLINEFOLIO_DIRTY 29
249
erofs_onlinefolio_init(struct folio * folio)250 void erofs_onlinefolio_init(struct folio *folio)
251 {
252 union {
253 atomic_t o;
254 void *v;
255 } u = { .o = ATOMIC_INIT(1) };
256
257 folio->private = u.v; /* valid only if file-backed folio is locked */
258 }
259
erofs_onlinefolio_split(struct folio * folio)260 void erofs_onlinefolio_split(struct folio *folio)
261 {
262 atomic_inc((atomic_t *)&folio->private);
263 }
264
erofs_onlinefolio_end(struct folio * folio,int err,bool dirty)265 void erofs_onlinefolio_end(struct folio *folio, int err, bool dirty)
266 {
267 int orig, v;
268
269 do {
270 orig = atomic_read((atomic_t *)&folio->private);
271 DBG_BUGON(orig <= 0);
272 v = dirty << EROFS_ONLINEFOLIO_DIRTY;
273 v |= (orig - 1) | (!!err << EROFS_ONLINEFOLIO_EIO);
274 } while (atomic_cmpxchg((atomic_t *)&folio->private, orig, v) != orig);
275
276 if (v & (BIT(EROFS_ONLINEFOLIO_DIRTY) - 1))
277 return;
278 folio->private = 0;
279 if (v & BIT(EROFS_ONLINEFOLIO_DIRTY))
280 flush_dcache_folio(folio);
281 folio_end_read(folio, !(v & BIT(EROFS_ONLINEFOLIO_EIO)));
282 }
283
erofs_iomap_begin(struct inode * inode,loff_t offset,loff_t length,unsigned int flags,struct iomap * iomap,struct iomap * srcmap)284 static int erofs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
285 unsigned int flags, struct iomap *iomap, struct iomap *srcmap)
286 {
287 int ret;
288 struct super_block *sb = inode->i_sb;
289 struct erofs_map_blocks map;
290 struct erofs_map_dev mdev;
291
292 map.m_la = offset;
293 map.m_llen = length;
294
295 ret = erofs_map_blocks(inode, &map);
296 if (ret < 0)
297 return ret;
298
299 mdev = (struct erofs_map_dev) {
300 .m_deviceid = map.m_deviceid,
301 .m_pa = map.m_pa,
302 };
303 ret = erofs_map_dev(sb, &mdev);
304 if (ret)
305 return ret;
306
307 iomap->offset = map.m_la;
308 if (flags & IOMAP_DAX)
309 iomap->dax_dev = mdev.m_dif->dax_dev;
310 else
311 iomap->bdev = mdev.m_bdev;
312 iomap->length = map.m_llen;
313 iomap->flags = 0;
314 iomap->private = NULL;
315
316 if (!(map.m_flags & EROFS_MAP_MAPPED)) {
317 iomap->type = IOMAP_HOLE;
318 iomap->addr = IOMAP_NULL_ADDR;
319 if (!iomap->length)
320 iomap->length = length;
321 return 0;
322 }
323
324 if (map.m_flags & EROFS_MAP_META) {
325 void *ptr;
326 struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
327
328 iomap->type = IOMAP_INLINE;
329 ptr = erofs_read_metabuf(&buf, sb, mdev.m_pa, EROFS_KMAP);
330 if (IS_ERR(ptr))
331 return PTR_ERR(ptr);
332 iomap->inline_data = ptr;
333 iomap->private = buf.base;
334 } else {
335 iomap->type = IOMAP_MAPPED;
336 iomap->addr = mdev.m_dif->fsoff + mdev.m_pa;
337 if (flags & IOMAP_DAX)
338 iomap->addr += mdev.m_dif->dax_part_off;
339 }
340 return 0;
341 }
342
erofs_iomap_end(struct inode * inode,loff_t pos,loff_t length,ssize_t written,unsigned int flags,struct iomap * iomap)343 static int erofs_iomap_end(struct inode *inode, loff_t pos, loff_t length,
344 ssize_t written, unsigned int flags, struct iomap *iomap)
345 {
346 void *ptr = iomap->private;
347
348 if (ptr) {
349 struct erofs_buf buf = {
350 .page = kmap_to_page(ptr),
351 .base = ptr,
352 .kmap_type = EROFS_KMAP,
353 };
354
355 DBG_BUGON(iomap->type != IOMAP_INLINE);
356 erofs_put_metabuf(&buf);
357 } else {
358 DBG_BUGON(iomap->type == IOMAP_INLINE);
359 }
360 return written;
361 }
362
363 static const struct iomap_ops erofs_iomap_ops = {
364 .iomap_begin = erofs_iomap_begin,
365 .iomap_end = erofs_iomap_end,
366 };
367
erofs_fiemap(struct inode * inode,struct fiemap_extent_info * fieinfo,u64 start,u64 len)368 int erofs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
369 u64 start, u64 len)
370 {
371 if (erofs_inode_is_data_compressed(EROFS_I(inode)->datalayout)) {
372 #ifdef CONFIG_EROFS_FS_ZIP
373 return iomap_fiemap(inode, fieinfo, start, len,
374 &z_erofs_iomap_report_ops);
375 #else
376 return -EOPNOTSUPP;
377 #endif
378 }
379 return iomap_fiemap(inode, fieinfo, start, len, &erofs_iomap_ops);
380 }
381
382 /*
383 * since we dont have write or truncate flows, so no inode
384 * locking needs to be held at the moment.
385 */
erofs_read_folio(struct file * file,struct folio * folio)386 static int erofs_read_folio(struct file *file, struct folio *folio)
387 {
388 trace_erofs_read_folio(folio, true);
389
390 return iomap_read_folio(folio, &erofs_iomap_ops);
391 }
392
erofs_readahead(struct readahead_control * rac)393 static void erofs_readahead(struct readahead_control *rac)
394 {
395 trace_erofs_readahead(rac->mapping->host, readahead_index(rac),
396 readahead_count(rac), true);
397
398 return iomap_readahead(rac, &erofs_iomap_ops);
399 }
400
erofs_bmap(struct address_space * mapping,sector_t block)401 static sector_t erofs_bmap(struct address_space *mapping, sector_t block)
402 {
403 return iomap_bmap(mapping, block, &erofs_iomap_ops);
404 }
405
erofs_file_read_iter(struct kiocb * iocb,struct iov_iter * to)406 static ssize_t erofs_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
407 {
408 struct inode *inode = file_inode(iocb->ki_filp);
409
410 /* no need taking (shared) inode lock since it's a ro filesystem */
411 if (!iov_iter_count(to))
412 return 0;
413
414 #ifdef CONFIG_FS_DAX
415 if (IS_DAX(inode))
416 return dax_iomap_rw(iocb, to, &erofs_iomap_ops);
417 #endif
418 if (iocb->ki_flags & IOCB_DIRECT) {
419 struct block_device *bdev = inode->i_sb->s_bdev;
420 unsigned int blksize_mask;
421
422 if (bdev)
423 blksize_mask = bdev_logical_block_size(bdev) - 1;
424 else
425 blksize_mask = i_blocksize(inode) - 1;
426
427 if ((iocb->ki_pos | iov_iter_count(to) |
428 iov_iter_alignment(to)) & blksize_mask)
429 return -EINVAL;
430
431 return iomap_dio_rw(iocb, to, &erofs_iomap_ops,
432 NULL, 0, NULL, 0);
433 }
434 return filemap_read(iocb, to, 0);
435 }
436
437 /* for uncompressed (aligned) files and raw access for other files */
438 const struct address_space_operations erofs_aops = {
439 .read_folio = erofs_read_folio,
440 .readahead = erofs_readahead,
441 .bmap = erofs_bmap,
442 .direct_IO = noop_direct_IO,
443 .release_folio = iomap_release_folio,
444 .invalidate_folio = iomap_invalidate_folio,
445 };
446
447 #ifdef CONFIG_FS_DAX
erofs_dax_huge_fault(struct vm_fault * vmf,unsigned int order)448 static vm_fault_t erofs_dax_huge_fault(struct vm_fault *vmf,
449 unsigned int order)
450 {
451 return dax_iomap_fault(vmf, order, NULL, NULL, &erofs_iomap_ops);
452 }
453
erofs_dax_fault(struct vm_fault * vmf)454 static vm_fault_t erofs_dax_fault(struct vm_fault *vmf)
455 {
456 return erofs_dax_huge_fault(vmf, 0);
457 }
458
459 static const struct vm_operations_struct erofs_dax_vm_ops = {
460 .fault = erofs_dax_fault,
461 .huge_fault = erofs_dax_huge_fault,
462 };
463
erofs_file_mmap(struct file * file,struct vm_area_struct * vma)464 static int erofs_file_mmap(struct file *file, struct vm_area_struct *vma)
465 {
466 if (!IS_DAX(file_inode(file)))
467 return generic_file_readonly_mmap(file, vma);
468
469 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
470 return -EINVAL;
471
472 vma->vm_ops = &erofs_dax_vm_ops;
473 vm_flags_set(vma, VM_HUGEPAGE);
474 return 0;
475 }
476 #else
477 #define erofs_file_mmap generic_file_readonly_mmap
478 #endif
479
480 const struct file_operations erofs_file_fops = {
481 .llseek = generic_file_llseek,
482 .read_iter = erofs_file_read_iter,
483 .mmap = erofs_file_mmap,
484 .get_unmapped_area = thp_get_unmapped_area,
485 .splice_read = filemap_splice_read,
486 };
487