Commit | Line | Data |
---|---|---|
29b24f6c | 1 | // SPDX-License-Identifier: GPL-2.0-only |
81781b02 | 2 | /* |
81781b02 | 3 | * Copyright (C) 2017-2018 HUAWEI, Inc. |
592e7cd0 | 4 | * https://www.huawei.com/ |
c5aa903a | 5 | * Copyright (C) 2021, Alibaba Cloud |
81781b02 GX |
6 | */ |
7 | #include "internal.h" | |
8 | #include <linux/prefetch.h> | |
06252e9c | 9 | #include <linux/dax.h> |
13f06f48 CY |
10 | #include <trace/events/erofs.h> |
11 | ||
fdf80a47 GX |
12 | void erofs_unmap_metabuf(struct erofs_buf *buf) |
13 | { | |
14 | if (buf->kmap_type == EROFS_KMAP) | |
15 | kunmap(buf->page); | |
16 | else if (buf->kmap_type == EROFS_KMAP_ATOMIC) | |
17 | kunmap_atomic(buf->base); | |
18 | buf->base = NULL; | |
19 | buf->kmap_type = EROFS_NO_KMAP; | |
20 | } | |
21 | ||
22 | void erofs_put_metabuf(struct erofs_buf *buf) | |
23 | { | |
24 | if (!buf->page) | |
25 | return; | |
26 | erofs_unmap_metabuf(buf); | |
27 | put_page(buf->page); | |
28 | buf->page = NULL; | |
29 | } | |
30 | ||
fe5de585 GX |
31 | void *erofs_bread(struct erofs_buf *buf, struct inode *inode, |
32 | erofs_blk_t blkaddr, enum erofs_kmap_type type) | |
fdf80a47 | 33 | { |
fe5de585 | 34 | struct address_space *const mapping = inode->i_mapping; |
fdf80a47 GX |
35 | erofs_off_t offset = blknr_to_addr(blkaddr); |
36 | pgoff_t index = offset >> PAGE_SHIFT; | |
37 | struct page *page = buf->page; | |
38 | ||
39 | if (!page || page->index != index) { | |
40 | erofs_put_metabuf(buf); | |
41 | page = read_cache_page_gfp(mapping, index, | |
42 | mapping_gfp_constraint(mapping, ~__GFP_FS)); | |
43 | if (IS_ERR(page)) | |
44 | return page; | |
45 | /* should already be PageUptodate, no need to lock page */ | |
46 | buf->page = page; | |
47 | } | |
48 | if (buf->kmap_type == EROFS_NO_KMAP) { | |
49 | if (type == EROFS_KMAP) | |
50 | buf->base = kmap(page); | |
51 | else if (type == EROFS_KMAP_ATOMIC) | |
52 | buf->base = kmap_atomic(page); | |
53 | buf->kmap_type = type; | |
54 | } else if (buf->kmap_type != type) { | |
55 | DBG_BUGON(1); | |
56 | return ERR_PTR(-EFAULT); | |
57 | } | |
58 | if (type == EROFS_NO_KMAP) | |
59 | return NULL; | |
60 | return buf->base + (offset & ~PAGE_MASK); | |
61 | } | |
62 | ||
fe5de585 GX |
63 | void *erofs_read_metabuf(struct erofs_buf *buf, struct super_block *sb, |
64 | erofs_blk_t blkaddr, enum erofs_kmap_type type) | |
65 | { | |
66 | return erofs_bread(buf, sb->s_bdev->bd_inode, blkaddr, type); | |
67 | } | |
68 | ||
81781b02 | 69 | static int erofs_map_blocks_flatmode(struct inode *inode, |
f0950b02 BD |
70 | struct erofs_map_blocks *map, |
71 | int flags) | |
81781b02 GX |
72 | { |
73 | erofs_blk_t nblocks, lastblk; | |
74 | u64 offset = map->m_la; | |
a5876e24 | 75 | struct erofs_inode *vi = EROFS_I(inode); |
8a765682 | 76 | bool tailendpacking = (vi->datalayout == EROFS_INODE_FLAT_INLINE); |
81781b02 | 77 | |
fdf80a47 | 78 | nblocks = DIV_ROUND_UP(inode->i_size, EROFS_BLKSIZ); |
8a765682 | 79 | lastblk = nblocks - tailendpacking; |
81781b02 | 80 | |
81781b02 GX |
81 | /* there is no hole in flatmode */ |
82 | map->m_flags = EROFS_MAP_MAPPED; | |
81781b02 GX |
83 | if (offset < blknr_to_addr(lastblk)) { |
84 | map->m_pa = blknr_to_addr(vi->raw_blkaddr) + map->m_la; | |
85 | map->m_plen = blknr_to_addr(lastblk) - offset; | |
8a765682 | 86 | } else if (tailendpacking) { |
81781b02 GX |
87 | /* 2 - inode inline B: inode, [xattrs], inline last blk... */ |
88 | struct erofs_sb_info *sbi = EROFS_SB(inode->i_sb); | |
89 | ||
90 | map->m_pa = iloc(sbi, vi->nid) + vi->inode_isize + | |
91 | vi->xattr_isize + erofs_blkoff(map->m_la); | |
92 | map->m_plen = inode->i_size - offset; | |
93 | ||
469407a3 GX |
94 | /* inline data should be located in the same meta block */ |
95 | if (erofs_blkoff(map->m_pa) + map->m_plen > EROFS_BLKSIZ) { | |
4f761fa2 GX |
96 | erofs_err(inode->i_sb, |
97 | "inline data cross block boundary @ nid %llu", | |
98 | vi->nid); | |
9141b60c | 99 | DBG_BUGON(1); |
469407a3 | 100 | return -EFSCORRUPTED; |
9141b60c | 101 | } |
81781b02 GX |
102 | map->m_flags |= EROFS_MAP_META; |
103 | } else { | |
4f761fa2 GX |
104 | erofs_err(inode->i_sb, |
105 | "internal error @ nid: %llu (size %llu), m_la 0x%llx", | |
106 | vi->nid, inode->i_size, map->m_la); | |
9141b60c | 107 | DBG_BUGON(1); |
469407a3 | 108 | return -EIO; |
81781b02 | 109 | } |
469407a3 | 110 | return 0; |
81781b02 GX |
111 | } |
112 | ||
94d78946 JX |
113 | int erofs_map_blocks(struct inode *inode, |
114 | struct erofs_map_blocks *map, int flags) | |
c5aa903a GX |
115 | { |
116 | struct super_block *sb = inode->i_sb; | |
117 | struct erofs_inode *vi = EROFS_I(inode); | |
118 | struct erofs_inode_chunk_index *idx; | |
fdf80a47 | 119 | struct erofs_buf buf = __EROFS_BUF_INITIALIZER; |
c5aa903a GX |
120 | u64 chunknr; |
121 | unsigned int unit; | |
122 | erofs_off_t pos; | |
fdf80a47 | 123 | void *kaddr; |
c5aa903a GX |
124 | int err = 0; |
125 | ||
469407a3 | 126 | trace_erofs_map_blocks_enter(inode, map, flags); |
dfeab2e9 | 127 | map->m_deviceid = 0; |
c5aa903a GX |
128 | if (map->m_la >= inode->i_size) { |
129 | /* leave out-of-bound access unmapped */ | |
130 | map->m_flags = 0; | |
131 | map->m_plen = 0; | |
132 | goto out; | |
133 | } | |
134 | ||
469407a3 GX |
135 | if (vi->datalayout != EROFS_INODE_CHUNK_BASED) { |
136 | err = erofs_map_blocks_flatmode(inode, map, flags); | |
137 | goto out; | |
138 | } | |
c5aa903a GX |
139 | |
140 | if (vi->chunkformat & EROFS_CHUNK_FORMAT_INDEXES) | |
141 | unit = sizeof(*idx); /* chunk index */ | |
142 | else | |
143 | unit = EROFS_BLOCK_MAP_ENTRY_SIZE; /* block map */ | |
144 | ||
145 | chunknr = map->m_la >> vi->chunkbits; | |
146 | pos = ALIGN(iloc(EROFS_SB(sb), vi->nid) + vi->inode_isize + | |
147 | vi->xattr_isize, unit) + unit * chunknr; | |
148 | ||
fdf80a47 GX |
149 | kaddr = erofs_read_metabuf(&buf, sb, erofs_blknr(pos), EROFS_KMAP); |
150 | if (IS_ERR(kaddr)) { | |
151 | err = PTR_ERR(kaddr); | |
469407a3 GX |
152 | goto out; |
153 | } | |
c5aa903a GX |
154 | map->m_la = chunknr << vi->chunkbits; |
155 | map->m_plen = min_t(erofs_off_t, 1UL << vi->chunkbits, | |
156 | roundup(inode->i_size - map->m_la, EROFS_BLKSIZ)); | |
157 | ||
158 | /* handle block map */ | |
159 | if (!(vi->chunkformat & EROFS_CHUNK_FORMAT_INDEXES)) { | |
fdf80a47 | 160 | __le32 *blkaddr = kaddr + erofs_blkoff(pos); |
c5aa903a GX |
161 | |
162 | if (le32_to_cpu(*blkaddr) == EROFS_NULL_ADDR) { | |
163 | map->m_flags = 0; | |
164 | } else { | |
165 | map->m_pa = blknr_to_addr(le32_to_cpu(*blkaddr)); | |
166 | map->m_flags = EROFS_MAP_MAPPED; | |
167 | } | |
168 | goto out_unlock; | |
169 | } | |
170 | /* parse chunk indexes */ | |
fdf80a47 | 171 | idx = kaddr + erofs_blkoff(pos); |
c5aa903a GX |
172 | switch (le32_to_cpu(idx->blkaddr)) { |
173 | case EROFS_NULL_ADDR: | |
174 | map->m_flags = 0; | |
175 | break; | |
176 | default: | |
dfeab2e9 GX |
177 | map->m_deviceid = le16_to_cpu(idx->device_id) & |
178 | EROFS_SB(sb)->device_id_mask; | |
c5aa903a GX |
179 | map->m_pa = blknr_to_addr(le32_to_cpu(idx->blkaddr)); |
180 | map->m_flags = EROFS_MAP_MAPPED; | |
181 | break; | |
182 | } | |
183 | out_unlock: | |
fdf80a47 | 184 | erofs_put_metabuf(&buf); |
c5aa903a | 185 | out: |
469407a3 GX |
186 | if (!err) |
187 | map->m_llen = map->m_plen; | |
188 | trace_erofs_map_blocks_exit(inode, map, flags, 0); | |
c5aa903a GX |
189 | return err; |
190 | } | |
191 | ||
dfeab2e9 GX |
192 | int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *map) |
193 | { | |
194 | struct erofs_dev_context *devs = EROFS_SB(sb)->devs; | |
195 | struct erofs_device_info *dif; | |
196 | int id; | |
197 | ||
198 | /* primary device by default */ | |
199 | map->m_bdev = sb->s_bdev; | |
200 | map->m_daxdev = EROFS_SB(sb)->dax_dev; | |
de205114 | 201 | map->m_dax_part_off = EROFS_SB(sb)->dax_part_off; |
955b478e | 202 | map->m_fscache = EROFS_SB(sb)->s_fscache; |
dfeab2e9 GX |
203 | |
204 | if (map->m_deviceid) { | |
205 | down_read(&devs->rwsem); | |
206 | dif = idr_find(&devs->tree, map->m_deviceid - 1); | |
207 | if (!dif) { | |
208 | up_read(&devs->rwsem); | |
209 | return -ENODEV; | |
210 | } | |
211 | map->m_bdev = dif->bdev; | |
212 | map->m_daxdev = dif->dax_dev; | |
de205114 | 213 | map->m_dax_part_off = dif->dax_part_off; |
955b478e | 214 | map->m_fscache = dif->fscache; |
dfeab2e9 GX |
215 | up_read(&devs->rwsem); |
216 | } else if (devs->extra_devices) { | |
217 | down_read(&devs->rwsem); | |
218 | idr_for_each_entry(&devs->tree, dif, id) { | |
219 | erofs_off_t startoff, length; | |
220 | ||
221 | if (!dif->mapped_blkaddr) | |
222 | continue; | |
223 | startoff = blknr_to_addr(dif->mapped_blkaddr); | |
224 | length = blknr_to_addr(dif->blocks); | |
225 | ||
226 | if (map->m_pa >= startoff && | |
227 | map->m_pa < startoff + length) { | |
228 | map->m_pa -= startoff; | |
229 | map->m_bdev = dif->bdev; | |
230 | map->m_daxdev = dif->dax_dev; | |
de205114 | 231 | map->m_dax_part_off = dif->dax_part_off; |
955b478e | 232 | map->m_fscache = dif->fscache; |
dfeab2e9 GX |
233 | break; |
234 | } | |
235 | } | |
236 | up_read(&devs->rwsem); | |
237 | } | |
238 | return 0; | |
239 | } | |
240 | ||
a08e67a0 HJ |
241 | static int erofs_iomap_begin(struct inode *inode, loff_t offset, loff_t length, |
242 | unsigned int flags, struct iomap *iomap, struct iomap *srcmap) | |
243 | { | |
244 | int ret; | |
245 | struct erofs_map_blocks map; | |
dfeab2e9 | 246 | struct erofs_map_dev mdev; |
a08e67a0 HJ |
247 | |
248 | map.m_la = offset; | |
249 | map.m_llen = length; | |
250 | ||
c5aa903a | 251 | ret = erofs_map_blocks(inode, &map, EROFS_GET_BLOCKS_RAW); |
a08e67a0 HJ |
252 | if (ret < 0) |
253 | return ret; | |
254 | ||
dfeab2e9 GX |
255 | mdev = (struct erofs_map_dev) { |
256 | .m_deviceid = map.m_deviceid, | |
257 | .m_pa = map.m_pa, | |
258 | }; | |
259 | ret = erofs_map_dev(inode->i_sb, &mdev); | |
260 | if (ret) | |
261 | return ret; | |
262 | ||
a08e67a0 | 263 | iomap->offset = map.m_la; |
e33f42b2 | 264 | if (flags & IOMAP_DAX) |
de205114 | 265 | iomap->dax_dev = mdev.m_daxdev; |
e33f42b2 | 266 | else |
de205114 | 267 | iomap->bdev = mdev.m_bdev; |
a08e67a0 HJ |
268 | iomap->length = map.m_llen; |
269 | iomap->flags = 0; | |
771c994e | 270 | iomap->private = NULL; |
a08e67a0 HJ |
271 | |
272 | if (!(map.m_flags & EROFS_MAP_MAPPED)) { | |
273 | iomap->type = IOMAP_HOLE; | |
274 | iomap->addr = IOMAP_NULL_ADDR; | |
275 | if (!iomap->length) | |
276 | iomap->length = length; | |
277 | return 0; | |
278 | } | |
279 | ||
a08e67a0 | 280 | if (map.m_flags & EROFS_MAP_META) { |
fdf80a47 GX |
281 | void *ptr; |
282 | struct erofs_buf buf = __EROFS_BUF_INITIALIZER; | |
771c994e GX |
283 | |
284 | iomap->type = IOMAP_INLINE; | |
fdf80a47 GX |
285 | ptr = erofs_read_metabuf(&buf, inode->i_sb, |
286 | erofs_blknr(mdev.m_pa), EROFS_KMAP); | |
287 | if (IS_ERR(ptr)) | |
288 | return PTR_ERR(ptr); | |
289 | iomap->inline_data = ptr + erofs_blkoff(mdev.m_pa); | |
290 | iomap->private = buf.base; | |
771c994e GX |
291 | } else { |
292 | iomap->type = IOMAP_MAPPED; | |
dfeab2e9 | 293 | iomap->addr = mdev.m_pa; |
e33f42b2 GX |
294 | if (flags & IOMAP_DAX) |
295 | iomap->addr += mdev.m_dax_part_off; | |
a08e67a0 | 296 | } |
a08e67a0 HJ |
297 | return 0; |
298 | } | |
299 | ||
771c994e GX |
300 | static int erofs_iomap_end(struct inode *inode, loff_t pos, loff_t length, |
301 | ssize_t written, unsigned int flags, struct iomap *iomap) | |
302 | { | |
fdf80a47 GX |
303 | void *ptr = iomap->private; |
304 | ||
305 | if (ptr) { | |
306 | struct erofs_buf buf = { | |
307 | .page = kmap_to_page(ptr), | |
308 | .base = ptr, | |
309 | .kmap_type = EROFS_KMAP, | |
310 | }; | |
771c994e | 311 | |
771c994e | 312 | DBG_BUGON(iomap->type != IOMAP_INLINE); |
fdf80a47 | 313 | erofs_put_metabuf(&buf); |
771c994e GX |
314 | } else { |
315 | DBG_BUGON(iomap->type == IOMAP_INLINE); | |
316 | } | |
317 | return written; | |
318 | } | |
319 | ||
a08e67a0 HJ |
320 | static const struct iomap_ops erofs_iomap_ops = { |
321 | .iomap_begin = erofs_iomap_begin, | |
771c994e | 322 | .iomap_end = erofs_iomap_end, |
a08e67a0 HJ |
323 | }; |
324 | ||
eadcd6b5 GX |
325 | int erofs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
326 | u64 start, u64 len) | |
327 | { | |
328 | if (erofs_inode_is_data_compressed(EROFS_I(inode)->datalayout)) { | |
329 | #ifdef CONFIG_EROFS_FS_ZIP | |
330 | return iomap_fiemap(inode, fieinfo, start, len, | |
331 | &z_erofs_iomap_report_ops); | |
332 | #else | |
333 | return -EOPNOTSUPP; | |
334 | #endif | |
335 | } | |
336 | return iomap_fiemap(inode, fieinfo, start, len, &erofs_iomap_ops); | |
337 | } | |
338 | ||
771c994e GX |
339 | /* |
340 | * since we dont have write or truncate flows, so no inode | |
341 | * locking needs to be held at the moment. | |
342 | */ | |
343 | static int erofs_readpage(struct file *file, struct page *page) | |
344 | { | |
345 | return iomap_readpage(page, &erofs_iomap_ops); | |
346 | } | |
347 | ||
348 | static void erofs_readahead(struct readahead_control *rac) | |
349 | { | |
350 | return iomap_readahead(rac, &erofs_iomap_ops); | |
351 | } | |
352 | ||
353 | static sector_t erofs_bmap(struct address_space *mapping, sector_t block) | |
354 | { | |
355 | return iomap_bmap(mapping, block, &erofs_iomap_ops); | |
356 | } | |
357 | ||
a08e67a0 HJ |
358 | static int erofs_prepare_dio(struct kiocb *iocb, struct iov_iter *to) |
359 | { | |
360 | struct inode *inode = file_inode(iocb->ki_filp); | |
361 | loff_t align = iocb->ki_pos | iov_iter_count(to) | | |
362 | iov_iter_alignment(to); | |
363 | struct block_device *bdev = inode->i_sb->s_bdev; | |
364 | unsigned int blksize_mask; | |
365 | ||
366 | if (bdev) | |
367 | blksize_mask = (1 << ilog2(bdev_logical_block_size(bdev))) - 1; | |
368 | else | |
369 | blksize_mask = (1 << inode->i_blkbits) - 1; | |
370 | ||
371 | if (align & blksize_mask) | |
372 | return -EINVAL; | |
a08e67a0 HJ |
373 | return 0; |
374 | } | |
375 | ||
376 | static ssize_t erofs_file_read_iter(struct kiocb *iocb, struct iov_iter *to) | |
377 | { | |
378 | /* no need taking (shared) inode lock since it's a ro filesystem */ | |
379 | if (!iov_iter_count(to)) | |
380 | return 0; | |
381 | ||
06252e9c GX |
382 | #ifdef CONFIG_FS_DAX |
383 | if (IS_DAX(iocb->ki_filp->f_mapping->host)) | |
384 | return dax_iomap_rw(iocb, to, &erofs_iomap_ops); | |
385 | #endif | |
a08e67a0 HJ |
386 | if (iocb->ki_flags & IOCB_DIRECT) { |
387 | int err = erofs_prepare_dio(iocb, to); | |
388 | ||
389 | if (!err) | |
390 | return iomap_dio_rw(iocb, to, &erofs_iomap_ops, | |
4fdccaa0 | 391 | NULL, 0, 0); |
a08e67a0 HJ |
392 | if (err < 0) |
393 | return err; | |
394 | } | |
395 | return filemap_read(iocb, to, 0); | |
396 | } | |
397 | ||
81781b02 GX |
398 | /* for uncompressed (aligned) files and raw access for other files */ |
399 | const struct address_space_operations erofs_raw_access_aops = { | |
771c994e GX |
400 | .readpage = erofs_readpage, |
401 | .readahead = erofs_readahead, | |
9da681e0 | 402 | .bmap = erofs_bmap, |
a08e67a0 HJ |
403 | .direct_IO = noop_direct_IO, |
404 | }; | |
405 | ||
06252e9c GX |
406 | #ifdef CONFIG_FS_DAX |
407 | static vm_fault_t erofs_dax_huge_fault(struct vm_fault *vmf, | |
408 | enum page_entry_size pe_size) | |
409 | { | |
410 | return dax_iomap_fault(vmf, pe_size, NULL, NULL, &erofs_iomap_ops); | |
411 | } | |
412 | ||
413 | static vm_fault_t erofs_dax_fault(struct vm_fault *vmf) | |
414 | { | |
415 | return erofs_dax_huge_fault(vmf, PE_SIZE_PTE); | |
416 | } | |
417 | ||
418 | static const struct vm_operations_struct erofs_dax_vm_ops = { | |
419 | .fault = erofs_dax_fault, | |
420 | .huge_fault = erofs_dax_huge_fault, | |
421 | }; | |
422 | ||
423 | static int erofs_file_mmap(struct file *file, struct vm_area_struct *vma) | |
424 | { | |
425 | if (!IS_DAX(file_inode(file))) | |
426 | return generic_file_readonly_mmap(file, vma); | |
427 | ||
428 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
429 | return -EINVAL; | |
430 | ||
431 | vma->vm_ops = &erofs_dax_vm_ops; | |
432 | vma->vm_flags |= VM_HUGEPAGE; | |
433 | return 0; | |
434 | } | |
435 | #else | |
436 | #define erofs_file_mmap generic_file_readonly_mmap | |
437 | #endif | |
438 | ||
a08e67a0 HJ |
439 | const struct file_operations erofs_file_fops = { |
440 | .llseek = generic_file_llseek, | |
441 | .read_iter = erofs_file_read_iter, | |
06252e9c | 442 | .mmap = erofs_file_mmap, |
a08e67a0 | 443 | .splice_read = generic_file_splice_read, |
81781b02 | 444 | }; |