projects/linux·blob master · refs

536 lines · 14.4 KiB

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/filelock.h>
9#include <linux/sched/mm.h>
10#include <trace/events/erofs.h>
11
12void erofs_unmap_metabuf(struct erofs_buf *buf)
13{
14 if (!buf->base)
15 return;
16 kunmap_local(buf->base);
17 buf->base = NULL;
18}
19
20void erofs_put_metabuf(struct erofs_buf *buf)
21{
22 if (!buf->page)
23 return;
24 erofs_unmap_metabuf(buf);
25 folio_put(page_folio(buf->page));
26 buf->page = NULL;
27}
28
29void *erofs_bread(struct erofs_buf *buf, erofs_off_t offset, bool need_kmap)
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->mapping != buf->mapping)
37 /* the cached folio belongs to another address_space */
38 folio = NULL;
39 else if (folio_file_page(folio, index) != buf->page)
40 erofs_unmap_metabuf(buf);
41 }
42 if (!folio || !folio_contains(folio, index)) {
43 erofs_put_metabuf(buf);
44 folio = read_cache_folio(buf->mapping, index,
45 buf->mc ? erofs_read_meta_folio : NULL, NULL);
46 if (IS_ERR(folio))
47 return folio;
48 }
49 buf->page = folio_file_page(folio, index);
50 if (!need_kmap)
51 return NULL;
52 if (!buf->base)
53 buf->base = kmap_local_page(buf->page);
54 return buf->base + ((buf->off + offset) & ~PAGE_MASK);
55}
56
57int erofs_init_metabuf(struct erofs_buf *buf, struct super_block *sb,
58 bool in_metabox)
59{
60 struct erofs_sb_info *sbi = EROFS_SB(sb);
61
62 buf->mc = false;
63 if (in_metabox) {
64 if (unlikely(!sbi->metabox_inode))
65 return -EFSCORRUPTED;
66 buf->mapping = sbi->metabox_inode->i_mapping;
67 return 0;
68 }
69 if (erofs_is_fileio_mode(sbi)) {
70 buf->mapping = sbi->managed_cache->i_mapping;
71 buf->mc = true;
72 } else {
73 buf->off = sbi->dif0.fsoff;
74 buf->mapping = sb->s_bdev->bd_mapping;
75 }
76 return 0;
77}
78
79void *erofs_read_metabuf(struct erofs_buf *buf, struct super_block *sb,
80 erofs_off_t offset, bool in_metabox)
81{
82 int err;
83
84 err = erofs_init_metabuf(buf, sb, in_metabox);
85 if (err)
86 return ERR_PTR(err);
87 return erofs_bread(buf, offset, true);
88}
89
90static int erofs_map_chunks(struct inode *inode, struct erofs_map_blocks *map)
91{
92 struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
93 struct super_block *sb = inode->i_sb;
94 struct erofs_inode *vi = EROFS_I(inode);
95 struct erofs_inode_chunk_index *idx;
96 unsigned int unit = vi->chunkformat & EROFS_CHUNK_FORMAT_INDEXES ?
97 sizeof(*idx) : EROFS_BLOCK_MAP_ENTRY_SIZE;
98 erofs_blk_t addrmask = (vi->chunkformat & EROFS_CHUNK_FORMAT_48BIT) ?
99 BIT_ULL(48) - 1 : BIT_ULL(32) - 1;
100 u64 nr = map->m_la >> vi->chunkbits, chunksize = 1ULL << vi->chunkbits;
101 erofs_off_t pos = ALIGN(erofs_iloc(inode) + vi->inode_isize +
102 vi->xattr_isize, unit) + unit * nr;
103 /* m_llen will be clamped to EOF in the end */
104 erofs_off_t endpos = round_up(pos + 1, sb->s_blocksize);
105 u64 last, addr;
106
107 idx = erofs_read_metabuf(&buf, sb, pos, erofs_inode_in_metabox(inode));
108 if (IS_ERR(idx))
109 return PTR_ERR(idx);
110
111 map->m_la = nr << vi->chunkbits;
112 map->m_llen = 0;
113 nr = 0;
114 do {
115 if (unit == EROFS_BLOCK_MAP_ENTRY_SIZE) {
116 addr = le32_to_cpu(((__le32 *)idx)[nr]);
117 if (addr == (u32)EROFS_NULL_ADDR)
118 addr = EROFS_NULL_ADDR;
119 } else {
120 addr = (((u64)le16_to_cpu(idx[nr].startblk_hi) << 32) |
121 le32_to_cpu(idx[nr].startblk_lo)) & addrmask;
122 if (addr ^ (EROFS_NULL_ADDR & addrmask))
123 addr |= (u64)(le16_to_cpu(idx[nr].device_id) &
124 EROFS_SB(sb)->device_id_mask) << 48;
125 else
126 addr = EROFS_NULL_ADDR;
127 }
128 if (!nr) {
129 last = addr;
130 continue;
131 }
132 /* expand and account the prior chunk here */
133 map->m_llen += chunksize;
134 if (last != EROFS_NULL_ADDR)
135 last += erofs_blknr(sb, chunksize);
136 } while (addr == last && pos + (++nr) * unit < endpos);
137
138 if (last != EROFS_NULL_ADDR) {
139 map->m_pa = erofs_pos(sb, last & addrmask) - map->m_llen;
140 map->m_deviceid = last >> 48;
141 map->m_flags = EROFS_MAP_MAPPED;
142 }
143 if (addr == last)
144 map->m_llen += chunksize;
145 map->m_llen = min_t(erofs_off_t, map->m_llen,
146 round_up(inode->i_size - map->m_la, sb->s_blocksize));
147 erofs_put_metabuf(&buf);
148 return 0;
149}
150
151int erofs_map_blocks(struct inode *inode, struct erofs_map_blocks *map)
152{
153 struct super_block *sb = inode->i_sb;
154 struct erofs_inode *vi = EROFS_I(inode);
155 bool tailinline = (vi->datalayout == EROFS_INODE_FLAT_INLINE);
156 erofs_off_t pos;
157 int err = 0;
158
159 trace_erofs_map_blocks_enter(inode, map, 0);
160 map->m_deviceid = 0;
161 map->m_flags = 0;
162 if (map->m_la >= inode->i_size)
163 goto out;
164 if (vi->datalayout == EROFS_INODE_CHUNK_BASED) {
165 err = erofs_map_chunks(inode, map);
166 } else if (tailinline || vi->startblk != EROFS_NULL_ADDR) {
167 pos = erofs_pos(sb, erofs_iblks(inode) - tailinline);
168 map->m_flags = EROFS_MAP_MAPPED;
169 if (map->m_la < pos) {
170 map->m_pa = erofs_pos(sb, vi->startblk) + map->m_la;
171 map->m_llen = pos - map->m_la;
172 } else {
173 map->m_pa = erofs_iloc(inode) + vi->inode_isize +
174 vi->xattr_isize + erofs_blkoff(sb, map->m_la);
175 map->m_llen = inode->i_size - map->m_la;
176 map->m_flags |= EROFS_MAP_META;
177 if (erofs_blkoff(sb, map->m_pa) + map->m_llen >
178 sb->s_blocksize) {
179 erofs_err(sb, "inline data across blocks @ nid %llu", vi->nid);
180 return -EFSCORRUPTED;
181 }
182 }
183 }
184out:
185 map->m_plen = err ? 0 : map->m_llen;
186 trace_erofs_map_blocks_exit(inode, map, 0, err);
187 return err;
188}
189
190static void erofs_fill_from_devinfo(struct erofs_map_dev *map,
191 struct super_block *sb, struct erofs_device_info *dif)
192{
193 map->m_sb = sb;
194 map->m_dif = dif;
195 map->m_bdev = NULL;
196 if (dif->file && S_ISBLK(file_inode(dif->file)->i_mode))
197 map->m_bdev = file_bdev(dif->file);
198}
199
200int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *map)
201{
202 struct erofs_dev_context *devs = EROFS_SB(sb)->devs;
203 struct erofs_device_info *dif;
204 erofs_off_t startoff;
205 int id;
206
207 erofs_fill_from_devinfo(map, sb, &EROFS_SB(sb)->dif0);
208 map->m_bdev = sb->s_bdev; /* use s_bdev for the primary device */
209 if (map->m_deviceid) {
210 down_read(&devs->rwsem);
211 dif = idr_find(&devs->tree, map->m_deviceid - 1);
212 if (!dif) {
213 up_read(&devs->rwsem);
214 return -ENODEV;
215 }
216 if (devs->flatdev) {
217 map->m_pa += erofs_pos(sb, dif->uniaddr);
218 up_read(&devs->rwsem);
219 return 0;
220 }
221 erofs_fill_from_devinfo(map, sb, dif);
222 up_read(&devs->rwsem);
223 } else if (devs->extra_devices && !devs->flatdev) {
224 down_read(&devs->rwsem);
225 idr_for_each_entry(&devs->tree, dif, id) {
226 if (!dif->uniaddr)
227 continue;
228
229 startoff = erofs_pos(sb, dif->uniaddr);
230 if (map->m_pa >= startoff &&
231 map->m_pa < startoff + erofs_pos(sb, dif->blocks)) {
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
250void 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
260void erofs_onlinefolio_split(struct folio *folio)
261{
262 atomic_inc((atomic_t *)&folio->private);
263}
264
265void 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
284static 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 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
288 struct erofs_iomap_iter_ctx *ctx = iter->private;
289 struct inode *realinode = ctx ? ctx->realinode : inode;
290 struct super_block *sb = realinode->i_sb;
291 struct erofs_map_blocks map;
292 struct erofs_map_dev mdev;
293 int ret;
294
295 map.m_la = offset;
296 map.m_llen = length;
297 ret = erofs_map_blocks(realinode, &map);
298 if (ret < 0)
299 return ret;
300
301 iomap->offset = map.m_la;
302 iomap->length = map.m_llen;
303 iomap->flags = 0;
304 iomap->addr = IOMAP_NULL_ADDR;
305 iomap->private = NULL;
306 if (!(map.m_flags & EROFS_MAP_MAPPED)) {
307 iomap->type = IOMAP_HOLE;
308 return 0;
309 }
310
311 if (!(map.m_flags & EROFS_MAP_META) || !erofs_inode_in_metabox(realinode)) {
312 mdev = (struct erofs_map_dev) {
313 .m_deviceid = map.m_deviceid,
314 .m_pa = map.m_pa,
315 };
316 ret = erofs_map_dev(sb, &mdev);
317 if (ret)
318 return ret;
319
320 iomap->private = mdev.m_dif;
321 if (flags & IOMAP_DAX)
322 iomap->dax_dev = mdev.m_dif->dax_dev;
323 else
324 iomap->bdev = mdev.m_bdev;
325 iomap->addr = mdev.m_dif->fsoff + mdev.m_pa;
326 if (flags & IOMAP_DAX)
327 iomap->addr += mdev.m_dif->dax_part_off;
328 }
329
330 if (map.m_flags & EROFS_MAP_META) {
331 iomap->type = IOMAP_INLINE;
332 /* read context should read the inlined data */
333 if (ctx) {
334 struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
335 void *ptr;
336
337 ptr = erofs_read_metabuf(&buf, sb, map.m_pa,
338 erofs_inode_in_metabox(realinode));
339 if (IS_ERR(ptr))
340 return PTR_ERR(ptr);
341 iomap->inline_data = ptr;
342 ctx->page = buf.page;
343 ctx->base = buf.base;
344 }
345 } else {
346 iomap->type = IOMAP_MAPPED;
347 }
348 return 0;
349}
350
351static int erofs_iomap_end(struct inode *inode, loff_t pos, loff_t length,
352 ssize_t written, unsigned int flags, struct iomap *iomap)
353{
354 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
355 struct erofs_iomap_iter_ctx *ctx = iter->private;
356
357 if (ctx && ctx->base) {
358 struct erofs_buf buf = {
359 .page = ctx->page,
360 .base = ctx->base,
361 };
362
363 DBG_BUGON(iomap->type != IOMAP_INLINE);
364 erofs_put_metabuf(&buf);
365 ctx->base = NULL;
366 }
367 return written;
368}
369
370static DEFINE_IOMAP_ITER_NEXT_END(erofs_iomap_next, erofs_iomap_begin,
371 erofs_iomap_end);
372
373static const struct iomap_ops erofs_iomap_ops = {
374 .iomap_next = erofs_iomap_next,
375};
376
377int erofs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
378 u64 start, u64 len)
379{
380 if (erofs_inode_is_data_compressed(EROFS_I(inode)->datalayout)) {
381 if (!IS_ENABLED(CONFIG_EROFS_FS_ZIP))
382 return -EOPNOTSUPP;
383 return iomap_fiemap(inode, fieinfo, start, len,
384 &z_erofs_iomap_report_ops);
385 }
386 return iomap_fiemap(inode, fieinfo, start, len, &erofs_iomap_ops);
387}
388
389/*
390 * since we dont have write or truncate flows, so no inode
391 * locking needs to be held at the moment.
392 */
393static int erofs_read_folio(struct file *file, struct folio *folio)
394{
395 bool need_iput;
396 struct erofs_iomap_iter_ctx iter_ctx = {
397 .realinode = erofs_real_inode(folio_inode(folio), &need_iput),
398 };
399 struct erofs_sb_info *sbi = EROFS_I_SB(iter_ctx.realinode);
400 struct iomap_read_folio_ctx read_ctx = {
401 .ops = erofs_is_fileio_mode(sbi) ?
402 &erofs_fileio_iomap_read_ops : &iomap_bio_read_ops,
403 .cur_folio = folio,
404 };
405
406 trace_erofs_read_folio(iter_ctx.realinode, folio, true);
407 iomap_read_folio(&erofs_iomap_ops, &read_ctx, &iter_ctx);
408 if (need_iput)
409 iput(iter_ctx.realinode);
410 return 0;
411}
412
413static void erofs_readahead(struct readahead_control *rac)
414{
415 bool need_iput;
416 struct erofs_iomap_iter_ctx iter_ctx = {
417 .realinode = erofs_real_inode(rac->mapping->host, &need_iput),
418 };
419 struct erofs_sb_info *sbi = EROFS_I_SB(iter_ctx.realinode);
420 struct iomap_read_folio_ctx read_ctx = {
421 .ops = erofs_is_fileio_mode(sbi) ?
422 &erofs_fileio_iomap_read_ops : &iomap_bio_read_ops,
423 .rac = rac,
424 };
425
426 trace_erofs_readahead(iter_ctx.realinode, readahead_index(rac),
427 readahead_count(rac), true);
428 iomap_readahead(&erofs_iomap_ops, &read_ctx, &iter_ctx);
429 if (need_iput)
430 iput(iter_ctx.realinode);
431}
432
433static sector_t erofs_bmap(struct address_space *mapping, sector_t block)
434{
435 return iomap_bmap(mapping, block, &erofs_iomap_ops);
436}
437
438static ssize_t erofs_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
439{
440 struct inode *inode = file_inode(iocb->ki_filp);
441
442 /* no need taking (shared) inode lock since it's a ro filesystem */
443 if (!iov_iter_count(to))
444 return 0;
445
446 if (IS_ENABLED(CONFIG_FS_DAX) && IS_DAX(inode))
447 return dax_iomap_rw(iocb, to, &erofs_iomap_ops);
448
449 if ((iocb->ki_flags & IOCB_DIRECT) && inode->i_sb->s_bdev) {
450 struct erofs_iomap_iter_ctx iter_ctx = {
451 .realinode = inode,
452 };
453
454 return iomap_dio_rw(iocb, to, &erofs_iomap_ops,
455 NULL, 0, &iter_ctx, 0);
456 }
457 return filemap_read(iocb, to, 0);
458}
459
460const struct address_space_operations erofs_aops = {
461 .read_folio = erofs_read_folio,
462 .readahead = erofs_readahead,
463 .bmap = erofs_bmap,
464 .direct_IO = noop_direct_IO,
465 .release_folio = iomap_release_folio,
466 .invalidate_folio = iomap_invalidate_folio,
467};
468
469#ifdef CONFIG_FS_DAX
470static vm_fault_t erofs_dax_huge_fault(struct vm_fault *vmf,
471 unsigned int order)
472{
473 return dax_iomap_fault(vmf, order, NULL, NULL, &erofs_iomap_ops);
474}
475
476static vm_fault_t erofs_dax_fault(struct vm_fault *vmf)
477{
478 return erofs_dax_huge_fault(vmf, 0);
479}
480
481static const struct vm_operations_struct erofs_dax_vm_ops = {
482 .fault = erofs_dax_fault,
483 .huge_fault = erofs_dax_huge_fault,
484};
485
486static int erofs_file_mmap_prepare(struct vm_area_desc *desc)
487{
488 if (!IS_DAX(file_inode(desc->file)))
489 return generic_file_readonly_mmap_prepare(desc);
490
491 if (vma_desc_test_all(desc, VMA_SHARED_BIT, VMA_MAYWRITE_BIT))
492 return -EINVAL;
493
494 desc->vm_ops = &erofs_dax_vm_ops;
495 vma_desc_set_flags(desc, VMA_HUGEPAGE_BIT);
496 return 0;
497}
498#else
499#define erofs_file_mmap_prepare generic_file_readonly_mmap_prepare
500#endif
501
502loff_t erofs_file_llseek(struct file *file, loff_t offset, int whence)
503{
504 struct inode *inode = file->f_mapping->host;
505 const struct iomap_ops *ops = &erofs_iomap_ops;
506
507 if (erofs_inode_is_data_compressed(EROFS_I(inode)->datalayout)) {
508 if (!IS_ENABLED(CONFIG_EROFS_FS_ZIP))
509 return generic_file_llseek(file, offset, whence);
510 ops = &z_erofs_iomap_report_ops;
511 }
512
513 if (whence == SEEK_HOLE)
514 offset = iomap_seek_hole(inode, offset, ops);
515 else if (whence == SEEK_DATA)
516 offset = iomap_seek_data(inode, offset, ops);
517 else
518 return generic_file_llseek(file, offset, whence);
519
520 if (offset < 0)
521 return offset;
522 return vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
523}
524
525const struct file_operations erofs_file_fops = {
526 .llseek = erofs_file_llseek,
527 .read_iter = erofs_file_read_iter,
528 .unlocked_ioctl = erofs_ioctl,
529#ifdef CONFIG_COMPAT
530 .compat_ioctl = erofs_compat_ioctl,
531#endif
532 .mmap_prepare = erofs_file_mmap_prepare,
533 .get_unmapped_area = thp_get_unmapped_area,
534 .splice_read = filemap_splice_read,
535 .setlease = generic_setlease,
536};