1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (C) 2017-2018 HUAWEI, Inc. 4 * http://www.huawei.com/ 5 * Created by Gao Xiang <gaoxiang25@huawei.com> 6 */ 7 #include "internal.h" 8 #include <linux/prefetch.h> 9 10 #include <trace/events/erofs.h> 11 12 static inline void read_endio(struct bio *bio) 13 { 14 struct super_block *const sb = bio->bi_private; 15 struct bio_vec *bvec; 16 blk_status_t err = bio->bi_status; 17 struct bvec_iter_all iter_all; 18 19 if (time_to_inject(EROFS_SB(sb), FAULT_READ_IO)) { 20 erofs_show_injection_info(FAULT_READ_IO); 21 err = BLK_STS_IOERR; 22 } 23 24 bio_for_each_segment_all(bvec, bio, iter_all) { 25 struct page *page = bvec->bv_page; 26 27 /* page is already locked */ 28 DBG_BUGON(PageUptodate(page)); 29 30 if (err) 31 SetPageError(page); 32 else 33 SetPageUptodate(page); 34 35 unlock_page(page); 36 /* page could be reclaimed now */ 37 } 38 bio_put(bio); 39 } 40 41 static struct bio *erofs_grab_raw_bio(struct super_block *sb, 42 erofs_blk_t blkaddr, 43 unsigned int nr_pages) 44 { 45 struct bio *bio = bio_alloc(GFP_NOIO, nr_pages); 46 47 bio->bi_end_io = read_endio; 48 bio_set_dev(bio, sb->s_bdev); 49 bio->bi_iter.bi_sector = (sector_t)blkaddr << LOG_SECTORS_PER_BLOCK; 50 bio->bi_private = sb; 51 return bio; 52 } 53 54 /* prio -- true is used for dir */ 55 struct page *__erofs_get_meta_page(struct super_block *sb, 56 erofs_blk_t blkaddr, bool prio, bool nofail) 57 { 58 struct inode *const bd_inode = sb->s_bdev->bd_inode; 59 struct address_space *const mapping = bd_inode->i_mapping; 60 /* prefer retrying in the allocator to blindly looping below */ 61 const gfp_t gfp = mapping_gfp_constraint(mapping, ~__GFP_FS) | 62 (nofail ? __GFP_NOFAIL : 0); 63 unsigned int io_retries = nofail ? EROFS_IO_MAX_RETRIES_NOFAIL : 0; 64 struct page *page; 65 int err; 66 67 repeat: 68 page = find_or_create_page(mapping, blkaddr, gfp); 69 if (!page) { 70 DBG_BUGON(nofail); 71 return ERR_PTR(-ENOMEM); 72 } 73 DBG_BUGON(!PageLocked(page)); 74 75 if (!PageUptodate(page)) { 76 struct bio *bio; 77 78 bio = erofs_grab_raw_bio(sb, blkaddr, 1); 79 80 if (bio_add_page(bio, page, PAGE_SIZE, 0) != PAGE_SIZE) { 81 err = -EFAULT; 82 goto err_out; 83 } 84 85 __submit_bio(bio, REQ_OP_READ, 86 REQ_META | (prio ? REQ_PRIO : 0)); 87 88 lock_page(page); 89 90 /* this page has been truncated by others */ 91 if (page->mapping != mapping) { 92 unlock_repeat: 93 unlock_page(page); 94 put_page(page); 95 goto repeat; 96 } 97 98 /* more likely a read error */ 99 if (!PageUptodate(page)) { 100 if (io_retries) { 101 --io_retries; 102 goto unlock_repeat; 103 } 104 err = -EIO; 105 goto err_out; 106 } 107 } 108 return page; 109 110 err_out: 111 unlock_page(page); 112 put_page(page); 113 return ERR_PTR(err); 114 } 115 116 static int erofs_map_blocks_flatmode(struct inode *inode, 117 struct erofs_map_blocks *map, 118 int flags) 119 { 120 int err = 0; 121 erofs_blk_t nblocks, lastblk; 122 u64 offset = map->m_la; 123 struct erofs_inode *vi = EROFS_I(inode); 124 bool tailendpacking = (vi->datalayout == EROFS_INODE_FLAT_INLINE); 125 126 trace_erofs_map_blocks_flatmode_enter(inode, map, flags); 127 128 nblocks = DIV_ROUND_UP(inode->i_size, PAGE_SIZE); 129 lastblk = nblocks - tailendpacking; 130 131 if (offset >= inode->i_size) { 132 /* leave out-of-bound access unmapped */ 133 map->m_flags = 0; 134 map->m_plen = 0; 135 goto out; 136 } 137 138 /* there is no hole in flatmode */ 139 map->m_flags = EROFS_MAP_MAPPED; 140 141 if (offset < blknr_to_addr(lastblk)) { 142 map->m_pa = blknr_to_addr(vi->raw_blkaddr) + map->m_la; 143 map->m_plen = blknr_to_addr(lastblk) - offset; 144 } else if (tailendpacking) { 145 /* 2 - inode inline B: inode, [xattrs], inline last blk... */ 146 struct erofs_sb_info *sbi = EROFS_SB(inode->i_sb); 147 148 map->m_pa = iloc(sbi, vi->nid) + vi->inode_isize + 149 vi->xattr_isize + erofs_blkoff(map->m_la); 150 map->m_plen = inode->i_size - offset; 151 152 /* inline data should be located in one meta block */ 153 if (erofs_blkoff(map->m_pa) + map->m_plen > PAGE_SIZE) { 154 errln("inline data cross block boundary @ nid %llu", 155 vi->nid); 156 DBG_BUGON(1); 157 err = -EFSCORRUPTED; 158 goto err_out; 159 } 160 161 map->m_flags |= EROFS_MAP_META; 162 } else { 163 errln("internal error @ nid: %llu (size %llu), m_la 0x%llx", 164 vi->nid, inode->i_size, map->m_la); 165 DBG_BUGON(1); 166 err = -EIO; 167 goto err_out; 168 } 169 170 out: 171 map->m_llen = map->m_plen; 172 173 err_out: 174 trace_erofs_map_blocks_flatmode_exit(inode, map, flags, 0); 175 return err; 176 } 177 178 int erofs_map_blocks(struct inode *inode, 179 struct erofs_map_blocks *map, int flags) 180 { 181 if (erofs_inode_is_data_compressed(EROFS_I(inode)->datalayout)) { 182 int err = z_erofs_map_blocks_iter(inode, map, flags); 183 184 if (map->mpage) { 185 put_page(map->mpage); 186 map->mpage = NULL; 187 } 188 return err; 189 } 190 return erofs_map_blocks_flatmode(inode, map, flags); 191 } 192 193 static inline struct bio *erofs_read_raw_page(struct bio *bio, 194 struct address_space *mapping, 195 struct page *page, 196 erofs_off_t *last_block, 197 unsigned int nblocks, 198 bool ra) 199 { 200 struct inode *const inode = mapping->host; 201 struct super_block *const sb = inode->i_sb; 202 erofs_off_t current_block = (erofs_off_t)page->index; 203 int err; 204 205 DBG_BUGON(!nblocks); 206 207 if (PageUptodate(page)) { 208 err = 0; 209 goto has_updated; 210 } 211 212 /* note that for readpage case, bio also equals to NULL */ 213 if (bio && 214 /* not continuous */ 215 *last_block + 1 != current_block) { 216 submit_bio_retry: 217 __submit_bio(bio, REQ_OP_READ, 0); 218 bio = NULL; 219 } 220 221 if (!bio) { 222 struct erofs_map_blocks map = { 223 .m_la = blknr_to_addr(current_block), 224 }; 225 erofs_blk_t blknr; 226 unsigned int blkoff; 227 228 err = erofs_map_blocks(inode, &map, EROFS_GET_BLOCKS_RAW); 229 if (err) 230 goto err_out; 231 232 /* zero out the holed page */ 233 if (!(map.m_flags & EROFS_MAP_MAPPED)) { 234 zero_user_segment(page, 0, PAGE_SIZE); 235 SetPageUptodate(page); 236 237 /* imply err = 0, see erofs_map_blocks */ 238 goto has_updated; 239 } 240 241 /* for RAW access mode, m_plen must be equal to m_llen */ 242 DBG_BUGON(map.m_plen != map.m_llen); 243 244 blknr = erofs_blknr(map.m_pa); 245 blkoff = erofs_blkoff(map.m_pa); 246 247 /* deal with inline page */ 248 if (map.m_flags & EROFS_MAP_META) { 249 void *vsrc, *vto; 250 struct page *ipage; 251 252 DBG_BUGON(map.m_plen > PAGE_SIZE); 253 254 ipage = erofs_get_meta_page(inode->i_sb, blknr, 0); 255 256 if (IS_ERR(ipage)) { 257 err = PTR_ERR(ipage); 258 goto err_out; 259 } 260 261 vsrc = kmap_atomic(ipage); 262 vto = kmap_atomic(page); 263 memcpy(vto, vsrc + blkoff, map.m_plen); 264 memset(vto + map.m_plen, 0, PAGE_SIZE - map.m_plen); 265 kunmap_atomic(vto); 266 kunmap_atomic(vsrc); 267 flush_dcache_page(page); 268 269 SetPageUptodate(page); 270 /* TODO: could we unlock the page earlier? */ 271 unlock_page(ipage); 272 put_page(ipage); 273 274 /* imply err = 0, see erofs_map_blocks */ 275 goto has_updated; 276 } 277 278 /* pa must be block-aligned for raw reading */ 279 DBG_BUGON(erofs_blkoff(map.m_pa)); 280 281 /* max # of continuous pages */ 282 if (nblocks > DIV_ROUND_UP(map.m_plen, PAGE_SIZE)) 283 nblocks = DIV_ROUND_UP(map.m_plen, PAGE_SIZE); 284 if (nblocks > BIO_MAX_PAGES) 285 nblocks = BIO_MAX_PAGES; 286 287 bio = erofs_grab_raw_bio(sb, blknr, nblocks); 288 } 289 290 err = bio_add_page(bio, page, PAGE_SIZE, 0); 291 /* out of the extent or bio is full */ 292 if (err < PAGE_SIZE) 293 goto submit_bio_retry; 294 295 *last_block = current_block; 296 297 /* shift in advance in case of it followed by too many gaps */ 298 if (bio->bi_iter.bi_size >= bio->bi_max_vecs * PAGE_SIZE) { 299 /* err should reassign to 0 after submitting */ 300 err = 0; 301 goto submit_bio_out; 302 } 303 304 return bio; 305 306 err_out: 307 /* for sync reading, set page error immediately */ 308 if (!ra) { 309 SetPageError(page); 310 ClearPageUptodate(page); 311 } 312 has_updated: 313 unlock_page(page); 314 315 /* if updated manually, continuous pages has a gap */ 316 if (bio) 317 submit_bio_out: 318 __submit_bio(bio, REQ_OP_READ, 0); 319 320 return err ? ERR_PTR(err) : NULL; 321 } 322 323 /* 324 * since we dont have write or truncate flows, so no inode 325 * locking needs to be held at the moment. 326 */ 327 static int erofs_raw_access_readpage(struct file *file, struct page *page) 328 { 329 erofs_off_t last_block; 330 struct bio *bio; 331 332 trace_erofs_readpage(page, true); 333 334 bio = erofs_read_raw_page(NULL, page->mapping, 335 page, &last_block, 1, false); 336 337 if (IS_ERR(bio)) 338 return PTR_ERR(bio); 339 340 DBG_BUGON(bio); /* since we have only one bio -- must be NULL */ 341 return 0; 342 } 343 344 static int erofs_raw_access_readpages(struct file *filp, 345 struct address_space *mapping, 346 struct list_head *pages, 347 unsigned int nr_pages) 348 { 349 erofs_off_t last_block; 350 struct bio *bio = NULL; 351 gfp_t gfp = readahead_gfp_mask(mapping); 352 struct page *page = list_last_entry(pages, struct page, lru); 353 354 trace_erofs_readpages(mapping->host, page, nr_pages, true); 355 356 for (; nr_pages; --nr_pages) { 357 page = list_entry(pages->prev, struct page, lru); 358 359 prefetchw(&page->flags); 360 list_del(&page->lru); 361 362 if (!add_to_page_cache_lru(page, mapping, page->index, gfp)) { 363 bio = erofs_read_raw_page(bio, mapping, page, 364 &last_block, nr_pages, true); 365 366 /* all the page errors are ignored when readahead */ 367 if (IS_ERR(bio)) { 368 pr_err("%s, readahead error at page %lu of nid %llu\n", 369 __func__, page->index, 370 EROFS_I(mapping->host)->nid); 371 372 bio = NULL; 373 } 374 } 375 376 /* pages could still be locked */ 377 put_page(page); 378 } 379 DBG_BUGON(!list_empty(pages)); 380 381 /* the rare case (end in gaps) */ 382 if (bio) 383 __submit_bio(bio, REQ_OP_READ, 0); 384 return 0; 385 } 386 387 static int erofs_get_block(struct inode *inode, sector_t iblock, 388 struct buffer_head *bh, int create) 389 { 390 struct erofs_map_blocks map = { 391 .m_la = iblock << 9, 392 }; 393 int err; 394 395 err = erofs_map_blocks(inode, &map, EROFS_GET_BLOCKS_RAW); 396 if (err) 397 return err; 398 399 if (map.m_flags & EROFS_MAP_MAPPED) 400 bh->b_blocknr = erofs_blknr(map.m_pa); 401 402 return err; 403 } 404 405 static sector_t erofs_bmap(struct address_space *mapping, sector_t block) 406 { 407 struct inode *inode = mapping->host; 408 409 if (EROFS_I(inode)->datalayout == EROFS_INODE_FLAT_INLINE) { 410 erofs_blk_t blks = i_size_read(inode) >> LOG_BLOCK_SIZE; 411 412 if (block >> LOG_SECTORS_PER_BLOCK >= blks) 413 return 0; 414 } 415 416 return generic_block_bmap(mapping, block, erofs_get_block); 417 } 418 419 /* for uncompressed (aligned) files and raw access for other files */ 420 const struct address_space_operations erofs_raw_access_aops = { 421 .readpage = erofs_raw_access_readpage, 422 .readpages = erofs_raw_access_readpages, 423 .bmap = erofs_bmap, 424 }; 425 426