drivers/staging/erofs/data.c

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