fs/ext4/page-io.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * linux/fs/ext4/page-io.c
   4  *
   5  * This contains the new page_io functions for ext4
   6  *
   7  * Written by Theodore Ts'o, 2010.
   8  */
   9
  10 #include <linux/fs.h>
  11 #include <linux/time.h>
  12 #include <linux/highuid.h>
  13 #include <linux/pagemap.h>
  14 #include <linux/quotaops.h>
  15 #include <linux/string.h>
  16 #include <linux/buffer_head.h>
  17 #include <linux/writeback.h>
  18 #include <linux/pagevec.h>
  19 #include <linux/mpage.h>
  20 #include <linux/namei.h>
  21 #include <linux/uio.h>
  22 #include <linux/bio.h>
  23 #include <linux/workqueue.h>
  24 #include <linux/kernel.h>
  25 #include <linux/slab.h>
  26 #include <linux/mm.h>
  27 #include <linux/backing-dev.h>
  28
  29 #include "ext4_jbd2.h"
  30 #include "xattr.h"
  31 #include "acl.h"
  32
  33 static struct kmem_cache *io_end_cachep;
  34 static struct kmem_cache *io_end_vec_cachep;
  35
  36 int __init ext4_init_pageio(void)
  37 {
  38         io_end_cachep = KMEM_CACHE(ext4_io_end, SLAB_RECLAIM_ACCOUNT);
  39         if (io_end_cachep == NULL)
  40                 return -ENOMEM;
  41
  42         io_end_vec_cachep = KMEM_CACHE(ext4_io_end_vec, 0);
  43         if (io_end_vec_cachep == NULL) {
  44                 kmem_cache_destroy(io_end_cachep);
  45                 return -ENOMEM;
  46         }
  47         return 0;
  48 }
  49
  50 void ext4_exit_pageio(void)
  51 {
  52         kmem_cache_destroy(io_end_cachep);
  53         kmem_cache_destroy(io_end_vec_cachep);
  54 }
  55
  56 struct ext4_io_end_vec *ext4_alloc_io_end_vec(ext4_io_end_t *io_end)
  57 {
  58         struct ext4_io_end_vec *io_end_vec;
  59
  60         io_end_vec = kmem_cache_zalloc(io_end_vec_cachep, GFP_NOFS);
  61         if (!io_end_vec)
  62                 return ERR_PTR(-ENOMEM);
  63         INIT_LIST_HEAD(&io_end_vec->list);
  64         list_add_tail(&io_end_vec->list, &io_end->list_vec);
  65         return io_end_vec;
  66 }
  67
  68 static void ext4_free_io_end_vec(ext4_io_end_t *io_end)
  69 {
  70         struct ext4_io_end_vec *io_end_vec, *tmp;
  71
  72         if (list_empty(&io_end->list_vec))
  73                 return;
  74         list_for_each_entry_safe(io_end_vec, tmp, &io_end->list_vec, list) {
  75                 list_del(&io_end_vec->list);
  76                 kmem_cache_free(io_end_vec_cachep, io_end_vec);
  77         }
  78 }
  79
  80 struct ext4_io_end_vec *ext4_last_io_end_vec(ext4_io_end_t *io_end)
  81 {
  82         BUG_ON(list_empty(&io_end->list_vec));
  83         return list_last_entry(&io_end->list_vec, struct ext4_io_end_vec, list);
  84 }
  85
  86 /*
  87  * Print an buffer I/O error compatible with the fs/buffer.c.  This
  88  * provides compatibility with dmesg scrapers that look for a specific
  89  * buffer I/O error message.  We really need a unified error reporting
  90  * structure to userspace ala Digital Unix's uerf system, but it's
  91  * probably not going to happen in my lifetime, due to LKML politics...
  92  */
  93 static void buffer_io_error(struct buffer_head *bh)
  94 {
  95         printk_ratelimited(KERN_ERR "Buffer I/O error on device %pg, logical block %llu\n",
  96                        bh->b_bdev,
  97                         (unsigned long long)bh->b_blocknr);
  98 }
  99
 100 static void ext4_finish_bio(struct bio *bio)
 101 {
 102         struct bio_vec *bvec;
 103         struct bvec_iter_all iter_all;
 104
 105         bio_for_each_segment_all(bvec, bio, iter_all) {
 106                 struct page *page = bvec->bv_page;
 107                 struct page *bounce_page = NULL;
 108                 struct buffer_head *bh, *head;
 109                 unsigned bio_start = bvec->bv_offset;
 110                 unsigned bio_end = bio_start + bvec->bv_len;
 111                 unsigned under_io = 0;
 112                 unsigned long flags;
 113
 114                 if (fscrypt_is_bounce_page(page)) {
 115                         bounce_page = page;
 116                         page = fscrypt_pagecache_page(bounce_page);
 117                 }
 118
 119                 if (bio->bi_status) {
 120                         SetPageError(page);
 121                         mapping_set_error(page->mapping, -EIO);
 122                 }
 123                 bh = head = page_buffers(page);
 124                 /*
 125                  * We check all buffers in the page under b_uptodate_lock
 126                  * to avoid races with other end io clearing async_write flags
 127                  */
 128                 spin_lock_irqsave(&head->b_uptodate_lock, flags);
 129                 do {
 130                         if (bh_offset(bh) < bio_start ||
 131                             bh_offset(bh) + bh->b_size > bio_end) {
 132                                 if (buffer_async_write(bh))
 133                                         under_io++;
 134                                 continue;
 135                         }
 136                         clear_buffer_async_write(bh);
 137                         if (bio->bi_status)
 138                                 buffer_io_error(bh);
 139                 } while ((bh = bh->b_this_page) != head);
 140                 spin_unlock_irqrestore(&head->b_uptodate_lock, flags);
 141                 if (!under_io) {
 142                         fscrypt_free_bounce_page(bounce_page);
 143                         end_page_writeback(page);
 144                 }
 145         }
 146 }
 147
 148 static void ext4_release_io_end(ext4_io_end_t *io_end)
 149 {
 150         struct bio *bio, *next_bio;
 151
 152         BUG_ON(!list_empty(&io_end->list));
 153         BUG_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
 154         WARN_ON(io_end->handle);
 155
 156         for (bio = io_end->bio; bio; bio = next_bio) {
 157                 next_bio = bio->bi_private;
 158                 ext4_finish_bio(bio);
 159                 bio_put(bio);
 160         }
 161         ext4_free_io_end_vec(io_end);
 162         kmem_cache_free(io_end_cachep, io_end);
 163 }
 164
 165 /*
 166  * Check a range of space and convert unwritten extents to written. Note that
 167  * we are protected from truncate touching same part of extent tree by the
 168  * fact that truncate code waits for all DIO to finish (thus exclusion from
 169  * direct IO is achieved) and also waits for PageWriteback bits. Thus we
 170  * cannot get to ext4_ext_truncate() before all IOs overlapping that range are
 171  * completed (happens from ext4_free_ioend()).
 172  */
 173 static int ext4_end_io_end(ext4_io_end_t *io_end)
 174 {
 175         struct inode *inode = io_end->inode;
 176         handle_t *handle = io_end->handle;
 177         int ret = 0;
 178
 179         ext4_debug("ext4_end_io_nolock: io_end 0x%p from inode %lu,list->next 0x%p,"
 180                    "list->prev 0x%p\n",
 181                    io_end, inode->i_ino, io_end->list.next, io_end->list.prev);
 182
 183         io_end->handle = NULL;  /* Following call will use up the handle */
 184         ret = ext4_convert_unwritten_io_end_vec(handle, io_end);
 185         if (ret < 0 && !ext4_forced_shutdown(EXT4_SB(inode->i_sb))) {
 186                 ext4_msg(inode->i_sb, KERN_EMERG,
 187                          "failed to convert unwritten extents to written "
 188                          "extents -- potential data loss!  "
 189                          "(inode %lu, error %d)", inode->i_ino, ret);
 190         }
 191         ext4_clear_io_unwritten_flag(io_end);
 192         ext4_release_io_end(io_end);
 193         return ret;
 194 }
 195
 196 static void dump_completed_IO(struct inode *inode, struct list_head *head)
 197 {
 198 #ifdef  EXT4FS_DEBUG
 199         struct list_head *cur, *before, *after;
 200         ext4_io_end_t *io_end, *io_end0, *io_end1;
 201
 202         if (list_empty(head))
 203                 return;
 204
 205         ext4_debug("Dump inode %lu completed io list\n", inode->i_ino);
 206         list_for_each_entry(io_end, head, list) {
 207                 cur = &io_end->list;
 208                 before = cur->prev;
 209                 io_end0 = container_of(before, ext4_io_end_t, list);
 210                 after = cur->next;
 211                 io_end1 = container_of(after, ext4_io_end_t, list);
 212
 213                 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
 214                             io_end, inode->i_ino, io_end0, io_end1);
 215         }
 216 #endif
 217 }
 218
 219 /* Add the io_end to per-inode completed end_io list. */
 220 static void ext4_add_complete_io(ext4_io_end_t *io_end)
 221 {
 222         struct ext4_inode_info *ei = EXT4_I(io_end->inode);
 223         struct ext4_sb_info *sbi = EXT4_SB(io_end->inode->i_sb);
 224         struct workqueue_struct *wq;
 225         unsigned long flags;
 226
 227         /* Only reserved conversions from writeback should enter here */
 228         WARN_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN));
 229         WARN_ON(!io_end->handle && sbi->s_journal);
 230         spin_lock_irqsave(&ei->i_completed_io_lock, flags);
 231         wq = sbi->rsv_conversion_wq;
 232         if (list_empty(&ei->i_rsv_conversion_list))
 233                 queue_work(wq, &ei->i_rsv_conversion_work);
 234         list_add_tail(&io_end->list, &ei->i_rsv_conversion_list);
 235         spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
 236 }
 237
 238 static int ext4_do_flush_completed_IO(struct inode *inode,
 239                                       struct list_head *head)
 240 {
 241         ext4_io_end_t *io_end;
 242         struct list_head unwritten;
 243         unsigned long flags;
 244         struct ext4_inode_info *ei = EXT4_I(inode);
 245         int err, ret = 0;
 246
 247         spin_lock_irqsave(&ei->i_completed_io_lock, flags);
 248         dump_completed_IO(inode, head);
 249         list_replace_init(head, &unwritten);
 250         spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
 251
 252         while (!list_empty(&unwritten)) {
 253                 io_end = list_entry(unwritten.next, ext4_io_end_t, list);
 254                 BUG_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN));
 255                 list_del_init(&io_end->list);
 256
 257                 err = ext4_end_io_end(io_end);
 258                 if (unlikely(!ret && err))
 259                         ret = err;
 260         }
 261         return ret;
 262 }
 263
 264 /*
 265  * work on completed IO, to convert unwritten extents to extents
 266  */
 267 void ext4_end_io_rsv_work(struct work_struct *work)
 268 {
 269         struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info,
 270                                                   i_rsv_conversion_work);
 271         ext4_do_flush_completed_IO(&ei->vfs_inode, &ei->i_rsv_conversion_list);
 272 }
 273
 274 ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
 275 {
 276         ext4_io_end_t *io_end = kmem_cache_zalloc(io_end_cachep, flags);
 277
 278         if (io_end) {
 279                 io_end->inode = inode;
 280                 INIT_LIST_HEAD(&io_end->list);
 281                 INIT_LIST_HEAD(&io_end->list_vec);
 282                 atomic_set(&io_end->count, 1);
 283         }
 284         return io_end;
 285 }
 286
 287 void ext4_put_io_end_defer(ext4_io_end_t *io_end)
 288 {
 289         if (atomic_dec_and_test(&io_end->count)) {
 290                 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN) ||
 291                                 list_empty(&io_end->list_vec)) {
 292                         ext4_release_io_end(io_end);
 293                         return;
 294                 }
 295                 ext4_add_complete_io(io_end);
 296         }
 297 }
 298
 299 int ext4_put_io_end(ext4_io_end_t *io_end)
 300 {
 301         int err = 0;
 302
 303         if (atomic_dec_and_test(&io_end->count)) {
 304                 if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
 305                         err = ext4_convert_unwritten_io_end_vec(io_end->handle,
 306                                                                 io_end);
 307                         io_end->handle = NULL;
 308                         ext4_clear_io_unwritten_flag(io_end);
 309                 }
 310                 ext4_release_io_end(io_end);
 311         }
 312         return err;
 313 }
 314
 315 ext4_io_end_t *ext4_get_io_end(ext4_io_end_t *io_end)
 316 {
 317         atomic_inc(&io_end->count);
 318         return io_end;
 319 }
 320
 321 /* BIO completion function for page writeback */
 322 static void ext4_end_bio(struct bio *bio)
 323 {
 324         ext4_io_end_t *io_end = bio->bi_private;
 325         sector_t bi_sector = bio->bi_iter.bi_sector;
 326         char b[BDEVNAME_SIZE];
 327
 328         if (WARN_ONCE(!io_end, "io_end is NULL: %s: sector %Lu len %u err %d\n",
 329                       bio_devname(bio, b),
 330                       (long long) bio->bi_iter.bi_sector,
 331                       (unsigned) bio_sectors(bio),
 332                       bio->bi_status)) {
 333                 ext4_finish_bio(bio);
 334                 bio_put(bio);
 335                 return;
 336         }
 337         bio->bi_end_io = NULL;
 338
 339         if (bio->bi_status) {
 340                 struct inode *inode = io_end->inode;
 341
 342                 ext4_warning(inode->i_sb, "I/O error %d writing to inode %lu "
 343                              "starting block %llu)",
 344                              bio->bi_status, inode->i_ino,
 345                              (unsigned long long)
 346                              bi_sector >> (inode->i_blkbits - 9));
 347                 mapping_set_error(inode->i_mapping,
 348                                 blk_status_to_errno(bio->bi_status));
 349         }
 350
 351         if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
 352                 /*
 353                  * Link bio into list hanging from io_end. We have to do it
 354                  * atomically as bio completions can be racing against each
 355                  * other.
 356                  */
 357                 bio->bi_private = xchg(&io_end->bio, bio);
 358                 ext4_put_io_end_defer(io_end);
 359         } else {
 360                 /*
 361                  * Drop io_end reference early. Inode can get freed once
 362                  * we finish the bio.
 363                  */
 364                 ext4_put_io_end_defer(io_end);
 365                 ext4_finish_bio(bio);
 366                 bio_put(bio);
 367         }
 368 }
 369
 370 void ext4_io_submit(struct ext4_io_submit *io)
 371 {
 372         struct bio *bio = io->io_bio;
 373
 374         if (bio) {
 375                 int io_op_flags = io->io_wbc->sync_mode == WB_SYNC_ALL ?
 376                                   REQ_SYNC : 0;
 377                 io->io_bio->bi_write_hint = io->io_end->inode->i_write_hint;
 378                 bio_set_op_attrs(io->io_bio, REQ_OP_WRITE, io_op_flags);
 379                 submit_bio(io->io_bio);
 380         }
 381         io->io_bio = NULL;
 382 }
 383
 384 void ext4_io_submit_init(struct ext4_io_submit *io,
 385                          struct writeback_control *wbc)
 386 {
 387         io->io_wbc = wbc;
 388         io->io_bio = NULL;
 389         io->io_end = NULL;
 390 }
 391
 392 static void io_submit_init_bio(struct ext4_io_submit *io,
 393                                struct buffer_head *bh)
 394 {
 395         struct bio *bio;
 396
 397         /*
 398          * bio_alloc will _always_ be able to allocate a bio if
 399          * __GFP_DIRECT_RECLAIM is set, see comments for bio_alloc_bioset().
 400          */
 401         bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES);
 402         fscrypt_set_bio_crypt_ctx_bh(bio, bh, GFP_NOIO);
 403         bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
 404         bio_set_dev(bio, bh->b_bdev);
 405         bio->bi_end_io = ext4_end_bio;
 406         bio->bi_private = ext4_get_io_end(io->io_end);
 407         io->io_bio = bio;
 408         io->io_next_block = bh->b_blocknr;
 409         wbc_init_bio(io->io_wbc, bio);
 410 }
 411
 412 static void io_submit_add_bh(struct ext4_io_submit *io,
 413                              struct inode *inode,
 414                              struct page *page,
 415                              struct buffer_head *bh)
 416 {
 417         int ret;
 418
 419         if (io->io_bio && (bh->b_blocknr != io->io_next_block ||
 420                            !fscrypt_mergeable_bio_bh(io->io_bio, bh))) {
 421 submit_and_retry:
 422                 ext4_io_submit(io);
 423         }
 424         if (io->io_bio == NULL) {
 425                 io_submit_init_bio(io, bh);
 426                 io->io_bio->bi_write_hint = inode->i_write_hint;
 427         }
 428         ret = bio_add_page(io->io_bio, page, bh->b_size, bh_offset(bh));
 429         if (ret != bh->b_size)
 430                 goto submit_and_retry;
 431         wbc_account_cgroup_owner(io->io_wbc, page, bh->b_size);
 432         io->io_next_block++;
 433 }
 434
 435 int ext4_bio_write_page(struct ext4_io_submit *io,
 436                         struct page *page,
 437                         int len,
 438                         bool keep_towrite)
 439 {
 440         struct page *bounce_page = NULL;
 441         struct inode *inode = page->mapping->host;
 442         unsigned block_start;
 443         struct buffer_head *bh, *head;
 444         int ret = 0;
 445         int nr_submitted = 0;
 446         int nr_to_submit = 0;
 447         struct writeback_control *wbc = io->io_wbc;
 448
 449         BUG_ON(!PageLocked(page));
 450         BUG_ON(PageWriteback(page));
 451
 452         if (keep_towrite)
 453                 set_page_writeback_keepwrite(page);
 454         else
 455                 set_page_writeback(page);
 456         ClearPageError(page);
 457
 458         /*
 459          * Comments copied from block_write_full_page:
 460          *
 461          * The page straddles i_size.  It must be zeroed out on each and every
 462          * writepage invocation because it may be mmapped.  "A file is mapped
 463          * in multiples of the page size.  For a file that is not a multiple of
 464          * the page size, the remaining memory is zeroed when mapped, and
 465          * writes to that region are not written out to the file."
 466          */
 467         if (len < PAGE_SIZE)
 468                 zero_user_segment(page, len, PAGE_SIZE);
 469         /*
 470          * In the first loop we prepare and mark buffers to submit. We have to
 471          * mark all buffers in the page before submitting so that
 472          * end_page_writeback() cannot be called from ext4_bio_end_io() when IO
 473          * on the first buffer finishes and we are still working on submitting
 474          * the second buffer.
 475          */
 476         bh = head = page_buffers(page);
 477         do {
 478                 block_start = bh_offset(bh);
 479                 if (block_start >= len) {
 480                         clear_buffer_dirty(bh);
 481                         set_buffer_uptodate(bh);
 482                         continue;
 483                 }
 484                 if (!buffer_dirty(bh) || buffer_delay(bh) ||
 485                     !buffer_mapped(bh) || buffer_unwritten(bh)) {
 486                         /* A hole? We can safely clear the dirty bit */
 487                         if (!buffer_mapped(bh))
 488                                 clear_buffer_dirty(bh);
 489                         if (io->io_bio)
 490                                 ext4_io_submit(io);
 491                         continue;
 492                 }
 493                 if (buffer_new(bh))
 494                         clear_buffer_new(bh);
 495                 set_buffer_async_write(bh);
 496                 nr_to_submit++;
 497         } while ((bh = bh->b_this_page) != head);
 498
 499         bh = head = page_buffers(page);
 500
 501         /*
 502          * If any blocks are being written to an encrypted file, encrypt them
 503          * into a bounce page.  For simplicity, just encrypt until the last
 504          * block which might be needed.  This may cause some unneeded blocks
 505          * (e.g. holes) to be unnecessarily encrypted, but this is rare and
 506          * can't happen in the common case of blocksize == PAGE_SIZE.
 507          */
 508         if (fscrypt_inode_uses_fs_layer_crypto(inode) && nr_to_submit) {
 509                 gfp_t gfp_flags = GFP_NOFS;
 510                 unsigned int enc_bytes = round_up(len, i_blocksize(inode));
 511
 512                 /*
 513                  * Since bounce page allocation uses a mempool, we can only use
 514                  * a waiting mask (i.e. request guaranteed allocation) on the
 515                  * first page of the bio.  Otherwise it can deadlock.
 516                  */
 517                 if (io->io_bio)
 518                         gfp_flags = GFP_NOWAIT | __GFP_NOWARN;
 519         retry_encrypt:
 520                 bounce_page = fscrypt_encrypt_pagecache_blocks(page, enc_bytes,
 521                                                                0, gfp_flags);
 522                 if (IS_ERR(bounce_page)) {
 523                         ret = PTR_ERR(bounce_page);
 524                         if (ret == -ENOMEM &&
 525                             (io->io_bio || wbc->sync_mode == WB_SYNC_ALL)) {
 526                                 gfp_flags = GFP_NOFS;
 527                                 if (io->io_bio)
 528                                         ext4_io_submit(io);
 529                                 else
 530                                         gfp_flags |= __GFP_NOFAIL;
 531                                 congestion_wait(BLK_RW_ASYNC, HZ/50);
 532                                 goto retry_encrypt;
 533                         }
 534
 535                         printk_ratelimited(KERN_ERR "%s: ret = %d\n", __func__, ret);
 536                         redirty_page_for_writepage(wbc, page);
 537                         do {
 538                                 clear_buffer_async_write(bh);
 539                                 bh = bh->b_this_page;
 540                         } while (bh != head);
 541                         goto unlock;
 542                 }
 543         }
 544
 545         /* Now submit buffers to write */
 546         do {
 547                 if (!buffer_async_write(bh))
 548                         continue;
 549                 io_submit_add_bh(io, inode,
 550                                  bounce_page ? bounce_page : page, bh);
 551                 nr_submitted++;
 552                 clear_buffer_dirty(bh);
 553         } while ((bh = bh->b_this_page) != head);
 554
 555 unlock:
 556         unlock_page(page);
 557         /* Nothing submitted - we have to end page writeback */
 558         if (!nr_submitted)
 559                 end_page_writeback(page);
 560         return ret;
 561 }