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
  35 int __init ext4_init_pageio(void)
  36 {
  37         io_end_cachep = KMEM_CACHE(ext4_io_end, SLAB_RECLAIM_ACCOUNT);
  38         if (io_end_cachep == NULL)
  39                 return -ENOMEM;
  40         return 0;
  41 }
  42
  43 void ext4_exit_pageio(void)
  44 {
  45         kmem_cache_destroy(io_end_cachep);
  46 }
  47
  48 /*
  49  * Print an buffer I/O error compatible with the fs/buffer.c.  This
  50  * provides compatibility with dmesg scrapers that look for a specific
  51  * buffer I/O error message.  We really need a unified error reporting
  52  * structure to userspace ala Digital Unix's uerf system, but it's
  53  * probably not going to happen in my lifetime, due to LKML politics...
  54  */
  55 static void buffer_io_error(struct buffer_head *bh)
  56 {
  57         printk_ratelimited(KERN_ERR "Buffer I/O error on device %pg, logical block %llu\n",
  58                        bh->b_bdev,
  59                         (unsigned long long)bh->b_blocknr);
  60 }
  61
  62 static void ext4_finish_bio(struct bio *bio)
  63 {
  64         int i;
  65         struct bio_vec *bvec;
  66         struct bvec_iter_all iter_all;
  67
  68         bio_for_each_segment_all(bvec, bio, i, iter_all) {
  69                 struct page *page = bvec->bv_page;
  70 #ifdef CONFIG_FS_ENCRYPTION
  71                 struct page *data_page = NULL;
  72 #endif
  73                 struct buffer_head *bh, *head;
  74                 unsigned bio_start = bvec->bv_offset;
  75                 unsigned bio_end = bio_start + bvec->bv_len;
  76                 unsigned under_io = 0;
  77                 unsigned long flags;
  78
  79                 if (!page)
  80                         continue;
  81
  82 #ifdef CONFIG_FS_ENCRYPTION
  83                 if (!page->mapping) {
  84                         /* The bounce data pages are unmapped. */
  85                         data_page = page;
  86                         fscrypt_pullback_bio_page(&page, false);
  87                 }
  88 #endif
  89
  90                 if (bio->bi_status) {
  91                         SetPageError(page);
  92                         mapping_set_error(page->mapping, -EIO);
  93                 }
  94                 bh = head = page_buffers(page);
  95                 /*
  96                  * We check all buffers in the page under BH_Uptodate_Lock
  97                  * to avoid races with other end io clearing async_write flags
  98                  */
  99                 local_irq_save(flags);
 100                 bit_spin_lock(BH_Uptodate_Lock, &head->b_state);
 101                 do {
 102                         if (bh_offset(bh) < bio_start ||
 103                             bh_offset(bh) + bh->b_size > bio_end) {
 104                                 if (buffer_async_write(bh))
 105                                         under_io++;
 106                                 continue;
 107                         }
 108                         clear_buffer_async_write(bh);
 109                         if (bio->bi_status)
 110                                 buffer_io_error(bh);
 111                 } while ((bh = bh->b_this_page) != head);
 112                 bit_spin_unlock(BH_Uptodate_Lock, &head->b_state);
 113                 local_irq_restore(flags);
 114                 if (!under_io) {
 115 #ifdef CONFIG_FS_ENCRYPTION
 116                         if (data_page)
 117                                 fscrypt_restore_control_page(data_page);
 118 #endif
 119                         end_page_writeback(page);
 120                 }
 121         }
 122 }
 123
 124 static void ext4_release_io_end(ext4_io_end_t *io_end)
 125 {
 126         struct bio *bio, *next_bio;
 127
 128         BUG_ON(!list_empty(&io_end->list));
 129         BUG_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
 130         WARN_ON(io_end->handle);
 131
 132         for (bio = io_end->bio; bio; bio = next_bio) {
 133                 next_bio = bio->bi_private;
 134                 ext4_finish_bio(bio);
 135                 bio_put(bio);
 136         }
 137         kmem_cache_free(io_end_cachep, io_end);
 138 }
 139
 140 /*
 141  * Check a range of space and convert unwritten extents to written. Note that
 142  * we are protected from truncate touching same part of extent tree by the
 143  * fact that truncate code waits for all DIO to finish (thus exclusion from
 144  * direct IO is achieved) and also waits for PageWriteback bits. Thus we
 145  * cannot get to ext4_ext_truncate() before all IOs overlapping that range are
 146  * completed (happens from ext4_free_ioend()).
 147  */
 148 static int ext4_end_io(ext4_io_end_t *io)
 149 {
 150         struct inode *inode = io->inode;
 151         loff_t offset = io->offset;
 152         ssize_t size = io->size;
 153         handle_t *handle = io->handle;
 154         int ret = 0;
 155
 156         ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
 157                    "list->prev 0x%p\n",
 158                    io, inode->i_ino, io->list.next, io->list.prev);
 159
 160         io->handle = NULL;      /* Following call will use up the handle */
 161         ret = ext4_convert_unwritten_extents(handle, inode, offset, size);
 162         if (ret < 0 && !ext4_forced_shutdown(EXT4_SB(inode->i_sb))) {
 163                 ext4_msg(inode->i_sb, KERN_EMERG,
 164                          "failed to convert unwritten extents to written "
 165                          "extents -- potential data loss!  "
 166                          "(inode %lu, offset %llu, size %zd, error %d)",
 167                          inode->i_ino, offset, size, ret);
 168         }
 169         ext4_clear_io_unwritten_flag(io);
 170         ext4_release_io_end(io);
 171         return ret;
 172 }
 173
 174 static void dump_completed_IO(struct inode *inode, struct list_head *head)
 175 {
 176 #ifdef  EXT4FS_DEBUG
 177         struct list_head *cur, *before, *after;
 178         ext4_io_end_t *io, *io0, *io1;
 179
 180         if (list_empty(head))
 181                 return;
 182
 183         ext4_debug("Dump inode %lu completed io list\n", inode->i_ino);
 184         list_for_each_entry(io, head, list) {
 185                 cur = &io->list;
 186                 before = cur->prev;
 187                 io0 = container_of(before, ext4_io_end_t, list);
 188                 after = cur->next;
 189                 io1 = container_of(after, ext4_io_end_t, list);
 190
 191                 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
 192                             io, inode->i_ino, io0, io1);
 193         }
 194 #endif
 195 }
 196
 197 /* Add the io_end to per-inode completed end_io list. */
 198 static void ext4_add_complete_io(ext4_io_end_t *io_end)
 199 {
 200         struct ext4_inode_info *ei = EXT4_I(io_end->inode);
 201         struct ext4_sb_info *sbi = EXT4_SB(io_end->inode->i_sb);
 202         struct workqueue_struct *wq;
 203         unsigned long flags;
 204
 205         /* Only reserved conversions from writeback should enter here */
 206         WARN_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN));
 207         WARN_ON(!io_end->handle && sbi->s_journal);
 208         spin_lock_irqsave(&ei->i_completed_io_lock, flags);
 209         wq = sbi->rsv_conversion_wq;
 210         if (list_empty(&ei->i_rsv_conversion_list))
 211                 queue_work(wq, &ei->i_rsv_conversion_work);
 212         list_add_tail(&io_end->list, &ei->i_rsv_conversion_list);
 213         spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
 214 }
 215
 216 static int ext4_do_flush_completed_IO(struct inode *inode,
 217                                       struct list_head *head)
 218 {
 219         ext4_io_end_t *io;
 220         struct list_head unwritten;
 221         unsigned long flags;
 222         struct ext4_inode_info *ei = EXT4_I(inode);
 223         int err, ret = 0;
 224
 225         spin_lock_irqsave(&ei->i_completed_io_lock, flags);
 226         dump_completed_IO(inode, head);
 227         list_replace_init(head, &unwritten);
 228         spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
 229
 230         while (!list_empty(&unwritten)) {
 231                 io = list_entry(unwritten.next, ext4_io_end_t, list);
 232                 BUG_ON(!(io->flag & EXT4_IO_END_UNWRITTEN));
 233                 list_del_init(&io->list);
 234
 235                 err = ext4_end_io(io);
 236                 if (unlikely(!ret && err))
 237                         ret = err;
 238         }
 239         return ret;
 240 }
 241
 242 /*
 243  * work on completed IO, to convert unwritten extents to extents
 244  */
 245 void ext4_end_io_rsv_work(struct work_struct *work)
 246 {
 247         struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info,
 248                                                   i_rsv_conversion_work);
 249         ext4_do_flush_completed_IO(&ei->vfs_inode, &ei->i_rsv_conversion_list);
 250 }
 251
 252 ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
 253 {
 254         ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags);
 255         if (io) {
 256                 io->inode = inode;
 257                 INIT_LIST_HEAD(&io->list);
 258                 atomic_set(&io->count, 1);
 259         }
 260         return io;
 261 }
 262
 263 void ext4_put_io_end_defer(ext4_io_end_t *io_end)
 264 {
 265         if (atomic_dec_and_test(&io_end->count)) {
 266                 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN) || !io_end->size) {
 267                         ext4_release_io_end(io_end);
 268                         return;
 269                 }
 270                 ext4_add_complete_io(io_end);
 271         }
 272 }
 273
 274 int ext4_put_io_end(ext4_io_end_t *io_end)
 275 {
 276         int err = 0;
 277
 278         if (atomic_dec_and_test(&io_end->count)) {
 279                 if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
 280                         err = ext4_convert_unwritten_extents(io_end->handle,
 281                                                 io_end->inode, io_end->offset,
 282                                                 io_end->size);
 283                         io_end->handle = NULL;
 284                         ext4_clear_io_unwritten_flag(io_end);
 285                 }
 286                 ext4_release_io_end(io_end);
 287         }
 288         return err;
 289 }
 290
 291 ext4_io_end_t *ext4_get_io_end(ext4_io_end_t *io_end)
 292 {
 293         atomic_inc(&io_end->count);
 294         return io_end;
 295 }
 296
 297 /* BIO completion function for page writeback */
 298 static void ext4_end_bio(struct bio *bio)
 299 {
 300         ext4_io_end_t *io_end = bio->bi_private;
 301         sector_t bi_sector = bio->bi_iter.bi_sector;
 302         char b[BDEVNAME_SIZE];
 303
 304         if (WARN_ONCE(!io_end, "io_end is NULL: %s: sector %Lu len %u err %d\n",
 305                       bio_devname(bio, b),
 306                       (long long) bio->bi_iter.bi_sector,
 307                       (unsigned) bio_sectors(bio),
 308                       bio->bi_status)) {
 309                 ext4_finish_bio(bio);
 310                 bio_put(bio);
 311                 return;
 312         }
 313         bio->bi_end_io = NULL;
 314
 315         if (bio->bi_status) {
 316                 struct inode *inode = io_end->inode;
 317
 318                 ext4_warning(inode->i_sb, "I/O error %d writing to inode %lu "
 319                              "(offset %llu size %ld starting block %llu)",
 320                              bio->bi_status, inode->i_ino,
 321                              (unsigned long long) io_end->offset,
 322                              (long) io_end->size,
 323                              (unsigned long long)
 324                              bi_sector >> (inode->i_blkbits - 9));
 325                 mapping_set_error(inode->i_mapping,
 326                                 blk_status_to_errno(bio->bi_status));
 327         }
 328
 329         if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
 330                 /*
 331                  * Link bio into list hanging from io_end. We have to do it
 332                  * atomically as bio completions can be racing against each
 333                  * other.
 334                  */
 335                 bio->bi_private = xchg(&io_end->bio, bio);
 336                 ext4_put_io_end_defer(io_end);
 337         } else {
 338                 /*
 339                  * Drop io_end reference early. Inode can get freed once
 340                  * we finish the bio.
 341                  */
 342                 ext4_put_io_end_defer(io_end);
 343                 ext4_finish_bio(bio);
 344                 bio_put(bio);
 345         }
 346 }
 347
 348 void ext4_io_submit(struct ext4_io_submit *io)
 349 {
 350         struct bio *bio = io->io_bio;
 351
 352         if (bio) {
 353                 int io_op_flags = io->io_wbc->sync_mode == WB_SYNC_ALL ?
 354                                   REQ_SYNC : 0;
 355                 io->io_bio->bi_write_hint = io->io_end->inode->i_write_hint;
 356                 bio_set_op_attrs(io->io_bio, REQ_OP_WRITE, io_op_flags);
 357                 submit_bio(io->io_bio);
 358         }
 359         io->io_bio = NULL;
 360 }
 361
 362 void ext4_io_submit_init(struct ext4_io_submit *io,
 363                          struct writeback_control *wbc)
 364 {
 365         io->io_wbc = wbc;
 366         io->io_bio = NULL;
 367         io->io_end = NULL;
 368 }
 369
 370 static int io_submit_init_bio(struct ext4_io_submit *io,
 371                               struct buffer_head *bh)
 372 {
 373         struct bio *bio;
 374
 375         bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES);
 376         if (!bio)
 377                 return -ENOMEM;
 378         bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
 379         bio_set_dev(bio, bh->b_bdev);
 380         bio->bi_end_io = ext4_end_bio;
 381         bio->bi_private = ext4_get_io_end(io->io_end);
 382         io->io_bio = bio;
 383         io->io_next_block = bh->b_blocknr;
 384         wbc_init_bio(io->io_wbc, bio);
 385         return 0;
 386 }
 387
 388 static int io_submit_add_bh(struct ext4_io_submit *io,
 389                             struct inode *inode,
 390                             struct page *page,
 391                             struct buffer_head *bh)
 392 {
 393         int ret;
 394
 395         if (io->io_bio && bh->b_blocknr != io->io_next_block) {
 396 submit_and_retry:
 397                 ext4_io_submit(io);
 398         }
 399         if (io->io_bio == NULL) {
 400                 ret = io_submit_init_bio(io, bh);
 401                 if (ret)
 402                         return ret;
 403                 io->io_bio->bi_write_hint = inode->i_write_hint;
 404         }
 405         ret = bio_add_page(io->io_bio, page, bh->b_size, bh_offset(bh));
 406         if (ret != bh->b_size)
 407                 goto submit_and_retry;
 408         wbc_account_io(io->io_wbc, page, bh->b_size);
 409         io->io_next_block++;
 410         return 0;
 411 }
 412
 413 int ext4_bio_write_page(struct ext4_io_submit *io,
 414                         struct page *page,
 415                         int len,
 416                         struct writeback_control *wbc,
 417                         bool keep_towrite)
 418 {
 419         struct page *data_page = NULL;
 420         struct inode *inode = page->mapping->host;
 421         unsigned block_start;
 422         struct buffer_head *bh, *head;
 423         int ret = 0;
 424         int nr_submitted = 0;
 425         int nr_to_submit = 0;
 426
 427         BUG_ON(!PageLocked(page));
 428         BUG_ON(PageWriteback(page));
 429
 430         if (keep_towrite)
 431                 set_page_writeback_keepwrite(page);
 432         else
 433                 set_page_writeback(page);
 434         ClearPageError(page);
 435
 436         /*
 437          * Comments copied from block_write_full_page:
 438          *
 439          * The page straddles i_size.  It must be zeroed out on each and every
 440          * writepage invocation because it may be mmapped.  "A file is mapped
 441          * in multiples of the page size.  For a file that is not a multiple of
 442          * the page size, the remaining memory is zeroed when mapped, and
 443          * writes to that region are not written out to the file."
 444          */
 445         if (len < PAGE_SIZE)
 446                 zero_user_segment(page, len, PAGE_SIZE);
 447         /*
 448          * In the first loop we prepare and mark buffers to submit. We have to
 449          * mark all buffers in the page before submitting so that
 450          * end_page_writeback() cannot be called from ext4_bio_end_io() when IO
 451          * on the first buffer finishes and we are still working on submitting
 452          * the second buffer.
 453          */
 454         bh = head = page_buffers(page);
 455         do {
 456                 block_start = bh_offset(bh);
 457                 if (block_start >= len) {
 458                         clear_buffer_dirty(bh);
 459                         set_buffer_uptodate(bh);
 460                         continue;
 461                 }
 462                 if (!buffer_dirty(bh) || buffer_delay(bh) ||
 463                     !buffer_mapped(bh) || buffer_unwritten(bh)) {
 464                         /* A hole? We can safely clear the dirty bit */
 465                         if (!buffer_mapped(bh))
 466                                 clear_buffer_dirty(bh);
 467                         if (io->io_bio)
 468                                 ext4_io_submit(io);
 469                         continue;
 470                 }
 471                 if (buffer_new(bh)) {
 472                         clear_buffer_new(bh);
 473                         clean_bdev_bh_alias(bh);
 474                 }
 475                 set_buffer_async_write(bh);
 476                 nr_to_submit++;
 477         } while ((bh = bh->b_this_page) != head);
 478
 479         bh = head = page_buffers(page);
 480
 481         if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode) && nr_to_submit) {
 482                 gfp_t gfp_flags = GFP_NOFS;
 483
 484         retry_encrypt:
 485                 data_page = fscrypt_encrypt_page(inode, page, PAGE_SIZE, 0,
 486                                                 page->index, gfp_flags);
 487                 if (IS_ERR(data_page)) {
 488                         ret = PTR_ERR(data_page);
 489                         if (ret == -ENOMEM && wbc->sync_mode == WB_SYNC_ALL) {
 490                                 if (io->io_bio) {
 491                                         ext4_io_submit(io);
 492                                         congestion_wait(BLK_RW_ASYNC, HZ/50);
 493                                 }
 494                                 gfp_flags |= __GFP_NOFAIL;
 495                                 goto retry_encrypt;
 496                         }
 497                         data_page = NULL;
 498                         goto out;
 499                 }
 500         }
 501
 502         /* Now submit buffers to write */
 503         do {
 504                 if (!buffer_async_write(bh))
 505                         continue;
 506                 ret = io_submit_add_bh(io, inode,
 507                                        data_page ? data_page : page, bh);
 508                 if (ret) {
 509                         /*
 510                          * We only get here on ENOMEM.  Not much else
 511                          * we can do but mark the page as dirty, and
 512                          * better luck next time.
 513                          */
 514                         break;
 515                 }
 516                 nr_submitted++;
 517                 clear_buffer_dirty(bh);
 518         } while ((bh = bh->b_this_page) != head);
 519
 520         /* Error stopped previous loop? Clean up buffers... */
 521         if (ret) {
 522         out:
 523                 if (data_page)
 524                         fscrypt_restore_control_page(data_page);
 525                 printk_ratelimited(KERN_ERR "%s: ret = %d\n", __func__, ret);
 526                 redirty_page_for_writepage(wbc, page);
 527                 do {
 528                         clear_buffer_async_write(bh);
 529                         bh = bh->b_this_page;
 530                 } while (bh != head);
 531         }
 532         unlock_page(page);
 533         /* Nothing submitted - we have to end page writeback */
 534         if (!nr_submitted)
 535                 end_page_writeback(page);
 536         return ret;
 537 }