fs/ext4/page-io.c

   1 /*
   2  * linux/fs/ext4/page-io.c
   3  *
   4  * This contains the new page_io functions for ext4
   5  *
   6  * Written by Theodore Ts'o, 2010.
   7  */
   8
   9 #include <linux/fs.h>
  10 #include <linux/time.h>
  11 #include <linux/highuid.h>
  12 #include <linux/pagemap.h>
  13 #include <linux/quotaops.h>
  14 #include <linux/string.h>
  15 #include <linux/buffer_head.h>
  16 #include <linux/writeback.h>
  17 #include <linux/pagevec.h>
  18 #include <linux/mpage.h>
  19 #include <linux/namei.h>
  20 #include <linux/aio.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
  28 #include "ext4_jbd2.h"
  29 #include "xattr.h"
  30 #include "acl.h"
  31
  32 static struct kmem_cache *io_end_cachep;
  33
  34 int __init ext4_init_pageio(void)
  35 {
  36         io_end_cachep = KMEM_CACHE(ext4_io_end, SLAB_RECLAIM_ACCOUNT);
  37         if (io_end_cachep == NULL)
  38                 return -ENOMEM;
  39         return 0;
  40 }
  41
  42 void ext4_exit_pageio(void)
  43 {
  44         kmem_cache_destroy(io_end_cachep);
  45 }
  46
  47 /*
  48  * Print an buffer I/O error compatible with the fs/buffer.c.  This
  49  * provides compatibility with dmesg scrapers that look for a specific
  50  * buffer I/O error message.  We really need a unified error reporting
  51  * structure to userspace ala Digital Unix's uerf system, but it's
  52  * probably not going to happen in my lifetime, due to LKML politics...
  53  */
  54 static void buffer_io_error(struct buffer_head *bh)
  55 {
  56         char b[BDEVNAME_SIZE];
  57         printk_ratelimited(KERN_ERR "Buffer I/O error on device %s, logical block %llu\n",
  58                         bdevname(bh->b_bdev, b),
  59                         (unsigned long long)bh->b_blocknr);
  60 }
  61
  62 static void ext4_finish_bio(struct bio *bio)
  63 {
  64         int i;
  65         int error = !test_bit(BIO_UPTODATE, &bio->bi_flags);
  66         struct bio_vec *bvec;
  67
  68         bio_for_each_segment_all(bvec, bio, i) {
  69                 struct page *page = bvec->bv_page;
  70                 struct buffer_head *bh, *head;
  71                 unsigned bio_start = bvec->bv_offset;
  72                 unsigned bio_end = bio_start + bvec->bv_len;
  73                 unsigned under_io = 0;
  74                 unsigned long flags;
  75
  76                 if (!page)
  77                         continue;
  78
  79                 if (error) {
  80                         SetPageError(page);
  81                         set_bit(AS_EIO, &page->mapping->flags);
  82                 }
  83                 bh = head = page_buffers(page);
  84                 /*
  85                  * We check all buffers in the page under BH_Uptodate_Lock
  86                  * to avoid races with other end io clearing async_write flags
  87                  */
  88                 local_irq_save(flags);
  89                 bit_spin_lock(BH_Uptodate_Lock, &head->b_state);
  90                 do {
  91                         if (bh_offset(bh) < bio_start ||
  92                             bh_offset(bh) + bh->b_size > bio_end) {
  93                                 if (buffer_async_write(bh))
  94                                         under_io++;
  95                                 continue;
  96                         }
  97                         clear_buffer_async_write(bh);
  98                         if (error)
  99                                 buffer_io_error(bh);
 100                 } while ((bh = bh->b_this_page) != head);
 101                 bit_spin_unlock(BH_Uptodate_Lock, &head->b_state);
 102                 local_irq_restore(flags);
 103                 if (!under_io)
 104                         end_page_writeback(page);
 105         }
 106 }
 107
 108 static void ext4_release_io_end(ext4_io_end_t *io_end)
 109 {
 110         struct bio *bio, *next_bio;
 111
 112         BUG_ON(!list_empty(&io_end->list));
 113         BUG_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
 114         WARN_ON(io_end->handle);
 115
 116         if (atomic_dec_and_test(&EXT4_I(io_end->inode)->i_ioend_count))
 117                 wake_up_all(ext4_ioend_wq(io_end->inode));
 118
 119         for (bio = io_end->bio; bio; bio = next_bio) {
 120                 next_bio = bio->bi_private;
 121                 ext4_finish_bio(bio);
 122                 bio_put(bio);
 123         }
 124         kmem_cache_free(io_end_cachep, io_end);
 125 }
 126
 127 static void ext4_clear_io_unwritten_flag(ext4_io_end_t *io_end)
 128 {
 129         struct inode *inode = io_end->inode;
 130
 131         io_end->flag &= ~EXT4_IO_END_UNWRITTEN;
 132         /* Wake up anyone waiting on unwritten extent conversion */
 133         if (atomic_dec_and_test(&EXT4_I(inode)->i_unwritten))
 134                 wake_up_all(ext4_ioend_wq(inode));
 135 }
 136
 137 /*
 138  * Check a range of space and convert unwritten extents to written. Note that
 139  * we are protected from truncate touching same part of extent tree by the
 140  * fact that truncate code waits for all DIO to finish (thus exclusion from
 141  * direct IO is achieved) and also waits for PageWriteback bits. Thus we
 142  * cannot get to ext4_ext_truncate() before all IOs overlapping that range are
 143  * completed (happens from ext4_free_ioend()).
 144  */
 145 static int ext4_end_io(ext4_io_end_t *io)
 146 {
 147         struct inode *inode = io->inode;
 148         loff_t offset = io->offset;
 149         ssize_t size = io->size;
 150         handle_t *handle = io->handle;
 151         int ret = 0;
 152
 153         ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
 154                    "list->prev 0x%p\n",
 155                    io, inode->i_ino, io->list.next, io->list.prev);
 156
 157         io->handle = NULL;      /* Following call will use up the handle */
 158         ret = ext4_convert_unwritten_extents(handle, inode, offset, size);
 159         if (ret < 0) {
 160                 ext4_msg(inode->i_sb, KERN_EMERG,
 161                          "failed to convert unwritten extents to written "
 162                          "extents -- potential data loss!  "
 163                          "(inode %lu, offset %llu, size %zd, error %d)",
 164                          inode->i_ino, offset, size, ret);
 165         }
 166         ext4_clear_io_unwritten_flag(io);
 167         ext4_release_io_end(io);
 168         return ret;
 169 }
 170
 171 static void dump_completed_IO(struct inode *inode, struct list_head *head)
 172 {
 173 #ifdef  EXT4FS_DEBUG
 174         struct list_head *cur, *before, *after;
 175         ext4_io_end_t *io, *io0, *io1;
 176
 177         if (list_empty(head))
 178                 return;
 179
 180         ext4_debug("Dump inode %lu completed io list\n", inode->i_ino);
 181         list_for_each_entry(io, head, list) {
 182                 cur = &io->list;
 183                 before = cur->prev;
 184                 io0 = container_of(before, ext4_io_end_t, list);
 185                 after = cur->next;
 186                 io1 = container_of(after, ext4_io_end_t, list);
 187
 188                 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
 189                             io, inode->i_ino, io0, io1);
 190         }
 191 #endif
 192 }
 193
 194 /* Add the io_end to per-inode completed end_io list. */
 195 static void ext4_add_complete_io(ext4_io_end_t *io_end)
 196 {
 197         struct ext4_inode_info *ei = EXT4_I(io_end->inode);
 198         struct ext4_sb_info *sbi = EXT4_SB(io_end->inode->i_sb);
 199         struct workqueue_struct *wq;
 200         unsigned long flags;
 201
 202         /* Only reserved conversions from writeback should enter here */
 203         WARN_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN));
 204         WARN_ON(!io_end->handle && sbi->s_journal);
 205         spin_lock_irqsave(&ei->i_completed_io_lock, flags);
 206         wq = sbi->rsv_conversion_wq;
 207         if (list_empty(&ei->i_rsv_conversion_list))
 208                 queue_work(wq, &ei->i_rsv_conversion_work);
 209         list_add_tail(&io_end->list, &ei->i_rsv_conversion_list);
 210         spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
 211 }
 212
 213 static int ext4_do_flush_completed_IO(struct inode *inode,
 214                                       struct list_head *head)
 215 {
 216         ext4_io_end_t *io;
 217         struct list_head unwritten;
 218         unsigned long flags;
 219         struct ext4_inode_info *ei = EXT4_I(inode);
 220         int err, ret = 0;
 221
 222         spin_lock_irqsave(&ei->i_completed_io_lock, flags);
 223         dump_completed_IO(inode, head);
 224         list_replace_init(head, &unwritten);
 225         spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
 226
 227         while (!list_empty(&unwritten)) {
 228                 io = list_entry(unwritten.next, ext4_io_end_t, list);
 229                 BUG_ON(!(io->flag & EXT4_IO_END_UNWRITTEN));
 230                 list_del_init(&io->list);
 231
 232                 err = ext4_end_io(io);
 233                 if (unlikely(!ret && err))
 234                         ret = err;
 235         }
 236         return ret;
 237 }
 238
 239 /*
 240  * work on completed IO, to convert unwritten extents to extents
 241  */
 242 void ext4_end_io_rsv_work(struct work_struct *work)
 243 {
 244         struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info,
 245                                                   i_rsv_conversion_work);
 246         ext4_do_flush_completed_IO(&ei->vfs_inode, &ei->i_rsv_conversion_list);
 247 }
 248
 249 ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
 250 {
 251         ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags);
 252         if (io) {
 253                 atomic_inc(&EXT4_I(inode)->i_ioend_count);
 254                 io->inode = inode;
 255                 INIT_LIST_HEAD(&io->list);
 256                 atomic_set(&io->count, 1);
 257         }
 258         return io;
 259 }
 260
 261 void ext4_put_io_end_defer(ext4_io_end_t *io_end)
 262 {
 263         if (atomic_dec_and_test(&io_end->count)) {
 264                 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN) || !io_end->size) {
 265                         ext4_release_io_end(io_end);
 266                         return;
 267                 }
 268                 ext4_add_complete_io(io_end);
 269         }
 270 }
 271
 272 int ext4_put_io_end(ext4_io_end_t *io_end)
 273 {
 274         int err = 0;
 275
 276         if (atomic_dec_and_test(&io_end->count)) {
 277                 if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
 278                         err = ext4_convert_unwritten_extents(io_end->handle,
 279                                                 io_end->inode, io_end->offset,
 280                                                 io_end->size);
 281                         io_end->handle = NULL;
 282                         ext4_clear_io_unwritten_flag(io_end);
 283                 }
 284                 ext4_release_io_end(io_end);
 285         }
 286         return err;
 287 }
 288
 289 ext4_io_end_t *ext4_get_io_end(ext4_io_end_t *io_end)
 290 {
 291         atomic_inc(&io_end->count);
 292         return io_end;
 293 }
 294
 295 /* BIO completion function for page writeback */
 296 static void ext4_end_bio(struct bio *bio, int error)
 297 {
 298         ext4_io_end_t *io_end = bio->bi_private;
 299         sector_t bi_sector = bio->bi_iter.bi_sector;
 300
 301         BUG_ON(!io_end);
 302         bio->bi_end_io = NULL;
 303         if (test_bit(BIO_UPTODATE, &bio->bi_flags))
 304                 error = 0;
 305
 306         if (error) {
 307                 struct inode *inode = io_end->inode;
 308
 309                 ext4_warning(inode->i_sb, "I/O error %d writing to inode %lu "
 310                              "(offset %llu size %ld starting block %llu)",
 311                              error, inode->i_ino,
 312                              (unsigned long long) io_end->offset,
 313                              (long) io_end->size,
 314                              (unsigned long long)
 315                              bi_sector >> (inode->i_blkbits - 9));
 316                 mapping_set_error(inode->i_mapping, error);
 317         }
 318
 319         if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
 320                 /*
 321                  * Link bio into list hanging from io_end. We have to do it
 322                  * atomically as bio completions can be racing against each
 323                  * other.
 324                  */
 325                 bio->bi_private = xchg(&io_end->bio, bio);
 326                 ext4_put_io_end_defer(io_end);
 327         } else {
 328                 /*
 329                  * Drop io_end reference early. Inode can get freed once
 330                  * we finish the bio.
 331                  */
 332                 ext4_put_io_end_defer(io_end);
 333                 ext4_finish_bio(bio);
 334                 bio_put(bio);
 335         }
 336 }
 337
 338 void ext4_io_submit(struct ext4_io_submit *io)
 339 {
 340         struct bio *bio = io->io_bio;
 341
 342         if (bio) {
 343                 bio_get(io->io_bio);
 344                 submit_bio(io->io_op, io->io_bio);
 345                 BUG_ON(bio_flagged(io->io_bio, BIO_EOPNOTSUPP));
 346                 bio_put(io->io_bio);
 347         }
 348         io->io_bio = NULL;
 349 }
 350
 351 void ext4_io_submit_init(struct ext4_io_submit *io,
 352                          struct writeback_control *wbc)
 353 {
 354         io->io_op = (wbc->sync_mode == WB_SYNC_ALL ?  WRITE_SYNC : WRITE);
 355         io->io_bio = NULL;
 356         io->io_end = NULL;
 357 }
 358
 359 static int io_submit_init_bio(struct ext4_io_submit *io,
 360                               struct buffer_head *bh)
 361 {
 362         int nvecs = bio_get_nr_vecs(bh->b_bdev);
 363         struct bio *bio;
 364
 365         bio = bio_alloc(GFP_NOIO, min(nvecs, BIO_MAX_PAGES));
 366         if (!bio)
 367                 return -ENOMEM;
 368         bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
 369         bio->bi_bdev = bh->b_bdev;
 370         bio->bi_end_io = ext4_end_bio;
 371         bio->bi_private = ext4_get_io_end(io->io_end);
 372         io->io_bio = bio;
 373         io->io_next_block = bh->b_blocknr;
 374         return 0;
 375 }
 376
 377 static int io_submit_add_bh(struct ext4_io_submit *io,
 378                             struct inode *inode,
 379                             struct buffer_head *bh)
 380 {
 381         int ret;
 382
 383         if (io->io_bio && bh->b_blocknr != io->io_next_block) {
 384 submit_and_retry:
 385                 ext4_io_submit(io);
 386         }
 387         if (io->io_bio == NULL) {
 388                 ret = io_submit_init_bio(io, bh);
 389                 if (ret)
 390                         return ret;
 391         }
 392         ret = bio_add_page(io->io_bio, bh->b_page, bh->b_size, bh_offset(bh));
 393         if (ret != bh->b_size)
 394                 goto submit_and_retry;
 395         io->io_next_block++;
 396         return 0;
 397 }
 398
 399 int ext4_bio_write_page(struct ext4_io_submit *io,
 400                         struct page *page,
 401                         int len,
 402                         struct writeback_control *wbc,
 403                         bool keep_towrite)
 404 {
 405         struct inode *inode = page->mapping->host;
 406         unsigned block_start, blocksize;
 407         struct buffer_head *bh, *head;
 408         int ret = 0;
 409         int nr_submitted = 0;
 410
 411         blocksize = 1 << inode->i_blkbits;
 412
 413         BUG_ON(!PageLocked(page));
 414         BUG_ON(PageWriteback(page));
 415
 416         if (keep_towrite)
 417                 set_page_writeback_keepwrite(page);
 418         else
 419                 set_page_writeback(page);
 420         ClearPageError(page);
 421
 422         /*
 423          * Comments copied from block_write_full_page:
 424          *
 425          * The page straddles i_size.  It must be zeroed out on each and every
 426          * writepage invocation because it may be mmapped.  "A file is mapped
 427          * in multiples of the page size.  For a file that is not a multiple of
 428          * the page size, the remaining memory is zeroed when mapped, and
 429          * writes to that region are not written out to the file."
 430          */
 431         if (len < PAGE_CACHE_SIZE)
 432                 zero_user_segment(page, len, PAGE_CACHE_SIZE);
 433         /*
 434          * In the first loop we prepare and mark buffers to submit. We have to
 435          * mark all buffers in the page before submitting so that
 436          * end_page_writeback() cannot be called from ext4_bio_end_io() when IO
 437          * on the first buffer finishes and we are still working on submitting
 438          * the second buffer.
 439          */
 440         bh = head = page_buffers(page);
 441         do {
 442                 block_start = bh_offset(bh);
 443                 if (block_start >= len) {
 444                         clear_buffer_dirty(bh);
 445                         set_buffer_uptodate(bh);
 446                         continue;
 447                 }
 448                 if (!buffer_dirty(bh) || buffer_delay(bh) ||
 449                     !buffer_mapped(bh) || buffer_unwritten(bh)) {
 450                         /* A hole? We can safely clear the dirty bit */
 451                         if (!buffer_mapped(bh))
 452                                 clear_buffer_dirty(bh);
 453                         if (io->io_bio)
 454                                 ext4_io_submit(io);
 455                         continue;
 456                 }
 457                 if (buffer_new(bh)) {
 458                         clear_buffer_new(bh);
 459                         unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
 460                 }
 461                 set_buffer_async_write(bh);
 462         } while ((bh = bh->b_this_page) != head);
 463
 464         /* Now submit buffers to write */
 465         bh = head = page_buffers(page);
 466         do {
 467                 if (!buffer_async_write(bh))
 468                         continue;
 469                 ret = io_submit_add_bh(io, inode, bh);
 470                 if (ret) {
 471                         /*
 472                          * We only get here on ENOMEM.  Not much else
 473                          * we can do but mark the page as dirty, and
 474                          * better luck next time.
 475                          */
 476                         redirty_page_for_writepage(wbc, page);
 477                         break;
 478                 }
 479                 nr_submitted++;
 480                 clear_buffer_dirty(bh);
 481         } while ((bh = bh->b_this_page) != head);
 482
 483         /* Error stopped previous loop? Clean up buffers... */
 484         if (ret) {
 485                 do {
 486                         clear_buffer_async_write(bh);
 487                         bh = bh->b_this_page;
 488                 } while (bh != head);
 489         }
 490         unlock_page(page);
 491         /* Nothing submitted - we have to end page writeback */
 492         if (!nr_submitted)
 493                 end_page_writeback(page);
 494         return ret;
 495 }