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/jbd2.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/aio.h>
  22 #include <linux/uio.h>
  23 #include <linux/bio.h>
  24 #include <linux/workqueue.h>
  25 #include <linux/kernel.h>
  26 #include <linux/slab.h>
  27 #include <linux/mm.h>
  28 #include <linux/ratelimit.h>
  29
  30 #include "ext4_jbd2.h"
  31 #include "xattr.h"
  32 #include "acl.h"
  33
  34 static struct kmem_cache *io_end_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         return 0;
  42 }
  43
  44 void ext4_exit_pageio(void)
  45 {
  46         kmem_cache_destroy(io_end_cachep);
  47 }
  48
  49 /*
  50  * Print an buffer I/O error compatible with the fs/buffer.c.  This
  51  * provides compatibility with dmesg scrapers that look for a specific
  52  * buffer I/O error message.  We really need a unified error reporting
  53  * structure to userspace ala Digital Unix's uerf system, but it's
  54  * probably not going to happen in my lifetime, due to LKML politics...
  55  */
  56 static void buffer_io_error(struct buffer_head *bh)
  57 {
  58         char b[BDEVNAME_SIZE];
  59         printk_ratelimited(KERN_ERR "Buffer I/O error on device %s, logical block %llu\n",
  60                         bdevname(bh->b_bdev, b),
  61                         (unsigned long long)bh->b_blocknr);
  62 }
  63
  64 static void ext4_finish_bio(struct bio *bio)
  65 {
  66         int i;
  67         int error = !test_bit(BIO_UPTODATE, &bio->bi_flags);
  68
  69         for (i = 0; i < bio->bi_vcnt; i++) {
  70                 struct bio_vec *bvec = &bio->bi_io_vec[i];
  71                 struct page *page = bvec->bv_page;
  72                 struct buffer_head *bh, *head;
  73                 unsigned bio_start = bvec->bv_offset;
  74                 unsigned bio_end = bio_start + bvec->bv_len;
  75                 unsigned under_io = 0;
  76                 unsigned long flags;
  77
  78                 if (!page)
  79                         continue;
  80
  81                 if (error) {
  82                         SetPageError(page);
  83                         set_bit(AS_EIO, &page->mapping->flags);
  84                 }
  85                 bh = head = page_buffers(page);
  86                 /*
  87                  * We check all buffers in the page under BH_Uptodate_Lock
  88                  * to avoid races with other end io clearing async_write flags
  89                  */
  90                 local_irq_save(flags);
  91                 bit_spin_lock(BH_Uptodate_Lock, &head->b_state);
  92                 do {
  93                         if (bh_offset(bh) < bio_start ||
  94                             bh_offset(bh) + bh->b_size > bio_end) {
  95                                 if (buffer_async_write(bh))
  96                                         under_io++;
  97                                 continue;
  98                         }
  99                         clear_buffer_async_write(bh);
 100                         if (error)
 101                                 buffer_io_error(bh);
 102                 } while ((bh = bh->b_this_page) != head);
 103                 bit_spin_unlock(BH_Uptodate_Lock, &head->b_state);
 104                 local_irq_restore(flags);
 105                 if (!under_io)
 106                         end_page_writeback(page);
 107         }
 108 }
 109
 110 static void ext4_release_io_end(ext4_io_end_t *io_end)
 111 {
 112         struct bio *bio, *next_bio;
 113
 114         BUG_ON(!list_empty(&io_end->list));
 115         BUG_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
 116         WARN_ON(io_end->handle);
 117
 118         if (atomic_dec_and_test(&EXT4_I(io_end->inode)->i_ioend_count))
 119                 wake_up_all(ext4_ioend_wq(io_end->inode));
 120
 121         for (bio = io_end->bio; bio; bio = next_bio) {
 122                 next_bio = bio->bi_private;
 123                 ext4_finish_bio(bio);
 124                 bio_put(bio);
 125         }
 126         if (io_end->flag & EXT4_IO_END_DIRECT)
 127                 inode_dio_done(io_end->inode);
 128         if (io_end->iocb)
 129                 aio_complete(io_end->iocb, io_end->result, 0);
 130         kmem_cache_free(io_end_cachep, io_end);
 131 }
 132
 133 static void ext4_clear_io_unwritten_flag(ext4_io_end_t *io_end)
 134 {
 135         struct inode *inode = io_end->inode;
 136
 137         io_end->flag &= ~EXT4_IO_END_UNWRITTEN;
 138         /* Wake up anyone waiting on unwritten extent conversion */
 139         if (atomic_dec_and_test(&EXT4_I(inode)->i_unwritten))
 140                 wake_up_all(ext4_ioend_wq(inode));
 141 }
 142
 143 /*
 144  * Check a range of space and convert unwritten extents to written. Note that
 145  * we are protected from truncate touching same part of extent tree by the
 146  * fact that truncate code waits for all DIO to finish (thus exclusion from
 147  * direct IO is achieved) and also waits for PageWriteback bits. Thus we
 148  * cannot get to ext4_ext_truncate() before all IOs overlapping that range are
 149  * completed (happens from ext4_free_ioend()).
 150  */
 151 static int ext4_end_io(ext4_io_end_t *io)
 152 {
 153         struct inode *inode = io->inode;
 154         loff_t offset = io->offset;
 155         ssize_t size = io->size;
 156         handle_t *handle = io->handle;
 157         int ret = 0;
 158
 159         ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
 160                    "list->prev 0x%p\n",
 161                    io, inode->i_ino, io->list.next, io->list.prev);
 162
 163         io->handle = NULL;      /* Following call will use up the handle */
 164         ret = ext4_convert_unwritten_extents(handle, inode, offset, size);
 165         if (ret < 0) {
 166                 ext4_msg(inode->i_sb, KERN_EMERG,
 167                          "failed to convert unwritten extents to written "
 168                          "extents -- potential data loss!  "
 169                          "(inode %lu, offset %llu, size %zd, error %d)",
 170                          inode->i_ino, offset, size, ret);
 171         }
 172         ext4_clear_io_unwritten_flag(io);
 173         ext4_release_io_end(io);
 174         return ret;
 175 }
 176
 177 static void dump_completed_IO(struct inode *inode, struct list_head *head)
 178 {
 179 #ifdef  EXT4FS_DEBUG
 180         struct list_head *cur, *before, *after;
 181         ext4_io_end_t *io, *io0, *io1;
 182
 183         if (list_empty(head))
 184                 return;
 185
 186         ext4_debug("Dump inode %lu completed io list\n", inode->i_ino);
 187         list_for_each_entry(io, head, list) {
 188                 cur = &io->list;
 189                 before = cur->prev;
 190                 io0 = container_of(before, ext4_io_end_t, list);
 191                 after = cur->next;
 192                 io1 = container_of(after, ext4_io_end_t, list);
 193
 194                 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
 195                             io, inode->i_ino, io0, io1);
 196         }
 197 #endif
 198 }
 199
 200 /* Add the io_end to per-inode completed end_io list. */
 201 static void ext4_add_complete_io(ext4_io_end_t *io_end)
 202 {
 203         struct ext4_inode_info *ei = EXT4_I(io_end->inode);
 204         struct workqueue_struct *wq;
 205         unsigned long flags;
 206
 207         BUG_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN));
 208         spin_lock_irqsave(&ei->i_completed_io_lock, flags);
 209         if (io_end->handle) {
 210                 wq = EXT4_SB(io_end->inode->i_sb)->rsv_conversion_wq;
 211                 if (list_empty(&ei->i_rsv_conversion_list))
 212                         queue_work(wq, &ei->i_rsv_conversion_work);
 213                 list_add_tail(&io_end->list, &ei->i_rsv_conversion_list);
 214         } else {
 215                 wq = EXT4_SB(io_end->inode->i_sb)->unrsv_conversion_wq;
 216                 if (list_empty(&ei->i_unrsv_conversion_list))
 217                         queue_work(wq, &ei->i_unrsv_conversion_work);
 218                 list_add_tail(&io_end->list, &ei->i_unrsv_conversion_list);
 219         }
 220         spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
 221 }
 222
 223 static int ext4_do_flush_completed_IO(struct inode *inode,
 224                                       struct list_head *head)
 225 {
 226         ext4_io_end_t *io;
 227         struct list_head unwritten;
 228         unsigned long flags;
 229         struct ext4_inode_info *ei = EXT4_I(inode);
 230         int err, ret = 0;
 231
 232         spin_lock_irqsave(&ei->i_completed_io_lock, flags);
 233         dump_completed_IO(inode, head);
 234         list_replace_init(head, &unwritten);
 235         spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
 236
 237         while (!list_empty(&unwritten)) {
 238                 io = list_entry(unwritten.next, ext4_io_end_t, list);
 239                 BUG_ON(!(io->flag & EXT4_IO_END_UNWRITTEN));
 240                 list_del_init(&io->list);
 241
 242                 err = ext4_end_io(io);
 243                 if (unlikely(!ret && err))
 244                         ret = err;
 245         }
 246         return ret;
 247 }
 248
 249 /*
 250  * work on completed IO, to convert unwritten extents to extents
 251  */
 252 void ext4_end_io_rsv_work(struct work_struct *work)
 253 {
 254         struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info,
 255                                                   i_rsv_conversion_work);
 256         ext4_do_flush_completed_IO(&ei->vfs_inode, &ei->i_rsv_conversion_list);
 257 }
 258
 259 void ext4_end_io_unrsv_work(struct work_struct *work)
 260 {
 261         struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info,
 262                                                   i_unrsv_conversion_work);
 263         ext4_do_flush_completed_IO(&ei->vfs_inode, &ei->i_unrsv_conversion_list);
 264 }
 265
 266 ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
 267 {
 268         ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags);
 269         if (io) {
 270                 atomic_inc(&EXT4_I(inode)->i_ioend_count);
 271                 io->inode = inode;
 272                 INIT_LIST_HEAD(&io->list);
 273                 atomic_set(&io->count, 1);
 274         }
 275         return io;
 276 }
 277
 278 void ext4_put_io_end_defer(ext4_io_end_t *io_end)
 279 {
 280         if (atomic_dec_and_test(&io_end->count)) {
 281                 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN) || !io_end->size) {
 282                         ext4_release_io_end(io_end);
 283                         return;
 284                 }
 285                 ext4_add_complete_io(io_end);
 286         }
 287 }
 288
 289 int ext4_put_io_end(ext4_io_end_t *io_end)
 290 {
 291         int err = 0;
 292
 293         if (atomic_dec_and_test(&io_end->count)) {
 294                 if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
 295                         err = ext4_convert_unwritten_extents(io_end->handle,
 296                                                 io_end->inode, io_end->offset,
 297                                                 io_end->size);
 298                         io_end->handle = NULL;
 299                         ext4_clear_io_unwritten_flag(io_end);
 300                 }
 301                 ext4_release_io_end(io_end);
 302         }
 303         return err;
 304 }
 305
 306 ext4_io_end_t *ext4_get_io_end(ext4_io_end_t *io_end)
 307 {
 308         atomic_inc(&io_end->count);
 309         return io_end;
 310 }
 311
 312 /* BIO completion function for page writeback */
 313 static void ext4_end_bio(struct bio *bio, int error)
 314 {
 315         ext4_io_end_t *io_end = bio->bi_private;
 316         sector_t bi_sector = bio->bi_sector;
 317
 318         BUG_ON(!io_end);
 319         bio->bi_end_io = NULL;
 320         if (test_bit(BIO_UPTODATE, &bio->bi_flags))
 321                 error = 0;
 322
 323         if (error) {
 324                 struct inode *inode = io_end->inode;
 325
 326                 ext4_warning(inode->i_sb, "I/O error writing to inode %lu "
 327                              "(offset %llu size %ld starting block %llu)",
 328                              inode->i_ino,
 329                              (unsigned long long) io_end->offset,
 330                              (long) io_end->size,
 331                              (unsigned long long)
 332                              bi_sector >> (inode->i_blkbits - 9));
 333         }
 334
 335         if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
 336                 /*
 337                  * Link bio into list hanging from io_end. We have to do it
 338                  * atomically as bio completions can be racing against each
 339                  * other.
 340                  */
 341                 bio->bi_private = xchg(&io_end->bio, bio);
 342                 ext4_put_io_end_defer(io_end);
 343         } else {
 344                 /*
 345                  * Drop io_end reference early. Inode can get freed once
 346                  * we finish the bio.
 347                  */
 348                 ext4_put_io_end_defer(io_end);
 349                 ext4_finish_bio(bio);
 350                 bio_put(bio);
 351         }
 352 }
 353
 354 void ext4_io_submit(struct ext4_io_submit *io)
 355 {
 356         struct bio *bio = io->io_bio;
 357
 358         if (bio) {
 359                 bio_get(io->io_bio);
 360                 submit_bio(io->io_op, io->io_bio);
 361                 BUG_ON(bio_flagged(io->io_bio, BIO_EOPNOTSUPP));
 362                 bio_put(io->io_bio);
 363         }
 364         io->io_bio = NULL;
 365 }
 366
 367 void ext4_io_submit_init(struct ext4_io_submit *io,
 368                          struct writeback_control *wbc)
 369 {
 370         io->io_op = (wbc->sync_mode == WB_SYNC_ALL ?  WRITE_SYNC : WRITE);
 371         io->io_bio = NULL;
 372         io->io_end = NULL;
 373 }
 374
 375 static int io_submit_init_bio(struct ext4_io_submit *io,
 376                               struct buffer_head *bh)
 377 {
 378         int nvecs = bio_get_nr_vecs(bh->b_bdev);
 379         struct bio *bio;
 380
 381         bio = bio_alloc(GFP_NOIO, min(nvecs, BIO_MAX_PAGES));
 382         if (!bio)
 383                 return -ENOMEM;
 384         bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
 385         bio->bi_bdev = bh->b_bdev;
 386         bio->bi_end_io = ext4_end_bio;
 387         bio->bi_private = ext4_get_io_end(io->io_end);
 388         io->io_bio = bio;
 389         io->io_next_block = bh->b_blocknr;
 390         return 0;
 391 }
 392
 393 static int io_submit_add_bh(struct ext4_io_submit *io,
 394                             struct inode *inode,
 395                             struct buffer_head *bh)
 396 {
 397         int ret;
 398
 399         if (io->io_bio && bh->b_blocknr != io->io_next_block) {
 400 submit_and_retry:
 401                 ext4_io_submit(io);
 402         }
 403         if (io->io_bio == NULL) {
 404                 ret = io_submit_init_bio(io, bh);
 405                 if (ret)
 406                         return ret;
 407         }
 408         ret = bio_add_page(io->io_bio, bh->b_page, bh->b_size, bh_offset(bh));
 409         if (ret != bh->b_size)
 410                 goto submit_and_retry;
 411         io->io_next_block++;
 412         return 0;
 413 }
 414
 415 int ext4_bio_write_page(struct ext4_io_submit *io,
 416                         struct page *page,
 417                         int len,
 418                         struct writeback_control *wbc)
 419 {
 420         struct inode *inode = page->mapping->host;
 421         unsigned block_start, blocksize;
 422         struct buffer_head *bh, *head;
 423         int ret = 0;
 424         int nr_submitted = 0;
 425
 426         blocksize = 1 << inode->i_blkbits;
 427
 428         BUG_ON(!PageLocked(page));
 429         BUG_ON(PageWriteback(page));
 430
 431         set_page_writeback(page);
 432         ClearPageError(page);
 433
 434         /*
 435          * In the first loop we prepare and mark buffers to submit. We have to
 436          * mark all buffers in the page before submitting so that
 437          * end_page_writeback() cannot be called from ext4_bio_end_io() when IO
 438          * on the first buffer finishes and we are still working on submitting
 439          * the second buffer.
 440          */
 441         bh = head = page_buffers(page);
 442         do {
 443                 block_start = bh_offset(bh);
 444                 if (block_start >= len) {
 445                         /*
 446                          * Comments copied from block_write_full_page_endio:
 447                          *
 448                          * The page straddles i_size.  It must be zeroed out on
 449                          * each and every writepage invocation because it may
 450                          * be mmapped.  "A file is mapped in multiples of the
 451                          * page size.  For a file that is not a multiple of
 452                          * the  page size, the remaining memory is zeroed when
 453                          * mapped, and writes to that region are not written
 454                          * out to the file."
 455                          */
 456                         zero_user_segment(page, block_start,
 457                                           block_start + blocksize);
 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                         unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
 474                 }
 475                 set_buffer_async_write(bh);
 476         } while ((bh = bh->b_this_page) != head);
 477
 478         /* Now submit buffers to write */
 479         bh = head = page_buffers(page);
 480         do {
 481                 if (!buffer_async_write(bh))
 482                         continue;
 483                 ret = io_submit_add_bh(io, inode, bh);
 484                 if (ret) {
 485                         /*
 486                          * We only get here on ENOMEM.  Not much else
 487                          * we can do but mark the page as dirty, and
 488                          * better luck next time.
 489                          */
 490                         redirty_page_for_writepage(wbc, page);
 491                         break;
 492                 }
 493                 nr_submitted++;
 494                 clear_buffer_dirty(bh);
 495         } while ((bh = bh->b_this_page) != head);
 496
 497         /* Error stopped previous loop? Clean up buffers... */
 498         if (ret) {
 499                 do {
 500                         clear_buffer_async_write(bh);
 501                         bh = bh->b_this_page;
 502                 } while (bh != head);
 503         }
 504         unlock_page(page);
 505         /* Nothing submitted - we have to end page writeback */
 506         if (!nr_submitted)
 507                 end_page_writeback(page);
 508         return ret;
 509 }