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
  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  * This function is called by ext4_evict_inode() to make sure there is
  50  * no more pending I/O completion work left to do.
  51  */
  52 void ext4_ioend_shutdown(struct inode *inode)
  53 {
  54         wait_queue_head_t *wq = ext4_ioend_wq(inode);
  55
  56         wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_ioend_count) == 0));
  57         /*
  58          * We need to make sure the work structure is finished being
  59          * used before we let the inode get destroyed.
  60          */
  61         if (work_pending(&EXT4_I(inode)->i_rsv_conversion_work))
  62                 cancel_work_sync(&EXT4_I(inode)->i_rsv_conversion_work);
  63         if (work_pending(&EXT4_I(inode)->i_unrsv_conversion_work))
  64                 cancel_work_sync(&EXT4_I(inode)->i_unrsv_conversion_work);
  65 }
  66
  67 /*
  68  * Print an buffer I/O error compatible with the fs/buffer.c.  This
  69  * provides compatibility with dmesg scrapers that look for a specific
  70  * buffer I/O error message.  We really need a unified error reporting
  71  * structure to userspace ala Digital Unix's uerf system, but it's
  72  * probably not going to happen in my lifetime, due to LKML politics...
  73  */
  74 static void buffer_io_error(struct buffer_head *bh)
  75 {
  76         char b[BDEVNAME_SIZE];
  77         printk(KERN_ERR "Buffer I/O error on device %s, logical block %llu\n",
  78                         bdevname(bh->b_bdev, b),
  79                         (unsigned long long)bh->b_blocknr);
  80 }
  81
  82 static void ext4_finish_bio(struct bio *bio)
  83 {
  84         int i;
  85         int error = !test_bit(BIO_UPTODATE, &bio->bi_flags);
  86
  87         for (i = 0; i < bio->bi_vcnt; i++) {
  88                 struct bio_vec *bvec = &bio->bi_io_vec[i];
  89                 struct page *page = bvec->bv_page;
  90                 struct buffer_head *bh, *head;
  91                 unsigned bio_start = bvec->bv_offset;
  92                 unsigned bio_end = bio_start + bvec->bv_len;
  93                 unsigned under_io = 0;
  94                 unsigned long flags;
  95
  96                 if (!page)
  97                         continue;
  98
  99                 if (error) {
 100                         SetPageError(page);
 101                         set_bit(AS_EIO, &page->mapping->flags);
 102                 }
 103                 bh = head = page_buffers(page);
 104                 /*
 105                  * We check all buffers in the page under BH_Uptodate_Lock
 106                  * to avoid races with other end io clearing async_write flags
 107                  */
 108                 local_irq_save(flags);
 109                 bit_spin_lock(BH_Uptodate_Lock, &head->b_state);
 110                 do {
 111                         if (bh_offset(bh) < bio_start ||
 112                             bh_offset(bh) + bh->b_size > bio_end) {
 113                                 if (buffer_async_write(bh))
 114                                         under_io++;
 115                                 continue;
 116                         }
 117                         clear_buffer_async_write(bh);
 118                         if (error)
 119                                 buffer_io_error(bh);
 120                 } while ((bh = bh->b_this_page) != head);
 121                 bit_spin_unlock(BH_Uptodate_Lock, &head->b_state);
 122                 local_irq_restore(flags);
 123                 if (!under_io)
 124                         end_page_writeback(page);
 125         }
 126 }
 127
 128 static void ext4_release_io_end(ext4_io_end_t *io_end)
 129 {
 130         struct bio *bio, *next_bio;
 131
 132         BUG_ON(!list_empty(&io_end->list));
 133         BUG_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
 134         WARN_ON(io_end->handle);
 135
 136         if (atomic_dec_and_test(&EXT4_I(io_end->inode)->i_ioend_count))
 137                 wake_up_all(ext4_ioend_wq(io_end->inode));
 138
 139         for (bio = io_end->bio; bio; bio = next_bio) {
 140                 next_bio = bio->bi_private;
 141                 ext4_finish_bio(bio);
 142                 bio_put(bio);
 143         }
 144         if (io_end->flag & EXT4_IO_END_DIRECT)
 145                 inode_dio_done(io_end->inode);
 146         if (io_end->iocb)
 147                 aio_complete(io_end->iocb, io_end->result, 0);
 148         kmem_cache_free(io_end_cachep, io_end);
 149 }
 150
 151 static void ext4_clear_io_unwritten_flag(ext4_io_end_t *io_end)
 152 {
 153         struct inode *inode = io_end->inode;
 154
 155         io_end->flag &= ~EXT4_IO_END_UNWRITTEN;
 156         /* Wake up anyone waiting on unwritten extent conversion */
 157         if (atomic_dec_and_test(&EXT4_I(inode)->i_unwritten))
 158                 wake_up_all(ext4_ioend_wq(inode));
 159 }
 160
 161 /*
 162  * Check a range of space and convert unwritten extents to written. Note that
 163  * we are protected from truncate touching same part of extent tree by the
 164  * fact that truncate code waits for all DIO to finish (thus exclusion from
 165  * direct IO is achieved) and also waits for PageWriteback bits. Thus we
 166  * cannot get to ext4_ext_truncate() before all IOs overlapping that range are
 167  * completed (happens from ext4_free_ioend()).
 168  */
 169 static int ext4_end_io(ext4_io_end_t *io)
 170 {
 171         struct inode *inode = io->inode;
 172         loff_t offset = io->offset;
 173         ssize_t size = io->size;
 174         handle_t *handle = io->handle;
 175         int ret = 0;
 176
 177         ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
 178                    "list->prev 0x%p\n",
 179                    io, inode->i_ino, io->list.next, io->list.prev);
 180
 181         io->handle = NULL;      /* Following call will use up the handle */
 182         ret = ext4_convert_unwritten_extents(handle, inode, offset, size);
 183         if (ret < 0) {
 184                 ext4_msg(inode->i_sb, KERN_EMERG,
 185                          "failed to convert unwritten extents to written "
 186                          "extents -- potential data loss!  "
 187                          "(inode %lu, offset %llu, size %zd, error %d)",
 188                          inode->i_ino, offset, size, ret);
 189         }
 190         ext4_clear_io_unwritten_flag(io);
 191         ext4_release_io_end(io);
 192         return ret;
 193 }
 194
 195 static void dump_completed_IO(struct inode *inode, struct list_head *head)
 196 {
 197 #ifdef  EXT4FS_DEBUG
 198         struct list_head *cur, *before, *after;
 199         ext4_io_end_t *io, *io0, *io1;
 200
 201         if (list_empty(head))
 202                 return;
 203
 204         ext4_debug("Dump inode %lu completed io list\n", inode->i_ino);
 205         list_for_each_entry(io, head, list) {
 206                 cur = &io->list;
 207                 before = cur->prev;
 208                 io0 = container_of(before, ext4_io_end_t, list);
 209                 after = cur->next;
 210                 io1 = container_of(after, ext4_io_end_t, list);
 211
 212                 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
 213                             io, inode->i_ino, io0, io1);
 214         }
 215 #endif
 216 }
 217
 218 /* Add the io_end to per-inode completed end_io list. */
 219 static void ext4_add_complete_io(ext4_io_end_t *io_end)
 220 {
 221         struct ext4_inode_info *ei = EXT4_I(io_end->inode);
 222         struct workqueue_struct *wq;
 223         unsigned long flags;
 224
 225         BUG_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN));
 226         spin_lock_irqsave(&ei->i_completed_io_lock, flags);
 227         if (io_end->handle) {
 228                 wq = EXT4_SB(io_end->inode->i_sb)->rsv_conversion_wq;
 229                 if (list_empty(&ei->i_rsv_conversion_list))
 230                         queue_work(wq, &ei->i_rsv_conversion_work);
 231                 list_add_tail(&io_end->list, &ei->i_rsv_conversion_list);
 232         } else {
 233                 wq = EXT4_SB(io_end->inode->i_sb)->unrsv_conversion_wq;
 234                 if (list_empty(&ei->i_unrsv_conversion_list))
 235                         queue_work(wq, &ei->i_unrsv_conversion_work);
 236                 list_add_tail(&io_end->list, &ei->i_unrsv_conversion_list);
 237         }
 238         spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
 239 }
 240
 241 static int ext4_do_flush_completed_IO(struct inode *inode,
 242                                       struct list_head *head)
 243 {
 244         ext4_io_end_t *io;
 245         struct list_head unwritten;
 246         unsigned long flags;
 247         struct ext4_inode_info *ei = EXT4_I(inode);
 248         int err, ret = 0;
 249
 250         spin_lock_irqsave(&ei->i_completed_io_lock, flags);
 251         dump_completed_IO(inode, head);
 252         list_replace_init(head, &unwritten);
 253         spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
 254
 255         while (!list_empty(&unwritten)) {
 256                 io = list_entry(unwritten.next, ext4_io_end_t, list);
 257                 BUG_ON(!(io->flag & EXT4_IO_END_UNWRITTEN));
 258                 list_del_init(&io->list);
 259
 260                 err = ext4_end_io(io);
 261                 if (unlikely(!ret && err))
 262                         ret = err;
 263         }
 264         return ret;
 265 }
 266
 267 /*
 268  * work on completed IO, to convert unwritten extents to extents
 269  */
 270 void ext4_end_io_rsv_work(struct work_struct *work)
 271 {
 272         struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info,
 273                                                   i_rsv_conversion_work);
 274         ext4_do_flush_completed_IO(&ei->vfs_inode, &ei->i_rsv_conversion_list);
 275 }
 276
 277 void ext4_end_io_unrsv_work(struct work_struct *work)
 278 {
 279         struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info,
 280                                                   i_unrsv_conversion_work);
 281         ext4_do_flush_completed_IO(&ei->vfs_inode, &ei->i_unrsv_conversion_list);
 282 }
 283
 284 int ext4_flush_unwritten_io(struct inode *inode)
 285 {
 286         int ret, err;
 287
 288         WARN_ON_ONCE(!mutex_is_locked(&inode->i_mutex) &&
 289                      !(inode->i_state & I_FREEING));
 290         ret = ext4_do_flush_completed_IO(inode,
 291                                          &EXT4_I(inode)->i_rsv_conversion_list);
 292         err = ext4_do_flush_completed_IO(inode,
 293                                          &EXT4_I(inode)->i_unrsv_conversion_list);
 294         if (!ret)
 295                 ret = err;
 296         ext4_unwritten_wait(inode);
 297         return ret;
 298 }
 299
 300 ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
 301 {
 302         ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags);
 303         if (io) {
 304                 atomic_inc(&EXT4_I(inode)->i_ioend_count);
 305                 io->inode = inode;
 306                 INIT_LIST_HEAD(&io->list);
 307                 atomic_set(&io->count, 1);
 308         }
 309         return io;
 310 }
 311
 312 void ext4_put_io_end_defer(ext4_io_end_t *io_end)
 313 {
 314         if (atomic_dec_and_test(&io_end->count)) {
 315                 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN) || !io_end->size) {
 316                         ext4_release_io_end(io_end);
 317                         return;
 318                 }
 319                 ext4_add_complete_io(io_end);
 320         }
 321 }
 322
 323 int ext4_put_io_end(ext4_io_end_t *io_end)
 324 {
 325         int err = 0;
 326
 327         if (atomic_dec_and_test(&io_end->count)) {
 328                 if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
 329                         err = ext4_convert_unwritten_extents(io_end->handle,
 330                                                 io_end->inode, io_end->offset,
 331                                                 io_end->size);
 332                         io_end->handle = NULL;
 333                         ext4_clear_io_unwritten_flag(io_end);
 334                 }
 335                 ext4_release_io_end(io_end);
 336         }
 337         return err;
 338 }
 339
 340 ext4_io_end_t *ext4_get_io_end(ext4_io_end_t *io_end)
 341 {
 342         atomic_inc(&io_end->count);
 343         return io_end;
 344 }
 345
 346 static void ext4_end_bio(struct bio *bio, int error)
 347 {
 348         ext4_io_end_t *io_end = bio->bi_private;
 349         sector_t bi_sector = bio->bi_sector;
 350
 351         BUG_ON(!io_end);
 352         bio->bi_end_io = NULL;
 353         if (test_bit(BIO_UPTODATE, &bio->bi_flags))
 354                 error = 0;
 355
 356         if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
 357                 /*
 358                  * Link bio into list hanging from io_end. We have to do it
 359                  * atomically as bio completions can be racing against each
 360                  * other.
 361                  */
 362                 bio->bi_private = xchg(&io_end->bio, bio);
 363         } else {
 364                 ext4_finish_bio(bio);
 365                 bio_put(bio);
 366         }
 367
 368         if (error) {
 369                 struct inode *inode = io_end->inode;
 370
 371                 ext4_warning(inode->i_sb, "I/O error writing to inode %lu "
 372                              "(offset %llu size %ld starting block %llu)",
 373                              inode->i_ino,
 374                              (unsigned long long) io_end->offset,
 375                              (long) io_end->size,
 376                              (unsigned long long)
 377                              bi_sector >> (inode->i_blkbits - 9));
 378         }
 379         ext4_put_io_end_defer(io_end);
 380 }
 381
 382 void ext4_io_submit(struct ext4_io_submit *io)
 383 {
 384         struct bio *bio = io->io_bio;
 385
 386         if (bio) {
 387                 bio_get(io->io_bio);
 388                 submit_bio(io->io_op, io->io_bio);
 389                 BUG_ON(bio_flagged(io->io_bio, BIO_EOPNOTSUPP));
 390                 bio_put(io->io_bio);
 391         }
 392         io->io_bio = NULL;
 393 }
 394
 395 void ext4_io_submit_init(struct ext4_io_submit *io,
 396                          struct writeback_control *wbc)
 397 {
 398         io->io_op = (wbc->sync_mode == WB_SYNC_ALL ?  WRITE_SYNC : WRITE);
 399         io->io_bio = NULL;
 400         io->io_end = NULL;
 401 }
 402
 403 static int io_submit_init_bio(struct ext4_io_submit *io,
 404                               struct buffer_head *bh)
 405 {
 406         int nvecs = bio_get_nr_vecs(bh->b_bdev);
 407         struct bio *bio;
 408
 409         bio = bio_alloc(GFP_NOIO, min(nvecs, BIO_MAX_PAGES));
 410         bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
 411         bio->bi_bdev = bh->b_bdev;
 412         bio->bi_end_io = ext4_end_bio;
 413         bio->bi_private = ext4_get_io_end(io->io_end);
 414         io->io_bio = bio;
 415         io->io_next_block = bh->b_blocknr;
 416         return 0;
 417 }
 418
 419 static int io_submit_add_bh(struct ext4_io_submit *io,
 420                             struct inode *inode,
 421                             struct buffer_head *bh)
 422 {
 423         int ret;
 424
 425         if (io->io_bio && bh->b_blocknr != io->io_next_block) {
 426 submit_and_retry:
 427                 ext4_io_submit(io);
 428         }
 429         if (io->io_bio == NULL) {
 430                 ret = io_submit_init_bio(io, bh);
 431                 if (ret)
 432                         return ret;
 433         }
 434         ret = bio_add_page(io->io_bio, bh->b_page, bh->b_size, bh_offset(bh));
 435         if (ret != bh->b_size)
 436                 goto submit_and_retry;
 437         io->io_next_block++;
 438         return 0;
 439 }
 440
 441 int ext4_bio_write_page(struct ext4_io_submit *io,
 442                         struct page *page,
 443                         int len,
 444                         struct writeback_control *wbc)
 445 {
 446         struct inode *inode = page->mapping->host;
 447         unsigned block_start, blocksize;
 448         struct buffer_head *bh, *head;
 449         int ret = 0;
 450         int nr_submitted = 0;
 451
 452         blocksize = 1 << inode->i_blkbits;
 453
 454         BUG_ON(!PageLocked(page));
 455         BUG_ON(PageWriteback(page));
 456
 457         set_page_writeback(page);
 458         ClearPageError(page);
 459
 460         /*
 461          * In the first loop we prepare and mark buffers to submit. We have to
 462          * mark all buffers in the page before submitting so that
 463          * end_page_writeback() cannot be called from ext4_bio_end_io() when IO
 464          * on the first buffer finishes and we are still working on submitting
 465          * the second buffer.
 466          */
 467         bh = head = page_buffers(page);
 468         do {
 469                 block_start = bh_offset(bh);
 470                 if (block_start >= len) {
 471                         /*
 472                          * Comments copied from block_write_full_page_endio:
 473                          *
 474                          * The page straddles i_size.  It must be zeroed out on
 475                          * each and every writepage invocation because it may
 476                          * be mmapped.  "A file is mapped in multiples of the
 477                          * page size.  For a file that is not a multiple of
 478                          * the  page size, the remaining memory is zeroed when
 479                          * mapped, and writes to that region are not written
 480                          * out to the file."
 481                          */
 482                         zero_user_segment(page, block_start,
 483                                           block_start + blocksize);
 484                         clear_buffer_dirty(bh);
 485                         set_buffer_uptodate(bh);
 486                         continue;
 487                 }
 488                 if (!buffer_dirty(bh) || buffer_delay(bh) ||
 489                     !buffer_mapped(bh) || buffer_unwritten(bh)) {
 490                         /* A hole? We can safely clear the dirty bit */
 491                         if (!buffer_mapped(bh))
 492                                 clear_buffer_dirty(bh);
 493                         if (io->io_bio)
 494                                 ext4_io_submit(io);
 495                         continue;
 496                 }
 497                 if (buffer_new(bh)) {
 498                         clear_buffer_new(bh);
 499                         unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
 500                 }
 501                 set_buffer_async_write(bh);
 502         } while ((bh = bh->b_this_page) != head);
 503
 504         /* Now submit buffers to write */
 505         bh = head = page_buffers(page);
 506         do {
 507                 if (!buffer_async_write(bh))
 508                         continue;
 509                 ret = io_submit_add_bh(io, inode, bh);
 510                 if (ret) {
 511                         /*
 512                          * We only get here on ENOMEM.  Not much else
 513                          * we can do but mark the page as dirty, and
 514                          * better luck next time.
 515                          */
 516                         redirty_page_for_writepage(wbc, page);
 517                         break;
 518                 }
 519                 nr_submitted++;
 520                 clear_buffer_dirty(bh);
 521         } while ((bh = bh->b_this_page) != head);
 522
 523         /* Error stopped previous loop? Clean up buffers... */
 524         if (ret) {
 525                 do {
 526                         clear_buffer_async_write(bh);
 527                         bh = bh->b_this_page;
 528                 } while (bh != head);
 529         }
 530         unlock_page(page);
 531         /* Nothing submitted - we have to end page writeback */
 532         if (!nr_submitted)
 533                 end_page_writeback(page);
 534         return ret;
 535 }