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/uio.h>
  21 #include <linux/bio.h>
  22 #include <linux/workqueue.h>
  23 #include <linux/kernel.h>
  24 #include <linux/slab.h>
  25 #include <linux/mm.h>
  26
  27 #include "ext4_jbd2.h"
  28 #include "xattr.h"
  29 #include "acl.h"
  30
  31 static struct kmem_cache *io_end_cachep;
  32
  33 int __init ext4_init_pageio(void)
  34 {
  35         io_end_cachep = KMEM_CACHE(ext4_io_end, SLAB_RECLAIM_ACCOUNT);
  36         if (io_end_cachep == NULL)
  37                 return -ENOMEM;
  38         return 0;
  39 }
  40
  41 void ext4_exit_pageio(void)
  42 {
  43         kmem_cache_destroy(io_end_cachep);
  44 }
  45
  46 /*
  47  * Print an buffer I/O error compatible with the fs/buffer.c.  This
  48  * provides compatibility with dmesg scrapers that look for a specific
  49  * buffer I/O error message.  We really need a unified error reporting
  50  * structure to userspace ala Digital Unix's uerf system, but it's
  51  * probably not going to happen in my lifetime, due to LKML politics...
  52  */
  53 static void buffer_io_error(struct buffer_head *bh)
  54 {
  55         char b[BDEVNAME_SIZE];
  56         printk_ratelimited(KERN_ERR "Buffer I/O error on device %s, logical block %llu\n",
  57                         bdevname(bh->b_bdev, b),
  58                         (unsigned long long)bh->b_blocknr);
  59 }
  60
  61 static void ext4_finish_bio(struct bio *bio)
  62 {
  63         int i;
  64         int error = !test_bit(BIO_UPTODATE, &bio->bi_flags);
  65         struct bio_vec *bvec;
  66
  67         bio_for_each_segment_all(bvec, bio, i) {
  68                 struct page *page = bvec->bv_page;
  69 #ifdef CONFIG_EXT4_FS_ENCRYPTION
  70                 struct page *data_page = NULL;
  71                 struct ext4_crypto_ctx *ctx = 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_EXT4_FS_ENCRYPTION
  83                 if (!page->mapping) {
  84                         /* The bounce data pages are unmapped. */
  85                         data_page = page;
  86                         ctx = (struct ext4_crypto_ctx *)page_private(data_page);
  87                         page = ctx->w.control_page;
  88                 }
  89 #endif
  90
  91                 if (error) {
  92                         SetPageError(page);
  93                         set_bit(AS_EIO, &page->mapping->flags);
  94                 }
  95                 bh = head = page_buffers(page);
  96                 /*
  97                  * We check all buffers in the page under BH_Uptodate_Lock
  98                  * to avoid races with other end io clearing async_write flags
  99                  */
 100                 local_irq_save(flags);
 101                 bit_spin_lock(BH_Uptodate_Lock, &head->b_state);
 102                 do {
 103                         if (bh_offset(bh) < bio_start ||
 104                             bh_offset(bh) + bh->b_size > bio_end) {
 105                                 if (buffer_async_write(bh))
 106                                         under_io++;
 107                                 continue;
 108                         }
 109                         clear_buffer_async_write(bh);
 110                         if (error)
 111                                 buffer_io_error(bh);
 112                 } while ((bh = bh->b_this_page) != head);
 113                 bit_spin_unlock(BH_Uptodate_Lock, &head->b_state);
 114                 local_irq_restore(flags);
 115                 if (!under_io) {
 116 #ifdef CONFIG_EXT4_FS_ENCRYPTION
 117                         if (ctx)
 118                                 ext4_restore_control_page(data_page);
 119 #endif
 120                         end_page_writeback(page);
 121                 }
 122         }
 123 }
 124
 125 static void ext4_release_io_end(ext4_io_end_t *io_end)
 126 {
 127         struct bio *bio, *next_bio;
 128
 129         BUG_ON(!list_empty(&io_end->list));
 130         BUG_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
 131         WARN_ON(io_end->handle);
 132
 133         if (atomic_dec_and_test(&EXT4_I(io_end->inode)->i_ioend_count))
 134                 wake_up_all(ext4_ioend_wq(io_end->inode));
 135
 136         for (bio = io_end->bio; bio; bio = next_bio) {
 137                 next_bio = bio->bi_private;
 138                 ext4_finish_bio(bio);
 139                 bio_put(bio);
 140         }
 141         kmem_cache_free(io_end_cachep, io_end);
 142 }
 143
 144 static void ext4_clear_io_unwritten_flag(ext4_io_end_t *io_end)
 145 {
 146         struct inode *inode = io_end->inode;
 147
 148         io_end->flag &= ~EXT4_IO_END_UNWRITTEN;
 149         /* Wake up anyone waiting on unwritten extent conversion */
 150         if (atomic_dec_and_test(&EXT4_I(inode)->i_unwritten))
 151                 wake_up_all(ext4_ioend_wq(inode));
 152 }
 153
 154 /*
 155  * Check a range of space and convert unwritten extents to written. Note that
 156  * we are protected from truncate touching same part of extent tree by the
 157  * fact that truncate code waits for all DIO to finish (thus exclusion from
 158  * direct IO is achieved) and also waits for PageWriteback bits. Thus we
 159  * cannot get to ext4_ext_truncate() before all IOs overlapping that range are
 160  * completed (happens from ext4_free_ioend()).
 161  */
 162 static int ext4_end_io(ext4_io_end_t *io)
 163 {
 164         struct inode *inode = io->inode;
 165         loff_t offset = io->offset;
 166         ssize_t size = io->size;
 167         handle_t *handle = io->handle;
 168         int ret = 0;
 169
 170         ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
 171                    "list->prev 0x%p\n",
 172                    io, inode->i_ino, io->list.next, io->list.prev);
 173
 174         io->handle = NULL;      /* Following call will use up the handle */
 175         ret = ext4_convert_unwritten_extents(handle, inode, offset, size);
 176         if (ret < 0) {
 177                 ext4_msg(inode->i_sb, KERN_EMERG,
 178                          "failed to convert unwritten extents to written "
 179                          "extents -- potential data loss!  "
 180                          "(inode %lu, offset %llu, size %zd, error %d)",
 181                          inode->i_ino, offset, size, ret);
 182         }
 183         ext4_clear_io_unwritten_flag(io);
 184         ext4_release_io_end(io);
 185         return ret;
 186 }
 187
 188 static void dump_completed_IO(struct inode *inode, struct list_head *head)
 189 {
 190 #ifdef  EXT4FS_DEBUG
 191         struct list_head *cur, *before, *after;
 192         ext4_io_end_t *io, *io0, *io1;
 193
 194         if (list_empty(head))
 195                 return;
 196
 197         ext4_debug("Dump inode %lu completed io list\n", inode->i_ino);
 198         list_for_each_entry(io, head, list) {
 199                 cur = &io->list;
 200                 before = cur->prev;
 201                 io0 = container_of(before, ext4_io_end_t, list);
 202                 after = cur->next;
 203                 io1 = container_of(after, ext4_io_end_t, list);
 204
 205                 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
 206                             io, inode->i_ino, io0, io1);
 207         }
 208 #endif
 209 }
 210
 211 /* Add the io_end to per-inode completed end_io list. */
 212 static void ext4_add_complete_io(ext4_io_end_t *io_end)
 213 {
 214         struct ext4_inode_info *ei = EXT4_I(io_end->inode);
 215         struct ext4_sb_info *sbi = EXT4_SB(io_end->inode->i_sb);
 216         struct workqueue_struct *wq;
 217         unsigned long flags;
 218
 219         /* Only reserved conversions from writeback should enter here */
 220         WARN_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN));
 221         WARN_ON(!io_end->handle && sbi->s_journal);
 222         spin_lock_irqsave(&ei->i_completed_io_lock, flags);
 223         wq = sbi->rsv_conversion_wq;
 224         if (list_empty(&ei->i_rsv_conversion_list))
 225                 queue_work(wq, &ei->i_rsv_conversion_work);
 226         list_add_tail(&io_end->list, &ei->i_rsv_conversion_list);
 227         spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
 228 }
 229
 230 static int ext4_do_flush_completed_IO(struct inode *inode,
 231                                       struct list_head *head)
 232 {
 233         ext4_io_end_t *io;
 234         struct list_head unwritten;
 235         unsigned long flags;
 236         struct ext4_inode_info *ei = EXT4_I(inode);
 237         int err, ret = 0;
 238
 239         spin_lock_irqsave(&ei->i_completed_io_lock, flags);
 240         dump_completed_IO(inode, head);
 241         list_replace_init(head, &unwritten);
 242         spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
 243
 244         while (!list_empty(&unwritten)) {
 245                 io = list_entry(unwritten.next, ext4_io_end_t, list);
 246                 BUG_ON(!(io->flag & EXT4_IO_END_UNWRITTEN));
 247                 list_del_init(&io->list);
 248
 249                 err = ext4_end_io(io);
 250                 if (unlikely(!ret && err))
 251                         ret = err;
 252         }
 253         return ret;
 254 }
 255
 256 /*
 257  * work on completed IO, to convert unwritten extents to extents
 258  */
 259 void ext4_end_io_rsv_work(struct work_struct *work)
 260 {
 261         struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info,
 262                                                   i_rsv_conversion_work);
 263         ext4_do_flush_completed_IO(&ei->vfs_inode, &ei->i_rsv_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_iter.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 %d writing to inode %lu "
 327                              "(offset %llu size %ld starting block %llu)",
 328                              error, 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                 mapping_set_error(inode->i_mapping, error);
 334         }
 335
 336         if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
 337                 /*
 338                  * Link bio into list hanging from io_end. We have to do it
 339                  * atomically as bio completions can be racing against each
 340                  * other.
 341                  */
 342                 bio->bi_private = xchg(&io_end->bio, bio);
 343                 ext4_put_io_end_defer(io_end);
 344         } else {
 345                 /*
 346                  * Drop io_end reference early. Inode can get freed once
 347                  * we finish the bio.
 348                  */
 349                 ext4_put_io_end_defer(io_end);
 350                 ext4_finish_bio(bio);
 351                 bio_put(bio);
 352         }
 353 }
 354
 355 void ext4_io_submit(struct ext4_io_submit *io)
 356 {
 357         struct bio *bio = io->io_bio;
 358
 359         if (bio) {
 360                 bio_get(io->io_bio);
 361                 submit_bio(io->io_op, io->io_bio);
 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_iter.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 page *page,
 396                             struct buffer_head *bh)
 397 {
 398         int ret;
 399
 400         if (io->io_bio && bh->b_blocknr != io->io_next_block) {
 401 submit_and_retry:
 402                 ext4_io_submit(io);
 403         }
 404         if (io->io_bio == NULL) {
 405                 ret = io_submit_init_bio(io, bh);
 406                 if (ret)
 407                         return ret;
 408         }
 409         ret = bio_add_page(io->io_bio, page, bh->b_size, bh_offset(bh));
 410         if (ret != bh->b_size)
 411                 goto submit_and_retry;
 412         io->io_next_block++;
 413         return 0;
 414 }
 415
 416 int ext4_bio_write_page(struct ext4_io_submit *io,
 417                         struct page *page,
 418                         int len,
 419                         struct writeback_control *wbc,
 420                         bool keep_towrite)
 421 {
 422         struct page *data_page = NULL;
 423         struct inode *inode = page->mapping->host;
 424         unsigned block_start, blocksize;
 425         struct buffer_head *bh, *head;
 426         int ret = 0;
 427         int nr_submitted = 0;
 428
 429         blocksize = 1 << inode->i_blkbits;
 430
 431         BUG_ON(!PageLocked(page));
 432         BUG_ON(PageWriteback(page));
 433
 434         if (keep_towrite)
 435                 set_page_writeback_keepwrite(page);
 436         else
 437                 set_page_writeback(page);
 438         ClearPageError(page);
 439
 440         /*
 441          * Comments copied from block_write_full_page:
 442          *
 443          * The page straddles i_size.  It must be zeroed out on each and every
 444          * writepage invocation because it may be mmapped.  "A file is mapped
 445          * in multiples of the page size.  For a file that is not a multiple of
 446          * the page size, the remaining memory is zeroed when mapped, and
 447          * writes to that region are not written out to the file."
 448          */
 449         if (len < PAGE_CACHE_SIZE)
 450                 zero_user_segment(page, len, PAGE_CACHE_SIZE);
 451         /*
 452          * In the first loop we prepare and mark buffers to submit. We have to
 453          * mark all buffers in the page before submitting so that
 454          * end_page_writeback() cannot be called from ext4_bio_end_io() when IO
 455          * on the first buffer finishes and we are still working on submitting
 456          * the second buffer.
 457          */
 458         bh = head = page_buffers(page);
 459         do {
 460                 block_start = bh_offset(bh);
 461                 if (block_start >= len) {
 462                         clear_buffer_dirty(bh);
 463                         set_buffer_uptodate(bh);
 464                         continue;
 465                 }
 466                 if (!buffer_dirty(bh) || buffer_delay(bh) ||
 467                     !buffer_mapped(bh) || buffer_unwritten(bh)) {
 468                         /* A hole? We can safely clear the dirty bit */
 469                         if (!buffer_mapped(bh))
 470                                 clear_buffer_dirty(bh);
 471                         if (io->io_bio)
 472                                 ext4_io_submit(io);
 473                         continue;
 474                 }
 475                 if (buffer_new(bh)) {
 476                         clear_buffer_new(bh);
 477                         unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
 478                 }
 479                 set_buffer_async_write(bh);
 480         } while ((bh = bh->b_this_page) != head);
 481
 482         bh = head = page_buffers(page);
 483
 484         if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode)) {
 485                 data_page = ext4_encrypt(inode, page);
 486                 if (IS_ERR(data_page)) {
 487                         ret = PTR_ERR(data_page);
 488                         data_page = NULL;
 489                         goto out;
 490                 }
 491         }
 492
 493         /* Now submit buffers to write */
 494         do {
 495                 if (!buffer_async_write(bh))
 496                         continue;
 497                 ret = io_submit_add_bh(io, inode,
 498                                        data_page ? data_page : page, bh);
 499                 if (ret) {
 500                         /*
 501                          * We only get here on ENOMEM.  Not much else
 502                          * we can do but mark the page as dirty, and
 503                          * better luck next time.
 504                          */
 505                         break;
 506                 }
 507                 nr_submitted++;
 508                 clear_buffer_dirty(bh);
 509         } while ((bh = bh->b_this_page) != head);
 510
 511         /* Error stopped previous loop? Clean up buffers... */
 512         if (ret) {
 513         out:
 514                 if (data_page)
 515                         ext4_restore_control_page(data_page);
 516                 printk_ratelimited(KERN_ERR "%s: ret = %d\n", __func__, ret);
 517                 redirty_page_for_writepage(wbc, page);
 518                 do {
 519                         clear_buffer_async_write(bh);
 520                         bh = bh->b_this_page;
 521                 } while (bh != head);
 522         }
 523         unlock_page(page);
 524         /* Nothing submitted - we have to end page writeback */
 525         if (!nr_submitted)
 526                 end_page_writeback(page);
 527         return ret;
 528 }