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 #include <linux/backing-dev.h>
  27 #include <linux/fscrypto.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
  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 #endif
  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 #ifdef CONFIG_EXT4_FS_ENCRYPTION
  82                 if (!page->mapping) {
  83                         /* The bounce data pages are unmapped. */
  84                         data_page = page;
  85                         fscrypt_pullback_bio_page(&page, false);
  86                 }
  87 #endif
  88
  89                 if (bio->bi_error) {
  90                         SetPageError(page);
  91                         set_bit(AS_EIO, &page->mapping->flags);
  92                 }
  93                 bh = head = page_buffers(page);
  94                 /*
  95                  * We check all buffers in the page under BH_Uptodate_Lock
  96                  * to avoid races with other end io clearing async_write flags
  97                  */
  98                 local_irq_save(flags);
  99                 bit_spin_lock(BH_Uptodate_Lock, &head->b_state);
 100                 do {
 101                         if (bh_offset(bh) < bio_start ||
 102                             bh_offset(bh) + bh->b_size > bio_end) {
 103                                 if (buffer_async_write(bh))
 104                                         under_io++;
 105                                 continue;
 106                         }
 107                         clear_buffer_async_write(bh);
 108                         if (bio->bi_error)
 109                                 buffer_io_error(bh);
 110                 } while ((bh = bh->b_this_page) != head);
 111                 bit_spin_unlock(BH_Uptodate_Lock, &head->b_state);
 112                 local_irq_restore(flags);
 113                 if (!under_io) {
 114 #ifdef CONFIG_EXT4_FS_ENCRYPTION
 115                         if (data_page)
 116                                 fscrypt_restore_control_page(data_page);
 117 #endif
 118                         end_page_writeback(page);
 119                 }
 120         }
 121 }
 122
 123 static void ext4_release_io_end(ext4_io_end_t *io_end)
 124 {
 125         struct bio *bio, *next_bio;
 126
 127         BUG_ON(!list_empty(&io_end->list));
 128         BUG_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
 129         WARN_ON(io_end->handle);
 130
 131         for (bio = io_end->bio; bio; bio = next_bio) {
 132                 next_bio = bio->bi_private;
 133                 ext4_finish_bio(bio);
 134                 bio_put(bio);
 135         }
 136         kmem_cache_free(io_end_cachep, io_end);
 137 }
 138
 139 /*
 140  * Check a range of space and convert unwritten extents to written. Note that
 141  * we are protected from truncate touching same part of extent tree by the
 142  * fact that truncate code waits for all DIO to finish (thus exclusion from
 143  * direct IO is achieved) and also waits for PageWriteback bits. Thus we
 144  * cannot get to ext4_ext_truncate() before all IOs overlapping that range are
 145  * completed (happens from ext4_free_ioend()).
 146  */
 147 static int ext4_end_io(ext4_io_end_t *io)
 148 {
 149         struct inode *inode = io->inode;
 150         loff_t offset = io->offset;
 151         ssize_t size = io->size;
 152         handle_t *handle = io->handle;
 153         int ret = 0;
 154
 155         ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
 156                    "list->prev 0x%p\n",
 157                    io, inode->i_ino, io->list.next, io->list.prev);
 158
 159         io->handle = NULL;      /* Following call will use up the handle */
 160         ret = ext4_convert_unwritten_extents(handle, inode, offset, size);
 161         if (ret < 0) {
 162                 ext4_msg(inode->i_sb, KERN_EMERG,
 163                          "failed to convert unwritten extents to written "
 164                          "extents -- potential data loss!  "
 165                          "(inode %lu, offset %llu, size %zd, error %d)",
 166                          inode->i_ino, offset, size, ret);
 167         }
 168         ext4_clear_io_unwritten_flag(io);
 169         ext4_release_io_end(io);
 170         return ret;
 171 }
 172
 173 static void dump_completed_IO(struct inode *inode, struct list_head *head)
 174 {
 175 #ifdef  EXT4FS_DEBUG
 176         struct list_head *cur, *before, *after;
 177         ext4_io_end_t *io, *io0, *io1;
 178
 179         if (list_empty(head))
 180                 return;
 181
 182         ext4_debug("Dump inode %lu completed io list\n", inode->i_ino);
 183         list_for_each_entry(io, head, list) {
 184                 cur = &io->list;
 185                 before = cur->prev;
 186                 io0 = container_of(before, ext4_io_end_t, list);
 187                 after = cur->next;
 188                 io1 = container_of(after, ext4_io_end_t, list);
 189
 190                 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
 191                             io, inode->i_ino, io0, io1);
 192         }
 193 #endif
 194 }
 195
 196 /* Add the io_end to per-inode completed end_io list. */
 197 static void ext4_add_complete_io(ext4_io_end_t *io_end)
 198 {
 199         struct ext4_inode_info *ei = EXT4_I(io_end->inode);
 200         struct ext4_sb_info *sbi = EXT4_SB(io_end->inode->i_sb);
 201         struct workqueue_struct *wq;
 202         unsigned long flags;
 203
 204         /* Only reserved conversions from writeback should enter here */
 205         WARN_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN));
 206         WARN_ON(!io_end->handle && sbi->s_journal);
 207         spin_lock_irqsave(&ei->i_completed_io_lock, flags);
 208         wq = sbi->rsv_conversion_wq;
 209         if (list_empty(&ei->i_rsv_conversion_list))
 210                 queue_work(wq, &ei->i_rsv_conversion_work);
 211         list_add_tail(&io_end->list, &ei->i_rsv_conversion_list);
 212         spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
 213 }
 214
 215 static int ext4_do_flush_completed_IO(struct inode *inode,
 216                                       struct list_head *head)
 217 {
 218         ext4_io_end_t *io;
 219         struct list_head unwritten;
 220         unsigned long flags;
 221         struct ext4_inode_info *ei = EXT4_I(inode);
 222         int err, ret = 0;
 223
 224         spin_lock_irqsave(&ei->i_completed_io_lock, flags);
 225         dump_completed_IO(inode, head);
 226         list_replace_init(head, &unwritten);
 227         spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
 228
 229         while (!list_empty(&unwritten)) {
 230                 io = list_entry(unwritten.next, ext4_io_end_t, list);
 231                 BUG_ON(!(io->flag & EXT4_IO_END_UNWRITTEN));
 232                 list_del_init(&io->list);
 233
 234                 err = ext4_end_io(io);
 235                 if (unlikely(!ret && err))
 236                         ret = err;
 237         }
 238         return ret;
 239 }
 240
 241 /*
 242  * work on completed IO, to convert unwritten extents to extents
 243  */
 244 void ext4_end_io_rsv_work(struct work_struct *work)
 245 {
 246         struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info,
 247                                                   i_rsv_conversion_work);
 248         ext4_do_flush_completed_IO(&ei->vfs_inode, &ei->i_rsv_conversion_list);
 249 }
 250
 251 ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
 252 {
 253         ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags);
 254         if (io) {
 255                 io->inode = inode;
 256                 INIT_LIST_HEAD(&io->list);
 257                 atomic_set(&io->count, 1);
 258         }
 259         return io;
 260 }
 261
 262 void ext4_put_io_end_defer(ext4_io_end_t *io_end)
 263 {
 264         if (atomic_dec_and_test(&io_end->count)) {
 265                 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN) || !io_end->size) {
 266                         ext4_release_io_end(io_end);
 267                         return;
 268                 }
 269                 ext4_add_complete_io(io_end);
 270         }
 271 }
 272
 273 int ext4_put_io_end(ext4_io_end_t *io_end)
 274 {
 275         int err = 0;
 276
 277         if (atomic_dec_and_test(&io_end->count)) {
 278                 if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
 279                         err = ext4_convert_unwritten_extents(io_end->handle,
 280                                                 io_end->inode, io_end->offset,
 281                                                 io_end->size);
 282                         io_end->handle = NULL;
 283                         ext4_clear_io_unwritten_flag(io_end);
 284                 }
 285                 ext4_release_io_end(io_end);
 286         }
 287         return err;
 288 }
 289
 290 ext4_io_end_t *ext4_get_io_end(ext4_io_end_t *io_end)
 291 {
 292         atomic_inc(&io_end->count);
 293         return io_end;
 294 }
 295
 296 /* BIO completion function for page writeback */
 297 static void ext4_end_bio(struct bio *bio)
 298 {
 299         ext4_io_end_t *io_end = bio->bi_private;
 300         sector_t bi_sector = bio->bi_iter.bi_sector;
 301
 302         BUG_ON(!io_end);
 303         bio->bi_end_io = NULL;
 304
 305         if (bio->bi_error) {
 306                 struct inode *inode = io_end->inode;
 307
 308                 ext4_warning(inode->i_sb, "I/O error %d writing to inode %lu "
 309                              "(offset %llu size %ld starting block %llu)",
 310                              bio->bi_error, inode->i_ino,
 311                              (unsigned long long) io_end->offset,
 312                              (long) io_end->size,
 313                              (unsigned long long)
 314                              bi_sector >> (inode->i_blkbits - 9));
 315                 mapping_set_error(inode->i_mapping, bio->bi_error);
 316         }
 317
 318         if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
 319                 /*
 320                  * Link bio into list hanging from io_end. We have to do it
 321                  * atomically as bio completions can be racing against each
 322                  * other.
 323                  */
 324                 bio->bi_private = xchg(&io_end->bio, bio);
 325                 ext4_put_io_end_defer(io_end);
 326         } else {
 327                 /*
 328                  * Drop io_end reference early. Inode can get freed once
 329                  * we finish the bio.
 330                  */
 331                 ext4_put_io_end_defer(io_end);
 332                 ext4_finish_bio(bio);
 333                 bio_put(bio);
 334         }
 335 }
 336
 337 void ext4_io_submit(struct ext4_io_submit *io)
 338 {
 339         struct bio *bio = io->io_bio;
 340
 341         if (bio) {
 342                 int io_op_flags = io->io_wbc->sync_mode == WB_SYNC_ALL ?
 343                                   WRITE_SYNC : 0;
 344                 bio_set_op_attrs(io->io_bio, REQ_OP_WRITE, io_op_flags);
 345                 submit_bio(io->io_bio);
 346         }
 347         io->io_bio = NULL;
 348 }
 349
 350 void ext4_io_submit_init(struct ext4_io_submit *io,
 351                          struct writeback_control *wbc)
 352 {
 353         io->io_wbc = wbc;
 354         io->io_bio = NULL;
 355         io->io_end = NULL;
 356 }
 357
 358 static int io_submit_init_bio(struct ext4_io_submit *io,
 359                               struct buffer_head *bh)
 360 {
 361         struct bio *bio;
 362
 363         bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES);
 364         if (!bio)
 365                 return -ENOMEM;
 366         wbc_init_bio(io->io_wbc, bio);
 367         bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
 368         bio->bi_bdev = bh->b_bdev;
 369         bio->bi_end_io = ext4_end_bio;
 370         bio->bi_private = ext4_get_io_end(io->io_end);
 371         io->io_bio = bio;
 372         io->io_next_block = bh->b_blocknr;
 373         return 0;
 374 }
 375
 376 static int io_submit_add_bh(struct ext4_io_submit *io,
 377                             struct inode *inode,
 378                             struct page *page,
 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, page, bh->b_size, bh_offset(bh));
 393         if (ret != bh->b_size)
 394                 goto submit_and_retry;
 395         wbc_account_io(io->io_wbc, page, bh->b_size);
 396         io->io_next_block++;
 397         return 0;
 398 }
 399
 400 int ext4_bio_write_page(struct ext4_io_submit *io,
 401                         struct page *page,
 402                         int len,
 403                         struct writeback_control *wbc,
 404                         bool keep_towrite)
 405 {
 406         struct page *data_page = NULL;
 407         struct inode *inode = page->mapping->host;
 408         unsigned block_start, blocksize;
 409         struct buffer_head *bh, *head;
 410         int ret = 0;
 411         int nr_submitted = 0;
 412         int nr_to_submit = 0;
 413
 414         blocksize = 1 << inode->i_blkbits;
 415
 416         BUG_ON(!PageLocked(page));
 417         BUG_ON(PageWriteback(page));
 418
 419         if (keep_towrite)
 420                 set_page_writeback_keepwrite(page);
 421         else
 422                 set_page_writeback(page);
 423         ClearPageError(page);
 424
 425         /*
 426          * Comments copied from block_write_full_page:
 427          *
 428          * The page straddles i_size.  It must be zeroed out on each and every
 429          * writepage invocation because it may be mmapped.  "A file is mapped
 430          * in multiples of the page size.  For a file that is not a multiple of
 431          * the page size, the remaining memory is zeroed when mapped, and
 432          * writes to that region are not written out to the file."
 433          */
 434         if (len < PAGE_SIZE)
 435                 zero_user_segment(page, len, PAGE_SIZE);
 436         /*
 437          * In the first loop we prepare and mark buffers to submit. We have to
 438          * mark all buffers in the page before submitting so that
 439          * end_page_writeback() cannot be called from ext4_bio_end_io() when IO
 440          * on the first buffer finishes and we are still working on submitting
 441          * the second buffer.
 442          */
 443         bh = head = page_buffers(page);
 444         do {
 445                 block_start = bh_offset(bh);
 446                 if (block_start >= len) {
 447                         clear_buffer_dirty(bh);
 448                         set_buffer_uptodate(bh);
 449                         continue;
 450                 }
 451                 if (!buffer_dirty(bh) || buffer_delay(bh) ||
 452                     !buffer_mapped(bh) || buffer_unwritten(bh)) {
 453                         /* A hole? We can safely clear the dirty bit */
 454                         if (!buffer_mapped(bh))
 455                                 clear_buffer_dirty(bh);
 456                         if (io->io_bio)
 457                                 ext4_io_submit(io);
 458                         continue;
 459                 }
 460                 if (buffer_new(bh)) {
 461                         clear_buffer_new(bh);
 462                         unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
 463                 }
 464                 set_buffer_async_write(bh);
 465                 nr_to_submit++;
 466         } while ((bh = bh->b_this_page) != head);
 467
 468         bh = head = page_buffers(page);
 469
 470         if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode) &&
 471             nr_to_submit) {
 472                 gfp_t gfp_flags = GFP_NOFS;
 473
 474         retry_encrypt:
 475                 data_page = fscrypt_encrypt_page(inode, page, gfp_flags);
 476                 if (IS_ERR(data_page)) {
 477                         ret = PTR_ERR(data_page);
 478                         if (ret == -ENOMEM && wbc->sync_mode == WB_SYNC_ALL) {
 479                                 if (io->io_bio) {
 480                                         ext4_io_submit(io);
 481                                         congestion_wait(BLK_RW_ASYNC, HZ/50);
 482                                 }
 483                                 gfp_flags |= __GFP_NOFAIL;
 484                                 goto retry_encrypt;
 485                         }
 486                         data_page = NULL;
 487                         goto out;
 488                 }
 489         }
 490
 491         /* Now submit buffers to write */
 492         do {
 493                 if (!buffer_async_write(bh))
 494                         continue;
 495                 ret = io_submit_add_bh(io, inode,
 496                                        data_page ? data_page : page, bh);
 497                 if (ret) {
 498                         /*
 499                          * We only get here on ENOMEM.  Not much else
 500                          * we can do but mark the page as dirty, and
 501                          * better luck next time.
 502                          */
 503                         break;
 504                 }
 505                 nr_submitted++;
 506                 clear_buffer_dirty(bh);
 507         } while ((bh = bh->b_this_page) != head);
 508
 509         /* Error stopped previous loop? Clean up buffers... */
 510         if (ret) {
 511         out:
 512                 if (data_page)
 513                         fscrypt_restore_control_page(data_page);
 514                 printk_ratelimited(KERN_ERR "%s: ret = %d\n", __func__, ret);
 515                 redirty_page_for_writepage(wbc, page);
 516                 do {
 517                         clear_buffer_async_write(bh);
 518                         bh = bh->b_this_page;
 519                 } while (bh != head);
 520         }
 521         unlock_page(page);
 522         /* Nothing submitted - we have to end page writeback */
 523         if (!nr_submitted)
 524                 end_page_writeback(page);
 525         return ret;
 526 }