LKBL

Journaling

System crash can leave FS in inconsistent state. This can be repaired using FS checkers, however they are slow. Modern FS prefer journaling also known as “Write-ahead” logging. It is a technique when before actual writing of data, operation is logged. This helps in case of crash where only the journal log is replayed instead of whole disk scan. Journal can be a file within FS or different device. Journal normally contains a “Transaction” start, followed by what needs to be updated, finally it has an end marker called “transaction” end. These set of information have enough information to replay in case of crash. A Journaled FS, first writes the journaled log (also only called log) and once this is successful it “checkpoints” or updates the metadata with the latest information.

Every write can be divide into two operations. 1) First, write to metadata, 2) Second, The Data itself. (block write)

There are two types of journal setup:-

1. Data journaling - Where both metadata and data are journaled. Downside, is that data is written twice.
2. Metadata journaling (ordered) - Where only metadata is journaled. In this case data is first wirtten followed by Metadata.

Cases where FS can in inconsistent state

Metadata Failed and Data failed

• FS is consistent. All failure.

Metadata failed but Data passed.

• Here there will no updated inode pointing to data.
• from FS point of view it is still consistent

Metadata passed but Data failed

• Inode points to latest updated data. Which is not there but garbage
• FS is in in-consistent state.

Write barrieres gurantees that the write before the issue of barrier would be executed first and would not be reorderd due to buffereing or cacheing. Journaling

Linux Kernel Block Layer

1. blk_queue_flag_set
2. blk_queue_flag_clear
3. blk_queue_flag_test_and_set
4. blk_rq_init
5. errno_to_blk_status
6. blk_status_to_errno
7. blk_dump_rq_flags
8. blk_sync_queue
9. blk_set_pm_only
10. blk_clear_pm_only
11. blk_put_queue
12. blk_set_queue_dying
13. blk_cleanup_queue
14. blk_alloc_queue
15. blk_alloc_queue_node
16. blk_get_queue
17. blk_get_request
18. blk_put_request
19. blk_init_request_from_bio
20. generic_make_request
21. direct_make_request
22. submit_bio
23. blk_insert_cloned_request
24. blk_rq_err_bytes
25. blk_steal_bios
26. blk_update_request
27. rq_flush_dcache_pages
28. blk_lld_busy
29. blk_rq_unprep_clone
30. blk_rq_prep_clone
31. kblockd_schedule_work
32. kblockd_schedule_work_on
33. kblockd_mod_delayed_work_on
34. blk_start_plug
35. blk_check_plugged
36. blk_finish_plug
37. blkdev_issue_discard
38. blkdev_issue_write_same
39. blkdev_issue_zeroout
40. fs_bio_set
41. bio_uninit
42. bio_init
43. bio_reset
44. bio_chain
45. bio_alloc_bioset
46. zero_fill_bio_iter
47. bio_put
48. bio_clone_fast
49. bio_add_pc_page
50. bio_add_page
51. submit_bio_wait
52. bio_advance
53. bio_copy_data_iter
54. bio_copy_data
55. bio_list_copy_data
56. bio_free_pages
57. bio_map_kern
58. generic_start_io_acct
59. generic_end_io_acct
60. bio_flush_dcache_pages
61. bio_endio
62. bio_split
63. bio_trim
64. bioset_exit
65. bioset_init
66. bioset_init_from_src
67. bio_disassociate_blkg
68. bio_associate_blkg_from_css
69. bio_associate_blkg
70. bio_clone_blkg_association