)]}'
{
  "commit": "cec47e3d4a861e1d942b3a580d0bbef2700d2bb2",
  "tree": "2f92b957d515a5d887fe0147984cda3203c8b8ea",
  "parents": [
    "f5db4af466e2dca0fe822019812d586ca910b00c"
  ],
  "author": {
    "name": "Kiyoshi Ueda",
    "email": "k-ueda@ct.jp.nec.com",
    "time": "Mon Jun 22 10:12:35 2009 +0100"
  },
  "committer": {
    "name": "Alasdair G Kergon",
    "email": "agk@redhat.com",
    "time": "Mon Jun 22 10:12:35 2009 +0100"
  },
  "message": "dm: prepare for request based option\n\nThis patch adds core functions for request-based dm.\n\nWhen struct mapped device (md) is initialized, md-\u003equeue has\nan I/O scheduler and the following functions are used for\nrequest-based dm as the queue functions:\n    make_request_fn: dm_make_request()\n    pref_fn:         dm_prep_fn()\n    request_fn:      dm_request_fn()\n    softirq_done_fn: dm_softirq_done()\n    lld_busy_fn:     dm_lld_busy()\nActual initializations are done in another patch (PATCH 2).\n\nBelow is a brief summary of how request-based dm behaves, including:\n  - making request from bio\n  - cloning, mapping and dispatching request\n  - completing request and bio\n  - suspending md\n  - resuming md\n\n  bio to request\n  \u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\n  md-\u003equeue-\u003emake_request_fn() (dm_make_request()) calls __make_request()\n  for a bio submitted to the md.\n  Then, the bio is kept in the queue as a new request or merged into\n  another request in the queue if possible.\n\n  Cloning and Mapping\n  \u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\n  Cloning and mapping are done in md-\u003equeue-\u003erequest_fn() (dm_request_fn()),\n  when requests are dispatched after they are sorted by the I/O scheduler.\n\n  dm_request_fn() checks busy state of underlying devices using\n  target\u0027s busy() function and stops dispatching requests to keep them\n  on the dm device\u0027s queue if busy.\n  It helps better I/O merging, since no merge is done for a request\n  once it is dispatched to underlying devices.\n\n  Actual cloning and mapping are done in dm_prep_fn() and map_request()\n  called from dm_request_fn().\n  dm_prep_fn() clones not only request but also bios of the request\n  so that dm can hold bio completion in error cases and prevent\n  the bio submitter from noticing the error.\n  (See the \"Completion\" section below for details.)\n\n  After the cloning, the clone is mapped by target\u0027s map_rq() function\n    and inserted to underlying device\u0027s queue using\n    blk_insert_cloned_request().\n\n  Completion\n  \u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\n  Request completion can be hooked by rq-\u003eend_io(), but then, all bios\n  in the request will have been completed even error cases, and the bio\n  submitter will have noticed the error.\n  To prevent the bio completion in error cases, request-based dm clones\n  both bio and request and hooks both bio-\u003ebi_end_io() and rq-\u003eend_io():\n      bio-\u003ebi_end_io(): end_clone_bio()\n      rq-\u003eend_io():     end_clone_request()\n\n  Summary of the request completion flow is below:\n  blk_end_request() for a clone request\n    \u003d\u003e blk_update_request()\n       \u003d\u003e bio-\u003ebi_end_io() \u003d\u003d end_clone_bio() for each clone bio\n          \u003d\u003e Free the clone bio\n          \u003d\u003e Success: Complete the original bio (blk_update_request())\n             Error:   Don\u0027t complete the original bio\n    \u003d\u003e blk_finish_request()\n       \u003d\u003e rq-\u003eend_io() \u003d\u003d end_clone_request()\n          \u003d\u003e blk_complete_request()\n             \u003d\u003e dm_softirq_done()\n                \u003d\u003e Free the clone request\n                \u003d\u003e Success: Complete the original request (blk_end_request())\n                   Error:   Requeue the original request\n\n  end_clone_bio() completes the original request on the size of\n  the original bio in successful cases.\n  Even if all bios in the original request are completed by that\n  completion, the original request must not be completed yet to keep\n  the ordering of request completion for the stacking.\n  So end_clone_bio() uses blk_update_request() instead of\n  blk_end_request().\n  In error cases, end_clone_bio() doesn\u0027t complete the original bio.\n  It just frees the cloned bio and gives over the error handling to\n  end_clone_request().\n\n  end_clone_request(), which is called with queue lock held, completes\n  the clone request and the original request in a softirq context\n  (dm_softirq_done()), which has no queue lock, to avoid a deadlock\n  issue on submission of another request during the completion:\n      - The submitted request may be mapped to the same device\n      - Request submission requires queue lock, but the queue lock\n        has been held by itself and it doesn\u0027t know that\n\n  The clone request has no clone bio when dm_softirq_done() is called.\n  So target drivers can\u0027t resubmit it again even error cases.\n  Instead, they can ask dm core for requeueing and remapping\n  the original request in that cases.\n\n  suspend\n  \u003d\u003d\u003d\u003d\u003d\u003d\u003d\n  Request-based dm uses stopping md-\u003equeue as suspend of the md.\n  For noflush suspend, just stops md-\u003equeue.\n\n  For flush suspend, inserts a marker request to the tail of md-\u003equeue.\n  And dispatches all requests in md-\u003equeue until the marker comes to\n  the front of md-\u003equeue.  Then, stops dispatching request and waits\n  for the all dispatched requests to complete.\n  After that, completes the marker request, stops md-\u003equeue and\n  wake up the waiter on the suspend queue, md-\u003ewait.\n\n  resume\n  \u003d\u003d\u003d\u003d\u003d\u003d\n  Starts md-\u003equeue.\n\nSigned-off-by: Kiyoshi Ueda \u003ck-ueda@ct.jp.nec.com\u003e\nSigned-off-by: Jun\u0027ichi Nomura \u003cj-nomura@ce.jp.nec.com\u003e\nSigned-off-by: Alasdair G Kergon \u003cagk@redhat.com\u003e\n",
  "tree_diff": [
    {
      "type": "modify",
      "old_id": "09a57113955e5fb0b04ece6e9d5a30d111b838d6",
      "old_mode": 33188,
      "old_path": "drivers/md/dm-table.c",
      "new_id": "c5f784419f238ab863acdeb563294e6271f64138",
      "new_mode": 33188,
      "new_path": "drivers/md/dm-table.c"
    },
    {
      "type": "modify",
      "old_id": "f609793a92d05c11d6ad0a0de00475f2ba369289",
      "old_mode": 33188,
      "old_path": "drivers/md/dm.c",
      "new_id": "be003e5fea3d8d1ba6c4cea4da758c885e623a06",
      "new_mode": 33188,
      "new_path": "drivers/md/dm.c"
    },
    {
      "type": "modify",
      "old_id": "604e85caadf64e9a9301a34282a2befd3fa980b6",
      "old_mode": 33188,
      "old_path": "drivers/md/dm.h",
      "new_id": "8dcabb1caff139289c12308c6a6af6f08b9f077e",
      "new_mode": 33188,
      "new_path": "drivers/md/dm.h"
    },
    {
      "type": "modify",
      "old_id": "e6bf3b8c7bf2811dccd4c64efe5adee187df5ad8",
      "old_mode": 33188,
      "old_path": "include/linux/device-mapper.h",
      "new_id": "0d6310657f32a41cc6f5af7150f13db9c35a51e7",
      "new_mode": 33188,
      "new_path": "include/linux/device-mapper.h"
    }
  ]
}
