linux内核遍历块设备,Linux通用块设备层 - Linux Kernel Exp .内核管理块设备要比管理字符设备复杂。因为字符设备仅仅需要控制一个位置-当前位...

我会带着你远行 2023-01-20 08:22 12阅读 0赞

Linux

http://www.ilinuxkernel.com

http://www.ilinuxkernel.com/

Linux

1 ........................................................................................................................... 42 ...................................................................................................... 52.1 Sectors .................................................................................................. 52.2 Blocks ....................................................................................................... 62.3 Segments .................................................................................................. 6

3 ............................................................................................................. 83.1 ................................................................................................ 83.2 bio ........................................................................................................... 123.3 bio .................................................................................. 15

4 ..................................................................................... 164.1 ....................................................................................... 164.2 ................................................................................... 19

4.2.1 ...................................................................................................... 194.2.2 ll\_rw\_lock .............................................................................................. 204.2.3 submit\_bh ............................................................................................. 224.2.4 generic\_make\_request ......................................................................... 244.2.5 \_\_generic\_make\_request ..................................................................... 26

Linux

1 LinuxI/O ........................................................................................................ 4

2 ................................................................................................................ 7

3 .................................................................................................... 10

4 biobio\_vecpage .................................................................... 14

Linux

read

1 LinuxI/O

1readVFSVirtual Filesystem Switch

2VFS

Mapping Layer

4Generic Block LayerI/O

Virtual Filesystem (VFS) Layer

Generic Block Layer

Buffer Cache (Page Cache)

I/O Scheduler Layer

Block Device Driver Block Device Driver

Hard Disk SSD

Direct IO

Request Queue

Kernel Space

Storage Media

Mapping Layer Disk Filesystem

Disk Filesystem

Linux

5I/OI/O Scheduler Layer

I/O

6Block Device Driver

12LinuxVFSVirtual Filesystem

Switchsys\_readsys\_write

VFS3Mapping Layer

chunk

CD-ROMSSD

2.1 Sectors

2512

512CD-ROM2k

512

Linux

sector indices3264

sector\_t

2.2 Blocks

VFS

block

inode

5121K4K

block

2.3 Segments

I/O

DMA

DMA/

DMA

/scatter-gatherDMA

/DMA

Linux

/DMAsegments

/DMA

generice block layer

physical segment

Page

Block B

ufferSector

Sector

Block B

uffer

Sector

Block B

uffer

Sector

Block B

uffer

Sector

Segment

512B

1KB

3KB

4KB

sectorI/O

mapping layer

block

Linux

/DMA

pagepage frame

4KB4KRAM4K

Generic Block Layer

CPU

zero-copy

I/O

LVMLogical Volume ManagerRAIDRedundant Array of

Inexpensive Disks

DMAI/O

3.1

Linux

block buffer

buffer headbuffer\_head

buffer\_head

include/linux/buffer\_head.h

00061: struct bbuuffffeerr\_\_hheeaadd \{ 00062: unsigned long b\_state; / \* buffer state bitmap (see above) \*/ 00063: struct buffer\_head \*b\_this\_page;/ \* circular list of page's buffers \*/ 00064: struct page \*b\_page; / \* the page this bh is mapped to \*/ 00065: 00066: sector\_t b\_blocknr; / \* start block number \*/ 00067: size\_t b\_size; / \* size of mapping \*/ 00068: char \*b\_data; / \* pointer to data within the page \*/ 00069: 00070: struct block\_device \*b\_bdev; 00071: bh\_end\_io\_t \*b\_end\_io; / \* I/ O completion \*/ 00072: void \*b\_private; / \* reserved for b\_end\_io \*/ 00073: struct list\_head b\_assoc\_buffers; / \* associated with another mapping \*/ 00074: struct address\_space \*b\_assoc\_map; / \* mapping this buffer is 00075: associated with \*/ 00076: atomic\_t b\_count; / \* users using this buffer\_head \*/ 00077: \};

b\_state

b\_this\_page

b\_page

b\_blocknr

b\_size

b\_data

b\_dev

b\_end\_ioI/O

b\_privateI/O

b\_assoc\_buffers

b\_assoc\_mapbufferaddress\_space

Linux

b\_count

b\_state1

bh\_state\_bitsinclude/linux/buffer\_head.h

00019: enum bbhh\_\_ssttaattee\_\_bbiittss \{ 00020: BH\_Uptodate, / \* Contains valid data \*/ 00021: BH\_Dirty, / \* Is dirty \*/ 00022: BH\_Lock,/ \* Is locked \*/ 00023: BH\_Req, / \* Has been submitted for I/ O \*/ 00024: BH\_Uptodate\_Lock,/ \* Used by the first bh in a page, to serialise 00025: \* IO completion of other buffers in the page 00026: \*/ 00027: 00028: BH\_Mapped, / \* Has a disk mapping \*/ 00029: BH\_New, / \* Disk mapping was newly created by get\_block \*/ 00030: BH\_Async\_Read, / \* Is under end\_buffer\_async\_read I/ O \*/ 00031: BH\_Async\_Write, / \* Is under end\_buffer\_async\_write I/ O \*/ 00032: BH\_Delay, / \* Buffer is not yet allocated on disk \*/ 00033: BH\_Boundary, / \* Block is followed by a discontiguity \*/ 00034: BH\_Write\_EIO, / \* I/ O error on write \*/ 00035: BH\_Ordered, / \* DEPRECATED: ordered write \*/ 00036: BH\_Eopnotsupp, / \* DEPRECATED: operation not supported (barrier) \*/ 00037: BH\_Unwritten, / \* Buffer is allocated on disk but not written \*/ 00038: BH\_Quiet, / \* Buffer Error Prinks to be quiet \*/ 00039: 00040: BH\_PrivateStart,/ \* not a state bit, but the first bit available 00041: \* for private allocation by other entities 00042: \*/ 00043: \};

1 bh\_state

BH\_Uptodate buffer BH\_Dirty buffer

BH\_Lock bufferI/O

Linux

BH\_Req bufferI/O BH\_Mapped buffer BH\_New bufferget\_block BH\_Async\_Read bufferend\_buffer\_async\_read BH\_Async\_Write bufferend\_buffer\_async\_write BH\_Delay buffer BH\_Boundary BH\_Write\_EIO buffer BH\_Ordered ordered BH\_Eopnosupp BH\_Unwritten buffer BH\_Quiet buffer

bh\_state\_bitsBH\_PrivateStart

I/OBH\_PrivateStart

b\_state

I/O

b\_countget\_bhput\_bh

include/linux/buffer\_head.h

00279: static inline void ggeett\_\_bbhh(struct buffer\_head \*bh00280: \{

)

00281: atomic\_inc(&bh->b\_count); 00282: \} 00283:

00284: static inline void ppuutt\_\_bbhh(struct buffer\_head \*bh00285: \{

)

00286: smp\_mb before\_atomic\_dec(); 00287: atomic\_dec(&bh->b\_count); 00288: \} 00289:

get\_bh

put\_bh

b\_blocknrb\_bdev

b\_pageb\_data

Linux

b\_pageb\_size

b\_datab\_data+b\_size

3.2 bio

I/Obioinclude/linux/bio.h

segmentI/O

bioI/OI/O

\-I/O

bio 00060: / \* 00061: \* main unit of I/ O for the block layer and lower layers (ie drivers and 00062: \* stacking drivers) 00063: \*/ 00064: struct bbiioo \{ 00065: sector\_t bi\_sector; / \* device address in 512 byte 00066: 00067: struct bio \*bi\_next; / \* request queue link \*/ 00068: struct block\_device \*bi\_bdev; 00069: unsigned long bi\_flags; / \* status, command, etc \*/ 00070: unsigned long bi\_rw; / \* bottom bits READ/ WRITE, 00071: \* top bits priority 00072: \*/ 00073: 00074: unsigned short bi\_vcnt; / \* how many bio\_vec's \*/ 00075: unsigned short bi\_idx; / \* current index into bvl\_vec \*/ 00076: 00077: / \* Number of segments in this BIO after 00078: \* physical address coalescing is performed. 00079: \*/ 00080: unsigned int bi\_phys\_segments; 00081: 00082: unsigned int bi\_size; / \* residual I/ O count \*/ 00083: 00084: / \* 00085: \* To keep track of the max segment size, we account for the 00086: \* sizes of the first and last mergeable segments in this bio. 00087: \*/ 00088: unsigned int bi\_seg\_front\_size; 00089: unsigned int bi\_seg\_back\_size; 00090: 00091: unsigned int bi\_max\_vecs; / \* max bvl\_vecs we can hold \*/ 00092: 00093: unsigned int bi\_comp\_cpu; / \* completion CPU \*/ 00094: 00095: atomic\_t bi\_cnt; / \* pin count \*/ 00096: 00097: struct bio\_vec \*bi\_io\_vec; / \* the actual vec list \*/ 00098:

Linux

00099: bio\_end\_io\_t \*bi\_end\_io; 00100: 00101: void \*bi\_private; 00102: \#if defined(CONFIG\_BLK\_DEV\_INTEGRITY) 00103: struct bio\_integrity\_payload \*bi\_integrity; / \* data integrity \*/ 00104: \#endif 00105: 00106: bio\_destructor\_t \*bi\_destructor; / \* destructor \*/ 00107: 00108: / \* 00109: \* We can inline a number of vecs at the end of the bio, to avoid 00110: \* double allocations for a small number of bio\_vecs. This member 00111: \* MUST obviously be kept at the very end of