The bio structure

The basic container for block I/O within the kernel is the bio structure, which is defined in <linux/bio.h>. This structure represents block I/O operations that are in flight (active) as a list of segments. A segment is a chunk of a buffer that is contiguous in memory. Thus, individual buffers need not be contiguous in memory. By allowing the buffers to be described in chunks, the bio structure provides the capability for the kernel to perform block I/O operations of even a single buffer from multiple locations in memory. Vector I/O such as this is called scatter-gather I/O.

Here is struct bio, defined in <linux/bio.h>, with comments added for each field:

struct bio {
        sector_t             bi_sector;         /* associated sector on disk */
        struct bio           *bi_next;          /* list of requests */
        struct block_device  *bi_bdev;          /* associated block device */
        unsigned long        bi_flags;          /* status and command flags */
        unsigned long        bi_rw;             /* read or write? */
        unsigned short       bi_vcnt;           /* number of bio_vecs off */
        unsigned short       bi_idx;            /* current index in bi_io_vec */
        unsigned short       bi_phys_segments;  /* number of segments after coalescing */
        unsigned short       bi_hw_segments;    /* number of segments after remapping */
        unsigned int         bi_size;           /* I/O count */
        unsigned int         bi_hw_front_size;  /* size of the first mergeable segment */
        unsigned int         bi_hw_back_size;   /* size of the last mergeable segment */
        unsigned int         bi_max_vecs;       /* maximum bio_vecs possible */
        struct bio_vec       *bi_io_vec;        /* bio_vec list */
        bio_end_io_t         *bi_end_io;        /* I/O completion method */
        atomic_t             bi_cnt;            /* usage counter */
        void                 *bi_private;       /* owner-private method */
        bio_destructor_t     *bi_destructor;    /* destructor method */
};

The primary purpose of a bio structure is to represent an in-flight block I/O operation. To this end, the majority of the fields in the structure are housekeeping related. The most important fields are bi_io_vecs, bi_vcnt, and bi_idx. Figure 13.2 shows the relationship between the bio structure and its friends.

The bi_io_vecs field points to an array of bio_vec structures. These structures are used as lists of individual segments in this specific block I/O operation. Each bio_vec is treated as a vector of the form <page, offset, len>, which describes a specific segment: the physical page on which it lies, the location of the block as an offset into the page, and the length of the block starting from the given offset. The full array of these vectors describes the entire buffer. The bio_vec structure is defined in <linux/bio.h>:

struct bio_vec {
        /* pointer to the physical page on which this buffer resides */
        struct page     *bv_page;

        /* the length in bytes of this buffer */
        unsigned int    bv_len;

        /* the byte offset within the page where the buffer resides */
        unsigned int    bv_offset;
};

In each given block I/O operation, there are bi_vcnt vectors in the bio_vec array starting with bi_io_vecs. As the block I/O operation is carried out, the bi_idx field is used to point to the current index into the array.

In summary, each block I/O request is represented by a bio structure. Each request comprises one or more blocks, which are stored in an array of bio_vec structures. These structures act as vectors and describe each segment's location in a physical page in memory. The first segment in the I/O operation is pointed to by b_io_vec. Each additional segment follows after the first, for a total of bi_vcnt segments in the list. As the block I/O layer submits segments in the request, the bi_idx field is updated to point to the current segment.

The bi_idx field is used to point to the current bio_vec in the list, which helps the block I/O layer keep track of partially completed block I/O operations. A more important usage, however, is to allow the splitting of bio structures. With this feature, drivers such as RAID (Redundant Array of Inexpensive Disks, a hard disk setup that allows single volumes to span multiple disks for performance and reliability purposes) can take a single bio structure, initially intended for a single device, and split it up among the multiple hard drives in the RAID array. All the RAID driver needs to do is copy the bio structure and update the bi_idx field to point to where the individual drive should start its operation.

The bio structure maintains a usage count in the bi_cnt field. When this field reaches zero, the structure is destroyed and the backing memory is freed. Two functions, which follow, manage the usage counters for you.

void bio_get(struct bio *bio)
void bio_put(struct bio *bio)

The former increments the usage count, whereas the latter decrements the usage count (and, if the count reaches zero, destroys the bio structure). Before manipulating an in-flight bio structure, be sure to increment its usage count to make sure it does not complete and deallocate out from under you. When you're finished, decrement the usage count in turn.

Finally, the bi_private field is a private field for the owner (that is, creator) of the structure. As a rule, you can read or write this field only if you allocated the bio structure.

The Old Versus the New

The difference between buffer heads and the new bio structure is important. The bio structure represents an I/O operation, which may include one or more pages in memory. On the other hand, the buffer_head structure represents a single buffer, which describes a single block on the disk. Because buffer heads are tied to a single disk block in a single page, buffer heads result in the unnecessary dividing of requests into block-sized chunks, only to later reassemble them. Because the bio structure is lightweight, it can describe discontiguous blocks, and does not unnecessarily split I/O operations.

Switching from struct buffer_head to struct bio provided other benefits, as well:

• The bio structure can easily represent high memory (see Chapter 11), because struct bio deals with only physical pages and not direct pointers.

• The bio structure can represent both normal page I/O and direct I/O (I/O operations that do not go through the page cachesee Chapter 15 for a discussion on the page cache).

• The bio structure makes it easy to perform scatter-gather (vectored) block I/O operations, with the data involved in the operation originating from multiple physical pages.

• The bio structure is much more lightweight than a buffer head because it contains only the minimum information needed to represent a block I/O operation and not unnecessary information related to the buffer itself.

The concept of buffer heads is still required, however; buffer heads function as descriptors, mapping disk blocks to pages. The bio structure does not contain any information about the state of a bufferit is simply an array of vectors describing one or more segments of data for a single block I/O operation, plus related information. In the current setup, the buffer_head structure is still needed to contain information about buffers while the bio structure describes in-flight I/O. Keeping the two structures separate allows each to remain as small as possible.