minix/servers/mfs/cache.c

681 lines
20 KiB
C
Raw Normal View History

2005-04-21 16:53:53 +02:00
/* The file system maintains a buffer cache to reduce the number of disk
* accesses needed. Whenever a read or write to the disk is done, a check is
* first made to see if the block is in the cache. This file manages the
* cache.
*
* The entry points into this file are:
* get_block: request to fetch a block for reading or writing from cache
* put_block: return a block previously requested with get_block
* alloc_zone: allocate a new zone (to increase the length of a file)
* free_zone: release a zone (when a file is removed)
* invalidate: remove all the cache blocks on some device
2005-10-12 17:06:47 +02:00
*
* Private functions:
* read_block: read or write a block from the disk itself
2005-04-21 16:53:53 +02:00
*/
#include "fs.h"
#include <minix/u64.h>
#include <minix/bdev.h>
#include <sys/param.h>
#include <stdlib.h>
#include <assert.h>
#include <minix/libminixfs.h>
#include <math.h>
2005-04-21 16:53:53 +02:00
#include "buf.h"
#include "super.h"
#include "inode.h"
2005-04-21 16:53:53 +02:00
2012-03-25 20:25:53 +02:00
static void rm_lru(struct buf *bp);
static void read_block(struct buf *);
2005-04-21 16:53:53 +02:00
2012-03-25 20:25:53 +02:00
static int vmcache = 0; /* are we using vm's secondary cache? (initially not) */
2012-03-25 20:25:53 +02:00
static block_t super_start = 0, super_end = 0;
2005-04-21 16:53:53 +02:00
/*===========================================================================*
* get_block *
*===========================================================================*/
2012-03-25 20:25:53 +02:00
struct buf *get_block(
register dev_t dev, /* on which device is the block? */
register block_t block, /* which block is wanted? */
int only_search /* if NO_READ, don't read, else act normal */
)
2005-04-21 16:53:53 +02:00
{
/* Check to see if the requested block is in the block cache. If so, return
* a pointer to it. If not, evict some other block and fetch it (unless
* 'only_search' is 1). All the blocks in the cache that are not in use
* are linked together in a chain, with 'front' pointing to the least recently
* used block and 'rear' to the most recently used block. If 'only_search' is
* 1, the block being requested will be overwritten in its entirety, so it is
* only necessary to see if it is in the cache; if it is not, any free buffer
* will do. It is not necessary to actually read the block in from disk.
* If 'only_search' is PREFETCH, the block need not be read from the disk,
* and the device is not to be marked on the block, so callers can tell if
* the block returned is valid.
* In addition to the LRU chain, there is also a hash chain to link together
* blocks whose block numbers end with the same bit strings, for fast lookup.
*/
int b;
static struct buf *bp, *prev_ptr;
u64_t yieldid = VM_BLOCKID_NONE, getid = make64(dev, block);
assert(buf_hash);
assert(buf);
assert(nr_bufs > 0);
ASSERT(fs_block_size > 0);
2005-04-21 16:53:53 +02:00
/* Search the hash chain for (dev, block). Do_read() can use
* get_block(NO_DEV ...) to get an unnamed block to fill with zeros when
* someone wants to read from a hole in a file, in which case this search
* is skipped
*/
if (dev != NO_DEV) {
b = BUFHASH(block);
2005-04-21 16:53:53 +02:00
bp = buf_hash[b];
while (bp != NULL) {
2005-04-21 16:53:53 +02:00
if (bp->b_blocknr == block && bp->b_dev == dev) {
/* Block needed has been found. */
if (bp->b_count == 0) rm_lru(bp);
bp->b_count++; /* record that block is in use */
2008-11-19 15:10:33 +01:00
ASSERT(bp->b_bytes == fs_block_size);
ASSERT(bp->b_dev == dev);
ASSERT(bp->b_dev != NO_DEV);
ASSERT(bp->bp);
2005-04-21 16:53:53 +02:00
return(bp);
} else {
/* This block is not the one sought. */
bp = bp->b_hash; /* move to next block on hash chain */
}
}
}
/* Desired block is not on available chain. Take oldest block ('front'). */
if ((bp = front) == NULL) panic("all buffers in use: %d", nr_bufs);
if(bp->b_bytes < fs_block_size) {
ASSERT(!bp->bp);
ASSERT(bp->b_bytes == 0);
if(!(bp->bp = alloc_contig( (size_t) fs_block_size, 0, NULL))) {
printf("MFS: couldn't allocate a new block.\n");
for(bp = front;
bp && bp->b_bytes < fs_block_size; bp = bp->b_next)
;
if(!bp) {
panic("no buffer available");
}
} else {
bp->b_bytes = fs_block_size;
}
}
ASSERT(bp);
ASSERT(bp->bp);
ASSERT(bp->b_bytes == fs_block_size);
ASSERT(bp->b_count == 0);
2005-04-21 16:53:53 +02:00
rm_lru(bp);
/* Remove the block that was just taken from its hash chain. */
b = BUFHASH(bp->b_blocknr);
2005-04-21 16:53:53 +02:00
prev_ptr = buf_hash[b];
if (prev_ptr == bp) {
buf_hash[b] = bp->b_hash;
} else {
/* The block just taken is not on the front of its hash chain. */
while (prev_ptr->b_hash != NULL)
2005-04-21 16:53:53 +02:00
if (prev_ptr->b_hash == bp) {
prev_ptr->b_hash = bp->b_hash; /* found it */
break;
} else {
prev_ptr = prev_ptr->b_hash; /* keep looking */
}
}
/* If the block taken is dirty, make it clean by writing it to the disk.
* Avoid hysteresis by flushing all other dirty blocks for the same device.
*/
if (bp->b_dev != NO_DEV) {
if (ISDIRTY(bp)) flushall(bp->b_dev);
/* Are we throwing out a block that contained something?
* Give it to VM for the second-layer cache.
*/
yieldid = make64(bp->b_dev, bp->b_blocknr);
assert(bp->b_bytes == fs_block_size);
bp->b_dev = NO_DEV;
2005-04-21 16:53:53 +02:00
}
/* Fill in block's parameters and add it to the hash chain where it goes. */
MARKCLEAN(bp); /* NO_DEV blocks may be marked dirty */
bp->b_dev = dev; /* fill in device number */
2005-04-21 16:53:53 +02:00
bp->b_blocknr = block; /* fill in block number */
bp->b_count++; /* record that block is being used */
b = BUFHASH(bp->b_blocknr);
2005-04-21 16:53:53 +02:00
bp->b_hash = buf_hash[b];
2005-04-21 16:53:53 +02:00
buf_hash[b] = bp; /* add to hash list */
if(dev == NO_DEV) {
if(vmcache && cmp64(yieldid, VM_BLOCKID_NONE) != 0) {
vm_yield_block_get_block(yieldid, VM_BLOCKID_NONE,
bp->bp, fs_block_size);
}
return(bp); /* If the caller wanted a NO_DEV block, work is done. */
}
2005-04-21 16:53:53 +02:00
/* Go get the requested block unless searching or prefetching. */
if(only_search == PREFETCH || only_search == NORMAL) {
/* Block is not found in our cache, but we do want it
* if it's in the vm cache.
*/
if(vmcache) {
/* If we can satisfy the PREFETCH or NORMAL request
* from the vm cache, work is done.
*/
if(vm_yield_block_get_block(yieldid, getid,
bp->bp, fs_block_size) == OK) {
return bp;
}
}
2005-04-21 16:53:53 +02:00
}
if(only_search == PREFETCH) {
/* PREFETCH: don't do i/o. */
bp->b_dev = NO_DEV;
} else if (only_search == NORMAL) {
read_block(bp);
} else if(only_search == NO_READ) {
/* we want this block, but its contents
* will be overwritten. VM has to forget
* about it.
*/
if(vmcache) {
vm_forgetblock(getid);
}
} else
panic("unexpected only_search value: %d", only_search);
assert(bp->bp);
2005-04-21 16:53:53 +02:00
return(bp); /* return the newly acquired block */
}
/*===========================================================================*
* put_block *
*===========================================================================*/
2012-03-25 20:25:53 +02:00
void put_block(bp, block_type)
2005-04-21 16:53:53 +02:00
register struct buf *bp; /* pointer to the buffer to be released */
int block_type; /* INODE_BLOCK, DIRECTORY_BLOCK, or whatever */
{
/* Return a block to the list of available blocks. Depending on 'block_type'
* it may be put on the front or rear of the LRU chain. Blocks that are
* expected to be needed again shortly (e.g., partially full data blocks)
* go on the rear; blocks that are unlikely to be needed again shortly
* (e.g., full data blocks) go on the front. Blocks whose loss can hurt
* the integrity of the file system (e.g., inode blocks) are written to
* disk immediately if they are dirty.
*/
if (bp == NULL) return; /* it is easier to check here than in caller */
2005-04-21 16:53:53 +02:00
bp->b_count--; /* there is one use fewer now */
if (bp->b_count != 0) return; /* block is still in use */
bufs_in_use--; /* one fewer block buffers in use */
/* Put this block back on the LRU chain. If the ONE_SHOT bit is set in
* 'block_type', the block is not likely to be needed again shortly, so put
* it on the front of the LRU chain where it will be the first one to be
* taken when a free buffer is needed later.
*/
2005-12-14 13:08:49 +01:00
if (bp->b_dev == DEV_RAM || (block_type & ONE_SHOT)) {
2005-04-21 16:53:53 +02:00
/* Block probably won't be needed quickly. Put it on front of chain.
* It will be the next block to be evicted from the cache.
*/
bp->b_prev = NULL;
2005-04-21 16:53:53 +02:00
bp->b_next = front;
if (front == NULL)
2005-04-21 16:53:53 +02:00
rear = bp; /* LRU chain was empty */
else
front->b_prev = bp;
front = bp;
}
else {
2005-04-21 16:53:53 +02:00
/* Block probably will be needed quickly. Put it on rear of chain.
* It will not be evicted from the cache for a long time.
*/
bp->b_prev = rear;
bp->b_next = NULL;
if (rear == NULL)
2005-04-21 16:53:53 +02:00
front = bp;
else
rear->b_next = bp;
rear = bp;
}
}
/*===========================================================================*
* alloc_zone *
*===========================================================================*/
2012-03-25 20:25:53 +02:00
zone_t alloc_zone(
dev_t dev, /* device where zone wanted */
zone_t z /* try to allocate new zone near this one */
)
2005-04-21 16:53:53 +02:00
{
/* Allocate a new zone on the indicated device and return its number. */
bit_t b, bit;
struct super_block *sp;
static int print_oos_msg = 1;
2005-04-21 16:53:53 +02:00
/* Note that the routine alloc_bit() returns 1 for the lowest possible
* zone, which corresponds to sp->s_firstdatazone. To convert a value
* between the bit number, 'b', used by alloc_bit() and the zone number, 'z',
* stored in the inode, use the formula:
* z = b + sp->s_firstdatazone - 1
* Alloc_bit() never returns 0, since this is used for NO_BIT (failure).
*/
sp = get_super(dev);
/* If z is 0, skip initial part of the map known to be fully in use. */
if (z == sp->s_firstdatazone) {
bit = sp->s_zsearch;
} else {
bit = (bit_t) (z - (sp->s_firstdatazone - 1));
2005-04-21 16:53:53 +02:00
}
b = alloc_bit(sp, ZMAP, bit);
if (b == NO_BIT) {
err_code = ENOSPC;
if (print_oos_msg)
printf("No space on device %d/%d\n", major(sp->s_dev),
minor(sp->s_dev));
print_oos_msg = 0; /* Don't repeat message */
2005-04-21 16:53:53 +02:00
return(NO_ZONE);
}
print_oos_msg = 1;
2005-04-21 16:53:53 +02:00
if (z == sp->s_firstdatazone) sp->s_zsearch = b; /* for next time */
return( (zone_t) (sp->s_firstdatazone - 1) + (zone_t) b);
2005-04-21 16:53:53 +02:00
}
/*===========================================================================*
* free_zone *
*===========================================================================*/
2012-03-25 20:25:53 +02:00
void free_zone(
dev_t dev, /* device where zone located */
zone_t numb /* zone to be returned */
)
2005-04-21 16:53:53 +02:00
{
/* Return a zone. */
register struct super_block *sp;
bit_t bit;
/* Locate the appropriate super_block and return bit. */
sp = get_super(dev);
if (numb < sp->s_firstdatazone || numb >= sp->s_zones) return;
bit = (bit_t) (numb - (zone_t) (sp->s_firstdatazone - 1));
2005-04-21 16:53:53 +02:00
free_bit(sp, ZMAP, bit);
if (bit < sp->s_zsearch) sp->s_zsearch = bit;
}
/*===========================================================================*
* read_block *
2005-04-21 16:53:53 +02:00
*===========================================================================*/
2012-03-25 20:25:53 +02:00
static void read_block(bp)
2005-04-21 16:53:53 +02:00
register struct buf *bp; /* buffer pointer */
{
/* Read or write a disk block. This is the only routine in which actual disk
* I/O is invoked. If an error occurs, a message is printed here, but the error
* is not reported to the caller. If the error occurred while purging a block
* from the cache, it is not clear what the caller could do about it anyway.
*/
int r, op_failed;
u64_t pos;
2005-04-21 16:53:53 +02:00
dev_t dev;
op_failed = 0;
2005-04-21 16:53:53 +02:00
if ( (dev = bp->b_dev) != NO_DEV) {
pos = mul64u(bp->b_blocknr, fs_block_size);
r = bdev_read(dev, pos, bp->b_data, fs_block_size,
BDEV_NOFLAGS);
if (r < 0) {
printf("MFS(%d) I/O error on device %d/%d, block %u\n",
SELF_E, major(dev), minor(dev), bp->b_blocknr);
op_failed = 1;
Split block/character protocols and libdriver This patch separates the character and block driver communication protocols. The old character protocol remains the same, but a new block protocol is introduced. The libdriver library is replaced by two new libraries: libchardriver and libblockdriver. Their exposed API, and drivers that use them, have been updated accordingly. Together, libbdev and libblockdriver now completely abstract away the message format used by the block protocol. As the memory driver is both a character and a block device driver, it now implements its own message loop. The most important semantic change made to the block protocol is that it is no longer possible to return both partial results and an error for a single transfer. This simplifies the interaction between the caller and the driver, as the I/O vector no longer needs to be copied back. Also, drivers are now no longer supposed to decide based on the layout of the I/O vector when a transfer should be cut short. Put simply, transfers are now supposed to either succeed completely, or result in an error. After this patch, the state of the various pieces is as follows: - block protocol: stable - libbdev API: stable for synchronous communication - libblockdriver API: needs slight revision (the drvlib/partition API in particular; the threading API will also change shortly) - character protocol: needs cleanup - libchardriver API: needs cleanup accordingly - driver restarts: largely unsupported until endpoint changes are reintroduced As a side effect, this patch eliminates several bugs, hacks, and gcc -Wall and -W warnings all over the place. It probably introduces a few new ones, too. Update warning: this patch changes the protocol between MFS and disk drivers, so in order to use old/new images, the MFS from the ramdisk must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
} else if (r != (ssize_t) fs_block_size) {
r = END_OF_FILE;
op_failed = 1;
}
if (op_failed) {
bp->b_dev = NO_DEV; /* invalidate block */
/* Report read errors to interested parties. */
rdwt_err = r;
}
2005-04-21 16:53:53 +02:00
}
}
/*===========================================================================*
* invalidate *
*===========================================================================*/
2012-03-25 20:25:53 +02:00
void invalidate(
dev_t device /* device whose blocks are to be purged */
)
2005-04-21 16:53:53 +02:00
{
/* Remove all the blocks belonging to some device from the cache. */
register struct buf *bp;
for (bp = &buf[0]; bp < &buf[nr_bufs]; bp++)
if (bp->b_dev == device) bp->b_dev = NO_DEV;
vm_forgetblocks();
2005-04-21 16:53:53 +02:00
}
/*===========================================================================*
* block_write_ok *
*===========================================================================*/
int block_write_ok(struct buf *bp)
{
if(superblock.s_dev != bp->b_dev) return 1;
if(bp->b_blocknr >= super_start && bp->b_blocknr <= super_end) {
printf("MFS: blocking write to superblock on mounted filesystem dev 0x%x.\n", bp->b_dev);
return 0;
}
if(superblock.s_rd_only) {
printf("MFS: blocking write to mounted readonly filesystem 0x%x.\n", bp->b_dev);
printf("This shouldn't happen.\n");
return 0;
}
return 1;
}
2005-09-11 18:45:46 +02:00
/*===========================================================================*
* flushall *
*===========================================================================*/
2012-03-25 20:25:53 +02:00
void flushall(
dev_t dev /* device to flush */
)
2005-04-21 16:53:53 +02:00
{
/* Flush all dirty blocks for one device. */
register struct buf *bp;
static struct buf **dirty; /* static so it isn't on stack */
static unsigned int dirtylistsize = 0;
2005-04-21 16:53:53 +02:00
int ndirty;
if(dirtylistsize != nr_bufs) {
if(dirtylistsize > 0) {
assert(dirty != NULL);
free(dirty);
}
if(!(dirty = malloc(sizeof(dirty[0])*nr_bufs)))
panic("couldn't allocate dirty buf list");
dirtylistsize = nr_bufs;
}
for (bp = &buf[0], ndirty = 0; bp < &buf[nr_bufs]; bp++) {
if (ISDIRTY(bp) && bp->b_dev == dev) {
if(!block_write_ok(bp)) {
printf("MFS: LATE: ignoring changes in block %d\n", bp->b_blocknr);
MARKCLEAN(bp);
continue;
}
dirty[ndirty++] = bp;
}
}
2005-04-21 16:53:53 +02:00
rw_scattered(dev, dirty, ndirty, WRITING);
}
/*===========================================================================*
* rw_scattered *
*===========================================================================*/
2012-03-25 20:25:53 +02:00
void rw_scattered(
dev_t dev, /* major-minor device number */
struct buf **bufq, /* pointer to array of buffers */
int bufqsize, /* number of buffers */
int rw_flag /* READING or WRITING */
)
2005-04-21 16:53:53 +02:00
{
/* Read or write scattered data from a device. */
register struct buf *bp;
int gap;
register int i;
register iovec_t *iop;
static iovec_t *iovec = NULL;
u64_t pos;
2005-04-21 16:53:53 +02:00
int j, r;
STATICINIT(iovec, NR_IOREQS);
2005-04-21 16:53:53 +02:00
/* (Shell) sort buffers on b_blocknr. */
gap = 1;
do
gap = 3 * gap + 1;
while (gap <= bufqsize);
while (gap != 1) {
gap /= 3;
for (j = gap; j < bufqsize; j++) {
for (i = j - gap;
i >= 0 && bufq[i]->b_blocknr > bufq[i + gap]->b_blocknr;
i -= gap) {
bp = bufq[i];
bufq[i] = bufq[i + gap];
bufq[i + gap] = bp;
}
}
}
/* Set up I/O vector and do I/O. The result of bdev I/O is OK if everything
2005-04-21 16:53:53 +02:00
* went fine, otherwise the error code for the first failed transfer.
*/
2005-04-21 16:53:53 +02:00
while (bufqsize > 0) {
for (j = 0, iop = iovec; j < NR_IOREQS && j < bufqsize; j++, iop++) {
bp = bufq[j];
if (bp->b_blocknr != (block_t) bufq[0]->b_blocknr + j) break;
2005-04-21 16:53:53 +02:00
iop->iov_addr = (vir_bytes) bp->b_data;
iop->iov_size = (vir_bytes) fs_block_size;
2005-04-21 16:53:53 +02:00
}
pos = mul64u(bufq[0]->b_blocknr, fs_block_size);
if (rw_flag == READING)
Split block/character protocols and libdriver This patch separates the character and block driver communication protocols. The old character protocol remains the same, but a new block protocol is introduced. The libdriver library is replaced by two new libraries: libchardriver and libblockdriver. Their exposed API, and drivers that use them, have been updated accordingly. Together, libbdev and libblockdriver now completely abstract away the message format used by the block protocol. As the memory driver is both a character and a block device driver, it now implements its own message loop. The most important semantic change made to the block protocol is that it is no longer possible to return both partial results and an error for a single transfer. This simplifies the interaction between the caller and the driver, as the I/O vector no longer needs to be copied back. Also, drivers are now no longer supposed to decide based on the layout of the I/O vector when a transfer should be cut short. Put simply, transfers are now supposed to either succeed completely, or result in an error. After this patch, the state of the various pieces is as follows: - block protocol: stable - libbdev API: stable for synchronous communication - libblockdriver API: needs slight revision (the drvlib/partition API in particular; the threading API will also change shortly) - character protocol: needs cleanup - libchardriver API: needs cleanup accordingly - driver restarts: largely unsupported until endpoint changes are reintroduced As a side effect, this patch eliminates several bugs, hacks, and gcc -Wall and -W warnings all over the place. It probably introduces a few new ones, too. Update warning: this patch changes the protocol between MFS and disk drivers, so in order to use old/new images, the MFS from the ramdisk must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
r = bdev_gather(dev, pos, iovec, j, BDEV_NOFLAGS);
else
Split block/character protocols and libdriver This patch separates the character and block driver communication protocols. The old character protocol remains the same, but a new block protocol is introduced. The libdriver library is replaced by two new libraries: libchardriver and libblockdriver. Their exposed API, and drivers that use them, have been updated accordingly. Together, libbdev and libblockdriver now completely abstract away the message format used by the block protocol. As the memory driver is both a character and a block device driver, it now implements its own message loop. The most important semantic change made to the block protocol is that it is no longer possible to return both partial results and an error for a single transfer. This simplifies the interaction between the caller and the driver, as the I/O vector no longer needs to be copied back. Also, drivers are now no longer supposed to decide based on the layout of the I/O vector when a transfer should be cut short. Put simply, transfers are now supposed to either succeed completely, or result in an error. After this patch, the state of the various pieces is as follows: - block protocol: stable - libbdev API: stable for synchronous communication - libblockdriver API: needs slight revision (the drvlib/partition API in particular; the threading API will also change shortly) - character protocol: needs cleanup - libchardriver API: needs cleanup accordingly - driver restarts: largely unsupported until endpoint changes are reintroduced As a side effect, this patch eliminates several bugs, hacks, and gcc -Wall and -W warnings all over the place. It probably introduces a few new ones, too. Update warning: this patch changes the protocol between MFS and disk drivers, so in order to use old/new images, the MFS from the ramdisk must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
r = bdev_scatter(dev, pos, iovec, j, BDEV_NOFLAGS);
2005-04-21 16:53:53 +02:00
Split block/character protocols and libdriver This patch separates the character and block driver communication protocols. The old character protocol remains the same, but a new block protocol is introduced. The libdriver library is replaced by two new libraries: libchardriver and libblockdriver. Their exposed API, and drivers that use them, have been updated accordingly. Together, libbdev and libblockdriver now completely abstract away the message format used by the block protocol. As the memory driver is both a character and a block device driver, it now implements its own message loop. The most important semantic change made to the block protocol is that it is no longer possible to return both partial results and an error for a single transfer. This simplifies the interaction between the caller and the driver, as the I/O vector no longer needs to be copied back. Also, drivers are now no longer supposed to decide based on the layout of the I/O vector when a transfer should be cut short. Put simply, transfers are now supposed to either succeed completely, or result in an error. After this patch, the state of the various pieces is as follows: - block protocol: stable - libbdev API: stable for synchronous communication - libblockdriver API: needs slight revision (the drvlib/partition API in particular; the threading API will also change shortly) - character protocol: needs cleanup - libchardriver API: needs cleanup accordingly - driver restarts: largely unsupported until endpoint changes are reintroduced As a side effect, this patch eliminates several bugs, hacks, and gcc -Wall and -W warnings all over the place. It probably introduces a few new ones, too. Update warning: this patch changes the protocol between MFS and disk drivers, so in order to use old/new images, the MFS from the ramdisk must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
/* Harvest the results. The driver may have returned an error, or it
* may have done less than what we asked for.
2005-04-21 16:53:53 +02:00
*/
Split block/character protocols and libdriver This patch separates the character and block driver communication protocols. The old character protocol remains the same, but a new block protocol is introduced. The libdriver library is replaced by two new libraries: libchardriver and libblockdriver. Their exposed API, and drivers that use them, have been updated accordingly. Together, libbdev and libblockdriver now completely abstract away the message format used by the block protocol. As the memory driver is both a character and a block device driver, it now implements its own message loop. The most important semantic change made to the block protocol is that it is no longer possible to return both partial results and an error for a single transfer. This simplifies the interaction between the caller and the driver, as the I/O vector no longer needs to be copied back. Also, drivers are now no longer supposed to decide based on the layout of the I/O vector when a transfer should be cut short. Put simply, transfers are now supposed to either succeed completely, or result in an error. After this patch, the state of the various pieces is as follows: - block protocol: stable - libbdev API: stable for synchronous communication - libblockdriver API: needs slight revision (the drvlib/partition API in particular; the threading API will also change shortly) - character protocol: needs cleanup - libchardriver API: needs cleanup accordingly - driver restarts: largely unsupported until endpoint changes are reintroduced As a side effect, this patch eliminates several bugs, hacks, and gcc -Wall and -W warnings all over the place. It probably introduces a few new ones, too. Update warning: this patch changes the protocol between MFS and disk drivers, so in order to use old/new images, the MFS from the ramdisk must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
if (r < 0) {
printf("MFS: I/O error %d on device %d/%d, block %u\n",
r, major(dev), minor(dev), bufq[0]->b_blocknr);
}
for (i = 0; i < j; i++) {
2005-04-21 16:53:53 +02:00
bp = bufq[i];
Split block/character protocols and libdriver This patch separates the character and block driver communication protocols. The old character protocol remains the same, but a new block protocol is introduced. The libdriver library is replaced by two new libraries: libchardriver and libblockdriver. Their exposed API, and drivers that use them, have been updated accordingly. Together, libbdev and libblockdriver now completely abstract away the message format used by the block protocol. As the memory driver is both a character and a block device driver, it now implements its own message loop. The most important semantic change made to the block protocol is that it is no longer possible to return both partial results and an error for a single transfer. This simplifies the interaction between the caller and the driver, as the I/O vector no longer needs to be copied back. Also, drivers are now no longer supposed to decide based on the layout of the I/O vector when a transfer should be cut short. Put simply, transfers are now supposed to either succeed completely, or result in an error. After this patch, the state of the various pieces is as follows: - block protocol: stable - libbdev API: stable for synchronous communication - libblockdriver API: needs slight revision (the drvlib/partition API in particular; the threading API will also change shortly) - character protocol: needs cleanup - libchardriver API: needs cleanup accordingly - driver restarts: largely unsupported until endpoint changes are reintroduced As a side effect, this patch eliminates several bugs, hacks, and gcc -Wall and -W warnings all over the place. It probably introduces a few new ones, too. Update warning: this patch changes the protocol between MFS and disk drivers, so in order to use old/new images, the MFS from the ramdisk must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
if (r < (ssize_t) fs_block_size) {
/* Transfer failed. */
if (i == 0) {
bp->b_dev = NO_DEV; /* Invalidate block */
Split block/character protocols and libdriver This patch separates the character and block driver communication protocols. The old character protocol remains the same, but a new block protocol is introduced. The libdriver library is replaced by two new libraries: libchardriver and libblockdriver. Their exposed API, and drivers that use them, have been updated accordingly. Together, libbdev and libblockdriver now completely abstract away the message format used by the block protocol. As the memory driver is both a character and a block device driver, it now implements its own message loop. The most important semantic change made to the block protocol is that it is no longer possible to return both partial results and an error for a single transfer. This simplifies the interaction between the caller and the driver, as the I/O vector no longer needs to be copied back. Also, drivers are now no longer supposed to decide based on the layout of the I/O vector when a transfer should be cut short. Put simply, transfers are now supposed to either succeed completely, or result in an error. After this patch, the state of the various pieces is as follows: - block protocol: stable - libbdev API: stable for synchronous communication - libblockdriver API: needs slight revision (the drvlib/partition API in particular; the threading API will also change shortly) - character protocol: needs cleanup - libchardriver API: needs cleanup accordingly - driver restarts: largely unsupported until endpoint changes are reintroduced As a side effect, this patch eliminates several bugs, hacks, and gcc -Wall and -W warnings all over the place. It probably introduces a few new ones, too. Update warning: this patch changes the protocol between MFS and disk drivers, so in order to use old/new images, the MFS from the ramdisk must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
vm_forgetblocks();
2005-04-21 16:53:53 +02:00
}
break;
}
if (rw_flag == READING) {
bp->b_dev = dev; /* validate block */
2005-04-21 16:53:53 +02:00
put_block(bp, PARTIAL_DATA_BLOCK);
} else {
MARKCLEAN(bp);
2005-04-21 16:53:53 +02:00
}
Split block/character protocols and libdriver This patch separates the character and block driver communication protocols. The old character protocol remains the same, but a new block protocol is introduced. The libdriver library is replaced by two new libraries: libchardriver and libblockdriver. Their exposed API, and drivers that use them, have been updated accordingly. Together, libbdev and libblockdriver now completely abstract away the message format used by the block protocol. As the memory driver is both a character and a block device driver, it now implements its own message loop. The most important semantic change made to the block protocol is that it is no longer possible to return both partial results and an error for a single transfer. This simplifies the interaction between the caller and the driver, as the I/O vector no longer needs to be copied back. Also, drivers are now no longer supposed to decide based on the layout of the I/O vector when a transfer should be cut short. Put simply, transfers are now supposed to either succeed completely, or result in an error. After this patch, the state of the various pieces is as follows: - block protocol: stable - libbdev API: stable for synchronous communication - libblockdriver API: needs slight revision (the drvlib/partition API in particular; the threading API will also change shortly) - character protocol: needs cleanup - libchardriver API: needs cleanup accordingly - driver restarts: largely unsupported until endpoint changes are reintroduced As a side effect, this patch eliminates several bugs, hacks, and gcc -Wall and -W warnings all over the place. It probably introduces a few new ones, too. Update warning: this patch changes the protocol between MFS and disk drivers, so in order to use old/new images, the MFS from the ramdisk must be used to mount all file systems.
2011-11-22 13:27:53 +01:00
r -= fs_block_size;
2005-04-21 16:53:53 +02:00
}
bufq += i;
bufqsize -= i;
if (rw_flag == READING) {
/* Don't bother reading more than the device is willing to
* give at this time. Don't forget to release those extras.
*/
while (bufqsize > 0) {
put_block(*bufq++, PARTIAL_DATA_BLOCK);
bufqsize--;
}
}
2005-09-11 18:45:46 +02:00
if (rw_flag == WRITING && i == 0) {
/* We're not making progress, this means we might keep
* looping. Buffers remain dirty if un-written. Buffers are
* lost if invalidate()d or LRU-removed while dirty. This
* is better than keeping unwritable blocks around forever..
*/
break;
}
2005-04-21 16:53:53 +02:00
}
}
/*===========================================================================*
* rm_lru *
*===========================================================================*/
2012-03-25 20:25:53 +02:00
static void rm_lru(bp)
2005-04-21 16:53:53 +02:00
struct buf *bp;
{
/* Remove a block from its LRU chain. */
struct buf *next_ptr, *prev_ptr;
bufs_in_use++;
next_ptr = bp->b_next; /* successor on LRU chain */
prev_ptr = bp->b_prev; /* predecessor on LRU chain */
if (prev_ptr != NULL)
2005-04-21 16:53:53 +02:00
prev_ptr->b_next = next_ptr;
else
front = next_ptr; /* this block was at front of chain */
if (next_ptr != NULL)
2005-04-21 16:53:53 +02:00
next_ptr->b_prev = prev_ptr;
else
rear = prev_ptr; /* this block was at rear of chain */
}
/*===========================================================================*
* cache_resize *
*===========================================================================*/
2012-03-25 20:25:53 +02:00
static void cache_resize(unsigned int blocksize, unsigned int bufs)
{
struct buf *bp;
struct inode *rip;
#define MINBUFS 10
assert(blocksize > 0);
assert(bufs >= MINBUFS);
for (bp = &buf[0]; bp < &buf[nr_bufs]; bp++)
if(bp->b_count != 0) panic("change blocksize with buffer in use");
for (rip = &inode[0]; rip < &inode[NR_INODES]; rip++)
if (rip->i_count > 0) panic("change blocksize with inode in use");
buf_pool(bufs);
fs_block_size = blocksize;
super_start = SUPER_BLOCK_BYTES / fs_block_size;
super_end = (SUPER_BLOCK_BYTES + _MIN_BLOCK_SIZE - 1) / fs_block_size;
}
/*===========================================================================*
* bufs_heuristic *
*===========================================================================*/
2012-03-25 20:25:53 +02:00
static int bufs_heuristic(struct super_block *sp)
{
u32_t btotal, bfree, bused;
blockstats(&btotal, &bfree, &bused);
return fs_bufs_heuristic(MINBUFS, btotal, bfree,
sp->s_block_size, major(sp->s_dev));
}
/*===========================================================================*
* set_blocksize *
*===========================================================================*/
2012-03-25 20:25:53 +02:00
void set_blocksize(struct super_block *sp)
{
int bufs;
cache_resize(sp->s_block_size, MINBUFS);
bufs = bufs_heuristic(sp);
cache_resize(sp->s_block_size, bufs);
/* Decide whether to use seconday cache or not.
* Only do this if
* - it's available, and
* - use of it hasn't been disabled for this fs, and
* - our main FS device isn't a memory device
*/
vmcache = 0;
if(vm_forgetblock(VM_BLOCKID_NONE) != ENOSYS &&
may_use_vmcache && major(sp->s_dev) != MEMORY_MAJOR) {
vmcache = 1;
}
}
/*===========================================================================*
* buf_pool *
*===========================================================================*/
2012-03-25 20:25:53 +02:00
void buf_pool(int new_nr_bufs)
{
/* Initialize the buffer pool. */
register struct buf *bp;
assert(new_nr_bufs >= MINBUFS);
if(nr_bufs > 0) {
assert(buf);
(void) fs_sync();
for (bp = &buf[0]; bp < &buf[nr_bufs]; bp++) {
if(bp->bp) {
assert(bp->b_bytes > 0);
free_contig(bp->bp, bp->b_bytes);
}
}
}
if(buf)
free(buf);
if(!(buf = calloc(sizeof(buf[0]), new_nr_bufs)))
panic("couldn't allocate buf list (%d)", new_nr_bufs);
if(buf_hash)
free(buf_hash);
if(!(buf_hash = calloc(sizeof(buf_hash[0]), new_nr_bufs)))
panic("couldn't allocate buf hash list (%d)", new_nr_bufs);
nr_bufs = new_nr_bufs;
bufs_in_use = 0;
front = &buf[0];
rear = &buf[nr_bufs - 1];
for (bp = &buf[0]; bp < &buf[nr_bufs]; bp++) {
bp->b_blocknr = NO_BLOCK;
bp->b_dev = NO_DEV;
bp->b_next = bp + 1;
bp->b_prev = bp - 1;
bp->bp = NULL;
bp->b_bytes = 0;
}
front->b_prev = NULL;
rear->b_next = NULL;
for (bp = &buf[0]; bp < &buf[nr_bufs]; bp++) bp->b_hash = bp->b_next;
buf_hash[0] = front;
vm_forgetblocks();
}