linux_dsm_epyc7002/fs/gfs2/bmap.c

1492 lines
37 KiB
C
Raw Normal View History

/*
* Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
* Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
*
* This copyrighted material is made available to anyone wishing to use,
* modify, copy, or redistribute it subject to the terms and conditions
* of the GNU General Public License version 2.
*/
#include <linux/spinlock.h>
#include <linux/completion.h>
#include <linux/buffer_head.h>
#include <linux/blkdev.h>
#include <linux/gfs2_ondisk.h>
#include <linux/crc32.h>
#include "gfs2.h"
#include "incore.h"
#include "bmap.h"
#include "glock.h"
#include "inode.h"
#include "meta_io.h"
#include "quota.h"
#include "rgrp.h"
#include "log.h"
GFS2: deallocation performance patch This patch is a performance improvement to GFS2's dealloc code. Rather than update the quota file and statfs file for every single block that's stripped off in unlink function do_strip, this patch keeps track and updates them once for every layer that's stripped. This is done entirely inside the existing transaction, so there should be no risk of corruption. The other functions that deallocate blocks will be unaffected because they are using wrapper functions that do the same thing that they do today. I tested this code on my roth cluster by creating 200 files in a directory, each of which is 100MB, then on four nodes, I simultaneously deleted the files, thus competing for GFS2 resources (but different files). The commands I used were: [root@roth-01]# time for i in `seq 1 4 200` ; do rm /mnt/gfs2/bigdir/gfs2.$i; done [root@roth-02]# time for i in `seq 2 4 200` ; do rm /mnt/gfs2/bigdir/gfs2.$i; done [root@roth-03]# time for i in `seq 3 4 200` ; do rm /mnt/gfs2/bigdir/gfs2.$i; done [root@roth-05]# time for i in `seq 4 4 200` ; do rm /mnt/gfs2/bigdir/gfs2.$i; done The performance increase was significant: roth-01 roth-02 roth-03 roth-05 --------- --------- --------- --------- old: real 0m34.027 0m25.021s 0m23.906s 0m35.646s new: real 0m22.379s 0m24.362s 0m24.133s 0m18.562s Total time spent deleting: old: 118.6s new: 89.4 For this particular case, this showed a 25% performance increase for GFS2 unlinks. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2011-02-24 04:11:33 +07:00
#include "super.h"
#include "trans.h"
[GFS2] Make journaled data files identical to normal files on disk This is a very large patch, with a few still to be resolved issues so you might want to check out the previous head of the tree since this is known to be unstable. Fixes for the various bugs will be forthcoming shortly. This patch removes the special data format which has been used up till now for journaled data files. Directories still retain the old format so that they will remain on disk compatible with earlier releases. As a result you can now do the following with journaled data files: 1) mmap them 2) export them over NFS 3) convert to/from normal files whenever you want to (the zero length restriction is gone) In addition the level at which GFS' locking is done has changed for all files (since they all now use the page cache) such that the locking is done at the page cache level rather than the level of the fs operations. This should mean that things like loopback mounts and other things which touch the page cache directly should now work. Current known issues: 1. There is a lock mode inversion problem related to the resource group hold function which needs to be resolved. 2. Any significant amount of I/O causes an oops with an offset of hex 320 (NULL pointer dereference) which appears to be related to a journaled data buffer appearing on a list where it shouldn't be. 3. Direct I/O writes are disabled for the time being (will reappear later) 4. There is probably a deadlock between the page lock and GFS' locks under certain combinations of mmap and fs operation I/O. 5. Issue relating to ref counting on internally used inodes causes a hang on umount (discovered before this patch, and not fixed by it) 6. One part of the directory metadata is different from GFS1 and will need to be resolved before next release. Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2006-02-08 18:50:51 +07:00
#include "dir.h"
#include "util.h"
#include "trace_gfs2.h"
/* This doesn't need to be that large as max 64 bit pointers in a 4k
* block is 512, so __u16 is fine for that. It saves stack space to
* keep it small.
*/
struct metapath {
struct buffer_head *mp_bh[GFS2_MAX_META_HEIGHT];
__u16 mp_list[GFS2_MAX_META_HEIGHT];
};
struct strip_mine {
int sm_first;
unsigned int sm_height;
};
/**
* gfs2_unstuffer_page - unstuff a stuffed inode into a block cached by a page
* @ip: the inode
* @dibh: the dinode buffer
* @block: the block number that was allocated
* @page: The (optional) page. This is looked up if @page is NULL
*
* Returns: errno
*/
static int gfs2_unstuffer_page(struct gfs2_inode *ip, struct buffer_head *dibh,
u64 block, struct page *page)
{
struct inode *inode = &ip->i_inode;
struct buffer_head *bh;
int release = 0;
if (!page || page->index) {
page = find_or_create_page(inode->i_mapping, 0, GFP_NOFS);
if (!page)
return -ENOMEM;
release = 1;
}
if (!PageUptodate(page)) {
void *kaddr = kmap(page);
u64 dsize = i_size_read(inode);
if (dsize > (dibh->b_size - sizeof(struct gfs2_dinode)))
dsize = dibh->b_size - sizeof(struct gfs2_dinode);
memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize);
mm, fs: get rid of PAGE_CACHE_* and page_cache_{get,release} macros PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} macros were introduced *long* time ago with promise that one day it will be possible to implement page cache with bigger chunks than PAGE_SIZE. This promise never materialized. And unlikely will. We have many places where PAGE_CACHE_SIZE assumed to be equal to PAGE_SIZE. And it's constant source of confusion on whether PAGE_CACHE_* or PAGE_* constant should be used in a particular case, especially on the border between fs and mm. Global switching to PAGE_CACHE_SIZE != PAGE_SIZE would cause to much breakage to be doable. Let's stop pretending that pages in page cache are special. They are not. The changes are pretty straight-forward: - <foo> << (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>; - <foo> >> (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>; - PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} -> PAGE_{SIZE,SHIFT,MASK,ALIGN}; - page_cache_get() -> get_page(); - page_cache_release() -> put_page(); This patch contains automated changes generated with coccinelle using script below. For some reason, coccinelle doesn't patch header files. I've called spatch for them manually. The only adjustment after coccinelle is revert of changes to PAGE_CAHCE_ALIGN definition: we are going to drop it later. There are few places in the code where coccinelle didn't reach. I'll fix them manually in a separate patch. Comments and documentation also will be addressed with the separate patch. virtual patch @@ expression E; @@ - E << (PAGE_CACHE_SHIFT - PAGE_SHIFT) + E @@ expression E; @@ - E >> (PAGE_CACHE_SHIFT - PAGE_SHIFT) + E @@ @@ - PAGE_CACHE_SHIFT + PAGE_SHIFT @@ @@ - PAGE_CACHE_SIZE + PAGE_SIZE @@ @@ - PAGE_CACHE_MASK + PAGE_MASK @@ expression E; @@ - PAGE_CACHE_ALIGN(E) + PAGE_ALIGN(E) @@ expression E; @@ - page_cache_get(E) + get_page(E) @@ expression E; @@ - page_cache_release(E) + put_page(E) Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Acked-by: Michal Hocko <mhocko@suse.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-04-01 19:29:47 +07:00
memset(kaddr + dsize, 0, PAGE_SIZE - dsize);
kunmap(page);
SetPageUptodate(page);
}
if (!page_has_buffers(page))
create_empty_buffers(page, 1 << inode->i_blkbits,
(1 << BH_Uptodate));
bh = page_buffers(page);
if (!buffer_mapped(bh))
map_bh(bh, inode->i_sb, block);
set_buffer_uptodate(bh);
if (!gfs2_is_jdata(ip))
mark_buffer_dirty(bh);
if (!gfs2_is_writeback(ip))
gfs2_trans_add_data(ip->i_gl, bh);
if (release) {
unlock_page(page);
mm, fs: get rid of PAGE_CACHE_* and page_cache_{get,release} macros PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} macros were introduced *long* time ago with promise that one day it will be possible to implement page cache with bigger chunks than PAGE_SIZE. This promise never materialized. And unlikely will. We have many places where PAGE_CACHE_SIZE assumed to be equal to PAGE_SIZE. And it's constant source of confusion on whether PAGE_CACHE_* or PAGE_* constant should be used in a particular case, especially on the border between fs and mm. Global switching to PAGE_CACHE_SIZE != PAGE_SIZE would cause to much breakage to be doable. Let's stop pretending that pages in page cache are special. They are not. The changes are pretty straight-forward: - <foo> << (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>; - <foo> >> (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>; - PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} -> PAGE_{SIZE,SHIFT,MASK,ALIGN}; - page_cache_get() -> get_page(); - page_cache_release() -> put_page(); This patch contains automated changes generated with coccinelle using script below. For some reason, coccinelle doesn't patch header files. I've called spatch for them manually. The only adjustment after coccinelle is revert of changes to PAGE_CAHCE_ALIGN definition: we are going to drop it later. There are few places in the code where coccinelle didn't reach. I'll fix them manually in a separate patch. Comments and documentation also will be addressed with the separate patch. virtual patch @@ expression E; @@ - E << (PAGE_CACHE_SHIFT - PAGE_SHIFT) + E @@ expression E; @@ - E >> (PAGE_CACHE_SHIFT - PAGE_SHIFT) + E @@ @@ - PAGE_CACHE_SHIFT + PAGE_SHIFT @@ @@ - PAGE_CACHE_SIZE + PAGE_SIZE @@ @@ - PAGE_CACHE_MASK + PAGE_MASK @@ expression E; @@ - PAGE_CACHE_ALIGN(E) + PAGE_ALIGN(E) @@ expression E; @@ - page_cache_get(E) + get_page(E) @@ expression E; @@ - page_cache_release(E) + put_page(E) Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Acked-by: Michal Hocko <mhocko@suse.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-04-01 19:29:47 +07:00
put_page(page);
}
return 0;
}
/**
* gfs2_unstuff_dinode - Unstuff a dinode when the data has grown too big
* @ip: The GFS2 inode to unstuff
* @page: The (optional) page. This is looked up if the @page is NULL
*
* This routine unstuffs a dinode and returns it to a "normal" state such
* that the height can be grown in the traditional way.
*
* Returns: errno
*/
int gfs2_unstuff_dinode(struct gfs2_inode *ip, struct page *page)
{
struct buffer_head *bh, *dibh;
struct gfs2_dinode *di;
u64 block = 0;
[GFS2] Make journaled data files identical to normal files on disk This is a very large patch, with a few still to be resolved issues so you might want to check out the previous head of the tree since this is known to be unstable. Fixes for the various bugs will be forthcoming shortly. This patch removes the special data format which has been used up till now for journaled data files. Directories still retain the old format so that they will remain on disk compatible with earlier releases. As a result you can now do the following with journaled data files: 1) mmap them 2) export them over NFS 3) convert to/from normal files whenever you want to (the zero length restriction is gone) In addition the level at which GFS' locking is done has changed for all files (since they all now use the page cache) such that the locking is done at the page cache level rather than the level of the fs operations. This should mean that things like loopback mounts and other things which touch the page cache directly should now work. Current known issues: 1. There is a lock mode inversion problem related to the resource group hold function which needs to be resolved. 2. Any significant amount of I/O causes an oops with an offset of hex 320 (NULL pointer dereference) which appears to be related to a journaled data buffer appearing on a list where it shouldn't be. 3. Direct I/O writes are disabled for the time being (will reappear later) 4. There is probably a deadlock between the page lock and GFS' locks under certain combinations of mmap and fs operation I/O. 5. Issue relating to ref counting on internally used inodes causes a hang on umount (discovered before this patch, and not fixed by it) 6. One part of the directory metadata is different from GFS1 and will need to be resolved before next release. Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2006-02-08 18:50:51 +07:00
int isdir = gfs2_is_dir(ip);
int error;
down_write(&ip->i_rw_mutex);
error = gfs2_meta_inode_buffer(ip, &dibh);
if (error)
goto out;
if (i_size_read(&ip->i_inode)) {
/* Get a free block, fill it with the stuffed data,
and write it out to disk */
unsigned int n = 1;
error = gfs2_alloc_blocks(ip, &block, &n, 0, NULL);
if (error)
goto out_brelse;
[GFS2] Make journaled data files identical to normal files on disk This is a very large patch, with a few still to be resolved issues so you might want to check out the previous head of the tree since this is known to be unstable. Fixes for the various bugs will be forthcoming shortly. This patch removes the special data format which has been used up till now for journaled data files. Directories still retain the old format so that they will remain on disk compatible with earlier releases. As a result you can now do the following with journaled data files: 1) mmap them 2) export them over NFS 3) convert to/from normal files whenever you want to (the zero length restriction is gone) In addition the level at which GFS' locking is done has changed for all files (since they all now use the page cache) such that the locking is done at the page cache level rather than the level of the fs operations. This should mean that things like loopback mounts and other things which touch the page cache directly should now work. Current known issues: 1. There is a lock mode inversion problem related to the resource group hold function which needs to be resolved. 2. Any significant amount of I/O causes an oops with an offset of hex 320 (NULL pointer dereference) which appears to be related to a journaled data buffer appearing on a list where it shouldn't be. 3. Direct I/O writes are disabled for the time being (will reappear later) 4. There is probably a deadlock between the page lock and GFS' locks under certain combinations of mmap and fs operation I/O. 5. Issue relating to ref counting on internally used inodes causes a hang on umount (discovered before this patch, and not fixed by it) 6. One part of the directory metadata is different from GFS1 and will need to be resolved before next release. Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2006-02-08 18:50:51 +07:00
if (isdir) {
gfs2_trans_add_unrevoke(GFS2_SB(&ip->i_inode), block, 1);
error = gfs2_dir_get_new_buffer(ip, block, &bh);
if (error)
goto out_brelse;
gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_meta_header),
dibh, sizeof(struct gfs2_dinode));
brelse(bh);
} else {
error = gfs2_unstuffer_page(ip, dibh, block, page);
if (error)
goto out_brelse;
}
}
/* Set up the pointer to the new block */
gfs2_trans_add_meta(ip->i_gl, dibh);
di = (struct gfs2_dinode *)dibh->b_data;
gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
if (i_size_read(&ip->i_inode)) {
*(__be64 *)(di + 1) = cpu_to_be64(block);
gfs2_add_inode_blocks(&ip->i_inode, 1);
di->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
}
ip->i_height = 1;
di->di_height = cpu_to_be16(1);
out_brelse:
brelse(dibh);
out:
up_write(&ip->i_rw_mutex);
return error;
}
/**
* find_metapath - Find path through the metadata tree
* @sdp: The superblock
* @mp: The metapath to return the result in
* @block: The disk block to look up
* @height: The pre-calculated height of the metadata tree
*
* This routine returns a struct metapath structure that defines a path
* through the metadata of inode "ip" to get to block "block".
*
* Example:
* Given: "ip" is a height 3 file, "offset" is 101342453, and this is a
* filesystem with a blocksize of 4096.
*
* find_metapath() would return a struct metapath structure set to:
* mp_offset = 101342453, mp_height = 3, mp_list[0] = 0, mp_list[1] = 48,
* and mp_list[2] = 165.
*
* That means that in order to get to the block containing the byte at
* offset 101342453, we would load the indirect block pointed to by pointer
* 0 in the dinode. We would then load the indirect block pointed to by
* pointer 48 in that indirect block. We would then load the data block
* pointed to by pointer 165 in that indirect block.
*
* ----------------------------------------
* | Dinode | |
* | | 4|
* | |0 1 2 3 4 5 9|
* | | 6|
* ----------------------------------------
* |
* |
* V
* ----------------------------------------
* | Indirect Block |
* | 5|
* | 4 4 4 4 4 5 5 1|
* |0 5 6 7 8 9 0 1 2|
* ----------------------------------------
* |
* |
* V
* ----------------------------------------
* | Indirect Block |
* | 1 1 1 1 1 5|
* | 6 6 6 6 6 1|
* |0 3 4 5 6 7 2|
* ----------------------------------------
* |
* |
* V
* ----------------------------------------
* | Data block containing offset |
* | 101342453 |
* | |
* | |
* ----------------------------------------
*
*/
static void find_metapath(const struct gfs2_sbd *sdp, u64 block,
struct metapath *mp, unsigned int height)
{
unsigned int i;
for (i = height; i--;)
mp->mp_list[i] = do_div(block, sdp->sd_inptrs);
}
static inline unsigned int metapath_branch_start(const struct metapath *mp)
{
if (mp->mp_list[0] == 0)
return 2;
return 1;
}
/**
* metapointer - Return pointer to start of metadata in a buffer
* @height: The metadata height (0 = dinode)
* @mp: The metapath
*
* Return a pointer to the block number of the next height of the metadata
* tree given a buffer containing the pointer to the current height of the
* metadata tree.
*/
static inline __be64 *metapointer(unsigned int height, const struct metapath *mp)
{
struct buffer_head *bh = mp->mp_bh[height];
unsigned int head_size = (height > 0) ?
sizeof(struct gfs2_meta_header) : sizeof(struct gfs2_dinode);
return ((__be64 *)(bh->b_data + head_size)) + mp->mp_list[height];
}
static void gfs2_metapath_ra(struct gfs2_glock *gl,
const struct buffer_head *bh, const __be64 *pos)
{
struct buffer_head *rabh;
const __be64 *endp = (const __be64 *)(bh->b_data + bh->b_size);
const __be64 *t;
for (t = pos; t < endp; t++) {
if (!*t)
continue;
rabh = gfs2_getbuf(gl, be64_to_cpu(*t), CREATE);
if (trylock_buffer(rabh)) {
if (!buffer_uptodate(rabh)) {
rabh->b_end_io = end_buffer_read_sync;
submit_bh(REQ_OP_READ, REQ_RAHEAD | REQ_META,
rabh);
continue;
}
unlock_buffer(rabh);
}
brelse(rabh);
}
}
/**
* lookup_metapath - Walk the metadata tree to a specific point
* @ip: The inode
* @mp: The metapath
*
* Assumes that the inode's buffer has already been looked up and
* hooked onto mp->mp_bh[0] and that the metapath has been initialised
* by find_metapath().
*
* If this function encounters part of the tree which has not been
* allocated, it returns the current height of the tree at the point
* at which it found the unallocated block. Blocks which are found are
* added to the mp->mp_bh[] list.
*
* Returns: error or height of metadata tree
*/
static int lookup_metapath(struct gfs2_inode *ip, struct metapath *mp)
{
unsigned int end_of_metadata = ip->i_height - 1;
unsigned int x;
__be64 *ptr;
u64 dblock;
int ret;
for (x = 0; x < end_of_metadata; x++) {
ptr = metapointer(x, mp);
dblock = be64_to_cpu(*ptr);
if (!dblock)
return x + 1;
ret = gfs2_meta_indirect_buffer(ip, x+1, dblock, &mp->mp_bh[x+1]);
if (ret)
return ret;
}
return ip->i_height;
}
static inline void release_metapath(struct metapath *mp)
{
int i;
for (i = 0; i < GFS2_MAX_META_HEIGHT; i++) {
if (mp->mp_bh[i] == NULL)
break;
brelse(mp->mp_bh[i]);
}
}
/**
* gfs2_extent_length - Returns length of an extent of blocks
* @start: Start of the buffer
* @len: Length of the buffer in bytes
* @ptr: Current position in the buffer
* @limit: Max extent length to return (0 = unlimited)
* @eob: Set to 1 if we hit "end of block"
*
* If the first block is zero (unallocated) it will return the number of
* unallocated blocks in the extent, otherwise it will return the number
* of contiguous blocks in the extent.
*
* Returns: The length of the extent (minimum of one block)
*/
static inline unsigned int gfs2_extent_length(void *start, unsigned int len, __be64 *ptr, size_t limit, int *eob)
{
const __be64 *end = (start + len);
const __be64 *first = ptr;
u64 d = be64_to_cpu(*ptr);
*eob = 0;
do {
ptr++;
if (ptr >= end)
break;
if (limit && --limit == 0)
break;
if (d)
d++;
} while(be64_to_cpu(*ptr) == d);
if (ptr >= end)
*eob = 1;
return (ptr - first);
}
static inline void bmap_lock(struct gfs2_inode *ip, int create)
{
if (create)
down_write(&ip->i_rw_mutex);
else
down_read(&ip->i_rw_mutex);
}
static inline void bmap_unlock(struct gfs2_inode *ip, int create)
{
if (create)
up_write(&ip->i_rw_mutex);
else
up_read(&ip->i_rw_mutex);
}
static inline __be64 *gfs2_indirect_init(struct metapath *mp,
struct gfs2_glock *gl, unsigned int i,
unsigned offset, u64 bn)
{
__be64 *ptr = (__be64 *)(mp->mp_bh[i - 1]->b_data +
((i > 1) ? sizeof(struct gfs2_meta_header) :
sizeof(struct gfs2_dinode)));
BUG_ON(i < 1);
BUG_ON(mp->mp_bh[i] != NULL);
mp->mp_bh[i] = gfs2_meta_new(gl, bn);
gfs2_trans_add_meta(gl, mp->mp_bh[i]);
gfs2_metatype_set(mp->mp_bh[i], GFS2_METATYPE_IN, GFS2_FORMAT_IN);
gfs2_buffer_clear_tail(mp->mp_bh[i], sizeof(struct gfs2_meta_header));
ptr += offset;
*ptr = cpu_to_be64(bn);
return ptr;
}
enum alloc_state {
ALLOC_DATA = 0,
ALLOC_GROW_DEPTH = 1,
ALLOC_GROW_HEIGHT = 2,
/* ALLOC_UNSTUFF = 3, TBD and rather complicated */
};
/**
* gfs2_bmap_alloc - Build a metadata tree of the requested height
* @inode: The GFS2 inode
* @lblock: The logical starting block of the extent
* @bh_map: This is used to return the mapping details
* @mp: The metapath
* @sheight: The starting height (i.e. whats already mapped)
* @height: The height to build to
* @maxlen: The max number of data blocks to alloc
*
* In this routine we may have to alloc:
* i) Indirect blocks to grow the metadata tree height
* ii) Indirect blocks to fill in lower part of the metadata tree
* iii) Data blocks
*
* The function is in two parts. The first part works out the total
* number of blocks which we need. The second part does the actual
* allocation asking for an extent at a time (if enough contiguous free
* blocks are available, there will only be one request per bmap call)
* and uses the state machine to initialise the blocks in order.
*
* Returns: errno on error
*/
static int gfs2_bmap_alloc(struct inode *inode, const sector_t lblock,
struct buffer_head *bh_map, struct metapath *mp,
const unsigned int sheight,
const unsigned int height,
const size_t maxlen)
{
struct gfs2_inode *ip = GFS2_I(inode);
struct gfs2_sbd *sdp = GFS2_SB(inode);
struct super_block *sb = sdp->sd_vfs;
struct buffer_head *dibh = mp->mp_bh[0];
u64 bn, dblock = 0;
unsigned n, i, blks, alloced = 0, iblks = 0, branch_start = 0;
unsigned dblks = 0;
unsigned ptrs_per_blk;
const unsigned end_of_metadata = height - 1;
int ret;
int eob = 0;
enum alloc_state state;
__be64 *ptr;
__be64 zero_bn = 0;
BUG_ON(sheight < 1);
BUG_ON(dibh == NULL);
gfs2_trans_add_meta(ip->i_gl, dibh);
if (height == sheight) {
struct buffer_head *bh;
/* Bottom indirect block exists, find unalloced extent size */
ptr = metapointer(end_of_metadata, mp);
bh = mp->mp_bh[end_of_metadata];
dblks = gfs2_extent_length(bh->b_data, bh->b_size, ptr, maxlen,
&eob);
BUG_ON(dblks < 1);
state = ALLOC_DATA;
} else {
/* Need to allocate indirect blocks */
ptrs_per_blk = height > 1 ? sdp->sd_inptrs : sdp->sd_diptrs;
dblks = min(maxlen, (size_t)(ptrs_per_blk -
mp->mp_list[end_of_metadata]));
if (height == ip->i_height) {
/* Writing into existing tree, extend tree down */
iblks = height - sheight;
state = ALLOC_GROW_DEPTH;
} else {
/* Building up tree height */
state = ALLOC_GROW_HEIGHT;
iblks = height - ip->i_height;
branch_start = metapath_branch_start(mp);
iblks += (height - branch_start);
}
}
/* start of the second part of the function (state machine) */
blks = dblks + iblks;
i = sheight;
do {
int error;
n = blks - alloced;
error = gfs2_alloc_blocks(ip, &bn, &n, 0, NULL);
if (error)
return error;
alloced += n;
if (state != ALLOC_DATA || gfs2_is_jdata(ip))
gfs2_trans_add_unrevoke(sdp, bn, n);
switch (state) {
/* Growing height of tree */
case ALLOC_GROW_HEIGHT:
if (i == 1) {
ptr = (__be64 *)(dibh->b_data +
sizeof(struct gfs2_dinode));
zero_bn = *ptr;
}
for (; i - 1 < height - ip->i_height && n > 0; i++, n--)
gfs2_indirect_init(mp, ip->i_gl, i, 0, bn++);
if (i - 1 == height - ip->i_height) {
i--;
gfs2_buffer_copy_tail(mp->mp_bh[i],
sizeof(struct gfs2_meta_header),
dibh, sizeof(struct gfs2_dinode));
gfs2_buffer_clear_tail(dibh,
sizeof(struct gfs2_dinode) +
sizeof(__be64));
ptr = (__be64 *)(mp->mp_bh[i]->b_data +
sizeof(struct gfs2_meta_header));
*ptr = zero_bn;
state = ALLOC_GROW_DEPTH;
for(i = branch_start; i < height; i++) {
if (mp->mp_bh[i] == NULL)
break;
brelse(mp->mp_bh[i]);
mp->mp_bh[i] = NULL;
}
i = branch_start;
}
if (n == 0)
break;
/* Branching from existing tree */
case ALLOC_GROW_DEPTH:
if (i > 1 && i < height)
gfs2_trans_add_meta(ip->i_gl, mp->mp_bh[i-1]);
for (; i < height && n > 0; i++, n--)
gfs2_indirect_init(mp, ip->i_gl, i,
mp->mp_list[i-1], bn++);
if (i == height)
state = ALLOC_DATA;
if (n == 0)
break;
/* Tree complete, adding data blocks */
case ALLOC_DATA:
BUG_ON(n > dblks);
BUG_ON(mp->mp_bh[end_of_metadata] == NULL);
gfs2_trans_add_meta(ip->i_gl, mp->mp_bh[end_of_metadata]);
dblks = n;
ptr = metapointer(end_of_metadata, mp);
dblock = bn;
while (n-- > 0)
*ptr++ = cpu_to_be64(bn++);
if (buffer_zeronew(bh_map)) {
ret = sb_issue_zeroout(sb, dblock, dblks,
GFP_NOFS);
if (ret) {
fs_err(sdp,
"Failed to zero data buffers\n");
clear_buffer_zeronew(bh_map);
}
}
break;
}
} while ((state != ALLOC_DATA) || !dblock);
ip->i_height = height;
gfs2_add_inode_blocks(&ip->i_inode, alloced);
gfs2_dinode_out(ip, mp->mp_bh[0]->b_data);
map_bh(bh_map, inode->i_sb, dblock);
bh_map->b_size = dblks << inode->i_blkbits;
set_buffer_new(bh_map);
return 0;
}
/**
* gfs2_block_map - Map a block from an inode to a disk block
* @inode: The inode
* @lblock: The logical block number
* @bh_map: The bh to be mapped
* @create: True if its ok to alloc blocks to satify the request
*
* Sets buffer_mapped() if successful, sets buffer_boundary() if a
* read of metadata will be required before the next block can be
* mapped. Sets buffer_new() if new blocks were allocated.
*
* Returns: errno
*/
int gfs2_block_map(struct inode *inode, sector_t lblock,
struct buffer_head *bh_map, int create)
{
struct gfs2_inode *ip = GFS2_I(inode);
struct gfs2_sbd *sdp = GFS2_SB(inode);
unsigned int bsize = sdp->sd_sb.sb_bsize;
const size_t maxlen = bh_map->b_size >> inode->i_blkbits;
const u64 *arr = sdp->sd_heightsize;
__be64 *ptr;
u64 size;
struct metapath mp;
int ret;
int eob;
unsigned int len;
struct buffer_head *bh;
u8 height;
BUG_ON(maxlen == 0);
memset(mp.mp_bh, 0, sizeof(mp.mp_bh));
bmap_lock(ip, create);
clear_buffer_mapped(bh_map);
clear_buffer_new(bh_map);
clear_buffer_boundary(bh_map);
trace_gfs2_bmap(ip, bh_map, lblock, create, 1);
if (gfs2_is_dir(ip)) {
bsize = sdp->sd_jbsize;
arr = sdp->sd_jheightsize;
}
ret = gfs2_meta_inode_buffer(ip, &mp.mp_bh[0]);
if (ret)
goto out;
height = ip->i_height;
size = (lblock + 1) * bsize;
while (size > arr[height])
height++;
find_metapath(sdp, lblock, &mp, height);
ret = 1;
if (height > ip->i_height || gfs2_is_stuffed(ip))
goto do_alloc;
ret = lookup_metapath(ip, &mp);
if (ret < 0)
goto out;
if (ret != ip->i_height)
goto do_alloc;
ptr = metapointer(ip->i_height - 1, &mp);
if (*ptr == 0)
goto do_alloc;
map_bh(bh_map, inode->i_sb, be64_to_cpu(*ptr));
bh = mp.mp_bh[ip->i_height - 1];
len = gfs2_extent_length(bh->b_data, bh->b_size, ptr, maxlen, &eob);
bh_map->b_size = (len << inode->i_blkbits);
if (eob)
set_buffer_boundary(bh_map);
ret = 0;
out:
release_metapath(&mp);
trace_gfs2_bmap(ip, bh_map, lblock, create, ret);
bmap_unlock(ip, create);
return ret;
do_alloc:
/* All allocations are done here, firstly check create flag */
if (!create) {
BUG_ON(gfs2_is_stuffed(ip));
ret = 0;
goto out;
}
/* At this point ret is the tree depth of already allocated blocks */
ret = gfs2_bmap_alloc(inode, lblock, bh_map, &mp, ret, height, maxlen);
goto out;
}
/*
* Deprecated: do not use in new code
*/
int gfs2_extent_map(struct inode *inode, u64 lblock, int *new, u64 *dblock, unsigned *extlen)
{
struct buffer_head bh = { .b_state = 0, .b_blocknr = 0 };
int ret;
int create = *new;
BUG_ON(!extlen);
BUG_ON(!dblock);
BUG_ON(!new);
bh.b_size = 1 << (inode->i_blkbits + (create ? 0 : 5));
ret = gfs2_block_map(inode, lblock, &bh, create);
*extlen = bh.b_size >> inode->i_blkbits;
*dblock = bh.b_blocknr;
if (buffer_new(&bh))
*new = 1;
else
*new = 0;
return ret;
}
/**
* do_strip - Look for a layer a particular layer of the file and strip it off
* @ip: the inode
* @dibh: the dinode buffer
* @bh: A buffer of pointers
* @top: The first pointer in the buffer
* @bottom: One more than the last pointer
* @height: the height this buffer is at
* @sm: a pointer to a struct strip_mine
*
* Returns: errno
*/
static int do_strip(struct gfs2_inode *ip, struct buffer_head *dibh,
struct buffer_head *bh, __be64 *top, __be64 *bottom,
unsigned int height, struct strip_mine *sm)
{
struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
struct gfs2_rgrp_list rlist;
u64 bn, bstart;
GFS2: deallocation performance patch This patch is a performance improvement to GFS2's dealloc code. Rather than update the quota file and statfs file for every single block that's stripped off in unlink function do_strip, this patch keeps track and updates them once for every layer that's stripped. This is done entirely inside the existing transaction, so there should be no risk of corruption. The other functions that deallocate blocks will be unaffected because they are using wrapper functions that do the same thing that they do today. I tested this code on my roth cluster by creating 200 files in a directory, each of which is 100MB, then on four nodes, I simultaneously deleted the files, thus competing for GFS2 resources (but different files). The commands I used were: [root@roth-01]# time for i in `seq 1 4 200` ; do rm /mnt/gfs2/bigdir/gfs2.$i; done [root@roth-02]# time for i in `seq 2 4 200` ; do rm /mnt/gfs2/bigdir/gfs2.$i; done [root@roth-03]# time for i in `seq 3 4 200` ; do rm /mnt/gfs2/bigdir/gfs2.$i; done [root@roth-05]# time for i in `seq 4 4 200` ; do rm /mnt/gfs2/bigdir/gfs2.$i; done The performance increase was significant: roth-01 roth-02 roth-03 roth-05 --------- --------- --------- --------- old: real 0m34.027 0m25.021s 0m23.906s 0m35.646s new: real 0m22.379s 0m24.362s 0m24.133s 0m18.562s Total time spent deleting: old: 118.6s new: 89.4 For this particular case, this showed a 25% performance increase for GFS2 unlinks. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2011-02-24 04:11:33 +07:00
u32 blen, btotal;
__be64 *p;
unsigned int rg_blocks = 0;
int metadata;
unsigned int revokes = 0;
int x;
int error;
error = gfs2_rindex_update(sdp);
if (error)
return error;
if (!*top)
sm->sm_first = 0;
if (height != sm->sm_height)
return 0;
if (sm->sm_first) {
top++;
sm->sm_first = 0;
}
metadata = (height != ip->i_height - 1);
if (metadata)
revokes = (height) ? sdp->sd_inptrs : sdp->sd_diptrs;
else if (ip->i_depth)
revokes = sdp->sd_inptrs;
memset(&rlist, 0, sizeof(struct gfs2_rgrp_list));
bstart = 0;
blen = 0;
for (p = top; p < bottom; p++) {
if (!*p)
continue;
bn = be64_to_cpu(*p);
if (bstart + blen == bn)
blen++;
else {
if (bstart)
gfs2_rlist_add(ip, &rlist, bstart);
bstart = bn;
blen = 1;
}
}
if (bstart)
gfs2_rlist_add(ip, &rlist, bstart);
else
goto out; /* Nothing to do */
gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE);
for (x = 0; x < rlist.rl_rgrps; x++) {
struct gfs2_rgrpd *rgd;
rgd = rlist.rl_ghs[x].gh_gl->gl_object;
rg_blocks += rgd->rd_length;
}
error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs);
if (error)
goto out_rlist;
if (gfs2_rs_active(&ip->i_res)) /* needs to be done with the rgrp glock held */
gfs2_rs_deltree(&ip->i_res);
error = gfs2_trans_begin(sdp, rg_blocks + RES_DINODE +
RES_INDIRECT + RES_STATFS + RES_QUOTA,
revokes);
if (error)
goto out_rg_gunlock;
down_write(&ip->i_rw_mutex);
gfs2_trans_add_meta(ip->i_gl, dibh);
gfs2_trans_add_meta(ip->i_gl, bh);
bstart = 0;
blen = 0;
GFS2: deallocation performance patch This patch is a performance improvement to GFS2's dealloc code. Rather than update the quota file and statfs file for every single block that's stripped off in unlink function do_strip, this patch keeps track and updates them once for every layer that's stripped. This is done entirely inside the existing transaction, so there should be no risk of corruption. The other functions that deallocate blocks will be unaffected because they are using wrapper functions that do the same thing that they do today. I tested this code on my roth cluster by creating 200 files in a directory, each of which is 100MB, then on four nodes, I simultaneously deleted the files, thus competing for GFS2 resources (but different files). The commands I used were: [root@roth-01]# time for i in `seq 1 4 200` ; do rm /mnt/gfs2/bigdir/gfs2.$i; done [root@roth-02]# time for i in `seq 2 4 200` ; do rm /mnt/gfs2/bigdir/gfs2.$i; done [root@roth-03]# time for i in `seq 3 4 200` ; do rm /mnt/gfs2/bigdir/gfs2.$i; done [root@roth-05]# time for i in `seq 4 4 200` ; do rm /mnt/gfs2/bigdir/gfs2.$i; done The performance increase was significant: roth-01 roth-02 roth-03 roth-05 --------- --------- --------- --------- old: real 0m34.027 0m25.021s 0m23.906s 0m35.646s new: real 0m22.379s 0m24.362s 0m24.133s 0m18.562s Total time spent deleting: old: 118.6s new: 89.4 For this particular case, this showed a 25% performance increase for GFS2 unlinks. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2011-02-24 04:11:33 +07:00
btotal = 0;
for (p = top; p < bottom; p++) {
if (!*p)
continue;
bn = be64_to_cpu(*p);
if (bstart + blen == bn)
blen++;
else {
if (bstart) {
__gfs2_free_blocks(ip, bstart, blen, metadata);
GFS2: deallocation performance patch This patch is a performance improvement to GFS2's dealloc code. Rather than update the quota file and statfs file for every single block that's stripped off in unlink function do_strip, this patch keeps track and updates them once for every layer that's stripped. This is done entirely inside the existing transaction, so there should be no risk of corruption. The other functions that deallocate blocks will be unaffected because they are using wrapper functions that do the same thing that they do today. I tested this code on my roth cluster by creating 200 files in a directory, each of which is 100MB, then on four nodes, I simultaneously deleted the files, thus competing for GFS2 resources (but different files). The commands I used were: [root@roth-01]# time for i in `seq 1 4 200` ; do rm /mnt/gfs2/bigdir/gfs2.$i; done [root@roth-02]# time for i in `seq 2 4 200` ; do rm /mnt/gfs2/bigdir/gfs2.$i; done [root@roth-03]# time for i in `seq 3 4 200` ; do rm /mnt/gfs2/bigdir/gfs2.$i; done [root@roth-05]# time for i in `seq 4 4 200` ; do rm /mnt/gfs2/bigdir/gfs2.$i; done The performance increase was significant: roth-01 roth-02 roth-03 roth-05 --------- --------- --------- --------- old: real 0m34.027 0m25.021s 0m23.906s 0m35.646s new: real 0m22.379s 0m24.362s 0m24.133s 0m18.562s Total time spent deleting: old: 118.6s new: 89.4 For this particular case, this showed a 25% performance increase for GFS2 unlinks. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2011-02-24 04:11:33 +07:00
btotal += blen;
}
bstart = bn;
blen = 1;
}
*p = 0;
gfs2_add_inode_blocks(&ip->i_inode, -1);
}
if (bstart) {
__gfs2_free_blocks(ip, bstart, blen, metadata);
GFS2: deallocation performance patch This patch is a performance improvement to GFS2's dealloc code. Rather than update the quota file and statfs file for every single block that's stripped off in unlink function do_strip, this patch keeps track and updates them once for every layer that's stripped. This is done entirely inside the existing transaction, so there should be no risk of corruption. The other functions that deallocate blocks will be unaffected because they are using wrapper functions that do the same thing that they do today. I tested this code on my roth cluster by creating 200 files in a directory, each of which is 100MB, then on four nodes, I simultaneously deleted the files, thus competing for GFS2 resources (but different files). The commands I used were: [root@roth-01]# time for i in `seq 1 4 200` ; do rm /mnt/gfs2/bigdir/gfs2.$i; done [root@roth-02]# time for i in `seq 2 4 200` ; do rm /mnt/gfs2/bigdir/gfs2.$i; done [root@roth-03]# time for i in `seq 3 4 200` ; do rm /mnt/gfs2/bigdir/gfs2.$i; done [root@roth-05]# time for i in `seq 4 4 200` ; do rm /mnt/gfs2/bigdir/gfs2.$i; done The performance increase was significant: roth-01 roth-02 roth-03 roth-05 --------- --------- --------- --------- old: real 0m34.027 0m25.021s 0m23.906s 0m35.646s new: real 0m22.379s 0m24.362s 0m24.133s 0m18.562s Total time spent deleting: old: 118.6s new: 89.4 For this particular case, this showed a 25% performance increase for GFS2 unlinks. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2011-02-24 04:11:33 +07:00
btotal += blen;
}
GFS2: deallocation performance patch This patch is a performance improvement to GFS2's dealloc code. Rather than update the quota file and statfs file for every single block that's stripped off in unlink function do_strip, this patch keeps track and updates them once for every layer that's stripped. This is done entirely inside the existing transaction, so there should be no risk of corruption. The other functions that deallocate blocks will be unaffected because they are using wrapper functions that do the same thing that they do today. I tested this code on my roth cluster by creating 200 files in a directory, each of which is 100MB, then on four nodes, I simultaneously deleted the files, thus competing for GFS2 resources (but different files). The commands I used were: [root@roth-01]# time for i in `seq 1 4 200` ; do rm /mnt/gfs2/bigdir/gfs2.$i; done [root@roth-02]# time for i in `seq 2 4 200` ; do rm /mnt/gfs2/bigdir/gfs2.$i; done [root@roth-03]# time for i in `seq 3 4 200` ; do rm /mnt/gfs2/bigdir/gfs2.$i; done [root@roth-05]# time for i in `seq 4 4 200` ; do rm /mnt/gfs2/bigdir/gfs2.$i; done The performance increase was significant: roth-01 roth-02 roth-03 roth-05 --------- --------- --------- --------- old: real 0m34.027 0m25.021s 0m23.906s 0m35.646s new: real 0m22.379s 0m24.362s 0m24.133s 0m18.562s Total time spent deleting: old: 118.6s new: 89.4 For this particular case, this showed a 25% performance increase for GFS2 unlinks. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2011-02-24 04:11:33 +07:00
gfs2_statfs_change(sdp, 0, +btotal, 0);
gfs2_quota_change(ip, -(s64)btotal, ip->i_inode.i_uid,
ip->i_inode.i_gid);
ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
gfs2_dinode_out(ip, dibh->b_data);
up_write(&ip->i_rw_mutex);
gfs2_trans_end(sdp);
out_rg_gunlock:
gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs);
out_rlist:
gfs2_rlist_free(&rlist);
out:
return error;
}
/**
* recursive_scan - recursively scan through the end of a file
* @ip: the inode
* @dibh: the dinode buffer
* @mp: the path through the metadata to the point to start
* @height: the height the recursion is at
* @block: the indirect block to look at
* @first: 1 if this is the first block
* @sm: data opaque to this function to pass to @bc
*
* When this is first called @height and @block should be zero and
* @first should be 1.
*
* Returns: errno
*/
static int recursive_scan(struct gfs2_inode *ip, struct buffer_head *dibh,
struct metapath *mp, unsigned int height,
u64 block, int first, struct strip_mine *sm)
{
struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
struct buffer_head *bh = NULL;
__be64 *top, *bottom;
u64 bn;
int error;
int mh_size = sizeof(struct gfs2_meta_header);
if (!height) {
error = gfs2_meta_inode_buffer(ip, &bh);
if (error)
return error;
dibh = bh;
top = (__be64 *)(bh->b_data + sizeof(struct gfs2_dinode)) + mp->mp_list[0];
bottom = (__be64 *)(bh->b_data + sizeof(struct gfs2_dinode)) + sdp->sd_diptrs;
} else {
error = gfs2_meta_indirect_buffer(ip, height, block, &bh);
if (error)
return error;
top = (__be64 *)(bh->b_data + mh_size) +
(first ? mp->mp_list[height] : 0);
bottom = (__be64 *)(bh->b_data + mh_size) + sdp->sd_inptrs;
}
error = do_strip(ip, dibh, bh, top, bottom, height, sm);
if (error)
goto out;
if (height < ip->i_height - 1) {
gfs2_metapath_ra(ip->i_gl, bh, top);
for (; top < bottom; top++, first = 0) {
if (!*top)
continue;
bn = be64_to_cpu(*top);
error = recursive_scan(ip, dibh, mp, height + 1, bn,
first, sm);
if (error)
break;
}
}
out:
brelse(bh);
return error;
}
/**
* gfs2_block_truncate_page - Deal with zeroing out data for truncate
*
* This is partly borrowed from ext3.
*/
static int gfs2_block_truncate_page(struct address_space *mapping, loff_t from)
{
struct inode *inode = mapping->host;
struct gfs2_inode *ip = GFS2_I(inode);
mm, fs: get rid of PAGE_CACHE_* and page_cache_{get,release} macros PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} macros were introduced *long* time ago with promise that one day it will be possible to implement page cache with bigger chunks than PAGE_SIZE. This promise never materialized. And unlikely will. We have many places where PAGE_CACHE_SIZE assumed to be equal to PAGE_SIZE. And it's constant source of confusion on whether PAGE_CACHE_* or PAGE_* constant should be used in a particular case, especially on the border between fs and mm. Global switching to PAGE_CACHE_SIZE != PAGE_SIZE would cause to much breakage to be doable. Let's stop pretending that pages in page cache are special. They are not. The changes are pretty straight-forward: - <foo> << (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>; - <foo> >> (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>; - PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} -> PAGE_{SIZE,SHIFT,MASK,ALIGN}; - page_cache_get() -> get_page(); - page_cache_release() -> put_page(); This patch contains automated changes generated with coccinelle using script below. For some reason, coccinelle doesn't patch header files. I've called spatch for them manually. The only adjustment after coccinelle is revert of changes to PAGE_CAHCE_ALIGN definition: we are going to drop it later. There are few places in the code where coccinelle didn't reach. I'll fix them manually in a separate patch. Comments and documentation also will be addressed with the separate patch. virtual patch @@ expression E; @@ - E << (PAGE_CACHE_SHIFT - PAGE_SHIFT) + E @@ expression E; @@ - E >> (PAGE_CACHE_SHIFT - PAGE_SHIFT) + E @@ @@ - PAGE_CACHE_SHIFT + PAGE_SHIFT @@ @@ - PAGE_CACHE_SIZE + PAGE_SIZE @@ @@ - PAGE_CACHE_MASK + PAGE_MASK @@ expression E; @@ - PAGE_CACHE_ALIGN(E) + PAGE_ALIGN(E) @@ expression E; @@ - page_cache_get(E) + get_page(E) @@ expression E; @@ - page_cache_release(E) + put_page(E) Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Acked-by: Michal Hocko <mhocko@suse.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-04-01 19:29:47 +07:00
unsigned long index = from >> PAGE_SHIFT;
unsigned offset = from & (PAGE_SIZE-1);
unsigned blocksize, iblock, length, pos;
struct buffer_head *bh;
struct page *page;
int err;
page = find_or_create_page(mapping, index, GFP_NOFS);
if (!page)
return 0;
blocksize = inode->i_sb->s_blocksize;
length = blocksize - (offset & (blocksize - 1));
mm, fs: get rid of PAGE_CACHE_* and page_cache_{get,release} macros PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} macros were introduced *long* time ago with promise that one day it will be possible to implement page cache with bigger chunks than PAGE_SIZE. This promise never materialized. And unlikely will. We have many places where PAGE_CACHE_SIZE assumed to be equal to PAGE_SIZE. And it's constant source of confusion on whether PAGE_CACHE_* or PAGE_* constant should be used in a particular case, especially on the border between fs and mm. Global switching to PAGE_CACHE_SIZE != PAGE_SIZE would cause to much breakage to be doable. Let's stop pretending that pages in page cache are special. They are not. The changes are pretty straight-forward: - <foo> << (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>; - <foo> >> (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>; - PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} -> PAGE_{SIZE,SHIFT,MASK,ALIGN}; - page_cache_get() -> get_page(); - page_cache_release() -> put_page(); This patch contains automated changes generated with coccinelle using script below. For some reason, coccinelle doesn't patch header files. I've called spatch for them manually. The only adjustment after coccinelle is revert of changes to PAGE_CAHCE_ALIGN definition: we are going to drop it later. There are few places in the code where coccinelle didn't reach. I'll fix them manually in a separate patch. Comments and documentation also will be addressed with the separate patch. virtual patch @@ expression E; @@ - E << (PAGE_CACHE_SHIFT - PAGE_SHIFT) + E @@ expression E; @@ - E >> (PAGE_CACHE_SHIFT - PAGE_SHIFT) + E @@ @@ - PAGE_CACHE_SHIFT + PAGE_SHIFT @@ @@ - PAGE_CACHE_SIZE + PAGE_SIZE @@ @@ - PAGE_CACHE_MASK + PAGE_MASK @@ expression E; @@ - PAGE_CACHE_ALIGN(E) + PAGE_ALIGN(E) @@ expression E; @@ - page_cache_get(E) + get_page(E) @@ expression E; @@ - page_cache_release(E) + put_page(E) Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Acked-by: Michal Hocko <mhocko@suse.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-04-01 19:29:47 +07:00
iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
if (!page_has_buffers(page))
create_empty_buffers(page, blocksize, 0);
/* Find the buffer that contains "offset" */
bh = page_buffers(page);
pos = blocksize;
while (offset >= pos) {
bh = bh->b_this_page;
iblock++;
pos += blocksize;
}
err = 0;
if (!buffer_mapped(bh)) {
gfs2_block_map(inode, iblock, bh, 0);
/* unmapped? It's a hole - nothing to do */
if (!buffer_mapped(bh))
goto unlock;
}
/* Ok, it's mapped. Make sure it's up-to-date */
if (PageUptodate(page))
set_buffer_uptodate(bh);
if (!buffer_uptodate(bh)) {
err = -EIO;
ll_rw_block(REQ_OP_READ, 0, 1, &bh);
wait_on_buffer(bh);
/* Uhhuh. Read error. Complain and punt. */
if (!buffer_uptodate(bh))
goto unlock;
err = 0;
}
if (!gfs2_is_writeback(ip))
gfs2_trans_add_data(ip->i_gl, bh);
Pagecache zeroing: zero_user_segment, zero_user_segments and zero_user Simplify page cache zeroing of segments of pages through 3 functions zero_user_segments(page, start1, end1, start2, end2) Zeros two segments of the page. It takes the position where to start and end the zeroing which avoids length calculations and makes code clearer. zero_user_segment(page, start, end) Same for a single segment. zero_user(page, start, length) Length variant for the case where we know the length. We remove the zero_user_page macro. Issues: 1. Its a macro. Inline functions are preferable. 2. The KM_USER0 macro is only defined for HIGHMEM. Having to treat this special case everywhere makes the code needlessly complex. The parameter for zeroing is always KM_USER0 except in one single case that we open code. Avoiding KM_USER0 makes a lot of code not having to be dealing with the special casing for HIGHMEM anymore. Dealing with kmap is only necessary for HIGHMEM configurations. In those configurations we use KM_USER0 like we do for a series of other functions defined in highmem.h. Since KM_USER0 is depends on HIGHMEM the existing zero_user_page function could not be a macro. zero_user_* functions introduced here can be be inline because that constant is not used when these functions are called. Also extract the flushing of the caches to be outside of the kmap. [akpm@linux-foundation.org: fix nfs and ntfs build] [akpm@linux-foundation.org: fix ntfs build some more] Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Steven French <sfrench@us.ibm.com> Cc: Michael Halcrow <mhalcrow@us.ibm.com> Cc: <linux-ext4@vger.kernel.org> Cc: Steven Whitehouse <swhiteho@redhat.com> Cc: Trond Myklebust <trond.myklebust@fys.uio.no> Cc: "J. Bruce Fields" <bfields@fieldses.org> Cc: Anton Altaparmakov <aia21@cantab.net> Cc: Mark Fasheh <mark.fasheh@oracle.com> Cc: David Chinner <dgc@sgi.com> Cc: Michael Halcrow <mhalcrow@us.ibm.com> Cc: Steven French <sfrench@us.ibm.com> Cc: Steven Whitehouse <swhiteho@redhat.com> Cc: Trond Myklebust <trond.myklebust@fys.uio.no> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-05 13:28:29 +07:00
zero_user(page, offset, length);
mark_buffer_dirty(bh);
unlock:
unlock_page(page);
mm, fs: get rid of PAGE_CACHE_* and page_cache_{get,release} macros PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} macros were introduced *long* time ago with promise that one day it will be possible to implement page cache with bigger chunks than PAGE_SIZE. This promise never materialized. And unlikely will. We have many places where PAGE_CACHE_SIZE assumed to be equal to PAGE_SIZE. And it's constant source of confusion on whether PAGE_CACHE_* or PAGE_* constant should be used in a particular case, especially on the border between fs and mm. Global switching to PAGE_CACHE_SIZE != PAGE_SIZE would cause to much breakage to be doable. Let's stop pretending that pages in page cache are special. They are not. The changes are pretty straight-forward: - <foo> << (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>; - <foo> >> (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>; - PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} -> PAGE_{SIZE,SHIFT,MASK,ALIGN}; - page_cache_get() -> get_page(); - page_cache_release() -> put_page(); This patch contains automated changes generated with coccinelle using script below. For some reason, coccinelle doesn't patch header files. I've called spatch for them manually. The only adjustment after coccinelle is revert of changes to PAGE_CAHCE_ALIGN definition: we are going to drop it later. There are few places in the code where coccinelle didn't reach. I'll fix them manually in a separate patch. Comments and documentation also will be addressed with the separate patch. virtual patch @@ expression E; @@ - E << (PAGE_CACHE_SHIFT - PAGE_SHIFT) + E @@ expression E; @@ - E >> (PAGE_CACHE_SHIFT - PAGE_SHIFT) + E @@ @@ - PAGE_CACHE_SHIFT + PAGE_SHIFT @@ @@ - PAGE_CACHE_SIZE + PAGE_SIZE @@ @@ - PAGE_CACHE_MASK + PAGE_MASK @@ expression E; @@ - PAGE_CACHE_ALIGN(E) + PAGE_ALIGN(E) @@ expression E; @@ - page_cache_get(E) + get_page(E) @@ expression E; @@ - page_cache_release(E) + put_page(E) Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Acked-by: Michal Hocko <mhocko@suse.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-04-01 19:29:47 +07:00
put_page(page);
return err;
}
#define GFS2_JTRUNC_REVOKES 8192
/**
* gfs2_journaled_truncate - Wrapper for truncate_pagecache for jdata files
* @inode: The inode being truncated
* @oldsize: The original (larger) size
* @newsize: The new smaller size
*
* With jdata files, we have to journal a revoke for each block which is
* truncated. As a result, we need to split this into separate transactions
* if the number of pages being truncated gets too large.
*/
static int gfs2_journaled_truncate(struct inode *inode, u64 oldsize, u64 newsize)
{
struct gfs2_sbd *sdp = GFS2_SB(inode);
u64 max_chunk = GFS2_JTRUNC_REVOKES * sdp->sd_vfs->s_blocksize;
u64 chunk;
int error;
while (oldsize != newsize) {
chunk = oldsize - newsize;
if (chunk > max_chunk)
chunk = max_chunk;
truncate_pagecache(inode, oldsize - chunk);
oldsize -= chunk;
gfs2_trans_end(sdp);
error = gfs2_trans_begin(sdp, RES_DINODE, GFS2_JTRUNC_REVOKES);
if (error)
return error;
}
return 0;
}
static int trunc_start(struct inode *inode, u64 oldsize, u64 newsize)
{
struct gfs2_inode *ip = GFS2_I(inode);
struct gfs2_sbd *sdp = GFS2_SB(inode);
struct address_space *mapping = inode->i_mapping;
struct buffer_head *dibh;
int journaled = gfs2_is_jdata(ip);
int error;
if (journaled)
error = gfs2_trans_begin(sdp, RES_DINODE + RES_JDATA, GFS2_JTRUNC_REVOKES);
else
error = gfs2_trans_begin(sdp, RES_DINODE, 0);
if (error)
return error;
error = gfs2_meta_inode_buffer(ip, &dibh);
if (error)
goto out;
gfs2_trans_add_meta(ip->i_gl, dibh);
if (gfs2_is_stuffed(ip)) {
gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode) + newsize);
} else {
if (newsize & (u64)(sdp->sd_sb.sb_bsize - 1)) {
error = gfs2_block_truncate_page(mapping, newsize);
if (error)
goto out_brelse;
}
ip->i_diskflags |= GFS2_DIF_TRUNC_IN_PROG;
}
i_size_write(inode, newsize);
ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
gfs2_dinode_out(ip, dibh->b_data);
if (journaled)
error = gfs2_journaled_truncate(inode, oldsize, newsize);
else
truncate_pagecache(inode, newsize);
if (error) {
brelse(dibh);
return error;
}
out_brelse:
brelse(dibh);
out:
gfs2_trans_end(sdp);
return error;
}
static int trunc_dealloc(struct gfs2_inode *ip, u64 size)
{
struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
unsigned int height = ip->i_height;
u64 lblock;
struct metapath mp;
int error;
if (!size)
lblock = 0;
[GFS2] Make journaled data files identical to normal files on disk This is a very large patch, with a few still to be resolved issues so you might want to check out the previous head of the tree since this is known to be unstable. Fixes for the various bugs will be forthcoming shortly. This patch removes the special data format which has been used up till now for journaled data files. Directories still retain the old format so that they will remain on disk compatible with earlier releases. As a result you can now do the following with journaled data files: 1) mmap them 2) export them over NFS 3) convert to/from normal files whenever you want to (the zero length restriction is gone) In addition the level at which GFS' locking is done has changed for all files (since they all now use the page cache) such that the locking is done at the page cache level rather than the level of the fs operations. This should mean that things like loopback mounts and other things which touch the page cache directly should now work. Current known issues: 1. There is a lock mode inversion problem related to the resource group hold function which needs to be resolved. 2. Any significant amount of I/O causes an oops with an offset of hex 320 (NULL pointer dereference) which appears to be related to a journaled data buffer appearing on a list where it shouldn't be. 3. Direct I/O writes are disabled for the time being (will reappear later) 4. There is probably a deadlock between the page lock and GFS' locks under certain combinations of mmap and fs operation I/O. 5. Issue relating to ref counting on internally used inodes causes a hang on umount (discovered before this patch, and not fixed by it) 6. One part of the directory metadata is different from GFS1 and will need to be resolved before next release. Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2006-02-08 18:50:51 +07:00
else
lblock = (size - 1) >> sdp->sd_sb.sb_bsize_shift;
find_metapath(sdp, lblock, &mp, ip->i_height);
error = gfs2_rindex_update(sdp);
if (error)
return error;
error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
if (error)
return error;
while (height--) {
struct strip_mine sm;
sm.sm_first = !!size;
sm.sm_height = height;
error = recursive_scan(ip, NULL, &mp, 0, 0, 1, &sm);
if (error)
break;
}
gfs2_quota_unhold(ip);
return error;
}
static int trunc_end(struct gfs2_inode *ip)
{
struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
struct buffer_head *dibh;
int error;
error = gfs2_trans_begin(sdp, RES_DINODE, 0);
if (error)
return error;
down_write(&ip->i_rw_mutex);
error = gfs2_meta_inode_buffer(ip, &dibh);
if (error)
goto out;
if (!i_size_read(&ip->i_inode)) {
ip->i_height = 0;
ip->i_goal = ip->i_no_addr;
gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
gfs2_ordered_del_inode(ip);
}
ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
ip->i_diskflags &= ~GFS2_DIF_TRUNC_IN_PROG;
gfs2_trans_add_meta(ip->i_gl, dibh);
gfs2_dinode_out(ip, dibh->b_data);
brelse(dibh);
out:
up_write(&ip->i_rw_mutex);
gfs2_trans_end(sdp);
return error;
}
/**
* do_shrink - make a file smaller
* @inode: the inode
* @oldsize: the current inode size
* @newsize: the size to make the file
*
* Called with an exclusive lock on @inode. The @size must
* be equal to or smaller than the current inode size.
*
* Returns: errno
*/
static int do_shrink(struct inode *inode, u64 oldsize, u64 newsize)
{
struct gfs2_inode *ip = GFS2_I(inode);
int error;
error = trunc_start(inode, oldsize, newsize);
if (error < 0)
return error;
if (gfs2_is_stuffed(ip))
return 0;
error = trunc_dealloc(ip, newsize);
if (error == 0)
error = trunc_end(ip);
return error;
}
void gfs2_trim_blocks(struct inode *inode)
{
u64 size = inode->i_size;
int ret;
ret = do_shrink(inode, size, size);
WARN_ON(ret != 0);
}
/**
* do_grow - Touch and update inode size
* @inode: The inode
* @size: The new size
*
* This function updates the timestamps on the inode and
* may also increase the size of the inode. This function
* must not be called with @size any smaller than the current
* inode size.
*
* Although it is not strictly required to unstuff files here,
* earlier versions of GFS2 have a bug in the stuffed file reading
* code which will result in a buffer overrun if the size is larger
* than the max stuffed file size. In order to prevent this from
* occurring, such files are unstuffed, but in other cases we can
* just update the inode size directly.
*
* Returns: 0 on success, or -ve on error
*/
static int do_grow(struct inode *inode, u64 size)
{
struct gfs2_inode *ip = GFS2_I(inode);
struct gfs2_sbd *sdp = GFS2_SB(inode);
struct gfs2_alloc_parms ap = { .target = 1, };
struct buffer_head *dibh;
int error;
int unstuff = 0;
if (gfs2_is_stuffed(ip) &&
(size > (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)))) {
error = gfs2_quota_lock_check(ip, &ap);
if (error)
return error;
error = gfs2_inplace_reserve(ip, &ap);
if (error)
goto do_grow_qunlock;
unstuff = 1;
}
error = gfs2_trans_begin(sdp, RES_DINODE + RES_STATFS + RES_RG_BIT +
(sdp->sd_args.ar_quota == GFS2_QUOTA_OFF ?
0 : RES_QUOTA), 0);
if (error)
goto do_grow_release;
if (unstuff) {
error = gfs2_unstuff_dinode(ip, NULL);
if (error)
goto do_end_trans;
}
error = gfs2_meta_inode_buffer(ip, &dibh);
if (error)
goto do_end_trans;
i_size_write(inode, size);
ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
gfs2_trans_add_meta(ip->i_gl, dibh);
gfs2_dinode_out(ip, dibh->b_data);
brelse(dibh);
do_end_trans:
gfs2_trans_end(sdp);
do_grow_release:
if (unstuff) {
gfs2_inplace_release(ip);
do_grow_qunlock:
gfs2_quota_unlock(ip);
}
return error;
}
/**
* gfs2_setattr_size - make a file a given size
* @inode: the inode
* @newsize: the size to make the file
*
* The file size can grow, shrink, or stay the same size. This
* is called holding i_mutex and an exclusive glock on the inode
* in question.
*
* Returns: errno
*/
int gfs2_setattr_size(struct inode *inode, u64 newsize)
{
struct gfs2_inode *ip = GFS2_I(inode);
int ret;
u64 oldsize;
BUG_ON(!S_ISREG(inode->i_mode));
ret = inode_newsize_ok(inode, newsize);
if (ret)
return ret;
inode_dio_wait(inode);
ret = gfs2_rsqa_alloc(ip);
if (ret)
goto out;
oldsize = inode->i_size;
if (newsize >= oldsize) {
ret = do_grow(inode, newsize);
goto out;
}
ret = do_shrink(inode, oldsize, newsize);
out:
gfs2_rsqa_delete(ip, NULL);
return ret;
}
int gfs2_truncatei_resume(struct gfs2_inode *ip)
{
int error;
error = trunc_dealloc(ip, i_size_read(&ip->i_inode));
if (!error)
error = trunc_end(ip);
return error;
}
int gfs2_file_dealloc(struct gfs2_inode *ip)
{
return trunc_dealloc(ip, 0);
}
/**
* gfs2_free_journal_extents - Free cached journal bmap info
* @jd: The journal
*
*/
void gfs2_free_journal_extents(struct gfs2_jdesc *jd)
{
struct gfs2_journal_extent *jext;
while(!list_empty(&jd->extent_list)) {
jext = list_entry(jd->extent_list.next, struct gfs2_journal_extent, list);
list_del(&jext->list);
kfree(jext);
}
}
/**
* gfs2_add_jextent - Add or merge a new extent to extent cache
* @jd: The journal descriptor
* @lblock: The logical block at start of new extent
* @dblock: The physical block at start of new extent
* @blocks: Size of extent in fs blocks
*
* Returns: 0 on success or -ENOMEM
*/
static int gfs2_add_jextent(struct gfs2_jdesc *jd, u64 lblock, u64 dblock, u64 blocks)
{
struct gfs2_journal_extent *jext;
if (!list_empty(&jd->extent_list)) {
jext = list_entry(jd->extent_list.prev, struct gfs2_journal_extent, list);
if ((jext->dblock + jext->blocks) == dblock) {
jext->blocks += blocks;
return 0;
}
}
jext = kzalloc(sizeof(struct gfs2_journal_extent), GFP_NOFS);
if (jext == NULL)
return -ENOMEM;
jext->dblock = dblock;
jext->lblock = lblock;
jext->blocks = blocks;
list_add_tail(&jext->list, &jd->extent_list);
jd->nr_extents++;
return 0;
}
/**
* gfs2_map_journal_extents - Cache journal bmap info
* @sdp: The super block
* @jd: The journal to map
*
* Create a reusable "extent" mapping from all logical
* blocks to all physical blocks for the given journal. This will save
* us time when writing journal blocks. Most journals will have only one
* extent that maps all their logical blocks. That's because gfs2.mkfs
* arranges the journal blocks sequentially to maximize performance.
* So the extent would map the first block for the entire file length.
* However, gfs2_jadd can happen while file activity is happening, so
* those journals may not be sequential. Less likely is the case where
* the users created their own journals by mounting the metafs and
* laying it out. But it's still possible. These journals might have
* several extents.
*
* Returns: 0 on success, or error on failure
*/
int gfs2_map_journal_extents(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd)
{
u64 lblock = 0;
u64 lblock_stop;
struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
struct buffer_head bh;
unsigned int shift = sdp->sd_sb.sb_bsize_shift;
u64 size;
int rc;
lblock_stop = i_size_read(jd->jd_inode) >> shift;
size = (lblock_stop - lblock) << shift;
jd->nr_extents = 0;
WARN_ON(!list_empty(&jd->extent_list));
do {
bh.b_state = 0;
bh.b_blocknr = 0;
bh.b_size = size;
rc = gfs2_block_map(jd->jd_inode, lblock, &bh, 0);
if (rc || !buffer_mapped(&bh))
goto fail;
rc = gfs2_add_jextent(jd, lblock, bh.b_blocknr, bh.b_size >> shift);
if (rc)
goto fail;
size -= bh.b_size;
lblock += (bh.b_size >> ip->i_inode.i_blkbits);
} while(size > 0);
fs_info(sdp, "journal %d mapped with %u extents\n", jd->jd_jid,
jd->nr_extents);
return 0;
fail:
fs_warn(sdp, "error %d mapping journal %u at offset %llu (extent %u)\n",
rc, jd->jd_jid,
(unsigned long long)(i_size_read(jd->jd_inode) - size),
jd->nr_extents);
fs_warn(sdp, "bmap=%d lblock=%llu block=%llu, state=0x%08lx, size=%llu\n",
rc, (unsigned long long)lblock, (unsigned long long)bh.b_blocknr,
bh.b_state, (unsigned long long)bh.b_size);
gfs2_free_journal_extents(jd);
return rc;
}
/**
* gfs2_write_alloc_required - figure out if a write will require an allocation
* @ip: the file being written to
* @offset: the offset to write to
* @len: the number of bytes being written
*
* Returns: 1 if an alloc is required, 0 otherwise
*/
int gfs2_write_alloc_required(struct gfs2_inode *ip, u64 offset,
unsigned int len)
{
struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
struct buffer_head bh;
unsigned int shift;
u64 lblock, lblock_stop, size;
u64 end_of_file;
if (!len)
return 0;
if (gfs2_is_stuffed(ip)) {
if (offset + len >
sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode))
return 1;
return 0;
}
shift = sdp->sd_sb.sb_bsize_shift;
BUG_ON(gfs2_is_dir(ip));
end_of_file = (i_size_read(&ip->i_inode) + sdp->sd_sb.sb_bsize - 1) >> shift;
lblock = offset >> shift;
lblock_stop = (offset + len + sdp->sd_sb.sb_bsize - 1) >> shift;
if (lblock_stop > end_of_file)
return 1;
size = (lblock_stop - lblock) << shift;
do {
bh.b_state = 0;
bh.b_size = size;
gfs2_block_map(&ip->i_inode, lblock, &bh, 0);
if (!buffer_mapped(&bh))
return 1;
size -= bh.b_size;
lblock += (bh.b_size >> ip->i_inode.i_blkbits);
} while(size > 0);
return 0;
}