Revision 2a18da7a9c7886f1c7307f8d3f23f24318583f03 authored by George Spelvin on 23 May 2016, 11:43:58 UTC, committed by George Spelvin on 28 May 2016, 19:45:29 UTC
Patch 0fed3ac866 improved the hash mixing, but the function is slower than necessary; there's a 7-instruction dependency chain (10 on x86) each loop iteration. Word-at-a-time access is a very tight loop (which is good, because link_path_walk() is one of the hottest code paths in the entire kernel), and the hash mixing function must not have a longer latency to avoid slowing it down. There do not appear to be any published fast hash functions that: 1) Operate on the input a word at a time, and 2) Don't need to know the length of the input beforehand, and 3) Have a single iterated mixing function, not needing conditional branches or unrolling to distinguish different loop iterations. One of the algorithms which comes closest is Yann Collet's xxHash, but that's two dependent multiplies per word, which is too much. The key insights in this design are: 1) Barring expensive ops like multiplies, to diffuse one input bit across 64 bits of hash state takes at least log2(64) = 6 sequentially dependent instructions. That is more cycles than we'd like. 2) An operation like "hash ^= hash << 13" requires a second temporary register anyway, and on a 2-operand machine like x86, it's three instructions. 3) A better use of a second register is to hold a two-word hash state. With careful design, no temporaries are needed at all, so it doesn't increase register pressure. And this gets rid of register copying on 2-operand machines, so the code is smaller and faster. 4) Using two words of state weakens the requirement for one-round mixing; we now have two rounds of mixing before cancellation is possible. 5) A two-word hash state also allows operations on both halves to be done in parallel, so on a superscalar processor we get more mixing in fewer cycles. I ended up using a mixing function inspired by the ChaCha and Speck round functions. It is 6 simple instructions and 3 cycles per iteration (assuming multiply by 9 can be done by an "lea" instruction): x ^= *input++; y ^= x; x = ROL(x, K1); x += y; y = ROL(y, K2); y *= 9; Not only is this reversible, two consecutive rounds are reversible: if you are given the initial and final states, but not the intermediate state, it is possible to compute both input words. This means that at least 3 words of input are required to create a collision. (It also has the property, used by hash_name() to avoid a branch, that it hashes all-zero to all-zero.) The rotate constants K1 and K2 were found by experiment. The search took a sample of random initial states (I used 1023) and considered the effect of flipping each of the 64 input bits on each of the 128 output bits two rounds later. Each of the 8192 pairs can be considered a biased coin, and adding up the Shannon entropy of all of them produces a score. The best-scoring shifts also did well in other tests (flipping bits in y, trying 3 or 4 rounds of mixing, flipping all 64*63/2 pairs of input bits), so the choice was made with the additional constraint that the sum of the shifts is odd and not too close to the word size. The final state is then folded into a 32-bit hash value by a less carefully optimized multiply-based scheme. This also has to be fast, as pathname components tend to be short (the most common case is one iteration!), but there's some room for latency, as there is a fair bit of intervening logic before the hash value is used for anything. (Performance verified with "bonnie++ -s 0 -n 1536:-2" on tmpfs. I need a better benchmark; the numbers seem to show a slight dip in performance between 4.6.0 and this patch, but they're too noisy to quote.) Special thanks to Bruce fields for diligent testing which uncovered a nasty fencepost error in an earlier version of this patch. [checkpatch.pl formatting complaints noted and respectfully disagreed with.] Signed-off-by: George Spelvin <linux@sciencehorizons.net> Tested-by: J. Bruce Fields <bfields@redhat.com>
1 parent ef703f4
sync.c
/*
* High-level sync()-related operations
*/
#include <linux/kernel.h>
#include <linux/file.h>
#include <linux/fs.h>
#include <linux/slab.h>
#include <linux/export.h>
#include <linux/namei.h>
#include <linux/sched.h>
#include <linux/writeback.h>
#include <linux/syscalls.h>
#include <linux/linkage.h>
#include <linux/pagemap.h>
#include <linux/quotaops.h>
#include <linux/backing-dev.h>
#include "internal.h"
#define VALID_FLAGS (SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE| \
SYNC_FILE_RANGE_WAIT_AFTER)
/*
* Do the filesystem syncing work. For simple filesystems
* writeback_inodes_sb(sb) just dirties buffers with inodes so we have to
* submit IO for these buffers via __sync_blockdev(). This also speeds up the
* wait == 1 case since in that case write_inode() functions do
* sync_dirty_buffer() and thus effectively write one block at a time.
*/
static int __sync_filesystem(struct super_block *sb, int wait)
{
if (wait)
sync_inodes_sb(sb);
else
writeback_inodes_sb(sb, WB_REASON_SYNC);
if (sb->s_op->sync_fs)
sb->s_op->sync_fs(sb, wait);
return __sync_blockdev(sb->s_bdev, wait);
}
/*
* Write out and wait upon all dirty data associated with this
* superblock. Filesystem data as well as the underlying block
* device. Takes the superblock lock.
*/
int sync_filesystem(struct super_block *sb)
{
int ret;
/*
* We need to be protected against the filesystem going from
* r/o to r/w or vice versa.
*/
WARN_ON(!rwsem_is_locked(&sb->s_umount));
/*
* No point in syncing out anything if the filesystem is read-only.
*/
if (sb->s_flags & MS_RDONLY)
return 0;
ret = __sync_filesystem(sb, 0);
if (ret < 0)
return ret;
return __sync_filesystem(sb, 1);
}
EXPORT_SYMBOL(sync_filesystem);
static void sync_inodes_one_sb(struct super_block *sb, void *arg)
{
if (!(sb->s_flags & MS_RDONLY))
sync_inodes_sb(sb);
}
static void sync_fs_one_sb(struct super_block *sb, void *arg)
{
if (!(sb->s_flags & MS_RDONLY) && sb->s_op->sync_fs)
sb->s_op->sync_fs(sb, *(int *)arg);
}
static void fdatawrite_one_bdev(struct block_device *bdev, void *arg)
{
filemap_fdatawrite(bdev->bd_inode->i_mapping);
}
static void fdatawait_one_bdev(struct block_device *bdev, void *arg)
{
/*
* We keep the error status of individual mapping so that
* applications can catch the writeback error using fsync(2).
* See filemap_fdatawait_keep_errors() for details.
*/
filemap_fdatawait_keep_errors(bdev->bd_inode->i_mapping);
}
/*
* Sync everything. We start by waking flusher threads so that most of
* writeback runs on all devices in parallel. Then we sync all inodes reliably
* which effectively also waits for all flusher threads to finish doing
* writeback. At this point all data is on disk so metadata should be stable
* and we tell filesystems to sync their metadata via ->sync_fs() calls.
* Finally, we writeout all block devices because some filesystems (e.g. ext2)
* just write metadata (such as inodes or bitmaps) to block device page cache
* and do not sync it on their own in ->sync_fs().
*/
SYSCALL_DEFINE0(sync)
{
int nowait = 0, wait = 1;
wakeup_flusher_threads(0, WB_REASON_SYNC);
iterate_supers(sync_inodes_one_sb, NULL);
iterate_supers(sync_fs_one_sb, &nowait);
iterate_supers(sync_fs_one_sb, &wait);
iterate_bdevs(fdatawrite_one_bdev, NULL);
iterate_bdevs(fdatawait_one_bdev, NULL);
if (unlikely(laptop_mode))
laptop_sync_completion();
return 0;
}
static void do_sync_work(struct work_struct *work)
{
int nowait = 0;
/*
* Sync twice to reduce the possibility we skipped some inodes / pages
* because they were temporarily locked
*/
iterate_supers(sync_inodes_one_sb, &nowait);
iterate_supers(sync_fs_one_sb, &nowait);
iterate_bdevs(fdatawrite_one_bdev, NULL);
iterate_supers(sync_inodes_one_sb, &nowait);
iterate_supers(sync_fs_one_sb, &nowait);
iterate_bdevs(fdatawrite_one_bdev, NULL);
printk("Emergency Sync complete\n");
kfree(work);
}
void emergency_sync(void)
{
struct work_struct *work;
work = kmalloc(sizeof(*work), GFP_ATOMIC);
if (work) {
INIT_WORK(work, do_sync_work);
schedule_work(work);
}
}
/*
* sync a single super
*/
SYSCALL_DEFINE1(syncfs, int, fd)
{
struct fd f = fdget(fd);
struct super_block *sb;
int ret;
if (!f.file)
return -EBADF;
sb = f.file->f_path.dentry->d_sb;
down_read(&sb->s_umount);
ret = sync_filesystem(sb);
up_read(&sb->s_umount);
fdput(f);
return ret;
}
/**
* vfs_fsync_range - helper to sync a range of data & metadata to disk
* @file: file to sync
* @start: offset in bytes of the beginning of data range to sync
* @end: offset in bytes of the end of data range (inclusive)
* @datasync: perform only datasync
*
* Write back data in range @start..@end and metadata for @file to disk. If
* @datasync is set only metadata needed to access modified file data is
* written.
*/
int vfs_fsync_range(struct file *file, loff_t start, loff_t end, int datasync)
{
struct inode *inode = file->f_mapping->host;
if (!file->f_op->fsync)
return -EINVAL;
if (!datasync && (inode->i_state & I_DIRTY_TIME)) {
spin_lock(&inode->i_lock);
inode->i_state &= ~I_DIRTY_TIME;
spin_unlock(&inode->i_lock);
mark_inode_dirty_sync(inode);
}
return file->f_op->fsync(file, start, end, datasync);
}
EXPORT_SYMBOL(vfs_fsync_range);
/**
* vfs_fsync - perform a fsync or fdatasync on a file
* @file: file to sync
* @datasync: only perform a fdatasync operation
*
* Write back data and metadata for @file to disk. If @datasync is
* set only metadata needed to access modified file data is written.
*/
int vfs_fsync(struct file *file, int datasync)
{
return vfs_fsync_range(file, 0, LLONG_MAX, datasync);
}
EXPORT_SYMBOL(vfs_fsync);
static int do_fsync(unsigned int fd, int datasync)
{
struct fd f = fdget(fd);
int ret = -EBADF;
if (f.file) {
ret = vfs_fsync(f.file, datasync);
fdput(f);
}
return ret;
}
SYSCALL_DEFINE1(fsync, unsigned int, fd)
{
return do_fsync(fd, 0);
}
SYSCALL_DEFINE1(fdatasync, unsigned int, fd)
{
return do_fsync(fd, 1);
}
/*
* sys_sync_file_range() permits finely controlled syncing over a segment of
* a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is
* zero then sys_sync_file_range() will operate from offset out to EOF.
*
* The flag bits are:
*
* SYNC_FILE_RANGE_WAIT_BEFORE: wait upon writeout of all pages in the range
* before performing the write.
*
* SYNC_FILE_RANGE_WRITE: initiate writeout of all those dirty pages in the
* range which are not presently under writeback. Note that this may block for
* significant periods due to exhaustion of disk request structures.
*
* SYNC_FILE_RANGE_WAIT_AFTER: wait upon writeout of all pages in the range
* after performing the write.
*
* Useful combinations of the flag bits are:
*
* SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages
* in the range which were dirty on entry to sys_sync_file_range() are placed
* under writeout. This is a start-write-for-data-integrity operation.
*
* SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
* are not presently under writeout. This is an asynchronous flush-to-disk
* operation. Not suitable for data integrity operations.
*
* SYNC_FILE_RANGE_WAIT_BEFORE (or SYNC_FILE_RANGE_WAIT_AFTER): wait for
* completion of writeout of all pages in the range. This will be used after an
* earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait
* for that operation to complete and to return the result.
*
* SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER:
* a traditional sync() operation. This is a write-for-data-integrity operation
* which will ensure that all pages in the range which were dirty on entry to
* sys_sync_file_range() are committed to disk.
*
*
* SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
* I/O errors or ENOSPC conditions and will return those to the caller, after
* clearing the EIO and ENOSPC flags in the address_space.
*
* It should be noted that none of these operations write out the file's
* metadata. So unless the application is strictly performing overwrites of
* already-instantiated disk blocks, there are no guarantees here that the data
* will be available after a crash.
*/
SYSCALL_DEFINE4(sync_file_range, int, fd, loff_t, offset, loff_t, nbytes,
unsigned int, flags)
{
int ret;
struct fd f;
struct address_space *mapping;
loff_t endbyte; /* inclusive */
umode_t i_mode;
ret = -EINVAL;
if (flags & ~VALID_FLAGS)
goto out;
endbyte = offset + nbytes;
if ((s64)offset < 0)
goto out;
if ((s64)endbyte < 0)
goto out;
if (endbyte < offset)
goto out;
if (sizeof(pgoff_t) == 4) {
if (offset >= (0x100000000ULL << PAGE_SHIFT)) {
/*
* The range starts outside a 32 bit machine's
* pagecache addressing capabilities. Let it "succeed"
*/
ret = 0;
goto out;
}
if (endbyte >= (0x100000000ULL << PAGE_SHIFT)) {
/*
* Out to EOF
*/
nbytes = 0;
}
}
if (nbytes == 0)
endbyte = LLONG_MAX;
else
endbyte--; /* inclusive */
ret = -EBADF;
f = fdget(fd);
if (!f.file)
goto out;
i_mode = file_inode(f.file)->i_mode;
ret = -ESPIPE;
if (!S_ISREG(i_mode) && !S_ISBLK(i_mode) && !S_ISDIR(i_mode) &&
!S_ISLNK(i_mode))
goto out_put;
mapping = f.file->f_mapping;
if (!mapping) {
ret = -EINVAL;
goto out_put;
}
ret = 0;
if (flags & SYNC_FILE_RANGE_WAIT_BEFORE) {
ret = filemap_fdatawait_range(mapping, offset, endbyte);
if (ret < 0)
goto out_put;
}
if (flags & SYNC_FILE_RANGE_WRITE) {
ret = __filemap_fdatawrite_range(mapping, offset, endbyte,
WB_SYNC_NONE);
if (ret < 0)
goto out_put;
}
if (flags & SYNC_FILE_RANGE_WAIT_AFTER)
ret = filemap_fdatawait_range(mapping, offset, endbyte);
out_put:
fdput(f);
out:
return ret;
}
/* It would be nice if people remember that not all the world's an i386
when they introduce new system calls */
SYSCALL_DEFINE4(sync_file_range2, int, fd, unsigned int, flags,
loff_t, offset, loff_t, nbytes)
{
return sys_sync_file_range(fd, offset, nbytes, flags);
}
![swh spinner](/static/img/swh-spinner.gif)
Computing file changes ...