d/freebsd-dev
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

vfs_vnops.c: Fix blksize for ZFS

Browse Source 
Since ZFS reports _PC_MIN_HOLE_SIZE as 512 (although it
appears that an unwritten region must be at least f_iosize
to remain unallocated), vn_generic_copy_file_range()
uses 4096 for the copy blksize for ZFS, reulting in slow copies.

For most other file systems, _PC_MIN_HOLE_SIZE and f_iosize
are the same value, so this patch modifies the code to
use f_iosize for most cases.  It also documents in comments
why the blksize is being set a certain way, so that the code
does not appear to be doing "magic math".

Reported by:	allanjude
Reviewed by:	allanjude, asomers
MFC after:	2 weeks
Differential Revision:	https://reviews.freebsd.org/D37076
This commit is contained in:
Rick Macklem 2022-11-16 17:37:22 -08:00
parent 5d42ef55de
commit 4ee16246f9
1 changed files with 29 additions and 22 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

51
sys/kern/vfs_vnops.c
View File

@ -3266,7 +3266,7 @@ vn_generic_copy_file_range(struct vnode *invp, off_t *inoffp,
int error, interrupted;
bool cantseek, readzeros, eof, lastblock, holetoeof;
ssize_t aresid, r = 0;
size_t copylen, len, rem, savlen;
size_t copylen, len, savlen;
char *dat;
long holein, holeout;
struct timespec curts, endts;
@ -3338,31 +3338,38 @@ vn_generic_copy_file_range(struct vnode *invp, off_t *inoffp,
if (error != 0)
goto out;
/*
* Set the blksize to the larger of the hole sizes for invp and outvp.
* If hole sizes aren't available, set the blksize to the larger
* f_iosize of invp and outvp.
* This code expects the hole sizes and f_iosizes to be powers of 2.
* This value is clipped at 4Kbytes and 1Mbyte.
*/
blksize = MAX(holein, holeout);
/* Clip len to end at an exact multiple of hole size. */
if (blksize > 1) {
rem = *inoffp % blksize;
if (rem > 0)
rem = blksize - rem;
if (len > rem && len - rem > blksize)
len = savlen = rounddown(len - rem, blksize) + rem;
}
if (blksize <= 1)
if (holein == 0 && holeout > 0) {
/*
* For this special case, the input data will be scanned
* for blocks of all 0 bytes. For these blocks, the
* write can be skipped for the output file to create
* an unallocated region.
* Therefore, use the appropriate size for the output file.
*/
blksize = holeout;
if (blksize <= 512) {
/*
* Use f_iosize, since ZFS reports a _PC_MIN_HOLE_SIZE
* of 512, although it actually only creates
* unallocated regions for blocks >= f_iosize.
*/
blksize = outvp->v_mount->mnt_stat.f_iosize;
}
} else {
/*
* Use the larger of the two f_iosize values. If they are
* not the same size, one will normally be an exact multiple of
* the other, since they are both likely to be a power of 2.
*/
blksize = MAX(invp->v_mount->mnt_stat.f_iosize,
outvp->v_mount->mnt_stat.f_iosize);
}
/* Clip to sane limits. */
if (blksize < 4096)
blksize = 4096;
else if (blksize > 1024 * 1024)
blksize = 1024 * 1024;
else if (blksize > maxphys)
blksize = maxphys;
dat = malloc(blksize, M_TEMP, M_WAITOK);
/*