[PATCH v2] erofs-utils: support per-inode compress pcluster

[Gao Xiang]

On Wed, Aug 25, 2021 at 09:17:38AM +0800, Gao Xiang wrote:
> On Wed, Aug 18, 2021 at 12:27:15PM +0800, Huang Jianan via Linux-erofs wrote:
> > Add an option to configure per-inode compression strategy. Each line
> > of the file should be in the following form:
> > 
> > <Regular-expression> <pcluster-in-bytes>
> > 
> > When pcluster is 0, it means that the file shouldn't be compressed.
> > 
> > Signed-off-by: Huang Jianan <huangjianan at oppo.com>
> 
> Sorry for the delay. Due to busy work, I will look into the details
> this weekend. Some comments in advance.
> 
> > ---
> > changes since v1:
> >  - rename c_pclusterblks to c_physical_clusterblks and place it in union
> >  - change cfg.c_physical_clusterblks > 1 to erofs_sb_has_big_pcluster()
> >    since it's per-inode compression strategy
> > 
> >  include/erofs/compress_rule.h |  25 ++++++++
> 
> How about calling "compress_hints"? Does it sound better?
> 
> >  include/erofs/config.h        |   1 +
> 
> ...
> 
> > index 0000000..497d662
> > --- /dev/null
> > +++ b/lib/compress_rule.c
> > @@ -0,0 +1,106 @@
> > +// SPDX-License-Identifier: GPL-2.0+
> > +/*
> > + * erofs-utils/lib/compress_rule.c
> > + *
> > + * Copyright (C), 2008-2021, OPPO Mobile Comm Corp., Ltd.
> > + * Created by Huang Jianan <huangjianan at oppo.com>
> > + */
> > +#include <string.h>
> > +#include <stdlib.h>
> > +#include "erofs/err.h"
> > +#include "erofs/list.h"
> > +#include "erofs/print.h"
> > +#include "erofs/compress_rule.h"
> > +
> > +static LIST_HEAD(compress_rule_head);
> > +
> > +static void dump_regerror(int errcode, const char *s, const regex_t *preg)
> > +{
> > +	char str[512];
> > +
> > +	regerror(errcode, preg, str, sizeof(str));
> > +	erofs_err("invalid regex %s (%s)\n", s, str);
> > +}
> > +
> > +static int erofs_insert_compress_rule(const char *s, unsigned int blks)
> > +{
> > +	struct erofs_compress_rule *r;
> > +	int ret = 0;
> > +
> > +	r = malloc(sizeof(struct erofs_compress_rule));
> > +	if (!r)
> > +		return -ENOMEM;
> > +
> > +	ret = regcomp(&r->reg, s, REG_EXTENDED|REG_NOSUB);
> > +	if (ret) {
> > +		dump_regerror(ret, s, &r->reg);
> > +		goto err;
> > +	}
> > +	r->c_physical_clusterblks = blks;
> > +
> > +	list_add_tail(&r->list, &compress_rule_head);
> > +	erofs_info("insert compress rule %s: %u", s, blks);
> > +	return ret;
> > +
> > +err:
> > +	free(r);
> > +	return ret;
> > +}
> > +
> > +unsigned int erofs_parse_pclusterblks(struct erofs_inode *inode)
> > +{
> > +	const char *s;
> > +	struct erofs_compress_rule *r;
> > +
> > +	if (inode->c_physical_clusterblks)
> > +		return inode->c_physical_clusterblks;
> > +
> > +	s = erofs_fspath(inode->i_srcpath);
> > +
> > +	list_for_each_entry(r, &compress_rule_head, list) {
> > +		int ret = regexec(&r->reg, s, (size_t)0, NULL, 0);
> > +
> > +		if (!ret) {
> > +			inode->c_physical_clusterblks = r->c_physical_clusterblks;
> > +			return r->c_physical_clusterblks;
> > +		}
> > +		if (ret > REG_NOMATCH)
> > +			dump_regerror(ret, s, &r->reg);
> > +	}
> > +
> > +	inode->c_physical_clusterblks = cfg.c_physical_clusterblks;
> > +	return cfg.c_physical_clusterblks;
> > +}
> > +
> > +int erofs_load_compress_rule()
> > +{
> > +	char buf[PATH_MAX + 100];
> > +	FILE* f;
> > +	int ret = 0;
> > +
> > +	if (!cfg.compress_rule_file)
> > +		return 0;
> > +
> > +	f = fopen(cfg.compress_rule_file, "r");
> > +	if (f == NULL)
> > +		return -errno;
> > +
> > +	while (fgets(buf, sizeof(buf), f)) {
> > +		char* line = buf;
> > +		char* s;
> > +		unsigned int blks;
> > +
> > +		s = strtok(line, " ");
> > +		blks = atoi(strtok(NULL, " "));
> > +		if (blks % EROFS_BLKSIZ) {
> 
> We might need to guarantee these are power of 2.

Oh, never mind. It's not necessary to leave pcluster power of 2.
(I need some wake-up coffee...)

> Also, how about just printing out warning message but using default "-C"
> value instead?
> 
> > +			erofs_err("invalid physical clustersize %u", blks);
> > +			ret = -EINVAL;
> > +			goto out;
> > +		}
> > +		erofs_insert_compress_rule(s, blks / EROFS_BLKSIZ);
> > +	}
> > +
> > +out:
> > +	fclose(f);
> > +	return ret;
> > +}
> 
> ...
> 
> > diff --git a/mkfs/main.c b/mkfs/main.c
> > index 10fe14d..467e875 100644
> > --- a/mkfs/main.c
> > +++ b/mkfs/main.c
> > @@ -23,6 +23,7 @@
> >  #include "erofs/xattr.h"
> >  #include "erofs/exclude.h"
> >  #include "erofs/block_list.h"
> > +#include "erofs/compress_rule.h"
> >  
> >  #ifdef HAVE_LIBUUID
> >  #include <uuid.h>
> > @@ -44,11 +45,12 @@ static struct option long_options[] = {
> >  	{"random-pclusterblks", no_argument, NULL, 8},
> >  #endif
> >  	{"max-extent-bytes", required_argument, NULL, 9},
> > +	{"compress-rule", required_argument, NULL, 10},
> >  #ifdef WITH_ANDROID
> > -	{"mount-point", required_argument, NULL, 10},
> > -	{"product-out", required_argument, NULL, 11},
> > -	{"fs-config-file", required_argument, NULL, 12},
> > -	{"block-list-file", required_argument, NULL, 13},
> > +	{"mount-point", required_argument, NULL, 20},
> > +	{"product-out", required_argument, NULL, 21},
> > +	{"fs-config-file", required_argument, NULL, 22},
> > +	{"block-list-file", required_argument, NULL, 23},
> 
> I think we might clean up these first with a separated patch.
> Use >= 256 for all of them instead.
> 
> Thanks,
> Gao Xiang