[lxc-devel] [PATCH lxcfs 4/5] cgfs: fix dorealloc's batch allocation

[Wolfgang Bumiller]

> On January 8, 2016 at 9:50 AM Wolfgang Bumiller <w.bumiller at proxmox.com>
> wrote:
> 
> 
> 
> > On January 7, 2016 at 8:20 PM Serge Hallyn <serge.hallyn at ubuntu.com> wrote:
> > Quoting Wolfgang Bumiller (w.bumiller at proxmox.com):
> > > > On January 7, 2016 at 7:42 PM Serge Hallyn <serge.hallyn at ubuntu.com>
> > > > wrote:
> > > > Quoting Wolfgang Bumiller (w.bumiller at proxmox.com):
> > > > > -	if (newlen % BATCH_SIZE <= oldlen % BATCH_SIZE)
> > > > > +	if (newlen <= oldlen)
> > > > 
> > > > This will result in extra reallocs (though innocuous).
> > > > Is there any reason not to just do
> > > > 
> > > > 	if (newlen / BATCH_SIZE <= oldlen / BATCH_SIZE)
> > > 
> > > Ah yes I missed that the old length is a string length not the
> > > capacity, so yes, the division makes sense. Just not the modulus.
> > > 
> > 
> > Ok, thanks - per your patch I think the following makes a nice cleanup:
> >
> > From a72193ff0dba319c3039372bc038708a75d0582f Mon Sep 17 00:00:00 2001
> > From: Serge Hallyn <serge.hallyn at ubuntu.com>
> > Date: Thu, 7 Jan 2016 11:17:17 -0800
> > Subject: [PATCH 1/1] dorealloc: avoid extra reallocs
> > 
> > The original check was very wrong, using % instead of /.  However
> > the length we track is the actual used length, not the allocated
> > length, which is always (len / BATCH_SIZE) + 1.  We don't want
> > to realloc when newlen is between oldlen and (oldlen / BATCH_SIZE) + 1)
> > 
> > Signed-off-by: Serge Hallyn <serge.hallyn at ubuntu.com>
> 
> Yes this seems fine and easier to read. Although the commit message
> mixes length and batch count again ;-).
> 
> Do you care about a Reviewed-by line? If so
> 
> Reviewed-by: Wolfgang Bumiller <w.bumiller at proxmox.com>
>
> > ---
> >  cgfs.c | 12 +++++++-----
> >  1 file changed, 7 insertions(+), 5 deletions(-)
> > 
> > diff --git a/cgfs.c b/cgfs.c
> > index ec0f630..f37bc2b 100644
> > --- a/cgfs.c
> > +++ b/cgfs.c
> > @@ -74,18 +74,20 @@ static inline void drop_trailing_newlines(char *s)
> >  #define BATCH_SIZE 50
> >  static void dorealloc(char **mem, size_t oldlen, size_t newlen)
> >  {
> > -	int batches;
> > -	if (newlen <= oldlen)
> > +	int newbatches = (newlen / BATCH_SIZE) + 1;
> > +	int oldbatches = (olden / BATCH_SIZE) + 1;
> > +
> > +	if (newbatches <= oldbatches)

Ah actually... if oldlen==0 (and newlen is < 50, as in most of the time
with the first line) then newbatches == oldbatches so the <= is true
and it won't do the first malloc().

realloc() also deals with NULL as first parameter so perhaps a
single conditional could do (unless that part of realloc() is not
standard C behavior or different on android or something?).

Like so:

>From 0d5dcaa2ad1e39b8f5c65eb5182ef7acd409c812 Mon Sep 17 00:00:00 2001
From: Wolfgang Bumiller <w.bumiller at proxmox.com>
Date: Fri, 8 Jan 2016 11:09:57 +0100
Subject: [PATCH lxcfs] cgfs: make dorealloc allocate the first batch, too

With a short first line the case can be
 *mem = NULL
 oldlen = 0
 newlen = 5 (anything < 50)
making newbatches == oldbatches == 1 causing the
 (newbatches <= oldbatches)
condition to be true.

Let realloc() handle *mem==NULL and use
(!*mem || newbatches > oldbatches) as the only condition.

Signed-off-by: Wolfgang Bumiller <w.bumiller at proxmox.com>
---
 cgfs.c | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)

diff --git a/cgfs.c b/cgfs.c
index 1f31ed1..f667456 100644
--- a/cgfs.c
+++ b/cgfs.c
@@ -77,14 +77,11 @@ static void dorealloc(char **mem, size_t oldlen, size_t
newlen)
 	int newbatches = (newlen / BATCH_SIZE) + 1;
 	int oldbatches = (oldlen / BATCH_SIZE) + 1;
 
-	if (newbatches <= oldbatches)
-		return;
-
 	if (!*mem) {
 		do {
 			*mem = malloc(newbatches * BATCH_SIZE);
 		} while (!*mem);
-	} else {
+	} else if (newbatches > oldbatches) {
 		char *tmp;
 		do {
 			tmp = realloc(*mem, newbatches * BATCH_SIZE);
-- 
2.1.4