From: David Howells <dhowells@redhat.com>
Date: Fri, 14 Jan 2022 14:13:59 +0000 (+0000)
Subject: cachefiles: Calculate the blockshift in terms of bytes, not pages
X-Git-Tag: v6.1-rc5~2144^2~7
X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=5638b067d370583c6c455f019129ce33340b4142;p=platform%2Fkernel%2Flinux-starfive.git

cachefiles: Calculate the blockshift in terms of bytes, not pages

Cachefiles keeps track of how much space is available on the backing
filesystem and refuses new writes permission to start if there isn't enough
(we especially don't want ENOSPC happening).  It also tracks the amount of
data pending in DIO writes (cache->b_writing) and reduces the amount of
free space available by this amount before deciding if it can set up a new
write.

However, the old fscache I/O API was very much page-granularity dependent
and, as such, cachefiles's cache->bshift was meant to be a multiplier to
get from PAGE_SIZE to block size (ie. a blocksize of 512 would give a shift
of 3 for a 4KiB page) - and this was incorrectly being used to turn the
number of bytes in a DIO write into a number of blocks, leading to a
massive over estimation of the amount of data in flight.

Fix this by changing cache->bshift to be a multiplier from bytes to
blocksize and deal with quantities of blocks, not quantities of pages.

Fix also the rounding in the calculation in cachefiles_write() which needs
a "- 1" inserting.

Fixes: 047487c947e8 ("cachefiles: Implement the I/O routines")
Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
cc: linux-cachefs@redhat.com
Link: https://lore.kernel.org/r/164251398954.3435901.7138806620218474123.stgit@warthog.procyon.org.uk/ # v1
---

diff --git a/fs/cachefiles/cache.c b/fs/cachefiles/cache.c
index ce4d478..1e9c7166 100644
--- a/fs/cachefiles/cache.c
+++ b/fs/cachefiles/cache.c
@@ -84,9 +84,7 @@ int cachefiles_add_cache(struct cachefiles_cache *cache)
 		goto error_unsupported;
 
 	cache->bsize = stats.f_bsize;
-	cache->bshift = 0;
-	if (stats.f_bsize < PAGE_SIZE)
-		cache->bshift = PAGE_SHIFT - ilog2(stats.f_bsize);
+	cache->bshift = ilog2(stats.f_bsize);
 
 	_debug("blksize %u (shift %u)",
 	       cache->bsize, cache->bshift);
@@ -106,7 +104,6 @@ int cachefiles_add_cache(struct cachefiles_cache *cache)
 	       (unsigned long long) cache->fcull,
 	       (unsigned long long) cache->fstop);
 
-	stats.f_blocks >>= cache->bshift;
 	do_div(stats.f_blocks, 100);
 	cache->bstop = stats.f_blocks * cache->bstop_percent;
 	cache->bcull = stats.f_blocks * cache->bcull_percent;
@@ -209,7 +206,7 @@ int cachefiles_has_space(struct cachefiles_cache *cache,
 		return ret;
 	}
 
-	b_avail = stats.f_bavail >> cache->bshift;
+	b_avail = stats.f_bavail;
 	b_writing = atomic_long_read(&cache->b_writing);
 	if (b_avail > b_writing)
 		b_avail -= b_writing;
diff --git a/fs/cachefiles/internal.h b/fs/cachefiles/internal.h
index 8dd54d9..c793d33 100644
--- a/fs/cachefiles/internal.h
+++ b/fs/cachefiles/internal.h
@@ -86,7 +86,7 @@ struct cachefiles_cache {
 	unsigned			bcull_percent;	/* when to start culling (% blocks) */
 	unsigned			bstop_percent;	/* when to stop allocating (% blocks) */
 	unsigned			bsize;		/* cache's block size */
-	unsigned			bshift;		/* min(ilog2(PAGE_SIZE / bsize), 0) */
+	unsigned			bshift;		/* ilog2(bsize) */
 	uint64_t			frun;		/* when to stop culling */
 	uint64_t			fcull;		/* when to start culling */
 	uint64_t			fstop;		/* when to stop allocating */
diff --git a/fs/cachefiles/io.c b/fs/cachefiles/io.c
index 60b1eac..04eb5273 100644
--- a/fs/cachefiles/io.c
+++ b/fs/cachefiles/io.c
@@ -264,7 +264,7 @@ static int cachefiles_write(struct netfs_cache_resources *cres,
 	ki->term_func		= term_func;
 	ki->term_func_priv	= term_func_priv;
 	ki->was_async		= true;
-	ki->b_writing		= (len + (1 << cache->bshift)) >> cache->bshift;
+	ki->b_writing		= (len + (1 << cache->bshift) - 1) >> cache->bshift;
 
 	if (ki->term_func)
 		ki->iocb.ki_complete = cachefiles_write_complete;