diff options
author | Meelap Shah <meelap@umich.edu> | 2007-07-17 11:04:39 (GMT) |
---|---|---|
committer | Linus Torvalds <torvalds@woody.linux-foundation.org> | 2007-07-17 17:23:07 (GMT) |
commit | c2f1a551dea8b37c2e0cb886885c250fb703e9d8 (patch) | |
tree | 11a5f256703d856017ceb2268bd02b7b510dee30 /fs/nfsd | |
parent | 1e5140279f31e47d58ed6036ee61ba7a65710e63 (diff) | |
download | linux-fsl-qoriq-c2f1a551dea8b37c2e0cb886885c250fb703e9d8.tar.xz |
knfsd: nfsd4: vary maximum delegation limit based on RAM size
Our original NFSv4 delegation policy was to give out a read delegation on any
open when it was possible to.
Since the lifetime of a delegation isn't limited to that of an open, a client
may quite reasonably hang on to a delegation as long as it has the inode
cached. This becomes an obvious problem the first time a client's inode cache
approaches the size of the server's total memory.
Our first quick solution was to add a hard-coded limit. This patch makes a
mild incremental improvement by varying that limit according to the server's
total memory size, allowing at most 4 delegations per megabyte of RAM.
My quick back-of-the-envelope calculation finds that in the worst case (where
every delegation is for a different inode), a delegation could take about
1.5K, which would make the worst case usage about 6% of memory. The new limit
works out to be about the same as the old on a 1-gig server.
[akpm@linux-foundation.org: Don't needlessly bloat vmlinux]
[akpm@linux-foundation.org: Make it right for highmem machines]
Signed-off-by: "J. Bruce Fields" <bfields@citi.umich.edu>
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Diffstat (limited to 'fs/nfsd')
-rw-r--r-- | fs/nfsd/nfs4state.c | 26 |
1 files changed, 25 insertions, 1 deletions
diff --git a/fs/nfsd/nfs4state.c b/fs/nfsd/nfs4state.c index 9cc31ea..4624988 100644 --- a/fs/nfsd/nfs4state.c +++ b/fs/nfsd/nfs4state.c @@ -49,6 +49,7 @@ #include <linux/nfsd/state.h> #include <linux/nfsd/xdr4.h> #include <linux/namei.h> +#include <linux/swap.h> #include <linux/mutex.h> #include <linux/lockd/bind.h> #include <linux/module.h> @@ -150,6 +151,7 @@ get_nfs4_file(struct nfs4_file *fi) } static int num_delegations; +unsigned int max_delegations; /* * Open owner state (share locks) @@ -193,7 +195,7 @@ alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_f struct nfs4_callback *cb = &stp->st_stateowner->so_client->cl_callback; dprintk("NFSD alloc_init_deleg\n"); - if (num_delegations > STATEID_HASH_SIZE * 4) + if (num_delegations > max_delegations) return NULL; dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL); if (dp == NULL) @@ -3197,6 +3199,27 @@ get_nfs4_grace_period(void) return max(user_lease_time, lease_time) * HZ; } +/* + * Since the lifetime of a delegation isn't limited to that of an open, a + * client may quite reasonably hang on to a delegation as long as it has + * the inode cached. This becomes an obvious problem the first time a + * client's inode cache approaches the size of the server's total memory. + * + * For now we avoid this problem by imposing a hard limit on the number + * of delegations, which varies according to the server's memory size. + */ +static void +set_max_delegations(void) +{ + /* + * Allow at most 4 delegations per megabyte of RAM. Quick + * estimates suggest that in the worst case (where every delegation + * is for a different inode), a delegation could take about 1.5K, + * giving a worst case usage of about 6% of memory. + */ + max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT); +} + /* initialization to perform when the nfsd service is started: */ static void @@ -3212,6 +3235,7 @@ __nfs4_state_start(void) grace_time/HZ); laundry_wq = create_singlethread_workqueue("nfsd4"); queue_delayed_work(laundry_wq, &laundromat_work, grace_time); + set_max_delegations(); } int |