| Message ID | 20160125164111.9466.79976.stgit@warthog.procyon.org.uk (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On 01/25/2016 11:41 AM, David Howells wrote: > Provide read-and-reset objects- and blocks-released counters for cachefilesd > to use to work out whether there's anything new that can be culled. > > One of the problems cachefilesd has is that if all the objects in the cache > are pinned by inodes lying dormant in the kernel inode cache, there isn't > anything for it to cull. In such a case, it just spins around walking the > filesystem tree and scanning for something to cull. This eats up a lot of > CPU time. > > By telling cachefilesd if there have been any releases, the daemon can > sleep until there is the possibility of something to do. > > cachefilesd finds this information by the following means: > > (1) When the control fd is read, the kernel presents a list of values of > interest. "freleased=N" and "breleased=N" are added to this list to > indicate the number of files released and number of blocks released > since the last read call. At this point the counters are reset. > > (2) POLLIN is signalled if the number of files released becomes greater > than 0. > > Note that by 'released' it just means that the kernel has released its > interest in those files for the moment, not necessarily that the files > should be deleted from the cache. > > Signed-off-by: David Howells <dhowells@redhat.com> Reviewed-by: Steve Dickson <steved@redhat.com> steved. > --- > > fs/cachefiles/daemon.c | 13 ++++++++++--- > fs/cachefiles/interface.c | 11 ++--------- > fs/cachefiles/internal.h | 4 ++++ > fs/cachefiles/namei.c | 28 +++++++++++++++++++++++----- > 4 files changed, 39 insertions(+), 17 deletions(-) > > diff --git a/fs/cachefiles/daemon.c b/fs/cachefiles/daemon.c > index 452e98dd7560..1ee54ffd3a24 100644 > --- a/fs/cachefiles/daemon.c > +++ b/fs/cachefiles/daemon.c > @@ -162,6 +162,8 @@ static ssize_t cachefiles_daemon_read(struct file *file, char __user *_buffer, > size_t buflen, loff_t *pos) > { > struct cachefiles_cache *cache = file->private_data; > + unsigned long long b_released; > + unsigned f_released; > char buffer[256]; > int n; > > @@ -174,6 +176,8 @@ static ssize_t cachefiles_daemon_read(struct file *file, char __user *_buffer, > cachefiles_has_space(cache, 0, 0); > > /* summarise */ > + f_released = atomic_xchg(&cache->f_released, 0); > + b_released = atomic_long_xchg(&cache->b_released, 0); > clear_bit(CACHEFILES_STATE_CHANGED, &cache->flags); > > n = snprintf(buffer, sizeof(buffer), > @@ -183,15 +187,18 @@ static ssize_t cachefiles_daemon_read(struct file *file, char __user *_buffer, > " fstop=%llx" > " brun=%llx" > " bcull=%llx" > - " bstop=%llx", > + " bstop=%llx" > + " freleased=%x" > + " breleased=%llx", > test_bit(CACHEFILES_CULLING, &cache->flags) ? '1' : '0', > (unsigned long long) cache->frun, > (unsigned long long) cache->fcull, > (unsigned long long) cache->fstop, > (unsigned long long) cache->brun, > (unsigned long long) cache->bcull, > - (unsigned long long) cache->bstop > - ); > + (unsigned long long) cache->bstop, > + f_released, > + b_released); > > if (n > buflen) > return -EMSGSIZE; > diff --git a/fs/cachefiles/interface.c b/fs/cachefiles/interface.c > index afa023dded5b..ee0cb116ff40 100644 > --- a/fs/cachefiles/interface.c > +++ b/fs/cachefiles/interface.c > @@ -291,15 +291,8 @@ static void cachefiles_drop_object(struct fscache_object *_object) > } > > /* note that the object is now inactive */ > - if (test_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags)) { > - write_lock(&cache->active_lock); > - if (!test_and_clear_bit(CACHEFILES_OBJECT_ACTIVE, > - &object->flags)) > - BUG(); > - rb_erase(&object->active_node, &cache->active_nodes); > - wake_up_bit(&object->flags, CACHEFILES_OBJECT_ACTIVE); > - write_unlock(&cache->active_lock); > - } > + if (test_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags)) > + cachefiles_mark_object_inactive(cache, object); > > dput(object->dentry); > object->dentry = NULL; > diff --git a/fs/cachefiles/internal.h b/fs/cachefiles/internal.h > index 9c4b737a54df..2fcde1a34b7c 100644 > --- a/fs/cachefiles/internal.h > +++ b/fs/cachefiles/internal.h > @@ -66,6 +66,8 @@ struct cachefiles_cache { > struct rb_root active_nodes; /* active nodes (can't be culled) */ > rwlock_t active_lock; /* lock for active_nodes */ > atomic_t gravecounter; /* graveyard uniquifier */ > + atomic_t f_released; /* number of objects released lately */ > + atomic_long_t b_released; /* number of blocks released lately */ > unsigned frun_percent; /* when to stop culling (% files) */ > unsigned fcull_percent; /* when to start culling (% files) */ > unsigned fstop_percent; /* when to stop allocating (% files) */ > @@ -157,6 +159,8 @@ extern char *cachefiles_cook_key(const u8 *raw, int keylen, uint8_t type); > /* > * namei.c > */ > +extern void cachefiles_mark_object_inactive(struct cachefiles_cache *cache, > + struct cachefiles_object *object); > extern int cachefiles_delete_object(struct cachefiles_cache *cache, > struct cachefiles_object *object); > extern int cachefiles_walk_to_object(struct cachefiles_object *parent, > diff --git a/fs/cachefiles/namei.c b/fs/cachefiles/namei.c > index c4b893453e0e..1ad16ae7ab96 100644 > --- a/fs/cachefiles/namei.c > +++ b/fs/cachefiles/namei.c > @@ -258,6 +258,28 @@ requeue: > } > > /* > + * Mark an object as being inactive. > + */ > +void cachefiles_mark_object_inactive(struct cachefiles_cache *cache, > + struct cachefiles_object *object) > +{ > + write_lock(&cache->active_lock); > + rb_erase(&object->active_node, &cache->active_nodes); > + clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags); > + write_unlock(&cache->active_lock); > + > + wake_up_bit(&object->flags, CACHEFILES_OBJECT_ACTIVE); > + > + /* This object can now be culled, so we need to let the daemon know > + * that there is something it can remove if it needs to. > + */ > + atomic_long_add(d_backing_inode(object->dentry)->i_blocks, > + &cache->b_released); > + if (atomic_inc_return(&cache->f_released)) > + cachefiles_state_changed(cache); > +} > + > +/* > * delete an object representation from the cache > * - file backed objects are unlinked > * - directory backed objects are stuffed into the graveyard for userspace to > @@ -684,11 +706,7 @@ mark_active_timed_out: > > check_error: > _debug("check error %d", ret); > - write_lock(&cache->active_lock); > - rb_erase(&object->active_node, &cache->active_nodes); > - clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags); > - wake_up_bit(&object->flags, CACHEFILES_OBJECT_ACTIVE); > - write_unlock(&cache->active_lock); > + cachefiles_mark_object_inactive(cache, object); > release_dentry: > dput(object->dentry); > object->dentry = NULL; > -- To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/fs/cachefiles/daemon.c b/fs/cachefiles/daemon.c index 452e98dd7560..1ee54ffd3a24 100644 --- a/fs/cachefiles/daemon.c +++ b/fs/cachefiles/daemon.c @@ -162,6 +162,8 @@ static ssize_t cachefiles_daemon_read(struct file *file, char __user *_buffer, size_t buflen, loff_t *pos) { struct cachefiles_cache *cache = file->private_data; + unsigned long long b_released; + unsigned f_released; char buffer[256]; int n; @@ -174,6 +176,8 @@ static ssize_t cachefiles_daemon_read(struct file *file, char __user *_buffer, cachefiles_has_space(cache, 0, 0); /* summarise */ + f_released = atomic_xchg(&cache->f_released, 0); + b_released = atomic_long_xchg(&cache->b_released, 0); clear_bit(CACHEFILES_STATE_CHANGED, &cache->flags); n = snprintf(buffer, sizeof(buffer), @@ -183,15 +187,18 @@ static ssize_t cachefiles_daemon_read(struct file *file, char __user *_buffer, " fstop=%llx" " brun=%llx" " bcull=%llx" - " bstop=%llx", + " bstop=%llx" + " freleased=%x" + " breleased=%llx", test_bit(CACHEFILES_CULLING, &cache->flags) ? '1' : '0', (unsigned long long) cache->frun, (unsigned long long) cache->fcull, (unsigned long long) cache->fstop, (unsigned long long) cache->brun, (unsigned long long) cache->bcull, - (unsigned long long) cache->bstop - ); + (unsigned long long) cache->bstop, + f_released, + b_released); if (n > buflen) return -EMSGSIZE; diff --git a/fs/cachefiles/interface.c b/fs/cachefiles/interface.c index afa023dded5b..ee0cb116ff40 100644 --- a/fs/cachefiles/interface.c +++ b/fs/cachefiles/interface.c @@ -291,15 +291,8 @@ static void cachefiles_drop_object(struct fscache_object *_object) } /* note that the object is now inactive */ - if (test_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags)) { - write_lock(&cache->active_lock); - if (!test_and_clear_bit(CACHEFILES_OBJECT_ACTIVE, - &object->flags)) - BUG(); - rb_erase(&object->active_node, &cache->active_nodes); - wake_up_bit(&object->flags, CACHEFILES_OBJECT_ACTIVE); - write_unlock(&cache->active_lock); - } + if (test_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags)) + cachefiles_mark_object_inactive(cache, object); dput(object->dentry); object->dentry = NULL; diff --git a/fs/cachefiles/internal.h b/fs/cachefiles/internal.h index 9c4b737a54df..2fcde1a34b7c 100644 --- a/fs/cachefiles/internal.h +++ b/fs/cachefiles/internal.h @@ -66,6 +66,8 @@ struct cachefiles_cache { struct rb_root active_nodes; /* active nodes (can't be culled) */ rwlock_t active_lock; /* lock for active_nodes */ atomic_t gravecounter; /* graveyard uniquifier */ + atomic_t f_released; /* number of objects released lately */ + atomic_long_t b_released; /* number of blocks released lately */ unsigned frun_percent; /* when to stop culling (% files) */ unsigned fcull_percent; /* when to start culling (% files) */ unsigned fstop_percent; /* when to stop allocating (% files) */ @@ -157,6 +159,8 @@ extern char *cachefiles_cook_key(const u8 *raw, int keylen, uint8_t type); /* * namei.c */ +extern void cachefiles_mark_object_inactive(struct cachefiles_cache *cache, + struct cachefiles_object *object); extern int cachefiles_delete_object(struct cachefiles_cache *cache, struct cachefiles_object *object); extern int cachefiles_walk_to_object(struct cachefiles_object *parent, diff --git a/fs/cachefiles/namei.c b/fs/cachefiles/namei.c index c4b893453e0e..1ad16ae7ab96 100644 --- a/fs/cachefiles/namei.c +++ b/fs/cachefiles/namei.c @@ -258,6 +258,28 @@ requeue: } /* + * Mark an object as being inactive. + */ +void cachefiles_mark_object_inactive(struct cachefiles_cache *cache, + struct cachefiles_object *object) +{ + write_lock(&cache->active_lock); + rb_erase(&object->active_node, &cache->active_nodes); + clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags); + write_unlock(&cache->active_lock); + + wake_up_bit(&object->flags, CACHEFILES_OBJECT_ACTIVE); + + /* This object can now be culled, so we need to let the daemon know + * that there is something it can remove if it needs to. + */ + atomic_long_add(d_backing_inode(object->dentry)->i_blocks, + &cache->b_released); + if (atomic_inc_return(&cache->f_released)) + cachefiles_state_changed(cache); +} + +/* * delete an object representation from the cache * - file backed objects are unlinked * - directory backed objects are stuffed into the graveyard for userspace to @@ -684,11 +706,7 @@ mark_active_timed_out: check_error: _debug("check error %d", ret); - write_lock(&cache->active_lock); - rb_erase(&object->active_node, &cache->active_nodes); - clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags); - wake_up_bit(&object->flags, CACHEFILES_OBJECT_ACTIVE); - write_unlock(&cache->active_lock); + cachefiles_mark_object_inactive(cache, object); release_dentry: dput(object->dentry); object->dentry = NULL;
Provide read-and-reset objects- and blocks-released counters for cachefilesd to use to work out whether there's anything new that can be culled. One of the problems cachefilesd has is that if all the objects in the cache are pinned by inodes lying dormant in the kernel inode cache, there isn't anything for it to cull. In such a case, it just spins around walking the filesystem tree and scanning for something to cull. This eats up a lot of CPU time. By telling cachefilesd if there have been any releases, the daemon can sleep until there is the possibility of something to do. cachefilesd finds this information by the following means: (1) When the control fd is read, the kernel presents a list of values of interest. "freleased=N" and "breleased=N" are added to this list to indicate the number of files released and number of blocks released since the last read call. At this point the counters are reset. (2) POLLIN is signalled if the number of files released becomes greater than 0. Note that by 'released' it just means that the kernel has released its interest in those files for the moment, not necessarily that the files should be deleted from the cache. Signed-off-by: David Howells <dhowells@redhat.com> --- fs/cachefiles/daemon.c | 13 ++++++++++--- fs/cachefiles/interface.c | 11 ++--------- fs/cachefiles/internal.h | 4 ++++ fs/cachefiles/namei.c | 28 +++++++++++++++++++++++----- 4 files changed, 39 insertions(+), 17 deletions(-) -- To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html