GFS2: change gfs2_quota_scan into a shrinker

Deallocation of gfs2_quota_data objects now happens on-demand through a
shrinker instead of routinely deallocating through the quotad daemon.

Signed-off-by: Abhijith Das <adas@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>

diff --git a/fs/gfs2/incore.h b/fs/gfs2/incore.h
index 608849d00021cc436f656d1afb7a0966aec63ba0..592aa5040d29d75fb03b0ada690457c03f4d9a10 100644 (file)
--- a/fs/gfs2/incore.h
+++ b/fs/gfs2/incore.h
@@ -283,7 +283,9 @@ enum {
 
 struct gfs2_quota_data {
        struct list_head qd_list;
-       unsigned int qd_count;
+       struct list_head qd_reclaim;
+
+       atomic_t qd_count;
 
        u32 qd_id;
        unsigned long qd_flags;         /* QDF_... */
@@ -303,7 +305,6 @@ struct gfs2_quota_data {
 
        u64 qd_sync_gen;
        unsigned long qd_last_warn;
-       unsigned long qd_last_touched;
 };
 
 struct gfs2_trans {
@@ -406,7 +407,6 @@ struct gfs2_tune {
        unsigned int gt_quota_warn_period; /* Secs between quota warn msgs */
        unsigned int gt_quota_scale_num; /* Numerator */
        unsigned int gt_quota_scale_den; /* Denominator */
-       unsigned int gt_quota_cache_secs;
        unsigned int gt_quota_quantum; /* Secs between syncs to quota file */
        unsigned int gt_new_files_jdata;
        unsigned int gt_max_readahead; /* Max bytes to read-ahead from disk */

diff --git a/fs/gfs2/main.c b/fs/gfs2/main.c
index 7cacfde321942c2149afbed5ba11a8ffa64077df..86fe06798711aa937d496468ba14b674fcd36a1f 100644 (file)
--- a/fs/gfs2/main.c
+++ b/fs/gfs2/main.c
@@ -23,6 +23,12 @@
 #include "sys.h"
 #include "util.h"
 #include "glock.h"
+#include "quota.h"
+
+static struct shrinker qd_shrinker = {
+       .shrink = gfs2_shrink_qd_memory,
+       .seeks = DEFAULT_SEEKS,
+};
 
 static void gfs2_init_inode_once(void *foo)
 {
@@ -100,6 +106,8 @@ static int __init init_gfs2_fs(void)
        if (!gfs2_quotad_cachep)
                goto fail;
 
+       register_shrinker(&qd_shrinker);
+
        error = register_filesystem(&gfs2_fs_type);
        if (error)
                goto fail;
@@ -117,6 +125,7 @@ static int __init init_gfs2_fs(void)
 fail_unregister:
        unregister_filesystem(&gfs2_fs_type);
 fail:
+       unregister_shrinker(&qd_shrinker);
        gfs2_glock_exit();
 
        if (gfs2_quotad_cachep)
@@ -145,6 +154,7 @@ fail:
 
 static void __exit exit_gfs2_fs(void)
 {
+       unregister_shrinker(&qd_shrinker);
        gfs2_glock_exit();
        gfs2_unregister_debugfs();
        unregister_filesystem(&gfs2_fs_type);

diff --git a/fs/gfs2/ops_address.c b/fs/gfs2/ops_address.c
index 4ddab67867eb273553b6afbf0a42399f56c51735..dde4ead2c3be57cce7639810a54bd779f74006d5 100644 (file)
--- a/fs/gfs2/ops_address.c
+++ b/fs/gfs2/ops_address.c
@@ -442,6 +442,7 @@ static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
         */
        if (unlikely(page->index)) {
                zero_user(page, 0, PAGE_CACHE_SIZE);
+               SetPageUptodate(page);
                return 0;
        }
 

diff --git a/fs/gfs2/ops_fstype.c b/fs/gfs2/ops_fstype.c
index 3eb49edae542dd17cf9f4597e583bf6a91e744cd..530d3f6f6ea8a9de1994731f3e9c9b94a7da666b 100644 (file)
--- a/fs/gfs2/ops_fstype.c
+++ b/fs/gfs2/ops_fstype.c
@@ -63,7 +63,6 @@ static void gfs2_tune_init(struct gfs2_tune *gt)
        gt->gt_quota_warn_period = 10;
        gt->gt_quota_scale_num = 1;
        gt->gt_quota_scale_den = 1;
-       gt->gt_quota_cache_secs = 300;
        gt->gt_quota_quantum = 60;
        gt->gt_new_files_jdata = 0;
        gt->gt_max_readahead = 1 << 18;

diff --git a/fs/gfs2/quota.c b/fs/gfs2/quota.c
index b08d09696b3e8d92fd8af400cfb819fa196d8229..2ada6e10d07b873b126bb894c903ed5fd29b6ca3 100644 (file)
--- a/fs/gfs2/quota.c
+++ b/fs/gfs2/quota.c
@@ -80,6 +80,53 @@ struct gfs2_quota_change_host {
        u32 qc_id;
 };
 
+static LIST_HEAD(qd_lru_list);
+static atomic_t qd_lru_count = ATOMIC_INIT(0);
+static spinlock_t qd_lru_lock = SPIN_LOCK_UNLOCKED;
+
+int gfs2_shrink_qd_memory(int nr, gfp_t gfp_mask)
+{
+       struct gfs2_quota_data *qd;
+       struct gfs2_sbd *sdp;
+
+       if (nr == 0)
+               goto out;
+
+       if (!(gfp_mask & __GFP_FS))
+               return -1;
+
+       spin_lock(&qd_lru_lock);
+       while (nr && !list_empty(&qd_lru_list)) {
+               qd = list_entry(qd_lru_list.next,
+                               struct gfs2_quota_data, qd_reclaim);
+               sdp = qd->qd_gl->gl_sbd;
+
+               /* Free from the filesystem-specific list */
+               list_del(&qd->qd_list);
+
+               spin_lock(&sdp->sd_quota_spin);
+               gfs2_assert_warn(sdp, !qd->qd_change);
+               gfs2_assert_warn(sdp, !qd->qd_slot_count);
+               gfs2_assert_warn(sdp, !qd->qd_bh_count);
+
+               gfs2_lvb_unhold(qd->qd_gl);
+               spin_unlock(&sdp->sd_quota_spin);
+               atomic_dec(&sdp->sd_quota_count);
+
+               /* Delete it from the common reclaim list */
+               list_del_init(&qd->qd_reclaim);
+               atomic_dec(&qd_lru_count);
+               spin_unlock(&qd_lru_lock);
+               kmem_cache_free(gfs2_quotad_cachep, qd);
+               spin_lock(&qd_lru_lock);
+               nr--;
+       }
+       spin_unlock(&qd_lru_lock);
+
+out:
+       return (atomic_read(&qd_lru_count) * sysctl_vfs_cache_pressure) / 100;
+}
+
 static u64 qd2offset(struct gfs2_quota_data *qd)
 {
        u64 offset;
@@ -100,11 +147,12 @@ static int qd_alloc(struct gfs2_sbd *sdp, int user, u32 id,
        if (!qd)
                return -ENOMEM;
 
-       qd->qd_count = 1;
+       atomic_set(&qd->qd_count, 1);
        qd->qd_id = id;
        if (user)
                set_bit(QDF_USER, &qd->qd_flags);
        qd->qd_slot = -1;
+       INIT_LIST_HEAD(&qd->qd_reclaim);
 
        error = gfs2_glock_get(sdp, 2 * (u64)id + !user,
                              &gfs2_quota_glops, CREATE, &qd->qd_gl);
@@ -135,11 +183,17 @@ static int qd_get(struct gfs2_sbd *sdp, int user, u32 id, int create,
 
        for (;;) {
                found = 0;
-               spin_lock(&sdp->sd_quota_spin);
+               spin_lock(&qd_lru_lock);
                list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
                        if (qd->qd_id == id &&
                            !test_bit(QDF_USER, &qd->qd_flags) == !user) {
-                               qd->qd_count++;
+                               if (!atomic_read(&qd->qd_count) &&
+                                   !list_empty(&qd->qd_reclaim)) {
+                                       /* Remove it from reclaim list */
+                                       list_del_init(&qd->qd_reclaim);
+                                       atomic_dec(&qd_lru_count);
+                               }
+                               atomic_inc(&qd->qd_count);
                                found = 1;
                                break;
                        }
@@ -155,7 +209,7 @@ static int qd_get(struct gfs2_sbd *sdp, int user, u32 id, int create,
                        new_qd = NULL;
                }
 
-               spin_unlock(&sdp->sd_quota_spin);
+               spin_unlock(&qd_lru_lock);
 
                if (qd || !create) {
                        if (new_qd) {
@@ -175,21 +229,18 @@ static int qd_get(struct gfs2_sbd *sdp, int user, u32 id, int create,
 static void qd_hold(struct gfs2_quota_data *qd)
 {
        struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
-
-       spin_lock(&sdp->sd_quota_spin);
-       gfs2_assert(sdp, qd->qd_count);
-       qd->qd_count++;
-       spin_unlock(&sdp->sd_quota_spin);
+       gfs2_assert(sdp, atomic_read(&qd->qd_count));
+       atomic_inc(&qd->qd_count);
 }
 
 static void qd_put(struct gfs2_quota_data *qd)
 {
-       struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
-       spin_lock(&sdp->sd_quota_spin);
-       gfs2_assert(sdp, qd->qd_count);
-       if (!--qd->qd_count)
-               qd->qd_last_touched = jiffies;
-       spin_unlock(&sdp->sd_quota_spin);
+       if (atomic_dec_and_lock(&qd->qd_count, &qd_lru_lock)) {
+               /* Add to the reclaim list */
+               list_add_tail(&qd->qd_reclaim, &qd_lru_list);
+               atomic_inc(&qd_lru_count);
+               spin_unlock(&qd_lru_lock);
+       }
 }
 
 static int slot_get(struct gfs2_quota_data *qd)
@@ -330,6 +381,7 @@ static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
        if (sdp->sd_vfs->s_flags & MS_RDONLY)
                return 0;
 
+       spin_lock(&qd_lru_lock);
        spin_lock(&sdp->sd_quota_spin);
 
        list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
@@ -341,8 +393,8 @@ static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
                list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
 
                set_bit(QDF_LOCKED, &qd->qd_flags);
-               gfs2_assert_warn(sdp, qd->qd_count);
-               qd->qd_count++;
+               gfs2_assert_warn(sdp, atomic_read(&qd->qd_count));
+               atomic_inc(&qd->qd_count);
                qd->qd_change_sync = qd->qd_change;
                gfs2_assert_warn(sdp, qd->qd_slot_count);
                qd->qd_slot_count++;
@@ -355,6 +407,7 @@ static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
                qd = NULL;
 
        spin_unlock(&sdp->sd_quota_spin);
+       spin_unlock(&qd_lru_lock);
 
        if (qd) {
                gfs2_assert_warn(sdp, qd->qd_change_sync);
@@ -379,24 +432,27 @@ static int qd_trylock(struct gfs2_quota_data *qd)
        if (sdp->sd_vfs->s_flags & MS_RDONLY)
                return 0;
 
+       spin_lock(&qd_lru_lock);
        spin_lock(&sdp->sd_quota_spin);
 
        if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
            !test_bit(QDF_CHANGE, &qd->qd_flags)) {
                spin_unlock(&sdp->sd_quota_spin);
+               spin_unlock(&qd_lru_lock);
                return 0;
        }
 
        list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
 
        set_bit(QDF_LOCKED, &qd->qd_flags);
-       gfs2_assert_warn(sdp, qd->qd_count);
-       qd->qd_count++;
+       gfs2_assert_warn(sdp, atomic_read(&qd->qd_count));
+       atomic_inc(&qd->qd_count);
        qd->qd_change_sync = qd->qd_change;
        gfs2_assert_warn(sdp, qd->qd_slot_count);
        qd->qd_slot_count++;
 
        spin_unlock(&sdp->sd_quota_spin);
+       spin_unlock(&qd_lru_lock);
 
        gfs2_assert_warn(sdp, qd->qd_change_sync);
        if (bh_get(qd)) {
@@ -802,8 +858,8 @@ restart:
                loff_t pos;
                gfs2_glock_dq_uninit(q_gh);
                error = gfs2_glock_nq_init(qd->qd_gl,
-                                         LM_ST_EXCLUSIVE, GL_NOCACHE,
-                                         q_gh);
+                                          LM_ST_EXCLUSIVE, GL_NOCACHE,
+                                          q_gh);
                if (error)
                        return error;
 
@@ -820,7 +876,6 @@ restart:
 
                gfs2_glock_dq_uninit(&i_gh);
 
-
                gfs2_quota_in(&q, buf);
                qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lvb;
                qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
@@ -1171,13 +1226,14 @@ int gfs2_quota_init(struct gfs2_sbd *sdp)
                        qd->qd_change = qc.qc_change;
                        qd->qd_slot = slot;
                        qd->qd_slot_count = 1;
-                       qd->qd_last_touched = jiffies;
 
+                       spin_lock(&qd_lru_lock);
                        spin_lock(&sdp->sd_quota_spin);
                        gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, slot, 1);
+                       spin_unlock(&sdp->sd_quota_spin);
                        list_add(&qd->qd_list, &sdp->sd_quota_list);
                        atomic_inc(&sdp->sd_quota_count);
-                       spin_unlock(&sdp->sd_quota_spin);
+                       spin_unlock(&qd_lru_lock);
 
                        found++;
                }
@@ -1197,61 +1253,39 @@ fail:
        return error;
 }
 
-static void gfs2_quota_scan(struct gfs2_sbd *sdp)
-{
-       struct gfs2_quota_data *qd, *safe;
-       LIST_HEAD(dead);
-
-       spin_lock(&sdp->sd_quota_spin);
-       list_for_each_entry_safe(qd, safe, &sdp->sd_quota_list, qd_list) {
-               if (!qd->qd_count &&
-                   time_after_eq(jiffies, qd->qd_last_touched +
-                               gfs2_tune_get(sdp, gt_quota_cache_secs) * HZ)) {
-                       list_move(&qd->qd_list, &dead);
-                       gfs2_assert_warn(sdp,
-                                        atomic_read(&sdp->sd_quota_count) > 0);
-                       atomic_dec(&sdp->sd_quota_count);
-               }
-       }
-       spin_unlock(&sdp->sd_quota_spin);
-
-       while (!list_empty(&dead)) {
-               qd = list_entry(dead.next, struct gfs2_quota_data, qd_list);
-               list_del(&qd->qd_list);
-
-               gfs2_assert_warn(sdp, !qd->qd_change);
-               gfs2_assert_warn(sdp, !qd->qd_slot_count);
-               gfs2_assert_warn(sdp, !qd->qd_bh_count);
-
-               gfs2_lvb_unhold(qd->qd_gl);
-               kmem_cache_free(gfs2_quotad_cachep, qd);
-       }
-}
-
 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
 {
        struct list_head *head = &sdp->sd_quota_list;
        struct gfs2_quota_data *qd;
        unsigned int x;
 
-       spin_lock(&sdp->sd_quota_spin);
+       spin_lock(&qd_lru_lock);
        while (!list_empty(head)) {
                qd = list_entry(head->prev, struct gfs2_quota_data, qd_list);
 
-               if (qd->qd_count > 1 ||
-                   (qd->qd_count && !test_bit(QDF_CHANGE, &qd->qd_flags))) {
-                       list_move(&qd->qd_list, head);
+               spin_lock(&sdp->sd_quota_spin);
+               if (atomic_read(&qd->qd_count) > 1 ||
+                   (atomic_read(&qd->qd_count) &&
+                    !test_bit(QDF_CHANGE, &qd->qd_flags))) {
                        spin_unlock(&sdp->sd_quota_spin);
+                       list_move(&qd->qd_list, head);
+                       spin_unlock(&qd_lru_lock);
                        schedule();
-                       spin_lock(&sdp->sd_quota_spin);
+                       spin_lock(&qd_lru_lock);
                        continue;
                }
+               spin_unlock(&sdp->sd_quota_spin);
 
                list_del(&qd->qd_list);
+               /* Also remove if this qd exists in the reclaim list */
+               if (!list_empty(&qd->qd_reclaim)) {
+                       list_del_init(&qd->qd_reclaim);
+                       atomic_dec(&qd_lru_count);
+               }
                atomic_dec(&sdp->sd_quota_count);
-               spin_unlock(&sdp->sd_quota_spin);
+               spin_unlock(&qd_lru_lock);
 
-               if (!qd->qd_count) {
+               if (!atomic_read(&qd->qd_count)) {
                        gfs2_assert_warn(sdp, !qd->qd_change);
                        gfs2_assert_warn(sdp, !qd->qd_slot_count);
                } else
@@ -1261,9 +1295,9 @@ void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
                gfs2_lvb_unhold(qd->qd_gl);
                kmem_cache_free(gfs2_quotad_cachep, qd);
 
-               spin_lock(&sdp->sd_quota_spin);
+               spin_lock(&qd_lru_lock);
        }
-       spin_unlock(&sdp->sd_quota_spin);
+       spin_unlock(&qd_lru_lock);
 
        gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_count));
 
@@ -1341,9 +1375,6 @@ int gfs2_quotad(void *data)
                quotad_check_timeo(sdp, "sync", gfs2_quota_sync, t,
                                   &quotad_timeo, &tune->gt_quota_quantum);
 
-               /* FIXME: This should be turned into a shrinker */
-               gfs2_quota_scan(sdp);
-
                /* Check for & recover partially truncated inodes */
                quotad_check_trunc_list(sdp);
 

diff --git a/fs/gfs2/quota.h b/fs/gfs2/quota.h
index cec9032be97da0bbcb439249c4c47ec90cf0cd77..0fa5fa63d0e8fce00a37c2ea4cbdc7a476fe5c3e 100644 (file)
--- a/fs/gfs2/quota.h
+++ b/fs/gfs2/quota.h
@@ -49,4 +49,6 @@ static inline int gfs2_quota_lock_check(struct gfs2_inode *ip)
        return ret;
 }
 
+extern int gfs2_shrink_qd_memory(int nr, gfp_t gfp_mask);
+
 #endif /* __QUOTA_DOT_H__ */

diff --git a/fs/gfs2/sys.c b/fs/gfs2/sys.c
index 26c1fa777a959f66f33755b977fa54f61ee04cfe..a58a120dac92f9fad611cc33f0d9ea17f592d4c9 100644 (file)
--- a/fs/gfs2/sys.c
+++ b/fs/gfs2/sys.c
@@ -373,7 +373,6 @@ TUNE_ATTR(complain_secs, 0);
 TUNE_ATTR(statfs_slow, 0);
 TUNE_ATTR(new_files_jdata, 0);
 TUNE_ATTR(quota_simul_sync, 1);
-TUNE_ATTR(quota_cache_secs, 1);
 TUNE_ATTR(stall_secs, 1);
 TUNE_ATTR(statfs_quantum, 1);
 TUNE_ATTR_DAEMON(recoverd_secs, recoverd_process);
@@ -389,7 +388,6 @@ static struct attribute *tune_attrs[] = {
        &tune_attr_complain_secs.attr,
        &tune_attr_statfs_slow.attr,
        &tune_attr_quota_simul_sync.attr,
-       &tune_attr_quota_cache_secs.attr,
        &tune_attr_stall_secs.attr,
        &tune_attr_statfs_quantum.attr,
        &tune_attr_recoverd_secs.attr,