X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fosd-zfs%2Fosd_quota.c;h=45b68ff556443e6b5593d8ca505948053952c61e;hb=refs%2Fchanges%2F93%2F27093%2F40;hp=7faa28732b8b574c02e94254b61c23682a6ac7ec;hpb=08aa217ce49aba1ded52e0f7adb8a607035123fd;p=fs%2Flustre-release.git

diff --git a/lustre/osd-zfs/osd_quota.c b/lustre/osd-zfs/osd_quota.c
index 7faa287..45b68ff 100644
--- a/lustre/osd-zfs/osd_quota.c
+++ b/lustre/osd-zfs/osd_quota.c
@@ -21,7 +21,7 @@
  * GPL HEADER END
  */
 /*
- * Copyright (c) 2012, Intel Corporation.
+ * Copyright (c) 2012, 2015, Intel Corporation.
  * Use is subject to license terms.
  *
  * Author: Johann Lombardi <johann@whamcloud.com>
@@ -29,18 +29,30 @@
 
 #include <lustre_quota.h>
 #include <obd.h>
-#include "udmu.h"
 #include "osd_internal.h"
 
 /**
- * Helper function to retrieve DMU object id from fid for accounting object
+ * Helper function to estimate the number of inodes in use for the given
+ * uid/gid/projid from the block usage
  */
-uint64_t osd_quota_fid2dmu(const struct lu_fid *fid)
+static uint64_t osd_objset_user_iused(struct osd_device *osd, uint64_t uidbytes)
 {
-	LASSERT(fid_is_acct(fid));
-	if (fid_oid(fid) == ACCT_GROUP_OID)
-		return DMU_GROUPUSED_OBJECT;
-	return DMU_USERUSED_OBJECT;
+	uint64_t refdbytes, availbytes, usedobjs, availobjs;
+	uint64_t uidobjs, bshift;
+
+	/* get fresh statfs info */
+	dmu_objset_space(osd->od_os, &refdbytes, &availbytes,
+			 &usedobjs, &availobjs);
+
+	/* estimate the number of objects based on the disk usage */
+	bshift = fls64(osd->od_max_blksz) - 1;
+	uidobjs = osd_objs_count_estimate(refdbytes, usedobjs,
+					  uidbytes >> bshift, bshift);
+	if (uidbytes > 0)
+		/* if we have at least 1 byte, we have at least one dnode ... */
+		uidobjs = max_t(uint64_t, uidobjs, 1);
+
+	return uidobjs;
 }
 
 /**
@@ -48,7 +60,7 @@ uint64_t osd_quota_fid2dmu(const struct lu_fid *fid)
  */
 
 /**
- * Return space usage consumed by a given uid or gid.
+ * Return space usage consumed by a given uid or gid or projid.
  * Block usage is accurrate since it is maintained by DMU itself.
  * However, DMU does not provide inode accounting, so the #inodes in use
  * is estimated from the block usage and statfs information.
@@ -58,7 +70,6 @@ uint64_t osd_quota_fid2dmu(const struct lu_fid *fid)
  * \param dtrec - is the record to fill with space usage information
  * \param dtkey - is the id the of the user or group for which we would
  *                like to access disk usage.
- * \param capa - is the capability, not used.
  *
  * \retval +ve - success : exact match
  * \retval -ve - failure
@@ -66,60 +77,66 @@ uint64_t osd_quota_fid2dmu(const struct lu_fid *fid)
 static int osd_acct_index_lookup(const struct lu_env *env,
 				struct dt_object *dtobj,
 				struct dt_rec *dtrec,
-				const struct dt_key *dtkey,
-				struct lustre_capa *capa)
+				const struct dt_key *dtkey)
 {
-	struct osd_thread_info	*info = osd_oti_get(env);
-	char			*buf  = info->oti_buf;
-	struct lquota_acct_rec	*rec  = (struct lquota_acct_rec *)dtrec;
-	struct osd_object	*obj = osd_dt_obj(dtobj);
-	struct osd_device	*osd = osd_obj2dev(obj);
-	int			 rc;
-	uint64_t		 oid;
+	struct osd_thread_info *info = osd_oti_get(env);
+	char *buf = info->oti_buf;
+	struct lquota_acct_rec *rec = (struct lquota_acct_rec *)dtrec;
+	struct osd_object *obj = osd_dt_obj(dtobj);
+	struct osd_device *osd = osd_obj2dev(obj);
+	dnode_t *dn = obj->oo_dn;
+	size_t buflen = sizeof(info->oti_buf);
+	int rc;
 	ENTRY;
 
 	rec->bspace = rec->ispace = 0;
 
-	/* convert the 64-bit uid/gid into a string */
-	sprintf(buf, "%llx", *((__u64 *)dtkey));
-	/* fetch DMU object ID (DMU_USERUSED_OBJECT/DMU_GROUPUSED_OBJECT) to be
-	 * used */
-	oid = osd_quota_fid2dmu(lu_object_fid(&dtobj->do_lu));
+	/* convert the 64-bit uid/gid/projid into a string */
+	snprintf(buf, buflen, "%llx", *((__u64 *)dtkey));
+	if (unlikely(!dn)) {
+		CDEBUG(D_QUOTA, "%s: miss accounting obj for %s\n",
+		       osd->od_svname, buf);
+
+		RETURN(-ENOENT);
+	}
 
 	/* disk usage (in bytes) is maintained by DMU.
 	 * DMU_USERUSED_OBJECT/DMU_GROUPUSED_OBJECT are special objects which
-	 * not associated with any dmu_but_t (see dnode_special_open()).
-	 * As a consequence, we cannot use udmu_zap_lookup() here since it
-	 * requires a valid oo_db. */
-	rc = -zap_lookup(osd->od_objset.os, oid, buf, sizeof(uint64_t), 1,
-			&rec->bspace);
-	if (rc == -ENOENT)
-		/* user/group has not created anything yet */
+	 * not associated with any dmu_but_t (see dnode_special_open()). */
+	rc = osd_zap_lookup(osd, dn->dn_object, dn, buf, sizeof(uint64_t), 1,
+			    &rec->bspace);
+	if (rc == -ENOENT) {
+		/* user/group/project has not created anything yet */
 		CDEBUG(D_QUOTA, "%s: id %s not found in DMU accounting ZAP\n",
 		       osd->od_svname, buf);
-	else if (rc)
+		/* -ENOENT is normal case, convert it as 1. */
+		rc = 1;
+	} else if (rc) {
 		RETURN(rc);
+	}
 
-	if (osd->od_quota_iused_est) {
+	if (!osd_dmu_userobj_accounting_available(osd)) {
 		if (rec->bspace != 0)
 			/* estimate #inodes in use */
-			rec->ispace = udmu_objset_user_iused(&osd->od_objset,
-							     rec->bspace);
-		RETURN(+1);
+			rec->ispace = osd_objset_user_iused(osd, rec->bspace);
+		rc = 1;
+	} else {
+		snprintf(buf, buflen, OSD_DMU_USEROBJ_PREFIX "%llx",
+			 *((__u64 *)dtkey));
+		rc = osd_zap_lookup(osd, dn->dn_object, dn, buf,
+				    sizeof(uint64_t), 1, &rec->ispace);
+		if (rc == -ENOENT) {
+			CDEBUG(D_QUOTA,
+			       "%s: id %s not found dnode accounting\n",
+			       osd->od_svname, buf);
+			/* -ENOENT is normal case, convert it as 1. */
+			rc = 1;
+		} else if (rc == 0) {
+			rc = 1;
+		}
 	}
 
-	/* as for inode accounting, it is not maintained by DMU, so we just
-	 * use our own ZAP to track inode usage */
-	rc = -zap_lookup(osd->od_objset.os, obj->oo_db->db_object,
-			 buf, sizeof(uint64_t), 1, &rec->ispace);
-	if (rc == -ENOENT)
-		/* user/group has not created any file yet */
-		CDEBUG(D_QUOTA, "%s: id %s not found in accounting ZAP\n",
-		       osd->od_svname, buf);
-	else if (rc)
-		RETURN(rc);
-
-	RETURN(+1);
+	RETURN(rc);
 }
 
 /**
@@ -127,36 +144,45 @@ static int osd_acct_index_lookup(const struct lu_env *env,
  *
  * \param  dt    - osd index object
  * \param  attr  - not used
- * \param  capa  - BYPASS_CAPA
  */
 static struct dt_it *osd_it_acct_init(const struct lu_env *env,
 				      struct dt_object *dt,
-				      __u32 attr,
-				      struct lustre_capa *capa)
+				      __u32 attr)
 {
-	struct osd_thread_info	*info = osd_oti_get(env);
-	struct osd_it_quota	*it;
-	struct lu_object	*lo   = &dt->do_lu;
-	struct osd_device	*osd  = osd_dev(lo->lo_dev);
-	int			 rc;
+	struct osd_thread_info *info = osd_oti_get(env);
+	struct osd_it_quota *it;
+	struct osd_object *obj = osd_dt_obj(dt);
+	struct osd_device *osd = osd_obj2dev(obj);
+	dnode_t *dn = obj->oo_dn;
+	int rc;
 	ENTRY;
 
-	LASSERT(lu_object_exists(lo));
+	if (unlikely(!dn)) {
+		CDEBUG(D_QUOTA, "%s: Not found in DMU accounting ZAP\n",
+		       osd->od_svname);
+
+		RETURN(ERR_PTR(-ENOENT));
+	}
 
 	if (info == NULL)
 		RETURN(ERR_PTR(-ENOMEM));
 
-	it = &info->oti_it_quota;
+	OBD_ALLOC_PTR(it);
+	if (it == NULL)
+		RETURN(ERR_PTR(-ENOMEM));
+
 	memset(it, 0, sizeof(*it));
-	it->oiq_oid = osd_quota_fid2dmu(lu_object_fid(lo));
+	it->oiq_oid = dn->dn_object;
 
 	/* initialize zap cursor */
-	rc = -udmu_zap_cursor_init(&it->oiq_zc, &osd->od_objset, it->oiq_oid,0);
-	if (rc)
+	rc = osd_zap_cursor_init(&it->oiq_zc, osd->od_os, it->oiq_oid, 0);
+	if (rc != 0) {
+		OBD_FREE_PTR(it);
 		RETURN(ERR_PTR(rc));
+	}
 
 	/* take object reference */
-	lu_object_get(lo);
+	lu_object_get(&dt->do_lu);
 	it->oiq_obj   = osd_dt_obj(dt);
 	it->oiq_reset = 1;
 
@@ -170,14 +196,40 @@ static struct dt_it *osd_it_acct_init(const struct lu_env *env,
  */
 static void osd_it_acct_fini(const struct lu_env *env, struct dt_it *di)
 {
-	struct osd_it_quota *it = (struct osd_it_quota *)di;
+	struct osd_it_quota	*it	= (struct osd_it_quota *)di;
 	ENTRY;
-	udmu_zap_cursor_fini(it->oiq_zc);
-	lu_object_put(env, &it->oiq_obj->oo_dt.do_lu);
+
+	osd_zap_cursor_fini(it->oiq_zc);
+	osd_object_put(env, it->oiq_obj);
+	OBD_FREE_PTR(it);
+
 	EXIT;
 }
 
 /**
+ * Locate the first entry that is for space accounting.
+ */
+static int osd_zap_locate(struct osd_it_quota *it, zap_attribute_t *za)
+{
+	int rc;
+	ENTRY;
+
+	while (1) {
+		rc = -zap_cursor_retrieve(it->oiq_zc, za);
+		if (rc)
+			break;
+
+		if (strncmp(za->za_name, OSD_DMU_USEROBJ_PREFIX,
+			    OSD_DMU_USEROBJ_PREFIX_LEN))
+			break;
+
+		zap_cursor_advance(it->oiq_zc);
+	}
+
+	RETURN(rc);
+}
+
+/**
  * Move on to the next valid entry.
  *
  * \param  di   - osd iterator
@@ -189,16 +241,16 @@ static void osd_it_acct_fini(const struct lu_env *env, struct dt_it *di)
 static int osd_it_acct_next(const struct lu_env *env, struct dt_it *di)
 {
 	struct osd_it_quota	*it = (struct osd_it_quota *)di;
+	zap_attribute_t		*za = &osd_oti_get(env)->oti_za;
 	int			 rc;
 	ENTRY;
 
 	if (it->oiq_reset == 0)
 		zap_cursor_advance(it->oiq_zc);
 	it->oiq_reset = 0;
-	rc = -udmu_zap_cursor_retrieve_key(env, it->oiq_zc, NULL, 32);
-	if (rc == -ENOENT) /* reached the end */
-		RETURN(+1);
-	RETURN(rc);
+
+	rc = osd_zap_locate(it, za);
+	RETURN(rc == -ENOENT ? 1 : rc);
 }
 
 /**
@@ -210,17 +262,19 @@ static struct dt_key *osd_it_acct_key(const struct lu_env *env,
 				      const struct dt_it *di)
 {
 	struct osd_it_quota	*it = (struct osd_it_quota *)di;
-	struct osd_thread_info	*info = osd_oti_get(env);
-	char			*buf  = info->oti_buf;
-	char			*p;
+	zap_attribute_t		*za = &osd_oti_get(env)->oti_za;
 	int			 rc;
 	ENTRY;
 
 	it->oiq_reset = 0;
-	rc = -udmu_zap_cursor_retrieve_key(env, it->oiq_zc, buf, 32);
+	rc = osd_zap_locate(it, za);
 	if (rc)
 		RETURN(ERR_PTR(rc));
-	it->oiq_id = simple_strtoull(buf, &p, 16);
+
+	rc = kstrtoull(za->za_name, 16, &it->oiq_id);
+	if (rc)
+		CERROR("couldn't parse name %s\n", za->za_name);
+
 	RETURN((struct dt_key *) &it->oiq_id);
 }
 
@@ -236,6 +290,47 @@ static int osd_it_acct_key_size(const struct lu_env *env,
 	RETURN((int)sizeof(uint64_t));
 }
 
+/*
+ * zap_cursor_retrieve read from current record.
+ * to read bytes we need to call zap_lookup explicitly.
+ */
+static int osd_zap_cursor_retrieve_value(const struct lu_env *env,
+					 struct osd_it_quota *it,
+					 char *buf, int buf_size,
+					 int *bytes_read)
+{
+	const struct lu_fid *fid = lu_object_fid(&it->oiq_obj->oo_dt.do_lu);
+	zap_attribute_t *za = &osd_oti_get(env)->oti_za;
+	zap_cursor_t *zc = it->oiq_zc;
+	struct osd_device *osd = osd_obj2dev(it->oiq_obj);
+	int rc, actual_size;
+
+	rc = -zap_cursor_retrieve(zc, za);
+	if (unlikely(rc != 0))
+		return rc;
+
+	if (unlikely(za->za_integer_length <= 0))
+		return -ERANGE;
+
+	actual_size = za->za_integer_length * za->za_num_integers;
+
+	if (actual_size > buf_size) {
+		actual_size = buf_size;
+		buf_size = actual_size / za->za_integer_length;
+	} else {
+		buf_size = za->za_num_integers;
+	}
+
+	/* use correct special ID to request bytes used */
+	rc = osd_zap_lookup(osd, fid_oid(fid) == ACCT_GROUP_OID ?
+			    DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT, NULL,
+			    za->za_name, za->za_integer_length, buf_size, buf);
+	if (likely(rc == 0))
+		*bytes_read = actual_size;
+
+	return rc;
+}
+
 /**
  * Return pointer to the record under iterator.
  *
@@ -247,7 +342,7 @@ static int osd_it_acct_rec(const struct lu_env *env,
 			   struct dt_rec *dtrec, __u32 attr)
 {
 	struct osd_thread_info	*info = osd_oti_get(env);
-	char			*buf  = info->oti_buf;
+	zap_attribute_t		*za = &info->oti_za;
 	struct osd_it_quota	*it = (struct osd_it_quota *)di;
 	struct lquota_acct_rec	*rec  = (struct lquota_acct_rec *)dtrec;
 	struct osd_object	*obj = it->oiq_obj;
@@ -260,33 +355,35 @@ static int osd_it_acct_rec(const struct lu_env *env,
 	rec->ispace = rec->bspace = 0;
 
 	/* retrieve block usage from the DMU accounting object */
-	rc = -udmu_zap_cursor_retrieve_value(env, it->oiq_zc,
-					     (char *)&rec->bspace,
-					     sizeof(uint64_t), &bytes_read);
+	rc = osd_zap_cursor_retrieve_value(env, it, (char *)&rec->bspace,
+					   sizeof(uint64_t), &bytes_read);
 	if (rc)
 		RETURN(rc);
 
-	if (osd->od_quota_iused_est) {
+	if (!osd_dmu_userobj_accounting_available(osd)) {
 		if (rec->bspace != 0)
 			/* estimate #inodes in use */
-			rec->ispace = udmu_objset_user_iused(&osd->od_objset,
-							     rec->bspace);
+			rec->ispace = osd_objset_user_iused(osd, rec->bspace);
 		RETURN(0);
 	}
 
 	/* retrieve key associated with the current cursor */
-	rc = -udmu_zap_cursor_retrieve_key(env, it->oiq_zc, buf, 32);
-	if (rc)
+	rc = -zap_cursor_retrieve(it->oiq_zc, za);
+	if (unlikely(rc != 0))
 		RETURN(rc);
 
-	/* inode accounting is not maintained by DMU, so we use our own ZAP to
-	 * track inode usage */
-	rc = -zap_lookup(osd->od_objset.os, it->oiq_obj->oo_db->db_object,
-			 buf, sizeof(uint64_t), 1, &rec->ispace);
+	/* inode accounting is maintained by DMU since 0.7.0 */
+	strncpy(info->oti_buf, OSD_DMU_USEROBJ_PREFIX,
+		OSD_DMU_USEROBJ_PREFIX_LEN);
+	strlcpy(info->oti_buf + OSD_DMU_USEROBJ_PREFIX_LEN, za->za_name,
+		sizeof(info->oti_buf) - OSD_DMU_USEROBJ_PREFIX_LEN);
+	rc = osd_zap_lookup(osd, it->oiq_obj->oo_dn->dn_object,
+			    it->oiq_obj->oo_dn, info->oti_buf, sizeof(uint64_t),
+			    1, &rec->ispace);
 	if (rc == -ENOENT)
 		/* user/group has not created any file yet */
 		CDEBUG(D_QUOTA, "%s: id %s not found in accounting ZAP\n",
-		       osd->od_svname, buf);
+		       osd->od_svname, info->oti_buf);
 	else if (rc)
 		RETURN(rc);
 
@@ -304,7 +401,7 @@ static __u64 osd_it_acct_store(const struct lu_env *env,
 	struct osd_it_quota *it = (struct osd_it_quota *)di;
 	ENTRY;
 	it->oiq_reset = 0;
-	RETURN(udmu_zap_cursor_serialize(it->oiq_zc));
+	RETURN(osd_zap_cursor_serialize(it->oiq_zc));
 }
 
 /**
@@ -323,23 +420,24 @@ static int osd_it_acct_load(const struct lu_env *env,
 {
 	struct osd_it_quota	*it  = (struct osd_it_quota *)di;
 	struct osd_device	*osd = osd_obj2dev(it->oiq_obj);
+	zap_attribute_t		*za = &osd_oti_get(env)->oti_za;
 	zap_cursor_t		*zc;
 	int			 rc;
 	ENTRY;
 
 	/* create new cursor pointing to the new hash */
-	rc = -udmu_zap_cursor_init(&zc, &osd->od_objset, it->oiq_oid, hash);
+	rc = osd_zap_cursor_init(&zc, osd->od_os, it->oiq_oid, hash);
 	if (rc)
 		RETURN(rc);
-	udmu_zap_cursor_fini(it->oiq_zc);
+	osd_zap_cursor_fini(it->oiq_zc);
 	it->oiq_zc = zc;
 	it->oiq_reset = 0;
 
-	rc = -udmu_zap_cursor_retrieve_key(env, it->oiq_zc, NULL, 32);
+	rc = osd_zap_locate(it, za);
 	if (rc == 0)
-		RETURN(+1);
+		rc = 1;
 	else if (rc == -ENOENT)
-		RETURN(0);
+		rc = 0;
 	RETURN(rc);
 }
 
@@ -348,7 +446,7 @@ static int osd_it_acct_load(const struct lu_env *env,
  * move to the first valid record.
  *
  * \param  di   - osd iterator
- * \param  key  - uid or gid
+ * \param  key  - uid or gid or projid
  *
  * \retval +ve  - di points to exact matched key
  * \retval 0    - di points to the first valid record
@@ -404,26 +502,27 @@ const struct dt_index_operations osd_acct_index_ops = {
  * \param osd    - is the osd_device
  * \param uid    - user id of the inode
  * \param gid    - group id of the inode
+ * \param projid - project id of the inode
  * \param space  - how many blocks/inodes will be consumed/released
  * \param oh     - osd transaction handle
- * \param is_blk - block quota or inode quota?
  * \param flags  - if the operation is write, return no user quota, no
  *                  group quota, or sync commit flags to the caller
- * \param force  - set to 1 when changes are performed by root user and thus
- *                  can't failed with EDQUOT
+ * \param osd_qid_declare_flags - indicate this is a inode/block accounting
+ *		    and whether changes are performed by root user
  *
  * \retval 0      - success
  * \retval -ve    - failure
  */
 int osd_declare_quota(const struct lu_env *env, struct osd_device *osd,
-		      qid_t uid, qid_t gid, long long space,
-		      struct osd_thandle *oh, bool is_blk, int *flags,
-		      bool force)
+		      qid_t uid, qid_t gid, qid_t projid, long long space,
+		      struct osd_thandle *oh, int *flags,
+		      enum osd_qid_declare_flags osd_qid_declare_flags)
 {
-	struct osd_thread_info	*info = osd_oti_get(env);
-	struct lquota_id_info	*qi = &info->oti_qi;
-	struct qsd_instance     *qsd = osd->od_quota_slave;
-	int			 rcu, rcg; /* user & group rc */
+	struct osd_thread_info *info = osd_oti_get(env);
+	struct lquota_id_info *qi = &info->oti_qi;
+	struct qsd_instance *qsd = osd->od_quota_slave;
+	int rcu, rcg, rcp = 0; /* user & group & project rc */
+	bool force = !!(osd_qid_declare_flags & OSD_QID_FORCE);
 	ENTRY;
 
 	if (unlikely(qsd == NULL))
@@ -434,9 +533,8 @@ int osd_declare_quota(const struct lu_env *env, struct osd_device *osd,
 	qi->lqi_id.qid_uid = uid;
 	qi->lqi_type       = USRQUOTA;
 	qi->lqi_space      = space;
-	qi->lqi_is_blk     = is_blk;
+	qi->lqi_is_blk     = !!(osd_qid_declare_flags & OSD_QID_BLK);
 	rcu = qsd_op_begin(env, qsd, &oh->ot_quota_trans, qi, flags);
-
 	if (force && (rcu == -EDQUOT || rcu == -EINPROGRESS))
 		/* ignore EDQUOT & EINPROGRESS when changes are done by root */
 		rcu = 0;
@@ -452,10 +550,23 @@ int osd_declare_quota(const struct lu_env *env, struct osd_device *osd,
 	qi->lqi_id.qid_gid = gid;
 	qi->lqi_type       = GRPQUOTA;
 	rcg = qsd_op_begin(env, qsd, &oh->ot_quota_trans, qi, flags);
-
 	if (force && (rcg == -EDQUOT || rcg == -EINPROGRESS))
 		/* as before, ignore EDQUOT & EINPROGRESS for root */
 		rcg = 0;
 
-	RETURN(rcu ? rcu : rcg);
+#ifdef ZFS_PROJINHERIT
+	if (rcg && (rcg != -EDQUOT || flags == NULL))
+		RETURN(rcg);
+
+	/* for project quota */
+	if (osd->od_projectused_dn) {
+		qi->lqi_id.qid_projid = projid;
+		qi->lqi_type = PRJQUOTA;
+		rcp = qsd_op_begin(env, qsd, &oh->ot_quota_trans, qi, flags);
+		if (force && (rcp == -EDQUOT || rcp == -EINPROGRESS))
+			rcp = 0;
+	}
+#endif
+
+	RETURN(rcu ? rcu : (rcg ? rcg : rcp));
 }