X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fosd-zfs%2Fosd_quota.c;h=d9867145facd3a5f40cc95478e1df69d0745d8e5;hb=3222fcd13868a9fcc520af53ae1ff739ba59908f;hp=68e8dd55ae7779b976e5001f0d96f76ce604f13c;hpb=67d925cc206c2d7b215a171fe9d09bd28f9bacb0;p=fs%2Flustre-release.git

diff --git a/lustre/osd-zfs/osd_quota.c b/lustre/osd-zfs/osd_quota.c
index 68e8dd5..d986714 100644
--- a/lustre/osd-zfs/osd_quota.c
+++ b/lustre/osd-zfs/osd_quota.c
@@ -21,15 +21,14 @@
  * GPL HEADER END
  */
 /*
- * Copyright (c) 2011 Whamcloud, Inc.
+ * Copyright (c) 2012, 2015, Intel Corporation.
  * Use is subject to license terms.
  *
  * Author: Johann Lombardi <johann@whamcloud.com>
  */
 
-#include <lquota.h>
+#include <lustre_quota.h>
 #include <obd.h>
-#include "udmu.h"
 #include "osd_internal.h"
 
 /**
@@ -44,6 +43,30 @@ uint64_t osd_quota_fid2dmu(const struct lu_fid *fid)
 }
 
 /**
+ * Helper function to estimate the number of inodes in use for a give uid/gid
+ * from the block usage
+ */
+static uint64_t osd_objset_user_iused(struct osd_device *osd, uint64_t uidbytes)
+{
+	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;
+}
+
+/**
  * Space Accounting Management
  */
 
@@ -58,7 +81,6 @@ 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
@@ -66,12 +88,11 @@ 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 acct_rec		*rec  = (struct acct_rec *)dtrec;
+	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;
@@ -91,31 +112,30 @@ static int osd_acct_index_lookup(const struct lu_env *env,
 	 * 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,
+	rc = -zap_lookup(osd->od_os, oid, buf, sizeof(uint64_t), 1,
 			&rec->bspace);
 	if (rc == -ENOENT)
 		/* user/group has not created anything yet */
 		CDEBUG(D_QUOTA, "%s: id %s not found in DMU accounting ZAP\n",
-			osd->od_dt_dev.dd_lu_dev.ld_obd->obd_name, buf);
+		       osd->od_svname, buf);
 	else if (rc)
 		RETURN(rc);
 
 	if (osd->od_quota_iused_est) {
 		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(+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,
+	rc = -zap_lookup(osd->od_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_dt_dev.dd_lu_dev.ld_obd->obd_name, buf);
+		       osd->od_svname, buf);
 	else if (rc)
 		RETURN(rc);
 
@@ -127,12 +147,10 @@ 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;
@@ -146,14 +164,19 @@ static struct dt_it *osd_it_acct_init(const struct lu_env *env,
 	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));
 
 	/* 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);
@@ -170,10 +193,13 @@ 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);
+
+	osd_zap_cursor_fini(it->oiq_zc);
 	lu_object_put(env, &it->oiq_obj->oo_dt.do_lu);
+	OBD_FREE_PTR(it);
+
 	EXIT;
 }
 
@@ -189,15 +215,16 @@ 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);
+	rc = -zap_cursor_retrieve(it->oiq_zc, za);
 	if (rc == -ENOENT) /* reached the end */
-		RETURN(+1);
+		rc = 1;
 	RETURN(rc);
 }
 
@@ -210,17 +237,16 @@ 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 = -zap_cursor_retrieve(it->oiq_zc, za);
 	if (rc)
 		RETURN(ERR_PTR(rc));
-	it->oiq_id = simple_strtoull(buf, &p, 16);
+	rc = kstrtoull(za->za_name, 16, &it->oiq_id);
+
 	RETURN((struct dt_key *) &it->oiq_id);
 }
 
@@ -236,6 +262,43 @@ 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,
+					 zap_cursor_t *zc,  char *buf,
+					 int buf_size, int *bytes_read)
+{
+	zap_attribute_t *za = &osd_oti_get(env)->oti_za;
+	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;
+	}
+
+	rc = -zap_lookup(zc->zc_objset, zc->zc_zapobj,
+			 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.
  *
@@ -247,9 +310,9 @@ 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 acct_rec		*rec  = (struct acct_rec *)dtrec;
+	struct lquota_acct_rec	*rec  = (struct lquota_acct_rec *)dtrec;
 	struct osd_object	*obj = it->oiq_obj;
 	struct osd_device	*osd = osd_obj2dev(obj);
 	int			 bytes_read;
@@ -260,33 +323,32 @@ 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->oiq_zc,
+					   (char *)&rec->bspace,
+					   sizeof(uint64_t), &bytes_read);
 	if (rc)
 		RETURN(rc);
 
 	if (osd->od_quota_iused_est) {
 		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);
+	rc = -zap_lookup(osd->od_os, it->oiq_obj->oo_db->db_object,
+			 za->za_name, 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_dt_dev.dd_lu_dev.ld_obd->obd_name, buf);
+		       osd->od_svname, za->za_name);
 	else if (rc)
 		RETURN(rc);
 
@@ -304,7 +366,7 @@ 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 +385,25 @@ 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 = -zap_cursor_retrieve(it->oiq_zc, za);
 	if (rc == 0)
-		RETURN(+1);
+		rc = 1;
 	else if (rc == -ENOENT)
-		RETURN(0);
+		rc = 0;
+
 	RETURN(rc);
 }
 
@@ -395,5 +459,67 @@ const struct dt_index_operations osd_acct_index_ops = {
 
 /**
  * Quota Enforcement Management
- * TODO
  */
+
+/*
+ * Wrapper for qsd_op_begin().
+ *
+ * \param env    - the environment passed by the caller
+ * \param osd    - is the osd_device
+ * \param uid    - user id of the inode
+ * \param gid    - group 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
+ *
+ * \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)
+{
+	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 */
+	ENTRY;
+
+	if (unlikely(qsd == NULL))
+		/* quota slave instance hasn't been allocated yet */
+		RETURN(0);
+
+	/* let's start with user quota */
+	qi->lqi_id.qid_uid = uid;
+	qi->lqi_type       = USRQUOTA;
+	qi->lqi_space      = space;
+	qi->lqi_is_blk     = is_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;
+
+	/* For non-fatal error, we want to continue to get the noquota flags
+	 * for group id. This is only for commit write, which has @flags passed
+	 * in. See osd_declare_write_commit().
+	 * When force is set to true, we also want to proceed with the gid */
+	if (rcu && (rcu != -EDQUOT || flags == NULL))
+		RETURN(rcu);
+
+	/* and now group quota */
+	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);
+}