Whamcloud - gitweb
LU-17744 ldiskfs: mballoc stats fixes
[fs/lustre-release.git] / lustre / ofd / ofd_objects.c
index bb92058..a3cdad5 100644 (file)
@@ -27,7 +27,6 @@
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
- * Lustre is a trademark of Sun Microsystems, Inc.
  *
  * lustre/ofd/ofd_objects.c
  *
@@ -42,6 +41,7 @@
 
 #include <dt_object.h>
 #include <lustre_lfsck.h>
+#include <lustre_export.h>
 
 #include "ofd_internal.h"
 
@@ -64,8 +64,6 @@ static int ofd_version_get_check(struct ofd_thread_info *info,
 {
        dt_obj_version_t curr_version;
 
-       LASSERT(ofd_object_exists(fo));
-
        if (info->fti_exp == NULL)
                RETURN(0);
 
@@ -147,6 +145,20 @@ int ofd_object_ff_load(const struct lu_env *env, struct ofd_object *fo)
        buf->lb_buf = ff;
        buf->lb_len = sizeof(*ff);
        rc = dt_xattr_get(env, ofd_object_child(fo), buf, XATTR_NAME_FID);
+       if (rc == -ERANGE) {
+               struct filter_fid *ff_new;
+
+               OBD_ALLOC(ff_new, sizeof(*ff) + FILTER_FID_EXTRA_SIZE);
+               if (!ff_new)
+                       return -ENOMEM;
+               buf->lb_buf = ff_new;
+               buf->lb_len = sizeof(*ff) + FILTER_FID_EXTRA_SIZE;
+               rc = dt_xattr_get(env, ofd_object_child(fo), buf,
+                                 XATTR_NAME_FID);
+               if (rc > 0)
+                       memcpy(ff, ff_new, sizeof(*ff));
+               OBD_FREE(ff_new, sizeof(*ff) + FILTER_FID_EXTRA_SIZE);
+       }
        if (rc < 0)
                return rc;
 
@@ -195,7 +207,7 @@ static int ofd_precreate_cb_add(const struct lu_env *env, struct thandle *th,
                return -ENOMEM;
 
        precreate = atomic_read(&oseq->os_precreate_in_progress);
-       atomic_inc(&oseq->os_refc);
+       refcount_inc(&oseq->os_refc);
        opc->opc_oseq = oseq;
        opc->opc_objects = objects;
        CDEBUG(D_OTHER, "Add %d to %d for "DFID", th_sync %d\n",
@@ -208,7 +220,7 @@ static int ofd_precreate_cb_add(const struct lu_env *env, struct thandle *th,
        dcb = &opc->opc_cb;
        dcb->dcb_func = ofd_cb_precreate;
        INIT_LIST_HEAD(&dcb->dcb_linkage);
-       strlcpy(dcb->dcb_name, "ofd_cb_precreate", sizeof(dcb->dcb_name));
+       strscpy(dcb->dcb_name, "ofd_cb_precreate", sizeof(dcb->dcb_name));
 
        rc = dt_trans_cb_add(th, dcb);
        if (rc) {
@@ -239,18 +251,20 @@ static int ofd_precreate_cb_add(const struct lu_env *env, struct thandle *th,
  * update the inode. The ctime = 0 case is also handled specially in
  * osd_inode_setattr(). See LU-221, LU-1042 for details.
  *
- * \param[in] env      execution environment
- * \param[in] ofd      OFD device
- * \param[in] id       object ID to start precreation from
- * \param[in] oseq     object sequence
- * \param[in] nr       number of objects to precreate
- * \param[in] sync     synchronous precreation flag
+ * \param[in] env              execution environment
+ * \param[in] ofd              OFD device
+ * \param[in] id               object ID to start precreation from
+ * \param[in] oseq             object sequence
+ * \param[in] nr               number of objects to precreate
+ * \param[in] sync             synchronous precreation flag
+ * \param[in] trans_local      start local transaction
  *
  * \retval             0 if successful
  * \retval             negative value on error
  */
 int ofd_precreate_objects(const struct lu_env *env, struct ofd_device *ofd,
-                         u64 id, struct ofd_seq *oseq, int nr, int sync)
+                         u64 id, struct ofd_seq *oseq, int nr, int sync,
+                         bool trans_local)
 {
        struct ofd_thread_info  *info = ofd_info(env);
        struct ofd_object       *fo = NULL;
@@ -267,14 +281,17 @@ int ofd_precreate_objects(const struct lu_env *env, struct ofd_device *ofd,
 
        ENTRY;
 
-       /* Don't create objects beyond the valid range for this SEQ */
+       /* Don't create objects beyond the valid range for this SEQ
+        * Last object to create is (id + nr - 1), but we move -1 on LHS
+        * to +1 on RHS to evaluate constant at compile time. LU-11186
+        */
        if (unlikely(fid_seq_is_mdt0(ostid_seq(&oseq->os_oi)) &&
-                    (id + nr) > IDIF_MAX_OID)) {
+                    id + nr > IDIF_MAX_OID + 1)) {
                CERROR("%s:"DOSTID" hit the IDIF_MAX_OID (1<<48)!\n",
                       ofd_name(ofd), id, ostid_seq(&oseq->os_oi));
                RETURN(rc = -ENOSPC);
        } else if (unlikely(!fid_seq_is_mdt0(ostid_seq(&oseq->os_oi)) &&
-                           (id + nr) > OBIF_MAX_OID)) {
+                           id + nr > OBIF_MAX_OID + 1)) {
                CERROR("%s:"DOSTID" hit the OBIF_MAX_OID (1<<32)!\n",
                       ofd_name(ofd), id, ostid_seq(&oseq->os_oi));
                RETURN(rc = -ENOSPC);
@@ -359,7 +376,11 @@ int ofd_precreate_objects(const struct lu_env *env, struct ofd_device *ofd,
                }
        }
 
-       rc = dt_trans_start_local(env, ofd->ofd_osd, th);
+       /* Only needed for MDS+OSS rolling upgrade interop with 2.16+older. */
+       if (unlikely(trans_local))
+               rc = dt_trans_start_local(env, ofd->ofd_osd, th);
+       else
+               rc = dt_trans_start(env, ofd->ofd_osd, th);
        if (rc)
                GOTO(trans_stop, rc);
 
@@ -379,7 +400,7 @@ int ofd_precreate_objects(const struct lu_env *env, struct ofd_device *ofd,
          * To make above mechanism to work, before OFD pre-create OST-objects,
          * it needs to update the LAST_ID file firstly, otherwise, the LFSCK
          * may cannot get latest last_id although new OST-object created. */
-       if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_SKIP_LASTID)) {
+       if (!CFS_FAIL_CHECK(OBD_FAIL_LFSCK_SKIP_LASTID)) {
                tmp = cpu_to_le64(id + nr - 1);
                dt_write_lock(env, oseq->os_lastid_obj, DT_LASTID);
                rc = dt_record_write(env, oseq->os_lastid_obj,
@@ -405,7 +426,7 @@ int ofd_precreate_objects(const struct lu_env *env, struct ofd_device *ofd,
                }
 
                if (likely(!ofd_object_exists(fo) &&
-                          !OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING))) {
+                          !CFS_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING))) {
                        next = ofd_object_child(fo);
                        LASSERT(next != NULL);
 
@@ -595,15 +616,17 @@ int ofd_object_ff_update(const struct lu_env *env, struct ofd_object *fo,
                       PFID(lu_object_fid(&fo->ofo_obj.do_lu)),
                       ff->ff_layout_version, oa->o_layout_version);
 
-               /* only the MDS has the authority to update layout version */
-               if (!(exp_connect_flags(ofd_info(env)->fti_exp) &
-                     OBD_CONNECT_MDS)) {
-                       CERROR(DFID": update layout version from client\n",
-                              PFID(&fo->ofo_ff.ff_parent));
-
-                       RETURN(-EPERM);
-               }
-
+               /**
+                * resync write from client on non-primary objects and
+                * resync start from MDS on primary objects will contain
+                * LU_LAYOUT_RESYNC flag in the @oa.
+                *
+                * The layout version checking for write/punch from client
+                * happens in ofd_verify_layout_version() before coming to
+                * here, so that resync with smaller layout version client
+                * will be rejected there, the biggest resync version will
+                * be recorded in the OFD objects.
+                */
                if (ff->ff_layout_version & LU_LAYOUT_RESYNC) {
                        /* this opens a new era of writing */
                        ff->ff_layout_version = 0;
@@ -611,7 +634,8 @@ int ofd_object_ff_update(const struct lu_env *env, struct ofd_object *fo,
                }
 
                /* it's not allowed to change it to a smaller value */
-               if (oa->o_layout_version < ff->ff_layout_version)
+               if (ofd_layout_version_less(oa->o_layout_version,
+                                           ff->ff_layout_version))
                        RETURN(-EINVAL);
 
                if (ff->ff_layout_version == 0 ||
@@ -711,11 +735,11 @@ int ofd_attr_set(const struct lu_env *env, struct ofd_object *fo,
                GOTO(unlock, rc = fl);
 
        if (fl) {
-               if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR1))
+               if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR1))
                        ff->ff_parent.f_oid = cpu_to_le32(1UL << 31);
-               else if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR2))
+               else if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR2))
                        le32_add_cpu(&ff->ff_parent.f_oid, -1);
-               else if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NOPFID))
+               else if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_NOPFID))
                        GOTO(unlock, rc);
 
                info->fti_buf.lb_buf = ff;
@@ -759,8 +783,8 @@ out:
  * \retval             negative value on error
  */
 int ofd_object_fallocate(const struct lu_env *env, struct ofd_object *fo,
-                       __u64 start, __u64 end, int mode, struct lu_attr *la,
-                       struct obdo *oa)
+                        __u64 start, __u64 end, int mode, struct lu_attr *la,
+                        struct obdo *oa)
 {
        struct ofd_thread_info *info = ofd_info(env);
        struct ofd_device *ofd = ofd_obj2dev(fo);
@@ -772,46 +796,76 @@ int ofd_object_fallocate(const struct lu_env *env, struct ofd_object *fo,
 
        ENTRY;
 
-       ofd_write_lock(env, fo);
        if (!ofd_object_exists(fo))
-               GOTO(unlock, rc = -ENOENT);
+               RETURN(-ENOENT);
 
        /* VBR: version recovery check */
        rc = ofd_version_get_check(info, fo);
        if (rc != 0)
-               GOTO(unlock, rc);
+               RETURN(rc);
 
        if (ff != NULL) {
                rc = ofd_object_ff_load(env, fo);
                if (rc == -ENODATA)
                        ff_needed = true;
                else if (rc < 0)
-                       GOTO(unlock, rc);
+                       RETURN(rc);
+
+               if (ff_needed) {
+                       if (oa->o_valid & OBD_MD_FLFID) {
+                               ff->ff_parent.f_seq = oa->o_parent_seq;
+                               ff->ff_parent.f_oid = oa->o_parent_oid;
+                               ff->ff_parent.f_ver = oa->o_stripe_idx;
+                       }
+                       if (oa->o_valid & OBD_MD_FLOSTLAYOUT)
+                               ff->ff_layout = oa->o_layout;
+                       if (oa->o_valid & OBD_MD_LAYOUT_VERSION)
+                               ff->ff_layout_version = oa->o_layout_version;
+                       filter_fid_cpu_to_le(ff, ff, sizeof(*ff));
+               }
        }
 
        th = ofd_trans_create(env, ofd);
        if (IS_ERR(th))
-               GOTO(unlock, rc = PTR_ERR(th));
+               RETURN(PTR_ERR(th));
 
        rc = dt_declare_attr_set(env, dob, la, th);
        if (rc)
                GOTO(stop, rc);
 
-       rc = dt_declare_falloc(env, dob, th);
+       rc = dt_declare_fallocate(env, dob, start, end, mode, th);
        if (rc)
                GOTO(stop, rc);
 
+       if (ff_needed) {
+               info->fti_buf.lb_buf = ff;
+               info->fti_buf.lb_len = sizeof(*ff);
+               rc = dt_declare_xattr_set(env, ofd_object_child(fo),
+                                         &info->fti_buf, XATTR_NAME_FID, 0,
+                                         th);
+               if (rc)
+                       GOTO(stop, rc);
+       }
+
        rc = ofd_trans_start(env, ofd, fo, th);
        if (rc)
                GOTO(stop, rc);
 
+       ofd_read_lock(env, fo);
+       if (!ofd_object_exists(fo))
+               GOTO(unlock, rc = -ENOENT);
+
+       if (la->la_valid & (LA_ATIME | LA_MTIME | LA_CTIME))
+               tgt_fmd_update(info->fti_exp, &fo->ofo_header.loh_fid,
+                              info->fti_xid);
+
        rc = dt_falloc(env, dob, start, end, mode, th);
        if (rc)
-               GOTO(stop, rc);
+               GOTO(unlock, rc);
 
        rc = dt_attr_set(env, dob, la, th);
        if (rc)
-               GOTO(stop, rc);
+               GOTO(unlock, rc);
 
        if (ff_needed) {
                rc = dt_xattr_set(env, ofd_object_child(fo), &info->fti_buf,
@@ -819,10 +873,10 @@ int ofd_object_fallocate(const struct lu_env *env, struct ofd_object *fo,
                if (!rc)
                        filter_fid_le_to_cpu(&fo->ofo_ff, ff, sizeof(*ff));
        }
+unlock:
+       ofd_read_unlock(env, fo);
 stop:
        ofd_trans_stop(env, ofd, th, rc);
-unlock:
-       ofd_write_unlock(env, fo);
        RETURN(rc);
 }
 
@@ -882,6 +936,11 @@ int ofd_object_punch(const struct lu_env *env, struct ofd_object *fo,
        if (IS_ERR(th))
                GOTO(out, rc = PTR_ERR(th));
 
+       if (oa->o_valid & OBD_MD_FLFLAGS && oa->o_flags & LUSTRE_ENCRYPT_FL) {
+               /* punch must be aware we are dealing with an encrypted file */
+               la->la_valid |= LA_FLAGS;
+               la->la_flags |= LUSTRE_ENCRYPT_FL;
+       }
        rc = dt_declare_attr_set(env, dob, la, th);
        if (rc)
                GOTO(stop, rc);
@@ -915,8 +974,6 @@ int ofd_object_punch(const struct lu_env *env, struct ofd_object *fo,
                rc = ofd_verify_layout_version(env, fo, oa);
                if (rc)
                        GOTO(unlock, rc);
-
-               oa->o_valid &= ~OBD_MD_LAYOUT_VERSION;
        }
 
        rc = dt_punch(env, dob, start, OBD_OBJECT_EOF, th);
@@ -932,11 +989,11 @@ int ofd_object_punch(const struct lu_env *env, struct ofd_object *fo,
                GOTO(unlock, rc);
 
        if (fl) {
-               if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR1))
+               if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR1))
                        ff->ff_parent.f_oid = cpu_to_le32(1UL << 31);
-               else if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR2))
+               else if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR2))
                        le32_add_cpu(&ff->ff_parent.f_oid, -1);
-               else if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NOPFID))
+               else if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_NOPFID))
                        GOTO(unlock, rc);
 
                info->fti_buf.lb_buf = ff;
@@ -1010,14 +1067,14 @@ int ofd_destroy(const struct lu_env *env, struct ofd_object *fo,
 
        ofd_write_lock(env, fo);
        if (!ofd_object_exists(fo))
-               GOTO(stop, rc = -ENOENT);
+               GOTO(unlock, rc = -ENOENT);
 
        tgt_fmd_drop(ofd_info(env)->fti_exp, &fo->ofo_header.loh_fid);
 
        dt_ref_del(env, ofd_object_child(fo), th);
        dt_destroy(env, ofd_object_child(fo), th);
+unlock:
        ofd_write_unlock(env, fo);
-
 stop:
        rc2 = ofd_trans_stop(env, ofd, th, rc);
        if (rc2)