Whamcloud - gitweb
LU-9440 osd-ldiskfs: efficient way to get ldiskfs backend name
[fs/lustre-release.git] / lustre / osd-ldiskfs / osd_oi.c
index 786e85b..cd1db06 100644 (file)
  *
  * You should have received a copy of the GNU General Public License
  * version 2 along with this program; If not, see
- * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
- *
- * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
- * CA 95054 USA or visit www.sun.com if you need additional information or
- * have any questions.
+ * http://www.gnu.org/licenses/gpl-2.0.html
  *
  * GPL HEADER END
  */
@@ -27,7 +23,7 @@
  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2012, 2013, Intel Corporation.
+ * Copyright (c) 2012, 2016, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -40,7 +36,7 @@
  * Author: Nikita Danilov <nikita@clusterfs.com>
  */
 
-#define DEBUG_SUBSYSTEM S_MDS
+#define DEBUG_SUBSYSTEM S_OSD
 
 #include <linux/module.h>
 
@@ -63,9 +59,8 @@
 #include "osd_scrub.h"
 
 static unsigned int osd_oi_count = OSD_OI_FID_NR;
-CFS_MODULE_PARM(osd_oi_count, "i", int, 0444,
-                "Number of Object Index containers to be created, "
-                "it's only valid for new filesystem.");
+module_param(osd_oi_count, int, 0444);
+MODULE_PARM_DESC(osd_oi_count, "Number of Object Index containers to be created, it's only valid for new filesystem.");
 
 /** to serialize concurrent OI index initialization */
 static struct mutex oi_init_lock;
@@ -118,7 +113,7 @@ static int osd_oi_index_create_one(struct osd_thread_info *info,
 
        dentry = osd_child_dentry_by_inode(env, dir, name, strlen(name));
        bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, NULL);
-       if (bh) {
+       if (!IS_ERR(bh)) {
                osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
                brelse(bh);
                inode = osd_iget(info, osd, id);
@@ -139,6 +134,9 @@ static int osd_oi_index_create_one(struct osd_thread_info *info,
                return PTR_ERR(inode);
        }
 
+       ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
+       unlock_new_inode(inode);
+
        if (feat->dif_flags & DT_IND_VARKEY)
                rc = iam_lvar_create(inode, feat->dif_keysize_max,
                                     feat->dif_ptrsize, feat->dif_recsize_max,
@@ -148,7 +146,7 @@ static int osd_oi_index_create_one(struct osd_thread_info *info,
                                     feat->dif_ptrsize, feat->dif_recsize_max,
                                     jh);
        dentry = osd_child_dentry_by_inode(env, dir, name, strlen(name));
-       rc = osd_ldiskfs_add_entry(jh, dentry, inode, NULL);
+       rc = osd_ldiskfs_add_entry(info, osd, jh, dentry, inode, NULL);
        ldiskfs_journal_stop(jh);
        iput(inode);
        return rc;
@@ -228,7 +226,6 @@ static int osd_oi_open(struct osd_thread_info *info, struct osd_device *osd,
         if (IS_ERR(inode))
                 RETURN(PTR_ERR(inode));
 
-       ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI);
        /* 'What the @fid is' is not imporatant, because these objects
         * have no OI mappings, and only are visible inside the OSD.*/
        lu_igif_build(&info->oti_fid, inode->i_ino, inode->i_generation);
@@ -319,12 +316,11 @@ osd_oi_table_open(struct osd_thread_info *info, struct osd_device *osd,
                        continue;
                }
 
-               CERROR("%.16s: can't open %s: rc = %d\n",
-                      LDISKFS_SB(osd_sb(osd))->s_es->s_volume_name, name, rc);
+               CERROR("%s: can't open %s: rc = %d\n",
+                      osd_dev2name(osd), name, rc);
                if (oi_count > 0)
-                       CERROR("%.16s: expect to open total %d OI files.\n",
-                              LDISKFS_SB(osd_sb(osd))->s_es->s_volume_name,
-                              oi_count);
+                       CERROR("%s: expect to open total %d OI files.\n",
+                              osd_dev2name(osd), oi_count);
                break;
        }
 
@@ -336,7 +332,52 @@ osd_oi_table_open(struct osd_thread_info *info, struct osd_device *osd,
        RETURN(count);
 }
 
-int osd_oi_init(struct osd_thread_info *info, struct osd_device *osd)
+static int osd_remove_oi_one(struct dentry *parent, const char *name,
+                            int namelen)
+{
+       struct dentry *child;
+       int rc;
+
+       child = ll_lookup_one_len(name, parent, namelen);
+       if (IS_ERR(child)) {
+               rc = PTR_ERR(child);
+       } else {
+               rc = ll_vfs_unlink(parent->d_inode, child);
+               dput(child);
+       }
+
+       return rc == -ENOENT ? 0 : rc;
+}
+
+static int osd_remove_ois(struct osd_thread_info *info, struct osd_device *osd)
+{
+       char name[16];
+       int namelen;
+       int rc;
+       int i;
+
+       for (i = 0; i < osd->od_scrub.os_file.sf_oi_count; i++) {
+               namelen = snprintf(name, sizeof(name), "%s.%d",
+                                  OSD_OI_NAME_BASE, i);
+               rc = osd_remove_oi_one(osd_sb(osd)->s_root, name, namelen);
+               if (rc != 0) {
+                       CERROR("%s: fail to remove the stale OI file %s: "
+                              "rc = %d\n", osd_dev2name(osd), name, rc);
+                       return rc;
+               }
+       }
+
+       namelen = snprintf(name, sizeof(name), "%s", OSD_OI_NAME_BASE);
+       rc = osd_remove_oi_one(osd_sb(osd)->s_root, name, namelen);
+       if (rc != 0)
+               CERROR("%s: fail to remove the stale OI file %s: rc = %d\n",
+                      osd_dev2name(osd), name, rc);
+
+       return rc;
+}
+
+int osd_oi_init(struct osd_thread_info *info, struct osd_device *osd,
+               bool restored)
 {
        struct osd_scrub  *scrub = &osd->od_scrub;
        struct scrub_file *sf = &scrub->os_file;
@@ -344,6 +385,12 @@ int osd_oi_init(struct osd_thread_info *info, struct osd_device *osd)
        int                rc;
        ENTRY;
 
+       if (restored) {
+               rc = osd_remove_ois(info, osd);
+               if (rc != 0)
+                       return rc;
+       }
+
        OBD_ALLOC(oi, sizeof(*oi) * OSD_OI_FID_NR_MAX);
        if (oi == NULL)
                RETURN(-ENOMEM);
@@ -368,6 +415,7 @@ int osd_oi_init(struct osd_thread_info *info, struct osd_device *osd)
        /* if previous failed then try found single OI from old filesystem */
        rc = osd_oi_open(info, osd, OSD_OI_NAME_BASE, &oi[0], false);
        if (rc == 0) { /* found single OI from old filesystem */
+               ldiskfs_clear_bit(0, sf->sf_oi_bitmap);
                if (sf->sf_success_count == 0)
                        /* XXX: There is one corner case that if the OI_scrub
                         *      file crashed or lost and we regard it upgrade,
@@ -384,9 +432,8 @@ int osd_oi_init(struct osd_thread_info *info, struct osd_device *osd)
                                        SF_UPGRADE);
                GOTO(out, rc = 1);
        } else if (rc != -ENOENT) {
-               CERROR("%.16s: can't open %s: rc = %d\n",
-                      LDISKFS_SB(osd_sb(osd))->s_es->s_volume_name,
-                      OSD_OI_NAME_BASE, rc);
+               CERROR("%s: can't open %s: rc = %d\n",
+                      osd_dev2name(osd), OSD_OI_NAME_BASE, rc);
                GOTO(out, rc);
        }
 
@@ -511,11 +558,23 @@ int fid_is_on_ost(struct osd_thread_info *info, struct osd_device *osd,
        if (!(flags & OI_CHECK_FLD))
                RETURN(0);
 
+       if (osd_seq_site(osd)->ss_server_fld == NULL)
+               RETURN(0);
+
        rc = osd_fld_lookup(info->oti_env, osd, fid_seq(fid), range);
        if (rc != 0) {
-               CERROR("%s: "DFID" lookup failed: rc = %d\n", osd_name(osd),
-                      PFID(fid), rc);
-               RETURN(rc);
+               /* During upgrade, OST FLDB might not be loaded because
+                * OST FLDB is not created until 2.6, so if some DNE
+                * filesystem upgrade from 2.5 to 2.7/2.8, they will
+                * not be able to find the sequence from local FLDB
+                * cache see fld_index_init(). */
+               if (rc == -ENOENT && osd->od_is_ost)
+                       RETURN(1);
+
+               if (rc != -ENOENT)
+                       CERROR("%s: lookup FLD "DFID": rc = %d\n",
+                              osd_name(osd), PFID(fid), rc);
+               RETURN(0);
        }
 
        if (fld_range_is_ost(range))
@@ -552,14 +611,23 @@ int osd_oi_lookup(struct osd_thread_info *info, struct osd_device *osd,
        if (fid_is_on_ost(info, osd, fid, flags) || fid_is_llog(fid))
                return osd_obj_map_lookup(info, osd, fid, id);
 
-       if (fid_is_fs_root(fid)) {
-               osd_id_gen(id, osd_sb(osd)->s_root->d_inode->i_ino,
-                          osd_sb(osd)->s_root->d_inode->i_generation);
-               return 0;
-       }
 
-       if (unlikely(fid_is_acct(fid)))
-               return osd_acct_obj_lookup(info, osd, fid, id);
+       if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
+               int rc;
+               if (fid_is_fs_root(fid)) {
+                       osd_id_gen(id, osd_sb(osd)->s_root->d_inode->i_ino,
+                                  osd_sb(osd)->s_root->d_inode->i_generation);
+                       return 0;
+               }
+               if (unlikely(fid_is_acct(fid)))
+                       return osd_acct_obj_lookup(info, osd, fid, id);
+
+               /* For other special FIDs, try OI first, then do spec lookup */
+               rc = __osd_oi_lookup(info, osd, fid, id);
+               if (rc == -ENOENT)
+                       return osd_obj_spec_lookup(info, osd, fid, id);
+               return rc;
+       }
 
        if (!osd->od_igif_inoi && fid_is_igif(fid)) {
                osd_id_gen(id, lu_igif_ino(fid), lu_igif_gen(fid));
@@ -571,11 +639,10 @@ int osd_oi_lookup(struct osd_thread_info *info, struct osd_device *osd,
 
 static int osd_oi_iam_refresh(struct osd_thread_info *oti, struct osd_oi *oi,
                             const struct dt_rec *rec, const struct dt_key *key,
-                            struct thandle *th, bool insert)
+                            handle_t *th, bool insert)
 {
        struct iam_container    *bag;
        struct iam_path_descr   *ipd;
-       struct osd_thandle      *oh;
        int                     rc;
        ENTRY;
 
@@ -588,14 +655,13 @@ static int osd_oi_iam_refresh(struct osd_thread_info *oti, struct osd_oi *oi,
        if (unlikely(ipd == NULL))
                RETURN(-ENOMEM);
 
-       oh = container_of0(th, struct osd_thandle, ot_super);
-       LASSERT(oh->ot_handle != NULL);
-       LASSERT(oh->ot_handle->h_transaction != NULL);
+       LASSERT(th != NULL);
+       LASSERT(th->h_transaction != NULL);
        if (insert)
-               rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
+               rc = iam_insert(th, bag, (const struct iam_key *)key,
                                (const struct iam_rec *)rec, ipd);
        else
-               rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)key,
+               rc = iam_update(th, bag, (const struct iam_key *)key,
                                (const struct iam_rec *)rec, ipd);
        osd_ipd_put(oti->oti_env, bag, ipd);
        LINVRNT(osd_invariant(obj));
@@ -604,7 +670,7 @@ static int osd_oi_iam_refresh(struct osd_thread_info *oti, struct osd_oi *oi,
 
 int osd_oi_insert(struct osd_thread_info *info, struct osd_device *osd,
                  const struct lu_fid *fid, const struct osd_inode_id *id,
-                 struct thandle *th, enum oi_check_flags flags)
+                 handle_t *th, enum oi_check_flags flags, bool *exist)
 {
        struct lu_fid       *oi_fid = &info->oti_fid2;
        struct osd_inode_id *oi_id  = &info->oti_id2;
@@ -623,7 +689,7 @@ int osd_oi_insert(struct osd_thread_info *info, struct osd_device *osd,
                               (const struct dt_key *)oi_fid, th, true);
        if (rc != 0) {
                struct inode *inode;
-               struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
+               struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
 
                if (rc != -EEXIST)
                        return rc;
@@ -633,7 +699,7 @@ int osd_oi_insert(struct osd_thread_info *info, struct osd_device *osd,
                        return rc;
 
                if (unlikely(osd_id_eq(id, oi_id)))
-                       return 0;
+                       return 1;
 
                /* Check whether the mapping for oi_id is valid or not. */
                inode = osd_iget(info, osd, oi_id);
@@ -644,7 +710,15 @@ int osd_oi_insert(struct osd_thread_info *info, struct osd_device *osd,
                        return rc;
                }
 
-               rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
+               /* The EA inode should NOT be in OI, old OI scrub may added
+                * such OI mapping by wrong, replace it. */
+               if (unlikely(osd_is_ea_inode(inode))) {
+                       iput(inode);
+                       goto update;
+               }
+
+               rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
+                                &info->oti_ost_attrs);
                iput(inode);
                if (rc == -ENODATA)
                        goto update;
@@ -654,9 +728,8 @@ int osd_oi_insert(struct osd_thread_info *info, struct osd_device *osd,
 
                if (!(lma->lma_compat & LMAC_NOT_IN_OI) &&
                    lu_fid_eq(fid, &lma->lma_self_fid)) {
-                       CERROR("%.16s: the FID "DFID" is used by two objects: "
-                              "%u/%u %u/%u\n",
-                              LDISKFS_SB(osd_sb(osd))->s_es->s_volume_name,
+                       CERROR("%s: the FID "DFID" is used by two objects: "
+                              "%u/%u %u/%u\n", osd_dev2name(osd),
                               PFID(fid), oi_id->oii_ino, oi_id->oii_gen,
                               id->oii_ino, id->oii_gen);
                        return -EEXIST;
@@ -669,6 +742,9 @@ update:
                                        (const struct dt_key *)oi_fid, th, false);
                if (rc != 0)
                        return rc;
+
+               if (exist != NULL)
+                       *exist = true;
        }
 
        if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE))
@@ -677,36 +753,34 @@ update:
 }
 
 static int osd_oi_iam_delete(struct osd_thread_info *oti, struct osd_oi *oi,
-                             const struct dt_key *key, struct thandle *handle)
+                            const struct dt_key *key, handle_t *th)
 {
-        struct iam_container  *bag;
-        struct iam_path_descr *ipd;
-        struct osd_thandle    *oh;
-        int                    rc;
-        ENTRY;
+       struct iam_container    *bag;
+       struct iam_path_descr   *ipd;
+       int                      rc;
+       ENTRY;
 
-        LASSERT(oi);
+       LASSERT(oi);
        LASSERT(oi->oi_inode);
        ll_vfs_dq_init(oi->oi_inode);
 
-        bag = &oi->oi_dir.od_container;
-        ipd = osd_idx_ipd_get(oti->oti_env, bag);
-        if (unlikely(ipd == NULL))
-                RETURN(-ENOMEM);
+       bag = &oi->oi_dir.od_container;
+       ipd = osd_idx_ipd_get(oti->oti_env, bag);
+       if (unlikely(ipd == NULL))
+               RETURN(-ENOMEM);
 
-        oh = container_of0(handle, struct osd_thandle, ot_super);
-        LASSERT(oh->ot_handle != NULL);
-        LASSERT(oh->ot_handle->h_transaction != NULL);
+       LASSERT(th != NULL);
+       LASSERT(th->h_transaction != NULL);
 
-        rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
-        osd_ipd_put(oti->oti_env, bag, ipd);
-        LINVRNT(osd_invariant(obj));
-        RETURN(rc);
+       rc = iam_delete(th, bag, (const struct iam_key *)key, ipd);
+       osd_ipd_put(oti->oti_env, bag, ipd);
+       LINVRNT(osd_invariant(obj));
+       RETURN(rc);
 }
 
 int osd_oi_delete(struct osd_thread_info *info,
                  struct osd_device *osd, const struct lu_fid *fid,
-                 struct thandle *th, enum oi_check_flags flags)
+                 handle_t *th, enum oi_check_flags flags)
 {
        struct lu_fid *oi_fid = &info->oti_fid2;
 
@@ -727,7 +801,7 @@ int osd_oi_delete(struct osd_thread_info *info,
 
 int osd_oi_update(struct osd_thread_info *info, struct osd_device *osd,
                  const struct lu_fid *fid, const struct osd_inode_id *id,
-                 struct thandle *th, enum oi_check_flags flags)
+                 handle_t *th, enum oi_check_flags flags)
 {
        struct lu_fid       *oi_fid = &info->oti_fid2;
        struct osd_inode_id *oi_id  = &info->oti_id2;