Whamcloud - gitweb
LU-1866 osd: ancillary work for initial OI scrub
[fs/lustre-release.git] / lustre / osd-ldiskfs / osd_oi.c
index d13a0fa..bb3efa4 100644 (file)
@@ -27,7 +27,7 @@
  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2012, Whamcloud, Inc.
+ * Copyright (c) 2011, 2012, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -69,7 +69,7 @@
 #include "osd_oi.h"
 /* osd_lookup(), struct osd_thread_info */
 #include "osd_internal.h"
-#include "osd_igif.h"
+#include "osd_scrub.h"
 
 static unsigned int osd_oi_count = OSD_OI_FID_NR;
 CFS_MODULE_PARM(osd_oi_count, "i", int, 0444,
@@ -77,7 +77,7 @@ CFS_MODULE_PARM(osd_oi_count, "i", int, 0444,
                 "it's only valid for new filesystem.");
 
 /** to serialize concurrent OI index initialization */
-static cfs_mutex_t oi_init_lock;
+static struct mutex oi_init_lock;
 
 static struct dt_index_features oi_feat = {
         .dif_flags       = DT_IND_UPDATE,
@@ -89,22 +89,25 @@ static struct dt_index_features oi_feat = {
 #define OSD_OI_NAME_BASE        "oi.16"
 
 static void osd_oi_table_put(struct osd_thread_info *info,
-                             struct osd_oi **oi_table, unsigned oi_count)
+                            struct osd_oi **oi_table, unsigned oi_count)
 {
-        struct iam_container *bag;
-        int                   i;
-
-        for (i = 0; i < oi_count; i++) {
-                LASSERT(oi_table[i] != NULL);
-                LASSERT(oi_table[i]->oi_inode != NULL);
-
-                bag = &(oi_table[i]->oi_dir.od_container);
-                if (bag->ic_object == oi_table[i]->oi_inode)
-                        iam_container_fini(bag);
-                iput(oi_table[i]->oi_inode);
-                oi_table[i]->oi_inode = NULL;
-                OBD_FREE_PTR(oi_table[i]);
-        }
+       struct iam_container *bag;
+       int                   i;
+
+       for (i = 0; i < oi_count; i++) {
+               if (oi_table[i] == NULL)
+                       continue;
+
+               LASSERT(oi_table[i]->oi_inode != NULL);
+
+               bag = &(oi_table[i]->oi_dir.od_container);
+               if (bag->ic_object == oi_table[i]->oi_inode)
+                       iam_container_fini(bag);
+               iput(oi_table[i]->oi_inode);
+               oi_table[i]->oi_inode = NULL;
+               OBD_FREE_PTR(oi_table[i]);
+               oi_table[i] = NULL;
+       }
 }
 
 static int osd_oi_index_create_one(struct osd_thread_info *info,
@@ -234,6 +237,7 @@ 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);
         OBD_ALLOC_PTR(oi);
         if (oi == NULL)
                 GOTO(out_inode, rc = -ENOMEM);
@@ -281,93 +285,143 @@ out_inode:
  */
 static int
 osd_oi_table_open(struct osd_thread_info *info, struct osd_device *osd,
-                  struct osd_oi **oi_table, unsigned oi_count, bool create)
+                 struct osd_oi **oi_table, unsigned oi_count, bool create)
 {
-        struct dt_device *dev = &osd->od_dt_dev;
-        int               count = 0;
-        int               rc = 0;
-        int               i;
-
-        /* NB: oi_count != 0 means that we have already created/known all OIs
-         * and have known exact number of OIs. */
-        LASSERT(oi_count <= OSD_OI_FID_NR_MAX);
-
-        for (i = 0; i < (oi_count != 0 ? oi_count : OSD_OI_FID_NR_MAX); i++) {
-                char name[12];
-
-                sprintf(name, "%s.%d", OSD_OI_NAME_BASE, i);
-                rc = osd_oi_open(info, osd, name, &oi_table[i], create);
-                if (rc == 0) {
-                        count++;
-                        continue;
-                }
-
-                if (rc == -ENOENT && oi_count == 0)
-                        return count;
-
-                CERROR("%s: can't open %s: rc = %d\n",
-                       dev->dd_lu_dev.ld_obd->obd_name, name, rc);
-                if (oi_count > 0) {
-                        CERROR("%s: expect to open total %d OI files.\n",
-                               dev->dd_lu_dev.ld_obd->obd_name, oi_count);
-                }
-                break;
-        }
+       struct scrub_file *sf = &osd->od_scrub.os_file;
+       int                count = 0;
+       int                rc = 0;
+       int                i;
+       ENTRY;
+
+       /* NB: oi_count != 0 means that we have already created/known all OIs
+        * and have known exact number of OIs. */
+       LASSERT(oi_count <= OSD_OI_FID_NR_MAX);
+
+       for (i = 0; i < (oi_count != 0 ? oi_count : OSD_OI_FID_NR_MAX); i++) {
+               char name[12];
+
+               if (oi_table[i] != NULL) {
+                       count++;
+                       continue;
+               }
 
-        if (rc < 0) {
-                osd_oi_table_put(info, oi_table, count);
-                return rc;
-        }
+               sprintf(name, "%s.%d", OSD_OI_NAME_BASE, i);
+               rc = osd_oi_open(info, osd, name, &oi_table[i], create);
+               if (rc == 0) {
+                       count++;
+                       continue;
+               }
+
+               if (rc == -ENOENT && create == false) {
+                       if (oi_count == 0)
+                               return count;
+
+                       rc = 0;
+                       ldiskfs_set_bit(i, sf->sf_oi_bitmap);
+                       continue;
+               }
+
+               CERROR("%.16s: can't open %s: rc = %d\n",
+                      LDISKFS_SB(osd_sb(osd))->s_es->s_volume_name, 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);
+               break;
+       }
 
-        return count;
+       if (rc < 0) {
+               osd_oi_table_put(info, oi_table, oi_count > 0 ? oi_count : i);
+               count = rc;
+       }
+
+       RETURN(count);
 }
 
 int osd_oi_init(struct osd_thread_info *info, struct osd_device *osd)
 {
-        struct dt_device *dev = &osd->od_dt_dev;
-        struct osd_oi   **oi;
-        int               rc;
+       struct osd_scrub  *scrub = &osd->od_scrub;
+       struct scrub_file *sf = &scrub->os_file;
+       struct osd_oi    **oi;
+       int                rc;
+       ENTRY;
+
+       OBD_ALLOC(oi, sizeof(*oi) * OSD_OI_FID_NR_MAX);
+       if (oi == NULL)
+               RETURN(-ENOMEM);
+
+       mutex_lock(&oi_init_lock);
+       /* try to open existing multiple OIs first */
+       rc = osd_oi_table_open(info, osd, oi, sf->sf_oi_count, false);
+       if (rc < 0)
+               GOTO(out, rc);
 
-        OBD_ALLOC(oi, sizeof(*oi) * OSD_OI_FID_NR_MAX);
-        if (oi == NULL)
-                return -ENOMEM;
-
-        cfs_mutex_lock(&oi_init_lock);
-        /* try to open existing multiple OIs first */
-        rc = osd_oi_table_open(info, osd, oi, 0, false);
-        if (rc != 0)
-                goto out;
-
-        /* 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 */
-                rc = 1;
-                goto out;
-        } else if (rc != -ENOENT) {
-                CERROR("%s: can't open %s: rc = %d\n",
-                       dev->dd_lu_dev.ld_obd->obd_name, OSD_OI_NAME_BASE, rc);
-                goto out;
-        }
+       if (rc > 0) {
+               if (rc == sf->sf_oi_count || sf->sf_oi_count == 0)
+                       GOTO(out, rc);
+
+               osd_scrub_file_reset(scrub,
+                                    LDISKFS_SB(osd_sb(osd))->s_es->s_uuid,
+                                    SF_RECREATED);
+               osd_oi_count = sf->sf_oi_count;
+               goto create;
+       }
+
+       /* 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 */
+               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);
+               GOTO(out, rc);
+       }
+
+       if (sf->sf_oi_count > 0) {
+               int i;
+
+               memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
+               for (i = 0; i < osd_oi_count; i++)
+                       ldiskfs_set_bit(i, sf->sf_oi_bitmap);
+               osd_scrub_file_reset(scrub,
+                                    LDISKFS_SB(osd_sb(osd))->s_es->s_uuid,
+                                    SF_RECREATED);
+       }
+       sf->sf_oi_count = osd_oi_count;
+
+create:
+       rc = osd_scrub_file_store(scrub);
+       if (rc < 0) {
+               osd_oi_table_put(info, oi, sf->sf_oi_count);
+               GOTO(out, rc);
+       }
+
+       /* No OIs exist, new filesystem, create OI objects */
+       rc = osd_oi_table_open(info, osd, oi, osd_oi_count, true);
+       LASSERT(ergo(rc >= 0, rc == osd_oi_count));
+
+       GOTO(out, rc);
 
-        /* No OIs exist, new filesystem, create OI objects */
-        rc = osd_oi_table_open(info, osd, oi, osd_oi_count, true);
-        LASSERT(ergo(rc >= 0, rc == osd_oi_count));
 out:
-        if (rc < 0) {
-                OBD_FREE(oi, sizeof(*oi) * OSD_OI_FID_NR_MAX);
-        } else {
-                LASSERT((rc & (rc - 1)) == 0);
-                osd->od_oi_table = oi;
-                osd->od_oi_count = rc;
-                rc = 0;
-        }
+       if (rc < 0) {
+               OBD_FREE(oi, sizeof(*oi) * OSD_OI_FID_NR_MAX);
+       } else {
+               LASSERT((rc & (rc - 1)) == 0);
+               osd->od_oi_table = oi;
+               osd->od_oi_count = rc;
+               rc = 0;
+       }
 
-        cfs_mutex_unlock(&oi_init_lock);
-        return rc;
+       mutex_unlock(&oi_init_lock);
+       return rc;
 }
 
 void osd_oi_fini(struct osd_thread_info *info, struct osd_device *osd)
 {
+       if (unlikely(osd->od_oi_table == NULL))
+               return;
+
         osd_oi_table_put(info, osd->od_oi_table, osd->od_oi_count);
 
         OBD_FREE(osd->od_oi_table,
@@ -388,7 +442,6 @@ static int osd_oi_iam_lookup(struct osd_thread_info *oti,
 {
         struct iam_container  *bag;
         struct iam_iterator   *it = &oti->oti_idx_it;
-        struct iam_rec        *iam_rec;
         struct iam_path_descr *ipd;
         int                    rc;
         ENTRY;
@@ -405,17 +458,8 @@ static int osd_oi_iam_lookup(struct osd_thread_info *oti,
         iam_it_init(it, bag, 0, ipd);
 
         rc = iam_it_get(it, (struct iam_key *)key);
-        if (rc >= 0) {
-                if (S_ISDIR(oi->oi_inode->i_mode))
-                        iam_rec = (struct iam_rec *)oti->oti_ldp;
-                else
-                        iam_rec = (struct iam_rec *)rec;
-
-                iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
-                if (S_ISDIR(oi->oi_inode->i_mode))
-                        osd_fid_unpack((struct lu_fid *)rec,
-                                       (struct osd_fid_pack *)iam_rec);
-        }
+       if (rc > 0)
+               iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)rec);
         iam_it_put(it);
         iam_it_fini(it);
         osd_ipd_put(oti->oti_env, bag, ipd);
@@ -425,6 +469,32 @@ static int osd_oi_iam_lookup(struct osd_thread_info *oti,
         RETURN(rc);
 }
 
+int fid_is_on_ost(struct osd_thread_info *info, struct osd_device *osd,
+                 const struct lu_fid *fid)
+{
+       struct lu_seq_range *range = &info->oti_seq_range;
+       int rc;
+       ENTRY;
+
+       if (fid_is_idif(fid) || fid_is_last_id(fid))
+               RETURN(1);
+
+       rc = osd_fld_lookup(info->oti_env, osd, fid, range);
+       if (rc != 0) {
+               CERROR("%s: Can not lookup fld for "DFID"\n",
+                      osd2lu_dev(osd)->ld_obd->obd_name, PFID(fid));
+               RETURN(rc);
+       }
+
+       CDEBUG(D_INFO, "fid "DFID" range "DRANGE"\n", PFID(fid),
+              PRANGE(range));
+
+       if (range->lsr_flags == LU_SEQ_RANGE_OST)
+               RETURN(1);
+
+       RETURN(0);
+}
+
 int __osd_oi_lookup(struct osd_thread_info *info, struct osd_device *osd,
                    const struct lu_fid *fid, struct osd_inode_id *id)
 {
@@ -444,25 +514,32 @@ int __osd_oi_lookup(struct osd_thread_info *info, struct osd_device *osd,
 }
 
 int osd_oi_lookup(struct osd_thread_info *info, struct osd_device *osd,
-                  const struct lu_fid *fid, struct osd_inode_id *id)
+                 const struct lu_fid *fid, struct osd_inode_id *id,
+                 bool check_fld)
 {
-        int                  rc = 0;
-
-        if (fid_is_idif(fid) || fid_seq(fid) == FID_SEQ_LLOG) {
-                /* old OSD obj id */
-                rc = osd_compat_objid_lookup(info, osd, fid, id);
-        } else if (fid_is_igif(fid)) {
-                lu_igif_to_id(fid, id);
-        } else if (fid_is_fs_root(fid)) {
+       int                  rc = 0;
+
+       if ((!fid_is_last_id(fid) && check_fld &&
+            fid_is_on_ost(info, osd, fid)) ||
+           fid_is_llog(fid)) {
+               /* old OSD obj id */
+               /* FIXME: actually for all of the OST object */
+               rc = osd_obj_map_lookup(info, osd, fid, id);
+       } else if (fid_is_igif(fid)) {
+               osd_id_gen(id, lu_igif_ino(fid), lu_igif_gen(fid));
+       } else 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);
        } else {
-               if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE))
-                       return osd_compat_spec_lookup(info, osd, fid, id);
+               if (unlikely(fid_is_acct(fid)))
+                       return osd_acct_obj_lookup(info, osd, fid, id);
+               else if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE) ||
+                        fid_is_last_id(fid))
+                       return osd_obj_spec_lookup(info, osd, fid, id);
 
                rc = __osd_oi_lookup(info, osd, fid, id);
-        }
-        return rc;
+       }
+       return rc;
 }
 
 static int osd_oi_iam_insert(struct osd_thread_info *oti, struct osd_oi *oi,
@@ -474,13 +551,11 @@ static int osd_oi_iam_insert(struct osd_thread_info *oti, struct osd_oi *oi,
         struct iam_path_descr *ipd;
         struct osd_thandle    *oh;
         int                    rc;
-#ifdef HAVE_QUOTA_SUPPORT
-        cfs_cap_t              save    = cfs_curproc_cap_pack();
-#endif
         ENTRY;
 
         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);
@@ -490,9 +565,6 @@ static int osd_oi_iam_insert(struct osd_thread_info *oti, struct osd_oi *oi,
         oh = container_of0(th, struct osd_thandle, ot_super);
         LASSERT(oh->ot_handle != NULL);
         LASSERT(oh->ot_handle->h_transaction != NULL);
-#ifdef HAVE_QUOTA_SUPPORT
-       cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
-#endif
         if (S_ISDIR(oi->oi_inode->i_mode))
                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec,
                              &oti->oti_fid);
@@ -500,9 +572,6 @@ static int osd_oi_iam_insert(struct osd_thread_info *oti, struct osd_oi *oi,
                 iam_rec = (struct iam_rec *) rec;
         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
                         iam_rec, ipd);
-#ifdef HAVE_QUOTA_SUPPORT
-        cfs_curproc_cap_unpack(save);
-#endif
         osd_ipd_put(oti->oti_env, bag, ipd);
         LINVRNT(osd_invariant(obj));
         RETURN(rc);
@@ -518,12 +587,14 @@ int osd_oi_insert(struct osd_thread_info *info, struct osd_device *osd,
        if (fid_is_igif(fid) || unlikely(fid_seq(fid) == FID_SEQ_DOT_LUSTRE))
                return 0;
 
-       if (fid_is_idif(fid) || fid_seq(fid) == FID_SEQ_LLOG)
-               return osd_compat_objid_insert(info, osd, fid, id, th);
+       if ((fid_is_on_ost(info, osd, fid) && !fid_is_last_id(fid)) ||
+            fid_is_llog(fid))
+               return osd_obj_map_insert(info, osd, fid, id, th);
 
        /* Server mount should not depends on OI files */
-       if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE))
-               return osd_compat_spec_insert(info, osd, fid, id, th);
+       if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE) ||
+           fid_is_last_id(fid))
+               return osd_obj_spec_insert(info, osd, fid, id, th);
 
        fid_cpu_to_be(oi_fid, fid);
        osd_id_pack(oi_id, id);
@@ -542,6 +613,8 @@ static int osd_oi_iam_delete(struct osd_thread_info *oti, struct osd_oi *oi,
         ENTRY;
 
         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);
@@ -564,10 +637,13 @@ int osd_oi_delete(struct osd_thread_info *info,
 {
        struct lu_fid *oi_fid = &info->oti_fid2;
 
+       if (fid_is_igif(fid) || fid_is_last_id(fid))
+               return 0;
+
        LASSERT(fid_seq(fid) != FID_SEQ_LOCAL_FILE);
 
-       if (fid_is_idif(fid) || fid_seq(fid) == FID_SEQ_LLOG)
-               return osd_compat_objid_delete(info, osd, fid, th);
+       if (fid_is_on_ost(info, osd, fid) || fid_is_llog(fid))
+               return osd_obj_map_delete(info, osd, fid, th);
 
        fid_cpu_to_be(oi_fid, fid);
        return osd_oi_iam_delete(info, osd_fid2oi(osd, fid),
@@ -585,6 +661,6 @@ int osd_oi_mod_init(void)
                 osd_oi_count = size_roundup_power2(osd_oi_count);
         }
 
-        cfs_mutex_init(&oi_init_lock);
+       mutex_init(&oi_init_lock);
         return 0;
 }