Whamcloud - gitweb
LU-11656 llite: fetch default layout for a directory
[fs/lustre-release.git] / lustre / llite / dir.c
index e85e975..e013f1b 100644 (file)
@@ -163,8 +163,7 @@ void ll_release_page(struct inode *inode, struct page *page,
 
        /* Always remove the page for striped dir, because the page is
         * built from temporarily in LMV layer */
-       if (inode != NULL && S_ISDIR(inode->i_mode) &&
-           ll_i2info(inode)->lli_lsm_md != NULL) {
+       if (inode && ll_dir_striped(inode)) {
                __free_page(page);
                return;
        }
@@ -172,7 +171,7 @@ void ll_release_page(struct inode *inode, struct page *page,
        if (remove) {
                lock_page(page);
                if (likely(page->mapping != NULL))
-                       truncate_complete_page(page->mapping, page);
+                       delete_from_page_cache(page);
                unlock_page(page);
        }
        put_page(page);
@@ -315,15 +314,16 @@ static int ll_iterate(struct file *filp, struct dir_context *ctx)
 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 #endif
 {
-       struct inode            *inode  = file_inode(filp);
-       struct ll_file_data     *lfd    = LUSTRE_FPRIVATE(filp);
-       struct ll_sb_info       *sbi    = ll_i2sbi(inode);
-       int                     hash64  = sbi->ll_flags & LL_SBI_64BIT_HASH;
-       int                     api32   = ll_need_32bit_api(sbi);
-       struct md_op_data       *op_data;
-       struct lu_fid           pfid = { 0 };
-       __u64                   pos;
-       int                     rc;
+       struct inode *inode = file_inode(filp);
+       struct ll_file_data *lfd = LUSTRE_FPRIVATE(filp);
+       struct ll_sb_info *sbi = ll_i2sbi(inode);
+       int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
+       int api32 = ll_need_32bit_api(sbi);
+       struct md_op_data *op_data;
+       struct lu_fid pfid = { 0 };
+       ktime_t kstart = ktime_get();
+       __u64 pos;
+       int rc;
        ENTRY;
 
        if (lfd != NULL)
@@ -341,12 +341,14 @@ static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
                 */
                GOTO(out, rc = 0);
 
-       if (unlikely(ll_i2info(inode)->lli_lsm_md != NULL)) {
-               /* This is only needed for striped dir to fill ..,
-                * see lmv_read_entry */
+       if (unlikely(ll_dir_striped(inode))) {
+               /*
+                * This is only needed for striped dir to fill ..,
+                * see lmv_read_page()
+                */
                if (file_dentry(filp)->d_parent != NULL &&
                    file_dentry(filp)->d_parent->d_inode != NULL) {
-                       __u64 ibits = MDS_INODELOCK_UPDATE;
+                       __u64 ibits = MDS_INODELOCK_LOOKUP;
                        struct inode *parent =
                                file_dentry(filp)->d_parent->d_inode;
 
@@ -367,6 +369,14 @@ static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
                                     LUSTRE_OPC_ANY, inode);
        if (IS_ERR(op_data))
                GOTO(out, rc = PTR_ERR(op_data));
+
+       /* foreign dirs are browsed out of Lustre */
+       if (unlikely(op_data->op_mea1 != NULL &&
+                    op_data->op_mea1->lsm_md_magic == LMV_MAGIC_FOREIGN)) {
+               ll_finish_md_op_data(op_data);
+               RETURN(-ENODATA);
+       }
+
        op_data->op_fid3 = pfid;
 
 #ifdef HAVE_DIR_CONTEXT
@@ -397,7 +407,8 @@ static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 
 out:
        if (!rc)
-               ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
+               ll_stats_ops_tally(sbi, LPROC_LL_READDIR,
+                                  ktime_us_delta(ktime_get(), kstart));
 
        RETURN(rc);
 }
@@ -454,14 +465,22 @@ static int ll_dir_setdirstripe(struct dentry *dparent, struct lmv_user_md *lump,
        int err;
        ENTRY;
 
-       if (unlikely(lump->lum_magic != LMV_USER_MAGIC &&
-                    lump->lum_magic != LMV_USER_MAGIC_SPECIFIC))
+       if (unlikely(!lmv_magic_supported(lump->lum_magic)))
                RETURN(-EINVAL);
 
-       CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) name %s "
-              "stripe_offset %d, stripe_count: %u\n",
-              PFID(ll_inode2fid(parent)), parent, dirname,
-              (int)lump->lum_stripe_offset, lump->lum_stripe_count);
+       if (lump->lum_magic != LMV_MAGIC_FOREIGN) {
+               CDEBUG(D_VFSTRACE,
+                      "VFS Op:inode="DFID"(%p) name %s stripe_offset %d, stripe_count: %u\n",
+                      PFID(ll_inode2fid(parent)), parent, dirname,
+                      (int)lump->lum_stripe_offset, lump->lum_stripe_count);
+       } else {
+               struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lump;
+
+               CDEBUG(D_VFSTRACE,
+                      "VFS Op:inode="DFID"(%p) name %s foreign, length %u, value '%.*s'\n",
+                      PFID(ll_inode2fid(parent)), parent, dirname,
+                      lfm->lfm_length, lfm->lfm_length, lfm->lfm_value);
+       }
 
        if (lump->lum_stripe_count > 1 &&
            !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_DIR_STRIPE))
@@ -471,8 +490,7 @@ static int ll_dir_setdirstripe(struct dentry *dparent, struct lmv_user_md *lump,
            !OBD_FAIL_CHECK(OBD_FAIL_LLITE_NO_CHECK_DEAD))
                RETURN(-ENOENT);
 
-       if (lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC) &&
-           lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC_SPECIFIC))
+       if (unlikely(!lmv_magic_supported(cpu_to_le32(lump->lum_magic))))
                lustre_swab_lmv_user_md(lump);
 
        if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
@@ -538,84 +556,67 @@ out_op_data:
 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
                      int set_default)
 {
-        struct ll_sb_info *sbi = ll_i2sbi(inode);
-        struct md_op_data *op_data;
-        struct ptlrpc_request *req = NULL;
-        int rc = 0;
+       struct ll_sb_info *sbi = ll_i2sbi(inode);
+       struct md_op_data *op_data;
+       struct ptlrpc_request *req = NULL;
+       int rc = 0;
 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
-        struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
-        struct obd_device *mgc = lsi->lsi_mgc;
+       struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
+       struct obd_device *mgc = lsi->lsi_mgc;
 #endif
-        int lum_size;
+       int lum_size;
        ENTRY;
 
-        if (lump != NULL) {
-                /*
-                 * This is coming from userspace, so should be in
-                 * local endian.  But the MDS would like it in little
-                 * endian, so we swab it before we send it.
-                 */
-                switch (lump->lmm_magic) {
-                case LOV_USER_MAGIC_V1: {
-                        if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
-                                lustre_swab_lov_user_md_v1(lump);
-                        lum_size = sizeof(struct lov_user_md_v1);
-                        break;
-                }
-                case LOV_USER_MAGIC_V3: {
-                        if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
-                                lustre_swab_lov_user_md_v3(
-                                        (struct lov_user_md_v3 *)lump);
-                        lum_size = sizeof(struct lov_user_md_v3);
-                        break;
-                }
-               case LOV_USER_MAGIC_COMP_V1: {
-                       if (lump->lmm_magic !=
-                           cpu_to_le32(LOV_USER_MAGIC_COMP_V1))
-                               lustre_swab_lov_comp_md_v1(
-                                       (struct lov_comp_md_v1 *)lump);
-                       lum_size = le32_to_cpu(
-                               ((struct lov_comp_md_v1 *)lump)->lcm_size);
+       if (lump != NULL) {
+               switch (lump->lmm_magic) {
+               case LOV_USER_MAGIC_V1:
+                       lum_size = sizeof(struct lov_user_md_v1);
                        break;
-               }
-               case LMV_USER_MAGIC: {
+               case LOV_USER_MAGIC_V3:
+                       lum_size = sizeof(struct lov_user_md_v3);
+                       break;
+               case LOV_USER_MAGIC_COMP_V1:
+                       lum_size = ((struct lov_comp_md_v1 *)lump)->lcm_size;
+                       break;
+               case LMV_USER_MAGIC:
                        if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
                                lustre_swab_lmv_user_md(
                                        (struct lmv_user_md *)lump);
                        lum_size = sizeof(struct lmv_user_md);
                        break;
-               }
                case LOV_USER_MAGIC_SPECIFIC: {
                        struct lov_user_md_v3 *v3 =
-                                       (struct lov_user_md_v3 *)lump;
+                               (struct lov_user_md_v3 *)lump;
                        if (v3->lmm_stripe_count > LOV_MAX_STRIPE_COUNT)
                                RETURN(-EINVAL);
-                       if (lump->lmm_magic !=
-                           cpu_to_le32(LOV_USER_MAGIC_SPECIFIC)) {
-                               lustre_swab_lov_user_md_v3(v3);
-                               lustre_swab_lov_user_md_objects(v3->lmm_objects,
-                                               v3->lmm_stripe_count);
-                       }
                        lum_size = lov_user_md_size(v3->lmm_stripe_count,
                                                    LOV_USER_MAGIC_SPECIFIC);
                        break;
                }
-               default: {
+               default:
                        CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
                                        " %#08x != %#08x nor %#08x\n",
                                        lump->lmm_magic, LOV_USER_MAGIC_V1,
                                        LOV_USER_MAGIC_V3);
                        RETURN(-EINVAL);
                }
-                }
-        } else {
-                lum_size = sizeof(struct lov_user_md_v1);
-        }
 
-        op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
-                                     LUSTRE_OPC_ANY, NULL);
-        if (IS_ERR(op_data))
-                RETURN(PTR_ERR(op_data));
+               /*
+                * This is coming from userspace, so should be in
+                * local endian.  But the MDS would like it in little
+                * endian, so we swab it before we send it.
+                */
+               if ((__swab32(lump->lmm_magic) & le32_to_cpu(LOV_MAGIC_MASK)) ==
+                   le32_to_cpu(LOV_MAGIC_MAGIC))
+                       lustre_swab_lov_user_md(lump, 0);
+       } else {
+               lum_size = sizeof(struct lov_user_md_v1);
+       }
+
+       op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
+                                    LUSTRE_OPC_ANY, NULL);
+       if (IS_ERR(op_data))
+               RETURN(PTR_ERR(op_data));
 
        /* swabbing is done in lov_setstripe() on server side */
        rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, &req);
@@ -646,8 +647,8 @@ int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 
                buf = param;
                /* Get fsname and assume devname to be -MDT0000. */
-               ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
-               strcat(buf, "-MDT0000.lov");
+               snprintf(buf, MGS_PARAM_MAXLEN, "%s-MDT0000.lov",
+                        sbi->ll_fsname);
                buf += strlen(buf);
 
                /* Set root stripesize */
@@ -678,16 +679,10 @@ end:
        RETURN(rc);
 }
 
-/**
- * This function will be used to get default LOV/LMV/Default LMV
- * @valid will be used to indicate which stripe it will retrieve
- *     OBD_MD_MEA              LMV stripe EA
- *     OBD_MD_DEFAULT_MEA      Default LMV stripe EA
- *     otherwise               Default LOV EA.
- * Each time, it can only retrieve 1 stripe EA
- **/
-int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
-                    struct ptlrpc_request **request, u64 valid)
+static int ll_dir_get_default_layout(struct inode *inode, void **plmm,
+                                    int *plmm_size,
+                                    struct ptlrpc_request **request, u64 valid,
+                                    enum get_default_layout_type type)
 {
        struct ll_sb_info *sbi = ll_i2sbi(inode);
        struct mdt_body   *body;
@@ -695,6 +690,7 @@ int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
        struct ptlrpc_request *req = NULL;
        int rc, lmm_size;
        struct md_op_data *op_data;
+       struct lu_fid fid;
        ENTRY;
 
        rc = ll_get_default_mdsize(sbi, &lmm_size);
@@ -708,11 +704,19 @@ int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
                RETURN(PTR_ERR(op_data));
 
        op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
+
+       if (type == GET_DEFAULT_LAYOUT_ROOT) {
+               lu_root_fid(&op_data->op_fid1);
+               fid = op_data->op_fid1;
+       } else {
+               fid = *ll_inode2fid(inode);
+       }
+
        rc = md_getattr(sbi->ll_md_exp, op_data, &req);
        ll_finish_md_op_data(op_data);
        if (rc < 0) {
-               CDEBUG(D_INFO, "md_getattr failed on inode "
-                      DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
+               CDEBUG(D_INFO, "md_getattr failed on inode "DFID": rc %d\n",
+                      PFID(&fid), rc);
                GOTO(out, rc);
        }
 
@@ -738,17 +742,11 @@ int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
        /* We don't swab objects for directories */
        switch (le32_to_cpu(lmm->lmm_magic)) {
        case LOV_MAGIC_V1:
-               if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
-                       lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
-               break;
        case LOV_MAGIC_V3:
-               if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
-                       lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
-               break;
        case LOV_MAGIC_COMP_V1:
+       case LOV_USER_MAGIC_SPECIFIC:
                if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
-                       lustre_swab_lov_comp_md_v1(
-                                       (struct lov_comp_md_v1 *)lmm);
+                       lustre_swab_lov_user_md((struct lov_user_md *)lmm, 0);
                break;
        case LMV_MAGIC_V1:
                if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
@@ -758,16 +756,17 @@ int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
                if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
                        lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
                break;
-       case LOV_USER_MAGIC_SPECIFIC: {
-               struct lov_user_md_v3 *v3 = (struct lov_user_md_v3 *)lmm;
-
-               if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC)) {
-                       lustre_swab_lov_user_md_v3(v3);
-                       lustre_swab_lov_user_md_objects(v3->lmm_objects,
-                                                       v3->lmm_stripe_count);
-                       }
+       case LMV_MAGIC_FOREIGN: {
+               struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lmm;
+
+               if (LMV_MAGIC_FOREIGN != cpu_to_le32(LMV_MAGIC_FOREIGN)) {
+                       __swab32s(&lfm->lfm_magic);
+                       __swab32s(&lfm->lfm_length);
+                       __swab32s(&lfm->lfm_type);
+                       __swab32s(&lfm->lfm_flags);
                }
                break;
+       }
        default:
                CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
                rc = -EPROTO;
@@ -779,6 +778,72 @@ out:
        return rc;
 }
 
+/**
+ * This function will be used to get default LOV/LMV/Default LMV
+ * @valid will be used to indicate which stripe it will retrieve.
+ * If the directory does not have its own default layout, then the
+ * function will request the default layout from root FID.
+ *     OBD_MD_MEA              LMV stripe EA
+ *     OBD_MD_DEFAULT_MEA      Default LMV stripe EA
+ *     otherwise               Default LOV EA.
+ * Each time, it can only retrieve 1 stripe EA
+ **/
+int ll_dir_getstripe_default(struct inode *inode, void **plmm, int *plmm_size,
+                            struct ptlrpc_request **request,
+                            struct ptlrpc_request **root_request,
+                            u64 valid)
+{
+       struct ptlrpc_request *req = NULL;
+       struct ptlrpc_request *root_req = NULL;
+       struct lov_mds_md *lmm = NULL;
+       int lmm_size = 0;
+       int rc = 0;
+       ENTRY;
+
+       rc = ll_dir_get_default_layout(inode, (void **)&lmm, &lmm_size,
+                                      &req, valid, 0);
+       if (rc == -ENODATA && !fid_is_root(ll_inode2fid(inode)) &&
+           !(valid & (OBD_MD_MEA|OBD_MD_DEFAULT_MEA)) && root_request != NULL)
+               rc = ll_dir_get_default_layout(inode, (void **)&lmm, &lmm_size,
+                                              &root_req, valid,
+                                              GET_DEFAULT_LAYOUT_ROOT);
+
+       *plmm = lmm;
+       *plmm_size = lmm_size;
+       *request = req;
+       if (root_request != NULL)
+               *root_request = root_req;
+
+       RETURN(rc);
+}
+
+/**
+ * This function will be used to get default LOV/LMV/Default LMV
+ * @valid will be used to indicate which stripe it will retrieve
+ *     OBD_MD_MEA              LMV stripe EA
+ *     OBD_MD_DEFAULT_MEA      Default LMV stripe EA
+ *     otherwise               Default LOV EA.
+ * Each time, it can only retrieve 1 stripe EA
+ **/
+int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
+                    struct ptlrpc_request **request, u64 valid)
+{
+       struct ptlrpc_request *req = NULL;
+       struct lov_mds_md *lmm = NULL;
+       int lmm_size = 0;
+       int rc = 0;
+       ENTRY;
+
+       rc = ll_dir_get_default_layout(inode, (void **)&lmm, &lmm_size,
+                                      &req, valid, 0);
+
+       *plmm = lmm;
+       *plmm_size = lmm_size;
+       *request = req;
+
+       RETURN(rc);
+}
+
 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
 {
        struct md_op_data       *op_data;
@@ -987,25 +1052,110 @@ progress:
 }
 
 
-static int copy_and_ioctl(int cmd, struct obd_export *exp,
-                         const void __user *data, size_t size)
+static int copy_and_ct_start(int cmd, struct obd_export *exp,
+                            const struct lustre_kernelcomm __user *data)
 {
-       void *copy;
+       struct lustre_kernelcomm *lk;
+       struct lustre_kernelcomm *tmp;
+       size_t size = sizeof(*lk);
+       size_t new_size;
+       int i;
        int rc;
 
-       OBD_ALLOC(copy, size);
-       if (copy == NULL)
+       /* copy data from userspace to get numbers of archive_id */
+       OBD_ALLOC(lk, size);
+       if (lk == NULL)
                return -ENOMEM;
 
-       if (copy_from_user(copy, data, size)) {
-               rc = -EFAULT;
-               goto out;
+       if (copy_from_user(lk, data, size))
+               GOTO(out_lk, rc = -EFAULT);
+
+       if (lk->lk_flags & LK_FLG_STOP)
+               goto do_ioctl;
+
+       if (!(lk->lk_flags & LK_FLG_DATANR)) {
+               __u32 archive_mask = lk->lk_data_count;
+               int count;
+
+               /* old hsm agent to old MDS */
+               if (!exp_connect_archive_id_array(exp))
+                       goto do_ioctl;
+
+               /* old hsm agent to new MDS */
+               lk->lk_flags |= LK_FLG_DATANR;
+
+               if (archive_mask == 0)
+                       goto do_ioctl;
+
+               count = hweight32(archive_mask);
+               new_size = offsetof(struct lustre_kernelcomm, lk_data[count]);
+               OBD_ALLOC(tmp, new_size);
+               if (tmp == NULL)
+                       GOTO(out_lk, rc = -ENOMEM);
+
+               memcpy(tmp, lk, size);
+               tmp->lk_data_count = count;
+               OBD_FREE(lk, size);
+               lk = tmp;
+               size = new_size;
+
+               count = 0;
+               for (i = 0; i < sizeof(archive_mask) * 8; i++) {
+                       if ((1 << i) & archive_mask) {
+                               lk->lk_data[count] = i + 1;
+                               count++;
+                       }
+               }
+               goto do_ioctl;
+       }
+
+       /* new hsm agent to new mds */
+       if (lk->lk_data_count > 0) {
+               new_size = offsetof(struct lustre_kernelcomm,
+                                   lk_data[lk->lk_data_count]);
+               OBD_ALLOC(tmp, new_size);
+               if (tmp == NULL)
+                       GOTO(out_lk, rc = -ENOMEM);
+
+               OBD_FREE(lk, size);
+               lk = tmp;
+               size = new_size;
+
+               if (copy_from_user(lk, data, size))
+                       GOTO(out_lk, rc = -EFAULT);
        }
 
-       rc = obd_iocontrol(cmd, exp, size, copy, NULL);
-out:
-       OBD_FREE(copy, size);
+       /* new hsm agent to old MDS */
+       if (!exp_connect_archive_id_array(exp)) {
+               __u32 archives = 0;
+
+               if (lk->lk_data_count > LL_HSM_ORIGIN_MAX_ARCHIVE)
+                       GOTO(out_lk, rc = -EINVAL);
+
+               for (i = 0; i < lk->lk_data_count; i++) {
+                       if (lk->lk_data[i] > LL_HSM_ORIGIN_MAX_ARCHIVE) {
+                               rc = -EINVAL;
+                               CERROR("%s: archive id %d requested but only "
+                                      "[0 - %zu] supported: rc = %d\n",
+                                      exp->exp_obd->obd_name, lk->lk_data[i],
+                                      LL_HSM_ORIGIN_MAX_ARCHIVE, rc);
+                               GOTO(out_lk, rc);
+                       }
+
+                       if (lk->lk_data[i] == 0) {
+                               archives = 0;
+                               break;
+                       }
 
+                       archives |= (1 << (lk->lk_data[i] - 1));
+               }
+               lk->lk_flags &= ~LK_FLG_DATANR;
+               lk->lk_data_count = archives;
+       }
+do_ioctl:
+       rc = obd_iocontrol(cmd, exp, size, lk, NULL);
+out_lk:
+       OBD_FREE(lk, size);
        return rc;
 }
 
@@ -1150,6 +1300,54 @@ out:
         RETURN(rc);
 }
 
+int ll_rmfid(struct file *file, void __user *arg)
+{
+       const struct fid_array __user *ufa = arg;
+       struct fid_array *lfa = NULL;
+       size_t size;
+       unsigned nr;
+       int i, rc, *rcs = NULL;
+       ENTRY;
+
+       if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
+           !(ll_i2sbi(file_inode(file))->ll_flags & LL_SBI_USER_FID2PATH))
+               RETURN(-EPERM);
+       /* Only need to get the buflen */
+       if (get_user(nr, &ufa->fa_nr))
+               RETURN(-EFAULT);
+       /* DoS protection */
+       if (nr > OBD_MAX_FIDS_IN_ARRAY)
+               RETURN(-E2BIG);
+
+       size = offsetof(struct fid_array, fa_fids[nr]);
+       OBD_ALLOC(lfa, size);
+       if (!lfa)
+               RETURN(-ENOMEM);
+       OBD_ALLOC(rcs, sizeof(int) * nr);
+       if (!rcs)
+               GOTO(free_lfa, rc = -ENOMEM);
+
+       if (copy_from_user(lfa, arg, size))
+               GOTO(free_rcs, rc = -EFAULT);
+
+       /* Call mdc_iocontrol */
+       rc = md_rmfid(ll_i2mdexp(file_inode(file)), lfa, rcs, NULL);
+       if (!rc) {
+               for (i = 0; i < nr; i++)
+                       if (rcs[i])
+                               lfa->fa_fids[i].f_ver = rcs[i];
+               if (copy_to_user(arg, lfa, size))
+                       rc = -EFAULT;
+       }
+
+free_rcs:
+       OBD_FREE(rcs, sizeof(int) * nr);
+free_lfa:
+       OBD_FREE(lfa, size);
+
+       RETURN(rc);
+}
+
 /* This function tries to get a single name component,
  * to send to the server. No actual path traversal involved,
  * so we limit to NAME_MAX */
@@ -1240,8 +1438,7 @@ static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
                rc = ll_get_fid_by_name(inode, filename, namelen, NULL, NULL);
                if (rc < 0) {
                        CERROR("%s: lookup %.*s failed: rc = %d\n",
-                              ll_get_fsname(inode->i_sb, NULL, 0), namelen,
-                              filename, rc);
+                              sbi->ll_fsname, namelen, filename, rc);
                        GOTO(out_free, rc);
                }
 out_free:
@@ -1277,9 +1474,22 @@ out_free:
                lum = (struct lmv_user_md *)data->ioc_inlbuf2;
                lumlen = data->ioc_inllen2;
 
-               if ((lum->lum_magic != LMV_USER_MAGIC &&
-                    lum->lum_magic != LMV_USER_MAGIC_SPECIFIC) ||
+               if (!lmv_magic_supported(lum->lum_magic)) {
+                       CERROR("%s: wrong lum magic %x : rc = %d\n", filename,
+                              lum->lum_magic, -EINVAL);
+                       GOTO(lmv_out_free, rc = -EINVAL);
+               }
+
+               if ((lum->lum_magic == LMV_USER_MAGIC ||
+                    lum->lum_magic == LMV_USER_MAGIC_SPECIFIC) &&
                    lumlen < sizeof(*lum)) {
+                       CERROR("%s: wrong lum size %d for magic %x : rc = %d\n",
+                              filename, lumlen, lum->lum_magic, -EINVAL);
+                       GOTO(lmv_out_free, rc = -EINVAL);
+               }
+
+               if (lum->lum_magic == LMV_MAGIC_FOREIGN &&
+                   lumlen < sizeof(struct lmv_foreign_md)) {
                        CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
                               filename, lum->lum_magic, lumlen, -EFAULT);
                        GOTO(lmv_out_free, rc = -EINVAL);
@@ -1321,13 +1531,13 @@ lmv_out_free:
                        (struct lov_user_md_v1 __user *)arg;
                struct lov_user_md_v3 __user *lumv3p =
                        (struct lov_user_md_v3 __user *)arg;
-               int lum_size;
+               int lum_size = 0;
 
                int set_default = 0;
 
-               CLASSERT(sizeof(struct lov_user_md_v3) >
-                        sizeof(struct lov_comp_md_v1));
-               CLASSERT(sizeof(*lumv3) == sizeof(*lumv3p));
+               BUILD_BUG_ON(sizeof(struct lov_user_md_v3) <=
+                            sizeof(struct lov_comp_md_v1));
+               BUILD_BUG_ON(sizeof(*lumv3) != sizeof(*lumv3p));
                /* first try with v1 which is smaller than v3 */
                if (copy_from_user(&lumv1, lumv1p, sizeof(lumv1)))
                        RETURN(-EFAULT);
@@ -1366,6 +1576,7 @@ out:
                                        (struct lmv_user_md __user *)arg;
                struct lmv_user_md      lum;
                struct ptlrpc_request   *request = NULL;
+               struct ptlrpc_request   *root_request = NULL;
                union lmv_mds_md        *lmm = NULL;
                int                     lmmsize;
                u64                     valid = 0;
@@ -1391,8 +1602,8 @@ out:
                else
                        RETURN(-EINVAL);
 
-               rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
-                                     valid);
+               rc = ll_dir_getstripe_default(inode, (void **)&lmm, &lmmsize,
+                                             &request, &root_request, valid);
                if (rc != 0)
                        GOTO(finish_req, rc);
 
@@ -1407,7 +1618,25 @@ out:
                        GOTO(finish_req, rc);
                }
 
-               stripe_count = lmv_mds_md_stripe_count_get(lmm);
+               /* if foreign LMV case, fake stripes number */
+               if (lmm->lmv_magic == LMV_MAGIC_FOREIGN) {
+                       struct lmv_foreign_md *lfm;
+
+                       lfm = (struct lmv_foreign_md *)lmm;
+                       if (lfm->lfm_length < XATTR_SIZE_MAX -
+                           offsetof(typeof(*lfm), lfm_value)) {
+                               __u32 size = lfm->lfm_length +
+                                            offsetof(typeof(*lfm), lfm_value);
+
+                               stripe_count = lmv_foreign_to_md_stripes(size);
+                       } else {
+                               CERROR("invalid %d foreign size returned\n",
+                                           lfm->lfm_length);
+                               return -EINVAL;
+                       }
+               } else {
+                       stripe_count = lmv_mds_md_stripe_count_get(lmm);
+               }
                if (max_stripe_count < stripe_count) {
                        lum.lum_stripe_count = stripe_count;
                        if (copy_to_user(ulmv, &lum, sizeof(lum)))
@@ -1415,6 +1644,19 @@ out:
                        GOTO(finish_req, rc = -E2BIG);
                }
 
+               /* enough room on user side and foreign case */
+               if (lmm->lmv_magic == LMV_MAGIC_FOREIGN) {
+                       struct lmv_foreign_md *lfm;
+                       __u32 size;
+
+                       lfm = (struct lmv_foreign_md *)lmm;
+                       size = lfm->lfm_length +
+                              offsetof(struct lmv_foreign_md, lfm_value);
+                       if (copy_to_user(ulmv, lfm, size))
+                               GOTO(finish_req, rc = -EFAULT);
+                       GOTO(finish_req, rc);
+               }
+
                lum_size = lmv_user_md_size(stripe_count,
                                            LMV_USER_MAGIC_SPECIFIC);
                OBD_ALLOC(tmp, lum_size);
@@ -1433,12 +1675,15 @@ out:
                        struct lu_fid   fid;
 
                        fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
-                       mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
-                       if (mdt_index < 0)
-                               GOTO(out_tmp, rc = mdt_index);
+                       if (fid_is_sane(&fid)) {
+                               mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
+                               if (mdt_index < 0)
+                                       GOTO(out_tmp, rc = mdt_index);
+
+                               tmp->lum_objects[i].lum_mds = mdt_index;
+                               tmp->lum_objects[i].lum_fid = fid;
+                       }
 
-                       tmp->lum_objects[i].lum_mds = mdt_index;
-                       tmp->lum_objects[i].lum_fid = fid;
                        tmp->lum_stripe_count++;
                }
 
@@ -1448,6 +1693,7 @@ out_tmp:
                OBD_FREE(tmp, lum_size);
 finish_req:
                ptlrpc_req_finished(request);
+               ptlrpc_req_finished(root_request);
                return rc;
        }
 
@@ -1478,6 +1724,8 @@ out_rmdir:
                         ll_putname(filename);
                RETURN(rc);
        }
+       case LL_IOC_RMFID:
+               RETURN(ll_rmfid(file, (void __user *)arg));
        case LL_IOC_LOV_SWAP_LAYOUTS:
                RETURN(-EPERM);
        case IOC_OBD_STATFS:
@@ -1485,38 +1733,50 @@ out_rmdir:
        case LL_IOC_LOV_GETSTRIPE:
        case LL_IOC_LOV_GETSTRIPE_NEW:
        case LL_IOC_MDC_GETINFO:
+       case LL_IOC_MDC_GETINFO_OLD:
        case IOC_MDC_GETFILEINFO:
+       case IOC_MDC_GETFILEINFO_OLD:
        case IOC_MDC_GETFILESTRIPE: {
                struct ptlrpc_request *request = NULL;
+               struct ptlrpc_request *root_request = NULL;
                struct lov_user_md __user *lump;
-                struct lov_mds_md *lmm = NULL;
-                struct mdt_body *body;
-                char *filename = NULL;
-                int lmmsize;
-
-                if (cmd == IOC_MDC_GETFILEINFO ||
-                    cmd == IOC_MDC_GETFILESTRIPE) {
+               struct lov_mds_md *lmm = NULL;
+               struct mdt_body *body;
+               char *filename = NULL;
+               lstat_t __user *statp = NULL;
+               lstatx_t __user *stxp = NULL;
+               __u64 __user *flagsp = NULL;
+               __u32 __user *lmmsizep = NULL;
+               struct lu_fid __user *fidp = NULL;
+               int lmmsize;
+
+               if (cmd == IOC_MDC_GETFILEINFO_OLD ||
+                   cmd == IOC_MDC_GETFILEINFO ||
+                   cmd == IOC_MDC_GETFILESTRIPE) {
                        filename = ll_getname((const char __user *)arg);
-                        if (IS_ERR(filename))
-                                RETURN(PTR_ERR(filename));
+                       if (IS_ERR(filename))
+                               RETURN(PTR_ERR(filename));
 
-                        rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
-                                                      &lmmsize, &request);
+                       rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
+                                                     &lmmsize, &request);
                } else {
-                       rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
-                                             &request, 0);
+                       rc = ll_dir_getstripe_default(inode, (void **)&lmm,
+                                                     &lmmsize, &request,
+                                                     &root_request, 0);
                }
 
-                if (request) {
-                        body = req_capsule_server_get(&request->rq_pill,
-                                                      &RMF_MDT_BODY);
-                        LASSERT(body != NULL);
-                } else {
-                        GOTO(out_req, rc);
-                }
+               if (request) {
+                       body = req_capsule_server_get(&request->rq_pill,
+                                                     &RMF_MDT_BODY);
+                       LASSERT(body != NULL);
+               } else {
+                       GOTO(out_req, rc);
+               }
 
                if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
-                                      cmd == LL_IOC_MDC_GETINFO)) {
+                                      cmd == LL_IOC_MDC_GETINFO ||
+                                      cmd == IOC_MDC_GETFILEINFO_OLD ||
+                                      cmd == LL_IOC_MDC_GETINFO_OLD)) {
                        lmmsize = 0;
                        rc = 0;
                }
@@ -1528,11 +1788,23 @@ out_rmdir:
                    cmd == LL_IOC_LOV_GETSTRIPE ||
                    cmd == LL_IOC_LOV_GETSTRIPE_NEW) {
                        lump = (struct lov_user_md __user *)arg;
-                } else {
+               } else if (cmd == IOC_MDC_GETFILEINFO_OLD ||
+                          cmd == LL_IOC_MDC_GETINFO_OLD){
+                       struct lov_user_mds_data_v1 __user *lmdp;
+
+                       lmdp = (struct lov_user_mds_data_v1 __user *)arg;
+                       statp = &lmdp->lmd_st;
+                       lump = &lmdp->lmd_lmm;
+               } else {
                        struct lov_user_mds_data __user *lmdp;
+
                        lmdp = (struct lov_user_mds_data __user *)arg;
-                        lump = &lmdp->lmd_lmm;
-                }
+                       fidp = &lmdp->lmd_fid;
+                       stxp = &lmdp->lmd_stx;
+                       flagsp = &lmdp->lmd_flags;
+                       lmmsizep = &lmdp->lmd_lmmsize;
+                       lump = &lmdp->lmd_lmm;
+               }
 
                if (lmmsize == 0) {
                        /* If the file has no striping then zero out *lump so
@@ -1545,9 +1817,9 @@ out_rmdir:
                        rc = -EOVERFLOW;
                }
 
-                if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
-                       struct lov_user_mds_data __user *lmdp;
-                        lstat_t st = { 0 };
+               if (cmd == IOC_MDC_GETFILEINFO_OLD ||
+                   cmd == LL_IOC_MDC_GETINFO_OLD) {
+                       lstat_t st = { 0 };
 
                        st.st_dev       = inode->i_sb->s_dev;
                        st.st_mode      = body->mbo_mode;
@@ -1565,18 +1837,75 @@ out_rmdir:
                                                sbi->ll_flags &
                                                LL_SBI_32BIT_API);
 
-                       lmdp = (struct lov_user_mds_data __user *)arg;
-                       if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
-                                GOTO(out_req, rc = -EFAULT);
-                }
+                       if (copy_to_user(statp, &st, sizeof(st)))
+                               GOTO(out_req, rc = -EFAULT);
+               } else if (cmd == IOC_MDC_GETFILEINFO ||
+                          cmd == LL_IOC_MDC_GETINFO) {
+                       lstatx_t stx = { 0 };
+                       __u64 valid = body->mbo_valid;
+
+                       stx.stx_blksize = PAGE_SIZE;
+                       stx.stx_nlink = body->mbo_nlink;
+                       stx.stx_uid = body->mbo_uid;
+                       stx.stx_gid = body->mbo_gid;
+                       stx.stx_mode = body->mbo_mode;
+                       stx.stx_ino = cl_fid_build_ino(&body->mbo_fid1,
+                                                      sbi->ll_flags &
+                                                      LL_SBI_32BIT_API);
+                       stx.stx_size = body->mbo_size;
+                       stx.stx_blocks = body->mbo_blocks;
+                       stx.stx_atime.tv_sec = body->mbo_atime;
+                       stx.stx_ctime.tv_sec = body->mbo_ctime;
+                       stx.stx_mtime.tv_sec = body->mbo_mtime;
+                       stx.stx_rdev_major = MAJOR(body->mbo_rdev);
+                       stx.stx_rdev_minor = MINOR(body->mbo_rdev);
+                       stx.stx_dev_major = MAJOR(inode->i_sb->s_dev);
+                       stx.stx_dev_minor = MINOR(inode->i_sb->s_dev);
+                       stx.stx_mask |= STATX_BASIC_STATS;
 
-                EXIT;
-        out_req:
-                ptlrpc_req_finished(request);
-                if (filename)
-                        ll_putname(filename);
-                return rc;
-        }
+                       /*
+                        * For a striped directory, the size and blocks returned
+                        * from MDT is not correct.
+                        * The size and blocks are aggregated by client across
+                        * all stripes.
+                        * Thus for a striped directory, do not return the valid
+                        * FLSIZE and FLBLOCKS flags to the caller.
+                        * However, this whould be better decided by the MDS
+                        * instead of the client.
+                        */
+                       if (cmd == LL_IOC_MDC_GETINFO &&
+                           ll_i2info(inode)->lli_lsm_md != NULL)
+                               valid &= ~(OBD_MD_FLSIZE | OBD_MD_FLBLOCKS);
+
+                       if (flagsp && copy_to_user(flagsp, &valid,
+                                                  sizeof(*flagsp)))
+                               GOTO(out_req, rc = -EFAULT);
+
+                       if (fidp && copy_to_user(fidp, &body->mbo_fid1,
+                                                sizeof(*fidp)))
+                               GOTO(out_req, rc = -EFAULT);
+
+                       if (!(valid & OBD_MD_FLSIZE))
+                               stx.stx_mask &= ~STATX_SIZE;
+                       if (!(valid & OBD_MD_FLBLOCKS))
+                               stx.stx_mask &= ~STATX_BLOCKS;
+
+                       if (stxp && copy_to_user(stxp, &stx, sizeof(stx)))
+                               GOTO(out_req, rc = -EFAULT);
+
+                       if (lmmsizep && copy_to_user(lmmsizep, &lmmsize,
+                                                    sizeof(*lmmsizep)))
+                               GOTO(out_req, rc = -EFAULT);
+               }
+
+               EXIT;
+out_req:
+               ptlrpc_req_finished(request);
+               ptlrpc_req_finished(root_request);
+               if (filename)
+                       ll_putname(filename);
+               return rc;
+       }
        case OBD_IOC_QUOTACTL: {
                 struct if_quotactl *qctl;
 
@@ -1738,8 +2067,8 @@ out_hur:
                if (!cfs_capable(CFS_CAP_SYS_ADMIN))
                        RETURN(-EPERM);
 
-               rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
-                                   sizeof(struct lustre_kernelcomm));
+               rc = copy_and_ct_start(cmd, sbi->ll_md_exp,
+                                      (struct lustre_kernelcomm __user *)arg);
                RETURN(rc);
 
        case LL_IOC_HSM_COPY_START: {
@@ -1824,6 +2153,43 @@ migrate_free:
                RETURN(ll_ioctl_fsgetxattr(inode, cmd, arg));
        case LL_IOC_FSSETXATTR:
                RETURN(ll_ioctl_fssetxattr(inode, cmd, arg));
+       case LL_IOC_PCC_DETACH_BY_FID: {
+               struct lu_pcc_detach_fid *detach;
+               struct lu_fid *fid;
+               struct inode *inode2;
+               unsigned long ino;
+
+               OBD_ALLOC_PTR(detach);
+               if (detach == NULL)
+                       RETURN(-ENOMEM);
+
+               if (copy_from_user(detach,
+                                  (const struct lu_pcc_detach_fid __user *)arg,
+                                  sizeof(*detach)))
+                       GOTO(out_detach, rc = -EFAULT);
+
+               fid = &detach->pccd_fid;
+               ino = cl_fid_build_ino(fid, ll_need_32bit_api(sbi));
+               inode2 = ilookup5(inode->i_sb, ino, ll_test_inode_by_fid, fid);
+               if (inode2 == NULL)
+                       /* Target inode is not in inode cache, and PCC file
+                        * has aleady released, return immdiately.
+                        */
+                       GOTO(out_detach, rc = 0);
+
+               if (!S_ISREG(inode2->i_mode))
+                       GOTO(out_iput, rc = -EINVAL);
+
+               if (!inode_owner_or_capable(inode2))
+                       GOTO(out_iput, rc = -EPERM);
+
+               rc = pcc_ioctl_detach(inode2, detach->pccd_opt);
+out_iput:
+               iput(inode2);
+out_detach:
+               OBD_FREE_PTR(detach);
+               RETURN(rc);
+       }
        default:
                RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
                                     (void __user *)arg));
@@ -1899,7 +2265,7 @@ const struct file_operations ll_dir_operations = {
        .release        = ll_dir_release,
        .read           = generic_read_dir,
 #ifdef HAVE_DIR_CONTEXT
-       .iterate        = ll_iterate,
+       .iterate_shared = ll_iterate,
 #else
        .readdir        = ll_readdir,
 #endif