Whamcloud - gitweb
LU-9859 libcfs: prepare for switch to container_of_safe()
[fs/lustre-release.git] / lustre / llite / llite_nfs.c
index 7615751..4907d68 100644 (file)
-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
+/*
+ * GPL HEADER START
  *
- *   NFS export of Lustre Light File System 
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
- *   Copyright (c) 2002, 2003 Cluster File Systems, Inc.
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 only,
+ * as published by the Free Software Foundation.
  *
- *   This file is part of Lustre, http://www.lustre.org.
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License version 2 for more details (a copy is included
+ * in the LICENSE file that accompanied this code).
  *
- *   Lustre is free software; you can redistribute it and/or
- *   modify it under the terms of version 2 of the GNU General Public
- *   License as published by the Free Software Foundation.
+ * You should have received a copy of the GNU General Public License
+ * version 2 along with this program; If not, see
+ * http://www.gnu.org/licenses/gpl-2.0.html
  *
- *   Lustre is distributed in the hope that it will be useful,
- *   but WITHOUT ANY WARRANTY; without even the implied warranty of
- *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
- *   GNU General Public License for more details.
+ * GPL HEADER END
+ */
+/*
+ * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Use is subject to license terms.
+ *
+ * Copyright (c) 2011, 2017, Intel Corporation.
+ */
+/*
+ * This file is part of Lustre, http://www.lustre.org/
+ * Lustre is a trademark of Sun Microsystems, Inc.
+ *
+ * lustre/lustre/llite/llite_nfs.c
+ *
+ * NFS export of Lustre Light File System
  *
- *   You should have received a copy of the GNU General Public License
- *   along with Lustre; if not, write to the Free Software
- *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ * Author: Yury Umanets <umka@clusterfs.com>
+ * Author: Huang Hua <huanghua@clusterfs.com>
  */
 
 #define DEBUG_SUBSYSTEM S_LLITE
-#include <linux/lustre_lite.h>
 #include "llite_internal.h"
+#include <linux/exportfs.h>
+
+u32 get_uuid2int(const char *name, int len)
+{
+       u32 key0 = 0x12a3fe2d, key1 = 0x37abe8f9;
+
+       while (len--) {
+               u32 key = key1 + (key0 ^ (*name++ * 7152373));
+
+               if (key & 0x80000000)
+                       key -= 0x7fffffff;
+
+               key1 = key0;
+               key0 = key;
+       }
+       return (key0 << 1);
+}
+
+struct inode *search_inode_for_lustre(struct super_block *sb,
+                                     const struct lu_fid *fid)
+{
+       struct ll_sb_info *sbi = ll_s2sbi(sb);
+       struct ptlrpc_request *req = NULL;
+       struct inode *inode = NULL;
+       int eadatalen = 0;
+       unsigned long hash = cl_fid_build_ino(fid, ll_need_32bit_api(sbi));
+       struct md_op_data *op_data;
+       int rc;
+
+       ENTRY;
+
+       CDEBUG(D_INFO, "searching inode for:(%lu,"DFID")\n", hash, PFID(fid));
+
+       inode = ilookup5(sb, hash, ll_test_inode_by_fid, (void *)fid);
+       if (inode)
+               RETURN(inode);
+
+       rc = ll_get_default_mdsize(sbi, &eadatalen);
+       if (rc)
+               RETURN(ERR_PTR(rc));
+
+       /*
+        * Because inode is NULL, ll_prep_md_op_data can not
+        * be used here. So we allocate op_data ourselves
+        */
+       OBD_ALLOC_PTR(op_data);
+       if (!op_data)
+               return ERR_PTR(-ENOMEM);
+
+       op_data->op_fid1 = *fid;
+       op_data->op_mode = eadatalen;
+       op_data->op_valid = OBD_MD_FLEASIZE;
+
+       /* mds_fid2dentry ignores f_type */
+       rc = md_getattr(sbi->ll_md_exp, op_data, &req);
+       OBD_FREE_PTR(op_data);
+       if (rc) {
+               /*
+                * Suppress erroneous/confusing messages when NFS
+                * is out of sync and requests old data.
+                */
+               CDEBUG(D_INFO, "can't get object attrs, fid "DFID", rc %d\n",
+                               PFID(fid), rc);
+               RETURN(ERR_PTR(rc));
+       }
+       rc = ll_prep_inode(&inode, req, sb, NULL);
+       ptlrpc_req_finished(req);
+       if (rc)
+               RETURN(ERR_PTR(rc));
+
+       RETURN(inode);
+}
+
+static struct dentry *
+ll_iget_for_nfs(struct super_block *sb, struct lu_fid *fid, struct lu_fid *parent)
+{
+       struct inode  *inode;
+       struct dentry *result;
+
+       ENTRY;
+
+       if (!fid_is_sane(fid))
+               RETURN(ERR_PTR(-ESTALE));
+
+       CDEBUG(D_INFO, "Get dentry for fid: "DFID"\n", PFID(fid));
+
+       inode = search_inode_for_lustre(sb, fid);
+       if (IS_ERR(inode))
+               RETURN(ERR_PTR(PTR_ERR(inode)));
+
+       if (is_bad_inode(inode)) {
+               /* we didn't find the right inode.. */
+               iput(inode);
+               RETURN(ERR_PTR(-ESTALE));
+       }
+
+       /* N.B. d_obtain_alias() drops inode ref on error */
+       result = d_obtain_alias(inode);
+       if (!IS_ERR(result)) {
+               int rc;
+
+               rc = ll_d_init(result);
+               if (rc < 0) {
+                       dput(result);
+                       result = ERR_PTR(rc);
+               } else {
+                       struct ll_dentry_data *ldd = ll_d2d(result);
+
+                       /*
+                        * Need to signal to the ll_file_open that
+                        * we came from NFS and so opencache needs to be
+                        * enabled for this one
+                        */
+                       spin_lock(&result->d_lock);
+                       ldd->lld_nfs_dentry = 1;
+                       spin_unlock(&result->d_lock);
+               }
+       }
+
+       RETURN(result);
+}
+
+#ifndef FILEID_INVALID
+#define FILEID_INVALID 0xff
+#endif
+#ifndef FILEID_LUSTRE
+#define FILEID_LUSTRE  0x97
+#endif
+
+/**
+ * \a connectable - is nfsd will connect himself or this should be done
+ *                  at lustre
+ *
+ * The return value is file handle type:
+ * 1 -- contains child file handle;
+ * 2 -- contains child file handle and parent file handle;
+ * 255 -- error.
+ */
+static int ll_encode_fh(struct inode *inode, u32 *fh, int *plen,
+                       struct inode *parent)
+{
+       int fileid_len = sizeof(struct lustre_file_handle) / 4;
+       struct lustre_file_handle *lfh = (void *)fh;
+
+       ENTRY;
+
+       CDEBUG(D_INFO, "%s: encoding for ("DFID") maxlen=%d minlen=%d\n",
+              ll_i2sbi(inode)->ll_fsname,
+              PFID(ll_inode2fid(inode)), *plen, fileid_len);
+
+       if (*plen < fileid_len) {
+               *plen = fileid_len;
+               RETURN(FILEID_INVALID);
+       }
+
+       lfh->lfh_child = *ll_inode2fid(inode);
+       if (parent)
+               lfh->lfh_parent = *ll_inode2fid(parent);
+       else
+               fid_zero(&lfh->lfh_parent);
+       *plen = fileid_len;
+
+       RETURN(FILEID_LUSTRE);
+}
 
-__u32 get_uuid2int(const char *name, int len)
+static int
+#ifndef HAVE_FILLDIR_USE_CTX
+ll_nfs_get_name_filldir(void *cookie, const char *name, int namelen,
+                       loff_t hash, u64 ino, unsigned type)
 {
-        __u32 key0 = 0x12a3fe2d, key1 = 0x37abe8f9;
-        while (len--) {
-                __u32 key = key1 + (key0 ^ (*name++ * 7152373));
-                if (key & 0x80000000) key -= 0x7fffffff;
-                key1 = key0;
-                key0 = key;
-        }
-        return (key0 << 1);
+       struct ll_getname_data *lgd = cookie;
+#else
+ll_nfs_get_name_filldir(struct dir_context *ctx, const char *name, int namelen,
+                       loff_t hash, u64 ino, unsigned type)
+{
+       struct ll_getname_data *lgd =
+               container_of(ctx, struct ll_getname_data, ctx);
+#endif /* HAVE_FILLDIR_USE_CTX */
+       /*
+        * It is hack to access lde_fid for comparison with lgd_fid.
+        * So the input 'name' must be part of the 'lu_dirent', and
+        * so must appear to be a non-const pointer to an empty array.
+        */
+       char (*n)[0] = (void *)name;
+       struct lu_dirent *lde = container_of0(n, struct lu_dirent, lde_name);
+       struct lu_fid fid;
+
+       fid_le_to_cpu(&fid, &lde->lde_fid);
+       if (lu_fid_eq(&fid, &lgd->lgd_fid)) {
+               memcpy(lgd->lgd_name, name, namelen);
+               lgd->lgd_name[namelen] = 0;
+               lgd->lgd_found = 1;
+       }
+        return lgd->lgd_found;
 }
 
-static struct inode * search_inode_for_lustre(struct super_block *sb,
-                                              unsigned long ino,
-                                              unsigned long generation,
-                                              int mode)
+static int ll_get_name(struct dentry *dentry, char *name,
+                       struct dentry *child)
 {
-        struct ptlrpc_request *req = NULL;
-        struct ll_sb_info *sbi = ll_s2sbi(sb);
-        struct ll_fid fid;
-        unsigned long valid = 0;
-        int eadatalen = 0, rc;
-        struct inode *inode = NULL;
-
-        inode = ILOOKUP(sb, ino, NULL, NULL);
-
-        if (inode)
-                return inode;
-        if (S_ISREG(mode)) {
-                eadatalen = obd_size_diskmd(sbi->ll_osc_exp, NULL);
-                valid |= OBD_MD_FLEASIZE;
-        }
-        fid.id = (__u64)ino;
-        fid.generation = generation;
-        fid.f_type = mode;
-
-        rc = md_getattr(sbi->ll_mdc_exp, &fid, valid, eadatalen, &req);
-        if (rc) {
-                CERROR("failure %d inode %lu\n", rc, ino);
-                return ERR_PTR(rc);
-        }
-
-        rc = ll_prep_inode(sbi->ll_osc_exp, sbi->ll_mdc_exp,
-                           &inode, req, 0, sb);
-        if (rc) {
-                ptlrpc_req_finished(req);
-                return ERR_PTR(rc);
-        }
-        ptlrpc_req_finished(req);
-
-        return inode;
+       struct inode *dir = dentry->d_inode;
+       struct ll_getname_data lgd = {
+               .lgd_name = name,
+               .lgd_fid = ll_i2info(child->d_inode)->lli_fid,
+#ifdef HAVE_DIR_CONTEXT
+               .ctx.actor = ll_nfs_get_name_filldir,
+#endif
+               .lgd_found = 0,
+       };
+       struct md_op_data *op_data;
+       u64 pos = 0;
+       int rc;
+
+       ENTRY;
+
+       if (!dir || !S_ISDIR(dir->i_mode))
+               GOTO(out, rc = -ENOTDIR);
+
+       if (!dir->i_fop)
+               GOTO(out, rc = -EINVAL);
+
+       op_data = ll_prep_md_op_data(NULL, dir, dir, NULL, 0, 0,
+                                    LUSTRE_OPC_ANY, dir);
+       if (IS_ERR(op_data))
+               GOTO(out, rc = PTR_ERR(op_data));
+
+       inode_lock(dir);
+#ifdef HAVE_DIR_CONTEXT
+       rc = ll_dir_read(dir, &pos, op_data, &lgd.ctx);
+#else
+       rc = ll_dir_read(dir, &pos, op_data, &lgd, ll_nfs_get_name_filldir);
+#endif
+       inode_unlock(dir);
+       ll_finish_md_op_data(op_data);
+       if (!rc && !lgd.lgd_found)
+               rc = -ENOENT;
+       EXIT;
+out:
+       return rc;
 }
 
-extern struct dentry_operations ll_d_ops;
+static struct dentry *ll_fh_to_dentry(struct super_block *sb, struct fid *fid,
+                                     int fh_len, int fh_type)
+{
+       struct lustre_file_handle *lfh = (struct lustre_file_handle *)fid;
+
+       if (fh_type != FILEID_LUSTRE)
+               RETURN(ERR_PTR(-EPROTO));
 
-static struct dentry *ll_iget_for_nfs(struct super_block *sb, unsigned long ino,
-                                      __u32 generation, umode_t mode)
+       RETURN(ll_iget_for_nfs(sb, &lfh->lfh_child, &lfh->lfh_parent));
+}
+
+static struct dentry *ll_fh_to_parent(struct super_block *sb, struct fid *fid,
+                                     int fh_len, int fh_type)
 {
-        struct inode *inode;
-        struct dentry *result;
-        struct list_head *lp;
-
-        if (ino == 0)
-                return ERR_PTR(-ESTALE);
-
-        inode = search_inode_for_lustre(sb, ino, generation, mode);
-        if (IS_ERR(inode)) {
-                return ERR_PTR(PTR_ERR(inode));
-        }
-        if (is_bad_inode(inode) 
-            || (generation && inode->i_generation != generation)
-            ){
-                /* we didn't find the right inode.. */
-              CERROR(" Inode %lu, Bad count: %lu %d or version  %u %u\n",
-                        inode->i_ino, 
-                        (unsigned long)inode->i_nlink, 
-                        atomic_read(&inode->i_count), 
-                        inode->i_generation, 
-                        generation);
-                iput(inode);
-                return ERR_PTR(-ESTALE);
-        }
-        
-        /* now to find a dentry.
-         * If possible, get a well-connected one
-         */
-        spin_lock(&dcache_lock);
-        for (lp = inode->i_dentry.next; lp != &inode->i_dentry ; lp=lp->next) {
-                result = list_entry(lp,struct dentry, d_alias);
-                if (!(result->d_flags & DCACHE_DISCONNECTED)) {
-                        dget_locked(result);
-                        ll_set_dflags(result, DCACHE_REFERENCED);
-                        spin_unlock(&dcache_lock);
-                        iput(inode);
-                        return result;
-                }
-        }
-        spin_unlock(&dcache_lock);
-        result = d_alloc_root(inode);
-        if (result == NULL) {
-                iput(inode);
-                return ERR_PTR(-ENOMEM);
-        }
-        result->d_flags |= DCACHE_DISCONNECTED;
-        
-        ll_set_dd(result);
-        result->d_op = &ll_d_ops;
-        return result;
+       struct lustre_file_handle *lfh = (struct lustre_file_handle *)fid;
+
+       if (fh_type != FILEID_LUSTRE)
+               RETURN(ERR_PTR(-EPROTO));
+
+       RETURN(ll_iget_for_nfs(sb, &lfh->lfh_parent, NULL));
 }
 
-struct dentry *ll_fh_to_dentry(struct super_block *sb, __u32 *data, int len,
-                               int fhtype, int parent)
+int ll_dir_get_parent_fid(struct inode *dir, struct lu_fid *parent_fid)
 {
-        switch (fhtype) {
-                case 2:
-                        if (len < 5)
-                                break;
-                        if (parent)
-                                return ll_iget_for_nfs(sb, data[3], 0, data[4]);
-                case 1:
-                        if (len < 3)
-                                break;
-                        if (parent)
-                                break;
-                        return ll_iget_for_nfs(sb, data[0], data[1], data[2]);
-                default: break;
-        }
-        return ERR_PTR(-EINVAL);
+       struct ptlrpc_request   *req = NULL;
+       struct ll_sb_info       *sbi;
+       struct mdt_body         *body;
+       static const char       dotdot[] = "..";
+       struct md_op_data       *op_data;
+       int                     rc;
+       int                     lmmsize;
+
+       ENTRY;
+
+       LASSERT(dir && S_ISDIR(dir->i_mode));
+
+       sbi = ll_s2sbi(dir->i_sb);
+
+       CDEBUG(D_INFO, "%s: getting parent for ("DFID")\n",
+              sbi->ll_fsname, PFID(ll_inode2fid(dir)));
+
+       rc = ll_get_default_mdsize(sbi, &lmmsize);
+       if (rc != 0)
+               RETURN(rc);
+
+       op_data = ll_prep_md_op_data(NULL, dir, NULL, dotdot,
+                                    strlen(dotdot), lmmsize,
+                                    LUSTRE_OPC_ANY, NULL);
+       if (IS_ERR(op_data))
+               RETURN(PTR_ERR(op_data));
+
+       rc = md_getattr_name(sbi->ll_md_exp, op_data, &req);
+       ll_finish_md_op_data(op_data);
+       if (rc != 0) {
+               CERROR("%s: failure inode "DFID" get parent: rc = %d\n",
+                      sbi->ll_fsname, PFID(ll_inode2fid(dir)), rc);
+               RETURN(rc);
+       }
+       body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
+
+       /*
+        * LU-3952: MDT may lost the FID of its parent, we should not crash
+        * the NFS server, ll_iget_for_nfs() will handle the error.
+        */
+       if (body->mbo_valid & OBD_MD_FLID) {
+               CDEBUG(D_INFO, "parent for "DFID" is "DFID"\n",
+                      PFID(ll_inode2fid(dir)), PFID(&body->mbo_fid1));
+               *parent_fid = body->mbo_fid1;
+       }
+
+       ptlrpc_req_finished(req);
+       RETURN(0);
 }
 
-int ll_dentry_to_fh(struct dentry *dentry, __u32 *datap, int *lenp,
-                    int need_parent)
+static struct dentry *ll_get_parent(struct dentry *dchild)
 {
-        if (*lenp < 3)
-                return 255;
-        *datap++ = dentry->d_inode->i_ino;
-        *datap++ = dentry->d_inode->i_generation;
-        *datap++ = (__u32)(S_IFMT & dentry->d_inode->i_mode);
-
-        if (*lenp == 3 || S_ISDIR(dentry->d_inode->i_mode)) {
-                *lenp = 3;
-                return 1;
-        }
-        if (dentry->d_parent) { 
-                *datap++ = dentry->d_parent->d_inode->i_ino;
-                *datap++ = (__u32)(S_IFMT & dentry->d_parent->d_inode->i_mode);
-         
-                *lenp = 5;
-                return 2;
-        }
-        *lenp = 3;
-        return 1;
+       struct lu_fid parent_fid = { 0 };
+       int rc;
+       struct dentry *dentry;
+
+       ENTRY;
+
+       rc = ll_dir_get_parent_fid(dchild->d_inode, &parent_fid);
+       if (rc != 0)
+               RETURN(ERR_PTR(rc));
+
+       dentry = ll_iget_for_nfs(dchild->d_inode->i_sb, &parent_fid, NULL);
+
+       RETURN(dentry);
 }
+
+struct export_operations lustre_export_operations = {
+       .get_parent = ll_get_parent,
+       .encode_fh  = ll_encode_fh,
+       .get_name   = ll_get_name,
+       .fh_to_dentry = ll_fh_to_dentry,
+       .fh_to_parent = ll_fh_to_parent,
+};