Whamcloud - gitweb
LU-6436 llite: NULL pointer dereference in cl_object_top()
[fs/lustre-release.git] / lustre / llite / namei.c
index b687d6e..de168ab 100644 (file)
@@ -23,7 +23,7 @@
  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2011, 2015, Intel Corporation.
+ * Copyright (c) 2011, 2016, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -51,7 +51,8 @@
 #include "llite_internal.h"
 
 static int ll_create_it(struct inode *dir, struct dentry *dentry,
-                       struct lookup_intent *it);
+                       struct lookup_intent *it,
+                       void *secctx, __u32 secctxlen);
 
 /* called from iget5_locked->find_inode() under inode_lock spinlock */
 static int ll_test_inode(struct inode *inode, void *opaque)
@@ -135,8 +136,12 @@ struct inode *ll_iget(struct super_block *sb, ino_t hash,
                        iput(inode);
                        inode = ERR_PTR(rc);
                } else {
+                       inode_has_no_xattr(inode);
                        unlock_new_inode(inode);
                }
+       } else if (is_bad_inode(inode)) {
+               iput(inode);
+               inode = ERR_PTR(-ESTALE);
        } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
                rc = ll_update_inode(inode, md);
                CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
@@ -271,12 +276,7 @@ int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 
                if (bits & MDS_INODELOCK_UPDATE) {
                        struct ll_inode_info *lli = ll_i2info(inode);
-
-                       spin_lock(&lli->lli_lock);
-                       LTIME_S(inode->i_mtime) = 0;
-                       LTIME_S(inode->i_atime) = 0;
-                       LTIME_S(inode->i_ctime) = 0;
-                       spin_unlock(&lli->lli_lock);
+                       lli->lli_update_atime = 1;
                }
 
                if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
@@ -546,7 +546,8 @@ out:
 }
 
 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
-                                  struct lookup_intent *it)
+                                  struct lookup_intent *it,
+                                  void **secctx, __u32 *secctxlen)
 {
        struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
        struct dentry *save = dentry, *retval;
@@ -602,6 +603,10 @@ static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
                                             &op_data->op_file_secctx_size);
                if (rc < 0)
                        GOTO(out, retval = ERR_PTR(rc));
+               if (secctx != NULL)
+                       *secctx = op_data->op_file_secctx;
+               if (secctxlen != NULL)
+                       *secctxlen = op_data->op_file_secctx_size;
        }
 
        rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
@@ -651,8 +656,15 @@ static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
        GOTO(out, retval = (dentry == save) ? NULL : dentry);
 
 out:
-       if (op_data != NULL && !IS_ERR(op_data))
+       if (op_data != NULL && !IS_ERR(op_data)) {
+               if (secctx != NULL && secctxlen != NULL) {
+                       /* caller needs sec ctx info, so reset it in op_data to
+                        * prevent it from being freed */
+                       op_data->op_file_secctx = NULL;
+                       op_data->op_file_secctx_size = 0;
+               }
                ll_finish_md_op_data(op_data);
+       }
 
        ptlrpc_req_finished(req);
        return retval;
@@ -682,7 +694,7 @@ static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
                itp = NULL;
        else
                itp = &it;
-       de = ll_lookup_it(parent, dentry, itp);
+       de = ll_lookup_it(parent, dentry, itp, NULL, NULL);
 
        if (itp != NULL)
                ll_intent_release(itp);
@@ -701,6 +713,8 @@ static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
        struct lookup_intent *it;
        struct dentry *de;
        long long lookup_flags = LOOKUP_OPEN;
+       void *secctx = NULL;
+       __u32 secctxlen = 0;
        int rc = 0;
        ENTRY;
 
@@ -741,7 +755,7 @@ static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
        it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
 
        /* Dentry added to dcache tree in ll_lookup_it */
-       de = ll_lookup_it(dir, dentry, it);
+       de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen);
        if (IS_ERR(de))
                rc = PTR_ERR(de);
        else if (de != NULL)
@@ -752,7 +766,8 @@ static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
        if (!rc) {
                if (it_disposition(it, DISP_OPEN_CREATE)) {
                        /* Dentry instantiated in ll_create_it. */
-                       rc = ll_create_it(dir, dentry, it);
+                       rc = ll_create_it(dir, dentry, it, secctx, secctxlen);
+                       security_release_secctx(secctx, secctxlen);
                        if (rc) {
                                /* We dget in ll_splice_alias. */
                                if (de != NULL)
@@ -792,7 +807,7 @@ out_release:
 
 #else /* !HAVE_IOP_ATOMIC_OPEN */
 static struct lookup_intent *
-ll_convert_intent(struct open_intent *oit, int lookup_flags)
+ll_convert_intent(struct open_intent *oit, int lookup_flags, bool is_readonly)
 {
        struct lookup_intent *it;
 
@@ -802,10 +817,12 @@ ll_convert_intent(struct open_intent *oit, int lookup_flags)
 
        if (lookup_flags & LOOKUP_OPEN) {
                it->it_op = IT_OPEN;
-               if (lookup_flags & LOOKUP_CREATE)
+               /* Avoid file creation for ro bind mount point(is_readonly) */
+               if ((lookup_flags & LOOKUP_CREATE) && !is_readonly)
                        it->it_op |= IT_CREAT;
                it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
-               it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags);
+               it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags &
+                                               ~(is_readonly ? O_CREAT : 0));
                it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
        } else {
                it->it_op = IT_GETATTR;
@@ -817,16 +834,16 @@ ll_convert_intent(struct open_intent *oit, int lookup_flags)
 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
                                    struct nameidata *nd)
 {
-        struct dentry *de;
-        ENTRY;
+       struct dentry *de;
+       ENTRY;
 
-        if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
-                struct lookup_intent *it;
+       if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
+               struct lookup_intent *it;
 
-                if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
-                        it = ll_d2d(dentry)->lld_it;
-                        ll_d2d(dentry)->lld_it = NULL;
-                } else {
+               if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
+                       it = ll_d2d(dentry)->lld_it;
+                       ll_d2d(dentry)->lld_it = NULL;
+               } else {
                        /*
                         * Optimize away (CREATE && !OPEN). Let .create handle
                         * the race. But only if we have write permissions
@@ -839,23 +856,25 @@ static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
                                              MAY_WRITE | MAY_EXEC) == 0))
                                RETURN(NULL);
 
-                        it = ll_convert_intent(&nd->intent.open, nd->flags);
-                        if (IS_ERR(it))
-                                RETURN((struct dentry *)it);
-                }
-
-               de = ll_lookup_it(parent, dentry, it);
-                if (de)
-                        dentry = de;
-                if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
-                        if (dentry->d_inode &&
-                            it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
-                                if (S_ISFIFO(dentry->d_inode->i_mode)) {
+                       it = ll_convert_intent(&nd->intent.open, nd->flags,
+                               (nd->path.mnt->mnt_flags & MNT_READONLY) ||
+                               (nd->path.mnt->mnt_sb->s_flags & MS_RDONLY));
+                       if (IS_ERR(it))
+                               RETURN((struct dentry *)it);
+               }
+
+               de = ll_lookup_it(parent, dentry, it, NULL, NULL);
+               if (de)
+                       dentry = de;
+               if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
+                       if (dentry->d_inode &&
+                           it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
+                               if (S_ISFIFO(dentry->d_inode->i_mode)) {
                                        /* We cannot call open here as it might
                                         * deadlock. This case is unreachable in
                                         * practice because of
                                         * OBD_CONNECT_NODEVOH. */
-                                } else {
+                               } else {
                                        struct file *filp;
 
                                        nd->intent.open.file->private_data = it;
@@ -867,24 +886,24 @@ static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
                                                        dput(de);
                                                de = (struct dentry *)filp;
                                        }
-                                }
-                        } else if (it_disposition(it, DISP_OPEN_CREATE)) {
-                                // XXX This can only reliably work on assumption
-                                // that there are NO hashed negative dentries.
-                                ll_d2d(dentry)->lld_it = it;
-                                it = NULL; /* Will be freed in ll_create_nd */
-                                /* We absolutely depend on ll_create_nd to be
-                                 * called to not leak this intent and possible
-                                 * data attached to it */
-                        }
-                }
-
-                if (it) {
-                        ll_intent_release(it);
-                        OBD_FREE(it, sizeof(*it));
-                }
-        } else {
-               de = ll_lookup_it(parent, dentry, NULL);
+                               }
+                       } else if (it_disposition(it, DISP_OPEN_CREATE)) {
+                               /* XXX This can only reliably work on assumption
+                                * that there are NO hashed negative dentries.*/
+                               ll_d2d(dentry)->lld_it = it;
+                               it = NULL; /* Will be freed in ll_create_nd */
+                               /* We absolutely depend on ll_create_nd to be
+                                * called to not leak this intent and possible
+                                * data attached to it */
+                       }
+               }
+
+               if (it) {
+                       ll_intent_release(it);
+                       OBD_FREE(it, sizeof(*it));
+               }
+       } else {
+               de = ll_lookup_it(parent, dentry, NULL, NULL, NULL);
        }
 
        RETURN(de);
@@ -909,7 +928,8 @@ static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
         if (rc)
                 GOTO(out, inode = ERR_PTR(rc));
 
-       LASSERT(ll_d_hlist_empty(&inode->i_dentry));
+       /* Pause to allow for a race with concurrent access by fid */
+       OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
 
         /* We asked for a lock on the directory, but were granted a
          * lock on the inode.  Since we finally have an inode pointer,
@@ -938,7 +958,8 @@ static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
  * with d_instantiate().
  */
 static int ll_create_it(struct inode *dir, struct dentry *dentry,
-                       struct lookup_intent *it)
+                       struct lookup_intent *it,
+                       void *secctx, __u32 secctxlen)
 {
        struct inode *inode;
        int rc = 0;
@@ -956,6 +977,18 @@ static int ll_create_it(struct inode *dir, struct dentry *dentry,
        if (IS_ERR(inode))
                RETURN(PTR_ERR(inode));
 
+       if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) &&
+           secctx != NULL) {
+               inode_lock(inode);
+               /* must be done before d_instantiate, because it calls
+                * security_d_instantiate, which means a getxattr if security
+                * context is not set yet */
+               rc = security_inode_notifysecctx(inode, secctx, secctxlen);
+               inode_unlock(inode);
+               if (rc)
+                       RETURN(rc);
+       }
+
        d_instantiate(dentry, inode);
 
        if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
@@ -1020,8 +1053,6 @@ again:
                        from_kuid(&init_user_ns, current_fsuid()),
                        from_kgid(&init_user_ns, current_fsgid()),
                        cfs_curproc_cap_pack(), rdev, &request);
-       ll_finish_md_op_data(op_data);
-       op_data = NULL;
        if (err < 0 && err != -EREMOTE)
                GOTO(err_exit, err);
 
@@ -1055,6 +1086,7 @@ again:
 
                ptlrpc_req_finished(request);
                request = NULL;
+               ll_finish_md_op_data(op_data);
                goto again;
        }
 
@@ -1066,6 +1098,19 @@ again:
        if (err)
                GOTO(err_exit, err);
 
+       if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
+               inode_lock(inode);
+               /* must be done before d_instantiate, because it calls
+                * security_d_instantiate, which means a getxattr if security
+                * context is not set yet */
+               err = security_inode_notifysecctx(inode,
+                                                 op_data->op_file_secctx,
+                                                 op_data->op_file_secctx_size);
+               inode_unlock(inode);
+               if (err)
+                       GOTO(err_exit, err);
+       }
+
        d_instantiate(dchild, inode);
 
        if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
@@ -1139,6 +1184,8 @@ static int ll_create_nd(struct inode *dir, struct dentry *dentry,
               dentry->d_name.name, PFID(ll_inode2fid(dir)),
               dir, mode, want_excl);
 
+       /* Using mknod(2) to create a regular file is designed to not recognize
+        * volatile file name, so we use ll_mknod() here. */
        rc = ll_mknod(dir, dentry, mode, 0);
 
        ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
@@ -1161,19 +1208,26 @@ static int ll_create_nd(struct inode *dir, struct dentry *dentry,
        if (lld != NULL)
                it = lld->lld_it;
 
-        if (!it)
-               return ll_mknod(dir, dentry, mode, 0);
+       if (!it) {
+               /* LU-8559: use LUSTRE_OPC_CREATE for non atomic open case
+                * so that volatile file name is recoginized.
+                * Mknod(2), however, is designed to not recognize volatile
+                * file name to avoid inode leak under orphan directory until
+                * MDT reboot */
+               return ll_new_node(dir, dentry, NULL, mode, 0,
+                                  LUSTRE_OPC_CREATE);
+       }
 
        lld->lld_it = NULL;
 
-        /* Was there an error? Propagate it! */
+       /* Was there an error? Propagate it! */
        if (it->it_status) {
                rc = it->it_status;
-                goto out;
-        }
+               goto out;
+       }
 
-       rc = ll_create_it(dir, dentry, it);
-        if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
+       rc = ll_create_it(dir, dentry, it, NULL, 0);
+       if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
                struct file *filp;
 
                nd->intent.open.file->private_data = it;
@@ -1370,15 +1424,25 @@ static int ll_unlink(struct inode *dir, struct dentry *dchild)
 }
 
 static int ll_rename(struct inode *src, struct dentry *src_dchild,
-                    struct inode *tgt, struct dentry *tgt_dchild)
+                    struct inode *tgt, struct dentry *tgt_dchild
+#ifdef HAVE_IOPS_RENAME_WITH_FLAGS
+                    , unsigned int flags
+#endif
+                    )
 {
        struct qstr *src_name = &src_dchild->d_name;
        struct qstr *tgt_name = &tgt_dchild->d_name;
-        struct ptlrpc_request *request = NULL;
-        struct ll_sb_info *sbi = ll_i2sbi(src);
-        struct md_op_data *op_data;
-        int err;
-        ENTRY;
+       struct ptlrpc_request *request = NULL;
+       struct ll_sb_info *sbi = ll_i2sbi(src);
+       struct md_op_data *op_data;
+       int err;
+       ENTRY;
+
+#ifdef HAVE_IOPS_RENAME_WITH_FLAGS
+       if (flags)
+               return -EINVAL;
+#endif
+
        CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
               "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
               src_name->len, src_name->name,
@@ -1418,28 +1482,33 @@ static int ll_rename(struct inode *src, struct dentry *src_dchild,
 }
 
 const struct inode_operations ll_dir_inode_operations = {
-       .mknod              = ll_mknod,
+       .mknod          = ll_mknod,
 #ifdef HAVE_IOP_ATOMIC_OPEN
-       .atomic_open        = ll_atomic_open,
+       .atomic_open    = ll_atomic_open,
 #endif
-       .lookup             = ll_lookup_nd,
-       .create             = ll_create_nd,
+       .lookup         = ll_lookup_nd,
+       .create         = ll_create_nd,
        /* We need all these non-raw things for NFSD, to not patch it. */
-       .unlink             = ll_unlink,
-       .mkdir              = ll_mkdir,
-       .rmdir              = ll_rmdir,
-       .symlink            = ll_symlink,
-       .link               = ll_link,
-       .rename             = ll_rename,
-       .setattr            = ll_setattr,
-       .getattr            = ll_getattr,
-       .permission         = ll_inode_permission,
-       .setxattr           = ll_setxattr,
-       .getxattr           = ll_getxattr,
-       .listxattr          = ll_listxattr,
-       .removexattr        = ll_removexattr,
+       .unlink         = ll_unlink,
+       .mkdir          = ll_mkdir,
+       .rmdir          = ll_rmdir,
+       .symlink        = ll_symlink,
+       .link           = ll_link,
+       .rename         = ll_rename,
+       .setattr        = ll_setattr,
+       .getattr        = ll_getattr,
+       .permission     = ll_inode_permission,
+#ifdef HAVE_IOP_XATTR
+       .setxattr       = ll_setxattr,
+       .getxattr       = ll_getxattr,
+       .removexattr    = ll_removexattr,
+#endif
+       .listxattr      = ll_listxattr,
 #ifdef HAVE_IOP_GET_ACL
-       .get_acl            = ll_get_acl,
+       .get_acl        = ll_get_acl,
+#endif
+#ifdef HAVE_IOP_SET_ACL
+       .set_acl        = ll_set_acl,
 #endif
 };
 
@@ -1447,11 +1516,16 @@ const struct inode_operations ll_special_inode_operations = {
        .setattr        = ll_setattr,
        .getattr        = ll_getattr,
        .permission     = ll_inode_permission,
-       .setxattr       = ll_setxattr,
-       .getxattr       = ll_getxattr,
-       .listxattr      = ll_listxattr,
+#ifdef HAVE_IOP_XATTR
+       .setxattr       = ll_setxattr,
+       .getxattr       = ll_getxattr,
        .removexattr    = ll_removexattr,
+#endif
+       .listxattr      = ll_listxattr,
 #ifdef HAVE_IOP_GET_ACL
-       .get_acl            = ll_get_acl,
+       .get_acl        = ll_get_acl,
+#endif
+#ifdef HAVE_IOP_SET_ACL
+       .set_acl        = ll_set_acl,
 #endif
 };