Whamcloud - gitweb
LU-14677 sec: migrate/extend/split on encrypted file
[fs/lustre-release.git] / lustre / llite / namei.c
index b84dfd7..29d4806 100644 (file)
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
- * Lustre is a trademark of Sun Microsystems, Inc.
  */
 
 #include <linux/fs.h>
 #include <linux/sched.h>
 #include <linux/mm.h>
+#include <linux/file.h>
 #include <linux/quotaops.h>
 #include <linux/highmem.h>
 #include <linux/pagemap.h>
-#include <linux/security.h>
 #include <linux/user_namespace.h>
 #include <linux/uidgid.h>
 
@@ -212,6 +211,8 @@ static int ll_dom_lock_cancel(struct inode *inode, struct ldlm_lock *lock)
        if (IS_ERR(env))
                RETURN(PTR_ERR(env));
 
+       OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_REPLAY_PAUSE, cfs_fail_val);
+
        /* reach MDC layer to flush data under  the DoM ldlm lock */
        rc = cl_object_flush(env, lli->lli_clob, lock);
        if (rc == -ENODATA) {
@@ -316,7 +317,7 @@ static void ll_lock_cancel_bits(struct ldlm_lock *lock, __u64 to_cancel)
        lli = ll_i2info(inode);
 
        if (bits & MDS_INODELOCK_UPDATE)
-               ll_file_set_flag(lli, LLIF_UPDATE_ATIME);
+               set_bit(LLIF_UPDATE_ATIME, &lli->lli_flags);
 
        if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
                CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
@@ -370,8 +371,7 @@ static void ll_lock_cancel_bits(struct ldlm_lock *lock, __u64 to_cancel)
        }
 
        if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
-           inode->i_sb->s_root != NULL &&
-           inode != inode->i_sb->s_root->d_inode)
+           !is_root_inode(inode))
                ll_prune_aliases(inode);
 
        if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM))
@@ -614,6 +614,27 @@ struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
        if (rc < 0)
                return ERR_PTR(rc);
        d_add(de, inode);
+
+       /* this needs only to be done for foreign symlink dirs as
+        * DCACHE_SYMLINK_TYPE is already set by d_flags_for_inode()
+        * kernel routine for files with symlink ops (ie, real symlink)
+        */
+       if (inode && S_ISDIR(inode->i_mode) &&
+           ll_sbi_has_foreign_symlink(ll_i2sbi(inode)) &&
+#ifdef HAVE_IOP_GET_LINK
+           inode->i_op->get_link) {
+#else
+           inode->i_op->follow_link) {
+#endif
+               CDEBUG(D_INFO, "%s: inode "DFID": faking foreign dir as a symlink\n",
+                      ll_i2sbi(inode)->ll_fsname, PFID(ll_inode2fid(inode)));
+               spin_lock(&de->d_lock);
+               /* like d_flags_for_inode() already does for files */
+               de->d_flags = (de->d_flags & ~DCACHE_ENTRY_TYPE) |
+                             DCACHE_SYMLINK_TYPE;
+               spin_unlock(&de->d_lock);
+       }
+
        CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
               de, de->d_inode, ll_d_count(de), de->d_flags);
         return de;
@@ -641,7 +662,7 @@ static int ll_lookup_it_finish(struct ptlrpc_request *request,
                struct mdt_body *body = req_capsule_server_get(pill,
                                                               &RMF_MDT_BODY);
 
-               rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
+               rc = ll_prep_inode(&inode, &request->rq_pill, (*de)->d_sb, it);
                if (rc)
                        RETURN(rc);
 
@@ -781,7 +802,7 @@ static int ll_lookup_it_finish(struct ptlrpc_request *request,
 out:
        if (rc != 0 && it->it_op & IT_OPEN) {
                ll_intent_drop_lock(it);
-               ll_open_cleanup((*de)->d_sb, request);
+               ll_open_cleanup((*de)->d_sb, &request->rq_pill);
        }
 
        return rc;
@@ -862,9 +883,90 @@ static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
                        *secctxlen = 0;
        }
        if (it->it_op & IT_CREAT && encrypt) {
-               rc = llcrypt_inherit_context(parent, NULL, op_data, false);
-               if (rc)
-                       GOTO(out, retval = ERR_PTR(rc));
+               /* Volatile file name may look like:
+                * <parent>/LUSTRE_VOLATILE_HDR:<mdt_index>:<random>:fd=<fd>
+                * where fd is opened descriptor of reference file.
+                */
+               if (unlikely(filename_is_volatile(dentry->d_name.name,
+                                                 dentry->d_name.len, NULL))) {
+                       int ctx_size = LLCRYPT_ENC_CTX_SIZE;
+                       struct lustre_sb_info *lsi;
+                       struct file *ref_file;
+                       struct inode *ref_inode;
+                       char *p, *q, *fd_str;
+                       void *ctx;
+                       int fd;
+
+                       p = strnstr(dentry->d_name.name, ":fd=",
+                                   dentry->d_name.len);
+                       if (!p || strlen(p + 4) == 0)
+                               GOTO(out, retval = ERR_PTR(-EINVAL));
+
+                       q = strchrnul(p + 4, ':');
+                       fd_str = kstrndup(p + 4, q - p - 4, GFP_NOFS);
+                       if (!fd_str)
+                               GOTO(out, retval = ERR_PTR(-ENOMEM));
+                       rc = kstrtouint(fd_str, 10, &fd);
+                       kfree(fd_str);
+                       if (rc)
+                               GOTO(inherit, rc = -EINVAL);
+
+                       ref_file = fget(fd);
+                       if (!ref_file)
+                               GOTO(inherit, rc = -EINVAL);
+
+                       ref_inode = file_inode(ref_file);
+                       if (!ref_inode) {
+                               fput(ref_file);
+                               GOTO(inherit, rc = -EINVAL);
+                       }
+
+                       lsi = s2lsi(ref_inode->i_sb);
+
+getctx:
+                       OBD_ALLOC(ctx, ctx_size);
+                       if (!ctx)
+                               GOTO(out, retval = ERR_PTR(-ENOMEM));
+
+#ifdef CONFIG_LL_ENCRYPTION
+                       rc = lsi->lsi_cop->get_context(ref_inode,
+                                                      ctx, ctx_size);
+#else
+                       rc = -ENODATA;
+#endif
+                       if (rc == -ERANGE) {
+                               OBD_FREE(ctx, ctx_size);
+                               ctx_size *= 2;
+                               goto getctx;
+                       }
+                       fput(ref_file);
+                       if (rc < 0) {
+                               OBD_FREE(ctx, ctx_size);
+                               GOTO(inherit, rc);
+                       }
+
+                       op_data->op_file_encctx_size = rc;
+                       if (rc == ctx_size) {
+                               op_data->op_file_encctx = ctx;
+                       } else {
+                               OBD_ALLOC(op_data->op_file_encctx,
+                                         op_data->op_file_encctx_size);
+                               if (!op_data->op_file_encctx) {
+                                       OBD_FREE(ctx, ctx_size);
+                                       GOTO(out, retval = ERR_PTR(-ENOMEM));
+                               }
+                               memcpy(op_data->op_file_encctx, ctx,
+                                      op_data->op_file_encctx_size);
+                               OBD_FREE(ctx, ctx_size);
+                       }
+
+               } else {
+inherit:
+                       rc = llcrypt_inherit_context(parent, NULL, op_data,
+                                                    false);
+                       if (rc)
+                               GOTO(out, retval = ERR_PTR(rc));
+               }
                if (encctx != NULL)
                        *encctx = op_data->op_file_encctx;
                if (encctxlen != NULL)
@@ -894,8 +996,6 @@ static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
        }
 
        if (pca && pca->pca_dataset) {
-               struct pcc_dataset *dataset = pca->pca_dataset;
-
                OBD_ALLOC_PTR(lum);
                if (lum == NULL)
                        GOTO(out, retval = ERR_PTR(-ENOMEM));
@@ -904,18 +1004,7 @@ static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
                lum->lmm_pattern = LOV_PATTERN_F_RELEASED | LOV_PATTERN_RAID0;
                op_data->op_data = lum;
                op_data->op_data_size = sizeof(*lum);
-               op_data->op_archive_id = dataset->pccd_rwid;
-
-               rc = obd_fid_alloc(NULL, ll_i2mdexp(parent), &op_data->op_fid2,
-                                  op_data);
-               if (rc)
-                       GOTO(out, retval = ERR_PTR(rc));
-
-               rc = pcc_inode_create(parent->i_sb, dataset, &op_data->op_fid2,
-                                     &pca->pca_dentry);
-               if (rc)
-                       GOTO(out, retval = ERR_PTR(rc));
-
+               op_data->op_archive_id = pca->pca_dataset->pccd_rwid;
                it->it_flags |= MDS_OPEN_PCC;
        }
 
@@ -950,6 +1039,14 @@ static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
        if (rc < 0)
                GOTO(out, retval = ERR_PTR(rc));
 
+       if (pca && pca->pca_dataset) {
+               rc = pcc_inode_create(parent->i_sb, pca->pca_dataset,
+                                     &op_data->op_fid2,
+                                     &pca->pca_dentry);
+               if (rc)
+                       GOTO(out, retval = ERR_PTR(rc));
+       }
+
        /* dir layout may change */
        ll_unlock_md_op_lsm(op_data);
        rc = ll_lookup_it_finish(req, it, parent, &dentry,
@@ -1137,6 +1234,15 @@ static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
 
        OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE2, cfs_fail_val);
 
+       /* We can only arrive at this path when we have no inode, so
+        * we only need to request open lock if it was requested
+        * for every open
+        */
+       if (ll_i2sbi(dir)->ll_oc_thrsh_count == 1 &&
+           exp_connect_flags2(ll_i2mdexp(dir)) &
+           OBD_CONNECT2_ATOMIC_OPEN_LOCK)
+               it->it_flags |= MDS_OPEN_LOCK;
+
        /* Dentry added to dcache tree in ll_lookup_it */
        de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen, &pca, encrypt,
                          &encctx, &encctxlen);
@@ -1152,7 +1258,7 @@ static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
                        /* Dentry instantiated in ll_create_it. */
                        rc = ll_create_it(dir, dentry, it, secctx, secctxlen,
                                          encrypt, encctx, encctxlen);
-                       security_release_secctx(secctx, secctxlen);
+                       ll_security_release_secctx(secctx, secctxlen);
                        llcrypt_free_ctx(encctx, encctxlen);
                        if (rc) {
                                /* We dget in ll_splice_alias. */
@@ -1195,7 +1301,9 @@ static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
                        }
                }
 
-               if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
+               /* check also if a foreign file is openable */
+               if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN) &&
+                   ll_foreign_is_openable(dentry, open_flags)) {
                        /* Open dentry. */
                        if (S_ISFIFO(dentry->d_inode->i_mode)) {
                                /* We cannot call open here as it might
@@ -1238,10 +1346,10 @@ static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
 
        LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
        request = it->it_request;
-        it_clear_disposition(it, DISP_ENQ_CREATE_REF);
-        rc = ll_prep_inode(&inode, request, dir->i_sb, it);
-        if (rc)
-                GOTO(out, inode = ERR_PTR(rc));
+       it_clear_disposition(it, DISP_ENQ_CREATE_REF);
+       rc = ll_prep_inode(&inode, &request->rq_pill, dir->i_sb, it);
+       if (rc)
+               GOTO(out, inode = ERR_PTR(rc));
 
        /* Pause to allow for a race with concurrent access by fid */
        OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
@@ -1437,6 +1545,10 @@ again:
                        md.default_lmv->lsm_md_master_mdt_index =
                                lum->lum_stripe_offset;
                        md.default_lmv->lsm_md_hash_type = lum->lum_hash_type;
+                       md.default_lmv->lsm_md_max_inherit =
+                               lum->lum_max_inherit;
+                       md.default_lmv->lsm_md_max_inherit_rr =
+                               lum->lum_max_inherit_rr;
 
                        err = ll_update_inode(dir, &md);
                        md_free_lustre_md(sbi->ll_md_exp, &md);
@@ -1470,7 +1582,7 @@ again:
 
        CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
 
-       err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
+       err = ll_prep_inode(&inode, &request->rq_pill, dchild->d_sb, NULL);
        if (err)
                GOTO(err_exit, err);
 
@@ -1681,6 +1793,10 @@ static int ll_rmdir(struct inode *dir, struct dentry *dchild)
        if (unlikely(d_mountpoint(dchild)))
                 RETURN(-EBUSY);
 
+       /* some foreign dir may not be allowed to be removed */
+       if (!ll_foreign_is_removable(dchild, false))
+               RETURN(-EPERM);
+
        op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
                                     S_IFDIR, LUSTRE_OPC_ANY, NULL);
        if (IS_ERR(op_data))
@@ -1692,13 +1808,25 @@ static int ll_rmdir(struct inode *dir, struct dentry *dchild)
        op_data->op_fid2 = op_data->op_fid3;
        rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
        ll_finish_md_op_data(op_data);
-       if (!rc)
-               ll_update_times(request, dir);
+       if (!rc) {
+               struct mdt_body *body;
 
-       ptlrpc_req_finished(request);
-       if (!rc)
+               ll_update_times(request, dir);
                ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
                                   ktime_us_delta(ktime_get(), kstart));
+
+               /*
+                * The server puts attributes in on the last unlink, use them
+                * to update the link count so the inode can be freed
+                * immediately.
+                */
+               body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
+               if (body->mbo_valid & OBD_MD_FLNLINK)
+                       set_nlink(dchild->d_inode, body->mbo_nlink);
+       }
+
+       ptlrpc_req_finished(request);
+
        RETURN(rc);
 }
 
@@ -1754,6 +1882,10 @@ static int ll_unlink(struct inode *dir, struct dentry *dchild)
        if (unlikely(d_mountpoint(dchild)))
                RETURN(-EBUSY);
 
+       /* some foreign file/dir may not be allowed to be unlinked */
+       if (!ll_foreign_is_removable(dchild, false))
+               RETURN(-EPERM);
+
        op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
                                     LUSTRE_OPC_ANY, NULL);
        if (IS_ERR(op_data))
@@ -1802,6 +1934,7 @@ static int ll_rename(struct inode *src, struct dentry *src_dchild,
        struct ll_sb_info *sbi = ll_i2sbi(src);
        struct md_op_data *op_data;
        ktime_t kstart = ktime_get();
+       umode_t mode = 0;
        int err;
        ENTRY;
 
@@ -1825,16 +1958,27 @@ static int ll_rename(struct inode *src, struct dentry *src_dchild,
 #endif
        if (err)
                RETURN(err);
+       /* we prevent an encrypted file from being renamed
+        * into an unencrypted dir
+        */
+       if (IS_ENCRYPTED(src) && !IS_ENCRYPTED(tgt))
+               RETURN(-EXDEV);
+
+       if (src_dchild->d_inode)
+               mode = src_dchild->d_inode->i_mode;
+
+       if (tgt_dchild->d_inode)
+               mode = tgt_dchild->d_inode->i_mode;
 
-       op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
+       op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, mode,
                                     LUSTRE_OPC_ANY, NULL);
        if (IS_ERR(op_data))
                RETURN(PTR_ERR(op_data));
 
-       if (src_dchild->d_inode != NULL)
+       if (src_dchild->d_inode)
                op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
 
-       if (tgt_dchild->d_inode != NULL)
+       if (tgt_dchild->d_inode)
                op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
 
        err = md_rename(sbi->ll_md_exp, op_data,