pvfs_open: the pvfs_odb_retry structs need to be children of the request
[samba.git] / source4 / ntvfs / posix / pvfs_open.c
index a01352f60cff54cf5064e5d9d850d85e799b5edf..d1cb0eb6034db4fbc89710753f64291d580f9837 100644 (file)
@@ -50,29 +50,10 @@ struct pvfs_file *pvfs_find_fd(struct pvfs_state *pvfs,
 */
 static int pvfs_dir_handle_destructor(struct pvfs_file_handle *h)
 {
-       int open_count;
-       char *path = NULL;
-
-       if (h->name->stream_name == NULL && 
-           pvfs_delete_on_close_set(h->pvfs, h, &open_count, &path) &&
-           open_count == 1) {
-               NTSTATUS status;
-               status = pvfs_xattr_unlink_hook(h->pvfs, path);
-               if (!NT_STATUS_IS_OK(status)) {
-                       DEBUG(0,("Warning: xattr unlink hook failed for '%s' - %s\n",
-                                path, nt_errstr(status)));
-               }
-               if (rmdir(path) != 0) {
-                       DEBUG(0,("pvfs_dir_handle_destructor: failed to rmdir '%s' - %s\n", 
-                                path, strerror(errno)));
-               }
-       }
-
-       talloc_free(path);
-
        if (h->have_opendb_entry) {
                struct odb_lock *lck;
                NTSTATUS status;
+               const char *delete_path = NULL;
 
                lck = odb_lock(h, h->pvfs->odb_context, &h->odb_locking_key);
                if (lck == NULL) {
@@ -80,12 +61,24 @@ static int pvfs_dir_handle_destructor(struct pvfs_file_handle *h)
                        return 0;
                }
 
-               status = odb_close_file(lck, h);
+               status = odb_close_file(lck, h, &delete_path);
                if (!NT_STATUS_IS_OK(status)) {
-                       DEBUG(0,("Unable to remove opendb entry for '%s' - %s\n", 
+                       DEBUG(0,("Unable to remove opendb entry for '%s' - %s\n",
                                 h->name->full_name, nt_errstr(status)));
                }
 
+               if (h->name->stream_name == NULL && delete_path) {
+                       status = pvfs_xattr_unlink_hook(h->pvfs, delete_path);
+                       if (!NT_STATUS_IS_OK(status)) {
+                               DEBUG(0,("Warning: xattr unlink hook failed for '%s' - %s\n",
+                                        delete_path, nt_errstr(status)));
+                       }
+                       if (rmdir(delete_path) != 0) {
+                               DEBUG(0,("pvfs_dir_handle_destructor: failed to rmdir '%s' - %s\n",
+                                        delete_path, strerror(errno)));
+                       }
+               }
+
                talloc_free(lck);
        }
 
@@ -151,8 +144,8 @@ static NTSTATUS pvfs_open_setup_eas_acl(struct pvfs_state *pvfs,
   form the lock context used for opendb locking. Note that we must
   zero here to take account of possible padding on some architectures
 */
-static NTSTATUS pvfs_locking_key(struct pvfs_filename *name, 
-                                TALLOC_CTX *mem_ctx, DATA_BLOB *key)
+NTSTATUS pvfs_locking_key(struct pvfs_filename *name,
+                         TALLOC_CTX *mem_ctx, DATA_BLOB *key)
 {
        struct {
                dev_t device;
@@ -297,10 +290,17 @@ static NTSTATUS pvfs_open_directory(struct pvfs_state *pvfs,
                }
                
                /* see if we are allowed to open at the same time as existing opens */
-               status = odb_open_file(lck, f->handle, name->full_name, name->stream_id,
-                                      share_access, access_mask, del_on_close, 
-                                      io->generic.in.open_disposition,
-                                      false, OPLOCK_NONE, NULL);
+               status = odb_can_open(lck, name->stream_id,
+                                     share_access, access_mask, del_on_close,
+                                     io->generic.in.open_disposition, false);
+               if (!NT_STATUS_IS_OK(status)) {
+                       talloc_free(lck);
+                       return status;
+               }
+
+               /* now really mark the file as open */
+               status = odb_open_file(lck, f->handle, name->full_name,
+                                      NULL, false, OPLOCK_NONE, NULL);
 
                if (!NT_STATUS_IS_OK(status)) {
                        talloc_free(lck);
@@ -351,10 +351,16 @@ static NTSTATUS pvfs_open_directory(struct pvfs_state *pvfs,
                        return NT_STATUS_INTERNAL_DB_CORRUPTION;
                }
 
-               status = odb_open_file(lck, f->handle, name->full_name, name->stream_id,
-                                      share_access, access_mask, del_on_close, 
-                                      io->generic.in.open_disposition,
-                                      false, OPLOCK_NONE, NULL);
+               status = odb_can_open(lck, name->stream_id,
+                                     share_access, access_mask, del_on_close,
+                                     io->generic.in.open_disposition, false);
+
+               if (!NT_STATUS_IS_OK(status)) {
+                       goto cleanup_delete;
+               }
+
+               status = odb_open_file(lck, f->handle, name->full_name,
+                                      NULL, false, OPLOCK_NONE, NULL);
 
                if (!NT_STATUS_IS_OK(status)) {
                        goto cleanup_delete;
@@ -410,9 +416,6 @@ cleanup_delete:
 */
 static int pvfs_handle_destructor(struct pvfs_file_handle *h)
 {
-       int open_count;
-       char *path = NULL;
-
        /* the write time is no longer sticky */
        if (h->sticky_write_time) {
                NTSTATUS status;
@@ -441,32 +444,10 @@ static int pvfs_handle_destructor(struct pvfs_file_handle *h)
                h->fd = -1;
        }
 
-       if (h->name->stream_name == NULL && 
-           h->open_completed &&
-           pvfs_delete_on_close_set(h->pvfs, h, &open_count, &path) &&
-           open_count == 1) {
-               NTSTATUS status;
-               status = pvfs_xattr_unlink_hook(h->pvfs, path);
-               if (!NT_STATUS_IS_OK(status)) {
-                       DEBUG(0,("Warning: xattr unlink hook failed for '%s' - %s\n",
-                                path, nt_errstr(status)));
-               }
-               if (unlink(path) != 0) {
-                       DEBUG(0,("pvfs_close: failed to delete '%s' - %s\n", 
-                                path, strerror(errno)));
-               } else {
-                       notify_trigger(h->pvfs->notify_context, 
-                                      NOTIFY_ACTION_REMOVED, 
-                                      FILE_NOTIFY_CHANGE_FILE_NAME,
-                                      path);
-               }
-       }
-
-       talloc_free(path);
-
        if (h->have_opendb_entry) {
                struct odb_lock *lck;
                NTSTATUS status;
+               const char *delete_path = NULL;
 
                lck = odb_lock(h, h->pvfs->odb_context, &h->odb_locking_key);
                if (lck == NULL) {
@@ -474,12 +455,30 @@ static int pvfs_handle_destructor(struct pvfs_file_handle *h)
                        return 0;
                }
 
-               status = odb_close_file(lck, h);
+               status = odb_close_file(lck, h, &delete_path);
                if (!NT_STATUS_IS_OK(status)) {
                        DEBUG(0,("Unable to remove opendb entry for '%s' - %s\n", 
                                 h->name->full_name, nt_errstr(status)));
                }
 
+               if (h->name->stream_name == NULL &&
+                   h->open_completed && delete_path) {
+                       status = pvfs_xattr_unlink_hook(h->pvfs, delete_path);
+                       if (!NT_STATUS_IS_OK(status)) {
+                               DEBUG(0,("Warning: xattr unlink hook failed for '%s' - %s\n",
+                                        delete_path, nt_errstr(status)));
+                       }
+                       if (unlink(delete_path) != 0) {
+                               DEBUG(0,("pvfs_close: failed to delete '%s' - %s\n",
+                                        delete_path, strerror(errno)));
+                       } else {
+                               notify_trigger(h->pvfs->notify_context,
+                                              NOTIFY_ACTION_REMOVED,
+                                              FILE_NOTIFY_CHANGE_FILE_NAME,
+                                              delete_path);
+                       }
+               }
+
                talloc_free(lck);
        }
 
@@ -558,6 +557,7 @@ static NTSTATUS pvfs_create_file(struct pvfs_state *pvfs,
        bool del_on_close;
        struct pvfs_filename *parent;
        uint32_t oplock_level = OPLOCK_NONE, oplock_granted;
+       bool allow_level_II_oplock = false;
 
        if ((io->ntcreatex.in.file_attr & FILE_ATTRIBUTE_READONLY) &&
            (create_options & NTCREATEX_OPTIONS_DELETE_ON_CLOSE)) {
@@ -574,7 +574,7 @@ static NTSTATUS pvfs_create_file(struct pvfs_state *pvfs,
                status = pvfs_locking_key(parent, req, &locking_key);
                NT_STATUS_NOT_OK_RETURN(status);
                status = odb_get_delete_on_close(pvfs->odb_context, &locking_key, 
-                                                &del_on_close, NULL, NULL);
+                                                &del_on_close);
                NT_STATUS_NOT_OK_RETURN(status);
                if (del_on_close) {
                        return NT_STATUS_DELETE_PENDING;
@@ -600,7 +600,7 @@ static NTSTATUS pvfs_create_file(struct pvfs_state *pvfs,
        mode = pvfs_fileperms(pvfs, attrib);
 
        /* create the file */
-       fd = open(name->full_name, flags | O_CREAT | O_EXCL, mode);
+       fd = open(name->full_name, flags | O_CREAT | O_EXCL| O_NONBLOCK, mode);
        if (fd == -1) {
                return pvfs_map_errno(pvfs, errno);
        }
@@ -672,20 +672,22 @@ static NTSTATUS pvfs_create_file(struct pvfs_state *pvfs,
                oplock_level = OPLOCK_EXCLUSIVE;
        }
 
-       status = odb_open_file(lck, f->handle, name->full_name, name->stream_id,
-                              share_access, access_mask, del_on_close, 
-                              io->generic.in.open_disposition,
-                              false, oplock_level, &oplock_granted);
-       talloc_free(lck);
+       if (req->client_caps & NTVFS_CLIENT_CAP_LEVEL_II_OPLOCKS) {
+               allow_level_II_oplock = true;
+       }
+
+       status = odb_can_open(lck, name->stream_id,
+                             share_access, access_mask, del_on_close,
+                             io->generic.in.open_disposition, false);
        if (!NT_STATUS_IS_OK(status)) {
+               talloc_free(lck);
                /* bad news, we must have hit a race - we don't delete the file
-                  here as the most likely scenario is that someone else created 
+                  here as the most likely scenario is that someone else created
                   the file at the same time */
                close(fd);
                return status;
        }
 
-
        f->ntvfs             = h;
        f->pvfs              = pvfs;
        f->pending_list      = NULL;
@@ -708,23 +710,34 @@ static NTSTATUS pvfs_create_file(struct pvfs_state *pvfs,
        f->handle->sticky_write_time = false;
        f->handle->open_completed    = false;
 
+       status = odb_open_file(lck, f->handle, name->full_name,
+                              &f->handle->fd, allow_level_II_oplock,
+                              oplock_level, &oplock_granted);
+       talloc_free(lck);
+       if (!NT_STATUS_IS_OK(status)) {
+               /* bad news, we must have hit a race - we don't delete the file
+                  here as the most likely scenario is that someone else created
+                  the file at the same time */
+               close(fd);
+               return status;
+       }
+
        DLIST_ADD(pvfs->files.list, f);
 
+       /* setup a destructor to avoid file descriptor leaks on
+          abnormal termination */
+       talloc_set_destructor(f, pvfs_fnum_destructor);
+       talloc_set_destructor(f->handle, pvfs_handle_destructor);
+
        if (pvfs->flags & PVFS_FLAG_FAKE_OPLOCKS) {
                oplock_granted = OPLOCK_BATCH;
        } else if (oplock_granted != OPLOCK_NONE) {
                status = pvfs_setup_oplock(f, oplock_granted);
                if (!NT_STATUS_IS_OK(status)) {
-                       talloc_free(lck);
                        return status;
                }
        }
 
-       /* setup a destructor to avoid file descriptor leaks on
-          abnormal termination */
-       talloc_set_destructor(f, pvfs_fnum_destructor);
-       talloc_set_destructor(f->handle, pvfs_handle_destructor);
-
        io->generic.out.oplock_level  = oplock_granted;
        io->generic.out.file.ntvfs    = f->ntvfs;
        io->generic.out.create_action = NTCREATEX_ACTION_CREATED;
@@ -810,7 +823,7 @@ static void pvfs_odb_retry_callback(void *_r, enum pvfs_wait_notice reason)
 
 /*
   setup for a retry of a request that was rejected
-  by odb_open_file() or odb_can_open()
+  by odb_can_open()
 */
 NTSTATUS pvfs_odb_retry_setup(struct ntvfs_module_context *ntvfs,
                              struct ntvfs_request *req,
@@ -845,7 +858,13 @@ NTSTATUS pvfs_odb_retry_setup(struct ntvfs_module_context *ntvfs,
 
        /* setup a pending lock */
        status = odb_open_file_pending(lck, r);
-       if (!NT_STATUS_IS_OK(status)) {
+       if (NT_STATUS_EQUAL(NT_STATUS_OBJECT_NAME_NOT_FOUND,status)) {
+               /*
+                * maybe only a unix application
+                * has the file open
+                */
+               data_blob_free(&r->odb_locking_key);
+       } else if (!NT_STATUS_IS_OK(status)) {
                return status;
        }
 
@@ -862,8 +881,6 @@ NTSTATUS pvfs_odb_retry_setup(struct ntvfs_module_context *ntvfs,
 
        talloc_steal(r, wait_handle);
 
-       talloc_steal(pvfs, r);
-
        return NT_STATUS_OK;
 }
 
@@ -878,8 +895,14 @@ static void pvfs_retry_open_sharing(struct pvfs_odb_retry *r,
                                    enum pvfs_wait_notice reason)
 {
        union smb_open *io = talloc_get_type(_io, union smb_open);
+       struct timeval *final_timeout = NULL;
        NTSTATUS status;
 
+       if (private_data) {
+               final_timeout = talloc_get_type(private_data,
+                                               struct timeval);
+       }
+
        /* w2k3 ignores SMBntcancel for outstanding open requests. It's probably
           just a bug in their server, but we better do the same */
        if (reason == PVFS_WAIT_CANCEL) {
@@ -887,6 +910,16 @@ static void pvfs_retry_open_sharing(struct pvfs_odb_retry *r,
        }
 
        if (reason == PVFS_WAIT_TIMEOUT) {
+               if (final_timeout &&
+                   !timeval_expired(final_timeout)) {
+                       /*
+                        * we need to retry periodictly
+                        * after an EAGAIN as there's
+                        * no way the kernel tell us
+                        * an oplock is released.
+                        */
+                       goto retry;
+               }
                /* if it timed out, then give the failure
                   immediately */
                talloc_free(r);
@@ -895,6 +928,7 @@ static void pvfs_retry_open_sharing(struct pvfs_odb_retry *r,
                return;
        }
 
+retry:
        talloc_free(r);
 
        /* try the open again, which could trigger another retry setup
@@ -1014,6 +1048,7 @@ static NTSTATUS pvfs_open_setup_retry(struct ntvfs_module_context *ntvfs,
        struct pvfs_state *pvfs = ntvfs->private_data;
        NTSTATUS status;
        struct timeval end_time;
+       struct timeval *final_timeout = NULL;
 
        if (io->generic.in.create_options & 
            (NTCREATEX_OPTIONS_PRIVATE_DENY_DOS | NTCREATEX_OPTIONS_PRIVATE_DENY_FCB)) {
@@ -1034,12 +1069,28 @@ static NTSTATUS pvfs_open_setup_retry(struct ntvfs_module_context *ntvfs,
        } else if (NT_STATUS_EQUAL(parent_status, NT_STATUS_OPLOCK_NOT_GRANTED)) {
                end_time = timeval_add(&req->statistics.request_time,
                                       pvfs->oplock_break_timeout, 0);
+       } else if (NT_STATUS_EQUAL(parent_status, STATUS_MORE_ENTRIES)) {
+               /*
+                * we got EAGAIN which means a unix application
+                * has an oplock or share mode
+                *
+                * we retry every 4/5 of the sharing violation delay
+                * to see if the unix application
+                * has released the oplock or share mode.
+                */
+               final_timeout = talloc(req, struct timeval);
+               NT_STATUS_HAVE_NO_MEMORY(final_timeout);
+               *final_timeout = timeval_add(&req->statistics.request_time,
+                                            pvfs->oplock_break_timeout,
+                                            0);
+               end_time = timeval_current_ofs(0, (pvfs->sharing_violation_delay*4)/5);
+               end_time = timeval_min(final_timeout, &end_time);
        } else {
                return NT_STATUS_INTERNAL_ERROR;
        }
 
-       return pvfs_odb_retry_setup(ntvfs, req, lck, end_time, io, NULL,
-                                   pvfs_retry_open_sharing);
+       return pvfs_odb_retry_setup(ntvfs, req, lck, end_time, io,
+                                   final_timeout, pvfs_retry_open_sharing);
 }
 
 /*
@@ -1062,6 +1113,7 @@ NTSTATUS pvfs_open(struct ntvfs_module_context *ntvfs,
        bool del_on_close;
        bool stream_existed, stream_truncate=false;
        uint32_t oplock_level = OPLOCK_NONE, oplock_granted;
+       bool allow_level_II_oplock = false;
 
        /* use the generic mapping code to avoid implementing all the
           different open calls. */
@@ -1256,11 +1308,14 @@ NTSTATUS pvfs_open(struct ntvfs_module_context *ntvfs,
                oplock_level = OPLOCK_EXCLUSIVE;
        }
 
+       if (req->client_caps & NTVFS_CLIENT_CAP_LEVEL_II_OPLOCKS) {
+               allow_level_II_oplock = true;
+       }
+
        /* see if we are allowed to open at the same time as existing opens */
-       status = odb_open_file(lck, f->handle, name->full_name, name->stream_id,
-                              share_access, access_mask, del_on_close,
-                              io->generic.in.open_disposition,
-                              false, oplock_level, &oplock_granted);
+       status = odb_can_open(lck, name->stream_id,
+                             share_access, access_mask, del_on_close,
+                             io->generic.in.open_disposition, false);
 
        /*
         * on a sharing violation we need to retry when the file is closed by
@@ -1279,18 +1334,6 @@ NTSTATUS pvfs_open(struct ntvfs_module_context *ntvfs,
                return status;
        }
 
-       if (pvfs->flags & PVFS_FLAG_FAKE_OPLOCKS) {
-               oplock_granted = OPLOCK_BATCH;
-       } else if (oplock_granted != OPLOCK_NONE) {
-               status = pvfs_setup_oplock(f, oplock_granted);
-               if (!NT_STATUS_IS_OK(status)) {
-                       talloc_free(lck);
-                       return status;
-               }
-       }
-
-       f->handle->have_opendb_entry = true;
-
        if (access_mask & (SEC_FILE_WRITE_DATA | SEC_FILE_APPEND_DATA)) {
                flags |= O_RDWR;
        } else {
@@ -1298,14 +1341,46 @@ NTSTATUS pvfs_open(struct ntvfs_module_context *ntvfs,
        }
 
        /* do the actual open */
-       fd = open(f->handle->name->full_name, flags);
+       fd = open(f->handle->name->full_name, flags | O_NONBLOCK);
        if (fd == -1) {
+               status = pvfs_map_errno(f->pvfs, errno);
+
+               /*
+                * STATUS_MORE_ENTRIES is EAGAIN or EWOULDBLOCK
+                */
+               if (NT_STATUS_EQUAL(status, STATUS_MORE_ENTRIES) &&
+                   (req->async_states->state & NTVFS_ASYNC_STATE_MAY_ASYNC)) {
+                       return pvfs_open_setup_retry(ntvfs, req, io, f, lck, status);
+               }
+
                talloc_free(lck);
-               return pvfs_map_errno(f->pvfs, errno);
+               return status;
        }
 
        f->handle->fd = fd;
 
+       /* now really mark the file as open */
+       status = odb_open_file(lck, f->handle, name->full_name,
+                              &f->handle->fd, allow_level_II_oplock,
+                              oplock_level, &oplock_granted);
+
+       if (!NT_STATUS_IS_OK(status)) {
+               talloc_free(lck);
+               return status;
+       }
+
+       if (pvfs->flags & PVFS_FLAG_FAKE_OPLOCKS) {
+               oplock_granted = OPLOCK_BATCH;
+       } else if (oplock_granted != OPLOCK_NONE) {
+               status = pvfs_setup_oplock(f, oplock_granted);
+               if (!NT_STATUS_IS_OK(status)) {
+                       talloc_free(lck);
+                       return status;
+               }
+       }
+
+       f->handle->have_opendb_entry = true;
+
        stream_existed = name->stream_exists;
 
        /* if this was a stream create then create the stream as well */
@@ -1738,14 +1813,13 @@ NTSTATUS pvfs_can_stat(struct pvfs_state *pvfs,
 /*
   determine if delete on close is set on 
 */
-bool pvfs_delete_on_close_set(struct pvfs_state *pvfs, struct pvfs_file_handle *h, 
-                             int *open_count, char **path)
+bool pvfs_delete_on_close_set(struct pvfs_state *pvfs, struct pvfs_file_handle *h)
 {
        NTSTATUS status;
        bool del_on_close;
 
        status = odb_get_delete_on_close(pvfs->odb_context, &h->odb_locking_key, 
-                                        &del_on_close, open_count, path);
+                                        &del_on_close);
        if (!NT_STATUS_IS_OK(status)) {
                DEBUG(1,("WARNING: unable to determine delete on close status for open file\n"));
                return false;