From 3da00ac7c9da8c6c413e72674041dc6847eb127c Mon Sep 17 00:00:00 2001 From: Jeremy Allison Date: Thu, 1 May 2014 10:58:51 -0700 Subject: [PATCH 1/4] s3 : smbd : Protect all possible code paths from fsp->op == NULL. In changes to come this will be possible for an INTERNAL_OPEN_ONLY. The protection was already in place for some code paths, this makes the coverage compete. Bug 10564 - Lock order violation and file lost https://bugzilla.samba.org/show_bug.cgi?id=10564 Signed-off-by: Jeremy Allison --- source3/locking/brlock.c | 16 ++++++++++++++-- source3/modules/vfs_btrfs.c | 5 +++++ source3/modules/vfs_default.c | 9 +++++++++ source3/smbd/aio.c | 10 ++++++++++ source3/smbd/scavenger.c | 3 +++ 5 files changed, 41 insertions(+), 2 deletions(-) diff --git a/source3/locking/brlock.c b/source3/locking/brlock.c index 78c2538..df689a7 100644 --- a/source3/locking/brlock.c +++ b/source3/locking/brlock.c @@ -1526,12 +1526,18 @@ void brl_close_fnum(struct messaging_context *msg_ctx, bool brl_mark_disconnected(struct files_struct *fsp) { uint32_t tid = fsp->conn->cnum; - uint64_t smblctx = fsp->op->global->open_persistent_id; + uint64_t smblctx; uint64_t fnum = fsp->fnum; unsigned int i; struct server_id self = messaging_server_id(fsp->conn->sconn->msg_ctx); struct byte_range_lock *br_lck = NULL; + if (fsp->op == NULL) { + return false; + } + + smblctx = fsp->op->global->open_persistent_id; + if (!fsp->op->global->durable) { return false; } @@ -1586,12 +1592,18 @@ bool brl_mark_disconnected(struct files_struct *fsp) bool brl_reconnect_disconnected(struct files_struct *fsp) { uint32_t tid = fsp->conn->cnum; - uint64_t smblctx = fsp->op->global->open_persistent_id; + uint64_t smblctx; uint64_t fnum = fsp->fnum; unsigned int i; struct server_id self = messaging_server_id(fsp->conn->sconn->msg_ctx); struct byte_range_lock *br_lck = NULL; + if (fsp->op == NULL) { + return false; + } + + smblctx = fsp->op->global->open_persistent_id; + if (!fsp->op->global->durable) { return false; } diff --git a/source3/modules/vfs_btrfs.c b/source3/modules/vfs_btrfs.c index 4ecf7ab..f062eba 100644 --- a/source3/modules/vfs_btrfs.c +++ b/source3/modules/vfs_btrfs.c @@ -98,6 +98,11 @@ static struct tevent_req *btrfs_copy_chunk_send(struct vfs_handle_struct *handle return tevent_req_post(req, ev); } + if (src_fsp->op == NULL || dest_fsp->op == NULL) { + tevent_req_nterror(req, NT_STATUS_INTERNAL_ERROR); + return tevent_req_post(req, ev); + } + init_strict_lock_struct(src_fsp, src_fsp->op->global->open_persistent_id, src_off, diff --git a/source3/modules/vfs_default.c b/source3/modules/vfs_default.c index 82d059c..f6200ed 100644 --- a/source3/modules/vfs_default.c +++ b/source3/modules/vfs_default.c @@ -1393,6 +1393,10 @@ static struct tevent_req *vfswrap_copy_chunk_send(struct vfs_handle_struct *hand off_t this_num = MIN(sizeof(vfs_cc_state->buf), num - vfs_cc_state->copied); + if (src_fsp->op == NULL) { + tevent_req_nterror(req, NT_STATUS_INTERNAL_ERROR); + return tevent_req_post(req, ev); + } init_strict_lock_struct(src_fsp, src_fsp->op->global->open_persistent_id, src_off, @@ -1426,6 +1430,11 @@ static struct tevent_req *vfswrap_copy_chunk_send(struct vfs_handle_struct *hand src_off += ret; + if (dest_fsp->op == NULL) { + tevent_req_nterror(req, NT_STATUS_INTERNAL_ERROR); + return tevent_req_post(req, ev); + } + init_strict_lock_struct(dest_fsp, dest_fsp->op->global->open_persistent_id, dest_off, diff --git a/source3/smbd/aio.c b/source3/smbd/aio.c index eec29f6..44d771e 100644 --- a/source3/smbd/aio.c +++ b/source3/smbd/aio.c @@ -688,6 +688,11 @@ NTSTATUS schedule_smb2_aio_read(connection_struct *conn, return NT_STATUS_RETRY; } + if (fsp->op == NULL) { + /* No AIO on internal opens. */ + return NT_STATUS_RETRY; + } + if ((!min_aio_read_size || (smb_maxcnt < min_aio_read_size)) && !SMB_VFS_AIO_FORCE(fsp)) { /* Too small a read for aio request. */ @@ -839,6 +844,11 @@ NTSTATUS schedule_aio_smb2_write(connection_struct *conn, return NT_STATUS_RETRY; } + if (fsp->op == NULL) { + /* No AIO on internal opens. */ + return NT_STATUS_RETRY; + } + if ((!min_aio_write_size || (in_data.length < min_aio_write_size)) && !SMB_VFS_AIO_FORCE(fsp)) { /* Too small a write for aio request. */ diff --git a/source3/smbd/scavenger.c b/source3/smbd/scavenger.c index e6e2878..122305e 100644 --- a/source3/smbd/scavenger.c +++ b/source3/smbd/scavenger.c @@ -418,6 +418,9 @@ void scavenger_schedule_disconnected(struct files_struct *fsp) struct scavenger_message msg; DATA_BLOB msg_blob; + if (fsp->op == NULL) { + return; + } nttime_to_timeval(&disconnect_time, fsp->op->global->disconnect_time); timeout_usec = 1000 * fsp->op->global->durable_timeout_msec; until = timeval_add(&disconnect_time, -- 1.9.1.423.g4596e3a From 8c960dfc85e61e8a3dd3e9b761301ca48af3c661 Mon Sep 17 00:00:00 2001 From: Jeremy Allison Date: Thu, 1 May 2014 11:01:03 -0700 Subject: [PATCH 2/4] s3: smbd : Ensure file_new doesn't call into smbXsrv_open_create() for INTERNAL_OPEN_ONLY. This causes deadlocks which cause smbd to crash if the locking database has already been locked for a compound operation we need to be atomic (as in the file rename case). Ensure INTERNAL_OPEN_ONLY opens are synonymous with req==NULL. INTERNAL_OPEN_ONLY opens leave a NO_OPLOCK record in the share mode database, so can be detected by other processes for share mode violation purposes (that they're doing an operation on the file they need to have real state inside the locking database) but have an fnum of FNUM_FIELD_INVALID and a local share_file_id of zero, as they will never be seen on the wire. Ensure validate_my_share_entries() ignores INTERNAL_OPEN_ONLY records (share_file_id == 0). Bug 10564 - Lock order violation and file lost https://bugzilla.samba.org/show_bug.cgi?id=10564 Signed-off-by: Jeremy Allison --- source3/smbd/files.c | 4 +++- source3/smbd/open.c | 14 +++++++++++--- 2 files changed, 14 insertions(+), 4 deletions(-) diff --git a/source3/smbd/files.c b/source3/smbd/files.c index d94ee11..34ec879 100644 --- a/source3/smbd/files.c +++ b/source3/smbd/files.c @@ -93,7 +93,7 @@ NTSTATUS file_new(struct smb_request *req, connection_struct *conn, GetTimeOfDay(&fsp->open_time); - if (sconn->conn) { + if (req) { struct smbXsrv_open *op = NULL; NTTIME now = timeval_to_nttime(&fsp->open_time); @@ -108,6 +108,8 @@ NTSTATUS file_new(struct smb_request *req, connection_struct *conn, op->compat = fsp; fsp->fnum = op->local_id; fsp->fh->gen_id = smbXsrv_open_hash(op); + } else { + /* INTERNAL_OPEN_ONLY */ } /* diff --git a/source3/smbd/open.c b/source3/smbd/open.c index 6e4f690..15fec95 100644 --- a/source3/smbd/open.c +++ b/source3/smbd/open.c @@ -1079,6 +1079,11 @@ static void validate_my_share_entries(struct smbd_server_connection *sconn, return; } + if (share_entry->share_file_id == 0) { + /* INTERNAL_OPEN_ONLY */ + return; + } + if (!is_valid_share_mode_entry(share_entry)) { return; } @@ -2088,9 +2093,12 @@ static NTSTATUS open_file_ntcreate(connection_struct *conn, create_options, (unsigned int)unx_mode, oplock_request, (unsigned int)private_flags)); - if ((req == NULL) && ((oplock_request & INTERNAL_OPEN_ONLY) == 0)) { - DEBUG(0, ("No smb request but not an internal only open!\n")); - return NT_STATUS_INTERNAL_ERROR; + if (req == NULL) { + /* Ensure req == NULL means INTERNAL_OPEN_ONLY */ + SMB_ASSERT(((oplock_request & INTERNAL_OPEN_ONLY) != 0)); + } else { + /* And req != NULL means no INTERNAL_OPEN_ONLY */ + SMB_ASSERT(((oplock_request & INTERNAL_OPEN_ONLY) == 0)); } /* -- 1.9.1.423.g4596e3a From 81cda374cb87d37bcac77a99fd490d85926f0e9d Mon Sep 17 00:00:00 2001 From: Jeremy Allison Date: Thu, 1 May 2014 11:07:44 -0700 Subject: [PATCH 3/4] s3: smbd: change file_set_dosmode() to use get_file_handle_for_metadata() instead of open_file_fchmod(). get_file_handle_for_metadata() is a new function that finds an existing open handle (fsp->fh->fd != -1) for a given dev/ino if there is one available, and uses INTERNAL_OPEN_ONLY with WRITE_DATA access if not. Allows open_file_fchmod() to be removed next. Bug 10564 - Lock order violation and file lost https://bugzilla.samba.org/show_bug.cgi?id=10564 Signed-off-by: Jeremy Allison --- source3/smbd/dosmode.c | 103 ++++++++++++++++++++++++++++++++++++++++++------- 1 file changed, 90 insertions(+), 13 deletions(-) diff --git a/source3/smbd/dosmode.c b/source3/smbd/dosmode.c index 2d07dd9..efadcb6 100644 --- a/source3/smbd/dosmode.c +++ b/source3/smbd/dosmode.c @@ -25,6 +25,11 @@ #include "smbd/smbd.h" #include "lib/param/loadparm.h" +static NTSTATUS get_file_handle_for_metadata(connection_struct *conn, + struct smb_filename *smb_fname, + files_struct **ret_fsp, + bool *need_close); + static uint32_t filter_mode_by_protocol(uint32_t mode) { if (get_Protocol() <= PROTOCOL_LANMAN2) { @@ -387,6 +392,7 @@ static bool set_ea_dos_attribute(connection_struct *conn, SAMBA_XATTR_DOS_ATTRIB, blob.data, blob.length, 0) == -1) { bool ret = false; + bool need_close = false; files_struct *fsp = NULL; if((errno != EPERM) && (errno != EACCES)) { @@ -418,14 +424,17 @@ static bool set_ea_dos_attribute(connection_struct *conn, } /* - * We need to open the file with write access whilst - * still in our current user context. This ensures we - * are not violating security in doing the setxattr. + * We need to get an open file handle to do the + * metadata operation under root. */ - if (!NT_STATUS_IS_OK(open_file_fchmod(conn, smb_fname, - &fsp))) + if (!NT_STATUS_IS_OK(get_file_handle_for_metadata(conn, + smb_fname, + &fsp, + &need_close))) { return false; + } + become_root(); if (SMB_VFS_FSETXATTR(fsp, SAMBA_XATTR_DOS_ATTRIB, blob.data, @@ -433,7 +442,9 @@ static bool set_ea_dos_attribute(connection_struct *conn, ret = true; } unbecome_root(); - close_file(NULL, fsp, NORMAL_CLOSE); + if (need_close) { + close_file(NULL, fsp, NORMAL_CLOSE); + } return ret; } DEBUG(10,("set_ea_dos_attribute: set EA 0x%x on file %s\n", @@ -711,6 +722,7 @@ int file_set_dosmode(connection_struct *conn, struct smb_filename *smb_fname, uint32_t old_mode; struct timespec new_create_timespec; files_struct *fsp = NULL; + bool need_close = false; if (!CAN_WRITE(conn)) { errno = EROFS; @@ -879,17 +891,23 @@ int file_set_dosmode(connection_struct *conn, struct smb_filename *smb_fname, } /* - * We need to open the file with write access whilst - * still in our current user context. This ensures we - * are not violating security in doing the fchmod. + * We need to get an open file handle to do the + * metadata operation under root. */ - if (!NT_STATUS_IS_OK(open_file_fchmod(conn, smb_fname, - &fsp))) - return -1; + + if (!NT_STATUS_IS_OK(get_file_handle_for_metadata(conn, + smb_fname, + &fsp, + &need_close))) { + return false; + } + become_root(); ret = SMB_VFS_FCHMOD(fsp, unixmode); unbecome_root(); - close_file(NULL, fsp, NORMAL_CLOSE); + if (need_close) { + close_file(NULL, fsp, NORMAL_CLOSE); + } if (!newfile) { notify_fname(conn, NOTIFY_ACTION_MODIFIED, FILE_NOTIFY_CHANGE_ATTRIBUTES, @@ -1125,3 +1143,62 @@ struct timespec get_change_timespec(connection_struct *conn, { return smb_fname->st.st_ex_mtime; } + +/**************************************************************************** + Get a real open file handle we can do meta-data operations on. As it's + going to be used under root access only on meta-data we should look for + any existing open file handle first, and use that in preference (also to + avoid kernel self-oplock breaks). If not use an INTERNAL_OPEN_ONLY handle. +****************************************************************************/ + +static NTSTATUS get_file_handle_for_metadata(connection_struct *conn, + struct smb_filename *smb_fname, + files_struct **ret_fsp, + bool *need_close) +{ + NTSTATUS status; + files_struct *fsp; + struct file_id file_id; + + *need_close = false; + + if (!VALID_STAT(smb_fname->st)) { + return NT_STATUS_INVALID_PARAMETER; + } + + file_id = vfs_file_id_from_sbuf(conn, &smb_fname->st); + + for(fsp = file_find_di_first(conn->sconn, file_id); + fsp; + fsp = file_find_di_next(fsp)) { + if (fsp->fh->fd != -1) { + *ret_fsp = fsp; + return NT_STATUS_OK; + } + } + + /* Opens an INTERNAL_OPEN_ONLY write handle. */ + status = SMB_VFS_CREATE_FILE( + conn, /* conn */ + NULL, /* req */ + 0, /* root_dir_fid */ + smb_fname, /* fname */ + FILE_WRITE_DATA, /* access_mask */ + (FILE_SHARE_READ | FILE_SHARE_WRITE | /* share_access */ + FILE_SHARE_DELETE), + FILE_OPEN, /* create_disposition*/ + 0, /* create_options */ + 0, /* file_attributes */ + INTERNAL_OPEN_ONLY, /* oplock_request */ + 0, /* allocation_size */ + 0, /* private_flags */ + NULL, /* sd */ + NULL, /* ea_list */ + ret_fsp, /* result */ + NULL); /* pinfo */ + + if (NT_STATUS_IS_OK(status)) { + *need_close = true; + } + return status; +} -- 1.9.1.423.g4596e3a From 192c4b03048e31070b318d181ba6b246998f11e3 Mon Sep 17 00:00:00 2001 From: Jeremy Allison Date: Thu, 1 May 2014 11:11:20 -0700 Subject: [PATCH 4/4] s3: smbd: Remove open_file_fchmod(). No longer used (hurrah!). Bug 10564 - Lock order violation and file lost https://bugzilla.samba.org/show_bug.cgi?id=10564 Signed-off-by: Jeremy Allison --- source3/smbd/open.c | 33 --------------------------------- source3/smbd/proto.h | 3 --- 2 files changed, 36 deletions(-) diff --git a/source3/smbd/open.c b/source3/smbd/open.c index 15fec95..5f7bff9 100644 --- a/source3/smbd/open.c +++ b/source3/smbd/open.c @@ -2838,39 +2838,6 @@ static NTSTATUS open_file_ntcreate(connection_struct *conn, return NT_STATUS_OK; } - -/**************************************************************************** - Open a file for for write to ensure that we can fchmod it. -****************************************************************************/ - -NTSTATUS open_file_fchmod(connection_struct *conn, - struct smb_filename *smb_fname, - files_struct **result) -{ - if (!VALID_STAT(smb_fname->st)) { - return NT_STATUS_INVALID_PARAMETER; - } - - return SMB_VFS_CREATE_FILE( - conn, /* conn */ - NULL, /* req */ - 0, /* root_dir_fid */ - smb_fname, /* fname */ - FILE_WRITE_DATA, /* access_mask */ - (FILE_SHARE_READ | FILE_SHARE_WRITE | /* share_access */ - FILE_SHARE_DELETE), - FILE_OPEN, /* create_disposition*/ - 0, /* create_options */ - 0, /* file_attributes */ - INTERNAL_OPEN_ONLY, /* oplock_request */ - 0, /* allocation_size */ - 0, /* private_flags */ - NULL, /* sd */ - NULL, /* ea_list */ - result, /* result */ - NULL); /* pinfo */ -} - static NTSTATUS mkdir_internal(connection_struct *conn, struct smb_filename *smb_dname, uint32 file_attributes) diff --git a/source3/smbd/proto.h b/source3/smbd/proto.h index 1e0d06d..6153a49 100644 --- a/source3/smbd/proto.h +++ b/source3/smbd/proto.h @@ -620,9 +620,6 @@ NTSTATUS change_dir_owner_to_parent(connection_struct *conn, SMB_STRUCT_STAT *psbuf); bool is_stat_open(uint32 access_mask); bool is_deferred_open_async(const void *ptr); -NTSTATUS open_file_fchmod(connection_struct *conn, - struct smb_filename *smb_fname, - files_struct **result); NTSTATUS create_directory(connection_struct *conn, struct smb_request *req, struct smb_filename *smb_dname); void msg_file_was_renamed(struct messaging_context *msg, -- 1.9.1.423.g4596e3a