cifs: avoid use of global locks for high contention data

During analysis of multichannel perf, it was seen that
the global locks cifs_tcp_ses_lock and GlobalMid_Lock, which
were shared between various data structures were causing a
lot of contention points.

With this change, we're breaking down the use of these locks
by introducing new locks at more granular levels. i.e.
server->srv_lock, ses->ses_lock and tcon->tc_lock to protect
the unprotected fields of server, session and tcon structs;
and server->mid_lock to protect mid related lists and entries
at server level.

Signed-off-by: Shyam Prasad N <sprasad@microsoft.com>
Signed-off-by: Steve French <stfrench@microsoft.com>

1 files changed, 73 insertions, 26 deletions

diff --git a/fs/cifs/cifsglob.h b/fs/cifs/cifsglob.h
index 54ea709fe0dd..3070407cafa7 100644
--- a/fs/cifs/cifsglob.h
+++ b/fs/cifs/cifsglob.h
@@ -605,6 +605,7 @@ inc_rfc1001_len(void *buf, int count)
 struct TCP_Server_Info {
 	struct list_head tcp_ses_list;
 	struct list_head smb_ses_list;
+	spinlock_t srv_lock;  /* protect anything here that is not protected */
 	__u64 conn_id; /* connection identifier (useful for debugging) */
 	int srv_count; /* reference counter */
 	/* 15 character server name + 0x20 16th byte indicating type = srv */
@@ -622,6 +623,7 @@ struct TCP_Server_Info {
 #endif
 	wait_queue_head_t response_q;
 	wait_queue_head_t request_q; /* if more than maxmpx to srvr must block*/
+	spinlock_t mid_lock;  /* protect mid queue and it's entries */
 	struct list_head pending_mid_q;
 	bool noblocksnd;		/* use blocking sendmsg */
 	bool noautotune;		/* do not autotune send buf sizes */
@@ -1008,6 +1010,7 @@ struct cifs_ses {
 	struct list_head rlist; /* reconnect list */
 	struct list_head tcon_list;
 	struct cifs_tcon *tcon_ipc;
+	spinlock_t ses_lock;  /* protect anything here that is not protected */
 	struct mutex session_mutex;
 	struct TCP_Server_Info *server;	/* pointer to server info */
 	int ses_count;		/* reference counter */
@@ -1169,6 +1172,7 @@ struct cifs_tcon {
 	struct list_head tcon_list;
 	int tc_count;
 	struct list_head rlist; /* reconnect list */
+	spinlock_t tc_lock;  /* protect anything here that is not protected */
 	atomic_t num_local_opens;  /* num of all opens including disconnected */
 	atomic_t num_remote_opens; /* num of all network opens on server */
 	struct list_head openFileList;
@@ -1899,33 +1903,78 @@ require use of the stronger protocol */
  */
 
 /****************************************************************************
- *  Locking notes.  All updates to global variables and lists should be
- *                  protected by spinlocks or semaphores.
+ * Here are all the locks (spinlock, mutex, semaphore) in cifs.ko, arranged according
+ * to the locking order. i.e. if two locks are to be held together, the lock that
+ * appears higher in this list needs to be taken before the other.
  *
- *  Spinlocks
- *  ---------
- *  GlobalMid_Lock protects:
- *	list operations on pending_mid_q and oplockQ
- *      updates to XID counters, multiplex id  and SMB sequence numbers
- *      list operations on global DnotifyReqList
- *      updates to ses->status and TCP_Server_Info->tcpStatus
- *      updates to server->CurrentMid
- *  tcp_ses_lock protects:
- *	list operations on tcp and SMB session lists
- *  tcon->open_file_lock protects the list of open files hanging off the tcon
- *  inode->open_file_lock protects the openFileList hanging off the inode
- *  cfile->file_info_lock protects counters and fields in cifs file struct
- *  f_owner.lock protects certain per file struct operations
- *  mapping->page_lock protects certain per page operations
+ * If you hold a lock that is lower in this list, and you need to take a higher lock
+ * (or if you think that one of the functions that you're calling may need to), first
+ * drop the lock you hold, pick up the higher lock, then the lower one. This will
+ * ensure that locks are picked up only in one direction in the below table
+ * (top to bottom).
  *
- *  Note that the cifs_tcon.open_file_lock should be taken before
- *  not after the cifsInodeInfo.open_file_lock
+ * Also, if you expect a function to be called with a lock held, explicitly document
+ * this in the comments on top of your function definition.
  *
- *  Semaphores
- *  ----------
- *  cifsInodeInfo->lock_sem protects:
- *	the list of locks held by the inode
+ * And also, try to keep the critical sections (lock hold time) to be as minimal as
+ * possible. Blocking / calling other functions with a lock held always increase
+ * the risk of a possible deadlock.
  *
+ * Following this rule will avoid unnecessary deadlocks, which can get really hard to
+ * debug. Also, any new lock that you introduce, please add to this list in the correct
+ * order.
+ *
+ * Please populate this list whenever you introduce new locks in your changes. Or in
+ * case I've missed some existing locks. Please ensure that it's added in the list
+ * based on the locking order expected.
+ *
+ * =====================================================================================
+ * Lock				Protects			Initialization fn
+ * =====================================================================================
+ * vol_list_lock
+ * vol_info->ctx_lock		vol_info->ctx
+ * cifs_sb_info->tlink_tree_lock	cifs_sb_info->tlink_tree	cifs_setup_cifs_sb
+ * TCP_Server_Info->		TCP_Server_Info			cifs_get_tcp_session
+ * reconnect_mutex
+ * TCP_Server_Info->srv_mutex	TCP_Server_Info			cifs_get_tcp_session
+ * cifs_ses->session_mutex		cifs_ses		sesInfoAlloc
+ *				cifs_tcon
+ * cifs_tcon->open_file_lock	cifs_tcon->openFileList		tconInfoAlloc
+ *				cifs_tcon->pending_opens
+ * cifs_tcon->stat_lock		cifs_tcon->bytes_read		tconInfoAlloc
+ *				cifs_tcon->bytes_written
+ * cifs_tcp_ses_lock		cifs_tcp_ses_list		sesInfoAlloc
+ * GlobalMid_Lock		GlobalMaxActiveXid		init_cifs
+ *				GlobalCurrentXid
+ *				GlobalTotalActiveXid
+ * TCP_Server_Info->srv_lock	(anything in struct not protected by another lock and can change)
+ * TCP_Server_Info->mid_lock	TCP_Server_Info->pending_mid_q	cifs_get_tcp_session
+ *				->CurrentMid
+ *				(any changes in mid_q_entry fields)
+ * TCP_Server_Info->req_lock	TCP_Server_Info->in_flight	cifs_get_tcp_session
+ *				->credits
+ *				->echo_credits
+ *				->oplock_credits
+ *				->reconnect_instance
+ * cifs_ses->ses_lock		(anything that is not protected by another lock and can change)
+ * cifs_ses->iface_lock		cifs_ses->iface_list		sesInfoAlloc
+ *				->iface_count
+ *				->iface_last_update
+ * cifs_ses->chan_lock		cifs_ses->chans
+ *				->chans_need_reconnect
+ *				->chans_in_reconnect
+ * cifs_tcon->tc_lock		(anything that is not protected by another lock and can change)
+ * cifsInodeInfo->open_file_lock	cifsInodeInfo->openFileList	cifs_alloc_inode
+ * cifsInodeInfo->writers_lock	cifsInodeInfo->writers		cifsInodeInfo_alloc
+ * cifsInodeInfo->lock_sem	cifsInodeInfo->llist		cifs_init_once
+ *				->can_cache_brlcks
+ * cifsInodeInfo->deferred_lock	cifsInodeInfo->deferred_closes	cifsInodeInfo_alloc
+ * cached_fid->fid_mutex		cifs_tcon->crfid		tconInfoAlloc
+ * cifsFileInfo->fh_mutex		cifsFileInfo			cifs_new_fileinfo
+ * cifsFileInfo->file_info_lock	cifsFileInfo->count		cifs_new_fileinfo
+ *				->invalidHandle			initiate_cifs_search
+ *				->oplock_break_cancelled
+ * cifs_aio_ctx->aio_mutex		cifs_aio_ctx			cifs_aio_ctx_alloc
  ****************************************************************************/
 
 #ifdef DECLARE_GLOBALS_HERE
@@ -1946,9 +1995,7 @@ extern struct list_head		cifs_tcp_ses_list;
 /*
  * This lock protects the cifs_tcp_ses_list, the list of smb sessions per
  * tcp session, and the list of tcon's per smb session. It also protects
- * the reference counters for the server, smb session, and tcon. It also
- * protects some fields in the TCP_Server_Info struct such as dstaddr. Finally,
- * changes to the tcon->tidStatus should be done while holding this lock.
+ * the reference counters for the server, smb session, and tcon.
  * generally the locks should be taken in order tcp_ses_lock before
  * tcon->open_file_lock and that before file->file_info_lock since the
  * structure order is cifs_socket-->cifs_ses-->cifs_tcon-->cifs_file