Clarify locking of cifs file and tcon structures and make more granular
authorSteve French <smfrench@gmail.com>
Thu, 22 Sep 2016 23:58:16 +0000 (18:58 -0500)
committerSteve French <smfrench@gmail.com>
Wed, 12 Oct 2016 17:08:32 +0000 (12:08 -0500)
Remove the global file_list_lock to simplify cifs/smb3 locking and
have spinlocks that more closely match the information they are
protecting.

Add new tcon->open_file_lock and file->file_info_lock spinlocks.
Locks continue to follow a heirachy,
cifs_socket --> cifs_ses --> cifs_tcon --> cifs_file
where global tcp_ses_lock still protects socket and cifs_ses, while the
the newer locks protect the lower level structure's information
(tcon and cifs_file respectively).

CC: Stable <stable@vger.kernel.org>
Signed-off-by: Steve French <steve.french@primarydata.com>
Signed-off-by: Pavel Shilovsky <pshilov@microsoft.com>
Reviewed-by: Aurelien Aptel <aaptel@suse.com>
Reviewed-by: Germano Percossi <germano.percossi@citrix.com>
fs/cifs/cifsfs.c
fs/cifs/cifsglob.h
fs/cifs/cifssmb.c
fs/cifs/file.c
fs/cifs/misc.c
fs/cifs/readdir.c
fs/cifs/smb2misc.c

index 978dbf049a778343a38b8ad71b2d39ae589134d4..d2b2703fed7dd57f17ffe1bb307c462dccd437fd 100644 (file)
@@ -1262,7 +1262,6 @@ init_cifs(void)
        GlobalTotalActiveXid = 0;
        GlobalMaxActiveXid = 0;
        spin_lock_init(&cifs_tcp_ses_lock);
-       spin_lock_init(&cifs_file_list_lock);
        spin_lock_init(&GlobalMid_Lock);
 
        get_random_bytes(&cifs_lock_secret, sizeof(cifs_lock_secret));
index 8f1d8c1e72bece412a926d0e20b9bdc95a7fa9ee..65f78b7a90620a02c82d19f020f2c087b3fad8e9 100644 (file)
@@ -833,6 +833,7 @@ struct cifs_tcon {
        struct list_head tcon_list;
        int tc_count;
        struct list_head openFileList;
+       spinlock_t open_file_lock; /* protects list above */
        struct cifs_ses *ses;   /* pointer to session associated with */
        char treeName[MAX_TREE_SIZE + 1]; /* UNC name of resource in ASCII */
        char *nativeFileSystem;
@@ -889,7 +890,7 @@ struct cifs_tcon {
 #endif /* CONFIG_CIFS_STATS2 */
        __u64    bytes_read;
        __u64    bytes_written;
-       spinlock_t stat_lock;
+       spinlock_t stat_lock;  /* protects the two fields above */
 #endif /* CONFIG_CIFS_STATS */
        FILE_SYSTEM_DEVICE_INFO fsDevInfo;
        FILE_SYSTEM_ATTRIBUTE_INFO fsAttrInfo; /* ok if fs name truncated */
@@ -1040,8 +1041,10 @@ struct cifs_fid_locks {
 };
 
 struct cifsFileInfo {
+       /* following two lists are protected by tcon->open_file_lock */
        struct list_head tlist; /* pointer to next fid owned by tcon */
        struct list_head flist; /* next fid (file instance) for this inode */
+       /* lock list below protected by cifsi->lock_sem */
        struct cifs_fid_locks *llist;   /* brlocks held by this fid */
        kuid_t uid;             /* allows finding which FileInfo structure */
        __u32 pid;              /* process id who opened file */
@@ -1049,11 +1052,12 @@ struct cifsFileInfo {
        /* BB add lock scope info here if needed */ ;
        /* lock scope id (0 if none) */
        struct dentry *dentry;
-       unsigned int f_flags;
        struct tcon_link *tlink;
+       unsigned int f_flags;
        bool invalidHandle:1;   /* file closed via session abend */
        bool oplock_break_cancelled:1;
-       int count;              /* refcount protected by cifs_file_list_lock */
+       int count;
+       spinlock_t file_info_lock; /* protects four flag/count fields above */
        struct mutex fh_mutex; /* prevents reopen race after dead ses*/
        struct cifs_search_info srch_inf;
        struct work_struct oplock_break; /* work for oplock breaks */
@@ -1120,7 +1124,7 @@ struct cifs_writedata {
 
 /*
  * Take a reference on the file private data. Must be called with
- * cifs_file_list_lock held.
+ * cfile->file_info_lock held.
  */
 static inline void
 cifsFileInfo_get_locked(struct cifsFileInfo *cifs_file)
@@ -1514,8 +1518,10 @@ require use of the stronger protocol */
  *  GlobalMid_Lock protects:
  *     list operations on pending_mid_q and oplockQ
  *      updates to XID counters, multiplex id  and SMB sequence numbers
- *  cifs_file_list_lock protects:
- *     list operations on tcp and SMB session lists and tCon lists
+ *  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
+ *  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
  *
@@ -1547,18 +1553,12 @@ GLOBAL_EXTERN struct list_head          cifs_tcp_ses_list;
  * tcp session, and the list of tcon's per smb session. It also protects
  * the reference counters for the server, smb session, and tcon. Finally,
  * changes to the tcon->tidStatus should be done while holding this lock.
+ * 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
  */
 GLOBAL_EXTERN spinlock_t               cifs_tcp_ses_lock;
 
-/*
- * This lock protects the cifs_file->llist and cifs_file->flist
- * list operations, and updates to some flags (cifs_file->invalidHandle)
- * It will be moved to either use the tcon->stat_lock or equivalent later.
- * If cifs_tcp_ses_lock and the lock below are both needed to be held, then
- * the cifs_tcp_ses_lock must be grabbed first and released last.
- */
-GLOBAL_EXTERN spinlock_t       cifs_file_list_lock;
-
 #ifdef CONFIG_CIFS_DNOTIFY_EXPERIMENTAL /* unused temporarily */
 /* Outstanding dir notify requests */
 GLOBAL_EXTERN struct list_head GlobalDnotifyReqList;
index f82d2823622f334cf83e677c042ead92a0379b0a..3f3185febc585f93fb9f88a0b0e96ec49b636f4a 100644 (file)
@@ -98,13 +98,13 @@ cifs_mark_open_files_invalid(struct cifs_tcon *tcon)
        struct list_head *tmp1;
 
        /* list all files open on tree connection and mark them invalid */
-       spin_lock(&cifs_file_list_lock);
+       spin_lock(&tcon->open_file_lock);
        list_for_each_safe(tmp, tmp1, &tcon->openFileList) {
                open_file = list_entry(tmp, struct cifsFileInfo, tlist);
                open_file->invalidHandle = true;
                open_file->oplock_break_cancelled = true;
        }
-       spin_unlock(&cifs_file_list_lock);
+       spin_unlock(&tcon->open_file_lock);
        /*
         * BB Add call to invalidate_inodes(sb) for all superblocks mounted
         * to this tcon.
index a95fe8b1afe93de6b091ce2015cd6f9dbb2f3f23..ee5ceae224118ceef87f6cab05344ce97ec1da14 100644 (file)
@@ -305,6 +305,7 @@ cifs_new_fileinfo(struct cifs_fid *fid, struct file *file,
        cfile->tlink = cifs_get_tlink(tlink);
        INIT_WORK(&cfile->oplock_break, cifs_oplock_break);
        mutex_init(&cfile->fh_mutex);
+       spin_lock_init(&cfile->file_info_lock);
 
        cifs_sb_active(inode->i_sb);
 
@@ -317,7 +318,7 @@ cifs_new_fileinfo(struct cifs_fid *fid, struct file *file,
                oplock = 0;
        }
 
-       spin_lock(&cifs_file_list_lock);
+       spin_lock(&tcon->open_file_lock);
        if (fid->pending_open->oplock != CIFS_OPLOCK_NO_CHANGE && oplock)
                oplock = fid->pending_open->oplock;
        list_del(&fid->pending_open->olist);
@@ -326,12 +327,13 @@ cifs_new_fileinfo(struct cifs_fid *fid, struct file *file,
        server->ops->set_fid(cfile, fid, oplock);
 
        list_add(&cfile->tlist, &tcon->openFileList);
+
        /* if readable file instance put first in list*/
        if (file->f_mode & FMODE_READ)
                list_add(&cfile->flist, &cinode->openFileList);
        else
                list_add_tail(&cfile->flist, &cinode->openFileList);
-       spin_unlock(&cifs_file_list_lock);
+       spin_unlock(&tcon->open_file_lock);
 
        if (fid->purge_cache)
                cifs_zap_mapping(inode);
@@ -343,16 +345,16 @@ cifs_new_fileinfo(struct cifs_fid *fid, struct file *file,
 struct cifsFileInfo *
 cifsFileInfo_get(struct cifsFileInfo *cifs_file)
 {
-       spin_lock(&cifs_file_list_lock);
+       spin_lock(&cifs_file->file_info_lock);
        cifsFileInfo_get_locked(cifs_file);
-       spin_unlock(&cifs_file_list_lock);
+       spin_unlock(&cifs_file->file_info_lock);
        return cifs_file;
 }
 
 /*
  * Release a reference on the file private data. This may involve closing
  * the filehandle out on the server. Must be called without holding
- * cifs_file_list_lock.
+ * tcon->open_file_lock and cifs_file->file_info_lock.
  */
 void cifsFileInfo_put(struct cifsFileInfo *cifs_file)
 {
@@ -367,11 +369,15 @@ void cifsFileInfo_put(struct cifsFileInfo *cifs_file)
        struct cifs_pending_open open;
        bool oplock_break_cancelled;
 
-       spin_lock(&cifs_file_list_lock);
+       spin_lock(&tcon->open_file_lock);
+
+       spin_lock(&cifs_file->file_info_lock);
        if (--cifs_file->count > 0) {
-               spin_unlock(&cifs_file_list_lock);
+               spin_unlock(&cifs_file->file_info_lock);
+               spin_unlock(&tcon->open_file_lock);
                return;
        }
+       spin_unlock(&cifs_file->file_info_lock);
 
        if (server->ops->get_lease_key)
                server->ops->get_lease_key(inode, &fid);
@@ -395,7 +401,8 @@ void cifsFileInfo_put(struct cifsFileInfo *cifs_file)
                        set_bit(CIFS_INO_INVALID_MAPPING, &cifsi->flags);
                cifs_set_oplock_level(cifsi, 0);
        }
-       spin_unlock(&cifs_file_list_lock);
+
+       spin_unlock(&tcon->open_file_lock);
 
        oplock_break_cancelled = cancel_work_sync(&cifs_file->oplock_break);
 
@@ -772,10 +779,10 @@ int cifs_closedir(struct inode *inode, struct file *file)
        server = tcon->ses->server;
 
        cifs_dbg(FYI, "Freeing private data in close dir\n");
-       spin_lock(&cifs_file_list_lock);
+       spin_lock(&cfile->file_info_lock);
        if (server->ops->dir_needs_close(cfile)) {
                cfile->invalidHandle = true;
-               spin_unlock(&cifs_file_list_lock);
+               spin_unlock(&cfile->file_info_lock);
                if (server->ops->close_dir)
                        rc = server->ops->close_dir(xid, tcon, &cfile->fid);
                else
@@ -784,7 +791,7 @@ int cifs_closedir(struct inode *inode, struct file *file)
                /* not much we can do if it fails anyway, ignore rc */
                rc = 0;
        } else
-               spin_unlock(&cifs_file_list_lock);
+               spin_unlock(&cfile->file_info_lock);
 
        buf = cfile->srch_inf.ntwrk_buf_start;
        if (buf) {
@@ -1728,12 +1735,13 @@ struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode,
 {
        struct cifsFileInfo *open_file = NULL;
        struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
+       struct cifs_tcon *tcon = cifs_sb_master_tcon(cifs_sb);
 
        /* only filter by fsuid on multiuser mounts */
        if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
                fsuid_only = false;
 
-       spin_lock(&cifs_file_list_lock);
+       spin_lock(&tcon->open_file_lock);
        /* we could simply get the first_list_entry since write-only entries
           are always at the end of the list but since the first entry might
           have a close pending, we go through the whole list */
@@ -1744,8 +1752,8 @@ struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode,
                        if (!open_file->invalidHandle) {
                                /* found a good file */
                                /* lock it so it will not be closed on us */
-                               cifsFileInfo_get_locked(open_file);
-                               spin_unlock(&cifs_file_list_lock);
+                               cifsFileInfo_get(open_file);
+                               spin_unlock(&tcon->open_file_lock);
                                return open_file;
                        } /* else might as well continue, and look for
                             another, or simply have the caller reopen it
@@ -1753,7 +1761,7 @@ struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode,
                } else /* write only file */
                        break; /* write only files are last so must be done */
        }
-       spin_unlock(&cifs_file_list_lock);
+       spin_unlock(&tcon->open_file_lock);
        return NULL;
 }
 
@@ -1762,6 +1770,7 @@ struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode,
 {
        struct cifsFileInfo *open_file, *inv_file = NULL;
        struct cifs_sb_info *cifs_sb;
+       struct cifs_tcon *tcon;
        bool any_available = false;
        int rc;
        unsigned int refind = 0;
@@ -1777,15 +1786,16 @@ struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode,
        }
 
        cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
+       tcon = cifs_sb_master_tcon(cifs_sb);
 
        /* only filter by fsuid on multiuser mounts */
        if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
                fsuid_only = false;
 
-       spin_lock(&cifs_file_list_lock);
+       spin_lock(&tcon->open_file_lock);
 refind_writable:
        if (refind > MAX_REOPEN_ATT) {
-               spin_unlock(&cifs_file_list_lock);
+               spin_unlock(&tcon->open_file_lock);
                return NULL;
        }
        list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
@@ -1796,8 +1806,8 @@ refind_writable:
                if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
                        if (!open_file->invalidHandle) {
                                /* found a good writable file */
-                               cifsFileInfo_get_locked(open_file);
-                               spin_unlock(&cifs_file_list_lock);
+                               cifsFileInfo_get(open_file);
+                               spin_unlock(&tcon->open_file_lock);
                                return open_file;
                        } else {
                                if (!inv_file)
@@ -1813,24 +1823,24 @@ refind_writable:
 
        if (inv_file) {
                any_available = false;
-               cifsFileInfo_get_locked(inv_file);
+               cifsFileInfo_get(inv_file);
        }
 
-       spin_unlock(&cifs_file_list_lock);
+       spin_unlock(&tcon->open_file_lock);
 
        if (inv_file) {
                rc = cifs_reopen_file(inv_file, false);
                if (!rc)
                        return inv_file;
                else {
-                       spin_lock(&cifs_file_list_lock);
+                       spin_lock(&tcon->open_file_lock);
                        list_move_tail(&inv_file->flist,
                                        &cifs_inode->openFileList);
-                       spin_unlock(&cifs_file_list_lock);
+                       spin_unlock(&tcon->open_file_lock);
                        cifsFileInfo_put(inv_file);
-                       spin_lock(&cifs_file_list_lock);
                        ++refind;
                        inv_file = NULL;
+                       spin_lock(&tcon->open_file_lock);
                        goto refind_writable;
                }
        }
@@ -3612,15 +3622,17 @@ static int cifs_readpage(struct file *file, struct page *page)
 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
 {
        struct cifsFileInfo *open_file;
+       struct cifs_tcon *tcon =
+               cifs_sb_master_tcon(CIFS_SB(cifs_inode->vfs_inode.i_sb));
 
-       spin_lock(&cifs_file_list_lock);
+       spin_lock(&tcon->open_file_lock);
        list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
                if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
-                       spin_unlock(&cifs_file_list_lock);
+                       spin_unlock(&tcon->open_file_lock);
                        return 1;
                }
        }
-       spin_unlock(&cifs_file_list_lock);
+       spin_unlock(&tcon->open_file_lock);
        return 0;
 }
 
index 813fe13c2ae175869cdc6db06a19392d09d7f05f..c6729156f9a00cf69938f8ec66c92fde561849ab 100644 (file)
@@ -120,6 +120,7 @@ tconInfoAlloc(void)
                ++ret_buf->tc_count;
                INIT_LIST_HEAD(&ret_buf->openFileList);
                INIT_LIST_HEAD(&ret_buf->tcon_list);
+               spin_lock_init(&ret_buf->open_file_lock);
 #ifdef CONFIG_CIFS_STATS
                spin_lock_init(&ret_buf->stat_lock);
 #endif
@@ -465,7 +466,7 @@ is_valid_oplock_break(char *buffer, struct TCP_Server_Info *srv)
                                continue;
 
                        cifs_stats_inc(&tcon->stats.cifs_stats.num_oplock_brks);
-                       spin_lock(&cifs_file_list_lock);
+                       spin_lock(&tcon->open_file_lock);
                        list_for_each(tmp2, &tcon->openFileList) {
                                netfile = list_entry(tmp2, struct cifsFileInfo,
                                                     tlist);
@@ -495,11 +496,11 @@ is_valid_oplock_break(char *buffer, struct TCP_Server_Info *srv)
                                           &netfile->oplock_break);
                                netfile->oplock_break_cancelled = false;
 
-                               spin_unlock(&cifs_file_list_lock);
+                               spin_unlock(&tcon->open_file_lock);
                                spin_unlock(&cifs_tcp_ses_lock);
                                return true;
                        }
-                       spin_unlock(&cifs_file_list_lock);
+                       spin_unlock(&tcon->open_file_lock);
                        spin_unlock(&cifs_tcp_ses_lock);
                        cifs_dbg(FYI, "No matching file for oplock break\n");
                        return true;
@@ -613,9 +614,9 @@ backup_cred(struct cifs_sb_info *cifs_sb)
 void
 cifs_del_pending_open(struct cifs_pending_open *open)
 {
-       spin_lock(&cifs_file_list_lock);
+       spin_lock(&tlink_tcon(open->tlink)->open_file_lock);
        list_del(&open->olist);
-       spin_unlock(&cifs_file_list_lock);
+       spin_unlock(&tlink_tcon(open->tlink)->open_file_lock);
 }
 
 void
@@ -635,7 +636,7 @@ void
 cifs_add_pending_open(struct cifs_fid *fid, struct tcon_link *tlink,
                      struct cifs_pending_open *open)
 {
-       spin_lock(&cifs_file_list_lock);
+       spin_lock(&tlink_tcon(tlink)->open_file_lock);
        cifs_add_pending_open_locked(fid, tlink, open);
-       spin_unlock(&cifs_file_list_lock);
+       spin_unlock(&tlink_tcon(open->tlink)->open_file_lock);
 }
index 65cf85dcda09bd773886c8aa6a86d82af1a033b5..8f6a2a5863b9d9275bfb6afb00fc16b867101275 100644 (file)
@@ -597,14 +597,14 @@ find_cifs_entry(const unsigned int xid, struct cifs_tcon *tcon, loff_t pos,
             is_dir_changed(file)) || (index_to_find < first_entry_in_buffer)) {
                /* close and restart search */
                cifs_dbg(FYI, "search backing up - close and restart search\n");
-               spin_lock(&cifs_file_list_lock);
+               spin_lock(&cfile->file_info_lock);
                if (server->ops->dir_needs_close(cfile)) {
                        cfile->invalidHandle = true;
-                       spin_unlock(&cifs_file_list_lock);
+                       spin_unlock(&cfile->file_info_lock);
                        if (server->ops->close_dir)
                                server->ops->close_dir(xid, tcon, &cfile->fid);
                } else
-                       spin_unlock(&cifs_file_list_lock);
+                       spin_unlock(&cfile->file_info_lock);
                if (cfile->srch_inf.ntwrk_buf_start) {
                        cifs_dbg(FYI, "freeing SMB ff cache buf on search rewind\n");
                        if (cfile->srch_inf.smallBuf)
index 389fb9f8c84e22308ac5fcb44c27f005bee6e27c..3d383489b9cf313395e94a0bff4a719501b4b309 100644 (file)
@@ -549,19 +549,19 @@ smb2_is_valid_lease_break(char *buffer)
                list_for_each(tmp1, &server->smb_ses_list) {
                        ses = list_entry(tmp1, struct cifs_ses, smb_ses_list);
 
-                       spin_lock(&cifs_file_list_lock);
                        list_for_each(tmp2, &ses->tcon_list) {
                                tcon = list_entry(tmp2, struct cifs_tcon,
                                                  tcon_list);
+                               spin_lock(&tcon->open_file_lock);
                                cifs_stats_inc(
                                    &tcon->stats.cifs_stats.num_oplock_brks);
                                if (smb2_tcon_has_lease(tcon, rsp, lw)) {
-                                       spin_unlock(&cifs_file_list_lock);
+                                       spin_unlock(&tcon->open_file_lock);
                                        spin_unlock(&cifs_tcp_ses_lock);
                                        return true;
                                }
+                               spin_unlock(&tcon->open_file_lock);
                        }
-                       spin_unlock(&cifs_file_list_lock);
                }
        }
        spin_unlock(&cifs_tcp_ses_lock);
@@ -603,7 +603,7 @@ smb2_is_valid_oplock_break(char *buffer, struct TCP_Server_Info *server)
                        tcon = list_entry(tmp1, struct cifs_tcon, tcon_list);
 
                        cifs_stats_inc(&tcon->stats.cifs_stats.num_oplock_brks);
-                       spin_lock(&cifs_file_list_lock);
+                       spin_lock(&tcon->open_file_lock);
                        list_for_each(tmp2, &tcon->openFileList) {
                                cfile = list_entry(tmp2, struct cifsFileInfo,
                                                     tlist);
@@ -615,7 +615,7 @@ smb2_is_valid_oplock_break(char *buffer, struct TCP_Server_Info *server)
 
                                cifs_dbg(FYI, "file id match, oplock break\n");
                                cinode = CIFS_I(d_inode(cfile->dentry));
-
+                               spin_lock(&cfile->file_info_lock);
                                if (!CIFS_CACHE_WRITE(cinode) &&
                                    rsp->OplockLevel == SMB2_OPLOCK_LEVEL_NONE)
                                        cfile->oplock_break_cancelled = true;
@@ -637,14 +637,14 @@ smb2_is_valid_oplock_break(char *buffer, struct TCP_Server_Info *server)
                                        clear_bit(
                                           CIFS_INODE_DOWNGRADE_OPLOCK_TO_L2,
                                           &cinode->flags);
-
+                               spin_unlock(&cfile->file_info_lock);
                                queue_work(cifsiod_wq, &cfile->oplock_break);
 
-                               spin_unlock(&cifs_file_list_lock);
+                               spin_unlock(&tcon->open_file_lock);
                                spin_unlock(&cifs_tcp_ses_lock);
                                return true;
                        }
-                       spin_unlock(&cifs_file_list_lock);
+                       spin_unlock(&tcon->open_file_lock);
                        spin_unlock(&cifs_tcp_ses_lock);
                        cifs_dbg(FYI, "No matching file for oplock break\n");
                        return true;