Whamcloud - gitweb
LU-12631 llite: report latency for filesystem ops
[fs/lustre-release.git] / lustre / llite / llite_lib.c
index 5ba34db..22fa391 100644 (file)
@@ -255,6 +255,7 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
                                   OBD_CONNECT2_FLR |
                                   OBD_CONNECT2_LOCK_CONVERT |
                                   OBD_CONNECT2_ARCHIVE_ID_ARRAY |
+                                  OBD_CONNECT2_INC_XID |
                                   OBD_CONNECT2_LSOM |
                                   OBD_CONNECT2_ASYNC_DISCARD |
                                   OBD_CONNECT2_PCC;
@@ -263,7 +264,7 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 #endif
-#ifdef CONFIG_FS_POSIX_ACL
+#ifdef CONFIG_LUSTRE_FS_POSIX_ACL
        data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
                                   OBD_CONNECT_LARGE_ACL;
 #endif
@@ -464,7 +465,8 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
        data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
 #endif
 
-       data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
+       data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD |
+                                  OBD_CONNECT2_INC_XID;
 
        if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
                data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
@@ -554,9 +556,7 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
        CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 
        sb->s_op = &lustre_super_operations;
-#ifdef HAVE_XATTR_HANDLER_FLAGS
        sb->s_xattr = ll_xattr_handlers;
-#endif
 #if THREAD_SIZE >= 8192 /*b=17630*/
        sb->s_export_op = &lustre_export_operations;
 #endif
@@ -600,7 +600,7 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
        ptlrpc_req_finished(request);
 
        if (IS_ERR(root)) {
-#ifdef CONFIG_FS_POSIX_ACL
+#ifdef CONFIG_LUSTRE_FS_POSIX_ACL
                if (lmd.posix_acl) {
                        posix_acl_release(lmd.posix_acl);
                        lmd.posix_acl = NULL;
@@ -608,7 +608,8 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 #endif
                err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
                root = NULL;
-               CERROR("lustre_lite: bad iget4 for root\n");
+               CERROR("%s: bad ll_iget() for root: rc = %d\n",
+                      sbi->ll_fsname, err);
                GOTO(out_root, err);
        }
 
@@ -632,9 +633,6 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
                       sbi->ll_fsname, err);
                GOTO(out_root, err);
        }
-#ifdef HAVE_DCACHE_LOCK
-       sb->s_root->d_op = &ll_d_ops;
-#endif
 
        sbi->ll_sdev_orig = sb->s_dev;
 
@@ -809,12 +807,11 @@ void ll_kill_super(struct super_block *sb)
         * put_super not affected real removing devices */
        if (sbi) {
                sb->s_dev = sbi->ll_sdev_orig;
-               sbi->ll_umounting = 1;
 
                /* wait running statahead threads to quit */
                while (atomic_read(&sbi->ll_sa_running) > 0) {
                        set_current_state(TASK_UNINTERRUPTIBLE);
-                       schedule_timeout(cfs_time_seconds(1 >> 3));
+                       schedule_timeout(cfs_time_seconds(1) >> 3);
                }
        }
 
@@ -842,38 +839,38 @@ static int ll_options(char *options, struct ll_sb_info *sbi)
 
        CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 
-        while (*s1) {
-                CDEBUG(D_SUPER, "next opt=%s\n", s1);
-                tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
-                if (tmp) {
-                        *flags |= tmp;
-                        goto next;
-                }
-                tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
-                if (tmp) {
-                        *flags |= tmp;
-                        goto next;
-                }
-                tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
-                if (tmp) {
-                        *flags |= tmp;
-                        goto next;
-                }
-                tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
-                if (tmp) {
-                        *flags &= ~tmp;
-                        goto next;
-                }
-                tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
-                if (tmp) {
-                        *flags |= tmp;
-                        goto next;
-                }
-                tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
-                if (tmp) {
-                        *flags &= ~tmp;
-                        goto next;
-                }
+       while (*s1) {
+               CDEBUG(D_SUPER, "next opt=%s\n", s1);
+               tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
+               if (tmp) {
+                       *flags |= tmp;
+                       goto next;
+               }
+               tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
+               if (tmp) {
+                       *flags = (*flags & ~LL_SBI_LOCALFLOCK) | tmp;
+                       goto next;
+               }
+               tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
+               if (tmp) {
+                       *flags = (*flags & ~LL_SBI_FLOCK) | tmp;
+                       goto next;
+               }
+               tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
+               if (tmp) {
+                       *flags &= ~tmp;
+                       goto next;
+               }
+               tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
+               if (tmp) {
+                       *flags |= tmp;
+                       goto next;
+               }
+               tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
+               if (tmp) {
+                       *flags &= ~tmp;
+                       goto next;
+               }
                tmp = ll_set_opt("context", s1, 1);
                if (tmp)
                        goto next;
@@ -988,7 +985,6 @@ void ll_lli_init(struct ll_inode_info *lli)
 
        LASSERT(lli->lli_vfs_inode.i_mode != 0);
        if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
-               mutex_init(&lli->lli_readdir_mutex);
                lli->lli_opendir_key = NULL;
                lli->lli_sai = NULL;
                spin_lock_init(&lli->lli_sa_lock);
@@ -1011,6 +1007,9 @@ void ll_lli_init(struct ll_inode_info *lli)
                mutex_init(&lli->lli_pcc_lock);
                lli->lli_pcc_state = PCC_STATE_FL_NONE;
                lli->lli_pcc_inode = NULL;
+               mutex_init(&lli->lli_group_mutex);
+               lli->lli_group_users = 0;
+               lli->lli_group_gid = 0;
        }
        mutex_init(&lli->lli_layout_mutex);
        memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
@@ -1091,10 +1090,9 @@ int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
        if (err)
                GOTO(out_free_cfg, err);
 
-#ifndef HAVE_DCACHE_LOCK
        /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
        sb->s_d_op = &ll_d_ops;
-#endif
+
        /* UUID handling */
        generate_random_uuid(uuid.b);
        snprintf(sbi->ll_sb_uuid.uuid, UUID_SIZE, "%pU", uuid.b);
@@ -1624,7 +1622,7 @@ void ll_clear_inode(struct inode *inode)
 
        ll_xattr_cache_destroy(inode);
 
-#ifdef CONFIG_FS_POSIX_ACL
+#ifdef CONFIG_LUSTRE_FS_POSIX_ACL
        forget_all_cached_acls(inode);
        if (lli->lli_posix_acl) {
                posix_acl_release(lli->lli_posix_acl);
@@ -1726,6 +1724,7 @@ int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
         struct inode *inode = dentry->d_inode;
         struct ll_inode_info *lli = ll_i2info(inode);
         struct md_op_data *op_data = NULL;
+       ktime_t kstart = ktime_get();
        int rc = 0;
 
        ENTRY;
@@ -1898,8 +1897,10 @@ out:
                inode_has_no_xattr(inode);
        }
 
-       ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
-                       LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
+       if (!rc)
+               ll_stats_ops_tally(ll_i2sbi(inode), attr->ia_valid & ATTR_SIZE ?
+                                       LPROC_LL_TRUNC : LPROC_LL_SETATTR,
+                                  ktime_us_delta(ktime_get(), kstart));
 
        return rc;
 }
@@ -1987,15 +1988,16 @@ int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
 out:
        RETURN(rc);
 }
+
 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
 {
        struct super_block *sb = de->d_sb;
        struct obd_statfs osfs;
        __u64 fsid = huge_encode_dev(sb->s_dev);
+       ktime_t kstart = ktime_get();
        int rc;
 
-       CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
-       ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STATFS, 1);
+       CDEBUG(D_VFSTRACE, "VFS Op:sb=%s (%p)\n", sb->s_id, sb);
 
        /* Some amount of caching on the client is allowed */
        rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
@@ -2018,11 +2020,15 @@ int ll_statfs(struct dentry *de, struct kstatfs *sfs)
                 }
         }
 
-        sfs->f_blocks = osfs.os_blocks;
-        sfs->f_bfree = osfs.os_bfree;
-        sfs->f_bavail = osfs.os_bavail;
+       sfs->f_blocks = osfs.os_blocks;
+       sfs->f_bfree = osfs.os_bfree;
+       sfs->f_bavail = osfs.os_bavail;
        sfs->f_fsid.val[0] = (__u32)fsid;
        sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
+
+       ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STATFS,
+                          ktime_us_delta(ktime_get(), kstart));
+
        return 0;
 }
 
@@ -2072,7 +2078,7 @@ int ll_update_inode(struct inode *inode, struct lustre_md *md)
                        return rc;
        }
 
-#ifdef CONFIG_FS_POSIX_ACL
+#ifdef CONFIG_LUSTRE_FS_POSIX_ACL
        if (body->mbo_valid & OBD_MD_FLACL) {
                spin_lock(&lli->lli_lock);
                if (lli->lli_posix_acl)
@@ -2236,6 +2242,8 @@ void ll_delete_inode(struct inode *inode)
        struct ll_inode_info *lli = ll_i2info(inode);
        struct address_space *mapping = &inode->i_data;
        unsigned long nrpages;
+       unsigned long flags;
+
        ENTRY;
 
        if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
@@ -2260,9 +2268,9 @@ void ll_delete_inode(struct inode *inode)
         */
        nrpages = mapping->nrpages;
        if (nrpages) {
-               xa_lock_irq(&mapping->i_pages);
+               xa_lock_irqsave(&mapping->i_pages, flags);
                nrpages = mapping->nrpages;
-               xa_unlock_irq(&mapping->i_pages);
+               xa_unlock_irqrestore(&mapping->i_pages, flags);
        } /* Workaround end */
 
        LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
@@ -2270,9 +2278,7 @@ void ll_delete_inode(struct inode *inode)
                 ll_i2sbi(inode)->ll_fsname,
                 PFID(ll_inode2fid(inode)), inode, nrpages);
 
-#ifdef HAVE_SBOPS_EVICT_INODE
        ll_clear_inode(inode);
-#endif
        clear_inode(inode);
 
         EXIT;
@@ -2405,16 +2411,16 @@ void ll_umount_begin(struct super_block *sb)
        }
        obd->obd_force = 1;
 
-        obd = class_exp2obd(sbi->ll_dt_exp);
-        if (obd == NULL) {
+       obd = class_exp2obd(sbi->ll_dt_exp);
+       if (obd == NULL) {
                CERROR("Invalid LOV connection handle %#llx\n",
-                       sbi->ll_dt_exp->exp_handle.h_cookie);
-                EXIT;
-                return;
-        }
-        obd->obd_force = 1;
+                      sbi->ll_dt_exp->exp_handle.h_cookie);
+               EXIT;
+               return;
+       }
+       obd->obd_force = 1;
 
-        OBD_ALLOC_PTR(ioc_data);
+       OBD_ALLOC_PTR(ioc_data);
        if (ioc_data) {
                obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
                              sizeof *ioc_data, ioc_data, NULL);
@@ -2559,7 +2565,7 @@ int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
                                             sbi->ll_flags & LL_SBI_32BIT_API),
                                 &md);
                if (IS_ERR(*inode)) {
-#ifdef CONFIG_FS_POSIX_ACL
+#ifdef CONFIG_LUSTRE_FS_POSIX_ACL
                         if (md.posix_acl) {
                                 posix_acl_release(md.posix_acl);
                                 md.posix_acl = NULL;
@@ -2764,7 +2770,7 @@ int ll_show_options(struct seq_file *seq, struct dentry *dentry)
 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
 #endif
 {
-        struct ll_sb_info *sbi;
+       struct ll_sb_info *sbi;
 
 #ifdef HAVE_SUPEROPS_USE_DENTRY
        LASSERT((seq != NULL) && (dentry != NULL));
@@ -2774,20 +2780,25 @@ int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
        sbi = ll_s2sbi(vfs->mnt_sb);
 #endif
 
-        if (sbi->ll_flags & LL_SBI_NOLCK)
-                seq_puts(seq, ",nolock");
-
-        if (sbi->ll_flags & LL_SBI_FLOCK)
-                seq_puts(seq, ",flock");
+       if (sbi->ll_flags & LL_SBI_NOLCK)
+               seq_puts(seq, ",nolock");
 
-        if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
-                seq_puts(seq, ",localflock");
+       /* "flock" is the default since 2.13, but it wasn't for many years,
+        * so it is still useful to print this to show it is enabled.
+        * Start to print "noflock" so it is now clear when flock is disabled.
+        */
+       if (sbi->ll_flags & LL_SBI_FLOCK)
+               seq_puts(seq, ",flock");
+       else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
+               seq_puts(seq, ",localflock");
+       else
+               seq_puts(seq, ",noflock");
 
-        if (sbi->ll_flags & LL_SBI_USER_XATTR)
-                seq_puts(seq, ",user_xattr");
+       if (sbi->ll_flags & LL_SBI_USER_XATTR)
+               seq_puts(seq, ",user_xattr");
 
-        if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
-                seq_puts(seq, ",lazystatfs");
+       if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
+               seq_puts(seq, ",lazystatfs");
 
        if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
                seq_puts(seq, ",user_fid2path");
@@ -2795,7 +2806,7 @@ int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
        if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
                seq_puts(seq, ",always_ping");
 
-        RETURN(0);
+       RETURN(0);
 }
 
 /**
@@ -2852,6 +2863,7 @@ void ll_dirty_page_discard_warn(struct page *page, int ioret)
                        path = ll_d_path(dentry, buf, PAGE_SIZE);
        }
 
+       /* The below message is checked in recovery-small.sh test_24b */
        CDEBUG(D_WARNING,
               "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
               "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
@@ -2880,12 +2892,12 @@ ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
        if (lum_size < 0)
                RETURN(lum_size);
 
-       OBD_ALLOC(*kbuf, lum_size);
+       OBD_ALLOC_LARGE(*kbuf, lum_size);
        if (*kbuf == NULL)
                RETURN(-ENOMEM);
 
        if (copy_from_user(*kbuf, md, lum_size) != 0) {
-               OBD_FREE(*kbuf, lum_size);
+               OBD_FREE_LARGE(*kbuf, lum_size);
                RETURN(-EFAULT);
        }
 
@@ -3017,13 +3029,8 @@ int ll_getparent(struct file *file, struct getparent __user *arg)
        if (rc < 0)
                GOTO(ldata_free, rc);
 
-#ifdef HAVE_XATTR_HANDLER_FLAGS
        rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
                           buf.lb_len, OBD_MD_FLXATTR);
-#else
-       rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
-                        buf.lb_len);
-#endif /* HAVE_XATTR_HANDLER_FLAGS */
        if (rc < 0)
                GOTO(lb_free, rc);