Whamcloud - gitweb
LU-744 llite: reimplement ll_get_fsname()
[fs/lustre-release.git] / lustre / llite / llite_lib.c
index ec264c4..08936e2 100644 (file)
 cfs_mem_cache_t *ll_file_data_slab;
 
 CFS_LIST_HEAD(ll_super_blocks);
-cfs_spinlock_t ll_sb_lock = CFS_SPIN_LOCK_UNLOCKED;
+DEFINE_SPINLOCK(ll_sb_lock);
 
 #ifndef MS_HAS_NEW_AOPS
 extern struct address_space_operations ll_aops;
-extern struct address_space_operations ll_dir_aops;
 #else
 extern struct address_space_operations_ext ll_aops;
-extern struct address_space_operations_ext ll_dir_aops;
 #endif
 
 #ifndef log2
@@ -227,6 +225,12 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 #ifdef CONFIG_FS_POSIX_ACL
         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 #endif
+
+       if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
+               /* flag mdc connection as lightweight, only used for test
+                * purpose, use with care */
+                data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
+
         data->ocd_ibits_known = MDS_INODELOCK_FULL;
         data->ocd_version = LUSTRE_VERSION_CODE;
 
@@ -372,6 +376,11 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
         else
                 sbi->ll_md_brw_size = CFS_PAGE_SIZE;
 
+       if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
+               LCONSOLE_INFO("Layout lock feature supported.\n");
+               sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
+       }
+
         obd = class_name2obd(dt);
         if (!obd) {
                 CERROR("DT %s: not setup or attached\n", dt);
@@ -458,7 +467,7 @@ 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;
-#if THREAD_SIZE >= 8192
+#if THREAD_SIZE >= 8192 /*b=17630*/
         sb->s_export_op = &lustre_export_operations;
 #endif
 
@@ -553,7 +562,6 @@ static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
         uuid = obd_get_uuid(sbi->ll_md_exp);
         if (uuid != NULL)
                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
-        sbi->ll_mnt = mnt;
 
         if (data != NULL)
                 OBD_FREE_PTR(data);
@@ -634,10 +642,10 @@ void lustre_dump_dentry(struct dentry *dentry, int recur)
         if (recur == 0)
                 return;
 
-       list_for_each(tmp, &dentry->d_subdirs) {
-                struct dentry *d = list_entry(tmp, struct dentry, d_child);
-                lustre_dump_dentry(d, recur - 1);
-        }
+       list_for_each(tmp, &dentry->d_subdirs) {
+               struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
+               lustre_dump_dentry(d, recur - 1);
+       }
 }
 
 void client_common_put_super(struct super_block *sb)
@@ -768,7 +776,7 @@ static int ll_options(char *options, int *flags)
                         *flags &= ~tmp;
                         goto next;
                 }
-#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,5,50,0)
+#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 50, 0)
                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
                 if (tmp) {
                         /* Ignore deprecated mount option.  The client will
@@ -792,6 +800,16 @@ static int ll_options(char *options, int *flags)
                         *flags |= tmp;
                         goto next;
                 }
+               tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
+               if (tmp) {
+                       *flags |= tmp;
+                       goto next;
+               }
+               tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
+               if (tmp) {
+                       *flags &= ~tmp;
+                       goto next;
+               }
 
                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
                 if (tmp) {
@@ -891,7 +909,6 @@ void ll_lli_init(struct ll_inode_info *lli)
                 cfs_mutex_init(&lli->lli_readdir_mutex);
                 lli->lli_opendir_key = NULL;
                 lli->lli_sai = NULL;
-                lli->lli_sa_pos = 0;
                 lli->lli_def_acl = NULL;
                 cfs_spin_lock_init(&lli->lli_sa_lock);
                 lli->lli_opendir_pid = 0;
@@ -901,13 +918,13 @@ void ll_lli_init(struct ll_inode_info *lli)
                 lli->lli_symlink_name = NULL;
                 cfs_init_rwsem(&lli->lli_trunc_sem);
                 cfs_mutex_init(&lli->lli_write_mutex);
-                lli->lli_async_rc = 0;
-                lli->lli_write_rc = 0;
-                cfs_init_rwsem(&lli->lli_glimpse_sem);
-                lli->lli_glimpse_time = 0;
-                CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
-                lli->lli_agl_index = 0;
-        }
+               cfs_init_rwsem(&lli->lli_glimpse_sem);
+               lli->lli_glimpse_time = 0;
+               CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
+               lli->lli_agl_index = 0;
+               lli->lli_async_rc = 0;
+       }
+       cfs_mutex_init(&lli->lli_layout_mutex);
 }
 
 static inline int ll_bdi_register(struct backing_dev_info *bdi)
@@ -976,7 +993,7 @@ int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
            Use the address of the super itself.*/
         cfg->cfg_instance = sb;
         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
-
+       cfg->cfg_callback = class_config_llog_handler;
         /* set up client obds */
         err = lustre_process_log(sb, profilenm, cfg);
         if (err < 0) {
@@ -1309,6 +1326,50 @@ static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
         return rc;
 }
 
+#ifndef HAVE_VFS_INODE_NEWSIZE_OK
+/**
+ * inode_newsize_ok - may this inode be truncated to a given size
+ * @inode:      the inode to be truncated
+ * @offset:     the new size to assign to the inode
+ * @Returns:    0 on success, -ve errno on failure
+ *
+ * inode_newsize_ok will check filesystem limits and ulimits to check that the
+ * new inode size is within limits. inode_newsize_ok will also send SIGXFSZ
+ * when necessary. Caller must not proceed with inode size change if failure is
+ * returned. @inode must be a file (not directory), with appropriate
+ * permissions to allow truncate (inode_newsize_ok does NOT check these
+ * conditions).
+ *
+ * inode_newsize_ok must be called with i_mutex held.
+ */
+int inode_newsize_ok(const struct inode *inode, loff_t offset)
+{
+       if (inode->i_size < offset) {
+               unsigned long limit;
+
+               limit = rlimit(RLIMIT_FSIZE);
+               if (limit != RLIM_INFINITY && offset > limit)
+                       goto out_sig;
+               if (offset > inode->i_sb->s_maxbytes)
+                       goto out_big;
+       } else {
+               /*
+                * truncation of in-use swapfiles is disallowed - it would
+                * cause subsequent swapout to scribble on the now-freed
+                * blocks.
+                */
+               if (IS_SWAPFILE(inode))
+                       return -ETXTBSY;
+       }
+
+       return 0;
+out_sig:
+       send_sig(SIGXFSZ, current, 0);
+out_big:
+       return -EFBIG;
+}
+#endif
+
 /* If this inode has objects allocated to it (lsm != NULL), then the OST
  * object(s) determine the file size and mtime.  Otherwise, the MDS will
  * keep these values until such a time that objects are allocated for it.
@@ -1388,15 +1449,15 @@ int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
         if (op_data == NULL)
                 RETURN(-ENOMEM);
 
-        if (!S_ISDIR(inode->i_mode)) {
-                if (ia_valid & ATTR_SIZE)
-                        UP_WRITE_I_ALLOC_SEM(inode);
-                UNLOCK_INODE_MUTEX(inode);
-                cfs_down_write(&lli->lli_trunc_sem);
-                LOCK_INODE_MUTEX(inode);
-                if (ia_valid & ATTR_SIZE)
-                        DOWN_WRITE_I_ALLOC_SEM(inode);
-        }
+       if (!S_ISDIR(inode->i_mode)) {
+               if (ia_valid & ATTR_SIZE)
+                       UP_WRITE_I_ALLOC_SEM(inode);
+               mutex_unlock(&inode->i_mutex);
+               cfs_down_write(&lli->lli_trunc_sem);
+               mutex_lock(&inode->i_mutex);
+               if (ia_valid & ATTR_SIZE)
+                       DOWN_WRITE_I_ALLOC_SEM(inode);
+       }
 
        /* We need a steady stripe configuration for setattr to avoid
         * confusion. */
@@ -1533,16 +1594,11 @@ int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
 
         RETURN(rc);
 }
-#ifndef HAVE_STATFS_DENTRY_PARAM
-int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
-{
-#else
 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
 {
-        struct super_block *sb = de->d_sb;
-#endif
-        struct obd_statfs osfs;
-        int rc;
+       struct super_block *sb = de->d_sb;
+       struct obd_statfs osfs;
+       int rc;
 
         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
@@ -1602,16 +1658,15 @@ void ll_inode_size_unlock(struct inode *inode)
 
 void ll_update_inode(struct inode *inode, struct lustre_md *md)
 {
-        struct ll_inode_info *lli = ll_i2info(inode);
-        struct mdt_body *body = md->body;
-        struct lov_stripe_md *lsm = md->lsm;
-        struct ll_sb_info *sbi = ll_i2sbi(inode);
-
-        LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
-        if (lsm != NULL) {
-                LASSERT(S_ISREG(inode->i_mode));
-
-                cfs_mutex_lock(&lli->lli_och_mutex);
+       struct ll_inode_info *lli = ll_i2info(inode);
+       struct mdt_body *body = md->body;
+       struct lov_stripe_md *lsm = md->lsm;
+       struct ll_sb_info *sbi = ll_i2sbi(inode);
+
+       LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
+       if (lsm != NULL) {
+               LASSERT(S_ISREG(inode->i_mode));
+               cfs_mutex_lock(&lli->lli_och_mutex);
                CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
                                lsm, inode->i_ino, inode->i_generation, inode);
                /* cl_file_inode_init must go before lli_has_smd or a race
@@ -1621,12 +1676,13 @@ void ll_update_inode(struct inode *inode, struct lustre_md *md)
                if (cl_file_inode_init(inode, md) == 0)
                        lli->lli_has_smd = true;
                cfs_mutex_unlock(&lli->lli_och_mutex);
+
                lli->lli_maxbytes = lsm->lsm_maxbytes;
                if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
                        lli->lli_maxbytes = MAX_LFS_FILESIZE;
                if (md->lsm != NULL)
                        obd_free_memmd(ll_i2dtexp(inode), &md->lsm);
-        }
+       }
 
         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
                 if (body->valid & OBD_MD_FLRMTPERM)
@@ -1673,9 +1729,6 @@ void ll_update_inode(struct inode *inode, struct lustre_md *md)
         } else {
                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
         }
-#ifdef HAVE_INODE_BLKSIZE
-        inode->i_blksize = 1<<inode->i_blkbits;
-#endif
         if (body->valid & OBD_MD_FLUID)
                 inode->i_uid = body->uid;
         if (body->valid & OBD_MD_FLGID)
@@ -1788,7 +1841,6 @@ void ll_read_inode2(struct inode *inode, void *opaque)
         } else if (S_ISDIR(inode->i_mode)) {
                 inode->i_op = &ll_dir_inode_operations;
                 inode->i_fop = &ll_dir_operations;
-                inode->i_mapping->a_ops = (struct address_space_operations *)&ll_dir_aops;
                 EXIT;
         } else if (S_ISLNK(inode->i_mode)) {
                 inode->i_op = &ll_fast_symlink_inode_operations;
@@ -1960,7 +2012,6 @@ void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
 void ll_umount_begin(struct super_block *sb)
 {
 #endif
-        struct lustre_sb_info *lsi = s2lsi(sb);
         struct ll_sb_info *sbi = ll_s2sbi(sb);
         struct obd_device *obd;
         struct obd_ioctl_data *ioc_data;
@@ -1973,9 +2024,6 @@ void ll_umount_begin(struct super_block *sb)
         }
 #endif
 
-        /* Tell the MGC we got umount -f */
-        lsi->lsi_flags |= LSI_UMOUNT_FORCE;
-
         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
                sb->s_count, atomic_read(&sb->s_active));
 
@@ -2064,8 +2112,9 @@ int ll_prep_inode(struct inode **inode,
                   struct ptlrpc_request *req,
                   struct super_block *sb)
 {
-        struct ll_sb_info *sbi = NULL;
-        struct lustre_md md;
+       struct ll_sb_info *sbi = NULL;
+       struct lustre_md md;
+       __u64 ibits;
         int rc;
         ENTRY;
 
@@ -2104,9 +2153,18 @@ int ll_prep_inode(struct inode **inode,
                 }
         }
 
+       /* sanity check for LAYOUT lock. */
+       ibits = MDS_INODELOCK_LAYOUT;
+       if (S_ISREG(md.body->mode) && sbi->ll_flags & LL_SBI_LAYOUT_LOCK &&
+           md.lsm != NULL && !ll_have_md_lock(*inode, &ibits, LCK_MINMODE)) {
+               CERROR("%s: inode "DFID" (%p) layout lock not granted.\n",
+                       ll_get_fsname(sb, NULL, 0),
+                       PFID(ll_inode2fid(*inode)), *inode);
+       }
+
 out:
-        md_free_lustre_md(sbi->ll_md_exp, &md);
-        RETURN(rc);
+       md_free_lustre_md(sbi->ll_md_exp, &md);
+       RETURN(rc);
 }
 
 int ll_obd_statfs(struct inode *inode, void *arg)
@@ -2116,6 +2174,7 @@ int ll_obd_statfs(struct inode *inode, void *arg)
         char *buf = NULL;
         struct obd_ioctl_data *data = NULL;
         __u32 type;
+       __u32 flags;
         int len = 0, rc;
 
         if (!inode || !(sbi = ll_i2sbi(inode)))
@@ -2137,14 +2196,15 @@ int ll_obd_statfs(struct inode *inode, void *arg)
                 GOTO(out_statfs, rc = -EINVAL);
 
         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
-        if (type == LL_STATFS_LMV)
+       if (type & LL_STATFS_LMV)
                 exp = sbi->ll_md_exp;
-        else if (type == LL_STATFS_LOV)
+       else if (type & LL_STATFS_LOV)
                 exp = sbi->ll_dt_exp;
         else
                 GOTO(out_statfs, rc = -ENODEV);
 
-        rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
+       flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
+       rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
         if (rc)
                 GOTO(out_statfs, rc);
 out_statfs:
@@ -2269,6 +2329,9 @@ int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
                 seq_puts(seq, ",lazystatfs");
 
+       if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
+               seq_puts(seq, ",user_fid2path");
+
         RETURN(0);
 }
 
@@ -2297,3 +2360,36 @@ int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
 
         RETURN(0);
 }
+
+/**
+ * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
+ * fsname will be returned in this buffer; otherwise, a static buffer will be
+ * used to store the fsname and returned to caller.
+ */
+char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
+{
+       static char fsname_static[MTI_NAME_MAXLEN];
+       struct lustre_sb_info *lsi = s2lsi(sb);
+       char *ptr;
+       int len;
+
+       if (buf == NULL) {
+               /* this means the caller wants to use static buffer
+                * and it doesn't care about race. Usually this is
+                * in error reporting path */
+               buf = fsname_static;
+               buflen = sizeof(fsname_static);
+       }
+
+       len = strlen(lsi->lsi_lmd->lmd_profile);
+       ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
+       if (ptr && (strcmp(ptr, "-client") == 0))
+               len -= 7;
+
+       if (unlikely(len >= buflen))
+               len = buflen - 1;
+       strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
+       buf[len] = '\0';
+
+       return buf;
+}