X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fllite%2Fllite_lib.c;h=f006c3802911be05534c596c5fb86ca7e3de22fc;hp=7968f4fe3a522cc0f8b34f80909da86ea2cd38a6;hb=ba2ecf04beab65a6a4c8399c6a740999de218ece;hpb=1285139134c14b329e743e14fd60de0fcecf0328 diff --git a/lustre/llite/llite_lib.c b/lustre/llite/llite_lib.c index 7968f4f..f006c38 100644 --- a/lustre/llite/llite_lib.c +++ b/lustre/llite/llite_lib.c @@ -27,14 +27,20 @@ #include #include #include +#include #include #include #include #include +#include +#include +#include +#include #include "llite_internal.h" kmem_cache_t *ll_file_data_slab; +kmem_cache_t *ll_intent_slab; extern struct address_space_operations ll_aops; extern struct address_space_operations ll_dir_aops; @@ -63,14 +69,34 @@ struct ll_sb_info *lustre_init_sbi(struct super_block *sb) INIT_LIST_HEAD(&sbi->ll_conn_chain); INIT_HLIST_HEAD(&sbi->ll_orphan_dentry_list); INIT_LIST_HEAD(&sbi->ll_mnt_list); + sema_init(&sbi->ll_gns_sem, 1); - init_completion(&sbi->ll_gns_completion); - sbi->ll_gns_state = LL_GNS_STATE_IDLE; + spin_lock_init(&sbi->ll_gns_lock); + INIT_LIST_HEAD(&sbi->ll_gns_sbi_head); + init_waitqueue_head(&sbi->ll_gns_waitq); + init_completion(&sbi->ll_gns_mount_finished); + + /* this later may be reset via /proc/fs/... */ + memcpy(sbi->ll_gns_oname, ".mntinfo", strlen(".mntinfo")); + sbi->ll_gns_oname[strlen(sbi->ll_gns_oname)] = '\0'; + + /* this later may be reset via /proc/fs/... */ + memcpy(sbi->ll_gns_upcall, "/usr/sbin/gns_upcall", + strlen("/usr/sbin/gns_upcall")); + sbi->ll_gns_upcall[strlen(sbi->ll_gns_upcall)] = '\0'; + + /* default values, may be changed via /proc/fs/... */ + sbi->ll_gns_state = LL_GNS_IDLE; + sbi->ll_gns_pending_dentry = NULL; + atomic_set(&sbi->ll_gns_enabled, 1); + sbi->ll_gns_tick = GNS_TICK_TIMEOUT; + sbi->ll_gns_timeout = GNS_MOUNT_TIMEOUT; + sbi->ll_gns_timer.data = (unsigned long)sbi; sbi->ll_gns_timer.function = ll_gns_timer_callback; init_timer(&sbi->ll_gns_timer); - INIT_LIST_HEAD(&sbi->ll_gns_sbi_head); - + //audit mask + sbi->ll_audit_mask = AUDIT_OFF; ll_set_sbi(sb, sbi); generate_random_uuid(uuid); @@ -94,7 +120,8 @@ void lustre_free_sbi(struct super_block *sb) int lustre_init_dt_desc(struct ll_sb_info *sbi) { - int valsize, rc; + __u32 valsize; + int rc = 0; ENTRY; valsize = sizeof(sbi->ll_dt_desc); @@ -104,36 +131,55 @@ int lustre_init_dt_desc(struct ll_sb_info *sbi) RETURN(rc); } -int lustre_common_fill_super(struct super_block *sb, char *lmv, char *lov) +static int lustre_connect_mds(struct super_block *sb, char *lmv, + struct obd_connect_data *data, + char *mds_security, int async, int pag) { struct ll_sb_info *sbi = ll_s2sbi(sb); - struct ptlrpc_request *request = NULL; - struct lustre_handle dt_conn = {0, }; struct lustre_handle md_conn = {0, }; - struct inode *root = NULL; - struct obd_device *obd; + struct obd_device *md_obd; struct obd_statfs osfs; - struct lustre_md md; - kdev_t devno; - int err; - ENTRY; - - obd = class_name2obd(lmv); - if (!obd) { + unsigned long sec_flags; + __u32 valsize; + int err = 0; + ENTRY; + + md_obd = class_name2obd(lmv); + if (!md_obd) { CERROR("MDC %s: not setup or attached\n", lmv); RETURN(-EINVAL); } - if (proc_lustre_fs_root) { - err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb, - lov, lmv); - if (err < 0) - CERROR("could not register mount in /proc/lustre"); + obd_set_info(md_obd->obd_self_export, strlen("async"), "async", + sizeof(async), &async); + + if (mds_security == NULL) + mds_security = "null"; + + err = obd_set_info(md_obd->obd_self_export, strlen("sec"), "sec", + strlen(mds_security), mds_security); + + if (err) { + CERROR("LMV %s: failed to set security %s, err %d\n", + lmv, mds_security, err); + RETURN(err); } - err = obd_connect(&md_conn, obd, &sbi->ll_sb_uuid, OBD_OPT_REAL_CLIENT); + if (pag) { + sec_flags = PTLRPC_SEC_FL_PAG; + err = obd_set_info(md_obd->obd_self_export, + strlen("sec_flags"), "sec_flags", + sizeof(sec_flags), &sec_flags); + if (err) { + OBD_FREE(data, sizeof(*data)); + RETURN(err); + } + } + + err = obd_connect(&md_conn, md_obd, &sbi->ll_sb_uuid, data, + OBD_OPT_REAL_CLIENT); if (err == -EBUSY) { - CERROR("An MDS (mdc %s) is performing recovery, of which this" + CERROR("An MDS (lmv %s) is performing recovery, of which this" " client is not a part. Please wait for recovery to " "complete, abort, or time out.\n", lmv); GOTO(out, err); @@ -141,69 +187,153 @@ int lustre_common_fill_super(struct super_block *sb, char *lmv, char *lov) CERROR("cannot connect to %s: rc = %d\n", lmv, err); GOTO(out, err); } + sbi->ll_md_exp = class_conn2export(&md_conn); - err = obd_statfs(obd, &osfs, jiffies - HZ); + + err = obd_statfs(md_obd, &osfs, jiffies - HZ); if (err) - GOTO(out_lmv, err); + GOTO(out_disconnect, err); if (!osfs.os_bsize) { CERROR("Invalid block size is detected."); - GOTO(out_lmv, err); + GOTO(out_disconnect, err); } sb->s_magic = LL_SUPER_MAGIC; sb->s_blocksize = osfs.os_bsize; sb->s_blocksize_bits = log2(osfs.os_bsize); sb->s_maxbytes = PAGE_CACHE_MAXBYTES; - - devno = get_uuid2int(sbi->ll_md_exp->exp_obd->obd_uuid.uuid, - strlen(sbi->ll_md_exp->exp_obd->obd_uuid.uuid)); - sb->s_dev = devno; + /* in 2.6.x FS is not allowed to form s_dev */ +#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0)) + { + kdev_t devno; + + devno = get_uuid2int((char *)sbi->ll_md_exp->exp_obd->obd_uuid.uuid, + strlen((char *)sbi->ll_md_exp->exp_obd->obd_uuid.uuid)); + + sb->s_dev = devno; + } +#endif + + /* after statfs, we are supposed to have connected to MDSs, + * so it's ok to check remote flag returned. + */ + valsize = sizeof(&sbi->ll_remote); + err = obd_get_info(sbi->ll_md_exp, strlen("remote_flag"), "remote_flag", + &valsize, &sbi->ll_remote); + if (err) { + CERROR("fail to obtain remote flag\n"); + GOTO(out_disconnect, err); + } + +out_disconnect: + if (err) + obd_disconnect(sbi->ll_md_exp, 0); +out: + RETURN(err); +} + +static int lustre_connect_ost(struct super_block *sb, char *lov, + struct obd_connect_data *data, + char *oss_security, int async, int pag) +{ + struct ll_sb_info *sbi = ll_s2sbi(sb); + struct lustre_handle dt_conn = {0, }; + struct obd_device *obd = NULL; + unsigned long sec_flags; + int err, mdsize; obd = class_name2obd(lov); if (!obd) { CERROR("OSC %s: not setup or attached\n", lov); - GOTO(out_lmv, err); + GOTO(out, err = -EINVAL); } + obd_set_info(obd->obd_self_export, strlen("async"), "async", + sizeof(async), &async); - err = obd_connect(&dt_conn, obd, &sbi->ll_sb_uuid, OBD_OPT_REAL_CLIENT); + if (oss_security == NULL) + oss_security = "null"; + + err = obd_set_info(obd->obd_self_export, strlen("sec"), "sec", + strlen(oss_security), oss_security); + if (err) { + CERROR("LOV %s: failed to set security %s, err %d\n", + lov, oss_security, err); + RETURN(err); + } + + /* FIXME Because of the async nature of file i/o, we never know + * who is actually dirty the pages; and any process have chance + * to trigger dirty-flushing within its own process context. So + * for simplicity we simply use root's credential, we suppose root + * always have credential. + */ + if (pag) + sec_flags = PTLRPC_SEC_FL_PAG; + else + sec_flags = PTLRPC_SEC_FL_OSS; + + err = obd_set_info(obd->obd_self_export, + strlen("sec_flags"), "sec_flags", + sizeof(sec_flags), &sec_flags); + if (err) { + OBD_FREE(data, sizeof(*data)); + RETURN(err); + } + + err = obd_connect(&dt_conn, obd, &sbi->ll_sb_uuid, data, 0); if (err == -EBUSY) { - CERROR("An OST (osc %s) is performing recovery, of which this" + CERROR("An OST (lov %s) is performing recovery, of which this" " client is not a part. Please wait for recovery to " "complete, abort, or time out.\n", lov); GOTO(out, err); } else if (err) { CERROR("cannot connect to %s: rc = %d\n", lov, err); - GOTO(out_lmv, err); + GOTO(out, err); } sbi->ll_dt_exp = class_conn2export(&dt_conn); err = lustre_init_dt_desc(sbi); - if (err == 0) { - int mdsize = obd_size_diskmd(sbi->ll_dt_exp, NULL); - obd_init_ea_size(sbi->ll_md_exp, mdsize, - sbi->ll_dt_desc.ld_tgt_count * - sizeof(struct llog_cookie)); + + if (err) { + CWARN("init dt_desc error %d \n", err); + GOTO(out, err = 0); } - + mdsize = obd_size_diskmd(sbi->ll_dt_exp, NULL); + obd_init_ea_size(sbi->ll_md_exp, mdsize, sbi->ll_dt_desc.ld_tgt_count * + sizeof(struct llog_cookie)); +out: + RETURN(err); +} + +extern struct dentry_operations ll_d_ops; + +static int lustre_init_root_inode(struct super_block *sb) +{ + struct ll_sb_info *sbi = ll_s2sbi(sb); + struct ptlrpc_request *request = NULL; + struct inode *root = NULL; + struct lustre_md md; + int err = 0; + ENTRY; + err = md_getstatus(sbi->ll_md_exp, &sbi->ll_rootid); if (err) { CERROR("cannot mds_connect: rc = %d\n", err); - GOTO(out_lov, err); + GOTO(out, err); } CDEBUG(D_SUPER, "rootid "DLID4"\n", OLID4(&sbi->ll_rootid)); sb->s_op = &lustre_super_operations; - /* make root inode - * XXX: move this to after cbd setup? */ + /* make root inode */ err = md_getattr(sbi->ll_md_exp, &sbi->ll_rootid, (OBD_MD_FLNOTOBD | OBD_MD_FLBLOCKS | OBD_MD_FID), - 0, &request); + NULL, NULL, 0, 0, NULL, &request); if (err) { CERROR("md_getattr failed for root: rc = %d\n", err); - GOTO(out_lov, err); + GOTO(out, err); } err = mdc_req2lustre_md(sbi->ll_md_exp, request, 0, @@ -211,7 +341,7 @@ int lustre_common_fill_super(struct super_block *sb, char *lmv, char *lov) if (err) { CERROR("failed to understand root inode md: rc = %d\n", err); ptlrpc_req_finished(request); - GOTO(out_lov, err); + GOTO(out, err); } LASSERT(id_ino(&sbi->ll_rootid) != 0); @@ -220,9 +350,75 @@ int lustre_common_fill_super(struct super_block *sb, char *lmv, char *lov) ptlrpc_req_finished(request); if (root == NULL || is_bad_inode(root)) { + if (md.lsm != NULL) + obd_free_memmd(sbi->ll_dt_exp, &md.lsm); + if (md.mea != NULL) + obd_free_memmd(sbi->ll_md_exp, + (struct lov_stripe_md**)&md.mea); CERROR("lustre_lite: bad iget4 for root\n"); GOTO(out_root, err = -EBADF); } + sb->s_root = d_alloc_root(root); + sb->s_root->d_op = &ll_d_ops; +out_root: + if (err) + iput(root); +out: + RETURN(err); +} + +int lustre_common_fill_super(struct super_block *sb, char *lmv, char *lov, + char *gkc, int async, char *mds_security, + char *oss_security, __u32 *nllu, int pag, + __u64 *remote) +{ + struct ll_sb_info *sbi = ll_s2sbi(sb); + struct obd_connect_data *data; + int err; + ENTRY; + + /*process the connect flags*/ + if ((*remote & (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) == + (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) { + CERROR("wrong remote flag "LPX64"\n", *remote); + RETURN(-EINVAL); + } + + OBD_ALLOC(data, sizeof(*data)); + if (!data) + RETURN(-ENOMEM); + + data->ocd_connect_flags |= *remote & (OBD_CONNECT_LOCAL | + OBD_CONNECT_REMOTE); + memcpy(data->ocd_nllu, nllu, sizeof(data->ocd_nllu)); + + if (proc_lustre_fs_root) { + err = lprocfs_register_mountpoint(proc_lustre_fs_root, + sb, lov, lmv); + if (err < 0) + CERROR("could not register mount in /proc/lustre"); + } + + /*connect mds */ + err = lustre_connect_mds(sb, lmv, data, mds_security, async, pag); + if (err) + GOTO(out, err); + + /*connect OST*/ + err = lustre_connect_ost(sb, lov, data, oss_security, async, pag); + if (err) + GOTO(out_lmv, err); + + /*connect GSS*/ + err = lustre_init_crypto(sb, gkc, data, async); + if (err) { + CERROR("Could not connect to GSS err %d\n", err); + err = 0; + } + + err = lustre_init_root_inode(sb); + if (err) + GOTO(out_gks, err); err = ll_close_thread_start(&sbi->ll_lcq); if (err) { @@ -232,32 +428,34 @@ int lustre_common_fill_super(struct super_block *sb, char *lmv, char *lov) ll_gns_add_timer(sbi); - /* making vm readahead 0 for 2.4.x. In the case of 2.6.x, backing dev - info assigned to inode mapping is used for determining maximal - readahead. */ -#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) + /* making vm readahead 0 for 2.4.x. In the case of 2.6.x, + backing dev info assigned to inode mapping is used for + determining maximal readahead. */ +#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \ + !defined(KERNEL_HAS_AS_MAX_READAHEAD) /* bug 2805 - set VM readahead to zero */ vm_max_readahead = vm_min_readahead = 0; #endif - - sb->s_root = d_alloc_root(root); - + sb->s_flags |= MS_POSIXACL; #ifdef S_PDIROPS CWARN("Enabling PDIROPS\n"); sb->s_flags |= S_PDIROPS; #endif - + if (data != NULL) + OBD_FREE(data, sizeof(*data)); RETURN(err); out_root: - if (root) - iput(root); -out_lov: - obd_disconnect(sbi->ll_dt_exp, 0); + if (sb->s_root) + dput(sb->s_root); +out_gks: + lustre_destroy_crypto(sb); out_lmv: obd_disconnect(sbi->ll_md_exp, 0); out: + if (data != NULL) + OBD_FREE(data, sizeof(*data)); lprocfs_unregister_mountpoint(sbi); - return err; + RETURN(err); } void lustre_common_put_super(struct super_block *sb) @@ -267,10 +465,12 @@ void lustre_common_put_super(struct super_block *sb) ENTRY; ll_gns_del_timer(sbi); - ll_close_thread_shutdown(sbi->ll_lcq); + ll_close_thread_stop(sbi->ll_lcq); list_del(&sbi->ll_conn_chain); obd_disconnect(sbi->ll_dt_exp, 0); + + lustre_destroy_crypto(sb); lprocfs_unregister_mountpoint(sbi); if (sbi->ll_proc_root) { @@ -281,14 +481,12 @@ void lustre_common_put_super(struct super_block *sb) obd_disconnect(sbi->ll_md_exp, 0); // We do this to get rid of orphaned dentries. That is not really trw. - spin_lock(&dcache_lock); hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) { struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash); - CWARN("orphan dentry %*s (%p) at unmount\n", - dentry->d_name.len, dentry->d_name.name, dentry); + CWARN("orphan dentry %.*s (%p->%p) at unmount\n", + dentry->d_name.len, dentry->d_name.name, dentry, next); shrink_dcache_parent(dentry); } - spin_unlock(&dcache_lock); EXIT; } @@ -327,7 +525,8 @@ int ll_set_opt(const char *opt, char *data, int fl) RETURN(fl); } -void ll_options(char *options, char **lov, char **lmv, int *flags) +void ll_options(char *options, char **lov, char **lmv, char **gkc, + char **mds_sec, char **oss_sec, int *async, int *flags) { char *this_char; #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0)) @@ -340,6 +539,7 @@ void ll_options(char *options, char **lov, char **lmv, int *flags) return; } + *async = 0; #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0)) for (this_char = strtok (options, ","); this_char != NULL; @@ -352,6 +552,16 @@ void ll_options(char *options, char **lov, char **lmv, int *flags) continue; if (!*lmv && (*lmv = ll_read_opt("mdc", this_char))) continue; + if (!*gkc && (*gkc = ll_read_opt("gkc", this_char))) + continue; + if (!strncmp(this_char, "lasync", strlen("lasync"))) { + *async = 1; + continue; + } + if (!*mds_sec && (*mds_sec = ll_read_opt("mds_sec", this_char))) + continue; + if (!*oss_sec && (*oss_sec = ll_read_opt("oss_sec", this_char))) + continue; if (!(*flags & LL_SBI_NOLCK) && ((*flags) = (*flags) | ll_set_opt("nolock", this_char, @@ -365,20 +575,36 @@ void ll_options(char *options, char **lov, char **lmv, int *flags) void ll_lli_init(struct ll_inode_info *lli) { sema_init(&lli->lli_open_sem, 1); + sema_init(&lli->lli_size_sem, 1); lli->lli_flags = 0; + lli->lli_size_pid = 0; lli->lli_maxbytes = PAGE_CACHE_MAXBYTES; spin_lock_init(&lli->lli_lock); INIT_LIST_HEAD(&lli->lli_pending_write_llaps); + INIT_LIST_HEAD(&lli->lli_close_item); lli->lli_inode_magic = LLI_INODE_MAGIC; memset(&lli->lli_id, 0, sizeof(lli->lli_id)); + sema_init(&lli->lli_och_sem, 1); + lli->lli_mds_read_och = lli->lli_mds_write_och = NULL; + lli->lli_mds_exec_och = NULL; + lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0; + lli->lli_open_fd_exec_count = 0; + lli->lli_audit_mask = AUDIT_OFF; + lli->lli_key_info = NULL; + init_waitqueue_head(&lli->lli_dirty_wait); + lli->lli_io_epoch = 0; + INIT_LIST_HEAD(&lli->lli_capas); } int ll_fill_super(struct super_block *sb, void *data, int silent) { struct ll_sb_info *sbi; - char *lov = NULL; - char *lmv = NULL; - int err; + char *lov = NULL, *lmv = NULL, *gkc = NULL; + char *mds_sec = NULL; + char *oss_sec = NULL; + int async, err; + __u32 nllu[2] = { NOBODY_UID, NOBODY_GID }; + __u64 remote_flag = 0; ENTRY; CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb); @@ -388,19 +614,16 @@ int ll_fill_super(struct super_block *sb, void *data, int silent) RETURN(-ENOMEM); sbi->ll_flags |= LL_SBI_READAHEAD; - ll_options(data, &lov, &lmv, &sbi->ll_flags); - - if (!lov) { - CERROR("no osc\n"); - GOTO(out, err = -EINVAL); - } + ll_options(data, &lov, &lmv, &gkc, &mds_sec, &oss_sec, + &async, &sbi->ll_flags); - if (!lmv) { - CERROR("no mdc\n"); + if (!lov || !lmv) { + CERROR("no osc %p or no mdc %p\n", lov, lmv); GOTO(out, err = -EINVAL); } - - err = lustre_common_fill_super(sb, lmv, lov); + + err = lustre_common_fill_super(sb, lmv, lov, gkc, async, mds_sec, + oss_sec, nllu, 0, &remote_flag); EXIT; out: if (err) @@ -410,13 +633,21 @@ out: OBD_FREE(lmv, strlen(lmv) + 1); if (lov) OBD_FREE(lov, strlen(lov) + 1); + if (mds_sec) + OBD_FREE(mds_sec, strlen(mds_sec) + 1); + if (oss_sec) + OBD_FREE(oss_sec, strlen(oss_sec) + 1); + if (gkc) + OBD_FREE(gkc, strlen(gkc) + 1); + return err; } /* ll_read_super */ static int lustre_process_log(struct lustre_mount_data *lmd, char *profile, struct config_llog_instance *cfg, int allow_recov) { - struct lustre_cfg lcfg; + struct lustre_cfg *lcfg = NULL; + struct lustre_cfg_bufs bufs; struct portals_cfg pcfg; char *peer = "MDS_PEER_UUID"; struct obd_device *obd; @@ -426,8 +657,7 @@ static int lustre_process_log(struct lustre_mount_data *lmd, char *profile, class_uuid_t uuid; struct obd_uuid lmv_uuid; struct llog_ctxt *ctxt; - int rc = 0; - int err; + int rc = 0, err = 0; ENTRY; if (lmd_bad_magic(lmd)) @@ -440,9 +670,9 @@ static int lustre_process_log(struct lustre_mount_data *lmd, char *profile, PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID); pcfg.pcfg_nal = lmd->lmd_nal; pcfg.pcfg_nid = lmd->lmd_local_nid; - err = libcfs_nal_cmd(&pcfg); - if (err <0) - GOTO(out, err); + rc = libcfs_nal_cmd(&pcfg); + if (rc < 0) + GOTO(out, rc); } if (lmd->lmd_nal == SOCKNAL || @@ -455,53 +685,72 @@ static int lustre_process_log(struct lustre_mount_data *lmd, char *profile, pcfg.pcfg_nid = lmd->lmd_server_nid; pcfg.pcfg_id = lmd->lmd_server_ipaddr; pcfg.pcfg_misc = lmd->lmd_port; - err = libcfs_nal_cmd(&pcfg); - if (err <0) - GOTO(out, err); + rc = libcfs_nal_cmd(&pcfg); + if (rc < 0) + GOTO(out, rc); } - - LCFG_INIT(lcfg, LCFG_ADD_UUID, name); - lcfg.lcfg_nid = lmd->lmd_server_nid; - lcfg.lcfg_inllen1 = strlen(peer) + 1; - lcfg.lcfg_inlbuf1 = peer; - lcfg.lcfg_nal = lmd->lmd_nal; - err = class_process_config(&lcfg); + lustre_cfg_bufs_reset(&bufs, name); + lustre_cfg_bufs_set_string(&bufs, 1, peer); + + lcfg = lustre_cfg_new(LCFG_ADD_UUID, &bufs); + lcfg->lcfg_nal = lmd->lmd_nal; + lcfg->lcfg_nid = lmd->lmd_server_nid; + LASSERT(lcfg->lcfg_nal); + LASSERT(lcfg->lcfg_nid); + err = class_process_config(lcfg); + lustre_cfg_free(lcfg); if (err < 0) GOTO(out_del_conn, err); - LCFG_INIT(lcfg, LCFG_ATTACH, name); - lcfg.lcfg_inlbuf1 = "mdc"; - lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1; - lcfg.lcfg_inlbuf2 = lmv_uuid.uuid; - lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1; - err = class_process_config(&lcfg); + lustre_cfg_bufs_reset(&bufs, name); + lustre_cfg_bufs_set_string(&bufs, 1, OBD_MDC_DEVICENAME); + lustre_cfg_bufs_set_string(&bufs, 2, (char *)lmv_uuid.uuid); + + lcfg = lustre_cfg_new(LCFG_ATTACH, &bufs); + err = class_process_config(lcfg); + lustre_cfg_free(lcfg); if (err < 0) GOTO(out_del_uuid, err); - LCFG_INIT(lcfg, LCFG_SETUP, name); - lcfg.lcfg_inlbuf1 = lmd->lmd_mds; - lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1; - lcfg.lcfg_inlbuf2 = peer; - lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1; - err = class_process_config(&lcfg); + lustre_cfg_bufs_reset(&bufs, name); + lustre_cfg_bufs_set_string(&bufs, 1, lmd->lmd_mds); + lustre_cfg_bufs_set_string(&bufs, 2, peer); + + lcfg = lustre_cfg_new(LCFG_SETUP, &bufs); + err = class_process_config(lcfg); + lustre_cfg_free(lcfg); if (err < 0) GOTO(out_detach, err); obd = class_name2obd(name); if (obd == NULL) - GOTO(out_cleanup, err = -EINVAL); + GOTO(out_cleanup, rc = -EINVAL); + + rc = obd_set_info(obd->obd_self_export, strlen("sec"), "sec", + strlen(lmd->lmd_mds_security), lmd->lmd_mds_security); + if (rc) + GOTO(out_cleanup, rc); + + if (lmd->lmd_pag) { + unsigned long sec_flags = PTLRPC_SEC_FL_PAG; + rc = obd_set_info(obd->obd_self_export, + strlen("sec_flags"), "sec_flags", + sizeof(sec_flags), &sec_flags); + if (rc) + GOTO(out_cleanup, rc); + } /* Disable initial recovery on this import */ - err = obd_set_info(obd->obd_self_export, - strlen("initial_recov"), "initial_recov", - sizeof(allow_recov), &allow_recov); - if (err) - GOTO(out_cleanup, err); + rc = obd_set_info(obd->obd_self_export, + strlen("initial_recov"), "initial_recov", + sizeof(allow_recov), &allow_recov); + if (rc) + GOTO(out_cleanup, rc); - err = obd_connect(&md_conn, obd, &lmv_uuid, 0); - if (err) { - CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, err); - GOTO(out_cleanup, err); + rc = obd_connect(&md_conn, obd, &lmv_uuid, NULL, OBD_OPT_REAL_CLIENT); + if (rc) { + CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, rc); + GOTO(out_cleanup, rc); } exp = class_conn2export(&md_conn); @@ -515,22 +764,26 @@ static int lustre_process_log(struct lustre_mount_data *lmd, char *profile, EXIT; out_cleanup: - LCFG_INIT(lcfg, LCFG_CLEANUP, name); - err = class_process_config(&lcfg); + lustre_cfg_bufs_reset(&bufs, name); + lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs); + err = class_process_config(lcfg); + lustre_cfg_free(lcfg); if (err < 0) GOTO(out, err); - out_detach: - LCFG_INIT(lcfg, LCFG_DETACH, name); - err = class_process_config(&lcfg); + lustre_cfg_bufs_reset(&bufs, name); + lcfg = lustre_cfg_new(LCFG_DETACH, &bufs); + err = class_process_config(lcfg); + lustre_cfg_free(lcfg); if (err < 0) GOTO(out, err); out_del_uuid: - LCFG_INIT(lcfg, LCFG_DEL_UUID, name); - lcfg.lcfg_inllen1 = strlen(peer) + 1; - lcfg.lcfg_inlbuf1 = peer; - err = class_process_config(&lcfg); + lustre_cfg_bufs_reset(&bufs, name); + lustre_cfg_bufs_set_string(&bufs, 1, peer); + lcfg = lustre_cfg_new(LCFG_DEL_UUID, &bufs); + err = class_process_config(lcfg); + lustre_cfg_free(lcfg); out_del_conn: if (lmd->lmd_nal == SOCKNAL || @@ -538,12 +791,16 @@ out_del_conn: lmd->lmd_nal == IIBNAL || lmd->lmd_nal == VIBNAL || lmd->lmd_nal == RANAL) { + int err2; + PCFG_INIT(pcfg, NAL_CMD_DEL_PEER); pcfg.pcfg_nal = lmd->lmd_nal; pcfg.pcfg_nid = lmd->lmd_server_nid; pcfg.pcfg_flags = 1; /* single_share */ - err = libcfs_nal_cmd(&pcfg); - if (err <0) + err2 = libcfs_nal_cmd(&pcfg); + if (err2 && !err) + err = err2; + if (err < 0) GOTO(out, err); } out: @@ -553,154 +810,220 @@ out: return rc; } -int lustre_fill_super(struct super_block *sb, void *data, int silent) +static void lustre_manual_cleanup(struct ll_sb_info *sbi) { - struct lustre_mount_data * lmd = data; - char *lov = NULL, *lmv = NULL; - struct ll_sb_info *sbi; - int err; - ENTRY; - - CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb); - if (lmd_bad_magic(lmd)) - RETURN(-EINVAL); - - sbi = lustre_init_sbi(sb); - if (!sbi) - RETURN(-ENOMEM); - - sbi->ll_flags |= LL_SBI_READAHEAD; + struct lustre_cfg *lcfg; + struct lustre_cfg_bufs bufs; + struct obd_device *obd; + int next = 0; - if (lmd->lmd_profile) { - struct lustre_profile *lprof; - struct config_llog_instance cfg; - int len; + while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL) { + int err; - if (lmd->lmd_mds[0] == '\0') { - CERROR("no mds name\n"); - GOTO(out_free, err = -EINVAL); + lustre_cfg_bufs_reset(&bufs, obd->obd_name); + lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs); + err = class_process_config(lcfg); + if (err) { + CERROR("cleanup failed: %s\n", obd->obd_name); + //continue; } - - OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd)); - if (sbi->ll_lmd == NULL) - GOTO(out_free, err = -ENOMEM); - memcpy(sbi->ll_lmd, lmd, sizeof(*lmd)); - - /* generate a string unique to this super, let's try - the address of the super itself.*/ - len = (sizeof(sb) * 2) + 1; - OBD_ALLOC(sbi->ll_instance, len); - if (sbi->ll_instance == NULL) - GOTO(out_free, err = -ENOMEM); - sprintf(sbi->ll_instance, "%p", sb); - - cfg.cfg_instance = sbi->ll_instance; - cfg.cfg_uuid = sbi->ll_sb_uuid; - cfg.cfg_local_nid = lmd->lmd_local_nid; - err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0); - if (err < 0) { - CERROR("Unable to process log: %s\n", lmd->lmd_profile); - GOTO(out_free, err); + + lcfg->lcfg_command = LCFG_DETACH; + err = class_process_config(lcfg); + lustre_cfg_free(lcfg); + if (err) { + CERROR("detach failed: %s\n", obd->obd_name); + //continue; } + } - lprof = class_get_profile(lmd->lmd_profile); - if (lprof == NULL) { - CERROR("No profile found: %s\n", lmd->lmd_profile); - GOTO(out_free, err = -EINVAL); - } - if (lov) - OBD_FREE(lov, strlen(lov) + 1); - OBD_ALLOC(lov, strlen(lprof->lp_lov) + - strlen(sbi->ll_instance) + 2); - sprintf(lov, "%s-%s", lprof->lp_lov, sbi->ll_instance); + if (sbi->ll_lmd != NULL) + class_del_profile(sbi->ll_lmd->lmd_profile); +} - if (lmv) - OBD_FREE(lmv, strlen(lmv) + 1); - OBD_ALLOC(lmv, strlen(lprof->lp_lmv) + - strlen(sbi->ll_instance) + 2); - sprintf(lmv, "%s-%s", lprof->lp_lmv, sbi->ll_instance); - } +static int lustre_process_profile(struct super_block *sb, + struct lustre_mount_data *lmd, + char **lov, char **lmv, char **gkc) +{ + struct ll_sb_info *sbi = ll_s2sbi(sb); + struct config_llog_instance cfg; + struct lustre_profile *lprof; + int len, err = 0; + ENTRY; - if (!lov) { - CERROR("no osc\n"); - GOTO(out_free, err = -EINVAL); + if (!lmd->lmd_profile) + RETURN(0); + + if (lmd->lmd_mds[0] == '\0') { + CERROR("no mds name\n"); + GOTO(out, err = -EINVAL); } + lmd->lmd_mds_security[sizeof(lmd->lmd_mds_security) - 1] = 0; + lmd->lmd_oss_security[sizeof(lmd->lmd_oss_security) - 1] = 0; + + OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd)); + if (sbi->ll_lmd == NULL) + GOTO(out, err = -ENOMEM); + memcpy(sbi->ll_lmd, lmd, sizeof(*lmd)); + + /* generate a string unique to this super, let's try the address of the + * super itself. */ + len = (sizeof(sb) * 2) + 1; + OBD_ALLOC(sbi->ll_instance, len); + if (sbi->ll_instance == NULL) + GOTO(out, err = -ENOMEM); + sprintf(sbi->ll_instance, "%p", sb); - if (!lmv) { - CERROR("no mdc\n"); - GOTO(out_free, err = -EINVAL); + cfg.cfg_instance = sbi->ll_instance; + cfg.cfg_uuid = sbi->ll_sb_uuid; + cfg.cfg_local_nid = lmd->lmd_local_nid; + cfg.cfg_flags |= CFG_MODIFY_UUID_FL; + err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0); + if (err < 0) { + CERROR("Unable to process log: %s\n", lmd->lmd_profile); + GOTO(out, err); } - err = lustre_common_fill_super(sb, lmv, lov); + lprof = class_get_profile(lmd->lmd_profile); + if (lprof == NULL) { + CERROR("No profile found: %s\n", lmd->lmd_profile); + GOTO(out, err = -EINVAL); + } - if (err) - GOTO(out_free, err); + OBD_ALLOC(*lov, strlen(lprof->lp_lov) + + strlen(sbi->ll_instance) + 2); + if (*lov == NULL) + GOTO(out, err = -ENOMEM); + + sprintf(*lov, "%s-%s", lprof->lp_lov, sbi->ll_instance); + + OBD_ALLOC(*lmv, strlen(lprof->lp_lmv) + + strlen(sbi->ll_instance) + 2); + if (*lmv == NULL) + GOTO(out_free_lov, err = -ENOMEM); + + sprintf(*lmv, "%s-%s", lprof->lp_lmv, sbi->ll_instance); + + if (lprof->lp_gkc) { + OBD_ALLOC(*gkc, strlen(lprof->lp_gkc) + + strlen(sbi->ll_instance) + 2); + if (*gkc == NULL) + GOTO(out_free_lmv, err = -ENOMEM); + + sprintf(*gkc, "%s-%s", lprof->lp_gkc, sbi->ll_instance); + } -out_dev: - if (lmv) - OBD_FREE(lmv, strlen(lmv) + 1); - if (lov) - OBD_FREE(lov, strlen(lov) + 1); - RETURN(err); +out_free_lmv: + OBD_FREE(*lmv, strlen(lprof->lp_lmv) + + strlen(sbi->ll_instance) + 2); +out_free_lov: + OBD_FREE(*lov, strlen(lprof->lp_lov) + + strlen(sbi->ll_instance) + 2); +out: + return err; +} -out_free: - if (sbi->ll_lmd) { - int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1; - int err; - - if (sbi->ll_instance != NULL) { - struct lustre_mount_data *lmd = sbi->ll_lmd; - char * cln_prof; - struct config_llog_instance cfg; - - cfg.cfg_instance = sbi->ll_instance; - cfg.cfg_uuid = sbi->ll_sb_uuid; - - OBD_ALLOC(cln_prof, len); - sprintf(cln_prof, "%s-clean", lmd->lmd_profile); +static int lustre_clean_profile(struct ll_sb_info *sbi, int force_umount) +{ + struct lustre_mount_data *lmd = sbi->ll_lmd; + struct config_llog_instance cfg; + char *cl_prof; + int len, err = 0; + ENTRY; + + if (!lmd) + RETURN(err); + + len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean") + 1; + + if (force_umount) { + CERROR("force umount, doing manual cleanup\n"); + lustre_manual_cleanup(sbi); + GOTO(free_lmd, 0); + + } + if (sbi->ll_instance != NULL) { + cfg.cfg_flags |= CFG_MODIFY_UUID_FL; + cfg.cfg_instance = sbi->ll_instance; + cfg.cfg_uuid = sbi->ll_sb_uuid; - err = lustre_process_log(lmd, cln_prof, &cfg, 0); - if (err < 0) - CERROR("Unable to process log: %s\n", cln_prof); - OBD_FREE(cln_prof, len); - OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance)+ 1); + OBD_ALLOC(cl_prof, len); + if (!cl_prof) { + CERROR("can't allocate memory, " + "skipping processing cleanup profile.\n"); + GOTO(free_lmd, err = -ENOMEM); + } + + sprintf(cl_prof, "%s-clean", lmd->lmd_profile); + err = lustre_process_log(lmd, cl_prof, &cfg, 0); + if (err < 0) { + CERROR("Unable to process log: %s\n", cl_prof); + lustre_manual_cleanup(sbi); } - OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd)); + OBD_FREE(cl_prof, len); } - lustre_free_sbi(sb); - - goto out_dev; -} /* lustre_fill_super */ + EXIT; +free_lmd: + if (sbi->ll_instance) + OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1); + OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd)); + return err; +} -static void lustre_manual_cleanup(struct ll_sb_info *sbi) +int lustre_fill_super(struct super_block *sb, void *data, int silent) { - struct lustre_cfg lcfg; - struct obd_device *obd; - int next = 0; + struct lustre_mount_data * lmd = data; + char *lov = NULL, *lmv = NULL, *gkc = NULL; + struct ll_sb_info *sbi; + int err; + ENTRY; - while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL) - { - int err; + CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb); + if (lmd_bad_magic(lmd)) + RETURN(-EINVAL); - LCFG_INIT(lcfg, LCFG_CLEANUP, obd->obd_name); - err = class_process_config(&lcfg); - if (err) { - CERROR("cleanup failed: %s\n", obd->obd_name); - //continue; - } + sbi = lustre_init_sbi(sb); + if (!sbi) + RETURN(-ENOMEM); - LCFG_INIT(lcfg, LCFG_DETACH, obd->obd_name); - err = class_process_config(&lcfg); - if (err) { - CERROR("detach failed: %s\n", obd->obd_name); - //continue; - } + sbi->ll_flags |= LL_SBI_READAHEAD; + + err = lustre_process_profile(sb, lmd, &lov, &lmv, &gkc); + if (err) { + CERROR("Can not process the profile err %d \n", err); + GOTO(out_free, err); + } + if (!lov || !lmv) { + CERROR("no osc %p or no mdc %p \n", lov, lmv); + GOTO(out_free, err = -EINVAL); } - if (sbi->ll_lmd != NULL) - class_del_profile(sbi->ll_lmd->lmd_profile); -} + err = lustre_common_fill_super(sb, lmv, lov, gkc, lmd->lmd_async, + lmd->lmd_mds_security, + lmd->lmd_oss_security, + &lmd->lmd_nllu, lmd->lmd_pag, + &lmd->lmd_remote_flag); + + if (err) + GOTO(out_free, err); + + EXIT; +out_dev: + if (lmv) + OBD_FREE(lmv, strlen(lmv) + 1); + if (lov) + OBD_FREE(lov, strlen(lov) + 1); + if (gkc) + OBD_FREE(gkc, strlen(gkc) + 1); + + return err; +out_free: + lustre_clean_profile(sbi, 0); + lustre_free_sbi(sb); + goto out_dev; + +} /* lustre_fill_super */ void lustre_put_super(struct super_block *sb) { @@ -716,37 +1039,7 @@ void lustre_put_super(struct super_block *sb) obd = NULL; lustre_common_put_super(sb); - if (sbi->ll_lmd != NULL) { - char * cln_prof; - int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1; - int err; - struct config_llog_instance cfg; - - if (force_umount) { - CERROR("force umount, doing manual cleanup\n"); - lustre_manual_cleanup(sbi); - GOTO(free_lmd, 0); - } - - cfg.cfg_instance = sbi->ll_instance; - cfg.cfg_uuid = sbi->ll_sb_uuid; - - OBD_ALLOC(cln_prof, len); - sprintf(cln_prof, "%s-clean", sbi->ll_lmd->lmd_profile); - - err = lustre_process_log(sbi->ll_lmd, cln_prof, &cfg, 0); - if (err < 0) { - CERROR("Unable to process log: %s, doing manual cleanup" - "\n", cln_prof); - lustre_manual_cleanup(sbi); - } - - OBD_FREE(cln_prof, len); - free_lmd: - OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd)); - OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1); - } - + lustre_clean_profile(sbi, force_umount); lustre_free_sbi(sb); EXIT; @@ -754,32 +1047,21 @@ void lustre_put_super(struct super_block *sb) int ll_process_config_update(struct ll_sb_info *sbi, int clean) { - struct obd_export *md_exp = sbi->ll_md_exp; struct lustre_mount_data *lmd = sbi->ll_lmd; - struct llog_ctxt *ctxt; - struct config_llog_instance cfg; char *profile = lmd->lmd_profile, *name = NULL; + struct config_llog_instance cfg; int rc, namelen = 0, version; + struct llog_ctxt *ctxt; ENTRY; if (profile == NULL) RETURN(0); if (lmd == NULL) { - CERROR("Client not mounted with zero-conf; cannot process " - "update log.\n"); + CERROR("Client not mounted with zero-conf; cannot " + "process update log.\n"); RETURN(0); } - rc = ldlm_cli_cancel_unused(md_exp->exp_obd->obd_namespace, NULL, - LDLM_FL_CONFIG_CHANGE, NULL); - if (rc != 0) - CWARN("ldlm_cli_cancel_unused(mdc): %d\n", rc); - - rc = obd_cancel_unused(sbi->ll_dt_exp, NULL, LDLM_FL_CONFIG_CHANGE, - NULL); - if (rc != 0) - CWARN("obd_cancel_unused(lov): %d\n", rc); - cfg.cfg_instance = sbi->ll_instance; cfg.cfg_uuid = sbi->ll_sb_uuid; cfg.cfg_local_nid = lmd->lmd_local_nid; @@ -799,7 +1081,7 @@ int ll_process_config_update(struct ll_sb_info *sbi, int clean) CWARN("Applying configuration log %s\n", name); - ctxt = llog_get_context(&md_exp->exp_obd->obd_llogs, + ctxt = llog_get_context(&sbi->ll_md_exp->exp_obd->obd_llogs, LLOG_CONFIG_REPL_CTXT); rc = class_config_process_llog(ctxt, name, &cfg); if (rc == 0) @@ -808,12 +1090,14 @@ int ll_process_config_update(struct ll_sb_info *sbi, int clean) if (rc == 0 && clean == 0) { struct lov_desc desc; - int rc, valsize; + __u32 valsize; + int rc = 0; + valsize = sizeof(desc); rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1, "lovdesc", &valsize, &desc); - rc = obd_init_ea_size(md_exp, + rc = obd_init_ea_size(sbi->ll_md_exp, obd_size_diskmd(sbi->ll_dt_exp, NULL), (desc.ld_tgt_count * sizeof(struct llog_cookie))); @@ -825,17 +1109,32 @@ int ll_process_config_update(struct ll_sb_info *sbi, int clean) struct inode *ll_inode_from_lock(struct ldlm_lock *lock) { struct inode *inode = NULL; - l_lock(&lock->l_resource->lr_namespace->ns_lock); + + /* NOTE: we depend on atomic igrab() -bzzz */ + lock_res_and_lock(lock); if (lock->l_ast_data) { struct ll_inode_info *lli = ll_i2info(lock->l_ast_data); if (lli->lli_inode_magic == LLI_INODE_MAGIC) { inode = igrab(lock->l_ast_data); } else { - CERROR("DEBUG: l_ast_data %p is bogus: magic %x\n", + struct timeval now; + do_gettimeofday(&now); + inode = lock->l_ast_data; + LDLM_ERROR(lock, "granted at %lu.%lu, now %lu.%lu", + lock->l_enqueued_time.tv_sec, + lock->l_enqueued_time.tv_usec, + now.tv_sec, now.tv_usec); + CDEBUG(inode->i_state & I_FREEING ? D_INFO : D_WARNING, + "l_ast_data %p is bogus: magic %0x8\n", lock->l_ast_data, lli->lli_inode_magic); + CDEBUG(D_ERROR, "i_state = 0x%lx, l_ast_data %p is bogus: magic %0x8\n", + inode->i_state, lock->l_ast_data, lli->lli_inode_magic); + inode = NULL; + unlock_res_and_lock(lock); + LBUG(); } } - l_unlock(&lock->l_resource->lr_namespace->ns_lock); + unlock_res_and_lock(lock); return inode; } @@ -851,22 +1150,34 @@ int null_if_equal(struct ldlm_lock *lock, void *data) return LDLM_ITER_CONTINUE; } +static void remote_acl_free(struct remote_acl *racl); + void ll_clear_inode(struct inode *inode) { struct lustre_id id; struct ll_inode_info *lli = ll_i2info(inode); struct ll_sb_info *sbi = ll_i2sbi(inode); + struct obd_capa *ocapa, *tmp; ENTRY; CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino, inode->i_generation, inode); - lli->lli_inode_magic = LLI_INODE_DEAD; + LASSERT(ll_is_inode_dirty(inode) == 0); ll_inode2id(&id, inode); clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags)); md_change_cbdata(sbi->ll_md_exp, &id, null_if_equal, inode); + LASSERT(!lli->lli_open_fd_write_count); + LASSERT(!lli->lli_open_fd_read_count); + LASSERT(!lli->lli_open_fd_exec_count); + if (lli->lli_mds_write_och) + ll_md_real_close(sbi->ll_md_exp, inode, FMODE_WRITE); + if (lli->lli_mds_exec_och) + ll_md_real_close(sbi->ll_md_exp, inode, FMODE_EXEC); + if (lli->lli_mds_read_och) + ll_md_real_close(sbi->ll_md_exp, inode, FMODE_READ); if (lli->lli_smd) obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd, null_if_equal, inode); @@ -882,12 +1193,33 @@ void ll_clear_inode(struct inode *inode) lli->lli_mea = NULL; } + LASSERT(sbi->ll_crypto_info != NULL); + ll_crypto_destroy_inode_key(inode); + if (lli->lli_symlink_name) { OBD_FREE(lli->lli_symlink_name, strlen(lli->lli_symlink_name) + 1); lli->lli_symlink_name = NULL; } + if (lli->lli_posix_acl) { + LASSERT(lli->lli_remote_acl == NULL); + posix_acl_release(lli->lli_posix_acl); + lli->lli_posix_acl = NULL; + } + + if (lli->lli_remote_acl) { + LASSERT(lli->lli_posix_acl == NULL); + remote_acl_free(lli->lli_remote_acl); + lli->lli_remote_acl = NULL; + } + + list_for_each_entry_safe(ocapa, tmp, &lli->lli_capas, u.client.lli_list) + capa_put(ocapa); + + LASSERT(!mapping_has_pages(inode->i_mapping)); + lli->lli_inode_magic = LLI_INODE_DEAD; + EXIT; } @@ -907,11 +1239,12 @@ void ll_clear_inode(struct inode *inode) int ll_setattr_raw(struct inode *inode, struct iattr *attr) { struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd; + struct ll_inode_info *lli = ll_i2info(inode); struct ll_sb_info *sbi = ll_i2sbi(inode); struct ptlrpc_request *request = NULL; struct mdc_op_data *op_data; int ia_valid = attr->ia_valid; - int rc = 0; + int err, rc = 0; ENTRY; CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino); @@ -951,30 +1284,43 @@ int ll_setattr_raw(struct inode *inode, struct iattr *attr) CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n", LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime), LTIME_S(CURRENT_TIME)); - if (lsm) - attr->ia_valid &= ~ATTR_SIZE; /* If only OST attributes being set on objects, don't do MDS RPC. * In that case, we need to check permissions and update the local * inode ourselves so we can call obdo_from_inode() always. */ - if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN | ATTR_RAW) : ~0)) { + if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN /*| ATTR_RAW*/) : ~0)) { struct lustre_md md; + void *key = NULL; + int key_size = 0; OBD_ALLOC(op_data, sizeof(*op_data)); if (op_data == NULL) RETURN(-ENOMEM); - ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0); - + ll_inode2mdc_data(op_data, inode, (OBD_MD_FLID | OBD_MD_MEA)); + + if (ia_valid & (ATTR_UID | ATTR_GID | ATTR_MODE)) { + rc = ll_crypto_get_mac(inode, attr, NULL, 0, &key, + &key_size); + if (rc) { + CERROR("can not get right mac, rc=%d\n", rc); + if (key && key_size) + OBD_FREE(key, key_size); + RETURN(rc); + } + } rc = md_setattr(sbi->ll_md_exp, op_data, - attr, NULL, 0, NULL, 0, &request); + attr, key, key_size, NULL, 0, NULL, + 0, &request); OBD_FREE(op_data, sizeof(*op_data)); + + if (key && key_size) + OBD_FREE(key, key_size); if (rc) { ptlrpc_req_finished(request); if (rc != -EPERM && rc != -EACCES) CERROR("md_setattr fails: rc = %d\n", rc); RETURN(rc); } - rc = mdc_req2lustre_md(sbi->ll_md_exp, request, 0, sbi->ll_dt_exp, &md); if (rc) { @@ -982,9 +1328,27 @@ int ll_setattr_raw(struct inode *inode, struct iattr *attr) RETURN(rc); } - /* Won't invoke vmtruncate as we already cleared ATTR_SIZE, - * but needed to set timestamps backwards on utime. */ - inode_setattr(inode, attr); + if (attr->ia_valid & ATTR_SIZE) { + rc = ll_set_trunc_capa(request, 0, inode); + if (rc) { + ptlrpc_req_finished(request); + RETURN(rc); + } + } + + /* We call inode_setattr to adjust timestamps, but we first + * clear ATTR_SIZE to avoid invoking vmtruncate. + * + * NB: ATTR_SIZE will only be set at this point if the size + * resides on the MDS, ie, this file has no objects. */ + attr->ia_valid &= ~ATTR_SIZE; + + /* + * assigning inode_setattr() to @err to disable warning that + * function's result should be checked by by caller. error is + * impossible here, as vmtruncate() control path is disabled. + */ + err = inode_setattr(inode, attr); ll_update_inode(inode, &md); ptlrpc_req_finished(request); @@ -1001,7 +1365,7 @@ int ll_setattr_raw(struct inode *inode, struct iattr *attr) /* from sys_utime() */ if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) { if (current->fsuid != inode->i_uid && - (rc=ll_permission(inode,MAY_WRITE,NULL))!=0) + (rc = ll_permission(inode, MAY_WRITE, NULL)) != 0) RETURN(rc); } else { /* from inode_change_ok() */ @@ -1011,8 +1375,16 @@ int ll_setattr_raw(struct inode *inode, struct iattr *attr) } } - /* Won't invoke vmtruncate, as we already cleared ATTR_SIZE */ - inode_setattr(inode, attr); + if (lsm) + attr->ia_valid &= ~ATTR_SIZE; + + /* won't invoke vmtruncate, as we already cleared ATTR_SIZE */ + err = inode_setattr(inode, attr); + /* + * assigning inode_setattr() to @err to disable warning that + * function's result should be checked by by caller. error is + * impossible here, as vmtruncate() control path is disabled. + */ } /* We really need to get our PW lock before we change inode->i_size. @@ -1032,30 +1404,28 @@ int ll_setattr_raw(struct inode *inode, struct iattr *attr) if (attr->ia_size == 0) ast_flags = LDLM_AST_DISCARD_DATA; - /* bug 1639: avoid write/truncate i_sem/DLM deadlock */ - LASSERT(atomic_read(&inode->i_sem.count) <= 0); - up(&inode->i_sem); rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh, ast_flags, &ll_i2sbi(inode)->ll_seek_stime); - down(&inode->i_sem); + if (rc != 0) RETURN(rc); + down(&lli->lli_size_sem); + lli->lli_size_pid = current->pid; rc = vmtruncate(inode, attr->ia_size); + if (rc != 0) { + LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0); + lli->lli_size_pid = 0; + up(&lli->lli_size_sem); + } - /* We need to drop the semaphore here, because this unlock may - * result in a cancellation, which will need the i_sem */ - up(&inode->i_sem); - /* unlock now as we don't mind others file lockers racing with - * the mds updates below? */ err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh); - down(&inode->i_sem); if (err) { CERROR("ll_extent_unlock failed: %d\n", err); if (!rc) rc = err; } - } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) { + } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET | ATTR_UID | ATTR_GID)) { struct obdo *oa = NULL; CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n", @@ -1068,20 +1438,36 @@ int ll_setattr_raw(struct inode *inode, struct iattr *attr) oa->o_id = lsm->lsm_object_id; oa->o_gr = lsm->lsm_object_gr; oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP; + + /* adding uid and gid, needed for quota */ + if (ia_valid & ATTR_UID) { + oa->o_uid = inode->i_uid; + oa->o_valid |= OBD_MD_FLUID; + } + + if (ia_valid & ATTR_GID) { + oa->o_gid = inode->i_gid; + oa->o_valid |= OBD_MD_FLGID; + } + + *(obdo_id(oa)) = lli->lli_id; + oa->o_valid |= OBD_MD_FLIFID; + obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME | OBD_MD_FLMTIME | OBD_MD_FLCTIME); - rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL); + rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL, NULL); obdo_free(oa); if (rc) CERROR("obd_setattr fails: rc = %d\n", rc); } + RETURN(rc); } int ll_setattr(struct dentry *de, struct iattr *attr) { - LBUG(); /* code is unused, but leave this in case of VFS changes */ - RETURN(-ENOSYS); + LASSERT(de->d_inode); + return ll_setattr_raw(de->d_inode, attr); } int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs, @@ -1094,7 +1480,7 @@ int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs, rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age); if (rc) { - CERROR("mdc_statfs fails: rc = %d\n", rc); + CERROR("obd_statfs fails: rc = %d\n", rc); RETURN(rc); } @@ -1164,16 +1550,438 @@ int ll_statfs(struct super_block *sb, struct kstatfs *sfs) return 0; } + +/******************************** + * remote acl * + ********************************/ + +static struct remote_acl *remote_acl_alloc(void) +{ + struct remote_acl *racl; + int i; + + OBD_ALLOC(racl, sizeof(*racl)); + if (!racl) + return NULL; + + spin_lock_init(&racl->ra_lock); + init_MUTEX(&racl->ra_update_sem); + + for (i = 0; i < REMOTE_ACL_HASHSIZE; i++) + INIT_LIST_HEAD(&racl->ra_perm_cache[i]); + + return racl; +} + +/* + * caller should guarantee no race here. + */ +static void remote_perm_flush_xperms(struct lustre_remote_perm *perm) +{ + struct remote_perm_setxid *xperm; + + while (!list_empty(&perm->lrp_setxid_perms)) { + xperm = list_entry(perm->lrp_setxid_perms.next, + struct remote_perm_setxid, + list); + list_del(&xperm->list); + OBD_FREE(xperm, sizeof(*xperm)); + } +} + +/* + * caller should guarantee no race here. + */ +static void remote_acl_flush(struct remote_acl *racl) +{ + struct list_head *head; + struct lustre_remote_perm *perm, *tmp; + int i; + + for (i = 0; i < REMOTE_ACL_HASHSIZE; i++) { + head = &racl->ra_perm_cache[i]; + + list_for_each_entry_safe(perm, tmp, head, lrp_list) { + remote_perm_flush_xperms(perm); + list_del(&perm->lrp_list); + OBD_FREE(perm, sizeof(*perm)); + } + } +} + +static void remote_acl_free(struct remote_acl *racl) +{ + if (!racl) + return; + + down(&racl->ra_update_sem); + spin_lock(&racl->ra_lock); + remote_acl_flush(racl); + spin_unlock(&racl->ra_lock); + up(&racl->ra_update_sem); + + OBD_FREE(racl, sizeof(*racl)); +} + +static inline int remote_acl_hashfunc(__u32 id) +{ + return (id & (REMOTE_ACL_HASHSIZE - 1)); +} + +static +int __remote_acl_check(struct remote_acl *racl, unsigned int *perm) +{ + struct list_head *head; + struct lustre_remote_perm *lperm; + struct remote_perm_setxid *xperm; + int found = 0, rc = -ENOENT; + + LASSERT(racl); + head = &racl->ra_perm_cache[remote_acl_hashfunc(current->uid)]; + spin_lock(&racl->ra_lock); + + list_for_each_entry(lperm, head, lrp_list) { + if (lperm->lrp_auth_uid == current->uid) { + found = 1; + break; + } + } + + if (!found) + goto out; + + if (lperm->lrp_auth_uid == current->fsuid && + lperm->lrp_auth_gid == current->fsgid) { + if (lperm->lrp_valid) { + *perm = lperm->lrp_perm; + rc = 0; + } + goto out; + } else if ((!lperm->lrp_setuid && + lperm->lrp_auth_uid != current->fsuid) || + (!lperm->lrp_setgid && + lperm->lrp_auth_gid != current->fsgid)) { + *perm = 0; + rc = 0; + goto out; + } + + list_for_each_entry(xperm, &lperm->lrp_setxid_perms, list) { + if (xperm->uid == current->fsuid && + xperm->gid == current->fsgid) { + *perm = xperm->perm; + rc = 0; + goto out; + } + } + +out: + spin_unlock(&racl->ra_lock); + return rc; +} + +static +int __remote_acl_update(struct remote_acl *racl, + struct mds_remote_perm *mperm, + struct lustre_remote_perm *lperm, + struct remote_perm_setxid *xperm) +{ + struct list_head *head; + struct lustre_remote_perm *lp; + struct remote_perm_setxid *xp; + int found = 0, setuid = 0, setgid = 0; + + LASSERT(racl); + LASSERT(mperm); + LASSERT(lperm); + LASSERT(current->uid == mperm->mrp_auth_uid); + + if (current->fsuid != mperm->mrp_auth_uid) + setuid = 1; + if (current->fsgid != mperm->mrp_auth_gid) + setgid = 1; + + head = &racl->ra_perm_cache[remote_acl_hashfunc(current->uid)]; + spin_lock(&racl->ra_lock); + + list_for_each_entry(lp, head, lrp_list) { + if (lp->lrp_auth_uid == current->uid) { + found = 1; + break; + } + } + + if (found) { + OBD_FREE(lperm, sizeof(*lperm)); + + if (!lp->lrp_valid && !setuid && !setgid) { + lp->lrp_perm = mperm->mrp_perm; + lp->lrp_valid = 1; + } + + /* sanity check for changes of setxid rules */ + if ((lp->lrp_setuid != 0) != (mperm->mrp_allow_setuid != 0)) { + CWARN("setuid changes: %d => %d\n", + (lp->lrp_setuid != 0), + (mperm->mrp_allow_setuid != 0)); + lp->lrp_setuid = (mperm->mrp_allow_setuid != 0); + } + + if ((lp->lrp_setgid != 0) != (mperm->mrp_allow_setgid != 0)) { + CWARN("setgid changes: %d => %d\n", + (lp->lrp_setgid != 0), + (mperm->mrp_allow_setgid != 0)); + lp->lrp_setgid = (mperm->mrp_allow_setgid != 0); + } + + if (!lp->lrp_setuid && !lp->lrp_setgid && + !list_empty(&lp->lrp_setxid_perms)) { + remote_perm_flush_xperms(lp); + } + } else { + /* initialize lperm and linked into hashtable + */ + INIT_LIST_HEAD(&lperm->lrp_setxid_perms); + lperm->lrp_auth_uid = mperm->mrp_auth_uid; + lperm->lrp_auth_gid = mperm->mrp_auth_gid; + lperm->lrp_setuid = (mperm->mrp_allow_setuid != 0); + lperm->lrp_setgid = (mperm->mrp_allow_setgid != 0); + list_add(&lperm->lrp_list, head); + + if (!setuid && !setgid) { + /* in this case, i'm the authenticated user, + * and mrp_perm is for me. + */ + lperm->lrp_perm = mperm->mrp_perm; + lperm->lrp_valid = 1; + spin_unlock(&racl->ra_lock); + + if (xperm) + OBD_FREE(xperm, sizeof(*xperm)); + return 0; + } + + lp = lperm; + /* fall through */ + } + + LASSERT(lp->lrp_setuid || lp->lrp_setgid || + list_empty(&lp->lrp_setxid_perms)); + + /* if no xperm supplied, we are all done here */ + if (!xperm) { + spin_unlock(&racl->ra_lock); + return 0; + } + + /* whether we allow setuid/setgid */ + if ((!lp->lrp_setuid && setuid) || (!lp->lrp_setgid && setgid)) { + OBD_FREE(xperm, sizeof(*xperm)); + spin_unlock(&racl->ra_lock); + return 0; + } + + /* traverse xperm list */ + list_for_each_entry(xp, &lp->lrp_setxid_perms, list) { + if (xp->uid == current->fsuid && + xp->gid == current->fsgid) { + if (xp->perm != mperm->mrp_perm) { + /* actually this should not happen */ + CWARN("perm changed: %o => %o\n", + xp->perm, mperm->mrp_perm); + xp->perm = mperm->mrp_perm; + } + OBD_FREE(xperm, sizeof(*xperm)); + spin_unlock(&racl->ra_lock); + return 0; + } + } + + /* finally insert this xperm */ + xperm->uid = current->fsuid; + xperm->gid = current->fsgid; + xperm->perm = mperm->mrp_perm; + list_add(&xperm->list, &lp->lrp_setxid_perms); + + spin_unlock(&racl->ra_lock); + return 0; +} + +/* + * remote_acl semaphore must be held by caller + */ +static +int remote_acl_update_locked(struct remote_acl *racl, + struct mds_remote_perm *mperm) +{ + struct lustre_remote_perm *lperm; + struct remote_perm_setxid *xperm; + int setuid = 0, setgid = 0; + + might_sleep(); + + if (current->uid != mperm->mrp_auth_uid) { + CERROR("current uid %u while authenticated as %u\n", + current->uid, mperm->mrp_auth_uid); + return -EINVAL; + } + + if (current->fsuid != mperm->mrp_auth_uid) + setuid = 1; + if (current->fsgid == mperm->mrp_auth_gid) + setgid = 1; + + OBD_ALLOC(lperm, sizeof(*lperm)); + if (!lperm) + return -ENOMEM; + + if ((setuid || setgid) && + !(setuid && !mperm->mrp_allow_setuid) && + !(setgid && !mperm->mrp_allow_setgid)) { + OBD_ALLOC(xperm, sizeof(*xperm)); + if (!xperm) { + OBD_FREE(lperm, sizeof(*lperm)); + return -ENOMEM; + } + } else + xperm = NULL; + + return __remote_acl_update(racl, mperm, lperm, xperm); +} + +/* + * return -EACCES at any error cases + */ +int ll_remote_acl_permission(struct inode *inode, int mode) +{ + struct ll_sb_info *sbi = ll_i2sbi(inode); + struct remote_acl *racl = ll_i2info(inode)->lli_remote_acl; + struct ptlrpc_request *req = NULL; + struct lustre_id id; + struct mds_remote_perm *mperm; + int rc = -EACCES, perm; + + if (!racl) + return -EACCES; + + if (__remote_acl_check(racl, &perm) == 0) { + return ((perm & mode) == mode ? 0 : -EACCES); + } + + might_sleep(); + + /* doing update + */ + down(&racl->ra_update_sem); + + /* we might lose the race when obtain semaphore, + * so check again. + */ + if (__remote_acl_check(racl, &perm) == 0) { + if ((perm & mode) == mode) + rc = 0; + goto out; + } + + /* really fetch from mds + */ + ll_inode2id(&id, inode); + if (md_access_check(sbi->ll_md_exp, &id, &req)) + goto out; + + /* status non-zero indicate there's more apparent error + * detected by mds, e.g. didn't allow this user at all. + * we simply ignore and didn't cache it. + */ + if (req->rq_repmsg->status) + goto out; + + mperm = lustre_swab_repbuf(req, 1, sizeof(*mperm), + lustre_swab_remote_perm); + LASSERT(mperm); + LASSERT_REPSWABBED(req, 1); + + if ((mperm->mrp_perm & mode) == mode) + rc = 0; + + remote_acl_update_locked(racl, mperm); +out: + if (req) + ptlrpc_req_finished(req); + + up(&racl->ra_update_sem); + return rc; +} + +int ll_remote_acl_update(struct inode *inode, struct mds_remote_perm *perm) +{ + struct remote_acl *racl = ll_i2info(inode)->lli_remote_acl; + int rc; + + LASSERT(perm); + + if (!racl) + return -EACCES; + + down(&racl->ra_update_sem); + rc = remote_acl_update_locked(racl, perm); + up(&racl->ra_update_sem); + + return rc; +} + +void ll_inode_invalidate_acl(struct inode *inode) +{ + struct ll_sb_info *sbi = ll_i2sbi(inode); + struct ll_inode_info *lli = ll_i2info(inode); + + if (sbi->ll_remote) { + struct remote_acl *racl = lli->lli_remote_acl; + + LASSERT(!lli->lli_posix_acl); + if (racl) { + down(&racl->ra_update_sem); + spin_lock(&racl->ra_lock); + remote_acl_flush(lli->lli_remote_acl); + spin_unlock(&racl->ra_lock); + up(&racl->ra_update_sem); + } + } else { + /* we can't invalide acl here: suppose we touch a new file + * under a dir, blocking ast on dir will lead to open failure + * on client, although succeed on mds. it's kind of weird, + * the real fix i think is improve client-vfs interaction. + * + * currently we just do nothing here. + */ + return; + + LASSERT(!lli->lli_remote_acl); + spin_lock(&lli->lli_lock); + posix_acl_release(lli->lli_posix_acl); + lli->lli_posix_acl = NULL; + spin_unlock(&lli->lli_lock); + } +} + void ll_update_inode(struct inode *inode, struct lustre_md *md) { struct ll_inode_info *lli = ll_i2info(inode); struct lov_stripe_md *lsm = md->lsm; struct mds_body *body = md->body; struct mea *mea = md->mea; + struct posix_acl *posix_acl = md->posix_acl; + struct ll_sb_info *sbi = ll_i2sbi(inode); + struct lustre_key *mkey = md->key; ENTRY; LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0)); - LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0)); + + if (md->lsm && md->lsm->lsm_magic != LOV_MAGIC) { + /* check for default striping info for dir. */ + LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0)); + } + if (lsm != NULL) { LASSERT(lsm->lsm_object_gr > 0); if (lli->lli_smd == NULL) { @@ -1236,6 +2044,25 @@ void ll_update_inode(struct inode *inode, struct lustre_md *md) if (body->valid & OBD_MD_FLGENER) id_gen(&lli->lli_id) = id_gen(&body->id1); + /* local/remote ACL */ + if (sbi->ll_remote) { + LASSERT(md->posix_acl == NULL); + if (md->remote_perm) { + ll_remote_acl_update(inode, md->remote_perm); + OBD_FREE(md->remote_perm, sizeof(*md->remote_perm)); + md->remote_perm = NULL; + } + } else { + LASSERT(md->remote_perm == NULL); + spin_lock(&lli->lli_lock); + if (posix_acl != NULL) { + if (lli->lli_posix_acl != NULL) + posix_acl_release(lli->lli_posix_acl); + lli->lli_posix_acl = posix_acl; + } + spin_unlock(&lli->lli_lock); + } + if (body->valid & OBD_MD_FLID) inode->i_ino = id_ino(&body->id1); if (body->valid & OBD_MD_FLGENER) @@ -1281,6 +2108,19 @@ void ll_update_inode(struct inode *inode, struct lustre_md *md) if (body->valid & OBD_MD_FLSIZE) set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags); + if (body->valid & OBD_MD_FLAUDIT) { + struct ll_sb_info * sbi = ll_s2sbi(inode->i_sb); + if (IS_AUDIT_OP(body->audit, AUDIT_FS)) + sbi->ll_audit_mask = body->audit; + else + lli->lli_audit_mask = body->audit; + } + + if (mkey != NULL) { + LASSERT(body->valid & OBD_MD_FLKEY); + ll_crypto_init_inode_key(inode, mkey); + } + #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0)) inode->i_dev = (kdev_t)id_group(&lli->lli_id); #endif @@ -1307,6 +2147,11 @@ void ll_read_inode2(struct inode *inode, void *opaque) LASSERT(!lli->lli_smd); + if (ll_i2sbi(inode)->ll_remote) { + lli->lli_remote_acl = remote_acl_alloc(); + /* if failed alloc, nobody will be able to access this inode */ + } + /* Core attributes from the MDS first. This is a new inode, and * the VFS doesn't zero times in the core inode so we have to do * it ourselves. They will be overwritten by either MDS or OST @@ -1401,7 +2246,8 @@ int ll_iocontrol(struct inode *inode, struct file *file, struct mds_body *body; ll_inode2id(&id, inode); - rc = md_getattr(sbi->ll_md_exp, &id, valid, 0, &req); + rc = md_getattr(sbi->ll_md_exp, &id, valid, NULL, NULL, + 0, 0, NULL, &req); if (rc) { CERROR("failure %d inode %lu\n", rc, inode->i_ino); RETURN(-abs(rc)); @@ -1438,14 +2284,14 @@ int ll_iocontrol(struct inode *inode, struct file *file, obdo_free(oa); RETURN(-ENOMEM); } - ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0); + ll_inode2mdc_data(op_data, inode, (OBD_MD_FLID | OBD_MD_MEA)); memset(&attr, 0x0, sizeof(attr)); attr.ia_attr_flags = flags; attr.ia_valid |= ATTR_ATTR_FLAG; rc = md_setattr(sbi->ll_md_exp, op_data, - &attr, NULL, 0, NULL, 0, &req); + &attr, NULL, 0, NULL, 0, NULL, 0, &req); OBD_FREE(op_data, sizeof(*op_data)); if (rc) { ptlrpc_req_finished(req); @@ -1459,9 +2305,11 @@ int ll_iocontrol(struct inode *inode, struct file *file, oa->o_id = lsm->lsm_object_id; oa->o_gr = lsm->lsm_object_gr; oa->o_flags = flags; - oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP; + *(obdo_id(oa)) = ll_i2info(inode)->lli_id; + oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP + | OBD_MD_FLIFID; - rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL); + rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL, NULL); obdo_free(oa); if (rc) { if (rc != -EPERM && rc != -EACCES) @@ -1495,8 +2343,8 @@ int ll_iocontrol(struct inode *inode, struct file *file, void ll_umount_begin(struct super_block *sb) { struct ll_sb_info *sbi = ll_s2sbi(sb); - struct obd_device *obd; struct obd_ioctl_data ioc_data = { 0 }; + struct obd_device *obd; ENTRY; CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb, @@ -1566,6 +2414,25 @@ int ll_prep_inode(struct obd_export *dt_exp, struct obd_export *md_exp, RETURN(rc); } +int ll_show_options(struct seq_file *m, struct vfsmount *mnt) +{ + struct ll_sb_info *sbi = ll_s2sbi(mnt->mnt_sb); + struct lustre_mount_data *lmd = sbi->ll_lmd; + + if (lmd) { + seq_printf(m, ",mds_sec=%s,oss_sec=%s", + lmd->lmd_mds_security, lmd->lmd_oss_security); + } + seq_printf(m, ",%s", sbi->ll_remote ? "remote" : "local"); + if (sbi->ll_remote && lmd) + seq_printf(m, ",nllu=%u:%u", lmd->lmd_nllu, lmd->lmd_nllg); + + if (lmd && lmd->lmd_pag) + seq_printf(m, ",pag"); + + return 0; +} + int ll_get_fid(struct obd_export *exp, struct lustre_id *idp, char *filename, struct lustre_id *ret) { @@ -1590,3 +2457,29 @@ int ll_get_fid(struct obd_export *exp, struct lustre_id *idp, return rc; } +int ll_flush_cred(struct inode *inode) +{ + struct ll_sb_info *sbi = ll_i2sbi(inode); + int rc = 0; + + /* XXX to avoid adding api, we simply use set_info() interface + * to notify underlying obds. set_info() is more like a ioctl() now... + */ + if (sbi->ll_md_exp) { + rc = obd_set_info(sbi->ll_md_exp, + strlen("flush_cred"), "flush_cred", + 0, NULL); + if (rc) + return rc; + } + + if (sbi->ll_dt_exp) { + rc = obd_set_info(sbi->ll_dt_exp, + strlen("flush_cred"), "flush_cred", + 0, NULL); + if (rc) + return rc; + } + + return rc; +}