diff options
Diffstat (limited to 'usr/src/uts/common/io/dls')
| -rw-r--r-- | usr/src/uts/common/io/dls/dls.c | 59 | ||||
| -rw-r--r-- | usr/src/uts/common/io/dls/dls_link.c | 1 | ||||
| -rw-r--r-- | usr/src/uts/common/io/dls/dls_mgmt.c | 423 | ||||
| -rw-r--r-- | usr/src/uts/common/io/dls/dls_stat.c | 172 |
4 files changed, 524 insertions, 131 deletions
diff --git a/usr/src/uts/common/io/dls/dls.c b/usr/src/uts/common/io/dls/dls.c index b26637203f..3fa65ef35d 100644 --- a/usr/src/uts/common/io/dls/dls.c +++ b/usr/src/uts/common/io/dls/dls.c @@ -250,7 +250,7 @@ dls_promisc(dld_str_t *dsp, uint32_t new_flags) { int err = 0; uint32_t old_flags = dsp->ds_promisc; - const uint32_t option_flags = DLS_PROMISC_RX_ONLY; + const uint32_t option_flags = DLS_PROMISC_RX_ONLY | DLS_PROMISC_FIXUPS; uint32_t old_type = old_flags & ~option_flags; uint32_t new_type = new_flags & ~option_flags; mac_client_promisc_type_t mptype = MAC_CLIENT_PROMISC_ALL; @@ -274,6 +274,8 @@ dls_promisc(dld_str_t *dsp, uint32_t new_flags) */ if (new_flags & DLS_PROMISC_RX_ONLY) mac_flags |= MAC_PROMISC_FLAGS_NO_TX_LOOP; + if (new_flags & DLS_PROMISC_FIXUPS) + mac_flags |= MAC_PROMISC_FLAGS_DO_FIXUPS; if (new_type == DLS_PROMISC_SAP) mac_flags |= MAC_PROMISC_FLAGS_NO_PHYS; @@ -643,6 +645,22 @@ boolean_t dls_accept_promisc(dld_str_t *dsp, mac_header_info_t *mhip, dls_rx_t *ds_rx, void **ds_rx_arg, boolean_t loopback) { + if (dsp->ds_promisc == 0) { + /* + * If there are active walkers of the mi_promisc_list when + * promiscuousness is disabled, ds_promisc will be cleared, + * but the DLS will remain on the mi_promisc_list until the + * walk is completed. If we do not recognize this case here, + * we won't properly execute the ds_promisc case in the common + * accept routine -- and we will potentially accept a packet + * that has originated with this DLS (which in turn can + * induce recursion and death by stack overflow). If + * ds_promisc is zero, we know that we are in this window -- + * and we refuse to accept the packet. + */ + return (B_FALSE); + } + return (dls_accept_common(dsp, mhip, ds_rx, ds_rx_arg, B_TRUE, loopback)); } @@ -673,7 +691,10 @@ dls_mac_active_set(dls_link_t *dlp) * Set the function to start receiving packets. */ mac_rx_set(dlp->dl_mch, i_dls_link_rx, dlp); + } else if (dlp->dl_exclusive == B_TRUE) { + return (EBUSY); } + dlp->dl_nactive++; return (0); } @@ -699,7 +720,11 @@ dls_active_set(dld_str_t *dsp) if (dsp->ds_passivestate == DLD_PASSIVE) return (0); - /* If we're already active, then there's nothing more to do. */ + if (dsp->ds_dlp->dl_exclusive == B_TRUE && + dsp->ds_passivestate != DLD_EXCLUSIVE) + return (EBUSY); + + /* If we're already active, we need to check the link's exclusivity */ if ((dsp->ds_nactive == 0) && ((err = dls_mac_active_set(dsp->ds_dlp)) != 0)) { /* except for ENXIO all other errors are mapped to EBUSY */ @@ -708,7 +733,8 @@ dls_active_set(dld_str_t *dsp) return (err); } - dsp->ds_passivestate = DLD_ACTIVE; + dsp->ds_passivestate = dsp->ds_dlp->dl_exclusive == B_TRUE ? + DLD_EXCLUSIVE : DLD_ACTIVE; dsp->ds_nactive++; return (0); } @@ -739,7 +765,32 @@ dls_active_clear(dld_str_t *dsp, boolean_t all) if (dsp->ds_nactive != 0) return; - ASSERT(dsp->ds_passivestate == DLD_ACTIVE); + ASSERT(dsp->ds_passivestate == DLD_ACTIVE || + dsp->ds_passivestate == DLD_EXCLUSIVE); dls_mac_active_clear(dsp->ds_dlp); + /* + * We verify below to ensure that no other part of DLS has mucked with + * our exclusive state. + */ + if (dsp->ds_passivestate == DLD_EXCLUSIVE) + VERIFY(dls_exclusive_set(dsp, B_FALSE) == 0); dsp->ds_passivestate = DLD_UNINITIALIZED; } + +int +dls_exclusive_set(dld_str_t *dsp, boolean_t enable) +{ + ASSERT(MAC_PERIM_HELD(dsp->ds_mh)); + + if (enable == B_FALSE) { + dsp->ds_dlp->dl_exclusive = B_FALSE; + return (0); + } + + if (dsp->ds_dlp->dl_nactive != 0) + return (EBUSY); + + dsp->ds_dlp->dl_exclusive = B_TRUE; + + return (0); +} diff --git a/usr/src/uts/common/io/dls/dls_link.c b/usr/src/uts/common/io/dls/dls_link.c index 4099d0b801..eee3569b10 100644 --- a/usr/src/uts/common/io/dls/dls_link.c +++ b/usr/src/uts/common/io/dls/dls_link.c @@ -686,6 +686,7 @@ i_dls_link_destroy(dls_link_t *dlp) dlp->dl_mnh = NULL; dlp->dl_unknowns = 0; dlp->dl_nonip_cnt = 0; + dlp->dl_exclusive = B_FALSE; kmem_cache_free(i_dls_link_cachep, dlp); } diff --git a/usr/src/uts/common/io/dls/dls_mgmt.c b/usr/src/uts/common/io/dls/dls_mgmt.c index 49e867a19e..90b65ab36a 100644 --- a/usr/src/uts/common/io/dls/dls_mgmt.c +++ b/usr/src/uts/common/io/dls/dls_mgmt.c @@ -21,6 +21,7 @@ /* * Copyright 2009 Sun Microsystems, Inc. All rights reserved. * Use is subject to license terms. + * Copyright 2019 Joyent, Inc. */ /* * Copyright (c) 2016 by Delphix. All rights reserved. @@ -85,6 +86,14 @@ static door_handle_t dls_mgmt_dh = NULL; /* dls_devnet_t dd_flags */ #define DD_CONDEMNED 0x1 #define DD_IMPLICIT_IPTUN 0x2 /* Implicitly-created ip*.*tun* tunnel */ +#define DD_INITIALIZING 0x4 + +/* + * If the link is marked as initializing or condemned then it should + * not be visible outside of the DLS framework. + */ +#define DD_NOT_VISIBLE(flags) ( \ + (flags & (DD_CONDEMNED | DD_INITIALIZING)) != 0) /* * This structure is used to keep the <linkid, macname> mapping. @@ -108,13 +117,14 @@ typedef struct dls_devnet_s { zoneid_t dd_zid; /* current zone */ boolean_t dd_prop_loaded; taskqid_t dd_prop_taskid; + boolean_t dd_transient; /* link goes away when zone does */ } dls_devnet_t; static int i_dls_devnet_create_iptun(const char *, const char *, datalink_id_t *); static int i_dls_devnet_destroy_iptun(datalink_id_t); -static int i_dls_devnet_setzid(dls_devnet_t *, zoneid_t, boolean_t); -static int dls_devnet_unset(const char *, datalink_id_t *, boolean_t); +static int i_dls_devnet_setzid(dls_devnet_t *, zoneid_t, boolean_t, boolean_t); +static int dls_devnet_unset(mac_handle_t, datalink_id_t *, boolean_t); /*ARGSUSED*/ static int @@ -134,9 +144,9 @@ i_dls_devnet_destructor(void *buf, void *arg) { dls_devnet_t *ddp = buf; - ASSERT(ddp->dd_ksp == NULL); - ASSERT(ddp->dd_ref == 0); - ASSERT(ddp->dd_tref == 0); + VERIFY(ddp->dd_ksp == NULL); + VERIFY(ddp->dd_ref == 0); + VERIFY(ddp->dd_tref == 0); mutex_destroy(&ddp->dd_mutex); cv_destroy(&ddp->dd_cv); } @@ -148,7 +158,12 @@ dls_zone_remove(datalink_id_t linkid, void *arg) dls_devnet_t *ddp; if (dls_devnet_hold_tmp(linkid, &ddp) == 0) { - (void) dls_devnet_setzid(ddp, GLOBAL_ZONEID); + /* + * Don't bother moving transient links back to the global zone + * since we will simply delete them in dls_devnet_unset. + */ + if (!ddp->dd_transient) + (void) dls_devnet_setzid(ddp, GLOBAL_ZONEID, B_FALSE); dls_devnet_rele_tmp(ddp); } return (0); @@ -529,6 +544,7 @@ dls_mgmt_get_linkid(const char *link, datalink_id_t *linkid) getlinkid.ld_cmd = DLMGMT_CMD_GETLINKID; (void) strlcpy(getlinkid.ld_link, link, MAXLINKNAMELEN); + getlinkid.ld_zoneid = getzoneid(); if ((err = i_dls_mgmt_upcall(&getlinkid, sizeof (getlinkid), &retval, sizeof (retval))) == 0) { @@ -537,6 +553,27 @@ dls_mgmt_get_linkid(const char *link, datalink_id_t *linkid) return (err); } +int +dls_mgmt_get_linkid_in_zone(const char *link, datalink_id_t *linkid, + zoneid_t zid) +{ + dlmgmt_door_getlinkid_t getlinkid; + dlmgmt_getlinkid_retval_t retval; + int err; + + ASSERT(getzoneid() == GLOBAL_ZONEID || zid == getzoneid()); + getlinkid.ld_cmd = DLMGMT_CMD_GETLINKID; + (void) strlcpy(getlinkid.ld_link, link, MAXLINKNAMELEN); + getlinkid.ld_zoneid = zid; + + if ((err = i_dls_mgmt_upcall(&getlinkid, sizeof (getlinkid), &retval, + sizeof (retval))) == 0) { + *linkid = retval.lr_linkid; + } + return (err); +} + + datalink_id_t dls_mgmt_get_next(datalink_id_t linkid, datalink_class_t class, datalink_media_t dmedia, uint32_t flags) @@ -736,13 +773,24 @@ dls_devnet_stat_update(kstat_t *ksp, int rw) * Create the "link" kstats. */ static void -dls_devnet_stat_create(dls_devnet_t *ddp, zoneid_t zoneid) +dls_devnet_stat_create(dls_devnet_t *ddp, zoneid_t zoneid, zoneid_t newzoneid) { kstat_t *ksp; + char *nm; + char kname[MAXLINKNAMELEN]; + + if (zoneid != newzoneid) { + ASSERT(zoneid == GLOBAL_ZONEID); + (void) snprintf(kname, sizeof (kname), "z%d_%s", newzoneid, + ddp->dd_linkname); + nm = kname; + } else { + nm = ddp->dd_linkname; + } - if (dls_stat_create("link", 0, ddp->dd_linkname, zoneid, + if (dls_stat_create("link", 0, nm, zoneid, dls_devnet_stat_update, (void *)(uintptr_t)ddp->dd_linkid, - &ksp) == 0) { + &ksp, newzoneid) == 0) { ASSERT(ksp != NULL); if (zoneid == ddp->dd_owner_zid) { ASSERT(ddp->dd_ksp == NULL); @@ -762,12 +810,12 @@ dls_devnet_stat_destroy(dls_devnet_t *ddp, zoneid_t zoneid) { if (zoneid == ddp->dd_owner_zid) { if (ddp->dd_ksp != NULL) { - kstat_delete(ddp->dd_ksp); + dls_stat_delete(ddp->dd_ksp); ddp->dd_ksp = NULL; } } else { if (ddp->dd_zone_ksp != NULL) { - kstat_delete(ddp->dd_zone_ksp); + dls_stat_delete(ddp->dd_zone_ksp); ddp->dd_zone_ksp = NULL; } } @@ -778,24 +826,38 @@ dls_devnet_stat_destroy(dls_devnet_t *ddp, zoneid_t zoneid) * and create the new set using the new name. */ static void -dls_devnet_stat_rename(dls_devnet_t *ddp) +dls_devnet_stat_rename(dls_devnet_t *ddp, boolean_t zoneinit) { if (ddp->dd_ksp != NULL) { - kstat_delete(ddp->dd_ksp); + dls_stat_delete(ddp->dd_ksp); ddp->dd_ksp = NULL; } - /* We can't rename a link while it's assigned to a non-global zone. */ + if (zoneinit && ddp->dd_zone_ksp != NULL) { + dls_stat_delete(ddp->dd_zone_ksp); + ddp->dd_zone_ksp = NULL; + } + /* + * We can't rename a link while it's assigned to a non-global zone + * unless we're first initializing the zone while readying it. + */ ASSERT(ddp->dd_zone_ksp == NULL); - dls_devnet_stat_create(ddp, ddp->dd_owner_zid); + dls_devnet_stat_create(ddp, ddp->dd_owner_zid, + (zoneinit ? ddp->dd_zid : ddp->dd_owner_zid)); + if (zoneinit) + dls_devnet_stat_create(ddp, ddp->dd_zid, ddp->dd_zid); } /* - * Associate a linkid with a given link (identified by macname) + * Associate the linkid with the link identified by macname. If this + * is called on behalf of a physical link then linkid may be + * DATALINK_INVALID_LINKID. Otherwise, if called on behalf of a + * virtual link, linkid must have a value. */ static int -dls_devnet_set(const char *macname, datalink_id_t linkid, zoneid_t zoneid, +dls_devnet_set(mac_handle_t mh, datalink_id_t linkid, zoneid_t zoneid, dls_devnet_t **ddpp) { + const char *macname = mac_name(mh); dls_devnet_t *ddp = NULL; datalink_class_t class; int err; @@ -828,17 +890,41 @@ dls_devnet_set(const char *macname, datalink_id_t linkid, zoneid_t zoneid, } /* - * This might be a physical link that has already - * been created, but which does not have a linkid - * because dlmgmtd was not running when it was created. + * If we arrive here we know we are attempting to set + * the linkid on a physical link. A virtual link + * should never arrive here because it should never + * call this function without a linkid. Virtual links + * are created through dlgmtmd and thus we know + * dlmgmtd is alive to assign it a linkid (search for + * uses of dladm_create_datalink_id() to prove this to + * yourself); we don't have the same guarantee for a + * physical link which may perform an upcall for a + * linkid while dlmgmtd is down but will continue + * creating a devnet without the linkid (see + * softmac_create_datalink() to see how physical link + * creation works). That is why there is no entry in + * the id hash but there is one in the macname hash -- + * softmac couldn't acquire a linkid the first time it + * called this function. + * + * Because of the check above, we also know that + * ddp->dd_linkid is not set. Following this, the link + * must still be in the DD_INITIALIZING state because + * that flag is removed IFF dd_linkid is set. This is + * why we can ASSERT the DD_INITIALIZING flag below if + * the call to i_dls_devnet_setzid() fails. */ if (linkid == DATALINK_INVALID_LINKID || class != DATALINK_CLASS_PHYS) { err = EINVAL; goto done; } + + ASSERT(ddp->dd_flags & DD_INITIALIZING); + } else { ddp = kmem_cache_alloc(i_dls_devnet_cachep, KM_SLEEP); + ddp->dd_flags = DD_INITIALIZING; ddp->dd_tref = 0; ddp->dd_ref++; ddp->dd_owner_zid = zoneid; @@ -875,8 +961,19 @@ done: rw_exit(&i_dls_devnet_lock); if (err == 0) { if (zoneid != GLOBAL_ZONEID && - (err = i_dls_devnet_setzid(ddp, zoneid, B_FALSE)) != 0) - (void) dls_devnet_unset(macname, &linkid, B_TRUE); + (err = i_dls_devnet_setzid(ddp, zoneid, B_FALSE, + B_FALSE)) != 0) { + /* + * At this point the link is marked as + * DD_INITIALIZING -- there can be no + * outstanding temp refs and therefore no need + * to wait for them. + */ + ASSERT(ddp->dd_flags & DD_INITIALIZING); + (void) dls_devnet_unset(mh, &linkid, B_FALSE); + return (err); + } + /* * The kstat subsystem holds its own locks (rather perimeter) * before calling the ks_update (dls_devnet_stat_update) entry @@ -884,20 +981,35 @@ done: * lock hierarchy is kstat locks -> i_dls_devnet_lock. */ if (stat_create) - dls_devnet_stat_create(ddp, zoneid); + dls_devnet_stat_create(ddp, zoneid, zoneid); if (ddpp != NULL) *ddpp = ddp; + + mutex_enter(&ddp->dd_mutex); + if (linkid != DATALINK_INVALID_LINKID && !ddp->dd_prop_loaded && + ddp->dd_prop_taskid == TASKQID_INVALID) { + ddp->dd_prop_taskid = taskq_dispatch(system_taskq, + dls_devnet_prop_task, ddp, TQ_SLEEP); + } + mutex_exit(&ddp->dd_mutex); + } return (err); } /* - * Disassociate a linkid with a given link (identified by macname) - * This waits until temporary references to the dls_devnet_t are gone. + * Disassociate the linkid from the link identified by macname. If + * wait is B_TRUE, wait until all temporary refs are released and the + * prop task is finished. + * + * If waiting then you SHOULD NOT call this from inside the MAC perim + * as deadlock will ensue. Otherwise, this function is safe to call + * from inside or outside the MAC perim. */ static int -dls_devnet_unset(const char *macname, datalink_id_t *id, boolean_t wait) +dls_devnet_unset(mac_handle_t mh, datalink_id_t *id, boolean_t wait) { + const char *macname = mac_name(mh); dls_devnet_t *ddp; int err; mod_hash_val_t val; @@ -918,21 +1030,62 @@ dls_devnet_unset(const char *macname, datalink_id_t *id, boolean_t wait) * deadlock. Return EBUSY if the asynchronous thread started for * property loading as part of the post attach hasn't yet completed. */ - ASSERT(ddp->dd_ref != 0); + VERIFY(ddp->dd_ref != 0); if ((ddp->dd_ref != 1) || (!wait && (ddp->dd_tref != 0 || ddp->dd_prop_taskid != 0))) { - mutex_exit(&ddp->dd_mutex); - rw_exit(&i_dls_devnet_lock); - return (EBUSY); + int zstatus = 0; + + /* + * There are a couple of alternatives that might be going on + * here; a) the zone is shutting down and it has a transient + * link assigned, in which case we want to clean it up instead + * of moving it back to the global zone, or b) its possible + * that we're trying to clean up an orphaned vnic that was + * delegated to a zone and which wasn't cleaned up properly + * when the zone went away. Check for either of these cases + * before we simply return EBUSY. + * + * zstatus indicates which situation we are dealing with: + * 0 - means return EBUSY + * 1 - means case (a), cleanup transient link + * -1 - means case (b), orphained VNIC + */ + if (ddp->dd_ref > 1 && ddp->dd_zid != GLOBAL_ZONEID) { + zone_t *zp; + + if ((zp = zone_find_by_id(ddp->dd_zid)) == NULL) { + zstatus = -1; + } else { + if (ddp->dd_transient) { + zone_status_t s = zone_status_get(zp); + + if (s >= ZONE_IS_SHUTTING_DOWN) + zstatus = 1; + } + zone_rele(zp); + } + } + + if (zstatus == 0) { + mutex_exit(&ddp->dd_mutex); + rw_exit(&i_dls_devnet_lock); + return (EBUSY); + } + + /* + * We want to delete the link, reset ref to 1; + */ + if (zstatus == -1) + /* Log a warning, but continue in this case */ + cmn_err(CE_WARN, "clear orphaned datalink: %s\n", + ddp->dd_linkname); + ddp->dd_ref = 1; } ddp->dd_flags |= DD_CONDEMNED; ddp->dd_ref--; *id = ddp->dd_linkid; - if (ddp->dd_zid != GLOBAL_ZONEID) - (void) i_dls_devnet_setzid(ddp, GLOBAL_ZONEID, B_FALSE); - /* * Remove this dls_devnet_t from the hash table. */ @@ -947,19 +1100,40 @@ dls_devnet_unset(const char *macname, datalink_id_t *id, boolean_t wait) } rw_exit(&i_dls_devnet_lock); + /* + * It is important to call i_dls_devnet_setzid() WITHOUT the + * i_dls_devnet_lock held. The setzid call grabs the MAC + * perim; thus causing DLS -> MAC lock ordering if performed + * with the i_dls_devnet_lock held. This forces consumers to + * grab the MAC perim before calling dls_devnet_unset() (the + * locking rules state MAC -> DLS order). By performing the + * setzid outside of the i_dls_devnet_lock consumers can + * safely call dls_devnet_unset() outside the MAC perim. + */ + if (ddp->dd_zid != GLOBAL_ZONEID) { + dls_devnet_stat_destroy(ddp, ddp->dd_zid); + (void) i_dls_devnet_setzid(ddp, GLOBAL_ZONEID, B_FALSE, + B_FALSE); + } + if (wait) { /* * Wait until all temporary references are released. + * The holders of the tref need the MAC perim to + * perform their work and release the tref. To avoid + * deadlock, assert that the perim is never held here. */ + ASSERT0(MAC_PERIM_HELD(mh)); while ((ddp->dd_tref != 0) || (ddp->dd_prop_taskid != 0)) cv_wait(&ddp->dd_cv, &ddp->dd_mutex); } else { - ASSERT(ddp->dd_tref == 0 && - ddp->dd_prop_taskid == (taskqid_t)NULL); + VERIFY(ddp->dd_tref == 0); + VERIFY(ddp->dd_prop_taskid == (taskqid_t)NULL); } - if (ddp->dd_linkid != DATALINK_INVALID_LINKID) + if (ddp->dd_linkid != DATALINK_INVALID_LINKID) { dls_devnet_stat_destroy(ddp, ddp->dd_owner_zid); + } ddp->dd_prop_loaded = B_FALSE; ddp->dd_linkid = DATALINK_INVALID_LINKID; @@ -1019,8 +1193,8 @@ dls_devnet_hold_common(datalink_id_t linkid, dls_devnet_t **ddpp, } mutex_enter(&ddp->dd_mutex); - ASSERT(ddp->dd_ref > 0); - if (ddp->dd_flags & DD_CONDEMNED) { + VERIFY(ddp->dd_ref > 0); + if (DD_NOT_VISIBLE(ddp->dd_flags)) { mutex_exit(&ddp->dd_mutex); rw_exit(&i_dls_devnet_lock); return (ENOENT); @@ -1087,8 +1261,8 @@ dls_devnet_hold_by_dev(dev_t dev, dls_dl_handle_t *ddhp) return (ENOENT); } mutex_enter(&ddp->dd_mutex); - ASSERT(ddp->dd_ref > 0); - if (ddp->dd_flags & DD_CONDEMNED) { + VERIFY(ddp->dd_ref > 0); + if (DD_NOT_VISIBLE(ddp->dd_flags)) { mutex_exit(&ddp->dd_mutex); rw_exit(&i_dls_devnet_lock); return (ENOENT); @@ -1105,7 +1279,7 @@ void dls_devnet_rele(dls_devnet_t *ddp) { mutex_enter(&ddp->dd_mutex); - ASSERT(ddp->dd_ref > 1); + VERIFY(ddp->dd_ref > 1); ddp->dd_ref--; if ((ddp->dd_flags & DD_IMPLICIT_IPTUN) && ddp->dd_ref == 1) { mutex_exit(&ddp->dd_mutex); @@ -1117,7 +1291,7 @@ dls_devnet_rele(dls_devnet_t *ddp) } static int -dls_devnet_hold_by_name(const char *link, dls_devnet_t **ddpp) +dls_devnet_hold_by_name(const char *link, dls_devnet_t **ddpp, zoneid_t zid) { char drv[MAXLINKNAMELEN]; uint_t ppa; @@ -1127,7 +1301,7 @@ dls_devnet_hold_by_name(const char *link, dls_devnet_t **ddpp) dls_dev_handle_t ddh; int err; - if ((err = dls_mgmt_get_linkid(link, &linkid)) == 0) + if ((err = dls_mgmt_get_linkid_in_zone(link, &linkid, zid)) == 0) return (dls_devnet_hold(linkid, ddpp)); /* @@ -1270,9 +1444,15 @@ dls_devnet_phydev(datalink_id_t vlanid, dev_t *devp) * * This case does not change the <link name, linkid> mapping, so the link's * kstats need to be updated with using name associated the given id2. + * + * The zoneinit parameter is used to allow us to create a VNIC in the global + * zone which is assigned to a non-global zone. Since there is a race condition + * in the create process if two VNICs have the same name, we need to rename it + * after it has been assigned to the zone. */ int -dls_devnet_rename(datalink_id_t id1, datalink_id_t id2, const char *link) +dls_devnet_rename(datalink_id_t id1, datalink_id_t id2, const char *link, + boolean_t zoneinit) { dls_dev_handle_t ddh = NULL; int err = 0; @@ -1317,10 +1497,12 @@ dls_devnet_rename(datalink_id_t id1, datalink_id_t id2, const char *link) } mutex_enter(&ddp->dd_mutex); - if (ddp->dd_ref > 1) { - mutex_exit(&ddp->dd_mutex); - err = EBUSY; - goto done; + if (!zoneinit) { + if (ddp->dd_ref > 1) { + mutex_exit(&ddp->dd_mutex); + err = EBUSY; + goto done; + } } mutex_exit(&ddp->dd_mutex); @@ -1331,7 +1513,15 @@ dls_devnet_rename(datalink_id_t id1, datalink_id_t id2, const char *link) /* rename mac client name and its flow if exists */ if ((err = mac_open(ddp->dd_mac, &mh)) != 0) goto done; - (void) mac_rename_primary(mh, link); + if (zoneinit) { + char tname[MAXLINKNAMELEN]; + + (void) snprintf(tname, sizeof (tname), "z%d_%s", + ddp->dd_zid, link); + (void) mac_rename_primary(mh, tname); + } else { + (void) mac_rename_primary(mh, link); + } mac_close(mh); goto done; } @@ -1398,7 +1588,7 @@ done: rw_exit(&i_dls_devnet_lock); if (err == 0) - dls_devnet_stat_rename(ddp); + dls_devnet_stat_rename(ddp, zoneinit); if (mph != NULL) mac_perim_exit(mph); @@ -1407,7 +1597,8 @@ done: } static int -i_dls_devnet_setzid(dls_devnet_t *ddp, zoneid_t new_zoneid, boolean_t setprop) +i_dls_devnet_setzid(dls_devnet_t *ddp, zoneid_t new_zoneid, boolean_t setprop, + boolean_t transient) { int err; mac_perim_handle_t mph; @@ -1436,10 +1627,18 @@ i_dls_devnet_setzid(dls_devnet_t *ddp, zoneid_t new_zoneid, boolean_t setprop) sizeof (retval)); if (err != 0) goto done; + + /* + * We set upcall_done only if the upcall is + * successful. This way, if dls_link_setzid() fails, + * we know another upcall must be done to reset the + * dlmgmtd state. + */ upcall_done = B_TRUE; } if ((err = dls_link_setzid(ddp->dd_mac, new_zoneid)) == 0) { ddp->dd_zid = new_zoneid; + ddp->dd_transient = transient; devnet_need_rebuild = B_TRUE; } @@ -1454,7 +1653,7 @@ done: } int -dls_devnet_setzid(dls_dl_handle_t ddh, zoneid_t new_zid) +dls_devnet_setzid(dls_dl_handle_t ddh, zoneid_t new_zid, boolean_t transient) { dls_devnet_t *ddp; int err; @@ -1476,7 +1675,7 @@ dls_devnet_setzid(dls_dl_handle_t ddh, zoneid_t new_zid) refheld = B_TRUE; } - if ((err = i_dls_devnet_setzid(ddh, new_zid, B_TRUE)) != 0) { + if ((err = i_dls_devnet_setzid(ddh, new_zid, B_TRUE, transient)) != 0) { if (refheld) dls_devnet_rele(ddp); return (err); @@ -1493,7 +1692,7 @@ dls_devnet_setzid(dls_dl_handle_t ddh, zoneid_t new_zid) if (old_zid != GLOBAL_ZONEID) dls_devnet_stat_destroy(ddh, old_zid); if (new_zid != GLOBAL_ZONEID) - dls_devnet_stat_create(ddh, new_zid); + dls_devnet_stat_create(ddh, new_zid, new_zid); return (0); } @@ -1531,15 +1730,19 @@ dls_devnet_islinkvisible(datalink_id_t linkid, zoneid_t zoneid) * Access a vanity naming node. */ int -dls_devnet_open(const char *link, dls_dl_handle_t *dhp, dev_t *devp) +dls_devnet_open_in_zone(const char *link, dls_dl_handle_t *dhp, dev_t *devp, + zoneid_t zid) { dls_devnet_t *ddp; dls_link_t *dlp; - zoneid_t zid = getzoneid(); + zoneid_t czid = getzoneid(); int err; mac_perim_handle_t mph; - if ((err = dls_devnet_hold_by_name(link, &ddp)) != 0) + if (czid != GLOBAL_ZONEID && czid != zid) + return (ENOENT); + + if ((err = dls_devnet_hold_by_name(link, &ddp, zid)) != 0) return (err); dls_devnet_prop_task_wait(ddp); @@ -1572,6 +1775,12 @@ dls_devnet_open(const char *link, dls_dl_handle_t *dhp, dev_t *devp) return (0); } +int +dls_devnet_open(const char *link, dls_dl_handle_t *dhp, dev_t *devp) +{ + return (dls_devnet_open_in_zone(link, dhp, devp, getzoneid())); +} + /* * Close access to a vanity naming node. */ @@ -1628,13 +1837,32 @@ dls_devnet_create(mac_handle_t mh, datalink_id_t linkid, zoneid_t zoneid) * we need to use the linkid to get the user name for the link * when we create the MAC client. */ - if ((err = dls_devnet_set(mac_name(mh), linkid, zoneid, &ddp)) == 0) { + if ((err = dls_devnet_set(mh, linkid, zoneid, &ddp)) == 0) { if ((err = dls_link_hold_create(mac_name(mh), &dlp)) != 0) { mac_perim_exit(mph); - (void) dls_devnet_unset(mac_name(mh), &linkid, B_TRUE); + (void) dls_devnet_unset(mh, &linkid, B_FALSE); return (err); } + + /* + * If dd_linkid is set then the link was successfully + * initialized. In this case we can remove the + * initializing flag and make the link visible to the + * rest of the system. + * + * If not set then we were called by softmac and it + * was unable to obtain a linkid for the physical link + * because dlmgmtd is down. In that case softmac will + * eventually obtain a linkid and call + * dls_devnet_recreate() to complete initialization. + */ + mutex_enter(&ddp->dd_mutex); + if (ddp->dd_linkid != DATALINK_INVALID_LINKID) + ddp->dd_flags &= ~DD_INITIALIZING; + mutex_exit(&ddp->dd_mutex); + } + mac_perim_exit(mph); return (err); } @@ -1648,8 +1876,19 @@ dls_devnet_create(mac_handle_t mh, datalink_id_t linkid, zoneid_t zoneid) int dls_devnet_recreate(mac_handle_t mh, datalink_id_t linkid) { - ASSERT(linkid != DATALINK_INVALID_LINKID); - return (dls_devnet_set(mac_name(mh), linkid, GLOBAL_ZONEID, NULL)); + dls_devnet_t *ddp; + int err; + + VERIFY(linkid != DATALINK_INVALID_LINKID); + if ((err = dls_devnet_set(mh, linkid, GLOBAL_ZONEID, &ddp)) == 0) { + mutex_enter(&ddp->dd_mutex); + if (ddp->dd_linkid != DATALINK_INVALID_LINKID) + ddp->dd_flags &= ~DD_INITIALIZING; + mutex_exit(&ddp->dd_mutex); + } + + return (err); + } int @@ -1659,15 +1898,52 @@ dls_devnet_destroy(mac_handle_t mh, datalink_id_t *idp, boolean_t wait) mac_perim_handle_t mph; *idp = DATALINK_INVALID_LINKID; - err = dls_devnet_unset(mac_name(mh), idp, wait); - if (err != 0 && err != ENOENT) + err = dls_devnet_unset(mh, idp, wait); + + /* + * We continue on in the face of ENOENT because the devnet + * unset and DLS link release are not atomic and we may have a + * scenario where there is no entry in i_dls_devnet_hash for + * the MAC name but there is an entry in i_dls_link_hash. For + * example, if the following occurred: + * + * 1. dls_devnet_unset() returns success, and + * + * 2. dls_link_rele_by_name() fails with ENOTEMPTY because + * flows still exist, and + * + * 3. dls_devnet_set() fails to set the zone id and calls + * dls_devnet_unset() -- leaving an entry in + * i_dls_link_hash but no corresponding entry in + * i_dls_devnet_hash. + * + * Even if #3 wasn't true the dls_devnet_set() may fail for + * different reasons in the future; the point is that it _can_ + * fail as part of its contract. We can't rely on it working + * so we must assume that these two pieces of state (devnet + * and link hashes), which should always be in sync, can get + * out of sync and thus even if we get ENOENT from the devnet + * hash we should still try to delete from the link hash just + * in case. + * + * We could prevent the ENOTEMPTY from dls_link_rele_by_name() + * by calling mac_disable() before calling + * dls_devnet_destroy() but that's not currently possible due + * to a long-standing bug. OpenSolaris 6791335: The semantics + * of mac_disable() were modified by Crossbow such that + * dls_devnet_destroy() needs to be called before + * mac_disable() can succeed. This is because of the implicit + * reference that dls has on the mac_impl_t. + */ + if (err != 0 && err != ENOENT) { return (err); + } mac_perim_enter_by_mh(mh, &mph); err = dls_link_rele_by_name(mac_name(mh)); - mac_perim_exit(mph); - if (err != 0) { + dls_devnet_t *ddp; + /* * XXX It is a general GLDv3 bug that dls_devnet_set() has to * be called to re-set the link when destroy fails. The @@ -1675,9 +1951,22 @@ dls_devnet_destroy(mac_handle_t mh, datalink_id_t *idp, boolean_t wait) * called from kernel context or from a zone other than that * which initially created the link. */ - (void) dls_devnet_set(mac_name(mh), *idp, crgetzoneid(CRED()), - NULL); + (void) dls_devnet_set(mh, *idp, crgetzoneid(CRED()), &ddp); + + /* + * You might think dd_linkid should always be set + * here, but in the case where dls_devnet_unset() + * returns ENOENT it will be DATALINK_INVALID_LINKID. + * Stay consistent with the rest of DLS and only + * remove the initializing flag if linkid is set. + */ + mutex_enter(&ddp->dd_mutex); + if (ddp->dd_linkid != DATALINK_INVALID_LINKID) + ddp->dd_flags &= ~DD_INITIALIZING; + mutex_exit(&ddp->dd_mutex); } + + mac_perim_exit(mph); return (err); } diff --git a/usr/src/uts/common/io/dls/dls_stat.c b/usr/src/uts/common/io/dls/dls_stat.c index 51e4be7260..82dceff278 100644 --- a/usr/src/uts/common/io/dls/dls_stat.c +++ b/usr/src/uts/common/io/dls/dls_stat.c @@ -21,6 +21,7 @@ /* * Copyright 2009 Sun Microsystems, Inc. All rights reserved. * Use is subject to license terms. + * Copyright 2011 Joyent, Inc. All rights reserved. */ /* @@ -30,30 +31,33 @@ #include <sys/dld_impl.h> #include <sys/mac_ether.h> -static mac_stat_info_t i_dls_si[] = { - { MAC_STAT_IFSPEED, "ifspeed", KSTAT_DATA_UINT64, 0 }, - { MAC_STAT_MULTIRCV, "multircv", KSTAT_DATA_UINT32, 0 }, - { MAC_STAT_BRDCSTRCV, "brdcstrcv", KSTAT_DATA_UINT32, 0 }, - { MAC_STAT_MULTIXMT, "multixmt", KSTAT_DATA_UINT32, 0 }, - { MAC_STAT_BRDCSTXMT, "brdcstxmt", KSTAT_DATA_UINT32, 0 }, - { MAC_STAT_NORCVBUF, "norcvbuf", KSTAT_DATA_UINT32, 0 }, - { MAC_STAT_IERRORS, "ierrors", KSTAT_DATA_UINT32, 0 }, - { MAC_STAT_NOXMTBUF, "noxmtbuf", KSTAT_DATA_UINT32, 0 }, - { MAC_STAT_OERRORS, "oerrors", KSTAT_DATA_UINT32, 0 }, - { MAC_STAT_COLLISIONS, "collisions", KSTAT_DATA_UINT32, 0 }, - { MAC_STAT_RBYTES, "rbytes", KSTAT_DATA_UINT32, 0 }, - { MAC_STAT_IPACKETS, "ipackets", KSTAT_DATA_UINT32, 0 }, - { MAC_STAT_OBYTES, "obytes", KSTAT_DATA_UINT32, 0 }, - { MAC_STAT_OPACKETS, "opackets", KSTAT_DATA_UINT32, 0 }, - { MAC_STAT_RBYTES, "rbytes64", KSTAT_DATA_UINT64, 0 }, - { MAC_STAT_IPACKETS, "ipackets64", KSTAT_DATA_UINT64, 0 }, - { MAC_STAT_OBYTES, "obytes64", KSTAT_DATA_UINT64, 0 }, - { MAC_STAT_OPACKETS, "opackets64", KSTAT_DATA_UINT64, 0 }, - { MAC_STAT_LINK_STATE, "link_state", KSTAT_DATA_UINT32, - (uint64_t)LINK_STATE_UNKNOWN} -}; - -#define STAT_INFO_COUNT (sizeof (i_dls_si) / sizeof (i_dls_si[0])) +/* + * structure for link kstats + */ +typedef struct { + kstat_named_t dk_ifspeed; + kstat_named_t dk_multircv; + kstat_named_t dk_brdcstrcv; + kstat_named_t dk_multixmt; + kstat_named_t dk_brdcstxmt; + kstat_named_t dk_norcvbuf; + kstat_named_t dk_ierrors; + kstat_named_t dk_noxmtbuf; + kstat_named_t dk_oerrors; + kstat_named_t dk_collisions; + kstat_named_t dk_rbytes; + kstat_named_t dk_ipackets; + kstat_named_t dk_obytes; + kstat_named_t dk_opackets; + kstat_named_t dk_rbytes64; + kstat_named_t dk_ipackets64; + kstat_named_t dk_obytes64; + kstat_named_t dk_opackets64; + kstat_named_t dk_link_state; + kstat_named_t dk_link_duplex; + kstat_named_t dk_unknowns; + kstat_named_t dk_zonename; +} dls_kstat_t; /* * Exported functions. @@ -61,42 +65,54 @@ static mac_stat_info_t i_dls_si[] = { int dls_stat_update(kstat_t *ksp, dls_link_t *dlp, int rw) { - kstat_named_t *knp; - uint_t i; - uint64_t val; + dls_kstat_t *dkp = ksp->ks_data; if (rw != KSTAT_READ) return (EACCES); - knp = (kstat_named_t *)ksp->ks_data; - for (i = 0; i < STAT_INFO_COUNT; i++) { - val = mac_stat_get(dlp->dl_mh, i_dls_si[i].msi_stat); - - switch (i_dls_si[i].msi_type) { - case KSTAT_DATA_UINT64: - knp->value.ui64 = val; - break; - case KSTAT_DATA_UINT32: - knp->value.ui32 = (uint32_t)val; - break; - default: - ASSERT(B_FALSE); - } - - knp++; - } + dkp->dk_ifspeed.value.ui64 = mac_stat_get(dlp->dl_mh, MAC_STAT_IFSPEED); + dkp->dk_multircv.value.ui32 = mac_stat_get(dlp->dl_mh, + MAC_STAT_MULTIRCV); + dkp->dk_brdcstrcv.value.ui32 = mac_stat_get(dlp->dl_mh, + MAC_STAT_BRDCSTRCV); + dkp->dk_multixmt.value.ui32 = mac_stat_get(dlp->dl_mh, + MAC_STAT_MULTIXMT); + dkp->dk_brdcstxmt.value.ui32 = mac_stat_get(dlp->dl_mh, + MAC_STAT_BRDCSTXMT); + dkp->dk_norcvbuf.value.ui32 = mac_stat_get(dlp->dl_mh, + MAC_STAT_NORCVBUF); + dkp->dk_ierrors.value.ui32 = mac_stat_get(dlp->dl_mh, MAC_STAT_IERRORS); + dkp->dk_noxmtbuf.value.ui32 = mac_stat_get(dlp->dl_mh, + MAC_STAT_NOXMTBUF); + dkp->dk_oerrors.value.ui32 = mac_stat_get(dlp->dl_mh, MAC_STAT_OERRORS); + dkp->dk_collisions.value.ui32 = mac_stat_get(dlp->dl_mh, + MAC_STAT_COLLISIONS); + dkp->dk_rbytes.value.ui32 = mac_stat_get(dlp->dl_mh, MAC_STAT_RBYTES); + dkp->dk_ipackets.value.ui32 = mac_stat_get(dlp->dl_mh, + MAC_STAT_IPACKETS); + dkp->dk_obytes.value.ui32 = mac_stat_get(dlp->dl_mh, MAC_STAT_OBYTES); + dkp->dk_opackets.value.ui32 = mac_stat_get(dlp->dl_mh, + MAC_STAT_OPACKETS); + dkp->dk_rbytes64.value.ui64 = mac_stat_get(dlp->dl_mh, MAC_STAT_RBYTES); + dkp->dk_ipackets64.value.ui64 = mac_stat_get(dlp->dl_mh, + MAC_STAT_IPACKETS); + dkp->dk_obytes64.value.ui64 = mac_stat_get(dlp->dl_mh, MAC_STAT_OBYTES); + dkp->dk_opackets64.value.ui64 = mac_stat_get(dlp->dl_mh, + MAC_STAT_OPACKETS); + dkp->dk_link_state.value.ui32 = mac_stat_get(dlp->dl_mh, + MAC_STAT_LINK_STATE); /* * Ethernet specific kstat "link_duplex" */ if (dlp->dl_mip->mi_nativemedia != DL_ETHER) { - knp->value.ui32 = LINK_DUPLEX_UNKNOWN; + dkp->dk_link_duplex.value.ui32 = LINK_DUPLEX_UNKNOWN; } else { - val = mac_stat_get(dlp->dl_mh, ETHER_STAT_LINK_DUPLEX); - knp->value.ui32 = (uint32_t)val; + dkp->dk_link_duplex.value.ui32 = + (uint32_t)mac_stat_get(dlp->dl_mh, ETHER_STAT_LINK_DUPLEX); } - knp++; - knp->value.ui32 = dlp->dl_unknowns; + + dkp->dk_unknowns.value.ui32 = dlp->dl_unknowns; return (0); } @@ -104,30 +120,66 @@ dls_stat_update(kstat_t *ksp, dls_link_t *dlp, int rw) int dls_stat_create(const char *module, int instance, const char *name, zoneid_t zoneid, int (*update)(struct kstat *, int), void *private, - kstat_t **kspp) + kstat_t **kspp, zoneid_t newzoneid) { kstat_t *ksp; - kstat_named_t *knp; - uint_t i; + zone_t *zone; + dls_kstat_t *dkp; if ((ksp = kstat_create_zone(module, instance, name, "net", - KSTAT_TYPE_NAMED, STAT_INFO_COUNT + 2, 0, zoneid)) == NULL) { + KSTAT_TYPE_NAMED, sizeof (dls_kstat_t) / sizeof (kstat_named_t), + KSTAT_FLAG_VIRTUAL, zoneid)) == NULL) { return (EINVAL); } ksp->ks_update = update; ksp->ks_private = private; + dkp = ksp->ks_data = kmem_zalloc(sizeof (dls_kstat_t), KM_SLEEP); + if ((zone = zone_find_by_id(newzoneid)) != NULL) { + ksp->ks_data_size += strlen(zone->zone_name) + 1; + } - knp = (kstat_named_t *)ksp->ks_data; - for (i = 0; i < STAT_INFO_COUNT; i++) { - kstat_named_init(knp, i_dls_si[i].msi_name, - i_dls_si[i].msi_type); - knp++; + kstat_named_init(&dkp->dk_ifspeed, "ifspeed", KSTAT_DATA_UINT64); + kstat_named_init(&dkp->dk_multircv, "multircv", KSTAT_DATA_UINT32); + kstat_named_init(&dkp->dk_brdcstrcv, "brdcstrcv", KSTAT_DATA_UINT32); + kstat_named_init(&dkp->dk_multixmt, "multixmt", KSTAT_DATA_UINT32); + kstat_named_init(&dkp->dk_brdcstxmt, "brdcstxmt", KSTAT_DATA_UINT32); + kstat_named_init(&dkp->dk_norcvbuf, "norcvbuf", KSTAT_DATA_UINT32); + kstat_named_init(&dkp->dk_ierrors, "ierrors", KSTAT_DATA_UINT32); + kstat_named_init(&dkp->dk_noxmtbuf, "noxmtbuf", KSTAT_DATA_UINT32); + kstat_named_init(&dkp->dk_oerrors, "oerrors", KSTAT_DATA_UINT32); + kstat_named_init(&dkp->dk_collisions, "collisions", KSTAT_DATA_UINT32); + kstat_named_init(&dkp->dk_rbytes, "rbytes", KSTAT_DATA_UINT32); + kstat_named_init(&dkp->dk_ipackets, "ipackets", KSTAT_DATA_UINT32); + kstat_named_init(&dkp->dk_obytes, "obytes", KSTAT_DATA_UINT32); + kstat_named_init(&dkp->dk_opackets, "opackets", KSTAT_DATA_UINT32); + kstat_named_init(&dkp->dk_rbytes64, "rbytes64", KSTAT_DATA_UINT64); + kstat_named_init(&dkp->dk_ipackets64, "ipackets64", KSTAT_DATA_UINT64); + kstat_named_init(&dkp->dk_obytes64, "obytes64", KSTAT_DATA_UINT64); + kstat_named_init(&dkp->dk_opackets64, "opackets64", KSTAT_DATA_UINT64); + kstat_named_init(&dkp->dk_link_state, "link_state", KSTAT_DATA_UINT32); + kstat_named_init(&dkp->dk_link_duplex, "link_duplex", + KSTAT_DATA_UINT32); + kstat_named_init(&dkp->dk_unknowns, "unknowns", KSTAT_DATA_UINT32); + kstat_named_init(&dkp->dk_zonename, "zonename", KSTAT_DATA_STRING); + + if (zone != NULL) { + kstat_named_setstr(&dkp->dk_zonename, zone->zone_name); + zone_rele(zone); } - kstat_named_init(knp++, "link_duplex", KSTAT_DATA_UINT32); - kstat_named_init(knp, "unknowns", KSTAT_DATA_UINT32); kstat_install(ksp); *kspp = ksp; return (0); } + +void +dls_stat_delete(kstat_t *ksp) +{ + void *data; + if (ksp != NULL) { + data = ksp->ks_data; + kstat_delete(ksp); + kmem_free(data, sizeof (dls_kstat_t)); + } +} |