nvme: In-kernel TLS support for TCP
Subsystems (6)
v8 cover nvme: In-kernel TLS support for TCP
Hi all,
with the merge of Chuck Levers handshake upcall mechanism and my tls_read_sock() implementation finally merged with net-next it's now time to restart on the actual issue, namely implementing in-kernel TLS support for nvme-tcp.
The patchset is based on the recent net-next git tree; everything after commit ba4a734e1aa0 ("net/tls: avoid TCP window full during ->read_sock()") should work.
You might want to add the patch
'nvmet-tcp: use 'spin_lock_bh' for state_lock()' before applying this patchset; otherwise results might be ... interesting.
It also requires the 'tlshd' userspace daemon
(https://github.com/oracle/ktls-utils)
for the actual TLS handshake.
Changes for nvme-cli are already included in the upstream repository.
Theory of operation:
A dedicated '.nvme' keyring is created to hold the pre-shared keys (PSKs) for the TLS handshake. Keys will have to be provisioned before TLS handshake is attempted; that can be done with the 'nvme gen-tls-key' command for nvme-cli (patches are already merged upstream).
After connection to the remote TCP port the client side will use the 'best' PSK (as inferred from the NVMe TCP spec) or the PSK specified by the '--tls_key' option to nvme-cli and call the TLS userspace daemon to initiate a TLS handshake.
The server side will then invoke the TLS userspace daemon to run the TLS handshake.
If the TLS handshake succeeds the userspace daemon will be activating kTLS on the socket, and control is passed back to the kernel.
This implementation currently does not implement the so-called 'secure concatenation' mode from NVMe-TCP; a TPAR is still pending with fixes for it, so I'll wait until it's published before posting patches for that.
Patchset can be found at:
git.kernel.org/pub/scm/linux/kernel/git/hare/nvme.git branch tls.v12
For testing I'm using this script, running on a nvme target with NQN 'nqn.test' and using 127.0.0.1 as a port:
modprobe nvmet-tcp
nvmetcli restore
modprobe nvme-tcp
./nvme gen-tls-key --subsysnqn=nqn.test -i
./nvme gen-tls-key --subsysnqn=nqn.2014-08.org.nvmexpress.discovery -i tlshd -c /etc/tlshd.conf
and then one can do a simple:
# nvme connect -t tcp -a 127.0.0.1 -s 4420 -n nqn.test --tls to start the connection.
As usual, comments and reviews are welcome.
Changes to v7:
- Include reviews from Simon
- Do not call sock_release() when sock_alloc_file() fails
Changes to v6:
- More reviews from Sagi
- Fixup non-tls connections
- Fixup nvme options handling
- Add reference counting to nvmet-tcp queues
Changes to v5:
- Include reviews from Sagi
- Split off nvmet tsas/treq handling
- Sanitize sock_file handling
Changes to v4:
- Split off network patches into a separate
patchset
- Handle TLS Alert notifications
Changes to v3:
- Really handle MSG_EOR for TLS
- Fixup MSG_SENDPAGE_NOTLAST handling
- Conditionally disable fabric option
Changes to v2:
- Included reviews from Sagi
- Removed MSG_SENDPAGE_NOTLAST
- Improved MSG_EOR handling for TLS
- Add config options NVME_TCP_TLS
and NVME_TARGET_TCP_TLS
Changes to the original RFC:
- Add a CONFIG_NVME_TLS config option
- Use a single PSK for the TLS handshake
- Make TLS connections mandatory
- Do not peek messages for the server
- Simplify data_ready callback
- Implement read_sock() for TLS
Hannes Reinecke (17):
nvme-keyring: register '.nvme' keyring
nvme-keyring: define a 'psk' keytype
nvme: add TCP TSAS definitions
nvme-tcp: add definitions for TLS cipher suites
nvme-keyring: implement nvme_tls_psk_default() security/keys: export key_lookup()
nvme-tcp: allocate socket file
nvme-tcp: enable TLS handshake upcall
nvme-tcp: control message handling for recvmsg()
nvme-fabrics: parse options 'keyring' and 'tls_key'
nvmet: make TCP sectype settable via configfs
nvmet-tcp: make nvmet_tcp_alloc_queue() a void function
nvmet-tcp: allocate socket file
nvmet-tcp: reference counting for queues
nvmet: Set 'TREQ' to 'required' when TLS is enabled
nvmet-tcp: enable TLS handshake upcall
nvmet-tcp: control messages for recvmsg()
drivers/nvme/common/Kconfig | 4 + drivers/nvme/common/Makefile | 3 +- drivers/nvme/common/keyring.c | 182 ++++++++++++++++++ drivers/nvme/host/Kconfig | 15 ++ drivers/nvme/host/core.c | 12 +- drivers/nvme/host/fabrics.c | 61 +++++- drivers/nvme/host/fabrics.h | 9 + drivers/nvme/host/nvme.h | 1 + drivers/nvme/host/sysfs.c | 21 +++ drivers/nvme/host/tcp.c | 172 ++++++++++++++++- drivers/nvme/target/Kconfig | 15 ++ drivers/nvme/target/configfs.c | 108 ++++++++++- drivers/nvme/target/nvmet.h | 1 + drivers/nvme/target/tcp.c | 335 ++++++++++++++++++++++++++++----- include/linux/nvme-keyring.h | 36 ++++ include/linux/nvme-tcp.h | 6 + include/linux/nvme.h | 10 + security/keys/key.c | 1 + 18 files changed, 935 insertions(+), 57 deletions(-) create mode 100644 drivers/nvme/common/keyring.c create mode 100644 include/linux/nvme-keyring.h
Signature · 1 line
2.35.3
PATCH v8 1/17 nvme-keyring: register '.nvme' keyring
Register a '.nvme' keyring to hold keys for TLS and DH-HMAC-CHAP and add a new config option NVME_KEYRING.
We need a separate keyring for NVMe as the configuration is done via individual commands (eg for configfs), and the usual per-session or per-process keyrings can't be used.
--- drivers/nvme/common/Kconfig | 4 ++++ drivers/nvme/common/Makefile | 3 ++- drivers/nvme/common/keyring.c | 40 +++++++++++++++++++++++++++++++++++ drivers/nvme/host/core.c | 10 +++++++-- include/linux/nvme-keyring.h | 28 ++++++++++++++++++++++++ 5 files changed, 82 insertions(+), 3 deletions(-) create mode 100644 drivers/nvme/common/keyring.c create mode 100644 include/linux/nvme-keyring.h
drivers/nvme/common/Kconfig +4 -0
@@ -2,3 +2,7 @@ config NVME_COMMON tristate + +config NVME_KEYRING + bool + select KEYS
drivers/nvme/common/Makefile +2 -1
@@ -4,4 +4,5 @@ ccflags-y += -I$(src) obj-$(CONFIG_NVME_COMMON) += nvme-common.o -nvme-common-y += auth.o +nvme-common-$(CONFIG_NVME_AUTH) += auth.o +nvme-common-$(CONFIG_NVME_KEYRING) += keyring.o
drivers/nvme/common/keyring.c new file +40 -0
@@ -0,0 +1,40 @@ +// SPDX-License-Identifier: GPL-2.0 +/* +* Copyright (c) 2023 Hannes Reinecke, SUSE Labs +*/ + +#include <linux/module.h> +#include <linux/seq_file.h> +#include <linux/key-type.h> +#include <keys/user-type.h> +#include <linux/nvme.h> + +static struct key *nvme_keyring; + +key_serial_t nvme_keyring_id (void ) +{ +return nvme_keyring->serial; +} +EXPORT_SYMBOL_GPL (nvme_keyring_id); + +int nvme_keyring_init (void ) +{ + nvme_keyring =keyring_alloc (".nvme" , +GLOBAL_ROOT_UID ,GLOBAL_ROOT_GID , +current_cred (), + (KEY_POS_ALL & ~KEY_POS_SETATTR ) | + (KEY_USR_ALL & ~KEY_USR_SETATTR ), +KEY_ALLOC_NOT_IN_QUOTA ,NULL ,NULL ); +if (IS_ERR (nvme_keyring)) +return PTR_ERR (nvme_keyring); + +return 0 ; +} +EXPORT_SYMBOL_GPL (nvme_keyring_init); + +void nvme_keyring_exit (void ) +{ +key_revoke (nvme_keyring); +key_put (nvme_keyring); +} +EXPORT_SYMBOL_GPL (nvme_keyring_exit);
drivers/nvme/host/core.c +8 -2
@@ -25,6 +25,7 @@#include "nvme.h" #include "fabrics.h" #include <linux/nvme-auth.h> +#include <linux/nvme-keyring.h> #define CREATE_TRACE_POINTS #include "trace.h"
@@ -4703,12 +4704,16 @@ static int __init nvme_core_init(void) result =PTR_ERR (nvme_ns_chr_class); goto unregister_generic_ns; } - - result =nvme_init_auth (); + result =nvme_keyring_init ();if (result) goto destroy_ns_chr; + result =nvme_init_auth (); +if (result) + goto keyring_exit;return 0 ; +keyring_exit: +nvme_keyring_exit (); destroy_ns_chr:class_destroy (nvme_ns_chr_class); unregister_generic_ns:
@@ -4732,6 +4737,7 @@ static int __init nvme_core_init(void) static void __exit nvme_core_exit(void) {nvme_exit_auth (); +nvme_keyring_exit ();class_destroy (nvme_ns_chr_class);class_destroy (nvme_subsys_class);class_destroy (nvme_class);
include/linux/nvme-keyring.h new file +28 -0
@@ -0,0 +1,28 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +/* +* Copyright (c) 2023 Hannes Reinecke, SUSE Labs +*/ + +#ifndef _NVME_KEYRING_H +#define _NVME_KEYRING_H + +#ifdef CONFIG_NVME_KEYRING + +key_serial_t nvme_keyring_id (void ); +int nvme_keyring_init (void ); +void nvme_keyring_exit (void ); + +#else + +static inline key_serial_t nvme_keyring_id (void ) +{ +return 0 ; +} +static inline int nvme_keyring_init (void ) +{ +return 0 ; +} +static inline void nvme_keyring_exit (void ) {} + +#endif /* !CONFIG_NVME_KEYRING */ +#endif /* _NVME_KEYRING_H */
Signature · 1 line
2.35.3
PATCH v8 2/17 nvme-keyring: define a 'psk' keytype
In reply to Hannes ReineckeDefine a 'psk' keytype to hold the NVMe TLS PSKs.
--- drivers/nvme/common/keyring.c | 94 +++++++++++++++++++++++++++++++++++ 1 file changed, 94 insertions(+)
drivers/nvme/common/keyring.c +94 -0
@@ -8,6 +8,8 @@#include <linux/key-type.h> #include <keys/user-type.h> #include <linux/nvme.h> +#include <linux/nvme-tcp.h> +#include <linux/nvme-keyring.h> static struct key *nvme_keyring;
@@ -17,8 +19,94 @@ key_serial_t nvme_keyring_id(void) }EXPORT_SYMBOL_GPL (nvme_keyring_id); +static void nvme_tls_psk_describe (const struct key *key,struct seq_file *m) +{ +seq_puts (m, key->description); +seq_printf (m,": %u" , key->datalen); +} + +static bool nvme_tls_psk_match (const struct key *key, +const struct key_match_data *match_data) +{ +const char *match_id; +size_t match_len; + +if (!key->description) { +pr_debug ("%s: no key description\n" , __func__); +return false; + } + match_len =strlen (key->description); +pr_debug ("%s: id %s len %zd\n" , __func__, key->description, match_len); + +if (!match_data->raw_data) { +pr_debug ("%s: no match data\n" , __func__); +return false; + } + match_id = match_data->raw_data; +pr_debug ("%s: match '%s' '%s' len %zd\n" , + __func__, match_id, key->description, match_len); +return !memcmp (key->description, match_id, match_len); +} + +static int nvme_tls_psk_match_preparse (struct key_match_data *match_data) +{ + match_data->lookup_type =KEYRING_SEARCH_LOOKUP_ITERATE ; + match_data->cmp = nvme_tls_psk_match; +return 0 ; +} + +static struct key_type nvme_tls_psk_key_type = { + .name ="psk" , + .flags =KEY_TYPE_NET_DOMAIN , + .preparse = user_preparse, + .free_preparse = user_free_preparse, + .match_preparse = nvme_tls_psk_match_preparse, + .instantiate = generic_key_instantiate, + .revoke = user_revoke, + .destroy = user_destroy, + .describe = nvme_tls_psk_describe, + .read = user_read, +}; + +static struct key *nvme_tls_psk_lookup (struct key *keyring, +const char *hostnqn,const char *subnqn, +int hmac,bool generated) +{ +char *identity; +size_t identity_len = (NVMF_NQN_SIZE ) *2 +11 ; +key_ref_t keyref; +key_serial_t keyring_id; + + identity =kzalloc (identity_len,GFP_KERNEL ); +if (!identity) +return ERR_PTR (-ENOMEM ); + +snprintf (identity, identity_len,"NVMe0%c%02d %s %s" , + generated ?'G' :'R' , hmac, hostnqn, subnqn); + +if (!keyring) + keyring = nvme_keyring; + keyring_id =key_serial (keyring); +pr_debug ("keyring %x lookup tls psk '%s'\n" , + keyring_id, identity); + keyref =keyring_search (make_key_ref (keyring, true), + &nvme_tls_psk_key_type, + identity, false); +if (IS_ERR (keyref)) { +pr_debug ("lookup tls psk '%s' failed, error %ld\n" , + identity,PTR_ERR (keyref)); +kfree (identity); +return ERR_PTR (-ENOKEY ); + } +kfree (identity); + +return key_ref_to_ptr (keyref); +} +int nvme_keyring_init (void ) { +int err; + nvme_keyring = keyring_alloc(".nvme", GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, current_cred(),
@@ -28,12 +116,18 @@ int nvme_keyring_init(void)if (IS_ERR (nvme_keyring))return PTR_ERR (nvme_keyring); + err =register_key_type (&nvme_tls_psk_key_type); +if (err) { +key_put (nvme_keyring); +return err; + }return 0 ; }EXPORT_SYMBOL_GPL (nvme_keyring_init);void nvme_keyring_exit (void ) { +unregister_key_type (&nvme_tls_psk_key_type);key_revoke (nvme_keyring);key_put (nvme_keyring); }
Signature · 1 line
2.35.3
PATCH v8 3/17 nvme: add TCP TSAS definitions
In reply to Hannes Reinecke--- include/linux/nvme.h | 10 ++++++++++ 1 file changed, 10 insertions(+)
include/linux/nvme.h +10 -0
@@ -108,6 +108,13 @@ enum { NVMF_RDMA_CMS_RDMA_CM = 1, /* Sockets based endpoint addressing */ }; +/* TSAS SECTYPE for TCP transport */ +enum { +NVMF_TCP_SECTYPE_NONE =0 ,/* No Security */ +NVMF_TCP_SECTYPE_TLS12 =1 ,/* TLSv1.2, NVMe-oF 1.1 and NVMe-TCP 3.6.1.1 */ +NVMF_TCP_SECTYPE_TLS13 =2 ,/* TLSv1.3, NVMe-oF 1.1 and NVMe-TCP 3.6.1.1 */ +}; +#define NVME_AQ_DEPTH 32#define NVME_NR_AEN_COMMANDS 1#define NVME_AQ_BLK_MQ_DEPTH (NVME_AQ_DEPTH - NVME_NR_AEN_COMMANDS)
@@ -1493,6 +1500,9 @@ struct nvmf_disc_rsp_page_entry {__u16 pkey;__u8 resv10[246 ]; } rdma; +struct tcp { +__u8 sectype; + } tcp; } tsas; };
Signature · 1 line
2.35.3
PATCH v8 4/17 nvme-tcp: add definitions for TLS cipher suites
In reply to Hannes Reinecke--- include/linux/nvme-tcp.h | 6 ++++++ 1 file changed, 6 insertions(+)
include/linux/nvme-tcp.h +6 -0
@@ -18,6 +18,12 @@ enum nvme_tcp_pfv { NVME_TCP_PFV_1_0 = 0x0, }; +enum nvme_tcp_tls_cipher { +NVME_TCP_TLS_CIPHER_INVALID =0 , +NVME_TCP_TLS_CIPHER_SHA256 =1 , +NVME_TCP_TLS_CIPHER_SHA384 =2 , +}; +enum nvme_tcp_fatal_error_status {NVME_TCP_FES_INVALID_PDU_HDR =0x01 ,NVME_TCP_FES_PDU_SEQ_ERR =0x02 ,
Signature · 1 line
2.35.3
PATCH v8 5/17 nvme-keyring: implement nvme_tls_psk_default()
In reply to Hannes ReineckeImplement a function to select the preferred PSK for TLS. A 'retained' PSK should be preferred over a 'generated' PSK, and SHA-384 should be preferred to SHA-256.
--- drivers/nvme/common/keyring.c | 48 +++++++++++++++++++++++++++++++++++ include/linux/nvme-keyring.h | 8 ++++++ 2 files changed, 56 insertions(+)
drivers/nvme/common/keyring.c +48 -0
@@ -5,6 +5,7 @@#include <linux/module.h> #include <linux/seq_file.h> +#include <linux/key.h> #include <linux/key-type.h> #include <keys/user-type.h> #include <linux/nvme.h>
@@ -103,6 +104,53 @@ static struct key *nvme_tls_psk_lookup(struct key *keyring,return key_ref_to_ptr (keyref); } +/* +* NVMe PSK priority list +* +* 'Retained' PSKs (ie 'generated == false') +* should be preferred to 'generated' PSKs, +* and SHA-384 should be preferred to SHA-256. +*/ +struct nvme_tls_psk_priority_list { +bool generated; +enum nvme_tcp_tls_cipher cipher; +} nvme_tls_psk_prio[] = { + { .generated = false, + .cipher =NVME_TCP_TLS_CIPHER_SHA384 , }, + { .generated = false, + .cipher =NVME_TCP_TLS_CIPHER_SHA256 , }, + { .generated = true, + .cipher =NVME_TCP_TLS_CIPHER_SHA384 , }, + { .generated = true, + .cipher =NVME_TCP_TLS_CIPHER_SHA256 , }, +}; + +/* +* nvme_tls_psk_default - Return the preferred PSK to use for TLS ClientHello +*/ +key_serial_t nvme_tls_psk_default (struct key *keyring, +const char *hostnqn,const char *subnqn) +{ +struct key *tls_key; +key_serial_t tls_key_id; +int prio; + +for (prio =0 ; prio <ARRAY_SIZE (nvme_tls_psk_prio); prio++) { +bool generated = nvme_tls_psk_prio[prio].generated; +enum nvme_tcp_tls_cipher cipher = nvme_tls_psk_prio[prio].cipher; + + tls_key =nvme_tls_psk_lookup (keyring, hostnqn, subnqn, + cipher, generated); +if (!IS_ERR (tls_key)) { + tls_key_id = tls_key->serial; +key_put (tls_key); +return tls_key_id; + } + } +return 0 ; +} +EXPORT_SYMBOL_GPL (nvme_tls_psk_default); +int nvme_keyring_init (void ) {int err;
include/linux/nvme-keyring.h +8 -0
@@ -8,12 +8,20 @@#ifdef CONFIG_NVME_KEYRING +key_serial_t nvme_tls_psk_default (struct key *keyring, +const char *hostnqn,const char *subnqn); +key_serial_t nvme_keyring_id (void );int nvme_keyring_init (void );void nvme_keyring_exit (void );#else +static inline key_serial_t nvme_tls_psk_default (struct key *keyring, +const char *hostnqn,const char *subnqn) +{ +return 0 ; +}static inline key_serial_t nvme_keyring_id (void ) {return 0 ;
Signature · 1 line
2.35.3
PATCH v8 6/17 security/keys: export key_lookup()
In reply to Hannes ReineckeFor in-kernel consumers one cannot readily assign a user (eg when running from a workqueue), so the normal key search permissions cannot be applied.
This patch exports the 'key_lookup()' function for a simple lookup of keys without checking for permissions.
--- security/keys/key.c | 1 + 1 file changed, 1 insertion(+)
security/keys/key.c +1 -0
@@ -693,6 +693,7 @@ struct key *key_lookup(key_serial_t id)spin_unlock (&key_serial_lock);return key; } +EXPORT_SYMBOL (key_lookup);/* * Find and lock the specified key type against removal.
Signature · 1 line
2.35.3
PATCH v8 7/17 nvme-tcp: allocate socket file
In reply to Hannes ReineckeWhen using the TLS upcall we need to allocate a socket file such that the userspace daemon is able to use the socket.
--- drivers/nvme/host/tcp.c | 13 +++++++++++-- 1 file changed, 11 insertions(+), 2 deletions(-)
drivers/nvme/host/tcp.c +11 -2
@@ -1338,7 +1338,9 @@ static void nvme_tcp_free_queue(struct nvme_ctrl *nctrl, int qid) } noreclaim_flag =memalloc_noreclaim_save (); -sock_release (queue->sock); +/* ->sock will be released by fput() */ +fput (queue->sock->file); + queue->sock =NULL ;memalloc_noreclaim_restore (noreclaim_flag);kfree (queue->pdu);
@@ -1512,6 +1514,7 @@ static int nvme_tcp_alloc_queue(struct nvme_ctrl *nctrl, int qid)struct nvme_tcp_ctrl *ctrl =to_tcp_ctrl (nctrl);struct nvme_tcp_queue *queue = &ctrl->queues[qid];int ret, rcv_pdu_size; +struct file *sock_file;mutex_init (&queue->queue_lock); queue->ctrl = ctrl;
@@ -1534,6 +1537,11 @@ static int nvme_tcp_alloc_queue(struct nvme_ctrl *nctrl, int qid) goto err_destroy_mutex; } + sock_file =sock_alloc_file (queue->sock,O_CLOEXEC ,NULL ); +if (IS_ERR (sock_file)) { + ret =PTR_ERR (sock_file); + goto err_destroy_mutex; + }nvme_tcp_reclassify_socket (queue->sock);/* Single syn retry */
@@ -1640,7 +1648,8 @@ static int nvme_tcp_alloc_queue(struct nvme_ctrl *nctrl, int qid)if (queue->hdr_digest || queue->data_digest)nvme_tcp_free_crypto (queue); err_sock: -sock_release (queue->sock); +/* ->sock will be released by fput() */ +fput (queue->sock->file); queue->sock =NULL ; err_destroy_mutex:mutex_destroy (&queue->send_mutex);
Signature · 1 line
2.35.3
PATCH v8 8/17 nvme-tcp: enable TLS handshake upcall
In reply to Hannes ReineckeAdd a fabrics option 'tls' and start the TLS handshake upcall with the default PSK. When TLS is started the PSK key serial number is displayed in the sysfs attribute 'tls_key'
--- drivers/nvme/host/Kconfig | 15 ++++ drivers/nvme/host/core.c | 2 +- drivers/nvme/host/fabrics.c | 12 ++++ drivers/nvme/host/fabrics.h | 3 + drivers/nvme/host/nvme.h | 1 + drivers/nvme/host/sysfs.c | 21 ++++++ drivers/nvme/host/tcp.c | 137 ++++++++++++++++++++++++++++++++++-- 7 files changed, 184 insertions(+), 7 deletions(-)
drivers/nvme/host/Kconfig +15 -0
@@ -92,6 +92,21 @@ config NVME_TCP If unsure, say N. +config NVME_TCP_TLS + bool "NVMe over Fabrics TCP TLS encryption support" + depends on NVME_TCP + select NVME_COMMON + select NVME_KEYRING + select NET_HANDSHAKE + select KEYS + help + Enables TLS encryption for NVMe TCP using the netlink handshake API. + + The TLS handshake daemon is availble at + https://github.com/oracle/ktls-utils. + + If unsure, say N. + config NVME_AUTH bool "NVM Express over Fabrics In-Band Authentication" depends on NVME_CORE
drivers/nvme/host/core.c +1 -1
@@ -4380,7 +4380,7 @@ static void nvme_free_ctrl(struct device *dev)if (!subsys || ctrl->instance != subsys->instance)ida_free (&nvme_instance_ida, ctrl->instance); - +key_put (ctrl->tls_key);nvme_free_cels (ctrl);nvme_mpath_uninit (ctrl);nvme_auth_stop (ctrl);
drivers/nvme/host/fabrics.c +12 -0
@@ -647,6 +647,9 @@ static const match_table_t opt_tokens = { { NVMF_OPT_DISCOVERY, "discovery" }, { NVMF_OPT_DHCHAP_SECRET, "dhchap_secret=%s" }, { NVMF_OPT_DHCHAP_CTRL_SECRET, "dhchap_ctrl_secret=%s" }, +#ifdef CONFIG_NVME_TCP_TLS + { NVMF_OPT_TLS, "tls" }, +#endif { NVMF_OPT_ERR, NULL } };
@@ -671,6 +674,7 @@ static int nvmf_parse_options(struct nvmf_ctrl_options *opts, opts->hdr_digest = false; opts->data_digest = false; opts->tos =-1 ;/* < 0 == use transport default */ + opts->tls = false; options = o =kstrdup (buf,GFP_KERNEL );if (!options)
@@ -955,6 +959,14 @@ static int nvmf_parse_options(struct nvmf_ctrl_options *opts,kfree (opts->dhchap_ctrl_secret); opts->dhchap_ctrl_secret = p;break ; +case NVMF_OPT_TLS : +if (!IS_ENABLED (CONFIG_NVME_TCP_TLS )) { +pr_err ("TLS is not supported\n" ); + ret = -EINVAL ; + goto out; + } + opts->tls = true; +break ;default :pr_warn ("unknown parameter or missing value '%s' in ctrl creation request\n" , p);
drivers/nvme/host/fabrics.h +3 -0
@@ -70,6 +70,7 @@ enum {NVMF_OPT_DISCOVERY =1 <<22 ,NVMF_OPT_DHCHAP_SECRET =1 <<23 ,NVMF_OPT_DHCHAP_CTRL_SECRET =1 <<24 , + NVMF_OPT_TLS = 1 << 25, };/**
@@ -102,6 +103,7 @@ enum {* @dhchap_secret: DH-HMAC-CHAP secret * @dhchap_ctrl_secret: DH-HMAC-CHAP controller secret for bi-directional * authentication +* @tls: Start TLS encrypted connections (TCP) * @disable_sqflow: disable controller sq flow control * @hdr_digest: generate/verify header digest (TCP) * @data_digest: generate/verify data digest (TCP)
@@ -128,6 +130,7 @@ struct nvmf_ctrl_options {struct nvmf_host *host;char *dhchap_secret;char *dhchap_ctrl_secret; +bool tls;bool disable_sqflow;bool hdr_digest;bool data_digest;
drivers/nvme/host/nvme.h +1 -0
@@ -357,6 +357,7 @@ struct nvme_ctrl {struct nvme_dhchap_key *ctrl_key;u16 transaction;#endif +struct key *tls_key;/* Power saving configuration */ u64 ps_max_latency_us;
drivers/nvme/host/sysfs.c +21 -0
@@ -527,6 +527,19 @@ static DEVICE_ATTR(dhchap_ctrl_secret, S_IRUGO | S_IWUSR, nvme_ctrl_dhchap_ctrl_secret_show, nvme_ctrl_dhchap_ctrl_secret_store);#endif +#ifdef CONFIG_NVME_TCP_TLS +static ssize_t tls_key_show (struct device *dev, +struct device_attribute *attr,char *buf) +{ +struct nvme_ctrl *ctrl =dev_get_drvdata (dev); + +if (!ctrl->tls_key) +return 0 ; +return sysfs_emit (buf,"%08x" ,key_serial (ctrl->tls_key)); +} +static DEVICE_ATTR_RO (tls_key); +#endif +static struct attribute *nvme_dev_attrs[] = { &dev_attr_reset_controller.attr, &dev_attr_rescan_controller.attr,
@@ -553,6 +566,9 @@ static struct attribute *nvme_dev_attrs[] = {#ifdef CONFIG_NVME_AUTH &dev_attr_dhchap_secret.attr, &dev_attr_dhchap_ctrl_secret.attr, +#endif +#ifdef CONFIG_NVME_TCP_TLS + &dev_attr_tls_key.attr, #endifNULL };
@@ -583,6 +599,11 @@ static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,if (a == &dev_attr_dhchap_ctrl_secret.attr && !ctrl->opts)return 0 ;#endif +#ifdef CONFIG_NVME_TCP_TLS +if (a == &dev_attr_tls_key.attr && + (!ctrl->opts ||strcmp (ctrl->opts->transport,"tcp" ))) +return 0 ; +#endif return a->mode; }
drivers/nvme/host/tcp.c +131 -6
191 lines not on this page · raw message
Signature · 1 line
2.35.3
Re: [PATCH 08/17] nvme-tcp: enable TLS handshake upcall
I *think* this looks good,
PATCH v8 9/17 nvme-tcp: control message handling for recvmsg()
In reply to Hannes ReineckekTLS is sending TLS ALERT messages as control messages for recvmsg(). As we can't do anything sensible with it just abort the connection and let the userspace agent to a re-negotiation.
--- drivers/nvme/host/tcp.c | 17 ++++++++++++++++- 1 file changed, 16 insertions(+), 1 deletion(-)
drivers/nvme/host/tcp.c +16 -1
@@ -14,6 +14,7 @@#include <net/sock.h> #include <net/tcp.h> #include <net/tls.h> +#include <net/tls_prot.h> #include <net/handshake.h> #include <linux/blk-mq.h> #include <crypto/hash.h>
@@ -1369,6 +1370,8 @@ static int nvme_tcp_init_connection(struct nvme_tcp_queue *queue) {struct nvme_tcp_icreq_pdu *icreq;struct nvme_tcp_icresp_pdu *icresp; +char cbuf[CMSG_LEN (sizeof (char ))] = {}; +u8 ctype;struct msghdr msg = {};struct kvec iov;bool ctrl_hdgst, ctrl_ddgst;
@@ -1406,11 +1409,23 @@ static int nvme_tcp_init_connection(struct nvme_tcp_queue *queue)memset (&msg,0 ,sizeof (msg)); iov.iov_base = icresp; iov.iov_len =sizeof (*icresp); +if (queue->ctrl->ctrl.opts->tls) { + msg.msg_control = cbuf; + msg.msg_controllen =sizeof (cbuf); + } ret =kernel_recvmsg (queue->sock, &msg, &iov,1 , iov.iov_len, msg.msg_flags);if (ret <0 ) goto free_icresp; - +if (queue->ctrl->ctrl.opts->tls) { + ctype =tls_get_record_type (queue->sock->sk, + (struct cmsghdr *)cbuf); +if (ctype !=TLS_RECORD_TYPE_DATA ) { +pr_err ("queue %d: unhandled TLS record %d\n" , +nvme_tcp_queue_id (queue), ctype); +return -ENOTCONN ; + } + } ret = -EINVAL ;if (icresp->hdr.type != nvme_tcp_icresp) { pr_err("queue %d: bad type returned %d\n",
Signature · 1 line
2.35.3
PATCH v8 10/17 nvme-fabrics: parse options 'keyring' and 'tls_key'
In reply to Hannes ReineckeParse the fabrics options 'keyring' and 'tls_key' and store the referenced keys in the options structure.
--- drivers/nvme/host/fabrics.c | 49 ++++++++++++++++++++++++++++++++++++- drivers/nvme/host/fabrics.h | 6 +++++ drivers/nvme/host/tcp.c | 11 ++++++--- 3 files changed, 62 insertions(+), 4 deletions(-)
drivers/nvme/host/fabrics.c +48 -1
@@ -643,6 +643,10 @@ static const match_table_t opt_tokens = { { NVMF_OPT_NR_WRITE_QUEUES, "nr_write_queues=%d" }, { NVMF_OPT_NR_POLL_QUEUES, "nr_poll_queues=%d" }, { NVMF_OPT_TOS, "tos=%d" }, +#ifdef CONFIG_NVME_TCP_TLS + { NVMF_OPT_KEYRING, "keyring=%d" }, + { NVMF_OPT_TLS_KEY, "tls_key=%d" }, +#endif { NVMF_OPT_FAIL_FAST_TMO, "fast_io_fail_tmo=%d" }, { NVMF_OPT_DISCOVERY, "discovery" }, { NVMF_OPT_DHCHAP_SECRET, "dhchap_secret=%s" },
@@ -660,9 +664,10 @@ static int nvmf_parse_options(struct nvmf_ctrl_options *opts,char *options, *o, *p;int token, ret =0 ;size_t nqnlen =0 ; -int ctrl_loss_tmo =NVMF_DEF_CTRL_LOSS_TMO ; +int ctrl_loss_tmo =NVMF_DEF_CTRL_LOSS_TMO , key_id;uuid_t hostid;char hostnqn[NVMF_NQN_SIZE ]; +struct key *key;/* Set defaults */ opts->queue_size =NVMF_DEF_QUEUE_SIZE ;
@@ -928,6 +933,46 @@ static int nvmf_parse_options(struct nvmf_ctrl_options *opts, } opts->tos = token;break ; +case NVMF_OPT_KEYRING : +if (!IS_ENABLED (CONFIG_NVME_TCP_TLS )) { +pr_err ("TLS is not supported\n" ); + ret = -EINVAL ; + goto out; + } +if (match_int (args, &key_id) || key_id <=0 ) { + ret = -EINVAL ; + goto out; + } + key =key_lookup (key_id); +if (IS_ERR (key)) { +pr_err ("Keyring %08x not found\n" , key_id); + ret =PTR_ERR (key); + goto out; + } +pr_debug ("Using keyring %08x\n" ,key_serial (key)); +key_put (opts->keyring); + opts->keyring = key; +break ; +case NVMF_OPT_TLS_KEY : +if (!IS_ENABLED (CONFIG_NVME_TCP_TLS )) { +pr_err ("TLS is not supported\n" ); + ret = -EINVAL ; + goto out; + } +if (match_int (args, &key_id) || key_id <=0 ) { + ret = -EINVAL ; + goto out; + } + key =key_lookup (key_id); +if (IS_ERR (key)) { +pr_err ("Key %08x not found\n" , key_id); + ret =PTR_ERR (key); + goto out; + } +pr_debug ("Using key %08x\n" ,key_serial (key)); +key_put (opts->tls_key); + opts->tls_key = key; +break ;case NVMF_OPT_DISCOVERY : opts->discovery_nqn = true;break ;
@@ -1168,6 +1213,8 @@ static int nvmf_check_allowed_opts(struct nvmf_ctrl_options *opts,void nvmf_free_options (struct nvmf_ctrl_options *opts) {nvmf_host_put (opts->host); +key_put (opts->keyring); +key_put (opts->tls_key);kfree (opts->transport);kfree (opts->traddr);kfree (opts->trsvcid);
drivers/nvme/host/fabrics.h +6 -0
@@ -71,6 +71,8 @@ enum {NVMF_OPT_DHCHAP_SECRET =1 <<23 ,NVMF_OPT_DHCHAP_CTRL_SECRET =1 <<24 ,NVMF_OPT_TLS =1 <<25 , +NVMF_OPT_KEYRING =1 <<26 , + NVMF_OPT_TLS_KEY = 1 << 27, };/**
@@ -103,6 +105,8 @@ enum {* @dhchap_secret: DH-HMAC-CHAP secret * @dhchap_ctrl_secret: DH-HMAC-CHAP controller secret for bi-directional * authentication +* @keyring: Keyring to use for key lookups +* @tls_key: TLS key for encrypted connections (TCP) * @tls: Start TLS encrypted connections (TCP) * @disable_sqflow: disable controller sq flow control * @hdr_digest: generate/verify header digest (TCP)
@@ -130,6 +134,8 @@ struct nvmf_ctrl_options {struct nvmf_host *host;char *dhchap_secret;char *dhchap_ctrl_secret; +struct key *keyring; +struct key *tls_key;bool tls;bool disable_sqflow;bool hdr_digest;
drivers/nvme/host/tcp.c +8 -3
@@ -1583,6 +1583,8 @@ static int nvme_tcp_start_tls(struct nvme_ctrl *nctrl,dev_dbg (nctrl->device,"queue %d: start TLS with key %x\n" , qid, pskid); +if (nctrl->opts->keyring) + keyring =key_serial (nctrl->opts->keyring);memset (&args,0 ,sizeof (args)); args.ta_sock = queue->sock; args.ta_done = nvme_tcp_tls_done;
@@ -1907,9 +1909,12 @@ static int nvme_tcp_alloc_admin_queue(struct nvme_ctrl *ctrl)key_serial_t pskid =0 ;if (ctrl->opts->tls) { - pskid =nvme_tls_psk_default (NULL , - ctrl->opts->host->nqn, - ctrl->opts->subsysnqn); +if (ctrl->opts->tls_key) + pskid =key_serial (ctrl->opts->tls_key); +else + pskid =nvme_tls_psk_default (ctrl->opts->keyring, + ctrl->opts->host->nqn, + ctrl->opts->subsysnqn);if (!pskid) {dev_err (ctrl->device,"no valid PSK found\n" ); ret = -ENOKEY ;
Signature · 1 line
2.35.3
Sagi Grimberg
maintainerRe: [PATCH 10/17] nvme-fabrics: parse options 'keyring' and 'tls_key'
Quoted 35 lines
drivers/nvme/host/fabrics.c +30
-0@@ -928,6 +933,46 @@ static int nvmf_parse_options(struct nvmf_ctrl_options *opts, } opts->tos = token;break ; +case NVMF_OPT_KEYRING : +if (!IS_ENABLED (CONFIG_NVME_TCP_TLS )) { +pr_err ("TLS is not supported\n" ); + ret = -EINVAL ; + goto out; + } +if (match_int (args, &key_id) || key_id <=0 ) { + ret = -EINVAL ; + goto out; + } + key =key_lookup (key_id); +if (IS_ERR (key)) { +pr_err ("Keyring %08x not found\n" , key_id); + ret =PTR_ERR (key); + goto out; + } +pr_debug ("Using keyring %08x\n" ,key_serial (key)); +key_put (opts->keyring); + opts->keyring = key; +break ; +case NVMF_OPT_TLS_KEY : +if (!IS_ENABLED (CONFIG_NVME_TCP_TLS )) { +pr_err ("TLS is not supported\n" ); + ret = -EINVAL ; + goto out; + } +if (match_int (args, &key_id) || key_id <=0 ) { + ret = -EINVAL ; + goto out; + }
drivers/nvme/host/fabrics.c static int nvmf_parse_options(struct nvmf_ctrl_options *opts, · line 975
Quoted 10 lines
drivers/nvme/host/fabrics.c +10
-0+ key =key_lookup (key_id); +if (IS_ERR (key)) { +pr_err ("Key %08x not found\n" , key_id); + ret =PTR_ERR (key); + goto out; + } +pr_debug ("Using key %08x\n" ,key_serial (key)); +key_put (opts->tls_key); + opts->tls_key = key; +break ;
Didn't we agree that we will share the code and simply return the key?
Maybe I missed further discussion on this?
PATCH v8 11/17 nvmet: make TCP sectype settable via configfs
In reply to Hannes ReineckeAdd a new configfs attribute 'addr_tsas' to make the TCP sectype settable via configfs.
--- drivers/nvme/target/configfs.c | 75 +++++++++++++++++++++++++++++++++- 1 file changed, 74 insertions(+), 1 deletion(-)
drivers/nvme/target/configfs.c +74 -1
@@ -174,11 +174,16 @@ static ssize_t nvmet_addr_treq_show(struct config_item *item, char *page)return snprintf (page,PAGE_SIZE ,"\n" ); } +static inline u8 nvmet_port_disc_addr_treq_mask (struct nvmet_port *port) +{ +return (port->disc_addr.treq & ~NVME_TREQ_SECURE_CHANNEL_MASK ); +} +static ssize_t nvmet_addr_treq_store (struct config_item *item,const char *page,size_t count) {struct nvmet_port *port =to_nvmet_port (item); -u8 treq = port->disc_addr.treq & ~NVME_TREQ_SECURE_CHANNEL_MASK ; +u8 treq =nvmet_port_disc_addr_treq_mask (port);int i;if (nvmet_is_port_enabled (port, __func__))
@@ -303,6 +308,11 @@ static void nvmet_port_init_tsas_rdma(struct nvmet_port *port) port->disc_addr.tsas.rdma.cms =NVMF_RDMA_CMS_RDMA_CM ; } +static void nvmet_port_init_tsas_tcp (struct nvmet_port *port,int sectype) +{ + port->disc_addr.tsas.tcp.sectype = sectype; +} +static ssize_t nvmet_addr_trtype_store (struct config_item *item,const char *page,size_t count) {
@@ -325,11 +335,73 @@ static ssize_t nvmet_addr_trtype_store(struct config_item *item, port->disc_addr.trtype = nvmet_transport[i].type;if (port->disc_addr.trtype ==NVMF_TRTYPE_RDMA )nvmet_port_init_tsas_rdma (port); +else if (port->disc_addr.trtype ==NVMF_TRTYPE_TCP ) +nvmet_port_init_tsas_tcp (port,NVMF_TCP_SECTYPE_NONE );return count; }CONFIGFS_ATTR (nvmet_, addr_trtype); +static const struct nvmet_type_name_map nvmet_addr_tsas_tcp[] = { + {NVMF_TCP_SECTYPE_NONE ,"none" }, + {NVMF_TCP_SECTYPE_TLS13 ,"tls1.3" }, +}; + +static const struct nvmet_type_name_map nvmet_addr_tsas_rdma[] = { + {NVMF_RDMA_QPTYPE_CONNECTED ,"connected" }, + {NVMF_RDMA_QPTYPE_DATAGRAM ,"datagram" }, +}; + +static ssize_t nvmet_addr_tsas_show (struct config_item *item, +char *page) +{ +struct nvmet_port *port =to_nvmet_port (item); +int i; + +if (port->disc_addr.trtype ==NVMF_TRTYPE_TCP ) { +for (i =0 ; i <ARRAY_SIZE (nvmet_addr_tsas_tcp); i++) { +if (port->disc_addr.tsas.tcp.sectype == nvmet_addr_tsas_tcp[i].type) +return sprintf (page,"%s\n" , nvmet_addr_tsas_tcp[i].name); + } + }else if (port->disc_addr.trtype ==NVMF_TRTYPE_RDMA ) { +for (i =0 ; i <ARRAY_SIZE (nvmet_addr_tsas_rdma); i++) { +if (port->disc_addr.tsas.rdma.qptype == nvmet_addr_tsas_rdma[i].type) +return sprintf (page,"%s\n" , nvmet_addr_tsas_rdma[i].name); + } + } +return sprintf (page,"reserved\n" ); +} + +static ssize_t nvmet_addr_tsas_store (struct config_item *item, +const char *page,size_t count) +{ +struct nvmet_port *port =to_nvmet_port (item); +u8 sectype; +int i; + +if (nvmet_is_port_enabled (port, __func__)) +return -EACCES ; + +if (port->disc_addr.trtype !=NVMF_TRTYPE_TCP ) +return -EINVAL ; + +for (i =0 ; i <ARRAY_SIZE (nvmet_addr_tsas_tcp); i++) { +if (sysfs_streq (page, nvmet_addr_tsas_tcp[i].name)) { + sectype = nvmet_addr_tsas_tcp[i].type; + goto found; + } + } + +pr_err ("Invalid value '%s' for tsas\n" , page); +return -EINVAL ; + +found: +nvmet_port_init_tsas_tcp (port, sectype); +return count; +} + +CONFIGFS_ATTR (nvmet_, addr_tsas); +/* * Namespace structures & file operation functions below */
@@ -1741,6 +1813,7 @@ static struct configfs_attribute *nvmet_port_attrs[] = { &nvmet_attr_addr_traddr, &nvmet_attr_addr_trsvcid, &nvmet_attr_addr_trtype, + &nvmet_attr_addr_tsas, &nvmet_attr_param_inline_data_size, #ifdef CONFIG_BLK_DEV_INTEGRITY &nvmet_attr_param_pi_enable,
Signature · 1 line
2.35.3
PATCH v8 12/17 nvmet-tcp: make nvmet_tcp_alloc_queue() a void function
In reply to Hannes ReineckeThe return value from nvmet_tcp_alloc_queue() are just used to figure out if sock_release() need to be called. So this patch moves sock_release() into nvmet_tcp_alloc_queue() and make it a void function.
--- drivers/nvme/target/tcp.c | 20 ++++++++++---------- 1 file changed, 10 insertions(+), 10 deletions(-)
drivers/nvme/target/tcp.c +10 -10
@@ -1621,15 +1621,17 @@ static int nvmet_tcp_set_queue_sock(struct nvmet_tcp_queue *queue)return ret; } -static int nvmet_tcp_alloc_queue (struct nvmet_tcp_port *port, +static void nvmet_tcp_alloc_queue (struct nvmet_tcp_port *port,struct socket *newsock) {struct nvmet_tcp_queue *queue;int ret; queue =kzalloc (sizeof (*queue),GFP_KERNEL ); -if (!queue) -return -ENOMEM ; +if (!queue) { + ret = -ENOMEM ; + goto out_release; + }INIT_WORK (&queue->release_work, nvmet_tcp_release_queue_work);INIT_WORK (&queue->io_work, nvmet_tcp_io_work);
@@ -1666,7 +1668,7 @@ static int nvmet_tcp_alloc_queue(struct nvmet_tcp_port *port,if (ret) goto out_destroy_sq; -return 0 ; +return ; out_destroy_sq:mutex_lock (&nvmet_tcp_queue_mutex);list_del_init (&queue->queue_list);
@@ -1678,7 +1680,9 @@ static int nvmet_tcp_alloc_queue(struct nvmet_tcp_port *port,ida_free (&nvmet_tcp_queue_ida, queue->idx); out_free_queue:kfree (queue); -return ret; +out_release: +pr_err ("failed to allocate queue, error %d\n" , ret); +sock_release (newsock); }static void nvmet_tcp_accept_work (struct work_struct *w)
@@ -1695,11 +1699,7 @@ static void nvmet_tcp_accept_work(struct work_struct *w)pr_warn ("failed to accept err=%d\n" , ret);return ; } - ret =nvmet_tcp_alloc_queue (port, newsock); -if (ret) { -pr_err ("failed to allocate queue\n" ); -sock_release (newsock); - } +nvmet_tcp_alloc_queue (port, newsock); } }
Signature · 1 line
2.35.3
PATCH v8 13/17 nvmet-tcp: allocate socket file
In reply to Hannes ReineckeFor the TLS upcall we need to allocate a socket file such that the userspace daemon is able to use the socket.
--- drivers/nvme/target/tcp.c | 18 ++++++++++++++---- 1 file changed, 14 insertions(+), 4 deletions(-)
drivers/nvme/target/tcp.c +14 -4
@@ -1493,12 +1493,12 @@ static void nvmet_tcp_release_queue_work(struct work_struct *w)nvmet_sq_destroy (&queue->nvme_sq);cancel_work_sync (&queue->io_work);nvmet_tcp_free_cmd_data_in_buffers (queue); -sock_release (queue->sock); +/* ->sock will be released by fput() */ +fput (queue->sock->file);nvmet_tcp_free_cmds (queue);if (queue->hdr_digest || queue->data_digest)nvmet_tcp_free_crypto (queue);ida_free (&nvmet_tcp_queue_ida, queue->idx); - page =virt_to_head_page (queue->pf_cache.va);__page_frag_cache_drain (page, queue->pf_cache.pagecnt_bias);kfree (queue);
@@ -1625,6 +1625,7 @@ static void nvmet_tcp_alloc_queue(struct nvmet_tcp_port *port, struct socket *newsock) {struct nvmet_tcp_queue *queue; +struct file *sock_file =NULL ;int ret; queue =kzalloc (sizeof (*queue),GFP_KERNEL );
@@ -1644,10 +1645,16 @@ static void nvmet_tcp_alloc_queue(struct nvmet_tcp_port *port,init_llist_head (&queue->resp_list);INIT_LIST_HEAD (&queue->resp_send_list); + sock_file =sock_alloc_file (queue->sock,O_CLOEXEC ,NULL ); +if (IS_ERR (sock_file)) { + ret =PTR_ERR (sock_file); + goto out_free_queue; + } + queue->idx =ida_alloc (&nvmet_tcp_queue_ida,GFP_KERNEL );if (queue->idx <0 ) { ret = queue->idx; - goto out_free_queue; + goto out_sock; } ret =nvmet_tcp_alloc_cmd (queue, &queue->connect);
@@ -1678,11 +1685,14 @@ static void nvmet_tcp_alloc_queue(struct nvmet_tcp_port *port,nvmet_tcp_free_cmd (&queue->connect); out_ida_remove:ida_free (&nvmet_tcp_queue_ida, queue->idx); +out_sock: +fput (queue->sock->file); out_free_queue:kfree (queue); out_release:pr_err ("failed to allocate queue, error %d\n" , ret); -sock_release (newsock); +if (!sock_file) +sock_release (newsock); }static void nvmet_tcp_accept_work (struct work_struct *w)
Signature · 1 line
2.35.3
PATCH v8 14/17 nvmet-tcp: reference counting for queues
In reply to Hannes ReineckeThe 'queue' structure is referenced from various places and used as an argument of asynchronous functions, so it's really hard to figure out if the queue is still valid when the asynchronous function triggers.
So add reference counting to validate the queue structure.
--- drivers/nvme/target/tcp.c | 74 ++++++++++++++++++++++++++++++--------- 1 file changed, 57 insertions(+), 17 deletions(-)
drivers/nvme/target/tcp.c +57 -17
@@ -127,6 +127,7 @@ enum nvmet_tcp_queue_state { };struct nvmet_tcp_queue { +struct kref kref;struct socket *sock;struct nvmet_tcp_port *port;struct work_struct io_work;
@@ -192,6 +193,9 @@ static struct workqueue_struct *nvmet_tcp_wq;static const struct nvmet_fabrics_ops nvmet_tcp_ops;static void nvmet_tcp_free_cmd (struct nvmet_tcp_cmd *c);static void nvmet_tcp_free_cmd_buffers (struct nvmet_tcp_cmd *cmd); +static int nvmet_tcp_get_queue (struct nvmet_tcp_queue *queue); +static void nvmet_tcp_put_queue (struct nvmet_tcp_queue *queue); +static void nvmet_tcp_data_ready (struct sock *sk);static inline u16 nvmet_tcp_cmd_tag (struct nvmet_tcp_queue *queue,struct nvmet_tcp_cmd *cmd)
@@ -1437,11 +1441,21 @@ static void nvmet_tcp_restore_socket_callbacks(struct nvmet_tcp_queue *queue)struct socket *sock = queue->sock;write_lock_bh (&sock->sk->sk_callback_lock); +/* +* Check if nvmet_tcp_set_queue_sock() has been called; +* if not the queue reference has not been increased +* and we're getting an refcount error on exit. +*/ +if (sock->sk->sk_data_ready != nvmet_tcp_data_ready) { +write_unlock_bh (&sock->sk->sk_callback_lock); +return ; + } sock->sk->sk_data_ready = queue->data_ready; sock->sk->sk_state_change = queue->state_change; sock->sk->sk_write_space = queue->write_space; sock->sk->sk_user_data =NULL ;write_unlock_bh (&sock->sk->sk_callback_lock); +nvmet_tcp_put_queue (queue); }static void nvmet_tcp_uninit_data_in_cmds (struct nvmet_tcp_queue *queue)
@@ -1474,6 +1488,30 @@ static void nvmet_tcp_free_cmd_data_in_buffers(struct nvmet_tcp_queue *queue)nvmet_tcp_free_cmd_buffers (&queue->connect); } +static void nvmet_tcp_release_queue_final (struct kref *kref) +{ + struct nvmet_tcp_queue *queue = container_of(kref, struct nvmet_tcp_queue, kref); + +WARN_ON (queue->state !=NVMET_TCP_Q_DISCONNECTING ); +nvmet_tcp_free_cmds (queue); +ida_free (&nvmet_tcp_queue_ida, queue->idx); +/* ->sock will be released by fput() */ +fput (queue->sock->file); +kfree (queue); +} + +static int nvmet_tcp_get_queue (struct nvmet_tcp_queue *queue) +{ +if (!queue) +return 0 ; +return kref_get_unless_zero (&queue->kref); +} + +static void nvmet_tcp_put_queue (struct nvmet_tcp_queue *queue) +{ +kref_put (&queue->kref, nvmet_tcp_release_queue_final); +} +static void nvmet_tcp_release_queue_work (struct work_struct *w) {struct page *page;
@@ -1493,15 +1531,11 @@ static void nvmet_tcp_release_queue_work(struct work_struct *w)nvmet_sq_destroy (&queue->nvme_sq);cancel_work_sync (&queue->io_work);nvmet_tcp_free_cmd_data_in_buffers (queue); -/* ->sock will be released by fput() */ -fput (queue->sock->file); -nvmet_tcp_free_cmds (queue);if (queue->hdr_digest || queue->data_digest)nvmet_tcp_free_crypto (queue); -ida_free (&nvmet_tcp_queue_ida, queue->idx); page =virt_to_head_page (queue->pf_cache.va);__page_frag_cache_drain (page, queue->pf_cache.pagecnt_bias); -kfree (queue); +nvmet_tcp_put_queue (queue); }static void nvmet_tcp_data_ready (struct sock *sk)
@@ -1512,8 +1546,10 @@ static void nvmet_tcp_data_ready(struct sock *sk)read_lock_bh (&sk->sk_callback_lock); queue = sk->sk_user_data; -if (likely (queue)) +if (likely (nvmet_tcp_get_queue (queue))) {queue_work_on (queue_cpu (queue), nvmet_tcp_wq, &queue->io_work); +nvmet_tcp_put_queue (queue); + }read_unlock_bh (&sk->sk_callback_lock); }
@@ -1523,18 +1559,16 @@ static void nvmet_tcp_write_space(struct sock *sk)read_lock_bh (&sk->sk_callback_lock); queue = sk->sk_user_data; -if (unlikely (!queue)) +if (unlikely (!nvmet_tcp_get_queue (queue))) goto out;if (unlikely (queue->state ==NVMET_TCP_Q_CONNECTING )) { queue->write_space (sk); - goto out; - } - -if (sk_stream_is_writeable (sk)) { + }else if (sk_stream_is_writeable (sk)) {clear_bit (SOCK_NOSPACE , &sk->sk_socket->flags);queue_work_on (queue_cpu (queue), nvmet_tcp_wq, &queue->io_work); } +nvmet_tcp_put_queue (queue); out:read_unlock_bh (&sk->sk_callback_lock); }
@@ -1545,7 +1579,7 @@ static void nvmet_tcp_state_change(struct sock *sk)read_lock_bh (&sk->sk_callback_lock); queue = sk->sk_user_data; -if (!queue) +if (!nvmet_tcp_get_queue (queue)) goto done;switch (sk->sk_state) {
@@ -1562,6 +1596,7 @@ static void nvmet_tcp_state_change(struct sock *sk)pr_warn ("queue %d unhandled state %d\n" , queue->idx, sk->sk_state); } +nvmet_tcp_put_queue (queue); done:read_unlock_bh (&sk->sk_callback_lock); }
@@ -1582,6 +1617,9 @@ static int nvmet_tcp_set_queue_sock(struct nvmet_tcp_queue *queue)if (ret <0 )return ret; +if (unlikely (!nvmet_tcp_get_queue (queue))) +return -ECONNRESET ; +/* * Cleanup whatever is sitting in the TCP transmit queue on socket * close. This is done to prevent stale data from being sent should
@@ -1603,6 +1641,7 @@ static int nvmet_tcp_set_queue_sock(struct nvmet_tcp_queue *queue)* If the socket is already closing, don't even start * consuming it */ +nvmet_tcp_put_queue (queue); ret = -ENOTCONN ; }else { sock->sk->sk_user_data = queue;
@@ -1636,6 +1675,7 @@ static void nvmet_tcp_alloc_queue(struct nvmet_tcp_port *port,INIT_WORK (&queue->release_work, nvmet_tcp_release_queue_work);INIT_WORK (&queue->io_work, nvmet_tcp_io_work); +kref_init (&queue->kref); queue->sock = newsock; queue->port = port; queue->nr_cmds =0 ;
@@ -1672,15 +1712,15 @@ static void nvmet_tcp_alloc_queue(struct nvmet_tcp_port *port,mutex_unlock (&nvmet_tcp_queue_mutex); ret =nvmet_tcp_set_queue_sock (queue); -if (ret) - goto out_destroy_sq; +if (!ret) +return ; -return ; -out_destroy_sq: + queue->state =NVMET_TCP_Q_DISCONNECTING ;mutex_lock (&nvmet_tcp_queue_mutex);list_del_init (&queue->queue_list);mutex_unlock (&nvmet_tcp_queue_mutex); -nvmet_sq_destroy (&queue->nvme_sq); +nvmet_tcp_put_queue (queue); +return ; out_free_connect:nvmet_tcp_free_cmd (&queue->connect); out_ida_remove:
Signature · 1 line
2.35.3
Sagi Grimberg
maintainerRe: [PATCH 14/17] nvmet-tcp: reference counting for queues
Quoted 36 lines · Hannes Reinecke
The 'queue' structure is referenced from various places and used as an argument of asynchronous functions, so it's really hard to figure out if the queue is still valid when the asynchronous function triggers.
So add reference counting to validate the queue structure.Signed-off-by Hannes Reinecke <[email protected]>--- drivers/nvme/target/tcp.c | 74 ++++++++++++++++++++++++++++++--------- 1 file changed, 57 insertions(+), 17 deletions(-)drivers/nvme/target/tcp.c +4
-0@@ -127,6 +127,7 @@ enum nvmet_tcp_queue_state { };struct nvmet_tcp_queue { +struct kref kref;struct socket *sock;struct nvmet_tcp_port *port;struct work_struct io_work;@@ -192,6 +193,9 @@ static struct workqueue_struct *nvmet_tcp_wq;static const struct nvmet_fabrics_ops nvmet_tcp_ops;static void nvmet_tcp_free_cmd (struct nvmet_tcp_cmd *c);static void nvmet_tcp_free_cmd_buffers (struct nvmet_tcp_cmd *cmd); +static int nvmet_tcp_get_queue (struct nvmet_tcp_queue *queue); +static void nvmet_tcp_put_queue (struct nvmet_tcp_queue *queue); +static void nvmet_tcp_data_ready (struct sock *sk);static inline u16 nvmet_tcp_cmd_tag (struct nvmet_tcp_queue *queue,struct nvmet_tcp_cmd *cmd)@@ -1437,11 +1441,21 @@ static void nvmet_tcp_restore_socket_callbacks(struct nvmet_tcp_queue *queue)struct socket *sock = queue->sock;
drivers/nvme/target/tcp.c static void nvmet_tcp_restore_socket_callbacks(struct nvmet_tcp_queue *queue) · line 1452
Quoted 10 lines
drivers/nvme/target/tcp.c +9
-0write_lock_bh (&sock->sk->sk_callback_lock); +/* +* Check if nvmet_tcp_set_queue_sock() has been called; +* if not the queue reference has not been increased +* and we're getting an refcount error on exit. +*/ +if (sock->sk->sk_data_ready != nvmet_tcp_data_ready) { +write_unlock_bh (&sock->sk->sk_callback_lock); +return ; + }
This is really ugly I think.
Quoted 44 lines
drivers/nvme/target/tcp.c +25
-0sock->sk->sk_data_ready = queue->data_ready; sock->sk->sk_state_change = queue->state_change; sock->sk->sk_write_space = queue->write_space; sock->sk->sk_user_data =NULL ;write_unlock_bh (&sock->sk->sk_callback_lock); +nvmet_tcp_put_queue (queue); }static void nvmet_tcp_uninit_data_in_cmds (struct nvmet_tcp_queue *queue)@@ -1474,6 +1488,30 @@ static void nvmet_tcp_free_cmd_data_in_buffers(struct nvmet_tcp_queue *queue)nvmet_tcp_free_cmd_buffers (&queue->connect); } +static void nvmet_tcp_release_queue_final (struct kref *kref) +{ + struct nvmet_tcp_queue *queue = container_of(kref, struct nvmet_tcp_queue, kref); + +WARN_ON (queue->state !=NVMET_TCP_Q_DISCONNECTING ); +nvmet_tcp_free_cmds (queue); +ida_free (&nvmet_tcp_queue_ida, queue->idx); +/* ->sock will be released by fput() */ +fput (queue->sock->file); +kfree (queue); +} + +static int nvmet_tcp_get_queue (struct nvmet_tcp_queue *queue) +{ +if (!queue) +return 0 ; +return kref_get_unless_zero (&queue->kref); +} + +static void nvmet_tcp_put_queue (struct nvmet_tcp_queue *queue) +{ +kref_put (&queue->kref, nvmet_tcp_release_queue_final); +} +static void nvmet_tcp_release_queue_work (struct work_struct *w) {struct page *page;@@ -1493,15 +1531,11 @@ static void nvmet_tcp_release_queue_work(struct work_struct *w)nvmet_sq_destroy (&queue->nvme_sq);cancel_work_sync (&queue->io_work);nvmet_tcp_free_cmd_data_in_buffers (queue);
drivers/nvme/target/tcp.c static void nvmet_tcp_release_queue_work(struct work_struct *w) · line 1538
Quoted 10 lines
drivers/nvme/target/tcp.c +1
-5-/* ->sock will be released by fput() */ -fput (queue->sock->file); -nvmet_tcp_free_cmds (queue);if (queue->hdr_digest || queue->data_digest)nvmet_tcp_free_crypto (queue); -ida_free (&nvmet_tcp_queue_ida, queue->idx); page =virt_to_head_page (queue->pf_cache.va);__page_frag_cache_drain (page, queue->pf_cache.pagecnt_bias); -kfree (queue); +nvmet_tcp_put_queue (queue);
What made you pick these vs. the others for before/after the final reference?
drivers/nvme/target/tcp.c static void nvmet_tcp_data_ready(struct sock *sk) · line 1552
Quoted 12 lines
drivers/nvme/target/tcp.c +3
-1}static void nvmet_tcp_data_ready (struct sock *sk)@@ -1512,8 +1546,10 @@ static void nvmet_tcp_data_ready(struct sock *sk)read_lock_bh (&sk->sk_callback_lock); queue = sk->sk_user_data; -if (likely (queue)) +if (likely (nvmet_tcp_get_queue (queue))) {queue_work_on (queue_cpu (queue), nvmet_tcp_wq, &queue->io_work); +nvmet_tcp_put_queue (queue); + }
No... Why?
The shutdown code should serialize perfectly without this. Why add a kref to the normal I/O path?
I thought we'd simply move release_work to do a kref_put and take an extra reference when we fire the tls handshake...
Quoted 12 lines
drivers/nvme/target/tcp.c +1
-1read_unlock_bh (&sk->sk_callback_lock); }@@ -1523,18 +1559,16 @@ static void nvmet_tcp_write_space(struct sock *sk)read_lock_bh (&sk->sk_callback_lock); queue = sk->sk_user_data; -if (unlikely (!queue)) +if (unlikely (!nvmet_tcp_get_queue (queue))) goto out;if (unlikely (queue->state ==NVMET_TCP_Q_CONNECTING )) {
drivers/nvme/target/tcp.c static void nvmet_tcp_write_space(struct sock *sk) · line 1571
Quoted 10 lines
drivers/nvme/target/tcp.c +2
-4queue->write_space (sk); - goto out; - } - -if (sk_stream_is_writeable (sk)) { + }else if (sk_stream_is_writeable (sk)) {clear_bit (SOCK_NOSPACE , &sk->sk_socket->flags);queue_work_on (queue_cpu (queue), nvmet_tcp_wq, &queue->io_work); } +nvmet_tcp_put_queue (queue);
Same here...
Why not simply exclude the kref get/put to the tls handshake where there is actually a race?
Quoted 21 lines
drivers/nvme/target/tcp.c +2
-1out:read_unlock_bh (&sk->sk_callback_lock); }@@ -1545,7 +1579,7 @@ static void nvmet_tcp_state_change(struct sock *sk)read_lock_bh (&sk->sk_callback_lock); queue = sk->sk_user_data; -if (!queue) +if (!nvmet_tcp_get_queue (queue)) goto done;switch (sk->sk_state) {@@ -1562,6 +1596,7 @@ static void nvmet_tcp_state_change(struct sock *sk)pr_warn ("queue %d unhandled state %d\n" , queue->idx, sk->sk_state); } +nvmet_tcp_put_queue (queue); done:read_unlock_bh (&sk->sk_callback_lock); }
drivers/nvme/target/tcp.c static int nvmet_tcp_set_queue_sock(struct nvmet_tcp_queue *queue) · line 1622
Quoted 6 lines
drivers/nvme/target/tcp.c +3
-0@@ -1582,6 +1617,9 @@ static int nvmet_tcp_set_queue_sock(struct nvmet_tcp_queue *queue)if (ret <0 )return ret; +if (unlikely (!nvmet_tcp_get_queue (queue))) +return -ECONNRESET ; +
Can this actually fail?
Quoted 40 lines
drivers/nvme/target/tcp.c +7
-5/* * Cleanup whatever is sitting in the TCP transmit queue on socket * close. This is done to prevent stale data from being sent should @@ -1603,6 +1641,7 @@ static int nvmet_tcp_set_queue_sock(struct nvmet_tcp_queue *queue)* If the socket is already closing, don't even start * consuming it */ +nvmet_tcp_put_queue (queue); ret = -ENOTCONN ; }else { sock->sk->sk_user_data = queue;@@ -1636,6 +1675,7 @@ static void nvmet_tcp_alloc_queue(struct nvmet_tcp_port *port,INIT_WORK (&queue->release_work, nvmet_tcp_release_queue_work);INIT_WORK (&queue->io_work, nvmet_tcp_io_work); +kref_init (&queue->kref); queue->sock = newsock; queue->port = port; queue->nr_cmds =0 ;@@ -1672,15 +1712,15 @@ static void nvmet_tcp_alloc_queue(struct nvmet_tcp_port *port,mutex_unlock (&nvmet_tcp_queue_mutex); ret =nvmet_tcp_set_queue_sock (queue); -if (ret) - goto out_destroy_sq; +if (!ret) +return ; -return ; -out_destroy_sq: + queue->state =NVMET_TCP_Q_DISCONNECTING ;mutex_lock (&nvmet_tcp_queue_mutex);list_del_init (&queue->queue_list);mutex_unlock (&nvmet_tcp_queue_mutex); -nvmet_sq_destroy (&queue->nvme_sq); +nvmet_tcp_put_queue (queue); +return ; out_free_connect:nvmet_tcp_free_cmd (&queue->connect); out_ida_remove:
Re: [PATCH 14/17] nvmet-tcp: reference counting for queues
Quoted 41 lines · Sagi Grimberg
Quoted 38 lines · Hannes Reinecke
The 'queue' structure is referenced from various places and used as an argument of asynchronous functions, so it's really hard to figure out if the queue is still valid when the asynchronous function triggers.
So add reference counting to validate the queue structure.Signed-off-by Hannes Reinecke <[email protected]>---
drivers/nvme/target/tcp.c| 74 ++++++++++++++++++++++++++++++---------1 file changed, 57 insertions(+), 17 deletions(-)drivers/nvme/target/tcp.c +7
-0@@ -127,6 +127,7 @@ enum nvmet_tcp_queue_state { }; struct nvmet_tcp_queue { + struct kref kref; struct socket *sock; struct nvmet_tcp_port *port; struct work_struct io_work;@@ -192,6 +193,9 @@ static struct workqueue_struct *nvmet_tcp_wq; static const struct nvmet_fabrics_ops nvmet_tcp_ops; static void nvmet_tcp_free_cmd(struct nvmet_tcp_cmd *c); static void nvmet_tcp_free_cmd_buffers(struct nvmet_tcp_cmd *cmd); +static int nvmet_tcp_get_queue (struct nvmet_tcp_queue *queue); +static void nvmet_tcp_put_queue (struct nvmet_tcp_queue *queue); +static void nvmet_tcp_data_ready (struct sock *sk); static inline u16 nvmet_tcp_cmd_tag(struct nvmet_tcp_queue *queue, struct nvmet_tcp_cmd *cmd)@@ -1437,11 +1441,21 @@ static void nvmet_tcp_restore_socket_callbacks(struct nvmet_tcp_queue *queue) struct socket *sock = queue->sock; write_lock_bh(&sock->sk->sk_callback_lock); + /* +* Check if nvmet_tcp_set_queue_sock() has been called; +* if not the queue reference has not been increased
Quoted 9 lines
drivers/nvme/target/tcp.c static void nvmet_tcp_restore_socket_callbacks(struct nvmet_tcp_queue *queue) · line 1452
Quoted 6 lines
drivers/nvme/target/tcp.c +6
-0+* and we're getting an refcount error on exit. +*/ + if (sock->sk->sk_data_ready != nvmet_tcp_data_ready) { + write_unlock_bh(&sock->sk->sk_callback_lock); + return; + }This is really ugly I think.
Me too, but what would be the alternative?
Quoted 50 lines
Quoted 50 lines
drivers/nvme/target/tcp.c +25
-0sock->sk->sk_data_ready = queue->data_ready; sock->sk->sk_state_change = queue->state_change; sock->sk->sk_write_space = queue->write_space; sock->sk->sk_user_data = NULL; write_unlock_bh(&sock->sk->sk_callback_lock); + nvmet_tcp_put_queue(queue); } static void nvmet_tcp_uninit_data_in_cmds(struct nvmet_tcp_queue *queue)@@ -1474,6 +1488,30 @@ static void nvmet_tcp_free_cmd_data_in_buffers(struct nvmet_tcp_queue *queue) nvmet_tcp_free_cmd_buffers(&queue->connect); } +static void nvmet_tcp_release_queue_final (struct kref *kref) +{ + struct nvmet_tcp_queue *queue = container_of(kref, struct nvmet_tcp_queue, kref); + + WARN_ON(queue->state != NVMET_TCP_Q_DISCONNECTING); + nvmet_tcp_free_cmds(queue); + ida_free(&nvmet_tcp_queue_ida, queue->idx); + /* ->sock will be released by fput() */ + fput(queue->sock->file); + kfree(queue); +} + +static int nvmet_tcp_get_queue (struct nvmet_tcp_queue *queue) +{ + if (!queue) + return 0; + return kref_get_unless_zero(&queue->kref); +} + +static void nvmet_tcp_put_queue (struct nvmet_tcp_queue *queue) +{ + kref_put(&queue->kref, nvmet_tcp_release_queue_final); +} + static void nvmet_tcp_release_queue_work(struct work_struct *w) {struct page *page;
drivers/nvme/target/tcp.c +0
-3@@ -1493,15 +1531,11 @@ static void nvmet_tcp_release_queue_work(struct work_struct *w) nvmet_sq_destroy(&queue->nvme_sq); cancel_work_sync(&queue->io_work); nvmet_tcp_free_cmd_data_in_buffers(queue); - /* ->sock will be released by fput() */ - fput(queue->sock->file); - nvmet_tcp_free_cmds(queue); if (queue->hdr_digest || queue->data_digest) nvmet_tcp_free_crypto(queue);
Quoted 9 lines
drivers/nvme/target/tcp.c static void nvmet_tcp_release_queue_work(struct work_struct *w) · line 1538
Quoted 5 lines
drivers/nvme/target/tcp.c +1
-2- ida_free(&nvmet_tcp_queue_ida, queue->idx); page = virt_to_head_page(queue->pf_cache.va); __page_frag_cache_drain(page, queue->pf_cache.pagecnt_bias); - kfree(queue); + nvmet_tcp_put_queue(queue);What made you pick these vs. the others for before/after the final reference?
I wanted to call 'nvmet_tcp_put_queue()' for a failed allocation in nvmet_tcp_alloc_queue(), and at that time the queue itself is not live.nvmet_tcp_release_queue() is only called on a live queue, so using that as the kref ->release() function would either limit it's usefulness or would require me to ensure that all calls in there can be made with a non-initialized argument.
Quoted 13 lines
drivers/nvme/target/tcp.c static void nvmet_tcp_data_ready(struct sock *sk) · line 1551
Quoted 10 lines
drivers/nvme/target/tcp.c +3
-1} static void nvmet_tcp_data_ready(struct sock *sk)@@ -1512,8 +1546,10 @@ static void nvmet_tcp_data_ready(struct sock *sk) read_lock_bh(&sk->sk_callback_lock); queue = sk->sk_user_data; - if (likely(queue)) + if (likely(nvmet_tcp_get_queue(queue))) { queue_work_on(queue_cpu(queue), nvmet_tcp_wq, &queue->io_work); + nvmet_tcp_put_queue(queue); + }No... Why?
Quoted 6 lines
The shutdown code should serialize perfectly without this. Why add a kref to the normal I/O path?
I thought we'd simply move
release_workto do akref_putand take an extra reference when we fire the tls handshake...
Because I feel ever so slightly unsure about using the sk_user_data argument. This function is completely asynchronous, and I can't really see how we can ensure that sk_user_data references a valid object. (If it were valid, why would we need to check for !queue ?)
If you have lifetime guarantees that the kref isn't required, by all means, please tell me, and we can drop the kref thing here. But I guess that would imply to _not_ having to check for (!queue) which is fine by me, too.
Quoted 23 lines
Quoted 10 lines
drivers/nvme/target/tcp.c +1
-1read_unlock_bh(&sk->sk_callback_lock); }@@ -1523,18 +1559,16 @@ static void nvmet_tcp_write_space(struct sock *sk) read_lock_bh(&sk->sk_callback_lock); queue = sk->sk_user_data; - if (unlikely(!queue)) + if (unlikely(!nvmet_tcp_get_queue(queue))) goto out; if (unlikely(queue->state == NVMET_TCP_Q_CONNECTING)) {drivers/nvme/target/tcp.c static void nvmet_tcp_write_space(struct sock · line 1570
Quoted 10 lines
drivers/nvme/target/tcp.c +2
-4queue->write_space(sk); - goto out; - } - - if (sk_stream_is_writeable(sk)) { + } else if (sk_stream_is_writeable(sk)) { clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags); queue_work_on(queue_cpu(queue), nvmet_tcp_wq, &queue->io_work); } + nvmet_tcp_put_queue(queue);Same here...
Quoted 3 lines
Why not simply exclude the kref get/put to the tls handshake where there is actually a race?
It's not abot the handshake, it's about socket closing. See above.
Quoted 19 lines
Quoted 19 lines
out:
read_unlock_bh(&sk->sk_callback_lock);
}drivers/nvme/target/tcp.c +2
-1@@ -1545,7 +1579,7 @@ static void nvmet_tcp_state_change(struct sock *sk) read_lock_bh(&sk->sk_callback_lock); queue = sk->sk_user_data; - if (!queue) + if (!nvmet_tcp_get_queue(queue)) goto done; switch (sk->sk_state) {@@ -1562,6 +1596,7 @@ static void nvmet_tcp_state_change(struct sock *sk) pr_warn("queue %d unhandled state %d\n", queue->idx, sk->sk_state); } + nvmet_tcp_put_queue(queue);done:
read_unlock_bh(&sk->sk_callback_lock);
}drivers/nvme/target/tcp.c +0
-0@@ -1582,6 +1617,9 @@ static int nvmet_tcp_set_queue_sock(struct
Quoted 9 lines
drivers/nvme/target/tcp.c static int nvmet_tcp_set_queue_sock(struct nvmet_tcp_queue *queue) · line 1622
Quoted 6 lines
drivers/nvme/target/tcp.c +3
-0nvmet_tcp_queue *queue) if (ret < 0) return ret; + if (unlikely(!nvmet_tcp_get_queue(queue))) + return -ECONNRESET; +Can this actually fail?
Yes. The socket might have been closed before this call. Unlikely, I know, but still...
Quoted 32 lines
Quoted 32 lines
drivers/nvme/target/tcp.c +1
-0/** Cleanup whatever is sitting in the TCP transmit queue on socket * close. This is done to prevent stale data from being sent should @@ -1603,6 +1641,7 @@ static int nvmet_tcp_set_queue_sock(structnvmet_tcp_queue *queue) * If the socket is already closing, don't even start * consuming it */ + nvmet_tcp_put_queue(queue); ret = -ENOTCONN; } else {
sock->sk->sk_user_data= queue;drivers/nvme/target/tcp.c +4
-4@@ -1636,6 +1675,7 @@ static void nvmet_tcp_alloc_queue(structnvmet_tcp_port *port, INIT_WORK(&queue->release_work, nvmet_tcp_release_queue_work); INIT_WORK(&queue->io_work, nvmet_tcp_io_work); + kref_init(&queue->kref); queue->sock = newsock; queue->port = port; queue->nr_cmds = 0;@@ -1672,15 +1712,15 @@ static void nvmet_tcp_alloc_queue(structnvmet_tcp_port *port, mutex_unlock(&nvmet_tcp_queue_mutex); ret = nvmet_tcp_set_queue_sock(queue); - if (ret) - goto out_destroy_sq; + if (!ret) + return; - return; -out_destroy_sq: + queue->state = NVMET_TCP_Q_DISCONNECTING; mutex_lock(&nvmet_tcp_queue_mutex);
Quoted 8 lines
Quoted 8 lines
drivers/nvme/target/tcp.c +2
-1list_del_init(&queue->queue_list); mutex_unlock(&nvmet_tcp_queue_mutex); - nvmet_sq_destroy(&queue->nvme_sq); + nvmet_tcp_put_queue(queue); + return;
out_free_connect:
nvmet_tcp_free_cmd(&queue->connect);
out_ida_remove:
But thanks for the review!
Cheers,
Hannes
Signature · 5 lines
Dr. Hannes Reinecke Kernel Storage Architect [email protected] +49 911 74053 688 SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg HRB 36809 (AG Nürnberg), Geschäftsführer: Ivo Totev, Andrew Myers, Andrew McDonald, Martje Boudien Moerman
Sagi Grimberg
maintainerRe: [PATCH 14/17] nvmet-tcp: reference counting for queues
Quoted 144 lines · Hannes Reinecke
Quoted 41 lines · Sagi Grimberg
Quoted 38 lines · Hannes Reinecke
The 'queue' structure is referenced from various places and used as an argument of asynchronous functions, so it's really hard to figure out if the queue is still valid when the asynchronous function triggers.
So add reference counting to validate the queue structure.Signed-off-by Hannes Reinecke <[email protected]>---
drivers/nvme/target/tcp.c| 74 ++++++++++++++++++++++++++++++---------1 file changed, 57 insertions(+), 17 deletions(-)drivers/nvme/target/tcp.c +7
-0@@ -127,6 +127,7 @@ enum nvmet_tcp_queue_state { }; struct nvmet_tcp_queue { + struct kref kref; struct socket *sock; struct nvmet_tcp_port *port; struct work_struct io_work;@@ -192,6 +193,9 @@ static struct workqueue_struct *nvmet_tcp_wq; static const struct nvmet_fabrics_ops nvmet_tcp_ops; static void nvmet_tcp_free_cmd(struct nvmet_tcp_cmd *c); static void nvmet_tcp_free_cmd_buffers(struct nvmet_tcp_cmd *cmd); +static int nvmet_tcp_get_queue (struct nvmet_tcp_queue *queue); +static void nvmet_tcp_put_queue (struct nvmet_tcp_queue *queue); +static void nvmet_tcp_data_ready (struct sock *sk); static inline u16 nvmet_tcp_cmd_tag(struct nvmet_tcp_queue *queue, struct nvmet_tcp_cmd *cmd)@@ -1437,11 +1441,21 @@ static void nvmet_tcp_restore_socket_callbacks(struct nvmet_tcp_queue *queue) struct socket *sock = queue->sock; write_lock_bh(&sock->sk->sk_callback_lock); + /* +* Check if nvmet_tcp_set_queue_sock() has been called; +* if not the queue reference has not been increased Quoted 9 lines
drivers/nvme/target/tcp.c static void nvmet_tcp_restore_socket_callbacks(struct nvmet_tcp_queue *queue) · line 1452
Quoted 6 lines
drivers/nvme/target/tcp.c +6
-0+* and we're getting an refcount error on exit. +*/ + if (sock->sk->sk_data_ready != nvmet_tcp_data_ready) { + write_unlock_bh(&sock->sk->sk_callback_lock); + return; + }This is really ugly I think.
Me too, but what would be the alternative?
Quoted 50 lines
Quoted 50 lines
drivers/nvme/target/tcp.c +25
-0sock->sk->sk_data_ready = queue->data_ready; sock->sk->sk_state_change = queue->state_change; sock->sk->sk_write_space = queue->write_space; sock->sk->sk_user_data = NULL; write_unlock_bh(&sock->sk->sk_callback_lock); + nvmet_tcp_put_queue(queue); } static void nvmet_tcp_uninit_data_in_cmds(struct nvmet_tcp_queue *queue)@@ -1474,6 +1488,30 @@ static void nvmet_tcp_free_cmd_data_in_buffers(struct nvmet_tcp_queue *queue) nvmet_tcp_free_cmd_buffers(&queue->connect); } +static void nvmet_tcp_release_queue_final (struct kref *kref) +{ + struct nvmet_tcp_queue *queue = container_of(kref, struct nvmet_tcp_queue, kref); + + WARN_ON(queue->state != NVMET_TCP_Q_DISCONNECTING); + nvmet_tcp_free_cmds(queue); + ida_free(&nvmet_tcp_queue_ida, queue->idx); + /* ->sock will be released by fput() */ + fput(queue->sock->file); + kfree(queue); +} + +static int nvmet_tcp_get_queue (struct nvmet_tcp_queue *queue) +{ + if (!queue) + return 0; + return kref_get_unless_zero(&queue->kref); +} + +static void nvmet_tcp_put_queue (struct nvmet_tcp_queue *queue) +{ + kref_put(&queue->kref, nvmet_tcp_release_queue_final); +} + static void nvmet_tcp_release_queue_work(struct work_struct *w) {struct page *page;
drivers/nvme/target/tcp.c +0
-3@@ -1493,15 +1531,11 @@ static void nvmet_tcp_release_queue_work(struct work_struct *w) nvmet_sq_destroy(&queue->nvme_sq); cancel_work_sync(&queue->io_work); nvmet_tcp_free_cmd_data_in_buffers(queue); - /* ->sock will be released by fput() */ - fput(queue->sock->file); - nvmet_tcp_free_cmds(queue); if (queue->hdr_digest || queue->data_digest) nvmet_tcp_free_crypto(queue);Quoted 9 lines
drivers/nvme/target/tcp.c static void nvmet_tcp_release_queue_work(struct work_struct *w) · line 1538
Quoted 5 lines
drivers/nvme/target/tcp.c +1
-2- ida_free(&nvmet_tcp_queue_ida, queue->idx); page = virt_to_head_page(queue->pf_cache.va); __page_frag_cache_drain(page, queue->pf_cache.pagecnt_bias); - kfree(queue); + nvmet_tcp_put_queue(queue);What made you pick these vs. the others for before/after the final reference?
I wanted to call '
nvmet_tcp_put_queue()' for a failed allocation innvmet_tcp_alloc_queue(), and at that time the queue itself is not live.nvmet_tcp_release_queue()is only called on a live queue, so using that as the kref ->release() function would either limit it's usefulness or would require me to ensure that all calls in there can be made with a non-initialized argument.Quoted 13 lines
drivers/nvme/target/tcp.c static void nvmet_tcp_data_ready(struct sock *sk) · line 1551
Quoted 10 lines
drivers/nvme/target/tcp.c +3
-1} static void nvmet_tcp_data_ready(struct sock *sk)@@ -1512,8 +1546,10 @@ static void nvmet_tcp_data_ready(struct sock *sk) read_lock_bh(&sk->sk_callback_lock); queue = sk->sk_user_data; - if (likely(queue)) + if (likely(nvmet_tcp_get_queue(queue))) { queue_work_on(queue_cpu(queue), nvmet_tcp_wq, &queue->io_work); + nvmet_tcp_put_queue(queue); + }No... Why?
Quoted 6 lines
The shutdown code should serialize perfectly without this. Why add a kref to the normal I/O path?
I thought we'd simply move
release_workto do akref_putand take an extra reference when we fire the tls handshake...Because I feel ever so slightly unsure about using the
sk_user_dataargument. This function is completely asynchronous, and I can't really see how we can ensure thatsk_user_datareferences a valid object. (If it were valid, why would we need to check for !queue ?)
Quoted 4 lines
If you have lifetime guarantees that the kref isn't required, by all means, please tell me, and we can drop the kref thing here. But I guess that would imply to _not_ having to check for (!queue) which is fine by me, too.
Something doesn't add up here.
What I think you need to do, is add another reference just for the tls handshake.
Then in the timeout handler you call tls_handshake_cancel():
- if you got %true back, you drop the reference and schedule a release work
- if you got %false back, you simply ignore the timeout because the .ta_done() was already triggered.
- in .ta_done() you drop the reference, cancel the timeout work and then continue or remove based on the status.
btw in the queue release you should call tls_handshake_cancel() as well.
PATCH v8 15/17 nvmet: Set 'TREQ' to 'required' when TLS is enabled
In reply to Hannes ReineckeThe current implementation does not support secure concatenation, so 'TREQ' is always set to 'required' when TLS is enabled.
--- drivers/nvme/target/configfs.c | 12 ++++++++++++ 1 file changed, 12 insertions(+)
drivers/nvme/target/configfs.c +12 -0
@@ -376,6 +376,7 @@ static ssize_t nvmet_addr_tsas_store(struct config_item *item, const char *page, size_t count) {struct nvmet_port *port =to_nvmet_port (item); +u8 treq =nvmet_port_disc_addr_treq_mask (port);u8 sectype;int i;
@@ -397,6 +398,17 @@ static ssize_t nvmet_addr_tsas_store(struct config_item *item, found:nvmet_port_init_tsas_tcp (port, sectype); +/* +* The TLS implementation currently does not support +* secure concatenation, so TREQ is always set to 'required' +* if TLS is enabled. +*/ +if (sectype ==NVMF_TCP_SECTYPE_TLS13 ) { + treq |=NVMF_TREQ_REQUIRED ; + }else { + treq |=NVMF_TREQ_NOT_SPECIFIED ; + } + port->disc_addr.treq = treq;return count; }
Signature · 1 line
2.35.3
Re: [PATCH 15/17] nvmet: Set 'TREQ' to 'required' when TLS is enabled
Looks fine,
PATCH v8 16/17 nvmet-tcp: enable TLS handshake upcall
In reply to Hannes ReineckeAdd functions to start the TLS handshake upcall when the TCP TSAS sectype is set to 'tls1.3' and add a config option NVME_TARGET_TCP_TLS.
And move the 'state' lock to an irq-safe lock as the 'done' callback from the TLS handshake will be running in interrupt context.
--- drivers/nvme/target/Kconfig | 15 ++++ drivers/nvme/target/configfs.c | 21 +++++ drivers/nvme/target/nvmet.h | 1 + drivers/nvme/target/tcp.c | 139 ++++++++++++++++++++++++++++++++- 4 files changed, 173 insertions(+), 3 deletions(-)
drivers/nvme/target/Kconfig +15 -0
@@ -84,6 +84,21 @@ config NVME_TARGET_TCP If unsure, say N. +config NVME_TARGET_TCP_TLS + bool "NVMe over Fabrics TCP target TLS encryption support" + depends on NVME_TARGET_TCP + select NVME_COMMON + select NVME_KEYRING + select NET_HANDSHAKE + select KEYS + help + Enables TLS encryption for the NVMe TCP target using the netlink handshake API. + + The TLS handshake daemon is availble at + https://github.com/oracle/ktls-utils. + + If unsure, say N. + config NVME_TARGET_AUTH bool "NVMe over Fabrics In-band Authentication support" depends on NVME_TARGET
drivers/nvme/target/configfs.c +21 -0
@@ -15,6 +15,7 @@#ifdef CONFIG_NVME_TARGET_AUTH #include <linux/nvme-auth.h> #endif +#include <linux/nvme-keyring.h> #include <crypto/hash.h> #include <crypto/kpp.h>
@@ -397,6 +398,17 @@ static ssize_t nvmet_addr_tsas_store(struct config_item *item,return -EINVAL ; found: +if (sectype ==NVMF_TCP_SECTYPE_TLS13 ) { +if (!IS_ENABLED (CONFIG_NVME_TARGET_TCP_TLS )) { +pr_err ("TLS is not supported\n" ); +return -EINVAL ; + } +if (!port->keyring) { +pr_err ("TLS keyring not configured\n" ); +return -EINVAL ; + } + } +nvmet_port_init_tsas_tcp (port, sectype);/* * The TLS implementation currently does not support
@@ -1815,6 +1827,7 @@ static void nvmet_port_release(struct config_item *item)flush_workqueue (nvmet_wq);list_del (&port->global_entry); +key_put (port->keyring);kfree (port->ana_state);kfree (port); }
@@ -1864,6 +1877,14 @@ static struct config_group *nvmet_ports_make(struct config_group *group,return ERR_PTR (-ENOMEM ); } +if (nvme_keyring_id ()) { + port->keyring =key_lookup (nvme_keyring_id ()); +if (IS_ERR (port->keyring)) { +pr_warn ("NVMe keyring not available, disabling TLS\n" ); + port->keyring =NULL ; + } + } +for (i =1 ; i <=NVMET_MAX_ANAGRPS ; i++) {if (i ==NVMET_DEFAULT_ANA_GRPID ) port->ana_state[1 ] =NVME_ANA_OPTIMIZED ;
drivers/nvme/target/nvmet.h +1 -0
@@ -158,6 +158,7 @@ struct nvmet_port {struct config_group ana_groups_group;struct nvmet_ana_group ana_default_group;enum nvme_ana_state *ana_state; +struct key *keyring;void *priv;bool enabled;int inline_data_size;
drivers/nvme/target/tcp.c +136 -3
200 lines not on this page · raw message
Signature · 1 line
2.35.3
PATCH v8 17/17 nvmet-tcp: control messages for recvmsg()
In reply to Hannes ReineckekTLS requires control messages for recvmsg() to relay any out-of-band TLS messages (eg TLS alerts) to the caller.
--- drivers/nvme/target/tcp.c | 94 +++++++++++++++++++++++++++++++++------ 1 file changed, 81 insertions(+), 13 deletions(-)
drivers/nvme/target/tcp.c +81 -13
@@ -14,6 +14,7 @@#include <net/sock.h> #include <net/tcp.h> #include <net/tls.h> +#include <net/tls_prot.h> #include <net/handshake.h> #include <linux/inet.h> #include <linux/llist.h>
@@ -118,6 +119,7 @@ struct nvmet_tcp_cmd {u32 pdu_len;u32 pdu_recv;int sg_idx; +char recv_cbuf[CMSG_LEN (sizeof (char ))];struct msghdr recv_msg;struct bio_vec *iov;u32 flags;
@@ -1122,20 +1124,65 @@ static inline bool nvmet_tcp_pdu_valid(u8 type)return false; } +static int nvmet_tcp_tls_record_ok (struct nvmet_tcp_queue *queue, +struct msghdr *msg,char *cbuf) +{ +struct cmsghdr *cmsg = (struct cmsghdr *)cbuf; +u8 ctype, level, description; +int ret =0 ; + + ctype =tls_get_record_type (queue->sock->sk, cmsg); +switch (ctype) { +case 0 : +break ; +case TLS_RECORD_TYPE_DATA : +break ; +case TLS_RECORD_TYPE_ALERT : +tls_alert_recv (queue->sock->sk, msg, &level, &description); +if (level ==TLS_ALERT_LEVEL_FATAL ) { +pr_err ("queue %d: TLS Alert desc %u\n" , + queue->idx, description); + ret = -ENOTCONN ; + }else { +pr_warn ("queue %d: TLS Alert desc %u\n" , + queue->idx, description); + ret = -EAGAIN ; + } +break ; +default : +/* discard this record type */ +pr_err ("queue %d: TLS record %d unhandled\n" , + queue->idx, ctype); + ret = -EAGAIN ; +break ; + } +return ret; +} +static int nvmet_tcp_try_recv_pdu (struct nvmet_tcp_queue *queue) {struct nvme_tcp_hdr *hdr = &queue->pdu.cmd.hdr; -int len; +int len, ret;struct kvec iov; +char cbuf[CMSG_LEN (sizeof (char ))] = {};struct msghdr msg = { .msg_flags =MSG_DONTWAIT }; recv: iov.iov_base = (void *)&queue->pdu + queue->offset; iov.iov_len = queue->left; +if (queue->tls_pskid) { + msg.msg_control = cbuf; + msg.msg_controllen =sizeof (cbuf); + } len =kernel_recvmsg (queue->sock, &msg, &iov,1 , iov.iov_len, msg.msg_flags);if (unlikely (len <0 ))return len; +if (queue->tls_pskid) { + ret =nvmet_tcp_tls_record_ok (queue, &msg, cbuf); +if (ret <0 ) +return ret; + } queue->offset += len; queue->left -= len;
@@ -1188,16 +1235,22 @@ static void nvmet_tcp_prep_recv_ddgst(struct nvmet_tcp_cmd *cmd)static int nvmet_tcp_try_recv_data (struct nvmet_tcp_queue *queue) {struct nvmet_tcp_cmd *cmd = queue->cmd; -int ret; +int len, ret;while (msg_data_left (&cmd->recv_msg)) { - ret =sock_recvmsg (cmd->queue->sock, &cmd->recv_msg, + len =sock_recvmsg (cmd->queue->sock, &cmd->recv_msg, cmd->recv_msg.msg_flags); -if (ret <=0 ) -return ret; +if (len <=0 ) +return len; +if (queue->tls_pskid) { + ret =nvmet_tcp_tls_record_ok (cmd->queue, + &cmd->recv_msg, cmd->recv_cbuf); +if (ret <0 ) +return ret; + } - cmd->pdu_recv += ret; - cmd->rbytes_done += ret; + cmd->pdu_recv += len; + cmd->rbytes_done += len; }if (queue->data_digest) {
@@ -1215,20 +1268,30 @@ static int nvmet_tcp_try_recv_data(struct nvmet_tcp_queue *queue)static int nvmet_tcp_try_recv_ddgst (struct nvmet_tcp_queue *queue) {struct nvmet_tcp_cmd *cmd = queue->cmd; -int ret; +int ret, len; +char cbuf[CMSG_LEN (sizeof (char ))] = {};struct msghdr msg = { .msg_flags =MSG_DONTWAIT };struct kvec iov = { .iov_base = (void *)&cmd->recv_ddgst + queue->offset, .iov_len = queue->left }; - ret =kernel_recvmsg (queue->sock, &msg, &iov,1 , +if (queue->tls_pskid) { + msg.msg_control = cbuf; + msg.msg_controllen =sizeof (cbuf); + } + len =kernel_recvmsg (queue->sock, &msg, &iov,1 , iov.iov_len, msg.msg_flags); -if (unlikely (ret <0 )) -return ret; +if (unlikely (len <0 )) +return len; +if (queue->tls_pskid) { + ret =nvmet_tcp_tls_record_ok (queue, &msg, cbuf); +if (ret <0 ) +return ret; + } - queue->offset += ret; - queue->left -= ret; + queue->offset += len; + queue->left -= len;if (queue->left)return -EAGAIN ;
@@ -1396,6 +1459,10 @@ static int nvmet_tcp_alloc_cmd(struct nvmet_tcp_queue *queue,if (!c->r2t_pdu) goto out_free_data; +if (queue->state ==NVMET_TCP_Q_TLS_HANDSHAKE ) { + c->recv_msg.msg_control = c->recv_cbuf; + c->recv_msg.msg_controllen =sizeof (c->recv_cbuf); + } c->recv_msg.msg_flags =MSG_DONTWAIT |MSG_NOSIGNAL ;list_add_tail (&c->entry, &queue->free_list);
@@ -1705,6 +1772,7 @@ static void nvmet_tcp_tls_handshake_done(void *data, int status, }if (!status) queue->tls_pskid = peerid; + queue->state =NVMET_TCP_Q_CONNECTING ;spin_unlock_bh (&queue->state_lock);
Signature · 1 line
2.35.3
Sagi Grimberg
maintainerRe: [PATCH 17/17] nvmet-tcp: control messages for recvmsg()
drivers/nvme/target/tcp.c static void nvmet_tcp_tls_handshake_done(void *data, int status, · line 1778
Quoted 9 lines
drivers/nvme/target/tcp.c +1
-0@@ -1705,6 +1772,7 @@ static void nvmet_tcp_tls_handshake_done(void *data, int status, }if (!status) queue->tls_pskid = peerid; + queue->state =NVMET_TCP_Q_CONNECTING ;spin_unlock_bh (&queue->state_lock);
Unneeded hunk here.