1 /* SPDX-License-Identifier: GPL-2.0 */
3 * Copyright (c) 2015-2016 HGST, a Western Digital Company.
9 #include <linux/dma-mapping.h>
10 #include <linux/types.h>
11 #include <linux/device.h>
12 #include <linux/kref.h>
13 #include <linux/percpu-refcount.h>
14 #include <linux/list.h>
15 #include <linux/mutex.h>
16 #include <linux/uuid.h>
17 #include <linux/nvme.h>
18 #include <linux/configfs.h>
19 #include <linux/rcupdate.h>
20 #include <linux/blkdev.h>
21 #include <linux/radix-tree.h>
22 #include <linux/t10-pi.h>
24 #define NVMET_DEFAULT_VS NVME_VS(1, 3, 0)
26 #define NVMET_ASYNC_EVENTS 4
27 #define NVMET_ERROR_LOG_SLOTS 128
28 #define NVMET_NO_ERROR_LOC ((u16)-1)
29 #define NVMET_DEFAULT_CTRL_MODEL "Linux"
30 #define NVMET_MN_MAX_SIZE 40
31 #define NVMET_SN_MAX_SIZE 20
34 * Supported optional AENs:
36 #define NVMET_AEN_CFG_OPTIONAL \
37 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_ANA_CHANGE)
38 #define NVMET_DISC_AEN_CFG_OPTIONAL \
39 (NVME_AEN_CFG_DISC_CHANGE)
42 * Plus mandatory SMART AENs (we'll never send them, but allow enabling them):
44 #define NVMET_AEN_CFG_ALL \
45 (NVME_SMART_CRIT_SPARE | NVME_SMART_CRIT_TEMPERATURE | \
46 NVME_SMART_CRIT_RELIABILITY | NVME_SMART_CRIT_MEDIA | \
47 NVME_SMART_CRIT_VOLATILE_MEMORY | NVMET_AEN_CFG_OPTIONAL)
49 /* Helper Macros when NVMe error is NVME_SC_CONNECT_INVALID_PARAM
50 * The 16 bit shift is to set IATTR bit to 1, which means offending
51 * offset starts in the data section of connect()
53 #define IPO_IATTR_CONNECT_DATA(x) \
54 (cpu_to_le32((1 << 16) | (offsetof(struct nvmf_connect_data, x))))
55 #define IPO_IATTR_CONNECT_SQE(x) \
56 (cpu_to_le32(offsetof(struct nvmf_connect_command, x)))
59 struct percpu_ref ref;
60 struct block_device *bdev;
72 struct nvmet_subsys *subsys;
73 const char *device_path;
75 struct config_group device_group;
76 struct config_group group;
78 struct completion disable_done;
80 struct kmem_cache *bvec_cache;
83 struct pci_dev *p2p_dev;
89 static inline struct nvmet_ns *to_nvmet_ns(struct config_item *item)
91 return container_of(to_config_group(item), struct nvmet_ns, group);
94 static inline struct device *nvmet_ns_dev(struct nvmet_ns *ns)
96 return ns->bdev ? disk_to_dev(ns->bdev->bd_disk) : NULL;
105 struct nvmet_ctrl *ctrl;
106 struct percpu_ref ref;
111 struct completion free_done;
112 struct completion confirm_done;
115 struct nvmet_ana_group {
116 struct config_group group;
117 struct nvmet_port *port;
121 static inline struct nvmet_ana_group *to_ana_group(struct config_item *item)
123 return container_of(to_config_group(item), struct nvmet_ana_group,
128 * struct nvmet_port - Common structure to keep port
129 * information for the target.
130 * @entry: Entry into referrals or transport list.
131 * @disc_addr: Address information is stored in a format defined
132 * for a discovery log page entry.
133 * @group: ConfigFS group for this element's folder.
134 * @priv: Private data for the transport.
137 struct list_head entry;
138 struct nvmf_disc_rsp_page_entry disc_addr;
139 struct config_group group;
140 struct config_group subsys_group;
141 struct list_head subsystems;
142 struct config_group referrals_group;
143 struct list_head referrals;
144 struct list_head global_entry;
145 struct config_group ana_groups_group;
146 struct nvmet_ana_group ana_default_group;
147 enum nvme_ana_state *ana_state;
150 int inline_data_size;
151 const struct nvmet_fabrics_ops *tr_ops;
155 static inline struct nvmet_port *to_nvmet_port(struct config_item *item)
157 return container_of(to_config_group(item), struct nvmet_port,
161 static inline struct nvmet_port *ana_groups_to_port(
162 struct config_item *item)
164 return container_of(to_config_group(item), struct nvmet_port,
169 struct nvmet_subsys *subsys;
170 struct nvmet_sq **sqs;
183 struct nvmet_port *port;
186 unsigned long aen_masked;
187 struct nvmet_req *async_event_cmds[NVMET_ASYNC_EVENTS];
188 unsigned int nr_async_event_cmds;
189 struct list_head async_events;
190 struct work_struct async_event_work;
192 struct list_head subsys_entry;
194 struct delayed_work ka_work;
195 struct work_struct fatal_err_work;
197 const struct nvmet_fabrics_ops *ops;
199 __le32 *changed_ns_list;
202 char subsysnqn[NVMF_NQN_FIELD_LEN];
203 char hostnqn[NVMF_NQN_FIELD_LEN];
205 struct device *p2p_client;
206 struct radix_tree_root p2p_ns_map;
208 spinlock_t error_lock;
210 struct nvme_error_slot slots[NVMET_ERROR_LOG_SLOTS];
214 struct nvmet_subsys {
215 enum nvme_subsys_type type;
220 struct xarray namespaces;
221 unsigned int nr_namespaces;
226 struct list_head ctrls;
228 struct list_head hosts;
234 char serial[NVMET_SN_MAX_SIZE];
235 bool subsys_discovered;
239 struct config_group group;
241 struct config_group namespaces_group;
242 struct config_group allowed_hosts_group;
246 #ifdef CONFIG_NVME_TARGET_PASSTHRU
247 struct nvme_ctrl *passthru_ctrl;
248 char *passthru_ctrl_path;
249 struct config_group passthru_group;
250 unsigned int admin_timeout;
251 unsigned int io_timeout;
252 #endif /* CONFIG_NVME_TARGET_PASSTHRU */
254 #ifdef CONFIG_BLK_DEV_ZONED
256 #endif /* CONFIG_BLK_DEV_ZONED */
259 static inline struct nvmet_subsys *to_subsys(struct config_item *item)
261 return container_of(to_config_group(item), struct nvmet_subsys, group);
264 static inline struct nvmet_subsys *namespaces_to_subsys(
265 struct config_item *item)
267 return container_of(to_config_group(item), struct nvmet_subsys,
272 struct config_group group;
275 static inline struct nvmet_host *to_host(struct config_item *item)
277 return container_of(to_config_group(item), struct nvmet_host, group);
280 static inline char *nvmet_host_name(struct nvmet_host *host)
282 return config_item_name(&host->group.cg_item);
285 struct nvmet_host_link {
286 struct list_head entry;
287 struct nvmet_host *host;
290 struct nvmet_subsys_link {
291 struct list_head entry;
292 struct nvmet_subsys *subsys;
296 struct nvmet_fabrics_ops {
297 struct module *owner;
301 #define NVMF_KEYED_SGLS (1 << 0)
302 #define NVMF_METADATA_SUPPORTED (1 << 1)
303 void (*queue_response)(struct nvmet_req *req);
304 int (*add_port)(struct nvmet_port *port);
305 void (*remove_port)(struct nvmet_port *port);
306 void (*delete_ctrl)(struct nvmet_ctrl *ctrl);
307 void (*disc_traddr)(struct nvmet_req *req,
308 struct nvmet_port *port, char *traddr);
309 u16 (*install_queue)(struct nvmet_sq *nvme_sq);
310 void (*discovery_chg)(struct nvmet_port *port);
311 u8 (*get_mdts)(const struct nvmet_ctrl *ctrl);
314 #define NVMET_MAX_INLINE_BIOVEC 8
315 #define NVMET_MAX_INLINE_DATA_LEN NVMET_MAX_INLINE_BIOVEC * PAGE_SIZE
318 struct nvme_command *cmd;
319 struct nvme_completion *cqe;
323 struct scatterlist *sg;
324 struct scatterlist *metadata_sg;
325 struct bio_vec inline_bvec[NVMET_MAX_INLINE_BIOVEC];
328 struct bio inline_bio;
333 struct bio_vec *bvec;
334 struct work_struct work;
337 struct bio inline_bio;
339 struct work_struct work;
342 #ifdef CONFIG_BLK_DEV_ZONED
344 struct bio inline_bio;
345 struct work_struct zmgmt_work;
347 #endif /* CONFIG_BLK_DEV_ZONED */
351 /* data length as parsed from the SGL descriptor: */
355 struct nvmet_port *port;
357 void (*execute)(struct nvmet_req *req);
358 const struct nvmet_fabrics_ops *ops;
360 struct pci_dev *p2p_dev;
361 struct device *p2p_client;
366 extern struct workqueue_struct *buffered_io_wq;
367 extern struct workqueue_struct *zbd_wq;
369 static inline void nvmet_set_result(struct nvmet_req *req, u32 result)
371 req->cqe->result.u32 = cpu_to_le32(result);
375 * NVMe command writes actually are DMA reads for us on the target side.
377 static inline enum dma_data_direction
378 nvmet_data_dir(struct nvmet_req *req)
380 return nvme_is_write(req->cmd) ? DMA_FROM_DEVICE : DMA_TO_DEVICE;
383 struct nvmet_async_event {
384 struct list_head entry;
390 static inline void nvmet_clear_aen_bit(struct nvmet_req *req, u32 bn)
392 int rae = le32_to_cpu(req->cmd->common.cdw10) & 1 << 15;
395 clear_bit(bn, &req->sq->ctrl->aen_masked);
398 static inline bool nvmet_aen_bit_disabled(struct nvmet_ctrl *ctrl, u32 bn)
400 if (!(READ_ONCE(ctrl->aen_enabled) & (1 << bn)))
402 return test_and_set_bit(bn, &ctrl->aen_masked);
405 void nvmet_get_feat_kato(struct nvmet_req *req);
406 void nvmet_get_feat_async_event(struct nvmet_req *req);
407 u16 nvmet_set_feat_kato(struct nvmet_req *req);
408 u16 nvmet_set_feat_async_event(struct nvmet_req *req, u32 mask);
409 void nvmet_execute_async_event(struct nvmet_req *req);
410 void nvmet_start_keep_alive_timer(struct nvmet_ctrl *ctrl);
411 void nvmet_stop_keep_alive_timer(struct nvmet_ctrl *ctrl);
413 u16 nvmet_parse_connect_cmd(struct nvmet_req *req);
414 void nvmet_bdev_set_limits(struct block_device *bdev, struct nvme_id_ns *id);
415 u16 nvmet_bdev_parse_io_cmd(struct nvmet_req *req);
416 u16 nvmet_file_parse_io_cmd(struct nvmet_req *req);
417 u16 nvmet_bdev_zns_parse_io_cmd(struct nvmet_req *req);
418 u16 nvmet_parse_admin_cmd(struct nvmet_req *req);
419 u16 nvmet_parse_discovery_cmd(struct nvmet_req *req);
420 u16 nvmet_parse_fabrics_cmd(struct nvmet_req *req);
422 bool nvmet_req_init(struct nvmet_req *req, struct nvmet_cq *cq,
423 struct nvmet_sq *sq, const struct nvmet_fabrics_ops *ops);
424 void nvmet_req_uninit(struct nvmet_req *req);
425 bool nvmet_check_transfer_len(struct nvmet_req *req, size_t len);
426 bool nvmet_check_data_len_lte(struct nvmet_req *req, size_t data_len);
427 void nvmet_req_complete(struct nvmet_req *req, u16 status);
428 int nvmet_req_alloc_sgls(struct nvmet_req *req);
429 void nvmet_req_free_sgls(struct nvmet_req *req);
431 void nvmet_execute_set_features(struct nvmet_req *req);
432 void nvmet_execute_get_features(struct nvmet_req *req);
433 void nvmet_execute_keep_alive(struct nvmet_req *req);
435 void nvmet_cq_setup(struct nvmet_ctrl *ctrl, struct nvmet_cq *cq, u16 qid,
437 void nvmet_sq_setup(struct nvmet_ctrl *ctrl, struct nvmet_sq *sq, u16 qid,
439 void nvmet_sq_destroy(struct nvmet_sq *sq);
440 int nvmet_sq_init(struct nvmet_sq *sq);
442 void nvmet_ctrl_fatal_error(struct nvmet_ctrl *ctrl);
444 void nvmet_update_cc(struct nvmet_ctrl *ctrl, u32 new);
445 u16 nvmet_alloc_ctrl(const char *subsysnqn, const char *hostnqn,
446 struct nvmet_req *req, u32 kato, struct nvmet_ctrl **ctrlp);
447 struct nvmet_ctrl *nvmet_ctrl_find_get(const char *subsysnqn,
448 const char *hostnqn, u16 cntlid,
449 struct nvmet_req *req);
450 void nvmet_ctrl_put(struct nvmet_ctrl *ctrl);
451 u16 nvmet_check_ctrl_status(struct nvmet_req *req);
453 struct nvmet_subsys *nvmet_subsys_alloc(const char *subsysnqn,
454 enum nvme_subsys_type type);
455 void nvmet_subsys_put(struct nvmet_subsys *subsys);
456 void nvmet_subsys_del_ctrls(struct nvmet_subsys *subsys);
458 u16 nvmet_req_find_ns(struct nvmet_req *req);
459 void nvmet_put_namespace(struct nvmet_ns *ns);
460 int nvmet_ns_enable(struct nvmet_ns *ns);
461 void nvmet_ns_disable(struct nvmet_ns *ns);
462 struct nvmet_ns *nvmet_ns_alloc(struct nvmet_subsys *subsys, u32 nsid);
463 void nvmet_ns_free(struct nvmet_ns *ns);
465 void nvmet_send_ana_event(struct nvmet_subsys *subsys,
466 struct nvmet_port *port);
467 void nvmet_port_send_ana_event(struct nvmet_port *port);
469 int nvmet_register_transport(const struct nvmet_fabrics_ops *ops);
470 void nvmet_unregister_transport(const struct nvmet_fabrics_ops *ops);
472 void nvmet_port_del_ctrls(struct nvmet_port *port,
473 struct nvmet_subsys *subsys);
475 int nvmet_enable_port(struct nvmet_port *port);
476 void nvmet_disable_port(struct nvmet_port *port);
478 void nvmet_referral_enable(struct nvmet_port *parent, struct nvmet_port *port);
479 void nvmet_referral_disable(struct nvmet_port *parent, struct nvmet_port *port);
481 u16 nvmet_copy_to_sgl(struct nvmet_req *req, off_t off, const void *buf,
483 u16 nvmet_copy_from_sgl(struct nvmet_req *req, off_t off, void *buf,
485 u16 nvmet_zero_sgl(struct nvmet_req *req, off_t off, size_t len);
487 u32 nvmet_get_log_page_len(struct nvme_command *cmd);
488 u64 nvmet_get_log_page_offset(struct nvme_command *cmd);
490 extern struct list_head *nvmet_ports;
491 void nvmet_port_disc_changed(struct nvmet_port *port,
492 struct nvmet_subsys *subsys);
493 void nvmet_subsys_disc_changed(struct nvmet_subsys *subsys,
494 struct nvmet_host *host);
495 void nvmet_add_async_event(struct nvmet_ctrl *ctrl, u8 event_type,
496 u8 event_info, u8 log_page);
498 #define NVMET_QUEUE_SIZE 1024
499 #define NVMET_NR_QUEUES 128
500 #define NVMET_MAX_CMD NVMET_QUEUE_SIZE
503 * Nice round number that makes a list of nsids fit into a page.
504 * Should become tunable at some point in the future.
506 #define NVMET_MAX_NAMESPACES 1024
509 * 0 is not a valid ANA group ID, so we start numbering at 1.
511 * ANA Group 1 exists without manual intervention, has namespaces assigned to it
512 * by default, and is available in an optimized state through all ports.
514 #define NVMET_MAX_ANAGRPS 128
515 #define NVMET_DEFAULT_ANA_GRPID 1
518 #define NVMET_DISC_KATO_MS 120000
520 int __init nvmet_init_configfs(void);
521 void __exit nvmet_exit_configfs(void);
523 int __init nvmet_init_discovery(void);
524 void nvmet_exit_discovery(void);
526 extern struct nvmet_subsys *nvmet_disc_subsys;
527 extern struct rw_semaphore nvmet_config_sem;
529 extern u32 nvmet_ana_group_enabled[NVMET_MAX_ANAGRPS + 1];
530 extern u64 nvmet_ana_chgcnt;
531 extern struct rw_semaphore nvmet_ana_sem;
533 bool nvmet_host_allowed(struct nvmet_subsys *subsys, const char *hostnqn);
535 int nvmet_bdev_ns_enable(struct nvmet_ns *ns);
536 int nvmet_file_ns_enable(struct nvmet_ns *ns);
537 void nvmet_bdev_ns_disable(struct nvmet_ns *ns);
538 void nvmet_file_ns_disable(struct nvmet_ns *ns);
539 u16 nvmet_bdev_flush(struct nvmet_req *req);
540 u16 nvmet_file_flush(struct nvmet_req *req);
541 void nvmet_ns_changed(struct nvmet_subsys *subsys, u32 nsid);
542 void nvmet_bdev_ns_revalidate(struct nvmet_ns *ns);
543 int nvmet_file_ns_revalidate(struct nvmet_ns *ns);
544 void nvmet_ns_revalidate(struct nvmet_ns *ns);
545 u16 blk_to_nvme_status(struct nvmet_req *req, blk_status_t blk_sts);
547 bool nvmet_bdev_zns_enable(struct nvmet_ns *ns);
548 void nvmet_execute_identify_cns_cs_ctrl(struct nvmet_req *req);
549 void nvmet_execute_identify_cns_cs_ns(struct nvmet_req *req);
550 void nvmet_bdev_execute_zone_mgmt_recv(struct nvmet_req *req);
551 void nvmet_bdev_execute_zone_mgmt_send(struct nvmet_req *req);
552 void nvmet_bdev_execute_zone_append(struct nvmet_req *req);
554 static inline u32 nvmet_rw_data_len(struct nvmet_req *req)
556 return ((u32)le16_to_cpu(req->cmd->rw.length) + 1) <<
557 req->ns->blksize_shift;
560 static inline u32 nvmet_rw_metadata_len(struct nvmet_req *req)
562 if (!IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY))
564 return ((u32)le16_to_cpu(req->cmd->rw.length) + 1) *
565 req->ns->metadata_size;
568 static inline u32 nvmet_dsm_len(struct nvmet_req *req)
570 return (le32_to_cpu(req->cmd->dsm.nr) + 1) *
571 sizeof(struct nvme_dsm_range);
574 static inline struct nvmet_subsys *nvmet_req_subsys(struct nvmet_req *req)
576 return req->sq->ctrl->subsys;
579 #ifdef CONFIG_NVME_TARGET_PASSTHRU
580 void nvmet_passthru_subsys_free(struct nvmet_subsys *subsys);
581 int nvmet_passthru_ctrl_enable(struct nvmet_subsys *subsys);
582 void nvmet_passthru_ctrl_disable(struct nvmet_subsys *subsys);
583 u16 nvmet_parse_passthru_admin_cmd(struct nvmet_req *req);
584 u16 nvmet_parse_passthru_io_cmd(struct nvmet_req *req);
585 static inline struct nvme_ctrl *nvmet_passthru_ctrl(struct nvmet_subsys *subsys)
587 return subsys->passthru_ctrl;
589 #else /* CONFIG_NVME_TARGET_PASSTHRU */
590 static inline void nvmet_passthru_subsys_free(struct nvmet_subsys *subsys)
593 static inline void nvmet_passthru_ctrl_disable(struct nvmet_subsys *subsys)
596 static inline u16 nvmet_parse_passthru_admin_cmd(struct nvmet_req *req)
600 static inline u16 nvmet_parse_passthru_io_cmd(struct nvmet_req *req)
604 static inline struct nvme_ctrl *nvmet_passthru_ctrl(struct nvmet_subsys *subsys)
608 #endif /* CONFIG_NVME_TARGET_PASSTHRU */
610 static inline struct nvme_ctrl *
611 nvmet_req_passthru_ctrl(struct nvmet_req *req)
613 return nvmet_passthru_ctrl(nvmet_req_subsys(req));
616 u16 errno_to_nvme_status(struct nvmet_req *req, int errno);
617 u16 nvmet_report_invalid_opcode(struct nvmet_req *req);
619 /* Convert a 32-bit number to a 16-bit 0's based number */
620 static inline __le16 to0based(u32 a)
622 return cpu_to_le16(max(1U, min(1U << 16, a)) - 1);
625 static inline bool nvmet_ns_has_pi(struct nvmet_ns *ns)
627 if (!IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY))
629 return ns->pi_type && ns->metadata_size == sizeof(struct t10_pi_tuple);
632 static inline __le64 nvmet_sect_to_lba(struct nvmet_ns *ns, sector_t sect)
634 return cpu_to_le64(sect >> (ns->blksize_shift - SECTOR_SHIFT));
637 static inline sector_t nvmet_lba_to_sect(struct nvmet_ns *ns, __le64 lba)
639 return le64_to_cpu(lba) << (ns->blksize_shift - SECTOR_SHIFT);
642 static inline bool nvmet_use_inline_bvec(struct nvmet_req *req)
644 return req->transfer_len <= NVMET_MAX_INLINE_DATA_LEN &&
645 req->sg_cnt <= NVMET_MAX_INLINE_BIOVEC;
648 static inline void nvmet_req_cns_error_complete(struct nvmet_req *req)
650 pr_debug("unhandled identify cns %d on qid %d\n",
651 req->cmd->identify.cns, req->sq->qid);
652 req->error_loc = offsetof(struct nvme_identify, cns);
653 nvmet_req_complete(req, NVME_SC_INVALID_FIELD | NVME_SC_DNR);
656 static inline void nvmet_req_bio_put(struct nvmet_req *req, struct bio *bio)
658 if (bio != &req->b.inline_bio)
662 #endif /* _NVMET_H */