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"
32 * Supported optional AENs:
34 #define NVMET_AEN_CFG_OPTIONAL \
35 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_ANA_CHANGE)
36 #define NVMET_DISC_AEN_CFG_OPTIONAL \
37 (NVME_AEN_CFG_DISC_CHANGE)
40 * Plus mandatory SMART AENs (we'll never send them, but allow enabling them):
42 #define NVMET_AEN_CFG_ALL \
43 (NVME_SMART_CRIT_SPARE | NVME_SMART_CRIT_TEMPERATURE | \
44 NVME_SMART_CRIT_RELIABILITY | NVME_SMART_CRIT_MEDIA | \
45 NVME_SMART_CRIT_VOLATILE_MEMORY | NVMET_AEN_CFG_OPTIONAL)
47 /* Helper Macros when NVMe error is NVME_SC_CONNECT_INVALID_PARAM
48 * The 16 bit shift is to set IATTR bit to 1, which means offending
49 * offset starts in the data section of connect()
51 #define IPO_IATTR_CONNECT_DATA(x) \
52 (cpu_to_le32((1 << 16) | (offsetof(struct nvmf_connect_data, x))))
53 #define IPO_IATTR_CONNECT_SQE(x) \
54 (cpu_to_le32(offsetof(struct nvmf_connect_command, x)))
57 struct percpu_ref ref;
58 struct block_device *bdev;
70 struct nvmet_subsys *subsys;
71 const char *device_path;
73 struct config_group device_group;
74 struct config_group group;
76 struct completion disable_done;
78 struct kmem_cache *bvec_cache;
81 struct pci_dev *p2p_dev;
86 static inline struct nvmet_ns *to_nvmet_ns(struct config_item *item)
88 return container_of(to_config_group(item), struct nvmet_ns, group);
91 static inline struct device *nvmet_ns_dev(struct nvmet_ns *ns)
93 return ns->bdev ? disk_to_dev(ns->bdev->bd_disk) : NULL;
102 struct nvmet_ctrl *ctrl;
103 struct percpu_ref ref;
108 struct completion free_done;
109 struct completion confirm_done;
112 struct nvmet_ana_group {
113 struct config_group group;
114 struct nvmet_port *port;
118 static inline struct nvmet_ana_group *to_ana_group(struct config_item *item)
120 return container_of(to_config_group(item), struct nvmet_ana_group,
125 * struct nvmet_port - Common structure to keep port
126 * information for the target.
127 * @entry: Entry into referrals or transport list.
128 * @disc_addr: Address information is stored in a format defined
129 * for a discovery log page entry.
130 * @group: ConfigFS group for this element's folder.
131 * @priv: Private data for the transport.
134 struct list_head entry;
135 struct nvmf_disc_rsp_page_entry disc_addr;
136 struct config_group group;
137 struct config_group subsys_group;
138 struct list_head subsystems;
139 struct config_group referrals_group;
140 struct list_head referrals;
141 struct list_head global_entry;
142 struct config_group ana_groups_group;
143 struct nvmet_ana_group ana_default_group;
144 enum nvme_ana_state *ana_state;
147 int inline_data_size;
148 const struct nvmet_fabrics_ops *tr_ops;
152 static inline struct nvmet_port *to_nvmet_port(struct config_item *item)
154 return container_of(to_config_group(item), struct nvmet_port,
158 static inline struct nvmet_port *ana_groups_to_port(
159 struct config_item *item)
161 return container_of(to_config_group(item), struct nvmet_port,
166 struct nvmet_subsys *subsys;
167 struct nvmet_sq **sqs;
180 struct nvmet_port *port;
183 unsigned long aen_masked;
184 struct nvmet_req *async_event_cmds[NVMET_ASYNC_EVENTS];
185 unsigned int nr_async_event_cmds;
186 struct list_head async_events;
187 struct work_struct async_event_work;
189 struct list_head subsys_entry;
191 struct delayed_work ka_work;
192 struct work_struct fatal_err_work;
194 const struct nvmet_fabrics_ops *ops;
196 __le32 *changed_ns_list;
199 char subsysnqn[NVMF_NQN_FIELD_LEN];
200 char hostnqn[NVMF_NQN_FIELD_LEN];
202 struct device *p2p_client;
203 struct radix_tree_root p2p_ns_map;
205 spinlock_t error_lock;
207 struct nvme_error_slot slots[NVMET_ERROR_LOG_SLOTS];
211 struct nvmet_subsys_model {
212 struct rcu_head rcuhead;
216 struct nvmet_subsys {
217 enum nvme_subsys_type type;
222 struct xarray namespaces;
223 unsigned int nr_namespaces;
224 unsigned int max_nsid;
228 struct list_head ctrls;
230 struct list_head hosts;
240 struct config_group group;
242 struct config_group namespaces_group;
243 struct config_group allowed_hosts_group;
245 struct nvmet_subsys_model __rcu *model;
247 #ifdef CONFIG_NVME_TARGET_PASSTHRU
248 struct nvme_ctrl *passthru_ctrl;
249 char *passthru_ctrl_path;
250 struct config_group passthru_group;
251 #endif /* CONFIG_NVME_TARGET_PASSTHRU */
254 static inline struct nvmet_subsys *to_subsys(struct config_item *item)
256 return container_of(to_config_group(item), struct nvmet_subsys, group);
259 static inline struct nvmet_subsys *namespaces_to_subsys(
260 struct config_item *item)
262 return container_of(to_config_group(item), struct nvmet_subsys,
267 struct config_group group;
270 static inline struct nvmet_host *to_host(struct config_item *item)
272 return container_of(to_config_group(item), struct nvmet_host, group);
275 static inline char *nvmet_host_name(struct nvmet_host *host)
277 return config_item_name(&host->group.cg_item);
280 struct nvmet_host_link {
281 struct list_head entry;
282 struct nvmet_host *host;
285 struct nvmet_subsys_link {
286 struct list_head entry;
287 struct nvmet_subsys *subsys;
291 struct nvmet_fabrics_ops {
292 struct module *owner;
296 #define NVMF_KEYED_SGLS (1 << 0)
297 #define NVMF_METADATA_SUPPORTED (1 << 1)
298 void (*queue_response)(struct nvmet_req *req);
299 int (*add_port)(struct nvmet_port *port);
300 void (*remove_port)(struct nvmet_port *port);
301 void (*delete_ctrl)(struct nvmet_ctrl *ctrl);
302 void (*disc_traddr)(struct nvmet_req *req,
303 struct nvmet_port *port, char *traddr);
304 u16 (*install_queue)(struct nvmet_sq *nvme_sq);
305 void (*discovery_chg)(struct nvmet_port *port);
306 u8 (*get_mdts)(const struct nvmet_ctrl *ctrl);
309 #define NVMET_MAX_INLINE_BIOVEC 8
310 #define NVMET_MAX_INLINE_DATA_LEN NVMET_MAX_INLINE_BIOVEC * PAGE_SIZE
313 struct nvme_command *cmd;
314 struct nvme_completion *cqe;
318 struct scatterlist *sg;
319 struct scatterlist *metadata_sg;
320 struct bio_vec inline_bvec[NVMET_MAX_INLINE_BIOVEC];
323 struct bio inline_bio;
328 struct bio_vec *bvec;
329 struct work_struct work;
333 struct work_struct work;
339 /* data length as parsed from the SGL descriptor: */
343 struct nvmet_port *port;
345 void (*execute)(struct nvmet_req *req);
346 const struct nvmet_fabrics_ops *ops;
348 struct pci_dev *p2p_dev;
349 struct device *p2p_client;
354 extern struct workqueue_struct *buffered_io_wq;
356 static inline void nvmet_set_result(struct nvmet_req *req, u32 result)
358 req->cqe->result.u32 = cpu_to_le32(result);
362 * NVMe command writes actually are DMA reads for us on the target side.
364 static inline enum dma_data_direction
365 nvmet_data_dir(struct nvmet_req *req)
367 return nvme_is_write(req->cmd) ? DMA_FROM_DEVICE : DMA_TO_DEVICE;
370 struct nvmet_async_event {
371 struct list_head entry;
377 static inline void nvmet_clear_aen_bit(struct nvmet_req *req, u32 bn)
379 int rae = le32_to_cpu(req->cmd->common.cdw10) & 1 << 15;
382 clear_bit(bn, &req->sq->ctrl->aen_masked);
385 static inline bool nvmet_aen_bit_disabled(struct nvmet_ctrl *ctrl, u32 bn)
387 if (!(READ_ONCE(ctrl->aen_enabled) & (1 << bn)))
389 return test_and_set_bit(bn, &ctrl->aen_masked);
392 void nvmet_get_feat_kato(struct nvmet_req *req);
393 void nvmet_get_feat_async_event(struct nvmet_req *req);
394 u16 nvmet_set_feat_kato(struct nvmet_req *req);
395 u16 nvmet_set_feat_async_event(struct nvmet_req *req, u32 mask);
396 void nvmet_execute_async_event(struct nvmet_req *req);
397 void nvmet_start_keep_alive_timer(struct nvmet_ctrl *ctrl);
398 void nvmet_stop_keep_alive_timer(struct nvmet_ctrl *ctrl);
400 u16 nvmet_parse_connect_cmd(struct nvmet_req *req);
401 void nvmet_bdev_set_limits(struct block_device *bdev, struct nvme_id_ns *id);
402 u16 nvmet_bdev_parse_io_cmd(struct nvmet_req *req);
403 u16 nvmet_file_parse_io_cmd(struct nvmet_req *req);
404 u16 nvmet_parse_admin_cmd(struct nvmet_req *req);
405 u16 nvmet_parse_discovery_cmd(struct nvmet_req *req);
406 u16 nvmet_parse_fabrics_cmd(struct nvmet_req *req);
408 bool nvmet_req_init(struct nvmet_req *req, struct nvmet_cq *cq,
409 struct nvmet_sq *sq, const struct nvmet_fabrics_ops *ops);
410 void nvmet_req_uninit(struct nvmet_req *req);
411 bool nvmet_check_transfer_len(struct nvmet_req *req, size_t len);
412 bool nvmet_check_data_len_lte(struct nvmet_req *req, size_t data_len);
413 void nvmet_req_complete(struct nvmet_req *req, u16 status);
414 int nvmet_req_alloc_sgls(struct nvmet_req *req);
415 void nvmet_req_free_sgls(struct nvmet_req *req);
417 void nvmet_execute_set_features(struct nvmet_req *req);
418 void nvmet_execute_get_features(struct nvmet_req *req);
419 void nvmet_execute_keep_alive(struct nvmet_req *req);
421 void nvmet_cq_setup(struct nvmet_ctrl *ctrl, struct nvmet_cq *cq, u16 qid,
423 void nvmet_sq_setup(struct nvmet_ctrl *ctrl, struct nvmet_sq *sq, u16 qid,
425 void nvmet_sq_destroy(struct nvmet_sq *sq);
426 int nvmet_sq_init(struct nvmet_sq *sq);
428 void nvmet_ctrl_fatal_error(struct nvmet_ctrl *ctrl);
430 void nvmet_update_cc(struct nvmet_ctrl *ctrl, u32 new);
431 u16 nvmet_alloc_ctrl(const char *subsysnqn, const char *hostnqn,
432 struct nvmet_req *req, u32 kato, struct nvmet_ctrl **ctrlp);
433 struct nvmet_ctrl *nvmet_ctrl_find_get(const char *subsysnqn,
434 const char *hostnqn, u16 cntlid,
435 struct nvmet_req *req);
436 void nvmet_ctrl_put(struct nvmet_ctrl *ctrl);
437 u16 nvmet_check_ctrl_status(struct nvmet_req *req, struct nvme_command *cmd);
439 struct nvmet_subsys *nvmet_subsys_alloc(const char *subsysnqn,
440 enum nvme_subsys_type type);
441 void nvmet_subsys_put(struct nvmet_subsys *subsys);
442 void nvmet_subsys_del_ctrls(struct nvmet_subsys *subsys);
444 struct nvmet_ns *nvmet_find_namespace(struct nvmet_ctrl *ctrl, __le32 nsid);
445 void nvmet_put_namespace(struct nvmet_ns *ns);
446 int nvmet_ns_enable(struct nvmet_ns *ns);
447 void nvmet_ns_disable(struct nvmet_ns *ns);
448 struct nvmet_ns *nvmet_ns_alloc(struct nvmet_subsys *subsys, u32 nsid);
449 void nvmet_ns_free(struct nvmet_ns *ns);
451 void nvmet_send_ana_event(struct nvmet_subsys *subsys,
452 struct nvmet_port *port);
453 void nvmet_port_send_ana_event(struct nvmet_port *port);
455 int nvmet_register_transport(const struct nvmet_fabrics_ops *ops);
456 void nvmet_unregister_transport(const struct nvmet_fabrics_ops *ops);
458 void nvmet_port_del_ctrls(struct nvmet_port *port,
459 struct nvmet_subsys *subsys);
461 int nvmet_enable_port(struct nvmet_port *port);
462 void nvmet_disable_port(struct nvmet_port *port);
464 void nvmet_referral_enable(struct nvmet_port *parent, struct nvmet_port *port);
465 void nvmet_referral_disable(struct nvmet_port *parent, struct nvmet_port *port);
467 u16 nvmet_copy_to_sgl(struct nvmet_req *req, off_t off, const void *buf,
469 u16 nvmet_copy_from_sgl(struct nvmet_req *req, off_t off, void *buf,
471 u16 nvmet_zero_sgl(struct nvmet_req *req, off_t off, size_t len);
473 u32 nvmet_get_log_page_len(struct nvme_command *cmd);
474 u64 nvmet_get_log_page_offset(struct nvme_command *cmd);
476 extern struct list_head *nvmet_ports;
477 void nvmet_port_disc_changed(struct nvmet_port *port,
478 struct nvmet_subsys *subsys);
479 void nvmet_subsys_disc_changed(struct nvmet_subsys *subsys,
480 struct nvmet_host *host);
481 void nvmet_add_async_event(struct nvmet_ctrl *ctrl, u8 event_type,
482 u8 event_info, u8 log_page);
484 #define NVMET_QUEUE_SIZE 1024
485 #define NVMET_NR_QUEUES 128
486 #define NVMET_MAX_CMD NVMET_QUEUE_SIZE
489 * Nice round number that makes a list of nsids fit into a page.
490 * Should become tunable at some point in the future.
492 #define NVMET_MAX_NAMESPACES 1024
495 * 0 is not a valid ANA group ID, so we start numbering at 1.
497 * ANA Group 1 exists without manual intervention, has namespaces assigned to it
498 * by default, and is available in an optimized state through all ports.
500 #define NVMET_MAX_ANAGRPS 128
501 #define NVMET_DEFAULT_ANA_GRPID 1
504 #define NVMET_DISC_KATO_MS 120000
506 int __init nvmet_init_configfs(void);
507 void __exit nvmet_exit_configfs(void);
509 int __init nvmet_init_discovery(void);
510 void nvmet_exit_discovery(void);
512 extern struct nvmet_subsys *nvmet_disc_subsys;
513 extern struct rw_semaphore nvmet_config_sem;
515 extern u32 nvmet_ana_group_enabled[NVMET_MAX_ANAGRPS + 1];
516 extern u64 nvmet_ana_chgcnt;
517 extern struct rw_semaphore nvmet_ana_sem;
519 bool nvmet_host_allowed(struct nvmet_subsys *subsys, const char *hostnqn);
521 int nvmet_bdev_ns_enable(struct nvmet_ns *ns);
522 int nvmet_file_ns_enable(struct nvmet_ns *ns);
523 void nvmet_bdev_ns_disable(struct nvmet_ns *ns);
524 void nvmet_file_ns_disable(struct nvmet_ns *ns);
525 u16 nvmet_bdev_flush(struct nvmet_req *req);
526 u16 nvmet_file_flush(struct nvmet_req *req);
527 void nvmet_ns_changed(struct nvmet_subsys *subsys, u32 nsid);
528 void nvmet_bdev_ns_revalidate(struct nvmet_ns *ns);
529 int nvmet_file_ns_revalidate(struct nvmet_ns *ns);
530 void nvmet_ns_revalidate(struct nvmet_ns *ns);
532 static inline u32 nvmet_rw_data_len(struct nvmet_req *req)
534 return ((u32)le16_to_cpu(req->cmd->rw.length) + 1) <<
535 req->ns->blksize_shift;
538 static inline u32 nvmet_rw_metadata_len(struct nvmet_req *req)
540 if (!IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY))
542 return ((u32)le16_to_cpu(req->cmd->rw.length) + 1) *
543 req->ns->metadata_size;
546 static inline u32 nvmet_dsm_len(struct nvmet_req *req)
548 return (le32_to_cpu(req->cmd->dsm.nr) + 1) *
549 sizeof(struct nvme_dsm_range);
552 #ifdef CONFIG_NVME_TARGET_PASSTHRU
553 void nvmet_passthru_subsys_free(struct nvmet_subsys *subsys);
554 int nvmet_passthru_ctrl_enable(struct nvmet_subsys *subsys);
555 void nvmet_passthru_ctrl_disable(struct nvmet_subsys *subsys);
556 u16 nvmet_parse_passthru_admin_cmd(struct nvmet_req *req);
557 u16 nvmet_parse_passthru_io_cmd(struct nvmet_req *req);
558 static inline struct nvme_ctrl *nvmet_passthru_ctrl(struct nvmet_subsys *subsys)
560 return subsys->passthru_ctrl;
562 #else /* CONFIG_NVME_TARGET_PASSTHRU */
563 static inline void nvmet_passthru_subsys_free(struct nvmet_subsys *subsys)
566 static inline void nvmet_passthru_ctrl_disable(struct nvmet_subsys *subsys)
569 static inline u16 nvmet_parse_passthru_admin_cmd(struct nvmet_req *req)
573 static inline u16 nvmet_parse_passthru_io_cmd(struct nvmet_req *req)
577 static inline struct nvme_ctrl *nvmet_passthru_ctrl(struct nvmet_subsys *subsys)
581 #endif /* CONFIG_NVME_TARGET_PASSTHRU */
583 static inline struct nvme_ctrl *
584 nvmet_req_passthru_ctrl(struct nvmet_req *req)
586 return nvmet_passthru_ctrl(req->sq->ctrl->subsys);
589 u16 errno_to_nvme_status(struct nvmet_req *req, int errno);
591 /* Convert a 32-bit number to a 16-bit 0's based number */
592 static inline __le16 to0based(u32 a)
594 return cpu_to_le16(max(1U, min(1U << 16, a)) - 1);
597 static inline bool nvmet_ns_has_pi(struct nvmet_ns *ns)
599 if (!IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY))
601 return ns->pi_type && ns->metadata_size == sizeof(struct t10_pi_tuple);
604 static inline __le64 nvmet_sect_to_lba(struct nvmet_ns *ns, sector_t sect)
606 return cpu_to_le64(sect >> (ns->blksize_shift - SECTOR_SHIFT));
609 static inline sector_t nvmet_lba_to_sect(struct nvmet_ns *ns, __le64 lba)
611 return le64_to_cpu(lba) << (ns->blksize_shift - SECTOR_SHIFT);
614 static inline bool nvmet_use_inline_bvec(struct nvmet_req *req)
616 return req->transfer_len <= NVMET_MAX_INLINE_DATA_LEN &&
617 req->sg_cnt <= NVMET_MAX_INLINE_BIOVEC;
620 #endif /* _NVMET_H */