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
33 * Supported optional AENs:
35 #define NVMET_AEN_CFG_OPTIONAL \
36 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_ANA_CHANGE)
37 #define NVMET_DISC_AEN_CFG_OPTIONAL \
38 (NVME_AEN_CFG_DISC_CHANGE)
41 * Plus mandatory SMART AENs (we'll never send them, but allow enabling them):
43 #define NVMET_AEN_CFG_ALL \
44 (NVME_SMART_CRIT_SPARE | NVME_SMART_CRIT_TEMPERATURE | \
45 NVME_SMART_CRIT_RELIABILITY | NVME_SMART_CRIT_MEDIA | \
46 NVME_SMART_CRIT_VOLATILE_MEMORY | NVMET_AEN_CFG_OPTIONAL)
48 /* Helper Macros when NVMe error is NVME_SC_CONNECT_INVALID_PARAM
49 * The 16 bit shift is to set IATTR bit to 1, which means offending
50 * offset starts in the data section of connect()
52 #define IPO_IATTR_CONNECT_DATA(x) \
53 (cpu_to_le32((1 << 16) | (offsetof(struct nvmf_connect_data, x))))
54 #define IPO_IATTR_CONNECT_SQE(x) \
55 (cpu_to_le32(offsetof(struct nvmf_connect_command, x)))
58 struct percpu_ref ref;
59 struct block_device *bdev;
71 struct nvmet_subsys *subsys;
72 const char *device_path;
74 struct config_group device_group;
75 struct config_group group;
77 struct completion disable_done;
79 struct kmem_cache *bvec_cache;
82 struct pci_dev *p2p_dev;
87 static inline struct nvmet_ns *to_nvmet_ns(struct config_item *item)
89 return container_of(to_config_group(item), struct nvmet_ns, group);
92 static inline struct device *nvmet_ns_dev(struct nvmet_ns *ns)
94 return ns->bdev ? disk_to_dev(ns->bdev->bd_disk) : NULL;
103 struct nvmet_ctrl *ctrl;
104 struct percpu_ref ref;
109 struct completion free_done;
110 struct completion confirm_done;
113 struct nvmet_ana_group {
114 struct config_group group;
115 struct nvmet_port *port;
119 static inline struct nvmet_ana_group *to_ana_group(struct config_item *item)
121 return container_of(to_config_group(item), struct nvmet_ana_group,
126 * struct nvmet_port - Common structure to keep port
127 * information for the target.
128 * @entry: Entry into referrals or transport list.
129 * @disc_addr: Address information is stored in a format defined
130 * for a discovery log page entry.
131 * @group: ConfigFS group for this element's folder.
132 * @priv: Private data for the transport.
135 struct list_head entry;
136 struct nvmf_disc_rsp_page_entry disc_addr;
137 struct config_group group;
138 struct config_group subsys_group;
139 struct list_head subsystems;
140 struct config_group referrals_group;
141 struct list_head referrals;
142 struct list_head global_entry;
143 struct config_group ana_groups_group;
144 struct nvmet_ana_group ana_default_group;
145 enum nvme_ana_state *ana_state;
148 int inline_data_size;
149 const struct nvmet_fabrics_ops *tr_ops;
153 static inline struct nvmet_port *to_nvmet_port(struct config_item *item)
155 return container_of(to_config_group(item), struct nvmet_port,
159 static inline struct nvmet_port *ana_groups_to_port(
160 struct config_item *item)
162 return container_of(to_config_group(item), struct nvmet_port,
167 struct nvmet_subsys *subsys;
168 struct nvmet_sq **sqs;
181 struct nvmet_port *port;
184 unsigned long aen_masked;
185 struct nvmet_req *async_event_cmds[NVMET_ASYNC_EVENTS];
186 unsigned int nr_async_event_cmds;
187 struct list_head async_events;
188 struct work_struct async_event_work;
190 struct list_head subsys_entry;
192 struct delayed_work ka_work;
193 struct work_struct fatal_err_work;
195 const struct nvmet_fabrics_ops *ops;
197 __le32 *changed_ns_list;
200 char subsysnqn[NVMF_NQN_FIELD_LEN];
201 char hostnqn[NVMF_NQN_FIELD_LEN];
203 struct device *p2p_client;
204 struct radix_tree_root p2p_ns_map;
206 spinlock_t error_lock;
208 struct nvme_error_slot slots[NVMET_ERROR_LOG_SLOTS];
212 struct nvmet_subsys {
213 enum nvme_subsys_type type;
218 struct xarray namespaces;
219 unsigned int nr_namespaces;
220 unsigned int max_nsid;
224 struct list_head ctrls;
226 struct list_head hosts;
236 struct config_group group;
238 struct config_group namespaces_group;
239 struct config_group allowed_hosts_group;
243 #ifdef CONFIG_NVME_TARGET_PASSTHRU
244 struct nvme_ctrl *passthru_ctrl;
245 char *passthru_ctrl_path;
246 struct config_group passthru_group;
247 unsigned int admin_timeout;
248 unsigned int io_timeout;
249 #endif /* CONFIG_NVME_TARGET_PASSTHRU */
252 static inline struct nvmet_subsys *to_subsys(struct config_item *item)
254 return container_of(to_config_group(item), struct nvmet_subsys, group);
257 static inline struct nvmet_subsys *namespaces_to_subsys(
258 struct config_item *item)
260 return container_of(to_config_group(item), struct nvmet_subsys,
265 struct config_group group;
268 static inline struct nvmet_host *to_host(struct config_item *item)
270 return container_of(to_config_group(item), struct nvmet_host, group);
273 static inline char *nvmet_host_name(struct nvmet_host *host)
275 return config_item_name(&host->group.cg_item);
278 struct nvmet_host_link {
279 struct list_head entry;
280 struct nvmet_host *host;
283 struct nvmet_subsys_link {
284 struct list_head entry;
285 struct nvmet_subsys *subsys;
289 struct nvmet_fabrics_ops {
290 struct module *owner;
294 #define NVMF_KEYED_SGLS (1 << 0)
295 #define NVMF_METADATA_SUPPORTED (1 << 1)
296 void (*queue_response)(struct nvmet_req *req);
297 int (*add_port)(struct nvmet_port *port);
298 void (*remove_port)(struct nvmet_port *port);
299 void (*delete_ctrl)(struct nvmet_ctrl *ctrl);
300 void (*disc_traddr)(struct nvmet_req *req,
301 struct nvmet_port *port, char *traddr);
302 u16 (*install_queue)(struct nvmet_sq *nvme_sq);
303 void (*discovery_chg)(struct nvmet_port *port);
304 u8 (*get_mdts)(const struct nvmet_ctrl *ctrl);
307 #define NVMET_MAX_INLINE_BIOVEC 8
308 #define NVMET_MAX_INLINE_DATA_LEN NVMET_MAX_INLINE_BIOVEC * PAGE_SIZE
311 struct nvme_command *cmd;
312 struct nvme_completion *cqe;
316 struct scatterlist *sg;
317 struct scatterlist *metadata_sg;
318 struct bio_vec inline_bvec[NVMET_MAX_INLINE_BIOVEC];
321 struct bio inline_bio;
326 struct bio_vec *bvec;
327 struct work_struct work;
330 struct bio inline_bio;
332 struct work_struct work;
338 /* data length as parsed from the SGL descriptor: */
342 struct nvmet_port *port;
344 void (*execute)(struct nvmet_req *req);
345 const struct nvmet_fabrics_ops *ops;
347 struct pci_dev *p2p_dev;
348 struct device *p2p_client;
353 extern struct workqueue_struct *buffered_io_wq;
355 static inline void nvmet_set_result(struct nvmet_req *req, u32 result)
357 req->cqe->result.u32 = cpu_to_le32(result);
361 * NVMe command writes actually are DMA reads for us on the target side.
363 static inline enum dma_data_direction
364 nvmet_data_dir(struct nvmet_req *req)
366 return nvme_is_write(req->cmd) ? DMA_FROM_DEVICE : DMA_TO_DEVICE;
369 struct nvmet_async_event {
370 struct list_head entry;
376 static inline void nvmet_clear_aen_bit(struct nvmet_req *req, u32 bn)
378 int rae = le32_to_cpu(req->cmd->common.cdw10) & 1 << 15;
381 clear_bit(bn, &req->sq->ctrl->aen_masked);
384 static inline bool nvmet_aen_bit_disabled(struct nvmet_ctrl *ctrl, u32 bn)
386 if (!(READ_ONCE(ctrl->aen_enabled) & (1 << bn)))
388 return test_and_set_bit(bn, &ctrl->aen_masked);
391 void nvmet_get_feat_kato(struct nvmet_req *req);
392 void nvmet_get_feat_async_event(struct nvmet_req *req);
393 u16 nvmet_set_feat_kato(struct nvmet_req *req);
394 u16 nvmet_set_feat_async_event(struct nvmet_req *req, u32 mask);
395 void nvmet_execute_async_event(struct nvmet_req *req);
396 void nvmet_start_keep_alive_timer(struct nvmet_ctrl *ctrl);
397 void nvmet_stop_keep_alive_timer(struct nvmet_ctrl *ctrl);
399 u16 nvmet_parse_connect_cmd(struct nvmet_req *req);
400 void nvmet_bdev_set_limits(struct block_device *bdev, struct nvme_id_ns *id);
401 u16 nvmet_bdev_parse_io_cmd(struct nvmet_req *req);
402 u16 nvmet_file_parse_io_cmd(struct nvmet_req *req);
403 u16 nvmet_parse_admin_cmd(struct nvmet_req *req);
404 u16 nvmet_parse_discovery_cmd(struct nvmet_req *req);
405 u16 nvmet_parse_fabrics_cmd(struct nvmet_req *req);
407 bool nvmet_req_init(struct nvmet_req *req, struct nvmet_cq *cq,
408 struct nvmet_sq *sq, const struct nvmet_fabrics_ops *ops);
409 void nvmet_req_uninit(struct nvmet_req *req);
410 bool nvmet_check_transfer_len(struct nvmet_req *req, size_t len);
411 bool nvmet_check_data_len_lte(struct nvmet_req *req, size_t data_len);
412 void nvmet_req_complete(struct nvmet_req *req, u16 status);
413 int nvmet_req_alloc_sgls(struct nvmet_req *req);
414 void nvmet_req_free_sgls(struct nvmet_req *req);
416 void nvmet_execute_set_features(struct nvmet_req *req);
417 void nvmet_execute_get_features(struct nvmet_req *req);
418 void nvmet_execute_keep_alive(struct nvmet_req *req);
420 void nvmet_cq_setup(struct nvmet_ctrl *ctrl, struct nvmet_cq *cq, u16 qid,
422 void nvmet_sq_setup(struct nvmet_ctrl *ctrl, struct nvmet_sq *sq, u16 qid,
424 void nvmet_sq_destroy(struct nvmet_sq *sq);
425 int nvmet_sq_init(struct nvmet_sq *sq);
427 void nvmet_ctrl_fatal_error(struct nvmet_ctrl *ctrl);
429 void nvmet_update_cc(struct nvmet_ctrl *ctrl, u32 new);
430 u16 nvmet_alloc_ctrl(const char *subsysnqn, const char *hostnqn,
431 struct nvmet_req *req, u32 kato, struct nvmet_ctrl **ctrlp);
432 struct nvmet_ctrl *nvmet_ctrl_find_get(const char *subsysnqn,
433 const char *hostnqn, u16 cntlid,
434 struct nvmet_req *req);
435 void nvmet_ctrl_put(struct nvmet_ctrl *ctrl);
436 u16 nvmet_check_ctrl_status(struct nvmet_req *req);
438 struct nvmet_subsys *nvmet_subsys_alloc(const char *subsysnqn,
439 enum nvme_subsys_type type);
440 void nvmet_subsys_put(struct nvmet_subsys *subsys);
441 void nvmet_subsys_del_ctrls(struct nvmet_subsys *subsys);
443 u16 nvmet_req_find_ns(struct nvmet_req *req);
444 void nvmet_put_namespace(struct nvmet_ns *ns);
445 int nvmet_ns_enable(struct nvmet_ns *ns);
446 void nvmet_ns_disable(struct nvmet_ns *ns);
447 struct nvmet_ns *nvmet_ns_alloc(struct nvmet_subsys *subsys, u32 nsid);
448 void nvmet_ns_free(struct nvmet_ns *ns);
450 void nvmet_send_ana_event(struct nvmet_subsys *subsys,
451 struct nvmet_port *port);
452 void nvmet_port_send_ana_event(struct nvmet_port *port);
454 int nvmet_register_transport(const struct nvmet_fabrics_ops *ops);
455 void nvmet_unregister_transport(const struct nvmet_fabrics_ops *ops);
457 void nvmet_port_del_ctrls(struct nvmet_port *port,
458 struct nvmet_subsys *subsys);
460 int nvmet_enable_port(struct nvmet_port *port);
461 void nvmet_disable_port(struct nvmet_port *port);
463 void nvmet_referral_enable(struct nvmet_port *parent, struct nvmet_port *port);
464 void nvmet_referral_disable(struct nvmet_port *parent, struct nvmet_port *port);
466 u16 nvmet_copy_to_sgl(struct nvmet_req *req, off_t off, const void *buf,
468 u16 nvmet_copy_from_sgl(struct nvmet_req *req, off_t off, void *buf,
470 u16 nvmet_zero_sgl(struct nvmet_req *req, off_t off, size_t len);
472 u32 nvmet_get_log_page_len(struct nvme_command *cmd);
473 u64 nvmet_get_log_page_offset(struct nvme_command *cmd);
475 extern struct list_head *nvmet_ports;
476 void nvmet_port_disc_changed(struct nvmet_port *port,
477 struct nvmet_subsys *subsys);
478 void nvmet_subsys_disc_changed(struct nvmet_subsys *subsys,
479 struct nvmet_host *host);
480 void nvmet_add_async_event(struct nvmet_ctrl *ctrl, u8 event_type,
481 u8 event_info, u8 log_page);
483 #define NVMET_QUEUE_SIZE 1024
484 #define NVMET_NR_QUEUES 128
485 #define NVMET_MAX_CMD NVMET_QUEUE_SIZE
488 * Nice round number that makes a list of nsids fit into a page.
489 * Should become tunable at some point in the future.
491 #define NVMET_MAX_NAMESPACES 1024
494 * 0 is not a valid ANA group ID, so we start numbering at 1.
496 * ANA Group 1 exists without manual intervention, has namespaces assigned to it
497 * by default, and is available in an optimized state through all ports.
499 #define NVMET_MAX_ANAGRPS 128
500 #define NVMET_DEFAULT_ANA_GRPID 1
503 #define NVMET_DISC_KATO_MS 120000
505 int __init nvmet_init_configfs(void);
506 void __exit nvmet_exit_configfs(void);
508 int __init nvmet_init_discovery(void);
509 void nvmet_exit_discovery(void);
511 extern struct nvmet_subsys *nvmet_disc_subsys;
512 extern struct rw_semaphore nvmet_config_sem;
514 extern u32 nvmet_ana_group_enabled[NVMET_MAX_ANAGRPS + 1];
515 extern u64 nvmet_ana_chgcnt;
516 extern struct rw_semaphore nvmet_ana_sem;
518 bool nvmet_host_allowed(struct nvmet_subsys *subsys, const char *hostnqn);
520 int nvmet_bdev_ns_enable(struct nvmet_ns *ns);
521 int nvmet_file_ns_enable(struct nvmet_ns *ns);
522 void nvmet_bdev_ns_disable(struct nvmet_ns *ns);
523 void nvmet_file_ns_disable(struct nvmet_ns *ns);
524 u16 nvmet_bdev_flush(struct nvmet_req *req);
525 u16 nvmet_file_flush(struct nvmet_req *req);
526 void nvmet_ns_changed(struct nvmet_subsys *subsys, u32 nsid);
527 void nvmet_bdev_ns_revalidate(struct nvmet_ns *ns);
528 int nvmet_file_ns_revalidate(struct nvmet_ns *ns);
529 void nvmet_ns_revalidate(struct nvmet_ns *ns);
531 static inline u32 nvmet_rw_data_len(struct nvmet_req *req)
533 return ((u32)le16_to_cpu(req->cmd->rw.length) + 1) <<
534 req->ns->blksize_shift;
537 static inline u32 nvmet_rw_metadata_len(struct nvmet_req *req)
539 if (!IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY))
541 return ((u32)le16_to_cpu(req->cmd->rw.length) + 1) *
542 req->ns->metadata_size;
545 static inline u32 nvmet_dsm_len(struct nvmet_req *req)
547 return (le32_to_cpu(req->cmd->dsm.nr) + 1) *
548 sizeof(struct nvme_dsm_range);
551 static inline struct nvmet_subsys *nvmet_req_subsys(struct nvmet_req *req)
553 return req->sq->ctrl->subsys;
556 #ifdef CONFIG_NVME_TARGET_PASSTHRU
557 void nvmet_passthru_subsys_free(struct nvmet_subsys *subsys);
558 int nvmet_passthru_ctrl_enable(struct nvmet_subsys *subsys);
559 void nvmet_passthru_ctrl_disable(struct nvmet_subsys *subsys);
560 u16 nvmet_parse_passthru_admin_cmd(struct nvmet_req *req);
561 u16 nvmet_parse_passthru_io_cmd(struct nvmet_req *req);
562 static inline struct nvme_ctrl *nvmet_passthru_ctrl(struct nvmet_subsys *subsys)
564 return subsys->passthru_ctrl;
566 #else /* CONFIG_NVME_TARGET_PASSTHRU */
567 static inline void nvmet_passthru_subsys_free(struct nvmet_subsys *subsys)
570 static inline void nvmet_passthru_ctrl_disable(struct nvmet_subsys *subsys)
573 static inline u16 nvmet_parse_passthru_admin_cmd(struct nvmet_req *req)
577 static inline u16 nvmet_parse_passthru_io_cmd(struct nvmet_req *req)
581 static inline struct nvme_ctrl *nvmet_passthru_ctrl(struct nvmet_subsys *subsys)
585 #endif /* CONFIG_NVME_TARGET_PASSTHRU */
587 static inline struct nvme_ctrl *
588 nvmet_req_passthru_ctrl(struct nvmet_req *req)
590 return nvmet_passthru_ctrl(nvmet_req_subsys(req));
593 u16 errno_to_nvme_status(struct nvmet_req *req, int errno);
594 u16 nvmet_report_invalid_opcode(struct nvmet_req *req);
596 /* Convert a 32-bit number to a 16-bit 0's based number */
597 static inline __le16 to0based(u32 a)
599 return cpu_to_le16(max(1U, min(1U << 16, a)) - 1);
602 static inline bool nvmet_ns_has_pi(struct nvmet_ns *ns)
604 if (!IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY))
606 return ns->pi_type && ns->metadata_size == sizeof(struct t10_pi_tuple);
609 static inline __le64 nvmet_sect_to_lba(struct nvmet_ns *ns, sector_t sect)
611 return cpu_to_le64(sect >> (ns->blksize_shift - SECTOR_SHIFT));
614 static inline sector_t nvmet_lba_to_sect(struct nvmet_ns *ns, __le64 lba)
616 return le64_to_cpu(lba) << (ns->blksize_shift - SECTOR_SHIFT);
619 static inline bool nvmet_use_inline_bvec(struct nvmet_req *req)
621 return req->transfer_len <= NVMET_MAX_INLINE_DATA_LEN &&
622 req->sg_cnt <= NVMET_MAX_INLINE_BIOVEC;
625 #endif /* _NVMET_H */