Back to home page

OSCL-LXR
 
 

     

0001 // SPDX-License-Identifier: GPL-2.0
0002 /*
0003  * Common code for the NVMe target.
0004  * Copyright (c) 2015-2016 HGST, a Western Digital Company.
0005  */
0006 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
0007 #include <linux/module.h>
0008 #include <linux/random.h>
0009 #include <linux/rculist.h>
0010 #include <linux/pci-p2pdma.h>
0011 #include <linux/scatterlist.h>
0012 
0013 #define CREATE_TRACE_POINTS
0014 #include "trace.h"
0015 
0016 #include "nvmet.h"
0017 
0018 struct workqueue_struct *buffered_io_wq;
0019 struct workqueue_struct *zbd_wq;
0020 static const struct nvmet_fabrics_ops *nvmet_transports[NVMF_TRTYPE_MAX];
0021 static DEFINE_IDA(cntlid_ida);
0022 
0023 struct workqueue_struct *nvmet_wq;
0024 EXPORT_SYMBOL_GPL(nvmet_wq);
0025 
0026 /*
0027  * This read/write semaphore is used to synchronize access to configuration
0028  * information on a target system that will result in discovery log page
0029  * information change for at least one host.
0030  * The full list of resources to protected by this semaphore is:
0031  *
0032  *  - subsystems list
0033  *  - per-subsystem allowed hosts list
0034  *  - allow_any_host subsystem attribute
0035  *  - nvmet_genctr
0036  *  - the nvmet_transports array
0037  *
0038  * When updating any of those lists/structures write lock should be obtained,
0039  * while when reading (popolating discovery log page or checking host-subsystem
0040  * link) read lock is obtained to allow concurrent reads.
0041  */
0042 DECLARE_RWSEM(nvmet_config_sem);
0043 
0044 u32 nvmet_ana_group_enabled[NVMET_MAX_ANAGRPS + 1];
0045 u64 nvmet_ana_chgcnt;
0046 DECLARE_RWSEM(nvmet_ana_sem);
0047 
0048 inline u16 errno_to_nvme_status(struct nvmet_req *req, int errno)
0049 {
0050     switch (errno) {
0051     case 0:
0052         return NVME_SC_SUCCESS;
0053     case -ENOSPC:
0054         req->error_loc = offsetof(struct nvme_rw_command, length);
0055         return NVME_SC_CAP_EXCEEDED | NVME_SC_DNR;
0056     case -EREMOTEIO:
0057         req->error_loc = offsetof(struct nvme_rw_command, slba);
0058         return  NVME_SC_LBA_RANGE | NVME_SC_DNR;
0059     case -EOPNOTSUPP:
0060         req->error_loc = offsetof(struct nvme_common_command, opcode);
0061         switch (req->cmd->common.opcode) {
0062         case nvme_cmd_dsm:
0063         case nvme_cmd_write_zeroes:
0064             return NVME_SC_ONCS_NOT_SUPPORTED | NVME_SC_DNR;
0065         default:
0066             return NVME_SC_INVALID_OPCODE | NVME_SC_DNR;
0067         }
0068         break;
0069     case -ENODATA:
0070         req->error_loc = offsetof(struct nvme_rw_command, nsid);
0071         return NVME_SC_ACCESS_DENIED;
0072     case -EIO:
0073         fallthrough;
0074     default:
0075         req->error_loc = offsetof(struct nvme_common_command, opcode);
0076         return NVME_SC_INTERNAL | NVME_SC_DNR;
0077     }
0078 }
0079 
0080 u16 nvmet_report_invalid_opcode(struct nvmet_req *req)
0081 {
0082     pr_debug("unhandled cmd %d on qid %d\n", req->cmd->common.opcode,
0083          req->sq->qid);
0084 
0085     req->error_loc = offsetof(struct nvme_common_command, opcode);
0086     return NVME_SC_INVALID_OPCODE | NVME_SC_DNR;
0087 }
0088 
0089 static struct nvmet_subsys *nvmet_find_get_subsys(struct nvmet_port *port,
0090         const char *subsysnqn);
0091 
0092 u16 nvmet_copy_to_sgl(struct nvmet_req *req, off_t off, const void *buf,
0093         size_t len)
0094 {
0095     if (sg_pcopy_from_buffer(req->sg, req->sg_cnt, buf, len, off) != len) {
0096         req->error_loc = offsetof(struct nvme_common_command, dptr);
0097         return NVME_SC_SGL_INVALID_DATA | NVME_SC_DNR;
0098     }
0099     return 0;
0100 }
0101 
0102 u16 nvmet_copy_from_sgl(struct nvmet_req *req, off_t off, void *buf, size_t len)
0103 {
0104     if (sg_pcopy_to_buffer(req->sg, req->sg_cnt, buf, len, off) != len) {
0105         req->error_loc = offsetof(struct nvme_common_command, dptr);
0106         return NVME_SC_SGL_INVALID_DATA | NVME_SC_DNR;
0107     }
0108     return 0;
0109 }
0110 
0111 u16 nvmet_zero_sgl(struct nvmet_req *req, off_t off, size_t len)
0112 {
0113     if (sg_zero_buffer(req->sg, req->sg_cnt, len, off) != len) {
0114         req->error_loc = offsetof(struct nvme_common_command, dptr);
0115         return NVME_SC_SGL_INVALID_DATA | NVME_SC_DNR;
0116     }
0117     return 0;
0118 }
0119 
0120 static u32 nvmet_max_nsid(struct nvmet_subsys *subsys)
0121 {
0122     struct nvmet_ns *cur;
0123     unsigned long idx;
0124     u32 nsid = 0;
0125 
0126     xa_for_each(&subsys->namespaces, idx, cur)
0127         nsid = cur->nsid;
0128 
0129     return nsid;
0130 }
0131 
0132 static u32 nvmet_async_event_result(struct nvmet_async_event *aen)
0133 {
0134     return aen->event_type | (aen->event_info << 8) | (aen->log_page << 16);
0135 }
0136 
0137 static void nvmet_async_events_failall(struct nvmet_ctrl *ctrl)
0138 {
0139     struct nvmet_req *req;
0140 
0141     mutex_lock(&ctrl->lock);
0142     while (ctrl->nr_async_event_cmds) {
0143         req = ctrl->async_event_cmds[--ctrl->nr_async_event_cmds];
0144         mutex_unlock(&ctrl->lock);
0145         nvmet_req_complete(req, NVME_SC_INTERNAL | NVME_SC_DNR);
0146         mutex_lock(&ctrl->lock);
0147     }
0148     mutex_unlock(&ctrl->lock);
0149 }
0150 
0151 static void nvmet_async_events_process(struct nvmet_ctrl *ctrl)
0152 {
0153     struct nvmet_async_event *aen;
0154     struct nvmet_req *req;
0155 
0156     mutex_lock(&ctrl->lock);
0157     while (ctrl->nr_async_event_cmds && !list_empty(&ctrl->async_events)) {
0158         aen = list_first_entry(&ctrl->async_events,
0159                        struct nvmet_async_event, entry);
0160         req = ctrl->async_event_cmds[--ctrl->nr_async_event_cmds];
0161         nvmet_set_result(req, nvmet_async_event_result(aen));
0162 
0163         list_del(&aen->entry);
0164         kfree(aen);
0165 
0166         mutex_unlock(&ctrl->lock);
0167         trace_nvmet_async_event(ctrl, req->cqe->result.u32);
0168         nvmet_req_complete(req, 0);
0169         mutex_lock(&ctrl->lock);
0170     }
0171     mutex_unlock(&ctrl->lock);
0172 }
0173 
0174 static void nvmet_async_events_free(struct nvmet_ctrl *ctrl)
0175 {
0176     struct nvmet_async_event *aen, *tmp;
0177 
0178     mutex_lock(&ctrl->lock);
0179     list_for_each_entry_safe(aen, tmp, &ctrl->async_events, entry) {
0180         list_del(&aen->entry);
0181         kfree(aen);
0182     }
0183     mutex_unlock(&ctrl->lock);
0184 }
0185 
0186 static void nvmet_async_event_work(struct work_struct *work)
0187 {
0188     struct nvmet_ctrl *ctrl =
0189         container_of(work, struct nvmet_ctrl, async_event_work);
0190 
0191     nvmet_async_events_process(ctrl);
0192 }
0193 
0194 void nvmet_add_async_event(struct nvmet_ctrl *ctrl, u8 event_type,
0195         u8 event_info, u8 log_page)
0196 {
0197     struct nvmet_async_event *aen;
0198 
0199     aen = kmalloc(sizeof(*aen), GFP_KERNEL);
0200     if (!aen)
0201         return;
0202 
0203     aen->event_type = event_type;
0204     aen->event_info = event_info;
0205     aen->log_page = log_page;
0206 
0207     mutex_lock(&ctrl->lock);
0208     list_add_tail(&aen->entry, &ctrl->async_events);
0209     mutex_unlock(&ctrl->lock);
0210 
0211     queue_work(nvmet_wq, &ctrl->async_event_work);
0212 }
0213 
0214 static void nvmet_add_to_changed_ns_log(struct nvmet_ctrl *ctrl, __le32 nsid)
0215 {
0216     u32 i;
0217 
0218     mutex_lock(&ctrl->lock);
0219     if (ctrl->nr_changed_ns > NVME_MAX_CHANGED_NAMESPACES)
0220         goto out_unlock;
0221 
0222     for (i = 0; i < ctrl->nr_changed_ns; i++) {
0223         if (ctrl->changed_ns_list[i] == nsid)
0224             goto out_unlock;
0225     }
0226 
0227     if (ctrl->nr_changed_ns == NVME_MAX_CHANGED_NAMESPACES) {
0228         ctrl->changed_ns_list[0] = cpu_to_le32(0xffffffff);
0229         ctrl->nr_changed_ns = U32_MAX;
0230         goto out_unlock;
0231     }
0232 
0233     ctrl->changed_ns_list[ctrl->nr_changed_ns++] = nsid;
0234 out_unlock:
0235     mutex_unlock(&ctrl->lock);
0236 }
0237 
0238 void nvmet_ns_changed(struct nvmet_subsys *subsys, u32 nsid)
0239 {
0240     struct nvmet_ctrl *ctrl;
0241 
0242     lockdep_assert_held(&subsys->lock);
0243 
0244     list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
0245         nvmet_add_to_changed_ns_log(ctrl, cpu_to_le32(nsid));
0246         if (nvmet_aen_bit_disabled(ctrl, NVME_AEN_BIT_NS_ATTR))
0247             continue;
0248         nvmet_add_async_event(ctrl, NVME_AER_TYPE_NOTICE,
0249                 NVME_AER_NOTICE_NS_CHANGED,
0250                 NVME_LOG_CHANGED_NS);
0251     }
0252 }
0253 
0254 void nvmet_send_ana_event(struct nvmet_subsys *subsys,
0255         struct nvmet_port *port)
0256 {
0257     struct nvmet_ctrl *ctrl;
0258 
0259     mutex_lock(&subsys->lock);
0260     list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
0261         if (port && ctrl->port != port)
0262             continue;
0263         if (nvmet_aen_bit_disabled(ctrl, NVME_AEN_BIT_ANA_CHANGE))
0264             continue;
0265         nvmet_add_async_event(ctrl, NVME_AER_TYPE_NOTICE,
0266                 NVME_AER_NOTICE_ANA, NVME_LOG_ANA);
0267     }
0268     mutex_unlock(&subsys->lock);
0269 }
0270 
0271 void nvmet_port_send_ana_event(struct nvmet_port *port)
0272 {
0273     struct nvmet_subsys_link *p;
0274 
0275     down_read(&nvmet_config_sem);
0276     list_for_each_entry(p, &port->subsystems, entry)
0277         nvmet_send_ana_event(p->subsys, port);
0278     up_read(&nvmet_config_sem);
0279 }
0280 
0281 int nvmet_register_transport(const struct nvmet_fabrics_ops *ops)
0282 {
0283     int ret = 0;
0284 
0285     down_write(&nvmet_config_sem);
0286     if (nvmet_transports[ops->type])
0287         ret = -EINVAL;
0288     else
0289         nvmet_transports[ops->type] = ops;
0290     up_write(&nvmet_config_sem);
0291 
0292     return ret;
0293 }
0294 EXPORT_SYMBOL_GPL(nvmet_register_transport);
0295 
0296 void nvmet_unregister_transport(const struct nvmet_fabrics_ops *ops)
0297 {
0298     down_write(&nvmet_config_sem);
0299     nvmet_transports[ops->type] = NULL;
0300     up_write(&nvmet_config_sem);
0301 }
0302 EXPORT_SYMBOL_GPL(nvmet_unregister_transport);
0303 
0304 void nvmet_port_del_ctrls(struct nvmet_port *port, struct nvmet_subsys *subsys)
0305 {
0306     struct nvmet_ctrl *ctrl;
0307 
0308     mutex_lock(&subsys->lock);
0309     list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
0310         if (ctrl->port == port)
0311             ctrl->ops->delete_ctrl(ctrl);
0312     }
0313     mutex_unlock(&subsys->lock);
0314 }
0315 
0316 int nvmet_enable_port(struct nvmet_port *port)
0317 {
0318     const struct nvmet_fabrics_ops *ops;
0319     int ret;
0320 
0321     lockdep_assert_held(&nvmet_config_sem);
0322 
0323     ops = nvmet_transports[port->disc_addr.trtype];
0324     if (!ops) {
0325         up_write(&nvmet_config_sem);
0326         request_module("nvmet-transport-%d", port->disc_addr.trtype);
0327         down_write(&nvmet_config_sem);
0328         ops = nvmet_transports[port->disc_addr.trtype];
0329         if (!ops) {
0330             pr_err("transport type %d not supported\n",
0331                 port->disc_addr.trtype);
0332             return -EINVAL;
0333         }
0334     }
0335 
0336     if (!try_module_get(ops->owner))
0337         return -EINVAL;
0338 
0339     /*
0340      * If the user requested PI support and the transport isn't pi capable,
0341      * don't enable the port.
0342      */
0343     if (port->pi_enable && !(ops->flags & NVMF_METADATA_SUPPORTED)) {
0344         pr_err("T10-PI is not supported by transport type %d\n",
0345                port->disc_addr.trtype);
0346         ret = -EINVAL;
0347         goto out_put;
0348     }
0349 
0350     ret = ops->add_port(port);
0351     if (ret)
0352         goto out_put;
0353 
0354     /* If the transport didn't set inline_data_size, then disable it. */
0355     if (port->inline_data_size < 0)
0356         port->inline_data_size = 0;
0357 
0358     port->enabled = true;
0359     port->tr_ops = ops;
0360     return 0;
0361 
0362 out_put:
0363     module_put(ops->owner);
0364     return ret;
0365 }
0366 
0367 void nvmet_disable_port(struct nvmet_port *port)
0368 {
0369     const struct nvmet_fabrics_ops *ops;
0370 
0371     lockdep_assert_held(&nvmet_config_sem);
0372 
0373     port->enabled = false;
0374     port->tr_ops = NULL;
0375 
0376     ops = nvmet_transports[port->disc_addr.trtype];
0377     ops->remove_port(port);
0378     module_put(ops->owner);
0379 }
0380 
0381 static void nvmet_keep_alive_timer(struct work_struct *work)
0382 {
0383     struct nvmet_ctrl *ctrl = container_of(to_delayed_work(work),
0384             struct nvmet_ctrl, ka_work);
0385     bool reset_tbkas = ctrl->reset_tbkas;
0386 
0387     ctrl->reset_tbkas = false;
0388     if (reset_tbkas) {
0389         pr_debug("ctrl %d reschedule traffic based keep-alive timer\n",
0390             ctrl->cntlid);
0391         queue_delayed_work(nvmet_wq, &ctrl->ka_work, ctrl->kato * HZ);
0392         return;
0393     }
0394 
0395     pr_err("ctrl %d keep-alive timer (%d seconds) expired!\n",
0396         ctrl->cntlid, ctrl->kato);
0397 
0398     nvmet_ctrl_fatal_error(ctrl);
0399 }
0400 
0401 void nvmet_start_keep_alive_timer(struct nvmet_ctrl *ctrl)
0402 {
0403     if (unlikely(ctrl->kato == 0))
0404         return;
0405 
0406     pr_debug("ctrl %d start keep-alive timer for %d secs\n",
0407         ctrl->cntlid, ctrl->kato);
0408 
0409     queue_delayed_work(nvmet_wq, &ctrl->ka_work, ctrl->kato * HZ);
0410 }
0411 
0412 void nvmet_stop_keep_alive_timer(struct nvmet_ctrl *ctrl)
0413 {
0414     if (unlikely(ctrl->kato == 0))
0415         return;
0416 
0417     pr_debug("ctrl %d stop keep-alive\n", ctrl->cntlid);
0418 
0419     cancel_delayed_work_sync(&ctrl->ka_work);
0420 }
0421 
0422 u16 nvmet_req_find_ns(struct nvmet_req *req)
0423 {
0424     u32 nsid = le32_to_cpu(req->cmd->common.nsid);
0425 
0426     req->ns = xa_load(&nvmet_req_subsys(req)->namespaces, nsid);
0427     if (unlikely(!req->ns)) {
0428         req->error_loc = offsetof(struct nvme_common_command, nsid);
0429         return NVME_SC_INVALID_NS | NVME_SC_DNR;
0430     }
0431 
0432     percpu_ref_get(&req->ns->ref);
0433     return NVME_SC_SUCCESS;
0434 }
0435 
0436 static void nvmet_destroy_namespace(struct percpu_ref *ref)
0437 {
0438     struct nvmet_ns *ns = container_of(ref, struct nvmet_ns, ref);
0439 
0440     complete(&ns->disable_done);
0441 }
0442 
0443 void nvmet_put_namespace(struct nvmet_ns *ns)
0444 {
0445     percpu_ref_put(&ns->ref);
0446 }
0447 
0448 static void nvmet_ns_dev_disable(struct nvmet_ns *ns)
0449 {
0450     nvmet_bdev_ns_disable(ns);
0451     nvmet_file_ns_disable(ns);
0452 }
0453 
0454 static int nvmet_p2pmem_ns_enable(struct nvmet_ns *ns)
0455 {
0456     int ret;
0457     struct pci_dev *p2p_dev;
0458 
0459     if (!ns->use_p2pmem)
0460         return 0;
0461 
0462     if (!ns->bdev) {
0463         pr_err("peer-to-peer DMA is not supported by non-block device namespaces\n");
0464         return -EINVAL;
0465     }
0466 
0467     if (!blk_queue_pci_p2pdma(ns->bdev->bd_disk->queue)) {
0468         pr_err("peer-to-peer DMA is not supported by the driver of %s\n",
0469                ns->device_path);
0470         return -EINVAL;
0471     }
0472 
0473     if (ns->p2p_dev) {
0474         ret = pci_p2pdma_distance(ns->p2p_dev, nvmet_ns_dev(ns), true);
0475         if (ret < 0)
0476             return -EINVAL;
0477     } else {
0478         /*
0479          * Right now we just check that there is p2pmem available so
0480          * we can report an error to the user right away if there
0481          * is not. We'll find the actual device to use once we
0482          * setup the controller when the port's device is available.
0483          */
0484 
0485         p2p_dev = pci_p2pmem_find(nvmet_ns_dev(ns));
0486         if (!p2p_dev) {
0487             pr_err("no peer-to-peer memory is available for %s\n",
0488                    ns->device_path);
0489             return -EINVAL;
0490         }
0491 
0492         pci_dev_put(p2p_dev);
0493     }
0494 
0495     return 0;
0496 }
0497 
0498 /*
0499  * Note: ctrl->subsys->lock should be held when calling this function
0500  */
0501 static void nvmet_p2pmem_ns_add_p2p(struct nvmet_ctrl *ctrl,
0502                     struct nvmet_ns *ns)
0503 {
0504     struct device *clients[2];
0505     struct pci_dev *p2p_dev;
0506     int ret;
0507 
0508     if (!ctrl->p2p_client || !ns->use_p2pmem)
0509         return;
0510 
0511     if (ns->p2p_dev) {
0512         ret = pci_p2pdma_distance(ns->p2p_dev, ctrl->p2p_client, true);
0513         if (ret < 0)
0514             return;
0515 
0516         p2p_dev = pci_dev_get(ns->p2p_dev);
0517     } else {
0518         clients[0] = ctrl->p2p_client;
0519         clients[1] = nvmet_ns_dev(ns);
0520 
0521         p2p_dev = pci_p2pmem_find_many(clients, ARRAY_SIZE(clients));
0522         if (!p2p_dev) {
0523             pr_err("no peer-to-peer memory is available that's supported by %s and %s\n",
0524                    dev_name(ctrl->p2p_client), ns->device_path);
0525             return;
0526         }
0527     }
0528 
0529     ret = radix_tree_insert(&ctrl->p2p_ns_map, ns->nsid, p2p_dev);
0530     if (ret < 0)
0531         pci_dev_put(p2p_dev);
0532 
0533     pr_info("using p2pmem on %s for nsid %d\n", pci_name(p2p_dev),
0534         ns->nsid);
0535 }
0536 
0537 bool nvmet_ns_revalidate(struct nvmet_ns *ns)
0538 {
0539     loff_t oldsize = ns->size;
0540 
0541     if (ns->bdev)
0542         nvmet_bdev_ns_revalidate(ns);
0543     else
0544         nvmet_file_ns_revalidate(ns);
0545 
0546     return oldsize != ns->size;
0547 }
0548 
0549 int nvmet_ns_enable(struct nvmet_ns *ns)
0550 {
0551     struct nvmet_subsys *subsys = ns->subsys;
0552     struct nvmet_ctrl *ctrl;
0553     int ret;
0554 
0555     mutex_lock(&subsys->lock);
0556     ret = 0;
0557 
0558     if (nvmet_is_passthru_subsys(subsys)) {
0559         pr_info("cannot enable both passthru and regular namespaces for a single subsystem");
0560         goto out_unlock;
0561     }
0562 
0563     if (ns->enabled)
0564         goto out_unlock;
0565 
0566     ret = -EMFILE;
0567     if (subsys->nr_namespaces == NVMET_MAX_NAMESPACES)
0568         goto out_unlock;
0569 
0570     ret = nvmet_bdev_ns_enable(ns);
0571     if (ret == -ENOTBLK)
0572         ret = nvmet_file_ns_enable(ns);
0573     if (ret)
0574         goto out_unlock;
0575 
0576     ret = nvmet_p2pmem_ns_enable(ns);
0577     if (ret)
0578         goto out_dev_disable;
0579 
0580     list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry)
0581         nvmet_p2pmem_ns_add_p2p(ctrl, ns);
0582 
0583     ret = percpu_ref_init(&ns->ref, nvmet_destroy_namespace,
0584                 0, GFP_KERNEL);
0585     if (ret)
0586         goto out_dev_put;
0587 
0588     if (ns->nsid > subsys->max_nsid)
0589         subsys->max_nsid = ns->nsid;
0590 
0591     ret = xa_insert(&subsys->namespaces, ns->nsid, ns, GFP_KERNEL);
0592     if (ret)
0593         goto out_restore_subsys_maxnsid;
0594 
0595     subsys->nr_namespaces++;
0596 
0597     nvmet_ns_changed(subsys, ns->nsid);
0598     ns->enabled = true;
0599     ret = 0;
0600 out_unlock:
0601     mutex_unlock(&subsys->lock);
0602     return ret;
0603 
0604 out_restore_subsys_maxnsid:
0605     subsys->max_nsid = nvmet_max_nsid(subsys);
0606     percpu_ref_exit(&ns->ref);
0607 out_dev_put:
0608     list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry)
0609         pci_dev_put(radix_tree_delete(&ctrl->p2p_ns_map, ns->nsid));
0610 out_dev_disable:
0611     nvmet_ns_dev_disable(ns);
0612     goto out_unlock;
0613 }
0614 
0615 void nvmet_ns_disable(struct nvmet_ns *ns)
0616 {
0617     struct nvmet_subsys *subsys = ns->subsys;
0618     struct nvmet_ctrl *ctrl;
0619 
0620     mutex_lock(&subsys->lock);
0621     if (!ns->enabled)
0622         goto out_unlock;
0623 
0624     ns->enabled = false;
0625     xa_erase(&ns->subsys->namespaces, ns->nsid);
0626     if (ns->nsid == subsys->max_nsid)
0627         subsys->max_nsid = nvmet_max_nsid(subsys);
0628 
0629     list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry)
0630         pci_dev_put(radix_tree_delete(&ctrl->p2p_ns_map, ns->nsid));
0631 
0632     mutex_unlock(&subsys->lock);
0633 
0634     /*
0635      * Now that we removed the namespaces from the lookup list, we
0636      * can kill the per_cpu ref and wait for any remaining references
0637      * to be dropped, as well as a RCU grace period for anyone only
0638      * using the namepace under rcu_read_lock().  Note that we can't
0639      * use call_rcu here as we need to ensure the namespaces have
0640      * been fully destroyed before unloading the module.
0641      */
0642     percpu_ref_kill(&ns->ref);
0643     synchronize_rcu();
0644     wait_for_completion(&ns->disable_done);
0645     percpu_ref_exit(&ns->ref);
0646 
0647     mutex_lock(&subsys->lock);
0648 
0649     subsys->nr_namespaces--;
0650     nvmet_ns_changed(subsys, ns->nsid);
0651     nvmet_ns_dev_disable(ns);
0652 out_unlock:
0653     mutex_unlock(&subsys->lock);
0654 }
0655 
0656 void nvmet_ns_free(struct nvmet_ns *ns)
0657 {
0658     nvmet_ns_disable(ns);
0659 
0660     down_write(&nvmet_ana_sem);
0661     nvmet_ana_group_enabled[ns->anagrpid]--;
0662     up_write(&nvmet_ana_sem);
0663 
0664     kfree(ns->device_path);
0665     kfree(ns);
0666 }
0667 
0668 struct nvmet_ns *nvmet_ns_alloc(struct nvmet_subsys *subsys, u32 nsid)
0669 {
0670     struct nvmet_ns *ns;
0671 
0672     ns = kzalloc(sizeof(*ns), GFP_KERNEL);
0673     if (!ns)
0674         return NULL;
0675 
0676     init_completion(&ns->disable_done);
0677 
0678     ns->nsid = nsid;
0679     ns->subsys = subsys;
0680 
0681     down_write(&nvmet_ana_sem);
0682     ns->anagrpid = NVMET_DEFAULT_ANA_GRPID;
0683     nvmet_ana_group_enabled[ns->anagrpid]++;
0684     up_write(&nvmet_ana_sem);
0685 
0686     uuid_gen(&ns->uuid);
0687     ns->buffered_io = false;
0688     ns->csi = NVME_CSI_NVM;
0689 
0690     return ns;
0691 }
0692 
0693 static void nvmet_update_sq_head(struct nvmet_req *req)
0694 {
0695     if (req->sq->size) {
0696         u32 old_sqhd, new_sqhd;
0697 
0698         do {
0699             old_sqhd = req->sq->sqhd;
0700             new_sqhd = (old_sqhd + 1) % req->sq->size;
0701         } while (cmpxchg(&req->sq->sqhd, old_sqhd, new_sqhd) !=
0702                     old_sqhd);
0703     }
0704     req->cqe->sq_head = cpu_to_le16(req->sq->sqhd & 0x0000FFFF);
0705 }
0706 
0707 static void nvmet_set_error(struct nvmet_req *req, u16 status)
0708 {
0709     struct nvmet_ctrl *ctrl = req->sq->ctrl;
0710     struct nvme_error_slot *new_error_slot;
0711     unsigned long flags;
0712 
0713     req->cqe->status = cpu_to_le16(status << 1);
0714 
0715     if (!ctrl || req->error_loc == NVMET_NO_ERROR_LOC)
0716         return;
0717 
0718     spin_lock_irqsave(&ctrl->error_lock, flags);
0719     ctrl->err_counter++;
0720     new_error_slot =
0721         &ctrl->slots[ctrl->err_counter % NVMET_ERROR_LOG_SLOTS];
0722 
0723     new_error_slot->error_count = cpu_to_le64(ctrl->err_counter);
0724     new_error_slot->sqid = cpu_to_le16(req->sq->qid);
0725     new_error_slot->cmdid = cpu_to_le16(req->cmd->common.command_id);
0726     new_error_slot->status_field = cpu_to_le16(status << 1);
0727     new_error_slot->param_error_location = cpu_to_le16(req->error_loc);
0728     new_error_slot->lba = cpu_to_le64(req->error_slba);
0729     new_error_slot->nsid = req->cmd->common.nsid;
0730     spin_unlock_irqrestore(&ctrl->error_lock, flags);
0731 
0732     /* set the more bit for this request */
0733     req->cqe->status |= cpu_to_le16(1 << 14);
0734 }
0735 
0736 static void __nvmet_req_complete(struct nvmet_req *req, u16 status)
0737 {
0738     struct nvmet_ns *ns = req->ns;
0739 
0740     if (!req->sq->sqhd_disabled)
0741         nvmet_update_sq_head(req);
0742     req->cqe->sq_id = cpu_to_le16(req->sq->qid);
0743     req->cqe->command_id = req->cmd->common.command_id;
0744 
0745     if (unlikely(status))
0746         nvmet_set_error(req, status);
0747 
0748     trace_nvmet_req_complete(req);
0749 
0750     req->ops->queue_response(req);
0751     if (ns)
0752         nvmet_put_namespace(ns);
0753 }
0754 
0755 void nvmet_req_complete(struct nvmet_req *req, u16 status)
0756 {
0757     __nvmet_req_complete(req, status);
0758     percpu_ref_put(&req->sq->ref);
0759 }
0760 EXPORT_SYMBOL_GPL(nvmet_req_complete);
0761 
0762 void nvmet_cq_setup(struct nvmet_ctrl *ctrl, struct nvmet_cq *cq,
0763         u16 qid, u16 size)
0764 {
0765     cq->qid = qid;
0766     cq->size = size;
0767 }
0768 
0769 void nvmet_sq_setup(struct nvmet_ctrl *ctrl, struct nvmet_sq *sq,
0770         u16 qid, u16 size)
0771 {
0772     sq->sqhd = 0;
0773     sq->qid = qid;
0774     sq->size = size;
0775 
0776     ctrl->sqs[qid] = sq;
0777 }
0778 
0779 static void nvmet_confirm_sq(struct percpu_ref *ref)
0780 {
0781     struct nvmet_sq *sq = container_of(ref, struct nvmet_sq, ref);
0782 
0783     complete(&sq->confirm_done);
0784 }
0785 
0786 void nvmet_sq_destroy(struct nvmet_sq *sq)
0787 {
0788     struct nvmet_ctrl *ctrl = sq->ctrl;
0789 
0790     /*
0791      * If this is the admin queue, complete all AERs so that our
0792      * queue doesn't have outstanding requests on it.
0793      */
0794     if (ctrl && ctrl->sqs && ctrl->sqs[0] == sq)
0795         nvmet_async_events_failall(ctrl);
0796     percpu_ref_kill_and_confirm(&sq->ref, nvmet_confirm_sq);
0797     wait_for_completion(&sq->confirm_done);
0798     wait_for_completion(&sq->free_done);
0799     percpu_ref_exit(&sq->ref);
0800     nvmet_auth_sq_free(sq);
0801 
0802     if (ctrl) {
0803         /*
0804          * The teardown flow may take some time, and the host may not
0805          * send us keep-alive during this period, hence reset the
0806          * traffic based keep-alive timer so we don't trigger a
0807          * controller teardown as a result of a keep-alive expiration.
0808          */
0809         ctrl->reset_tbkas = true;
0810         sq->ctrl->sqs[sq->qid] = NULL;
0811         nvmet_ctrl_put(ctrl);
0812         sq->ctrl = NULL; /* allows reusing the queue later */
0813     }
0814 }
0815 EXPORT_SYMBOL_GPL(nvmet_sq_destroy);
0816 
0817 static void nvmet_sq_free(struct percpu_ref *ref)
0818 {
0819     struct nvmet_sq *sq = container_of(ref, struct nvmet_sq, ref);
0820 
0821     complete(&sq->free_done);
0822 }
0823 
0824 int nvmet_sq_init(struct nvmet_sq *sq)
0825 {
0826     int ret;
0827 
0828     ret = percpu_ref_init(&sq->ref, nvmet_sq_free, 0, GFP_KERNEL);
0829     if (ret) {
0830         pr_err("percpu_ref init failed!\n");
0831         return ret;
0832     }
0833     init_completion(&sq->free_done);
0834     init_completion(&sq->confirm_done);
0835 
0836     return 0;
0837 }
0838 EXPORT_SYMBOL_GPL(nvmet_sq_init);
0839 
0840 static inline u16 nvmet_check_ana_state(struct nvmet_port *port,
0841         struct nvmet_ns *ns)
0842 {
0843     enum nvme_ana_state state = port->ana_state[ns->anagrpid];
0844 
0845     if (unlikely(state == NVME_ANA_INACCESSIBLE))
0846         return NVME_SC_ANA_INACCESSIBLE;
0847     if (unlikely(state == NVME_ANA_PERSISTENT_LOSS))
0848         return NVME_SC_ANA_PERSISTENT_LOSS;
0849     if (unlikely(state == NVME_ANA_CHANGE))
0850         return NVME_SC_ANA_TRANSITION;
0851     return 0;
0852 }
0853 
0854 static inline u16 nvmet_io_cmd_check_access(struct nvmet_req *req)
0855 {
0856     if (unlikely(req->ns->readonly)) {
0857         switch (req->cmd->common.opcode) {
0858         case nvme_cmd_read:
0859         case nvme_cmd_flush:
0860             break;
0861         default:
0862             return NVME_SC_NS_WRITE_PROTECTED;
0863         }
0864     }
0865 
0866     return 0;
0867 }
0868 
0869 static u16 nvmet_parse_io_cmd(struct nvmet_req *req)
0870 {
0871     struct nvme_command *cmd = req->cmd;
0872     u16 ret;
0873 
0874     if (nvme_is_fabrics(cmd))
0875         return nvmet_parse_fabrics_io_cmd(req);
0876 
0877     if (unlikely(!nvmet_check_auth_status(req)))
0878         return NVME_SC_AUTH_REQUIRED | NVME_SC_DNR;
0879 
0880     ret = nvmet_check_ctrl_status(req);
0881     if (unlikely(ret))
0882         return ret;
0883 
0884     if (nvmet_is_passthru_req(req))
0885         return nvmet_parse_passthru_io_cmd(req);
0886 
0887     ret = nvmet_req_find_ns(req);
0888     if (unlikely(ret))
0889         return ret;
0890 
0891     ret = nvmet_check_ana_state(req->port, req->ns);
0892     if (unlikely(ret)) {
0893         req->error_loc = offsetof(struct nvme_common_command, nsid);
0894         return ret;
0895     }
0896     ret = nvmet_io_cmd_check_access(req);
0897     if (unlikely(ret)) {
0898         req->error_loc = offsetof(struct nvme_common_command, nsid);
0899         return ret;
0900     }
0901 
0902     switch (req->ns->csi) {
0903     case NVME_CSI_NVM:
0904         if (req->ns->file)
0905             return nvmet_file_parse_io_cmd(req);
0906         return nvmet_bdev_parse_io_cmd(req);
0907     case NVME_CSI_ZNS:
0908         if (IS_ENABLED(CONFIG_BLK_DEV_ZONED))
0909             return nvmet_bdev_zns_parse_io_cmd(req);
0910         return NVME_SC_INVALID_IO_CMD_SET;
0911     default:
0912         return NVME_SC_INVALID_IO_CMD_SET;
0913     }
0914 }
0915 
0916 bool nvmet_req_init(struct nvmet_req *req, struct nvmet_cq *cq,
0917         struct nvmet_sq *sq, const struct nvmet_fabrics_ops *ops)
0918 {
0919     u8 flags = req->cmd->common.flags;
0920     u16 status;
0921 
0922     req->cq = cq;
0923     req->sq = sq;
0924     req->ops = ops;
0925     req->sg = NULL;
0926     req->metadata_sg = NULL;
0927     req->sg_cnt = 0;
0928     req->metadata_sg_cnt = 0;
0929     req->transfer_len = 0;
0930     req->metadata_len = 0;
0931     req->cqe->status = 0;
0932     req->cqe->sq_head = 0;
0933     req->ns = NULL;
0934     req->error_loc = NVMET_NO_ERROR_LOC;
0935     req->error_slba = 0;
0936 
0937     /* no support for fused commands yet */
0938     if (unlikely(flags & (NVME_CMD_FUSE_FIRST | NVME_CMD_FUSE_SECOND))) {
0939         req->error_loc = offsetof(struct nvme_common_command, flags);
0940         status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
0941         goto fail;
0942     }
0943 
0944     /*
0945      * For fabrics, PSDT field shall describe metadata pointer (MPTR) that
0946      * contains an address of a single contiguous physical buffer that is
0947      * byte aligned.
0948      */
0949     if (unlikely((flags & NVME_CMD_SGL_ALL) != NVME_CMD_SGL_METABUF)) {
0950         req->error_loc = offsetof(struct nvme_common_command, flags);
0951         status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
0952         goto fail;
0953     }
0954 
0955     if (unlikely(!req->sq->ctrl))
0956         /* will return an error for any non-connect command: */
0957         status = nvmet_parse_connect_cmd(req);
0958     else if (likely(req->sq->qid != 0))
0959         status = nvmet_parse_io_cmd(req);
0960     else
0961         status = nvmet_parse_admin_cmd(req);
0962 
0963     if (status)
0964         goto fail;
0965 
0966     trace_nvmet_req_init(req, req->cmd);
0967 
0968     if (unlikely(!percpu_ref_tryget_live(&sq->ref))) {
0969         status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
0970         goto fail;
0971     }
0972 
0973     if (sq->ctrl)
0974         sq->ctrl->reset_tbkas = true;
0975 
0976     return true;
0977 
0978 fail:
0979     __nvmet_req_complete(req, status);
0980     return false;
0981 }
0982 EXPORT_SYMBOL_GPL(nvmet_req_init);
0983 
0984 void nvmet_req_uninit(struct nvmet_req *req)
0985 {
0986     percpu_ref_put(&req->sq->ref);
0987     if (req->ns)
0988         nvmet_put_namespace(req->ns);
0989 }
0990 EXPORT_SYMBOL_GPL(nvmet_req_uninit);
0991 
0992 bool nvmet_check_transfer_len(struct nvmet_req *req, size_t len)
0993 {
0994     if (unlikely(len != req->transfer_len)) {
0995         req->error_loc = offsetof(struct nvme_common_command, dptr);
0996         nvmet_req_complete(req, NVME_SC_SGL_INVALID_DATA | NVME_SC_DNR);
0997         return false;
0998     }
0999 
1000     return true;
1001 }
1002 EXPORT_SYMBOL_GPL(nvmet_check_transfer_len);
1003 
1004 bool nvmet_check_data_len_lte(struct nvmet_req *req, size_t data_len)
1005 {
1006     if (unlikely(data_len > req->transfer_len)) {
1007         req->error_loc = offsetof(struct nvme_common_command, dptr);
1008         nvmet_req_complete(req, NVME_SC_SGL_INVALID_DATA | NVME_SC_DNR);
1009         return false;
1010     }
1011 
1012     return true;
1013 }
1014 
1015 static unsigned int nvmet_data_transfer_len(struct nvmet_req *req)
1016 {
1017     return req->transfer_len - req->metadata_len;
1018 }
1019 
1020 static int nvmet_req_alloc_p2pmem_sgls(struct pci_dev *p2p_dev,
1021         struct nvmet_req *req)
1022 {
1023     req->sg = pci_p2pmem_alloc_sgl(p2p_dev, &req->sg_cnt,
1024             nvmet_data_transfer_len(req));
1025     if (!req->sg)
1026         goto out_err;
1027 
1028     if (req->metadata_len) {
1029         req->metadata_sg = pci_p2pmem_alloc_sgl(p2p_dev,
1030                 &req->metadata_sg_cnt, req->metadata_len);
1031         if (!req->metadata_sg)
1032             goto out_free_sg;
1033     }
1034 
1035     req->p2p_dev = p2p_dev;
1036 
1037     return 0;
1038 out_free_sg:
1039     pci_p2pmem_free_sgl(req->p2p_dev, req->sg);
1040 out_err:
1041     return -ENOMEM;
1042 }
1043 
1044 static struct pci_dev *nvmet_req_find_p2p_dev(struct nvmet_req *req)
1045 {
1046     if (!IS_ENABLED(CONFIG_PCI_P2PDMA) ||
1047         !req->sq->ctrl || !req->sq->qid || !req->ns)
1048         return NULL;
1049     return radix_tree_lookup(&req->sq->ctrl->p2p_ns_map, req->ns->nsid);
1050 }
1051 
1052 int nvmet_req_alloc_sgls(struct nvmet_req *req)
1053 {
1054     struct pci_dev *p2p_dev = nvmet_req_find_p2p_dev(req);
1055 
1056     if (p2p_dev && !nvmet_req_alloc_p2pmem_sgls(p2p_dev, req))
1057         return 0;
1058 
1059     req->sg = sgl_alloc(nvmet_data_transfer_len(req), GFP_KERNEL,
1060                 &req->sg_cnt);
1061     if (unlikely(!req->sg))
1062         goto out;
1063 
1064     if (req->metadata_len) {
1065         req->metadata_sg = sgl_alloc(req->metadata_len, GFP_KERNEL,
1066                          &req->metadata_sg_cnt);
1067         if (unlikely(!req->metadata_sg))
1068             goto out_free;
1069     }
1070 
1071     return 0;
1072 out_free:
1073     sgl_free(req->sg);
1074 out:
1075     return -ENOMEM;
1076 }
1077 EXPORT_SYMBOL_GPL(nvmet_req_alloc_sgls);
1078 
1079 void nvmet_req_free_sgls(struct nvmet_req *req)
1080 {
1081     if (req->p2p_dev) {
1082         pci_p2pmem_free_sgl(req->p2p_dev, req->sg);
1083         if (req->metadata_sg)
1084             pci_p2pmem_free_sgl(req->p2p_dev, req->metadata_sg);
1085         req->p2p_dev = NULL;
1086     } else {
1087         sgl_free(req->sg);
1088         if (req->metadata_sg)
1089             sgl_free(req->metadata_sg);
1090     }
1091 
1092     req->sg = NULL;
1093     req->metadata_sg = NULL;
1094     req->sg_cnt = 0;
1095     req->metadata_sg_cnt = 0;
1096 }
1097 EXPORT_SYMBOL_GPL(nvmet_req_free_sgls);
1098 
1099 static inline bool nvmet_cc_en(u32 cc)
1100 {
1101     return (cc >> NVME_CC_EN_SHIFT) & 0x1;
1102 }
1103 
1104 static inline u8 nvmet_cc_css(u32 cc)
1105 {
1106     return (cc >> NVME_CC_CSS_SHIFT) & 0x7;
1107 }
1108 
1109 static inline u8 nvmet_cc_mps(u32 cc)
1110 {
1111     return (cc >> NVME_CC_MPS_SHIFT) & 0xf;
1112 }
1113 
1114 static inline u8 nvmet_cc_ams(u32 cc)
1115 {
1116     return (cc >> NVME_CC_AMS_SHIFT) & 0x7;
1117 }
1118 
1119 static inline u8 nvmet_cc_shn(u32 cc)
1120 {
1121     return (cc >> NVME_CC_SHN_SHIFT) & 0x3;
1122 }
1123 
1124 static inline u8 nvmet_cc_iosqes(u32 cc)
1125 {
1126     return (cc >> NVME_CC_IOSQES_SHIFT) & 0xf;
1127 }
1128 
1129 static inline u8 nvmet_cc_iocqes(u32 cc)
1130 {
1131     return (cc >> NVME_CC_IOCQES_SHIFT) & 0xf;
1132 }
1133 
1134 static inline bool nvmet_css_supported(u8 cc_css)
1135 {
1136     switch (cc_css << NVME_CC_CSS_SHIFT) {
1137     case NVME_CC_CSS_NVM:
1138     case NVME_CC_CSS_CSI:
1139         return true;
1140     default:
1141         return false;
1142     }
1143 }
1144 
1145 static void nvmet_start_ctrl(struct nvmet_ctrl *ctrl)
1146 {
1147     lockdep_assert_held(&ctrl->lock);
1148 
1149     /*
1150      * Only I/O controllers should verify iosqes,iocqes.
1151      * Strictly speaking, the spec says a discovery controller
1152      * should verify iosqes,iocqes are zeroed, however that
1153      * would break backwards compatibility, so don't enforce it.
1154      */
1155     if (!nvmet_is_disc_subsys(ctrl->subsys) &&
1156         (nvmet_cc_iosqes(ctrl->cc) != NVME_NVM_IOSQES ||
1157          nvmet_cc_iocqes(ctrl->cc) != NVME_NVM_IOCQES)) {
1158         ctrl->csts = NVME_CSTS_CFS;
1159         return;
1160     }
1161 
1162     if (nvmet_cc_mps(ctrl->cc) != 0 ||
1163         nvmet_cc_ams(ctrl->cc) != 0 ||
1164         !nvmet_css_supported(nvmet_cc_css(ctrl->cc))) {
1165         ctrl->csts = NVME_CSTS_CFS;
1166         return;
1167     }
1168 
1169     ctrl->csts = NVME_CSTS_RDY;
1170 
1171     /*
1172      * Controllers that are not yet enabled should not really enforce the
1173      * keep alive timeout, but we still want to track a timeout and cleanup
1174      * in case a host died before it enabled the controller.  Hence, simply
1175      * reset the keep alive timer when the controller is enabled.
1176      */
1177     if (ctrl->kato)
1178         mod_delayed_work(system_wq, &ctrl->ka_work, ctrl->kato * HZ);
1179 }
1180 
1181 static void nvmet_clear_ctrl(struct nvmet_ctrl *ctrl)
1182 {
1183     lockdep_assert_held(&ctrl->lock);
1184 
1185     /* XXX: tear down queues? */
1186     ctrl->csts &= ~NVME_CSTS_RDY;
1187     ctrl->cc = 0;
1188 }
1189 
1190 void nvmet_update_cc(struct nvmet_ctrl *ctrl, u32 new)
1191 {
1192     u32 old;
1193 
1194     mutex_lock(&ctrl->lock);
1195     old = ctrl->cc;
1196     ctrl->cc = new;
1197 
1198     if (nvmet_cc_en(new) && !nvmet_cc_en(old))
1199         nvmet_start_ctrl(ctrl);
1200     if (!nvmet_cc_en(new) && nvmet_cc_en(old))
1201         nvmet_clear_ctrl(ctrl);
1202     if (nvmet_cc_shn(new) && !nvmet_cc_shn(old)) {
1203         nvmet_clear_ctrl(ctrl);
1204         ctrl->csts |= NVME_CSTS_SHST_CMPLT;
1205     }
1206     if (!nvmet_cc_shn(new) && nvmet_cc_shn(old))
1207         ctrl->csts &= ~NVME_CSTS_SHST_CMPLT;
1208     mutex_unlock(&ctrl->lock);
1209 }
1210 
1211 static void nvmet_init_cap(struct nvmet_ctrl *ctrl)
1212 {
1213     /* command sets supported: NVMe command set: */
1214     ctrl->cap = (1ULL << 37);
1215     /* Controller supports one or more I/O Command Sets */
1216     ctrl->cap |= (1ULL << 43);
1217     /* CC.EN timeout in 500msec units: */
1218     ctrl->cap |= (15ULL << 24);
1219     /* maximum queue entries supported: */
1220     if (ctrl->ops->get_max_queue_size)
1221         ctrl->cap |= ctrl->ops->get_max_queue_size(ctrl) - 1;
1222     else
1223         ctrl->cap |= NVMET_QUEUE_SIZE - 1;
1224 
1225     if (nvmet_is_passthru_subsys(ctrl->subsys))
1226         nvmet_passthrough_override_cap(ctrl);
1227 }
1228 
1229 struct nvmet_ctrl *nvmet_ctrl_find_get(const char *subsysnqn,
1230                        const char *hostnqn, u16 cntlid,
1231                        struct nvmet_req *req)
1232 {
1233     struct nvmet_ctrl *ctrl = NULL;
1234     struct nvmet_subsys *subsys;
1235 
1236     subsys = nvmet_find_get_subsys(req->port, subsysnqn);
1237     if (!subsys) {
1238         pr_warn("connect request for invalid subsystem %s!\n",
1239             subsysnqn);
1240         req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(subsysnqn);
1241         goto out;
1242     }
1243 
1244     mutex_lock(&subsys->lock);
1245     list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
1246         if (ctrl->cntlid == cntlid) {
1247             if (strncmp(hostnqn, ctrl->hostnqn, NVMF_NQN_SIZE)) {
1248                 pr_warn("hostnqn mismatch.\n");
1249                 continue;
1250             }
1251             if (!kref_get_unless_zero(&ctrl->ref))
1252                 continue;
1253 
1254             /* ctrl found */
1255             goto found;
1256         }
1257     }
1258 
1259     ctrl = NULL; /* ctrl not found */
1260     pr_warn("could not find controller %d for subsys %s / host %s\n",
1261         cntlid, subsysnqn, hostnqn);
1262     req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(cntlid);
1263 
1264 found:
1265     mutex_unlock(&subsys->lock);
1266     nvmet_subsys_put(subsys);
1267 out:
1268     return ctrl;
1269 }
1270 
1271 u16 nvmet_check_ctrl_status(struct nvmet_req *req)
1272 {
1273     if (unlikely(!(req->sq->ctrl->cc & NVME_CC_ENABLE))) {
1274         pr_err("got cmd %d while CC.EN == 0 on qid = %d\n",
1275                req->cmd->common.opcode, req->sq->qid);
1276         return NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR;
1277     }
1278 
1279     if (unlikely(!(req->sq->ctrl->csts & NVME_CSTS_RDY))) {
1280         pr_err("got cmd %d while CSTS.RDY == 0 on qid = %d\n",
1281                req->cmd->common.opcode, req->sq->qid);
1282         return NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR;
1283     }
1284 
1285     if (unlikely(!nvmet_check_auth_status(req))) {
1286         pr_warn("qid %d not authenticated\n", req->sq->qid);
1287         return NVME_SC_AUTH_REQUIRED | NVME_SC_DNR;
1288     }
1289     return 0;
1290 }
1291 
1292 bool nvmet_host_allowed(struct nvmet_subsys *subsys, const char *hostnqn)
1293 {
1294     struct nvmet_host_link *p;
1295 
1296     lockdep_assert_held(&nvmet_config_sem);
1297 
1298     if (subsys->allow_any_host)
1299         return true;
1300 
1301     if (nvmet_is_disc_subsys(subsys)) /* allow all access to disc subsys */
1302         return true;
1303 
1304     list_for_each_entry(p, &subsys->hosts, entry) {
1305         if (!strcmp(nvmet_host_name(p->host), hostnqn))
1306             return true;
1307     }
1308 
1309     return false;
1310 }
1311 
1312 /*
1313  * Note: ctrl->subsys->lock should be held when calling this function
1314  */
1315 static void nvmet_setup_p2p_ns_map(struct nvmet_ctrl *ctrl,
1316         struct nvmet_req *req)
1317 {
1318     struct nvmet_ns *ns;
1319     unsigned long idx;
1320 
1321     if (!req->p2p_client)
1322         return;
1323 
1324     ctrl->p2p_client = get_device(req->p2p_client);
1325 
1326     xa_for_each(&ctrl->subsys->namespaces, idx, ns)
1327         nvmet_p2pmem_ns_add_p2p(ctrl, ns);
1328 }
1329 
1330 /*
1331  * Note: ctrl->subsys->lock should be held when calling this function
1332  */
1333 static void nvmet_release_p2p_ns_map(struct nvmet_ctrl *ctrl)
1334 {
1335     struct radix_tree_iter iter;
1336     void __rcu **slot;
1337 
1338     radix_tree_for_each_slot(slot, &ctrl->p2p_ns_map, &iter, 0)
1339         pci_dev_put(radix_tree_deref_slot(slot));
1340 
1341     put_device(ctrl->p2p_client);
1342 }
1343 
1344 static void nvmet_fatal_error_handler(struct work_struct *work)
1345 {
1346     struct nvmet_ctrl *ctrl =
1347             container_of(work, struct nvmet_ctrl, fatal_err_work);
1348 
1349     pr_err("ctrl %d fatal error occurred!\n", ctrl->cntlid);
1350     ctrl->ops->delete_ctrl(ctrl);
1351 }
1352 
1353 u16 nvmet_alloc_ctrl(const char *subsysnqn, const char *hostnqn,
1354         struct nvmet_req *req, u32 kato, struct nvmet_ctrl **ctrlp)
1355 {
1356     struct nvmet_subsys *subsys;
1357     struct nvmet_ctrl *ctrl;
1358     int ret;
1359     u16 status;
1360 
1361     status = NVME_SC_CONNECT_INVALID_PARAM | NVME_SC_DNR;
1362     subsys = nvmet_find_get_subsys(req->port, subsysnqn);
1363     if (!subsys) {
1364         pr_warn("connect request for invalid subsystem %s!\n",
1365             subsysnqn);
1366         req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(subsysnqn);
1367         req->error_loc = offsetof(struct nvme_common_command, dptr);
1368         goto out;
1369     }
1370 
1371     down_read(&nvmet_config_sem);
1372     if (!nvmet_host_allowed(subsys, hostnqn)) {
1373         pr_info("connect by host %s for subsystem %s not allowed\n",
1374             hostnqn, subsysnqn);
1375         req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(hostnqn);
1376         up_read(&nvmet_config_sem);
1377         status = NVME_SC_CONNECT_INVALID_HOST | NVME_SC_DNR;
1378         req->error_loc = offsetof(struct nvme_common_command, dptr);
1379         goto out_put_subsystem;
1380     }
1381     up_read(&nvmet_config_sem);
1382 
1383     status = NVME_SC_INTERNAL;
1384     ctrl = kzalloc(sizeof(*ctrl), GFP_KERNEL);
1385     if (!ctrl)
1386         goto out_put_subsystem;
1387     mutex_init(&ctrl->lock);
1388 
1389     ctrl->port = req->port;
1390     ctrl->ops = req->ops;
1391 
1392 #ifdef CONFIG_NVME_TARGET_PASSTHRU
1393     /* By default, set loop targets to clear IDS by default */
1394     if (ctrl->port->disc_addr.trtype == NVMF_TRTYPE_LOOP)
1395         subsys->clear_ids = 1;
1396 #endif
1397 
1398     INIT_WORK(&ctrl->async_event_work, nvmet_async_event_work);
1399     INIT_LIST_HEAD(&ctrl->async_events);
1400     INIT_RADIX_TREE(&ctrl->p2p_ns_map, GFP_KERNEL);
1401     INIT_WORK(&ctrl->fatal_err_work, nvmet_fatal_error_handler);
1402     INIT_DELAYED_WORK(&ctrl->ka_work, nvmet_keep_alive_timer);
1403 
1404     memcpy(ctrl->subsysnqn, subsysnqn, NVMF_NQN_SIZE);
1405     memcpy(ctrl->hostnqn, hostnqn, NVMF_NQN_SIZE);
1406 
1407     kref_init(&ctrl->ref);
1408     ctrl->subsys = subsys;
1409     nvmet_init_cap(ctrl);
1410     WRITE_ONCE(ctrl->aen_enabled, NVMET_AEN_CFG_OPTIONAL);
1411 
1412     ctrl->changed_ns_list = kmalloc_array(NVME_MAX_CHANGED_NAMESPACES,
1413             sizeof(__le32), GFP_KERNEL);
1414     if (!ctrl->changed_ns_list)
1415         goto out_free_ctrl;
1416 
1417     ctrl->sqs = kcalloc(subsys->max_qid + 1,
1418             sizeof(struct nvmet_sq *),
1419             GFP_KERNEL);
1420     if (!ctrl->sqs)
1421         goto out_free_changed_ns_list;
1422 
1423     if (subsys->cntlid_min > subsys->cntlid_max)
1424         goto out_free_sqs;
1425 
1426     ret = ida_alloc_range(&cntlid_ida,
1427                  subsys->cntlid_min, subsys->cntlid_max,
1428                  GFP_KERNEL);
1429     if (ret < 0) {
1430         status = NVME_SC_CONNECT_CTRL_BUSY | NVME_SC_DNR;
1431         goto out_free_sqs;
1432     }
1433     ctrl->cntlid = ret;
1434 
1435     /*
1436      * Discovery controllers may use some arbitrary high value
1437      * in order to cleanup stale discovery sessions
1438      */
1439     if (nvmet_is_disc_subsys(ctrl->subsys) && !kato)
1440         kato = NVMET_DISC_KATO_MS;
1441 
1442     /* keep-alive timeout in seconds */
1443     ctrl->kato = DIV_ROUND_UP(kato, 1000);
1444 
1445     ctrl->err_counter = 0;
1446     spin_lock_init(&ctrl->error_lock);
1447 
1448     nvmet_start_keep_alive_timer(ctrl);
1449 
1450     mutex_lock(&subsys->lock);
1451     list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
1452     nvmet_setup_p2p_ns_map(ctrl, req);
1453     mutex_unlock(&subsys->lock);
1454 
1455     *ctrlp = ctrl;
1456     return 0;
1457 
1458 out_free_sqs:
1459     kfree(ctrl->sqs);
1460 out_free_changed_ns_list:
1461     kfree(ctrl->changed_ns_list);
1462 out_free_ctrl:
1463     kfree(ctrl);
1464 out_put_subsystem:
1465     nvmet_subsys_put(subsys);
1466 out:
1467     return status;
1468 }
1469 
1470 static void nvmet_ctrl_free(struct kref *ref)
1471 {
1472     struct nvmet_ctrl *ctrl = container_of(ref, struct nvmet_ctrl, ref);
1473     struct nvmet_subsys *subsys = ctrl->subsys;
1474 
1475     mutex_lock(&subsys->lock);
1476     nvmet_release_p2p_ns_map(ctrl);
1477     list_del(&ctrl->subsys_entry);
1478     mutex_unlock(&subsys->lock);
1479 
1480     nvmet_stop_keep_alive_timer(ctrl);
1481 
1482     flush_work(&ctrl->async_event_work);
1483     cancel_work_sync(&ctrl->fatal_err_work);
1484 
1485     nvmet_destroy_auth(ctrl);
1486 
1487     ida_free(&cntlid_ida, ctrl->cntlid);
1488 
1489     nvmet_async_events_free(ctrl);
1490     kfree(ctrl->sqs);
1491     kfree(ctrl->changed_ns_list);
1492     kfree(ctrl);
1493 
1494     nvmet_subsys_put(subsys);
1495 }
1496 
1497 void nvmet_ctrl_put(struct nvmet_ctrl *ctrl)
1498 {
1499     kref_put(&ctrl->ref, nvmet_ctrl_free);
1500 }
1501 
1502 void nvmet_ctrl_fatal_error(struct nvmet_ctrl *ctrl)
1503 {
1504     mutex_lock(&ctrl->lock);
1505     if (!(ctrl->csts & NVME_CSTS_CFS)) {
1506         ctrl->csts |= NVME_CSTS_CFS;
1507         queue_work(nvmet_wq, &ctrl->fatal_err_work);
1508     }
1509     mutex_unlock(&ctrl->lock);
1510 }
1511 EXPORT_SYMBOL_GPL(nvmet_ctrl_fatal_error);
1512 
1513 static struct nvmet_subsys *nvmet_find_get_subsys(struct nvmet_port *port,
1514         const char *subsysnqn)
1515 {
1516     struct nvmet_subsys_link *p;
1517 
1518     if (!port)
1519         return NULL;
1520 
1521     if (!strcmp(NVME_DISC_SUBSYS_NAME, subsysnqn)) {
1522         if (!kref_get_unless_zero(&nvmet_disc_subsys->ref))
1523             return NULL;
1524         return nvmet_disc_subsys;
1525     }
1526 
1527     down_read(&nvmet_config_sem);
1528     list_for_each_entry(p, &port->subsystems, entry) {
1529         if (!strncmp(p->subsys->subsysnqn, subsysnqn,
1530                 NVMF_NQN_SIZE)) {
1531             if (!kref_get_unless_zero(&p->subsys->ref))
1532                 break;
1533             up_read(&nvmet_config_sem);
1534             return p->subsys;
1535         }
1536     }
1537     up_read(&nvmet_config_sem);
1538     return NULL;
1539 }
1540 
1541 struct nvmet_subsys *nvmet_subsys_alloc(const char *subsysnqn,
1542         enum nvme_subsys_type type)
1543 {
1544     struct nvmet_subsys *subsys;
1545     char serial[NVMET_SN_MAX_SIZE / 2];
1546     int ret;
1547 
1548     subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
1549     if (!subsys)
1550         return ERR_PTR(-ENOMEM);
1551 
1552     subsys->ver = NVMET_DEFAULT_VS;
1553     /* generate a random serial number as our controllers are ephemeral: */
1554     get_random_bytes(&serial, sizeof(serial));
1555     bin2hex(subsys->serial, &serial, sizeof(serial));
1556 
1557     subsys->model_number = kstrdup(NVMET_DEFAULT_CTRL_MODEL, GFP_KERNEL);
1558     if (!subsys->model_number) {
1559         ret = -ENOMEM;
1560         goto free_subsys;
1561     }
1562 
1563     switch (type) {
1564     case NVME_NQN_NVME:
1565         subsys->max_qid = NVMET_NR_QUEUES;
1566         break;
1567     case NVME_NQN_DISC:
1568     case NVME_NQN_CURR:
1569         subsys->max_qid = 0;
1570         break;
1571     default:
1572         pr_err("%s: Unknown Subsystem type - %d\n", __func__, type);
1573         ret = -EINVAL;
1574         goto free_mn;
1575     }
1576     subsys->type = type;
1577     subsys->subsysnqn = kstrndup(subsysnqn, NVMF_NQN_SIZE,
1578             GFP_KERNEL);
1579     if (!subsys->subsysnqn) {
1580         ret = -ENOMEM;
1581         goto free_mn;
1582     }
1583     subsys->cntlid_min = NVME_CNTLID_MIN;
1584     subsys->cntlid_max = NVME_CNTLID_MAX;
1585     kref_init(&subsys->ref);
1586 
1587     mutex_init(&subsys->lock);
1588     xa_init(&subsys->namespaces);
1589     INIT_LIST_HEAD(&subsys->ctrls);
1590     INIT_LIST_HEAD(&subsys->hosts);
1591 
1592     return subsys;
1593 
1594 free_mn:
1595     kfree(subsys->model_number);
1596 free_subsys:
1597     kfree(subsys);
1598     return ERR_PTR(ret);
1599 }
1600 
1601 static void nvmet_subsys_free(struct kref *ref)
1602 {
1603     struct nvmet_subsys *subsys =
1604         container_of(ref, struct nvmet_subsys, ref);
1605 
1606     WARN_ON_ONCE(!xa_empty(&subsys->namespaces));
1607 
1608     xa_destroy(&subsys->namespaces);
1609     nvmet_passthru_subsys_free(subsys);
1610 
1611     kfree(subsys->subsysnqn);
1612     kfree(subsys->model_number);
1613     kfree(subsys);
1614 }
1615 
1616 void nvmet_subsys_del_ctrls(struct nvmet_subsys *subsys)
1617 {
1618     struct nvmet_ctrl *ctrl;
1619 
1620     mutex_lock(&subsys->lock);
1621     list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry)
1622         ctrl->ops->delete_ctrl(ctrl);
1623     mutex_unlock(&subsys->lock);
1624 }
1625 
1626 void nvmet_subsys_put(struct nvmet_subsys *subsys)
1627 {
1628     kref_put(&subsys->ref, nvmet_subsys_free);
1629 }
1630 
1631 static int __init nvmet_init(void)
1632 {
1633     int error;
1634 
1635     nvmet_ana_group_enabled[NVMET_DEFAULT_ANA_GRPID] = 1;
1636 
1637     zbd_wq = alloc_workqueue("nvmet-zbd-wq", WQ_MEM_RECLAIM, 0);
1638     if (!zbd_wq)
1639         return -ENOMEM;
1640 
1641     buffered_io_wq = alloc_workqueue("nvmet-buffered-io-wq",
1642             WQ_MEM_RECLAIM, 0);
1643     if (!buffered_io_wq) {
1644         error = -ENOMEM;
1645         goto out_free_zbd_work_queue;
1646     }
1647 
1648     nvmet_wq = alloc_workqueue("nvmet-wq", WQ_MEM_RECLAIM, 0);
1649     if (!nvmet_wq) {
1650         error = -ENOMEM;
1651         goto out_free_buffered_work_queue;
1652     }
1653 
1654     error = nvmet_init_discovery();
1655     if (error)
1656         goto out_free_nvmet_work_queue;
1657 
1658     error = nvmet_init_configfs();
1659     if (error)
1660         goto out_exit_discovery;
1661     return 0;
1662 
1663 out_exit_discovery:
1664     nvmet_exit_discovery();
1665 out_free_nvmet_work_queue:
1666     destroy_workqueue(nvmet_wq);
1667 out_free_buffered_work_queue:
1668     destroy_workqueue(buffered_io_wq);
1669 out_free_zbd_work_queue:
1670     destroy_workqueue(zbd_wq);
1671     return error;
1672 }
1673 
1674 static void __exit nvmet_exit(void)
1675 {
1676     nvmet_exit_configfs();
1677     nvmet_exit_discovery();
1678     ida_destroy(&cntlid_ida);
1679     destroy_workqueue(nvmet_wq);
1680     destroy_workqueue(buffered_io_wq);
1681     destroy_workqueue(zbd_wq);
1682 
1683     BUILD_BUG_ON(sizeof(struct nvmf_disc_rsp_page_entry) != 1024);
1684     BUILD_BUG_ON(sizeof(struct nvmf_disc_rsp_page_hdr) != 1024);
1685 }
1686 
1687 module_init(nvmet_init);
1688 module_exit(nvmet_exit);
1689 
1690 MODULE_LICENSE("GPL v2");
[ Source navigation ] [ Diff markup ] [ Identifier search ] [ general search ]

This page was automatically generated by the 2.1.0 LXR engine.

The LXR team
Valid CSS 2.1! Valid HTML 4.01!