diff options
Diffstat (limited to 'drivers/infiniband/core/ucm.c')
| -rw-r--r-- | drivers/infiniband/core/ucm.c | 715 |
1 files changed, 424 insertions, 291 deletions
diff --git a/drivers/infiniband/core/ucm.c b/drivers/infiniband/core/ucm.c index 79595826ccc..f2f63933e8a 100644 --- a/drivers/infiniband/core/ucm.c +++ b/drivers/infiniband/core/ucm.c @@ -29,56 +29,103 @@ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE * SOFTWARE. - * - * $Id: ucm.c 2594 2005-06-13 19:46:02Z libor $ */ + +#include <linux/completion.h> #include <linux/init.h> #include <linux/fs.h> #include <linux/module.h> #include <linux/device.h> #include <linux/err.h> #include <linux/poll.h> +#include <linux/sched.h> #include <linux/file.h> #include <linux/mount.h> #include <linux/cdev.h> +#include <linux/idr.h> +#include <linux/mutex.h> +#include <linux/slab.h> #include <asm/uaccess.h> -#include "ucm.h" +#include <rdma/ib_cm.h> +#include <rdma/ib_user_cm.h> +#include <rdma/ib_marshall.h> MODULE_AUTHOR("Libor Michalek"); MODULE_DESCRIPTION("InfiniBand userspace Connection Manager access"); MODULE_LICENSE("Dual BSD/GPL"); -static int ucm_debug_level; +struct ib_ucm_device { + int devnum; + struct cdev cdev; + struct device dev; + struct ib_device *ib_dev; +}; + +struct ib_ucm_file { + struct mutex file_mutex; + struct file *filp; + struct ib_ucm_device *device; + + struct list_head ctxs; + struct list_head events; + wait_queue_head_t poll_wait; +}; + +struct ib_ucm_context { + int id; + struct completion comp; + atomic_t ref; + int events_reported; -module_param_named(debug_level, ucm_debug_level, int, 0644); -MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0"); + struct ib_ucm_file *file; + struct ib_cm_id *cm_id; + __u64 uid; + + struct list_head events; /* list of pending events. */ + struct list_head file_list; /* member in file ctx list */ +}; + +struct ib_ucm_event { + struct ib_ucm_context *ctx; + struct list_head file_list; /* member in file event list */ + struct list_head ctx_list; /* member in ctx event list */ + + struct ib_cm_id *cm_id; + struct ib_ucm_event_resp resp; + void *data; + void *info; + int data_len; + int info_len; +}; enum { IB_UCM_MAJOR = 231, - IB_UCM_MINOR = 255 + IB_UCM_BASE_MINOR = 224, + IB_UCM_MAX_DEVICES = 32 }; -#define IB_UCM_DEV MKDEV(IB_UCM_MAJOR, IB_UCM_MINOR) +#define IB_UCM_BASE_DEV MKDEV(IB_UCM_MAJOR, IB_UCM_BASE_MINOR) -#define PFX "UCM: " +static void ib_ucm_add_one(struct ib_device *device); +static void ib_ucm_remove_one(struct ib_device *device); -#define ucm_dbg(format, arg...) \ - do { \ - if (ucm_debug_level > 0) \ - printk(KERN_DEBUG PFX format, ## arg); \ - } while (0) +static struct ib_client ucm_client = { + .name = "ucm", + .add = ib_ucm_add_one, + .remove = ib_ucm_remove_one +}; -static struct semaphore ctx_id_mutex; -static struct idr ctx_id_table; -static int ctx_id_rover = 0; +static DEFINE_MUTEX(ctx_id_mutex); +static DEFINE_IDR(ctx_id_table); +static DECLARE_BITMAP(dev_map, IB_UCM_MAX_DEVICES); static struct ib_ucm_context *ib_ucm_ctx_get(struct ib_ucm_file *file, int id) { struct ib_ucm_context *ctx; - down(&ctx_id_mutex); + mutex_lock(&ctx_id_mutex); ctx = idr_find(&ctx_id_table, id); if (!ctx) ctx = ERR_PTR(-ENOENT); @@ -86,7 +133,7 @@ static struct ib_ucm_context *ib_ucm_ctx_get(struct ib_ucm_file *file, int id) ctx = ERR_PTR(-EINVAL); else atomic_inc(&ctx->ref); - up(&ctx_id_mutex); + mutex_unlock(&ctx_id_mutex); return ctx; } @@ -94,36 +141,19 @@ static struct ib_ucm_context *ib_ucm_ctx_get(struct ib_ucm_file *file, int id) static void ib_ucm_ctx_put(struct ib_ucm_context *ctx) { if (atomic_dec_and_test(&ctx->ref)) - wake_up(&ctx->wait); + complete(&ctx->comp); } -static ssize_t ib_ucm_destroy_ctx(struct ib_ucm_file *file, int id) +static inline int ib_ucm_new_cm_id(int event) { - struct ib_ucm_context *ctx; - struct ib_ucm_event *uevent; - - down(&ctx_id_mutex); - ctx = idr_find(&ctx_id_table, id); - if (!ctx) - ctx = ERR_PTR(-ENOENT); - else if (ctx->file != file) - ctx = ERR_PTR(-EINVAL); - else - idr_remove(&ctx_id_table, ctx->id); - up(&ctx_id_mutex); - - if (IS_ERR(ctx)) - return PTR_ERR(ctx); - - atomic_dec(&ctx->ref); - wait_event(ctx->wait, !atomic_read(&ctx->ref)); + return event == IB_CM_REQ_RECEIVED || event == IB_CM_SIDR_REQ_RECEIVED; +} - /* No new events will be generated after destroying the cm_id. */ - if (!IS_ERR(ctx->cm_id)) - ib_destroy_cm_id(ctx->cm_id); +static void ib_ucm_cleanup_events(struct ib_ucm_context *ctx) +{ + struct ib_ucm_event *uevent; - /* Cleanup events not yet reported to the user. */ - down(&file->mutex); + mutex_lock(&ctx->file->file_mutex); list_del(&ctx->file_list); while (!list_empty(&ctx->events)) { @@ -131,100 +161,48 @@ static ssize_t ib_ucm_destroy_ctx(struct ib_ucm_file *file, int id) struct ib_ucm_event, ctx_list); list_del(&uevent->file_list); list_del(&uevent->ctx_list); + mutex_unlock(&ctx->file->file_mutex); /* clear incoming connections. */ - if (uevent->cm_id) + if (ib_ucm_new_cm_id(uevent->resp.event)) ib_destroy_cm_id(uevent->cm_id); kfree(uevent); + mutex_lock(&ctx->file->file_mutex); } - up(&file->mutex); - - kfree(ctx); - return 0; + mutex_unlock(&ctx->file->file_mutex); } static struct ib_ucm_context *ib_ucm_ctx_alloc(struct ib_ucm_file *file) { struct ib_ucm_context *ctx; - int result; - ctx = kmalloc(sizeof(*ctx), GFP_KERNEL); + ctx = kzalloc(sizeof *ctx, GFP_KERNEL); if (!ctx) return NULL; atomic_set(&ctx->ref, 1); - init_waitqueue_head(&ctx->wait); + init_completion(&ctx->comp); ctx->file = file; - INIT_LIST_HEAD(&ctx->events); - list_add_tail(&ctx->file_list, &file->ctxs); - - ctx_id_rover = (ctx_id_rover + 1) & INT_MAX; -retry: - result = idr_pre_get(&ctx_id_table, GFP_KERNEL); - if (!result) + mutex_lock(&ctx_id_mutex); + ctx->id = idr_alloc(&ctx_id_table, ctx, 0, 0, GFP_KERNEL); + mutex_unlock(&ctx_id_mutex); + if (ctx->id < 0) goto error; - down(&ctx_id_mutex); - result = idr_get_new_above(&ctx_id_table, ctx, ctx_id_rover, &ctx->id); - up(&ctx_id_mutex); - - if (result == -EAGAIN) - goto retry; - if (result) - goto error; - - ucm_dbg("Allocated CM ID <%d>\n", ctx->id); - + list_add_tail(&ctx->file_list, &file->ctxs); return ctx; + error: - list_del(&ctx->file_list); kfree(ctx); - return NULL; } -/* - * Event portion of the API, handle CM events - * and allow event polling. - */ -static void ib_ucm_event_path_get(struct ib_ucm_path_rec *upath, - struct ib_sa_path_rec *kpath) -{ - if (!kpath || !upath) - return; - - memcpy(upath->dgid, kpath->dgid.raw, sizeof *upath->dgid); - memcpy(upath->sgid, kpath->sgid.raw, sizeof *upath->sgid); - - upath->dlid = kpath->dlid; - upath->slid = kpath->slid; - upath->raw_traffic = kpath->raw_traffic; - upath->flow_label = kpath->flow_label; - upath->hop_limit = kpath->hop_limit; - upath->traffic_class = kpath->traffic_class; - upath->reversible = kpath->reversible; - upath->numb_path = kpath->numb_path; - upath->pkey = kpath->pkey; - upath->sl = kpath->sl; - upath->mtu_selector = kpath->mtu_selector; - upath->mtu = kpath->mtu; - upath->rate_selector = kpath->rate_selector; - upath->rate = kpath->rate; - upath->packet_life_time = kpath->packet_life_time; - upath->preference = kpath->preference; - - upath->packet_life_time_selector = - kpath->packet_life_time_selector; -} -static void ib_ucm_event_req_get(struct ib_ucm_context *ctx, - struct ib_ucm_req_event_resp *ureq, +static void ib_ucm_event_req_get(struct ib_ucm_req_event_resp *ureq, struct ib_cm_req_event_param *kreq) { - ureq->listen_id = ctx->id; - ureq->remote_ca_guid = kreq->remote_ca_guid; ureq->remote_qkey = kreq->remote_qkey; ureq->remote_qpn = kreq->remote_qpn; @@ -238,9 +216,12 @@ static void ib_ucm_event_req_get(struct ib_ucm_context *ctx, ureq->retry_count = kreq->retry_count; ureq->rnr_retry_count = kreq->rnr_retry_count; ureq->srq = kreq->srq; + ureq->port = kreq->port; - ib_ucm_event_path_get(&ureq->primary_path, kreq->primary_path); - ib_ucm_event_path_get(&ureq->alternate_path, kreq->alternate_path); + ib_copy_path_rec_to_user(&ureq->primary_path, kreq->primary_path); + if (kreq->alternate_path) + ib_copy_path_rec_to_user(&ureq->alternate_path, + kreq->alternate_path); } static void ib_ucm_event_rep_get(struct ib_ucm_rep_event_resp *urep, @@ -259,14 +240,6 @@ static void ib_ucm_event_rep_get(struct ib_ucm_rep_event_resp *urep, urep->srq = krep->srq; } -static void ib_ucm_event_sidr_req_get(struct ib_ucm_context *ctx, - struct ib_ucm_sidr_req_event_resp *ureq, - struct ib_cm_sidr_req_event_param *kreq) -{ - ureq->listen_id = ctx->id; - ureq->pkey = kreq->pkey; -} - static void ib_ucm_event_sidr_rep_get(struct ib_ucm_sidr_rep_event_resp *urep, struct ib_cm_sidr_rep_event_param *krep) { @@ -275,15 +248,14 @@ static void ib_ucm_event_sidr_rep_get(struct ib_ucm_sidr_rep_event_resp *urep, urep->qpn = krep->qpn; }; -static int ib_ucm_event_process(struct ib_ucm_context *ctx, - struct ib_cm_event *evt, +static int ib_ucm_event_process(struct ib_cm_event *evt, struct ib_ucm_event *uvt) { void *info = NULL; switch (evt->event) { case IB_CM_REQ_RECEIVED: - ib_ucm_event_req_get(ctx, &uvt->resp.u.req_resp, + ib_ucm_event_req_get(&uvt->resp.u.req_resp, &evt->param.req_rcvd); uvt->data_len = IB_CM_REQ_PRIVATE_DATA_SIZE; uvt->resp.present = IB_UCM_PRES_PRIMARY; @@ -319,8 +291,8 @@ static int ib_ucm_event_process(struct ib_ucm_context *ctx, info = evt->param.rej_rcvd.ari; break; case IB_CM_LAP_RECEIVED: - ib_ucm_event_path_get(&uvt->resp.u.lap_resp.path, - evt->param.lap_rcvd.alternate_path); + ib_copy_path_rec_to_user(&uvt->resp.u.lap_resp.path, + evt->param.lap_rcvd.alternate_path); uvt->data_len = IB_CM_LAP_PRIVATE_DATA_SIZE; uvt->resp.present = IB_UCM_PRES_ALTERNATE; break; @@ -331,8 +303,10 @@ static int ib_ucm_event_process(struct ib_ucm_context *ctx, info = evt->param.apr_rcvd.apr_info; break; case IB_CM_SIDR_REQ_RECEIVED: - ib_ucm_event_sidr_req_get(ctx, &uvt->resp.u.sidr_req_resp, - &evt->param.sidr_req_rcvd); + uvt->resp.u.sidr_req_resp.pkey = + evt->param.sidr_req_rcvd.pkey; + uvt->resp.u.sidr_req_resp.port = + evt->param.sidr_req_rcvd.port; uvt->data_len = IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE; break; case IB_CM_SIDR_REP_RECEIVED: @@ -348,20 +322,18 @@ static int ib_ucm_event_process(struct ib_ucm_context *ctx, } if (uvt->data_len) { - uvt->data = kmalloc(uvt->data_len, GFP_KERNEL); + uvt->data = kmemdup(evt->private_data, uvt->data_len, GFP_KERNEL); if (!uvt->data) goto err1; - memcpy(uvt->data, evt->private_data, uvt->data_len); uvt->resp.present |= IB_UCM_PRES_DATA; } if (uvt->info_len) { - uvt->info = kmalloc(uvt->info_len, GFP_KERNEL); + uvt->info = kmemdup(info, uvt->info_len, GFP_KERNEL); if (!uvt->info) goto err2; - memcpy(uvt->info, info, uvt->info_len); uvt->resp.present |= IB_UCM_PRES_INFO; } return 0; @@ -378,43 +350,35 @@ static int ib_ucm_event_handler(struct ib_cm_id *cm_id, struct ib_ucm_event *uevent; struct ib_ucm_context *ctx; int result = 0; - int id; ctx = cm_id->context; - if (event->event == IB_CM_REQ_RECEIVED || - event->event == IB_CM_SIDR_REQ_RECEIVED) - id = IB_UCM_CM_ID_INVALID; - else - id = ctx->id; - - uevent = kmalloc(sizeof(*uevent), GFP_KERNEL); + uevent = kzalloc(sizeof *uevent, GFP_KERNEL); if (!uevent) goto err1; - memset(uevent, 0, sizeof(*uevent)); - uevent->resp.id = id; + uevent->ctx = ctx; + uevent->cm_id = cm_id; + uevent->resp.uid = ctx->uid; + uevent->resp.id = ctx->id; uevent->resp.event = event->event; - result = ib_ucm_event_process(ctx, event, uevent); + result = ib_ucm_event_process(event, uevent); if (result) goto err2; - uevent->ctx = ctx; - uevent->cm_id = (id == IB_UCM_CM_ID_INVALID) ? cm_id : NULL; - - down(&ctx->file->mutex); + mutex_lock(&ctx->file->file_mutex); list_add_tail(&uevent->file_list, &ctx->file->events); list_add_tail(&uevent->ctx_list, &ctx->events); wake_up_interruptible(&ctx->file->poll_wait); - up(&ctx->file->mutex); + mutex_unlock(&ctx->file->file_mutex); return 0; err2: kfree(uevent); err1: /* Destroy new cm_id's */ - return (id == IB_UCM_CM_ID_INVALID); + return ib_ucm_new_cm_id(event->event); } static ssize_t ib_ucm_event(struct ib_ucm_file *file, @@ -423,61 +387,43 @@ static ssize_t ib_ucm_event(struct ib_ucm_file *file, { struct ib_ucm_context *ctx; struct ib_ucm_event_get cmd; - struct ib_ucm_event *uevent = NULL; + struct ib_ucm_event *uevent; int result = 0; - DEFINE_WAIT(wait); if (out_len < sizeof(struct ib_ucm_event_resp)) return -ENOSPC; if (copy_from_user(&cmd, inbuf, sizeof(cmd))) return -EFAULT; - /* - * wait - */ - down(&file->mutex); + mutex_lock(&file->file_mutex); while (list_empty(&file->events)) { + mutex_unlock(&file->file_mutex); - if (file->filp->f_flags & O_NONBLOCK) { - result = -EAGAIN; - break; - } - - if (signal_pending(current)) { - result = -ERESTARTSYS; - break; - } - - prepare_to_wait(&file->poll_wait, &wait, TASK_INTERRUPTIBLE); + if (file->filp->f_flags & O_NONBLOCK) + return -EAGAIN; - up(&file->mutex); - schedule(); - down(&file->mutex); + if (wait_event_interruptible(file->poll_wait, + !list_empty(&file->events))) + return -ERESTARTSYS; - finish_wait(&file->poll_wait, &wait); + mutex_lock(&file->file_mutex); } - if (result) - goto done; - uevent = list_entry(file->events.next, struct ib_ucm_event, file_list); - if (!uevent->cm_id) - goto user; + if (ib_ucm_new_cm_id(uevent->resp.event)) { + ctx = ib_ucm_ctx_alloc(file); + if (!ctx) { + result = -ENOMEM; + goto done; + } - ctx = ib_ucm_ctx_alloc(file); - if (!ctx) { - result = -ENOMEM; - goto done; + ctx->cm_id = uevent->cm_id; + ctx->cm_id->context = ctx; + uevent->resp.id = ctx->id; } - ctx->cm_id = uevent->cm_id; - ctx->cm_id->context = ctx; - - uevent->resp.id = ctx->id; - -user: if (copy_to_user((void __user *)(unsigned long)cmd.response, &uevent->resp, sizeof(uevent->resp))) { result = -EFAULT; @@ -485,12 +431,10 @@ user: } if (uevent->data) { - if (cmd.data_len < uevent->data_len) { result = -ENOMEM; goto done; } - if (copy_to_user((void __user *)(unsigned long)cmd.data, uevent->data, uevent->data_len)) { result = -EFAULT; @@ -499,12 +443,10 @@ user: } if (uevent->info) { - if (cmd.info_len < uevent->info_len) { result = -ENOMEM; goto done; } - if (copy_to_user((void __user *)(unsigned long)cmd.info, uevent->info, uevent->info_len)) { result = -EFAULT; @@ -514,16 +456,16 @@ user: list_del(&uevent->file_list); list_del(&uevent->ctx_list); + uevent->ctx->events_reported++; kfree(uevent->data); kfree(uevent->info); kfree(uevent); done: - up(&file->mutex); + mutex_unlock(&file->file_mutex); return result; } - static ssize_t ib_ucm_create_id(struct ib_ucm_file *file, const char __user *inbuf, int in_len, int out_len) @@ -539,29 +481,35 @@ static ssize_t ib_ucm_create_id(struct ib_ucm_file *file, if (copy_from_user(&cmd, inbuf, sizeof(cmd))) return -EFAULT; - down(&file->mutex); + mutex_lock(&file->file_mutex); ctx = ib_ucm_ctx_alloc(file); - up(&file->mutex); + mutex_unlock(&file->file_mutex); if (!ctx) return -ENOMEM; - ctx->cm_id = ib_create_cm_id(ib_ucm_event_handler, ctx); + ctx->uid = cmd.uid; + ctx->cm_id = ib_create_cm_id(file->device->ib_dev, + ib_ucm_event_handler, ctx); if (IS_ERR(ctx->cm_id)) { result = PTR_ERR(ctx->cm_id); - goto err; + goto err1; } resp.id = ctx->id; if (copy_to_user((void __user *)(unsigned long)cmd.response, &resp, sizeof(resp))) { result = -EFAULT; - goto err; + goto err2; } - return 0; -err: - ib_ucm_destroy_ctx(file, ctx->id); +err2: + ib_destroy_cm_id(ctx->cm_id); +err1: + mutex_lock(&ctx_id_mutex); + idr_remove(&ctx_id_table, ctx->id); + mutex_unlock(&ctx_id_mutex); + kfree(ctx); return result; } @@ -570,11 +518,44 @@ static ssize_t ib_ucm_destroy_id(struct ib_ucm_file *file, int in_len, int out_len) { struct ib_ucm_destroy_id cmd; + struct ib_ucm_destroy_id_resp resp; + struct ib_ucm_context *ctx; + int result = 0; + + if (out_len < sizeof(resp)) + return -ENOSPC; if (copy_from_user(&cmd, inbuf, sizeof(cmd))) return -EFAULT; - return ib_ucm_destroy_ctx(file, cmd.id); + mutex_lock(&ctx_id_mutex); + ctx = idr_find(&ctx_id_table, cmd.id); + if (!ctx) + ctx = ERR_PTR(-ENOENT); + else if (ctx->file != file) + ctx = ERR_PTR(-EINVAL); + else + idr_remove(&ctx_id_table, ctx->id); + mutex_unlock(&ctx_id_mutex); + + if (IS_ERR(ctx)) + return PTR_ERR(ctx); + + ib_ucm_ctx_put(ctx); + wait_for_completion(&ctx->comp); + + /* No new events will be generated after destroying the cm_id. */ + ib_destroy_cm_id(ctx->cm_id); + /* Cleanup events not yet reported to the user. */ + ib_ucm_cleanup_events(ctx); + + resp.events_reported = ctx->events_reported; + if (copy_to_user((void __user *)(unsigned long)cmd.response, + &resp, sizeof(resp))) + result = -EFAULT; + + kfree(ctx); + return result; } static ssize_t ib_ucm_attr_id(struct ib_ucm_file *file, @@ -609,6 +590,55 @@ static ssize_t ib_ucm_attr_id(struct ib_ucm_file *file, return result; } +static ssize_t ib_ucm_init_qp_attr(struct ib_ucm_file *file, + const char __user *inbuf, + int in_len, int out_len) +{ + struct ib_uverbs_qp_attr resp; + struct ib_ucm_init_qp_attr cmd; + struct ib_ucm_context *ctx; + struct ib_qp_attr qp_attr; + int result = 0; + + if (out_len < sizeof(resp)) + return -ENOSPC; + + if (copy_from_user(&cmd, inbuf, sizeof(cmd))) + return -EFAULT; + + ctx = ib_ucm_ctx_get(file, cmd.id); + if (IS_ERR(ctx)) + return PTR_ERR(ctx); + + resp.qp_attr_mask = 0; + memset(&qp_attr, 0, sizeof qp_attr); + qp_attr.qp_state = cmd.qp_state; + result = ib_cm_init_qp_attr(ctx->cm_id, &qp_attr, &resp.qp_attr_mask); + if (result) + goto out; + + ib_copy_qp_attr_to_user(&resp, &qp_attr); + + if (copy_to_user((void __user *)(unsigned long)cmd.response, + &resp, sizeof(resp))) + result = -EFAULT; + +out: + ib_ucm_ctx_put(ctx); + return result; +} + +static int ucm_validate_listen(__be64 service_id, __be64 service_mask) +{ + service_id &= service_mask; + + if (((service_id & IB_CMA_SERVICE_ID_MASK) == IB_CMA_SERVICE_ID) || + ((service_id & IB_SDP_SERVICE_ID_MASK) == IB_SDP_SERVICE_ID)) + return -EINVAL; + + return 0; +} + static ssize_t ib_ucm_listen(struct ib_ucm_file *file, const char __user *inbuf, int in_len, int out_len) @@ -624,16 +654,22 @@ static ssize_t ib_ucm_listen(struct ib_ucm_file *file, if (IS_ERR(ctx)) return PTR_ERR(ctx); - result = ib_cm_listen(ctx->cm_id, cmd.service_id, cmd.service_mask); + result = ucm_validate_listen(cmd.service_id, cmd.service_mask); + if (result) + goto out; + + result = ib_cm_listen(ctx->cm_id, cmd.service_id, cmd.service_mask, + NULL); +out: ib_ucm_ctx_put(ctx); return result; } -static ssize_t ib_ucm_establish(struct ib_ucm_file *file, - const char __user *inbuf, - int in_len, int out_len) +static ssize_t ib_ucm_notify(struct ib_ucm_file *file, + const char __user *inbuf, + int in_len, int out_len) { - struct ib_ucm_establish cmd; + struct ib_ucm_notify cmd; struct ib_ucm_context *ctx; int result; @@ -644,7 +680,7 @@ static ssize_t ib_ucm_establish(struct ib_ucm_file *file, if (IS_ERR(ctx)) return PTR_ERR(ctx); - result = ib_cm_establish(ctx->cm_id); + result = ib_cm_notify(ctx->cm_id, (enum ib_event_type) cmd.event); ib_ucm_ctx_put(ctx); return result; } @@ -658,14 +694,9 @@ static int ib_ucm_alloc_data(const void **dest, u64 src, u32 len) if (!len) return 0; - data = kmalloc(len, GFP_KERNEL); - if (!data) - return -ENOMEM; - - if (copy_from_user(data, (void __user *)(unsigned long)src, len)) { - kfree(data); - return -EFAULT; - } + data = memdup_user((void __user *)(unsigned long)src, len); + if (IS_ERR(data)) + return PTR_ERR(data); *dest = data; return 0; @@ -673,7 +704,7 @@ static int ib_ucm_alloc_data(const void **dest, u64 src, u32 len) static int ib_ucm_path_get(struct ib_sa_path_rec **path, u64 src) { - struct ib_ucm_path_rec ucm_path; + struct ib_user_path_rec upath; struct ib_sa_path_rec *sa_path; *path = NULL; @@ -685,36 +716,14 @@ static int ib_ucm_path_get(struct ib_sa_path_rec **path, u64 src) if (!sa_path) return -ENOMEM; - if (copy_from_user(&ucm_path, (void __user *)(unsigned long)src, - sizeof(ucm_path))) { + if (copy_from_user(&upath, (void __user *)(unsigned long)src, + sizeof(upath))) { kfree(sa_path); return -EFAULT; } - memcpy(sa_path->dgid.raw, ucm_path.dgid, sizeof sa_path->dgid); - memcpy(sa_path->sgid.raw, ucm_path.sgid, sizeof sa_path->sgid); - - sa_path->dlid = ucm_path.dlid; - sa_path->slid = ucm_path.slid; - sa_path->raw_traffic = ucm_path.raw_traffic; - sa_path->flow_label = ucm_path.flow_label; - sa_path->hop_limit = ucm_path.hop_limit; - sa_path->traffic_class = ucm_path.traffic_class; - sa_path->reversible = ucm_path.reversible; - sa_path->numb_path = ucm_path.numb_path; - sa_path->pkey = ucm_path.pkey; - sa_path->sl = ucm_path.sl; - sa_path->mtu_selector = ucm_path.mtu_selector; - sa_path->mtu = ucm_path.mtu; - sa_path->rate_selector = ucm_path.rate_selector; - sa_path->rate = ucm_path.rate; - sa_path->packet_life_time = ucm_path.packet_life_time; - sa_path->preference = ucm_path.preference; - - sa_path->packet_life_time_selector = - ucm_path.packet_life_time_selector; - + ib_copy_path_rec_from_user(sa_path, &upath); *path = sa_path; return 0; } @@ -800,7 +809,6 @@ static ssize_t ib_ucm_send_rep(struct ib_ucm_file *file, param.private_data_len = cmd.len; param.responder_resources = cmd.responder_resources; param.initiator_depth = cmd.initiator_depth; - param.target_ack_delay = cmd.target_ack_delay; param.failover_accepted = cmd.failover_accepted; param.flow_control = cmd.flow_control; param.rnr_retry_count = cmd.rnr_retry_count; @@ -808,6 +816,7 @@ static ssize_t ib_ucm_send_rep(struct ib_ucm_file *file, ctx = ib_ucm_ctx_get(file, cmd.id); if (!IS_ERR(ctx)) { + ctx->uid = cmd.uid; result = ib_send_cm_rep(ctx->cm_id, ¶m); ib_ucm_ctx_put(ctx); } else @@ -1009,7 +1018,6 @@ static ssize_t ib_ucm_send_sidr_req(struct ib_ucm_file *file, param.service_id = cmd.sid; param.timeout_ms = cmd.timeout; param.max_cm_retries = cmd.max_cm_retries; - param.pkey = cmd.pkey; ctx = ib_ucm_ctx_get(file, cmd.id); if (!IS_ERR(ctx)) { @@ -1073,7 +1081,7 @@ static ssize_t (*ucm_cmd_table[])(struct ib_ucm_file *file, [IB_USER_CM_CMD_DESTROY_ID] = ib_ucm_destroy_id, [IB_USER_CM_CMD_ATTR_ID] = ib_ucm_attr_id, [IB_USER_CM_CMD_LISTEN] = ib_ucm_listen, - [IB_USER_CM_CMD_ESTABLISH] = ib_ucm_establish, + [IB_USER_CM_CMD_NOTIFY] = ib_ucm_notify, [IB_USER_CM_CMD_SEND_REQ] = ib_ucm_send_req, [IB_USER_CM_CMD_SEND_REP] = ib_ucm_send_rep, [IB_USER_CM_CMD_SEND_RTU] = ib_ucm_send_rtu, @@ -1086,6 +1094,7 @@ static ssize_t (*ucm_cmd_table[])(struct ib_ucm_file *file, [IB_USER_CM_CMD_SEND_SIDR_REQ] = ib_ucm_send_sidr_req, [IB_USER_CM_CMD_SEND_SIDR_REP] = ib_ucm_send_sidr_rep, [IB_USER_CM_CMD_EVENT] = ib_ucm_event, + [IB_USER_CM_CMD_INIT_QP_ATTR] = ib_ucm_init_qp_attr, }; static ssize_t ib_ucm_write(struct file *filp, const char __user *buf, @@ -1101,10 +1110,7 @@ static ssize_t ib_ucm_write(struct file *filp, const char __user *buf, if (copy_from_user(&hdr, buf, sizeof(hdr))) return -EFAULT; - ucm_dbg("Write. cmd <%d> in <%d> out <%d> len <%Zu>\n", - hdr.cmd, hdr.in, hdr.out, len); - - if (hdr.cmd < 0 || hdr.cmd >= ARRAY_SIZE(ucm_cmd_table)) + if (hdr.cmd >= ARRAY_SIZE(ucm_cmd_table)) return -EINVAL; if (hdr.in + sizeof(hdr) > len) @@ -1132,6 +1138,14 @@ static unsigned int ib_ucm_poll(struct file *filp, return mask; } +/* + * ib_ucm_open() does not need the BKL: + * + * - no global state is referred to; + * - there is no ioctl method to race against; + * - no further module initialization is required for open to work + * after the device is registered. + */ static int ib_ucm_open(struct inode *inode, struct file *filp) { struct ib_ucm_file *file; @@ -1144,14 +1158,13 @@ static int ib_ucm_open(struct inode *inode, struct file *filp) INIT_LIST_HEAD(&file->ctxs); init_waitqueue_head(&file->poll_wait); - init_MUTEX(&file->mutex); + mutex_init(&file->file_mutex); filp->private_data = file; file->filp = filp; + file->device = container_of(inode->i_cdev, struct ib_ucm_device, cdev); - ucm_dbg("Created struct\n"); - - return 0; + return nonseekable_open(inode, filp); } static int ib_ucm_close(struct inode *inode, struct file *filp) @@ -1159,78 +1172,198 @@ static int ib_ucm_close(struct inode *inode, struct file *filp) struct ib_ucm_file *file = filp->private_data; struct ib_ucm_context *ctx; - down(&file->mutex); + mutex_lock(&file->file_mutex); while (!list_empty(&file->ctxs)) { - ctx = list_entry(file->ctxs.next, struct ib_ucm_context, file_list); + mutex_unlock(&file->file_mutex); + + mutex_lock(&ctx_id_mutex); + idr_remove(&ctx_id_table, ctx->id); + mutex_unlock(&ctx_id_mutex); + + ib_destroy_cm_id(ctx->cm_id); + ib_ucm_cleanup_events(ctx); + kfree(ctx); - up(&file->mutex); - ib_ucm_destroy_ctx(file, ctx->id); - down(&file->mutex); + mutex_lock(&file->file_mutex); } - up(&file->mutex); + mutex_unlock(&file->file_mutex); kfree(file); return 0; } -static struct file_operations ib_ucm_fops = { - .owner = THIS_MODULE, - .open = ib_ucm_open, +static void ib_ucm_release_dev(struct device *dev) +{ + struct ib_ucm_device *ucm_dev; + + ucm_dev = container_of(dev, struct ib_ucm_device, dev); + cdev_del(&ucm_dev->cdev); + if (ucm_dev->devnum < IB_UCM_MAX_DEVICES) + clear_bit(ucm_dev->devnum, dev_map); + else + clear_bit(ucm_dev->devnum - IB_UCM_MAX_DEVICES, dev_map); + kfree(ucm_dev); +} + +static const struct file_operations ucm_fops = { + .owner = THIS_MODULE, + .open = ib_ucm_open, .release = ib_ucm_close, - .write = ib_ucm_write, + .write = ib_ucm_write, .poll = ib_ucm_poll, + .llseek = no_llseek, }; +static ssize_t show_ibdev(struct device *dev, struct device_attribute *attr, + char *buf) +{ + struct ib_ucm_device *ucm_dev; -static struct class *ib_ucm_class; -static struct cdev ib_ucm_cdev; + ucm_dev = container_of(dev, struct ib_ucm_device, dev); + return sprintf(buf, "%s\n", ucm_dev->ib_dev->name); +} +static DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL); -static int __init ib_ucm_init(void) +static dev_t overflow_maj; +static DECLARE_BITMAP(overflow_map, IB_UCM_MAX_DEVICES); +static int find_overflow_devnum(void) { - int result; + int ret; + + if (!overflow_maj) { + ret = alloc_chrdev_region(&overflow_maj, 0, IB_UCM_MAX_DEVICES, + "infiniband_cm"); + if (ret) { + printk(KERN_ERR "ucm: couldn't register dynamic device number\n"); + return ret; + } + } + + ret = find_first_zero_bit(overflow_map, IB_UCM_MAX_DEVICES); + if (ret >= IB_UCM_MAX_DEVICES) + return -1; + + return ret; +} + +static void ib_ucm_add_one(struct ib_device *device) +{ + int devnum; + dev_t base; + struct ib_ucm_device *ucm_dev; + + if (!device->alloc_ucontext || + rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB) + return; - result = register_chrdev_region(IB_UCM_DEV, 1, "infiniband_cm"); - if (result) { - ucm_dbg("Error <%d> registering dev\n", result); - goto err_chr; + ucm_dev = kzalloc(sizeof *ucm_dev, GFP_KERNEL); + if (!ucm_dev) + return; + + ucm_dev->ib_dev = device; + + devnum = find_first_zero_bit(dev_map, IB_UCM_MAX_DEVICES); + if (devnum >= IB_UCM_MAX_DEVICES) { + devnum = find_overflow_devnum(); + if (devnum < 0) + goto err; + + ucm_dev->devnum = devnum + IB_UCM_MAX_DEVICES; + base = devnum + overflow_maj; + set_bit(devnum, overflow_map); + } else { + ucm_dev->devnum = devnum; + base = devnum + IB_UCM_BASE_DEV; + set_bit(devnum, dev_map); } - cdev_init(&ib_ucm_cdev, &ib_ucm_fops); + cdev_init(&ucm_dev->cdev, &ucm_fops); + ucm_dev->cdev.owner = THIS_MODULE; + kobject_set_name(&ucm_dev->cdev.kobj, "ucm%d", ucm_dev->devnum); + if (cdev_add(&ucm_dev->cdev, base, 1)) + goto err; - result = cdev_add(&ib_ucm_cdev, IB_UCM_DEV, 1); - if (result) { - ucm_dbg("Error <%d> adding cdev\n", result); + ucm_dev->dev.class = &cm_class; + ucm_dev->dev.parent = device->dma_device; + ucm_dev->dev.devt = ucm_dev->cdev.dev; + ucm_dev->dev.release = ib_ucm_release_dev; + dev_set_name(&ucm_dev->dev, "ucm%d", ucm_dev->devnum); + if (device_register(&ucm_dev->dev)) goto err_cdev; - } - ib_ucm_class = class_create(THIS_MODULE, "infiniband_cm"); - if (IS_ERR(ib_ucm_class)) { - result = PTR_ERR(ib_ucm_class); - ucm_dbg("Error <%d> creating class\n", result); - goto err_class; - } + if (device_create_file(&ucm_dev->dev, &dev_attr_ibdev)) + goto err_dev; + + ib_set_client_data(device, &ucm_client, ucm_dev); + return; + +err_dev: + device_unregister(&ucm_dev->dev); +err_cdev: + cdev_del(&ucm_dev->cdev); + if (ucm_dev->devnum < IB_UCM_MAX_DEVICES) + clear_bit(devnum, dev_map); + else + clear_bit(devnum, overflow_map); +err: + kfree(ucm_dev); + return; +} - class_device_create(ib_ucm_class, IB_UCM_DEV, NULL, "ucm"); +static void ib_ucm_remove_one(struct ib_device *device) +{ + struct ib_ucm_device *ucm_dev = ib_get_client_data(device, &ucm_client); - idr_init(&ctx_id_table); - init_MUTEX(&ctx_id_mutex); + if (!ucm_dev) + return; + device_unregister(&ucm_dev->dev); +} + +static CLASS_ATTR_STRING(abi_version, S_IRUGO, + __stringify(IB_USER_CM_ABI_VERSION)); + +static int __init ib_ucm_init(void) +{ + int ret; + + ret = register_chrdev_region(IB_UCM_BASE_DEV, IB_UCM_MAX_DEVICES, + "infiniband_cm"); + if (ret) { + printk(KERN_ERR "ucm: couldn't register device number\n"); + goto error1; + } + + ret = class_create_file(&cm_class, &class_attr_abi_version.attr); + if (ret) { + printk(KERN_ERR "ucm: couldn't create abi_version attribute\n"); + goto error2; + } + + ret = ib_register_client(&ucm_client); + if (ret) { + printk(KERN_ERR "ucm: couldn't register client\n"); + goto error3; + } return 0; -err_class: - cdev_del(&ib_ucm_cdev); -err_cdev: - unregister_chrdev_region(IB_UCM_DEV, 1); -err_chr: - return result; + +error3: + class_remove_file(&cm_class, &class_attr_abi_version.attr); +error2: + unregister_chrdev_region(IB_UCM_BASE_DEV, IB_UCM_MAX_DEVICES); +error1: + return ret; } static void __exit ib_ucm_cleanup(void) { - class_device_destroy(ib_ucm_class, IB_UCM_DEV); - class_destroy(ib_ucm_class); - cdev_del(&ib_ucm_cdev); - unregister_chrdev_region(IB_UCM_DEV, 1); + ib_unregister_client(&ucm_client); + class_remove_file(&cm_class, &class_attr_abi_version.attr); + unregister_chrdev_region(IB_UCM_BASE_DEV, IB_UCM_MAX_DEVICES); + if (overflow_maj) + unregister_chrdev_region(overflow_maj, IB_UCM_MAX_DEVICES); + idr_destroy(&ctx_id_table); } module_init(ib_ucm_init); |
