Viewing: lib-md.c
// SPDX-License-Identifier: GPL-2.0
/* Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
* Copyright (c) 2012, 2017, Intel Corporation.
*/
/* This file is part of Lustre, http://www.lustre.org/
*
* Memory Descriptor management routines
*/
#define DEBUG_SUBSYSTEM S_LNET
#include <linux/lnet/lib-lnet.h>
/* must be called with lnet_res_lock held */
void
lnet_md_unlink(struct lnet_libmd *md)
{
if ((md->md_flags & LNET_MD_FLAG_ZOMBIE) == 0) {
/* first unlink attempt... */
struct lnet_me *me = md->md_me;
md->md_flags |= LNET_MD_FLAG_ZOMBIE;
/* Disassociate from ME (if any), and unlink it if it was created
* with LNET_UNLINK */
if (me != NULL) {
/* detach MD from portal */
lnet_ptl_detach_md(me, md);
if (me->me_unlink == LNET_UNLINK)
lnet_me_unlink(me);
}
/* ensure all future handle lookups fail */
lnet_res_lh_invalidate(&md->md_lh);
}
if (md->md_refcount != 0) {
CDEBUG(D_NET, "Queueing unlink of md %p\n", md);
return;
}
CDEBUG(D_NET, "Unlinking md %p\n", md);
LASSERT(!list_empty(&md->md_list));
list_del_init(&md->md_list);
LASSERT(!(md->md_flags & LNET_MD_FLAG_HANDLING));
lnet_md_free(md);
}
struct page *
lnet_get_first_page(struct lnet_libmd *md, unsigned int offset)
{
unsigned int niov;
struct bio_vec *kiov;
/*
* if the md_options has a bulk handle then we want to look at the
* bulk md because that's the data which we will be DMAing
*/
if (md && (md->md_options & LNET_MD_BULK_HANDLE) != 0 &&
!LNetMDHandleIsInvalid(md->md_bulk_handle))
md = lnet_handle2md(&md->md_bulk_handle);
if (!md || md->md_niov == 0)
return NULL;
kiov = md->md_kiov;
niov = md->md_niov;
while (offset >= kiov->bv_len) {
offset -= kiov->bv_len;
niov--;
kiov++;
if (niov == 0) {
CERROR("offset %d goes beyond kiov\n", offset);
return NULL;
}
}
return kiov->bv_page;
}
int
lnet_cpt_of_md(struct lnet_libmd *md, unsigned int offset)
{
struct page *page;
int cpt = CFS_CPT_ANY;
page = lnet_get_first_page(md, offset);
if (!page) {
CDEBUG(D_NET, "Couldn't resolve first page of md %p with offset %u\n",
md, offset);
goto out;
}
cpt = cfs_cpt_of_node(lnet_cpt_table(), page_to_nid(page));
out:
return cpt;
}
static int lnet_md_validate(const struct lnet_md *umd);
static struct lnet_libmd *
lnet_md_build(const struct lnet_md *umd, int unlink)
{
int i;
unsigned int niov;
int total_length = 0;
struct lnet_libmd *lmd;
unsigned int size;
if (lnet_md_validate(umd) != 0)
return ERR_PTR(-EINVAL);
if (umd->umd_options & LNET_MD_KIOV)
niov = umd->umd_length;
else
niov = DIV_ROUND_UP(
offset_in_page(umd->umd_start) + umd->umd_length,
PAGE_SIZE);
size = offsetof(struct lnet_libmd, md_kiov[niov]);
if (size <= LNET_SMALL_MD_SIZE) {
lmd = kmem_cache_zalloc(lnet_small_mds_cachep, GFP_NOFS);
if (lmd) {
LIBCFS_MEM_MSG(lmd, size, "slab-alloced");
} else {
CDEBUG(D_MALLOC, "failed to allocate 'md' of size %u\n",
size);
}
} else {
LIBCFS_ALLOC(lmd, size);
}
if (!lmd)
return ERR_PTR(-ENOMEM);
lmd->md_niov = niov;
INIT_LIST_HEAD(&lmd->md_list);
lmd->md_me = NULL;
lmd->md_start = umd->umd_start;
lmd->md_offset = 0;
lmd->md_max_size = umd->umd_max_size;
lmd->md_options = umd->umd_options;
lmd->md_user_ptr = umd->umd_user_ptr;
lmd->md_handler = NULL;
lmd->md_threshold = umd->umd_threshold;
lmd->md_refcount = 0;
lmd->md_flags = (unlink == LNET_UNLINK) ? LNET_MD_FLAG_AUTO_UNLINK : 0;
lmd->md_bulk_handle = umd->umd_bulk_handle;
if (umd->umd_options & LNET_MD_GPU_ADDR)
lmd->md_flags |= LNET_MD_FLAG_GPU;
if (umd->umd_options & LNET_MD_KIOV) {
memcpy(lmd->md_kiov, umd->umd_start,
niov * sizeof(lmd->md_kiov[0]));
for (i = 0; i < (int)niov; i++) {
/* We take the page pointer on trust */
if (lmd->md_kiov[i].bv_offset +
lmd->md_kiov[i].bv_len > PAGE_SIZE) {
lnet_md_free(lmd);
return ERR_PTR(-EINVAL); /* invalid length */
}
total_length += lmd->md_kiov[i].bv_len;
}
lmd->md_length = total_length;
if ((umd->umd_options & LNET_MD_MAX_SIZE) && /* max size used */
(umd->umd_max_size < 0 ||
umd->umd_max_size > total_length)) { /* illegal max_size */
lnet_md_free(lmd);
return ERR_PTR(-EINVAL);
}
} else { /* contiguous - split into pages */
void *pa = umd->umd_start;
int len = umd->umd_length;
lmd->md_length = len;
i = 0;
while (len) {
struct page *p;
int plen;
if (is_vmalloc_addr(pa))
p = vmalloc_to_page(pa);
else
p = virt_to_page(pa);
plen = min_t(int, len, PAGE_SIZE - offset_in_page(pa));
lmd->md_kiov[i].bv_page = p;
lmd->md_kiov[i].bv_offset = offset_in_page(pa);
lmd->md_kiov[i].bv_len = plen;
len -= plen;
pa += plen;
i += 1;
}
WARN(!(lmd->md_options & LNET_MD_GNILND) && i > LNET_MAX_IOV,
"Max IOV exceeded: %d should be < %d\n",
i, LNET_MAX_IOV);
if ((umd->umd_options & LNET_MD_MAX_SIZE) && /* max size used */
(umd->umd_max_size < 0 ||
umd->umd_max_size > (int)umd->umd_length)) {
lnet_md_free(lmd);
return ERR_PTR(-EINVAL);
}
lmd->md_options |= LNET_MD_KIOV;
}
return lmd;
}
/* must be called with resource lock held */
static void
lnet_md_link(struct lnet_libmd *md, lnet_handler_t handler, int cpt)
{
struct lnet_res_container *container = the_lnet.ln_md_containers[cpt];
/* NB we are passed an allocated, but inactive md.
* Caller may lnet_md_unlink() it, or may lnet_md_free() it.
*/
/* This implementation doesn't know how to create START events or
* disable END events. Best to LASSERT our caller is compliant so
* we find out quickly... */
/* TODO - reevaluate what should be here in light of
* the removal of the start and end events
* maybe there we shouldn't even allow LNET_EQ_NONE!)
* LASSERT (handler != NULL);
*/
md->md_handler = handler;
lnet_res_lh_initialize(container, &md->md_lh);
LASSERT(list_empty(&md->md_list));
list_add(&md->md_list, &container->rec_active);
}
void lnet_assert_handler_unused(lnet_handler_t handler)
{
struct lnet_res_container *container;
int cpt;
if (!handler)
return;
cfs_percpt_for_each(container, cpt, the_lnet.ln_md_containers) {
struct lnet_libmd *md;
lnet_res_lock(cpt);
list_for_each_entry(md, &container->rec_active, md_list) {
LASSERT(md->md_handler != handler);
}
lnet_res_unlock(cpt);
}
}
EXPORT_SYMBOL(lnet_assert_handler_unused);
/* must be called with lnet_res_lock held */
void
lnet_md_deconstruct(struct lnet_libmd *lmd, struct lnet_event *ev)
{
ev->md_start = lmd->md_start;
ev->md_options = lmd->md_options;
ev->md_user_ptr = lmd->md_user_ptr;
}
static int
lnet_md_validate(const struct lnet_md *umd)
{
if (umd->umd_start == NULL && umd->umd_length != 0) {
CERROR("MD start pointer can not be NULL with length %u\n",
umd->umd_length);
return -EINVAL;
}
if ((umd->umd_options & LNET_MD_KIOV) &&
umd->umd_length > LNET_MAX_IOV) {
CERROR("Invalid option: too many fragments %u, %d max\n",
umd->umd_length, LNET_MAX_IOV);
return -EINVAL;
}
return 0;
}
/**
* LNetMDAttach() - Create a memory descriptor and attach it to a ME
* @me: An ME to associate the new MD with.
* @umd: Provides initial values for the user-visible parts of a MD.
* Other than its use for initialization, there is no linkage between this
* structure and the MD maintained by the LNet.
* @unlink: A flag to indicate whether the MD is automatically unlinked
* when it becomes inactive, either because the operation threshold
* drops to zero or because the available memory becomes less than
* @umd.umd_max_size. (Note that the check for unlinking a MD only
* occurs after the completion of a successful operation on the MD.)
* The value LNET_UNLINK enables auto unlinking; the value LNET_RETAIN
* disables it.
* @handle: On successful returns, a handle to the newly created MD is saved
* here. This handle can be used later in LNetMDUnlink(). [out]
*
* The ME will either be linked to the new MD, or it will be freed.
*
* Return:
* * %0 On success.
* * %-EINVAL If @umd is not valid.
* * %-ENOMEM If new MD cannot be allocated.
*/
int
LNetMDAttach(struct lnet_me *me, const struct lnet_md *umd,
enum lnet_unlink unlink, struct lnet_handle_md *handle)
{
LIST_HEAD(matches);
LIST_HEAD(drops);
struct lnet_libmd *md;
int cpt;
LASSERT(the_lnet.ln_refcount > 0);
LASSERT(!me->me_md);
if ((umd->umd_options & (LNET_MD_OP_GET | LNET_MD_OP_PUT)) == 0) {
CERROR("Invalid option: no MD_OP set\n");
md = ERR_PTR(-EINVAL);
} else
md = lnet_md_build(umd, unlink);
cpt = me->me_cpt;
lnet_res_lock(cpt);
if (IS_ERR(md)) {
lnet_me_unlink(me);
lnet_res_unlock(cpt);
return PTR_ERR(md);
}
lnet_md_link(md, umd->umd_handler, cpt);
/* attach this MD to portal of ME and check if it matches any
* blocked msgs on this portal */
lnet_ptl_attach_md(me, md, &matches, &drops);
lnet_md2handle(handle, md);
lnet_res_unlock(cpt);
lnet_drop_delayed_msg_list(&drops, "Bad match");
lnet_recv_delayed_msg_list(&matches);
return 0;
}
EXPORT_SYMBOL(LNetMDAttach);
/**
* LNetMDBind() - Create a "free floating" memory descriptor (RDMA)
* @umd: memory descriptor
* @unlink: memory descripto for unlink
* @handle: On successful returns, a handle to the newly created MD is saved
* here. This handle can be used later in LNetMDUnlink(), LNetPut(),
* and LNetGet() operations [out]
*
* Create a "free floating" memory descriptor - a MD that is not associated
* with a ME. Such MDs are usually used in LNetPut() and LNetGet() operations.
*
* @umd, @unlink See the discussion for LNetMDAttach().
*
* Return:
* * %0 On success.
* * %-EINVAL If @umd is not valid.
* * %-ENOMEM If new MD cannot be allocated.
*/
int
LNetMDBind(const struct lnet_md *umd, enum lnet_unlink unlink,
struct lnet_handle_md *handle)
{
struct lnet_libmd *md;
int cpt;
int rc;
LASSERT(the_lnet.ln_refcount > 0);
if ((umd->umd_options & (LNET_MD_OP_GET | LNET_MD_OP_PUT)) != 0) {
CERROR("Invalid option: GET|PUT illegal on active MDs\n");
return -EINVAL;
}
md = lnet_md_build(umd, unlink);
if (IS_ERR(md))
return PTR_ERR(md);
if (md->md_length > LNET_MTU) {
CERROR("Invalid length: too big transfer size %u, %d max\n",
md->md_length, LNET_MTU);
rc = -EINVAL;
goto out_free;
}
cpt = lnet_res_lock_current();
lnet_md_link(md, umd->umd_handler, cpt);
lnet_md2handle(handle, md);
lnet_res_unlock(cpt);
return 0;
out_free:
lnet_md_free(md);
return rc;
}
EXPORT_SYMBOL(LNetMDBind);
/**
* LNetMDUnlink() - Unlink the memory descriptor from any ME
* @mdh: A handle for the MD to be unlinked.
*
* Unlink the memory descriptor from any ME it may be linked to and release
* the internal resources associated with it. As a result, active messages
* associated with the MD may get aborted.
*
* This function does not free the memory region associated with the MD;
* i.e., the memory the user allocated for this MD. If the ME associated with
* this MD is not NULL and was created with auto unlink enabled, the ME is
* unlinked as well (see LNetMEAttach()).
*
* Explicitly unlinking a MD via this function call has the same behavior as
* a MD that has been automatically unlinked, except that no LNET_EVENT_UNLINK
* is generated in the latter case.
*
* An unlinked event can be reported in two ways:
* - If there's no pending operations on the MD, it's unlinked immediately
* and an LNET_EVENT_UNLINK event is logged before this function returns.
* - Otherwise, the MD is only marked for deletion when this function
* returns, and the unlinked event will be piggybacked on the event of
* the completion of the last operation by setting the unlinked field of
* the event. No dedicated LNET_EVENT_UNLINK event is generated.
*
* Note that in both cases the unlinked field of the event is always set; no
* more event will happen on the MD after such an event is logged.
*
* Return:
* * %0 On success.
* * %-ENOENT If @mdh does not point to a valid MD object.
*/
int
LNetMDUnlink(struct lnet_handle_md mdh)
{
struct lnet_event ev;
struct lnet_libmd *md = NULL;
lnet_handler_t handler = NULL;
int cpt;
LASSERT(the_lnet.ln_refcount > 0);
cpt = lnet_cpt_of_cookie(mdh.cookie);
lnet_res_lock(cpt);
while (!md) {
md = lnet_handle2md(&mdh);
if (!md) {
lnet_res_unlock(cpt);
return -ENOENT;
}
if (md->md_refcount == 0 &&
md->md_flags & LNET_MD_FLAG_HANDLING) {
/* Race with unlocked call to ->md_handler. */
lnet_md_wait_handling(md, cpt);
md = NULL;
}
}
md->md_flags |= LNET_MD_FLAG_ABORTED;
/* If the MD is busy, lnet_md_unlink just marks it for deletion, and
* when the LND is done, the completion event flags that the MD was
* unlinked. Otherwise, we enqueue an event now... */
if (md->md_handler && md->md_refcount == 0) {
lnet_build_unlink_event(md, &ev);
handler = md->md_handler;
}
if (md->md_rspt_ptr != NULL)
lnet_detach_rsp_tracker(md, cpt);
lnet_md_unlink(md);
lnet_res_unlock(cpt);
if (handler)
handler(&ev);
return 0;
}
EXPORT_SYMBOL(LNetMDUnlink);
