On Apr 24, 2014, at 10:50 AM, Jeff Layton <jlayton@xxxxxxxxxxxxxxx> wrote:
> On Tue, 22 Apr 2014 17:29:13 -0400
> Weston Andros Adamson <dros@xxxxxxxxxxxxxxx> wrote:
>
>> Add "page groups" - a circular list of nfs requests (struct nfs_page)
>> that all reference the same page. This gives nfs read and write paths
>> the ability to account for sub-page regions independently. This
>> somewhat follows the design of struct buffer_head's sub-page
>> accounting.
>>
>> Only "head" requests are ever added/removed from the inode list in
>> the buffered write path. "head" and "sub" requests are treated the
>> same through the read path and the rest of the write/commit path.
>> Requests are given an extra reference across the life of the list.
>>
>> Page groups are never rejoined after being split. If the read/write
>> request fails and the client falls back to another path (ie revert
>> to MDS in PNFS case), the already split requests are pushed through
>> the recoalescing code again, which may split them further and then
>> coalesce them into properly sized requests on the wire. Fragmentation
>> shouldn't be a problem with the current design, because we flush all
>> requests in page group when a non-contiguous request is added, so
>> the only time resplitting should occur is on a resend of a read or
>> write.
>>
>> This patch lays the groundwork for sub-page splitting, but does not
>> actually do any splitting. For now all page groups have one request
>> as pg_test functions don't yet split pages. There are several related
>> patches that are needed support multiple requests per page group.
>>
>> Signed-off-by: Weston Andros Adamson <dros@xxxxxxxxxxxxxxx>
>> ---
>> fs/nfs/direct.c | 7 +-
>> fs/nfs/pagelist.c | 218 ++++++++++++++++++++++++++++++++++++++++++++---
>> fs/nfs/read.c | 4 +-
>> fs/nfs/write.c | 12 ++-
>> include/linux/nfs_page.h | 12 ++-
>> 5 files changed, 231 insertions(+), 22 deletions(-)
>>
>> diff --git a/fs/nfs/direct.c b/fs/nfs/direct.c
>> index a0c30c5..9d968ca 100644
>> --- a/fs/nfs/direct.c
>> +++ b/fs/nfs/direct.c
>> @@ -380,7 +380,7 @@ static ssize_t nfs_direct_read_schedule_segment(struct nfs_pageio_descriptor *de
>> struct nfs_page *req;
>> unsigned int req_len = min_t(size_t, bytes, PAGE_SIZE - pgbase);
>> /* XXX do we need to do the eof zeroing found in async_filler? */
>> - req = nfs_create_request(dreq->ctx, pagevec[i],
>> + req = nfs_create_request(dreq->ctx, pagevec[i], NULL,
>> pgbase, req_len);
>> if (IS_ERR(req)) {
>> result = PTR_ERR(req);
>> @@ -749,7 +749,7 @@ static ssize_t nfs_direct_write_schedule_segment(struct nfs_pageio_descriptor *d
>> struct nfs_page *req;
>> unsigned int req_len = min_t(size_t, bytes, PAGE_SIZE - pgbase);
>>
>> - req = nfs_create_request(dreq->ctx, pagevec[i],
>> + req = nfs_create_request(dreq->ctx, pagevec[i], NULL,
>> pgbase, req_len);
>> if (IS_ERR(req)) {
>> result = PTR_ERR(req);
>> @@ -827,6 +827,8 @@ static void nfs_direct_write_completion(struct nfs_pgio_header *hdr)
>> spin_unlock(&dreq->lock);
>>
>> while (!list_empty(&hdr->pages)) {
>> + bool do_destroy = true;
>> +
>> req = nfs_list_entry(hdr->pages.next);
>> nfs_list_remove_request(req);
>> switch (bit) {
>> @@ -834,6 +836,7 @@ static void nfs_direct_write_completion(struct nfs_pgio_header *hdr)
>> case NFS_IOHDR_NEED_COMMIT:
>> kref_get(&req->wb_kref);
>> nfs_mark_request_commit(req, hdr->lseg, &cinfo);
>> + do_destroy = false;
>> }
>> nfs_unlock_and_release_request(req);
>> }
>> diff --git a/fs/nfs/pagelist.c b/fs/nfs/pagelist.c
>> index ac4fb64..8cb8e14 100644
>> --- a/fs/nfs/pagelist.c
>> +++ b/fs/nfs/pagelist.c
>> @@ -26,6 +26,8 @@
>>
>> static struct kmem_cache *nfs_page_cachep;
>>
>> +static void nfs_free_request(struct nfs_page *);
>> +
>> bool nfs_pgarray_set(struct nfs_page_array *p, unsigned int pagecount)
>> {
>> p->npages = pagecount;
>> @@ -133,10 +135,145 @@ nfs_iocounter_wait(struct nfs_io_counter *c)
>> return __nfs_iocounter_wait(c);
>> }
>>
>> +/*
>> + * nfs_page_group_lock - lock the head of the page group
>> + * @req - request in group that is to be locked
>> + *
>> + * this lock must be held if modifying the page group list
>> + */
>> +void
>> +nfs_page_group_lock(struct nfs_page *req)
>> +{
>> + struct nfs_page *head = req->wb_head;
>> + int err = -EAGAIN;
>> +
>> + WARN_ON_ONCE(head != head->wb_head);
>> +
>> + while (err)
>> + err = wait_on_bit_lock(&head->wb_flags, PG_HEADLOCK,
>> + nfs_wait_bit_killable, TASK_KILLABLE);
>> +}
>> +
>> +/*
>> + * nfs_page_group_unlock - unlock the head of the page group
>> + * @req - request in group that is to be unlocked
>> + */
>> +void
>> +nfs_page_group_unlock(struct nfs_page *req)
>> +{
>> + struct nfs_page *head = req->wb_head;
>> +
>> + WARN_ON_ONCE(head != head->wb_head);
>> +
>> + smp_mb__before_clear_bit();
>> + clear_bit(PG_HEADLOCK, &head->wb_flags);
>> + smp_mb__after_clear_bit();
>> + wake_up_bit(&head->wb_flags, PG_HEADLOCK);
>> +}
>> +
>> +/*
>> + * nfs_page_group_sync_on_bit_locked
>> + *
>> + * must be called with page group lock held
>> + */
>> +static bool
>> +nfs_page_group_sync_on_bit_locked(struct nfs_page *req, unsigned int bit)
>> +{
>> + struct nfs_page *head = req->wb_head;
>> + struct nfs_page *tmp;
>> +
>> + WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_flags));
>> + WARN_ON_ONCE(test_and_set_bit(bit, &req->wb_flags));
>> +
>> + tmp = req->wb_this_page;
>> + while (tmp != req) {
>> + if (!test_bit(bit, &tmp->wb_flags))
>> + return false;
>> + tmp = tmp->wb_this_page;
>> + }
>> +
>> + /* true! reset all bits */
>> + tmp = req;
>> + do {
>> + clear_bit(bit, &tmp->wb_flags);
>> + tmp = tmp->wb_this_page;
>> + } while (tmp != req);
>> +
>> + return true;
>> +}
>> +
>> +/*
>> + * nfs_page_group_sync_on_bit - set bit on current request, but only
>> + * return true if the bit is set for all requests in page group
>> + * @req - request in page group
>> + * @bit - PG_* bit that is used to sync page group
>> + */
>> +bool nfs_page_group_sync_on_bit(struct nfs_page *req, unsigned int bit)
>> +{
>> + bool ret;
>> +
>> + nfs_page_group_lock(req);
>> + ret = nfs_page_group_sync_on_bit_locked(req, bit);
>> + nfs_page_group_unlock(req);
>> +
>> + return ret;
>> +}
>> +
>> +/*
>> + * nfs_page_group_init - Initialize the page group linkage for @req
>> + * @req - a new nfs request
>> + * @prev - the previous request in page group, or NULL if @req is the first
>> + * or only request in the group (the head).
>> + */
>> +static inline void
>> +nfs_page_group_init(struct nfs_page *req, struct nfs_page *prev)
>> +{
>> + WARN_ON_ONCE(prev == req);
>> +
>> + if (!prev) {
>> + req->wb_head = req;
>> + req->wb_this_page = req;
>> + } else {
>> + WARN_ON_ONCE(prev->wb_this_page != prev->wb_head);
>> + WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &prev->wb_head->wb_flags));
>> + req->wb_head = prev->wb_head;
>> + req->wb_this_page = prev->wb_this_page;
>> + prev->wb_this_page = req;
>> + }
>> +}
>> +
>> +/*
>> + * nfs_page_group_destroy - sync the destruction of page groups
>> + * @req - request that no longer needs the page group
>> + *
>> + * releases the page group reference from each member once all
>> + * members have called this function.
>> + */
>> +static void
>> +nfs_page_group_destroy(struct kref *kref)
>> +{
>> + struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref);
>> + struct nfs_page *tmp, *next;
>> +
>> + if (!nfs_page_group_sync_on_bit(req, PG_TEARDOWN))
>> + return;
>> +
>> + tmp = req;
>> + do {
>> + next = tmp->wb_this_page;
>> + /* unlink and free */
>> + tmp->wb_this_page = tmp;
>> + tmp->wb_head = tmp;
>> + nfs_free_request(tmp);
>> + tmp = next;
>> + } while (tmp != req);
>> +}
>> +
>> /**
>> * nfs_create_request - Create an NFS read/write request.
>> * @ctx: open context to use
>> * @page: page to write
>> + * @last: last nfs request created for this page group or NULL if head
>> * @offset: starting offset within the page for the write
>> * @count: number of bytes to read/write
>> *
>> @@ -146,7 +283,8 @@ nfs_iocounter_wait(struct nfs_io_counter *c)
>> */
>> struct nfs_page *
>> nfs_create_request(struct nfs_open_context *ctx, struct page *page,
>> - unsigned int offset, unsigned int count)
>> + struct nfs_page *last, unsigned int offset,
>> + unsigned int count)
>> {
>> struct nfs_page *req;
>> struct nfs_lock_context *l_ctx;
>> @@ -178,6 +316,7 @@ nfs_create_request(struct nfs_open_context *ctx, struct page *page,
>> req->wb_bytes = count;
>> req->wb_context = get_nfs_open_context(ctx);
>> kref_init(&req->wb_kref);
>> + nfs_page_group_init(req, last);
>> return req;
>> }
>>
>> @@ -235,16 +374,22 @@ static void nfs_clear_request(struct nfs_page *req)
>> }
>> }
>>
>> -
>> /**
>> * nfs_release_request - Release the count on an NFS read/write request
>> * @req: request to release
>> *
>> * Note: Should never be called with the spinlock held!
>> */
>> -static void nfs_free_request(struct kref *kref)
>> +static void nfs_free_request(struct nfs_page *req)
>> {
>> - struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref);
>> + WARN_ON_ONCE(req->wb_this_page != req);
>> +
>> + /* extra debug: make sure no sync bits are still set */
>> + WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags));
>> + WARN_ON_ONCE(test_bit(PG_UNLOCKPAGE, &req->wb_flags));
>> + WARN_ON_ONCE(test_bit(PG_UPTODATE, &req->wb_flags));
>> + WARN_ON_ONCE(test_bit(PG_WB_END, &req->wb_flags));
>> + WARN_ON_ONCE(test_bit(PG_REMOVE, &req->wb_flags));
>>
>> /* Release struct file and open context */
>> nfs_clear_request(req);
>> @@ -253,7 +398,7 @@ static void nfs_free_request(struct kref *kref)
>>
>> void nfs_release_request(struct nfs_page *req)
>> {
>> - kref_put(&req->wb_kref, nfs_free_request);
>> + kref_put(&req->wb_kref, nfs_page_group_destroy);
>> }
>>
>> static int nfs_wait_bit_uninterruptible(void *word)
>> @@ -439,21 +584,66 @@ static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
>> * @desc: destination io descriptor
>> * @req: request
>> *
>> + * This may split a request into subrequests which are all part of the
>> + * same page group.
>> + *
>> * Returns true if the request 'req' was successfully coalesced into the
>> * existing list of pages 'desc'.
>> */
>> static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
>> struct nfs_page *req)
>> {
>> - while (!nfs_pageio_do_add_request(desc, req)) {
>> - desc->pg_moreio = 1;
>> - nfs_pageio_doio(desc);
>> - if (desc->pg_error < 0)
>> - return 0;
>> - desc->pg_moreio = 0;
>> - if (desc->pg_recoalesce)
>> - return 0;
>> - }
>> + struct nfs_page *subreq;
>> + unsigned int bytes_left = 0;
>> + unsigned int offset, pgbase;
>> +
>> + nfs_page_group_lock(req);
>> +
>> + subreq = req;
>> + bytes_left = subreq->wb_bytes;
>> + offset = subreq->wb_offset;
>> + pgbase = subreq->wb_pgbase;
>> +
>> + do {
>> + if (!nfs_pageio_do_add_request(desc, subreq)) {
>> + /* make sure pg_test call(s) did nothing */
>> + WARN_ON_ONCE(subreq->wb_bytes != bytes_left);
>> + WARN_ON_ONCE(subreq->wb_offset != offset);
>> + WARN_ON_ONCE(subreq->wb_pgbase != pgbase);
>> +
>> + nfs_page_group_unlock(req);
>> + desc->pg_moreio = 1;
>> + nfs_pageio_doio(desc);
>> + if (desc->pg_error < 0)
>> + return 0;
>> + desc->pg_moreio = 0;
>> + if (desc->pg_recoalesce)
>> + return 0;
>> + /* retry add_request for this subreq */
>> + nfs_page_group_lock(req);
>> + continue;
>> + }
>> +
>> + /* check for buggy pg_test call(s) */
>> + WARN_ON_ONCE(subreq->wb_bytes + subreq->wb_pgbase > PAGE_SIZE);
>> + WARN_ON_ONCE(subreq->wb_bytes > bytes_left);
>> + WARN_ON_ONCE(subreq->wb_bytes == 0);
>> +
>> + bytes_left -= subreq->wb_bytes;
>> + offset += subreq->wb_bytes;
>> + pgbase += subreq->wb_bytes;
>> +
>> + if (bytes_left) {
>> + subreq = nfs_create_request(req->wb_context,
>> + req->wb_page,
>> + subreq, pgbase, bytes_left);
>> + nfs_lock_request(subreq);
>> + subreq->wb_offset = offset;
>> + subreq->wb_index = req->wb_index;
>> + }
>> + } while (bytes_left > 0);
>> +
>> + nfs_page_group_unlock(req);
>> return 1;
>> }
>>
>> diff --git a/fs/nfs/read.c b/fs/nfs/read.c
>> index 95a0855..ee0a3cd 100644
>> --- a/fs/nfs/read.c
>> +++ b/fs/nfs/read.c
>> @@ -139,7 +139,7 @@ int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode,
>> len = nfs_page_length(page);
>> if (len == 0)
>> return nfs_return_empty_page(page);
>> - new = nfs_create_request(ctx, page, 0, len);
>> + new = nfs_create_request(ctx, page, NULL, 0, len);
>> if (IS_ERR(new)) {
>> unlock_page(page);
>> return PTR_ERR(new);
>> @@ -600,7 +600,7 @@ readpage_async_filler(void *data, struct page *page)
>> if (len == 0)
>> return nfs_return_empty_page(page);
>>
>> - new = nfs_create_request(desc->ctx, page, 0, len);
>> + new = nfs_create_request(desc->ctx, page, NULL, 0, len);
>> if (IS_ERR(new))
>> goto out_error;
>>
>> diff --git a/fs/nfs/write.c b/fs/nfs/write.c
>> index ca20ec7..d1453f2 100644
>> --- a/fs/nfs/write.c
>> +++ b/fs/nfs/write.c
>> @@ -461,7 +461,7 @@ static void nfs_inode_remove_request(struct nfs_page *req)
>> }
>> nfsi->npages--;
>> spin_unlock(&inode->i_lock);
>> - nfs_release_request(req);
>> + nfs_release_request(head);
>> }
>>
>> static void
>> @@ -625,6 +625,7 @@ static void nfs_write_completion(struct nfs_pgio_header *hdr)
>> {
>> struct nfs_commit_info cinfo;
>> unsigned long bytes = 0;
>> + bool do_destroy;
>>
>> if (test_bit(NFS_IOHDR_REDO, &hdr->flags))
>> goto out;
>> @@ -654,6 +655,7 @@ remove_req:
>> next:
>> nfs_unlock_request(req);
>> nfs_end_page_writeback(req->wb_page);
>> + do_destroy = !test_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags);
>> nfs_release_request(req);
>> }
>> out:
>> @@ -758,6 +760,10 @@ static struct nfs_page *nfs_try_to_update_request(struct inode *inode,
>> if (req == NULL)
>> goto out_unlock;
>>
>> + /* should be handled by nfs_flush_incompatible */
>> + WARN_ON_ONCE(req->wb_head != req);
>> + WARN_ON_ONCE(req->wb_this_page != req);
>> +
>> rqend = req->wb_offset + req->wb_bytes;
>> /*
>> * Tell the caller to flush out the request if
>> @@ -819,7 +825,7 @@ static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx,
>> req = nfs_try_to_update_request(inode, page, offset, bytes);
>> if (req != NULL)
>> goto out;
>> - req = nfs_create_request(ctx, page, offset, bytes);
>> + req = nfs_create_request(ctx, page, NULL, offset, bytes);
>> if (IS_ERR(req))
>> goto out;
>> nfs_inode_add_request(inode, req);
>> @@ -863,6 +869,8 @@ int nfs_flush_incompatible(struct file *file, struct page *page)
>> return 0;
>> l_ctx = req->wb_lock_context;
>> do_flush = req->wb_page != page || req->wb_context != ctx;
>> + /* for now, flush if more than 1 request in page_group */
>> + do_flush |= req->wb_this_page != req;
>> if (l_ctx && ctx->dentry->d_inode->i_flock != NULL) {
>> do_flush |= l_ctx->lockowner.l_owner != current->files
>> || l_ctx->lockowner.l_pid != current->tgid;
>> diff --git a/include/linux/nfs_page.h b/include/linux/nfs_page.h
>> index 214e098..1fb161b 100644
>> --- a/include/linux/nfs_page.h
>> +++ b/include/linux/nfs_page.h
>> @@ -26,6 +26,8 @@ enum {
>> PG_MAPPED, /* page private set for buffered io */
>> PG_CLEAN, /* write succeeded */
>> PG_COMMIT_TO_DS, /* used by pnfs layouts */
>> + PG_HEADLOCK, /* page group lock of wb_head */
>> + PG_TEARDOWN, /* page group sync for destroy */
>> };
>>
>> struct nfs_inode;
>> @@ -41,6 +43,8 @@ struct nfs_page {
>> struct kref wb_kref; /* reference count */
>> unsigned long wb_flags;
>> struct nfs_write_verifier wb_verf; /* Commit cookie */
>> + struct nfs_page *wb_this_page; /* list of reqs for this page */
>> + struct nfs_page *wb_head; /* head pointer for req list */
>
> Hmm ok, so to make sure I understand...
>
> So page->private will point to the "head" req (struct page_private).
Only in the buffered write case. Page->private is not set for read path / direct i/o path.
> Then we'll have a singly-linked list of reqs hanging off of
> wb_this_page. Is that right?
>
> If so, then it seems like it would be clearer to use a standard
> list_head here. If you need to get to the wb_head, you could always do
> something like this:
>
> list_first_entry(&req->wb_page->wb_this_page);
Well, wb_page is a struct page and doesn’t have wb_this_page (which is in struct
nfs_page), but I see where you’re going with this.
A strategy like this only works if we always have page->private pointing to the head
request. We chose not to go that way because it messes with the buffered
write path’s setting / clearing of page private which interacts with the swappable
nfs pages code that everyone seems to be afraid to touch ;)
So we decided to go this route (not messing with page_private) as a first step - we
certainly could add it later, but the current approach makes things less complex.
>
> ...and could even turn that into a macro or static inline for some
> syntactic sugar. It's a little more pointer chasing to find the head,
> but it seems like that would be clearer than using yet another
> linked-list implementation.
So, I’m not against using list_head.. I didn’t go that route initially because I was:
1) following the buffer_head example, which rolls it’s own list
2) trying to grow nfs_page as little as possible - but we might have room within
the allocator bucket it currently lives in…
3) not sure list_head is suitable for a circular list (I haven’t ever looked into it).
and until we have a way to find the head request (via page private, etc) without
walking the circular list (chicken / egg problem needing to grab head lock before walking
list to find the head to lock it), we’ll still need the head pointer.
Thoughts?
-dros
>
>> };
>>
>> struct nfs_pageio_descriptor;
>> @@ -75,9 +79,10 @@ struct nfs_pageio_descriptor {
>>
>> extern struct nfs_page *nfs_create_request(struct nfs_open_context *ctx,
>> struct page *page,
>> + struct nfs_page *last,
>> unsigned int offset,
>> unsigned int count);
>> -extern void nfs_release_request(struct nfs_page *req);
>> +extern void nfs_release_request(struct nfs_page *);
>>
>>
>> extern void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
>> @@ -95,7 +100,10 @@ extern size_t nfs_generic_pg_test(struct nfs_pageio_descriptor *desc,
>> struct nfs_page *req);
>> extern int nfs_wait_on_request(struct nfs_page *);
>> extern void nfs_unlock_request(struct nfs_page *req);
>> -extern void nfs_unlock_and_release_request(struct nfs_page *req);
>> +extern void nfs_unlock_and_release_request(struct nfs_page *);
>> +extern void nfs_page_group_lock(struct nfs_page *);
>> +extern void nfs_page_group_unlock(struct nfs_page *);
>> +extern bool nfs_page_group_sync_on_bit(struct nfs_page *, unsigned int);
>>
>> /*
>> * Lock the page of an asynchronous request
>
>
> --
> Jeff Layton <jlayton@xxxxxxxxxxxxxxx>
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
