From df9f86faf3ee610527ed02031fe7dd3c8b752e44 Mon Sep 17 00:00:00 2001 From: Sage Weil Date: Mon, 1 Nov 2010 15:49:23 -0700 Subject: ceph: fix small seq message skipping If the client gets out of sync with the server message sequence number, we normally skip low seq messages (ones we already received). The skip code was also incrementing the expected seq, such that all subsequent messages also appeared old and got skipped, and an eventual timeout on the osd connection. This resulted in some lagging requests and console messages like [233480.882885] ceph: skipping osd22 10.138.138.13:6804 seq 2016, expected 2017 [233480.882919] ceph: skipping osd22 10.138.138.13:6804 seq 2017, expected 2018 [233480.882963] ceph: skipping osd22 10.138.138.13:6804 seq 2018, expected 2019 [233480.883488] ceph: skipping osd22 10.138.138.13:6804 seq 2019, expected 2020 [233485.219558] ceph: skipping osd22 10.138.138.13:6804 seq 2020, expected 2021 [233485.906595] ceph: skipping osd22 10.138.138.13:6804 seq 2021, expected 2022 [233490.379536] ceph: skipping osd22 10.138.138.13:6804 seq 2022, expected 2023 [233495.523260] ceph: skipping osd22 10.138.138.13:6804 seq 2023, expected 2024 [233495.923194] ceph: skipping osd22 10.138.138.13:6804 seq 2024, expected 2025 [233500.534614] ceph: tid 6023602 timed out on osd22, will reset osd Reported-by: Theodore Ts'o Signed-off-by: Sage Weil --- net/ceph/messenger.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) (limited to 'net/ceph') diff --git a/net/ceph/messenger.c b/net/ceph/messenger.c index 0e8157ee5d43..d379abf873bc 100644 --- a/net/ceph/messenger.c +++ b/net/ceph/messenger.c @@ -1532,14 +1532,13 @@ static int read_partial_message(struct ceph_connection *con) /* verify seq# */ seq = le64_to_cpu(con->in_hdr.seq); if ((s64)seq - (s64)con->in_seq < 1) { - pr_info("skipping %s%lld %s seq %lld, expected %lld\n", + pr_info("skipping %s%lld %s seq %lld expected %lld\n", ENTITY_NAME(con->peer_name), ceph_pr_addr(&con->peer_addr.in_addr), seq, con->in_seq + 1); con->in_base_pos = -front_len - middle_len - data_len - sizeof(m->footer); con->in_tag = CEPH_MSGR_TAG_READY; - con->in_seq++; return 0; } else if ((s64)seq - (s64)con->in_seq > 1) { pr_err("read_partial_message bad seq %lld expected %lld\n", -- cgit v1.2.3 From e98b6fed84d0f0155d7b398e0dfeac74c792f2d0 Mon Sep 17 00:00:00 2001 From: Sage Weil Date: Tue, 9 Nov 2010 12:24:53 -0800 Subject: ceph: fix comment, remove extraneous args The offset/length arguments aren't used. Signed-off-by: Sage Weil --- net/ceph/pagevec.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) (limited to 'net/ceph') diff --git a/net/ceph/pagevec.c b/net/ceph/pagevec.c index 54caf0687155..ac34feeb2b3a 100644 --- a/net/ceph/pagevec.c +++ b/net/ceph/pagevec.c @@ -13,8 +13,7 @@ * build a vector of user pages */ struct page **ceph_get_direct_page_vector(const char __user *data, - int num_pages, - loff_t off, size_t len) + int num_pages) { struct page **pages; int rc; -- cgit v1.2.3 From b7495fc2ff941db6a118a93ab8d61149e3f4cef8 Mon Sep 17 00:00:00 2001 From: Sage Weil Date: Tue, 9 Nov 2010 12:43:12 -0800 Subject: ceph: make page alignment explicit in osd interface We used to infer alignment of IOs within a page based on the file offset, which assumed they matched. This broke with direct IO that was not aligned to pages (e.g., 512-byte aligned IO). We were also trusting the alignment specified in the OSD reply, which could have been adjusted by the server. Explicitly specify the page alignment when setting up OSD IO requests. Signed-off-by: Sage Weil --- net/ceph/osd_client.c | 22 ++++++++++++++-------- 1 file changed, 14 insertions(+), 8 deletions(-) (limited to 'net/ceph') diff --git a/net/ceph/osd_client.c b/net/ceph/osd_client.c index 79391994b3ed..6c096239660c 100644 --- a/net/ceph/osd_client.c +++ b/net/ceph/osd_client.c @@ -71,6 +71,7 @@ void ceph_calc_raw_layout(struct ceph_osd_client *osdc, op->extent.length = objlen; } req->r_num_pages = calc_pages_for(off, *plen); + req->r_page_alignment = off & ~PAGE_MASK; if (op->op == CEPH_OSD_OP_WRITE) op->payload_len = *plen; @@ -419,7 +420,8 @@ struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc, u32 truncate_seq, u64 truncate_size, struct timespec *mtime, - bool use_mempool, int num_reply) + bool use_mempool, int num_reply, + int page_align) { struct ceph_osd_req_op ops[3]; struct ceph_osd_request *req; @@ -447,6 +449,10 @@ struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc, calc_layout(osdc, vino, layout, off, plen, req, ops); req->r_file_layout = *layout; /* keep a copy */ + /* in case it differs from natural alignment that calc_layout + filled in for us */ + req->r_page_alignment = page_align; + ceph_osdc_build_request(req, off, plen, ops, snapc, mtime, @@ -1489,7 +1495,7 @@ int ceph_osdc_readpages(struct ceph_osd_client *osdc, struct ceph_vino vino, struct ceph_file_layout *layout, u64 off, u64 *plen, u32 truncate_seq, u64 truncate_size, - struct page **pages, int num_pages) + struct page **pages, int num_pages, int page_align) { struct ceph_osd_request *req; int rc = 0; @@ -1499,15 +1505,15 @@ int ceph_osdc_readpages(struct ceph_osd_client *osdc, req = ceph_osdc_new_request(osdc, layout, vino, off, plen, CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ, NULL, 0, truncate_seq, truncate_size, NULL, - false, 1); + false, 1, page_align); if (!req) return -ENOMEM; /* it may be a short read due to an object boundary */ req->r_pages = pages; - dout("readpages final extent is %llu~%llu (%d pages)\n", - off, *plen, req->r_num_pages); + dout("readpages final extent is %llu~%llu (%d pages align %d)\n", + off, *plen, req->r_num_pages, page_align); rc = ceph_osdc_start_request(osdc, req, false); if (!rc) @@ -1533,6 +1539,7 @@ int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino, { struct ceph_osd_request *req; int rc = 0; + int page_align = off & ~PAGE_MASK; BUG_ON(vino.snap != CEPH_NOSNAP); req = ceph_osdc_new_request(osdc, layout, vino, off, &len, @@ -1541,7 +1548,7 @@ int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino, CEPH_OSD_FLAG_WRITE, snapc, do_sync, truncate_seq, truncate_size, mtime, - nofail, 1); + nofail, 1, page_align); if (!req) return -ENOMEM; @@ -1638,8 +1645,7 @@ static struct ceph_msg *get_reply(struct ceph_connection *con, m = ceph_msg_get(req->r_reply); if (data_len > 0) { - unsigned data_off = le16_to_cpu(hdr->data_off); - int want = calc_pages_for(data_off & ~PAGE_MASK, data_len); + int want = calc_pages_for(req->r_page_alignment, data_len); if (unlikely(req->r_num_pages < want)) { pr_warning("tid %lld reply %d > expected %d pages\n", -- cgit v1.2.3 From c5c6b19d4b8f5431fca05f28ae9e141045022149 Mon Sep 17 00:00:00 2001 From: Sage Weil Date: Tue, 9 Nov 2010 12:40:00 -0800 Subject: ceph: explicitly specify page alignment in network messages The alignment used for reading data into or out of pages used to be taken from the data_off field in the message header. This only worked as long as the page alignment matched the object offset, breaking direct io to non-page aligned offsets. Instead, explicitly specify the page alignment next to the page vector in the ceph_msg struct, and use that instead of the message header (which probably shouldn't be trusted). The alloc_msg callback is responsible for filling in this field properly when it sets up the page vector. Signed-off-by: Sage Weil --- net/ceph/messenger.c | 10 +++++----- net/ceph/osd_client.c | 3 +++ 2 files changed, 8 insertions(+), 5 deletions(-) (limited to 'net/ceph') diff --git a/net/ceph/messenger.c b/net/ceph/messenger.c index d379abf873bc..1c7a2ec4f3cc 100644 --- a/net/ceph/messenger.c +++ b/net/ceph/messenger.c @@ -540,8 +540,7 @@ static void prepare_write_message(struct ceph_connection *con) /* initialize page iterator */ con->out_msg_pos.page = 0; if (m->pages) - con->out_msg_pos.page_pos = - le16_to_cpu(m->hdr.data_off) & ~PAGE_MASK; + con->out_msg_pos.page_pos = m->page_alignment; else con->out_msg_pos.page_pos = 0; con->out_msg_pos.data_pos = 0; @@ -1491,7 +1490,7 @@ static int read_partial_message(struct ceph_connection *con) struct ceph_msg *m = con->in_msg; int ret; int to, left; - unsigned front_len, middle_len, data_len, data_off; + unsigned front_len, middle_len, data_len; int datacrc = con->msgr->nocrc; int skip; u64 seq; @@ -1527,7 +1526,6 @@ static int read_partial_message(struct ceph_connection *con) data_len = le32_to_cpu(con->in_hdr.data_len); if (data_len > CEPH_MSG_MAX_DATA_LEN) return -EIO; - data_off = le16_to_cpu(con->in_hdr.data_off); /* verify seq# */ seq = le64_to_cpu(con->in_hdr.seq); @@ -1575,7 +1573,7 @@ static int read_partial_message(struct ceph_connection *con) con->in_msg_pos.page = 0; if (m->pages) - con->in_msg_pos.page_pos = data_off & ~PAGE_MASK; + con->in_msg_pos.page_pos = m->page_alignment; else con->in_msg_pos.page_pos = 0; con->in_msg_pos.data_pos = 0; @@ -2300,6 +2298,7 @@ struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags) /* data */ m->nr_pages = 0; + m->page_alignment = 0; m->pages = NULL; m->pagelist = NULL; m->bio = NULL; @@ -2369,6 +2368,7 @@ static struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con, type, front_len); return NULL; } + msg->page_alignment = le16_to_cpu(hdr->data_off); } memcpy(&msg->hdr, &con->in_hdr, sizeof(con->in_hdr)); diff --git a/net/ceph/osd_client.c b/net/ceph/osd_client.c index 6c096239660c..3e20a122ffa2 100644 --- a/net/ceph/osd_client.c +++ b/net/ceph/osd_client.c @@ -391,6 +391,8 @@ void ceph_osdc_build_request(struct ceph_osd_request *req, req->r_request->hdr.data_len = cpu_to_le32(data_len); } + req->r_request->page_alignment = req->r_page_alignment; + BUG_ON(p > msg->front.iov_base + msg->front.iov_len); msg_size = p - msg->front.iov_base; msg->front.iov_len = msg_size; @@ -1657,6 +1659,7 @@ static struct ceph_msg *get_reply(struct ceph_connection *con, } m->pages = req->r_pages; m->nr_pages = req->r_num_pages; + m->page_alignment = req->r_page_alignment; #ifdef CONFIG_BLOCK m->bio = req->r_bio; #endif -- cgit v1.2.3 From d96c9043d1588f04c7f467167f653c07d83232d5 Mon Sep 17 00:00:00 2001 From: Sage Weil Date: Mon, 13 Dec 2010 20:30:28 -0800 Subject: ceph: fix msgr_init error path create_workqueue() returns NULL on failure. Signed-off-by: Sage Weil --- net/ceph/messenger.c | 8 +++----- 1 file changed, 3 insertions(+), 5 deletions(-) (limited to 'net/ceph') diff --git a/net/ceph/messenger.c b/net/ceph/messenger.c index 1c7a2ec4f3cc..b6ff4a1519ab 100644 --- a/net/ceph/messenger.c +++ b/net/ceph/messenger.c @@ -97,11 +97,9 @@ struct workqueue_struct *ceph_msgr_wq; int ceph_msgr_init(void) { ceph_msgr_wq = create_workqueue("ceph-msgr"); - if (IS_ERR(ceph_msgr_wq)) { - int ret = PTR_ERR(ceph_msgr_wq); - pr_err("msgr_init failed to create workqueue: %d\n", ret); - ceph_msgr_wq = NULL; - return ret; + if (!ceph_msgr_wq) { + pr_err("msgr_init failed to create workqueue\n"); + return -ENOMEM; } return 0; } -- cgit v1.2.3 From b6aa5901c7a2bd90d0b6b9866300d2648b2568f3 Mon Sep 17 00:00:00 2001 From: Henry C Chang Date: Wed, 15 Dec 2010 20:45:41 -0800 Subject: ceph: mark user pages dirty on direct-io reads For read operation, we have to set the argument _write_ of get_user_pages to 1 since we will write data to pages. Also, we need to SetPageDirty before releasing these pages. Signed-off-by: Henry C Chang Signed-off-by: Sage Weil --- net/ceph/pagevec.c | 11 +++++++---- 1 file changed, 7 insertions(+), 4 deletions(-) (limited to 'net/ceph') diff --git a/net/ceph/pagevec.c b/net/ceph/pagevec.c index ac34feeb2b3a..01947a5d03b7 100644 --- a/net/ceph/pagevec.c +++ b/net/ceph/pagevec.c @@ -13,7 +13,7 @@ * build a vector of user pages */ struct page **ceph_get_direct_page_vector(const char __user *data, - int num_pages) + int num_pages, bool write_page) { struct page **pages; int rc; @@ -24,7 +24,7 @@ struct page **ceph_get_direct_page_vector(const char __user *data, down_read(¤t->mm->mmap_sem); rc = get_user_pages(current, current->mm, (unsigned long)data, - num_pages, 0, 0, pages, NULL); + num_pages, write_page, 0, pages, NULL); up_read(¤t->mm->mmap_sem); if (rc < 0) goto fail; @@ -36,12 +36,15 @@ fail: } EXPORT_SYMBOL(ceph_get_direct_page_vector); -void ceph_put_page_vector(struct page **pages, int num_pages) +void ceph_put_page_vector(struct page **pages, int num_pages, bool dirty) { int i; - for (i = 0; i < num_pages; i++) + for (i = 0; i < num_pages; i++) { + if (dirty) + set_page_dirty_lock(pages[i]); put_page(pages[i]); + } kfree(pages); } EXPORT_SYMBOL(ceph_put_page_vector); -- cgit v1.2.3 From 361cf40519a491f68b28ad90225e4611c4bf8e12 Mon Sep 17 00:00:00 2001 From: Henry C Chang Date: Fri, 17 Dec 2010 09:55:59 -0800 Subject: ceph: handle partial result from get_user_pages The get_user_pages() helper can return fewer than the requested pages. Error out in that case, and clean up the partial result. Signed-off-by: Henry C Chang Signed-off-by: Sage Weil --- net/ceph/pagevec.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'net/ceph') diff --git a/net/ceph/pagevec.c b/net/ceph/pagevec.c index 01947a5d03b7..1a040e64c69f 100644 --- a/net/ceph/pagevec.c +++ b/net/ceph/pagevec.c @@ -26,12 +26,12 @@ struct page **ceph_get_direct_page_vector(const char __user *data, rc = get_user_pages(current, current->mm, (unsigned long)data, num_pages, write_page, 0, pages, NULL); up_read(¤t->mm->mmap_sem); - if (rc < 0) + if (rc < num_pages) goto fail; return pages; fail: - kfree(pages); + ceph_put_page_vector(pages, rc > 0 ? rc : 0, false); return ERR_PTR(rc); } EXPORT_SYMBOL(ceph_get_direct_page_vector); -- cgit v1.2.3