Re: [RFC PATCH 18/18] ceph: Don't use data_pages

From: Xiubo Li
Date: Sun Aug 27 2023 - 21:34:44 EST

Next message: Jijie Shao: "Re: [PATCH V2 drivers/perf: hisi:] drivers/perf: hisi: Update HiSilicon PMU maintainers"
Previous message: Xiubo Li: "Re: [RFC PATCH 00/18] ceph, rbd: Collapse all the I/O types down to something iov_iter-based"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

David,

Please let me know if this is ready and I will review it carefully and have a test for them. Haven't got a time to go through this yet.

Thanks

- Xiubo

On 8/4/23 21:13, David Howells wrote:

---
fs/ceph/addr.c | 16 +++++-----------
fs/ceph/file.c | 34 +++++++++++++++------------------
include/linux/ceph/osd_client.h | 1 +
net/ceph/osd_client.c | 16 ++--------------
4 files changed, 23 insertions(+), 44 deletions(-)

diff --git a/fs/ceph/addr.c b/fs/ceph/addr.c
index 7571606cf61f..7557f4a85ef0 100644
--- a/fs/ceph/addr.c
+++ b/fs/ceph/addr.c
@@ -277,11 +277,6 @@ static void finish_netfs_read(struct ceph_osd_request *req)
}
}
- if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
- ceph_put_page_vector(osd_data->pages,
- calc_pages_for(osd_data->offset,
- osd_data->length), false);
- }
netfs_subreq_terminated(subreq, err, false);
iput(req->r_inode);
ceph_dec_osd_stopping_blocker(fsc->mdsc);
@@ -2007,7 +2002,7 @@ static int __ceph_pool_perm_get(struct ceph_inode_info *ci,
struct ceph_osd_request *rd_req = NULL, *wr_req = NULL;
struct rb_node **p, *parent;
struct ceph_pool_perm *perm;
- struct page **pages;
+ struct ceph_databuf *dbuf;
size_t pool_ns_len;
int err = 0, err2 = 0, have = 0;
@@ -2107,14 +2102,13 @@ static int __ceph_pool_perm_get(struct ceph_inode_info *ci,
goto out_unlock;
/* one page should be large enough for STAT data */
- pages = ceph_alloc_page_vector(1, GFP_KERNEL);
- if (IS_ERR(pages)) {
- err = PTR_ERR(pages);
+ dbuf = ceph_databuf_alloc(1, PAGE_SIZE, GFP_KERNEL);
+ if (!dbuf) {
+ err = -ENOMEM;
goto out_unlock;
}
- osd_req_op_raw_data_in_pages(rd_req, 0, pages, PAGE_SIZE,
- 0, false, true);
+ osd_req_op_raw_data_in_databuf(rd_req, 0, dbuf);
ceph_osdc_start_request(&fsc->client->osdc, rd_req);
wr_req->r_mtime = ci->netfs.inode.i_mtime;
diff --git a/fs/ceph/file.c b/fs/ceph/file.c
index 5d16469a3690..caf557187ca8 100644
--- a/fs/ceph/file.c
+++ b/fs/ceph/file.c
@@ -977,6 +977,7 @@ ssize_t __ceph_sync_read(struct inode *inode, loff_t *ki_pos,
struct ceph_fs_client *fsc = ceph_inode_to_fs_client(inode);
struct ceph_client *cl = fsc->client;
struct ceph_osd_client *osdc = &fsc->client->osdc;
+ struct ceph_databuf *dbuf;
ssize_t ret;
u64 off = *ki_pos;
u64 len = iov_iter_count(to);
@@ -1041,16 +1042,14 @@ ssize_t __ceph_sync_read(struct inode *inode, loff_t *ki_pos,
num_pages = calc_pages_for(read_off, read_len);
page_off = offset_in_page(off);
- pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
- if (IS_ERR(pages)) {
+ dbuf = ceph_databuf_alloc(num_pages, read_len, GFP_KERNEL);
+ if (!dbuf) {
ceph_osdc_put_request(req);
- ret = PTR_ERR(pages);
+ ret = -ENOMEM;
break;
}
- osd_req_op_extent_osd_data_pages(req, 0, pages, read_len,
- offset_in_page(read_off),
- false, false);
+ osd_req_op_extent_osd_databuf(req, 0, dbuf);
op = &req->r_ops[0];
if (sparse) {
@@ -1137,7 +1136,7 @@ ssize_t __ceph_sync_read(struct inode *inode, loff_t *ki_pos,
break;
}
}
- ceph_release_page_vector(pages, num_pages);
+ ceph_databuf_release(dbuf);
if (ret < 0) {
if (ret == -EBLOCKLISTED)
@@ -1625,7 +1624,7 @@ ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos,
struct ceph_client *cl = fsc->client;
struct ceph_osd_client *osdc = &fsc->client->osdc;
struct ceph_osd_request *req;
- struct page **pages;
+ struct ceph_databuf *dbuf = NULL;
u64 len;
int num_pages;
int written = 0;
@@ -1691,9 +1690,9 @@ ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos,
* an array of pagecache pages.
*/
num_pages = calc_pages_for(write_pos, write_len);
- pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
- if (IS_ERR(pages)) {
- ret = PTR_ERR(pages);
+ dbuf = ceph_databuf_alloc(num_pages, num_pages * PAGE_SIZE, GFP_KERNEL);
+ if (!dbuf) {
+ ret = -ENOMEM;
break;
}
@@ -1722,7 +1721,6 @@ ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos,
NULL, ci->i_truncate_seq,
ci->i_truncate_size, false);
if (IS_ERR(req)) {
- ceph_release_page_vector(pages, num_pages);
ret = PTR_ERR(req);
break;
}
@@ -1730,7 +1728,6 @@ ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos,
/* Something is misaligned! */
if (read_len != CEPH_FSCRYPT_BLOCK_SIZE) {
ceph_osdc_put_request(req);
- ceph_release_page_vector(pages, num_pages);
ret = -EIO;
break;
}
@@ -1739,15 +1736,14 @@ ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos,
op = &req->r_ops[0];
if (first) {
- osd_req_op_extent_osd_data_pages(req, 0, pages,
- CEPH_FSCRYPT_BLOCK_SIZE,
- offset_in_page(first_pos),
- false, false);
+ iov_iter_advance(&dbuf->iter, offset_in_page(first_pos));
+ osd_req_op_extent_osd_databuf(req, 0, dbuf,
+ CEPH_FSCRYPT_BLOCK_SIZE);
+ dbuf = NULL;
/* We only expect a single extent here */
ret = __ceph_alloc_sparse_ext_map(op, 1);
if (ret) {
ceph_osdc_put_request(req);
- ceph_release_page_vector(pages, num_pages);
break;
}
}
@@ -1766,7 +1762,6 @@ ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos,
ret = __ceph_alloc_sparse_ext_map(op, 1);
if (ret) {
ceph_osdc_put_request(req);
- ceph_release_page_vector(pages, num_pages);
break;
}
@@ -1998,6 +1993,7 @@ ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos,
}
+ ceph_databuf_release(dbuf);
if (ret != -EOLDSNAPC && written > 0) {
ret = written;
iocb->ki_pos = pos;
diff --git a/include/linux/ceph/osd_client.h b/include/linux/ceph/osd_client.h
index 1a1137787487..c26a7866695a 100644
--- a/include/linux/ceph/osd_client.h
+++ b/include/linux/ceph/osd_client.h
@@ -110,6 +110,7 @@ struct ceph_osd_data {
enum ceph_osd_data_type type;
struct ceph_databuf *dbuf;
struct iov_iter iter;
+ size_t length;
};
struct ceph_osd_req_op {
diff --git a/net/ceph/osd_client.c b/net/ceph/osd_client.c
index 6fb78ae14f03..95daf4cdb07b 100644
--- a/net/ceph/osd_client.c
+++ b/net/ceph/osd_client.c
@@ -127,6 +127,7 @@ static void ceph_osd_iter_init(struct ceph_osd_data *osd_data,
{
osd_data->type = CEPH_OSD_DATA_TYPE_ITER;
osd_data->iter = *iter;
+ osd_data->length = iter->count;
}
/*
@@ -239,19 +240,6 @@ void osd_req_op_cls_response_databuf(struct ceph_osd_request *osd_req,
}
EXPORT_SYMBOL(osd_req_op_cls_response_databuf);
-static u64 ceph_osd_data_length(struct ceph_osd_data *osd_data)
-{
- switch (osd_data->type) {
- case CEPH_OSD_DATA_TYPE_NONE:
- return 0;
- case CEPH_OSD_DATA_TYPE_ITER:
- return iov_iter_count(&osd_data->iter);
- default:
- WARN(true, "unrecognized data type %d\n", (int)osd_data->type);
- return 0;
- }
-}
-
static void ceph_osd_data_release(struct ceph_osd_data *osd_data)
{
ceph_osd_data_init(osd_data);
@@ -4475,7 +4463,7 @@ static void handle_watch_notify(struct ceph_osd_client *osdc,
if (data) {
if (lreq->reply) {
WARN_ON(data->type !=
- CEPH_MSG_DATA_PAGES);
+ CEPH_MSG_DATA_DATABUF);
*lreq->preply_pages = data->pages;
*lreq->preply_len = data->length;
data->own_pages = false;

Next message: Jijie Shao: "Re: [PATCH V2 drivers/perf: hisi:] drivers/perf: hisi: Update HiSilicon PMU maintainers"
Previous message: Xiubo Li: "Re: [RFC PATCH 00/18] ceph, rbd: Collapse all the I/O types down to something iov_iter-based"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]