Skip to contentSkip to conversation

libceph: add support for sparse reads

PATCH New Jeff Layton 1 participant 8 messages Mar 25, 2022

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

v5 cover libceph: add support for sparse reads

Yet another revised version of the sparse read patches. The first 4 and last patch are essentially the same as the ones I posted earlier this week, modulo a couple of small changes. Patches 5 and 6 add support for sparse reads to ms_mode=secure and ms_mode=legacy mounts.

This series is necessary for the fscrypt integration work. In fact, I was able to run the fscrypt series I posted earlier this week on top of this and it worked just fine.

Again, I'm mostly looking for feedback from Ilya at this point, but I'll probably go ahead and merge this into the testing branch later today (unless anyone pipes up with objections).

Jeff Layton (7):
libceph: add spinlock around osd->o_requests
libceph: define struct ceph_sparse_extent and add some helpers
libceph: add sparse read support to msgr2 crc state machine
libceph: add sparse read support to OSD client
libceph: support sparse reads on msgr2 secure codepath
libceph: add sparse read support to msgr1
ceph: add new mount option to enable sparse reads

 fs/ceph/addr.c                  |  18 +-
 fs/ceph/file.c                  |  51 +++++-
 fs/ceph/super.c                 |  16 +-
 fs/ceph/super.h                 |   8 +
 include/linux/ceph/messenger.h  |  33 ++++
 include/linux/ceph/osd_client.h |  71 +++++++-
 net/ceph/messenger.c            |   1 +
 net/ceph/messenger_v1.c         |  98 ++++++++++-
 net/ceph/messenger_v2.c         | 289 +++++++++++++++++++++++++++++---
 net/ceph/osd_client.c           | 266 ++++++++++++++++++++++++++++-
 10 files changed, 807 insertions(+), 44 deletions(-)
Signature · 1 line
2.35.1
#1 lore raw

PATCH v5 1/7 libceph: add spinlock around osd->o_requests

In a later patch, we're going to need to search for a request in the rbtree, but taking the o_mutex is inconvenient as we already hold the con mutex at the point where we need it.

Add a new spinlock that we take when inserting and erasing entries from the o_requests tree. Search of the rbtree can be done with either the mutex or the spinlock, but insertion and removal requires both.

---
 include/linux/ceph/osd_client.h | 8 +++++++-
 net/ceph/osd_client.c           | 5 +++++
 2 files changed, 12 insertions(+), 1 deletion(-)
include/linux/ceph/osd_client.h +7 -1
index 3431011f364d..3122c1a3205f 100644
@@ -29,7 +29,12 @@ typedef void (*ceph_osdc_callback_t)(struct ceph_osd_request *);
 
 #define CEPH_HOMELESS_OSD	-1
 
-/* a given osd we're communicating with */
+/*
+ * A given osd we're communicating with.
+ *
+ * Note that the o_requests tree can be searched while holding the "lock" mutex
+ * or the "o_requests_lock" spinlock. Insertion or removal requires both!
+ */
 struct ceph_osd {
 	refcount_t o_ref;
 	struct ceph_osd_client *o_osdc;
@@ -37,6 +42,7 @@ struct ceph_osd {
 	int o_incarnation;
 	struct rb_node o_node;
 	struct ceph_connection o_con;
+	spinlock_t o_requests_lock;
 	struct rb_root o_requests;
 	struct rb_root o_linger_requests;
 	struct rb_root o_backoff_mappings;
net/ceph/osd_client.c +5 -0
index 83eb97c94e83..17c792b32343 100644
@@ -1198,6 +1198,7 @@ static void osd_init(struct ceph_osd *osd)
 {
 	refcount_set(&osd->o_ref, 1);
 	RB_CLEAR_NODE(&osd->o_node);
+	spin_lock_init(&osd->o_requests_lock);
 	osd->o_requests = RB_ROOT;
 	osd->o_linger_requests = RB_ROOT;
 	osd->o_backoff_mappings = RB_ROOT;
@@ -1427,7 +1428,9 @@ static void link_request(struct ceph_osd *osd, struct ceph_osd_request *req)
 		atomic_inc(&osd->o_osdc->num_homeless);
 
 	get_osd(osd);
+	spin_lock(&osd->o_requests_lock);
 	insert_request(&osd->o_requests, req);
+	spin_unlock(&osd->o_requests_lock);
 	req->r_osd = osd;
 }
 
@@ -1439,7 +1442,9 @@ static void unlink_request(struct ceph_osd *osd, struct ceph_osd_request *req)
 	     req, req->r_tid);
 
 	req->r_osd = NULL;
+	spin_lock(&osd->o_requests_lock);
 	erase_request(&osd->o_requests, req);
+	spin_unlock(&osd->o_requests_lock);
 	put_osd(osd);
 
 	if (!osd_homeless(osd))
Signature · 1 line
2.35.1
#2 lore raw

PATCH v5 2/7 libceph: define struct ceph_sparse_extent and add some helpers

In reply to Jeff Layton

When the OSD sends back a sparse read reply, it contains an array of these structures. Define the structure and add a couple of helpers for dealing with them.

Also add a place in struct ceph_osd_req_op to store the extent buffer, and code to free it if it's populated when the req is torn down.

---
 include/linux/ceph/osd_client.h | 31 ++++++++++++++++++++++++++++++-
 net/ceph/osd_client.c           | 13 +++++++++++++
 2 files changed, 43 insertions(+), 1 deletion(-)
include/linux/ceph/osd_client.h +30 -1
index 3122c1a3205f..12f38295693d 100644
@@ -29,6 +29,17 @@ typedef void (*ceph_osdc_callback_t)(struct ceph_osd_request *);
 
 #define CEPH_HOMELESS_OSD	-1
 
+/*
+ * A single extent in a SPARSE_READ reply.
+ *
+ * Note that these come from the OSD as little-endian values. On BE arches,
+ * we convert them in-place after receipt.
+ */
+struct ceph_sparse_extent {
+	u64	off;
+	u64	len;
+} __packed;
+
 /*
  * A given osd we're communicating with.
  *
@@ -104,6 +115,8 @@ struct ceph_osd_req_op {
 			u64 offset, length;
 			u64 truncate_size;
 			u32 truncate_seq;
+			int sparse_ext_cnt;
+			struct ceph_sparse_extent *sparse_ext;
 			struct ceph_osd_data osd_data;
 		} extent;
 		struct {
@@ -507,6 +520,8 @@ extern struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *,
 				      u32 truncate_seq, u64 truncate_size,
 				      bool use_mempool);
 
+int ceph_alloc_sparse_ext_map(struct ceph_osd_req_op *op, int cnt);
+
 extern void ceph_osdc_get_request(struct ceph_osd_request *req);
 extern void ceph_osdc_put_request(struct ceph_osd_request *req);
 
@@ -562,5 +577,19 @@ int ceph_osdc_list_watchers(struct ceph_osd_client *osdc,
 			    struct ceph_object_locator *oloc,
 			    struct ceph_watch_item **watchers,
 			    u32 *num_watchers);
-#endif
 
+/* Find offset into the buffer of the end of the extent map */
+static inline u64 ceph_sparse_ext_map_end(struct ceph_osd_req_op *op)
+{
+	struct ceph_sparse_extent *ext;
+
+	/* No extents? No data */
+	if (op->extent.sparse_ext_cnt == 0)
+		return 0;
+
+	ext = &op->extent.sparse_ext[op->extent.sparse_ext_cnt - 1];
+
+	return ext->off + ext->len - op->extent.offset;
+}
+
+#endif
net/ceph/osd_client.c +13 -0
index 17c792b32343..7fc96c98c17b 100644
@@ -378,6 +378,7 @@ static void osd_req_op_data_release(struct ceph_osd_request *osd_req,
 	case CEPH_OSD_OP_READ:
 	case CEPH_OSD_OP_WRITE:
 	case CEPH_OSD_OP_WRITEFULL:
+		kfree(op->extent.sparse_ext);
 		ceph_osd_data_release(&op->extent.osd_data);
 		break;
 	case CEPH_OSD_OP_CALL:
@@ -1141,6 +1142,18 @@ struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
 }
 EXPORT_SYMBOL(ceph_osdc_new_request);
 
+int ceph_alloc_sparse_ext_map(struct ceph_osd_req_op *op, int cnt)
+{
+	op->extent.sparse_ext_cnt = cnt;
+	op->extent.sparse_ext = kmalloc_array(cnt,
+					      sizeof(*op->extent.sparse_ext),
+					      GFP_NOFS);
+	if (!op->extent.sparse_ext)
+		return -ENOMEM;
+	return 0;
+}
+EXPORT_SYMBOL(ceph_alloc_sparse_ext_map);
+
 /*
  * We keep osd requests in an rbtree, sorted by ->r_tid.
  */
Signature · 1 line
2.35.1
#3 lore raw

PATCH v5 3/7 libceph: add sparse read support to msgr2 crc state machine

In reply to Jeff Layton

Add support for a new sparse_read ceph_connection operation. The idea is that the client driver can define this operation use it to do special handling for incoming reads.

The alloc_msg routine will look at the request and determine whether the reply is expected to be sparse. If it is, then we'll dispatch to a different set of state machine states that will repeatedly call the driver's sparse_read op to get length and placement info for reading the extent map, and the extents themselves.

This necessitates adding some new field to some other structs:

- The msg gets a new bool to track whether it's a sparse_read request.

- A new field is added to the cursor to track the amount remaining in the current extent. This is used to cap the read from the socket into the msg_data

- Handing a revoke with all of this is particularly difficult, so I've added a new data_len_remain field to the v2 connection info, and then use that to skip that much on a revoke. We may want to expand the use of that to the normal read path as well, just for consistency's sake.

---
 include/linux/ceph/messenger.h |  29 ++++++
 net/ceph/messenger.c           |   1 +
 net/ceph/messenger_v2.c        | 169 +++++++++++++++++++++++++++++++--
 3 files changed, 190 insertions(+), 9 deletions(-)
include/linux/ceph/messenger.h +29 -0
index e7f2fb2fc207..25213eb1d348 100644
@@ -17,6 +17,7 @@
 
 struct ceph_msg;
 struct ceph_connection;
+struct ceph_msg_data_cursor;
 
 /*
  * Ceph defines these callbacks for handling connection events.
@@ -70,6 +71,31 @@ struct ceph_connection_operations {
 				      int used_proto, int result,
 				      const int *allowed_protos, int proto_cnt,
 				      const int *allowed_modes, int mode_cnt);
+
+	/**
+	 * sparse_read: read sparse data
+	 * @con: connection we're reading from
+	 * @cursor: data cursor for reading extents
+	 * @len: len of the data that msgr should read
+	 * @buf: optional buffer to read into
+	 *
+	 * This should be called more than once, each time setting up to
+	 * receive an extent into the current cursor position, and zeroing
+	 * the holes between them.
+	 *
+	 * Returns 1 if there is more data to be read, 0 if reading is
+	 * complete, or -errno if there was an error.
+	 *
+	 * If @buf is set on a 1 return, then the data should be read into
+	 * the provided buffer. Otherwise, it should be read into the cursor.
+	 *
+	 * The sparse read operation is expected to initialize the cursor
+	 * with a length covering up to the end of the last extent.
+	 */
+	int (*sparse_read)(struct ceph_connection *con,
+			   struct ceph_msg_data_cursor *cursor,
+			   u64 *len, char **buf);
+
 };
 
 /* use format string %s%lld */
@@ -207,6 +233,7 @@ struct ceph_msg_data_cursor {
 
 	struct ceph_msg_data	*data;		/* current data item */
 	size_t			resid;		/* bytes not yet consumed */
+	int			sr_resid;	/* residual sparse_read len */
 	bool			last_piece;	/* current is last piece */
 	bool			need_crc;	/* crc update needed */
 	union {
@@ -252,6 +279,7 @@ struct ceph_msg {
 	struct kref kref;
 	bool more_to_follow;
 	bool needs_out_seq;
+	bool sparse_read;
 	int front_alloc_len;
 
 	struct ceph_msgpool *pool;
@@ -396,6 +424,7 @@ struct ceph_connection_v2_info {
 
 	void *conn_bufs[16];
 	int conn_buf_cnt;
+	int data_len_remain;
 
 	struct kvec in_sign_kvecs[8];
 	struct kvec out_sign_kvecs[8];
net/ceph/messenger.c +1 -0
index d3bb656308b4..bf4e7f5751ee 100644
@@ -1034,6 +1034,7 @@ void ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor,
 
 	cursor->total_resid = length;
 	cursor->data = msg->data;
+	cursor->sr_resid = 0;
 
 	__ceph_msg_data_cursor_init(cursor);
 }
net/ceph/messenger_v2.c +160 -9

211 lines not on this page · raw message

Signature · 1 line
2.35.1
#4 lore raw

PATCH v5 4/7 libceph: add sparse read support to OSD client

In reply to Jeff Layton

Have get_reply check for the presence of sparse read ops in the request and set the sparse_read boolean in the msg. That will queue the messenger layer to use the sparse read codepath instead of the normal data receive.

Add a new sparse_read operation for the OSD client, driven by its own state machine. The messenger will repeatedly call the sparse_read operation, and it will pass back the necessary info to set up to read the next extent of data, while zero-filling the sparse regions.

The state machine will stop at the end of the last extent, and will attach the extent map buffer to the ceph_osd_req_op so that the caller can use it.

---
 include/linux/ceph/osd_client.h |  32 +++++
 net/ceph/osd_client.c           | 248 +++++++++++++++++++++++++++++++-
 2 files changed, 276 insertions(+), 4 deletions(-)
include/linux/ceph/osd_client.h +32 -0
index 12f38295693d..cf663423d51f 100644
@@ -40,6 +40,36 @@ struct ceph_sparse_extent {
 	u64	len;
 } __packed;
 
+/* Sparse read state machine state values */
+enum ceph_sparse_read_state {
+	CEPH_SPARSE_READ_HDR	= 0,
+	CEPH_SPARSE_READ_EXTENTS,
+	CEPH_SPARSE_READ_DATA_LEN,
+	CEPH_SPARSE_READ_DATA,
+};
+
+/*
+ * A SPARSE_READ reply is a 32-bit count of extents, followed by an array of
+ * 64-bit offset/length pairs, and then all of the actual file data
+ * concatenated after it (sans holes).
+ *
+ * Unfortunately, we don't know how long the extent array is until we've
+ * started reading the data section of the reply. The caller should send down
+ * a destination buffer for the array, but we'll alloc one if it's too small
+ * or if the caller doesn't.
+ */
+struct ceph_sparse_read {
+	enum ceph_sparse_read_state	sr_state;	/* state machine state */
+	u64				sr_req_off;	/* orig request offset */
+	u64				sr_req_len;	/* orig request length */
+	u64				sr_pos;		/* current pos in buffer */
+	int				sr_index;	/* current extent index */
+	__le32				sr_datalen;	/* length of actual data */
+	u32				sr_count;	/* extent count in reply */
+	int				sr_ext_len;	/* length of extent array */
+	struct ceph_sparse_extent	*sr_extent;	/* extent array */
+};
+
 /*
  * A given osd we're communicating with.
  *
@@ -48,6 +78,7 @@ struct ceph_sparse_extent {
  */
 struct ceph_osd {
 	refcount_t o_ref;
+	int o_sparse_op_idx;
 	struct ceph_osd_client *o_osdc;
 	int o_osd;
 	int o_incarnation;
@@ -63,6 +94,7 @@ struct ceph_osd {
 	unsigned long lru_ttl;
 	struct list_head o_keepalive_item;
 	struct mutex lock;
+	struct ceph_sparse_read	o_sparse_read;
 };
 
 #define CEPH_OSD_SLAB_OPS	2
net/ceph/osd_client.c +244 -4

342 lines not on this page · raw message

Signature · 1 line
2.35.1
#5 lore raw

PATCH v5 5/7 libceph: support sparse reads on msgr2 secure codepath

In reply to Jeff Layton

Add a new init_sgs_pages helper that populates the scatterlist from an arbitrary point in an array of pages.

Change setup_message_sgs to take an optional pointer to an array of pages. If that's set, then the scatterlist will be set using that array instead of the cursor.

When given a sparse read on a secure connection, decrypt the data in-place rather than into the final destination, by passing it the in_enc_pages array.

After decrypting, run the sparse_read state machine in a loop, copying data from the decrypted pages until it's complete.

---
 net/ceph/messenger_v2.c | 120 ++++++++++++++++++++++++++++++++++++----
 1 file changed, 110 insertions(+), 10 deletions(-)
net/ceph/messenger_v2.c +110 -10
index 4fc115fd63f3..c9f6b67fcdda 100644
@@ -963,12 +963,48 @@ static void init_sgs_cursor(struct scatterlist **sg,
 	}
 }
 
+/**
+ * init_sgs_pages: set up scatterlist on an array of page pointers
+ * @sg: 	scatterlist to populate
+ * @pages: 	pointer to page array
+ * @dpos:	position in the array to start (bytes)
+ * @dlen:	len to add to sg (bytes)
+ * @pad:	pointer to pad destination (if any)
+ *
+ * Populate the scatterlist from the page array, starting at an arbitrary
+ * byte in the array and running for a specified length.
+ */
+static void init_sgs_pages(struct scatterlist **sg, struct page **pages,
+			     int dpos, int dlen, u8 *pad)
+{
+	int idx = dpos >> PAGE_SHIFT;
+	int off = offset_in_page(dpos);
+	int resid = dlen;
+
+	do {
+		int len = min(resid, (int)PAGE_SIZE - off);
+
+		sg_set_page(*sg, pages[idx], len, off);
+		*sg = sg_next(*sg);
+		off = 0;
+		++idx;
+		resid -= len;
+	} while (resid);
+
+	if (need_padding(dlen)) {
+		sg_set_buf(*sg, pad, padding_len(dlen));
+		*sg = sg_next(*sg);
+	}
+}
+
 static int setup_message_sgs(struct sg_table *sgt, struct ceph_msg *msg,
 			     u8 *front_pad, u8 *middle_pad, u8 *data_pad,
-			     void *epilogue, bool add_tag)
+			     void *epilogue, struct page **pages, int dpos,
+			     bool add_tag)
 {
 	struct ceph_msg_data_cursor cursor;
 	struct scatterlist *cur_sg;
+	int dlen = data_len(msg);
 	int sg_cnt;
 	int ret;
 
@@ -982,9 +1018,15 @@ static int setup_message_sgs(struct sg_table *sgt, struct ceph_msg *msg,
 	if (middle_len(msg))
 		sg_cnt += calc_sg_cnt(msg->middle->vec.iov_base,
 				      middle_len(msg));
-	if (data_len(msg)) {
-		ceph_msg_data_cursor_init(&cursor, msg, data_len(msg));
-		sg_cnt += calc_sg_cnt_cursor(&cursor);
+	if (dlen) {
+		if (pages) {
+			sg_cnt += calc_pages_for(dpos, dlen);
+			if (need_padding(dlen))
+				sg_cnt++;
+		} else {
+			ceph_msg_data_cursor_init(&cursor, msg, dlen);
+			sg_cnt += calc_sg_cnt_cursor(&cursor);
+		}
 	}
 
 	ret = sg_alloc_table(sgt, sg_cnt, GFP_NOIO);
@@ -998,9 +1040,13 @@ static int setup_message_sgs(struct sg_table *sgt, struct ceph_msg *msg,
 	if (middle_len(msg))
 		init_sgs(&cur_sg, msg->middle->vec.iov_base, middle_len(msg),
 			 middle_pad);
-	if (data_len(msg)) {
-		ceph_msg_data_cursor_init(&cursor, msg, data_len(msg));
-		init_sgs_cursor(&cur_sg, &cursor, data_pad);
+	if (dlen) {
+		if (pages) {
+			init_sgs_pages(&cur_sg, pages, dpos, dlen, data_pad);
+		} else {
+			ceph_msg_data_cursor_init(&cursor, msg, dlen);
+			init_sgs_cursor(&cur_sg, &cursor, data_pad);
+		}
 	}
 
 	WARN_ON(!sg_is_last(cur_sg));
@@ -1035,10 +1081,53 @@ static int decrypt_control_remainder(struct ceph_connection *con)
 			 padded_len(rem_len) + CEPH_GCM_TAG_LEN);
 }
 
+/* Process sparse read data that lives in a buffer */
+static int process_v2_sparse_read(struct ceph_connection *con, struct page **pages, int spos)
+{
+	struct ceph_msg_data_cursor *cursor = &con->v2.in_cursor;
+	int ret;
+
+	for (;;) {
+		u64 elen;
+		char *buf = NULL;
+
+		ret = con->ops->sparse_read(con, cursor, &elen, &buf);
+		if (ret <= 0)
+			return ret;
+
+		dout("%s: sparse_read return elen %llx buf %p\n", __func__, elen, buf);
+
+		do {
+			int idx = spos >> PAGE_SHIFT;
+			int soff = offset_in_page(spos);
+			struct page *spage = con->v2.in_enc_pages[idx];
+			int len = min_t(int, elen, PAGE_SIZE - soff);
+
+			if (buf) {
+				memcpy_from_page(buf, spage, soff, len);
+				buf += len;
+			} else {
+				struct bio_vec bv;
+
+				get_bvec_at(cursor, &bv);
+				len = min_t(int, len, bv.bv_len);
+				memcpy_page(bv.bv_page, bv.bv_offset,
+					    spage, soff, len);
+				ceph_msg_data_advance(cursor, len);
+			}
+			spos += len;
+			elen -= len;
+		} while (elen);
+	}
+}
+
 static int decrypt_tail(struct ceph_connection *con)
 {
 	struct sg_table enc_sgt = {};
 	struct sg_table sgt = {};
+	struct page **pages = NULL;
+	bool sparse = con->in_msg->sparse_read;
+	int dpos = 0;
 	int tail_len;
 	int ret;
 
@@ -1049,9 +1138,14 @@ static int decrypt_tail(struct ceph_connection *con)
 	if (ret)
 		goto out;
 
+	if (sparse) {
+		dpos = padded_len(front_len(con->in_msg) + padded_len(middle_len(con->in_msg)));
+		pages = con->v2.in_enc_pages;
+	}
+
 	ret = setup_message_sgs(&sgt, con->in_msg, FRONT_PAD(con->v2.in_buf),
-			MIDDLE_PAD(con->v2.in_buf), DATA_PAD(con->v2.in_buf),
-			con->v2.in_buf, true);
+				MIDDLE_PAD(con->v2.in_buf), DATA_PAD(con->v2.in_buf),
+				con->v2.in_buf, pages, dpos, true);
 	if (ret)
 		goto out;
 
@@ -1061,6 +1155,12 @@ static int decrypt_tail(struct ceph_connection *con)
 	if (ret)
 		goto out;
 
+	if (sparse && data_len(con->in_msg)) {
+		ret = process_v2_sparse_read(con, con->v2.in_enc_pages, dpos);
+		if (ret)
+			goto out;
+	}
+
 	WARN_ON(!con->v2.in_enc_page_cnt);
 	ceph_release_page_vector(con->v2.in_enc_pages,
 				 con->v2.in_enc_page_cnt);
@@ -1584,7 +1684,7 @@ static int prepare_message_secure(struct ceph_connection *con)
 
 	encode_epilogue_secure(con, false);
 	ret = setup_message_sgs(&sgt, con->out_msg, zerop, zerop, zerop,
-				&con->v2.out_epil, false);
+				&con->v2.out_epil, NULL, 0, false);
 	if (ret)
 		goto out;
 
Signature · 1 line
2.35.1
#6 lore raw

PATCH v5 6/7 libceph: add sparse read support to msgr1

In reply to Jeff Layton

Add 2 new fields to ceph_connection_v1_info to track the necessary info in sparse reads. Skip initializing the cursor for a sparse read.

Break out read_partial_message_section into a wrapper around a new read_partial_message_chunk function that doesn't zero out the crc first.

Add new helper functions to drive receiving into the destinations provided by the sparse_read state machine.

---
 include/linux/ceph/messenger.h |  4 ++
 net/ceph/messenger_v1.c        | 98 +++++++++++++++++++++++++++++++---
 2 files changed, 94 insertions(+), 8 deletions(-)
include/linux/ceph/messenger.h +4 -0
index 25213eb1d348..3697049e1465 100644
@@ -338,6 +338,10 @@ struct ceph_connection_v1_info {
 
 	int in_base_pos;     /* bytes read */
 
+	/* sparse reads */
+	struct kvec in_sr_kvec; /* current location to receive into */
+	u64 in_sr_len;		/* amount of data in this extent */
+
 	/* message in temps */
 	u8 in_tag;           /* protocol control byte */
 	struct ceph_msg_header in_hdr;
net/ceph/messenger_v1.c +90 -8

126 lines not on this page · raw message

Signature · 1 line
2.35.1
#7 lore raw

PATCH v5 7/7 ceph: add new mount option to enable sparse reads

In reply to Jeff Layton

Add a new mount option that has the client issue sparse reads instead of normal ones. The callers now preallocate an sparse extent buffer that the libceph receive code can populate and hand back after the operation completes.

After a successful sparse read, we can't use the req->r_result value to determine the amount of data "read", so instead we set the received length to be from the end of the last extent in the buffer. Any interstitial holes will have been filled by the receive code.

---
 fs/ceph/addr.c  | 18 +++++++++++++++--
 fs/ceph/file.c  | 51 +++++++++++++++++++++++++++++++++++++++++--------
 fs/ceph/super.c | 16 +++++++++++++++-
 fs/ceph/super.h |  8 ++++++++
 4 files changed, 82 insertions(+), 11 deletions(-)
fs/ceph/addr.c +16 -2
index 752c421c9922..15bc455bc87f 100644
@@ -220,8 +220,10 @@ static void finish_netfs_read(struct ceph_osd_request *req)
 	struct ceph_fs_client *fsc = ceph_inode_to_client(req->r_inode);
 	struct ceph_osd_data *osd_data = osd_req_op_extent_osd_data(req, 0);
 	struct netfs_read_subrequest *subreq = req->r_priv;
+	struct ceph_osd_req_op *op = &req->r_ops[0];
 	int num_pages;
 	int err = req->r_result;
+	bool sparse = (op->op == CEPH_OSD_OP_SPARSE_READ);
 
 	ceph_update_read_metrics(&fsc->mdsc->metric, req->r_start_latency,
 				 req->r_end_latency, osd_data->length, err);
@@ -230,7 +232,9 @@ static void finish_netfs_read(struct ceph_osd_request *req)
 	     subreq->len, i_size_read(req->r_inode));
 
 	/* no object means success but no data */
-	if (err == -ENOENT)
+	if (sparse && err >= 0)
+		err = ceph_sparse_ext_map_end(op);
+	else if (err == -ENOENT)
 		err = 0;
 	else if (err == -EBLOCKLISTED)
 		fsc->blocklisted = true;
@@ -311,13 +315,14 @@ static void ceph_netfs_issue_op(struct netfs_read_subrequest *subreq)
 	size_t page_off;
 	int err = 0;
 	u64 len = subreq->len;
+	bool sparse = ceph_test_mount_opt(fsc, SPARSEREAD);
 
 	if (ci->i_inline_version != CEPH_INLINE_NONE &&
 	    ceph_netfs_issue_op_inline(subreq))
 		return;
 
 	req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, vino, subreq->start, &len,
-			0, 1, CEPH_OSD_OP_READ,
+			0, 1, sparse ? CEPH_OSD_OP_SPARSE_READ : CEPH_OSD_OP_READ,
 			CEPH_OSD_FLAG_READ | fsc->client->osdc.client->options->read_from_replica,
 			NULL, ci->i_truncate_seq, ci->i_truncate_size, false);
 	if (IS_ERR(req)) {
@@ -326,6 +331,15 @@ static void ceph_netfs_issue_op(struct netfs_read_subrequest *subreq)
 		goto out;
 	}
 
+	if (sparse) {
+		err = ceph_alloc_sparse_ext_map(&req->r_ops[0],
+					CEPH_SPARSE_EXT_ARRAY_INITIAL);
+		if (err) {
+			ceph_osdc_put_request(req);
+			goto out;
+		}
+	}
+
 	dout("%s: pos=%llu orig_len=%zu len=%llu\n", __func__, subreq->start, subreq->len, len);
 	iov_iter_xarray(&iter, READ, &rreq->mapping->i_pages, subreq->start, len);
 	err = iov_iter_get_pages_alloc(&iter, &pages, len, &page_off);
fs/ceph/file.c +43 -8

132 lines not on this page · raw message

fs/ceph/super.c +15 -1
index 978463fa822c..a859921bbe96 100644
@@ -163,6 +163,7 @@ enum {
 	Opt_copyfrom,
 	Opt_wsync,
 	Opt_pagecache,
+	Opt_sparseread,
 };
 
 enum ceph_recover_session_mode {
@@ -205,6 +206,7 @@ static const struct fs_parameter_spec ceph_mount_parameters[] = {
 	fsparam_u32	("wsize",			Opt_wsize),
 	fsparam_flag_no	("wsync",			Opt_wsync),
 	fsparam_flag_no	("pagecache",			Opt_pagecache),
+	fsparam_flag_no	("sparseread",			Opt_sparseread),
 	{}
 };
 
@@ -574,6 +576,12 @@ static int ceph_parse_mount_param(struct fs_context *fc,
 		else
 			fsopt->flags &= ~CEPH_MOUNT_OPT_NOPAGECACHE;
 		break;
+	case Opt_sparseread:
+		if (result.negated)
+			fsopt->flags &= ~CEPH_MOUNT_OPT_SPARSEREAD;
+		else
+			fsopt->flags |= CEPH_MOUNT_OPT_SPARSEREAD;
+		break;
 	default:
 		BUG();
 	}
@@ -708,9 +716,10 @@ static int ceph_show_options(struct seq_file *m, struct dentry *root)
 
 	if (!(fsopt->flags & CEPH_MOUNT_OPT_ASYNC_DIROPS))
 		seq_puts(m, ",wsync");
-
 	if (fsopt->flags & CEPH_MOUNT_OPT_NOPAGECACHE)
 		seq_puts(m, ",nopagecache");
+	if (fsopt->flags & CEPH_MOUNT_OPT_SPARSEREAD)
+		seq_puts(m, ",sparseread");
 
 	if (fsopt->wsize != CEPH_MAX_WRITE_SIZE)
 		seq_printf(m, ",wsize=%u", fsopt->wsize);
@@ -1289,6 +1298,11 @@ static int ceph_reconfigure_fc(struct fs_context *fc)
 	else
 		ceph_clear_mount_opt(fsc, ASYNC_DIROPS);
 
+	if (fsopt->flags & CEPH_MOUNT_OPT_SPARSEREAD)
+		ceph_set_mount_opt(fsc, SPARSEREAD);
+	else
+		ceph_clear_mount_opt(fsc, SPARSEREAD);
+
 	if (strcmp_null(fsc->mount_options->mon_addr, fsopt->mon_addr)) {
 		kfree(fsc->mount_options->mon_addr);
 		fsc->mount_options->mon_addr = fsopt->mon_addr;
fs/ceph/super.h +8 -0
index 250aefecd628..4999207a5466 100644
@@ -43,6 +43,7 @@
 #define CEPH_MOUNT_OPT_NOCOPYFROM      (1<<14) /* don't use RADOS 'copy-from' op */
 #define CEPH_MOUNT_OPT_ASYNC_DIROPS    (1<<15) /* allow async directory ops */
 #define CEPH_MOUNT_OPT_NOPAGECACHE     (1<<16) /* bypass pagecache altogether */
+#define CEPH_MOUNT_OPT_SPARSEREAD      (1<<17) /* always do sparse reads */
 
 #define CEPH_MOUNT_OPT_DEFAULT			\
 	(CEPH_MOUNT_OPT_DCACHE |		\
@@ -75,6 +76,13 @@
 #define CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT      5  /* cap release delay */
 #define CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT     60  /* cap release delay */
 
+/*
+ * How big an extent array should we preallocate for a sparse read? This is
+ * just a starting value.  If we get more than this back from the OSD, the
+ * receiver will reallocate.
+ */
+#define CEPH_SPARSE_EXT_ARRAY_INITIAL	16
+
 struct ceph_mount_options {
 	unsigned int flags;
 
Signature · 1 line
2.35.1