From: David Howells <dhowells@redhat.com>
To: Viacheslav Dubeyko <slava@dubeyko.com>,
Alex Markuze <amarkuze@redhat.com>
Cc: David Howells <dhowells@redhat.com>,
Ilya Dryomov <idryomov@gmail.com>,
Jeff Layton <jlayton@kernel.org>,
Dongsheng Yang <dongsheng.yang@easystack.cn>,
ceph-devel@vger.kernel.org, linux-fsdevel@vger.kernel.org,
linux-block@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: [RFC PATCH 32/35] netfs: Add some more RMW support for ceph
Date: Thu, 13 Mar 2025 23:33:24 +0000 [thread overview]
Message-ID: <20250313233341.1675324-33-dhowells@redhat.com> (raw)
In-Reply-To: <20250313233341.1675324-1-dhowells@redhat.com>
Add some support for RMW in ceph:
(1) Add netfs_unbuffered_read_from_inode() to allow reading from an inode
without having a file pointer so that truncate can modify a
now-partial tail block of a content-encrypted file.
This takes an additional argument to cause it to fail or give a short
read if a hole is encountered. This is noted on the request with
NETFS_RREQ_NO_READ_HOLE for the filesystem to pick up.
(2) Set NETFS_RREQ_RMW when doing an RMW as part of a request.
(3) Provide a ->rmw_read_done() op for netfslib to tell the filesystem
that it has completed the read required for RMW.
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Jeff Layton <jlayton@kernel.org>
cc: Viacheslav Dubeyko <slava@dubeyko.com>
cc: Alex Markuze <amarkuze@redhat.com>
cc: Ilya Dryomov <idryomov@gmail.com>
cc: ceph-devel@vger.kernel.org
cc: linux-fsdevel@vger.kernel.org
---
fs/netfs/direct_read.c | 75 ++++++++++++++++++++++++++++++++++++
fs/netfs/direct_write.c | 1 +
fs/netfs/main.c | 1 +
fs/netfs/objects.c | 1 +
fs/netfs/read_collect.c | 2 +
fs/netfs/write_retry.c | 3 ++
include/linux/netfs.h | 7 ++++
include/trace/events/netfs.h | 3 ++
8 files changed, 93 insertions(+)
diff --git a/fs/netfs/direct_read.c b/fs/netfs/direct_read.c
index 5e4bd1e5a378..4061f934dfe6 100644
--- a/fs/netfs/direct_read.c
+++ b/fs/netfs/direct_read.c
@@ -373,3 +373,78 @@ ssize_t netfs_unbuffered_read_iter(struct kiocb *iocb, struct iov_iter *iter)
return ret;
}
EXPORT_SYMBOL(netfs_unbuffered_read_iter);
+
+/**
+ * netfs_unbuffered_read_from_inode - Perform an unbuffered sync I/O read
+ * @inode: The inode being accessed
+ * @pos: The file position to read from
+ * @iter: The output buffer (also specifies read length)
+ * @nohole: True to return short/ENODATA if hole encountered
+ *
+ * Perform a synchronous unbuffered I/O from the inode to the output buffer.
+ * No use is made of the pagecache. The output buffer must be suitably aligned
+ * if content encryption is to be used. If @nohole is true then the read will
+ * stop short if a hole is encountered and return -ENODATA if the read begins
+ * with a hole.
+ *
+ * The caller must hold any appropriate locks.
+ */
+ssize_t netfs_unbuffered_read_from_inode(struct inode *inode, loff_t pos,
+ struct iov_iter *iter, bool nohole)
+{
+ struct netfs_io_request *rreq;
+ ssize_t ret;
+ size_t orig_count = iov_iter_count(iter);
+
+ _enter("");
+
+ if (WARN_ON(user_backed_iter(iter)))
+ return -EIO;
+
+ if (!orig_count)
+ return 0; /* Don't update atime */
+
+ ret = filemap_write_and_wait_range(inode->i_mapping, pos, orig_count);
+ if (ret < 0)
+ return ret;
+ inode_update_time(inode, S_ATIME);
+
+ rreq = netfs_alloc_request(inode->i_mapping, NULL, pos, orig_count,
+ NULL, NETFS_UNBUFFERED_READ);
+ if (IS_ERR(rreq))
+ return PTR_ERR(rreq);
+
+ ret = -EIO;
+ if (test_bit(NETFS_RREQ_CONTENT_ENCRYPTION, &rreq->flags) &&
+ WARN_ON(!netfs_is_crypto_aligned(rreq, iter)))
+ goto out;
+
+ netfs_stat(&netfs_n_rh_dio_read);
+ trace_netfs_read(rreq, rreq->start, rreq->len,
+ netfs_read_trace_unbuffered_read_from_inode);
+
+ rreq->buffer.iter = *iter;
+ rreq->len = orig_count;
+ rreq->direct_bv_unpin = false;
+ iov_iter_advance(iter, orig_count);
+
+ if (nohole)
+ __set_bit(NETFS_RREQ_NO_READ_HOLE, &rreq->flags);
+
+ /* We're going to do the crypto in place in the destination buffer. */
+ if (test_bit(NETFS_RREQ_CONTENT_ENCRYPTION, &rreq->flags))
+ __set_bit(NETFS_RREQ_CRYPT_IN_PLACE, &rreq->flags);
+
+ ret = netfs_dispatch_unbuffered_reads(rreq);
+
+ if (!rreq->submitted) {
+ netfs_put_request(rreq, false, netfs_rreq_trace_put_no_submit);
+ goto out;
+ }
+
+ ret = netfs_wait_for_read(rreq);
+out:
+ netfs_put_request(rreq, false, netfs_rreq_trace_put_return);
+ return ret;
+}
+EXPORT_SYMBOL(netfs_unbuffered_read_from_inode);
diff --git a/fs/netfs/direct_write.c b/fs/netfs/direct_write.c
index 83c5c06c4710..a99722f90c71 100644
--- a/fs/netfs/direct_write.c
+++ b/fs/netfs/direct_write.c
@@ -145,6 +145,7 @@ static ssize_t netfs_write_through_bounce_buffer(struct netfs_io_request *wreq,
wreq->start = gstart;
wreq->len = gend - gstart;
+ __set_bit(NETFS_RREQ_RMW, &ictx->flags);
if (gstart >= end) {
/* At or after EOF, nothing to read. */
} else {
diff --git a/fs/netfs/main.c b/fs/netfs/main.c
index 07f8cffbda8c..0900dea53e4a 100644
--- a/fs/netfs/main.c
+++ b/fs/netfs/main.c
@@ -39,6 +39,7 @@ static const char *netfs_origins[nr__netfs_io_origin] = {
[NETFS_READ_GAPS] = "RG",
[NETFS_READ_SINGLE] = "R1",
[NETFS_READ_FOR_WRITE] = "RW",
+ [NETFS_UNBUFFERED_READ] = "UR",
[NETFS_DIO_READ] = "DR",
[NETFS_WRITEBACK] = "WB",
[NETFS_WRITEBACK_SINGLE] = "W1",
diff --git a/fs/netfs/objects.c b/fs/netfs/objects.c
index 4606e830c116..958c4d460d07 100644
--- a/fs/netfs/objects.c
+++ b/fs/netfs/objects.c
@@ -60,6 +60,7 @@ struct netfs_io_request *netfs_alloc_request(struct address_space *mapping,
origin == NETFS_READ_GAPS ||
origin == NETFS_READ_SINGLE ||
origin == NETFS_READ_FOR_WRITE ||
+ origin == NETFS_UNBUFFERED_READ ||
origin == NETFS_DIO_READ) {
INIT_WORK(&rreq->work, netfs_read_collection_worker);
rreq->io_streams[0].avail = true;
diff --git a/fs/netfs/read_collect.c b/fs/netfs/read_collect.c
index 0a0bff90ca9e..013a90738dcd 100644
--- a/fs/netfs/read_collect.c
+++ b/fs/netfs/read_collect.c
@@ -462,6 +462,7 @@ static void netfs_read_collection(struct netfs_io_request *rreq)
//netfs_rreq_is_still_valid(rreq);
switch (rreq->origin) {
+ case NETFS_UNBUFFERED_READ:
case NETFS_DIO_READ:
case NETFS_READ_GAPS:
case NETFS_RMW_READ:
@@ -681,6 +682,7 @@ ssize_t netfs_wait_for_read(struct netfs_io_request *rreq)
if (ret == 0) {
ret = rreq->transferred;
switch (rreq->origin) {
+ case NETFS_UNBUFFERED_READ:
case NETFS_DIO_READ:
case NETFS_READ_SINGLE:
ret = rreq->transferred;
diff --git a/fs/netfs/write_retry.c b/fs/netfs/write_retry.c
index f727b48e2bfe..9e4e79d5a403 100644
--- a/fs/netfs/write_retry.c
+++ b/fs/netfs/write_retry.c
@@ -386,6 +386,9 @@ ssize_t netfs_rmw_read(struct netfs_io_request *wreq, struct file *file,
ret = 0;
}
+ if (ret == 0 && rreq->netfs_ops->rmw_read_done)
+ rreq->netfs_ops->rmw_read_done(wreq, rreq);
+
error:
netfs_put_request(rreq, false, netfs_rreq_trace_put_return);
return ret;
diff --git a/include/linux/netfs.h b/include/linux/netfs.h
index 9d17d4bd9753..4049c985b9b4 100644
--- a/include/linux/netfs.h
+++ b/include/linux/netfs.h
@@ -220,6 +220,7 @@ enum netfs_io_origin {
NETFS_READ_GAPS, /* This read is a synchronous read to fill gaps */
NETFS_READ_SINGLE, /* This read should be treated as a single object */
NETFS_READ_FOR_WRITE, /* This read is to prepare a write */
+ NETFS_UNBUFFERED_READ, /* This is an unbuffered I/O read */
NETFS_DIO_READ, /* This is a direct I/O read */
NETFS_WRITEBACK, /* This write was triggered by writepages */
NETFS_WRITEBACK_SINGLE, /* This monolithic write was triggered by writepages */
@@ -308,6 +309,9 @@ struct netfs_io_request {
#define NETFS_RREQ_CONTENT_ENCRYPTION 16 /* Content encryption is in use */
#define NETFS_RREQ_CRYPT_IN_PLACE 17 /* Do decryption in place */
#define NETFS_RREQ_PUT_RMW_TAIL 18 /* Need to put ->rmw_tail */
+#define NETFS_RREQ_RMW 19 /* Performing RMW cycle */
+#define NETFS_RREQ_REPEAT_RMW 20 /* Need to perform an RMW cycle */
+#define NETFS_RREQ_NO_READ_HOLE 21 /* Give short read/error if hole encountered */
#define NETFS_RREQ_USE_PGPRIV2 31 /* [DEPRECATED] Use PG_private_2 to mark
* write to cache on read */
const struct netfs_request_ops *netfs_ops;
@@ -336,6 +340,7 @@ struct netfs_request_ops {
/* Modification handling */
void (*update_i_size)(struct inode *inode, loff_t i_size);
void (*post_modify)(struct inode *inode, void *fs_priv);
+ void (*rmw_read_done)(struct netfs_io_request *wreq, struct netfs_io_request *rreq);
/* Write request handling */
void (*begin_writeback)(struct netfs_io_request *wreq);
@@ -432,6 +437,8 @@ ssize_t netfs_unbuffered_read_iter_locked(struct kiocb *iocb, struct iov_iter *i
ssize_t netfs_unbuffered_read_iter(struct kiocb *iocb, struct iov_iter *iter);
ssize_t netfs_buffered_read_iter(struct kiocb *iocb, struct iov_iter *iter);
ssize_t netfs_file_read_iter(struct kiocb *iocb, struct iov_iter *iter);
+ssize_t netfs_unbuffered_read_from_inode(struct inode *inode, loff_t pos,
+ struct iov_iter *iter, bool nohole);
/* High-level write API */
ssize_t netfs_perform_write(struct kiocb *iocb, struct iov_iter *iter,
diff --git a/include/trace/events/netfs.h b/include/trace/events/netfs.h
index 74af82d773bd..9254c6f0e604 100644
--- a/include/trace/events/netfs.h
+++ b/include/trace/events/netfs.h
@@ -23,6 +23,7 @@
EM(netfs_read_trace_read_gaps, "READ-GAPS") \
EM(netfs_read_trace_read_single, "READ-SNGL") \
EM(netfs_read_trace_prefetch_for_write, "PREFETCHW") \
+ EM(netfs_read_trace_unbuffered_read_from_inode, "READ-INOD") \
E_(netfs_read_trace_write_begin, "WRITEBEGN")
#define netfs_write_traces \
@@ -38,6 +39,7 @@
EM(NETFS_READ_GAPS, "RG") \
EM(NETFS_READ_SINGLE, "R1") \
EM(NETFS_READ_FOR_WRITE, "RW") \
+ EM(NETFS_UNBUFFERED_READ, "UR") \
EM(NETFS_DIO_READ, "DR") \
EM(NETFS_WRITEBACK, "WB") \
EM(NETFS_WRITEBACK_SINGLE, "W1") \
@@ -104,6 +106,7 @@
EM(netfs_sreq_trace_io_progress, "IO ") \
EM(netfs_sreq_trace_limited, "LIMIT") \
EM(netfs_sreq_trace_need_clear, "N-CLR") \
+ EM(netfs_sreq_trace_need_rmw, "N-RMW") \
EM(netfs_sreq_trace_partial_read, "PARTR") \
EM(netfs_sreq_trace_need_retry, "ND-RT") \
EM(netfs_sreq_trace_pending, "PEND ") \
next prev parent reply other threads:[~2025-03-13 23:35 UTC|newest]
Thread overview: 72+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-03-13 23:32 [RFC PATCH 00/35] ceph, rbd, netfs: Make ceph fully use netfslib David Howells
2025-03-13 23:32 ` [RFC PATCH 01/35] ceph: Fix incorrect flush end position calculation David Howells
2025-03-13 23:32 ` [RFC PATCH 02/35] libceph: Rename alignment to offset David Howells
2025-03-14 19:04 ` Viacheslav Dubeyko
2025-03-14 20:01 ` David Howells
2025-03-13 23:32 ` [RFC PATCH 03/35] libceph: Add a new data container type, ceph_databuf David Howells
2025-03-14 20:06 ` Viacheslav Dubeyko
2025-03-17 11:27 ` David Howells
2025-03-13 23:32 ` [RFC PATCH 04/35] ceph: Convert ceph_mds_request::r_pagelist to a databuf David Howells
2025-03-14 22:27 ` slava
2025-03-17 11:52 ` David Howells
2025-03-20 20:34 ` Viacheslav Dubeyko
2025-03-20 22:01 ` David Howells
2025-03-13 23:32 ` [RFC PATCH 05/35] libceph: Add functions to add ceph_databufs to requests David Howells
2025-03-13 23:32 ` [RFC PATCH 06/35] rbd: Use ceph_databuf for rbd_obj_read_sync() David Howells
2025-03-17 19:08 ` Viacheslav Dubeyko
2025-04-11 13:48 ` David Howells
2025-03-13 23:32 ` [RFC PATCH 07/35] libceph: Change ceph_osdc_call()'s reply to a ceph_databuf David Howells
2025-03-17 19:41 ` Viacheslav Dubeyko
2025-03-17 22:12 ` David Howells
2025-03-13 23:33 ` [RFC PATCH 08/35] libceph: Unexport osd_req_op_cls_request_data_pages() David Howells
2025-03-13 23:33 ` [RFC PATCH 09/35] libceph: Remove osd_req_op_cls_response_data_pages() David Howells
2025-03-13 23:33 ` [RFC PATCH 10/35] libceph: Convert notify_id_pages to a ceph_databuf David Howells
2025-03-13 23:33 ` [RFC PATCH 11/35] ceph: Use ceph_databuf in DIO David Howells
2025-03-17 20:03 ` Viacheslav Dubeyko
2025-03-17 22:26 ` David Howells
2025-03-13 23:33 ` [RFC PATCH 12/35] libceph: Bypass the messenger-v1 Tx loop for databuf/iter data blobs David Howells
2025-03-13 23:33 ` [RFC PATCH 13/35] rbd: Switch from using bvec_iter to iov_iter David Howells
2025-03-18 19:38 ` Viacheslav Dubeyko
2025-03-18 22:13 ` David Howells
2025-03-13 23:33 ` [RFC PATCH 14/35] libceph: Remove bvec and bio data container types David Howells
2025-03-13 23:33 ` [RFC PATCH 15/35] libceph: Make osd_req_op_cls_init() use a ceph_databuf and map it David Howells
2025-03-13 23:33 ` [RFC PATCH 16/35] libceph: Convert req_page of ceph_osdc_call() to ceph_databuf David Howells
2025-03-13 23:33 ` [RFC PATCH 17/35] libceph, rbd: Use ceph_databuf encoding start/stop David Howells
2025-03-18 19:59 ` Viacheslav Dubeyko
2025-03-18 22:19 ` David Howells
2025-03-20 21:45 ` Viacheslav Dubeyko
2025-03-13 23:33 ` [RFC PATCH 18/35] libceph, rbd: Convert some page arrays to ceph_databuf David Howells
2025-03-18 20:02 ` Viacheslav Dubeyko
2025-03-18 22:25 ` David Howells
2025-03-13 23:33 ` [RFC PATCH 19/35] libceph, ceph: Convert users of ceph_pagelist " David Howells
2025-03-18 20:09 ` Viacheslav Dubeyko
2025-03-18 22:27 ` David Howells
2025-03-13 23:33 ` [RFC PATCH 20/35] libceph: Remove ceph_pagelist David Howells
2025-03-13 23:33 ` [RFC PATCH 21/35] libceph: Make notify code use ceph_databuf_enc_start/stop David Howells
2025-03-18 20:12 ` Viacheslav Dubeyko
2025-03-18 22:36 ` David Howells
2025-03-13 23:33 ` [RFC PATCH 22/35] libceph, rbd: Convert ceph_osdc_notify() reply to ceph_databuf David Howells
2025-03-19 0:08 ` Viacheslav Dubeyko
2025-03-20 14:44 ` David Howells
2025-03-13 23:33 ` [RFC PATCH 23/35] rbd: Use ceph_databuf_enc_start/stop() David Howells
2025-03-19 0:32 ` Viacheslav Dubeyko
2025-03-20 14:59 ` Why use plain numbers and totals rather than predef'd constants for RPC sizes? David Howells
2025-03-20 21:48 ` Viacheslav Dubeyko
2025-03-13 23:33 ` [RFC PATCH 24/35] ceph: Make ceph_calc_file_object_mapping() return size as size_t David Howells
2025-03-13 23:33 ` [RFC PATCH 25/35] ceph: Wrap POSIX_FADV_WILLNEED to get caps David Howells
2025-03-13 23:33 ` [RFC PATCH 26/35] ceph: Kill ceph_rw_context David Howells
2025-03-13 23:33 ` [RFC PATCH 27/35] netfs: Pass extra write context to write functions David Howells
2025-03-13 23:33 ` [RFC PATCH 28/35] netfs: Adjust group handling David Howells
2025-03-19 18:57 ` Viacheslav Dubeyko
2025-03-20 15:22 ` David Howells
2025-03-13 23:33 ` [RFC PATCH 29/35] netfs: Allow fs-private data to be handed through to request alloc David Howells
2025-03-13 23:33 ` [RFC PATCH 30/35] netfs: Make netfs_page_mkwrite() use folio_mkwrite_check_truncate() David Howells
2025-03-13 23:33 ` [RFC PATCH 31/35] netfs: Fix netfs_unbuffered_read() to return ssize_t rather than int David Howells
2025-03-13 23:33 ` David Howells [this message]
2025-03-19 19:14 ` [RFC PATCH 32/35] netfs: Add some more RMW support for ceph Viacheslav Dubeyko
2025-03-20 15:25 ` David Howells
2025-03-13 23:33 ` [RFC PATCH 33/35] ceph: Use netfslib [INCOMPLETE] David Howells
2025-03-19 19:54 ` Viacheslav Dubeyko
2025-03-20 15:38 ` David Howells
2025-03-13 23:33 ` [RFC PATCH 34/35] ceph: Enable multipage folios for ceph files David Howells
2025-03-13 23:33 ` [RFC PATCH 35/35] ceph: Remove old I/O API bits David Howells
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20250313233341.1675324-33-dhowells@redhat.com \
--to=dhowells@redhat.com \
--cc=amarkuze@redhat.com \
--cc=ceph-devel@vger.kernel.org \
--cc=dongsheng.yang@easystack.cn \
--cc=idryomov@gmail.com \
--cc=jlayton@kernel.org \
--cc=linux-block@vger.kernel.org \
--cc=linux-fsdevel@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=slava@dubeyko.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.