netfs: Speed up buffered reading

Improve the efficiency of buffered reads in a number of ways:

 (1) Overhaul the algorithm in general so that it's a lot more compact and
     split the read submission code between buffered and unbuffered
     versions.  The unbuffered version can be vastly simplified.

 (2) Read-result collection is handed off to a work queue rather than being
     done in the I/O thread.  Multiple subrequests can be processes
     simultaneously.

 (3) When a subrequest is collected, any folios it fully spans are
     collected and "spare" data on either side is donated to either the
     previous or the next subrequest in the sequence.

Notes:

 (*) Readahead expansion is massively slows down fio, presumably because it
     causes a load of extra allocations, both folio and xarray, up front
     before RPC requests can be transmitted.

 (*) RDMA with cifs does appear to work, both with SIW and RXE.

 (*) PG_private_2-based reading and copy-to-cache is split out into its own
     file and altered to use folio_queue.  Note that the copy to the cache
     now creates a new write transaction against the cache and adds the
     folios to be copied into it.  This allows it to use part of the
     writeback I/O code.

Signed-off-by: David Howells <dhowells@redhat.com>
cc: Jeff Layton <jlayton@kernel.org>
cc: netfs@lists.linux.dev
cc: linux-fsdevel@vger.kernel.org
Link: https://lore.kernel.org/r/20240814203850.2240469-20-dhowells@redhat.com/ # v2
Signed-off-by: Christian Brauner <brauner@kernel.org>

3 files changed, 134 insertions, 13 deletions

diff --git a/include/linux/folio_queue.h b/include/linux/folio_queue.h
index 52773613bf23..955680c3bb5f 100644
--- a/include/linux/folio_queue.h
+++ b/include/linux/folio_queue.h
@@ -27,6 +27,7 @@ struct folio_queue {
 	struct folio_queue	*prev;		/* Previous queue segment of NULL */
 	unsigned long		marks;		/* 1-bit mark per folio */
 	unsigned long		marks2;		/* Second 1-bit mark per folio */
+	unsigned long		marks3;		/* Third 1-bit mark per folio */
 #if PAGEVEC_SIZE > BITS_PER_LONG
 #error marks is not big enough
 #endif
@@ -39,6 +40,7 @@ static inline void folioq_init(struct folio_queue *folioq)
 	folioq->prev = NULL;
 	folioq->marks = 0;
 	folioq->marks2 = 0;
+	folioq->marks3 = 0;
 }
 
 static inline unsigned int folioq_nr_slots(const struct folio_queue *folioq)
@@ -87,6 +89,21 @@ static inline void folioq_unmark2(struct folio_queue *folioq, unsigned int slot)
 	clear_bit(slot, &folioq->marks2);
 }
 
+static inline bool folioq_is_marked3(const struct folio_queue *folioq, unsigned int slot)
+{
+	return test_bit(slot, &folioq->marks3);
+}
+
+static inline void folioq_mark3(struct folio_queue *folioq, unsigned int slot)
+{
+	set_bit(slot, &folioq->marks3);
+}
+
+static inline void folioq_unmark3(struct folio_queue *folioq, unsigned int slot)
+{
+	clear_bit(slot, &folioq->marks3);
+}
+
 static inline unsigned int __folio_order(struct folio *folio)
 {
 	if (!folio_test_large(folio))
@@ -133,6 +150,7 @@ static inline void folioq_clear(struct folio_queue *folioq, unsigned int slot)
 	folioq->vec.folios[slot] = NULL;
 	folioq_unmark(folioq, slot);
 	folioq_unmark2(folioq, slot);
+	folioq_unmark3(folioq, slot);
 }
 
 #endif /* _LINUX_FOLIO_QUEUE_H */
diff --git a/include/linux/netfs.h b/include/linux/netfs.h
index bd0e3d147822..c0f0c9c87d86 100644
--- a/include/linux/netfs.h
+++ b/include/linux/netfs.h
@@ -178,36 +178,43 @@ struct netfs_io_subrequest {
 	unsigned long long	start;		/* Where to start the I/O */
 	size_t			len;		/* Size of the I/O */
 	size_t			transferred;	/* Amount of data transferred */
+	size_t			consumed;	/* Amount of read data consumed */
+	size_t			prev_donated;	/* Amount of data donated from previous subreq */
+	size_t			next_donated;	/* Amount of data donated from next subreq */
 	refcount_t		ref;
 	short			error;		/* 0 or error that occurred */
 	unsigned short		debug_index;	/* Index in list (for debugging output) */
 	unsigned int		nr_segs;	/* Number of segs in io_iter */
 	enum netfs_io_source	source;		/* Where to read from/write to */
 	unsigned char		stream_nr;	/* I/O stream this belongs to */
+	unsigned char		curr_folioq_slot; /* Folio currently being read */
+	unsigned char		curr_folio_order; /* Order of folio */
+	struct folio_queue	*curr_folioq;	/* Queue segment in which current folio resides */
 	unsigned long		flags;
 #define NETFS_SREQ_COPY_TO_CACHE	0	/* Set if should copy the data to the cache */
 #define NETFS_SREQ_CLEAR_TAIL		1	/* Set if the rest of the read should be cleared */
-#define NETFS_SREQ_SHORT_IO		2	/* Set if the I/O was short */
 #define NETFS_SREQ_SEEK_DATA_READ	3	/* Set if ->read() should SEEK_DATA first */
 #define NETFS_SREQ_NO_PROGRESS		4	/* Set if we didn't manage to read any data */
 #define NETFS_SREQ_ONDEMAND		5	/* Set if it's from on-demand read mode */
 #define NETFS_SREQ_BOUNDARY		6	/* Set if ends on hard boundary (eg. ceph object) */
+#define NETFS_SREQ_HIT_EOF		7	/* Set if short due to EOF */
 #define NETFS_SREQ_IN_PROGRESS		8	/* Unlocked when the subrequest completes */
 #define NETFS_SREQ_NEED_RETRY		9	/* Set if the filesystem requests a retry */
 #define NETFS_SREQ_RETRYING		10	/* Set if we're retrying */
 #define NETFS_SREQ_FAILED		11	/* Set if the subreq failed unretryably */
-#define NETFS_SREQ_HIT_EOF		12	/* Set if we hit the EOF */
 };
 
 enum netfs_io_origin {
 	NETFS_READAHEAD,		/* This read was triggered by readahead */
 	NETFS_READPAGE,			/* This read is a synchronous read */
+	NETFS_READ_GAPS,		/* This read is a synchronous read to fill gaps */
 	NETFS_READ_FOR_WRITE,		/* This read is to prepare a write */
 	NETFS_DIO_READ,			/* This is a direct I/O read */
 	NETFS_WRITEBACK,		/* This write was triggered by writepages */
 	NETFS_WRITETHROUGH,		/* This write was made by netfs_perform_write() */
 	NETFS_UNBUFFERED_WRITE,		/* This is an unbuffered write */
 	NETFS_DIO_WRITE,		/* This is a direct I/O write */
+	NETFS_PGPRIV2_COPY_TO_CACHE,	/* [DEPRECATED] This is writing read data to the cache */
 	nr__netfs_io_origin
 } __mode(byte);
 
@@ -224,6 +231,7 @@ struct netfs_io_request {
 	struct address_space	*mapping;	/* The mapping being accessed */
 	struct kiocb		*iocb;		/* AIO completion vector */
 	struct netfs_cache_resources cache_resources;
+	struct readahead_control *ractl;	/* Readahead descriptor */
 	struct list_head	proc_link;	/* Link in netfs_iorequests */
 	struct list_head	subrequests;	/* Contributory I/O operations */
 	struct netfs_io_stream	io_streams[2];	/* Streams of parallel I/O operations */
@@ -244,12 +252,10 @@ struct netfs_io_request {
 	unsigned int		nr_group_rel;	/* Number of refs to release on ->group */
 	spinlock_t		lock;		/* Lock for queuing subreqs */
 	atomic_t		nr_outstanding;	/* Number of ops in progress */
-	atomic_t		nr_copy_ops;	/* Number of copy-to-cache ops in progress */
-	size_t			upper_len;	/* Length can be extended to here */
 	unsigned long long	submitted;	/* Amount submitted for I/O so far */
 	unsigned long long	len;		/* Length of the request */
 	size_t			transferred;	/* Amount to be indicated as transferred */
-	short			error;		/* 0 or error that occurred */
+	long			error;		/* 0 or error that occurred */
 	enum netfs_io_origin	origin;		/* Origin of the request */
 	bool			direct_bv_unpin; /* T if direct_bv[] must be unpinned */
 	u8			buffer_head_slot; /* First slot in ->buffer */
@@ -260,9 +266,9 @@ struct netfs_io_request {
 	unsigned long long	collected_to;	/* Point we've collected to */
 	unsigned long long	cleaned_to;	/* Position we've cleaned folios to */
 	pgoff_t			no_unlock_folio; /* Don't unlock this folio after read */
+	size_t			prev_donated;	/* Fallback for subreq->prev_donated */
 	refcount_t		ref;
 	unsigned long		flags;
-#define NETFS_RREQ_INCOMPLETE_IO	0	/* Some ioreqs terminated short or with error */
 #define NETFS_RREQ_COPY_TO_CACHE	1	/* Need to write to the cache */
 #define NETFS_RREQ_NO_UNLOCK_FOLIO	2	/* Don't unlock no_unlock_folio on completion */
 #define NETFS_RREQ_DONT_UNLOCK_FOLIOS	3	/* Don't unlock the folios on completion */
@@ -274,6 +280,7 @@ struct netfs_io_request {
 #define NETFS_RREQ_PAUSE		11	/* Pause subrequest generation */
 #define NETFS_RREQ_USE_IO_ITER		12	/* Use ->io_iter rather than ->i_pages */
 #define NETFS_RREQ_ALL_QUEUED		13	/* All subreqs are now queued */
+#define NETFS_RREQ_NEED_RETRY		14	/* Need to try retrying */
 #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;
@@ -292,7 +299,7 @@ struct netfs_request_ops {
 
 	/* Read request handling */
 	void (*expand_readahead)(struct netfs_io_request *rreq);
-	bool (*clamp_length)(struct netfs_io_subrequest *subreq);
+	int (*prepare_read)(struct netfs_io_subrequest *subreq);
 	void (*issue_read)(struct netfs_io_subrequest *subreq);
 	bool (*is_still_valid)(struct netfs_io_request *rreq);
 	int (*check_write_begin)(struct file *file, loff_t pos, unsigned len,
@@ -422,7 +429,10 @@ bool netfs_release_folio(struct folio *folio, gfp_t gfp);
 vm_fault_t netfs_page_mkwrite(struct vm_fault *vmf, struct netfs_group *netfs_group);
 
 /* (Sub)request management API. */
-void netfs_subreq_terminated(struct netfs_io_subrequest *, ssize_t, bool);
+void netfs_read_subreq_progress(struct netfs_io_subrequest *subreq,
+				bool was_async);
+void netfs_read_subreq_terminated(struct netfs_io_subrequest *subreq,
+				  int error, bool was_async);
 void netfs_get_subrequest(struct netfs_io_subrequest *subreq,
 			  enum netfs_sreq_ref_trace what);
 void netfs_put_subrequest(struct netfs_io_subrequest *subreq,
diff --git a/include/trace/events/netfs.h b/include/trace/events/netfs.h
index 58bf23002fc1..7b26463cb98f 100644
--- a/include/trace/events/netfs.h
+++ b/include/trace/events/netfs.h
@@ -20,6 +20,7 @@
 	EM(netfs_read_trace_expanded,		"EXPANDED ")	\
 	EM(netfs_read_trace_readahead,		"READAHEAD")	\
 	EM(netfs_read_trace_readpage,		"READPAGE ")	\
+	EM(netfs_read_trace_read_gaps,		"READ-GAPS")	\
 	EM(netfs_read_trace_prefetch_for_write,	"PREFETCHW")	\
 	E_(netfs_read_trace_write_begin,	"WRITEBEGN")
 
@@ -33,12 +34,14 @@
 #define netfs_rreq_origins					\
 	EM(NETFS_READAHEAD,			"RA")		\
 	EM(NETFS_READPAGE,			"RP")		\
+	EM(NETFS_READ_GAPS,			"RG")		\
 	EM(NETFS_READ_FOR_WRITE,		"RW")		\
 	EM(NETFS_DIO_READ,			"DR")		\
 	EM(NETFS_WRITEBACK,			"WB")		\
 	EM(NETFS_WRITETHROUGH,			"WT")		\
 	EM(NETFS_UNBUFFERED_WRITE,		"UW")		\
-	E_(NETFS_DIO_WRITE,			"DW")
+	EM(NETFS_DIO_WRITE,			"DW")		\
+	E_(NETFS_PGPRIV2_COPY_TO_CACHE,		"2C")
 
 #define netfs_rreq_traces					\
 	EM(netfs_rreq_trace_assess,		"ASSESS ")	\
@@ -69,15 +72,25 @@
 	E_(NETFS_INVALID_WRITE,			"INVL")
 
 #define netfs_sreq_traces					\
+	EM(netfs_sreq_trace_add_donations,	"+DON ")	\
+	EM(netfs_sreq_trace_added,		"ADD  ")	\
+	EM(netfs_sreq_trace_clear,		"CLEAR")	\
 	EM(netfs_sreq_trace_discard,		"DSCRD")	\
+	EM(netfs_sreq_trace_donate_to_prev,	"DON-P")	\
+	EM(netfs_sreq_trace_donate_to_next,	"DON-N")	\
 	EM(netfs_sreq_trace_download_instead,	"RDOWN")	\
 	EM(netfs_sreq_trace_fail,		"FAIL ")	\
 	EM(netfs_sreq_trace_free,		"FREE ")	\
+	EM(netfs_sreq_trace_hit_eof,		"EOF  ")	\
+	EM(netfs_sreq_trace_io_progress,	"IO   ")	\
 	EM(netfs_sreq_trace_limited,		"LIMIT")	\
 	EM(netfs_sreq_trace_prepare,		"PREP ")	\
 	EM(netfs_sreq_trace_prep_failed,	"PRPFL")	\
-	EM(netfs_sreq_trace_resubmit_short,	"SHORT")	\
+	EM(netfs_sreq_trace_progress,		"PRGRS")	\
+	EM(netfs_sreq_trace_reprep_failed,	"REPFL")	\
 	EM(netfs_sreq_trace_retry,		"RETRY")	\
+	EM(netfs_sreq_trace_short,		"SHORT")	\
+	EM(netfs_sreq_trace_split,		"SPLIT")	\
 	EM(netfs_sreq_trace_submit,		"SUBMT")	\
 	EM(netfs_sreq_trace_terminated,		"TERM ")	\
 	EM(netfs_sreq_trace_write,		"WRITE")	\
@@ -118,7 +131,7 @@
 	EM(netfs_sreq_trace_new,		"NEW        ")	\
 	EM(netfs_sreq_trace_put_cancel,		"PUT CANCEL ")	\
 	EM(netfs_sreq_trace_put_clear,		"PUT CLEAR  ")	\
-	EM(netfs_sreq_trace_put_discard,	"PUT DISCARD")	\
+	EM(netfs_sreq_trace_put_consumed,	"PUT CONSUME")	\
 	EM(netfs_sreq_trace_put_done,		"PUT DONE   ")	\
 	EM(netfs_sreq_trace_put_failed,		"PUT FAILED ")	\
 	EM(netfs_sreq_trace_put_merged,		"PUT MERGED ")	\
@@ -138,6 +151,7 @@
 	EM(netfs_flush_content,			"flush")	\
 	EM(netfs_streaming_filled_page,		"mod-streamw-f") \
 	EM(netfs_streaming_cont_filled_page,	"mod-streamw-f+") \
+	EM(netfs_folio_trace_abandon,		"abandon")	\
 	EM(netfs_folio_trace_cancel_copy,	"cancel-copy")	\
 	EM(netfs_folio_trace_clear,		"clear")	\
 	EM(netfs_folio_trace_clear_cc,		"clear-cc")	\
@@ -154,7 +168,11 @@
 	EM(netfs_folio_trace_mkwrite_plus,	"mkwrite+")	\
 	EM(netfs_folio_trace_not_under_wback,	"!wback")	\
 	EM(netfs_folio_trace_put,		"put")		\
+	EM(netfs_folio_trace_read,		"read")		\
+	EM(netfs_folio_trace_read_done,		"read-done")	\
 	EM(netfs_folio_trace_read_gaps,		"read-gaps")	\
+	EM(netfs_folio_trace_read_put,		"read-put")	\
+	EM(netfs_folio_trace_read_unlock,	"read-unlock")	\
 	EM(netfs_folio_trace_redirtied,		"redirtied")	\
 	EM(netfs_folio_trace_store,		"store")	\
 	EM(netfs_folio_trace_store_copy,	"store-copy")	\
@@ -167,6 +185,12 @@
 	EM(netfs_contig_trace_jump,		"-->JUMP-->")	\
 	E_(netfs_contig_trace_unlock,		"Unlock")
 
+#define netfs_donate_traces					\
+	EM(netfs_trace_donate_tail_to_prev,	"tail-to-prev")	\
+	EM(netfs_trace_donate_to_prev,		"to-prev")	\
+	EM(netfs_trace_donate_to_next,		"to-next")	\
+	E_(netfs_trace_donate_to_deferred_next,	"defer-next")
+
 #ifndef __NETFS_DECLARE_TRACE_ENUMS_ONCE_ONLY
 #define __NETFS_DECLARE_TRACE_ENUMS_ONCE_ONLY
 
@@ -184,6 +208,7 @@ enum netfs_rreq_ref_trace { netfs_rreq_ref_traces } __mode(byte);
 enum netfs_sreq_ref_trace { netfs_sreq_ref_traces } __mode(byte);
 enum netfs_folio_trace { netfs_folio_traces } __mode(byte);
 enum netfs_collect_contig_trace { netfs_collect_contig_traces } __mode(byte);
+enum netfs_donate_trace { netfs_donate_traces } __mode(byte);
 
 #endif
 
@@ -206,6 +231,7 @@ netfs_rreq_ref_traces;
 netfs_sreq_ref_traces;
 netfs_folio_traces;
 netfs_collect_contig_traces;
+netfs_donate_traces;
 
 /*
  * Now redefine the EM() and E_() macros to map the enums to the strings that
@@ -226,6 +252,7 @@ TRACE_EVENT(netfs_read,
 	    TP_STRUCT__entry(
 		    __field(unsigned int,		rreq		)
 		    __field(unsigned int,		cookie		)
+		    __field(loff_t,			i_size		)
 		    __field(loff_t,			start		)
 		    __field(size_t,			len		)
 		    __field(enum netfs_read_trace,	what		)
@@ -235,18 +262,19 @@ TRACE_EVENT(netfs_read,
 	    TP_fast_assign(
 		    __entry->rreq	= rreq->debug_id;
 		    __entry->cookie	= rreq->cache_resources.debug_id;
+		    __entry->i_size	= rreq->i_size;
 		    __entry->start	= start;
 		    __entry->len	= len;
 		    __entry->what	= what;
 		    __entry->netfs_inode = rreq->inode->i_ino;
 			   ),
 
-	    TP_printk("R=%08x %s c=%08x ni=%x s=%llx %zx",
+	    TP_printk("R=%08x %s c=%08x ni=%x s=%llx l=%zx sz=%llx",
 		      __entry->rreq,
 		      __print_symbolic(__entry->what, netfs_read_traces),
 		      __entry->cookie,
 		      __entry->netfs_inode,
-		      __entry->start, __entry->len)
+		      __entry->start, __entry->len, __entry->i_size)
 	    );
 
 TRACE_EVENT(netfs_rreq,
@@ -651,6 +679,71 @@ TRACE_EVENT(netfs_collect_stream,
 		      __entry->collected_to, __entry->front)
 	    );
 
+TRACE_EVENT(netfs_progress,
+	    TP_PROTO(const struct netfs_io_subrequest *subreq,
+		     unsigned long long start, size_t avail, size_t part),
+
+	    TP_ARGS(subreq, start, avail, part),
+
+	    TP_STRUCT__entry(
+		    __field(unsigned int,		rreq)
+		    __field(unsigned int,		subreq)
+		    __field(unsigned int,		consumed)
+		    __field(unsigned int,		transferred)
+		    __field(unsigned long long,		f_start)
+		    __field(unsigned int,		f_avail)
+		    __field(unsigned int,		f_part)
+		    __field(unsigned char,		slot)
+			     ),
+
+	    TP_fast_assign(
+		    __entry->rreq	= subreq->rreq->debug_id;
+		    __entry->subreq	= subreq->debug_index;
+		    __entry->consumed	= subreq->consumed;
+		    __entry->transferred = subreq->transferred;
+		    __entry->f_start	= start;
+		    __entry->f_avail	= avail;
+		    __entry->f_part	= part;
+		    __entry->slot	= subreq->curr_folioq_slot;
+			   ),
+
+	    TP_printk("R=%08x[%02x] s=%llx ct=%x/%x pa=%x/%x sl=%x",
+		      __entry->rreq, __entry->subreq, __entry->f_start,
+		      __entry->consumed, __entry->transferred,
+		      __entry->f_part, __entry->f_avail,  __entry->slot)
+	    );
+
+TRACE_EVENT(netfs_donate,
+	    TP_PROTO(const struct netfs_io_request *rreq,
+		     const struct netfs_io_subrequest *from,
+		     const struct netfs_io_subrequest *to,
+		     size_t amount,
+		     enum netfs_donate_trace trace),
+
+	    TP_ARGS(rreq, from, to, amount, trace),
+
+	    TP_STRUCT__entry(
+		    __field(unsigned int,		rreq)
+		    __field(unsigned int,		from)
+		    __field(unsigned int,		to)
+		    __field(unsigned int,		amount)
+		    __field(enum netfs_donate_trace,	trace)
+			     ),
+
+	    TP_fast_assign(
+		    __entry->rreq	= rreq->debug_id;
+		    __entry->from	= from->debug_index;
+		    __entry->to		= to ? to->debug_index : -1;
+		    __entry->amount	= amount;
+		    __entry->trace	= trace;
+			   ),
+
+	    TP_printk("R=%08x[%02x] -> [%02x] %s am=%x",
+		      __entry->rreq, __entry->from, __entry->to,
+		      __print_symbolic(__entry->trace, netfs_donate_traces),
+		      __entry->amount)
+	    );
+
 #undef EM
 #undef E_
 #endif /* _TRACE_NETFS_H */