Merge tag '9p-for-6.1' of https://github.com/martinetd/linux

Pull 9p updates from Dominique Martinet:
 "Smaller buffers for small messages and fixes.

  The highlight of this is Christian's patch to allocate smaller buffers
  for most metadata requests: 9p with a big msize would try to allocate
  large buffers when just 4 or 8k would be more than enough; this brings
  in nice performance improvements.

  There's also a few fixes for problems reported by syzkaller (thanks to
  Schspa Shi, Tetsuo Handa for tests and feedback/patches) as well as
  some minor cleanup"

* tag '9p-for-6.1' of https://github.com/martinetd/linux:
  net/9p: clarify trans_fd parse_opt failure handling
  net/9p: add __init/__exit annotations to module init/exit funcs
  net/9p: use a dedicated spinlock for trans_fd
  9p/trans_fd: always use O_NONBLOCK read/write
  net/9p: allocate appropriate reduced message buffers
  net/9p: add 'pooled_rbuffers' flag to struct p9_trans_module
  net/9p: add p9_msg_buf_size()
  9p: add P9_ERRMAX for 9p2000 and 9p2000.u
  net/9p: split message size argument into 't_size' and 'r_size' pair
  9p: trans_fd/p9_conn_cancel: drop client lock earlier

include/net/9p/9p.h

@@ -331,6 +331,9 @@ enum p9_qid_t {
 /* size of header for zero copy read/write */
 #define P9_ZC_HDR_SZ 4096
 
+/* maximum length of an error string */
+#define P9_ERRMAX 128
+
 /**
  * struct p9_qid - file system entity information
  * @type: 8-bit type &p9_qid_t

include/net/9p/transport.h

@@ -19,6 +19,10 @@
  * @list: used to maintain a list of currently available transports
  * @name: the human-readable name of the transport
  * @maxsize: transport provided maximum packet size
+ * @pooled_rbuffers: currently only set for RDMA transport which pulls the
+ *                   response buffers from a shared pool, and accordingly
+ *                   we're less flexible when choosing the response message
+ *                   size in this case
  * @def: set if this transport should be considered the default
  * @create: member function to create a new connection on this transport
  * @close: member function to discard a connection on this transport
@@ -38,6 +42,7 @@ struct p9_trans_module {
 	struct list_head list;
 	char *name;		/* name of transport */
 	int maxsize;		/* max message size of transport */
+	bool pooled_rbuffers;
 	int def;		/* this transport should be default */
 	struct module *owner;
 	int (*create)(struct p9_client *client,

net/9p/client.c

@@ -255,24 +255,42 @@ static struct kmem_cache *p9_req_cache;
  * p9_tag_alloc - Allocate a new request.
  * @c: Client session.
  * @type: Transaction type.
- * @max_size: Maximum packet size for this request.
+ * @t_size: Buffer size for holding this request
+ * (automatic calculation by format template if 0).
+ * @r_size: Buffer size for holding server's reply on this request
+ * (automatic calculation by format template if 0).
+ * @fmt: Format template for assembling 9p request message
+ * (see p9pdu_vwritef).
+ * @ap: Variable arguments to be fed to passed format template
+ * (see p9pdu_vwritef).
  *
  * Context: Process context.
  * Return: Pointer to new request.
  */
 static struct p9_req_t *
-p9_tag_alloc(struct p9_client *c, int8_t type, unsigned int max_size)
+p9_tag_alloc(struct p9_client *c, int8_t type, uint t_size, uint r_size,
+	      const char *fmt, va_list ap)
 {
 	struct p9_req_t *req = kmem_cache_alloc(p9_req_cache, GFP_NOFS);
-	int alloc_msize = min(c->msize, max_size);
+	int alloc_tsize;
+	int alloc_rsize;
 	int tag;
+	va_list apc;
+
+	va_copy(apc, ap);
+	alloc_tsize = min_t(size_t, c->msize,
+			    t_size ?: p9_msg_buf_size(c, type, fmt, apc));
+	va_end(apc);
+
+	alloc_rsize = min_t(size_t, c->msize,
+			    r_size ?: p9_msg_buf_size(c, type + 1, fmt, ap));
 
 	if (!req)
 		return ERR_PTR(-ENOMEM);
 
-	if (p9_fcall_init(c, &req->tc, alloc_msize))
+	if (p9_fcall_init(c, &req->tc, alloc_tsize))
 		goto free_req;
-	if (p9_fcall_init(c, &req->rc, alloc_msize))
+	if (p9_fcall_init(c, &req->rc, alloc_rsize))
 		goto free;
 
 	p9pdu_reset(&req->tc);
@@ -592,11 +610,12 @@ static int p9_client_flush(struct p9_client *c, struct p9_req_t *oldreq)
 }
 
 static struct p9_req_t *p9_client_prepare_req(struct p9_client *c,
-					      int8_t type, int req_size,
+					      int8_t type, uint t_size, uint r_size,
 					      const char *fmt, va_list ap)
 {
 	int err;
 	struct p9_req_t *req;
+	va_list apc;
 
 	p9_debug(P9_DEBUG_MUX, "client %p op %d\n", c, type);
 
@@ -608,7 +627,9 @@ static struct p9_req_t *p9_client_prepare_req(struct p9_client *c,
 	if (c->status == BeginDisconnect && type != P9_TCLUNK)
 		return ERR_PTR(-EIO);
 
-	req = p9_tag_alloc(c, type, req_size);
+	va_copy(apc, ap);
+	req = p9_tag_alloc(c, type, t_size, r_size, fmt, apc);
+	va_end(apc);
 	if (IS_ERR(req))
 		return req;
 
@@ -643,9 +664,18 @@ p9_client_rpc(struct p9_client *c, int8_t type, const char *fmt, ...)
 	int sigpending, err;
 	unsigned long flags;
 	struct p9_req_t *req;
+	/* Passing zero for tsize/rsize to p9_client_prepare_req() tells it to
+	 * auto determine an appropriate (small) request/response size
+	 * according to actual message data being sent. Currently RDMA
+	 * transport is excluded from this response message size optimization,
+	 * as it would not cope with it, due to its pooled response buffers
+	 * (using an optimized request size for RDMA as well though).
+	 */
+	const uint tsize = 0;
+	const uint rsize = c->trans_mod->pooled_rbuffers ? c->msize : 0;
 
 	va_start(ap, fmt);
-	req = p9_client_prepare_req(c, type, c->msize, fmt, ap);
+	req = p9_client_prepare_req(c, type, tsize, rsize, fmt, ap);
 	va_end(ap);
 	if (IS_ERR(req))
 		return req;
@@ -743,7 +773,7 @@ static struct p9_req_t *p9_client_zc_rpc(struct p9_client *c, int8_t type,
 	/* We allocate a inline protocol data of only 4k bytes.
 	 * The actual content is passed in zero-copy fashion.
 	 */
-	req = p9_client_prepare_req(c, type, P9_ZC_HDR_SZ, fmt, ap);
+	req = p9_client_prepare_req(c, type, P9_ZC_HDR_SZ, P9_ZC_HDR_SZ, fmt, ap);
 	va_end(ap);
 	if (IS_ERR(req))
 		return req;

net/9p/protocol.c

@@ -23,6 +23,173 @@
 
 #include <trace/events/9p.h>
 
+/* len[2] text[len] */
+#define P9_STRLEN(s) \
+	(2 + min_t(size_t, s ? strlen(s) : 0, USHRT_MAX))
+
+/**
+ * p9_msg_buf_size - Returns a buffer size sufficiently large to hold the
+ * intended 9p message.
+ * @c: client
+ * @type: message type
+ * @fmt: format template for assembling request message
+ * (see p9pdu_vwritef)
+ * @ap: variable arguments to be fed to passed format template
+ * (see p9pdu_vwritef)
+ *
+ * Note: Even for response types (P9_R*) the format template and variable
+ * arguments must always be for the originating request type (P9_T*).
+ */
+size_t p9_msg_buf_size(struct p9_client *c, enum p9_msg_t type,
+			const char *fmt, va_list ap)
+{
+	/* size[4] type[1] tag[2] */
+	const int hdr = 4 + 1 + 2;
+	/* ename[s] errno[4] */
+	const int rerror_size = hdr + P9_ERRMAX + 4;
+	/* ecode[4] */
+	const int rlerror_size = hdr + 4;
+	const int err_size =
+		c->proto_version == p9_proto_2000L ? rlerror_size : rerror_size;
+
+	static_assert(NAME_MAX <= 4*1024, "p9_msg_buf_size() currently assumes "
+				  "a max. allowed directory entry name length of 4k");
+
+	switch (type) {
+
+	/* message types not used at all */
+	case P9_TERROR:
+	case P9_TLERROR:
+	case P9_TAUTH:
+	case P9_RAUTH:
+		BUG();
+
+	/* variable length & potentially large message types */
+	case P9_TATTACH:
+		BUG_ON(strcmp("ddss?u", fmt));
+		va_arg(ap, int32_t);
+		va_arg(ap, int32_t);
+		{
+			const char *uname = va_arg(ap, const char *);
+			const char *aname = va_arg(ap, const char *);
+			/* fid[4] afid[4] uname[s] aname[s] n_uname[4] */
+			return hdr + 4 + 4 + P9_STRLEN(uname) + P9_STRLEN(aname) + 4;
+		}
+	case P9_TWALK:
+		BUG_ON(strcmp("ddT", fmt));
+		va_arg(ap, int32_t);
+		va_arg(ap, int32_t);
+		{
+			uint i, nwname = va_arg(ap, int);
+			size_t wname_all;
+			const char **wnames = va_arg(ap, const char **);
+			for (i = 0, wname_all = 0; i < nwname; ++i) {
+				wname_all += P9_STRLEN(wnames[i]);
+			}
+			/* fid[4] newfid[4] nwname[2] nwname*(wname[s]) */
+			return hdr + 4 + 4 + 2 + wname_all;
+		}
+	case P9_RWALK:
+		BUG_ON(strcmp("ddT", fmt));
+		va_arg(ap, int32_t);
+		va_arg(ap, int32_t);
+		{
+			uint nwname = va_arg(ap, int);
+			/* nwqid[2] nwqid*(wqid[13]) */
+			return max_t(size_t, hdr + 2 + nwname * 13, err_size);
+		}
+	case P9_TCREATE:
+		BUG_ON(strcmp("dsdb?s", fmt));
+		va_arg(ap, int32_t);
+		{
+			const char *name = va_arg(ap, const char *);
+			if (c->proto_version == p9_proto_legacy) {
+				/* fid[4] name[s] perm[4] mode[1] */
+				return hdr + 4 + P9_STRLEN(name) + 4 + 1;
+			} else {
+				va_arg(ap, int32_t);
+				va_arg(ap, int);
+				{
+					const char *ext = va_arg(ap, const char *);
+					/* fid[4] name[s] perm[4] mode[1] extension[s] */
+					return hdr + 4 + P9_STRLEN(name) + 4 + 1 + P9_STRLEN(ext);
+				}
+			}
+		}
+	case P9_TLCREATE:
+		BUG_ON(strcmp("dsddg", fmt));
+		va_arg(ap, int32_t);
+		{
+			const char *name = va_arg(ap, const char *);
+			/* fid[4] name[s] flags[4] mode[4] gid[4] */
+			return hdr + 4 + P9_STRLEN(name) + 4 + 4 + 4;
+		}
+	case P9_RREAD:
+	case P9_RREADDIR:
+		BUG_ON(strcmp("dqd", fmt));
+		va_arg(ap, int32_t);
+		va_arg(ap, int64_t);
+		{
+			const int32_t count = va_arg(ap, int32_t);
+			/* count[4] data[count] */
+			return max_t(size_t, hdr + 4 + count, err_size);
+		}
+	case P9_TWRITE:
+		BUG_ON(strcmp("dqV", fmt));
+		va_arg(ap, int32_t);
+		va_arg(ap, int64_t);
+		{
+			const int32_t count = va_arg(ap, int32_t);
+			/* fid[4] offset[8] count[4] data[count] */
+			return hdr + 4 + 8 + 4 + count;
+		}
+	case P9_TRENAMEAT:
+		BUG_ON(strcmp("dsds", fmt));
+		va_arg(ap, int32_t);
+		{
+			const char *oldname, *newname;
+			oldname = va_arg(ap, const char *);
+			va_arg(ap, int32_t);
+			newname = va_arg(ap, const char *);
+			/* olddirfid[4] oldname[s] newdirfid[4] newname[s] */
+			return hdr + 4 + P9_STRLEN(oldname) + 4 + P9_STRLEN(newname);
+		}
+	case P9_TSYMLINK:
+		BUG_ON(strcmp("dssg", fmt));
+		va_arg(ap, int32_t);
+		{
+			const char *name = va_arg(ap, const char *);
+			const char *symtgt = va_arg(ap, const char *);
+			/* fid[4] name[s] symtgt[s] gid[4] */
+			return hdr + 4 + P9_STRLEN(name) + P9_STRLEN(symtgt) + 4;
+		}
+
+	case P9_RERROR:
+		return rerror_size;
+	case P9_RLERROR:
+		return rlerror_size;
+
+	/* small message types */
+	case P9_TWSTAT:
+	case P9_RSTAT:
+	case P9_RREADLINK:
+	case P9_TXATTRWALK:
+	case P9_TXATTRCREATE:
+	case P9_TLINK:
+	case P9_TMKDIR:
+	case P9_TMKNOD:
+	case P9_TRENAME:
+	case P9_TUNLINKAT:
+	case P9_TLOCK:
+		return 8 * 1024;
+
+	/* tiny message types */
+	default:
+		return 4 * 1024;
+
+	}
+}
+
 static int
 p9pdu_writef(struct p9_fcall *pdu, int proto_version, const char *fmt, ...);
 

net/9p/protocol.h

@@ -8,6 +8,8 @@
  *  Copyright (C) 2008 by IBM, Corp.
  */
 
+size_t p9_msg_buf_size(struct p9_client *c, enum p9_msg_t type,
+			const char *fmt, va_list ap);
 int p9pdu_vwritef(struct p9_fcall *pdu, int proto_version, const char *fmt,
 		  va_list ap);
 int p9pdu_readf(struct p9_fcall *pdu, int proto_version, const char *fmt, ...);

net/9p/trans_fd.c

@@ -91,6 +91,7 @@ struct p9_poll_wait {
  * @mux_list: list link for mux to manage multiple connections (?)
  * @client: reference to client instance for this connection
  * @err: error state
+ * @req_lock: lock protecting req_list and requests statuses
  * @req_list: accounting for requests which have been sent
  * @unsent_req_list: accounting for requests that haven't been sent
  * @rreq: read request
@@ -114,6 +115,7 @@ struct p9_conn {
 	struct list_head mux_list;
 	struct p9_client *client;
 	int err;
+	spinlock_t req_lock;
 	struct list_head req_list;
 	struct list_head unsent_req_list;
 	struct p9_req_t *rreq;
@@ -189,10 +191,10 @@ static void p9_conn_cancel(struct p9_conn *m, int err)
 
 	p9_debug(P9_DEBUG_ERROR, "mux %p err %d\n", m, err);
 
-	spin_lock(&m->client->lock);
+	spin_lock(&m->req_lock);
 
 	if (m->err) {
-		spin_unlock(&m->client->lock);
+		spin_unlock(&m->req_lock);
 		return;
 	}
 
@@ -205,6 +207,8 @@ static void p9_conn_cancel(struct p9_conn *m, int err)
 		list_move(&req->req_list, &cancel_list);
 	}
 
+	spin_unlock(&m->req_lock);
+
 	list_for_each_entry_safe(req, rtmp, &cancel_list, req_list) {
 		p9_debug(P9_DEBUG_ERROR, "call back req %p\n", req);
 		list_del(&req->req_list);
@@ -212,7 +216,6 @@ static void p9_conn_cancel(struct p9_conn *m, int err)
 			req->t_err = err;
 		p9_client_cb(m->client, req, REQ_STATUS_ERROR);
 	}
-	spin_unlock(&m->client->lock);
 }
 
 static __poll_t
@@ -359,7 +362,7 @@ static void p9_read_work(struct work_struct *work)
 	if ((m->rreq) && (m->rc.offset == m->rc.capacity)) {
 		p9_debug(P9_DEBUG_TRANS, "got new packet\n");
 		m->rreq->rc.size = m->rc.offset;
-		spin_lock(&m->client->lock);
+		spin_lock(&m->req_lock);
 		if (m->rreq->status == REQ_STATUS_SENT) {
 			list_del(&m->rreq->req_list);
 			p9_client_cb(m->client, m->rreq, REQ_STATUS_RCVD);
@@ -368,14 +371,14 @@ static void p9_read_work(struct work_struct *work)
 			p9_debug(P9_DEBUG_TRANS,
 				 "Ignore replies associated with a cancelled request\n");
 		} else {
-			spin_unlock(&m->client->lock);
+			spin_unlock(&m->req_lock);
 			p9_debug(P9_DEBUG_ERROR,
 				 "Request tag %d errored out while we were reading the reply\n",
 				 m->rc.tag);
 			err = -EIO;
 			goto error;
 		}
-		spin_unlock(&m->client->lock);
+		spin_unlock(&m->req_lock);
 		m->rc.sdata = NULL;
 		m->rc.offset = 0;
 		m->rc.capacity = 0;
@@ -453,10 +456,10 @@ static void p9_write_work(struct work_struct *work)
 	}
 
 	if (!m->wsize) {
-		spin_lock(&m->client->lock);
+		spin_lock(&m->req_lock);
 		if (list_empty(&m->unsent_req_list)) {
 			clear_bit(Wworksched, &m->wsched);
-			spin_unlock(&m->client->lock);
+			spin_unlock(&m->req_lock);
 			return;
 		}
 
@@ -471,7 +474,7 @@ static void p9_write_work(struct work_struct *work)
 		m->wpos = 0;
 		p9_req_get(req);
 		m->wreq = req;
-		spin_unlock(&m->client->lock);
+		spin_unlock(&m->req_lock);
 	}
 
 	p9_debug(P9_DEBUG_TRANS, "mux %p pos %d size %d\n",
@@ -588,6 +591,7 @@ static void p9_conn_create(struct p9_client *client)
 	INIT_LIST_HEAD(&m->mux_list);
 	m->client = client;
 
+	spin_lock_init(&m->req_lock);
 	INIT_LIST_HEAD(&m->req_list);
 	INIT_LIST_HEAD(&m->unsent_req_list);
 	INIT_WORK(&m->rq, p9_read_work);
@@ -669,10 +673,10 @@ static int p9_fd_request(struct p9_client *client, struct p9_req_t *req)
 	if (m->err < 0)
 		return m->err;
 
-	spin_lock(&client->lock);
+	spin_lock(&m->req_lock);
 	req->status = REQ_STATUS_UNSENT;
 	list_add_tail(&req->req_list, &m->unsent_req_list);
-	spin_unlock(&client->lock);
+	spin_unlock(&m->req_lock);
 
 	if (test_and_clear_bit(Wpending, &m->wsched))
 		n = EPOLLOUT;
@@ -687,11 +691,13 @@ static int p9_fd_request(struct p9_client *client, struct p9_req_t *req)
 
 static int p9_fd_cancel(struct p9_client *client, struct p9_req_t *req)
 {
+	struct p9_trans_fd *ts = client->trans;
+	struct p9_conn *m = &ts->conn;
 	int ret = 1;
 
 	p9_debug(P9_DEBUG_TRANS, "client %p req %p\n", client, req);
 
-	spin_lock(&client->lock);
+	spin_lock(&m->req_lock);
 
 	if (req->status == REQ_STATUS_UNSENT) {
 		list_del(&req->req_list);
@@ -699,21 +705,24 @@ static int p9_fd_cancel(struct p9_client *client, struct p9_req_t *req)
 		p9_req_put(client, req);
 		ret = 0;
 	}
-	spin_unlock(&client->lock);
+	spin_unlock(&m->req_lock);
 
 	return ret;
 }
 
 static int p9_fd_cancelled(struct p9_client *client, struct p9_req_t *req)
 {
+	struct p9_trans_fd *ts = client->trans;
+	struct p9_conn *m = &ts->conn;
+
 	p9_debug(P9_DEBUG_TRANS, "client %p req %p\n", client, req);
 
-	spin_lock(&client->lock);
+	spin_lock(&m->req_lock);
 	/* Ignore cancelled request if message has been received
 	 * before lock.
 	 */
 	if (req->status == REQ_STATUS_RCVD) {
-		spin_unlock(&client->lock);
+		spin_unlock(&m->req_lock);
 		return 0;
 	}
 
@@ -722,7 +731,8 @@ static int p9_fd_cancelled(struct p9_client *client, struct p9_req_t *req)
 	 */
 	list_del(&req->req_list);
 	req->status = REQ_STATUS_FLSHD;
-	spin_unlock(&client->lock);
+	spin_unlock(&m->req_lock);
+
 	p9_req_put(client, req);
 
 	return 0;
@@ -821,11 +831,14 @@ static int p9_fd_open(struct p9_client *client, int rfd, int wfd)
 		goto out_free_ts;
 	if (!(ts->rd->f_mode & FMODE_READ))
 		goto out_put_rd;
+	/* prevent workers from hanging on IO when fd is a pipe */
+	ts->rd->f_flags |= O_NONBLOCK;
 	ts->wr = fget(wfd);
 	if (!ts->wr)
 		goto out_put_rd;
 	if (!(ts->wr->f_mode & FMODE_WRITE))
 		goto out_put_wr;
+	ts->wr->f_flags |= O_NONBLOCK;
 
 	client->trans = ts;
 	client->status = Connected;
@@ -1061,7 +1074,9 @@ p9_fd_create(struct p9_client *client, const char *addr, char *args)
 	int err;
 	struct p9_fd_opts opts;
 
-	parse_opts(args, &opts);
+	err = parse_opts(args, &opts);
+	if (err < 0)
+		return err;
 	client->trans_opts.fd.rfd = opts.rfd;
 	client->trans_opts.fd.wfd = opts.wfd;
 
@@ -1082,6 +1097,7 @@ p9_fd_create(struct p9_client *client, const char *addr, char *args)
 static struct p9_trans_module p9_tcp_trans = {
 	.name = "tcp",
 	.maxsize = MAX_SOCK_BUF,
+	.pooled_rbuffers = false,
 	.def = 0,
 	.create = p9_fd_create_tcp,
 	.close = p9_fd_close,

net/9p/trans_rdma.c

@@ -739,6 +739,7 @@ rdma_create_trans(struct p9_client *client, const char *addr, char *args)
 static struct p9_trans_module p9_rdma_trans = {
 	.name = "rdma",
 	.maxsize = P9_RDMA_MAXSIZE,
+	.pooled_rbuffers = true,
 	.def = 0,
 	.owner = THIS_MODULE,
 	.create = rdma_create_trans,

net/9p/trans_virtio.c

@@ -802,6 +802,7 @@ static struct p9_trans_module p9_virtio_trans = {
 	 * page in zero copy.
 	 */
 	.maxsize = PAGE_SIZE * (VIRTQUEUE_NUM - 3),
+	.pooled_rbuffers = false,
 	.def = 1,
 	.owner = THIS_MODULE,
 };

net/9p/trans_xen.c

@@ -246,6 +246,7 @@ static irqreturn_t xen_9pfs_front_event_handler(int irq, void *r)
 static struct p9_trans_module p9_xen_trans = {
 	.name = "xen",
 	.maxsize = 1 << (XEN_9PFS_RING_ORDER + XEN_PAGE_SHIFT - 2),
+	.pooled_rbuffers = false,
 	.def = 1,
 	.create = p9_xen_create,
 	.close = p9_xen_close,
@@ -510,7 +511,7 @@ static struct xenbus_driver xen_9pfs_front_driver = {
 	.otherend_changed = xen_9pfs_front_changed,
 };
 
-static int p9_trans_xen_init(void)
+static int __init p9_trans_xen_init(void)
 {
 	int rc;
 
@@ -529,7 +530,7 @@ static int p9_trans_xen_init(void)
 module_init(p9_trans_xen_init);
 MODULE_ALIAS_9P("xen");
 
-static void p9_trans_xen_exit(void)
+static void __exit p9_trans_xen_exit(void)
 {
 	v9fs_unregister_trans(&p9_xen_trans);
 	return xenbus_unregister_driver(&xen_9pfs_front_driver);