CIFS: Use multicredits for SMB 2.1/3 reads

If we negotiate SMB 2.1 and higher version of the protocol and
a server supports large read buffer size, we need to consume 1
credit per 65536 bytes. So, we need to know how many credits
we have and obtain the required number of them before constructing
a readdata structure in readpages and user read.

Reviewed-by: Shirish Pargaonkar <spargaonkar@suse.com>
Signed-off-by: Pavel Shilovsky <pshilovsky@samba.org>
Signed-off-by: Steve French <smfrench@gmail.com>

4 files changed, 59 insertions, 12 deletions

diff --git a/fs/cifs/cifsglob.h b/fs/cifs/cifsglob.h
index 54ca2b985a0d..f33ff4c7b8a8 100644
--- a/fs/cifs/cifsglob.h
+++ b/fs/cifs/cifsglob.h
@@ -1068,6 +1068,7 @@ struct cifs_readdata {
         struct kvec                     iov;
         unsigned int                    pagesz;
         unsigned int                    tailsz;
+        unsigned int                    credits;
         unsigned int                    nr_pages;
         struct page                     *pages[];
 };
diff --git a/fs/cifs/file.c b/fs/cifs/file.c
index 00b2a254ff1c..ebdeb56f8d30 100644
--- a/fs/cifs/file.c
+++ b/fs/cifs/file.c
@@ -2917,7 +2917,7 @@ cifs_send_async_read(loff_t offset, size_t len, struct cifsFileInfo *open_file,
                      struct cifs_sb_info *cifs_sb, struct list_head *rdata_list)
 {
         struct cifs_readdata *rdata;
-        unsigned int npages;
+        unsigned int npages, rsize, credits;
         size_t cur_len;
         int rc;
         pid_t pid;
@@ -2931,13 +2931,19 @@ cifs_send_async_read(loff_t offset, size_t len, struct cifsFileInfo *open_file,
                 pid = current->tgid;
 
         do {
-                cur_len = min_t(const size_t, len, cifs_sb->rsize);
+                rc = server->ops->wait_mtu_credits(server, cifs_sb->rsize,
+                                                   &rsize, &credits);
+                if (rc)
+                        break;
+
+                cur_len = min_t(const size_t, len, rsize);
                 npages = DIV_ROUND_UP(cur_len, PAGE_SIZE);
 
                 /* allocate a readdata struct */
                 rdata = cifs_readdata_alloc(npages,
                                             cifs_uncached_readv_complete);
                 if (!rdata) {
+                        add_credits_and_wake_if(server, credits, 0);
                         rc = -ENOMEM;
                         break;
                 }
@@ -2953,12 +2959,14 @@ cifs_send_async_read(loff_t offset, size_t len, struct cifsFileInfo *open_file,
                 rdata->pid = pid;
                 rdata->pagesz = PAGE_SIZE;
                 rdata->read_into_pages = cifs_uncached_read_into_pages;
+                rdata->credits = credits;
 
                 if (!rdata->cfile->invalidHandle ||
                     !cifs_reopen_file(rdata->cfile, true))
                         rc = server->ops->async_readv(rdata);
 error:
                 if (rc) {
+                        add_credits_and_wake_if(server, rdata->credits, 0);
                         kref_put(&rdata->refcount,
                                  cifs_uncached_readdata_release);
                         if (rc == -EAGAIN)
@@ -3458,10 +3466,16 @@ static int cifs_readpages(struct file *file, struct address_space *mapping,
          * the rdata->pages, then we want them in increasing order.
          */
         while (!list_empty(page_list)) {
-                unsigned int i, nr_pages, bytes;
+                unsigned int i, nr_pages, bytes, rsize;
                 loff_t offset;
                 struct page *page, *tpage;
                 struct cifs_readdata *rdata;
+                unsigned credits;
+
+                rc = server->ops->wait_mtu_credits(server, cifs_sb->rsize,
+                                                   &rsize, &credits);
+                if (rc)
+                        break;
 
                 /*
                  * Give up immediately if rsize is too small to read an entire
@@ -3469,13 +3483,17 @@ static int cifs_readpages(struct file *file, struct address_space *mapping,
                  * reach this point however since we set ra_pages to 0 when the
                  * rsize is smaller than a cache page.
                  */
-                if (unlikely(cifs_sb->rsize < PAGE_CACHE_SIZE))
+                if (unlikely(rsize < PAGE_CACHE_SIZE)) {
+                        add_credits_and_wake_if(server, credits, 0);
                         return 0;
+                }
 
-                rc = readpages_get_pages(mapping, page_list, cifs_sb->rsize,
-                                         &tmplist, &nr_pages, &offset, &bytes);
-                if (rc)
+                rc = readpages_get_pages(mapping, page_list, rsize, &tmplist,
+                                         &nr_pages, &offset, &bytes);
+                if (rc) {
+                        add_credits_and_wake_if(server, credits, 0);
                         break;
+                }
 
                 rdata = cifs_readdata_alloc(nr_pages, cifs_readv_complete);
                 if (!rdata) {
@@ -3487,6 +3505,7 @@ static int cifs_readpages(struct file *file, struct address_space *mapping,
                                 page_cache_release(page);
                         }
                         rc = -ENOMEM;
+                        add_credits_and_wake_if(server, credits, 0);
                         break;
                 }
 
@@ -3497,6 +3516,7 @@ static int cifs_readpages(struct file *file, struct address_space *mapping,
                 rdata->pid = pid;
                 rdata->pagesz = PAGE_CACHE_SIZE;
                 rdata->read_into_pages = cifs_readpages_read_into_pages;
+                rdata->credits = credits;
 
                 list_for_each_entry_safe(page, tpage, &tmplist, lru) {
                         list_del(&page->lru);
@@ -3507,6 +3527,7 @@ static int cifs_readpages(struct file *file, struct address_space *mapping,
                     !cifs_reopen_file(rdata->cfile, true))
                         rc = server->ops->async_readv(rdata);
                 if (rc) {
+                        add_credits_and_wake_if(server, rdata->credits, 0);
                         for (i = 0; i < rdata->nr_pages; i++) {
                                 page = rdata->pages[i];
                                 lru_cache_add_file(page);
diff --git a/fs/cifs/smb2ops.c b/fs/cifs/smb2ops.c
index 3427c1fa3806..d0210a8e9829 100644
--- a/fs/cifs/smb2ops.c
+++ b/fs/cifs/smb2ops.c
@@ -245,8 +245,9 @@ smb2_negotiate_rsize(struct cifs_tcon *tcon, struct smb_vol *volume_info)
         /* start with specified rsize, or default */
         rsize = volume_info->rsize ? volume_info->rsize : CIFS_DEFAULT_IOSIZE;
         rsize = min_t(unsigned int, rsize, server->max_read);
-        /* set it to the maximum buffer size value we can send with 1 credit */
-        rsize = min_t(unsigned int, rsize, SMB2_MAX_BUFFER_SIZE);
+
+        if (!(server->capabilities & SMB2_GLOBAL_CAP_LARGE_MTU))
+                rsize = min_t(unsigned int, rsize, SMB2_MAX_BUFFER_SIZE);
 
         return rsize;
 }
diff --git a/fs/cifs/smb2pdu.c b/fs/cifs/smb2pdu.c
index be7b2eb55134..c31e5a060338 100644
--- a/fs/cifs/smb2pdu.c
+++ b/fs/cifs/smb2pdu.c
@@ -1747,11 +1747,12 @@ smb2_readv_callback(struct mid_q_entry *mid)
 int
 smb2_async_readv(struct cifs_readdata *rdata)
 {
-        int rc;
+        int rc, flags = 0;
         struct smb2_hdr *buf;
         struct cifs_io_parms io_parms;
         struct smb_rqst rqst = { .rq_iov = &rdata->iov,
                                  .rq_nvec = 1 };
+        struct TCP_Server_Info *server;
 
         cifs_dbg(FYI, "%s: offset=%llu bytes=%u\n",
                  __func__, rdata->offset, rdata->bytes);
@@ -1762,18 +1763,41 @@ smb2_async_readv(struct cifs_readdata *rdata)
         io_parms.persistent_fid = rdata->cfile->fid.persistent_fid;
         io_parms.volatile_fid = rdata->cfile->fid.volatile_fid;
         io_parms.pid = rdata->pid;
+
+        server = io_parms.tcon->ses->server;
+
         rc = smb2_new_read_req(&rdata->iov, &io_parms, 0, 0);
-        if (rc)
+        if (rc) {
+                if (rc == -EAGAIN && rdata->credits) {
+                        /* credits was reset by reconnect */
+                        rdata->credits = 0;
+                        /* reduce in_flight value since we won't send the req */
+                        spin_lock(&server->req_lock);
+                        server->in_flight--;
+                        spin_unlock(&server->req_lock);
+                }
                 return rc;
+        }
 
         buf = (struct smb2_hdr *)rdata->iov.iov_base;
         /* 4 for rfc1002 length field */
         rdata->iov.iov_len = get_rfc1002_length(rdata->iov.iov_base) + 4;
 
+        if (rdata->credits) {
+                buf->CreditCharge = cpu_to_le16(DIV_ROUND_UP(rdata->bytes,
+                                                SMB2_MAX_BUFFER_SIZE));
+                spin_lock(&server->req_lock);
+                server->credits += rdata->credits -
+                                                le16_to_cpu(buf->CreditCharge);
+                spin_unlock(&server->req_lock);
+                wake_up(&server->request_q);
+                flags = CIFS_HAS_CREDITS;
+        }
+
         kref_get(&rdata->refcount);
         rc = cifs_call_async(io_parms.tcon->ses->server, &rqst,
                              cifs_readv_receive, smb2_readv_callback,
-                             rdata, 0);
+                             rdata, flags);
         if (rc) {
                 kref_put(&rdata->refcount, cifs_readdata_release);
                 cifs_stats_fail_inc(io_parms.tcon, SMB2_READ_HE);