aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorTrond Myklebust <trond.myklebust@primarydata.com>2015-02-03 16:01:27 -0500
committerTrond Myklebust <trond.myklebust@primarydata.com>2015-02-03 16:01:27 -0500
commite2c63e091e29786a34ecf42c169e627a3d1d96d7 (patch)
tree4e96723b969bd6b3d39613de601f18da678cca71
parentcc3ea893cbb386e657d775f299f61bb9ba3bc435 (diff)
parent7c13789e3e6c66dbcaade1760087429240eb3d27 (diff)
Merge branch 'flexfiles'
* flexfiles: (53 commits) pnfs: lookup new lseg at lseg boundary nfs41: .init_read and .init_write can be called with valid pg_lseg pnfs: Update documentation on the Layout Drivers pnfs/flexfiles: Add the FlexFile Layout Driver nfs: count DIO good bytes correctly with mirroring nfs41: wait for LAYOUTRETURN before retrying LAYOUTGET nfs: add a helper to set NFS_ODIRECT_RESCHED_WRITES to direct writes nfs41: add NFS_LAYOUT_RETRY_LAYOUTGET to layout header flags nfs/flexfiles: send layoutreturn before freeing lseg nfs41: introduce NFS_LAYOUT_RETURN_BEFORE_CLOSE nfs41: allow async version layoutreturn nfs41: add range to layoutreturn args pnfs: allow LD to ask to resend read through pnfs nfs: add nfs_pgio_current_mirror helper nfs: only reset desc->pg_mirror_idx when mirroring is supported nfs41: add a debug warning if we destroy an unempty layout pnfs: fail comparison when bucket verifier not set nfs: mirroring support for direct io nfs: add mirroring support to pgio layer pnfs: pass ds_commit_idx through the commit path ... Conflicts: fs/nfs/pnfs.c fs/nfs/pnfs.h
Diffstat
-rw-r--r--Documentation/filesystems/nfs/pnfs.txt13
-rw-r--r--fs/nfs/Kconfig5
-rw-r--r--fs/nfs/Makefile3
-rw-r--r--fs/nfs/blocklayout/blocklayout.c2
-rw-r--r--fs/nfs/direct.c112
-rw-r--r--fs/nfs/filelayout/filelayout.c315
-rw-r--r--fs/nfs/filelayout/filelayout.h40
-rw-r--r--fs/nfs/filelayout/filelayoutdev.c469
-rw-r--r--fs/nfs/flexfilelayout/Makefile5
-rw-r--r--fs/nfs/flexfilelayout/flexfilelayout.c1574
-rw-r--r--fs/nfs/flexfilelayout/flexfilelayout.h155
-rw-r--r--fs/nfs/flexfilelayout/flexfilelayoutdev.c552
-rw-r--r--fs/nfs/idmap.c3
-rw-r--r--fs/nfs/internal.h31
-rw-r--r--fs/nfs/nfs2xdr.c10
-rw-r--r--fs/nfs/nfs3_fs.h2
-rw-r--r--fs/nfs/nfs3client.c41
-rw-r--r--fs/nfs/nfs3proc.c9
-rw-r--r--fs/nfs/nfs3super.c2
-rw-r--r--fs/nfs/nfs3xdr.c3
-rw-r--r--fs/nfs/nfs4_fs.h6
-rw-r--r--fs/nfs/nfs4client.c7
-rw-r--r--fs/nfs/nfs4proc.c48
-rw-r--r--fs/nfs/nfs4super.c3
-rw-r--r--fs/nfs/nfs4xdr.c9
-rw-r--r--fs/nfs/objlayout/objio_osd.c5
-rw-r--r--fs/nfs/pagelist.c294
-rw-r--r--fs/nfs/pnfs.c471
-rw-r--r--fs/nfs/pnfs.h125
-rw-r--r--fs/nfs/pnfs_nfs.c840
-rw-r--r--fs/nfs/read.c33
-rw-r--r--fs/nfs/write.c52
-rw-r--r--include/linux/nfs4.h1
-rw-r--r--include/linux/nfs_fs_sb.h9
-rw-r--r--include/linux/nfs_idmap.h2
-rw-r--r--include/linux/nfs_page.h22
-rw-r--r--include/linux/nfs_xdr.h6
-rw-r--r--include/linux/sunrpc/metrics.h4
-rw-r--r--net/sunrpc/stats.c26
39 files changed, 4268 insertions, 1041 deletions
diff --git a/Documentation/filesystems/nfs/pnfs.txt b/Documentation/filesystems/nfs/pnfs.txt
index adc81a35fe2d..44a9f2493a88 100644
--- a/Documentation/filesystems/nfs/pnfs.txt
+++ b/Documentation/filesystems/nfs/pnfs.txt
@@ -57,15 +57,16 @@ bit is set, preventing any new lsegs from being added.
57layout drivers 57layout drivers
58-------------- 58--------------
59 59
60PNFS utilizes what is called layout drivers. The STD defines 3 basic 60PNFS utilizes what is called layout drivers. The STD defines 4 basic
61layout types: "files" "objects" and "blocks". For each of these types 61layout types: "files", "objects", "blocks", and "flexfiles". For each
62there is a layout-driver with a common function-vectors table which 62of these types there is a layout-driver with a common function-vectors
63are called by the nfs-client pnfs-core to implement the different layout 63table which are called by the nfs-client pnfs-core to implement the
64types. 64different layout types.
65 65
66Files-layout-driver code is in: fs/nfs/nfs4filelayout.c && nfs4filelayoutdev.c 66Files-layout-driver code is in: fs/nfs/filelayout/.. directory
67Objects-layout-deriver code is in: fs/nfs/objlayout/.. directory 67Objects-layout-deriver code is in: fs/nfs/objlayout/.. directory
68Blocks-layout-deriver code is in: fs/nfs/blocklayout/.. directory 68Blocks-layout-deriver code is in: fs/nfs/blocklayout/.. directory
69Flexfiles-layout-driver code is in: fs/nfs/flexfilelayout/.. directory
69 70
70objects-layout setup 71objects-layout setup
71-------------------- 72--------------------
diff --git a/fs/nfs/Kconfig b/fs/nfs/Kconfig
index 3dece03f2fc8..c7abc10279af 100644
--- a/fs/nfs/Kconfig
+++ b/fs/nfs/Kconfig
@@ -128,6 +128,11 @@ config PNFS_OBJLAYOUT
128 depends on NFS_V4_1 && SCSI_OSD_ULD 128 depends on NFS_V4_1 && SCSI_OSD_ULD
129 default NFS_V4 129 default NFS_V4
130 130
131config PNFS_FLEXFILE_LAYOUT
132 tristate
133 depends on NFS_V4_1 && NFS_V3
134 default m
135
131config NFS_V4_1_IMPLEMENTATION_ID_DOMAIN 136config NFS_V4_1_IMPLEMENTATION_ID_DOMAIN
132 string "NFSv4.1 Implementation ID Domain" 137 string "NFSv4.1 Implementation ID Domain"
133 depends on NFS_V4_1 138 depends on NFS_V4_1
diff --git a/fs/nfs/Makefile b/fs/nfs/Makefile
index 04cb830fa09f..1e987acf20c9 100644
--- a/fs/nfs/Makefile
+++ b/fs/nfs/Makefile
@@ -27,9 +27,10 @@ nfsv4-y := nfs4proc.o nfs4xdr.o nfs4state.o nfs4renewd.o nfs4super.o nfs4file.o
27 dns_resolve.o nfs4trace.o 27 dns_resolve.o nfs4trace.o
28nfsv4-$(CONFIG_NFS_USE_LEGACY_DNS) += cache_lib.o 28nfsv4-$(CONFIG_NFS_USE_LEGACY_DNS) += cache_lib.o
29nfsv4-$(CONFIG_SYSCTL) += nfs4sysctl.o 29nfsv4-$(CONFIG_SYSCTL) += nfs4sysctl.o
30nfsv4-$(CONFIG_NFS_V4_1) += pnfs.o pnfs_dev.o 30nfsv4-$(CONFIG_NFS_V4_1) += pnfs.o pnfs_dev.o pnfs_nfs.o
31nfsv4-$(CONFIG_NFS_V4_2) += nfs42proc.o 31nfsv4-$(CONFIG_NFS_V4_2) += nfs42proc.o
32 32
33obj-$(CONFIG_PNFS_FILE_LAYOUT) += filelayout/ 33obj-$(CONFIG_PNFS_FILE_LAYOUT) += filelayout/
34obj-$(CONFIG_PNFS_OBJLAYOUT) += objlayout/ 34obj-$(CONFIG_PNFS_OBJLAYOUT) += objlayout/
35obj-$(CONFIG_PNFS_BLOCK) += blocklayout/ 35obj-$(CONFIG_PNFS_BLOCK) += blocklayout/
36obj-$(CONFIG_PNFS_FLEXFILE_LAYOUT) += flexfilelayout/
diff --git a/fs/nfs/blocklayout/blocklayout.c b/fs/nfs/blocklayout/blocklayout.c
index 77fec6a55f57..1cac3c175d18 100644
--- a/fs/nfs/blocklayout/blocklayout.c
+++ b/fs/nfs/blocklayout/blocklayout.c
@@ -860,12 +860,14 @@ static const struct nfs_pageio_ops bl_pg_read_ops = {
860 .pg_init = bl_pg_init_read, 860 .pg_init = bl_pg_init_read,
861 .pg_test = bl_pg_test_read, 861 .pg_test = bl_pg_test_read,
862 .pg_doio = pnfs_generic_pg_readpages, 862 .pg_doio = pnfs_generic_pg_readpages,
863 .pg_cleanup = pnfs_generic_pg_cleanup,
863}; 864};
864 865
865static const struct nfs_pageio_ops bl_pg_write_ops = { 866static const struct nfs_pageio_ops bl_pg_write_ops = {
866 .pg_init = bl_pg_init_write, 867 .pg_init = bl_pg_init_write,
867 .pg_test = bl_pg_test_write, 868 .pg_test = bl_pg_test_write,
868 .pg_doio = pnfs_generic_pg_writepages, 869 .pg_doio = pnfs_generic_pg_writepages,
870 .pg_cleanup = pnfs_generic_pg_cleanup,
869}; 871};
870 872
871static struct pnfs_layoutdriver_type blocklayout_type = { 873static struct pnfs_layoutdriver_type blocklayout_type = {
diff --git a/fs/nfs/direct.c b/fs/nfs/direct.c
index 294692ff83b1..7077521acdf4 100644
--- a/fs/nfs/direct.c
+++ b/fs/nfs/direct.c
@@ -66,6 +66,10 @@ static struct kmem_cache *nfs_direct_cachep;
66/* 66/*
67 * This represents a set of asynchronous requests that we're waiting on 67 * This represents a set of asynchronous requests that we're waiting on
68 */ 68 */
69struct nfs_direct_mirror {
70 ssize_t count;
71};
72
69struct nfs_direct_req { 73struct nfs_direct_req {
70 struct kref kref; /* release manager */ 74 struct kref kref; /* release manager */
71 75
@@ -78,8 +82,13 @@ struct nfs_direct_req {
78 /* completion state */ 82 /* completion state */
79 atomic_t io_count; /* i/os we're waiting for */ 83 atomic_t io_count; /* i/os we're waiting for */
80 spinlock_t lock; /* protect completion state */ 84 spinlock_t lock; /* protect completion state */
85
86 struct nfs_direct_mirror mirrors[NFS_PAGEIO_DESCRIPTOR_MIRROR_MAX];
87 int mirror_count;
88
81 ssize_t count, /* bytes actually processed */ 89 ssize_t count, /* bytes actually processed */
82 bytes_left, /* bytes left to be sent */ 90 bytes_left, /* bytes left to be sent */
91 io_start, /* start of IO */
83 error; /* any reported error */ 92 error; /* any reported error */
84 struct completion completion; /* wait for i/o completion */ 93 struct completion completion; /* wait for i/o completion */
85 94
@@ -108,26 +117,56 @@ static inline int put_dreq(struct nfs_direct_req *dreq)
108 return atomic_dec_and_test(&dreq->io_count); 117 return atomic_dec_and_test(&dreq->io_count);
109} 118}
110 119
120void nfs_direct_set_resched_writes(struct nfs_direct_req *dreq)
121{
122 dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
123}
124EXPORT_SYMBOL_GPL(nfs_direct_set_resched_writes);
125
126static void
127nfs_direct_good_bytes(struct nfs_direct_req *dreq, struct nfs_pgio_header *hdr)
128{
129 int i;
130 ssize_t count;
131
132 WARN_ON_ONCE(hdr->pgio_mirror_idx >= dreq->mirror_count);
133
134 count = dreq->mirrors[hdr->pgio_mirror_idx].count;
135 if (count + dreq->io_start < hdr->io_start + hdr->good_bytes) {
136 count = hdr->io_start + hdr->good_bytes - dreq->io_start;
137 dreq->mirrors[hdr->pgio_mirror_idx].count = count;
138 }
139
140 /* update the dreq->count by finding the minimum agreed count from all
141 * mirrors */
142 count = dreq->mirrors[0].count;
143
144 for (i = 1; i < dreq->mirror_count; i++)
145 count = min(count, dreq->mirrors[i].count);
146
147 dreq->count = count;
148}
149
111/* 150/*
112 * nfs_direct_select_verf - select the right verifier 151 * nfs_direct_select_verf - select the right verifier
113 * @dreq - direct request possibly spanning multiple servers 152 * @dreq - direct request possibly spanning multiple servers
114 * @ds_clp - nfs_client of data server or NULL if MDS / non-pnfs 153 * @ds_clp - nfs_client of data server or NULL if MDS / non-pnfs
115 * @ds_idx - index of data server in data server list, only valid if ds_clp set 154 * @commit_idx - commit bucket index for the DS
116 * 155 *
117 * returns the correct verifier to use given the role of the server 156 * returns the correct verifier to use given the role of the server
118 */ 157 */
119static struct nfs_writeverf * 158static struct nfs_writeverf *
120nfs_direct_select_verf(struct nfs_direct_req *dreq, 159nfs_direct_select_verf(struct nfs_direct_req *dreq,
121 struct nfs_client *ds_clp, 160 struct nfs_client *ds_clp,
122 int ds_idx) 161 int commit_idx)
123{ 162{
124 struct nfs_writeverf *verfp = &dreq->verf; 163 struct nfs_writeverf *verfp = &dreq->verf;
125 164
126#ifdef CONFIG_NFS_V4_1 165#ifdef CONFIG_NFS_V4_1
127 if (ds_clp) { 166 if (ds_clp) {
128 /* pNFS is in use, use the DS verf */ 167 /* pNFS is in use, use the DS verf */
129 if (ds_idx >= 0 && ds_idx < dreq->ds_cinfo.nbuckets) 168 if (commit_idx >= 0 && commit_idx < dreq->ds_cinfo.nbuckets)
130 verfp = &dreq->ds_cinfo.buckets[ds_idx].direct_verf; 169 verfp = &dreq->ds_cinfo.buckets[commit_idx].direct_verf;
131 else 170 else
132 WARN_ON_ONCE(1); 171 WARN_ON_ONCE(1);
133 } 172 }
@@ -148,8 +187,7 @@ static void nfs_direct_set_hdr_verf(struct nfs_direct_req *dreq,
148{ 187{
149 struct nfs_writeverf *verfp; 188 struct nfs_writeverf *verfp;
150 189
151 verfp = nfs_direct_select_verf(dreq, hdr->ds_clp, 190 verfp = nfs_direct_select_verf(dreq, hdr->ds_clp, hdr->ds_commit_idx);
152 hdr->ds_idx);
153 WARN_ON_ONCE(verfp->committed >= 0); 191 WARN_ON_ONCE(verfp->committed >= 0);
154 memcpy(verfp, &hdr->verf, sizeof(struct nfs_writeverf)); 192 memcpy(verfp, &hdr->verf, sizeof(struct nfs_writeverf));
155 WARN_ON_ONCE(verfp->committed < 0); 193 WARN_ON_ONCE(verfp->committed < 0);
@@ -169,8 +207,7 @@ static int nfs_direct_set_or_cmp_hdr_verf(struct nfs_direct_req *dreq,
169{ 207{
170 struct nfs_writeverf *verfp; 208 struct nfs_writeverf *verfp;
171 209
172 verfp = nfs_direct_select_verf(dreq, hdr->ds_clp, 210 verfp = nfs_direct_select_verf(dreq, hdr->ds_clp, hdr->ds_commit_idx);
173 hdr->ds_idx);
174 if (verfp->committed < 0) { 211 if (verfp->committed < 0) {
175 nfs_direct_set_hdr_verf(dreq, hdr); 212 nfs_direct_set_hdr_verf(dreq, hdr);
176 return 0; 213 return 0;
@@ -193,7 +230,11 @@ static int nfs_direct_cmp_commit_data_verf(struct nfs_direct_req *dreq,
193 230
194 verfp = nfs_direct_select_verf(dreq, data->ds_clp, 231 verfp = nfs_direct_select_verf(dreq, data->ds_clp,
195 data->ds_commit_index); 232 data->ds_commit_index);
196 WARN_ON_ONCE(verfp->committed < 0); 233
234 /* verifier not set so always fail */
235 if (verfp->committed < 0)
236 return 1;
237
197 return memcmp(verfp, &data->verf, sizeof(struct nfs_writeverf)); 238 return memcmp(verfp, &data->verf, sizeof(struct nfs_writeverf));
198} 239}
199 240
@@ -249,6 +290,18 @@ void nfs_init_cinfo_from_dreq(struct nfs_commit_info *cinfo,
249 cinfo->completion_ops = &nfs_direct_commit_completion_ops; 290 cinfo->completion_ops = &nfs_direct_commit_completion_ops;
250} 291}
251 292
293static inline void nfs_direct_setup_mirroring(struct nfs_direct_req *dreq,
294 struct nfs_pageio_descriptor *pgio,
295 struct nfs_page *req)
296{
297 int mirror_count = 1;
298
299 if (pgio->pg_ops->pg_get_mirror_count)
300 mirror_count = pgio->pg_ops->pg_get_mirror_count(pgio, req);
301
302 dreq->mirror_count = mirror_count;
303}
304
252static inline struct nfs_direct_req *nfs_direct_req_alloc(void) 305static inline struct nfs_direct_req *nfs_direct_req_alloc(void)
253{ 306{
254 struct nfs_direct_req *dreq; 307 struct nfs_direct_req *dreq;
@@ -263,6 +316,7 @@ static inline struct nfs_direct_req *nfs_direct_req_alloc(void)
263 INIT_LIST_HEAD(&dreq->mds_cinfo.list); 316 INIT_LIST_HEAD(&dreq->mds_cinfo.list);
264 dreq->verf.committed = NFS_INVALID_STABLE_HOW; /* not set yet */ 317 dreq->verf.committed = NFS_INVALID_STABLE_HOW; /* not set yet */
265 INIT_WORK(&dreq->work, nfs_direct_write_schedule_work); 318 INIT_WORK(&dreq->work, nfs_direct_write_schedule_work);
319 dreq->mirror_count = 1;
266 spin_lock_init(&dreq->lock); 320 spin_lock_init(&dreq->lock);
267 321
268 return dreq; 322 return dreq;
@@ -369,7 +423,8 @@ static void nfs_direct_read_completion(struct nfs_pgio_header *hdr)
369 if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && (hdr->good_bytes == 0)) 423 if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && (hdr->good_bytes == 0))
370 dreq->error = hdr->error; 424 dreq->error = hdr->error;
371 else 425 else
372 dreq->count += hdr->good_bytes; 426 nfs_direct_good_bytes(dreq, hdr);
427
373 spin_unlock(&dreq->lock); 428 spin_unlock(&dreq->lock);
374 429
375 while (!list_empty(&hdr->pages)) { 430 while (!list_empty(&hdr->pages)) {
@@ -547,6 +602,7 @@ ssize_t nfs_file_direct_read(struct kiocb *iocb, struct iov_iter *iter,
547 602
548 dreq->inode = inode; 603 dreq->inode = inode;
549 dreq->bytes_left = count; 604 dreq->bytes_left = count;
605 dreq->io_start = pos;
550 dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp)); 606 dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp));
551 l_ctx = nfs_get_lock_context(dreq->ctx); 607 l_ctx = nfs_get_lock_context(dreq->ctx);
552 if (IS_ERR(l_ctx)) { 608 if (IS_ERR(l_ctx)) {
@@ -579,6 +635,20 @@ out:
579 return result; 635 return result;
580} 636}
581 637
638static void
639nfs_direct_write_scan_commit_list(struct inode *inode,
640 struct list_head *list,
641 struct nfs_commit_info *cinfo)
642{
643 spin_lock(cinfo->lock);
644#ifdef CONFIG_NFS_V4_1
645 if (cinfo->ds != NULL && cinfo->ds->nwritten != 0)
646 NFS_SERVER(inode)->pnfs_curr_ld->recover_commit_reqs(list, cinfo);
647#endif
648 nfs_scan_commit_list(&cinfo->mds->list, list, cinfo, 0);
649 spin_unlock(cinfo->lock);
650}
651
582static void nfs_direct_write_reschedule(struct nfs_direct_req *dreq) 652static void nfs_direct_write_reschedule(struct nfs_direct_req *dreq)
583{ 653{
584 struct nfs_pageio_descriptor desc; 654 struct nfs_pageio_descriptor desc;
@@ -586,20 +656,23 @@ static void nfs_direct_write_reschedule(struct nfs_direct_req *dreq)
586 LIST_HEAD(reqs); 656 LIST_HEAD(reqs);
587 struct nfs_commit_info cinfo; 657 struct nfs_commit_info cinfo;
588 LIST_HEAD(failed); 658 LIST_HEAD(failed);
659 int i;
589 660
590 nfs_init_cinfo_from_dreq(&cinfo, dreq); 661 nfs_init_cinfo_from_dreq(&cinfo, dreq);
591 pnfs_recover_commit_reqs(dreq->inode, &reqs, &cinfo); 662 nfs_direct_write_scan_commit_list(dreq->inode, &reqs, &cinfo);
592 spin_lock(cinfo.lock);
593 nfs_scan_commit_list(&cinfo.mds->list, &reqs, &cinfo, 0);
594 spin_unlock(cinfo.lock);
595 663
596 dreq->count = 0; 664 dreq->count = 0;
665 for (i = 0; i < dreq->mirror_count; i++)
666 dreq->mirrors[i].count = 0;
597 get_dreq(dreq); 667 get_dreq(dreq);
598 668
599 nfs_pageio_init_write(&desc, dreq->inode, FLUSH_STABLE, false, 669 nfs_pageio_init_write(&desc, dreq->inode, FLUSH_STABLE, false,
600 &nfs_direct_write_completion_ops); 670 &nfs_direct_write_completion_ops);
601 desc.pg_dreq = dreq; 671 desc.pg_dreq = dreq;
602 672
673 req = nfs_list_entry(reqs.next);
674 nfs_direct_setup_mirroring(dreq, &desc, req);
675
603 list_for_each_entry_safe(req, tmp, &reqs, wb_list) { 676 list_for_each_entry_safe(req, tmp, &reqs, wb_list) {
604 if (!nfs_pageio_add_request(&desc, req)) { 677 if (!nfs_pageio_add_request(&desc, req)) {
605 nfs_list_remove_request(req); 678 nfs_list_remove_request(req);
@@ -646,7 +719,7 @@ static void nfs_direct_commit_complete(struct nfs_commit_data *data)
646 nfs_list_remove_request(req); 719 nfs_list_remove_request(req);
647 if (dreq->flags == NFS_ODIRECT_RESCHED_WRITES) { 720 if (dreq->flags == NFS_ODIRECT_RESCHED_WRITES) {
648 /* Note the rewrite will go through mds */ 721 /* Note the rewrite will go through mds */
649 nfs_mark_request_commit(req, NULL, &cinfo); 722 nfs_mark_request_commit(req, NULL, &cinfo, 0);
650 } else 723 } else
651 nfs_release_request(req); 724 nfs_release_request(req);
652 nfs_unlock_and_release_request(req); 725 nfs_unlock_and_release_request(req);
@@ -721,7 +794,7 @@ static void nfs_direct_write_completion(struct nfs_pgio_header *hdr)
721 dreq->error = hdr->error; 794 dreq->error = hdr->error;
722 } 795 }
723 if (dreq->error == 0) { 796 if (dreq->error == 0) {
724 dreq->count += hdr->good_bytes; 797 nfs_direct_good_bytes(dreq, hdr);
725 if (nfs_write_need_commit(hdr)) { 798 if (nfs_write_need_commit(hdr)) {
726 if (dreq->flags == NFS_ODIRECT_RESCHED_WRITES) 799 if (dreq->flags == NFS_ODIRECT_RESCHED_WRITES)
727 request_commit = true; 800 request_commit = true;
@@ -745,7 +818,8 @@ static void nfs_direct_write_completion(struct nfs_pgio_header *hdr)
745 nfs_list_remove_request(req); 818 nfs_list_remove_request(req);
746 if (request_commit) { 819 if (request_commit) {
747 kref_get(&req->wb_kref); 820 kref_get(&req->wb_kref);
748 nfs_mark_request_commit(req, hdr->lseg, &cinfo); 821 nfs_mark_request_commit(req, hdr->lseg, &cinfo,
822 hdr->ds_commit_idx);
749 } 823 }
750 nfs_unlock_and_release_request(req); 824 nfs_unlock_and_release_request(req);
751 } 825 }
@@ -826,6 +900,9 @@ static ssize_t nfs_direct_write_schedule_iovec(struct nfs_direct_req *dreq,
826 result = PTR_ERR(req); 900 result = PTR_ERR(req);
827 break; 901 break;
828 } 902 }
903
904 nfs_direct_setup_mirroring(dreq, &desc, req);
905
829 nfs_lock_request(req); 906 nfs_lock_request(req);
830 req->wb_index = pos >> PAGE_SHIFT; 907 req->wb_index = pos >> PAGE_SHIFT;
831 req->wb_offset = pos & ~PAGE_MASK; 908 req->wb_offset = pos & ~PAGE_MASK;
@@ -934,6 +1011,7 @@ ssize_t nfs_file_direct_write(struct kiocb *iocb, struct iov_iter *iter,
934 1011
935 dreq->inode = inode; 1012 dreq->inode = inode;
936 dreq->bytes_left = count; 1013 dreq->bytes_left = count;
1014 dreq->io_start = pos;
937 dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp)); 1015 dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp));
938 l_ctx = nfs_get_lock_context(dreq->ctx); 1016 l_ctx = nfs_get_lock_context(dreq->ctx);
939 if (IS_ERR(l_ctx)) { 1017 if (IS_ERR(l_ctx)) {
diff --git a/fs/nfs/filelayout/filelayout.c b/fs/nfs/filelayout/filelayout.c
index 7afb52f6a25a..3c9769441f36 100644
--- a/fs/nfs/filelayout/filelayout.c
+++ b/fs/nfs/filelayout/filelayout.c
@@ -118,13 +118,6 @@ static void filelayout_reset_read(struct nfs_pgio_header *hdr)
118 } 118 }
119} 119}
120 120
121static void filelayout_fenceme(struct inode *inode, struct pnfs_layout_hdr *lo)
122{
123 if (!test_and_clear_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
124 return;
125 pnfs_return_layout(inode);
126}
127
128static int filelayout_async_handle_error(struct rpc_task *task, 121static int filelayout_async_handle_error(struct rpc_task *task,
129 struct nfs4_state *state, 122 struct nfs4_state *state,
130 struct nfs_client *clp, 123 struct nfs_client *clp,
@@ -207,7 +200,7 @@ static int filelayout_async_handle_error(struct rpc_task *task,
207 dprintk("%s DS connection error %d\n", __func__, 200 dprintk("%s DS connection error %d\n", __func__,
208 task->tk_status); 201 task->tk_status);
209 nfs4_mark_deviceid_unavailable(devid); 202 nfs4_mark_deviceid_unavailable(devid);
210 set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags); 203 pnfs_error_mark_layout_for_return(inode, lseg);
211 rpc_wake_up(&tbl->slot_tbl_waitq); 204 rpc_wake_up(&tbl->slot_tbl_waitq);
212 /* fall through */ 205 /* fall through */
213 default: 206 default:
@@ -339,16 +332,6 @@ static void filelayout_read_count_stats(struct rpc_task *task, void *data)
339 rpc_count_iostats(task, NFS_SERVER(hdr->inode)->client->cl_metrics); 332 rpc_count_iostats(task, NFS_SERVER(hdr->inode)->client->cl_metrics);
340} 333}
341 334
342static void filelayout_read_release(void *data)
343{
344 struct nfs_pgio_header *hdr = data;
345 struct pnfs_layout_hdr *lo = hdr->lseg->pls_layout;
346
347 filelayout_fenceme(lo->plh_inode, lo);
348 nfs_put_client(hdr->ds_clp);
349 hdr->mds_ops->rpc_release(data);
350}
351
352static int filelayout_write_done_cb(struct rpc_task *task, 335static int filelayout_write_done_cb(struct rpc_task *task,
353 struct nfs_pgio_header *hdr) 336 struct nfs_pgio_header *hdr)
354{ 337{
@@ -371,17 +354,6 @@ static int filelayout_write_done_cb(struct rpc_task *task,
371 return 0; 354 return 0;
372} 355}
373 356
374/* Fake up some data that will cause nfs_commit_release to retry the writes. */
375static void prepare_to_resend_writes(struct nfs_commit_data *data)
376{
377 struct nfs_page *first = nfs_list_entry(data->pages.next);
378
379 data->task.tk_status = 0;
380 memcpy(&data->verf.verifier, &first->wb_verf,
381 sizeof(data->verf.verifier));
382 data->verf.verifier.data[0]++; /* ensure verifier mismatch */
383}
384
385static int filelayout_commit_done_cb(struct rpc_task *task, 357static int filelayout_commit_done_cb(struct rpc_task *task,
386 struct nfs_commit_data *data) 358 struct nfs_commit_data *data)
387{ 359{
@@ -393,7 +365,7 @@ static int filelayout_commit_done_cb(struct rpc_task *task,
393 365
394 switch (err) { 366 switch (err) {
395 case -NFS4ERR_RESET_TO_MDS: 367 case -NFS4ERR_RESET_TO_MDS:
396 prepare_to_resend_writes(data); 368 pnfs_generic_prepare_to_resend_writes(data);
397 return -EAGAIN; 369 return -EAGAIN;
398 case -EAGAIN: 370 case -EAGAIN:
399 rpc_restart_call_prepare(task); 371 rpc_restart_call_prepare(task);
@@ -451,16 +423,6 @@ static void filelayout_write_count_stats(struct rpc_task *task, void *data)
451 rpc_count_iostats(task, NFS_SERVER(hdr->inode)->client->cl_metrics); 423 rpc_count_iostats(task, NFS_SERVER(hdr->inode)->client->cl_metrics);
452} 424}
453 425
454static void filelayout_write_release(void *data)
455{
456 struct nfs_pgio_header *hdr = data;
457 struct pnfs_layout_hdr *lo = hdr->lseg->pls_layout;
458
459 filelayout_fenceme(lo->plh_inode, lo);
460 nfs_put_client(hdr->ds_clp);
461 hdr->mds_ops->rpc_release(data);
462}
463
464static void filelayout_commit_prepare(struct rpc_task *task, void *data) 426static void filelayout_commit_prepare(struct rpc_task *task, void *data)
465{ 427{
466 struct nfs_commit_data *wdata = data; 428 struct nfs_commit_data *wdata = data;
@@ -471,14 +433,6 @@ static void filelayout_commit_prepare(struct rpc_task *task, void *data)
471 task); 433 task);
472} 434}
473 435
474static void filelayout_write_commit_done(struct rpc_task *task, void *data)
475{
476 struct nfs_commit_data *wdata = data;
477
478 /* Note this may cause RPC to be resent */
479 wdata->mds_ops->rpc_call_done(task, data);
480}
481
482static void filelayout_commit_count_stats(struct rpc_task *task, void *data) 436static void filelayout_commit_count_stats(struct rpc_task *task, void *data)
483{ 437{
484 struct nfs_commit_data *cdata = data; 438 struct nfs_commit_data *cdata = data;
@@ -486,35 +440,25 @@ static void filelayout_commit_count_stats(struct rpc_task *task, void *data)
486 rpc_count_iostats(task, NFS_SERVER(cdata->inode)->client->cl_metrics); 440 rpc_count_iostats(task, NFS_SERVER(cdata->inode)->client->cl_metrics);
487} 441}
488 442
489static void filelayout_commit_release(void *calldata)
490{
491 struct nfs_commit_data *data = calldata;
492
493 data->completion_ops->completion(data);
494 pnfs_put_lseg(data->lseg);
495 nfs_put_client(data->ds_clp);
496 nfs_commitdata_release(data);
497}
498
499static const struct rpc_call_ops filelayout_read_call_ops = { 443static const struct rpc_call_ops filelayout_read_call_ops = {
500 .rpc_call_prepare = filelayout_read_prepare, 444 .rpc_call_prepare = filelayout_read_prepare,
501 .rpc_call_done = filelayout_read_call_done, 445 .rpc_call_done = filelayout_read_call_done,
502 .rpc_count_stats = filelayout_read_count_stats, 446 .rpc_count_stats = filelayout_read_count_stats,
503 .rpc_release = filelayout_read_release, 447 .rpc_release = pnfs_generic_rw_release,
504}; 448};
505 449
506static const struct rpc_call_ops filelayout_write_call_ops = { 450static const struct rpc_call_ops filelayout_write_call_ops = {
507 .rpc_call_prepare = filelayout_write_prepare, 451 .rpc_call_prepare = filelayout_write_prepare,
508 .rpc_call_done = filelayout_write_call_done, 452 .rpc_call_done = filelayout_write_call_done,
509 .rpc_count_stats = filelayout_write_count_stats, 453 .rpc_count_stats = filelayout_write_count_stats,
510 .rpc_release = filelayout_write_release, 454 .rpc_release = pnfs_generic_rw_release,
511}; 455};
512 456
513static const struct rpc_call_ops filelayout_commit_call_ops = { 457static const struct rpc_call_ops filelayout_commit_call_ops = {
514 .rpc_call_prepare = filelayout_commit_prepare, 458 .rpc_call_prepare = filelayout_commit_prepare,
515 .rpc_call_done = filelayout_write_commit_done, 459 .rpc_call_done = pnfs_generic_write_commit_done,
516 .rpc_count_stats = filelayout_commit_count_stats, 460 .rpc_count_stats = filelayout_commit_count_stats,
517 .rpc_release = filelayout_commit_release, 461 .rpc_release = pnfs_generic_commit_release,
518}; 462};
519 463
520static enum pnfs_try_status 464static enum pnfs_try_status
@@ -548,7 +492,7 @@ filelayout_read_pagelist(struct nfs_pgio_header *hdr)
548 /* No multipath support. Use first DS */ 492 /* No multipath support. Use first DS */
549 atomic_inc(&ds->ds_clp->cl_count); 493 atomic_inc(&ds->ds_clp->cl_count);
550 hdr->ds_clp = ds->ds_clp; 494 hdr->ds_clp = ds->ds_clp;
551 hdr->ds_idx = idx; 495 hdr->ds_commit_idx = idx;
552 fh = nfs4_fl_select_ds_fh(lseg, j); 496 fh = nfs4_fl_select_ds_fh(lseg, j);
553 if (fh) 497 if (fh)
554 hdr->args.fh = fh; 498 hdr->args.fh = fh;
@@ -557,8 +501,9 @@ filelayout_read_pagelist(struct nfs_pgio_header *hdr)
557 hdr->mds_offset = offset; 501 hdr->mds_offset = offset;
558 502
559 /* Perform an asynchronous read to ds */ 503 /* Perform an asynchronous read to ds */
560 nfs_initiate_pgio(ds_clnt, hdr, 504 nfs_initiate_pgio(ds_clnt, hdr, hdr->cred,
561 &filelayout_read_call_ops, 0, RPC_TASK_SOFTCONN); 505 NFS_PROTO(hdr->inode), &filelayout_read_call_ops,
506 0, RPC_TASK_SOFTCONN);
562 return PNFS_ATTEMPTED; 507 return PNFS_ATTEMPTED;
563} 508}
564 509
@@ -591,16 +536,16 @@ filelayout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
591 hdr->pgio_done_cb = filelayout_write_done_cb; 536 hdr->pgio_done_cb = filelayout_write_done_cb;
592 atomic_inc(&ds->ds_clp->cl_count); 537 atomic_inc(&ds->ds_clp->cl_count);
593 hdr->ds_clp = ds->ds_clp; 538 hdr->ds_clp = ds->ds_clp;
594 hdr->ds_idx = idx; 539 hdr->ds_commit_idx = idx;
595 fh = nfs4_fl_select_ds_fh(lseg, j); 540 fh = nfs4_fl_select_ds_fh(lseg, j);
596 if (fh) 541 if (fh)
597 hdr->args.fh = fh; 542 hdr->args.fh = fh;
598 hdr->args.offset = filelayout_get_dserver_offset(lseg, offset); 543 hdr->args.offset = filelayout_get_dserver_offset(lseg, offset);
599 544
600 /* Perform an asynchronous write */ 545 /* Perform an asynchronous write */
601 nfs_initiate_pgio(ds_clnt, hdr, 546 nfs_initiate_pgio(ds_clnt, hdr, hdr->cred,
602 &filelayout_write_call_ops, sync, 547 NFS_PROTO(hdr->inode), &filelayout_write_call_ops,
603 RPC_TASK_SOFTCONN); 548 sync, RPC_TASK_SOFTCONN);
604 return PNFS_ATTEMPTED; 549 return PNFS_ATTEMPTED;
605} 550}
606 551
@@ -988,12 +933,14 @@ static const struct nfs_pageio_ops filelayout_pg_read_ops = {
988 .pg_init = filelayout_pg_init_read, 933 .pg_init = filelayout_pg_init_read,
989 .pg_test = filelayout_pg_test, 934 .pg_test = filelayout_pg_test,
990 .pg_doio = pnfs_generic_pg_readpages, 935 .pg_doio = pnfs_generic_pg_readpages,
936 .pg_cleanup = pnfs_generic_pg_cleanup,
991}; 937};
992 938
993static const struct nfs_pageio_ops filelayout_pg_write_ops = { 939static const struct nfs_pageio_ops filelayout_pg_write_ops = {
994 .pg_init = filelayout_pg_init_write, 940 .pg_init = filelayout_pg_init_write,
995 .pg_test = filelayout_pg_test, 941 .pg_test = filelayout_pg_test,
996 .pg_doio = pnfs_generic_pg_writepages, 942 .pg_doio = pnfs_generic_pg_writepages,
943 .pg_cleanup = pnfs_generic_pg_cleanup,
997}; 944};
998 945
999static u32 select_bucket_index(struct nfs4_filelayout_segment *fl, u32 j) 946static u32 select_bucket_index(struct nfs4_filelayout_segment *fl, u32 j)
@@ -1004,37 +951,11 @@ static u32 select_bucket_index(struct nfs4_filelayout_segment *fl, u32 j)
1004 return j; 951 return j;
1005} 952}
1006 953
1007/* The generic layer is about to remove the req from the commit list.
1008 * If this will make the bucket empty, it will need to put the lseg reference.
1009 * Note this is must be called holding the inode (/cinfo) lock
1010 */
1011static void
1012filelayout_clear_request_commit(struct nfs_page *req,
1013 struct nfs_commit_info *cinfo)
1014{
1015 struct pnfs_layout_segment *freeme = NULL;
1016
1017 if (!test_and_clear_bit(PG_COMMIT_TO_DS, &req->wb_flags))
1018 goto out;
1019 cinfo->ds->nwritten--;
1020 if (list_is_singular(&req->wb_list)) {
1021 struct pnfs_commit_bucket *bucket;
1022
1023 bucket = list_first_entry(&req->wb_list,
1024 struct pnfs_commit_bucket,
1025 written);
1026 freeme = bucket->wlseg;
1027 bucket->wlseg = NULL;
1028 }
1029out:
1030 nfs_request_remove_commit_list(req, cinfo);
1031 pnfs_put_lseg_locked(freeme);
1032}
1033
1034static void 954static void
1035filelayout_mark_request_commit(struct nfs_page *req, 955filelayout_mark_request_commit(struct nfs_page *req,
1036 struct pnfs_layout_segment *lseg, 956 struct pnfs_layout_segment *lseg,
1037 struct nfs_commit_info *cinfo) 957 struct nfs_commit_info *cinfo,
958 u32 ds_commit_idx)
1038 959
1039{ 960{
1040 struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg); 961 struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
@@ -1064,7 +985,7 @@ filelayout_mark_request_commit(struct nfs_page *req,
1064 * is normally transferred to the COMMIT call and released 985 * is normally transferred to the COMMIT call and released
1065 * there. It could also be released if the last req is pulled 986 * there. It could also be released if the last req is pulled
1066 * off due to a rewrite, in which case it will be done in 987 * off due to a rewrite, in which case it will be done in
1067 * filelayout_clear_request_commit 988 * pnfs_generic_clear_request_commit
1068 */ 989 */
1069 buckets[i].wlseg = pnfs_get_lseg(lseg); 990 buckets[i].wlseg = pnfs_get_lseg(lseg);
1070 } 991 }
@@ -1138,101 +1059,15 @@ static int filelayout_initiate_commit(struct nfs_commit_data *data, int how)
1138 fh = select_ds_fh_from_commit(lseg, data->ds_commit_index); 1059 fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1139 if (fh) 1060 if (fh)
1140 data->args.fh = fh; 1061 data->args.fh = fh;
1141 return nfs_initiate_commit(ds_clnt, data, 1062 return nfs_initiate_commit(ds_clnt, data, NFS_PROTO(data->inode),
1142 &filelayout_commit_call_ops, how, 1063 &filelayout_commit_call_ops, how,
1143 RPC_TASK_SOFTCONN); 1064 RPC_TASK_SOFTCONN);
1144out_err: 1065out_err:
1145 prepare_to_resend_writes(data); 1066 pnfs_generic_prepare_to_resend_writes(data);
1146 filelayout_commit_release(data); 1067 pnfs_generic_commit_release(data);
1147 return -EAGAIN; 1068 return -EAGAIN;
1148} 1069}
1149 1070
1150static int
1151transfer_commit_list(struct list_head *src, struct list_head *dst,
1152 struct nfs_commit_info *cinfo, int max)
1153{
1154 struct nfs_page *req, *tmp;
1155 int ret = 0;
1156
1157 list_for_each_entry_safe(req, tmp, src, wb_list) {
1158 if (!nfs_lock_request(req))
1159 continue;
1160 kref_get(&req->wb_kref);
1161 if (cond_resched_lock(cinfo->lock))
1162 list_safe_reset_next(req, tmp, wb_list);
1163 nfs_request_remove_commit_list(req, cinfo);
1164 clear_bit(PG_COMMIT_TO_DS, &req->wb_flags);
1165 nfs_list_add_request(req, dst);
1166 ret++;
1167 if ((ret == max) && !cinfo->dreq)
1168 break;
1169 }
1170 return ret;
1171}
1172
1173/* Note called with cinfo->lock held. */
1174static int
1175filelayout_scan_ds_commit_list(struct pnfs_commit_bucket *bucket,
1176 struct nfs_commit_info *cinfo,
1177 int max)
1178{
1179 struct list_head *src = &bucket->written;
1180 struct list_head *dst = &bucket->committing;
1181 int ret;
1182
1183 ret = transfer_commit_list(src, dst, cinfo, max);
1184 if (ret) {
1185 cinfo->ds->nwritten -= ret;
1186 cinfo->ds->ncommitting += ret;
1187 bucket->clseg = bucket->wlseg;
1188 if (list_empty(src))
1189 bucket->wlseg = NULL;
1190 else
1191 pnfs_get_lseg(bucket->clseg);
1192 }
1193 return ret;
1194}
1195
1196/* Move reqs from written to committing lists, returning count of number moved.
1197 * Note called with cinfo->lock held.
1198 */
1199static int filelayout_scan_commit_lists(struct nfs_commit_info *cinfo,
1200 int max)
1201{
1202 int i, rv = 0, cnt;
1203
1204 for (i = 0; i < cinfo->ds->nbuckets && max != 0; i++) {
1205 cnt = filelayout_scan_ds_commit_list(&cinfo->ds->buckets[i],
1206 cinfo, max);
1207 max -= cnt;
1208 rv += cnt;
1209 }
1210 return rv;
1211}
1212
1213/* Pull everything off the committing lists and dump into @dst */
1214static void filelayout_recover_commit_reqs(struct list_head *dst,
1215 struct nfs_commit_info *cinfo)
1216{
1217 struct pnfs_commit_bucket *b;
1218 struct pnfs_layout_segment *freeme;
1219 int i;
1220
1221restart:
1222 spin_lock(cinfo->lock);
1223 for (i = 0, b = cinfo->ds->buckets; i < cinfo->ds->nbuckets; i++, b++) {
1224 if (transfer_commit_list(&b->written, dst, cinfo, 0)) {
1225 freeme = b->wlseg;
1226 b->wlseg = NULL;
1227 spin_unlock(cinfo->lock);
1228 pnfs_put_lseg(freeme);
1229 goto restart;
1230 }
1231 }
1232 cinfo->ds->nwritten = 0;
1233 spin_unlock(cinfo->lock);
1234}
1235
1236/* filelayout_search_commit_reqs - Search lists in @cinfo for the head reqest 1071/* filelayout_search_commit_reqs - Search lists in @cinfo for the head reqest
1237 * for @page 1072 * for @page
1238 * @cinfo - commit info for current inode 1073 * @cinfo - commit info for current inode
@@ -1263,108 +1098,14 @@ filelayout_search_commit_reqs(struct nfs_commit_info *cinfo, struct page *page)
1263 return NULL; 1098 return NULL;
1264} 1099}
1265 1100
1266static void filelayout_retry_commit(struct nfs_commit_info *cinfo, int idx)
1267{
1268 struct pnfs_ds_commit_info *fl_cinfo = cinfo->ds;
1269 struct pnfs_commit_bucket *bucket;
1270 struct pnfs_layout_segment *freeme;
1271 int i;
1272
1273 for (i = idx; i < fl_cinfo->nbuckets; i++) {
1274 bucket = &fl_cinfo->buckets[i];
1275 if (list_empty(&bucket->committing))
1276 continue;
1277 nfs_retry_commit(&bucket->committing, bucket->clseg, cinfo);
1278 spin_lock(cinfo->lock);
1279 freeme = bucket->clseg;
1280 bucket->clseg = NULL;
1281 spin_unlock(cinfo->lock);
1282 pnfs_put_lseg(freeme);
1283 }
1284}
1285
1286static unsigned int
1287alloc_ds_commits(struct nfs_commit_info *cinfo, struct list_head *list)
1288{
1289 struct pnfs_ds_commit_info *fl_cinfo;
1290 struct pnfs_commit_bucket *bucket;
1291 struct nfs_commit_data *data;
1292 int i;
1293 unsigned int nreq = 0;
1294
1295 fl_cinfo = cinfo->ds;
1296 bucket = fl_cinfo->buckets;
1297 for (i = 0; i < fl_cinfo->nbuckets; i++, bucket++) {
1298 if (list_empty(&bucket->committing))
1299 continue;
1300 data = nfs_commitdata_alloc();
1301 if (!data)
1302 break;
1303 data->ds_commit_index = i;
1304 spin_lock(cinfo->lock);
1305 data->lseg = bucket->clseg;
1306 bucket->clseg = NULL;
1307 spin_unlock(cinfo->lock);
1308 list_add(&data->pages, list);
1309 nreq++;
1310 }
1311
1312 /* Clean up on error */
1313 filelayout_retry_commit(cinfo, i);
1314 /* Caller will clean up entries put on list */
1315 return nreq;
1316}
1317
1318/* This follows nfs_commit_list pretty closely */
1319static int 1101static int
1320filelayout_commit_pagelist(struct inode *inode, struct list_head *mds_pages, 1102filelayout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1321 int how, struct nfs_commit_info *cinfo) 1103 int how, struct nfs_commit_info *cinfo)
1322{ 1104{
1323 struct nfs_commit_data *data, *tmp; 1105 return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
1324 LIST_HEAD(list); 1106 filelayout_initiate_commit);
1325 unsigned int nreq = 0;
1326
1327 if (!list_empty(mds_pages)) {
1328 data = nfs_commitdata_alloc();
1329 if (data != NULL) {
1330 data->lseg = NULL;
1331 list_add(&data->pages, &list);
1332 nreq++;
1333 } else {
1334 nfs_retry_commit(mds_pages, NULL, cinfo);
1335 filelayout_retry_commit(cinfo, 0);
1336 cinfo->completion_ops->error_cleanup(NFS_I(inode));
1337 return -ENOMEM;
1338 }
1339 }
1340
1341 nreq += alloc_ds_commits(cinfo, &list);
1342
1343 if (nreq == 0) {
1344 cinfo->completion_ops->error_cleanup(NFS_I(inode));
1345 goto out;
1346 }
1347
1348 atomic_add(nreq, &cinfo->mds->rpcs_out);
1349
1350 list_for_each_entry_safe(data, tmp, &list, pages) {
1351 list_del_init(&data->pages);
1352 if (!data->lseg) {
1353 nfs_init_commit(data, mds_pages, NULL, cinfo);
1354 nfs_initiate_commit(NFS_CLIENT(inode), data,
1355 data->mds_ops, how, 0);
1356 } else {
1357 struct pnfs_commit_bucket *buckets;
1358
1359 buckets = cinfo->ds->buckets;
1360 nfs_init_commit(data, &buckets[data->ds_commit_index].committing, data->lseg, cinfo);
1361 filelayout_initiate_commit(data, how);
1362 }
1363 }
1364out:
1365 cinfo->ds->ncommitting = 0;
1366 return PNFS_ATTEMPTED;
1367} 1107}
1108
1368static struct nfs4_deviceid_node * 1109static struct nfs4_deviceid_node *
1369filelayout_alloc_deviceid_node(struct nfs_server *server, 1110filelayout_alloc_deviceid_node(struct nfs_server *server,
1370 struct pnfs_device *pdev, gfp_t gfp_flags) 1111 struct pnfs_device *pdev, gfp_t gfp_flags)
@@ -1421,9 +1162,9 @@ static struct pnfs_layoutdriver_type filelayout_type = {
1421 .pg_write_ops = &filelayout_pg_write_ops, 1162 .pg_write_ops = &filelayout_pg_write_ops,
1422 .get_ds_info = &filelayout_get_ds_info, 1163 .get_ds_info = &filelayout_get_ds_info,
1423 .mark_request_commit = filelayout_mark_request_commit, 1164 .mark_request_commit = filelayout_mark_request_commit,
1424 .clear_request_commit = filelayout_clear_request_commit, 1165 .clear_request_commit = pnfs_generic_clear_request_commit,
1425 .scan_commit_lists = filelayout_scan_commit_lists, 1166 .scan_commit_lists = pnfs_generic_scan_commit_lists,
1426 .recover_commit_reqs = filelayout_recover_commit_reqs, 1167 .recover_commit_reqs = pnfs_generic_recover_commit_reqs,
1427 .search_commit_reqs = filelayout_search_commit_reqs, 1168 .search_commit_reqs = filelayout_search_commit_reqs,
1428 .commit_pagelist = filelayout_commit_pagelist, 1169 .commit_pagelist = filelayout_commit_pagelist,
1429 .read_pagelist = filelayout_read_pagelist, 1170 .read_pagelist = filelayout_read_pagelist,
diff --git a/fs/nfs/filelayout/filelayout.h b/fs/nfs/filelayout/filelayout.h
index 7c9f800c49d7..2896cb833a11 100644
--- a/fs/nfs/filelayout/filelayout.h
+++ b/fs/nfs/filelayout/filelayout.h
@@ -33,13 +33,6 @@
33#include "../pnfs.h" 33#include "../pnfs.h"
34 34
35/* 35/*
36 * Default data server connection timeout and retrans vaules.
37 * Set by module paramters dataserver_timeo and dataserver_retrans.
38 */
39#define NFS4_DEF_DS_TIMEO 600 /* in tenths of a second */
40#define NFS4_DEF_DS_RETRANS 5
41
42/*
43 * Field testing shows we need to support up to 4096 stripe indices. 36 * Field testing shows we need to support up to 4096 stripe indices.
44 * We store each index as a u8 (u32 on the wire) to keep the memory footprint 37 * We store each index as a u8 (u32 on the wire) to keep the memory footprint
45 * reasonable. This in turn means we support a maximum of 256 38 * reasonable. This in turn means we support a maximum of 256
@@ -48,32 +41,11 @@
48#define NFS4_PNFS_MAX_STRIPE_CNT 4096 41#define NFS4_PNFS_MAX_STRIPE_CNT 4096
49#define NFS4_PNFS_MAX_MULTI_CNT 256 /* 256 fit into a u8 stripe_index */ 42#define NFS4_PNFS_MAX_MULTI_CNT 256 /* 256 fit into a u8 stripe_index */
50 43
51/* error codes for internal use */
52#define NFS4ERR_RESET_TO_MDS 12001
53
54enum stripetype4 { 44enum stripetype4 {
55 STRIPE_SPARSE = 1, 45 STRIPE_SPARSE = 1,
56 STRIPE_DENSE = 2 46 STRIPE_DENSE = 2
57}; 47};
58 48
59/* Individual ip address */
60struct nfs4_pnfs_ds_addr {
61 struct sockaddr_storage da_addr;
62 size_t da_addrlen;
63 struct list_head da_node; /* nfs4_pnfs_dev_hlist dev_dslist */
64 char *da_remotestr; /* human readable addr+port */
65};
66
67struct nfs4_pnfs_ds {
68 struct list_head ds_node; /* nfs4_pnfs_dev_hlist dev_dslist */
69 char *ds_remotestr; /* comma sep list of addrs */
70 struct list_head ds_addrs;
71 struct nfs_client *ds_clp;
72 atomic_t ds_count;
73 unsigned long ds_state;
74#define NFS4DS_CONNECTING 0 /* ds is establishing connection */
75};
76
77struct nfs4_file_layout_dsaddr { 49struct nfs4_file_layout_dsaddr {
78 struct nfs4_deviceid_node id_node; 50 struct nfs4_deviceid_node id_node;
79 u32 stripe_count; 51 u32 stripe_count;
@@ -119,17 +91,6 @@ FILELAYOUT_DEVID_NODE(struct pnfs_layout_segment *lseg)
119 return &FILELAYOUT_LSEG(lseg)->dsaddr->id_node; 91 return &FILELAYOUT_LSEG(lseg)->dsaddr->id_node;
120} 92}
121 93
122static inline void
123filelayout_mark_devid_invalid(struct nfs4_deviceid_node *node)
124{
125 u32 *p = (u32 *)&node->deviceid;
126
127 printk(KERN_WARNING "NFS: Deviceid [%x%x%x%x] marked out of use.\n",
128 p[0], p[1], p[2], p[3]);
129
130 set_bit(NFS_DEVICEID_INVALID, &node->flags);
131}
132
133static inline bool 94static inline bool
134filelayout_test_devid_invalid(struct nfs4_deviceid_node *node) 95filelayout_test_devid_invalid(struct nfs4_deviceid_node *node)
135{ 96{
@@ -142,7 +103,6 @@ filelayout_test_devid_unavailable(struct nfs4_deviceid_node *node);
142extern struct nfs_fh * 103extern struct nfs_fh *
143nfs4_fl_select_ds_fh(struct pnfs_layout_segment *lseg, u32 j); 104nfs4_fl_select_ds_fh(struct pnfs_layout_segment *lseg, u32 j);
144 105
145extern void print_ds(struct nfs4_pnfs_ds *ds);
146u32 nfs4_fl_calc_j_index(struct pnfs_layout_segment *lseg, loff_t offset); 106u32 nfs4_fl_calc_j_index(struct pnfs_layout_segment *lseg, loff_t offset);
147u32 nfs4_fl_calc_ds_index(struct pnfs_layout_segment *lseg, u32 j); 107u32 nfs4_fl_calc_ds_index(struct pnfs_layout_segment *lseg, u32 j);
148struct nfs4_pnfs_ds *nfs4_fl_prepare_ds(struct pnfs_layout_segment *lseg, 108struct nfs4_pnfs_ds *nfs4_fl_prepare_ds(struct pnfs_layout_segment *lseg,
diff --git a/fs/nfs/filelayout/filelayoutdev.c b/fs/nfs/filelayout/filelayoutdev.c
index bfecac781f19..4f372e224603 100644
--- a/fs/nfs/filelayout/filelayoutdev.c
+++ b/fs/nfs/filelayout/filelayoutdev.c
@@ -31,7 +31,6 @@
31#include <linux/nfs_fs.h> 31#include <linux/nfs_fs.h>
32#include <linux/vmalloc.h> 32#include <linux/vmalloc.h>
33#include <linux/module.h> 33#include <linux/module.h>
34#include <linux/sunrpc/addr.h>
35 34
36#include "../internal.h" 35#include "../internal.h"
37#include "../nfs4session.h" 36#include "../nfs4session.h"
@@ -42,183 +41,6 @@
42static unsigned int dataserver_timeo = NFS4_DEF_DS_TIMEO; 41static unsigned int dataserver_timeo = NFS4_DEF_DS_TIMEO;
43static unsigned int dataserver_retrans = NFS4_DEF_DS_RETRANS; 42static unsigned int dataserver_retrans = NFS4_DEF_DS_RETRANS;
44 43
45/*
46 * Data server cache
47 *
48 * Data servers can be mapped to different device ids.
49 * nfs4_pnfs_ds reference counting
50 * - set to 1 on allocation
51 * - incremented when a device id maps a data server already in the cache.
52 * - decremented when deviceid is removed from the cache.
53 */
54static DEFINE_SPINLOCK(nfs4_ds_cache_lock);
55static LIST_HEAD(nfs4_data_server_cache);
56
57/* Debug routines */
58void
59print_ds(struct nfs4_pnfs_ds *ds)
60{
61 if (ds == NULL) {
62 printk("%s NULL device\n", __func__);
63 return;
64 }
65 printk(" ds %s\n"
66 " ref count %d\n"
67 " client %p\n"
68 " cl_exchange_flags %x\n",
69 ds->ds_remotestr,
70 atomic_read(&ds->ds_count), ds->ds_clp,
71 ds->ds_clp ? ds->ds_clp->cl_exchange_flags : 0);
72}
73
74static bool
75same_sockaddr(struct sockaddr *addr1, struct sockaddr *addr2)
76{
77 struct sockaddr_in *a, *b;
78 struct sockaddr_in6 *a6, *b6;
79
80 if (addr1->sa_family != addr2->sa_family)
81 return false;
82
83 switch (addr1->sa_family) {
84 case AF_INET:
85 a = (struct sockaddr_in *)addr1;
86 b = (struct sockaddr_in *)addr2;
87
88 if (a->sin_addr.s_addr == b->sin_addr.s_addr &&
89 a->sin_port == b->sin_port)
90 return true;
91 break;
92
93 case AF_INET6:
94 a6 = (struct sockaddr_in6 *)addr1;
95 b6 = (struct sockaddr_in6 *)addr2;
96
97 /* LINKLOCAL addresses must have matching scope_id */
98 if (ipv6_addr_src_scope(&a6->sin6_addr) ==
99 IPV6_ADDR_SCOPE_LINKLOCAL &&
100 a6->sin6_scope_id != b6->sin6_scope_id)
101 return false;
102
103 if (ipv6_addr_equal(&a6->sin6_addr, &b6->sin6_addr) &&
104 a6->sin6_port == b6->sin6_port)
105 return true;
106 break;
107
108 default:
109 dprintk("%s: unhandled address family: %u\n",
110 __func__, addr1->sa_family);
111 return false;
112 }
113
114 return false;
115}
116
117static bool
118_same_data_server_addrs_locked(const struct list_head *dsaddrs1,
119 const struct list_head *dsaddrs2)
120{
121 struct nfs4_pnfs_ds_addr *da1, *da2;
122
123 /* step through both lists, comparing as we go */
124 for (da1 = list_first_entry(dsaddrs1, typeof(*da1), da_node),
125 da2 = list_first_entry(dsaddrs2, typeof(*da2), da_node);
126 da1 != NULL && da2 != NULL;
127 da1 = list_entry(da1->da_node.next, typeof(*da1), da_node),
128 da2 = list_entry(da2->da_node.next, typeof(*da2), da_node)) {
129 if (!same_sockaddr((struct sockaddr *)&da1->da_addr,
130 (struct sockaddr *)&da2->da_addr))
131 return false;
132 }
133 if (da1 == NULL && da2 == NULL)
134 return true;
135
136 return false;
137}
138
139/*
140 * Lookup DS by addresses. nfs4_ds_cache_lock is held
141 */
142static struct nfs4_pnfs_ds *
143_data_server_lookup_locked(const struct list_head *dsaddrs)
144{
145 struct nfs4_pnfs_ds *ds;
146
147 list_for_each_entry(ds, &nfs4_data_server_cache, ds_node)
148 if (_same_data_server_addrs_locked(&ds->ds_addrs, dsaddrs))
149 return ds;
150 return NULL;
151}
152
153/*
154 * Create an rpc connection to the nfs4_pnfs_ds data server
155 * Currently only supports IPv4 and IPv6 addresses
156 */
157static int
158nfs4_ds_connect(struct nfs_server *mds_srv, struct nfs4_pnfs_ds *ds)
159{
160 struct nfs_client *clp = ERR_PTR(-EIO);
161 struct nfs4_pnfs_ds_addr *da;
162 int status = 0;
163
164 dprintk("--> %s DS %s au_flavor %d\n", __func__, ds->ds_remotestr,
165 mds_srv->nfs_client->cl_rpcclient->cl_auth->au_flavor);
166
167 list_for_each_entry(da, &ds->ds_addrs, da_node) {
168 dprintk("%s: DS %s: trying address %s\n",
169 __func__, ds->ds_remotestr, da->da_remotestr);
170
171 clp = nfs4_set_ds_client(mds_srv->nfs_client,
172 (struct sockaddr *)&da->da_addr,
173 da->da_addrlen, IPPROTO_TCP,
174 dataserver_timeo, dataserver_retrans);
175 if (!IS_ERR(clp))
176 break;
177 }
178
179 if (IS_ERR(clp)) {
180 status = PTR_ERR(clp);
181 goto out;
182 }
183
184 status = nfs4_init_ds_session(clp, mds_srv->nfs_client->cl_lease_time);
185 if (status)
186 goto out_put;
187
188 smp_wmb();
189 ds->ds_clp = clp;
190 dprintk("%s [new] addr: %s\n", __func__, ds->ds_remotestr);
191out:
192 return status;
193out_put:
194 nfs_put_client(clp);
195 goto out;
196}
197
198static void
199destroy_ds(struct nfs4_pnfs_ds *ds)
200{
201 struct nfs4_pnfs_ds_addr *da;
202
203 dprintk("--> %s\n", __func__);
204 ifdebug(FACILITY)
205 print_ds(ds);
206
207 nfs_put_client(ds->ds_clp);
208
209 while (!list_empty(&ds->ds_addrs)) {
210 da = list_first_entry(&ds->ds_addrs,
211 struct nfs4_pnfs_ds_addr,
212 da_node);
213 list_del_init(&da->da_node);
214 kfree(da->da_remotestr);
215 kfree(da);
216 }
217
218 kfree(ds->ds_remotestr);
219 kfree(ds);
220}
221
222void 44void
223nfs4_fl_free_deviceid(struct nfs4_file_layout_dsaddr *dsaddr) 45nfs4_fl_free_deviceid(struct nfs4_file_layout_dsaddr *dsaddr)
224{ 46{
@@ -229,259 +51,13 @@ nfs4_fl_free_deviceid(struct nfs4_file_layout_dsaddr *dsaddr)
229 51
230 for (i = 0; i < dsaddr->ds_num; i++) { 52 for (i = 0; i < dsaddr->ds_num; i++) {
231 ds = dsaddr->ds_list[i]; 53 ds = dsaddr->ds_list[i];
232 if (ds != NULL) { 54 if (ds != NULL)
233 if (atomic_dec_and_lock(&ds->ds_count, 55 nfs4_pnfs_ds_put(ds);
234 &nfs4_ds_cache_lock)) {
235 list_del_init(&ds->ds_node);
236 spin_unlock(&nfs4_ds_cache_lock);
237 destroy_ds(ds);
238 }
239 }
240 } 56 }
241 kfree(dsaddr->stripe_indices); 57 kfree(dsaddr->stripe_indices);
242 kfree(dsaddr); 58 kfree(dsaddr);
243} 59}
244 60
245/*
246 * Create a string with a human readable address and port to avoid
247 * complicated setup around many dprinks.
248 */
249static char *
250nfs4_pnfs_remotestr(struct list_head *dsaddrs, gfp_t gfp_flags)
251{
252 struct nfs4_pnfs_ds_addr *da;
253 char *remotestr;
254 size_t len;
255 char *p;
256
257 len = 3; /* '{', '}' and eol */
258 list_for_each_entry(da, dsaddrs, da_node) {
259 len += strlen(da->da_remotestr) + 1; /* string plus comma */
260 }
261
262 remotestr = kzalloc(len, gfp_flags);
263 if (!remotestr)
264 return NULL;
265
266 p = remotestr;
267 *(p++) = '{';
268 len--;
269 list_for_each_entry(da, dsaddrs, da_node) {
270 size_t ll = strlen(da->da_remotestr);
271
272 if (ll > len)
273 goto out_err;
274
275 memcpy(p, da->da_remotestr, ll);
276 p += ll;
277 len -= ll;
278
279 if (len < 1)
280 goto out_err;
281 (*p++) = ',';
282 len--;
283 }
284 if (len < 2)
285 goto out_err;
286 *(p++) = '}';
287 *p = '\0';
288 return remotestr;
289out_err:
290 kfree(remotestr);
291 return NULL;
292}
293
294static struct nfs4_pnfs_ds *
295nfs4_pnfs_ds_add(struct list_head *dsaddrs, gfp_t gfp_flags)
296{
297 struct nfs4_pnfs_ds *tmp_ds, *ds = NULL;
298 char *remotestr;
299
300 if (list_empty(dsaddrs)) {
301 dprintk("%s: no addresses defined\n", __func__);
302 goto out;
303 }
304
305 ds = kzalloc(sizeof(*ds), gfp_flags);
306 if (!ds)
307 goto out;
308
309 /* this is only used for debugging, so it's ok if its NULL */
310 remotestr = nfs4_pnfs_remotestr(dsaddrs, gfp_flags);
311
312 spin_lock(&nfs4_ds_cache_lock);
313 tmp_ds = _data_server_lookup_locked(dsaddrs);
314 if (tmp_ds == NULL) {
315 INIT_LIST_HEAD(&ds->ds_addrs);
316 list_splice_init(dsaddrs, &ds->ds_addrs);
317 ds->ds_remotestr = remotestr;
318 atomic_set(&ds->ds_count, 1);
319 INIT_LIST_HEAD(&ds->ds_node);
320 ds->ds_clp = NULL;
321 list_add(&ds->ds_node, &nfs4_data_server_cache);
322 dprintk("%s add new data server %s\n", __func__,
323 ds->ds_remotestr);
324 } else {
325 kfree(remotestr);
326 kfree(ds);
327 atomic_inc(&tmp_ds->ds_count);
328 dprintk("%s data server %s found, inc'ed ds_count to %d\n",
329 __func__, tmp_ds->ds_remotestr,
330 atomic_read(&tmp_ds->ds_count));
331 ds = tmp_ds;
332 }
333 spin_unlock(&nfs4_ds_cache_lock);
334out:
335 return ds;
336}
337
338/*
339 * Currently only supports ipv4, ipv6 and one multi-path address.
340 */
341static struct nfs4_pnfs_ds_addr *
342decode_ds_addr(struct net *net, struct xdr_stream *streamp, gfp_t gfp_flags)
343{
344 struct nfs4_pnfs_ds_addr *da = NULL;
345 char *buf, *portstr;
346 __be16 port;
347 int nlen, rlen;
348 int tmp[2];
349 __be32 *p;
350 char *netid, *match_netid;
351 size_t len, match_netid_len;
352 char *startsep = "";
353 char *endsep = "";
354
355
356 /* r_netid */
357 p = xdr_inline_decode(streamp, 4);
358 if (unlikely(!p))
359 goto out_err;
360 nlen = be32_to_cpup(p++);
361
362 p = xdr_inline_decode(streamp, nlen);
363 if (unlikely(!p))
364 goto out_err;
365
366 netid = kmalloc(nlen+1, gfp_flags);
367 if (unlikely(!netid))
368 goto out_err;
369
370 netid[nlen] = '\0';
371 memcpy(netid, p, nlen);
372
373 /* r_addr: ip/ip6addr with port in dec octets - see RFC 5665 */
374 p = xdr_inline_decode(streamp, 4);
375 if (unlikely(!p))
376 goto out_free_netid;
377 rlen = be32_to_cpup(p);
378
379 p = xdr_inline_decode(streamp, rlen);
380 if (unlikely(!p))
381 goto out_free_netid;
382
383 /* port is ".ABC.DEF", 8 chars max */
384 if (rlen > INET6_ADDRSTRLEN + IPV6_SCOPE_ID_LEN + 8) {
385 dprintk("%s: Invalid address, length %d\n", __func__,
386 rlen);
387 goto out_free_netid;
388 }
389 buf = kmalloc(rlen + 1, gfp_flags);
390 if (!buf) {
391 dprintk("%s: Not enough memory\n", __func__);
392 goto out_free_netid;
393 }
394 buf[rlen] = '\0';
395 memcpy(buf, p, rlen);
396
397 /* replace port '.' with '-' */
398 portstr = strrchr(buf, '.');
399 if (!portstr) {
400 dprintk("%s: Failed finding expected dot in port\n",
401 __func__);
402 goto out_free_buf;
403 }
404 *portstr = '-';
405
406 /* find '.' between address and port */
407 portstr = strrchr(buf, '.');
408 if (!portstr) {
409 dprintk("%s: Failed finding expected dot between address and "
410 "port\n", __func__);
411 goto out_free_buf;
412 }
413 *portstr = '\0';
414
415 da = kzalloc(sizeof(*da), gfp_flags);
416 if (unlikely(!da))
417 goto out_free_buf;
418
419 INIT_LIST_HEAD(&da->da_node);
420
421 if (!rpc_pton(net, buf, portstr-buf, (struct sockaddr *)&da->da_addr,
422 sizeof(da->da_addr))) {
423 dprintk("%s: error parsing address %s\n", __func__, buf);
424 goto out_free_da;
425 }
426
427 portstr++;
428 sscanf(portstr, "%d-%d", &tmp[0], &tmp[1]);
429 port = htons((tmp[0] << 8) | (tmp[1]));
430
431 switch (da->da_addr.ss_family) {
432 case AF_INET:
433 ((struct sockaddr_in *)&da->da_addr)->sin_port = port;
434 da->da_addrlen = sizeof(struct sockaddr_in);
435 match_netid = "tcp";
436 match_netid_len = 3;
437 break;
438
439 case AF_INET6:
440 ((struct sockaddr_in6 *)&da->da_addr)->sin6_port = port;
441 da->da_addrlen = sizeof(struct sockaddr_in6);
442 match_netid = "tcp6";
443 match_netid_len = 4;
444 startsep = "[";
445 endsep = "]";
446 break;
447
448 default:
449 dprintk("%s: unsupported address family: %u\n",
450 __func__, da->da_addr.ss_family);
451 goto out_free_da;
452 }
453
454 if (nlen != match_netid_len || strncmp(netid, match_netid, nlen)) {
455 dprintk("%s: ERROR: r_netid \"%s\" != \"%s\"\n",
456 __func__, netid, match_netid);
457 goto out_free_da;
458 }
459
460 /* save human readable address */
461 len = strlen(startsep) + strlen(buf) + strlen(endsep) + 7;
462 da->da_remotestr = kzalloc(len, gfp_flags);
463
464 /* NULL is ok, only used for dprintk */
465 if (da->da_remotestr)
466 snprintf(da->da_remotestr, len, "%s%s%s:%u", startsep,
467 buf, endsep, ntohs(port));
468
469 dprintk("%s: Parsed DS addr %s\n", __func__, da->da_remotestr);
470 kfree(buf);
471 kfree(netid);
472 return da;
473
474out_free_da:
475 kfree(da);
476out_free_buf:
477 dprintk("%s: Error parsing DS addr: %s\n", __func__, buf);
478 kfree(buf);
479out_free_netid:
480 kfree(netid);
481out_err:
482 return NULL;
483}
484
485/* Decode opaque device data and return the result */ 61/* Decode opaque device data and return the result */
486struct nfs4_file_layout_dsaddr * 62struct nfs4_file_layout_dsaddr *
487nfs4_fl_alloc_deviceid_node(struct nfs_server *server, struct pnfs_device *pdev, 63nfs4_fl_alloc_deviceid_node(struct nfs_server *server, struct pnfs_device *pdev,
@@ -584,8 +160,8 @@ nfs4_fl_alloc_deviceid_node(struct nfs_server *server, struct pnfs_device *pdev,
584 160
585 mp_count = be32_to_cpup(p); /* multipath count */ 161 mp_count = be32_to_cpup(p); /* multipath count */
586 for (j = 0; j < mp_count; j++) { 162 for (j = 0; j < mp_count; j++) {
587 da = decode_ds_addr(server->nfs_client->cl_net, 163 da = nfs4_decode_mp_ds_addr(server->nfs_client->cl_net,
588 &stream, gfp_flags); 164 &stream, gfp_flags);
589 if (da) 165 if (da)
590 list_add_tail(&da->da_node, &dsaddrs); 166 list_add_tail(&da->da_node, &dsaddrs);
591 } 167 }
@@ -681,22 +257,7 @@ nfs4_fl_select_ds_fh(struct pnfs_layout_segment *lseg, u32 j)
681 return flseg->fh_array[i]; 257 return flseg->fh_array[i];
682} 258}
683 259
684static void nfs4_wait_ds_connect(struct nfs4_pnfs_ds *ds) 260/* Upon return, either ds is connected, or ds is NULL */
685{
686 might_sleep();
687 wait_on_bit_action(&ds->ds_state, NFS4DS_CONNECTING,
688 nfs_wait_bit_killable, TASK_KILLABLE);
689}
690
691static void nfs4_clear_ds_conn_bit(struct nfs4_pnfs_ds *ds)
692{
693 smp_mb__before_atomic();
694 clear_bit(NFS4DS_CONNECTING, &ds->ds_state);
695 smp_mb__after_atomic();
696 wake_up_bit(&ds->ds_state, NFS4DS_CONNECTING);
697}
698
699
700struct nfs4_pnfs_ds * 261struct nfs4_pnfs_ds *
701nfs4_fl_prepare_ds(struct pnfs_layout_segment *lseg, u32 ds_idx) 262nfs4_fl_prepare_ds(struct pnfs_layout_segment *lseg, u32 ds_idx)
702{ 263{
@@ -704,29 +265,23 @@ nfs4_fl_prepare_ds(struct pnfs_layout_segment *lseg, u32 ds_idx)
704 struct nfs4_pnfs_ds *ds = dsaddr->ds_list[ds_idx]; 265 struct nfs4_pnfs_ds *ds = dsaddr->ds_list[ds_idx];
705 struct nfs4_deviceid_node *devid = FILELAYOUT_DEVID_NODE(lseg); 266 struct nfs4_deviceid_node *devid = FILELAYOUT_DEVID_NODE(lseg);
706 struct nfs4_pnfs_ds *ret = ds; 267 struct nfs4_pnfs_ds *ret = ds;
268 struct nfs_server *s = NFS_SERVER(lseg->pls_layout->plh_inode);
707 269
708 if (ds == NULL) { 270 if (ds == NULL) {
709 printk(KERN_ERR "NFS: %s: No data server for offset index %d\n", 271 printk(KERN_ERR "NFS: %s: No data server for offset index %d\n",
710 __func__, ds_idx); 272 __func__, ds_idx);
711 filelayout_mark_devid_invalid(devid); 273 pnfs_generic_mark_devid_invalid(devid);
712 goto out; 274 goto out;
713 } 275 }
714 smp_rmb(); 276 smp_rmb();
715 if (ds->ds_clp) 277 if (ds->ds_clp)
716 goto out_test_devid; 278 goto out_test_devid;
717 279
718 if (test_and_set_bit(NFS4DS_CONNECTING, &ds->ds_state) == 0) { 280 nfs4_pnfs_ds_connect(s, ds, devid, dataserver_timeo,
719 struct nfs_server *s = NFS_SERVER(lseg->pls_layout->plh_inode); 281 dataserver_retrans, 4,
720 int err; 282 s->nfs_client->cl_minorversion,
721 283 s->nfs_client->cl_rpcclient->cl_auth->au_flavor);
722 err = nfs4_ds_connect(s, ds); 284
723 if (err)
724 nfs4_mark_deviceid_unavailable(devid);
725 nfs4_clear_ds_conn_bit(ds);
726 } else {
727 /* Either ds is connected, or ds is NULL */
728 nfs4_wait_ds_connect(ds);
729 }
730out_test_devid: 285out_test_devid:
731 if (filelayout_test_devid_unavailable(devid)) 286 if (filelayout_test_devid_unavailable(devid))
732 ret = NULL; 287 ret = NULL;
diff --git a/fs/nfs/flexfilelayout/Makefile b/fs/nfs/flexfilelayout/Makefile
new file mode 100644
index 000000000000..1d2c9f6bbcd4
--- /dev/null
+++ b/fs/nfs/flexfilelayout/Makefile
@@ -0,0 +1,5 @@
1#
2# Makefile for the pNFS Flexfile Layout Driver kernel module
3#
4obj-$(CONFIG_PNFS_FLEXFILE_LAYOUT) += nfs_layout_flexfiles.o
5nfs_layout_flexfiles-y := flexfilelayout.o flexfilelayoutdev.o
diff --git a/fs/nfs/flexfilelayout/flexfilelayout.c b/fs/nfs/flexfilelayout/flexfilelayout.c
new file mode 100644
index 000000000000..f29fb7d7e8f8
--- /dev/null
+++ b/fs/nfs/flexfilelayout/flexfilelayout.c
@@ -0,0 +1,1574 @@
1/*
2 * Module for pnfs flexfile layout driver.
3 *
4 * Copyright (c) 2014, Primary Data, Inc. All rights reserved.
5 *
6 * Tao Peng <bergwolf@primarydata.com>
7 */
8
9#include <linux/nfs_fs.h>
10#include <linux/nfs_page.h>
11#include <linux/module.h>
12
13#include <linux/sunrpc/metrics.h>
14#include <linux/nfs_idmap.h>
15
16#include "flexfilelayout.h"
17#include "../nfs4session.h"
18#include "../internal.h"
19#include "../delegation.h"
20#include "../nfs4trace.h"
21#include "../iostat.h"
22#include "../nfs.h"
23
24#define NFSDBG_FACILITY NFSDBG_PNFS_LD
25
26#define FF_LAYOUT_POLL_RETRY_MAX (15*HZ)
27
28static struct pnfs_layout_hdr *
29ff_layout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
30{
31 struct nfs4_flexfile_layout *ffl;
32
33 ffl = kzalloc(sizeof(*ffl), gfp_flags);
34 if (ffl) {
35 INIT_LIST_HEAD(&ffl->error_list);
36 return &ffl->generic_hdr;
37 } else
38 return NULL;
39}
40
41static void
42ff_layout_free_layout_hdr(struct pnfs_layout_hdr *lo)
43{
44 struct nfs4_ff_layout_ds_err *err, *n;
45
46 list_for_each_entry_safe(err, n, &FF_LAYOUT_FROM_HDR(lo)->error_list,
47 list) {
48 list_del(&err->list);
49 kfree(err);
50 }
51 kfree(FF_LAYOUT_FROM_HDR(lo));
52}
53
54static int decode_stateid(struct xdr_stream *xdr, nfs4_stateid *stateid)
55{
56 __be32 *p;
57
58 p = xdr_inline_decode(xdr, NFS4_STATEID_SIZE);
59 if (unlikely(p == NULL))
60 return -ENOBUFS;
61 memcpy(stateid, p, NFS4_STATEID_SIZE);
62 dprintk("%s: stateid id= [%x%x%x%x]\n", __func__,
63 p[0], p[1], p[2], p[3]);
64 return 0;
65}
66
67static int decode_deviceid(struct xdr_stream *xdr, struct nfs4_deviceid *devid)
68{
69 __be32 *p;
70
71 p = xdr_inline_decode(xdr, NFS4_DEVICEID4_SIZE);
72 if (unlikely(!p))
73 return -ENOBUFS;
74 memcpy(devid, p, NFS4_DEVICEID4_SIZE);
75 nfs4_print_deviceid(devid);
76 return 0;
77}
78
79static int decode_nfs_fh(struct xdr_stream *xdr, struct nfs_fh *fh)
80{
81 __be32 *p;
82
83 p = xdr_inline_decode(xdr, 4);
84 if (unlikely(!p))
85 return -ENOBUFS;
86 fh->size = be32_to_cpup(p++);
87 if (fh->size > sizeof(struct nfs_fh)) {
88 printk(KERN_ERR "NFS flexfiles: Too big fh received %d\n",
89 fh->size);
90 return -EOVERFLOW;
91 }
92 /* fh.data */
93 p = xdr_inline_decode(xdr, fh->size);
94 if (unlikely(!p))
95 return -ENOBUFS;
96 memcpy(&fh->data, p, fh->size);
97 dprintk("%s: fh len %d\n", __func__, fh->size);
98
99 return 0;
100}
101
102/*
103 * Currently only stringified uids and gids are accepted.
104 * I.e., kerberos is not supported to the DSes, so no pricipals.
105 *
106 * That means that one common function will suffice, but when
107 * principals are added, this should be split to accomodate
108 * calls to both nfs_map_name_to_uid() and nfs_map_group_to_gid().
109 */
110static int
111decode_name(struct xdr_stream *xdr, u32 *id)
112{
113 __be32 *p;
114 int len;
115
116 /* opaque_length(4)*/
117 p = xdr_inline_decode(xdr, 4);
118 if (unlikely(!p))
119 return -ENOBUFS;
120 len = be32_to_cpup(p++);
121 if (len < 0)
122 return -EINVAL;
123
124 dprintk("%s: len %u\n", __func__, len);
125
126 /* opaque body */
127 p = xdr_inline_decode(xdr, len);
128 if (unlikely(!p))
129 return -ENOBUFS;
130
131 if (!nfs_map_string_to_numeric((char *)p, len, id))
132 return -EINVAL;
133
134 return 0;
135}
136
137static void ff_layout_free_mirror_array(struct nfs4_ff_layout_segment *fls)
138{
139 int i;
140
141 if (fls->mirror_array) {
142 for (i = 0; i < fls->mirror_array_cnt; i++) {
143 /* normally mirror_ds is freed in
144 * .free_deviceid_node but we still do it here
145 * for .alloc_lseg error path */
146 if (fls->mirror_array[i]) {
147 kfree(fls->mirror_array[i]->fh_versions);
148 nfs4_ff_layout_put_deviceid(fls->mirror_array[i]->mirror_ds);
149 kfree(fls->mirror_array[i]);
150 }
151 }
152 kfree(fls->mirror_array);
153 fls->mirror_array = NULL;
154 }
155}
156
157static int ff_layout_check_layout(struct nfs4_layoutget_res *lgr)
158{
159 int ret = 0;
160
161 dprintk("--> %s\n", __func__);
162
163 /* FIXME: remove this check when layout segment support is added */
164 if (lgr->range.offset != 0 ||
165 lgr->range.length != NFS4_MAX_UINT64) {
166 dprintk("%s Only whole file layouts supported. Use MDS i/o\n",
167 __func__);
168 ret = -EINVAL;
169 }
170
171 dprintk("--> %s returns %d\n", __func__, ret);
172 return ret;
173}
174
175static void _ff_layout_free_lseg(struct nfs4_ff_layout_segment *fls)
176{
177 if (fls) {
178 ff_layout_free_mirror_array(fls);
179 kfree(fls);
180 }
181}
182
183static void ff_layout_sort_mirrors(struct nfs4_ff_layout_segment *fls)
184{
185 struct nfs4_ff_layout_mirror *tmp;
186 int i, j;
187
188 for (i = 0; i < fls->mirror_array_cnt - 1; i++) {
189 for (j = i + 1; j < fls->mirror_array_cnt; j++)
190 if (fls->mirror_array[i]->efficiency <
191 fls->mirror_array[j]->efficiency) {
192 tmp = fls->mirror_array[i];
193 fls->mirror_array[i] = fls->mirror_array[j];
194 fls->mirror_array[j] = tmp;
195 }
196 }
197}
198
199static struct pnfs_layout_segment *
200ff_layout_alloc_lseg(struct pnfs_layout_hdr *lh,
201 struct nfs4_layoutget_res *lgr,
202 gfp_t gfp_flags)
203{
204 struct pnfs_layout_segment *ret;
205 struct nfs4_ff_layout_segment *fls = NULL;
206 struct xdr_stream stream;
207 struct xdr_buf buf;
208 struct page *scratch;
209 u64 stripe_unit;
210 u32 mirror_array_cnt;
211 __be32 *p;
212 int i, rc;
213
214 dprintk("--> %s\n", __func__);
215 scratch = alloc_page(gfp_flags);
216 if (!scratch)
217 return ERR_PTR(-ENOMEM);
218
219 xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages,
220 lgr->layoutp->len);
221 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
222
223 /* stripe unit and mirror_array_cnt */
224 rc = -EIO;
225 p = xdr_inline_decode(&stream, 8 + 4);
226 if (!p)
227 goto out_err_free;
228
229 p = xdr_decode_hyper(p, &stripe_unit);
230 mirror_array_cnt = be32_to_cpup(p++);
231 dprintk("%s: stripe_unit=%llu mirror_array_cnt=%u\n", __func__,
232 stripe_unit, mirror_array_cnt);
233
234 if (mirror_array_cnt > NFS4_FLEXFILE_LAYOUT_MAX_MIRROR_CNT ||
235 mirror_array_cnt == 0)
236 goto out_err_free;
237
238 rc = -ENOMEM;
239 fls = kzalloc(sizeof(*fls), gfp_flags);
240 if (!fls)
241 goto out_err_free;
242
243 fls->mirror_array_cnt = mirror_array_cnt;
244 fls->stripe_unit = stripe_unit;
245 fls->mirror_array = kcalloc(fls->mirror_array_cnt,
246 sizeof(fls->mirror_array[0]), gfp_flags);
247 if (fls->mirror_array == NULL)
248 goto out_err_free;
249
250 for (i = 0; i < fls->mirror_array_cnt; i++) {
251 struct nfs4_deviceid devid;
252 struct nfs4_deviceid_node *idnode;
253 u32 ds_count;
254 u32 fh_count;
255 int j;
256
257 rc = -EIO;
258 p = xdr_inline_decode(&stream, 4);
259 if (!p)
260 goto out_err_free;
261 ds_count = be32_to_cpup(p);
262
263 /* FIXME: allow for striping? */
264 if (ds_count != 1)
265 goto out_err_free;
266
267 fls->mirror_array[i] =
268 kzalloc(sizeof(struct nfs4_ff_layout_mirror),
269 gfp_flags);
270 if (fls->mirror_array[i] == NULL) {
271 rc = -ENOMEM;
272 goto out_err_free;
273 }
274
275 spin_lock_init(&fls->mirror_array[i]->lock);
276 fls->mirror_array[i]->ds_count = ds_count;
277
278 /* deviceid */
279 rc = decode_deviceid(&stream, &devid);
280 if (rc)
281 goto out_err_free;
282
283 idnode = nfs4_find_get_deviceid(NFS_SERVER(lh->plh_inode),
284 &devid, lh->plh_lc_cred,
285 gfp_flags);
286 /*
287 * upon success, mirror_ds is allocated by previous
288 * getdeviceinfo, or newly by .alloc_deviceid_node
289 * nfs4_find_get_deviceid failure is indeed getdeviceinfo falure
290 */
291 if (idnode)
292 fls->mirror_array[i]->mirror_ds =
293 FF_LAYOUT_MIRROR_DS(idnode);
294 else
295 goto out_err_free;
296
297 /* efficiency */
298 rc = -EIO;
299 p = xdr_inline_decode(&stream, 4);
300 if (!p)
301 goto out_err_free;
302 fls->mirror_array[i]->efficiency = be32_to_cpup(p);
303
304 /* stateid */
305 rc = decode_stateid(&stream, &fls->mirror_array[i]->stateid);
306 if (rc)
307 goto out_err_free;
308
309 /* fh */
310 p = xdr_inline_decode(&stream, 4);
311 if (!p)
312 goto out_err_free;
313 fh_count = be32_to_cpup(p);
314
315 fls->mirror_array[i]->fh_versions =
316 kzalloc(fh_count * sizeof(struct nfs_fh),
317 gfp_flags);
318 if (fls->mirror_array[i]->fh_versions == NULL) {
319 rc = -ENOMEM;
320 goto out_err_free;
321 }
322
323 for (j = 0; j < fh_count; j++) {
324 rc = decode_nfs_fh(&stream,
325 &fls->mirror_array[i]->fh_versions[j]);
326 if (rc)
327 goto out_err_free;
328 }
329
330 fls->mirror_array[i]->fh_versions_cnt = fh_count;
331
332 /* user */
333 rc = decode_name(&stream, &fls->mirror_array[i]->uid);
334 if (rc)
335 goto out_err_free;
336
337 /* group */
338 rc = decode_name(&stream, &fls->mirror_array[i]->gid);
339 if (rc)
340 goto out_err_free;
341
342 dprintk("%s: uid %d gid %d\n", __func__,
343 fls->mirror_array[i]->uid,
344 fls->mirror_array[i]->gid);
345 }
346
347 ff_layout_sort_mirrors(fls);
348 rc = ff_layout_check_layout(lgr);
349 if (rc)
350 goto out_err_free;
351
352 ret = &fls->generic_hdr;
353 dprintk("<-- %s (success)\n", __func__);
354out_free_page:
355 __free_page(scratch);
356 return ret;
357out_err_free:
358 _ff_layout_free_lseg(fls);
359 ret = ERR_PTR(rc);
360 dprintk("<-- %s (%d)\n", __func__, rc);
361 goto out_free_page;
362}
363
364static bool ff_layout_has_rw_segments(struct pnfs_layout_hdr *layout)
365{
366 struct pnfs_layout_segment *lseg;
367
368 list_for_each_entry(lseg, &layout->plh_segs, pls_list)
369 if (lseg->pls_range.iomode == IOMODE_RW)
370 return true;
371
372 return false;
373}
374
375static void
376ff_layout_free_lseg(struct pnfs_layout_segment *lseg)
377{
378 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
379 int i;
380
381 dprintk("--> %s\n", __func__);
382
383 for (i = 0; i < fls->mirror_array_cnt; i++) {
384 if (fls->mirror_array[i]) {
385 nfs4_ff_layout_put_deviceid(fls->mirror_array[i]->mirror_ds);
386 fls->mirror_array[i]->mirror_ds = NULL;
387 if (fls->mirror_array[i]->cred) {
388 put_rpccred(fls->mirror_array[i]->cred);
389 fls->mirror_array[i]->cred = NULL;
390 }
391 }
392 }
393
394 if (lseg->pls_range.iomode == IOMODE_RW) {
395 struct nfs4_flexfile_layout *ffl;
396 struct inode *inode;
397
398 ffl = FF_LAYOUT_FROM_HDR(lseg->pls_layout);
399 inode = ffl->generic_hdr.plh_inode;
400 spin_lock(&inode->i_lock);
401 if (!ff_layout_has_rw_segments(lseg->pls_layout)) {
402 ffl->commit_info.nbuckets = 0;
403 kfree(ffl->commit_info.buckets);
404 ffl->commit_info.buckets = NULL;
405 }
406 spin_unlock(&inode->i_lock);
407 }
408 _ff_layout_free_lseg(fls);
409}
410
411/* Return 1 until we have multiple lsegs support */
412static int
413ff_layout_get_lseg_count(struct nfs4_ff_layout_segment *fls)
414{
415 return 1;
416}
417
418static int
419ff_layout_alloc_commit_info(struct pnfs_layout_segment *lseg,
420 struct nfs_commit_info *cinfo,
421 gfp_t gfp_flags)
422{
423 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
424 struct pnfs_commit_bucket *buckets;
425 int size;
426
427 if (cinfo->ds->nbuckets != 0) {
428 /* This assumes there is only one RW lseg per file.
429 * To support multiple lseg per file, we need to
430 * change struct pnfs_commit_bucket to allow dynamic
431 * increasing nbuckets.
432 */
433 return 0;
434 }
435
436 size = ff_layout_get_lseg_count(fls) * FF_LAYOUT_MIRROR_COUNT(lseg);
437
438 buckets = kcalloc(size, sizeof(struct pnfs_commit_bucket),
439 gfp_flags);
440 if (!buckets)
441 return -ENOMEM;
442 else {
443 int i;
444
445 spin_lock(cinfo->lock);
446 if (cinfo->ds->nbuckets != 0)
447 kfree(buckets);
448 else {
449 cinfo->ds->buckets = buckets;
450 cinfo->ds->nbuckets = size;
451 for (i = 0; i < size; i++) {
452 INIT_LIST_HEAD(&buckets[i].written);
453 INIT_LIST_HEAD(&buckets[i].committing);
454 /* mark direct verifier as unset */
455 buckets[i].direct_verf.committed =
456 NFS_INVALID_STABLE_HOW;
457 }
458 }
459 spin_unlock(cinfo->lock);
460 return 0;
461 }
462}
463
464static struct nfs4_pnfs_ds *
465ff_layout_choose_best_ds_for_read(struct nfs_pageio_descriptor *pgio,
466 int *best_idx)
467{
468 struct nfs4_ff_layout_segment *fls;
469 struct nfs4_pnfs_ds *ds;
470 int idx;
471
472 fls = FF_LAYOUT_LSEG(pgio->pg_lseg);
473 /* mirrors are sorted by efficiency */
474 for (idx = 0; idx < fls->mirror_array_cnt; idx++) {
475 ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, idx, false);
476 if (ds) {
477 *best_idx = idx;
478 return ds;
479 }
480 }
481
482 return NULL;
483}
484
485static void
486ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio,
487 struct nfs_page *req)
488{
489 struct nfs_pgio_mirror *pgm;
490 struct nfs4_ff_layout_mirror *mirror;
491 struct nfs4_pnfs_ds *ds;
492 int ds_idx;
493
494 /* Use full layout for now */
495 if (!pgio->pg_lseg)
496 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
497 req->wb_context,
498 0,
499 NFS4_MAX_UINT64,
500 IOMODE_READ,
501 GFP_KERNEL);
502 /* If no lseg, fall back to read through mds */
503 if (pgio->pg_lseg == NULL)
504 goto out_mds;
505
506 ds = ff_layout_choose_best_ds_for_read(pgio, &ds_idx);
507 if (!ds)
508 goto out_mds;
509 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, ds_idx);
510
511 pgio->pg_mirror_idx = ds_idx;
512
513 /* read always uses only one mirror - idx 0 for pgio layer */
514 pgm = &pgio->pg_mirrors[0];
515 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].rsize;
516
517 return;
518out_mds:
519 pnfs_put_lseg(pgio->pg_lseg);
520 pgio->pg_lseg = NULL;
521 nfs_pageio_reset_read_mds(pgio);
522}
523
524static void
525ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio,
526 struct nfs_page *req)
527{
528 struct nfs4_ff_layout_mirror *mirror;
529 struct nfs_pgio_mirror *pgm;
530 struct nfs_commit_info cinfo;
531 struct nfs4_pnfs_ds *ds;
532 int i;
533 int status;
534
535 if (!pgio->pg_lseg)
536 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
537 req->wb_context,
538 0,
539 NFS4_MAX_UINT64,
540 IOMODE_RW,
541 GFP_NOFS);
542 /* If no lseg, fall back to write through mds */
543 if (pgio->pg_lseg == NULL)
544 goto out_mds;
545
546 nfs_init_cinfo(&cinfo, pgio->pg_inode, pgio->pg_dreq);
547 status = ff_layout_alloc_commit_info(pgio->pg_lseg, &cinfo, GFP_NOFS);
548 if (status < 0)
549 goto out_mds;
550
551 /* Use a direct mapping of ds_idx to pgio mirror_idx */
552 if (WARN_ON_ONCE(pgio->pg_mirror_count !=
553 FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg)))
554 goto out_mds;
555
556 for (i = 0; i < pgio->pg_mirror_count; i++) {
557 ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, i, true);
558 if (!ds)
559 goto out_mds;
560 pgm = &pgio->pg_mirrors[i];
561 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i);
562 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].wsize;
563 }
564
565 return;
566
567out_mds:
568 pnfs_put_lseg(pgio->pg_lseg);
569 pgio->pg_lseg = NULL;
570 nfs_pageio_reset_write_mds(pgio);
571}
572
573static unsigned int
574ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio,
575 struct nfs_page *req)
576{
577 if (!pgio->pg_lseg)
578 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
579 req->wb_context,
580 0,
581 NFS4_MAX_UINT64,
582 IOMODE_RW,
583 GFP_NOFS);
584 if (pgio->pg_lseg)
585 return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg);
586
587 /* no lseg means that pnfs is not in use, so no mirroring here */
588 pnfs_put_lseg(pgio->pg_lseg);
589 pgio->pg_lseg = NULL;
590 nfs_pageio_reset_write_mds(pgio);
591 return 1;
592}
593
594static const struct nfs_pageio_ops ff_layout_pg_read_ops = {
595 .pg_init = ff_layout_pg_init_read,
596 .pg_test = pnfs_generic_pg_test,
597 .pg_doio = pnfs_generic_pg_readpages,
598 .pg_cleanup = pnfs_generic_pg_cleanup,
599};
600
601static const struct nfs_pageio_ops ff_layout_pg_write_ops = {
602 .pg_init = ff_layout_pg_init_write,
603 .pg_test = pnfs_generic_pg_test,
604 .pg_doio = pnfs_generic_pg_writepages,
605 .pg_get_mirror_count = ff_layout_pg_get_mirror_count_write,
606 .pg_cleanup = pnfs_generic_pg_cleanup,
607};
608
609static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs)
610{
611 struct rpc_task *task = &hdr->task;
612
613 pnfs_layoutcommit_inode(hdr->inode, false);
614
615 if (retry_pnfs) {
616 dprintk("%s Reset task %5u for i/o through pNFS "
617 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
618 hdr->task.tk_pid,
619 hdr->inode->i_sb->s_id,
620 (unsigned long long)NFS_FILEID(hdr->inode),
621 hdr->args.count,
622 (unsigned long long)hdr->args.offset);
623
624 if (!hdr->dreq) {
625 struct nfs_open_context *ctx;
626
627 ctx = nfs_list_entry(hdr->pages.next)->wb_context;
628 set_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
629 hdr->completion_ops->error_cleanup(&hdr->pages);
630 } else {
631 nfs_direct_set_resched_writes(hdr->dreq);
632 /* fake unstable write to let common nfs resend pages */
633 hdr->verf.committed = NFS_UNSTABLE;
634 hdr->good_bytes = 0;
635 }
636 return;
637 }
638
639 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
640 dprintk("%s Reset task %5u for i/o through MDS "
641 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
642 hdr->task.tk_pid,
643 hdr->inode->i_sb->s_id,
644 (unsigned long long)NFS_FILEID(hdr->inode),
645 hdr->args.count,
646 (unsigned long long)hdr->args.offset);
647
648 task->tk_status = pnfs_write_done_resend_to_mds(hdr);
649 }
650}
651
652static void ff_layout_reset_read(struct nfs_pgio_header *hdr)
653{
654 struct rpc_task *task = &hdr->task;
655
656 pnfs_layoutcommit_inode(hdr->inode, false);
657
658 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
659 dprintk("%s Reset task %5u for i/o through MDS "
660 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
661 hdr->task.tk_pid,
662 hdr->inode->i_sb->s_id,
663 (unsigned long long)NFS_FILEID(hdr->inode),
664 hdr->args.count,
665 (unsigned long long)hdr->args.offset);
666
667 task->tk_status = pnfs_read_done_resend_to_mds(hdr);
668 }
669}
670
671static int ff_layout_async_handle_error_v4(struct rpc_task *task,
672 struct nfs4_state *state,
673 struct nfs_client *clp,
674 struct pnfs_layout_segment *lseg,
675 int idx)
676{
677 struct pnfs_layout_hdr *lo = lseg->pls_layout;
678 struct inode *inode = lo->plh_inode;
679 struct nfs_server *mds_server = NFS_SERVER(inode);
680
681 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
682 struct nfs_client *mds_client = mds_server->nfs_client;
683 struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
684
685 if (task->tk_status >= 0)
686 return 0;
687
688 switch (task->tk_status) {
689 /* MDS state errors */
690 case -NFS4ERR_DELEG_REVOKED:
691 case -NFS4ERR_ADMIN_REVOKED:
692 case -NFS4ERR_BAD_STATEID:
693 if (state == NULL)
694 break;
695 nfs_remove_bad_delegation(state->inode);
696 case -NFS4ERR_OPENMODE:
697 if (state == NULL)
698 break;
699 if (nfs4_schedule_stateid_recovery(mds_server, state) < 0)
700 goto out_bad_stateid;
701 goto wait_on_recovery;
702 case -NFS4ERR_EXPIRED:
703 if (state != NULL) {
704 if (nfs4_schedule_stateid_recovery(mds_server, state) < 0)
705 goto out_bad_stateid;
706 }
707 nfs4_schedule_lease_recovery(mds_client);
708 goto wait_on_recovery;
709 /* DS session errors */
710 case -NFS4ERR_BADSESSION:
711 case -NFS4ERR_BADSLOT:
712 case -NFS4ERR_BAD_HIGH_SLOT:
713 case -NFS4ERR_DEADSESSION:
714 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
715 case -NFS4ERR_SEQ_FALSE_RETRY:
716 case -NFS4ERR_SEQ_MISORDERED:
717 dprintk("%s ERROR %d, Reset session. Exchangeid "
718 "flags 0x%x\n", __func__, task->tk_status,
719 clp->cl_exchange_flags);
720 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
721 break;
722 case -NFS4ERR_DELAY:
723 case -NFS4ERR_GRACE:
724 rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX);
725 break;
726 case -NFS4ERR_RETRY_UNCACHED_REP:
727 break;
728 /* Invalidate Layout errors */
729 case -NFS4ERR_PNFS_NO_LAYOUT:
730 case -ESTALE: /* mapped NFS4ERR_STALE */
731 case -EBADHANDLE: /* mapped NFS4ERR_BADHANDLE */
732 case -EISDIR: /* mapped NFS4ERR_ISDIR */
733 case -NFS4ERR_FHEXPIRED:
734 case -NFS4ERR_WRONG_TYPE:
735 dprintk("%s Invalid layout error %d\n", __func__,
736 task->tk_status);
737 /*
738 * Destroy layout so new i/o will get a new layout.
739 * Layout will not be destroyed until all current lseg
740 * references are put. Mark layout as invalid to resend failed
741 * i/o and all i/o waiting on the slot table to the MDS until
742 * layout is destroyed and a new valid layout is obtained.
743 */
744 pnfs_destroy_layout(NFS_I(inode));
745 rpc_wake_up(&tbl->slot_tbl_waitq);
746 goto reset;
747 /* RPC connection errors */
748 case -ECONNREFUSED:
749 case -EHOSTDOWN:
750 case -EHOSTUNREACH:
751 case -ENETUNREACH:
752 case -EIO:
753 case -ETIMEDOUT:
754 case -EPIPE:
755 dprintk("%s DS connection error %d\n", __func__,
756 task->tk_status);
757 nfs4_mark_deviceid_unavailable(devid);
758 rpc_wake_up(&tbl->slot_tbl_waitq);
759 /* fall through */
760 default:
761 if (ff_layout_has_available_ds(lseg))
762 return -NFS4ERR_RESET_TO_PNFS;
763reset:
764 dprintk("%s Retry through MDS. Error %d\n", __func__,
765 task->tk_status);
766 return -NFS4ERR_RESET_TO_MDS;
767 }
768out:
769 task->tk_status = 0;
770 return -EAGAIN;
771out_bad_stateid:
772 task->tk_status = -EIO;
773 return 0;
774wait_on_recovery:
775 rpc_sleep_on(&mds_client->cl_rpcwaitq, task, NULL);
776 if (test_bit(NFS4CLNT_MANAGER_RUNNING, &mds_client->cl_state) == 0)
777 rpc_wake_up_queued_task(&mds_client->cl_rpcwaitq, task);
778 goto out;
779}
780
781/* Retry all errors through either pNFS or MDS except for -EJUKEBOX */
782static int ff_layout_async_handle_error_v3(struct rpc_task *task,
783 struct pnfs_layout_segment *lseg,
784 int idx)
785{
786 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
787
788 if (task->tk_status >= 0)
789 return 0;
790
791 if (task->tk_status != -EJUKEBOX) {
792 dprintk("%s DS connection error %d\n", __func__,
793 task->tk_status);
794 nfs4_mark_deviceid_unavailable(devid);
795 if (ff_layout_has_available_ds(lseg))
796 return -NFS4ERR_RESET_TO_PNFS;
797 else
798 return -NFS4ERR_RESET_TO_MDS;
799 }
800
801 if (task->tk_status == -EJUKEBOX)
802 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY);
803 task->tk_status = 0;
804 rpc_restart_call(task);
805 rpc_delay(task, NFS_JUKEBOX_RETRY_TIME);
806 return -EAGAIN;
807}
808
809static int ff_layout_async_handle_error(struct rpc_task *task,
810 struct nfs4_state *state,
811 struct nfs_client *clp,
812 struct pnfs_layout_segment *lseg,
813 int idx)
814{
815 int vers = clp->cl_nfs_mod->rpc_vers->number;
816
817 switch (vers) {
818 case 3:
819 return ff_layout_async_handle_error_v3(task, lseg, idx);
820 case 4:
821 return ff_layout_async_handle_error_v4(task, state, clp,
822 lseg, idx);
823 default:
824 /* should never happen */
825 WARN_ON_ONCE(1);
826 return 0;
827 }
828}
829
830static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg,
831 int idx, u64 offset, u64 length,
832 u32 status, int opnum)
833{
834 struct nfs4_ff_layout_mirror *mirror;
835 int err;
836
837 mirror = FF_LAYOUT_COMP(lseg, idx);
838 err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
839 mirror, offset, length, status, opnum,
840 GFP_NOIO);
841 dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status);
842}
843
844/* NFS_PROTO call done callback routines */
845
846static int ff_layout_read_done_cb(struct rpc_task *task,
847 struct nfs_pgio_header *hdr)
848{
849 struct inode *inode;
850 int err;
851
852 trace_nfs4_pnfs_read(hdr, task->tk_status);
853 if (task->tk_status == -ETIMEDOUT && !hdr->res.op_status)
854 hdr->res.op_status = NFS4ERR_NXIO;
855 if (task->tk_status < 0 && hdr->res.op_status)
856 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
857 hdr->args.offset, hdr->args.count,
858 hdr->res.op_status, OP_READ);
859 err = ff_layout_async_handle_error(task, hdr->args.context->state,
860 hdr->ds_clp, hdr->lseg,
861 hdr->pgio_mirror_idx);
862
863 switch (err) {
864 case -NFS4ERR_RESET_TO_PNFS:
865 set_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE,
866 &hdr->lseg->pls_layout->plh_flags);
867 pnfs_read_resend_pnfs(hdr);
868 return task->tk_status;
869 case -NFS4ERR_RESET_TO_MDS:
870 inode = hdr->lseg->pls_layout->plh_inode;
871 pnfs_error_mark_layout_for_return(inode, hdr->lseg);
872 ff_layout_reset_read(hdr);
873 return task->tk_status;
874 case -EAGAIN:
875 rpc_restart_call_prepare(task);
876 return -EAGAIN;
877 }
878
879 return 0;
880}
881
882/*
883 * We reference the rpc_cred of the first WRITE that triggers the need for
884 * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
885 * rfc5661 is not clear about which credential should be used.
886 *
887 * Flexlayout client should treat DS replied FILE_SYNC as DATA_SYNC, so
888 * to follow http://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751
889 * we always send layoutcommit after DS writes.
890 */
891static void
892ff_layout_set_layoutcommit(struct nfs_pgio_header *hdr)
893{
894 pnfs_set_layoutcommit(hdr);
895 dprintk("%s inode %lu pls_end_pos %lu\n", __func__, hdr->inode->i_ino,
896 (unsigned long) NFS_I(hdr->inode)->layout->plh_lwb);
897}
898
899static bool
900ff_layout_reset_to_mds(struct pnfs_layout_segment *lseg, int idx)
901{
902 /* No mirroring for now */
903 struct nfs4_deviceid_node *node = FF_LAYOUT_DEVID_NODE(lseg, idx);
904
905 return ff_layout_test_devid_unavailable(node);
906}
907
908static int ff_layout_read_prepare_common(struct rpc_task *task,
909 struct nfs_pgio_header *hdr)
910{
911 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
912 rpc_exit(task, -EIO);
913 return -EIO;
914 }
915 if (ff_layout_reset_to_mds(hdr->lseg, hdr->pgio_mirror_idx)) {
916 dprintk("%s task %u reset io to MDS\n", __func__, task->tk_pid);
917 if (ff_layout_has_available_ds(hdr->lseg))
918 pnfs_read_resend_pnfs(hdr);
919 else
920 ff_layout_reset_read(hdr);
921 rpc_exit(task, 0);
922 return -EAGAIN;
923 }
924 hdr->pgio_done_cb = ff_layout_read_done_cb;
925
926 return 0;
927}
928
929/*
930 * Call ops for the async read/write cases
931 * In the case of dense layouts, the offset needs to be reset to its
932 * original value.
933 */
934static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data)
935{
936 struct nfs_pgio_header *hdr = data;
937
938 if (ff_layout_read_prepare_common(task, hdr))
939 return;
940
941 rpc_call_start(task);
942}
943
944static int ff_layout_setup_sequence(struct nfs_client *ds_clp,
945 struct nfs4_sequence_args *args,
946 struct nfs4_sequence_res *res,
947 struct rpc_task *task)
948{
949 if (ds_clp->cl_session)
950 return nfs41_setup_sequence(ds_clp->cl_session,
951 args,
952 res,
953 task);
954 return nfs40_setup_sequence(ds_clp->cl_slot_tbl,
955 args,
956 res,
957 task);
958}
959
960static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data)
961{
962 struct nfs_pgio_header *hdr = data;
963
964 if (ff_layout_read_prepare_common(task, hdr))
965 return;
966
967 if (ff_layout_setup_sequence(hdr->ds_clp,
968 &hdr->args.seq_args,
969 &hdr->res.seq_res,
970 task))
971 return;
972
973 if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context,
974 hdr->args.lock_context, FMODE_READ) == -EIO)
975 rpc_exit(task, -EIO); /* lost lock, terminate I/O */
976}
977
978static void ff_layout_read_call_done(struct rpc_task *task, void *data)
979{
980 struct nfs_pgio_header *hdr = data;
981
982 dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
983
984 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
985 task->tk_status == 0) {
986 nfs4_sequence_done(task, &hdr->res.seq_res);
987 return;
988 }
989
990 /* Note this may cause RPC to be resent */
991 hdr->mds_ops->rpc_call_done(task, hdr);
992}
993
994static void ff_layout_read_count_stats(struct rpc_task *task, void *data)
995{
996 struct nfs_pgio_header *hdr = data;
997
998 rpc_count_iostats_metrics(task,
999 &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]);
1000}
1001
1002static int ff_layout_write_done_cb(struct rpc_task *task,
1003 struct nfs_pgio_header *hdr)
1004{
1005 struct inode *inode;
1006 int err;
1007
1008 trace_nfs4_pnfs_write(hdr, task->tk_status);
1009 if (task->tk_status == -ETIMEDOUT && !hdr->res.op_status)
1010 hdr->res.op_status = NFS4ERR_NXIO;
1011 if (task->tk_status < 0 && hdr->res.op_status)
1012 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1013 hdr->args.offset, hdr->args.count,
1014 hdr->res.op_status, OP_WRITE);
1015 err = ff_layout_async_handle_error(task, hdr->args.context->state,
1016 hdr->ds_clp, hdr->lseg,
1017 hdr->pgio_mirror_idx);
1018
1019 switch (err) {
1020 case -NFS4ERR_RESET_TO_PNFS:
1021 case -NFS4ERR_RESET_TO_MDS:
1022 inode = hdr->lseg->pls_layout->plh_inode;
1023 pnfs_error_mark_layout_for_return(inode, hdr->lseg);
1024 if (err == -NFS4ERR_RESET_TO_PNFS) {
1025 pnfs_set_retry_layoutget(hdr->lseg->pls_layout);
1026 ff_layout_reset_write(hdr, true);
1027 } else {
1028 pnfs_clear_retry_layoutget(hdr->lseg->pls_layout);
1029 ff_layout_reset_write(hdr, false);
1030 }
1031 return task->tk_status;
1032 case -EAGAIN:
1033 rpc_restart_call_prepare(task);
1034 return -EAGAIN;
1035 }
1036
1037 if (hdr->res.verf->committed == NFS_FILE_SYNC ||
1038 hdr->res.verf->committed == NFS_DATA_SYNC)
1039 ff_layout_set_layoutcommit(hdr);
1040
1041 return 0;
1042}
1043
1044static int ff_layout_commit_done_cb(struct rpc_task *task,
1045 struct nfs_commit_data *data)
1046{
1047 struct inode *inode;
1048 int err;
1049
1050 trace_nfs4_pnfs_commit_ds(data, task->tk_status);
1051 if (task->tk_status == -ETIMEDOUT && !data->res.op_status)
1052 data->res.op_status = NFS4ERR_NXIO;
1053 if (task->tk_status < 0 && data->res.op_status)
1054 ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index,
1055 data->args.offset, data->args.count,
1056 data->res.op_status, OP_COMMIT);
1057 err = ff_layout_async_handle_error(task, NULL, data->ds_clp,
1058 data->lseg, data->ds_commit_index);
1059
1060 switch (err) {
1061 case -NFS4ERR_RESET_TO_PNFS:
1062 case -NFS4ERR_RESET_TO_MDS:
1063 inode = data->lseg->pls_layout->plh_inode;
1064 pnfs_error_mark_layout_for_return(inode, data->lseg);
1065 if (err == -NFS4ERR_RESET_TO_PNFS)
1066 pnfs_set_retry_layoutget(data->lseg->pls_layout);
1067 else
1068 pnfs_clear_retry_layoutget(data->lseg->pls_layout);
1069 pnfs_generic_prepare_to_resend_writes(data);
1070 return -EAGAIN;
1071 case -EAGAIN:
1072 rpc_restart_call_prepare(task);
1073 return -EAGAIN;
1074 }
1075
1076 if (data->verf.committed == NFS_UNSTABLE)
1077 pnfs_commit_set_layoutcommit(data);
1078
1079 return 0;
1080}
1081
1082static int ff_layout_write_prepare_common(struct rpc_task *task,
1083 struct nfs_pgio_header *hdr)
1084{
1085 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1086 rpc_exit(task, -EIO);
1087 return -EIO;
1088 }
1089
1090 if (ff_layout_reset_to_mds(hdr->lseg, hdr->pgio_mirror_idx)) {
1091 bool retry_pnfs;
1092
1093 retry_pnfs = ff_layout_has_available_ds(hdr->lseg);
1094 dprintk("%s task %u reset io to %s\n", __func__,
1095 task->tk_pid, retry_pnfs ? "pNFS" : "MDS");
1096 ff_layout_reset_write(hdr, retry_pnfs);
1097 rpc_exit(task, 0);
1098 return -EAGAIN;
1099 }
1100
1101 return 0;
1102}
1103
1104static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data)
1105{
1106 struct nfs_pgio_header *hdr = data;
1107
1108 if (ff_layout_write_prepare_common(task, hdr))
1109 return;
1110
1111 rpc_call_start(task);
1112}
1113
1114static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data)
1115{
1116 struct nfs_pgio_header *hdr = data;
1117
1118 if (ff_layout_write_prepare_common(task, hdr))
1119 return;
1120
1121 if (ff_layout_setup_sequence(hdr->ds_clp,
1122 &hdr->args.seq_args,
1123 &hdr->res.seq_res,
1124 task))
1125 return;
1126
1127 if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context,
1128 hdr->args.lock_context, FMODE_WRITE) == -EIO)
1129 rpc_exit(task, -EIO); /* lost lock, terminate I/O */
1130}
1131
1132static void ff_layout_write_call_done(struct rpc_task *task, void *data)
1133{
1134 struct nfs_pgio_header *hdr = data;
1135
1136 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1137 task->tk_status == 0) {
1138 nfs4_sequence_done(task, &hdr->res.seq_res);
1139 return;
1140 }
1141
1142 /* Note this may cause RPC to be resent */
1143 hdr->mds_ops->rpc_call_done(task, hdr);
1144}
1145
1146static void ff_layout_write_count_stats(struct rpc_task *task, void *data)
1147{
1148 struct nfs_pgio_header *hdr = data;
1149
1150 rpc_count_iostats_metrics(task,
1151 &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]);
1152}
1153
1154static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data)
1155{
1156 rpc_call_start(task);
1157}
1158
1159static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data)
1160{
1161 struct nfs_commit_data *wdata = data;
1162
1163 ff_layout_setup_sequence(wdata->ds_clp,
1164 &wdata->args.seq_args,
1165 &wdata->res.seq_res,
1166 task);
1167}
1168
1169static void ff_layout_commit_count_stats(struct rpc_task *task, void *data)
1170{
1171 struct nfs_commit_data *cdata = data;
1172
1173 rpc_count_iostats_metrics(task,
1174 &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]);
1175}
1176
1177static const struct rpc_call_ops ff_layout_read_call_ops_v3 = {
1178 .rpc_call_prepare = ff_layout_read_prepare_v3,
1179 .rpc_call_done = ff_layout_read_call_done,
1180 .rpc_count_stats = ff_layout_read_count_stats,
1181 .rpc_release = pnfs_generic_rw_release,
1182};
1183
1184static const struct rpc_call_ops ff_layout_read_call_ops_v4 = {
1185 .rpc_call_prepare = ff_layout_read_prepare_v4,
1186 .rpc_call_done = ff_layout_read_call_done,
1187 .rpc_count_stats = ff_layout_read_count_stats,
1188 .rpc_release = pnfs_generic_rw_release,
1189};
1190
1191static const struct rpc_call_ops ff_layout_write_call_ops_v3 = {
1192 .rpc_call_prepare = ff_layout_write_prepare_v3,
1193 .rpc_call_done = ff_layout_write_call_done,
1194 .rpc_count_stats = ff_layout_write_count_stats,
1195 .rpc_release = pnfs_generic_rw_release,
1196};
1197
1198static const struct rpc_call_ops ff_layout_write_call_ops_v4 = {
1199 .rpc_call_prepare = ff_layout_write_prepare_v4,
1200 .rpc_call_done = ff_layout_write_call_done,
1201 .rpc_count_stats = ff_layout_write_count_stats,
1202 .rpc_release = pnfs_generic_rw_release,
1203};
1204
1205static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = {
1206 .rpc_call_prepare = ff_layout_commit_prepare_v3,
1207 .rpc_call_done = pnfs_generic_write_commit_done,
1208 .rpc_count_stats = ff_layout_commit_count_stats,
1209 .rpc_release = pnfs_generic_commit_release,
1210};
1211
1212static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = {
1213 .rpc_call_prepare = ff_layout_commit_prepare_v4,
1214 .rpc_call_done = pnfs_generic_write_commit_done,
1215 .rpc_count_stats = ff_layout_commit_count_stats,
1216 .rpc_release = pnfs_generic_commit_release,
1217};
1218
1219static enum pnfs_try_status
1220ff_layout_read_pagelist(struct nfs_pgio_header *hdr)
1221{
1222 struct pnfs_layout_segment *lseg = hdr->lseg;
1223 struct nfs4_pnfs_ds *ds;
1224 struct rpc_clnt *ds_clnt;
1225 struct rpc_cred *ds_cred;
1226 loff_t offset = hdr->args.offset;
1227 u32 idx = hdr->pgio_mirror_idx;
1228 int vers;
1229 struct nfs_fh *fh;
1230
1231 dprintk("--> %s ino %lu pgbase %u req %Zu@%llu\n",
1232 __func__, hdr->inode->i_ino,
1233 hdr->args.pgbase, (size_t)hdr->args.count, offset);
1234
1235 ds = nfs4_ff_layout_prepare_ds(lseg, idx, false);
1236 if (!ds)
1237 goto out_failed;
1238
1239 ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1240 hdr->inode);
1241 if (IS_ERR(ds_clnt))
1242 goto out_failed;
1243
1244 ds_cred = ff_layout_get_ds_cred(lseg, idx, hdr->cred);
1245 if (IS_ERR(ds_cred))
1246 goto out_failed;
1247
1248 vers = nfs4_ff_layout_ds_version(lseg, idx);
1249
1250 dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__,
1251 ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count), vers);
1252
1253 atomic_inc(&ds->ds_clp->cl_count);
1254 hdr->ds_clp = ds->ds_clp;
1255 fh = nfs4_ff_layout_select_ds_fh(lseg, idx);
1256 if (fh)
1257 hdr->args.fh = fh;
1258
1259 /*
1260 * Note that if we ever decide to split across DSes,
1261 * then we may need to handle dense-like offsets.
1262 */
1263 hdr->args.offset = offset;
1264 hdr->mds_offset = offset;
1265
1266 /* Perform an asynchronous read to ds */
1267 nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1268 vers == 3 ? &ff_layout_read_call_ops_v3 :
1269 &ff_layout_read_call_ops_v4,
1270 0, RPC_TASK_SOFTCONN);
1271
1272 return PNFS_ATTEMPTED;
1273
1274out_failed:
1275 if (ff_layout_has_available_ds(lseg))
1276 return PNFS_TRY_AGAIN;
1277 return PNFS_NOT_ATTEMPTED;
1278}
1279
1280/* Perform async writes. */
1281static enum pnfs_try_status
1282ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
1283{
1284 struct pnfs_layout_segment *lseg = hdr->lseg;
1285 struct nfs4_pnfs_ds *ds;
1286 struct rpc_clnt *ds_clnt;
1287 struct rpc_cred *ds_cred;
1288 loff_t offset = hdr->args.offset;
1289 int vers;
1290 struct nfs_fh *fh;
1291 int idx = hdr->pgio_mirror_idx;
1292
1293 ds = nfs4_ff_layout_prepare_ds(lseg, idx, true);
1294 if (!ds)
1295 return PNFS_NOT_ATTEMPTED;
1296
1297 ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1298 hdr->inode);
1299 if (IS_ERR(ds_clnt))
1300 return PNFS_NOT_ATTEMPTED;
1301
1302 ds_cred = ff_layout_get_ds_cred(lseg, idx, hdr->cred);
1303 if (IS_ERR(ds_cred))
1304 return PNFS_NOT_ATTEMPTED;
1305
1306 vers = nfs4_ff_layout_ds_version(lseg, idx);
1307
1308 dprintk("%s ino %lu sync %d req %Zu@%llu DS: %s cl_count %d vers %d\n",
1309 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
1310 offset, ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count),
1311 vers);
1312
1313 hdr->pgio_done_cb = ff_layout_write_done_cb;
1314 atomic_inc(&ds->ds_clp->cl_count);
1315 hdr->ds_clp = ds->ds_clp;
1316 hdr->ds_commit_idx = idx;
1317 fh = nfs4_ff_layout_select_ds_fh(lseg, idx);
1318 if (fh)
1319 hdr->args.fh = fh;
1320
1321 /*
1322 * Note that if we ever decide to split across DSes,
1323 * then we may need to handle dense-like offsets.
1324 */
1325 hdr->args.offset = offset;
1326
1327 /* Perform an asynchronous write */
1328 nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1329 vers == 3 ? &ff_layout_write_call_ops_v3 :
1330 &ff_layout_write_call_ops_v4,
1331 sync, RPC_TASK_SOFTCONN);
1332 return PNFS_ATTEMPTED;
1333}
1334
1335static void
1336ff_layout_mark_request_commit(struct nfs_page *req,
1337 struct pnfs_layout_segment *lseg,
1338 struct nfs_commit_info *cinfo,
1339 u32 ds_commit_idx)
1340{
1341 struct list_head *list;
1342 struct pnfs_commit_bucket *buckets;
1343
1344 spin_lock(cinfo->lock);
1345 buckets = cinfo->ds->buckets;
1346 list = &buckets[ds_commit_idx].written;
1347 if (list_empty(list)) {
1348 /* Non-empty buckets hold a reference on the lseg. That ref
1349 * is normally transferred to the COMMIT call and released
1350 * there. It could also be released if the last req is pulled
1351 * off due to a rewrite, in which case it will be done in
1352 * pnfs_common_clear_request_commit
1353 */
1354 WARN_ON_ONCE(buckets[ds_commit_idx].wlseg != NULL);
1355 buckets[ds_commit_idx].wlseg = pnfs_get_lseg(lseg);
1356 }
1357 set_bit(PG_COMMIT_TO_DS, &req->wb_flags);
1358 cinfo->ds->nwritten++;
1359
1360 /* nfs_request_add_commit_list(). We need to add req to list without
1361 * dropping cinfo lock.
1362 */
1363 set_bit(PG_CLEAN, &(req)->wb_flags);
1364 nfs_list_add_request(req, list);
1365 cinfo->mds->ncommit++;
1366 spin_unlock(cinfo->lock);
1367 if (!cinfo->dreq) {
1368 inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
1369 inc_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info,
1370 BDI_RECLAIMABLE);
1371 __mark_inode_dirty(req->wb_context->dentry->d_inode,
1372 I_DIRTY_DATASYNC);
1373 }
1374}
1375
1376static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1377{
1378 return i;
1379}
1380
1381static struct nfs_fh *
1382select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1383{
1384 struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1385
1386 /* FIXME: Assume that there is only one NFS version available
1387 * for the DS.
1388 */
1389 return &flseg->mirror_array[i]->fh_versions[0];
1390}
1391
1392static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how)
1393{
1394 struct pnfs_layout_segment *lseg = data->lseg;
1395 struct nfs4_pnfs_ds *ds;
1396 struct rpc_clnt *ds_clnt;
1397 struct rpc_cred *ds_cred;
1398 u32 idx;
1399 int vers;
1400 struct nfs_fh *fh;
1401
1402 idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
1403 ds = nfs4_ff_layout_prepare_ds(lseg, idx, true);
1404 if (!ds)
1405 goto out_err;
1406
1407 ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1408 data->inode);
1409 if (IS_ERR(ds_clnt))
1410 goto out_err;
1411
1412 ds_cred = ff_layout_get_ds_cred(lseg, idx, data->cred);
1413 if (IS_ERR(ds_cred))
1414 goto out_err;
1415
1416 vers = nfs4_ff_layout_ds_version(lseg, idx);
1417
1418 dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__,
1419 data->inode->i_ino, how, atomic_read(&ds->ds_clp->cl_count),
1420 vers);
1421 data->commit_done_cb = ff_layout_commit_done_cb;
1422 data->cred = ds_cred;
1423 atomic_inc(&ds->ds_clp->cl_count);
1424 data->ds_clp = ds->ds_clp;
1425 fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1426 if (fh)
1427 data->args.fh = fh;
1428 return nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops,
1429 vers == 3 ? &ff_layout_commit_call_ops_v3 :
1430 &ff_layout_commit_call_ops_v4,
1431 how, RPC_TASK_SOFTCONN);
1432out_err:
1433 pnfs_generic_prepare_to_resend_writes(data);
1434 pnfs_generic_commit_release(data);
1435 return -EAGAIN;
1436}
1437
1438static int
1439ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1440 int how, struct nfs_commit_info *cinfo)
1441{
1442 return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
1443 ff_layout_initiate_commit);
1444}
1445
1446static struct pnfs_ds_commit_info *
1447ff_layout_get_ds_info(struct inode *inode)
1448{
1449 struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
1450
1451 if (layout == NULL)
1452 return NULL;
1453
1454 return &FF_LAYOUT_FROM_HDR(layout)->commit_info;
1455}
1456
1457static void
1458ff_layout_free_deveiceid_node(struct nfs4_deviceid_node *d)
1459{
1460 nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds,
1461 id_node));
1462}
1463
1464static int ff_layout_encode_ioerr(struct nfs4_flexfile_layout *flo,
1465 struct xdr_stream *xdr,
1466 const struct nfs4_layoutreturn_args *args)
1467{
1468 struct pnfs_layout_hdr *hdr = &flo->generic_hdr;
1469 __be32 *start;
1470 int count = 0, ret = 0;
1471
1472 start = xdr_reserve_space(xdr, 4);
1473 if (unlikely(!start))
1474 return -E2BIG;
1475
1476 /* This assume we always return _ALL_ layouts */
1477 spin_lock(&hdr->plh_inode->i_lock);
1478 ret = ff_layout_encode_ds_ioerr(flo, xdr, &count, &args->range);
1479 spin_unlock(&hdr->plh_inode->i_lock);
1480
1481 *start = cpu_to_be32(count);
1482
1483 return ret;
1484}
1485
1486/* report nothing for now */
1487static void ff_layout_encode_iostats(struct nfs4_flexfile_layout *flo,
1488 struct xdr_stream *xdr,
1489 const struct nfs4_layoutreturn_args *args)
1490{
1491 __be32 *p;
1492
1493 p = xdr_reserve_space(xdr, 4);
1494 if (likely(p))
1495 *p = cpu_to_be32(0);
1496}
1497
1498static struct nfs4_deviceid_node *
1499ff_layout_alloc_deviceid_node(struct nfs_server *server,
1500 struct pnfs_device *pdev, gfp_t gfp_flags)
1501{
1502 struct nfs4_ff_layout_ds *dsaddr;
1503
1504 dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags);
1505 if (!dsaddr)
1506 return NULL;
1507 return &dsaddr->id_node;
1508}
1509
1510static void
1511ff_layout_encode_layoutreturn(struct pnfs_layout_hdr *lo,
1512 struct xdr_stream *xdr,
1513 const struct nfs4_layoutreturn_args *args)
1514{
1515 struct nfs4_flexfile_layout *flo = FF_LAYOUT_FROM_HDR(lo);
1516 __be32 *start;
1517
1518 dprintk("%s: Begin\n", __func__);
1519 start = xdr_reserve_space(xdr, 4);
1520 BUG_ON(!start);
1521
1522 if (ff_layout_encode_ioerr(flo, xdr, args))
1523 goto out;
1524
1525 ff_layout_encode_iostats(flo, xdr, args);
1526out:
1527 *start = cpu_to_be32((xdr->p - start - 1) * 4);
1528 dprintk("%s: Return\n", __func__);
1529}
1530
1531static struct pnfs_layoutdriver_type flexfilelayout_type = {
1532 .id = LAYOUT_FLEX_FILES,
1533 .name = "LAYOUT_FLEX_FILES",
1534 .owner = THIS_MODULE,
1535 .alloc_layout_hdr = ff_layout_alloc_layout_hdr,
1536 .free_layout_hdr = ff_layout_free_layout_hdr,
1537 .alloc_lseg = ff_layout_alloc_lseg,
1538 .free_lseg = ff_layout_free_lseg,
1539 .pg_read_ops = &ff_layout_pg_read_ops,
1540 .pg_write_ops = &ff_layout_pg_write_ops,
1541 .get_ds_info = ff_layout_get_ds_info,
1542 .free_deviceid_node = ff_layout_free_deveiceid_node,
1543 .mark_request_commit = ff_layout_mark_request_commit,
1544 .clear_request_commit = pnfs_generic_clear_request_commit,
1545 .scan_commit_lists = pnfs_generic_scan_commit_lists,
1546 .recover_commit_reqs = pnfs_generic_recover_commit_reqs,
1547 .commit_pagelist = ff_layout_commit_pagelist,
1548 .read_pagelist = ff_layout_read_pagelist,
1549 .write_pagelist = ff_layout_write_pagelist,
1550 .alloc_deviceid_node = ff_layout_alloc_deviceid_node,
1551 .encode_layoutreturn = ff_layout_encode_layoutreturn,
1552};
1553
1554static int __init nfs4flexfilelayout_init(void)
1555{
1556 printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n",
1557 __func__);
1558 return pnfs_register_layoutdriver(&flexfilelayout_type);
1559}
1560
1561static void __exit nfs4flexfilelayout_exit(void)
1562{
1563 printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n",
1564 __func__);
1565 pnfs_unregister_layoutdriver(&flexfilelayout_type);
1566}
1567
1568MODULE_ALIAS("nfs-layouttype4-4");
1569
1570MODULE_LICENSE("GPL");
1571MODULE_DESCRIPTION("The NFSv4 flexfile layout driver");
1572
1573module_init(nfs4flexfilelayout_init);
1574module_exit(nfs4flexfilelayout_exit);
diff --git a/fs/nfs/flexfilelayout/flexfilelayout.h b/fs/nfs/flexfilelayout/flexfilelayout.h
new file mode 100644
index 000000000000..070f20445b2d
--- /dev/null
+++ b/fs/nfs/flexfilelayout/flexfilelayout.h
@@ -0,0 +1,155 @@
1/*
2 * NFSv4 flexfile layout driver data structures.
3 *
4 * Copyright (c) 2014, Primary Data, Inc. All rights reserved.
5 *
6 * Tao Peng <bergwolf@primarydata.com>
7 */
8
9#ifndef FS_NFS_NFS4FLEXFILELAYOUT_H
10#define FS_NFS_NFS4FLEXFILELAYOUT_H
11
12#include "../pnfs.h"
13
14/* XXX: Let's filter out insanely large mirror count for now to avoid oom
15 * due to network error etc. */
16#define NFS4_FLEXFILE_LAYOUT_MAX_MIRROR_CNT 4096
17
18struct nfs4_ff_ds_version {
19 u32 version;
20 u32 minor_version;
21 u32 rsize;
22 u32 wsize;
23 bool tightly_coupled;
24};
25
26/* chained in global deviceid hlist */
27struct nfs4_ff_layout_ds {
28 struct nfs4_deviceid_node id_node;
29 u32 ds_versions_cnt;
30 struct nfs4_ff_ds_version *ds_versions;
31 struct nfs4_pnfs_ds *ds;
32};
33
34struct nfs4_ff_layout_ds_err {
35 struct list_head list; /* linked in mirror error_list */
36 u64 offset;
37 u64 length;
38 int status;
39 enum nfs_opnum4 opnum;
40 nfs4_stateid stateid;
41 struct nfs4_deviceid deviceid;
42};
43
44struct nfs4_ff_layout_mirror {
45 u32 ds_count;
46 u32 efficiency;
47 struct nfs4_ff_layout_ds *mirror_ds;
48 u32 fh_versions_cnt;
49 struct nfs_fh *fh_versions;
50 nfs4_stateid stateid;
51 struct nfs4_string user_name;
52 struct nfs4_string group_name;
53 u32 uid;
54 u32 gid;
55 struct rpc_cred *cred;
56 spinlock_t lock;
57};
58
59struct nfs4_ff_layout_segment {
60 struct pnfs_layout_segment generic_hdr;
61 u64 stripe_unit;
62 u32 mirror_array_cnt;
63 struct nfs4_ff_layout_mirror **mirror_array;
64};
65
66struct nfs4_flexfile_layout {
67 struct pnfs_layout_hdr generic_hdr;
68 struct pnfs_ds_commit_info commit_info;
69 struct list_head error_list; /* nfs4_ff_layout_ds_err */
70};
71
72static inline struct nfs4_flexfile_layout *
73FF_LAYOUT_FROM_HDR(struct pnfs_layout_hdr *lo)
74{
75 return container_of(lo, struct nfs4_flexfile_layout, generic_hdr);
76}
77
78static inline struct nfs4_ff_layout_segment *
79FF_LAYOUT_LSEG(struct pnfs_layout_segment *lseg)
80{
81 return container_of(lseg,
82 struct nfs4_ff_layout_segment,
83 generic_hdr);
84}
85
86static inline struct nfs4_deviceid_node *
87FF_LAYOUT_DEVID_NODE(struct pnfs_layout_segment *lseg, u32 idx)
88{
89 if (idx >= FF_LAYOUT_LSEG(lseg)->mirror_array_cnt ||
90 FF_LAYOUT_LSEG(lseg)->mirror_array[idx] == NULL ||
91 FF_LAYOUT_LSEG(lseg)->mirror_array[idx]->mirror_ds == NULL)
92 return NULL;
93 return &FF_LAYOUT_LSEG(lseg)->mirror_array[idx]->mirror_ds->id_node;
94}
95
96static inline struct nfs4_ff_layout_ds *
97FF_LAYOUT_MIRROR_DS(struct nfs4_deviceid_node *node)
98{
99 return container_of(node, struct nfs4_ff_layout_ds, id_node);
100}
101
102static inline struct nfs4_ff_layout_mirror *
103FF_LAYOUT_COMP(struct pnfs_layout_segment *lseg, u32 idx)
104{
105 if (idx >= FF_LAYOUT_LSEG(lseg)->mirror_array_cnt)
106 return NULL;
107 return FF_LAYOUT_LSEG(lseg)->mirror_array[idx];
108}
109
110static inline u32
111FF_LAYOUT_MIRROR_COUNT(struct pnfs_layout_segment *lseg)
112{
113 return FF_LAYOUT_LSEG(lseg)->mirror_array_cnt;
114}
115
116static inline bool
117ff_layout_test_devid_unavailable(struct nfs4_deviceid_node *node)
118{
119 return nfs4_test_deviceid_unavailable(node);
120}
121
122static inline int
123nfs4_ff_layout_ds_version(struct pnfs_layout_segment *lseg, u32 ds_idx)
124{
125 return FF_LAYOUT_COMP(lseg, ds_idx)->mirror_ds->ds_versions[0].version;
126}
127
128struct nfs4_ff_layout_ds *
129nfs4_ff_alloc_deviceid_node(struct nfs_server *server, struct pnfs_device *pdev,
130 gfp_t gfp_flags);
131void nfs4_ff_layout_put_deviceid(struct nfs4_ff_layout_ds *mirror_ds);
132void nfs4_ff_layout_free_deviceid(struct nfs4_ff_layout_ds *mirror_ds);
133int ff_layout_track_ds_error(struct nfs4_flexfile_layout *flo,
134 struct nfs4_ff_layout_mirror *mirror, u64 offset,
135 u64 length, int status, enum nfs_opnum4 opnum,
136 gfp_t gfp_flags);
137int ff_layout_encode_ds_ioerr(struct nfs4_flexfile_layout *flo,
138 struct xdr_stream *xdr, int *count,
139 const struct pnfs_layout_range *range);
140struct nfs_fh *
141nfs4_ff_layout_select_ds_fh(struct pnfs_layout_segment *lseg, u32 mirror_idx);
142
143struct nfs4_pnfs_ds *
144nfs4_ff_layout_prepare_ds(struct pnfs_layout_segment *lseg, u32 ds_idx,
145 bool fail_return);
146
147struct rpc_clnt *
148nfs4_ff_find_or_create_ds_client(struct pnfs_layout_segment *lseg,
149 u32 ds_idx,
150 struct nfs_client *ds_clp,
151 struct inode *inode);
152struct rpc_cred *ff_layout_get_ds_cred(struct pnfs_layout_segment *lseg,
153 u32 ds_idx, struct rpc_cred *mdscred);
154bool ff_layout_has_available_ds(struct pnfs_layout_segment *lseg);
155#endif /* FS_NFS_NFS4FLEXFILELAYOUT_H */
diff --git a/fs/nfs/flexfilelayout/flexfilelayoutdev.c b/fs/nfs/flexfilelayout/flexfilelayoutdev.c
new file mode 100644
index 000000000000..3bbb16b3066f
--- /dev/null
+++ b/fs/nfs/flexfilelayout/flexfilelayoutdev.c
@@ -0,0 +1,552 @@
1/*
2 * Device operations for the pnfs nfs4 file layout driver.
3 *
4 * Copyright (c) 2014, Primary Data, Inc. All rights reserved.
5 *
6 * Tao Peng <bergwolf@primarydata.com>
7 */
8
9#include <linux/nfs_fs.h>
10#include <linux/vmalloc.h>
11#include <linux/module.h>
12#include <linux/sunrpc/addr.h>
13
14#include "../internal.h"
15#include "../nfs4session.h"
16#include "flexfilelayout.h"
17
18#define NFSDBG_FACILITY NFSDBG_PNFS_LD
19
20static unsigned int dataserver_timeo = NFS4_DEF_DS_TIMEO;
21static unsigned int dataserver_retrans = NFS4_DEF_DS_RETRANS;
22
23void nfs4_ff_layout_put_deviceid(struct nfs4_ff_layout_ds *mirror_ds)
24{
25 if (mirror_ds)
26 nfs4_put_deviceid_node(&mirror_ds->id_node);
27}
28
29void nfs4_ff_layout_free_deviceid(struct nfs4_ff_layout_ds *mirror_ds)
30{
31 nfs4_print_deviceid(&mirror_ds->id_node.deviceid);
32 nfs4_pnfs_ds_put(mirror_ds->ds);
33 kfree(mirror_ds);
34}
35
36/* Decode opaque device data and construct new_ds using it */
37struct nfs4_ff_layout_ds *
38nfs4_ff_alloc_deviceid_node(struct nfs_server *server, struct pnfs_device *pdev,
39 gfp_t gfp_flags)
40{
41 struct xdr_stream stream;
42 struct xdr_buf buf;
43 struct page *scratch;
44 struct list_head dsaddrs;
45 struct nfs4_pnfs_ds_addr *da;
46 struct nfs4_ff_layout_ds *new_ds = NULL;
47 struct nfs4_ff_ds_version *ds_versions = NULL;
48 u32 mp_count;
49 u32 version_count;
50 __be32 *p;
51 int i, ret = -ENOMEM;
52
53 /* set up xdr stream */
54 scratch = alloc_page(gfp_flags);
55 if (!scratch)
56 goto out_err;
57
58 new_ds = kzalloc(sizeof(struct nfs4_ff_layout_ds), gfp_flags);
59 if (!new_ds)
60 goto out_scratch;
61
62 nfs4_init_deviceid_node(&new_ds->id_node,
63 server,
64 &pdev->dev_id);
65 INIT_LIST_HEAD(&dsaddrs);
66
67 xdr_init_decode_pages(&stream, &buf, pdev->pages, pdev->pglen);
68 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
69
70 /* multipath count */
71 p = xdr_inline_decode(&stream, 4);
72 if (unlikely(!p))
73 goto out_err_drain_dsaddrs;
74 mp_count = be32_to_cpup(p);
75 dprintk("%s: multipath ds count %d\n", __func__, mp_count);
76
77 for (i = 0; i < mp_count; i++) {
78 /* multipath ds */
79 da = nfs4_decode_mp_ds_addr(server->nfs_client->cl_net,
80 &stream, gfp_flags);
81 if (da)
82 list_add_tail(&da->da_node, &dsaddrs);
83 }
84 if (list_empty(&dsaddrs)) {
85 dprintk("%s: no suitable DS addresses found\n",
86 __func__);
87 ret = -ENOMEDIUM;
88 goto out_err_drain_dsaddrs;
89 }
90
91 /* version count */
92 p = xdr_inline_decode(&stream, 4);
93 if (unlikely(!p))
94 goto out_err_drain_dsaddrs;
95 version_count = be32_to_cpup(p);
96 dprintk("%s: version count %d\n", __func__, version_count);
97
98 ds_versions = kzalloc(version_count * sizeof(struct nfs4_ff_ds_version),
99 gfp_flags);
100 if (!ds_versions)
101 goto out_scratch;
102
103 for (i = 0; i < version_count; i++) {
104 /* 20 = version(4) + minor_version(4) + rsize(4) + wsize(4) +
105 * tightly_coupled(4) */
106 p = xdr_inline_decode(&stream, 20);
107 if (unlikely(!p))
108 goto out_err_drain_dsaddrs;
109 ds_versions[i].version = be32_to_cpup(p++);
110 ds_versions[i].minor_version = be32_to_cpup(p++);
111 ds_versions[i].rsize = nfs_block_size(be32_to_cpup(p++), NULL);
112 ds_versions[i].wsize = nfs_block_size(be32_to_cpup(p++), NULL);
113 ds_versions[i].tightly_coupled = be32_to_cpup(p);
114
115 if (ds_versions[i].rsize > NFS_MAX_FILE_IO_SIZE)
116 ds_versions[i].rsize = NFS_MAX_FILE_IO_SIZE;
117 if (ds_versions[i].wsize > NFS_MAX_FILE_IO_SIZE)
118 ds_versions[i].wsize = NFS_MAX_FILE_IO_SIZE;
119
120 if (ds_versions[i].version != 3 || ds_versions[i].minor_version != 0) {
121 dprintk("%s: [%d] unsupported ds version %d-%d\n", __func__,
122 i, ds_versions[i].version,
123 ds_versions[i].minor_version);
124 ret = -EPROTONOSUPPORT;
125 goto out_err_drain_dsaddrs;
126 }
127
128 dprintk("%s: [%d] vers %u minor_ver %u rsize %u wsize %u coupled %d\n",
129 __func__, i, ds_versions[i].version,
130 ds_versions[i].minor_version,
131 ds_versions[i].rsize,
132 ds_versions[i].wsize,
133 ds_versions[i].tightly_coupled);
134 }
135
136 new_ds->ds_versions = ds_versions;
137 new_ds->ds_versions_cnt = version_count;
138
139 new_ds->ds = nfs4_pnfs_ds_add(&dsaddrs, gfp_flags);
140 if (!new_ds->ds)
141 goto out_err_drain_dsaddrs;
142
143 /* If DS was already in cache, free ds addrs */
144 while (!list_empty(&dsaddrs)) {
145 da = list_first_entry(&dsaddrs,
146 struct nfs4_pnfs_ds_addr,
147 da_node);
148 list_del_init(&da->da_node);
149 kfree(da->da_remotestr);
150 kfree(da);
151 }
152
153 __free_page(scratch);
154 return new_ds;
155
156out_err_drain_dsaddrs:
157 while (!list_empty(&dsaddrs)) {
158 da = list_first_entry(&dsaddrs, struct nfs4_pnfs_ds_addr,
159 da_node);
160 list_del_init(&da->da_node);
161 kfree(da->da_remotestr);
162 kfree(da);
163 }
164
165 kfree(ds_versions);
166out_scratch:
167 __free_page(scratch);
168out_err:
169 kfree(new_ds);
170
171 dprintk("%s ERROR: returning %d\n", __func__, ret);
172 return NULL;
173}
174
175static u64
176end_offset(u64 start, u64 len)
177{
178 u64 end;
179
180 end = start + len;
181 return end >= start ? end : NFS4_MAX_UINT64;
182}
183
184static void extend_ds_error(struct nfs4_ff_layout_ds_err *err,
185 u64 offset, u64 length)
186{
187 u64 end;
188
189 end = max_t(u64, end_offset(err->offset, err->length),
190 end_offset(offset, length));
191 err->offset = min_t(u64, err->offset, offset);
192 err->length = end - err->offset;
193}
194
195static bool ds_error_can_merge(struct nfs4_ff_layout_ds_err *err, u64 offset,
196 u64 length, int status, enum nfs_opnum4 opnum,
197 nfs4_stateid *stateid,
198 struct nfs4_deviceid *deviceid)
199{
200 return err->status == status && err->opnum == opnum &&
201 nfs4_stateid_match(&err->stateid, stateid) &&
202 !memcmp(&err->deviceid, deviceid, sizeof(*deviceid)) &&
203 end_offset(err->offset, err->length) >= offset &&
204 err->offset <= end_offset(offset, length);
205}
206
207static bool merge_ds_error(struct nfs4_ff_layout_ds_err *old,
208 struct nfs4_ff_layout_ds_err *new)
209{
210 if (!ds_error_can_merge(old, new->offset, new->length, new->status,
211 new->opnum, &new->stateid, &new->deviceid))
212 return false;
213
214 extend_ds_error(old, new->offset, new->length);
215 return true;
216}
217
218static bool
219ff_layout_add_ds_error_locked(struct nfs4_flexfile_layout *flo,
220 struct nfs4_ff_layout_ds_err *dserr)
221{
222 struct nfs4_ff_layout_ds_err *err;
223
224 list_for_each_entry(err, &flo->error_list, list) {
225 if (merge_ds_error(err, dserr)) {
226 return true;
227 }
228 }
229
230 list_add(&dserr->list, &flo->error_list);
231 return false;
232}
233
234static bool
235ff_layout_update_ds_error(struct nfs4_flexfile_layout *flo, u64 offset,
236 u64 length, int status, enum nfs_opnum4 opnum,
237 nfs4_stateid *stateid, struct nfs4_deviceid *deviceid)
238{
239 bool found = false;
240 struct nfs4_ff_layout_ds_err *err;
241
242 list_for_each_entry(err, &flo->error_list, list) {
243 if (ds_error_can_merge(err, offset, length, status, opnum,
244 stateid, deviceid)) {
245 found = true;
246 extend_ds_error(err, offset, length);
247 break;
248 }
249 }
250
251 return found;
252}
253
254int ff_layout_track_ds_error(struct nfs4_flexfile_layout *flo,
255 struct nfs4_ff_layout_mirror *mirror, u64 offset,
256 u64 length, int status, enum nfs_opnum4 opnum,
257 gfp_t gfp_flags)
258{
259 struct nfs4_ff_layout_ds_err *dserr;
260 bool needfree;
261
262 if (status == 0)
263 return 0;
264
265 if (mirror->mirror_ds == NULL)
266 return -EINVAL;
267
268 spin_lock(&flo->generic_hdr.plh_inode->i_lock);
269 if (ff_layout_update_ds_error(flo, offset, length, status, opnum,
270 &mirror->stateid,
271 &mirror->mirror_ds->id_node.deviceid)) {
272 spin_unlock(&flo->generic_hdr.plh_inode->i_lock);
273 return 0;
274 }
275 spin_unlock(&flo->generic_hdr.plh_inode->i_lock);
276 dserr = kmalloc(sizeof(*dserr), gfp_flags);
277 if (!dserr)
278 return -ENOMEM;
279
280 INIT_LIST_HEAD(&dserr->list);
281 dserr->offset = offset;
282 dserr->length = length;
283 dserr->status = status;
284 dserr->opnum = opnum;
285 nfs4_stateid_copy(&dserr->stateid, &mirror->stateid);
286 memcpy(&dserr->deviceid, &mirror->mirror_ds->id_node.deviceid,
287 NFS4_DEVICEID4_SIZE);
288
289 spin_lock(&flo->generic_hdr.plh_inode->i_lock);
290 needfree = ff_layout_add_ds_error_locked(flo, dserr);
291 spin_unlock(&flo->generic_hdr.plh_inode->i_lock);
292 if (needfree)
293 kfree(dserr);
294
295 return 0;
296}
297
298/* currently we only support AUTH_NONE and AUTH_SYS */
299static rpc_authflavor_t
300nfs4_ff_layout_choose_authflavor(struct nfs4_ff_layout_mirror *mirror)
301{
302 if (mirror->uid == (u32)-1)
303 return RPC_AUTH_NULL;
304 return RPC_AUTH_UNIX;
305}
306
307/* fetch cred for NFSv3 DS */
308static int ff_layout_update_mirror_cred(struct nfs4_ff_layout_mirror *mirror,
309 struct nfs4_pnfs_ds *ds)
310{
311 if (ds->ds_clp && !mirror->cred &&
312 mirror->mirror_ds->ds_versions[0].version == 3) {
313 struct rpc_auth *auth = ds->ds_clp->cl_rpcclient->cl_auth;
314 struct rpc_cred *cred;
315 struct auth_cred acred = {
316 .uid = make_kuid(&init_user_ns, mirror->uid),
317 .gid = make_kgid(&init_user_ns, mirror->gid),
318 };
319
320 /* AUTH_NULL ignores acred */
321 cred = auth->au_ops->lookup_cred(auth, &acred, 0);
322 if (IS_ERR(cred)) {
323 dprintk("%s: lookup_cred failed with %ld\n",
324 __func__, PTR_ERR(cred));
325 return PTR_ERR(cred);
326 } else {
327 mirror->cred = cred;
328 }
329 }
330 return 0;
331}
332
333struct nfs_fh *
334nfs4_ff_layout_select_ds_fh(struct pnfs_layout_segment *lseg, u32 mirror_idx)
335{
336 struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, mirror_idx);
337 struct nfs_fh *fh = NULL;
338 struct nfs4_deviceid_node *devid;
339
340 if (mirror == NULL || mirror->mirror_ds == NULL ||
341 mirror->mirror_ds->ds == NULL) {
342 printk(KERN_ERR "NFS: %s: No data server for mirror offset index %d\n",
343 __func__, mirror_idx);
344 if (mirror && mirror->mirror_ds) {
345 devid = &mirror->mirror_ds->id_node;
346 pnfs_generic_mark_devid_invalid(devid);
347 }
348 goto out;
349 }
350
351 /* FIXME: For now assume there is only 1 version available for the DS */
352 fh = &mirror->fh_versions[0];
353out:
354 return fh;
355}
356
357/* Upon return, either ds is connected, or ds is NULL */
358struct nfs4_pnfs_ds *
359nfs4_ff_layout_prepare_ds(struct pnfs_layout_segment *lseg, u32 ds_idx,
360 bool fail_return)
361{
362 struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, ds_idx);
363 struct nfs4_pnfs_ds *ds = NULL;
364 struct nfs4_deviceid_node *devid;
365 struct inode *ino = lseg->pls_layout->plh_inode;
366 struct nfs_server *s = NFS_SERVER(ino);
367 unsigned int max_payload;
368 rpc_authflavor_t flavor;
369
370 if (mirror == NULL || mirror->mirror_ds == NULL ||
371 mirror->mirror_ds->ds == NULL) {
372 printk(KERN_ERR "NFS: %s: No data server for offset index %d\n",
373 __func__, ds_idx);
374 if (mirror && mirror->mirror_ds) {
375 devid = &mirror->mirror_ds->id_node;
376 pnfs_generic_mark_devid_invalid(devid);
377 }
378 goto out;
379 }
380
381 devid = &mirror->mirror_ds->id_node;
382 if (ff_layout_test_devid_unavailable(devid))
383 goto out;
384
385 ds = mirror->mirror_ds->ds;
386 /* matching smp_wmb() in _nfs4_pnfs_v3/4_ds_connect */
387 smp_rmb();
388 if (ds->ds_clp)
389 goto out;
390
391 flavor = nfs4_ff_layout_choose_authflavor(mirror);
392
393 /* FIXME: For now we assume the server sent only one version of NFS
394 * to use for the DS.
395 */
396 nfs4_pnfs_ds_connect(s, ds, devid, dataserver_timeo,
397 dataserver_retrans,
398 mirror->mirror_ds->ds_versions[0].version,
399 mirror->mirror_ds->ds_versions[0].minor_version,
400 flavor);
401
402 /* connect success, check rsize/wsize limit */
403 if (ds->ds_clp) {
404 max_payload =
405 nfs_block_size(rpc_max_payload(ds->ds_clp->cl_rpcclient),
406 NULL);
407 if (mirror->mirror_ds->ds_versions[0].rsize > max_payload)
408 mirror->mirror_ds->ds_versions[0].rsize = max_payload;
409 if (mirror->mirror_ds->ds_versions[0].wsize > max_payload)
410 mirror->mirror_ds->ds_versions[0].wsize = max_payload;
411 } else {
412 ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
413 mirror, lseg->pls_range.offset,
414 lseg->pls_range.length, NFS4ERR_NXIO,
415 OP_ILLEGAL, GFP_NOIO);
416 if (fail_return) {
417 pnfs_error_mark_layout_for_return(ino, lseg);
418 if (ff_layout_has_available_ds(lseg))
419 pnfs_set_retry_layoutget(lseg->pls_layout);
420 else
421 pnfs_clear_retry_layoutget(lseg->pls_layout);
422
423 } else {
424 if (ff_layout_has_available_ds(lseg))
425 set_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE,
426 &lseg->pls_layout->plh_flags);
427 else {
428 pnfs_error_mark_layout_for_return(ino, lseg);
429 pnfs_clear_retry_layoutget(lseg->pls_layout);
430 }
431 }
432 }
433
434 if (ff_layout_update_mirror_cred(mirror, ds))
435 ds = NULL;
436out:
437 return ds;
438}
439
440struct rpc_cred *
441ff_layout_get_ds_cred(struct pnfs_layout_segment *lseg, u32 ds_idx,
442 struct rpc_cred *mdscred)
443{
444 struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, ds_idx);
445 struct rpc_cred *cred = ERR_PTR(-EINVAL);
446
447 if (!nfs4_ff_layout_prepare_ds(lseg, ds_idx, true))
448 goto out;
449
450 if (mirror && mirror->cred)
451 cred = mirror->cred;
452 else
453 cred = mdscred;
454out:
455 return cred;
456}
457
458/**
459* Find or create a DS rpc client with th MDS server rpc client auth flavor
460* in the nfs_client cl_ds_clients list.
461*/
462struct rpc_clnt *
463nfs4_ff_find_or_create_ds_client(struct pnfs_layout_segment *lseg, u32 ds_idx,
464 struct nfs_client *ds_clp, struct inode *inode)
465{
466 struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, ds_idx);
467
468 switch (mirror->mirror_ds->ds_versions[0].version) {
469 case 3:
470 /* For NFSv3 DS, flavor is set when creating DS connections */
471 return ds_clp->cl_rpcclient;
472 case 4:
473 return nfs4_find_or_create_ds_client(ds_clp, inode);
474 default:
475 BUG();
476 }
477}
478
479static bool is_range_intersecting(u64 offset1, u64 length1,
480 u64 offset2, u64 length2)
481{
482 u64 end1 = end_offset(offset1, length1);
483 u64 end2 = end_offset(offset2, length2);
484
485 return (end1 == NFS4_MAX_UINT64 || end1 > offset2) &&
486 (end2 == NFS4_MAX_UINT64 || end2 > offset1);
487}
488
489/* called with inode i_lock held */
490int ff_layout_encode_ds_ioerr(struct nfs4_flexfile_layout *flo,
491 struct xdr_stream *xdr, int *count,
492 const struct pnfs_layout_range *range)
493{
494 struct nfs4_ff_layout_ds_err *err, *n;
495 __be32 *p;
496
497 list_for_each_entry_safe(err, n, &flo->error_list, list) {
498 if (!is_range_intersecting(err->offset, err->length,
499 range->offset, range->length))
500 continue;
501 /* offset(8) + length(8) + stateid(NFS4_STATEID_SIZE)
502 * + deviceid(NFS4_DEVICEID4_SIZE) + status(4) + opnum(4)
503 */
504 p = xdr_reserve_space(xdr,
505 24 + NFS4_STATEID_SIZE + NFS4_DEVICEID4_SIZE);
506 if (unlikely(!p))
507 return -ENOBUFS;
508 p = xdr_encode_hyper(p, err->offset);
509 p = xdr_encode_hyper(p, err->length);
510 p = xdr_encode_opaque_fixed(p, &err->stateid,
511 NFS4_STATEID_SIZE);
512 p = xdr_encode_opaque_fixed(p, &err->deviceid,
513 NFS4_DEVICEID4_SIZE);
514 *p++ = cpu_to_be32(err->status);
515 *p++ = cpu_to_be32(err->opnum);
516 *count += 1;
517 list_del(&err->list);
518 kfree(err);
519 dprintk("%s: offset %llu length %llu status %d op %d count %d\n",
520 __func__, err->offset, err->length, err->status,
521 err->opnum, *count);
522 }
523
524 return 0;
525}
526
527bool ff_layout_has_available_ds(struct pnfs_layout_segment *lseg)
528{
529 struct nfs4_ff_layout_mirror *mirror;
530 struct nfs4_deviceid_node *devid;
531 int idx;
532
533 for (idx = 0; idx < FF_LAYOUT_MIRROR_COUNT(lseg); idx++) {
534 mirror = FF_LAYOUT_COMP(lseg, idx);
535 if (mirror && mirror->mirror_ds) {
536 devid = &mirror->mirror_ds->id_node;
537 if (!ff_layout_test_devid_unavailable(devid))
538 return true;
539 }
540 }
541
542 return false;
543}
544
545module_param(dataserver_retrans, uint, 0644);
546MODULE_PARM_DESC(dataserver_retrans, "The number of times the NFSv4.1 client "
547 "retries a request before it attempts further "
548 " recovery action.");
549module_param(dataserver_timeo, uint, 0644);
550MODULE_PARM_DESC(dataserver_timeo, "The time (in tenths of a second) the "
551 "NFSv4.1 client waits for a response from a "
552 " data server before it retries an NFS request.");
diff --git a/fs/nfs/idmap.c b/fs/nfs/idmap.c
index 2f5db844c172..857e2a99acc8 100644
--- a/fs/nfs/idmap.c
+++ b/fs/nfs/idmap.c
@@ -152,7 +152,7 @@ void nfs_fattr_map_and_free_names(struct nfs_server *server, struct nfs_fattr *f
152 nfs_fattr_free_group_name(fattr); 152 nfs_fattr_free_group_name(fattr);
153} 153}
154 154
155static int nfs_map_string_to_numeric(const char *name, size_t namelen, __u32 *res) 155int nfs_map_string_to_numeric(const char *name, size_t namelen, __u32 *res)
156{ 156{
157 unsigned long val; 157 unsigned long val;
158 char buf[16]; 158 char buf[16];
@@ -166,6 +166,7 @@ static int nfs_map_string_to_numeric(const char *name, size_t namelen, __u32 *re
166 *res = val; 166 *res = val;
167 return 1; 167 return 1;
168} 168}
169EXPORT_SYMBOL_GPL(nfs_map_string_to_numeric);
169 170
170static int nfs_map_numeric_to_string(__u32 id, char *buf, size_t buflen) 171static int nfs_map_numeric_to_string(__u32 id, char *buf, size_t buflen)
171{ 172{
diff --git a/fs/nfs/internal.h b/fs/nfs/internal.h
index b6f34bfa6fe8..a98cf2006179 100644
--- a/fs/nfs/internal.h
+++ b/fs/nfs/internal.h
@@ -6,6 +6,7 @@
6#include <linux/mount.h> 6#include <linux/mount.h>
7#include <linux/security.h> 7#include <linux/security.h>
8#include <linux/crc32.h> 8#include <linux/crc32.h>
9#include <linux/nfs_page.h>
9 10
10#define NFS_MS_MASK (MS_RDONLY|MS_NOSUID|MS_NODEV|MS_NOEXEC|MS_SYNCHRONOUS) 11#define NFS_MS_MASK (MS_RDONLY|MS_NOSUID|MS_NODEV|MS_NOEXEC|MS_SYNCHRONOUS)
11 12
@@ -187,9 +188,15 @@ extern struct nfs_client *nfs4_set_ds_client(struct nfs_client* mds_clp,
187 const struct sockaddr *ds_addr, 188 const struct sockaddr *ds_addr,
188 int ds_addrlen, int ds_proto, 189 int ds_addrlen, int ds_proto,
189 unsigned int ds_timeo, 190 unsigned int ds_timeo,
190 unsigned int ds_retrans); 191 unsigned int ds_retrans,
192 u32 minor_version,
193 rpc_authflavor_t au_flavor);
191extern struct rpc_clnt *nfs4_find_or_create_ds_client(struct nfs_client *, 194extern struct rpc_clnt *nfs4_find_or_create_ds_client(struct nfs_client *,
192 struct inode *); 195 struct inode *);
196extern struct nfs_client *nfs3_set_ds_client(struct nfs_client *mds_clp,
197 const struct sockaddr *ds_addr, int ds_addrlen,
198 int ds_proto, unsigned int ds_timeo,
199 unsigned int ds_retrans, rpc_authflavor_t au_flavor);
193#ifdef CONFIG_PROC_FS 200#ifdef CONFIG_PROC_FS
194extern int __init nfs_fs_proc_init(void); 201extern int __init nfs_fs_proc_init(void);
195extern void nfs_fs_proc_exit(void); 202extern void nfs_fs_proc_exit(void);
@@ -242,9 +249,12 @@ struct nfs_pgio_header *nfs_pgio_header_alloc(const struct nfs_rw_ops *);
242void nfs_pgio_header_free(struct nfs_pgio_header *); 249void nfs_pgio_header_free(struct nfs_pgio_header *);
243void nfs_pgio_data_destroy(struct nfs_pgio_header *); 250void nfs_pgio_data_destroy(struct nfs_pgio_header *);
244int nfs_generic_pgio(struct nfs_pageio_descriptor *, struct nfs_pgio_header *); 251int nfs_generic_pgio(struct nfs_pageio_descriptor *, struct nfs_pgio_header *);
245int nfs_initiate_pgio(struct rpc_clnt *, struct nfs_pgio_header *, 252int nfs_initiate_pgio(struct rpc_clnt *clnt, struct nfs_pgio_header *hdr,
246 const struct rpc_call_ops *, int, int); 253 struct rpc_cred *cred, const struct nfs_rpc_ops *rpc_ops,
254 const struct rpc_call_ops *call_ops, int how, int flags);
247void nfs_free_request(struct nfs_page *req); 255void nfs_free_request(struct nfs_page *req);
256struct nfs_pgio_mirror *
257nfs_pgio_current_mirror(struct nfs_pageio_descriptor *desc);
248 258
249static inline void nfs_iocounter_init(struct nfs_io_counter *c) 259static inline void nfs_iocounter_init(struct nfs_io_counter *c)
250{ 260{
@@ -252,6 +262,12 @@ static inline void nfs_iocounter_init(struct nfs_io_counter *c)
252 atomic_set(&c->io_count, 0); 262 atomic_set(&c->io_count, 0);
253} 263}
254 264
265static inline bool nfs_pgio_has_mirroring(struct nfs_pageio_descriptor *desc)
266{
267 WARN_ON_ONCE(desc->pg_mirror_count < 1);
268 return desc->pg_mirror_count > 1;
269}
270
255/* nfs2xdr.c */ 271/* nfs2xdr.c */
256extern struct rpc_procinfo nfs_procedures[]; 272extern struct rpc_procinfo nfs_procedures[];
257extern int nfs2_decode_dirent(struct xdr_stream *, 273extern int nfs2_decode_dirent(struct xdr_stream *,
@@ -427,6 +443,7 @@ extern void nfs_write_prepare(struct rpc_task *task, void *calldata);
427extern void nfs_commit_prepare(struct rpc_task *task, void *calldata); 443extern void nfs_commit_prepare(struct rpc_task *task, void *calldata);
428extern int nfs_initiate_commit(struct rpc_clnt *clnt, 444extern int nfs_initiate_commit(struct rpc_clnt *clnt,
429 struct nfs_commit_data *data, 445 struct nfs_commit_data *data,
446 const struct nfs_rpc_ops *nfs_ops,
430 const struct rpc_call_ops *call_ops, 447 const struct rpc_call_ops *call_ops,
431 int how, int flags); 448 int how, int flags);
432extern void nfs_init_commit(struct nfs_commit_data *data, 449extern void nfs_init_commit(struct nfs_commit_data *data,
@@ -440,13 +457,15 @@ int nfs_scan_commit(struct inode *inode, struct list_head *dst,
440 struct nfs_commit_info *cinfo); 457 struct nfs_commit_info *cinfo);
441void nfs_mark_request_commit(struct nfs_page *req, 458void nfs_mark_request_commit(struct nfs_page *req,
442 struct pnfs_layout_segment *lseg, 459 struct pnfs_layout_segment *lseg,
443 struct nfs_commit_info *cinfo); 460 struct nfs_commit_info *cinfo,
461 u32 ds_commit_idx);
444int nfs_write_need_commit(struct nfs_pgio_header *); 462int nfs_write_need_commit(struct nfs_pgio_header *);
445int nfs_generic_commit_list(struct inode *inode, struct list_head *head, 463int nfs_generic_commit_list(struct inode *inode, struct list_head *head,
446 int how, struct nfs_commit_info *cinfo); 464 int how, struct nfs_commit_info *cinfo);
447void nfs_retry_commit(struct list_head *page_list, 465void nfs_retry_commit(struct list_head *page_list,
448 struct pnfs_layout_segment *lseg, 466 struct pnfs_layout_segment *lseg,
449 struct nfs_commit_info *cinfo); 467 struct nfs_commit_info *cinfo,
468 u32 ds_commit_idx);
450void nfs_commitdata_release(struct nfs_commit_data *data); 469void nfs_commitdata_release(struct nfs_commit_data *data);
451void nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst, 470void nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst,
452 struct nfs_commit_info *cinfo); 471 struct nfs_commit_info *cinfo);
@@ -457,6 +476,7 @@ void nfs_init_cinfo(struct nfs_commit_info *cinfo,
457 struct nfs_direct_req *dreq); 476 struct nfs_direct_req *dreq);
458int nfs_key_timeout_notify(struct file *filp, struct inode *inode); 477int nfs_key_timeout_notify(struct file *filp, struct inode *inode);
459bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx); 478bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx);
479void nfs_pageio_stop_mirroring(struct nfs_pageio_descriptor *pgio);
460 480
461#ifdef CONFIG_MIGRATION 481#ifdef CONFIG_MIGRATION
462extern int nfs_migrate_page(struct address_space *, 482extern int nfs_migrate_page(struct address_space *,
@@ -480,6 +500,7 @@ static inline void nfs_inode_dio_wait(struct inode *inode)
480 inode_dio_wait(inode); 500 inode_dio_wait(inode);
481} 501}
482extern ssize_t nfs_dreq_bytes_left(struct nfs_direct_req *dreq); 502extern ssize_t nfs_dreq_bytes_left(struct nfs_direct_req *dreq);
503extern void nfs_direct_set_resched_writes(struct nfs_direct_req *dreq);
483 504
484/* nfs4proc.c */ 505/* nfs4proc.c */
485extern void __nfs4_read_done_cb(struct nfs_pgio_header *); 506extern void __nfs4_read_done_cb(struct nfs_pgio_header *);
diff --git a/fs/nfs/nfs2xdr.c b/fs/nfs/nfs2xdr.c
index 5f61b83f4a1c..b4e03ed8599d 100644
--- a/fs/nfs/nfs2xdr.c
+++ b/fs/nfs/nfs2xdr.c
@@ -481,7 +481,8 @@ out_overflow:
481 * void; 481 * void;
482 * }; 482 * };
483 */ 483 */
484static int decode_attrstat(struct xdr_stream *xdr, struct nfs_fattr *result) 484static int decode_attrstat(struct xdr_stream *xdr, struct nfs_fattr *result,
485 __u32 *op_status)
485{ 486{
486 enum nfs_stat status; 487 enum nfs_stat status;
487 int error; 488 int error;
@@ -489,6 +490,8 @@ static int decode_attrstat(struct xdr_stream *xdr, struct nfs_fattr *result)
489 error = decode_stat(xdr, &status); 490 error = decode_stat(xdr, &status);
490 if (unlikely(error)) 491 if (unlikely(error))
491 goto out; 492 goto out;
493 if (op_status)
494 *op_status = status;
492 if (status != NFS_OK) 495 if (status != NFS_OK)
493 goto out_default; 496 goto out_default;
494 error = decode_fattr(xdr, result); 497 error = decode_fattr(xdr, result);
@@ -808,7 +811,7 @@ out_default:
808static int nfs2_xdr_dec_attrstat(struct rpc_rqst *req, struct xdr_stream *xdr, 811static int nfs2_xdr_dec_attrstat(struct rpc_rqst *req, struct xdr_stream *xdr,
809 struct nfs_fattr *result) 812 struct nfs_fattr *result)
810{ 813{
811 return decode_attrstat(xdr, result); 814 return decode_attrstat(xdr, result, NULL);
812} 815}
813 816
814static int nfs2_xdr_dec_diropres(struct rpc_rqst *req, struct xdr_stream *xdr, 817static int nfs2_xdr_dec_diropres(struct rpc_rqst *req, struct xdr_stream *xdr,
@@ -865,6 +868,7 @@ static int nfs2_xdr_dec_readres(struct rpc_rqst *req, struct xdr_stream *xdr,
865 error = decode_stat(xdr, &status); 868 error = decode_stat(xdr, &status);
866 if (unlikely(error)) 869 if (unlikely(error))
867 goto out; 870 goto out;
871 result->op_status = status;
868 if (status != NFS_OK) 872 if (status != NFS_OK)
869 goto out_default; 873 goto out_default;
870 error = decode_fattr(xdr, result->fattr); 874 error = decode_fattr(xdr, result->fattr);
@@ -882,7 +886,7 @@ static int nfs2_xdr_dec_writeres(struct rpc_rqst *req, struct xdr_stream *xdr,
882{ 886{
883 /* All NFSv2 writes are "file sync" writes */ 887 /* All NFSv2 writes are "file sync" writes */
884 result->verf->committed = NFS_FILE_SYNC; 888 result->verf->committed = NFS_FILE_SYNC;
885 return decode_attrstat(xdr, result->fattr); 889 return decode_attrstat(xdr, result->fattr, &result->op_status);
886} 890}
887 891
888/** 892/**
diff --git a/fs/nfs/nfs3_fs.h b/fs/nfs/nfs3_fs.h
index 333ae4068506..e134d6548ab7 100644
--- a/fs/nfs/nfs3_fs.h
+++ b/fs/nfs/nfs3_fs.h
@@ -30,5 +30,7 @@ struct nfs_server *nfs3_create_server(struct nfs_mount_info *, struct nfs_subver
30struct nfs_server *nfs3_clone_server(struct nfs_server *, struct nfs_fh *, 30struct nfs_server *nfs3_clone_server(struct nfs_server *, struct nfs_fh *,
31 struct nfs_fattr *, rpc_authflavor_t); 31 struct nfs_fattr *, rpc_authflavor_t);
32 32
33/* nfs3super.c */
34extern struct nfs_subversion nfs_v3;
33 35
34#endif /* __LINUX_FS_NFS_NFS3_FS_H */ 36#endif /* __LINUX_FS_NFS_NFS3_FS_H */
diff --git a/fs/nfs/nfs3client.c b/fs/nfs/nfs3client.c
index 8c1b437c5403..9e9fa347a948 100644
--- a/fs/nfs/nfs3client.c
+++ b/fs/nfs/nfs3client.c
@@ -1,5 +1,6 @@
1#include <linux/nfs_fs.h> 1#include <linux/nfs_fs.h>
2#include <linux/nfs_mount.h> 2#include <linux/nfs_mount.h>
3#include <linux/sunrpc/addr.h>
3#include "internal.h" 4#include "internal.h"
4#include "nfs3_fs.h" 5#include "nfs3_fs.h"
5 6
@@ -64,3 +65,43 @@ struct nfs_server *nfs3_clone_server(struct nfs_server *source,
64 nfs_init_server_aclclient(server); 65 nfs_init_server_aclclient(server);
65 return server; 66 return server;
66} 67}
68
69/*
70 * Set up a pNFS Data Server client over NFSv3.
71 *
72 * Return any existing nfs_client that matches server address,port,version
73 * and minorversion.
74 *
75 * For a new nfs_client, use a soft mount (default), a low retrans and a
76 * low timeout interval so that if a connection is lost, we retry through
77 * the MDS.
78 */
79struct nfs_client *nfs3_set_ds_client(struct nfs_client *mds_clp,
80 const struct sockaddr *ds_addr, int ds_addrlen,
81 int ds_proto, unsigned int ds_timeo, unsigned int ds_retrans,
82 rpc_authflavor_t au_flavor)
83{
84 struct nfs_client_initdata cl_init = {
85 .addr = ds_addr,
86 .addrlen = ds_addrlen,
87 .nfs_mod = &nfs_v3,
88 .proto = ds_proto,
89 .net = mds_clp->cl_net,
90 };
91 struct rpc_timeout ds_timeout;
92 struct nfs_client *clp;
93 char buf[INET6_ADDRSTRLEN + 1];
94
95 /* fake a hostname because lockd wants it */
96 if (rpc_ntop(ds_addr, buf, sizeof(buf)) <= 0)
97 return ERR_PTR(-EINVAL);
98 cl_init.hostname = buf;
99
100 /* Use the MDS nfs_client cl_ipaddr. */
101 nfs_init_timeout_values(&ds_timeout, ds_proto, ds_timeo, ds_retrans);
102 clp = nfs_get_client(&cl_init, &ds_timeout, mds_clp->cl_ipaddr,
103 au_flavor);
104
105 return clp;
106}
107EXPORT_SYMBOL_GPL(nfs3_set_ds_client);
diff --git a/fs/nfs/nfs3proc.c b/fs/nfs/nfs3proc.c
index 524f9f837408..78e557c3ab87 100644
--- a/fs/nfs/nfs3proc.c
+++ b/fs/nfs/nfs3proc.c
@@ -800,6 +800,9 @@ static int nfs3_read_done(struct rpc_task *task, struct nfs_pgio_header *hdr)
800{ 800{
801 struct inode *inode = hdr->inode; 801 struct inode *inode = hdr->inode;
802 802
803 if (hdr->pgio_done_cb != NULL)
804 return hdr->pgio_done_cb(task, hdr);
805
803 if (nfs3_async_handle_jukebox(task, inode)) 806 if (nfs3_async_handle_jukebox(task, inode))
804 return -EAGAIN; 807 return -EAGAIN;
805 808
@@ -825,6 +828,9 @@ static int nfs3_write_done(struct rpc_task *task, struct nfs_pgio_header *hdr)
825{ 828{
826 struct inode *inode = hdr->inode; 829 struct inode *inode = hdr->inode;
827 830
831 if (hdr->pgio_done_cb != NULL)
832 return hdr->pgio_done_cb(task, hdr);
833
828 if (nfs3_async_handle_jukebox(task, inode)) 834 if (nfs3_async_handle_jukebox(task, inode))
829 return -EAGAIN; 835 return -EAGAIN;
830 if (task->tk_status >= 0) 836 if (task->tk_status >= 0)
@@ -845,6 +851,9 @@ static void nfs3_proc_commit_rpc_prepare(struct rpc_task *task, struct nfs_commi
845 851
846static int nfs3_commit_done(struct rpc_task *task, struct nfs_commit_data *data) 852static int nfs3_commit_done(struct rpc_task *task, struct nfs_commit_data *data)
847{ 853{
854 if (data->commit_done_cb != NULL)
855 return data->commit_done_cb(task, data);
856
848 if (nfs3_async_handle_jukebox(task, data->inode)) 857 if (nfs3_async_handle_jukebox(task, data->inode))
849 return -EAGAIN; 858 return -EAGAIN;
850 nfs_refresh_inode(data->inode, data->res.fattr); 859 nfs_refresh_inode(data->inode, data->res.fattr);
diff --git a/fs/nfs/nfs3super.c b/fs/nfs/nfs3super.c
index 6af29c2da352..5c4394e4656b 100644
--- a/fs/nfs/nfs3super.c
+++ b/fs/nfs/nfs3super.c
@@ -7,7 +7,7 @@
7#include "nfs3_fs.h" 7#include "nfs3_fs.h"
8#include "nfs.h" 8#include "nfs.h"
9 9
10static struct nfs_subversion nfs_v3 = { 10struct nfs_subversion nfs_v3 = {
11 .owner = THIS_MODULE, 11 .owner = THIS_MODULE,
12 .nfs_fs = &nfs_fs_type, 12 .nfs_fs = &nfs_fs_type,
13 .rpc_vers = &nfs_version3, 13 .rpc_vers = &nfs_version3,
diff --git a/fs/nfs/nfs3xdr.c b/fs/nfs/nfs3xdr.c
index 8f4cbe7f4aa8..2a932fdc57cb 100644
--- a/fs/nfs/nfs3xdr.c
+++ b/fs/nfs/nfs3xdr.c
@@ -1636,6 +1636,7 @@ static int nfs3_xdr_dec_read3res(struct rpc_rqst *req, struct xdr_stream *xdr,
1636 error = decode_post_op_attr(xdr, result->fattr); 1636 error = decode_post_op_attr(xdr, result->fattr);
1637 if (unlikely(error)) 1637 if (unlikely(error))
1638 goto out; 1638 goto out;
1639 result->op_status = status;
1639 if (status != NFS3_OK) 1640 if (status != NFS3_OK)
1640 goto out_status; 1641 goto out_status;
1641 error = decode_read3resok(xdr, result); 1642 error = decode_read3resok(xdr, result);
@@ -1708,6 +1709,7 @@ static int nfs3_xdr_dec_write3res(struct rpc_rqst *req, struct xdr_stream *xdr,
1708 error = decode_wcc_data(xdr, result->fattr); 1709 error = decode_wcc_data(xdr, result->fattr);
1709 if (unlikely(error)) 1710 if (unlikely(error))
1710 goto out; 1711 goto out;
1712 result->op_status = status;
1711 if (status != NFS3_OK) 1713 if (status != NFS3_OK)
1712 goto out_status; 1714 goto out_status;
1713 error = decode_write3resok(xdr, result); 1715 error = decode_write3resok(xdr, result);
@@ -2323,6 +2325,7 @@ static int nfs3_xdr_dec_commit3res(struct rpc_rqst *req,
2323 error = decode_wcc_data(xdr, result->fattr); 2325 error = decode_wcc_data(xdr, result->fattr);
2324 if (unlikely(error)) 2326 if (unlikely(error))
2325 goto out; 2327 goto out;
2328 result->op_status = status;
2326 if (status != NFS3_OK) 2329 if (status != NFS3_OK)
2327 goto out_status; 2330 goto out_status;
2328 error = decode_writeverf3(xdr, &result->verf->verifier); 2331 error = decode_writeverf3(xdr, &result->verf->verifier);
diff --git a/fs/nfs/nfs4_fs.h b/fs/nfs/nfs4_fs.h
index e57290a66b76..fdef424b0cd3 100644
--- a/fs/nfs/nfs4_fs.h
+++ b/fs/nfs/nfs4_fs.h
@@ -446,6 +446,12 @@ extern void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid);
446extern void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid); 446extern void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid);
447extern void nfs_release_seqid(struct nfs_seqid *seqid); 447extern void nfs_release_seqid(struct nfs_seqid *seqid);
448extern void nfs_free_seqid(struct nfs_seqid *seqid); 448extern void nfs_free_seqid(struct nfs_seqid *seqid);
449extern int nfs40_setup_sequence(struct nfs4_slot_table *tbl,
450 struct nfs4_sequence_args *args,
451 struct nfs4_sequence_res *res,
452 struct rpc_task *task);
453extern int nfs4_sequence_done(struct rpc_task *task,
454 struct nfs4_sequence_res *res);
449 455
450extern void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp); 456extern void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp);
451 457
diff --git a/fs/nfs/nfs4client.c b/fs/nfs/nfs4client.c
index 706ad10b8186..8646af9b11d2 100644
--- a/fs/nfs/nfs4client.c
+++ b/fs/nfs/nfs4client.c
@@ -849,14 +849,15 @@ error:
849 */ 849 */
850struct nfs_client *nfs4_set_ds_client(struct nfs_client* mds_clp, 850struct nfs_client *nfs4_set_ds_client(struct nfs_client* mds_clp,
851 const struct sockaddr *ds_addr, int ds_addrlen, 851 const struct sockaddr *ds_addr, int ds_addrlen,
852 int ds_proto, unsigned int ds_timeo, unsigned int ds_retrans) 852 int ds_proto, unsigned int ds_timeo, unsigned int ds_retrans,
853 u32 minor_version, rpc_authflavor_t au_flavor)
853{ 854{
854 struct nfs_client_initdata cl_init = { 855 struct nfs_client_initdata cl_init = {
855 .addr = ds_addr, 856 .addr = ds_addr,
856 .addrlen = ds_addrlen, 857 .addrlen = ds_addrlen,
857 .nfs_mod = &nfs_v4, 858 .nfs_mod = &nfs_v4,
858 .proto = ds_proto, 859 .proto = ds_proto,
859 .minorversion = mds_clp->cl_minorversion, 860 .minorversion = minor_version,
860 .net = mds_clp->cl_net, 861 .net = mds_clp->cl_net,
861 }; 862 };
862 struct rpc_timeout ds_timeout; 863 struct rpc_timeout ds_timeout;
@@ -874,7 +875,7 @@ struct nfs_client *nfs4_set_ds_client(struct nfs_client* mds_clp,
874 */ 875 */
875 nfs_init_timeout_values(&ds_timeout, ds_proto, ds_timeo, ds_retrans); 876 nfs_init_timeout_values(&ds_timeout, ds_proto, ds_timeo, ds_retrans);
876 clp = nfs_get_client(&cl_init, &ds_timeout, mds_clp->cl_ipaddr, 877 clp = nfs_get_client(&cl_init, &ds_timeout, mds_clp->cl_ipaddr,
877 mds_clp->cl_rpcclient->cl_auth->au_flavor); 878 au_flavor);
878 879
879 dprintk("<-- %s %p\n", __func__, clp); 880 dprintk("<-- %s %p\n", __func__, clp);
880 return clp; 881 return clp;
diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index 0f75b9276726..6e1c9b2d92c5 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -495,12 +495,11 @@ static void nfs4_set_sequence_privileged(struct nfs4_sequence_args *args)
495 args->sa_privileged = 1; 495 args->sa_privileged = 1;
496} 496}
497 497
498static int nfs40_setup_sequence(const struct nfs_server *server, 498int nfs40_setup_sequence(struct nfs4_slot_table *tbl,
499 struct nfs4_sequence_args *args, 499 struct nfs4_sequence_args *args,
500 struct nfs4_sequence_res *res, 500 struct nfs4_sequence_res *res,
501 struct rpc_task *task) 501 struct rpc_task *task)
502{ 502{
503 struct nfs4_slot_table *tbl = server->nfs_client->cl_slot_tbl;
504 struct nfs4_slot *slot; 503 struct nfs4_slot *slot;
505 504
506 /* slot already allocated? */ 505 /* slot already allocated? */
@@ -535,6 +534,7 @@ out_sleep:
535 spin_unlock(&tbl->slot_tbl_lock); 534 spin_unlock(&tbl->slot_tbl_lock);
536 return -EAGAIN; 535 return -EAGAIN;
537} 536}
537EXPORT_SYMBOL_GPL(nfs40_setup_sequence);
538 538
539static int nfs40_sequence_done(struct rpc_task *task, 539static int nfs40_sequence_done(struct rpc_task *task,
540 struct nfs4_sequence_res *res) 540 struct nfs4_sequence_res *res)
@@ -694,8 +694,7 @@ out_retry:
694} 694}
695EXPORT_SYMBOL_GPL(nfs41_sequence_done); 695EXPORT_SYMBOL_GPL(nfs41_sequence_done);
696 696
697static int nfs4_sequence_done(struct rpc_task *task, 697int nfs4_sequence_done(struct rpc_task *task, struct nfs4_sequence_res *res)
698 struct nfs4_sequence_res *res)
699{ 698{
700 if (res->sr_slot == NULL) 699 if (res->sr_slot == NULL)
701 return 1; 700 return 1;
@@ -703,6 +702,7 @@ static int nfs4_sequence_done(struct rpc_task *task,
703 return nfs40_sequence_done(task, res); 702 return nfs40_sequence_done(task, res);
704 return nfs41_sequence_done(task, res); 703 return nfs41_sequence_done(task, res);
705} 704}
705EXPORT_SYMBOL_GPL(nfs4_sequence_done);
706 706
707int nfs41_setup_sequence(struct nfs4_session *session, 707int nfs41_setup_sequence(struct nfs4_session *session,
708 struct nfs4_sequence_args *args, 708 struct nfs4_sequence_args *args,
@@ -777,7 +777,8 @@ static int nfs4_setup_sequence(const struct nfs_server *server,
777 int ret = 0; 777 int ret = 0;
778 778
779 if (!session) 779 if (!session)
780 return nfs40_setup_sequence(server, args, res, task); 780 return nfs40_setup_sequence(server->nfs_client->cl_slot_tbl,
781 args, res, task);
781 782
782 dprintk("--> %s clp %p session %p sr_slot %u\n", 783 dprintk("--> %s clp %p session %p sr_slot %u\n",
783 __func__, session->clp, session, res->sr_slot ? 784 __func__, session->clp, session, res->sr_slot ?
@@ -818,14 +819,16 @@ static int nfs4_setup_sequence(const struct nfs_server *server,
818 struct nfs4_sequence_res *res, 819 struct nfs4_sequence_res *res,
819 struct rpc_task *task) 820 struct rpc_task *task)
820{ 821{
821 return nfs40_setup_sequence(server, args, res, task); 822 return nfs40_setup_sequence(server->nfs_client->cl_slot_tbl,
823 args, res, task);
822} 824}
823 825
824static int nfs4_sequence_done(struct rpc_task *task, 826int nfs4_sequence_done(struct rpc_task *task,
825 struct nfs4_sequence_res *res) 827 struct nfs4_sequence_res *res)
826{ 828{
827 return nfs40_sequence_done(task, res); 829 return nfs40_sequence_done(task, res);
828} 830}
831EXPORT_SYMBOL_GPL(nfs4_sequence_done);
829 832
830#endif /* !CONFIG_NFS_V4_1 */ 833#endif /* !CONFIG_NFS_V4_1 */
831 834
@@ -1712,8 +1715,8 @@ static void nfs4_open_confirm_prepare(struct rpc_task *task, void *calldata)
1712{ 1715{
1713 struct nfs4_opendata *data = calldata; 1716 struct nfs4_opendata *data = calldata;
1714 1717
1715 nfs40_setup_sequence(data->o_arg.server, &data->c_arg.seq_args, 1718 nfs40_setup_sequence(data->o_arg.server->nfs_client->cl_slot_tbl,
1716 &data->c_res.seq_res, task); 1719 &data->c_arg.seq_args, &data->c_res.seq_res, task);
1717} 1720}
1718 1721
1719static void nfs4_open_confirm_done(struct rpc_task *task, void *calldata) 1722static void nfs4_open_confirm_done(struct rpc_task *task, void *calldata)
@@ -5994,8 +5997,8 @@ static void nfs4_release_lockowner_prepare(struct rpc_task *task, void *calldata
5994{ 5997{
5995 struct nfs_release_lockowner_data *data = calldata; 5998 struct nfs_release_lockowner_data *data = calldata;
5996 struct nfs_server *server = data->server; 5999 struct nfs_server *server = data->server;
5997 nfs40_setup_sequence(server, &data->args.seq_args, 6000 nfs40_setup_sequence(server->nfs_client->cl_slot_tbl,
5998 &data->res.seq_res, task); 6001 &data->args.seq_args, &data->res.seq_res, task);
5999 data->args.lock_owner.clientid = server->nfs_client->cl_clientid; 6002 data->args.lock_owner.clientid = server->nfs_client->cl_clientid;
6000 data->timestamp = jiffies; 6003 data->timestamp = jiffies;
6001} 6004}
@@ -7557,6 +7560,7 @@ nfs4_layoutget_prepare(struct rpc_task *task, void *calldata)
7557 return; 7560 return;
7558 if (pnfs_choose_layoutget_stateid(&lgp->args.stateid, 7561 if (pnfs_choose_layoutget_stateid(&lgp->args.stateid,
7559 NFS_I(lgp->args.inode)->layout, 7562 NFS_I(lgp->args.inode)->layout,
7563 &lgp->args.range,
7560 lgp->args.ctx->state)) { 7564 lgp->args.ctx->state)) {
7561 rpc_exit(task, NFS4_OK); 7565 rpc_exit(task, NFS4_OK);
7562 } 7566 }
@@ -7812,6 +7816,9 @@ static void nfs4_layoutreturn_release(void *calldata)
7812 spin_lock(&lo->plh_inode->i_lock); 7816 spin_lock(&lo->plh_inode->i_lock);
7813 if (lrp->res.lrs_present) 7817 if (lrp->res.lrs_present)
7814 pnfs_set_layout_stateid(lo, &lrp->res.stateid, true); 7818 pnfs_set_layout_stateid(lo, &lrp->res.stateid, true);
7819 pnfs_clear_layoutreturn_waitbit(lo);
7820 clear_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE, &lo->plh_flags);
7821 rpc_wake_up(&NFS_SERVER(lo->plh_inode)->roc_rpcwaitq);
7815 lo->plh_block_lgets--; 7822 lo->plh_block_lgets--;
7816 spin_unlock(&lo->plh_inode->i_lock); 7823 spin_unlock(&lo->plh_inode->i_lock);
7817 pnfs_put_layout_hdr(lrp->args.layout); 7824 pnfs_put_layout_hdr(lrp->args.layout);
@@ -7825,7 +7832,7 @@ static const struct rpc_call_ops nfs4_layoutreturn_call_ops = {
7825 .rpc_release = nfs4_layoutreturn_release, 7832 .rpc_release = nfs4_layoutreturn_release,
7826}; 7833};
7827 7834
7828int nfs4_proc_layoutreturn(struct nfs4_layoutreturn *lrp) 7835int nfs4_proc_layoutreturn(struct nfs4_layoutreturn *lrp, bool sync)
7829{ 7836{
7830 struct rpc_task *task; 7837 struct rpc_task *task;
7831 struct rpc_message msg = { 7838 struct rpc_message msg = {
@@ -7839,16 +7846,23 @@ int nfs4_proc_layoutreturn(struct nfs4_layoutreturn *lrp)
7839 .rpc_message = &msg, 7846 .rpc_message = &msg,
7840 .callback_ops = &nfs4_layoutreturn_call_ops, 7847 .callback_ops = &nfs4_layoutreturn_call_ops,
7841 .callback_data = lrp, 7848 .callback_data = lrp,
7849 .flags = RPC_TASK_ASYNC,
7842 }; 7850 };
7843 int status; 7851 int status = 0;
7844 7852
7845 dprintk("--> %s\n", __func__); 7853 dprintk("--> %s\n", __func__);
7846 nfs4_init_sequence(&lrp->args.seq_args, &lrp->res.seq_res, 1); 7854 nfs4_init_sequence(&lrp->args.seq_args, &lrp->res.seq_res, 1);
7847 task = rpc_run_task(&task_setup_data); 7855 task = rpc_run_task(&task_setup_data);
7848 if (IS_ERR(task)) 7856 if (IS_ERR(task))
7849 return PTR_ERR(task); 7857 return PTR_ERR(task);
7858 if (sync == false)
7859 goto out;
7860 status = nfs4_wait_for_completion_rpc_task(task);
7861 if (status != 0)
7862 goto out;
7850 status = task->tk_status; 7863 status = task->tk_status;
7851 trace_nfs4_layoutreturn(lrp->args.inode, status); 7864 trace_nfs4_layoutreturn(lrp->args.inode, status);
7865out:
7852 dprintk("<-- %s status=%d\n", __func__, status); 7866 dprintk("<-- %s status=%d\n", __func__, status);
7853 rpc_put_task(task); 7867 rpc_put_task(task);
7854 return status; 7868 return status;
diff --git a/fs/nfs/nfs4super.c b/fs/nfs/nfs4super.c
index 6f340f02f2ba..48cea3c30e5d 100644
--- a/fs/nfs/nfs4super.c
+++ b/fs/nfs/nfs4super.c
@@ -346,6 +346,9 @@ out:
346 346
347static void __exit exit_nfs_v4(void) 347static void __exit exit_nfs_v4(void)
348{ 348{
349 /* Not called in the _init(), conditionally loaded */
350 nfs4_pnfs_v3_ds_connect_unload();
351
349 unregister_nfs_version(&nfs_v4); 352 unregister_nfs_version(&nfs_v4);
350 nfs4_unregister_sysctl(); 353 nfs4_unregister_sysctl();
351 nfs_idmap_quit(); 354 nfs_idmap_quit();
diff --git a/fs/nfs/nfs4xdr.c b/fs/nfs/nfs4xdr.c
index 41253393171f..a2329d69502b 100644
--- a/fs/nfs/nfs4xdr.c
+++ b/fs/nfs/nfs4xdr.c
@@ -2011,11 +2011,11 @@ encode_layoutreturn(struct xdr_stream *xdr,
2011 p = reserve_space(xdr, 16); 2011 p = reserve_space(xdr, 16);
2012 *p++ = cpu_to_be32(0); /* reclaim. always 0 for now */ 2012 *p++ = cpu_to_be32(0); /* reclaim. always 0 for now */
2013 *p++ = cpu_to_be32(args->layout_type); 2013 *p++ = cpu_to_be32(args->layout_type);
2014 *p++ = cpu_to_be32(IOMODE_ANY); 2014 *p++ = cpu_to_be32(args->range.iomode);
2015 *p = cpu_to_be32(RETURN_FILE); 2015 *p = cpu_to_be32(RETURN_FILE);
2016 p = reserve_space(xdr, 16); 2016 p = reserve_space(xdr, 16);
2017 p = xdr_encode_hyper(p, 0); 2017 p = xdr_encode_hyper(p, args->range.offset);
2018 p = xdr_encode_hyper(p, NFS4_MAX_UINT64); 2018 p = xdr_encode_hyper(p, args->range.length);
2019 spin_lock(&args->inode->i_lock); 2019 spin_lock(&args->inode->i_lock);
2020 encode_nfs4_stateid(xdr, &args->stateid); 2020 encode_nfs4_stateid(xdr, &args->stateid);
2021 spin_unlock(&args->inode->i_lock); 2021 spin_unlock(&args->inode->i_lock);
@@ -6566,6 +6566,7 @@ static int nfs4_xdr_dec_read(struct rpc_rqst *rqstp, struct xdr_stream *xdr,
6566 int status; 6566 int status;
6567 6567
6568 status = decode_compound_hdr(xdr, &hdr); 6568 status = decode_compound_hdr(xdr, &hdr);
6569 res->op_status = hdr.status;
6569 if (status) 6570 if (status)
6570 goto out; 6571 goto out;
6571 status = decode_sequence(xdr, &res->seq_res, rqstp); 6572 status = decode_sequence(xdr, &res->seq_res, rqstp);
@@ -6591,6 +6592,7 @@ static int nfs4_xdr_dec_write(struct rpc_rqst *rqstp, struct xdr_stream *xdr,
6591 int status; 6592 int status;
6592 6593
6593 status = decode_compound_hdr(xdr, &hdr); 6594 status = decode_compound_hdr(xdr, &hdr);
6595 res->op_status = hdr.status;
6594 if (status) 6596 if (status)
6595 goto out; 6597 goto out;
6596 status = decode_sequence(xdr, &res->seq_res, rqstp); 6598 status = decode_sequence(xdr, &res->seq_res, rqstp);
@@ -6620,6 +6622,7 @@ static int nfs4_xdr_dec_commit(struct rpc_rqst *rqstp, struct xdr_stream *xdr,
6620 int status; 6622 int status;
6621 6623
6622 status = decode_compound_hdr(xdr, &hdr); 6624 status = decode_compound_hdr(xdr, &hdr);
6625 res->op_status = hdr.status;
6623 if (status) 6626 if (status)
6624 goto out; 6627 goto out;
6625 status = decode_sequence(xdr, &res->seq_res, rqstp); 6628 status = decode_sequence(xdr, &res->seq_res, rqstp);
diff --git a/fs/nfs/objlayout/objio_osd.c b/fs/nfs/objlayout/objio_osd.c
index 9e5bc42180e4..24e1d7403c0b 100644
--- a/fs/nfs/objlayout/objio_osd.c
+++ b/fs/nfs/objlayout/objio_osd.c
@@ -537,11 +537,12 @@ int objio_write_pagelist(struct nfs_pgio_header *hdr, int how)
537static size_t objio_pg_test(struct nfs_pageio_descriptor *pgio, 537static size_t objio_pg_test(struct nfs_pageio_descriptor *pgio,
538 struct nfs_page *prev, struct nfs_page *req) 538 struct nfs_page *prev, struct nfs_page *req)
539{ 539{
540 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(pgio);
540 unsigned int size; 541 unsigned int size;
541 542
542 size = pnfs_generic_pg_test(pgio, prev, req); 543 size = pnfs_generic_pg_test(pgio, prev, req);
543 544
544 if (!size || pgio->pg_count + req->wb_bytes > 545 if (!size || mirror->pg_count + req->wb_bytes >
545 (unsigned long)pgio->pg_layout_private) 546 (unsigned long)pgio->pg_layout_private)
546 return 0; 547 return 0;
547 548
@@ -607,12 +608,14 @@ static const struct nfs_pageio_ops objio_pg_read_ops = {
607 .pg_init = objio_init_read, 608 .pg_init = objio_init_read,
608 .pg_test = objio_pg_test, 609 .pg_test = objio_pg_test,
609 .pg_doio = pnfs_generic_pg_readpages, 610 .pg_doio = pnfs_generic_pg_readpages,
611 .pg_cleanup = pnfs_generic_pg_cleanup,
610}; 612};
611 613
612static const struct nfs_pageio_ops objio_pg_write_ops = { 614static const struct nfs_pageio_ops objio_pg_write_ops = {
613 .pg_init = objio_init_write, 615 .pg_init = objio_init_write,
614 .pg_test = objio_pg_test, 616 .pg_test = objio_pg_test,
615 .pg_doio = pnfs_generic_pg_writepages, 617 .pg_doio = pnfs_generic_pg_writepages,
618 .pg_cleanup = pnfs_generic_pg_cleanup,
616}; 619};
617 620
618static struct pnfs_layoutdriver_type objlayout_type = { 621static struct pnfs_layoutdriver_type objlayout_type = {
diff --git a/fs/nfs/pagelist.c b/fs/nfs/pagelist.c
index 2b5e769beb16..960c99f75d3f 100644
--- a/fs/nfs/pagelist.c
+++ b/fs/nfs/pagelist.c
@@ -42,21 +42,35 @@ static bool nfs_pgarray_set(struct nfs_page_array *p, unsigned int pagecount)
42 return p->pagevec != NULL; 42 return p->pagevec != NULL;
43} 43}
44 44
45struct nfs_pgio_mirror *
46nfs_pgio_current_mirror(struct nfs_pageio_descriptor *desc)
47{
48 return nfs_pgio_has_mirroring(desc) ?
49 &desc->pg_mirrors[desc->pg_mirror_idx] :
50 &desc->pg_mirrors[0];
51}
52EXPORT_SYMBOL_GPL(nfs_pgio_current_mirror);
53
45void nfs_pgheader_init(struct nfs_pageio_descriptor *desc, 54void nfs_pgheader_init(struct nfs_pageio_descriptor *desc,
46 struct nfs_pgio_header *hdr, 55 struct nfs_pgio_header *hdr,
47 void (*release)(struct nfs_pgio_header *hdr)) 56 void (*release)(struct nfs_pgio_header *hdr))
48{ 57{
49 hdr->req = nfs_list_entry(desc->pg_list.next); 58 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
59
60
61 hdr->req = nfs_list_entry(mirror->pg_list.next);
50 hdr->inode = desc->pg_inode; 62 hdr->inode = desc->pg_inode;
51 hdr->cred = hdr->req->wb_context->cred; 63 hdr->cred = hdr->req->wb_context->cred;
52 hdr->io_start = req_offset(hdr->req); 64 hdr->io_start = req_offset(hdr->req);
53 hdr->good_bytes = desc->pg_count; 65 hdr->good_bytes = mirror->pg_count;
54 hdr->dreq = desc->pg_dreq; 66 hdr->dreq = desc->pg_dreq;
55 hdr->layout_private = desc->pg_layout_private; 67 hdr->layout_private = desc->pg_layout_private;
56 hdr->release = release; 68 hdr->release = release;
57 hdr->completion_ops = desc->pg_completion_ops; 69 hdr->completion_ops = desc->pg_completion_ops;
58 if (hdr->completion_ops->init_hdr) 70 if (hdr->completion_ops->init_hdr)
59 hdr->completion_ops->init_hdr(hdr); 71 hdr->completion_ops->init_hdr(hdr);
72
73 hdr->pgio_mirror_idx = desc->pg_mirror_idx;
60} 74}
61EXPORT_SYMBOL_GPL(nfs_pgheader_init); 75EXPORT_SYMBOL_GPL(nfs_pgheader_init);
62 76
@@ -480,7 +494,10 @@ nfs_wait_on_request(struct nfs_page *req)
480size_t nfs_generic_pg_test(struct nfs_pageio_descriptor *desc, 494size_t nfs_generic_pg_test(struct nfs_pageio_descriptor *desc,
481 struct nfs_page *prev, struct nfs_page *req) 495 struct nfs_page *prev, struct nfs_page *req)
482{ 496{
483 if (desc->pg_count > desc->pg_bsize) { 497 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
498
499
500 if (mirror->pg_count > mirror->pg_bsize) {
484 /* should never happen */ 501 /* should never happen */
485 WARN_ON_ONCE(1); 502 WARN_ON_ONCE(1);
486 return 0; 503 return 0;
@@ -490,11 +507,11 @@ size_t nfs_generic_pg_test(struct nfs_pageio_descriptor *desc,
490 * Limit the request size so that we can still allocate a page array 507 * Limit the request size so that we can still allocate a page array
491 * for it without upsetting the slab allocator. 508 * for it without upsetting the slab allocator.
492 */ 509 */
493 if (((desc->pg_count + req->wb_bytes) >> PAGE_SHIFT) * 510 if (((mirror->pg_count + req->wb_bytes) >> PAGE_SHIFT) *
494 sizeof(struct page) > PAGE_SIZE) 511 sizeof(struct page) > PAGE_SIZE)
495 return 0; 512 return 0;
496 513
497 return min(desc->pg_bsize - desc->pg_count, (size_t)req->wb_bytes); 514 return min(mirror->pg_bsize - mirror->pg_count, (size_t)req->wb_bytes);
498} 515}
499EXPORT_SYMBOL_GPL(nfs_generic_pg_test); 516EXPORT_SYMBOL_GPL(nfs_generic_pg_test);
500 517
@@ -597,13 +614,14 @@ static void nfs_pgio_prepare(struct rpc_task *task, void *calldata)
597} 614}
598 615
599int nfs_initiate_pgio(struct rpc_clnt *clnt, struct nfs_pgio_header *hdr, 616int nfs_initiate_pgio(struct rpc_clnt *clnt, struct nfs_pgio_header *hdr,
617 struct rpc_cred *cred, const struct nfs_rpc_ops *rpc_ops,
600 const struct rpc_call_ops *call_ops, int how, int flags) 618 const struct rpc_call_ops *call_ops, int how, int flags)
601{ 619{
602 struct rpc_task *task; 620 struct rpc_task *task;
603 struct rpc_message msg = { 621 struct rpc_message msg = {
604 .rpc_argp = &hdr->args, 622 .rpc_argp = &hdr->args,
605 .rpc_resp = &hdr->res, 623 .rpc_resp = &hdr->res,
606 .rpc_cred = hdr->cred, 624 .rpc_cred = cred,
607 }; 625 };
608 struct rpc_task_setup task_setup_data = { 626 struct rpc_task_setup task_setup_data = {
609 .rpc_client = clnt, 627 .rpc_client = clnt,
@@ -616,7 +634,7 @@ int nfs_initiate_pgio(struct rpc_clnt *clnt, struct nfs_pgio_header *hdr,
616 }; 634 };
617 int ret = 0; 635 int ret = 0;
618 636
619 hdr->rw_ops->rw_initiate(hdr, &msg, &task_setup_data, how); 637 hdr->rw_ops->rw_initiate(hdr, &msg, rpc_ops, &task_setup_data, how);
620 638
621 dprintk("NFS: %5u initiated pgio call " 639 dprintk("NFS: %5u initiated pgio call "
622 "(req %s/%llu, %u bytes @ offset %llu)\n", 640 "(req %s/%llu, %u bytes @ offset %llu)\n",
@@ -650,10 +668,18 @@ EXPORT_SYMBOL_GPL(nfs_initiate_pgio);
650static int nfs_pgio_error(struct nfs_pageio_descriptor *desc, 668static int nfs_pgio_error(struct nfs_pageio_descriptor *desc,
651 struct nfs_pgio_header *hdr) 669 struct nfs_pgio_header *hdr)
652{ 670{
671 struct nfs_pgio_mirror *mirror;
672 u32 midx;
673
653 set_bit(NFS_IOHDR_REDO, &hdr->flags); 674 set_bit(NFS_IOHDR_REDO, &hdr->flags);
654 nfs_pgio_data_destroy(hdr); 675 nfs_pgio_data_destroy(hdr);
655 hdr->completion_ops->completion(hdr); 676 hdr->completion_ops->completion(hdr);
656 desc->pg_completion_ops->error_cleanup(&desc->pg_list); 677 /* TODO: Make sure it's right to clean up all mirrors here
678 * and not just hdr->pgio_mirror_idx */
679 for (midx = 0; midx < desc->pg_mirror_count; midx++) {
680 mirror = &desc->pg_mirrors[midx];
681 desc->pg_completion_ops->error_cleanup(&mirror->pg_list);
682 }
657 return -ENOMEM; 683 return -ENOMEM;
658} 684}
659 685
@@ -670,6 +696,17 @@ static void nfs_pgio_release(void *calldata)
670 hdr->completion_ops->completion(hdr); 696 hdr->completion_ops->completion(hdr);
671} 697}
672 698
699static void nfs_pageio_mirror_init(struct nfs_pgio_mirror *mirror,
700 unsigned int bsize)
701{
702 INIT_LIST_HEAD(&mirror->pg_list);
703 mirror->pg_bytes_written = 0;
704 mirror->pg_count = 0;
705 mirror->pg_bsize = bsize;
706 mirror->pg_base = 0;
707 mirror->pg_recoalesce = 0;
708}
709
673/** 710/**
674 * nfs_pageio_init - initialise a page io descriptor 711 * nfs_pageio_init - initialise a page io descriptor
675 * @desc: pointer to descriptor 712 * @desc: pointer to descriptor
@@ -686,13 +723,10 @@ void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
686 size_t bsize, 723 size_t bsize,
687 int io_flags) 724 int io_flags)
688{ 725{
689 INIT_LIST_HEAD(&desc->pg_list); 726 struct nfs_pgio_mirror *new;
690 desc->pg_bytes_written = 0; 727 int i;
691 desc->pg_count = 0; 728
692 desc->pg_bsize = bsize;
693 desc->pg_base = 0;
694 desc->pg_moreio = 0; 729 desc->pg_moreio = 0;
695 desc->pg_recoalesce = 0;
696 desc->pg_inode = inode; 730 desc->pg_inode = inode;
697 desc->pg_ops = pg_ops; 731 desc->pg_ops = pg_ops;
698 desc->pg_completion_ops = compl_ops; 732 desc->pg_completion_ops = compl_ops;
@@ -702,6 +736,26 @@ void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
702 desc->pg_lseg = NULL; 736 desc->pg_lseg = NULL;
703 desc->pg_dreq = NULL; 737 desc->pg_dreq = NULL;
704 desc->pg_layout_private = NULL; 738 desc->pg_layout_private = NULL;
739 desc->pg_bsize = bsize;
740
741 desc->pg_mirror_count = 1;
742 desc->pg_mirror_idx = 0;
743
744 if (pg_ops->pg_get_mirror_count) {
745 /* until we have a request, we don't have an lseg and no
746 * idea how many mirrors there will be */
747 new = kcalloc(NFS_PAGEIO_DESCRIPTOR_MIRROR_MAX,
748 sizeof(struct nfs_pgio_mirror), GFP_KERNEL);
749 desc->pg_mirrors_dynamic = new;
750 desc->pg_mirrors = new;
751
752 for (i = 0; i < NFS_PAGEIO_DESCRIPTOR_MIRROR_MAX; i++)
753 nfs_pageio_mirror_init(&desc->pg_mirrors[i], bsize);
754 } else {
755 desc->pg_mirrors_dynamic = NULL;
756 desc->pg_mirrors = desc->pg_mirrors_static;
757 nfs_pageio_mirror_init(&desc->pg_mirrors[0], bsize);
758 }
705} 759}
706EXPORT_SYMBOL_GPL(nfs_pageio_init); 760EXPORT_SYMBOL_GPL(nfs_pageio_init);
707 761
@@ -737,14 +791,16 @@ static void nfs_pgio_result(struct rpc_task *task, void *calldata)
737int nfs_generic_pgio(struct nfs_pageio_descriptor *desc, 791int nfs_generic_pgio(struct nfs_pageio_descriptor *desc,
738 struct nfs_pgio_header *hdr) 792 struct nfs_pgio_header *hdr)
739{ 793{
794 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
795
740 struct nfs_page *req; 796 struct nfs_page *req;
741 struct page **pages, 797 struct page **pages,
742 *last_page; 798 *last_page;
743 struct list_head *head = &desc->pg_list; 799 struct list_head *head = &mirror->pg_list;
744 struct nfs_commit_info cinfo; 800 struct nfs_commit_info cinfo;
745 unsigned int pagecount, pageused; 801 unsigned int pagecount, pageused;
746 802
747 pagecount = nfs_page_array_len(desc->pg_base, desc->pg_count); 803 pagecount = nfs_page_array_len(mirror->pg_base, mirror->pg_count);
748 if (!nfs_pgarray_set(&hdr->page_array, pagecount)) 804 if (!nfs_pgarray_set(&hdr->page_array, pagecount))
749 return nfs_pgio_error(desc, hdr); 805 return nfs_pgio_error(desc, hdr);
750 806
@@ -772,7 +828,7 @@ int nfs_generic_pgio(struct nfs_pageio_descriptor *desc,
772 desc->pg_ioflags &= ~FLUSH_COND_STABLE; 828 desc->pg_ioflags &= ~FLUSH_COND_STABLE;
773 829
774 /* Set up the argument struct */ 830 /* Set up the argument struct */
775 nfs_pgio_rpcsetup(hdr, desc->pg_count, 0, desc->pg_ioflags, &cinfo); 831 nfs_pgio_rpcsetup(hdr, mirror->pg_count, 0, desc->pg_ioflags, &cinfo);
776 desc->pg_rpc_callops = &nfs_pgio_common_ops; 832 desc->pg_rpc_callops = &nfs_pgio_common_ops;
777 return 0; 833 return 0;
778} 834}
@@ -780,23 +836,74 @@ EXPORT_SYMBOL_GPL(nfs_generic_pgio);
780 836
781static int nfs_generic_pg_pgios(struct nfs_pageio_descriptor *desc) 837static int nfs_generic_pg_pgios(struct nfs_pageio_descriptor *desc)
782{ 838{
839 struct nfs_pgio_mirror *mirror;
783 struct nfs_pgio_header *hdr; 840 struct nfs_pgio_header *hdr;
784 int ret; 841 int ret;
785 842
843 mirror = nfs_pgio_current_mirror(desc);
844
786 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops); 845 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
787 if (!hdr) { 846 if (!hdr) {
788 desc->pg_completion_ops->error_cleanup(&desc->pg_list); 847 /* TODO: make sure this is right with mirroring - or
848 * should it back out all mirrors? */
849 desc->pg_completion_ops->error_cleanup(&mirror->pg_list);
789 return -ENOMEM; 850 return -ENOMEM;
790 } 851 }
791 nfs_pgheader_init(desc, hdr, nfs_pgio_header_free); 852 nfs_pgheader_init(desc, hdr, nfs_pgio_header_free);
792 ret = nfs_generic_pgio(desc, hdr); 853 ret = nfs_generic_pgio(desc, hdr);
793 if (ret == 0) 854 if (ret == 0)
794 ret = nfs_initiate_pgio(NFS_CLIENT(hdr->inode), 855 ret = nfs_initiate_pgio(NFS_CLIENT(hdr->inode),
795 hdr, desc->pg_rpc_callops, 856 hdr,
857 hdr->cred,
858 NFS_PROTO(hdr->inode),
859 desc->pg_rpc_callops,
796 desc->pg_ioflags, 0); 860 desc->pg_ioflags, 0);
797 return ret; 861 return ret;
798} 862}
799 863
864/*
865 * nfs_pageio_setup_mirroring - determine if mirroring is to be used
866 * by calling the pg_get_mirror_count op
867 */
868static int nfs_pageio_setup_mirroring(struct nfs_pageio_descriptor *pgio,
869 struct nfs_page *req)
870{
871 int mirror_count = 1;
872
873 if (!pgio->pg_ops->pg_get_mirror_count)
874 return 0;
875
876 mirror_count = pgio->pg_ops->pg_get_mirror_count(pgio, req);
877
878 if (!mirror_count || mirror_count > NFS_PAGEIO_DESCRIPTOR_MIRROR_MAX)
879 return -EINVAL;
880
881 if (WARN_ON_ONCE(!pgio->pg_mirrors_dynamic))
882 return -EINVAL;
883
884 pgio->pg_mirror_count = mirror_count;
885
886 return 0;
887}
888
889/*
890 * nfs_pageio_stop_mirroring - stop using mirroring (set mirror count to 1)
891 */
892void nfs_pageio_stop_mirroring(struct nfs_pageio_descriptor *pgio)
893{
894 pgio->pg_mirror_count = 1;
895 pgio->pg_mirror_idx = 0;
896}
897
898static void nfs_pageio_cleanup_mirroring(struct nfs_pageio_descriptor *pgio)
899{
900 pgio->pg_mirror_count = 1;
901 pgio->pg_mirror_idx = 0;
902 pgio->pg_mirrors = pgio->pg_mirrors_static;
903 kfree(pgio->pg_mirrors_dynamic);
904 pgio->pg_mirrors_dynamic = NULL;
905}
906
800static bool nfs_match_open_context(const struct nfs_open_context *ctx1, 907static bool nfs_match_open_context(const struct nfs_open_context *ctx1,
801 const struct nfs_open_context *ctx2) 908 const struct nfs_open_context *ctx2)
802{ 909{
@@ -863,19 +970,22 @@ static bool nfs_can_coalesce_requests(struct nfs_page *prev,
863static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc, 970static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
864 struct nfs_page *req) 971 struct nfs_page *req)
865{ 972{
973 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
974
866 struct nfs_page *prev = NULL; 975 struct nfs_page *prev = NULL;
867 if (desc->pg_count != 0) { 976
868 prev = nfs_list_entry(desc->pg_list.prev); 977 if (mirror->pg_count != 0) {
978 prev = nfs_list_entry(mirror->pg_list.prev);
869 } else { 979 } else {
870 if (desc->pg_ops->pg_init) 980 if (desc->pg_ops->pg_init)
871 desc->pg_ops->pg_init(desc, req); 981 desc->pg_ops->pg_init(desc, req);
872 desc->pg_base = req->wb_pgbase; 982 mirror->pg_base = req->wb_pgbase;
873 } 983 }
874 if (!nfs_can_coalesce_requests(prev, req, desc)) 984 if (!nfs_can_coalesce_requests(prev, req, desc))
875 return 0; 985 return 0;
876 nfs_list_remove_request(req); 986 nfs_list_remove_request(req);
877 nfs_list_add_request(req, &desc->pg_list); 987 nfs_list_add_request(req, &mirror->pg_list);
878 desc->pg_count += req->wb_bytes; 988 mirror->pg_count += req->wb_bytes;
879 return 1; 989 return 1;
880} 990}
881 991
@@ -884,16 +994,19 @@ static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
884 */ 994 */
885static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc) 995static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
886{ 996{
887 if (!list_empty(&desc->pg_list)) { 997 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
998
999
1000 if (!list_empty(&mirror->pg_list)) {
888 int error = desc->pg_ops->pg_doio(desc); 1001 int error = desc->pg_ops->pg_doio(desc);
889 if (error < 0) 1002 if (error < 0)
890 desc->pg_error = error; 1003 desc->pg_error = error;
891 else 1004 else
892 desc->pg_bytes_written += desc->pg_count; 1005 mirror->pg_bytes_written += mirror->pg_count;
893 } 1006 }
894 if (list_empty(&desc->pg_list)) { 1007 if (list_empty(&mirror->pg_list)) {
895 desc->pg_count = 0; 1008 mirror->pg_count = 0;
896 desc->pg_base = 0; 1009 mirror->pg_base = 0;
897 } 1010 }
898} 1011}
899 1012
@@ -911,6 +1024,8 @@ static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
911static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc, 1024static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
912 struct nfs_page *req) 1025 struct nfs_page *req)
913{ 1026{
1027 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1028
914 struct nfs_page *subreq; 1029 struct nfs_page *subreq;
915 unsigned int bytes_left = 0; 1030 unsigned int bytes_left = 0;
916 unsigned int offset, pgbase; 1031 unsigned int offset, pgbase;
@@ -934,7 +1049,7 @@ static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
934 nfs_pageio_doio(desc); 1049 nfs_pageio_doio(desc);
935 if (desc->pg_error < 0) 1050 if (desc->pg_error < 0)
936 return 0; 1051 return 0;
937 if (desc->pg_recoalesce) 1052 if (mirror->pg_recoalesce)
938 return 0; 1053 return 0;
939 /* retry add_request for this subreq */ 1054 /* retry add_request for this subreq */
940 nfs_page_group_lock(req, false); 1055 nfs_page_group_lock(req, false);
@@ -972,14 +1087,16 @@ err_ptr:
972 1087
973static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc) 1088static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc)
974{ 1089{
1090 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
975 LIST_HEAD(head); 1091 LIST_HEAD(head);
976 1092
977 do { 1093 do {
978 list_splice_init(&desc->pg_list, &head); 1094 list_splice_init(&mirror->pg_list, &head);
979 desc->pg_bytes_written -= desc->pg_count; 1095 mirror->pg_bytes_written -= mirror->pg_count;
980 desc->pg_count = 0; 1096 mirror->pg_count = 0;
981 desc->pg_base = 0; 1097 mirror->pg_base = 0;
982 desc->pg_recoalesce = 0; 1098 mirror->pg_recoalesce = 0;
1099
983 desc->pg_moreio = 0; 1100 desc->pg_moreio = 0;
984 1101
985 while (!list_empty(&head)) { 1102 while (!list_empty(&head)) {
@@ -993,11 +1110,11 @@ static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc)
993 return 0; 1110 return 0;
994 break; 1111 break;
995 } 1112 }
996 } while (desc->pg_recoalesce); 1113 } while (mirror->pg_recoalesce);
997 return 1; 1114 return 1;
998} 1115}
999 1116
1000int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc, 1117static int nfs_pageio_add_request_mirror(struct nfs_pageio_descriptor *desc,
1001 struct nfs_page *req) 1118 struct nfs_page *req)
1002{ 1119{
1003 int ret; 1120 int ret;
@@ -1010,9 +1127,80 @@ int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
1010 break; 1127 break;
1011 ret = nfs_do_recoalesce(desc); 1128 ret = nfs_do_recoalesce(desc);
1012 } while (ret); 1129 } while (ret);
1130
1013 return ret; 1131 return ret;
1014} 1132}
1015 1133
1134int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
1135 struct nfs_page *req)
1136{
1137 u32 midx;
1138 unsigned int pgbase, offset, bytes;
1139 struct nfs_page *dupreq, *lastreq;
1140
1141 pgbase = req->wb_pgbase;
1142 offset = req->wb_offset;
1143 bytes = req->wb_bytes;
1144
1145 nfs_pageio_setup_mirroring(desc, req);
1146
1147 for (midx = 0; midx < desc->pg_mirror_count; midx++) {
1148 if (midx) {
1149 nfs_page_group_lock(req, false);
1150
1151 /* find the last request */
1152 for (lastreq = req->wb_head;
1153 lastreq->wb_this_page != req->wb_head;
1154 lastreq = lastreq->wb_this_page)
1155 ;
1156
1157 dupreq = nfs_create_request(req->wb_context,
1158 req->wb_page, lastreq, pgbase, bytes);
1159
1160 if (IS_ERR(dupreq)) {
1161 nfs_page_group_unlock(req);
1162 return 0;
1163 }
1164
1165 nfs_lock_request(dupreq);
1166 nfs_page_group_unlock(req);
1167 dupreq->wb_offset = offset;
1168 dupreq->wb_index = req->wb_index;
1169 } else
1170 dupreq = req;
1171
1172 if (nfs_pgio_has_mirroring(desc))
1173 desc->pg_mirror_idx = midx;
1174 if (!nfs_pageio_add_request_mirror(desc, dupreq))
1175 return 0;
1176 }
1177
1178 return 1;
1179}
1180
1181/*
1182 * nfs_pageio_complete_mirror - Complete I/O on the current mirror of an
1183 * nfs_pageio_descriptor
1184 * @desc: pointer to io descriptor
1185 */
1186static void nfs_pageio_complete_mirror(struct nfs_pageio_descriptor *desc,
1187 u32 mirror_idx)
1188{
1189 struct nfs_pgio_mirror *mirror = &desc->pg_mirrors[mirror_idx];
1190 u32 restore_idx = desc->pg_mirror_idx;
1191
1192 if (nfs_pgio_has_mirroring(desc))
1193 desc->pg_mirror_idx = mirror_idx;
1194 for (;;) {
1195 nfs_pageio_doio(desc);
1196 if (!mirror->pg_recoalesce)
1197 break;
1198 if (!nfs_do_recoalesce(desc))
1199 break;
1200 }
1201 desc->pg_mirror_idx = restore_idx;
1202}
1203
1016/* 1204/*
1017 * nfs_pageio_resend - Transfer requests to new descriptor and resend 1205 * nfs_pageio_resend - Transfer requests to new descriptor and resend
1018 * @hdr - the pgio header to move request from 1206 * @hdr - the pgio header to move request from
@@ -1046,18 +1234,19 @@ int nfs_pageio_resend(struct nfs_pageio_descriptor *desc,
1046EXPORT_SYMBOL_GPL(nfs_pageio_resend); 1234EXPORT_SYMBOL_GPL(nfs_pageio_resend);
1047 1235
1048/** 1236/**
1049 * nfs_pageio_complete - Complete I/O on an nfs_pageio_descriptor 1237 * nfs_pageio_complete - Complete I/O then cleanup an nfs_pageio_descriptor
1050 * @desc: pointer to io descriptor 1238 * @desc: pointer to io descriptor
1051 */ 1239 */
1052void nfs_pageio_complete(struct nfs_pageio_descriptor *desc) 1240void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
1053{ 1241{
1054 for (;;) { 1242 u32 midx;
1055 nfs_pageio_doio(desc); 1243
1056 if (!desc->pg_recoalesce) 1244 for (midx = 0; midx < desc->pg_mirror_count; midx++)
1057 break; 1245 nfs_pageio_complete_mirror(desc, midx);
1058 if (!nfs_do_recoalesce(desc)) 1246
1059 break; 1247 if (desc->pg_ops->pg_cleanup)
1060 } 1248 desc->pg_ops->pg_cleanup(desc);
1249 nfs_pageio_cleanup_mirroring(desc);
1061} 1250}
1062 1251
1063/** 1252/**
@@ -1073,10 +1262,17 @@ void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
1073 */ 1262 */
1074void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index) 1263void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index)
1075{ 1264{
1076 if (!list_empty(&desc->pg_list)) { 1265 struct nfs_pgio_mirror *mirror;
1077 struct nfs_page *prev = nfs_list_entry(desc->pg_list.prev); 1266 struct nfs_page *prev;
1078 if (index != prev->wb_index + 1) 1267 u32 midx;
1079 nfs_pageio_complete(desc); 1268
1269 for (midx = 0; midx < desc->pg_mirror_count; midx++) {
1270 mirror = &desc->pg_mirrors[midx];
1271 if (!list_empty(&mirror->pg_list)) {
1272 prev = nfs_list_entry(mirror->pg_list.prev);
1273 if (index != prev->wb_index + 1)
1274 nfs_pageio_complete_mirror(desc, midx);
1275 }
1080 } 1276 }
1081} 1277}
1082 1278
diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c
index 4d69076a6028..703501d3ed19 100644
--- a/fs/nfs/pnfs.c
+++ b/fs/nfs/pnfs.c
@@ -51,6 +51,10 @@ static DEFINE_SPINLOCK(pnfs_spinlock);
51 */ 51 */
52static LIST_HEAD(pnfs_modules_tbl); 52static LIST_HEAD(pnfs_modules_tbl);
53 53
54static int
55pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, nfs4_stateid stateid,
56 enum pnfs_iomode iomode, bool sync);
57
54/* Return the registered pnfs layout driver module matching given id */ 58/* Return the registered pnfs layout driver module matching given id */
55static struct pnfs_layoutdriver_type * 59static struct pnfs_layoutdriver_type *
56find_pnfs_driver_locked(u32 id) 60find_pnfs_driver_locked(u32 id)
@@ -239,6 +243,8 @@ pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo)
239 struct inode *inode = lo->plh_inode; 243 struct inode *inode = lo->plh_inode;
240 244
241 if (atomic_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) { 245 if (atomic_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) {
246 if (!list_empty(&lo->plh_segs))
247 WARN_ONCE(1, "NFS: BUG unfreed layout segments.\n");
242 pnfs_detach_layout_hdr(lo); 248 pnfs_detach_layout_hdr(lo);
243 spin_unlock(&inode->i_lock); 249 spin_unlock(&inode->i_lock);
244 pnfs_free_layout_hdr(lo); 250 pnfs_free_layout_hdr(lo);
@@ -338,6 +344,65 @@ pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo,
338 rpc_wake_up(&NFS_SERVER(inode)->roc_rpcwaitq); 344 rpc_wake_up(&NFS_SERVER(inode)->roc_rpcwaitq);
339} 345}
340 346
347/* Return true if layoutreturn is needed */
348static bool
349pnfs_layout_need_return(struct pnfs_layout_hdr *lo,
350 struct pnfs_layout_segment *lseg)
351{
352 struct pnfs_layout_segment *s;
353
354 if (!test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
355 return false;
356
357 list_for_each_entry(s, &lo->plh_segs, pls_list)
358 if (s != lseg && test_bit(NFS_LSEG_LAYOUTRETURN, &s->pls_flags))
359 return false;
360
361 return true;
362}
363
364static void pnfs_layoutreturn_free_lseg(struct work_struct *work)
365{
366 struct pnfs_layout_segment *lseg;
367 struct pnfs_layout_hdr *lo;
368 struct inode *inode;
369
370 lseg = container_of(work, struct pnfs_layout_segment, pls_work);
371 WARN_ON(atomic_read(&lseg->pls_refcount));
372 lo = lseg->pls_layout;
373 inode = lo->plh_inode;
374
375 spin_lock(&inode->i_lock);
376 if (pnfs_layout_need_return(lo, lseg)) {
377 nfs4_stateid stateid;
378 enum pnfs_iomode iomode;
379
380 stateid = lo->plh_stateid;
381 iomode = lo->plh_return_iomode;
382 /* decreased in pnfs_send_layoutreturn() */
383 lo->plh_block_lgets++;
384 lo->plh_return_iomode = 0;
385 spin_unlock(&inode->i_lock);
386
387 pnfs_send_layoutreturn(lo, stateid, iomode, true);
388 spin_lock(&inode->i_lock);
389 } else
390 /* match pnfs_get_layout_hdr #2 in pnfs_put_lseg */
391 pnfs_put_layout_hdr(lo);
392 pnfs_layout_remove_lseg(lo, lseg);
393 spin_unlock(&inode->i_lock);
394 pnfs_free_lseg(lseg);
395 /* match pnfs_get_layout_hdr #1 in pnfs_put_lseg */
396 pnfs_put_layout_hdr(lo);
397}
398
399static void
400pnfs_layoutreturn_free_lseg_async(struct pnfs_layout_segment *lseg)
401{
402 INIT_WORK(&lseg->pls_work, pnfs_layoutreturn_free_lseg);
403 queue_work(nfsiod_workqueue, &lseg->pls_work);
404}
405
341void 406void
342pnfs_put_lseg(struct pnfs_layout_segment *lseg) 407pnfs_put_lseg(struct pnfs_layout_segment *lseg)
343{ 408{
@@ -354,10 +419,17 @@ pnfs_put_lseg(struct pnfs_layout_segment *lseg)
354 inode = lo->plh_inode; 419 inode = lo->plh_inode;
355 if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) { 420 if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
356 pnfs_get_layout_hdr(lo); 421 pnfs_get_layout_hdr(lo);
357 pnfs_layout_remove_lseg(lo, lseg); 422 if (pnfs_layout_need_return(lo, lseg)) {
358 spin_unlock(&inode->i_lock); 423 spin_unlock(&inode->i_lock);
359 pnfs_free_lseg(lseg); 424 /* hdr reference dropped in nfs4_layoutreturn_release */
360 pnfs_put_layout_hdr(lo); 425 pnfs_get_layout_hdr(lo);
426 pnfs_layoutreturn_free_lseg_async(lseg);
427 } else {
428 pnfs_layout_remove_lseg(lo, lseg);
429 spin_unlock(&inode->i_lock);
430 pnfs_free_lseg(lseg);
431 pnfs_put_layout_hdr(lo);
432 }
361 } 433 }
362} 434}
363EXPORT_SYMBOL_GPL(pnfs_put_lseg); 435EXPORT_SYMBOL_GPL(pnfs_put_lseg);
@@ -544,6 +616,7 @@ pnfs_destroy_layout(struct nfs_inode *nfsi)
544 pnfs_get_layout_hdr(lo); 616 pnfs_get_layout_hdr(lo);
545 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED); 617 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED);
546 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED); 618 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED);
619 pnfs_clear_retry_layoutget(lo);
547 spin_unlock(&nfsi->vfs_inode.i_lock); 620 spin_unlock(&nfsi->vfs_inode.i_lock);
548 pnfs_free_lseg_list(&tmp_list); 621 pnfs_free_lseg_list(&tmp_list);
549 pnfs_put_layout_hdr(lo); 622 pnfs_put_layout_hdr(lo);
@@ -741,25 +814,37 @@ pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo,
741 return !pnfs_seqid_is_newer(seqid, lo->plh_barrier); 814 return !pnfs_seqid_is_newer(seqid, lo->plh_barrier);
742} 815}
743 816
817static bool
818pnfs_layout_returning(const struct pnfs_layout_hdr *lo,
819 struct pnfs_layout_range *range)
820{
821 return test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags) &&
822 (lo->plh_return_iomode == IOMODE_ANY ||
823 lo->plh_return_iomode == range->iomode);
824}
825
744/* lget is set to 1 if called from inside send_layoutget call chain */ 826/* lget is set to 1 if called from inside send_layoutget call chain */
745static bool 827static bool
746pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo, int lget) 828pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo,
829 struct pnfs_layout_range *range, int lget)
747{ 830{
748 return lo->plh_block_lgets || 831 return lo->plh_block_lgets ||
749 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) || 832 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) ||
750 (list_empty(&lo->plh_segs) && 833 (list_empty(&lo->plh_segs) &&
751 (atomic_read(&lo->plh_outstanding) > lget)); 834 (atomic_read(&lo->plh_outstanding) > lget)) ||
835 pnfs_layout_returning(lo, range);
752} 836}
753 837
754int 838int
755pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo, 839pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo,
840 struct pnfs_layout_range *range,
756 struct nfs4_state *open_state) 841 struct nfs4_state *open_state)
757{ 842{
758 int status = 0; 843 int status = 0;
759 844
760 dprintk("--> %s\n", __func__); 845 dprintk("--> %s\n", __func__);
761 spin_lock(&lo->plh_inode->i_lock); 846 spin_lock(&lo->plh_inode->i_lock);
762 if (pnfs_layoutgets_blocked(lo, 1)) { 847 if (pnfs_layoutgets_blocked(lo, range, 1)) {
763 status = -EAGAIN; 848 status = -EAGAIN;
764 } else if (!nfs4_valid_open_stateid(open_state)) { 849 } else if (!nfs4_valid_open_stateid(open_state)) {
765 status = -EBADF; 850 status = -EBADF;
@@ -826,7 +911,9 @@ send_layoutget(struct pnfs_layout_hdr *lo,
826 pnfs_layout_io_set_failed(lo, range->iomode); 911 pnfs_layout_io_set_failed(lo, range->iomode);
827 } 912 }
828 return NULL; 913 return NULL;
829 } 914 } else
915 pnfs_layout_clear_fail_bit(lo,
916 pnfs_iomode_to_fail_bit(range->iomode));
830 917
831 return lseg; 918 return lseg;
832} 919}
@@ -846,6 +933,49 @@ static void pnfs_clear_layoutcommit(struct inode *inode,
846 } 933 }
847} 934}
848 935
936void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo)
937{
938 clear_bit_unlock(NFS_LAYOUT_RETURN, &lo->plh_flags);
939 smp_mb__after_atomic();
940 wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN);
941}
942
943static int
944pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, nfs4_stateid stateid,
945 enum pnfs_iomode iomode, bool sync)
946{
947 struct inode *ino = lo->plh_inode;
948 struct nfs4_layoutreturn *lrp;
949 int status = 0;
950
951 lrp = kzalloc(sizeof(*lrp), GFP_KERNEL);
952 if (unlikely(lrp == NULL)) {
953 status = -ENOMEM;
954 spin_lock(&ino->i_lock);
955 lo->plh_block_lgets--;
956 pnfs_clear_layoutreturn_waitbit(lo);
957 rpc_wake_up(&NFS_SERVER(ino)->roc_rpcwaitq);
958 spin_unlock(&ino->i_lock);
959 pnfs_put_layout_hdr(lo);
960 goto out;
961 }
962
963 lrp->args.stateid = stateid;
964 lrp->args.layout_type = NFS_SERVER(ino)->pnfs_curr_ld->id;
965 lrp->args.inode = ino;
966 lrp->args.range.iomode = iomode;
967 lrp->args.range.offset = 0;
968 lrp->args.range.length = NFS4_MAX_UINT64;
969 lrp->args.layout = lo;
970 lrp->clp = NFS_SERVER(ino)->nfs_client;
971 lrp->cred = lo->plh_lc_cred;
972
973 status = nfs4_proc_layoutreturn(lrp, sync);
974out:
975 dprintk("<-- %s status: %d\n", __func__, status);
976 return status;
977}
978
849/* 979/*
850 * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr 980 * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
851 * when the layout segment list is empty. 981 * when the layout segment list is empty.
@@ -860,7 +990,6 @@ _pnfs_return_layout(struct inode *ino)
860 struct pnfs_layout_hdr *lo = NULL; 990 struct pnfs_layout_hdr *lo = NULL;
861 struct nfs_inode *nfsi = NFS_I(ino); 991 struct nfs_inode *nfsi = NFS_I(ino);
862 LIST_HEAD(tmp_list); 992 LIST_HEAD(tmp_list);
863 struct nfs4_layoutreturn *lrp;
864 nfs4_stateid stateid; 993 nfs4_stateid stateid;
865 int status = 0, empty; 994 int status = 0, empty;
866 995
@@ -902,24 +1031,7 @@ _pnfs_return_layout(struct inode *ino)
902 spin_unlock(&ino->i_lock); 1031 spin_unlock(&ino->i_lock);
903 pnfs_free_lseg_list(&tmp_list); 1032 pnfs_free_lseg_list(&tmp_list);
904 1033
905 lrp = kzalloc(sizeof(*lrp), GFP_KERNEL); 1034 status = pnfs_send_layoutreturn(lo, stateid, IOMODE_ANY, true);
906 if (unlikely(lrp == NULL)) {
907 status = -ENOMEM;
908 spin_lock(&ino->i_lock);
909 lo->plh_block_lgets--;
910 spin_unlock(&ino->i_lock);
911 pnfs_put_layout_hdr(lo);
912 goto out;
913 }
914
915 lrp->args.stateid = stateid;
916 lrp->args.layout_type = NFS_SERVER(ino)->pnfs_curr_ld->id;
917 lrp->args.inode = ino;
918 lrp->args.layout = lo;
919 lrp->clp = NFS_SERVER(ino)->nfs_client;
920 lrp->cred = lo->plh_lc_cred;
921
922 status = nfs4_proc_layoutreturn(lrp);
923out: 1035out:
924 dprintk("<-- %s status: %d\n", __func__, status); 1036 dprintk("<-- %s status: %d\n", __func__, status);
925 return status; 1037 return status;
@@ -960,8 +1072,9 @@ bool pnfs_roc(struct inode *ino)
960 struct nfs4_state *state; 1072 struct nfs4_state *state;
961 struct pnfs_layout_hdr *lo; 1073 struct pnfs_layout_hdr *lo;
962 struct pnfs_layout_segment *lseg, *tmp; 1074 struct pnfs_layout_segment *lseg, *tmp;
1075 nfs4_stateid stateid;
963 LIST_HEAD(tmp_list); 1076 LIST_HEAD(tmp_list);
964 bool found = false; 1077 bool found = false, layoutreturn = false;
965 1078
966 spin_lock(&ino->i_lock); 1079 spin_lock(&ino->i_lock);
967 lo = nfsi->layout; 1080 lo = nfsi->layout;
@@ -980,6 +1093,8 @@ bool pnfs_roc(struct inode *ino)
980 goto out_noroc; 1093 goto out_noroc;
981 } 1094 }
982 1095
1096 goto out_noroc;
1097 pnfs_clear_retry_layoutget(lo);
983 list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list) 1098 list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list)
984 if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) { 1099 if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
985 mark_lseg_invalid(lseg, &tmp_list); 1100 mark_lseg_invalid(lseg, &tmp_list);
@@ -994,7 +1109,19 @@ bool pnfs_roc(struct inode *ino)
994 return true; 1109 return true;
995 1110
996out_noroc: 1111out_noroc:
1112 if (lo) {
1113 stateid = lo->plh_stateid;
1114 layoutreturn =
1115 test_and_clear_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE,
1116 &lo->plh_flags);
1117 if (layoutreturn) {
1118 lo->plh_block_lgets++;
1119 pnfs_get_layout_hdr(lo);
1120 }
1121 }
997 spin_unlock(&ino->i_lock); 1122 spin_unlock(&ino->i_lock);
1123 if (layoutreturn)
1124 pnfs_send_layoutreturn(lo, stateid, IOMODE_ANY, true);
998 return false; 1125 return false;
999} 1126}
1000 1127
@@ -1029,8 +1156,9 @@ bool pnfs_roc_drain(struct inode *ino, u32 *barrier, struct rpc_task *task)
1029 struct nfs_inode *nfsi = NFS_I(ino); 1156 struct nfs_inode *nfsi = NFS_I(ino);
1030 struct pnfs_layout_hdr *lo; 1157 struct pnfs_layout_hdr *lo;
1031 struct pnfs_layout_segment *lseg; 1158 struct pnfs_layout_segment *lseg;
1159 nfs4_stateid stateid;
1032 u32 current_seqid; 1160 u32 current_seqid;
1033 bool found = false; 1161 bool found = false, layoutreturn = false;
1034 1162
1035 spin_lock(&ino->i_lock); 1163 spin_lock(&ino->i_lock);
1036 list_for_each_entry(lseg, &nfsi->layout->plh_segs, pls_list) 1164 list_for_each_entry(lseg, &nfsi->layout->plh_segs, pls_list)
@@ -1047,7 +1175,21 @@ bool pnfs_roc_drain(struct inode *ino, u32 *barrier, struct rpc_task *task)
1047 */ 1175 */
1048 *barrier = current_seqid + atomic_read(&lo->plh_outstanding); 1176 *barrier = current_seqid + atomic_read(&lo->plh_outstanding);
1049out: 1177out:
1178 if (!found) {
1179 stateid = lo->plh_stateid;
1180 layoutreturn =
1181 test_and_clear_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE,
1182 &lo->plh_flags);
1183 if (layoutreturn) {
1184 lo->plh_block_lgets++;
1185 pnfs_get_layout_hdr(lo);
1186 }
1187 }
1050 spin_unlock(&ino->i_lock); 1188 spin_unlock(&ino->i_lock);
1189 if (layoutreturn) {
1190 rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
1191 pnfs_send_layoutreturn(lo, stateid, IOMODE_ANY, false);
1192 }
1051 return found; 1193 return found;
1052} 1194}
1053 1195
@@ -1194,6 +1336,7 @@ pnfs_find_lseg(struct pnfs_layout_hdr *lo,
1194 1336
1195 list_for_each_entry(lseg, &lo->plh_segs, pls_list) { 1337 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
1196 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) && 1338 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
1339 !test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) &&
1197 pnfs_lseg_range_match(&lseg->pls_range, range)) { 1340 pnfs_lseg_range_match(&lseg->pls_range, range)) {
1198 ret = pnfs_get_lseg(lseg); 1341 ret = pnfs_get_lseg(lseg);
1199 break; 1342 break;
@@ -1282,6 +1425,35 @@ static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx,
1282 return ret; 1425 return ret;
1283} 1426}
1284 1427
1428/* stop waiting if someone clears NFS_LAYOUT_RETRY_LAYOUTGET bit. */
1429static int pnfs_layoutget_retry_bit_wait(struct wait_bit_key *key)
1430{
1431 if (!test_bit(NFS_LAYOUT_RETRY_LAYOUTGET, key->flags))
1432 return 1;
1433 return nfs_wait_bit_killable(key);
1434}
1435
1436static bool pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo)
1437{
1438 /*
1439 * send layoutcommit as it can hold up layoutreturn due to lseg
1440 * reference
1441 */
1442 pnfs_layoutcommit_inode(lo->plh_inode, false);
1443 return !wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN,
1444 pnfs_layoutget_retry_bit_wait,
1445 TASK_UNINTERRUPTIBLE);
1446}
1447
1448static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo)
1449{
1450 unsigned long *bitlock = &lo->plh_flags;
1451
1452 clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock);
1453 smp_mb__after_atomic();
1454 wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET);
1455}
1456
1285/* 1457/*
1286 * Layout segment is retreived from the server if not cached. 1458 * Layout segment is retreived from the server if not cached.
1287 * The appropriate layout segment is referenced and returned to the caller. 1459 * The appropriate layout segment is referenced and returned to the caller.
@@ -1312,6 +1484,8 @@ pnfs_update_layout(struct inode *ino,
1312 if (pnfs_within_mdsthreshold(ctx, ino, iomode)) 1484 if (pnfs_within_mdsthreshold(ctx, ino, iomode))
1313 goto out; 1485 goto out;
1314 1486
1487lookup_again:
1488 first = false;
1315 spin_lock(&ino->i_lock); 1489 spin_lock(&ino->i_lock);
1316 lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags); 1490 lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
1317 if (lo == NULL) { 1491 if (lo == NULL) {
@@ -1326,27 +1500,62 @@ pnfs_update_layout(struct inode *ino,
1326 } 1500 }
1327 1501
1328 /* if LAYOUTGET already failed once we don't try again */ 1502 /* if LAYOUTGET already failed once we don't try again */
1329 if (pnfs_layout_io_test_failed(lo, iomode)) 1503 if (pnfs_layout_io_test_failed(lo, iomode) &&
1504 !pnfs_should_retry_layoutget(lo))
1330 goto out_unlock; 1505 goto out_unlock;
1331 1506
1332 /* Check to see if the layout for the given range already exists */ 1507 first = list_empty(&lo->plh_segs);
1333 lseg = pnfs_find_lseg(lo, &arg); 1508 if (first) {
1334 if (lseg) 1509 /* The first layoutget for the file. Need to serialize per
1335 goto out_unlock; 1510 * RFC 5661 Errata 3208.
1511 */
1512 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET,
1513 &lo->plh_flags)) {
1514 spin_unlock(&ino->i_lock);
1515 wait_on_bit(&lo->plh_flags, NFS_LAYOUT_FIRST_LAYOUTGET,
1516 TASK_UNINTERRUPTIBLE);
1517 pnfs_put_layout_hdr(lo);
1518 goto lookup_again;
1519 }
1520 } else {
1521 /* Check to see if the layout for the given range
1522 * already exists
1523 */
1524 lseg = pnfs_find_lseg(lo, &arg);
1525 if (lseg)
1526 goto out_unlock;
1527 }
1336 1528
1337 if (pnfs_layoutgets_blocked(lo, 0)) 1529 /*
1530 * Because we free lsegs before sending LAYOUTRETURN, we need to wait
1531 * for LAYOUTRETURN even if first is true.
1532 */
1533 if (!lseg && pnfs_should_retry_layoutget(lo) &&
1534 test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
1535 spin_unlock(&ino->i_lock);
1536 dprintk("%s wait for layoutreturn\n", __func__);
1537 if (pnfs_prepare_to_retry_layoutget(lo)) {
1538 if (first)
1539 pnfs_clear_first_layoutget(lo);
1540 pnfs_put_layout_hdr(lo);
1541 dprintk("%s retrying\n", __func__);
1542 goto lookup_again;
1543 }
1544 goto out_put_layout_hdr;
1545 }
1546
1547 if (pnfs_layoutgets_blocked(lo, &arg, 0))
1338 goto out_unlock; 1548 goto out_unlock;
1339 atomic_inc(&lo->plh_outstanding); 1549 atomic_inc(&lo->plh_outstanding);
1340
1341 first = list_empty(&lo->plh_layouts) ? true : false;
1342 spin_unlock(&ino->i_lock); 1550 spin_unlock(&ino->i_lock);
1343 1551
1344 if (first) { 1552 if (list_empty(&lo->plh_layouts)) {
1345 /* The lo must be on the clp list if there is any 1553 /* The lo must be on the clp list if there is any
1346 * chance of a CB_LAYOUTRECALL(FILE) coming in. 1554 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1347 */ 1555 */
1348 spin_lock(&clp->cl_lock); 1556 spin_lock(&clp->cl_lock);
1349 list_add_tail(&lo->plh_layouts, &server->layouts); 1557 if (list_empty(&lo->plh_layouts))
1558 list_add_tail(&lo->plh_layouts, &server->layouts);
1350 spin_unlock(&clp->cl_lock); 1559 spin_unlock(&clp->cl_lock);
1351 } 1560 }
1352 1561
@@ -1359,8 +1568,11 @@ pnfs_update_layout(struct inode *ino,
1359 arg.length = PAGE_CACHE_ALIGN(arg.length); 1568 arg.length = PAGE_CACHE_ALIGN(arg.length);
1360 1569
1361 lseg = send_layoutget(lo, ctx, &arg, gfp_flags); 1570 lseg = send_layoutget(lo, ctx, &arg, gfp_flags);
1571 pnfs_clear_retry_layoutget(lo);
1362 atomic_dec(&lo->plh_outstanding); 1572 atomic_dec(&lo->plh_outstanding);
1363out_put_layout_hdr: 1573out_put_layout_hdr:
1574 if (first)
1575 pnfs_clear_first_layoutget(lo);
1364 pnfs_put_layout_hdr(lo); 1576 pnfs_put_layout_hdr(lo);
1365out: 1577out:
1366 dprintk("%s: inode %s/%llu pNFS layout segment %s for " 1578 dprintk("%s: inode %s/%llu pNFS layout segment %s for "
@@ -1409,7 +1621,7 @@ pnfs_layout_process(struct nfs4_layoutget *lgp)
1409 goto out_forget_reply; 1621 goto out_forget_reply;
1410 } 1622 }
1411 1623
1412 if (pnfs_layoutgets_blocked(lo, 1)) { 1624 if (pnfs_layoutgets_blocked(lo, &lgp->args.range, 1)) {
1413 dprintk("%s forget reply due to state\n", __func__); 1625 dprintk("%s forget reply due to state\n", __func__);
1414 goto out_forget_reply; 1626 goto out_forget_reply;
1415 } 1627 }
@@ -1456,24 +1668,79 @@ out_forget_reply:
1456 goto out; 1668 goto out;
1457} 1669}
1458 1670
1671static void
1672pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo,
1673 struct list_head *tmp_list,
1674 struct pnfs_layout_range *return_range)
1675{
1676 struct pnfs_layout_segment *lseg, *next;
1677
1678 dprintk("%s:Begin lo %p\n", __func__, lo);
1679
1680 if (list_empty(&lo->plh_segs))
1681 return;
1682
1683 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
1684 if (should_free_lseg(&lseg->pls_range, return_range)) {
1685 dprintk("%s: marking lseg %p iomode %d "
1686 "offset %llu length %llu\n", __func__,
1687 lseg, lseg->pls_range.iomode,
1688 lseg->pls_range.offset,
1689 lseg->pls_range.length);
1690 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
1691 mark_lseg_invalid(lseg, tmp_list);
1692 }
1693}
1694
1695void pnfs_error_mark_layout_for_return(struct inode *inode,
1696 struct pnfs_layout_segment *lseg)
1697{
1698 struct pnfs_layout_hdr *lo = NFS_I(inode)->layout;
1699 int iomode = pnfs_iomode_to_fail_bit(lseg->pls_range.iomode);
1700 struct pnfs_layout_range range = {
1701 .iomode = lseg->pls_range.iomode,
1702 .offset = 0,
1703 .length = NFS4_MAX_UINT64,
1704 };
1705 LIST_HEAD(free_me);
1706
1707 spin_lock(&inode->i_lock);
1708 /* set failure bit so that pnfs path will be retried later */
1709 pnfs_layout_set_fail_bit(lo, iomode);
1710 set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags);
1711 if (lo->plh_return_iomode == 0)
1712 lo->plh_return_iomode = range.iomode;
1713 else if (lo->plh_return_iomode != range.iomode)
1714 lo->plh_return_iomode = IOMODE_ANY;
1715 /*
1716 * mark all matching lsegs so that we are sure to have no live
1717 * segments at hand when sending layoutreturn. See pnfs_put_lseg()
1718 * for how it works.
1719 */
1720 pnfs_mark_matching_lsegs_return(lo, &free_me, &range);
1721 spin_unlock(&inode->i_lock);
1722 pnfs_free_lseg_list(&free_me);
1723}
1724EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return);
1725
1459void 1726void
1460pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) 1727pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
1461{ 1728{
1462 u64 rd_size = req->wb_bytes; 1729 u64 rd_size = req->wb_bytes;
1463 1730
1464 WARN_ON_ONCE(pgio->pg_lseg != NULL); 1731 if (pgio->pg_lseg == NULL) {
1465 1732 if (pgio->pg_dreq == NULL)
1466 if (pgio->pg_dreq == NULL) 1733 rd_size = i_size_read(pgio->pg_inode) - req_offset(req);
1467 rd_size = i_size_read(pgio->pg_inode) - req_offset(req); 1734 else
1468 else 1735 rd_size = nfs_dreq_bytes_left(pgio->pg_dreq);
1469 rd_size = nfs_dreq_bytes_left(pgio->pg_dreq); 1736
1470 1737 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
1471 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, 1738 req->wb_context,
1472 req->wb_context, 1739 req_offset(req),
1473 req_offset(req), 1740 rd_size,
1474 rd_size, 1741 IOMODE_READ,
1475 IOMODE_READ, 1742 GFP_KERNEL);
1476 GFP_KERNEL); 1743 }
1477 /* If no lseg, fall back to read through mds */ 1744 /* If no lseg, fall back to read through mds */
1478 if (pgio->pg_lseg == NULL) 1745 if (pgio->pg_lseg == NULL)
1479 nfs_pageio_reset_read_mds(pgio); 1746 nfs_pageio_reset_read_mds(pgio);
@@ -1485,27 +1752,36 @@ void
1485pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio, 1752pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio,
1486 struct nfs_page *req, u64 wb_size) 1753 struct nfs_page *req, u64 wb_size)
1487{ 1754{
1488 WARN_ON_ONCE(pgio->pg_lseg != NULL); 1755 if (pgio->pg_lseg == NULL)
1489 1756 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
1490 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, 1757 req->wb_context,
1491 req->wb_context, 1758 req_offset(req),
1492 req_offset(req), 1759 wb_size,
1493 wb_size, 1760 IOMODE_RW,
1494 IOMODE_RW, 1761 GFP_NOFS);
1495 GFP_NOFS);
1496 /* If no lseg, fall back to write through mds */ 1762 /* If no lseg, fall back to write through mds */
1497 if (pgio->pg_lseg == NULL) 1763 if (pgio->pg_lseg == NULL)
1498 nfs_pageio_reset_write_mds(pgio); 1764 nfs_pageio_reset_write_mds(pgio);
1499} 1765}
1500EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write); 1766EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);
1501 1767
1768void
1769pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc)
1770{
1771 if (desc->pg_lseg) {
1772 pnfs_put_lseg(desc->pg_lseg);
1773 desc->pg_lseg = NULL;
1774 }
1775}
1776EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup);
1777
1502/* 1778/*
1503 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number 1779 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
1504 * of bytes (maximum @req->wb_bytes) that can be coalesced. 1780 * of bytes (maximum @req->wb_bytes) that can be coalesced.
1505 */ 1781 */
1506size_t 1782size_t
1507pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev, 1783pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio,
1508 struct nfs_page *req) 1784 struct nfs_page *prev, struct nfs_page *req)
1509{ 1785{
1510 unsigned int size; 1786 unsigned int size;
1511 u64 seg_end, req_start, seg_left; 1787 u64 seg_end, req_start, seg_left;
@@ -1529,10 +1805,16 @@ pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
1529 seg_end = end_offset(pgio->pg_lseg->pls_range.offset, 1805 seg_end = end_offset(pgio->pg_lseg->pls_range.offset,
1530 pgio->pg_lseg->pls_range.length); 1806 pgio->pg_lseg->pls_range.length);
1531 req_start = req_offset(req); 1807 req_start = req_offset(req);
1532 WARN_ON_ONCE(req_start > seg_end); 1808 WARN_ON_ONCE(req_start >= seg_end);
1533 /* start of request is past the last byte of this segment */ 1809 /* start of request is past the last byte of this segment */
1534 if (req_start >= seg_end) 1810 if (req_start >= seg_end) {
1811 /* reference the new lseg */
1812 if (pgio->pg_ops->pg_cleanup)
1813 pgio->pg_ops->pg_cleanup(pgio);
1814 if (pgio->pg_ops->pg_init)
1815 pgio->pg_ops->pg_init(pgio, req);
1535 return 0; 1816 return 0;
1817 }
1536 1818
1537 /* adjust 'size' iff there are fewer bytes left in the 1819 /* adjust 'size' iff there are fewer bytes left in the
1538 * segment than what nfs_generic_pg_test returned */ 1820 * segment than what nfs_generic_pg_test returned */
@@ -1587,10 +1869,12 @@ static void
1587pnfs_write_through_mds(struct nfs_pageio_descriptor *desc, 1869pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
1588 struct nfs_pgio_header *hdr) 1870 struct nfs_pgio_header *hdr)
1589{ 1871{
1872 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1873
1590 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { 1874 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1591 list_splice_tail_init(&hdr->pages, &desc->pg_list); 1875 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
1592 nfs_pageio_reset_write_mds(desc); 1876 nfs_pageio_reset_write_mds(desc);
1593 desc->pg_recoalesce = 1; 1877 mirror->pg_recoalesce = 1;
1594 } 1878 }
1595 nfs_pgio_data_destroy(hdr); 1879 nfs_pgio_data_destroy(hdr);
1596} 1880}
@@ -1624,11 +1908,9 @@ pnfs_do_write(struct nfs_pageio_descriptor *desc,
1624 struct pnfs_layout_segment *lseg = desc->pg_lseg; 1908 struct pnfs_layout_segment *lseg = desc->pg_lseg;
1625 enum pnfs_try_status trypnfs; 1909 enum pnfs_try_status trypnfs;
1626 1910
1627 desc->pg_lseg = NULL;
1628 trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how); 1911 trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how);
1629 if (trypnfs == PNFS_NOT_ATTEMPTED) 1912 if (trypnfs == PNFS_NOT_ATTEMPTED)
1630 pnfs_write_through_mds(desc, hdr); 1913 pnfs_write_through_mds(desc, hdr);
1631 pnfs_put_lseg(lseg);
1632} 1914}
1633 1915
1634static void pnfs_writehdr_free(struct nfs_pgio_header *hdr) 1916static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
@@ -1641,24 +1923,23 @@ EXPORT_SYMBOL_GPL(pnfs_writehdr_free);
1641int 1923int
1642pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc) 1924pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
1643{ 1925{
1926 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1927
1644 struct nfs_pgio_header *hdr; 1928 struct nfs_pgio_header *hdr;
1645 int ret; 1929 int ret;
1646 1930
1647 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops); 1931 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
1648 if (!hdr) { 1932 if (!hdr) {
1649 desc->pg_completion_ops->error_cleanup(&desc->pg_list); 1933 desc->pg_completion_ops->error_cleanup(&mirror->pg_list);
1650 pnfs_put_lseg(desc->pg_lseg);
1651 desc->pg_lseg = NULL;
1652 return -ENOMEM; 1934 return -ENOMEM;
1653 } 1935 }
1654 nfs_pgheader_init(desc, hdr, pnfs_writehdr_free); 1936 nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
1937
1655 hdr->lseg = pnfs_get_lseg(desc->pg_lseg); 1938 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
1656 ret = nfs_generic_pgio(desc, hdr); 1939 ret = nfs_generic_pgio(desc, hdr);
1657 if (ret != 0) { 1940 if (!ret)
1658 pnfs_put_lseg(desc->pg_lseg);
1659 desc->pg_lseg = NULL;
1660 } else
1661 pnfs_do_write(desc, hdr, desc->pg_ioflags); 1941 pnfs_do_write(desc, hdr, desc->pg_ioflags);
1942
1662 return ret; 1943 return ret;
1663} 1944}
1664EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages); 1945EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);
@@ -1703,10 +1984,12 @@ static void
1703pnfs_read_through_mds(struct nfs_pageio_descriptor *desc, 1984pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
1704 struct nfs_pgio_header *hdr) 1985 struct nfs_pgio_header *hdr)
1705{ 1986{
1987 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1988
1706 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { 1989 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1707 list_splice_tail_init(&hdr->pages, &desc->pg_list); 1990 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
1708 nfs_pageio_reset_read_mds(desc); 1991 nfs_pageio_reset_read_mds(desc);
1709 desc->pg_recoalesce = 1; 1992 mirror->pg_recoalesce = 1;
1710 } 1993 }
1711 nfs_pgio_data_destroy(hdr); 1994 nfs_pgio_data_destroy(hdr);
1712} 1995}
@@ -1735,18 +2018,29 @@ pnfs_try_to_read_data(struct nfs_pgio_header *hdr,
1735 return trypnfs; 2018 return trypnfs;
1736} 2019}
1737 2020
2021/* Resend all requests through pnfs. */
2022int pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr)
2023{
2024 struct nfs_pageio_descriptor pgio;
2025
2026 nfs_pageio_init_read(&pgio, hdr->inode, false, hdr->completion_ops);
2027 return nfs_pageio_resend(&pgio, hdr);
2028}
2029EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs);
2030
1738static void 2031static void
1739pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr) 2032pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr)
1740{ 2033{
1741 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; 2034 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
1742 struct pnfs_layout_segment *lseg = desc->pg_lseg; 2035 struct pnfs_layout_segment *lseg = desc->pg_lseg;
1743 enum pnfs_try_status trypnfs; 2036 enum pnfs_try_status trypnfs;
2037 int err = 0;
1744 2038
1745 desc->pg_lseg = NULL;
1746 trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg); 2039 trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg);
1747 if (trypnfs == PNFS_NOT_ATTEMPTED) 2040 if (trypnfs == PNFS_TRY_AGAIN)
2041 err = pnfs_read_resend_pnfs(hdr);
2042 if (trypnfs == PNFS_NOT_ATTEMPTED || err)
1748 pnfs_read_through_mds(desc, hdr); 2043 pnfs_read_through_mds(desc, hdr);
1749 pnfs_put_lseg(lseg);
1750} 2044}
1751 2045
1752static void pnfs_readhdr_free(struct nfs_pgio_header *hdr) 2046static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
@@ -1759,24 +2053,20 @@ EXPORT_SYMBOL_GPL(pnfs_readhdr_free);
1759int 2053int
1760pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc) 2054pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
1761{ 2055{
2056 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2057
1762 struct nfs_pgio_header *hdr; 2058 struct nfs_pgio_header *hdr;
1763 int ret; 2059 int ret;
1764 2060
1765 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops); 2061 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
1766 if (!hdr) { 2062 if (!hdr) {
1767 desc->pg_completion_ops->error_cleanup(&desc->pg_list); 2063 desc->pg_completion_ops->error_cleanup(&mirror->pg_list);
1768 ret = -ENOMEM; 2064 return -ENOMEM;
1769 pnfs_put_lseg(desc->pg_lseg);
1770 desc->pg_lseg = NULL;
1771 return ret;
1772 } 2065 }
1773 nfs_pgheader_init(desc, hdr, pnfs_readhdr_free); 2066 nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
1774 hdr->lseg = pnfs_get_lseg(desc->pg_lseg); 2067 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
1775 ret = nfs_generic_pgio(desc, hdr); 2068 ret = nfs_generic_pgio(desc, hdr);
1776 if (ret != 0) { 2069 if (!ret)
1777 pnfs_put_lseg(desc->pg_lseg);
1778 desc->pg_lseg = NULL;
1779 } else
1780 pnfs_do_read(desc, hdr); 2070 pnfs_do_read(desc, hdr);
1781 return ret; 2071 return ret;
1782} 2072}
@@ -1982,6 +2272,7 @@ clear_layoutcommitting:
1982 pnfs_clear_layoutcommitting(inode); 2272 pnfs_clear_layoutcommitting(inode);
1983 goto out; 2273 goto out;
1984} 2274}
2275EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode);
1985 2276
1986struct nfs4_threshold *pnfs_mdsthreshold_alloc(void) 2277struct nfs4_threshold *pnfs_mdsthreshold_alloc(void)
1987{ 2278{
diff --git a/fs/nfs/pnfs.h b/fs/nfs/pnfs.h
index a98d8fd9637f..797cd6253adf 100644
--- a/fs/nfs/pnfs.h
+++ b/fs/nfs/pnfs.h
@@ -38,6 +38,25 @@ enum {
38 NFS_LSEG_VALID = 0, /* cleared when lseg is recalled/returned */ 38 NFS_LSEG_VALID = 0, /* cleared when lseg is recalled/returned */
39 NFS_LSEG_ROC, /* roc bit received from server */ 39 NFS_LSEG_ROC, /* roc bit received from server */
40 NFS_LSEG_LAYOUTCOMMIT, /* layoutcommit bit set for layoutcommit */ 40 NFS_LSEG_LAYOUTCOMMIT, /* layoutcommit bit set for layoutcommit */
41 NFS_LSEG_LAYOUTRETURN, /* layoutreturn bit set for layoutreturn */
42};
43
44/* Individual ip address */
45struct nfs4_pnfs_ds_addr {
46 struct sockaddr_storage da_addr;
47 size_t da_addrlen;
48 struct list_head da_node; /* nfs4_pnfs_dev_hlist dev_dslist */
49 char *da_remotestr; /* human readable addr+port */
50};
51
52struct nfs4_pnfs_ds {
53 struct list_head ds_node; /* nfs4_pnfs_dev_hlist dev_dslist */
54 char *ds_remotestr; /* comma sep list of addrs */
55 struct list_head ds_addrs;
56 struct nfs_client *ds_clp;
57 atomic_t ds_count;
58 unsigned long ds_state;
59#define NFS4DS_CONNECTING 0 /* ds is establishing connection */
41}; 60};
42 61
43struct pnfs_layout_segment { 62struct pnfs_layout_segment {
@@ -53,19 +72,34 @@ struct pnfs_layout_segment {
53enum pnfs_try_status { 72enum pnfs_try_status {
54 PNFS_ATTEMPTED = 0, 73 PNFS_ATTEMPTED = 0,
55 PNFS_NOT_ATTEMPTED = 1, 74 PNFS_NOT_ATTEMPTED = 1,
75 PNFS_TRY_AGAIN = 2,
56}; 76};
57 77
58#ifdef CONFIG_NFS_V4_1 78#ifdef CONFIG_NFS_V4_1
59 79
60#define LAYOUT_NFSV4_1_MODULE_PREFIX "nfs-layouttype4" 80#define LAYOUT_NFSV4_1_MODULE_PREFIX "nfs-layouttype4"
61 81
82/*
83 * Default data server connection timeout and retrans vaules.
84 * Set by module parameters dataserver_timeo and dataserver_retrans.
85 */
86#define NFS4_DEF_DS_TIMEO 600 /* in tenths of a second */
87#define NFS4_DEF_DS_RETRANS 5
88
89/* error codes for internal use */
90#define NFS4ERR_RESET_TO_MDS 12001
91#define NFS4ERR_RESET_TO_PNFS 12002
92
62enum { 93enum {
63 NFS_LAYOUT_RO_FAILED = 0, /* get ro layout failed stop trying */ 94 NFS_LAYOUT_RO_FAILED = 0, /* get ro layout failed stop trying */
64 NFS_LAYOUT_RW_FAILED, /* get rw layout failed stop trying */ 95 NFS_LAYOUT_RW_FAILED, /* get rw layout failed stop trying */
65 NFS_LAYOUT_BULK_RECALL, /* bulk recall affecting layout */ 96 NFS_LAYOUT_BULK_RECALL, /* bulk recall affecting layout */
66 NFS_LAYOUT_ROC, /* some lseg had roc bit set */ 97 NFS_LAYOUT_ROC, /* some lseg had roc bit set */
67 NFS_LAYOUT_RETURN, /* Return this layout ASAP */ 98 NFS_LAYOUT_RETURN, /* Return this layout ASAP */
99 NFS_LAYOUT_RETURN_BEFORE_CLOSE, /* Return this layout before close */
68 NFS_LAYOUT_INVALID_STID, /* layout stateid id is invalid */ 100 NFS_LAYOUT_INVALID_STID, /* layout stateid id is invalid */
101 NFS_LAYOUT_FIRST_LAYOUTGET, /* Serialize first layoutget */
102 NFS_LAYOUT_RETRY_LAYOUTGET, /* Retry layoutget */
69}; 103};
70 104
71enum layoutdriver_policy_flags { 105enum layoutdriver_policy_flags {
@@ -106,7 +140,8 @@ struct pnfs_layoutdriver_type {
106 struct pnfs_ds_commit_info *(*get_ds_info) (struct inode *inode); 140 struct pnfs_ds_commit_info *(*get_ds_info) (struct inode *inode);
107 void (*mark_request_commit) (struct nfs_page *req, 141 void (*mark_request_commit) (struct nfs_page *req,
108 struct pnfs_layout_segment *lseg, 142 struct pnfs_layout_segment *lseg,
109 struct nfs_commit_info *cinfo); 143 struct nfs_commit_info *cinfo,
144 u32 ds_commit_idx);
110 void (*clear_request_commit) (struct nfs_page *req, 145 void (*clear_request_commit) (struct nfs_page *req,
111 struct nfs_commit_info *cinfo); 146 struct nfs_commit_info *cinfo);
112 int (*scan_commit_lists) (struct nfs_commit_info *cinfo, 147 int (*scan_commit_lists) (struct nfs_commit_info *cinfo,
@@ -154,6 +189,7 @@ struct pnfs_layout_hdr {
154 u32 plh_barrier; /* ignore lower seqids */ 189 u32 plh_barrier; /* ignore lower seqids */
155 unsigned long plh_retry_timestamp; 190 unsigned long plh_retry_timestamp;
156 unsigned long plh_flags; 191 unsigned long plh_flags;
192 enum pnfs_iomode plh_return_iomode;
157 loff_t plh_lwb; /* last write byte for layoutcommit */ 193 loff_t plh_lwb; /* last write byte for layoutcommit */
158 struct rpc_cred *plh_lc_cred; /* layoutcommit cred */ 194 struct rpc_cred *plh_lc_cred; /* layoutcommit cred */
159 struct inode *plh_inode; 195 struct inode *plh_inode;
@@ -185,7 +221,7 @@ extern int nfs4_proc_getdeviceinfo(struct nfs_server *server,
185 struct pnfs_device *dev, 221 struct pnfs_device *dev,
186 struct rpc_cred *cred); 222 struct rpc_cred *cred);
187extern struct pnfs_layout_segment* nfs4_proc_layoutget(struct nfs4_layoutget *lgp, gfp_t gfp_flags); 223extern struct pnfs_layout_segment* nfs4_proc_layoutget(struct nfs4_layoutget *lgp, gfp_t gfp_flags);
188extern int nfs4_proc_layoutreturn(struct nfs4_layoutreturn *lrp); 224extern int nfs4_proc_layoutreturn(struct nfs4_layoutreturn *lrp, bool sync);
189 225
190/* pnfs.c */ 226/* pnfs.c */
191void pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo); 227void pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo);
@@ -198,6 +234,7 @@ void pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *, struct nfs_page *
198int pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc); 234int pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc);
199void pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio, 235void pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio,
200 struct nfs_page *req, u64 wb_size); 236 struct nfs_page *req, u64 wb_size);
237void pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *);
201int pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc); 238int pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc);
202size_t pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio, 239size_t pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio,
203 struct nfs_page *prev, struct nfs_page *req); 240 struct nfs_page *prev, struct nfs_page *req);
@@ -217,6 +254,7 @@ void pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo,
217 bool update_barrier); 254 bool update_barrier);
218int pnfs_choose_layoutget_stateid(nfs4_stateid *dst, 255int pnfs_choose_layoutget_stateid(nfs4_stateid *dst,
219 struct pnfs_layout_hdr *lo, 256 struct pnfs_layout_hdr *lo,
257 struct pnfs_layout_range *range,
220 struct nfs4_state *open_state); 258 struct nfs4_state *open_state);
221int pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo, 259int pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
222 struct list_head *tmp_list, 260 struct list_head *tmp_list,
@@ -233,17 +271,21 @@ int _pnfs_return_layout(struct inode *);
233int pnfs_commit_and_return_layout(struct inode *); 271int pnfs_commit_and_return_layout(struct inode *);
234void pnfs_ld_write_done(struct nfs_pgio_header *); 272void pnfs_ld_write_done(struct nfs_pgio_header *);
235void pnfs_ld_read_done(struct nfs_pgio_header *); 273void pnfs_ld_read_done(struct nfs_pgio_header *);
274int pnfs_read_resend_pnfs(struct nfs_pgio_header *);
236struct pnfs_layout_segment *pnfs_update_layout(struct inode *ino, 275struct pnfs_layout_segment *pnfs_update_layout(struct inode *ino,
237 struct nfs_open_context *ctx, 276 struct nfs_open_context *ctx,
238 loff_t pos, 277 loff_t pos,
239 u64 count, 278 u64 count,
240 enum pnfs_iomode iomode, 279 enum pnfs_iomode iomode,
241 gfp_t gfp_flags); 280 gfp_t gfp_flags);
281void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo);
242 282
243void nfs4_deviceid_mark_client_invalid(struct nfs_client *clp); 283void nfs4_deviceid_mark_client_invalid(struct nfs_client *clp);
244int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *); 284int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *);
245int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *); 285int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *);
246struct nfs4_threshold *pnfs_mdsthreshold_alloc(void); 286struct nfs4_threshold *pnfs_mdsthreshold_alloc(void);
287void pnfs_error_mark_layout_for_return(struct inode *inode,
288 struct pnfs_layout_segment *lseg);
247 289
248/* nfs4_deviceid_flags */ 290/* nfs4_deviceid_flags */
249enum { 291enum {
@@ -275,6 +317,34 @@ void nfs4_mark_deviceid_unavailable(struct nfs4_deviceid_node *node);
275bool nfs4_test_deviceid_unavailable(struct nfs4_deviceid_node *node); 317bool nfs4_test_deviceid_unavailable(struct nfs4_deviceid_node *node);
276void nfs4_deviceid_purge_client(const struct nfs_client *); 318void nfs4_deviceid_purge_client(const struct nfs_client *);
277 319
320/* pnfs_nfs.c */
321void pnfs_generic_clear_request_commit(struct nfs_page *req,
322 struct nfs_commit_info *cinfo);
323void pnfs_generic_commit_release(void *calldata);
324void pnfs_generic_prepare_to_resend_writes(struct nfs_commit_data *data);
325void pnfs_generic_rw_release(void *data);
326void pnfs_generic_recover_commit_reqs(struct list_head *dst,
327 struct nfs_commit_info *cinfo);
328int pnfs_generic_commit_pagelist(struct inode *inode,
329 struct list_head *mds_pages,
330 int how,
331 struct nfs_commit_info *cinfo,
332 int (*initiate_commit)(struct nfs_commit_data *data,
333 int how));
334int pnfs_generic_scan_commit_lists(struct nfs_commit_info *cinfo, int max);
335void pnfs_generic_write_commit_done(struct rpc_task *task, void *data);
336void nfs4_pnfs_ds_put(struct nfs4_pnfs_ds *ds);
337struct nfs4_pnfs_ds *nfs4_pnfs_ds_add(struct list_head *dsaddrs,
338 gfp_t gfp_flags);
339void nfs4_pnfs_v3_ds_connect_unload(void);
340void nfs4_pnfs_ds_connect(struct nfs_server *mds_srv, struct nfs4_pnfs_ds *ds,
341 struct nfs4_deviceid_node *devid, unsigned int timeo,
342 unsigned int retrans, u32 version, u32 minor_version,
343 rpc_authflavor_t au_flavor);
344struct nfs4_pnfs_ds_addr *nfs4_decode_mp_ds_addr(struct net *net,
345 struct xdr_stream *xdr,
346 gfp_t gfp_flags);
347
278static inline bool nfs_have_layout(struct inode *inode) 348static inline bool nfs_have_layout(struct inode *inode)
279{ 349{
280 return NFS_I(inode)->layout != NULL; 350 return NFS_I(inode)->layout != NULL;
@@ -287,6 +357,26 @@ nfs4_get_deviceid(struct nfs4_deviceid_node *d)
287 return d; 357 return d;
288} 358}
289 359
360static inline void pnfs_set_retry_layoutget(struct pnfs_layout_hdr *lo)
361{
362 if (!test_and_set_bit(NFS_LAYOUT_RETRY_LAYOUTGET, &lo->plh_flags))
363 atomic_inc(&lo->plh_refcount);
364}
365
366static inline void pnfs_clear_retry_layoutget(struct pnfs_layout_hdr *lo)
367{
368 if (test_and_clear_bit(NFS_LAYOUT_RETRY_LAYOUTGET, &lo->plh_flags)) {
369 atomic_dec(&lo->plh_refcount);
370 /* wake up waiters for LAYOUTRETURN as that is not needed */
371 wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN);
372 }
373}
374
375static inline bool pnfs_should_retry_layoutget(struct pnfs_layout_hdr *lo)
376{
377 return test_bit(NFS_LAYOUT_RETRY_LAYOUTGET, &lo->plh_flags);
378}
379
290static inline struct pnfs_layout_segment * 380static inline struct pnfs_layout_segment *
291pnfs_get_lseg(struct pnfs_layout_segment *lseg) 381pnfs_get_lseg(struct pnfs_layout_segment *lseg)
292{ 382{
@@ -322,16 +412,22 @@ pnfs_get_ds_info(struct inode *inode)
322 return ld->get_ds_info(inode); 412 return ld->get_ds_info(inode);
323} 413}
324 414
415static inline void
416pnfs_generic_mark_devid_invalid(struct nfs4_deviceid_node *node)
417{
418 set_bit(NFS_DEVICEID_INVALID, &node->flags);
419}
420
325static inline bool 421static inline bool
326pnfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 422pnfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg,
327 struct nfs_commit_info *cinfo) 423 struct nfs_commit_info *cinfo, u32 ds_commit_idx)
328{ 424{
329 struct inode *inode = req->wb_context->dentry->d_inode; 425 struct inode *inode = req->wb_context->dentry->d_inode;
330 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld; 426 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
331 427
332 if (lseg == NULL || ld->mark_request_commit == NULL) 428 if (lseg == NULL || ld->mark_request_commit == NULL)
333 return false; 429 return false;
334 ld->mark_request_commit(req, lseg, cinfo); 430 ld->mark_request_commit(req, lseg, cinfo, ds_commit_idx);
335 return true; 431 return true;
336} 432}
337 433
@@ -357,15 +453,6 @@ pnfs_scan_commit_lists(struct inode *inode, struct nfs_commit_info *cinfo,
357 return NFS_SERVER(inode)->pnfs_curr_ld->scan_commit_lists(cinfo, max); 453 return NFS_SERVER(inode)->pnfs_curr_ld->scan_commit_lists(cinfo, max);
358} 454}
359 455
360static inline void
361pnfs_recover_commit_reqs(struct inode *inode, struct list_head *list,
362 struct nfs_commit_info *cinfo)
363{
364 if (cinfo->ds == NULL || cinfo->ds->nwritten == 0)
365 return;
366 NFS_SERVER(inode)->pnfs_curr_ld->recover_commit_reqs(list, cinfo);
367}
368
369static inline struct nfs_page * 456static inline struct nfs_page *
370pnfs_search_commit_reqs(struct inode *inode, struct nfs_commit_info *cinfo, 457pnfs_search_commit_reqs(struct inode *inode, struct nfs_commit_info *cinfo,
371 struct page *page) 458 struct page *page)
@@ -523,7 +610,7 @@ pnfs_get_ds_info(struct inode *inode)
523 610
524static inline bool 611static inline bool
525pnfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 612pnfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg,
526 struct nfs_commit_info *cinfo) 613 struct nfs_commit_info *cinfo, u32 ds_commit_idx)
527{ 614{
528 return false; 615 return false;
529} 616}
@@ -541,12 +628,6 @@ pnfs_scan_commit_lists(struct inode *inode, struct nfs_commit_info *cinfo,
541 return 0; 628 return 0;
542} 629}
543 630
544static inline void
545pnfs_recover_commit_reqs(struct inode *inode, struct list_head *list,
546 struct nfs_commit_info *cinfo)
547{
548}
549
550static inline struct nfs_page * 631static inline struct nfs_page *
551pnfs_search_commit_reqs(struct inode *inode, struct nfs_commit_info *cinfo, 632pnfs_search_commit_reqs(struct inode *inode, struct nfs_commit_info *cinfo,
552 struct page *page) 633 struct page *page)
@@ -578,6 +659,10 @@ static inline struct nfs4_threshold *pnfs_mdsthreshold_alloc(void)
578 return NULL; 659 return NULL;
579} 660}
580 661
662static inline void nfs4_pnfs_v3_ds_connect_unload(void)
663{
664}
665
581#endif /* CONFIG_NFS_V4_1 */ 666#endif /* CONFIG_NFS_V4_1 */
582 667
583#endif /* FS_NFS_PNFS_H */ 668#endif /* FS_NFS_PNFS_H */
diff --git a/fs/nfs/pnfs_nfs.c b/fs/nfs/pnfs_nfs.c
new file mode 100644
index 000000000000..fdc4f6562bb7
--- /dev/null
+++ b/fs/nfs/pnfs_nfs.c
@@ -0,0 +1,840 @@
1/*
2 * Common NFS I/O operations for the pnfs file based
3 * layout drivers.
4 *
5 * Copyright (c) 2014, Primary Data, Inc. All rights reserved.
6 *
7 * Tom Haynes <loghyr@primarydata.com>
8 */
9
10#include <linux/nfs_fs.h>
11#include <linux/nfs_page.h>
12#include <linux/sunrpc/addr.h>
13#include <linux/module.h>
14
15#include "nfs4session.h"
16#include "internal.h"
17#include "pnfs.h"
18
19#define NFSDBG_FACILITY NFSDBG_PNFS
20
21void pnfs_generic_rw_release(void *data)
22{
23 struct nfs_pgio_header *hdr = data;
24
25 nfs_put_client(hdr->ds_clp);
26 hdr->mds_ops->rpc_release(data);
27}
28EXPORT_SYMBOL_GPL(pnfs_generic_rw_release);
29
30/* Fake up some data that will cause nfs_commit_release to retry the writes. */
31void pnfs_generic_prepare_to_resend_writes(struct nfs_commit_data *data)
32{
33 struct nfs_page *first = nfs_list_entry(data->pages.next);
34
35 data->task.tk_status = 0;
36 memcpy(&data->verf.verifier, &first->wb_verf,
37 sizeof(data->verf.verifier));
38 data->verf.verifier.data[0]++; /* ensure verifier mismatch */
39}
40EXPORT_SYMBOL_GPL(pnfs_generic_prepare_to_resend_writes);
41
42void pnfs_generic_write_commit_done(struct rpc_task *task, void *data)
43{
44 struct nfs_commit_data *wdata = data;
45
46 /* Note this may cause RPC to be resent */
47 wdata->mds_ops->rpc_call_done(task, data);
48}
49EXPORT_SYMBOL_GPL(pnfs_generic_write_commit_done);
50
51void pnfs_generic_commit_release(void *calldata)
52{
53 struct nfs_commit_data *data = calldata;
54
55 data->completion_ops->completion(data);
56 pnfs_put_lseg(data->lseg);
57 nfs_put_client(data->ds_clp);
58 nfs_commitdata_release(data);
59}
60EXPORT_SYMBOL_GPL(pnfs_generic_commit_release);
61
62/* The generic layer is about to remove the req from the commit list.
63 * If this will make the bucket empty, it will need to put the lseg reference.
64 * Note this must be called holding the inode (/cinfo) lock
65 */
66void
67pnfs_generic_clear_request_commit(struct nfs_page *req,
68 struct nfs_commit_info *cinfo)
69{
70 struct pnfs_layout_segment *freeme = NULL;
71
72 if (!test_and_clear_bit(PG_COMMIT_TO_DS, &req->wb_flags))
73 goto out;
74 cinfo->ds->nwritten--;
75 if (list_is_singular(&req->wb_list)) {
76 struct pnfs_commit_bucket *bucket;
77
78 bucket = list_first_entry(&req->wb_list,
79 struct pnfs_commit_bucket,
80 written);
81 freeme = bucket->wlseg;
82 bucket->wlseg = NULL;
83 }
84out:
85 nfs_request_remove_commit_list(req, cinfo);
86 pnfs_put_lseg_locked(freeme);
87}
88EXPORT_SYMBOL_GPL(pnfs_generic_clear_request_commit);
89
90static int
91pnfs_generic_transfer_commit_list(struct list_head *src, struct list_head *dst,
92 struct nfs_commit_info *cinfo, int max)
93{
94 struct nfs_page *req, *tmp;
95 int ret = 0;
96
97 list_for_each_entry_safe(req, tmp, src, wb_list) {
98 if (!nfs_lock_request(req))
99 continue;
100 kref_get(&req->wb_kref);
101 if (cond_resched_lock(cinfo->lock))
102 list_safe_reset_next(req, tmp, wb_list);
103 nfs_request_remove_commit_list(req, cinfo);
104 clear_bit(PG_COMMIT_TO_DS, &req->wb_flags);
105 nfs_list_add_request(req, dst);
106 ret++;
107 if ((ret == max) && !cinfo->dreq)
108 break;
109 }
110 return ret;
111}
112
113static int
114pnfs_generic_scan_ds_commit_list(struct pnfs_commit_bucket *bucket,
115 struct nfs_commit_info *cinfo,
116 int max)
117{
118 struct list_head *src = &bucket->written;
119 struct list_head *dst = &bucket->committing;
120 int ret;
121
122 lockdep_assert_held(cinfo->lock);
123 ret = pnfs_generic_transfer_commit_list(src, dst, cinfo, max);
124 if (ret) {
125 cinfo->ds->nwritten -= ret;
126 cinfo->ds->ncommitting += ret;
127 bucket->clseg = bucket->wlseg;
128 if (list_empty(src))
129 bucket->wlseg = NULL;
130 else
131 pnfs_get_lseg(bucket->clseg);
132 }
133 return ret;
134}
135
136/* Move reqs from written to committing lists, returning count
137 * of number moved.
138 */
139int pnfs_generic_scan_commit_lists(struct nfs_commit_info *cinfo,
140 int max)
141{
142 int i, rv = 0, cnt;
143
144 lockdep_assert_held(cinfo->lock);
145 for (i = 0; i < cinfo->ds->nbuckets && max != 0; i++) {
146 cnt = pnfs_generic_scan_ds_commit_list(&cinfo->ds->buckets[i],
147 cinfo, max);
148 max -= cnt;
149 rv += cnt;
150 }
151 return rv;
152}
153EXPORT_SYMBOL_GPL(pnfs_generic_scan_commit_lists);
154
155/* Pull everything off the committing lists and dump into @dst. */
156void pnfs_generic_recover_commit_reqs(struct list_head *dst,
157 struct nfs_commit_info *cinfo)
158{
159 struct pnfs_commit_bucket *b;
160 struct pnfs_layout_segment *freeme;
161 int i;
162
163 lockdep_assert_held(cinfo->lock);
164restart:
165 for (i = 0, b = cinfo->ds->buckets; i < cinfo->ds->nbuckets; i++, b++) {
166 if (pnfs_generic_transfer_commit_list(&b->written, dst,
167 cinfo, 0)) {
168 freeme = b->wlseg;
169 b->wlseg = NULL;
170 spin_unlock(cinfo->lock);
171 pnfs_put_lseg(freeme);
172 spin_lock(cinfo->lock);
173 goto restart;
174 }
175 }
176 cinfo->ds->nwritten = 0;
177}
178EXPORT_SYMBOL_GPL(pnfs_generic_recover_commit_reqs);
179
180static void pnfs_generic_retry_commit(struct nfs_commit_info *cinfo, int idx)
181{
182 struct pnfs_ds_commit_info *fl_cinfo = cinfo->ds;
183 struct pnfs_commit_bucket *bucket;
184 struct pnfs_layout_segment *freeme;
185 int i;
186
187 for (i = idx; i < fl_cinfo->nbuckets; i++) {
188 bucket = &fl_cinfo->buckets[i];
189 if (list_empty(&bucket->committing))
190 continue;
191 nfs_retry_commit(&bucket->committing, bucket->clseg, cinfo, i);
192 spin_lock(cinfo->lock);
193 freeme = bucket->clseg;
194 bucket->clseg = NULL;
195 spin_unlock(cinfo->lock);
196 pnfs_put_lseg(freeme);
197 }
198}
199
200static unsigned int
201pnfs_generic_alloc_ds_commits(struct nfs_commit_info *cinfo,
202 struct list_head *list)
203{
204 struct pnfs_ds_commit_info *fl_cinfo;
205 struct pnfs_commit_bucket *bucket;
206 struct nfs_commit_data *data;
207 int i;
208 unsigned int nreq = 0;
209
210 fl_cinfo = cinfo->ds;
211 bucket = fl_cinfo->buckets;
212 for (i = 0; i < fl_cinfo->nbuckets; i++, bucket++) {
213 if (list_empty(&bucket->committing))
214 continue;
215 data = nfs_commitdata_alloc();
216 if (!data)
217 break;
218 data->ds_commit_index = i;
219 spin_lock(cinfo->lock);
220 data->lseg = bucket->clseg;
221 bucket->clseg = NULL;
222 spin_unlock(cinfo->lock);
223 list_add(&data->pages, list);
224 nreq++;
225 }
226
227 /* Clean up on error */
228 pnfs_generic_retry_commit(cinfo, i);
229 return nreq;
230}
231
232/* This follows nfs_commit_list pretty closely */
233int
234pnfs_generic_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
235 int how, struct nfs_commit_info *cinfo,
236 int (*initiate_commit)(struct nfs_commit_data *data,
237 int how))
238{
239 struct nfs_commit_data *data, *tmp;
240 LIST_HEAD(list);
241 unsigned int nreq = 0;
242
243 if (!list_empty(mds_pages)) {
244 data = nfs_commitdata_alloc();
245 if (data != NULL) {
246 data->lseg = NULL;
247 list_add(&data->pages, &list);
248 nreq++;
249 } else {
250 nfs_retry_commit(mds_pages, NULL, cinfo, 0);
251 pnfs_generic_retry_commit(cinfo, 0);
252 cinfo->completion_ops->error_cleanup(NFS_I(inode));
253 return -ENOMEM;
254 }
255 }
256
257 nreq += pnfs_generic_alloc_ds_commits(cinfo, &list);
258
259 if (nreq == 0) {
260 cinfo->completion_ops->error_cleanup(NFS_I(inode));
261 goto out;
262 }
263
264 atomic_add(nreq, &cinfo->mds->rpcs_out);
265
266 list_for_each_entry_safe(data, tmp, &list, pages) {
267 list_del_init(&data->pages);
268 if (!data->lseg) {
269 nfs_init_commit(data, mds_pages, NULL, cinfo);
270 nfs_initiate_commit(NFS_CLIENT(inode), data,
271 NFS_PROTO(data->inode),
272 data->mds_ops, how, 0);
273 } else {
274 struct pnfs_commit_bucket *buckets;
275
276 buckets = cinfo->ds->buckets;
277 nfs_init_commit(data,
278 &buckets[data->ds_commit_index].committing,
279 data->lseg,
280 cinfo);
281 initiate_commit(data, how);
282 }
283 }
284out:
285 cinfo->ds->ncommitting = 0;
286 return PNFS_ATTEMPTED;
287}
288EXPORT_SYMBOL_GPL(pnfs_generic_commit_pagelist);
289
290/*
291 * Data server cache
292 *
293 * Data servers can be mapped to different device ids.
294 * nfs4_pnfs_ds reference counting
295 * - set to 1 on allocation
296 * - incremented when a device id maps a data server already in the cache.
297 * - decremented when deviceid is removed from the cache.
298 */
299static DEFINE_SPINLOCK(nfs4_ds_cache_lock);
300static LIST_HEAD(nfs4_data_server_cache);
301
302/* Debug routines */
303static void
304print_ds(struct nfs4_pnfs_ds *ds)
305{
306 if (ds == NULL) {
307 printk(KERN_WARNING "%s NULL device\n", __func__);
308 return;
309 }
310 printk(KERN_WARNING " ds %s\n"
311 " ref count %d\n"
312 " client %p\n"
313 " cl_exchange_flags %x\n",
314 ds->ds_remotestr,
315 atomic_read(&ds->ds_count), ds->ds_clp,
316 ds->ds_clp ? ds->ds_clp->cl_exchange_flags : 0);
317}
318
319static bool
320same_sockaddr(struct sockaddr *addr1, struct sockaddr *addr2)
321{
322 struct sockaddr_in *a, *b;
323 struct sockaddr_in6 *a6, *b6;
324
325 if (addr1->sa_family != addr2->sa_family)
326 return false;
327
328 switch (addr1->sa_family) {
329 case AF_INET:
330 a = (struct sockaddr_in *)addr1;
331 b = (struct sockaddr_in *)addr2;
332
333 if (a->sin_addr.s_addr == b->sin_addr.s_addr &&
334 a->sin_port == b->sin_port)
335 return true;
336 break;
337
338 case AF_INET6:
339 a6 = (struct sockaddr_in6 *)addr1;
340 b6 = (struct sockaddr_in6 *)addr2;
341
342 /* LINKLOCAL addresses must have matching scope_id */
343 if (ipv6_addr_src_scope(&a6->sin6_addr) ==
344 IPV6_ADDR_SCOPE_LINKLOCAL &&
345 a6->sin6_scope_id != b6->sin6_scope_id)
346 return false;
347
348 if (ipv6_addr_equal(&a6->sin6_addr, &b6->sin6_addr) &&
349 a6->sin6_port == b6->sin6_port)
350 return true;
351 break;
352
353 default:
354 dprintk("%s: unhandled address family: %u\n",
355 __func__, addr1->sa_family);
356 return false;
357 }
358
359 return false;
360}
361
362static bool
363_same_data_server_addrs_locked(const struct list_head *dsaddrs1,
364 const struct list_head *dsaddrs2)
365{
366 struct nfs4_pnfs_ds_addr *da1, *da2;
367
368 /* step through both lists, comparing as we go */
369 for (da1 = list_first_entry(dsaddrs1, typeof(*da1), da_node),
370 da2 = list_first_entry(dsaddrs2, typeof(*da2), da_node);
371 da1 != NULL && da2 != NULL;
372 da1 = list_entry(da1->da_node.next, typeof(*da1), da_node),
373 da2 = list_entry(da2->da_node.next, typeof(*da2), da_node)) {
374 if (!same_sockaddr((struct sockaddr *)&da1->da_addr,
375 (struct sockaddr *)&da2->da_addr))
376 return false;
377 }
378 if (da1 == NULL && da2 == NULL)
379 return true;
380
381 return false;
382}
383
384/*
385 * Lookup DS by addresses. nfs4_ds_cache_lock is held
386 */
387static struct nfs4_pnfs_ds *
388_data_server_lookup_locked(const struct list_head *dsaddrs)
389{
390 struct nfs4_pnfs_ds *ds;
391
392 list_for_each_entry(ds, &nfs4_data_server_cache, ds_node)
393 if (_same_data_server_addrs_locked(&ds->ds_addrs, dsaddrs))
394 return ds;
395 return NULL;
396}
397
398static void destroy_ds(struct nfs4_pnfs_ds *ds)
399{
400 struct nfs4_pnfs_ds_addr *da;
401
402 dprintk("--> %s\n", __func__);
403 ifdebug(FACILITY)
404 print_ds(ds);
405
406 nfs_put_client(ds->ds_clp);
407
408 while (!list_empty(&ds->ds_addrs)) {
409 da = list_first_entry(&ds->ds_addrs,
410 struct nfs4_pnfs_ds_addr,
411 da_node);
412 list_del_init(&da->da_node);
413 kfree(da->da_remotestr);
414 kfree(da);
415 }
416
417 kfree(ds->ds_remotestr);
418 kfree(ds);
419}
420
421void nfs4_pnfs_ds_put(struct nfs4_pnfs_ds *ds)
422{
423 if (atomic_dec_and_lock(&ds->ds_count,
424 &nfs4_ds_cache_lock)) {
425 list_del_init(&ds->ds_node);
426 spin_unlock(&nfs4_ds_cache_lock);
427 destroy_ds(ds);
428 }
429}
430EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_put);
431
432/*
433 * Create a string with a human readable address and port to avoid
434 * complicated setup around many dprinks.
435 */
436static char *
437nfs4_pnfs_remotestr(struct list_head *dsaddrs, gfp_t gfp_flags)
438{
439 struct nfs4_pnfs_ds_addr *da;
440 char *remotestr;
441 size_t len;
442 char *p;
443
444 len = 3; /* '{', '}' and eol */
445 list_for_each_entry(da, dsaddrs, da_node) {
446 len += strlen(da->da_remotestr) + 1; /* string plus comma */
447 }
448
449 remotestr = kzalloc(len, gfp_flags);
450 if (!remotestr)
451 return NULL;
452
453 p = remotestr;
454 *(p++) = '{';
455 len--;
456 list_for_each_entry(da, dsaddrs, da_node) {
457 size_t ll = strlen(da->da_remotestr);
458
459 if (ll > len)
460 goto out_err;
461
462 memcpy(p, da->da_remotestr, ll);
463 p += ll;
464 len -= ll;
465
466 if (len < 1)
467 goto out_err;
468 (*p++) = ',';
469 len--;
470 }
471 if (len < 2)
472 goto out_err;
473 *(p++) = '}';
474 *p = '\0';
475 return remotestr;
476out_err:
477 kfree(remotestr);
478 return NULL;
479}
480
481/*
482 * Given a list of multipath struct nfs4_pnfs_ds_addr, add it to ds cache if
483 * uncached and return cached struct nfs4_pnfs_ds.
484 */
485struct nfs4_pnfs_ds *
486nfs4_pnfs_ds_add(struct list_head *dsaddrs, gfp_t gfp_flags)
487{
488 struct nfs4_pnfs_ds *tmp_ds, *ds = NULL;
489 char *remotestr;
490
491 if (list_empty(dsaddrs)) {
492 dprintk("%s: no addresses defined\n", __func__);
493 goto out;
494 }
495
496 ds = kzalloc(sizeof(*ds), gfp_flags);
497 if (!ds)
498 goto out;
499
500 /* this is only used for debugging, so it's ok if its NULL */
501 remotestr = nfs4_pnfs_remotestr(dsaddrs, gfp_flags);
502
503 spin_lock(&nfs4_ds_cache_lock);
504 tmp_ds = _data_server_lookup_locked(dsaddrs);
505 if (tmp_ds == NULL) {
506 INIT_LIST_HEAD(&ds->ds_addrs);
507 list_splice_init(dsaddrs, &ds->ds_addrs);
508 ds->ds_remotestr = remotestr;
509 atomic_set(&ds->ds_count, 1);
510 INIT_LIST_HEAD(&ds->ds_node);
511 ds->ds_clp = NULL;
512 list_add(&ds->ds_node, &nfs4_data_server_cache);
513 dprintk("%s add new data server %s\n", __func__,
514 ds->ds_remotestr);
515 } else {
516 kfree(remotestr);
517 kfree(ds);
518 atomic_inc(&tmp_ds->ds_count);
519 dprintk("%s data server %s found, inc'ed ds_count to %d\n",
520 __func__, tmp_ds->ds_remotestr,
521 atomic_read(&tmp_ds->ds_count));
522 ds = tmp_ds;
523 }
524 spin_unlock(&nfs4_ds_cache_lock);
525out:
526 return ds;
527}
528EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_add);
529
530static void nfs4_wait_ds_connect(struct nfs4_pnfs_ds *ds)
531{
532 might_sleep();
533 wait_on_bit(&ds->ds_state, NFS4DS_CONNECTING,
534 TASK_KILLABLE);
535}
536
537static void nfs4_clear_ds_conn_bit(struct nfs4_pnfs_ds *ds)
538{
539 smp_mb__before_atomic();
540 clear_bit(NFS4DS_CONNECTING, &ds->ds_state);
541 smp_mb__after_atomic();
542 wake_up_bit(&ds->ds_state, NFS4DS_CONNECTING);
543}
544
545static struct nfs_client *(*get_v3_ds_connect)(
546 struct nfs_client *mds_clp,
547 const struct sockaddr *ds_addr,
548 int ds_addrlen,
549 int ds_proto,
550 unsigned int ds_timeo,
551 unsigned int ds_retrans,
552 rpc_authflavor_t au_flavor);
553
554static bool load_v3_ds_connect(void)
555{
556 if (!get_v3_ds_connect) {
557 get_v3_ds_connect = symbol_request(nfs3_set_ds_client);
558 WARN_ON_ONCE(!get_v3_ds_connect);
559 }
560
561 return(get_v3_ds_connect != NULL);
562}
563
564void __exit nfs4_pnfs_v3_ds_connect_unload(void)
565{
566 if (get_v3_ds_connect) {
567 symbol_put(nfs3_set_ds_client);
568 get_v3_ds_connect = NULL;
569 }
570}
571EXPORT_SYMBOL_GPL(nfs4_pnfs_v3_ds_connect_unload);
572
573static int _nfs4_pnfs_v3_ds_connect(struct nfs_server *mds_srv,
574 struct nfs4_pnfs_ds *ds,
575 unsigned int timeo,
576 unsigned int retrans,
577 rpc_authflavor_t au_flavor)
578{
579 struct nfs_client *clp = ERR_PTR(-EIO);
580 struct nfs4_pnfs_ds_addr *da;
581 int status = 0;
582
583 dprintk("--> %s DS %s au_flavor %d\n", __func__,
584 ds->ds_remotestr, au_flavor);
585
586 if (!load_v3_ds_connect())
587 goto out;
588
589 list_for_each_entry(da, &ds->ds_addrs, da_node) {
590 dprintk("%s: DS %s: trying address %s\n",
591 __func__, ds->ds_remotestr, da->da_remotestr);
592
593 clp = get_v3_ds_connect(mds_srv->nfs_client,
594 (struct sockaddr *)&da->da_addr,
595 da->da_addrlen, IPPROTO_TCP,
596 timeo, retrans, au_flavor);
597 if (!IS_ERR(clp))
598 break;
599 }
600
601 if (IS_ERR(clp)) {
602 status = PTR_ERR(clp);
603 goto out;
604 }
605
606 smp_wmb();
607 ds->ds_clp = clp;
608 dprintk("%s [new] addr: %s\n", __func__, ds->ds_remotestr);
609out:
610 return status;
611}
612
613static int _nfs4_pnfs_v4_ds_connect(struct nfs_server *mds_srv,
614 struct nfs4_pnfs_ds *ds,
615 unsigned int timeo,
616 unsigned int retrans,
617 u32 minor_version,
618 rpc_authflavor_t au_flavor)
619{
620 struct nfs_client *clp = ERR_PTR(-EIO);
621 struct nfs4_pnfs_ds_addr *da;
622 int status = 0;
623
624 dprintk("--> %s DS %s au_flavor %d\n", __func__, ds->ds_remotestr,
625 au_flavor);
626
627 list_for_each_entry(da, &ds->ds_addrs, da_node) {
628 dprintk("%s: DS %s: trying address %s\n",
629 __func__, ds->ds_remotestr, da->da_remotestr);
630
631 clp = nfs4_set_ds_client(mds_srv->nfs_client,
632 (struct sockaddr *)&da->da_addr,
633 da->da_addrlen, IPPROTO_TCP,
634 timeo, retrans, minor_version,
635 au_flavor);
636 if (!IS_ERR(clp))
637 break;
638 }
639
640 if (IS_ERR(clp)) {
641 status = PTR_ERR(clp);
642 goto out;
643 }
644
645 status = nfs4_init_ds_session(clp, mds_srv->nfs_client->cl_lease_time);
646 if (status)
647 goto out_put;
648
649 smp_wmb();
650 ds->ds_clp = clp;
651 dprintk("%s [new] addr: %s\n", __func__, ds->ds_remotestr);
652out:
653 return status;
654out_put:
655 nfs_put_client(clp);
656 goto out;
657}
658
659/*
660 * Create an rpc connection to the nfs4_pnfs_ds data server.
661 * Currently only supports IPv4 and IPv6 addresses.
662 * If connection fails, make devid unavailable.
663 */
664void nfs4_pnfs_ds_connect(struct nfs_server *mds_srv, struct nfs4_pnfs_ds *ds,
665 struct nfs4_deviceid_node *devid, unsigned int timeo,
666 unsigned int retrans, u32 version,
667 u32 minor_version, rpc_authflavor_t au_flavor)
668{
669 if (test_and_set_bit(NFS4DS_CONNECTING, &ds->ds_state) == 0) {
670 int err = 0;
671
672 if (version == 3) {
673 err = _nfs4_pnfs_v3_ds_connect(mds_srv, ds, timeo,
674 retrans, au_flavor);
675 } else if (version == 4) {
676 err = _nfs4_pnfs_v4_ds_connect(mds_srv, ds, timeo,
677 retrans, minor_version,
678 au_flavor);
679 } else {
680 dprintk("%s: unsupported DS version %d\n", __func__,
681 version);
682 err = -EPROTONOSUPPORT;
683 }
684
685 if (err)
686 nfs4_mark_deviceid_unavailable(devid);
687 nfs4_clear_ds_conn_bit(ds);
688 } else {
689 nfs4_wait_ds_connect(ds);
690 }
691}
692EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_connect);
693
694/*
695 * Currently only supports ipv4, ipv6 and one multi-path address.
696 */
697struct nfs4_pnfs_ds_addr *
698nfs4_decode_mp_ds_addr(struct net *net, struct xdr_stream *xdr, gfp_t gfp_flags)
699{
700 struct nfs4_pnfs_ds_addr *da = NULL;
701 char *buf, *portstr;
702 __be16 port;
703 int nlen, rlen;
704 int tmp[2];
705 __be32 *p;
706 char *netid, *match_netid;
707 size_t len, match_netid_len;
708 char *startsep = "";
709 char *endsep = "";
710
711
712 /* r_netid */
713 p = xdr_inline_decode(xdr, 4);
714 if (unlikely(!p))
715 goto out_err;
716 nlen = be32_to_cpup(p++);
717
718 p = xdr_inline_decode(xdr, nlen);
719 if (unlikely(!p))
720 goto out_err;
721
722 netid = kmalloc(nlen+1, gfp_flags);
723 if (unlikely(!netid))
724 goto out_err;
725
726 netid[nlen] = '\0';
727 memcpy(netid, p, nlen);
728
729 /* r_addr: ip/ip6addr with port in dec octets - see RFC 5665 */
730 p = xdr_inline_decode(xdr, 4);
731 if (unlikely(!p))
732 goto out_free_netid;
733 rlen = be32_to_cpup(p);
734
735 p = xdr_inline_decode(xdr, rlen);
736 if (unlikely(!p))
737 goto out_free_netid;
738
739 /* port is ".ABC.DEF", 8 chars max */
740 if (rlen > INET6_ADDRSTRLEN + IPV6_SCOPE_ID_LEN + 8) {
741 dprintk("%s: Invalid address, length %d\n", __func__,
742 rlen);
743 goto out_free_netid;
744 }
745 buf = kmalloc(rlen + 1, gfp_flags);
746 if (!buf) {
747 dprintk("%s: Not enough memory\n", __func__);
748 goto out_free_netid;
749 }
750 buf[rlen] = '\0';
751 memcpy(buf, p, rlen);
752
753 /* replace port '.' with '-' */
754 portstr = strrchr(buf, '.');
755 if (!portstr) {
756 dprintk("%s: Failed finding expected dot in port\n",
757 __func__);
758 goto out_free_buf;
759 }
760 *portstr = '-';
761
762 /* find '.' between address and port */
763 portstr = strrchr(buf, '.');
764 if (!portstr) {
765 dprintk("%s: Failed finding expected dot between address and "
766 "port\n", __func__);
767 goto out_free_buf;
768 }
769 *portstr = '\0';
770
771 da = kzalloc(sizeof(*da), gfp_flags);
772 if (unlikely(!da))
773 goto out_free_buf;
774
775 INIT_LIST_HEAD(&da->da_node);
776
777 if (!rpc_pton(net, buf, portstr-buf, (struct sockaddr *)&da->da_addr,
778 sizeof(da->da_addr))) {
779 dprintk("%s: error parsing address %s\n", __func__, buf);
780 goto out_free_da;
781 }
782
783 portstr++;
784 sscanf(portstr, "%d-%d", &tmp[0], &tmp[1]);
785 port = htons((tmp[0] << 8) | (tmp[1]));
786
787 switch (da->da_addr.ss_family) {
788 case AF_INET:
789 ((struct sockaddr_in *)&da->da_addr)->sin_port = port;
790 da->da_addrlen = sizeof(struct sockaddr_in);
791 match_netid = "tcp";
792 match_netid_len = 3;
793 break;
794
795 case AF_INET6:
796 ((struct sockaddr_in6 *)&da->da_addr)->sin6_port = port;
797 da->da_addrlen = sizeof(struct sockaddr_in6);
798 match_netid = "tcp6";
799 match_netid_len = 4;
800 startsep = "[";
801 endsep = "]";
802 break;
803
804 default:
805 dprintk("%s: unsupported address family: %u\n",
806 __func__, da->da_addr.ss_family);
807 goto out_free_da;
808 }
809
810 if (nlen != match_netid_len || strncmp(netid, match_netid, nlen)) {
811 dprintk("%s: ERROR: r_netid \"%s\" != \"%s\"\n",
812 __func__, netid, match_netid);
813 goto out_free_da;
814 }
815
816 /* save human readable address */
817 len = strlen(startsep) + strlen(buf) + strlen(endsep) + 7;
818 da->da_remotestr = kzalloc(len, gfp_flags);
819
820 /* NULL is ok, only used for dprintk */
821 if (da->da_remotestr)
822 snprintf(da->da_remotestr, len, "%s%s%s:%u", startsep,
823 buf, endsep, ntohs(port));
824
825 dprintk("%s: Parsed DS addr %s\n", __func__, da->da_remotestr);
826 kfree(buf);
827 kfree(netid);
828 return da;
829
830out_free_da:
831 kfree(da);
832out_free_buf:
833 dprintk("%s: Error parsing DS addr: %s\n", __func__, buf);
834 kfree(buf);
835out_free_netid:
836 kfree(netid);
837out_err:
838 return NULL;
839}
840EXPORT_SYMBOL_GPL(nfs4_decode_mp_ds_addr);
diff --git a/fs/nfs/read.c b/fs/nfs/read.c
index c91a4799c562..568ecf0a880f 100644
--- a/fs/nfs/read.c
+++ b/fs/nfs/read.c
@@ -70,8 +70,15 @@ EXPORT_SYMBOL_GPL(nfs_pageio_init_read);
70 70
71void nfs_pageio_reset_read_mds(struct nfs_pageio_descriptor *pgio) 71void nfs_pageio_reset_read_mds(struct nfs_pageio_descriptor *pgio)
72{ 72{
73 struct nfs_pgio_mirror *mirror;
74
73 pgio->pg_ops = &nfs_pgio_rw_ops; 75 pgio->pg_ops = &nfs_pgio_rw_ops;
74 pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->rsize; 76
77 /* read path should never have more than one mirror */
78 WARN_ON_ONCE(pgio->pg_mirror_count != 1);
79
80 mirror = &pgio->pg_mirrors[0];
81 mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->rsize;
75} 82}
76EXPORT_SYMBOL_GPL(nfs_pageio_reset_read_mds); 83EXPORT_SYMBOL_GPL(nfs_pageio_reset_read_mds);
77 84
@@ -81,6 +88,7 @@ int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode,
81 struct nfs_page *new; 88 struct nfs_page *new;
82 unsigned int len; 89 unsigned int len;
83 struct nfs_pageio_descriptor pgio; 90 struct nfs_pageio_descriptor pgio;
91 struct nfs_pgio_mirror *pgm;
84 92
85 len = nfs_page_length(page); 93 len = nfs_page_length(page);
86 if (len == 0) 94 if (len == 0)
@@ -97,7 +105,13 @@ int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode,
97 &nfs_async_read_completion_ops); 105 &nfs_async_read_completion_ops);
98 nfs_pageio_add_request(&pgio, new); 106 nfs_pageio_add_request(&pgio, new);
99 nfs_pageio_complete(&pgio); 107 nfs_pageio_complete(&pgio);
100 NFS_I(inode)->read_io += pgio.pg_bytes_written; 108
109 /* It doesn't make sense to do mirrored reads! */
110 WARN_ON_ONCE(pgio.pg_mirror_count != 1);
111
112 pgm = &pgio.pg_mirrors[0];
113 NFS_I(inode)->read_io += pgm->pg_bytes_written;
114
101 return 0; 115 return 0;
102} 116}
103 117
@@ -168,13 +182,14 @@ out:
168 182
169static void nfs_initiate_read(struct nfs_pgio_header *hdr, 183static void nfs_initiate_read(struct nfs_pgio_header *hdr,
170 struct rpc_message *msg, 184 struct rpc_message *msg,
185 const struct nfs_rpc_ops *rpc_ops,
171 struct rpc_task_setup *task_setup_data, int how) 186 struct rpc_task_setup *task_setup_data, int how)
172{ 187{
173 struct inode *inode = hdr->inode; 188 struct inode *inode = hdr->inode;
174 int swap_flags = IS_SWAPFILE(inode) ? NFS_RPC_SWAPFLAGS : 0; 189 int swap_flags = IS_SWAPFILE(inode) ? NFS_RPC_SWAPFLAGS : 0;
175 190
176 task_setup_data->flags |= swap_flags; 191 task_setup_data->flags |= swap_flags;
177 NFS_PROTO(inode)->read_setup(hdr, msg); 192 rpc_ops->read_setup(hdr, msg);
178} 193}
179 194
180static void 195static void
@@ -351,6 +366,7 @@ int nfs_readpages(struct file *filp, struct address_space *mapping,
351 struct list_head *pages, unsigned nr_pages) 366 struct list_head *pages, unsigned nr_pages)
352{ 367{
353 struct nfs_pageio_descriptor pgio; 368 struct nfs_pageio_descriptor pgio;
369 struct nfs_pgio_mirror *pgm;
354 struct nfs_readdesc desc = { 370 struct nfs_readdesc desc = {
355 .pgio = &pgio, 371 .pgio = &pgio,
356 }; 372 };
@@ -386,10 +402,15 @@ int nfs_readpages(struct file *filp, struct address_space *mapping,
386 &nfs_async_read_completion_ops); 402 &nfs_async_read_completion_ops);
387 403
388 ret = read_cache_pages(mapping, pages, readpage_async_filler, &desc); 404 ret = read_cache_pages(mapping, pages, readpage_async_filler, &desc);
389
390 nfs_pageio_complete(&pgio); 405 nfs_pageio_complete(&pgio);
391 NFS_I(inode)->read_io += pgio.pg_bytes_written; 406
392 npages = (pgio.pg_bytes_written + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 407 /* It doesn't make sense to do mirrored reads! */
408 WARN_ON_ONCE(pgio.pg_mirror_count != 1);
409
410 pgm = &pgio.pg_mirrors[0];
411 NFS_I(inode)->read_io += pgm->pg_bytes_written;
412 npages = (pgm->pg_bytes_written + PAGE_CACHE_SIZE - 1) >>
413 PAGE_CACHE_SHIFT;
393 nfs_add_stats(inode, NFSIOS_READPAGES, npages); 414 nfs_add_stats(inode, NFSIOS_READPAGES, npages);
394read_complete: 415read_complete:
395 put_nfs_open_context(desc.ctx); 416 put_nfs_open_context(desc.ctx);
diff --git a/fs/nfs/write.c b/fs/nfs/write.c
index af3af685a9e3..ceacfeeb28c2 100644
--- a/fs/nfs/write.c
+++ b/fs/nfs/write.c
@@ -473,13 +473,18 @@ try_again:
473 do { 473 do {
474 /* 474 /*
475 * Subrequests are always contiguous, non overlapping 475 * Subrequests are always contiguous, non overlapping
476 * and in order. If not, it's a programming error. 476 * and in order - but may be repeated (mirrored writes).
477 */ 477 */
478 WARN_ON_ONCE(subreq->wb_offset != 478 if (subreq->wb_offset == (head->wb_offset + total_bytes)) {
479 (head->wb_offset + total_bytes)); 479 /* keep track of how many bytes this group covers */
480 480 total_bytes += subreq->wb_bytes;
481 /* keep track of how many bytes this group covers */ 481 } else if (WARN_ON_ONCE(subreq->wb_offset < head->wb_offset ||
482 total_bytes += subreq->wb_bytes; 482 ((subreq->wb_offset + subreq->wb_bytes) >
483 (head->wb_offset + total_bytes)))) {
484 nfs_page_group_unlock(head);
485 spin_unlock(&inode->i_lock);
486 return ERR_PTR(-EIO);
487 }
483 488
484 if (!nfs_lock_request(subreq)) { 489 if (!nfs_lock_request(subreq)) {
485 /* releases page group bit lock and 490 /* releases page group bit lock and
@@ -842,9 +847,9 @@ EXPORT_SYMBOL_GPL(nfs_init_cinfo);
842 */ 847 */
843void 848void
844nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 849nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg,
845 struct nfs_commit_info *cinfo) 850 struct nfs_commit_info *cinfo, u32 ds_commit_idx)
846{ 851{
847 if (pnfs_mark_request_commit(req, lseg, cinfo)) 852 if (pnfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx))
848 return; 853 return;
849 nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo); 854 nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo);
850} 855}
@@ -900,7 +905,8 @@ static void nfs_write_completion(struct nfs_pgio_header *hdr)
900 } 905 }
901 if (nfs_write_need_commit(hdr)) { 906 if (nfs_write_need_commit(hdr)) {
902 memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); 907 memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf));
903 nfs_mark_request_commit(req, hdr->lseg, &cinfo); 908 nfs_mark_request_commit(req, hdr->lseg, &cinfo,
909 hdr->pgio_mirror_idx);
904 goto next; 910 goto next;
905 } 911 }
906remove_req: 912remove_req:
@@ -1240,15 +1246,15 @@ static int flush_task_priority(int how)
1240 1246
1241static void nfs_initiate_write(struct nfs_pgio_header *hdr, 1247static void nfs_initiate_write(struct nfs_pgio_header *hdr,
1242 struct rpc_message *msg, 1248 struct rpc_message *msg,
1249 const struct nfs_rpc_ops *rpc_ops,
1243 struct rpc_task_setup *task_setup_data, int how) 1250 struct rpc_task_setup *task_setup_data, int how)
1244{ 1251{
1245 struct inode *inode = hdr->inode;
1246 int priority = flush_task_priority(how); 1252 int priority = flush_task_priority(how);
1247 1253
1248 task_setup_data->priority = priority; 1254 task_setup_data->priority = priority;
1249 NFS_PROTO(inode)->write_setup(hdr, msg); 1255 rpc_ops->write_setup(hdr, msg);
1250 1256
1251 nfs4_state_protect_write(NFS_SERVER(inode)->nfs_client, 1257 nfs4_state_protect_write(NFS_SERVER(hdr->inode)->nfs_client,
1252 &task_setup_data->rpc_client, msg, hdr); 1258 &task_setup_data->rpc_client, msg, hdr);
1253} 1259}
1254 1260
@@ -1298,8 +1304,14 @@ EXPORT_SYMBOL_GPL(nfs_pageio_init_write);
1298 1304
1299void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) 1305void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio)
1300{ 1306{
1307 struct nfs_pgio_mirror *mirror;
1308
1301 pgio->pg_ops = &nfs_pgio_rw_ops; 1309 pgio->pg_ops = &nfs_pgio_rw_ops;
1302 pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; 1310
1311 nfs_pageio_stop_mirroring(pgio);
1312
1313 mirror = &pgio->pg_mirrors[0];
1314 mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize;
1303} 1315}
1304EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); 1316EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds);
1305 1317
@@ -1465,6 +1477,7 @@ void nfs_commitdata_release(struct nfs_commit_data *data)
1465EXPORT_SYMBOL_GPL(nfs_commitdata_release); 1477EXPORT_SYMBOL_GPL(nfs_commitdata_release);
1466 1478
1467int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, 1479int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data,
1480 const struct nfs_rpc_ops *nfs_ops,
1468 const struct rpc_call_ops *call_ops, 1481 const struct rpc_call_ops *call_ops,
1469 int how, int flags) 1482 int how, int flags)
1470{ 1483{
@@ -1486,7 +1499,7 @@ int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data,
1486 .priority = priority, 1499 .priority = priority,
1487 }; 1500 };
1488 /* Set up the initial task struct. */ 1501 /* Set up the initial task struct. */
1489 NFS_PROTO(data->inode)->commit_setup(data, &msg); 1502 nfs_ops->commit_setup(data, &msg);
1490 1503
1491 dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); 1504 dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid);
1492 1505
@@ -1554,14 +1567,15 @@ EXPORT_SYMBOL_GPL(nfs_init_commit);
1554 1567
1555void nfs_retry_commit(struct list_head *page_list, 1568void nfs_retry_commit(struct list_head *page_list,
1556 struct pnfs_layout_segment *lseg, 1569 struct pnfs_layout_segment *lseg,
1557 struct nfs_commit_info *cinfo) 1570 struct nfs_commit_info *cinfo,
1571 u32 ds_commit_idx)
1558{ 1572{
1559 struct nfs_page *req; 1573 struct nfs_page *req;
1560 1574
1561 while (!list_empty(page_list)) { 1575 while (!list_empty(page_list)) {
1562 req = nfs_list_entry(page_list->next); 1576 req = nfs_list_entry(page_list->next);
1563 nfs_list_remove_request(req); 1577 nfs_list_remove_request(req);
1564 nfs_mark_request_commit(req, lseg, cinfo); 1578 nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx);
1565 if (!cinfo->dreq) { 1579 if (!cinfo->dreq) {
1566 dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); 1580 dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
1567 dec_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info, 1581 dec_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info,
@@ -1589,10 +1603,10 @@ nfs_commit_list(struct inode *inode, struct list_head *head, int how,
1589 /* Set up the argument struct */ 1603 /* Set up the argument struct */
1590 nfs_init_commit(data, head, NULL, cinfo); 1604 nfs_init_commit(data, head, NULL, cinfo);
1591 atomic_inc(&cinfo->mds->rpcs_out); 1605 atomic_inc(&cinfo->mds->rpcs_out);
1592 return nfs_initiate_commit(NFS_CLIENT(inode), data, data->mds_ops, 1606 return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode),
1593 how, 0); 1607 data->mds_ops, how, 0);
1594 out_bad: 1608 out_bad:
1595 nfs_retry_commit(head, NULL, cinfo); 1609 nfs_retry_commit(head, NULL, cinfo, 0);
1596 cinfo->completion_ops->error_cleanup(NFS_I(inode)); 1610 cinfo->completion_ops->error_cleanup(NFS_I(inode));
1597 return -ENOMEM; 1611 return -ENOMEM;
1598} 1612}
diff --git a/include/linux/nfs4.h b/include/linux/nfs4.h
index 022b761dbf0a..de7c91ca427e 100644
--- a/include/linux/nfs4.h
+++ b/include/linux/nfs4.h
@@ -516,6 +516,7 @@ enum pnfs_layouttype {
516 LAYOUT_NFSV4_1_FILES = 1, 516 LAYOUT_NFSV4_1_FILES = 1,
517 LAYOUT_OSD2_OBJECTS = 2, 517 LAYOUT_OSD2_OBJECTS = 2,
518 LAYOUT_BLOCK_VOLUME = 3, 518 LAYOUT_BLOCK_VOLUME = 3,
519 LAYOUT_FLEX_FILES = 4,
519}; 520};
520 521
521/* used for both layout return and recall */ 522/* used for both layout return and recall */
diff --git a/include/linux/nfs_fs_sb.h b/include/linux/nfs_fs_sb.h
index ddea982355f3..5e1273d4de14 100644
--- a/include/linux/nfs_fs_sb.h
+++ b/include/linux/nfs_fs_sb.h
@@ -77,10 +77,6 @@ struct nfs_client {
77 /* Client owner identifier */ 77 /* Client owner identifier */
78 const char * cl_owner_id; 78 const char * cl_owner_id;
79 79
80 /* Our own IP address, as a null-terminated string.
81 * This is used to generate the mv0 callback address.
82 */
83 char cl_ipaddr[48];
84 u32 cl_cb_ident; /* v4.0 callback identifier */ 80 u32 cl_cb_ident; /* v4.0 callback identifier */
85 const struct nfs4_minor_version_ops *cl_mvops; 81 const struct nfs4_minor_version_ops *cl_mvops;
86 unsigned long cl_mig_gen; 82 unsigned long cl_mig_gen;
@@ -108,6 +104,11 @@ struct nfs_client {
108#define NFS_SP4_MACH_CRED_COMMIT 6 /* COMMIT */ 104#define NFS_SP4_MACH_CRED_COMMIT 6 /* COMMIT */
109#endif /* CONFIG_NFS_V4 */ 105#endif /* CONFIG_NFS_V4 */
110 106
107 /* Our own IP address, as a null-terminated string.
108 * This is used to generate the mv0 callback address.
109 */
110 char cl_ipaddr[48];
111
111#ifdef CONFIG_NFS_FSCACHE 112#ifdef CONFIG_NFS_FSCACHE
112 struct fscache_cookie *fscache; /* client index cache cookie */ 113 struct fscache_cookie *fscache; /* client index cache cookie */
113#endif 114#endif
diff --git a/include/linux/nfs_idmap.h b/include/linux/nfs_idmap.h
index 0f4b79da6584..333844e38f66 100644
--- a/include/linux/nfs_idmap.h
+++ b/include/linux/nfs_idmap.h
@@ -73,5 +73,7 @@ int nfs_map_group_to_gid(const struct nfs_server *, const char *, size_t, kgid_t
73int nfs_map_uid_to_name(const struct nfs_server *, kuid_t, char *, size_t); 73int nfs_map_uid_to_name(const struct nfs_server *, kuid_t, char *, size_t);
74int nfs_map_gid_to_group(const struct nfs_server *, kgid_t, char *, size_t); 74int nfs_map_gid_to_group(const struct nfs_server *, kgid_t, char *, size_t);
75 75
76int nfs_map_string_to_numeric(const char *name, size_t namelen, __u32 *res);
77
76extern unsigned int nfs_idmap_cache_timeout; 78extern unsigned int nfs_idmap_cache_timeout;
77#endif /* NFS_IDMAP_H */ 79#endif /* NFS_IDMAP_H */
diff --git a/include/linux/nfs_page.h b/include/linux/nfs_page.h
index 6c3e06ee2fb7..3eb072dbce83 100644
--- a/include/linux/nfs_page.h
+++ b/include/linux/nfs_page.h
@@ -58,6 +58,9 @@ struct nfs_pageio_ops {
58 size_t (*pg_test)(struct nfs_pageio_descriptor *, struct nfs_page *, 58 size_t (*pg_test)(struct nfs_pageio_descriptor *, struct nfs_page *,
59 struct nfs_page *); 59 struct nfs_page *);
60 int (*pg_doio)(struct nfs_pageio_descriptor *); 60 int (*pg_doio)(struct nfs_pageio_descriptor *);
61 unsigned int (*pg_get_mirror_count)(struct nfs_pageio_descriptor *,
62 struct nfs_page *);
63 void (*pg_cleanup)(struct nfs_pageio_descriptor *);
61}; 64};
62 65
63struct nfs_rw_ops { 66struct nfs_rw_ops {
@@ -69,18 +72,21 @@ struct nfs_rw_ops {
69 struct inode *); 72 struct inode *);
70 void (*rw_result)(struct rpc_task *, struct nfs_pgio_header *); 73 void (*rw_result)(struct rpc_task *, struct nfs_pgio_header *);
71 void (*rw_initiate)(struct nfs_pgio_header *, struct rpc_message *, 74 void (*rw_initiate)(struct nfs_pgio_header *, struct rpc_message *,
75 const struct nfs_rpc_ops *,
72 struct rpc_task_setup *, int); 76 struct rpc_task_setup *, int);
73}; 77};
74 78
75struct nfs_pageio_descriptor { 79struct nfs_pgio_mirror {
76 struct list_head pg_list; 80 struct list_head pg_list;
77 unsigned long pg_bytes_written; 81 unsigned long pg_bytes_written;
78 size_t pg_count; 82 size_t pg_count;
79 size_t pg_bsize; 83 size_t pg_bsize;
80 unsigned int pg_base; 84 unsigned int pg_base;
81 unsigned char pg_moreio : 1, 85 unsigned char pg_recoalesce : 1;
82 pg_recoalesce : 1; 86};
83 87
88struct nfs_pageio_descriptor {
89 unsigned char pg_moreio : 1;
84 struct inode *pg_inode; 90 struct inode *pg_inode;
85 const struct nfs_pageio_ops *pg_ops; 91 const struct nfs_pageio_ops *pg_ops;
86 const struct nfs_rw_ops *pg_rw_ops; 92 const struct nfs_rw_ops *pg_rw_ops;
@@ -91,8 +97,18 @@ struct nfs_pageio_descriptor {
91 struct pnfs_layout_segment *pg_lseg; 97 struct pnfs_layout_segment *pg_lseg;
92 struct nfs_direct_req *pg_dreq; 98 struct nfs_direct_req *pg_dreq;
93 void *pg_layout_private; 99 void *pg_layout_private;
100 unsigned int pg_bsize; /* default bsize for mirrors */
101
102 u32 pg_mirror_count;
103 struct nfs_pgio_mirror *pg_mirrors;
104 struct nfs_pgio_mirror pg_mirrors_static[1];
105 struct nfs_pgio_mirror *pg_mirrors_dynamic;
106 u32 pg_mirror_idx; /* current mirror */
94}; 107};
95 108
109/* arbitrarily selected limit to number of mirrors */
110#define NFS_PAGEIO_DESCRIPTOR_MIRROR_MAX 16
111
96#define NFS_WBACK_BUSY(req) (test_bit(PG_BUSY,&(req)->wb_flags)) 112#define NFS_WBACK_BUSY(req) (test_bit(PG_BUSY,&(req)->wb_flags))
97 113
98extern struct nfs_page *nfs_create_request(struct nfs_open_context *ctx, 114extern struct nfs_page *nfs_create_request(struct nfs_open_context *ctx,
diff --git a/include/linux/nfs_xdr.h b/include/linux/nfs_xdr.h
index e5c3b620a609..81401125ab2d 100644
--- a/include/linux/nfs_xdr.h
+++ b/include/linux/nfs_xdr.h
@@ -293,6 +293,7 @@ struct nfs4_layoutreturn_args {
293 struct nfs4_sequence_args seq_args; 293 struct nfs4_sequence_args seq_args;
294 struct pnfs_layout_hdr *layout; 294 struct pnfs_layout_hdr *layout;
295 struct inode *inode; 295 struct inode *inode;
296 struct pnfs_layout_range range;
296 nfs4_stateid stateid; 297 nfs4_stateid stateid;
297 __u32 layout_type; 298 __u32 layout_type;
298}; 299};
@@ -514,6 +515,7 @@ struct nfs_pgio_res {
514 struct nfs4_sequence_res seq_res; 515 struct nfs4_sequence_res seq_res;
515 struct nfs_fattr * fattr; 516 struct nfs_fattr * fattr;
516 __u32 count; 517 __u32 count;
518 __u32 op_status;
517 int eof; /* used by read */ 519 int eof; /* used by read */
518 struct nfs_writeverf * verf; /* used by write */ 520 struct nfs_writeverf * verf; /* used by write */
519 const struct nfs_server *server; /* used by write */ 521 const struct nfs_server *server; /* used by write */
@@ -533,6 +535,7 @@ struct nfs_commitargs {
533 535
534struct nfs_commitres { 536struct nfs_commitres {
535 struct nfs4_sequence_res seq_res; 537 struct nfs4_sequence_res seq_res;
538 __u32 op_status;
536 struct nfs_fattr *fattr; 539 struct nfs_fattr *fattr;
537 struct nfs_writeverf *verf; 540 struct nfs_writeverf *verf;
538 const struct nfs_server *server; 541 const struct nfs_server *server;
@@ -1326,7 +1329,8 @@ struct nfs_pgio_header {
1326 __u64 mds_offset; /* Filelayout dense stripe */ 1329 __u64 mds_offset; /* Filelayout dense stripe */
1327 struct nfs_page_array page_array; 1330 struct nfs_page_array page_array;
1328 struct nfs_client *ds_clp; /* pNFS data server */ 1331 struct nfs_client *ds_clp; /* pNFS data server */
1329 int ds_idx; /* ds index if ds_clp is set */ 1332 int ds_commit_idx; /* ds index if ds_clp is set */
1333 int pgio_mirror_idx;/* mirror index in pgio layer */
1330}; 1334};
1331 1335
1332struct nfs_mds_commit_info { 1336struct nfs_mds_commit_info {
diff --git a/include/linux/sunrpc/metrics.h b/include/linux/sunrpc/metrics.h
index eecb5a71e6c0..7e61a17030a4 100644
--- a/include/linux/sunrpc/metrics.h
+++ b/include/linux/sunrpc/metrics.h
@@ -79,6 +79,8 @@ struct rpc_clnt;
79struct rpc_iostats * rpc_alloc_iostats(struct rpc_clnt *); 79struct rpc_iostats * rpc_alloc_iostats(struct rpc_clnt *);
80void rpc_count_iostats(const struct rpc_task *, 80void rpc_count_iostats(const struct rpc_task *,
81 struct rpc_iostats *); 81 struct rpc_iostats *);
82void rpc_count_iostats_metrics(const struct rpc_task *,
83 struct rpc_iostats *);
82void rpc_print_iostats(struct seq_file *, struct rpc_clnt *); 84void rpc_print_iostats(struct seq_file *, struct rpc_clnt *);
83void rpc_free_iostats(struct rpc_iostats *); 85void rpc_free_iostats(struct rpc_iostats *);
84 86
@@ -87,6 +89,8 @@ void rpc_free_iostats(struct rpc_iostats *);
87static inline struct rpc_iostats *rpc_alloc_iostats(struct rpc_clnt *clnt) { return NULL; } 89static inline struct rpc_iostats *rpc_alloc_iostats(struct rpc_clnt *clnt) { return NULL; }
88static inline void rpc_count_iostats(const struct rpc_task *task, 90static inline void rpc_count_iostats(const struct rpc_task *task,
89 struct rpc_iostats *stats) {} 91 struct rpc_iostats *stats) {}
92static inline void rpc_count_iostats_metrics(const struct rpc_task *,
93 struct rpc_iostats *) {}
90static inline void rpc_print_iostats(struct seq_file *seq, struct rpc_clnt *clnt) {} 94static inline void rpc_print_iostats(struct seq_file *seq, struct rpc_clnt *clnt) {}
91static inline void rpc_free_iostats(struct rpc_iostats *stats) {} 95static inline void rpc_free_iostats(struct rpc_iostats *stats) {}
92 96
diff --git a/net/sunrpc/stats.c b/net/sunrpc/stats.c
index 9711a155bc50..2ecb994314c1 100644
--- a/net/sunrpc/stats.c
+++ b/net/sunrpc/stats.c
@@ -140,22 +140,20 @@ void rpc_free_iostats(struct rpc_iostats *stats)
140EXPORT_SYMBOL_GPL(rpc_free_iostats); 140EXPORT_SYMBOL_GPL(rpc_free_iostats);
141 141
142/** 142/**
143 * rpc_count_iostats - tally up per-task stats 143 * rpc_count_iostats_metrics - tally up per-task stats
144 * @task: completed rpc_task 144 * @task: completed rpc_task
145 * @stats: array of stat structures 145 * @op_metrics: stat structure for OP that will accumulate stats from @task
146 */ 146 */
147void rpc_count_iostats(const struct rpc_task *task, struct rpc_iostats *stats) 147void rpc_count_iostats_metrics(const struct rpc_task *task,
148 struct rpc_iostats *op_metrics)
148{ 149{
149 struct rpc_rqst *req = task->tk_rqstp; 150 struct rpc_rqst *req = task->tk_rqstp;
150 struct rpc_iostats *op_metrics;
151 ktime_t delta, now; 151 ktime_t delta, now;
152 152
153 if (!stats || !req) 153 if (!op_metrics || !req)
154 return; 154 return;
155 155
156 now = ktime_get(); 156 now = ktime_get();
157 op_metrics = &stats[task->tk_msg.rpc_proc->p_statidx];
158
159 spin_lock(&op_metrics->om_lock); 157 spin_lock(&op_metrics->om_lock);
160 158
161 op_metrics->om_ops++; 159 op_metrics->om_ops++;
@@ -175,6 +173,20 @@ void rpc_count_iostats(const struct rpc_task *task, struct rpc_iostats *stats)
175 173
176 spin_unlock(&op_metrics->om_lock); 174 spin_unlock(&op_metrics->om_lock);
177} 175}
176EXPORT_SYMBOL_GPL(rpc_count_iostats_metrics);
177
178/**
179 * rpc_count_iostats - tally up per-task stats
180 * @task: completed rpc_task
181 * @stats: array of stat structures
182 *
183 * Uses the statidx from @task
184 */
185void rpc_count_iostats(const struct rpc_task *task, struct rpc_iostats *stats)
186{
187 rpc_count_iostats_metrics(task,
188 &stats[task->tk_msg.rpc_proc->p_statidx]);
189}
178EXPORT_SYMBOL_GPL(rpc_count_iostats); 190EXPORT_SYMBOL_GPL(rpc_count_iostats);
179 191
180static void _print_name(struct seq_file *seq, unsigned int op, 192static void _print_name(struct seq_file *seq, unsigned int op,
generated by cgit v1.2.2 (git 2.25.0) at 2025-02-19 22:23:39 -0500