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_rcu(mirror_ds, id_node.rcu);
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			if (cmpxchg(&mirror->cred, NULL, cred))
328				put_rpccred(cred);
329		}
330	}
331	return 0;
332}
333
334struct nfs_fh *
335nfs4_ff_layout_select_ds_fh(struct pnfs_layout_segment *lseg, u32 mirror_idx)
336{
337	struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, mirror_idx);
338	struct nfs_fh *fh = NULL;
339	struct nfs4_deviceid_node *devid;
340
341	if (mirror == NULL || mirror->mirror_ds == NULL ||
342	    mirror->mirror_ds->ds == NULL) {
343		printk(KERN_ERR "NFS: %s: No data server for mirror offset index %d\n",
344			__func__, mirror_idx);
345		if (mirror && mirror->mirror_ds) {
346			devid = &mirror->mirror_ds->id_node;
347			pnfs_generic_mark_devid_invalid(devid);
348		}
349		goto out;
350	}
351
352	/* FIXME: For now assume there is only 1 version available for the DS */
353	fh = &mirror->fh_versions[0];
354out:
355	return fh;
356}
357
358/* Upon return, either ds is connected, or ds is NULL */
359struct nfs4_pnfs_ds *
360nfs4_ff_layout_prepare_ds(struct pnfs_layout_segment *lseg, u32 ds_idx,
361			  bool fail_return)
362{
363	struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, ds_idx);
364	struct nfs4_pnfs_ds *ds = NULL;
365	struct nfs4_deviceid_node *devid;
366	struct inode *ino = lseg->pls_layout->plh_inode;
367	struct nfs_server *s = NFS_SERVER(ino);
368	unsigned int max_payload;
369	rpc_authflavor_t flavor;
370
371	if (mirror == NULL || mirror->mirror_ds == NULL ||
372	    mirror->mirror_ds->ds == NULL) {
373		printk(KERN_ERR "NFS: %s: No data server for offset index %d\n",
374			__func__, ds_idx);
375		if (mirror && mirror->mirror_ds) {
376			devid = &mirror->mirror_ds->id_node;
377			pnfs_generic_mark_devid_invalid(devid);
378		}
379		goto out;
380	}
381
382	devid = &mirror->mirror_ds->id_node;
383	if (ff_layout_test_devid_unavailable(devid))
384		goto out;
385
386	ds = mirror->mirror_ds->ds;
387	/* matching smp_wmb() in _nfs4_pnfs_v3/4_ds_connect */
388	smp_rmb();
389	if (ds->ds_clp)
390		goto out_update_creds;
391
392	flavor = nfs4_ff_layout_choose_authflavor(mirror);
393
394	/* FIXME: For now we assume the server sent only one version of NFS
395	 * to use for the DS.
396	 */
397	nfs4_pnfs_ds_connect(s, ds, devid, dataserver_timeo,
398			     dataserver_retrans,
399			     mirror->mirror_ds->ds_versions[0].version,
400			     mirror->mirror_ds->ds_versions[0].minor_version,
401			     flavor);
402
403	/* connect success, check rsize/wsize limit */
404	if (ds->ds_clp) {
405		max_payload =
406			nfs_block_size(rpc_max_payload(ds->ds_clp->cl_rpcclient),
407				       NULL);
408		if (mirror->mirror_ds->ds_versions[0].rsize > max_payload)
409			mirror->mirror_ds->ds_versions[0].rsize = max_payload;
410		if (mirror->mirror_ds->ds_versions[0].wsize > max_payload)
411			mirror->mirror_ds->ds_versions[0].wsize = max_payload;
412	} else {
413		ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
414					 mirror, lseg->pls_range.offset,
415					 lseg->pls_range.length, NFS4ERR_NXIO,
416					 OP_ILLEGAL, GFP_NOIO);
417		if (fail_return) {
418			pnfs_error_mark_layout_for_return(ino, lseg);
419			if (ff_layout_has_available_ds(lseg))
420				pnfs_set_retry_layoutget(lseg->pls_layout);
421			else
422				pnfs_clear_retry_layoutget(lseg->pls_layout);
423
424		} else {
425			if (ff_layout_has_available_ds(lseg))
426				set_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE,
427					&lseg->pls_layout->plh_flags);
428			else {
429				pnfs_error_mark_layout_for_return(ino, lseg);
430				pnfs_clear_retry_layoutget(lseg->pls_layout);
431			}
432		}
433	}
434out_update_creds:
435	if (ff_layout_update_mirror_cred(mirror, ds))
436		ds = NULL;
437out:
438	return ds;
439}
440
441struct rpc_cred *
442ff_layout_get_ds_cred(struct pnfs_layout_segment *lseg, u32 ds_idx,
443		      struct rpc_cred *mdscred)
444{
445	struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, ds_idx);
446	struct rpc_cred *cred = ERR_PTR(-EINVAL);
447
448	if (!nfs4_ff_layout_prepare_ds(lseg, ds_idx, true))
449		goto out;
450
451	if (mirror && mirror->cred)
452		cred = mirror->cred;
453	else
454		cred = mdscred;
455out:
456	return cred;
457}
458
459/**
460* Find or create a DS rpc client with th MDS server rpc client auth flavor
461* in the nfs_client cl_ds_clients list.
462*/
463struct rpc_clnt *
464nfs4_ff_find_or_create_ds_client(struct pnfs_layout_segment *lseg, u32 ds_idx,
465				 struct nfs_client *ds_clp, struct inode *inode)
466{
467	struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, ds_idx);
468
469	switch (mirror->mirror_ds->ds_versions[0].version) {
470	case 3:
471		/* For NFSv3 DS, flavor is set when creating DS connections */
472		return ds_clp->cl_rpcclient;
473	case 4:
474		return nfs4_find_or_create_ds_client(ds_clp, inode);
475	default:
476		BUG();
477	}
478}
479
480static bool is_range_intersecting(u64 offset1, u64 length1,
481				  u64 offset2, u64 length2)
482{
483	u64 end1 = end_offset(offset1, length1);
484	u64 end2 = end_offset(offset2, length2);
485
486	return (end1 == NFS4_MAX_UINT64 || end1 > offset2) &&
487	       (end2 == NFS4_MAX_UINT64 || end2 > offset1);
488}
489
490/* called with inode i_lock held */
491int ff_layout_encode_ds_ioerr(struct nfs4_flexfile_layout *flo,
492			      struct xdr_stream *xdr, int *count,
493			      const struct pnfs_layout_range *range)
494{
495	struct nfs4_ff_layout_ds_err *err, *n;
496	__be32 *p;
497
498	list_for_each_entry_safe(err, n, &flo->error_list, list) {
499		if (!is_range_intersecting(err->offset, err->length,
500					   range->offset, range->length))
501			continue;
502		/* offset(8) + length(8) + stateid(NFS4_STATEID_SIZE)
503		 * + array length + deviceid(NFS4_DEVICEID4_SIZE)
504		 * + status(4) + opnum(4)
505		 */
506		p = xdr_reserve_space(xdr,
507				28 + NFS4_STATEID_SIZE + NFS4_DEVICEID4_SIZE);
508		if (unlikely(!p))
509			return -ENOBUFS;
510		p = xdr_encode_hyper(p, err->offset);
511		p = xdr_encode_hyper(p, err->length);
512		p = xdr_encode_opaque_fixed(p, &err->stateid,
513					    NFS4_STATEID_SIZE);
514		/* Encode 1 error */
515		*p++ = cpu_to_be32(1);
516		p = xdr_encode_opaque_fixed(p, &err->deviceid,
517					    NFS4_DEVICEID4_SIZE);
518		*p++ = cpu_to_be32(err->status);
519		*p++ = cpu_to_be32(err->opnum);
520		*count += 1;
521		list_del(&err->list);
522		dprintk("%s: offset %llu length %llu status %d op %d count %d\n",
523			__func__, err->offset, err->length, err->status,
524			err->opnum, *count);
525		kfree(err);
526	}
527
528	return 0;
529}
530
531bool ff_layout_has_available_ds(struct pnfs_layout_segment *lseg)
532{
533	struct nfs4_ff_layout_mirror *mirror;
534	struct nfs4_deviceid_node *devid;
535	int idx;
536
537	for (idx = 0; idx < FF_LAYOUT_MIRROR_COUNT(lseg); idx++) {
538		mirror = FF_LAYOUT_COMP(lseg, idx);
539		if (mirror && mirror->mirror_ds) {
540			devid = &mirror->mirror_ds->id_node;
541			if (!ff_layout_test_devid_unavailable(devid))
542				return true;
543		}
544	}
545
546	return false;
547}
548
549module_param(dataserver_retrans, uint, 0644);
550MODULE_PARM_DESC(dataserver_retrans, "The  number of times the NFSv4.1 client "
551			"retries a request before it attempts further "
552			" recovery  action.");
553module_param(dataserver_timeo, uint, 0644);
554MODULE_PARM_DESC(dataserver_timeo, "The time (in tenths of a second) the "
555			"NFSv4.1  client  waits for a response from a "
556			" data server before it retries an NFS request.");
557