xref: /linux/fs/nfs/flexfilelayout/flexfilelayoutdev.c (revision 0883c2c06fb5bcf5b9e008270827e63c09a88c1e)
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 
20 static unsigned int dataserver_timeo = NFS4_DEF_DS_TIMEO;
21 static unsigned int dataserver_retrans = NFS4_DEF_DS_RETRANS;
22 
23 void 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 
29 void 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 */
37 struct nfs4_ff_layout_ds *
38 nfs4_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 
156 out_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);
166 out_scratch:
167 	__free_page(scratch);
168 out_err:
169 	kfree(new_ds);
170 
171 	dprintk("%s ERROR: returning %d\n", __func__, ret);
172 	return NULL;
173 }
174 
175 static void ff_layout_mark_devid_invalid(struct pnfs_layout_segment *lseg,
176 		struct nfs4_deviceid_node *devid)
177 {
178 	nfs4_mark_deviceid_unavailable(devid);
179 	if (!ff_layout_has_available_ds(lseg))
180 		pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode,
181 				lseg);
182 }
183 
184 static bool ff_layout_mirror_valid(struct pnfs_layout_segment *lseg,
185 		struct nfs4_ff_layout_mirror *mirror)
186 {
187 	if (mirror == NULL || mirror->mirror_ds == NULL) {
188 		pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode,
189 					lseg);
190 		return false;
191 	}
192 	if (mirror->mirror_ds->ds == NULL) {
193 		struct nfs4_deviceid_node *devid;
194 		devid = &mirror->mirror_ds->id_node;
195 		ff_layout_mark_devid_invalid(lseg, devid);
196 		return false;
197 	}
198 	return true;
199 }
200 
201 static u64
202 end_offset(u64 start, u64 len)
203 {
204 	u64 end;
205 
206 	end = start + len;
207 	return end >= start ? end : NFS4_MAX_UINT64;
208 }
209 
210 static void extend_ds_error(struct nfs4_ff_layout_ds_err *err,
211 			    u64 offset, u64 length)
212 {
213 	u64 end;
214 
215 	end = max_t(u64, end_offset(err->offset, err->length),
216 		    end_offset(offset, length));
217 	err->offset = min_t(u64, err->offset, offset);
218 	err->length = end - err->offset;
219 }
220 
221 static int
222 ff_ds_error_match(const struct nfs4_ff_layout_ds_err *e1,
223 		const struct nfs4_ff_layout_ds_err *e2)
224 {
225 	int ret;
226 
227 	if (e1->opnum != e2->opnum)
228 		return e1->opnum < e2->opnum ? -1 : 1;
229 	if (e1->status != e2->status)
230 		return e1->status < e2->status ? -1 : 1;
231 	ret = memcmp(e1->stateid.data, e2->stateid.data,
232 			sizeof(e1->stateid.data));
233 	if (ret != 0)
234 		return ret;
235 	ret = memcmp(&e1->deviceid, &e2->deviceid, sizeof(e1->deviceid));
236 	if (ret != 0)
237 		return ret;
238 	if (end_offset(e1->offset, e1->length) < e2->offset)
239 		return -1;
240 	if (e1->offset > end_offset(e2->offset, e2->length))
241 		return 1;
242 	/* If ranges overlap or are contiguous, they are the same */
243 	return 0;
244 }
245 
246 static void
247 ff_layout_add_ds_error_locked(struct nfs4_flexfile_layout *flo,
248 			      struct nfs4_ff_layout_ds_err *dserr)
249 {
250 	struct nfs4_ff_layout_ds_err *err, *tmp;
251 	struct list_head *head = &flo->error_list;
252 	int match;
253 
254 	/* Do insertion sort w/ merges */
255 	list_for_each_entry_safe(err, tmp, &flo->error_list, list) {
256 		match = ff_ds_error_match(err, dserr);
257 		if (match < 0)
258 			continue;
259 		if (match > 0) {
260 			/* Add entry "dserr" _before_ entry "err" */
261 			head = &err->list;
262 			break;
263 		}
264 		/* Entries match, so merge "err" into "dserr" */
265 		extend_ds_error(dserr, err->offset, err->length);
266 		list_del(&err->list);
267 		kfree(err);
268 	}
269 
270 	list_add_tail(&dserr->list, head);
271 }
272 
273 int ff_layout_track_ds_error(struct nfs4_flexfile_layout *flo,
274 			     struct nfs4_ff_layout_mirror *mirror, u64 offset,
275 			     u64 length, int status, enum nfs_opnum4 opnum,
276 			     gfp_t gfp_flags)
277 {
278 	struct nfs4_ff_layout_ds_err *dserr;
279 
280 	if (status == 0)
281 		return 0;
282 
283 	if (mirror->mirror_ds == NULL)
284 		return -EINVAL;
285 
286 	dserr = kmalloc(sizeof(*dserr), gfp_flags);
287 	if (!dserr)
288 		return -ENOMEM;
289 
290 	INIT_LIST_HEAD(&dserr->list);
291 	dserr->offset = offset;
292 	dserr->length = length;
293 	dserr->status = status;
294 	dserr->opnum = opnum;
295 	nfs4_stateid_copy(&dserr->stateid, &mirror->stateid);
296 	memcpy(&dserr->deviceid, &mirror->mirror_ds->id_node.deviceid,
297 	       NFS4_DEVICEID4_SIZE);
298 
299 	spin_lock(&flo->generic_hdr.plh_inode->i_lock);
300 	ff_layout_add_ds_error_locked(flo, dserr);
301 	spin_unlock(&flo->generic_hdr.plh_inode->i_lock);
302 
303 	return 0;
304 }
305 
306 static struct rpc_cred *
307 ff_layout_get_mirror_cred(struct nfs4_ff_layout_mirror *mirror, u32 iomode)
308 {
309 	struct rpc_cred *cred, __rcu **pcred;
310 
311 	if (iomode == IOMODE_READ)
312 		pcred = &mirror->ro_cred;
313 	else
314 		pcred = &mirror->rw_cred;
315 
316 	rcu_read_lock();
317 	do {
318 		cred = rcu_dereference(*pcred);
319 		if (!cred)
320 			break;
321 
322 		cred = get_rpccred_rcu(cred);
323 	} while(!cred);
324 	rcu_read_unlock();
325 	return cred;
326 }
327 
328 struct nfs_fh *
329 nfs4_ff_layout_select_ds_fh(struct pnfs_layout_segment *lseg, u32 mirror_idx)
330 {
331 	struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, mirror_idx);
332 	struct nfs_fh *fh = NULL;
333 
334 	if (!ff_layout_mirror_valid(lseg, mirror)) {
335 		pr_err_ratelimited("NFS: %s: No data server for mirror offset index %d\n",
336 			__func__, mirror_idx);
337 		goto out;
338 	}
339 
340 	/* FIXME: For now assume there is only 1 version available for the DS */
341 	fh = &mirror->fh_versions[0];
342 out:
343 	return fh;
344 }
345 
346 /**
347  * nfs4_ff_layout_prepare_ds - prepare a DS connection for an RPC call
348  * @lseg: the layout segment we're operating on
349  * @ds_idx: index of the DS to use
350  * @fail_return: return layout on connect failure?
351  *
352  * Try to prepare a DS connection to accept an RPC call. This involves
353  * selecting a mirror to use and connecting the client to it if it's not
354  * already connected.
355  *
356  * Since we only need a single functioning mirror to satisfy a read, we don't
357  * want to return the layout if there is one. For writes though, any down
358  * mirror should result in a LAYOUTRETURN. @fail_return is how we distinguish
359  * between the two cases.
360  *
361  * Returns a pointer to a connected DS object on success or NULL on failure.
362  */
363 struct nfs4_pnfs_ds *
364 nfs4_ff_layout_prepare_ds(struct pnfs_layout_segment *lseg, u32 ds_idx,
365 			  bool fail_return)
366 {
367 	struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, ds_idx);
368 	struct nfs4_pnfs_ds *ds = NULL;
369 	struct nfs4_deviceid_node *devid;
370 	struct inode *ino = lseg->pls_layout->plh_inode;
371 	struct nfs_server *s = NFS_SERVER(ino);
372 	unsigned int max_payload;
373 
374 	if (!ff_layout_mirror_valid(lseg, mirror)) {
375 		pr_err_ratelimited("NFS: %s: No data server for offset index %d\n",
376 			__func__, ds_idx);
377 		goto out;
378 	}
379 
380 	devid = &mirror->mirror_ds->id_node;
381 	if (ff_layout_test_devid_unavailable(devid))
382 		goto out;
383 
384 	ds = mirror->mirror_ds->ds;
385 	/* matching smp_wmb() in _nfs4_pnfs_v3/4_ds_connect */
386 	smp_rmb();
387 	if (ds->ds_clp)
388 		goto out;
389 
390 	/* FIXME: For now we assume the server sent only one version of NFS
391 	 * to use for the DS.
392 	 */
393 	nfs4_pnfs_ds_connect(s, ds, devid, dataserver_timeo,
394 			     dataserver_retrans,
395 			     mirror->mirror_ds->ds_versions[0].version,
396 			     mirror->mirror_ds->ds_versions[0].minor_version,
397 			     RPC_AUTH_UNIX);
398 
399 	/* connect success, check rsize/wsize limit */
400 	if (ds->ds_clp) {
401 		max_payload =
402 			nfs_block_size(rpc_max_payload(ds->ds_clp->cl_rpcclient),
403 				       NULL);
404 		if (mirror->mirror_ds->ds_versions[0].rsize > max_payload)
405 			mirror->mirror_ds->ds_versions[0].rsize = max_payload;
406 		if (mirror->mirror_ds->ds_versions[0].wsize > max_payload)
407 			mirror->mirror_ds->ds_versions[0].wsize = max_payload;
408 	} else {
409 		ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
410 					 mirror, lseg->pls_range.offset,
411 					 lseg->pls_range.length, NFS4ERR_NXIO,
412 					 OP_ILLEGAL, GFP_NOIO);
413 		if (fail_return || !ff_layout_has_available_ds(lseg))
414 			pnfs_error_mark_layout_for_return(ino, lseg);
415 		ds = NULL;
416 	}
417 out:
418 	return ds;
419 }
420 
421 struct rpc_cred *
422 ff_layout_get_ds_cred(struct pnfs_layout_segment *lseg, u32 ds_idx,
423 		      struct rpc_cred *mdscred)
424 {
425 	struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, ds_idx);
426 	struct rpc_cred *cred;
427 
428 	if (mirror) {
429 		cred = ff_layout_get_mirror_cred(mirror, lseg->pls_range.iomode);
430 		if (!cred)
431 			cred = get_rpccred(mdscred);
432 	} else {
433 		cred = get_rpccred(mdscred);
434 	}
435 	return cred;
436 }
437 
438 /**
439 * Find or create a DS rpc client with th MDS server rpc client auth flavor
440 * in the nfs_client cl_ds_clients list.
441 */
442 struct rpc_clnt *
443 nfs4_ff_find_or_create_ds_client(struct pnfs_layout_segment *lseg, u32 ds_idx,
444 				 struct nfs_client *ds_clp, struct inode *inode)
445 {
446 	struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, ds_idx);
447 
448 	switch (mirror->mirror_ds->ds_versions[0].version) {
449 	case 3:
450 		/* For NFSv3 DS, flavor is set when creating DS connections */
451 		return ds_clp->cl_rpcclient;
452 	case 4:
453 		return nfs4_find_or_create_ds_client(ds_clp, inode);
454 	default:
455 		BUG();
456 	}
457 }
458 
459 static bool is_range_intersecting(u64 offset1, u64 length1,
460 				  u64 offset2, u64 length2)
461 {
462 	u64 end1 = end_offset(offset1, length1);
463 	u64 end2 = end_offset(offset2, length2);
464 
465 	return (end1 == NFS4_MAX_UINT64 || end1 > offset2) &&
466 	       (end2 == NFS4_MAX_UINT64 || end2 > offset1);
467 }
468 
469 /* called with inode i_lock held */
470 int ff_layout_encode_ds_ioerr(struct nfs4_flexfile_layout *flo,
471 			      struct xdr_stream *xdr, int *count,
472 			      const struct pnfs_layout_range *range)
473 {
474 	struct nfs4_ff_layout_ds_err *err, *n;
475 	__be32 *p;
476 
477 	list_for_each_entry_safe(err, n, &flo->error_list, list) {
478 		if (!is_range_intersecting(err->offset, err->length,
479 					   range->offset, range->length))
480 			continue;
481 		/* offset(8) + length(8) + stateid(NFS4_STATEID_SIZE)
482 		 * + array length + deviceid(NFS4_DEVICEID4_SIZE)
483 		 * + status(4) + opnum(4)
484 		 */
485 		p = xdr_reserve_space(xdr,
486 				28 + NFS4_STATEID_SIZE + NFS4_DEVICEID4_SIZE);
487 		if (unlikely(!p))
488 			return -ENOBUFS;
489 		p = xdr_encode_hyper(p, err->offset);
490 		p = xdr_encode_hyper(p, err->length);
491 		p = xdr_encode_opaque_fixed(p, &err->stateid,
492 					    NFS4_STATEID_SIZE);
493 		/* Encode 1 error */
494 		*p++ = cpu_to_be32(1);
495 		p = xdr_encode_opaque_fixed(p, &err->deviceid,
496 					    NFS4_DEVICEID4_SIZE);
497 		*p++ = cpu_to_be32(err->status);
498 		*p++ = cpu_to_be32(err->opnum);
499 		*count += 1;
500 		list_del(&err->list);
501 		dprintk("%s: offset %llu length %llu status %d op %d count %d\n",
502 			__func__, err->offset, err->length, err->status,
503 			err->opnum, *count);
504 		kfree(err);
505 	}
506 
507 	return 0;
508 }
509 
510 static bool ff_read_layout_has_available_ds(struct pnfs_layout_segment *lseg)
511 {
512 	struct nfs4_ff_layout_mirror *mirror;
513 	struct nfs4_deviceid_node *devid;
514 	u32 idx;
515 
516 	for (idx = 0; idx < FF_LAYOUT_MIRROR_COUNT(lseg); idx++) {
517 		mirror = FF_LAYOUT_COMP(lseg, idx);
518 		if (mirror && mirror->mirror_ds) {
519 			devid = &mirror->mirror_ds->id_node;
520 			if (!ff_layout_test_devid_unavailable(devid))
521 				return true;
522 		}
523 	}
524 
525 	return false;
526 }
527 
528 static bool ff_rw_layout_has_available_ds(struct pnfs_layout_segment *lseg)
529 {
530 	struct nfs4_ff_layout_mirror *mirror;
531 	struct nfs4_deviceid_node *devid;
532 	u32 idx;
533 
534 	for (idx = 0; idx < FF_LAYOUT_MIRROR_COUNT(lseg); idx++) {
535 		mirror = FF_LAYOUT_COMP(lseg, idx);
536 		if (!mirror || !mirror->mirror_ds)
537 			return false;
538 		devid = &mirror->mirror_ds->id_node;
539 		if (ff_layout_test_devid_unavailable(devid))
540 			return false;
541 	}
542 
543 	return FF_LAYOUT_MIRROR_COUNT(lseg) != 0;
544 }
545 
546 bool ff_layout_has_available_ds(struct pnfs_layout_segment *lseg)
547 {
548 	if (lseg->pls_range.iomode == IOMODE_READ)
549 		return  ff_read_layout_has_available_ds(lseg);
550 	/* Note: RW layout needs all mirrors available */
551 	return ff_rw_layout_has_available_ds(lseg);
552 }
553 
554 bool ff_layout_avoid_mds_available_ds(struct pnfs_layout_segment *lseg)
555 {
556 	return ff_layout_no_fallback_to_mds(lseg) ||
557 	       ff_layout_has_available_ds(lseg);
558 }
559 
560 bool ff_layout_avoid_read_on_rw(struct pnfs_layout_segment *lseg)
561 {
562 	return lseg->pls_range.iomode == IOMODE_RW &&
563 	       ff_layout_no_read_on_rw(lseg);
564 }
565 
566 module_param(dataserver_retrans, uint, 0644);
567 MODULE_PARM_DESC(dataserver_retrans, "The  number of times the NFSv4.1 client "
568 			"retries a request before it attempts further "
569 			" recovery  action.");
570 module_param(dataserver_timeo, uint, 0644);
571 MODULE_PARM_DESC(dataserver_timeo, "The time (in tenths of a second) the "
572 			"NFSv4.1  client  waits for a response from a "
573 			" data server before it retries an NFS request.");
574