1/*
2 *  Device operations for the pnfs client.
3 *
4 *  Copyright (c) 2002
5 *  The Regents of the University of Michigan
6 *  All Rights Reserved
7 *
8 *  Dean Hildebrand <dhildebz@umich.edu>
9 *  Garth Goodson   <Garth.Goodson@netapp.com>
10 *
11 *  Permission is granted to use, copy, create derivative works, and
12 *  redistribute this software and such derivative works for any purpose,
13 *  so long as the name of the University of Michigan is not used in
14 *  any advertising or publicity pertaining to the use or distribution
15 *  of this software without specific, written prior authorization. If
16 *  the above copyright notice or any other identification of the
17 *  University of Michigan is included in any copy of any portion of
18 *  this software, then the disclaimer below must also be included.
19 *
20 *  This software is provided as is, without representation or warranty
21 *  of any kind either express or implied, including without limitation
22 *  the implied warranties of merchantability, fitness for a particular
23 *  purpose, or noninfringement.  The Regents of the University of
24 *  Michigan shall not be liable for any damages, including special,
25 *  indirect, incidental, or consequential damages, with respect to any
26 *  claim arising out of or in connection with the use of the software,
27 *  even if it has been or is hereafter advised of the possibility of
28 *  such damages.
29 */
30
31#include <linux/export.h>
32#include <linux/nfs_fs.h>
33#include "nfs4session.h"
34#include "internal.h"
35#include "pnfs.h"
36
37#include "nfs4trace.h"
38
39#define NFSDBG_FACILITY		NFSDBG_PNFS
40
41/*
42 * Device ID RCU cache. A device ID is unique per server and layout type.
43 */
44#define NFS4_DEVICE_ID_HASH_BITS	5
45#define NFS4_DEVICE_ID_HASH_SIZE	(1 << NFS4_DEVICE_ID_HASH_BITS)
46#define NFS4_DEVICE_ID_HASH_MASK	(NFS4_DEVICE_ID_HASH_SIZE - 1)
47
48
49static struct hlist_head nfs4_deviceid_cache[NFS4_DEVICE_ID_HASH_SIZE];
50static DEFINE_SPINLOCK(nfs4_deviceid_lock);
51
52#ifdef NFS_DEBUG
53void
54nfs4_print_deviceid(const struct nfs4_deviceid *id)
55{
56	u32 *p = (u32 *)id;
57
58	dprintk("%s: device id= [%x%x%x%x]\n", __func__,
59		p[0], p[1], p[2], p[3]);
60}
61EXPORT_SYMBOL_GPL(nfs4_print_deviceid);
62#endif
63
64static inline u32
65nfs4_deviceid_hash(const struct nfs4_deviceid *id)
66{
67	unsigned char *cptr = (unsigned char *)id->data;
68	unsigned int nbytes = NFS4_DEVICEID4_SIZE;
69	u32 x = 0;
70
71	while (nbytes--) {
72		x *= 37;
73		x += *cptr++;
74	}
75	return x & NFS4_DEVICE_ID_HASH_MASK;
76}
77
78static struct nfs4_deviceid_node *
79_lookup_deviceid(const struct pnfs_layoutdriver_type *ld,
80		 const struct nfs_client *clp, const struct nfs4_deviceid *id,
81		 long hash)
82{
83	struct nfs4_deviceid_node *d;
84
85	hlist_for_each_entry_rcu(d, &nfs4_deviceid_cache[hash], node)
86		if (d->ld == ld && d->nfs_client == clp &&
87		    !memcmp(&d->deviceid, id, sizeof(*id))) {
88			if (atomic_read(&d->ref))
89				return d;
90			else
91				continue;
92		}
93	return NULL;
94}
95
96static struct nfs4_deviceid_node *
97nfs4_get_device_info(struct nfs_server *server,
98		const struct nfs4_deviceid *dev_id,
99		const struct cred *cred, gfp_t gfp_flags)
100{
101	struct nfs4_deviceid_node *d = NULL;
102	struct pnfs_device *pdev = NULL;
103	struct page **pages = NULL;
104	u32 max_resp_sz;
105	int max_pages;
106	int rc, i;
107
108	/*
109	 * Use the session max response size as the basis for setting
110	 * GETDEVICEINFO's maxcount
111	 */
112	max_resp_sz = server->nfs_client->cl_session->fc_attrs.max_resp_sz;
113	if (server->pnfs_curr_ld->max_deviceinfo_size &&
114	    server->pnfs_curr_ld->max_deviceinfo_size < max_resp_sz)
115		max_resp_sz = server->pnfs_curr_ld->max_deviceinfo_size;
116	max_pages = nfs_page_array_len(0, max_resp_sz);
117	dprintk("%s: server %p max_resp_sz %u max_pages %d\n",
118		__func__, server, max_resp_sz, max_pages);
119
120	pdev = kzalloc(sizeof(*pdev), gfp_flags);
121	if (!pdev)
122		return NULL;
123
124	pages = kcalloc(max_pages, sizeof(struct page *), gfp_flags);
125	if (!pages)
126		goto out_free_pdev;
127
128	for (i = 0; i < max_pages; i++) {
129		pages[i] = alloc_page(gfp_flags);
130		if (!pages[i])
131			goto out_free_pages;
132	}
133
134	memcpy(&pdev->dev_id, dev_id, sizeof(*dev_id));
135	pdev->layout_type = server->pnfs_curr_ld->id;
136	pdev->pages = pages;
137	pdev->pgbase = 0;
138	pdev->pglen = max_resp_sz;
139	pdev->mincount = 0;
140	pdev->maxcount = max_resp_sz - nfs41_maxgetdevinfo_overhead;
141
142	rc = nfs4_proc_getdeviceinfo(server, pdev, cred);
143	dprintk("%s getdevice info returns %d\n", __func__, rc);
144	if (rc)
145		goto out_free_pages;
146
147	/*
148	 * Found new device, need to decode it and then add it to the
149	 * list of known devices for this mountpoint.
150	 */
151	d = server->pnfs_curr_ld->alloc_deviceid_node(server, pdev,
152			gfp_flags);
153	if (d && pdev->nocache)
154		set_bit(NFS_DEVICEID_NOCACHE, &d->flags);
155
156out_free_pages:
157	while (--i >= 0)
158		__free_page(pages[i]);
159	kfree(pages);
160out_free_pdev:
161	kfree(pdev);
162	dprintk("<-- %s d %p\n", __func__, d);
163	return d;
164}
165
166/*
167 * Lookup a deviceid in cache and get a reference count on it if found
168 *
169 * @clp nfs_client associated with deviceid
170 * @id deviceid to look up
171 */
172static struct nfs4_deviceid_node *
173__nfs4_find_get_deviceid(struct nfs_server *server,
174		const struct nfs4_deviceid *id, long hash)
175{
176	struct nfs4_deviceid_node *d;
177
178	rcu_read_lock();
179	d = _lookup_deviceid(server->pnfs_curr_ld, server->nfs_client, id,
180			hash);
181	if (d != NULL && !atomic_inc_not_zero(&d->ref))
182		d = NULL;
183	rcu_read_unlock();
184	return d;
185}
186
187struct nfs4_deviceid_node *
188nfs4_find_get_deviceid(struct nfs_server *server,
189		const struct nfs4_deviceid *id, const struct cred *cred,
190		gfp_t gfp_mask)
191{
192	long hash = nfs4_deviceid_hash(id);
193	struct nfs4_deviceid_node *d, *new;
194
195	d = __nfs4_find_get_deviceid(server, id, hash);
196	if (d)
197		goto found;
198
199	new = nfs4_get_device_info(server, id, cred, gfp_mask);
200	if (!new) {
201		trace_nfs4_find_deviceid(server, id, -ENOENT);
202		return new;
203	}
204
205	spin_lock(&nfs4_deviceid_lock);
206	d = __nfs4_find_get_deviceid(server, id, hash);
207	if (d) {
208		spin_unlock(&nfs4_deviceid_lock);
209		server->pnfs_curr_ld->free_deviceid_node(new);
210	} else {
211		atomic_inc(&new->ref);
212		hlist_add_head_rcu(&new->node, &nfs4_deviceid_cache[hash]);
213		spin_unlock(&nfs4_deviceid_lock);
214		d = new;
215	}
216found:
217	trace_nfs4_find_deviceid(server, id, 0);
218	return d;
219}
220EXPORT_SYMBOL_GPL(nfs4_find_get_deviceid);
221
222/*
223 * Remove a deviceid from cache
224 *
225 * @clp nfs_client associated with deviceid
226 * @id the deviceid to unhash
227 *
228 * @ret the unhashed node, if found and dereferenced to zero, NULL otherwise.
229 */
230void
231nfs4_delete_deviceid(const struct pnfs_layoutdriver_type *ld,
232			 const struct nfs_client *clp, const struct nfs4_deviceid *id)
233{
234	struct nfs4_deviceid_node *d;
235
236	spin_lock(&nfs4_deviceid_lock);
237	rcu_read_lock();
238	d = _lookup_deviceid(ld, clp, id, nfs4_deviceid_hash(id));
239	rcu_read_unlock();
240	if (!d) {
241		spin_unlock(&nfs4_deviceid_lock);
242		return;
243	}
244	hlist_del_init_rcu(&d->node);
245	clear_bit(NFS_DEVICEID_NOCACHE, &d->flags);
246	spin_unlock(&nfs4_deviceid_lock);
247
248	/* balance the initial ref set in pnfs_insert_deviceid */
249	nfs4_put_deviceid_node(d);
250}
251EXPORT_SYMBOL_GPL(nfs4_delete_deviceid);
252
253void
254nfs4_init_deviceid_node(struct nfs4_deviceid_node *d, struct nfs_server *server,
255			const struct nfs4_deviceid *id)
256{
257	INIT_HLIST_NODE(&d->node);
258	INIT_HLIST_NODE(&d->tmpnode);
259	d->ld = server->pnfs_curr_ld;
260	d->nfs_client = server->nfs_client;
261	d->flags = 0;
262	d->deviceid = *id;
263	atomic_set(&d->ref, 1);
264}
265EXPORT_SYMBOL_GPL(nfs4_init_deviceid_node);
266
267/*
268 * Dereference a deviceid node and delete it when its reference count drops
269 * to zero.
270 *
271 * @d deviceid node to put
272 *
273 * return true iff the node was deleted
274 * Note that since the test for d->ref == 0 is sufficient to establish
275 * that the node is no longer hashed in the global device id cache.
276 */
277bool
278nfs4_put_deviceid_node(struct nfs4_deviceid_node *d)
279{
280	if (test_bit(NFS_DEVICEID_NOCACHE, &d->flags)) {
281		if (atomic_add_unless(&d->ref, -1, 2))
282			return false;
283		nfs4_delete_deviceid(d->ld, d->nfs_client, &d->deviceid);
284	}
285	if (!atomic_dec_and_test(&d->ref))
286		return false;
287	trace_nfs4_deviceid_free(d->nfs_client, &d->deviceid);
288	d->ld->free_deviceid_node(d);
289	return true;
290}
291EXPORT_SYMBOL_GPL(nfs4_put_deviceid_node);
292
293void
294nfs4_mark_deviceid_available(struct nfs4_deviceid_node *node)
295{
296	if (test_bit(NFS_DEVICEID_UNAVAILABLE, &node->flags)) {
297		clear_bit(NFS_DEVICEID_UNAVAILABLE, &node->flags);
298		smp_mb__after_atomic();
299	}
300}
301EXPORT_SYMBOL_GPL(nfs4_mark_deviceid_available);
302
303void
304nfs4_mark_deviceid_unavailable(struct nfs4_deviceid_node *node)
305{
306	node->timestamp_unavailable = jiffies;
307	smp_mb__before_atomic();
308	set_bit(NFS_DEVICEID_UNAVAILABLE, &node->flags);
309	smp_mb__after_atomic();
310}
311EXPORT_SYMBOL_GPL(nfs4_mark_deviceid_unavailable);
312
313bool
314nfs4_test_deviceid_unavailable(struct nfs4_deviceid_node *node)
315{
316	if (test_bit(NFS_DEVICEID_UNAVAILABLE, &node->flags)) {
317		unsigned long start, end;
318
319		end = jiffies;
320		start = end - PNFS_DEVICE_RETRY_TIMEOUT;
321		if (time_in_range(node->timestamp_unavailable, start, end))
322			return true;
323		clear_bit(NFS_DEVICEID_UNAVAILABLE, &node->flags);
324		smp_mb__after_atomic();
325	}
326	return false;
327}
328EXPORT_SYMBOL_GPL(nfs4_test_deviceid_unavailable);
329
330static void
331_deviceid_purge_client(const struct nfs_client *clp, long hash)
332{
333	struct nfs4_deviceid_node *d;
334	HLIST_HEAD(tmp);
335
336	spin_lock(&nfs4_deviceid_lock);
337	rcu_read_lock();
338	hlist_for_each_entry_rcu(d, &nfs4_deviceid_cache[hash], node)
339		if (d->nfs_client == clp && atomic_read(&d->ref)) {
340			hlist_del_init_rcu(&d->node);
341			hlist_add_head(&d->tmpnode, &tmp);
342			clear_bit(NFS_DEVICEID_NOCACHE, &d->flags);
343		}
344	rcu_read_unlock();
345	spin_unlock(&nfs4_deviceid_lock);
346
347	if (hlist_empty(&tmp))
348		return;
349
350	while (!hlist_empty(&tmp)) {
351		d = hlist_entry(tmp.first, struct nfs4_deviceid_node, tmpnode);
352		hlist_del(&d->tmpnode);
353		nfs4_put_deviceid_node(d);
354	}
355}
356
357void
358nfs4_deviceid_purge_client(const struct nfs_client *clp)
359{
360	long h;
361
362	if (!(clp->cl_exchange_flags & EXCHGID4_FLAG_USE_PNFS_MDS))
363		return;
364	for (h = 0; h < NFS4_DEVICE_ID_HASH_SIZE; h++)
365		_deviceid_purge_client(clp, h);
366}
367
368/*
369 * Stop use of all deviceids associated with an nfs_client
370 */
371void
372nfs4_deviceid_mark_client_invalid(struct nfs_client *clp)
373{
374	struct nfs4_deviceid_node *d;
375	int i;
376
377	rcu_read_lock();
378	for (i = 0; i < NFS4_DEVICE_ID_HASH_SIZE; i ++){
379		hlist_for_each_entry_rcu(d, &nfs4_deviceid_cache[i], node)
380			if (d->nfs_client == clp)
381				set_bit(NFS_DEVICEID_INVALID, &d->flags);
382	}
383	rcu_read_unlock();
384}
385