1 /*
2  *  Device operations for the pnfs nfs4 file layout driver.
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/nfs_fs.h>
32 #include <linux/vmalloc.h>
33 #include <linux/module.h>
34 
35 #include "../internal.h"
36 #include "../nfs4session.h"
37 #include "filelayout.h"
38 
39 #define NFSDBG_FACILITY		NFSDBG_PNFS_LD
40 
41 static unsigned int dataserver_timeo = NFS4_DEF_DS_TIMEO;
42 static unsigned int dataserver_retrans = NFS4_DEF_DS_RETRANS;
43 
44 void
45 nfs4_fl_free_deviceid(struct nfs4_file_layout_dsaddr *dsaddr)
46 {
47 	struct nfs4_pnfs_ds *ds;
48 	int i;
49 
50 	nfs4_print_deviceid(&dsaddr->id_node.deviceid);
51 
52 	for (i = 0; i < dsaddr->ds_num; i++) {
53 		ds = dsaddr->ds_list[i];
54 		if (ds != NULL)
55 			nfs4_pnfs_ds_put(ds);
56 	}
57 	kfree(dsaddr->stripe_indices);
58 	kfree_rcu(dsaddr, id_node.rcu);
59 }
60 
61 /* Decode opaque device data and return the result */
62 struct nfs4_file_layout_dsaddr *
63 nfs4_fl_alloc_deviceid_node(struct nfs_server *server, struct pnfs_device *pdev,
64 		gfp_t gfp_flags)
65 {
66 	int i;
67 	u32 cnt, num;
68 	u8 *indexp;
69 	__be32 *p;
70 	u8 *stripe_indices;
71 	u8 max_stripe_index;
72 	struct nfs4_file_layout_dsaddr *dsaddr = NULL;
73 	struct xdr_stream stream;
74 	struct xdr_buf buf;
75 	struct page *scratch;
76 	struct list_head dsaddrs;
77 	struct nfs4_pnfs_ds_addr *da;
78 
79 	/* set up xdr stream */
80 	scratch = alloc_page(gfp_flags);
81 	if (!scratch)
82 		goto out_err;
83 
84 	xdr_init_decode_pages(&stream, &buf, pdev->pages, pdev->pglen);
85 	xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
86 
87 	/* Get the stripe count (number of stripe index) */
88 	p = xdr_inline_decode(&stream, 4);
89 	if (unlikely(!p))
90 		goto out_err_free_scratch;
91 
92 	cnt = be32_to_cpup(p);
93 	dprintk("%s stripe count  %d\n", __func__, cnt);
94 	if (cnt > NFS4_PNFS_MAX_STRIPE_CNT) {
95 		printk(KERN_WARNING "NFS: %s: stripe count %d greater than "
96 		       "supported maximum %d\n", __func__,
97 			cnt, NFS4_PNFS_MAX_STRIPE_CNT);
98 		goto out_err_free_scratch;
99 	}
100 
101 	/* read stripe indices */
102 	stripe_indices = kcalloc(cnt, sizeof(u8), gfp_flags);
103 	if (!stripe_indices)
104 		goto out_err_free_scratch;
105 
106 	p = xdr_inline_decode(&stream, cnt << 2);
107 	if (unlikely(!p))
108 		goto out_err_free_stripe_indices;
109 
110 	indexp = &stripe_indices[0];
111 	max_stripe_index = 0;
112 	for (i = 0; i < cnt; i++) {
113 		*indexp = be32_to_cpup(p++);
114 		max_stripe_index = max(max_stripe_index, *indexp);
115 		indexp++;
116 	}
117 
118 	/* Check the multipath list count */
119 	p = xdr_inline_decode(&stream, 4);
120 	if (unlikely(!p))
121 		goto out_err_free_stripe_indices;
122 
123 	num = be32_to_cpup(p);
124 	dprintk("%s ds_num %u\n", __func__, num);
125 	if (num > NFS4_PNFS_MAX_MULTI_CNT) {
126 		printk(KERN_WARNING "NFS: %s: multipath count %d greater than "
127 			"supported maximum %d\n", __func__,
128 			num, NFS4_PNFS_MAX_MULTI_CNT);
129 		goto out_err_free_stripe_indices;
130 	}
131 
132 	/* validate stripe indices are all < num */
133 	if (max_stripe_index >= num) {
134 		printk(KERN_WARNING "NFS: %s: stripe index %u >= num ds %u\n",
135 			__func__, max_stripe_index, num);
136 		goto out_err_free_stripe_indices;
137 	}
138 
139 	dsaddr = kzalloc(sizeof(*dsaddr) +
140 			(sizeof(struct nfs4_pnfs_ds *) * (num - 1)),
141 			gfp_flags);
142 	if (!dsaddr)
143 		goto out_err_free_stripe_indices;
144 
145 	dsaddr->stripe_count = cnt;
146 	dsaddr->stripe_indices = stripe_indices;
147 	stripe_indices = NULL;
148 	dsaddr->ds_num = num;
149 	nfs4_init_deviceid_node(&dsaddr->id_node, server, &pdev->dev_id);
150 
151 	INIT_LIST_HEAD(&dsaddrs);
152 
153 	for (i = 0; i < dsaddr->ds_num; i++) {
154 		int j;
155 		u32 mp_count;
156 
157 		p = xdr_inline_decode(&stream, 4);
158 		if (unlikely(!p))
159 			goto out_err_free_deviceid;
160 
161 		mp_count = be32_to_cpup(p); /* multipath count */
162 		for (j = 0; j < mp_count; j++) {
163 			da = nfs4_decode_mp_ds_addr(server->nfs_client->cl_net,
164 						    &stream, gfp_flags);
165 			if (da)
166 				list_add_tail(&da->da_node, &dsaddrs);
167 		}
168 		if (list_empty(&dsaddrs)) {
169 			dprintk("%s: no suitable DS addresses found\n",
170 				__func__);
171 			goto out_err_free_deviceid;
172 		}
173 
174 		dsaddr->ds_list[i] = nfs4_pnfs_ds_add(&dsaddrs, gfp_flags);
175 		if (!dsaddr->ds_list[i])
176 			goto out_err_drain_dsaddrs;
177 
178 		/* If DS was already in cache, free ds addrs */
179 		while (!list_empty(&dsaddrs)) {
180 			da = list_first_entry(&dsaddrs,
181 					      struct nfs4_pnfs_ds_addr,
182 					      da_node);
183 			list_del_init(&da->da_node);
184 			kfree(da->da_remotestr);
185 			kfree(da);
186 		}
187 	}
188 
189 	__free_page(scratch);
190 	return dsaddr;
191 
192 out_err_drain_dsaddrs:
193 	while (!list_empty(&dsaddrs)) {
194 		da = list_first_entry(&dsaddrs, struct nfs4_pnfs_ds_addr,
195 				      da_node);
196 		list_del_init(&da->da_node);
197 		kfree(da->da_remotestr);
198 		kfree(da);
199 	}
200 out_err_free_deviceid:
201 	nfs4_fl_free_deviceid(dsaddr);
202 	/* stripe_indicies was part of dsaddr */
203 	goto out_err_free_scratch;
204 out_err_free_stripe_indices:
205 	kfree(stripe_indices);
206 out_err_free_scratch:
207 	__free_page(scratch);
208 out_err:
209 	dprintk("%s ERROR: returning NULL\n", __func__);
210 	return NULL;
211 }
212 
213 void
214 nfs4_fl_put_deviceid(struct nfs4_file_layout_dsaddr *dsaddr)
215 {
216 	nfs4_put_deviceid_node(&dsaddr->id_node);
217 }
218 
219 /*
220  * Want res = (offset - layout->pattern_offset)/ layout->stripe_unit
221  * Then: ((res + fsi) % dsaddr->stripe_count)
222  */
223 u32
224 nfs4_fl_calc_j_index(struct pnfs_layout_segment *lseg, loff_t offset)
225 {
226 	struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
227 	u64 tmp;
228 
229 	tmp = offset - flseg->pattern_offset;
230 	do_div(tmp, flseg->stripe_unit);
231 	tmp += flseg->first_stripe_index;
232 	return do_div(tmp, flseg->dsaddr->stripe_count);
233 }
234 
235 u32
236 nfs4_fl_calc_ds_index(struct pnfs_layout_segment *lseg, u32 j)
237 {
238 	return FILELAYOUT_LSEG(lseg)->dsaddr->stripe_indices[j];
239 }
240 
241 struct nfs_fh *
242 nfs4_fl_select_ds_fh(struct pnfs_layout_segment *lseg, u32 j)
243 {
244 	struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
245 	u32 i;
246 
247 	if (flseg->stripe_type == STRIPE_SPARSE) {
248 		if (flseg->num_fh == 1)
249 			i = 0;
250 		else if (flseg->num_fh == 0)
251 			/* Use the MDS OPEN fh set in nfs_read_rpcsetup */
252 			return NULL;
253 		else
254 			i = nfs4_fl_calc_ds_index(lseg, j);
255 	} else
256 		i = j;
257 	return flseg->fh_array[i];
258 }
259 
260 /* Upon return, either ds is connected, or ds is NULL */
261 struct nfs4_pnfs_ds *
262 nfs4_fl_prepare_ds(struct pnfs_layout_segment *lseg, u32 ds_idx)
263 {
264 	struct nfs4_file_layout_dsaddr *dsaddr = FILELAYOUT_LSEG(lseg)->dsaddr;
265 	struct nfs4_pnfs_ds *ds = dsaddr->ds_list[ds_idx];
266 	struct nfs4_deviceid_node *devid = FILELAYOUT_DEVID_NODE(lseg);
267 	struct nfs4_pnfs_ds *ret = ds;
268 	struct nfs_server *s = NFS_SERVER(lseg->pls_layout->plh_inode);
269 
270 	if (ds == NULL) {
271 		printk(KERN_ERR "NFS: %s: No data server for offset index %d\n",
272 			__func__, ds_idx);
273 		pnfs_generic_mark_devid_invalid(devid);
274 		goto out;
275 	}
276 	smp_rmb();
277 	if (ds->ds_clp)
278 		goto out_test_devid;
279 
280 	nfs4_pnfs_ds_connect(s, ds, devid, dataserver_timeo,
281 			     dataserver_retrans, 4,
282 			     s->nfs_client->cl_minorversion,
283 			     s->nfs_client->cl_rpcclient->cl_auth->au_flavor);
284 
285 out_test_devid:
286 	if (filelayout_test_devid_unavailable(devid))
287 		ret = NULL;
288 out:
289 	return ret;
290 }
291 
292 module_param(dataserver_retrans, uint, 0644);
293 MODULE_PARM_DESC(dataserver_retrans, "The  number of times the NFSv4.1 client "
294 			"retries a request before it attempts further "
295 			" recovery  action.");
296 module_param(dataserver_timeo, uint, 0644);
297 MODULE_PARM_DESC(dataserver_timeo, "The time (in tenths of a second) the "
298 			"NFSv4.1  client  waits for a response from a "
299 			" data server before it retries an NFS request.");
300