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 #include "../nfs4trace.h"
39
40 #define NFSDBG_FACILITY NFSDBG_PNFS_LD
41
42 static unsigned int dataserver_timeo = NFS4_DEF_DS_TIMEO;
43 static unsigned int dataserver_retrans = NFS4_DEF_DS_RETRANS;
44
45 void
nfs4_fl_free_deviceid(struct nfs4_file_layout_dsaddr * dsaddr)46 nfs4_fl_free_deviceid(struct nfs4_file_layout_dsaddr *dsaddr)
47 {
48 struct nfs4_pnfs_ds *ds;
49 int i;
50
51 nfs4_print_deviceid(&dsaddr->id_node.deviceid);
52
53 for (i = 0; i < dsaddr->ds_num; i++) {
54 ds = dsaddr->ds_list[i];
55 if (ds != NULL)
56 nfs4_pnfs_ds_put(ds);
57 }
58 kfree(dsaddr->stripe_indices);
59 kfree_rcu(dsaddr, id_node.rcu);
60 }
61
62 /* Decode opaque device data and return the result */
63 struct nfs4_file_layout_dsaddr *
nfs4_fl_alloc_deviceid_node(struct nfs_server * server,struct pnfs_device * pdev,gfp_t gfp_flags)64 nfs4_fl_alloc_deviceid_node(struct nfs_server *server, struct pnfs_device *pdev,
65 gfp_t gfp_flags)
66 {
67 int i;
68 u32 cnt, num;
69 u8 *indexp;
70 __be32 *p;
71 u8 *stripe_indices;
72 u8 max_stripe_index;
73 struct nfs4_file_layout_dsaddr *dsaddr = NULL;
74 struct xdr_stream stream;
75 struct xdr_buf buf;
76 struct folio *scratch;
77 struct list_head dsaddrs;
78 struct nfs4_pnfs_ds_addr *da;
79 struct net *net = server->nfs_client->cl_net;
80
81 /* set up xdr stream */
82 scratch = folio_alloc(gfp_flags, 0);
83 if (!scratch)
84 goto out_err;
85
86 xdr_init_decode_pages(&stream, &buf, pdev->pages, pdev->pglen);
87 xdr_set_scratch_folio(&stream, scratch);
88
89 /* Get the stripe count (number of stripe index) */
90 p = xdr_inline_decode(&stream, 4);
91 if (unlikely(!p))
92 goto out_err_free_scratch;
93
94 cnt = be32_to_cpup(p);
95 dprintk("%s stripe count %d\n", __func__, cnt);
96 if (cnt > NFS4_PNFS_MAX_STRIPE_CNT) {
97 printk(KERN_WARNING "NFS: %s: stripe count %d greater than "
98 "supported maximum %d\n", __func__,
99 cnt, NFS4_PNFS_MAX_STRIPE_CNT);
100 goto out_err_free_scratch;
101 }
102
103 /* read stripe indices */
104 stripe_indices = kcalloc(cnt, sizeof(u8), gfp_flags);
105 if (!stripe_indices)
106 goto out_err_free_scratch;
107
108 p = xdr_inline_decode(&stream, cnt << 2);
109 if (unlikely(!p))
110 goto out_err_free_stripe_indices;
111
112 indexp = &stripe_indices[0];
113 max_stripe_index = 0;
114 for (i = 0; i < cnt; i++) {
115 *indexp = be32_to_cpup(p++);
116 max_stripe_index = max(max_stripe_index, *indexp);
117 indexp++;
118 }
119
120 /* Check the multipath list count */
121 p = xdr_inline_decode(&stream, 4);
122 if (unlikely(!p))
123 goto out_err_free_stripe_indices;
124
125 num = be32_to_cpup(p);
126 dprintk("%s ds_num %u\n", __func__, num);
127 if (num > NFS4_PNFS_MAX_MULTI_CNT) {
128 printk(KERN_WARNING "NFS: %s: multipath count %d greater than "
129 "supported maximum %d\n", __func__,
130 num, NFS4_PNFS_MAX_MULTI_CNT);
131 goto out_err_free_stripe_indices;
132 }
133
134 /* validate stripe indices are all < num */
135 if (max_stripe_index >= num) {
136 printk(KERN_WARNING "NFS: %s: stripe index %u >= num ds %u\n",
137 __func__, max_stripe_index, num);
138 goto out_err_free_stripe_indices;
139 }
140
141 dsaddr = kzalloc_flex(*dsaddr, ds_list, num, 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(net, &stream, gfp_flags);
164 if (da)
165 list_add_tail(&da->da_node, &dsaddrs);
166 }
167 if (list_empty(&dsaddrs)) {
168 dprintk("%s: no suitable DS addresses found\n",
169 __func__);
170 goto out_err_free_deviceid;
171 }
172
173 dsaddr->ds_list[i] = nfs4_pnfs_ds_add(net, &dsaddrs, 4,
174 gfp_flags);
175 if (!dsaddr->ds_list[i])
176 goto out_err_drain_dsaddrs;
177 trace_fl_getdevinfo(server, &pdev->dev_id, dsaddr->ds_list[i]->ds_remotestr);
178
179 /* If DS was already in cache, free ds addrs */
180 while (!list_empty(&dsaddrs)) {
181 da = list_first_entry(&dsaddrs,
182 struct nfs4_pnfs_ds_addr,
183 da_node);
184 list_del_init(&da->da_node);
185 kfree(da->da_remotestr);
186 kfree(da);
187 }
188 }
189
190 folio_put(scratch);
191 return dsaddr;
192
193 out_err_drain_dsaddrs:
194 while (!list_empty(&dsaddrs)) {
195 da = list_first_entry(&dsaddrs, struct nfs4_pnfs_ds_addr,
196 da_node);
197 list_del_init(&da->da_node);
198 kfree(da->da_remotestr);
199 kfree(da);
200 }
201 out_err_free_deviceid:
202 nfs4_fl_free_deviceid(dsaddr);
203 /* stripe_indicies was part of dsaddr */
204 goto out_err_free_scratch;
205 out_err_free_stripe_indices:
206 kfree(stripe_indices);
207 out_err_free_scratch:
208 folio_put(scratch);
209 out_err:
210 dprintk("%s ERROR: returning NULL\n", __func__);
211 return NULL;
212 }
213
214 void
nfs4_fl_put_deviceid(struct nfs4_file_layout_dsaddr * dsaddr)215 nfs4_fl_put_deviceid(struct nfs4_file_layout_dsaddr *dsaddr)
216 {
217 nfs4_put_deviceid_node(&dsaddr->id_node);
218 }
219
220 /*
221 * Want res = (offset - layout->pattern_offset)/ layout->stripe_unit
222 * Then: ((res + fsi) % dsaddr->stripe_count)
223 */
224 u32
nfs4_fl_calc_j_index(struct pnfs_layout_segment * lseg,loff_t offset)225 nfs4_fl_calc_j_index(struct pnfs_layout_segment *lseg, loff_t offset)
226 {
227 struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
228 u64 tmp;
229
230 tmp = offset - flseg->pattern_offset;
231 do_div(tmp, flseg->stripe_unit);
232 tmp += flseg->first_stripe_index;
233 return do_div(tmp, flseg->dsaddr->stripe_count);
234 }
235
236 u32
nfs4_fl_calc_ds_index(struct pnfs_layout_segment * lseg,u32 j)237 nfs4_fl_calc_ds_index(struct pnfs_layout_segment *lseg, u32 j)
238 {
239 return FILELAYOUT_LSEG(lseg)->dsaddr->stripe_indices[j];
240 }
241
242 struct nfs_fh *
nfs4_fl_select_ds_fh(struct pnfs_layout_segment * lseg,u32 j)243 nfs4_fl_select_ds_fh(struct pnfs_layout_segment *lseg, u32 j)
244 {
245 struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
246 u32 i;
247
248 if (flseg->stripe_type == STRIPE_SPARSE) {
249 if (flseg->num_fh == 1)
250 i = 0;
251 else if (flseg->num_fh == 0)
252 /* Use the MDS OPEN fh set in nfs_read_rpcsetup */
253 return NULL;
254 else
255 i = nfs4_fl_calc_ds_index(lseg, j);
256 } else
257 i = j;
258 return flseg->fh_array[i];
259 }
260
261 /* Upon return, either ds is connected, or ds is NULL */
262 struct nfs4_pnfs_ds *
nfs4_fl_prepare_ds(struct pnfs_layout_segment * lseg,u32 ds_idx)263 nfs4_fl_prepare_ds(struct pnfs_layout_segment *lseg, u32 ds_idx)
264 {
265 struct nfs4_file_layout_dsaddr *dsaddr = FILELAYOUT_LSEG(lseg)->dsaddr;
266 struct nfs4_pnfs_ds *ds = dsaddr->ds_list[ds_idx];
267 struct nfs4_deviceid_node *devid = FILELAYOUT_DEVID_NODE(lseg);
268 struct nfs4_pnfs_ds *ret = ds;
269 struct nfs_server *s = NFS_SERVER(lseg->pls_layout->plh_inode);
270 int status;
271
272 if (ds == NULL) {
273 printk(KERN_ERR "NFS: %s: No data server for offset index %d\n",
274 __func__, ds_idx);
275 pnfs_generic_mark_devid_invalid(devid);
276 goto out;
277 }
278 smp_rmb();
279 if (ds->ds_clp)
280 goto out_test_devid;
281
282 status = nfs4_pnfs_ds_connect(s, ds, devid, dataserver_timeo,
283 dataserver_retrans, 4,
284 s->nfs_client->cl_minorversion, true);
285 if (status) {
286 nfs4_mark_deviceid_unavailable(devid);
287 ret = NULL;
288 goto out;
289 }
290
291 out_test_devid:
292 if (ret->ds_clp == NULL ||
293 filelayout_test_devid_unavailable(devid))
294 ret = NULL;
295 out:
296 return ret;
297 }
298
299 module_param(dataserver_retrans, uint, 0644);
300 MODULE_PARM_DESC(dataserver_retrans, "The number of times the NFSv4.1 client "
301 "retries a request before it attempts further "
302 " recovery action.");
303 module_param(dataserver_timeo, uint, 0644);
304 MODULE_PARM_DESC(dataserver_timeo, "The time (in tenths of a second) the "
305 "NFSv4.1 client waits for a response from a "
306 " data server before it retries an NFS request.");
307