1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (c) 2014-2016 Christoph Hellwig.
4 */
5 #include <linux/exportfs.h>
6 #include <linux/iomap.h>
7 #include <linux/slab.h>
8 #include <linux/pr.h>
9
10 #include <linux/nfsd/debug.h>
11
12 #include "blocklayoutxdr.h"
13 #include "pnfs.h"
14 #include "filecache.h"
15
16 #define NFSDDBG_FACILITY NFSDDBG_PNFS
17
18
19 static __be32
nfsd4_block_proc_layoutget(struct inode * inode,const struct svc_fh * fhp,struct nfsd4_layoutget * args)20 nfsd4_block_proc_layoutget(struct inode *inode, const struct svc_fh *fhp,
21 struct nfsd4_layoutget *args)
22 {
23 struct nfsd4_layout_seg *seg = &args->lg_seg;
24 struct super_block *sb = inode->i_sb;
25 u32 block_size = i_blocksize(inode);
26 struct pnfs_block_extent *bex;
27 struct iomap iomap;
28 u32 device_generation = 0;
29 int error;
30
31 if (seg->offset & (block_size - 1)) {
32 dprintk("pnfsd: I/O misaligned\n");
33 goto out_layoutunavailable;
34 }
35
36 /*
37 * Some clients barf on non-zero block numbers for NONE or INVALID
38 * layouts, so make sure to zero the whole structure.
39 */
40 error = -ENOMEM;
41 bex = kzalloc(sizeof(*bex), GFP_KERNEL);
42 if (!bex)
43 goto out_error;
44 args->lg_content = bex;
45
46 error = sb->s_export_op->map_blocks(inode, seg->offset, seg->length,
47 &iomap, seg->iomode != IOMODE_READ,
48 &device_generation);
49 if (error) {
50 if (error == -ENXIO)
51 goto out_layoutunavailable;
52 goto out_error;
53 }
54
55 if (iomap.length < args->lg_minlength) {
56 dprintk("pnfsd: extent smaller than minlength\n");
57 goto out_layoutunavailable;
58 }
59
60 switch (iomap.type) {
61 case IOMAP_MAPPED:
62 if (seg->iomode == IOMODE_READ)
63 bex->es = PNFS_BLOCK_READ_DATA;
64 else
65 bex->es = PNFS_BLOCK_READWRITE_DATA;
66 bex->soff = iomap.addr;
67 break;
68 case IOMAP_UNWRITTEN:
69 if (seg->iomode & IOMODE_RW) {
70 /*
71 * Crack monkey special case from section 2.3.1.
72 */
73 if (args->lg_minlength == 0) {
74 dprintk("pnfsd: no soup for you!\n");
75 goto out_layoutunavailable;
76 }
77
78 bex->es = PNFS_BLOCK_INVALID_DATA;
79 bex->soff = iomap.addr;
80 break;
81 }
82 fallthrough;
83 case IOMAP_HOLE:
84 if (seg->iomode == IOMODE_READ) {
85 bex->es = PNFS_BLOCK_NONE_DATA;
86 break;
87 }
88 fallthrough;
89 case IOMAP_DELALLOC:
90 default:
91 WARN(1, "pnfsd: filesystem returned %d extent\n", iomap.type);
92 goto out_layoutunavailable;
93 }
94
95 error = nfsd4_set_deviceid(&bex->vol_id, fhp, device_generation);
96 if (error)
97 goto out_error;
98 bex->foff = iomap.offset;
99 bex->len = iomap.length;
100
101 seg->offset = iomap.offset;
102 seg->length = iomap.length;
103
104 dprintk("GET: 0x%llx:0x%llx %d\n", bex->foff, bex->len, bex->es);
105 return 0;
106
107 out_error:
108 seg->length = 0;
109 return nfserrno(error);
110 out_layoutunavailable:
111 seg->length = 0;
112 return nfserr_layoutunavailable;
113 }
114
115 static __be32
nfsd4_block_commit_blocks(struct inode * inode,struct nfsd4_layoutcommit * lcp,struct iomap * iomaps,int nr_iomaps)116 nfsd4_block_commit_blocks(struct inode *inode, struct nfsd4_layoutcommit *lcp,
117 struct iomap *iomaps, int nr_iomaps)
118 {
119 loff_t new_size = lcp->lc_last_wr + 1;
120 struct iattr iattr = { .ia_valid = 0 };
121 int error;
122
123 if (lcp->lc_mtime.tv_nsec == UTIME_NOW ||
124 timespec64_compare(&lcp->lc_mtime, &inode->i_mtime) < 0)
125 lcp->lc_mtime = current_time(inode);
126 iattr.ia_valid |= ATTR_ATIME | ATTR_CTIME | ATTR_MTIME;
127 iattr.ia_atime = iattr.ia_ctime = iattr.ia_mtime = lcp->lc_mtime;
128
129 if (new_size > i_size_read(inode)) {
130 iattr.ia_valid |= ATTR_SIZE;
131 iattr.ia_size = new_size;
132 }
133
134 error = inode->i_sb->s_export_op->commit_blocks(inode, iomaps,
135 nr_iomaps, &iattr);
136 kfree(iomaps);
137 return nfserrno(error);
138 }
139
140 #ifdef CONFIG_NFSD_BLOCKLAYOUT
141 static int
nfsd4_block_get_device_info_simple(struct super_block * sb,struct nfsd4_getdeviceinfo * gdp)142 nfsd4_block_get_device_info_simple(struct super_block *sb,
143 struct nfsd4_getdeviceinfo *gdp)
144 {
145 struct pnfs_block_deviceaddr *dev;
146 struct pnfs_block_volume *b;
147
148 dev = kzalloc(sizeof(struct pnfs_block_deviceaddr) +
149 sizeof(struct pnfs_block_volume), GFP_KERNEL);
150 if (!dev)
151 return -ENOMEM;
152 gdp->gd_device = dev;
153
154 dev->nr_volumes = 1;
155 b = &dev->volumes[0];
156
157 b->type = PNFS_BLOCK_VOLUME_SIMPLE;
158 b->simple.sig_len = PNFS_BLOCK_UUID_LEN;
159 return sb->s_export_op->get_uuid(sb, b->simple.sig, &b->simple.sig_len,
160 &b->simple.offset);
161 }
162
163 static __be32
nfsd4_block_proc_getdeviceinfo(struct super_block * sb,struct svc_rqst * rqstp,struct nfs4_client * clp,struct nfsd4_getdeviceinfo * gdp)164 nfsd4_block_proc_getdeviceinfo(struct super_block *sb,
165 struct svc_rqst *rqstp,
166 struct nfs4_client *clp,
167 struct nfsd4_getdeviceinfo *gdp)
168 {
169 if (bdev_is_partition(sb->s_bdev))
170 return nfserr_inval;
171 return nfserrno(nfsd4_block_get_device_info_simple(sb, gdp));
172 }
173
174 static __be32
nfsd4_block_proc_layoutcommit(struct inode * inode,struct nfsd4_layoutcommit * lcp)175 nfsd4_block_proc_layoutcommit(struct inode *inode,
176 struct nfsd4_layoutcommit *lcp)
177 {
178 struct iomap *iomaps;
179 int nr_iomaps;
180
181 nr_iomaps = nfsd4_block_decode_layoutupdate(lcp->lc_up_layout,
182 lcp->lc_up_len, &iomaps, i_blocksize(inode));
183 if (nr_iomaps < 0)
184 return nfserrno(nr_iomaps);
185
186 return nfsd4_block_commit_blocks(inode, lcp, iomaps, nr_iomaps);
187 }
188
189 const struct nfsd4_layout_ops bl_layout_ops = {
190 /*
191 * Pretend that we send notification to the client. This is a blatant
192 * lie to force recent Linux clients to cache our device IDs.
193 * We rarely ever change the device ID, so the harm of leaking deviceids
194 * for a while isn't too bad. Unfortunately RFC5661 is a complete mess
195 * in this regard, but I filed errata 4119 for this a while ago, and
196 * hopefully the Linux client will eventually start caching deviceids
197 * without this again.
198 */
199 .notify_types =
200 NOTIFY_DEVICEID4_DELETE | NOTIFY_DEVICEID4_CHANGE,
201 .proc_getdeviceinfo = nfsd4_block_proc_getdeviceinfo,
202 .encode_getdeviceinfo = nfsd4_block_encode_getdeviceinfo,
203 .proc_layoutget = nfsd4_block_proc_layoutget,
204 .encode_layoutget = nfsd4_block_encode_layoutget,
205 .proc_layoutcommit = nfsd4_block_proc_layoutcommit,
206 };
207 #endif /* CONFIG_NFSD_BLOCKLAYOUT */
208
209 #ifdef CONFIG_NFSD_SCSILAYOUT
210 #define NFSD_MDS_PR_KEY 0x0100000000000000ULL
211
212 /*
213 * We use the client ID as a unique key for the reservations.
214 * This allows us to easily fence a client when recalls fail.
215 */
nfsd4_scsi_pr_key(struct nfs4_client * clp)216 static u64 nfsd4_scsi_pr_key(struct nfs4_client *clp)
217 {
218 return ((u64)clp->cl_clientid.cl_boot << 32) | clp->cl_clientid.cl_id;
219 }
220
221 static const u8 designator_types[] = {
222 PS_DESIGNATOR_EUI64,
223 PS_DESIGNATOR_NAA,
224 };
225
226 static int
nfsd4_block_get_unique_id(struct gendisk * disk,struct pnfs_block_volume * b)227 nfsd4_block_get_unique_id(struct gendisk *disk, struct pnfs_block_volume *b)
228 {
229 int ret, i;
230
231 for (i = 0; i < ARRAY_SIZE(designator_types); i++) {
232 u8 type = designator_types[i];
233
234 ret = disk->fops->get_unique_id(disk, b->scsi.designator, type);
235 if (ret > 0) {
236 b->scsi.code_set = PS_CODE_SET_BINARY;
237 b->scsi.designator_type = type;
238 b->scsi.designator_len = ret;
239 return 0;
240 }
241 }
242
243 return -EINVAL;
244 }
245
246 static int
nfsd4_block_get_device_info_scsi(struct super_block * sb,struct nfs4_client * clp,struct nfsd4_getdeviceinfo * gdp)247 nfsd4_block_get_device_info_scsi(struct super_block *sb,
248 struct nfs4_client *clp,
249 struct nfsd4_getdeviceinfo *gdp)
250 {
251 struct pnfs_block_deviceaddr *dev;
252 struct pnfs_block_volume *b;
253 const struct pr_ops *ops;
254 int ret;
255
256 dev = kzalloc(sizeof(struct pnfs_block_deviceaddr) +
257 sizeof(struct pnfs_block_volume), GFP_KERNEL);
258 if (!dev)
259 return -ENOMEM;
260 gdp->gd_device = dev;
261
262 dev->nr_volumes = 1;
263 b = &dev->volumes[0];
264
265 b->type = PNFS_BLOCK_VOLUME_SCSI;
266 b->scsi.pr_key = nfsd4_scsi_pr_key(clp);
267
268 ret = nfsd4_block_get_unique_id(sb->s_bdev->bd_disk, b);
269 if (ret < 0)
270 goto out_free_dev;
271
272 ret = -EINVAL;
273 ops = sb->s_bdev->bd_disk->fops->pr_ops;
274 if (!ops) {
275 pr_err("pNFS: device %s does not support PRs.\n",
276 sb->s_id);
277 goto out_free_dev;
278 }
279
280 ret = ops->pr_register(sb->s_bdev, 0, NFSD_MDS_PR_KEY, true);
281 if (ret) {
282 pr_err("pNFS: failed to register key for device %s.\n",
283 sb->s_id);
284 goto out_free_dev;
285 }
286
287 ret = ops->pr_reserve(sb->s_bdev, NFSD_MDS_PR_KEY,
288 PR_EXCLUSIVE_ACCESS_REG_ONLY, 0);
289 if (ret) {
290 pr_err("pNFS: failed to reserve device %s.\n",
291 sb->s_id);
292 goto out_free_dev;
293 }
294
295 return 0;
296
297 out_free_dev:
298 kfree(dev);
299 return ret;
300 }
301
302 static __be32
nfsd4_scsi_proc_getdeviceinfo(struct super_block * sb,struct svc_rqst * rqstp,struct nfs4_client * clp,struct nfsd4_getdeviceinfo * gdp)303 nfsd4_scsi_proc_getdeviceinfo(struct super_block *sb,
304 struct svc_rqst *rqstp,
305 struct nfs4_client *clp,
306 struct nfsd4_getdeviceinfo *gdp)
307 {
308 if (bdev_is_partition(sb->s_bdev))
309 return nfserr_inval;
310 return nfserrno(nfsd4_block_get_device_info_scsi(sb, clp, gdp));
311 }
312 static __be32
nfsd4_scsi_proc_layoutcommit(struct inode * inode,struct nfsd4_layoutcommit * lcp)313 nfsd4_scsi_proc_layoutcommit(struct inode *inode,
314 struct nfsd4_layoutcommit *lcp)
315 {
316 struct iomap *iomaps;
317 int nr_iomaps;
318
319 nr_iomaps = nfsd4_scsi_decode_layoutupdate(lcp->lc_up_layout,
320 lcp->lc_up_len, &iomaps, i_blocksize(inode));
321 if (nr_iomaps < 0)
322 return nfserrno(nr_iomaps);
323
324 return nfsd4_block_commit_blocks(inode, lcp, iomaps, nr_iomaps);
325 }
326
327 static void
nfsd4_scsi_fence_client(struct nfs4_layout_stateid * ls)328 nfsd4_scsi_fence_client(struct nfs4_layout_stateid *ls)
329 {
330 struct nfs4_client *clp = ls->ls_stid.sc_client;
331 struct block_device *bdev = ls->ls_file->nf_file->f_path.mnt->mnt_sb->s_bdev;
332
333 bdev->bd_disk->fops->pr_ops->pr_preempt(bdev, NFSD_MDS_PR_KEY,
334 nfsd4_scsi_pr_key(clp), 0, true);
335 }
336
337 const struct nfsd4_layout_ops scsi_layout_ops = {
338 /*
339 * Pretend that we send notification to the client. This is a blatant
340 * lie to force recent Linux clients to cache our device IDs.
341 * We rarely ever change the device ID, so the harm of leaking deviceids
342 * for a while isn't too bad. Unfortunately RFC5661 is a complete mess
343 * in this regard, but I filed errata 4119 for this a while ago, and
344 * hopefully the Linux client will eventually start caching deviceids
345 * without this again.
346 */
347 .notify_types =
348 NOTIFY_DEVICEID4_DELETE | NOTIFY_DEVICEID4_CHANGE,
349 .proc_getdeviceinfo = nfsd4_scsi_proc_getdeviceinfo,
350 .encode_getdeviceinfo = nfsd4_block_encode_getdeviceinfo,
351 .proc_layoutget = nfsd4_block_proc_layoutget,
352 .encode_layoutget = nfsd4_block_encode_layoutget,
353 .proc_layoutcommit = nfsd4_scsi_proc_layoutcommit,
354 .fence_client = nfsd4_scsi_fence_client,
355 };
356 #endif /* CONFIG_NFSD_SCSILAYOUT */
357