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 #include "vfs.h"
16
17 #define NFSDDBG_FACILITY NFSDDBG_PNFS
18
19
20 static __be32
nfsd4_block_proc_layoutget(struct svc_rqst * rqstp,struct inode * inode,const struct svc_fh * fhp,struct nfsd4_layoutget * args)21 nfsd4_block_proc_layoutget(struct svc_rqst *rqstp, struct inode *inode,
22 const struct svc_fh *fhp, struct nfsd4_layoutget *args)
23 {
24 struct nfsd4_layout_seg *seg = &args->lg_seg;
25 struct super_block *sb = inode->i_sb;
26 u32 block_size = i_blocksize(inode);
27 struct pnfs_block_extent *bex;
28 struct iomap iomap;
29 u32 device_generation = 0;
30 int error;
31
32 if (locks_in_grace(SVC_NET(rqstp)))
33 return nfserr_grace;
34
35 if (seg->offset & (block_size - 1)) {
36 dprintk("pnfsd: I/O misaligned\n");
37 goto out_layoutunavailable;
38 }
39
40 /*
41 * Some clients barf on non-zero block numbers for NONE or INVALID
42 * layouts, so make sure to zero the whole structure.
43 */
44 error = -ENOMEM;
45 bex = kzalloc(sizeof(*bex), GFP_KERNEL);
46 if (!bex)
47 goto out_error;
48 args->lg_content = bex;
49
50 error = sb->s_export_op->map_blocks(inode, seg->offset, seg->length,
51 &iomap, seg->iomode != IOMODE_READ,
52 &device_generation);
53 if (error) {
54 if (error == -ENXIO)
55 goto out_layoutunavailable;
56 goto out_error;
57 }
58
59 if (iomap.length < args->lg_minlength) {
60 dprintk("pnfsd: extent smaller than minlength\n");
61 goto out_layoutunavailable;
62 }
63
64 switch (iomap.type) {
65 case IOMAP_MAPPED:
66 if (seg->iomode == IOMODE_READ)
67 bex->es = PNFS_BLOCK_READ_DATA;
68 else
69 bex->es = PNFS_BLOCK_READWRITE_DATA;
70 bex->soff = iomap.addr;
71 break;
72 case IOMAP_UNWRITTEN:
73 if (seg->iomode & IOMODE_RW) {
74 /*
75 * Crack monkey special case from section 2.3.1.
76 */
77 if (args->lg_minlength == 0) {
78 dprintk("pnfsd: no soup for you!\n");
79 goto out_layoutunavailable;
80 }
81
82 bex->es = PNFS_BLOCK_INVALID_DATA;
83 bex->soff = iomap.addr;
84 break;
85 }
86 fallthrough;
87 case IOMAP_HOLE:
88 if (seg->iomode == IOMODE_READ) {
89 bex->es = PNFS_BLOCK_NONE_DATA;
90 break;
91 }
92 fallthrough;
93 case IOMAP_DELALLOC:
94 default:
95 WARN(1, "pnfsd: filesystem returned %d extent\n", iomap.type);
96 goto out_layoutunavailable;
97 }
98
99 error = nfsd4_set_deviceid(&bex->vol_id, fhp, device_generation);
100 if (error)
101 goto out_error;
102 bex->foff = iomap.offset;
103 bex->len = iomap.length;
104
105 seg->offset = iomap.offset;
106 seg->length = iomap.length;
107
108 dprintk("GET: 0x%llx:0x%llx %d\n", bex->foff, bex->len, bex->es);
109 return 0;
110
111 out_error:
112 seg->length = 0;
113 return nfserrno(error);
114 out_layoutunavailable:
115 seg->length = 0;
116 return nfserr_layoutunavailable;
117 }
118
119 static __be32
nfsd4_block_commit_blocks(struct inode * inode,struct nfsd4_layoutcommit * lcp,struct iomap * iomaps,int nr_iomaps)120 nfsd4_block_commit_blocks(struct inode *inode, struct nfsd4_layoutcommit *lcp,
121 struct iomap *iomaps, int nr_iomaps)
122 {
123 struct timespec64 mtime = inode_get_mtime(inode);
124 struct iattr iattr = { .ia_valid = 0 };
125 int error;
126
127 if (lcp->lc_mtime.tv_nsec == UTIME_NOW ||
128 timespec64_compare(&lcp->lc_mtime, &mtime) < 0)
129 lcp->lc_mtime = current_time(inode);
130 iattr.ia_valid |= ATTR_ATIME | ATTR_CTIME | ATTR_MTIME;
131 iattr.ia_atime = iattr.ia_ctime = iattr.ia_mtime = lcp->lc_mtime;
132
133 if (lcp->lc_size_chg) {
134 iattr.ia_valid |= ATTR_SIZE;
135 iattr.ia_size = lcp->lc_newsize;
136 }
137
138 error = inode->i_sb->s_export_op->commit_blocks(inode, iomaps,
139 nr_iomaps, &iattr);
140 kfree(iomaps);
141 return nfserrno(error);
142 }
143
144 #ifdef CONFIG_NFSD_BLOCKLAYOUT
145 static int
nfsd4_block_get_device_info_simple(struct super_block * sb,struct nfsd4_getdeviceinfo * gdp)146 nfsd4_block_get_device_info_simple(struct super_block *sb,
147 struct nfsd4_getdeviceinfo *gdp)
148 {
149 struct pnfs_block_deviceaddr *dev;
150 struct pnfs_block_volume *b;
151
152 dev = kzalloc(struct_size(dev, volumes, 1), GFP_KERNEL);
153 if (!dev)
154 return -ENOMEM;
155 gdp->gd_device = dev;
156
157 dev->nr_volumes = 1;
158 b = &dev->volumes[0];
159
160 b->type = PNFS_BLOCK_VOLUME_SIMPLE;
161 b->simple.sig_len = PNFS_BLOCK_UUID_LEN;
162 return sb->s_export_op->get_uuid(sb, b->simple.sig, &b->simple.sig_len,
163 &b->simple.offset);
164 }
165
166 static __be32
nfsd4_block_proc_getdeviceinfo(struct super_block * sb,struct svc_rqst * rqstp,struct nfs4_client * clp,struct nfsd4_getdeviceinfo * gdp)167 nfsd4_block_proc_getdeviceinfo(struct super_block *sb,
168 struct svc_rqst *rqstp,
169 struct nfs4_client *clp,
170 struct nfsd4_getdeviceinfo *gdp)
171 {
172 if (bdev_is_partition(sb->s_bdev))
173 return nfserr_inval;
174 return nfserrno(nfsd4_block_get_device_info_simple(sb, gdp));
175 }
176
177 static __be32
nfsd4_block_proc_layoutcommit(struct inode * inode,struct svc_rqst * rqstp,struct nfsd4_layoutcommit * lcp)178 nfsd4_block_proc_layoutcommit(struct inode *inode, struct svc_rqst *rqstp,
179 struct nfsd4_layoutcommit *lcp)
180 {
181 struct iomap *iomaps;
182 int nr_iomaps;
183 __be32 nfserr;
184
185 rqstp->rq_arg = lcp->lc_up_layout;
186 svcxdr_init_decode(rqstp);
187
188 nfserr = nfsd4_block_decode_layoutupdate(&rqstp->rq_arg_stream,
189 &iomaps, &nr_iomaps, i_blocksize(inode));
190 if (nfserr != nfs_ok)
191 return nfserr;
192
193 return nfsd4_block_commit_blocks(inode, lcp, iomaps, nr_iomaps);
194 }
195
196 const struct nfsd4_layout_ops bl_layout_ops = {
197 /*
198 * Pretend that we send notification to the client. This is a blatant
199 * lie to force recent Linux clients to cache our device IDs.
200 * We rarely ever change the device ID, so the harm of leaking deviceids
201 * for a while isn't too bad. Unfortunately RFC5661 is a complete mess
202 * in this regard, but I filed errata 4119 for this a while ago, and
203 * hopefully the Linux client will eventually start caching deviceids
204 * without this again.
205 */
206 .notify_types =
207 NOTIFY_DEVICEID4_DELETE | NOTIFY_DEVICEID4_CHANGE,
208 .proc_getdeviceinfo = nfsd4_block_proc_getdeviceinfo,
209 .encode_getdeviceinfo = nfsd4_block_encode_getdeviceinfo,
210 .proc_layoutget = nfsd4_block_proc_layoutget,
211 .encode_layoutget = nfsd4_block_encode_layoutget,
212 .proc_layoutcommit = nfsd4_block_proc_layoutcommit,
213 };
214 #endif /* CONFIG_NFSD_BLOCKLAYOUT */
215
216 #ifdef CONFIG_NFSD_SCSILAYOUT
217 #define NFSD_MDS_PR_KEY 0x0100000000000000ULL
218
219 /*
220 * We use the client ID as a unique key for the reservations.
221 * This allows us to easily fence a client when recalls fail.
222 */
nfsd4_scsi_pr_key(struct nfs4_client * clp)223 static u64 nfsd4_scsi_pr_key(struct nfs4_client *clp)
224 {
225 return ((u64)clp->cl_clientid.cl_boot << 32) | clp->cl_clientid.cl_id;
226 }
227
228 static const u8 designator_types[] = {
229 PS_DESIGNATOR_EUI64,
230 PS_DESIGNATOR_NAA,
231 };
232
233 static int
nfsd4_block_get_unique_id(struct gendisk * disk,struct pnfs_block_volume * b)234 nfsd4_block_get_unique_id(struct gendisk *disk, struct pnfs_block_volume *b)
235 {
236 int ret, i;
237
238 for (i = 0; i < ARRAY_SIZE(designator_types); i++) {
239 u8 type = designator_types[i];
240
241 ret = disk->fops->get_unique_id(disk, b->scsi.designator, type);
242 if (ret > 0) {
243 b->scsi.code_set = PS_CODE_SET_BINARY;
244 b->scsi.designator_type = type;
245 b->scsi.designator_len = ret;
246 return 0;
247 }
248 }
249
250 return -EINVAL;
251 }
252
253 static int
nfsd4_block_get_device_info_scsi(struct super_block * sb,struct nfs4_client * clp,struct nfsd4_getdeviceinfo * gdp)254 nfsd4_block_get_device_info_scsi(struct super_block *sb,
255 struct nfs4_client *clp,
256 struct nfsd4_getdeviceinfo *gdp)
257 {
258 struct pnfs_block_deviceaddr *dev;
259 struct pnfs_block_volume *b;
260 const struct pr_ops *ops;
261 int ret;
262
263 dev = kzalloc(struct_size(dev, volumes, 1), GFP_KERNEL);
264 if (!dev)
265 return -ENOMEM;
266 gdp->gd_device = dev;
267
268 dev->nr_volumes = 1;
269 b = &dev->volumes[0];
270
271 b->type = PNFS_BLOCK_VOLUME_SCSI;
272 b->scsi.pr_key = nfsd4_scsi_pr_key(clp);
273
274 ret = nfsd4_block_get_unique_id(sb->s_bdev->bd_disk, b);
275 if (ret < 0)
276 goto out_free_dev;
277
278 ret = -EINVAL;
279 ops = sb->s_bdev->bd_disk->fops->pr_ops;
280 if (!ops) {
281 pr_err("pNFS: device %s does not support PRs.\n",
282 sb->s_id);
283 goto out_free_dev;
284 }
285
286 ret = ops->pr_register(sb->s_bdev, 0, NFSD_MDS_PR_KEY, true);
287 if (ret) {
288 pr_err("pNFS: failed to register key for device %s.\n",
289 sb->s_id);
290 goto out_free_dev;
291 }
292
293 ret = ops->pr_reserve(sb->s_bdev, NFSD_MDS_PR_KEY,
294 PR_EXCLUSIVE_ACCESS_REG_ONLY, 0);
295 if (ret) {
296 pr_err("pNFS: failed to reserve device %s.\n",
297 sb->s_id);
298 goto out_free_dev;
299 }
300
301 return 0;
302
303 out_free_dev:
304 kfree(dev);
305 gdp->gd_device = NULL;
306 return ret;
307 }
308
309 static __be32
nfsd4_scsi_proc_getdeviceinfo(struct super_block * sb,struct svc_rqst * rqstp,struct nfs4_client * clp,struct nfsd4_getdeviceinfo * gdp)310 nfsd4_scsi_proc_getdeviceinfo(struct super_block *sb,
311 struct svc_rqst *rqstp,
312 struct nfs4_client *clp,
313 struct nfsd4_getdeviceinfo *gdp)
314 {
315 if (bdev_is_partition(sb->s_bdev))
316 return nfserr_inval;
317 return nfserrno(nfsd4_block_get_device_info_scsi(sb, clp, gdp));
318 }
319 static __be32
nfsd4_scsi_proc_layoutcommit(struct inode * inode,struct svc_rqst * rqstp,struct nfsd4_layoutcommit * lcp)320 nfsd4_scsi_proc_layoutcommit(struct inode *inode, struct svc_rqst *rqstp,
321 struct nfsd4_layoutcommit *lcp)
322 {
323 struct iomap *iomaps;
324 int nr_iomaps;
325 __be32 nfserr;
326
327 rqstp->rq_arg = lcp->lc_up_layout;
328 svcxdr_init_decode(rqstp);
329
330 nfserr = nfsd4_scsi_decode_layoutupdate(&rqstp->rq_arg_stream,
331 &iomaps, &nr_iomaps, i_blocksize(inode));
332 if (nfserr != nfs_ok)
333 return nfserr;
334
335 return nfsd4_block_commit_blocks(inode, lcp, iomaps, nr_iomaps);
336 }
337
338 static void
nfsd4_scsi_fence_client(struct nfs4_layout_stateid * ls,struct nfsd_file * file)339 nfsd4_scsi_fence_client(struct nfs4_layout_stateid *ls, struct nfsd_file *file)
340 {
341 struct nfs4_client *clp = ls->ls_stid.sc_client;
342 struct block_device *bdev = file->nf_file->f_path.mnt->mnt_sb->s_bdev;
343
344 bdev->bd_disk->fops->pr_ops->pr_preempt(bdev, NFSD_MDS_PR_KEY,
345 nfsd4_scsi_pr_key(clp), 0, true);
346 }
347
348 const struct nfsd4_layout_ops scsi_layout_ops = {
349 /*
350 * Pretend that we send notification to the client. This is a blatant
351 * lie to force recent Linux clients to cache our device IDs.
352 * We rarely ever change the device ID, so the harm of leaking deviceids
353 * for a while isn't too bad. Unfortunately RFC5661 is a complete mess
354 * in this regard, but I filed errata 4119 for this a while ago, and
355 * hopefully the Linux client will eventually start caching deviceids
356 * without this again.
357 */
358 .notify_types =
359 NOTIFY_DEVICEID4_DELETE | NOTIFY_DEVICEID4_CHANGE,
360 .proc_getdeviceinfo = nfsd4_scsi_proc_getdeviceinfo,
361 .encode_getdeviceinfo = nfsd4_block_encode_getdeviceinfo,
362 .proc_layoutget = nfsd4_block_proc_layoutget,
363 .encode_layoutget = nfsd4_block_encode_layoutget,
364 .proc_layoutcommit = nfsd4_scsi_proc_layoutcommit,
365 .fence_client = nfsd4_scsi_fence_client,
366 };
367 #endif /* CONFIG_NFSD_SCSILAYOUT */
368