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