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
20static __be32
21nfsd4_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(node: 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(size: 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(id: &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
108out_error:
109 seg->length = 0;
110 return nfserrno(errno: error);
111out_layoutunavailable:
112 seg->length = 0;
113 return nfserr_layoutunavailable;
114}
115
116static __be32
117nfsd4_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(lhs: &lcp->lc_mtime, rhs: &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(objp: iomaps);
139 return nfserrno(errno: error);
140}
141
142#ifdef CONFIG_NFSD_BLOCKLAYOUT
143static int
144nfsd4_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(size: sizeof(struct pnfs_block_deviceaddr) +
151 sizeof(struct pnfs_block_volume), GFP_KERNEL);
152 if (!dev)
153 return -ENOMEM;
154 gdp->gd_device = dev;
155
156 dev->nr_volumes = 1;
157 b = &dev->volumes[0];
158
159 b->type = PNFS_BLOCK_VOLUME_SIMPLE;
160 b->simple.sig_len = PNFS_BLOCK_UUID_LEN;
161 return sb->s_export_op->get_uuid(sb, b->simple.sig, &b->simple.sig_len,
162 &b->simple.offset);
163}
164
165static __be32
166nfsd4_block_proc_getdeviceinfo(struct super_block *sb,
167 struct svc_rqst *rqstp,
168 struct nfs4_client *clp,
169 struct nfsd4_getdeviceinfo *gdp)
170{
171 if (bdev_is_partition(bdev: sb->s_bdev))
172 return nfserr_inval;
173 return nfserrno(errno: nfsd4_block_get_device_info_simple(sb, gdp));
174}
175
176static __be32
177nfsd4_block_proc_layoutcommit(struct inode *inode,
178 struct nfsd4_layoutcommit *lcp)
179{
180 struct iomap *iomaps;
181 int nr_iomaps;
182
183 nr_iomaps = nfsd4_block_decode_layoutupdate(p: lcp->lc_up_layout,
184 len: lcp->lc_up_len, iomapp: &iomaps, block_size: i_blocksize(node: inode));
185 if (nr_iomaps < 0)
186 return nfserrno(errno: nr_iomaps);
187
188 return nfsd4_block_commit_blocks(inode, lcp, iomaps, nr_iomaps);
189}
190
191const struct nfsd4_layout_ops bl_layout_ops = {
192 /*
193 * Pretend that we send notification to the client. This is a blatant
194 * lie to force recent Linux clients to cache our device IDs.
195 * We rarely ever change the device ID, so the harm of leaking deviceids
196 * for a while isn't too bad. Unfortunately RFC5661 is a complete mess
197 * in this regard, but I filed errata 4119 for this a while ago, and
198 * hopefully the Linux client will eventually start caching deviceids
199 * without this again.
200 */
201 .notify_types =
202 NOTIFY_DEVICEID4_DELETE | NOTIFY_DEVICEID4_CHANGE,
203 .proc_getdeviceinfo = nfsd4_block_proc_getdeviceinfo,
204 .encode_getdeviceinfo = nfsd4_block_encode_getdeviceinfo,
205 .proc_layoutget = nfsd4_block_proc_layoutget,
206 .encode_layoutget = nfsd4_block_encode_layoutget,
207 .proc_layoutcommit = nfsd4_block_proc_layoutcommit,
208};
209#endif /* CONFIG_NFSD_BLOCKLAYOUT */
210
211#ifdef CONFIG_NFSD_SCSILAYOUT
212#define NFSD_MDS_PR_KEY 0x0100000000000000ULL
213
214/*
215 * We use the client ID as a unique key for the reservations.
216 * This allows us to easily fence a client when recalls fail.
217 */
218static u64 nfsd4_scsi_pr_key(struct nfs4_client *clp)
219{
220 return ((u64)clp->cl_clientid.cl_boot << 32) | clp->cl_clientid.cl_id;
221}
222
223static const u8 designator_types[] = {
224 PS_DESIGNATOR_EUI64,
225 PS_DESIGNATOR_NAA,
226};
227
228static int
229nfsd4_block_get_unique_id(struct gendisk *disk, struct pnfs_block_volume *b)
230{
231 int ret, i;
232
233 for (i = 0; i < ARRAY_SIZE(designator_types); i++) {
234 u8 type = designator_types[i];
235
236 ret = disk->fops->get_unique_id(disk, b->scsi.designator, type);
237 if (ret > 0) {
238 b->scsi.code_set = PS_CODE_SET_BINARY;
239 b->scsi.designator_type = type;
240 b->scsi.designator_len = ret;
241 return 0;
242 }
243 }
244
245 return -EINVAL;
246}
247
248static int
249nfsd4_block_get_device_info_scsi(struct super_block *sb,
250 struct nfs4_client *clp,
251 struct nfsd4_getdeviceinfo *gdp)
252{
253 struct pnfs_block_deviceaddr *dev;
254 struct pnfs_block_volume *b;
255 const struct pr_ops *ops;
256 int ret;
257
258 dev = kzalloc(size: sizeof(struct pnfs_block_deviceaddr) +
259 sizeof(struct pnfs_block_volume), 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(disk: 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
299out_free_dev:
300 kfree(objp: dev);
301 gdp->gd_device = NULL;
302 return ret;
303}
304
305static __be32
306nfsd4_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(bdev: sb->s_bdev))
312 return nfserr_inval;
313 return nfserrno(errno: nfsd4_block_get_device_info_scsi(sb, clp, gdp));
314}
315static __be32
316nfsd4_scsi_proc_layoutcommit(struct inode *inode,
317 struct nfsd4_layoutcommit *lcp)
318{
319 struct iomap *iomaps;
320 int nr_iomaps;
321
322 nr_iomaps = nfsd4_scsi_decode_layoutupdate(p: lcp->lc_up_layout,
323 len: lcp->lc_up_len, iomapp: &iomaps, block_size: i_blocksize(node: inode));
324 if (nr_iomaps < 0)
325 return nfserrno(errno: nr_iomaps);
326
327 return nfsd4_block_commit_blocks(inode, lcp, iomaps, nr_iomaps);
328}
329
330static void
331nfsd4_scsi_fence_client(struct nfs4_layout_stateid *ls)
332{
333 struct nfs4_client *clp = ls->ls_stid.sc_client;
334 struct block_device *bdev = ls->ls_file->nf_file->f_path.mnt->mnt_sb->s_bdev;
335
336 bdev->bd_disk->fops->pr_ops->pr_preempt(bdev, NFSD_MDS_PR_KEY,
337 nfsd4_scsi_pr_key(clp), 0, true);
338}
339
340const struct nfsd4_layout_ops scsi_layout_ops = {
341 /*
342 * Pretend that we send notification to the client. This is a blatant
343 * lie to force recent Linux clients to cache our device IDs.
344 * We rarely ever change the device ID, so the harm of leaking deviceids
345 * for a while isn't too bad. Unfortunately RFC5661 is a complete mess
346 * in this regard, but I filed errata 4119 for this a while ago, and
347 * hopefully the Linux client will eventually start caching deviceids
348 * without this again.
349 */
350 .notify_types =
351 NOTIFY_DEVICEID4_DELETE | NOTIFY_DEVICEID4_CHANGE,
352 .proc_getdeviceinfo = nfsd4_scsi_proc_getdeviceinfo,
353 .encode_getdeviceinfo = nfsd4_block_encode_getdeviceinfo,
354 .proc_layoutget = nfsd4_block_proc_layoutget,
355 .encode_layoutget = nfsd4_block_encode_layoutget,
356 .proc_layoutcommit = nfsd4_scsi_proc_layoutcommit,
357 .fence_client = nfsd4_scsi_fence_client,
358};
359#endif /* CONFIG_NFSD_SCSILAYOUT */
360

source code of linux/fs/nfsd/blocklayout.c