1/*
2 * Copyright (c) 2014 Christoph Hellwig.
3 */
4#include <linux/exportfs.h>
5#include <linux/genhd.h>
6#include <linux/slab.h>
7
8#include <linux/nfsd/debug.h>
9
10#include "blocklayoutxdr.h"
11#include "pnfs.h"
12
13#define NFSDDBG_FACILITY	NFSDDBG_PNFS
14
15
16static int
17nfsd4_block_get_device_info_simple(struct super_block *sb,
18		struct nfsd4_getdeviceinfo *gdp)
19{
20	struct pnfs_block_deviceaddr *dev;
21	struct pnfs_block_volume *b;
22
23	dev = kzalloc(sizeof(struct pnfs_block_deviceaddr) +
24		      sizeof(struct pnfs_block_volume), GFP_KERNEL);
25	if (!dev)
26		return -ENOMEM;
27	gdp->gd_device = dev;
28
29	dev->nr_volumes = 1;
30	b = &dev->volumes[0];
31
32	b->type = PNFS_BLOCK_VOLUME_SIMPLE;
33	b->simple.sig_len = PNFS_BLOCK_UUID_LEN;
34	return sb->s_export_op->get_uuid(sb, b->simple.sig, &b->simple.sig_len,
35			&b->simple.offset);
36}
37
38static __be32
39nfsd4_block_proc_getdeviceinfo(struct super_block *sb,
40		struct nfsd4_getdeviceinfo *gdp)
41{
42	if (sb->s_bdev != sb->s_bdev->bd_contains)
43		return nfserr_inval;
44	return nfserrno(nfsd4_block_get_device_info_simple(sb, gdp));
45}
46
47static __be32
48nfsd4_block_proc_layoutget(struct inode *inode, const struct svc_fh *fhp,
49		struct nfsd4_layoutget *args)
50{
51	struct nfsd4_layout_seg *seg = &args->lg_seg;
52	struct super_block *sb = inode->i_sb;
53	u32 block_size = (1 << inode->i_blkbits);
54	struct pnfs_block_extent *bex;
55	struct iomap iomap;
56	u32 device_generation = 0;
57	int error;
58
59	if (seg->offset & (block_size - 1)) {
60		dprintk("pnfsd: I/O misaligned\n");
61		goto out_layoutunavailable;
62	}
63
64	/*
65	 * Some clients barf on non-zero block numbers for NONE or INVALID
66	 * layouts, so make sure to zero the whole structure.
67	 */
68	error = -ENOMEM;
69	bex = kzalloc(sizeof(*bex), GFP_KERNEL);
70	if (!bex)
71		goto out_error;
72	args->lg_content = bex;
73
74	error = sb->s_export_op->map_blocks(inode, seg->offset, seg->length,
75					    &iomap, seg->iomode != IOMODE_READ,
76					    &device_generation);
77	if (error) {
78		if (error == -ENXIO)
79			goto out_layoutunavailable;
80		goto out_error;
81	}
82
83	if (iomap.length < args->lg_minlength) {
84		dprintk("pnfsd: extent smaller than minlength\n");
85		goto out_layoutunavailable;
86	}
87
88	switch (iomap.type) {
89	case IOMAP_MAPPED:
90		if (seg->iomode == IOMODE_READ)
91			bex->es = PNFS_BLOCK_READ_DATA;
92		else
93			bex->es = PNFS_BLOCK_READWRITE_DATA;
94		bex->soff = (iomap.blkno << 9);
95		break;
96	case IOMAP_UNWRITTEN:
97		if (seg->iomode & IOMODE_RW) {
98			/*
99			 * Crack monkey special case from section 2.3.1.
100			 */
101			if (args->lg_minlength == 0) {
102				dprintk("pnfsd: no soup for you!\n");
103				goto out_layoutunavailable;
104			}
105
106			bex->es = PNFS_BLOCK_INVALID_DATA;
107			bex->soff = (iomap.blkno << 9);
108			break;
109		}
110		/*FALLTHRU*/
111	case IOMAP_HOLE:
112		if (seg->iomode == IOMODE_READ) {
113			bex->es = PNFS_BLOCK_NONE_DATA;
114			break;
115		}
116		/*FALLTHRU*/
117	case IOMAP_DELALLOC:
118	default:
119		WARN(1, "pnfsd: filesystem returned %d extent\n", iomap.type);
120		goto out_layoutunavailable;
121	}
122
123	error = nfsd4_set_deviceid(&bex->vol_id, fhp, device_generation);
124	if (error)
125		goto out_error;
126	bex->foff = iomap.offset;
127	bex->len = iomap.length;
128
129	seg->offset = iomap.offset;
130	seg->length = iomap.length;
131
132	dprintk("GET: 0x%llx:0x%llx %d\n", bex->foff, bex->len, bex->es);
133	return 0;
134
135out_error:
136	seg->length = 0;
137	return nfserrno(error);
138out_layoutunavailable:
139	seg->length = 0;
140	return nfserr_layoutunavailable;
141}
142
143static __be32
144nfsd4_block_proc_layoutcommit(struct inode *inode,
145		struct nfsd4_layoutcommit *lcp)
146{
147	loff_t new_size = lcp->lc_last_wr + 1;
148	struct iattr iattr = { .ia_valid = 0 };
149	struct iomap *iomaps;
150	int nr_iomaps;
151	int error;
152
153	nr_iomaps = nfsd4_block_decode_layoutupdate(lcp->lc_up_layout,
154			lcp->lc_up_len, &iomaps, 1 << inode->i_blkbits);
155	if (nr_iomaps < 0)
156		return nfserrno(nr_iomaps);
157
158	if (lcp->lc_mtime.tv_nsec == UTIME_NOW ||
159	    timespec_compare(&lcp->lc_mtime, &inode->i_mtime) < 0)
160		lcp->lc_mtime = current_fs_time(inode->i_sb);
161	iattr.ia_valid |= ATTR_ATIME | ATTR_CTIME | ATTR_MTIME;
162	iattr.ia_atime = iattr.ia_ctime = iattr.ia_mtime = lcp->lc_mtime;
163
164	if (new_size > i_size_read(inode)) {
165		iattr.ia_valid |= ATTR_SIZE;
166		iattr.ia_size = new_size;
167	}
168
169	error = inode->i_sb->s_export_op->commit_blocks(inode, iomaps,
170			nr_iomaps, &iattr);
171	kfree(iomaps);
172	return nfserrno(error);
173}
174
175const struct nfsd4_layout_ops bl_layout_ops = {
176	/*
177	 * Pretend that we send notification to the client.  This is a blatant
178	 * lie to force recent Linux clients to cache our device IDs.
179	 * We rarely ever change the device ID, so the harm of leaking deviceids
180	 * for a while isn't too bad.  Unfortunately RFC5661 is a complete mess
181	 * in this regard, but I filed errata 4119 for this a while ago, and
182	 * hopefully the Linux client will eventually start caching deviceids
183	 * without this again.
184	 */
185	.notify_types		=
186			NOTIFY_DEVICEID4_DELETE | NOTIFY_DEVICEID4_CHANGE,
187	.proc_getdeviceinfo	= nfsd4_block_proc_getdeviceinfo,
188	.encode_getdeviceinfo	= nfsd4_block_encode_getdeviceinfo,
189	.proc_layoutget		= nfsd4_block_proc_layoutget,
190	.encode_layoutget	= nfsd4_block_encode_layoutget,
191	.proc_layoutcommit	= nfsd4_block_proc_layoutcommit,
192};
193