1#include <linux/ceph/ceph_debug.h>
2#include <linux/in.h>
3
4#include "super.h"
5#include "mds_client.h"
6#include "ioctl.h"
7
8
9/*
10 * ioctls
11 */
12
13/*
14 * get and set the file layout
15 */
16static long ceph_ioctl_get_layout(struct file *file, void __user *arg)
17{
18	struct ceph_inode_info *ci = ceph_inode(file_inode(file));
19	struct ceph_ioctl_layout l;
20	int err;
21
22	err = ceph_do_getattr(file_inode(file), CEPH_STAT_CAP_LAYOUT, false);
23	if (!err) {
24		l.stripe_unit = ceph_file_layout_su(ci->i_layout);
25		l.stripe_count = ceph_file_layout_stripe_count(ci->i_layout);
26		l.object_size = ceph_file_layout_object_size(ci->i_layout);
27		l.data_pool = le32_to_cpu(ci->i_layout.fl_pg_pool);
28		l.preferred_osd = (s32)-1;
29		if (copy_to_user(arg, &l, sizeof(l)))
30			return -EFAULT;
31	}
32
33	return err;
34}
35
36static long __validate_layout(struct ceph_mds_client *mdsc,
37			      struct ceph_ioctl_layout *l)
38{
39	int i, err;
40
41	/* validate striping parameters */
42	if ((l->object_size & ~PAGE_MASK) ||
43	    (l->stripe_unit & ~PAGE_MASK) ||
44	    ((unsigned)l->stripe_unit != 0 &&
45	     ((unsigned)l->object_size % (unsigned)l->stripe_unit)))
46		return -EINVAL;
47
48	/* make sure it's a valid data pool */
49	mutex_lock(&mdsc->mutex);
50	err = -EINVAL;
51	for (i = 0; i < mdsc->mdsmap->m_num_data_pg_pools; i++)
52		if (mdsc->mdsmap->m_data_pg_pools[i] == l->data_pool) {
53			err = 0;
54			break;
55		}
56	mutex_unlock(&mdsc->mutex);
57	if (err)
58		return err;
59
60	return 0;
61}
62
63static long ceph_ioctl_set_layout(struct file *file, void __user *arg)
64{
65	struct inode *inode = file_inode(file);
66	struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
67	struct ceph_mds_request *req;
68	struct ceph_ioctl_layout l;
69	struct ceph_inode_info *ci = ceph_inode(file_inode(file));
70	struct ceph_ioctl_layout nl;
71	int err;
72
73	if (copy_from_user(&l, arg, sizeof(l)))
74		return -EFAULT;
75
76	/* validate changed params against current layout */
77	err = ceph_do_getattr(file_inode(file), CEPH_STAT_CAP_LAYOUT, false);
78	if (err)
79		return err;
80
81	memset(&nl, 0, sizeof(nl));
82	if (l.stripe_count)
83		nl.stripe_count = l.stripe_count;
84	else
85		nl.stripe_count = ceph_file_layout_stripe_count(ci->i_layout);
86	if (l.stripe_unit)
87		nl.stripe_unit = l.stripe_unit;
88	else
89		nl.stripe_unit = ceph_file_layout_su(ci->i_layout);
90	if (l.object_size)
91		nl.object_size = l.object_size;
92	else
93		nl.object_size = ceph_file_layout_object_size(ci->i_layout);
94	if (l.data_pool)
95		nl.data_pool = l.data_pool;
96	else
97		nl.data_pool = ceph_file_layout_pg_pool(ci->i_layout);
98
99	/* this is obsolete, and always -1 */
100	nl.preferred_osd = le64_to_cpu(-1);
101
102	err = __validate_layout(mdsc, &nl);
103	if (err)
104		return err;
105
106	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETLAYOUT,
107				       USE_AUTH_MDS);
108	if (IS_ERR(req))
109		return PTR_ERR(req);
110	req->r_inode = inode;
111	ihold(inode);
112	req->r_num_caps = 1;
113
114	req->r_inode_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL;
115
116	req->r_args.setlayout.layout.fl_stripe_unit =
117		cpu_to_le32(l.stripe_unit);
118	req->r_args.setlayout.layout.fl_stripe_count =
119		cpu_to_le32(l.stripe_count);
120	req->r_args.setlayout.layout.fl_object_size =
121		cpu_to_le32(l.object_size);
122	req->r_args.setlayout.layout.fl_pg_pool = cpu_to_le32(l.data_pool);
123
124	err = ceph_mdsc_do_request(mdsc, NULL, req);
125	ceph_mdsc_put_request(req);
126	return err;
127}
128
129/*
130 * Set a layout policy on a directory inode. All items in the tree
131 * rooted at this inode will inherit this layout on creation,
132 * (It doesn't apply retroactively )
133 * unless a subdirectory has its own layout policy.
134 */
135static long ceph_ioctl_set_layout_policy (struct file *file, void __user *arg)
136{
137	struct inode *inode = file_inode(file);
138	struct ceph_mds_request *req;
139	struct ceph_ioctl_layout l;
140	int err;
141	struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
142
143	/* copy and validate */
144	if (copy_from_user(&l, arg, sizeof(l)))
145		return -EFAULT;
146
147	err = __validate_layout(mdsc, &l);
148	if (err)
149		return err;
150
151	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETDIRLAYOUT,
152				       USE_AUTH_MDS);
153
154	if (IS_ERR(req))
155		return PTR_ERR(req);
156	req->r_inode = inode;
157	ihold(inode);
158	req->r_num_caps = 1;
159
160	req->r_args.setlayout.layout.fl_stripe_unit =
161			cpu_to_le32(l.stripe_unit);
162	req->r_args.setlayout.layout.fl_stripe_count =
163			cpu_to_le32(l.stripe_count);
164	req->r_args.setlayout.layout.fl_object_size =
165			cpu_to_le32(l.object_size);
166	req->r_args.setlayout.layout.fl_pg_pool =
167			cpu_to_le32(l.data_pool);
168
169	err = ceph_mdsc_do_request(mdsc, inode, req);
170	ceph_mdsc_put_request(req);
171	return err;
172}
173
174/*
175 * Return object name, size/offset information, and location (OSD
176 * number, network address) for a given file offset.
177 */
178static long ceph_ioctl_get_dataloc(struct file *file, void __user *arg)
179{
180	struct ceph_ioctl_dataloc dl;
181	struct inode *inode = file_inode(file);
182	struct ceph_inode_info *ci = ceph_inode(inode);
183	struct ceph_osd_client *osdc =
184		&ceph_sb_to_client(inode->i_sb)->client->osdc;
185	struct ceph_object_locator oloc;
186	struct ceph_object_id oid;
187	u64 len = 1, olen;
188	u64 tmp;
189	struct ceph_pg pgid;
190	int r;
191
192	/* copy and validate */
193	if (copy_from_user(&dl, arg, sizeof(dl)))
194		return -EFAULT;
195
196	down_read(&osdc->map_sem);
197	r = ceph_calc_file_object_mapping(&ci->i_layout, dl.file_offset, len,
198					  &dl.object_no, &dl.object_offset,
199					  &olen);
200	if (r < 0) {
201		up_read(&osdc->map_sem);
202		return -EIO;
203	}
204	dl.file_offset -= dl.object_offset;
205	dl.object_size = ceph_file_layout_object_size(ci->i_layout);
206	dl.block_size = ceph_file_layout_su(ci->i_layout);
207
208	/* block_offset = object_offset % block_size */
209	tmp = dl.object_offset;
210	dl.block_offset = do_div(tmp, dl.block_size);
211
212	snprintf(dl.object_name, sizeof(dl.object_name), "%llx.%08llx",
213		 ceph_ino(inode), dl.object_no);
214
215	oloc.pool = ceph_file_layout_pg_pool(ci->i_layout);
216	ceph_oid_set_name(&oid, dl.object_name);
217
218	r = ceph_oloc_oid_to_pg(osdc->osdmap, &oloc, &oid, &pgid);
219	if (r < 0) {
220		up_read(&osdc->map_sem);
221		return r;
222	}
223
224	dl.osd = ceph_calc_pg_primary(osdc->osdmap, pgid);
225	if (dl.osd >= 0) {
226		struct ceph_entity_addr *a =
227			ceph_osd_addr(osdc->osdmap, dl.osd);
228		if (a)
229			memcpy(&dl.osd_addr, &a->in_addr, sizeof(dl.osd_addr));
230	} else {
231		memset(&dl.osd_addr, 0, sizeof(dl.osd_addr));
232	}
233	up_read(&osdc->map_sem);
234
235	/* send result back to user */
236	if (copy_to_user(arg, &dl, sizeof(dl)))
237		return -EFAULT;
238
239	return 0;
240}
241
242static long ceph_ioctl_lazyio(struct file *file)
243{
244	struct ceph_file_info *fi = file->private_data;
245	struct inode *inode = file_inode(file);
246	struct ceph_inode_info *ci = ceph_inode(inode);
247
248	if ((fi->fmode & CEPH_FILE_MODE_LAZY) == 0) {
249		spin_lock(&ci->i_ceph_lock);
250		ci->i_nr_by_mode[fi->fmode]--;
251		fi->fmode |= CEPH_FILE_MODE_LAZY;
252		ci->i_nr_by_mode[fi->fmode]++;
253		spin_unlock(&ci->i_ceph_lock);
254		dout("ioctl_layzio: file %p marked lazy\n", file);
255
256		ceph_check_caps(ci, 0, NULL);
257	} else {
258		dout("ioctl_layzio: file %p already lazy\n", file);
259	}
260	return 0;
261}
262
263static long ceph_ioctl_syncio(struct file *file)
264{
265	struct ceph_file_info *fi = file->private_data;
266
267	fi->flags |= CEPH_F_SYNC;
268	return 0;
269}
270
271long ceph_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
272{
273	dout("ioctl file %p cmd %u arg %lu\n", file, cmd, arg);
274	switch (cmd) {
275	case CEPH_IOC_GET_LAYOUT:
276		return ceph_ioctl_get_layout(file, (void __user *)arg);
277
278	case CEPH_IOC_SET_LAYOUT:
279		return ceph_ioctl_set_layout(file, (void __user *)arg);
280
281	case CEPH_IOC_SET_LAYOUT_POLICY:
282		return ceph_ioctl_set_layout_policy(file, (void __user *)arg);
283
284	case CEPH_IOC_GET_DATALOC:
285		return ceph_ioctl_get_dataloc(file, (void __user *)arg);
286
287	case CEPH_IOC_LAZYIO:
288		return ceph_ioctl_lazyio(file);
289
290	case CEPH_IOC_SYNCIO:
291		return ceph_ioctl_syncio(file);
292	}
293
294	return -ENOTTY;
295}
296