Loading...
1#include <linux/in.h>
2
3#include "super.h"
4#include "mds_client.h"
5#include <linux/ceph/ceph_debug.h>
6
7#include "ioctl.h"
8
9
10/*
11 * ioctls
12 */
13
14/*
15 * get and set the file layout
16 */
17static long ceph_ioctl_get_layout(struct file *file, void __user *arg)
18{
19 struct ceph_inode_info *ci = ceph_inode(file->f_dentry->d_inode);
20 struct ceph_ioctl_layout l;
21 int err;
22
23 err = ceph_do_getattr(file->f_dentry->d_inode, CEPH_STAT_CAP_LAYOUT);
24 if (!err) {
25 l.stripe_unit = ceph_file_layout_su(ci->i_layout);
26 l.stripe_count = ceph_file_layout_stripe_count(ci->i_layout);
27 l.object_size = ceph_file_layout_object_size(ci->i_layout);
28 l.data_pool = le32_to_cpu(ci->i_layout.fl_pg_pool);
29 l.preferred_osd = (s32)-1;
30 if (copy_to_user(arg, &l, sizeof(l)))
31 return -EFAULT;
32 }
33
34 return err;
35}
36
37static long __validate_layout(struct ceph_mds_client *mdsc,
38 struct ceph_ioctl_layout *l)
39{
40 int i, err;
41
42 /* validate striping parameters */
43 if ((l->object_size & ~PAGE_MASK) ||
44 (l->stripe_unit & ~PAGE_MASK) ||
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->f_dentry->d_inode;
66 struct inode *parent_inode;
67 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
68 struct ceph_mds_request *req;
69 struct ceph_ioctl_layout l;
70 struct ceph_inode_info *ci = ceph_inode(file->f_dentry->d_inode);
71 struct ceph_ioctl_layout nl;
72 int err;
73
74 if (copy_from_user(&l, arg, sizeof(l)))
75 return -EFAULT;
76
77 /* validate changed params against current layout */
78 err = ceph_do_getattr(file->f_dentry->d_inode, CEPH_STAT_CAP_LAYOUT);
79 if (err)
80 return err;
81
82 memset(&nl, 0, sizeof(nl));
83 if (l.stripe_count)
84 nl.stripe_count = l.stripe_count;
85 else
86 nl.stripe_count = ceph_file_layout_stripe_count(ci->i_layout);
87 if (l.stripe_unit)
88 nl.stripe_unit = l.stripe_unit;
89 else
90 nl.stripe_unit = ceph_file_layout_su(ci->i_layout);
91 if (l.object_size)
92 nl.object_size = l.object_size;
93 else
94 nl.object_size = ceph_file_layout_object_size(ci->i_layout);
95 if (l.data_pool)
96 nl.data_pool = l.data_pool;
97 else
98 nl.data_pool = ceph_file_layout_pg_pool(ci->i_layout);
99
100 /* this is obsolete, and always -1 */
101 nl.preferred_osd = le64_to_cpu(-1);
102
103 err = __validate_layout(mdsc, &nl);
104 if (err)
105 return err;
106
107 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETLAYOUT,
108 USE_AUTH_MDS);
109 if (IS_ERR(req))
110 return PTR_ERR(req);
111 req->r_inode = inode;
112 ihold(inode);
113 req->r_inode_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL;
114
115 req->r_args.setlayout.layout.fl_stripe_unit =
116 cpu_to_le32(l.stripe_unit);
117 req->r_args.setlayout.layout.fl_stripe_count =
118 cpu_to_le32(l.stripe_count);
119 req->r_args.setlayout.layout.fl_object_size =
120 cpu_to_le32(l.object_size);
121 req->r_args.setlayout.layout.fl_pg_pool = cpu_to_le32(l.data_pool);
122
123 parent_inode = ceph_get_dentry_parent_inode(file->f_dentry);
124 err = ceph_mdsc_do_request(mdsc, parent_inode, req);
125 iput(parent_inode);
126 ceph_mdsc_put_request(req);
127 return err;
128}
129
130/*
131 * Set a layout policy on a directory inode. All items in the tree
132 * rooted at this inode will inherit this layout on creation,
133 * (It doesn't apply retroactively )
134 * unless a subdirectory has its own layout policy.
135 */
136static long ceph_ioctl_set_layout_policy (struct file *file, void __user *arg)
137{
138 struct inode *inode = file->f_dentry->d_inode;
139 struct ceph_mds_request *req;
140 struct ceph_ioctl_layout l;
141 int err;
142 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
143
144 /* copy and validate */
145 if (copy_from_user(&l, arg, sizeof(l)))
146 return -EFAULT;
147
148 err = __validate_layout(mdsc, &l);
149 if (err)
150 return err;
151
152 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETDIRLAYOUT,
153 USE_AUTH_MDS);
154
155 if (IS_ERR(req))
156 return PTR_ERR(req);
157 req->r_inode = inode;
158 ihold(inode);
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->f_dentry->d_inode;
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 u64 len = 1, olen;
186 u64 tmp;
187 struct ceph_object_layout ol;
188 struct ceph_pg pgid;
189
190 /* copy and validate */
191 if (copy_from_user(&dl, arg, sizeof(dl)))
192 return -EFAULT;
193
194 down_read(&osdc->map_sem);
195 ceph_calc_file_object_mapping(&ci->i_layout, dl.file_offset, &len,
196 &dl.object_no, &dl.object_offset, &olen);
197 dl.file_offset -= dl.object_offset;
198 dl.object_size = ceph_file_layout_object_size(ci->i_layout);
199 dl.block_size = ceph_file_layout_su(ci->i_layout);
200
201 /* block_offset = object_offset % block_size */
202 tmp = dl.object_offset;
203 dl.block_offset = do_div(tmp, dl.block_size);
204
205 snprintf(dl.object_name, sizeof(dl.object_name), "%llx.%08llx",
206 ceph_ino(inode), dl.object_no);
207 ceph_calc_object_layout(&ol, dl.object_name, &ci->i_layout,
208 osdc->osdmap);
209
210 pgid = ol.ol_pgid;
211 dl.osd = ceph_calc_pg_primary(osdc->osdmap, pgid);
212 if (dl.osd >= 0) {
213 struct ceph_entity_addr *a =
214 ceph_osd_addr(osdc->osdmap, dl.osd);
215 if (a)
216 memcpy(&dl.osd_addr, &a->in_addr, sizeof(dl.osd_addr));
217 } else {
218 memset(&dl.osd_addr, 0, sizeof(dl.osd_addr));
219 }
220 up_read(&osdc->map_sem);
221
222 /* send result back to user */
223 if (copy_to_user(arg, &dl, sizeof(dl)))
224 return -EFAULT;
225
226 return 0;
227}
228
229static long ceph_ioctl_lazyio(struct file *file)
230{
231 struct ceph_file_info *fi = file->private_data;
232 struct inode *inode = file->f_dentry->d_inode;
233 struct ceph_inode_info *ci = ceph_inode(inode);
234
235 if ((fi->fmode & CEPH_FILE_MODE_LAZY) == 0) {
236 spin_lock(&ci->i_ceph_lock);
237 ci->i_nr_by_mode[fi->fmode]--;
238 fi->fmode |= CEPH_FILE_MODE_LAZY;
239 ci->i_nr_by_mode[fi->fmode]++;
240 spin_unlock(&ci->i_ceph_lock);
241 dout("ioctl_layzio: file %p marked lazy\n", file);
242
243 ceph_check_caps(ci, 0, NULL);
244 } else {
245 dout("ioctl_layzio: file %p already lazy\n", file);
246 }
247 return 0;
248}
249
250static long ceph_ioctl_syncio(struct file *file)
251{
252 struct ceph_file_info *fi = file->private_data;
253
254 fi->flags |= CEPH_F_SYNC;
255 return 0;
256}
257
258long ceph_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
259{
260 dout("ioctl file %p cmd %u arg %lu\n", file, cmd, arg);
261 switch (cmd) {
262 case CEPH_IOC_GET_LAYOUT:
263 return ceph_ioctl_get_layout(file, (void __user *)arg);
264
265 case CEPH_IOC_SET_LAYOUT:
266 return ceph_ioctl_set_layout(file, (void __user *)arg);
267
268 case CEPH_IOC_SET_LAYOUT_POLICY:
269 return ceph_ioctl_set_layout_policy(file, (void __user *)arg);
270
271 case CEPH_IOC_GET_DATALOC:
272 return ceph_ioctl_get_dataloc(file, (void __user *)arg);
273
274 case CEPH_IOC_LAZYIO:
275 return ceph_ioctl_lazyio(file);
276
277 case CEPH_IOC_SYNCIO:
278 return ceph_ioctl_syncio(file);
279 }
280
281 return -ENOTTY;
282}
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 = ci->i_layout.stripe_unit;
25 l.stripe_count = ci->i_layout.stripe_count;
26 l.object_size = ci->i_layout.object_size;
27 l.data_pool = ci->i_layout.pool_id;
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 = ci->i_layout.stripe_count;
86 if (l.stripe_unit)
87 nl.stripe_unit = l.stripe_unit;
88 else
89 nl.stripe_unit = ci->i_layout.stripe_unit;
90 if (l.object_size)
91 nl.object_size = l.object_size;
92 else
93 nl.object_size = ci->i_layout.object_size;
94 if (l.data_pool)
95 nl.data_pool = l.data_pool;
96 else
97 nl.data_pool = ci->i_layout.pool_id;
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 CEPH_DEFINE_OID_ONSTACK(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->lock);
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->lock);
202 return -EIO;
203 }
204 dl.file_offset -= dl.object_offset;
205 dl.object_size = ci->i_layout.object_size;
206 dl.block_size = ci->i_layout.stripe_unit;
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 = ci->i_layout.pool_id;
216 oloc.pool_ns = ceph_try_get_string(ci->i_layout.pool_ns);
217 ceph_oid_printf(&oid, "%s", dl.object_name);
218
219 r = ceph_object_locator_to_pg(osdc->osdmap, &oid, &oloc, &pgid);
220
221 ceph_oloc_destroy(&oloc);
222 if (r < 0) {
223 up_read(&osdc->lock);
224 return r;
225 }
226
227 dl.osd = ceph_pg_to_acting_primary(osdc->osdmap, &pgid);
228 if (dl.osd >= 0) {
229 struct ceph_entity_addr *a =
230 ceph_osd_addr(osdc->osdmap, dl.osd);
231 if (a)
232 memcpy(&dl.osd_addr, &a->in_addr, sizeof(dl.osd_addr));
233 } else {
234 memset(&dl.osd_addr, 0, sizeof(dl.osd_addr));
235 }
236 up_read(&osdc->lock);
237
238 /* send result back to user */
239 if (copy_to_user(arg, &dl, sizeof(dl)))
240 return -EFAULT;
241
242 return 0;
243}
244
245static long ceph_ioctl_lazyio(struct file *file)
246{
247 struct ceph_file_info *fi = file->private_data;
248 struct inode *inode = file_inode(file);
249 struct ceph_inode_info *ci = ceph_inode(inode);
250
251 if ((fi->fmode & CEPH_FILE_MODE_LAZY) == 0) {
252 spin_lock(&ci->i_ceph_lock);
253 fi->fmode |= CEPH_FILE_MODE_LAZY;
254 ci->i_nr_by_mode[ffs(CEPH_FILE_MODE_LAZY)]++;
255 spin_unlock(&ci->i_ceph_lock);
256 dout("ioctl_layzio: file %p marked lazy\n", file);
257
258 ceph_check_caps(ci, 0, NULL);
259 } else {
260 dout("ioctl_layzio: file %p already lazy\n", file);
261 }
262 return 0;
263}
264
265static long ceph_ioctl_syncio(struct file *file)
266{
267 struct ceph_file_info *fi = file->private_data;
268
269 fi->flags |= CEPH_F_SYNC;
270 return 0;
271}
272
273long ceph_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
274{
275 dout("ioctl file %p cmd %u arg %lu\n", file, cmd, arg);
276 switch (cmd) {
277 case CEPH_IOC_GET_LAYOUT:
278 return ceph_ioctl_get_layout(file, (void __user *)arg);
279
280 case CEPH_IOC_SET_LAYOUT:
281 return ceph_ioctl_set_layout(file, (void __user *)arg);
282
283 case CEPH_IOC_SET_LAYOUT_POLICY:
284 return ceph_ioctl_set_layout_policy(file, (void __user *)arg);
285
286 case CEPH_IOC_GET_DATALOC:
287 return ceph_ioctl_get_dataloc(file, (void __user *)arg);
288
289 case CEPH_IOC_LAZYIO:
290 return ceph_ioctl_lazyio(file);
291
292 case CEPH_IOC_SYNCIO:
293 return ceph_ioctl_syncio(file);
294 }
295
296 return -ENOTTY;
297}