ceph: add CEPH_MDS_OP_SETDIRLAYOUT and associated ioctl.
[linux-2.6.git] / fs / ceph / ioctl.c
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  */
17 static 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 =
30                         (s32)le32_to_cpu(ci->i_layout.fl_pg_preferred);
31                 if (copy_to_user(arg, &l, sizeof(l)))
32                         return -EFAULT;
33         }
34
35         return err;
36 }
37
38 static long ceph_ioctl_set_layout(struct file *file, void __user *arg)
39 {
40         struct inode *inode = file->f_dentry->d_inode;
41         struct inode *parent_inode = file->f_dentry->d_parent->d_inode;
42         struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
43         struct ceph_mds_request *req;
44         struct ceph_ioctl_layout l;
45         int err, i;
46
47         /* copy and validate */
48         if (copy_from_user(&l, arg, sizeof(l)))
49                 return -EFAULT;
50
51         if ((l.object_size & ~PAGE_MASK) ||
52             (l.stripe_unit & ~PAGE_MASK) ||
53             !l.stripe_unit ||
54             (l.object_size &&
55              (unsigned)l.object_size % (unsigned)l.stripe_unit))
56                 return -EINVAL;
57
58         /* make sure it's a valid data pool */
59         if (l.data_pool > 0) {
60                 mutex_lock(&mdsc->mutex);
61                 err = -EINVAL;
62                 for (i = 0; i < mdsc->mdsmap->m_num_data_pg_pools; i++)
63                         if (mdsc->mdsmap->m_data_pg_pools[i] == l.data_pool) {
64                                 err = 0;
65                                 break;
66                         }
67                 mutex_unlock(&mdsc->mutex);
68                 if (err)
69                         return err;
70         }
71
72         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETLAYOUT,
73                                        USE_AUTH_MDS);
74         if (IS_ERR(req))
75                 return PTR_ERR(req);
76         req->r_inode = igrab(inode);
77         req->r_inode_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL;
78
79         req->r_args.setlayout.layout.fl_stripe_unit =
80                 cpu_to_le32(l.stripe_unit);
81         req->r_args.setlayout.layout.fl_stripe_count =
82                 cpu_to_le32(l.stripe_count);
83         req->r_args.setlayout.layout.fl_object_size =
84                 cpu_to_le32(l.object_size);
85         req->r_args.setlayout.layout.fl_pg_pool = cpu_to_le32(l.data_pool);
86         req->r_args.setlayout.layout.fl_pg_preferred =
87                 cpu_to_le32(l.preferred_osd);
88
89         err = ceph_mdsc_do_request(mdsc, parent_inode, req);
90         ceph_mdsc_put_request(req);
91         return err;
92 }
93
94 /*
95  * Set a layout policy on a directory inode. All items in the tree
96  * rooted at this inode will inherit this layout on creation,
97  * (It doesn't apply retroactively )
98  * unless a subdirectory has its own layout policy.
99  */
100 static long ceph_ioctl_set_layout_policy (struct file *file, void __user *arg)
101 {
102         struct inode *inode = file->f_dentry->d_inode;
103         struct ceph_mds_request *req;
104         struct ceph_ioctl_layout l;
105         int err, i;
106         struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
107
108         /* copy and validate */
109         if (copy_from_user(&l, arg, sizeof(l)))
110                 return -EFAULT;
111
112         if ((l.object_size & ~PAGE_MASK) ||
113             (l.stripe_unit & ~PAGE_MASK) ||
114             !l.stripe_unit ||
115             (l.object_size &&
116                 (unsigned)l.object_size % (unsigned)l.stripe_unit))
117                 return -EINVAL;
118
119         /* make sure it's a valid data pool */
120         if (l.data_pool > 0) {
121                 mutex_lock(&mdsc->mutex);
122                 err = -EINVAL;
123                 for (i = 0; i < mdsc->mdsmap->m_num_data_pg_pools; i++)
124                         if (mdsc->mdsmap->m_data_pg_pools[i] == l.data_pool) {
125                                 err = 0;
126                                 break;
127                         }
128                 mutex_unlock(&mdsc->mutex);
129                 if (err)
130                         return err;
131         }
132
133         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETDIRLAYOUT,
134                                        USE_AUTH_MDS);
135
136         if (IS_ERR(req))
137                 return PTR_ERR(req);
138         req->r_inode = igrab(inode);
139
140         req->r_args.setlayout.layout.fl_stripe_unit =
141                         cpu_to_le32(l.stripe_unit);
142         req->r_args.setlayout.layout.fl_stripe_count =
143                         cpu_to_le32(l.stripe_count);
144         req->r_args.setlayout.layout.fl_object_size =
145                         cpu_to_le32(l.object_size);
146         req->r_args.setlayout.layout.fl_pg_pool =
147                         cpu_to_le32(l.data_pool);
148         req->r_args.setlayout.layout.fl_pg_preferred =
149                         cpu_to_le32(l.preferred_osd);
150
151         err = ceph_mdsc_do_request(mdsc, inode, req);
152         ceph_mdsc_put_request(req);
153         return err;
154 }
155
156 /*
157  * Return object name, size/offset information, and location (OSD
158  * number, network address) for a given file offset.
159  */
160 static long ceph_ioctl_get_dataloc(struct file *file, void __user *arg)
161 {
162         struct ceph_ioctl_dataloc dl;
163         struct inode *inode = file->f_dentry->d_inode;
164         struct ceph_inode_info *ci = ceph_inode(inode);
165         struct ceph_osd_client *osdc =
166                 &ceph_sb_to_client(inode->i_sb)->client->osdc;
167         u64 len = 1, olen;
168         u64 tmp;
169         struct ceph_object_layout ol;
170         struct ceph_pg pgid;
171
172         /* copy and validate */
173         if (copy_from_user(&dl, arg, sizeof(dl)))
174                 return -EFAULT;
175
176         down_read(&osdc->map_sem);
177         ceph_calc_file_object_mapping(&ci->i_layout, dl.file_offset, &len,
178                                       &dl.object_no, &dl.object_offset, &olen);
179         dl.file_offset -= dl.object_offset;
180         dl.object_size = ceph_file_layout_object_size(ci->i_layout);
181         dl.block_size = ceph_file_layout_su(ci->i_layout);
182
183         /* block_offset = object_offset % block_size */
184         tmp = dl.object_offset;
185         dl.block_offset = do_div(tmp, dl.block_size);
186
187         snprintf(dl.object_name, sizeof(dl.object_name), "%llx.%08llx",
188                  ceph_ino(inode), dl.object_no);
189         ceph_calc_object_layout(&ol, dl.object_name, &ci->i_layout,
190                                 osdc->osdmap);
191
192         pgid = ol.ol_pgid;
193         dl.osd = ceph_calc_pg_primary(osdc->osdmap, pgid);
194         if (dl.osd >= 0) {
195                 struct ceph_entity_addr *a =
196                         ceph_osd_addr(osdc->osdmap, dl.osd);
197                 if (a)
198                         memcpy(&dl.osd_addr, &a->in_addr, sizeof(dl.osd_addr));
199         } else {
200                 memset(&dl.osd_addr, 0, sizeof(dl.osd_addr));
201         }
202         up_read(&osdc->map_sem);
203
204         /* send result back to user */
205         if (copy_to_user(arg, &dl, sizeof(dl)))
206                 return -EFAULT;
207
208         return 0;
209 }
210
211 static long ceph_ioctl_lazyio(struct file *file)
212 {
213         struct ceph_file_info *fi = file->private_data;
214         struct inode *inode = file->f_dentry->d_inode;
215         struct ceph_inode_info *ci = ceph_inode(inode);
216
217         if ((fi->fmode & CEPH_FILE_MODE_LAZY) == 0) {
218                 spin_lock(&inode->i_lock);
219                 ci->i_nr_by_mode[fi->fmode]--;
220                 fi->fmode |= CEPH_FILE_MODE_LAZY;
221                 ci->i_nr_by_mode[fi->fmode]++;
222                 spin_unlock(&inode->i_lock);
223                 dout("ioctl_layzio: file %p marked lazy\n", file);
224
225                 ceph_check_caps(ci, 0, NULL);
226         } else {
227                 dout("ioctl_layzio: file %p already lazy\n", file);
228         }
229         return 0;
230 }
231
232 long ceph_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
233 {
234         dout("ioctl file %p cmd %u arg %lu\n", file, cmd, arg);
235         switch (cmd) {
236         case CEPH_IOC_GET_LAYOUT:
237                 return ceph_ioctl_get_layout(file, (void __user *)arg);
238
239         case CEPH_IOC_SET_LAYOUT:
240                 return ceph_ioctl_set_layout(file, (void __user *)arg);
241
242         case CEPH_IOC_SET_LAYOUT_POLICY:
243                 return ceph_ioctl_set_layout_policy(file, (void __user *)arg);
244
245         case CEPH_IOC_GET_DATALOC:
246                 return ceph_ioctl_get_dataloc(file, (void __user *)arg);
247
248         case CEPH_IOC_LAZYIO:
249                 return ceph_ioctl_lazyio(file);
250         }
251
252         return -ENOTTY;
253 }