ioctl.c 7.8 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291
  1. #include <linux/in.h>
  2. #include "super.h"
  3. #include "mds_client.h"
  4. #include <linux/ceph/ceph_debug.h>
  5. #include "ioctl.h"
  6. /*
  7. * ioctls
  8. */
  9. /*
  10. * get and set the file layout
  11. */
  12. static long ceph_ioctl_get_layout(struct file *file, void __user *arg)
  13. {
  14. struct ceph_inode_info *ci = ceph_inode(file->f_dentry->d_inode);
  15. struct ceph_ioctl_layout l;
  16. int err;
  17. err = ceph_do_getattr(file->f_dentry->d_inode, CEPH_STAT_CAP_LAYOUT);
  18. if (!err) {
  19. l.stripe_unit = ceph_file_layout_su(ci->i_layout);
  20. l.stripe_count = ceph_file_layout_stripe_count(ci->i_layout);
  21. l.object_size = ceph_file_layout_object_size(ci->i_layout);
  22. l.data_pool = le32_to_cpu(ci->i_layout.fl_pg_pool);
  23. l.preferred_osd =
  24. (s32)le32_to_cpu(ci->i_layout.fl_pg_preferred);
  25. if (copy_to_user(arg, &l, sizeof(l)))
  26. return -EFAULT;
  27. }
  28. return err;
  29. }
  30. static long ceph_ioctl_set_layout(struct file *file, void __user *arg)
  31. {
  32. struct inode *inode = file->f_dentry->d_inode;
  33. struct inode *parent_inode;
  34. struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
  35. struct ceph_mds_request *req;
  36. struct ceph_ioctl_layout l;
  37. struct ceph_inode_info *ci = ceph_inode(file->f_dentry->d_inode);
  38. struct ceph_ioctl_layout nl;
  39. int err, i;
  40. if (copy_from_user(&l, arg, sizeof(l)))
  41. return -EFAULT;
  42. /* validate changed params against current layout */
  43. err = ceph_do_getattr(file->f_dentry->d_inode, CEPH_STAT_CAP_LAYOUT);
  44. if (!err) {
  45. nl.stripe_unit = ceph_file_layout_su(ci->i_layout);
  46. nl.stripe_count = ceph_file_layout_stripe_count(ci->i_layout);
  47. nl.object_size = ceph_file_layout_object_size(ci->i_layout);
  48. nl.data_pool = le32_to_cpu(ci->i_layout.fl_pg_pool);
  49. nl.preferred_osd =
  50. (s32)le32_to_cpu(ci->i_layout.fl_pg_preferred);
  51. } else
  52. return err;
  53. if (l.stripe_count)
  54. nl.stripe_count = l.stripe_count;
  55. if (l.stripe_unit)
  56. nl.stripe_unit = l.stripe_unit;
  57. if (l.object_size)
  58. nl.object_size = l.object_size;
  59. if (l.data_pool)
  60. nl.data_pool = l.data_pool;
  61. if (l.preferred_osd)
  62. nl.preferred_osd = l.preferred_osd;
  63. if ((nl.object_size & ~PAGE_MASK) ||
  64. (nl.stripe_unit & ~PAGE_MASK) ||
  65. ((unsigned)nl.object_size % (unsigned)nl.stripe_unit))
  66. return -EINVAL;
  67. /* make sure it's a valid data pool */
  68. if (l.data_pool > 0) {
  69. mutex_lock(&mdsc->mutex);
  70. err = -EINVAL;
  71. for (i = 0; i < mdsc->mdsmap->m_num_data_pg_pools; i++)
  72. if (mdsc->mdsmap->m_data_pg_pools[i] == l.data_pool) {
  73. err = 0;
  74. break;
  75. }
  76. mutex_unlock(&mdsc->mutex);
  77. if (err)
  78. return err;
  79. }
  80. req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETLAYOUT,
  81. USE_AUTH_MDS);
  82. if (IS_ERR(req))
  83. return PTR_ERR(req);
  84. req->r_inode = inode;
  85. ihold(inode);
  86. req->r_inode_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL;
  87. req->r_args.setlayout.layout.fl_stripe_unit =
  88. cpu_to_le32(l.stripe_unit);
  89. req->r_args.setlayout.layout.fl_stripe_count =
  90. cpu_to_le32(l.stripe_count);
  91. req->r_args.setlayout.layout.fl_object_size =
  92. cpu_to_le32(l.object_size);
  93. req->r_args.setlayout.layout.fl_pg_pool = cpu_to_le32(l.data_pool);
  94. req->r_args.setlayout.layout.fl_pg_preferred =
  95. cpu_to_le32(l.preferred_osd);
  96. parent_inode = ceph_get_dentry_parent_inode(file->f_dentry);
  97. err = ceph_mdsc_do_request(mdsc, parent_inode, req);
  98. iput(parent_inode);
  99. ceph_mdsc_put_request(req);
  100. return err;
  101. }
  102. /*
  103. * Set a layout policy on a directory inode. All items in the tree
  104. * rooted at this inode will inherit this layout on creation,
  105. * (It doesn't apply retroactively )
  106. * unless a subdirectory has its own layout policy.
  107. */
  108. static long ceph_ioctl_set_layout_policy (struct file *file, void __user *arg)
  109. {
  110. struct inode *inode = file->f_dentry->d_inode;
  111. struct ceph_mds_request *req;
  112. struct ceph_ioctl_layout l;
  113. int err, i;
  114. struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
  115. /* copy and validate */
  116. if (copy_from_user(&l, arg, sizeof(l)))
  117. return -EFAULT;
  118. if ((l.object_size & ~PAGE_MASK) ||
  119. (l.stripe_unit & ~PAGE_MASK) ||
  120. !l.stripe_unit ||
  121. (l.object_size &&
  122. (unsigned)l.object_size % (unsigned)l.stripe_unit))
  123. return -EINVAL;
  124. /* make sure it's a valid data pool */
  125. if (l.data_pool > 0) {
  126. mutex_lock(&mdsc->mutex);
  127. err = -EINVAL;
  128. for (i = 0; i < mdsc->mdsmap->m_num_data_pg_pools; i++)
  129. if (mdsc->mdsmap->m_data_pg_pools[i] == l.data_pool) {
  130. err = 0;
  131. break;
  132. }
  133. mutex_unlock(&mdsc->mutex);
  134. if (err)
  135. return err;
  136. }
  137. req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETDIRLAYOUT,
  138. USE_AUTH_MDS);
  139. if (IS_ERR(req))
  140. return PTR_ERR(req);
  141. req->r_inode = inode;
  142. ihold(inode);
  143. req->r_args.setlayout.layout.fl_stripe_unit =
  144. cpu_to_le32(l.stripe_unit);
  145. req->r_args.setlayout.layout.fl_stripe_count =
  146. cpu_to_le32(l.stripe_count);
  147. req->r_args.setlayout.layout.fl_object_size =
  148. cpu_to_le32(l.object_size);
  149. req->r_args.setlayout.layout.fl_pg_pool =
  150. cpu_to_le32(l.data_pool);
  151. req->r_args.setlayout.layout.fl_pg_preferred =
  152. cpu_to_le32(l.preferred_osd);
  153. err = ceph_mdsc_do_request(mdsc, inode, req);
  154. ceph_mdsc_put_request(req);
  155. return err;
  156. }
  157. /*
  158. * Return object name, size/offset information, and location (OSD
  159. * number, network address) for a given file offset.
  160. */
  161. static long ceph_ioctl_get_dataloc(struct file *file, void __user *arg)
  162. {
  163. struct ceph_ioctl_dataloc dl;
  164. struct inode *inode = file->f_dentry->d_inode;
  165. struct ceph_inode_info *ci = ceph_inode(inode);
  166. struct ceph_osd_client *osdc =
  167. &ceph_sb_to_client(inode->i_sb)->client->osdc;
  168. u64 len = 1, olen;
  169. u64 tmp;
  170. struct ceph_object_layout ol;
  171. struct ceph_pg pgid;
  172. /* copy and validate */
  173. if (copy_from_user(&dl, arg, sizeof(dl)))
  174. return -EFAULT;
  175. down_read(&osdc->map_sem);
  176. ceph_calc_file_object_mapping(&ci->i_layout, dl.file_offset, &len,
  177. &dl.object_no, &dl.object_offset, &olen);
  178. dl.file_offset -= dl.object_offset;
  179. dl.object_size = ceph_file_layout_object_size(ci->i_layout);
  180. dl.block_size = ceph_file_layout_su(ci->i_layout);
  181. /* block_offset = object_offset % block_size */
  182. tmp = dl.object_offset;
  183. dl.block_offset = do_div(tmp, dl.block_size);
  184. snprintf(dl.object_name, sizeof(dl.object_name), "%llx.%08llx",
  185. ceph_ino(inode), dl.object_no);
  186. ceph_calc_object_layout(&ol, dl.object_name, &ci->i_layout,
  187. osdc->osdmap);
  188. pgid = ol.ol_pgid;
  189. dl.osd = ceph_calc_pg_primary(osdc->osdmap, pgid);
  190. if (dl.osd >= 0) {
  191. struct ceph_entity_addr *a =
  192. ceph_osd_addr(osdc->osdmap, dl.osd);
  193. if (a)
  194. memcpy(&dl.osd_addr, &a->in_addr, sizeof(dl.osd_addr));
  195. } else {
  196. memset(&dl.osd_addr, 0, sizeof(dl.osd_addr));
  197. }
  198. up_read(&osdc->map_sem);
  199. /* send result back to user */
  200. if (copy_to_user(arg, &dl, sizeof(dl)))
  201. return -EFAULT;
  202. return 0;
  203. }
  204. static long ceph_ioctl_lazyio(struct file *file)
  205. {
  206. struct ceph_file_info *fi = file->private_data;
  207. struct inode *inode = file->f_dentry->d_inode;
  208. struct ceph_inode_info *ci = ceph_inode(inode);
  209. if ((fi->fmode & CEPH_FILE_MODE_LAZY) == 0) {
  210. spin_lock(&ci->i_ceph_lock);
  211. ci->i_nr_by_mode[fi->fmode]--;
  212. fi->fmode |= CEPH_FILE_MODE_LAZY;
  213. ci->i_nr_by_mode[fi->fmode]++;
  214. spin_unlock(&ci->i_ceph_lock);
  215. dout("ioctl_layzio: file %p marked lazy\n", file);
  216. ceph_check_caps(ci, 0, NULL);
  217. } else {
  218. dout("ioctl_layzio: file %p already lazy\n", file);
  219. }
  220. return 0;
  221. }
  222. static long ceph_ioctl_syncio(struct file *file)
  223. {
  224. struct ceph_file_info *fi = file->private_data;
  225. fi->flags |= CEPH_F_SYNC;
  226. return 0;
  227. }
  228. long ceph_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
  229. {
  230. dout("ioctl file %p cmd %u arg %lu\n", file, cmd, arg);
  231. switch (cmd) {
  232. case CEPH_IOC_GET_LAYOUT:
  233. return ceph_ioctl_get_layout(file, (void __user *)arg);
  234. case CEPH_IOC_SET_LAYOUT:
  235. return ceph_ioctl_set_layout(file, (void __user *)arg);
  236. case CEPH_IOC_SET_LAYOUT_POLICY:
  237. return ceph_ioctl_set_layout_policy(file, (void __user *)arg);
  238. case CEPH_IOC_GET_DATALOC:
  239. return ceph_ioctl_get_dataloc(file, (void __user *)arg);
  240. case CEPH_IOC_LAZYIO:
  241. return ceph_ioctl_lazyio(file);
  242. case CEPH_IOC_SYNCIO:
  243. return ceph_ioctl_syncio(file);
  244. }
  245. return -ENOTTY;
  246. }