export.c 6.6 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263
  1. #include <linux/ceph/ceph_debug.h>
  2. #include <linux/exportfs.h>
  3. #include <linux/slab.h>
  4. #include <asm/unaligned.h>
  5. #include "super.h"
  6. #include "mds_client.h"
  7. /*
  8. * NFS export support
  9. *
  10. * NFS re-export of a ceph mount is, at present, only semireliable.
  11. * The basic issue is that the Ceph architectures doesn't lend itself
  12. * well to generating filehandles that will remain valid forever.
  13. *
  14. * So, we do our best. If you're lucky, your inode will be in the
  15. * client's cache. If it's not, and you have a connectable fh, then
  16. * the MDS server may be able to find it for you. Otherwise, you get
  17. * ESTALE.
  18. *
  19. * There are ways to this more reliable, but in the non-connectable fh
  20. * case, we won't every work perfectly, and in the connectable case,
  21. * some changes are needed on the MDS side to work better.
  22. */
  23. /*
  24. * Basic fh
  25. */
  26. struct ceph_nfs_fh {
  27. u64 ino;
  28. } __attribute__ ((packed));
  29. /*
  30. * Larger 'connectable' fh that includes parent ino and name hash.
  31. * Use this whenever possible, as it works more reliably.
  32. */
  33. struct ceph_nfs_confh {
  34. u64 ino, parent_ino;
  35. u32 parent_name_hash;
  36. } __attribute__ ((packed));
  37. /*
  38. * The presence of @parent_inode here tells us whether NFS wants a
  39. * connectable file handle. However, we want to make a connectionable
  40. * file handle unconditionally so that the MDS gets as much of a hint
  41. * as possible. That means we only use @parent_dentry to indicate
  42. * whether nfsd wants a connectable fh, and whether we should indicate
  43. * failure from a too-small @max_len.
  44. */
  45. static int ceph_encode_fh(struct inode *inode, u32 *rawfh, int *max_len,
  46. struct inode *parent_inode)
  47. {
  48. int type;
  49. struct ceph_nfs_fh *fh = (void *)rawfh;
  50. struct ceph_nfs_confh *cfh = (void *)rawfh;
  51. int connected_handle_length = sizeof(*cfh)/4;
  52. int handle_length = sizeof(*fh)/4;
  53. struct dentry *dentry = d_find_alias(inode);
  54. struct dentry *parent;
  55. /* don't re-export snaps */
  56. if (ceph_snap(inode) != CEPH_NOSNAP)
  57. return -EINVAL;
  58. /* if we found an alias, generate a connectable fh */
  59. if (*max_len >= connected_handle_length && dentry) {
  60. dout("encode_fh %p connectable\n", dentry);
  61. spin_lock(&dentry->d_lock);
  62. parent = dentry->d_parent;
  63. cfh->ino = ceph_ino(inode);
  64. cfh->parent_ino = ceph_ino(parent->d_inode);
  65. cfh->parent_name_hash = ceph_dentry_hash(parent->d_inode,
  66. dentry);
  67. *max_len = connected_handle_length;
  68. type = 2;
  69. spin_unlock(&dentry->d_lock);
  70. } else if (*max_len >= handle_length) {
  71. if (parent_inode) {
  72. /* nfsd wants connectable */
  73. *max_len = connected_handle_length;
  74. type = 255;
  75. } else {
  76. dout("encode_fh %p\n", dentry);
  77. fh->ino = ceph_ino(inode);
  78. *max_len = handle_length;
  79. type = 1;
  80. }
  81. } else {
  82. *max_len = handle_length;
  83. type = 255;
  84. }
  85. return type;
  86. }
  87. /*
  88. * convert regular fh to dentry
  89. *
  90. * FIXME: we should try harder by querying the mds for the ino.
  91. */
  92. static struct dentry *__fh_to_dentry(struct super_block *sb,
  93. struct ceph_nfs_fh *fh)
  94. {
  95. struct ceph_mds_client *mdsc = ceph_sb_to_client(sb)->mdsc;
  96. struct inode *inode;
  97. struct dentry *dentry;
  98. struct ceph_vino vino;
  99. int err;
  100. dout("__fh_to_dentry %llx\n", fh->ino);
  101. vino.ino = fh->ino;
  102. vino.snap = CEPH_NOSNAP;
  103. inode = ceph_find_inode(sb, vino);
  104. if (!inode) {
  105. struct ceph_mds_request *req;
  106. req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPINO,
  107. USE_ANY_MDS);
  108. if (IS_ERR(req))
  109. return ERR_CAST(req);
  110. req->r_ino1 = vino;
  111. req->r_num_caps = 1;
  112. err = ceph_mdsc_do_request(mdsc, NULL, req);
  113. inode = req->r_target_inode;
  114. if (inode)
  115. ihold(inode);
  116. ceph_mdsc_put_request(req);
  117. if (!inode)
  118. return ERR_PTR(-ESTALE);
  119. }
  120. dentry = d_obtain_alias(inode);
  121. if (IS_ERR(dentry)) {
  122. pr_err("fh_to_dentry %llx -- inode %p but ENOMEM\n",
  123. fh->ino, inode);
  124. iput(inode);
  125. return dentry;
  126. }
  127. err = ceph_init_dentry(dentry);
  128. if (err < 0) {
  129. iput(inode);
  130. return ERR_PTR(err);
  131. }
  132. dout("__fh_to_dentry %llx %p dentry %p\n", fh->ino, inode, dentry);
  133. return dentry;
  134. }
  135. /*
  136. * convert connectable fh to dentry
  137. */
  138. static struct dentry *__cfh_to_dentry(struct super_block *sb,
  139. struct ceph_nfs_confh *cfh)
  140. {
  141. struct ceph_mds_client *mdsc = ceph_sb_to_client(sb)->mdsc;
  142. struct inode *inode;
  143. struct dentry *dentry;
  144. struct ceph_vino vino;
  145. int err;
  146. dout("__cfh_to_dentry %llx (%llx/%x)\n",
  147. cfh->ino, cfh->parent_ino, cfh->parent_name_hash);
  148. vino.ino = cfh->ino;
  149. vino.snap = CEPH_NOSNAP;
  150. inode = ceph_find_inode(sb, vino);
  151. if (!inode) {
  152. struct ceph_mds_request *req;
  153. req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPHASH,
  154. USE_ANY_MDS);
  155. if (IS_ERR(req))
  156. return ERR_CAST(req);
  157. req->r_ino1 = vino;
  158. req->r_ino2.ino = cfh->parent_ino;
  159. req->r_ino2.snap = CEPH_NOSNAP;
  160. req->r_path2 = kmalloc(16, GFP_NOFS);
  161. snprintf(req->r_path2, 16, "%d", cfh->parent_name_hash);
  162. req->r_num_caps = 1;
  163. err = ceph_mdsc_do_request(mdsc, NULL, req);
  164. inode = req->r_target_inode;
  165. if (inode)
  166. ihold(inode);
  167. ceph_mdsc_put_request(req);
  168. if (!inode)
  169. return ERR_PTR(err ? err : -ESTALE);
  170. }
  171. dentry = d_obtain_alias(inode);
  172. if (IS_ERR(dentry)) {
  173. pr_err("cfh_to_dentry %llx -- inode %p but ENOMEM\n",
  174. cfh->ino, inode);
  175. iput(inode);
  176. return dentry;
  177. }
  178. err = ceph_init_dentry(dentry);
  179. if (err < 0) {
  180. iput(inode);
  181. return ERR_PTR(err);
  182. }
  183. dout("__cfh_to_dentry %llx %p dentry %p\n", cfh->ino, inode, dentry);
  184. return dentry;
  185. }
  186. static struct dentry *ceph_fh_to_dentry(struct super_block *sb, struct fid *fid,
  187. int fh_len, int fh_type)
  188. {
  189. if (fh_type == 1)
  190. return __fh_to_dentry(sb, (struct ceph_nfs_fh *)fid->raw);
  191. else
  192. return __cfh_to_dentry(sb, (struct ceph_nfs_confh *)fid->raw);
  193. }
  194. /*
  195. * get parent, if possible.
  196. *
  197. * FIXME: we could do better by querying the mds to discover the
  198. * parent.
  199. */
  200. static struct dentry *ceph_fh_to_parent(struct super_block *sb,
  201. struct fid *fid,
  202. int fh_len, int fh_type)
  203. {
  204. struct ceph_nfs_confh *cfh = (void *)fid->raw;
  205. struct ceph_vino vino;
  206. struct inode *inode;
  207. struct dentry *dentry;
  208. int err;
  209. if (fh_type == 1)
  210. return ERR_PTR(-ESTALE);
  211. pr_debug("fh_to_parent %llx/%d\n", cfh->parent_ino,
  212. cfh->parent_name_hash);
  213. vino.ino = cfh->ino;
  214. vino.snap = CEPH_NOSNAP;
  215. inode = ceph_find_inode(sb, vino);
  216. if (!inode)
  217. return ERR_PTR(-ESTALE);
  218. dentry = d_obtain_alias(inode);
  219. if (IS_ERR(dentry)) {
  220. pr_err("fh_to_parent %llx -- inode %p but ENOMEM\n",
  221. cfh->ino, inode);
  222. iput(inode);
  223. return dentry;
  224. }
  225. err = ceph_init_dentry(dentry);
  226. if (err < 0) {
  227. iput(inode);
  228. return ERR_PTR(err);
  229. }
  230. dout("fh_to_parent %llx %p dentry %p\n", cfh->ino, inode, dentry);
  231. return dentry;
  232. }
  233. const struct export_operations ceph_export_ops = {
  234. .encode_fh = ceph_encode_fh,
  235. .fh_to_dentry = ceph_fh_to_dentry,
  236. .fh_to_parent = ceph_fh_to_parent,
  237. };