export.c 6.9 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274
  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, int fh_len)
  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. if (fh_len < sizeof(*fh) / 4)
  101. return ERR_PTR(-ESTALE);
  102. dout("__fh_to_dentry %llx\n", fh->ino);
  103. vino.ino = fh->ino;
  104. vino.snap = CEPH_NOSNAP;
  105. inode = ceph_find_inode(sb, vino);
  106. if (!inode) {
  107. struct ceph_mds_request *req;
  108. req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPINO,
  109. USE_ANY_MDS);
  110. if (IS_ERR(req))
  111. return ERR_CAST(req);
  112. req->r_ino1 = vino;
  113. req->r_num_caps = 1;
  114. err = ceph_mdsc_do_request(mdsc, NULL, req);
  115. inode = req->r_target_inode;
  116. if (inode)
  117. ihold(inode);
  118. ceph_mdsc_put_request(req);
  119. if (!inode)
  120. return ERR_PTR(-ESTALE);
  121. }
  122. dentry = d_obtain_alias(inode);
  123. if (IS_ERR(dentry)) {
  124. pr_err("fh_to_dentry %llx -- inode %p but ENOMEM\n",
  125. fh->ino, inode);
  126. iput(inode);
  127. return dentry;
  128. }
  129. err = ceph_init_dentry(dentry);
  130. if (err < 0) {
  131. iput(inode);
  132. return ERR_PTR(err);
  133. }
  134. dout("__fh_to_dentry %llx %p dentry %p\n", fh->ino, inode, dentry);
  135. return dentry;
  136. }
  137. /*
  138. * convert connectable fh to dentry
  139. */
  140. static struct dentry *__cfh_to_dentry(struct super_block *sb,
  141. struct ceph_nfs_confh *cfh, int fh_len)
  142. {
  143. struct ceph_mds_client *mdsc = ceph_sb_to_client(sb)->mdsc;
  144. struct inode *inode;
  145. struct dentry *dentry;
  146. struct ceph_vino vino;
  147. int err;
  148. if (fh_len < sizeof(*cfh) / 4)
  149. return ERR_PTR(-ESTALE);
  150. dout("__cfh_to_dentry %llx (%llx/%x)\n",
  151. cfh->ino, cfh->parent_ino, cfh->parent_name_hash);
  152. vino.ino = cfh->ino;
  153. vino.snap = CEPH_NOSNAP;
  154. inode = ceph_find_inode(sb, vino);
  155. if (!inode) {
  156. struct ceph_mds_request *req;
  157. req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPHASH,
  158. USE_ANY_MDS);
  159. if (IS_ERR(req))
  160. return ERR_CAST(req);
  161. req->r_ino1 = vino;
  162. req->r_ino2.ino = cfh->parent_ino;
  163. req->r_ino2.snap = CEPH_NOSNAP;
  164. req->r_path2 = kmalloc(16, GFP_NOFS);
  165. snprintf(req->r_path2, 16, "%d", cfh->parent_name_hash);
  166. req->r_num_caps = 1;
  167. err = ceph_mdsc_do_request(mdsc, NULL, req);
  168. inode = req->r_target_inode;
  169. if (inode)
  170. ihold(inode);
  171. ceph_mdsc_put_request(req);
  172. if (!inode)
  173. return ERR_PTR(err ? err : -ESTALE);
  174. }
  175. dentry = d_obtain_alias(inode);
  176. if (IS_ERR(dentry)) {
  177. pr_err("cfh_to_dentry %llx -- inode %p but ENOMEM\n",
  178. cfh->ino, inode);
  179. iput(inode);
  180. return dentry;
  181. }
  182. err = ceph_init_dentry(dentry);
  183. if (err < 0) {
  184. iput(inode);
  185. return ERR_PTR(err);
  186. }
  187. dout("__cfh_to_dentry %llx %p dentry %p\n", cfh->ino, inode, dentry);
  188. return dentry;
  189. }
  190. static struct dentry *ceph_fh_to_dentry(struct super_block *sb, struct fid *fid,
  191. int fh_len, int fh_type)
  192. {
  193. if (fh_type == 1)
  194. return __fh_to_dentry(sb, (struct ceph_nfs_fh *)fid->raw,
  195. fh_len);
  196. else
  197. return __cfh_to_dentry(sb, (struct ceph_nfs_confh *)fid->raw,
  198. fh_len);
  199. }
  200. /*
  201. * get parent, if possible.
  202. *
  203. * FIXME: we could do better by querying the mds to discover the
  204. * parent.
  205. */
  206. static struct dentry *ceph_fh_to_parent(struct super_block *sb,
  207. struct fid *fid,
  208. int fh_len, int fh_type)
  209. {
  210. struct ceph_nfs_confh *cfh = (void *)fid->raw;
  211. struct ceph_vino vino;
  212. struct inode *inode;
  213. struct dentry *dentry;
  214. int err;
  215. if (fh_type == 1)
  216. return ERR_PTR(-ESTALE);
  217. if (fh_len < sizeof(*cfh) / 4)
  218. return ERR_PTR(-ESTALE);
  219. pr_debug("fh_to_parent %llx/%d\n", cfh->parent_ino,
  220. cfh->parent_name_hash);
  221. vino.ino = cfh->ino;
  222. vino.snap = CEPH_NOSNAP;
  223. inode = ceph_find_inode(sb, vino);
  224. if (!inode)
  225. return ERR_PTR(-ESTALE);
  226. dentry = d_obtain_alias(inode);
  227. if (IS_ERR(dentry)) {
  228. pr_err("fh_to_parent %llx -- inode %p but ENOMEM\n",
  229. cfh->ino, inode);
  230. iput(inode);
  231. return dentry;
  232. }
  233. err = ceph_init_dentry(dentry);
  234. if (err < 0) {
  235. iput(inode);
  236. return ERR_PTR(err);
  237. }
  238. dout("fh_to_parent %llx %p dentry %p\n", cfh->ino, inode, dentry);
  239. return dentry;
  240. }
  241. const struct export_operations ceph_export_ops = {
  242. .encode_fh = ceph_encode_fh,
  243. .fh_to_dentry = ceph_fh_to_dentry,
  244. .fh_to_parent = ceph_fh_to_parent,
  245. };