diff options
| -rw-r--r-- | fs/ceph/export.c | 167 |
1 files changed, 32 insertions, 135 deletions
diff --git a/fs/ceph/export.c b/fs/ceph/export.c index 16796be53ca5..976d3411d5ed 100644 --- a/fs/ceph/export.c +++ b/fs/ceph/export.c | |||
| @@ -8,23 +8,6 @@ | |||
| 8 | #include "mds_client.h" | 8 | #include "mds_client.h" |
| 9 | 9 | ||
| 10 | /* | 10 | /* |
| 11 | * NFS export support | ||
| 12 | * | ||
| 13 | * NFS re-export of a ceph mount is, at present, only semireliable. | ||
| 14 | * The basic issue is that the Ceph architectures doesn't lend itself | ||
| 15 | * well to generating filehandles that will remain valid forever. | ||
| 16 | * | ||
| 17 | * So, we do our best. If you're lucky, your inode will be in the | ||
| 18 | * client's cache. If it's not, and you have a connectable fh, then | ||
| 19 | * the MDS server may be able to find it for you. Otherwise, you get | ||
| 20 | * ESTALE. | ||
| 21 | * | ||
| 22 | * There are ways to this more reliable, but in the non-connectable fh | ||
| 23 | * case, we won't every work perfectly, and in the connectable case, | ||
| 24 | * some changes are needed on the MDS side to work better. | ||
| 25 | */ | ||
| 26 | |||
| 27 | /* | ||
| 28 | * Basic fh | 11 | * Basic fh |
| 29 | */ | 12 | */ |
| 30 | struct ceph_nfs_fh { | 13 | struct ceph_nfs_fh { |
| @@ -32,22 +15,12 @@ struct ceph_nfs_fh { | |||
| 32 | } __attribute__ ((packed)); | 15 | } __attribute__ ((packed)); |
| 33 | 16 | ||
| 34 | /* | 17 | /* |
| 35 | * Larger 'connectable' fh that includes parent ino and name hash. | 18 | * Larger fh that includes parent ino. |
| 36 | * Use this whenever possible, as it works more reliably. | ||
| 37 | */ | 19 | */ |
| 38 | struct ceph_nfs_confh { | 20 | struct ceph_nfs_confh { |
| 39 | u64 ino, parent_ino; | 21 | u64 ino, parent_ino; |
| 40 | u32 parent_name_hash; | ||
| 41 | } __attribute__ ((packed)); | 22 | } __attribute__ ((packed)); |
| 42 | 23 | ||
| 43 | /* | ||
| 44 | * The presence of @parent_inode here tells us whether NFS wants a | ||
| 45 | * connectable file handle. However, we want to make a connectionable | ||
| 46 | * file handle unconditionally so that the MDS gets as much of a hint | ||
| 47 | * as possible. That means we only use @parent_dentry to indicate | ||
| 48 | * whether nfsd wants a connectable fh, and whether we should indicate | ||
| 49 | * failure from a too-small @max_len. | ||
| 50 | */ | ||
| 51 | static int ceph_encode_fh(struct inode *inode, u32 *rawfh, int *max_len, | 24 | static int ceph_encode_fh(struct inode *inode, u32 *rawfh, int *max_len, |
| 52 | struct inode *parent_inode) | 25 | struct inode *parent_inode) |
| 53 | { | 26 | { |
| @@ -56,54 +29,36 @@ static int ceph_encode_fh(struct inode *inode, u32 *rawfh, int *max_len, | |||
| 56 | struct ceph_nfs_confh *cfh = (void *)rawfh; | 29 | struct ceph_nfs_confh *cfh = (void *)rawfh; |
| 57 | int connected_handle_length = sizeof(*cfh)/4; | 30 | int connected_handle_length = sizeof(*cfh)/4; |
| 58 | int handle_length = sizeof(*fh)/4; | 31 | int handle_length = sizeof(*fh)/4; |
| 59 | struct dentry *dentry; | ||
| 60 | struct dentry *parent; | ||
| 61 | 32 | ||
| 62 | /* don't re-export snaps */ | 33 | /* don't re-export snaps */ |
| 63 | if (ceph_snap(inode) != CEPH_NOSNAP) | 34 | if (ceph_snap(inode) != CEPH_NOSNAP) |
| 64 | return -EINVAL; | 35 | return -EINVAL; |
| 65 | 36 | ||
| 66 | dentry = d_find_alias(inode); | 37 | if (parent_inode && (*max_len < connected_handle_length)) { |
| 38 | *max_len = connected_handle_length; | ||
| 39 | return FILEID_INVALID; | ||
| 40 | } else if (*max_len < handle_length) { | ||
| 41 | *max_len = handle_length; | ||
| 42 | return FILEID_INVALID; | ||
| 43 | } | ||
| 67 | 44 | ||
| 68 | /* if we found an alias, generate a connectable fh */ | 45 | if (parent_inode) { |
| 69 | if (*max_len >= connected_handle_length && dentry) { | 46 | dout("encode_fh %llx with parent %llx\n", |
| 70 | dout("encode_fh %p connectable\n", dentry); | 47 | ceph_ino(inode), ceph_ino(parent_inode)); |
| 71 | spin_lock(&dentry->d_lock); | ||
| 72 | parent = dentry->d_parent; | ||
| 73 | cfh->ino = ceph_ino(inode); | 48 | cfh->ino = ceph_ino(inode); |
| 74 | cfh->parent_ino = ceph_ino(parent->d_inode); | 49 | cfh->parent_ino = ceph_ino(parent_inode); |
| 75 | cfh->parent_name_hash = ceph_dentry_hash(parent->d_inode, | ||
| 76 | dentry); | ||
| 77 | *max_len = connected_handle_length; | 50 | *max_len = connected_handle_length; |
| 78 | type = 2; | 51 | type = FILEID_INO32_GEN_PARENT; |
| 79 | spin_unlock(&dentry->d_lock); | ||
| 80 | } else if (*max_len >= handle_length) { | ||
| 81 | if (parent_inode) { | ||
| 82 | /* nfsd wants connectable */ | ||
| 83 | *max_len = connected_handle_length; | ||
| 84 | type = FILEID_INVALID; | ||
| 85 | } else { | ||
| 86 | dout("encode_fh %p\n", dentry); | ||
| 87 | fh->ino = ceph_ino(inode); | ||
| 88 | *max_len = handle_length; | ||
| 89 | type = 1; | ||
| 90 | } | ||
| 91 | } else { | 52 | } else { |
| 53 | dout("encode_fh %llx\n", ceph_ino(inode)); | ||
| 54 | fh->ino = ceph_ino(inode); | ||
| 92 | *max_len = handle_length; | 55 | *max_len = handle_length; |
| 93 | type = FILEID_INVALID; | 56 | type = FILEID_INO32_GEN; |
| 94 | } | 57 | } |
| 95 | if (dentry) | ||
| 96 | dput(dentry); | ||
| 97 | return type; | 58 | return type; |
| 98 | } | 59 | } |
| 99 | 60 | ||
| 100 | /* | 61 | static struct dentry *__fh_to_dentry(struct super_block *sb, u64 ino) |
| 101 | * convert regular fh to dentry | ||
| 102 | * | ||
| 103 | * FIXME: we should try harder by querying the mds for the ino. | ||
| 104 | */ | ||
| 105 | static struct dentry *__fh_to_dentry(struct super_block *sb, | ||
| 106 | struct ceph_nfs_fh *fh, int fh_len) | ||
| 107 | { | 62 | { |
| 108 | struct ceph_mds_client *mdsc = ceph_sb_to_client(sb)->mdsc; | 63 | struct ceph_mds_client *mdsc = ceph_sb_to_client(sb)->mdsc; |
| 109 | struct inode *inode; | 64 | struct inode *inode; |
| @@ -111,11 +66,7 @@ static struct dentry *__fh_to_dentry(struct super_block *sb, | |||
| 111 | struct ceph_vino vino; | 66 | struct ceph_vino vino; |
| 112 | int err; | 67 | int err; |
| 113 | 68 | ||
| 114 | if (fh_len < sizeof(*fh) / 4) | 69 | vino.ino = ino; |
| 115 | return ERR_PTR(-ESTALE); | ||
| 116 | |||
| 117 | dout("__fh_to_dentry %llx\n", fh->ino); | ||
| 118 | vino.ino = fh->ino; | ||
| 119 | vino.snap = CEPH_NOSNAP; | 70 | vino.snap = CEPH_NOSNAP; |
| 120 | inode = ceph_find_inode(sb, vino); | 71 | inode = ceph_find_inode(sb, vino); |
| 121 | if (!inode) { | 72 | if (!inode) { |
| @@ -139,89 +90,35 @@ static struct dentry *__fh_to_dentry(struct super_block *sb, | |||
| 139 | 90 | ||
| 140 | dentry = d_obtain_alias(inode); | 91 | dentry = d_obtain_alias(inode); |
| 141 | if (IS_ERR(dentry)) { | 92 | if (IS_ERR(dentry)) { |
| 142 | pr_err("fh_to_dentry %llx -- inode %p but ENOMEM\n", | ||
| 143 | fh->ino, inode); | ||
| 144 | iput(inode); | 93 | iput(inode); |
| 145 | return dentry; | 94 | return dentry; |
| 146 | } | 95 | } |
| 147 | err = ceph_init_dentry(dentry); | 96 | err = ceph_init_dentry(dentry); |
| 148 | if (err < 0) { | 97 | if (err < 0) { |
| 149 | iput(inode); | 98 | dput(dentry); |
| 150 | return ERR_PTR(err); | 99 | return ERR_PTR(err); |
| 151 | } | 100 | } |
| 152 | dout("__fh_to_dentry %llx %p dentry %p\n", fh->ino, inode, dentry); | 101 | dout("__fh_to_dentry %llx %p dentry %p\n", ino, inode, dentry); |
| 153 | return dentry; | 102 | return dentry; |
| 154 | } | 103 | } |
| 155 | 104 | ||
| 156 | /* | 105 | /* |
| 157 | * convert connectable fh to dentry | 106 | * convert regular fh to dentry |
| 158 | */ | 107 | */ |
| 159 | static struct dentry *__cfh_to_dentry(struct super_block *sb, | 108 | static struct dentry *ceph_fh_to_dentry(struct super_block *sb, |
| 160 | struct ceph_nfs_confh *cfh, int fh_len) | 109 | struct fid *fid, |
| 110 | int fh_len, int fh_type) | ||
| 161 | { | 111 | { |
| 162 | struct ceph_mds_client *mdsc = ceph_sb_to_client(sb)->mdsc; | 112 | struct ceph_nfs_fh *fh = (void *)fid->raw; |
| 163 | struct inode *inode; | ||
| 164 | struct dentry *dentry; | ||
| 165 | struct ceph_vino vino; | ||
| 166 | int err; | ||
| 167 | |||
| 168 | if (fh_len < sizeof(*cfh) / 4) | ||
| 169 | return ERR_PTR(-ESTALE); | ||
| 170 | |||
| 171 | dout("__cfh_to_dentry %llx (%llx/%x)\n", | ||
| 172 | cfh->ino, cfh->parent_ino, cfh->parent_name_hash); | ||
| 173 | |||
| 174 | vino.ino = cfh->ino; | ||
| 175 | vino.snap = CEPH_NOSNAP; | ||
| 176 | inode = ceph_find_inode(sb, vino); | ||
| 177 | if (!inode) { | ||
| 178 | struct ceph_mds_request *req; | ||
| 179 | |||
| 180 | req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPHASH, | ||
| 181 | USE_ANY_MDS); | ||
| 182 | if (IS_ERR(req)) | ||
| 183 | return ERR_CAST(req); | ||
| 184 | |||
| 185 | req->r_ino1 = vino; | ||
| 186 | req->r_ino2.ino = cfh->parent_ino; | ||
| 187 | req->r_ino2.snap = CEPH_NOSNAP; | ||
| 188 | req->r_path2 = kmalloc(16, GFP_NOFS); | ||
| 189 | snprintf(req->r_path2, 16, "%d", cfh->parent_name_hash); | ||
| 190 | req->r_num_caps = 1; | ||
| 191 | err = ceph_mdsc_do_request(mdsc, NULL, req); | ||
| 192 | inode = req->r_target_inode; | ||
| 193 | if (inode) | ||
| 194 | ihold(inode); | ||
| 195 | ceph_mdsc_put_request(req); | ||
| 196 | if (!inode) | ||
| 197 | return ERR_PTR(err ? err : -ESTALE); | ||
| 198 | } | ||
| 199 | 113 | ||
| 200 | dentry = d_obtain_alias(inode); | 114 | if (fh_type != FILEID_INO32_GEN && |
| 201 | if (IS_ERR(dentry)) { | 115 | fh_type != FILEID_INO32_GEN_PARENT) |
| 202 | pr_err("cfh_to_dentry %llx -- inode %p but ENOMEM\n", | 116 | return NULL; |
| 203 | cfh->ino, inode); | 117 | if (fh_len < sizeof(*fh) / 4) |
| 204 | iput(inode); | 118 | return NULL; |
| 205 | return dentry; | ||
| 206 | } | ||
| 207 | err = ceph_init_dentry(dentry); | ||
| 208 | if (err < 0) { | ||
| 209 | iput(inode); | ||
| 210 | return ERR_PTR(err); | ||
| 211 | } | ||
| 212 | dout("__cfh_to_dentry %llx %p dentry %p\n", cfh->ino, inode, dentry); | ||
| 213 | return dentry; | ||
| 214 | } | ||
| 215 | 119 | ||
| 216 | static struct dentry *ceph_fh_to_dentry(struct super_block *sb, struct fid *fid, | 120 | dout("fh_to_dentry %llx\n", fh->ino); |
| 217 | int fh_len, int fh_type) | 121 | return __fh_to_dentry(sb, fh->ino); |
| 218 | { | ||
| 219 | if (fh_type == 1) | ||
| 220 | return __fh_to_dentry(sb, (struct ceph_nfs_fh *)fid->raw, | ||
| 221 | fh_len); | ||
| 222 | else | ||
| 223 | return __cfh_to_dentry(sb, (struct ceph_nfs_confh *)fid->raw, | ||
| 224 | fh_len); | ||
| 225 | } | 122 | } |
| 226 | 123 | ||
| 227 | /* | 124 | /* |
