fs/ceph/export.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/ceph/ceph_debug.h>
   3
   4 #include <linux/exportfs.h>
   5 #include <linux/slab.h>
   6 #include <asm/unaligned.h>
   7
   8 #include "super.h"
   9 #include "mds_client.h"
  10
  11 /*
  12  * Basic fh
  13  */
  14 struct ceph_nfs_fh {
  15         u64 ino;
  16 } __attribute__ ((packed));
  17
  18 /*
  19  * Larger fh that includes parent ino.
  20  */
  21 struct ceph_nfs_confh {
  22         u64 ino, parent_ino;
  23 } __attribute__ ((packed));
  24
  25 /*
  26  * fh for snapped inode
  27  */
  28 struct ceph_nfs_snapfh {
  29         u64 ino;
  30         u64 snapid;
  31         u64 parent_ino;
  32         u32 hash;
  33 } __attribute__ ((packed));
  34
  35 static int ceph_encode_snapfh(struct inode *inode, u32 *rawfh, int *max_len,
  36                               struct inode *parent_inode)
  37 {
  38         static const int snap_handle_length =
  39                 sizeof(struct ceph_nfs_snapfh) >> 2;
  40         struct ceph_nfs_snapfh *sfh = (void *)rawfh;
  41         u64 snapid = ceph_snap(inode);
  42         int ret;
  43         bool no_parent = true;
  44
  45         if (*max_len < snap_handle_length) {
  46                 *max_len = snap_handle_length;
  47                 ret = FILEID_INVALID;
  48                 goto out;
  49         }
  50
  51         ret =  -EINVAL;
  52         if (snapid != CEPH_SNAPDIR) {
  53                 struct inode *dir;
  54                 struct dentry *dentry = d_find_alias(inode);
  55                 if (!dentry)
  56                         goto out;
  57
  58                 rcu_read_lock();
  59                 dir = d_inode_rcu(dentry->d_parent);
  60                 if (ceph_snap(dir) != CEPH_SNAPDIR) {
  61                         sfh->parent_ino = ceph_ino(dir);
  62                         sfh->hash = ceph_dentry_hash(dir, dentry);
  63                         no_parent = false;
  64                 }
  65                 rcu_read_unlock();
  66                 dput(dentry);
  67         }
  68
  69         if (no_parent) {
  70                 if (!S_ISDIR(inode->i_mode))
  71                         goto out;
  72                 sfh->parent_ino = sfh->ino;
  73                 sfh->hash = 0;
  74         }
  75         sfh->ino = ceph_ino(inode);
  76         sfh->snapid = snapid;
  77
  78         *max_len = snap_handle_length;
  79         ret = FILEID_BTRFS_WITH_PARENT;
  80 out:
  81         dout("encode_snapfh %llx.%llx ret=%d\n", ceph_vinop(inode), ret);
  82         return ret;
  83 }
  84
  85 static int ceph_encode_fh(struct inode *inode, u32 *rawfh, int *max_len,
  86                           struct inode *parent_inode)
  87 {
  88         static const int handle_length =
  89                 sizeof(struct ceph_nfs_fh) >> 2;
  90         static const int connected_handle_length =
  91                 sizeof(struct ceph_nfs_confh) >> 2;
  92         int type;
  93
  94         if (ceph_snap(inode) != CEPH_NOSNAP)
  95                 return ceph_encode_snapfh(inode, rawfh, max_len, parent_inode);
  96
  97         if (parent_inode && (*max_len < connected_handle_length)) {
  98                 *max_len = connected_handle_length;
  99                 return FILEID_INVALID;
 100         } else if (*max_len < handle_length) {
 101                 *max_len = handle_length;
 102                 return FILEID_INVALID;
 103         }
 104
 105         if (parent_inode) {
 106                 struct ceph_nfs_confh *cfh = (void *)rawfh;
 107                 dout("encode_fh %llx with parent %llx\n",
 108                      ceph_ino(inode), ceph_ino(parent_inode));
 109                 cfh->ino = ceph_ino(inode);
 110                 cfh->parent_ino = ceph_ino(parent_inode);
 111                 *max_len = connected_handle_length;
 112                 type = FILEID_INO32_GEN_PARENT;
 113         } else {
 114                 struct ceph_nfs_fh *fh = (void *)rawfh;
 115                 dout("encode_fh %llx\n", ceph_ino(inode));
 116                 fh->ino = ceph_ino(inode);
 117                 *max_len = handle_length;
 118                 type = FILEID_INO32_GEN;
 119         }
 120         return type;
 121 }
 122
 123 static struct inode *__lookup_inode(struct super_block *sb, u64 ino)
 124 {
 125         struct ceph_mds_client *mdsc = ceph_sb_to_client(sb)->mdsc;
 126         struct inode *inode;
 127         struct ceph_vino vino;
 128         int err;
 129
 130         vino.ino = ino;
 131         vino.snap = CEPH_NOSNAP;
 132         inode = ceph_find_inode(sb, vino);
 133         if (!inode) {
 134                 struct ceph_mds_request *req;
 135                 int mask;
 136
 137                 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPINO,
 138                                                USE_ANY_MDS);
 139                 if (IS_ERR(req))
 140                         return ERR_CAST(req);
 141
 142                 mask = CEPH_STAT_CAP_INODE;
 143                 if (ceph_security_xattr_wanted(d_inode(sb->s_root)))
 144                         mask |= CEPH_CAP_XATTR_SHARED;
 145                 req->r_args.lookupino.mask = cpu_to_le32(mask);
 146
 147                 req->r_ino1 = vino;
 148                 req->r_num_caps = 1;
 149                 err = ceph_mdsc_do_request(mdsc, NULL, req);
 150                 inode = req->r_target_inode;
 151                 if (inode)
 152                         ihold(inode);
 153                 ceph_mdsc_put_request(req);
 154                 if (!inode)
 155                         return err < 0 ? ERR_PTR(err) : ERR_PTR(-ESTALE);
 156         }
 157         return inode;
 158 }
 159
 160 struct inode *ceph_lookup_inode(struct super_block *sb, u64 ino)
 161 {
 162         struct inode *inode = __lookup_inode(sb, ino);
 163         if (IS_ERR(inode))
 164                 return inode;
 165         if (inode->i_nlink == 0) {
 166                 iput(inode);
 167                 return ERR_PTR(-ESTALE);
 168         }
 169         return inode;
 170 }
 171
 172 static struct dentry *__fh_to_dentry(struct super_block *sb, u64 ino)
 173 {
 174         struct inode *inode = __lookup_inode(sb, ino);
 175         int err;
 176
 177         if (IS_ERR(inode))
 178                 return ERR_CAST(inode);
 179         /* We need LINK caps to reliably check i_nlink */
 180         err = ceph_do_getattr(inode, CEPH_CAP_LINK_SHARED, false);
 181         if (err) {
 182                 iput(inode);
 183                 return ERR_PTR(err);
 184         }
 185         /* -ESTALE if inode as been unlinked and no file is open */
 186         if ((inode->i_nlink == 0) && (atomic_read(&inode->i_count) == 1)) {
 187                 iput(inode);
 188                 return ERR_PTR(-ESTALE);
 189         }
 190         return d_obtain_alias(inode);
 191 }
 192
 193 static struct dentry *__snapfh_to_dentry(struct super_block *sb,
 194                                           struct ceph_nfs_snapfh *sfh,
 195                                           bool want_parent)
 196 {
 197         struct ceph_mds_client *mdsc = ceph_sb_to_client(sb)->mdsc;
 198         struct ceph_mds_request *req;
 199         struct inode *inode;
 200         struct ceph_vino vino;
 201         int mask;
 202         int err;
 203         bool unlinked = false;
 204
 205         if (want_parent) {
 206                 vino.ino = sfh->parent_ino;
 207                 if (sfh->snapid == CEPH_SNAPDIR)
 208                         vino.snap = CEPH_NOSNAP;
 209                 else if (sfh->ino == sfh->parent_ino)
 210                         vino.snap = CEPH_SNAPDIR;
 211                 else
 212                         vino.snap = sfh->snapid;
 213         } else {
 214                 vino.ino = sfh->ino;
 215                 vino.snap = sfh->snapid;
 216         }
 217         inode = ceph_find_inode(sb, vino);
 218         if (inode)
 219                 return d_obtain_alias(inode);
 220
 221         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPINO,
 222                                        USE_ANY_MDS);
 223         if (IS_ERR(req))
 224                 return ERR_CAST(req);
 225
 226         mask = CEPH_STAT_CAP_INODE;
 227         if (ceph_security_xattr_wanted(d_inode(sb->s_root)))
 228                 mask |= CEPH_CAP_XATTR_SHARED;
 229         req->r_args.lookupino.mask = cpu_to_le32(mask);
 230         if (vino.snap < CEPH_NOSNAP) {
 231                 req->r_args.lookupino.snapid = cpu_to_le64(vino.snap);
 232                 if (!want_parent && sfh->ino != sfh->parent_ino) {
 233                         req->r_args.lookupino.parent =
 234                                         cpu_to_le64(sfh->parent_ino);
 235                         req->r_args.lookupino.hash =
 236                                         cpu_to_le32(sfh->hash);
 237                 }
 238         }
 239
 240         req->r_ino1 = vino;
 241         req->r_num_caps = 1;
 242         err = ceph_mdsc_do_request(mdsc, NULL, req);
 243         inode = req->r_target_inode;
 244         if (inode) {
 245                 if (vino.snap == CEPH_SNAPDIR) {
 246                         if (inode->i_nlink == 0)
 247                                 unlinked = true;
 248                         inode = ceph_get_snapdir(inode);
 249                 } else if (ceph_snap(inode) == vino.snap) {
 250                         ihold(inode);
 251                 } else {
 252                         /* mds does not support lookup snapped inode */
 253                         err = -EOPNOTSUPP;
 254                         inode = NULL;
 255                 }
 256         }
 257         ceph_mdsc_put_request(req);
 258
 259         if (want_parent) {
 260                 dout("snapfh_to_parent %llx.%llx\n err=%d\n",
 261                      vino.ino, vino.snap, err);
 262         } else {
 263                 dout("snapfh_to_dentry %llx.%llx parent %llx hash %x err=%d",
 264                       vino.ino, vino.snap, sfh->parent_ino, sfh->hash, err);
 265         }
 266         if (!inode)
 267                 return ERR_PTR(-ESTALE);
 268         /* see comments in ceph_get_parent() */
 269         return unlinked ? d_obtain_root(inode) : d_obtain_alias(inode);
 270 }
 271
 272 /*
 273  * convert regular fh to dentry
 274  */
 275 static struct dentry *ceph_fh_to_dentry(struct super_block *sb,
 276                                         struct fid *fid,
 277                                         int fh_len, int fh_type)
 278 {
 279         struct ceph_nfs_fh *fh = (void *)fid->raw;
 280
 281         if (fh_type == FILEID_BTRFS_WITH_PARENT) {
 282                 struct ceph_nfs_snapfh *sfh = (void *)fid->raw;
 283                 return __snapfh_to_dentry(sb, sfh, false);
 284         }
 285
 286         if (fh_type != FILEID_INO32_GEN  &&
 287             fh_type != FILEID_INO32_GEN_PARENT)
 288                 return NULL;
 289         if (fh_len < sizeof(*fh) / 4)
 290                 return NULL;
 291
 292         dout("fh_to_dentry %llx\n", fh->ino);
 293         return __fh_to_dentry(sb, fh->ino);
 294 }
 295
 296 static struct dentry *__get_parent(struct super_block *sb,
 297                                    struct dentry *child, u64 ino)
 298 {
 299         struct ceph_mds_client *mdsc = ceph_sb_to_client(sb)->mdsc;
 300         struct ceph_mds_request *req;
 301         struct inode *inode;
 302         int mask;
 303         int err;
 304
 305         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPPARENT,
 306                                        USE_ANY_MDS);
 307         if (IS_ERR(req))
 308                 return ERR_CAST(req);
 309
 310         if (child) {
 311                 req->r_inode = d_inode(child);
 312                 ihold(d_inode(child));
 313         } else {
 314                 req->r_ino1 = (struct ceph_vino) {
 315                         .ino = ino,
 316                         .snap = CEPH_NOSNAP,
 317                 };
 318         }
 319
 320         mask = CEPH_STAT_CAP_INODE;
 321         if (ceph_security_xattr_wanted(d_inode(sb->s_root)))
 322                 mask |= CEPH_CAP_XATTR_SHARED;
 323         req->r_args.getattr.mask = cpu_to_le32(mask);
 324
 325         req->r_num_caps = 1;
 326         err = ceph_mdsc_do_request(mdsc, NULL, req);
 327         if (err) {
 328                 ceph_mdsc_put_request(req);
 329                 return ERR_PTR(err);
 330         }
 331
 332         inode = req->r_target_inode;
 333         if (inode)
 334                 ihold(inode);
 335         ceph_mdsc_put_request(req);
 336         if (!inode)
 337                 return ERR_PTR(-ENOENT);
 338
 339         return d_obtain_alias(inode);
 340 }
 341
 342 static struct dentry *ceph_get_parent(struct dentry *child)
 343 {
 344         struct inode *inode = d_inode(child);
 345         struct dentry *dn;
 346
 347         if (ceph_snap(inode) != CEPH_NOSNAP) {
 348                 struct inode* dir;
 349                 bool unlinked = false;
 350                 /* do not support non-directory */
 351                 if (!d_is_dir(child)) {
 352                         dn = ERR_PTR(-EINVAL);
 353                         goto out;
 354                 }
 355                 dir = __lookup_inode(inode->i_sb, ceph_ino(inode));
 356                 if (IS_ERR(dir)) {
 357                         dn = ERR_CAST(dir);
 358                         goto out;
 359                 }
 360                 /* There can be multiple paths to access snapped inode.
 361                  * For simplicity, treat snapdir of head inode as parent */
 362                 if (ceph_snap(inode) != CEPH_SNAPDIR) {
 363                         struct inode *snapdir = ceph_get_snapdir(dir);
 364                         if (dir->i_nlink == 0)
 365                                 unlinked = true;
 366                         iput(dir);
 367                         if (IS_ERR(snapdir)) {
 368                                 dn = ERR_CAST(snapdir);
 369                                 goto out;
 370                         }
 371                         dir = snapdir;
 372                 }
 373                 /* If directory has already been deleted, futher get_parent
 374                  * will fail. Do not mark snapdir dentry as disconnected,
 375                  * this prevent exportfs from doing futher get_parent. */
 376                 if (unlinked)
 377                         dn = d_obtain_root(dir);
 378                 else
 379                         dn = d_obtain_alias(dir);
 380         } else {
 381                 dn = __get_parent(child->d_sb, child, 0);
 382         }
 383 out:
 384         dout("get_parent %p ino %llx.%llx err=%ld\n",
 385              child, ceph_vinop(inode), (long)PTR_ERR_OR_ZERO(dn));
 386         return dn;
 387 }
 388
 389 /*
 390  * convert regular fh to parent
 391  */
 392 static struct dentry *ceph_fh_to_parent(struct super_block *sb,
 393                                         struct fid *fid,
 394                                         int fh_len, int fh_type)
 395 {
 396         struct ceph_nfs_confh *cfh = (void *)fid->raw;
 397         struct dentry *dentry;
 398
 399         if (fh_type == FILEID_BTRFS_WITH_PARENT) {
 400                 struct ceph_nfs_snapfh *sfh = (void *)fid->raw;
 401                 return __snapfh_to_dentry(sb, sfh, true);
 402         }
 403
 404         if (fh_type != FILEID_INO32_GEN_PARENT)
 405                 return NULL;
 406         if (fh_len < sizeof(*cfh) / 4)
 407                 return NULL;
 408
 409         dout("fh_to_parent %llx\n", cfh->parent_ino);
 410         dentry = __get_parent(sb, NULL, cfh->ino);
 411         if (unlikely(dentry == ERR_PTR(-ENOENT)))
 412                 dentry = __fh_to_dentry(sb, cfh->parent_ino);
 413         return dentry;
 414 }
 415
 416 static int __get_snap_name(struct dentry *parent, char *name,
 417                            struct dentry *child)
 418 {
 419         struct inode *inode = d_inode(child);
 420         struct inode *dir = d_inode(parent);
 421         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
 422         struct ceph_mds_request *req = NULL;
 423         char *last_name = NULL;
 424         unsigned next_offset = 2;
 425         int err = -EINVAL;
 426
 427         if (ceph_ino(inode) != ceph_ino(dir))
 428                 goto out;
 429         if (ceph_snap(inode) == CEPH_SNAPDIR) {
 430                 if (ceph_snap(dir) == CEPH_NOSNAP) {
 431                         strcpy(name, fsc->mount_options->snapdir_name);
 432                         err = 0;
 433                 }
 434                 goto out;
 435         }
 436         if (ceph_snap(dir) != CEPH_SNAPDIR)
 437                 goto out;
 438
 439         while (1) {
 440                 struct ceph_mds_reply_info_parsed *rinfo;
 441                 struct ceph_mds_reply_dir_entry *rde;
 442                 int i;
 443
 444                 req = ceph_mdsc_create_request(fsc->mdsc, CEPH_MDS_OP_LSSNAP,
 445                                                USE_AUTH_MDS);
 446                 if (IS_ERR(req)) {
 447                         err = PTR_ERR(req);
 448                         req = NULL;
 449                         goto out;
 450                 }
 451                 err = ceph_alloc_readdir_reply_buffer(req, inode);
 452                 if (err)
 453                         goto out;
 454
 455                 req->r_direct_mode = USE_AUTH_MDS;
 456                 req->r_readdir_offset = next_offset;
 457                 req->r_args.readdir.flags =
 458                                 cpu_to_le16(CEPH_READDIR_REPLY_BITFLAGS);
 459                 if (last_name) {
 460                         req->r_path2 = last_name;
 461                         last_name = NULL;
 462                 }
 463
 464                 req->r_inode = dir;
 465                 ihold(dir);
 466                 req->r_dentry = dget(parent);
 467
 468                 inode_lock(dir);
 469                 err = ceph_mdsc_do_request(fsc->mdsc, NULL, req);
 470                 inode_unlock(dir);
 471
 472                 if (err < 0)
 473                         goto out;
 474
 475                 rinfo = &req->r_reply_info;
 476                 for (i = 0; i < rinfo->dir_nr; i++) {
 477                         rde = rinfo->dir_entries + i;
 478                         BUG_ON(!rde->inode.in);
 479                         if (ceph_snap(inode) ==
 480                             le64_to_cpu(rde->inode.in->snapid)) {
 481                                 memcpy(name, rde->name, rde->name_len);
 482                                 name[rde->name_len] = '\0';
 483                                 err = 0;
 484                                 goto out;
 485                         }
 486                 }
 487
 488                 if (rinfo->dir_end)
 489                         break;
 490
 491                 BUG_ON(rinfo->dir_nr <= 0);
 492                 rde = rinfo->dir_entries + (rinfo->dir_nr - 1);
 493                 next_offset += rinfo->dir_nr;
 494                 last_name = kstrndup(rde->name, rde->name_len, GFP_KERNEL);
 495                 if (!last_name) {
 496                         err = -ENOMEM;
 497                         goto out;
 498                 }
 499
 500                 ceph_mdsc_put_request(req);
 501                 req = NULL;
 502         }
 503         err = -ENOENT;
 504 out:
 505         if (req)
 506                 ceph_mdsc_put_request(req);
 507         kfree(last_name);
 508         dout("get_snap_name %p ino %llx.%llx err=%d\n",
 509              child, ceph_vinop(inode), err);
 510         return err;
 511 }
 512
 513 static int ceph_get_name(struct dentry *parent, char *name,
 514                          struct dentry *child)
 515 {
 516         struct ceph_mds_client *mdsc;
 517         struct ceph_mds_request *req;
 518         struct inode *inode = d_inode(child);
 519         int err;
 520
 521         if (ceph_snap(inode) != CEPH_NOSNAP)
 522                 return __get_snap_name(parent, name, child);
 523
 524         mdsc = ceph_inode_to_client(inode)->mdsc;
 525         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPNAME,
 526                                        USE_ANY_MDS);
 527         if (IS_ERR(req))
 528                 return PTR_ERR(req);
 529
 530         inode_lock(d_inode(parent));
 531
 532         req->r_inode = inode;
 533         ihold(inode);
 534         req->r_ino2 = ceph_vino(d_inode(parent));
 535         req->r_parent = d_inode(parent);
 536         set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
 537         req->r_num_caps = 2;
 538         err = ceph_mdsc_do_request(mdsc, NULL, req);
 539
 540         inode_unlock(d_inode(parent));
 541
 542         if (!err) {
 543                 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
 544                 memcpy(name, rinfo->dname, rinfo->dname_len);
 545                 name[rinfo->dname_len] = 0;
 546                 dout("get_name %p ino %llx.%llx name %s\n",
 547                      child, ceph_vinop(inode), name);
 548         } else {
 549                 dout("get_name %p ino %llx.%llx err %d\n",
 550                      child, ceph_vinop(inode), err);
 551         }
 552
 553         ceph_mdsc_put_request(req);
 554         return err;
 555 }
 556
 557 const struct export_operations ceph_export_ops = {
 558         .encode_fh = ceph_encode_fh,
 559         .fh_to_dentry = ceph_fh_to_dentry,
 560         .fh_to_parent = ceph_fh_to_parent,
 561         .get_parent = ceph_get_parent,
 562         .get_name = ceph_get_name,
 563 };