fs/ceph/dir.c

   1 #include <linux/ceph/ceph_debug.h>
   2
   3 #include <linux/spinlock.h>
   4 #include <linux/fs_struct.h>
   5 #include <linux/namei.h>
   6 #include <linux/slab.h>
   7 #include <linux/sched.h>
   8
   9 #include "super.h"
  10 #include "mds_client.h"
  11
  12 /*
  13  * Directory operations: readdir, lookup, create, link, unlink,
  14  * rename, etc.
  15  */
  16
  17 /*
  18  * Ceph MDS operations are specified in terms of a base ino and
  19  * relative path.  Thus, the client can specify an operation on a
  20  * specific inode (e.g., a getattr due to fstat(2)), or as a path
  21  * relative to, say, the root directory.
  22  *
  23  * Normally, we limit ourselves to strict inode ops (no path component)
  24  * or dentry operations (a single path component relative to an ino).  The
  25  * exception to this is open_root_dentry(), which will open the mount
  26  * point by name.
  27  */
  28
  29 const struct dentry_operations ceph_dentry_ops;
  30
  31 /*
  32  * Initialize ceph dentry state.
  33  */
  34 int ceph_init_dentry(struct dentry *dentry)
  35 {
  36         struct ceph_dentry_info *di;
  37
  38         if (dentry->d_fsdata)
  39                 return 0;
  40
  41         di = kmem_cache_zalloc(ceph_dentry_cachep, GFP_KERNEL);
  42         if (!di)
  43                 return -ENOMEM;          /* oh well */
  44
  45         spin_lock(&dentry->d_lock);
  46         if (dentry->d_fsdata) {
  47                 /* lost a race */
  48                 kmem_cache_free(ceph_dentry_cachep, di);
  49                 goto out_unlock;
  50         }
  51
  52         if (ceph_snap(d_inode(dentry->d_parent)) == CEPH_NOSNAP)
  53                 d_set_d_op(dentry, &ceph_dentry_ops);
  54         else if (ceph_snap(d_inode(dentry->d_parent)) == CEPH_SNAPDIR)
  55                 d_set_d_op(dentry, &ceph_snapdir_dentry_ops);
  56         else
  57                 d_set_d_op(dentry, &ceph_snap_dentry_ops);
  58
  59         di->dentry = dentry;
  60         di->lease_session = NULL;
  61         dentry->d_time = jiffies;
  62         /* avoid reordering d_fsdata setup so that the check above is safe */
  63         smp_mb();
  64         dentry->d_fsdata = di;
  65         ceph_dentry_lru_add(dentry);
  66 out_unlock:
  67         spin_unlock(&dentry->d_lock);
  68         return 0;
  69 }
  70
  71 /*
  72  * for readdir, we encode the directory frag and offset within that
  73  * frag into f_pos.
  74  */
  75 static unsigned fpos_frag(loff_t p)
  76 {
  77         return p >> 32;
  78 }
  79 static unsigned fpos_off(loff_t p)
  80 {
  81         return p & 0xffffffff;
  82 }
  83
  84 static int fpos_cmp(loff_t l, loff_t r)
  85 {
  86         int v = ceph_frag_compare(fpos_frag(l), fpos_frag(r));
  87         if (v)
  88                 return v;
  89         return (int)(fpos_off(l) - fpos_off(r));
  90 }
  91
  92 /*
  93  * make note of the last dentry we read, so we can
  94  * continue at the same lexicographical point,
  95  * regardless of what dir changes take place on the
  96  * server.
  97  */
  98 static int note_last_dentry(struct ceph_file_info *fi, const char *name,
  99                             int len, unsigned next_offset)
 100 {
 101         char *buf = kmalloc(len+1, GFP_KERNEL);
 102         if (!buf)
 103                 return -ENOMEM;
 104         kfree(fi->last_name);
 105         fi->last_name = buf;
 106         memcpy(fi->last_name, name, len);
 107         fi->last_name[len] = 0;
 108         fi->next_offset = next_offset;
 109         dout("note_last_dentry '%s'\n", fi->last_name);
 110         return 0;
 111 }
 112
 113 /*
 114  * When possible, we try to satisfy a readdir by peeking at the
 115  * dcache.  We make this work by carefully ordering dentries on
 116  * d_child when we initially get results back from the MDS, and
 117  * falling back to a "normal" sync readdir if any dentries in the dir
 118  * are dropped.
 119  *
 120  * Complete dir indicates that we have all dentries in the dir.  It is
 121  * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
 122  * the MDS if/when the directory is modified).
 123  */
 124 static int __dcache_readdir(struct file *file,  struct dir_context *ctx,
 125                             u32 shared_gen)
 126 {
 127         struct ceph_file_info *fi = file->private_data;
 128         struct dentry *parent = file->f_path.dentry;
 129         struct inode *dir = d_inode(parent);
 130         struct dentry *dentry, *last = NULL;
 131         struct ceph_dentry_info *di;
 132         unsigned nsize = PAGE_CACHE_SIZE / sizeof(struct dentry *);
 133         int err = 0;
 134         loff_t ptr_pos = 0;
 135         struct ceph_readdir_cache_control cache_ctl = {};
 136
 137         dout("__dcache_readdir %p v%u at %llu\n", dir, shared_gen, ctx->pos);
 138
 139         /* we can calculate cache index for the first dirfrag */
 140         if (ceph_frag_is_leftmost(fpos_frag(ctx->pos))) {
 141                 cache_ctl.index = fpos_off(ctx->pos) - 2;
 142                 BUG_ON(cache_ctl.index < 0);
 143                 ptr_pos = cache_ctl.index * sizeof(struct dentry *);
 144         }
 145
 146         while (true) {
 147                 pgoff_t pgoff;
 148                 bool emit_dentry;
 149
 150                 if (ptr_pos >= i_size_read(dir)) {
 151                         fi->flags |= CEPH_F_ATEND;
 152                         err = 0;
 153                         break;
 154                 }
 155
 156                 err = -EAGAIN;
 157                 pgoff = ptr_pos >> PAGE_CACHE_SHIFT;
 158                 if (!cache_ctl.page || pgoff != page_index(cache_ctl.page)) {
 159                         ceph_readdir_cache_release(&cache_ctl);
 160                         cache_ctl.page = find_lock_page(&dir->i_data, pgoff);
 161                         if (!cache_ctl.page) {
 162                                 dout(" page %lu not found\n", pgoff);
 163                                 break;
 164                         }
 165                         /* reading/filling the cache are serialized by
 166                          * i_mutex, no need to use page lock */
 167                         unlock_page(cache_ctl.page);
 168                         cache_ctl.dentries = kmap(cache_ctl.page);
 169                 }
 170
 171                 rcu_read_lock();
 172                 spin_lock(&parent->d_lock);
 173                 /* check i_size again here, because empty directory can be
 174                  * marked as complete while not holding the i_mutex. */
 175                 if (ceph_dir_is_complete_ordered(dir) &&
 176                     ptr_pos < i_size_read(dir))
 177                         dentry = cache_ctl.dentries[cache_ctl.index % nsize];
 178                 else
 179                         dentry = NULL;
 180                 spin_unlock(&parent->d_lock);
 181                 if (dentry && !lockref_get_not_dead(&dentry->d_lockref))
 182                         dentry = NULL;
 183                 rcu_read_unlock();
 184                 if (!dentry)
 185                         break;
 186
 187                 emit_dentry = false;
 188                 di = ceph_dentry(dentry);
 189                 spin_lock(&dentry->d_lock);
 190                 if (di->lease_shared_gen == shared_gen &&
 191                     d_really_is_positive(dentry) &&
 192                     ceph_snap(d_inode(dentry)) != CEPH_SNAPDIR &&
 193                     ceph_ino(d_inode(dentry)) != CEPH_INO_CEPH &&
 194                     fpos_cmp(ctx->pos, di->offset) <= 0) {
 195                         emit_dentry = true;
 196                 }
 197                 spin_unlock(&dentry->d_lock);
 198
 199                 if (emit_dentry) {
 200                         dout(" %llu (%llu) dentry %p %pd %p\n", di->offset, ctx->pos,
 201                              dentry, dentry, d_inode(dentry));
 202                         ctx->pos = di->offset;
 203                         if (!dir_emit(ctx, dentry->d_name.name,
 204                                       dentry->d_name.len,
 205                                       ceph_translate_ino(dentry->d_sb,
 206                                                          d_inode(dentry)->i_ino),
 207                                       d_inode(dentry)->i_mode >> 12)) {
 208                                 dput(dentry);
 209                                 err = 0;
 210                                 break;
 211                         }
 212                         ctx->pos++;
 213
 214                         if (last)
 215                                 dput(last);
 216                         last = dentry;
 217                 } else {
 218                         dput(dentry);
 219                 }
 220
 221                 cache_ctl.index++;
 222                 ptr_pos += sizeof(struct dentry *);
 223         }
 224         ceph_readdir_cache_release(&cache_ctl);
 225         if (last) {
 226                 int ret;
 227                 di = ceph_dentry(last);
 228                 ret = note_last_dentry(fi, last->d_name.name, last->d_name.len,
 229                                        fpos_off(di->offset) + 1);
 230                 if (ret < 0)
 231                         err = ret;
 232                 dput(last);
 233         }
 234         return err;
 235 }
 236
 237 static int ceph_readdir(struct file *file, struct dir_context *ctx)
 238 {
 239         struct ceph_file_info *fi = file->private_data;
 240         struct inode *inode = file_inode(file);
 241         struct ceph_inode_info *ci = ceph_inode(inode);
 242         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
 243         struct ceph_mds_client *mdsc = fsc->mdsc;
 244         unsigned frag = fpos_frag(ctx->pos);
 245         int off = fpos_off(ctx->pos);
 246         int err;
 247         u32 ftype;
 248         struct ceph_mds_reply_info_parsed *rinfo;
 249
 250         dout("readdir %p file %p frag %u off %u\n", inode, file, frag, off);
 251         if (fi->flags & CEPH_F_ATEND)
 252                 return 0;
 253
 254         /* always start with . and .. */
 255         if (ctx->pos == 0) {
 256                 dout("readdir off 0 -> '.'\n");
 257                 if (!dir_emit(ctx, ".", 1,
 258                             ceph_translate_ino(inode->i_sb, inode->i_ino),
 259                             inode->i_mode >> 12))
 260                         return 0;
 261                 ctx->pos = 1;
 262                 off = 1;
 263         }
 264         if (ctx->pos == 1) {
 265                 ino_t ino = parent_ino(file->f_path.dentry);
 266                 dout("readdir off 1 -> '..'\n");
 267                 if (!dir_emit(ctx, "..", 2,
 268                             ceph_translate_ino(inode->i_sb, ino),
 269                             inode->i_mode >> 12))
 270                         return 0;
 271                 ctx->pos = 2;
 272                 off = 2;
 273         }
 274
 275         /* can we use the dcache? */
 276         spin_lock(&ci->i_ceph_lock);
 277         if (ceph_test_mount_opt(fsc, DCACHE) &&
 278             !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
 279             ceph_snap(inode) != CEPH_SNAPDIR &&
 280             __ceph_dir_is_complete_ordered(ci) &&
 281             __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
 282                 u32 shared_gen = ci->i_shared_gen;
 283                 spin_unlock(&ci->i_ceph_lock);
 284                 err = __dcache_readdir(file, ctx, shared_gen);
 285                 if (err != -EAGAIN)
 286                         return err;
 287                 frag = fpos_frag(ctx->pos);
 288                 off = fpos_off(ctx->pos);
 289         } else {
 290                 spin_unlock(&ci->i_ceph_lock);
 291         }
 292
 293         /* proceed with a normal readdir */
 294 more:
 295         /* do we have the correct frag content buffered? */
 296         if (fi->frag != frag || fi->last_readdir == NULL) {
 297                 struct ceph_mds_request *req;
 298                 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
 299                         CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
 300
 301                 /* discard old result, if any */
 302                 if (fi->last_readdir) {
 303                         ceph_mdsc_put_request(fi->last_readdir);
 304                         fi->last_readdir = NULL;
 305                 }
 306
 307                 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
 308                      ceph_vinop(inode), frag, fi->last_name);
 309                 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
 310                 if (IS_ERR(req))
 311                         return PTR_ERR(req);
 312                 err = ceph_alloc_readdir_reply_buffer(req, inode);
 313                 if (err) {
 314                         ceph_mdsc_put_request(req);
 315                         return err;
 316                 }
 317                 /* hints to request -> mds selection code */
 318                 req->r_direct_mode = USE_AUTH_MDS;
 319                 req->r_direct_hash = ceph_frag_value(frag);
 320                 req->r_direct_is_hash = true;
 321                 if (fi->last_name) {
 322                         req->r_path2 = kstrdup(fi->last_name, GFP_KERNEL);
 323                         if (!req->r_path2) {
 324                                 ceph_mdsc_put_request(req);
 325                                 return -ENOMEM;
 326                         }
 327                 }
 328                 req->r_dir_release_cnt = fi->dir_release_count;
 329                 req->r_dir_ordered_cnt = fi->dir_ordered_count;
 330                 req->r_readdir_cache_idx = fi->readdir_cache_idx;
 331                 req->r_readdir_offset = fi->next_offset;
 332                 req->r_args.readdir.frag = cpu_to_le32(frag);
 333
 334                 req->r_inode = inode;
 335                 ihold(inode);
 336                 req->r_dentry = dget(file->f_path.dentry);
 337                 err = ceph_mdsc_do_request(mdsc, NULL, req);
 338                 if (err < 0) {
 339                         ceph_mdsc_put_request(req);
 340                         return err;
 341                 }
 342                 dout("readdir got and parsed readdir result=%d"
 343                      " on frag %x, end=%d, complete=%d\n", err, frag,
 344                      (int)req->r_reply_info.dir_end,
 345                      (int)req->r_reply_info.dir_complete);
 346
 347
 348                 /* note next offset and last dentry name */
 349                 rinfo = &req->r_reply_info;
 350                 if (le32_to_cpu(rinfo->dir_dir->frag) != frag) {
 351                         frag = le32_to_cpu(rinfo->dir_dir->frag);
 352                         off = req->r_readdir_offset;
 353                         fi->next_offset = off;
 354                 }
 355
 356                 fi->frag = frag;
 357                 fi->offset = fi->next_offset;
 358                 fi->last_readdir = req;
 359
 360                 if (req->r_did_prepopulate) {
 361                         fi->readdir_cache_idx = req->r_readdir_cache_idx;
 362                         if (fi->readdir_cache_idx < 0) {
 363                                 /* preclude from marking dir ordered */
 364                                 fi->dir_ordered_count = 0;
 365                         } else if (ceph_frag_is_leftmost(frag) && off == 2) {
 366                                 /* note dir version at start of readdir so
 367                                  * we can tell if any dentries get dropped */
 368                                 fi->dir_release_count = req->r_dir_release_cnt;
 369                                 fi->dir_ordered_count = req->r_dir_ordered_cnt;
 370                         }
 371                 } else {
 372                         dout("readdir !did_prepopulate");
 373                         /* disable readdir cache */
 374                         fi->readdir_cache_idx = -1;
 375                         /* preclude from marking dir complete */
 376                         fi->dir_release_count = 0;
 377                 }
 378
 379                 if (req->r_reply_info.dir_end) {
 380                         kfree(fi->last_name);
 381                         fi->last_name = NULL;
 382                         if (ceph_frag_is_rightmost(frag))
 383                                 fi->next_offset = 2;
 384                         else
 385                                 fi->next_offset = 0;
 386                 } else {
 387                         err = note_last_dentry(fi,
 388                                        rinfo->dir_dname[rinfo->dir_nr-1],
 389                                        rinfo->dir_dname_len[rinfo->dir_nr-1],
 390                                        fi->next_offset + rinfo->dir_nr);
 391                         if (err)
 392                                 return err;
 393                 }
 394         }
 395
 396         rinfo = &fi->last_readdir->r_reply_info;
 397         dout("readdir frag %x num %d off %d chunkoff %d\n", frag,
 398              rinfo->dir_nr, off, fi->offset);
 399
 400         ctx->pos = ceph_make_fpos(frag, off);
 401         while (off >= fi->offset && off - fi->offset < rinfo->dir_nr) {
 402                 struct ceph_mds_reply_inode *in =
 403                         rinfo->dir_in[off - fi->offset].in;
 404                 struct ceph_vino vino;
 405                 ino_t ino;
 406
 407                 dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
 408                      off, off - fi->offset, rinfo->dir_nr, ctx->pos,
 409                      rinfo->dir_dname_len[off - fi->offset],
 410                      rinfo->dir_dname[off - fi->offset], in);
 411                 BUG_ON(!in);
 412                 ftype = le32_to_cpu(in->mode) >> 12;
 413                 vino.ino = le64_to_cpu(in->ino);
 414                 vino.snap = le64_to_cpu(in->snapid);
 415                 ino = ceph_vino_to_ino(vino);
 416                 if (!dir_emit(ctx,
 417                             rinfo->dir_dname[off - fi->offset],
 418                             rinfo->dir_dname_len[off - fi->offset],
 419                             ceph_translate_ino(inode->i_sb, ino), ftype)) {
 420                         dout("filldir stopping us...\n");
 421                         return 0;
 422                 }
 423                 off++;
 424                 ctx->pos++;
 425         }
 426
 427         if (fi->last_name) {
 428                 ceph_mdsc_put_request(fi->last_readdir);
 429                 fi->last_readdir = NULL;
 430                 goto more;
 431         }
 432
 433         /* more frags? */
 434         if (!ceph_frag_is_rightmost(frag)) {
 435                 frag = ceph_frag_next(frag);
 436                 off = 0;
 437                 ctx->pos = ceph_make_fpos(frag, off);
 438                 dout("readdir next frag is %x\n", frag);
 439                 goto more;
 440         }
 441         fi->flags |= CEPH_F_ATEND;
 442
 443         /*
 444          * if dir_release_count still matches the dir, no dentries
 445          * were released during the whole readdir, and we should have
 446          * the complete dir contents in our cache.
 447          */
 448         if (atomic64_read(&ci->i_release_count) == fi->dir_release_count) {
 449                 spin_lock(&ci->i_ceph_lock);
 450                 if (fi->dir_ordered_count == atomic64_read(&ci->i_ordered_count)) {
 451                         dout(" marking %p complete and ordered\n", inode);
 452                         /* use i_size to track number of entries in
 453                          * readdir cache */
 454                         BUG_ON(fi->readdir_cache_idx < 0);
 455                         i_size_write(inode, fi->readdir_cache_idx *
 456                                      sizeof(struct dentry*));
 457                 } else {
 458                         dout(" marking %p complete\n", inode);
 459                 }
 460                 __ceph_dir_set_complete(ci, fi->dir_release_count,
 461                                         fi->dir_ordered_count);
 462                 spin_unlock(&ci->i_ceph_lock);
 463         }
 464
 465         dout("readdir %p file %p done.\n", inode, file);
 466         return 0;
 467 }
 468
 469 static void reset_readdir(struct ceph_file_info *fi, unsigned frag)
 470 {
 471         if (fi->last_readdir) {
 472                 ceph_mdsc_put_request(fi->last_readdir);
 473                 fi->last_readdir = NULL;
 474         }
 475         kfree(fi->last_name);
 476         fi->last_name = NULL;
 477         fi->dir_release_count = 0;
 478         fi->readdir_cache_idx = -1;
 479         if (ceph_frag_is_leftmost(frag))
 480                 fi->next_offset = 2;  /* compensate for . and .. */
 481         else
 482                 fi->next_offset = 0;
 483         fi->flags &= ~CEPH_F_ATEND;
 484 }
 485
 486 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int whence)
 487 {
 488         struct ceph_file_info *fi = file->private_data;
 489         struct inode *inode = file->f_mapping->host;
 490         loff_t old_offset = ceph_make_fpos(fi->frag, fi->next_offset);
 491         loff_t retval;
 492
 493         inode_lock(inode);
 494         retval = -EINVAL;
 495         switch (whence) {
 496         case SEEK_CUR:
 497                 offset += file->f_pos;
 498         case SEEK_SET:
 499                 break;
 500         case SEEK_END:
 501                 retval = -EOPNOTSUPP;
 502         default:
 503                 goto out;
 504         }
 505
 506         if (offset >= 0) {
 507                 if (offset != file->f_pos) {
 508                         file->f_pos = offset;
 509                         file->f_version = 0;
 510                         fi->flags &= ~CEPH_F_ATEND;
 511                 }
 512                 retval = offset;
 513
 514                 if (offset == 0 ||
 515                     fpos_frag(offset) != fi->frag ||
 516                     fpos_off(offset) < fi->offset) {
 517                         /* discard buffered readdir content on seekdir(0), or
 518                          * seek to new frag, or seek prior to current chunk */
 519                         dout("dir_llseek dropping %p content\n", file);
 520                         reset_readdir(fi, fpos_frag(offset));
 521                 } else if (fpos_cmp(offset, old_offset) > 0) {
 522                         /* reset dir_release_count if we did a forward seek */
 523                         fi->dir_release_count = 0;
 524                         fi->readdir_cache_idx = -1;
 525                 }
 526         }
 527 out:
 528         inode_unlock(inode);
 529         return retval;
 530 }
 531
 532 /*
 533  * Handle lookups for the hidden .snap directory.
 534  */
 535 int ceph_handle_snapdir(struct ceph_mds_request *req,
 536                         struct dentry *dentry, int err)
 537 {
 538         struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
 539         struct inode *parent = d_inode(dentry->d_parent); /* we hold i_mutex */
 540
 541         /* .snap dir? */
 542         if (err == -ENOENT &&
 543             ceph_snap(parent) == CEPH_NOSNAP &&
 544             strcmp(dentry->d_name.name,
 545                    fsc->mount_options->snapdir_name) == 0) {
 546                 struct inode *inode = ceph_get_snapdir(parent);
 547                 dout("ENOENT on snapdir %p '%pd', linking to snapdir %p\n",
 548                      dentry, dentry, inode);
 549                 BUG_ON(!d_unhashed(dentry));
 550                 d_add(dentry, inode);
 551                 err = 0;
 552         }
 553         return err;
 554 }
 555
 556 /*
 557  * Figure out final result of a lookup/open request.
 558  *
 559  * Mainly, make sure we return the final req->r_dentry (if it already
 560  * existed) in place of the original VFS-provided dentry when they
 561  * differ.
 562  *
 563  * Gracefully handle the case where the MDS replies with -ENOENT and
 564  * no trace (which it may do, at its discretion, e.g., if it doesn't
 565  * care to issue a lease on the negative dentry).
 566  */
 567 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
 568                                   struct dentry *dentry, int err)
 569 {
 570         if (err == -ENOENT) {
 571                 /* no trace? */
 572                 err = 0;
 573                 if (!req->r_reply_info.head->is_dentry) {
 574                         dout("ENOENT and no trace, dentry %p inode %p\n",
 575                              dentry, d_inode(dentry));
 576                         if (d_really_is_positive(dentry)) {
 577                                 d_drop(dentry);
 578                                 err = -ENOENT;
 579                         } else {
 580                                 d_add(dentry, NULL);
 581                         }
 582                 }
 583         }
 584         if (err)
 585                 dentry = ERR_PTR(err);
 586         else if (dentry != req->r_dentry)
 587                 dentry = dget(req->r_dentry);   /* we got spliced */
 588         else
 589                 dentry = NULL;
 590         return dentry;
 591 }
 592
 593 static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
 594 {
 595         return ceph_ino(inode) == CEPH_INO_ROOT &&
 596                 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
 597 }
 598
 599 /*
 600  * Look up a single dir entry.  If there is a lookup intent, inform
 601  * the MDS so that it gets our 'caps wanted' value in a single op.
 602  */
 603 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
 604                                   unsigned int flags)
 605 {
 606         struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
 607         struct ceph_mds_client *mdsc = fsc->mdsc;
 608         struct ceph_mds_request *req;
 609         int op;
 610         int mask;
 611         int err;
 612
 613         dout("lookup %p dentry %p '%pd'\n",
 614              dir, dentry, dentry);
 615
 616         if (dentry->d_name.len > NAME_MAX)
 617                 return ERR_PTR(-ENAMETOOLONG);
 618
 619         err = ceph_init_dentry(dentry);
 620         if (err < 0)
 621                 return ERR_PTR(err);
 622
 623         /* can we conclude ENOENT locally? */
 624         if (d_really_is_negative(dentry)) {
 625                 struct ceph_inode_info *ci = ceph_inode(dir);
 626                 struct ceph_dentry_info *di = ceph_dentry(dentry);
 627
 628                 spin_lock(&ci->i_ceph_lock);
 629                 dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
 630                 if (strncmp(dentry->d_name.name,
 631                             fsc->mount_options->snapdir_name,
 632                             dentry->d_name.len) &&
 633                     !is_root_ceph_dentry(dir, dentry) &&
 634                     ceph_test_mount_opt(fsc, DCACHE) &&
 635                     __ceph_dir_is_complete(ci) &&
 636                     (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
 637                         spin_unlock(&ci->i_ceph_lock);
 638                         dout(" dir %p complete, -ENOENT\n", dir);
 639                         d_add(dentry, NULL);
 640                         di->lease_shared_gen = ci->i_shared_gen;
 641                         return NULL;
 642                 }
 643                 spin_unlock(&ci->i_ceph_lock);
 644         }
 645
 646         op = ceph_snap(dir) == CEPH_SNAPDIR ?
 647                 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
 648         req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
 649         if (IS_ERR(req))
 650                 return ERR_CAST(req);
 651         req->r_dentry = dget(dentry);
 652         req->r_num_caps = 2;
 653
 654         mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
 655         if (ceph_security_xattr_wanted(dir))
 656                 mask |= CEPH_CAP_XATTR_SHARED;
 657         req->r_args.getattr.mask = cpu_to_le32(mask);
 658
 659         req->r_locked_dir = dir;
 660         err = ceph_mdsc_do_request(mdsc, NULL, req);
 661         err = ceph_handle_snapdir(req, dentry, err);
 662         dentry = ceph_finish_lookup(req, dentry, err);
 663         ceph_mdsc_put_request(req);  /* will dput(dentry) */
 664         dout("lookup result=%p\n", dentry);
 665         return dentry;
 666 }
 667
 668 /*
 669  * If we do a create but get no trace back from the MDS, follow up with
 670  * a lookup (the VFS expects us to link up the provided dentry).
 671  */
 672 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
 673 {
 674         struct dentry *result = ceph_lookup(dir, dentry, 0);
 675
 676         if (result && !IS_ERR(result)) {
 677                 /*
 678                  * We created the item, then did a lookup, and found
 679                  * it was already linked to another inode we already
 680                  * had in our cache (and thus got spliced). To not
 681                  * confuse VFS (especially when inode is a directory),
 682                  * we don't link our dentry to that inode, return an
 683                  * error instead.
 684                  *
 685                  * This event should be rare and it happens only when
 686                  * we talk to old MDS. Recent MDS does not send traceless
 687                  * reply for request that creates new inode.
 688                  */
 689                 d_drop(result);
 690                 return -ESTALE;
 691         }
 692         return PTR_ERR(result);
 693 }
 694
 695 static int ceph_mknod(struct inode *dir, struct dentry *dentry,
 696                       umode_t mode, dev_t rdev)
 697 {
 698         struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
 699         struct ceph_mds_client *mdsc = fsc->mdsc;
 700         struct ceph_mds_request *req;
 701         struct ceph_acls_info acls = {};
 702         int err;
 703
 704         if (ceph_snap(dir) != CEPH_NOSNAP)
 705                 return -EROFS;
 706
 707         err = ceph_pre_init_acls(dir, &mode, &acls);
 708         if (err < 0)
 709                 return err;
 710
 711         dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n",
 712              dir, dentry, mode, rdev);
 713         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
 714         if (IS_ERR(req)) {
 715                 err = PTR_ERR(req);
 716                 goto out;
 717         }
 718         req->r_dentry = dget(dentry);
 719         req->r_num_caps = 2;
 720         req->r_locked_dir = dir;
 721         req->r_args.mknod.mode = cpu_to_le32(mode);
 722         req->r_args.mknod.rdev = cpu_to_le32(rdev);
 723         req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
 724         req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
 725         if (acls.pagelist) {
 726                 req->r_pagelist = acls.pagelist;
 727                 acls.pagelist = NULL;
 728         }
 729         err = ceph_mdsc_do_request(mdsc, dir, req);
 730         if (!err && !req->r_reply_info.head->is_dentry)
 731                 err = ceph_handle_notrace_create(dir, dentry);
 732         ceph_mdsc_put_request(req);
 733 out:
 734         if (!err)
 735                 ceph_init_inode_acls(d_inode(dentry), &acls);
 736         else
 737                 d_drop(dentry);
 738         ceph_release_acls_info(&acls);
 739         return err;
 740 }
 741
 742 static int ceph_create(struct inode *dir, struct dentry *dentry, umode_t mode,
 743                        bool excl)
 744 {
 745         return ceph_mknod(dir, dentry, mode, 0);
 746 }
 747
 748 static int ceph_symlink(struct inode *dir, struct dentry *dentry,
 749                             const char *dest)
 750 {
 751         struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
 752         struct ceph_mds_client *mdsc = fsc->mdsc;
 753         struct ceph_mds_request *req;
 754         int err;
 755
 756         if (ceph_snap(dir) != CEPH_NOSNAP)
 757                 return -EROFS;
 758
 759         dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
 760         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
 761         if (IS_ERR(req)) {
 762                 err = PTR_ERR(req);
 763                 goto out;
 764         }
 765         req->r_path2 = kstrdup(dest, GFP_KERNEL);
 766         if (!req->r_path2) {
 767                 err = -ENOMEM;
 768                 ceph_mdsc_put_request(req);
 769                 goto out;
 770         }
 771         req->r_locked_dir = dir;
 772         req->r_dentry = dget(dentry);
 773         req->r_num_caps = 2;
 774         req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
 775         req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
 776         err = ceph_mdsc_do_request(mdsc, dir, req);
 777         if (!err && !req->r_reply_info.head->is_dentry)
 778                 err = ceph_handle_notrace_create(dir, dentry);
 779         ceph_mdsc_put_request(req);
 780 out:
 781         if (err)
 782                 d_drop(dentry);
 783         return err;
 784 }
 785
 786 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
 787 {
 788         struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
 789         struct ceph_mds_client *mdsc = fsc->mdsc;
 790         struct ceph_mds_request *req;
 791         struct ceph_acls_info acls = {};
 792         int err = -EROFS;
 793         int op;
 794
 795         if (ceph_snap(dir) == CEPH_SNAPDIR) {
 796                 /* mkdir .snap/foo is a MKSNAP */
 797                 op = CEPH_MDS_OP_MKSNAP;
 798                 dout("mksnap dir %p snap '%pd' dn %p\n", dir,
 799                      dentry, dentry);
 800         } else if (ceph_snap(dir) == CEPH_NOSNAP) {
 801                 dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode);
 802                 op = CEPH_MDS_OP_MKDIR;
 803         } else {
 804                 goto out;
 805         }
 806
 807         mode |= S_IFDIR;
 808         err = ceph_pre_init_acls(dir, &mode, &acls);
 809         if (err < 0)
 810                 goto out;
 811
 812         req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
 813         if (IS_ERR(req)) {
 814                 err = PTR_ERR(req);
 815                 goto out;
 816         }
 817
 818         req->r_dentry = dget(dentry);
 819         req->r_num_caps = 2;
 820         req->r_locked_dir = dir;
 821         req->r_args.mkdir.mode = cpu_to_le32(mode);
 822         req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
 823         req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
 824         if (acls.pagelist) {
 825                 req->r_pagelist = acls.pagelist;
 826                 acls.pagelist = NULL;
 827         }
 828         err = ceph_mdsc_do_request(mdsc, dir, req);
 829         if (!err &&
 830             !req->r_reply_info.head->is_target &&
 831             !req->r_reply_info.head->is_dentry)
 832                 err = ceph_handle_notrace_create(dir, dentry);
 833         ceph_mdsc_put_request(req);
 834 out:
 835         if (!err)
 836                 ceph_init_inode_acls(d_inode(dentry), &acls);
 837         else
 838                 d_drop(dentry);
 839         ceph_release_acls_info(&acls);
 840         return err;
 841 }
 842
 843 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
 844                      struct dentry *dentry)
 845 {
 846         struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
 847         struct ceph_mds_client *mdsc = fsc->mdsc;
 848         struct ceph_mds_request *req;
 849         int err;
 850
 851         if (ceph_snap(dir) != CEPH_NOSNAP)
 852                 return -EROFS;
 853
 854         dout("link in dir %p old_dentry %p dentry %p\n", dir,
 855              old_dentry, dentry);
 856         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
 857         if (IS_ERR(req)) {
 858                 d_drop(dentry);
 859                 return PTR_ERR(req);
 860         }
 861         req->r_dentry = dget(dentry);
 862         req->r_num_caps = 2;
 863         req->r_old_dentry = dget(old_dentry);
 864         req->r_locked_dir = dir;
 865         req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
 866         req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
 867         /* release LINK_SHARED on source inode (mds will lock it) */
 868         req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
 869         err = ceph_mdsc_do_request(mdsc, dir, req);
 870         if (err) {
 871                 d_drop(dentry);
 872         } else if (!req->r_reply_info.head->is_dentry) {
 873                 ihold(d_inode(old_dentry));
 874                 d_instantiate(dentry, d_inode(old_dentry));
 875         }
 876         ceph_mdsc_put_request(req);
 877         return err;
 878 }
 879
 880 /*
 881  * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps.  If it
 882  * looks like the link count will hit 0, drop any other caps (other
 883  * than PIN) we don't specifically want (due to the file still being
 884  * open).
 885  */
 886 static int drop_caps_for_unlink(struct inode *inode)
 887 {
 888         struct ceph_inode_info *ci = ceph_inode(inode);
 889         int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
 890
 891         spin_lock(&ci->i_ceph_lock);
 892         if (inode->i_nlink == 1) {
 893                 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
 894                 ci->i_ceph_flags |= CEPH_I_NODELAY;
 895         }
 896         spin_unlock(&ci->i_ceph_lock);
 897         return drop;
 898 }
 899
 900 /*
 901  * rmdir and unlink are differ only by the metadata op code
 902  */
 903 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
 904 {
 905         struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
 906         struct ceph_mds_client *mdsc = fsc->mdsc;
 907         struct inode *inode = d_inode(dentry);
 908         struct ceph_mds_request *req;
 909         int err = -EROFS;
 910         int op;
 911
 912         if (ceph_snap(dir) == CEPH_SNAPDIR) {
 913                 /* rmdir .snap/foo is RMSNAP */
 914                 dout("rmsnap dir %p '%pd' dn %p\n", dir, dentry, dentry);
 915                 op = CEPH_MDS_OP_RMSNAP;
 916         } else if (ceph_snap(dir) == CEPH_NOSNAP) {
 917                 dout("unlink/rmdir dir %p dn %p inode %p\n",
 918                      dir, dentry, inode);
 919                 op = d_is_dir(dentry) ?
 920                         CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
 921         } else
 922                 goto out;
 923         req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
 924         if (IS_ERR(req)) {
 925                 err = PTR_ERR(req);
 926                 goto out;
 927         }
 928         req->r_dentry = dget(dentry);
 929         req->r_num_caps = 2;
 930         req->r_locked_dir = dir;
 931         req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
 932         req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
 933         req->r_inode_drop = drop_caps_for_unlink(inode);
 934         err = ceph_mdsc_do_request(mdsc, dir, req);
 935         if (!err && !req->r_reply_info.head->is_dentry)
 936                 d_delete(dentry);
 937         ceph_mdsc_put_request(req);
 938 out:
 939         return err;
 940 }
 941
 942 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
 943                        struct inode *new_dir, struct dentry *new_dentry)
 944 {
 945         struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
 946         struct ceph_mds_client *mdsc = fsc->mdsc;
 947         struct ceph_mds_request *req;
 948         int op = CEPH_MDS_OP_RENAME;
 949         int err;
 950
 951         if (ceph_snap(old_dir) != ceph_snap(new_dir))
 952                 return -EXDEV;
 953         if (ceph_snap(old_dir) != CEPH_NOSNAP) {
 954                 if (old_dir == new_dir && ceph_snap(old_dir) == CEPH_SNAPDIR)
 955                         op = CEPH_MDS_OP_RENAMESNAP;
 956                 else
 957                         return -EROFS;
 958         }
 959         dout("rename dir %p dentry %p to dir %p dentry %p\n",
 960              old_dir, old_dentry, new_dir, new_dentry);
 961         req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
 962         if (IS_ERR(req))
 963                 return PTR_ERR(req);
 964         ihold(old_dir);
 965         req->r_dentry = dget(new_dentry);
 966         req->r_num_caps = 2;
 967         req->r_old_dentry = dget(old_dentry);
 968         req->r_old_dentry_dir = old_dir;
 969         req->r_locked_dir = new_dir;
 970         req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
 971         req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
 972         req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
 973         req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
 974         /* release LINK_RDCACHE on source inode (mds will lock it) */
 975         req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
 976         if (d_really_is_positive(new_dentry))
 977                 req->r_inode_drop = drop_caps_for_unlink(d_inode(new_dentry));
 978         err = ceph_mdsc_do_request(mdsc, old_dir, req);
 979         if (!err && !req->r_reply_info.head->is_dentry) {
 980                 /*
 981                  * Normally d_move() is done by fill_trace (called by
 982                  * do_request, above).  If there is no trace, we need
 983                  * to do it here.
 984                  */
 985
 986                 /* d_move screws up sibling dentries' offsets */
 987                 ceph_dir_clear_complete(old_dir);
 988                 ceph_dir_clear_complete(new_dir);
 989
 990                 d_move(old_dentry, new_dentry);
 991
 992                 /* ensure target dentry is invalidated, despite
 993                    rehashing bug in vfs_rename_dir */
 994                 ceph_invalidate_dentry_lease(new_dentry);
 995         }
 996         ceph_mdsc_put_request(req);
 997         return err;
 998 }
 999
1000 /*
1001  * Ensure a dentry lease will no longer revalidate.
1002  */
1003 void ceph_invalidate_dentry_lease(struct dentry *dentry)
1004 {
1005         spin_lock(&dentry->d_lock);
1006         dentry->d_time = jiffies;
1007         ceph_dentry(dentry)->lease_shared_gen = 0;
1008         spin_unlock(&dentry->d_lock);
1009 }
1010
1011 /*
1012  * Check if dentry lease is valid.  If not, delete the lease.  Try to
1013  * renew if the least is more than half up.
1014  */
1015 static int dentry_lease_is_valid(struct dentry *dentry)
1016 {
1017         struct ceph_dentry_info *di;
1018         struct ceph_mds_session *s;
1019         int valid = 0;
1020         u32 gen;
1021         unsigned long ttl;
1022         struct ceph_mds_session *session = NULL;
1023         struct inode *dir = NULL;
1024         u32 seq = 0;
1025
1026         spin_lock(&dentry->d_lock);
1027         di = ceph_dentry(dentry);
1028         if (di->lease_session) {
1029                 s = di->lease_session;
1030                 spin_lock(&s->s_gen_ttl_lock);
1031                 gen = s->s_cap_gen;
1032                 ttl = s->s_cap_ttl;
1033                 spin_unlock(&s->s_gen_ttl_lock);
1034
1035                 if (di->lease_gen == gen &&
1036                     time_before(jiffies, dentry->d_time) &&
1037                     time_before(jiffies, ttl)) {
1038                         valid = 1;
1039                         if (di->lease_renew_after &&
1040                             time_after(jiffies, di->lease_renew_after)) {
1041                                 /* we should renew */
1042                                 dir = d_inode(dentry->d_parent);
1043                                 session = ceph_get_mds_session(s);
1044                                 seq = di->lease_seq;
1045                                 di->lease_renew_after = 0;
1046                                 di->lease_renew_from = jiffies;
1047                         }
1048                 }
1049         }
1050         spin_unlock(&dentry->d_lock);
1051
1052         if (session) {
1053                 ceph_mdsc_lease_send_msg(session, dir, dentry,
1054                                          CEPH_MDS_LEASE_RENEW, seq);
1055                 ceph_put_mds_session(session);
1056         }
1057         dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
1058         return valid;
1059 }
1060
1061 /*
1062  * Check if directory-wide content lease/cap is valid.
1063  */
1064 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
1065 {
1066         struct ceph_inode_info *ci = ceph_inode(dir);
1067         struct ceph_dentry_info *di = ceph_dentry(dentry);
1068         int valid = 0;
1069
1070         spin_lock(&ci->i_ceph_lock);
1071         if (ci->i_shared_gen == di->lease_shared_gen)
1072                 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
1073         spin_unlock(&ci->i_ceph_lock);
1074         dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
1075              dir, (unsigned)ci->i_shared_gen, dentry,
1076              (unsigned)di->lease_shared_gen, valid);
1077         return valid;
1078 }
1079
1080 /*
1081  * Check if cached dentry can be trusted.
1082  */
1083 static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)
1084 {
1085         int valid = 0;
1086         struct dentry *parent;
1087         struct inode *dir;
1088
1089         if (flags & LOOKUP_RCU)
1090                 return -ECHILD;
1091
1092         dout("d_revalidate %p '%pd' inode %p offset %lld\n", dentry,
1093              dentry, d_inode(dentry), ceph_dentry(dentry)->offset);
1094
1095         parent = dget_parent(dentry);
1096         dir = d_inode(parent);
1097
1098         /* always trust cached snapped dentries, snapdir dentry */
1099         if (ceph_snap(dir) != CEPH_NOSNAP) {
1100                 dout("d_revalidate %p '%pd' inode %p is SNAPPED\n", dentry,
1101                      dentry, d_inode(dentry));
1102                 valid = 1;
1103         } else if (d_really_is_positive(dentry) &&
1104                    ceph_snap(d_inode(dentry)) == CEPH_SNAPDIR) {
1105                 valid = 1;
1106         } else if (dentry_lease_is_valid(dentry) ||
1107                    dir_lease_is_valid(dir, dentry)) {
1108                 if (d_really_is_positive(dentry))
1109                         valid = ceph_is_any_caps(d_inode(dentry));
1110                 else
1111                         valid = 1;
1112         }
1113
1114         if (!valid) {
1115                 struct ceph_mds_client *mdsc =
1116                         ceph_sb_to_client(dir->i_sb)->mdsc;
1117                 struct ceph_mds_request *req;
1118                 int op, mask, err;
1119
1120                 op = ceph_snap(dir) == CEPH_SNAPDIR ?
1121                         CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
1122                 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
1123                 if (!IS_ERR(req)) {
1124                         req->r_dentry = dget(dentry);
1125                         req->r_num_caps = 2;
1126
1127                         mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
1128                         if (ceph_security_xattr_wanted(dir))
1129                                 mask |= CEPH_CAP_XATTR_SHARED;
1130                         req->r_args.getattr.mask = mask;
1131
1132                         req->r_locked_dir = dir;
1133                         err = ceph_mdsc_do_request(mdsc, NULL, req);
1134                         if (err == 0 || err == -ENOENT) {
1135                                 if (dentry == req->r_dentry) {
1136                                         valid = !d_unhashed(dentry);
1137                                 } else {
1138                                         d_invalidate(req->r_dentry);
1139                                         err = -EAGAIN;
1140                                 }
1141                         }
1142                         ceph_mdsc_put_request(req);
1143                         dout("d_revalidate %p lookup result=%d\n",
1144                              dentry, err);
1145                 }
1146         }
1147
1148         dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid");
1149         if (valid) {
1150                 ceph_dentry_lru_touch(dentry);
1151         } else {
1152                 ceph_dir_clear_complete(dir);
1153         }
1154
1155         dput(parent);
1156         return valid;
1157 }
1158
1159 /*
1160  * Release our ceph_dentry_info.
1161  */
1162 static void ceph_d_release(struct dentry *dentry)
1163 {
1164         struct ceph_dentry_info *di = ceph_dentry(dentry);
1165
1166         dout("d_release %p\n", dentry);
1167         ceph_dentry_lru_del(dentry);
1168         if (di->lease_session)
1169                 ceph_put_mds_session(di->lease_session);
1170         kmem_cache_free(ceph_dentry_cachep, di);
1171         dentry->d_fsdata = NULL;
1172 }
1173
1174 static int ceph_snapdir_d_revalidate(struct dentry *dentry,
1175                                           unsigned int flags)
1176 {
1177         /*
1178          * Eventually, we'll want to revalidate snapped metadata
1179          * too... probably...
1180          */
1181         return 1;
1182 }
1183
1184 /*
1185  * When the VFS prunes a dentry from the cache, we need to clear the
1186  * complete flag on the parent directory.
1187  *
1188  * Called under dentry->d_lock.
1189  */
1190 static void ceph_d_prune(struct dentry *dentry)
1191 {
1192         dout("ceph_d_prune %p\n", dentry);
1193
1194         /* do we have a valid parent? */
1195         if (IS_ROOT(dentry))
1196                 return;
1197
1198         /* if we are not hashed, we don't affect dir's completeness */
1199         if (d_unhashed(dentry))
1200                 return;
1201
1202         /*
1203          * we hold d_lock, so d_parent is stable, and d_fsdata is never
1204          * cleared until d_release
1205          */
1206         ceph_dir_clear_complete(d_inode(dentry->d_parent));
1207 }
1208
1209 /*
1210  * read() on a dir.  This weird interface hack only works if mounted
1211  * with '-o dirstat'.
1212  */
1213 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1214                              loff_t *ppos)
1215 {
1216         struct ceph_file_info *cf = file->private_data;
1217         struct inode *inode = file_inode(file);
1218         struct ceph_inode_info *ci = ceph_inode(inode);
1219         int left;
1220         const int bufsize = 1024;
1221
1222         if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1223                 return -EISDIR;
1224
1225         if (!cf->dir_info) {
1226                 cf->dir_info = kmalloc(bufsize, GFP_KERNEL);
1227                 if (!cf->dir_info)
1228                         return -ENOMEM;
1229                 cf->dir_info_len =
1230                         snprintf(cf->dir_info, bufsize,
1231                                 "entries:   %20lld\n"
1232                                 " files:    %20lld\n"
1233                                 " subdirs:  %20lld\n"
1234                                 "rentries:  %20lld\n"
1235                                 " rfiles:   %20lld\n"
1236                                 " rsubdirs: %20lld\n"
1237                                 "rbytes:    %20lld\n"
1238                                 "rctime:    %10ld.%09ld\n",
1239                                 ci->i_files + ci->i_subdirs,
1240                                 ci->i_files,
1241                                 ci->i_subdirs,
1242                                 ci->i_rfiles + ci->i_rsubdirs,
1243                                 ci->i_rfiles,
1244                                 ci->i_rsubdirs,
1245                                 ci->i_rbytes,
1246                                 (long)ci->i_rctime.tv_sec,
1247                                 (long)ci->i_rctime.tv_nsec);
1248         }
1249
1250         if (*ppos >= cf->dir_info_len)
1251                 return 0;
1252         size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1253         left = copy_to_user(buf, cf->dir_info + *ppos, size);
1254         if (left == size)
1255                 return -EFAULT;
1256         *ppos += (size - left);
1257         return size - left;
1258 }
1259
1260 /*
1261  * We maintain a private dentry LRU.
1262  *
1263  * FIXME: this needs to be changed to a per-mds lru to be useful.
1264  */
1265 void ceph_dentry_lru_add(struct dentry *dn)
1266 {
1267         struct ceph_dentry_info *di = ceph_dentry(dn);
1268         struct ceph_mds_client *mdsc;
1269
1270         dout("dentry_lru_add %p %p '%pd'\n", di, dn, dn);
1271         mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1272         spin_lock(&mdsc->dentry_lru_lock);
1273         list_add_tail(&di->lru, &mdsc->dentry_lru);
1274         mdsc->num_dentry++;
1275         spin_unlock(&mdsc->dentry_lru_lock);
1276 }
1277
1278 void ceph_dentry_lru_touch(struct dentry *dn)
1279 {
1280         struct ceph_dentry_info *di = ceph_dentry(dn);
1281         struct ceph_mds_client *mdsc;
1282
1283         dout("dentry_lru_touch %p %p '%pd' (offset %lld)\n", di, dn, dn,
1284              di->offset);
1285         mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1286         spin_lock(&mdsc->dentry_lru_lock);
1287         list_move_tail(&di->lru, &mdsc->dentry_lru);
1288         spin_unlock(&mdsc->dentry_lru_lock);
1289 }
1290
1291 void ceph_dentry_lru_del(struct dentry *dn)
1292 {
1293         struct ceph_dentry_info *di = ceph_dentry(dn);
1294         struct ceph_mds_client *mdsc;
1295
1296         dout("dentry_lru_del %p %p '%pd'\n", di, dn, dn);
1297         mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1298         spin_lock(&mdsc->dentry_lru_lock);
1299         list_del_init(&di->lru);
1300         mdsc->num_dentry--;
1301         spin_unlock(&mdsc->dentry_lru_lock);
1302 }
1303
1304 /*
1305  * Return name hash for a given dentry.  This is dependent on
1306  * the parent directory's hash function.
1307  */
1308 unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn)
1309 {
1310         struct ceph_inode_info *dci = ceph_inode(dir);
1311
1312         switch (dci->i_dir_layout.dl_dir_hash) {
1313         case 0: /* for backward compat */
1314         case CEPH_STR_HASH_LINUX:
1315                 return dn->d_name.hash;
1316
1317         default:
1318                 return ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
1319                                      dn->d_name.name, dn->d_name.len);
1320         }
1321 }
1322
1323 const struct file_operations ceph_dir_fops = {
1324         .read = ceph_read_dir,
1325         .iterate = ceph_readdir,
1326         .llseek = ceph_dir_llseek,
1327         .open = ceph_open,
1328         .release = ceph_release,
1329         .unlocked_ioctl = ceph_ioctl,
1330         .fsync = ceph_fsync,
1331 };
1332
1333 const struct file_operations ceph_snapdir_fops = {
1334         .iterate = ceph_readdir,
1335         .llseek = ceph_dir_llseek,
1336         .open = ceph_open,
1337         .release = ceph_release,
1338 };
1339
1340 const struct inode_operations ceph_dir_iops = {
1341         .lookup = ceph_lookup,
1342         .permission = ceph_permission,
1343         .getattr = ceph_getattr,
1344         .setattr = ceph_setattr,
1345         .setxattr = ceph_setxattr,
1346         .getxattr = ceph_getxattr,
1347         .listxattr = ceph_listxattr,
1348         .removexattr = ceph_removexattr,
1349         .get_acl = ceph_get_acl,
1350         .set_acl = ceph_set_acl,
1351         .mknod = ceph_mknod,
1352         .symlink = ceph_symlink,
1353         .mkdir = ceph_mkdir,
1354         .link = ceph_link,
1355         .unlink = ceph_unlink,
1356         .rmdir = ceph_unlink,
1357         .rename = ceph_rename,
1358         .create = ceph_create,
1359         .atomic_open = ceph_atomic_open,
1360 };
1361
1362 const struct inode_operations ceph_snapdir_iops = {
1363         .lookup = ceph_lookup,
1364         .permission = ceph_permission,
1365         .getattr = ceph_getattr,
1366         .mkdir = ceph_mkdir,
1367         .rmdir = ceph_unlink,
1368         .rename = ceph_rename,
1369 };
1370
1371 const struct dentry_operations ceph_dentry_ops = {
1372         .d_revalidate = ceph_d_revalidate,
1373         .d_release = ceph_d_release,
1374         .d_prune = ceph_d_prune,
1375 };
1376
1377 const struct dentry_operations ceph_snapdir_dentry_ops = {
1378         .d_revalidate = ceph_snapdir_d_revalidate,
1379         .d_release = ceph_d_release,
1380 };
1381
1382 const struct dentry_operations ceph_snap_dentry_ops = {
1383         .d_release = ceph_d_release,
1384         .d_prune = ceph_d_prune,
1385 };