1 #include "ceph_debug.h" 2 3 #include <linux/spinlock.h> 4 #include <linux/fs_struct.h> 5 #include <linux/namei.h> 6 #include <linux/sched.h> 7 8 #include "super.h" 9 10 /* 11 * Directory operations: readdir, lookup, create, link, unlink, 12 * rename, etc. 13 */ 14 15 /* 16 * Ceph MDS operations are specified in terms of a base ino and 17 * relative path. Thus, the client can specify an operation on a 18 * specific inode (e.g., a getattr due to fstat(2)), or as a path 19 * relative to, say, the root directory. 20 * 21 * Normally, we limit ourselves to strict inode ops (no path component) 22 * or dentry operations (a single path component relative to an ino). The 23 * exception to this is open_root_dentry(), which will open the mount 24 * point by name. 25 */ 26 27 const struct inode_operations ceph_dir_iops; 28 const struct file_operations ceph_dir_fops; 29 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 if (ceph_snap(dentry->d_parent->d_inode) == CEPH_NOSNAP) 42 dentry->d_op = &ceph_dentry_ops; 43 else if (ceph_snap(dentry->d_parent->d_inode) == CEPH_SNAPDIR) 44 dentry->d_op = &ceph_snapdir_dentry_ops; 45 else 46 dentry->d_op = &ceph_snap_dentry_ops; 47 48 di = kmem_cache_alloc(ceph_dentry_cachep, GFP_NOFS); 49 if (!di) 50 return -ENOMEM; /* oh well */ 51 52 spin_lock(&dentry->d_lock); 53 if (dentry->d_fsdata) /* lost a race */ 54 goto out_unlock; 55 di->dentry = dentry; 56 di->lease_session = NULL; 57 dentry->d_fsdata = di; 58 dentry->d_time = jiffies; 59 ceph_dentry_lru_add(dentry); 60 out_unlock: 61 spin_unlock(&dentry->d_lock); 62 return 0; 63 } 64 65 66 67 /* 68 * for readdir, we encode the directory frag and offset within that 69 * frag into f_pos. 70 */ 71 static unsigned fpos_frag(loff_t p) 72 { 73 return p >> 32; 74 } 75 static unsigned fpos_off(loff_t p) 76 { 77 return p & 0xffffffff; 78 } 79 80 /* 81 * When possible, we try to satisfy a readdir by peeking at the 82 * dcache. We make this work by carefully ordering dentries on 83 * d_u.d_child when we initially get results back from the MDS, and 84 * falling back to a "normal" sync readdir if any dentries in the dir 85 * are dropped. 86 * 87 * I_COMPLETE tells indicates we have all dentries in the dir. It is 88 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by 89 * the MDS if/when the directory is modified). 90 */ 91 static int __dcache_readdir(struct file *filp, 92 void *dirent, filldir_t filldir) 93 { 94 struct inode *inode = filp->f_dentry->d_inode; 95 struct ceph_file_info *fi = filp->private_data; 96 struct dentry *parent = filp->f_dentry; 97 struct inode *dir = parent->d_inode; 98 struct list_head *p; 99 struct dentry *dentry, *last; 100 struct ceph_dentry_info *di; 101 int err = 0; 102 103 /* claim ref on last dentry we returned */ 104 last = fi->dentry; 105 fi->dentry = NULL; 106 107 dout("__dcache_readdir %p at %llu (last %p)\n", dir, filp->f_pos, 108 last); 109 110 spin_lock(&dcache_lock); 111 112 /* start at beginning? */ 113 if (filp->f_pos == 2 || (last && 114 filp->f_pos < ceph_dentry(last)->offset)) { 115 if (list_empty(&parent->d_subdirs)) 116 goto out_unlock; 117 p = parent->d_subdirs.prev; 118 dout(" initial p %p/%p\n", p->prev, p->next); 119 } else { 120 p = last->d_u.d_child.prev; 121 } 122 123 more: 124 dentry = list_entry(p, struct dentry, d_u.d_child); 125 di = ceph_dentry(dentry); 126 while (1) { 127 dout(" p %p/%p d_subdirs %p/%p\n", p->prev, p->next, 128 parent->d_subdirs.prev, parent->d_subdirs.next); 129 if (p == &parent->d_subdirs) { 130 fi->at_end = 1; 131 goto out_unlock; 132 } 133 if (!d_unhashed(dentry) && dentry->d_inode && 134 ceph_snap(dentry->d_inode) != CEPH_SNAPDIR && 135 ceph_ino(dentry->d_inode) != CEPH_INO_CEPH && 136 filp->f_pos <= di->offset) 137 break; 138 dout(" skipping %p %.*s at %llu (%llu)%s%s\n", dentry, 139 dentry->d_name.len, dentry->d_name.name, di->offset, 140 filp->f_pos, d_unhashed(dentry) ? " unhashed" : "", 141 !dentry->d_inode ? " null" : ""); 142 p = p->prev; 143 dentry = list_entry(p, struct dentry, d_u.d_child); 144 di = ceph_dentry(dentry); 145 } 146 147 atomic_inc(&dentry->d_count); 148 spin_unlock(&dcache_lock); 149 spin_unlock(&inode->i_lock); 150 151 dout(" %llu (%llu) dentry %p %.*s %p\n", di->offset, filp->f_pos, 152 dentry, dentry->d_name.len, dentry->d_name.name, dentry->d_inode); 153 filp->f_pos = di->offset; 154 err = filldir(dirent, dentry->d_name.name, 155 dentry->d_name.len, di->offset, 156 dentry->d_inode->i_ino, 157 dentry->d_inode->i_mode >> 12); 158 159 if (last) { 160 if (err < 0) { 161 /* remember our position */ 162 fi->dentry = last; 163 fi->next_offset = di->offset; 164 } else { 165 dput(last); 166 } 167 last = NULL; 168 } 169 170 spin_lock(&inode->i_lock); 171 spin_lock(&dcache_lock); 172 173 if (err < 0) 174 goto out_unlock; 175 176 last = dentry; 177 178 p = p->prev; 179 filp->f_pos++; 180 181 /* make sure a dentry wasn't dropped while we didn't have dcache_lock */ 182 if ((ceph_inode(dir)->i_ceph_flags & CEPH_I_COMPLETE)) 183 goto more; 184 dout(" lost I_COMPLETE on %p; falling back to mds\n", dir); 185 err = -EAGAIN; 186 187 out_unlock: 188 spin_unlock(&dcache_lock); 189 190 if (last) { 191 spin_unlock(&inode->i_lock); 192 dput(last); 193 spin_lock(&inode->i_lock); 194 } 195 196 return err; 197 } 198 199 /* 200 * make note of the last dentry we read, so we can 201 * continue at the same lexicographical point, 202 * regardless of what dir changes take place on the 203 * server. 204 */ 205 static int note_last_dentry(struct ceph_file_info *fi, const char *name, 206 int len) 207 { 208 kfree(fi->last_name); 209 fi->last_name = kmalloc(len+1, GFP_NOFS); 210 if (!fi->last_name) 211 return -ENOMEM; 212 memcpy(fi->last_name, name, len); 213 fi->last_name[len] = 0; 214 dout("note_last_dentry '%s'\n", fi->last_name); 215 return 0; 216 } 217 218 static int ceph_readdir(struct file *filp, void *dirent, filldir_t filldir) 219 { 220 struct ceph_file_info *fi = filp->private_data; 221 struct inode *inode = filp->f_dentry->d_inode; 222 struct ceph_inode_info *ci = ceph_inode(inode); 223 struct ceph_client *client = ceph_inode_to_client(inode); 224 struct ceph_mds_client *mdsc = &client->mdsc; 225 unsigned frag = fpos_frag(filp->f_pos); 226 int off = fpos_off(filp->f_pos); 227 int err; 228 u32 ftype; 229 struct ceph_mds_reply_info_parsed *rinfo; 230 const int max_entries = client->mount_args->max_readdir; 231 232 dout("readdir %p filp %p frag %u off %u\n", inode, filp, frag, off); 233 if (fi->at_end) 234 return 0; 235 236 /* always start with . and .. */ 237 if (filp->f_pos == 0) { 238 /* note dir version at start of readdir so we can tell 239 * if any dentries get dropped */ 240 fi->dir_release_count = ci->i_release_count; 241 242 dout("readdir off 0 -> '.'\n"); 243 if (filldir(dirent, ".", 1, ceph_make_fpos(0, 0), 244 inode->i_ino, inode->i_mode >> 12) < 0) 245 return 0; 246 filp->f_pos = 1; 247 off = 1; 248 } 249 if (filp->f_pos == 1) { 250 dout("readdir off 1 -> '..'\n"); 251 if (filldir(dirent, "..", 2, ceph_make_fpos(0, 1), 252 filp->f_dentry->d_parent->d_inode->i_ino, 253 inode->i_mode >> 12) < 0) 254 return 0; 255 filp->f_pos = 2; 256 off = 2; 257 } 258 259 /* can we use the dcache? */ 260 spin_lock(&inode->i_lock); 261 if ((filp->f_pos == 2 || fi->dentry) && 262 !ceph_test_opt(client, NOASYNCREADDIR) && 263 (ci->i_ceph_flags & CEPH_I_COMPLETE) && 264 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) { 265 err = __dcache_readdir(filp, dirent, filldir); 266 if (err != -EAGAIN) { 267 spin_unlock(&inode->i_lock); 268 return err; 269 } 270 } 271 spin_unlock(&inode->i_lock); 272 if (fi->dentry) { 273 err = note_last_dentry(fi, fi->dentry->d_name.name, 274 fi->dentry->d_name.len); 275 if (err) 276 return err; 277 dput(fi->dentry); 278 fi->dentry = NULL; 279 } 280 281 /* proceed with a normal readdir */ 282 283 more: 284 /* do we have the correct frag content buffered? */ 285 if (fi->frag != frag || fi->last_readdir == NULL) { 286 struct ceph_mds_request *req; 287 int op = ceph_snap(inode) == CEPH_SNAPDIR ? 288 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR; 289 290 /* discard old result, if any */ 291 if (fi->last_readdir) 292 ceph_mdsc_put_request(fi->last_readdir); 293 294 /* requery frag tree, as the frag topology may have changed */ 295 frag = ceph_choose_frag(ceph_inode(inode), frag, NULL, NULL); 296 297 dout("readdir fetching %llx.%llx frag %x offset '%s'\n", 298 ceph_vinop(inode), frag, fi->last_name); 299 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS); 300 if (IS_ERR(req)) 301 return PTR_ERR(req); 302 req->r_inode = igrab(inode); 303 req->r_dentry = dget(filp->f_dentry); 304 /* hints to request -> mds selection code */ 305 req->r_direct_mode = USE_AUTH_MDS; 306 req->r_direct_hash = ceph_frag_value(frag); 307 req->r_direct_is_hash = true; 308 req->r_path2 = kstrdup(fi->last_name, GFP_NOFS); 309 req->r_readdir_offset = fi->next_offset; 310 req->r_args.readdir.frag = cpu_to_le32(frag); 311 req->r_args.readdir.max_entries = cpu_to_le32(max_entries); 312 req->r_num_caps = max_entries; 313 err = ceph_mdsc_do_request(mdsc, NULL, req); 314 if (err < 0) { 315 ceph_mdsc_put_request(req); 316 return err; 317 } 318 dout("readdir got and parsed readdir result=%d" 319 " on frag %x, end=%d, complete=%d\n", err, frag, 320 (int)req->r_reply_info.dir_end, 321 (int)req->r_reply_info.dir_complete); 322 323 if (!req->r_did_prepopulate) { 324 dout("readdir !did_prepopulate"); 325 fi->dir_release_count--; /* preclude I_COMPLETE */ 326 } 327 328 /* note next offset and last dentry name */ 329 fi->offset = fi->next_offset; 330 fi->last_readdir = req; 331 332 if (req->r_reply_info.dir_end) { 333 kfree(fi->last_name); 334 fi->last_name = NULL; 335 fi->next_offset = 0; 336 } else { 337 rinfo = &req->r_reply_info; 338 err = note_last_dentry(fi, 339 rinfo->dir_dname[rinfo->dir_nr-1], 340 rinfo->dir_dname_len[rinfo->dir_nr-1]); 341 if (err) 342 return err; 343 fi->next_offset += rinfo->dir_nr; 344 } 345 } 346 347 rinfo = &fi->last_readdir->r_reply_info; 348 dout("readdir frag %x num %d off %d chunkoff %d\n", frag, 349 rinfo->dir_nr, off, fi->offset); 350 while (off - fi->offset >= 0 && off - fi->offset < rinfo->dir_nr) { 351 u64 pos = ceph_make_fpos(frag, off); 352 struct ceph_mds_reply_inode *in = 353 rinfo->dir_in[off - fi->offset].in; 354 dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n", 355 off, off - fi->offset, rinfo->dir_nr, pos, 356 rinfo->dir_dname_len[off - fi->offset], 357 rinfo->dir_dname[off - fi->offset], in); 358 BUG_ON(!in); 359 ftype = le32_to_cpu(in->mode) >> 12; 360 if (filldir(dirent, 361 rinfo->dir_dname[off - fi->offset], 362 rinfo->dir_dname_len[off - fi->offset], 363 pos, 364 le64_to_cpu(in->ino), 365 ftype) < 0) { 366 dout("filldir stopping us...\n"); 367 return 0; 368 } 369 off++; 370 filp->f_pos = pos + 1; 371 } 372 373 if (fi->last_name) { 374 ceph_mdsc_put_request(fi->last_readdir); 375 fi->last_readdir = NULL; 376 goto more; 377 } 378 379 /* more frags? */ 380 if (!ceph_frag_is_rightmost(frag)) { 381 frag = ceph_frag_next(frag); 382 off = 0; 383 filp->f_pos = ceph_make_fpos(frag, off); 384 dout("readdir next frag is %x\n", frag); 385 goto more; 386 } 387 fi->at_end = 1; 388 389 /* 390 * if dir_release_count still matches the dir, no dentries 391 * were released during the whole readdir, and we should have 392 * the complete dir contents in our cache. 393 */ 394 spin_lock(&inode->i_lock); 395 if (ci->i_release_count == fi->dir_release_count) { 396 dout(" marking %p complete\n", inode); 397 ci->i_ceph_flags |= CEPH_I_COMPLETE; 398 ci->i_max_offset = filp->f_pos; 399 } 400 spin_unlock(&inode->i_lock); 401 402 dout("readdir %p filp %p done.\n", inode, filp); 403 return 0; 404 } 405 406 static void reset_readdir(struct ceph_file_info *fi) 407 { 408 if (fi->last_readdir) { 409 ceph_mdsc_put_request(fi->last_readdir); 410 fi->last_readdir = NULL; 411 } 412 kfree(fi->last_name); 413 fi->next_offset = 2; /* compensate for . and .. */ 414 if (fi->dentry) { 415 dput(fi->dentry); 416 fi->dentry = NULL; 417 } 418 fi->at_end = 0; 419 } 420 421 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int origin) 422 { 423 struct ceph_file_info *fi = file->private_data; 424 struct inode *inode = file->f_mapping->host; 425 loff_t old_offset = offset; 426 loff_t retval; 427 428 mutex_lock(&inode->i_mutex); 429 switch (origin) { 430 case SEEK_END: 431 offset += inode->i_size + 2; /* FIXME */ 432 break; 433 case SEEK_CUR: 434 offset += file->f_pos; 435 } 436 retval = -EINVAL; 437 if (offset >= 0 && offset <= inode->i_sb->s_maxbytes) { 438 if (offset != file->f_pos) { 439 file->f_pos = offset; 440 file->f_version = 0; 441 fi->at_end = 0; 442 } 443 retval = offset; 444 445 /* 446 * discard buffered readdir content on seekdir(0), or 447 * seek to new frag, or seek prior to current chunk. 448 */ 449 if (offset == 0 || 450 fpos_frag(offset) != fpos_frag(old_offset) || 451 fpos_off(offset) < fi->offset) { 452 dout("dir_llseek dropping %p content\n", file); 453 reset_readdir(fi); 454 } 455 456 /* bump dir_release_count if we did a forward seek */ 457 if (offset > old_offset) 458 fi->dir_release_count--; 459 } 460 mutex_unlock(&inode->i_mutex); 461 return retval; 462 } 463 464 /* 465 * Process result of a lookup/open request. 466 * 467 * Mainly, make sure we return the final req->r_dentry (if it already 468 * existed) in place of the original VFS-provided dentry when they 469 * differ. 470 * 471 * Gracefully handle the case where the MDS replies with -ENOENT and 472 * no trace (which it may do, at its discretion, e.g., if it doesn't 473 * care to issue a lease on the negative dentry). 474 */ 475 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req, 476 struct dentry *dentry, int err) 477 { 478 struct ceph_client *client = ceph_client(dentry->d_sb); 479 struct inode *parent = dentry->d_parent->d_inode; 480 481 /* .snap dir? */ 482 if (err == -ENOENT && 483 ceph_vino(parent).ino != CEPH_INO_ROOT && /* no .snap in root dir */ 484 strcmp(dentry->d_name.name, 485 client->mount_args->snapdir_name) == 0) { 486 struct inode *inode = ceph_get_snapdir(parent); 487 dout("ENOENT on snapdir %p '%.*s', linking to snapdir %p\n", 488 dentry, dentry->d_name.len, dentry->d_name.name, inode); 489 d_add(dentry, inode); 490 err = 0; 491 } 492 493 if (err == -ENOENT) { 494 /* no trace? */ 495 err = 0; 496 if (!req->r_reply_info.head->is_dentry) { 497 dout("ENOENT and no trace, dentry %p inode %p\n", 498 dentry, dentry->d_inode); 499 if (dentry->d_inode) { 500 d_drop(dentry); 501 err = -ENOENT; 502 } else { 503 d_add(dentry, NULL); 504 } 505 } 506 } 507 if (err) 508 dentry = ERR_PTR(err); 509 else if (dentry != req->r_dentry) 510 dentry = dget(req->r_dentry); /* we got spliced */ 511 else 512 dentry = NULL; 513 return dentry; 514 } 515 516 static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry) 517 { 518 return ceph_ino(inode) == CEPH_INO_ROOT && 519 strncmp(dentry->d_name.name, ".ceph", 5) == 0; 520 } 521 522 /* 523 * Look up a single dir entry. If there is a lookup intent, inform 524 * the MDS so that it gets our 'caps wanted' value in a single op. 525 */ 526 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry, 527 struct nameidata *nd) 528 { 529 struct ceph_client *client = ceph_sb_to_client(dir->i_sb); 530 struct ceph_mds_client *mdsc = &client->mdsc; 531 struct ceph_mds_request *req; 532 int op; 533 int err; 534 535 dout("lookup %p dentry %p '%.*s'\n", 536 dir, dentry, dentry->d_name.len, dentry->d_name.name); 537 538 if (dentry->d_name.len > NAME_MAX) 539 return ERR_PTR(-ENAMETOOLONG); 540 541 err = ceph_init_dentry(dentry); 542 if (err < 0) 543 return ERR_PTR(err); 544 545 /* open (but not create!) intent? */ 546 if (nd && 547 (nd->flags & LOOKUP_OPEN) && 548 (nd->flags & LOOKUP_CONTINUE) == 0 && /* only open last component */ 549 !(nd->intent.open.flags & O_CREAT)) { 550 int mode = nd->intent.open.create_mode & ~current->fs->umask; 551 return ceph_lookup_open(dir, dentry, nd, mode, 1); 552 } 553 554 /* can we conclude ENOENT locally? */ 555 if (dentry->d_inode == NULL) { 556 struct ceph_inode_info *ci = ceph_inode(dir); 557 struct ceph_dentry_info *di = ceph_dentry(dentry); 558 559 spin_lock(&dir->i_lock); 560 dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags); 561 if (strncmp(dentry->d_name.name, 562 client->mount_args->snapdir_name, 563 dentry->d_name.len) && 564 !is_root_ceph_dentry(dir, dentry) && 565 (ci->i_ceph_flags & CEPH_I_COMPLETE) && 566 (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) { 567 di->offset = ci->i_max_offset++; 568 spin_unlock(&dir->i_lock); 569 dout(" dir %p complete, -ENOENT\n", dir); 570 d_add(dentry, NULL); 571 di->lease_shared_gen = ci->i_shared_gen; 572 return NULL; 573 } 574 spin_unlock(&dir->i_lock); 575 } 576 577 op = ceph_snap(dir) == CEPH_SNAPDIR ? 578 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP; 579 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS); 580 if (IS_ERR(req)) 581 return ERR_PTR(PTR_ERR(req)); 582 req->r_dentry = dget(dentry); 583 req->r_num_caps = 2; 584 /* we only need inode linkage */ 585 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE); 586 req->r_locked_dir = dir; 587 err = ceph_mdsc_do_request(mdsc, NULL, req); 588 dentry = ceph_finish_lookup(req, dentry, err); 589 ceph_mdsc_put_request(req); /* will dput(dentry) */ 590 dout("lookup result=%p\n", dentry); 591 return dentry; 592 } 593 594 /* 595 * If we do a create but get no trace back from the MDS, follow up with 596 * a lookup (the VFS expects us to link up the provided dentry). 597 */ 598 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry) 599 { 600 struct dentry *result = ceph_lookup(dir, dentry, NULL); 601 602 if (result && !IS_ERR(result)) { 603 /* 604 * We created the item, then did a lookup, and found 605 * it was already linked to another inode we already 606 * had in our cache (and thus got spliced). Link our 607 * dentry to that inode, but don't hash it, just in 608 * case the VFS wants to dereference it. 609 */ 610 BUG_ON(!result->d_inode); 611 d_instantiate(dentry, result->d_inode); 612 return 0; 613 } 614 return PTR_ERR(result); 615 } 616 617 static int ceph_mknod(struct inode *dir, struct dentry *dentry, 618 int mode, dev_t rdev) 619 { 620 struct ceph_client *client = ceph_sb_to_client(dir->i_sb); 621 struct ceph_mds_client *mdsc = &client->mdsc; 622 struct ceph_mds_request *req; 623 int err; 624 625 if (ceph_snap(dir) != CEPH_NOSNAP) 626 return -EROFS; 627 628 dout("mknod in dir %p dentry %p mode 0%o rdev %d\n", 629 dir, dentry, mode, rdev); 630 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS); 631 if (IS_ERR(req)) { 632 d_drop(dentry); 633 return PTR_ERR(req); 634 } 635 req->r_dentry = dget(dentry); 636 req->r_num_caps = 2; 637 req->r_locked_dir = dir; 638 req->r_args.mknod.mode = cpu_to_le32(mode); 639 req->r_args.mknod.rdev = cpu_to_le32(rdev); 640 req->r_dentry_drop = CEPH_CAP_FILE_SHARED; 641 req->r_dentry_unless = CEPH_CAP_FILE_EXCL; 642 err = ceph_mdsc_do_request(mdsc, dir, req); 643 if (!err && !req->r_reply_info.head->is_dentry) 644 err = ceph_handle_notrace_create(dir, dentry); 645 ceph_mdsc_put_request(req); 646 if (err) 647 d_drop(dentry); 648 return err; 649 } 650 651 static int ceph_create(struct inode *dir, struct dentry *dentry, int mode, 652 struct nameidata *nd) 653 { 654 dout("create in dir %p dentry %p name '%.*s'\n", 655 dir, dentry, dentry->d_name.len, dentry->d_name.name); 656 657 if (ceph_snap(dir) != CEPH_NOSNAP) 658 return -EROFS; 659 660 if (nd) { 661 BUG_ON((nd->flags & LOOKUP_OPEN) == 0); 662 dentry = ceph_lookup_open(dir, dentry, nd, mode, 0); 663 /* hrm, what should i do here if we get aliased? */ 664 if (IS_ERR(dentry)) 665 return PTR_ERR(dentry); 666 return 0; 667 } 668 669 /* fall back to mknod */ 670 return ceph_mknod(dir, dentry, (mode & ~S_IFMT) | S_IFREG, 0); 671 } 672 673 static int ceph_symlink(struct inode *dir, struct dentry *dentry, 674 const char *dest) 675 { 676 struct ceph_client *client = ceph_sb_to_client(dir->i_sb); 677 struct ceph_mds_client *mdsc = &client->mdsc; 678 struct ceph_mds_request *req; 679 int err; 680 681 if (ceph_snap(dir) != CEPH_NOSNAP) 682 return -EROFS; 683 684 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest); 685 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS); 686 if (IS_ERR(req)) { 687 d_drop(dentry); 688 return PTR_ERR(req); 689 } 690 req->r_dentry = dget(dentry); 691 req->r_num_caps = 2; 692 req->r_path2 = kstrdup(dest, GFP_NOFS); 693 req->r_locked_dir = dir; 694 req->r_dentry_drop = CEPH_CAP_FILE_SHARED; 695 req->r_dentry_unless = CEPH_CAP_FILE_EXCL; 696 err = ceph_mdsc_do_request(mdsc, dir, req); 697 if (!err && !req->r_reply_info.head->is_dentry) 698 err = ceph_handle_notrace_create(dir, dentry); 699 ceph_mdsc_put_request(req); 700 if (err) 701 d_drop(dentry); 702 return err; 703 } 704 705 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, int mode) 706 { 707 struct ceph_client *client = ceph_sb_to_client(dir->i_sb); 708 struct ceph_mds_client *mdsc = &client->mdsc; 709 struct ceph_mds_request *req; 710 int err = -EROFS; 711 int op; 712 713 if (ceph_snap(dir) == CEPH_SNAPDIR) { 714 /* mkdir .snap/foo is a MKSNAP */ 715 op = CEPH_MDS_OP_MKSNAP; 716 dout("mksnap dir %p snap '%.*s' dn %p\n", dir, 717 dentry->d_name.len, dentry->d_name.name, dentry); 718 } else if (ceph_snap(dir) == CEPH_NOSNAP) { 719 dout("mkdir dir %p dn %p mode 0%o\n", dir, dentry, mode); 720 op = CEPH_MDS_OP_MKDIR; 721 } else { 722 goto out; 723 } 724 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS); 725 if (IS_ERR(req)) { 726 err = PTR_ERR(req); 727 goto out; 728 } 729 730 req->r_dentry = dget(dentry); 731 req->r_num_caps = 2; 732 req->r_locked_dir = dir; 733 req->r_args.mkdir.mode = cpu_to_le32(mode); 734 req->r_dentry_drop = CEPH_CAP_FILE_SHARED; 735 req->r_dentry_unless = CEPH_CAP_FILE_EXCL; 736 err = ceph_mdsc_do_request(mdsc, dir, req); 737 if (!err && !req->r_reply_info.head->is_dentry) 738 err = ceph_handle_notrace_create(dir, dentry); 739 ceph_mdsc_put_request(req); 740 out: 741 if (err < 0) 742 d_drop(dentry); 743 return err; 744 } 745 746 static int ceph_link(struct dentry *old_dentry, struct inode *dir, 747 struct dentry *dentry) 748 { 749 struct ceph_client *client = ceph_sb_to_client(dir->i_sb); 750 struct ceph_mds_client *mdsc = &client->mdsc; 751 struct ceph_mds_request *req; 752 int err; 753 754 if (ceph_snap(dir) != CEPH_NOSNAP) 755 return -EROFS; 756 757 dout("link in dir %p old_dentry %p dentry %p\n", dir, 758 old_dentry, dentry); 759 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS); 760 if (IS_ERR(req)) { 761 d_drop(dentry); 762 return PTR_ERR(req); 763 } 764 req->r_dentry = dget(dentry); 765 req->r_num_caps = 2; 766 req->r_old_dentry = dget(old_dentry); /* or inode? hrm. */ 767 req->r_locked_dir = dir; 768 req->r_dentry_drop = CEPH_CAP_FILE_SHARED; 769 req->r_dentry_unless = CEPH_CAP_FILE_EXCL; 770 err = ceph_mdsc_do_request(mdsc, dir, req); 771 if (err) 772 d_drop(dentry); 773 else if (!req->r_reply_info.head->is_dentry) 774 d_instantiate(dentry, igrab(old_dentry->d_inode)); 775 ceph_mdsc_put_request(req); 776 return err; 777 } 778 779 /* 780 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it 781 * looks like the link count will hit 0, drop any other caps (other 782 * than PIN) we don't specifically want (due to the file still being 783 * open). 784 */ 785 static int drop_caps_for_unlink(struct inode *inode) 786 { 787 struct ceph_inode_info *ci = ceph_inode(inode); 788 int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL; 789 790 spin_lock(&inode->i_lock); 791 if (inode->i_nlink == 1) { 792 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN); 793 ci->i_ceph_flags |= CEPH_I_NODELAY; 794 } 795 spin_unlock(&inode->i_lock); 796 return drop; 797 } 798 799 /* 800 * rmdir and unlink are differ only by the metadata op code 801 */ 802 static int ceph_unlink(struct inode *dir, struct dentry *dentry) 803 { 804 struct ceph_client *client = ceph_sb_to_client(dir->i_sb); 805 struct ceph_mds_client *mdsc = &client->mdsc; 806 struct inode *inode = dentry->d_inode; 807 struct ceph_mds_request *req; 808 int err = -EROFS; 809 int op; 810 811 if (ceph_snap(dir) == CEPH_SNAPDIR) { 812 /* rmdir .snap/foo is RMSNAP */ 813 dout("rmsnap dir %p '%.*s' dn %p\n", dir, dentry->d_name.len, 814 dentry->d_name.name, dentry); 815 op = CEPH_MDS_OP_RMSNAP; 816 } else if (ceph_snap(dir) == CEPH_NOSNAP) { 817 dout("unlink/rmdir dir %p dn %p inode %p\n", 818 dir, dentry, inode); 819 op = ((dentry->d_inode->i_mode & S_IFMT) == S_IFDIR) ? 820 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK; 821 } else 822 goto out; 823 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS); 824 if (IS_ERR(req)) { 825 err = PTR_ERR(req); 826 goto out; 827 } 828 req->r_dentry = dget(dentry); 829 req->r_num_caps = 2; 830 req->r_locked_dir = dir; 831 req->r_dentry_drop = CEPH_CAP_FILE_SHARED; 832 req->r_dentry_unless = CEPH_CAP_FILE_EXCL; 833 req->r_inode_drop = drop_caps_for_unlink(inode); 834 err = ceph_mdsc_do_request(mdsc, dir, req); 835 if (!err && !req->r_reply_info.head->is_dentry) 836 d_delete(dentry); 837 ceph_mdsc_put_request(req); 838 out: 839 return err; 840 } 841 842 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry, 843 struct inode *new_dir, struct dentry *new_dentry) 844 { 845 struct ceph_client *client = ceph_sb_to_client(old_dir->i_sb); 846 struct ceph_mds_client *mdsc = &client->mdsc; 847 struct ceph_mds_request *req; 848 int err; 849 850 if (ceph_snap(old_dir) != ceph_snap(new_dir)) 851 return -EXDEV; 852 if (ceph_snap(old_dir) != CEPH_NOSNAP || 853 ceph_snap(new_dir) != CEPH_NOSNAP) 854 return -EROFS; 855 dout("rename dir %p dentry %p to dir %p dentry %p\n", 856 old_dir, old_dentry, new_dir, new_dentry); 857 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RENAME, USE_AUTH_MDS); 858 if (IS_ERR(req)) 859 return PTR_ERR(req); 860 req->r_dentry = dget(new_dentry); 861 req->r_num_caps = 2; 862 req->r_old_dentry = dget(old_dentry); 863 req->r_locked_dir = new_dir; 864 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED; 865 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL; 866 req->r_dentry_drop = CEPH_CAP_FILE_SHARED; 867 req->r_dentry_unless = CEPH_CAP_FILE_EXCL; 868 /* release LINK_RDCACHE on source inode (mds will lock it) */ 869 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED; 870 if (new_dentry->d_inode) 871 req->r_inode_drop = drop_caps_for_unlink(new_dentry->d_inode); 872 err = ceph_mdsc_do_request(mdsc, old_dir, req); 873 if (!err && !req->r_reply_info.head->is_dentry) { 874 /* 875 * Normally d_move() is done by fill_trace (called by 876 * do_request, above). If there is no trace, we need 877 * to do it here. 878 */ 879 d_move(old_dentry, new_dentry); 880 } 881 ceph_mdsc_put_request(req); 882 return err; 883 } 884 885 886 /* 887 * Check if dentry lease is valid. If not, delete the lease. Try to 888 * renew if the least is more than half up. 889 */ 890 static int dentry_lease_is_valid(struct dentry *dentry) 891 { 892 struct ceph_dentry_info *di; 893 struct ceph_mds_session *s; 894 int valid = 0; 895 u32 gen; 896 unsigned long ttl; 897 struct ceph_mds_session *session = NULL; 898 struct inode *dir = NULL; 899 u32 seq = 0; 900 901 spin_lock(&dentry->d_lock); 902 di = ceph_dentry(dentry); 903 if (di && di->lease_session) { 904 s = di->lease_session; 905 spin_lock(&s->s_cap_lock); 906 gen = s->s_cap_gen; 907 ttl = s->s_cap_ttl; 908 spin_unlock(&s->s_cap_lock); 909 910 if (di->lease_gen == gen && 911 time_before(jiffies, dentry->d_time) && 912 time_before(jiffies, ttl)) { 913 valid = 1; 914 if (di->lease_renew_after && 915 time_after(jiffies, di->lease_renew_after)) { 916 /* we should renew */ 917 dir = dentry->d_parent->d_inode; 918 session = ceph_get_mds_session(s); 919 seq = di->lease_seq; 920 di->lease_renew_after = 0; 921 di->lease_renew_from = jiffies; 922 } 923 } 924 } 925 spin_unlock(&dentry->d_lock); 926 927 if (session) { 928 ceph_mdsc_lease_send_msg(session, dir, dentry, 929 CEPH_MDS_LEASE_RENEW, seq); 930 ceph_put_mds_session(session); 931 } 932 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid); 933 return valid; 934 } 935 936 /* 937 * Check if directory-wide content lease/cap is valid. 938 */ 939 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry) 940 { 941 struct ceph_inode_info *ci = ceph_inode(dir); 942 struct ceph_dentry_info *di = ceph_dentry(dentry); 943 int valid = 0; 944 945 spin_lock(&dir->i_lock); 946 if (ci->i_shared_gen == di->lease_shared_gen) 947 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1); 948 spin_unlock(&dir->i_lock); 949 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n", 950 dir, (unsigned)ci->i_shared_gen, dentry, 951 (unsigned)di->lease_shared_gen, valid); 952 return valid; 953 } 954 955 /* 956 * Check if cached dentry can be trusted. 957 */ 958 static int ceph_d_revalidate(struct dentry *dentry, struct nameidata *nd) 959 { 960 struct inode *dir = dentry->d_parent->d_inode; 961 962 dout("d_revalidate %p '%.*s' inode %p\n", dentry, 963 dentry->d_name.len, dentry->d_name.name, dentry->d_inode); 964 965 /* always trust cached snapped dentries, snapdir dentry */ 966 if (ceph_snap(dir) != CEPH_NOSNAP) { 967 dout("d_revalidate %p '%.*s' inode %p is SNAPPED\n", dentry, 968 dentry->d_name.len, dentry->d_name.name, dentry->d_inode); 969 goto out_touch; 970 } 971 if (dentry->d_inode && ceph_snap(dentry->d_inode) == CEPH_SNAPDIR) 972 goto out_touch; 973 974 if (dentry_lease_is_valid(dentry) || 975 dir_lease_is_valid(dir, dentry)) 976 goto out_touch; 977 978 dout("d_revalidate %p invalid\n", dentry); 979 d_drop(dentry); 980 return 0; 981 out_touch: 982 ceph_dentry_lru_touch(dentry); 983 return 1; 984 } 985 986 /* 987 * When a dentry is released, clear the dir I_COMPLETE if it was part 988 * of the current dir gen. 989 */ 990 static void ceph_dentry_release(struct dentry *dentry) 991 { 992 struct ceph_dentry_info *di = ceph_dentry(dentry); 993 struct inode *parent_inode = dentry->d_parent->d_inode; 994 995 if (parent_inode) { 996 struct ceph_inode_info *ci = ceph_inode(parent_inode); 997 998 spin_lock(&parent_inode->i_lock); 999 if (ci->i_shared_gen == di->lease_shared_gen) { 1000 dout(" clearing %p complete (d_release)\n", 1001 parent_inode); 1002 ci->i_ceph_flags &= ~CEPH_I_COMPLETE; 1003 ci->i_release_count++; 1004 } 1005 spin_unlock(&parent_inode->i_lock); 1006 } 1007 if (di) { 1008 ceph_dentry_lru_del(dentry); 1009 if (di->lease_session) 1010 ceph_put_mds_session(di->lease_session); 1011 kmem_cache_free(ceph_dentry_cachep, di); 1012 dentry->d_fsdata = NULL; 1013 } 1014 } 1015 1016 static int ceph_snapdir_d_revalidate(struct dentry *dentry, 1017 struct nameidata *nd) 1018 { 1019 /* 1020 * Eventually, we'll want to revalidate snapped metadata 1021 * too... probably... 1022 */ 1023 return 1; 1024 } 1025 1026 1027 1028 /* 1029 * read() on a dir. This weird interface hack only works if mounted 1030 * with '-o dirstat'. 1031 */ 1032 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size, 1033 loff_t *ppos) 1034 { 1035 struct ceph_file_info *cf = file->private_data; 1036 struct inode *inode = file->f_dentry->d_inode; 1037 struct ceph_inode_info *ci = ceph_inode(inode); 1038 int left; 1039 1040 if (!ceph_test_opt(ceph_client(inode->i_sb), DIRSTAT)) 1041 return -EISDIR; 1042 1043 if (!cf->dir_info) { 1044 cf->dir_info = kmalloc(1024, GFP_NOFS); 1045 if (!cf->dir_info) 1046 return -ENOMEM; 1047 cf->dir_info_len = 1048 sprintf(cf->dir_info, 1049 "entries: %20lld\n" 1050 " files: %20lld\n" 1051 " subdirs: %20lld\n" 1052 "rentries: %20lld\n" 1053 " rfiles: %20lld\n" 1054 " rsubdirs: %20lld\n" 1055 "rbytes: %20lld\n" 1056 "rctime: %10ld.%09ld\n", 1057 ci->i_files + ci->i_subdirs, 1058 ci->i_files, 1059 ci->i_subdirs, 1060 ci->i_rfiles + ci->i_rsubdirs, 1061 ci->i_rfiles, 1062 ci->i_rsubdirs, 1063 ci->i_rbytes, 1064 (long)ci->i_rctime.tv_sec, 1065 (long)ci->i_rctime.tv_nsec); 1066 } 1067 1068 if (*ppos >= cf->dir_info_len) 1069 return 0; 1070 size = min_t(unsigned, size, cf->dir_info_len-*ppos); 1071 left = copy_to_user(buf, cf->dir_info + *ppos, size); 1072 if (left == size) 1073 return -EFAULT; 1074 *ppos += (size - left); 1075 return size - left; 1076 } 1077 1078 /* 1079 * an fsync() on a dir will wait for any uncommitted directory 1080 * operations to commit. 1081 */ 1082 static int ceph_dir_fsync(struct file *file, struct dentry *dentry, 1083 int datasync) 1084 { 1085 struct inode *inode = dentry->d_inode; 1086 struct ceph_inode_info *ci = ceph_inode(inode); 1087 struct list_head *head = &ci->i_unsafe_dirops; 1088 struct ceph_mds_request *req; 1089 u64 last_tid; 1090 int ret = 0; 1091 1092 dout("dir_fsync %p\n", inode); 1093 spin_lock(&ci->i_unsafe_lock); 1094 if (list_empty(head)) 1095 goto out; 1096 1097 req = list_entry(head->prev, 1098 struct ceph_mds_request, r_unsafe_dir_item); 1099 last_tid = req->r_tid; 1100 1101 do { 1102 ceph_mdsc_get_request(req); 1103 spin_unlock(&ci->i_unsafe_lock); 1104 dout("dir_fsync %p wait on tid %llu (until %llu)\n", 1105 inode, req->r_tid, last_tid); 1106 if (req->r_timeout) { 1107 ret = wait_for_completion_timeout( 1108 &req->r_safe_completion, req->r_timeout); 1109 if (ret > 0) 1110 ret = 0; 1111 else if (ret == 0) 1112 ret = -EIO; /* timed out */ 1113 } else { 1114 wait_for_completion(&req->r_safe_completion); 1115 } 1116 spin_lock(&ci->i_unsafe_lock); 1117 ceph_mdsc_put_request(req); 1118 1119 if (ret || list_empty(head)) 1120 break; 1121 req = list_entry(head->next, 1122 struct ceph_mds_request, r_unsafe_dir_item); 1123 } while (req->r_tid < last_tid); 1124 out: 1125 spin_unlock(&ci->i_unsafe_lock); 1126 return ret; 1127 } 1128 1129 /* 1130 * We maintain a private dentry LRU. 1131 * 1132 * FIXME: this needs to be changed to a per-mds lru to be useful. 1133 */ 1134 void ceph_dentry_lru_add(struct dentry *dn) 1135 { 1136 struct ceph_dentry_info *di = ceph_dentry(dn); 1137 struct ceph_mds_client *mdsc; 1138 1139 dout("dentry_lru_add %p %p '%.*s'\n", di, dn, 1140 dn->d_name.len, dn->d_name.name); 1141 if (di) { 1142 mdsc = &ceph_client(dn->d_sb)->mdsc; 1143 spin_lock(&mdsc->dentry_lru_lock); 1144 list_add_tail(&di->lru, &mdsc->dentry_lru); 1145 mdsc->num_dentry++; 1146 spin_unlock(&mdsc->dentry_lru_lock); 1147 } 1148 } 1149 1150 void ceph_dentry_lru_touch(struct dentry *dn) 1151 { 1152 struct ceph_dentry_info *di = ceph_dentry(dn); 1153 struct ceph_mds_client *mdsc; 1154 1155 dout("dentry_lru_touch %p %p '%.*s'\n", di, dn, 1156 dn->d_name.len, dn->d_name.name); 1157 if (di) { 1158 mdsc = &ceph_client(dn->d_sb)->mdsc; 1159 spin_lock(&mdsc->dentry_lru_lock); 1160 list_move_tail(&di->lru, &mdsc->dentry_lru); 1161 spin_unlock(&mdsc->dentry_lru_lock); 1162 } 1163 } 1164 1165 void ceph_dentry_lru_del(struct dentry *dn) 1166 { 1167 struct ceph_dentry_info *di = ceph_dentry(dn); 1168 struct ceph_mds_client *mdsc; 1169 1170 dout("dentry_lru_del %p %p '%.*s'\n", di, dn, 1171 dn->d_name.len, dn->d_name.name); 1172 if (di) { 1173 mdsc = &ceph_client(dn->d_sb)->mdsc; 1174 spin_lock(&mdsc->dentry_lru_lock); 1175 list_del_init(&di->lru); 1176 mdsc->num_dentry--; 1177 spin_unlock(&mdsc->dentry_lru_lock); 1178 } 1179 } 1180 1181 const struct file_operations ceph_dir_fops = { 1182 .read = ceph_read_dir, 1183 .readdir = ceph_readdir, 1184 .llseek = ceph_dir_llseek, 1185 .open = ceph_open, 1186 .release = ceph_release, 1187 .unlocked_ioctl = ceph_ioctl, 1188 .fsync = ceph_dir_fsync, 1189 }; 1190 1191 const struct inode_operations ceph_dir_iops = { 1192 .lookup = ceph_lookup, 1193 .permission = ceph_permission, 1194 .getattr = ceph_getattr, 1195 .setattr = ceph_setattr, 1196 .setxattr = ceph_setxattr, 1197 .getxattr = ceph_getxattr, 1198 .listxattr = ceph_listxattr, 1199 .removexattr = ceph_removexattr, 1200 .mknod = ceph_mknod, 1201 .symlink = ceph_symlink, 1202 .mkdir = ceph_mkdir, 1203 .link = ceph_link, 1204 .unlink = ceph_unlink, 1205 .rmdir = ceph_unlink, 1206 .rename = ceph_rename, 1207 .create = ceph_create, 1208 }; 1209 1210 struct dentry_operations ceph_dentry_ops = { 1211 .d_revalidate = ceph_d_revalidate, 1212 .d_release = ceph_dentry_release, 1213 }; 1214 1215 struct dentry_operations ceph_snapdir_dentry_ops = { 1216 .d_revalidate = ceph_snapdir_d_revalidate, 1217 }; 1218 1219 struct dentry_operations ceph_snap_dentry_ops = { 1220 }; 1221