1 #include <linux/ceph/ceph_debug.h> 2 3 #include <linux/module.h> 4 #include <linux/fs.h> 5 #include <linux/smp_lock.h> 6 #include <linux/slab.h> 7 #include <linux/string.h> 8 #include <linux/uaccess.h> 9 #include <linux/kernel.h> 10 #include <linux/namei.h> 11 #include <linux/writeback.h> 12 #include <linux/vmalloc.h> 13 #include <linux/pagevec.h> 14 15 #include "super.h" 16 #include "mds_client.h" 17 #include <linux/ceph/decode.h> 18 19 /* 20 * Ceph inode operations 21 * 22 * Implement basic inode helpers (get, alloc) and inode ops (getattr, 23 * setattr, etc.), xattr helpers, and helpers for assimilating 24 * metadata returned by the MDS into our cache. 25 * 26 * Also define helpers for doing asynchronous writeback, invalidation, 27 * and truncation for the benefit of those who can't afford to block 28 * (typically because they are in the message handler path). 29 */ 30 31 static const struct inode_operations ceph_symlink_iops; 32 33 static void ceph_invalidate_work(struct work_struct *work); 34 static void ceph_writeback_work(struct work_struct *work); 35 static void ceph_vmtruncate_work(struct work_struct *work); 36 37 /* 38 * find or create an inode, given the ceph ino number 39 */ 40 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino) 41 { 42 struct inode *inode; 43 ino_t t = ceph_vino_to_ino(vino); 44 45 inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino); 46 if (inode == NULL) 47 return ERR_PTR(-ENOMEM); 48 if (inode->i_state & I_NEW) { 49 dout("get_inode created new inode %p %llx.%llx ino %llx\n", 50 inode, ceph_vinop(inode), (u64)inode->i_ino); 51 unlock_new_inode(inode); 52 } 53 54 dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino, 55 vino.snap, inode); 56 return inode; 57 } 58 59 /* 60 * get/constuct snapdir inode for a given directory 61 */ 62 struct inode *ceph_get_snapdir(struct inode *parent) 63 { 64 struct ceph_vino vino = { 65 .ino = ceph_ino(parent), 66 .snap = CEPH_SNAPDIR, 67 }; 68 struct inode *inode = ceph_get_inode(parent->i_sb, vino); 69 struct ceph_inode_info *ci = ceph_inode(inode); 70 71 BUG_ON(!S_ISDIR(parent->i_mode)); 72 if (IS_ERR(inode)) 73 return inode; 74 inode->i_mode = parent->i_mode; 75 inode->i_uid = parent->i_uid; 76 inode->i_gid = parent->i_gid; 77 inode->i_op = &ceph_dir_iops; 78 inode->i_fop = &ceph_dir_fops; 79 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */ 80 ci->i_rbytes = 0; 81 return inode; 82 } 83 84 const struct inode_operations ceph_file_iops = { 85 .permission = ceph_permission, 86 .setattr = ceph_setattr, 87 .getattr = ceph_getattr, 88 .setxattr = ceph_setxattr, 89 .getxattr = ceph_getxattr, 90 .listxattr = ceph_listxattr, 91 .removexattr = ceph_removexattr, 92 }; 93 94 95 /* 96 * We use a 'frag tree' to keep track of the MDS's directory fragments 97 * for a given inode (usually there is just a single fragment). We 98 * need to know when a child frag is delegated to a new MDS, or when 99 * it is flagged as replicated, so we can direct our requests 100 * accordingly. 101 */ 102 103 /* 104 * find/create a frag in the tree 105 */ 106 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci, 107 u32 f) 108 { 109 struct rb_node **p; 110 struct rb_node *parent = NULL; 111 struct ceph_inode_frag *frag; 112 int c; 113 114 p = &ci->i_fragtree.rb_node; 115 while (*p) { 116 parent = *p; 117 frag = rb_entry(parent, struct ceph_inode_frag, node); 118 c = ceph_frag_compare(f, frag->frag); 119 if (c < 0) 120 p = &(*p)->rb_left; 121 else if (c > 0) 122 p = &(*p)->rb_right; 123 else 124 return frag; 125 } 126 127 frag = kmalloc(sizeof(*frag), GFP_NOFS); 128 if (!frag) { 129 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx " 130 "frag %x\n", &ci->vfs_inode, 131 ceph_vinop(&ci->vfs_inode), f); 132 return ERR_PTR(-ENOMEM); 133 } 134 frag->frag = f; 135 frag->split_by = 0; 136 frag->mds = -1; 137 frag->ndist = 0; 138 139 rb_link_node(&frag->node, parent, p); 140 rb_insert_color(&frag->node, &ci->i_fragtree); 141 142 dout("get_or_create_frag added %llx.%llx frag %x\n", 143 ceph_vinop(&ci->vfs_inode), f); 144 return frag; 145 } 146 147 /* 148 * find a specific frag @f 149 */ 150 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f) 151 { 152 struct rb_node *n = ci->i_fragtree.rb_node; 153 154 while (n) { 155 struct ceph_inode_frag *frag = 156 rb_entry(n, struct ceph_inode_frag, node); 157 int c = ceph_frag_compare(f, frag->frag); 158 if (c < 0) 159 n = n->rb_left; 160 else if (c > 0) 161 n = n->rb_right; 162 else 163 return frag; 164 } 165 return NULL; 166 } 167 168 /* 169 * Choose frag containing the given value @v. If @pfrag is 170 * specified, copy the frag delegation info to the caller if 171 * it is present. 172 */ 173 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v, 174 struct ceph_inode_frag *pfrag, 175 int *found) 176 { 177 u32 t = ceph_frag_make(0, 0); 178 struct ceph_inode_frag *frag; 179 unsigned nway, i; 180 u32 n; 181 182 if (found) 183 *found = 0; 184 185 mutex_lock(&ci->i_fragtree_mutex); 186 while (1) { 187 WARN_ON(!ceph_frag_contains_value(t, v)); 188 frag = __ceph_find_frag(ci, t); 189 if (!frag) 190 break; /* t is a leaf */ 191 if (frag->split_by == 0) { 192 if (pfrag) 193 memcpy(pfrag, frag, sizeof(*pfrag)); 194 if (found) 195 *found = 1; 196 break; 197 } 198 199 /* choose child */ 200 nway = 1 << frag->split_by; 201 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t, 202 frag->split_by, nway); 203 for (i = 0; i < nway; i++) { 204 n = ceph_frag_make_child(t, frag->split_by, i); 205 if (ceph_frag_contains_value(n, v)) { 206 t = n; 207 break; 208 } 209 } 210 BUG_ON(i == nway); 211 } 212 dout("choose_frag(%x) = %x\n", v, t); 213 214 mutex_unlock(&ci->i_fragtree_mutex); 215 return t; 216 } 217 218 /* 219 * Process dirfrag (delegation) info from the mds. Include leaf 220 * fragment in tree ONLY if ndist > 0. Otherwise, only 221 * branches/splits are included in i_fragtree) 222 */ 223 static int ceph_fill_dirfrag(struct inode *inode, 224 struct ceph_mds_reply_dirfrag *dirinfo) 225 { 226 struct ceph_inode_info *ci = ceph_inode(inode); 227 struct ceph_inode_frag *frag; 228 u32 id = le32_to_cpu(dirinfo->frag); 229 int mds = le32_to_cpu(dirinfo->auth); 230 int ndist = le32_to_cpu(dirinfo->ndist); 231 int i; 232 int err = 0; 233 234 mutex_lock(&ci->i_fragtree_mutex); 235 if (ndist == 0) { 236 /* no delegation info needed. */ 237 frag = __ceph_find_frag(ci, id); 238 if (!frag) 239 goto out; 240 if (frag->split_by == 0) { 241 /* tree leaf, remove */ 242 dout("fill_dirfrag removed %llx.%llx frag %x" 243 " (no ref)\n", ceph_vinop(inode), id); 244 rb_erase(&frag->node, &ci->i_fragtree); 245 kfree(frag); 246 } else { 247 /* tree branch, keep and clear */ 248 dout("fill_dirfrag cleared %llx.%llx frag %x" 249 " referral\n", ceph_vinop(inode), id); 250 frag->mds = -1; 251 frag->ndist = 0; 252 } 253 goto out; 254 } 255 256 257 /* find/add this frag to store mds delegation info */ 258 frag = __get_or_create_frag(ci, id); 259 if (IS_ERR(frag)) { 260 /* this is not the end of the world; we can continue 261 with bad/inaccurate delegation info */ 262 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n", 263 ceph_vinop(inode), le32_to_cpu(dirinfo->frag)); 264 err = -ENOMEM; 265 goto out; 266 } 267 268 frag->mds = mds; 269 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP); 270 for (i = 0; i < frag->ndist; i++) 271 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]); 272 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n", 273 ceph_vinop(inode), frag->frag, frag->ndist); 274 275 out: 276 mutex_unlock(&ci->i_fragtree_mutex); 277 return err; 278 } 279 280 281 /* 282 * initialize a newly allocated inode. 283 */ 284 struct inode *ceph_alloc_inode(struct super_block *sb) 285 { 286 struct ceph_inode_info *ci; 287 int i; 288 289 ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS); 290 if (!ci) 291 return NULL; 292 293 dout("alloc_inode %p\n", &ci->vfs_inode); 294 295 ci->i_version = 0; 296 ci->i_time_warp_seq = 0; 297 ci->i_ceph_flags = 0; 298 ci->i_release_count = 0; 299 ci->i_symlink = NULL; 300 301 ci->i_fragtree = RB_ROOT; 302 mutex_init(&ci->i_fragtree_mutex); 303 304 ci->i_xattrs.blob = NULL; 305 ci->i_xattrs.prealloc_blob = NULL; 306 ci->i_xattrs.dirty = false; 307 ci->i_xattrs.index = RB_ROOT; 308 ci->i_xattrs.count = 0; 309 ci->i_xattrs.names_size = 0; 310 ci->i_xattrs.vals_size = 0; 311 ci->i_xattrs.version = 0; 312 ci->i_xattrs.index_version = 0; 313 314 ci->i_caps = RB_ROOT; 315 ci->i_auth_cap = NULL; 316 ci->i_dirty_caps = 0; 317 ci->i_flushing_caps = 0; 318 INIT_LIST_HEAD(&ci->i_dirty_item); 319 INIT_LIST_HEAD(&ci->i_flushing_item); 320 ci->i_cap_flush_seq = 0; 321 ci->i_cap_flush_last_tid = 0; 322 memset(&ci->i_cap_flush_tid, 0, sizeof(ci->i_cap_flush_tid)); 323 init_waitqueue_head(&ci->i_cap_wq); 324 ci->i_hold_caps_min = 0; 325 ci->i_hold_caps_max = 0; 326 INIT_LIST_HEAD(&ci->i_cap_delay_list); 327 ci->i_cap_exporting_mds = 0; 328 ci->i_cap_exporting_mseq = 0; 329 ci->i_cap_exporting_issued = 0; 330 INIT_LIST_HEAD(&ci->i_cap_snaps); 331 ci->i_head_snapc = NULL; 332 ci->i_snap_caps = 0; 333 334 for (i = 0; i < CEPH_FILE_MODE_NUM; i++) 335 ci->i_nr_by_mode[i] = 0; 336 337 ci->i_truncate_seq = 0; 338 ci->i_truncate_size = 0; 339 ci->i_truncate_pending = 0; 340 341 ci->i_max_size = 0; 342 ci->i_reported_size = 0; 343 ci->i_wanted_max_size = 0; 344 ci->i_requested_max_size = 0; 345 346 ci->i_pin_ref = 0; 347 ci->i_rd_ref = 0; 348 ci->i_rdcache_ref = 0; 349 ci->i_wr_ref = 0; 350 ci->i_wrbuffer_ref = 0; 351 ci->i_wrbuffer_ref_head = 0; 352 ci->i_shared_gen = 0; 353 ci->i_rdcache_gen = 0; 354 ci->i_rdcache_revoking = 0; 355 356 INIT_LIST_HEAD(&ci->i_unsafe_writes); 357 INIT_LIST_HEAD(&ci->i_unsafe_dirops); 358 spin_lock_init(&ci->i_unsafe_lock); 359 360 ci->i_snap_realm = NULL; 361 INIT_LIST_HEAD(&ci->i_snap_realm_item); 362 INIT_LIST_HEAD(&ci->i_snap_flush_item); 363 364 INIT_WORK(&ci->i_wb_work, ceph_writeback_work); 365 INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work); 366 367 INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work); 368 369 return &ci->vfs_inode; 370 } 371 372 void ceph_destroy_inode(struct inode *inode) 373 { 374 struct ceph_inode_info *ci = ceph_inode(inode); 375 struct ceph_inode_frag *frag; 376 struct rb_node *n; 377 378 dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode)); 379 380 ceph_queue_caps_release(inode); 381 382 /* 383 * we may still have a snap_realm reference if there are stray 384 * caps in i_cap_exporting_issued or i_snap_caps. 385 */ 386 if (ci->i_snap_realm) { 387 struct ceph_mds_client *mdsc = 388 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc; 389 struct ceph_snap_realm *realm = ci->i_snap_realm; 390 391 dout(" dropping residual ref to snap realm %p\n", realm); 392 spin_lock(&realm->inodes_with_caps_lock); 393 list_del_init(&ci->i_snap_realm_item); 394 spin_unlock(&realm->inodes_with_caps_lock); 395 ceph_put_snap_realm(mdsc, realm); 396 } 397 398 kfree(ci->i_symlink); 399 while ((n = rb_first(&ci->i_fragtree)) != NULL) { 400 frag = rb_entry(n, struct ceph_inode_frag, node); 401 rb_erase(n, &ci->i_fragtree); 402 kfree(frag); 403 } 404 405 __ceph_destroy_xattrs(ci); 406 if (ci->i_xattrs.blob) 407 ceph_buffer_put(ci->i_xattrs.blob); 408 if (ci->i_xattrs.prealloc_blob) 409 ceph_buffer_put(ci->i_xattrs.prealloc_blob); 410 411 kmem_cache_free(ceph_inode_cachep, ci); 412 } 413 414 415 /* 416 * Helpers to fill in size, ctime, mtime, and atime. We have to be 417 * careful because either the client or MDS may have more up to date 418 * info, depending on which capabilities are held, and whether 419 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime 420 * and size are monotonically increasing, except when utimes() or 421 * truncate() increments the corresponding _seq values.) 422 */ 423 int ceph_fill_file_size(struct inode *inode, int issued, 424 u32 truncate_seq, u64 truncate_size, u64 size) 425 { 426 struct ceph_inode_info *ci = ceph_inode(inode); 427 int queue_trunc = 0; 428 429 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 || 430 (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) { 431 dout("size %lld -> %llu\n", inode->i_size, size); 432 inode->i_size = size; 433 inode->i_blocks = (size + (1<<9) - 1) >> 9; 434 ci->i_reported_size = size; 435 if (truncate_seq != ci->i_truncate_seq) { 436 dout("truncate_seq %u -> %u\n", 437 ci->i_truncate_seq, truncate_seq); 438 ci->i_truncate_seq = truncate_seq; 439 /* 440 * If we hold relevant caps, or in the case where we're 441 * not the only client referencing this file and we 442 * don't hold those caps, then we need to check whether 443 * the file is either opened or mmaped 444 */ 445 if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_RD| 446 CEPH_CAP_FILE_WR|CEPH_CAP_FILE_BUFFER| 447 CEPH_CAP_FILE_EXCL| 448 CEPH_CAP_FILE_LAZYIO)) || 449 mapping_mapped(inode->i_mapping) || 450 __ceph_caps_file_wanted(ci)) { 451 ci->i_truncate_pending++; 452 queue_trunc = 1; 453 } 454 } 455 } 456 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 && 457 ci->i_truncate_size != truncate_size) { 458 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size, 459 truncate_size); 460 ci->i_truncate_size = truncate_size; 461 } 462 return queue_trunc; 463 } 464 465 void ceph_fill_file_time(struct inode *inode, int issued, 466 u64 time_warp_seq, struct timespec *ctime, 467 struct timespec *mtime, struct timespec *atime) 468 { 469 struct ceph_inode_info *ci = ceph_inode(inode); 470 int warn = 0; 471 472 if (issued & (CEPH_CAP_FILE_EXCL| 473 CEPH_CAP_FILE_WR| 474 CEPH_CAP_FILE_BUFFER| 475 CEPH_CAP_AUTH_EXCL| 476 CEPH_CAP_XATTR_EXCL)) { 477 if (timespec_compare(ctime, &inode->i_ctime) > 0) { 478 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n", 479 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 480 ctime->tv_sec, ctime->tv_nsec); 481 inode->i_ctime = *ctime; 482 } 483 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) { 484 /* the MDS did a utimes() */ 485 dout("mtime %ld.%09ld -> %ld.%09ld " 486 "tw %d -> %d\n", 487 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 488 mtime->tv_sec, mtime->tv_nsec, 489 ci->i_time_warp_seq, (int)time_warp_seq); 490 491 inode->i_mtime = *mtime; 492 inode->i_atime = *atime; 493 ci->i_time_warp_seq = time_warp_seq; 494 } else if (time_warp_seq == ci->i_time_warp_seq) { 495 /* nobody did utimes(); take the max */ 496 if (timespec_compare(mtime, &inode->i_mtime) > 0) { 497 dout("mtime %ld.%09ld -> %ld.%09ld inc\n", 498 inode->i_mtime.tv_sec, 499 inode->i_mtime.tv_nsec, 500 mtime->tv_sec, mtime->tv_nsec); 501 inode->i_mtime = *mtime; 502 } 503 if (timespec_compare(atime, &inode->i_atime) > 0) { 504 dout("atime %ld.%09ld -> %ld.%09ld inc\n", 505 inode->i_atime.tv_sec, 506 inode->i_atime.tv_nsec, 507 atime->tv_sec, atime->tv_nsec); 508 inode->i_atime = *atime; 509 } 510 } else if (issued & CEPH_CAP_FILE_EXCL) { 511 /* we did a utimes(); ignore mds values */ 512 } else { 513 warn = 1; 514 } 515 } else { 516 /* we have no write|excl caps; whatever the MDS says is true */ 517 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) { 518 inode->i_ctime = *ctime; 519 inode->i_mtime = *mtime; 520 inode->i_atime = *atime; 521 ci->i_time_warp_seq = time_warp_seq; 522 } else { 523 warn = 1; 524 } 525 } 526 if (warn) /* time_warp_seq shouldn't go backwards */ 527 dout("%p mds time_warp_seq %llu < %u\n", 528 inode, time_warp_seq, ci->i_time_warp_seq); 529 } 530 531 /* 532 * Populate an inode based on info from mds. May be called on new or 533 * existing inodes. 534 */ 535 static int fill_inode(struct inode *inode, 536 struct ceph_mds_reply_info_in *iinfo, 537 struct ceph_mds_reply_dirfrag *dirinfo, 538 struct ceph_mds_session *session, 539 unsigned long ttl_from, int cap_fmode, 540 struct ceph_cap_reservation *caps_reservation) 541 { 542 struct ceph_mds_reply_inode *info = iinfo->in; 543 struct ceph_inode_info *ci = ceph_inode(inode); 544 int i; 545 int issued, implemented; 546 struct timespec mtime, atime, ctime; 547 u32 nsplits; 548 struct ceph_buffer *xattr_blob = NULL; 549 int err = 0; 550 int queue_trunc = 0; 551 552 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n", 553 inode, ceph_vinop(inode), le64_to_cpu(info->version), 554 ci->i_version); 555 556 /* 557 * prealloc xattr data, if it looks like we'll need it. only 558 * if len > 4 (meaning there are actually xattrs; the first 4 559 * bytes are the xattr count). 560 */ 561 if (iinfo->xattr_len > 4) { 562 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS); 563 if (!xattr_blob) 564 pr_err("fill_inode ENOMEM xattr blob %d bytes\n", 565 iinfo->xattr_len); 566 } 567 568 spin_lock(&inode->i_lock); 569 570 /* 571 * provided version will be odd if inode value is projected, 572 * even if stable. skip the update if we have newer stable 573 * info (ours>=theirs, e.g. due to racing mds replies), unless 574 * we are getting projected (unstable) info (in which case the 575 * version is odd, and we want ours>theirs). 576 * us them 577 * 2 2 skip 578 * 3 2 skip 579 * 3 3 update 580 */ 581 if (le64_to_cpu(info->version) > 0 && 582 (ci->i_version & ~1) >= le64_to_cpu(info->version)) 583 goto no_change; 584 585 issued = __ceph_caps_issued(ci, &implemented); 586 issued |= implemented | __ceph_caps_dirty(ci); 587 588 /* update inode */ 589 ci->i_version = le64_to_cpu(info->version); 590 inode->i_version++; 591 inode->i_rdev = le32_to_cpu(info->rdev); 592 593 if ((issued & CEPH_CAP_AUTH_EXCL) == 0) { 594 inode->i_mode = le32_to_cpu(info->mode); 595 inode->i_uid = le32_to_cpu(info->uid); 596 inode->i_gid = le32_to_cpu(info->gid); 597 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode, 598 inode->i_uid, inode->i_gid); 599 } 600 601 if ((issued & CEPH_CAP_LINK_EXCL) == 0) 602 inode->i_nlink = le32_to_cpu(info->nlink); 603 604 /* be careful with mtime, atime, size */ 605 ceph_decode_timespec(&atime, &info->atime); 606 ceph_decode_timespec(&mtime, &info->mtime); 607 ceph_decode_timespec(&ctime, &info->ctime); 608 queue_trunc = ceph_fill_file_size(inode, issued, 609 le32_to_cpu(info->truncate_seq), 610 le64_to_cpu(info->truncate_size), 611 le64_to_cpu(info->size)); 612 ceph_fill_file_time(inode, issued, 613 le32_to_cpu(info->time_warp_seq), 614 &ctime, &mtime, &atime); 615 616 /* only update max_size on auth cap */ 617 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) && 618 ci->i_max_size != le64_to_cpu(info->max_size)) { 619 dout("max_size %lld -> %llu\n", ci->i_max_size, 620 le64_to_cpu(info->max_size)); 621 ci->i_max_size = le64_to_cpu(info->max_size); 622 } 623 624 ci->i_layout = info->layout; 625 inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1; 626 627 /* xattrs */ 628 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */ 629 if ((issued & CEPH_CAP_XATTR_EXCL) == 0 && 630 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) { 631 if (ci->i_xattrs.blob) 632 ceph_buffer_put(ci->i_xattrs.blob); 633 ci->i_xattrs.blob = xattr_blob; 634 if (xattr_blob) 635 memcpy(ci->i_xattrs.blob->vec.iov_base, 636 iinfo->xattr_data, iinfo->xattr_len); 637 ci->i_xattrs.version = le64_to_cpu(info->xattr_version); 638 xattr_blob = NULL; 639 } 640 641 inode->i_mapping->a_ops = &ceph_aops; 642 inode->i_mapping->backing_dev_info = 643 &ceph_sb_to_client(inode->i_sb)->backing_dev_info; 644 645 switch (inode->i_mode & S_IFMT) { 646 case S_IFIFO: 647 case S_IFBLK: 648 case S_IFCHR: 649 case S_IFSOCK: 650 init_special_inode(inode, inode->i_mode, inode->i_rdev); 651 inode->i_op = &ceph_file_iops; 652 break; 653 case S_IFREG: 654 inode->i_op = &ceph_file_iops; 655 inode->i_fop = &ceph_file_fops; 656 break; 657 case S_IFLNK: 658 inode->i_op = &ceph_symlink_iops; 659 if (!ci->i_symlink) { 660 int symlen = iinfo->symlink_len; 661 char *sym; 662 663 BUG_ON(symlen != inode->i_size); 664 spin_unlock(&inode->i_lock); 665 666 err = -ENOMEM; 667 sym = kmalloc(symlen+1, GFP_NOFS); 668 if (!sym) 669 goto out; 670 memcpy(sym, iinfo->symlink, symlen); 671 sym[symlen] = 0; 672 673 spin_lock(&inode->i_lock); 674 if (!ci->i_symlink) 675 ci->i_symlink = sym; 676 else 677 kfree(sym); /* lost a race */ 678 } 679 break; 680 case S_IFDIR: 681 inode->i_op = &ceph_dir_iops; 682 inode->i_fop = &ceph_dir_fops; 683 684 ci->i_files = le64_to_cpu(info->files); 685 ci->i_subdirs = le64_to_cpu(info->subdirs); 686 ci->i_rbytes = le64_to_cpu(info->rbytes); 687 ci->i_rfiles = le64_to_cpu(info->rfiles); 688 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs); 689 ceph_decode_timespec(&ci->i_rctime, &info->rctime); 690 691 /* set dir completion flag? */ 692 if (ci->i_files == 0 && ci->i_subdirs == 0 && 693 ceph_snap(inode) == CEPH_NOSNAP && 694 (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED) && 695 (issued & CEPH_CAP_FILE_EXCL) == 0 && 696 (ci->i_ceph_flags & CEPH_I_COMPLETE) == 0) { 697 dout(" marking %p complete (empty)\n", inode); 698 ci->i_ceph_flags |= CEPH_I_COMPLETE; 699 ci->i_max_offset = 2; 700 } 701 702 /* it may be better to set st_size in getattr instead? */ 703 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), RBYTES)) 704 inode->i_size = ci->i_rbytes; 705 break; 706 default: 707 pr_err("fill_inode %llx.%llx BAD mode 0%o\n", 708 ceph_vinop(inode), inode->i_mode); 709 } 710 711 no_change: 712 spin_unlock(&inode->i_lock); 713 714 /* queue truncate if we saw i_size decrease */ 715 if (queue_trunc) 716 ceph_queue_vmtruncate(inode); 717 718 /* populate frag tree */ 719 /* FIXME: move me up, if/when version reflects fragtree changes */ 720 nsplits = le32_to_cpu(info->fragtree.nsplits); 721 mutex_lock(&ci->i_fragtree_mutex); 722 for (i = 0; i < nsplits; i++) { 723 u32 id = le32_to_cpu(info->fragtree.splits[i].frag); 724 struct ceph_inode_frag *frag = __get_or_create_frag(ci, id); 725 726 if (IS_ERR(frag)) 727 continue; 728 frag->split_by = le32_to_cpu(info->fragtree.splits[i].by); 729 dout(" frag %x split by %d\n", frag->frag, frag->split_by); 730 } 731 mutex_unlock(&ci->i_fragtree_mutex); 732 733 /* were we issued a capability? */ 734 if (info->cap.caps) { 735 if (ceph_snap(inode) == CEPH_NOSNAP) { 736 ceph_add_cap(inode, session, 737 le64_to_cpu(info->cap.cap_id), 738 cap_fmode, 739 le32_to_cpu(info->cap.caps), 740 le32_to_cpu(info->cap.wanted), 741 le32_to_cpu(info->cap.seq), 742 le32_to_cpu(info->cap.mseq), 743 le64_to_cpu(info->cap.realm), 744 info->cap.flags, 745 caps_reservation); 746 } else { 747 spin_lock(&inode->i_lock); 748 dout(" %p got snap_caps %s\n", inode, 749 ceph_cap_string(le32_to_cpu(info->cap.caps))); 750 ci->i_snap_caps |= le32_to_cpu(info->cap.caps); 751 if (cap_fmode >= 0) 752 __ceph_get_fmode(ci, cap_fmode); 753 spin_unlock(&inode->i_lock); 754 } 755 } else if (cap_fmode >= 0) { 756 pr_warning("mds issued no caps on %llx.%llx\n", 757 ceph_vinop(inode)); 758 __ceph_get_fmode(ci, cap_fmode); 759 } 760 761 /* update delegation info? */ 762 if (dirinfo) 763 ceph_fill_dirfrag(inode, dirinfo); 764 765 err = 0; 766 767 out: 768 if (xattr_blob) 769 ceph_buffer_put(xattr_blob); 770 return err; 771 } 772 773 /* 774 * caller should hold session s_mutex. 775 */ 776 static void update_dentry_lease(struct dentry *dentry, 777 struct ceph_mds_reply_lease *lease, 778 struct ceph_mds_session *session, 779 unsigned long from_time) 780 { 781 struct ceph_dentry_info *di = ceph_dentry(dentry); 782 long unsigned duration = le32_to_cpu(lease->duration_ms); 783 long unsigned ttl = from_time + (duration * HZ) / 1000; 784 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000; 785 struct inode *dir; 786 787 /* only track leases on regular dentries */ 788 if (dentry->d_op != &ceph_dentry_ops) 789 return; 790 791 spin_lock(&dentry->d_lock); 792 dout("update_dentry_lease %p mask %d duration %lu ms ttl %lu\n", 793 dentry, le16_to_cpu(lease->mask), duration, ttl); 794 795 /* make lease_rdcache_gen match directory */ 796 dir = dentry->d_parent->d_inode; 797 di->lease_shared_gen = ceph_inode(dir)->i_shared_gen; 798 799 if (lease->mask == 0) 800 goto out_unlock; 801 802 if (di->lease_gen == session->s_cap_gen && 803 time_before(ttl, dentry->d_time)) 804 goto out_unlock; /* we already have a newer lease. */ 805 806 if (di->lease_session && di->lease_session != session) 807 goto out_unlock; 808 809 ceph_dentry_lru_touch(dentry); 810 811 if (!di->lease_session) 812 di->lease_session = ceph_get_mds_session(session); 813 di->lease_gen = session->s_cap_gen; 814 di->lease_seq = le32_to_cpu(lease->seq); 815 di->lease_renew_after = half_ttl; 816 di->lease_renew_from = 0; 817 dentry->d_time = ttl; 818 out_unlock: 819 spin_unlock(&dentry->d_lock); 820 return; 821 } 822 823 /* 824 * Set dentry's directory position based on the current dir's max, and 825 * order it in d_subdirs, so that dcache_readdir behaves. 826 */ 827 static void ceph_set_dentry_offset(struct dentry *dn) 828 { 829 struct dentry *dir = dn->d_parent; 830 struct inode *inode = dn->d_parent->d_inode; 831 struct ceph_dentry_info *di; 832 833 BUG_ON(!inode); 834 835 di = ceph_dentry(dn); 836 837 spin_lock(&inode->i_lock); 838 if ((ceph_inode(inode)->i_ceph_flags & CEPH_I_COMPLETE) == 0) { 839 spin_unlock(&inode->i_lock); 840 return; 841 } 842 di->offset = ceph_inode(inode)->i_max_offset++; 843 spin_unlock(&inode->i_lock); 844 845 spin_lock(&dcache_lock); 846 spin_lock(&dn->d_lock); 847 list_move(&dn->d_u.d_child, &dir->d_subdirs); 848 dout("set_dentry_offset %p %lld (%p %p)\n", dn, di->offset, 849 dn->d_u.d_child.prev, dn->d_u.d_child.next); 850 spin_unlock(&dn->d_lock); 851 spin_unlock(&dcache_lock); 852 } 853 854 /* 855 * splice a dentry to an inode. 856 * caller must hold directory i_mutex for this to be safe. 857 * 858 * we will only rehash the resulting dentry if @prehash is 859 * true; @prehash will be set to false (for the benefit of 860 * the caller) if we fail. 861 */ 862 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in, 863 bool *prehash, bool set_offset) 864 { 865 struct dentry *realdn; 866 867 BUG_ON(dn->d_inode); 868 869 /* dn must be unhashed */ 870 if (!d_unhashed(dn)) 871 d_drop(dn); 872 realdn = d_materialise_unique(dn, in); 873 if (IS_ERR(realdn)) { 874 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n", 875 PTR_ERR(realdn), dn, in, ceph_vinop(in)); 876 if (prehash) 877 *prehash = false; /* don't rehash on error */ 878 dn = realdn; /* note realdn contains the error */ 879 goto out; 880 } else if (realdn) { 881 dout("dn %p (%d) spliced with %p (%d) " 882 "inode %p ino %llx.%llx\n", 883 dn, atomic_read(&dn->d_count), 884 realdn, atomic_read(&realdn->d_count), 885 realdn->d_inode, ceph_vinop(realdn->d_inode)); 886 dput(dn); 887 dn = realdn; 888 } else { 889 BUG_ON(!ceph_dentry(dn)); 890 dout("dn %p attached to %p ino %llx.%llx\n", 891 dn, dn->d_inode, ceph_vinop(dn->d_inode)); 892 } 893 if ((!prehash || *prehash) && d_unhashed(dn)) 894 d_rehash(dn); 895 if (set_offset) 896 ceph_set_dentry_offset(dn); 897 out: 898 return dn; 899 } 900 901 /* 902 * Incorporate results into the local cache. This is either just 903 * one inode, or a directory, dentry, and possibly linked-to inode (e.g., 904 * after a lookup). 905 * 906 * A reply may contain 907 * a directory inode along with a dentry. 908 * and/or a target inode 909 * 910 * Called with snap_rwsem (read). 911 */ 912 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req, 913 struct ceph_mds_session *session) 914 { 915 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 916 struct inode *in = NULL; 917 struct ceph_mds_reply_inode *ininfo; 918 struct ceph_vino vino; 919 struct ceph_fs_client *fsc = ceph_sb_to_client(sb); 920 int i = 0; 921 int err = 0; 922 923 dout("fill_trace %p is_dentry %d is_target %d\n", req, 924 rinfo->head->is_dentry, rinfo->head->is_target); 925 926 #if 0 927 /* 928 * Debugging hook: 929 * 930 * If we resend completed ops to a recovering mds, we get no 931 * trace. Since that is very rare, pretend this is the case 932 * to ensure the 'no trace' handlers in the callers behave. 933 * 934 * Fill in inodes unconditionally to avoid breaking cap 935 * invariants. 936 */ 937 if (rinfo->head->op & CEPH_MDS_OP_WRITE) { 938 pr_info("fill_trace faking empty trace on %lld %s\n", 939 req->r_tid, ceph_mds_op_name(rinfo->head->op)); 940 if (rinfo->head->is_dentry) { 941 rinfo->head->is_dentry = 0; 942 err = fill_inode(req->r_locked_dir, 943 &rinfo->diri, rinfo->dirfrag, 944 session, req->r_request_started, -1); 945 } 946 if (rinfo->head->is_target) { 947 rinfo->head->is_target = 0; 948 ininfo = rinfo->targeti.in; 949 vino.ino = le64_to_cpu(ininfo->ino); 950 vino.snap = le64_to_cpu(ininfo->snapid); 951 in = ceph_get_inode(sb, vino); 952 err = fill_inode(in, &rinfo->targeti, NULL, 953 session, req->r_request_started, 954 req->r_fmode); 955 iput(in); 956 } 957 } 958 #endif 959 960 if (!rinfo->head->is_target && !rinfo->head->is_dentry) { 961 dout("fill_trace reply is empty!\n"); 962 if (rinfo->head->result == 0 && req->r_locked_dir) 963 ceph_invalidate_dir_request(req); 964 return 0; 965 } 966 967 if (rinfo->head->is_dentry) { 968 struct inode *dir = req->r_locked_dir; 969 970 err = fill_inode(dir, &rinfo->diri, rinfo->dirfrag, 971 session, req->r_request_started, -1, 972 &req->r_caps_reservation); 973 if (err < 0) 974 return err; 975 } 976 977 /* 978 * ignore null lease/binding on snapdir ENOENT, or else we 979 * will have trouble splicing in the virtual snapdir later 980 */ 981 if (rinfo->head->is_dentry && !req->r_aborted && 982 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name, 983 fsc->mount_options->snapdir_name, 984 req->r_dentry->d_name.len))) { 985 /* 986 * lookup link rename : null -> possibly existing inode 987 * mknod symlink mkdir : null -> new inode 988 * unlink : linked -> null 989 */ 990 struct inode *dir = req->r_locked_dir; 991 struct dentry *dn = req->r_dentry; 992 bool have_dir_cap, have_lease; 993 994 BUG_ON(!dn); 995 BUG_ON(!dir); 996 BUG_ON(dn->d_parent->d_inode != dir); 997 BUG_ON(ceph_ino(dir) != 998 le64_to_cpu(rinfo->diri.in->ino)); 999 BUG_ON(ceph_snap(dir) != 1000 le64_to_cpu(rinfo->diri.in->snapid)); 1001 1002 /* do we have a lease on the whole dir? */ 1003 have_dir_cap = 1004 (le32_to_cpu(rinfo->diri.in->cap.caps) & 1005 CEPH_CAP_FILE_SHARED); 1006 1007 /* do we have a dn lease? */ 1008 have_lease = have_dir_cap || 1009 (le16_to_cpu(rinfo->dlease->mask) & 1010 CEPH_LOCK_DN); 1011 1012 if (!have_lease) 1013 dout("fill_trace no dentry lease or dir cap\n"); 1014 1015 /* rename? */ 1016 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) { 1017 dout(" src %p '%.*s' dst %p '%.*s'\n", 1018 req->r_old_dentry, 1019 req->r_old_dentry->d_name.len, 1020 req->r_old_dentry->d_name.name, 1021 dn, dn->d_name.len, dn->d_name.name); 1022 dout("fill_trace doing d_move %p -> %p\n", 1023 req->r_old_dentry, dn); 1024 1025 /* d_move screws up d_subdirs order */ 1026 ceph_i_clear(dir, CEPH_I_COMPLETE); 1027 1028 d_move(req->r_old_dentry, dn); 1029 dout(" src %p '%.*s' dst %p '%.*s'\n", 1030 req->r_old_dentry, 1031 req->r_old_dentry->d_name.len, 1032 req->r_old_dentry->d_name.name, 1033 dn, dn->d_name.len, dn->d_name.name); 1034 1035 /* ensure target dentry is invalidated, despite 1036 rehashing bug in vfs_rename_dir */ 1037 ceph_invalidate_dentry_lease(dn); 1038 1039 /* take overwritten dentry's readdir offset */ 1040 dout("dn %p gets %p offset %lld (old offset %lld)\n", 1041 req->r_old_dentry, dn, ceph_dentry(dn)->offset, 1042 ceph_dentry(req->r_old_dentry)->offset); 1043 ceph_dentry(req->r_old_dentry)->offset = 1044 ceph_dentry(dn)->offset; 1045 1046 dn = req->r_old_dentry; /* use old_dentry */ 1047 in = dn->d_inode; 1048 } 1049 1050 /* null dentry? */ 1051 if (!rinfo->head->is_target) { 1052 dout("fill_trace null dentry\n"); 1053 if (dn->d_inode) { 1054 dout("d_delete %p\n", dn); 1055 d_delete(dn); 1056 } else { 1057 dout("d_instantiate %p NULL\n", dn); 1058 d_instantiate(dn, NULL); 1059 if (have_lease && d_unhashed(dn)) 1060 d_rehash(dn); 1061 update_dentry_lease(dn, rinfo->dlease, 1062 session, 1063 req->r_request_started); 1064 } 1065 goto done; 1066 } 1067 1068 /* attach proper inode */ 1069 ininfo = rinfo->targeti.in; 1070 vino.ino = le64_to_cpu(ininfo->ino); 1071 vino.snap = le64_to_cpu(ininfo->snapid); 1072 in = dn->d_inode; 1073 if (!in) { 1074 in = ceph_get_inode(sb, vino); 1075 if (IS_ERR(in)) { 1076 pr_err("fill_trace bad get_inode " 1077 "%llx.%llx\n", vino.ino, vino.snap); 1078 err = PTR_ERR(in); 1079 d_delete(dn); 1080 goto done; 1081 } 1082 dn = splice_dentry(dn, in, &have_lease, true); 1083 if (IS_ERR(dn)) { 1084 err = PTR_ERR(dn); 1085 goto done; 1086 } 1087 req->r_dentry = dn; /* may have spliced */ 1088 igrab(in); 1089 } else if (ceph_ino(in) == vino.ino && 1090 ceph_snap(in) == vino.snap) { 1091 igrab(in); 1092 } else { 1093 dout(" %p links to %p %llx.%llx, not %llx.%llx\n", 1094 dn, in, ceph_ino(in), ceph_snap(in), 1095 vino.ino, vino.snap); 1096 have_lease = false; 1097 in = NULL; 1098 } 1099 1100 if (have_lease) 1101 update_dentry_lease(dn, rinfo->dlease, session, 1102 req->r_request_started); 1103 dout(" final dn %p\n", dn); 1104 i++; 1105 } else if (req->r_op == CEPH_MDS_OP_LOOKUPSNAP || 1106 req->r_op == CEPH_MDS_OP_MKSNAP) { 1107 struct dentry *dn = req->r_dentry; 1108 1109 /* fill out a snapdir LOOKUPSNAP dentry */ 1110 BUG_ON(!dn); 1111 BUG_ON(!req->r_locked_dir); 1112 BUG_ON(ceph_snap(req->r_locked_dir) != CEPH_SNAPDIR); 1113 ininfo = rinfo->targeti.in; 1114 vino.ino = le64_to_cpu(ininfo->ino); 1115 vino.snap = le64_to_cpu(ininfo->snapid); 1116 in = ceph_get_inode(sb, vino); 1117 if (IS_ERR(in)) { 1118 pr_err("fill_inode get_inode badness %llx.%llx\n", 1119 vino.ino, vino.snap); 1120 err = PTR_ERR(in); 1121 d_delete(dn); 1122 goto done; 1123 } 1124 dout(" linking snapped dir %p to dn %p\n", in, dn); 1125 dn = splice_dentry(dn, in, NULL, true); 1126 if (IS_ERR(dn)) { 1127 err = PTR_ERR(dn); 1128 goto done; 1129 } 1130 req->r_dentry = dn; /* may have spliced */ 1131 igrab(in); 1132 rinfo->head->is_dentry = 1; /* fool notrace handlers */ 1133 } 1134 1135 if (rinfo->head->is_target) { 1136 vino.ino = le64_to_cpu(rinfo->targeti.in->ino); 1137 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid); 1138 1139 if (in == NULL || ceph_ino(in) != vino.ino || 1140 ceph_snap(in) != vino.snap) { 1141 in = ceph_get_inode(sb, vino); 1142 if (IS_ERR(in)) { 1143 err = PTR_ERR(in); 1144 goto done; 1145 } 1146 } 1147 req->r_target_inode = in; 1148 1149 err = fill_inode(in, 1150 &rinfo->targeti, NULL, 1151 session, req->r_request_started, 1152 (le32_to_cpu(rinfo->head->result) == 0) ? 1153 req->r_fmode : -1, 1154 &req->r_caps_reservation); 1155 if (err < 0) { 1156 pr_err("fill_inode badness %p %llx.%llx\n", 1157 in, ceph_vinop(in)); 1158 goto done; 1159 } 1160 } 1161 1162 done: 1163 dout("fill_trace done err=%d\n", err); 1164 return err; 1165 } 1166 1167 /* 1168 * Prepopulate our cache with readdir results, leases, etc. 1169 */ 1170 int ceph_readdir_prepopulate(struct ceph_mds_request *req, 1171 struct ceph_mds_session *session) 1172 { 1173 struct dentry *parent = req->r_dentry; 1174 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1175 struct qstr dname; 1176 struct dentry *dn; 1177 struct inode *in; 1178 int err = 0, i; 1179 struct inode *snapdir = NULL; 1180 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base; 1181 u64 frag = le32_to_cpu(rhead->args.readdir.frag); 1182 struct ceph_dentry_info *di; 1183 1184 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) { 1185 snapdir = ceph_get_snapdir(parent->d_inode); 1186 parent = d_find_alias(snapdir); 1187 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n", 1188 rinfo->dir_nr, parent); 1189 } else { 1190 dout("readdir_prepopulate %d items under dn %p\n", 1191 rinfo->dir_nr, parent); 1192 if (rinfo->dir_dir) 1193 ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir); 1194 } 1195 1196 for (i = 0; i < rinfo->dir_nr; i++) { 1197 struct ceph_vino vino; 1198 1199 dname.name = rinfo->dir_dname[i]; 1200 dname.len = rinfo->dir_dname_len[i]; 1201 dname.hash = full_name_hash(dname.name, dname.len); 1202 1203 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino); 1204 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid); 1205 1206 retry_lookup: 1207 dn = d_lookup(parent, &dname); 1208 dout("d_lookup on parent=%p name=%.*s got %p\n", 1209 parent, dname.len, dname.name, dn); 1210 1211 if (!dn) { 1212 dn = d_alloc(parent, &dname); 1213 dout("d_alloc %p '%.*s' = %p\n", parent, 1214 dname.len, dname.name, dn); 1215 if (dn == NULL) { 1216 dout("d_alloc badness\n"); 1217 err = -ENOMEM; 1218 goto out; 1219 } 1220 err = ceph_init_dentry(dn); 1221 if (err < 0) { 1222 dput(dn); 1223 goto out; 1224 } 1225 } else if (dn->d_inode && 1226 (ceph_ino(dn->d_inode) != vino.ino || 1227 ceph_snap(dn->d_inode) != vino.snap)) { 1228 dout(" dn %p points to wrong inode %p\n", 1229 dn, dn->d_inode); 1230 d_delete(dn); 1231 dput(dn); 1232 goto retry_lookup; 1233 } else { 1234 /* reorder parent's d_subdirs */ 1235 spin_lock(&dcache_lock); 1236 spin_lock(&dn->d_lock); 1237 list_move(&dn->d_u.d_child, &parent->d_subdirs); 1238 spin_unlock(&dn->d_lock); 1239 spin_unlock(&dcache_lock); 1240 } 1241 1242 di = dn->d_fsdata; 1243 di->offset = ceph_make_fpos(frag, i + req->r_readdir_offset); 1244 1245 /* inode */ 1246 if (dn->d_inode) { 1247 in = dn->d_inode; 1248 } else { 1249 in = ceph_get_inode(parent->d_sb, vino); 1250 if (IS_ERR(in)) { 1251 dout("new_inode badness\n"); 1252 d_delete(dn); 1253 dput(dn); 1254 err = PTR_ERR(in); 1255 goto out; 1256 } 1257 dn = splice_dentry(dn, in, NULL, false); 1258 if (IS_ERR(dn)) 1259 dn = NULL; 1260 } 1261 1262 if (fill_inode(in, &rinfo->dir_in[i], NULL, session, 1263 req->r_request_started, -1, 1264 &req->r_caps_reservation) < 0) { 1265 pr_err("fill_inode badness on %p\n", in); 1266 goto next_item; 1267 } 1268 if (dn) 1269 update_dentry_lease(dn, rinfo->dir_dlease[i], 1270 req->r_session, 1271 req->r_request_started); 1272 next_item: 1273 if (dn) 1274 dput(dn); 1275 } 1276 req->r_did_prepopulate = true; 1277 1278 out: 1279 if (snapdir) { 1280 iput(snapdir); 1281 dput(parent); 1282 } 1283 dout("readdir_prepopulate done\n"); 1284 return err; 1285 } 1286 1287 int ceph_inode_set_size(struct inode *inode, loff_t size) 1288 { 1289 struct ceph_inode_info *ci = ceph_inode(inode); 1290 int ret = 0; 1291 1292 spin_lock(&inode->i_lock); 1293 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size); 1294 inode->i_size = size; 1295 inode->i_blocks = (size + (1 << 9) - 1) >> 9; 1296 1297 /* tell the MDS if we are approaching max_size */ 1298 if ((size << 1) >= ci->i_max_size && 1299 (ci->i_reported_size << 1) < ci->i_max_size) 1300 ret = 1; 1301 1302 spin_unlock(&inode->i_lock); 1303 return ret; 1304 } 1305 1306 /* 1307 * Write back inode data in a worker thread. (This can't be done 1308 * in the message handler context.) 1309 */ 1310 void ceph_queue_writeback(struct inode *inode) 1311 { 1312 if (queue_work(ceph_inode_to_client(inode)->wb_wq, 1313 &ceph_inode(inode)->i_wb_work)) { 1314 dout("ceph_queue_writeback %p\n", inode); 1315 igrab(inode); 1316 } else { 1317 dout("ceph_queue_writeback %p failed\n", inode); 1318 } 1319 } 1320 1321 static void ceph_writeback_work(struct work_struct *work) 1322 { 1323 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1324 i_wb_work); 1325 struct inode *inode = &ci->vfs_inode; 1326 1327 dout("writeback %p\n", inode); 1328 filemap_fdatawrite(&inode->i_data); 1329 iput(inode); 1330 } 1331 1332 /* 1333 * queue an async invalidation 1334 */ 1335 void ceph_queue_invalidate(struct inode *inode) 1336 { 1337 if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq, 1338 &ceph_inode(inode)->i_pg_inv_work)) { 1339 dout("ceph_queue_invalidate %p\n", inode); 1340 igrab(inode); 1341 } else { 1342 dout("ceph_queue_invalidate %p failed\n", inode); 1343 } 1344 } 1345 1346 /* 1347 * invalidate any pages that are not dirty or under writeback. this 1348 * includes pages that are clean and mapped. 1349 */ 1350 static void ceph_invalidate_nondirty_pages(struct address_space *mapping) 1351 { 1352 struct pagevec pvec; 1353 pgoff_t next = 0; 1354 int i; 1355 1356 pagevec_init(&pvec, 0); 1357 while (pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { 1358 for (i = 0; i < pagevec_count(&pvec); i++) { 1359 struct page *page = pvec.pages[i]; 1360 pgoff_t index; 1361 int skip_page = 1362 (PageDirty(page) || PageWriteback(page)); 1363 1364 if (!skip_page) 1365 skip_page = !trylock_page(page); 1366 1367 /* 1368 * We really shouldn't be looking at the ->index of an 1369 * unlocked page. But we're not allowed to lock these 1370 * pages. So we rely upon nobody altering the ->index 1371 * of this (pinned-by-us) page. 1372 */ 1373 index = page->index; 1374 if (index > next) 1375 next = index; 1376 next++; 1377 1378 if (skip_page) 1379 continue; 1380 1381 generic_error_remove_page(mapping, page); 1382 unlock_page(page); 1383 } 1384 pagevec_release(&pvec); 1385 cond_resched(); 1386 } 1387 } 1388 1389 /* 1390 * Invalidate inode pages in a worker thread. (This can't be done 1391 * in the message handler context.) 1392 */ 1393 static void ceph_invalidate_work(struct work_struct *work) 1394 { 1395 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1396 i_pg_inv_work); 1397 struct inode *inode = &ci->vfs_inode; 1398 u32 orig_gen; 1399 int check = 0; 1400 1401 spin_lock(&inode->i_lock); 1402 dout("invalidate_pages %p gen %d revoking %d\n", inode, 1403 ci->i_rdcache_gen, ci->i_rdcache_revoking); 1404 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) { 1405 /* nevermind! */ 1406 spin_unlock(&inode->i_lock); 1407 goto out; 1408 } 1409 orig_gen = ci->i_rdcache_gen; 1410 spin_unlock(&inode->i_lock); 1411 1412 ceph_invalidate_nondirty_pages(inode->i_mapping); 1413 1414 spin_lock(&inode->i_lock); 1415 if (orig_gen == ci->i_rdcache_gen && 1416 orig_gen == ci->i_rdcache_revoking) { 1417 dout("invalidate_pages %p gen %d successful\n", inode, 1418 ci->i_rdcache_gen); 1419 ci->i_rdcache_revoking--; 1420 check = 1; 1421 } else { 1422 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n", 1423 inode, orig_gen, ci->i_rdcache_gen, 1424 ci->i_rdcache_revoking); 1425 } 1426 spin_unlock(&inode->i_lock); 1427 1428 if (check) 1429 ceph_check_caps(ci, 0, NULL); 1430 out: 1431 iput(inode); 1432 } 1433 1434 1435 /* 1436 * called by trunc_wq; take i_mutex ourselves 1437 * 1438 * We also truncate in a separate thread as well. 1439 */ 1440 static void ceph_vmtruncate_work(struct work_struct *work) 1441 { 1442 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1443 i_vmtruncate_work); 1444 struct inode *inode = &ci->vfs_inode; 1445 1446 dout("vmtruncate_work %p\n", inode); 1447 mutex_lock(&inode->i_mutex); 1448 __ceph_do_pending_vmtruncate(inode); 1449 mutex_unlock(&inode->i_mutex); 1450 iput(inode); 1451 } 1452 1453 /* 1454 * Queue an async vmtruncate. If we fail to queue work, we will handle 1455 * the truncation the next time we call __ceph_do_pending_vmtruncate. 1456 */ 1457 void ceph_queue_vmtruncate(struct inode *inode) 1458 { 1459 struct ceph_inode_info *ci = ceph_inode(inode); 1460 1461 if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq, 1462 &ci->i_vmtruncate_work)) { 1463 dout("ceph_queue_vmtruncate %p\n", inode); 1464 igrab(inode); 1465 } else { 1466 dout("ceph_queue_vmtruncate %p failed, pending=%d\n", 1467 inode, ci->i_truncate_pending); 1468 } 1469 } 1470 1471 /* 1472 * called with i_mutex held. 1473 * 1474 * Make sure any pending truncation is applied before doing anything 1475 * that may depend on it. 1476 */ 1477 void __ceph_do_pending_vmtruncate(struct inode *inode) 1478 { 1479 struct ceph_inode_info *ci = ceph_inode(inode); 1480 u64 to; 1481 int wrbuffer_refs, wake = 0; 1482 1483 retry: 1484 spin_lock(&inode->i_lock); 1485 if (ci->i_truncate_pending == 0) { 1486 dout("__do_pending_vmtruncate %p none pending\n", inode); 1487 spin_unlock(&inode->i_lock); 1488 return; 1489 } 1490 1491 /* 1492 * make sure any dirty snapped pages are flushed before we 1493 * possibly truncate them.. so write AND block! 1494 */ 1495 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) { 1496 dout("__do_pending_vmtruncate %p flushing snaps first\n", 1497 inode); 1498 spin_unlock(&inode->i_lock); 1499 filemap_write_and_wait_range(&inode->i_data, 0, 1500 inode->i_sb->s_maxbytes); 1501 goto retry; 1502 } 1503 1504 to = ci->i_truncate_size; 1505 wrbuffer_refs = ci->i_wrbuffer_ref; 1506 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode, 1507 ci->i_truncate_pending, to); 1508 spin_unlock(&inode->i_lock); 1509 1510 truncate_inode_pages(inode->i_mapping, to); 1511 1512 spin_lock(&inode->i_lock); 1513 ci->i_truncate_pending--; 1514 if (ci->i_truncate_pending == 0) 1515 wake = 1; 1516 spin_unlock(&inode->i_lock); 1517 1518 if (wrbuffer_refs == 0) 1519 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); 1520 if (wake) 1521 wake_up_all(&ci->i_cap_wq); 1522 } 1523 1524 1525 /* 1526 * symlinks 1527 */ 1528 static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd) 1529 { 1530 struct ceph_inode_info *ci = ceph_inode(dentry->d_inode); 1531 nd_set_link(nd, ci->i_symlink); 1532 return NULL; 1533 } 1534 1535 static const struct inode_operations ceph_symlink_iops = { 1536 .readlink = generic_readlink, 1537 .follow_link = ceph_sym_follow_link, 1538 }; 1539 1540 /* 1541 * setattr 1542 */ 1543 int ceph_setattr(struct dentry *dentry, struct iattr *attr) 1544 { 1545 struct inode *inode = dentry->d_inode; 1546 struct ceph_inode_info *ci = ceph_inode(inode); 1547 struct inode *parent_inode = dentry->d_parent->d_inode; 1548 const unsigned int ia_valid = attr->ia_valid; 1549 struct ceph_mds_request *req; 1550 struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc; 1551 int issued; 1552 int release = 0, dirtied = 0; 1553 int mask = 0; 1554 int err = 0; 1555 1556 if (ceph_snap(inode) != CEPH_NOSNAP) 1557 return -EROFS; 1558 1559 __ceph_do_pending_vmtruncate(inode); 1560 1561 err = inode_change_ok(inode, attr); 1562 if (err != 0) 1563 return err; 1564 1565 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR, 1566 USE_AUTH_MDS); 1567 if (IS_ERR(req)) 1568 return PTR_ERR(req); 1569 1570 spin_lock(&inode->i_lock); 1571 issued = __ceph_caps_issued(ci, NULL); 1572 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued)); 1573 1574 if (ia_valid & ATTR_UID) { 1575 dout("setattr %p uid %d -> %d\n", inode, 1576 inode->i_uid, attr->ia_uid); 1577 if (issued & CEPH_CAP_AUTH_EXCL) { 1578 inode->i_uid = attr->ia_uid; 1579 dirtied |= CEPH_CAP_AUTH_EXCL; 1580 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1581 attr->ia_uid != inode->i_uid) { 1582 req->r_args.setattr.uid = cpu_to_le32(attr->ia_uid); 1583 mask |= CEPH_SETATTR_UID; 1584 release |= CEPH_CAP_AUTH_SHARED; 1585 } 1586 } 1587 if (ia_valid & ATTR_GID) { 1588 dout("setattr %p gid %d -> %d\n", inode, 1589 inode->i_gid, attr->ia_gid); 1590 if (issued & CEPH_CAP_AUTH_EXCL) { 1591 inode->i_gid = attr->ia_gid; 1592 dirtied |= CEPH_CAP_AUTH_EXCL; 1593 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1594 attr->ia_gid != inode->i_gid) { 1595 req->r_args.setattr.gid = cpu_to_le32(attr->ia_gid); 1596 mask |= CEPH_SETATTR_GID; 1597 release |= CEPH_CAP_AUTH_SHARED; 1598 } 1599 } 1600 if (ia_valid & ATTR_MODE) { 1601 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode, 1602 attr->ia_mode); 1603 if (issued & CEPH_CAP_AUTH_EXCL) { 1604 inode->i_mode = attr->ia_mode; 1605 dirtied |= CEPH_CAP_AUTH_EXCL; 1606 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1607 attr->ia_mode != inode->i_mode) { 1608 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode); 1609 mask |= CEPH_SETATTR_MODE; 1610 release |= CEPH_CAP_AUTH_SHARED; 1611 } 1612 } 1613 1614 if (ia_valid & ATTR_ATIME) { 1615 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode, 1616 inode->i_atime.tv_sec, inode->i_atime.tv_nsec, 1617 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec); 1618 if (issued & CEPH_CAP_FILE_EXCL) { 1619 ci->i_time_warp_seq++; 1620 inode->i_atime = attr->ia_atime; 1621 dirtied |= CEPH_CAP_FILE_EXCL; 1622 } else if ((issued & CEPH_CAP_FILE_WR) && 1623 timespec_compare(&inode->i_atime, 1624 &attr->ia_atime) < 0) { 1625 inode->i_atime = attr->ia_atime; 1626 dirtied |= CEPH_CAP_FILE_WR; 1627 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1628 !timespec_equal(&inode->i_atime, &attr->ia_atime)) { 1629 ceph_encode_timespec(&req->r_args.setattr.atime, 1630 &attr->ia_atime); 1631 mask |= CEPH_SETATTR_ATIME; 1632 release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD | 1633 CEPH_CAP_FILE_WR; 1634 } 1635 } 1636 if (ia_valid & ATTR_MTIME) { 1637 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode, 1638 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 1639 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec); 1640 if (issued & CEPH_CAP_FILE_EXCL) { 1641 ci->i_time_warp_seq++; 1642 inode->i_mtime = attr->ia_mtime; 1643 dirtied |= CEPH_CAP_FILE_EXCL; 1644 } else if ((issued & CEPH_CAP_FILE_WR) && 1645 timespec_compare(&inode->i_mtime, 1646 &attr->ia_mtime) < 0) { 1647 inode->i_mtime = attr->ia_mtime; 1648 dirtied |= CEPH_CAP_FILE_WR; 1649 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1650 !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) { 1651 ceph_encode_timespec(&req->r_args.setattr.mtime, 1652 &attr->ia_mtime); 1653 mask |= CEPH_SETATTR_MTIME; 1654 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | 1655 CEPH_CAP_FILE_WR; 1656 } 1657 } 1658 if (ia_valid & ATTR_SIZE) { 1659 dout("setattr %p size %lld -> %lld\n", inode, 1660 inode->i_size, attr->ia_size); 1661 if (attr->ia_size > inode->i_sb->s_maxbytes) { 1662 err = -EINVAL; 1663 goto out; 1664 } 1665 if ((issued & CEPH_CAP_FILE_EXCL) && 1666 attr->ia_size > inode->i_size) { 1667 inode->i_size = attr->ia_size; 1668 inode->i_blocks = 1669 (attr->ia_size + (1 << 9) - 1) >> 9; 1670 inode->i_ctime = attr->ia_ctime; 1671 ci->i_reported_size = attr->ia_size; 1672 dirtied |= CEPH_CAP_FILE_EXCL; 1673 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1674 attr->ia_size != inode->i_size) { 1675 req->r_args.setattr.size = cpu_to_le64(attr->ia_size); 1676 req->r_args.setattr.old_size = 1677 cpu_to_le64(inode->i_size); 1678 mask |= CEPH_SETATTR_SIZE; 1679 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | 1680 CEPH_CAP_FILE_WR; 1681 } 1682 } 1683 1684 /* these do nothing */ 1685 if (ia_valid & ATTR_CTIME) { 1686 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME| 1687 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0; 1688 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode, 1689 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 1690 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec, 1691 only ? "ctime only" : "ignored"); 1692 inode->i_ctime = attr->ia_ctime; 1693 if (only) { 1694 /* 1695 * if kernel wants to dirty ctime but nothing else, 1696 * we need to choose a cap to dirty under, or do 1697 * a almost-no-op setattr 1698 */ 1699 if (issued & CEPH_CAP_AUTH_EXCL) 1700 dirtied |= CEPH_CAP_AUTH_EXCL; 1701 else if (issued & CEPH_CAP_FILE_EXCL) 1702 dirtied |= CEPH_CAP_FILE_EXCL; 1703 else if (issued & CEPH_CAP_XATTR_EXCL) 1704 dirtied |= CEPH_CAP_XATTR_EXCL; 1705 else 1706 mask |= CEPH_SETATTR_CTIME; 1707 } 1708 } 1709 if (ia_valid & ATTR_FILE) 1710 dout("setattr %p ATTR_FILE ... hrm!\n", inode); 1711 1712 if (dirtied) { 1713 __ceph_mark_dirty_caps(ci, dirtied); 1714 inode->i_ctime = CURRENT_TIME; 1715 } 1716 1717 release &= issued; 1718 spin_unlock(&inode->i_lock); 1719 1720 if (mask) { 1721 req->r_inode = igrab(inode); 1722 req->r_inode_drop = release; 1723 req->r_args.setattr.mask = cpu_to_le32(mask); 1724 req->r_num_caps = 1; 1725 err = ceph_mdsc_do_request(mdsc, parent_inode, req); 1726 } 1727 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err, 1728 ceph_cap_string(dirtied), mask); 1729 1730 ceph_mdsc_put_request(req); 1731 __ceph_do_pending_vmtruncate(inode); 1732 return err; 1733 out: 1734 spin_unlock(&inode->i_lock); 1735 ceph_mdsc_put_request(req); 1736 return err; 1737 } 1738 1739 /* 1740 * Verify that we have a lease on the given mask. If not, 1741 * do a getattr against an mds. 1742 */ 1743 int ceph_do_getattr(struct inode *inode, int mask) 1744 { 1745 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); 1746 struct ceph_mds_client *mdsc = fsc->mdsc; 1747 struct ceph_mds_request *req; 1748 int err; 1749 1750 if (ceph_snap(inode) == CEPH_SNAPDIR) { 1751 dout("do_getattr inode %p SNAPDIR\n", inode); 1752 return 0; 1753 } 1754 1755 dout("do_getattr inode %p mask %s mode 0%o\n", inode, ceph_cap_string(mask), inode->i_mode); 1756 if (ceph_caps_issued_mask(ceph_inode(inode), mask, 1)) 1757 return 0; 1758 1759 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS); 1760 if (IS_ERR(req)) 1761 return PTR_ERR(req); 1762 req->r_inode = igrab(inode); 1763 req->r_num_caps = 1; 1764 req->r_args.getattr.mask = cpu_to_le32(mask); 1765 err = ceph_mdsc_do_request(mdsc, NULL, req); 1766 ceph_mdsc_put_request(req); 1767 dout("do_getattr result=%d\n", err); 1768 return err; 1769 } 1770 1771 1772 /* 1773 * Check inode permissions. We verify we have a valid value for 1774 * the AUTH cap, then call the generic handler. 1775 */ 1776 int ceph_permission(struct inode *inode, int mask) 1777 { 1778 int err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED); 1779 1780 if (!err) 1781 err = generic_permission(inode, mask, NULL); 1782 return err; 1783 } 1784 1785 /* 1786 * Get all attributes. Hopefully somedata we'll have a statlite() 1787 * and can limit the fields we require to be accurate. 1788 */ 1789 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry, 1790 struct kstat *stat) 1791 { 1792 struct inode *inode = dentry->d_inode; 1793 struct ceph_inode_info *ci = ceph_inode(inode); 1794 int err; 1795 1796 err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL); 1797 if (!err) { 1798 generic_fillattr(inode, stat); 1799 stat->ino = inode->i_ino; 1800 if (ceph_snap(inode) != CEPH_NOSNAP) 1801 stat->dev = ceph_snap(inode); 1802 else 1803 stat->dev = 0; 1804 if (S_ISDIR(inode->i_mode)) { 1805 stat->size = ci->i_rbytes; 1806 stat->blocks = 0; 1807 stat->blksize = 65536; 1808 } 1809 } 1810 return err; 1811 } 1812