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 if (timespec_compare(ctime, &inode->i_ctime) > 0) { 476 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n", 477 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 478 ctime->tv_sec, ctime->tv_nsec); 479 inode->i_ctime = *ctime; 480 } 481 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) { 482 /* the MDS did a utimes() */ 483 dout("mtime %ld.%09ld -> %ld.%09ld " 484 "tw %d -> %d\n", 485 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 486 mtime->tv_sec, mtime->tv_nsec, 487 ci->i_time_warp_seq, (int)time_warp_seq); 488 489 inode->i_mtime = *mtime; 490 inode->i_atime = *atime; 491 ci->i_time_warp_seq = time_warp_seq; 492 } else if (time_warp_seq == ci->i_time_warp_seq) { 493 /* nobody did utimes(); take the max */ 494 if (timespec_compare(mtime, &inode->i_mtime) > 0) { 495 dout("mtime %ld.%09ld -> %ld.%09ld inc\n", 496 inode->i_mtime.tv_sec, 497 inode->i_mtime.tv_nsec, 498 mtime->tv_sec, mtime->tv_nsec); 499 inode->i_mtime = *mtime; 500 } 501 if (timespec_compare(atime, &inode->i_atime) > 0) { 502 dout("atime %ld.%09ld -> %ld.%09ld inc\n", 503 inode->i_atime.tv_sec, 504 inode->i_atime.tv_nsec, 505 atime->tv_sec, atime->tv_nsec); 506 inode->i_atime = *atime; 507 } 508 } else if (issued & CEPH_CAP_FILE_EXCL) { 509 /* we did a utimes(); ignore mds values */ 510 } else { 511 warn = 1; 512 } 513 } else { 514 /* we have no write caps; whatever the MDS says is true */ 515 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) { 516 inode->i_ctime = *ctime; 517 inode->i_mtime = *mtime; 518 inode->i_atime = *atime; 519 ci->i_time_warp_seq = time_warp_seq; 520 } else { 521 warn = 1; 522 } 523 } 524 if (warn) /* time_warp_seq shouldn't go backwards */ 525 dout("%p mds time_warp_seq %llu < %u\n", 526 inode, time_warp_seq, ci->i_time_warp_seq); 527 } 528 529 /* 530 * Populate an inode based on info from mds. May be called on new or 531 * existing inodes. 532 */ 533 static int fill_inode(struct inode *inode, 534 struct ceph_mds_reply_info_in *iinfo, 535 struct ceph_mds_reply_dirfrag *dirinfo, 536 struct ceph_mds_session *session, 537 unsigned long ttl_from, int cap_fmode, 538 struct ceph_cap_reservation *caps_reservation) 539 { 540 struct ceph_mds_reply_inode *info = iinfo->in; 541 struct ceph_inode_info *ci = ceph_inode(inode); 542 int i; 543 int issued, implemented; 544 struct timespec mtime, atime, ctime; 545 u32 nsplits; 546 struct ceph_buffer *xattr_blob = NULL; 547 int err = 0; 548 int queue_trunc = 0; 549 550 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n", 551 inode, ceph_vinop(inode), le64_to_cpu(info->version), 552 ci->i_version); 553 554 /* 555 * prealloc xattr data, if it looks like we'll need it. only 556 * if len > 4 (meaning there are actually xattrs; the first 4 557 * bytes are the xattr count). 558 */ 559 if (iinfo->xattr_len > 4) { 560 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS); 561 if (!xattr_blob) 562 pr_err("fill_inode ENOMEM xattr blob %d bytes\n", 563 iinfo->xattr_len); 564 } 565 566 spin_lock(&inode->i_lock); 567 568 /* 569 * provided version will be odd if inode value is projected, 570 * even if stable. skip the update if we have newer stable 571 * info (ours>=theirs, e.g. due to racing mds replies), unless 572 * we are getting projected (unstable) info (in which case the 573 * version is odd, and we want ours>theirs). 574 * us them 575 * 2 2 skip 576 * 3 2 skip 577 * 3 3 update 578 */ 579 if (le64_to_cpu(info->version) > 0 && 580 (ci->i_version & ~1) >= le64_to_cpu(info->version)) 581 goto no_change; 582 583 issued = __ceph_caps_issued(ci, &implemented); 584 issued |= implemented | __ceph_caps_dirty(ci); 585 586 /* update inode */ 587 ci->i_version = le64_to_cpu(info->version); 588 inode->i_version++; 589 inode->i_rdev = le32_to_cpu(info->rdev); 590 591 if ((issued & CEPH_CAP_AUTH_EXCL) == 0) { 592 inode->i_mode = le32_to_cpu(info->mode); 593 inode->i_uid = le32_to_cpu(info->uid); 594 inode->i_gid = le32_to_cpu(info->gid); 595 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode, 596 inode->i_uid, inode->i_gid); 597 } 598 599 if ((issued & CEPH_CAP_LINK_EXCL) == 0) 600 inode->i_nlink = le32_to_cpu(info->nlink); 601 602 /* be careful with mtime, atime, size */ 603 ceph_decode_timespec(&atime, &info->atime); 604 ceph_decode_timespec(&mtime, &info->mtime); 605 ceph_decode_timespec(&ctime, &info->ctime); 606 queue_trunc = ceph_fill_file_size(inode, issued, 607 le32_to_cpu(info->truncate_seq), 608 le64_to_cpu(info->truncate_size), 609 le64_to_cpu(info->size)); 610 ceph_fill_file_time(inode, issued, 611 le32_to_cpu(info->time_warp_seq), 612 &ctime, &mtime, &atime); 613 614 /* only update max_size on auth cap */ 615 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) && 616 ci->i_max_size != le64_to_cpu(info->max_size)) { 617 dout("max_size %lld -> %llu\n", ci->i_max_size, 618 le64_to_cpu(info->max_size)); 619 ci->i_max_size = le64_to_cpu(info->max_size); 620 } 621 622 ci->i_layout = info->layout; 623 inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1; 624 625 /* xattrs */ 626 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */ 627 if ((issued & CEPH_CAP_XATTR_EXCL) == 0 && 628 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) { 629 if (ci->i_xattrs.blob) 630 ceph_buffer_put(ci->i_xattrs.blob); 631 ci->i_xattrs.blob = xattr_blob; 632 if (xattr_blob) 633 memcpy(ci->i_xattrs.blob->vec.iov_base, 634 iinfo->xattr_data, iinfo->xattr_len); 635 ci->i_xattrs.version = le64_to_cpu(info->xattr_version); 636 xattr_blob = NULL; 637 } 638 639 inode->i_mapping->a_ops = &ceph_aops; 640 inode->i_mapping->backing_dev_info = 641 &ceph_sb_to_client(inode->i_sb)->backing_dev_info; 642 643 switch (inode->i_mode & S_IFMT) { 644 case S_IFIFO: 645 case S_IFBLK: 646 case S_IFCHR: 647 case S_IFSOCK: 648 init_special_inode(inode, inode->i_mode, inode->i_rdev); 649 inode->i_op = &ceph_file_iops; 650 break; 651 case S_IFREG: 652 inode->i_op = &ceph_file_iops; 653 inode->i_fop = &ceph_file_fops; 654 break; 655 case S_IFLNK: 656 inode->i_op = &ceph_symlink_iops; 657 if (!ci->i_symlink) { 658 int symlen = iinfo->symlink_len; 659 char *sym; 660 661 BUG_ON(symlen != inode->i_size); 662 spin_unlock(&inode->i_lock); 663 664 err = -ENOMEM; 665 sym = kmalloc(symlen+1, GFP_NOFS); 666 if (!sym) 667 goto out; 668 memcpy(sym, iinfo->symlink, symlen); 669 sym[symlen] = 0; 670 671 spin_lock(&inode->i_lock); 672 if (!ci->i_symlink) 673 ci->i_symlink = sym; 674 else 675 kfree(sym); /* lost a race */ 676 } 677 break; 678 case S_IFDIR: 679 inode->i_op = &ceph_dir_iops; 680 inode->i_fop = &ceph_dir_fops; 681 682 ci->i_files = le64_to_cpu(info->files); 683 ci->i_subdirs = le64_to_cpu(info->subdirs); 684 ci->i_rbytes = le64_to_cpu(info->rbytes); 685 ci->i_rfiles = le64_to_cpu(info->rfiles); 686 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs); 687 ceph_decode_timespec(&ci->i_rctime, &info->rctime); 688 689 /* set dir completion flag? */ 690 if (ci->i_files == 0 && ci->i_subdirs == 0 && 691 ceph_snap(inode) == CEPH_NOSNAP && 692 (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED) && 693 (issued & CEPH_CAP_FILE_EXCL) == 0 && 694 (ci->i_ceph_flags & CEPH_I_COMPLETE) == 0) { 695 dout(" marking %p complete (empty)\n", inode); 696 ci->i_ceph_flags |= CEPH_I_COMPLETE; 697 ci->i_max_offset = 2; 698 } 699 700 /* it may be better to set st_size in getattr instead? */ 701 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), RBYTES)) 702 inode->i_size = ci->i_rbytes; 703 break; 704 default: 705 pr_err("fill_inode %llx.%llx BAD mode 0%o\n", 706 ceph_vinop(inode), inode->i_mode); 707 } 708 709 no_change: 710 spin_unlock(&inode->i_lock); 711 712 /* queue truncate if we saw i_size decrease */ 713 if (queue_trunc) 714 ceph_queue_vmtruncate(inode); 715 716 /* populate frag tree */ 717 /* FIXME: move me up, if/when version reflects fragtree changes */ 718 nsplits = le32_to_cpu(info->fragtree.nsplits); 719 mutex_lock(&ci->i_fragtree_mutex); 720 for (i = 0; i < nsplits; i++) { 721 u32 id = le32_to_cpu(info->fragtree.splits[i].frag); 722 struct ceph_inode_frag *frag = __get_or_create_frag(ci, id); 723 724 if (IS_ERR(frag)) 725 continue; 726 frag->split_by = le32_to_cpu(info->fragtree.splits[i].by); 727 dout(" frag %x split by %d\n", frag->frag, frag->split_by); 728 } 729 mutex_unlock(&ci->i_fragtree_mutex); 730 731 /* were we issued a capability? */ 732 if (info->cap.caps) { 733 if (ceph_snap(inode) == CEPH_NOSNAP) { 734 ceph_add_cap(inode, session, 735 le64_to_cpu(info->cap.cap_id), 736 cap_fmode, 737 le32_to_cpu(info->cap.caps), 738 le32_to_cpu(info->cap.wanted), 739 le32_to_cpu(info->cap.seq), 740 le32_to_cpu(info->cap.mseq), 741 le64_to_cpu(info->cap.realm), 742 info->cap.flags, 743 caps_reservation); 744 } else { 745 spin_lock(&inode->i_lock); 746 dout(" %p got snap_caps %s\n", inode, 747 ceph_cap_string(le32_to_cpu(info->cap.caps))); 748 ci->i_snap_caps |= le32_to_cpu(info->cap.caps); 749 if (cap_fmode >= 0) 750 __ceph_get_fmode(ci, cap_fmode); 751 spin_unlock(&inode->i_lock); 752 } 753 } else if (cap_fmode >= 0) { 754 pr_warning("mds issued no caps on %llx.%llx\n", 755 ceph_vinop(inode)); 756 __ceph_get_fmode(ci, cap_fmode); 757 } 758 759 /* update delegation info? */ 760 if (dirinfo) 761 ceph_fill_dirfrag(inode, dirinfo); 762 763 err = 0; 764 765 out: 766 if (xattr_blob) 767 ceph_buffer_put(xattr_blob); 768 return err; 769 } 770 771 /* 772 * caller should hold session s_mutex. 773 */ 774 static void update_dentry_lease(struct dentry *dentry, 775 struct ceph_mds_reply_lease *lease, 776 struct ceph_mds_session *session, 777 unsigned long from_time) 778 { 779 struct ceph_dentry_info *di = ceph_dentry(dentry); 780 long unsigned duration = le32_to_cpu(lease->duration_ms); 781 long unsigned ttl = from_time + (duration * HZ) / 1000; 782 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000; 783 struct inode *dir; 784 785 /* only track leases on regular dentries */ 786 if (dentry->d_op != &ceph_dentry_ops) 787 return; 788 789 spin_lock(&dentry->d_lock); 790 dout("update_dentry_lease %p mask %d duration %lu ms ttl %lu\n", 791 dentry, le16_to_cpu(lease->mask), duration, ttl); 792 793 /* make lease_rdcache_gen match directory */ 794 dir = dentry->d_parent->d_inode; 795 di->lease_shared_gen = ceph_inode(dir)->i_shared_gen; 796 797 if (lease->mask == 0) 798 goto out_unlock; 799 800 if (di->lease_gen == session->s_cap_gen && 801 time_before(ttl, dentry->d_time)) 802 goto out_unlock; /* we already have a newer lease. */ 803 804 if (di->lease_session && di->lease_session != session) 805 goto out_unlock; 806 807 ceph_dentry_lru_touch(dentry); 808 809 if (!di->lease_session) 810 di->lease_session = ceph_get_mds_session(session); 811 di->lease_gen = session->s_cap_gen; 812 di->lease_seq = le32_to_cpu(lease->seq); 813 di->lease_renew_after = half_ttl; 814 di->lease_renew_from = 0; 815 dentry->d_time = ttl; 816 out_unlock: 817 spin_unlock(&dentry->d_lock); 818 return; 819 } 820 821 /* 822 * Set dentry's directory position based on the current dir's max, and 823 * order it in d_subdirs, so that dcache_readdir behaves. 824 */ 825 static void ceph_set_dentry_offset(struct dentry *dn) 826 { 827 struct dentry *dir = dn->d_parent; 828 struct inode *inode = dn->d_parent->d_inode; 829 struct ceph_dentry_info *di; 830 831 BUG_ON(!inode); 832 833 di = ceph_dentry(dn); 834 835 spin_lock(&inode->i_lock); 836 if ((ceph_inode(inode)->i_ceph_flags & CEPH_I_COMPLETE) == 0) { 837 spin_unlock(&inode->i_lock); 838 return; 839 } 840 di->offset = ceph_inode(inode)->i_max_offset++; 841 spin_unlock(&inode->i_lock); 842 843 spin_lock(&dcache_lock); 844 spin_lock(&dn->d_lock); 845 list_move(&dn->d_u.d_child, &dir->d_subdirs); 846 dout("set_dentry_offset %p %lld (%p %p)\n", dn, di->offset, 847 dn->d_u.d_child.prev, dn->d_u.d_child.next); 848 spin_unlock(&dn->d_lock); 849 spin_unlock(&dcache_lock); 850 } 851 852 /* 853 * splice a dentry to an inode. 854 * caller must hold directory i_mutex for this to be safe. 855 * 856 * we will only rehash the resulting dentry if @prehash is 857 * true; @prehash will be set to false (for the benefit of 858 * the caller) if we fail. 859 */ 860 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in, 861 bool *prehash, bool set_offset) 862 { 863 struct dentry *realdn; 864 865 BUG_ON(dn->d_inode); 866 867 /* dn must be unhashed */ 868 if (!d_unhashed(dn)) 869 d_drop(dn); 870 realdn = d_materialise_unique(dn, in); 871 if (IS_ERR(realdn)) { 872 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n", 873 PTR_ERR(realdn), dn, in, ceph_vinop(in)); 874 if (prehash) 875 *prehash = false; /* don't rehash on error */ 876 dn = realdn; /* note realdn contains the error */ 877 goto out; 878 } else if (realdn) { 879 dout("dn %p (%d) spliced with %p (%d) " 880 "inode %p ino %llx.%llx\n", 881 dn, atomic_read(&dn->d_count), 882 realdn, atomic_read(&realdn->d_count), 883 realdn->d_inode, ceph_vinop(realdn->d_inode)); 884 dput(dn); 885 dn = realdn; 886 } else { 887 BUG_ON(!ceph_dentry(dn)); 888 dout("dn %p attached to %p ino %llx.%llx\n", 889 dn, dn->d_inode, ceph_vinop(dn->d_inode)); 890 } 891 if ((!prehash || *prehash) && d_unhashed(dn)) 892 d_rehash(dn); 893 if (set_offset) 894 ceph_set_dentry_offset(dn); 895 out: 896 return dn; 897 } 898 899 /* 900 * Incorporate results into the local cache. This is either just 901 * one inode, or a directory, dentry, and possibly linked-to inode (e.g., 902 * after a lookup). 903 * 904 * A reply may contain 905 * a directory inode along with a dentry. 906 * and/or a target inode 907 * 908 * Called with snap_rwsem (read). 909 */ 910 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req, 911 struct ceph_mds_session *session) 912 { 913 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 914 struct inode *in = NULL; 915 struct ceph_mds_reply_inode *ininfo; 916 struct ceph_vino vino; 917 struct ceph_fs_client *fsc = ceph_sb_to_client(sb); 918 int i = 0; 919 int err = 0; 920 921 dout("fill_trace %p is_dentry %d is_target %d\n", req, 922 rinfo->head->is_dentry, rinfo->head->is_target); 923 924 #if 0 925 /* 926 * Debugging hook: 927 * 928 * If we resend completed ops to a recovering mds, we get no 929 * trace. Since that is very rare, pretend this is the case 930 * to ensure the 'no trace' handlers in the callers behave. 931 * 932 * Fill in inodes unconditionally to avoid breaking cap 933 * invariants. 934 */ 935 if (rinfo->head->op & CEPH_MDS_OP_WRITE) { 936 pr_info("fill_trace faking empty trace on %lld %s\n", 937 req->r_tid, ceph_mds_op_name(rinfo->head->op)); 938 if (rinfo->head->is_dentry) { 939 rinfo->head->is_dentry = 0; 940 err = fill_inode(req->r_locked_dir, 941 &rinfo->diri, rinfo->dirfrag, 942 session, req->r_request_started, -1); 943 } 944 if (rinfo->head->is_target) { 945 rinfo->head->is_target = 0; 946 ininfo = rinfo->targeti.in; 947 vino.ino = le64_to_cpu(ininfo->ino); 948 vino.snap = le64_to_cpu(ininfo->snapid); 949 in = ceph_get_inode(sb, vino); 950 err = fill_inode(in, &rinfo->targeti, NULL, 951 session, req->r_request_started, 952 req->r_fmode); 953 iput(in); 954 } 955 } 956 #endif 957 958 if (!rinfo->head->is_target && !rinfo->head->is_dentry) { 959 dout("fill_trace reply is empty!\n"); 960 if (rinfo->head->result == 0 && req->r_locked_dir) 961 ceph_invalidate_dir_request(req); 962 return 0; 963 } 964 965 if (rinfo->head->is_dentry) { 966 struct inode *dir = req->r_locked_dir; 967 968 err = fill_inode(dir, &rinfo->diri, rinfo->dirfrag, 969 session, req->r_request_started, -1, 970 &req->r_caps_reservation); 971 if (err < 0) 972 return err; 973 } 974 975 /* 976 * ignore null lease/binding on snapdir ENOENT, or else we 977 * will have trouble splicing in the virtual snapdir later 978 */ 979 if (rinfo->head->is_dentry && !req->r_aborted && 980 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name, 981 fsc->mount_options->snapdir_name, 982 req->r_dentry->d_name.len))) { 983 /* 984 * lookup link rename : null -> possibly existing inode 985 * mknod symlink mkdir : null -> new inode 986 * unlink : linked -> null 987 */ 988 struct inode *dir = req->r_locked_dir; 989 struct dentry *dn = req->r_dentry; 990 bool have_dir_cap, have_lease; 991 992 BUG_ON(!dn); 993 BUG_ON(!dir); 994 BUG_ON(dn->d_parent->d_inode != dir); 995 BUG_ON(ceph_ino(dir) != 996 le64_to_cpu(rinfo->diri.in->ino)); 997 BUG_ON(ceph_snap(dir) != 998 le64_to_cpu(rinfo->diri.in->snapid)); 999 1000 /* do we have a lease on the whole dir? */ 1001 have_dir_cap = 1002 (le32_to_cpu(rinfo->diri.in->cap.caps) & 1003 CEPH_CAP_FILE_SHARED); 1004 1005 /* do we have a dn lease? */ 1006 have_lease = have_dir_cap || 1007 (le16_to_cpu(rinfo->dlease->mask) & 1008 CEPH_LOCK_DN); 1009 1010 if (!have_lease) 1011 dout("fill_trace no dentry lease or dir cap\n"); 1012 1013 /* rename? */ 1014 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) { 1015 dout(" src %p '%.*s' dst %p '%.*s'\n", 1016 req->r_old_dentry, 1017 req->r_old_dentry->d_name.len, 1018 req->r_old_dentry->d_name.name, 1019 dn, dn->d_name.len, dn->d_name.name); 1020 dout("fill_trace doing d_move %p -> %p\n", 1021 req->r_old_dentry, dn); 1022 1023 /* d_move screws up d_subdirs order */ 1024 ceph_i_clear(dir, CEPH_I_COMPLETE); 1025 1026 d_move(req->r_old_dentry, dn); 1027 dout(" src %p '%.*s' dst %p '%.*s'\n", 1028 req->r_old_dentry, 1029 req->r_old_dentry->d_name.len, 1030 req->r_old_dentry->d_name.name, 1031 dn, dn->d_name.len, dn->d_name.name); 1032 1033 /* ensure target dentry is invalidated, despite 1034 rehashing bug in vfs_rename_dir */ 1035 ceph_invalidate_dentry_lease(dn); 1036 1037 /* take overwritten dentry's readdir offset */ 1038 dout("dn %p gets %p offset %lld (old offset %lld)\n", 1039 req->r_old_dentry, dn, ceph_dentry(dn)->offset, 1040 ceph_dentry(req->r_old_dentry)->offset); 1041 ceph_dentry(req->r_old_dentry)->offset = 1042 ceph_dentry(dn)->offset; 1043 1044 dn = req->r_old_dentry; /* use old_dentry */ 1045 in = dn->d_inode; 1046 } 1047 1048 /* null dentry? */ 1049 if (!rinfo->head->is_target) { 1050 dout("fill_trace null dentry\n"); 1051 if (dn->d_inode) { 1052 dout("d_delete %p\n", dn); 1053 d_delete(dn); 1054 } else { 1055 dout("d_instantiate %p NULL\n", dn); 1056 d_instantiate(dn, NULL); 1057 if (have_lease && d_unhashed(dn)) 1058 d_rehash(dn); 1059 update_dentry_lease(dn, rinfo->dlease, 1060 session, 1061 req->r_request_started); 1062 } 1063 goto done; 1064 } 1065 1066 /* attach proper inode */ 1067 ininfo = rinfo->targeti.in; 1068 vino.ino = le64_to_cpu(ininfo->ino); 1069 vino.snap = le64_to_cpu(ininfo->snapid); 1070 in = dn->d_inode; 1071 if (!in) { 1072 in = ceph_get_inode(sb, vino); 1073 if (IS_ERR(in)) { 1074 pr_err("fill_trace bad get_inode " 1075 "%llx.%llx\n", vino.ino, vino.snap); 1076 err = PTR_ERR(in); 1077 d_delete(dn); 1078 goto done; 1079 } 1080 dn = splice_dentry(dn, in, &have_lease, true); 1081 if (IS_ERR(dn)) { 1082 err = PTR_ERR(dn); 1083 goto done; 1084 } 1085 req->r_dentry = dn; /* may have spliced */ 1086 igrab(in); 1087 } else if (ceph_ino(in) == vino.ino && 1088 ceph_snap(in) == vino.snap) { 1089 igrab(in); 1090 } else { 1091 dout(" %p links to %p %llx.%llx, not %llx.%llx\n", 1092 dn, in, ceph_ino(in), ceph_snap(in), 1093 vino.ino, vino.snap); 1094 have_lease = false; 1095 in = NULL; 1096 } 1097 1098 if (have_lease) 1099 update_dentry_lease(dn, rinfo->dlease, session, 1100 req->r_request_started); 1101 dout(" final dn %p\n", dn); 1102 i++; 1103 } else if (req->r_op == CEPH_MDS_OP_LOOKUPSNAP || 1104 req->r_op == CEPH_MDS_OP_MKSNAP) { 1105 struct dentry *dn = req->r_dentry; 1106 1107 /* fill out a snapdir LOOKUPSNAP dentry */ 1108 BUG_ON(!dn); 1109 BUG_ON(!req->r_locked_dir); 1110 BUG_ON(ceph_snap(req->r_locked_dir) != CEPH_SNAPDIR); 1111 ininfo = rinfo->targeti.in; 1112 vino.ino = le64_to_cpu(ininfo->ino); 1113 vino.snap = le64_to_cpu(ininfo->snapid); 1114 in = ceph_get_inode(sb, vino); 1115 if (IS_ERR(in)) { 1116 pr_err("fill_inode get_inode badness %llx.%llx\n", 1117 vino.ino, vino.snap); 1118 err = PTR_ERR(in); 1119 d_delete(dn); 1120 goto done; 1121 } 1122 dout(" linking snapped dir %p to dn %p\n", in, dn); 1123 dn = splice_dentry(dn, in, NULL, true); 1124 if (IS_ERR(dn)) { 1125 err = PTR_ERR(dn); 1126 goto done; 1127 } 1128 req->r_dentry = dn; /* may have spliced */ 1129 igrab(in); 1130 rinfo->head->is_dentry = 1; /* fool notrace handlers */ 1131 } 1132 1133 if (rinfo->head->is_target) { 1134 vino.ino = le64_to_cpu(rinfo->targeti.in->ino); 1135 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid); 1136 1137 if (in == NULL || ceph_ino(in) != vino.ino || 1138 ceph_snap(in) != vino.snap) { 1139 in = ceph_get_inode(sb, vino); 1140 if (IS_ERR(in)) { 1141 err = PTR_ERR(in); 1142 goto done; 1143 } 1144 } 1145 req->r_target_inode = in; 1146 1147 err = fill_inode(in, 1148 &rinfo->targeti, NULL, 1149 session, req->r_request_started, 1150 (le32_to_cpu(rinfo->head->result) == 0) ? 1151 req->r_fmode : -1, 1152 &req->r_caps_reservation); 1153 if (err < 0) { 1154 pr_err("fill_inode badness %p %llx.%llx\n", 1155 in, ceph_vinop(in)); 1156 goto done; 1157 } 1158 } 1159 1160 done: 1161 dout("fill_trace done err=%d\n", err); 1162 return err; 1163 } 1164 1165 /* 1166 * Prepopulate our cache with readdir results, leases, etc. 1167 */ 1168 int ceph_readdir_prepopulate(struct ceph_mds_request *req, 1169 struct ceph_mds_session *session) 1170 { 1171 struct dentry *parent = req->r_dentry; 1172 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1173 struct qstr dname; 1174 struct dentry *dn; 1175 struct inode *in; 1176 int err = 0, i; 1177 struct inode *snapdir = NULL; 1178 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base; 1179 u64 frag = le32_to_cpu(rhead->args.readdir.frag); 1180 struct ceph_dentry_info *di; 1181 1182 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) { 1183 snapdir = ceph_get_snapdir(parent->d_inode); 1184 parent = d_find_alias(snapdir); 1185 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n", 1186 rinfo->dir_nr, parent); 1187 } else { 1188 dout("readdir_prepopulate %d items under dn %p\n", 1189 rinfo->dir_nr, parent); 1190 if (rinfo->dir_dir) 1191 ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir); 1192 } 1193 1194 for (i = 0; i < rinfo->dir_nr; i++) { 1195 struct ceph_vino vino; 1196 1197 dname.name = rinfo->dir_dname[i]; 1198 dname.len = rinfo->dir_dname_len[i]; 1199 dname.hash = full_name_hash(dname.name, dname.len); 1200 1201 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino); 1202 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid); 1203 1204 retry_lookup: 1205 dn = d_lookup(parent, &dname); 1206 dout("d_lookup on parent=%p name=%.*s got %p\n", 1207 parent, dname.len, dname.name, dn); 1208 1209 if (!dn) { 1210 dn = d_alloc(parent, &dname); 1211 dout("d_alloc %p '%.*s' = %p\n", parent, 1212 dname.len, dname.name, dn); 1213 if (dn == NULL) { 1214 dout("d_alloc badness\n"); 1215 err = -ENOMEM; 1216 goto out; 1217 } 1218 err = ceph_init_dentry(dn); 1219 if (err < 0) { 1220 dput(dn); 1221 goto out; 1222 } 1223 } else if (dn->d_inode && 1224 (ceph_ino(dn->d_inode) != vino.ino || 1225 ceph_snap(dn->d_inode) != vino.snap)) { 1226 dout(" dn %p points to wrong inode %p\n", 1227 dn, dn->d_inode); 1228 d_delete(dn); 1229 dput(dn); 1230 goto retry_lookup; 1231 } else { 1232 /* reorder parent's d_subdirs */ 1233 spin_lock(&dcache_lock); 1234 spin_lock(&dn->d_lock); 1235 list_move(&dn->d_u.d_child, &parent->d_subdirs); 1236 spin_unlock(&dn->d_lock); 1237 spin_unlock(&dcache_lock); 1238 } 1239 1240 di = dn->d_fsdata; 1241 di->offset = ceph_make_fpos(frag, i + req->r_readdir_offset); 1242 1243 /* inode */ 1244 if (dn->d_inode) { 1245 in = dn->d_inode; 1246 } else { 1247 in = ceph_get_inode(parent->d_sb, vino); 1248 if (IS_ERR(in)) { 1249 dout("new_inode badness\n"); 1250 d_delete(dn); 1251 dput(dn); 1252 err = PTR_ERR(in); 1253 goto out; 1254 } 1255 dn = splice_dentry(dn, in, NULL, false); 1256 if (IS_ERR(dn)) 1257 dn = NULL; 1258 } 1259 1260 if (fill_inode(in, &rinfo->dir_in[i], NULL, session, 1261 req->r_request_started, -1, 1262 &req->r_caps_reservation) < 0) { 1263 pr_err("fill_inode badness on %p\n", in); 1264 goto next_item; 1265 } 1266 if (dn) 1267 update_dentry_lease(dn, rinfo->dir_dlease[i], 1268 req->r_session, 1269 req->r_request_started); 1270 next_item: 1271 if (dn) 1272 dput(dn); 1273 } 1274 req->r_did_prepopulate = true; 1275 1276 out: 1277 if (snapdir) { 1278 iput(snapdir); 1279 dput(parent); 1280 } 1281 dout("readdir_prepopulate done\n"); 1282 return err; 1283 } 1284 1285 int ceph_inode_set_size(struct inode *inode, loff_t size) 1286 { 1287 struct ceph_inode_info *ci = ceph_inode(inode); 1288 int ret = 0; 1289 1290 spin_lock(&inode->i_lock); 1291 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size); 1292 inode->i_size = size; 1293 inode->i_blocks = (size + (1 << 9) - 1) >> 9; 1294 1295 /* tell the MDS if we are approaching max_size */ 1296 if ((size << 1) >= ci->i_max_size && 1297 (ci->i_reported_size << 1) < ci->i_max_size) 1298 ret = 1; 1299 1300 spin_unlock(&inode->i_lock); 1301 return ret; 1302 } 1303 1304 /* 1305 * Write back inode data in a worker thread. (This can't be done 1306 * in the message handler context.) 1307 */ 1308 void ceph_queue_writeback(struct inode *inode) 1309 { 1310 if (queue_work(ceph_inode_to_client(inode)->wb_wq, 1311 &ceph_inode(inode)->i_wb_work)) { 1312 dout("ceph_queue_writeback %p\n", inode); 1313 igrab(inode); 1314 } else { 1315 dout("ceph_queue_writeback %p failed\n", inode); 1316 } 1317 } 1318 1319 static void ceph_writeback_work(struct work_struct *work) 1320 { 1321 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1322 i_wb_work); 1323 struct inode *inode = &ci->vfs_inode; 1324 1325 dout("writeback %p\n", inode); 1326 filemap_fdatawrite(&inode->i_data); 1327 iput(inode); 1328 } 1329 1330 /* 1331 * queue an async invalidation 1332 */ 1333 void ceph_queue_invalidate(struct inode *inode) 1334 { 1335 if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq, 1336 &ceph_inode(inode)->i_pg_inv_work)) { 1337 dout("ceph_queue_invalidate %p\n", inode); 1338 igrab(inode); 1339 } else { 1340 dout("ceph_queue_invalidate %p failed\n", inode); 1341 } 1342 } 1343 1344 /* 1345 * invalidate any pages that are not dirty or under writeback. this 1346 * includes pages that are clean and mapped. 1347 */ 1348 static void ceph_invalidate_nondirty_pages(struct address_space *mapping) 1349 { 1350 struct pagevec pvec; 1351 pgoff_t next = 0; 1352 int i; 1353 1354 pagevec_init(&pvec, 0); 1355 while (pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { 1356 for (i = 0; i < pagevec_count(&pvec); i++) { 1357 struct page *page = pvec.pages[i]; 1358 pgoff_t index; 1359 int skip_page = 1360 (PageDirty(page) || PageWriteback(page)); 1361 1362 if (!skip_page) 1363 skip_page = !trylock_page(page); 1364 1365 /* 1366 * We really shouldn't be looking at the ->index of an 1367 * unlocked page. But we're not allowed to lock these 1368 * pages. So we rely upon nobody altering the ->index 1369 * of this (pinned-by-us) page. 1370 */ 1371 index = page->index; 1372 if (index > next) 1373 next = index; 1374 next++; 1375 1376 if (skip_page) 1377 continue; 1378 1379 generic_error_remove_page(mapping, page); 1380 unlock_page(page); 1381 } 1382 pagevec_release(&pvec); 1383 cond_resched(); 1384 } 1385 } 1386 1387 /* 1388 * Invalidate inode pages in a worker thread. (This can't be done 1389 * in the message handler context.) 1390 */ 1391 static void ceph_invalidate_work(struct work_struct *work) 1392 { 1393 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1394 i_pg_inv_work); 1395 struct inode *inode = &ci->vfs_inode; 1396 u32 orig_gen; 1397 int check = 0; 1398 1399 spin_lock(&inode->i_lock); 1400 dout("invalidate_pages %p gen %d revoking %d\n", inode, 1401 ci->i_rdcache_gen, ci->i_rdcache_revoking); 1402 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) { 1403 /* nevermind! */ 1404 spin_unlock(&inode->i_lock); 1405 goto out; 1406 } 1407 orig_gen = ci->i_rdcache_gen; 1408 spin_unlock(&inode->i_lock); 1409 1410 ceph_invalidate_nondirty_pages(inode->i_mapping); 1411 1412 spin_lock(&inode->i_lock); 1413 if (orig_gen == ci->i_rdcache_gen && 1414 orig_gen == ci->i_rdcache_revoking) { 1415 dout("invalidate_pages %p gen %d successful\n", inode, 1416 ci->i_rdcache_gen); 1417 ci->i_rdcache_revoking--; 1418 check = 1; 1419 } else { 1420 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n", 1421 inode, orig_gen, ci->i_rdcache_gen, 1422 ci->i_rdcache_revoking); 1423 } 1424 spin_unlock(&inode->i_lock); 1425 1426 if (check) 1427 ceph_check_caps(ci, 0, NULL); 1428 out: 1429 iput(inode); 1430 } 1431 1432 1433 /* 1434 * called by trunc_wq; take i_mutex ourselves 1435 * 1436 * We also truncate in a separate thread as well. 1437 */ 1438 static void ceph_vmtruncate_work(struct work_struct *work) 1439 { 1440 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1441 i_vmtruncate_work); 1442 struct inode *inode = &ci->vfs_inode; 1443 1444 dout("vmtruncate_work %p\n", inode); 1445 mutex_lock(&inode->i_mutex); 1446 __ceph_do_pending_vmtruncate(inode); 1447 mutex_unlock(&inode->i_mutex); 1448 iput(inode); 1449 } 1450 1451 /* 1452 * Queue an async vmtruncate. If we fail to queue work, we will handle 1453 * the truncation the next time we call __ceph_do_pending_vmtruncate. 1454 */ 1455 void ceph_queue_vmtruncate(struct inode *inode) 1456 { 1457 struct ceph_inode_info *ci = ceph_inode(inode); 1458 1459 if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq, 1460 &ci->i_vmtruncate_work)) { 1461 dout("ceph_queue_vmtruncate %p\n", inode); 1462 igrab(inode); 1463 } else { 1464 dout("ceph_queue_vmtruncate %p failed, pending=%d\n", 1465 inode, ci->i_truncate_pending); 1466 } 1467 } 1468 1469 /* 1470 * called with i_mutex held. 1471 * 1472 * Make sure any pending truncation is applied before doing anything 1473 * that may depend on it. 1474 */ 1475 void __ceph_do_pending_vmtruncate(struct inode *inode) 1476 { 1477 struct ceph_inode_info *ci = ceph_inode(inode); 1478 u64 to; 1479 int wrbuffer_refs, wake = 0; 1480 1481 retry: 1482 spin_lock(&inode->i_lock); 1483 if (ci->i_truncate_pending == 0) { 1484 dout("__do_pending_vmtruncate %p none pending\n", inode); 1485 spin_unlock(&inode->i_lock); 1486 return; 1487 } 1488 1489 /* 1490 * make sure any dirty snapped pages are flushed before we 1491 * possibly truncate them.. so write AND block! 1492 */ 1493 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) { 1494 dout("__do_pending_vmtruncate %p flushing snaps first\n", 1495 inode); 1496 spin_unlock(&inode->i_lock); 1497 filemap_write_and_wait_range(&inode->i_data, 0, 1498 inode->i_sb->s_maxbytes); 1499 goto retry; 1500 } 1501 1502 to = ci->i_truncate_size; 1503 wrbuffer_refs = ci->i_wrbuffer_ref; 1504 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode, 1505 ci->i_truncate_pending, to); 1506 spin_unlock(&inode->i_lock); 1507 1508 truncate_inode_pages(inode->i_mapping, to); 1509 1510 spin_lock(&inode->i_lock); 1511 ci->i_truncate_pending--; 1512 if (ci->i_truncate_pending == 0) 1513 wake = 1; 1514 spin_unlock(&inode->i_lock); 1515 1516 if (wrbuffer_refs == 0) 1517 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); 1518 if (wake) 1519 wake_up_all(&ci->i_cap_wq); 1520 } 1521 1522 1523 /* 1524 * symlinks 1525 */ 1526 static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd) 1527 { 1528 struct ceph_inode_info *ci = ceph_inode(dentry->d_inode); 1529 nd_set_link(nd, ci->i_symlink); 1530 return NULL; 1531 } 1532 1533 static const struct inode_operations ceph_symlink_iops = { 1534 .readlink = generic_readlink, 1535 .follow_link = ceph_sym_follow_link, 1536 }; 1537 1538 /* 1539 * setattr 1540 */ 1541 int ceph_setattr(struct dentry *dentry, struct iattr *attr) 1542 { 1543 struct inode *inode = dentry->d_inode; 1544 struct ceph_inode_info *ci = ceph_inode(inode); 1545 struct inode *parent_inode = dentry->d_parent->d_inode; 1546 const unsigned int ia_valid = attr->ia_valid; 1547 struct ceph_mds_request *req; 1548 struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc; 1549 int issued; 1550 int release = 0, dirtied = 0; 1551 int mask = 0; 1552 int err = 0; 1553 1554 if (ceph_snap(inode) != CEPH_NOSNAP) 1555 return -EROFS; 1556 1557 __ceph_do_pending_vmtruncate(inode); 1558 1559 err = inode_change_ok(inode, attr); 1560 if (err != 0) 1561 return err; 1562 1563 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR, 1564 USE_AUTH_MDS); 1565 if (IS_ERR(req)) 1566 return PTR_ERR(req); 1567 1568 spin_lock(&inode->i_lock); 1569 issued = __ceph_caps_issued(ci, NULL); 1570 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued)); 1571 1572 if (ia_valid & ATTR_UID) { 1573 dout("setattr %p uid %d -> %d\n", inode, 1574 inode->i_uid, attr->ia_uid); 1575 if (issued & CEPH_CAP_AUTH_EXCL) { 1576 inode->i_uid = attr->ia_uid; 1577 dirtied |= CEPH_CAP_AUTH_EXCL; 1578 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1579 attr->ia_uid != inode->i_uid) { 1580 req->r_args.setattr.uid = cpu_to_le32(attr->ia_uid); 1581 mask |= CEPH_SETATTR_UID; 1582 release |= CEPH_CAP_AUTH_SHARED; 1583 } 1584 } 1585 if (ia_valid & ATTR_GID) { 1586 dout("setattr %p gid %d -> %d\n", inode, 1587 inode->i_gid, attr->ia_gid); 1588 if (issued & CEPH_CAP_AUTH_EXCL) { 1589 inode->i_gid = attr->ia_gid; 1590 dirtied |= CEPH_CAP_AUTH_EXCL; 1591 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1592 attr->ia_gid != inode->i_gid) { 1593 req->r_args.setattr.gid = cpu_to_le32(attr->ia_gid); 1594 mask |= CEPH_SETATTR_GID; 1595 release |= CEPH_CAP_AUTH_SHARED; 1596 } 1597 } 1598 if (ia_valid & ATTR_MODE) { 1599 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode, 1600 attr->ia_mode); 1601 if (issued & CEPH_CAP_AUTH_EXCL) { 1602 inode->i_mode = attr->ia_mode; 1603 dirtied |= CEPH_CAP_AUTH_EXCL; 1604 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1605 attr->ia_mode != inode->i_mode) { 1606 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode); 1607 mask |= CEPH_SETATTR_MODE; 1608 release |= CEPH_CAP_AUTH_SHARED; 1609 } 1610 } 1611 1612 if (ia_valid & ATTR_ATIME) { 1613 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode, 1614 inode->i_atime.tv_sec, inode->i_atime.tv_nsec, 1615 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec); 1616 if (issued & CEPH_CAP_FILE_EXCL) { 1617 ci->i_time_warp_seq++; 1618 inode->i_atime = attr->ia_atime; 1619 dirtied |= CEPH_CAP_FILE_EXCL; 1620 } else if ((issued & CEPH_CAP_FILE_WR) && 1621 timespec_compare(&inode->i_atime, 1622 &attr->ia_atime) < 0) { 1623 inode->i_atime = attr->ia_atime; 1624 dirtied |= CEPH_CAP_FILE_WR; 1625 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1626 !timespec_equal(&inode->i_atime, &attr->ia_atime)) { 1627 ceph_encode_timespec(&req->r_args.setattr.atime, 1628 &attr->ia_atime); 1629 mask |= CEPH_SETATTR_ATIME; 1630 release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD | 1631 CEPH_CAP_FILE_WR; 1632 } 1633 } 1634 if (ia_valid & ATTR_MTIME) { 1635 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode, 1636 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 1637 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec); 1638 if (issued & CEPH_CAP_FILE_EXCL) { 1639 ci->i_time_warp_seq++; 1640 inode->i_mtime = attr->ia_mtime; 1641 dirtied |= CEPH_CAP_FILE_EXCL; 1642 } else if ((issued & CEPH_CAP_FILE_WR) && 1643 timespec_compare(&inode->i_mtime, 1644 &attr->ia_mtime) < 0) { 1645 inode->i_mtime = attr->ia_mtime; 1646 dirtied |= CEPH_CAP_FILE_WR; 1647 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1648 !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) { 1649 ceph_encode_timespec(&req->r_args.setattr.mtime, 1650 &attr->ia_mtime); 1651 mask |= CEPH_SETATTR_MTIME; 1652 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | 1653 CEPH_CAP_FILE_WR; 1654 } 1655 } 1656 if (ia_valid & ATTR_SIZE) { 1657 dout("setattr %p size %lld -> %lld\n", inode, 1658 inode->i_size, attr->ia_size); 1659 if (attr->ia_size > inode->i_sb->s_maxbytes) { 1660 err = -EINVAL; 1661 goto out; 1662 } 1663 if ((issued & CEPH_CAP_FILE_EXCL) && 1664 attr->ia_size > inode->i_size) { 1665 inode->i_size = attr->ia_size; 1666 inode->i_blocks = 1667 (attr->ia_size + (1 << 9) - 1) >> 9; 1668 inode->i_ctime = attr->ia_ctime; 1669 ci->i_reported_size = attr->ia_size; 1670 dirtied |= CEPH_CAP_FILE_EXCL; 1671 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1672 attr->ia_size != inode->i_size) { 1673 req->r_args.setattr.size = cpu_to_le64(attr->ia_size); 1674 req->r_args.setattr.old_size = 1675 cpu_to_le64(inode->i_size); 1676 mask |= CEPH_SETATTR_SIZE; 1677 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | 1678 CEPH_CAP_FILE_WR; 1679 } 1680 } 1681 1682 /* these do nothing */ 1683 if (ia_valid & ATTR_CTIME) { 1684 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME| 1685 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0; 1686 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode, 1687 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 1688 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec, 1689 only ? "ctime only" : "ignored"); 1690 inode->i_ctime = attr->ia_ctime; 1691 if (only) { 1692 /* 1693 * if kernel wants to dirty ctime but nothing else, 1694 * we need to choose a cap to dirty under, or do 1695 * a almost-no-op setattr 1696 */ 1697 if (issued & CEPH_CAP_AUTH_EXCL) 1698 dirtied |= CEPH_CAP_AUTH_EXCL; 1699 else if (issued & CEPH_CAP_FILE_EXCL) 1700 dirtied |= CEPH_CAP_FILE_EXCL; 1701 else if (issued & CEPH_CAP_XATTR_EXCL) 1702 dirtied |= CEPH_CAP_XATTR_EXCL; 1703 else 1704 mask |= CEPH_SETATTR_CTIME; 1705 } 1706 } 1707 if (ia_valid & ATTR_FILE) 1708 dout("setattr %p ATTR_FILE ... hrm!\n", inode); 1709 1710 if (dirtied) { 1711 __ceph_mark_dirty_caps(ci, dirtied); 1712 inode->i_ctime = CURRENT_TIME; 1713 } 1714 1715 release &= issued; 1716 spin_unlock(&inode->i_lock); 1717 1718 if (mask) { 1719 req->r_inode = igrab(inode); 1720 req->r_inode_drop = release; 1721 req->r_args.setattr.mask = cpu_to_le32(mask); 1722 req->r_num_caps = 1; 1723 err = ceph_mdsc_do_request(mdsc, parent_inode, req); 1724 } 1725 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err, 1726 ceph_cap_string(dirtied), mask); 1727 1728 ceph_mdsc_put_request(req); 1729 __ceph_do_pending_vmtruncate(inode); 1730 return err; 1731 out: 1732 spin_unlock(&inode->i_lock); 1733 ceph_mdsc_put_request(req); 1734 return err; 1735 } 1736 1737 /* 1738 * Verify that we have a lease on the given mask. If not, 1739 * do a getattr against an mds. 1740 */ 1741 int ceph_do_getattr(struct inode *inode, int mask) 1742 { 1743 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); 1744 struct ceph_mds_client *mdsc = fsc->mdsc; 1745 struct ceph_mds_request *req; 1746 int err; 1747 1748 if (ceph_snap(inode) == CEPH_SNAPDIR) { 1749 dout("do_getattr inode %p SNAPDIR\n", inode); 1750 return 0; 1751 } 1752 1753 dout("do_getattr inode %p mask %s\n", inode, ceph_cap_string(mask)); 1754 if (ceph_caps_issued_mask(ceph_inode(inode), mask, 1)) 1755 return 0; 1756 1757 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS); 1758 if (IS_ERR(req)) 1759 return PTR_ERR(req); 1760 req->r_inode = igrab(inode); 1761 req->r_num_caps = 1; 1762 req->r_args.getattr.mask = cpu_to_le32(mask); 1763 err = ceph_mdsc_do_request(mdsc, NULL, req); 1764 ceph_mdsc_put_request(req); 1765 dout("do_getattr result=%d\n", err); 1766 return err; 1767 } 1768 1769 1770 /* 1771 * Check inode permissions. We verify we have a valid value for 1772 * the AUTH cap, then call the generic handler. 1773 */ 1774 int ceph_permission(struct inode *inode, int mask) 1775 { 1776 int err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED); 1777 1778 if (!err) 1779 err = generic_permission(inode, mask, NULL); 1780 return err; 1781 } 1782 1783 /* 1784 * Get all attributes. Hopefully somedata we'll have a statlite() 1785 * and can limit the fields we require to be accurate. 1786 */ 1787 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry, 1788 struct kstat *stat) 1789 { 1790 struct inode *inode = dentry->d_inode; 1791 struct ceph_inode_info *ci = ceph_inode(inode); 1792 int err; 1793 1794 err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL); 1795 if (!err) { 1796 generic_fillattr(inode, stat); 1797 stat->ino = inode->i_ino; 1798 if (ceph_snap(inode) != CEPH_NOSNAP) 1799 stat->dev = ceph_snap(inode); 1800 else 1801 stat->dev = 0; 1802 if (S_ISDIR(inode->i_mode)) { 1803 stat->size = ci->i_rbytes; 1804 stat->blocks = 0; 1805 stat->blksize = 65536; 1806 } 1807 } 1808 return err; 1809 } 1810