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