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