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 ERR_PTR(PTR_ERR(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 mapping_mapped(inode->i_mapping) || 448 __ceph_caps_file_wanted(ci)) { 449 ci->i_truncate_pending++; 450 queue_trunc = 1; 451 } 452 } 453 } 454 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 && 455 ci->i_truncate_size != truncate_size) { 456 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size, 457 truncate_size); 458 ci->i_truncate_size = truncate_size; 459 } 460 return queue_trunc; 461 } 462 463 void ceph_fill_file_time(struct inode *inode, int issued, 464 u64 time_warp_seq, struct timespec *ctime, 465 struct timespec *mtime, struct timespec *atime) 466 { 467 struct ceph_inode_info *ci = ceph_inode(inode); 468 int warn = 0; 469 470 if (issued & (CEPH_CAP_FILE_EXCL| 471 CEPH_CAP_FILE_WR| 472 CEPH_CAP_FILE_BUFFER)) { 473 if (timespec_compare(ctime, &inode->i_ctime) > 0) { 474 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n", 475 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 476 ctime->tv_sec, ctime->tv_nsec); 477 inode->i_ctime = *ctime; 478 } 479 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) { 480 /* the MDS did a utimes() */ 481 dout("mtime %ld.%09ld -> %ld.%09ld " 482 "tw %d -> %d\n", 483 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 484 mtime->tv_sec, mtime->tv_nsec, 485 ci->i_time_warp_seq, (int)time_warp_seq); 486 487 inode->i_mtime = *mtime; 488 inode->i_atime = *atime; 489 ci->i_time_warp_seq = time_warp_seq; 490 } else if (time_warp_seq == ci->i_time_warp_seq) { 491 /* nobody did utimes(); take the max */ 492 if (timespec_compare(mtime, &inode->i_mtime) > 0) { 493 dout("mtime %ld.%09ld -> %ld.%09ld inc\n", 494 inode->i_mtime.tv_sec, 495 inode->i_mtime.tv_nsec, 496 mtime->tv_sec, mtime->tv_nsec); 497 inode->i_mtime = *mtime; 498 } 499 if (timespec_compare(atime, &inode->i_atime) > 0) { 500 dout("atime %ld.%09ld -> %ld.%09ld inc\n", 501 inode->i_atime.tv_sec, 502 inode->i_atime.tv_nsec, 503 atime->tv_sec, atime->tv_nsec); 504 inode->i_atime = *atime; 505 } 506 } else if (issued & CEPH_CAP_FILE_EXCL) { 507 /* we did a utimes(); ignore mds values */ 508 } else { 509 warn = 1; 510 } 511 } else { 512 /* we have no write caps; whatever the MDS says is true */ 513 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) { 514 inode->i_ctime = *ctime; 515 inode->i_mtime = *mtime; 516 inode->i_atime = *atime; 517 ci->i_time_warp_seq = time_warp_seq; 518 } else { 519 warn = 1; 520 } 521 } 522 if (warn) /* time_warp_seq shouldn't go backwards */ 523 dout("%p mds time_warp_seq %llu < %u\n", 524 inode, time_warp_seq, ci->i_time_warp_seq); 525 } 526 527 /* 528 * Populate an inode based on info from mds. May be called on new or 529 * existing inodes. 530 */ 531 static int fill_inode(struct inode *inode, 532 struct ceph_mds_reply_info_in *iinfo, 533 struct ceph_mds_reply_dirfrag *dirinfo, 534 struct ceph_mds_session *session, 535 unsigned long ttl_from, int cap_fmode, 536 struct ceph_cap_reservation *caps_reservation) 537 { 538 struct ceph_mds_reply_inode *info = iinfo->in; 539 struct ceph_inode_info *ci = ceph_inode(inode); 540 int i; 541 int issued, implemented; 542 struct timespec mtime, atime, ctime; 543 u32 nsplits; 544 struct ceph_buffer *xattr_blob = NULL; 545 int err = 0; 546 int queue_trunc = 0; 547 548 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n", 549 inode, ceph_vinop(inode), le64_to_cpu(info->version), 550 ci->i_version); 551 552 /* 553 * prealloc xattr data, if it looks like we'll need it. only 554 * if len > 4 (meaning there are actually xattrs; the first 4 555 * bytes are the xattr count). 556 */ 557 if (iinfo->xattr_len > 4) { 558 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS); 559 if (!xattr_blob) 560 pr_err("fill_inode ENOMEM xattr blob %d bytes\n", 561 iinfo->xattr_len); 562 } 563 564 spin_lock(&inode->i_lock); 565 566 /* 567 * provided version will be odd if inode value is projected, 568 * even if stable. skip the update if we have a newer info 569 * (e.g., due to inode info racing form multiple MDSs), or if 570 * we are getting projected (unstable) inode info. 571 */ 572 if (le64_to_cpu(info->version) > 0 && 573 (ci->i_version & ~1) > le64_to_cpu(info->version)) 574 goto no_change; 575 576 issued = __ceph_caps_issued(ci, &implemented); 577 issued |= implemented | __ceph_caps_dirty(ci); 578 579 /* update inode */ 580 ci->i_version = le64_to_cpu(info->version); 581 inode->i_version++; 582 inode->i_rdev = le32_to_cpu(info->rdev); 583 584 if ((issued & CEPH_CAP_AUTH_EXCL) == 0) { 585 inode->i_mode = le32_to_cpu(info->mode); 586 inode->i_uid = le32_to_cpu(info->uid); 587 inode->i_gid = le32_to_cpu(info->gid); 588 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode, 589 inode->i_uid, inode->i_gid); 590 } 591 592 if ((issued & CEPH_CAP_LINK_EXCL) == 0) 593 inode->i_nlink = le32_to_cpu(info->nlink); 594 595 /* be careful with mtime, atime, size */ 596 ceph_decode_timespec(&atime, &info->atime); 597 ceph_decode_timespec(&mtime, &info->mtime); 598 ceph_decode_timespec(&ctime, &info->ctime); 599 queue_trunc = ceph_fill_file_size(inode, issued, 600 le32_to_cpu(info->truncate_seq), 601 le64_to_cpu(info->truncate_size), 602 le64_to_cpu(info->size)); 603 ceph_fill_file_time(inode, issued, 604 le32_to_cpu(info->time_warp_seq), 605 &ctime, &mtime, &atime); 606 607 ci->i_max_size = le64_to_cpu(info->max_size); 608 ci->i_layout = info->layout; 609 inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1; 610 611 /* xattrs */ 612 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */ 613 if ((issued & CEPH_CAP_XATTR_EXCL) == 0 && 614 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) { 615 if (ci->i_xattrs.blob) 616 ceph_buffer_put(ci->i_xattrs.blob); 617 ci->i_xattrs.blob = xattr_blob; 618 if (xattr_blob) 619 memcpy(ci->i_xattrs.blob->vec.iov_base, 620 iinfo->xattr_data, iinfo->xattr_len); 621 ci->i_xattrs.version = le64_to_cpu(info->xattr_version); 622 xattr_blob = NULL; 623 } 624 625 inode->i_mapping->a_ops = &ceph_aops; 626 inode->i_mapping->backing_dev_info = 627 &ceph_sb_to_client(inode->i_sb)->backing_dev_info; 628 629 switch (inode->i_mode & S_IFMT) { 630 case S_IFIFO: 631 case S_IFBLK: 632 case S_IFCHR: 633 case S_IFSOCK: 634 init_special_inode(inode, inode->i_mode, inode->i_rdev); 635 inode->i_op = &ceph_file_iops; 636 break; 637 case S_IFREG: 638 inode->i_op = &ceph_file_iops; 639 inode->i_fop = &ceph_file_fops; 640 break; 641 case S_IFLNK: 642 inode->i_op = &ceph_symlink_iops; 643 if (!ci->i_symlink) { 644 int symlen = iinfo->symlink_len; 645 char *sym; 646 647 BUG_ON(symlen != inode->i_size); 648 spin_unlock(&inode->i_lock); 649 650 err = -ENOMEM; 651 sym = kmalloc(symlen+1, GFP_NOFS); 652 if (!sym) 653 goto out; 654 memcpy(sym, iinfo->symlink, symlen); 655 sym[symlen] = 0; 656 657 spin_lock(&inode->i_lock); 658 if (!ci->i_symlink) 659 ci->i_symlink = sym; 660 else 661 kfree(sym); /* lost a race */ 662 } 663 break; 664 case S_IFDIR: 665 inode->i_op = &ceph_dir_iops; 666 inode->i_fop = &ceph_dir_fops; 667 668 ci->i_files = le64_to_cpu(info->files); 669 ci->i_subdirs = le64_to_cpu(info->subdirs); 670 ci->i_rbytes = le64_to_cpu(info->rbytes); 671 ci->i_rfiles = le64_to_cpu(info->rfiles); 672 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs); 673 ceph_decode_timespec(&ci->i_rctime, &info->rctime); 674 675 /* set dir completion flag? */ 676 if (ci->i_files == 0 && ci->i_subdirs == 0 && 677 ceph_snap(inode) == CEPH_NOSNAP && 678 (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED) && 679 (ci->i_ceph_flags & CEPH_I_COMPLETE) == 0) { 680 dout(" marking %p complete (empty)\n", inode); 681 ci->i_ceph_flags |= CEPH_I_COMPLETE; 682 ci->i_max_offset = 2; 683 } 684 685 /* it may be better to set st_size in getattr instead? */ 686 if (ceph_test_opt(ceph_sb_to_client(inode->i_sb), RBYTES)) 687 inode->i_size = ci->i_rbytes; 688 break; 689 default: 690 pr_err("fill_inode %llx.%llx BAD mode 0%o\n", 691 ceph_vinop(inode), inode->i_mode); 692 } 693 694 no_change: 695 spin_unlock(&inode->i_lock); 696 697 /* queue truncate if we saw i_size decrease */ 698 if (queue_trunc) 699 ceph_queue_vmtruncate(inode); 700 701 /* populate frag tree */ 702 /* FIXME: move me up, if/when version reflects fragtree changes */ 703 nsplits = le32_to_cpu(info->fragtree.nsplits); 704 mutex_lock(&ci->i_fragtree_mutex); 705 for (i = 0; i < nsplits; i++) { 706 u32 id = le32_to_cpu(info->fragtree.splits[i].frag); 707 struct ceph_inode_frag *frag = __get_or_create_frag(ci, id); 708 709 if (IS_ERR(frag)) 710 continue; 711 frag->split_by = le32_to_cpu(info->fragtree.splits[i].by); 712 dout(" frag %x split by %d\n", frag->frag, frag->split_by); 713 } 714 mutex_unlock(&ci->i_fragtree_mutex); 715 716 /* were we issued a capability? */ 717 if (info->cap.caps) { 718 if (ceph_snap(inode) == CEPH_NOSNAP) { 719 ceph_add_cap(inode, session, 720 le64_to_cpu(info->cap.cap_id), 721 cap_fmode, 722 le32_to_cpu(info->cap.caps), 723 le32_to_cpu(info->cap.wanted), 724 le32_to_cpu(info->cap.seq), 725 le32_to_cpu(info->cap.mseq), 726 le64_to_cpu(info->cap.realm), 727 info->cap.flags, 728 caps_reservation); 729 } else { 730 spin_lock(&inode->i_lock); 731 dout(" %p got snap_caps %s\n", inode, 732 ceph_cap_string(le32_to_cpu(info->cap.caps))); 733 ci->i_snap_caps |= le32_to_cpu(info->cap.caps); 734 if (cap_fmode >= 0) 735 __ceph_get_fmode(ci, cap_fmode); 736 spin_unlock(&inode->i_lock); 737 } 738 } else if (cap_fmode >= 0) { 739 pr_warning("mds issued no caps on %llx.%llx\n", 740 ceph_vinop(inode)); 741 __ceph_get_fmode(ci, cap_fmode); 742 } 743 744 /* update delegation info? */ 745 if (dirinfo) 746 ceph_fill_dirfrag(inode, dirinfo); 747 748 err = 0; 749 750 out: 751 if (xattr_blob) 752 ceph_buffer_put(xattr_blob); 753 return err; 754 } 755 756 /* 757 * caller should hold session s_mutex. 758 */ 759 static void update_dentry_lease(struct dentry *dentry, 760 struct ceph_mds_reply_lease *lease, 761 struct ceph_mds_session *session, 762 unsigned long from_time) 763 { 764 struct ceph_dentry_info *di = ceph_dentry(dentry); 765 long unsigned duration = le32_to_cpu(lease->duration_ms); 766 long unsigned ttl = from_time + (duration * HZ) / 1000; 767 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000; 768 struct inode *dir; 769 770 /* only track leases on regular dentries */ 771 if (dentry->d_op != &ceph_dentry_ops) 772 return; 773 774 spin_lock(&dentry->d_lock); 775 dout("update_dentry_lease %p mask %d duration %lu ms ttl %lu\n", 776 dentry, le16_to_cpu(lease->mask), duration, ttl); 777 778 /* make lease_rdcache_gen match directory */ 779 dir = dentry->d_parent->d_inode; 780 di->lease_shared_gen = ceph_inode(dir)->i_shared_gen; 781 782 if (lease->mask == 0) 783 goto out_unlock; 784 785 if (di->lease_gen == session->s_cap_gen && 786 time_before(ttl, dentry->d_time)) 787 goto out_unlock; /* we already have a newer lease. */ 788 789 if (di->lease_session && di->lease_session != session) 790 goto out_unlock; 791 792 ceph_dentry_lru_touch(dentry); 793 794 if (!di->lease_session) 795 di->lease_session = ceph_get_mds_session(session); 796 di->lease_gen = session->s_cap_gen; 797 di->lease_seq = le32_to_cpu(lease->seq); 798 di->lease_renew_after = half_ttl; 799 di->lease_renew_from = 0; 800 dentry->d_time = ttl; 801 out_unlock: 802 spin_unlock(&dentry->d_lock); 803 return; 804 } 805 806 /* 807 * splice a dentry to an inode. 808 * caller must hold directory i_mutex for this to be safe. 809 * 810 * we will only rehash the resulting dentry if @prehash is 811 * true; @prehash will be set to false (for the benefit of 812 * the caller) if we fail. 813 */ 814 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in, 815 bool *prehash) 816 { 817 struct dentry *realdn; 818 819 /* dn must be unhashed */ 820 if (!d_unhashed(dn)) 821 d_drop(dn); 822 realdn = d_materialise_unique(dn, in); 823 if (IS_ERR(realdn)) { 824 pr_err("splice_dentry error %p inode %p ino %llx.%llx\n", 825 dn, in, ceph_vinop(in)); 826 if (prehash) 827 *prehash = false; /* don't rehash on error */ 828 dn = realdn; /* note realdn contains the error */ 829 goto out; 830 } else if (realdn) { 831 dout("dn %p (%d) spliced with %p (%d) " 832 "inode %p ino %llx.%llx\n", 833 dn, atomic_read(&dn->d_count), 834 realdn, atomic_read(&realdn->d_count), 835 realdn->d_inode, ceph_vinop(realdn->d_inode)); 836 dput(dn); 837 dn = realdn; 838 } else { 839 BUG_ON(!ceph_dentry(dn)); 840 841 dout("dn %p attached to %p ino %llx.%llx\n", 842 dn, dn->d_inode, ceph_vinop(dn->d_inode)); 843 } 844 if ((!prehash || *prehash) && d_unhashed(dn)) 845 d_rehash(dn); 846 out: 847 return dn; 848 } 849 850 /* 851 * Set dentry's directory position based on the current dir's max, and 852 * order it in d_subdirs, so that dcache_readdir behaves. 853 */ 854 static void ceph_set_dentry_offset(struct dentry *dn) 855 { 856 struct dentry *dir = dn->d_parent; 857 struct inode *inode = dn->d_parent->d_inode; 858 struct ceph_dentry_info *di; 859 860 BUG_ON(!inode); 861 862 di = ceph_dentry(dn); 863 864 spin_lock(&inode->i_lock); 865 if ((ceph_inode(inode)->i_ceph_flags & CEPH_I_COMPLETE) == 0) { 866 spin_unlock(&inode->i_lock); 867 return; 868 } 869 di->offset = ceph_inode(inode)->i_max_offset++; 870 spin_unlock(&inode->i_lock); 871 872 spin_lock(&dcache_lock); 873 spin_lock(&dn->d_lock); 874 list_move_tail(&dir->d_subdirs, &dn->d_u.d_child); 875 dout("set_dentry_offset %p %lld (%p %p)\n", dn, di->offset, 876 dn->d_u.d_child.prev, dn->d_u.d_child.next); 877 spin_unlock(&dn->d_lock); 878 spin_unlock(&dcache_lock); 879 } 880 881 /* 882 * Incorporate results into the local cache. This is either just 883 * one inode, or a directory, dentry, and possibly linked-to inode (e.g., 884 * after a lookup). 885 * 886 * A reply may contain 887 * a directory inode along with a dentry. 888 * and/or a target inode 889 * 890 * Called with snap_rwsem (read). 891 */ 892 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req, 893 struct ceph_mds_session *session) 894 { 895 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 896 struct inode *in = NULL; 897 struct ceph_mds_reply_inode *ininfo; 898 struct ceph_vino vino; 899 struct ceph_client *client = ceph_sb_to_client(sb); 900 int i = 0; 901 int err = 0; 902 903 dout("fill_trace %p is_dentry %d is_target %d\n", req, 904 rinfo->head->is_dentry, rinfo->head->is_target); 905 906 #if 0 907 /* 908 * Debugging hook: 909 * 910 * If we resend completed ops to a recovering mds, we get no 911 * trace. Since that is very rare, pretend this is the case 912 * to ensure the 'no trace' handlers in the callers behave. 913 * 914 * Fill in inodes unconditionally to avoid breaking cap 915 * invariants. 916 */ 917 if (rinfo->head->op & CEPH_MDS_OP_WRITE) { 918 pr_info("fill_trace faking empty trace on %lld %s\n", 919 req->r_tid, ceph_mds_op_name(rinfo->head->op)); 920 if (rinfo->head->is_dentry) { 921 rinfo->head->is_dentry = 0; 922 err = fill_inode(req->r_locked_dir, 923 &rinfo->diri, rinfo->dirfrag, 924 session, req->r_request_started, -1); 925 } 926 if (rinfo->head->is_target) { 927 rinfo->head->is_target = 0; 928 ininfo = rinfo->targeti.in; 929 vino.ino = le64_to_cpu(ininfo->ino); 930 vino.snap = le64_to_cpu(ininfo->snapid); 931 in = ceph_get_inode(sb, vino); 932 err = fill_inode(in, &rinfo->targeti, NULL, 933 session, req->r_request_started, 934 req->r_fmode); 935 iput(in); 936 } 937 } 938 #endif 939 940 if (!rinfo->head->is_target && !rinfo->head->is_dentry) { 941 dout("fill_trace reply is empty!\n"); 942 if (rinfo->head->result == 0 && req->r_locked_dir) { 943 struct ceph_inode_info *ci = 944 ceph_inode(req->r_locked_dir); 945 dout(" clearing %p complete (empty trace)\n", 946 req->r_locked_dir); 947 spin_lock(&req->r_locked_dir->i_lock); 948 ci->i_ceph_flags &= ~CEPH_I_COMPLETE; 949 ci->i_release_count++; 950 spin_unlock(&req->r_locked_dir->i_lock); 951 952 if (req->r_dentry) 953 ceph_invalidate_dentry_lease(req->r_dentry); 954 if (req->r_old_dentry) 955 ceph_invalidate_dentry_lease(req->r_old_dentry); 956 } 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 client->mount_args->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 ceph_dentry(req->r_old_dentry)->offset = 1034 ceph_dentry(dn)->offset; 1035 1036 dn = req->r_old_dentry; /* use old_dentry */ 1037 in = dn->d_inode; 1038 } 1039 1040 /* null dentry? */ 1041 if (!rinfo->head->is_target) { 1042 dout("fill_trace null dentry\n"); 1043 if (dn->d_inode) { 1044 dout("d_delete %p\n", dn); 1045 d_delete(dn); 1046 } else { 1047 dout("d_instantiate %p NULL\n", dn); 1048 d_instantiate(dn, NULL); 1049 if (have_lease && d_unhashed(dn)) 1050 d_rehash(dn); 1051 update_dentry_lease(dn, rinfo->dlease, 1052 session, 1053 req->r_request_started); 1054 } 1055 goto done; 1056 } 1057 1058 /* attach proper inode */ 1059 ininfo = rinfo->targeti.in; 1060 vino.ino = le64_to_cpu(ininfo->ino); 1061 vino.snap = le64_to_cpu(ininfo->snapid); 1062 if (!dn->d_inode) { 1063 in = ceph_get_inode(sb, vino); 1064 if (IS_ERR(in)) { 1065 pr_err("fill_trace bad get_inode " 1066 "%llx.%llx\n", vino.ino, vino.snap); 1067 err = PTR_ERR(in); 1068 d_delete(dn); 1069 goto done; 1070 } 1071 dn = splice_dentry(dn, in, &have_lease); 1072 if (IS_ERR(dn)) { 1073 err = PTR_ERR(dn); 1074 goto done; 1075 } 1076 req->r_dentry = dn; /* may have spliced */ 1077 ceph_set_dentry_offset(dn); 1078 igrab(in); 1079 } else if (ceph_ino(in) == vino.ino && 1080 ceph_snap(in) == vino.snap) { 1081 igrab(in); 1082 } else { 1083 dout(" %p links to %p %llx.%llx, not %llx.%llx\n", 1084 dn, in, ceph_ino(in), ceph_snap(in), 1085 vino.ino, vino.snap); 1086 have_lease = false; 1087 in = NULL; 1088 } 1089 1090 if (have_lease) 1091 update_dentry_lease(dn, rinfo->dlease, session, 1092 req->r_request_started); 1093 dout(" final dn %p\n", dn); 1094 i++; 1095 } else if (req->r_op == CEPH_MDS_OP_LOOKUPSNAP || 1096 req->r_op == CEPH_MDS_OP_MKSNAP) { 1097 struct dentry *dn = req->r_dentry; 1098 1099 /* fill out a snapdir LOOKUPSNAP dentry */ 1100 BUG_ON(!dn); 1101 BUG_ON(!req->r_locked_dir); 1102 BUG_ON(ceph_snap(req->r_locked_dir) != CEPH_SNAPDIR); 1103 ininfo = rinfo->targeti.in; 1104 vino.ino = le64_to_cpu(ininfo->ino); 1105 vino.snap = le64_to_cpu(ininfo->snapid); 1106 in = ceph_get_inode(sb, vino); 1107 if (IS_ERR(in)) { 1108 pr_err("fill_inode get_inode badness %llx.%llx\n", 1109 vino.ino, vino.snap); 1110 err = PTR_ERR(in); 1111 d_delete(dn); 1112 goto done; 1113 } 1114 dout(" linking snapped dir %p to dn %p\n", in, dn); 1115 dn = splice_dentry(dn, in, NULL); 1116 if (IS_ERR(dn)) { 1117 err = PTR_ERR(dn); 1118 goto done; 1119 } 1120 ceph_set_dentry_offset(dn); 1121 req->r_dentry = dn; /* may have spliced */ 1122 igrab(in); 1123 rinfo->head->is_dentry = 1; /* fool notrace handlers */ 1124 } 1125 1126 if (rinfo->head->is_target) { 1127 vino.ino = le64_to_cpu(rinfo->targeti.in->ino); 1128 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid); 1129 1130 if (in == NULL || ceph_ino(in) != vino.ino || 1131 ceph_snap(in) != vino.snap) { 1132 in = ceph_get_inode(sb, vino); 1133 if (IS_ERR(in)) { 1134 err = PTR_ERR(in); 1135 goto done; 1136 } 1137 } 1138 req->r_target_inode = in; 1139 1140 err = fill_inode(in, 1141 &rinfo->targeti, NULL, 1142 session, req->r_request_started, 1143 (le32_to_cpu(rinfo->head->result) == 0) ? 1144 req->r_fmode : -1, 1145 &req->r_caps_reservation); 1146 if (err < 0) { 1147 pr_err("fill_inode badness %p %llx.%llx\n", 1148 in, ceph_vinop(in)); 1149 goto done; 1150 } 1151 } 1152 1153 done: 1154 dout("fill_trace done err=%d\n", err); 1155 return err; 1156 } 1157 1158 /* 1159 * Prepopulate our cache with readdir results, leases, etc. 1160 */ 1161 int ceph_readdir_prepopulate(struct ceph_mds_request *req, 1162 struct ceph_mds_session *session) 1163 { 1164 struct dentry *parent = req->r_dentry; 1165 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1166 struct qstr dname; 1167 struct dentry *dn; 1168 struct inode *in; 1169 int err = 0, i; 1170 struct inode *snapdir = NULL; 1171 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base; 1172 u64 frag = le32_to_cpu(rhead->args.readdir.frag); 1173 struct ceph_dentry_info *di; 1174 1175 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) { 1176 snapdir = ceph_get_snapdir(parent->d_inode); 1177 parent = d_find_alias(snapdir); 1178 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n", 1179 rinfo->dir_nr, parent); 1180 } else { 1181 dout("readdir_prepopulate %d items under dn %p\n", 1182 rinfo->dir_nr, parent); 1183 if (rinfo->dir_dir) 1184 ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir); 1185 } 1186 1187 for (i = 0; i < rinfo->dir_nr; i++) { 1188 struct ceph_vino vino; 1189 1190 dname.name = rinfo->dir_dname[i]; 1191 dname.len = rinfo->dir_dname_len[i]; 1192 dname.hash = full_name_hash(dname.name, dname.len); 1193 1194 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino); 1195 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid); 1196 1197 retry_lookup: 1198 dn = d_lookup(parent, &dname); 1199 dout("d_lookup on parent=%p name=%.*s got %p\n", 1200 parent, dname.len, dname.name, dn); 1201 1202 if (!dn) { 1203 dn = d_alloc(parent, &dname); 1204 dout("d_alloc %p '%.*s' = %p\n", parent, 1205 dname.len, dname.name, dn); 1206 if (dn == NULL) { 1207 dout("d_alloc badness\n"); 1208 err = -ENOMEM; 1209 goto out; 1210 } 1211 err = ceph_init_dentry(dn); 1212 if (err < 0) 1213 goto out; 1214 } else if (dn->d_inode && 1215 (ceph_ino(dn->d_inode) != vino.ino || 1216 ceph_snap(dn->d_inode) != vino.snap)) { 1217 dout(" dn %p points to wrong inode %p\n", 1218 dn, dn->d_inode); 1219 d_delete(dn); 1220 dput(dn); 1221 goto retry_lookup; 1222 } else { 1223 /* reorder parent's d_subdirs */ 1224 spin_lock(&dcache_lock); 1225 spin_lock(&dn->d_lock); 1226 list_move(&dn->d_u.d_child, &parent->d_subdirs); 1227 spin_unlock(&dn->d_lock); 1228 spin_unlock(&dcache_lock); 1229 } 1230 1231 di = dn->d_fsdata; 1232 di->offset = ceph_make_fpos(frag, i + req->r_readdir_offset); 1233 1234 /* inode */ 1235 if (dn->d_inode) { 1236 in = dn->d_inode; 1237 } else { 1238 in = ceph_get_inode(parent->d_sb, vino); 1239 if (in == NULL) { 1240 dout("new_inode badness\n"); 1241 d_delete(dn); 1242 dput(dn); 1243 err = -ENOMEM; 1244 goto out; 1245 } 1246 dn = splice_dentry(dn, in, NULL); 1247 } 1248 1249 if (fill_inode(in, &rinfo->dir_in[i], NULL, session, 1250 req->r_request_started, -1, 1251 &req->r_caps_reservation) < 0) { 1252 pr_err("fill_inode badness on %p\n", in); 1253 dput(dn); 1254 continue; 1255 } 1256 update_dentry_lease(dn, rinfo->dir_dlease[i], 1257 req->r_session, req->r_request_started); 1258 dput(dn); 1259 } 1260 req->r_did_prepopulate = true; 1261 1262 out: 1263 if (snapdir) { 1264 iput(snapdir); 1265 dput(parent); 1266 } 1267 dout("readdir_prepopulate done\n"); 1268 return err; 1269 } 1270 1271 int ceph_inode_set_size(struct inode *inode, loff_t size) 1272 { 1273 struct ceph_inode_info *ci = ceph_inode(inode); 1274 int ret = 0; 1275 1276 spin_lock(&inode->i_lock); 1277 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size); 1278 inode->i_size = size; 1279 inode->i_blocks = (size + (1 << 9) - 1) >> 9; 1280 1281 /* tell the MDS if we are approaching max_size */ 1282 if ((size << 1) >= ci->i_max_size && 1283 (ci->i_reported_size << 1) < ci->i_max_size) 1284 ret = 1; 1285 1286 spin_unlock(&inode->i_lock); 1287 return ret; 1288 } 1289 1290 /* 1291 * Write back inode data in a worker thread. (This can't be done 1292 * in the message handler context.) 1293 */ 1294 void ceph_queue_writeback(struct inode *inode) 1295 { 1296 if (queue_work(ceph_inode_to_client(inode)->wb_wq, 1297 &ceph_inode(inode)->i_wb_work)) { 1298 dout("ceph_queue_writeback %p\n", inode); 1299 igrab(inode); 1300 } else { 1301 dout("ceph_queue_writeback %p failed\n", inode); 1302 } 1303 } 1304 1305 static void ceph_writeback_work(struct work_struct *work) 1306 { 1307 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1308 i_wb_work); 1309 struct inode *inode = &ci->vfs_inode; 1310 1311 dout("writeback %p\n", inode); 1312 filemap_fdatawrite(&inode->i_data); 1313 iput(inode); 1314 } 1315 1316 /* 1317 * queue an async invalidation 1318 */ 1319 void ceph_queue_invalidate(struct inode *inode) 1320 { 1321 if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq, 1322 &ceph_inode(inode)->i_pg_inv_work)) { 1323 dout("ceph_queue_invalidate %p\n", inode); 1324 igrab(inode); 1325 } else { 1326 dout("ceph_queue_invalidate %p failed\n", inode); 1327 } 1328 } 1329 1330 /* 1331 * invalidate any pages that are not dirty or under writeback. this 1332 * includes pages that are clean and mapped. 1333 */ 1334 static void ceph_invalidate_nondirty_pages(struct address_space *mapping) 1335 { 1336 struct pagevec pvec; 1337 pgoff_t next = 0; 1338 int i; 1339 1340 pagevec_init(&pvec, 0); 1341 while (pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { 1342 for (i = 0; i < pagevec_count(&pvec); i++) { 1343 struct page *page = pvec.pages[i]; 1344 pgoff_t index; 1345 int skip_page = 1346 (PageDirty(page) || PageWriteback(page)); 1347 1348 if (!skip_page) 1349 skip_page = !trylock_page(page); 1350 1351 /* 1352 * We really shouldn't be looking at the ->index of an 1353 * unlocked page. But we're not allowed to lock these 1354 * pages. So we rely upon nobody altering the ->index 1355 * of this (pinned-by-us) page. 1356 */ 1357 index = page->index; 1358 if (index > next) 1359 next = index; 1360 next++; 1361 1362 if (skip_page) 1363 continue; 1364 1365 generic_error_remove_page(mapping, page); 1366 unlock_page(page); 1367 } 1368 pagevec_release(&pvec); 1369 cond_resched(); 1370 } 1371 } 1372 1373 /* 1374 * Invalidate inode pages in a worker thread. (This can't be done 1375 * in the message handler context.) 1376 */ 1377 static void ceph_invalidate_work(struct work_struct *work) 1378 { 1379 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1380 i_pg_inv_work); 1381 struct inode *inode = &ci->vfs_inode; 1382 u32 orig_gen; 1383 int check = 0; 1384 1385 spin_lock(&inode->i_lock); 1386 dout("invalidate_pages %p gen %d revoking %d\n", inode, 1387 ci->i_rdcache_gen, ci->i_rdcache_revoking); 1388 if (ci->i_rdcache_gen == 0 || 1389 ci->i_rdcache_revoking != ci->i_rdcache_gen) { 1390 BUG_ON(ci->i_rdcache_revoking > ci->i_rdcache_gen); 1391 /* nevermind! */ 1392 ci->i_rdcache_revoking = 0; 1393 spin_unlock(&inode->i_lock); 1394 goto out; 1395 } 1396 orig_gen = ci->i_rdcache_gen; 1397 spin_unlock(&inode->i_lock); 1398 1399 ceph_invalidate_nondirty_pages(inode->i_mapping); 1400 1401 spin_lock(&inode->i_lock); 1402 if (orig_gen == ci->i_rdcache_gen) { 1403 dout("invalidate_pages %p gen %d successful\n", inode, 1404 ci->i_rdcache_gen); 1405 ci->i_rdcache_gen = 0; 1406 ci->i_rdcache_revoking = 0; 1407 check = 1; 1408 } else { 1409 dout("invalidate_pages %p gen %d raced, gen now %d\n", 1410 inode, orig_gen, ci->i_rdcache_gen); 1411 } 1412 spin_unlock(&inode->i_lock); 1413 1414 if (check) 1415 ceph_check_caps(ci, 0, NULL); 1416 out: 1417 iput(inode); 1418 } 1419 1420 1421 /* 1422 * called by trunc_wq; take i_mutex ourselves 1423 * 1424 * We also truncate in a separate thread as well. 1425 */ 1426 static void ceph_vmtruncate_work(struct work_struct *work) 1427 { 1428 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1429 i_vmtruncate_work); 1430 struct inode *inode = &ci->vfs_inode; 1431 1432 dout("vmtruncate_work %p\n", inode); 1433 mutex_lock(&inode->i_mutex); 1434 __ceph_do_pending_vmtruncate(inode); 1435 mutex_unlock(&inode->i_mutex); 1436 iput(inode); 1437 } 1438 1439 /* 1440 * Queue an async vmtruncate. If we fail to queue work, we will handle 1441 * the truncation the next time we call __ceph_do_pending_vmtruncate. 1442 */ 1443 void ceph_queue_vmtruncate(struct inode *inode) 1444 { 1445 struct ceph_inode_info *ci = ceph_inode(inode); 1446 1447 if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq, 1448 &ci->i_vmtruncate_work)) { 1449 dout("ceph_queue_vmtruncate %p\n", inode); 1450 igrab(inode); 1451 } else { 1452 dout("ceph_queue_vmtruncate %p failed, pending=%d\n", 1453 inode, ci->i_truncate_pending); 1454 } 1455 } 1456 1457 /* 1458 * called with i_mutex held. 1459 * 1460 * Make sure any pending truncation is applied before doing anything 1461 * that may depend on it. 1462 */ 1463 void __ceph_do_pending_vmtruncate(struct inode *inode) 1464 { 1465 struct ceph_inode_info *ci = ceph_inode(inode); 1466 u64 to; 1467 int wrbuffer_refs, wake = 0; 1468 1469 retry: 1470 spin_lock(&inode->i_lock); 1471 if (ci->i_truncate_pending == 0) { 1472 dout("__do_pending_vmtruncate %p none pending\n", inode); 1473 spin_unlock(&inode->i_lock); 1474 return; 1475 } 1476 1477 /* 1478 * make sure any dirty snapped pages are flushed before we 1479 * possibly truncate them.. so write AND block! 1480 */ 1481 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) { 1482 dout("__do_pending_vmtruncate %p flushing snaps first\n", 1483 inode); 1484 spin_unlock(&inode->i_lock); 1485 filemap_write_and_wait_range(&inode->i_data, 0, 1486 inode->i_sb->s_maxbytes); 1487 goto retry; 1488 } 1489 1490 to = ci->i_truncate_size; 1491 wrbuffer_refs = ci->i_wrbuffer_ref; 1492 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode, 1493 ci->i_truncate_pending, to); 1494 spin_unlock(&inode->i_lock); 1495 1496 truncate_inode_pages(inode->i_mapping, to); 1497 1498 spin_lock(&inode->i_lock); 1499 ci->i_truncate_pending--; 1500 if (ci->i_truncate_pending == 0) 1501 wake = 1; 1502 spin_unlock(&inode->i_lock); 1503 1504 if (wrbuffer_refs == 0) 1505 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); 1506 if (wake) 1507 wake_up(&ci->i_cap_wq); 1508 } 1509 1510 1511 /* 1512 * symlinks 1513 */ 1514 static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd) 1515 { 1516 struct ceph_inode_info *ci = ceph_inode(dentry->d_inode); 1517 nd_set_link(nd, ci->i_symlink); 1518 return NULL; 1519 } 1520 1521 static const struct inode_operations ceph_symlink_iops = { 1522 .readlink = generic_readlink, 1523 .follow_link = ceph_sym_follow_link, 1524 }; 1525 1526 /* 1527 * setattr 1528 */ 1529 int ceph_setattr(struct dentry *dentry, struct iattr *attr) 1530 { 1531 struct inode *inode = dentry->d_inode; 1532 struct ceph_inode_info *ci = ceph_inode(inode); 1533 struct inode *parent_inode = dentry->d_parent->d_inode; 1534 const unsigned int ia_valid = attr->ia_valid; 1535 struct ceph_mds_request *req; 1536 struct ceph_mds_client *mdsc = &ceph_sb_to_client(dentry->d_sb)->mdsc; 1537 int issued; 1538 int release = 0, dirtied = 0; 1539 int mask = 0; 1540 int err = 0; 1541 1542 if (ceph_snap(inode) != CEPH_NOSNAP) 1543 return -EROFS; 1544 1545 __ceph_do_pending_vmtruncate(inode); 1546 1547 err = inode_change_ok(inode, attr); 1548 if (err != 0) 1549 return err; 1550 1551 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR, 1552 USE_AUTH_MDS); 1553 if (IS_ERR(req)) 1554 return PTR_ERR(req); 1555 1556 spin_lock(&inode->i_lock); 1557 issued = __ceph_caps_issued(ci, NULL); 1558 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued)); 1559 1560 if (ia_valid & ATTR_UID) { 1561 dout("setattr %p uid %d -> %d\n", inode, 1562 inode->i_uid, attr->ia_uid); 1563 if (issued & CEPH_CAP_AUTH_EXCL) { 1564 inode->i_uid = attr->ia_uid; 1565 dirtied |= CEPH_CAP_AUTH_EXCL; 1566 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1567 attr->ia_uid != inode->i_uid) { 1568 req->r_args.setattr.uid = cpu_to_le32(attr->ia_uid); 1569 mask |= CEPH_SETATTR_UID; 1570 release |= CEPH_CAP_AUTH_SHARED; 1571 } 1572 } 1573 if (ia_valid & ATTR_GID) { 1574 dout("setattr %p gid %d -> %d\n", inode, 1575 inode->i_gid, attr->ia_gid); 1576 if (issued & CEPH_CAP_AUTH_EXCL) { 1577 inode->i_gid = attr->ia_gid; 1578 dirtied |= CEPH_CAP_AUTH_EXCL; 1579 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1580 attr->ia_gid != inode->i_gid) { 1581 req->r_args.setattr.gid = cpu_to_le32(attr->ia_gid); 1582 mask |= CEPH_SETATTR_GID; 1583 release |= CEPH_CAP_AUTH_SHARED; 1584 } 1585 } 1586 if (ia_valid & ATTR_MODE) { 1587 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode, 1588 attr->ia_mode); 1589 if (issued & CEPH_CAP_AUTH_EXCL) { 1590 inode->i_mode = attr->ia_mode; 1591 dirtied |= CEPH_CAP_AUTH_EXCL; 1592 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1593 attr->ia_mode != inode->i_mode) { 1594 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode); 1595 mask |= CEPH_SETATTR_MODE; 1596 release |= CEPH_CAP_AUTH_SHARED; 1597 } 1598 } 1599 1600 if (ia_valid & ATTR_ATIME) { 1601 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode, 1602 inode->i_atime.tv_sec, inode->i_atime.tv_nsec, 1603 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec); 1604 if (issued & CEPH_CAP_FILE_EXCL) { 1605 ci->i_time_warp_seq++; 1606 inode->i_atime = attr->ia_atime; 1607 dirtied |= CEPH_CAP_FILE_EXCL; 1608 } else if ((issued & CEPH_CAP_FILE_WR) && 1609 timespec_compare(&inode->i_atime, 1610 &attr->ia_atime) < 0) { 1611 inode->i_atime = attr->ia_atime; 1612 dirtied |= CEPH_CAP_FILE_WR; 1613 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1614 !timespec_equal(&inode->i_atime, &attr->ia_atime)) { 1615 ceph_encode_timespec(&req->r_args.setattr.atime, 1616 &attr->ia_atime); 1617 mask |= CEPH_SETATTR_ATIME; 1618 release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD | 1619 CEPH_CAP_FILE_WR; 1620 } 1621 } 1622 if (ia_valid & ATTR_MTIME) { 1623 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode, 1624 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 1625 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec); 1626 if (issued & CEPH_CAP_FILE_EXCL) { 1627 ci->i_time_warp_seq++; 1628 inode->i_mtime = attr->ia_mtime; 1629 dirtied |= CEPH_CAP_FILE_EXCL; 1630 } else if ((issued & CEPH_CAP_FILE_WR) && 1631 timespec_compare(&inode->i_mtime, 1632 &attr->ia_mtime) < 0) { 1633 inode->i_mtime = attr->ia_mtime; 1634 dirtied |= CEPH_CAP_FILE_WR; 1635 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1636 !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) { 1637 ceph_encode_timespec(&req->r_args.setattr.mtime, 1638 &attr->ia_mtime); 1639 mask |= CEPH_SETATTR_MTIME; 1640 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | 1641 CEPH_CAP_FILE_WR; 1642 } 1643 } 1644 if (ia_valid & ATTR_SIZE) { 1645 dout("setattr %p size %lld -> %lld\n", inode, 1646 inode->i_size, attr->ia_size); 1647 if (attr->ia_size > inode->i_sb->s_maxbytes) { 1648 err = -EINVAL; 1649 goto out; 1650 } 1651 if ((issued & CEPH_CAP_FILE_EXCL) && 1652 attr->ia_size > inode->i_size) { 1653 inode->i_size = attr->ia_size; 1654 inode->i_blocks = 1655 (attr->ia_size + (1 << 9) - 1) >> 9; 1656 inode->i_ctime = attr->ia_ctime; 1657 ci->i_reported_size = attr->ia_size; 1658 dirtied |= CEPH_CAP_FILE_EXCL; 1659 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1660 attr->ia_size != inode->i_size) { 1661 req->r_args.setattr.size = cpu_to_le64(attr->ia_size); 1662 req->r_args.setattr.old_size = 1663 cpu_to_le64(inode->i_size); 1664 mask |= CEPH_SETATTR_SIZE; 1665 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | 1666 CEPH_CAP_FILE_WR; 1667 } 1668 } 1669 1670 /* these do nothing */ 1671 if (ia_valid & ATTR_CTIME) { 1672 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME| 1673 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0; 1674 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode, 1675 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 1676 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec, 1677 only ? "ctime only" : "ignored"); 1678 inode->i_ctime = attr->ia_ctime; 1679 if (only) { 1680 /* 1681 * if kernel wants to dirty ctime but nothing else, 1682 * we need to choose a cap to dirty under, or do 1683 * a almost-no-op setattr 1684 */ 1685 if (issued & CEPH_CAP_AUTH_EXCL) 1686 dirtied |= CEPH_CAP_AUTH_EXCL; 1687 else if (issued & CEPH_CAP_FILE_EXCL) 1688 dirtied |= CEPH_CAP_FILE_EXCL; 1689 else if (issued & CEPH_CAP_XATTR_EXCL) 1690 dirtied |= CEPH_CAP_XATTR_EXCL; 1691 else 1692 mask |= CEPH_SETATTR_CTIME; 1693 } 1694 } 1695 if (ia_valid & ATTR_FILE) 1696 dout("setattr %p ATTR_FILE ... hrm!\n", inode); 1697 1698 if (dirtied) { 1699 __ceph_mark_dirty_caps(ci, dirtied); 1700 inode->i_ctime = CURRENT_TIME; 1701 } 1702 1703 release &= issued; 1704 spin_unlock(&inode->i_lock); 1705 1706 if (mask) { 1707 req->r_inode = igrab(inode); 1708 req->r_inode_drop = release; 1709 req->r_args.setattr.mask = cpu_to_le32(mask); 1710 req->r_num_caps = 1; 1711 err = ceph_mdsc_do_request(mdsc, parent_inode, req); 1712 } 1713 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err, 1714 ceph_cap_string(dirtied), mask); 1715 1716 ceph_mdsc_put_request(req); 1717 __ceph_do_pending_vmtruncate(inode); 1718 return err; 1719 out: 1720 spin_unlock(&inode->i_lock); 1721 ceph_mdsc_put_request(req); 1722 return err; 1723 } 1724 1725 /* 1726 * Verify that we have a lease on the given mask. If not, 1727 * do a getattr against an mds. 1728 */ 1729 int ceph_do_getattr(struct inode *inode, int mask) 1730 { 1731 struct ceph_client *client = ceph_sb_to_client(inode->i_sb); 1732 struct ceph_mds_client *mdsc = &client->mdsc; 1733 struct ceph_mds_request *req; 1734 int err; 1735 1736 if (ceph_snap(inode) == CEPH_SNAPDIR) { 1737 dout("do_getattr inode %p SNAPDIR\n", inode); 1738 return 0; 1739 } 1740 1741 dout("do_getattr inode %p mask %s\n", inode, ceph_cap_string(mask)); 1742 if (ceph_caps_issued_mask(ceph_inode(inode), mask, 1)) 1743 return 0; 1744 1745 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS); 1746 if (IS_ERR(req)) 1747 return PTR_ERR(req); 1748 req->r_inode = igrab(inode); 1749 req->r_num_caps = 1; 1750 req->r_args.getattr.mask = cpu_to_le32(mask); 1751 err = ceph_mdsc_do_request(mdsc, NULL, req); 1752 ceph_mdsc_put_request(req); 1753 dout("do_getattr result=%d\n", err); 1754 return err; 1755 } 1756 1757 1758 /* 1759 * Check inode permissions. We verify we have a valid value for 1760 * the AUTH cap, then call the generic handler. 1761 */ 1762 int ceph_permission(struct inode *inode, int mask) 1763 { 1764 int err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED); 1765 1766 if (!err) 1767 err = generic_permission(inode, mask, NULL); 1768 return err; 1769 } 1770 1771 /* 1772 * Get all attributes. Hopefully somedata we'll have a statlite() 1773 * and can limit the fields we require to be accurate. 1774 */ 1775 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry, 1776 struct kstat *stat) 1777 { 1778 struct inode *inode = dentry->d_inode; 1779 struct ceph_inode_info *ci = ceph_inode(inode); 1780 int err; 1781 1782 err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL); 1783 if (!err) { 1784 generic_fillattr(inode, stat); 1785 stat->ino = inode->i_ino; 1786 if (ceph_snap(inode) != CEPH_NOSNAP) 1787 stat->dev = ceph_snap(inode); 1788 else 1789 stat->dev = 0; 1790 if (S_ISDIR(inode->i_mode)) { 1791 stat->size = ci->i_rbytes; 1792 stat->blocks = 0; 1793 stat->blksize = 65536; 1794 } 1795 } 1796 return err; 1797 } 1798