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