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 14 #include "super.h" 15 #include "decode.h" 16 17 /* 18 * Ceph inode operations 19 * 20 * Implement basic inode helpers (get, alloc) and inode ops (getattr, 21 * setattr, etc.), xattr helpers, and helpers for assimilating 22 * metadata returned by the MDS into our cache. 23 * 24 * Also define helpers for doing asynchronous writeback, invalidation, 25 * and truncation for the benefit of those who can't afford to block 26 * (typically because they are in the message handler path). 27 */ 28 29 static const struct inode_operations ceph_symlink_iops; 30 31 static void ceph_inode_invalidate_pages(struct work_struct *work); 32 33 /* 34 * find or create an inode, given the ceph ino number 35 */ 36 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino) 37 { 38 struct inode *inode; 39 ino_t t = ceph_vino_to_ino(vino); 40 41 inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino); 42 if (inode == NULL) 43 return ERR_PTR(-ENOMEM); 44 if (inode->i_state & I_NEW) { 45 dout("get_inode created new inode %p %llx.%llx ino %llx\n", 46 inode, ceph_vinop(inode), (u64)inode->i_ino); 47 unlock_new_inode(inode); 48 } 49 50 dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino, 51 vino.snap, inode); 52 return inode; 53 } 54 55 /* 56 * get/constuct snapdir inode for a given directory 57 */ 58 struct inode *ceph_get_snapdir(struct inode *parent) 59 { 60 struct ceph_vino vino = { 61 .ino = ceph_ino(parent), 62 .snap = CEPH_SNAPDIR, 63 }; 64 struct inode *inode = ceph_get_inode(parent->i_sb, vino); 65 struct ceph_inode_info *ci = ceph_inode(inode); 66 67 BUG_ON(!S_ISDIR(parent->i_mode)); 68 if (IS_ERR(inode)) 69 return ERR_PTR(PTR_ERR(inode)); 70 inode->i_mode = parent->i_mode; 71 inode->i_uid = parent->i_uid; 72 inode->i_gid = parent->i_gid; 73 inode->i_op = &ceph_dir_iops; 74 inode->i_fop = &ceph_dir_fops; 75 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */ 76 ci->i_rbytes = 0; 77 return inode; 78 } 79 80 const struct inode_operations ceph_file_iops = { 81 .permission = ceph_permission, 82 .setattr = ceph_setattr, 83 .getattr = ceph_getattr, 84 .setxattr = ceph_setxattr, 85 .getxattr = ceph_getxattr, 86 .listxattr = ceph_listxattr, 87 .removexattr = ceph_removexattr, 88 }; 89 90 91 /* 92 * We use a 'frag tree' to keep track of the MDS's directory fragments 93 * for a given inode (usually there is just a single fragment). We 94 * need to know when a child frag is delegated to a new MDS, or when 95 * it is flagged as replicated, so we can direct our requests 96 * accordingly. 97 */ 98 99 /* 100 * find/create a frag in the tree 101 */ 102 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci, 103 u32 f) 104 { 105 struct rb_node **p; 106 struct rb_node *parent = NULL; 107 struct ceph_inode_frag *frag; 108 int c; 109 110 p = &ci->i_fragtree.rb_node; 111 while (*p) { 112 parent = *p; 113 frag = rb_entry(parent, struct ceph_inode_frag, node); 114 c = ceph_frag_compare(f, frag->frag); 115 if (c < 0) 116 p = &(*p)->rb_left; 117 else if (c > 0) 118 p = &(*p)->rb_right; 119 else 120 return frag; 121 } 122 123 frag = kmalloc(sizeof(*frag), GFP_NOFS); 124 if (!frag) { 125 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx " 126 "frag %x\n", &ci->vfs_inode, 127 ceph_vinop(&ci->vfs_inode), f); 128 return ERR_PTR(-ENOMEM); 129 } 130 frag->frag = f; 131 frag->split_by = 0; 132 frag->mds = -1; 133 frag->ndist = 0; 134 135 rb_link_node(&frag->node, parent, p); 136 rb_insert_color(&frag->node, &ci->i_fragtree); 137 138 dout("get_or_create_frag added %llx.%llx frag %x\n", 139 ceph_vinop(&ci->vfs_inode), f); 140 return frag; 141 } 142 143 /* 144 * find a specific frag @f 145 */ 146 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f) 147 { 148 struct rb_node *n = ci->i_fragtree.rb_node; 149 150 while (n) { 151 struct ceph_inode_frag *frag = 152 rb_entry(n, struct ceph_inode_frag, node); 153 int c = ceph_frag_compare(f, frag->frag); 154 if (c < 0) 155 n = n->rb_left; 156 else if (c > 0) 157 n = n->rb_right; 158 else 159 return frag; 160 } 161 return NULL; 162 } 163 164 /* 165 * Choose frag containing the given value @v. If @pfrag is 166 * specified, copy the frag delegation info to the caller if 167 * it is present. 168 */ 169 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v, 170 struct ceph_inode_frag *pfrag, 171 int *found) 172 { 173 u32 t = ceph_frag_make(0, 0); 174 struct ceph_inode_frag *frag; 175 unsigned nway, i; 176 u32 n; 177 178 if (found) 179 *found = 0; 180 181 mutex_lock(&ci->i_fragtree_mutex); 182 while (1) { 183 WARN_ON(!ceph_frag_contains_value(t, v)); 184 frag = __ceph_find_frag(ci, t); 185 if (!frag) 186 break; /* t is a leaf */ 187 if (frag->split_by == 0) { 188 if (pfrag) 189 memcpy(pfrag, frag, sizeof(*pfrag)); 190 if (found) 191 *found = 1; 192 break; 193 } 194 195 /* choose child */ 196 nway = 1 << frag->split_by; 197 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t, 198 frag->split_by, nway); 199 for (i = 0; i < nway; i++) { 200 n = ceph_frag_make_child(t, frag->split_by, i); 201 if (ceph_frag_contains_value(n, v)) { 202 t = n; 203 break; 204 } 205 } 206 BUG_ON(i == nway); 207 } 208 dout("choose_frag(%x) = %x\n", v, t); 209 210 mutex_unlock(&ci->i_fragtree_mutex); 211 return t; 212 } 213 214 /* 215 * Process dirfrag (delegation) info from the mds. Include leaf 216 * fragment in tree ONLY if ndist > 0. Otherwise, only 217 * branches/splits are included in i_fragtree) 218 */ 219 static int ceph_fill_dirfrag(struct inode *inode, 220 struct ceph_mds_reply_dirfrag *dirinfo) 221 { 222 struct ceph_inode_info *ci = ceph_inode(inode); 223 struct ceph_inode_frag *frag; 224 u32 id = le32_to_cpu(dirinfo->frag); 225 int mds = le32_to_cpu(dirinfo->auth); 226 int ndist = le32_to_cpu(dirinfo->ndist); 227 int i; 228 int err = 0; 229 230 mutex_lock(&ci->i_fragtree_mutex); 231 if (ndist == 0) { 232 /* no delegation info needed. */ 233 frag = __ceph_find_frag(ci, id); 234 if (!frag) 235 goto out; 236 if (frag->split_by == 0) { 237 /* tree leaf, remove */ 238 dout("fill_dirfrag removed %llx.%llx frag %x" 239 " (no ref)\n", ceph_vinop(inode), id); 240 rb_erase(&frag->node, &ci->i_fragtree); 241 kfree(frag); 242 } else { 243 /* tree branch, keep and clear */ 244 dout("fill_dirfrag cleared %llx.%llx frag %x" 245 " referral\n", ceph_vinop(inode), id); 246 frag->mds = -1; 247 frag->ndist = 0; 248 } 249 goto out; 250 } 251 252 253 /* find/add this frag to store mds delegation info */ 254 frag = __get_or_create_frag(ci, id); 255 if (IS_ERR(frag)) { 256 /* this is not the end of the world; we can continue 257 with bad/inaccurate delegation info */ 258 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n", 259 ceph_vinop(inode), le32_to_cpu(dirinfo->frag)); 260 err = -ENOMEM; 261 goto out; 262 } 263 264 frag->mds = mds; 265 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP); 266 for (i = 0; i < frag->ndist; i++) 267 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]); 268 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n", 269 ceph_vinop(inode), frag->frag, frag->ndist); 270 271 out: 272 mutex_unlock(&ci->i_fragtree_mutex); 273 return err; 274 } 275 276 277 /* 278 * initialize a newly allocated inode. 279 */ 280 struct inode *ceph_alloc_inode(struct super_block *sb) 281 { 282 struct ceph_inode_info *ci; 283 int i; 284 285 ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS); 286 if (!ci) 287 return NULL; 288 289 dout("alloc_inode %p\n", &ci->vfs_inode); 290 291 ci->i_version = 0; 292 ci->i_time_warp_seq = 0; 293 ci->i_ceph_flags = 0; 294 ci->i_release_count = 0; 295 ci->i_symlink = NULL; 296 297 ci->i_fragtree = RB_ROOT; 298 mutex_init(&ci->i_fragtree_mutex); 299 300 ci->i_xattrs.blob = NULL; 301 ci->i_xattrs.prealloc_blob = NULL; 302 ci->i_xattrs.dirty = false; 303 ci->i_xattrs.index = RB_ROOT; 304 ci->i_xattrs.count = 0; 305 ci->i_xattrs.names_size = 0; 306 ci->i_xattrs.vals_size = 0; 307 ci->i_xattrs.version = 0; 308 ci->i_xattrs.index_version = 0; 309 310 ci->i_caps = RB_ROOT; 311 ci->i_auth_cap = NULL; 312 ci->i_dirty_caps = 0; 313 ci->i_flushing_caps = 0; 314 INIT_LIST_HEAD(&ci->i_dirty_item); 315 INIT_LIST_HEAD(&ci->i_flushing_item); 316 ci->i_cap_flush_seq = 0; 317 ci->i_cap_flush_last_tid = 0; 318 memset(&ci->i_cap_flush_tid, 0, sizeof(ci->i_cap_flush_tid)); 319 init_waitqueue_head(&ci->i_cap_wq); 320 ci->i_hold_caps_min = 0; 321 ci->i_hold_caps_max = 0; 322 INIT_LIST_HEAD(&ci->i_cap_delay_list); 323 ci->i_cap_exporting_mds = 0; 324 ci->i_cap_exporting_mseq = 0; 325 ci->i_cap_exporting_issued = 0; 326 INIT_LIST_HEAD(&ci->i_cap_snaps); 327 ci->i_head_snapc = NULL; 328 ci->i_snap_caps = 0; 329 330 for (i = 0; i < CEPH_FILE_MODE_NUM; i++) 331 ci->i_nr_by_mode[i] = 0; 332 333 ci->i_truncate_seq = 0; 334 ci->i_truncate_size = 0; 335 ci->i_truncate_pending = 0; 336 337 ci->i_max_size = 0; 338 ci->i_reported_size = 0; 339 ci->i_wanted_max_size = 0; 340 ci->i_requested_max_size = 0; 341 342 ci->i_pin_ref = 0; 343 ci->i_rd_ref = 0; 344 ci->i_rdcache_ref = 0; 345 ci->i_wr_ref = 0; 346 ci->i_wrbuffer_ref = 0; 347 ci->i_wrbuffer_ref_head = 0; 348 ci->i_shared_gen = 0; 349 ci->i_rdcache_gen = 0; 350 ci->i_rdcache_revoking = 0; 351 352 INIT_LIST_HEAD(&ci->i_unsafe_writes); 353 INIT_LIST_HEAD(&ci->i_unsafe_dirops); 354 spin_lock_init(&ci->i_unsafe_lock); 355 356 ci->i_snap_realm = NULL; 357 INIT_LIST_HEAD(&ci->i_snap_realm_item); 358 INIT_LIST_HEAD(&ci->i_snap_flush_item); 359 360 INIT_WORK(&ci->i_wb_work, ceph_inode_writeback); 361 INIT_WORK(&ci->i_pg_inv_work, ceph_inode_invalidate_pages); 362 363 INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work); 364 365 return &ci->vfs_inode; 366 } 367 368 void ceph_destroy_inode(struct inode *inode) 369 { 370 struct ceph_inode_info *ci = ceph_inode(inode); 371 struct ceph_inode_frag *frag; 372 struct rb_node *n; 373 374 dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode)); 375 376 ceph_queue_caps_release(inode); 377 378 kfree(ci->i_symlink); 379 while ((n = rb_first(&ci->i_fragtree)) != NULL) { 380 frag = rb_entry(n, struct ceph_inode_frag, node); 381 rb_erase(n, &ci->i_fragtree); 382 kfree(frag); 383 } 384 385 __ceph_destroy_xattrs(ci); 386 if (ci->i_xattrs.blob) 387 ceph_buffer_put(ci->i_xattrs.blob); 388 if (ci->i_xattrs.prealloc_blob) 389 ceph_buffer_put(ci->i_xattrs.prealloc_blob); 390 391 kmem_cache_free(ceph_inode_cachep, ci); 392 } 393 394 395 /* 396 * Helpers to fill in size, ctime, mtime, and atime. We have to be 397 * careful because either the client or MDS may have more up to date 398 * info, depending on which capabilities are held, and whether 399 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime 400 * and size are monotonically increasing, except when utimes() or 401 * truncate() increments the corresponding _seq values.) 402 */ 403 int ceph_fill_file_size(struct inode *inode, int issued, 404 u32 truncate_seq, u64 truncate_size, u64 size) 405 { 406 struct ceph_inode_info *ci = ceph_inode(inode); 407 int queue_trunc = 0; 408 409 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 || 410 (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) { 411 dout("size %lld -> %llu\n", inode->i_size, size); 412 inode->i_size = size; 413 inode->i_blocks = (size + (1<<9) - 1) >> 9; 414 ci->i_reported_size = size; 415 if (truncate_seq != ci->i_truncate_seq) { 416 dout("truncate_seq %u -> %u\n", 417 ci->i_truncate_seq, truncate_seq); 418 ci->i_truncate_seq = truncate_seq; 419 /* 420 * If we hold relevant caps, or in the case where we're 421 * not the only client referencing this file and we 422 * don't hold those caps, then we need to check whether 423 * the file is either opened or mmaped 424 */ 425 if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_RD| 426 CEPH_CAP_FILE_WR|CEPH_CAP_FILE_BUFFER| 427 CEPH_CAP_FILE_EXCL)) || 428 mapping_mapped(inode->i_mapping) || 429 __ceph_caps_file_wanted(ci)) { 430 ci->i_truncate_pending++; 431 queue_trunc = 1; 432 } 433 } 434 } 435 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 && 436 ci->i_truncate_size != truncate_size) { 437 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size, 438 truncate_size); 439 ci->i_truncate_size = truncate_size; 440 } 441 return queue_trunc; 442 } 443 444 void ceph_fill_file_time(struct inode *inode, int issued, 445 u64 time_warp_seq, struct timespec *ctime, 446 struct timespec *mtime, struct timespec *atime) 447 { 448 struct ceph_inode_info *ci = ceph_inode(inode); 449 int warn = 0; 450 451 if (issued & (CEPH_CAP_FILE_EXCL| 452 CEPH_CAP_FILE_WR| 453 CEPH_CAP_FILE_BUFFER)) { 454 if (timespec_compare(ctime, &inode->i_ctime) > 0) { 455 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n", 456 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 457 ctime->tv_sec, ctime->tv_nsec); 458 inode->i_ctime = *ctime; 459 } 460 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) { 461 /* the MDS did a utimes() */ 462 dout("mtime %ld.%09ld -> %ld.%09ld " 463 "tw %d -> %d\n", 464 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 465 mtime->tv_sec, mtime->tv_nsec, 466 ci->i_time_warp_seq, (int)time_warp_seq); 467 468 inode->i_mtime = *mtime; 469 inode->i_atime = *atime; 470 ci->i_time_warp_seq = time_warp_seq; 471 } else if (time_warp_seq == ci->i_time_warp_seq) { 472 /* nobody did utimes(); take the max */ 473 if (timespec_compare(mtime, &inode->i_mtime) > 0) { 474 dout("mtime %ld.%09ld -> %ld.%09ld inc\n", 475 inode->i_mtime.tv_sec, 476 inode->i_mtime.tv_nsec, 477 mtime->tv_sec, mtime->tv_nsec); 478 inode->i_mtime = *mtime; 479 } 480 if (timespec_compare(atime, &inode->i_atime) > 0) { 481 dout("atime %ld.%09ld -> %ld.%09ld inc\n", 482 inode->i_atime.tv_sec, 483 inode->i_atime.tv_nsec, 484 atime->tv_sec, atime->tv_nsec); 485 inode->i_atime = *atime; 486 } 487 } else if (issued & CEPH_CAP_FILE_EXCL) { 488 /* we did a utimes(); ignore mds values */ 489 } else { 490 warn = 1; 491 } 492 } else { 493 /* we have no write caps; whatever the MDS says is true */ 494 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) { 495 inode->i_ctime = *ctime; 496 inode->i_mtime = *mtime; 497 inode->i_atime = *atime; 498 ci->i_time_warp_seq = time_warp_seq; 499 } else { 500 warn = 1; 501 } 502 } 503 if (warn) /* time_warp_seq shouldn't go backwards */ 504 dout("%p mds time_warp_seq %llu < %u\n", 505 inode, time_warp_seq, ci->i_time_warp_seq); 506 } 507 508 /* 509 * Populate an inode based on info from mds. May be called on new or 510 * existing inodes. 511 */ 512 static int fill_inode(struct inode *inode, 513 struct ceph_mds_reply_info_in *iinfo, 514 struct ceph_mds_reply_dirfrag *dirinfo, 515 struct ceph_mds_session *session, 516 unsigned long ttl_from, int cap_fmode, 517 struct ceph_cap_reservation *caps_reservation) 518 { 519 struct ceph_mds_reply_inode *info = iinfo->in; 520 struct ceph_inode_info *ci = ceph_inode(inode); 521 int i; 522 int issued, implemented; 523 struct timespec mtime, atime, ctime; 524 u32 nsplits; 525 struct ceph_buffer *xattr_blob = NULL; 526 int err = 0; 527 int queue_trunc = 0; 528 529 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n", 530 inode, ceph_vinop(inode), le64_to_cpu(info->version), 531 ci->i_version); 532 533 /* 534 * prealloc xattr data, if it looks like we'll need it. only 535 * if len > 4 (meaning there are actually xattrs; the first 4 536 * bytes are the xattr count). 537 */ 538 if (iinfo->xattr_len > 4) { 539 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS); 540 if (!xattr_blob) 541 pr_err("fill_inode ENOMEM xattr blob %d bytes\n", 542 iinfo->xattr_len); 543 } 544 545 spin_lock(&inode->i_lock); 546 547 /* 548 * provided version will be odd if inode value is projected, 549 * even if stable. skip the update if we have a newer info 550 * (e.g., due to inode info racing form multiple MDSs), or if 551 * we are getting projected (unstable) inode info. 552 */ 553 if (le64_to_cpu(info->version) > 0 && 554 (ci->i_version & ~1) > le64_to_cpu(info->version)) 555 goto no_change; 556 557 issued = __ceph_caps_issued(ci, &implemented); 558 issued |= implemented | __ceph_caps_dirty(ci); 559 560 /* update inode */ 561 ci->i_version = le64_to_cpu(info->version); 562 inode->i_version++; 563 inode->i_rdev = le32_to_cpu(info->rdev); 564 565 if ((issued & CEPH_CAP_AUTH_EXCL) == 0) { 566 inode->i_mode = le32_to_cpu(info->mode); 567 inode->i_uid = le32_to_cpu(info->uid); 568 inode->i_gid = le32_to_cpu(info->gid); 569 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode, 570 inode->i_uid, inode->i_gid); 571 } 572 573 if ((issued & CEPH_CAP_LINK_EXCL) == 0) 574 inode->i_nlink = le32_to_cpu(info->nlink); 575 576 /* be careful with mtime, atime, size */ 577 ceph_decode_timespec(&atime, &info->atime); 578 ceph_decode_timespec(&mtime, &info->mtime); 579 ceph_decode_timespec(&ctime, &info->ctime); 580 queue_trunc = ceph_fill_file_size(inode, issued, 581 le32_to_cpu(info->truncate_seq), 582 le64_to_cpu(info->truncate_size), 583 le64_to_cpu(info->size)); 584 ceph_fill_file_time(inode, issued, 585 le32_to_cpu(info->time_warp_seq), 586 &ctime, &mtime, &atime); 587 588 ci->i_max_size = le64_to_cpu(info->max_size); 589 ci->i_layout = info->layout; 590 inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1; 591 592 /* xattrs */ 593 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */ 594 if ((issued & CEPH_CAP_XATTR_EXCL) == 0 && 595 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) { 596 if (ci->i_xattrs.blob) 597 ceph_buffer_put(ci->i_xattrs.blob); 598 ci->i_xattrs.blob = xattr_blob; 599 if (xattr_blob) 600 memcpy(ci->i_xattrs.blob->vec.iov_base, 601 iinfo->xattr_data, iinfo->xattr_len); 602 ci->i_xattrs.version = le64_to_cpu(info->xattr_version); 603 } 604 605 inode->i_mapping->a_ops = &ceph_aops; 606 inode->i_mapping->backing_dev_info = 607 &ceph_client(inode->i_sb)->backing_dev_info; 608 609 switch (inode->i_mode & S_IFMT) { 610 case S_IFIFO: 611 case S_IFBLK: 612 case S_IFCHR: 613 case S_IFSOCK: 614 init_special_inode(inode, inode->i_mode, inode->i_rdev); 615 inode->i_op = &ceph_file_iops; 616 break; 617 case S_IFREG: 618 inode->i_op = &ceph_file_iops; 619 inode->i_fop = &ceph_file_fops; 620 break; 621 case S_IFLNK: 622 inode->i_op = &ceph_symlink_iops; 623 if (!ci->i_symlink) { 624 int symlen = iinfo->symlink_len; 625 char *sym; 626 627 BUG_ON(symlen != inode->i_size); 628 spin_unlock(&inode->i_lock); 629 630 err = -ENOMEM; 631 sym = kmalloc(symlen+1, GFP_NOFS); 632 if (!sym) 633 goto out; 634 memcpy(sym, iinfo->symlink, symlen); 635 sym[symlen] = 0; 636 637 spin_lock(&inode->i_lock); 638 if (!ci->i_symlink) 639 ci->i_symlink = sym; 640 else 641 kfree(sym); /* lost a race */ 642 } 643 break; 644 case S_IFDIR: 645 inode->i_op = &ceph_dir_iops; 646 inode->i_fop = &ceph_dir_fops; 647 648 ci->i_files = le64_to_cpu(info->files); 649 ci->i_subdirs = le64_to_cpu(info->subdirs); 650 ci->i_rbytes = le64_to_cpu(info->rbytes); 651 ci->i_rfiles = le64_to_cpu(info->rfiles); 652 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs); 653 ceph_decode_timespec(&ci->i_rctime, &info->rctime); 654 655 /* set dir completion flag? */ 656 if (ci->i_files == 0 && ci->i_subdirs == 0 && 657 ceph_snap(inode) == CEPH_NOSNAP && 658 (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED)) { 659 dout(" marking %p complete (empty)\n", inode); 660 ci->i_ceph_flags |= CEPH_I_COMPLETE; 661 ci->i_max_offset = 2; 662 } 663 664 /* it may be better to set st_size in getattr instead? */ 665 if (ceph_test_opt(ceph_client(inode->i_sb), RBYTES)) 666 inode->i_size = ci->i_rbytes; 667 break; 668 default: 669 pr_err("fill_inode %llx.%llx BAD mode 0%o\n", 670 ceph_vinop(inode), inode->i_mode); 671 } 672 673 no_change: 674 spin_unlock(&inode->i_lock); 675 676 /* queue truncate if we saw i_size decrease */ 677 if (queue_trunc) 678 if (queue_work(ceph_client(inode->i_sb)->trunc_wq, 679 &ci->i_vmtruncate_work)) 680 igrab(inode); 681 682 /* populate frag tree */ 683 /* FIXME: move me up, if/when version reflects fragtree changes */ 684 nsplits = le32_to_cpu(info->fragtree.nsplits); 685 mutex_lock(&ci->i_fragtree_mutex); 686 for (i = 0; i < nsplits; i++) { 687 u32 id = le32_to_cpu(info->fragtree.splits[i].frag); 688 struct ceph_inode_frag *frag = __get_or_create_frag(ci, id); 689 690 if (IS_ERR(frag)) 691 continue; 692 frag->split_by = le32_to_cpu(info->fragtree.splits[i].by); 693 dout(" frag %x split by %d\n", frag->frag, frag->split_by); 694 } 695 mutex_unlock(&ci->i_fragtree_mutex); 696 697 /* were we issued a capability? */ 698 if (info->cap.caps) { 699 if (ceph_snap(inode) == CEPH_NOSNAP) { 700 ceph_add_cap(inode, session, 701 le64_to_cpu(info->cap.cap_id), 702 cap_fmode, 703 le32_to_cpu(info->cap.caps), 704 le32_to_cpu(info->cap.wanted), 705 le32_to_cpu(info->cap.seq), 706 le32_to_cpu(info->cap.mseq), 707 le64_to_cpu(info->cap.realm), 708 info->cap.flags, 709 caps_reservation); 710 } else { 711 spin_lock(&inode->i_lock); 712 dout(" %p got snap_caps %s\n", inode, 713 ceph_cap_string(le32_to_cpu(info->cap.caps))); 714 ci->i_snap_caps |= le32_to_cpu(info->cap.caps); 715 if (cap_fmode >= 0) 716 __ceph_get_fmode(ci, cap_fmode); 717 spin_unlock(&inode->i_lock); 718 } 719 } 720 721 /* update delegation info? */ 722 if (dirinfo) 723 ceph_fill_dirfrag(inode, dirinfo); 724 725 err = 0; 726 727 out: 728 if (xattr_blob) 729 ceph_buffer_put(xattr_blob); 730 return err; 731 } 732 733 /* 734 * caller should hold session s_mutex. 735 */ 736 static void update_dentry_lease(struct dentry *dentry, 737 struct ceph_mds_reply_lease *lease, 738 struct ceph_mds_session *session, 739 unsigned long from_time) 740 { 741 struct ceph_dentry_info *di = ceph_dentry(dentry); 742 long unsigned duration = le32_to_cpu(lease->duration_ms); 743 long unsigned ttl = from_time + (duration * HZ) / 1000; 744 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000; 745 struct inode *dir; 746 747 /* only track leases on regular dentries */ 748 if (dentry->d_op != &ceph_dentry_ops) 749 return; 750 751 spin_lock(&dentry->d_lock); 752 dout("update_dentry_lease %p mask %d duration %lu ms ttl %lu\n", 753 dentry, le16_to_cpu(lease->mask), duration, ttl); 754 755 /* make lease_rdcache_gen match directory */ 756 dir = dentry->d_parent->d_inode; 757 di->lease_shared_gen = ceph_inode(dir)->i_shared_gen; 758 759 if (lease->mask == 0) 760 goto out_unlock; 761 762 if (di->lease_gen == session->s_cap_gen && 763 time_before(ttl, dentry->d_time)) 764 goto out_unlock; /* we already have a newer lease. */ 765 766 if (di->lease_session && di->lease_session != session) 767 goto out_unlock; 768 769 ceph_dentry_lru_touch(dentry); 770 771 if (!di->lease_session) 772 di->lease_session = ceph_get_mds_session(session); 773 di->lease_gen = session->s_cap_gen; 774 di->lease_seq = le32_to_cpu(lease->seq); 775 di->lease_renew_after = half_ttl; 776 di->lease_renew_from = 0; 777 dentry->d_time = ttl; 778 out_unlock: 779 spin_unlock(&dentry->d_lock); 780 return; 781 } 782 783 /* 784 * splice a dentry to an inode. 785 * caller must hold directory i_mutex for this to be safe. 786 * 787 * we will only rehash the resulting dentry if @prehash is 788 * true; @prehash will be set to false (for the benefit of 789 * the caller) if we fail. 790 */ 791 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in, 792 bool *prehash) 793 { 794 struct dentry *realdn; 795 796 /* dn must be unhashed */ 797 if (!d_unhashed(dn)) 798 d_drop(dn); 799 realdn = d_materialise_unique(dn, in); 800 if (IS_ERR(realdn)) { 801 pr_err("splice_dentry error %p inode %p ino %llx.%llx\n", 802 dn, in, ceph_vinop(in)); 803 if (prehash) 804 *prehash = false; /* don't rehash on error */ 805 dn = realdn; /* note realdn contains the error */ 806 goto out; 807 } else if (realdn) { 808 dout("dn %p (%d) spliced with %p (%d) " 809 "inode %p ino %llx.%llx\n", 810 dn, atomic_read(&dn->d_count), 811 realdn, atomic_read(&realdn->d_count), 812 realdn->d_inode, ceph_vinop(realdn->d_inode)); 813 dput(dn); 814 dn = realdn; 815 } else { 816 BUG_ON(!ceph_dentry(dn)); 817 818 dout("dn %p attached to %p ino %llx.%llx\n", 819 dn, dn->d_inode, ceph_vinop(dn->d_inode)); 820 } 821 if ((!prehash || *prehash) && d_unhashed(dn)) 822 d_rehash(dn); 823 out: 824 return dn; 825 } 826 827 /* 828 * Set dentry's directory position based on the current dir's max, and 829 * order it in d_subdirs, so that dcache_readdir behaves. 830 */ 831 static void ceph_set_dentry_offset(struct dentry *dn) 832 { 833 struct dentry *dir = dn->d_parent; 834 struct inode *inode = dn->d_parent->d_inode; 835 struct ceph_dentry_info *di; 836 837 BUG_ON(!inode); 838 839 di = ceph_dentry(dn); 840 841 spin_lock(&inode->i_lock); 842 di->offset = ceph_inode(inode)->i_max_offset++; 843 spin_unlock(&inode->i_lock); 844 845 spin_lock(&dcache_lock); 846 spin_lock(&dn->d_lock); 847 list_move_tail(&dir->d_subdirs, &dn->d_u.d_child); 848 dout("set_dentry_offset %p %lld (%p %p)\n", dn, di->offset, 849 dn->d_u.d_child.prev, dn->d_u.d_child.next); 850 spin_unlock(&dn->d_lock); 851 spin_unlock(&dcache_lock); 852 } 853 854 /* 855 * Incorporate results into the local cache. This is either just 856 * one inode, or a directory, dentry, and possibly linked-to inode (e.g., 857 * after a lookup). 858 * 859 * A reply may contain 860 * a directory inode along with a dentry. 861 * and/or a target inode 862 * 863 * Called with snap_rwsem (read). 864 */ 865 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req, 866 struct ceph_mds_session *session) 867 { 868 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 869 struct inode *in = NULL; 870 struct ceph_mds_reply_inode *ininfo; 871 struct ceph_vino vino; 872 int i = 0; 873 int err = 0; 874 875 dout("fill_trace %p is_dentry %d is_target %d\n", req, 876 rinfo->head->is_dentry, rinfo->head->is_target); 877 878 #if 0 879 /* 880 * Debugging hook: 881 * 882 * If we resend completed ops to a recovering mds, we get no 883 * trace. Since that is very rare, pretend this is the case 884 * to ensure the 'no trace' handlers in the callers behave. 885 * 886 * Fill in inodes unconditionally to avoid breaking cap 887 * invariants. 888 */ 889 if (rinfo->head->op & CEPH_MDS_OP_WRITE) { 890 pr_info("fill_trace faking empty trace on %lld %s\n", 891 req->r_tid, ceph_mds_op_name(rinfo->head->op)); 892 if (rinfo->head->is_dentry) { 893 rinfo->head->is_dentry = 0; 894 err = fill_inode(req->r_locked_dir, 895 &rinfo->diri, rinfo->dirfrag, 896 session, req->r_request_started, -1); 897 } 898 if (rinfo->head->is_target) { 899 rinfo->head->is_target = 0; 900 ininfo = rinfo->targeti.in; 901 vino.ino = le64_to_cpu(ininfo->ino); 902 vino.snap = le64_to_cpu(ininfo->snapid); 903 in = ceph_get_inode(sb, vino); 904 err = fill_inode(in, &rinfo->targeti, NULL, 905 session, req->r_request_started, 906 req->r_fmode); 907 iput(in); 908 } 909 } 910 #endif 911 912 if (!rinfo->head->is_target && !rinfo->head->is_dentry) { 913 dout("fill_trace reply is empty!\n"); 914 if (rinfo->head->result == 0 && req->r_locked_dir) { 915 struct ceph_inode_info *ci = 916 ceph_inode(req->r_locked_dir); 917 dout(" clearing %p complete (empty trace)\n", 918 req->r_locked_dir); 919 ci->i_ceph_flags &= ~CEPH_I_COMPLETE; 920 ci->i_release_count++; 921 } 922 return 0; 923 } 924 925 if (rinfo->head->is_dentry) { 926 struct inode *dir = req->r_locked_dir; 927 928 err = fill_inode(dir, &rinfo->diri, rinfo->dirfrag, 929 session, req->r_request_started, -1, 930 &req->r_caps_reservation); 931 if (err < 0) 932 return err; 933 } 934 935 if (rinfo->head->is_dentry && !req->r_aborted) { 936 /* 937 * lookup link rename : null -> possibly existing inode 938 * mknod symlink mkdir : null -> new inode 939 * unlink : linked -> null 940 */ 941 struct inode *dir = req->r_locked_dir; 942 struct dentry *dn = req->r_dentry; 943 bool have_dir_cap, have_lease; 944 945 BUG_ON(!dn); 946 BUG_ON(!dir); 947 BUG_ON(dn->d_parent->d_inode != dir); 948 BUG_ON(ceph_ino(dir) != 949 le64_to_cpu(rinfo->diri.in->ino)); 950 BUG_ON(ceph_snap(dir) != 951 le64_to_cpu(rinfo->diri.in->snapid)); 952 953 /* do we have a lease on the whole dir? */ 954 have_dir_cap = 955 (le32_to_cpu(rinfo->diri.in->cap.caps) & 956 CEPH_CAP_FILE_SHARED); 957 958 /* do we have a dn lease? */ 959 have_lease = have_dir_cap || 960 (le16_to_cpu(rinfo->dlease->mask) & 961 CEPH_LOCK_DN); 962 963 if (!have_lease) 964 dout("fill_trace no dentry lease or dir cap\n"); 965 966 /* rename? */ 967 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) { 968 dout(" src %p '%.*s' dst %p '%.*s'\n", 969 req->r_old_dentry, 970 req->r_old_dentry->d_name.len, 971 req->r_old_dentry->d_name.name, 972 dn, dn->d_name.len, dn->d_name.name); 973 dout("fill_trace doing d_move %p -> %p\n", 974 req->r_old_dentry, dn); 975 d_move(req->r_old_dentry, dn); 976 dout(" src %p '%.*s' dst %p '%.*s'\n", 977 req->r_old_dentry, 978 req->r_old_dentry->d_name.len, 979 req->r_old_dentry->d_name.name, 980 dn, dn->d_name.len, dn->d_name.name); 981 /* ensure target dentry is invalidated, despite 982 rehashing bug in vfs_rename_dir */ 983 dn->d_time = jiffies; 984 ceph_dentry(dn)->lease_shared_gen = 0; 985 /* take overwritten dentry's readdir offset */ 986 ceph_dentry(req->r_old_dentry)->offset = 987 ceph_dentry(dn)->offset; 988 dn = req->r_old_dentry; /* use old_dentry */ 989 in = dn->d_inode; 990 } 991 992 /* null dentry? */ 993 if (!rinfo->head->is_target) { 994 dout("fill_trace null dentry\n"); 995 if (dn->d_inode) { 996 dout("d_delete %p\n", dn); 997 d_delete(dn); 998 } else { 999 dout("d_instantiate %p NULL\n", dn); 1000 d_instantiate(dn, NULL); 1001 if (have_lease && d_unhashed(dn)) 1002 d_rehash(dn); 1003 update_dentry_lease(dn, rinfo->dlease, 1004 session, 1005 req->r_request_started); 1006 } 1007 goto done; 1008 } 1009 1010 /* attach proper inode */ 1011 ininfo = rinfo->targeti.in; 1012 vino.ino = le64_to_cpu(ininfo->ino); 1013 vino.snap = le64_to_cpu(ininfo->snapid); 1014 if (!dn->d_inode) { 1015 in = ceph_get_inode(sb, vino); 1016 if (IS_ERR(in)) { 1017 pr_err("fill_trace bad get_inode " 1018 "%llx.%llx\n", vino.ino, vino.snap); 1019 err = PTR_ERR(in); 1020 d_delete(dn); 1021 goto done; 1022 } 1023 dn = splice_dentry(dn, in, &have_lease); 1024 if (IS_ERR(dn)) { 1025 err = PTR_ERR(dn); 1026 goto done; 1027 } 1028 req->r_dentry = dn; /* may have spliced */ 1029 ceph_set_dentry_offset(dn); 1030 igrab(in); 1031 } else if (ceph_ino(in) == vino.ino && 1032 ceph_snap(in) == vino.snap) { 1033 igrab(in); 1034 } else { 1035 dout(" %p links to %p %llx.%llx, not %llx.%llx\n", 1036 dn, in, ceph_ino(in), ceph_snap(in), 1037 vino.ino, vino.snap); 1038 have_lease = false; 1039 in = NULL; 1040 } 1041 1042 if (have_lease) 1043 update_dentry_lease(dn, rinfo->dlease, session, 1044 req->r_request_started); 1045 dout(" final dn %p\n", dn); 1046 i++; 1047 } else if (req->r_op == CEPH_MDS_OP_LOOKUPSNAP || 1048 req->r_op == CEPH_MDS_OP_MKSNAP) { 1049 struct dentry *dn = req->r_dentry; 1050 1051 /* fill out a snapdir LOOKUPSNAP dentry */ 1052 BUG_ON(!dn); 1053 BUG_ON(!req->r_locked_dir); 1054 BUG_ON(ceph_snap(req->r_locked_dir) != CEPH_SNAPDIR); 1055 ininfo = rinfo->targeti.in; 1056 vino.ino = le64_to_cpu(ininfo->ino); 1057 vino.snap = le64_to_cpu(ininfo->snapid); 1058 in = ceph_get_inode(sb, vino); 1059 if (IS_ERR(in)) { 1060 pr_err("fill_inode get_inode badness %llx.%llx\n", 1061 vino.ino, vino.snap); 1062 err = PTR_ERR(in); 1063 d_delete(dn); 1064 goto done; 1065 } 1066 dout(" linking snapped dir %p to dn %p\n", in, dn); 1067 dn = splice_dentry(dn, in, NULL); 1068 if (IS_ERR(dn)) { 1069 err = PTR_ERR(dn); 1070 goto done; 1071 } 1072 ceph_set_dentry_offset(dn); 1073 req->r_dentry = dn; /* may have spliced */ 1074 igrab(in); 1075 rinfo->head->is_dentry = 1; /* fool notrace handlers */ 1076 } 1077 1078 if (rinfo->head->is_target) { 1079 vino.ino = le64_to_cpu(rinfo->targeti.in->ino); 1080 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid); 1081 1082 if (in == NULL || ceph_ino(in) != vino.ino || 1083 ceph_snap(in) != vino.snap) { 1084 in = ceph_get_inode(sb, vino); 1085 if (IS_ERR(in)) { 1086 err = PTR_ERR(in); 1087 goto done; 1088 } 1089 } 1090 req->r_target_inode = in; 1091 1092 err = fill_inode(in, 1093 &rinfo->targeti, NULL, 1094 session, req->r_request_started, 1095 (le32_to_cpu(rinfo->head->result) == 0) ? 1096 req->r_fmode : -1, 1097 &req->r_caps_reservation); 1098 if (err < 0) { 1099 pr_err("fill_inode badness %p %llx.%llx\n", 1100 in, ceph_vinop(in)); 1101 goto done; 1102 } 1103 } 1104 1105 done: 1106 dout("fill_trace done err=%d\n", err); 1107 return err; 1108 } 1109 1110 /* 1111 * Prepopulate our cache with readdir results, leases, etc. 1112 */ 1113 int ceph_readdir_prepopulate(struct ceph_mds_request *req, 1114 struct ceph_mds_session *session) 1115 { 1116 struct dentry *parent = req->r_dentry; 1117 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1118 struct qstr dname; 1119 struct dentry *dn; 1120 struct inode *in; 1121 int err = 0, i; 1122 struct inode *snapdir = NULL; 1123 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base; 1124 u64 frag = le32_to_cpu(rhead->args.readdir.frag); 1125 struct ceph_dentry_info *di; 1126 1127 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) { 1128 snapdir = ceph_get_snapdir(parent->d_inode); 1129 parent = d_find_alias(snapdir); 1130 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n", 1131 rinfo->dir_nr, parent); 1132 } else { 1133 dout("readdir_prepopulate %d items under dn %p\n", 1134 rinfo->dir_nr, parent); 1135 if (rinfo->dir_dir) 1136 ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir); 1137 } 1138 1139 for (i = 0; i < rinfo->dir_nr; i++) { 1140 struct ceph_vino vino; 1141 1142 dname.name = rinfo->dir_dname[i]; 1143 dname.len = rinfo->dir_dname_len[i]; 1144 dname.hash = full_name_hash(dname.name, dname.len); 1145 1146 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino); 1147 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid); 1148 1149 retry_lookup: 1150 dn = d_lookup(parent, &dname); 1151 dout("d_lookup on parent=%p name=%.*s got %p\n", 1152 parent, dname.len, dname.name, dn); 1153 1154 if (!dn) { 1155 dn = d_alloc(parent, &dname); 1156 dout("d_alloc %p '%.*s' = %p\n", parent, 1157 dname.len, dname.name, dn); 1158 if (dn == NULL) { 1159 dout("d_alloc badness\n"); 1160 err = -ENOMEM; 1161 goto out; 1162 } 1163 err = ceph_init_dentry(dn); 1164 if (err < 0) 1165 goto out; 1166 } else if (dn->d_inode && 1167 (ceph_ino(dn->d_inode) != vino.ino || 1168 ceph_snap(dn->d_inode) != vino.snap)) { 1169 dout(" dn %p points to wrong inode %p\n", 1170 dn, dn->d_inode); 1171 d_delete(dn); 1172 dput(dn); 1173 goto retry_lookup; 1174 } else { 1175 /* reorder parent's d_subdirs */ 1176 spin_lock(&dcache_lock); 1177 spin_lock(&dn->d_lock); 1178 list_move(&dn->d_u.d_child, &parent->d_subdirs); 1179 spin_unlock(&dn->d_lock); 1180 spin_unlock(&dcache_lock); 1181 } 1182 1183 di = dn->d_fsdata; 1184 di->offset = ceph_make_fpos(frag, i + req->r_readdir_offset); 1185 1186 /* inode */ 1187 if (dn->d_inode) { 1188 in = dn->d_inode; 1189 } else { 1190 in = ceph_get_inode(parent->d_sb, vino); 1191 if (in == NULL) { 1192 dout("new_inode badness\n"); 1193 d_delete(dn); 1194 dput(dn); 1195 err = -ENOMEM; 1196 goto out; 1197 } 1198 dn = splice_dentry(dn, in, NULL); 1199 } 1200 1201 if (fill_inode(in, &rinfo->dir_in[i], NULL, session, 1202 req->r_request_started, -1, 1203 &req->r_caps_reservation) < 0) { 1204 pr_err("fill_inode badness on %p\n", in); 1205 dput(dn); 1206 continue; 1207 } 1208 update_dentry_lease(dn, rinfo->dir_dlease[i], 1209 req->r_session, req->r_request_started); 1210 dput(dn); 1211 } 1212 req->r_did_prepopulate = true; 1213 1214 out: 1215 if (snapdir) { 1216 iput(snapdir); 1217 dput(parent); 1218 } 1219 dout("readdir_prepopulate done\n"); 1220 return err; 1221 } 1222 1223 int ceph_inode_set_size(struct inode *inode, loff_t size) 1224 { 1225 struct ceph_inode_info *ci = ceph_inode(inode); 1226 int ret = 0; 1227 1228 spin_lock(&inode->i_lock); 1229 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size); 1230 inode->i_size = size; 1231 inode->i_blocks = (size + (1 << 9) - 1) >> 9; 1232 1233 /* tell the MDS if we are approaching max_size */ 1234 if ((size << 1) >= ci->i_max_size && 1235 (ci->i_reported_size << 1) < ci->i_max_size) 1236 ret = 1; 1237 1238 spin_unlock(&inode->i_lock); 1239 return ret; 1240 } 1241 1242 /* 1243 * Write back inode data in a worker thread. (This can't be done 1244 * in the message handler context.) 1245 */ 1246 void ceph_inode_writeback(struct work_struct *work) 1247 { 1248 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1249 i_wb_work); 1250 struct inode *inode = &ci->vfs_inode; 1251 1252 dout("writeback %p\n", inode); 1253 filemap_fdatawrite(&inode->i_data); 1254 iput(inode); 1255 } 1256 1257 /* 1258 * Invalidate inode pages in a worker thread. (This can't be done 1259 * in the message handler context.) 1260 */ 1261 static void ceph_inode_invalidate_pages(struct work_struct *work) 1262 { 1263 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1264 i_pg_inv_work); 1265 struct inode *inode = &ci->vfs_inode; 1266 u32 orig_gen; 1267 int check = 0; 1268 1269 spin_lock(&inode->i_lock); 1270 dout("invalidate_pages %p gen %d revoking %d\n", inode, 1271 ci->i_rdcache_gen, ci->i_rdcache_revoking); 1272 if (ci->i_rdcache_gen == 0 || 1273 ci->i_rdcache_revoking != ci->i_rdcache_gen) { 1274 BUG_ON(ci->i_rdcache_revoking > ci->i_rdcache_gen); 1275 /* nevermind! */ 1276 ci->i_rdcache_revoking = 0; 1277 spin_unlock(&inode->i_lock); 1278 goto out; 1279 } 1280 orig_gen = ci->i_rdcache_gen; 1281 spin_unlock(&inode->i_lock); 1282 1283 truncate_inode_pages(&inode->i_data, 0); 1284 1285 spin_lock(&inode->i_lock); 1286 if (orig_gen == ci->i_rdcache_gen) { 1287 dout("invalidate_pages %p gen %d successful\n", inode, 1288 ci->i_rdcache_gen); 1289 ci->i_rdcache_gen = 0; 1290 ci->i_rdcache_revoking = 0; 1291 check = 1; 1292 } else { 1293 dout("invalidate_pages %p gen %d raced, gen now %d\n", 1294 inode, orig_gen, ci->i_rdcache_gen); 1295 } 1296 spin_unlock(&inode->i_lock); 1297 1298 if (check) 1299 ceph_check_caps(ci, 0, NULL); 1300 out: 1301 iput(inode); 1302 } 1303 1304 1305 /* 1306 * called by trunc_wq; take i_mutex ourselves 1307 * 1308 * We also truncate in a separate thread as well. 1309 */ 1310 void ceph_vmtruncate_work(struct work_struct *work) 1311 { 1312 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1313 i_vmtruncate_work); 1314 struct inode *inode = &ci->vfs_inode; 1315 1316 dout("vmtruncate_work %p\n", inode); 1317 mutex_lock(&inode->i_mutex); 1318 __ceph_do_pending_vmtruncate(inode); 1319 mutex_unlock(&inode->i_mutex); 1320 iput(inode); 1321 } 1322 1323 /* 1324 * called with i_mutex held. 1325 * 1326 * Make sure any pending truncation is applied before doing anything 1327 * that may depend on it. 1328 */ 1329 void __ceph_do_pending_vmtruncate(struct inode *inode) 1330 { 1331 struct ceph_inode_info *ci = ceph_inode(inode); 1332 u64 to; 1333 int wrbuffer_refs, wake = 0; 1334 1335 retry: 1336 spin_lock(&inode->i_lock); 1337 if (ci->i_truncate_pending == 0) { 1338 dout("__do_pending_vmtruncate %p none pending\n", inode); 1339 spin_unlock(&inode->i_lock); 1340 return; 1341 } 1342 1343 /* 1344 * make sure any dirty snapped pages are flushed before we 1345 * possibly truncate them.. so write AND block! 1346 */ 1347 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) { 1348 dout("__do_pending_vmtruncate %p flushing snaps first\n", 1349 inode); 1350 spin_unlock(&inode->i_lock); 1351 filemap_write_and_wait_range(&inode->i_data, 0, 1352 inode->i_sb->s_maxbytes); 1353 goto retry; 1354 } 1355 1356 to = ci->i_truncate_size; 1357 wrbuffer_refs = ci->i_wrbuffer_ref; 1358 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode, 1359 ci->i_truncate_pending, to); 1360 spin_unlock(&inode->i_lock); 1361 1362 truncate_inode_pages(inode->i_mapping, to); 1363 1364 spin_lock(&inode->i_lock); 1365 ci->i_truncate_pending--; 1366 if (ci->i_truncate_pending == 0) 1367 wake = 1; 1368 spin_unlock(&inode->i_lock); 1369 1370 if (wrbuffer_refs == 0) 1371 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); 1372 if (wake) 1373 wake_up(&ci->i_cap_wq); 1374 } 1375 1376 1377 /* 1378 * symlinks 1379 */ 1380 static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd) 1381 { 1382 struct ceph_inode_info *ci = ceph_inode(dentry->d_inode); 1383 nd_set_link(nd, ci->i_symlink); 1384 return NULL; 1385 } 1386 1387 static const struct inode_operations ceph_symlink_iops = { 1388 .readlink = generic_readlink, 1389 .follow_link = ceph_sym_follow_link, 1390 }; 1391 1392 /* 1393 * setattr 1394 */ 1395 int ceph_setattr(struct dentry *dentry, struct iattr *attr) 1396 { 1397 struct inode *inode = dentry->d_inode; 1398 struct ceph_inode_info *ci = ceph_inode(inode); 1399 struct inode *parent_inode = dentry->d_parent->d_inode; 1400 const unsigned int ia_valid = attr->ia_valid; 1401 struct ceph_mds_request *req; 1402 struct ceph_mds_client *mdsc = &ceph_client(dentry->d_sb)->mdsc; 1403 int issued; 1404 int release = 0, dirtied = 0; 1405 int mask = 0; 1406 int err = 0; 1407 1408 if (ceph_snap(inode) != CEPH_NOSNAP) 1409 return -EROFS; 1410 1411 __ceph_do_pending_vmtruncate(inode); 1412 1413 err = inode_change_ok(inode, attr); 1414 if (err != 0) 1415 return err; 1416 1417 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR, 1418 USE_AUTH_MDS); 1419 if (IS_ERR(req)) 1420 return PTR_ERR(req); 1421 1422 spin_lock(&inode->i_lock); 1423 issued = __ceph_caps_issued(ci, NULL); 1424 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued)); 1425 1426 if (ia_valid & ATTR_UID) { 1427 dout("setattr %p uid %d -> %d\n", inode, 1428 inode->i_uid, attr->ia_uid); 1429 if (issued & CEPH_CAP_AUTH_EXCL) { 1430 inode->i_uid = attr->ia_uid; 1431 dirtied |= CEPH_CAP_AUTH_EXCL; 1432 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1433 attr->ia_uid != inode->i_uid) { 1434 req->r_args.setattr.uid = cpu_to_le32(attr->ia_uid); 1435 mask |= CEPH_SETATTR_UID; 1436 release |= CEPH_CAP_AUTH_SHARED; 1437 } 1438 } 1439 if (ia_valid & ATTR_GID) { 1440 dout("setattr %p gid %d -> %d\n", inode, 1441 inode->i_gid, attr->ia_gid); 1442 if (issued & CEPH_CAP_AUTH_EXCL) { 1443 inode->i_gid = attr->ia_gid; 1444 dirtied |= CEPH_CAP_AUTH_EXCL; 1445 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1446 attr->ia_gid != inode->i_gid) { 1447 req->r_args.setattr.gid = cpu_to_le32(attr->ia_gid); 1448 mask |= CEPH_SETATTR_GID; 1449 release |= CEPH_CAP_AUTH_SHARED; 1450 } 1451 } 1452 if (ia_valid & ATTR_MODE) { 1453 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode, 1454 attr->ia_mode); 1455 if (issued & CEPH_CAP_AUTH_EXCL) { 1456 inode->i_mode = attr->ia_mode; 1457 dirtied |= CEPH_CAP_AUTH_EXCL; 1458 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1459 attr->ia_mode != inode->i_mode) { 1460 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode); 1461 mask |= CEPH_SETATTR_MODE; 1462 release |= CEPH_CAP_AUTH_SHARED; 1463 } 1464 } 1465 1466 if (ia_valid & ATTR_ATIME) { 1467 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode, 1468 inode->i_atime.tv_sec, inode->i_atime.tv_nsec, 1469 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec); 1470 if (issued & CEPH_CAP_FILE_EXCL) { 1471 ci->i_time_warp_seq++; 1472 inode->i_atime = attr->ia_atime; 1473 dirtied |= CEPH_CAP_FILE_EXCL; 1474 } else if ((issued & CEPH_CAP_FILE_WR) && 1475 timespec_compare(&inode->i_atime, 1476 &attr->ia_atime) < 0) { 1477 inode->i_atime = attr->ia_atime; 1478 dirtied |= CEPH_CAP_FILE_WR; 1479 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1480 !timespec_equal(&inode->i_atime, &attr->ia_atime)) { 1481 ceph_encode_timespec(&req->r_args.setattr.atime, 1482 &attr->ia_atime); 1483 mask |= CEPH_SETATTR_ATIME; 1484 release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD | 1485 CEPH_CAP_FILE_WR; 1486 } 1487 } 1488 if (ia_valid & ATTR_MTIME) { 1489 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode, 1490 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 1491 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec); 1492 if (issued & CEPH_CAP_FILE_EXCL) { 1493 ci->i_time_warp_seq++; 1494 inode->i_mtime = attr->ia_mtime; 1495 dirtied |= CEPH_CAP_FILE_EXCL; 1496 } else if ((issued & CEPH_CAP_FILE_WR) && 1497 timespec_compare(&inode->i_mtime, 1498 &attr->ia_mtime) < 0) { 1499 inode->i_mtime = attr->ia_mtime; 1500 dirtied |= CEPH_CAP_FILE_WR; 1501 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1502 !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) { 1503 ceph_encode_timespec(&req->r_args.setattr.mtime, 1504 &attr->ia_mtime); 1505 mask |= CEPH_SETATTR_MTIME; 1506 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | 1507 CEPH_CAP_FILE_WR; 1508 } 1509 } 1510 if (ia_valid & ATTR_SIZE) { 1511 dout("setattr %p size %lld -> %lld\n", inode, 1512 inode->i_size, attr->ia_size); 1513 if (attr->ia_size > inode->i_sb->s_maxbytes) { 1514 err = -EINVAL; 1515 goto out; 1516 } 1517 if ((issued & CEPH_CAP_FILE_EXCL) && 1518 attr->ia_size > inode->i_size) { 1519 inode->i_size = attr->ia_size; 1520 inode->i_blocks = 1521 (attr->ia_size + (1 << 9) - 1) >> 9; 1522 inode->i_ctime = attr->ia_ctime; 1523 ci->i_reported_size = attr->ia_size; 1524 dirtied |= CEPH_CAP_FILE_EXCL; 1525 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1526 attr->ia_size != inode->i_size) { 1527 req->r_args.setattr.size = cpu_to_le64(attr->ia_size); 1528 req->r_args.setattr.old_size = 1529 cpu_to_le64(inode->i_size); 1530 mask |= CEPH_SETATTR_SIZE; 1531 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | 1532 CEPH_CAP_FILE_WR; 1533 } 1534 } 1535 1536 /* these do nothing */ 1537 if (ia_valid & ATTR_CTIME) { 1538 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME| 1539 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0; 1540 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode, 1541 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 1542 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec, 1543 only ? "ctime only" : "ignored"); 1544 inode->i_ctime = attr->ia_ctime; 1545 if (only) { 1546 /* 1547 * if kernel wants to dirty ctime but nothing else, 1548 * we need to choose a cap to dirty under, or do 1549 * a almost-no-op setattr 1550 */ 1551 if (issued & CEPH_CAP_AUTH_EXCL) 1552 dirtied |= CEPH_CAP_AUTH_EXCL; 1553 else if (issued & CEPH_CAP_FILE_EXCL) 1554 dirtied |= CEPH_CAP_FILE_EXCL; 1555 else if (issued & CEPH_CAP_XATTR_EXCL) 1556 dirtied |= CEPH_CAP_XATTR_EXCL; 1557 else 1558 mask |= CEPH_SETATTR_CTIME; 1559 } 1560 } 1561 if (ia_valid & ATTR_FILE) 1562 dout("setattr %p ATTR_FILE ... hrm!\n", inode); 1563 1564 if (dirtied) { 1565 __ceph_mark_dirty_caps(ci, dirtied); 1566 inode->i_ctime = CURRENT_TIME; 1567 } 1568 1569 release &= issued; 1570 spin_unlock(&inode->i_lock); 1571 1572 if (mask) { 1573 req->r_inode = igrab(inode); 1574 req->r_inode_drop = release; 1575 req->r_args.setattr.mask = cpu_to_le32(mask); 1576 req->r_num_caps = 1; 1577 err = ceph_mdsc_do_request(mdsc, parent_inode, req); 1578 } 1579 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err, 1580 ceph_cap_string(dirtied), mask); 1581 1582 ceph_mdsc_put_request(req); 1583 __ceph_do_pending_vmtruncate(inode); 1584 return err; 1585 out: 1586 spin_unlock(&inode->i_lock); 1587 ceph_mdsc_put_request(req); 1588 return err; 1589 } 1590 1591 /* 1592 * Verify that we have a lease on the given mask. If not, 1593 * do a getattr against an mds. 1594 */ 1595 int ceph_do_getattr(struct inode *inode, int mask) 1596 { 1597 struct ceph_client *client = ceph_sb_to_client(inode->i_sb); 1598 struct ceph_mds_client *mdsc = &client->mdsc; 1599 struct ceph_mds_request *req; 1600 int err; 1601 1602 if (ceph_snap(inode) == CEPH_SNAPDIR) { 1603 dout("do_getattr inode %p SNAPDIR\n", inode); 1604 return 0; 1605 } 1606 1607 dout("do_getattr inode %p mask %s\n", inode, ceph_cap_string(mask)); 1608 if (ceph_caps_issued_mask(ceph_inode(inode), mask, 1)) 1609 return 0; 1610 1611 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS); 1612 if (IS_ERR(req)) 1613 return PTR_ERR(req); 1614 req->r_inode = igrab(inode); 1615 req->r_num_caps = 1; 1616 req->r_args.getattr.mask = cpu_to_le32(mask); 1617 err = ceph_mdsc_do_request(mdsc, NULL, req); 1618 ceph_mdsc_put_request(req); 1619 dout("do_getattr result=%d\n", err); 1620 return err; 1621 } 1622 1623 1624 /* 1625 * Check inode permissions. We verify we have a valid value for 1626 * the AUTH cap, then call the generic handler. 1627 */ 1628 int ceph_permission(struct inode *inode, int mask) 1629 { 1630 int err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED); 1631 1632 if (!err) 1633 err = generic_permission(inode, mask, NULL); 1634 return err; 1635 } 1636 1637 /* 1638 * Get all attributes. Hopefully somedata we'll have a statlite() 1639 * and can limit the fields we require to be accurate. 1640 */ 1641 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry, 1642 struct kstat *stat) 1643 { 1644 struct inode *inode = dentry->d_inode; 1645 struct ceph_inode_info *ci = ceph_inode(inode); 1646 int err; 1647 1648 err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL); 1649 if (!err) { 1650 generic_fillattr(inode, stat); 1651 stat->ino = inode->i_ino; 1652 if (ceph_snap(inode) != CEPH_NOSNAP) 1653 stat->dev = ceph_snap(inode); 1654 else 1655 stat->dev = 0; 1656 if (S_ISDIR(inode->i_mode)) { 1657 stat->size = ci->i_rbytes; 1658 stat->blocks = 0; 1659 stat->blksize = 65536; 1660 } 1661 } 1662 return err; 1663 } 1664