1/* 2 * Copyright (c) 2004-2008 Apple Inc. All rights reserved. 3 * 4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ 5 * 6 * This file contains Original Code and/or Modifications of Original Code 7 * as defined in and that are subject to the Apple Public Source License 8 * Version 2.0 (the 'License'). You may not use this file except in 9 * compliance with the License. The rights granted to you under the License 10 * may not be used to create, or enable the creation or redistribution of, 11 * unlawful or unlicensed copies of an Apple operating system, or to 12 * circumvent, violate, or enable the circumvention or violation of, any 13 * terms of an Apple operating system software license agreement. 14 * 15 * Please obtain a copy of the License at 16 * http://www.opensource.apple.com/apsl/ and read it before using this file. 17 * 18 * The Original Code and all software distributed under the License are 19 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER 20 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, 21 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, 22 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. 23 * Please see the License for the specific language governing rights and 24 * limitations under the License. 25 * 26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ 27 */ 28#include <stdarg.h> 29#include <sys/param.h> 30#include <sys/systm.h> 31#include <sys/event.h> // for kqueue related stuff 32#include <sys/fsevents.h> 33 34#if CONFIG_FSE 35#include <sys/namei.h> 36#include <sys/filedesc.h> 37#include <sys/kernel.h> 38#include <sys/file_internal.h> 39#include <sys/stat.h> 40#include <sys/vnode_internal.h> 41#include <sys/mount_internal.h> 42#include <sys/proc_internal.h> 43#include <sys/kauth.h> 44#include <sys/uio.h> 45#include <sys/malloc.h> 46#include <sys/dirent.h> 47#include <sys/attr.h> 48#include <sys/sysctl.h> 49#include <sys/ubc.h> 50#include <machine/cons.h> 51#include <miscfs/specfs/specdev.h> 52#include <miscfs/devfs/devfs.h> 53#include <sys/filio.h> 54#include <kern/locks.h> 55#include <libkern/OSAtomic.h> 56#include <kern/zalloc.h> 57#include <mach/mach_time.h> 58#include <kern/thread_call.h> 59#include <kern/clock.h> 60 61#include <security/audit/audit.h> 62#include <bsm/audit_kevents.h> 63 64#include <pexpert/pexpert.h> 65 66typedef struct kfs_event { 67 LIST_ENTRY(kfs_event) kevent_list; 68 int16_t type; // type code of this event 69 u_int16_t flags, // per-event flags 70 len; // the length of the path in "str" 71 int32_t refcount; // number of clients referencing this 72 pid_t pid; // pid of the process that did the op 73 74 uint64_t abstime; // when this event happened (mach_absolute_time()) 75 ino64_t ino; 76 dev_t dev; 77 int32_t mode; 78 uid_t uid; 79 gid_t gid; 80 81 const char *str; 82 83 struct kfs_event *dest; // if this is a two-file op 84} kfs_event; 85 86// flags for the flags field 87#define KFSE_COMBINED_EVENTS 0x0001 88#define KFSE_CONTAINS_DROPPED_EVENTS 0x0002 89#define KFSE_RECYCLED_EVENT 0x0004 90#define KFSE_BEING_CREATED 0x0008 91 92LIST_HEAD(kfse_list, kfs_event) kfse_list_head = LIST_HEAD_INITIALIZER(x); 93int num_events_outstanding = 0; 94int num_pending_rename = 0; 95 96 97struct fsevent_handle; 98 99typedef struct fs_event_watcher { 100 int8_t *event_list; // the events we're interested in 101 int32_t num_events; 102 dev_t *devices_not_to_watch; // report events from devices not in this list 103 uint32_t num_devices; 104 int32_t flags; 105 kfs_event **event_queue; 106 int32_t eventq_size; // number of event pointers in queue 107 int32_t num_readers; 108 int32_t rd; // read index into the event_queue 109 int32_t wr; // write index into the event_queue 110 int32_t blockers; 111 int32_t my_id; 112 uint32_t num_dropped; 113 uint64_t max_event_id; 114 struct fsevent_handle *fseh; 115 pid_t pid; 116 char proc_name[(2 * MAXCOMLEN) + 1]; 117} fs_event_watcher; 118 119// fs_event_watcher flags 120#define WATCHER_DROPPED_EVENTS 0x0001 121#define WATCHER_CLOSING 0x0002 122#define WATCHER_WANTS_COMPACT_EVENTS 0x0004 123#define WATCHER_WANTS_EXTENDED_INFO 0x0008 124#define WATCHER_APPLE_SYSTEM_SERVICE 0x0010 // fseventsd, coreservicesd, mds 125 126#define MAX_WATCHERS 8 127static fs_event_watcher *watcher_table[MAX_WATCHERS]; 128 129#define DEFAULT_MAX_KFS_EVENTS 4096 130static int max_kfs_events = DEFAULT_MAX_KFS_EVENTS; 131 132// we allocate kfs_event structures out of this zone 133static zone_t event_zone; 134static int fs_event_init = 0; 135 136// 137// this array records whether anyone is interested in a 138// particular type of event. if no one is, we bail out 139// early from the event delivery 140// 141static int16_t fs_event_type_watchers[FSE_MAX_EVENTS]; 142 143static int watcher_add_event(fs_event_watcher *watcher, kfs_event *kfse); 144static void fsevents_wakeup(fs_event_watcher *watcher); 145 146// 147// Locks 148// 149static lck_grp_attr_t * fsevent_group_attr; 150static lck_attr_t * fsevent_lock_attr; 151static lck_grp_t * fsevent_mutex_group; 152 153static lck_grp_t * fsevent_rw_group; 154 155static lck_rw_t event_handling_lock; // handles locking for event manipulation and recycling 156static lck_mtx_t watch_table_lock; 157static lck_mtx_t event_buf_lock; 158static lck_mtx_t event_writer_lock; 159 160 161/* Explicitly declare qsort so compiler doesn't complain */ 162__private_extern__ void qsort( 163 void * array, 164 size_t nmembers, 165 size_t member_size, 166 int (*)(const void *, const void *)); 167 168static int 169is_ignored_directory(const char *path) { 170 171 if (!path) { 172 return 0; 173 } 174 175#define IS_TLD(x) strnstr((char *) path, x, MAXPATHLEN) 176 if (IS_TLD("/.Spotlight-V100/") || 177 IS_TLD("/.MobileBackups/") || 178 IS_TLD("/Backups.backupdb/")) { 179 return 1; 180 } 181#undef IS_TLD 182 183 return 0; 184} 185 186static void 187fsevents_internal_init(void) 188{ 189 int i; 190 191 if (fs_event_init++ != 0) { 192 return; 193 } 194 195 for(i=0; i < FSE_MAX_EVENTS; i++) { 196 fs_event_type_watchers[i] = 0; 197 } 198 199 memset(watcher_table, 0, sizeof(watcher_table)); 200 201 fsevent_lock_attr = lck_attr_alloc_init(); 202 fsevent_group_attr = lck_grp_attr_alloc_init(); 203 fsevent_mutex_group = lck_grp_alloc_init("fsevent-mutex", fsevent_group_attr); 204 fsevent_rw_group = lck_grp_alloc_init("fsevent-rw", fsevent_group_attr); 205 206 lck_mtx_init(&watch_table_lock, fsevent_mutex_group, fsevent_lock_attr); 207 lck_mtx_init(&event_buf_lock, fsevent_mutex_group, fsevent_lock_attr); 208 lck_mtx_init(&event_writer_lock, fsevent_mutex_group, fsevent_lock_attr); 209 210 lck_rw_init(&event_handling_lock, fsevent_rw_group, fsevent_lock_attr); 211 212 PE_get_default("kern.maxkfsevents", &max_kfs_events, sizeof(max_kfs_events)); 213 214 event_zone = zinit(sizeof(kfs_event), 215 max_kfs_events * sizeof(kfs_event), 216 max_kfs_events * sizeof(kfs_event), 217 "fs-event-buf"); 218 if (event_zone == NULL) { 219 printf("fsevents: failed to initialize the event zone.\n"); 220 } 221 222 // mark the zone as exhaustible so that it will not 223 // ever grow beyond what we initially filled it with 224 zone_change(event_zone, Z_EXHAUST, TRUE); 225 zone_change(event_zone, Z_COLLECT, FALSE); 226 zone_change(event_zone, Z_CALLERACCT, FALSE); 227 228 if (zfill(event_zone, max_kfs_events) < max_kfs_events) { 229 printf("fsevents: failed to pre-fill the event zone.\n"); 230 } 231 232} 233 234static void 235lock_watch_table(void) 236{ 237 lck_mtx_lock(&watch_table_lock); 238} 239 240static void 241unlock_watch_table(void) 242{ 243 lck_mtx_unlock(&watch_table_lock); 244} 245 246static void 247lock_fs_event_list(void) 248{ 249 lck_mtx_lock(&event_buf_lock); 250} 251 252static void 253unlock_fs_event_list(void) 254{ 255 lck_mtx_unlock(&event_buf_lock); 256} 257 258// forward prototype 259static void release_event_ref(kfs_event *kfse); 260 261static int 262watcher_cares_about_dev(fs_event_watcher *watcher, dev_t dev) 263{ 264 unsigned int i; 265 266 // if devices_not_to_watch is NULL then we care about all 267 // events from all devices 268 if (watcher->devices_not_to_watch == NULL) { 269 return 1; 270 } 271 272 for(i=0; i < watcher->num_devices; i++) { 273 if (dev == watcher->devices_not_to_watch[i]) { 274 // found a match! that means we do not 275 // want events from this device. 276 return 0; 277 } 278 } 279 280 // if we're here it's not in the devices_not_to_watch[] 281 // list so that means we do care about it 282 return 1; 283} 284 285 286int 287need_fsevent(int type, vnode_t vp) 288{ 289 if (type >= 0 && type < FSE_MAX_EVENTS && fs_event_type_watchers[type] == 0) 290 return (0); 291 292 // events in /dev aren't really interesting... 293 if (vp->v_tag == VT_DEVFS) { 294 return (0); 295 } 296 297 return 1; 298} 299 300 301#define is_throw_away(x) ((x) == FSE_STAT_CHANGED || (x) == FSE_CONTENT_MODIFIED) 302 303 304// Ways that an event can be reused: 305// 306// "combined" events mean that there were two events for 307// the same vnode or path and we're combining both events 308// into a single event. The primary event gets a bit that 309// marks it as having been combined. The secondary event 310// is essentially dropped and the kfse structure reused. 311// 312// "collapsed" means that multiple events below a given 313// directory are collapsed into a single event. in this 314// case, the directory that we collapse into and all of 315// its children must be re-scanned. 316// 317// "recycled" means that we're completely blowing away 318// the event since there are other events that have info 319// about the same vnode or path (and one of those other 320// events will be marked as combined or collapsed as 321// appropriate). 322// 323#define KFSE_COMBINED 0x0001 324#define KFSE_COLLAPSED 0x0002 325#define KFSE_RECYCLED 0x0004 326 327int num_dropped = 0; 328int num_parent_switch = 0; 329int num_recycled_rename = 0; 330 331static struct timeval last_print; 332 333// 334// These variables are used to track coalescing multiple identical 335// events for the same vnode/pathname. If we get the same event 336// type and same vnode/pathname as the previous event, we just drop 337// the event since it's superfluous. This improves some micro- 338// benchmarks considerably and actually has a real-world impact on 339// tests like a Finder copy where multiple stat-changed events can 340// get coalesced. 341// 342static int last_event_type=-1; 343static void *last_ptr=NULL; 344static char last_str[MAXPATHLEN]; 345static int last_nlen=0; 346static int last_vid=-1; 347static uint64_t last_coalesced_time=0; 348static void *last_event_ptr=NULL; 349int last_coalesced = 0; 350static mach_timebase_info_data_t sTimebaseInfo = { 0, 0 }; 351 352 353int 354add_fsevent(int type, vfs_context_t ctx, ...) 355{ 356 struct proc *p = vfs_context_proc(ctx); 357 int i, arg_type, ret; 358 kfs_event *kfse, *kfse_dest=NULL, *cur; 359 fs_event_watcher *watcher; 360 va_list ap; 361 int error = 0, did_alloc=0; 362 dev_t dev = 0; 363 uint64_t now, elapsed; 364 char *pathbuff=NULL; 365 int pathbuff_len; 366 367 368 369 va_start(ap, ctx); 370 371 // ignore bogus event types.. 372 if (type < 0 || type >= FSE_MAX_EVENTS) { 373 return EINVAL; 374 } 375 376 // if no one cares about this type of event, bail out 377 if (fs_event_type_watchers[type] == 0) { 378 va_end(ap); 379 380 return 0; 381 } 382 383 now = mach_absolute_time(); 384 385 // find a free event and snag it for our use 386 // NOTE: do not do anything that would block until 387 // the lock is dropped. 388 lock_fs_event_list(); 389 390 // 391 // check if this event is identical to the previous one... 392 // (as long as it's not an event type that can never be the 393 // same as a previous event) 394 // 395 if (type != FSE_CREATE_FILE && type != FSE_DELETE && type != FSE_RENAME && type != FSE_EXCHANGE && type != FSE_CHOWN && type != FSE_DOCID_CHANGED && type != FSE_DOCID_CREATED) { 396 void *ptr=NULL; 397 int vid=0, was_str=0, nlen=0; 398 399 for(arg_type=va_arg(ap, int32_t); arg_type != FSE_ARG_DONE; arg_type=va_arg(ap, int32_t)) { 400 switch(arg_type) { 401 case FSE_ARG_VNODE: { 402 ptr = va_arg(ap, void *); 403 vid = vnode_vid((struct vnode *)ptr); 404 last_str[0] = '\0'; 405 break; 406 } 407 case FSE_ARG_STRING: { 408 nlen = va_arg(ap, int32_t); 409 ptr = va_arg(ap, void *); 410 was_str = 1; 411 break; 412 } 413 } 414 if (ptr != NULL) { 415 break; 416 } 417 } 418 419 if ( sTimebaseInfo.denom == 0 ) { 420 (void) clock_timebase_info(&sTimebaseInfo); 421 } 422 423 elapsed = (now - last_coalesced_time); 424 if (sTimebaseInfo.denom != sTimebaseInfo.numer) { 425 if (sTimebaseInfo.denom == 1) { 426 elapsed *= sTimebaseInfo.numer; 427 } else { 428 // this could overflow... the worst that will happen is that we'll 429 // send (or not send) an extra event so I'm not going to worry about 430 // doing the math right like dtrace_abs_to_nano() does. 431 elapsed = (elapsed * sTimebaseInfo.numer) / (uint64_t)sTimebaseInfo.denom; 432 } 433 } 434 435 if (type == last_event_type 436 && (elapsed < 1000000000) 437 && 438 ((vid && vid == last_vid && last_ptr == ptr) 439 || 440 (last_str[0] && last_nlen == nlen && ptr && strcmp(last_str, ptr) == 0)) 441 ) { 442 443 last_coalesced++; 444 unlock_fs_event_list(); 445 va_end(ap); 446 447 return 0; 448 } else { 449 last_ptr = ptr; 450 if (was_str) { 451 strlcpy(last_str, ptr, sizeof(last_str)); 452 } 453 last_nlen = nlen; 454 last_vid = vid; 455 last_event_type = type; 456 last_coalesced_time = now; 457 } 458 } 459 va_start(ap, ctx); 460 461 462 kfse = zalloc_noblock(event_zone); 463 if (kfse && (type == FSE_RENAME || type == FSE_EXCHANGE)) { 464 kfse_dest = zalloc_noblock(event_zone); 465 if (kfse_dest == NULL) { 466 did_alloc = 1; 467 zfree(event_zone, kfse); 468 kfse = NULL; 469 } 470 } 471 472 473 if (kfse == NULL) { // yikes! no free events 474 unlock_fs_event_list(); 475 lock_watch_table(); 476 477 for(i=0; i < MAX_WATCHERS; i++) { 478 watcher = watcher_table[i]; 479 if (watcher == NULL) { 480 continue; 481 } 482 483 watcher->flags |= WATCHER_DROPPED_EVENTS; 484 fsevents_wakeup(watcher); 485 } 486 unlock_watch_table(); 487 488 { 489 struct timeval current_tv; 490 491 num_dropped++; 492 493 // only print a message at most once every 5 seconds 494 microuptime(¤t_tv); 495 if ((current_tv.tv_sec - last_print.tv_sec) > 10) { 496 int ii; 497 void *junkptr=zalloc_noblock(event_zone), *listhead=kfse_list_head.lh_first; 498 499 printf("add_fsevent: event queue is full! dropping events (num dropped events: %d; num events outstanding: %d).\n", num_dropped, num_events_outstanding); 500 printf("add_fsevent: kfse_list head %p ; num_pending_rename %d\n", listhead, num_pending_rename); 501 printf("add_fsevent: zalloc sez: %p\n", junkptr); 502 printf("add_fsevent: event_zone info: %d 0x%x\n", ((int *)event_zone)[0], ((int *)event_zone)[1]); 503 for(ii=0; ii < MAX_WATCHERS; ii++) { 504 if (watcher_table[ii] == NULL) { 505 continue; 506 } 507 508 printf("add_fsevent: watcher %s %p: rd %4d wr %4d q_size %4d flags 0x%x\n", 509 watcher_table[ii]->proc_name, 510 watcher_table[ii], 511 watcher_table[ii]->rd, watcher_table[ii]->wr, 512 watcher_table[ii]->eventq_size, watcher_table[ii]->flags); 513 } 514 515 last_print = current_tv; 516 if (junkptr) { 517 zfree(event_zone, junkptr); 518 } 519 } 520 } 521 522 if (pathbuff) { 523 release_pathbuff(pathbuff); 524 pathbuff = NULL; 525 } 526 return ENOSPC; 527 } 528 529 memset(kfse, 0, sizeof(kfs_event)); 530 kfse->refcount = 1; 531 OSBitOrAtomic16(KFSE_BEING_CREATED, &kfse->flags); 532 533 last_event_ptr = kfse; 534 kfse->type = type; 535 kfse->abstime = now; 536 kfse->pid = p->p_pid; 537 if (type == FSE_RENAME || type == FSE_EXCHANGE) { 538 memset(kfse_dest, 0, sizeof(kfs_event)); 539 kfse_dest->refcount = 1; 540 OSBitOrAtomic16(KFSE_BEING_CREATED, &kfse_dest->flags); 541 kfse_dest->type = type; 542 kfse_dest->pid = p->p_pid; 543 kfse_dest->abstime = now; 544 545 kfse->dest = kfse_dest; 546 } 547 548 num_events_outstanding++; 549 if (kfse->type == FSE_RENAME) { 550 num_pending_rename++; 551 } 552 LIST_INSERT_HEAD(&kfse_list_head, kfse, kevent_list); 553 554 if (kfse->refcount < 1) { 555 panic("add_fsevent: line %d: kfse recount %d but should be at least 1\n", __LINE__, kfse->refcount); 556 } 557 558 unlock_fs_event_list(); // at this point it's safe to unlock 559 560 // 561 // now process the arguments passed in and copy them into 562 // the kfse 563 // 564 565 cur = kfse; 566 567 if (type == FSE_DOCID_CREATED || type == FSE_DOCID_CHANGED) { 568 uint64_t val; 569 570 // 571 // These events are special and not like the other events. They only 572 // have a dev_t, src inode #, dest inode #, and a doc-id. We use the 573 // fields that we can in the kfse but have to overlay the dest inode 574 // number and the doc-id on the other fields. 575 // 576 577 // First the dev_t 578 arg_type = va_arg(ap, int32_t); 579 if (arg_type == FSE_ARG_DEV) { 580 cur->dev = (dev_t)(va_arg(ap, dev_t)); 581 } else { 582 cur->dev = (dev_t)0xbadc0de1; 583 } 584 585 // next the source inode # 586 arg_type = va_arg(ap, int32_t); 587 if (arg_type == FSE_ARG_INO) { 588 cur->ino = (ino64_t)(va_arg(ap, ino64_t)); 589 } else { 590 cur->ino = 0xbadc0de2; 591 } 592 593 // now the dest inode # 594 arg_type = va_arg(ap, int32_t); 595 if (arg_type == FSE_ARG_INO) { 596 val = (ino64_t)(va_arg(ap, ino64_t)); 597 } else { 598 val = 0xbadc0de2; 599 } 600 // overlay the dest inode number on the str/dest pointer fields 601 memcpy(&cur->str, &val, sizeof(ino64_t)); 602 603 604 // and last the document-id 605 arg_type = va_arg(ap, int32_t); 606 if (arg_type == FSE_ARG_INT32) { 607 val = (uint64_t)va_arg(ap, uint32_t); 608 } else if (arg_type == FSE_ARG_INT64) { 609 val = (uint64_t)va_arg(ap, uint64_t); 610 } else { 611 val = 0xbadc0de3; 612 } 613 614 // the docid is 64-bit and overlays the uid/gid fields 615 memcpy(&cur->uid, &val, sizeof(uint64_t)); 616 617 goto done_with_args; 618 } 619 620 for(arg_type=va_arg(ap, int32_t); arg_type != FSE_ARG_DONE; arg_type=va_arg(ap, int32_t)) 621 622 switch(arg_type) { 623 case FSE_ARG_VNODE: { 624 // this expands out into multiple arguments to the client 625 struct vnode *vp; 626 struct vnode_attr va; 627 628 if (kfse->str != NULL) { 629 cur = kfse_dest; 630 } 631 632 vp = va_arg(ap, struct vnode *); 633 if (vp == NULL) { 634 panic("add_fsevent: you can't pass me a NULL vnode ptr (type %d)!\n", 635 cur->type); 636 } 637 638 VATTR_INIT(&va); 639 VATTR_WANTED(&va, va_fsid); 640 VATTR_WANTED(&va, va_fileid); 641 VATTR_WANTED(&va, va_mode); 642 VATTR_WANTED(&va, va_uid); 643 VATTR_WANTED(&va, va_gid); 644 if ((ret = vnode_getattr(vp, &va, vfs_context_kernel())) != 0) { 645 // printf("add_fsevent: failed to getattr on vp %p (%d)\n", cur->fref.vp, ret); 646 cur->str = NULL; 647 error = EINVAL; 648 goto clean_up; 649 } 650 651 cur->dev = dev = (dev_t)va.va_fsid; 652 cur->ino = (ino64_t)va.va_fileid; 653 cur->mode = (int32_t)vnode_vttoif(vnode_vtype(vp)) | va.va_mode; 654 cur->uid = va.va_uid; 655 cur->gid = va.va_gid; 656 657 // if we haven't gotten the path yet, get it. 658 if (pathbuff == NULL) { 659 pathbuff = get_pathbuff(); 660 pathbuff_len = MAXPATHLEN; 661 662 pathbuff[0] = '\0'; 663 if ((ret = vn_getpath(vp, pathbuff, &pathbuff_len)) != 0 || pathbuff[0] == '\0') { 664 665 cur->flags |= KFSE_CONTAINS_DROPPED_EVENTS; 666 667 do { 668 if (vp->v_parent != NULL) { 669 vp = vp->v_parent; 670 } else if (vp->v_mount) { 671 strlcpy(pathbuff, vp->v_mount->mnt_vfsstat.f_mntonname, MAXPATHLEN); 672 break; 673 } else { 674 vp = NULL; 675 } 676 677 if (vp == NULL) { 678 break; 679 } 680 681 pathbuff_len = MAXPATHLEN; 682 ret = vn_getpath(vp, pathbuff, &pathbuff_len); 683 } while (ret == ENOSPC); 684 685 if (ret != 0 || vp == NULL) { 686 error = ENOENT; 687 goto clean_up; 688 } 689 } 690 } 691 692 // store the path by adding it to the global string table 693 cur->len = pathbuff_len; 694 cur->str = vfs_addname(pathbuff, pathbuff_len, 0, 0); 695 if (cur->str == NULL || cur->str[0] == '\0') { 696 panic("add_fsevent: was not able to add path %s to event %p.\n", pathbuff, cur); 697 } 698 699 release_pathbuff(pathbuff); 700 pathbuff = NULL; 701 702 break; 703 } 704 705 case FSE_ARG_FINFO: { 706 fse_info *fse; 707 708 fse = va_arg(ap, fse_info *); 709 710 cur->dev = dev = (dev_t)fse->dev; 711 cur->ino = (ino64_t)fse->ino; 712 cur->mode = (int32_t)fse->mode; 713 cur->uid = (uid_t)fse->uid; 714 cur->gid = (uid_t)fse->gid; 715 // if it's a hard-link and this is the last link, flag it 716 if ((fse->mode & FSE_MODE_HLINK) && fse->nlink == 0) { 717 cur->mode |= FSE_MODE_LAST_HLINK; 718 } 719 if (cur->mode & FSE_TRUNCATED_PATH) { 720 cur->flags |= KFSE_CONTAINS_DROPPED_EVENTS; 721 cur->mode &= ~FSE_TRUNCATED_PATH; 722 } 723 break; 724 } 725 726 case FSE_ARG_STRING: 727 if (kfse->str != NULL) { 728 cur = kfse_dest; 729 } 730 731 cur->len = (int16_t)(va_arg(ap, int32_t) & 0x7fff); 732 if (cur->len >= 1) { 733 cur->str = vfs_addname(va_arg(ap, char *), cur->len, 0, 0); 734 } else { 735 printf("add_fsevent: funny looking string length: %d\n", (int)cur->len); 736 cur->len = 2; 737 cur->str = vfs_addname("/", cur->len, 0, 0); 738 } 739 if (cur->str[0] == 0) { 740 printf("add_fsevent: bogus looking string (len %d)\n", cur->len); 741 } 742 break; 743 744 case FSE_ARG_INT32: { 745 uint32_t ival = (uint32_t)va_arg(ap, int32_t); 746 kfse->uid = (ino64_t)ival; 747 break; 748 } 749 750 default: 751 printf("add_fsevent: unknown type %d\n", arg_type); 752 // just skip one 32-bit word and hope we sync up... 753 (void)va_arg(ap, int32_t); 754 } 755 756done_with_args: 757 va_end(ap); 758 759 OSBitAndAtomic16(~KFSE_BEING_CREATED, &kfse->flags); 760 if (kfse_dest) { 761 OSBitAndAtomic16(~KFSE_BEING_CREATED, &kfse_dest->flags); 762 } 763 764 // 765 // now we have to go and let everyone know that 766 // is interested in this type of event 767 // 768 lock_watch_table(); 769 770 for(i=0; i < MAX_WATCHERS; i++) { 771 watcher = watcher_table[i]; 772 if (watcher == NULL) { 773 continue; 774 } 775 776 if ( watcher->event_list[type] == FSE_REPORT 777 && watcher_cares_about_dev(watcher, dev)) { 778 779 if (watcher_add_event(watcher, kfse) != 0) { 780 watcher->num_dropped++; 781 continue; 782 } 783 } 784 785 // if (kfse->refcount < 1) { 786 // panic("add_fsevent: line %d: kfse recount %d but should be at least 1\n", __LINE__, kfse->refcount); 787 // } 788 } 789 790 unlock_watch_table(); 791 792 clean_up: 793 794 if (pathbuff) { 795 release_pathbuff(pathbuff); 796 pathbuff = NULL; 797 } 798 799 release_event_ref(kfse); 800 801 return error; 802} 803 804 805static void 806release_event_ref(kfs_event *kfse) 807{ 808 int old_refcount; 809 kfs_event copy, dest_copy; 810 811 812 old_refcount = OSAddAtomic(-1, &kfse->refcount); 813 if (old_refcount > 1) { 814 return; 815 } 816 817 lock_fs_event_list(); 818 if (last_event_ptr == kfse) { 819 last_event_ptr = NULL; 820 last_event_type = -1; 821 last_coalesced_time = 0; 822 } 823 824 if (kfse->refcount < 0) { 825 panic("release_event_ref: bogus kfse refcount %d\n", kfse->refcount); 826 } 827 828 if (kfse->refcount > 0 || kfse->type == FSE_INVALID) { 829 // This is very subtle. Either of these conditions can 830 // be true if an event got recycled while we were waiting 831 // on the fs_event_list lock or the event got recycled, 832 // delivered, _and_ free'd by someone else while we were 833 // waiting on the fs event list lock. In either case 834 // we need to just unlock the list and return without 835 // doing anything because if the refcount is > 0 then 836 // someone else will take care of free'ing it and when 837 // the kfse->type is invalid then someone else already 838 // has handled free'ing the event (while we were blocked 839 // on the event list lock). 840 // 841 unlock_fs_event_list(); 842 return; 843 } 844 845 // 846 // make a copy of this so we can free things without 847 // holding the fs_event_buf lock 848 // 849 copy = *kfse; 850 if (kfse->dest && OSAddAtomic(-1, &kfse->dest->refcount) == 1) { 851 dest_copy = *kfse->dest; 852 } else { 853 dest_copy.str = NULL; 854 dest_copy.len = 0; 855 dest_copy.type = FSE_INVALID; 856 } 857 858 kfse->pid = kfse->type; // save this off for debugging... 859 kfse->uid = (uid_t)(long)kfse->str; // save this off for debugging... 860 kfse->gid = (gid_t)(long)current_thread(); 861 862 kfse->str = (char *)0xdeadbeef; // XXXdbg - catch any cheaters... 863 864 if (dest_copy.type != FSE_INVALID) { 865 kfse->dest->str = (char *)0xbadc0de; // XXXdbg - catch any cheaters... 866 kfse->dest->type = FSE_INVALID; 867 868 if (kfse->dest->kevent_list.le_prev != NULL) { 869 num_events_outstanding--; 870 LIST_REMOVE(kfse->dest, kevent_list); 871 memset(&kfse->dest->kevent_list, 0xa5, sizeof(kfse->dest->kevent_list)); 872 } 873 874 zfree(event_zone, kfse->dest); 875 } 876 877 // mark this fsevent as invalid 878 { 879 int otype; 880 881 otype = kfse->type; 882 kfse->type = FSE_INVALID; 883 884 if (kfse->kevent_list.le_prev != NULL) { 885 num_events_outstanding--; 886 if (otype == FSE_RENAME) { 887 num_pending_rename--; 888 } 889 LIST_REMOVE(kfse, kevent_list); 890 memset(&kfse->kevent_list, 0, sizeof(kfse->kevent_list)); 891 } 892 } 893 894 zfree(event_zone, kfse); 895 896 unlock_fs_event_list(); 897 898 // if we have a pointer in the union 899 if (copy.str && copy.type != FSE_DOCID_CHANGED) { 900 if (copy.len == 0) { // and it's not a string 901 panic("%s:%d: no more fref.vp!\n", __FILE__, __LINE__); 902 // vnode_rele_ext(copy.fref.vp, O_EVTONLY, 0); 903 } else { // else it's a string 904 vfs_removename(copy.str); 905 } 906 } 907 908 if (dest_copy.type != FSE_INVALID && dest_copy.str) { 909 if (dest_copy.len == 0) { 910 panic("%s:%d: no more fref.vp!\n", __FILE__, __LINE__); 911 // vnode_rele_ext(dest_copy.fref.vp, O_EVTONLY, 0); 912 } else { 913 vfs_removename(dest_copy.str); 914 } 915 } 916} 917 918static int 919add_watcher(int8_t *event_list, int32_t num_events, int32_t eventq_size, fs_event_watcher **watcher_out, void *fseh) 920{ 921 int i; 922 fs_event_watcher *watcher; 923 924 if (eventq_size <= 0 || eventq_size > 100*max_kfs_events) { 925 eventq_size = max_kfs_events; 926 } 927 928 // Note: the event_queue follows the fs_event_watcher struct 929 // in memory so we only have to do one allocation 930 MALLOC(watcher, 931 fs_event_watcher *, 932 sizeof(fs_event_watcher) + eventq_size * sizeof(kfs_event *), 933 M_TEMP, M_WAITOK); 934 if (watcher == NULL) { 935 return ENOMEM; 936 } 937 938 watcher->event_list = event_list; 939 watcher->num_events = num_events; 940 watcher->devices_not_to_watch = NULL; 941 watcher->num_devices = 0; 942 watcher->flags = 0; 943 watcher->event_queue = (kfs_event **)&watcher[1]; 944 watcher->eventq_size = eventq_size; 945 watcher->rd = 0; 946 watcher->wr = 0; 947 watcher->blockers = 0; 948 watcher->num_readers = 0; 949 watcher->max_event_id = 0; 950 watcher->fseh = fseh; 951 watcher->pid = proc_selfpid(); 952 proc_selfname(watcher->proc_name, sizeof(watcher->proc_name)); 953 954 watcher->num_dropped = 0; // XXXdbg - debugging 955 956 if (!strncmp(watcher->proc_name, "fseventsd", sizeof(watcher->proc_name)) || 957 !strncmp(watcher->proc_name, "coreservicesd", sizeof(watcher->proc_name)) || 958 !strncmp(watcher->proc_name, "mds", sizeof(watcher->proc_name))) { 959 watcher->flags |= WATCHER_APPLE_SYSTEM_SERVICE; 960 } else { 961 printf("fsevents: watcher %s (pid: %d) - Using /dev/fsevents directly is unsupported. Migrate to FSEventsFramework\n", 962 watcher->proc_name, watcher->pid); 963 } 964 965 lock_watch_table(); 966 967 // now update the global list of who's interested in 968 // events of a particular type... 969 for(i=0; i < num_events; i++) { 970 if (event_list[i] != FSE_IGNORE && i < FSE_MAX_EVENTS) { 971 fs_event_type_watchers[i]++; 972 } 973 } 974 975 for(i=0; i < MAX_WATCHERS; i++) { 976 if (watcher_table[i] == NULL) { 977 watcher->my_id = i; 978 watcher_table[i] = watcher; 979 break; 980 } 981 } 982 983 if (i > MAX_WATCHERS) { 984 printf("fsevents: too many watchers!\n"); 985 unlock_watch_table(); 986 return ENOSPC; 987 } 988 989 unlock_watch_table(); 990 991 *watcher_out = watcher; 992 993 return 0; 994} 995 996 997 998static void 999remove_watcher(fs_event_watcher *target) 1000{ 1001 int i, j, counter=0; 1002 fs_event_watcher *watcher; 1003 kfs_event *kfse; 1004 1005 lock_watch_table(); 1006 1007 for(j=0; j < MAX_WATCHERS; j++) { 1008 watcher = watcher_table[j]; 1009 if (watcher != target) { 1010 continue; 1011 } 1012 1013 watcher_table[j] = NULL; 1014 1015 for(i=0; i < watcher->num_events; i++) { 1016 if (watcher->event_list[i] != FSE_IGNORE && i < FSE_MAX_EVENTS) { 1017 fs_event_type_watchers[i]--; 1018 } 1019 } 1020 1021 if (watcher->flags & WATCHER_CLOSING) { 1022 unlock_watch_table(); 1023 return; 1024 } 1025 1026 // printf("fsevents: removing watcher %p (rd %d wr %d num_readers %d flags 0x%x)\n", watcher, watcher->rd, watcher->wr, watcher->num_readers, watcher->flags); 1027 watcher->flags |= WATCHER_CLOSING; 1028 OSAddAtomic(1, &watcher->num_readers); 1029 1030 unlock_watch_table(); 1031 1032 while (watcher->num_readers > 1 && counter++ < 5000) { 1033 lock_watch_table(); 1034 fsevents_wakeup(watcher); // in case they're asleep 1035 unlock_watch_table(); 1036 1037 tsleep(watcher, PRIBIO, "fsevents-close", 1); 1038 } 1039 if (counter++ >= 5000) { 1040 // printf("fsevents: close: still have readers! (%d)\n", watcher->num_readers); 1041 panic("fsevents: close: still have readers! (%d)\n", watcher->num_readers); 1042 } 1043 1044 // drain the event_queue 1045 1046 lck_rw_lock_exclusive(&event_handling_lock); 1047 while(watcher->rd != watcher->wr) { 1048 kfse = watcher->event_queue[watcher->rd]; 1049 watcher->event_queue[watcher->rd] = NULL; 1050 watcher->rd = (watcher->rd+1) % watcher->eventq_size; 1051 OSSynchronizeIO(); 1052 if (kfse != NULL && kfse->type != FSE_INVALID && kfse->refcount >= 1) { 1053 release_event_ref(kfse); 1054 } 1055 } 1056 lck_rw_unlock_exclusive(&event_handling_lock); 1057 1058 if (watcher->event_list) { 1059 FREE(watcher->event_list, M_TEMP); 1060 watcher->event_list = NULL; 1061 } 1062 if (watcher->devices_not_to_watch) { 1063 FREE(watcher->devices_not_to_watch, M_TEMP); 1064 watcher->devices_not_to_watch = NULL; 1065 } 1066 FREE(watcher, M_TEMP); 1067 1068 return; 1069 } 1070 1071 unlock_watch_table(); 1072} 1073 1074 1075#define EVENT_DELAY_IN_MS 10 1076static thread_call_t event_delivery_timer = NULL; 1077static int timer_set = 0; 1078 1079 1080static void 1081delayed_event_delivery(__unused void *param0, __unused void *param1) 1082{ 1083 int i; 1084 1085 lock_watch_table(); 1086 1087 for(i=0; i < MAX_WATCHERS; i++) { 1088 if (watcher_table[i] != NULL && watcher_table[i]->rd != watcher_table[i]->wr) { 1089 fsevents_wakeup(watcher_table[i]); 1090 } 1091 } 1092 1093 timer_set = 0; 1094 1095 unlock_watch_table(); 1096} 1097 1098 1099// 1100// The watch table must be locked before calling this function. 1101// 1102static void 1103schedule_event_wakeup(void) 1104{ 1105 uint64_t deadline; 1106 1107 if (event_delivery_timer == NULL) { 1108 event_delivery_timer = thread_call_allocate((thread_call_func_t)delayed_event_delivery, NULL); 1109 } 1110 1111 clock_interval_to_deadline(EVENT_DELAY_IN_MS, 1000 * 1000, &deadline); 1112 1113 thread_call_enter_delayed(event_delivery_timer, deadline); 1114 timer_set = 1; 1115} 1116 1117 1118 1119#define MAX_NUM_PENDING 16 1120 1121// 1122// NOTE: the watch table must be locked before calling 1123// this routine. 1124// 1125static int 1126watcher_add_event(fs_event_watcher *watcher, kfs_event *kfse) 1127{ 1128 if (kfse->abstime > watcher->max_event_id) { 1129 watcher->max_event_id = kfse->abstime; 1130 } 1131 1132 if (((watcher->wr + 1) % watcher->eventq_size) == watcher->rd) { 1133 watcher->flags |= WATCHER_DROPPED_EVENTS; 1134 fsevents_wakeup(watcher); 1135 return ENOSPC; 1136 } 1137 1138 OSAddAtomic(1, &kfse->refcount); 1139 watcher->event_queue[watcher->wr] = kfse; 1140 OSSynchronizeIO(); 1141 watcher->wr = (watcher->wr + 1) % watcher->eventq_size; 1142 1143 // 1144 // wake up the watcher if there are more than MAX_NUM_PENDING events. 1145 // otherwise schedule a timer (if one isn't already set) which will 1146 // send any pending events if no more are received in the next 1147 // EVENT_DELAY_IN_MS milli-seconds. 1148 // 1149 int32_t num_pending = 0; 1150 if (watcher->rd < watcher->wr) { 1151 num_pending = watcher->wr - watcher->rd; 1152 } 1153 1154 if (watcher->rd > watcher->wr) { 1155 num_pending = watcher->wr + watcher->eventq_size - watcher->rd; 1156 } 1157 1158 if (num_pending > (watcher->eventq_size*3/4) && !(watcher->flags & WATCHER_APPLE_SYSTEM_SERVICE)) { 1159 /* Non-Apple Service is falling behind, start dropping events for this process */ 1160 lck_rw_lock_exclusive(&event_handling_lock); 1161 while (watcher->rd != watcher->wr) { 1162 kfse = watcher->event_queue[watcher->rd]; 1163 watcher->event_queue[watcher->rd] = NULL; 1164 watcher->rd = (watcher->rd+1) % watcher->eventq_size; 1165 OSSynchronizeIO(); 1166 if (kfse != NULL && kfse->type != FSE_INVALID && kfse->refcount >= 1) { 1167 release_event_ref(kfse); 1168 } 1169 } 1170 watcher->flags |= WATCHER_DROPPED_EVENTS; 1171 lck_rw_unlock_exclusive(&event_handling_lock); 1172 1173 printf("fsevents: watcher falling behind: %s (pid: %d) rd: %4d wr: %4d q_size: %4d flags: 0x%x\n", 1174 watcher->proc_name, watcher->pid, watcher->rd, watcher->wr, 1175 watcher->eventq_size, watcher->flags); 1176 1177 fsevents_wakeup(watcher); 1178 } else if (num_pending > MAX_NUM_PENDING) { 1179 fsevents_wakeup(watcher); 1180 } else if (timer_set == 0) { 1181 schedule_event_wakeup(); 1182 } 1183 1184 return 0; 1185} 1186 1187static int 1188fill_buff(uint16_t type, int32_t size, const void *data, 1189 char *buff, int32_t *_buff_idx, int32_t buff_sz, 1190 struct uio *uio) 1191{ 1192 int32_t amt, error = 0, buff_idx = *_buff_idx; 1193 uint16_t tmp; 1194 1195 // 1196 // the +1 on the size is to guarantee that the main data 1197 // copy loop will always copy at least 1 byte 1198 // 1199 if ((buff_sz - buff_idx) <= (int)(2*sizeof(uint16_t) + 1)) { 1200 if (buff_idx > uio_resid(uio)) { 1201 error = ENOSPC; 1202 goto get_out; 1203 } 1204 1205 error = uiomove(buff, buff_idx, uio); 1206 if (error) { 1207 goto get_out; 1208 } 1209 buff_idx = 0; 1210 } 1211 1212 // copy out the header (type & size) 1213 memcpy(&buff[buff_idx], &type, sizeof(uint16_t)); 1214 buff_idx += sizeof(uint16_t); 1215 1216 tmp = size & 0xffff; 1217 memcpy(&buff[buff_idx], &tmp, sizeof(uint16_t)); 1218 buff_idx += sizeof(uint16_t); 1219 1220 // now copy the body of the data, flushing along the way 1221 // if the buffer fills up. 1222 // 1223 while(size > 0) { 1224 amt = (size < (buff_sz - buff_idx)) ? size : (buff_sz - buff_idx); 1225 memcpy(&buff[buff_idx], data, amt); 1226 1227 size -= amt; 1228 buff_idx += amt; 1229 data = (const char *)data + amt; 1230 if (size > (buff_sz - buff_idx)) { 1231 if (buff_idx > uio_resid(uio)) { 1232 error = ENOSPC; 1233 goto get_out; 1234 } 1235 error = uiomove(buff, buff_idx, uio); 1236 if (error) { 1237 goto get_out; 1238 } 1239 buff_idx = 0; 1240 } 1241 1242 if (amt == 0) { // just in case... 1243 break; 1244 } 1245 } 1246 1247 get_out: 1248 *_buff_idx = buff_idx; 1249 1250 return error; 1251} 1252 1253 1254static int copy_out_kfse(fs_event_watcher *watcher, kfs_event *kfse, struct uio *uio) __attribute__((noinline)); 1255 1256static int 1257copy_out_kfse(fs_event_watcher *watcher, kfs_event *kfse, struct uio *uio) 1258{ 1259 int error; 1260 uint16_t tmp16; 1261 int32_t type; 1262 kfs_event *cur; 1263 char evbuff[512]; 1264 int evbuff_idx = 0; 1265 1266 if (kfse->type == FSE_INVALID) { 1267 panic("fsevents: copy_out_kfse: asked to copy out an invalid event (kfse %p, refcount %d fref ptr %p)\n", kfse, kfse->refcount, kfse->str); 1268 } 1269 1270 if (kfse->flags & KFSE_BEING_CREATED) { 1271 return 0; 1272 } 1273 1274 if (kfse->type == FSE_RENAME && kfse->dest == NULL) { 1275 // 1276 // This can happen if an event gets recycled but we had a 1277 // pointer to it in our event queue. The event is the 1278 // destination of a rename which we'll process separately 1279 // (that is, another kfse points to this one so it's ok 1280 // to skip this guy because we'll process it when we process 1281 // the other one) 1282 error = 0; 1283 goto get_out; 1284 } 1285 1286 if (watcher->flags & WATCHER_WANTS_EXTENDED_INFO) { 1287 1288 type = (kfse->type & 0xfff); 1289 1290 if (kfse->flags & KFSE_CONTAINS_DROPPED_EVENTS) { 1291 type |= (FSE_CONTAINS_DROPPED_EVENTS << FSE_FLAG_SHIFT); 1292 } else if (kfse->flags & KFSE_COMBINED_EVENTS) { 1293 type |= (FSE_COMBINED_EVENTS << FSE_FLAG_SHIFT); 1294 } 1295 1296 } else { 1297 type = (int32_t)kfse->type; 1298 } 1299 1300 // copy out the type of the event 1301 memcpy(evbuff, &type, sizeof(int32_t)); 1302 evbuff_idx += sizeof(int32_t); 1303 1304 // copy out the pid of the person that generated the event 1305 memcpy(&evbuff[evbuff_idx], &kfse->pid, sizeof(pid_t)); 1306 evbuff_idx += sizeof(pid_t); 1307 1308 cur = kfse; 1309 1310 copy_again: 1311 1312 if (kfse->type == FSE_DOCID_CHANGED || kfse->type == FSE_DOCID_CREATED) { 1313 dev_t dev = cur->dev; 1314 ino_t ino = cur->ino; 1315 uint64_t ival; 1316 1317 error = fill_buff(FSE_ARG_DEV, sizeof(dev_t), &dev, evbuff, &evbuff_idx, sizeof(evbuff), uio); 1318 if (error != 0) { 1319 goto get_out; 1320 } 1321 1322 error = fill_buff(FSE_ARG_INO, sizeof(ino_t), &ino, evbuff, &evbuff_idx, sizeof(evbuff), uio); 1323 if (error != 0) { 1324 goto get_out; 1325 } 1326 1327 memcpy(&ino, &cur->str, sizeof(ino_t)); 1328 error = fill_buff(FSE_ARG_INO, sizeof(ino_t), &ino, evbuff, &evbuff_idx, sizeof(evbuff), uio); 1329 if (error != 0) { 1330 goto get_out; 1331 } 1332 1333 memcpy(&ival, &cur->uid, sizeof(uint64_t)); // the docid gets stuffed into the ino field 1334 error = fill_buff(FSE_ARG_INT64, sizeof(uint64_t), &ival, evbuff, &evbuff_idx, sizeof(evbuff), uio); 1335 if (error != 0) { 1336 goto get_out; 1337 } 1338 1339 goto done; 1340 } 1341 1342 if (cur->str == NULL || cur->str[0] == '\0') { 1343 printf("copy_out_kfse:2: empty/short path (%s)\n", cur->str); 1344 error = fill_buff(FSE_ARG_STRING, 2, "/", evbuff, &evbuff_idx, sizeof(evbuff), uio); 1345 } else { 1346 error = fill_buff(FSE_ARG_STRING, cur->len, cur->str, evbuff, &evbuff_idx, sizeof(evbuff), uio); 1347 } 1348 if (error != 0) { 1349 goto get_out; 1350 } 1351 1352 if (cur->dev == 0 && cur->ino == 0) { 1353 // this happens when a rename event happens and the 1354 // destination of the rename did not previously exist. 1355 // it thus has no other file info so skip copying out 1356 // the stuff below since it isn't initialized 1357 goto done; 1358 } 1359 1360 1361 if (watcher->flags & WATCHER_WANTS_COMPACT_EVENTS) { 1362 int32_t finfo_size; 1363 1364 finfo_size = sizeof(dev_t) + sizeof(ino64_t) + sizeof(int32_t) + sizeof(uid_t) + sizeof(gid_t); 1365 error = fill_buff(FSE_ARG_FINFO, finfo_size, &cur->ino, evbuff, &evbuff_idx, sizeof(evbuff), uio); 1366 if (error != 0) { 1367 goto get_out; 1368 } 1369 } else { 1370 ino_t ino; 1371 1372 error = fill_buff(FSE_ARG_DEV, sizeof(dev_t), &cur->dev, evbuff, &evbuff_idx, sizeof(evbuff), uio); 1373 if (error != 0) { 1374 goto get_out; 1375 } 1376 1377 ino = (ino_t)cur->ino; 1378 error = fill_buff(FSE_ARG_INO, sizeof(ino_t), &ino, evbuff, &evbuff_idx, sizeof(evbuff), uio); 1379 if (error != 0) { 1380 goto get_out; 1381 } 1382 1383 error = fill_buff(FSE_ARG_MODE, sizeof(int32_t), &cur->mode, evbuff, &evbuff_idx, sizeof(evbuff), uio); 1384 if (error != 0) { 1385 goto get_out; 1386 } 1387 1388 error = fill_buff(FSE_ARG_UID, sizeof(uid_t), &cur->uid, evbuff, &evbuff_idx, sizeof(evbuff), uio); 1389 if (error != 0) { 1390 goto get_out; 1391 } 1392 1393 error = fill_buff(FSE_ARG_GID, sizeof(gid_t), &cur->gid, evbuff, &evbuff_idx, sizeof(evbuff), uio); 1394 if (error != 0) { 1395 goto get_out; 1396 } 1397 } 1398 1399 1400 if (cur->dest) { 1401 cur = cur->dest; 1402 goto copy_again; 1403 } 1404 1405 done: 1406 // very last thing: the time stamp 1407 error = fill_buff(FSE_ARG_INT64, sizeof(uint64_t), &cur->abstime, evbuff, &evbuff_idx, sizeof(evbuff), uio); 1408 if (error != 0) { 1409 goto get_out; 1410 } 1411 1412 // check if the FSE_ARG_DONE will fit 1413 if (sizeof(uint16_t) > sizeof(evbuff) - evbuff_idx) { 1414 if (evbuff_idx > uio_resid(uio)) { 1415 error = ENOSPC; 1416 goto get_out; 1417 } 1418 error = uiomove(evbuff, evbuff_idx, uio); 1419 if (error) { 1420 goto get_out; 1421 } 1422 evbuff_idx = 0; 1423 } 1424 1425 tmp16 = FSE_ARG_DONE; 1426 memcpy(&evbuff[evbuff_idx], &tmp16, sizeof(uint16_t)); 1427 evbuff_idx += sizeof(uint16_t); 1428 1429 // flush any remaining data in the buffer (and hopefully 1430 // in most cases this is the only uiomove we'll do) 1431 if (evbuff_idx > uio_resid(uio)) { 1432 error = ENOSPC; 1433 } else { 1434 error = uiomove(evbuff, evbuff_idx, uio); 1435 } 1436 1437 get_out: 1438 1439 return error; 1440} 1441 1442 1443 1444static int 1445fmod_watch(fs_event_watcher *watcher, struct uio *uio) 1446{ 1447 int error=0; 1448 user_ssize_t last_full_event_resid; 1449 kfs_event *kfse; 1450 uint16_t tmp16; 1451 int skipped; 1452 1453 last_full_event_resid = uio_resid(uio); 1454 1455 // need at least 2048 bytes of space (maxpathlen + 1 event buf) 1456 if (uio_resid(uio) < 2048 || watcher == NULL) { 1457 return EINVAL; 1458 } 1459 1460 if (watcher->flags & WATCHER_CLOSING) { 1461 return 0; 1462 } 1463 1464 if (OSAddAtomic(1, &watcher->num_readers) != 0) { 1465 // don't allow multiple threads to read from the fd at the same time 1466 OSAddAtomic(-1, &watcher->num_readers); 1467 return EAGAIN; 1468 } 1469 1470 restart_watch: 1471 if (watcher->rd == watcher->wr) { 1472 if (watcher->flags & WATCHER_CLOSING) { 1473 OSAddAtomic(-1, &watcher->num_readers); 1474 return 0; 1475 } 1476 OSAddAtomic(1, &watcher->blockers); 1477 1478 // there's nothing to do, go to sleep 1479 error = tsleep((caddr_t)watcher, PUSER|PCATCH, "fsevents_empty", 0); 1480 1481 OSAddAtomic(-1, &watcher->blockers); 1482 1483 if (error != 0 || (watcher->flags & WATCHER_CLOSING)) { 1484 OSAddAtomic(-1, &watcher->num_readers); 1485 return error; 1486 } 1487 } 1488 1489 // if we dropped events, return that as an event first 1490 if (watcher->flags & WATCHER_DROPPED_EVENTS) { 1491 int32_t val = FSE_EVENTS_DROPPED; 1492 1493 error = uiomove((caddr_t)&val, sizeof(int32_t), uio); 1494 if (error == 0) { 1495 val = 0; // a fake pid 1496 error = uiomove((caddr_t)&val, sizeof(int32_t), uio); 1497 1498 tmp16 = FSE_ARG_DONE; // makes it a consistent msg 1499 error = uiomove((caddr_t)&tmp16, sizeof(int16_t), uio); 1500 1501 last_full_event_resid = uio_resid(uio); 1502 } 1503 1504 if (error) { 1505 OSAddAtomic(-1, &watcher->num_readers); 1506 return error; 1507 } 1508 1509 watcher->flags &= ~WATCHER_DROPPED_EVENTS; 1510 } 1511 1512 skipped = 0; 1513 1514 lck_rw_lock_shared(&event_handling_lock); 1515 while (uio_resid(uio) > 0 && watcher->rd != watcher->wr) { 1516 if (watcher->flags & WATCHER_CLOSING) { 1517 break; 1518 } 1519 1520 // 1521 // check if the event is something of interest to us 1522 // (since it may have been recycled/reused and changed 1523 // its type or which device it is for) 1524 // 1525 kfse = watcher->event_queue[watcher->rd]; 1526 if (!kfse || kfse->type == FSE_INVALID || kfse->refcount < 1) { 1527 break; 1528 } 1529 1530 if (watcher->event_list[kfse->type] == FSE_REPORT && watcher_cares_about_dev(watcher, kfse->dev)) { 1531 1532 if (!(watcher->flags & WATCHER_APPLE_SYSTEM_SERVICE) && kfse->type != FSE_DOCID_CHANGED && is_ignored_directory(kfse->str)) { 1533 // If this is not an Apple System Service, skip specified directories 1534 // radar://12034844 1535 error = 0; 1536 skipped = 1; 1537 } else { 1538 1539 skipped = 0; 1540 if (last_event_ptr == kfse) { 1541 last_event_ptr = NULL; 1542 last_event_type = -1; 1543 last_coalesced_time = 0; 1544 } 1545 error = copy_out_kfse(watcher, kfse, uio); 1546 if (error != 0) { 1547 // if an event won't fit or encountered an error while 1548 // we were copying it out, then backup to the last full 1549 // event and just bail out. if the error was ENOENT 1550 // then we can continue regular processing, otherwise 1551 // we should unlock things and return. 1552 uio_setresid(uio, last_full_event_resid); 1553 if (error != ENOENT) { 1554 lck_rw_unlock_shared(&event_handling_lock); 1555 error = 0; 1556 goto get_out; 1557 } 1558 } 1559 1560 last_full_event_resid = uio_resid(uio); 1561 } 1562 } 1563 1564 watcher->event_queue[watcher->rd] = NULL; 1565 watcher->rd = (watcher->rd + 1) % watcher->eventq_size; 1566 OSSynchronizeIO(); 1567 release_event_ref(kfse); 1568 } 1569 lck_rw_unlock_shared(&event_handling_lock); 1570 1571 if (skipped && error == 0) { 1572 goto restart_watch; 1573 } 1574 1575 get_out: 1576 OSAddAtomic(-1, &watcher->num_readers); 1577 1578 return error; 1579} 1580 1581 1582// release any references we might have on vnodes which are 1583// the mount point passed to us (so that it can be cleanly 1584// unmounted). 1585// 1586// since we don't want to lose the events we'll convert the 1587// vnode refs to full paths. 1588// 1589void 1590fsevent_unmount(__unused struct mount *mp) 1591{ 1592 // we no longer maintain pointers to vnodes so 1593 // there is nothing to do... 1594} 1595 1596 1597// 1598// /dev/fsevents device code 1599// 1600static int fsevents_installed = 0; 1601 1602typedef struct fsevent_handle { 1603 UInt32 flags; 1604 SInt32 active; 1605 fs_event_watcher *watcher; 1606 struct klist knotes; 1607 struct selinfo si; 1608} fsevent_handle; 1609 1610#define FSEH_CLOSING 0x0001 1611 1612static int 1613fseventsf_read(struct fileproc *fp, struct uio *uio, 1614 __unused int flags, __unused vfs_context_t ctx) 1615{ 1616 fsevent_handle *fseh = (struct fsevent_handle *)fp->f_fglob->fg_data; 1617 int error; 1618 1619 error = fmod_watch(fseh->watcher, uio); 1620 1621 return error; 1622} 1623 1624 1625static int 1626fseventsf_write(__unused struct fileproc *fp, __unused struct uio *uio, 1627 __unused int flags, __unused vfs_context_t ctx) 1628{ 1629 return EIO; 1630} 1631 1632#pragma pack(push, 4) 1633typedef struct ext_fsevent_dev_filter_args { 1634 uint32_t num_devices; 1635 user_addr_t devices; 1636} ext_fsevent_dev_filter_args; 1637#pragma pack(pop) 1638 1639#define NEW_FSEVENTS_DEVICE_FILTER _IOW('s', 100, ext_fsevent_dev_filter_args) 1640 1641typedef struct old_fsevent_dev_filter_args { 1642 uint32_t num_devices; 1643 int32_t devices; 1644} old_fsevent_dev_filter_args; 1645 1646#define OLD_FSEVENTS_DEVICE_FILTER _IOW('s', 100, old_fsevent_dev_filter_args) 1647 1648#if __LP64__ 1649/* need this in spite of the padding due to alignment of devices */ 1650typedef struct fsevent_dev_filter_args32 { 1651 uint32_t num_devices; 1652 uint32_t devices; 1653 int32_t pad1; 1654} fsevent_dev_filter_args32; 1655#endif 1656 1657static int 1658fseventsf_ioctl(struct fileproc *fp, u_long cmd, caddr_t data, vfs_context_t ctx) 1659{ 1660 fsevent_handle *fseh = (struct fsevent_handle *)fp->f_fglob->fg_data; 1661 int ret = 0; 1662 ext_fsevent_dev_filter_args *devfilt_args, _devfilt_args; 1663 1664 if (proc_is64bit(vfs_context_proc(ctx))) { 1665 devfilt_args = (ext_fsevent_dev_filter_args *)data; 1666 } 1667 else if (cmd == OLD_FSEVENTS_DEVICE_FILTER) { 1668 old_fsevent_dev_filter_args *udev_filt_args = (old_fsevent_dev_filter_args *)data; 1669 1670 devfilt_args = &_devfilt_args; 1671 memset(devfilt_args, 0, sizeof(ext_fsevent_dev_filter_args)); 1672 1673 devfilt_args->num_devices = udev_filt_args->num_devices; 1674 devfilt_args->devices = CAST_USER_ADDR_T(udev_filt_args->devices); 1675 } 1676 else { 1677#if __LP64__ 1678 fsevent_dev_filter_args32 *udev_filt_args = (fsevent_dev_filter_args32 *)data; 1679#else 1680 fsevent_dev_filter_args *udev_filt_args = (fsevent_dev_filter_args *)data; 1681#endif 1682 1683 devfilt_args = &_devfilt_args; 1684 memset(devfilt_args, 0, sizeof(ext_fsevent_dev_filter_args)); 1685 1686 devfilt_args->num_devices = udev_filt_args->num_devices; 1687 devfilt_args->devices = CAST_USER_ADDR_T(udev_filt_args->devices); 1688 } 1689 1690 OSAddAtomic(1, &fseh->active); 1691 if (fseh->flags & FSEH_CLOSING) { 1692 OSAddAtomic(-1, &fseh->active); 1693 return 0; 1694 } 1695 1696 switch (cmd) { 1697 case FIONBIO: 1698 case FIOASYNC: 1699 break; 1700 1701 case FSEVENTS_WANT_COMPACT_EVENTS: { 1702 fseh->watcher->flags |= WATCHER_WANTS_COMPACT_EVENTS; 1703 break; 1704 } 1705 1706 case FSEVENTS_WANT_EXTENDED_INFO: { 1707 fseh->watcher->flags |= WATCHER_WANTS_EXTENDED_INFO; 1708 break; 1709 } 1710 1711 case FSEVENTS_GET_CURRENT_ID: { 1712 *(uint64_t *)data = fseh->watcher->max_event_id; 1713 ret = 0; 1714 break; 1715 } 1716 1717 case OLD_FSEVENTS_DEVICE_FILTER: 1718 case NEW_FSEVENTS_DEVICE_FILTER: { 1719 int new_num_devices; 1720 dev_t *devices_not_to_watch, *tmp=NULL; 1721 1722 if (devfilt_args->num_devices > 256) { 1723 ret = EINVAL; 1724 break; 1725 } 1726 1727 new_num_devices = devfilt_args->num_devices; 1728 if (new_num_devices == 0) { 1729 tmp = fseh->watcher->devices_not_to_watch; 1730 1731 lock_watch_table(); 1732 fseh->watcher->devices_not_to_watch = NULL; 1733 fseh->watcher->num_devices = new_num_devices; 1734 unlock_watch_table(); 1735 1736 if (tmp) { 1737 FREE(tmp, M_TEMP); 1738 } 1739 break; 1740 } 1741 1742 MALLOC(devices_not_to_watch, dev_t *, 1743 new_num_devices * sizeof(dev_t), 1744 M_TEMP, M_WAITOK); 1745 if (devices_not_to_watch == NULL) { 1746 ret = ENOMEM; 1747 break; 1748 } 1749 1750 ret = copyin(devfilt_args->devices, 1751 (void *)devices_not_to_watch, 1752 new_num_devices * sizeof(dev_t)); 1753 if (ret) { 1754 FREE(devices_not_to_watch, M_TEMP); 1755 break; 1756 } 1757 1758 lock_watch_table(); 1759 fseh->watcher->num_devices = new_num_devices; 1760 tmp = fseh->watcher->devices_not_to_watch; 1761 fseh->watcher->devices_not_to_watch = devices_not_to_watch; 1762 unlock_watch_table(); 1763 1764 if (tmp) { 1765 FREE(tmp, M_TEMP); 1766 } 1767 1768 break; 1769 } 1770 1771 default: 1772 ret = EINVAL; 1773 break; 1774 } 1775 1776 OSAddAtomic(-1, &fseh->active); 1777 return (ret); 1778} 1779 1780 1781static int 1782fseventsf_select(struct fileproc *fp, int which, __unused void *wql, vfs_context_t ctx) 1783{ 1784 fsevent_handle *fseh = (struct fsevent_handle *)fp->f_fglob->fg_data; 1785 int ready = 0; 1786 1787 if ((which != FREAD) || (fseh->watcher->flags & WATCHER_CLOSING)) { 1788 return 0; 1789 } 1790 1791 1792 // if there's nothing in the queue, we're not ready 1793 if (fseh->watcher->rd != fseh->watcher->wr) { 1794 ready = 1; 1795 } 1796 1797 if (!ready) { 1798 selrecord(vfs_context_proc(ctx), &fseh->si, wql); 1799 } 1800 1801 return ready; 1802} 1803 1804 1805#if NOTUSED 1806static int 1807fseventsf_stat(__unused struct fileproc *fp, __unused struct stat *sb, __unused vfs_context_t ctx) 1808{ 1809 return ENOTSUP; 1810} 1811#endif 1812 1813static int 1814fseventsf_close(struct fileglob *fg, __unused vfs_context_t ctx) 1815{ 1816 fsevent_handle *fseh = (struct fsevent_handle *)fg->fg_data; 1817 fs_event_watcher *watcher; 1818 1819 OSBitOrAtomic(FSEH_CLOSING, &fseh->flags); 1820 while (OSAddAtomic(0, &fseh->active) > 0) { 1821 tsleep((caddr_t)fseh->watcher, PRIBIO, "fsevents-close", 1); 1822 } 1823 1824 watcher = fseh->watcher; 1825 fg->fg_data = NULL; 1826 fseh->watcher = NULL; 1827 1828 remove_watcher(watcher); 1829 FREE(fseh, M_TEMP); 1830 1831 return 0; 1832} 1833 1834static void 1835filt_fsevent_detach(struct knote *kn) 1836{ 1837 fsevent_handle *fseh = (struct fsevent_handle *)kn->kn_hook; 1838 1839 lock_watch_table(); 1840 1841 KNOTE_DETACH(&fseh->knotes, kn); 1842 1843 unlock_watch_table(); 1844} 1845 1846/* 1847 * Determine whether this knote should be active 1848 * 1849 * This is kind of subtle. 1850 * --First, notice if the vnode has been revoked: in so, override hint 1851 * --EVFILT_READ knotes are checked no matter what the hint is 1852 * --Other knotes activate based on hint. 1853 * --If hint is revoke, set special flags and activate 1854 */ 1855static int 1856filt_fsevent(struct knote *kn, long hint) 1857{ 1858 fsevent_handle *fseh = (struct fsevent_handle *)kn->kn_hook; 1859 int activate = 0; 1860 int32_t rd, wr, amt; 1861 1862 if (NOTE_REVOKE == hint) { 1863 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 1864 activate = 1; 1865 } 1866 1867 rd = fseh->watcher->rd; 1868 wr = fseh->watcher->wr; 1869 if (rd <= wr) { 1870 amt = wr - rd; 1871 } else { 1872 amt = fseh->watcher->eventq_size - (rd - wr); 1873 } 1874 1875 switch(kn->kn_filter) { 1876 case EVFILT_READ: 1877 kn->kn_data = amt; 1878 1879 if (kn->kn_data != 0) { 1880 activate = 1; 1881 } 1882 break; 1883 case EVFILT_VNODE: 1884 /* Check events this note matches against the hint */ 1885 if (kn->kn_sfflags & hint) { 1886 kn->kn_fflags |= hint; /* Set which event occurred */ 1887 } 1888 if (kn->kn_fflags != 0) { 1889 activate = 1; 1890 } 1891 break; 1892 default: { 1893 // nothing to do... 1894 break; 1895 } 1896 } 1897 1898 return (activate); 1899} 1900 1901 1902struct filterops fsevent_filtops = { 1903 .f_isfd = 1, 1904 .f_attach = NULL, 1905 .f_detach = filt_fsevent_detach, 1906 .f_event = filt_fsevent 1907}; 1908 1909static int 1910fseventsf_kqfilter(__unused struct fileproc *fp, __unused struct knote *kn, __unused vfs_context_t ctx) 1911{ 1912 fsevent_handle *fseh = (struct fsevent_handle *)fp->f_fglob->fg_data; 1913 1914 kn->kn_hook = (void*)fseh; 1915 kn->kn_hookid = 1; 1916 kn->kn_fop = &fsevent_filtops; 1917 1918 lock_watch_table(); 1919 1920 KNOTE_ATTACH(&fseh->knotes, kn); 1921 1922 unlock_watch_table(); 1923 return 0; 1924} 1925 1926 1927static int 1928fseventsf_drain(struct fileproc *fp, __unused vfs_context_t ctx) 1929{ 1930 int counter = 0; 1931 fsevent_handle *fseh = (struct fsevent_handle *)fp->f_fglob->fg_data; 1932 1933 fseh->watcher->flags |= WATCHER_CLOSING; 1934 1935 // if there are people still waiting, sleep for 10ms to 1936 // let them clean up and get out of there. however we 1937 // also don't want to get stuck forever so if they don't 1938 // exit after 5 seconds we're tearing things down anyway. 1939 while(fseh->watcher->blockers && counter++ < 500) { 1940 // issue wakeup in case anyone is blocked waiting for an event 1941 // do this each time we wakeup in case the blocker missed 1942 // the wakeup due to the unprotected test of WATCHER_CLOSING 1943 // and decision to tsleep in fmod_watch... this bit of 1944 // latency is a decent tradeoff against not having to 1945 // take and drop a lock in fmod_watch 1946 lock_watch_table(); 1947 fsevents_wakeup(fseh->watcher); 1948 unlock_watch_table(); 1949 1950 tsleep((caddr_t)fseh->watcher, PRIBIO, "watcher-close", 1); 1951 } 1952 1953 return 0; 1954} 1955 1956 1957static int 1958fseventsopen(__unused dev_t dev, __unused int flag, __unused int mode, __unused struct proc *p) 1959{ 1960 if (!kauth_cred_issuser(kauth_cred_get())) { 1961 return EPERM; 1962 } 1963 1964 return 0; 1965} 1966 1967static int 1968fseventsclose(__unused dev_t dev, __unused int flag, __unused int mode, __unused struct proc *p) 1969{ 1970 return 0; 1971} 1972 1973static int 1974fseventsread(__unused dev_t dev, __unused struct uio *uio, __unused int ioflag) 1975{ 1976 return EIO; 1977} 1978 1979 1980static int 1981parse_buffer_and_add_events(const char *buffer, int bufsize, vfs_context_t ctx, long *remainder) 1982{ 1983 const fse_info *finfo, *dest_finfo; 1984 const char *path, *ptr, *dest_path, *event_start=buffer; 1985 int path_len, type, dest_path_len, err = 0; 1986 1987 1988 ptr = buffer; 1989 while ((ptr+sizeof(int)+sizeof(fse_info)+1) < buffer+bufsize) { 1990 type = *(const int *)ptr; 1991 if (type < 0 || type >= FSE_MAX_EVENTS) { 1992 err = EINVAL; 1993 break; 1994 } 1995 1996 ptr += sizeof(int); 1997 1998 finfo = (const fse_info *)ptr; 1999 ptr += sizeof(fse_info); 2000 2001 path = ptr; 2002 while(ptr < buffer+bufsize && *ptr != '\0') { 2003 ptr++; 2004 } 2005 2006 if (ptr >= buffer+bufsize) { 2007 break; 2008 } 2009 2010 ptr++; // advance over the trailing '\0' 2011 2012 path_len = ptr - path; 2013 2014 if (type != FSE_RENAME && type != FSE_EXCHANGE) { 2015 event_start = ptr; // record where the next event starts 2016 2017 err = add_fsevent(type, ctx, FSE_ARG_STRING, path_len, path, FSE_ARG_FINFO, finfo, FSE_ARG_DONE); 2018 if (err) { 2019 break; 2020 } 2021 continue; 2022 } 2023 2024 // 2025 // if we're here we have to slurp up the destination finfo 2026 // and path so that we can pass them to the add_fsevent() 2027 // call. basically it's a copy of the above code. 2028 // 2029 dest_finfo = (const fse_info *)ptr; 2030 ptr += sizeof(fse_info); 2031 2032 dest_path = ptr; 2033 while(ptr < buffer+bufsize && *ptr != '\0') { 2034 ptr++; 2035 } 2036 2037 if (ptr >= buffer+bufsize) { 2038 break; 2039 } 2040 2041 ptr++; // advance over the trailing '\0' 2042 event_start = ptr; // record where the next event starts 2043 2044 dest_path_len = ptr - dest_path; 2045 // 2046 // If the destination inode number is non-zero, generate a rename 2047 // with both source and destination FSE_ARG_FINFO. Otherwise generate 2048 // a rename with only one FSE_ARG_FINFO. If you need to inject an 2049 // exchange with an inode of zero, just make that inode (and its path) 2050 // come in as the first one, not the second. 2051 // 2052 if (dest_finfo->ino) { 2053 err = add_fsevent(type, ctx, 2054 FSE_ARG_STRING, path_len, path, FSE_ARG_FINFO, finfo, 2055 FSE_ARG_STRING, dest_path_len, dest_path, FSE_ARG_FINFO, dest_finfo, 2056 FSE_ARG_DONE); 2057 } else { 2058 err = add_fsevent(type, ctx, 2059 FSE_ARG_STRING, path_len, path, FSE_ARG_FINFO, finfo, 2060 FSE_ARG_STRING, dest_path_len, dest_path, 2061 FSE_ARG_DONE); 2062 } 2063 2064 if (err) { 2065 break; 2066 } 2067 2068 } 2069 2070 // if the last event wasn't complete, set the remainder 2071 // to be the last event start boundary. 2072 // 2073 *remainder = (long)((buffer+bufsize) - event_start); 2074 2075 return err; 2076} 2077 2078 2079// 2080// Note: this buffer size can not ever be less than 2081// 2*MAXPATHLEN + 2*sizeof(fse_info) + sizeof(int) 2082// because that is the max size for a single event. 2083// I made it 4k to be a "nice" size. making it 2084// smaller is not a good idea. 2085// 2086#define WRITE_BUFFER_SIZE 4096 2087char *write_buffer=NULL; 2088 2089static int 2090fseventswrite(__unused dev_t dev, struct uio *uio, __unused int ioflag) 2091{ 2092 int error=0, count; 2093 vfs_context_t ctx = vfs_context_current(); 2094 long offset=0, remainder; 2095 2096 lck_mtx_lock(&event_writer_lock); 2097 2098 if (write_buffer == NULL) { 2099 if (kmem_alloc(kernel_map, (vm_offset_t *)&write_buffer, WRITE_BUFFER_SIZE)) { 2100 lck_mtx_unlock(&event_writer_lock); 2101 return ENOMEM; 2102 } 2103 } 2104 2105 // 2106 // this loop copies in and processes the events written. 2107 // it takes care to copy in reasonable size chunks and 2108 // process them. if there is an event that spans a chunk 2109 // boundary we're careful to copy those bytes down to the 2110 // beginning of the buffer and read the next chunk in just 2111 // after it. 2112 // 2113 while(uio_resid(uio)) { 2114 if (uio_resid(uio) > (WRITE_BUFFER_SIZE-offset)) { 2115 count = WRITE_BUFFER_SIZE - offset; 2116 } else { 2117 count = uio_resid(uio); 2118 } 2119 2120 error = uiomove(write_buffer+offset, count, uio); 2121 if (error) { 2122 break; 2123 } 2124 2125 // printf("fsevents: write: copied in %d bytes (offset: %ld)\n", count, offset); 2126 error = parse_buffer_and_add_events(write_buffer, offset+count, ctx, &remainder); 2127 if (error) { 2128 break; 2129 } 2130 2131 // 2132 // if there's any remainder, copy it down to the beginning 2133 // of the buffer so that it will get processed the next time 2134 // through the loop. note that the remainder always starts 2135 // at an event boundary. 2136 // 2137 if (remainder != 0) { 2138 // printf("fsevents: write: an event spanned a %d byte boundary. remainder: %ld\n", 2139 // WRITE_BUFFER_SIZE, remainder); 2140 memmove(write_buffer, (write_buffer+count+offset) - remainder, remainder); 2141 offset = remainder; 2142 } else { 2143 offset = 0; 2144 } 2145 } 2146 2147 lck_mtx_unlock(&event_writer_lock); 2148 2149 return error; 2150} 2151 2152 2153static const struct fileops fsevents_fops = { 2154 DTYPE_FSEVENTS, 2155 fseventsf_read, 2156 fseventsf_write, 2157 fseventsf_ioctl, 2158 fseventsf_select, 2159 fseventsf_close, 2160 fseventsf_kqfilter, 2161 fseventsf_drain 2162}; 2163 2164typedef struct ext_fsevent_clone_args { 2165 user_addr_t event_list; 2166 int32_t num_events; 2167 int32_t event_queue_depth; 2168 user_addr_t fd; 2169} ext_fsevent_clone_args; 2170 2171typedef struct old_fsevent_clone_args { 2172 uint32_t event_list; 2173 int32_t num_events; 2174 int32_t event_queue_depth; 2175 uint32_t fd; 2176} old_fsevent_clone_args; 2177 2178#define OLD_FSEVENTS_CLONE _IOW('s', 1, old_fsevent_clone_args) 2179 2180static int 2181fseventsioctl(__unused dev_t dev, u_long cmd, caddr_t data, __unused int flag, struct proc *p) 2182{ 2183 struct fileproc *f; 2184 int fd, error; 2185 fsevent_handle *fseh = NULL; 2186 ext_fsevent_clone_args *fse_clone_args, _fse_clone; 2187 int8_t *event_list; 2188 int is64bit = proc_is64bit(p); 2189 2190 switch (cmd) { 2191 case OLD_FSEVENTS_CLONE: { 2192 old_fsevent_clone_args *old_args = (old_fsevent_clone_args *)data; 2193 2194 fse_clone_args = &_fse_clone; 2195 memset(fse_clone_args, 0, sizeof(ext_fsevent_clone_args)); 2196 2197 fse_clone_args->event_list = CAST_USER_ADDR_T(old_args->event_list); 2198 fse_clone_args->num_events = old_args->num_events; 2199 fse_clone_args->event_queue_depth = old_args->event_queue_depth; 2200 fse_clone_args->fd = CAST_USER_ADDR_T(old_args->fd); 2201 goto handle_clone; 2202 } 2203 2204 case FSEVENTS_CLONE: 2205 if (is64bit) { 2206 fse_clone_args = (ext_fsevent_clone_args *)data; 2207 } else { 2208 fsevent_clone_args *ufse_clone = (fsevent_clone_args *)data; 2209 2210 fse_clone_args = &_fse_clone; 2211 memset(fse_clone_args, 0, sizeof(ext_fsevent_clone_args)); 2212 2213 fse_clone_args->event_list = CAST_USER_ADDR_T(ufse_clone->event_list); 2214 fse_clone_args->num_events = ufse_clone->num_events; 2215 fse_clone_args->event_queue_depth = ufse_clone->event_queue_depth; 2216 fse_clone_args->fd = CAST_USER_ADDR_T(ufse_clone->fd); 2217 } 2218 2219 handle_clone: 2220 if (fse_clone_args->num_events < 0 || fse_clone_args->num_events > 4096) { 2221 return EINVAL; 2222 } 2223 2224 MALLOC(fseh, fsevent_handle *, sizeof(fsevent_handle), 2225 M_TEMP, M_WAITOK); 2226 if (fseh == NULL) { 2227 return ENOMEM; 2228 } 2229 memset(fseh, 0, sizeof(fsevent_handle)); 2230 2231 klist_init(&fseh->knotes); 2232 2233 MALLOC(event_list, int8_t *, 2234 fse_clone_args->num_events * sizeof(int8_t), 2235 M_TEMP, M_WAITOK); 2236 if (event_list == NULL) { 2237 FREE(fseh, M_TEMP); 2238 return ENOMEM; 2239 } 2240 2241 error = copyin(fse_clone_args->event_list, 2242 (void *)event_list, 2243 fse_clone_args->num_events * sizeof(int8_t)); 2244 if (error) { 2245 FREE(event_list, M_TEMP); 2246 FREE(fseh, M_TEMP); 2247 return error; 2248 } 2249 2250 error = add_watcher(event_list, 2251 fse_clone_args->num_events, 2252 fse_clone_args->event_queue_depth, 2253 &fseh->watcher, 2254 fseh); 2255 if (error) { 2256 FREE(event_list, M_TEMP); 2257 FREE(fseh, M_TEMP); 2258 return error; 2259 } 2260 2261 fseh->watcher->fseh = fseh; 2262 2263 error = falloc(p, &f, &fd, vfs_context_current()); 2264 if (error) { 2265 FREE(event_list, M_TEMP); 2266 FREE(fseh, M_TEMP); 2267 return (error); 2268 } 2269 proc_fdlock(p); 2270 f->f_fglob->fg_flag = FREAD | FWRITE; 2271 f->f_fglob->fg_ops = &fsevents_fops; 2272 f->f_fglob->fg_data = (caddr_t) fseh; 2273 proc_fdunlock(p); 2274 error = copyout((void *)&fd, fse_clone_args->fd, sizeof(int32_t)); 2275 if (error != 0) { 2276 fp_free(p, fd, f); 2277 } else { 2278 proc_fdlock(p); 2279 procfdtbl_releasefd(p, fd, NULL); 2280 fp_drop(p, fd, f, 1); 2281 proc_fdunlock(p); 2282 } 2283 break; 2284 2285 default: 2286 error = EINVAL; 2287 break; 2288 } 2289 2290 return error; 2291} 2292 2293static void 2294fsevents_wakeup(fs_event_watcher *watcher) 2295{ 2296 selwakeup(&watcher->fseh->si); 2297 KNOTE(&watcher->fseh->knotes, NOTE_WRITE|NOTE_NONE); 2298 wakeup((caddr_t)watcher); 2299} 2300 2301 2302/* 2303 * A struct describing which functions will get invoked for certain 2304 * actions. 2305 */ 2306static struct cdevsw fsevents_cdevsw = 2307{ 2308 fseventsopen, /* open */ 2309 fseventsclose, /* close */ 2310 fseventsread, /* read */ 2311 fseventswrite, /* write */ 2312 fseventsioctl, /* ioctl */ 2313 (stop_fcn_t *)&nulldev, /* stop */ 2314 (reset_fcn_t *)&nulldev, /* reset */ 2315 NULL, /* tty's */ 2316 eno_select, /* select */ 2317 eno_mmap, /* mmap */ 2318 eno_strat, /* strategy */ 2319 eno_getc, /* getc */ 2320 eno_putc, /* putc */ 2321 0 /* type */ 2322}; 2323 2324 2325/* 2326 * Called to initialize our device, 2327 * and to register ourselves with devfs 2328 */ 2329 2330void 2331fsevents_init(void) 2332{ 2333 int ret; 2334 2335 if (fsevents_installed) { 2336 return; 2337 } 2338 2339 fsevents_installed = 1; 2340 2341 ret = cdevsw_add(-1, &fsevents_cdevsw); 2342 if (ret < 0) { 2343 fsevents_installed = 0; 2344 return; 2345 } 2346 2347 devfs_make_node(makedev (ret, 0), DEVFS_CHAR, 2348 UID_ROOT, GID_WHEEL, 0644, "fsevents", 0); 2349 2350 fsevents_internal_init(); 2351} 2352 2353 2354char * 2355get_pathbuff(void) 2356{ 2357 char *path; 2358 2359 MALLOC_ZONE(path, char *, MAXPATHLEN, M_NAMEI, M_WAITOK); 2360 return path; 2361} 2362 2363void 2364release_pathbuff(char *path) 2365{ 2366 2367 if (path == NULL) { 2368 return; 2369 } 2370 FREE_ZONE(path, MAXPATHLEN, M_NAMEI); 2371} 2372 2373int 2374get_fse_info(struct vnode *vp, fse_info *fse, __unused vfs_context_t ctx) 2375{ 2376 struct vnode_attr va; 2377 2378 VATTR_INIT(&va); 2379 VATTR_WANTED(&va, va_fsid); 2380 VATTR_WANTED(&va, va_fileid); 2381 VATTR_WANTED(&va, va_mode); 2382 VATTR_WANTED(&va, va_uid); 2383 VATTR_WANTED(&va, va_gid); 2384 if (vp->v_flag & VISHARDLINK) { 2385 if (vp->v_type == VDIR) { 2386 VATTR_WANTED(&va, va_dirlinkcount); 2387 } else { 2388 VATTR_WANTED(&va, va_nlink); 2389 } 2390 } 2391 2392 if (vnode_getattr(vp, &va, vfs_context_kernel()) != 0) { 2393 memset(fse, 0, sizeof(fse_info)); 2394 return -1; 2395 } 2396 2397 return vnode_get_fse_info_from_vap(vp, fse, &va); 2398} 2399 2400int 2401vnode_get_fse_info_from_vap(vnode_t vp, fse_info *fse, struct vnode_attr *vap) 2402{ 2403 fse->ino = (ino64_t)vap->va_fileid; 2404 fse->dev = (dev_t)vap->va_fsid; 2405 fse->mode = (int32_t)vnode_vttoif(vnode_vtype(vp)) | vap->va_mode; 2406 fse->uid = (uid_t)vap->va_uid; 2407 fse->gid = (gid_t)vap->va_gid; 2408 if (vp->v_flag & VISHARDLINK) { 2409 fse->mode |= FSE_MODE_HLINK; 2410 if (vp->v_type == VDIR) { 2411 fse->nlink = (uint64_t)vap->va_dirlinkcount; 2412 } else { 2413 fse->nlink = (uint64_t)vap->va_nlink; 2414 } 2415 } 2416 2417 return 0; 2418} 2419 2420void 2421create_fsevent_from_kevent(vnode_t vp, uint32_t kevents, struct vnode_attr *vap) 2422{ 2423 int fsevent_type=FSE_CONTENT_MODIFIED, len; // the default is the most pessimistic 2424 char pathbuf[MAXPATHLEN]; 2425 fse_info fse; 2426 2427 2428 if (kevents & VNODE_EVENT_DELETE) { 2429 fsevent_type = FSE_DELETE; 2430 } else if (kevents & (VNODE_EVENT_EXTEND|VNODE_EVENT_WRITE)) { 2431 fsevent_type = FSE_CONTENT_MODIFIED; 2432 } else if (kevents & VNODE_EVENT_LINK) { 2433 fsevent_type = FSE_CREATE_FILE; 2434 } else if (kevents & VNODE_EVENT_RENAME) { 2435 fsevent_type = FSE_CREATE_FILE; // XXXdbg - should use FSE_RENAME but we don't have the destination info; 2436 } else if (kevents & (VNODE_EVENT_FILE_CREATED|VNODE_EVENT_FILE_REMOVED|VNODE_EVENT_DIR_CREATED|VNODE_EVENT_DIR_REMOVED)) { 2437 fsevent_type = FSE_STAT_CHANGED; // XXXdbg - because vp is a dir and the thing created/removed lived inside it 2438 } else { // a catch all for VNODE_EVENT_PERMS, VNODE_EVENT_ATTRIB and anything else 2439 fsevent_type = FSE_STAT_CHANGED; 2440 } 2441 2442 // printf("convert_kevent: kevents 0x%x fsevent type 0x%x (for %s)\n", kevents, fsevent_type, vp->v_name ? vp->v_name : "(no-name)"); 2443 2444 fse.dev = vap->va_fsid; 2445 fse.ino = vap->va_fileid; 2446 fse.mode = vnode_vttoif(vnode_vtype(vp)) | (uint32_t)vap->va_mode; 2447 if (vp->v_flag & VISHARDLINK) { 2448 fse.mode |= FSE_MODE_HLINK; 2449 if (vp->v_type == VDIR) { 2450 fse.nlink = vap->va_dirlinkcount; 2451 } else { 2452 fse.nlink = vap->va_nlink; 2453 } 2454 } 2455 2456 if (vp->v_type == VDIR) { 2457 fse.mode |= FSE_REMOTE_DIR_EVENT; 2458 } 2459 2460 2461 fse.uid = vap->va_uid; 2462 fse.gid = vap->va_gid; 2463 2464 len = sizeof(pathbuf); 2465 if (vn_getpath(vp, pathbuf, &len) == 0) { 2466 add_fsevent(fsevent_type, vfs_context_current(), FSE_ARG_STRING, len, pathbuf, FSE_ARG_FINFO, &fse, FSE_ARG_DONE); 2467 } 2468 return; 2469} 2470 2471#else /* CONFIG_FSE */ 2472/* 2473 * The get_pathbuff and release_pathbuff routines are used in places not 2474 * related to fsevents, and it's a handy abstraction, so define trivial 2475 * versions that don't cache a pool of buffers. This way, we don't have 2476 * to conditionalize the callers, and they still get the advantage of the 2477 * pool of buffers if CONFIG_FSE is turned on. 2478 */ 2479char * 2480get_pathbuff(void) 2481{ 2482 char *path; 2483 MALLOC_ZONE(path, char *, MAXPATHLEN, M_NAMEI, M_WAITOK); 2484 return path; 2485} 2486 2487void 2488release_pathbuff(char *path) 2489{ 2490 FREE_ZONE(path, MAXPATHLEN, M_NAMEI); 2491} 2492#endif /* CONFIG_FSE */ 2493