]> git.saurik.com Git - apple/xnu.git/blame_incremental - bsd/kern/decmpfs.c
xnu-7195.101.1.tar.gz
[apple/xnu.git] / bsd / kern / decmpfs.c
... / ...
CommitLineData
1/*
2 * Copyright (c) 2008-2018 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#if !FS_COMPRESSION
29
30/* We need these symbols even though compression is turned off */
31
32#define UNUSED_SYMBOL(x) asm(".global _" #x "\n.set _" #x ", 0\n");
33
34UNUSED_SYMBOL(register_decmpfs_decompressor)
35UNUSED_SYMBOL(unregister_decmpfs_decompressor)
36UNUSED_SYMBOL(decmpfs_init)
37UNUSED_SYMBOL(decmpfs_read_compressed)
38UNUSED_SYMBOL(decmpfs_cnode_cmp_type)
39UNUSED_SYMBOL(decmpfs_cnode_get_vnode_state)
40UNUSED_SYMBOL(decmpfs_cnode_get_vnode_cached_size)
41UNUSED_SYMBOL(decmpfs_cnode_get_vnode_cached_nchildren)
42UNUSED_SYMBOL(decmpfs_cnode_get_vnode_cached_total_size)
43UNUSED_SYMBOL(decmpfs_lock_compressed_data)
44UNUSED_SYMBOL(decmpfs_cnode_free)
45UNUSED_SYMBOL(decmpfs_cnode_alloc)
46UNUSED_SYMBOL(decmpfs_cnode_destroy)
47UNUSED_SYMBOL(decmpfs_decompress_file)
48UNUSED_SYMBOL(decmpfs_unlock_compressed_data)
49UNUSED_SYMBOL(decmpfs_cnode_init)
50UNUSED_SYMBOL(decmpfs_cnode_set_vnode_state)
51UNUSED_SYMBOL(decmpfs_hides_xattr)
52UNUSED_SYMBOL(decmpfs_ctx)
53UNUSED_SYMBOL(decmpfs_file_is_compressed)
54UNUSED_SYMBOL(decmpfs_update_attributes)
55UNUSED_SYMBOL(decmpfs_hides_rsrc)
56UNUSED_SYMBOL(decmpfs_pagein_compressed)
57UNUSED_SYMBOL(decmpfs_validate_compressed_file)
58
59#else /* FS_COMPRESSION */
60#include <sys/kernel.h>
61#include <sys/vnode_internal.h>
62#include <sys/file_internal.h>
63#include <sys/stat.h>
64#include <sys/fcntl.h>
65#include <sys/xattr.h>
66#include <sys/namei.h>
67#include <sys/user.h>
68#include <sys/mount_internal.h>
69#include <sys/ubc.h>
70#include <sys/decmpfs.h>
71#include <sys/uio_internal.h>
72#include <libkern/OSByteOrder.h>
73#include <libkern/section_keywords.h>
74
75#include <ptrauth.h>
76
77#pragma mark --- debugging ---
78
79#define COMPRESSION_DEBUG 0
80#define COMPRESSION_DEBUG_VERBOSE 0
81#define MALLOC_DEBUG 0
82
83#if COMPRESSION_DEBUG
84static char*
85vnpath(vnode_t vp, char *path, int len)
86{
87 int origlen = len;
88 path[0] = 0;
89 vn_getpath(vp, path, &len);
90 path[origlen - 1] = 0;
91 return path;
92}
93#endif
94
95#define ErrorLog(x, args...) \
96 printf("%s:%d:%s: " x, __FILE_NAME__, __LINE__, __FUNCTION__, ## args)
97#if COMPRESSION_DEBUG
98#define ErrorLogWithPath(x, args...) do { \
99 char *path = zalloc(ZV_NAMEI); \
100 printf("%s:%d:%s: %s: " x, __FILE_NAME__, __LINE__, __FUNCTION__, \
101 vnpath(vp, path, PATH_MAX), ## args); \
102 zfree(ZV_NAMEI, path); \
103} while(0)
104#else
105#define ErrorLogWithPath(x, args...) do { \
106 (void*)vp; \
107 printf("%s:%d:%s: %s: " x, __FILE_NAME__, __LINE__, __FUNCTION__, \
108 "<private>", ## args); \
109} while(0)
110#endif
111
112#if COMPRESSION_DEBUG
113#define DebugLog ErrorLog
114#define DebugLogWithPath ErrorLogWithPath
115#else
116#define DebugLog(x...) do { } while(0)
117#define DebugLogWithPath(x...) do { } while(0)
118#endif
119
120#if COMPRESSION_DEBUG_VERBOSE
121#define VerboseLog ErrorLog
122#define VerboseLogWithPath ErrorLogWithPath
123#else
124#define VerboseLog(x...) do { } while(0)
125#define VerboseLogWithPath(x...) do { } while(0)
126#endif
127
128#pragma mark --- globals ---
129
130static LCK_GRP_DECLARE(decmpfs_lockgrp, "VFSCOMP");
131static LCK_RW_DECLARE(decompressorsLock, &decmpfs_lockgrp);
132static LCK_MTX_DECLARE(decompress_channel_mtx, &decmpfs_lockgrp);
133
134static const decmpfs_registration *decompressors[CMP_MAX]; /* the registered compressors */
135static int decompress_channel; /* channel used by decompress_file to wake up waiters */
136
137vfs_context_t decmpfs_ctx;
138
139#pragma mark --- decmp_get_func ---
140
141#define offsetof_func(func) ((uintptr_t)offsetof(decmpfs_registration, func))
142
143static void *
144_func_from_offset(uint32_t type, uintptr_t offset, uint32_t discriminator)
145{
146 /* get the function at the given offset in the registration for the given type */
147 const decmpfs_registration *reg = decompressors[type];
148
149 switch (reg->decmpfs_registration) {
150 case DECMPFS_REGISTRATION_VERSION_V1:
151 if (offset > offsetof_func(free_data)) {
152 return NULL;
153 }
154 break;
155 case DECMPFS_REGISTRATION_VERSION_V3:
156 if (offset > offsetof_func(get_flags)) {
157 return NULL;
158 }
159 break;
160 default:
161 return NULL;
162 }
163
164 void *ptr = *(void * const *)((const void *)reg + offset);
165 if (ptr != NULL) {
166 /* Resign as a function-in-void* */
167 ptr = ptrauth_auth_and_resign(ptr, ptrauth_key_asia, discriminator, ptrauth_key_asia, 0);
168 }
169 return ptr;
170}
171
172extern void IOServicePublishResource( const char * property, boolean_t value );
173extern boolean_t IOServiceWaitForMatchingResource( const char * property, uint64_t timeout );
174extern boolean_t IOCatalogueMatchingDriversPresent( const char * property );
175
176static void *
177_decmp_get_func(vnode_t vp, uint32_t type, uintptr_t offset, uint32_t discriminator)
178{
179 /*
180 * this function should be called while holding a shared lock to decompressorsLock,
181 * and will return with the lock held
182 */
183
184 if (type >= CMP_MAX) {
185 return NULL;
186 }
187
188 if (decompressors[type] != NULL) {
189 // the compressor has already registered but the function might be null
190 return _func_from_offset(type, offset, discriminator);
191 }
192
193 // does IOKit know about a kext that is supposed to provide this type?
194 char providesName[80];
195 snprintf(providesName, sizeof(providesName), "com.apple.AppleFSCompression.providesType%u", type);
196 if (IOCatalogueMatchingDriversPresent(providesName)) {
197 // there is a kext that says it will register for this type, so let's wait for it
198 char resourceName[80];
199 uint64_t delay = 10000000ULL; // 10 milliseconds.
200 snprintf(resourceName, sizeof(resourceName), "com.apple.AppleFSCompression.Type%u", type);
201 ErrorLogWithPath("waiting for %s\n", resourceName);
202 while (decompressors[type] == NULL) {
203 lck_rw_unlock_shared(&decompressorsLock); // we have to unlock to allow the kext to register
204 if (IOServiceWaitForMatchingResource(resourceName, delay)) {
205 lck_rw_lock_shared(&decompressorsLock);
206 break;
207 }
208 if (!IOCatalogueMatchingDriversPresent(providesName)) {
209 //
210 ErrorLogWithPath("the kext with %s is no longer present\n", providesName);
211 lck_rw_lock_shared(&decompressorsLock);
212 break;
213 }
214 ErrorLogWithPath("still waiting for %s\n", resourceName);
215 delay *= 2;
216 lck_rw_lock_shared(&decompressorsLock);
217 }
218 // IOKit says the kext is loaded, so it should be registered too!
219 if (decompressors[type] == NULL) {
220 ErrorLogWithPath("we found %s, but the type still isn't registered\n", providesName);
221 return NULL;
222 }
223 // it's now registered, so let's return the function
224 return _func_from_offset(type, offset, discriminator);
225 }
226
227 // the compressor hasn't registered, so it never will unless someone manually kextloads it
228 ErrorLogWithPath("tried to access a compressed file of unregistered type %d\n", type);
229 return NULL;
230}
231
232#define decmp_get_func(vp, type, func) (typeof(decompressors[0]->func))_decmp_get_func(vp, type, offsetof_func(func), ptrauth_function_pointer_type_discriminator(typeof(decompressors[0]->func)))
233
234#pragma mark --- utilities ---
235
236#if COMPRESSION_DEBUG
237static int
238vnsize(vnode_t vp, uint64_t *size)
239{
240 struct vnode_attr va;
241 VATTR_INIT(&va);
242 VATTR_WANTED(&va, va_data_size);
243 int error = vnode_getattr(vp, &va, decmpfs_ctx);
244 if (error != 0) {
245 ErrorLogWithPath("vnode_getattr err %d\n", error);
246 return error;
247 }
248 *size = va.va_data_size;
249 return 0;
250}
251#endif /* COMPRESSION_DEBUG */
252
253#pragma mark --- cnode routines ---
254
255ZONE_DECLARE(decmpfs_cnode_zone, "decmpfs_cnode",
256 sizeof(struct decmpfs_cnode), ZC_NONE);
257
258decmpfs_cnode *
259decmpfs_cnode_alloc(void)
260{
261 return zalloc(decmpfs_cnode_zone);
262}
263
264void
265decmpfs_cnode_free(decmpfs_cnode *dp)
266{
267 zfree(decmpfs_cnode_zone, dp);
268}
269
270void
271decmpfs_cnode_init(decmpfs_cnode *cp)
272{
273 memset(cp, 0, sizeof(*cp));
274 lck_rw_init(&cp->compressed_data_lock, &decmpfs_lockgrp, NULL);
275}
276
277void
278decmpfs_cnode_destroy(decmpfs_cnode *cp)
279{
280 lck_rw_destroy(&cp->compressed_data_lock, &decmpfs_lockgrp);
281}
282
283bool
284decmpfs_trylock_compressed_data(decmpfs_cnode *cp, int exclusive)
285{
286 void *thread = current_thread();
287 bool retval = false;
288
289 if (cp->lockowner == thread) {
290 /* this thread is already holding an exclusive lock, so bump the count */
291 cp->lockcount++;
292 retval = true;
293 } else if (exclusive) {
294 if ((retval = lck_rw_try_lock_exclusive(&cp->compressed_data_lock))) {
295 cp->lockowner = thread;
296 cp->lockcount = 1;
297 }
298 } else {
299 if ((retval = lck_rw_try_lock_shared(&cp->compressed_data_lock))) {
300 cp->lockowner = (void *)-1;
301 }
302 }
303 return retval;
304}
305
306void
307decmpfs_lock_compressed_data(decmpfs_cnode *cp, int exclusive)
308{
309 void *thread = current_thread();
310
311 if (cp->lockowner == thread) {
312 /* this thread is already holding an exclusive lock, so bump the count */
313 cp->lockcount++;
314 } else if (exclusive) {
315 lck_rw_lock_exclusive(&cp->compressed_data_lock);
316 cp->lockowner = thread;
317 cp->lockcount = 1;
318 } else {
319 lck_rw_lock_shared(&cp->compressed_data_lock);
320 cp->lockowner = (void *)-1;
321 }
322}
323
324void
325decmpfs_unlock_compressed_data(decmpfs_cnode *cp, __unused int exclusive)
326{
327 void *thread = current_thread();
328
329 if (cp->lockowner == thread) {
330 /* this thread is holding an exclusive lock, so decrement the count */
331 if ((--cp->lockcount) > 0) {
332 /* the caller still has outstanding locks, so we're done */
333 return;
334 }
335 cp->lockowner = NULL;
336 }
337
338 lck_rw_done(&cp->compressed_data_lock);
339}
340
341uint32_t
342decmpfs_cnode_get_vnode_state(decmpfs_cnode *cp)
343{
344 return cp->cmp_state;
345}
346
347void
348decmpfs_cnode_set_vnode_state(decmpfs_cnode *cp, uint32_t state, int skiplock)
349{
350 if (!skiplock) {
351 decmpfs_lock_compressed_data(cp, 1);
352 }
353 cp->cmp_state = (uint8_t)state;
354 if (state == FILE_TYPE_UNKNOWN) {
355 /* clear out the compression type too */
356 cp->cmp_type = 0;
357 }
358 if (!skiplock) {
359 decmpfs_unlock_compressed_data(cp, 1);
360 }
361}
362
363static void
364decmpfs_cnode_set_vnode_cmp_type(decmpfs_cnode *cp, uint32_t cmp_type, int skiplock)
365{
366 if (!skiplock) {
367 decmpfs_lock_compressed_data(cp, 1);
368 }
369 cp->cmp_type = cmp_type;
370 if (!skiplock) {
371 decmpfs_unlock_compressed_data(cp, 1);
372 }
373}
374
375static void
376decmpfs_cnode_set_vnode_minimal_xattr(decmpfs_cnode *cp, int minimal_xattr, int skiplock)
377{
378 if (!skiplock) {
379 decmpfs_lock_compressed_data(cp, 1);
380 }
381 cp->cmp_minimal_xattr = !!minimal_xattr;
382 if (!skiplock) {
383 decmpfs_unlock_compressed_data(cp, 1);
384 }
385}
386
387uint64_t
388decmpfs_cnode_get_vnode_cached_size(decmpfs_cnode *cp)
389{
390 return cp->uncompressed_size;
391}
392
393uint64_t
394decmpfs_cnode_get_vnode_cached_nchildren(decmpfs_cnode *cp)
395{
396 return cp->nchildren;
397}
398
399uint64_t
400decmpfs_cnode_get_vnode_cached_total_size(decmpfs_cnode *cp)
401{
402 return cp->total_size;
403}
404
405void
406decmpfs_cnode_set_vnode_cached_size(decmpfs_cnode *cp, uint64_t size)
407{
408 while (1) {
409 uint64_t old = cp->uncompressed_size;
410 if (OSCompareAndSwap64(old, size, (UInt64*)&cp->uncompressed_size)) {
411 return;
412 } else {
413 /* failed to write our value, so loop */
414 }
415 }
416}
417
418void
419decmpfs_cnode_set_vnode_cached_nchildren(decmpfs_cnode *cp, uint64_t nchildren)
420{
421 while (1) {
422 uint64_t old = cp->nchildren;
423 if (OSCompareAndSwap64(old, nchildren, (UInt64*)&cp->nchildren)) {
424 return;
425 } else {
426 /* failed to write our value, so loop */
427 }
428 }
429}
430
431void
432decmpfs_cnode_set_vnode_cached_total_size(decmpfs_cnode *cp, uint64_t total_sz)
433{
434 while (1) {
435 uint64_t old = cp->total_size;
436 if (OSCompareAndSwap64(old, total_sz, (UInt64*)&cp->total_size)) {
437 return;
438 } else {
439 /* failed to write our value, so loop */
440 }
441 }
442}
443
444static uint64_t
445decmpfs_cnode_get_decompression_flags(decmpfs_cnode *cp)
446{
447 return cp->decompression_flags;
448}
449
450static void
451decmpfs_cnode_set_decompression_flags(decmpfs_cnode *cp, uint64_t flags)
452{
453 while (1) {
454 uint64_t old = cp->decompression_flags;
455 if (OSCompareAndSwap64(old, flags, (UInt64*)&cp->decompression_flags)) {
456 return;
457 } else {
458 /* failed to write our value, so loop */
459 }
460 }
461}
462
463uint32_t
464decmpfs_cnode_cmp_type(decmpfs_cnode *cp)
465{
466 return cp->cmp_type;
467}
468
469#pragma mark --- decmpfs state routines ---
470
471static int
472decmpfs_fetch_compressed_header(vnode_t vp, decmpfs_cnode *cp, decmpfs_header **hdrOut, int returnInvalid, size_t *hdr_size)
473{
474 /*
475 * fetches vp's compression xattr, converting it into a decmpfs_header; returns 0 or errno
476 * if returnInvalid == 1, returns the header even if the type was invalid (out of range),
477 * and return ERANGE in that case
478 */
479
480 size_t read_size = 0;
481 size_t attr_size = 0;
482 size_t alloc_size = 0;
483 uio_t attr_uio = NULL;
484 int err = 0;
485 char *data = NULL;
486 const bool no_additional_data = ((cp != NULL)
487 && (cp->cmp_type != 0)
488 && (cp->cmp_minimal_xattr != 0));
489 char uio_buf[UIO_SIZEOF(1)];
490 decmpfs_header *hdr = NULL;
491
492 /*
493 * Trace the following parameters on entry with event-id 0x03120004
494 *
495 * @vp->v_id: vnode-id for which to fetch compressed header.
496 * @no_additional_data: If set true then xattr didn't have any extra data.
497 * @returnInvalid: return the header even though the type is out of range.
498 */
499 DECMPFS_EMIT_TRACE_ENTRY(DECMPDBG_FETCH_COMPRESSED_HEADER, vp->v_id,
500 no_additional_data, returnInvalid);
501
502 if (no_additional_data) {
503 /* this file's xattr didn't have any extra data when we fetched it, so we can synthesize a header from the data in the cnode */
504
505 alloc_size = sizeof(decmpfs_header);
506 data = kheap_alloc(KHEAP_TEMP, alloc_size, Z_WAITOK);
507 if (!data) {
508 err = ENOMEM;
509 goto out;
510 }
511 hdr = (decmpfs_header*)data;
512 hdr->attr_size = sizeof(decmpfs_disk_header);
513 hdr->compression_magic = DECMPFS_MAGIC;
514 hdr->compression_type = cp->cmp_type;
515 if (hdr->compression_type == DATALESS_PKG_CMPFS_TYPE) {
516 if (!vnode_isdir(vp)) {
517 err = EINVAL;
518 goto out;
519 }
520 hdr->_size.value = DECMPFS_PKG_VALUE_FROM_SIZE_COUNT(
521 decmpfs_cnode_get_vnode_cached_size(cp),
522 decmpfs_cnode_get_vnode_cached_nchildren(cp));
523 } else if (vnode_isdir(vp)) {
524 hdr->_size.value = decmpfs_cnode_get_vnode_cached_nchildren(cp);
525 } else {
526 hdr->_size.value = decmpfs_cnode_get_vnode_cached_size(cp);
527 }
528 } else {
529 /* figure out how big the xattr is on disk */
530 err = vn_getxattr(vp, DECMPFS_XATTR_NAME, NULL, &attr_size, XATTR_NOSECURITY, decmpfs_ctx);
531 if (err != 0) {
532 goto out;
533 }
534 alloc_size = attr_size + sizeof(hdr->attr_size);
535
536 if (attr_size < sizeof(decmpfs_disk_header) || attr_size > MAX_DECMPFS_XATTR_SIZE) {
537 err = EINVAL;
538 goto out;
539 }
540
541 /* allocation includes space for the extra attr_size field of a compressed_header */
542 data = kheap_alloc(KHEAP_TEMP, alloc_size, Z_WAITOK);
543 if (!data) {
544 err = ENOMEM;
545 goto out;
546 }
547
548 /* read the xattr into our buffer, skipping over the attr_size field at the beginning */
549 attr_uio = uio_createwithbuffer(1, 0, UIO_SYSSPACE, UIO_READ, &uio_buf[0], sizeof(uio_buf));
550 uio_addiov(attr_uio, CAST_USER_ADDR_T(data + sizeof(hdr->attr_size)), attr_size);
551
552 err = vn_getxattr(vp, DECMPFS_XATTR_NAME, attr_uio, &read_size, XATTR_NOSECURITY, decmpfs_ctx);
553 if (err != 0) {
554 goto out;
555 }
556 if (read_size != attr_size) {
557 err = EINVAL;
558 goto out;
559 }
560 hdr = (decmpfs_header*)data;
561 hdr->attr_size = (uint32_t)attr_size;
562 /* swap the fields to native endian */
563 hdr->compression_magic = OSSwapLittleToHostInt32(hdr->compression_magic);
564 hdr->compression_type = OSSwapLittleToHostInt32(hdr->compression_type);
565 hdr->uncompressed_size = OSSwapLittleToHostInt64(hdr->uncompressed_size);
566 }
567
568 if (hdr->compression_magic != DECMPFS_MAGIC) {
569 ErrorLogWithPath("invalid compression_magic 0x%08x, should be 0x%08x\n", hdr->compression_magic, DECMPFS_MAGIC);
570 err = EINVAL;
571 goto out;
572 }
573
574 /*
575 * Special-case the DATALESS compressor here; that is a valid type,
576 * even through there will never be an entry in the decompressor
577 * handler table for it. If we don't do this, then the cmp_state
578 * for this cnode will end up being marked NOT_COMPRESSED, and
579 * we'll be stuck in limbo.
580 */
581 if (hdr->compression_type >= CMP_MAX && !decmpfs_type_is_dataless(hdr->compression_type)) {
582 if (returnInvalid) {
583 /* return the header even though the type is out of range */
584 err = ERANGE;
585 } else {
586 ErrorLogWithPath("compression_type %d out of range\n", hdr->compression_type);
587 err = EINVAL;
588 }
589 goto out;
590 }
591
592out:
593 if (err && (err != ERANGE)) {
594 DebugLogWithPath("err %d\n", err);
595 kheap_free(KHEAP_TEMP, data, alloc_size);
596 *hdrOut = NULL;
597 } else {
598 *hdrOut = hdr;
599 *hdr_size = alloc_size;
600 }
601 /*
602 * Trace the following parameters on return with event-id 0x03120004.
603 *
604 * @vp->v_id: vnode-id for which to fetch compressed header.
605 * @err: value returned from this function.
606 */
607 DECMPFS_EMIT_TRACE_RETURN(DECMPDBG_FETCH_COMPRESSED_HEADER, vp->v_id, err);
608 return err;
609}
610
611static int
612decmpfs_fast_get_state(decmpfs_cnode *cp)
613{
614 /*
615 * return the cached state
616 * this should *only* be called when we know that decmpfs_file_is_compressed has already been called,
617 * because this implies that the cached state is valid
618 */
619 int cmp_state = decmpfs_cnode_get_vnode_state(cp);
620
621 switch (cmp_state) {
622 case FILE_IS_NOT_COMPRESSED:
623 case FILE_IS_COMPRESSED:
624 case FILE_IS_CONVERTING:
625 return cmp_state;
626 case FILE_TYPE_UNKNOWN:
627 /*
628 * we should only get here if decmpfs_file_is_compressed was not called earlier on this vnode,
629 * which should not be possible
630 */
631 ErrorLog("decmpfs_fast_get_state called on unknown file\n");
632 return FILE_IS_NOT_COMPRESSED;
633 default:
634 /* */
635 ErrorLog("unknown cmp_state %d\n", cmp_state);
636 return FILE_IS_NOT_COMPRESSED;
637 }
638}
639
640static int
641decmpfs_fast_file_is_compressed(decmpfs_cnode *cp)
642{
643 int cmp_state = decmpfs_cnode_get_vnode_state(cp);
644
645 switch (cmp_state) {
646 case FILE_IS_NOT_COMPRESSED:
647 return 0;
648 case FILE_IS_COMPRESSED:
649 case FILE_IS_CONVERTING:
650 return 1;
651 case FILE_TYPE_UNKNOWN:
652 /*
653 * we should only get here if decmpfs_file_is_compressed was not called earlier on this vnode,
654 * which should not be possible
655 */
656 ErrorLog("decmpfs_fast_get_state called on unknown file\n");
657 return 0;
658 default:
659 /* */
660 ErrorLog("unknown cmp_state %d\n", cmp_state);
661 return 0;
662 }
663}
664
665errno_t
666decmpfs_validate_compressed_file(vnode_t vp, decmpfs_cnode *cp)
667{
668 /* give a compressor a chance to indicate that a compressed file is invalid */
669 decmpfs_header *hdr = NULL;
670 size_t alloc_size = 0;
671 errno_t err = decmpfs_fetch_compressed_header(vp, cp, &hdr, 0, &alloc_size);
672
673 if (err) {
674 /* we couldn't get the header */
675 if (decmpfs_fast_get_state(cp) == FILE_IS_NOT_COMPRESSED) {
676 /* the file is no longer compressed, so return success */
677 err = 0;
678 }
679 goto out;
680 }
681
682 if (!decmpfs_type_is_dataless(hdr->compression_type)) {
683 lck_rw_lock_shared(&decompressorsLock);
684 decmpfs_validate_compressed_file_func validate = decmp_get_func(vp, hdr->compression_type, validate);
685 if (validate) { /* make sure this validation function is valid */
686 /* is the data okay? */
687 err = validate(vp, decmpfs_ctx, hdr);
688 } else if (decmp_get_func(vp, hdr->compression_type, fetch) == NULL) {
689 /* the type isn't registered */
690 err = EIO;
691 } else {
692 /* no validate registered, so nothing to do */
693 err = 0;
694 }
695 lck_rw_unlock_shared(&decompressorsLock);
696 }
697out:
698 if (hdr != NULL) {
699 kheap_free(KHEAP_TEMP, hdr, alloc_size);
700 }
701#if COMPRESSION_DEBUG
702 if (err) {
703 DebugLogWithPath("decmpfs_validate_compressed_file ret %d, vp->v_flag %d\n", err, vp->v_flag);
704 }
705#endif
706 return err;
707}
708
709int
710decmpfs_file_is_compressed(vnode_t vp, decmpfs_cnode *cp)
711{
712 /*
713 * determines whether vp points to a compressed file
714 *
715 * to speed up this operation, we cache the result in the cnode, and do as little as possible
716 * in the case where the cnode already has a valid cached state
717 *
718 */
719
720 int ret = 0;
721 int error = 0;
722 uint32_t cmp_state;
723 struct vnode_attr va_fetch;
724 decmpfs_header *hdr = NULL;
725 size_t alloc_size = 0;
726 mount_t mp = NULL;
727 int cnode_locked = 0;
728 int saveInvalid = 0; // save the header data even though the type was out of range
729 uint64_t decompression_flags = 0;
730 bool is_mounted, is_local_fs;
731
732 if (vnode_isnamedstream(vp)) {
733 /*
734 * named streams can't be compressed
735 * since named streams of the same file share the same cnode,
736 * we don't want to get/set the state in the cnode, just return 0
737 */
738 return 0;
739 }
740
741 /* examine the cached a state in this cnode */
742 cmp_state = decmpfs_cnode_get_vnode_state(cp);
743 switch (cmp_state) {
744 case FILE_IS_NOT_COMPRESSED:
745 return 0;
746 case FILE_IS_COMPRESSED:
747 return 1;
748 case FILE_IS_CONVERTING:
749 /* treat the file as compressed, because this gives us a way to block future reads until decompression is done */
750 return 1;
751 case FILE_TYPE_UNKNOWN:
752 /* the first time we encountered this vnode, so we need to check it out */
753 break;
754 default:
755 /* unknown state, assume file is not compressed */
756 ErrorLogWithPath("unknown cmp_state %d\n", cmp_state);
757 return 0;
758 }
759
760 is_mounted = false;
761 is_local_fs = false;
762 mp = vnode_mount(vp);
763 if (mp) {
764 is_mounted = true;
765 }
766 if (is_mounted) {
767 is_local_fs = ((mp->mnt_flag & MNT_LOCAL));
768 }
769 /*
770 * Trace the following parameters on entry with event-id 0x03120014.
771 *
772 * @vp->v_id: vnode-id of the file being queried.
773 * @is_mounted: set to true if @vp belongs to a mounted fs.
774 * @is_local_fs: set to true if @vp belongs to local fs.
775 */
776 DECMPFS_EMIT_TRACE_ENTRY(DECMPDBG_FILE_IS_COMPRESSED, vp->v_id,
777 is_mounted, is_local_fs);
778
779 if (!is_mounted) {
780 /*
781 * this should only be true before we mount the root filesystem
782 * we short-cut this return to avoid the call to getattr below, which
783 * will fail before root is mounted
784 */
785 ret = FILE_IS_NOT_COMPRESSED;
786 goto done;
787 }
788
789 if (!is_local_fs) {
790 /* compression only supported on local filesystems */
791 ret = FILE_IS_NOT_COMPRESSED;
792 goto done;
793 }
794
795 /* lock our cnode data so that another caller doesn't change the state under us */
796 decmpfs_lock_compressed_data(cp, 1);
797 cnode_locked = 1;
798
799 VATTR_INIT(&va_fetch);
800 VATTR_WANTED(&va_fetch, va_flags);
801 error = vnode_getattr(vp, &va_fetch, decmpfs_ctx);
802 if (error) {
803 /* failed to get the bsd flags so the file is not compressed */
804 ret = FILE_IS_NOT_COMPRESSED;
805 goto done;
806 }
807 if (va_fetch.va_flags & UF_COMPRESSED) {
808 /* UF_COMPRESSED is on, make sure the file has the DECMPFS_XATTR_NAME xattr */
809 error = decmpfs_fetch_compressed_header(vp, cp, &hdr, 1, &alloc_size);
810 if ((hdr != NULL) && (error == ERANGE)) {
811 saveInvalid = 1;
812 }
813 if (error) {
814 /* failed to get the xattr so the file is not compressed */
815 ret = FILE_IS_NOT_COMPRESSED;
816 goto done;
817 }
818 /*
819 * We got the xattr, so the file is at least tagged compressed.
820 * For DATALESS, regular files and directories can be "compressed".
821 * For all other types, only files are allowed.
822 */
823 if (!vnode_isreg(vp) &&
824 !(decmpfs_type_is_dataless(hdr->compression_type) && vnode_isdir(vp))) {
825 ret = FILE_IS_NOT_COMPRESSED;
826 goto done;
827 }
828 ret = FILE_IS_COMPRESSED;
829 goto done;
830 }
831 /* UF_COMPRESSED isn't on, so the file isn't compressed */
832 ret = FILE_IS_NOT_COMPRESSED;
833
834done:
835 if (((ret == FILE_IS_COMPRESSED) || saveInvalid) && hdr) {
836 /*
837 * cache the uncompressed size away in the cnode
838 */
839
840 if (!cnode_locked) {
841 /*
842 * we should never get here since the only place ret is set to FILE_IS_COMPRESSED
843 * is after the call to decmpfs_lock_compressed_data above
844 */
845 decmpfs_lock_compressed_data(cp, 1);
846 cnode_locked = 1;
847 }
848
849 if (vnode_isdir(vp)) {
850 decmpfs_cnode_set_vnode_cached_size(cp, 64);
851 decmpfs_cnode_set_vnode_cached_nchildren(cp, decmpfs_get_directory_entries(hdr));
852 if (hdr->compression_type == DATALESS_PKG_CMPFS_TYPE) {
853 decmpfs_cnode_set_vnode_cached_total_size(cp, DECMPFS_PKG_SIZE(hdr->_size));
854 }
855 } else {
856 decmpfs_cnode_set_vnode_cached_size(cp, hdr->uncompressed_size);
857 }
858 decmpfs_cnode_set_vnode_state(cp, ret, 1);
859 decmpfs_cnode_set_vnode_cmp_type(cp, hdr->compression_type, 1);
860 /* remember if the xattr's size was equal to the minimal xattr */
861 if (hdr->attr_size == sizeof(decmpfs_disk_header)) {
862 decmpfs_cnode_set_vnode_minimal_xattr(cp, 1, 1);
863 }
864 if (ret == FILE_IS_COMPRESSED) {
865 /* update the ubc's size for this file */
866 ubc_setsize(vp, hdr->uncompressed_size);
867
868 /* update the decompression flags in the decmpfs cnode */
869 lck_rw_lock_shared(&decompressorsLock);
870 decmpfs_get_decompression_flags_func get_flags = decmp_get_func(vp, hdr->compression_type, get_flags);
871 if (get_flags) {
872 decompression_flags = get_flags(vp, decmpfs_ctx, hdr);
873 }
874 lck_rw_unlock_shared(&decompressorsLock);
875 decmpfs_cnode_set_decompression_flags(cp, decompression_flags);
876 }
877 } else {
878 /* we might have already taken the lock above; if so, skip taking it again by passing cnode_locked as the skiplock parameter */
879 decmpfs_cnode_set_vnode_state(cp, ret, cnode_locked);
880 }
881
882 if (cnode_locked) {
883 decmpfs_unlock_compressed_data(cp, 1);
884 }
885
886 if (hdr != NULL) {
887 kheap_free(KHEAP_TEMP, hdr, alloc_size);
888 }
889
890 /*
891 * Trace the following parameters on return with event-id 0x03120014.
892 *
893 * @vp->v_id: vnode-id of the file being queried.
894 * @return: set to 1 is file is compressed.
895 */
896 switch (ret) {
897 case FILE_IS_NOT_COMPRESSED:
898 DECMPFS_EMIT_TRACE_RETURN(DECMPDBG_FILE_IS_COMPRESSED, vp->v_id, 0);
899 return 0;
900 case FILE_IS_COMPRESSED:
901 case FILE_IS_CONVERTING:
902 DECMPFS_EMIT_TRACE_RETURN(DECMPDBG_FILE_IS_COMPRESSED, vp->v_id, 1);
903 return 1;
904 default:
905 /* unknown state, assume file is not compressed */
906 DECMPFS_EMIT_TRACE_RETURN(DECMPDBG_FILE_IS_COMPRESSED, vp->v_id, 0);
907 ErrorLogWithPath("unknown ret %d\n", ret);
908 return 0;
909 }
910}
911
912int
913decmpfs_update_attributes(vnode_t vp, struct vnode_attr *vap)
914{
915 int error = 0;
916
917 if (VATTR_IS_ACTIVE(vap, va_flags)) {
918 /* the BSD flags are being updated */
919 if (vap->va_flags & UF_COMPRESSED) {
920 /* the compressed bit is being set, did it change? */
921 struct vnode_attr va_fetch;
922 int old_flags = 0;
923 VATTR_INIT(&va_fetch);
924 VATTR_WANTED(&va_fetch, va_flags);
925 error = vnode_getattr(vp, &va_fetch, decmpfs_ctx);
926 if (error) {
927 return error;
928 }
929
930 old_flags = va_fetch.va_flags;
931
932 if (!(old_flags & UF_COMPRESSED)) {
933 /*
934 * Compression bit was turned on, make sure the file has the DECMPFS_XATTR_NAME attribute.
935 * This precludes anyone from using the UF_COMPRESSED bit for anything else, and it enforces
936 * an order of operation -- you must first do the setxattr and then the chflags.
937 */
938
939 if (VATTR_IS_ACTIVE(vap, va_data_size)) {
940 /*
941 * don't allow the caller to set the BSD flag and the size in the same call
942 * since this doesn't really make sense
943 */
944 vap->va_flags &= ~UF_COMPRESSED;
945 return 0;
946 }
947
948 decmpfs_header *hdr = NULL;
949 size_t alloc_size = 0;
950 error = decmpfs_fetch_compressed_header(vp, NULL, &hdr, 1, &alloc_size);
951 if (error == 0) {
952 /*
953 * Allow the flag to be set since the decmpfs attribute
954 * is present.
955 *
956 * If we're creating a dataless file we do not want to
957 * truncate it to zero which allows the file resolver to
958 * have more control over when truncation should happen.
959 * All other types of compressed files are truncated to
960 * zero.
961 */
962 if (!decmpfs_type_is_dataless(hdr->compression_type)) {
963 VATTR_SET_ACTIVE(vap, va_data_size);
964 vap->va_data_size = 0;
965 }
966 } else if (error == ERANGE) {
967 /* the file had a decmpfs attribute but the type was out of range, so don't muck with the file's data size */
968 } else {
969 /* no DECMPFS_XATTR_NAME attribute, so deny the update */
970 vap->va_flags &= ~UF_COMPRESSED;
971 }
972 if (hdr != NULL) {
973 kheap_free(KHEAP_TEMP, hdr, alloc_size);
974 }
975 }
976 }
977 }
978
979 return 0;
980}
981
982static int
983wait_for_decompress(decmpfs_cnode *cp)
984{
985 int state;
986 lck_mtx_lock(&decompress_channel_mtx);
987 do {
988 state = decmpfs_fast_get_state(cp);
989 if (state != FILE_IS_CONVERTING) {
990 /* file is not decompressing */
991 lck_mtx_unlock(&decompress_channel_mtx);
992 return state;
993 }
994 msleep((caddr_t)&decompress_channel, &decompress_channel_mtx, PINOD, "wait_for_decompress", NULL);
995 } while (1);
996}
997
998#pragma mark --- decmpfs hide query routines ---
999
1000int
1001decmpfs_hides_rsrc(vfs_context_t ctx, decmpfs_cnode *cp)
1002{
1003 /*
1004 * WARNING!!!
1005 * callers may (and do) pass NULL for ctx, so we should only use it
1006 * for this equality comparison
1007 *
1008 * This routine should only be called after a file has already been through decmpfs_file_is_compressed
1009 */
1010
1011 if (ctx == decmpfs_ctx) {
1012 return 0;
1013 }
1014
1015 if (!decmpfs_fast_file_is_compressed(cp)) {
1016 return 0;
1017 }
1018
1019 /* all compressed files hide their resource fork */
1020 return 1;
1021}
1022
1023int
1024decmpfs_hides_xattr(vfs_context_t ctx, decmpfs_cnode *cp, const char *xattr)
1025{
1026 /*
1027 * WARNING!!!
1028 * callers may (and do) pass NULL for ctx, so we should only use it
1029 * for this equality comparison
1030 *
1031 * This routine should only be called after a file has already been through decmpfs_file_is_compressed
1032 */
1033
1034 if (ctx == decmpfs_ctx) {
1035 return 0;
1036 }
1037 if (strncmp(xattr, XATTR_RESOURCEFORK_NAME, sizeof(XATTR_RESOURCEFORK_NAME) - 1) == 0) {
1038 return decmpfs_hides_rsrc(ctx, cp);
1039 }
1040 if (!decmpfs_fast_file_is_compressed(cp)) {
1041 /* file is not compressed, so don't hide this xattr */
1042 return 0;
1043 }
1044 if (strncmp(xattr, DECMPFS_XATTR_NAME, sizeof(DECMPFS_XATTR_NAME) - 1) == 0) {
1045 /* it's our xattr, so hide it */
1046 return 1;
1047 }
1048 /* don't hide this xattr */
1049 return 0;
1050}
1051
1052#pragma mark --- registration/validation routines ---
1053
1054static inline int
1055registration_valid(const decmpfs_registration *registration)
1056{
1057 return registration && ((registration->decmpfs_registration == DECMPFS_REGISTRATION_VERSION_V1) || (registration->decmpfs_registration == DECMPFS_REGISTRATION_VERSION_V3));
1058}
1059
1060errno_t
1061register_decmpfs_decompressor(uint32_t compression_type, const decmpfs_registration *registration)
1062{
1063 /* called by kexts to register decompressors */
1064
1065 errno_t ret = 0;
1066 int locked = 0;
1067 char resourceName[80];
1068
1069 if ((compression_type >= CMP_MAX) || !registration_valid(registration)) {
1070 ret = EINVAL;
1071 goto out;
1072 }
1073
1074 lck_rw_lock_exclusive(&decompressorsLock); locked = 1;
1075
1076 /* make sure the registration for this type is zero */
1077 if (decompressors[compression_type] != NULL) {
1078 ret = EEXIST;
1079 goto out;
1080 }
1081 decompressors[compression_type] = registration;
1082 snprintf(resourceName, sizeof(resourceName), "com.apple.AppleFSCompression.Type%u", compression_type);
1083 IOServicePublishResource(resourceName, TRUE);
1084
1085out:
1086 if (locked) {
1087 lck_rw_unlock_exclusive(&decompressorsLock);
1088 }
1089 return ret;
1090}
1091
1092errno_t
1093unregister_decmpfs_decompressor(uint32_t compression_type, decmpfs_registration *registration)
1094{
1095 /* called by kexts to unregister decompressors */
1096
1097 errno_t ret = 0;
1098 int locked = 0;
1099 char resourceName[80];
1100
1101 if ((compression_type >= CMP_MAX) || !registration_valid(registration)) {
1102 ret = EINVAL;
1103 goto out;
1104 }
1105
1106 lck_rw_lock_exclusive(&decompressorsLock); locked = 1;
1107 if (decompressors[compression_type] != registration) {
1108 ret = EEXIST;
1109 goto out;
1110 }
1111 decompressors[compression_type] = NULL;
1112 snprintf(resourceName, sizeof(resourceName), "com.apple.AppleFSCompression.Type%u", compression_type);
1113 IOServicePublishResource(resourceName, FALSE);
1114
1115out:
1116 if (locked) {
1117 lck_rw_unlock_exclusive(&decompressorsLock);
1118 }
1119 return ret;
1120}
1121
1122static int
1123compression_type_valid(vnode_t vp, decmpfs_header *hdr)
1124{
1125 /* fast pre-check to determine if the given compressor has checked in */
1126 int ret = 0;
1127
1128 /* every compressor must have at least a fetch function */
1129 lck_rw_lock_shared(&decompressorsLock);
1130 if (decmp_get_func(vp, hdr->compression_type, fetch) != NULL) {
1131 ret = 1;
1132 }
1133 lck_rw_unlock_shared(&decompressorsLock);
1134
1135 return ret;
1136}
1137
1138#pragma mark --- compression/decompression routines ---
1139
1140static int
1141decmpfs_fetch_uncompressed_data(vnode_t vp, decmpfs_cnode *cp, decmpfs_header *hdr, off_t offset, user_ssize_t size, int nvec, decmpfs_vector *vec, uint64_t *bytes_read)
1142{
1143 /* get the uncompressed bytes for the specified region of vp by calling out to the registered compressor */
1144
1145 int err = 0;
1146
1147 *bytes_read = 0;
1148
1149 if (offset >= (off_t)hdr->uncompressed_size) {
1150 /* reading past end of file; nothing to do */
1151 err = 0;
1152 goto out;
1153 }
1154 if (offset < 0) {
1155 /* tried to read from before start of file */
1156 err = EINVAL;
1157 goto out;
1158 }
1159 if (hdr->uncompressed_size - offset < size) {
1160 /* adjust size so we don't read past the end of the file */
1161 size = (user_ssize_t)(hdr->uncompressed_size - offset);
1162 }
1163 if (size == 0) {
1164 /* nothing to read */
1165 err = 0;
1166 goto out;
1167 }
1168
1169 /*
1170 * Trace the following parameters on entry with event-id 0x03120008.
1171 *
1172 * @vp->v_id: vnode-id of the file being decompressed.
1173 * @hdr->compression_type: compression type.
1174 * @offset: offset from where to fetch uncompressed data.
1175 * @size: amount of uncompressed data to fetch.
1176 *
1177 * Please NOTE: @offset and @size can overflow in theory but
1178 * here it is safe.
1179 */
1180 DECMPFS_EMIT_TRACE_ENTRY(DECMPDBG_FETCH_UNCOMPRESSED_DATA, vp->v_id,
1181 hdr->compression_type, (int)offset, (int)size);
1182 lck_rw_lock_shared(&decompressorsLock);
1183 decmpfs_fetch_uncompressed_data_func fetch = decmp_get_func(vp, hdr->compression_type, fetch);
1184 if (fetch) {
1185 err = fetch(vp, decmpfs_ctx, hdr, offset, size, nvec, vec, bytes_read);
1186 lck_rw_unlock_shared(&decompressorsLock);
1187 if (err == 0) {
1188 uint64_t decompression_flags = decmpfs_cnode_get_decompression_flags(cp);
1189 if (decompression_flags & DECMPFS_FLAGS_FORCE_FLUSH_ON_DECOMPRESS) {
1190#if !defined(__i386__) && !defined(__x86_64__)
1191 int i;
1192 for (i = 0; i < nvec; i++) {
1193 assert(vec[i].size >= 0 && vec[i].size <= UINT_MAX);
1194 flush_dcache64((addr64_t)(uintptr_t)vec[i].buf, (unsigned int)vec[i].size, FALSE);
1195 }
1196#endif
1197 }
1198 }
1199 } else {
1200 err = ENOTSUP;
1201 lck_rw_unlock_shared(&decompressorsLock);
1202 }
1203 /*
1204 * Trace the following parameters on return with event-id 0x03120008.
1205 *
1206 * @vp->v_id: vnode-id of the file being decompressed.
1207 * @bytes_read: amount of uncompressed bytes fetched in bytes.
1208 * @err: value returned from this function.
1209 *
1210 * Please NOTE: @bytes_read can overflow in theory but here it is safe.
1211 */
1212 DECMPFS_EMIT_TRACE_RETURN(DECMPDBG_FETCH_UNCOMPRESSED_DATA, vp->v_id,
1213 (int)*bytes_read, err);
1214out:
1215 return err;
1216}
1217
1218static kern_return_t
1219commit_upl(upl_t upl, upl_offset_t pl_offset, size_t uplSize, int flags, int abort)
1220{
1221 kern_return_t kr = 0;
1222
1223#if CONFIG_IOSCHED
1224 upl_unmark_decmp(upl);
1225#endif /* CONFIG_IOSCHED */
1226
1227 /* commit the upl pages */
1228 if (abort) {
1229 VerboseLog("aborting upl, flags 0x%08x\n", flags);
1230 kr = ubc_upl_abort_range(upl, pl_offset, (upl_size_t)uplSize, flags);
1231 if (kr != KERN_SUCCESS) {
1232 ErrorLog("ubc_upl_abort_range error %d\n", (int)kr);
1233 }
1234 } else {
1235 VerboseLog("committing upl, flags 0x%08x\n", flags | UPL_COMMIT_CLEAR_DIRTY);
1236 kr = ubc_upl_commit_range(upl, pl_offset, (upl_size_t)uplSize, flags | UPL_COMMIT_CLEAR_DIRTY | UPL_COMMIT_WRITTEN_BY_KERNEL);
1237 if (kr != KERN_SUCCESS) {
1238 ErrorLog("ubc_upl_commit_range error %d\n", (int)kr);
1239 }
1240 }
1241 return kr;
1242}
1243
1244
1245errno_t
1246decmpfs_pagein_compressed(struct vnop_pagein_args *ap, int *is_compressed, decmpfs_cnode *cp)
1247{
1248 /* handles a page-in request from vfs for a compressed file */
1249
1250 int err = 0;
1251 vnode_t vp = ap->a_vp;
1252 upl_t pl = ap->a_pl;
1253 upl_offset_t pl_offset = ap->a_pl_offset;
1254 off_t f_offset = ap->a_f_offset;
1255 size_t size = ap->a_size;
1256 int flags = ap->a_flags;
1257 off_t uplPos = 0;
1258 user_ssize_t uplSize = 0;
1259 size_t verify_block_size = 0;
1260 void *data = NULL;
1261 decmpfs_header *hdr = NULL;
1262 size_t alloc_size = 0;
1263 uint64_t cachedSize = 0;
1264 int cmpdata_locked = 0;
1265 bool file_tail_page_valid = false;
1266 int num_valid_pages = 0;
1267 int num_invalid_pages = 0;
1268
1269 if (!decmpfs_trylock_compressed_data(cp, 0)) {
1270 return EAGAIN;
1271 }
1272 cmpdata_locked = 1;
1273
1274
1275 if (flags & ~(UPL_IOSYNC | UPL_NOCOMMIT | UPL_NORDAHEAD)) {
1276 DebugLogWithPath("pagein: unknown flags 0x%08x\n", (flags & ~(UPL_IOSYNC | UPL_NOCOMMIT | UPL_NORDAHEAD)));
1277 }
1278
1279 err = decmpfs_fetch_compressed_header(vp, cp, &hdr, 0, &alloc_size);
1280 if (err != 0) {
1281 goto out;
1282 }
1283
1284 cachedSize = hdr->uncompressed_size;
1285
1286 if (!compression_type_valid(vp, hdr)) {
1287 /* compressor not registered */
1288 err = ENOTSUP;
1289 goto out;
1290 }
1291
1292 /*
1293 * If the verify block size is larger than the page size, the UPL needs
1294 * to be aligned to it, Since the UPL has been created by the filesystem,
1295 * we will only check if the passed in UPL length conforms to the
1296 * alignment requirements.
1297 */
1298 err = VNOP_VERIFY(vp, f_offset, NULL, 0, &verify_block_size,
1299 VNODE_VERIFY_DEFAULT, NULL);
1300 if (err) {
1301 goto out;
1302 } else if (verify_block_size) {
1303 if (verify_block_size & (verify_block_size - 1)) {
1304 ErrorLogWithPath("verify block size is not power of 2, no verification will be done\n");
1305 err = EINVAL;
1306 } else if (size % verify_block_size) {
1307 ErrorLogWithPath("upl size is not a multiple of verify block size\n");
1308 err = EINVAL;
1309 }
1310 if (err) {
1311 goto out;
1312 }
1313 }
1314
1315#if CONFIG_IOSCHED
1316 /* Mark the UPL as the requesting UPL for decompression */
1317 upl_mark_decmp(pl);
1318#endif /* CONFIG_IOSCHED */
1319
1320 /* map the upl so we can fetch into it */
1321 kern_return_t kr = ubc_upl_map(pl, (vm_offset_t*)&data);
1322 if ((kr != KERN_SUCCESS) || (data == NULL)) {
1323 err = ENOSPC;
1324 data = NULL;
1325#if CONFIG_IOSCHED
1326 upl_unmark_decmp(pl);
1327#endif /* CONFIG_IOSCHED */
1328 goto out;
1329 }
1330
1331 uplPos = f_offset;
1332 uplSize = size;
1333
1334 /* clip the size to the size of the file */
1335 if ((uint64_t)uplPos + uplSize > cachedSize) {
1336 /* truncate the read to the size of the file */
1337 uplSize = (user_ssize_t)(cachedSize - uplPos);
1338 }
1339
1340 /* do the fetch */
1341 decmpfs_vector vec;
1342
1343decompress:
1344 /* the mapped data pointer points to the first page of the page list, so we want to start filling in at an offset of pl_offset */
1345 vec = (decmpfs_vector) {
1346 .buf = (char*)data + pl_offset,
1347 .size = size,
1348 };
1349
1350 uint64_t did_read = 0;
1351 if (decmpfs_fast_get_state(cp) == FILE_IS_CONVERTING) {
1352 ErrorLogWithPath("unexpected pagein during decompress\n");
1353 /*
1354 * if the file is converting, this must be a recursive call to pagein from underneath a call to decmpfs_decompress_file;
1355 * pretend that it succeeded but don't do anything since we're just going to write over the pages anyway
1356 */
1357 err = 0;
1358 } else {
1359 if (!verify_block_size || (verify_block_size <= PAGE_SIZE)) {
1360 err = decmpfs_fetch_uncompressed_data(vp, cp, hdr, uplPos, uplSize, 1, &vec, &did_read);
1361 } else {
1362 off_t l_uplPos = uplPos;
1363 off_t l_pl_offset = pl_offset;
1364 user_ssize_t l_uplSize = uplSize;
1365 upl_page_info_t *pl_info = ubc_upl_pageinfo(pl);
1366
1367 err = 0;
1368 /*
1369 * When the system page size is less than the "verify block size",
1370 * the UPL passed may not consist solely of absent pages.
1371 * We have to detect the "absent" pages and only decompress
1372 * into those absent/invalid page ranges.
1373 *
1374 * Things that will change in each iteration of the loop :
1375 *
1376 * l_pl_offset = where we are inside the UPL [0, caller_upl_created_size)
1377 * l_uplPos = the file offset the l_pl_offset corresponds to.
1378 * l_uplSize = the size of the upl still unprocessed;
1379 *
1380 * In this picture, we have to do the transfer on 2 ranges
1381 * (One 2 page range and one 3 page range) and the loop
1382 * below will skip the first two pages and then identify
1383 * the next two as invalid and fill those in and
1384 * then skip the next one and then do the last pages.
1385 *
1386 * uplPos(file_offset)
1387 * | uplSize
1388 * 0 V<--------------> file_size
1389 * |--------------------------------------------------->
1390 * | | |V|V|I|I|V|I|I|I|
1391 * ^
1392 * | upl
1393 * <------------------->
1394 * |
1395 * pl_offset
1396 *
1397 * uplSize will be clipped in case the UPL range exceeds
1398 * the file size.
1399 *
1400 */
1401 while (l_uplSize) {
1402 uint64_t l_did_read = 0;
1403 int pl_offset_pg = (int)(l_pl_offset / PAGE_SIZE);
1404 int pages_left_in_upl;
1405 int start_pg;
1406 int last_pg;
1407
1408 /*
1409 * l_uplSize may start off less than the size of the upl,
1410 * we have to round it up to PAGE_SIZE to calculate
1411 * how many more pages are left.
1412 */
1413 pages_left_in_upl = (int)(round_page((vm_offset_t)l_uplSize) / PAGE_SIZE);
1414
1415 /*
1416 * scan from the beginning of the upl looking for the first
1417 * non-valid page.... this will become the first page in
1418 * the request we're going to make to
1419 * 'decmpfs_fetch_uncompressed_data'... if all
1420 * of the pages are valid, we won't call through
1421 * to 'decmpfs_fetch_uncompressed_data'
1422 */
1423 for (start_pg = 0; start_pg < pages_left_in_upl; start_pg++) {
1424 if (!upl_valid_page(pl_info, pl_offset_pg + start_pg)) {
1425 break;
1426 }
1427 }
1428
1429 num_valid_pages += start_pg;
1430
1431 /*
1432 * scan from the starting invalid page looking for
1433 * a valid page before the end of the upl is
1434 * reached, if we find one, then it will be the
1435 * last page of the request to 'decmpfs_fetch_uncompressed_data'
1436 */
1437 for (last_pg = start_pg; last_pg < pages_left_in_upl; last_pg++) {
1438 if (upl_valid_page(pl_info, pl_offset_pg + last_pg)) {
1439 break;
1440 }
1441 }
1442
1443 if (start_pg < last_pg) {
1444 off_t inval_offset = start_pg * PAGE_SIZE;
1445 int inval_pages = last_pg - start_pg;
1446 int inval_size = inval_pages * PAGE_SIZE;
1447 decmpfs_vector l_vec;
1448
1449 num_invalid_pages += inval_pages;
1450 if (inval_offset) {
1451 did_read += inval_offset;
1452 l_pl_offset += inval_offset;
1453 l_uplPos += inval_offset;
1454 l_uplSize -= inval_offset;
1455 }
1456
1457 l_vec = (decmpfs_vector) {
1458 .buf = (char*)data + l_pl_offset,
1459 .size = inval_size,
1460 };
1461
1462 err = decmpfs_fetch_uncompressed_data(vp, cp, hdr, l_uplPos,
1463 MIN(l_uplSize, inval_size), 1, &l_vec, &l_did_read);
1464
1465 if (!err && (l_did_read != inval_size) && (l_uplSize > inval_size)) {
1466 ErrorLogWithPath("Unexpected size fetch of decompressed data, l_uplSize = %d, l_did_read = %d, inval_size = %d\n",
1467 (int)l_uplSize, (int)l_did_read, (int)inval_size);
1468 err = EINVAL;
1469 }
1470 } else {
1471 /* no invalid pages left */
1472 l_did_read = l_uplSize;
1473 if (uplSize < size) {
1474 file_tail_page_valid = true;
1475 }
1476 }
1477
1478 if (err) {
1479 break;
1480 }
1481
1482 did_read += l_did_read;
1483 l_pl_offset += l_did_read;
1484 l_uplPos += l_did_read;
1485 l_uplSize -= l_did_read;
1486 }
1487 }
1488 }
1489 if (err) {
1490 DebugLogWithPath("decmpfs_fetch_uncompressed_data err %d\n", err);
1491 int cmp_state = decmpfs_fast_get_state(cp);
1492 if (cmp_state == FILE_IS_CONVERTING) {
1493 DebugLogWithPath("cmp_state == FILE_IS_CONVERTING\n");
1494 cmp_state = wait_for_decompress(cp);
1495 if (cmp_state == FILE_IS_COMPRESSED) {
1496 DebugLogWithPath("cmp_state == FILE_IS_COMPRESSED\n");
1497 /* a decompress was attempted but it failed, let's try calling fetch again */
1498 goto decompress;
1499 }
1500 }
1501 if (cmp_state == FILE_IS_NOT_COMPRESSED) {
1502 DebugLogWithPath("cmp_state == FILE_IS_NOT_COMPRESSED\n");
1503 /* the file was decompressed after we started reading it */
1504 *is_compressed = 0; /* instruct caller to fall back to its normal path */
1505 }
1506 }
1507
1508 /* zero out whatever we didn't read, and zero out the end of the last page(s) */
1509 uint64_t total_size = (size + (PAGE_SIZE - 1)) & ~(PAGE_SIZE - 1);
1510 if (did_read < total_size && !(verify_block_size && err)) {
1511 uint64_t rounded_up_did_read = file_tail_page_valid ? (uint64_t)(round_page((vm_offset_t)did_read)) : did_read;
1512 memset((char*)vec.buf + rounded_up_did_read, 0, (size_t)(total_size - rounded_up_did_read));
1513 }
1514
1515 if (!err && verify_block_size) {
1516 size_t cur_verify_block_size = verify_block_size;
1517
1518 if ((err = VNOP_VERIFY(vp, uplPos, vec.buf, size, &cur_verify_block_size, 0, NULL))) {
1519 ErrorLogWithPath("Verification failed with error %d, uplPos = %lld, uplSize = %d, did_read = %d, total_size = %d, valid_pages = %d, invalid_pages = %d, tail_page_valid = %d\n",
1520 err, (long long)uplPos, (int)uplSize, (int)did_read, (int)total_size, num_valid_pages, num_invalid_pages, file_tail_page_valid);
1521 }
1522 /* XXX : If the verify block size changes, redo the read */
1523 }
1524
1525#if CONFIG_IOSCHED
1526 upl_unmark_decmp(pl);
1527#endif /* CONFIG_IOSCHED */
1528
1529 kr = ubc_upl_unmap(pl); data = NULL; /* make sure to set data to NULL so we don't try to unmap again below */
1530 if (kr != KERN_SUCCESS) {
1531 ErrorLogWithPath("ubc_upl_unmap error %d\n", (int)kr);
1532 } else {
1533 if (!err) {
1534 /* commit our pages */
1535 kr = commit_upl(pl, pl_offset, (size_t)total_size, UPL_COMMIT_FREE_ON_EMPTY, 0);
1536 }
1537 }
1538
1539out:
1540 if (data) {
1541 ubc_upl_unmap(pl);
1542 }
1543 if (hdr != NULL) {
1544 kheap_free(KHEAP_TEMP, hdr, alloc_size);
1545 }
1546 if (cmpdata_locked) {
1547 decmpfs_unlock_compressed_data(cp, 0);
1548 }
1549 if (err) {
1550#if 0
1551 if (err != ENXIO && err != ENOSPC) {
1552 char *path = zalloc(ZV_NAMEI);
1553 panic("%s: decmpfs_pagein_compressed: err %d", vnpath(vp, path, PATH_MAX), err);
1554 zfree(ZV_NAMEI, path);
1555 }
1556#endif /* 0 */
1557 ErrorLogWithPath("err %d\n", err);
1558 }
1559 return err;
1560}
1561
1562errno_t
1563decmpfs_read_compressed(struct vnop_read_args *ap, int *is_compressed, decmpfs_cnode *cp)
1564{
1565 /* handles a read request from vfs for a compressed file */
1566
1567 uio_t uio = ap->a_uio;
1568 vnode_t vp = ap->a_vp;
1569 int err = 0;
1570 int countInt = 0;
1571 off_t uplPos = 0;
1572 user_ssize_t uplSize = 0;
1573 user_ssize_t uplRemaining = 0;
1574 off_t curUplPos = 0;
1575 user_ssize_t curUplSize = 0;
1576 kern_return_t kr = KERN_SUCCESS;
1577 int abort_read = 0;
1578 void *data = NULL;
1579 uint64_t did_read = 0;
1580 upl_t upl = NULL;
1581 upl_page_info_t *pli = NULL;
1582 decmpfs_header *hdr = NULL;
1583 size_t alloc_size = 0;
1584 uint64_t cachedSize = 0;
1585 off_t uioPos = 0;
1586 user_ssize_t uioRemaining = 0;
1587 size_t verify_block_size = 0;
1588 size_t alignment_size = PAGE_SIZE;
1589 int cmpdata_locked = 0;
1590
1591 decmpfs_lock_compressed_data(cp, 0); cmpdata_locked = 1;
1592
1593 uplPos = uio_offset(uio);
1594 uplSize = uio_resid(uio);
1595 VerboseLogWithPath("uplPos %lld uplSize %lld\n", uplPos, uplSize);
1596
1597 cachedSize = decmpfs_cnode_get_vnode_cached_size(cp);
1598
1599 if ((uint64_t)uplPos + uplSize > cachedSize) {
1600 /* truncate the read to the size of the file */
1601 uplSize = (user_ssize_t)(cachedSize - uplPos);
1602 }
1603
1604 /* give the cluster layer a chance to fill in whatever it already has */
1605 countInt = (uplSize > INT_MAX) ? INT_MAX : (int)uplSize;
1606 err = cluster_copy_ubc_data(vp, uio, &countInt, 0);
1607 if (err != 0) {
1608 goto out;
1609 }
1610
1611 /* figure out what's left */
1612 uioPos = uio_offset(uio);
1613 uioRemaining = uio_resid(uio);
1614 if ((uint64_t)uioPos + uioRemaining > cachedSize) {
1615 /* truncate the read to the size of the file */
1616 uioRemaining = (user_ssize_t)(cachedSize - uioPos);
1617 }
1618
1619 if (uioRemaining <= 0) {
1620 /* nothing left */
1621 goto out;
1622 }
1623
1624 err = decmpfs_fetch_compressed_header(vp, cp, &hdr, 0, &alloc_size);
1625 if (err != 0) {
1626 goto out;
1627 }
1628 if (!compression_type_valid(vp, hdr)) {
1629 err = ENOTSUP;
1630 goto out;
1631 }
1632
1633 uplPos = uioPos;
1634 uplSize = uioRemaining;
1635#if COMPRESSION_DEBUG
1636 DebugLogWithPath("uplPos %lld uplSize %lld\n", (uint64_t)uplPos, (uint64_t)uplSize);
1637#endif
1638
1639 lck_rw_lock_shared(&decompressorsLock);
1640 decmpfs_adjust_fetch_region_func adjust_fetch = decmp_get_func(vp, hdr->compression_type, adjust_fetch);
1641 if (adjust_fetch) {
1642 /* give the compressor a chance to adjust the portion of the file that we read */
1643 adjust_fetch(vp, decmpfs_ctx, hdr, &uplPos, &uplSize);
1644 VerboseLogWithPath("adjusted uplPos %lld uplSize %lld\n", (uint64_t)uplPos, (uint64_t)uplSize);
1645 }
1646 lck_rw_unlock_shared(&decompressorsLock);
1647
1648 /* clip the adjusted size to the size of the file */
1649 if ((uint64_t)uplPos + uplSize > cachedSize) {
1650 /* truncate the read to the size of the file */
1651 uplSize = (user_ssize_t)(cachedSize - uplPos);
1652 }
1653
1654 if (uplSize <= 0) {
1655 /* nothing left */
1656 goto out;
1657 }
1658
1659 /*
1660 * since we're going to create a upl for the given region of the file,
1661 * make sure we're on page boundaries
1662 */
1663
1664 /* If the verify block size is larger than the page size, the UPL needs to aligned to it */
1665 err = VNOP_VERIFY(vp, uplPos, NULL, 0, &verify_block_size, VNODE_VERIFY_DEFAULT, NULL);
1666 if (err) {
1667 goto out;
1668 } else if (verify_block_size) {
1669 if (verify_block_size & (verify_block_size - 1)) {
1670 ErrorLogWithPath("verify block size is not power of 2, no verification will be done\n");
1671 verify_block_size = 0;
1672 } else if (verify_block_size > PAGE_SIZE) {
1673 alignment_size = verify_block_size;
1674 }
1675 }
1676
1677 if (uplPos & (alignment_size - 1)) {
1678 /* round position down to page boundary */
1679 uplSize += (uplPos & (alignment_size - 1));
1680 uplPos &= ~(alignment_size - 1);
1681 }
1682
1683 /* round size up to alignement_size multiple */
1684 uplSize = (uplSize + (alignment_size - 1)) & ~(alignment_size - 1);
1685
1686 VerboseLogWithPath("new uplPos %lld uplSize %lld\n", (uint64_t)uplPos, (uint64_t)uplSize);
1687
1688 uplRemaining = uplSize;
1689 curUplPos = uplPos;
1690 curUplSize = 0;
1691
1692 while (uplRemaining > 0) {
1693 /* start after the last upl */
1694 curUplPos += curUplSize;
1695
1696 /* clip to max upl size */
1697 curUplSize = uplRemaining;
1698 if (curUplSize > MAX_UPL_SIZE_BYTES) {
1699 curUplSize = MAX_UPL_SIZE_BYTES;
1700 }
1701
1702 /* create the upl */
1703 kr = ubc_create_upl_kernel(vp, curUplPos, (int)curUplSize, &upl, &pli, UPL_SET_LITE, VM_KERN_MEMORY_FILE);
1704 if (kr != KERN_SUCCESS) {
1705 ErrorLogWithPath("ubc_create_upl error %d\n", (int)kr);
1706 err = EINVAL;
1707 goto out;
1708 }
1709 VerboseLogWithPath("curUplPos %lld curUplSize %lld\n", (uint64_t)curUplPos, (uint64_t)curUplSize);
1710
1711#if CONFIG_IOSCHED
1712 /* Mark the UPL as the requesting UPL for decompression */
1713 upl_mark_decmp(upl);
1714#endif /* CONFIG_IOSCHED */
1715
1716 /* map the upl */
1717 kr = ubc_upl_map(upl, (vm_offset_t*)&data);
1718 if (kr != KERN_SUCCESS) {
1719 commit_upl(upl, 0, curUplSize, UPL_ABORT_FREE_ON_EMPTY, 1);
1720#if 0
1721 char *path = zalloc(ZV_NAMEI);
1722 panic("%s: decmpfs_read_compressed: ubc_upl_map error %d", vnpath(vp, path, PATH_MAX), (int)kr);
1723 zfree(ZV_NAMEI, path);
1724#else /* 0 */
1725 ErrorLogWithPath("ubc_upl_map kr=0x%x\n", (int)kr);
1726#endif /* 0 */
1727 err = EINVAL;
1728 goto out;
1729 }
1730
1731 /* make sure the map succeeded */
1732 if (!data) {
1733 commit_upl(upl, 0, curUplSize, UPL_ABORT_FREE_ON_EMPTY, 1);
1734
1735 ErrorLogWithPath("ubc_upl_map mapped null\n");
1736 err = EINVAL;
1737 goto out;
1738 }
1739
1740 /* fetch uncompressed data into the mapped upl */
1741 decmpfs_vector vec;
1742decompress:
1743 vec = (decmpfs_vector){ .buf = data, .size = curUplSize };
1744 err = decmpfs_fetch_uncompressed_data(vp, cp, hdr, curUplPos, curUplSize, 1, &vec, &did_read);
1745 if (err) {
1746 ErrorLogWithPath("decmpfs_fetch_uncompressed_data err %d\n", err);
1747
1748 /* maybe the file is converting to decompressed */
1749 int cmp_state = decmpfs_fast_get_state(cp);
1750 if (cmp_state == FILE_IS_CONVERTING) {
1751 ErrorLogWithPath("cmp_state == FILE_IS_CONVERTING\n");
1752 cmp_state = wait_for_decompress(cp);
1753 if (cmp_state == FILE_IS_COMPRESSED) {
1754 ErrorLogWithPath("cmp_state == FILE_IS_COMPRESSED\n");
1755 /* a decompress was attempted but it failed, let's try fetching again */
1756 goto decompress;
1757 }
1758 }
1759 if (cmp_state == FILE_IS_NOT_COMPRESSED) {
1760 ErrorLogWithPath("cmp_state == FILE_IS_NOT_COMPRESSED\n");
1761 /* the file was decompressed after we started reading it */
1762 abort_read = 1; /* we're not going to commit our data */
1763 *is_compressed = 0; /* instruct caller to fall back to its normal path */
1764 }
1765 kr = KERN_FAILURE;
1766 did_read = 0;
1767 }
1768
1769 /* zero out the remainder of the last page */
1770 memset((char*)data + did_read, 0, (size_t)(curUplSize - did_read));
1771 if (!err && verify_block_size) {
1772 size_t cur_verify_block_size = verify_block_size;
1773
1774 if ((err = VNOP_VERIFY(vp, curUplPos, data, curUplSize, &cur_verify_block_size, 0, NULL))) {
1775 ErrorLogWithPath("Verification failed with error %d\n", err);
1776 abort_read = 1;
1777 }
1778 /* XXX : If the verify block size changes, redo the read */
1779 }
1780
1781 kr = ubc_upl_unmap(upl);
1782 if (kr == KERN_SUCCESS) {
1783 if (abort_read) {
1784 kr = commit_upl(upl, 0, curUplSize, UPL_ABORT_FREE_ON_EMPTY, 1);
1785 } else {
1786 VerboseLogWithPath("uioPos %lld uioRemaining %lld\n", (uint64_t)uioPos, (uint64_t)uioRemaining);
1787 if (uioRemaining) {
1788 off_t uplOff = uioPos - curUplPos;
1789 if (uplOff < 0) {
1790 ErrorLogWithPath("uplOff %lld should never be negative\n", (int64_t)uplOff);
1791 err = EINVAL;
1792 } else if (uplOff > INT_MAX) {
1793 ErrorLogWithPath("uplOff %lld too large\n", (int64_t)uplOff);
1794 err = EINVAL;
1795 } else {
1796 off_t count = curUplPos + curUplSize - uioPos;
1797 if (count < 0) {
1798 /* this upl is entirely before the uio */
1799 } else {
1800 if (count > uioRemaining) {
1801 count = uioRemaining;
1802 }
1803 int icount = (count > INT_MAX) ? INT_MAX : (int)count;
1804 int io_resid = icount;
1805 err = cluster_copy_upl_data(uio, upl, (int)uplOff, &io_resid);
1806 int copied = icount - io_resid;
1807 VerboseLogWithPath("uplOff %lld count %lld copied %lld\n", (uint64_t)uplOff, (uint64_t)count, (uint64_t)copied);
1808 if (err) {
1809 ErrorLogWithPath("cluster_copy_upl_data err %d\n", err);
1810 }
1811 uioPos += copied;
1812 uioRemaining -= copied;
1813 }
1814 }
1815 }
1816 kr = commit_upl(upl, 0, curUplSize, UPL_COMMIT_FREE_ON_EMPTY | UPL_COMMIT_INACTIVATE, 0);
1817 if (err) {
1818 goto out;
1819 }
1820 }
1821 } else {
1822 ErrorLogWithPath("ubc_upl_unmap error %d\n", (int)kr);
1823 }
1824
1825 uplRemaining -= curUplSize;
1826 }
1827
1828out:
1829
1830 if (hdr != NULL) {
1831 kheap_free(KHEAP_TEMP, hdr, alloc_size);
1832 }
1833 if (cmpdata_locked) {
1834 decmpfs_unlock_compressed_data(cp, 0);
1835 }
1836 if (err) {/* something went wrong */
1837 ErrorLogWithPath("err %d\n", err);
1838 return err;
1839 }
1840
1841#if COMPRESSION_DEBUG
1842 uplSize = uio_resid(uio);
1843 if (uplSize) {
1844 VerboseLogWithPath("still %lld bytes to copy\n", uplSize);
1845 }
1846#endif
1847 return 0;
1848}
1849
1850int
1851decmpfs_free_compressed_data(vnode_t vp, decmpfs_cnode *cp)
1852{
1853 /*
1854 * call out to the decompressor to free remove any data associated with this compressed file
1855 * then delete the file's compression xattr
1856 */
1857 decmpfs_header *hdr = NULL;
1858 size_t alloc_size = 0;
1859
1860 /*
1861 * Trace the following parameters on entry with event-id 0x03120010.
1862 *
1863 * @vp->v_id: vnode-id of the file for which to free compressed data.
1864 */
1865 DECMPFS_EMIT_TRACE_ENTRY(DECMPDBG_FREE_COMPRESSED_DATA, vp->v_id);
1866
1867 int err = decmpfs_fetch_compressed_header(vp, cp, &hdr, 0, &alloc_size);
1868 if (err) {
1869 ErrorLogWithPath("decmpfs_fetch_compressed_header err %d\n", err);
1870 } else {
1871 lck_rw_lock_shared(&decompressorsLock);
1872 decmpfs_free_compressed_data_func free_data = decmp_get_func(vp, hdr->compression_type, free_data);
1873 if (free_data) {
1874 err = free_data(vp, decmpfs_ctx, hdr);
1875 } else {
1876 /* nothing to do, so no error */
1877 err = 0;
1878 }
1879 lck_rw_unlock_shared(&decompressorsLock);
1880
1881 if (err != 0) {
1882 ErrorLogWithPath("decompressor err %d\n", err);
1883 }
1884 }
1885 /*
1886 * Trace the following parameters on return with event-id 0x03120010.
1887 *
1888 * @vp->v_id: vnode-id of the file for which to free compressed data.
1889 * @err: value returned from this function.
1890 */
1891 DECMPFS_EMIT_TRACE_RETURN(DECMPDBG_FREE_COMPRESSED_DATA, vp->v_id, err);
1892
1893 /* delete the xattr */
1894 err = vn_removexattr(vp, DECMPFS_XATTR_NAME, 0, decmpfs_ctx);
1895
1896 if (hdr != NULL) {
1897 kheap_free(KHEAP_TEMP, hdr, alloc_size);
1898 }
1899 return err;
1900}
1901
1902#pragma mark --- file conversion routines ---
1903
1904static int
1905unset_compressed_flag(vnode_t vp)
1906{
1907 int err = 0;
1908 struct vnode_attr va;
1909 int new_bsdflags = 0;
1910
1911 VATTR_INIT(&va);
1912 VATTR_WANTED(&va, va_flags);
1913 err = vnode_getattr(vp, &va, decmpfs_ctx);
1914
1915 if (err != 0) {
1916 ErrorLogWithPath("vnode_getattr err %d\n", err);
1917 } else {
1918 new_bsdflags = va.va_flags & ~UF_COMPRESSED;
1919
1920 VATTR_INIT(&va);
1921 VATTR_SET(&va, va_flags, new_bsdflags);
1922 err = vnode_setattr(vp, &va, decmpfs_ctx);
1923 if (err != 0) {
1924 ErrorLogWithPath("vnode_setattr err %d\n", err);
1925 }
1926 }
1927 return err;
1928}
1929
1930int
1931decmpfs_decompress_file(vnode_t vp, decmpfs_cnode *cp, off_t toSize, int truncate_okay, int skiplock)
1932{
1933 /* convert a compressed file to an uncompressed file */
1934
1935 int err = 0;
1936 char *data = NULL;
1937 uio_t uio_w = 0;
1938 off_t offset = 0;
1939 uint32_t old_state = 0;
1940 uint32_t new_state = 0;
1941 int update_file_state = 0;
1942 size_t allocSize = 0;
1943 decmpfs_header *hdr = NULL;
1944 size_t hdr_size = 0;
1945 int cmpdata_locked = 0;
1946 off_t remaining = 0;
1947 uint64_t uncompressed_size = 0;
1948
1949 /*
1950 * Trace the following parameters on entry with event-id 0x03120000.
1951 *
1952 * @vp->v_id: vnode-id of the file being decompressed.
1953 * @toSize: uncompress given bytes of the file.
1954 * @truncate_okay: on error it is OK to truncate.
1955 * @skiplock: compressed data is locked, skip locking again.
1956 *
1957 * Please NOTE: @toSize can overflow in theory but here it is safe.
1958 */
1959 DECMPFS_EMIT_TRACE_ENTRY(DECMPDBG_DECOMPRESS_FILE, vp->v_id,
1960 (int)toSize, truncate_okay, skiplock);
1961
1962 if (!skiplock) {
1963 decmpfs_lock_compressed_data(cp, 1); cmpdata_locked = 1;
1964 }
1965
1966decompress:
1967 old_state = decmpfs_fast_get_state(cp);
1968
1969 switch (old_state) {
1970 case FILE_IS_NOT_COMPRESSED:
1971 {
1972 /* someone else decompressed the file */
1973 err = 0;
1974 goto out;
1975 }
1976
1977 case FILE_TYPE_UNKNOWN:
1978 {
1979 /* the file is in an unknown state, so update the state and retry */
1980 (void)decmpfs_file_is_compressed(vp, cp);
1981
1982 /* try again */
1983 goto decompress;
1984 }
1985
1986 case FILE_IS_COMPRESSED:
1987 {
1988 /* the file is compressed, so decompress it */
1989 break;
1990 }
1991
1992 default:
1993 {
1994 /*
1995 * this shouldn't happen since multiple calls to decmpfs_decompress_file lock each other out,
1996 * and when decmpfs_decompress_file returns, the state should be always be set back to
1997 * FILE_IS_NOT_COMPRESSED or FILE_IS_UNKNOWN
1998 */
1999 err = EINVAL;
2000 goto out;
2001 }
2002 }
2003
2004 err = decmpfs_fetch_compressed_header(vp, cp, &hdr, 0, &hdr_size);
2005 if (err != 0) {
2006 goto out;
2007 }
2008
2009 uncompressed_size = hdr->uncompressed_size;
2010 if (toSize == -1) {
2011 toSize = hdr->uncompressed_size;
2012 }
2013
2014 if (toSize == 0) {
2015 /* special case truncating the file to zero bytes */
2016 goto nodecmp;
2017 } else if ((uint64_t)toSize > hdr->uncompressed_size) {
2018 /* the caller is trying to grow the file, so we should decompress all the data */
2019 toSize = hdr->uncompressed_size;
2020 }
2021
2022 allocSize = MIN(64 * 1024, (size_t)toSize);
2023 data = kheap_alloc(KHEAP_TEMP, allocSize, Z_WAITOK);
2024 if (!data) {
2025 err = ENOMEM;
2026 goto out;
2027 }
2028
2029 uio_w = uio_create(1, 0LL, UIO_SYSSPACE, UIO_WRITE);
2030 if (!uio_w) {
2031 err = ENOMEM;
2032 goto out;
2033 }
2034 uio_w->uio_flags |= UIO_FLAGS_IS_COMPRESSED_FILE;
2035
2036 remaining = toSize;
2037
2038 /* tell the buffer cache that this is an empty file */
2039 ubc_setsize(vp, 0);
2040
2041 /* if we got here, we need to decompress the file */
2042 decmpfs_cnode_set_vnode_state(cp, FILE_IS_CONVERTING, 1);
2043
2044 while (remaining > 0) {
2045 /* loop decompressing data from the file and writing it into the data fork */
2046
2047 uint64_t bytes_read = 0;
2048 decmpfs_vector vec = { .buf = data, .size = (user_ssize_t)MIN(allocSize, remaining) };
2049 err = decmpfs_fetch_uncompressed_data(vp, cp, hdr, offset, vec.size, 1, &vec, &bytes_read);
2050 if (err != 0) {
2051 ErrorLogWithPath("decmpfs_fetch_uncompressed_data err %d\n", err);
2052 goto out;
2053 }
2054
2055 if (bytes_read == 0) {
2056 /* we're done reading data */
2057 break;
2058 }
2059
2060 uio_reset(uio_w, offset, UIO_SYSSPACE, UIO_WRITE);
2061 err = uio_addiov(uio_w, CAST_USER_ADDR_T(data), (user_size_t)bytes_read);
2062 if (err != 0) {
2063 ErrorLogWithPath("uio_addiov err %d\n", err);
2064 err = ENOMEM;
2065 goto out;
2066 }
2067
2068 err = VNOP_WRITE(vp, uio_w, 0, decmpfs_ctx);
2069 if (err != 0) {
2070 /* if the write failed, truncate the file to zero bytes */
2071 ErrorLogWithPath("VNOP_WRITE err %d\n", err);
2072 break;
2073 }
2074 offset += bytes_read;
2075 remaining -= bytes_read;
2076 }
2077
2078 if (err == 0) {
2079 if (offset != toSize) {
2080 ErrorLogWithPath("file decompressed to %lld instead of %lld\n", offset, toSize);
2081 err = EINVAL;
2082 goto out;
2083 }
2084 }
2085
2086 if (err == 0) {
2087 /* sync the data and metadata */
2088 err = VNOP_FSYNC(vp, MNT_WAIT, decmpfs_ctx);
2089 if (err != 0) {
2090 ErrorLogWithPath("VNOP_FSYNC err %d\n", err);
2091 goto out;
2092 }
2093 }
2094
2095 if (err != 0) {
2096 /* write, setattr, or fsync failed */
2097 ErrorLogWithPath("aborting decompress, err %d\n", err);
2098 if (truncate_okay) {
2099 /* truncate anything we might have written */
2100 int error = vnode_setsize(vp, 0, 0, decmpfs_ctx);
2101 ErrorLogWithPath("vnode_setsize err %d\n", error);
2102 }
2103 goto out;
2104 }
2105
2106nodecmp:
2107 /* if we're truncating the file to zero bytes, we'll skip ahead to here */
2108
2109 /* unset the compressed flag */
2110 unset_compressed_flag(vp);
2111
2112 /* free the compressed data associated with this file */
2113 err = decmpfs_free_compressed_data(vp, cp);
2114 if (err != 0) {
2115 ErrorLogWithPath("decmpfs_free_compressed_data err %d\n", err);
2116 }
2117
2118 /*
2119 * even if free_compressed_data or vnode_getattr/vnode_setattr failed, return success
2120 * since we succeeded in writing all of the file data to the data fork
2121 */
2122 err = 0;
2123
2124 /* if we got this far, the file was successfully decompressed */
2125 update_file_state = 1;
2126 new_state = FILE_IS_NOT_COMPRESSED;
2127
2128#if COMPRESSION_DEBUG
2129 {
2130 uint64_t filesize = 0;
2131 vnsize(vp, &filesize);
2132 DebugLogWithPath("new file size %lld\n", filesize);
2133 }
2134#endif
2135
2136out:
2137 if (hdr != NULL) {
2138 kheap_free(KHEAP_TEMP, hdr, hdr_size);
2139 }
2140 kheap_free(KHEAP_TEMP, data, allocSize);
2141 if (uio_w) {
2142 uio_free(uio_w);
2143 }
2144
2145 if (err != 0) {
2146 /* if there was a failure, reset compression flags to unknown and clear the buffer cache data */
2147 update_file_state = 1;
2148 new_state = FILE_TYPE_UNKNOWN;
2149 if (uncompressed_size) {
2150 ubc_setsize(vp, 0);
2151 ubc_setsize(vp, uncompressed_size);
2152 }
2153 }
2154
2155 if (update_file_state) {
2156 lck_mtx_lock(&decompress_channel_mtx);
2157 decmpfs_cnode_set_vnode_state(cp, new_state, 1);
2158 wakeup((caddr_t)&decompress_channel); /* wake up anyone who might have been waiting for decompression */
2159 lck_mtx_unlock(&decompress_channel_mtx);
2160 }
2161
2162 if (cmpdata_locked) {
2163 decmpfs_unlock_compressed_data(cp, 1);
2164 }
2165 /*
2166 * Trace the following parameters on return with event-id 0x03120000.
2167 *
2168 * @vp->v_id: vnode-id of the file being decompressed.
2169 * @err: value returned from this function.
2170 */
2171 DECMPFS_EMIT_TRACE_RETURN(DECMPDBG_DECOMPRESS_FILE, vp->v_id, err);
2172 return err;
2173}
2174
2175#pragma mark --- Type1 compressor ---
2176
2177/*
2178 * The "Type1" compressor stores the data fork directly in the compression xattr
2179 */
2180
2181static int
2182decmpfs_validate_compressed_file_Type1(__unused vnode_t vp, __unused vfs_context_t ctx, decmpfs_header *hdr)
2183{
2184 int err = 0;
2185
2186 if (hdr->uncompressed_size + sizeof(decmpfs_disk_header) != (uint64_t)hdr->attr_size) {
2187 err = EINVAL;
2188 goto out;
2189 }
2190out:
2191 return err;
2192}
2193
2194static int
2195decmpfs_fetch_uncompressed_data_Type1(__unused vnode_t vp, __unused vfs_context_t ctx, decmpfs_header *hdr, off_t offset, user_ssize_t size, int nvec, decmpfs_vector *vec, uint64_t *bytes_read)
2196{
2197 int err = 0;
2198 int i;
2199 user_ssize_t remaining;
2200
2201 if (hdr->uncompressed_size + sizeof(decmpfs_disk_header) != (uint64_t)hdr->attr_size) {
2202 err = EINVAL;
2203 goto out;
2204 }
2205
2206#if COMPRESSION_DEBUG
2207 static int dummy = 0; // prevent syslog from coalescing printfs
2208 DebugLogWithPath("%d memcpy %lld at %lld\n", dummy++, size, (uint64_t)offset);
2209#endif
2210
2211 remaining = size;
2212 for (i = 0; (i < nvec) && (remaining > 0); i++) {
2213 user_ssize_t curCopy = vec[i].size;
2214 if (curCopy > remaining) {
2215 curCopy = remaining;
2216 }
2217 memcpy(vec[i].buf, hdr->attr_bytes + offset, curCopy);
2218 offset += curCopy;
2219 remaining -= curCopy;
2220 }
2221
2222 if ((bytes_read) && (err == 0)) {
2223 *bytes_read = (size - remaining);
2224 }
2225
2226out:
2227 return err;
2228}
2229
2230SECURITY_READ_ONLY_EARLY(static decmpfs_registration) Type1Reg =
2231{
2232 .decmpfs_registration = DECMPFS_REGISTRATION_VERSION,
2233 .validate = decmpfs_validate_compressed_file_Type1,
2234 .adjust_fetch = NULL,/* no adjust necessary */
2235 .fetch = decmpfs_fetch_uncompressed_data_Type1,
2236 .free_data = NULL,/* no free necessary */
2237 .get_flags = NULL/* no flags */
2238};
2239
2240#pragma mark --- decmpfs initialization ---
2241
2242void
2243decmpfs_init(void)
2244{
2245 static int done = 0;
2246 if (done) {
2247 return;
2248 }
2249
2250 decmpfs_ctx = vfs_context_create(vfs_context_kernel());
2251
2252 register_decmpfs_decompressor(CMP_Type1, &Type1Reg);
2253
2254 done = 1;
2255}
2256#endif /* FS_COMPRESSION */