• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright 2019 Google LLC
4  */
5 #include <linux/crc32.h>
6 #include <linux/file.h>
7 #include <linux/fsverity.h>
8 #include <linux/gfp.h>
9 #include <linux/kobject.h>
10 #include <linux/ktime.h>
11 #include <linux/lz4.h>
12 #include <linux/mm.h>
13 #include <linux/namei.h>
14 #include <linux/pagemap.h>
15 #include <linux/slab.h>
16 #include <linux/types.h>
17 #include <linux/workqueue.h>
18 
19 #include "data_mgmt.h"
20 #include "format.h"
21 #include "integrity.h"
22 #include "sysfs.h"
23 #include "verity.h"
24 
25 static int incfs_scan_metadata_chain(struct data_file *df);
26 
log_wake_up_all(struct work_struct * work)27 static void log_wake_up_all(struct work_struct *work)
28 {
29 	struct delayed_work *dw = container_of(work, struct delayed_work, work);
30 	struct read_log *rl = container_of(dw, struct read_log, ml_wakeup_work);
31 	wake_up_all(&rl->ml_notif_wq);
32 }
33 
zstd_free_workspace(struct work_struct * work)34 static void zstd_free_workspace(struct work_struct *work)
35 {
36 	struct delayed_work *dw = container_of(work, struct delayed_work, work);
37 	struct mount_info *mi =
38 		container_of(dw, struct mount_info, mi_zstd_cleanup_work);
39 
40 	mutex_lock(&mi->mi_zstd_workspace_mutex);
41 	kvfree(mi->mi_zstd_workspace);
42 	mi->mi_zstd_workspace = NULL;
43 	mi->mi_zstd_stream = NULL;
44 	mutex_unlock(&mi->mi_zstd_workspace_mutex);
45 }
46 
incfs_alloc_mount_info(struct super_block * sb,struct mount_options * options,struct path * backing_dir_path)47 struct mount_info *incfs_alloc_mount_info(struct super_block *sb,
48 					  struct mount_options *options,
49 					  struct path *backing_dir_path)
50 {
51 	struct mount_info *mi = NULL;
52 	int error = 0;
53 	struct incfs_sysfs_node *node;
54 
55 	mi = kzalloc(sizeof(*mi), GFP_NOFS);
56 	if (!mi)
57 		return ERR_PTR(-ENOMEM);
58 
59 	mi->mi_sb = sb;
60 	mi->mi_backing_dir_path = *backing_dir_path;
61 	mi->mi_owner = get_current_cred();
62 	path_get(&mi->mi_backing_dir_path);
63 	mutex_init(&mi->mi_dir_struct_mutex);
64 	init_waitqueue_head(&mi->mi_pending_reads_notif_wq);
65 	init_waitqueue_head(&mi->mi_log.ml_notif_wq);
66 	init_waitqueue_head(&mi->mi_blocks_written_notif_wq);
67 	atomic_set(&mi->mi_blocks_written, 0);
68 	INIT_DELAYED_WORK(&mi->mi_log.ml_wakeup_work, log_wake_up_all);
69 	spin_lock_init(&mi->mi_log.rl_lock);
70 	spin_lock_init(&mi->pending_read_lock);
71 	INIT_LIST_HEAD(&mi->mi_reads_list_head);
72 	spin_lock_init(&mi->mi_per_uid_read_timeouts_lock);
73 	mutex_init(&mi->mi_zstd_workspace_mutex);
74 	INIT_DELAYED_WORK(&mi->mi_zstd_cleanup_work, zstd_free_workspace);
75 	mutex_init(&mi->mi_le_mutex);
76 
77 	node = incfs_add_sysfs_node(options->sysfs_name, mi);
78 	if (IS_ERR(node)) {
79 		error = PTR_ERR(node);
80 		goto err;
81 	}
82 	mi->mi_sysfs_node = node;
83 
84 	error = incfs_realloc_mount_info(mi, options);
85 	if (error)
86 		goto err;
87 
88 	return mi;
89 
90 err:
91 	incfs_free_mount_info(mi);
92 	return ERR_PTR(error);
93 }
94 
incfs_realloc_mount_info(struct mount_info * mi,struct mount_options * options)95 int incfs_realloc_mount_info(struct mount_info *mi,
96 			     struct mount_options *options)
97 {
98 	void *new_buffer = NULL;
99 	void *old_buffer;
100 	size_t new_buffer_size = 0;
101 
102 	if (options->read_log_pages != mi->mi_options.read_log_pages) {
103 		struct read_log_state log_state;
104 		/*
105 		 * Even though having two buffers allocated at once isn't
106 		 * usually good, allocating a multipage buffer under a spinlock
107 		 * is even worse, so let's optimize for the shorter lock
108 		 * duration. It's not end of the world if we fail to increase
109 		 * the buffer size anyway.
110 		 */
111 		if (options->read_log_pages > 0) {
112 			new_buffer_size = PAGE_SIZE * options->read_log_pages;
113 			new_buffer = kzalloc(new_buffer_size, GFP_NOFS);
114 			if (!new_buffer)
115 				return -ENOMEM;
116 		}
117 
118 		spin_lock(&mi->mi_log.rl_lock);
119 		old_buffer = mi->mi_log.rl_ring_buf;
120 		mi->mi_log.rl_ring_buf = new_buffer;
121 		mi->mi_log.rl_size = new_buffer_size;
122 		log_state = (struct read_log_state){
123 			.generation_id = mi->mi_log.rl_head.generation_id + 1,
124 		};
125 		mi->mi_log.rl_head = log_state;
126 		mi->mi_log.rl_tail = log_state;
127 		spin_unlock(&mi->mi_log.rl_lock);
128 
129 		kfree(old_buffer);
130 	}
131 
132 	if (options->sysfs_name && !mi->mi_sysfs_node)
133 		mi->mi_sysfs_node = incfs_add_sysfs_node(options->sysfs_name,
134 							 mi);
135 	else if (!options->sysfs_name && mi->mi_sysfs_node) {
136 		incfs_free_sysfs_node(mi->mi_sysfs_node);
137 		mi->mi_sysfs_node = NULL;
138 	} else if (options->sysfs_name &&
139 		strcmp(options->sysfs_name,
140 		       kobject_name(&mi->mi_sysfs_node->isn_sysfs_node))) {
141 		incfs_free_sysfs_node(mi->mi_sysfs_node);
142 		mi->mi_sysfs_node = incfs_add_sysfs_node(options->sysfs_name,
143 							 mi);
144 	}
145 
146 	if (IS_ERR(mi->mi_sysfs_node)) {
147 		int err = PTR_ERR(mi->mi_sysfs_node);
148 
149 		mi->mi_sysfs_node = NULL;
150 		return err;
151 	}
152 
153 	mi->mi_options = *options;
154 	return 0;
155 }
156 
incfs_free_mount_info(struct mount_info * mi)157 void incfs_free_mount_info(struct mount_info *mi)
158 {
159 	int i;
160 	if (!mi)
161 		return;
162 
163 	flush_delayed_work(&mi->mi_log.ml_wakeup_work);
164 	flush_delayed_work(&mi->mi_zstd_cleanup_work);
165 
166 	dput(mi->mi_index_dir);
167 	dput(mi->mi_incomplete_dir);
168 	path_put(&mi->mi_backing_dir_path);
169 	mutex_destroy(&mi->mi_dir_struct_mutex);
170 	mutex_destroy(&mi->mi_zstd_workspace_mutex);
171 	put_cred(mi->mi_owner);
172 	kfree(mi->mi_log.rl_ring_buf);
173 	for (i = 0; i < ARRAY_SIZE(mi->pseudo_file_xattr); ++i)
174 		kfree(mi->pseudo_file_xattr[i].data);
175 	kfree(mi->mi_per_uid_read_timeouts);
176 	incfs_free_sysfs_node(mi->mi_sysfs_node);
177 	kfree(mi);
178 }
179 
data_file_segment_init(struct data_file_segment * segment)180 static void data_file_segment_init(struct data_file_segment *segment)
181 {
182 	init_waitqueue_head(&segment->new_data_arrival_wq);
183 	init_rwsem(&segment->rwsem);
184 	INIT_LIST_HEAD(&segment->reads_list_head);
185 }
186 
file_id_to_str(incfs_uuid_t id)187 char *file_id_to_str(incfs_uuid_t id)
188 {
189 	char *result = kmalloc(1 + sizeof(id.bytes) * 2, GFP_NOFS);
190 	char *end;
191 
192 	if (!result)
193 		return NULL;
194 
195 	end = bin2hex(result, id.bytes, sizeof(id.bytes));
196 	*end = 0;
197 	return result;
198 }
199 
incfs_lookup_dentry(struct dentry * parent,const char * name)200 struct dentry *incfs_lookup_dentry(struct dentry *parent, const char *name)
201 {
202 	struct inode *inode;
203 	struct dentry *result = NULL;
204 
205 	if (!parent)
206 		return ERR_PTR(-EFAULT);
207 
208 	inode = d_inode(parent);
209 	inode_lock_nested(inode, I_MUTEX_PARENT);
210 	result = lookup_one_len(name, parent, strlen(name));
211 	inode_unlock(inode);
212 
213 	if (IS_ERR(result))
214 		pr_warn("%s err:%ld\n", __func__, PTR_ERR(result));
215 
216 	return result;
217 }
218 
handle_mapped_file(struct mount_info * mi,struct data_file * df)219 static struct data_file *handle_mapped_file(struct mount_info *mi,
220 					    struct data_file *df)
221 {
222 	char *file_id_str;
223 	struct dentry *index_file_dentry;
224 	struct path path;
225 	struct file *bf;
226 	struct data_file *result = NULL;
227 	const struct cred *old_cred;
228 
229 	file_id_str = file_id_to_str(df->df_id);
230 	if (!file_id_str)
231 		return ERR_PTR(-ENOENT);
232 
233 	index_file_dentry = incfs_lookup_dentry(mi->mi_index_dir,
234 						file_id_str);
235 	kfree(file_id_str);
236 	if (!index_file_dentry)
237 		return ERR_PTR(-ENOENT);
238 	if (IS_ERR(index_file_dentry))
239 		return (struct data_file *)index_file_dentry;
240 	if (!d_really_is_positive(index_file_dentry)) {
241 		result = ERR_PTR(-ENOENT);
242 		goto out;
243 	}
244 
245 	path = (struct path) {
246 		.mnt = mi->mi_backing_dir_path.mnt,
247 		.dentry = index_file_dentry
248 	};
249 
250 	old_cred = override_creds(mi->mi_owner);
251 	bf = dentry_open(&path, O_RDWR | O_NOATIME | O_LARGEFILE,
252 			 current_cred());
253 	revert_creds(old_cred);
254 
255 	if (IS_ERR(bf)) {
256 		result = (struct data_file *)bf;
257 		goto out;
258 	}
259 
260 	result = incfs_open_data_file(mi, bf);
261 	fput(bf);
262 	if (IS_ERR(result))
263 		goto out;
264 
265 	result->df_mapped_offset = df->df_metadata_off;
266 
267 out:
268 	dput(index_file_dentry);
269 	return result;
270 }
271 
incfs_open_data_file(struct mount_info * mi,struct file * bf)272 struct data_file *incfs_open_data_file(struct mount_info *mi, struct file *bf)
273 {
274 	struct data_file *df = NULL;
275 	struct backing_file_context *bfc = NULL;
276 	int md_records;
277 	u64 size;
278 	int error = 0;
279 	int i;
280 
281 	if (!bf || !mi)
282 		return ERR_PTR(-EFAULT);
283 
284 	if (!S_ISREG(bf->f_inode->i_mode))
285 		return ERR_PTR(-EBADF);
286 
287 	bfc = incfs_alloc_bfc(mi, bf);
288 	if (IS_ERR(bfc))
289 		return ERR_CAST(bfc);
290 
291 	df = kzalloc(sizeof(*df), GFP_NOFS);
292 	if (!df) {
293 		error = -ENOMEM;
294 		goto out;
295 	}
296 
297 	mutex_init(&df->df_enable_verity);
298 
299 	df->df_backing_file_context = bfc;
300 	df->df_mount_info = mi;
301 	for (i = 0; i < ARRAY_SIZE(df->df_segments); i++)
302 		data_file_segment_init(&df->df_segments[i]);
303 
304 	error = incfs_read_file_header(bfc, &df->df_metadata_off, &df->df_id,
305 				       &size, &df->df_header_flags);
306 
307 	if (error)
308 		goto out;
309 
310 	df->df_size = size;
311 	if (size > 0)
312 		df->df_data_block_count = get_blocks_count_for_size(size);
313 
314 	if (df->df_header_flags & INCFS_FILE_MAPPED) {
315 		struct data_file *mapped_df = handle_mapped_file(mi, df);
316 
317 		incfs_free_data_file(df);
318 		return mapped_df;
319 	}
320 
321 	md_records = incfs_scan_metadata_chain(df);
322 	if (md_records < 0)
323 		error = md_records;
324 
325 out:
326 	if (error) {
327 		incfs_free_bfc(bfc);
328 		if (df)
329 			df->df_backing_file_context = NULL;
330 		incfs_free_data_file(df);
331 		return ERR_PTR(error);
332 	}
333 	return df;
334 }
335 
incfs_free_data_file(struct data_file * df)336 void incfs_free_data_file(struct data_file *df)
337 {
338 	u32 data_blocks_written, hash_blocks_written;
339 
340 	if (!df)
341 		return;
342 
343 	data_blocks_written = atomic_read(&df->df_data_blocks_written);
344 	hash_blocks_written = atomic_read(&df->df_hash_blocks_written);
345 
346 	if (data_blocks_written != df->df_initial_data_blocks_written ||
347 	    hash_blocks_written != df->df_initial_hash_blocks_written) {
348 		struct backing_file_context *bfc = df->df_backing_file_context;
349 		int error = -1;
350 
351 		if (bfc && !mutex_lock_interruptible(&bfc->bc_mutex)) {
352 			error = incfs_write_status_to_backing_file(
353 						df->df_backing_file_context,
354 						df->df_status_offset,
355 						data_blocks_written,
356 						hash_blocks_written);
357 			mutex_unlock(&bfc->bc_mutex);
358 		}
359 
360 		if (error)
361 			/* Nothing can be done, just warn */
362 			pr_warn("incfs: failed to write status to backing file\n");
363 	}
364 
365 	incfs_free_mtree(df->df_hash_tree);
366 	incfs_free_bfc(df->df_backing_file_context);
367 	kfree(df->df_signature);
368 	kfree(df->df_verity_file_digest.data);
369 	kfree(df->df_verity_signature);
370 	mutex_destroy(&df->df_enable_verity);
371 	kfree(df);
372 }
373 
make_inode_ready_for_data_ops(struct mount_info * mi,struct inode * inode,struct file * backing_file)374 int make_inode_ready_for_data_ops(struct mount_info *mi,
375 				struct inode *inode,
376 				struct file *backing_file)
377 {
378 	struct inode_info *node = get_incfs_node(inode);
379 	struct data_file *df = NULL;
380 	int err = 0;
381 
382 	inode_lock(inode);
383 	if (S_ISREG(inode->i_mode)) {
384 		if (!node->n_file) {
385 			df = incfs_open_data_file(mi, backing_file);
386 
387 			if (IS_ERR(df))
388 				err = PTR_ERR(df);
389 			else
390 				node->n_file = df;
391 		}
392 	} else
393 		err = -EBADF;
394 	inode_unlock(inode);
395 	return err;
396 }
397 
incfs_open_dir_file(struct mount_info * mi,struct file * bf)398 struct dir_file *incfs_open_dir_file(struct mount_info *mi, struct file *bf)
399 {
400 	struct dir_file *dir = NULL;
401 
402 	if (!S_ISDIR(bf->f_inode->i_mode))
403 		return ERR_PTR(-EBADF);
404 
405 	dir = kzalloc(sizeof(*dir), GFP_NOFS);
406 	if (!dir)
407 		return ERR_PTR(-ENOMEM);
408 
409 	dir->backing_dir = get_file(bf);
410 	dir->mount_info = mi;
411 	return dir;
412 }
413 
incfs_free_dir_file(struct dir_file * dir)414 void incfs_free_dir_file(struct dir_file *dir)
415 {
416 	if (!dir)
417 		return;
418 	if (dir->backing_dir)
419 		fput(dir->backing_dir);
420 	kfree(dir);
421 }
422 
zstd_decompress_safe(struct mount_info * mi,struct mem_range src,struct mem_range dst)423 static ssize_t zstd_decompress_safe(struct mount_info *mi,
424 				    struct mem_range src, struct mem_range dst)
425 {
426 	ssize_t result;
427 	ZSTD_inBuffer inbuf = {.src = src.data,	.size = src.len};
428 	ZSTD_outBuffer outbuf = {.dst = dst.data, .size = dst.len};
429 
430 	result = mutex_lock_interruptible(&mi->mi_zstd_workspace_mutex);
431 	if (result)
432 		return result;
433 
434 	if (!mi->mi_zstd_stream) {
435 		unsigned int workspace_size = ZSTD_DStreamWorkspaceBound(
436 						INCFS_DATA_FILE_BLOCK_SIZE);
437 		void *workspace = kvmalloc(workspace_size, GFP_NOFS);
438 		ZSTD_DStream *stream;
439 
440 		if (!workspace) {
441 			result = -ENOMEM;
442 			goto out;
443 		}
444 
445 		stream = ZSTD_initDStream(INCFS_DATA_FILE_BLOCK_SIZE, workspace,
446 				  workspace_size);
447 		if (!stream) {
448 			kvfree(workspace);
449 			result = -EIO;
450 			goto out;
451 		}
452 
453 		mi->mi_zstd_workspace = workspace;
454 		mi->mi_zstd_stream = stream;
455 	}
456 
457 	result = ZSTD_decompressStream(mi->mi_zstd_stream, &outbuf, &inbuf) ?
458 		-EBADMSG : outbuf.pos;
459 
460 	mod_delayed_work(system_wq, &mi->mi_zstd_cleanup_work,
461 			 msecs_to_jiffies(5000));
462 
463 out:
464 	mutex_unlock(&mi->mi_zstd_workspace_mutex);
465 	return result;
466 }
467 
decompress(struct mount_info * mi,struct mem_range src,struct mem_range dst,int alg)468 static ssize_t decompress(struct mount_info *mi,
469 			  struct mem_range src, struct mem_range dst, int alg)
470 {
471 	int result;
472 
473 	switch (alg) {
474 	case INCFS_BLOCK_COMPRESSED_LZ4:
475 		result = LZ4_decompress_safe(src.data, dst.data, src.len,
476 					     dst.len);
477 		if (result < 0)
478 			return -EBADMSG;
479 		return result;
480 
481 	case INCFS_BLOCK_COMPRESSED_ZSTD:
482 		return zstd_decompress_safe(mi, src, dst);
483 
484 	default:
485 		WARN_ON(true);
486 		return -EOPNOTSUPP;
487 	}
488 }
489 
log_read_one_record(struct read_log * rl,struct read_log_state * rs)490 static void log_read_one_record(struct read_log *rl, struct read_log_state *rs)
491 {
492 	union log_record *record =
493 		(union log_record *)((u8 *)rl->rl_ring_buf + rs->next_offset);
494 	size_t record_size;
495 
496 	switch (record->full_record.type) {
497 	case FULL:
498 		rs->base_record = record->full_record;
499 		record_size = sizeof(record->full_record);
500 		break;
501 
502 	case SAME_FILE:
503 		rs->base_record.block_index =
504 			record->same_file.block_index;
505 		rs->base_record.absolute_ts_us +=
506 			record->same_file.relative_ts_us;
507 		rs->base_record.uid = record->same_file.uid;
508 		record_size = sizeof(record->same_file);
509 		break;
510 
511 	case SAME_FILE_CLOSE_BLOCK:
512 		rs->base_record.block_index +=
513 			record->same_file_close_block.block_index_delta;
514 		rs->base_record.absolute_ts_us +=
515 			record->same_file_close_block.relative_ts_us;
516 		record_size = sizeof(record->same_file_close_block);
517 		break;
518 
519 	case SAME_FILE_CLOSE_BLOCK_SHORT:
520 		rs->base_record.block_index +=
521 			record->same_file_close_block_short.block_index_delta;
522 		rs->base_record.absolute_ts_us +=
523 		   record->same_file_close_block_short.relative_ts_tens_us * 10;
524 		record_size = sizeof(record->same_file_close_block_short);
525 		break;
526 
527 	case SAME_FILE_NEXT_BLOCK:
528 		++rs->base_record.block_index;
529 		rs->base_record.absolute_ts_us +=
530 			record->same_file_next_block.relative_ts_us;
531 		record_size = sizeof(record->same_file_next_block);
532 		break;
533 
534 	case SAME_FILE_NEXT_BLOCK_SHORT:
535 		++rs->base_record.block_index;
536 		rs->base_record.absolute_ts_us +=
537 		    record->same_file_next_block_short.relative_ts_tens_us * 10;
538 		record_size = sizeof(record->same_file_next_block_short);
539 		break;
540 	}
541 
542 	rs->next_offset += record_size;
543 	if (rs->next_offset > rl->rl_size - sizeof(*record)) {
544 		rs->next_offset = 0;
545 		++rs->current_pass_no;
546 	}
547 	++rs->current_record_no;
548 }
549 
log_block_read(struct mount_info * mi,incfs_uuid_t * id,int block_index)550 static void log_block_read(struct mount_info *mi, incfs_uuid_t *id,
551 			   int block_index)
552 {
553 	struct read_log *log = &mi->mi_log;
554 	struct read_log_state *head, *tail;
555 	s64 now_us;
556 	s64 relative_us;
557 	union log_record record;
558 	size_t record_size;
559 	uid_t uid = current_uid().val;
560 	int block_delta;
561 	bool same_file, same_uid;
562 	bool next_block, close_block, very_close_block;
563 	bool close_time, very_close_time, very_very_close_time;
564 
565 	/*
566 	 * This may read the old value, but it's OK to delay the logging start
567 	 * right after the configuration update.
568 	 */
569 	if (READ_ONCE(log->rl_size) == 0)
570 		return;
571 
572 	now_us = ktime_to_us(ktime_get());
573 
574 	spin_lock(&log->rl_lock);
575 	if (log->rl_size == 0) {
576 		spin_unlock(&log->rl_lock);
577 		return;
578 	}
579 
580 	head = &log->rl_head;
581 	tail = &log->rl_tail;
582 	relative_us = now_us - head->base_record.absolute_ts_us;
583 
584 	same_file = !memcmp(id, &head->base_record.file_id,
585 			    sizeof(incfs_uuid_t));
586 	same_uid = uid == head->base_record.uid;
587 
588 	block_delta = block_index - head->base_record.block_index;
589 	next_block = block_delta == 1;
590 	very_close_block = block_delta >= S8_MIN && block_delta <= S8_MAX;
591 	close_block = block_delta >= S16_MIN && block_delta <= S16_MAX;
592 
593 	very_very_close_time = relative_us < (1 << 5) * 10;
594 	very_close_time = relative_us < (1 << 13);
595 	close_time = relative_us < (1 << 16);
596 
597 	if (same_file && same_uid && next_block && very_very_close_time) {
598 		record.same_file_next_block_short =
599 			(struct same_file_next_block_short){
600 				.type = SAME_FILE_NEXT_BLOCK_SHORT,
601 				.relative_ts_tens_us = div_s64(relative_us, 10),
602 			};
603 		record_size = sizeof(struct same_file_next_block_short);
604 	} else if (same_file && same_uid && next_block && very_close_time) {
605 		record.same_file_next_block = (struct same_file_next_block){
606 			.type = SAME_FILE_NEXT_BLOCK,
607 			.relative_ts_us = relative_us,
608 		};
609 		record_size = sizeof(struct same_file_next_block);
610 	} else if (same_file && same_uid && very_close_block &&
611 		   very_very_close_time) {
612 		record.same_file_close_block_short =
613 			(struct same_file_close_block_short){
614 				.type = SAME_FILE_CLOSE_BLOCK_SHORT,
615 				.relative_ts_tens_us = div_s64(relative_us, 10),
616 				.block_index_delta = block_delta,
617 			};
618 		record_size = sizeof(struct same_file_close_block_short);
619 	} else if (same_file && same_uid && close_block && very_close_time) {
620 		record.same_file_close_block = (struct same_file_close_block){
621 				.type = SAME_FILE_CLOSE_BLOCK,
622 				.relative_ts_us = relative_us,
623 				.block_index_delta = block_delta,
624 			};
625 		record_size = sizeof(struct same_file_close_block);
626 	} else if (same_file && close_time) {
627 		record.same_file = (struct same_file){
628 			.type = SAME_FILE,
629 			.block_index = block_index,
630 			.relative_ts_us = relative_us,
631 			.uid = uid,
632 		};
633 		record_size = sizeof(struct same_file);
634 	} else {
635 		record.full_record = (struct full_record){
636 			.type = FULL,
637 			.block_index = block_index,
638 			.file_id = *id,
639 			.absolute_ts_us = now_us,
640 			.uid = uid,
641 		};
642 		head->base_record.file_id = *id;
643 		record_size = sizeof(struct full_record);
644 	}
645 
646 	head->base_record.block_index = block_index;
647 	head->base_record.absolute_ts_us = now_us;
648 
649 	/* Advance tail beyond area we are going to overwrite */
650 	while (tail->current_pass_no < head->current_pass_no &&
651 	       tail->next_offset < head->next_offset + record_size)
652 		log_read_one_record(log, tail);
653 
654 	memcpy(((u8 *)log->rl_ring_buf) + head->next_offset, &record,
655 	       record_size);
656 	head->next_offset += record_size;
657 	if (head->next_offset > log->rl_size - sizeof(record)) {
658 		head->next_offset = 0;
659 		++head->current_pass_no;
660 	}
661 	++head->current_record_no;
662 
663 	spin_unlock(&log->rl_lock);
664 	schedule_delayed_work(&log->ml_wakeup_work, msecs_to_jiffies(16));
665 }
666 
validate_hash_tree(struct backing_file_context * bfc,struct file * f,int block_index,struct mem_range data,u8 * buf)667 static int validate_hash_tree(struct backing_file_context *bfc, struct file *f,
668 			      int block_index, struct mem_range data, u8 *buf)
669 {
670 	struct data_file *df = get_incfs_data_file(f);
671 	u8 stored_digest[INCFS_MAX_HASH_SIZE] = {};
672 	u8 calculated_digest[INCFS_MAX_HASH_SIZE] = {};
673 	struct mtree *tree = NULL;
674 	struct incfs_df_signature *sig = NULL;
675 	int digest_size;
676 	int hash_block_index = block_index;
677 	int lvl;
678 	int res;
679 	loff_t hash_block_offset[INCFS_MAX_MTREE_LEVELS];
680 	size_t hash_offset_in_block[INCFS_MAX_MTREE_LEVELS];
681 	int hash_per_block;
682 	pgoff_t file_pages;
683 
684 	/*
685 	 * Memory barrier to make sure tree is fully present if added via enable
686 	 * verity
687 	 */
688 	tree = smp_load_acquire(&df->df_hash_tree);
689 	sig = df->df_signature;
690 	if (!tree || !sig)
691 		return 0;
692 
693 	digest_size = tree->alg->digest_size;
694 	hash_per_block = INCFS_DATA_FILE_BLOCK_SIZE / digest_size;
695 	for (lvl = 0; lvl < tree->depth; lvl++) {
696 		loff_t lvl_off = tree->hash_level_suboffset[lvl];
697 
698 		hash_block_offset[lvl] =
699 			lvl_off + round_down(hash_block_index * digest_size,
700 					     INCFS_DATA_FILE_BLOCK_SIZE);
701 		hash_offset_in_block[lvl] = hash_block_index * digest_size %
702 					    INCFS_DATA_FILE_BLOCK_SIZE;
703 		hash_block_index /= hash_per_block;
704 	}
705 
706 	memcpy(stored_digest, tree->root_hash, digest_size);
707 
708 	file_pages = DIV_ROUND_UP(df->df_size, INCFS_DATA_FILE_BLOCK_SIZE);
709 	for (lvl = tree->depth - 1; lvl >= 0; lvl--) {
710 		pgoff_t hash_page =
711 			file_pages +
712 			hash_block_offset[lvl] / INCFS_DATA_FILE_BLOCK_SIZE;
713 		struct page *page = find_get_page_flags(
714 			f->f_inode->i_mapping, hash_page, FGP_ACCESSED);
715 
716 		if (page && PageChecked(page)) {
717 			u8 *addr = kmap_atomic(page);
718 
719 			memcpy(stored_digest, addr + hash_offset_in_block[lvl],
720 			       digest_size);
721 
722 			kunmap_atomic(addr);
723 			put_page(page);
724 			continue;
725 		}
726 
727 		if (page)
728 			put_page(page);
729 
730 		res = incfs_kread(bfc, buf, INCFS_DATA_FILE_BLOCK_SIZE,
731 				  hash_block_offset[lvl] + sig->hash_offset);
732 		if (res < 0)
733 			return res;
734 		if (res != INCFS_DATA_FILE_BLOCK_SIZE)
735 			return -EIO;
736 		res = incfs_calc_digest(tree->alg,
737 					range(buf, INCFS_DATA_FILE_BLOCK_SIZE),
738 					range(calculated_digest, digest_size));
739 		if (res)
740 			return res;
741 
742 		if (memcmp(stored_digest, calculated_digest, digest_size)) {
743 			int i;
744 			bool zero = true;
745 
746 			pr_warn("incfs: Hash mismatch lvl:%d blk:%d\n",
747 				lvl, block_index);
748 			for (i = 0; i < digest_size; i++)
749 				if (stored_digest[i]) {
750 					zero = false;
751 					break;
752 				}
753 
754 			if (zero)
755 				pr_debug("Note saved_digest all zero - did you forget to load the hashes?\n");
756 			return -EBADMSG;
757 		}
758 
759 		memcpy(stored_digest, buf + hash_offset_in_block[lvl],
760 		       digest_size);
761 
762 		page = grab_cache_page(f->f_inode->i_mapping, hash_page);
763 		if (page) {
764 			u8 *addr = kmap_atomic(page);
765 
766 			memcpy(addr, buf, INCFS_DATA_FILE_BLOCK_SIZE);
767 			kunmap_atomic(addr);
768 			SetPageChecked(page);
769 			SetPageUptodate(page);
770 			unlock_page(page);
771 			put_page(page);
772 		}
773 	}
774 
775 	res = incfs_calc_digest(tree->alg, data,
776 				range(calculated_digest, digest_size));
777 	if (res)
778 		return res;
779 
780 	if (memcmp(stored_digest, calculated_digest, digest_size)) {
781 		pr_debug("Leaf hash mismatch blk:%d\n", block_index);
782 		return -EBADMSG;
783 	}
784 
785 	return 0;
786 }
787 
get_file_segment(struct data_file * df,int block_index)788 static struct data_file_segment *get_file_segment(struct data_file *df,
789 						  int block_index)
790 {
791 	int seg_idx = block_index % ARRAY_SIZE(df->df_segments);
792 
793 	return &df->df_segments[seg_idx];
794 }
795 
is_data_block_present(struct data_file_block * block)796 static bool is_data_block_present(struct data_file_block *block)
797 {
798 	return (block->db_backing_file_data_offset != 0) &&
799 	       (block->db_stored_size != 0);
800 }
801 
convert_data_file_block(struct incfs_blockmap_entry * bme,struct data_file_block * res_block)802 static void convert_data_file_block(struct incfs_blockmap_entry *bme,
803 				    struct data_file_block *res_block)
804 {
805 	u16 flags = le16_to_cpu(bme->me_flags);
806 
807 	res_block->db_backing_file_data_offset =
808 		le16_to_cpu(bme->me_data_offset_hi);
809 	res_block->db_backing_file_data_offset <<= 32;
810 	res_block->db_backing_file_data_offset |=
811 		le32_to_cpu(bme->me_data_offset_lo);
812 	res_block->db_stored_size = le16_to_cpu(bme->me_data_size);
813 	res_block->db_comp_alg = flags & INCFS_BLOCK_COMPRESSED_MASK;
814 }
815 
get_data_file_block(struct data_file * df,int index,struct data_file_block * res_block)816 static int get_data_file_block(struct data_file *df, int index,
817 			       struct data_file_block *res_block)
818 {
819 	struct incfs_blockmap_entry bme = {};
820 	struct backing_file_context *bfc = NULL;
821 	loff_t blockmap_off = 0;
822 	int error = 0;
823 
824 	if (!df || !res_block)
825 		return -EFAULT;
826 
827 	blockmap_off = df->df_blockmap_off;
828 	bfc = df->df_backing_file_context;
829 
830 	if (index < 0 || blockmap_off == 0)
831 		return -EINVAL;
832 
833 	error = incfs_read_blockmap_entry(bfc, index, blockmap_off, &bme);
834 	if (error)
835 		return error;
836 
837 	convert_data_file_block(&bme, res_block);
838 	return 0;
839 }
840 
check_room_for_one_range(u32 size,u32 size_out)841 static int check_room_for_one_range(u32 size, u32 size_out)
842 {
843 	if (size_out + sizeof(struct incfs_filled_range) > size)
844 		return -ERANGE;
845 	return 0;
846 }
847 
copy_one_range(struct incfs_filled_range * range,void __user * buffer,u32 size,u32 * size_out)848 static int copy_one_range(struct incfs_filled_range *range, void __user *buffer,
849 			  u32 size, u32 *size_out)
850 {
851 	int error = check_room_for_one_range(size, *size_out);
852 	if (error)
853 		return error;
854 
855 	if (copy_to_user(((char __user *)buffer) + *size_out, range,
856 				sizeof(*range)))
857 		return -EFAULT;
858 
859 	*size_out += sizeof(*range);
860 	return 0;
861 }
862 
863 #define READ_BLOCKMAP_ENTRIES 512
incfs_get_filled_blocks(struct data_file * df,struct incfs_file_data * fd,struct incfs_get_filled_blocks_args * arg)864 int incfs_get_filled_blocks(struct data_file *df,
865 			    struct incfs_file_data *fd,
866 			    struct incfs_get_filled_blocks_args *arg)
867 {
868 	int error = 0;
869 	bool in_range = false;
870 	struct incfs_filled_range range;
871 	void __user *buffer = u64_to_user_ptr(arg->range_buffer);
872 	u32 size = arg->range_buffer_size;
873 	u32 end_index =
874 		arg->end_index ? arg->end_index : df->df_total_block_count;
875 	u32 *size_out = &arg->range_buffer_size_out;
876 	int i = READ_BLOCKMAP_ENTRIES - 1;
877 	int entries_read = 0;
878 	struct incfs_blockmap_entry *bme;
879 	int data_blocks_filled = 0;
880 	int hash_blocks_filled = 0;
881 
882 	*size_out = 0;
883 	if (end_index > df->df_total_block_count)
884 		end_index = df->df_total_block_count;
885 	arg->total_blocks_out = df->df_total_block_count;
886 	arg->data_blocks_out = df->df_data_block_count;
887 
888 	if (atomic_read(&df->df_data_blocks_written) ==
889 	    df->df_data_block_count) {
890 		pr_debug("File marked full, fast get_filled_blocks");
891 		if (arg->start_index > end_index) {
892 			arg->index_out = arg->start_index;
893 			return 0;
894 		}
895 		arg->index_out = arg->start_index;
896 
897 		error = check_room_for_one_range(size, *size_out);
898 		if (error)
899 			return error;
900 
901 		range = (struct incfs_filled_range){
902 			.begin = arg->start_index,
903 			.end = end_index,
904 		};
905 
906 		error = copy_one_range(&range, buffer, size, size_out);
907 		if (error)
908 			return error;
909 		arg->index_out = end_index;
910 		return 0;
911 	}
912 
913 	bme = kzalloc(sizeof(*bme) * READ_BLOCKMAP_ENTRIES,
914 		      GFP_NOFS | __GFP_COMP);
915 	if (!bme)
916 		return -ENOMEM;
917 
918 	for (arg->index_out = arg->start_index; arg->index_out < end_index;
919 	     ++arg->index_out) {
920 		struct data_file_block dfb;
921 
922 		if (++i == READ_BLOCKMAP_ENTRIES) {
923 			entries_read = incfs_read_blockmap_entries(
924 				df->df_backing_file_context, bme,
925 				arg->index_out, READ_BLOCKMAP_ENTRIES,
926 				df->df_blockmap_off);
927 			if (entries_read < 0) {
928 				error = entries_read;
929 				break;
930 			}
931 
932 			i = 0;
933 		}
934 
935 		if (i >= entries_read) {
936 			error = -EIO;
937 			break;
938 		}
939 
940 		convert_data_file_block(bme + i, &dfb);
941 
942 		if (is_data_block_present(&dfb)) {
943 			if (arg->index_out >= df->df_data_block_count)
944 				++hash_blocks_filled;
945 			else
946 				++data_blocks_filled;
947 		}
948 
949 		if (is_data_block_present(&dfb) == in_range)
950 			continue;
951 
952 		if (!in_range) {
953 			error = check_room_for_one_range(size, *size_out);
954 			if (error)
955 				break;
956 			in_range = true;
957 			range.begin = arg->index_out;
958 		} else {
959 			range.end = arg->index_out;
960 			error = copy_one_range(&range, buffer, size, size_out);
961 			if (error) {
962 				/* there will be another try out of the loop,
963 				 * it will reset the index_out if it fails too
964 				 */
965 				break;
966 			}
967 			in_range = false;
968 		}
969 	}
970 
971 	if (in_range) {
972 		range.end = arg->index_out;
973 		error = copy_one_range(&range, buffer, size, size_out);
974 		if (error)
975 			arg->index_out = range.begin;
976 	}
977 
978 	if (arg->start_index == 0) {
979 		fd->fd_get_block_pos = 0;
980 		fd->fd_filled_data_blocks = 0;
981 		fd->fd_filled_hash_blocks = 0;
982 	}
983 
984 	if (arg->start_index == fd->fd_get_block_pos) {
985 		fd->fd_get_block_pos = arg->index_out + 1;
986 		fd->fd_filled_data_blocks += data_blocks_filled;
987 		fd->fd_filled_hash_blocks += hash_blocks_filled;
988 	}
989 
990 	if (fd->fd_get_block_pos == df->df_total_block_count + 1) {
991 		if (fd->fd_filled_data_blocks >
992 		   atomic_read(&df->df_data_blocks_written))
993 			atomic_set(&df->df_data_blocks_written,
994 				   fd->fd_filled_data_blocks);
995 
996 		if (fd->fd_filled_hash_blocks >
997 		   atomic_read(&df->df_hash_blocks_written))
998 			atomic_set(&df->df_hash_blocks_written,
999 				   fd->fd_filled_hash_blocks);
1000 	}
1001 
1002 	kfree(bme);
1003 	return error;
1004 }
1005 
is_read_done(struct pending_read * read)1006 static bool is_read_done(struct pending_read *read)
1007 {
1008 	return atomic_read_acquire(&read->done) != 0;
1009 }
1010 
set_read_done(struct pending_read * read)1011 static void set_read_done(struct pending_read *read)
1012 {
1013 	atomic_set_release(&read->done, 1);
1014 }
1015 
1016 /*
1017  * Notifies a given data file about pending read from a given block.
1018  * Returns a new pending read entry.
1019  */
add_pending_read(struct data_file * df,int block_index)1020 static struct pending_read *add_pending_read(struct data_file *df,
1021 					     int block_index)
1022 {
1023 	struct pending_read *result = NULL;
1024 	struct data_file_segment *segment = NULL;
1025 	struct mount_info *mi = NULL;
1026 
1027 	segment = get_file_segment(df, block_index);
1028 	mi = df->df_mount_info;
1029 
1030 	result = kzalloc(sizeof(*result), GFP_NOFS);
1031 	if (!result)
1032 		return NULL;
1033 
1034 	result->file_id = df->df_id;
1035 	result->block_index = block_index;
1036 	result->timestamp_us = ktime_to_us(ktime_get());
1037 	result->uid = current_uid().val;
1038 
1039 	spin_lock(&mi->pending_read_lock);
1040 
1041 	result->serial_number = ++mi->mi_last_pending_read_number;
1042 	mi->mi_pending_reads_count++;
1043 
1044 	list_add_rcu(&result->mi_reads_list, &mi->mi_reads_list_head);
1045 	list_add_rcu(&result->segment_reads_list, &segment->reads_list_head);
1046 
1047 	spin_unlock(&mi->pending_read_lock);
1048 
1049 	wake_up_all(&mi->mi_pending_reads_notif_wq);
1050 	return result;
1051 }
1052 
free_pending_read_entry(struct rcu_head * entry)1053 static void free_pending_read_entry(struct rcu_head *entry)
1054 {
1055 	struct pending_read *read;
1056 
1057 	read = container_of(entry, struct pending_read, rcu);
1058 
1059 	kfree(read);
1060 }
1061 
1062 /* Notifies a given data file that pending read is completed. */
remove_pending_read(struct data_file * df,struct pending_read * read)1063 static void remove_pending_read(struct data_file *df, struct pending_read *read)
1064 {
1065 	struct mount_info *mi = NULL;
1066 
1067 	if (!df || !read) {
1068 		WARN_ON(!df);
1069 		WARN_ON(!read);
1070 		return;
1071 	}
1072 
1073 	mi = df->df_mount_info;
1074 
1075 	spin_lock(&mi->pending_read_lock);
1076 
1077 	list_del_rcu(&read->mi_reads_list);
1078 	list_del_rcu(&read->segment_reads_list);
1079 
1080 	mi->mi_pending_reads_count--;
1081 
1082 	spin_unlock(&mi->pending_read_lock);
1083 
1084 	/* Don't free. Wait for readers */
1085 	call_rcu(&read->rcu, free_pending_read_entry);
1086 }
1087 
notify_pending_reads(struct mount_info * mi,struct data_file_segment * segment,int index)1088 static void notify_pending_reads(struct mount_info *mi,
1089 		struct data_file_segment *segment,
1090 		int index)
1091 {
1092 	struct pending_read *entry = NULL;
1093 
1094 	/* Notify pending reads waiting for this block. */
1095 	rcu_read_lock();
1096 	list_for_each_entry_rcu(entry, &segment->reads_list_head,
1097 						segment_reads_list) {
1098 		if (entry->block_index == index)
1099 			set_read_done(entry);
1100 	}
1101 	rcu_read_unlock();
1102 	wake_up_all(&segment->new_data_arrival_wq);
1103 
1104 	atomic_inc(&mi->mi_blocks_written);
1105 	wake_up_all(&mi->mi_blocks_written_notif_wq);
1106 }
1107 
wait_for_data_block(struct data_file * df,int block_index,struct data_file_block * res_block,struct incfs_read_data_file_timeouts * timeouts,unsigned int * delayed_min_us)1108 static int wait_for_data_block(struct data_file *df, int block_index,
1109 			       struct data_file_block *res_block,
1110 			       struct incfs_read_data_file_timeouts *timeouts,
1111 			       unsigned int *delayed_min_us)
1112 {
1113 	struct data_file_block block = {};
1114 	struct data_file_segment *segment = NULL;
1115 	struct pending_read *read = NULL;
1116 	struct mount_info *mi = NULL;
1117 	int error;
1118 	int wait_res = 0;
1119 	unsigned int delayed_pending_us = 0;
1120 	bool delayed_pending = false;
1121 
1122 	if (!df || !res_block)
1123 		return -EFAULT;
1124 
1125 	if (block_index < 0 || block_index >= df->df_data_block_count)
1126 		return -EINVAL;
1127 
1128 	if (df->df_blockmap_off <= 0 || !df->df_mount_info)
1129 		return -ENODATA;
1130 
1131 	mi = df->df_mount_info;
1132 	segment = get_file_segment(df, block_index);
1133 
1134 	error = down_read_killable(&segment->rwsem);
1135 	if (error)
1136 		return error;
1137 
1138 	/* Look up the given block */
1139 	error = get_data_file_block(df, block_index, &block);
1140 
1141 	up_read(&segment->rwsem);
1142 
1143 	if (error)
1144 		return error;
1145 
1146 	/* If the block was found, just return it. No need to wait. */
1147 	if (is_data_block_present(&block)) {
1148 		*res_block = block;
1149 		if (timeouts && timeouts->min_time_us) {
1150 			*delayed_min_us = timeouts->min_time_us;
1151 			goto out;
1152 		}
1153 		return 0;
1154 	} else {
1155 		/* If it's not found, create a pending read */
1156 		if (timeouts && timeouts->max_pending_time_us) {
1157 			read = add_pending_read(df, block_index);
1158 			if (!read)
1159 				return -ENOMEM;
1160 		} else {
1161 			log_block_read(mi, &df->df_id, block_index);
1162 			return -ETIME;
1163 		}
1164 	}
1165 
1166 	/* Rest of function only applies if timeouts != NULL */
1167 	if (!timeouts) {
1168 		pr_warn("incfs: timeouts unexpectedly NULL\n");
1169 		return -EFSCORRUPTED;
1170 	}
1171 
1172 	/* Wait for notifications about block's arrival */
1173 	wait_res =
1174 		wait_event_interruptible_timeout(segment->new_data_arrival_wq,
1175 			(is_read_done(read)),
1176 			usecs_to_jiffies(timeouts->max_pending_time_us));
1177 
1178 	/* Woke up, the pending read is no longer needed. */
1179 	remove_pending_read(df, read);
1180 
1181 	if (wait_res == 0) {
1182 		/* Wait has timed out */
1183 		log_block_read(mi, &df->df_id, block_index);
1184 		return -ETIME;
1185 	}
1186 	if (wait_res < 0) {
1187 		/*
1188 		 * Only ERESTARTSYS is really expected here when a signal
1189 		 * comes while we wait.
1190 		 */
1191 		return wait_res;
1192 	}
1193 
1194 	delayed_pending = true;
1195 	delayed_pending_us = timeouts->max_pending_time_us -
1196 				jiffies_to_usecs(wait_res);
1197 	if (timeouts->min_pending_time_us > delayed_pending_us)
1198 		*delayed_min_us = timeouts->min_pending_time_us -
1199 					     delayed_pending_us;
1200 
1201 	error = down_read_killable(&segment->rwsem);
1202 	if (error)
1203 		return error;
1204 
1205 	/*
1206 	 * Re-read blocks info now, it has just arrived and
1207 	 * should be available.
1208 	 */
1209 	error = get_data_file_block(df, block_index, &block);
1210 	if (!error) {
1211 		if (is_data_block_present(&block))
1212 			*res_block = block;
1213 		else {
1214 			/*
1215 			 * Somehow wait finished successfully but block still
1216 			 * can't be found. It's not normal.
1217 			 */
1218 			pr_warn("incfs: Wait succeeded but block not found.\n");
1219 			error = -ENODATA;
1220 		}
1221 	}
1222 	up_read(&segment->rwsem);
1223 
1224 out:
1225 	if (error)
1226 		return error;
1227 
1228 	if (delayed_pending) {
1229 		mi->mi_reads_delayed_pending++;
1230 		mi->mi_reads_delayed_pending_us +=
1231 			delayed_pending_us;
1232 	}
1233 
1234 	if (delayed_min_us && *delayed_min_us) {
1235 		mi->mi_reads_delayed_min++;
1236 		mi->mi_reads_delayed_min_us += *delayed_min_us;
1237 	}
1238 
1239 	return 0;
1240 }
1241 
incfs_update_sysfs_error(struct file * file,int index,int result,struct mount_info * mi,struct data_file * df)1242 static int incfs_update_sysfs_error(struct file *file, int index, int result,
1243 				struct mount_info *mi, struct data_file *df)
1244 {
1245 	int error;
1246 
1247 	if (result >= 0)
1248 		return 0;
1249 
1250 	error = mutex_lock_interruptible(&mi->mi_le_mutex);
1251 	if (error)
1252 		return error;
1253 
1254 	mi->mi_le_file_id = df->df_id;
1255 	mi->mi_le_time_us = ktime_to_us(ktime_get());
1256 	mi->mi_le_page = index;
1257 	mi->mi_le_errno = result;
1258 	mi->mi_le_uid = current_uid().val;
1259 	mutex_unlock(&mi->mi_le_mutex);
1260 
1261 	return 0;
1262 }
1263 
incfs_read_data_file_block(struct mem_range dst,struct file * f,int index,struct mem_range tmp,struct incfs_read_data_file_timeouts * timeouts,unsigned int * delayed_min_us)1264 ssize_t incfs_read_data_file_block(struct mem_range dst, struct file *f,
1265 			int index, struct mem_range tmp,
1266 			struct incfs_read_data_file_timeouts *timeouts,
1267 			unsigned int *delayed_min_us)
1268 {
1269 	loff_t pos;
1270 	ssize_t result;
1271 	size_t bytes_to_read;
1272 	struct mount_info *mi = NULL;
1273 	struct backing_file_context *bfc = NULL;
1274 	struct data_file_block block = {};
1275 	struct data_file *df = get_incfs_data_file(f);
1276 
1277 	if (!dst.data || !df || !tmp.data)
1278 		return -EFAULT;
1279 
1280 	if (tmp.len < 2 * INCFS_DATA_FILE_BLOCK_SIZE)
1281 		return -ERANGE;
1282 
1283 	mi = df->df_mount_info;
1284 	bfc = df->df_backing_file_context;
1285 
1286 	result = wait_for_data_block(df, index, &block, timeouts,
1287 				     delayed_min_us);
1288 	if (result < 0)
1289 		goto out;
1290 
1291 	pos = block.db_backing_file_data_offset;
1292 	if (block.db_comp_alg == COMPRESSION_NONE) {
1293 		bytes_to_read = min(dst.len, block.db_stored_size);
1294 		result = incfs_kread(bfc, dst.data, bytes_to_read, pos);
1295 
1296 		/* Some data was read, but not enough */
1297 		if (result >= 0 && result != bytes_to_read)
1298 			result = -EIO;
1299 	} else {
1300 		bytes_to_read = min(tmp.len, block.db_stored_size);
1301 		result = incfs_kread(bfc, tmp.data, bytes_to_read, pos);
1302 		if (result == bytes_to_read) {
1303 			result =
1304 				decompress(mi, range(tmp.data, bytes_to_read),
1305 					   dst, block.db_comp_alg);
1306 			if (result < 0) {
1307 				const char *name =
1308 				    bfc->bc_file->f_path.dentry->d_name.name;
1309 
1310 				pr_warn_once("incfs: Decompression error. %s",
1311 					     name);
1312 			}
1313 		} else if (result >= 0) {
1314 			/* Some data was read, but not enough */
1315 			result = -EIO;
1316 		}
1317 	}
1318 
1319 	if (result > 0) {
1320 		int err = validate_hash_tree(bfc, f, index, dst, tmp.data);
1321 
1322 		if (err < 0)
1323 			result = err;
1324 	}
1325 
1326 	if (result >= 0)
1327 		log_block_read(mi, &df->df_id, index);
1328 
1329 out:
1330 	if (result == -ETIME)
1331 		mi->mi_reads_failed_timed_out++;
1332 	else if (result == -EBADMSG)
1333 		mi->mi_reads_failed_hash_verification++;
1334 	else if (result < 0)
1335 		mi->mi_reads_failed_other++;
1336 
1337 	incfs_update_sysfs_error(f, index, result, mi, df);
1338 
1339 	return result;
1340 }
1341 
incfs_read_merkle_tree_blocks(struct mem_range dst,struct data_file * df,size_t offset)1342 ssize_t incfs_read_merkle_tree_blocks(struct mem_range dst,
1343 				      struct data_file *df, size_t offset)
1344 {
1345 	struct backing_file_context *bfc = NULL;
1346 	struct incfs_df_signature *sig = NULL;
1347 	size_t to_read = dst.len;
1348 
1349 	if (!dst.data || !df)
1350 		return -EFAULT;
1351 
1352 	sig = df->df_signature;
1353 	bfc = df->df_backing_file_context;
1354 
1355 	if (offset > sig->hash_size)
1356 		return -ERANGE;
1357 
1358 	if (offset + to_read > sig->hash_size)
1359 		to_read = sig->hash_size - offset;
1360 
1361 	return incfs_kread(bfc, dst.data, to_read, sig->hash_offset + offset);
1362 }
1363 
incfs_process_new_data_block(struct data_file * df,struct incfs_fill_block * block,u8 * data,bool * complete)1364 int incfs_process_new_data_block(struct data_file *df,
1365 				 struct incfs_fill_block *block, u8 *data,
1366 				 bool *complete)
1367 {
1368 	struct mount_info *mi = NULL;
1369 	struct backing_file_context *bfc = NULL;
1370 	struct data_file_segment *segment = NULL;
1371 	struct data_file_block existing_block = {};
1372 	u16 flags = 0;
1373 	int error = 0;
1374 
1375 	if (!df || !block)
1376 		return -EFAULT;
1377 
1378 	bfc = df->df_backing_file_context;
1379 	mi = df->df_mount_info;
1380 
1381 	if (block->block_index >= df->df_data_block_count)
1382 		return -ERANGE;
1383 
1384 	segment = get_file_segment(df, block->block_index);
1385 	if (!segment)
1386 		return -EFAULT;
1387 
1388 	if (block->compression == COMPRESSION_LZ4)
1389 		flags |= INCFS_BLOCK_COMPRESSED_LZ4;
1390 	else if (block->compression == COMPRESSION_ZSTD)
1391 		flags |= INCFS_BLOCK_COMPRESSED_ZSTD;
1392 	else if (block->compression)
1393 		return -EINVAL;
1394 
1395 	error = down_read_killable(&segment->rwsem);
1396 	if (error)
1397 		return error;
1398 
1399 	error = get_data_file_block(df, block->block_index, &existing_block);
1400 
1401 	up_read(&segment->rwsem);
1402 
1403 	if (error)
1404 		return error;
1405 	if (is_data_block_present(&existing_block))
1406 		/* Block is already present, nothing to do here */
1407 		return 0;
1408 
1409 	error = down_write_killable(&segment->rwsem);
1410 	if (error)
1411 		return error;
1412 
1413 	/* Recheck inside write lock */
1414 	error = get_data_file_block(df, block->block_index, &existing_block);
1415 	if (error)
1416 		goto out_up_write;
1417 
1418 	if (is_data_block_present(&existing_block))
1419 		goto out_up_write;
1420 
1421 	error = mutex_lock_interruptible(&bfc->bc_mutex);
1422 	if (error)
1423 		goto out_up_write;
1424 
1425 	error = incfs_write_data_block_to_backing_file(bfc,
1426 			range(data, block->data_len), block->block_index,
1427 			df->df_blockmap_off, flags);
1428 	if (error)
1429 		goto out_mutex_unlock;
1430 
1431 	if (atomic_inc_return(&df->df_data_blocks_written)
1432 			>= df->df_data_block_count)
1433 		*complete = true;
1434 
1435 out_mutex_unlock:
1436 	mutex_unlock(&bfc->bc_mutex);
1437 	if (!error)
1438 		notify_pending_reads(mi, segment, block->block_index);
1439 
1440 out_up_write:
1441 	up_write(&segment->rwsem);
1442 
1443 	if (error)
1444 		pr_debug("%d error: %d\n", block->block_index, error);
1445 	return error;
1446 }
1447 
incfs_read_file_signature(struct data_file * df,struct mem_range dst)1448 int incfs_read_file_signature(struct data_file *df, struct mem_range dst)
1449 {
1450 	struct backing_file_context *bfc = df->df_backing_file_context;
1451 	struct incfs_df_signature *sig;
1452 	int read_res = 0;
1453 
1454 	if (!dst.data)
1455 		return -EFAULT;
1456 
1457 	sig = df->df_signature;
1458 	if (!sig)
1459 		return 0;
1460 
1461 	if (dst.len < sig->sig_size)
1462 		return -E2BIG;
1463 
1464 	read_res = incfs_kread(bfc, dst.data, sig->sig_size, sig->sig_offset);
1465 
1466 	if (read_res < 0)
1467 		return read_res;
1468 
1469 	if (read_res != sig->sig_size)
1470 		return -EIO;
1471 
1472 	return read_res;
1473 }
1474 
incfs_process_new_hash_block(struct data_file * df,struct incfs_fill_block * block,u8 * data)1475 int incfs_process_new_hash_block(struct data_file *df,
1476 				 struct incfs_fill_block *block, u8 *data)
1477 {
1478 	struct backing_file_context *bfc = NULL;
1479 	struct mount_info *mi = NULL;
1480 	struct mtree *hash_tree = NULL;
1481 	struct incfs_df_signature *sig = NULL;
1482 	loff_t hash_area_base = 0;
1483 	loff_t hash_area_size = 0;
1484 	int error = 0;
1485 
1486 	if (!df || !block)
1487 		return -EFAULT;
1488 
1489 	if (!(block->flags & INCFS_BLOCK_FLAGS_HASH))
1490 		return -EINVAL;
1491 
1492 	bfc = df->df_backing_file_context;
1493 	mi = df->df_mount_info;
1494 
1495 	if (!df)
1496 		return -ENOENT;
1497 
1498 	hash_tree = df->df_hash_tree;
1499 	sig = df->df_signature;
1500 	if (!hash_tree || !sig || sig->hash_offset == 0)
1501 		return -ENOTSUPP;
1502 
1503 	hash_area_base = sig->hash_offset;
1504 	hash_area_size = sig->hash_size;
1505 	if (hash_area_size < block->block_index * INCFS_DATA_FILE_BLOCK_SIZE
1506 				+ block->data_len) {
1507 		/* Hash block goes beyond dedicated hash area of this file. */
1508 		return -ERANGE;
1509 	}
1510 
1511 	error = mutex_lock_interruptible(&bfc->bc_mutex);
1512 	if (!error) {
1513 		error = incfs_write_hash_block_to_backing_file(
1514 			bfc, range(data, block->data_len), block->block_index,
1515 			hash_area_base, df->df_blockmap_off, df->df_size);
1516 		mutex_unlock(&bfc->bc_mutex);
1517 	}
1518 	if (!error)
1519 		atomic_inc(&df->df_hash_blocks_written);
1520 
1521 	return error;
1522 }
1523 
process_blockmap_md(struct incfs_blockmap * bm,struct metadata_handler * handler)1524 static int process_blockmap_md(struct incfs_blockmap *bm,
1525 			       struct metadata_handler *handler)
1526 {
1527 	struct data_file *df = handler->context;
1528 	int error = 0;
1529 	loff_t base_off = le64_to_cpu(bm->m_base_offset);
1530 	u32 block_count = le32_to_cpu(bm->m_block_count);
1531 
1532 	if (!df)
1533 		return -EFAULT;
1534 
1535 	if (df->df_data_block_count > block_count)
1536 		return -EBADMSG;
1537 
1538 	df->df_total_block_count = block_count;
1539 	df->df_blockmap_off = base_off;
1540 	return error;
1541 }
1542 
process_file_signature_md(struct incfs_file_signature * sg,struct metadata_handler * handler)1543 static int process_file_signature_md(struct incfs_file_signature *sg,
1544 				struct metadata_handler *handler)
1545 {
1546 	struct data_file *df = handler->context;
1547 	struct mtree *hash_tree = NULL;
1548 	int error = 0;
1549 	struct incfs_df_signature *signature =
1550 		kzalloc(sizeof(*signature), GFP_NOFS);
1551 	void *buf = NULL;
1552 	ssize_t read;
1553 
1554 	if (!signature)
1555 		return -ENOMEM;
1556 
1557 	if (!df || !df->df_backing_file_context ||
1558 	    !df->df_backing_file_context->bc_file) {
1559 		error = -ENOENT;
1560 		goto out;
1561 	}
1562 
1563 	signature->hash_offset = le64_to_cpu(sg->sg_hash_tree_offset);
1564 	signature->hash_size = le32_to_cpu(sg->sg_hash_tree_size);
1565 	signature->sig_offset = le64_to_cpu(sg->sg_sig_offset);
1566 	signature->sig_size = le32_to_cpu(sg->sg_sig_size);
1567 
1568 	buf = kzalloc(signature->sig_size, GFP_NOFS);
1569 	if (!buf) {
1570 		error = -ENOMEM;
1571 		goto out;
1572 	}
1573 
1574 	read = incfs_kread(df->df_backing_file_context, buf,
1575 			   signature->sig_size, signature->sig_offset);
1576 	if (read < 0) {
1577 		error = read;
1578 		goto out;
1579 	}
1580 
1581 	if (read != signature->sig_size) {
1582 		error = -EINVAL;
1583 		goto out;
1584 	}
1585 
1586 	hash_tree = incfs_alloc_mtree(range(buf, signature->sig_size),
1587 				      df->df_data_block_count);
1588 	if (IS_ERR(hash_tree)) {
1589 		error = PTR_ERR(hash_tree);
1590 		hash_tree = NULL;
1591 		goto out;
1592 	}
1593 	if (hash_tree->hash_tree_area_size != signature->hash_size) {
1594 		error = -EINVAL;
1595 		goto out;
1596 	}
1597 	if (signature->hash_size > 0 &&
1598 	    handler->md_record_offset <= signature->hash_offset) {
1599 		error = -EINVAL;
1600 		goto out;
1601 	}
1602 	if (handler->md_record_offset <= signature->sig_offset) {
1603 		error = -EINVAL;
1604 		goto out;
1605 	}
1606 	df->df_hash_tree = hash_tree;
1607 	hash_tree = NULL;
1608 	df->df_signature = signature;
1609 	signature = NULL;
1610 out:
1611 	incfs_free_mtree(hash_tree);
1612 	kfree(signature);
1613 	kfree(buf);
1614 
1615 	return error;
1616 }
1617 
process_status_md(struct incfs_status * is,struct metadata_handler * handler)1618 static int process_status_md(struct incfs_status *is,
1619 			     struct metadata_handler *handler)
1620 {
1621 	struct data_file *df = handler->context;
1622 
1623 	df->df_initial_data_blocks_written =
1624 		le32_to_cpu(is->is_data_blocks_written);
1625 	atomic_set(&df->df_data_blocks_written,
1626 		   df->df_initial_data_blocks_written);
1627 
1628 	df->df_initial_hash_blocks_written =
1629 		le32_to_cpu(is->is_hash_blocks_written);
1630 	atomic_set(&df->df_hash_blocks_written,
1631 		   df->df_initial_hash_blocks_written);
1632 
1633 	df->df_status_offset = handler->md_record_offset;
1634 	return 0;
1635 }
1636 
process_file_verity_signature_md(struct incfs_file_verity_signature * vs,struct metadata_handler * handler)1637 static int process_file_verity_signature_md(
1638 		struct incfs_file_verity_signature *vs,
1639 		struct metadata_handler *handler)
1640 {
1641 	struct data_file *df = handler->context;
1642 	struct incfs_df_verity_signature *verity_signature;
1643 
1644 	if (!df)
1645 		return -EFAULT;
1646 
1647 	verity_signature = kzalloc(sizeof(*verity_signature), GFP_NOFS);
1648 	if (!verity_signature)
1649 		return -ENOMEM;
1650 
1651 	verity_signature->offset = le64_to_cpu(vs->vs_offset);
1652 	verity_signature->size = le32_to_cpu(vs->vs_size);
1653 	if (verity_signature->size > FS_VERITY_MAX_SIGNATURE_SIZE) {
1654 		kfree(verity_signature);
1655 		return -EFAULT;
1656 	}
1657 
1658 	df->df_verity_signature = verity_signature;
1659 	return 0;
1660 }
1661 
incfs_scan_metadata_chain(struct data_file * df)1662 static int incfs_scan_metadata_chain(struct data_file *df)
1663 {
1664 	struct metadata_handler *handler = NULL;
1665 	int result = 0;
1666 	int records_count = 0;
1667 	int error = 0;
1668 	struct backing_file_context *bfc = NULL;
1669 	int nondata_block_count;
1670 
1671 	if (!df || !df->df_backing_file_context)
1672 		return -EFAULT;
1673 
1674 	bfc = df->df_backing_file_context;
1675 
1676 	handler = kzalloc(sizeof(*handler), GFP_NOFS);
1677 	if (!handler)
1678 		return -ENOMEM;
1679 
1680 	handler->md_record_offset = df->df_metadata_off;
1681 	handler->context = df;
1682 	handler->handle_blockmap = process_blockmap_md;
1683 	handler->handle_signature = process_file_signature_md;
1684 	handler->handle_status = process_status_md;
1685 	handler->handle_verity_signature = process_file_verity_signature_md;
1686 
1687 	while (handler->md_record_offset > 0) {
1688 		error = incfs_read_next_metadata_record(bfc, handler);
1689 		if (error) {
1690 			pr_warn("incfs: Error during reading incfs-metadata record. Offset: %lld Record #%d Error code: %d\n",
1691 				handler->md_record_offset, records_count + 1,
1692 				-error);
1693 			break;
1694 		}
1695 		records_count++;
1696 	}
1697 	if (error) {
1698 		pr_warn("incfs: Error %d after reading %d incfs-metadata records.\n",
1699 			 -error, records_count);
1700 		result = error;
1701 	} else
1702 		result = records_count;
1703 
1704 	nondata_block_count = df->df_total_block_count -
1705 		df->df_data_block_count;
1706 	if (df->df_hash_tree) {
1707 		int hash_block_count = get_blocks_count_for_size(
1708 			df->df_hash_tree->hash_tree_area_size);
1709 
1710 		/*
1711 		 * Files that were created with a hash tree have the hash tree
1712 		 * included in the block map, i.e. nondata_block_count ==
1713 		 * hash_block_count.  Files whose hash tree was added by
1714 		 * FS_IOC_ENABLE_VERITY will still have the original block
1715 		 * count, i.e. nondata_block_count == 0.
1716 		 */
1717 		if (nondata_block_count != hash_block_count &&
1718 		    nondata_block_count != 0)
1719 			result = -EINVAL;
1720 	} else if (nondata_block_count != 0) {
1721 		result = -EINVAL;
1722 	}
1723 
1724 	kfree(handler);
1725 	return result;
1726 }
1727 
1728 /*
1729  * Quickly checks if there are pending reads with a serial number larger
1730  * than a given one.
1731  */
incfs_fresh_pending_reads_exist(struct mount_info * mi,int last_number)1732 bool incfs_fresh_pending_reads_exist(struct mount_info *mi, int last_number)
1733 {
1734 	bool result = false;
1735 
1736 	spin_lock(&mi->pending_read_lock);
1737 	result = (mi->mi_last_pending_read_number > last_number) &&
1738 		(mi->mi_pending_reads_count > 0);
1739 	spin_unlock(&mi->pending_read_lock);
1740 	return result;
1741 }
1742 
incfs_collect_pending_reads(struct mount_info * mi,int sn_lowerbound,struct incfs_pending_read_info * reads,struct incfs_pending_read_info2 * reads2,int reads_size,int * new_max_sn)1743 int incfs_collect_pending_reads(struct mount_info *mi, int sn_lowerbound,
1744 				struct incfs_pending_read_info *reads,
1745 				struct incfs_pending_read_info2 *reads2,
1746 				int reads_size, int *new_max_sn)
1747 {
1748 	int reported_reads = 0;
1749 	struct pending_read *entry = NULL;
1750 
1751 	if (!mi)
1752 		return -EFAULT;
1753 
1754 	if (reads_size <= 0)
1755 		return 0;
1756 
1757 	if (!incfs_fresh_pending_reads_exist(mi, sn_lowerbound))
1758 		return 0;
1759 
1760 	rcu_read_lock();
1761 
1762 	list_for_each_entry_rcu(entry, &mi->mi_reads_list_head, mi_reads_list) {
1763 		if (entry->serial_number <= sn_lowerbound)
1764 			continue;
1765 
1766 		if (reads) {
1767 			reads[reported_reads].file_id = entry->file_id;
1768 			reads[reported_reads].block_index = entry->block_index;
1769 			reads[reported_reads].serial_number =
1770 				entry->serial_number;
1771 			reads[reported_reads].timestamp_us =
1772 				entry->timestamp_us;
1773 		}
1774 
1775 		if (reads2) {
1776 			reads2[reported_reads].file_id = entry->file_id;
1777 			reads2[reported_reads].block_index = entry->block_index;
1778 			reads2[reported_reads].serial_number =
1779 				entry->serial_number;
1780 			reads2[reported_reads].timestamp_us =
1781 				entry->timestamp_us;
1782 			reads2[reported_reads].uid = entry->uid;
1783 		}
1784 
1785 		if (entry->serial_number > *new_max_sn)
1786 			*new_max_sn = entry->serial_number;
1787 
1788 		reported_reads++;
1789 		if (reported_reads >= reads_size)
1790 			break;
1791 	}
1792 
1793 	rcu_read_unlock();
1794 
1795 	return reported_reads;
1796 }
1797 
incfs_get_log_state(struct mount_info * mi)1798 struct read_log_state incfs_get_log_state(struct mount_info *mi)
1799 {
1800 	struct read_log *log = &mi->mi_log;
1801 	struct read_log_state result;
1802 
1803 	spin_lock(&log->rl_lock);
1804 	result = log->rl_head;
1805 	spin_unlock(&log->rl_lock);
1806 	return result;
1807 }
1808 
incfs_get_uncollected_logs_count(struct mount_info * mi,const struct read_log_state * state)1809 int incfs_get_uncollected_logs_count(struct mount_info *mi,
1810 				     const struct read_log_state *state)
1811 {
1812 	struct read_log *log = &mi->mi_log;
1813 	u32 generation;
1814 	u64 head_no, tail_no;
1815 
1816 	spin_lock(&log->rl_lock);
1817 	tail_no = log->rl_tail.current_record_no;
1818 	head_no = log->rl_head.current_record_no;
1819 	generation = log->rl_head.generation_id;
1820 	spin_unlock(&log->rl_lock);
1821 
1822 	if (generation != state->generation_id)
1823 		return head_no - tail_no;
1824 	else
1825 		return head_no - max_t(u64, tail_no, state->current_record_no);
1826 }
1827 
incfs_collect_logged_reads(struct mount_info * mi,struct read_log_state * state,struct incfs_pending_read_info * reads,struct incfs_pending_read_info2 * reads2,int reads_size)1828 int incfs_collect_logged_reads(struct mount_info *mi,
1829 			       struct read_log_state *state,
1830 			       struct incfs_pending_read_info *reads,
1831 			       struct incfs_pending_read_info2 *reads2,
1832 			       int reads_size)
1833 {
1834 	int dst_idx;
1835 	struct read_log *log = &mi->mi_log;
1836 	struct read_log_state *head, *tail;
1837 
1838 	spin_lock(&log->rl_lock);
1839 	head = &log->rl_head;
1840 	tail = &log->rl_tail;
1841 
1842 	if (state->generation_id != head->generation_id) {
1843 		pr_debug("read ptr is wrong generation: %u/%u",
1844 			 state->generation_id, head->generation_id);
1845 
1846 		*state = (struct read_log_state){
1847 			.generation_id = head->generation_id,
1848 		};
1849 	}
1850 
1851 	if (state->current_record_no < tail->current_record_no) {
1852 		pr_debug("read ptr is behind, moving: %u/%u -> %u/%u\n",
1853 			 (u32)state->next_offset,
1854 			 (u32)state->current_pass_no,
1855 			 (u32)tail->next_offset, (u32)tail->current_pass_no);
1856 
1857 		*state = *tail;
1858 	}
1859 
1860 	for (dst_idx = 0; dst_idx < reads_size; dst_idx++) {
1861 		if (state->current_record_no == head->current_record_no)
1862 			break;
1863 
1864 		log_read_one_record(log, state);
1865 
1866 		if (reads)
1867 			reads[dst_idx] = (struct incfs_pending_read_info) {
1868 				.file_id = state->base_record.file_id,
1869 				.block_index = state->base_record.block_index,
1870 				.serial_number = state->current_record_no,
1871 				.timestamp_us =
1872 					state->base_record.absolute_ts_us,
1873 			};
1874 
1875 		if (reads2)
1876 			reads2[dst_idx] = (struct incfs_pending_read_info2) {
1877 				.file_id = state->base_record.file_id,
1878 				.block_index = state->base_record.block_index,
1879 				.serial_number = state->current_record_no,
1880 				.timestamp_us =
1881 					state->base_record.absolute_ts_us,
1882 				.uid = state->base_record.uid,
1883 			};
1884 	}
1885 
1886 	spin_unlock(&log->rl_lock);
1887 	return dst_idx;
1888 }
1889 
1890