xref: /linux/fs/nfs/blocklayout/extent_tree.c (revision 4949009eb8d40a441dcddcd96e101e77d31cf1b2)
1 /*
2  * Copyright (c) 2014 Christoph Hellwig.
3  */
4 
5 #include <linux/vmalloc.h>
6 
7 #include "blocklayout.h"
8 
9 #define NFSDBG_FACILITY		NFSDBG_PNFS_LD
10 
11 static inline struct pnfs_block_extent *
12 ext_node(struct rb_node *node)
13 {
14 	return rb_entry(node, struct pnfs_block_extent, be_node);
15 }
16 
17 static struct pnfs_block_extent *
18 ext_tree_first(struct rb_root *root)
19 {
20 	struct rb_node *node = rb_first(root);
21 	return node ? ext_node(node) : NULL;
22 }
23 
24 static struct pnfs_block_extent *
25 ext_tree_prev(struct pnfs_block_extent *be)
26 {
27 	struct rb_node *node = rb_prev(&be->be_node);
28 	return node ? ext_node(node) : NULL;
29 }
30 
31 static struct pnfs_block_extent *
32 ext_tree_next(struct pnfs_block_extent *be)
33 {
34 	struct rb_node *node = rb_next(&be->be_node);
35 	return node ? ext_node(node) : NULL;
36 }
37 
38 static inline sector_t
39 ext_f_end(struct pnfs_block_extent *be)
40 {
41 	return be->be_f_offset + be->be_length;
42 }
43 
44 static struct pnfs_block_extent *
45 __ext_tree_search(struct rb_root *root, sector_t start)
46 {
47 	struct rb_node *node = root->rb_node;
48 	struct pnfs_block_extent *be = NULL;
49 
50 	while (node) {
51 		be = ext_node(node);
52 		if (start < be->be_f_offset)
53 			node = node->rb_left;
54 		else if (start >= ext_f_end(be))
55 			node = node->rb_right;
56 		else
57 			return be;
58 	}
59 
60 	if (be) {
61 		if (start < be->be_f_offset)
62 			return be;
63 
64 		if (start >= ext_f_end(be))
65 			return ext_tree_next(be);
66 	}
67 
68 	return NULL;
69 }
70 
71 static bool
72 ext_can_merge(struct pnfs_block_extent *be1, struct pnfs_block_extent *be2)
73 {
74 	if (be1->be_state != be2->be_state)
75 		return false;
76 	if (be1->be_device != be2->be_device)
77 		return false;
78 
79 	if (be1->be_f_offset + be1->be_length != be2->be_f_offset)
80 		return false;
81 
82 	if (be1->be_state != PNFS_BLOCK_NONE_DATA &&
83 	    (be1->be_v_offset + be1->be_length != be2->be_v_offset))
84 		return false;
85 
86 	if (be1->be_state == PNFS_BLOCK_INVALID_DATA &&
87 	    be1->be_tag != be2->be_tag)
88 		return false;
89 
90 	return true;
91 }
92 
93 static struct pnfs_block_extent *
94 ext_try_to_merge_left(struct rb_root *root, struct pnfs_block_extent *be)
95 {
96 	struct pnfs_block_extent *left = ext_tree_prev(be);
97 
98 	if (left && ext_can_merge(left, be)) {
99 		left->be_length += be->be_length;
100 		rb_erase(&be->be_node, root);
101 		nfs4_put_deviceid_node(be->be_device);
102 		kfree(be);
103 		return left;
104 	}
105 
106 	return be;
107 }
108 
109 static struct pnfs_block_extent *
110 ext_try_to_merge_right(struct rb_root *root, struct pnfs_block_extent *be)
111 {
112 	struct pnfs_block_extent *right = ext_tree_next(be);
113 
114 	if (right && ext_can_merge(be, right)) {
115 		be->be_length += right->be_length;
116 		rb_erase(&right->be_node, root);
117 		nfs4_put_deviceid_node(right->be_device);
118 		kfree(right);
119 	}
120 
121 	return be;
122 }
123 
124 static void
125 __ext_tree_insert(struct rb_root *root,
126 		struct pnfs_block_extent *new, bool merge_ok)
127 {
128 	struct rb_node **p = &root->rb_node, *parent = NULL;
129 	struct pnfs_block_extent *be;
130 
131 	while (*p) {
132 		parent = *p;
133 		be = ext_node(parent);
134 
135 		if (new->be_f_offset < be->be_f_offset) {
136 			if (merge_ok && ext_can_merge(new, be)) {
137 				be->be_f_offset = new->be_f_offset;
138 				if (be->be_state != PNFS_BLOCK_NONE_DATA)
139 					be->be_v_offset = new->be_v_offset;
140 				be->be_length += new->be_length;
141 				be = ext_try_to_merge_left(root, be);
142 				goto free_new;
143 			}
144 			p = &(*p)->rb_left;
145 		} else if (new->be_f_offset >= ext_f_end(be)) {
146 			if (merge_ok && ext_can_merge(be, new)) {
147 				be->be_length += new->be_length;
148 				be = ext_try_to_merge_right(root, be);
149 				goto free_new;
150 			}
151 			p = &(*p)->rb_right;
152 		} else {
153 			BUG();
154 		}
155 	}
156 
157 	rb_link_node(&new->be_node, parent, p);
158 	rb_insert_color(&new->be_node, root);
159 	return;
160 free_new:
161 	nfs4_put_deviceid_node(new->be_device);
162 	kfree(new);
163 }
164 
165 static int
166 __ext_tree_remove(struct rb_root *root, sector_t start, sector_t end)
167 {
168 	struct pnfs_block_extent *be;
169 	sector_t len1 = 0, len2 = 0;
170 	sector_t orig_v_offset;
171 	sector_t orig_len;
172 
173 	be = __ext_tree_search(root, start);
174 	if (!be)
175 		return 0;
176 	if (be->be_f_offset >= end)
177 		return 0;
178 
179 	orig_v_offset = be->be_v_offset;
180 	orig_len = be->be_length;
181 
182 	if (start > be->be_f_offset)
183 		len1 = start - be->be_f_offset;
184 	if (ext_f_end(be) > end)
185 		len2 = ext_f_end(be) - end;
186 
187 	if (len2 > 0) {
188 		if (len1 > 0) {
189 			struct pnfs_block_extent *new;
190 
191 			new = kzalloc(sizeof(*new), GFP_ATOMIC);
192 			if (!new)
193 				return -ENOMEM;
194 
195 			be->be_length = len1;
196 
197 			new->be_f_offset = end;
198 			if (be->be_state != PNFS_BLOCK_NONE_DATA) {
199 				new->be_v_offset =
200 					orig_v_offset + orig_len - len2;
201 			}
202 			new->be_length = len2;
203 			new->be_state = be->be_state;
204 			new->be_tag = be->be_tag;
205 			new->be_device = nfs4_get_deviceid(be->be_device);
206 
207 			__ext_tree_insert(root, new, true);
208 		} else {
209 			be->be_f_offset = end;
210 			if (be->be_state != PNFS_BLOCK_NONE_DATA) {
211 				be->be_v_offset =
212 					orig_v_offset + orig_len - len2;
213 			}
214 			be->be_length = len2;
215 		}
216 	} else {
217 		if (len1 > 0) {
218 			be->be_length = len1;
219 			be = ext_tree_next(be);
220 		}
221 
222 		while (be && ext_f_end(be) <= end) {
223 			struct pnfs_block_extent *next = ext_tree_next(be);
224 
225 			rb_erase(&be->be_node, root);
226 			nfs4_put_deviceid_node(be->be_device);
227 			kfree(be);
228 			be = next;
229 		}
230 
231 		if (be && be->be_f_offset < end) {
232 			len1 = ext_f_end(be) - end;
233 			be->be_f_offset = end;
234 			if (be->be_state != PNFS_BLOCK_NONE_DATA)
235 				be->be_v_offset += be->be_length - len1;
236 			be->be_length = len1;
237 		}
238 	}
239 
240 	return 0;
241 }
242 
243 int
244 ext_tree_insert(struct pnfs_block_layout *bl, struct pnfs_block_extent *new)
245 {
246 	struct pnfs_block_extent *be;
247 	struct rb_root *root;
248 	int err = 0;
249 
250 	switch (new->be_state) {
251 	case PNFS_BLOCK_READWRITE_DATA:
252 	case PNFS_BLOCK_INVALID_DATA:
253 		root = &bl->bl_ext_rw;
254 		break;
255 	case PNFS_BLOCK_READ_DATA:
256 	case PNFS_BLOCK_NONE_DATA:
257 		root = &bl->bl_ext_ro;
258 		break;
259 	default:
260 		dprintk("invalid extent type\n");
261 		return -EINVAL;
262 	}
263 
264 	spin_lock(&bl->bl_ext_lock);
265 retry:
266 	be = __ext_tree_search(root, new->be_f_offset);
267 	if (!be || be->be_f_offset >= ext_f_end(new)) {
268 		__ext_tree_insert(root, new, true);
269 	} else if (new->be_f_offset >= be->be_f_offset) {
270 		if (ext_f_end(new) <= ext_f_end(be)) {
271 			nfs4_put_deviceid_node(new->be_device);
272 			kfree(new);
273 		} else {
274 			sector_t new_len = ext_f_end(new) - ext_f_end(be);
275 			sector_t diff = new->be_length - new_len;
276 
277 			new->be_f_offset += diff;
278 			new->be_v_offset += diff;
279 			new->be_length = new_len;
280 			goto retry;
281 		}
282 	} else if (ext_f_end(new) <= ext_f_end(be)) {
283 		new->be_length = be->be_f_offset - new->be_f_offset;
284 		__ext_tree_insert(root, new, true);
285 	} else {
286 		struct pnfs_block_extent *split;
287 		sector_t new_len = ext_f_end(new) - ext_f_end(be);
288 		sector_t diff = new->be_length - new_len;
289 
290 		split = kmemdup(new, sizeof(*new), GFP_ATOMIC);
291 		if (!split) {
292 			err = -EINVAL;
293 			goto out;
294 		}
295 
296 		split->be_length = be->be_f_offset - split->be_f_offset;
297 		split->be_device = nfs4_get_deviceid(new->be_device);
298 		__ext_tree_insert(root, split, true);
299 
300 		new->be_f_offset += diff;
301 		new->be_v_offset += diff;
302 		new->be_length = new_len;
303 		goto retry;
304 	}
305 out:
306 	spin_unlock(&bl->bl_ext_lock);
307 	return err;
308 }
309 
310 static bool
311 __ext_tree_lookup(struct rb_root *root, sector_t isect,
312 		struct pnfs_block_extent *ret)
313 {
314 	struct rb_node *node;
315 	struct pnfs_block_extent *be;
316 
317 	node = root->rb_node;
318 	while (node) {
319 		be = ext_node(node);
320 		if (isect < be->be_f_offset)
321 			node = node->rb_left;
322 		else if (isect >= ext_f_end(be))
323 			node = node->rb_right;
324 		else {
325 			*ret = *be;
326 			return true;
327 		}
328 	}
329 
330 	return false;
331 }
332 
333 bool
334 ext_tree_lookup(struct pnfs_block_layout *bl, sector_t isect,
335 	    struct pnfs_block_extent *ret, bool rw)
336 {
337 	bool found = false;
338 
339 	spin_lock(&bl->bl_ext_lock);
340 	if (!rw)
341 		found = __ext_tree_lookup(&bl->bl_ext_ro, isect, ret);
342 	if (!found)
343 		found = __ext_tree_lookup(&bl->bl_ext_rw, isect, ret);
344 	spin_unlock(&bl->bl_ext_lock);
345 
346 	return found;
347 }
348 
349 int ext_tree_remove(struct pnfs_block_layout *bl, bool rw,
350 		sector_t start, sector_t end)
351 {
352 	int err, err2;
353 
354 	spin_lock(&bl->bl_ext_lock);
355 	err = __ext_tree_remove(&bl->bl_ext_ro, start, end);
356 	if (rw) {
357 		err2 = __ext_tree_remove(&bl->bl_ext_rw, start, end);
358 		if (!err)
359 			err = err2;
360 	}
361 	spin_unlock(&bl->bl_ext_lock);
362 
363 	return err;
364 }
365 
366 static int
367 ext_tree_split(struct rb_root *root, struct pnfs_block_extent *be,
368 		sector_t split)
369 {
370 	struct pnfs_block_extent *new;
371 	sector_t orig_len = be->be_length;
372 
373 	new = kzalloc(sizeof(*new), GFP_ATOMIC);
374 	if (!new)
375 		return -ENOMEM;
376 
377 	be->be_length = split - be->be_f_offset;
378 
379 	new->be_f_offset = split;
380 	if (be->be_state != PNFS_BLOCK_NONE_DATA)
381 		new->be_v_offset = be->be_v_offset + be->be_length;
382 	new->be_length = orig_len - be->be_length;
383 	new->be_state = be->be_state;
384 	new->be_tag = be->be_tag;
385 	new->be_device = nfs4_get_deviceid(be->be_device);
386 
387 	__ext_tree_insert(root, new, false);
388 	return 0;
389 }
390 
391 int
392 ext_tree_mark_written(struct pnfs_block_layout *bl, sector_t start,
393 		sector_t len)
394 {
395 	struct rb_root *root = &bl->bl_ext_rw;
396 	sector_t end = start + len;
397 	struct pnfs_block_extent *be;
398 	int err = 0;
399 
400 	spin_lock(&bl->bl_ext_lock);
401 	/*
402 	 * First remove all COW extents or holes from written to range.
403 	 */
404 	err = __ext_tree_remove(&bl->bl_ext_ro, start, end);
405 	if (err)
406 		goto out;
407 
408 	/*
409 	 * Then mark all invalid extents in the range as written to.
410 	 */
411 	for (be = __ext_tree_search(root, start); be; be = ext_tree_next(be)) {
412 		if (be->be_f_offset >= end)
413 			break;
414 
415 		if (be->be_state != PNFS_BLOCK_INVALID_DATA || be->be_tag)
416 			continue;
417 
418 		if (be->be_f_offset < start) {
419 			struct pnfs_block_extent *left = ext_tree_prev(be);
420 
421 			if (left && ext_can_merge(left, be)) {
422 				sector_t diff = start - be->be_f_offset;
423 
424 				left->be_length += diff;
425 
426 				be->be_f_offset += diff;
427 				be->be_v_offset += diff;
428 				be->be_length -= diff;
429 			} else {
430 				err = ext_tree_split(root, be, start);
431 				if (err)
432 					goto out;
433 			}
434 		}
435 
436 		if (ext_f_end(be) > end) {
437 			struct pnfs_block_extent *right = ext_tree_next(be);
438 
439 			if (right && ext_can_merge(be, right)) {
440 				sector_t diff = end - be->be_f_offset;
441 
442 				be->be_length -= diff;
443 
444 				right->be_f_offset -= diff;
445 				right->be_v_offset -= diff;
446 				right->be_length += diff;
447 			} else {
448 				err = ext_tree_split(root, be, end);
449 				if (err)
450 					goto out;
451 			}
452 		}
453 
454 		if (be->be_f_offset >= start && ext_f_end(be) <= end) {
455 			be->be_tag = EXTENT_WRITTEN;
456 			be = ext_try_to_merge_left(root, be);
457 			be = ext_try_to_merge_right(root, be);
458 		}
459 	}
460 out:
461 	spin_unlock(&bl->bl_ext_lock);
462 	return err;
463 }
464 
465 static void ext_tree_free_commitdata(struct nfs4_layoutcommit_args *arg,
466 		size_t buffer_size)
467 {
468 	if (arg->layoutupdate_pages != &arg->layoutupdate_page) {
469 		int nr_pages = DIV_ROUND_UP(buffer_size, PAGE_SIZE), i;
470 
471 		for (i = 0; i < nr_pages; i++)
472 			put_page(arg->layoutupdate_pages[i]);
473 		kfree(arg->layoutupdate_pages);
474 	} else {
475 		put_page(arg->layoutupdate_page);
476 	}
477 }
478 
479 static int ext_tree_encode_commit(struct pnfs_block_layout *bl, __be32 *p,
480 		size_t buffer_size, size_t *count)
481 {
482 	struct pnfs_block_extent *be;
483 	int ret = 0;
484 
485 	spin_lock(&bl->bl_ext_lock);
486 	for (be = ext_tree_first(&bl->bl_ext_rw); be; be = ext_tree_next(be)) {
487 		if (be->be_state != PNFS_BLOCK_INVALID_DATA ||
488 		    be->be_tag != EXTENT_WRITTEN)
489 			continue;
490 
491 		(*count)++;
492 		if (*count * BL_EXTENT_SIZE > buffer_size) {
493 			/* keep counting.. */
494 			ret = -ENOSPC;
495 			continue;
496 		}
497 
498 		p = xdr_encode_opaque_fixed(p, be->be_device->deviceid.data,
499 				NFS4_DEVICEID4_SIZE);
500 		p = xdr_encode_hyper(p, be->be_f_offset << SECTOR_SHIFT);
501 		p = xdr_encode_hyper(p, be->be_length << SECTOR_SHIFT);
502 		p = xdr_encode_hyper(p, 0LL);
503 		*p++ = cpu_to_be32(PNFS_BLOCK_READWRITE_DATA);
504 
505 		be->be_tag = EXTENT_COMMITTING;
506 	}
507 	spin_unlock(&bl->bl_ext_lock);
508 
509 	return ret;
510 }
511 
512 int
513 ext_tree_prepare_commit(struct nfs4_layoutcommit_args *arg)
514 {
515 	struct pnfs_block_layout *bl = BLK_LO2EXT(NFS_I(arg->inode)->layout);
516 	size_t count = 0, buffer_size = PAGE_SIZE;
517 	__be32 *start_p;
518 	int ret;
519 
520 	dprintk("%s enter\n", __func__);
521 
522 	arg->layoutupdate_page = alloc_page(GFP_NOFS);
523 	if (!arg->layoutupdate_page)
524 		return -ENOMEM;
525 	start_p = page_address(arg->layoutupdate_page);
526 	arg->layoutupdate_pages = &arg->layoutupdate_page;
527 
528 retry:
529 	ret = ext_tree_encode_commit(bl, start_p + 1, buffer_size, &count);
530 	if (unlikely(ret)) {
531 		ext_tree_free_commitdata(arg, buffer_size);
532 
533 		buffer_size = sizeof(__be32) + BL_EXTENT_SIZE * count;
534 		count = 0;
535 
536 		arg->layoutupdate_pages =
537 			kcalloc(DIV_ROUND_UP(buffer_size, PAGE_SIZE),
538 				sizeof(struct page *), GFP_NOFS);
539 		if (!arg->layoutupdate_pages)
540 			return -ENOMEM;
541 
542 		start_p = __vmalloc(buffer_size, GFP_NOFS, PAGE_KERNEL);
543 		if (!start_p) {
544 			kfree(arg->layoutupdate_pages);
545 			return -ENOMEM;
546 		}
547 
548 		goto retry;
549 	}
550 
551 	*start_p = cpu_to_be32(count);
552 	arg->layoutupdate_len = sizeof(__be32) + BL_EXTENT_SIZE * count;
553 
554 	if (unlikely(arg->layoutupdate_pages != &arg->layoutupdate_page)) {
555 		__be32 *p = start_p;
556 		int i = 0;
557 
558 		for (p = start_p;
559 		     p < start_p + arg->layoutupdate_len;
560 		     p += PAGE_SIZE) {
561 			arg->layoutupdate_pages[i++] = vmalloc_to_page(p);
562 		}
563 	}
564 
565 	dprintk("%s found %zu ranges\n", __func__, count);
566 	return 0;
567 }
568 
569 void
570 ext_tree_mark_committed(struct nfs4_layoutcommit_args *arg, int status)
571 {
572 	struct pnfs_block_layout *bl = BLK_LO2EXT(NFS_I(arg->inode)->layout);
573 	struct rb_root *root = &bl->bl_ext_rw;
574 	struct pnfs_block_extent *be;
575 
576 	dprintk("%s status %d\n", __func__, status);
577 
578 	ext_tree_free_commitdata(arg, arg->layoutupdate_len);
579 
580 	spin_lock(&bl->bl_ext_lock);
581 	for (be = ext_tree_first(root); be; be = ext_tree_next(be)) {
582 		if (be->be_state != PNFS_BLOCK_INVALID_DATA ||
583 		    be->be_tag != EXTENT_COMMITTING)
584 			continue;
585 
586 		if (status) {
587 			/*
588 			 * Mark as written and try again.
589 			 *
590 			 * XXX: some real error handling here wouldn't hurt..
591 			 */
592 			be->be_tag = EXTENT_WRITTEN;
593 		} else {
594 			be->be_state = PNFS_BLOCK_READWRITE_DATA;
595 			be->be_tag = 0;
596 		}
597 
598 		be = ext_try_to_merge_left(root, be);
599 		be = ext_try_to_merge_right(root, be);
600 	}
601 	spin_unlock(&bl->bl_ext_lock);
602 }
603