blob: 3247d2422beae75d71d95345287d78135e5bc50f [file] [log] [blame]
Gao Xiang29b24f62019-07-31 23:57:31 +08001// SPDX-License-Identifier: GPL-2.0-only
Gao Xiang02827e12018-07-26 20:21:58 +08002/*
Gao Xiang02827e12018-07-26 20:21:58 +08003 * Copyright (C) 2018 HUAWEI, Inc.
Alexander A. Klimov592e7cd2020-07-13 15:09:44 +02004 * https://www.huawei.com/
Gao Xiang06a304c2022-07-15 23:41:51 +08005 * Copyright (C) 2022 Alibaba Cloud
Gao Xiang02827e12018-07-26 20:21:58 +08006 */
Gao Xiang27481232019-06-24 15:22:54 +08007#include "compress.h"
Gao Xiang3883a792018-07-26 20:22:06 +08008#include <linux/prefetch.h>
Christoph Hellwig99486c52022-09-15 10:41:59 +01009#include <linux/psi.h>
Sandeep Dhavale3fffb582023-02-08 17:33:22 +080010#include <linux/cpuhotplug.h>
Chen Gong284db122018-09-18 22:27:27 +080011#include <trace/events/erofs.h>
12
Gao Xianga9a94d92023-02-04 17:30:38 +080013#define Z_EROFS_PCLUSTER_MAX_PAGES (Z_EROFS_PCLUSTER_MAX_SIZE / PAGE_SIZE)
14#define Z_EROFS_INLINE_BVECS 2
15
16/*
17 * let's leave a type here in case of introducing
18 * another tagged pointer later.
19 */
20typedef void *z_erofs_next_pcluster_t;
21
22struct z_erofs_bvec {
23 struct page *page;
24 int offset;
25 unsigned int end;
26};
27
28#define __Z_EROFS_BVSET(name, total) \
29struct name { \
30 /* point to the next page which contains the following bvecs */ \
31 struct page *nextpage; \
32 struct z_erofs_bvec bvec[total]; \
33}
34__Z_EROFS_BVSET(z_erofs_bvset,);
35__Z_EROFS_BVSET(z_erofs_bvset_inline, Z_EROFS_INLINE_BVECS);
36
37/*
38 * Structure fields follow one of the following exclusion rules.
39 *
40 * I: Modifiable by initialization/destruction paths and read-only
41 * for everyone else;
42 *
43 * L: Field should be protected by the pcluster lock;
44 *
45 * A: Field should be accessed / updated in atomic for parallelized code.
46 */
47struct z_erofs_pcluster {
48 struct erofs_workgroup obj;
49 struct mutex lock;
50
51 /* A: point to next chained pcluster or TAILs */
52 z_erofs_next_pcluster_t next;
53
54 /* L: the maximum decompression size of this round */
55 unsigned int length;
56
57 /* L: total number of bvecs */
58 unsigned int vcnt;
59
60 /* I: page offset of start position of decompression */
61 unsigned short pageofs_out;
62
63 /* I: page offset of inline compressed data */
64 unsigned short pageofs_in;
65
66 union {
67 /* L: inline a certain number of bvec for bootstrap */
68 struct z_erofs_bvset_inline bvset;
69
70 /* I: can be used to free the pcluster by RCU. */
71 struct rcu_head rcu;
72 };
73
74 union {
75 /* I: physical cluster size in pages */
76 unsigned short pclusterpages;
77
78 /* I: tailpacking inline compressed size */
79 unsigned short tailpacking_size;
80 };
81
82 /* I: compression algorithm format */
83 unsigned char algorithmformat;
84
85 /* L: whether partial decompression or not */
86 bool partial;
87
88 /* L: indicate several pageofs_outs or not */
89 bool multibases;
90
91 /* A: compressed bvecs (can be cached or inplaced pages) */
92 struct z_erofs_bvec compressed_bvecs[];
93};
94
95/* let's avoid the valid 32-bit kernel addresses */
96
97/* the chained workgroup has't submitted io (still open) */
98#define Z_EROFS_PCLUSTER_TAIL ((void *)0x5F0ECAFE)
99/* the chained workgroup has already submitted io */
100#define Z_EROFS_PCLUSTER_TAIL_CLOSED ((void *)0x5F0EDEAD)
101
102#define Z_EROFS_PCLUSTER_NIL (NULL)
103
104struct z_erofs_decompressqueue {
105 struct super_block *sb;
106 atomic_t pending_bios;
107 z_erofs_next_pcluster_t head;
108
109 union {
110 struct completion done;
111 struct work_struct work;
Sandeep Dhavale3fffb582023-02-08 17:33:22 +0800112 struct kthread_work kthread_work;
Gao Xianga9a94d92023-02-04 17:30:38 +0800113 } u;
114 bool eio, sync;
115};
116
117static inline bool z_erofs_is_inline_pcluster(struct z_erofs_pcluster *pcl)
118{
119 return !pcl->obj.index;
120}
121
122static inline unsigned int z_erofs_pclusterpages(struct z_erofs_pcluster *pcl)
123{
124 if (z_erofs_is_inline_pcluster(pcl))
125 return 1;
126 return pcl->pclusterpages;
127}
128
129/*
130 * bit 30: I/O error occurred on this page
131 * bit 0 - 29: remaining parts to complete this page
132 */
133#define Z_EROFS_PAGE_EIO (1 << 30)
134
135static inline void z_erofs_onlinepage_init(struct page *page)
136{
137 union {
138 atomic_t o;
139 unsigned long v;
140 } u = { .o = ATOMIC_INIT(1) };
141
142 set_page_private(page, u.v);
143 smp_wmb();
144 SetPagePrivate(page);
145}
146
147static inline void z_erofs_onlinepage_split(struct page *page)
148{
149 atomic_inc((atomic_t *)&page->private);
150}
151
152static inline void z_erofs_page_mark_eio(struct page *page)
153{
154 int orig;
155
156 do {
157 orig = atomic_read((atomic_t *)&page->private);
158 } while (atomic_cmpxchg((atomic_t *)&page->private, orig,
159 orig | Z_EROFS_PAGE_EIO) != orig);
160}
161
162static inline void z_erofs_onlinepage_endio(struct page *page)
163{
164 unsigned int v;
165
166 DBG_BUGON(!PagePrivate(page));
167 v = atomic_dec_return((atomic_t *)&page->private);
168 if (!(v & ~Z_EROFS_PAGE_EIO)) {
169 set_page_private(page, 0);
170 ClearPagePrivate(page);
171 if (!(v & Z_EROFS_PAGE_EIO))
172 SetPageUptodate(page);
173 unlock_page(page);
174 }
175}
176
177#define Z_EROFS_ONSTACK_PAGES 32
178
Gao Xiang672e5472018-12-08 00:19:14 +0800179/*
Gao Xiang9f6cc762021-04-07 12:39:20 +0800180 * since pclustersize is variable for big pcluster feature, introduce slab
181 * pools implementation for different pcluster sizes.
182 */
183struct z_erofs_pcluster_slab {
184 struct kmem_cache *slab;
185 unsigned int maxpages;
186 char name[48];
187};
188
189#define _PCLP(n) { .maxpages = n }
190
191static struct z_erofs_pcluster_slab pcluster_pool[] __read_mostly = {
192 _PCLP(1), _PCLP(4), _PCLP(16), _PCLP(64), _PCLP(128),
193 _PCLP(Z_EROFS_PCLUSTER_MAX_PAGES)
194};
195
Gao Xiang06a304c2022-07-15 23:41:51 +0800196struct z_erofs_bvec_iter {
197 struct page *bvpage;
198 struct z_erofs_bvset *bvset;
199 unsigned int nr, cur;
200};
201
202static struct page *z_erofs_bvec_iter_end(struct z_erofs_bvec_iter *iter)
203{
204 if (iter->bvpage)
205 kunmap_local(iter->bvset);
206 return iter->bvpage;
207}
208
209static struct page *z_erofs_bvset_flip(struct z_erofs_bvec_iter *iter)
210{
211 unsigned long base = (unsigned long)((struct z_erofs_bvset *)0)->bvec;
212 /* have to access nextpage in advance, otherwise it will be unmapped */
213 struct page *nextpage = iter->bvset->nextpage;
214 struct page *oldpage;
215
216 DBG_BUGON(!nextpage);
217 oldpage = z_erofs_bvec_iter_end(iter);
218 iter->bvpage = nextpage;
219 iter->bvset = kmap_local_page(nextpage);
220 iter->nr = (PAGE_SIZE - base) / sizeof(struct z_erofs_bvec);
221 iter->cur = 0;
222 return oldpage;
223}
224
225static void z_erofs_bvec_iter_begin(struct z_erofs_bvec_iter *iter,
226 struct z_erofs_bvset_inline *bvset,
227 unsigned int bootstrap_nr,
228 unsigned int cur)
229{
230 *iter = (struct z_erofs_bvec_iter) {
231 .nr = bootstrap_nr,
232 .bvset = (struct z_erofs_bvset *)bvset,
233 };
234
235 while (cur > iter->nr) {
236 cur -= iter->nr;
237 z_erofs_bvset_flip(iter);
238 }
239 iter->cur = cur;
240}
241
242static int z_erofs_bvec_enqueue(struct z_erofs_bvec_iter *iter,
243 struct z_erofs_bvec *bvec,
244 struct page **candidate_bvpage)
245{
246 if (iter->cur == iter->nr) {
247 if (!*candidate_bvpage)
248 return -EAGAIN;
249
250 DBG_BUGON(iter->bvset->nextpage);
251 iter->bvset->nextpage = *candidate_bvpage;
252 z_erofs_bvset_flip(iter);
253
254 iter->bvset->nextpage = NULL;
255 *candidate_bvpage = NULL;
256 }
257 iter->bvset->bvec[iter->cur++] = *bvec;
258 return 0;
259}
260
261static void z_erofs_bvec_dequeue(struct z_erofs_bvec_iter *iter,
262 struct z_erofs_bvec *bvec,
263 struct page **old_bvpage)
264{
265 if (iter->cur == iter->nr)
266 *old_bvpage = z_erofs_bvset_flip(iter);
267 else
268 *old_bvpage = NULL;
269 *bvec = iter->bvset->bvec[iter->cur++];
270}
271
Gao Xiang9f6cc762021-04-07 12:39:20 +0800272static void z_erofs_destroy_pcluster_pool(void)
273{
274 int i;
275
276 for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) {
277 if (!pcluster_pool[i].slab)
278 continue;
279 kmem_cache_destroy(pcluster_pool[i].slab);
280 pcluster_pool[i].slab = NULL;
281 }
282}
283
284static int z_erofs_create_pcluster_pool(void)
285{
286 struct z_erofs_pcluster_slab *pcs;
287 struct z_erofs_pcluster *a;
288 unsigned int size;
289
290 for (pcs = pcluster_pool;
291 pcs < pcluster_pool + ARRAY_SIZE(pcluster_pool); ++pcs) {
Gao Xianged722fb2022-07-15 23:41:54 +0800292 size = struct_size(a, compressed_bvecs, pcs->maxpages);
Gao Xiang9f6cc762021-04-07 12:39:20 +0800293
294 sprintf(pcs->name, "erofs_pcluster-%u", pcs->maxpages);
295 pcs->slab = kmem_cache_create(pcs->name, size, 0,
296 SLAB_RECLAIM_ACCOUNT, NULL);
297 if (pcs->slab)
298 continue;
299
300 z_erofs_destroy_pcluster_pool();
301 return -ENOMEM;
302 }
303 return 0;
304}
305
306static struct z_erofs_pcluster *z_erofs_alloc_pcluster(unsigned int nrpages)
307{
308 int i;
309
310 for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) {
311 struct z_erofs_pcluster_slab *pcs = pcluster_pool + i;
312 struct z_erofs_pcluster *pcl;
313
314 if (nrpages > pcs->maxpages)
315 continue;
316
317 pcl = kmem_cache_zalloc(pcs->slab, GFP_NOFS);
318 if (!pcl)
319 return ERR_PTR(-ENOMEM);
320 pcl->pclusterpages = nrpages;
321 return pcl;
322 }
323 return ERR_PTR(-EINVAL);
324}
325
326static void z_erofs_free_pcluster(struct z_erofs_pcluster *pcl)
327{
Yue Hucecf8642021-12-29 07:29:19 +0800328 unsigned int pclusterpages = z_erofs_pclusterpages(pcl);
Gao Xiang9f6cc762021-04-07 12:39:20 +0800329 int i;
330
331 for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) {
332 struct z_erofs_pcluster_slab *pcs = pcluster_pool + i;
333
Yue Hucecf8642021-12-29 07:29:19 +0800334 if (pclusterpages > pcs->maxpages)
Gao Xiang9f6cc762021-04-07 12:39:20 +0800335 continue;
336
337 kmem_cache_free(pcs->slab, pcl);
338 return;
339 }
340 DBG_BUGON(1);
341}
342
Gao Xiang3883a792018-07-26 20:22:06 +0800343static struct workqueue_struct *z_erofs_workqueue __read_mostly;
Gao Xiang3883a792018-07-26 20:22:06 +0800344
Sandeep Dhavale3fffb582023-02-08 17:33:22 +0800345#ifdef CONFIG_EROFS_FS_PCPU_KTHREAD
346static struct kthread_worker __rcu **z_erofs_pcpu_workers;
347
348static void erofs_destroy_percpu_workers(void)
Gao Xiang3883a792018-07-26 20:22:06 +0800349{
Sandeep Dhavale3fffb582023-02-08 17:33:22 +0800350 struct kthread_worker *worker;
351 unsigned int cpu;
352
353 for_each_possible_cpu(cpu) {
354 worker = rcu_dereference_protected(
355 z_erofs_pcpu_workers[cpu], 1);
356 rcu_assign_pointer(z_erofs_pcpu_workers[cpu], NULL);
357 if (worker)
358 kthread_destroy_worker(worker);
359 }
360 kfree(z_erofs_pcpu_workers);
Gao Xiang3883a792018-07-26 20:22:06 +0800361}
362
Sandeep Dhavale3fffb582023-02-08 17:33:22 +0800363static struct kthread_worker *erofs_init_percpu_worker(int cpu)
Gao Xiang3883a792018-07-26 20:22:06 +0800364{
Sandeep Dhavale3fffb582023-02-08 17:33:22 +0800365 struct kthread_worker *worker =
366 kthread_create_worker_on_cpu(cpu, 0, "erofs_worker/%u", cpu);
Gao Xiang3883a792018-07-26 20:22:06 +0800367
Sandeep Dhavale3fffb582023-02-08 17:33:22 +0800368 if (IS_ERR(worker))
369 return worker;
370 if (IS_ENABLED(CONFIG_EROFS_FS_PCPU_KTHREAD_HIPRI))
371 sched_set_fifo_low(worker->task);
372 else
373 sched_set_normal(worker->task, 0);
374 return worker;
375}
376
377static int erofs_init_percpu_workers(void)
378{
379 struct kthread_worker *worker;
380 unsigned int cpu;
381
382 z_erofs_pcpu_workers = kcalloc(num_possible_cpus(),
383 sizeof(struct kthread_worker *), GFP_ATOMIC);
384 if (!z_erofs_pcpu_workers)
385 return -ENOMEM;
386
387 for_each_online_cpu(cpu) { /* could miss cpu{off,on}line? */
388 worker = erofs_init_percpu_worker(cpu);
389 if (!IS_ERR(worker))
390 rcu_assign_pointer(z_erofs_pcpu_workers[cpu], worker);
391 }
392 return 0;
393}
394#else
395static inline void erofs_destroy_percpu_workers(void) {}
396static inline int erofs_init_percpu_workers(void) { return 0; }
397#endif
398
399#if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_EROFS_FS_PCPU_KTHREAD)
400static DEFINE_SPINLOCK(z_erofs_pcpu_worker_lock);
401static enum cpuhp_state erofs_cpuhp_state;
402
403static int erofs_cpu_online(unsigned int cpu)
404{
405 struct kthread_worker *worker, *old;
406
407 worker = erofs_init_percpu_worker(cpu);
408 if (IS_ERR(worker))
409 return PTR_ERR(worker);
410
411 spin_lock(&z_erofs_pcpu_worker_lock);
412 old = rcu_dereference_protected(z_erofs_pcpu_workers[cpu],
413 lockdep_is_held(&z_erofs_pcpu_worker_lock));
414 if (!old)
415 rcu_assign_pointer(z_erofs_pcpu_workers[cpu], worker);
416 spin_unlock(&z_erofs_pcpu_worker_lock);
417 if (old)
418 kthread_destroy_worker(worker);
419 return 0;
420}
421
422static int erofs_cpu_offline(unsigned int cpu)
423{
424 struct kthread_worker *worker;
425
426 spin_lock(&z_erofs_pcpu_worker_lock);
427 worker = rcu_dereference_protected(z_erofs_pcpu_workers[cpu],
428 lockdep_is_held(&z_erofs_pcpu_worker_lock));
429 rcu_assign_pointer(z_erofs_pcpu_workers[cpu], NULL);
430 spin_unlock(&z_erofs_pcpu_worker_lock);
431
432 synchronize_rcu();
433 if (worker)
434 kthread_destroy_worker(worker);
435 return 0;
436}
437
438static int erofs_cpu_hotplug_init(void)
439{
440 int state;
441
442 state = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN,
443 "fs/erofs:online", erofs_cpu_online, erofs_cpu_offline);
444 if (state < 0)
445 return state;
446
447 erofs_cpuhp_state = state;
448 return 0;
449}
450
451static void erofs_cpu_hotplug_destroy(void)
452{
453 if (erofs_cpuhp_state)
454 cpuhp_remove_state_nocalls(erofs_cpuhp_state);
455}
456#else /* !CONFIG_HOTPLUG_CPU || !CONFIG_EROFS_FS_PCPU_KTHREAD */
457static inline int erofs_cpu_hotplug_init(void) { return 0; }
458static inline void erofs_cpu_hotplug_destroy(void) {}
459#endif
460
461void z_erofs_exit_zip_subsystem(void)
462{
463 erofs_cpu_hotplug_destroy();
464 erofs_destroy_percpu_workers();
465 destroy_workqueue(z_erofs_workqueue);
466 z_erofs_destroy_pcluster_pool();
Gao Xiang3883a792018-07-26 20:22:06 +0800467}
468
Gao Xiang0a0b7e62018-10-09 21:43:53 +0800469int __init z_erofs_init_zip_subsystem(void)
Gao Xiang3883a792018-07-26 20:22:06 +0800470{
Gao Xiang9f6cc762021-04-07 12:39:20 +0800471 int err = z_erofs_create_pcluster_pool();
Gao Xiang3883a792018-07-26 20:22:06 +0800472
Gao Xiang9f6cc762021-04-07 12:39:20 +0800473 if (err)
Sandeep Dhavale3fffb582023-02-08 17:33:22 +0800474 goto out_error_pcluster_pool;
475
476 z_erofs_workqueue = alloc_workqueue("erofs_worker",
477 WQ_UNBOUND | WQ_HIGHPRI, num_possible_cpus());
Dan Carpenter8d1b80a72023-02-16 15:13:04 +0300478 if (!z_erofs_workqueue) {
479 err = -ENOMEM;
Sandeep Dhavale3fffb582023-02-08 17:33:22 +0800480 goto out_error_workqueue_init;
Dan Carpenter8d1b80a72023-02-16 15:13:04 +0300481 }
Sandeep Dhavale3fffb582023-02-08 17:33:22 +0800482
483 err = erofs_init_percpu_workers();
Gao Xiang9f6cc762021-04-07 12:39:20 +0800484 if (err)
Sandeep Dhavale3fffb582023-02-08 17:33:22 +0800485 goto out_error_pcpu_worker;
486
487 err = erofs_cpu_hotplug_init();
488 if (err < 0)
489 goto out_error_cpuhp_init;
490 return err;
491
492out_error_cpuhp_init:
493 erofs_destroy_percpu_workers();
494out_error_pcpu_worker:
495 destroy_workqueue(z_erofs_workqueue);
496out_error_workqueue_init:
497 z_erofs_destroy_pcluster_pool();
498out_error_pcluster_pool:
Gao Xiang9f6cc762021-04-07 12:39:20 +0800499 return err;
Gao Xiang3883a792018-07-26 20:22:06 +0800500}
501
Gao Xiangdb166fc2022-07-15 23:41:57 +0800502enum z_erofs_pclustermode {
503 Z_EROFS_PCLUSTER_INFLIGHT,
Gao Xiang3883a792018-07-26 20:22:06 +0800504 /*
Gao Xiangdb166fc2022-07-15 23:41:57 +0800505 * The current pclusters was the tail of an exist chain, in addition
506 * that the previous processed chained pclusters are all decided to
Gao Xiang97e86a82019-07-31 23:57:47 +0800507 * be hooked up to it.
Gao Xiangdb166fc2022-07-15 23:41:57 +0800508 * A new chain will be created for the remaining pclusters which are
509 * not processed yet, so different from Z_EROFS_PCLUSTER_FOLLOWED,
510 * the next pcluster cannot reuse the whole page safely for inplace I/O
511 * in the following scenario:
Gao Xianga1121522019-02-27 13:33:32 +0800512 * ________________________________________________________________
513 * | tail (partial) page | head (partial) page |
Gao Xiangdb166fc2022-07-15 23:41:57 +0800514 * | (belongs to the next pcl) | (belongs to the current pcl) |
515 * |_______PCLUSTER_FOLLOWED______|________PCLUSTER_HOOKED__________|
Gao Xianga1121522019-02-27 13:33:32 +0800516 */
Gao Xiangdb166fc2022-07-15 23:41:57 +0800517 Z_EROFS_PCLUSTER_HOOKED,
Gao Xiang0b964602021-03-22 02:32:27 +0800518 /*
Gao Xiangdb166fc2022-07-15 23:41:57 +0800519 * a weak form of Z_EROFS_PCLUSTER_FOLLOWED, the difference is that it
Gao Xiang0b964602021-03-22 02:32:27 +0800520 * could be dispatched into bypass queue later due to uptodated managed
521 * pages. All related online pages cannot be reused for inplace I/O (or
Gao Xiang387bab82022-07-15 23:41:52 +0800522 * bvpage) since it can be directly decoded without I/O submission.
Gao Xiang0b964602021-03-22 02:32:27 +0800523 */
Gao Xiangdb166fc2022-07-15 23:41:57 +0800524 Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE,
Gao Xianga1121522019-02-27 13:33:32 +0800525 /*
Gao Xiang97e86a82019-07-31 23:57:47 +0800526 * The current collection has been linked with the owned chain, and
527 * could also be linked with the remaining collections, which means
528 * if the processing page is the tail page of the collection, thus
529 * the current collection can safely use the whole page (since
530 * the previous collection is under control) for in-place I/O, as
531 * illustrated below:
Gao Xianga1121522019-02-27 13:33:32 +0800532 * ________________________________________________________________
Gao Xiang97e86a82019-07-31 23:57:47 +0800533 * | tail (partial) page | head (partial) page |
534 * | (of the current cl) | (of the previous collection) |
Gao Xiangdb166fc2022-07-15 23:41:57 +0800535 * | PCLUSTER_FOLLOWED or | |
536 * |_____PCLUSTER_HOOKED__|___________PCLUSTER_FOLLOWED____________|
Gao Xianga1121522019-02-27 13:33:32 +0800537 *
Gao Xiang97e86a82019-07-31 23:57:47 +0800538 * [ (*) the above page can be used as inplace I/O. ]
Gao Xiang3883a792018-07-26 20:22:06 +0800539 */
Gao Xiangdb166fc2022-07-15 23:41:57 +0800540 Z_EROFS_PCLUSTER_FOLLOWED,
Gao Xiang3883a792018-07-26 20:22:06 +0800541};
542
Gao Xiang5c6dcc52022-03-02 03:49:50 +0800543struct z_erofs_decompress_frontend {
544 struct inode *const inode;
545 struct erofs_map_blocks map;
Gao Xiang06a304c2022-07-15 23:41:51 +0800546 struct z_erofs_bvec_iter biter;
Gao Xiang3883a792018-07-26 20:22:06 +0800547
Gao Xiang06a304c2022-07-15 23:41:51 +0800548 struct page *candidate_bvpage;
Gao Xiangbfc4ccb2019-08-21 11:09:08 +0800549 struct z_erofs_pcluster *pcl, *tailpcl;
Gao Xiang97e86a82019-07-31 23:57:47 +0800550 z_erofs_next_pcluster_t owned_head;
Gao Xiangdb166fc2022-07-15 23:41:57 +0800551 enum z_erofs_pclustermode mode;
Gao Xiang97e86a82019-07-31 23:57:47 +0800552
Gao Xiang6ea5aad2020-09-19 15:27:30 +0800553 bool readahead;
Gao Xiang97e86a82019-07-31 23:57:47 +0800554 /* used for applying cache strategy on the fly */
555 bool backmost;
556 erofs_off_t headoffset;
Gao Xianged722fb2022-07-15 23:41:54 +0800557
558 /* a pointer used to pick up inplace I/O pages */
559 unsigned int icur;
Gao Xiang97e86a82019-07-31 23:57:47 +0800560};
561
Gao Xiang97e86a82019-07-31 23:57:47 +0800562#define DECOMPRESS_FRONTEND_INIT(__i) { \
Gao Xiang5c6dcc52022-03-02 03:49:50 +0800563 .inode = __i, .owned_head = Z_EROFS_PCLUSTER_TAIL, \
Gao Xiangdb166fc2022-07-15 23:41:57 +0800564 .mode = Z_EROFS_PCLUSTER_FOLLOWED, .backmost = true }
Gao Xiang97e86a82019-07-31 23:57:47 +0800565
Gao Xiang1282dea2022-12-06 14:03:52 +0800566static bool z_erofs_should_alloc_cache(struct z_erofs_decompress_frontend *fe)
567{
568 unsigned int cachestrategy = EROFS_I_SB(fe->inode)->opt.cache_strategy;
569
570 if (cachestrategy <= EROFS_ZIP_CACHE_DISABLED)
571 return false;
572
573 if (fe->backmost)
574 return true;
575
576 if (cachestrategy >= EROFS_ZIP_CACHE_READAROUND &&
577 fe->map.m_la < fe->headoffset)
578 return true;
579
580 return false;
581}
582
Gao Xiang6f39d1e2022-03-02 03:49:51 +0800583static void z_erofs_bind_cache(struct z_erofs_decompress_frontend *fe,
Gao Xiang6f39d1e2022-03-02 03:49:51 +0800584 struct page **pagepool)
Gao Xiang105d4ad2018-07-26 20:22:07 +0800585{
Gao Xiang6f39d1e2022-03-02 03:49:51 +0800586 struct address_space *mc = MNGD_MAPPING(EROFS_I_SB(fe->inode));
Gao Xiang5c6dcc52022-03-02 03:49:50 +0800587 struct z_erofs_pcluster *pcl = fe->pcl;
Gao Xiang1282dea2022-12-06 14:03:52 +0800588 bool shouldalloc = z_erofs_should_alloc_cache(fe);
Gao Xiang92e6efd2018-12-08 00:19:16 +0800589 bool standalone = true;
Gao Xiang6f39d1e2022-03-02 03:49:51 +0800590 /*
591 * optimistic allocation without direct reclaim since inplace I/O
592 * can be used if low memory otherwise.
593 */
Gao Xiang1825c8d2020-12-09 20:37:17 +0800594 gfp_t gfp = (mapping_gfp_mask(mc) & ~__GFP_DIRECT_RECLAIM) |
595 __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
Gao Xianged722fb2022-07-15 23:41:54 +0800596 unsigned int i;
Gao Xiang105d4ad2018-07-26 20:22:07 +0800597
Gao Xiangdb166fc2022-07-15 23:41:57 +0800598 if (fe->mode < Z_EROFS_PCLUSTER_FOLLOWED)
Gao Xiang92e6efd2018-12-08 00:19:16 +0800599 return;
Gao Xiang105d4ad2018-07-26 20:22:07 +0800600
Gao Xianged722fb2022-07-15 23:41:54 +0800601 for (i = 0; i < pcl->pclusterpages; ++i) {
Gao Xiang92e6efd2018-12-08 00:19:16 +0800602 struct page *page;
Gao Xiangb1ed2202023-02-04 17:30:37 +0800603 void *t; /* mark pages just found for debugging */
Gao Xiang1825c8d2020-12-09 20:37:17 +0800604 struct page *newpage = NULL;
Gao Xiang92e6efd2018-12-08 00:19:16 +0800605
606 /* the compressed page was loaded before */
Gao Xianged722fb2022-07-15 23:41:54 +0800607 if (READ_ONCE(pcl->compressed_bvecs[i].page))
Gao Xiang105d4ad2018-07-26 20:22:07 +0800608 continue;
609
Gao Xianged722fb2022-07-15 23:41:54 +0800610 page = find_get_page(mc, pcl->obj.index + i);
Gao Xiang92e6efd2018-12-08 00:19:16 +0800611
612 if (page) {
Gao Xiangb1ed2202023-02-04 17:30:37 +0800613 t = (void *)((unsigned long)page | 1);
Gao Xiang0b964602021-03-22 02:32:27 +0800614 } else {
615 /* I/O is needed, no possible to decompress directly */
Gao Xiang92e6efd2018-12-08 00:19:16 +0800616 standalone = false;
Gao Xiang1282dea2022-12-06 14:03:52 +0800617 if (!shouldalloc)
Gao Xiang0b964602021-03-22 02:32:27 +0800618 continue;
Gao Xiang1282dea2022-12-06 14:03:52 +0800619
620 /*
621 * try to use cached I/O if page allocation
622 * succeeds or fallback to in-place I/O instead
623 * to avoid any direct reclaim.
624 */
625 newpage = erofs_allocpage(pagepool, gfp);
626 if (!newpage)
627 continue;
628 set_page_private(newpage, Z_EROFS_PREALLOCATED_PAGE);
Gao Xiangb1ed2202023-02-04 17:30:37 +0800629 t = (void *)((unsigned long)newpage | 1);
Gao Xiang105d4ad2018-07-26 20:22:07 +0800630 }
631
Gao Xiangb1ed2202023-02-04 17:30:37 +0800632 if (!cmpxchg_relaxed(&pcl->compressed_bvecs[i].page, NULL, t))
Gao Xiang105d4ad2018-07-26 20:22:07 +0800633 continue;
634
Gao Xiangeaa91722021-10-22 17:01:20 +0800635 if (page)
Gao Xiang92e6efd2018-12-08 00:19:16 +0800636 put_page(page);
Gao Xiangeaa91722021-10-22 17:01:20 +0800637 else if (newpage)
638 erofs_pagepool_add(pagepool, newpage);
Gao Xiang105d4ad2018-07-26 20:22:07 +0800639 }
Gao Xiang92e6efd2018-12-08 00:19:16 +0800640
Gao Xiang0b964602021-03-22 02:32:27 +0800641 /*
642 * don't do inplace I/O if all compressed pages are available in
643 * managed cache since it can be moved to the bypass queue instead.
644 */
645 if (standalone)
Gao Xiangdb166fc2022-07-15 23:41:57 +0800646 fe->mode = Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE;
Gao Xiang105d4ad2018-07-26 20:22:07 +0800647}
648
649/* called by erofs_shrinker to get rid of all compressed_pages */
Gao Xiang47e541a2018-07-29 13:34:58 +0800650int erofs_try_to_free_all_cached_pages(struct erofs_sb_info *sbi,
Gao Xiang97e86a82019-07-31 23:57:47 +0800651 struct erofs_workgroup *grp)
Gao Xiang105d4ad2018-07-26 20:22:07 +0800652{
Gao Xiang97e86a82019-07-31 23:57:47 +0800653 struct z_erofs_pcluster *const pcl =
654 container_of(grp, struct z_erofs_pcluster, obj);
Gao Xiang105d4ad2018-07-26 20:22:07 +0800655 int i;
656
Yue Hucecf8642021-12-29 07:29:19 +0800657 DBG_BUGON(z_erofs_is_inline_pcluster(pcl));
Gao Xiang105d4ad2018-07-26 20:22:07 +0800658 /*
659 * refcount of workgroup is now freezed as 1,
660 * therefore no need to worry about available decompression users.
661 */
Gao Xiang9f6cc762021-04-07 12:39:20 +0800662 for (i = 0; i < pcl->pclusterpages; ++i) {
Gao Xianged722fb2022-07-15 23:41:54 +0800663 struct page *page = pcl->compressed_bvecs[i].page;
Gao Xiang105d4ad2018-07-26 20:22:07 +0800664
Gao Xiang97e86a82019-07-31 23:57:47 +0800665 if (!page)
Gao Xiang105d4ad2018-07-26 20:22:07 +0800666 continue;
667
668 /* block other users from reclaiming or migrating the page */
669 if (!trylock_page(page))
670 return -EBUSY;
671
Yue Huf4d4e5f2021-08-10 14:54:50 +0800672 if (!erofs_page_is_managed(sbi, page))
Gao Xiang97e86a82019-07-31 23:57:47 +0800673 continue;
Gao Xiang105d4ad2018-07-26 20:22:07 +0800674
Gao Xiang97e86a82019-07-31 23:57:47 +0800675 /* barrier is implied in the following 'unlock_page' */
Gao Xianged722fb2022-07-15 23:41:54 +0800676 WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL);
Gao Xiang6aaa7b02020-12-08 17:58:32 +0800677 detach_page_private(page);
Gao Xiang105d4ad2018-07-26 20:22:07 +0800678 unlock_page(page);
Gao Xiang105d4ad2018-07-26 20:22:07 +0800679 }
680 return 0;
681}
682
Yue Hud252ff32021-08-10 15:24:16 +0800683int erofs_try_to_free_cached_page(struct page *page)
Gao Xiang105d4ad2018-07-26 20:22:07 +0800684{
Gao Xiang97e86a82019-07-31 23:57:47 +0800685 struct z_erofs_pcluster *const pcl = (void *)page_private(page);
Gao Xianged722fb2022-07-15 23:41:54 +0800686 int ret, i;
Gao Xiang105d4ad2018-07-26 20:22:07 +0800687
Gao Xianged722fb2022-07-15 23:41:54 +0800688 if (!erofs_workgroup_try_to_freeze(&pcl->obj, 1))
689 return 0;
Gao Xiang105d4ad2018-07-26 20:22:07 +0800690
Gao Xianged722fb2022-07-15 23:41:54 +0800691 ret = 0;
692 DBG_BUGON(z_erofs_is_inline_pcluster(pcl));
693 for (i = 0; i < pcl->pclusterpages; ++i) {
694 if (pcl->compressed_bvecs[i].page == page) {
695 WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL);
696 ret = 1;
697 break;
Gao Xiang105d4ad2018-07-26 20:22:07 +0800698 }
Gao Xiang105d4ad2018-07-26 20:22:07 +0800699 }
Gao Xianged722fb2022-07-15 23:41:54 +0800700 erofs_workgroup_unfreeze(&pcl->obj, 1);
701 if (ret)
702 detach_page_private(page);
Gao Xiang105d4ad2018-07-26 20:22:07 +0800703 return ret;
704}
Gao Xiang105d4ad2018-07-26 20:22:07 +0800705
Gao Xiang5c6dcc52022-03-02 03:49:50 +0800706static bool z_erofs_try_inplace_io(struct z_erofs_decompress_frontend *fe,
Gao Xianged722fb2022-07-15 23:41:54 +0800707 struct z_erofs_bvec *bvec)
Gao Xiang3883a792018-07-26 20:22:06 +0800708{
Gao Xiang5c6dcc52022-03-02 03:49:50 +0800709 struct z_erofs_pcluster *const pcl = fe->pcl;
Gao Xiang97e86a82019-07-31 23:57:47 +0800710
Gao Xianged722fb2022-07-15 23:41:54 +0800711 while (fe->icur > 0) {
712 if (!cmpxchg(&pcl->compressed_bvecs[--fe->icur].page,
713 NULL, bvec->page)) {
714 pcl->compressed_bvecs[fe->icur] = *bvec;
Gao Xiang3883a792018-07-26 20:22:06 +0800715 return true;
Gao Xianged722fb2022-07-15 23:41:54 +0800716 }
717 }
Gao Xiang3883a792018-07-26 20:22:06 +0800718 return false;
719}
720
Gao Xiang87ca34a2022-05-29 13:54:23 +0800721/* callers must be with pcluster lock held */
Gao Xiang5c6dcc52022-03-02 03:49:50 +0800722static int z_erofs_attach_page(struct z_erofs_decompress_frontend *fe,
Gao Xiang5b220b22022-07-15 23:41:56 +0800723 struct z_erofs_bvec *bvec, bool exclusive)
Gao Xiang3883a792018-07-26 20:22:06 +0800724{
725 int ret;
Gao Xiang3883a792018-07-26 20:22:06 +0800726
Gao Xiangdb166fc2022-07-15 23:41:57 +0800727 if (exclusive) {
Gao Xiang06a304c2022-07-15 23:41:51 +0800728 /* give priority for inplaceio to use file pages first */
Gao Xianged722fb2022-07-15 23:41:54 +0800729 if (z_erofs_try_inplace_io(fe, bvec))
Gao Xiang06a304c2022-07-15 23:41:51 +0800730 return 0;
731 /* otherwise, check if it can be used as a bvpage */
Gao Xiangdb166fc2022-07-15 23:41:57 +0800732 if (fe->mode >= Z_EROFS_PCLUSTER_FOLLOWED &&
Gao Xiang06a304c2022-07-15 23:41:51 +0800733 !fe->candidate_bvpage)
734 fe->candidate_bvpage = bvec->page;
735 }
736 ret = z_erofs_bvec_enqueue(&fe->biter, bvec, &fe->candidate_bvpage);
737 fe->pcl->vcnt += (ret >= 0);
738 return ret;
Gao Xiang3883a792018-07-26 20:22:06 +0800739}
740
Gao Xiang5c6dcc52022-03-02 03:49:50 +0800741static void z_erofs_try_to_claim_pcluster(struct z_erofs_decompress_frontend *f)
Gao Xiang3883a792018-07-26 20:22:06 +0800742{
Gao Xiang5c6dcc52022-03-02 03:49:50 +0800743 struct z_erofs_pcluster *pcl = f->pcl;
744 z_erofs_next_pcluster_t *owned_head = &f->owned_head;
Gao Xiang3883a792018-07-26 20:22:06 +0800745
Gao Xiang473e15b2020-12-08 17:58:34 +0800746 /* type 1, nil pcluster (this pcluster doesn't belong to any chain.) */
747 if (cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_NIL,
748 *owned_head) == Z_EROFS_PCLUSTER_NIL) {
Gao Xiang97e86a82019-07-31 23:57:47 +0800749 *owned_head = &pcl->next;
Gao Xiang473e15b2020-12-08 17:58:34 +0800750 /* so we can attach this pcluster to our submission chain. */
Gao Xiangdb166fc2022-07-15 23:41:57 +0800751 f->mode = Z_EROFS_PCLUSTER_FOLLOWED;
Gao Xiang473e15b2020-12-08 17:58:34 +0800752 return;
Gao Xianga1121522019-02-27 13:33:32 +0800753 }
Gao Xiang473e15b2020-12-08 17:58:34 +0800754
755 /*
756 * type 2, link to the end of an existing open chain, be careful
757 * that its submission is controlled by the original attached chain.
758 */
Gao Xiang267f2492022-07-15 23:42:03 +0800759 if (*owned_head != &pcl->next && pcl != f->tailpcl &&
760 cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_TAIL,
Gao Xiang473e15b2020-12-08 17:58:34 +0800761 *owned_head) == Z_EROFS_PCLUSTER_TAIL) {
762 *owned_head = Z_EROFS_PCLUSTER_TAIL;
Gao Xiangdb166fc2022-07-15 23:41:57 +0800763 f->mode = Z_EROFS_PCLUSTER_HOOKED;
Gao Xiang5c6dcc52022-03-02 03:49:50 +0800764 f->tailpcl = NULL;
Gao Xiang473e15b2020-12-08 17:58:34 +0800765 return;
766 }
767 /* type 3, it belongs to a chain, but it isn't the end of the chain */
Gao Xiangdb166fc2022-07-15 23:41:57 +0800768 f->mode = Z_EROFS_PCLUSTER_INFLIGHT;
Gao Xiang3883a792018-07-26 20:22:06 +0800769}
770
Gao Xiang83a386c2022-07-15 23:41:48 +0800771static int z_erofs_register_pcluster(struct z_erofs_decompress_frontend *fe)
Gao Xiang3883a792018-07-26 20:22:06 +0800772{
Gao Xiang83a386c2022-07-15 23:41:48 +0800773 struct erofs_map_blocks *map = &fe->map;
Yue Hucecf8642021-12-29 07:29:19 +0800774 bool ztailpacking = map->m_flags & EROFS_MAP_META;
Gao Xiang97e86a82019-07-31 23:57:47 +0800775 struct z_erofs_pcluster *pcl;
Gao Xiang64094a02020-02-20 10:46:42 +0800776 struct erofs_workgroup *grp;
Gao Xiang97e86a82019-07-31 23:57:47 +0800777 int err;
Gao Xiange5e3abb2018-09-19 13:49:07 +0800778
Chen Zhongjinc42c0ff2022-12-05 11:49:57 +0800779 if (!(map->m_flags & EROFS_MAP_ENCODED) ||
780 (!ztailpacking && !(map->m_pa >> PAGE_SHIFT))) {
Gao Xiang8f899262021-10-09 04:08:37 +0800781 DBG_BUGON(1);
782 return -EFSCORRUPTED;
783 }
784
Gao Xiang9f6cc762021-04-07 12:39:20 +0800785 /* no available pcluster, let's allocate one */
Yue Hucecf8642021-12-29 07:29:19 +0800786 pcl = z_erofs_alloc_pcluster(ztailpacking ? 1 :
787 map->m_plen >> PAGE_SHIFT);
Gao Xiang9f6cc762021-04-07 12:39:20 +0800788 if (IS_ERR(pcl))
789 return PTR_ERR(pcl);
Gao Xiang3883a792018-07-26 20:22:06 +0800790
Gao Xiang64094a02020-02-20 10:46:42 +0800791 atomic_set(&pcl->obj.refcount, 1);
Gao Xiang8f899262021-10-09 04:08:37 +0800792 pcl->algorithmformat = map->m_algorithmformat;
Gao Xiang2bfab9c2022-07-15 23:42:02 +0800793 pcl->length = 0;
794 pcl->partial = true;
Gao Xiang3883a792018-07-26 20:22:06 +0800795
Gao Xiang97e86a82019-07-31 23:57:47 +0800796 /* new pclusters should be claimed as type 1, primary and followed */
Gao Xiang5c6dcc52022-03-02 03:49:50 +0800797 pcl->next = fe->owned_head;
Gao Xiang87ca34a2022-05-29 13:54:23 +0800798 pcl->pageofs_out = map->m_la & ~PAGE_MASK;
Gao Xiangdb166fc2022-07-15 23:41:57 +0800799 fe->mode = Z_EROFS_PCLUSTER_FOLLOWED;
Gao Xiang97e86a82019-07-31 23:57:47 +0800800
Gao Xiang23edf3a2018-11-23 01:21:47 +0800801 /*
802 * lock all primary followed works before visible to others
Gao Xiang97e86a82019-07-31 23:57:47 +0800803 * and mutex_trylock *never* fails for a new pcluster.
Gao Xiang23edf3a2018-11-23 01:21:47 +0800804 */
Gao Xiang87ca34a2022-05-29 13:54:23 +0800805 mutex_init(&pcl->lock);
806 DBG_BUGON(!mutex_trylock(&pcl->lock));
Gao Xiang23edf3a2018-11-23 01:21:47 +0800807
Yue Hucecf8642021-12-29 07:29:19 +0800808 if (ztailpacking) {
809 pcl->obj.index = 0; /* which indicates ztailpacking */
810 pcl->pageofs_in = erofs_blkoff(map->m_pa);
811 pcl->tailpacking_size = map->m_plen;
812 } else {
813 pcl->obj.index = map->m_pa >> PAGE_SHIFT;
Gao Xiang64094a02020-02-20 10:46:42 +0800814
Gao Xiang83a386c2022-07-15 23:41:48 +0800815 grp = erofs_insert_workgroup(fe->inode->i_sb, &pcl->obj);
Yue Hucecf8642021-12-29 07:29:19 +0800816 if (IS_ERR(grp)) {
817 err = PTR_ERR(grp);
818 goto err_out;
819 }
820
821 if (grp != &pcl->obj) {
Gao Xiang5c6dcc52022-03-02 03:49:50 +0800822 fe->pcl = container_of(grp,
Yue Hucecf8642021-12-29 07:29:19 +0800823 struct z_erofs_pcluster, obj);
824 err = -EEXIST;
825 goto err_out;
826 }
Gao Xiang3883a792018-07-26 20:22:06 +0800827 }
Gao Xiangbfc4ccb2019-08-21 11:09:08 +0800828 /* used to check tail merging loop due to corrupted images */
Gao Xiang5c6dcc52022-03-02 03:49:50 +0800829 if (fe->owned_head == Z_EROFS_PCLUSTER_TAIL)
830 fe->tailpcl = pcl;
831 fe->owned_head = &pcl->next;
832 fe->pcl = pcl;
Gao Xiang9e579fc2019-10-08 20:56:12 +0800833 return 0;
Gao Xiang64094a02020-02-20 10:46:42 +0800834
835err_out:
Gao Xiang87ca34a2022-05-29 13:54:23 +0800836 mutex_unlock(&pcl->lock);
Gao Xiang9f6cc762021-04-07 12:39:20 +0800837 z_erofs_free_pcluster(pcl);
Gao Xiang64094a02020-02-20 10:46:42 +0800838 return err;
Gao Xiang3883a792018-07-26 20:22:06 +0800839}
840
Gao Xiang83a386c2022-07-15 23:41:48 +0800841static int z_erofs_collector_begin(struct z_erofs_decompress_frontend *fe)
Gao Xiang3883a792018-07-26 20:22:06 +0800842{
Gao Xiang83a386c2022-07-15 23:41:48 +0800843 struct erofs_map_blocks *map = &fe->map;
Gao Xiang0d823b42022-07-15 23:41:49 +0800844 struct erofs_workgroup *grp = NULL;
Gao Xiang9e579fc2019-10-08 20:56:12 +0800845 int ret;
Gao Xiang3883a792018-07-26 20:22:06 +0800846
Gao Xiang87ca34a2022-05-29 13:54:23 +0800847 DBG_BUGON(fe->pcl);
Gao Xiang3883a792018-07-26 20:22:06 +0800848
Gao Xiang87ca34a2022-05-29 13:54:23 +0800849 /* must be Z_EROFS_PCLUSTER_TAIL or pointed to previous pcluster */
Gao Xiang5c6dcc52022-03-02 03:49:50 +0800850 DBG_BUGON(fe->owned_head == Z_EROFS_PCLUSTER_NIL);
851 DBG_BUGON(fe->owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED);
Gao Xiang3883a792018-07-26 20:22:06 +0800852
Gao Xiang0d823b42022-07-15 23:41:49 +0800853 if (!(map->m_flags & EROFS_MAP_META)) {
854 grp = erofs_find_workgroup(fe->inode->i_sb,
855 map->m_pa >> PAGE_SHIFT);
856 } else if ((map->m_pa & ~PAGE_MASK) + map->m_plen > PAGE_SIZE) {
857 DBG_BUGON(1);
858 return -EFSCORRUPTED;
Gao Xiang97e86a82019-07-31 23:57:47 +0800859 }
Gao Xiang3883a792018-07-26 20:22:06 +0800860
Gao Xiang64094a02020-02-20 10:46:42 +0800861 if (grp) {
Gao Xiang5c6dcc52022-03-02 03:49:50 +0800862 fe->pcl = container_of(grp, struct z_erofs_pcluster, obj);
Gao Xiang0d823b42022-07-15 23:41:49 +0800863 ret = -EEXIST;
Gao Xiang64094a02020-02-20 10:46:42 +0800864 } else {
Gao Xiang83a386c2022-07-15 23:41:48 +0800865 ret = z_erofs_register_pcluster(fe);
Gao Xiang3883a792018-07-26 20:22:06 +0800866 }
867
Gao Xiang0d823b42022-07-15 23:41:49 +0800868 if (ret == -EEXIST) {
Gao Xiang267f2492022-07-15 23:42:03 +0800869 mutex_lock(&fe->pcl->lock);
870 /* used to check tail merging loop due to corrupted images */
871 if (fe->owned_head == Z_EROFS_PCLUSTER_TAIL)
872 fe->tailpcl = fe->pcl;
873
874 z_erofs_try_to_claim_pcluster(fe);
Gao Xiang0d823b42022-07-15 23:41:49 +0800875 } else if (ret) {
Gao Xiang9e579fc2019-10-08 20:56:12 +0800876 return ret;
Gao Xiang64094a02020-02-20 10:46:42 +0800877 }
Gao Xiang06a304c2022-07-15 23:41:51 +0800878 z_erofs_bvec_iter_begin(&fe->biter, &fe->pcl->bvset,
Gao Xiang387bab82022-07-15 23:41:52 +0800879 Z_EROFS_INLINE_BVECS, fe->pcl->vcnt);
Gao Xiang81382f52021-04-07 12:39:21 +0800880 /* since file-backed online pages are traversed in reverse order */
Gao Xianged722fb2022-07-15 23:41:54 +0800881 fe->icur = z_erofs_pclusterpages(fe->pcl);
Gao Xiang3883a792018-07-26 20:22:06 +0800882 return 0;
883}
884
885/*
Gao Xiang97e86a82019-07-31 23:57:47 +0800886 * keep in mind that no referenced pclusters will be freed
887 * only after a RCU grace period.
Gao Xiang3883a792018-07-26 20:22:06 +0800888 */
889static void z_erofs_rcu_callback(struct rcu_head *head)
890{
Gao Xiang87ca34a2022-05-29 13:54:23 +0800891 z_erofs_free_pcluster(container_of(head,
892 struct z_erofs_pcluster, rcu));
Gao Xiang3883a792018-07-26 20:22:06 +0800893}
894
895void erofs_workgroup_free_rcu(struct erofs_workgroup *grp)
896{
Gao Xiang97e86a82019-07-31 23:57:47 +0800897 struct z_erofs_pcluster *const pcl =
898 container_of(grp, struct z_erofs_pcluster, obj);
Gao Xiang3883a792018-07-26 20:22:06 +0800899
Gao Xiang87ca34a2022-05-29 13:54:23 +0800900 call_rcu(&pcl->rcu, z_erofs_rcu_callback);
Gao Xiang3883a792018-07-26 20:22:06 +0800901}
902
Gao Xiang5c6dcc52022-03-02 03:49:50 +0800903static bool z_erofs_collector_end(struct z_erofs_decompress_frontend *fe)
Gao Xiang3883a792018-07-26 20:22:06 +0800904{
Gao Xiang87ca34a2022-05-29 13:54:23 +0800905 struct z_erofs_pcluster *pcl = fe->pcl;
Gao Xiang3883a792018-07-26 20:22:06 +0800906
Gao Xiang87ca34a2022-05-29 13:54:23 +0800907 if (!pcl)
Gao Xiang3883a792018-07-26 20:22:06 +0800908 return false;
909
Gao Xiang06a304c2022-07-15 23:41:51 +0800910 z_erofs_bvec_iter_end(&fe->biter);
Gao Xiang87ca34a2022-05-29 13:54:23 +0800911 mutex_unlock(&pcl->lock);
Gao Xiang3883a792018-07-26 20:22:06 +0800912
Gao Xiang06a304c2022-07-15 23:41:51 +0800913 if (fe->candidate_bvpage) {
914 DBG_BUGON(z_erofs_is_shortlived_page(fe->candidate_bvpage));
915 fe->candidate_bvpage = NULL;
916 }
917
Gao Xiang3883a792018-07-26 20:22:06 +0800918 /*
Gao Xiang97e86a82019-07-31 23:57:47 +0800919 * if all pending pages are added, don't hold its reference
920 * any longer if the pcluster isn't hosted by ourselves.
Gao Xiang3883a792018-07-26 20:22:06 +0800921 */
Gao Xiangdb166fc2022-07-15 23:41:57 +0800922 if (fe->mode < Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE)
Gao Xiang87ca34a2022-05-29 13:54:23 +0800923 erofs_workgroup_put(&pcl->obj);
Gao Xiang3883a792018-07-26 20:22:06 +0800924
Gao Xiang87ca34a2022-05-29 13:54:23 +0800925 fe->pcl = NULL;
Gao Xiang3883a792018-07-26 20:22:06 +0800926 return true;
927}
928
Yue Hub15b2e32022-09-23 10:11:22 +0800929static int z_erofs_read_fragment(struct inode *inode, erofs_off_t pos,
930 struct page *page, unsigned int pageofs,
931 unsigned int len)
932{
933 struct inode *packed_inode = EROFS_I_SB(inode)->packed_inode;
934 struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
935 u8 *src, *dst;
936 unsigned int i, cnt;
937
Yue Hue5126de2022-10-21 16:53:25 +0800938 if (!packed_inode)
939 return -EFSCORRUPTED;
940
Yue Hub15b2e32022-09-23 10:11:22 +0800941 pos += EROFS_I(inode)->z_fragmentoff;
942 for (i = 0; i < len; i += cnt) {
943 cnt = min_t(unsigned int, len - i,
944 EROFS_BLKSIZ - erofs_blkoff(pos));
945 src = erofs_bread(&buf, packed_inode,
946 erofs_blknr(pos), EROFS_KMAP);
947 if (IS_ERR(src)) {
948 erofs_put_metabuf(&buf);
949 return PTR_ERR(src);
950 }
951
952 dst = kmap_local_page(page);
953 memcpy(dst + pageofs + i, src + erofs_blkoff(pos), cnt);
954 kunmap_local(dst);
955 pos += cnt;
956 }
957 erofs_put_metabuf(&buf);
958 return 0;
959}
960
Gao Xiang97e86a82019-07-31 23:57:47 +0800961static int z_erofs_do_read_page(struct z_erofs_decompress_frontend *fe,
Gao Xiangeaa91722021-10-22 17:01:20 +0800962 struct page *page, struct page **pagepool)
Gao Xiang3883a792018-07-26 20:22:06 +0800963{
Gao Xiang97e86a82019-07-31 23:57:47 +0800964 struct inode *const inode = fe->inode;
Chao Yu3b423412019-01-15 09:42:21 +0800965 struct erofs_map_blocks *const map = &fe->map;
Gao Xiang3883a792018-07-26 20:22:06 +0800966 const loff_t offset = page_offset(page);
Gao Xiang5b220b22022-07-15 23:41:56 +0800967 bool tight = true, exclusive;
Gao Xiang2bfab9c2022-07-15 23:42:02 +0800968 unsigned int cur, end, spiltted;
Gao Xiang1e05ff32018-09-18 22:27:25 +0800969 int err = 0;
Gao Xiang3883a792018-07-26 20:22:06 +0800970
971 /* register locked file pages as online pages in pack */
972 z_erofs_onlinepage_init(page);
973
974 spiltted = 0;
975 end = PAGE_SIZE;
976repeat:
977 cur = end - 1;
978
Gao Xiang39397a42022-05-29 13:54:24 +0800979 if (offset + cur < map->m_la ||
980 offset + cur >= map->m_la + map->m_llen) {
981 erofs_dbg("out-of-range map @ pos %llu", offset + cur);
982
983 if (z_erofs_collector_end(fe))
984 fe->backmost = false;
985 map->m_la = offset + cur;
986 map->m_llen = 0;
987 err = z_erofs_map_blocks_iter(inode, map, 0);
988 if (err)
Gao Xiang67148552022-07-15 23:41:55 +0800989 goto out;
Gao Xiang39397a42022-05-29 13:54:24 +0800990 } else {
991 if (fe->pcl)
992 goto hitted;
Gao Xiang87ca34a2022-05-29 13:54:23 +0800993 /* didn't get a valid pcluster previously (very rare) */
Gao Xiang1e5ceea2019-02-27 13:33:31 +0800994 }
Gao Xiang3883a792018-07-26 20:22:06 +0800995
Yue Hub15b2e32022-09-23 10:11:22 +0800996 if (!(map->m_flags & EROFS_MAP_MAPPED) ||
997 map->m_flags & EROFS_MAP_FRAGMENT)
Gao Xiang3883a792018-07-26 20:22:06 +0800998 goto hitted;
999
Gao Xiang83a386c2022-07-15 23:41:48 +08001000 err = z_erofs_collector_begin(fe);
Gao Xiang8d8a09b2019-08-30 00:38:27 +08001001 if (err)
Gao Xiang67148552022-07-15 23:41:55 +08001002 goto out;
Gao Xiang3883a792018-07-26 20:22:06 +08001003
Gao Xiang5c6dcc52022-03-02 03:49:50 +08001004 if (z_erofs_is_inline_pcluster(fe->pcl)) {
Gao Xiang09c54372022-01-02 12:00:17 +08001005 void *mp;
Gao Xiang105d4ad2018-07-26 20:22:07 +08001006
Gao Xiang09c54372022-01-02 12:00:17 +08001007 mp = erofs_read_metabuf(&fe->map.buf, inode->i_sb,
1008 erofs_blknr(map->m_pa), EROFS_NO_KMAP);
1009 if (IS_ERR(mp)) {
1010 err = PTR_ERR(mp);
Yue Hucecf8642021-12-29 07:29:19 +08001011 erofs_err(inode->i_sb,
1012 "failed to get inline page, err %d", err);
Gao Xiang67148552022-07-15 23:41:55 +08001013 goto out;
Yue Hucecf8642021-12-29 07:29:19 +08001014 }
Gao Xiang09c54372022-01-02 12:00:17 +08001015 get_page(fe->map.buf.page);
Gao Xianged722fb2022-07-15 23:41:54 +08001016 WRITE_ONCE(fe->pcl->compressed_bvecs[0].page,
1017 fe->map.buf.page);
Gao Xiangdb166fc2022-07-15 23:41:57 +08001018 fe->mode = Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE;
Yue Hucecf8642021-12-29 07:29:19 +08001019 } else {
Gao Xiang6f39d1e2022-03-02 03:49:51 +08001020 /* bind cache first when cached decompression is preferred */
Gao Xiang1282dea2022-12-06 14:03:52 +08001021 z_erofs_bind_cache(fe, pagepool);
Yue Hucecf8642021-12-29 07:29:19 +08001022 }
Gao Xiang3883a792018-07-26 20:22:06 +08001023hitted:
Gao Xiangdc76ea82019-09-22 18:04:34 +08001024 /*
1025 * Ensure the current partial page belongs to this submit chain rather
1026 * than other concurrent submit chains or the noio(bypass) chain since
1027 * those chains are handled asynchronously thus the page cannot be used
Gao Xiang387bab82022-07-15 23:41:52 +08001028 * for inplace I/O or bvpage (should be processed in a strict order.)
Gao Xiangdc76ea82019-09-22 18:04:34 +08001029 */
Gao Xiangdb166fc2022-07-15 23:41:57 +08001030 tight &= (fe->mode >= Z_EROFS_PCLUSTER_HOOKED &&
1031 fe->mode != Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE);
Gao Xiangdc76ea82019-09-22 18:04:34 +08001032
Thomas Weißschuh7dd68b12018-09-10 21:41:14 +02001033 cur = end - min_t(unsigned int, offset + end - map->m_la, end);
Gao Xiang8d8a09b2019-08-30 00:38:27 +08001034 if (!(map->m_flags & EROFS_MAP_MAPPED)) {
Gao Xiang3883a792018-07-26 20:22:06 +08001035 zero_user_segment(page, cur, end);
1036 goto next_part;
1037 }
Yue Hub15b2e32022-09-23 10:11:22 +08001038 if (map->m_flags & EROFS_MAP_FRAGMENT) {
1039 unsigned int pageofs, skip, len;
1040
1041 if (offset > map->m_la) {
1042 pageofs = 0;
1043 skip = offset - map->m_la;
1044 } else {
1045 pageofs = map->m_la & ~PAGE_MASK;
1046 skip = 0;
1047 }
1048 len = min_t(unsigned int, map->m_llen - skip, end - cur);
1049 err = z_erofs_read_fragment(inode, skip, page, pageofs, len);
1050 if (err)
1051 goto out;
1052 ++spiltted;
1053 tight = false;
1054 goto next_part;
1055 }
Gao Xiang3883a792018-07-26 20:22:06 +08001056
Gao Xiang5b220b22022-07-15 23:41:56 +08001057 exclusive = (!cur && (!spiltted || tight));
Gao Xianga1121522019-02-27 13:33:32 +08001058 if (cur)
Gao Xiangdb166fc2022-07-15 23:41:57 +08001059 tight &= (fe->mode >= Z_EROFS_PCLUSTER_FOLLOWED);
Gao Xianga1121522019-02-27 13:33:32 +08001060
Gao Xiang3883a792018-07-26 20:22:06 +08001061retry:
Gao Xiang06a304c2022-07-15 23:41:51 +08001062 err = z_erofs_attach_page(fe, &((struct z_erofs_bvec) {
1063 .page = page,
1064 .offset = offset - map->m_la,
1065 .end = end,
Gao Xiang5b220b22022-07-15 23:41:56 +08001066 }), exclusive);
Gao Xiang06a304c2022-07-15 23:41:51 +08001067 /* should allocate an additional short-lived page for bvset */
1068 if (err == -EAGAIN && !fe->candidate_bvpage) {
1069 fe->candidate_bvpage = alloc_page(GFP_NOFS | __GFP_NOFAIL);
1070 set_page_private(fe->candidate_bvpage,
1071 Z_EROFS_SHORTLIVED_PAGE);
1072 goto retry;
Gao Xiang3883a792018-07-26 20:22:06 +08001073 }
1074
Gao Xiang06a304c2022-07-15 23:41:51 +08001075 if (err) {
1076 DBG_BUGON(err == -EAGAIN && fe->candidate_bvpage);
Gao Xiang67148552022-07-15 23:41:55 +08001077 goto out;
Gao Xiang06a304c2022-07-15 23:41:51 +08001078 }
Gao Xiang3883a792018-07-26 20:22:06 +08001079
Gao Xiang67148552022-07-15 23:41:55 +08001080 z_erofs_onlinepage_split(page);
Gao Xiang1e05ff32018-09-18 22:27:25 +08001081 /* bump up the number of spiltted parts of a page */
1082 ++spiltted;
Gao Xiang267f2492022-07-15 23:42:03 +08001083 if (fe->pcl->pageofs_out != (map->m_la & ~PAGE_MASK))
1084 fe->pcl->multibases = true;
Gao Xiang2bfab9c2022-07-15 23:42:02 +08001085 if (fe->pcl->length < offset + end - map->m_la) {
1086 fe->pcl->length = offset + end - map->m_la;
1087 fe->pcl->pageofs_out = map->m_la & ~PAGE_MASK;
1088 }
Gao Xiange7933272022-10-14 14:49:15 +08001089 if ((map->m_flags & EROFS_MAP_FULL_MAPPED) &&
1090 !(map->m_flags & EROFS_MAP_PARTIAL_REF) &&
1091 fe->pcl->length == map->m_llen)
1092 fe->pcl->partial = false;
Gao Xiang3883a792018-07-26 20:22:06 +08001093next_part:
Gao Xiang2bfab9c2022-07-15 23:42:02 +08001094 /* shorten the remaining extent to update progress */
Gao Xiang3883a792018-07-26 20:22:06 +08001095 map->m_llen = offset + cur - map->m_la;
Gao Xiang2bfab9c2022-07-15 23:42:02 +08001096 map->m_flags &= ~EROFS_MAP_FULL_MAPPED;
Gao Xiang3883a792018-07-26 20:22:06 +08001097
Kristaps ÄŒivkulis2bc75962018-08-05 18:21:01 +03001098 end = cur;
1099 if (end > 0)
Gao Xiang3883a792018-07-26 20:22:06 +08001100 goto repeat;
1101
Gao Xiang1e05ff32018-09-18 22:27:25 +08001102out:
Gao Xiang67148552022-07-15 23:41:55 +08001103 if (err)
1104 z_erofs_page_mark_eio(page);
Gao Xiang3883a792018-07-26 20:22:06 +08001105 z_erofs_onlinepage_endio(page);
1106
Gao Xiang4f761fa2019-09-04 10:09:09 +08001107 erofs_dbg("%s, finish page: %pK spiltted: %u map->m_llen %llu",
1108 __func__, page, spiltted, map->m_llen);
Gao Xiang3883a792018-07-26 20:22:06 +08001109 return err;
1110}
1111
Huang Jianan40452ff2021-12-06 22:35:52 +08001112static bool z_erofs_get_sync_decompress_policy(struct erofs_sb_info *sbi,
1113 unsigned int readahead_pages)
1114{
Matthew Wilcox (Oracle)a2e20a22022-04-29 11:12:16 -04001115 /* auto: enable for read_folio, disable for readahead */
Huang Jianan40452ff2021-12-06 22:35:52 +08001116 if ((sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_AUTO) &&
1117 !readahead_pages)
1118 return true;
1119
1120 if ((sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_FORCE_ON) &&
1121 (readahead_pages <= sbi->opt.max_sync_decompress_pages))
1122 return true;
1123
1124 return false;
1125}
1126
Gao Xiang6aaa7b02020-12-08 17:58:32 +08001127static bool z_erofs_page_is_invalidated(struct page *page)
1128{
1129 return !page->mapping && !z_erofs_is_shortlived_page(page);
1130}
1131
Gao Xiang4f056872022-07-15 23:41:59 +08001132struct z_erofs_decompress_backend {
1133 struct page *onstack_pages[Z_EROFS_ONSTACK_PAGES];
1134 struct super_block *sb;
1135 struct z_erofs_pcluster *pcl;
1136
1137 /* pages with the longest decompressed length for deduplication */
1138 struct page **decompressed_pages;
1139 /* pages to keep the compressed data */
1140 struct page **compressed_pages;
1141
Gao Xiang267f2492022-07-15 23:42:03 +08001142 struct list_head decompressed_secondary_bvecs;
Gao Xiang4f056872022-07-15 23:41:59 +08001143 struct page **pagepool;
Gao Xiang2bfab9c2022-07-15 23:42:02 +08001144 unsigned int onstack_used, nr_pages;
Gao Xiang4f056872022-07-15 23:41:59 +08001145};
1146
Gao Xiang267f2492022-07-15 23:42:03 +08001147struct z_erofs_bvec_item {
1148 struct z_erofs_bvec bvec;
1149 struct list_head list;
1150};
1151
1152static void z_erofs_do_decompressed_bvec(struct z_erofs_decompress_backend *be,
1153 struct z_erofs_bvec *bvec)
Gao Xiang3fe96ee2022-07-15 23:42:01 +08001154{
Gao Xiang267f2492022-07-15 23:42:03 +08001155 struct z_erofs_bvec_item *item;
Gao Xiang3fe96ee2022-07-15 23:42:01 +08001156
Gao Xiang267f2492022-07-15 23:42:03 +08001157 if (!((bvec->offset + be->pcl->pageofs_out) & ~PAGE_MASK)) {
1158 unsigned int pgnr;
Gao Xiang3fe96ee2022-07-15 23:42:01 +08001159
Gao Xiang267f2492022-07-15 23:42:03 +08001160 pgnr = (bvec->offset + be->pcl->pageofs_out) >> PAGE_SHIFT;
1161 DBG_BUGON(pgnr >= be->nr_pages);
Gao Xiang63bbb852022-10-12 12:50:56 +08001162 if (!be->decompressed_pages[pgnr]) {
1163 be->decompressed_pages[pgnr] = bvec->page;
Gao Xiang267f2492022-07-15 23:42:03 +08001164 return;
Gao Xiang63bbb852022-10-12 12:50:56 +08001165 }
Gao Xiang3fe96ee2022-07-15 23:42:01 +08001166 }
Gao Xiang267f2492022-07-15 23:42:03 +08001167
1168 /* (cold path) one pcluster is requested multiple times */
1169 item = kmalloc(sizeof(*item), GFP_KERNEL | __GFP_NOFAIL);
1170 item->bvec = *bvec;
1171 list_add(&item->list, &be->decompressed_secondary_bvecs);
Gao Xiang3fe96ee2022-07-15 23:42:01 +08001172}
1173
Gao Xiang267f2492022-07-15 23:42:03 +08001174static void z_erofs_fill_other_copies(struct z_erofs_decompress_backend *be,
1175 int err)
1176{
1177 unsigned int off0 = be->pcl->pageofs_out;
1178 struct list_head *p, *n;
1179
1180 list_for_each_safe(p, n, &be->decompressed_secondary_bvecs) {
1181 struct z_erofs_bvec_item *bvi;
1182 unsigned int end, cur;
1183 void *dst, *src;
1184
1185 bvi = container_of(p, struct z_erofs_bvec_item, list);
1186 cur = bvi->bvec.offset < 0 ? -bvi->bvec.offset : 0;
1187 end = min_t(unsigned int, be->pcl->length - bvi->bvec.offset,
1188 bvi->bvec.end);
1189 dst = kmap_local_page(bvi->bvec.page);
1190 while (cur < end) {
1191 unsigned int pgnr, scur, len;
1192
1193 pgnr = (bvi->bvec.offset + cur + off0) >> PAGE_SHIFT;
1194 DBG_BUGON(pgnr >= be->nr_pages);
1195
1196 scur = bvi->bvec.offset + cur -
1197 ((pgnr << PAGE_SHIFT) - off0);
1198 len = min_t(unsigned int, end - cur, PAGE_SIZE - scur);
1199 if (!be->decompressed_pages[pgnr]) {
1200 err = -EFSCORRUPTED;
1201 cur += len;
1202 continue;
1203 }
1204 src = kmap_local_page(be->decompressed_pages[pgnr]);
1205 memcpy(dst + cur, src + scur, len);
1206 kunmap_local(src);
1207 cur += len;
1208 }
1209 kunmap_local(dst);
1210 if (err)
1211 z_erofs_page_mark_eio(bvi->bvec.page);
1212 z_erofs_onlinepage_endio(bvi->bvec.page);
1213 list_del(p);
1214 kfree(bvi);
1215 }
1216}
1217
1218static void z_erofs_parse_out_bvecs(struct z_erofs_decompress_backend *be)
Gao Xiang42fec232022-07-15 23:41:50 +08001219{
Gao Xiang4f056872022-07-15 23:41:59 +08001220 struct z_erofs_pcluster *pcl = be->pcl;
Gao Xiang06a304c2022-07-15 23:41:51 +08001221 struct z_erofs_bvec_iter biter;
1222 struct page *old_bvpage;
Gao Xiang267f2492022-07-15 23:42:03 +08001223 int i;
Gao Xiang42fec232022-07-15 23:41:50 +08001224
Gao Xiang387bab82022-07-15 23:41:52 +08001225 z_erofs_bvec_iter_begin(&biter, &pcl->bvset, Z_EROFS_INLINE_BVECS, 0);
Gao Xiang42fec232022-07-15 23:41:50 +08001226 for (i = 0; i < pcl->vcnt; ++i) {
Gao Xiang06a304c2022-07-15 23:41:51 +08001227 struct z_erofs_bvec bvec;
Gao Xiang42fec232022-07-15 23:41:50 +08001228
Gao Xiang06a304c2022-07-15 23:41:51 +08001229 z_erofs_bvec_dequeue(&biter, &bvec, &old_bvpage);
Gao Xiang42fec232022-07-15 23:41:50 +08001230
Gao Xiang06a304c2022-07-15 23:41:51 +08001231 if (old_bvpage)
Gao Xiang4f056872022-07-15 23:41:59 +08001232 z_erofs_put_shortlivedpage(be->pagepool, old_bvpage);
Gao Xiang42fec232022-07-15 23:41:50 +08001233
Gao Xiang06a304c2022-07-15 23:41:51 +08001234 DBG_BUGON(z_erofs_page_is_invalidated(bvec.page));
Gao Xiang267f2492022-07-15 23:42:03 +08001235 z_erofs_do_decompressed_bvec(be, &bvec);
Gao Xiang42fec232022-07-15 23:41:50 +08001236 }
Gao Xiang06a304c2022-07-15 23:41:51 +08001237
1238 old_bvpage = z_erofs_bvec_iter_end(&biter);
1239 if (old_bvpage)
Gao Xiang4f056872022-07-15 23:41:59 +08001240 z_erofs_put_shortlivedpage(be->pagepool, old_bvpage);
Gao Xiang42fec232022-07-15 23:41:50 +08001241}
1242
Gao Xiang4f056872022-07-15 23:41:59 +08001243static int z_erofs_parse_in_bvecs(struct z_erofs_decompress_backend *be,
1244 bool *overlapped)
Gao Xiang67139e32022-07-15 23:41:53 +08001245{
Gao Xiang4f056872022-07-15 23:41:59 +08001246 struct z_erofs_pcluster *pcl = be->pcl;
Gao Xiang67139e32022-07-15 23:41:53 +08001247 unsigned int pclusterpages = z_erofs_pclusterpages(pcl);
Gao Xiang67139e32022-07-15 23:41:53 +08001248 int i, err = 0;
1249
Gao Xiang67139e32022-07-15 23:41:53 +08001250 *overlapped = false;
Gao Xiang67139e32022-07-15 23:41:53 +08001251 for (i = 0; i < pclusterpages; ++i) {
Gao Xianged722fb2022-07-15 23:41:54 +08001252 struct z_erofs_bvec *bvec = &pcl->compressed_bvecs[i];
1253 struct page *page = bvec->page;
Gao Xiang67139e32022-07-15 23:41:53 +08001254
1255 /* compressed pages ought to be present before decompressing */
1256 if (!page) {
1257 DBG_BUGON(1);
1258 continue;
1259 }
Gao Xiangfe3e5912022-07-15 23:42:00 +08001260 be->compressed_pages[i] = page;
Gao Xiang67139e32022-07-15 23:41:53 +08001261
1262 if (z_erofs_is_inline_pcluster(pcl)) {
1263 if (!PageUptodate(page))
1264 err = -EIO;
1265 continue;
1266 }
1267
1268 DBG_BUGON(z_erofs_page_is_invalidated(page));
1269 if (!z_erofs_is_shortlived_page(page)) {
Gao Xiang4f056872022-07-15 23:41:59 +08001270 if (erofs_page_is_managed(EROFS_SB(be->sb), page)) {
Gao Xiang67139e32022-07-15 23:41:53 +08001271 if (!PageUptodate(page))
1272 err = -EIO;
1273 continue;
1274 }
Gao Xiang267f2492022-07-15 23:42:03 +08001275 z_erofs_do_decompressed_bvec(be, bvec);
Gao Xiang67139e32022-07-15 23:41:53 +08001276 *overlapped = true;
1277 }
Gao Xiang67139e32022-07-15 23:41:53 +08001278 }
1279
Gao Xiangfe3e5912022-07-15 23:42:00 +08001280 if (err)
Gao Xiang4f056872022-07-15 23:41:59 +08001281 return err;
Gao Xiang4f056872022-07-15 23:41:59 +08001282 return 0;
Gao Xiang67139e32022-07-15 23:41:53 +08001283}
1284
Gao Xiang4f056872022-07-15 23:41:59 +08001285static int z_erofs_decompress_pcluster(struct z_erofs_decompress_backend *be,
1286 int err)
Gao Xiang3883a792018-07-26 20:22:06 +08001287{
Gao Xiang4f056872022-07-15 23:41:59 +08001288 struct erofs_sb_info *const sbi = EROFS_SB(be->sb);
1289 struct z_erofs_pcluster *pcl = be->pcl;
Yue Hucecf8642021-12-29 07:29:19 +08001290 unsigned int pclusterpages = z_erofs_pclusterpages(pcl);
Gao Xiang2bfab9c2022-07-15 23:42:02 +08001291 unsigned int i, inputsize;
Gao Xiang67148552022-07-15 23:41:55 +08001292 int err2;
Gao Xiang2bfab9c2022-07-15 23:42:02 +08001293 struct page *page;
1294 bool overlapped;
Gao Xiang3883a792018-07-26 20:22:06 +08001295
Gao Xiang87ca34a2022-05-29 13:54:23 +08001296 mutex_lock(&pcl->lock);
Gao Xiang2bfab9c2022-07-15 23:42:02 +08001297 be->nr_pages = PAGE_ALIGN(pcl->length + pcl->pageofs_out) >> PAGE_SHIFT;
Gao Xiang3883a792018-07-26 20:22:06 +08001298
Gao Xiangfe3e5912022-07-15 23:42:00 +08001299 /* allocate (de)compressed page arrays if cannot be kept on stack */
1300 be->decompressed_pages = NULL;
1301 be->compressed_pages = NULL;
1302 be->onstack_used = 0;
Gao Xiang2bfab9c2022-07-15 23:42:02 +08001303 if (be->nr_pages <= Z_EROFS_ONSTACK_PAGES) {
Gao Xiang4f056872022-07-15 23:41:59 +08001304 be->decompressed_pages = be->onstack_pages;
Gao Xiang2bfab9c2022-07-15 23:42:02 +08001305 be->onstack_used = be->nr_pages;
Gao Xiang4f056872022-07-15 23:41:59 +08001306 memset(be->decompressed_pages, 0,
Gao Xiang2bfab9c2022-07-15 23:42:02 +08001307 sizeof(struct page *) * be->nr_pages);
Gao Xiangfe3e5912022-07-15 23:42:00 +08001308 }
1309
1310 if (pclusterpages + be->onstack_used <= Z_EROFS_ONSTACK_PAGES)
1311 be->compressed_pages = be->onstack_pages + be->onstack_used;
1312
1313 if (!be->decompressed_pages)
Gao Xiang4f056872022-07-15 23:41:59 +08001314 be->decompressed_pages =
Gao Xiang12724ba2023-01-10 15:49:27 +08001315 kcalloc(be->nr_pages, sizeof(struct page *),
1316 GFP_KERNEL | __GFP_NOFAIL);
Gao Xiangfe3e5912022-07-15 23:42:00 +08001317 if (!be->compressed_pages)
1318 be->compressed_pages =
Gao Xiang12724ba2023-01-10 15:49:27 +08001319 kcalloc(pclusterpages, sizeof(struct page *),
1320 GFP_KERNEL | __GFP_NOFAIL);
Gao Xiang3883a792018-07-26 20:22:06 +08001321
Gao Xiang267f2492022-07-15 23:42:03 +08001322 z_erofs_parse_out_bvecs(be);
Gao Xiang4f056872022-07-15 23:41:59 +08001323 err2 = z_erofs_parse_in_bvecs(be, &overlapped);
1324 if (err2)
1325 err = err2;
Gao Xiang8d8a09b2019-08-30 00:38:27 +08001326 if (err)
Gao Xiang11152492019-03-25 11:40:07 +08001327 goto out;
1328
Yue Hucecf8642021-12-29 07:29:19 +08001329 if (z_erofs_is_inline_pcluster(pcl))
1330 inputsize = pcl->tailpacking_size;
1331 else
1332 inputsize = pclusterpages * PAGE_SIZE;
1333
Gao Xiang88aaf5a2019-06-24 15:22:57 +08001334 err = z_erofs_decompress(&(struct z_erofs_decompress_req) {
Gao Xiang4f056872022-07-15 23:41:59 +08001335 .sb = be->sb,
1336 .in = be->compressed_pages,
1337 .out = be->decompressed_pages,
Yue Hucecf8642021-12-29 07:29:19 +08001338 .pageofs_in = pcl->pageofs_in,
Gao Xiang87ca34a2022-05-29 13:54:23 +08001339 .pageofs_out = pcl->pageofs_out,
Gao Xiang9f6cc762021-04-07 12:39:20 +08001340 .inputsize = inputsize,
Gao Xiang2bfab9c2022-07-15 23:42:02 +08001341 .outputsize = pcl->length,
Gao Xiang97e86a82019-07-31 23:57:47 +08001342 .alg = pcl->algorithmformat,
Gao Xiang88aaf5a2019-06-24 15:22:57 +08001343 .inplace_io = overlapped,
Gao Xiang2bfab9c2022-07-15 23:42:02 +08001344 .partial_decoding = pcl->partial,
Gao Xiang267f2492022-07-15 23:42:03 +08001345 .fillgaps = pcl->multibases,
Gao Xiang4f056872022-07-15 23:41:59 +08001346 }, be->pagepool);
Gao Xiang3883a792018-07-26 20:22:06 +08001347
1348out:
Yue Hucecf8642021-12-29 07:29:19 +08001349 /* must handle all compressed pages before actual file pages */
1350 if (z_erofs_is_inline_pcluster(pcl)) {
Gao Xianged722fb2022-07-15 23:41:54 +08001351 page = pcl->compressed_bvecs[0].page;
1352 WRITE_ONCE(pcl->compressed_bvecs[0].page, NULL);
Yue Hucecf8642021-12-29 07:29:19 +08001353 put_page(page);
1354 } else {
1355 for (i = 0; i < pclusterpages; ++i) {
Gao Xianged722fb2022-07-15 23:41:54 +08001356 page = pcl->compressed_bvecs[i].page;
Gao Xiang3883a792018-07-26 20:22:06 +08001357
Yue Hucecf8642021-12-29 07:29:19 +08001358 if (erofs_page_is_managed(sbi, page))
1359 continue;
Gao Xiangd61fbb62019-03-25 11:40:08 +08001360
Yue Hucecf8642021-12-29 07:29:19 +08001361 /* recycle all individual short-lived pages */
Gao Xiang4f056872022-07-15 23:41:59 +08001362 (void)z_erofs_put_shortlivedpage(be->pagepool, page);
Gao Xianged722fb2022-07-15 23:41:54 +08001363 WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL);
Yue Hucecf8642021-12-29 07:29:19 +08001364 }
Gao Xiang3883a792018-07-26 20:22:06 +08001365 }
Gao Xiangfe3e5912022-07-15 23:42:00 +08001366 if (be->compressed_pages < be->onstack_pages ||
1367 be->compressed_pages >= be->onstack_pages + Z_EROFS_ONSTACK_PAGES)
Gao Xiang12724ba2023-01-10 15:49:27 +08001368 kfree(be->compressed_pages);
Gao Xiang267f2492022-07-15 23:42:03 +08001369 z_erofs_fill_other_copies(be, err);
Gao Xiang3883a792018-07-26 20:22:06 +08001370
Gao Xiang2bfab9c2022-07-15 23:42:02 +08001371 for (i = 0; i < be->nr_pages; ++i) {
Gao Xiang4f056872022-07-15 23:41:59 +08001372 page = be->decompressed_pages[i];
Gao Xiangaf692e12019-02-27 13:33:30 +08001373 if (!page)
1374 continue;
1375
Gao Xiang6aaa7b02020-12-08 17:58:32 +08001376 DBG_BUGON(z_erofs_page_is_invalidated(page));
Gao Xiangaf692e12019-02-27 13:33:30 +08001377
Gao Xiang6aaa7b02020-12-08 17:58:32 +08001378 /* recycle all individual short-lived pages */
Gao Xiang4f056872022-07-15 23:41:59 +08001379 if (z_erofs_put_shortlivedpage(be->pagepool, page))
Gao Xiangaf692e12019-02-27 13:33:30 +08001380 continue;
Gao Xiang67148552022-07-15 23:41:55 +08001381 if (err)
1382 z_erofs_page_mark_eio(page);
Gao Xiangaf692e12019-02-27 13:33:30 +08001383 z_erofs_onlinepage_endio(page);
1384 }
1385
Gao Xiang4f056872022-07-15 23:41:59 +08001386 if (be->decompressed_pages != be->onstack_pages)
Gao Xiang12724ba2023-01-10 15:49:27 +08001387 kfree(be->decompressed_pages);
Gao Xiang3883a792018-07-26 20:22:06 +08001388
Gao Xiang2bfab9c2022-07-15 23:42:02 +08001389 pcl->length = 0;
1390 pcl->partial = true;
Gao Xiang267f2492022-07-15 23:42:03 +08001391 pcl->multibases = false;
Gao Xiang06a304c2022-07-15 23:41:51 +08001392 pcl->bvset.nextpage = NULL;
Gao Xiang87ca34a2022-05-29 13:54:23 +08001393 pcl->vcnt = 0;
Gao Xiang3883a792018-07-26 20:22:06 +08001394
Gao Xiang87ca34a2022-05-29 13:54:23 +08001395 /* pcluster lock MUST be taken before the following line */
Gao Xiang97e86a82019-07-31 23:57:47 +08001396 WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_NIL);
Gao Xiang87ca34a2022-05-29 13:54:23 +08001397 mutex_unlock(&pcl->lock);
Gao Xiang3883a792018-07-26 20:22:06 +08001398 return err;
1399}
1400
Gao Xiang0c638f72019-11-08 11:37:33 +08001401static void z_erofs_decompress_queue(const struct z_erofs_decompressqueue *io,
Gao Xiangeaa91722021-10-22 17:01:20 +08001402 struct page **pagepool)
Gao Xiang3883a792018-07-26 20:22:06 +08001403{
Gao Xiang4f056872022-07-15 23:41:59 +08001404 struct z_erofs_decompress_backend be = {
1405 .sb = io->sb,
1406 .pagepool = pagepool,
Gao Xiang267f2492022-07-15 23:42:03 +08001407 .decompressed_secondary_bvecs =
1408 LIST_HEAD_INIT(be.decompressed_secondary_bvecs),
Gao Xiang4f056872022-07-15 23:41:59 +08001409 };
Gao Xiang97e86a82019-07-31 23:57:47 +08001410 z_erofs_next_pcluster_t owned = io->head;
Gao Xiang3883a792018-07-26 20:22:06 +08001411
Gao Xiang97e86a82019-07-31 23:57:47 +08001412 while (owned != Z_EROFS_PCLUSTER_TAIL_CLOSED) {
Gao Xiang4f056872022-07-15 23:41:59 +08001413 /* impossible that 'owned' equals Z_EROFS_WORK_TPTR_TAIL */
Gao Xiang97e86a82019-07-31 23:57:47 +08001414 DBG_BUGON(owned == Z_EROFS_PCLUSTER_TAIL);
Gao Xiang4f056872022-07-15 23:41:59 +08001415 /* impossible that 'owned' equals Z_EROFS_PCLUSTER_NIL */
Gao Xiang97e86a82019-07-31 23:57:47 +08001416 DBG_BUGON(owned == Z_EROFS_PCLUSTER_NIL);
Gao Xiang3883a792018-07-26 20:22:06 +08001417
Gao Xiang4f056872022-07-15 23:41:59 +08001418 be.pcl = container_of(owned, struct z_erofs_pcluster, next);
1419 owned = READ_ONCE(be.pcl->next);
Gao Xiang3883a792018-07-26 20:22:06 +08001420
Gao Xiang4f056872022-07-15 23:41:59 +08001421 z_erofs_decompress_pcluster(&be, io->eio ? -EIO : 0);
1422 erofs_workgroup_put(&be.pcl->obj);
Gao Xiang3978c8e2018-08-06 11:27:53 +08001423 }
Gao Xiang3883a792018-07-26 20:22:06 +08001424}
1425
Gao Xiang0c638f72019-11-08 11:37:33 +08001426static void z_erofs_decompressqueue_work(struct work_struct *work)
Gao Xiang3883a792018-07-26 20:22:06 +08001427{
Gao Xianga4b1fab2019-10-08 20:56:15 +08001428 struct z_erofs_decompressqueue *bgq =
1429 container_of(work, struct z_erofs_decompressqueue, u.work);
Gao Xiangeaa91722021-10-22 17:01:20 +08001430 struct page *pagepool = NULL;
Gao Xiang3883a792018-07-26 20:22:06 +08001431
Gao Xianga4b1fab2019-10-08 20:56:15 +08001432 DBG_BUGON(bgq->head == Z_EROFS_PCLUSTER_TAIL_CLOSED);
Gao Xiang0c638f72019-11-08 11:37:33 +08001433 z_erofs_decompress_queue(bgq, &pagepool);
Gao Xiangeaa91722021-10-22 17:01:20 +08001434 erofs_release_pages(&pagepool);
Gao Xianga4b1fab2019-10-08 20:56:15 +08001435 kvfree(bgq);
Gao Xiang3883a792018-07-26 20:22:06 +08001436}
1437
Sandeep Dhavale3fffb582023-02-08 17:33:22 +08001438#ifdef CONFIG_EROFS_FS_PCPU_KTHREAD
1439static void z_erofs_decompressqueue_kthread_work(struct kthread_work *work)
1440{
1441 z_erofs_decompressqueue_work((struct work_struct *)work);
1442}
1443#endif
1444
Gao Xiang78658272022-01-21 17:14:12 +08001445static void z_erofs_decompress_kickoff(struct z_erofs_decompressqueue *io,
Gao Xiangcdba5502023-02-04 17:30:36 +08001446 int bios)
Gao Xiang78658272022-01-21 17:14:12 +08001447{
1448 struct erofs_sb_info *const sbi = EROFS_SB(io->sb);
1449
1450 /* wake up the caller thread for sync decompression */
Gao Xiangcdba5502023-02-04 17:30:36 +08001451 if (io->sync) {
Gao Xiang78658272022-01-21 17:14:12 +08001452 if (!atomic_add_return(bios, &io->pending_bios))
Hongyu Jin60b30052022-04-01 19:55:27 +08001453 complete(&io->u.done);
Gao Xiang78658272022-01-21 17:14:12 +08001454 return;
1455 }
1456
1457 if (atomic_add_return(bios, &io->pending_bios))
1458 return;
Sandeep Dhavale3fffb582023-02-08 17:33:22 +08001459 /* Use (kthread_)work and sync decompression for atomic contexts only */
Gao Xiang78658272022-01-21 17:14:12 +08001460 if (in_atomic() || irqs_disabled()) {
Sandeep Dhavale3fffb582023-02-08 17:33:22 +08001461#ifdef CONFIG_EROFS_FS_PCPU_KTHREAD
1462 struct kthread_worker *worker;
1463
1464 rcu_read_lock();
1465 worker = rcu_dereference(
1466 z_erofs_pcpu_workers[raw_smp_processor_id()]);
1467 if (!worker) {
1468 INIT_WORK(&io->u.work, z_erofs_decompressqueue_work);
1469 queue_work(z_erofs_workqueue, &io->u.work);
1470 } else {
1471 kthread_queue_work(worker, &io->u.kthread_work);
1472 }
1473 rcu_read_unlock();
1474#else
Gao Xiang78658272022-01-21 17:14:12 +08001475 queue_work(z_erofs_workqueue, &io->u.work);
Sandeep Dhavale3fffb582023-02-08 17:33:22 +08001476#endif
Gao Xiang78658272022-01-21 17:14:12 +08001477 /* enable sync decompression for readahead */
1478 if (sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_AUTO)
1479 sbi->opt.sync_decompress = EROFS_SYNC_DECOMPRESS_FORCE_ON;
1480 return;
1481 }
1482 z_erofs_decompressqueue_work(&io->u.work);
1483}
1484
Gao Xiang97e86a82019-07-31 23:57:47 +08001485static struct page *pickup_page_for_submission(struct z_erofs_pcluster *pcl,
1486 unsigned int nr,
Gao Xiangeaa91722021-10-22 17:01:20 +08001487 struct page **pagepool,
Gao Xiang9f2731d2022-03-11 02:27:43 +08001488 struct address_space *mc)
Gao Xiang9248fce2018-12-08 00:19:15 +08001489{
Gao Xiang97e86a82019-07-31 23:57:47 +08001490 const pgoff_t index = pcl->obj.index;
Gao Xiang9f2731d2022-03-11 02:27:43 +08001491 gfp_t gfp = mapping_gfp_mask(mc);
Gao Xiang9248fce2018-12-08 00:19:15 +08001492 bool tocache = false;
1493
1494 struct address_space *mapping;
1495 struct page *oldpage, *page;
Gao Xiang92e6efd2018-12-08 00:19:16 +08001496 int justfound;
1497
Gao Xiang9248fce2018-12-08 00:19:15 +08001498repeat:
Gao Xianged722fb2022-07-15 23:41:54 +08001499 page = READ_ONCE(pcl->compressed_bvecs[nr].page);
Gao Xiang9248fce2018-12-08 00:19:15 +08001500 oldpage = page;
1501
1502 if (!page)
1503 goto out_allocpage;
1504
Gao Xiangb1ed2202023-02-04 17:30:37 +08001505 justfound = (unsigned long)page & 1UL;
1506 page = (struct page *)((unsigned long)page & ~1UL);
Gao Xiang92e6efd2018-12-08 00:19:16 +08001507
Gao Xiang1825c8d2020-12-09 20:37:17 +08001508 /*
1509 * preallocated cached pages, which is used to avoid direct reclaim
1510 * otherwise, it will go inplace I/O path instead.
1511 */
1512 if (page->private == Z_EROFS_PREALLOCATED_PAGE) {
Gao Xianged722fb2022-07-15 23:41:54 +08001513 WRITE_ONCE(pcl->compressed_bvecs[nr].page, page);
Gao Xiang1825c8d2020-12-09 20:37:17 +08001514 set_page_private(page, 0);
1515 tocache = true;
1516 goto out_tocache;
1517 }
Gao Xiang9248fce2018-12-08 00:19:15 +08001518 mapping = READ_ONCE(page->mapping);
1519
1520 /*
Gao Xiang6aaa7b02020-12-08 17:58:32 +08001521 * file-backed online pages in plcuster are all locked steady,
Gao Xiang9248fce2018-12-08 00:19:15 +08001522 * therefore it is impossible for `mapping' to be NULL.
1523 */
1524 if (mapping && mapping != mc)
1525 /* ought to be unmanaged pages */
1526 goto out;
1527
Gao Xiang6aaa7b02020-12-08 17:58:32 +08001528 /* directly return for shortlived page as well */
1529 if (z_erofs_is_shortlived_page(page))
1530 goto out;
1531
Gao Xiang9248fce2018-12-08 00:19:15 +08001532 lock_page(page);
1533
Gao Xiang92e6efd2018-12-08 00:19:16 +08001534 /* only true if page reclaim goes wrong, should never happen */
1535 DBG_BUGON(justfound && PagePrivate(page));
1536
Gao Xiang9248fce2018-12-08 00:19:15 +08001537 /* the page is still in manage cache */
1538 if (page->mapping == mc) {
Gao Xianged722fb2022-07-15 23:41:54 +08001539 WRITE_ONCE(pcl->compressed_bvecs[nr].page, page);
Gao Xiang9248fce2018-12-08 00:19:15 +08001540
1541 if (!PagePrivate(page)) {
Gao Xiang92e6efd2018-12-08 00:19:16 +08001542 /*
1543 * impossible to be !PagePrivate(page) for
1544 * the current restriction as well if
Gao Xianged722fb2022-07-15 23:41:54 +08001545 * the page is already in compressed_bvecs[].
Gao Xiang92e6efd2018-12-08 00:19:16 +08001546 */
1547 DBG_BUGON(!justfound);
1548
1549 justfound = 0;
Gao Xiang97e86a82019-07-31 23:57:47 +08001550 set_page_private(page, (unsigned long)pcl);
Gao Xiang9248fce2018-12-08 00:19:15 +08001551 SetPagePrivate(page);
1552 }
1553
1554 /* no need to submit io if it is already up-to-date */
1555 if (PageUptodate(page)) {
1556 unlock_page(page);
1557 page = NULL;
1558 }
1559 goto out;
1560 }
1561
1562 /*
1563 * the managed page has been truncated, it's unsafe to
1564 * reuse this one, let's allocate a new cache-managed page.
1565 */
1566 DBG_BUGON(page->mapping);
Gao Xiang92e6efd2018-12-08 00:19:16 +08001567 DBG_BUGON(!justfound);
Gao Xiang9248fce2018-12-08 00:19:15 +08001568
1569 tocache = true;
1570 unlock_page(page);
1571 put_page(page);
1572out_allocpage:
Gao Xiang5ddcee12019-11-21 21:59:54 +08001573 page = erofs_allocpage(pagepool, gfp | __GFP_NOFAIL);
Gao Xianged722fb2022-07-15 23:41:54 +08001574 if (oldpage != cmpxchg(&pcl->compressed_bvecs[nr].page,
1575 oldpage, page)) {
Gao Xiangeaa91722021-10-22 17:01:20 +08001576 erofs_pagepool_add(pagepool, page);
Gao Xiang5ddcee12019-11-21 21:59:54 +08001577 cond_resched();
1578 goto repeat;
1579 }
Gao Xiang1825c8d2020-12-09 20:37:17 +08001580out_tocache:
Gao Xiangbf225072020-12-08 17:58:33 +08001581 if (!tocache || add_to_page_cache_lru(page, mc, index + nr, gfp)) {
1582 /* turn into temporary page if fails (1 ref) */
1583 set_page_private(page, Z_EROFS_SHORTLIVED_PAGE);
1584 goto out;
Gao Xianga30573b2020-10-22 22:57:21 +08001585 }
Gao Xiangbf225072020-12-08 17:58:33 +08001586 attach_page_private(page, pcl);
1587 /* drop a refcount added by allocpage (then we have 2 refs here) */
1588 put_page(page);
1589
Gao Xiang9248fce2018-12-08 00:19:15 +08001590out: /* the only exit (for tracing and debugging) */
1591 return page;
1592}
1593
Gao Xiangcdba5502023-02-04 17:30:36 +08001594static struct z_erofs_decompressqueue *jobqueue_init(struct super_block *sb,
1595 struct z_erofs_decompressqueue *fgq, bool *fg)
Gao Xiang3883a792018-07-26 20:22:06 +08001596{
Gao Xianga4b1fab2019-10-08 20:56:15 +08001597 struct z_erofs_decompressqueue *q;
Gao Xiang3883a792018-07-26 20:22:06 +08001598
Gao Xianga4b1fab2019-10-08 20:56:15 +08001599 if (fg && !*fg) {
1600 q = kvzalloc(sizeof(*q), GFP_KERNEL | __GFP_NOWARN);
1601 if (!q) {
1602 *fg = true;
1603 goto fg_out;
1604 }
Sandeep Dhavale3fffb582023-02-08 17:33:22 +08001605#ifdef CONFIG_EROFS_FS_PCPU_KTHREAD
1606 kthread_init_work(&q->u.kthread_work,
1607 z_erofs_decompressqueue_kthread_work);
1608#else
Gao Xiang0c638f72019-11-08 11:37:33 +08001609 INIT_WORK(&q->u.work, z_erofs_decompressqueue_work);
Sandeep Dhavale3fffb582023-02-08 17:33:22 +08001610#endif
Gao Xianga4b1fab2019-10-08 20:56:15 +08001611 } else {
1612fg_out:
1613 q = fgq;
Hongyu Jin60b30052022-04-01 19:55:27 +08001614 init_completion(&fgq->u.done);
Gao Xianga4b1fab2019-10-08 20:56:15 +08001615 atomic_set(&fgq->pending_bios, 0);
Gao Xiang67148552022-07-15 23:41:55 +08001616 q->eio = false;
Gao Xiangcdba5502023-02-04 17:30:36 +08001617 q->sync = true;
Gao Xiang3883a792018-07-26 20:22:06 +08001618 }
Gao Xianga4b1fab2019-10-08 20:56:15 +08001619 q->sb = sb;
1620 q->head = Z_EROFS_PCLUSTER_TAIL_CLOSED;
1621 return q;
Gao Xiang3883a792018-07-26 20:22:06 +08001622}
1623
Gao Xiang97e86a82019-07-31 23:57:47 +08001624/* define decompression jobqueue types */
Gao Xiang7146a4f2018-12-08 00:19:18 +08001625enum {
Gao Xiang7146a4f2018-12-08 00:19:18 +08001626 JQ_BYPASS,
Gao Xiang7146a4f2018-12-08 00:19:18 +08001627 JQ_SUBMIT,
1628 NR_JOBQUEUES,
1629};
1630
Gao Xiang97e86a82019-07-31 23:57:47 +08001631static void move_to_bypass_jobqueue(struct z_erofs_pcluster *pcl,
1632 z_erofs_next_pcluster_t qtail[],
1633 z_erofs_next_pcluster_t owned_head)
Gao Xiang7146a4f2018-12-08 00:19:18 +08001634{
Gao Xiang97e86a82019-07-31 23:57:47 +08001635 z_erofs_next_pcluster_t *const submit_qtail = qtail[JQ_SUBMIT];
1636 z_erofs_next_pcluster_t *const bypass_qtail = qtail[JQ_BYPASS];
Gao Xiang7146a4f2018-12-08 00:19:18 +08001637
Gao Xiang97e86a82019-07-31 23:57:47 +08001638 DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED);
1639 if (owned_head == Z_EROFS_PCLUSTER_TAIL)
1640 owned_head = Z_EROFS_PCLUSTER_TAIL_CLOSED;
Gao Xiang7146a4f2018-12-08 00:19:18 +08001641
Gao Xiang97e86a82019-07-31 23:57:47 +08001642 WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_TAIL_CLOSED);
Gao Xiang7146a4f2018-12-08 00:19:18 +08001643
1644 WRITE_ONCE(*submit_qtail, owned_head);
Gao Xiang97e86a82019-07-31 23:57:47 +08001645 WRITE_ONCE(*bypass_qtail, &pcl->next);
Gao Xiang7146a4f2018-12-08 00:19:18 +08001646
Gao Xiang97e86a82019-07-31 23:57:47 +08001647 qtail[JQ_BYPASS] = &pcl->next;
Gao Xiang7146a4f2018-12-08 00:19:18 +08001648}
1649
Gao Xiang78658272022-01-21 17:14:12 +08001650static void z_erofs_decompressqueue_endio(struct bio *bio)
1651{
Gao Xiangcdba5502023-02-04 17:30:36 +08001652 struct z_erofs_decompressqueue *q = bio->bi_private;
Gao Xiang78658272022-01-21 17:14:12 +08001653 blk_status_t err = bio->bi_status;
1654 struct bio_vec *bvec;
1655 struct bvec_iter_all iter_all;
1656
1657 bio_for_each_segment_all(bvec, bio, iter_all) {
1658 struct page *page = bvec->bv_page;
1659
1660 DBG_BUGON(PageUptodate(page));
1661 DBG_BUGON(z_erofs_page_is_invalidated(page));
1662
Gao Xiang78658272022-01-21 17:14:12 +08001663 if (erofs_page_is_managed(EROFS_SB(q->sb), page)) {
1664 if (!err)
1665 SetPageUptodate(page);
1666 unlock_page(page);
1667 }
1668 }
Gao Xiang67148552022-07-15 23:41:55 +08001669 if (err)
1670 q->eio = true;
Gao Xiangcdba5502023-02-04 17:30:36 +08001671 z_erofs_decompress_kickoff(q, -1);
Gao Xiang78658272022-01-21 17:14:12 +08001672 bio_put(bio);
1673}
1674
Gao Xiang83a386c2022-07-15 23:41:48 +08001675static void z_erofs_submit_queue(struct z_erofs_decompress_frontend *f,
Gao Xiangeaa91722021-10-22 17:01:20 +08001676 struct page **pagepool,
Gao Xiang0c638f72019-11-08 11:37:33 +08001677 struct z_erofs_decompressqueue *fgq,
1678 bool *force_fg)
Gao Xiang3883a792018-07-26 20:22:06 +08001679{
Gao Xiang83a386c2022-07-15 23:41:48 +08001680 struct super_block *sb = f->inode->i_sb;
1681 struct address_space *mc = MNGD_MAPPING(EROFS_SB(sb));
Gao Xiang97e86a82019-07-31 23:57:47 +08001682 z_erofs_next_pcluster_t qtail[NR_JOBQUEUES];
Gao Xianga4b1fab2019-10-08 20:56:15 +08001683 struct z_erofs_decompressqueue *q[NR_JOBQUEUES];
Gao Xiang5c6dcc52022-03-02 03:49:50 +08001684 z_erofs_next_pcluster_t owned_head = f->owned_head;
Gao Xiangdfeab2e2021-10-14 16:10:10 +08001685 /* bio is NULL initially, so no need to initialize last_{index,bdev} */
Kees Cook3f649ab2020-06-03 13:09:38 -07001686 pgoff_t last_index;
Gao Xiangdfeab2e2021-10-14 16:10:10 +08001687 struct block_device *last_bdev;
Gao Xiang1e4a2952020-01-21 14:48:19 +08001688 unsigned int nr_bios = 0;
1689 struct bio *bio = NULL;
Johannes Weiner82e60d02022-11-03 17:34:31 -04001690 unsigned long pflags;
1691 int memstall = 0;
Gao Xiang3883a792018-07-26 20:22:06 +08001692
Gao Xiangcdba5502023-02-04 17:30:36 +08001693 /*
1694 * if managed cache is enabled, bypass jobqueue is needed,
1695 * no need to read from device for all pclusters in this queue.
1696 */
1697 q[JQ_BYPASS] = jobqueue_init(sb, fgq + JQ_BYPASS, NULL);
1698 q[JQ_SUBMIT] = jobqueue_init(sb, fgq + JQ_SUBMIT, force_fg);
1699
Gao Xianga4b1fab2019-10-08 20:56:15 +08001700 qtail[JQ_BYPASS] = &q[JQ_BYPASS]->head;
1701 qtail[JQ_SUBMIT] = &q[JQ_SUBMIT]->head;
Gao Xiang3883a792018-07-26 20:22:06 +08001702
1703 /* by default, all need io submission */
Gao Xiang7146a4f2018-12-08 00:19:18 +08001704 q[JQ_SUBMIT]->head = owned_head;
Gao Xiang3883a792018-07-26 20:22:06 +08001705
1706 do {
Gao Xiangdfeab2e2021-10-14 16:10:10 +08001707 struct erofs_map_dev mdev;
Gao Xiang97e86a82019-07-31 23:57:47 +08001708 struct z_erofs_pcluster *pcl;
Gao Xiang1e4a2952020-01-21 14:48:19 +08001709 pgoff_t cur, end;
1710 unsigned int i = 0;
1711 bool bypass = true;
Gao Xiang3883a792018-07-26 20:22:06 +08001712
1713 /* no possible 'owned_head' equals the following */
Gao Xiang97e86a82019-07-31 23:57:47 +08001714 DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED);
1715 DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_NIL);
Gao Xiang3883a792018-07-26 20:22:06 +08001716
Gao Xiang97e86a82019-07-31 23:57:47 +08001717 pcl = container_of(owned_head, struct z_erofs_pcluster, next);
1718
Yue Hucecf8642021-12-29 07:29:19 +08001719 /* close the main owned chain at first */
1720 owned_head = cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_TAIL,
1721 Z_EROFS_PCLUSTER_TAIL_CLOSED);
1722 if (z_erofs_is_inline_pcluster(pcl)) {
1723 move_to_bypass_jobqueue(pcl, qtail, owned_head);
1724 continue;
1725 }
1726
Gao Xiangdfeab2e2021-10-14 16:10:10 +08001727 /* no device id here, thus it will always succeed */
1728 mdev = (struct erofs_map_dev) {
1729 .m_pa = blknr_to_addr(pcl->obj.index),
1730 };
1731 (void)erofs_map_dev(sb, &mdev);
1732
1733 cur = erofs_blknr(mdev.m_pa);
Gao Xiang9f6cc762021-04-07 12:39:20 +08001734 end = cur + pcl->pclusterpages;
Gao Xiang3883a792018-07-26 20:22:06 +08001735
Gao Xiang1e4a2952020-01-21 14:48:19 +08001736 do {
1737 struct page *page;
Gao Xiang9248fce2018-12-08 00:19:15 +08001738
Gao Xiang1e4a2952020-01-21 14:48:19 +08001739 page = pickup_page_for_submission(pcl, i++, pagepool,
Gao Xiang83a386c2022-07-15 23:41:48 +08001740 mc);
Gao Xiang1e4a2952020-01-21 14:48:19 +08001741 if (!page)
1742 continue;
Gao Xiang3883a792018-07-26 20:22:06 +08001743
Gao Xiangdfeab2e2021-10-14 16:10:10 +08001744 if (bio && (cur != last_index + 1 ||
1745 last_bdev != mdev.m_bdev)) {
Gao Xiang3883a792018-07-26 20:22:06 +08001746submit_bio_retry:
Gao Xiang1e4a2952020-01-21 14:48:19 +08001747 submit_bio(bio);
Johannes Weiner82e60d02022-11-03 17:34:31 -04001748 if (memstall) {
1749 psi_memstall_leave(&pflags);
1750 memstall = 0;
1751 }
Gao Xiang1e4a2952020-01-21 14:48:19 +08001752 bio = NULL;
1753 }
Gao Xiang3883a792018-07-26 20:22:06 +08001754
Johannes Weiner82e60d02022-11-03 17:34:31 -04001755 if (unlikely(PageWorkingset(page)) && !memstall) {
Christoph Hellwig99486c52022-09-15 10:41:59 +01001756 psi_memstall_enter(&pflags);
Johannes Weiner82e60d02022-11-03 17:34:31 -04001757 memstall = 1;
1758 }
Christoph Hellwig99486c52022-09-15 10:41:59 +01001759
Gao Xiang1e4a2952020-01-21 14:48:19 +08001760 if (!bio) {
Christoph Hellwig07888c662022-01-24 10:11:05 +01001761 bio = bio_alloc(mdev.m_bdev, BIO_MAX_VECS,
1762 REQ_OP_READ, GFP_NOIO);
Gao Xiang1e4a2952020-01-21 14:48:19 +08001763 bio->bi_end_io = z_erofs_decompressqueue_endio;
Gao Xiangdfeab2e2021-10-14 16:10:10 +08001764
Gao Xiangdfeab2e2021-10-14 16:10:10 +08001765 last_bdev = mdev.m_bdev;
Gao Xiang1e4a2952020-01-21 14:48:19 +08001766 bio->bi_iter.bi_sector = (sector_t)cur <<
1767 LOG_SECTORS_PER_BLOCK;
Gao Xiangcdba5502023-02-04 17:30:36 +08001768 bio->bi_private = q[JQ_SUBMIT];
Gao Xiang6ea5aad2020-09-19 15:27:30 +08001769 if (f->readahead)
1770 bio->bi_opf |= REQ_RAHEAD;
Gao Xiang1e4a2952020-01-21 14:48:19 +08001771 ++nr_bios;
1772 }
Gao Xiang94e4e152019-09-04 10:09:04 +08001773
Gao Xiang6c3e4852020-09-19 15:27:28 +08001774 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
Gao Xiang1e4a2952020-01-21 14:48:19 +08001775 goto submit_bio_retry;
Gao Xiang3883a792018-07-26 20:22:06 +08001776
Gao Xiang1e4a2952020-01-21 14:48:19 +08001777 last_index = cur;
1778 bypass = false;
1779 } while (++cur < end);
Gao Xiang3883a792018-07-26 20:22:06 +08001780
Gao Xiang1e4a2952020-01-21 14:48:19 +08001781 if (!bypass)
Gao Xiang97e86a82019-07-31 23:57:47 +08001782 qtail[JQ_SUBMIT] = &pcl->next;
Gao Xiang7146a4f2018-12-08 00:19:18 +08001783 else
Gao Xiang97e86a82019-07-31 23:57:47 +08001784 move_to_bypass_jobqueue(pcl, qtail, owned_head);
1785 } while (owned_head != Z_EROFS_PCLUSTER_TAIL);
Gao Xiang3883a792018-07-26 20:22:06 +08001786
Christoph Hellwig99486c52022-09-15 10:41:59 +01001787 if (bio) {
Gao Xiang94e4e152019-09-04 10:09:04 +08001788 submit_bio(bio);
Johannes Weiner82e60d02022-11-03 17:34:31 -04001789 if (memstall)
1790 psi_memstall_leave(&pflags);
Christoph Hellwig99486c52022-09-15 10:41:59 +01001791 }
Gao Xiang3883a792018-07-26 20:22:06 +08001792
Gao Xiang587a67b72020-01-21 14:47:47 +08001793 /*
1794 * although background is preferred, no one is pending for submission.
Sandeep Dhavale3fffb582023-02-08 17:33:22 +08001795 * don't issue decompression but drop it directly instead.
Gao Xiang587a67b72020-01-21 14:47:47 +08001796 */
1797 if (!*force_fg && !nr_bios) {
1798 kvfree(q[JQ_SUBMIT]);
Gao Xiang1e4a2952020-01-21 14:48:19 +08001799 return;
Gao Xiang587a67b72020-01-21 14:47:47 +08001800 }
Gao Xiangcdba5502023-02-04 17:30:36 +08001801 z_erofs_decompress_kickoff(q[JQ_SUBMIT], nr_bios);
Gao Xiang3883a792018-07-26 20:22:06 +08001802}
1803
Gao Xiang83a386c2022-07-15 23:41:48 +08001804static void z_erofs_runqueue(struct z_erofs_decompress_frontend *f,
Gao Xiangeaa91722021-10-22 17:01:20 +08001805 struct page **pagepool, bool force_fg)
Gao Xiang3883a792018-07-26 20:22:06 +08001806{
Gao Xianga4b1fab2019-10-08 20:56:15 +08001807 struct z_erofs_decompressqueue io[NR_JOBQUEUES];
Gao Xiang3883a792018-07-26 20:22:06 +08001808
Gao Xiang5c6dcc52022-03-02 03:49:50 +08001809 if (f->owned_head == Z_EROFS_PCLUSTER_TAIL)
Gao Xiang3883a792018-07-26 20:22:06 +08001810 return;
Gao Xiang83a386c2022-07-15 23:41:48 +08001811 z_erofs_submit_queue(f, pagepool, io, &force_fg);
Gao Xiang3883a792018-07-26 20:22:06 +08001812
Gao Xiang0c638f72019-11-08 11:37:33 +08001813 /* handle bypass queue (no i/o pclusters) immediately */
1814 z_erofs_decompress_queue(&io[JQ_BYPASS], pagepool);
Gao Xiang4279f3f2019-07-31 23:57:49 +08001815
Gao Xiang3883a792018-07-26 20:22:06 +08001816 if (!force_fg)
1817 return;
1818
1819 /* wait until all bios are completed */
Hongyu Jin60b30052022-04-01 19:55:27 +08001820 wait_for_completion_io(&io[JQ_SUBMIT].u.done);
Gao Xiang3883a792018-07-26 20:22:06 +08001821
Gao Xiang0c638f72019-11-08 11:37:33 +08001822 /* handle synchronous decompress queue in the caller context */
1823 z_erofs_decompress_queue(&io[JQ_SUBMIT], pagepool);
Gao Xiang3883a792018-07-26 20:22:06 +08001824}
1825
Gao Xiang38629292021-10-09 04:08:39 +08001826/*
1827 * Since partial uptodate is still unimplemented for now, we have to use
1828 * approximate readmore strategies as a start.
1829 */
1830static void z_erofs_pcluster_readmore(struct z_erofs_decompress_frontend *f,
1831 struct readahead_control *rac,
1832 erofs_off_t end,
Gao Xiangeaa91722021-10-22 17:01:20 +08001833 struct page **pagepool,
Gao Xiang38629292021-10-09 04:08:39 +08001834 bool backmost)
1835{
1836 struct inode *inode = f->inode;
1837 struct erofs_map_blocks *map = &f->map;
1838 erofs_off_t cur;
1839 int err;
1840
1841 if (backmost) {
1842 map->m_la = end;
Gao Xiang622cead2021-10-11 05:31:45 +08001843 err = z_erofs_map_blocks_iter(inode, map,
1844 EROFS_GET_BLOCKS_READMORE);
Gao Xiang38629292021-10-09 04:08:39 +08001845 if (err)
1846 return;
1847
1848 /* expend ra for the trailing edge if readahead */
1849 if (rac) {
1850 loff_t newstart = readahead_pos(rac);
1851
1852 cur = round_up(map->m_la + map->m_llen, PAGE_SIZE);
1853 readahead_expand(rac, newstart, cur - newstart);
1854 return;
1855 }
1856 end = round_up(end, PAGE_SIZE);
1857 } else {
1858 end = round_up(map->m_la, PAGE_SIZE);
1859
1860 if (!map->m_llen)
1861 return;
1862 }
1863
1864 cur = map->m_la + map->m_llen - 1;
1865 while (cur >= end) {
1866 pgoff_t index = cur >> PAGE_SHIFT;
1867 struct page *page;
1868
1869 page = erofs_grab_cache_page_nowait(inode->i_mapping, index);
Gao Xiangaa793b462022-05-29 13:54:25 +08001870 if (page) {
1871 if (PageUptodate(page)) {
1872 unlock_page(page);
1873 } else {
1874 err = z_erofs_do_read_page(f, page, pagepool);
1875 if (err)
1876 erofs_err(inode->i_sb,
1877 "readmore error at page %lu @ nid %llu",
1878 index, EROFS_I(inode)->nid);
1879 }
Gao Xiang38629292021-10-09 04:08:39 +08001880 put_page(page);
Gao Xiang38629292021-10-09 04:08:39 +08001881 }
1882
Gao Xiang38629292021-10-09 04:08:39 +08001883 if (cur < PAGE_SIZE)
1884 break;
1885 cur = (index << PAGE_SHIFT) - 1;
1886 }
1887}
1888
Matthew Wilcox (Oracle)a2e20a22022-04-29 11:12:16 -04001889static int z_erofs_read_folio(struct file *file, struct folio *folio)
Gao Xiang3883a792018-07-26 20:22:06 +08001890{
Matthew Wilcox (Oracle)a2e20a22022-04-29 11:12:16 -04001891 struct page *page = &folio->page;
Gao Xiang3883a792018-07-26 20:22:06 +08001892 struct inode *const inode = page->mapping->host;
Huang Jianan40452ff2021-12-06 22:35:52 +08001893 struct erofs_sb_info *const sbi = EROFS_I_SB(inode);
Gao Xiang97e86a82019-07-31 23:57:47 +08001894 struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode);
Gao Xiangeaa91722021-10-22 17:01:20 +08001895 struct page *pagepool = NULL;
Gao Xiang3883a792018-07-26 20:22:06 +08001896 int err;
Gao Xiang3883a792018-07-26 20:22:06 +08001897
Gao Xiangba9ce772018-11-23 01:15:58 +08001898 trace_erofs_readpage(page, false);
Gao Xiangf0c519f2018-11-23 01:21:49 +08001899 f.headoffset = (erofs_off_t)page->index << PAGE_SHIFT;
1900
Gao Xiang38629292021-10-09 04:08:39 +08001901 z_erofs_pcluster_readmore(&f, NULL, f.headoffset + PAGE_SIZE - 1,
1902 &pagepool, true);
Gao Xiang1825c8d2020-12-09 20:37:17 +08001903 err = z_erofs_do_read_page(&f, page, &pagepool);
Gao Xiang38629292021-10-09 04:08:39 +08001904 z_erofs_pcluster_readmore(&f, NULL, 0, &pagepool, false);
1905
Gao Xiang5c6dcc52022-03-02 03:49:50 +08001906 (void)z_erofs_collector_end(&f);
Gao Xiang3883a792018-07-26 20:22:06 +08001907
Gao Xiangee451972019-08-19 18:34:21 +08001908 /* if some compressed cluster ready, need submit them anyway */
Gao Xiang83a386c2022-07-15 23:41:48 +08001909 z_erofs_runqueue(&f, &pagepool,
Huang Jianan40452ff2021-12-06 22:35:52 +08001910 z_erofs_get_sync_decompress_policy(sbi, 0));
Gao Xiangee451972019-08-19 18:34:21 +08001911
1912 if (err)
Gao Xiang4f761fa2019-09-04 10:09:09 +08001913 erofs_err(inode->i_sb, "failed to read, err [%d]", err);
Gao Xiangee451972019-08-19 18:34:21 +08001914
Gao Xiang09c54372022-01-02 12:00:17 +08001915 erofs_put_metabuf(&f.map.buf);
Gao Xiangeaa91722021-10-22 17:01:20 +08001916 erofs_release_pages(&pagepool);
Gao Xiangee451972019-08-19 18:34:21 +08001917 return err;
Gao Xiang3883a792018-07-26 20:22:06 +08001918}
1919
Matthew Wilcox (Oracle)06150902020-06-01 21:47:13 -07001920static void z_erofs_readahead(struct readahead_control *rac)
Gao Xiang3883a792018-07-26 20:22:06 +08001921{
Matthew Wilcox (Oracle)06150902020-06-01 21:47:13 -07001922 struct inode *const inode = rac->mapping->host;
Gao Xiang5fb76bb2018-09-20 00:06:56 +08001923 struct erofs_sb_info *const sbi = EROFS_I_SB(inode);
Gao Xiang97e86a82019-07-31 23:57:47 +08001924 struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode);
Gao Xiangeaa91722021-10-22 17:01:20 +08001925 struct page *pagepool = NULL, *head = NULL, *page;
Gao Xiang38629292021-10-09 04:08:39 +08001926 unsigned int nr_pages;
Gao Xiang3883a792018-07-26 20:22:06 +08001927
Gao Xiang6ea5aad2020-09-19 15:27:30 +08001928 f.readahead = true;
Matthew Wilcox (Oracle)06150902020-06-01 21:47:13 -07001929 f.headoffset = readahead_pos(rac);
Gao Xiangf0c519f2018-11-23 01:21:49 +08001930
Gao Xiang38629292021-10-09 04:08:39 +08001931 z_erofs_pcluster_readmore(&f, rac, f.headoffset +
1932 readahead_length(rac) - 1, &pagepool, true);
1933 nr_pages = readahead_count(rac);
1934 trace_erofs_readpages(inode, readahead_index(rac), nr_pages, false);
1935
Matthew Wilcox (Oracle)06150902020-06-01 21:47:13 -07001936 while ((page = readahead_page(rac))) {
Gao Xiang3883a792018-07-26 20:22:06 +08001937 set_page_private(page, (unsigned long)head);
1938 head = page;
1939 }
1940
Cristian Sicilia42d40b42018-11-12 21:43:57 +01001941 while (head) {
Gao Xiang3883a792018-07-26 20:22:06 +08001942 struct page *page = head;
1943 int err;
1944
1945 /* traversal in reverse order */
1946 head = (void *)page_private(page);
1947
Gao Xiang1825c8d2020-12-09 20:37:17 +08001948 err = z_erofs_do_read_page(&f, page, &pagepool);
Gao Xianga5876e22019-09-04 10:08:56 +08001949 if (err)
Gao Xiang4f761fa2019-09-04 10:09:09 +08001950 erofs_err(inode->i_sb,
1951 "readahead error at page %lu @ nid %llu",
1952 page->index, EROFS_I(inode)->nid);
Gao Xiang3883a792018-07-26 20:22:06 +08001953 put_page(page);
1954 }
Gao Xiang38629292021-10-09 04:08:39 +08001955 z_erofs_pcluster_readmore(&f, rac, 0, &pagepool, false);
Gao Xiang5c6dcc52022-03-02 03:49:50 +08001956 (void)z_erofs_collector_end(&f);
Gao Xiang3883a792018-07-26 20:22:06 +08001957
Gao Xiang83a386c2022-07-15 23:41:48 +08001958 z_erofs_runqueue(&f, &pagepool,
Huang Jianan40452ff2021-12-06 22:35:52 +08001959 z_erofs_get_sync_decompress_policy(sbi, nr_pages));
Gao Xiang09c54372022-01-02 12:00:17 +08001960 erofs_put_metabuf(&f.map.buf);
Gao Xiangeaa91722021-10-22 17:01:20 +08001961 erofs_release_pages(&pagepool);
Gao Xiang3883a792018-07-26 20:22:06 +08001962}
1963
Gao Xiang0c638f72019-11-08 11:37:33 +08001964const struct address_space_operations z_erofs_aops = {
Matthew Wilcox (Oracle)a2e20a22022-04-29 11:12:16 -04001965 .read_folio = z_erofs_read_folio,
Matthew Wilcox (Oracle)06150902020-06-01 21:47:13 -07001966 .readahead = z_erofs_readahead,
Gao Xiang3883a792018-07-26 20:22:06 +08001967};