Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0-only
2 : /*
3 : * Copyright (C) 1993 Linus Torvalds
4 : * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
5 : * SMP-safe vmalloc/vfree/ioremap, Tigran Aivazian <tigran@veritas.com>, May 2000
6 : * Major rework to support vmap/vunmap, Christoph Hellwig, SGI, August 2002
7 : * Numa awareness, Christoph Lameter, SGI, June 2005
8 : * Improving global KVA allocator, Uladzislau Rezki, Sony, May 2019
9 : */
10 :
11 : #include <linux/vmalloc.h>
12 : #include <linux/mm.h>
13 : #include <linux/module.h>
14 : #include <linux/highmem.h>
15 : #include <linux/sched/signal.h>
16 : #include <linux/slab.h>
17 : #include <linux/spinlock.h>
18 : #include <linux/interrupt.h>
19 : #include <linux/proc_fs.h>
20 : #include <linux/seq_file.h>
21 : #include <linux/set_memory.h>
22 : #include <linux/debugobjects.h>
23 : #include <linux/kallsyms.h>
24 : #include <linux/list.h>
25 : #include <linux/notifier.h>
26 : #include <linux/rbtree.h>
27 : #include <linux/xarray.h>
28 : #include <linux/io.h>
29 : #include <linux/rcupdate.h>
30 : #include <linux/pfn.h>
31 : #include <linux/kmemleak.h>
32 : #include <linux/atomic.h>
33 : #include <linux/compiler.h>
34 : #include <linux/memcontrol.h>
35 : #include <linux/llist.h>
36 : #include <linux/bitops.h>
37 : #include <linux/rbtree_augmented.h>
38 : #include <linux/overflow.h>
39 : #include <linux/pgtable.h>
40 : #include <linux/uaccess.h>
41 : #include <linux/hugetlb.h>
42 : #include <linux/sched/mm.h>
43 : #include <asm/tlbflush.h>
44 : #include <asm/shmparam.h>
45 :
46 : #define CREATE_TRACE_POINTS
47 : #include <trace/events/vmalloc.h>
48 :
49 : #include "internal.h"
50 : #include "pgalloc-track.h"
51 :
52 : #ifdef CONFIG_HAVE_ARCH_HUGE_VMAP
53 : static unsigned int __ro_after_init ioremap_max_page_shift = BITS_PER_LONG - 1;
54 :
55 : static int __init set_nohugeiomap(char *str)
56 : {
57 : ioremap_max_page_shift = PAGE_SHIFT;
58 : return 0;
59 : }
60 : early_param("nohugeiomap", set_nohugeiomap);
61 : #else /* CONFIG_HAVE_ARCH_HUGE_VMAP */
62 : static const unsigned int ioremap_max_page_shift = PAGE_SHIFT;
63 : #endif /* CONFIG_HAVE_ARCH_HUGE_VMAP */
64 :
65 : #ifdef CONFIG_HAVE_ARCH_HUGE_VMALLOC
66 : static bool __ro_after_init vmap_allow_huge = true;
67 :
68 : static int __init set_nohugevmalloc(char *str)
69 : {
70 : vmap_allow_huge = false;
71 : return 0;
72 : }
73 : early_param("nohugevmalloc", set_nohugevmalloc);
74 : #else /* CONFIG_HAVE_ARCH_HUGE_VMALLOC */
75 : static const bool vmap_allow_huge = false;
76 : #endif /* CONFIG_HAVE_ARCH_HUGE_VMALLOC */
77 :
78 259 : bool is_vmalloc_addr(const void *x)
79 : {
80 264 : unsigned long addr = (unsigned long)kasan_reset_tag(x);
81 :
82 259 : return addr >= VMALLOC_START && addr < VMALLOC_END;
83 : }
84 : EXPORT_SYMBOL(is_vmalloc_addr);
85 :
86 : struct vfree_deferred {
87 : struct llist_head list;
88 : struct work_struct wq;
89 : };
90 : static DEFINE_PER_CPU(struct vfree_deferred, vfree_deferred);
91 :
92 : /*** Page table manipulation functions ***/
93 0 : static int vmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
94 : phys_addr_t phys_addr, pgprot_t prot,
95 : unsigned int max_page_shift, pgtbl_mod_mask *mask)
96 : {
97 : pte_t *pte;
98 : u64 pfn;
99 0 : unsigned long size = PAGE_SIZE;
100 :
101 0 : pfn = phys_addr >> PAGE_SHIFT;
102 0 : pte = pte_alloc_kernel_track(pmd, addr, mask);
103 0 : if (!pte)
104 : return -ENOMEM;
105 : do {
106 0 : BUG_ON(!pte_none(*pte));
107 :
108 : #ifdef CONFIG_HUGETLB_PAGE
109 : size = arch_vmap_pte_range_map_size(addr, end, pfn, max_page_shift);
110 : if (size != PAGE_SIZE) {
111 : pte_t entry = pfn_pte(pfn, prot);
112 :
113 : entry = arch_make_huge_pte(entry, ilog2(size), 0);
114 : set_huge_pte_at(&init_mm, addr, pte, entry);
115 : pfn += PFN_DOWN(size);
116 : continue;
117 : }
118 : #endif
119 0 : set_pte_at(&init_mm, addr, pte, pfn_pte(pfn, prot));
120 0 : pfn++;
121 0 : } while (pte += PFN_DOWN(size), addr += size, addr != end);
122 0 : *mask |= PGTBL_PTE_MODIFIED;
123 : return 0;
124 : }
125 :
126 : static int vmap_try_huge_pmd(pmd_t *pmd, unsigned long addr, unsigned long end,
127 : phys_addr_t phys_addr, pgprot_t prot,
128 : unsigned int max_page_shift)
129 : {
130 : if (max_page_shift < PMD_SHIFT)
131 : return 0;
132 :
133 : if (!arch_vmap_pmd_supported(prot))
134 : return 0;
135 :
136 : if ((end - addr) != PMD_SIZE)
137 : return 0;
138 :
139 : if (!IS_ALIGNED(addr, PMD_SIZE))
140 : return 0;
141 :
142 : if (!IS_ALIGNED(phys_addr, PMD_SIZE))
143 : return 0;
144 :
145 : if (pmd_present(*pmd) && !pmd_free_pte_page(pmd, addr))
146 : return 0;
147 :
148 : return pmd_set_huge(pmd, phys_addr, prot);
149 : }
150 :
151 0 : static int vmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end,
152 : phys_addr_t phys_addr, pgprot_t prot,
153 : unsigned int max_page_shift, pgtbl_mod_mask *mask)
154 : {
155 : pmd_t *pmd;
156 : unsigned long next;
157 :
158 0 : pmd = pmd_alloc_track(&init_mm, pud, addr, mask);
159 0 : if (!pmd)
160 : return -ENOMEM;
161 : do {
162 0 : next = pmd_addr_end(addr, end);
163 :
164 0 : if (vmap_try_huge_pmd(pmd, addr, next, phys_addr, prot,
165 : max_page_shift)) {
166 : *mask |= PGTBL_PMD_MODIFIED;
167 : continue;
168 : }
169 :
170 0 : if (vmap_pte_range(pmd, addr, next, phys_addr, prot, max_page_shift, mask))
171 : return -ENOMEM;
172 0 : } while (pmd++, phys_addr += (next - addr), addr = next, addr != end);
173 : return 0;
174 : }
175 :
176 : static int vmap_try_huge_pud(pud_t *pud, unsigned long addr, unsigned long end,
177 : phys_addr_t phys_addr, pgprot_t prot,
178 : unsigned int max_page_shift)
179 : {
180 : if (max_page_shift < PUD_SHIFT)
181 : return 0;
182 :
183 : if (!arch_vmap_pud_supported(prot))
184 : return 0;
185 :
186 : if ((end - addr) != PUD_SIZE)
187 : return 0;
188 :
189 : if (!IS_ALIGNED(addr, PUD_SIZE))
190 : return 0;
191 :
192 : if (!IS_ALIGNED(phys_addr, PUD_SIZE))
193 : return 0;
194 :
195 : if (pud_present(*pud) && !pud_free_pmd_page(pud, addr))
196 : return 0;
197 :
198 : return pud_set_huge(pud, phys_addr, prot);
199 : }
200 :
201 : static int vmap_pud_range(p4d_t *p4d, unsigned long addr, unsigned long end,
202 : phys_addr_t phys_addr, pgprot_t prot,
203 : unsigned int max_page_shift, pgtbl_mod_mask *mask)
204 : {
205 : pud_t *pud;
206 : unsigned long next;
207 :
208 0 : pud = pud_alloc_track(&init_mm, p4d, addr, mask);
209 : if (!pud)
210 : return -ENOMEM;
211 : do {
212 0 : next = pud_addr_end(addr, end);
213 :
214 0 : if (vmap_try_huge_pud(pud, addr, next, phys_addr, prot,
215 : max_page_shift)) {
216 : *mask |= PGTBL_PUD_MODIFIED;
217 : continue;
218 : }
219 :
220 0 : if (vmap_pmd_range(pud, addr, next, phys_addr, prot,
221 : max_page_shift, mask))
222 : return -ENOMEM;
223 0 : } while (pud++, phys_addr += (next - addr), addr = next, addr != end);
224 : return 0;
225 : }
226 :
227 : static int vmap_try_huge_p4d(p4d_t *p4d, unsigned long addr, unsigned long end,
228 : phys_addr_t phys_addr, pgprot_t prot,
229 : unsigned int max_page_shift)
230 : {
231 : if (max_page_shift < P4D_SHIFT)
232 : return 0;
233 :
234 : if (!arch_vmap_p4d_supported(prot))
235 : return 0;
236 :
237 : if ((end - addr) != P4D_SIZE)
238 : return 0;
239 :
240 : if (!IS_ALIGNED(addr, P4D_SIZE))
241 : return 0;
242 :
243 : if (!IS_ALIGNED(phys_addr, P4D_SIZE))
244 : return 0;
245 :
246 : if (p4d_present(*p4d) && !p4d_free_pud_page(p4d, addr))
247 : return 0;
248 :
249 : return p4d_set_huge(p4d, phys_addr, prot);
250 : }
251 :
252 0 : static int vmap_p4d_range(pgd_t *pgd, unsigned long addr, unsigned long end,
253 : phys_addr_t phys_addr, pgprot_t prot,
254 : unsigned int max_page_shift, pgtbl_mod_mask *mask)
255 : {
256 : p4d_t *p4d;
257 : unsigned long next;
258 :
259 0 : p4d = p4d_alloc_track(&init_mm, pgd, addr, mask);
260 0 : if (!p4d)
261 : return -ENOMEM;
262 : do {
263 0 : next = p4d_addr_end(addr, end);
264 :
265 0 : if (vmap_try_huge_p4d(p4d, addr, next, phys_addr, prot,
266 : max_page_shift)) {
267 : *mask |= PGTBL_P4D_MODIFIED;
268 : continue;
269 : }
270 :
271 0 : if (vmap_pud_range(p4d, addr, next, phys_addr, prot,
272 : max_page_shift, mask))
273 : return -ENOMEM;
274 0 : } while (p4d++, phys_addr += (next - addr), addr = next, addr != end);
275 : return 0;
276 : }
277 :
278 0 : static int vmap_range_noflush(unsigned long addr, unsigned long end,
279 : phys_addr_t phys_addr, pgprot_t prot,
280 : unsigned int max_page_shift)
281 : {
282 : pgd_t *pgd;
283 : unsigned long start;
284 : unsigned long next;
285 : int err;
286 0 : pgtbl_mod_mask mask = 0;
287 :
288 : might_sleep();
289 0 : BUG_ON(addr >= end);
290 :
291 0 : start = addr;
292 0 : pgd = pgd_offset_k(addr);
293 : do {
294 0 : next = pgd_addr_end(addr, end);
295 0 : err = vmap_p4d_range(pgd, addr, next, phys_addr, prot,
296 : max_page_shift, &mask);
297 0 : if (err)
298 : break;
299 0 : } while (pgd++, phys_addr += (next - addr), addr = next, addr != end);
300 :
301 : if (mask & ARCH_PAGE_TABLE_SYNC_MASK)
302 : arch_sync_kernel_mappings(start, end);
303 :
304 0 : return err;
305 : }
306 :
307 0 : int ioremap_page_range(unsigned long addr, unsigned long end,
308 : phys_addr_t phys_addr, pgprot_t prot)
309 : {
310 : int err;
311 :
312 0 : err = vmap_range_noflush(addr, end, phys_addr, pgprot_nx(prot),
313 : ioremap_max_page_shift);
314 0 : flush_cache_vmap(addr, end);
315 : if (!err)
316 : kmsan_ioremap_page_range(addr, end, phys_addr, prot,
317 : ioremap_max_page_shift);
318 0 : return err;
319 : }
320 :
321 337 : static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
322 : pgtbl_mod_mask *mask)
323 : {
324 : pte_t *pte;
325 :
326 337 : pte = pte_offset_kernel(pmd, addr);
327 : do {
328 85468 : pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
329 42734 : WARN_ON(!pte_none(ptent) && !pte_present(ptent));
330 42734 : } while (pte++, addr += PAGE_SIZE, addr != end);
331 337 : *mask |= PGTBL_PTE_MODIFIED;
332 337 : }
333 :
334 258 : static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end,
335 : pgtbl_mod_mask *mask)
336 : {
337 : pmd_t *pmd;
338 : unsigned long next;
339 : int cleared;
340 :
341 258 : pmd = pmd_offset(pud, addr);
342 : do {
343 338 : next = pmd_addr_end(addr, end);
344 :
345 338 : cleared = pmd_clear_huge(pmd);
346 338 : if (cleared || pmd_bad(*pmd))
347 1 : *mask |= PGTBL_PMD_MODIFIED;
348 :
349 : if (cleared)
350 : continue;
351 338 : if (pmd_none_or_clear_bad(pmd))
352 1 : continue;
353 337 : vunmap_pte_range(pmd, addr, next, mask);
354 :
355 337 : cond_resched();
356 338 : } while (pmd++, addr = next, addr != end);
357 258 : }
358 :
359 258 : static void vunmap_pud_range(p4d_t *p4d, unsigned long addr, unsigned long end,
360 : pgtbl_mod_mask *mask)
361 : {
362 : pud_t *pud;
363 : unsigned long next;
364 : int cleared;
365 :
366 258 : pud = pud_offset(p4d, addr);
367 : do {
368 258 : next = pud_addr_end(addr, end);
369 :
370 258 : cleared = pud_clear_huge(pud);
371 258 : if (cleared || pud_bad(*pud))
372 0 : *mask |= PGTBL_PUD_MODIFIED;
373 :
374 : if (cleared)
375 : continue;
376 258 : if (pud_none_or_clear_bad(pud))
377 0 : continue;
378 258 : vunmap_pmd_range(pud, addr, next, mask);
379 258 : } while (pud++, addr = next, addr != end);
380 258 : }
381 :
382 : static void vunmap_p4d_range(pgd_t *pgd, unsigned long addr, unsigned long end,
383 : pgtbl_mod_mask *mask)
384 : {
385 : p4d_t *p4d;
386 : unsigned long next;
387 :
388 258 : p4d = p4d_offset(pgd, addr);
389 : do {
390 258 : next = p4d_addr_end(addr, end);
391 :
392 258 : p4d_clear_huge(p4d);
393 258 : if (p4d_bad(*p4d))
394 : *mask |= PGTBL_P4D_MODIFIED;
395 :
396 258 : if (p4d_none_or_clear_bad(p4d))
397 : continue;
398 258 : vunmap_pud_range(p4d, addr, next, mask);
399 258 : } while (p4d++, addr = next, addr != end);
400 : }
401 :
402 : /*
403 : * vunmap_range_noflush is similar to vunmap_range, but does not
404 : * flush caches or TLBs.
405 : *
406 : * The caller is responsible for calling flush_cache_vmap() before calling
407 : * this function, and flush_tlb_kernel_range after it has returned
408 : * successfully (and before the addresses are expected to cause a page fault
409 : * or be re-mapped for something else, if TLB flushes are being delayed or
410 : * coalesced).
411 : *
412 : * This is an internal function only. Do not use outside mm/.
413 : */
414 258 : void __vunmap_range_noflush(unsigned long start, unsigned long end)
415 : {
416 : unsigned long next;
417 : pgd_t *pgd;
418 258 : unsigned long addr = start;
419 258 : pgtbl_mod_mask mask = 0;
420 :
421 258 : BUG_ON(addr >= end);
422 516 : pgd = pgd_offset_k(addr);
423 : do {
424 258 : next = pgd_addr_end(addr, end);
425 258 : if (pgd_bad(*pgd))
426 : mask |= PGTBL_PGD_MODIFIED;
427 258 : if (pgd_none_or_clear_bad(pgd))
428 : continue;
429 : vunmap_p4d_range(pgd, addr, next, &mask);
430 258 : } while (pgd++, addr = next, addr != end);
431 :
432 : if (mask & ARCH_PAGE_TABLE_SYNC_MASK)
433 : arch_sync_kernel_mappings(start, end);
434 258 : }
435 :
436 0 : void vunmap_range_noflush(unsigned long start, unsigned long end)
437 : {
438 258 : kmsan_vunmap_range_noflush(start, end);
439 258 : __vunmap_range_noflush(start, end);
440 0 : }
441 :
442 : /**
443 : * vunmap_range - unmap kernel virtual addresses
444 : * @addr: start of the VM area to unmap
445 : * @end: end of the VM area to unmap (non-inclusive)
446 : *
447 : * Clears any present PTEs in the virtual address range, flushes TLBs and
448 : * caches. Any subsequent access to the address before it has been re-mapped
449 : * is a kernel bug.
450 : */
451 0 : void vunmap_range(unsigned long addr, unsigned long end)
452 : {
453 0 : flush_cache_vunmap(addr, end);
454 0 : vunmap_range_noflush(addr, end);
455 0 : flush_tlb_kernel_range(addr, end);
456 0 : }
457 :
458 353 : static int vmap_pages_pte_range(pmd_t *pmd, unsigned long addr,
459 : unsigned long end, pgprot_t prot, struct page **pages, int *nr,
460 : pgtbl_mod_mask *mask)
461 : {
462 : pte_t *pte;
463 :
464 : /*
465 : * nr is a running index into the array which helps higher level
466 : * callers keep track of where we're up to.
467 : */
468 :
469 706 : pte = pte_alloc_kernel_track(pmd, addr, mask);
470 353 : if (!pte)
471 : return -ENOMEM;
472 : do {
473 42541 : struct page *page = pages[*nr];
474 :
475 42541 : if (WARN_ON(!pte_none(*pte)))
476 : return -EBUSY;
477 42541 : if (WARN_ON(!page))
478 : return -ENOMEM;
479 85082 : if (WARN_ON(!pfn_valid(page_to_pfn(page))))
480 : return -EINVAL;
481 :
482 85082 : set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
483 42541 : (*nr)++;
484 42541 : } while (pte++, addr += PAGE_SIZE, addr != end);
485 353 : *mask |= PGTBL_PTE_MODIFIED;
486 353 : return 0;
487 : }
488 :
489 274 : static int vmap_pages_pmd_range(pud_t *pud, unsigned long addr,
490 : unsigned long end, pgprot_t prot, struct page **pages, int *nr,
491 : pgtbl_mod_mask *mask)
492 : {
493 : pmd_t *pmd;
494 : unsigned long next;
495 :
496 274 : pmd = pmd_alloc_track(&init_mm, pud, addr, mask);
497 274 : if (!pmd)
498 : return -ENOMEM;
499 : do {
500 353 : next = pmd_addr_end(addr, end);
501 353 : if (vmap_pages_pte_range(pmd, addr, next, prot, pages, nr, mask))
502 : return -ENOMEM;
503 353 : } while (pmd++, addr = next, addr != end);
504 : return 0;
505 : }
506 :
507 : static int vmap_pages_pud_range(p4d_t *p4d, unsigned long addr,
508 : unsigned long end, pgprot_t prot, struct page **pages, int *nr,
509 : pgtbl_mod_mask *mask)
510 : {
511 : pud_t *pud;
512 : unsigned long next;
513 :
514 548 : pud = pud_alloc_track(&init_mm, p4d, addr, mask);
515 : if (!pud)
516 : return -ENOMEM;
517 : do {
518 274 : next = pud_addr_end(addr, end);
519 274 : if (vmap_pages_pmd_range(pud, addr, next, prot, pages, nr, mask))
520 : return -ENOMEM;
521 274 : } while (pud++, addr = next, addr != end);
522 : return 0;
523 : }
524 :
525 274 : static int vmap_pages_p4d_range(pgd_t *pgd, unsigned long addr,
526 : unsigned long end, pgprot_t prot, struct page **pages, int *nr,
527 : pgtbl_mod_mask *mask)
528 : {
529 : p4d_t *p4d;
530 : unsigned long next;
531 :
532 548 : p4d = p4d_alloc_track(&init_mm, pgd, addr, mask);
533 274 : if (!p4d)
534 : return -ENOMEM;
535 : do {
536 274 : next = p4d_addr_end(addr, end);
537 274 : if (vmap_pages_pud_range(p4d, addr, next, prot, pages, nr, mask))
538 : return -ENOMEM;
539 274 : } while (p4d++, addr = next, addr != end);
540 274 : return 0;
541 : }
542 :
543 274 : static int vmap_small_pages_range_noflush(unsigned long addr, unsigned long end,
544 : pgprot_t prot, struct page **pages)
545 : {
546 274 : unsigned long start = addr;
547 : pgd_t *pgd;
548 : unsigned long next;
549 274 : int err = 0;
550 274 : int nr = 0;
551 274 : pgtbl_mod_mask mask = 0;
552 :
553 274 : BUG_ON(addr >= end);
554 548 : pgd = pgd_offset_k(addr);
555 : do {
556 274 : next = pgd_addr_end(addr, end);
557 274 : if (pgd_bad(*pgd))
558 : mask |= PGTBL_PGD_MODIFIED;
559 274 : err = vmap_pages_p4d_range(pgd, addr, next, prot, pages, &nr, &mask);
560 274 : if (err)
561 : return err;
562 274 : } while (pgd++, addr = next, addr != end);
563 :
564 : if (mask & ARCH_PAGE_TABLE_SYNC_MASK)
565 : arch_sync_kernel_mappings(start, end);
566 :
567 : return 0;
568 : }
569 :
570 : /*
571 : * vmap_pages_range_noflush is similar to vmap_pages_range, but does not
572 : * flush caches.
573 : *
574 : * The caller is responsible for calling flush_cache_vmap() after this
575 : * function returns successfully and before the addresses are accessed.
576 : *
577 : * This is an internal function only. Do not use outside mm/.
578 : */
579 274 : int __vmap_pages_range_noflush(unsigned long addr, unsigned long end,
580 : pgprot_t prot, struct page **pages, unsigned int page_shift)
581 : {
582 274 : unsigned int i, nr = (end - addr) >> PAGE_SHIFT;
583 :
584 274 : WARN_ON(page_shift < PAGE_SHIFT);
585 :
586 : if (!IS_ENABLED(CONFIG_HAVE_ARCH_HUGE_VMALLOC) ||
587 : page_shift == PAGE_SHIFT)
588 274 : return vmap_small_pages_range_noflush(addr, end, prot, pages);
589 :
590 : for (i = 0; i < nr; i += 1U << (page_shift - PAGE_SHIFT)) {
591 : int err;
592 :
593 : err = vmap_range_noflush(addr, addr + (1UL << page_shift),
594 : page_to_phys(pages[i]), prot,
595 : page_shift);
596 : if (err)
597 : return err;
598 :
599 : addr += 1UL << page_shift;
600 : }
601 :
602 : return 0;
603 : }
604 :
605 0 : int vmap_pages_range_noflush(unsigned long addr, unsigned long end,
606 : pgprot_t prot, struct page **pages, unsigned int page_shift)
607 : {
608 274 : kmsan_vmap_pages_range_noflush(addr, end, prot, pages, page_shift);
609 274 : return __vmap_pages_range_noflush(addr, end, prot, pages, page_shift);
610 : }
611 :
612 : /**
613 : * vmap_pages_range - map pages to a kernel virtual address
614 : * @addr: start of the VM area to map
615 : * @end: end of the VM area to map (non-inclusive)
616 : * @prot: page protection flags to use
617 : * @pages: pages to map (always PAGE_SIZE pages)
618 : * @page_shift: maximum shift that the pages may be mapped with, @pages must
619 : * be aligned and contiguous up to at least this shift.
620 : *
621 : * RETURNS:
622 : * 0 on success, -errno on failure.
623 : */
624 274 : static int vmap_pages_range(unsigned long addr, unsigned long end,
625 : pgprot_t prot, struct page **pages, unsigned int page_shift)
626 : {
627 : int err;
628 :
629 274 : err = vmap_pages_range_noflush(addr, end, prot, pages, page_shift);
630 274 : flush_cache_vmap(addr, end);
631 274 : return err;
632 : }
633 :
634 0 : int is_vmalloc_or_module_addr(const void *x)
635 : {
636 : /*
637 : * ARM, x86-64 and sparc64 put modules in a special place,
638 : * and fall back on vmalloc() if that fails. Others
639 : * just put it in the vmalloc space.
640 : */
641 : #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
642 : unsigned long addr = (unsigned long)kasan_reset_tag(x);
643 : if (addr >= MODULES_VADDR && addr < MODULES_END)
644 : return 1;
645 : #endif
646 5 : return is_vmalloc_addr(x);
647 : }
648 : EXPORT_SYMBOL_GPL(is_vmalloc_or_module_addr);
649 :
650 : /*
651 : * Walk a vmap address to the struct page it maps. Huge vmap mappings will
652 : * return the tail page that corresponds to the base page address, which
653 : * matches small vmap mappings.
654 : */
655 0 : struct page *vmalloc_to_page(const void *vmalloc_addr)
656 : {
657 0 : unsigned long addr = (unsigned long) vmalloc_addr;
658 0 : struct page *page = NULL;
659 0 : pgd_t *pgd = pgd_offset_k(addr);
660 : p4d_t *p4d;
661 : pud_t *pud;
662 : pmd_t *pmd;
663 : pte_t *ptep, pte;
664 :
665 : /*
666 : * XXX we might need to change this if we add VIRTUAL_BUG_ON for
667 : * architectures that do not vmalloc module space
668 : */
669 : VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr));
670 :
671 : if (pgd_none(*pgd))
672 : return NULL;
673 0 : if (WARN_ON_ONCE(pgd_leaf(*pgd)))
674 : return NULL; /* XXX: no allowance for huge pgd */
675 0 : if (WARN_ON_ONCE(pgd_bad(*pgd)))
676 : return NULL;
677 :
678 0 : p4d = p4d_offset(pgd, addr);
679 : if (p4d_none(*p4d))
680 : return NULL;
681 : if (p4d_leaf(*p4d))
682 : return p4d_page(*p4d) + ((addr & ~P4D_MASK) >> PAGE_SHIFT);
683 0 : if (WARN_ON_ONCE(p4d_bad(*p4d)))
684 : return NULL;
685 :
686 0 : pud = pud_offset(p4d, addr);
687 0 : if (pud_none(*pud))
688 : return NULL;
689 : if (pud_leaf(*pud))
690 : return pud_page(*pud) + ((addr & ~PUD_MASK) >> PAGE_SHIFT);
691 0 : if (WARN_ON_ONCE(pud_bad(*pud)))
692 : return NULL;
693 :
694 0 : pmd = pmd_offset(pud, addr);
695 0 : if (pmd_none(*pmd))
696 : return NULL;
697 : if (pmd_leaf(*pmd))
698 : return pmd_page(*pmd) + ((addr & ~PMD_MASK) >> PAGE_SHIFT);
699 0 : if (WARN_ON_ONCE(pmd_bad(*pmd)))
700 : return NULL;
701 :
702 0 : ptep = pte_offset_map(pmd, addr);
703 0 : pte = *ptep;
704 0 : if (pte_present(pte))
705 0 : page = pte_page(pte);
706 : pte_unmap(ptep);
707 :
708 : return page;
709 : }
710 : EXPORT_SYMBOL(vmalloc_to_page);
711 :
712 : /*
713 : * Map a vmalloc()-space virtual address to the physical page frame number.
714 : */
715 0 : unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
716 : {
717 0 : return page_to_pfn(vmalloc_to_page(vmalloc_addr));
718 : }
719 : EXPORT_SYMBOL(vmalloc_to_pfn);
720 :
721 :
722 : /*** Global kva allocator ***/
723 :
724 : #define DEBUG_AUGMENT_PROPAGATE_CHECK 0
725 : #define DEBUG_AUGMENT_LOWEST_MATCH_CHECK 0
726 :
727 :
728 : static DEFINE_SPINLOCK(vmap_area_lock);
729 : static DEFINE_SPINLOCK(free_vmap_area_lock);
730 : /* Export for kexec only */
731 : LIST_HEAD(vmap_area_list);
732 : static struct rb_root vmap_area_root = RB_ROOT;
733 : static bool vmap_initialized __read_mostly;
734 :
735 : static struct rb_root purge_vmap_area_root = RB_ROOT;
736 : static LIST_HEAD(purge_vmap_area_list);
737 : static DEFINE_SPINLOCK(purge_vmap_area_lock);
738 :
739 : /*
740 : * This kmem_cache is used for vmap_area objects. Instead of
741 : * allocating from slab we reuse an object from this cache to
742 : * make things faster. Especially in "no edge" splitting of
743 : * free block.
744 : */
745 : static struct kmem_cache *vmap_area_cachep;
746 :
747 : /*
748 : * This linked list is used in pair with free_vmap_area_root.
749 : * It gives O(1) access to prev/next to perform fast coalescing.
750 : */
751 : static LIST_HEAD(free_vmap_area_list);
752 :
753 : /*
754 : * This augment red-black tree represents the free vmap space.
755 : * All vmap_area objects in this tree are sorted by va->va_start
756 : * address. It is used for allocation and merging when a vmap
757 : * object is released.
758 : *
759 : * Each vmap_area node contains a maximum available free block
760 : * of its sub-tree, right or left. Therefore it is possible to
761 : * find a lowest match of free area.
762 : */
763 : static struct rb_root free_vmap_area_root = RB_ROOT;
764 :
765 : /*
766 : * Preload a CPU with one object for "no edge" split case. The
767 : * aim is to get rid of allocations from the atomic context, thus
768 : * to use more permissive allocation masks.
769 : */
770 : static DEFINE_PER_CPU(struct vmap_area *, ne_fit_preload_node);
771 :
772 : static __always_inline unsigned long
773 : va_size(struct vmap_area *va)
774 : {
775 1686 : return (va->va_end - va->va_start);
776 : }
777 :
778 : static __always_inline unsigned long
779 : get_subtree_max_size(struct rb_node *node)
780 : {
781 : struct vmap_area *va;
782 :
783 4016 : va = rb_entry_safe(node, struct vmap_area, rb_node);
784 4016 : return va ? va->subtree_max_size : 0;
785 : }
786 :
787 2001 : RB_DECLARE_CALLBACKS_MAX(static, free_vmap_area_rb_augment_cb,
788 : struct vmap_area, rb_node, unsigned long, subtree_max_size, va_size)
789 :
790 : static void purge_vmap_area_lazy(void);
791 : static BLOCKING_NOTIFIER_HEAD(vmap_notify_list);
792 : static void drain_vmap_area_work(struct work_struct *work);
793 : static DECLARE_WORK(drain_vmap_work, drain_vmap_area_work);
794 :
795 : static atomic_long_t nr_vmalloc_pages;
796 :
797 0 : unsigned long vmalloc_nr_pages(void)
798 : {
799 0 : return atomic_long_read(&nr_vmalloc_pages);
800 : }
801 :
802 : /* Look up the first VA which satisfies addr < va_end, NULL if none. */
803 : static struct vmap_area *find_vmap_area_exceed_addr(unsigned long addr)
804 : {
805 0 : struct vmap_area *va = NULL;
806 0 : struct rb_node *n = vmap_area_root.rb_node;
807 :
808 0 : addr = (unsigned long)kasan_reset_tag((void *)addr);
809 :
810 0 : while (n) {
811 : struct vmap_area *tmp;
812 :
813 0 : tmp = rb_entry(n, struct vmap_area, rb_node);
814 0 : if (tmp->va_end > addr) {
815 0 : va = tmp;
816 0 : if (tmp->va_start <= addr)
817 : break;
818 :
819 0 : n = n->rb_left;
820 : } else
821 0 : n = n->rb_right;
822 : }
823 :
824 : return va;
825 : }
826 :
827 : static struct vmap_area *__find_vmap_area(unsigned long addr, struct rb_root *root)
828 : {
829 274 : struct rb_node *n = root->rb_node;
830 :
831 274 : addr = (unsigned long)kasan_reset_tag((void *)addr);
832 :
833 1604 : while (n) {
834 : struct vmap_area *va;
835 :
836 1604 : va = rb_entry(n, struct vmap_area, rb_node);
837 1604 : if (addr < va->va_start)
838 0 : n = n->rb_left;
839 1604 : else if (addr >= va->va_end)
840 1330 : n = n->rb_right;
841 : else
842 : return va;
843 : }
844 :
845 : return NULL;
846 : }
847 :
848 : /*
849 : * This function returns back addresses of parent node
850 : * and its left or right link for further processing.
851 : *
852 : * Otherwise NULL is returned. In that case all further
853 : * steps regarding inserting of conflicting overlap range
854 : * have to be declined and actually considered as a bug.
855 : */
856 : static __always_inline struct rb_node **
857 : find_va_links(struct vmap_area *va,
858 : struct rb_root *root, struct rb_node *from,
859 : struct rb_node **parent)
860 : {
861 : struct vmap_area *tmp_va;
862 : struct rb_node **link;
863 :
864 275 : if (root) {
865 538 : link = &root->rb_node;
866 538 : if (unlikely(!*link)) {
867 : *parent = NULL;
868 : return link;
869 : }
870 : } else {
871 : link = &from;
872 : }
873 :
874 : /*
875 : * Go to the bottom of the tree. When we hit the last point
876 : * we end up with parent rb_node and correct direction, i name
877 : * it link, where the new va->rb_node will be attached to.
878 : */
879 : do {
880 1650 : tmp_va = rb_entry(*link, struct vmap_area, rb_node);
881 :
882 : /*
883 : * During the traversal we also do some sanity check.
884 : * Trigger the BUG() if there are sides(left/right)
885 : * or full overlaps.
886 : */
887 1650 : if (va->va_end <= tmp_va->va_start)
888 21 : link = &(*link)->rb_left;
889 1629 : else if (va->va_start >= tmp_va->va_end)
890 1629 : link = &(*link)->rb_right;
891 : else {
892 0 : WARN(1, "vmalloc bug: 0x%lx-0x%lx overlaps with 0x%lx-0x%lx\n",
893 : va->va_start, va->va_end, tmp_va->va_start, tmp_va->va_end);
894 :
895 : return NULL;
896 : }
897 1650 : } while (*link);
898 :
899 546 : *parent = &tmp_va->rb_node;
900 : return link;
901 : }
902 :
903 : static __always_inline struct list_head *
904 : get_va_next_sibling(struct rb_node *parent, struct rb_node **link)
905 : {
906 : struct list_head *list;
907 :
908 263 : if (unlikely(!parent))
909 : /*
910 : * The red-black tree where we try to find VA neighbors
911 : * before merging or inserting is empty, i.e. it means
912 : * there is no free vmap space. Normally it does not
913 : * happen but we handle this case anyway.
914 : */
915 : return NULL;
916 :
917 257 : list = &rb_entry(parent, struct vmap_area, rb_node)->list;
918 257 : return (&parent->rb_right == link ? list->next : list);
919 : }
920 :
921 : static __always_inline void
922 : __link_va(struct vmap_area *va, struct rb_root *root,
923 : struct rb_node *parent, struct rb_node **link,
924 : struct list_head *head, bool augment)
925 : {
926 : /*
927 : * VA is still not in the list, but we can
928 : * identify its future previous list_head node.
929 : */
930 297 : if (likely(parent)) {
931 289 : head = &rb_entry(parent, struct vmap_area, rb_node)->list;
932 289 : if (&parent->rb_right != link)
933 8 : head = head->prev;
934 : }
935 :
936 : /* Insert to the rb-tree */
937 594 : rb_link_node(&va->rb_node, parent, link);
938 : if (augment) {
939 : /*
940 : * Some explanation here. Just perform simple insertion
941 : * to the tree. We do not set va->subtree_max_size to
942 : * its current size before calling rb_insert_augmented().
943 : * It is because we populate the tree from the bottom
944 : * to parent levels when the node _is_ in the tree.
945 : *
946 : * Therefore we set subtree_max_size to zero after insertion,
947 : * to let __augment_tree_propagate_from() puts everything to
948 : * the correct order later on.
949 : */
950 17 : rb_insert_augmented(&va->rb_node,
951 : root, &free_vmap_area_rb_augment_cb);
952 17 : va->subtree_max_size = 0;
953 : } else {
954 280 : rb_insert_color(&va->rb_node, root);
955 : }
956 :
957 : /* Address-sort this list */
958 314 : list_add(&va->list, head);
959 : }
960 :
961 : static __always_inline void
962 : link_va(struct vmap_area *va, struct rb_root *root,
963 : struct rb_node *parent, struct rb_node **link,
964 : struct list_head *head)
965 : {
966 : __link_va(va, root, parent, link, head, false);
967 : }
968 :
969 : static __always_inline void
970 : link_va_augment(struct vmap_area *va, struct rb_root *root,
971 : struct rb_node *parent, struct rb_node **link,
972 : struct list_head *head)
973 : {
974 17 : __link_va(va, root, parent, link, head, true);
975 : }
976 :
977 : static __always_inline void
978 : __unlink_va(struct vmap_area *va, struct rb_root *root, bool augment)
979 : {
980 258 : if (WARN_ON(RB_EMPTY_NODE(&va->rb_node)))
981 : return;
982 :
983 : if (augment)
984 0 : rb_erase_augmented(&va->rb_node,
985 : root, &free_vmap_area_rb_augment_cb);
986 : else
987 258 : rb_erase(&va->rb_node, root);
988 :
989 516 : list_del_init(&va->list);
990 258 : RB_CLEAR_NODE(&va->rb_node);
991 : }
992 :
993 : static __always_inline void
994 : unlink_va(struct vmap_area *va, struct rb_root *root)
995 : {
996 0 : __unlink_va(va, root, false);
997 : }
998 :
999 : static __always_inline void
1000 : unlink_va_augment(struct vmap_area *va, struct rb_root *root)
1001 : {
1002 0 : __unlink_va(va, root, true);
1003 : }
1004 :
1005 : #if DEBUG_AUGMENT_PROPAGATE_CHECK
1006 : /*
1007 : * Gets called when remove the node and rotate.
1008 : */
1009 : static __always_inline unsigned long
1010 : compute_subtree_max_size(struct vmap_area *va)
1011 : {
1012 : return max3(va_size(va),
1013 : get_subtree_max_size(va->rb_node.rb_left),
1014 : get_subtree_max_size(va->rb_node.rb_right));
1015 : }
1016 :
1017 : static void
1018 : augment_tree_propagate_check(void)
1019 : {
1020 : struct vmap_area *va;
1021 : unsigned long computed_size;
1022 :
1023 : list_for_each_entry(va, &free_vmap_area_list, list) {
1024 : computed_size = compute_subtree_max_size(va);
1025 : if (computed_size != va->subtree_max_size)
1026 : pr_emerg("tree is corrupted: %lu, %lu\n",
1027 : va_size(va), va->subtree_max_size);
1028 : }
1029 : }
1030 : #endif
1031 :
1032 : /*
1033 : * This function populates subtree_max_size from bottom to upper
1034 : * levels starting from VA point. The propagation must be done
1035 : * when VA size is modified by changing its va_start/va_end. Or
1036 : * in case of newly inserting of VA to the tree.
1037 : *
1038 : * It means that __augment_tree_propagate_from() must be called:
1039 : * - After VA has been inserted to the tree(free path);
1040 : * - After VA has been shrunk(allocation path);
1041 : * - After VA has been increased(merging path).
1042 : *
1043 : * Please note that, it does not mean that upper parent nodes
1044 : * and their subtree_max_size are recalculated all the time up
1045 : * to the root node.
1046 : *
1047 : * 4--8
1048 : * /\
1049 : * / \
1050 : * / \
1051 : * 2--2 8--8
1052 : *
1053 : * For example if we modify the node 4, shrinking it to 2, then
1054 : * no any modification is required. If we shrink the node 2 to 1
1055 : * its subtree_max_size is updated only, and set to 1. If we shrink
1056 : * the node 8 to 6, then its subtree_max_size is set to 6 and parent
1057 : * node becomes 4--6.
1058 : */
1059 : static __always_inline void
1060 : augment_tree_propagate_from(struct vmap_area *va)
1061 : {
1062 : /*
1063 : * Populate the tree from bottom towards the root until
1064 : * the calculated maximum available size of checked node
1065 : * is equal to its current one.
1066 : */
1067 296 : free_vmap_area_rb_augment_cb_propagate(&va->rb_node, NULL);
1068 :
1069 : #if DEBUG_AUGMENT_PROPAGATE_CHECK
1070 : augment_tree_propagate_check();
1071 : #endif
1072 : }
1073 :
1074 : static void
1075 274 : insert_vmap_area(struct vmap_area *va,
1076 : struct rb_root *root, struct list_head *head)
1077 : {
1078 : struct rb_node **link;
1079 : struct rb_node *parent;
1080 :
1081 274 : link = find_va_links(va, root, NULL, &parent);
1082 274 : if (link)
1083 274 : link_va(va, root, parent, link, head);
1084 274 : }
1085 :
1086 : static void
1087 17 : insert_vmap_area_augment(struct vmap_area *va,
1088 : struct rb_node *from, struct rb_root *root,
1089 : struct list_head *head)
1090 : {
1091 : struct rb_node **link;
1092 : struct rb_node *parent;
1093 :
1094 17 : if (from)
1095 : link = find_va_links(va, NULL, from, &parent);
1096 : else
1097 : link = find_va_links(va, root, NULL, &parent);
1098 :
1099 17 : if (link) {
1100 34 : link_va_augment(va, root, parent, link, head);
1101 : augment_tree_propagate_from(va);
1102 : }
1103 17 : }
1104 :
1105 : /*
1106 : * Merge de-allocated chunk of VA memory with previous
1107 : * and next free blocks. If coalesce is not done a new
1108 : * free area is inserted. If VA has been merged, it is
1109 : * freed.
1110 : *
1111 : * Please note, it can return NULL in case of overlap
1112 : * ranges, followed by WARN() report. Despite it is a
1113 : * buggy behaviour, a system can be alive and keep
1114 : * ongoing.
1115 : */
1116 : static __always_inline struct vmap_area *
1117 : __merge_or_add_vmap_area(struct vmap_area *va,
1118 : struct rb_root *root, struct list_head *head, bool augment)
1119 : {
1120 : struct vmap_area *sibling;
1121 : struct list_head *next;
1122 : struct rb_node **link;
1123 : struct rb_node *parent;
1124 263 : bool merged = false;
1125 :
1126 : /*
1127 : * Find a place in the tree where VA potentially will be
1128 : * inserted, unless it is merged with its sibling/siblings.
1129 : */
1130 263 : link = find_va_links(va, root, NULL, &parent);
1131 263 : if (!link)
1132 : return NULL;
1133 :
1134 : /*
1135 : * Get next node of VA to check if merging can be done.
1136 : */
1137 526 : next = get_va_next_sibling(parent, link);
1138 263 : if (unlikely(next == NULL))
1139 : goto insert;
1140 :
1141 : /*
1142 : * start end
1143 : * | |
1144 : * |<------VA------>|<-----Next----->|
1145 : * | |
1146 : * start end
1147 : */
1148 257 : if (next != head) {
1149 5 : sibling = list_entry(next, struct vmap_area, list);
1150 5 : if (sibling->va_start == va->va_end) {
1151 5 : sibling->va_start = va->va_start;
1152 :
1153 : /* Free vmap_area object. */
1154 5 : kmem_cache_free(vmap_area_cachep, va);
1155 :
1156 : /* Point to the new merged area. */
1157 5 : va = sibling;
1158 5 : merged = true;
1159 : }
1160 : }
1161 :
1162 : /*
1163 : * start end
1164 : * | |
1165 : * |<-----Prev----->|<------VA------>|
1166 : * | |
1167 : * start end
1168 : */
1169 257 : if (next->prev != head) {
1170 257 : sibling = list_entry(next->prev, struct vmap_area, list);
1171 257 : if (sibling->va_end == va->va_start) {
1172 : /*
1173 : * If both neighbors are coalesced, it is important
1174 : * to unlink the "next" node first, followed by merging
1175 : * with "previous" one. Otherwise the tree might not be
1176 : * fully populated if a sibling's augmented value is
1177 : * "normalized" because of rotation operations.
1178 : */
1179 252 : if (merged)
1180 0 : __unlink_va(va, root, augment);
1181 :
1182 252 : sibling->va_end = va->va_end;
1183 :
1184 : /* Free vmap_area object. */
1185 252 : kmem_cache_free(vmap_area_cachep, va);
1186 :
1187 : /* Point to the new merged area. */
1188 252 : va = sibling;
1189 252 : merged = true;
1190 : }
1191 : }
1192 :
1193 : insert:
1194 263 : if (!merged)
1195 6 : __link_va(va, root, parent, link, head, augment);
1196 :
1197 : return va;
1198 : }
1199 :
1200 : static __always_inline struct vmap_area *
1201 : merge_or_add_vmap_area(struct vmap_area *va,
1202 : struct rb_root *root, struct list_head *head)
1203 : {
1204 258 : return __merge_or_add_vmap_area(va, root, head, false);
1205 : }
1206 :
1207 : static __always_inline struct vmap_area *
1208 : merge_or_add_vmap_area_augment(struct vmap_area *va,
1209 : struct rb_root *root, struct list_head *head)
1210 : {
1211 5 : va = __merge_or_add_vmap_area(va, root, head, true);
1212 5 : if (va)
1213 : augment_tree_propagate_from(va);
1214 :
1215 : return va;
1216 : }
1217 :
1218 : static __always_inline bool
1219 : is_within_this_va(struct vmap_area *va, unsigned long size,
1220 : unsigned long align, unsigned long vstart)
1221 : {
1222 : unsigned long nva_start_addr;
1223 :
1224 2145 : if (va->va_start > vstart)
1225 1871 : nva_start_addr = ALIGN(va->va_start, align);
1226 : else
1227 274 : nva_start_addr = ALIGN(vstart, align);
1228 :
1229 : /* Can be overflowed due to big size or alignment. */
1230 2145 : if (nva_start_addr + size < nva_start_addr ||
1231 : nva_start_addr < vstart)
1232 : return false;
1233 :
1234 2145 : return (nva_start_addr + size <= va->va_end);
1235 : }
1236 :
1237 : /*
1238 : * Find the first free block(lowest start address) in the tree,
1239 : * that will accomplish the request corresponding to passing
1240 : * parameters. Please note, with an alignment bigger than PAGE_SIZE,
1241 : * a search length is adjusted to account for worst case alignment
1242 : * overhead.
1243 : */
1244 : static __always_inline struct vmap_area *
1245 : find_vmap_lowest_match(struct rb_root *root, unsigned long size,
1246 : unsigned long align, unsigned long vstart, bool adjust_search_size)
1247 : {
1248 : struct vmap_area *va;
1249 : struct rb_node *node;
1250 : unsigned long length;
1251 :
1252 : /* Start from the root. */
1253 274 : node = root->rb_node;
1254 :
1255 : /* Adjust the search size for alignment overhead. */
1256 274 : length = adjust_search_size ? size + align - 1 : size;
1257 :
1258 2145 : while (node) {
1259 2145 : va = rb_entry(node, struct vmap_area, rb_node);
1260 :
1261 4830 : if (get_subtree_max_size(node->rb_left) >= length &&
1262 540 : vstart < va->va_start) {
1263 : node = node->rb_left;
1264 : } else {
1265 1605 : if (is_within_this_va(va, size, align, vstart))
1266 : return va;
1267 :
1268 : /*
1269 : * Does not make sense to go deeper towards the right
1270 : * sub-tree if it does not have a free block that is
1271 : * equal or bigger to the requested search length.
1272 : */
1273 2664 : if (get_subtree_max_size(node->rb_right) >= length) {
1274 1059 : node = node->rb_right;
1275 1059 : continue;
1276 : }
1277 :
1278 : /*
1279 : * OK. We roll back and find the first right sub-tree,
1280 : * that will satisfy the search criteria. It can happen
1281 : * due to "vstart" restriction or an alignment overhead
1282 : * that is bigger then PAGE_SIZE.
1283 : */
1284 540 : while ((node = rb_parent(node))) {
1285 540 : va = rb_entry(node, struct vmap_area, rb_node);
1286 540 : if (is_within_this_va(va, size, align, vstart))
1287 : return va;
1288 :
1289 1078 : if (get_subtree_max_size(node->rb_right) >= length &&
1290 : vstart <= va->va_start) {
1291 : /*
1292 : * Shift the vstart forward. Please note, we update it with
1293 : * parent's start address adding "1" because we do not want
1294 : * to enter same sub-tree after it has already been checked
1295 : * and no suitable free block found there.
1296 : */
1297 272 : vstart = va->va_start + 1;
1298 272 : node = node->rb_right;
1299 : break;
1300 : }
1301 : }
1302 : }
1303 : }
1304 :
1305 : return NULL;
1306 : }
1307 :
1308 : #if DEBUG_AUGMENT_LOWEST_MATCH_CHECK
1309 : #include <linux/random.h>
1310 :
1311 : static struct vmap_area *
1312 : find_vmap_lowest_linear_match(struct list_head *head, unsigned long size,
1313 : unsigned long align, unsigned long vstart)
1314 : {
1315 : struct vmap_area *va;
1316 :
1317 : list_for_each_entry(va, head, list) {
1318 : if (!is_within_this_va(va, size, align, vstart))
1319 : continue;
1320 :
1321 : return va;
1322 : }
1323 :
1324 : return NULL;
1325 : }
1326 :
1327 : static void
1328 : find_vmap_lowest_match_check(struct rb_root *root, struct list_head *head,
1329 : unsigned long size, unsigned long align)
1330 : {
1331 : struct vmap_area *va_1, *va_2;
1332 : unsigned long vstart;
1333 : unsigned int rnd;
1334 :
1335 : get_random_bytes(&rnd, sizeof(rnd));
1336 : vstart = VMALLOC_START + rnd;
1337 :
1338 : va_1 = find_vmap_lowest_match(root, size, align, vstart, false);
1339 : va_2 = find_vmap_lowest_linear_match(head, size, align, vstart);
1340 :
1341 : if (va_1 != va_2)
1342 : pr_emerg("not lowest: t: 0x%p, l: 0x%p, v: 0x%lx\n",
1343 : va_1, va_2, vstart);
1344 : }
1345 : #endif
1346 :
1347 : enum fit_type {
1348 : NOTHING_FIT = 0,
1349 : FL_FIT_TYPE = 1, /* full fit */
1350 : LE_FIT_TYPE = 2, /* left edge fit */
1351 : RE_FIT_TYPE = 3, /* right edge fit */
1352 : NE_FIT_TYPE = 4 /* no edge fit */
1353 : };
1354 :
1355 : static __always_inline enum fit_type
1356 : classify_va_fit_type(struct vmap_area *va,
1357 : unsigned long nva_start_addr, unsigned long size)
1358 : {
1359 : enum fit_type type;
1360 :
1361 : /* Check if it is within VA. */
1362 548 : if (nva_start_addr < va->va_start ||
1363 274 : nva_start_addr + size > va->va_end)
1364 : return NOTHING_FIT;
1365 :
1366 : /* Now classify. */
1367 274 : if (va->va_start == nva_start_addr) {
1368 258 : if (va->va_end == nva_start_addr + size)
1369 : type = FL_FIT_TYPE;
1370 : else
1371 258 : type = LE_FIT_TYPE;
1372 16 : } else if (va->va_end == nva_start_addr + size) {
1373 : type = RE_FIT_TYPE;
1374 : } else {
1375 16 : type = NE_FIT_TYPE;
1376 : }
1377 :
1378 : return type;
1379 : }
1380 :
1381 : static __always_inline int
1382 : adjust_va_to_fit_type(struct rb_root *root, struct list_head *head,
1383 : struct vmap_area *va, unsigned long nva_start_addr,
1384 : unsigned long size)
1385 : {
1386 274 : struct vmap_area *lva = NULL;
1387 274 : enum fit_type type = classify_va_fit_type(va, nva_start_addr, size);
1388 :
1389 274 : if (type == FL_FIT_TYPE) {
1390 : /*
1391 : * No need to split VA, it fully fits.
1392 : *
1393 : * | |
1394 : * V NVA V
1395 : * |---------------|
1396 : */
1397 0 : unlink_va_augment(va, root);
1398 0 : kmem_cache_free(vmap_area_cachep, va);
1399 274 : } else if (type == LE_FIT_TYPE) {
1400 : /*
1401 : * Split left edge of fit VA.
1402 : *
1403 : * | |
1404 : * V NVA V R
1405 : * |-------|-------|
1406 : */
1407 258 : va->va_start += size;
1408 16 : } else if (type == RE_FIT_TYPE) {
1409 : /*
1410 : * Split right edge of fit VA.
1411 : *
1412 : * | |
1413 : * L V NVA V
1414 : * |-------|-------|
1415 : */
1416 0 : va->va_end = nva_start_addr;
1417 16 : } else if (type == NE_FIT_TYPE) {
1418 : /*
1419 : * Split no edge of fit VA.
1420 : *
1421 : * | |
1422 : * L V NVA V R
1423 : * |---|-------|---|
1424 : */
1425 16 : lva = __this_cpu_xchg(ne_fit_preload_node, NULL);
1426 16 : if (unlikely(!lva)) {
1427 : /*
1428 : * For percpu allocator we do not do any pre-allocation
1429 : * and leave it as it is. The reason is it most likely
1430 : * never ends up with NE_FIT_TYPE splitting. In case of
1431 : * percpu allocations offsets and sizes are aligned to
1432 : * fixed align request, i.e. RE_FIT_TYPE and FL_FIT_TYPE
1433 : * are its main fitting cases.
1434 : *
1435 : * There are a few exceptions though, as an example it is
1436 : * a first allocation (early boot up) when we have "one"
1437 : * big free space that has to be split.
1438 : *
1439 : * Also we can hit this path in case of regular "vmap"
1440 : * allocations, if "this" current CPU was not preloaded.
1441 : * See the comment in alloc_vmap_area() why. If so, then
1442 : * GFP_NOWAIT is used instead to get an extra object for
1443 : * split purpose. That is rare and most time does not
1444 : * occur.
1445 : *
1446 : * What happens if an allocation gets failed. Basically,
1447 : * an "overflow" path is triggered to purge lazily freed
1448 : * areas to free some memory, then, the "retry" path is
1449 : * triggered to repeat one more time. See more details
1450 : * in alloc_vmap_area() function.
1451 : */
1452 0 : lva = kmem_cache_alloc(vmap_area_cachep, GFP_NOWAIT);
1453 0 : if (!lva)
1454 : return -1;
1455 : }
1456 :
1457 : /*
1458 : * Build the remainder.
1459 : */
1460 16 : lva->va_start = va->va_start;
1461 16 : lva->va_end = nva_start_addr;
1462 :
1463 : /*
1464 : * Shrink this VA to remaining size.
1465 : */
1466 16 : va->va_start = nva_start_addr + size;
1467 : } else {
1468 : return -1;
1469 : }
1470 :
1471 274 : if (type != FL_FIT_TYPE) {
1472 274 : augment_tree_propagate_from(va);
1473 :
1474 274 : if (lva) /* type == NE_FIT_TYPE */
1475 16 : insert_vmap_area_augment(lva, &va->rb_node, root, head);
1476 : }
1477 :
1478 : return 0;
1479 : }
1480 :
1481 : /*
1482 : * Returns a start address of the newly allocated area, if success.
1483 : * Otherwise a vend is returned that indicates failure.
1484 : */
1485 : static __always_inline unsigned long
1486 : __alloc_vmap_area(struct rb_root *root, struct list_head *head,
1487 : unsigned long size, unsigned long align,
1488 : unsigned long vstart, unsigned long vend)
1489 : {
1490 274 : bool adjust_search_size = true;
1491 : unsigned long nva_start_addr;
1492 : struct vmap_area *va;
1493 : int ret;
1494 :
1495 : /*
1496 : * Do not adjust when:
1497 : * a) align <= PAGE_SIZE, because it does not make any sense.
1498 : * All blocks(their start addresses) are at least PAGE_SIZE
1499 : * aligned anyway;
1500 : * b) a short range where a requested size corresponds to exactly
1501 : * specified [vstart:vend] interval and an alignment > PAGE_SIZE.
1502 : * With adjusted search length an allocation would not succeed.
1503 : */
1504 274 : if (align <= PAGE_SIZE || (align > PAGE_SIZE && (vend - vstart) == size))
1505 258 : adjust_search_size = false;
1506 :
1507 548 : va = find_vmap_lowest_match(root, size, align, vstart, adjust_search_size);
1508 274 : if (unlikely(!va))
1509 : return vend;
1510 :
1511 274 : if (va->va_start > vstart)
1512 273 : nva_start_addr = ALIGN(va->va_start, align);
1513 : else
1514 1 : nva_start_addr = ALIGN(vstart, align);
1515 :
1516 : /* Check the "vend" restriction. */
1517 274 : if (nva_start_addr + size > vend)
1518 : return vend;
1519 :
1520 : /* Update the free vmap_area. */
1521 274 : ret = adjust_va_to_fit_type(root, head, va, nva_start_addr, size);
1522 274 : if (WARN_ON_ONCE(ret))
1523 : return vend;
1524 :
1525 : #if DEBUG_AUGMENT_LOWEST_MATCH_CHECK
1526 : find_vmap_lowest_match_check(root, head, size, align);
1527 : #endif
1528 :
1529 : return nva_start_addr;
1530 : }
1531 :
1532 : /*
1533 : * Free a region of KVA allocated by alloc_vmap_area
1534 : */
1535 0 : static void free_vmap_area(struct vmap_area *va)
1536 : {
1537 : /*
1538 : * Remove from the busy tree/list.
1539 : */
1540 0 : spin_lock(&vmap_area_lock);
1541 0 : unlink_va(va, &vmap_area_root);
1542 0 : spin_unlock(&vmap_area_lock);
1543 :
1544 : /*
1545 : * Insert/Merge it back to the free tree/list.
1546 : */
1547 0 : spin_lock(&free_vmap_area_lock);
1548 0 : merge_or_add_vmap_area_augment(va, &free_vmap_area_root, &free_vmap_area_list);
1549 0 : spin_unlock(&free_vmap_area_lock);
1550 0 : }
1551 :
1552 : static inline void
1553 274 : preload_this_cpu_lock(spinlock_t *lock, gfp_t gfp_mask, int node)
1554 : {
1555 274 : struct vmap_area *va = NULL;
1556 :
1557 : /*
1558 : * Preload this CPU with one extra vmap_area object. It is used
1559 : * when fit type of free area is NE_FIT_TYPE. It guarantees that
1560 : * a CPU that does an allocation is preloaded.
1561 : *
1562 : * We do it in non-atomic context, thus it allows us to use more
1563 : * permissive allocation masks to be more stable under low memory
1564 : * condition and high memory pressure.
1565 : */
1566 274 : if (!this_cpu_read(ne_fit_preload_node))
1567 17 : va = kmem_cache_alloc_node(vmap_area_cachep, gfp_mask, node);
1568 :
1569 274 : spin_lock(lock);
1570 :
1571 274 : if (va && __this_cpu_cmpxchg(ne_fit_preload_node, NULL, va))
1572 0 : kmem_cache_free(vmap_area_cachep, va);
1573 274 : }
1574 :
1575 : /*
1576 : * Allocate a region of KVA of the specified size and alignment, within the
1577 : * vstart and vend.
1578 : */
1579 274 : static struct vmap_area *alloc_vmap_area(unsigned long size,
1580 : unsigned long align,
1581 : unsigned long vstart, unsigned long vend,
1582 : int node, gfp_t gfp_mask,
1583 : unsigned long va_flags)
1584 : {
1585 : struct vmap_area *va;
1586 : unsigned long freed;
1587 : unsigned long addr;
1588 274 : int purged = 0;
1589 : int ret;
1590 :
1591 548 : if (unlikely(!size || offset_in_page(size) || !is_power_of_2(align)))
1592 : return ERR_PTR(-EINVAL);
1593 :
1594 274 : if (unlikely(!vmap_initialized))
1595 : return ERR_PTR(-EBUSY);
1596 :
1597 : might_sleep();
1598 274 : gfp_mask = gfp_mask & GFP_RECLAIM_MASK;
1599 :
1600 274 : va = kmem_cache_alloc_node(vmap_area_cachep, gfp_mask, node);
1601 274 : if (unlikely(!va))
1602 : return ERR_PTR(-ENOMEM);
1603 :
1604 : /*
1605 : * Only scan the relevant parts containing pointers to other objects
1606 : * to avoid false negatives.
1607 : */
1608 : kmemleak_scan_area(&va->rb_node, SIZE_MAX, gfp_mask);
1609 :
1610 : retry:
1611 274 : preload_this_cpu_lock(&free_vmap_area_lock, gfp_mask, node);
1612 274 : addr = __alloc_vmap_area(&free_vmap_area_root, &free_vmap_area_list,
1613 : size, align, vstart, vend);
1614 274 : spin_unlock(&free_vmap_area_lock);
1615 :
1616 274 : trace_alloc_vmap_area(addr, size, align, vstart, vend, addr == vend);
1617 :
1618 : /*
1619 : * If an allocation fails, the "vend" address is
1620 : * returned. Therefore trigger the overflow path.
1621 : */
1622 274 : if (unlikely(addr == vend))
1623 : goto overflow;
1624 :
1625 274 : va->va_start = addr;
1626 274 : va->va_end = addr + size;
1627 274 : va->vm = NULL;
1628 274 : va->flags = va_flags;
1629 :
1630 274 : spin_lock(&vmap_area_lock);
1631 274 : insert_vmap_area(va, &vmap_area_root, &vmap_area_list);
1632 274 : spin_unlock(&vmap_area_lock);
1633 :
1634 274 : BUG_ON(!IS_ALIGNED(va->va_start, align));
1635 274 : BUG_ON(va->va_start < vstart);
1636 274 : BUG_ON(va->va_end > vend);
1637 :
1638 : ret = kasan_populate_vmalloc(addr, size);
1639 : if (ret) {
1640 : free_vmap_area(va);
1641 : return ERR_PTR(ret);
1642 : }
1643 :
1644 : return va;
1645 :
1646 : overflow:
1647 0 : if (!purged) {
1648 0 : purge_vmap_area_lazy();
1649 0 : purged = 1;
1650 0 : goto retry;
1651 : }
1652 :
1653 0 : freed = 0;
1654 0 : blocking_notifier_call_chain(&vmap_notify_list, 0, &freed);
1655 :
1656 0 : if (freed > 0) {
1657 : purged = 0;
1658 : goto retry;
1659 : }
1660 :
1661 0 : if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit())
1662 0 : pr_warn("vmap allocation for size %lu failed: use vmalloc=<size> to increase size\n",
1663 : size);
1664 :
1665 0 : kmem_cache_free(vmap_area_cachep, va);
1666 0 : return ERR_PTR(-EBUSY);
1667 : }
1668 :
1669 0 : int register_vmap_purge_notifier(struct notifier_block *nb)
1670 : {
1671 0 : return blocking_notifier_chain_register(&vmap_notify_list, nb);
1672 : }
1673 : EXPORT_SYMBOL_GPL(register_vmap_purge_notifier);
1674 :
1675 0 : int unregister_vmap_purge_notifier(struct notifier_block *nb)
1676 : {
1677 0 : return blocking_notifier_chain_unregister(&vmap_notify_list, nb);
1678 : }
1679 : EXPORT_SYMBOL_GPL(unregister_vmap_purge_notifier);
1680 :
1681 : /*
1682 : * lazy_max_pages is the maximum amount of virtual address space we gather up
1683 : * before attempting to purge with a TLB flush.
1684 : *
1685 : * There is a tradeoff here: a larger number will cover more kernel page tables
1686 : * and take slightly longer to purge, but it will linearly reduce the number of
1687 : * global TLB flushes that must be performed. It would seem natural to scale
1688 : * this number up linearly with the number of CPUs (because vmapping activity
1689 : * could also scale linearly with the number of CPUs), however it is likely
1690 : * that in practice, workloads might be constrained in other ways that mean
1691 : * vmap activity will not scale linearly with CPUs. Also, I want to be
1692 : * conservative and not introduce a big latency on huge systems, so go with
1693 : * a less aggressive log scale. It will still be an improvement over the old
1694 : * code, and it will be simple to change the scale factor if we find that it
1695 : * becomes a problem on bigger systems.
1696 : */
1697 : static unsigned long lazy_max_pages(void)
1698 : {
1699 : unsigned int log;
1700 :
1701 268 : log = fls(num_online_cpus());
1702 :
1703 268 : return log * (32UL * 1024 * 1024 / PAGE_SIZE);
1704 : }
1705 :
1706 : static atomic_long_t vmap_lazy_nr = ATOMIC_LONG_INIT(0);
1707 :
1708 : /*
1709 : * Serialize vmap purging. There is no actual critical section protected
1710 : * by this lock, but we want to avoid concurrent calls for performance
1711 : * reasons and to make the pcpu_get_vm_areas more deterministic.
1712 : */
1713 : static DEFINE_MUTEX(vmap_purge_lock);
1714 :
1715 : /* for per-CPU blocks */
1716 : static void purge_fragmented_blocks_allcpus(void);
1717 :
1718 : /*
1719 : * Purges all lazily-freed vmap areas.
1720 : */
1721 5 : static bool __purge_vmap_area_lazy(unsigned long start, unsigned long end)
1722 : {
1723 : unsigned long resched_threshold;
1724 5 : unsigned int num_purged_areas = 0;
1725 : struct list_head local_purge_list;
1726 : struct vmap_area *va, *n_va;
1727 :
1728 : lockdep_assert_held(&vmap_purge_lock);
1729 :
1730 5 : spin_lock(&purge_vmap_area_lock);
1731 5 : purge_vmap_area_root = RB_ROOT;
1732 5 : list_replace_init(&purge_vmap_area_list, &local_purge_list);
1733 5 : spin_unlock(&purge_vmap_area_lock);
1734 :
1735 5 : if (unlikely(list_empty(&local_purge_list)))
1736 : goto out;
1737 :
1738 5 : start = min(start,
1739 : list_first_entry(&local_purge_list,
1740 : struct vmap_area, list)->va_start);
1741 :
1742 5 : end = max(end,
1743 : list_last_entry(&local_purge_list,
1744 : struct vmap_area, list)->va_end);
1745 :
1746 5 : flush_tlb_kernel_range(start, end);
1747 5 : resched_threshold = lazy_max_pages() << 1;
1748 :
1749 5 : spin_lock(&free_vmap_area_lock);
1750 10 : list_for_each_entry_safe(va, n_va, &local_purge_list, list) {
1751 5 : unsigned long nr = (va->va_end - va->va_start) >> PAGE_SHIFT;
1752 5 : unsigned long orig_start = va->va_start;
1753 5 : unsigned long orig_end = va->va_end;
1754 :
1755 : /*
1756 : * Finally insert or merge lazily-freed area. It is
1757 : * detached and there is no need to "unlink" it from
1758 : * anything.
1759 : */
1760 5 : va = merge_or_add_vmap_area_augment(va, &free_vmap_area_root,
1761 : &free_vmap_area_list);
1762 :
1763 5 : if (!va)
1764 0 : continue;
1765 :
1766 10 : if (is_vmalloc_or_module_addr((void *)orig_start))
1767 : kasan_release_vmalloc(orig_start, orig_end,
1768 : va->va_start, va->va_end);
1769 :
1770 10 : atomic_long_sub(nr, &vmap_lazy_nr);
1771 5 : num_purged_areas++;
1772 :
1773 5 : if (atomic_long_read(&vmap_lazy_nr) < resched_threshold)
1774 5 : cond_resched_lock(&free_vmap_area_lock);
1775 : }
1776 : spin_unlock(&free_vmap_area_lock);
1777 :
1778 : out:
1779 5 : trace_purge_vmap_area_lazy(start, end, num_purged_areas);
1780 5 : return num_purged_areas > 0;
1781 : }
1782 :
1783 : /*
1784 : * Kick off a purge of the outstanding lazy areas.
1785 : */
1786 0 : static void purge_vmap_area_lazy(void)
1787 : {
1788 0 : mutex_lock(&vmap_purge_lock);
1789 0 : purge_fragmented_blocks_allcpus();
1790 0 : __purge_vmap_area_lazy(ULONG_MAX, 0);
1791 0 : mutex_unlock(&vmap_purge_lock);
1792 0 : }
1793 :
1794 5 : static void drain_vmap_area_work(struct work_struct *work)
1795 : {
1796 : unsigned long nr_lazy;
1797 :
1798 : do {
1799 5 : mutex_lock(&vmap_purge_lock);
1800 5 : __purge_vmap_area_lazy(ULONG_MAX, 0);
1801 5 : mutex_unlock(&vmap_purge_lock);
1802 :
1803 : /* Recheck if further work is required. */
1804 5 : nr_lazy = atomic_long_read(&vmap_lazy_nr);
1805 5 : } while (nr_lazy > lazy_max_pages());
1806 5 : }
1807 :
1808 : /*
1809 : * Free a vmap area, caller ensuring that the area has been unmapped,
1810 : * unlinked and flush_cache_vunmap had been called for the correct
1811 : * range previously.
1812 : */
1813 258 : static void free_vmap_area_noflush(struct vmap_area *va)
1814 : {
1815 258 : unsigned long nr_lazy_max = lazy_max_pages();
1816 258 : unsigned long va_start = va->va_start;
1817 : unsigned long nr_lazy;
1818 :
1819 516 : if (WARN_ON_ONCE(!list_empty(&va->list)))
1820 : return;
1821 :
1822 516 : nr_lazy = atomic_long_add_return((va->va_end - va->va_start) >>
1823 : PAGE_SHIFT, &vmap_lazy_nr);
1824 :
1825 : /*
1826 : * Merge or place it to the purge tree/list.
1827 : */
1828 258 : spin_lock(&purge_vmap_area_lock);
1829 258 : merge_or_add_vmap_area(va,
1830 : &purge_vmap_area_root, &purge_vmap_area_list);
1831 258 : spin_unlock(&purge_vmap_area_lock);
1832 :
1833 258 : trace_free_vmap_area_noflush(va_start, nr_lazy, nr_lazy_max);
1834 :
1835 : /* After this point, we may free va at any time */
1836 258 : if (unlikely(nr_lazy > nr_lazy_max))
1837 : schedule_work(&drain_vmap_work);
1838 : }
1839 :
1840 : /*
1841 : * Free and unmap a vmap area
1842 : */
1843 258 : static void free_unmap_vmap_area(struct vmap_area *va)
1844 : {
1845 258 : flush_cache_vunmap(va->va_start, va->va_end);
1846 516 : vunmap_range_noflush(va->va_start, va->va_end);
1847 : if (debug_pagealloc_enabled_static())
1848 : flush_tlb_kernel_range(va->va_start, va->va_end);
1849 :
1850 258 : free_vmap_area_noflush(va);
1851 258 : }
1852 :
1853 0 : struct vmap_area *find_vmap_area(unsigned long addr)
1854 : {
1855 : struct vmap_area *va;
1856 :
1857 16 : spin_lock(&vmap_area_lock);
1858 16 : va = __find_vmap_area(addr, &vmap_area_root);
1859 16 : spin_unlock(&vmap_area_lock);
1860 :
1861 0 : return va;
1862 : }
1863 :
1864 258 : static struct vmap_area *find_unlink_vmap_area(unsigned long addr)
1865 : {
1866 : struct vmap_area *va;
1867 :
1868 258 : spin_lock(&vmap_area_lock);
1869 258 : va = __find_vmap_area(addr, &vmap_area_root);
1870 258 : if (va)
1871 : unlink_va(va, &vmap_area_root);
1872 258 : spin_unlock(&vmap_area_lock);
1873 :
1874 258 : return va;
1875 : }
1876 :
1877 : /*** Per cpu kva allocator ***/
1878 :
1879 : /*
1880 : * vmap space is limited especially on 32 bit architectures. Ensure there is
1881 : * room for at least 16 percpu vmap blocks per CPU.
1882 : */
1883 : /*
1884 : * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
1885 : * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
1886 : * instead (we just need a rough idea)
1887 : */
1888 : #if BITS_PER_LONG == 32
1889 : #define VMALLOC_SPACE (128UL*1024*1024)
1890 : #else
1891 : #define VMALLOC_SPACE (128UL*1024*1024*1024)
1892 : #endif
1893 :
1894 : #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
1895 : #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
1896 : #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
1897 : #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
1898 : #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
1899 : #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
1900 : #define VMAP_BBMAP_BITS \
1901 : VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
1902 : VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
1903 : VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
1904 :
1905 : #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
1906 :
1907 : #define VMAP_RAM 0x1 /* indicates vm_map_ram area*/
1908 : #define VMAP_BLOCK 0x2 /* mark out the vmap_block sub-type*/
1909 : #define VMAP_FLAGS_MASK 0x3
1910 :
1911 : struct vmap_block_queue {
1912 : spinlock_t lock;
1913 : struct list_head free;
1914 : };
1915 :
1916 : struct vmap_block {
1917 : spinlock_t lock;
1918 : struct vmap_area *va;
1919 : unsigned long free, dirty;
1920 : DECLARE_BITMAP(used_map, VMAP_BBMAP_BITS);
1921 : unsigned long dirty_min, dirty_max; /*< dirty range */
1922 : struct list_head free_list;
1923 : struct rcu_head rcu_head;
1924 : struct list_head purge;
1925 : };
1926 :
1927 : /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
1928 : static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
1929 :
1930 : /*
1931 : * XArray of vmap blocks, indexed by address, to quickly find a vmap block
1932 : * in the free path. Could get rid of this if we change the API to return a
1933 : * "cookie" from alloc, to be passed to free. But no big deal yet.
1934 : */
1935 : static DEFINE_XARRAY(vmap_blocks);
1936 :
1937 : /*
1938 : * We should probably have a fallback mechanism to allocate virtual memory
1939 : * out of partially filled vmap blocks. However vmap block sizing should be
1940 : * fairly reasonable according to the vmalloc size, so it shouldn't be a
1941 : * big problem.
1942 : */
1943 :
1944 : static unsigned long addr_to_vb_idx(unsigned long addr)
1945 : {
1946 0 : addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
1947 0 : addr /= VMAP_BLOCK_SIZE;
1948 : return addr;
1949 : }
1950 :
1951 0 : static void *vmap_block_vaddr(unsigned long va_start, unsigned long pages_off)
1952 : {
1953 : unsigned long addr;
1954 :
1955 0 : addr = va_start + (pages_off << PAGE_SHIFT);
1956 0 : BUG_ON(addr_to_vb_idx(addr) != addr_to_vb_idx(va_start));
1957 0 : return (void *)addr;
1958 : }
1959 :
1960 : /**
1961 : * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this
1962 : * block. Of course pages number can't exceed VMAP_BBMAP_BITS
1963 : * @order: how many 2^order pages should be occupied in newly allocated block
1964 : * @gfp_mask: flags for the page level allocator
1965 : *
1966 : * Return: virtual address in a newly allocated block or ERR_PTR(-errno)
1967 : */
1968 0 : static void *new_vmap_block(unsigned int order, gfp_t gfp_mask)
1969 : {
1970 : struct vmap_block_queue *vbq;
1971 : struct vmap_block *vb;
1972 : struct vmap_area *va;
1973 : unsigned long vb_idx;
1974 : int node, err;
1975 : void *vaddr;
1976 :
1977 0 : node = numa_node_id();
1978 :
1979 0 : vb = kmalloc_node(sizeof(struct vmap_block),
1980 : gfp_mask & GFP_RECLAIM_MASK, node);
1981 0 : if (unlikely(!vb))
1982 : return ERR_PTR(-ENOMEM);
1983 :
1984 0 : va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
1985 0 : VMALLOC_START, VMALLOC_END,
1986 : node, gfp_mask,
1987 : VMAP_RAM|VMAP_BLOCK);
1988 0 : if (IS_ERR(va)) {
1989 0 : kfree(vb);
1990 0 : return ERR_CAST(va);
1991 : }
1992 :
1993 0 : vaddr = vmap_block_vaddr(va->va_start, 0);
1994 0 : spin_lock_init(&vb->lock);
1995 0 : vb->va = va;
1996 : /* At least something should be left free */
1997 0 : BUG_ON(VMAP_BBMAP_BITS <= (1UL << order));
1998 0 : bitmap_zero(vb->used_map, VMAP_BBMAP_BITS);
1999 0 : vb->free = VMAP_BBMAP_BITS - (1UL << order);
2000 0 : vb->dirty = 0;
2001 0 : vb->dirty_min = VMAP_BBMAP_BITS;
2002 0 : vb->dirty_max = 0;
2003 0 : bitmap_set(vb->used_map, 0, (1UL << order));
2004 0 : INIT_LIST_HEAD(&vb->free_list);
2005 :
2006 0 : vb_idx = addr_to_vb_idx(va->va_start);
2007 0 : err = xa_insert(&vmap_blocks, vb_idx, vb, gfp_mask);
2008 0 : if (err) {
2009 0 : kfree(vb);
2010 0 : free_vmap_area(va);
2011 0 : return ERR_PTR(err);
2012 : }
2013 :
2014 0 : vbq = raw_cpu_ptr(&vmap_block_queue);
2015 0 : spin_lock(&vbq->lock);
2016 0 : list_add_tail_rcu(&vb->free_list, &vbq->free);
2017 0 : spin_unlock(&vbq->lock);
2018 :
2019 0 : return vaddr;
2020 : }
2021 :
2022 0 : static void free_vmap_block(struct vmap_block *vb)
2023 : {
2024 : struct vmap_block *tmp;
2025 :
2026 0 : tmp = xa_erase(&vmap_blocks, addr_to_vb_idx(vb->va->va_start));
2027 0 : BUG_ON(tmp != vb);
2028 :
2029 0 : spin_lock(&vmap_area_lock);
2030 0 : unlink_va(vb->va, &vmap_area_root);
2031 0 : spin_unlock(&vmap_area_lock);
2032 :
2033 0 : free_vmap_area_noflush(vb->va);
2034 0 : kfree_rcu(vb, rcu_head);
2035 0 : }
2036 :
2037 0 : static void purge_fragmented_blocks(int cpu)
2038 : {
2039 0 : LIST_HEAD(purge);
2040 : struct vmap_block *vb;
2041 : struct vmap_block *n_vb;
2042 0 : struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
2043 :
2044 : rcu_read_lock();
2045 0 : list_for_each_entry_rcu(vb, &vbq->free, free_list) {
2046 :
2047 0 : if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
2048 0 : continue;
2049 :
2050 0 : spin_lock(&vb->lock);
2051 0 : if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
2052 0 : vb->free = 0; /* prevent further allocs after releasing lock */
2053 0 : vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */
2054 0 : vb->dirty_min = 0;
2055 0 : vb->dirty_max = VMAP_BBMAP_BITS;
2056 0 : spin_lock(&vbq->lock);
2057 0 : list_del_rcu(&vb->free_list);
2058 0 : spin_unlock(&vbq->lock);
2059 0 : spin_unlock(&vb->lock);
2060 0 : list_add_tail(&vb->purge, &purge);
2061 : } else
2062 0 : spin_unlock(&vb->lock);
2063 : }
2064 : rcu_read_unlock();
2065 :
2066 0 : list_for_each_entry_safe(vb, n_vb, &purge, purge) {
2067 0 : list_del(&vb->purge);
2068 0 : free_vmap_block(vb);
2069 : }
2070 0 : }
2071 :
2072 : static void purge_fragmented_blocks_allcpus(void)
2073 : {
2074 : int cpu;
2075 :
2076 0 : for_each_possible_cpu(cpu)
2077 0 : purge_fragmented_blocks(cpu);
2078 : }
2079 :
2080 0 : static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
2081 : {
2082 : struct vmap_block_queue *vbq;
2083 : struct vmap_block *vb;
2084 0 : void *vaddr = NULL;
2085 : unsigned int order;
2086 :
2087 0 : BUG_ON(offset_in_page(size));
2088 0 : BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
2089 0 : if (WARN_ON(size == 0)) {
2090 : /*
2091 : * Allocating 0 bytes isn't what caller wants since
2092 : * get_order(0) returns funny result. Just warn and terminate
2093 : * early.
2094 : */
2095 : return NULL;
2096 : }
2097 0 : order = get_order(size);
2098 :
2099 : rcu_read_lock();
2100 0 : vbq = raw_cpu_ptr(&vmap_block_queue);
2101 0 : list_for_each_entry_rcu(vb, &vbq->free, free_list) {
2102 : unsigned long pages_off;
2103 :
2104 0 : spin_lock(&vb->lock);
2105 0 : if (vb->free < (1UL << order)) {
2106 0 : spin_unlock(&vb->lock);
2107 0 : continue;
2108 : }
2109 :
2110 0 : pages_off = VMAP_BBMAP_BITS - vb->free;
2111 0 : vaddr = vmap_block_vaddr(vb->va->va_start, pages_off);
2112 0 : vb->free -= 1UL << order;
2113 0 : bitmap_set(vb->used_map, pages_off, (1UL << order));
2114 0 : if (vb->free == 0) {
2115 0 : spin_lock(&vbq->lock);
2116 0 : list_del_rcu(&vb->free_list);
2117 0 : spin_unlock(&vbq->lock);
2118 : }
2119 :
2120 0 : spin_unlock(&vb->lock);
2121 : break;
2122 : }
2123 :
2124 : rcu_read_unlock();
2125 :
2126 : /* Allocate new block if nothing was found */
2127 0 : if (!vaddr)
2128 0 : vaddr = new_vmap_block(order, gfp_mask);
2129 :
2130 : return vaddr;
2131 : }
2132 :
2133 0 : static void vb_free(unsigned long addr, unsigned long size)
2134 : {
2135 : unsigned long offset;
2136 : unsigned int order;
2137 : struct vmap_block *vb;
2138 :
2139 0 : BUG_ON(offset_in_page(size));
2140 0 : BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
2141 :
2142 0 : flush_cache_vunmap(addr, addr + size);
2143 :
2144 0 : order = get_order(size);
2145 0 : offset = (addr & (VMAP_BLOCK_SIZE - 1)) >> PAGE_SHIFT;
2146 0 : vb = xa_load(&vmap_blocks, addr_to_vb_idx(addr));
2147 0 : spin_lock(&vb->lock);
2148 0 : bitmap_clear(vb->used_map, offset, (1UL << order));
2149 0 : spin_unlock(&vb->lock);
2150 :
2151 0 : vunmap_range_noflush(addr, addr + size);
2152 :
2153 : if (debug_pagealloc_enabled_static())
2154 : flush_tlb_kernel_range(addr, addr + size);
2155 :
2156 0 : spin_lock(&vb->lock);
2157 :
2158 : /* Expand dirty range */
2159 0 : vb->dirty_min = min(vb->dirty_min, offset);
2160 0 : vb->dirty_max = max(vb->dirty_max, offset + (1UL << order));
2161 :
2162 0 : vb->dirty += 1UL << order;
2163 0 : if (vb->dirty == VMAP_BBMAP_BITS) {
2164 0 : BUG_ON(vb->free);
2165 0 : spin_unlock(&vb->lock);
2166 0 : free_vmap_block(vb);
2167 : } else
2168 0 : spin_unlock(&vb->lock);
2169 0 : }
2170 :
2171 0 : static void _vm_unmap_aliases(unsigned long start, unsigned long end, int flush)
2172 : {
2173 : int cpu;
2174 :
2175 0 : if (unlikely(!vmap_initialized))
2176 : return;
2177 :
2178 : might_sleep();
2179 :
2180 0 : for_each_possible_cpu(cpu) {
2181 0 : struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
2182 : struct vmap_block *vb;
2183 :
2184 : rcu_read_lock();
2185 0 : list_for_each_entry_rcu(vb, &vbq->free, free_list) {
2186 0 : spin_lock(&vb->lock);
2187 0 : if (vb->dirty && vb->dirty != VMAP_BBMAP_BITS) {
2188 0 : unsigned long va_start = vb->va->va_start;
2189 : unsigned long s, e;
2190 :
2191 0 : s = va_start + (vb->dirty_min << PAGE_SHIFT);
2192 0 : e = va_start + (vb->dirty_max << PAGE_SHIFT);
2193 :
2194 0 : start = min(s, start);
2195 0 : end = max(e, end);
2196 :
2197 0 : flush = 1;
2198 : }
2199 0 : spin_unlock(&vb->lock);
2200 : }
2201 : rcu_read_unlock();
2202 : }
2203 :
2204 0 : mutex_lock(&vmap_purge_lock);
2205 0 : purge_fragmented_blocks_allcpus();
2206 0 : if (!__purge_vmap_area_lazy(start, end) && flush)
2207 0 : flush_tlb_kernel_range(start, end);
2208 0 : mutex_unlock(&vmap_purge_lock);
2209 : }
2210 :
2211 : /**
2212 : * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
2213 : *
2214 : * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
2215 : * to amortize TLB flushing overheads. What this means is that any page you
2216 : * have now, may, in a former life, have been mapped into kernel virtual
2217 : * address by the vmap layer and so there might be some CPUs with TLB entries
2218 : * still referencing that page (additional to the regular 1:1 kernel mapping).
2219 : *
2220 : * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
2221 : * be sure that none of the pages we have control over will have any aliases
2222 : * from the vmap layer.
2223 : */
2224 0 : void vm_unmap_aliases(void)
2225 : {
2226 0 : unsigned long start = ULONG_MAX, end = 0;
2227 0 : int flush = 0;
2228 :
2229 0 : _vm_unmap_aliases(start, end, flush);
2230 0 : }
2231 : EXPORT_SYMBOL_GPL(vm_unmap_aliases);
2232 :
2233 : /**
2234 : * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
2235 : * @mem: the pointer returned by vm_map_ram
2236 : * @count: the count passed to that vm_map_ram call (cannot unmap partial)
2237 : */
2238 0 : void vm_unmap_ram(const void *mem, unsigned int count)
2239 : {
2240 0 : unsigned long size = (unsigned long)count << PAGE_SHIFT;
2241 0 : unsigned long addr = (unsigned long)kasan_reset_tag(mem);
2242 : struct vmap_area *va;
2243 :
2244 : might_sleep();
2245 0 : BUG_ON(!addr);
2246 0 : BUG_ON(addr < VMALLOC_START);
2247 0 : BUG_ON(addr > VMALLOC_END);
2248 0 : BUG_ON(!PAGE_ALIGNED(addr));
2249 :
2250 0 : kasan_poison_vmalloc(mem, size);
2251 :
2252 0 : if (likely(count <= VMAP_MAX_ALLOC)) {
2253 0 : debug_check_no_locks_freed(mem, size);
2254 0 : vb_free(addr, size);
2255 0 : return;
2256 : }
2257 :
2258 0 : va = find_unlink_vmap_area(addr);
2259 0 : if (WARN_ON_ONCE(!va))
2260 : return;
2261 :
2262 0 : debug_check_no_locks_freed((void *)va->va_start,
2263 0 : (va->va_end - va->va_start));
2264 0 : free_unmap_vmap_area(va);
2265 : }
2266 : EXPORT_SYMBOL(vm_unmap_ram);
2267 :
2268 : /**
2269 : * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
2270 : * @pages: an array of pointers to the pages to be mapped
2271 : * @count: number of pages
2272 : * @node: prefer to allocate data structures on this node
2273 : *
2274 : * If you use this function for less than VMAP_MAX_ALLOC pages, it could be
2275 : * faster than vmap so it's good. But if you mix long-life and short-life
2276 : * objects with vm_map_ram(), it could consume lots of address space through
2277 : * fragmentation (especially on a 32bit machine). You could see failures in
2278 : * the end. Please use this function for short-lived objects.
2279 : *
2280 : * Returns: a pointer to the address that has been mapped, or %NULL on failure
2281 : */
2282 0 : void *vm_map_ram(struct page **pages, unsigned int count, int node)
2283 : {
2284 0 : unsigned long size = (unsigned long)count << PAGE_SHIFT;
2285 : unsigned long addr;
2286 : void *mem;
2287 :
2288 0 : if (likely(count <= VMAP_MAX_ALLOC)) {
2289 0 : mem = vb_alloc(size, GFP_KERNEL);
2290 0 : if (IS_ERR(mem))
2291 : return NULL;
2292 : addr = (unsigned long)mem;
2293 : } else {
2294 : struct vmap_area *va;
2295 0 : va = alloc_vmap_area(size, PAGE_SIZE,
2296 0 : VMALLOC_START, VMALLOC_END,
2297 : node, GFP_KERNEL, VMAP_RAM);
2298 0 : if (IS_ERR(va))
2299 : return NULL;
2300 :
2301 0 : addr = va->va_start;
2302 0 : mem = (void *)addr;
2303 : }
2304 :
2305 0 : if (vmap_pages_range(addr, addr + size, PAGE_KERNEL,
2306 : pages, PAGE_SHIFT) < 0) {
2307 0 : vm_unmap_ram(mem, count);
2308 0 : return NULL;
2309 : }
2310 :
2311 : /*
2312 : * Mark the pages as accessible, now that they are mapped.
2313 : * With hardware tag-based KASAN, marking is skipped for
2314 : * non-VM_ALLOC mappings, see __kasan_unpoison_vmalloc().
2315 : */
2316 : mem = kasan_unpoison_vmalloc(mem, size, KASAN_VMALLOC_PROT_NORMAL);
2317 :
2318 : return mem;
2319 : }
2320 : EXPORT_SYMBOL(vm_map_ram);
2321 :
2322 : static struct vm_struct *vmlist __initdata;
2323 :
2324 : static inline unsigned int vm_area_page_order(struct vm_struct *vm)
2325 : {
2326 : #ifdef CONFIG_HAVE_ARCH_HUGE_VMALLOC
2327 : return vm->page_order;
2328 : #else
2329 : return 0;
2330 : #endif
2331 : }
2332 :
2333 274 : static inline void set_vm_area_page_order(struct vm_struct *vm, unsigned int order)
2334 : {
2335 : #ifdef CONFIG_HAVE_ARCH_HUGE_VMALLOC
2336 : vm->page_order = order;
2337 : #else
2338 274 : BUG_ON(order != 0);
2339 : #endif
2340 274 : }
2341 :
2342 : /**
2343 : * vm_area_add_early - add vmap area early during boot
2344 : * @vm: vm_struct to add
2345 : *
2346 : * This function is used to add fixed kernel vm area to vmlist before
2347 : * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
2348 : * should contain proper values and the other fields should be zero.
2349 : *
2350 : * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
2351 : */
2352 0 : void __init vm_area_add_early(struct vm_struct *vm)
2353 : {
2354 : struct vm_struct *tmp, **p;
2355 :
2356 0 : BUG_ON(vmap_initialized);
2357 0 : for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
2358 0 : if (tmp->addr >= vm->addr) {
2359 0 : BUG_ON(tmp->addr < vm->addr + vm->size);
2360 : break;
2361 : } else
2362 0 : BUG_ON(tmp->addr + tmp->size > vm->addr);
2363 : }
2364 0 : vm->next = *p;
2365 0 : *p = vm;
2366 0 : }
2367 :
2368 : /**
2369 : * vm_area_register_early - register vmap area early during boot
2370 : * @vm: vm_struct to register
2371 : * @align: requested alignment
2372 : *
2373 : * This function is used to register kernel vm area before
2374 : * vmalloc_init() is called. @vm->size and @vm->flags should contain
2375 : * proper values on entry and other fields should be zero. On return,
2376 : * vm->addr contains the allocated address.
2377 : *
2378 : * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
2379 : */
2380 0 : void __init vm_area_register_early(struct vm_struct *vm, size_t align)
2381 : {
2382 0 : unsigned long addr = ALIGN(VMALLOC_START, align);
2383 : struct vm_struct *cur, **p;
2384 :
2385 0 : BUG_ON(vmap_initialized);
2386 :
2387 0 : for (p = &vmlist; (cur = *p) != NULL; p = &cur->next) {
2388 0 : if ((unsigned long)cur->addr - addr >= vm->size)
2389 : break;
2390 0 : addr = ALIGN((unsigned long)cur->addr + cur->size, align);
2391 : }
2392 :
2393 0 : BUG_ON(addr > VMALLOC_END - vm->size);
2394 0 : vm->addr = (void *)addr;
2395 0 : vm->next = *p;
2396 0 : *p = vm;
2397 0 : kasan_populate_early_vm_area_shadow(vm->addr, vm->size);
2398 0 : }
2399 :
2400 1 : static void vmap_init_free_space(void)
2401 : {
2402 1 : unsigned long vmap_start = 1;
2403 1 : const unsigned long vmap_end = ULONG_MAX;
2404 : struct vmap_area *busy, *free;
2405 :
2406 : /*
2407 : * B F B B B F
2408 : * -|-----|.....|-----|-----|-----|.....|-
2409 : * | The KVA space |
2410 : * |<--------------------------------->|
2411 : */
2412 1 : list_for_each_entry(busy, &vmap_area_list, list) {
2413 0 : if (busy->va_start - vmap_start > 0) {
2414 0 : free = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT);
2415 0 : if (!WARN_ON_ONCE(!free)) {
2416 0 : free->va_start = vmap_start;
2417 0 : free->va_end = busy->va_start;
2418 :
2419 0 : insert_vmap_area_augment(free, NULL,
2420 : &free_vmap_area_root,
2421 : &free_vmap_area_list);
2422 : }
2423 : }
2424 :
2425 0 : vmap_start = busy->va_end;
2426 : }
2427 :
2428 1 : if (vmap_end - vmap_start > 0) {
2429 2 : free = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT);
2430 1 : if (!WARN_ON_ONCE(!free)) {
2431 1 : free->va_start = vmap_start;
2432 1 : free->va_end = vmap_end;
2433 :
2434 1 : insert_vmap_area_augment(free, NULL,
2435 : &free_vmap_area_root,
2436 : &free_vmap_area_list);
2437 : }
2438 : }
2439 1 : }
2440 :
2441 : static inline void setup_vmalloc_vm_locked(struct vm_struct *vm,
2442 : struct vmap_area *va, unsigned long flags, const void *caller)
2443 : {
2444 274 : vm->flags = flags;
2445 274 : vm->addr = (void *)va->va_start;
2446 274 : vm->size = va->va_end - va->va_start;
2447 274 : vm->caller = caller;
2448 274 : va->vm = vm;
2449 : }
2450 :
2451 : static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
2452 : unsigned long flags, const void *caller)
2453 : {
2454 274 : spin_lock(&vmap_area_lock);
2455 274 : setup_vmalloc_vm_locked(vm, va, flags, caller);
2456 274 : spin_unlock(&vmap_area_lock);
2457 : }
2458 :
2459 : static void clear_vm_uninitialized_flag(struct vm_struct *vm)
2460 : {
2461 : /*
2462 : * Before removing VM_UNINITIALIZED,
2463 : * we should make sure that vm has proper values.
2464 : * Pair with smp_rmb() in show_numa_info().
2465 : */
2466 274 : smp_wmb();
2467 274 : vm->flags &= ~VM_UNINITIALIZED;
2468 : }
2469 :
2470 274 : static struct vm_struct *__get_vm_area_node(unsigned long size,
2471 : unsigned long align, unsigned long shift, unsigned long flags,
2472 : unsigned long start, unsigned long end, int node,
2473 : gfp_t gfp_mask, const void *caller)
2474 : {
2475 : struct vmap_area *va;
2476 : struct vm_struct *area;
2477 274 : unsigned long requested_size = size;
2478 :
2479 274 : BUG_ON(in_interrupt());
2480 274 : size = ALIGN(size, 1ul << shift);
2481 274 : if (unlikely(!size))
2482 : return NULL;
2483 :
2484 274 : if (flags & VM_IOREMAP)
2485 0 : align = 1ul << clamp_t(int, get_count_order_long(size),
2486 : PAGE_SHIFT, IOREMAP_MAX_ORDER);
2487 :
2488 274 : area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
2489 274 : if (unlikely(!area))
2490 : return NULL;
2491 :
2492 274 : if (!(flags & VM_NO_GUARD))
2493 274 : size += PAGE_SIZE;
2494 :
2495 274 : va = alloc_vmap_area(size, align, start, end, node, gfp_mask, 0);
2496 274 : if (IS_ERR(va)) {
2497 0 : kfree(area);
2498 0 : return NULL;
2499 : }
2500 :
2501 274 : setup_vmalloc_vm(area, va, flags, caller);
2502 :
2503 : /*
2504 : * Mark pages for non-VM_ALLOC mappings as accessible. Do it now as a
2505 : * best-effort approach, as they can be mapped outside of vmalloc code.
2506 : * For VM_ALLOC mappings, the pages are marked as accessible after
2507 : * getting mapped in __vmalloc_node_range().
2508 : * With hardware tag-based KASAN, marking is skipped for
2509 : * non-VM_ALLOC mappings, see __kasan_unpoison_vmalloc().
2510 : */
2511 274 : if (!(flags & VM_ALLOC))
2512 : area->addr = kasan_unpoison_vmalloc(area->addr, requested_size,
2513 : KASAN_VMALLOC_PROT_NORMAL);
2514 :
2515 : return area;
2516 : }
2517 :
2518 0 : struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
2519 : unsigned long start, unsigned long end,
2520 : const void *caller)
2521 : {
2522 0 : return __get_vm_area_node(size, 1, PAGE_SHIFT, flags, start, end,
2523 : NUMA_NO_NODE, GFP_KERNEL, caller);
2524 : }
2525 :
2526 : /**
2527 : * get_vm_area - reserve a contiguous kernel virtual area
2528 : * @size: size of the area
2529 : * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
2530 : *
2531 : * Search an area of @size in the kernel virtual mapping area,
2532 : * and reserved it for out purposes. Returns the area descriptor
2533 : * on success or %NULL on failure.
2534 : *
2535 : * Return: the area descriptor on success or %NULL on failure.
2536 : */
2537 0 : struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
2538 : {
2539 0 : return __get_vm_area_node(size, 1, PAGE_SHIFT, flags,
2540 0 : VMALLOC_START, VMALLOC_END,
2541 : NUMA_NO_NODE, GFP_KERNEL,
2542 0 : __builtin_return_address(0));
2543 : }
2544 :
2545 0 : struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
2546 : const void *caller)
2547 : {
2548 0 : return __get_vm_area_node(size, 1, PAGE_SHIFT, flags,
2549 0 : VMALLOC_START, VMALLOC_END,
2550 : NUMA_NO_NODE, GFP_KERNEL, caller);
2551 : }
2552 :
2553 : /**
2554 : * find_vm_area - find a continuous kernel virtual area
2555 : * @addr: base address
2556 : *
2557 : * Search for the kernel VM area starting at @addr, and return it.
2558 : * It is up to the caller to do all required locking to keep the returned
2559 : * pointer valid.
2560 : *
2561 : * Return: the area descriptor on success or %NULL on failure.
2562 : */
2563 16 : struct vm_struct *find_vm_area(const void *addr)
2564 : {
2565 : struct vmap_area *va;
2566 :
2567 32 : va = find_vmap_area((unsigned long)addr);
2568 16 : if (!va)
2569 : return NULL;
2570 :
2571 16 : return va->vm;
2572 : }
2573 :
2574 : /**
2575 : * remove_vm_area - find and remove a continuous kernel virtual area
2576 : * @addr: base address
2577 : *
2578 : * Search for the kernel VM area starting at @addr, and remove it.
2579 : * This function returns the found VM area, but using it is NOT safe
2580 : * on SMP machines, except for its size or flags.
2581 : *
2582 : * Return: the area descriptor on success or %NULL on failure.
2583 : */
2584 258 : struct vm_struct *remove_vm_area(const void *addr)
2585 : {
2586 : struct vmap_area *va;
2587 : struct vm_struct *vm;
2588 :
2589 : might_sleep();
2590 :
2591 258 : if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n",
2592 : addr))
2593 : return NULL;
2594 :
2595 258 : va = find_unlink_vmap_area((unsigned long)addr);
2596 258 : if (!va || !va->vm)
2597 : return NULL;
2598 258 : vm = va->vm;
2599 :
2600 258 : debug_check_no_locks_freed(vm->addr, get_vm_area_size(vm));
2601 258 : debug_check_no_obj_freed(vm->addr, get_vm_area_size(vm));
2602 258 : kasan_free_module_shadow(vm);
2603 258 : kasan_poison_vmalloc(vm->addr, get_vm_area_size(vm));
2604 :
2605 258 : free_unmap_vmap_area(va);
2606 258 : return vm;
2607 : }
2608 :
2609 0 : static inline void set_area_direct_map(const struct vm_struct *area,
2610 : int (*set_direct_map)(struct page *page))
2611 : {
2612 : int i;
2613 :
2614 : /* HUGE_VMALLOC passes small pages to set_direct_map */
2615 0 : for (i = 0; i < area->nr_pages; i++)
2616 0 : if (page_address(area->pages[i]))
2617 0 : set_direct_map(area->pages[i]);
2618 0 : }
2619 :
2620 : /*
2621 : * Flush the vm mapping and reset the direct map.
2622 : */
2623 0 : static void vm_reset_perms(struct vm_struct *area)
2624 : {
2625 0 : unsigned long start = ULONG_MAX, end = 0;
2626 0 : unsigned int page_order = vm_area_page_order(area);
2627 0 : int flush_dmap = 0;
2628 : int i;
2629 :
2630 : /*
2631 : * Find the start and end range of the direct mappings to make sure that
2632 : * the vm_unmap_aliases() flush includes the direct map.
2633 : */
2634 0 : for (i = 0; i < area->nr_pages; i += 1U << page_order) {
2635 0 : unsigned long addr = (unsigned long)page_address(area->pages[i]);
2636 :
2637 0 : if (addr) {
2638 : unsigned long page_size;
2639 :
2640 0 : page_size = PAGE_SIZE << page_order;
2641 0 : start = min(addr, start);
2642 0 : end = max(addr + page_size, end);
2643 0 : flush_dmap = 1;
2644 : }
2645 : }
2646 :
2647 : /*
2648 : * Set direct map to something invalid so that it won't be cached if
2649 : * there are any accesses after the TLB flush, then flush the TLB and
2650 : * reset the direct map permissions to the default.
2651 : */
2652 0 : set_area_direct_map(area, set_direct_map_invalid_noflush);
2653 0 : _vm_unmap_aliases(start, end, flush_dmap);
2654 0 : set_area_direct_map(area, set_direct_map_default_noflush);
2655 0 : }
2656 :
2657 0 : static void delayed_vfree_work(struct work_struct *w)
2658 : {
2659 0 : struct vfree_deferred *p = container_of(w, struct vfree_deferred, wq);
2660 : struct llist_node *t, *llnode;
2661 :
2662 0 : llist_for_each_safe(llnode, t, llist_del_all(&p->list))
2663 0 : vfree(llnode);
2664 0 : }
2665 :
2666 : /**
2667 : * vfree_atomic - release memory allocated by vmalloc()
2668 : * @addr: memory base address
2669 : *
2670 : * This one is just like vfree() but can be called in any atomic context
2671 : * except NMIs.
2672 : */
2673 0 : void vfree_atomic(const void *addr)
2674 : {
2675 0 : struct vfree_deferred *p = raw_cpu_ptr(&vfree_deferred);
2676 :
2677 0 : BUG_ON(in_nmi());
2678 0 : kmemleak_free(addr);
2679 :
2680 : /*
2681 : * Use raw_cpu_ptr() because this can be called from preemptible
2682 : * context. Preemption is absolutely fine here, because the llist_add()
2683 : * implementation is lockless, so it works even if we are adding to
2684 : * another cpu's list. schedule_work() should be fine with this too.
2685 : */
2686 0 : if (addr && llist_add((struct llist_node *)addr, &p->list))
2687 0 : schedule_work(&p->wq);
2688 0 : }
2689 :
2690 : /**
2691 : * vfree - Release memory allocated by vmalloc()
2692 : * @addr: Memory base address
2693 : *
2694 : * Free the virtually continuous memory area starting at @addr, as obtained
2695 : * from one of the vmalloc() family of APIs. This will usually also free the
2696 : * physical memory underlying the virtual allocation, but that memory is
2697 : * reference counted, so it will not be freed until the last user goes away.
2698 : *
2699 : * If @addr is NULL, no operation is performed.
2700 : *
2701 : * Context:
2702 : * May sleep if called *not* from interrupt context.
2703 : * Must not be called in NMI context (strictly speaking, it could be
2704 : * if we have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
2705 : * conventions for vfree() arch-dependent would be a really bad idea).
2706 : */
2707 258 : void vfree(const void *addr)
2708 : {
2709 : struct vm_struct *vm;
2710 : int i;
2711 :
2712 258 : if (unlikely(in_interrupt())) {
2713 0 : vfree_atomic(addr);
2714 0 : return;
2715 : }
2716 :
2717 258 : BUG_ON(in_nmi());
2718 258 : kmemleak_free(addr);
2719 : might_sleep();
2720 :
2721 258 : if (!addr)
2722 : return;
2723 :
2724 258 : vm = remove_vm_area(addr);
2725 258 : if (unlikely(!vm)) {
2726 0 : WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
2727 : addr);
2728 0 : return;
2729 : }
2730 :
2731 258 : if (unlikely(vm->flags & VM_FLUSH_RESET_PERMS))
2732 0 : vm_reset_perms(vm);
2733 42477 : for (i = 0; i < vm->nr_pages; i++) {
2734 42477 : struct page *page = vm->pages[i];
2735 :
2736 42477 : BUG_ON(!page);
2737 42477 : mod_memcg_page_state(page, MEMCG_VMALLOC, -1);
2738 : /*
2739 : * High-order allocs for huge vmallocs are split, so
2740 : * can be freed as an array of order-0 allocations
2741 : */
2742 42477 : __free_pages(page, 0);
2743 42477 : cond_resched();
2744 : }
2745 516 : atomic_long_sub(vm->nr_pages, &nr_vmalloc_pages);
2746 258 : kvfree(vm->pages);
2747 258 : kfree(vm);
2748 : }
2749 : EXPORT_SYMBOL(vfree);
2750 :
2751 : /**
2752 : * vunmap - release virtual mapping obtained by vmap()
2753 : * @addr: memory base address
2754 : *
2755 : * Free the virtually contiguous memory area starting at @addr,
2756 : * which was created from the page array passed to vmap().
2757 : *
2758 : * Must not be called in interrupt context.
2759 : */
2760 0 : void vunmap(const void *addr)
2761 : {
2762 : struct vm_struct *vm;
2763 :
2764 0 : BUG_ON(in_interrupt());
2765 : might_sleep();
2766 :
2767 0 : if (!addr)
2768 : return;
2769 0 : vm = remove_vm_area(addr);
2770 0 : if (unlikely(!vm)) {
2771 0 : WARN(1, KERN_ERR "Trying to vunmap() nonexistent vm area (%p)\n",
2772 : addr);
2773 0 : return;
2774 : }
2775 0 : kfree(vm);
2776 : }
2777 : EXPORT_SYMBOL(vunmap);
2778 :
2779 : /**
2780 : * vmap - map an array of pages into virtually contiguous space
2781 : * @pages: array of page pointers
2782 : * @count: number of pages to map
2783 : * @flags: vm_area->flags
2784 : * @prot: page protection for the mapping
2785 : *
2786 : * Maps @count pages from @pages into contiguous kernel virtual space.
2787 : * If @flags contains %VM_MAP_PUT_PAGES the ownership of the pages array itself
2788 : * (which must be kmalloc or vmalloc memory) and one reference per pages in it
2789 : * are transferred from the caller to vmap(), and will be freed / dropped when
2790 : * vfree() is called on the return value.
2791 : *
2792 : * Return: the address of the area or %NULL on failure
2793 : */
2794 0 : void *vmap(struct page **pages, unsigned int count,
2795 : unsigned long flags, pgprot_t prot)
2796 : {
2797 : struct vm_struct *area;
2798 : unsigned long addr;
2799 : unsigned long size; /* In bytes */
2800 :
2801 : might_sleep();
2802 :
2803 0 : if (WARN_ON_ONCE(flags & VM_FLUSH_RESET_PERMS))
2804 : return NULL;
2805 :
2806 : /*
2807 : * Your top guard is someone else's bottom guard. Not having a top
2808 : * guard compromises someone else's mappings too.
2809 : */
2810 0 : if (WARN_ON_ONCE(flags & VM_NO_GUARD))
2811 0 : flags &= ~VM_NO_GUARD;
2812 :
2813 0 : if (count > totalram_pages())
2814 : return NULL;
2815 :
2816 0 : size = (unsigned long)count << PAGE_SHIFT;
2817 0 : area = get_vm_area_caller(size, flags, __builtin_return_address(0));
2818 0 : if (!area)
2819 : return NULL;
2820 :
2821 0 : addr = (unsigned long)area->addr;
2822 0 : if (vmap_pages_range(addr, addr + size, pgprot_nx(prot),
2823 : pages, PAGE_SHIFT) < 0) {
2824 0 : vunmap(area->addr);
2825 0 : return NULL;
2826 : }
2827 :
2828 0 : if (flags & VM_MAP_PUT_PAGES) {
2829 0 : area->pages = pages;
2830 0 : area->nr_pages = count;
2831 : }
2832 0 : return area->addr;
2833 : }
2834 : EXPORT_SYMBOL(vmap);
2835 :
2836 : #ifdef CONFIG_VMAP_PFN
2837 : struct vmap_pfn_data {
2838 : unsigned long *pfns;
2839 : pgprot_t prot;
2840 : unsigned int idx;
2841 : };
2842 :
2843 : static int vmap_pfn_apply(pte_t *pte, unsigned long addr, void *private)
2844 : {
2845 : struct vmap_pfn_data *data = private;
2846 :
2847 : if (WARN_ON_ONCE(pfn_valid(data->pfns[data->idx])))
2848 : return -EINVAL;
2849 : *pte = pte_mkspecial(pfn_pte(data->pfns[data->idx++], data->prot));
2850 : return 0;
2851 : }
2852 :
2853 : /**
2854 : * vmap_pfn - map an array of PFNs into virtually contiguous space
2855 : * @pfns: array of PFNs
2856 : * @count: number of pages to map
2857 : * @prot: page protection for the mapping
2858 : *
2859 : * Maps @count PFNs from @pfns into contiguous kernel virtual space and returns
2860 : * the start address of the mapping.
2861 : */
2862 : void *vmap_pfn(unsigned long *pfns, unsigned int count, pgprot_t prot)
2863 : {
2864 : struct vmap_pfn_data data = { .pfns = pfns, .prot = pgprot_nx(prot) };
2865 : struct vm_struct *area;
2866 :
2867 : area = get_vm_area_caller(count * PAGE_SIZE, VM_IOREMAP,
2868 : __builtin_return_address(0));
2869 : if (!area)
2870 : return NULL;
2871 : if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
2872 : count * PAGE_SIZE, vmap_pfn_apply, &data)) {
2873 : free_vm_area(area);
2874 : return NULL;
2875 : }
2876 : return area->addr;
2877 : }
2878 : EXPORT_SYMBOL_GPL(vmap_pfn);
2879 : #endif /* CONFIG_VMAP_PFN */
2880 :
2881 : static inline unsigned int
2882 274 : vm_area_alloc_pages(gfp_t gfp, int nid,
2883 : unsigned int order, unsigned int nr_pages, struct page **pages)
2884 : {
2885 274 : unsigned int nr_allocated = 0;
2886 : struct page *page;
2887 : int i;
2888 :
2889 : /*
2890 : * For order-0 pages we make use of bulk allocator, if
2891 : * the page array is partly or not at all populated due
2892 : * to fails, fallback to a single page allocator that is
2893 : * more permissive.
2894 : */
2895 274 : if (!order) {
2896 274 : gfp_t bulk_gfp = gfp & ~__GFP_NOFAIL;
2897 :
2898 1144 : while (nr_allocated < nr_pages) {
2899 : unsigned int nr, nr_pages_request;
2900 :
2901 : /*
2902 : * A maximum allowed request is hard-coded and is 100
2903 : * pages per call. That is done in order to prevent a
2904 : * long preemption off scenario in the bulk-allocator
2905 : * so the range is [1:100].
2906 : */
2907 596 : nr_pages_request = min(100U, nr_pages - nr_allocated);
2908 :
2909 : /* memory allocation should consider mempolicy, we can't
2910 : * wrongly use nearest node when nid == NUMA_NO_NODE,
2911 : * otherwise memory may be allocated in only one node,
2912 : * but mempolicy wants to alloc memory by interleaving.
2913 : */
2914 : if (IS_ENABLED(CONFIG_NUMA) && nid == NUMA_NO_NODE)
2915 : nr = alloc_pages_bulk_array_mempolicy(bulk_gfp,
2916 : nr_pages_request,
2917 : pages + nr_allocated);
2918 :
2919 : else
2920 1192 : nr = alloc_pages_bulk_array_node(bulk_gfp, nid,
2921 : nr_pages_request,
2922 596 : pages + nr_allocated);
2923 :
2924 596 : nr_allocated += nr;
2925 596 : cond_resched();
2926 :
2927 : /*
2928 : * If zero or pages were obtained partly,
2929 : * fallback to a single page allocator.
2930 : */
2931 596 : if (nr != nr_pages_request)
2932 : break;
2933 : }
2934 : }
2935 :
2936 : /* High-order pages or fallback path if "bulk" fails. */
2937 :
2938 274 : while (nr_allocated < nr_pages) {
2939 0 : if (fatal_signal_pending(current))
2940 : break;
2941 :
2942 0 : if (nid == NUMA_NO_NODE)
2943 0 : page = alloc_pages(gfp, order);
2944 : else
2945 0 : page = alloc_pages_node(nid, gfp, order);
2946 0 : if (unlikely(!page))
2947 : break;
2948 : /*
2949 : * Higher order allocations must be able to be treated as
2950 : * indepdenent small pages by callers (as they can with
2951 : * small-page vmallocs). Some drivers do their own refcounting
2952 : * on vmalloc_to_page() pages, some use page->mapping,
2953 : * page->lru, etc.
2954 : */
2955 0 : if (order)
2956 0 : split_page(page, order);
2957 :
2958 : /*
2959 : * Careful, we allocate and map page-order pages, but
2960 : * tracking is done per PAGE_SIZE page so as to keep the
2961 : * vm_struct APIs independent of the physical/mapped size.
2962 : */
2963 0 : for (i = 0; i < (1U << order); i++)
2964 0 : pages[nr_allocated + i] = page + i;
2965 :
2966 0 : cond_resched();
2967 0 : nr_allocated += 1U << order;
2968 : }
2969 :
2970 274 : return nr_allocated;
2971 : }
2972 :
2973 274 : static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
2974 : pgprot_t prot, unsigned int page_shift,
2975 : int node)
2976 : {
2977 274 : const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
2978 274 : bool nofail = gfp_mask & __GFP_NOFAIL;
2979 274 : unsigned long addr = (unsigned long)area->addr;
2980 548 : unsigned long size = get_vm_area_size(area);
2981 : unsigned long array_size;
2982 274 : unsigned int nr_small_pages = size >> PAGE_SHIFT;
2983 : unsigned int page_order;
2984 : unsigned int flags;
2985 : int ret;
2986 :
2987 274 : array_size = (unsigned long)nr_small_pages * sizeof(struct page *);
2988 :
2989 274 : if (!(gfp_mask & (GFP_DMA | GFP_DMA32)))
2990 274 : gfp_mask |= __GFP_HIGHMEM;
2991 :
2992 : /* Please note that the recursion is strictly bounded. */
2993 274 : if (array_size > PAGE_SIZE) {
2994 1 : area->pages = __vmalloc_node(array_size, 1, nested_gfp, node,
2995 : area->caller);
2996 : } else {
2997 273 : area->pages = kmalloc_node(array_size, nested_gfp, node);
2998 : }
2999 :
3000 274 : if (!area->pages) {
3001 0 : warn_alloc(gfp_mask, NULL,
3002 : "vmalloc error: size %lu, failed to allocated page array size %lu",
3003 : nr_small_pages * PAGE_SIZE, array_size);
3004 0 : free_vm_area(area);
3005 0 : return NULL;
3006 : }
3007 :
3008 274 : set_vm_area_page_order(area, page_shift - PAGE_SHIFT);
3009 274 : page_order = vm_area_page_order(area);
3010 :
3011 274 : area->nr_pages = vm_area_alloc_pages(gfp_mask | __GFP_NOWARN,
3012 : node, page_order, nr_small_pages, area->pages);
3013 :
3014 548 : atomic_long_add(area->nr_pages, &nr_vmalloc_pages);
3015 274 : if (gfp_mask & __GFP_ACCOUNT) {
3016 : int i;
3017 :
3018 0 : for (i = 0; i < area->nr_pages; i++)
3019 0 : mod_memcg_page_state(area->pages[i], MEMCG_VMALLOC, 1);
3020 : }
3021 :
3022 : /*
3023 : * If not enough pages were obtained to accomplish an
3024 : * allocation request, free them via vfree() if any.
3025 : */
3026 274 : if (area->nr_pages != nr_small_pages) {
3027 0 : warn_alloc(gfp_mask, NULL,
3028 : "vmalloc error: size %lu, page order %u, failed to allocate pages",
3029 0 : area->nr_pages * PAGE_SIZE, page_order);
3030 0 : goto fail;
3031 : }
3032 :
3033 : /*
3034 : * page tables allocations ignore external gfp mask, enforce it
3035 : * by the scope API
3036 : */
3037 274 : if ((gfp_mask & (__GFP_FS | __GFP_IO)) == __GFP_IO)
3038 0 : flags = memalloc_nofs_save();
3039 274 : else if ((gfp_mask & (__GFP_FS | __GFP_IO)) == 0)
3040 0 : flags = memalloc_noio_save();
3041 :
3042 : do {
3043 274 : ret = vmap_pages_range(addr, addr + size, prot, area->pages,
3044 : page_shift);
3045 274 : if (nofail && (ret < 0))
3046 0 : schedule_timeout_uninterruptible(1);
3047 274 : } while (nofail && (ret < 0));
3048 :
3049 274 : if ((gfp_mask & (__GFP_FS | __GFP_IO)) == __GFP_IO)
3050 : memalloc_nofs_restore(flags);
3051 274 : else if ((gfp_mask & (__GFP_FS | __GFP_IO)) == 0)
3052 : memalloc_noio_restore(flags);
3053 :
3054 274 : if (ret < 0) {
3055 0 : warn_alloc(gfp_mask, NULL,
3056 : "vmalloc error: size %lu, failed to map pages",
3057 0 : area->nr_pages * PAGE_SIZE);
3058 0 : goto fail;
3059 : }
3060 :
3061 274 : return area->addr;
3062 :
3063 : fail:
3064 0 : vfree(area->addr);
3065 0 : return NULL;
3066 : }
3067 :
3068 : /**
3069 : * __vmalloc_node_range - allocate virtually contiguous memory
3070 : * @size: allocation size
3071 : * @align: desired alignment
3072 : * @start: vm area range start
3073 : * @end: vm area range end
3074 : * @gfp_mask: flags for the page level allocator
3075 : * @prot: protection mask for the allocated pages
3076 : * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD)
3077 : * @node: node to use for allocation or NUMA_NO_NODE
3078 : * @caller: caller's return address
3079 : *
3080 : * Allocate enough pages to cover @size from the page level
3081 : * allocator with @gfp_mask flags. Please note that the full set of gfp
3082 : * flags are not supported. GFP_KERNEL, GFP_NOFS and GFP_NOIO are all
3083 : * supported.
3084 : * Zone modifiers are not supported. From the reclaim modifiers
3085 : * __GFP_DIRECT_RECLAIM is required (aka GFP_NOWAIT is not supported)
3086 : * and only __GFP_NOFAIL is supported (i.e. __GFP_NORETRY and
3087 : * __GFP_RETRY_MAYFAIL are not supported).
3088 : *
3089 : * __GFP_NOWARN can be used to suppress failures messages.
3090 : *
3091 : * Map them into contiguous kernel virtual space, using a pagetable
3092 : * protection of @prot.
3093 : *
3094 : * Return: the address of the area or %NULL on failure
3095 : */
3096 274 : void *__vmalloc_node_range(unsigned long size, unsigned long align,
3097 : unsigned long start, unsigned long end, gfp_t gfp_mask,
3098 : pgprot_t prot, unsigned long vm_flags, int node,
3099 : const void *caller)
3100 : {
3101 : struct vm_struct *area;
3102 : void *ret;
3103 274 : kasan_vmalloc_flags_t kasan_flags = KASAN_VMALLOC_NONE;
3104 274 : unsigned long real_size = size;
3105 274 : unsigned long real_align = align;
3106 274 : unsigned int shift = PAGE_SHIFT;
3107 :
3108 274 : if (WARN_ON_ONCE(!size))
3109 : return NULL;
3110 :
3111 548 : if ((size >> PAGE_SHIFT) > totalram_pages()) {
3112 0 : warn_alloc(gfp_mask, NULL,
3113 : "vmalloc error: size %lu, exceeds total pages",
3114 : real_size);
3115 0 : return NULL;
3116 : }
3117 :
3118 : if (vmap_allow_huge && (vm_flags & VM_ALLOW_HUGE_VMAP)) {
3119 : unsigned long size_per_node;
3120 :
3121 : /*
3122 : * Try huge pages. Only try for PAGE_KERNEL allocations,
3123 : * others like modules don't yet expect huge pages in
3124 : * their allocations due to apply_to_page_range not
3125 : * supporting them.
3126 : */
3127 :
3128 : size_per_node = size;
3129 : if (node == NUMA_NO_NODE)
3130 : size_per_node /= num_online_nodes();
3131 : if (arch_vmap_pmd_supported(prot) && size_per_node >= PMD_SIZE)
3132 : shift = PMD_SHIFT;
3133 : else
3134 : shift = arch_vmap_pte_supported_shift(size_per_node);
3135 :
3136 : align = max(real_align, 1UL << shift);
3137 : size = ALIGN(real_size, 1UL << shift);
3138 : }
3139 :
3140 : again:
3141 274 : area = __get_vm_area_node(real_size, align, shift, VM_ALLOC |
3142 : VM_UNINITIALIZED | vm_flags, start, end, node,
3143 : gfp_mask, caller);
3144 274 : if (!area) {
3145 0 : bool nofail = gfp_mask & __GFP_NOFAIL;
3146 0 : warn_alloc(gfp_mask, NULL,
3147 : "vmalloc error: size %lu, vm_struct allocation failed%s",
3148 : real_size, (nofail) ? ". Retrying." : "");
3149 0 : if (nofail) {
3150 0 : schedule_timeout_uninterruptible(1);
3151 0 : goto again;
3152 : }
3153 : goto fail;
3154 : }
3155 :
3156 : /*
3157 : * Prepare arguments for __vmalloc_area_node() and
3158 : * kasan_unpoison_vmalloc().
3159 : */
3160 274 : if (pgprot_val(prot) == pgprot_val(PAGE_KERNEL)) {
3161 : if (kasan_hw_tags_enabled()) {
3162 : /*
3163 : * Modify protection bits to allow tagging.
3164 : * This must be done before mapping.
3165 : */
3166 : prot = arch_vmap_pgprot_tagged(prot);
3167 :
3168 : /*
3169 : * Skip page_alloc poisoning and zeroing for physical
3170 : * pages backing VM_ALLOC mapping. Memory is instead
3171 : * poisoned and zeroed by kasan_unpoison_vmalloc().
3172 : */
3173 : gfp_mask |= __GFP_SKIP_KASAN_UNPOISON | __GFP_SKIP_ZERO;
3174 : }
3175 :
3176 : /* Take note that the mapping is PAGE_KERNEL. */
3177 : kasan_flags |= KASAN_VMALLOC_PROT_NORMAL;
3178 : }
3179 :
3180 : /* Allocate physical pages and map them into vmalloc space. */
3181 274 : ret = __vmalloc_area_node(area, gfp_mask, prot, shift, node);
3182 274 : if (!ret)
3183 : goto fail;
3184 :
3185 : /*
3186 : * Mark the pages as accessible, now that they are mapped.
3187 : * The condition for setting KASAN_VMALLOC_INIT should complement the
3188 : * one in post_alloc_hook() with regards to the __GFP_SKIP_ZERO check
3189 : * to make sure that memory is initialized under the same conditions.
3190 : * Tag-based KASAN modes only assign tags to normal non-executable
3191 : * allocations, see __kasan_unpoison_vmalloc().
3192 : */
3193 274 : kasan_flags |= KASAN_VMALLOC_VM_ALLOC;
3194 548 : if (!want_init_on_free() && want_init_on_alloc(gfp_mask) &&
3195 : (gfp_mask & __GFP_SKIP_ZERO))
3196 : kasan_flags |= KASAN_VMALLOC_INIT;
3197 : /* KASAN_VMALLOC_PROT_NORMAL already set if required. */
3198 274 : area->addr = kasan_unpoison_vmalloc(area->addr, real_size, kasan_flags);
3199 :
3200 : /*
3201 : * In this function, newly allocated vm_struct has VM_UNINITIALIZED
3202 : * flag. It means that vm_struct is not fully initialized.
3203 : * Now, it is fully initialized, so remove this flag here.
3204 : */
3205 274 : clear_vm_uninitialized_flag(area);
3206 :
3207 274 : size = PAGE_ALIGN(size);
3208 : if (!(vm_flags & VM_DEFER_KMEMLEAK))
3209 274 : kmemleak_vmalloc(area, size, gfp_mask);
3210 :
3211 274 : return area->addr;
3212 :
3213 : fail:
3214 : if (shift > PAGE_SHIFT) {
3215 : shift = PAGE_SHIFT;
3216 : align = real_align;
3217 : size = real_size;
3218 : goto again;
3219 : }
3220 :
3221 : return NULL;
3222 : }
3223 :
3224 : /**
3225 : * __vmalloc_node - allocate virtually contiguous memory
3226 : * @size: allocation size
3227 : * @align: desired alignment
3228 : * @gfp_mask: flags for the page level allocator
3229 : * @node: node to use for allocation or NUMA_NO_NODE
3230 : * @caller: caller's return address
3231 : *
3232 : * Allocate enough pages to cover @size from the page level allocator with
3233 : * @gfp_mask flags. Map them into contiguous kernel virtual space.
3234 : *
3235 : * Reclaim modifiers in @gfp_mask - __GFP_NORETRY, __GFP_RETRY_MAYFAIL
3236 : * and __GFP_NOFAIL are not supported
3237 : *
3238 : * Any use of gfp flags outside of GFP_KERNEL should be consulted
3239 : * with mm people.
3240 : *
3241 : * Return: pointer to the allocated memory or %NULL on error
3242 : */
3243 258 : void *__vmalloc_node(unsigned long size, unsigned long align,
3244 : gfp_t gfp_mask, int node, const void *caller)
3245 : {
3246 258 : return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
3247 258 : gfp_mask, PAGE_KERNEL, 0, node, caller);
3248 : }
3249 : /*
3250 : * This is only for performance analysis of vmalloc and stress purpose.
3251 : * It is required by vmalloc test module, therefore do not use it other
3252 : * than that.
3253 : */
3254 : #ifdef CONFIG_TEST_VMALLOC_MODULE
3255 : EXPORT_SYMBOL_GPL(__vmalloc_node);
3256 : #endif
3257 :
3258 0 : void *__vmalloc(unsigned long size, gfp_t gfp_mask)
3259 : {
3260 0 : return __vmalloc_node(size, 1, gfp_mask, NUMA_NO_NODE,
3261 0 : __builtin_return_address(0));
3262 : }
3263 : EXPORT_SYMBOL(__vmalloc);
3264 :
3265 : /**
3266 : * vmalloc - allocate virtually contiguous memory
3267 : * @size: allocation size
3268 : *
3269 : * Allocate enough pages to cover @size from the page level
3270 : * allocator and map them into contiguous kernel virtual space.
3271 : *
3272 : * For tight control over page level allocator and protection flags
3273 : * use __vmalloc() instead.
3274 : *
3275 : * Return: pointer to the allocated memory or %NULL on error
3276 : */
3277 102 : void *vmalloc(unsigned long size)
3278 : {
3279 102 : return __vmalloc_node(size, 1, GFP_KERNEL, NUMA_NO_NODE,
3280 102 : __builtin_return_address(0));
3281 : }
3282 : EXPORT_SYMBOL(vmalloc);
3283 :
3284 : /**
3285 : * vmalloc_huge - allocate virtually contiguous memory, allow huge pages
3286 : * @size: allocation size
3287 : * @gfp_mask: flags for the page level allocator
3288 : *
3289 : * Allocate enough pages to cover @size from the page level
3290 : * allocator and map them into contiguous kernel virtual space.
3291 : * If @size is greater than or equal to PMD_SIZE, allow using
3292 : * huge pages for the memory
3293 : *
3294 : * Return: pointer to the allocated memory or %NULL on error
3295 : */
3296 0 : void *vmalloc_huge(unsigned long size, gfp_t gfp_mask)
3297 : {
3298 0 : return __vmalloc_node_range(size, 1, VMALLOC_START, VMALLOC_END,
3299 0 : gfp_mask, PAGE_KERNEL, VM_ALLOW_HUGE_VMAP,
3300 0 : NUMA_NO_NODE, __builtin_return_address(0));
3301 : }
3302 : EXPORT_SYMBOL_GPL(vmalloc_huge);
3303 :
3304 : /**
3305 : * vzalloc - allocate virtually contiguous memory with zero fill
3306 : * @size: allocation size
3307 : *
3308 : * Allocate enough pages to cover @size from the page level
3309 : * allocator and map them into contiguous kernel virtual space.
3310 : * The memory allocated is set to zero.
3311 : *
3312 : * For tight control over page level allocator and protection flags
3313 : * use __vmalloc() instead.
3314 : *
3315 : * Return: pointer to the allocated memory or %NULL on error
3316 : */
3317 155 : void *vzalloc(unsigned long size)
3318 : {
3319 155 : return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_ZERO, NUMA_NO_NODE,
3320 155 : __builtin_return_address(0));
3321 : }
3322 : EXPORT_SYMBOL(vzalloc);
3323 :
3324 : /**
3325 : * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
3326 : * @size: allocation size
3327 : *
3328 : * The resulting memory area is zeroed so it can be mapped to userspace
3329 : * without leaking data.
3330 : *
3331 : * Return: pointer to the allocated memory or %NULL on error
3332 : */
3333 0 : void *vmalloc_user(unsigned long size)
3334 : {
3335 0 : return __vmalloc_node_range(size, SHMLBA, VMALLOC_START, VMALLOC_END,
3336 0 : GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL,
3337 : VM_USERMAP, NUMA_NO_NODE,
3338 0 : __builtin_return_address(0));
3339 : }
3340 : EXPORT_SYMBOL(vmalloc_user);
3341 :
3342 : /**
3343 : * vmalloc_node - allocate memory on a specific node
3344 : * @size: allocation size
3345 : * @node: numa node
3346 : *
3347 : * Allocate enough pages to cover @size from the page level
3348 : * allocator and map them into contiguous kernel virtual space.
3349 : *
3350 : * For tight control over page level allocator and protection flags
3351 : * use __vmalloc() instead.
3352 : *
3353 : * Return: pointer to the allocated memory or %NULL on error
3354 : */
3355 0 : void *vmalloc_node(unsigned long size, int node)
3356 : {
3357 0 : return __vmalloc_node(size, 1, GFP_KERNEL, node,
3358 0 : __builtin_return_address(0));
3359 : }
3360 : EXPORT_SYMBOL(vmalloc_node);
3361 :
3362 : /**
3363 : * vzalloc_node - allocate memory on a specific node with zero fill
3364 : * @size: allocation size
3365 : * @node: numa node
3366 : *
3367 : * Allocate enough pages to cover @size from the page level
3368 : * allocator and map them into contiguous kernel virtual space.
3369 : * The memory allocated is set to zero.
3370 : *
3371 : * Return: pointer to the allocated memory or %NULL on error
3372 : */
3373 0 : void *vzalloc_node(unsigned long size, int node)
3374 : {
3375 0 : return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_ZERO, node,
3376 0 : __builtin_return_address(0));
3377 : }
3378 : EXPORT_SYMBOL(vzalloc_node);
3379 :
3380 : #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
3381 : #define GFP_VMALLOC32 (GFP_DMA32 | GFP_KERNEL)
3382 : #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
3383 : #define GFP_VMALLOC32 (GFP_DMA | GFP_KERNEL)
3384 : #else
3385 : /*
3386 : * 64b systems should always have either DMA or DMA32 zones. For others
3387 : * GFP_DMA32 should do the right thing and use the normal zone.
3388 : */
3389 : #define GFP_VMALLOC32 (GFP_DMA32 | GFP_KERNEL)
3390 : #endif
3391 :
3392 : /**
3393 : * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
3394 : * @size: allocation size
3395 : *
3396 : * Allocate enough 32bit PA addressable pages to cover @size from the
3397 : * page level allocator and map them into contiguous kernel virtual space.
3398 : *
3399 : * Return: pointer to the allocated memory or %NULL on error
3400 : */
3401 0 : void *vmalloc_32(unsigned long size)
3402 : {
3403 0 : return __vmalloc_node(size, 1, GFP_VMALLOC32, NUMA_NO_NODE,
3404 0 : __builtin_return_address(0));
3405 : }
3406 : EXPORT_SYMBOL(vmalloc_32);
3407 :
3408 : /**
3409 : * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
3410 : * @size: allocation size
3411 : *
3412 : * The resulting memory area is 32bit addressable and zeroed so it can be
3413 : * mapped to userspace without leaking data.
3414 : *
3415 : * Return: pointer to the allocated memory or %NULL on error
3416 : */
3417 0 : void *vmalloc_32_user(unsigned long size)
3418 : {
3419 0 : return __vmalloc_node_range(size, SHMLBA, VMALLOC_START, VMALLOC_END,
3420 0 : GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
3421 : VM_USERMAP, NUMA_NO_NODE,
3422 0 : __builtin_return_address(0));
3423 : }
3424 : EXPORT_SYMBOL(vmalloc_32_user);
3425 :
3426 : /*
3427 : * small helper routine , copy contents to buf from addr.
3428 : * If the page is not present, fill zero.
3429 : */
3430 :
3431 0 : static int aligned_vread(char *buf, char *addr, unsigned long count)
3432 : {
3433 : struct page *p;
3434 0 : int copied = 0;
3435 :
3436 0 : while (count) {
3437 : unsigned long offset, length;
3438 :
3439 0 : offset = offset_in_page(addr);
3440 0 : length = PAGE_SIZE - offset;
3441 0 : if (length > count)
3442 0 : length = count;
3443 0 : p = vmalloc_to_page(addr);
3444 : /*
3445 : * To do safe access to this _mapped_ area, we need
3446 : * lock. But adding lock here means that we need to add
3447 : * overhead of vmalloc()/vfree() calls for this _debug_
3448 : * interface, rarely used. Instead of that, we'll use
3449 : * kmap() and get small overhead in this access function.
3450 : */
3451 0 : if (p) {
3452 : /* We can expect USER0 is not used -- see vread() */
3453 0 : void *map = kmap_atomic(p);
3454 0 : memcpy(buf, map + offset, length);
3455 : kunmap_atomic(map);
3456 : } else
3457 0 : memset(buf, 0, length);
3458 :
3459 0 : addr += length;
3460 0 : buf += length;
3461 0 : copied += length;
3462 0 : count -= length;
3463 : }
3464 0 : return copied;
3465 : }
3466 :
3467 0 : static void vmap_ram_vread(char *buf, char *addr, int count, unsigned long flags)
3468 : {
3469 : char *start;
3470 : struct vmap_block *vb;
3471 : unsigned long offset;
3472 : unsigned int rs, re, n;
3473 :
3474 : /*
3475 : * If it's area created by vm_map_ram() interface directly, but
3476 : * not further subdividing and delegating management to vmap_block,
3477 : * handle it here.
3478 : */
3479 0 : if (!(flags & VMAP_BLOCK)) {
3480 0 : aligned_vread(buf, addr, count);
3481 0 : return;
3482 : }
3483 :
3484 : /*
3485 : * Area is split into regions and tracked with vmap_block, read out
3486 : * each region and zero fill the hole between regions.
3487 : */
3488 0 : vb = xa_load(&vmap_blocks, addr_to_vb_idx((unsigned long)addr));
3489 0 : if (!vb)
3490 : goto finished;
3491 :
3492 0 : spin_lock(&vb->lock);
3493 0 : if (bitmap_empty(vb->used_map, VMAP_BBMAP_BITS)) {
3494 0 : spin_unlock(&vb->lock);
3495 : goto finished;
3496 : }
3497 0 : for_each_set_bitrange(rs, re, vb->used_map, VMAP_BBMAP_BITS) {
3498 0 : if (!count)
3499 : break;
3500 0 : start = vmap_block_vaddr(vb->va->va_start, rs);
3501 0 : while (addr < start) {
3502 0 : if (count == 0)
3503 : goto unlock;
3504 0 : *buf = '\0';
3505 0 : buf++;
3506 0 : addr++;
3507 0 : count--;
3508 : }
3509 : /*it could start reading from the middle of used region*/
3510 0 : offset = offset_in_page(addr);
3511 0 : n = ((re - rs + 1) << PAGE_SHIFT) - offset;
3512 0 : if (n > count)
3513 0 : n = count;
3514 0 : aligned_vread(buf, start+offset, n);
3515 :
3516 0 : buf += n;
3517 0 : addr += n;
3518 0 : count -= n;
3519 : }
3520 : unlock:
3521 0 : spin_unlock(&vb->lock);
3522 :
3523 : finished:
3524 : /* zero-fill the left dirty or free regions */
3525 0 : if (count)
3526 0 : memset(buf, 0, count);
3527 : }
3528 :
3529 : /**
3530 : * vread() - read vmalloc area in a safe way.
3531 : * @buf: buffer for reading data
3532 : * @addr: vm address.
3533 : * @count: number of bytes to be read.
3534 : *
3535 : * This function checks that addr is a valid vmalloc'ed area, and
3536 : * copy data from that area to a given buffer. If the given memory range
3537 : * of [addr...addr+count) includes some valid address, data is copied to
3538 : * proper area of @buf. If there are memory holes, they'll be zero-filled.
3539 : * IOREMAP area is treated as memory hole and no copy is done.
3540 : *
3541 : * If [addr...addr+count) doesn't includes any intersects with alive
3542 : * vm_struct area, returns 0. @buf should be kernel's buffer.
3543 : *
3544 : * Note: In usual ops, vread() is never necessary because the caller
3545 : * should know vmalloc() area is valid and can use memcpy().
3546 : * This is for routines which have to access vmalloc area without
3547 : * any information, as /proc/kcore.
3548 : *
3549 : * Return: number of bytes for which addr and buf should be increased
3550 : * (same number as @count) or %0 if [addr...addr+count) doesn't
3551 : * include any intersection with valid vmalloc area
3552 : */
3553 0 : long vread(char *buf, char *addr, unsigned long count)
3554 : {
3555 : struct vmap_area *va;
3556 : struct vm_struct *vm;
3557 0 : char *vaddr, *buf_start = buf;
3558 0 : unsigned long buflen = count;
3559 : unsigned long n, size, flags;
3560 :
3561 0 : addr = kasan_reset_tag(addr);
3562 :
3563 : /* Don't allow overflow */
3564 0 : if ((unsigned long) addr + count < count)
3565 0 : count = -(unsigned long) addr;
3566 :
3567 0 : spin_lock(&vmap_area_lock);
3568 0 : va = find_vmap_area_exceed_addr((unsigned long)addr);
3569 0 : if (!va)
3570 : goto finished;
3571 :
3572 : /* no intersects with alive vmap_area */
3573 0 : if ((unsigned long)addr + count <= va->va_start)
3574 : goto finished;
3575 :
3576 0 : list_for_each_entry_from(va, &vmap_area_list, list) {
3577 0 : if (!count)
3578 : break;
3579 :
3580 0 : vm = va->vm;
3581 0 : flags = va->flags & VMAP_FLAGS_MASK;
3582 : /*
3583 : * VMAP_BLOCK indicates a sub-type of vm_map_ram area, need
3584 : * be set together with VMAP_RAM.
3585 : */
3586 0 : WARN_ON(flags == VMAP_BLOCK);
3587 :
3588 0 : if (!vm && !flags)
3589 0 : continue;
3590 :
3591 0 : if (vm && (vm->flags & VM_UNINITIALIZED))
3592 0 : continue;
3593 : /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
3594 0 : smp_rmb();
3595 :
3596 0 : vaddr = (char *) va->va_start;
3597 0 : size = vm ? get_vm_area_size(vm) : va_size(va);
3598 :
3599 0 : if (addr >= vaddr + size)
3600 0 : continue;
3601 0 : while (addr < vaddr) {
3602 0 : if (count == 0)
3603 : goto finished;
3604 0 : *buf = '\0';
3605 0 : buf++;
3606 0 : addr++;
3607 0 : count--;
3608 : }
3609 0 : n = vaddr + size - addr;
3610 0 : if (n > count)
3611 0 : n = count;
3612 :
3613 0 : if (flags & VMAP_RAM)
3614 0 : vmap_ram_vread(buf, addr, n, flags);
3615 0 : else if (!(vm->flags & VM_IOREMAP))
3616 0 : aligned_vread(buf, addr, n);
3617 : else /* IOREMAP area is treated as memory hole */
3618 0 : memset(buf, 0, n);
3619 0 : buf += n;
3620 0 : addr += n;
3621 0 : count -= n;
3622 : }
3623 : finished:
3624 0 : spin_unlock(&vmap_area_lock);
3625 :
3626 0 : if (buf == buf_start)
3627 : return 0;
3628 : /* zero-fill memory holes */
3629 0 : if (buf != buf_start + buflen)
3630 0 : memset(buf, 0, buflen - (buf - buf_start));
3631 :
3632 0 : return buflen;
3633 : }
3634 :
3635 : /**
3636 : * remap_vmalloc_range_partial - map vmalloc pages to userspace
3637 : * @vma: vma to cover
3638 : * @uaddr: target user address to start at
3639 : * @kaddr: virtual address of vmalloc kernel memory
3640 : * @pgoff: offset from @kaddr to start at
3641 : * @size: size of map area
3642 : *
3643 : * Returns: 0 for success, -Exxx on failure
3644 : *
3645 : * This function checks that @kaddr is a valid vmalloc'ed area,
3646 : * and that it is big enough to cover the range starting at
3647 : * @uaddr in @vma. Will return failure if that criteria isn't
3648 : * met.
3649 : *
3650 : * Similar to remap_pfn_range() (see mm/memory.c)
3651 : */
3652 0 : int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr,
3653 : void *kaddr, unsigned long pgoff,
3654 : unsigned long size)
3655 : {
3656 : struct vm_struct *area;
3657 : unsigned long off;
3658 : unsigned long end_index;
3659 :
3660 0 : if (check_shl_overflow(pgoff, PAGE_SHIFT, &off))
3661 : return -EINVAL;
3662 :
3663 0 : size = PAGE_ALIGN(size);
3664 :
3665 0 : if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr))
3666 : return -EINVAL;
3667 :
3668 0 : area = find_vm_area(kaddr);
3669 0 : if (!area)
3670 : return -EINVAL;
3671 :
3672 0 : if (!(area->flags & (VM_USERMAP | VM_DMA_COHERENT)))
3673 : return -EINVAL;
3674 :
3675 0 : if (check_add_overflow(size, off, &end_index) ||
3676 : end_index > get_vm_area_size(area))
3677 : return -EINVAL;
3678 0 : kaddr += off;
3679 :
3680 : do {
3681 0 : struct page *page = vmalloc_to_page(kaddr);
3682 : int ret;
3683 :
3684 0 : ret = vm_insert_page(vma, uaddr, page);
3685 0 : if (ret)
3686 : return ret;
3687 :
3688 0 : uaddr += PAGE_SIZE;
3689 0 : kaddr += PAGE_SIZE;
3690 0 : size -= PAGE_SIZE;
3691 0 : } while (size > 0);
3692 :
3693 0 : vm_flags_set(vma, VM_DONTEXPAND | VM_DONTDUMP);
3694 :
3695 0 : return 0;
3696 : }
3697 :
3698 : /**
3699 : * remap_vmalloc_range - map vmalloc pages to userspace
3700 : * @vma: vma to cover (map full range of vma)
3701 : * @addr: vmalloc memory
3702 : * @pgoff: number of pages into addr before first page to map
3703 : *
3704 : * Returns: 0 for success, -Exxx on failure
3705 : *
3706 : * This function checks that addr is a valid vmalloc'ed area, and
3707 : * that it is big enough to cover the vma. Will return failure if
3708 : * that criteria isn't met.
3709 : *
3710 : * Similar to remap_pfn_range() (see mm/memory.c)
3711 : */
3712 0 : int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
3713 : unsigned long pgoff)
3714 : {
3715 0 : return remap_vmalloc_range_partial(vma, vma->vm_start,
3716 : addr, pgoff,
3717 0 : vma->vm_end - vma->vm_start);
3718 : }
3719 : EXPORT_SYMBOL(remap_vmalloc_range);
3720 :
3721 0 : void free_vm_area(struct vm_struct *area)
3722 : {
3723 : struct vm_struct *ret;
3724 0 : ret = remove_vm_area(area->addr);
3725 0 : BUG_ON(ret != area);
3726 0 : kfree(area);
3727 0 : }
3728 : EXPORT_SYMBOL_GPL(free_vm_area);
3729 :
3730 : #ifdef CONFIG_SMP
3731 : static struct vmap_area *node_to_va(struct rb_node *n)
3732 : {
3733 : return rb_entry_safe(n, struct vmap_area, rb_node);
3734 : }
3735 :
3736 : /**
3737 : * pvm_find_va_enclose_addr - find the vmap_area @addr belongs to
3738 : * @addr: target address
3739 : *
3740 : * Returns: vmap_area if it is found. If there is no such area
3741 : * the first highest(reverse order) vmap_area is returned
3742 : * i.e. va->va_start < addr && va->va_end < addr or NULL
3743 : * if there are no any areas before @addr.
3744 : */
3745 : static struct vmap_area *
3746 : pvm_find_va_enclose_addr(unsigned long addr)
3747 : {
3748 : struct vmap_area *va, *tmp;
3749 : struct rb_node *n;
3750 :
3751 : n = free_vmap_area_root.rb_node;
3752 : va = NULL;
3753 :
3754 : while (n) {
3755 : tmp = rb_entry(n, struct vmap_area, rb_node);
3756 : if (tmp->va_start <= addr) {
3757 : va = tmp;
3758 : if (tmp->va_end >= addr)
3759 : break;
3760 :
3761 : n = n->rb_right;
3762 : } else {
3763 : n = n->rb_left;
3764 : }
3765 : }
3766 :
3767 : return va;
3768 : }
3769 :
3770 : /**
3771 : * pvm_determine_end_from_reverse - find the highest aligned address
3772 : * of free block below VMALLOC_END
3773 : * @va:
3774 : * in - the VA we start the search(reverse order);
3775 : * out - the VA with the highest aligned end address.
3776 : * @align: alignment for required highest address
3777 : *
3778 : * Returns: determined end address within vmap_area
3779 : */
3780 : static unsigned long
3781 : pvm_determine_end_from_reverse(struct vmap_area **va, unsigned long align)
3782 : {
3783 : unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
3784 : unsigned long addr;
3785 :
3786 : if (likely(*va)) {
3787 : list_for_each_entry_from_reverse((*va),
3788 : &free_vmap_area_list, list) {
3789 : addr = min((*va)->va_end & ~(align - 1), vmalloc_end);
3790 : if ((*va)->va_start < addr)
3791 : return addr;
3792 : }
3793 : }
3794 :
3795 : return 0;
3796 : }
3797 :
3798 : /**
3799 : * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
3800 : * @offsets: array containing offset of each area
3801 : * @sizes: array containing size of each area
3802 : * @nr_vms: the number of areas to allocate
3803 : * @align: alignment, all entries in @offsets and @sizes must be aligned to this
3804 : *
3805 : * Returns: kmalloc'd vm_struct pointer array pointing to allocated
3806 : * vm_structs on success, %NULL on failure
3807 : *
3808 : * Percpu allocator wants to use congruent vm areas so that it can
3809 : * maintain the offsets among percpu areas. This function allocates
3810 : * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
3811 : * be scattered pretty far, distance between two areas easily going up
3812 : * to gigabytes. To avoid interacting with regular vmallocs, these
3813 : * areas are allocated from top.
3814 : *
3815 : * Despite its complicated look, this allocator is rather simple. It
3816 : * does everything top-down and scans free blocks from the end looking
3817 : * for matching base. While scanning, if any of the areas do not fit the
3818 : * base address is pulled down to fit the area. Scanning is repeated till
3819 : * all the areas fit and then all necessary data structures are inserted
3820 : * and the result is returned.
3821 : */
3822 : struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
3823 : const size_t *sizes, int nr_vms,
3824 : size_t align)
3825 : {
3826 : const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
3827 : const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
3828 : struct vmap_area **vas, *va;
3829 : struct vm_struct **vms;
3830 : int area, area2, last_area, term_area;
3831 : unsigned long base, start, size, end, last_end, orig_start, orig_end;
3832 : bool purged = false;
3833 :
3834 : /* verify parameters and allocate data structures */
3835 : BUG_ON(offset_in_page(align) || !is_power_of_2(align));
3836 : for (last_area = 0, area = 0; area < nr_vms; area++) {
3837 : start = offsets[area];
3838 : end = start + sizes[area];
3839 :
3840 : /* is everything aligned properly? */
3841 : BUG_ON(!IS_ALIGNED(offsets[area], align));
3842 : BUG_ON(!IS_ALIGNED(sizes[area], align));
3843 :
3844 : /* detect the area with the highest address */
3845 : if (start > offsets[last_area])
3846 : last_area = area;
3847 :
3848 : for (area2 = area + 1; area2 < nr_vms; area2++) {
3849 : unsigned long start2 = offsets[area2];
3850 : unsigned long end2 = start2 + sizes[area2];
3851 :
3852 : BUG_ON(start2 < end && start < end2);
3853 : }
3854 : }
3855 : last_end = offsets[last_area] + sizes[last_area];
3856 :
3857 : if (vmalloc_end - vmalloc_start < last_end) {
3858 : WARN_ON(true);
3859 : return NULL;
3860 : }
3861 :
3862 : vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL);
3863 : vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL);
3864 : if (!vas || !vms)
3865 : goto err_free2;
3866 :
3867 : for (area = 0; area < nr_vms; area++) {
3868 : vas[area] = kmem_cache_zalloc(vmap_area_cachep, GFP_KERNEL);
3869 : vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
3870 : if (!vas[area] || !vms[area])
3871 : goto err_free;
3872 : }
3873 : retry:
3874 : spin_lock(&free_vmap_area_lock);
3875 :
3876 : /* start scanning - we scan from the top, begin with the last area */
3877 : area = term_area = last_area;
3878 : start = offsets[area];
3879 : end = start + sizes[area];
3880 :
3881 : va = pvm_find_va_enclose_addr(vmalloc_end);
3882 : base = pvm_determine_end_from_reverse(&va, align) - end;
3883 :
3884 : while (true) {
3885 : /*
3886 : * base might have underflowed, add last_end before
3887 : * comparing.
3888 : */
3889 : if (base + last_end < vmalloc_start + last_end)
3890 : goto overflow;
3891 :
3892 : /*
3893 : * Fitting base has not been found.
3894 : */
3895 : if (va == NULL)
3896 : goto overflow;
3897 :
3898 : /*
3899 : * If required width exceeds current VA block, move
3900 : * base downwards and then recheck.
3901 : */
3902 : if (base + end > va->va_end) {
3903 : base = pvm_determine_end_from_reverse(&va, align) - end;
3904 : term_area = area;
3905 : continue;
3906 : }
3907 :
3908 : /*
3909 : * If this VA does not fit, move base downwards and recheck.
3910 : */
3911 : if (base + start < va->va_start) {
3912 : va = node_to_va(rb_prev(&va->rb_node));
3913 : base = pvm_determine_end_from_reverse(&va, align) - end;
3914 : term_area = area;
3915 : continue;
3916 : }
3917 :
3918 : /*
3919 : * This area fits, move on to the previous one. If
3920 : * the previous one is the terminal one, we're done.
3921 : */
3922 : area = (area + nr_vms - 1) % nr_vms;
3923 : if (area == term_area)
3924 : break;
3925 :
3926 : start = offsets[area];
3927 : end = start + sizes[area];
3928 : va = pvm_find_va_enclose_addr(base + end);
3929 : }
3930 :
3931 : /* we've found a fitting base, insert all va's */
3932 : for (area = 0; area < nr_vms; area++) {
3933 : int ret;
3934 :
3935 : start = base + offsets[area];
3936 : size = sizes[area];
3937 :
3938 : va = pvm_find_va_enclose_addr(start);
3939 : if (WARN_ON_ONCE(va == NULL))
3940 : /* It is a BUG(), but trigger recovery instead. */
3941 : goto recovery;
3942 :
3943 : ret = adjust_va_to_fit_type(&free_vmap_area_root,
3944 : &free_vmap_area_list,
3945 : va, start, size);
3946 : if (WARN_ON_ONCE(unlikely(ret)))
3947 : /* It is a BUG(), but trigger recovery instead. */
3948 : goto recovery;
3949 :
3950 : /* Allocated area. */
3951 : va = vas[area];
3952 : va->va_start = start;
3953 : va->va_end = start + size;
3954 : }
3955 :
3956 : spin_unlock(&free_vmap_area_lock);
3957 :
3958 : /* populate the kasan shadow space */
3959 : for (area = 0; area < nr_vms; area++) {
3960 : if (kasan_populate_vmalloc(vas[area]->va_start, sizes[area]))
3961 : goto err_free_shadow;
3962 : }
3963 :
3964 : /* insert all vm's */
3965 : spin_lock(&vmap_area_lock);
3966 : for (area = 0; area < nr_vms; area++) {
3967 : insert_vmap_area(vas[area], &vmap_area_root, &vmap_area_list);
3968 :
3969 : setup_vmalloc_vm_locked(vms[area], vas[area], VM_ALLOC,
3970 : pcpu_get_vm_areas);
3971 : }
3972 : spin_unlock(&vmap_area_lock);
3973 :
3974 : /*
3975 : * Mark allocated areas as accessible. Do it now as a best-effort
3976 : * approach, as they can be mapped outside of vmalloc code.
3977 : * With hardware tag-based KASAN, marking is skipped for
3978 : * non-VM_ALLOC mappings, see __kasan_unpoison_vmalloc().
3979 : */
3980 : for (area = 0; area < nr_vms; area++)
3981 : vms[area]->addr = kasan_unpoison_vmalloc(vms[area]->addr,
3982 : vms[area]->size, KASAN_VMALLOC_PROT_NORMAL);
3983 :
3984 : kfree(vas);
3985 : return vms;
3986 :
3987 : recovery:
3988 : /*
3989 : * Remove previously allocated areas. There is no
3990 : * need in removing these areas from the busy tree,
3991 : * because they are inserted only on the final step
3992 : * and when pcpu_get_vm_areas() is success.
3993 : */
3994 : while (area--) {
3995 : orig_start = vas[area]->va_start;
3996 : orig_end = vas[area]->va_end;
3997 : va = merge_or_add_vmap_area_augment(vas[area], &free_vmap_area_root,
3998 : &free_vmap_area_list);
3999 : if (va)
4000 : kasan_release_vmalloc(orig_start, orig_end,
4001 : va->va_start, va->va_end);
4002 : vas[area] = NULL;
4003 : }
4004 :
4005 : overflow:
4006 : spin_unlock(&free_vmap_area_lock);
4007 : if (!purged) {
4008 : purge_vmap_area_lazy();
4009 : purged = true;
4010 :
4011 : /* Before "retry", check if we recover. */
4012 : for (area = 0; area < nr_vms; area++) {
4013 : if (vas[area])
4014 : continue;
4015 :
4016 : vas[area] = kmem_cache_zalloc(
4017 : vmap_area_cachep, GFP_KERNEL);
4018 : if (!vas[area])
4019 : goto err_free;
4020 : }
4021 :
4022 : goto retry;
4023 : }
4024 :
4025 : err_free:
4026 : for (area = 0; area < nr_vms; area++) {
4027 : if (vas[area])
4028 : kmem_cache_free(vmap_area_cachep, vas[area]);
4029 :
4030 : kfree(vms[area]);
4031 : }
4032 : err_free2:
4033 : kfree(vas);
4034 : kfree(vms);
4035 : return NULL;
4036 :
4037 : err_free_shadow:
4038 : spin_lock(&free_vmap_area_lock);
4039 : /*
4040 : * We release all the vmalloc shadows, even the ones for regions that
4041 : * hadn't been successfully added. This relies on kasan_release_vmalloc
4042 : * being able to tolerate this case.
4043 : */
4044 : for (area = 0; area < nr_vms; area++) {
4045 : orig_start = vas[area]->va_start;
4046 : orig_end = vas[area]->va_end;
4047 : va = merge_or_add_vmap_area_augment(vas[area], &free_vmap_area_root,
4048 : &free_vmap_area_list);
4049 : if (va)
4050 : kasan_release_vmalloc(orig_start, orig_end,
4051 : va->va_start, va->va_end);
4052 : vas[area] = NULL;
4053 : kfree(vms[area]);
4054 : }
4055 : spin_unlock(&free_vmap_area_lock);
4056 : kfree(vas);
4057 : kfree(vms);
4058 : return NULL;
4059 : }
4060 :
4061 : /**
4062 : * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
4063 : * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
4064 : * @nr_vms: the number of allocated areas
4065 : *
4066 : * Free vm_structs and the array allocated by pcpu_get_vm_areas().
4067 : */
4068 : void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
4069 : {
4070 : int i;
4071 :
4072 : for (i = 0; i < nr_vms; i++)
4073 : free_vm_area(vms[i]);
4074 : kfree(vms);
4075 : }
4076 : #endif /* CONFIG_SMP */
4077 :
4078 : #ifdef CONFIG_PRINTK
4079 0 : bool vmalloc_dump_obj(void *object)
4080 : {
4081 : struct vm_struct *vm;
4082 0 : void *objp = (void *)PAGE_ALIGN((unsigned long)object);
4083 :
4084 0 : vm = find_vm_area(objp);
4085 0 : if (!vm)
4086 : return false;
4087 0 : pr_cont(" %u-page vmalloc region starting at %#lx allocated at %pS\n",
4088 : vm->nr_pages, (unsigned long)vm->addr, vm->caller);
4089 0 : return true;
4090 : }
4091 : #endif
4092 :
4093 : #ifdef CONFIG_PROC_FS
4094 0 : static void *s_start(struct seq_file *m, loff_t *pos)
4095 : __acquires(&vmap_purge_lock)
4096 : __acquires(&vmap_area_lock)
4097 : {
4098 0 : mutex_lock(&vmap_purge_lock);
4099 0 : spin_lock(&vmap_area_lock);
4100 :
4101 0 : return seq_list_start(&vmap_area_list, *pos);
4102 : }
4103 :
4104 0 : static void *s_next(struct seq_file *m, void *p, loff_t *pos)
4105 : {
4106 0 : return seq_list_next(p, &vmap_area_list, pos);
4107 : }
4108 :
4109 0 : static void s_stop(struct seq_file *m, void *p)
4110 : __releases(&vmap_area_lock)
4111 : __releases(&vmap_purge_lock)
4112 : {
4113 0 : spin_unlock(&vmap_area_lock);
4114 0 : mutex_unlock(&vmap_purge_lock);
4115 0 : }
4116 :
4117 : static void show_numa_info(struct seq_file *m, struct vm_struct *v)
4118 : {
4119 : if (IS_ENABLED(CONFIG_NUMA)) {
4120 : unsigned int nr, *counters = m->private;
4121 : unsigned int step = 1U << vm_area_page_order(v);
4122 :
4123 : if (!counters)
4124 : return;
4125 :
4126 : if (v->flags & VM_UNINITIALIZED)
4127 : return;
4128 : /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
4129 : smp_rmb();
4130 :
4131 : memset(counters, 0, nr_node_ids * sizeof(unsigned int));
4132 :
4133 : for (nr = 0; nr < v->nr_pages; nr += step)
4134 : counters[page_to_nid(v->pages[nr])] += step;
4135 : for_each_node_state(nr, N_HIGH_MEMORY)
4136 : if (counters[nr])
4137 : seq_printf(m, " N%u=%u", nr, counters[nr]);
4138 : }
4139 : }
4140 :
4141 0 : static void show_purge_info(struct seq_file *m)
4142 : {
4143 : struct vmap_area *va;
4144 :
4145 0 : spin_lock(&purge_vmap_area_lock);
4146 0 : list_for_each_entry(va, &purge_vmap_area_list, list) {
4147 0 : seq_printf(m, "0x%pK-0x%pK %7ld unpurged vm_area\n",
4148 : (void *)va->va_start, (void *)va->va_end,
4149 0 : va->va_end - va->va_start);
4150 : }
4151 0 : spin_unlock(&purge_vmap_area_lock);
4152 0 : }
4153 :
4154 0 : static int s_show(struct seq_file *m, void *p)
4155 : {
4156 : struct vmap_area *va;
4157 : struct vm_struct *v;
4158 :
4159 0 : va = list_entry(p, struct vmap_area, list);
4160 :
4161 0 : if (!va->vm) {
4162 0 : if (va->flags & VMAP_RAM)
4163 0 : seq_printf(m, "0x%pK-0x%pK %7ld vm_map_ram\n",
4164 : (void *)va->va_start, (void *)va->va_end,
4165 0 : va->va_end - va->va_start);
4166 :
4167 : goto final;
4168 : }
4169 :
4170 0 : v = va->vm;
4171 :
4172 0 : seq_printf(m, "0x%pK-0x%pK %7ld",
4173 0 : v->addr, v->addr + v->size, v->size);
4174 :
4175 0 : if (v->caller)
4176 0 : seq_printf(m, " %pS", v->caller);
4177 :
4178 0 : if (v->nr_pages)
4179 0 : seq_printf(m, " pages=%d", v->nr_pages);
4180 :
4181 0 : if (v->phys_addr)
4182 0 : seq_printf(m, " phys=%pa", &v->phys_addr);
4183 :
4184 0 : if (v->flags & VM_IOREMAP)
4185 0 : seq_puts(m, " ioremap");
4186 :
4187 0 : if (v->flags & VM_ALLOC)
4188 0 : seq_puts(m, " vmalloc");
4189 :
4190 0 : if (v->flags & VM_MAP)
4191 0 : seq_puts(m, " vmap");
4192 :
4193 0 : if (v->flags & VM_USERMAP)
4194 0 : seq_puts(m, " user");
4195 :
4196 0 : if (v->flags & VM_DMA_COHERENT)
4197 0 : seq_puts(m, " dma-coherent");
4198 :
4199 0 : if (is_vmalloc_addr(v->pages))
4200 0 : seq_puts(m, " vpages");
4201 :
4202 0 : show_numa_info(m, v);
4203 0 : seq_putc(m, '\n');
4204 :
4205 : /*
4206 : * As a final step, dump "unpurged" areas.
4207 : */
4208 : final:
4209 0 : if (list_is_last(&va->list, &vmap_area_list))
4210 0 : show_purge_info(m);
4211 :
4212 0 : return 0;
4213 : }
4214 :
4215 : static const struct seq_operations vmalloc_op = {
4216 : .start = s_start,
4217 : .next = s_next,
4218 : .stop = s_stop,
4219 : .show = s_show,
4220 : };
4221 :
4222 1 : static int __init proc_vmalloc_init(void)
4223 : {
4224 : if (IS_ENABLED(CONFIG_NUMA))
4225 : proc_create_seq_private("vmallocinfo", 0400, NULL,
4226 : &vmalloc_op,
4227 : nr_node_ids * sizeof(unsigned int), NULL);
4228 : else
4229 1 : proc_create_seq("vmallocinfo", 0400, NULL, &vmalloc_op);
4230 1 : return 0;
4231 : }
4232 : module_init(proc_vmalloc_init);
4233 :
4234 : #endif
4235 :
4236 1 : void __init vmalloc_init(void)
4237 : {
4238 : struct vmap_area *va;
4239 : struct vm_struct *tmp;
4240 : int i;
4241 :
4242 : /*
4243 : * Create the cache for vmap_area objects.
4244 : */
4245 1 : vmap_area_cachep = KMEM_CACHE(vmap_area, SLAB_PANIC);
4246 :
4247 2 : for_each_possible_cpu(i) {
4248 : struct vmap_block_queue *vbq;
4249 : struct vfree_deferred *p;
4250 :
4251 1 : vbq = &per_cpu(vmap_block_queue, i);
4252 1 : spin_lock_init(&vbq->lock);
4253 2 : INIT_LIST_HEAD(&vbq->free);
4254 1 : p = &per_cpu(vfree_deferred, i);
4255 2 : init_llist_head(&p->list);
4256 2 : INIT_WORK(&p->wq, delayed_vfree_work);
4257 : }
4258 :
4259 : /* Import existing vmlist entries. */
4260 1 : for (tmp = vmlist; tmp; tmp = tmp->next) {
4261 0 : va = kmem_cache_zalloc(vmap_area_cachep, GFP_NOWAIT);
4262 0 : if (WARN_ON_ONCE(!va))
4263 0 : continue;
4264 :
4265 0 : va->va_start = (unsigned long)tmp->addr;
4266 0 : va->va_end = va->va_start + tmp->size;
4267 0 : va->vm = tmp;
4268 0 : insert_vmap_area(va, &vmap_area_root, &vmap_area_list);
4269 : }
4270 :
4271 : /*
4272 : * Now we can initialize a free vmap space.
4273 : */
4274 1 : vmap_init_free_space();
4275 1 : vmap_initialized = true;
4276 1 : }
|
