Annotation of sys/uvm/uvm_swap.c, Revision 1.1.1.1
1.1 nbrk 1: /* $OpenBSD: uvm_swap.c,v 1.72 2007/06/18 21:51:15 pedro Exp $ */
2: /* $NetBSD: uvm_swap.c,v 1.40 2000/11/17 11:39:39 mrg Exp $ */
3:
4: /*
5: * Copyright (c) 1995, 1996, 1997 Matthew R. Green
6: * All rights reserved.
7: *
8: * Redistribution and use in source and binary forms, with or without
9: * modification, are permitted provided that the following conditions
10: * are met:
11: * 1. Redistributions of source code must retain the above copyright
12: * notice, this list of conditions and the following disclaimer.
13: * 2. Redistributions in binary form must reproduce the above copyright
14: * notice, this list of conditions and the following disclaimer in the
15: * documentation and/or other materials provided with the distribution.
16: * 3. The name of the author may not be used to endorse or promote products
17: * derived from this software without specific prior written permission.
18: *
19: * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
20: * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
21: * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
22: * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
23: * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
24: * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
25: * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
26: * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
27: * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28: * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29: * SUCH DAMAGE.
30: *
31: * from: NetBSD: vm_swap.c,v 1.52 1997/12/02 13:47:37 pk Exp
32: * from: Id: uvm_swap.c,v 1.1.2.42 1998/02/02 20:38:06 chuck Exp
33: */
34:
35: #include <sys/param.h>
36: #include <sys/systm.h>
37: #include <sys/buf.h>
38: #include <sys/conf.h>
39: #include <sys/proc.h>
40: #include <sys/namei.h>
41: #include <sys/disklabel.h>
42: #include <sys/errno.h>
43: #include <sys/kernel.h>
44: #include <sys/malloc.h>
45: #include <sys/vnode.h>
46: #include <sys/file.h>
47: #include <sys/extent.h>
48: #include <sys/mount.h>
49: #include <sys/pool.h>
50: #include <sys/syscallargs.h>
51: #include <sys/swap.h>
52:
53: #include <uvm/uvm.h>
54: #ifdef UVM_SWAP_ENCRYPT
55: #include <sys/syslog.h>
56: #endif
57:
58: #include <miscfs/specfs/specdev.h>
59:
60: /*
61: * uvm_swap.c: manage configuration and i/o to swap space.
62: */
63:
64: /*
65: * swap space is managed in the following way:
66: *
67: * each swap partition or file is described by a "swapdev" structure.
68: * each "swapdev" structure contains a "swapent" structure which contains
69: * information that is passed up to the user (via system calls).
70: *
71: * each swap partition is assigned a "priority" (int) which controls
72: * swap partition usage.
73: *
74: * the system maintains a global data structure describing all swap
75: * partitions/files. there is a sorted LIST of "swappri" structures
76: * which describe "swapdev"'s at that priority. this LIST is headed
77: * by the "swap_priority" global var. each "swappri" contains a
78: * CIRCLEQ of "swapdev" structures at that priority.
79: *
80: * locking:
81: * - swap_syscall_lock (sleep lock): this lock serializes the swapctl
82: * system call and prevents the swap priority list from changing
83: * while we are in the middle of a system call (e.g. SWAP_STATS).
84: * - uvm.swap_data_lock (simple_lock): this lock protects all swap data
85: * structures including the priority list, the swapdev structures,
86: * and the swapmap extent.
87: *
88: * each swap device has the following info:
89: * - swap device in use (could be disabled, preventing future use)
90: * - swap enabled (allows new allocations on swap)
91: * - map info in /dev/drum
92: * - vnode pointer
93: * for swap files only:
94: * - block size
95: * - max byte count in buffer
96: * - buffer
97: * - credentials to use when doing i/o to file
98: *
99: * userland controls and configures swap with the swapctl(2) system call.
100: * the sys_swapctl performs the following operations:
101: * [1] SWAP_NSWAP: returns the number of swap devices currently configured
102: * [2] SWAP_STATS: given a pointer to an array of swapent structures
103: * (passed in via "arg") of a size passed in via "misc" ... we load
104: * the current swap config into the array.
105: * [3] SWAP_ON: given a pathname in arg (could be device or file) and a
106: * priority in "misc", start swapping on it.
107: * [4] SWAP_OFF: as SWAP_ON, but stops swapping to a device
108: * [5] SWAP_CTL: changes the priority of a swap device (new priority in
109: * "misc")
110: */
111:
112: /*
113: * swapdev: describes a single swap partition/file
114: *
115: * note the following should be true:
116: * swd_inuse <= swd_nblks [number of blocks in use is <= total blocks]
117: * swd_nblks <= swd_mapsize [because mapsize includes miniroot+disklabel]
118: */
119: struct swapdev {
120: struct swapent swd_se;
121: #define swd_dev swd_se.se_dev /* device id */
122: #define swd_flags swd_se.se_flags /* flags:inuse/enable/fake */
123: #define swd_priority swd_se.se_priority /* our priority */
124: #define swd_inuse swd_se.se_inuse /* our priority */
125: #define swd_nblks swd_se.se_nblks /* our priority */
126: char *swd_path; /* saved pathname of device */
127: int swd_pathlen; /* length of pathname */
128: int swd_npages; /* #pages we can use */
129: int swd_npginuse; /* #pages in use */
130: int swd_npgbad; /* #pages bad */
131: int swd_drumoffset; /* page0 offset in drum */
132: int swd_drumsize; /* #pages in drum */
133: struct extent *swd_ex; /* extent for this swapdev */
134: char swd_exname[12]; /* name of extent above */
135: struct vnode *swd_vp; /* backing vnode */
136: CIRCLEQ_ENTRY(swapdev) swd_next; /* priority circleq */
137:
138: int swd_bsize; /* blocksize (bytes) */
139: int swd_maxactive; /* max active i/o reqs */
140: struct buf swd_tab; /* buffer list */
141: struct ucred *swd_cred; /* cred for file access */
142: #ifdef UVM_SWAP_ENCRYPT
143: #define SWD_KEY_SHIFT 7 /* One key per 0.5 MByte */
144: #define SWD_KEY(x,y) &((x)->swd_keys[((y) - (x)->swd_drumoffset) >> SWD_KEY_SHIFT])
145:
146: #define SWD_DCRYPT_SHIFT 5
147: #define SWD_DCRYPT_BITS 32
148: #define SWD_DCRYPT_MASK (SWD_DCRYPT_BITS - 1)
149: #define SWD_DCRYPT_OFF(x) ((x) >> SWD_DCRYPT_SHIFT)
150: #define SWD_DCRYPT_BIT(x) ((x) & SWD_DCRYPT_MASK)
151: #define SWD_DCRYPT_SIZE(x) (SWD_DCRYPT_OFF((x) + SWD_DCRYPT_MASK) * sizeof(u_int32_t))
152: u_int32_t *swd_decrypt; /* bitmap for decryption */
153: struct swap_key *swd_keys; /* keys for different parts */
154: int swd_nkeys; /* active keys */
155: #endif
156: };
157:
158: /*
159: * swap device priority entry; the list is kept sorted on `spi_priority'.
160: */
161: struct swappri {
162: int spi_priority; /* priority */
163: CIRCLEQ_HEAD(spi_swapdev, swapdev) spi_swapdev;
164: /* circleq of swapdevs at this priority */
165: LIST_ENTRY(swappri) spi_swappri; /* global list of pri's */
166: };
167:
168: /*
169: * The following two structures are used to keep track of data transfers
170: * on swap devices associated with regular files.
171: * NOTE: this code is more or less a copy of vnd.c; we use the same
172: * structure names here to ease porting..
173: */
174: struct vndxfer {
175: struct buf *vx_bp; /* Pointer to parent buffer */
176: struct swapdev *vx_sdp;
177: int vx_error;
178: int vx_pending; /* # of pending aux buffers */
179: int vx_flags;
180: #define VX_BUSY 1
181: #define VX_DEAD 2
182: };
183:
184: struct vndbuf {
185: struct buf vb_buf;
186: struct vndxfer *vb_xfer;
187: };
188:
189:
190: /*
191: * We keep a of pool vndbuf's and vndxfer structures.
192: */
193: struct pool vndxfer_pool;
194: struct pool vndbuf_pool;
195:
196: #define getvndxfer(vnx) do { \
197: int s = splbio(); \
198: vnx = pool_get(&vndxfer_pool, PR_WAITOK); \
199: splx(s); \
200: } while (0)
201:
202: #define putvndxfer(vnx) { \
203: pool_put(&vndxfer_pool, (void *)(vnx)); \
204: }
205:
206: #define getvndbuf(vbp) do { \
207: int s = splbio(); \
208: vbp = pool_get(&vndbuf_pool, PR_WAITOK); \
209: splx(s); \
210: } while (0)
211:
212: #define putvndbuf(vbp) { \
213: pool_put(&vndbuf_pool, (void *)(vbp)); \
214: }
215:
216: /* /dev/drum */
217: bdev_decl(sw);
218: cdev_decl(sw);
219:
220: /*
221: * local variables
222: */
223: static struct extent *swapmap; /* controls the mapping of /dev/drum */
224:
225: /* list of all active swap devices [by priority] */
226: LIST_HEAD(swap_priority, swappri);
227: static struct swap_priority swap_priority;
228:
229: /* locks */
230: struct rwlock swap_syscall_lock = RWLOCK_INITIALIZER("swplk");
231:
232: /*
233: * prototypes
234: */
235: static void swapdrum_add(struct swapdev *, int);
236: static struct swapdev *swapdrum_getsdp(int);
237:
238: static struct swapdev *swaplist_find(struct vnode *, int);
239: static void swaplist_insert(struct swapdev *,
240: struct swappri *, int);
241: static void swaplist_trim(void);
242:
243: static int swap_on(struct proc *, struct swapdev *);
244: static int swap_off(struct proc *, struct swapdev *);
245:
246: static void sw_reg_strategy(struct swapdev *, struct buf *, int);
247: static void sw_reg_iodone(struct buf *);
248: static void sw_reg_start(struct swapdev *);
249:
250: static int uvm_swap_io(struct vm_page **, int, int, int);
251:
252: static void swapmount(void);
253:
254: #ifdef UVM_SWAP_ENCRYPT
255: /* for swap encrypt */
256: boolean_t uvm_swap_allocpages(struct vm_page **, int);
257: void uvm_swap_markdecrypt(struct swapdev *, int, int, int);
258: boolean_t uvm_swap_needdecrypt(struct swapdev *, int);
259: void uvm_swap_initcrypt(struct swapdev *, int);
260: #endif
261:
262: /*
263: * uvm_swap_init: init the swap system data structures and locks
264: *
265: * => called at boot time from init_main.c after the filesystems
266: * are brought up (which happens after uvm_init())
267: */
268: void
269: uvm_swap_init()
270: {
271: UVMHIST_FUNC("uvm_swap_init");
272:
273: UVMHIST_CALLED(pdhist);
274: /*
275: * first, init the swap list, its counter, and its lock.
276: * then get a handle on the vnode for /dev/drum by using
277: * the its dev_t number ("swapdev", from MD conf.c).
278: */
279:
280: LIST_INIT(&swap_priority);
281: uvmexp.nswapdev = 0;
282: simple_lock_init(&uvm.swap_data_lock);
283:
284: if (!swapdev_vp && bdevvp(swapdev, &swapdev_vp))
285: panic("uvm_swap_init: can't get vnode for swap device");
286:
287: /*
288: * create swap block resource map to map /dev/drum. the range
289: * from 1 to INT_MAX allows 2 gigablocks of swap space. note
290: * that block 0 is reserved (used to indicate an allocation
291: * failure, or no allocation).
292: */
293: swapmap = extent_create("swapmap", 1, INT_MAX,
294: M_VMSWAP, 0, 0, EX_NOWAIT);
295: if (swapmap == 0)
296: panic("uvm_swap_init: extent_create failed");
297:
298: /*
299: * allocate pools for structures used for swapping to files.
300: */
301:
302:
303: pool_init(&vndxfer_pool, sizeof(struct vndxfer), 0, 0, 0, "swp vnx",
304: NULL);
305:
306: pool_init(&vndbuf_pool, sizeof(struct vndbuf), 0, 0, 0, "swp vnd",
307: NULL);
308:
309: /*
310: * Setup the initial swap partition
311: */
312: swapmount();
313:
314: /*
315: * done!
316: */
317: UVMHIST_LOG(pdhist, "<- done", 0, 0, 0, 0);
318: }
319:
320: #ifdef UVM_SWAP_ENCRYPT
321: void
322: uvm_swap_initcrypt_all(void)
323: {
324: struct swapdev *sdp;
325: struct swappri *spp;
326:
327: simple_lock(&uvm.swap_data_lock);
328:
329: LIST_FOREACH(spp, &swap_priority, spi_swappri) {
330: CIRCLEQ_FOREACH(sdp, &spp->spi_swapdev, swd_next)
331: if (sdp->swd_decrypt == NULL)
332: uvm_swap_initcrypt(sdp, sdp->swd_npages);
333: }
334: simple_unlock(&uvm.swap_data_lock);
335: }
336:
337: void
338: uvm_swap_initcrypt(struct swapdev *sdp, int npages)
339: {
340: /*
341: * keep information if a page needs to be decrypted when we get it
342: * from the swap device.
343: * We cannot chance a malloc later, if we are doing ASYNC puts,
344: * we may not call malloc with M_WAITOK. This consumes only
345: * 8KB memory for a 256MB swap partition.
346: */
347: sdp->swd_decrypt = malloc(SWD_DCRYPT_SIZE(npages), M_VMSWAP, M_WAITOK);
348: memset(sdp->swd_decrypt, 0, SWD_DCRYPT_SIZE(npages));
349: sdp->swd_keys = malloc((npages >> SWD_KEY_SHIFT) * sizeof(struct swap_key),
350: M_VMSWAP, M_WAITOK);
351: memset(sdp->swd_keys, 0, (npages >> SWD_KEY_SHIFT) * sizeof(struct swap_key));
352: sdp->swd_nkeys = 0;
353: }
354:
355: boolean_t
356: uvm_swap_allocpages(struct vm_page **pps, int npages)
357: {
358: int i, s;
359: int minus, reserve;
360: boolean_t fail;
361:
362: /* Estimate if we will succeed */
363: s = uvm_lock_fpageq();
364:
365: minus = uvmexp.free - npages;
366: reserve = uvmexp.reserve_kernel;
367: fail = uvmexp.free - npages < uvmexp.reserve_kernel;
368:
369: uvm_unlock_fpageq(s);
370:
371: if (fail)
372: return FALSE;
373:
374: /* Get new pages */
375: for (i = 0; i < npages; i++) {
376: pps[i] = uvm_pagealloc(NULL, 0, NULL, 0);
377: if (pps[i] == NULL)
378: break;
379: }
380:
381: /* On failure free and return */
382: if (i < npages) {
383: uvm_swap_freepages(pps, i);
384: return FALSE;
385: }
386:
387: return TRUE;
388: }
389:
390: void
391: uvm_swap_freepages(struct vm_page **pps, int npages)
392: {
393: int i;
394:
395: uvm_lock_pageq();
396: for (i = 0; i < npages; i++)
397: uvm_pagefree(pps[i]);
398: uvm_unlock_pageq();
399: }
400:
401: /*
402: * Mark pages on the swap device for later decryption
403: */
404:
405: void
406: uvm_swap_markdecrypt(struct swapdev *sdp, int startslot, int npages,
407: int decrypt)
408: {
409: int pagestart, i;
410: int off, bit;
411:
412: if (!sdp)
413: return;
414:
415: pagestart = startslot - sdp->swd_drumoffset;
416: for (i = 0; i < npages; i++, pagestart++) {
417: off = SWD_DCRYPT_OFF(pagestart);
418: bit = SWD_DCRYPT_BIT(pagestart);
419: if (decrypt)
420: /* pages read need decryption */
421: sdp->swd_decrypt[off] |= 1 << bit;
422: else
423: /* pages read do not need decryption */
424: sdp->swd_decrypt[off] &= ~(1 << bit);
425: }
426: }
427:
428: /*
429: * Check if the page that we got from disk needs to be decrypted
430: */
431:
432: boolean_t
433: uvm_swap_needdecrypt(struct swapdev *sdp, int off)
434: {
435: if (!sdp)
436: return FALSE;
437:
438: off -= sdp->swd_drumoffset;
439: return sdp->swd_decrypt[SWD_DCRYPT_OFF(off)] & (1 << SWD_DCRYPT_BIT(off)) ?
440: TRUE : FALSE;
441: }
442: #endif /* UVM_SWAP_ENCRYPT */
443: /*
444: * swaplist functions: functions that operate on the list of swap
445: * devices on the system.
446: */
447:
448: /*
449: * swaplist_insert: insert swap device "sdp" into the global list
450: *
451: * => caller must hold both swap_syscall_lock and uvm.swap_data_lock
452: * => caller must provide a newly malloc'd swappri structure (we will
453: * FREE it if we don't need it... this it to prevent malloc blocking
454: * here while adding swap)
455: */
456: static void
457: swaplist_insert(sdp, newspp, priority)
458: struct swapdev *sdp;
459: struct swappri *newspp;
460: int priority;
461: {
462: struct swappri *spp, *pspp;
463: UVMHIST_FUNC("swaplist_insert"); UVMHIST_CALLED(pdhist);
464:
465: /*
466: * find entry at or after which to insert the new device.
467: */
468: for (pspp = NULL, spp = LIST_FIRST(&swap_priority); spp != NULL;
469: spp = LIST_NEXT(spp, spi_swappri)) {
470: if (priority <= spp->spi_priority)
471: break;
472: pspp = spp;
473: }
474:
475: /*
476: * new priority?
477: */
478: if (spp == NULL || spp->spi_priority != priority) {
479: spp = newspp; /* use newspp! */
480: UVMHIST_LOG(pdhist, "created new swappri = %ld",
481: priority, 0, 0, 0);
482:
483: spp->spi_priority = priority;
484: CIRCLEQ_INIT(&spp->spi_swapdev);
485:
486: if (pspp)
487: LIST_INSERT_AFTER(pspp, spp, spi_swappri);
488: else
489: LIST_INSERT_HEAD(&swap_priority, spp, spi_swappri);
490: } else {
491: /* we don't need a new priority structure, free it */
492: FREE(newspp, M_VMSWAP);
493: }
494:
495: /*
496: * priority found (or created). now insert on the priority's
497: * circleq list and bump the total number of swapdevs.
498: */
499: sdp->swd_priority = priority;
500: CIRCLEQ_INSERT_TAIL(&spp->spi_swapdev, sdp, swd_next);
501: uvmexp.nswapdev++;
502: }
503:
504: /*
505: * swaplist_find: find and optionally remove a swap device from the
506: * global list.
507: *
508: * => caller must hold both swap_syscall_lock and uvm.swap_data_lock
509: * => we return the swapdev we found (and removed)
510: */
511: static struct swapdev *
512: swaplist_find(vp, remove)
513: struct vnode *vp;
514: boolean_t remove;
515: {
516: struct swapdev *sdp;
517: struct swappri *spp;
518:
519: /*
520: * search the lists for the requested vp
521: */
522: for (spp = LIST_FIRST(&swap_priority); spp != NULL;
523: spp = LIST_NEXT(spp, spi_swappri)) {
524: for (sdp = CIRCLEQ_FIRST(&spp->spi_swapdev);
525: sdp != (void *)&spp->spi_swapdev;
526: sdp = CIRCLEQ_NEXT(sdp, swd_next))
527: if (sdp->swd_vp == vp) {
528: if (remove) {
529: CIRCLEQ_REMOVE(&spp->spi_swapdev,
530: sdp, swd_next);
531: uvmexp.nswapdev--;
532: }
533: return(sdp);
534: }
535: }
536: return (NULL);
537: }
538:
539:
540: /*
541: * swaplist_trim: scan priority list for empty priority entries and kill
542: * them.
543: *
544: * => caller must hold both swap_syscall_lock and uvm.swap_data_lock
545: */
546: static void
547: swaplist_trim()
548: {
549: struct swappri *spp, *nextspp;
550:
551: for (spp = LIST_FIRST(&swap_priority); spp != NULL; spp = nextspp) {
552: nextspp = LIST_NEXT(spp, spi_swappri);
553: if (CIRCLEQ_FIRST(&spp->spi_swapdev) !=
554: (void *)&spp->spi_swapdev)
555: continue;
556: LIST_REMOVE(spp, spi_swappri);
557: free(spp, M_VMSWAP);
558: }
559: }
560:
561: /*
562: * swapdrum_add: add a "swapdev"'s blocks into /dev/drum's area.
563: *
564: * => caller must hold swap_syscall_lock
565: * => uvm.swap_data_lock should be unlocked (we may sleep)
566: */
567: static void
568: swapdrum_add(sdp, npages)
569: struct swapdev *sdp;
570: int npages;
571: {
572: u_long result;
573:
574: if (extent_alloc(swapmap, npages, EX_NOALIGN, 0, EX_NOBOUNDARY,
575: EX_WAITOK, &result))
576: panic("swapdrum_add");
577:
578: sdp->swd_drumoffset = result;
579: sdp->swd_drumsize = npages;
580: }
581:
582: /*
583: * swapdrum_getsdp: given a page offset in /dev/drum, convert it back
584: * to the "swapdev" that maps that section of the drum.
585: *
586: * => each swapdev takes one big contig chunk of the drum
587: * => caller must hold uvm.swap_data_lock
588: */
589: static struct swapdev *
590: swapdrum_getsdp(pgno)
591: int pgno;
592: {
593: struct swapdev *sdp;
594: struct swappri *spp;
595:
596: for (spp = LIST_FIRST(&swap_priority); spp != NULL;
597: spp = LIST_NEXT(spp, spi_swappri))
598: for (sdp = CIRCLEQ_FIRST(&spp->spi_swapdev);
599: sdp != (void *)&spp->spi_swapdev;
600: sdp = CIRCLEQ_NEXT(sdp, swd_next))
601: if (pgno >= sdp->swd_drumoffset &&
602: pgno < (sdp->swd_drumoffset + sdp->swd_drumsize)) {
603: return sdp;
604: }
605: return NULL;
606: }
607:
608:
609: /*
610: * sys_swapctl: main entry point for swapctl(2) system call
611: * [with two helper functions: swap_on and swap_off]
612: */
613: int
614: sys_swapctl(p, v, retval)
615: struct proc *p;
616: void *v;
617: register_t *retval;
618: {
619: struct sys_swapctl_args /* {
620: syscallarg(int) cmd;
621: syscallarg(void *) arg;
622: syscallarg(int) misc;
623: } */ *uap = (struct sys_swapctl_args *)v;
624: struct vnode *vp;
625: struct nameidata nd;
626: struct swappri *spp;
627: struct swapdev *sdp;
628: struct swapent *sep;
629: char userpath[MAXPATHLEN];
630: size_t len;
631: int count, error, misc;
632: int priority;
633: UVMHIST_FUNC("sys_swapctl"); UVMHIST_CALLED(pdhist);
634:
635: misc = SCARG(uap, misc);
636:
637: /*
638: * ensure serialized syscall access by grabbing the swap_syscall_lock
639: */
640: rw_enter_write(&swap_syscall_lock);
641:
642: /*
643: * we handle the non-priv NSWAP and STATS request first.
644: *
645: * SWAP_NSWAP: return number of config'd swap devices
646: * [can also be obtained with uvmexp sysctl]
647: */
648: if (SCARG(uap, cmd) == SWAP_NSWAP) {
649: UVMHIST_LOG(pdhist, "<- done SWAP_NSWAP=%ld", uvmexp.nswapdev,
650: 0, 0, 0);
651: *retval = uvmexp.nswapdev;
652: error = 0;
653: goto out;
654: }
655:
656: /*
657: * SWAP_STATS: get stats on current # of configured swap devs
658: *
659: * note that the swap_priority list can't change as long
660: * as we are holding the swap_syscall_lock. we don't want
661: * to grab the uvm.swap_data_lock because we may fault&sleep during
662: * copyout() and we don't want to be holding that lock then!
663: */
664: if (SCARG(uap, cmd) == SWAP_STATS
665: #if defined(COMPAT_13)
666: || SCARG(uap, cmd) == SWAP_OSTATS
667: #endif
668: ) {
669: sep = (struct swapent *)SCARG(uap, arg);
670: count = 0;
671:
672: for (spp = LIST_FIRST(&swap_priority); spp != NULL;
673: spp = LIST_NEXT(spp, spi_swappri)) {
674: for (sdp = CIRCLEQ_FIRST(&spp->spi_swapdev);
675: sdp != (void *)&spp->spi_swapdev && misc-- > 0;
676: sdp = CIRCLEQ_NEXT(sdp, swd_next)) {
677: sdp->swd_inuse =
678: btodb((u_int64_t)sdp->swd_npginuse <<
679: PAGE_SHIFT);
680: error = copyout(&sdp->swd_se, sep,
681: sizeof(struct swapent));
682:
683: /* now copy out the path if necessary */
684: #if defined(COMPAT_13)
685: if (error == 0 && SCARG(uap, cmd) == SWAP_STATS)
686: #else
687: if (error == 0)
688: #endif
689: error = copyout(sdp->swd_path,
690: &sep->se_path, sdp->swd_pathlen);
691:
692: if (error)
693: goto out;
694: count++;
695: #if defined(COMPAT_13)
696: if (SCARG(uap, cmd) == SWAP_OSTATS)
697: ((struct oswapent *)sep)++;
698: else
699: #endif
700: sep++;
701: }
702: }
703:
704: UVMHIST_LOG(pdhist, "<- done SWAP_STATS", 0, 0, 0, 0);
705:
706: *retval = count;
707: error = 0;
708: goto out;
709: }
710:
711: /*
712: * all other requests require superuser privs. verify.
713: */
714: if ((error = suser(p, 0)))
715: goto out;
716:
717: /*
718: * at this point we expect a path name in arg. we will
719: * use namei() to gain a vnode reference (vref), and lock
720: * the vnode (VOP_LOCK).
721: *
722: * XXX: a NULL arg means use the root vnode pointer (e.g. for
723: * miniroot)
724: */
725: if (SCARG(uap, arg) == NULL) {
726: vp = rootvp; /* miniroot */
727: if (vget(vp, LK_EXCLUSIVE, p)) {
728: error = EBUSY;
729: goto out;
730: }
731: if (SCARG(uap, cmd) == SWAP_ON &&
732: copystr("miniroot", userpath, sizeof userpath, &len))
733: panic("swapctl: miniroot copy failed");
734: } else {
735: int space;
736: char *where;
737:
738: if (SCARG(uap, cmd) == SWAP_ON) {
739: if ((error = copyinstr(SCARG(uap, arg), userpath,
740: sizeof userpath, &len)))
741: goto out;
742: space = UIO_SYSSPACE;
743: where = userpath;
744: } else {
745: space = UIO_USERSPACE;
746: where = (char *)SCARG(uap, arg);
747: }
748: NDINIT(&nd, LOOKUP, FOLLOW|LOCKLEAF, space, where, p);
749: if ((error = namei(&nd)))
750: goto out;
751: vp = nd.ni_vp;
752: }
753: /* note: "vp" is referenced and locked */
754:
755: error = 0; /* assume no error */
756: switch(SCARG(uap, cmd)) {
757:
758: case SWAP_DUMPDEV:
759: if (vp->v_type != VBLK) {
760: error = ENOTBLK;
761: break;
762: }
763: dumpdev = vp->v_rdev;
764: break;
765:
766: case SWAP_CTL:
767: /*
768: * get new priority, remove old entry (if any) and then
769: * reinsert it in the correct place. finally, prune out
770: * any empty priority structures.
771: */
772: priority = SCARG(uap, misc);
773: spp = malloc(sizeof *spp, M_VMSWAP, M_WAITOK);
774: simple_lock(&uvm.swap_data_lock);
775: if ((sdp = swaplist_find(vp, 1)) == NULL) {
776: error = ENOENT;
777: } else {
778: swaplist_insert(sdp, spp, priority);
779: swaplist_trim();
780: }
781: simple_unlock(&uvm.swap_data_lock);
782: if (error)
783: free(spp, M_VMSWAP);
784: break;
785:
786: case SWAP_ON:
787:
788: /*
789: * check for duplicates. if none found, then insert a
790: * dummy entry on the list to prevent someone else from
791: * trying to enable this device while we are working on
792: * it.
793: */
794:
795: priority = SCARG(uap, misc);
796: simple_lock(&uvm.swap_data_lock);
797: if ((sdp = swaplist_find(vp, 0)) != NULL) {
798: error = EBUSY;
799: simple_unlock(&uvm.swap_data_lock);
800: break;
801: }
802: sdp = malloc(sizeof *sdp, M_VMSWAP, M_WAITOK);
803: spp = malloc(sizeof *spp, M_VMSWAP, M_WAITOK);
804: memset(sdp, 0, sizeof(*sdp));
805: sdp->swd_flags = SWF_FAKE; /* placeholder only */
806: sdp->swd_vp = vp;
807: sdp->swd_dev = (vp->v_type == VBLK) ? vp->v_rdev : NODEV;
808:
809: /*
810: * XXX Is NFS elaboration necessary?
811: */
812: if (vp->v_type == VREG) {
813: sdp->swd_cred = crdup(p->p_ucred);
814: }
815:
816: swaplist_insert(sdp, spp, priority);
817: simple_unlock(&uvm.swap_data_lock);
818:
819: sdp->swd_pathlen = len;
820: sdp->swd_path = malloc(sdp->swd_pathlen, M_VMSWAP, M_WAITOK);
821: if (copystr(userpath, sdp->swd_path, sdp->swd_pathlen, 0) != 0)
822: panic("swapctl: copystr");
823:
824: /*
825: * we've now got a FAKE placeholder in the swap list.
826: * now attempt to enable swap on it. if we fail, undo
827: * what we've done and kill the fake entry we just inserted.
828: * if swap_on is a success, it will clear the SWF_FAKE flag
829: */
830:
831: if ((error = swap_on(p, sdp)) != 0) {
832: simple_lock(&uvm.swap_data_lock);
833: (void) swaplist_find(vp, 1); /* kill fake entry */
834: swaplist_trim();
835: simple_unlock(&uvm.swap_data_lock);
836: if (vp->v_type == VREG) {
837: crfree(sdp->swd_cred);
838: }
839: free(sdp->swd_path, M_VMSWAP);
840: free(sdp, M_VMSWAP);
841: break;
842: }
843: break;
844:
845: case SWAP_OFF:
846: simple_lock(&uvm.swap_data_lock);
847: if ((sdp = swaplist_find(vp, 0)) == NULL) {
848: simple_unlock(&uvm.swap_data_lock);
849: error = ENXIO;
850: break;
851: }
852:
853: /*
854: * If a device isn't in use or enabled, we
855: * can't stop swapping from it (again).
856: */
857: if ((sdp->swd_flags & (SWF_INUSE|SWF_ENABLE)) == 0) {
858: simple_unlock(&uvm.swap_data_lock);
859: error = EBUSY;
860: break;
861: }
862:
863: /*
864: * do the real work.
865: */
866: error = swap_off(p, sdp);
867: break;
868:
869: default:
870: error = EINVAL;
871: }
872:
873: /*
874: * done! release the ref gained by namei() and unlock.
875: */
876: vput(vp);
877:
878: out:
879: rw_exit_write(&swap_syscall_lock);
880:
881: UVMHIST_LOG(pdhist, "<- done! error=%ld", error, 0, 0, 0);
882: return (error);
883: }
884:
885: /*
886: * swap_on: attempt to enable a swapdev for swapping. note that the
887: * swapdev is already on the global list, but disabled (marked
888: * SWF_FAKE).
889: *
890: * => we avoid the start of the disk (to protect disk labels)
891: * => we also avoid the miniroot, if we are swapping to root.
892: * => caller should leave uvm.swap_data_lock unlocked, we may lock it
893: * if needed.
894: */
895: static int
896: swap_on(p, sdp)
897: struct proc *p;
898: struct swapdev *sdp;
899: {
900: static int count = 0; /* static */
901: struct vnode *vp;
902: int error, npages, nblocks, size;
903: long addr;
904: struct vattr va;
905: #if defined(NFSCLIENT)
906: extern int (**nfsv2_vnodeop_p)(void *);
907: #endif /* defined(NFSCLIENT) */
908: dev_t dev;
909: UVMHIST_FUNC("swap_on"); UVMHIST_CALLED(pdhist);
910:
911: /*
912: * we want to enable swapping on sdp. the swd_vp contains
913: * the vnode we want (locked and ref'd), and the swd_dev
914: * contains the dev_t of the file, if it a block device.
915: */
916:
917: vp = sdp->swd_vp;
918: dev = sdp->swd_dev;
919:
920: /*
921: * open the swap file (mostly useful for block device files to
922: * let device driver know what is up).
923: *
924: * we skip the open/close for root on swap because the root
925: * has already been opened when root was mounted (mountroot).
926: */
927: if (vp != rootvp) {
928: if ((error = VOP_OPEN(vp, FREAD|FWRITE, p->p_ucred, p)))
929: return (error);
930: }
931:
932: /* XXX this only works for block devices */
933: UVMHIST_LOG(pdhist, " dev=%ld, major(dev)=%ld", dev, major(dev), 0,0);
934:
935: /*
936: * we now need to determine the size of the swap area. for
937: * block specials we can call the d_psize function.
938: * for normal files, we must stat [get attrs].
939: *
940: * we put the result in nblks.
941: * for normal files, we also want the filesystem block size
942: * (which we get with statfs).
943: */
944: switch (vp->v_type) {
945: case VBLK:
946: if (bdevsw[major(dev)].d_psize == 0 ||
947: (nblocks = (*bdevsw[major(dev)].d_psize)(dev)) == -1) {
948: error = ENXIO;
949: goto bad;
950: }
951: break;
952:
953: case VREG:
954: if ((error = VOP_GETATTR(vp, &va, p->p_ucred, p)))
955: goto bad;
956: nblocks = (int)btodb(va.va_size);
957: if ((error =
958: VFS_STATFS(vp->v_mount, &vp->v_mount->mnt_stat, p)) != 0)
959: goto bad;
960:
961: sdp->swd_bsize = vp->v_mount->mnt_stat.f_iosize;
962: /*
963: * limit the max # of outstanding I/O requests we issue
964: * at any one time. take it easy on NFS servers.
965: */
966: #if defined(NFSCLIENT)
967: if (vp->v_op == nfsv2_vnodeop_p)
968: sdp->swd_maxactive = 2; /* XXX */
969: else
970: #endif /* defined(NFSCLIENT) */
971: sdp->swd_maxactive = 8; /* XXX */
972: break;
973:
974: default:
975: error = ENXIO;
976: goto bad;
977: }
978:
979: /*
980: * save nblocks in a safe place and convert to pages.
981: */
982:
983: sdp->swd_nblks = nblocks;
984: npages = dbtob((u_int64_t)nblocks) >> PAGE_SHIFT;
985:
986: /*
987: * for block special files, we want to make sure that leave
988: * the disklabel and bootblocks alone, so we arrange to skip
989: * over them (arbitrarily choosing to skip PAGE_SIZE bytes).
990: * note that because of this the "size" can be less than the
991: * actual number of blocks on the device.
992: */
993: if (vp->v_type == VBLK) {
994: /* we use pages 1 to (size - 1) [inclusive] */
995: size = npages - 1;
996: addr = 1;
997: } else {
998: /* we use pages 0 to (size - 1) [inclusive] */
999: size = npages;
1000: addr = 0;
1001: }
1002:
1003: /*
1004: * make sure we have enough blocks for a reasonable sized swap
1005: * area. we want at least one page.
1006: */
1007:
1008: if (size < 1) {
1009: UVMHIST_LOG(pdhist, " size <= 1!!", 0, 0, 0, 0);
1010: error = EINVAL;
1011: goto bad;
1012: }
1013:
1014: UVMHIST_LOG(pdhist, " dev=%lx: size=%ld addr=0x%lx\n",
1015: dev, size, addr, 0);
1016:
1017: /*
1018: * now we need to allocate an extent to manage this swap device
1019: */
1020: snprintf(sdp->swd_exname, sizeof(sdp->swd_exname), "swap0x%04x",
1021: count++);
1022:
1023: /* note that extent_create's 3rd arg is inclusive, thus "- 1" */
1024: sdp->swd_ex = extent_create(sdp->swd_exname, 0, npages - 1, M_VMSWAP,
1025: 0, 0, EX_WAITOK);
1026: /* allocate the `saved' region from the extent so it won't be used */
1027: if (addr) {
1028: if (extent_alloc_region(sdp->swd_ex, 0, addr, EX_WAITOK))
1029: panic("disklabel region");
1030: }
1031:
1032: /*
1033: * if the vnode we are swapping to is the root vnode
1034: * (i.e. we are swapping to the miniroot) then we want
1035: * to make sure we don't overwrite it. do a statfs to
1036: * find its size and skip over it.
1037: */
1038: if (vp == rootvp) {
1039: struct mount *mp;
1040: struct statfs *sp;
1041: int rootblocks, rootpages;
1042:
1043: mp = rootvnode->v_mount;
1044: sp = &mp->mnt_stat;
1045: rootblocks = sp->f_blocks * btodb(sp->f_bsize);
1046: rootpages = round_page(dbtob(rootblocks)) >> PAGE_SHIFT;
1047: if (rootpages > size)
1048: panic("swap_on: miniroot larger than swap?");
1049:
1050: if (extent_alloc_region(sdp->swd_ex, addr,
1051: rootpages, EX_WAITOK))
1052: panic("swap_on: unable to preserve miniroot");
1053:
1054: size -= rootpages;
1055: printf("Preserved %d pages of miniroot ", rootpages);
1056: printf("leaving %d pages of swap\n", size);
1057: }
1058:
1059: /*
1060: * add a ref to vp to reflect usage as a swap device.
1061: */
1062: vref(vp);
1063:
1064: #ifdef UVM_SWAP_ENCRYPT
1065: if (uvm_doswapencrypt)
1066: uvm_swap_initcrypt(sdp, npages);
1067: #endif
1068: /*
1069: * now add the new swapdev to the drum and enable.
1070: */
1071: simple_lock(&uvm.swap_data_lock);
1072: swapdrum_add(sdp, npages);
1073: sdp->swd_npages = size;
1074: sdp->swd_flags &= ~SWF_FAKE; /* going live */
1075: sdp->swd_flags |= (SWF_INUSE|SWF_ENABLE);
1076: uvmexp.swpages += size;
1077: simple_unlock(&uvm.swap_data_lock);
1078: return (0);
1079:
1080: bad:
1081: /*
1082: * failure: close device if necessary and return error.
1083: */
1084: if (vp != rootvp)
1085: (void)VOP_CLOSE(vp, FREAD|FWRITE, p->p_ucred, p);
1086: return (error);
1087: }
1088:
1089: /*
1090: * swap_off: stop swapping on swapdev
1091: *
1092: * => swap data should be locked, we will unlock.
1093: */
1094: static int
1095: swap_off(p, sdp)
1096: struct proc *p;
1097: struct swapdev *sdp;
1098: {
1099: int error;
1100: UVMHIST_FUNC("swap_off"); UVMHIST_CALLED(pdhist);
1101: UVMHIST_LOG(pdhist, " dev=%lx", sdp->swd_dev,0,0,0);
1102:
1103: /* disable the swap area being removed */
1104: sdp->swd_flags &= ~SWF_ENABLE;
1105: simple_unlock(&uvm.swap_data_lock);
1106:
1107: /*
1108: * the idea is to find all the pages that are paged out to this
1109: * device, and page them all in. in uvm, swap-backed pageable
1110: * memory can take two forms: aobjs and anons. call the
1111: * swapoff hook for each subsystem to bring in pages.
1112: */
1113:
1114: if (uao_swap_off(sdp->swd_drumoffset,
1115: sdp->swd_drumoffset + sdp->swd_drumsize) ||
1116: amap_swap_off(sdp->swd_drumoffset,
1117: sdp->swd_drumoffset + sdp->swd_drumsize)) {
1118:
1119: error = ENOMEM;
1120: } else if (sdp->swd_npginuse > sdp->swd_npgbad) {
1121: error = EBUSY;
1122: }
1123:
1124: if (error) {
1125: simple_lock(&uvm.swap_data_lock);
1126: sdp->swd_flags |= SWF_ENABLE;
1127: simple_unlock(&uvm.swap_data_lock);
1128: return (error);
1129: }
1130:
1131: /*
1132: * done with the vnode and saved creds.
1133: * drop our ref on the vnode before calling VOP_CLOSE()
1134: * so that spec_close() can tell if this is the last close.
1135: */
1136: if (sdp->swd_vp->v_type == VREG) {
1137: crfree(sdp->swd_cred);
1138: }
1139: vrele(sdp->swd_vp);
1140: if (sdp->swd_vp != rootvp) {
1141: (void) VOP_CLOSE(sdp->swd_vp, FREAD|FWRITE, p->p_ucred, p);
1142: }
1143:
1144: simple_lock(&uvm.swap_data_lock);
1145: uvmexp.swpages -= sdp->swd_npages;
1146:
1147: if (swaplist_find(sdp->swd_vp, 1) == NULL)
1148: panic("swap_off: swapdev not in list");
1149: swaplist_trim();
1150:
1151: /*
1152: * free all resources!
1153: */
1154: extent_free(swapmap, sdp->swd_drumoffset, sdp->swd_drumsize,
1155: EX_WAITOK);
1156: extent_destroy(sdp->swd_ex);
1157: free(sdp, M_VMSWAP);
1158: simple_unlock(&uvm.swap_data_lock);
1159: return (0);
1160: }
1161:
1162: /*
1163: * /dev/drum interface and i/o functions
1164: */
1165:
1166: /*
1167: * swread: the read function for the drum (just a call to physio)
1168: */
1169: /*ARGSUSED*/
1170: int
1171: swread(dev, uio, ioflag)
1172: dev_t dev;
1173: struct uio *uio;
1174: int ioflag;
1175: {
1176: UVMHIST_FUNC("swread"); UVMHIST_CALLED(pdhist);
1177:
1178: UVMHIST_LOG(pdhist, " dev=%lx offset=%lx",
1179: dev, (u_long)uio->uio_offset, 0, 0);
1180: return (physio(swstrategy, NULL, dev, B_READ, minphys, uio));
1181: }
1182:
1183: /*
1184: * swwrite: the write function for the drum (just a call to physio)
1185: */
1186: /*ARGSUSED*/
1187: int
1188: swwrite(dev, uio, ioflag)
1189: dev_t dev;
1190: struct uio *uio;
1191: int ioflag;
1192: {
1193: UVMHIST_FUNC("swwrite"); UVMHIST_CALLED(pdhist);
1194:
1195: UVMHIST_LOG(pdhist, " dev=%lx offset=%lx",
1196: dev, (u_long)uio->uio_offset, 0, 0);
1197: return (physio(swstrategy, NULL, dev, B_WRITE, minphys, uio));
1198: }
1199:
1200: /*
1201: * swstrategy: perform I/O on the drum
1202: *
1203: * => we must map the i/o request from the drum to the correct swapdev.
1204: */
1205: void
1206: swstrategy(bp)
1207: struct buf *bp;
1208: {
1209: struct swapdev *sdp;
1210: int s, pageno, bn;
1211: UVMHIST_FUNC("swstrategy"); UVMHIST_CALLED(pdhist);
1212:
1213: /*
1214: * convert block number to swapdev. note that swapdev can't
1215: * be yanked out from under us because we are holding resources
1216: * in it (i.e. the blocks we are doing I/O on).
1217: */
1218: pageno = dbtob((int64_t)bp->b_blkno) >> PAGE_SHIFT;
1219: simple_lock(&uvm.swap_data_lock);
1220: sdp = swapdrum_getsdp(pageno);
1221: simple_unlock(&uvm.swap_data_lock);
1222: if (sdp == NULL) {
1223: bp->b_error = EINVAL;
1224: bp->b_flags |= B_ERROR;
1225: s = splbio();
1226: biodone(bp);
1227: splx(s);
1228: UVMHIST_LOG(pdhist, " failed to get swap device", 0, 0, 0, 0);
1229: return;
1230: }
1231:
1232: /*
1233: * convert drum page number to block number on this swapdev.
1234: */
1235:
1236: pageno -= sdp->swd_drumoffset; /* page # on swapdev */
1237: bn = btodb((u_int64_t)pageno << PAGE_SHIFT); /* convert to diskblock */
1238:
1239: UVMHIST_LOG(pdhist, " %s: mapoff=%lx bn=0x%lx bcount=%ld",
1240: ((bp->b_flags & B_READ) == 0) ? "write" : "read",
1241: sdp->swd_drumoffset, bn, bp->b_bcount);
1242:
1243: /*
1244: * for block devices we finish up here.
1245: * for regular files we have to do more work which we delegate
1246: * to sw_reg_strategy().
1247: */
1248:
1249: switch (sdp->swd_vp->v_type) {
1250: default:
1251: panic("swstrategy: vnode type 0x%x", sdp->swd_vp->v_type);
1252:
1253: case VBLK:
1254:
1255: /*
1256: * must convert "bp" from an I/O on /dev/drum to an I/O
1257: * on the swapdev (sdp).
1258: */
1259: s = splbio();
1260: buf_replacevnode(bp, sdp->swd_vp);
1261:
1262: bp->b_blkno = bn;
1263: splx(s);
1264: VOP_STRATEGY(bp);
1265: return;
1266:
1267: case VREG:
1268: /*
1269: * delegate to sw_reg_strategy function.
1270: */
1271: sw_reg_strategy(sdp, bp, bn);
1272: return;
1273: }
1274: /* NOTREACHED */
1275: }
1276:
1277: /*
1278: * sw_reg_strategy: handle swap i/o to regular files
1279: */
1280: static void
1281: sw_reg_strategy(sdp, bp, bn)
1282: struct swapdev *sdp;
1283: struct buf *bp;
1284: int bn;
1285: {
1286: struct vnode *vp;
1287: struct vndxfer *vnx;
1288: daddr64_t nbn;
1289: caddr_t addr;
1290: off_t byteoff;
1291: int s, off, nra, error, sz, resid;
1292: UVMHIST_FUNC("sw_reg_strategy"); UVMHIST_CALLED(pdhist);
1293:
1294: /*
1295: * allocate a vndxfer head for this transfer and point it to
1296: * our buffer.
1297: */
1298: getvndxfer(vnx);
1299: vnx->vx_flags = VX_BUSY;
1300: vnx->vx_error = 0;
1301: vnx->vx_pending = 0;
1302: vnx->vx_bp = bp;
1303: vnx->vx_sdp = sdp;
1304:
1305: /*
1306: * setup for main loop where we read filesystem blocks into
1307: * our buffer.
1308: */
1309: error = 0;
1310: bp->b_resid = bp->b_bcount; /* nothing transferred yet! */
1311: addr = bp->b_data; /* current position in buffer */
1312: byteoff = dbtob((u_int64_t)bn);
1313:
1314: for (resid = bp->b_resid; resid; resid -= sz) {
1315: struct vndbuf *nbp;
1316:
1317: /*
1318: * translate byteoffset into block number. return values:
1319: * vp = vnode of underlying device
1320: * nbn = new block number (on underlying vnode dev)
1321: * nra = num blocks we can read-ahead (excludes requested
1322: * block)
1323: */
1324: nra = 0;
1325: error = VOP_BMAP(sdp->swd_vp, byteoff / sdp->swd_bsize,
1326: &vp, &nbn, &nra);
1327:
1328: if (error == 0 && nbn == (daddr64_t)-1) {
1329: /*
1330: * this used to just set error, but that doesn't
1331: * do the right thing. Instead, it causes random
1332: * memory errors. The panic() should remain until
1333: * this condition doesn't destabilize the system.
1334: */
1335: #if 1
1336: panic("sw_reg_strategy: swap to sparse file");
1337: #else
1338: error = EIO; /* failure */
1339: #endif
1340: }
1341:
1342: /*
1343: * punt if there was an error or a hole in the file.
1344: * we must wait for any i/o ops we have already started
1345: * to finish before returning.
1346: *
1347: * XXX we could deal with holes here but it would be
1348: * a hassle (in the write case).
1349: */
1350: if (error) {
1351: s = splbio();
1352: vnx->vx_error = error; /* pass error up */
1353: goto out;
1354: }
1355:
1356: /*
1357: * compute the size ("sz") of this transfer (in bytes).
1358: */
1359: off = byteoff % sdp->swd_bsize;
1360: sz = (1 + nra) * sdp->swd_bsize - off;
1361: if (sz > resid)
1362: sz = resid;
1363:
1364: UVMHIST_LOG(pdhist, "sw_reg_strategy: "
1365: "vp %p/%p offset 0x%lx/0x%llx",
1366: sdp->swd_vp, vp, (u_long)byteoff, nbn);
1367:
1368: /*
1369: * now get a buf structure. note that the vb_buf is
1370: * at the front of the nbp structure so that you can
1371: * cast pointers between the two structure easily.
1372: */
1373: getvndbuf(nbp);
1374: nbp->vb_buf.b_flags = bp->b_flags | B_CALL;
1375: nbp->vb_buf.b_bcount = sz;
1376: nbp->vb_buf.b_bufsize = sz;
1377: nbp->vb_buf.b_error = 0;
1378: nbp->vb_buf.b_data = addr;
1379: nbp->vb_buf.b_blkno = nbn + btodb(off);
1380: nbp->vb_buf.b_proc = bp->b_proc;
1381: nbp->vb_buf.b_iodone = sw_reg_iodone;
1382: nbp->vb_buf.b_vp = NULLVP;
1383: nbp->vb_buf.b_vnbufs.le_next = NOLIST;
1384: LIST_INIT(&nbp->vb_buf.b_dep);
1385:
1386: /*
1387: * set b_dirtyoff/end and b_validoff/end. this is
1388: * required by the NFS client code (otherwise it will
1389: * just discard our I/O request).
1390: */
1391: if (bp->b_dirtyend == 0) {
1392: nbp->vb_buf.b_dirtyoff = 0;
1393: nbp->vb_buf.b_dirtyend = sz;
1394: } else {
1395: nbp->vb_buf.b_dirtyoff =
1396: max(0, bp->b_dirtyoff - (bp->b_bcount-resid));
1397: nbp->vb_buf.b_dirtyend =
1398: min(sz,
1399: max(0, bp->b_dirtyend - (bp->b_bcount-resid)));
1400: }
1401: if (bp->b_validend == 0) {
1402: nbp->vb_buf.b_validoff = 0;
1403: nbp->vb_buf.b_validend = sz;
1404: } else {
1405: nbp->vb_buf.b_validoff =
1406: max(0, bp->b_validoff - (bp->b_bcount-resid));
1407: nbp->vb_buf.b_validend =
1408: min(sz,
1409: max(0, bp->b_validend - (bp->b_bcount-resid)));
1410: }
1411:
1412: nbp->vb_xfer = vnx; /* patch it back in to vnx */
1413:
1414: /*
1415: * Just sort by block number
1416: */
1417: nbp->vb_buf.b_cylinder = nbp->vb_buf.b_blkno;
1418: s = splbio();
1419: if (vnx->vx_error != 0) {
1420: putvndbuf(nbp);
1421: goto out;
1422: }
1423: vnx->vx_pending++;
1424:
1425: /* assoc new buffer with underlying vnode */
1426: bgetvp(vp, &nbp->vb_buf);
1427:
1428: /* sort it in and start I/O if we are not over our limit */
1429: disksort(&sdp->swd_tab, &nbp->vb_buf);
1430: sw_reg_start(sdp);
1431: splx(s);
1432:
1433: /*
1434: * advance to the next I/O
1435: */
1436: byteoff += sz;
1437: addr += sz;
1438: }
1439:
1440: s = splbio();
1441:
1442: out: /* Arrive here at splbio */
1443: vnx->vx_flags &= ~VX_BUSY;
1444: if (vnx->vx_pending == 0) {
1445: if (vnx->vx_error != 0) {
1446: bp->b_error = vnx->vx_error;
1447: bp->b_flags |= B_ERROR;
1448: }
1449: putvndxfer(vnx);
1450: biodone(bp);
1451: }
1452: splx(s);
1453: }
1454:
1455: /*
1456: * sw_reg_start: start an I/O request on the requested swapdev
1457: *
1458: * => reqs are sorted by disksort (above)
1459: */
1460: static void
1461: sw_reg_start(sdp)
1462: struct swapdev *sdp;
1463: {
1464: struct buf *bp;
1465: UVMHIST_FUNC("sw_reg_start"); UVMHIST_CALLED(pdhist);
1466:
1467: /* recursion control */
1468: if ((sdp->swd_flags & SWF_BUSY) != 0)
1469: return;
1470:
1471: sdp->swd_flags |= SWF_BUSY;
1472:
1473: while (sdp->swd_tab.b_active < sdp->swd_maxactive) {
1474: bp = sdp->swd_tab.b_actf;
1475: if (bp == NULL)
1476: break;
1477: sdp->swd_tab.b_actf = bp->b_actf;
1478: sdp->swd_tab.b_active++;
1479:
1480: UVMHIST_LOG(pdhist,
1481: "sw_reg_start: bp %p vp %p blkno 0x%lx cnt 0x%lx",
1482: bp, bp->b_vp, bp->b_blkno, bp->b_bcount);
1483: if ((bp->b_flags & B_READ) == 0)
1484: bp->b_vp->v_numoutput++;
1485:
1486: VOP_STRATEGY(bp);
1487: }
1488: sdp->swd_flags &= ~SWF_BUSY;
1489: }
1490:
1491: /*
1492: * sw_reg_iodone: one of our i/o's has completed and needs post-i/o cleanup
1493: *
1494: * => note that we can recover the vndbuf struct by casting the buf ptr
1495: */
1496: static void
1497: sw_reg_iodone(bp)
1498: struct buf *bp;
1499: {
1500: struct vndbuf *vbp = (struct vndbuf *) bp;
1501: struct vndxfer *vnx = vbp->vb_xfer;
1502: struct buf *pbp = vnx->vx_bp; /* parent buffer */
1503: struct swapdev *sdp = vnx->vx_sdp;
1504: int resid;
1505: UVMHIST_FUNC("sw_reg_iodone"); UVMHIST_CALLED(pdhist);
1506:
1507: UVMHIST_LOG(pdhist, " vbp=%p vp=%p blkno=0x%lx addr=%p",
1508: vbp, vbp->vb_buf.b_vp, vbp->vb_buf.b_blkno, vbp->vb_buf.b_data);
1509: UVMHIST_LOG(pdhist, " cnt=%lx resid=%lx",
1510: vbp->vb_buf.b_bcount, vbp->vb_buf.b_resid, 0, 0);
1511:
1512: splassert(IPL_BIO);
1513:
1514: resid = vbp->vb_buf.b_bcount - vbp->vb_buf.b_resid;
1515: pbp->b_resid -= resid;
1516: vnx->vx_pending--;
1517:
1518: if (vbp->vb_buf.b_error) {
1519: UVMHIST_LOG(pdhist, " got error=%ld !",
1520: vbp->vb_buf.b_error, 0, 0, 0);
1521:
1522: /* pass error upward */
1523: vnx->vx_error = vbp->vb_buf.b_error;
1524: }
1525:
1526: /*
1527: * disassociate this buffer from the vnode (if any).
1528: */
1529: if (vbp->vb_buf.b_vp != NULL) {
1530: brelvp(&vbp->vb_buf);
1531: }
1532:
1533: /*
1534: * kill vbp structure
1535: */
1536: putvndbuf(vbp);
1537:
1538: /*
1539: * wrap up this transaction if it has run to completion or, in
1540: * case of an error, when all auxiliary buffers have returned.
1541: */
1542: if (vnx->vx_error != 0) {
1543: /* pass error upward */
1544: pbp->b_flags |= B_ERROR;
1545: pbp->b_error = vnx->vx_error;
1546: if ((vnx->vx_flags & VX_BUSY) == 0 && vnx->vx_pending == 0) {
1547: putvndxfer(vnx);
1548: biodone(pbp);
1549: }
1550: } else if (pbp->b_resid == 0) {
1551: KASSERT(vnx->vx_pending == 0);
1552: if ((vnx->vx_flags & VX_BUSY) == 0) {
1553: UVMHIST_LOG(pdhist, " iodone error=%ld !",
1554: pbp, vnx->vx_error, 0, 0);
1555: putvndxfer(vnx);
1556: biodone(pbp);
1557: }
1558: }
1559:
1560: /*
1561: * done! start next swapdev I/O if one is pending
1562: */
1563: sdp->swd_tab.b_active--;
1564: sw_reg_start(sdp);
1565: }
1566:
1567:
1568: /*
1569: * uvm_swap_alloc: allocate space on swap
1570: *
1571: * => allocation is done "round robin" down the priority list, as we
1572: * allocate in a priority we "rotate" the circle queue.
1573: * => space can be freed with uvm_swap_free
1574: * => we return the page slot number in /dev/drum (0 == invalid slot)
1575: * => we lock uvm.swap_data_lock
1576: * => XXXMRG: "LESSOK" INTERFACE NEEDED TO EXTENT SYSTEM
1577: */
1578: int
1579: uvm_swap_alloc(nslots, lessok)
1580: int *nslots; /* IN/OUT */
1581: boolean_t lessok;
1582: {
1583: struct swapdev *sdp;
1584: struct swappri *spp;
1585: u_long result;
1586: UVMHIST_FUNC("uvm_swap_alloc"); UVMHIST_CALLED(pdhist);
1587:
1588: /*
1589: * no swap devices configured yet? definite failure.
1590: */
1591: if (uvmexp.nswapdev < 1)
1592: return 0;
1593:
1594: /*
1595: * lock data lock, convert slots into blocks, and enter loop
1596: */
1597: simple_lock(&uvm.swap_data_lock);
1598:
1599: ReTry: /* XXXMRG */
1600: for (spp = LIST_FIRST(&swap_priority); spp != NULL;
1601: spp = LIST_NEXT(spp, spi_swappri)) {
1602: for (sdp = CIRCLEQ_FIRST(&spp->spi_swapdev);
1603: sdp != (void *)&spp->spi_swapdev;
1604: sdp = CIRCLEQ_NEXT(sdp,swd_next)) {
1605: /* if it's not enabled, then we can't swap from it */
1606: if ((sdp->swd_flags & SWF_ENABLE) == 0)
1607: continue;
1608: if (sdp->swd_npginuse + *nslots > sdp->swd_npages)
1609: continue;
1610: if (extent_alloc(sdp->swd_ex, *nslots, EX_NOALIGN, 0,
1611: EX_NOBOUNDARY, EX_MALLOCOK|EX_NOWAIT,
1612: &result) != 0) {
1613: continue;
1614: }
1615:
1616: /*
1617: * successful allocation! now rotate the circleq.
1618: */
1619: CIRCLEQ_REMOVE(&spp->spi_swapdev, sdp, swd_next);
1620: CIRCLEQ_INSERT_TAIL(&spp->spi_swapdev, sdp, swd_next);
1621: sdp->swd_npginuse += *nslots;
1622: uvmexp.swpginuse += *nslots;
1623: simple_unlock(&uvm.swap_data_lock);
1624: /* done! return drum slot number */
1625: UVMHIST_LOG(pdhist,
1626: "success! returning %ld slots starting at %ld",
1627: *nslots, result + sdp->swd_drumoffset, 0, 0);
1628: return(result + sdp->swd_drumoffset);
1629: }
1630: }
1631:
1632: /* XXXMRG: BEGIN HACK */
1633: if (*nslots > 1 && lessok) {
1634: *nslots = 1;
1635: goto ReTry; /* XXXMRG: ugh! extent should support this for us */
1636: }
1637: /* XXXMRG: END HACK */
1638:
1639: simple_unlock(&uvm.swap_data_lock);
1640: return 0; /* failed */
1641: }
1642:
1643: /*
1644: * uvm_swap_markbad: keep track of swap ranges where we've had i/o errors
1645: *
1646: * => we lock uvm.swap_data_lock
1647: */
1648: void
1649: uvm_swap_markbad(startslot, nslots)
1650: int startslot;
1651: int nslots;
1652: {
1653: struct swapdev *sdp;
1654: UVMHIST_FUNC("uvm_swap_markbad"); UVMHIST_CALLED(pdhist);
1655:
1656: simple_lock(&uvm.swap_data_lock);
1657: sdp = swapdrum_getsdp(startslot);
1658: if (sdp != NULL) {
1659: /*
1660: * we just keep track of how many pages have been marked bad
1661: * in this device, to make everything add up in swap_off().
1662: * we assume here that the range of slots will all be within
1663: * one swap device.
1664: */
1665: sdp->swd_npgbad += nslots;
1666: UVMHIST_LOG(pdhist, "now %ld bad", sdp->swd_npgbad, 0,0,0);
1667: }
1668: simple_unlock(&uvm.swap_data_lock);
1669: }
1670:
1671: /*
1672: * uvm_swap_free: free swap slots
1673: *
1674: * => this can be all or part of an allocation made by uvm_swap_alloc
1675: * => we lock uvm.swap_data_lock
1676: */
1677: void
1678: uvm_swap_free(startslot, nslots)
1679: int startslot;
1680: int nslots;
1681: {
1682: struct swapdev *sdp;
1683: UVMHIST_FUNC("uvm_swap_free"); UVMHIST_CALLED(pdhist);
1684:
1685: UVMHIST_LOG(pdhist, "freeing %ld slots starting at %ld", nslots,
1686: startslot, 0, 0);
1687:
1688: /*
1689: * ignore attempts to free the "bad" slot.
1690: */
1691:
1692: if (startslot == SWSLOT_BAD) {
1693: return;
1694: }
1695:
1696: /*
1697: * convert drum slot offset back to sdp, free the blocks
1698: * in the extent, and return. must hold pri lock to do
1699: * lookup and access the extent.
1700: */
1701:
1702: simple_lock(&uvm.swap_data_lock);
1703: sdp = swapdrum_getsdp(startslot);
1704: KASSERT(uvmexp.nswapdev >= 1);
1705: KASSERT(sdp != NULL);
1706: KASSERT(sdp->swd_npginuse >= nslots);
1707: if (extent_free(sdp->swd_ex, startslot - sdp->swd_drumoffset, nslots,
1708: EX_MALLOCOK|EX_NOWAIT) != 0) {
1709: printf("warning: resource shortage: %d pages of swap lost\n",
1710: nslots);
1711: }
1712:
1713: sdp->swd_npginuse -= nslots;
1714: uvmexp.swpginuse -= nslots;
1715: #ifdef UVM_SWAP_ENCRYPT
1716: {
1717: int i;
1718: if (swap_encrypt_initialized) {
1719: /* Dereference keys */
1720: for (i = 0; i < nslots; i++)
1721: if (uvm_swap_needdecrypt(sdp, startslot + i))
1722: SWAP_KEY_PUT(sdp, SWD_KEY(sdp, startslot + i));
1723:
1724: /* Mark range as not decrypt */
1725: uvm_swap_markdecrypt(sdp, startslot, nslots, 0);
1726: }
1727: }
1728: #endif /* UVM_SWAP_ENCRYPT */
1729: simple_unlock(&uvm.swap_data_lock);
1730: }
1731:
1732: /*
1733: * uvm_swap_put: put any number of pages into a contig place on swap
1734: *
1735: * => can be sync or async
1736: * => XXXMRG: consider making it an inline or macro
1737: */
1738: int
1739: uvm_swap_put(swslot, ppsp, npages, flags)
1740: int swslot;
1741: struct vm_page **ppsp;
1742: int npages;
1743: int flags;
1744: {
1745: int result;
1746:
1747: result = uvm_swap_io(ppsp, swslot, npages, B_WRITE |
1748: ((flags & PGO_SYNCIO) ? 0 : B_ASYNC));
1749:
1750: return (result);
1751: }
1752:
1753: /*
1754: * uvm_swap_get: get a single page from swap
1755: *
1756: * => usually a sync op (from fault)
1757: * => XXXMRG: consider making it an inline or macro
1758: */
1759: int
1760: uvm_swap_get(page, swslot, flags)
1761: struct vm_page *page;
1762: int swslot, flags;
1763: {
1764: int result;
1765:
1766: uvmexp.nswget++;
1767: KASSERT(flags & PGO_SYNCIO);
1768: if (swslot == SWSLOT_BAD) {
1769: return VM_PAGER_ERROR;
1770: }
1771:
1772: /*
1773: * this page is (about to be) no longer only in swap.
1774: */
1775: simple_lock(&uvm.swap_data_lock);
1776: uvmexp.swpgonly--;
1777: simple_unlock(&uvm.swap_data_lock);
1778:
1779: result = uvm_swap_io(&page, swslot, 1, B_READ |
1780: ((flags & PGO_SYNCIO) ? 0 : B_ASYNC));
1781:
1782: if (result != VM_PAGER_OK && result != VM_PAGER_PEND) {
1783: /*
1784: * oops, the read failed so it really is still only in swap.
1785: */
1786: simple_lock(&uvm.swap_data_lock);
1787: uvmexp.swpgonly++;
1788: simple_unlock(&uvm.swap_data_lock);
1789: }
1790:
1791: return (result);
1792: }
1793:
1794: /*
1795: * uvm_swap_io: do an i/o operation to swap
1796: */
1797:
1798: static int
1799: uvm_swap_io(pps, startslot, npages, flags)
1800: struct vm_page **pps;
1801: int startslot, npages, flags;
1802: {
1803: daddr64_t startblk;
1804: struct buf *bp;
1805: vaddr_t kva;
1806: int result, s, mapinflags, pflag;
1807: boolean_t write, async;
1808: #ifdef UVM_SWAP_ENCRYPT
1809: vaddr_t dstkva;
1810: struct vm_page *tpps[MAXBSIZE >> PAGE_SHIFT];
1811: struct swapdev *sdp;
1812: int encrypt = 0;
1813: #endif
1814: UVMHIST_FUNC("uvm_swap_io"); UVMHIST_CALLED(pdhist);
1815:
1816: UVMHIST_LOG(pdhist, "<- called, startslot=%ld, npages=%ld, flags=%ld",
1817: startslot, npages, flags, 0);
1818:
1819: write = (flags & B_READ) == 0;
1820: async = (flags & B_ASYNC) != 0;
1821:
1822: /*
1823: * convert starting drum slot to block number
1824: */
1825: startblk = btodb((u_int64_t)startslot << PAGE_SHIFT);
1826:
1827: /*
1828: * first, map the pages into the kernel (XXX: currently required
1829: * by buffer system).
1830: */
1831: mapinflags = !write ? UVMPAGER_MAPIN_READ : UVMPAGER_MAPIN_WRITE;
1832: if (!async)
1833: mapinflags |= UVMPAGER_MAPIN_WAITOK;
1834: kva = uvm_pagermapin(pps, npages, mapinflags);
1835: if (kva == 0)
1836: return (VM_PAGER_AGAIN);
1837:
1838: #ifdef UVM_SWAP_ENCRYPT
1839: if (write) {
1840: /*
1841: * Check if we need to do swap encryption on old pages.
1842: * Later we need a different scheme, that swap encrypts
1843: * all pages of a process that had at least one page swap
1844: * encrypted. Then we might not need to copy all pages
1845: * in the cluster, and avoid the memory overheard in
1846: * swapping.
1847: */
1848: if (uvm_doswapencrypt)
1849: encrypt = 1;
1850: }
1851:
1852: if (swap_encrypt_initialized || encrypt) {
1853: /*
1854: * we need to know the swap device that we are swapping to/from
1855: * to see if the pages need to be marked for decryption or
1856: * actually need to be decrypted.
1857: * XXX - does this information stay the same over the whole
1858: * execution of this function?
1859: */
1860: simple_lock(&uvm.swap_data_lock);
1861: sdp = swapdrum_getsdp(startslot);
1862: simple_unlock(&uvm.swap_data_lock);
1863: }
1864:
1865: /*
1866: * encrypt to swap
1867: */
1868: if (write && encrypt) {
1869: int i, opages;
1870: caddr_t src, dst;
1871: struct swap_key *key;
1872: u_int64_t block;
1873: int swmapflags;
1874:
1875: /* We always need write access. */
1876: swmapflags = UVMPAGER_MAPIN_READ;
1877: if (!async)
1878: swmapflags |= UVMPAGER_MAPIN_WAITOK;
1879:
1880: if (!uvm_swap_allocpages(tpps, npages)) {
1881: uvm_pagermapout(kva, npages);
1882: return (VM_PAGER_AGAIN);
1883: }
1884:
1885: dstkva = uvm_pagermapin(tpps, npages, swmapflags);
1886: if (dstkva == 0) {
1887: uvm_pagermapout(kva, npages);
1888: uvm_swap_freepages(tpps, npages);
1889: return (VM_PAGER_AGAIN);
1890: }
1891:
1892: src = (caddr_t) kva;
1893: dst = (caddr_t) dstkva;
1894: block = startblk;
1895: for (i = 0; i < npages; i++) {
1896: key = SWD_KEY(sdp, startslot + i);
1897: SWAP_KEY_GET(sdp, key); /* add reference */
1898:
1899: /* mark for async writes */
1900: atomic_setbits_int(&tpps[i]->pg_flags, PQ_ENCRYPT);
1901: swap_encrypt(key, src, dst, block, 1 << PAGE_SHIFT);
1902: src += 1 << PAGE_SHIFT;
1903: dst += 1 << PAGE_SHIFT;
1904: block += btodb(1 << PAGE_SHIFT);
1905: }
1906:
1907: uvm_pagermapout(kva, npages);
1908:
1909: /* dispose of pages we dont use anymore */
1910: opages = npages;
1911: uvm_pager_dropcluster(NULL, NULL, pps, &opages,
1912: PGO_PDFREECLUST);
1913:
1914: kva = dstkva;
1915: }
1916: #endif /* UVM_SWAP_ENCRYPT */
1917:
1918: /*
1919: * now allocate a buf for the i/o.
1920: * [make sure we don't put the pagedaemon to sleep...]
1921: */
1922: s = splbio();
1923: pflag = (async || curproc == uvm.pagedaemon_proc) ? 0 : PR_WAITOK;
1924: bp = pool_get(&bufpool, pflag);
1925: splx(s);
1926:
1927: /*
1928: * if we failed to get a swapbuf, return "try again"
1929: */
1930: if (bp == NULL) {
1931: #ifdef UVM_SWAP_ENCRYPT
1932: if (write && encrypt) {
1933: int i;
1934:
1935: /* swap encrypt needs cleanup */
1936: for (i = 0; i < npages; i++)
1937: SWAP_KEY_PUT(sdp, SWD_KEY(sdp, startslot + i));
1938:
1939: uvm_pagermapout(kva, npages);
1940: uvm_swap_freepages(tpps, npages);
1941: }
1942: #endif
1943: return (VM_PAGER_AGAIN);
1944: }
1945:
1946: #ifdef UVM_SWAP_ENCRYPT
1947: /*
1948: * prevent ASYNC reads.
1949: * uvm_swap_io is only called from uvm_swap_get, uvm_swap_get
1950: * assumes that all gets are SYNCIO. Just make sure here.
1951: * XXXARTUBC - might not be true anymore.
1952: */
1953: if (!write) {
1954: flags &= ~B_ASYNC;
1955: async = 0;
1956: }
1957: #endif
1958: /*
1959: * fill in the bp. we currently route our i/o through
1960: * /dev/drum's vnode [swapdev_vp].
1961: */
1962: bp->b_flags = B_BUSY | B_NOCACHE | (flags & (B_READ|B_ASYNC));
1963: bp->b_proc = &proc0; /* XXX */
1964: bp->b_vnbufs.le_next = NOLIST;
1965: bp->b_data = (caddr_t)kva;
1966: bp->b_blkno = startblk;
1967: LIST_INIT(&bp->b_dep);
1968: s = splbio();
1969: bp->b_vp = NULL;
1970: buf_replacevnode(bp, swapdev_vp);
1971: splx(s);
1972: bp->b_bufsize = bp->b_bcount = npages << PAGE_SHIFT;
1973:
1974: /*
1975: * for pageouts we must set "dirtyoff" [NFS client code needs it].
1976: * and we bump v_numoutput (counter of number of active outputs).
1977: */
1978: if (write) {
1979: bp->b_dirtyoff = 0;
1980: bp->b_dirtyend = npages << PAGE_SHIFT;
1981: #ifdef UVM_SWAP_ENCRYPT
1982: /* mark the pages in the drum for decryption */
1983: if (swap_encrypt_initialized)
1984: uvm_swap_markdecrypt(sdp, startslot, npages, encrypt);
1985: #endif
1986: s = splbio();
1987: swapdev_vp->v_numoutput++;
1988: splx(s);
1989: }
1990:
1991: /*
1992: * for async ops we must set up the iodone handler.
1993: */
1994: if (async) {
1995: bp->b_flags |= B_CALL | (curproc == uvm.pagedaemon_proc ?
1996: B_PDAEMON : 0);
1997: bp->b_iodone = uvm_aio_biodone;
1998: UVMHIST_LOG(pdhist, "doing async!", 0, 0, 0, 0);
1999: }
2000: UVMHIST_LOG(pdhist,
2001: "about to start io: data = %p blkno = 0x%lx, bcount = %ld",
2002: bp->b_data, bp->b_blkno, bp->b_bcount, 0);
2003:
2004: /*
2005: * now we start the I/O, and if async, return.
2006: */
2007: VOP_STRATEGY(bp);
2008: if (async)
2009: return (VM_PAGER_PEND);
2010:
2011: /*
2012: * must be sync i/o. wait for it to finish
2013: */
2014: (void) biowait(bp);
2015: result = (bp->b_flags & B_ERROR) ? VM_PAGER_ERROR : VM_PAGER_OK;
2016:
2017: #ifdef UVM_SWAP_ENCRYPT
2018: /*
2019: * decrypt swap
2020: */
2021: if (swap_encrypt_initialized &&
2022: (bp->b_flags & B_READ) && !(bp->b_flags & B_ERROR)) {
2023: int i;
2024: caddr_t data = bp->b_data;
2025: u_int64_t block = startblk;
2026: struct swap_key *key = NULL;
2027:
2028: for (i = 0; i < npages; i++) {
2029: /* Check if we need to decrypt */
2030: if (uvm_swap_needdecrypt(sdp, startslot + i)) {
2031: key = SWD_KEY(sdp, startslot + i);
2032: swap_decrypt(key, data, data, block,
2033: 1 << PAGE_SHIFT);
2034: }
2035: data += 1 << PAGE_SHIFT;
2036: block += btodb(1 << PAGE_SHIFT);
2037: }
2038: }
2039: #endif
2040: /*
2041: * kill the pager mapping
2042: */
2043: uvm_pagermapout(kva, npages);
2044:
2045: #ifdef UVM_SWAP_ENCRYPT
2046: /*
2047: * Not anymore needed, free after encryption
2048: */
2049: if ((bp->b_flags & B_READ) == 0 && encrypt)
2050: uvm_swap_freepages(tpps, npages);
2051: #endif
2052: /*
2053: * now dispose of the buf
2054: */
2055: s = splbio();
2056: if (bp->b_vp)
2057: brelvp(bp);
2058:
2059: if (write && bp->b_vp)
2060: vwakeup(bp->b_vp);
2061: pool_put(&bufpool, bp);
2062: splx(s);
2063:
2064: /*
2065: * finally return.
2066: */
2067: UVMHIST_LOG(pdhist, "<- done (sync) result=%ld", result, 0, 0, 0);
2068: return (result);
2069: }
2070:
2071: static void
2072: swapmount()
2073: {
2074: struct swapdev *sdp;
2075: struct swappri *spp;
2076: struct vnode *vp;
2077: dev_t swap_dev = swdevt[0].sw_dev;
2078:
2079: /*
2080: * No locking here since we happen to know that we will just be called
2081: * once before any other process has forked.
2082: */
2083:
2084: if (swap_dev == NODEV) {
2085: printf("swapmount: no device\n");
2086: return;
2087: }
2088:
2089: if (bdevvp(swap_dev, &vp)) {
2090: printf("swapmount: no device 2\n");
2091: return;
2092: }
2093:
2094: sdp = malloc(sizeof(*sdp), M_VMSWAP, M_WAITOK);
2095: spp = malloc(sizeof(*spp), M_VMSWAP, M_WAITOK);
2096: memset(sdp, 0, sizeof(*sdp));
2097:
2098: sdp->swd_flags = SWF_FAKE;
2099: sdp->swd_dev = swap_dev;
2100: sdp->swd_vp = vp;
2101: swaplist_insert(sdp, spp, 0);
2102: sdp->swd_pathlen = strlen("swap_device") + 1;
2103: sdp->swd_path = malloc(sdp->swd_pathlen, M_VMSWAP, M_WAITOK);
2104: if (copystr("swap_device", sdp->swd_path, sdp->swd_pathlen, 0))
2105: panic("swapmount: copystr");
2106:
2107: if (swap_on(curproc, sdp)) {
2108: swaplist_find(vp, 1);
2109: swaplist_trim();
2110: vput(sdp->swd_vp);
2111: free(sdp->swd_path, M_VMSWAP);
2112: free(sdp, M_VMSWAP);
2113: return;
2114: }
2115:
2116: VOP_UNLOCK(vp, 0, curproc);
2117: }
CVSweb