Home | History | Annotate | Line # | Download | only in uvm
uvm_vnode.c revision 1.101
      1  1.101   mlelstv /*	$NetBSD: uvm_vnode.c,v 1.101 2015/12/06 08:53:22 mlelstv Exp $	*/
      2    1.1       mrg 
      3    1.1       mrg /*
      4    1.1       mrg  * Copyright (c) 1997 Charles D. Cranor and Washington University.
      5    1.1       mrg  * Copyright (c) 1991, 1993
      6   1.49       chs  *      The Regents of the University of California.
      7    1.1       mrg  * Copyright (c) 1990 University of Utah.
      8    1.1       mrg  *
      9    1.1       mrg  * All rights reserved.
     10    1.1       mrg  *
     11    1.1       mrg  * This code is derived from software contributed to Berkeley by
     12    1.1       mrg  * the Systems Programming Group of the University of Utah Computer
     13    1.1       mrg  * Science Department.
     14    1.1       mrg  *
     15    1.1       mrg  * Redistribution and use in source and binary forms, with or without
     16    1.1       mrg  * modification, are permitted provided that the following conditions
     17    1.1       mrg  * are met:
     18    1.1       mrg  * 1. Redistributions of source code must retain the above copyright
     19    1.1       mrg  *    notice, this list of conditions and the following disclaimer.
     20    1.1       mrg  * 2. Redistributions in binary form must reproduce the above copyright
     21    1.1       mrg  *    notice, this list of conditions and the following disclaimer in the
     22    1.1       mrg  *    documentation and/or other materials provided with the distribution.
     23   1.94     chuck  * 3. Neither the name of the University nor the names of its contributors
     24    1.1       mrg  *    may be used to endorse or promote products derived from this software
     25    1.1       mrg  *    without specific prior written permission.
     26    1.1       mrg  *
     27    1.1       mrg  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     28    1.1       mrg  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     29    1.1       mrg  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     30    1.1       mrg  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     31    1.1       mrg  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     32    1.1       mrg  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     33    1.1       mrg  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     34    1.1       mrg  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     35    1.1       mrg  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     36    1.1       mrg  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     37    1.1       mrg  * SUCH DAMAGE.
     38    1.1       mrg  *
     39    1.1       mrg  *      @(#)vnode_pager.c       8.8 (Berkeley) 2/13/94
     40    1.3       mrg  * from: Id: uvm_vnode.c,v 1.1.2.26 1998/02/02 20:38:07 chuck Exp
     41    1.1       mrg  */
     42    1.1       mrg 
     43   1.55     lukem /*
     44   1.55     lukem  * uvm_vnode.c: the vnode pager.
     45   1.55     lukem  */
     46   1.55     lukem 
     47   1.55     lukem #include <sys/cdefs.h>
     48  1.101   mlelstv __KERNEL_RCSID(0, "$NetBSD: uvm_vnode.c,v 1.101 2015/12/06 08:53:22 mlelstv Exp $");
     49   1.55     lukem 
     50  1.100     pooka #ifdef _KERNEL_OPT
     51    1.4       mrg #include "opt_uvmhist.h"
     52  1.100     pooka #endif
     53    1.1       mrg 
     54    1.1       mrg #include <sys/param.h>
     55    1.1       mrg #include <sys/systm.h>
     56   1.37       chs #include <sys/kernel.h>
     57    1.1       mrg #include <sys/vnode.h>
     58   1.13   thorpej #include <sys/disklabel.h>
     59   1.13   thorpej #include <sys/ioctl.h>
     60   1.13   thorpej #include <sys/fcntl.h>
     61   1.13   thorpej #include <sys/conf.h>
     62   1.37       chs #include <sys/pool.h>
     63   1.37       chs #include <sys/mount.h>
     64   1.13   thorpej 
     65   1.13   thorpej #include <miscfs/specfs/specdev.h>
     66    1.1       mrg 
     67    1.1       mrg #include <uvm/uvm.h>
     68   1.68      yamt #include <uvm/uvm_readahead.h>
     69    1.1       mrg 
     70   1.99      matt #ifdef UVMHIST
     71   1.99      matt UVMHIST_DEFINE(ubchist);
     72   1.99      matt #endif
     73   1.99      matt 
     74    1.1       mrg /*
     75    1.1       mrg  * functions
     76    1.1       mrg  */
     77    1.1       mrg 
     78   1.66   thorpej static void	uvn_detach(struct uvm_object *);
     79   1.66   thorpej static int	uvn_get(struct uvm_object *, voff_t, struct vm_page **, int *,
     80   1.66   thorpej 			int, vm_prot_t, int, int);
     81   1.66   thorpej static int	uvn_put(struct uvm_object *, voff_t, voff_t, int);
     82   1.66   thorpej static void	uvn_reference(struct uvm_object *);
     83   1.52       chs 
     84   1.66   thorpej static int	uvn_findpage(struct uvm_object *, voff_t, struct vm_page **,
     85   1.66   thorpej 			     int);
     86    1.1       mrg 
     87    1.1       mrg /*
     88    1.1       mrg  * master pager structure
     89    1.1       mrg  */
     90    1.1       mrg 
     91   1.89      yamt const struct uvm_pagerops uvm_vnodeops = {
     92   1.88      yamt 	.pgo_reference = uvn_reference,
     93   1.88      yamt 	.pgo_detach = uvn_detach,
     94   1.88      yamt 	.pgo_get = uvn_get,
     95   1.88      yamt 	.pgo_put = uvn_put,
     96    1.1       mrg };
     97    1.1       mrg 
     98    1.1       mrg /*
     99    1.1       mrg  * the ops!
    100    1.1       mrg  */
    101    1.1       mrg 
    102    1.1       mrg /*
    103    1.1       mrg  * uvn_reference
    104    1.1       mrg  *
    105    1.1       mrg  * duplicate a reference to a VM object.  Note that the reference
    106   1.49       chs  * count must already be at least one (the passed in reference) so
    107    1.1       mrg  * there is no chance of the uvn being killed or locked out here.
    108    1.1       mrg  *
    109   1.49       chs  * => caller must call with object unlocked.
    110    1.1       mrg  * => caller must be using the same accessprot as was used at attach time
    111    1.1       mrg  */
    112    1.1       mrg 
    113   1.66   thorpej static void
    114   1.65   thorpej uvn_reference(struct uvm_object *uobj)
    115    1.1       mrg {
    116   1.93     pooka 	vref((struct vnode *)uobj);
    117    1.1       mrg }
    118    1.1       mrg 
    119   1.52       chs 
    120    1.1       mrg /*
    121    1.1       mrg  * uvn_detach
    122    1.1       mrg  *
    123    1.1       mrg  * remove a reference to a VM object.
    124    1.1       mrg  *
    125    1.1       mrg  * => caller must call with object unlocked and map locked.
    126    1.1       mrg  */
    127   1.52       chs 
    128   1.66   thorpej static void
    129   1.65   thorpej uvn_detach(struct uvm_object *uobj)
    130    1.8       mrg {
    131   1.37       chs 	vrele((struct vnode *)uobj);
    132    1.1       mrg }
    133    1.1       mrg 
    134    1.1       mrg /*
    135    1.1       mrg  * uvn_put: flush page data to backing store.
    136    1.1       mrg  *
    137   1.53  sommerfe  * => object must be locked on entry!   VOP_PUTPAGES must unlock it.
    138    1.1       mrg  * => flags: PGO_SYNCIO -- use sync. I/O
    139    1.1       mrg  * => note: caller must set PG_CLEAN and pmap_clear_modify (if needed)
    140    1.1       mrg  */
    141    1.1       mrg 
    142   1.66   thorpej static int
    143   1.65   thorpej uvn_put(struct uvm_object *uobj, voff_t offlo, voff_t offhi, int flags)
    144    1.1       mrg {
    145   1.37       chs 	struct vnode *vp = (struct vnode *)uobj;
    146   1.37       chs 	int error;
    147    1.1       mrg 
    148   1.96     rmind 	KASSERT(mutex_owned(vp->v_interlock));
    149   1.54       chs 	error = VOP_PUTPAGES(vp, offlo, offhi, flags);
    150   1.90        ad 
    151   1.48       chs 	return error;
    152    1.1       mrg }
    153    1.1       mrg 
    154    1.1       mrg 
    155    1.1       mrg /*
    156    1.1       mrg  * uvn_get: get pages (synchronously) from backing store
    157    1.1       mrg  *
    158    1.1       mrg  * => prefer map unlocked (not required)
    159    1.1       mrg  * => object must be locked!  we will _unlock_ it before starting any I/O.
    160    1.1       mrg  * => flags: PGO_ALLPAGES: get all of the pages
    161    1.1       mrg  *           PGO_LOCKED: fault data structures are locked
    162    1.1       mrg  * => NOTE: offset is the offset of pps[0], _NOT_ pps[centeridx]
    163    1.1       mrg  * => NOTE: caller must check for released pages!!
    164    1.1       mrg  */
    165   1.49       chs 
    166   1.66   thorpej static int
    167   1.65   thorpej uvn_get(struct uvm_object *uobj, voff_t offset,
    168   1.65   thorpej     struct vm_page **pps /* IN/OUT */,
    169   1.65   thorpej     int *npagesp /* IN (OUT if PGO_LOCKED)*/,
    170   1.65   thorpej     int centeridx, vm_prot_t access_type, int advice, int flags)
    171    1.8       mrg {
    172   1.37       chs 	struct vnode *vp = (struct vnode *)uobj;
    173   1.37       chs 	int error;
    174   1.67      yamt 
    175   1.37       chs 	UVMHIST_FUNC("uvn_get"); UVMHIST_CALLED(ubchist);
    176   1.37       chs 
    177   1.37       chs 	UVMHIST_LOG(ubchist, "vp %p off 0x%x", vp, (int)offset, 0,0);
    178   1.68      yamt 
    179   1.98    martin 	if (vp->v_type == VREG && (access_type & VM_PROT_WRITE) == 0
    180   1.98    martin 	    && (flags & PGO_LOCKED) == 0) {
    181   1.68      yamt 		vn_ra_allocctx(vp);
    182   1.68      yamt 		uvm_ra_request(vp->v_ractx, advice, uobj, offset,
    183   1.68      yamt 		    *npagesp << PAGE_SHIFT);
    184   1.68      yamt 	}
    185   1.68      yamt 
    186   1.37       chs 	error = VOP_GETPAGES(vp, offset, pps, npagesp, centeridx,
    187   1.37       chs 			     access_type, advice, flags);
    188   1.67      yamt 
    189   1.96     rmind 	KASSERT(((flags & PGO_LOCKED) != 0 && mutex_owned(vp->v_interlock)) ||
    190   1.90        ad 	    (flags & PGO_LOCKED) == 0);
    191   1.48       chs 	return error;
    192   1.37       chs }
    193    1.8       mrg 
    194    1.8       mrg 
    195   1.37       chs /*
    196   1.37       chs  * uvn_findpages:
    197   1.37       chs  * return the page for the uobj and offset requested, allocating if needed.
    198   1.37       chs  * => uobj must be locked.
    199   1.52       chs  * => returned pages will be BUSY.
    200   1.37       chs  */
    201    1.1       mrg 
    202   1.58     enami int
    203   1.65   thorpej uvn_findpages(struct uvm_object *uobj, voff_t offset, int *npagesp,
    204   1.65   thorpej     struct vm_page **pgs, int flags)
    205   1.37       chs {
    206   1.58     enami 	int i, count, found, npages, rv;
    207    1.8       mrg 
    208   1.58     enami 	count = found = 0;
    209   1.37       chs 	npages = *npagesp;
    210   1.52       chs 	if (flags & UFP_BACKWARD) {
    211   1.52       chs 		for (i = npages - 1; i >= 0; i--, offset -= PAGE_SIZE) {
    212   1.52       chs 			rv = uvn_findpage(uobj, offset, &pgs[i], flags);
    213   1.58     enami 			if (rv == 0) {
    214   1.58     enami 				if (flags & UFP_DIRTYONLY)
    215   1.58     enami 					break;
    216   1.58     enami 			} else
    217   1.58     enami 				found++;
    218   1.52       chs 			count++;
    219   1.52       chs 		}
    220   1.52       chs 	} else {
    221   1.52       chs 		for (i = 0; i < npages; i++, offset += PAGE_SIZE) {
    222   1.52       chs 			rv = uvn_findpage(uobj, offset, &pgs[i], flags);
    223   1.58     enami 			if (rv == 0) {
    224   1.58     enami 				if (flags & UFP_DIRTYONLY)
    225   1.58     enami 					break;
    226   1.58     enami 			} else
    227   1.58     enami 				found++;
    228   1.52       chs 			count++;
    229   1.52       chs 		}
    230   1.37       chs 	}
    231   1.52       chs 	*npagesp = count;
    232   1.58     enami 	return (found);
    233   1.37       chs }
    234    1.8       mrg 
    235   1.66   thorpej static int
    236   1.65   thorpej uvn_findpage(struct uvm_object *uobj, voff_t offset, struct vm_page **pgp,
    237   1.65   thorpej     int flags)
    238   1.37       chs {
    239   1.37       chs 	struct vm_page *pg;
    240   1.79   thorpej 	bool dirty;
    241   1.37       chs 	UVMHIST_FUNC("uvn_findpage"); UVMHIST_CALLED(ubchist);
    242   1.37       chs 	UVMHIST_LOG(ubchist, "vp %p off 0x%lx", uobj, offset,0,0);
    243    1.8       mrg 
    244   1.96     rmind 	KASSERT(mutex_owned(uobj->vmobjlock));
    245   1.96     rmind 
    246   1.37       chs 	if (*pgp != NULL) {
    247   1.37       chs 		UVMHIST_LOG(ubchist, "dontcare", 0,0,0,0);
    248   1.37       chs 		return 0;
    249   1.37       chs 	}
    250   1.37       chs 	for (;;) {
    251   1.37       chs 		/* look for an existing page */
    252   1.37       chs 		pg = uvm_pagelookup(uobj, offset);
    253   1.37       chs 
    254   1.52       chs 		/* nope?  allocate one now */
    255   1.37       chs 		if (pg == NULL) {
    256   1.37       chs 			if (flags & UFP_NOALLOC) {
    257   1.37       chs 				UVMHIST_LOG(ubchist, "noalloc", 0,0,0,0);
    258   1.37       chs 				return 0;
    259   1.37       chs 			}
    260   1.97      matt 			pg = uvm_pagealloc(uobj, offset, NULL,
    261   1.97      matt 			    UVM_FLAG_COLORMATCH);
    262   1.37       chs 			if (pg == NULL) {
    263   1.37       chs 				if (flags & UFP_NOWAIT) {
    264   1.37       chs 					UVMHIST_LOG(ubchist, "nowait",0,0,0,0);
    265   1.37       chs 					return 0;
    266    1.8       mrg 				}
    267   1.96     rmind 				mutex_exit(uobj->vmobjlock);
    268   1.37       chs 				uvm_wait("uvn_fp1");
    269   1.96     rmind 				mutex_enter(uobj->vmobjlock);
    270   1.37       chs 				continue;
    271   1.47       chs 			}
    272   1.97      matt 			UVMHIST_LOG(ubchist, "alloced %p (color %u)", pg,
    273   1.97      matt 			    VM_PGCOLOR_BUCKET(pg), 0,0);
    274   1.37       chs 			break;
    275   1.37       chs 		} else if (flags & UFP_NOCACHE) {
    276   1.37       chs 			UVMHIST_LOG(ubchist, "nocache",0,0,0,0);
    277   1.37       chs 			return 0;
    278    1.8       mrg 		}
    279    1.8       mrg 
    280   1.37       chs 		/* page is there, see if we need to wait on it */
    281   1.52       chs 		if ((pg->flags & PG_BUSY) != 0) {
    282   1.37       chs 			if (flags & UFP_NOWAIT) {
    283   1.37       chs 				UVMHIST_LOG(ubchist, "nowait",0,0,0,0);
    284   1.37       chs 				return 0;
    285   1.37       chs 			}
    286   1.37       chs 			pg->flags |= PG_WANTED;
    287   1.97      matt 			UVMHIST_LOG(ubchist, "wait %p (color %u)", pg,
    288   1.97      matt 			    VM_PGCOLOR_BUCKET(pg), 0,0);
    289   1.96     rmind 			UVM_UNLOCK_AND_WAIT(pg, uobj->vmobjlock, 0,
    290   1.37       chs 					    "uvn_fp2", 0);
    291   1.96     rmind 			mutex_enter(uobj->vmobjlock);
    292   1.37       chs 			continue;
    293    1.8       mrg 		}
    294   1.49       chs 
    295   1.37       chs 		/* skip PG_RDONLY pages if requested */
    296   1.37       chs 		if ((flags & UFP_NORDONLY) && (pg->flags & PG_RDONLY)) {
    297   1.37       chs 			UVMHIST_LOG(ubchist, "nordonly",0,0,0,0);
    298   1.37       chs 			return 0;
    299    1.8       mrg 		}
    300    1.8       mrg 
    301   1.52       chs 		/* stop on clean pages if requested */
    302   1.52       chs 		if (flags & UFP_DIRTYONLY) {
    303   1.52       chs 			dirty = pmap_clear_modify(pg) ||
    304   1.52       chs 				(pg->flags & PG_CLEAN) == 0;
    305   1.52       chs 			pg->flags |= PG_CLEAN;
    306   1.52       chs 			if (!dirty) {
    307   1.58     enami 				UVMHIST_LOG(ubchist, "dirtonly", 0,0,0,0);
    308   1.52       chs 				return 0;
    309   1.52       chs 			}
    310   1.52       chs 		}
    311   1.52       chs 
    312   1.37       chs 		/* mark the page BUSY and we're done. */
    313   1.37       chs 		pg->flags |= PG_BUSY;
    314   1.37       chs 		UVM_PAGE_OWN(pg, "uvn_findpage");
    315   1.97      matt 		UVMHIST_LOG(ubchist, "found %p (color %u)",
    316   1.97      matt 		    pg, VM_PGCOLOR_BUCKET(pg), 0,0);
    317   1.37       chs 		break;
    318    1.8       mrg 	}
    319   1.37       chs 	*pgp = pg;
    320   1.37       chs 	return 1;
    321    1.1       mrg }
    322    1.1       mrg 
    323    1.1       mrg /*
    324   1.52       chs  * uvm_vnp_setsize: grow or shrink a vnode uobj
    325    1.1       mrg  *
    326    1.1       mrg  * grow   => just update size value
    327    1.1       mrg  * shrink => toss un-needed pages
    328    1.1       mrg  *
    329   1.49       chs  * => we assume that the caller has a reference of some sort to the
    330    1.1       mrg  *	vnode in question so that it will not be yanked out from under
    331    1.1       mrg  *	us.
    332    1.1       mrg  */
    333    1.1       mrg 
    334    1.8       mrg void
    335   1.65   thorpej uvm_vnp_setsize(struct vnode *vp, voff_t newsize)
    336    1.8       mrg {
    337   1.52       chs 	struct uvm_object *uobj = &vp->v_uobj;
    338   1.46     enami 	voff_t pgend = round_page(newsize);
    339   1.72      yamt 	voff_t oldsize;
    340   1.37       chs 	UVMHIST_FUNC("uvm_vnp_setsize"); UVMHIST_CALLED(ubchist);
    341   1.37       chs 
    342   1.96     rmind 	mutex_enter(uobj->vmobjlock);
    343   1.52       chs 	UVMHIST_LOG(ubchist, "vp %p old 0x%x new 0x%x",
    344   1.52       chs 	    vp, vp->v_size, newsize, 0);
    345    1.1       mrg 
    346    1.8       mrg 	/*
    347   1.37       chs 	 * now check if the size has changed: if we shrink we had better
    348   1.37       chs 	 * toss some pages...
    349    1.8       mrg 	 */
    350    1.1       mrg 
    351  1.101   mlelstv 	KASSERT(newsize != VSIZENOTSET && newsize >= 0);
    352   1.85     pooka 	KASSERT(vp->v_size <= vp->v_writesize);
    353   1.85     pooka 	KASSERT(vp->v_size == vp->v_writesize ||
    354   1.85     pooka 	    newsize == vp->v_writesize || newsize <= vp->v_size);
    355   1.85     pooka 
    356   1.85     pooka 	oldsize = vp->v_writesize;
    357   1.85     pooka 
    358  1.101   mlelstv 	/*
    359  1.101   mlelstv 	 * check wether size shrinks
    360  1.101   mlelstv 	 * if old size hasn't been set, there are no pages to drop
    361  1.101   mlelstv 	 * if there was an integer overflow in pgend, then this is no shrink
    362  1.101   mlelstv 	 */
    363  1.101   mlelstv 	if (oldsize > pgend && oldsize != VSIZENOTSET && pgend >= 0) {
    364   1.57       chs 		(void) uvn_put(uobj, pgend, 0, PGO_FREE | PGO_SYNCIO);
    365   1.96     rmind 		mutex_enter(uobj->vmobjlock);
    366    1.8       mrg 	}
    367   1.82      yamt 	vp->v_size = vp->v_writesize = newsize;
    368   1.96     rmind 	mutex_exit(uobj->vmobjlock);
    369    1.1       mrg }
    370    1.1       mrg 
    371   1.82      yamt void
    372   1.82      yamt uvm_vnp_setwritesize(struct vnode *vp, voff_t newsize)
    373   1.82      yamt {
    374   1.82      yamt 
    375   1.96     rmind 	mutex_enter(vp->v_interlock);
    376  1.101   mlelstv 	KASSERT(newsize != VSIZENOTSET && newsize >= 0);
    377   1.82      yamt 	KASSERT(vp->v_size != VSIZENOTSET);
    378   1.82      yamt 	KASSERT(vp->v_writesize != VSIZENOTSET);
    379   1.82      yamt 	KASSERT(vp->v_size <= vp->v_writesize);
    380   1.82      yamt 	KASSERT(vp->v_size <= newsize);
    381   1.82      yamt 	vp->v_writesize = newsize;
    382   1.96     rmind 	mutex_exit(vp->v_interlock);
    383   1.82      yamt }
    384   1.82      yamt 
    385   1.79   thorpej bool
    386   1.75      yamt uvn_text_p(struct uvm_object *uobj)
    387   1.75      yamt {
    388   1.75      yamt 	struct vnode *vp = (struct vnode *)uobj;
    389   1.75      yamt 
    390   1.86        ad 	return (vp->v_iflag & VI_EXECMAP) != 0;
    391   1.75      yamt }
    392   1.75      yamt 
    393   1.79   thorpej bool
    394   1.75      yamt uvn_clean_p(struct uvm_object *uobj)
    395   1.75      yamt {
    396   1.75      yamt 	struct vnode *vp = (struct vnode *)uobj;
    397   1.75      yamt 
    398   1.86        ad 	return (vp->v_iflag & VI_ONWORKLST) == 0;
    399   1.75      yamt }
    400   1.75      yamt 
    401   1.79   thorpej bool
    402   1.75      yamt uvn_needs_writefault_p(struct uvm_object *uobj)
    403   1.75      yamt {
    404   1.75      yamt 	struct vnode *vp = (struct vnode *)uobj;
    405   1.75      yamt 
    406   1.75      yamt 	return uvn_clean_p(uobj) ||
    407   1.86        ad 	    (vp->v_iflag & (VI_WRMAP|VI_WRMAPDIRTY)) == VI_WRMAP;
    408   1.75      yamt }
    409