Date: Mon, 21 Oct 2019 00:08:35 +0000 (UTC) From: Kyle Evans <kevans@FreeBSD.org> To: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-stable@freebsd.org, svn-src-stable-12@freebsd.org Subject: svn commit: r353787 - in stable/12/sys: kern vm Message-ID: <201910210008.x9L08ZQt027865@repo.freebsd.org>
next in thread | raw e-mail | index | archive | help
Author: kevans Date: Mon Oct 21 00:08:34 2019 New Revision: 353787 URL: https://svnweb.freebsd.org/changeset/base/353787 Log: MFC r351650, r351795-r351796: writemapping accounting for posixshm r351650: posixshm: switch to OBJT_SWAP in advance of other changes Future changes to posixshm will start tracking writeable mappings in order to support file sealing. Tracking writeable mappings for an OBJT_DEFAULT object is complicated as it may be swapped out and converted to an OBJT_SWAP. One may generically add this tracking for vm_object, but this is difficult to do without increasing memory footprint of vm_object and blowing up memory usage by a significant amount. On the other hand, the swap pager can be expanded to track writeable mappings without increasing vm_object size. This change is currently in D21456. Switch over to OBJT_SWAP in advance of the other changes to the swap pager and posixshm. r351795: vm pager: writemapping accounting for OBJT_SWAP Currently writemapping accounting is only done for vnode_pager which does some accounting on the underlying vnode. Extend this to allow accounting to be possible for any of the pager types. New pageops are added to update/release writecount that need to be implemented for any pager wishing to do said accounting, and we implement these methods now for both vnode_pager (unchanged) and swap_pager. The primary motivation for this is to allow other systems with OBJT_SWAP objects to check if their objects have any write mappings and reject operations with EBUSY if so. posixshm will be the first to do so in order to reject adding write seals to the shmfd if any writable mappings exist. r351796: posixshm: start counting writeable mappings r351650 switched posixshm to using OBJT_SWAP for shm_object r351795 added support to the swap_pager for tracking writeable mappings Take advantage of this and start tracking writeable mappings; fd sealing will use this to reject a seal on writing with EBUSY if any such mapping exist. Modified: stable/12/sys/kern/uipc_shm.c stable/12/sys/kern/vfs_vnops.c stable/12/sys/vm/swap_pager.c stable/12/sys/vm/vm_map.c stable/12/sys/vm/vm_map.h stable/12/sys/vm/vm_mmap.c stable/12/sys/vm/vm_object.h stable/12/sys/vm/vm_pager.h stable/12/sys/vm/vnode_pager.c stable/12/sys/vm/vnode_pager.h Directory Properties: stable/12/ (props changed) Modified: stable/12/sys/kern/uipc_shm.c ============================================================================== --- stable/12/sys/kern/uipc_shm.c Sun Oct 20 22:55:47 2019 (r353786) +++ stable/12/sys/kern/uipc_shm.c Mon Oct 21 00:08:34 2019 (r353787) @@ -554,7 +554,7 @@ shm_alloc(struct ucred *ucred, mode_t mode) shmfd->shm_uid = ucred->cr_uid; shmfd->shm_gid = ucred->cr_gid; shmfd->shm_mode = mode; - shmfd->shm_object = vm_pager_allocate(OBJT_DEFAULT, NULL, + shmfd->shm_object = vm_pager_allocate(OBJT_SWAP, NULL, shmfd->shm_size, VM_PROT_DEFAULT, 0, ucred); KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate")); shmfd->shm_object->pg_color = 0; @@ -900,6 +900,7 @@ shm_mmap(struct file *fp, vm_map_t map, vm_offset_t *a struct shmfd *shmfd; vm_prot_t maxprot; int error; + bool writecnt; shmfd = fp->f_data; maxprot = VM_PROT_NONE; @@ -910,10 +911,10 @@ shm_mmap(struct file *fp, vm_map_t map, vm_offset_t *a if ((fp->f_flag & FWRITE) != 0) maxprot |= VM_PROT_WRITE; + writecnt = (flags & MAP_SHARED) != 0 && (prot & VM_PROT_WRITE) != 0; + /* Don't permit shared writable mappings on read-only descriptors. */ - if ((flags & MAP_SHARED) != 0 && - (maxprot & VM_PROT_WRITE) == 0 && - (prot & VM_PROT_WRITE) != 0) + if (writecnt && (maxprot & VM_PROT_WRITE) == 0) return (EACCES); maxprot &= cap_maxprot; @@ -936,10 +937,16 @@ shm_mmap(struct file *fp, vm_map_t map, vm_offset_t *a mtx_unlock(&shm_timestamp_lock); vm_object_reference(shmfd->shm_object); + if (writecnt) + vm_pager_update_writecount(shmfd->shm_object, 0, objsize); error = vm_mmap_object(map, addr, objsize, prot, maxprot, flags, - shmfd->shm_object, foff, FALSE, td); - if (error != 0) + shmfd->shm_object, foff, writecnt, td); + if (error != 0) { + if (writecnt) + vm_pager_release_writecount(shmfd->shm_object, 0, + objsize); vm_object_deallocate(shmfd->shm_object); + } return (error); } Modified: stable/12/sys/kern/vfs_vnops.c ============================================================================== --- stable/12/sys/kern/vfs_vnops.c Sun Oct 20 22:55:47 2019 (r353786) +++ stable/12/sys/kern/vfs_vnops.c Mon Oct 21 00:08:34 2019 (r353787) @@ -86,7 +86,7 @@ __FBSDID("$FreeBSD$"); #include <vm/vm_map.h> #include <vm/vm_object.h> #include <vm/vm_page.h> -#include <vm/vnode_pager.h> +#include <vm/vm_pager.h> #ifdef HWPMC_HOOKS #include <sys/pmckern.h> @@ -2483,7 +2483,7 @@ vn_mmap(struct file *fp, vm_map_t map, vm_offset_t *ad * writecount, then undo that now. */ if (writecounted) - vnode_pager_release_writecount(object, 0, size); + vm_pager_release_writecount(object, 0, size); vm_object_deallocate(object); } #ifdef HWPMC_HOOKS Modified: stable/12/sys/vm/swap_pager.c ============================================================================== --- stable/12/sys/vm/swap_pager.c Sun Oct 20 22:55:47 2019 (r353786) +++ stable/12/sys/vm/swap_pager.c Mon Oct 21 00:08:34 2019 (r353787) @@ -375,6 +375,10 @@ static boolean_t static void swap_pager_init(void); static void swap_pager_unswapped(vm_page_t); static void swap_pager_swapoff(struct swdevt *sp); +static void swap_pager_update_writecount(vm_object_t object, + vm_offset_t start, vm_offset_t end); +static void swap_pager_release_writecount(vm_object_t object, + vm_offset_t start, vm_offset_t end); struct pagerops swappagerops = { .pgo_init = swap_pager_init, /* early system initialization of pager */ @@ -385,6 +389,8 @@ struct pagerops swappagerops = { .pgo_putpages = swap_pager_putpages, /* pageout */ .pgo_haspage = swap_pager_haspage, /* get backing store status for page */ .pgo_pageunswapped = swap_pager_unswapped, /* remove swap related to page */ + .pgo_update_writecount = swap_pager_update_writecount, + .pgo_release_writecount = swap_pager_release_writecount, }; /* @@ -610,6 +616,7 @@ swap_pager_alloc_init(void *handle, struct ucred *cred object = vm_object_allocate(OBJT_SWAP, OFF_TO_IDX(offset + PAGE_MASK + size)); + object->un_pager.swp.writemappings = 0; object->handle = handle; if (cred != NULL) { object->cred = cred; @@ -1920,6 +1927,7 @@ swp_pager_meta_build(vm_object_t object, vm_pindex_t p atomic_thread_fence_rel(); object->type = OBJT_SWAP; + object->un_pager.swp.writemappings = 0; KASSERT(object->handle == NULL, ("default pager with handle")); } @@ -3004,4 +3012,28 @@ sysctl_swap_async_max(SYSCTL_HANDLER_ARGS) mtx_unlock(&pbuf_mtx); return (0); +} + +static void +swap_pager_update_writecount(vm_object_t object, vm_offset_t start, + vm_offset_t end) +{ + + VM_OBJECT_WLOCK(object); + KASSERT((object->flags & OBJ_NOSPLIT) != 0, + ("Splittable object with writecount")); + object->un_pager.swp.writemappings += (vm_ooffset_t)end - start; + VM_OBJECT_WUNLOCK(object); +} + +static void +swap_pager_release_writecount(vm_object_t object, vm_offset_t start, + vm_offset_t end) +{ + + VM_OBJECT_WLOCK(object); + KASSERT((object->flags & OBJ_NOSPLIT) != 0, + ("Splittable object with writecount")); + object->un_pager.swp.writemappings -= (vm_ooffset_t)end - start; + VM_OBJECT_WUNLOCK(object); } Modified: stable/12/sys/vm/vm_map.c ============================================================================== --- stable/12/sys/vm/vm_map.c Sun Oct 20 22:55:47 2019 (r353786) +++ stable/12/sys/vm/vm_map.c Mon Oct 21 00:08:34 2019 (r353787) @@ -574,10 +574,10 @@ vm_map_process_deferred(void) td->td_map_def_user = NULL; while (entry != NULL) { next = entry->next; - MPASS((entry->eflags & (MAP_ENTRY_VN_WRITECNT | - MAP_ENTRY_VN_EXEC)) != (MAP_ENTRY_VN_WRITECNT | + MPASS((entry->eflags & (MAP_ENTRY_WRITECNT | + MAP_ENTRY_VN_EXEC)) != (MAP_ENTRY_WRITECNT | MAP_ENTRY_VN_EXEC)); - if ((entry->eflags & MAP_ENTRY_VN_WRITECNT) != 0) { + if ((entry->eflags & MAP_ENTRY_WRITECNT) != 0) { /* * Decrement the object's writemappings and * possibly the vnode's v_writecount. @@ -586,7 +586,7 @@ vm_map_process_deferred(void) ("Submap with writecount")); object = entry->object.vm_object; KASSERT(object != NULL, ("No object for writecount")); - vnode_pager_release_writecount(object, entry->start, + vm_pager_release_writecount(object, entry->start, entry->end); } vm_map_entry_set_vnode_text(entry, false); @@ -1484,8 +1484,8 @@ vm_map_insert(vm_map_t map, vm_object_t object, vm_oof protoeflags |= MAP_ENTRY_GROWS_DOWN; if (cow & MAP_STACK_GROWS_UP) protoeflags |= MAP_ENTRY_GROWS_UP; - if (cow & MAP_VN_WRITECOUNT) - protoeflags |= MAP_ENTRY_VN_WRITECNT; + if (cow & MAP_WRITECOUNT) + protoeflags |= MAP_ENTRY_WRITECNT; if (cow & MAP_VN_EXEC) protoeflags |= MAP_ENTRY_VN_EXEC; if ((cow & MAP_CREATE_GUARD) != 0) @@ -2196,7 +2196,7 @@ _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_map_entry_set_vnode_text(new_entry, true); /* * The object->un_pager.vnp.writemappings for the - * object of MAP_ENTRY_VN_WRITECNT type entry shall be + * object of MAP_ENTRY_WRITECNT type entry shall be * kept as is here. The virtual pages are * re-distributed among the clipped entries, so the sum is * left the same. @@ -3758,20 +3758,20 @@ vm_map_copy_entry( dst_entry->eflags |= MAP_ENTRY_COW | MAP_ENTRY_NEEDS_COPY; dst_entry->offset = src_entry->offset; - if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) { + if (src_entry->eflags & MAP_ENTRY_WRITECNT) { /* - * MAP_ENTRY_VN_WRITECNT cannot + * MAP_ENTRY_WRITECNT cannot * indicate write reference from * src_entry, since the entry is * marked as needs copy. Allocate a * fake entry that is used to - * decrement object->un_pager.vnp.writecount + * decrement object->un_pager writecount * at the appropriate time. Attach * fake_entry to the deferred list. */ fake_entry = vm_map_entry_create(dst_map); - fake_entry->eflags = MAP_ENTRY_VN_WRITECNT; - src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT; + fake_entry->eflags = MAP_ENTRY_WRITECNT; + src_entry->eflags &= ~MAP_ENTRY_WRITECNT; vm_object_reference(src_object); fake_entry->object.vm_object = src_object; fake_entry->start = src_entry->start; @@ -3954,7 +3954,7 @@ vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_c * not relock it later for the assertion * correctness. */ - if (old_entry->eflags & MAP_ENTRY_VN_WRITECNT && + if (old_entry->eflags & MAP_ENTRY_WRITECNT && object->type == OBJT_VNODE) { KASSERT(((struct vnode *)object->handle)-> v_writecount > 0, @@ -3974,8 +3974,8 @@ vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_c MAP_ENTRY_IN_TRANSITION); new_entry->wiring_thread = NULL; new_entry->wired_count = 0; - if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) { - vnode_pager_update_writecount(object, + if (new_entry->eflags & MAP_ENTRY_WRITECNT) { + vm_pager_update_writecount(object, new_entry->start, new_entry->end); } vm_map_entry_set_vnode_text(new_entry, true); @@ -4006,7 +4006,7 @@ vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_c * Copied entry is COW over the old object. */ new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED | - MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT); + MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_WRITECNT); new_entry->wiring_thread = NULL; new_entry->wired_count = 0; new_entry->object.vm_object = NULL; @@ -4030,7 +4030,7 @@ vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_c new_entry->end = old_entry->end; new_entry->eflags = old_entry->eflags & ~(MAP_ENTRY_USER_WIRED | MAP_ENTRY_IN_TRANSITION | - MAP_ENTRY_VN_WRITECNT | MAP_ENTRY_VN_EXEC); + MAP_ENTRY_WRITECNT | MAP_ENTRY_VN_EXEC); new_entry->protection = old_entry->protection; new_entry->max_protection = old_entry->max_protection; new_entry->inheritance = VM_INHERIT_ZERO; Modified: stable/12/sys/vm/vm_map.h ============================================================================== --- stable/12/sys/vm/vm_map.h Sun Oct 20 22:55:47 2019 (r353786) +++ stable/12/sys/vm/vm_map.h Mon Oct 21 00:08:34 2019 (r353787) @@ -145,7 +145,7 @@ struct vm_map_entry { #define MAP_ENTRY_GROWS_UP 0x00002000 /* bottom-up stacks */ #define MAP_ENTRY_WIRE_SKIPPED 0x00004000 -#define MAP_ENTRY_VN_WRITECNT 0x00008000 /* writeable vnode +#define MAP_ENTRY_WRITECNT 0x00008000 /* tracked writeable mapping */ #define MAP_ENTRY_GUARD 0x00010000 #define MAP_ENTRY_STACK_GAP_DN 0x00020000 @@ -349,7 +349,7 @@ long vmspace_resident_count(struct vmspace *vmspace); #define MAP_CREATE_GUARD 0x00000080 #define MAP_DISABLE_COREDUMP 0x00000100 #define MAP_PREFAULT_MADVISE 0x00000200 /* from (user) madvise request */ -#define MAP_VN_WRITECOUNT 0x00000400 +#define MAP_WRITECOUNT 0x00000400 #define MAP_REMAP 0x00000800 #define MAP_STACK_GROWS_DOWN 0x00001000 #define MAP_STACK_GROWS_UP 0x00002000 Modified: stable/12/sys/vm/vm_mmap.c ============================================================================== --- stable/12/sys/vm/vm_mmap.c Sun Oct 20 22:55:47 2019 (r353786) +++ stable/12/sys/vm/vm_mmap.c Mon Oct 21 00:08:34 2019 (r353787) @@ -1242,7 +1242,7 @@ vm_mmap_vnode(struct thread *td, vm_size_t objsize, } if (writex) { *writecounted = TRUE; - vnode_pager_update_writecount(obj, 0, objsize); + vm_pager_update_writecount(obj, 0, objsize); } } else { error = EINVAL; @@ -1298,7 +1298,7 @@ vm_mmap_vnode(struct thread *td, vm_size_t objsize, done: if (error != 0 && *writecounted) { *writecounted = FALSE; - vnode_pager_update_writecount(obj, objsize, 0); + vm_pager_update_writecount(obj, objsize, 0); } vput(vp); return (error); @@ -1433,7 +1433,7 @@ vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t siz * writecount, then undo that now. */ if (writecounted) - vnode_pager_release_writecount(object, 0, size); + vm_pager_release_writecount(object, 0, size); vm_object_deallocate(object); } return (error); @@ -1522,7 +1522,7 @@ vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_siz if (flags & MAP_SHARED) docow |= MAP_INHERIT_SHARE; if (writecounted) - docow |= MAP_VN_WRITECOUNT; + docow |= MAP_WRITECOUNT; if (flags & MAP_STACK) { if (object != NULL) return (EINVAL); Modified: stable/12/sys/vm/vm_object.h ============================================================================== --- stable/12/sys/vm/vm_object.h Sun Oct 20 22:55:47 2019 (r353786) +++ stable/12/sys/vm/vm_object.h Mon Oct 21 00:08:34 2019 (r353787) @@ -167,6 +167,7 @@ struct vm_object { struct { void *swp_tmpfs; struct pctrie swp_blks; + vm_ooffset_t writemappings; } swp; } un_pager; struct ucred *cred; Modified: stable/12/sys/vm/vm_pager.h ============================================================================== --- stable/12/sys/vm/vm_pager.h Sun Oct 20 22:55:47 2019 (r353786) +++ stable/12/sys/vm/vm_pager.h Mon Oct 21 00:08:34 2019 (r353787) @@ -61,6 +61,7 @@ typedef boolean_t pgo_haspage_t(vm_object_t, vm_pindex typedef int pgo_populate_t(vm_object_t, vm_pindex_t, int, vm_prot_t, vm_pindex_t *, vm_pindex_t *); typedef void pgo_pageunswapped_t(vm_page_t); +typedef void pgo_writecount_t(vm_object_t, vm_offset_t, vm_offset_t); struct pagerops { pgo_init_t *pgo_init; /* Initialize pager. */ @@ -72,6 +73,9 @@ struct pagerops { pgo_haspage_t *pgo_haspage; /* Query page. */ pgo_populate_t *pgo_populate; /* Bulk spec pagein. */ pgo_pageunswapped_t *pgo_pageunswapped; + /* Operations for specialized writecount handling */ + pgo_writecount_t *pgo_update_writecount; + pgo_writecount_t *pgo_release_writecount; }; extern struct pagerops defaultpagerops; @@ -190,6 +194,26 @@ vm_pager_page_unswapped(vm_page_t m) VM_OBJECT_ASSERT_LOCKED(m->object); if (pagertab[m->object->type]->pgo_pageunswapped) (*pagertab[m->object->type]->pgo_pageunswapped)(m); +} + +static __inline void +vm_pager_update_writecount(vm_object_t object, vm_offset_t start, + vm_offset_t end) +{ + + if (pagertab[object->type]->pgo_update_writecount) + pagertab[object->type]->pgo_update_writecount(object, start, + end); +} + +static __inline void +vm_pager_release_writecount(vm_object_t object, vm_offset_t start, + vm_offset_t end) +{ + + if (pagertab[object->type]->pgo_release_writecount) + pagertab[object->type]->pgo_release_writecount(object, start, + end); } struct cdev_pager_ops { Modified: stable/12/sys/vm/vnode_pager.c ============================================================================== --- stable/12/sys/vm/vnode_pager.c Sun Oct 20 22:55:47 2019 (r353786) +++ stable/12/sys/vm/vnode_pager.c Mon Oct 21 00:08:34 2019 (r353787) @@ -97,6 +97,10 @@ static vm_object_t vnode_pager_alloc(void *, vm_ooffse vm_ooffset_t, struct ucred *cred); static int vnode_pager_generic_getpages_done(struct buf *); static void vnode_pager_generic_getpages_done_async(struct buf *); +static void vnode_pager_update_writecount(vm_object_t, vm_offset_t, + vm_offset_t); +static void vnode_pager_release_writecount(vm_object_t, vm_offset_t, + vm_offset_t); struct pagerops vnodepagerops = { .pgo_alloc = vnode_pager_alloc, @@ -105,6 +109,8 @@ struct pagerops vnodepagerops = { .pgo_getpages_async = vnode_pager_getpages_async, .pgo_putpages = vnode_pager_putpages, .pgo_haspage = vnode_pager_haspage, + .pgo_update_writecount = vnode_pager_update_writecount, + .pgo_release_writecount = vnode_pager_release_writecount, }; int vnode_pbuf_freecnt; @@ -1499,7 +1505,7 @@ done: VM_OBJECT_WUNLOCK(obj); } -void +static void vnode_pager_update_writecount(vm_object_t object, vm_offset_t start, vm_offset_t end) { @@ -1528,7 +1534,7 @@ vnode_pager_update_writecount(vm_object_t object, vm_o VM_OBJECT_WUNLOCK(object); } -void +static void vnode_pager_release_writecount(vm_object_t object, vm_offset_t start, vm_offset_t end) { Modified: stable/12/sys/vm/vnode_pager.h ============================================================================== --- stable/12/sys/vm/vnode_pager.h Sun Oct 20 22:55:47 2019 (r353786) +++ stable/12/sys/vm/vnode_pager.h Mon Oct 21 00:08:34 2019 (r353787) @@ -50,12 +50,8 @@ int vnode_pager_generic_putpages(struct vnode *vp, vm_ int vnode_pager_local_getpages(struct vop_getpages_args *ap); int vnode_pager_local_getpages_async(struct vop_getpages_async_args *ap); int vnode_pager_putpages_ioflags(int pager_flags); -void vnode_pager_release_writecount(vm_object_t object, vm_offset_t start, - vm_offset_t end); void vnode_pager_undirty_pages(vm_page_t *ma, int *rtvals, int written, off_t eof, int lpos); -void vnode_pager_update_writecount(vm_object_t object, vm_offset_t start, - vm_offset_t end); #endif /* _KERNEL */ #endif /* _VNODE_PAGER_ */
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201910210008.x9L08ZQt027865>