]> git.saurik.com Git - apple/xnu.git/blobdiff - bsd/dev/dtrace/dtrace_ptss.c
xnu-1228.tar.gz
[apple/xnu.git] / bsd / dev / dtrace / dtrace_ptss.c
diff --git a/bsd/dev/dtrace/dtrace_ptss.c b/bsd/dev/dtrace/dtrace_ptss.c
new file mode 100644 (file)
index 0000000..8e2ec27
--- /dev/null
@@ -0,0 +1,309 @@
+/*
+ * Copyright (c) 2000-2006 Apple Computer, Inc. All rights reserved.
+ *
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
+ * 
+ * This file contains Original Code and/or Modifications of Original Code
+ * as defined in and that are subject to the Apple Public Source License
+ * Version 2.0 (the 'License'). You may not use this file except in
+ * compliance with the License. The rights granted to you under the License
+ * may not be used to create, or enable the creation or redistribution of,
+ * unlawful or unlicensed copies of an Apple operating system, or to
+ * circumvent, violate, or enable the circumvention or violation of, any
+ * terms of an Apple operating system software license agreement.
+ * 
+ * Please obtain a copy of the License at
+ * http://www.opensource.apple.com/apsl/ and read it before using this file.
+ * 
+ * The Original Code and all software distributed under the License are
+ * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
+ * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
+ * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
+ * Please see the License for the specific language governing rights and
+ * limitations under the License.
+ * 
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
+ */
+
+#include <sys/types.h>
+#include <sys/proc.h>
+#include <sys/proc_internal.h>
+#include <sys/systm.h>
+#include <sys/user.h>
+#include <sys/dtrace_ptss.h>
+
+#include <mach/vm_param.h>
+#include <mach/mach_vm.h>
+
+#include <kern/task.h>
+
+#include <vm/vm_map.h>
+
+/*
+ * This function requires the sprlock to be held
+ *
+ * In general, it will not block. If it needs to allocate a new
+ * page of memory, the underlying kernel _MALLOC may block.
+ */
+struct dtrace_ptss_page_entry*
+dtrace_ptss_claim_entry_locked(struct proc* p) {
+       lck_mtx_assert(&p->p_dtrace_sprlock, LCK_MTX_ASSERT_OWNED);
+
+       struct dtrace_ptss_page_entry* entry = NULL;
+
+       while (TRUE) {
+               struct dtrace_ptss_page_entry* temp = p->p_dtrace_ptss_free_list;
+
+               if (temp == NULL) {
+                       // Nothing on the free list. Allocate a new page, its okay if multiple threads race here.
+                       struct dtrace_ptss_page* page = dtrace_ptss_allocate_page(p);
+
+                       // Make sure we actually got a page
+                       if (page == NULL)
+                               return NULL;
+
+                       // Add the page to the page list
+                       page->next = p->p_dtrace_ptss_pages;
+                       p->p_dtrace_ptss_pages = page;
+
+                       // CAS the entries onto the free list.
+                       do {
+                               page->entries[DTRACE_PTSS_ENTRIES_PER_PAGE-1].next = p->p_dtrace_ptss_free_list;
+                       } while (!OSCompareAndSwap((UInt32)page->entries[DTRACE_PTSS_ENTRIES_PER_PAGE-1].next,
+                                                  (UInt32)&page->entries[0],
+                                                  (volatile UInt32 *)&p->p_dtrace_ptss_free_list));
+                                
+                       // Now that we've added to the free list, try again.
+                       continue;
+               }
+
+               // Claim temp
+               if (!OSCompareAndSwap((UInt32)temp, (UInt32)temp->next, (volatile UInt32 *)&p->p_dtrace_ptss_free_list))
+                       continue;
+
+               // At this point, we own temp.
+               entry = temp;
+
+               break;
+       }
+
+       return entry;
+}
+
+/*
+ * This function does not require any locks to be held on entry.
+ */
+struct dtrace_ptss_page_entry*
+dtrace_ptss_claim_entry(struct proc* p) {
+       // Verify no locks held on entry
+       lck_mtx_assert(&p->p_dtrace_sprlock, LCK_MTX_ASSERT_NOTOWNED);
+       lck_mtx_assert(&p->p_mlock, LCK_MTX_ASSERT_NOTOWNED);
+
+       struct dtrace_ptss_page_entry* entry = NULL;
+
+       while (TRUE) {
+               struct dtrace_ptss_page_entry* temp = p->p_dtrace_ptss_free_list;
+
+               if (temp == NULL) {
+                       lck_mtx_lock(&p->p_dtrace_sprlock);
+                       temp = dtrace_ptss_claim_entry_locked(p);
+                       lck_mtx_unlock(&p->p_dtrace_sprlock);
+                       return temp;
+               }
+
+               // Claim temp
+               if (!OSCompareAndSwap((UInt32)temp, (UInt32)temp->next, (volatile UInt32 *)&p->p_dtrace_ptss_free_list))
+                       continue;
+
+               // At this point, we own temp.
+               entry = temp;
+
+               break;
+       }
+
+       return entry;
+}
+
+/*
+ * This function does not require any locks to be held on entry.
+ */
+void
+dtrace_ptss_release_entry(struct proc* p, struct dtrace_ptss_page_entry* e) {
+       if (p && e) {
+               do {
+                       e->next = p->p_dtrace_ptss_free_list;
+               } while (!OSCompareAndSwap((UInt32)e->next, (UInt32)e, (volatile UInt32 *)&p->p_dtrace_ptss_free_list));
+       }
+}
+
+/*
+ * This function allocates a new page in the target process's address space.
+ * 
+ * It returns a dtrace_ptss_page that has its entries chained, with the last
+ * entries next field set to NULL. It does not add the page or the entries to
+ * the process's page/entry lists.
+ *
+ * This function does not require that any locks be held when it is invoked.
+ */
+struct dtrace_ptss_page*
+dtrace_ptss_allocate_page(struct proc* p)
+{
+       // Allocate the kernel side data
+       struct dtrace_ptss_page* ptss_page = _MALLOC(sizeof(struct dtrace_ptss_page), M_TEMP, M_ZERO | M_WAITOK);
+       if (ptss_page == NULL)
+               return NULL;
+
+       // Now allocate a page in user space and set its protections to allow execute.
+       task_t task = p->task;
+       vm_map_t map = get_task_map_reference(task);
+
+       mach_vm_address_t addr = 0LL;
+       mach_vm_size_t size = PAGE_SIZE; // We need some way to assert that this matches vm_map_round_page() !!!
+
+       kern_return_t kr = mach_vm_allocate(map, &addr, size, VM_FLAGS_ANYWHERE);
+       if (kr != KERN_SUCCESS) {
+               goto err;
+       }
+
+       kr = mach_vm_protect(map, addr, size, 0, VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
+       if (kr != KERN_SUCCESS) {
+               mach_vm_deallocate(map, addr, size);
+               goto err;
+       }       
+
+       // Chain the page entries.
+       int i;
+       for (i=0; i<DTRACE_PTSS_ENTRIES_PER_PAGE; i++) {
+               ptss_page->entries[i].addr = addr + (i * DTRACE_PTSS_SCRATCH_SPACE_PER_THREAD);
+               ptss_page->entries[i].next = &ptss_page->entries[i+1];
+       }
+
+       // The last entry should point to NULL
+       ptss_page->entries[DTRACE_PTSS_ENTRIES_PER_PAGE-1].next = NULL;
+
+       vm_map_deallocate(map);
+
+       return ptss_page;
+
+err:
+       _FREE(ptss_page, M_TEMP);
+
+       vm_map_deallocate(map);
+
+       return NULL;
+}
+
+/*
+ * This function frees an existing page in the target process's address space.
+ * 
+ * It does not alter any of the process's page/entry lists.
+ *
+ * TODO: Inline in dtrace_ptrace_exec_exit?
+ */
+void
+dtrace_ptss_free_page(struct proc* p, struct dtrace_ptss_page* ptss_page)
+{
+       // Grab the task and get a reference to its vm_map
+       task_t task = p->task;
+       vm_map_t map = get_task_map_reference(task);
+
+       mach_vm_address_t addr = ptss_page->entries[0].addr;
+       mach_vm_size_t size = PAGE_SIZE; // We need some way to assert that this matches vm_map_round_page() !!!
+
+       // Silent failures, no point in checking return code.
+       mach_vm_deallocate(map, addr, size);
+
+       vm_map_deallocate(map);
+}
+
+/*
+ * This function assumes that the target process has been
+ * suspended, and the proc_lock & sprlock is held 
+ */
+void
+dtrace_ptss_enable(struct proc* p) {
+       lck_mtx_assert(&p->p_dtrace_sprlock, LCK_MTX_ASSERT_OWNED);
+       lck_mtx_assert(&p->p_mlock, LCK_MTX_ASSERT_OWNED);
+
+       struct uthread* uth;
+       /*
+        * XXX There has been a concern raised about holding the proc_lock
+        * while calling dtrace_ptss_claim_entry(), due to the fact
+        * that dtrace_ptss_claim_entry() can potentially malloc.
+        */
+       TAILQ_FOREACH(uth, &p->p_uthlist, uu_list) {
+               uth->t_dtrace_scratch = dtrace_ptss_claim_entry_locked(p);
+       }
+}
+
+/*
+ * This function is not thread safe.
+ *
+ * It assumes the sprlock is held, and the proc_lock is not.
+ */
+void
+dtrace_ptss_exec_exit(struct proc* p) {
+       /*
+        * Should hold sprlock to touch the pages list. Must not
+        * hold the proc lock to avoid deadlock.
+        */
+       lck_mtx_assert(&p->p_dtrace_sprlock, LCK_MTX_ASSERT_OWNED);
+       lck_mtx_assert(&p->p_mlock, LCK_MTX_ASSERT_NOTOWNED);
+
+       p->p_dtrace_ptss_free_list = NULL;
+
+       struct dtrace_ptss_page* temp = p->p_dtrace_ptss_pages;
+       p->p_dtrace_ptss_pages = NULL;
+
+       while (temp != NULL) {
+               struct dtrace_ptss_page* next = temp->next;
+               
+               // Do we need to specifically mach_vm_deallocate the user pages?
+               // This can be called when the process is exiting, I believe the proc's
+               // vm_map_t may already be toast.
+               
+               // Must be certain to free the kernel memory!
+               _FREE(temp, M_TEMP);
+               temp = next;
+       }
+}
+
+/*
+ * This function is not thread safe. It is not used for vfork.
+ *
+ * The child proc ptss fields are initialized to NULL at fork time.
+ * Pages allocated in the parent are copied as part of the vm_map copy, though.
+ * We need to deallocate those pages.
+ *
+ * Parent and child sprlock should be held, and proc_lock must NOT be held.
+ */
+void
+dtrace_ptss_fork(struct proc* parent, struct proc* child) {
+       // The child should not have any pages/entries allocated at this point.
+       // ASSERT(child->p_dtrace_ptss_pages == NULL);
+       // ASSERT(child->p_dtrace_ptss_free_list == NULL);
+
+       /*
+        * The parent's sprlock should be held, to protect its pages list
+        * from changing while the child references it. The child's sprlock
+        * must also be held, because we are modifying its pages list.
+        * Finally, to prevent a deadlock with the fasttrap cleanup code,
+        * neither the parent or child proc_lock should be held.
+        */
+       lck_mtx_assert(&parent->p_dtrace_sprlock, LCK_MTX_ASSERT_OWNED);
+       lck_mtx_assert(&parent->p_mlock, LCK_MTX_ASSERT_NOTOWNED);
+       lck_mtx_assert(&child->p_dtrace_sprlock, LCK_MTX_ASSERT_OWNED);
+       lck_mtx_assert(&child->p_mlock, LCK_MTX_ASSERT_NOTOWNED);
+
+       // Get page list from *PARENT*
+       struct dtrace_ptss_page* temp = parent->p_dtrace_ptss_pages;
+
+       while (temp != NULL) {          
+               // Freeing the page in the *CHILD*
+               dtrace_ptss_free_page(child, temp);
+
+               // Do not free the kernel memory, it belong to the parent.
+               temp = temp->next;
+       }
+}