]> git.saurik.com Git - apple/xnu.git/blobdiff - bsd/vfs/vfs_bio.c
xnu-2782.10.72.tar.gz
[apple/xnu.git] / bsd / vfs / vfs_bio.c
index 53f4f5576ba5474cedb90bf6188049aeb7efc85b..c6e919d9e806e19ca306a9c397a5da52a2207649 100644 (file)
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2000-2010 Apple Inc. All rights reserved.
+ * Copyright (c) 2000-2014 Apple Inc. All rights reserved.
  *
  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  * 
@@ -91,7 +91,8 @@
 #endif /* DIAGNOSTIC */
 #include <kern/task.h>
 #include <kern/zalloc.h>
-#include <kern/lock.h>
+#include <kern/locks.h>
+#include <kern/thread.h>
 
 #include <sys/fslog.h>         /* fslog_io_error() */
 
 #include <sys/ubc_internal.h>
 
 #include <sys/sdt.h>
-
-
-#if BALANCE_QUEUES
-static __inline__ void bufqinc(int q);
-static __inline__ void bufqdec(int q);
-#endif
+#include <sys/cprotect.h>
 
 int    bcleanbuf(buf_t bp, boolean_t discard);
 static int     brecover_data(buf_t bp);
@@ -137,8 +133,8 @@ static buf_t        buf_create_shadow_internal(buf_t bp, boolean_t force_copy,
 __private_extern__ int  bdwrite_internal(buf_t, int);
 
 /* zone allocated buffer headers */
-static void    bufzoneinit(void) __attribute__((section("__TEXT, initcode")));
-static void    bcleanbuf_thread_init(void) __attribute__((section("__TEXT, initcode")));
+static void    bufzoneinit(void);
+static void    bcleanbuf_thread_init(void);
 static void    bcleanbuf_thread(void);
 
 static zone_t  buf_hdr_zone;
@@ -189,17 +185,6 @@ buf_timestamp(void)
 /*
  * Insq/Remq for the buffer free lists.
  */
-#if BALANCE_QUEUES
-#define        binsheadfree(bp, dp, whichq)    do { \
-                                   TAILQ_INSERT_HEAD(dp, bp, b_freelist); \
-                                       bufqinc((whichq));      \
-                               } while (0)
-
-#define        binstailfree(bp, dp, whichq)    do { \
-                                   TAILQ_INSERT_TAIL(dp, bp, b_freelist); \
-                                       bufqinc((whichq));      \
-                               } while (0)
-#else
 #define        binsheadfree(bp, dp, whichq)    do { \
                                    TAILQ_INSERT_HEAD(dp, bp, b_freelist); \
                                } while (0)
@@ -207,8 +192,6 @@ buf_timestamp(void)
 #define        binstailfree(bp, dp, whichq)    do { \
                                    TAILQ_INSERT_TAIL(dp, bp, b_freelist); \
                                } while (0)
-#endif
-
 
 #define BHASHENTCHECK(bp)      \
        if ((bp)->b_hash.le_prev != (struct buf **)0xdeadbeef)  \
@@ -382,32 +365,129 @@ buf_markfua(buf_t bp) {
         SET(bp->b_flags, B_FUA);
 }
 
-#ifdef CONFIG_PROTECT
+#if CONFIG_PROTECT
+void
+buf_setcpaddr(buf_t bp, struct cprotect *entry) {
+       bp->b_attr.ba_cpentry = entry;
+}
+
+void
+buf_setcpoff (buf_t bp, uint64_t foffset) {
+       bp->b_attr.ba_cp_file_off = foffset;
+}
+
 void *
-buf_getcpaddr(buf_t bp) {
-       return bp->b_cpentry;
+bufattr_cpaddr(bufattr_t bap) {
+       return (bap->ba_cpentry);
 }
 
-void 
-buf_setcpaddr(buf_t bp, void *cp_entry_addr) {
-       bp->b_cpentry = (struct cprotect *) cp_entry_addr;
+uint64_t
+bufattr_cpoff(bufattr_t bap) {
+       return (bap->ba_cp_file_off);
+}
+
+void
+bufattr_setcpaddr(bufattr_t bap, void *cp_entry_addr) {
+        bap->ba_cpentry = cp_entry_addr;
+}
+
+void
+bufattr_setcpoff(bufattr_t bap, uint64_t foffset) {
+        bap->ba_cp_file_off = foffset;
 }
 
 #else
 void *
-buf_getcpaddr(buf_t bp __unused) {
-       return NULL;
+bufattr_cpaddr(bufattr_t bap __unused) {
+        return NULL;
 }
 
-void 
-buf_setcpaddr(buf_t bp __unused, void *cp_entry_addr __unused) {
+uint64_t
+bufattr_cpoff(bufattr_t bap __unused) {
+       return 0;
+}
+
+void
+bufattr_setcpaddr(bufattr_t bap __unused, void *cp_entry_addr __unused) {
+}
+
+void
+bufattr_setcpoff(__unused bufattr_t bap, __unused uint64_t foffset) {
        return;
 }
 #endif /* CONFIG_PROTECT */
 
+bufattr_t
+bufattr_alloc() {
+       bufattr_t bap;
+       MALLOC(bap, bufattr_t, sizeof(struct bufattr), M_TEMP, M_WAITOK);
+       if (bap == NULL)
+               return NULL;
+
+       bzero(bap, sizeof(struct bufattr));
+       return bap;
+}
+
+void
+bufattr_free(bufattr_t bap) {
+       if (bap)
+               FREE(bap, M_TEMP);
+}
+
+bufattr_t
+bufattr_dup(bufattr_t bap) {
+       bufattr_t new_bufattr;
+       MALLOC(new_bufattr, bufattr_t, sizeof(struct bufattr), M_TEMP, M_WAITOK);
+       if (new_bufattr == NULL)
+               return NULL;
+
+       /* Copy the provided one into the new copy */
+       memcpy (new_bufattr, bap, sizeof(struct bufattr));
+       return new_bufattr;
+}
+
+int
+bufattr_rawencrypted(bufattr_t bap) {
+       if ( (bap->ba_flags & BA_RAW_ENCRYPTED_IO) )
+               return 1;
+       return 0;
+}
+
 int
 bufattr_throttled(bufattr_t bap) {
-       if ( (bap->ba_flags & BA_THROTTLED_IO) )
+       return (GET_BUFATTR_IO_TIER(bap));
+}
+
+int
+bufattr_passive(bufattr_t bap) {
+       if ( (bap->ba_flags & BA_PASSIVE) )
+               return 1;
+       return 0;
+}
+
+int
+bufattr_nocache(bufattr_t bap) {
+       if ( (bap->ba_flags & BA_NOCACHE) )
+               return 1;
+       return 0;
+}
+
+int
+bufattr_meta(bufattr_t bap) {
+       if ( (bap->ba_flags & BA_META) )
+               return 1;
+       return 0;
+}
+
+void
+bufattr_markmeta(bufattr_t bap) {
+       SET(bap->ba_flags,  BA_META);
+}
+
+int
+bufattr_delayidlesleep(bufattr_t bap) 
+{
+       if ( (bap->ba_flags & BA_DELAYIDLESLEEP) )
                return 1;
        return 0;
 }
@@ -417,6 +497,54 @@ buf_attr(buf_t bp) {
        return &bp->b_attr;
 }
 
+void 
+buf_markstatic(buf_t bp __unused) {
+       SET(bp->b_flags, B_STATICCONTENT);
+}
+
+int
+buf_static(buf_t bp) {
+    if ( (bp->b_flags & B_STATICCONTENT) )
+        return 1;
+    return 0;
+}
+
+void 
+bufattr_markgreedymode(bufattr_t bap) {
+       SET(bap->ba_flags, BA_GREEDY_MODE);
+}
+
+int
+bufattr_greedymode(bufattr_t bap) {
+    if ( (bap->ba_flags & BA_GREEDY_MODE) )
+        return 1;
+    return 0;
+}
+
+void 
+bufattr_markisochronous(bufattr_t bap) {
+       SET(bap->ba_flags, BA_ISOCHRONOUS);
+}
+
+int
+bufattr_isochronous(bufattr_t bap) {
+    if ( (bap->ba_flags & BA_ISOCHRONOUS) )
+        return 1;
+    return 0;
+}
+
+void 
+bufattr_markquickcomplete(bufattr_t bap) {
+       SET(bap->ba_flags, BA_QUICK_COMPLETE);
+}
+
+int
+bufattr_quickcomplete(bufattr_t bap) {
+    if ( (bap->ba_flags & BA_QUICK_COMPLETE) )
+        return 1;
+    return 0;
+}
+
 errno_t
 buf_error(buf_t bp) {
         
@@ -1102,7 +1230,7 @@ buf_strategy(vnode_t devvp, void *ap)
         errno_t error;
 #if CONFIG_DTRACE
        int dtrace_io_start_flag = 0;    /* We only want to trip the io:::start
-                                         * probe once, with the true phisical
+                                         * probe once, with the true physical
                                          * block in place (b_blkno)
                                          */
 
@@ -1135,7 +1263,7 @@ buf_strategy(vnode_t devvp, void *ap)
                        return (cluster_bp(bp));
                }
                if (bp->b_blkno == bp->b_lblkno) {
-                       off_t   f_offset;
+                   off_t       f_offset;
                        size_t  contig_bytes;
                  
                        if ((error = VNOP_BLKTOOFF(vp, bp->b_lblkno, &f_offset))) {
@@ -1143,21 +1271,22 @@ buf_strategy(vnode_t devvp, void *ap)
                                buf_seterror(bp, error);
                                buf_biodone(bp);
 
-                               return (error);
+                           return (error);
                        }
-                       if ((error = VNOP_BLOCKMAP(vp, f_offset, bp->b_bcount, &bp->b_blkno, &contig_bytes, NULL, bmap_flags, NULL))) {
+
+               if ((error = VNOP_BLOCKMAP(vp, f_offset, bp->b_bcount, &bp->b_blkno, &contig_bytes, NULL, bmap_flags, NULL))) {
                                DTRACE_IO1(start, buf_t, bp);
                                buf_seterror(bp, error);
                                buf_biodone(bp);
 
                                return (error);
                        }
-                       
+
                        DTRACE_IO1(start, buf_t, bp);
 #if CONFIG_DTRACE
                        dtrace_io_start_flag = 1;
 #endif /* CONFIG_DTRACE */                     
-                       
+
                        if ((bp->b_blkno == -1) || (contig_bytes == 0)) {
                                /* Set block number to force biodone later */
                                bp->b_blkno = -1;
@@ -1186,15 +1315,45 @@ buf_strategy(vnode_t devvp, void *ap)
                DTRACE_IO1(start, buf_t, bp);
 #endif /* CONFIG_DTRACE */
        
+#if CONFIG_PROTECT
+       /* Capture f_offset in the bufattr*/
+       if (bp->b_attr.ba_cpentry != 0) {
+               /* No need to go here for older EAs */
+               if(bp->b_attr.ba_cpentry->cp_flags & CP_OFF_IV_ENABLED) {
+                       off_t f_offset;
+                       if ((error = VNOP_BLKTOOFF(bp->b_vp, bp->b_lblkno, &f_offset)))
+                               return error;
+
+                       /* 
+                        * Attach the file offset to this buffer.  The
+                        * bufattr attributes will be passed down the stack
+                        * until they reach IOFlashStorage.  IOFlashStorage
+                        * will retain the offset in a local variable when it
+                        * issues its I/Os to the NAND controller.       
+                        * 
+                        * Note that LwVM may end up splitting this I/O 
+                        * into sub-I/Os if it crosses a chunk boundary.  In this
+                        * case, LwVM will update this field when it dispatches
+                        * each I/O to IOFlashStorage.  But from our perspective
+                        * we have only issued a single I/O.
+                        */
+                       bufattr_setcpoff (&(bp->b_attr), (u_int64_t)f_offset);
+                       CP_DEBUG((CPDBG_OFFSET_IO | DBG_FUNC_NONE), (uint32_t) f_offset, (uint32_t) bp->b_lblkno, (uint32_t) bp->b_blkno, (uint32_t) bp->b_bcount, 0);
+               }
+       }
+#endif
+
        /*
         * we can issue the I/O because...
         * either B_CLUSTER is set which
         * means that the I/O is properly set
         * up to be a multiple of the page size, or
         * we were able to successfully set up the
-        * phsyical block mapping
+        * physical block mapping
         */
-       return (VOCALL(devvp->v_op, VOFFSET(vnop_strategy), ap));
+       error = VOCALL(devvp->v_op, VOFFSET(vnop_strategy), ap);
+       DTRACE_FSINFO(strategy, vnode_t, vp);
+       return (error);
 }
 
 
@@ -1489,12 +1648,20 @@ try_dirty_list:
 
 void
 buf_flushdirtyblks(vnode_t vp, int wait, int flags, const char *msg) {
+
+       (void) buf_flushdirtyblks_skipinfo(vp, wait, flags, msg);
+       return;
+}
+
+int
+buf_flushdirtyblks_skipinfo(vnode_t vp, int wait, int flags, const char *msg) {
        buf_t   bp;
        int     writes_issued = 0;
        errno_t error;
        int     busy = 0;
        struct  buflists local_iterblkhd;
        int     lock_flags = BAC_NOWAIT | BAC_REMOVE;
+       int any_locked = 0;
 
        if (flags & BUF_SKIP_LOCKED)
                lock_flags |= BAC_SKIP_LOCKED;
@@ -1508,11 +1675,26 @@ loop:
                        bp = LIST_FIRST(&local_iterblkhd);
                        LIST_REMOVE(bp, b_vnbufs);
                        LIST_INSERT_HEAD(&vp->v_dirtyblkhd, bp, b_vnbufs);
-                       
-                       if ((error = buf_acquire_locked(bp, lock_flags, 0, 0)) == EBUSY)
-                               busy++;
-                       if (error)
-                               continue;
+
+                       if ((error = buf_acquire_locked(bp, lock_flags, 0, 0)) == EBUSY) {
+                               busy++;
+                       }
+                       if (error) {
+                               /* 
+                                * If we passed in BUF_SKIP_LOCKED or BUF_SKIP_NONLOCKED,
+                                * we may want to do somethign differently if a locked or unlocked
+                                * buffer was encountered (depending on the arg specified).
+                                * In this case, we know that one of those two was set, and the
+                                * buf acquisition failed above.  
+                                * 
+                                * If it failed with EDEADLK, then save state which can be emitted
+                                * later on to the caller.  Most callers should not care.
+                                */
+                               if (error == EDEADLK) {
+                                       any_locked++;
+                               }
+                               continue;
+                       }
                        lck_mtx_unlock(buf_mtxp);
 
                        bp->b_flags &= ~B_LOCKED;
@@ -1558,6 +1740,8 @@ loop:
                        goto loop;
                }
        }
+
+       return any_locked;
 }
 
 
@@ -1654,9 +1838,6 @@ bremfree_locked(buf_t bp)
        }
        TAILQ_REMOVE(dp, bp, b_freelist);
 
-#if BALANCE_QUEUES
-       bufqdec(whichq);
-#endif
        if (whichq == BQ_LAUNDRY)
                blaundrycnt--;
 
@@ -1834,17 +2015,8 @@ bufinit(void)
                panic("Couldn't register buffer cache callout for vm pressure!\n");
        }
 
-#if BALANCE_QUEUES
-       {
-       static void bufq_balance_thread_init(void) __attribute__((section("__TEXT, initcode")));
-       /* create a thread to do dynamic buffer queue balancing */
-       bufq_balance_thread_init();
-       }
-#endif /* notyet */
 }
 
-
-
 /*
  * Zones for the meta data buffers
  */
@@ -1935,8 +2107,9 @@ bio_doread(vnode_t vp, daddr64_t blkno, int size, kauth_cred_t cred, int async,
                trace(TR_BREADMISS, pack(vp, size), blkno);
 
                /* Pay for the read. */
-               if (p && p->p_stats) 
+               if (p && p->p_stats) 
                        OSIncrementAtomicLong(&p->p_stats->p_ru.ru_inblock);            /* XXX */
+               }
 
                if (async) {
                        /*
@@ -2070,9 +2243,10 @@ buf_bwrite(buf_t bp)
                 */
                if (wasdelayed)
                        buf_reassign(bp, vp);
-               else
-               if (p && p->p_stats) 
-                       OSIncrementAtomicLong(&p->p_stats->p_ru.ru_oublock);    /* XXX */
+               else 
+                       if (p && p->p_stats) {
+                               OSIncrementAtomicLong(&p->p_stats->p_ru.ru_oublock);    /* XXX */
+                       }
        }
        trace(TR_BUFWRITE, pack(vp, bp->b_bcount), bp->b_lblkno);
 
@@ -2096,8 +2270,9 @@ buf_bwrite(buf_t bp)
                if (wasdelayed)
                        buf_reassign(bp, vp);
                else
-               if (p && p->p_stats) 
-                       OSIncrementAtomicLong(&p->p_stats->p_ru.ru_oublock);    /* XXX */
+                       if (p && p->p_stats) { 
+                               OSIncrementAtomicLong(&p->p_stats->p_ru.ru_oublock);    /* XXX */
+                       }
 
                /* Release the buffer. */
                // XXXdbg - only if the unused bit is set
@@ -2152,8 +2327,9 @@ bdwrite_internal(buf_t bp, int return_error)
         */
        if (!ISSET(bp->b_flags, B_DELWRI)) {
                SET(bp->b_flags, B_DELWRI);
-               if (p && p->p_stats) 
+               if (p && p->p_stats) 
                        OSIncrementAtomicLong(&p->p_stats->p_ru.ru_oublock);    /* XXX */
+               }
                OSAddAtomicLong(1, &nbdwrite);
                buf_reassign(bp, vp);
        }
@@ -2267,6 +2443,8 @@ buf_brelse_shadow(buf_t bp)
        buf_t   bp_data;
        int     data_ref = 0;
 #endif
+       int need_wakeup = 0;
+
        lck_mtx_lock_spin(buf_mtxp);
 
        bp_head = (buf_t)bp->b_orig;
@@ -2334,8 +2512,16 @@ buf_brelse_shadow(buf_t bp)
 
                        bp_return = bp_head;
                }
+               if (ISSET(bp_head->b_lflags, BL_WANTED_REF)) {
+                       CLR(bp_head->b_lflags, BL_WANTED_REF);
+                       need_wakeup = 1;
+               }
        }
        lck_mtx_unlock(buf_mtxp);
+
+       if (need_wakeup)
+               wakeup(bp_head);
+
 #ifdef BUF_MAKE_PRIVATE        
        if (bp == bp_data && data_ref == 0)
                buf_free_meta_store(bp);
@@ -2663,6 +2849,30 @@ incore_locked(vnode_t vp, daddr64_t blkno, struct bufhashhdr *dp)
 }
 
 
+void
+buf_wait_for_shadow_io(vnode_t vp, daddr64_t blkno)
+{
+       buf_t bp;
+       struct  bufhashhdr *dp;
+
+       dp = BUFHASH(vp, blkno);
+
+       lck_mtx_lock_spin(buf_mtxp);
+
+       for (;;) {
+               if ((bp = incore_locked(vp, blkno, dp)) == NULL)
+                       break;
+
+               if (bp->b_shadow_ref == 0)
+                       break;
+
+               SET(bp->b_lflags, BL_WANTED_REF);
+
+               (void) msleep(bp, buf_mtxp, PSPIN | (PRIBIO+1), "buf_wait_for_shadow", NULL);
+       }
+       lck_mtx_unlock(buf_mtxp);
+}
+       
 /* XXX FIXME -- Update the comment to reflect the UBC changes (please) -- */
 /*
  * Get a block of requested size that is associated with
@@ -3027,6 +3237,80 @@ buf_clear_redundancy_flags(buf_t bp, uint32_t flags)
        CLR(bp->b_redundancy_flags, flags);
 }
 
+
+
+static void *
+recycle_buf_from_pool(int nsize)
+{
+       buf_t   bp;
+       void    *ptr = NULL;
+
+       lck_mtx_lock_spin(buf_mtxp);
+
+       TAILQ_FOREACH(bp, &bufqueues[BQ_META], b_freelist) {
+               if (ISSET(bp->b_flags, B_DELWRI) || bp->b_bufsize != nsize)
+                       continue;
+               ptr = (void *)bp->b_datap;
+               bp->b_bufsize = 0;
+
+               bcleanbuf(bp, TRUE);
+               break;
+       }
+       lck_mtx_unlock(buf_mtxp);
+
+       return (ptr);
+}
+
+
+
+int zalloc_nopagewait_failed = 0;
+int recycle_buf_failed = 0;
+
+static void *
+grab_memory_for_meta_buf(int nsize)
+{
+       zone_t z;
+       void *ptr;
+       boolean_t was_vmpriv;
+
+       z = getbufzone(nsize);
+
+       /*
+        * make sure we're NOT priviliged so that
+        * if a vm_page_grab is needed, it won't
+        * block if we're out of free pages... if
+        * it blocks, then we can't honor the
+        * nopagewait request
+        */
+       was_vmpriv = set_vm_privilege(FALSE);
+
+       ptr = zalloc_nopagewait(z);
+
+       if (was_vmpriv == TRUE)
+               set_vm_privilege(TRUE);
+
+       if (ptr == NULL) {
+
+               zalloc_nopagewait_failed++;
+
+               ptr = recycle_buf_from_pool(nsize);
+
+               if (ptr == NULL) {
+
+                       recycle_buf_failed++;
+
+                       if (was_vmpriv == FALSE)
+                               set_vm_privilege(TRUE);
+
+                       ptr = zalloc(z);
+
+                       if (was_vmpriv == FALSE)
+                               set_vm_privilege(FALSE);
+               }
+       }
+       return (ptr);
+}
+
 /*
  * With UBC, there is no need to expand / shrink the file data 
  * buffer. The VM uses the same pages, hence no waste.
@@ -3052,7 +3336,6 @@ allocbuf(buf_t bp, int size)
                panic("allocbuf: buffer larger than MAXBSIZE requested");
 
        if (ISSET(bp->b_flags, B_META)) {
-               zone_t zprev, z;
                int    nsize = roundup(size, MINMETA);
 
                if (bp->b_datap) {
@@ -3060,14 +3343,16 @@ allocbuf(buf_t bp, int size)
 
                        if (ISSET(bp->b_flags, B_ZALLOC)) {
                                if (bp->b_bufsize < nsize) {
+                                       zone_t zprev;
+
                                        /* reallocate to a bigger size */
 
                                        zprev = getbufzone(bp->b_bufsize);
                                        if (nsize <= MAXMETA) {
                                                desired_size = nsize;
-                                               z = getbufzone(nsize);
+
                                                /* b_datap not really a ptr */
-                                               *(void **)(&bp->b_datap) = zalloc(z);
+                                               *(void **)(&bp->b_datap) = grab_memory_for_meta_buf(nsize);
                                        } else {
                                                bp->b_datap = (uintptr_t)NULL;
                                                kmem_alloc_kobject(kernel_map, (vm_offset_t *)&bp->b_datap, desired_size);
@@ -3094,9 +3379,9 @@ allocbuf(buf_t bp, int size)
                        /* new allocation */
                        if (nsize <= MAXMETA) {
                                desired_size = nsize;
-                               z = getbufzone(nsize);
+
                                /* b_datap not really a ptr */
-                               *(void **)(&bp->b_datap) = zalloc(z);
+                               *(void **)(&bp->b_datap) = grab_memory_for_meta_buf(nsize);
                                SET(bp->b_flags, B_ZALLOC);
                        } else
                                kmem_alloc_kobject(kernel_map, (vm_offset_t *)&bp->b_datap, desired_size);
@@ -3374,6 +3659,8 @@ bcleanbuf(buf_t bp, boolean_t discard)
        trace(TR_BRELSE, pack(bp->b_vp, bp->b_bufsize), bp->b_lblkno);
 
        buf_release_credentials(bp);
+       
+       bp->b_redundancy_flags = 0;
 
        /* If discarding, just move to the empty queue */
        if (discard) {
@@ -3409,9 +3696,6 @@ bcleanbuf(buf_t bp, boolean_t discard)
                bp->b_bcount = 0;
                bp->b_dirtyoff = bp->b_dirtyend = 0;
                bp->b_validoff = bp->b_validend = 0;
-#ifdef CONFIG_PROTECT
-               bp->b_cpentry = 0;
-#endif
                bzero(&bp->b_attr, sizeof(struct bufattr));
 
                lck_mtx_lock_spin(buf_mtxp);
@@ -3605,13 +3889,12 @@ buf_biowait(buf_t bp)
  * (for swap pager, that puts swap buffers on the free lists (!!!),
  * for the vn device, that puts malloc'd buffers on the free lists!)
  */
-extern struct timeval priority_IO_timestamp_for_root;
-extern int hard_throttle_on_root;
 
 void
 buf_biodone(buf_t bp)
 {
        mount_t mp;
+       struct bufattr *bap;
        
        KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 387)) | DBG_FUNC_START,
                     bp, bp->b_datap, bp->b_flags, 0, 0);
@@ -3619,9 +3902,7 @@ buf_biodone(buf_t bp)
        if (ISSET(bp->b_flags, B_DONE))
                panic("biodone already");
 
-       if (ISSET(bp->b_flags, B_ERROR)) {
-               fslog_io_error(bp);
-       }
+       bap = &bp->b_attr;
 
        if (bp->b_vp && bp->b_vp->v_mount) {
                mp = bp->b_vp->v_mount;
@@ -3636,8 +3917,9 @@ buf_biodone(buf_t bp)
                INCR_PENDING_IO(-(pending_io_t)buf_count(bp), mp->mnt_pending_read_size);
        }
 
-        if (kdebug_enable) {
-               int    code = DKIO_DONE;
+       if (kdebug_enable) {
+               int code    = DKIO_DONE;
+               int io_tier = GET_BUFATTR_IO_TIER(bap);
 
                if (bp->b_flags & B_READ)
                        code |= DKIO_READ;
@@ -3649,21 +3931,20 @@ buf_biodone(buf_t bp)
                else if (bp->b_flags & B_PAGEIO)
                        code |= DKIO_PAGING;
 
-               if (bp->b_flags & B_THROTTLED_IO)
+               if (io_tier != 0)
                        code |= DKIO_THROTTLE;
-               else if (bp->b_flags & B_PASSIVE)
+
+               code |= ((io_tier << DKIO_TIER_SHIFT) & DKIO_TIER_MASK);
+
+               if (bp->b_flags & B_PASSIVE)
                        code |= DKIO_PASSIVE;
 
-               KERNEL_DEBUG_CONSTANT(FSDBG_CODE(DBG_DKRW, code) | DBG_FUNC_NONE,
-                              bp, (uintptr_t)bp->b_vp,
-                                     bp->b_resid, bp->b_error, 0);
+               if (bap->ba_flags & BA_NOCACHE)
+                       code |= DKIO_NOCACHE;
+
+               KERNEL_DEBUG_CONSTANT_IST(KDEBUG_COMMON, FSDBG_CODE(DBG_DKRW, code) | DBG_FUNC_NONE,
+                                         buf_kernel_addrperm_addr(bp), (uintptr_t)VM_KERNEL_ADDRPERM(bp->b_vp), bp->b_resid, bp->b_error, 0);
         }
-       if ((bp->b_vp != NULLVP) &&
-           ((bp->b_flags & (B_IOSTREAMING | B_PAGEIO | B_READ)) == (B_PAGEIO | B_READ)) &&
-           (bp->b_vp->v_mount->mnt_kern_flag & MNTK_ROOTDEV)) {
-               microuptime(&priority_IO_timestamp_for_root);
-               hard_throttle_on_root = 0;
-       }
 
        /*
         * I/O was done, so don't believe
@@ -3671,8 +3952,11 @@ buf_biodone(buf_t bp)
         * and we need to reset the THROTTLED/PASSIVE
         * indicators
         */
-       CLR(bp->b_flags, (B_WASDIRTY | B_THROTTLED_IO | B_PASSIVE));
-       CLR(bp->b_attr.ba_flags, (BA_THROTTLED_IO));
+       CLR(bp->b_flags, (B_WASDIRTY | B_PASSIVE));
+       CLR(bap->ba_flags, (BA_META | BA_NOCACHE | BA_DELAYIDLESLEEP));
+
+       SET_BUFATTR_IO_TIER(bap, 0);
+
        DTRACE_IO1(done, buf_t, bp);
 
        if (!ISSET(bp->b_flags, B_READ) && !ISSET(bp->b_flags, B_RAW))
@@ -3748,6 +4032,18 @@ biodone_done:
                  (uintptr_t)bp, (uintptr_t)bp->b_datap, bp->b_flags, 0, 0);
 }
 
+/*
+ * Obfuscate buf pointers.
+ */
+vm_offset_t
+buf_kernel_addrperm_addr(void * addr)
+{
+       if ((vm_offset_t)addr == 0)
+               return 0;
+       else
+               return ((vm_offset_t)addr + buf_kernel_addrperm);
+}
+
 /*
  * Return a count of buffers on the "locked" queue.
  */
@@ -3769,6 +4065,7 @@ count_lock_queue(void)
 
 /*
  * Return a count of 'busy' buffers. Used at the time of shutdown.
+ * note: This is also called from the mach side in debug context in kdp.c
  */
 int
 count_busy_buffers(void)
@@ -3864,9 +4161,6 @@ alloc_io_buf(vnode_t vp, int priv)
        bp->b_bufsize = 0;
        bp->b_upl = NULL;
        bp->b_vp = vp;
-#ifdef CONFIG_PROTECT
-       bp->b_cpentry = 0;
-#endif
        bzero(&bp->b_attr, sizeof(struct bufattr));
 
        if (vp && (vp->v_type == VBLK || vp->v_type == VCHR))
@@ -3889,6 +4183,9 @@ free_io_buf(buf_t bp)
        bp->b_vp = NULL;
        bp->b_flags = B_INVAL;
 
+       /* Zero out the bufattr and its flags before relinquishing this iobuf */
+       bzero (&bp->b_attr, sizeof(struct bufattr));
+       
        lck_mtx_lock_spin(iobuffer_mtxp);
 
        binsheadfree(bp, &iobufqueue, -1);
@@ -4085,7 +4382,7 @@ buffer_cache_gc(int all)
        boolean_t did_large_zfree = FALSE;
        boolean_t need_wakeup = FALSE;
        int now = buf_timestamp();
-       uint32_t found = 0, total_found = 0;
+       uint32_t found = 0;
        struct bqueues privq;
        int thresh_hold = BUF_STALE_THRESHHOLD;
 
@@ -4093,11 +4390,15 @@ buffer_cache_gc(int all)
                thresh_hold = 0;
        /* 
         * We only care about metadata (incore storage comes from zalloc()).
-        * No more than 1024 buffers total, and only those not accessed within the
-        * last 30s.  We will also only examine 128 buffers during a single grab
-        * of the lock in order to limit lock hold time.
+        * Unless "all" is set (used to evict meta data buffers in preparation
+        * for deep sleep), we only evict up to BUF_MAX_GC_BATCH_SIZE buffers
+        * that have not been accessed in the last BUF_STALE_THRESHOLD seconds.
+        * BUF_MAX_GC_BATCH_SIZE controls both the hold time of the global lock
+        * "buf_mtxp" and the length of time we spend compute bound in the GC
+        * thread which calls this function
         */
        lck_mtx_lock(buf_mtxp);
+
        do {
                found = 0;
                TAILQ_INIT(&privq);
@@ -4179,7 +4480,6 @@ buffer_cache_gc(int all)
                        bp->b_whichq = BQ_EMPTY;
                        BLISTNONE(bp);
                }
-
                lck_mtx_lock(buf_mtxp);
 
                /* Back under lock, move them all to invalid hash and clear busy */
@@ -4199,9 +4499,8 @@ buffer_cache_gc(int all)
 
                /* And do a big bulk move to the empty queue */
                TAILQ_CONCAT(&bufqueues[BQ_EMPTY], &privq, b_freelist);
-               total_found += found;
 
-       } while ((all || (total_found < BUF_MAX_GC_COUNT)) && (found == BUF_MAX_GC_BATCH_SIZE));
+       } while (all && (found == BUF_MAX_GC_BATCH_SIZE));
 
        lck_mtx_unlock(buf_mtxp);
 
@@ -4296,334 +4595,3 @@ bflushq(int whichq, mount_t mp)
        return (total_writes);
 }
 #endif
-
-
-#if BALANCE_QUEUES
-
-/* XXX move this to a separate file */
-
-/*
- * NOTE: THIS CODE HAS NOT BEEN UPDATED
- * WITH RESPECT TO THE NEW LOCKING MODEL
- */
-   
-
-/*
- * Dynamic Scaling of the Buffer Queues
- */
-
-typedef long long blsize_t;
-
-blsize_t MAXNBUF; /* initialize to (sane_size / PAGE_SIZE) */
-/* Global tunable limits */
-blsize_t nbufh;                        /* number of buffer headers */
-blsize_t nbuflow;              /* minimum number of buffer headers required */
-blsize_t nbufhigh;             /* maximum number of buffer headers allowed */
-blsize_t nbuftarget;   /* preferred number of buffer headers */
-
-/*
- * assertions:
- *
- * 1.  0 < nbuflow <= nbufh <= nbufhigh
- * 2.  nbufhigh <= MAXNBUF
- * 3.  0 < nbuflow <= nbuftarget <= nbufhigh
- * 4.  nbufh can not be set by sysctl().
- */
-
-/* Per queue tunable limits */
-
-struct bufqlim {
-       blsize_t        bl_nlow;        /* minimum number of buffer headers required */
-       blsize_t        bl_num;         /* number of buffer headers on the queue */
-       blsize_t        bl_nlhigh;      /* maximum number of buffer headers allowed */
-       blsize_t        bl_target;      /* preferred number of buffer headers */
-       long    bl_stale;       /* Seconds after which a buffer is considered stale */
-} bufqlim[BQUEUES];
-
-/*
- * assertions:
- *
- * 1.  0 <= bl_nlow <= bl_num <= bl_nlhigh
- * 2.  bl_nlhigh <= MAXNBUF
- * 3.  bufqlim[BQ_META].bl_nlow != 0
- * 4.  bufqlim[BQ_META].bl_nlow > (number of possible concurrent 
- *                                                                     file system IO operations)
- * 5.  bl_num can not be set by sysctl().
- * 6.  bl_nhigh <= nbufhigh
- */
-
-/*
- * Rationale:
- * ----------
- * Defining it blsize_t as long permits 2^31 buffer headers per queue.
- * Which can describe (2^31 * PAGE_SIZE) memory per queue.
- * 
- * These limits are exported to by means of sysctl().
- * It was decided to define blsize_t as a 64 bit quantity.
- * This will make sure that we will not be required to change it
- * as long as we do not exceed 64 bit address space for the kernel.
- * 
- * low and high numbers parameters initialized at compile time
- * and boot arguments can be used to override them. sysctl() 
- * would not change the value. sysctl() can get all the values 
- * but can set only target. num is the current level.
- *
- * Advantages of having a "bufqscan" thread doing the balancing are, 
- * Keep enough bufs on BQ_EMPTY.
- *     getnewbuf() by default will always select a buffer from the BQ_EMPTY.
- *             getnewbuf() perfoms best if a buffer was found there.
- *             Also this minimizes the possibility of starting IO
- *             from getnewbuf(). That's a performance win, too.
- *
- *     Localize complex logic [balancing as well as time aging]
- *             to balancebufq().
- *
- *     Simplify getnewbuf() logic by elimination of time aging code.
- */
-
-/* 
- * Algorithm:
- * -----------
- * The goal of the dynamic scaling of the buffer queues to to keep
- * the size of the LRU close to bl_target. Buffers on a queue would
- * be time aged.
- *
- * There would be a thread which will be responsible for "balancing"
- * the buffer cache queues.
- *
- * The scan order would be:    AGE, LRU, META, EMPTY.
- */
-
-long bufqscanwait = 0;
-
-static void bufqscan_thread();
-static int balancebufq(int q);
-static int btrimempty(int n);
-static __inline__ int initbufqscan(void);
-static __inline__ int nextbufq(int q);
-static void buqlimprt(int all);
-
-
-static __inline__ void
-bufqinc(int q)
-{
-       if ((q < 0) || (q >= BQUEUES))
-               return;
-
-       bufqlim[q].bl_num++;
-       return;
-}
-
-static __inline__ void
-bufqdec(int q)
-{
-       if ((q < 0) || (q >= BQUEUES))
-               return; 
-
-       bufqlim[q].bl_num--;
-       return;
-}
-
-static void
-bufq_balance_thread_init(void)
-{
-       thread_t        thread = THREAD_NULL;
-
-       if (bufqscanwait++ == 0) {
-
-               /* Initalize globals */
-               MAXNBUF = (sane_size / PAGE_SIZE);
-               nbufh = nbuf_headers;
-               nbuflow = min(nbufh, 100);
-               nbufhigh = min(MAXNBUF, max(nbufh, 2048));
-               nbuftarget = (sane_size >> 5) / PAGE_SIZE;
-               nbuftarget = max(nbuflow, nbuftarget);
-               nbuftarget = min(nbufhigh, nbuftarget);
-
-               /*
-                * Initialize the bufqlim 
-                */ 
-
-               /* LOCKED queue */
-               bufqlim[BQ_LOCKED].bl_nlow = 0;
-               bufqlim[BQ_LOCKED].bl_nlhigh = 32;
-               bufqlim[BQ_LOCKED].bl_target = 0;
-               bufqlim[BQ_LOCKED].bl_stale = 30;
-
-               /* LRU queue */
-               bufqlim[BQ_LRU].bl_nlow = 0;
-               bufqlim[BQ_LRU].bl_nlhigh = nbufhigh/4;
-               bufqlim[BQ_LRU].bl_target = nbuftarget/4;
-               bufqlim[BQ_LRU].bl_stale = LRU_IS_STALE;
-
-               /* AGE queue */
-               bufqlim[BQ_AGE].bl_nlow = 0;
-               bufqlim[BQ_AGE].bl_nlhigh = nbufhigh/4;
-               bufqlim[BQ_AGE].bl_target = nbuftarget/4;
-               bufqlim[BQ_AGE].bl_stale = AGE_IS_STALE;
-
-               /* EMPTY queue */
-               bufqlim[BQ_EMPTY].bl_nlow = 0;
-               bufqlim[BQ_EMPTY].bl_nlhigh = nbufhigh/4;
-               bufqlim[BQ_EMPTY].bl_target = nbuftarget/4;
-               bufqlim[BQ_EMPTY].bl_stale = 600000;
-
-               /* META queue */
-               bufqlim[BQ_META].bl_nlow = 0;
-               bufqlim[BQ_META].bl_nlhigh = nbufhigh/4;
-               bufqlim[BQ_META].bl_target = nbuftarget/4;
-               bufqlim[BQ_META].bl_stale = META_IS_STALE;
-
-               /* LAUNDRY queue */
-               bufqlim[BQ_LOCKED].bl_nlow = 0;
-               bufqlim[BQ_LOCKED].bl_nlhigh = 32;
-               bufqlim[BQ_LOCKED].bl_target = 0;
-               bufqlim[BQ_LOCKED].bl_stale = 30;
-
-               buqlimprt(1);
-       }
-
-       /* create worker thread */
-       kernel_thread_start((thread_continue_t)bufqscan_thread, NULL, &thread);
-       thread_deallocate(thread);
-}
-
-/* The workloop for the buffer balancing thread */
-static void
-bufqscan_thread()
-{
-       int moretodo = 0;
-
-       for(;;) {
-               do {
-                       int q;  /* buffer queue to process */
-               
-                       q = initbufqscan();
-                       for (; q; ) {
-                               moretodo |= balancebufq(q);
-                               q = nextbufq(q);
-                       }
-               } while (moretodo);
-
-#if DIAGNOSTIC
-               vfs_bufstats();
-               buqlimprt(0);
-#endif
-               (void)tsleep((void *)&bufqscanwait, PRIBIO, "bufqscanwait", 60 * hz);
-               moretodo = 0;
-       }
-}
-
-/* Seed for the buffer queue balancing */
-static __inline__ int
-initbufqscan()
-{
-       /* Start with AGE queue */
-       return (BQ_AGE);
-}
-
-/* Pick next buffer queue to balance */
-static __inline__ int
-nextbufq(int q)
-{
-       int order[] = { BQ_AGE, BQ_LRU, BQ_META, BQ_EMPTY, 0 };
-       
-       q++;
-       q %= sizeof(order);
-       return (order[q]);
-}
-
-/* function to balance the buffer queues */
-static int
-balancebufq(int q)
-{
-       int moretodo = 0;
-       int n, t;
-       
-       /* reject invalid q */
-       if ((q < 0) || (q >= BQUEUES))
-               goto out;
-
-       /* LOCKED or LAUNDRY queue MUST not be balanced */
-       if ((q == BQ_LOCKED) || (q == BQ_LAUNDRY))
-               goto out;
-
-       n = (bufqlim[q].bl_num - bufqlim[q].bl_target);
-
-       /* If queue has less than target nothing more to do */
-       if (n < 0)
-               goto out;
-
-       if ( n > 8 ) {
-               /* Balance only a small amount (12.5%) at a time */
-               n >>= 3;
-       }
-
-       /* EMPTY queue needs special handling */
-       if (q == BQ_EMPTY) {
-               moretodo |= btrimempty(n);
-               goto out;
-       }
-
-       t = buf_timestamp():
-       
-       for (; n > 0; n--) {
-               struct buf *bp = bufqueues[q].tqh_first;
-               if (!bp)
-                       break;
-               
-               /* check if it's stale */
-               if ((t - bp->b_timestamp) > bufqlim[q].bl_stale) {
-                       if (bcleanbuf(bp, FALSE)) {
-                               /* buf_bawrite() issued, bp not ready */
-                               moretodo = 1;
-                       } else {
-                               /* release the cleaned buffer to BQ_EMPTY */
-                               SET(bp->b_flags, B_INVAL);
-                               buf_brelse(bp);
-                       }
-               } else
-                       break;          
-       }
-
-out:
-       return (moretodo);              
-}
-
-static int
-btrimempty(int n)
-{
-       /*
-        * When struct buf are allocated dynamically, this would
-        * reclaim upto 'n' struct buf from the empty queue.
-        */
-        
-        return (0);
-}
-
-static void
-buqlimprt(int all)
-{
-       int i;
-    static char *bname[BQUEUES] =
-               { "LOCKED", "LRU", "AGE", "EMPTY", "META", "LAUNDRY" };
-
-       if (all)
-               for (i = 0; i < BQUEUES; i++) {
-                       printf("%s : ", bname[i]);
-                       printf("min = %ld, ", (long)bufqlim[i].bl_nlow);
-                       printf("cur = %ld, ", (long)bufqlim[i].bl_num);
-                       printf("max = %ld, ", (long)bufqlim[i].bl_nlhigh);
-                       printf("target = %ld, ", (long)bufqlim[i].bl_target);
-                       printf("stale after %ld seconds\n", bufqlim[i].bl_stale);
-               }
-       else
-               for (i = 0; i < BQUEUES; i++) {
-                       printf("%s : ", bname[i]);
-                       printf("cur = %ld, ", (long)bufqlim[i].bl_num);
-               }
-}
-
-#endif
-
-