xnu-3247.1.106.tar.gz

[apple/xnu.git] / bsd / sys / buf_internal.h

diff --git a/bsd/sys/buf_internal.h b/bsd/sys/buf_internal.h
index 7282821467306ee22c30f4f093c5cac2d12840d2..6ff3284bc2b7256550522183c48854b612e1780b 100644 (file)
--- a/bsd/sys/buf_internal.h
+++ b/bsd/sys/buf_internal.h
@@ -1,5 +1,5 @@
 /*
 /*

- * Copyright (c) 2000-2002 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2000-2014 Apple Inc. All rights reserved.

  *
  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  * 
  *
  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  * 


@@ -79,9 +79,23 @@
 #include <sys/buf.h>
 #include <sys/lock.h>
 
 #include <sys/buf.h>
 #include <sys/lock.h>
 

+#if CONFIG_PROTECT
+#include <sys/cprotect.h>
+#endif

 
 #define NOLIST ((struct buf *)0x87654321)
 
 
 #define NOLIST ((struct buf *)0x87654321)
 

+/*
+ * Attributes of an I/O to be used by lower layers
+ */
+struct bufattr {
+#if CONFIG_PROTECT
+       struct cpx *ba_cpx;
+       uint64_t ba_cp_file_off;
+#endif
+       uint64_t ba_flags;      /* flags. Some are only in-use on embedded devices */
+};
+

 /*
  * The buffer header describes an I/O operation in the kernel.
  */
 /*
  * The buffer header describes an I/O operation in the kernel.
  */


@@ -90,66 +104,117 @@ struct buf {
        LIST_ENTRY(buf) b_vnbufs;       /* Buffer's associated vnode. */
        TAILQ_ENTRY(buf) b_freelist;    /* Free list position if not active. */
        int     b_timestamp;            /* timestamp for queuing operation */
        LIST_ENTRY(buf) b_vnbufs;       /* Buffer's associated vnode. */
        TAILQ_ENTRY(buf) b_freelist;    /* Free list position if not active. */
        int     b_timestamp;            /* timestamp for queuing operation */

-       long    b_whichq;               /* the free list the buffer belongs to */
-       volatile long   b_flags;        /* B_* flags. */
-       volatile long   b_lflags;       /* BL_BUSY | BL_WANTED flags... protected by buf_mtx */
+       int     b_whichq;               /* the free list the buffer belongs to */
+       volatile uint32_t       b_flags;        /* B_* flags. */
+       volatile uint32_t       b_lflags;       /* BL_BUSY | BL_WANTED flags... protected by buf_mtx */

        int     b_error;                /* errno value. */
        int     b_error;                /* errno value. */

-       long    b_bufsize;              /* Allocated buffer size. */
-       long    b_bcount;               /* Valid bytes in buffer. */
-       long    b_resid;                /* Remaining I/O. */
+       int     b_bufsize;              /* Allocated buffer size. */
+       int     b_bcount;               /* Valid bytes in buffer. */
+       int     b_resid;                /* Remaining I/O. */

        dev_t   b_dev;                  /* Device associated with buffer. */
        dev_t   b_dev;                  /* Device associated with buffer. */

-        uintptr_t      b_datap;        /* Memory, superblocks, indirect etc.*/
+       uintptr_t       b_datap;        /* Memory, superblocks, indirect etc.*/

        daddr64_t       b_lblkno;       /* Logical block number. */
        daddr64_t       b_blkno;        /* Underlying physical block number. */
        void    (*b_iodone)(buf_t, void *);     /* Function to call upon completion. */
        daddr64_t       b_lblkno;       /* Logical block number. */
        daddr64_t       b_blkno;        /* Underlying physical block number. */
        void    (*b_iodone)(buf_t, void *);     /* Function to call upon completion. */

-       vnode_t b_vp;                   /* Device vnode. */
-       struct  ucred *b_rcred;         /* Read credentials reference. */
-       struct  ucred *b_wcred;         /* Write credentials reference. */
+       vnode_t b_vp;                   /* File vnode for data, device vnode for metadata. */
+       kauth_cred_t b_rcred;           /* Read credentials reference. */
+       kauth_cred_t b_wcred;           /* Write credentials reference. */

        void *  b_upl;                  /* Pointer to UPL */
        buf_t   b_real_bp;              /* used to track bp generated through cluster_bp */
        TAILQ_ENTRY(buf)        b_act;  /* Device driver queue when active */
        void *  b_drvdata;              /* Device driver private use */
        void *  b_upl;                  /* Pointer to UPL */
        buf_t   b_real_bp;              /* used to track bp generated through cluster_bp */
        TAILQ_ENTRY(buf)        b_act;  /* Device driver queue when active */
        void *  b_drvdata;              /* Device driver private use */

-        void *  b_fsprivate;           /* filesystem private use */
+       void *  b_fsprivate;            /* filesystem private use */

        void *  b_transaction;          /* journal private use */
        int     b_dirtyoff;             /* Offset in buffer of dirty region. */
        int     b_dirtyend;             /* Offset of end of dirty region. */
        int     b_validoff;             /* Offset in buffer of valid region. */
        int     b_validend;             /* Offset of end of valid region. */
        void *  b_transaction;          /* journal private use */
        int     b_dirtyoff;             /* Offset in buffer of dirty region. */
        int     b_dirtyend;             /* Offset of end of dirty region. */
        int     b_validoff;             /* Offset in buffer of valid region. */
        int     b_validend;             /* Offset of end of valid region. */

+
+       /* store extra information related to redundancy of data, such as
+        * which redundancy copy to use, etc
+        */
+       uint32_t b_redundancy_flags;
+

        proc_t  b_proc;                 /* Associated proc; NULL if kernel. */
        proc_t  b_proc;                 /* Associated proc; NULL if kernel. */

+#ifdef BUF_MAKE_PRIVATE
+       buf_t   b_data_store;
+#endif
+       struct bufattr b_attr;

 #ifdef JOE_DEBUG
         void * b_owner;
         int     b_tag;
         void *  b_lastbrelse;
 #ifdef JOE_DEBUG
         void * b_owner;
         int     b_tag;
         void *  b_lastbrelse;

-        int    b_stackbrelse[6];
-        int    b_stackgetblk[6];
+        void * b_stackbrelse[6];
+        void * b_stackgetblk[6];

 #endif
 };
 
 #endif
 };
 

+extern vm_offset_t buf_kernel_addrperm;

 
 /* cluster_io definitions for use with io bufs */
 #define b_uploffset  b_bufsize
 
 /* cluster_io definitions for use with io bufs */
 #define b_uploffset  b_bufsize

+#define b_orig      b_freelist.tqe_prev
+#define b_shadow     b_freelist.tqe_next
+#define        b_shadow_ref b_validoff
+#ifdef BUF_MAKE_PRIVATE
+#define b_data_ref   b_validend
+#endif

 #define b_trans_head b_freelist.tqe_prev
 #define b_trans_next b_freelist.tqe_next
 #define b_iostate    b_rcred
 #define b_trans_head b_freelist.tqe_prev
 #define b_trans_next b_freelist.tqe_next
 #define b_iostate    b_rcred

+#define b_cliodone   b_wcred

 
 /*
 
 /*

- * These flags are kept in b_lflags... 
+ * These flags are kept in b_lflags...

  * buf_mtxp must be held before examining/updating
  */
 #define        BL_BUSY         0x00000001      /* I/O in progress. */
 #define        BL_WANTED       0x00000002      /* Process wants this buffer. */
 #define BL_IOBUF       0x00000004      /* buffer allocated via 'buf_alloc' */
  * buf_mtxp must be held before examining/updating
  */
 #define        BL_BUSY         0x00000001      /* I/O in progress. */
 #define        BL_WANTED       0x00000002      /* Process wants this buffer. */
 #define BL_IOBUF       0x00000004      /* buffer allocated via 'buf_alloc' */

+#define BL_WANTDEALLOC 0x00000010      /* buffer should be put on empty list when clean */
+#define BL_SHADOW      0x00000020
+#define BL_EXTERNAL    0x00000040
+#define BL_WAITSHADOW  0x00000080
+#define BL_IOBUF_ALLOC 0x00000100
+#define BL_WANTED_REF  0x00000200

 
 

+/*
+ * Parameters for buffer cache garbage collection 
+ */
+#define BUF_STALE_THRESHHOLD   30      /* Collect if untouched in the last 30 seconds */
+#define BUF_MAX_GC_BATCH_SIZE  64      /* Under a single grab of the lock */

 
 /*
  * mask used by buf_flags... these are the readable external flags
  */
 
 /*
  * mask used by buf_flags... these are the readable external flags
  */

-#define BUF_X_RDFLAGS (B_CLUSTER | B_PHYS | B_LOCKED | B_DELWRI | B_ASYNC |\
-                       B_READ | B_WRITE | B_META | B_PAGEIO)
+#define BUF_X_RDFLAGS (B_PHYS | B_RAW | B_LOCKED | B_ASYNC | B_READ | B_WRITE | B_PAGEIO |\
+                      B_META | B_CLUSTER | B_DELWRI | B_FUA | B_PASSIVE | B_IOSTREAMING |\
+                      B_ENCRYPTED_IO | B_STATICCONTENT)

 /*
  * mask used by buf_clearflags/buf_setflags... these are the writable external flags
  */
 /*
  * mask used by buf_clearflags/buf_setflags... these are the writable external flags
  */

-#define BUF_X_WRFLAGS (B_LOCKED | B_NOCACHE | B_ASYNC | B_READ | B_WRITE | B_PAGEIO)
+#define BUF_X_WRFLAGS (B_PHYS | B_RAW | B_LOCKED | B_ASYNC | B_READ | B_WRITE | B_PAGEIO |\
+                      B_NOCACHE | B_FUA | B_PASSIVE | B_IOSTREAMING)
+
+#if 0
+/* b_flags defined in buf.h */
+#define        B_WRITE         0x00000000      /* Write buffer (pseudo flag). */
+#define        B_READ          0x00000001      /* Read buffer. */
+#define        B_ASYNC         0x00000002      /* Start I/O, do not wait. */
+#define        B_NOCACHE       0x00000004      /* Do not cache block after use. */
+#define        B_DELWRI        0x00000008      /* Delay I/O until buffer reused. */
+#define        B_LOCKED        0x00000010      /* Locked in core (not reusable). */
+#define        B_PHYS          0x00000020      /* I/O to user memory. */
+#define B_CLUSTER      0x00000040      /* UPL based I/O generated by cluster layer */
+#define        B_PAGEIO        0x00000080      /* Page in/out */
+#define        B_META          0x00000100      /* buffer contains meta-data. */
+#define        B_RAW           0x00000200      /* Set by physio for raw transfers. */
+#define        B_FUA           0x00000400      /* Write-through disk cache(if supported) */
+#define B_PASSIVE      0x00000800      /* PASSIVE I/Os are ignored by THROTTLE I/O */
+#define        B_IOSTREAMING   0x00001000      /* sequential access pattern detected */
+#define B_ENCRYPTED_IO 0x00004000      /* Encrypted I/O */
+#define B_STATICCONTENT 0x00008000     /* Buffer is likely to remain unaltered */
+#endif

 
 /*
  * These flags are kept in b_flags... access is lockless
 
 /*
  * These flags are kept in b_flags... access is lockless


@@ -166,7 +231,7 @@ struct buf {
 #define        B_AGE           0x00200000      /* Move to age queue when I/O done. */
 #define B_FILTER       0x00400000      /* call b_iodone from biodone as an in-line filter */
 #define        B_CALL          0x00800000      /* Call b_iodone from biodone, assumes b_iodone consumes bp */
 #define        B_AGE           0x00200000      /* Move to age queue when I/O done. */
 #define B_FILTER       0x00400000      /* call b_iodone from biodone as an in-line filter */
 #define        B_CALL          0x00800000      /* Call b_iodone from biodone, assumes b_iodone consumes bp */

-#define        B_RAW           0x01000000      /* Set by physio for raw transfers. */
+#define B_EOT          0x01000000      /* last buffer in a transaction list created by cluster_io */

 #define        B_WASDIRTY      0x02000000      /* page was found dirty in the VM cache */
 #define        B_HDRALLOC      0x04000000      /* zone allocated buffer header */
 #define        B_ZALLOC        0x08000000      /* b_datap is zalloc()ed */
 #define        B_WASDIRTY      0x02000000      /* page was found dirty in the VM cache */
 #define        B_HDRALLOC      0x04000000      /* zone allocated buffer header */
 #define        B_ZALLOC        0x08000000      /* b_datap is zalloc()ed */


@@ -177,25 +242,50 @@ struct buf {
 /*
  * private flags used by by the cluster layer
  */
 /*
  * private flags used by by the cluster layer
  */

-#define B_NEED_IODONE   0x20000000     /* need biodone on the real_bp associated with a cluster_io */
+#define B_TWANTED      0x20000000      /* but_t that is part of a cluster level transaction is wanted */

 #define B_COMMIT_UPL    0x40000000     /* commit/abort the UPL on I/O success/failure */
 #define B_COMMIT_UPL    0x40000000     /* commit/abort the UPL on I/O success/failure */

-/*
- * can we deprecate?
- */
-#define        B_TAPE          0x80000000      /* Magnetic tape I/O. */
-
+#define B_TDONE                0x80000000      /* buf_t that is part of a cluster level transaction has completed */

 
 /* Flags to low-level allocation routines. */
 #define B_CLRBUF       0x01    /* Request allocated buffer be cleared. */
 #define B_SYNC         0x02    /* Do all allocations synchronously. */
 #define B_NOBUFF       0x04    /* Do not allocate struct buf */
 
 
 /* Flags to low-level allocation routines. */
 #define B_CLRBUF       0x01    /* Request allocated buffer be cleared. */
 #define B_SYNC         0x02    /* Do all allocations synchronously. */
 #define B_NOBUFF       0x04    /* Do not allocate struct buf */
 

+/*
+ * ba_flags (Buffer Attribute flags)
+ * Some of these may be in-use only on embedded devices.
+ */
+#define BA_RAW_ENCRYPTED_IO     0x00000001
+#define BA_THROTTLED_IO         0x00000002
+#define BA_DELAYIDLESLEEP       0x00000004     /* Process is marked to delay idle sleep on disk IO */
+#define BA_NOCACHE             0x00000008
+#define BA_META                        0x00000010
+#define BA_GREEDY_MODE         0x00000020      /* High speed writes that consume more storage */
+#define BA_QUICK_COMPLETE      0x00000040      /* Request quick completion at expense of storage efficiency */
+#define BA_PASSIVE             0x00000080
+
+/*
+ * Note: IO_TIERs consume 0x0100, 0x0200, 0x0400, 0x0800
+ * These are now in-use by the I/O tiering system.
+ */ 
+#define BA_IO_TIER_MASK                0x00000f00
+#define BA_IO_TIER_SHIFT       8
+
+#define BA_ISOCHRONOUS         0x00001000 /* device specific isochronous throughput to media */
+
+
+#define GET_BUFATTR_IO_TIER(bap)       ((bap->ba_flags & BA_IO_TIER_MASK) >> BA_IO_TIER_SHIFT)
+#define SET_BUFATTR_IO_TIER(bap, tier)                                         \
+do {                                                                           \
+       (bap)->ba_flags &= (~BA_IO_TIER_MASK);                                  \
+       (bap)->ba_flags |= (((tier) << BA_IO_TIER_SHIFT) & BA_IO_TIER_MASK);    \
+} while(0)

 
 

-extern int niobuf;             /* The number of IO buffer headers for cluster IO */
-extern int nbuf;               /* The number of buffer headers */
-extern struct buf *buf;                /* The buffer headers. */
+extern int niobuf_headers;             /* The number of IO buffer headers for cluster IO */
+extern int nbuf_headers;               /* The number of buffer headers */

 extern int max_nbuf_headers;           /* The max number of buffer headers */
 extern int nbuf_hashelements;          /* The number of elements in bufhash */
 extern int max_nbuf_headers;           /* The max number of buffer headers */
 extern int nbuf_hashelements;          /* The number of elements in bufhash */

+extern struct buf *buf_headers;                /* The buffer headers. */

 
 
 /*
 
 
 /*


@@ -213,14 +303,14 @@ extern int nbuf_hashelements;             /* The number of elements in bufhash */
 
 __BEGIN_DECLS
 
 
 __BEGIN_DECLS
 

+buf_t  buf_create_shadow_priv(buf_t bp, boolean_t force_copy, uintptr_t external_storage, void (*iodone)(buf_t, void *), void *arg);
+

 buf_t  alloc_io_buf(vnode_t, int);
 void   free_io_buf(buf_t);
 
 int    allocbuf(struct buf *, int);
 void   bufinit(void);
 
 buf_t  alloc_io_buf(vnode_t, int);
 void   free_io_buf(buf_t);
 
 int    allocbuf(struct buf *, int);
 void   bufinit(void);
 

-void   buf_setfilter(buf_t, void (*)(buf_t, void *), void *, void **, void **);
-

 /*
  * Flags for buf_acquire
  */
 /*
  * Flags for buf_acquire
  */


@@ -229,6 +319,9 @@ void        buf_setfilter(buf_t, void (*)(buf_t, void *), void *, void **, void **);
 #define BAC_SKIP_NONLOCKED     0x04    /* Don't return LOCKED buffers */
 #define BAC_SKIP_LOCKED                0x08    /* Only return LOCKED buffers */
 
 #define BAC_SKIP_NONLOCKED     0x04    /* Don't return LOCKED buffers */
 #define BAC_SKIP_LOCKED                0x08    /* Only return LOCKED buffers */
 

+void   buf_list_lock(void);
+void   buf_list_unlock(void);
+

 void   cluster_init(void);
 void   buf_drop(buf_t);
 errno_t        buf_acquire(buf_t, int, int, int);
 void   cluster_init(void);
 void   buf_drop(buf_t);
 errno_t        buf_acquire(buf_t, int, int, int);


@@ -236,6 +329,16 @@ errno_t    buf_acquire(buf_t, int, int, int);
 int    count_busy_buffers(void);
 int    count_lock_queue(void);
 
 int    count_busy_buffers(void);
 int    count_lock_queue(void);
 

+int buf_flushdirtyblks_skipinfo (vnode_t, int, int, const char *);
+void buf_wait_for_shadow_io (vnode_t, daddr64_t);
+
+#ifdef BUF_MAKE_PRIVATE
+errno_t        buf_make_private(buf_t bp);
+#endif
+
+#ifdef CONFIG_PROTECT
+void buf_setcpoff (buf_t, uint64_t);
+#endif

 
 __END_DECLS
 
 
 __END_DECLS