Index: head/sys/kern/vfs_bio.c =================================================================== --- head/sys/kern/vfs_bio.c (revision 288430) +++ head/sys/kern/vfs_bio.c (revision 288431) @@ -1,4597 +1,4611 @@ /*- * Copyright (c) 2004 Poul-Henning Kamp * Copyright (c) 1994,1997 John S. Dyson * Copyright (c) 2013 The FreeBSD Foundation * All rights reserved. * * Portions of this software were developed by Konstantin Belousov * under sponsorship from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /* * this file contains a new buffer I/O scheme implementing a coherent * VM object and buffer cache scheme. Pains have been taken to make * sure that the performance degradation associated with schemes such * as this is not realized. * * Author: John S. Dyson * Significant help during the development and debugging phases * had been provided by David Greenman, also of the FreeBSD core team. * * see man buf(9) for more info. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "opt_compat.h" #include "opt_swap.h" static MALLOC_DEFINE(M_BIOBUF, "biobuf", "BIO buffer"); struct bio_ops bioops; /* I/O operation notification */ struct buf_ops buf_ops_bio = { .bop_name = "buf_ops_bio", .bop_write = bufwrite, .bop_strategy = bufstrategy, .bop_sync = bufsync, .bop_bdflush = bufbdflush, }; static struct buf *buf; /* buffer header pool */ extern struct buf *swbuf; /* Swap buffer header pool. */ caddr_t unmapped_buf; /* Used below and for softdep flushing threads in ufs/ffs/ffs_softdep.c */ struct proc *bufdaemonproc; static int inmem(struct vnode *vp, daddr_t blkno); static void vm_hold_free_pages(struct buf *bp, int newbsize); static void vm_hold_load_pages(struct buf *bp, vm_offset_t from, vm_offset_t to); static void vfs_page_set_valid(struct buf *bp, vm_ooffset_t off, vm_page_t m); static void vfs_page_set_validclean(struct buf *bp, vm_ooffset_t off, vm_page_t m); static void vfs_clean_pages_dirty_buf(struct buf *bp); static void vfs_setdirty_locked_object(struct buf *bp); static void vfs_vmio_invalidate(struct buf *bp); static void vfs_vmio_truncate(struct buf *bp, int npages); static void vfs_vmio_extend(struct buf *bp, int npages, int size); static int vfs_bio_clcheck(struct vnode *vp, int size, daddr_t lblkno, daddr_t blkno); static int buf_flush(struct vnode *vp, int); static int flushbufqueues(struct vnode *, int, int); static void buf_daemon(void); static void bremfreel(struct buf *bp); static __inline void bd_wakeup(void); static int sysctl_runningspace(SYSCTL_HANDLER_ARGS); #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) static int sysctl_bufspace(SYSCTL_HANDLER_ARGS); #endif int vmiodirenable = TRUE; SYSCTL_INT(_vfs, OID_AUTO, vmiodirenable, CTLFLAG_RW, &vmiodirenable, 0, "Use the VM system for directory writes"); long runningbufspace; SYSCTL_LONG(_vfs, OID_AUTO, runningbufspace, CTLFLAG_RD, &runningbufspace, 0, "Amount of presently outstanding async buffer io"); static long bufspace; #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) SYSCTL_PROC(_vfs, OID_AUTO, bufspace, CTLTYPE_LONG|CTLFLAG_MPSAFE|CTLFLAG_RD, &bufspace, 0, sysctl_bufspace, "L", "Virtual memory used for buffers"); #else SYSCTL_LONG(_vfs, OID_AUTO, bufspace, CTLFLAG_RD, &bufspace, 0, "Physical memory used for buffers"); #endif static long bufkvaspace; SYSCTL_LONG(_vfs, OID_AUTO, bufkvaspace, CTLFLAG_RD, &bufkvaspace, 0, "Kernel virtual memory used for buffers"); static long maxbufspace; SYSCTL_LONG(_vfs, OID_AUTO, maxbufspace, CTLFLAG_RD, &maxbufspace, 0, "Maximum allowed value of bufspace (including buf_daemon)"); static long bufmallocspace; SYSCTL_LONG(_vfs, OID_AUTO, bufmallocspace, CTLFLAG_RD, &bufmallocspace, 0, "Amount of malloced memory for buffers"); static long maxbufmallocspace; SYSCTL_LONG(_vfs, OID_AUTO, maxmallocbufspace, CTLFLAG_RW, &maxbufmallocspace, 0, "Maximum amount of malloced memory for buffers"); static long lobufspace; SYSCTL_LONG(_vfs, OID_AUTO, lobufspace, CTLFLAG_RD, &lobufspace, 0, "Minimum amount of buffers we want to have"); long hibufspace; SYSCTL_LONG(_vfs, OID_AUTO, hibufspace, CTLFLAG_RD, &hibufspace, 0, "Maximum allowed value of bufspace (excluding buf_daemon)"); static int bufreusecnt; SYSCTL_INT(_vfs, OID_AUTO, bufreusecnt, CTLFLAG_RW, &bufreusecnt, 0, "Number of times we have reused a buffer"); static int buffreekvacnt; SYSCTL_INT(_vfs, OID_AUTO, buffreekvacnt, CTLFLAG_RW, &buffreekvacnt, 0, "Number of times we have freed the KVA space from some buffer"); static int bufdefragcnt; SYSCTL_INT(_vfs, OID_AUTO, bufdefragcnt, CTLFLAG_RW, &bufdefragcnt, 0, "Number of times we have had to repeat buffer allocation to defragment"); static long lorunningspace; SYSCTL_PROC(_vfs, OID_AUTO, lorunningspace, CTLTYPE_LONG | CTLFLAG_MPSAFE | CTLFLAG_RW, &lorunningspace, 0, sysctl_runningspace, "L", "Minimum preferred space used for in-progress I/O"); static long hirunningspace; SYSCTL_PROC(_vfs, OID_AUTO, hirunningspace, CTLTYPE_LONG | CTLFLAG_MPSAFE | CTLFLAG_RW, &hirunningspace, 0, sysctl_runningspace, "L", "Maximum amount of space to use for in-progress I/O"); int dirtybufferflushes; SYSCTL_INT(_vfs, OID_AUTO, dirtybufferflushes, CTLFLAG_RW, &dirtybufferflushes, 0, "Number of bdwrite to bawrite conversions to limit dirty buffers"); int bdwriteskip; SYSCTL_INT(_vfs, OID_AUTO, bdwriteskip, CTLFLAG_RW, &bdwriteskip, 0, "Number of buffers supplied to bdwrite with snapshot deadlock risk"); int altbufferflushes; SYSCTL_INT(_vfs, OID_AUTO, altbufferflushes, CTLFLAG_RW, &altbufferflushes, 0, "Number of fsync flushes to limit dirty buffers"); static int recursiveflushes; SYSCTL_INT(_vfs, OID_AUTO, recursiveflushes, CTLFLAG_RW, &recursiveflushes, 0, "Number of flushes skipped due to being recursive"); static int numdirtybuffers; SYSCTL_INT(_vfs, OID_AUTO, numdirtybuffers, CTLFLAG_RD, &numdirtybuffers, 0, "Number of buffers that are dirty (has unwritten changes) at the moment"); static int lodirtybuffers; SYSCTL_INT(_vfs, OID_AUTO, lodirtybuffers, CTLFLAG_RW, &lodirtybuffers, 0, "How many buffers we want to have free before bufdaemon can sleep"); static int hidirtybuffers; SYSCTL_INT(_vfs, OID_AUTO, hidirtybuffers, CTLFLAG_RW, &hidirtybuffers, 0, "When the number of dirty buffers is considered severe"); int dirtybufthresh; SYSCTL_INT(_vfs, OID_AUTO, dirtybufthresh, CTLFLAG_RW, &dirtybufthresh, 0, "Number of bdwrite to bawrite conversions to clear dirty buffers"); static int numfreebuffers; SYSCTL_INT(_vfs, OID_AUTO, numfreebuffers, CTLFLAG_RD, &numfreebuffers, 0, "Number of free buffers"); static int lofreebuffers; SYSCTL_INT(_vfs, OID_AUTO, lofreebuffers, CTLFLAG_RW, &lofreebuffers, 0, "XXX Unused"); static int hifreebuffers; SYSCTL_INT(_vfs, OID_AUTO, hifreebuffers, CTLFLAG_RW, &hifreebuffers, 0, "XXX Complicatedly unused"); static int getnewbufcalls; SYSCTL_INT(_vfs, OID_AUTO, getnewbufcalls, CTLFLAG_RW, &getnewbufcalls, 0, "Number of calls to getnewbuf"); static int getnewbufrestarts; SYSCTL_INT(_vfs, OID_AUTO, getnewbufrestarts, CTLFLAG_RW, &getnewbufrestarts, 0, "Number of times getnewbuf has had to restart a buffer aquisition"); static int mappingrestarts; SYSCTL_INT(_vfs, OID_AUTO, mappingrestarts, CTLFLAG_RW, &mappingrestarts, 0, "Number of times getblk has had to restart a buffer mapping for " "unmapped buffer"); static int flushbufqtarget = 100; SYSCTL_INT(_vfs, OID_AUTO, flushbufqtarget, CTLFLAG_RW, &flushbufqtarget, 0, "Amount of work to do in flushbufqueues when helping bufdaemon"); static long notbufdflushes; SYSCTL_LONG(_vfs, OID_AUTO, notbufdflushes, CTLFLAG_RD, ¬bufdflushes, 0, "Number of dirty buffer flushes done by the bufdaemon helpers"); static long barrierwrites; SYSCTL_LONG(_vfs, OID_AUTO, barrierwrites, CTLFLAG_RW, &barrierwrites, 0, "Number of barrier writes"); SYSCTL_INT(_vfs, OID_AUTO, unmapped_buf_allowed, CTLFLAG_RD, &unmapped_buf_allowed, 0, "Permit the use of the unmapped i/o"); /* * Lock for the non-dirty bufqueues */ static struct mtx_padalign bqclean; /* * Lock for the dirty queue. */ static struct mtx_padalign bqdirty; /* * This lock synchronizes access to bd_request. */ static struct mtx_padalign bdlock; /* * This lock protects the runningbufreq and synchronizes runningbufwakeup and * waitrunningbufspace(). */ static struct mtx_padalign rbreqlock; /* * Lock that protects needsbuffer and the sleeps/wakeups surrounding it. */ static struct rwlock_padalign nblock; /* * Lock that protects bdirtywait. */ static struct mtx_padalign bdirtylock; /* * Wakeup point for bufdaemon, as well as indicator of whether it is already * active. Set to 1 when the bufdaemon is already "on" the queue, 0 when it * is idling. */ static int bd_request; /* * Request for the buf daemon to write more buffers than is indicated by * lodirtybuf. This may be necessary to push out excess dependencies or * defragment the address space where a simple count of the number of dirty * buffers is insufficient to characterize the demand for flushing them. */ static int bd_speedupreq; /* * bogus page -- for I/O to/from partially complete buffers * this is a temporary solution to the problem, but it is not * really that bad. it would be better to split the buffer * for input in the case of buffers partially already in memory, * but the code is intricate enough already. */ vm_page_t bogus_page; /* * Synchronization (sleep/wakeup) variable for active buffer space requests. * Set when wait starts, cleared prior to wakeup(). * Used in runningbufwakeup() and waitrunningbufspace(). */ static int runningbufreq; /* * Synchronization (sleep/wakeup) variable for buffer requests. * Can contain the VFS_BIO_NEED flags defined below; setting/clearing is done * by and/or. * Used in numdirtywakeup(), bufspacewakeup(), bufcountadd(), bwillwrite(), * getnewbuf(), and getblk(). */ static volatile int needsbuffer; /* * Synchronization for bwillwrite() waiters. */ static int bdirtywait; /* * Definitions for the buffer free lists. */ #define BUFFER_QUEUES 4 /* number of free buffer queues */ #define QUEUE_NONE 0 /* on no queue */ #define QUEUE_CLEAN 1 /* non-B_DELWRI buffers */ #define QUEUE_DIRTY 2 /* B_DELWRI buffers */ #define QUEUE_EMPTY 3 /* empty buffer headers */ #define QUEUE_SENTINEL 1024 /* not an queue index, but mark for sentinel */ /* Queues for free buffers with various properties */ static TAILQ_HEAD(bqueues, buf) bufqueues[BUFFER_QUEUES] = { { 0 } }; #ifdef INVARIANTS static int bq_len[BUFFER_QUEUES]; #endif /* * Single global constant for BUF_WMESG, to avoid getting multiple references. * buf_wmesg is referred from macros. */ const char *buf_wmesg = BUF_WMESG; #define VFS_BIO_NEED_ANY 0x01 /* any freeable buffer */ #define VFS_BIO_NEED_FREE 0x04 /* wait for free bufs, hi hysteresis */ #define VFS_BIO_NEED_BUFSPACE 0x08 /* wait for buf space, lo hysteresis */ static int sysctl_runningspace(SYSCTL_HANDLER_ARGS) { long value; int error; value = *(long *)arg1; error = sysctl_handle_long(oidp, &value, 0, req); if (error != 0 || req->newptr == NULL) return (error); mtx_lock(&rbreqlock); if (arg1 == &hirunningspace) { if (value < lorunningspace) error = EINVAL; else hirunningspace = value; } else { KASSERT(arg1 == &lorunningspace, ("%s: unknown arg1", __func__)); if (value > hirunningspace) error = EINVAL; else lorunningspace = value; } mtx_unlock(&rbreqlock); return (error); } #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) static int sysctl_bufspace(SYSCTL_HANDLER_ARGS) { long lvalue; int ivalue; if (sizeof(int) == sizeof(long) || req->oldlen >= sizeof(long)) return (sysctl_handle_long(oidp, arg1, arg2, req)); lvalue = *(long *)arg1; if (lvalue > INT_MAX) /* On overflow, still write out a long to trigger ENOMEM. */ return (sysctl_handle_long(oidp, &lvalue, 0, req)); ivalue = lvalue; return (sysctl_handle_int(oidp, &ivalue, 0, req)); } #endif /* * bqlock: * * Return the appropriate queue lock based on the index. */ static inline struct mtx * bqlock(int qindex) { if (qindex == QUEUE_DIRTY) return (struct mtx *)(&bqdirty); return (struct mtx *)(&bqclean); } /* * bdirtywakeup: * * Wakeup any bwillwrite() waiters. */ static void bdirtywakeup(void) { mtx_lock(&bdirtylock); if (bdirtywait) { bdirtywait = 0; wakeup(&bdirtywait); } mtx_unlock(&bdirtylock); } /* * bdirtysub: * * Decrement the numdirtybuffers count by one and wakeup any * threads blocked in bwillwrite(). */ static void bdirtysub(void) { if (atomic_fetchadd_int(&numdirtybuffers, -1) == (lodirtybuffers + hidirtybuffers) / 2) bdirtywakeup(); } /* * bdirtyadd: * * Increment the numdirtybuffers count by one and wakeup the buf * daemon if needed. */ static void bdirtyadd(void) { /* * Only do the wakeup once as we cross the boundary. The * buf daemon will keep running until the condition clears. */ if (atomic_fetchadd_int(&numdirtybuffers, 1) == (lodirtybuffers + hidirtybuffers) / 2) bd_wakeup(); } /* * bufspacewakeup: * * Called when buffer space is potentially available for recovery. * getnewbuf() will block on this flag when it is unable to free * sufficient buffer space. Buffer space becomes recoverable when * bp's get placed back in the queues. */ static __inline void bufspacewakeup(void) { int need_wakeup, on; /* * If someone is waiting for bufspace, wake them up. Even * though we may not have freed the kva space yet, the waiting * process will be able to now. */ rw_rlock(&nblock); for (;;) { need_wakeup = 0; on = needsbuffer; if ((on & VFS_BIO_NEED_BUFSPACE) == 0) break; need_wakeup = 1; if (atomic_cmpset_rel_int(&needsbuffer, on, on & ~VFS_BIO_NEED_BUFSPACE)) break; } if (need_wakeup) wakeup(__DEVOLATILE(void *, &needsbuffer)); rw_runlock(&nblock); } /* * bufspaceadjust: * * Adjust the reported bufspace for a KVA managed buffer, possibly * waking any waiters. */ static void bufspaceadjust(struct buf *bp, int bufsize) { int diff; KASSERT((bp->b_flags & B_MALLOC) == 0, ("bufspaceadjust: malloc buf %p", bp)); diff = bufsize - bp->b_bufsize; if (diff < 0) { atomic_subtract_long(&bufspace, -diff); bufspacewakeup(); } else atomic_add_long(&bufspace, diff); bp->b_bufsize = bufsize; } /* * bufmallocadjust: * * Adjust the reported bufspace for a malloc managed buffer, possibly * waking any waiters. */ static void bufmallocadjust(struct buf *bp, int bufsize) { int diff; KASSERT((bp->b_flags & B_MALLOC) != 0, ("bufmallocadjust: non-malloc buf %p", bp)); diff = bufsize - bp->b_bufsize; if (diff < 0) { atomic_subtract_long(&bufmallocspace, -diff); bufspacewakeup(); } else atomic_add_long(&bufmallocspace, diff); bp->b_bufsize = bufsize; } /* * runningwakeup: * * Wake up processes that are waiting on asynchronous writes to fall * below lorunningspace. */ static void runningwakeup(void) { mtx_lock(&rbreqlock); if (runningbufreq) { runningbufreq = 0; wakeup(&runningbufreq); } mtx_unlock(&rbreqlock); } /* * runningbufwakeup: * * Decrement the outstanding write count according. */ void runningbufwakeup(struct buf *bp) { long space, bspace; bspace = bp->b_runningbufspace; if (bspace == 0) return; space = atomic_fetchadd_long(&runningbufspace, -bspace); KASSERT(space >= bspace, ("runningbufspace underflow %ld %ld", space, bspace)); bp->b_runningbufspace = 0; /* * Only acquire the lock and wakeup on the transition from exceeding * the threshold to falling below it. */ if (space < lorunningspace) return; if (space - bspace > lorunningspace) return; runningwakeup(); } /* * bufcountadd: * * Called when a buffer has been added to one of the free queues to * account for the buffer and to wakeup anyone waiting for free buffers. * This typically occurs when large amounts of metadata are being handled * by the buffer cache ( else buffer space runs out first, usually ). */ static __inline void bufcountadd(struct buf *bp) { int mask, need_wakeup, old, on; KASSERT((bp->b_flags & B_INFREECNT) == 0, ("buf %p already counted as free", bp)); bp->b_flags |= B_INFREECNT; old = atomic_fetchadd_int(&numfreebuffers, 1); KASSERT(old >= 0 && old < nbuf, ("numfreebuffers climbed to %d", old + 1)); mask = VFS_BIO_NEED_ANY; if (numfreebuffers >= hifreebuffers) mask |= VFS_BIO_NEED_FREE; rw_rlock(&nblock); for (;;) { need_wakeup = 0; on = needsbuffer; if (on == 0) break; need_wakeup = 1; if (atomic_cmpset_rel_int(&needsbuffer, on, on & ~mask)) break; } if (need_wakeup) wakeup(__DEVOLATILE(void *, &needsbuffer)); rw_runlock(&nblock); } /* * bufcountsub: * * Decrement the numfreebuffers count as needed. */ static void bufcountsub(struct buf *bp) { int old; /* * Fixup numfreebuffers count. If the buffer is invalid or not * delayed-write, the buffer was free and we must decrement * numfreebuffers. */ if ((bp->b_flags & B_INVAL) || (bp->b_flags & B_DELWRI) == 0) { KASSERT((bp->b_flags & B_INFREECNT) != 0, ("buf %p not counted in numfreebuffers", bp)); bp->b_flags &= ~B_INFREECNT; old = atomic_fetchadd_int(&numfreebuffers, -1); KASSERT(old > 0, ("numfreebuffers dropped to %d", old - 1)); } } /* * waitrunningbufspace() * * runningbufspace is a measure of the amount of I/O currently * running. This routine is used in async-write situations to * prevent creating huge backups of pending writes to a device. * Only asynchronous writes are governed by this function. * * This does NOT turn an async write into a sync write. It waits * for earlier writes to complete and generally returns before the * caller's write has reached the device. */ void waitrunningbufspace(void) { mtx_lock(&rbreqlock); while (runningbufspace > hirunningspace) { runningbufreq = 1; msleep(&runningbufreq, &rbreqlock, PVM, "wdrain", 0); } mtx_unlock(&rbreqlock); } /* * vfs_buf_test_cache: * * Called when a buffer is extended. This function clears the B_CACHE * bit if the newly extended portion of the buffer does not contain * valid data. */ static __inline void vfs_buf_test_cache(struct buf *bp, vm_ooffset_t foff, vm_offset_t off, vm_offset_t size, vm_page_t m) { VM_OBJECT_ASSERT_LOCKED(m->object); if (bp->b_flags & B_CACHE) { int base = (foff + off) & PAGE_MASK; if (vm_page_is_valid(m, base, size) == 0) bp->b_flags &= ~B_CACHE; } } /* Wake up the buffer daemon if necessary */ static __inline void bd_wakeup(void) { mtx_lock(&bdlock); if (bd_request == 0) { bd_request = 1; wakeup(&bd_request); } mtx_unlock(&bdlock); } /* * bd_speedup - speedup the buffer cache flushing code */ void bd_speedup(void) { int needwake; mtx_lock(&bdlock); needwake = 0; if (bd_speedupreq == 0 || bd_request == 0) needwake = 1; bd_speedupreq = 1; bd_request = 1; if (needwake) wakeup(&bd_request); mtx_unlock(&bdlock); } #ifndef NSWBUF_MIN #define NSWBUF_MIN 16 #endif #ifdef __i386__ #define TRANSIENT_DENOM 5 #else #define TRANSIENT_DENOM 10 #endif /* * Calculating buffer cache scaling values and reserve space for buffer * headers. This is called during low level kernel initialization and * may be called more then once. We CANNOT write to the memory area * being reserved at this time. */ caddr_t kern_vfs_bio_buffer_alloc(caddr_t v, long physmem_est) { int tuned_nbuf; long maxbuf, maxbuf_sz, buf_sz, biotmap_sz; /* * physmem_est is in pages. Convert it to kilobytes (assumes * PAGE_SIZE is >= 1K) */ physmem_est = physmem_est * (PAGE_SIZE / 1024); /* * The nominal buffer size (and minimum KVA allocation) is BKVASIZE. * For the first 64MB of ram nominally allocate sufficient buffers to * cover 1/4 of our ram. Beyond the first 64MB allocate additional * buffers to cover 1/10 of our ram over 64MB. When auto-sizing * the buffer cache we limit the eventual kva reservation to * maxbcache bytes. * * factor represents the 1/4 x ram conversion. */ if (nbuf == 0) { int factor = 4 * BKVASIZE / 1024; nbuf = 50; if (physmem_est > 4096) nbuf += min((physmem_est - 4096) / factor, 65536 / factor); if (physmem_est > 65536) nbuf += min((physmem_est - 65536) * 2 / (factor * 5), 32 * 1024 * 1024 / (factor * 5)); if (maxbcache && nbuf > maxbcache / BKVASIZE) nbuf = maxbcache / BKVASIZE; tuned_nbuf = 1; } else tuned_nbuf = 0; /* XXX Avoid unsigned long overflows later on with maxbufspace. */ maxbuf = (LONG_MAX / 3) / BKVASIZE; if (nbuf > maxbuf) { if (!tuned_nbuf) printf("Warning: nbufs lowered from %d to %ld\n", nbuf, maxbuf); nbuf = maxbuf; } /* * Ideal allocation size for the transient bio submap is 10% * of the maximal space buffer map. This roughly corresponds * to the amount of the buffer mapped for typical UFS load. * * Clip the buffer map to reserve space for the transient * BIOs, if its extent is bigger than 90% (80% on i386) of the * maximum buffer map extent on the platform. * * The fall-back to the maxbuf in case of maxbcache unset, * allows to not trim the buffer KVA for the architectures * with ample KVA space. */ if (bio_transient_maxcnt == 0 && unmapped_buf_allowed) { maxbuf_sz = maxbcache != 0 ? maxbcache : maxbuf * BKVASIZE; buf_sz = (long)nbuf * BKVASIZE; if (buf_sz < maxbuf_sz / TRANSIENT_DENOM * (TRANSIENT_DENOM - 1)) { /* * There is more KVA than memory. Do not * adjust buffer map size, and assign the rest * of maxbuf to transient map. */ biotmap_sz = maxbuf_sz - buf_sz; } else { /* * Buffer map spans all KVA we could afford on * this platform. Give 10% (20% on i386) of * the buffer map to the transient bio map. */ biotmap_sz = buf_sz / TRANSIENT_DENOM; buf_sz -= biotmap_sz; } if (biotmap_sz / INT_MAX > MAXPHYS) bio_transient_maxcnt = INT_MAX; else bio_transient_maxcnt = biotmap_sz / MAXPHYS; /* * Artifically limit to 1024 simultaneous in-flight I/Os * using the transient mapping. */ if (bio_transient_maxcnt > 1024) bio_transient_maxcnt = 1024; if (tuned_nbuf) nbuf = buf_sz / BKVASIZE; } /* * swbufs are used as temporary holders for I/O, such as paging I/O. * We have no less then 16 and no more then 256. */ nswbuf = min(nbuf / 4, 256); TUNABLE_INT_FETCH("kern.nswbuf", &nswbuf); if (nswbuf < NSWBUF_MIN) nswbuf = NSWBUF_MIN; /* * Reserve space for the buffer cache buffers */ swbuf = (void *)v; v = (caddr_t)(swbuf + nswbuf); buf = (void *)v; v = (caddr_t)(buf + nbuf); return(v); } /* Initialize the buffer subsystem. Called before use of any buffers. */ void bufinit(void) { struct buf *bp; int i; CTASSERT(MAXBCACHEBUF >= MAXBSIZE); mtx_init(&bqclean, "bufq clean lock", NULL, MTX_DEF); mtx_init(&bqdirty, "bufq dirty lock", NULL, MTX_DEF); mtx_init(&rbreqlock, "runningbufspace lock", NULL, MTX_DEF); rw_init(&nblock, "needsbuffer lock"); mtx_init(&bdlock, "buffer daemon lock", NULL, MTX_DEF); mtx_init(&bdirtylock, "dirty buf lock", NULL, MTX_DEF); /* next, make a null set of free lists */ for (i = 0; i < BUFFER_QUEUES; i++) TAILQ_INIT(&bufqueues[i]); unmapped_buf = (caddr_t)kva_alloc(MAXPHYS); /* finally, initialize each buffer header and stick on empty q */ for (i = 0; i < nbuf; i++) { bp = &buf[i]; bzero(bp, sizeof *bp); bp->b_flags = B_INVAL | B_INFREECNT; bp->b_rcred = NOCRED; bp->b_wcred = NOCRED; bp->b_qindex = QUEUE_EMPTY; bp->b_xflags = 0; bp->b_data = bp->b_kvabase = unmapped_buf; LIST_INIT(&bp->b_dep); BUF_LOCKINIT(bp); TAILQ_INSERT_TAIL(&bufqueues[QUEUE_EMPTY], bp, b_freelist); #ifdef INVARIANTS bq_len[QUEUE_EMPTY]++; #endif } /* * maxbufspace is the absolute maximum amount of buffer space we are * allowed to reserve in KVM and in real terms. The absolute maximum * is nominally used by buf_daemon. hibufspace is the nominal maximum * used by most other processes. The differential is required to * ensure that buf_daemon is able to run when other processes might * be blocked waiting for buffer space. * * maxbufspace is based on BKVASIZE. Allocating buffers larger then * this may result in KVM fragmentation which is not handled optimally * by the system. */ maxbufspace = (long)nbuf * BKVASIZE; hibufspace = lmax(3 * maxbufspace / 4, maxbufspace - MAXBCACHEBUF * 10); lobufspace = hibufspace - MAXBCACHEBUF; /* * Note: The 16 MiB upper limit for hirunningspace was chosen * arbitrarily and may need further tuning. It corresponds to * 128 outstanding write IO requests (if IO size is 128 KiB), * which fits with many RAID controllers' tagged queuing limits. * The lower 1 MiB limit is the historical upper limit for * hirunningspace. */ hirunningspace = lmax(lmin(roundup(hibufspace / 64, MAXBCACHEBUF), 16 * 1024 * 1024), 1024 * 1024); lorunningspace = roundup((hirunningspace * 2) / 3, MAXBCACHEBUF); /* * Limit the amount of malloc memory since it is wired permanently into * the kernel space. Even though this is accounted for in the buffer * allocation, we don't want the malloced region to grow uncontrolled. * The malloc scheme improves memory utilization significantly on average * (small) directories. */ maxbufmallocspace = hibufspace / 20; /* * Reduce the chance of a deadlock occuring by limiting the number * of delayed-write dirty buffers we allow to stack up. */ hidirtybuffers = nbuf / 4 + 20; dirtybufthresh = hidirtybuffers * 9 / 10; numdirtybuffers = 0; /* * To support extreme low-memory systems, make sure hidirtybuffers cannot * eat up all available buffer space. This occurs when our minimum cannot * be met. We try to size hidirtybuffers to 3/4 our buffer space assuming * BKVASIZE'd buffers. */ while ((long)hidirtybuffers * BKVASIZE > 3 * hibufspace / 4) { hidirtybuffers >>= 1; } lodirtybuffers = hidirtybuffers / 2; /* * Try to keep the number of free buffers in the specified range, * and give special processes (e.g. like buf_daemon) access to an * emergency reserve. */ lofreebuffers = nbuf / 18 + 5; hifreebuffers = 2 * lofreebuffers; numfreebuffers = nbuf; bogus_page = vm_page_alloc(NULL, 0, VM_ALLOC_NOOBJ | VM_ALLOC_NORMAL | VM_ALLOC_WIRED); } #ifdef INVARIANTS static inline void vfs_buf_check_mapped(struct buf *bp) { KASSERT(bp->b_kvabase != unmapped_buf, ("mapped buf: b_kvabase was not updated %p", bp)); KASSERT(bp->b_data != unmapped_buf, ("mapped buf: b_data was not updated %p", bp)); KASSERT(bp->b_data < unmapped_buf || bp->b_data >= unmapped_buf + MAXPHYS, ("b_data + b_offset unmapped %p", bp)); } static inline void vfs_buf_check_unmapped(struct buf *bp) { KASSERT(bp->b_data == unmapped_buf, ("unmapped buf: corrupted b_data %p", bp)); } #define BUF_CHECK_MAPPED(bp) vfs_buf_check_mapped(bp) #define BUF_CHECK_UNMAPPED(bp) vfs_buf_check_unmapped(bp) #else #define BUF_CHECK_MAPPED(bp) do {} while (0) #define BUF_CHECK_UNMAPPED(bp) do {} while (0) #endif static int isbufbusy(struct buf *bp) { if (((bp->b_flags & (B_INVAL | B_PERSISTENT)) == 0 && BUF_ISLOCKED(bp)) || ((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI)) return (1); return (0); } /* * Shutdown the system cleanly to prepare for reboot, halt, or power off. */ void bufshutdown(int show_busybufs) { static int first_buf_printf = 1; struct buf *bp; int iter, nbusy, pbusy; #ifndef PREEMPTION int subiter; #endif /* * Sync filesystems for shutdown */ wdog_kern_pat(WD_LASTVAL); sys_sync(curthread, NULL); /* * With soft updates, some buffers that are * written will be remarked as dirty until other * buffers are written. */ for (iter = pbusy = 0; iter < 20; iter++) { nbusy = 0; for (bp = &buf[nbuf]; --bp >= buf; ) if (isbufbusy(bp)) nbusy++; if (nbusy == 0) { if (first_buf_printf) printf("All buffers synced."); break; } if (first_buf_printf) { printf("Syncing disks, buffers remaining... "); first_buf_printf = 0; } printf("%d ", nbusy); if (nbusy < pbusy) iter = 0; pbusy = nbusy; wdog_kern_pat(WD_LASTVAL); sys_sync(curthread, NULL); #ifdef PREEMPTION /* * Drop Giant and spin for a while to allow * interrupt threads to run. */ DROP_GIANT(); DELAY(50000 * iter); PICKUP_GIANT(); #else /* * Drop Giant and context switch several times to * allow interrupt threads to run. */ DROP_GIANT(); for (subiter = 0; subiter < 50 * iter; subiter++) { thread_lock(curthread); mi_switch(SW_VOL, NULL); thread_unlock(curthread); DELAY(1000); } PICKUP_GIANT(); #endif } printf("\n"); /* * Count only busy local buffers to prevent forcing * a fsck if we're just a client of a wedged NFS server */ nbusy = 0; for (bp = &buf[nbuf]; --bp >= buf; ) { if (isbufbusy(bp)) { #if 0 /* XXX: This is bogus. We should probably have a BO_REMOTE flag instead */ if (bp->b_dev == NULL) { TAILQ_REMOVE(&mountlist, bp->b_vp->v_mount, mnt_list); continue; } #endif nbusy++; if (show_busybufs > 0) { printf( "%d: buf:%p, vnode:%p, flags:%0x, blkno:%jd, lblkno:%jd, buflock:", nbusy, bp, bp->b_vp, bp->b_flags, (intmax_t)bp->b_blkno, (intmax_t)bp->b_lblkno); BUF_LOCKPRINTINFO(bp); if (show_busybufs > 1) vn_printf(bp->b_vp, "vnode content: "); } } } if (nbusy) { /* * Failed to sync all blocks. Indicate this and don't * unmount filesystems (thus forcing an fsck on reboot). */ printf("Giving up on %d buffers\n", nbusy); DELAY(5000000); /* 5 seconds */ } else { if (!first_buf_printf) printf("Final sync complete\n"); /* * Unmount filesystems */ if (panicstr == 0) vfs_unmountall(); } swapoff_all(); DELAY(100000); /* wait for console output to finish */ } static void bpmap_qenter(struct buf *bp) { BUF_CHECK_MAPPED(bp); /* * bp->b_data is relative to bp->b_offset, but * bp->b_offset may be offset into the first page. */ bp->b_data = (caddr_t)trunc_page((vm_offset_t)bp->b_data); pmap_qenter((vm_offset_t)bp->b_data, bp->b_pages, bp->b_npages); bp->b_data = (caddr_t)((vm_offset_t)bp->b_data | (vm_offset_t)(bp->b_offset & PAGE_MASK)); } /* * binsfree: * * Insert the buffer into the appropriate free list. */ static void binsfree(struct buf *bp, int qindex) { struct mtx *olock, *nlock; BUF_ASSERT_XLOCKED(bp); nlock = bqlock(qindex); /* Handle delayed bremfree() processing. */ if (bp->b_flags & B_REMFREE) { olock = bqlock(bp->b_qindex); mtx_lock(olock); bremfreel(bp); if (olock != nlock) { mtx_unlock(olock); mtx_lock(nlock); } } else mtx_lock(nlock); if (bp->b_qindex != QUEUE_NONE) panic("binsfree: free buffer onto another queue???"); bp->b_qindex = qindex; if (bp->b_flags & B_AGE) TAILQ_INSERT_HEAD(&bufqueues[bp->b_qindex], bp, b_freelist); else TAILQ_INSERT_TAIL(&bufqueues[bp->b_qindex], bp, b_freelist); #ifdef INVARIANTS bq_len[bp->b_qindex]++; #endif mtx_unlock(nlock); /* * Something we can maybe free or reuse. */ if (bp->b_bufsize && !(bp->b_flags & B_DELWRI)) bufspacewakeup(); if ((bp->b_flags & B_INVAL) || !(bp->b_flags & B_DELWRI)) bufcountadd(bp); } /* * bremfree: * * Mark the buffer for removal from the appropriate free list. * */ void bremfree(struct buf *bp) { CTR3(KTR_BUF, "bremfree(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT((bp->b_flags & B_REMFREE) == 0, ("bremfree: buffer %p already marked for delayed removal.", bp)); KASSERT(bp->b_qindex != QUEUE_NONE, ("bremfree: buffer %p not on a queue.", bp)); BUF_ASSERT_XLOCKED(bp); bp->b_flags |= B_REMFREE; bufcountsub(bp); } /* * bremfreef: * * Force an immediate removal from a free list. Used only in nfs when * it abuses the b_freelist pointer. */ void bremfreef(struct buf *bp) { struct mtx *qlock; qlock = bqlock(bp->b_qindex); mtx_lock(qlock); bremfreel(bp); mtx_unlock(qlock); } /* * bremfreel: * * Removes a buffer from the free list, must be called with the * correct qlock held. */ static void bremfreel(struct buf *bp) { CTR3(KTR_BUF, "bremfreel(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(bp->b_qindex != QUEUE_NONE, ("bremfreel: buffer %p not on a queue.", bp)); BUF_ASSERT_XLOCKED(bp); mtx_assert(bqlock(bp->b_qindex), MA_OWNED); TAILQ_REMOVE(&bufqueues[bp->b_qindex], bp, b_freelist); #ifdef INVARIANTS KASSERT(bq_len[bp->b_qindex] >= 1, ("queue %d underflow", bp->b_qindex)); bq_len[bp->b_qindex]--; #endif bp->b_qindex = QUEUE_NONE; /* * If this was a delayed bremfree() we only need to remove the buffer * from the queue and return the stats are already done. */ if (bp->b_flags & B_REMFREE) { bp->b_flags &= ~B_REMFREE; return; } bufcountsub(bp); } /* * bufkvafree: * * Free the kva allocation for a buffer. * */ static void bufkvafree(struct buf *bp) { #ifdef INVARIANTS if (bp->b_kvasize == 0) { KASSERT(bp->b_kvabase == unmapped_buf && bp->b_data == unmapped_buf, ("Leaked KVA space on %p", bp)); } else if (buf_mapped(bp)) BUF_CHECK_MAPPED(bp); else BUF_CHECK_UNMAPPED(bp); #endif if (bp->b_kvasize == 0) return; vmem_free(buffer_arena, (vm_offset_t)bp->b_kvabase, bp->b_kvasize); atomic_subtract_long(&bufkvaspace, bp->b_kvasize); atomic_add_int(&buffreekvacnt, 1); bp->b_data = bp->b_kvabase = unmapped_buf; bp->b_kvasize = 0; } /* * bufkvaalloc: * * Allocate the buffer KVA and set b_kvasize and b_kvabase. */ static int bufkvaalloc(struct buf *bp, int maxsize, int gbflags) { vm_offset_t addr; int error; KASSERT((gbflags & GB_UNMAPPED) == 0 || (gbflags & GB_KVAALLOC) != 0, ("Invalid gbflags 0x%x in %s", gbflags, __func__)); bufkvafree(bp); addr = 0; error = vmem_alloc(buffer_arena, maxsize, M_BESTFIT | M_NOWAIT, &addr); if (error != 0) { /* * Buffer map is too fragmented. Request the caller * to defragment the map. */ atomic_add_int(&bufdefragcnt, 1); return (error); } bp->b_kvabase = (caddr_t)addr; bp->b_kvasize = maxsize; atomic_add_long(&bufkvaspace, bp->b_kvasize); if ((gbflags & GB_UNMAPPED) != 0) { bp->b_data = unmapped_buf; BUF_CHECK_UNMAPPED(bp); } else { bp->b_data = bp->b_kvabase; BUF_CHECK_MAPPED(bp); } return (0); } /* * Attempt to initiate asynchronous I/O on read-ahead blocks. We must * clear BIO_ERROR and B_INVAL prior to initiating I/O . If B_CACHE is set, * the buffer is valid and we do not have to do anything. */ void breada(struct vnode * vp, daddr_t * rablkno, int * rabsize, int cnt, struct ucred * cred) { struct buf *rabp; int i; for (i = 0; i < cnt; i++, rablkno++, rabsize++) { if (inmem(vp, *rablkno)) continue; rabp = getblk(vp, *rablkno, *rabsize, 0, 0, 0); if ((rabp->b_flags & B_CACHE) == 0) { if (!TD_IS_IDLETHREAD(curthread)) curthread->td_ru.ru_inblock++; rabp->b_flags |= B_ASYNC; rabp->b_flags &= ~B_INVAL; rabp->b_ioflags &= ~BIO_ERROR; rabp->b_iocmd = BIO_READ; if (rabp->b_rcred == NOCRED && cred != NOCRED) rabp->b_rcred = crhold(cred); vfs_busy_pages(rabp, 0); BUF_KERNPROC(rabp); rabp->b_iooffset = dbtob(rabp->b_blkno); bstrategy(rabp); } else { brelse(rabp); } } } /* * Entry point for bread() and breadn() via #defines in sys/buf.h. * * Get a buffer with the specified data. Look in the cache first. We * must clear BIO_ERROR and B_INVAL prior to initiating I/O. If B_CACHE * is set, the buffer is valid and we do not have to do anything, see * getblk(). Also starts asynchronous I/O on read-ahead blocks. */ int breadn_flags(struct vnode *vp, daddr_t blkno, int size, daddr_t *rablkno, int *rabsize, int cnt, struct ucred *cred, int flags, struct buf **bpp) { struct buf *bp; int rv = 0, readwait = 0; CTR3(KTR_BUF, "breadn(%p, %jd, %d)", vp, blkno, size); /* * Can only return NULL if GB_LOCK_NOWAIT flag is specified. */ *bpp = bp = getblk(vp, blkno, size, 0, 0, flags); if (bp == NULL) return (EBUSY); /* if not found in cache, do some I/O */ if ((bp->b_flags & B_CACHE) == 0) { if (!TD_IS_IDLETHREAD(curthread)) curthread->td_ru.ru_inblock++; bp->b_iocmd = BIO_READ; bp->b_flags &= ~B_INVAL; bp->b_ioflags &= ~BIO_ERROR; if (bp->b_rcred == NOCRED && cred != NOCRED) bp->b_rcred = crhold(cred); vfs_busy_pages(bp, 0); bp->b_iooffset = dbtob(bp->b_blkno); bstrategy(bp); ++readwait; } breada(vp, rablkno, rabsize, cnt, cred); if (readwait) { rv = bufwait(bp); } return (rv); } /* * Write, release buffer on completion. (Done by iodone * if async). Do not bother writing anything if the buffer * is invalid. * * Note that we set B_CACHE here, indicating that buffer is * fully valid and thus cacheable. This is true even of NFS * now so we set it generally. This could be set either here * or in biodone() since the I/O is synchronous. We put it * here. */ int bufwrite(struct buf *bp) { int oldflags; struct vnode *vp; long space; int vp_md; CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); if ((bp->b_bufobj->bo_flag & BO_DEAD) != 0) { bp->b_flags |= B_INVAL | B_RELBUF; bp->b_flags &= ~B_CACHE; brelse(bp); return (ENXIO); } if (bp->b_flags & B_INVAL) { brelse(bp); return (0); } if (bp->b_flags & B_BARRIER) barrierwrites++; oldflags = bp->b_flags; BUF_ASSERT_HELD(bp); if (bp->b_pin_count > 0) bunpin_wait(bp); KASSERT(!(bp->b_vflags & BV_BKGRDINPROG), ("FFS background buffer should not get here %p", bp)); vp = bp->b_vp; if (vp) vp_md = vp->v_vflag & VV_MD; else vp_md = 0; /* * Mark the buffer clean. Increment the bufobj write count * before bundirty() call, to prevent other thread from seeing * empty dirty list and zero counter for writes in progress, * falsely indicating that the bufobj is clean. */ bufobj_wref(bp->b_bufobj); bundirty(bp); bp->b_flags &= ~B_DONE; bp->b_ioflags &= ~BIO_ERROR; bp->b_flags |= B_CACHE; bp->b_iocmd = BIO_WRITE; vfs_busy_pages(bp, 1); /* * Normal bwrites pipeline writes */ bp->b_runningbufspace = bp->b_bufsize; space = atomic_fetchadd_long(&runningbufspace, bp->b_runningbufspace); if (!TD_IS_IDLETHREAD(curthread)) curthread->td_ru.ru_oublock++; if (oldflags & B_ASYNC) BUF_KERNPROC(bp); bp->b_iooffset = dbtob(bp->b_blkno); bstrategy(bp); if ((oldflags & B_ASYNC) == 0) { int rtval = bufwait(bp); brelse(bp); return (rtval); } else if (space > hirunningspace) { /* * don't allow the async write to saturate the I/O * system. We will not deadlock here because * we are blocking waiting for I/O that is already in-progress * to complete. We do not block here if it is the update * or syncer daemon trying to clean up as that can lead * to deadlock. */ if ((curthread->td_pflags & TDP_NORUNNINGBUF) == 0 && !vp_md) waitrunningbufspace(); } return (0); } void bufbdflush(struct bufobj *bo, struct buf *bp) { struct buf *nbp; if (bo->bo_dirty.bv_cnt > dirtybufthresh + 10) { (void) VOP_FSYNC(bp->b_vp, MNT_NOWAIT, curthread); altbufferflushes++; } else if (bo->bo_dirty.bv_cnt > dirtybufthresh) { BO_LOCK(bo); /* * Try to find a buffer to flush. */ TAILQ_FOREACH(nbp, &bo->bo_dirty.bv_hd, b_bobufs) { if ((nbp->b_vflags & BV_BKGRDINPROG) || BUF_LOCK(nbp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) continue; if (bp == nbp) panic("bdwrite: found ourselves"); BO_UNLOCK(bo); /* Don't countdeps with the bo lock held. */ if (buf_countdeps(nbp, 0)) { BO_LOCK(bo); BUF_UNLOCK(nbp); continue; } if (nbp->b_flags & B_CLUSTEROK) { vfs_bio_awrite(nbp); } else { bremfree(nbp); bawrite(nbp); } dirtybufferflushes++; break; } if (nbp == NULL) BO_UNLOCK(bo); } } /* * Delayed write. (Buffer is marked dirty). Do not bother writing * anything if the buffer is marked invalid. * * Note that since the buffer must be completely valid, we can safely * set B_CACHE. In fact, we have to set B_CACHE here rather then in * biodone() in order to prevent getblk from writing the buffer * out synchronously. */ void bdwrite(struct buf *bp) { struct thread *td = curthread; struct vnode *vp; struct bufobj *bo; CTR3(KTR_BUF, "bdwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp)); KASSERT((bp->b_flags & B_BARRIER) == 0, ("Barrier request in delayed write %p", bp)); BUF_ASSERT_HELD(bp); if (bp->b_flags & B_INVAL) { brelse(bp); return; } /* * If we have too many dirty buffers, don't create any more. * If we are wildly over our limit, then force a complete * cleanup. Otherwise, just keep the situation from getting * out of control. Note that we have to avoid a recursive * disaster and not try to clean up after our own cleanup! */ vp = bp->b_vp; bo = bp->b_bufobj; if ((td->td_pflags & (TDP_COWINPROGRESS|TDP_INBDFLUSH)) == 0) { td->td_pflags |= TDP_INBDFLUSH; BO_BDFLUSH(bo, bp); td->td_pflags &= ~TDP_INBDFLUSH; } else recursiveflushes++; bdirty(bp); /* * Set B_CACHE, indicating that the buffer is fully valid. This is * true even of NFS now. */ bp->b_flags |= B_CACHE; /* * This bmap keeps the system from needing to do the bmap later, * perhaps when the system is attempting to do a sync. Since it * is likely that the indirect block -- or whatever other datastructure * that the filesystem needs is still in memory now, it is a good * thing to do this. Note also, that if the pageout daemon is * requesting a sync -- there might not be enough memory to do * the bmap then... So, this is important to do. */ if (vp->v_type != VCHR && bp->b_lblkno == bp->b_blkno) { VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, NULL, NULL); } /* * Set the *dirty* buffer range based upon the VM system dirty * pages. * * Mark the buffer pages as clean. We need to do this here to * satisfy the vnode_pager and the pageout daemon, so that it * thinks that the pages have been "cleaned". Note that since * the pages are in a delayed write buffer -- the VFS layer * "will" see that the pages get written out on the next sync, * or perhaps the cluster will be completed. */ vfs_clean_pages_dirty_buf(bp); bqrelse(bp); /* * note: we cannot initiate I/O from a bdwrite even if we wanted to, * due to the softdep code. */ } /* * bdirty: * * Turn buffer into delayed write request. We must clear BIO_READ and * B_RELBUF, and we must set B_DELWRI. We reassign the buffer to * itself to properly update it in the dirty/clean lists. We mark it * B_DONE to ensure that any asynchronization of the buffer properly * clears B_DONE ( else a panic will occur later ). * * bdirty() is kinda like bdwrite() - we have to clear B_INVAL which * might have been set pre-getblk(). Unlike bwrite/bdwrite, bdirty() * should only be called if the buffer is known-good. * * Since the buffer is not on a queue, we do not update the numfreebuffers * count. * * The buffer must be on QUEUE_NONE. */ void bdirty(struct buf *bp) { CTR3(KTR_BUF, "bdirty(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp)); KASSERT(bp->b_flags & B_REMFREE || bp->b_qindex == QUEUE_NONE, ("bdirty: buffer %p still on queue %d", bp, bp->b_qindex)); BUF_ASSERT_HELD(bp); bp->b_flags &= ~(B_RELBUF); bp->b_iocmd = BIO_WRITE; if ((bp->b_flags & B_DELWRI) == 0) { bp->b_flags |= /* XXX B_DONE | */ B_DELWRI; reassignbuf(bp); bdirtyadd(); } } /* * bundirty: * * Clear B_DELWRI for buffer. * * Since the buffer is not on a queue, we do not update the numfreebuffers * count. * * The buffer must be on QUEUE_NONE. */ void bundirty(struct buf *bp) { CTR3(KTR_BUF, "bundirty(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp)); KASSERT(bp->b_flags & B_REMFREE || bp->b_qindex == QUEUE_NONE, ("bundirty: buffer %p still on queue %d", bp, bp->b_qindex)); BUF_ASSERT_HELD(bp); if (bp->b_flags & B_DELWRI) { bp->b_flags &= ~B_DELWRI; reassignbuf(bp); bdirtysub(); } /* * Since it is now being written, we can clear its deferred write flag. */ bp->b_flags &= ~B_DEFERRED; } /* * bawrite: * * Asynchronous write. Start output on a buffer, but do not wait for * it to complete. The buffer is released when the output completes. * * bwrite() ( or the VOP routine anyway ) is responsible for handling * B_INVAL buffers. Not us. */ void bawrite(struct buf *bp) { bp->b_flags |= B_ASYNC; (void) bwrite(bp); } /* * babarrierwrite: * * Asynchronous barrier write. Start output on a buffer, but do not * wait for it to complete. Place a write barrier after this write so * that this buffer and all buffers written before it are committed to * the disk before any buffers written after this write are committed * to the disk. The buffer is released when the output completes. */ void babarrierwrite(struct buf *bp) { bp->b_flags |= B_ASYNC | B_BARRIER; (void) bwrite(bp); } /* * bbarrierwrite: * * Synchronous barrier write. Start output on a buffer and wait for * it to complete. Place a write barrier after this write so that * this buffer and all buffers written before it are committed to * the disk before any buffers written after this write are committed * to the disk. The buffer is released when the output completes. */ int bbarrierwrite(struct buf *bp) { bp->b_flags |= B_BARRIER; return (bwrite(bp)); } /* * bwillwrite: * * Called prior to the locking of any vnodes when we are expecting to * write. We do not want to starve the buffer cache with too many * dirty buffers so we block here. By blocking prior to the locking * of any vnodes we attempt to avoid the situation where a locked vnode * prevents the various system daemons from flushing related buffers. */ void bwillwrite(void) { if (numdirtybuffers >= hidirtybuffers) { mtx_lock(&bdirtylock); while (numdirtybuffers >= hidirtybuffers) { bdirtywait = 1; msleep(&bdirtywait, &bdirtylock, (PRIBIO + 4), "flswai", 0); } mtx_unlock(&bdirtylock); } } /* * Return true if we have too many dirty buffers. */ int buf_dirty_count_severe(void) { return(numdirtybuffers >= hidirtybuffers); } /* * brelse: * * Release a busy buffer and, if requested, free its resources. The * buffer will be stashed in the appropriate bufqueue[] allowing it * to be accessed later as a cache entity or reused for other purposes. */ void brelse(struct buf *bp) { int qindex; CTR3(KTR_BUF, "brelse(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(!(bp->b_flags & (B_CLUSTER|B_PAGING)), ("brelse: inappropriate B_PAGING or B_CLUSTER bp %p", bp)); + KASSERT((bp->b_flags & B_VMIO) != 0 || (bp->b_flags & B_NOREUSE) == 0, + ("brelse: non-VMIO buffer marked NOREUSE")); if (BUF_LOCKRECURSED(bp)) { /* * Do not process, in particular, do not handle the * B_INVAL/B_RELBUF and do not release to free list. */ BUF_UNLOCK(bp); return; } if (bp->b_flags & B_MANAGED) { bqrelse(bp); return; } if ((bp->b_vflags & (BV_BKGRDINPROG | BV_BKGRDERR)) == BV_BKGRDERR) { BO_LOCK(bp->b_bufobj); bp->b_vflags &= ~BV_BKGRDERR; BO_UNLOCK(bp->b_bufobj); bdirty(bp); } if (bp->b_iocmd == BIO_WRITE && (bp->b_ioflags & BIO_ERROR) && bp->b_error == EIO && !(bp->b_flags & B_INVAL)) { /* * Failed write, redirty. Must clear BIO_ERROR to prevent * pages from being scrapped. If the error is anything * other than an I/O error (EIO), assume that retrying * is futile. */ bp->b_ioflags &= ~BIO_ERROR; bdirty(bp); } else if ((bp->b_flags & (B_NOCACHE | B_INVAL)) || (bp->b_ioflags & BIO_ERROR) || (bp->b_bufsize <= 0)) { /* * Either a failed I/O or we were asked to free or not * cache the buffer. */ bp->b_flags |= B_INVAL; if (!LIST_EMPTY(&bp->b_dep)) buf_deallocate(bp); if (bp->b_flags & B_DELWRI) bdirtysub(); bp->b_flags &= ~(B_DELWRI | B_CACHE); if ((bp->b_flags & B_VMIO) == 0) { allocbuf(bp, 0); if (bp->b_vp) brelvp(bp); } } /* * We must clear B_RELBUF if B_DELWRI is set. If vfs_vmio_truncate() * is called with B_DELWRI set, the underlying pages may wind up * getting freed causing a previous write (bdwrite()) to get 'lost' * because pages associated with a B_DELWRI bp are marked clean. * * We still allow the B_INVAL case to call vfs_vmio_truncate(), even * if B_DELWRI is set. */ if (bp->b_flags & B_DELWRI) bp->b_flags &= ~B_RELBUF; /* * VMIO buffer rundown. It is not very necessary to keep a VMIO buffer * constituted, not even NFS buffers now. Two flags effect this. If * B_INVAL, the struct buf is invalidated but the VM object is kept * around ( i.e. so it is trivial to reconstitute the buffer later ). * * If BIO_ERROR or B_NOCACHE is set, pages in the VM object will be * invalidated. BIO_ERROR cannot be set for a failed write unless the * buffer is also B_INVAL because it hits the re-dirtying code above. * * Normally we can do this whether a buffer is B_DELWRI or not. If * the buffer is an NFS buffer, it is tracking piecemeal writes or * the commit state and we cannot afford to lose the buffer. If the * buffer has a background write in progress, we need to keep it * around to prevent it from being reconstituted and starting a second * background write. */ if ((bp->b_flags & B_VMIO) && (bp->b_flags & B_NOCACHE || (bp->b_ioflags & BIO_ERROR && bp->b_iocmd == BIO_READ)) && !(bp->b_vp->v_mount != NULL && (bp->b_vp->v_mount->mnt_vfc->vfc_flags & VFCF_NETWORK) != 0 && !vn_isdisk(bp->b_vp, NULL) && (bp->b_flags & B_DELWRI))) { vfs_vmio_invalidate(bp); allocbuf(bp, 0); } - if ((bp->b_flags & (B_INVAL | B_RELBUF)) != 0) { + if ((bp->b_flags & (B_INVAL | B_RELBUF)) != 0 || + (bp->b_flags & (B_DELWRI | B_NOREUSE)) == B_NOREUSE) { allocbuf(bp, 0); + bp->b_flags &= ~B_NOREUSE; if (bp->b_vp != NULL) brelvp(bp); } /* * If the buffer has junk contents signal it and eventually * clean up B_DELWRI and diassociate the vnode so that gbincore() * doesn't find it. */ if (bp->b_bufsize == 0 || (bp->b_ioflags & BIO_ERROR) != 0 || (bp->b_flags & (B_INVAL | B_NOCACHE | B_RELBUF)) != 0) bp->b_flags |= B_INVAL; if (bp->b_flags & B_INVAL) { if (bp->b_flags & B_DELWRI) bundirty(bp); if (bp->b_vp) brelvp(bp); } /* buffers with no memory */ if (bp->b_bufsize == 0) { bp->b_xflags &= ~(BX_BKGRDWRITE | BX_ALTDATA); if (bp->b_vflags & BV_BKGRDINPROG) panic("losing buffer 1"); bufkvafree(bp); qindex = QUEUE_EMPTY; bp->b_flags |= B_AGE; /* buffers with junk contents */ } else if (bp->b_flags & (B_INVAL | B_NOCACHE | B_RELBUF) || (bp->b_ioflags & BIO_ERROR)) { bp->b_xflags &= ~(BX_BKGRDWRITE | BX_ALTDATA); if (bp->b_vflags & BV_BKGRDINPROG) panic("losing buffer 2"); qindex = QUEUE_CLEAN; bp->b_flags |= B_AGE; /* remaining buffers */ } else if (bp->b_flags & B_DELWRI) qindex = QUEUE_DIRTY; else qindex = QUEUE_CLEAN; binsfree(bp, qindex); bp->b_flags &= ~(B_ASYNC | B_NOCACHE | B_AGE | B_RELBUF | B_DIRECT); if ((bp->b_flags & B_DELWRI) == 0 && (bp->b_xflags & BX_VNDIRTY)) panic("brelse: not dirty"); /* unlock */ BUF_UNLOCK(bp); } /* * Release a buffer back to the appropriate queue but do not try to free * it. The buffer is expected to be used again soon. * * bqrelse() is used by bdwrite() to requeue a delayed write, and used by * biodone() to requeue an async I/O on completion. It is also used when * known good buffers need to be requeued but we think we may need the data * again soon. * * XXX we should be able to leave the B_RELBUF hint set on completion. */ void bqrelse(struct buf *bp) { int qindex; CTR3(KTR_BUF, "bqrelse(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(!(bp->b_flags & (B_CLUSTER|B_PAGING)), ("bqrelse: inappropriate B_PAGING or B_CLUSTER bp %p", bp)); if (BUF_LOCKRECURSED(bp)) { /* do not release to free list */ BUF_UNLOCK(bp); return; } bp->b_flags &= ~(B_ASYNC | B_NOCACHE | B_AGE | B_RELBUF); if (bp->b_flags & B_MANAGED) { if (bp->b_flags & B_REMFREE) bremfreef(bp); goto out; } /* buffers with stale but valid contents */ if ((bp->b_flags & B_DELWRI) != 0 || (bp->b_vflags & (BV_BKGRDINPROG | BV_BKGRDERR)) == BV_BKGRDERR) { BO_LOCK(bp->b_bufobj); bp->b_vflags &= ~BV_BKGRDERR; BO_UNLOCK(bp->b_bufobj); qindex = QUEUE_DIRTY; } else { if ((bp->b_flags & B_DELWRI) == 0 && (bp->b_xflags & BX_VNDIRTY)) panic("bqrelse: not dirty"); + if ((bp->b_flags & B_NOREUSE) != 0) { + brelse(bp); + return; + } qindex = QUEUE_CLEAN; } binsfree(bp, qindex); out: /* unlock */ BUF_UNLOCK(bp); } /* * Complete I/O to a VMIO backed page. Validate the pages as appropriate, * restore bogus pages. */ static void vfs_vmio_iodone(struct buf *bp) { vm_ooffset_t foff; vm_page_t m; vm_object_t obj; struct vnode *vp; int bogus, i, iosize; obj = bp->b_bufobj->bo_object; KASSERT(obj->paging_in_progress >= bp->b_npages, ("vfs_vmio_iodone: paging in progress(%d) < b_npages(%d)", obj->paging_in_progress, bp->b_npages)); vp = bp->b_vp; KASSERT(vp->v_holdcnt > 0, ("vfs_vmio_iodone: vnode %p has zero hold count", vp)); KASSERT(vp->v_object != NULL, ("vfs_vmio_iodone: vnode %p has no vm_object", vp)); foff = bp->b_offset; KASSERT(bp->b_offset != NOOFFSET, ("vfs_vmio_iodone: bp %p has no buffer offset", bp)); bogus = 0; iosize = bp->b_bcount - bp->b_resid; VM_OBJECT_WLOCK(obj); for (i = 0; i < bp->b_npages; i++) { int resid; resid = ((foff + PAGE_SIZE) & ~(off_t)PAGE_MASK) - foff; if (resid > iosize) resid = iosize; /* * cleanup bogus pages, restoring the originals */ m = bp->b_pages[i]; if (m == bogus_page) { bogus = 1; m = vm_page_lookup(obj, OFF_TO_IDX(foff)); if (m == NULL) panic("biodone: page disappeared!"); bp->b_pages[i] = m; } else if ((bp->b_iocmd == BIO_READ) && resid > 0) { /* * In the write case, the valid and clean bits are * already changed correctly ( see bdwrite() ), so we * only need to do this here in the read case. */ KASSERT((m->dirty & vm_page_bits(foff & PAGE_MASK, resid)) == 0, ("vfs_vmio_iodone: page %p " "has unexpected dirty bits", m)); vfs_page_set_valid(bp, foff, m); } KASSERT(OFF_TO_IDX(foff) == m->pindex, ("vfs_vmio_iodone: foff(%jd)/pindex(%ju) mismatch", (intmax_t)foff, (uintmax_t)m->pindex)); vm_page_sunbusy(m); vm_object_pip_subtract(obj, 1); foff = (foff + PAGE_SIZE) & ~(off_t)PAGE_MASK; iosize -= resid; } vm_object_pip_wakeupn(obj, 0); VM_OBJECT_WUNLOCK(obj); if (bogus && buf_mapped(bp)) { BUF_CHECK_MAPPED(bp); pmap_qenter(trunc_page((vm_offset_t)bp->b_data), bp->b_pages, bp->b_npages); } } /* * Unwire a page held by a buf and place it on the appropriate vm queue. */ static void vfs_vmio_unwire(struct buf *bp, vm_page_t m) { bool freed; vm_page_lock(m); if (vm_page_unwire(m, PQ_NONE)) { /* * Determine if the page should be freed before adding * it to the inactive queue. */ if (m->valid == 0) { freed = !vm_page_busied(m); if (freed) vm_page_free(m); } else if ((bp->b_flags & B_DIRECT) != 0) freed = vm_page_try_to_free(m); else freed = false; if (!freed) { /* - * In order to maintain LRU page ordering, put - * the page at the tail of the inactive queue. + * If the page is unlikely to be reused, let the + * VM know. Otherwise, maintain LRU page + * ordering and put the page at the tail of the + * inactive queue. */ - vm_page_deactivate(m); + if ((bp->b_flags & B_NOREUSE) != 0) + vm_page_deactivate_noreuse(m); + else + vm_page_deactivate(m); } } vm_page_unlock(m); } /* * Perform page invalidation when a buffer is released. The fully invalid * pages will be reclaimed later in vfs_vmio_truncate(). */ static void vfs_vmio_invalidate(struct buf *bp) { vm_object_t obj; vm_page_t m; int i, resid, poffset, presid; if (buf_mapped(bp)) { BUF_CHECK_MAPPED(bp); pmap_qremove(trunc_page((vm_offset_t)bp->b_data), bp->b_npages); } else BUF_CHECK_UNMAPPED(bp); /* * Get the base offset and length of the buffer. Note that * in the VMIO case if the buffer block size is not * page-aligned then b_data pointer may not be page-aligned. * But our b_pages[] array *IS* page aligned. * * block sizes less then DEV_BSIZE (usually 512) are not * supported due to the page granularity bits (m->valid, * m->dirty, etc...). * * See man buf(9) for more information */ obj = bp->b_bufobj->bo_object; resid = bp->b_bufsize; poffset = bp->b_offset & PAGE_MASK; VM_OBJECT_WLOCK(obj); for (i = 0; i < bp->b_npages; i++) { m = bp->b_pages[i]; if (m == bogus_page) panic("vfs_vmio_invalidate: Unexpected bogus page."); bp->b_pages[i] = NULL; presid = resid > (PAGE_SIZE - poffset) ? (PAGE_SIZE - poffset) : resid; KASSERT(presid >= 0, ("brelse: extra page")); while (vm_page_xbusied(m)) { vm_page_lock(m); VM_OBJECT_WUNLOCK(obj); vm_page_busy_sleep(m, "mbncsh"); VM_OBJECT_WLOCK(obj); } if (pmap_page_wired_mappings(m) == 0) vm_page_set_invalid(m, poffset, presid); vfs_vmio_unwire(bp, m); resid -= presid; poffset = 0; } VM_OBJECT_WUNLOCK(obj); bp->b_npages = 0; } /* * Page-granular truncation of an existing VMIO buffer. */ static void vfs_vmio_truncate(struct buf *bp, int desiredpages) { vm_object_t obj; vm_page_t m; int i; if (bp->b_npages == desiredpages) return; if (buf_mapped(bp)) { BUF_CHECK_MAPPED(bp); pmap_qremove((vm_offset_t)trunc_page((vm_offset_t)bp->b_data) + (desiredpages << PAGE_SHIFT), bp->b_npages - desiredpages); } else BUF_CHECK_UNMAPPED(bp); obj = bp->b_bufobj->bo_object; if (obj != NULL) VM_OBJECT_WLOCK(obj); for (i = desiredpages; i < bp->b_npages; i++) { m = bp->b_pages[i]; KASSERT(m != bogus_page, ("allocbuf: bogus page found")); bp->b_pages[i] = NULL; vfs_vmio_unwire(bp, m); } if (obj != NULL) VM_OBJECT_WUNLOCK(obj); bp->b_npages = desiredpages; } /* * Byte granular extension of VMIO buffers. */ static void vfs_vmio_extend(struct buf *bp, int desiredpages, int size) { /* * We are growing the buffer, possibly in a * byte-granular fashion. */ vm_object_t obj; vm_offset_t toff; vm_offset_t tinc; vm_page_t m; /* * Step 1, bring in the VM pages from the object, allocating * them if necessary. We must clear B_CACHE if these pages * are not valid for the range covered by the buffer. */ obj = bp->b_bufobj->bo_object; VM_OBJECT_WLOCK(obj); while (bp->b_npages < desiredpages) { /* * We must allocate system pages since blocking * here could interfere with paging I/O, no * matter which process we are. * * Only exclusive busy can be tested here. * Blocking on shared busy might lead to * deadlocks once allocbuf() is called after * pages are vfs_busy_pages(). */ m = vm_page_grab(obj, OFF_TO_IDX(bp->b_offset) + bp->b_npages, VM_ALLOC_NOBUSY | VM_ALLOC_SYSTEM | VM_ALLOC_WIRED | VM_ALLOC_IGN_SBUSY | VM_ALLOC_COUNT(desiredpages - bp->b_npages)); if (m->valid == 0) bp->b_flags &= ~B_CACHE; bp->b_pages[bp->b_npages] = m; ++bp->b_npages; } /* * Step 2. We've loaded the pages into the buffer, * we have to figure out if we can still have B_CACHE * set. Note that B_CACHE is set according to the * byte-granular range ( bcount and size ), not the * aligned range ( newbsize ). * * The VM test is against m->valid, which is DEV_BSIZE * aligned. Needless to say, the validity of the data * needs to also be DEV_BSIZE aligned. Note that this * fails with NFS if the server or some other client * extends the file's EOF. If our buffer is resized, * B_CACHE may remain set! XXX */ toff = bp->b_bcount; tinc = PAGE_SIZE - ((bp->b_offset + toff) & PAGE_MASK); while ((bp->b_flags & B_CACHE) && toff < size) { vm_pindex_t pi; if (tinc > (size - toff)) tinc = size - toff; pi = ((bp->b_offset & PAGE_MASK) + toff) >> PAGE_SHIFT; m = bp->b_pages[pi]; vfs_buf_test_cache(bp, bp->b_offset, toff, tinc, m); toff += tinc; tinc = PAGE_SIZE; } VM_OBJECT_WUNLOCK(obj); /* * Step 3, fixup the KVA pmap. */ if (buf_mapped(bp)) bpmap_qenter(bp); else BUF_CHECK_UNMAPPED(bp); } /* * Check to see if a block at a particular lbn is available for a clustered * write. */ static int vfs_bio_clcheck(struct vnode *vp, int size, daddr_t lblkno, daddr_t blkno) { struct buf *bpa; int match; match = 0; /* If the buf isn't in core skip it */ if ((bpa = gbincore(&vp->v_bufobj, lblkno)) == NULL) return (0); /* If the buf is busy we don't want to wait for it */ if (BUF_LOCK(bpa, LK_EXCLUSIVE | LK_NOWAIT, NULL) != 0) return (0); /* Only cluster with valid clusterable delayed write buffers */ if ((bpa->b_flags & (B_DELWRI | B_CLUSTEROK | B_INVAL)) != (B_DELWRI | B_CLUSTEROK)) goto done; if (bpa->b_bufsize != size) goto done; /* * Check to see if it is in the expected place on disk and that the * block has been mapped. */ if ((bpa->b_blkno != bpa->b_lblkno) && (bpa->b_blkno == blkno)) match = 1; done: BUF_UNLOCK(bpa); return (match); } /* * vfs_bio_awrite: * * Implement clustered async writes for clearing out B_DELWRI buffers. * This is much better then the old way of writing only one buffer at * a time. Note that we may not be presented with the buffers in the * correct order, so we search for the cluster in both directions. */ int vfs_bio_awrite(struct buf *bp) { struct bufobj *bo; int i; int j; daddr_t lblkno = bp->b_lblkno; struct vnode *vp = bp->b_vp; int ncl; int nwritten; int size; int maxcl; int gbflags; bo = &vp->v_bufobj; gbflags = (bp->b_data == unmapped_buf) ? GB_UNMAPPED : 0; /* * right now we support clustered writing only to regular files. If * we find a clusterable block we could be in the middle of a cluster * rather then at the beginning. */ if ((vp->v_type == VREG) && (vp->v_mount != 0) && /* Only on nodes that have the size info */ (bp->b_flags & (B_CLUSTEROK | B_INVAL)) == B_CLUSTEROK) { size = vp->v_mount->mnt_stat.f_iosize; maxcl = MAXPHYS / size; BO_RLOCK(bo); for (i = 1; i < maxcl; i++) if (vfs_bio_clcheck(vp, size, lblkno + i, bp->b_blkno + ((i * size) >> DEV_BSHIFT)) == 0) break; for (j = 1; i + j <= maxcl && j <= lblkno; j++) if (vfs_bio_clcheck(vp, size, lblkno - j, bp->b_blkno - ((j * size) >> DEV_BSHIFT)) == 0) break; BO_RUNLOCK(bo); --j; ncl = i + j; /* * this is a possible cluster write */ if (ncl != 1) { BUF_UNLOCK(bp); nwritten = cluster_wbuild(vp, size, lblkno - j, ncl, gbflags); return (nwritten); } } bremfree(bp); bp->b_flags |= B_ASYNC; /* * default (old) behavior, writing out only one block * * XXX returns b_bufsize instead of b_bcount for nwritten? */ nwritten = bp->b_bufsize; (void) bwrite(bp); return (nwritten); } /* * Ask the bufdaemon for help, or act as bufdaemon itself, when a * locked vnode is supplied. */ static void getnewbuf_bufd_help(struct vnode *vp, int gbflags, int slpflag, int slptimeo, int defrag) { struct thread *td; char *waitmsg; int error, fl, flags, norunbuf; mtx_assert(&bqclean, MA_OWNED); if (defrag) { flags = VFS_BIO_NEED_BUFSPACE; waitmsg = "nbufkv"; } else if (bufspace >= hibufspace) { waitmsg = "nbufbs"; flags = VFS_BIO_NEED_BUFSPACE; } else { waitmsg = "newbuf"; flags = VFS_BIO_NEED_ANY; } atomic_set_int(&needsbuffer, flags); mtx_unlock(&bqclean); bd_speedup(); /* heeeelp */ if ((gbflags & GB_NOWAIT_BD) != 0) return; td = curthread; rw_wlock(&nblock); while ((needsbuffer & flags) != 0) { if (vp != NULL && vp->v_type != VCHR && (td->td_pflags & TDP_BUFNEED) == 0) { rw_wunlock(&nblock); /* * getblk() is called with a vnode locked, and * some majority of the dirty buffers may as * well belong to the vnode. Flushing the * buffers there would make a progress that * cannot be achieved by the buf_daemon, that * cannot lock the vnode. */ norunbuf = ~(TDP_BUFNEED | TDP_NORUNNINGBUF) | (td->td_pflags & TDP_NORUNNINGBUF); /* * Play bufdaemon. The getnewbuf() function * may be called while the thread owns lock * for another dirty buffer for the same * vnode, which makes it impossible to use * VOP_FSYNC() there, due to the buffer lock * recursion. */ td->td_pflags |= TDP_BUFNEED | TDP_NORUNNINGBUF; fl = buf_flush(vp, flushbufqtarget); td->td_pflags &= norunbuf; rw_wlock(&nblock); if (fl != 0) continue; if ((needsbuffer & flags) == 0) break; } error = rw_sleep(__DEVOLATILE(void *, &needsbuffer), &nblock, (PRIBIO + 4) | slpflag, waitmsg, slptimeo); if (error != 0) break; } rw_wunlock(&nblock); } static void getnewbuf_reuse_bp(struct buf *bp, int qindex) { CTR6(KTR_BUF, "getnewbuf(%p) vp %p flags %X kvasize %d bufsize %d " "queue %d (recycling)", bp, bp->b_vp, bp->b_flags, bp->b_kvasize, bp->b_bufsize, qindex); mtx_assert(&bqclean, MA_NOTOWNED); /* * Note: we no longer distinguish between VMIO and non-VMIO * buffers. */ - KASSERT((bp->b_flags & B_DELWRI) == 0, - ("delwri buffer %p found in queue %d", bp, qindex)); + KASSERT((bp->b_flags & (B_DELWRI | B_NOREUSE)) == 0, + ("invalid buffer %p flags %#x found in queue %d", bp, bp->b_flags, + qindex)); /* * When recycling a clean buffer we have to truncate it and * release the vnode. */ if (qindex == QUEUE_CLEAN) { allocbuf(bp, 0); if (bp->b_vp != NULL) brelvp(bp); } /* * Get the rest of the buffer freed up. b_kva* is still valid * after this operation. */ if (bp->b_rcred != NOCRED) { crfree(bp->b_rcred); bp->b_rcred = NOCRED; } if (bp->b_wcred != NOCRED) { crfree(bp->b_wcred); bp->b_wcred = NOCRED; } if (!LIST_EMPTY(&bp->b_dep)) buf_deallocate(bp); if (bp->b_vflags & BV_BKGRDINPROG) panic("losing buffer 3"); KASSERT(bp->b_vp == NULL, ("bp: %p still has vnode %p. qindex: %d", bp, bp->b_vp, qindex)); KASSERT((bp->b_xflags & (BX_VNCLEAN|BX_VNDIRTY)) == 0, ("bp: %p still on a buffer list. xflags %X", bp, bp->b_xflags)); KASSERT(bp->b_npages == 0, ("bp: %p still has %d vm pages\n", bp, bp->b_npages)); bp->b_flags = 0; bp->b_ioflags = 0; bp->b_xflags = 0; KASSERT((bp->b_flags & B_INFREECNT) == 0, ("buf %p still counted as free?", bp)); bp->b_vflags = 0; bp->b_vp = NULL; bp->b_blkno = bp->b_lblkno = 0; bp->b_offset = NOOFFSET; bp->b_iodone = 0; bp->b_error = 0; bp->b_resid = 0; bp->b_bcount = 0; bp->b_npages = 0; bp->b_dirtyoff = bp->b_dirtyend = 0; bp->b_bufobj = NULL; bp->b_pin_count = 0; bp->b_data = bp->b_kvabase; bp->b_fsprivate1 = NULL; bp->b_fsprivate2 = NULL; bp->b_fsprivate3 = NULL; LIST_INIT(&bp->b_dep); } static struct buf * getnewbuf_scan(int maxsize, int defrag, int unmapped, int metadata) { struct buf *bp, *nbp; int nqindex, qindex, pass; KASSERT(!unmapped || !defrag, ("both unmapped and defrag")); pass = 0; restart: if (pass != 0) atomic_add_int(&getnewbufrestarts, 1); nbp = NULL; mtx_lock(&bqclean); /* * If we're not defragging or low on bufspace attempt to make a new * buf from a header. */ if (defrag == 0 && bufspace + maxsize < hibufspace) { nqindex = QUEUE_EMPTY; nbp = TAILQ_FIRST(&bufqueues[nqindex]); } /* * All available buffers might be clean or we need to start recycling. */ if (nbp == NULL) { nqindex = QUEUE_CLEAN; nbp = TAILQ_FIRST(&bufqueues[QUEUE_CLEAN]); } /* * Run scan, possibly freeing data and/or kva mappings on the fly * depending. */ while ((bp = nbp) != NULL) { qindex = nqindex; /* * Calculate next bp (we can only use it if we do not * release the bqlock) */ if ((nbp = TAILQ_NEXT(bp, b_freelist)) == NULL) { switch (qindex) { case QUEUE_EMPTY: nqindex = QUEUE_CLEAN; nbp = TAILQ_FIRST(&bufqueues[nqindex]); if (nbp != NULL) break; /* FALLTHROUGH */ case QUEUE_CLEAN: if (metadata && pass == 0) { pass = 1; nqindex = QUEUE_EMPTY; nbp = TAILQ_FIRST(&bufqueues[nqindex]); } /* * nbp is NULL. */ break; } } /* * If we are defragging then we need a buffer with * b_kvasize != 0. This situation occurs when we * have many unmapped bufs. */ if (defrag && bp->b_kvasize == 0) continue; /* * Start freeing the bp. This is somewhat involved. nbp * remains valid only for QUEUE_EMPTY[KVA] bp's. */ if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL) != 0) continue; /* * BKGRDINPROG can only be set with the buf and bufobj * locks both held. We tolerate a race to clear it here. */ if (bp->b_vflags & BV_BKGRDINPROG) { BUF_UNLOCK(bp); continue; } /* * Requeue the background write buffer with error. */ if ((bp->b_vflags & BV_BKGRDERR) != 0) { bremfreel(bp); mtx_unlock(&bqclean); bqrelse(bp); continue; } KASSERT(bp->b_qindex == qindex, ("getnewbuf: inconsistent queue %d bp %p", qindex, bp)); bremfreel(bp); mtx_unlock(&bqclean); /* * NOTE: nbp is now entirely invalid. We can only restart * the scan from this point on. */ getnewbuf_reuse_bp(bp, qindex); mtx_assert(&bqclean, MA_NOTOWNED); /* * If we are defragging then free the buffer. */ if (defrag) { bp->b_flags |= B_INVAL; brelse(bp); defrag = 0; goto restart; } /* * Notify any waiters for the buffer lock about * identity change by freeing the buffer. */ if (qindex == QUEUE_CLEAN && BUF_LOCKWAITERS(bp)) { bp->b_flags |= B_INVAL; brelse(bp); goto restart; } if (metadata) break; /* * If we are overcomitted then recover the buffer and its * KVM space. This occurs in rare situations when multiple * processes are blocked in getnewbuf() or allocbuf(). */ if (bufspace >= hibufspace && bp->b_kvasize != 0) { bp->b_flags |= B_INVAL; brelse(bp); goto restart; } break; } return (bp); } /* * getnewbuf: * * Find and initialize a new buffer header, freeing up existing buffers * in the bufqueues as necessary. The new buffer is returned locked. * * Important: B_INVAL is not set. If the caller wishes to throw the * buffer away, the caller must set B_INVAL prior to calling brelse(). * * We block if: * We have insufficient buffer headers * We have insufficient buffer space * buffer_arena is too fragmented ( space reservation fails ) * If we have to flush dirty buffers ( but we try to avoid this ) */ static struct buf * getnewbuf(struct vnode *vp, int slpflag, int slptimeo, int size, int maxsize, int gbflags) { struct buf *bp; int defrag, metadata; KASSERT((gbflags & (GB_UNMAPPED | GB_KVAALLOC)) != GB_KVAALLOC, ("GB_KVAALLOC only makes sense with GB_UNMAPPED")); if (!unmapped_buf_allowed) gbflags &= ~(GB_UNMAPPED | GB_KVAALLOC); defrag = 0; if (vp == NULL || (vp->v_vflag & (VV_MD | VV_SYSTEM)) != 0 || vp->v_type == VCHR) metadata = 1; else metadata = 0; /* * We can't afford to block since we might be holding a vnode lock, * which may prevent system daemons from running. We deal with * low-memory situations by proactively returning memory and running * async I/O rather then sync I/O. */ atomic_add_int(&getnewbufcalls, 1); restart: bp = getnewbuf_scan(maxsize, defrag, (gbflags & (GB_UNMAPPED | GB_KVAALLOC)) == GB_UNMAPPED, metadata); if (bp != NULL) defrag = 0; /* * If we exhausted our list, sleep as appropriate. We may have to * wakeup various daemons and write out some dirty buffers. * * Generally we are sleeping due to insufficient buffer space. */ if (bp == NULL) { mtx_assert(&bqclean, MA_OWNED); getnewbuf_bufd_help(vp, gbflags, slpflag, slptimeo, defrag); mtx_assert(&bqclean, MA_NOTOWNED); } else if ((gbflags & (GB_UNMAPPED | GB_KVAALLOC)) == GB_UNMAPPED) { mtx_assert(&bqclean, MA_NOTOWNED); bufkvafree(bp); atomic_add_int(&bufreusecnt, 1); } else { mtx_assert(&bqclean, MA_NOTOWNED); /* * We finally have a valid bp. We aren't quite out of the * woods, we still have to reserve kva space. In order to * keep fragmentation sane we only allocate kva in BKVASIZE * chunks. */ maxsize = (maxsize + BKVAMASK) & ~BKVAMASK; if (maxsize != bp->b_kvasize && bufkvaalloc(bp, maxsize, gbflags)) { defrag = 1; bp->b_flags |= B_INVAL; brelse(bp); goto restart; } else if ((gbflags & (GB_UNMAPPED | GB_KVAALLOC)) == (GB_UNMAPPED | GB_KVAALLOC)) { bp->b_data = unmapped_buf; BUF_CHECK_UNMAPPED(bp); } atomic_add_int(&bufreusecnt, 1); } return (bp); } /* * buf_daemon: * * buffer flushing daemon. Buffers are normally flushed by the * update daemon but if it cannot keep up this process starts to * take the load in an attempt to prevent getnewbuf() from blocking. */ static struct kproc_desc buf_kp = { "bufdaemon", buf_daemon, &bufdaemonproc }; SYSINIT(bufdaemon, SI_SUB_KTHREAD_BUF, SI_ORDER_FIRST, kproc_start, &buf_kp); static int buf_flush(struct vnode *vp, int target) { int flushed; flushed = flushbufqueues(vp, target, 0); if (flushed == 0) { /* * Could not find any buffers without rollback * dependencies, so just write the first one * in the hopes of eventually making progress. */ if (vp != NULL && target > 2) target /= 2; flushbufqueues(vp, target, 1); } return (flushed); } static void buf_daemon() { int lodirty; /* * This process needs to be suspended prior to shutdown sync. */ EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, bufdaemonproc, SHUTDOWN_PRI_LAST); /* * This process is allowed to take the buffer cache to the limit */ curthread->td_pflags |= TDP_NORUNNINGBUF | TDP_BUFNEED; mtx_lock(&bdlock); for (;;) { bd_request = 0; mtx_unlock(&bdlock); kproc_suspend_check(bufdaemonproc); lodirty = lodirtybuffers; if (bd_speedupreq) { lodirty = numdirtybuffers / 2; bd_speedupreq = 0; } /* * Do the flush. Limit the amount of in-transit I/O we * allow to build up, otherwise we would completely saturate * the I/O system. */ while (numdirtybuffers > lodirty) { if (buf_flush(NULL, numdirtybuffers - lodirty) == 0) break; kern_yield(PRI_USER); } /* * Only clear bd_request if we have reached our low water * mark. The buf_daemon normally waits 1 second and * then incrementally flushes any dirty buffers that have * built up, within reason. * * If we were unable to hit our low water mark and couldn't * find any flushable buffers, we sleep for a short period * to avoid endless loops on unlockable buffers. */ mtx_lock(&bdlock); if (numdirtybuffers <= lodirtybuffers) { /* * We reached our low water mark, reset the * request and sleep until we are needed again. * The sleep is just so the suspend code works. */ bd_request = 0; /* * Do an extra wakeup in case dirty threshold * changed via sysctl and the explicit transition * out of shortfall was missed. */ bdirtywakeup(); if (runningbufspace <= lorunningspace) runningwakeup(); msleep(&bd_request, &bdlock, PVM, "psleep", hz); } else { /* * We couldn't find any flushable dirty buffers but * still have too many dirty buffers, we * have to sleep and try again. (rare) */ msleep(&bd_request, &bdlock, PVM, "qsleep", hz / 10); } } } /* * flushbufqueues: * * Try to flush a buffer in the dirty queue. We must be careful to * free up B_INVAL buffers instead of write them, which NFS is * particularly sensitive to. */ static int flushwithdeps = 0; SYSCTL_INT(_vfs, OID_AUTO, flushwithdeps, CTLFLAG_RW, &flushwithdeps, 0, "Number of buffers flushed with dependecies that require rollbacks"); static int flushbufqueues(struct vnode *lvp, int target, int flushdeps) { struct buf *sentinel; struct vnode *vp; struct mount *mp; struct buf *bp; int hasdeps; int flushed; int queue; int error; bool unlock; flushed = 0; queue = QUEUE_DIRTY; bp = NULL; sentinel = malloc(sizeof(struct buf), M_TEMP, M_WAITOK | M_ZERO); sentinel->b_qindex = QUEUE_SENTINEL; mtx_lock(&bqdirty); TAILQ_INSERT_HEAD(&bufqueues[queue], sentinel, b_freelist); mtx_unlock(&bqdirty); while (flushed != target) { maybe_yield(); mtx_lock(&bqdirty); bp = TAILQ_NEXT(sentinel, b_freelist); if (bp != NULL) { TAILQ_REMOVE(&bufqueues[queue], sentinel, b_freelist); TAILQ_INSERT_AFTER(&bufqueues[queue], bp, sentinel, b_freelist); } else { mtx_unlock(&bqdirty); break; } /* * Skip sentinels inserted by other invocations of the * flushbufqueues(), taking care to not reorder them. * * Only flush the buffers that belong to the * vnode locked by the curthread. */ if (bp->b_qindex == QUEUE_SENTINEL || (lvp != NULL && bp->b_vp != lvp)) { mtx_unlock(&bqdirty); continue; } error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL); mtx_unlock(&bqdirty); if (error != 0) continue; if (bp->b_pin_count > 0) { BUF_UNLOCK(bp); continue; } /* * BKGRDINPROG can only be set with the buf and bufobj * locks both held. We tolerate a race to clear it here. */ if ((bp->b_vflags & BV_BKGRDINPROG) != 0 || (bp->b_flags & B_DELWRI) == 0) { BUF_UNLOCK(bp); continue; } if (bp->b_flags & B_INVAL) { bremfreef(bp); brelse(bp); flushed++; continue; } if (!LIST_EMPTY(&bp->b_dep) && buf_countdeps(bp, 0)) { if (flushdeps == 0) { BUF_UNLOCK(bp); continue; } hasdeps = 1; } else hasdeps = 0; /* * We must hold the lock on a vnode before writing * one of its buffers. Otherwise we may confuse, or * in the case of a snapshot vnode, deadlock the * system. * * The lock order here is the reverse of the normal * of vnode followed by buf lock. This is ok because * the NOWAIT will prevent deadlock. */ vp = bp->b_vp; if (vn_start_write(vp, &mp, V_NOWAIT) != 0) { BUF_UNLOCK(bp); continue; } if (lvp == NULL) { unlock = true; error = vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT); } else { ASSERT_VOP_LOCKED(vp, "getbuf"); unlock = false; error = VOP_ISLOCKED(vp) == LK_EXCLUSIVE ? 0 : vn_lock(vp, LK_TRYUPGRADE); } if (error == 0) { CTR3(KTR_BUF, "flushbufqueue(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); if (curproc == bufdaemonproc) { vfs_bio_awrite(bp); } else { bremfree(bp); bwrite(bp); notbufdflushes++; } vn_finished_write(mp); if (unlock) VOP_UNLOCK(vp, 0); flushwithdeps += hasdeps; flushed++; /* * Sleeping on runningbufspace while holding * vnode lock leads to deadlock. */ if (curproc == bufdaemonproc && runningbufspace > hirunningspace) waitrunningbufspace(); continue; } vn_finished_write(mp); BUF_UNLOCK(bp); } mtx_lock(&bqdirty); TAILQ_REMOVE(&bufqueues[queue], sentinel, b_freelist); mtx_unlock(&bqdirty); free(sentinel, M_TEMP); return (flushed); } /* * Check to see if a block is currently memory resident. */ struct buf * incore(struct bufobj *bo, daddr_t blkno) { struct buf *bp; BO_RLOCK(bo); bp = gbincore(bo, blkno); BO_RUNLOCK(bo); return (bp); } /* * Returns true if no I/O is needed to access the * associated VM object. This is like incore except * it also hunts around in the VM system for the data. */ static int inmem(struct vnode * vp, daddr_t blkno) { vm_object_t obj; vm_offset_t toff, tinc, size; vm_page_t m; vm_ooffset_t off; ASSERT_VOP_LOCKED(vp, "inmem"); if (incore(&vp->v_bufobj, blkno)) return 1; if (vp->v_mount == NULL) return 0; obj = vp->v_object; if (obj == NULL) return (0); size = PAGE_SIZE; if (size > vp->v_mount->mnt_stat.f_iosize) size = vp->v_mount->mnt_stat.f_iosize; off = (vm_ooffset_t)blkno * (vm_ooffset_t)vp->v_mount->mnt_stat.f_iosize; VM_OBJECT_RLOCK(obj); for (toff = 0; toff < vp->v_mount->mnt_stat.f_iosize; toff += tinc) { m = vm_page_lookup(obj, OFF_TO_IDX(off + toff)); if (!m) goto notinmem; tinc = size; if (tinc > PAGE_SIZE - ((toff + off) & PAGE_MASK)) tinc = PAGE_SIZE - ((toff + off) & PAGE_MASK); if (vm_page_is_valid(m, (vm_offset_t) ((toff + off) & PAGE_MASK), tinc) == 0) goto notinmem; } VM_OBJECT_RUNLOCK(obj); return 1; notinmem: VM_OBJECT_RUNLOCK(obj); return (0); } /* * Set the dirty range for a buffer based on the status of the dirty * bits in the pages comprising the buffer. The range is limited * to the size of the buffer. * * Tell the VM system that the pages associated with this buffer * are clean. This is used for delayed writes where the data is * going to go to disk eventually without additional VM intevention. * * Note that while we only really need to clean through to b_bcount, we * just go ahead and clean through to b_bufsize. */ static void vfs_clean_pages_dirty_buf(struct buf *bp) { vm_ooffset_t foff, noff, eoff; vm_page_t m; int i; if ((bp->b_flags & B_VMIO) == 0 || bp->b_bufsize == 0) return; foff = bp->b_offset; KASSERT(bp->b_offset != NOOFFSET, ("vfs_clean_pages_dirty_buf: no buffer offset")); VM_OBJECT_WLOCK(bp->b_bufobj->bo_object); vfs_drain_busy_pages(bp); vfs_setdirty_locked_object(bp); for (i = 0; i < bp->b_npages; i++) { noff = (foff + PAGE_SIZE) & ~(off_t)PAGE_MASK; eoff = noff; if (eoff > bp->b_offset + bp->b_bufsize) eoff = bp->b_offset + bp->b_bufsize; m = bp->b_pages[i]; vfs_page_set_validclean(bp, foff, m); /* vm_page_clear_dirty(m, foff & PAGE_MASK, eoff - foff); */ foff = noff; } VM_OBJECT_WUNLOCK(bp->b_bufobj->bo_object); } static void vfs_setdirty_locked_object(struct buf *bp) { vm_object_t object; int i; object = bp->b_bufobj->bo_object; VM_OBJECT_ASSERT_WLOCKED(object); /* * We qualify the scan for modified pages on whether the * object has been flushed yet. */ if ((object->flags & OBJ_MIGHTBEDIRTY) != 0) { vm_offset_t boffset; vm_offset_t eoffset; /* * test the pages to see if they have been modified directly * by users through the VM system. */ for (i = 0; i < bp->b_npages; i++) vm_page_test_dirty(bp->b_pages[i]); /* * Calculate the encompassing dirty range, boffset and eoffset, * (eoffset - boffset) bytes. */ for (i = 0; i < bp->b_npages; i++) { if (bp->b_pages[i]->dirty) break; } boffset = (i << PAGE_SHIFT) - (bp->b_offset & PAGE_MASK); for (i = bp->b_npages - 1; i >= 0; --i) { if (bp->b_pages[i]->dirty) { break; } } eoffset = ((i + 1) << PAGE_SHIFT) - (bp->b_offset & PAGE_MASK); /* * Fit it to the buffer. */ if (eoffset > bp->b_bcount) eoffset = bp->b_bcount; /* * If we have a good dirty range, merge with the existing * dirty range. */ if (boffset < eoffset) { if (bp->b_dirtyoff > boffset) bp->b_dirtyoff = boffset; if (bp->b_dirtyend < eoffset) bp->b_dirtyend = eoffset; } } } /* * Allocate the KVA mapping for an existing buffer. * If an unmapped buffer is provided but a mapped buffer is requested, take * also care to properly setup mappings between pages and KVA. */ static void bp_unmapped_get_kva(struct buf *bp, daddr_t blkno, int size, int gbflags) { struct buf *scratch_bp; int bsize, maxsize, need_mapping, need_kva; off_t offset; need_mapping = bp->b_data == unmapped_buf && (gbflags & GB_UNMAPPED) == 0; need_kva = bp->b_kvabase == unmapped_buf && bp->b_data == unmapped_buf && (gbflags & GB_KVAALLOC) != 0; if (!need_mapping && !need_kva) return; BUF_CHECK_UNMAPPED(bp); if (need_mapping && bp->b_kvabase != unmapped_buf) { /* * Buffer is not mapped, but the KVA was already * reserved at the time of the instantiation. Use the * allocated space. */ goto has_addr; } /* * Calculate the amount of the address space we would reserve * if the buffer was mapped. */ bsize = vn_isdisk(bp->b_vp, NULL) ? DEV_BSIZE : bp->b_bufobj->bo_bsize; KASSERT(bsize != 0, ("bsize == 0, check bo->bo_bsize")); offset = blkno * bsize; maxsize = size + (offset & PAGE_MASK); maxsize = imax(maxsize, bsize); mapping_loop: if (bufkvaalloc(bp, maxsize, gbflags)) { /* * Request defragmentation. getnewbuf() returns us the * allocated space by the scratch buffer KVA. */ scratch_bp = getnewbuf(bp->b_vp, 0, 0, size, maxsize, gbflags | (GB_UNMAPPED | GB_KVAALLOC)); if (scratch_bp == NULL) { if ((gbflags & GB_NOWAIT_BD) != 0) { /* * XXXKIB: defragmentation cannot * succeed, not sure what else to do. */ panic("GB_NOWAIT_BD and GB_UNMAPPED %p", bp); } atomic_add_int(&mappingrestarts, 1); goto mapping_loop; } KASSERT(scratch_bp->b_kvabase != unmapped_buf, ("scratch bp has no KVA %p", scratch_bp)); /* Grab pointers. */ bp->b_kvabase = scratch_bp->b_kvabase; bp->b_kvasize = scratch_bp->b_kvasize; bp->b_data = scratch_bp->b_data; /* Get rid of the scratch buffer. */ scratch_bp->b_kvasize = 0; scratch_bp->b_flags |= B_INVAL; scratch_bp->b_data = scratch_bp->b_kvabase = unmapped_buf; brelse(scratch_bp); } has_addr: if (need_mapping) { /* b_offset is handled by bpmap_qenter. */ bp->b_data = bp->b_kvabase; BUF_CHECK_MAPPED(bp); bpmap_qenter(bp); } } /* * getblk: * * Get a block given a specified block and offset into a file/device. * The buffers B_DONE bit will be cleared on return, making it almost * ready for an I/O initiation. B_INVAL may or may not be set on * return. The caller should clear B_INVAL prior to initiating a * READ. * * For a non-VMIO buffer, B_CACHE is set to the opposite of B_INVAL for * an existing buffer. * * For a VMIO buffer, B_CACHE is modified according to the backing VM. * If getblk()ing a previously 0-sized invalid buffer, B_CACHE is set * and then cleared based on the backing VM. If the previous buffer is * non-0-sized but invalid, B_CACHE will be cleared. * * If getblk() must create a new buffer, the new buffer is returned with * both B_INVAL and B_CACHE clear unless it is a VMIO buffer, in which * case it is returned with B_INVAL clear and B_CACHE set based on the * backing VM. * * getblk() also forces a bwrite() for any B_DELWRI buffer whos * B_CACHE bit is clear. * * What this means, basically, is that the caller should use B_CACHE to * determine whether the buffer is fully valid or not and should clear * B_INVAL prior to issuing a read. If the caller intends to validate * the buffer by loading its data area with something, the caller needs * to clear B_INVAL. If the caller does this without issuing an I/O, * the caller should set B_CACHE ( as an optimization ), else the caller * should issue the I/O and biodone() will set B_CACHE if the I/O was * a write attempt or if it was a successfull read. If the caller * intends to issue a READ, the caller must clear B_INVAL and BIO_ERROR * prior to issuing the READ. biodone() will *not* clear B_INVAL. */ struct buf * getblk(struct vnode *vp, daddr_t blkno, int size, int slpflag, int slptimeo, int flags) { struct buf *bp; struct bufobj *bo; int bsize, error, maxsize, vmio; off_t offset; CTR3(KTR_BUF, "getblk(%p, %ld, %d)", vp, (long)blkno, size); KASSERT((flags & (GB_UNMAPPED | GB_KVAALLOC)) != GB_KVAALLOC, ("GB_KVAALLOC only makes sense with GB_UNMAPPED")); ASSERT_VOP_LOCKED(vp, "getblk"); if (size > MAXBCACHEBUF) panic("getblk: size(%d) > MAXBCACHEBUF(%d)\n", size, MAXBCACHEBUF); if (!unmapped_buf_allowed) flags &= ~(GB_UNMAPPED | GB_KVAALLOC); bo = &vp->v_bufobj; loop: BO_RLOCK(bo); bp = gbincore(bo, blkno); if (bp != NULL) { int lockflags; /* * Buffer is in-core. If the buffer is not busy nor managed, * it must be on a queue. */ lockflags = LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK; if (flags & GB_LOCK_NOWAIT) lockflags |= LK_NOWAIT; error = BUF_TIMELOCK(bp, lockflags, BO_LOCKPTR(bo), "getblk", slpflag, slptimeo); /* * If we slept and got the lock we have to restart in case * the buffer changed identities. */ if (error == ENOLCK) goto loop; /* We timed out or were interrupted. */ else if (error) return (NULL); /* If recursed, assume caller knows the rules. */ else if (BUF_LOCKRECURSED(bp)) goto end; /* * The buffer is locked. B_CACHE is cleared if the buffer is * invalid. Otherwise, for a non-VMIO buffer, B_CACHE is set * and for a VMIO buffer B_CACHE is adjusted according to the * backing VM cache. */ if (bp->b_flags & B_INVAL) bp->b_flags &= ~B_CACHE; else if ((bp->b_flags & (B_VMIO | B_INVAL)) == 0) bp->b_flags |= B_CACHE; if (bp->b_flags & B_MANAGED) MPASS(bp->b_qindex == QUEUE_NONE); else bremfree(bp); /* * check for size inconsistencies for non-VMIO case. */ if (bp->b_bcount != size) { if ((bp->b_flags & B_VMIO) == 0 || (size > bp->b_kvasize)) { if (bp->b_flags & B_DELWRI) { /* * If buffer is pinned and caller does * not want sleep waiting for it to be * unpinned, bail out * */ if (bp->b_pin_count > 0) { if (flags & GB_LOCK_NOWAIT) { bqrelse(bp); return (NULL); } else { bunpin_wait(bp); } } bp->b_flags |= B_NOCACHE; bwrite(bp); } else { if (LIST_EMPTY(&bp->b_dep)) { bp->b_flags |= B_RELBUF; brelse(bp); } else { bp->b_flags |= B_NOCACHE; bwrite(bp); } } goto loop; } } /* * Handle the case of unmapped buffer which should * become mapped, or the buffer for which KVA * reservation is requested. */ bp_unmapped_get_kva(bp, blkno, size, flags); /* * If the size is inconsistant in the VMIO case, we can resize * the buffer. This might lead to B_CACHE getting set or * cleared. If the size has not changed, B_CACHE remains * unchanged from its previous state. */ allocbuf(bp, size); KASSERT(bp->b_offset != NOOFFSET, ("getblk: no buffer offset")); /* * A buffer with B_DELWRI set and B_CACHE clear must * be committed before we can return the buffer in * order to prevent the caller from issuing a read * ( due to B_CACHE not being set ) and overwriting * it. * * Most callers, including NFS and FFS, need this to * operate properly either because they assume they * can issue a read if B_CACHE is not set, or because * ( for example ) an uncached B_DELWRI might loop due * to softupdates re-dirtying the buffer. In the latter * case, B_CACHE is set after the first write completes, * preventing further loops. * NOTE! b*write() sets B_CACHE. If we cleared B_CACHE * above while extending the buffer, we cannot allow the * buffer to remain with B_CACHE set after the write * completes or it will represent a corrupt state. To * deal with this we set B_NOCACHE to scrap the buffer * after the write. * * We might be able to do something fancy, like setting * B_CACHE in bwrite() except if B_DELWRI is already set, * so the below call doesn't set B_CACHE, but that gets real * confusing. This is much easier. */ if ((bp->b_flags & (B_CACHE|B_DELWRI)) == B_DELWRI) { bp->b_flags |= B_NOCACHE; bwrite(bp); goto loop; } bp->b_flags &= ~B_DONE; } else { /* * Buffer is not in-core, create new buffer. The buffer * returned by getnewbuf() is locked. Note that the returned * buffer is also considered valid (not marked B_INVAL). */ BO_RUNLOCK(bo); /* * If the user does not want us to create the buffer, bail out * here. */ if (flags & GB_NOCREAT) return NULL; if (numfreebuffers == 0 && TD_IS_IDLETHREAD(curthread)) return NULL; bsize = vn_isdisk(vp, NULL) ? DEV_BSIZE : bo->bo_bsize; KASSERT(bsize != 0, ("bsize == 0, check bo->bo_bsize")); offset = blkno * bsize; vmio = vp->v_object != NULL; if (vmio) { maxsize = size + (offset & PAGE_MASK); } else { maxsize = size; /* Do not allow non-VMIO notmapped buffers. */ flags &= ~(GB_UNMAPPED | GB_KVAALLOC); } maxsize = imax(maxsize, bsize); bp = getnewbuf(vp, slpflag, slptimeo, size, maxsize, flags); if (bp == NULL) { if (slpflag || slptimeo) return NULL; goto loop; } /* * This code is used to make sure that a buffer is not * created while the getnewbuf routine is blocked. * This can be a problem whether the vnode is locked or not. * If the buffer is created out from under us, we have to * throw away the one we just created. * * Note: this must occur before we associate the buffer * with the vp especially considering limitations in * the splay tree implementation when dealing with duplicate * lblkno's. */ BO_LOCK(bo); if (gbincore(bo, blkno)) { BO_UNLOCK(bo); bp->b_flags |= B_INVAL; brelse(bp); goto loop; } /* * Insert the buffer into the hash, so that it can * be found by incore. */ bp->b_blkno = bp->b_lblkno = blkno; bp->b_offset = offset; bgetvp(vp, bp); BO_UNLOCK(bo); /* * set B_VMIO bit. allocbuf() the buffer bigger. Since the * buffer size starts out as 0, B_CACHE will be set by * allocbuf() for the VMIO case prior to it testing the * backing store for validity. */ if (vmio) { bp->b_flags |= B_VMIO; KASSERT(vp->v_object == bp->b_bufobj->bo_object, ("ARGH! different b_bufobj->bo_object %p %p %p\n", bp, vp->v_object, bp->b_bufobj->bo_object)); } else { bp->b_flags &= ~B_VMIO; KASSERT(bp->b_bufobj->bo_object == NULL, ("ARGH! has b_bufobj->bo_object %p %p\n", bp, bp->b_bufobj->bo_object)); BUF_CHECK_MAPPED(bp); } allocbuf(bp, size); bp->b_flags &= ~B_DONE; } CTR4(KTR_BUF, "getblk(%p, %ld, %d) = %p", vp, (long)blkno, size, bp); BUF_ASSERT_HELD(bp); end: KASSERT(bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); return (bp); } /* * Get an empty, disassociated buffer of given size. The buffer is initially * set to B_INVAL. */ struct buf * geteblk(int size, int flags) { struct buf *bp; int maxsize; maxsize = (size + BKVAMASK) & ~BKVAMASK; while ((bp = getnewbuf(NULL, 0, 0, size, maxsize, flags)) == NULL) { if ((flags & GB_NOWAIT_BD) && (curthread->td_pflags & TDP_BUFNEED) != 0) return (NULL); } allocbuf(bp, size); bp->b_flags |= B_INVAL; /* b_dep cleared by getnewbuf() */ BUF_ASSERT_HELD(bp); return (bp); } /* * Truncate the backing store for a non-vmio buffer. */ static void vfs_nonvmio_truncate(struct buf *bp, int newbsize) { if (bp->b_flags & B_MALLOC) { /* * malloced buffers are not shrunk */ if (newbsize == 0) { bufmallocadjust(bp, 0); free(bp->b_data, M_BIOBUF); bp->b_data = bp->b_kvabase; bp->b_flags &= ~B_MALLOC; } return; } vm_hold_free_pages(bp, newbsize); bufspaceadjust(bp, newbsize); } /* * Extend the backing for a non-VMIO buffer. */ static void vfs_nonvmio_extend(struct buf *bp, int newbsize) { caddr_t origbuf; int origbufsize; /* * We only use malloced memory on the first allocation. * and revert to page-allocated memory when the buffer * grows. * * There is a potential smp race here that could lead * to bufmallocspace slightly passing the max. It * is probably extremely rare and not worth worrying * over. */ if (bp->b_bufsize == 0 && newbsize <= PAGE_SIZE/2 && bufmallocspace < maxbufmallocspace) { bp->b_data = malloc(newbsize, M_BIOBUF, M_WAITOK); bp->b_flags |= B_MALLOC; bufmallocadjust(bp, newbsize); return; } /* * If the buffer is growing on its other-than-first * allocation then we revert to the page-allocation * scheme. */ origbuf = NULL; origbufsize = 0; if (bp->b_flags & B_MALLOC) { origbuf = bp->b_data; origbufsize = bp->b_bufsize; bp->b_data = bp->b_kvabase; bufmallocadjust(bp, 0); bp->b_flags &= ~B_MALLOC; newbsize = round_page(newbsize); } vm_hold_load_pages(bp, (vm_offset_t) bp->b_data + bp->b_bufsize, (vm_offset_t) bp->b_data + newbsize); if (origbuf != NULL) { bcopy(origbuf, bp->b_data, origbufsize); free(origbuf, M_BIOBUF); } bufspaceadjust(bp, newbsize); } /* * This code constitutes the buffer memory from either anonymous system * memory (in the case of non-VMIO operations) or from an associated * VM object (in the case of VMIO operations). This code is able to * resize a buffer up or down. * * Note that this code is tricky, and has many complications to resolve * deadlock or inconsistant data situations. Tread lightly!!! * There are B_CACHE and B_DELWRI interactions that must be dealt with by * the caller. Calling this code willy nilly can result in the loss of data. * * allocbuf() only adjusts B_CACHE for VMIO buffers. getblk() deals with * B_CACHE for the non-VMIO case. */ int allocbuf(struct buf *bp, int size) { int newbsize; BUF_ASSERT_HELD(bp); if (bp->b_bcount == size) return (1); if (bp->b_kvasize != 0 && bp->b_kvasize < size) panic("allocbuf: buffer too small"); newbsize = (size + DEV_BSIZE - 1) & ~(DEV_BSIZE - 1); if ((bp->b_flags & B_VMIO) == 0) { if ((bp->b_flags & B_MALLOC) == 0) newbsize = round_page(newbsize); /* * Just get anonymous memory from the kernel. Don't * mess with B_CACHE. */ if (newbsize < bp->b_bufsize) vfs_nonvmio_truncate(bp, newbsize); else if (newbsize > bp->b_bufsize) vfs_nonvmio_extend(bp, newbsize); } else { int desiredpages; desiredpages = (size == 0) ? 0 : num_pages((bp->b_offset & PAGE_MASK) + newbsize); if (bp->b_flags & B_MALLOC) panic("allocbuf: VMIO buffer can't be malloced"); /* * Set B_CACHE initially if buffer is 0 length or will become * 0-length. */ if (size == 0 || bp->b_bufsize == 0) bp->b_flags |= B_CACHE; if (newbsize < bp->b_bufsize) vfs_vmio_truncate(bp, desiredpages); /* XXX This looks as if it should be newbsize > b_bufsize */ else if (size > bp->b_bcount) vfs_vmio_extend(bp, desiredpages, size); bufspaceadjust(bp, newbsize); } bp->b_bcount = size; /* requested buffer size. */ return (1); } extern int inflight_transient_maps; void biodone(struct bio *bp) { struct mtx *mtxp; void (*done)(struct bio *); vm_offset_t start, end; if ((bp->bio_flags & BIO_TRANSIENT_MAPPING) != 0) { bp->bio_flags &= ~BIO_TRANSIENT_MAPPING; bp->bio_flags |= BIO_UNMAPPED; start = trunc_page((vm_offset_t)bp->bio_data); end = round_page((vm_offset_t)bp->bio_data + bp->bio_length); bp->bio_data = unmapped_buf; pmap_qremove(start, OFF_TO_IDX(end - start)); vmem_free(transient_arena, start, end - start); atomic_add_int(&inflight_transient_maps, -1); } done = bp->bio_done; if (done == NULL) { mtxp = mtx_pool_find(mtxpool_sleep, bp); mtx_lock(mtxp); bp->bio_flags |= BIO_DONE; wakeup(bp); mtx_unlock(mtxp); } else { bp->bio_flags |= BIO_DONE; done(bp); } } /* * Wait for a BIO to finish. */ int biowait(struct bio *bp, const char *wchan) { struct mtx *mtxp; mtxp = mtx_pool_find(mtxpool_sleep, bp); mtx_lock(mtxp); while ((bp->bio_flags & BIO_DONE) == 0) msleep(bp, mtxp, PRIBIO, wchan, 0); mtx_unlock(mtxp); if (bp->bio_error != 0) return (bp->bio_error); if (!(bp->bio_flags & BIO_ERROR)) return (0); return (EIO); } void biofinish(struct bio *bp, struct devstat *stat, int error) { if (error) { bp->bio_error = error; bp->bio_flags |= BIO_ERROR; } if (stat != NULL) devstat_end_transaction_bio(stat, bp); biodone(bp); } /* * bufwait: * * Wait for buffer I/O completion, returning error status. The buffer * is left locked and B_DONE on return. B_EINTR is converted into an EINTR * error and cleared. */ int bufwait(struct buf *bp) { if (bp->b_iocmd == BIO_READ) bwait(bp, PRIBIO, "biord"); else bwait(bp, PRIBIO, "biowr"); if (bp->b_flags & B_EINTR) { bp->b_flags &= ~B_EINTR; return (EINTR); } if (bp->b_ioflags & BIO_ERROR) { return (bp->b_error ? bp->b_error : EIO); } else { return (0); } } /* * bufdone: * * Finish I/O on a buffer, optionally calling a completion function. * This is usually called from an interrupt so process blocking is * not allowed. * * biodone is also responsible for setting B_CACHE in a B_VMIO bp. * In a non-VMIO bp, B_CACHE will be set on the next getblk() * assuming B_INVAL is clear. * * For the VMIO case, we set B_CACHE if the op was a read and no * read error occured, or if the op was a write. B_CACHE is never * set if the buffer is invalid or otherwise uncacheable. * * biodone does not mess with B_INVAL, allowing the I/O routine or the * initiator to leave B_INVAL set to brelse the buffer out of existance * in the biodone routine. */ void bufdone(struct buf *bp) { struct bufobj *dropobj; void (*biodone)(struct buf *); CTR3(KTR_BUF, "bufdone(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); dropobj = NULL; KASSERT(!(bp->b_flags & B_DONE), ("biodone: bp %p already done", bp)); BUF_ASSERT_HELD(bp); runningbufwakeup(bp); if (bp->b_iocmd == BIO_WRITE) dropobj = bp->b_bufobj; /* call optional completion function if requested */ if (bp->b_iodone != NULL) { biodone = bp->b_iodone; bp->b_iodone = NULL; (*biodone) (bp); if (dropobj) bufobj_wdrop(dropobj); return; } bufdone_finish(bp); if (dropobj) bufobj_wdrop(dropobj); } void bufdone_finish(struct buf *bp) { BUF_ASSERT_HELD(bp); if (!LIST_EMPTY(&bp->b_dep)) buf_complete(bp); if (bp->b_flags & B_VMIO) { /* * Set B_CACHE if the op was a normal read and no error * occured. B_CACHE is set for writes in the b*write() * routines. */ if (bp->b_iocmd == BIO_READ && !(bp->b_flags & (B_INVAL|B_NOCACHE)) && !(bp->b_ioflags & BIO_ERROR)) bp->b_flags |= B_CACHE; vfs_vmio_iodone(bp); } /* * For asynchronous completions, release the buffer now. The brelse * will do a wakeup there if necessary - so no need to do a wakeup * here in the async case. The sync case always needs to do a wakeup. */ if (bp->b_flags & B_ASYNC) { if ((bp->b_flags & (B_NOCACHE | B_INVAL | B_RELBUF)) || (bp->b_ioflags & BIO_ERROR)) brelse(bp); else bqrelse(bp); } else bdone(bp); } /* * This routine is called in lieu of iodone in the case of * incomplete I/O. This keeps the busy status for pages * consistant. */ void vfs_unbusy_pages(struct buf *bp) { int i; vm_object_t obj; vm_page_t m; runningbufwakeup(bp); if (!(bp->b_flags & B_VMIO)) return; obj = bp->b_bufobj->bo_object; VM_OBJECT_WLOCK(obj); for (i = 0; i < bp->b_npages; i++) { m = bp->b_pages[i]; if (m == bogus_page) { m = vm_page_lookup(obj, OFF_TO_IDX(bp->b_offset) + i); if (!m) panic("vfs_unbusy_pages: page missing\n"); bp->b_pages[i] = m; if (buf_mapped(bp)) { BUF_CHECK_MAPPED(bp); pmap_qenter(trunc_page((vm_offset_t)bp->b_data), bp->b_pages, bp->b_npages); } else BUF_CHECK_UNMAPPED(bp); } vm_object_pip_subtract(obj, 1); vm_page_sunbusy(m); } vm_object_pip_wakeupn(obj, 0); VM_OBJECT_WUNLOCK(obj); } /* * vfs_page_set_valid: * * Set the valid bits in a page based on the supplied offset. The * range is restricted to the buffer's size. * * This routine is typically called after a read completes. */ static void vfs_page_set_valid(struct buf *bp, vm_ooffset_t off, vm_page_t m) { vm_ooffset_t eoff; /* * Compute the end offset, eoff, such that [off, eoff) does not span a * page boundary and eoff is not greater than the end of the buffer. * The end of the buffer, in this case, is our file EOF, not the * allocation size of the buffer. */ eoff = (off + PAGE_SIZE) & ~(vm_ooffset_t)PAGE_MASK; if (eoff > bp->b_offset + bp->b_bcount) eoff = bp->b_offset + bp->b_bcount; /* * Set valid range. This is typically the entire buffer and thus the * entire page. */ if (eoff > off) vm_page_set_valid_range(m, off & PAGE_MASK, eoff - off); } /* * vfs_page_set_validclean: * * Set the valid bits and clear the dirty bits in a page based on the * supplied offset. The range is restricted to the buffer's size. */ static void vfs_page_set_validclean(struct buf *bp, vm_ooffset_t off, vm_page_t m) { vm_ooffset_t soff, eoff; /* * Start and end offsets in buffer. eoff - soff may not cross a * page boundry or cross the end of the buffer. The end of the * buffer, in this case, is our file EOF, not the allocation size * of the buffer. */ soff = off; eoff = (off + PAGE_SIZE) & ~(off_t)PAGE_MASK; if (eoff > bp->b_offset + bp->b_bcount) eoff = bp->b_offset + bp->b_bcount; /* * Set valid range. This is typically the entire buffer and thus the * entire page. */ if (eoff > soff) { vm_page_set_validclean( m, (vm_offset_t) (soff & PAGE_MASK), (vm_offset_t) (eoff - soff) ); } } /* * Ensure that all buffer pages are not exclusive busied. If any page is * exclusive busy, drain it. */ void vfs_drain_busy_pages(struct buf *bp) { vm_page_t m; int i, last_busied; VM_OBJECT_ASSERT_WLOCKED(bp->b_bufobj->bo_object); last_busied = 0; for (i = 0; i < bp->b_npages; i++) { m = bp->b_pages[i]; if (vm_page_xbusied(m)) { for (; last_busied < i; last_busied++) vm_page_sbusy(bp->b_pages[last_busied]); while (vm_page_xbusied(m)) { vm_page_lock(m); VM_OBJECT_WUNLOCK(bp->b_bufobj->bo_object); vm_page_busy_sleep(m, "vbpage"); VM_OBJECT_WLOCK(bp->b_bufobj->bo_object); } } } for (i = 0; i < last_busied; i++) vm_page_sunbusy(bp->b_pages[i]); } /* * This routine is called before a device strategy routine. * It is used to tell the VM system that paging I/O is in * progress, and treat the pages associated with the buffer * almost as being exclusive busy. Also the object paging_in_progress * flag is handled to make sure that the object doesn't become * inconsistant. * * Since I/O has not been initiated yet, certain buffer flags * such as BIO_ERROR or B_INVAL may be in an inconsistant state * and should be ignored. */ void vfs_busy_pages(struct buf *bp, int clear_modify) { int i, bogus; vm_object_t obj; vm_ooffset_t foff; vm_page_t m; if (!(bp->b_flags & B_VMIO)) return; obj = bp->b_bufobj->bo_object; foff = bp->b_offset; KASSERT(bp->b_offset != NOOFFSET, ("vfs_busy_pages: no buffer offset")); VM_OBJECT_WLOCK(obj); vfs_drain_busy_pages(bp); if (bp->b_bufsize != 0) vfs_setdirty_locked_object(bp); bogus = 0; for (i = 0; i < bp->b_npages; i++) { m = bp->b_pages[i]; if ((bp->b_flags & B_CLUSTER) == 0) { vm_object_pip_add(obj, 1); vm_page_sbusy(m); } /* * When readying a buffer for a read ( i.e * clear_modify == 0 ), it is important to do * bogus_page replacement for valid pages in * partially instantiated buffers. Partially * instantiated buffers can, in turn, occur when * reconstituting a buffer from its VM backing store * base. We only have to do this if B_CACHE is * clear ( which causes the I/O to occur in the * first place ). The replacement prevents the read * I/O from overwriting potentially dirty VM-backed * pages. XXX bogus page replacement is, uh, bogus. * It may not work properly with small-block devices. * We need to find a better way. */ if (clear_modify) { pmap_remove_write(m); vfs_page_set_validclean(bp, foff, m); } else if (m->valid == VM_PAGE_BITS_ALL && (bp->b_flags & B_CACHE) == 0) { bp->b_pages[i] = bogus_page; bogus++; } foff = (foff + PAGE_SIZE) & ~(off_t)PAGE_MASK; } VM_OBJECT_WUNLOCK(obj); if (bogus && buf_mapped(bp)) { BUF_CHECK_MAPPED(bp); pmap_qenter(trunc_page((vm_offset_t)bp->b_data), bp->b_pages, bp->b_npages); } } /* * vfs_bio_set_valid: * * Set the range within the buffer to valid. The range is * relative to the beginning of the buffer, b_offset. Note that * b_offset itself may be offset from the beginning of the first * page. */ void vfs_bio_set_valid(struct buf *bp, int base, int size) { int i, n; vm_page_t m; if (!(bp->b_flags & B_VMIO)) return; /* * Fixup base to be relative to beginning of first page. * Set initial n to be the maximum number of bytes in the * first page that can be validated. */ base += (bp->b_offset & PAGE_MASK); n = PAGE_SIZE - (base & PAGE_MASK); VM_OBJECT_WLOCK(bp->b_bufobj->bo_object); for (i = base / PAGE_SIZE; size > 0 && i < bp->b_npages; ++i) { m = bp->b_pages[i]; if (n > size) n = size; vm_page_set_valid_range(m, base & PAGE_MASK, n); base += n; size -= n; n = PAGE_SIZE; } VM_OBJECT_WUNLOCK(bp->b_bufobj->bo_object); } /* * vfs_bio_clrbuf: * * If the specified buffer is a non-VMIO buffer, clear the entire * buffer. If the specified buffer is a VMIO buffer, clear and * validate only the previously invalid portions of the buffer. * This routine essentially fakes an I/O, so we need to clear * BIO_ERROR and B_INVAL. * * Note that while we only theoretically need to clear through b_bcount, * we go ahead and clear through b_bufsize. */ void vfs_bio_clrbuf(struct buf *bp) { int i, j, mask, sa, ea, slide; if ((bp->b_flags & (B_VMIO | B_MALLOC)) != B_VMIO) { clrbuf(bp); return; } bp->b_flags &= ~B_INVAL; bp->b_ioflags &= ~BIO_ERROR; VM_OBJECT_WLOCK(bp->b_bufobj->bo_object); if ((bp->b_npages == 1) && (bp->b_bufsize < PAGE_SIZE) && (bp->b_offset & PAGE_MASK) == 0) { if (bp->b_pages[0] == bogus_page) goto unlock; mask = (1 << (bp->b_bufsize / DEV_BSIZE)) - 1; VM_OBJECT_ASSERT_WLOCKED(bp->b_pages[0]->object); if ((bp->b_pages[0]->valid & mask) == mask) goto unlock; if ((bp->b_pages[0]->valid & mask) == 0) { pmap_zero_page_area(bp->b_pages[0], 0, bp->b_bufsize); bp->b_pages[0]->valid |= mask; goto unlock; } } sa = bp->b_offset & PAGE_MASK; slide = 0; for (i = 0; i < bp->b_npages; i++, sa = 0) { slide = imin(slide + PAGE_SIZE, bp->b_offset + bp->b_bufsize); ea = slide & PAGE_MASK; if (ea == 0) ea = PAGE_SIZE; if (bp->b_pages[i] == bogus_page) continue; j = sa / DEV_BSIZE; mask = ((1 << ((ea - sa) / DEV_BSIZE)) - 1) << j; VM_OBJECT_ASSERT_WLOCKED(bp->b_pages[i]->object); if ((bp->b_pages[i]->valid & mask) == mask) continue; if ((bp->b_pages[i]->valid & mask) == 0) pmap_zero_page_area(bp->b_pages[i], sa, ea - sa); else { for (; sa < ea; sa += DEV_BSIZE, j++) { if ((bp->b_pages[i]->valid & (1 << j)) == 0) { pmap_zero_page_area(bp->b_pages[i], sa, DEV_BSIZE); } } } bp->b_pages[i]->valid |= mask; } unlock: VM_OBJECT_WUNLOCK(bp->b_bufobj->bo_object); bp->b_resid = 0; } void vfs_bio_bzero_buf(struct buf *bp, int base, int size) { vm_page_t m; int i, n; if (buf_mapped(bp)) { BUF_CHECK_MAPPED(bp); bzero(bp->b_data + base, size); } else { BUF_CHECK_UNMAPPED(bp); n = PAGE_SIZE - (base & PAGE_MASK); for (i = base / PAGE_SIZE; size > 0 && i < bp->b_npages; ++i) { m = bp->b_pages[i]; if (n > size) n = size; pmap_zero_page_area(m, base & PAGE_MASK, n); base += n; size -= n; n = PAGE_SIZE; } } } /* * vm_hold_load_pages and vm_hold_free_pages get pages into * a buffers address space. The pages are anonymous and are * not associated with a file object. */ static void vm_hold_load_pages(struct buf *bp, vm_offset_t from, vm_offset_t to) { vm_offset_t pg; vm_page_t p; int index; BUF_CHECK_MAPPED(bp); to = round_page(to); from = round_page(from); index = (from - trunc_page((vm_offset_t)bp->b_data)) >> PAGE_SHIFT; for (pg = from; pg < to; pg += PAGE_SIZE, index++) { tryagain: /* * note: must allocate system pages since blocking here * could interfere with paging I/O, no matter which * process we are. */ p = vm_page_alloc(NULL, 0, VM_ALLOC_SYSTEM | VM_ALLOC_NOOBJ | VM_ALLOC_WIRED | VM_ALLOC_COUNT((to - pg) >> PAGE_SHIFT)); if (p == NULL) { VM_WAIT; goto tryagain; } pmap_qenter(pg, &p, 1); bp->b_pages[index] = p; } bp->b_npages = index; } /* Return pages associated with this buf to the vm system */ static void vm_hold_free_pages(struct buf *bp, int newbsize) { vm_offset_t from; vm_page_t p; int index, newnpages; BUF_CHECK_MAPPED(bp); from = round_page((vm_offset_t)bp->b_data + newbsize); newnpages = (from - trunc_page((vm_offset_t)bp->b_data)) >> PAGE_SHIFT; if (bp->b_npages > newnpages) pmap_qremove(from, bp->b_npages - newnpages); for (index = newnpages; index < bp->b_npages; index++) { p = bp->b_pages[index]; bp->b_pages[index] = NULL; if (vm_page_sbusied(p)) printf("vm_hold_free_pages: blkno: %jd, lblkno: %jd\n", (intmax_t)bp->b_blkno, (intmax_t)bp->b_lblkno); p->wire_count--; vm_page_free(p); atomic_subtract_int(&vm_cnt.v_wire_count, 1); } bp->b_npages = newnpages; } /* * Map an IO request into kernel virtual address space. * * All requests are (re)mapped into kernel VA space. * Notice that we use b_bufsize for the size of the buffer * to be mapped. b_bcount might be modified by the driver. * * Note that even if the caller determines that the address space should * be valid, a race or a smaller-file mapped into a larger space may * actually cause vmapbuf() to fail, so all callers of vmapbuf() MUST * check the return value. * * This function only works with pager buffers. */ int vmapbuf(struct buf *bp, int mapbuf) { vm_prot_t prot; int pidx; if (bp->b_bufsize < 0) return (-1); prot = VM_PROT_READ; if (bp->b_iocmd == BIO_READ) prot |= VM_PROT_WRITE; /* Less backwards than it looks */ if ((pidx = vm_fault_quick_hold_pages(&curproc->p_vmspace->vm_map, (vm_offset_t)bp->b_data, bp->b_bufsize, prot, bp->b_pages, btoc(MAXPHYS))) < 0) return (-1); bp->b_npages = pidx; bp->b_offset = ((vm_offset_t)bp->b_data) & PAGE_MASK; if (mapbuf || !unmapped_buf_allowed) { pmap_qenter((vm_offset_t)bp->b_kvabase, bp->b_pages, pidx); bp->b_data = bp->b_kvabase + bp->b_offset; } else bp->b_data = unmapped_buf; return(0); } /* * Free the io map PTEs associated with this IO operation. * We also invalidate the TLB entries and restore the original b_addr. * * This function only works with pager buffers. */ void vunmapbuf(struct buf *bp) { int npages; npages = bp->b_npages; if (buf_mapped(bp)) pmap_qremove(trunc_page((vm_offset_t)bp->b_data), npages); vm_page_unhold_pages(bp->b_pages, npages); bp->b_data = unmapped_buf; } void bdone(struct buf *bp) { struct mtx *mtxp; mtxp = mtx_pool_find(mtxpool_sleep, bp); mtx_lock(mtxp); bp->b_flags |= B_DONE; wakeup(bp); mtx_unlock(mtxp); } void bwait(struct buf *bp, u_char pri, const char *wchan) { struct mtx *mtxp; mtxp = mtx_pool_find(mtxpool_sleep, bp); mtx_lock(mtxp); while ((bp->b_flags & B_DONE) == 0) msleep(bp, mtxp, pri, wchan, 0); mtx_unlock(mtxp); } int bufsync(struct bufobj *bo, int waitfor) { return (VOP_FSYNC(bo->__bo_vnode, waitfor, curthread)); } void bufstrategy(struct bufobj *bo, struct buf *bp) { int i = 0; struct vnode *vp; vp = bp->b_vp; KASSERT(vp == bo->bo_private, ("Inconsistent vnode bufstrategy")); KASSERT(vp->v_type != VCHR && vp->v_type != VBLK, ("Wrong vnode in bufstrategy(bp=%p, vp=%p)", bp, vp)); i = VOP_STRATEGY(vp, bp); KASSERT(i == 0, ("VOP_STRATEGY failed bp=%p vp=%p", bp, bp->b_vp)); } void bufobj_wrefl(struct bufobj *bo) { KASSERT(bo != NULL, ("NULL bo in bufobj_wref")); ASSERT_BO_WLOCKED(bo); bo->bo_numoutput++; } void bufobj_wref(struct bufobj *bo) { KASSERT(bo != NULL, ("NULL bo in bufobj_wref")); BO_LOCK(bo); bo->bo_numoutput++; BO_UNLOCK(bo); } void bufobj_wdrop(struct bufobj *bo) { KASSERT(bo != NULL, ("NULL bo in bufobj_wdrop")); BO_LOCK(bo); KASSERT(bo->bo_numoutput > 0, ("bufobj_wdrop non-positive count")); if ((--bo->bo_numoutput == 0) && (bo->bo_flag & BO_WWAIT)) { bo->bo_flag &= ~BO_WWAIT; wakeup(&bo->bo_numoutput); } BO_UNLOCK(bo); } int bufobj_wwait(struct bufobj *bo, int slpflag, int timeo) { int error; KASSERT(bo != NULL, ("NULL bo in bufobj_wwait")); ASSERT_BO_WLOCKED(bo); error = 0; while (bo->bo_numoutput) { bo->bo_flag |= BO_WWAIT; error = msleep(&bo->bo_numoutput, BO_LOCKPTR(bo), slpflag | (PRIBIO + 1), "bo_wwait", timeo); if (error) break; } return (error); } void bpin(struct buf *bp) { struct mtx *mtxp; mtxp = mtx_pool_find(mtxpool_sleep, bp); mtx_lock(mtxp); bp->b_pin_count++; mtx_unlock(mtxp); } void bunpin(struct buf *bp) { struct mtx *mtxp; mtxp = mtx_pool_find(mtxpool_sleep, bp); mtx_lock(mtxp); if (--bp->b_pin_count == 0) wakeup(bp); mtx_unlock(mtxp); } void bunpin_wait(struct buf *bp) { struct mtx *mtxp; mtxp = mtx_pool_find(mtxpool_sleep, bp); mtx_lock(mtxp); while (bp->b_pin_count > 0) msleep(bp, mtxp, PRIBIO, "bwunpin", 0); mtx_unlock(mtxp); } /* * Set bio_data or bio_ma for struct bio from the struct buf. */ void bdata2bio(struct buf *bp, struct bio *bip) { if (!buf_mapped(bp)) { KASSERT(unmapped_buf_allowed, ("unmapped")); bip->bio_ma = bp->b_pages; bip->bio_ma_n = bp->b_npages; bip->bio_data = unmapped_buf; bip->bio_ma_offset = (vm_offset_t)bp->b_offset & PAGE_MASK; bip->bio_flags |= BIO_UNMAPPED; KASSERT(round_page(bip->bio_ma_offset + bip->bio_length) / PAGE_SIZE == bp->b_npages, ("Buffer %p too short: %d %lld %d", bp, bip->bio_ma_offset, (long long)bip->bio_length, bip->bio_ma_n)); } else { bip->bio_data = bp->b_data; bip->bio_ma = NULL; } } #include "opt_ddb.h" #ifdef DDB #include /* DDB command to show buffer data */ DB_SHOW_COMMAND(buffer, db_show_buffer) { /* get args */ struct buf *bp = (struct buf *)addr; if (!have_addr) { db_printf("usage: show buffer \n"); return; } db_printf("buf at %p\n", bp); db_printf("b_flags = 0x%b, b_xflags=0x%b, b_vflags=0x%b\n", (u_int)bp->b_flags, PRINT_BUF_FLAGS, (u_int)bp->b_xflags, PRINT_BUF_XFLAGS, (u_int)bp->b_vflags, PRINT_BUF_VFLAGS); db_printf( "b_error = %d, b_bufsize = %ld, b_bcount = %ld, b_resid = %ld\n" "b_bufobj = (%p), b_data = %p, b_blkno = %jd, b_lblkno = %jd, " "b_dep = %p\n", bp->b_error, bp->b_bufsize, bp->b_bcount, bp->b_resid, bp->b_bufobj, bp->b_data, (intmax_t)bp->b_blkno, (intmax_t)bp->b_lblkno, bp->b_dep.lh_first); db_printf("b_kvabase = %p, b_kvasize = %d\n", bp->b_kvabase, bp->b_kvasize); if (bp->b_npages) { int i; db_printf("b_npages = %d, pages(OBJ, IDX, PA): ", bp->b_npages); for (i = 0; i < bp->b_npages; i++) { vm_page_t m; m = bp->b_pages[i]; db_printf("(%p, 0x%lx, 0x%lx)", (void *)m->object, (u_long)m->pindex, (u_long)VM_PAGE_TO_PHYS(m)); if ((i + 1) < bp->b_npages) db_printf(","); } db_printf("\n"); } db_printf(" "); BUF_LOCKPRINTINFO(bp); } DB_SHOW_COMMAND(lockedbufs, lockedbufs) { struct buf *bp; int i; for (i = 0; i < nbuf; i++) { bp = &buf[i]; if (BUF_ISLOCKED(bp)) { db_show_buffer((uintptr_t)bp, 1, 0, NULL); db_printf("\n"); } } } DB_SHOW_COMMAND(vnodebufs, db_show_vnodebufs) { struct vnode *vp; struct buf *bp; if (!have_addr) { db_printf("usage: show vnodebufs \n"); return; } vp = (struct vnode *)addr; db_printf("Clean buffers:\n"); TAILQ_FOREACH(bp, &vp->v_bufobj.bo_clean.bv_hd, b_bobufs) { db_show_buffer((uintptr_t)bp, 1, 0, NULL); db_printf("\n"); } db_printf("Dirty buffers:\n"); TAILQ_FOREACH(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs) { db_show_buffer((uintptr_t)bp, 1, 0, NULL); db_printf("\n"); } } DB_COMMAND(countfreebufs, db_coundfreebufs) { struct buf *bp; int i, used = 0, nfree = 0; if (have_addr) { db_printf("usage: countfreebufs\n"); return; } for (i = 0; i < nbuf; i++) { bp = &buf[i]; if ((bp->b_flags & B_INFREECNT) != 0) nfree++; else used++; } db_printf("Counted %d free, %d used (%d tot)\n", nfree, used, nfree + used); db_printf("numfreebuffers is %d\n", numfreebuffers); } #endif /* DDB */ Index: head/sys/kern/vfs_default.c =================================================================== --- head/sys/kern/vfs_default.c (revision 288430) +++ head/sys/kern/vfs_default.c (revision 288431) @@ -1,1288 +1,1322 @@ /*- * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed * to Berkeley by John Heidemann of the UCLA Ficus project. * * Source: * @(#)i405_init.c 2.10 92/04/27 UCLA Ficus project * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static int vop_nolookup(struct vop_lookup_args *); static int vop_norename(struct vop_rename_args *); static int vop_nostrategy(struct vop_strategy_args *); static int get_next_dirent(struct vnode *vp, struct dirent **dpp, char *dirbuf, int dirbuflen, off_t *off, char **cpos, int *len, int *eofflag, struct thread *td); static int dirent_exists(struct vnode *vp, const char *dirname, struct thread *td); #define DIRENT_MINSIZE (sizeof(struct dirent) - (MAXNAMLEN+1) + 4) static int vop_stdis_text(struct vop_is_text_args *ap); static int vop_stdset_text(struct vop_set_text_args *ap); static int vop_stdunset_text(struct vop_unset_text_args *ap); static int vop_stdget_writecount(struct vop_get_writecount_args *ap); static int vop_stdadd_writecount(struct vop_add_writecount_args *ap); static int vop_stdgetpages_async(struct vop_getpages_async_args *ap); /* * This vnode table stores what we want to do if the filesystem doesn't * implement a particular VOP. * * If there is no specific entry here, we will return EOPNOTSUPP. * * Note that every filesystem has to implement either vop_access * or vop_accessx; failing to do so will result in immediate crash * due to stack overflow, as vop_stdaccess() calls vop_stdaccessx(), * which calls vop_stdaccess() etc. */ struct vop_vector default_vnodeops = { .vop_default = NULL, .vop_bypass = VOP_EOPNOTSUPP, .vop_access = vop_stdaccess, .vop_accessx = vop_stdaccessx, .vop_advise = vop_stdadvise, .vop_advlock = vop_stdadvlock, .vop_advlockasync = vop_stdadvlockasync, .vop_advlockpurge = vop_stdadvlockpurge, .vop_allocate = vop_stdallocate, .vop_bmap = vop_stdbmap, .vop_close = VOP_NULL, .vop_fsync = VOP_NULL, .vop_getpages = vop_stdgetpages, .vop_getpages_async = vop_stdgetpages_async, .vop_getwritemount = vop_stdgetwritemount, .vop_inactive = VOP_NULL, .vop_ioctl = VOP_ENOTTY, .vop_kqfilter = vop_stdkqfilter, .vop_islocked = vop_stdislocked, .vop_lock1 = vop_stdlock, .vop_lookup = vop_nolookup, .vop_open = VOP_NULL, .vop_pathconf = VOP_EINVAL, .vop_poll = vop_nopoll, .vop_putpages = vop_stdputpages, .vop_readlink = VOP_EINVAL, .vop_rename = vop_norename, .vop_revoke = VOP_PANIC, .vop_strategy = vop_nostrategy, .vop_unlock = vop_stdunlock, .vop_vptocnp = vop_stdvptocnp, .vop_vptofh = vop_stdvptofh, .vop_unp_bind = vop_stdunp_bind, .vop_unp_connect = vop_stdunp_connect, .vop_unp_detach = vop_stdunp_detach, .vop_is_text = vop_stdis_text, .vop_set_text = vop_stdset_text, .vop_unset_text = vop_stdunset_text, .vop_get_writecount = vop_stdget_writecount, .vop_add_writecount = vop_stdadd_writecount, }; /* * Series of placeholder functions for various error returns for * VOPs. */ int vop_eopnotsupp(struct vop_generic_args *ap) { /* printf("vop_notsupp[%s]\n", ap->a_desc->vdesc_name); */ return (EOPNOTSUPP); } int vop_ebadf(struct vop_generic_args *ap) { return (EBADF); } int vop_enotty(struct vop_generic_args *ap) { return (ENOTTY); } int vop_einval(struct vop_generic_args *ap) { return (EINVAL); } int vop_enoent(struct vop_generic_args *ap) { return (ENOENT); } int vop_null(struct vop_generic_args *ap) { return (0); } /* * Helper function to panic on some bad VOPs in some filesystems. */ int vop_panic(struct vop_generic_args *ap) { panic("filesystem goof: vop_panic[%s]", ap->a_desc->vdesc_name); } /* * vop_std and vop_no are default functions for use by * filesystems that need the "default reasonable" implementation for a * particular operation. * * The documentation for the operations they implement exists (if it exists) * in the VOP_(9) manpage (all uppercase). */ /* * Default vop for filesystems that do not support name lookup */ static int vop_nolookup(ap) struct vop_lookup_args /* { struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; } */ *ap; { *ap->a_vpp = NULL; return (ENOTDIR); } /* * vop_norename: * * Handle unlock and reference counting for arguments of vop_rename * for filesystems that do not implement rename operation. */ static int vop_norename(struct vop_rename_args *ap) { vop_rename_fail(ap); return (EOPNOTSUPP); } /* * vop_nostrategy: * * Strategy routine for VFS devices that have none. * * BIO_ERROR and B_INVAL must be cleared prior to calling any strategy * routine. Typically this is done for a BIO_READ strategy call. * Typically B_INVAL is assumed to already be clear prior to a write * and should not be cleared manually unless you just made the buffer * invalid. BIO_ERROR should be cleared either way. */ static int vop_nostrategy (struct vop_strategy_args *ap) { printf("No strategy for buffer at %p\n", ap->a_bp); vprint("vnode", ap->a_vp); ap->a_bp->b_ioflags |= BIO_ERROR; ap->a_bp->b_error = EOPNOTSUPP; bufdone(ap->a_bp); return (EOPNOTSUPP); } static int get_next_dirent(struct vnode *vp, struct dirent **dpp, char *dirbuf, int dirbuflen, off_t *off, char **cpos, int *len, int *eofflag, struct thread *td) { int error, reclen; struct uio uio; struct iovec iov; struct dirent *dp; KASSERT(VOP_ISLOCKED(vp), ("vp %p is not locked", vp)); KASSERT(vp->v_type == VDIR, ("vp %p is not a directory", vp)); if (*len == 0) { iov.iov_base = dirbuf; iov.iov_len = dirbuflen; uio.uio_iov = &iov; uio.uio_iovcnt = 1; uio.uio_offset = *off; uio.uio_resid = dirbuflen; uio.uio_segflg = UIO_SYSSPACE; uio.uio_rw = UIO_READ; uio.uio_td = td; *eofflag = 0; #ifdef MAC error = mac_vnode_check_readdir(td->td_ucred, vp); if (error == 0) #endif error = VOP_READDIR(vp, &uio, td->td_ucred, eofflag, NULL, NULL); if (error) return (error); *off = uio.uio_offset; *cpos = dirbuf; *len = (dirbuflen - uio.uio_resid); if (*len == 0) return (ENOENT); } dp = (struct dirent *)(*cpos); reclen = dp->d_reclen; *dpp = dp; /* check for malformed directory.. */ if (reclen < DIRENT_MINSIZE) return (EINVAL); *cpos += reclen; *len -= reclen; return (0); } /* * Check if a named file exists in a given directory vnode. */ static int dirent_exists(struct vnode *vp, const char *dirname, struct thread *td) { char *dirbuf, *cpos; int error, eofflag, dirbuflen, len, found; off_t off; struct dirent *dp; struct vattr va; KASSERT(VOP_ISLOCKED(vp), ("vp %p is not locked", vp)); KASSERT(vp->v_type == VDIR, ("vp %p is not a directory", vp)); found = 0; error = VOP_GETATTR(vp, &va, td->td_ucred); if (error) return (found); dirbuflen = DEV_BSIZE; if (dirbuflen < va.va_blocksize) dirbuflen = va.va_blocksize; dirbuf = (char *)malloc(dirbuflen, M_TEMP, M_WAITOK); off = 0; len = 0; do { error = get_next_dirent(vp, &dp, dirbuf, dirbuflen, &off, &cpos, &len, &eofflag, td); if (error) goto out; if (dp->d_type != DT_WHT && dp->d_fileno != 0 && strcmp(dp->d_name, dirname) == 0) { found = 1; goto out; } } while (len > 0 || !eofflag); out: free(dirbuf, M_TEMP); return (found); } int vop_stdaccess(struct vop_access_args *ap) { KASSERT((ap->a_accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0, ("invalid bit in accmode")); return (VOP_ACCESSX(ap->a_vp, ap->a_accmode, ap->a_cred, ap->a_td)); } int vop_stdaccessx(struct vop_accessx_args *ap) { int error; accmode_t accmode = ap->a_accmode; error = vfs_unixify_accmode(&accmode); if (error != 0) return (error); if (accmode == 0) return (0); return (VOP_ACCESS(ap->a_vp, accmode, ap->a_cred, ap->a_td)); } /* * Advisory record locking support */ int vop_stdadvlock(struct vop_advlock_args *ap) { struct vnode *vp; struct vattr vattr; int error; vp = ap->a_vp; if (ap->a_fl->l_whence == SEEK_END) { /* * The NFSv4 server must avoid doing a vn_lock() here, since it * can deadlock the nfsd threads, due to a LOR. Fortunately * the NFSv4 server always uses SEEK_SET and this code is * only required for the SEEK_END case. */ vn_lock(vp, LK_SHARED | LK_RETRY); error = VOP_GETATTR(vp, &vattr, curthread->td_ucred); VOP_UNLOCK(vp, 0); if (error) return (error); } else vattr.va_size = 0; return (lf_advlock(ap, &(vp->v_lockf), vattr.va_size)); } int vop_stdadvlockasync(struct vop_advlockasync_args *ap) { struct vnode *vp; struct vattr vattr; int error; vp = ap->a_vp; if (ap->a_fl->l_whence == SEEK_END) { /* The size argument is only needed for SEEK_END. */ vn_lock(vp, LK_SHARED | LK_RETRY); error = VOP_GETATTR(vp, &vattr, curthread->td_ucred); VOP_UNLOCK(vp, 0); if (error) return (error); } else vattr.va_size = 0; return (lf_advlockasync(ap, &(vp->v_lockf), vattr.va_size)); } int vop_stdadvlockpurge(struct vop_advlockpurge_args *ap) { struct vnode *vp; vp = ap->a_vp; lf_purgelocks(vp, &vp->v_lockf); return (0); } /* * vop_stdpathconf: * * Standard implementation of POSIX pathconf, to get information about limits * for a filesystem. * Override per filesystem for the case where the filesystem has smaller * limits. */ int vop_stdpathconf(ap) struct vop_pathconf_args /* { struct vnode *a_vp; int a_name; int *a_retval; } */ *ap; { switch (ap->a_name) { case _PC_NAME_MAX: *ap->a_retval = NAME_MAX; return (0); case _PC_PATH_MAX: *ap->a_retval = PATH_MAX; return (0); case _PC_LINK_MAX: *ap->a_retval = LINK_MAX; return (0); case _PC_MAX_CANON: *ap->a_retval = MAX_CANON; return (0); case _PC_MAX_INPUT: *ap->a_retval = MAX_INPUT; return (0); case _PC_PIPE_BUF: *ap->a_retval = PIPE_BUF; return (0); case _PC_CHOWN_RESTRICTED: *ap->a_retval = 1; return (0); case _PC_VDISABLE: *ap->a_retval = _POSIX_VDISABLE; return (0); default: return (EINVAL); } /* NOTREACHED */ } /* * Standard lock, unlock and islocked functions. */ int vop_stdlock(ap) struct vop_lock1_args /* { struct vnode *a_vp; int a_flags; char *file; int line; } */ *ap; { struct vnode *vp = ap->a_vp; return (_lockmgr_args(vp->v_vnlock, ap->a_flags, VI_MTX(vp), LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, ap->a_file, ap->a_line)); } /* See above. */ int vop_stdunlock(ap) struct vop_unlock_args /* { struct vnode *a_vp; int a_flags; } */ *ap; { struct vnode *vp = ap->a_vp; return (lockmgr(vp->v_vnlock, ap->a_flags | LK_RELEASE, VI_MTX(vp))); } /* See above. */ int vop_stdislocked(ap) struct vop_islocked_args /* { struct vnode *a_vp; } */ *ap; { return (lockstatus(ap->a_vp->v_vnlock)); } /* * Return true for select/poll. */ int vop_nopoll(ap) struct vop_poll_args /* { struct vnode *a_vp; int a_events; struct ucred *a_cred; struct thread *a_td; } */ *ap; { return (poll_no_poll(ap->a_events)); } /* * Implement poll for local filesystems that support it. */ int vop_stdpoll(ap) struct vop_poll_args /* { struct vnode *a_vp; int a_events; struct ucred *a_cred; struct thread *a_td; } */ *ap; { if (ap->a_events & ~POLLSTANDARD) return (vn_pollrecord(ap->a_vp, ap->a_td, ap->a_events)); return (ap->a_events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM)); } /* * Return our mount point, as we will take charge of the writes. */ int vop_stdgetwritemount(ap) struct vop_getwritemount_args /* { struct vnode *a_vp; struct mount **a_mpp; } */ *ap; { struct mount *mp; /* * XXX Since this is called unlocked we may be recycled while * attempting to ref the mount. If this is the case or mountpoint * will be set to NULL. We only have to prevent this call from * returning with a ref to an incorrect mountpoint. It is not * harmful to return with a ref to our previous mountpoint. */ mp = ap->a_vp->v_mount; if (mp != NULL) { vfs_ref(mp); if (mp != ap->a_vp->v_mount) { vfs_rel(mp); mp = NULL; } } *(ap->a_mpp) = mp; return (0); } /* XXX Needs good comment and VOP_BMAP(9) manpage */ int vop_stdbmap(ap) struct vop_bmap_args /* { struct vnode *a_vp; daddr_t a_bn; struct bufobj **a_bop; daddr_t *a_bnp; int *a_runp; int *a_runb; } */ *ap; { if (ap->a_bop != NULL) *ap->a_bop = &ap->a_vp->v_bufobj; if (ap->a_bnp != NULL) *ap->a_bnp = ap->a_bn * btodb(ap->a_vp->v_mount->mnt_stat.f_iosize); if (ap->a_runp != NULL) *ap->a_runp = 0; if (ap->a_runb != NULL) *ap->a_runb = 0; return (0); } int vop_stdfsync(ap) struct vop_fsync_args /* { struct vnode *a_vp; struct ucred *a_cred; int a_waitfor; struct thread *a_td; } */ *ap; { struct vnode *vp = ap->a_vp; struct buf *bp; struct bufobj *bo; struct buf *nbp; int error = 0; int maxretry = 1000; /* large, arbitrarily chosen */ bo = &vp->v_bufobj; BO_LOCK(bo); loop1: /* * MARK/SCAN initialization to avoid infinite loops. */ TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) { bp->b_vflags &= ~BV_SCANNED; bp->b_error = 0; } /* * Flush all dirty buffers associated with a vnode. */ loop2: TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { if ((bp->b_vflags & BV_SCANNED) != 0) continue; bp->b_vflags |= BV_SCANNED; if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { if (ap->a_waitfor != MNT_WAIT) continue; if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_INTERLOCK | LK_SLEEPFAIL, BO_LOCKPTR(bo)) != 0) { BO_LOCK(bo); goto loop1; } BO_LOCK(bo); } BO_UNLOCK(bo); KASSERT(bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); if ((bp->b_flags & B_DELWRI) == 0) panic("fsync: not dirty"); if ((vp->v_object != NULL) && (bp->b_flags & B_CLUSTEROK)) { vfs_bio_awrite(bp); } else { bremfree(bp); bawrite(bp); } BO_LOCK(bo); goto loop2; } /* * If synchronous the caller expects us to completely resolve all * dirty buffers in the system. Wait for in-progress I/O to * complete (which could include background bitmap writes), then * retry if dirty blocks still exist. */ if (ap->a_waitfor == MNT_WAIT) { bufobj_wwait(bo, 0, 0); if (bo->bo_dirty.bv_cnt > 0) { /* * If we are unable to write any of these buffers * then we fail now rather than trying endlessly * to write them out. */ TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) if ((error = bp->b_error) == 0) continue; if (error == 0 && --maxretry >= 0) goto loop1; error = EAGAIN; } } BO_UNLOCK(bo); if (error == EAGAIN) vprint("fsync: giving up on dirty", vp); return (error); } /* XXX Needs good comment and more info in the manpage (VOP_GETPAGES(9)). */ int vop_stdgetpages(ap) struct vop_getpages_args /* { struct vnode *a_vp; vm_page_t *a_m; int a_count; int a_reqpage; } */ *ap; { return vnode_pager_generic_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_reqpage, NULL, NULL); } static int vop_stdgetpages_async(struct vop_getpages_async_args *ap) { int error; error = VOP_GETPAGES(ap->a_vp, ap->a_m, ap->a_count, ap->a_reqpage); ap->a_iodone(ap->a_arg, ap->a_m, ap->a_reqpage, error); return (error); } int vop_stdkqfilter(struct vop_kqfilter_args *ap) { return vfs_kqfilter(ap); } /* XXX Needs good comment and more info in the manpage (VOP_PUTPAGES(9)). */ int vop_stdputpages(ap) struct vop_putpages_args /* { struct vnode *a_vp; vm_page_t *a_m; int a_count; int a_sync; int *a_rtvals; } */ *ap; { return vnode_pager_generic_putpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_sync, ap->a_rtvals); } int vop_stdvptofh(struct vop_vptofh_args *ap) { return (EOPNOTSUPP); } int vop_stdvptocnp(struct vop_vptocnp_args *ap) { struct vnode *vp = ap->a_vp; struct vnode **dvp = ap->a_vpp; struct ucred *cred = ap->a_cred; char *buf = ap->a_buf; int *buflen = ap->a_buflen; char *dirbuf, *cpos; int i, error, eofflag, dirbuflen, flags, locked, len, covered; off_t off; ino_t fileno; struct vattr va; struct nameidata nd; struct thread *td; struct dirent *dp; struct vnode *mvp; i = *buflen; error = 0; covered = 0; td = curthread; if (vp->v_type != VDIR) return (ENOENT); error = VOP_GETATTR(vp, &va, cred); if (error) return (error); VREF(vp); locked = VOP_ISLOCKED(vp); VOP_UNLOCK(vp, 0); NDINIT_ATVP(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE, "..", vp, td); flags = FREAD; error = vn_open_cred(&nd, &flags, 0, VN_OPEN_NOAUDIT, cred, NULL); if (error) { vn_lock(vp, locked | LK_RETRY); return (error); } NDFREE(&nd, NDF_ONLY_PNBUF); mvp = *dvp = nd.ni_vp; if (vp->v_mount != (*dvp)->v_mount && ((*dvp)->v_vflag & VV_ROOT) && ((*dvp)->v_mount->mnt_flag & MNT_UNION)) { *dvp = (*dvp)->v_mount->mnt_vnodecovered; VREF(mvp); VOP_UNLOCK(mvp, 0); vn_close(mvp, FREAD, cred, td); VREF(*dvp); vn_lock(*dvp, LK_SHARED | LK_RETRY); covered = 1; } fileno = va.va_fileid; dirbuflen = DEV_BSIZE; if (dirbuflen < va.va_blocksize) dirbuflen = va.va_blocksize; dirbuf = (char *)malloc(dirbuflen, M_TEMP, M_WAITOK); if ((*dvp)->v_type != VDIR) { error = ENOENT; goto out; } off = 0; len = 0; do { /* call VOP_READDIR of parent */ error = get_next_dirent(*dvp, &dp, dirbuf, dirbuflen, &off, &cpos, &len, &eofflag, td); if (error) goto out; if ((dp->d_type != DT_WHT) && (dp->d_fileno == fileno)) { if (covered) { VOP_UNLOCK(*dvp, 0); vn_lock(mvp, LK_SHARED | LK_RETRY); if (dirent_exists(mvp, dp->d_name, td)) { error = ENOENT; VOP_UNLOCK(mvp, 0); vn_lock(*dvp, LK_SHARED | LK_RETRY); goto out; } VOP_UNLOCK(mvp, 0); vn_lock(*dvp, LK_SHARED | LK_RETRY); } i -= dp->d_namlen; if (i < 0) { error = ENOMEM; goto out; } if (dp->d_namlen == 1 && dp->d_name[0] == '.') { error = ENOENT; } else { bcopy(dp->d_name, buf + i, dp->d_namlen); error = 0; } goto out; } } while (len > 0 || !eofflag); error = ENOENT; out: free(dirbuf, M_TEMP); if (!error) { *buflen = i; vref(*dvp); } if (covered) { vput(*dvp); vrele(mvp); } else { VOP_UNLOCK(mvp, 0); vn_close(mvp, FREAD, cred, td); } vn_lock(vp, locked | LK_RETRY); return (error); } int vop_stdallocate(struct vop_allocate_args *ap) { #ifdef __notyet__ struct statfs sfs; #endif struct iovec aiov; struct vattr vattr, *vap; struct uio auio; off_t fsize, len, cur, offset; uint8_t *buf; struct thread *td; struct vnode *vp; size_t iosize; int error; buf = NULL; error = 0; td = curthread; vap = &vattr; vp = ap->a_vp; len = *ap->a_len; offset = *ap->a_offset; error = VOP_GETATTR(vp, vap, td->td_ucred); if (error != 0) goto out; fsize = vap->va_size; iosize = vap->va_blocksize; if (iosize == 0) iosize = BLKDEV_IOSIZE; if (iosize > MAXPHYS) iosize = MAXPHYS; buf = malloc(iosize, M_TEMP, M_WAITOK); #ifdef __notyet__ /* * Check if the filesystem sets f_maxfilesize; if not use * VOP_SETATTR to perform the check. */ error = VFS_STATFS(vp->v_mount, &sfs, td); if (error != 0) goto out; if (sfs.f_maxfilesize) { if (offset > sfs.f_maxfilesize || len > sfs.f_maxfilesize || offset + len > sfs.f_maxfilesize) { error = EFBIG; goto out; } } else #endif if (offset + len > vap->va_size) { /* * Test offset + len against the filesystem's maxfilesize. */ VATTR_NULL(vap); vap->va_size = offset + len; error = VOP_SETATTR(vp, vap, td->td_ucred); if (error != 0) goto out; VATTR_NULL(vap); vap->va_size = fsize; error = VOP_SETATTR(vp, vap, td->td_ucred); if (error != 0) goto out; } for (;;) { /* * Read and write back anything below the nominal file * size. There's currently no way outside the filesystem * to know whether this area is sparse or not. */ cur = iosize; if ((offset % iosize) != 0) cur -= (offset % iosize); if (cur > len) cur = len; if (offset < fsize) { aiov.iov_base = buf; aiov.iov_len = cur; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_offset = offset; auio.uio_resid = cur; auio.uio_segflg = UIO_SYSSPACE; auio.uio_rw = UIO_READ; auio.uio_td = td; error = VOP_READ(vp, &auio, 0, td->td_ucred); if (error != 0) break; if (auio.uio_resid > 0) { bzero(buf + cur - auio.uio_resid, auio.uio_resid); } } else { bzero(buf, cur); } aiov.iov_base = buf; aiov.iov_len = cur; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_offset = offset; auio.uio_resid = cur; auio.uio_segflg = UIO_SYSSPACE; auio.uio_rw = UIO_WRITE; auio.uio_td = td; error = VOP_WRITE(vp, &auio, 0, td->td_ucred); if (error != 0) break; len -= cur; offset += cur; if (len == 0) break; if (should_yield()) break; } out: *ap->a_len = len; *ap->a_offset = offset; free(buf, M_TEMP); return (error); } int vop_stdadvise(struct vop_advise_args *ap) { + struct buf *bp; + struct buflists *bl; struct vnode *vp; + daddr_t bn, startn, endn; off_t start, end; - int error; + int bsize, error; vp = ap->a_vp; switch (ap->a_advice) { case POSIX_FADV_WILLNEED: /* * Do nothing for now. Filesystems should provide a * custom method which starts an asynchronous read of * the requested region. */ error = 0; break; case POSIX_FADV_DONTNEED: - /* - * Flush any open FS buffers and then remove pages - * from the backing VM object. Using vinvalbuf() here - * is a bit heavy-handed as it flushes all buffers for - * the given vnode, not just the buffers covering the - * requested range. - */ error = 0; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); if (vp->v_iflag & VI_DOOMED) { VOP_UNLOCK(vp, 0); break; } - vinvalbuf(vp, V_CLEANONLY, 0, 0); + + /* + * Deactivate pages in the specified range from the backing VM + * object. Pages that are resident in the buffer cache will + * remain wired until their corresponding buffers are released + * below. + */ if (vp->v_object != NULL) { start = trunc_page(ap->a_start); end = round_page(ap->a_end); VM_OBJECT_WLOCK(vp->v_object); - vm_object_page_cache(vp->v_object, OFF_TO_IDX(start), + vm_object_page_noreuse(vp->v_object, OFF_TO_IDX(start), OFF_TO_IDX(end)); VM_OBJECT_WUNLOCK(vp->v_object); + } + + BO_RLOCK(&vp->v_bufobj); + bsize = vp->v_bufobj.bo_bsize; + startn = ap->a_start / bsize; + if (ap->a_end == OFF_MAX) { + endn = -1; + bl = &vp->v_bufobj.bo_clean.bv_hd; + if (!TAILQ_EMPTY(bl)) + endn = TAILQ_LAST(bl, buflists)->b_lblkno; + bl = &vp->v_bufobj.bo_dirty.bv_hd; + if (!TAILQ_EMPTY(bl) && + endn < TAILQ_LAST(bl, buflists)->b_lblkno) + endn = TAILQ_LAST(bl, buflists)->b_lblkno; + } else + endn = ap->a_end / bsize; + BO_RUNLOCK(&vp->v_bufobj); + /* + * In the VMIO case, use the B_NOREUSE flag to hint that the + * pages backing each buffer in the range are unlikely to be + * reused. Dirty buffers will have the hint applied once + * they've been written. + */ + for (bn = startn; bn <= endn; bn++) { + bp = getblk(vp, bn, bsize, 0, 0, GB_NOCREAT | + GB_UNMAPPED); + if (bp == NULL) + continue; + bp->b_flags |= B_RELBUF; + if (vp->v_object != NULL) + bp->b_flags |= B_NOREUSE; + brelse(bp); } VOP_UNLOCK(vp, 0); break; default: error = EINVAL; break; } return (error); } int vop_stdunp_bind(struct vop_unp_bind_args *ap) { ap->a_vp->v_socket = ap->a_socket; return (0); } int vop_stdunp_connect(struct vop_unp_connect_args *ap) { *ap->a_socket = ap->a_vp->v_socket; return (0); } int vop_stdunp_detach(struct vop_unp_detach_args *ap) { ap->a_vp->v_socket = NULL; return (0); } static int vop_stdis_text(struct vop_is_text_args *ap) { return ((ap->a_vp->v_vflag & VV_TEXT) != 0); } static int vop_stdset_text(struct vop_set_text_args *ap) { ap->a_vp->v_vflag |= VV_TEXT; return (0); } static int vop_stdunset_text(struct vop_unset_text_args *ap) { ap->a_vp->v_vflag &= ~VV_TEXT; return (0); } static int vop_stdget_writecount(struct vop_get_writecount_args *ap) { *ap->a_writecount = ap->a_vp->v_writecount; return (0); } static int vop_stdadd_writecount(struct vop_add_writecount_args *ap) { ap->a_vp->v_writecount += ap->a_inc; return (0); } /* * vfs default ops * used to fill the vfs function table to get reasonable default return values. */ int vfs_stdroot (mp, flags, vpp) struct mount *mp; int flags; struct vnode **vpp; { return (EOPNOTSUPP); } int vfs_stdstatfs (mp, sbp) struct mount *mp; struct statfs *sbp; { return (EOPNOTSUPP); } int vfs_stdquotactl (mp, cmds, uid, arg) struct mount *mp; int cmds; uid_t uid; void *arg; { return (EOPNOTSUPP); } int vfs_stdsync(mp, waitfor) struct mount *mp; int waitfor; { struct vnode *vp, *mvp; struct thread *td; int error, lockreq, allerror = 0; td = curthread; lockreq = LK_EXCLUSIVE | LK_INTERLOCK; if (waitfor != MNT_WAIT) lockreq |= LK_NOWAIT; /* * Force stale buffer cache information to be flushed. */ loop: MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { if (vp->v_bufobj.bo_dirty.bv_cnt == 0) { VI_UNLOCK(vp); continue; } if ((error = vget(vp, lockreq, td)) != 0) { if (error == ENOENT) { MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); goto loop; } continue; } error = VOP_FSYNC(vp, waitfor, td); if (error) allerror = error; vput(vp); } return (allerror); } int vfs_stdnosync (mp, waitfor) struct mount *mp; int waitfor; { return (0); } int vfs_stdvget (mp, ino, flags, vpp) struct mount *mp; ino_t ino; int flags; struct vnode **vpp; { return (EOPNOTSUPP); } int vfs_stdfhtovp (mp, fhp, flags, vpp) struct mount *mp; struct fid *fhp; int flags; struct vnode **vpp; { return (EOPNOTSUPP); } int vfs_stdinit (vfsp) struct vfsconf *vfsp; { return (0); } int vfs_stduninit (vfsp) struct vfsconf *vfsp; { return(0); } int vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace, attrname) struct mount *mp; int cmd; struct vnode *filename_vp; int attrnamespace; const char *attrname; { if (filename_vp != NULL) VOP_UNLOCK(filename_vp, 0); return (EOPNOTSUPP); } int vfs_stdsysctl(mp, op, req) struct mount *mp; fsctlop_t op; struct sysctl_req *req; { return (EOPNOTSUPP); } /* end of vfs default ops */ Index: head/sys/kern/vfs_syscalls.c =================================================================== --- head/sys/kern/vfs_syscalls.c (revision 288430) +++ head/sys/kern/vfs_syscalls.c (revision 288431) @@ -1,4671 +1,4669 @@ /*- * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)vfs_syscalls.c 8.13 (Berkeley) 4/15/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_capsicum.h" #include "opt_compat.h" #include "opt_ktrace.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef KTRACE #include #endif #include #include #include #include #include #include #include #include MALLOC_DEFINE(M_FADVISE, "fadvise", "posix_fadvise(2) information"); SDT_PROVIDER_DEFINE(vfs); SDT_PROBE_DEFINE2(vfs, , stat, mode, "char *", "int"); SDT_PROBE_DEFINE2(vfs, , stat, reg, "char *", "int"); static int kern_chflagsat(struct thread *td, int fd, const char *path, enum uio_seg pathseg, u_long flags, int atflag); static int setfflags(struct thread *td, struct vnode *, u_long); static int getutimes(const struct timeval *, enum uio_seg, struct timespec *); static int getutimens(const struct timespec *, enum uio_seg, struct timespec *, int *); static int setutimes(struct thread *td, struct vnode *, const struct timespec *, int, int); static int vn_access(struct vnode *vp, int user_flags, struct ucred *cred, struct thread *td); /* * The module initialization routine for POSIX asynchronous I/O will * set this to the version of AIO that it implements. (Zero means * that it is not implemented.) This value is used here by pathconf() * and in kern_descrip.c by fpathconf(). */ int async_io_version; /* * Sync each mounted filesystem. */ #ifndef _SYS_SYSPROTO_H_ struct sync_args { int dummy; }; #endif /* ARGSUSED */ int sys_sync(td, uap) struct thread *td; struct sync_args *uap; { struct mount *mp, *nmp; int save; mtx_lock(&mountlist_mtx); for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) { nmp = TAILQ_NEXT(mp, mnt_list); continue; } if ((mp->mnt_flag & MNT_RDONLY) == 0 && vn_start_write(NULL, &mp, V_NOWAIT) == 0) { save = curthread_pflags_set(TDP_SYNCIO); vfs_msync(mp, MNT_NOWAIT); VFS_SYNC(mp, MNT_NOWAIT); curthread_pflags_restore(save); vn_finished_write(mp); } mtx_lock(&mountlist_mtx); nmp = TAILQ_NEXT(mp, mnt_list); vfs_unbusy(mp); } mtx_unlock(&mountlist_mtx); return (0); } /* * Change filesystem quotas. */ #ifndef _SYS_SYSPROTO_H_ struct quotactl_args { char *path; int cmd; int uid; caddr_t arg; }; #endif int sys_quotactl(td, uap) struct thread *td; register struct quotactl_args /* { char *path; int cmd; int uid; caddr_t arg; } */ *uap; { struct mount *mp; struct nameidata nd; int error; AUDIT_ARG_CMD(uap->cmd); AUDIT_ARG_UID(uap->uid); if (!prison_allow(td->td_ucred, PR_ALLOW_QUOTAS)) return (EPERM); NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_USERSPACE, uap->path, td); if ((error = namei(&nd)) != 0) return (error); NDFREE(&nd, NDF_ONLY_PNBUF); mp = nd.ni_vp->v_mount; vfs_ref(mp); vput(nd.ni_vp); error = vfs_busy(mp, 0); vfs_rel(mp); if (error != 0) return (error); error = VFS_QUOTACTL(mp, uap->cmd, uap->uid, uap->arg); /* * Since quota on operation typically needs to open quota * file, the Q_QUOTAON handler needs to unbusy the mount point * before calling into namei. Otherwise, unmount might be * started between two vfs_busy() invocations (first is our, * second is from mount point cross-walk code in lookup()), * causing deadlock. * * Require that Q_QUOTAON handles the vfs_busy() reference on * its own, always returning with ubusied mount point. */ if ((uap->cmd >> SUBCMDSHIFT) != Q_QUOTAON) vfs_unbusy(mp); return (error); } /* * Used by statfs conversion routines to scale the block size up if * necessary so that all of the block counts are <= 'max_size'. Note * that 'max_size' should be a bitmask, i.e. 2^n - 1 for some non-zero * value of 'n'. */ void statfs_scale_blocks(struct statfs *sf, long max_size) { uint64_t count; int shift; KASSERT(powerof2(max_size + 1), ("%s: invalid max_size", __func__)); /* * Attempt to scale the block counts to give a more accurate * overview to userland of the ratio of free space to used * space. To do this, find the largest block count and compute * a divisor that lets it fit into a signed integer <= max_size. */ if (sf->f_bavail < 0) count = -sf->f_bavail; else count = sf->f_bavail; count = MAX(sf->f_blocks, MAX(sf->f_bfree, count)); if (count <= max_size) return; count >>= flsl(max_size); shift = 0; while (count > 0) { shift++; count >>=1; } sf->f_bsize <<= shift; sf->f_blocks >>= shift; sf->f_bfree >>= shift; sf->f_bavail >>= shift; } /* * Get filesystem statistics. */ #ifndef _SYS_SYSPROTO_H_ struct statfs_args { char *path; struct statfs *buf; }; #endif int sys_statfs(td, uap) struct thread *td; register struct statfs_args /* { char *path; struct statfs *buf; } */ *uap; { struct statfs sf; int error; error = kern_statfs(td, uap->path, UIO_USERSPACE, &sf); if (error == 0) error = copyout(&sf, uap->buf, sizeof(sf)); return (error); } int kern_statfs(struct thread *td, char *path, enum uio_seg pathseg, struct statfs *buf) { struct mount *mp; struct statfs *sp, sb; struct nameidata nd; int error; NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF | AUDITVNODE1, pathseg, path, td); error = namei(&nd); if (error != 0) return (error); mp = nd.ni_vp->v_mount; vfs_ref(mp); NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_vp); error = vfs_busy(mp, 0); vfs_rel(mp); if (error != 0) return (error); #ifdef MAC error = mac_mount_check_stat(td->td_ucred, mp); if (error != 0) goto out; #endif /* * Set these in case the underlying filesystem fails to do so. */ sp = &mp->mnt_stat; sp->f_version = STATFS_VERSION; sp->f_namemax = NAME_MAX; sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK; error = VFS_STATFS(mp, sp); if (error != 0) goto out; if (priv_check(td, PRIV_VFS_GENERATION)) { bcopy(sp, &sb, sizeof(sb)); sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0; prison_enforce_statfs(td->td_ucred, mp, &sb); sp = &sb; } *buf = *sp; out: vfs_unbusy(mp); return (error); } /* * Get filesystem statistics. */ #ifndef _SYS_SYSPROTO_H_ struct fstatfs_args { int fd; struct statfs *buf; }; #endif int sys_fstatfs(td, uap) struct thread *td; register struct fstatfs_args /* { int fd; struct statfs *buf; } */ *uap; { struct statfs sf; int error; error = kern_fstatfs(td, uap->fd, &sf); if (error == 0) error = copyout(&sf, uap->buf, sizeof(sf)); return (error); } int kern_fstatfs(struct thread *td, int fd, struct statfs *buf) { struct file *fp; struct mount *mp; struct statfs *sp, sb; struct vnode *vp; cap_rights_t rights; int error; AUDIT_ARG_FD(fd); error = getvnode(td, fd, cap_rights_init(&rights, CAP_FSTATFS), &fp); if (error != 0) return (error); vp = fp->f_vnode; vn_lock(vp, LK_SHARED | LK_RETRY); #ifdef AUDIT AUDIT_ARG_VNODE1(vp); #endif mp = vp->v_mount; if (mp) vfs_ref(mp); VOP_UNLOCK(vp, 0); fdrop(fp, td); if (mp == NULL) { error = EBADF; goto out; } error = vfs_busy(mp, 0); vfs_rel(mp); if (error != 0) return (error); #ifdef MAC error = mac_mount_check_stat(td->td_ucred, mp); if (error != 0) goto out; #endif /* * Set these in case the underlying filesystem fails to do so. */ sp = &mp->mnt_stat; sp->f_version = STATFS_VERSION; sp->f_namemax = NAME_MAX; sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK; error = VFS_STATFS(mp, sp); if (error != 0) goto out; if (priv_check(td, PRIV_VFS_GENERATION)) { bcopy(sp, &sb, sizeof(sb)); sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0; prison_enforce_statfs(td->td_ucred, mp, &sb); sp = &sb; } *buf = *sp; out: if (mp) vfs_unbusy(mp); return (error); } /* * Get statistics on all filesystems. */ #ifndef _SYS_SYSPROTO_H_ struct getfsstat_args { struct statfs *buf; long bufsize; int flags; }; #endif int sys_getfsstat(td, uap) struct thread *td; register struct getfsstat_args /* { struct statfs *buf; long bufsize; int flags; } */ *uap; { size_t count; int error; error = kern_getfsstat(td, &uap->buf, uap->bufsize, &count, UIO_USERSPACE, uap->flags); if (error == 0) td->td_retval[0] = count; return (error); } /* * If (bufsize > 0 && bufseg == UIO_SYSSPACE) * The caller is responsible for freeing memory which will be allocated * in '*buf'. */ int kern_getfsstat(struct thread *td, struct statfs **buf, size_t bufsize, size_t *countp, enum uio_seg bufseg, int flags) { struct mount *mp, *nmp; struct statfs *sfsp, *sp, sb; size_t count, maxcount; int error; maxcount = bufsize / sizeof(struct statfs); if (bufsize == 0) sfsp = NULL; else if (bufseg == UIO_USERSPACE) sfsp = *buf; else /* if (bufseg == UIO_SYSSPACE) */ { count = 0; mtx_lock(&mountlist_mtx); TAILQ_FOREACH(mp, &mountlist, mnt_list) { count++; } mtx_unlock(&mountlist_mtx); if (maxcount > count) maxcount = count; sfsp = *buf = malloc(maxcount * sizeof(struct statfs), M_TEMP, M_WAITOK); } count = 0; mtx_lock(&mountlist_mtx); for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { if (prison_canseemount(td->td_ucred, mp) != 0) { nmp = TAILQ_NEXT(mp, mnt_list); continue; } #ifdef MAC if (mac_mount_check_stat(td->td_ucred, mp) != 0) { nmp = TAILQ_NEXT(mp, mnt_list); continue; } #endif if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) { nmp = TAILQ_NEXT(mp, mnt_list); continue; } if (sfsp && count < maxcount) { sp = &mp->mnt_stat; /* * Set these in case the underlying filesystem * fails to do so. */ sp->f_version = STATFS_VERSION; sp->f_namemax = NAME_MAX; sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK; /* * If MNT_NOWAIT or MNT_LAZY is specified, do not * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY * overrides MNT_WAIT. */ if (((flags & (MNT_LAZY|MNT_NOWAIT)) == 0 || (flags & MNT_WAIT)) && (error = VFS_STATFS(mp, sp))) { mtx_lock(&mountlist_mtx); nmp = TAILQ_NEXT(mp, mnt_list); vfs_unbusy(mp); continue; } if (priv_check(td, PRIV_VFS_GENERATION)) { bcopy(sp, &sb, sizeof(sb)); sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0; prison_enforce_statfs(td->td_ucred, mp, &sb); sp = &sb; } if (bufseg == UIO_SYSSPACE) bcopy(sp, sfsp, sizeof(*sp)); else /* if (bufseg == UIO_USERSPACE) */ { error = copyout(sp, sfsp, sizeof(*sp)); if (error != 0) { vfs_unbusy(mp); return (error); } } sfsp++; } count++; mtx_lock(&mountlist_mtx); nmp = TAILQ_NEXT(mp, mnt_list); vfs_unbusy(mp); } mtx_unlock(&mountlist_mtx); if (sfsp && count > maxcount) *countp = maxcount; else *countp = count; return (0); } #ifdef COMPAT_FREEBSD4 /* * Get old format filesystem statistics. */ static void cvtstatfs(struct statfs *, struct ostatfs *); #ifndef _SYS_SYSPROTO_H_ struct freebsd4_statfs_args { char *path; struct ostatfs *buf; }; #endif int freebsd4_statfs(td, uap) struct thread *td; struct freebsd4_statfs_args /* { char *path; struct ostatfs *buf; } */ *uap; { struct ostatfs osb; struct statfs sf; int error; error = kern_statfs(td, uap->path, UIO_USERSPACE, &sf); if (error != 0) return (error); cvtstatfs(&sf, &osb); return (copyout(&osb, uap->buf, sizeof(osb))); } /* * Get filesystem statistics. */ #ifndef _SYS_SYSPROTO_H_ struct freebsd4_fstatfs_args { int fd; struct ostatfs *buf; }; #endif int freebsd4_fstatfs(td, uap) struct thread *td; struct freebsd4_fstatfs_args /* { int fd; struct ostatfs *buf; } */ *uap; { struct ostatfs osb; struct statfs sf; int error; error = kern_fstatfs(td, uap->fd, &sf); if (error != 0) return (error); cvtstatfs(&sf, &osb); return (copyout(&osb, uap->buf, sizeof(osb))); } /* * Get statistics on all filesystems. */ #ifndef _SYS_SYSPROTO_H_ struct freebsd4_getfsstat_args { struct ostatfs *buf; long bufsize; int flags; }; #endif int freebsd4_getfsstat(td, uap) struct thread *td; register struct freebsd4_getfsstat_args /* { struct ostatfs *buf; long bufsize; int flags; } */ *uap; { struct statfs *buf, *sp; struct ostatfs osb; size_t count, size; int error; count = uap->bufsize / sizeof(struct ostatfs); size = count * sizeof(struct statfs); error = kern_getfsstat(td, &buf, size, &count, UIO_SYSSPACE, uap->flags); if (size > 0) { sp = buf; while (count > 0 && error == 0) { cvtstatfs(sp, &osb); error = copyout(&osb, uap->buf, sizeof(osb)); sp++; uap->buf++; count--; } free(buf, M_TEMP); } if (error == 0) td->td_retval[0] = count; return (error); } /* * Implement fstatfs() for (NFS) file handles. */ #ifndef _SYS_SYSPROTO_H_ struct freebsd4_fhstatfs_args { struct fhandle *u_fhp; struct ostatfs *buf; }; #endif int freebsd4_fhstatfs(td, uap) struct thread *td; struct freebsd4_fhstatfs_args /* { struct fhandle *u_fhp; struct ostatfs *buf; } */ *uap; { struct ostatfs osb; struct statfs sf; fhandle_t fh; int error; error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t)); if (error != 0) return (error); error = kern_fhstatfs(td, fh, &sf); if (error != 0) return (error); cvtstatfs(&sf, &osb); return (copyout(&osb, uap->buf, sizeof(osb))); } /* * Convert a new format statfs structure to an old format statfs structure. */ static void cvtstatfs(nsp, osp) struct statfs *nsp; struct ostatfs *osp; { statfs_scale_blocks(nsp, LONG_MAX); bzero(osp, sizeof(*osp)); osp->f_bsize = nsp->f_bsize; osp->f_iosize = MIN(nsp->f_iosize, LONG_MAX); osp->f_blocks = nsp->f_blocks; osp->f_bfree = nsp->f_bfree; osp->f_bavail = nsp->f_bavail; osp->f_files = MIN(nsp->f_files, LONG_MAX); osp->f_ffree = MIN(nsp->f_ffree, LONG_MAX); osp->f_owner = nsp->f_owner; osp->f_type = nsp->f_type; osp->f_flags = nsp->f_flags; osp->f_syncwrites = MIN(nsp->f_syncwrites, LONG_MAX); osp->f_asyncwrites = MIN(nsp->f_asyncwrites, LONG_MAX); osp->f_syncreads = MIN(nsp->f_syncreads, LONG_MAX); osp->f_asyncreads = MIN(nsp->f_asyncreads, LONG_MAX); strlcpy(osp->f_fstypename, nsp->f_fstypename, MIN(MFSNAMELEN, OMFSNAMELEN)); strlcpy(osp->f_mntonname, nsp->f_mntonname, MIN(MNAMELEN, OMNAMELEN)); strlcpy(osp->f_mntfromname, nsp->f_mntfromname, MIN(MNAMELEN, OMNAMELEN)); osp->f_fsid = nsp->f_fsid; } #endif /* COMPAT_FREEBSD4 */ /* * Change current working directory to a given file descriptor. */ #ifndef _SYS_SYSPROTO_H_ struct fchdir_args { int fd; }; #endif int sys_fchdir(td, uap) struct thread *td; struct fchdir_args /* { int fd; } */ *uap; { struct vnode *vp, *tdp; struct mount *mp; struct file *fp; cap_rights_t rights; int error; AUDIT_ARG_FD(uap->fd); error = getvnode(td, uap->fd, cap_rights_init(&rights, CAP_FCHDIR), &fp); if (error != 0) return (error); vp = fp->f_vnode; VREF(vp); fdrop(fp, td); vn_lock(vp, LK_SHARED | LK_RETRY); AUDIT_ARG_VNODE1(vp); error = change_dir(vp, td); while (!error && (mp = vp->v_mountedhere) != NULL) { if (vfs_busy(mp, 0)) continue; error = VFS_ROOT(mp, LK_SHARED, &tdp); vfs_unbusy(mp); if (error != 0) break; vput(vp); vp = tdp; } if (error != 0) { vput(vp); return (error); } VOP_UNLOCK(vp, 0); pwd_chdir(td, vp); return (0); } /* * Change current working directory (``.''). */ #ifndef _SYS_SYSPROTO_H_ struct chdir_args { char *path; }; #endif int sys_chdir(td, uap) struct thread *td; struct chdir_args /* { char *path; } */ *uap; { return (kern_chdir(td, uap->path, UIO_USERSPACE)); } int kern_chdir(struct thread *td, char *path, enum uio_seg pathseg) { struct nameidata nd; int error; NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF | AUDITVNODE1, pathseg, path, td); if ((error = namei(&nd)) != 0) return (error); if ((error = change_dir(nd.ni_vp, td)) != 0) { vput(nd.ni_vp); NDFREE(&nd, NDF_ONLY_PNBUF); return (error); } VOP_UNLOCK(nd.ni_vp, 0); NDFREE(&nd, NDF_ONLY_PNBUF); pwd_chdir(td, nd.ni_vp); return (0); } /* * Change notion of root (``/'') directory. */ #ifndef _SYS_SYSPROTO_H_ struct chroot_args { char *path; }; #endif int sys_chroot(td, uap) struct thread *td; struct chroot_args /* { char *path; } */ *uap; { struct nameidata nd; int error; error = priv_check(td, PRIV_VFS_CHROOT); if (error != 0) return (error); NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF | AUDITVNODE1, UIO_USERSPACE, uap->path, td); error = namei(&nd); if (error != 0) goto error; error = change_dir(nd.ni_vp, td); if (error != 0) goto e_vunlock; #ifdef MAC error = mac_vnode_check_chroot(td->td_ucred, nd.ni_vp); if (error != 0) goto e_vunlock; #endif VOP_UNLOCK(nd.ni_vp, 0); error = pwd_chroot(td, nd.ni_vp); vrele(nd.ni_vp); NDFREE(&nd, NDF_ONLY_PNBUF); return (error); e_vunlock: vput(nd.ni_vp); error: NDFREE(&nd, NDF_ONLY_PNBUF); return (error); } /* * Common routine for chroot and chdir. Callers must provide a locked vnode * instance. */ int change_dir(vp, td) struct vnode *vp; struct thread *td; { #ifdef MAC int error; #endif ASSERT_VOP_LOCKED(vp, "change_dir(): vp not locked"); if (vp->v_type != VDIR) return (ENOTDIR); #ifdef MAC error = mac_vnode_check_chdir(td->td_ucred, vp); if (error != 0) return (error); #endif return (VOP_ACCESS(vp, VEXEC, td->td_ucred, td)); } static __inline void flags_to_rights(int flags, cap_rights_t *rightsp) { if (flags & O_EXEC) { cap_rights_set(rightsp, CAP_FEXECVE); } else { switch ((flags & O_ACCMODE)) { case O_RDONLY: cap_rights_set(rightsp, CAP_READ); break; case O_RDWR: cap_rights_set(rightsp, CAP_READ); /* FALLTHROUGH */ case O_WRONLY: cap_rights_set(rightsp, CAP_WRITE); if (!(flags & (O_APPEND | O_TRUNC))) cap_rights_set(rightsp, CAP_SEEK); break; } } if (flags & O_CREAT) cap_rights_set(rightsp, CAP_CREATE); if (flags & O_TRUNC) cap_rights_set(rightsp, CAP_FTRUNCATE); if (flags & (O_SYNC | O_FSYNC)) cap_rights_set(rightsp, CAP_FSYNC); if (flags & (O_EXLOCK | O_SHLOCK)) cap_rights_set(rightsp, CAP_FLOCK); } /* * Check permissions, allocate an open file structure, and call the device * open routine if any. */ #ifndef _SYS_SYSPROTO_H_ struct open_args { char *path; int flags; int mode; }; #endif int sys_open(td, uap) struct thread *td; register struct open_args /* { char *path; int flags; int mode; } */ *uap; { return (kern_openat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->flags, uap->mode)); } #ifndef _SYS_SYSPROTO_H_ struct openat_args { int fd; char *path; int flag; int mode; }; #endif int sys_openat(struct thread *td, struct openat_args *uap) { return (kern_openat(td, uap->fd, uap->path, UIO_USERSPACE, uap->flag, uap->mode)); } int kern_openat(struct thread *td, int fd, char *path, enum uio_seg pathseg, int flags, int mode) { struct proc *p = td->td_proc; struct filedesc *fdp = p->p_fd; struct file *fp; struct vnode *vp; struct nameidata nd; cap_rights_t rights; int cmode, error, indx; indx = -1; AUDIT_ARG_FFLAGS(flags); AUDIT_ARG_MODE(mode); /* XXX: audit dirfd */ cap_rights_init(&rights, CAP_LOOKUP); flags_to_rights(flags, &rights); /* * Only one of the O_EXEC, O_RDONLY, O_WRONLY and O_RDWR flags * may be specified. */ if (flags & O_EXEC) { if (flags & O_ACCMODE) return (EINVAL); } else if ((flags & O_ACCMODE) == O_ACCMODE) { return (EINVAL); } else { flags = FFLAGS(flags); } /* * Allocate the file descriptor, but don't install a descriptor yet. */ error = falloc_noinstall(td, &fp); if (error != 0) return (error); /* * An extra reference on `fp' has been held for us by * falloc_noinstall(). */ /* Set the flags early so the finit in devfs can pick them up. */ fp->f_flag = flags & FMASK; cmode = ((mode & ~fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT; NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | AUDITVNODE1, pathseg, path, fd, &rights, td); td->td_dupfd = -1; /* XXX check for fdopen */ error = vn_open(&nd, &flags, cmode, fp); if (error != 0) { /* * If the vn_open replaced the method vector, something * wonderous happened deep below and we just pass it up * pretending we know what we do. */ if (error == ENXIO && fp->f_ops != &badfileops) goto success; /* * Handle special fdopen() case. bleh. * * Don't do this for relative (capability) lookups; we don't * understand exactly what would happen, and we don't think * that it ever should. */ if (nd.ni_strictrelative == 0 && (error == ENODEV || error == ENXIO) && td->td_dupfd >= 0) { error = dupfdopen(td, fdp, td->td_dupfd, flags, error, &indx); if (error == 0) goto success; } goto bad; } td->td_dupfd = 0; NDFREE(&nd, NDF_ONLY_PNBUF); vp = nd.ni_vp; /* * Store the vnode, for any f_type. Typically, the vnode use * count is decremented by direct call to vn_closefile() for * files that switched type in the cdevsw fdopen() method. */ fp->f_vnode = vp; /* * If the file wasn't claimed by devfs bind it to the normal * vnode operations here. */ if (fp->f_ops == &badfileops) { KASSERT(vp->v_type != VFIFO, ("Unexpected fifo.")); fp->f_seqcount = 1; finit(fp, (flags & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE, vp, &vnops); } VOP_UNLOCK(vp, 0); if (flags & O_TRUNC) { error = fo_truncate(fp, 0, td->td_ucred, td); if (error != 0) goto bad; } success: /* * If we haven't already installed the FD (for dupfdopen), do so now. */ if (indx == -1) { struct filecaps *fcaps; #ifdef CAPABILITIES if (nd.ni_strictrelative == 1) fcaps = &nd.ni_filecaps; else #endif fcaps = NULL; error = finstall(td, fp, &indx, flags, fcaps); /* On success finstall() consumes fcaps. */ if (error != 0) { filecaps_free(&nd.ni_filecaps); goto bad; } } else { filecaps_free(&nd.ni_filecaps); } /* * Release our private reference, leaving the one associated with * the descriptor table intact. */ fdrop(fp, td); td->td_retval[0] = indx; return (0); bad: KASSERT(indx == -1, ("indx=%d, should be -1", indx)); fdrop(fp, td); return (error); } #ifdef COMPAT_43 /* * Create a file. */ #ifndef _SYS_SYSPROTO_H_ struct ocreat_args { char *path; int mode; }; #endif int ocreat(td, uap) struct thread *td; register struct ocreat_args /* { char *path; int mode; } */ *uap; { return (kern_openat(td, AT_FDCWD, uap->path, UIO_USERSPACE, O_WRONLY | O_CREAT | O_TRUNC, uap->mode)); } #endif /* COMPAT_43 */ /* * Create a special file. */ #ifndef _SYS_SYSPROTO_H_ struct mknod_args { char *path; int mode; int dev; }; #endif int sys_mknod(td, uap) struct thread *td; register struct mknod_args /* { char *path; int mode; int dev; } */ *uap; { return (kern_mknodat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->mode, uap->dev)); } #ifndef _SYS_SYSPROTO_H_ struct mknodat_args { int fd; char *path; mode_t mode; dev_t dev; }; #endif int sys_mknodat(struct thread *td, struct mknodat_args *uap) { return (kern_mknodat(td, uap->fd, uap->path, UIO_USERSPACE, uap->mode, uap->dev)); } int kern_mknodat(struct thread *td, int fd, char *path, enum uio_seg pathseg, int mode, int dev) { struct vnode *vp; struct mount *mp; struct vattr vattr; struct nameidata nd; cap_rights_t rights; int error, whiteout = 0; AUDIT_ARG_MODE(mode); AUDIT_ARG_DEV(dev); switch (mode & S_IFMT) { case S_IFCHR: case S_IFBLK: error = priv_check(td, PRIV_VFS_MKNOD_DEV); break; case S_IFMT: error = priv_check(td, PRIV_VFS_MKNOD_BAD); break; case S_IFWHT: error = priv_check(td, PRIV_VFS_MKNOD_WHT); break; case S_IFIFO: if (dev == 0) return (kern_mkfifoat(td, fd, path, pathseg, mode)); /* FALLTHROUGH */ default: error = EINVAL; break; } if (error != 0) return (error); restart: bwillwrite(); NDINIT_ATRIGHTS(&nd, CREATE, LOCKPARENT | SAVENAME | AUDITVNODE1 | NOCACHE, pathseg, path, fd, cap_rights_init(&rights, CAP_MKNODAT), td); if ((error = namei(&nd)) != 0) return (error); vp = nd.ni_vp; if (vp != NULL) { NDFREE(&nd, NDF_ONLY_PNBUF); if (vp == nd.ni_dvp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); vrele(vp); return (EEXIST); } else { VATTR_NULL(&vattr); vattr.va_mode = (mode & ALLPERMS) & ~td->td_proc->p_fd->fd_cmask; vattr.va_rdev = dev; whiteout = 0; switch (mode & S_IFMT) { case S_IFMT: /* used by badsect to flag bad sectors */ vattr.va_type = VBAD; break; case S_IFCHR: vattr.va_type = VCHR; break; case S_IFBLK: vattr.va_type = VBLK; break; case S_IFWHT: whiteout = 1; break; default: panic("kern_mknod: invalid mode"); } } if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0) return (error); goto restart; } #ifdef MAC if (error == 0 && !whiteout) error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, &vattr); #endif if (error == 0) { if (whiteout) error = VOP_WHITEOUT(nd.ni_dvp, &nd.ni_cnd, CREATE); else { error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); if (error == 0) vput(nd.ni_vp); } } NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); vn_finished_write(mp); return (error); } /* * Create a named pipe. */ #ifndef _SYS_SYSPROTO_H_ struct mkfifo_args { char *path; int mode; }; #endif int sys_mkfifo(td, uap) struct thread *td; register struct mkfifo_args /* { char *path; int mode; } */ *uap; { return (kern_mkfifoat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->mode)); } #ifndef _SYS_SYSPROTO_H_ struct mkfifoat_args { int fd; char *path; mode_t mode; }; #endif int sys_mkfifoat(struct thread *td, struct mkfifoat_args *uap) { return (kern_mkfifoat(td, uap->fd, uap->path, UIO_USERSPACE, uap->mode)); } int kern_mkfifoat(struct thread *td, int fd, char *path, enum uio_seg pathseg, int mode) { struct mount *mp; struct vattr vattr; struct nameidata nd; cap_rights_t rights; int error; AUDIT_ARG_MODE(mode); restart: bwillwrite(); NDINIT_ATRIGHTS(&nd, CREATE, LOCKPARENT | SAVENAME | AUDITVNODE1 | NOCACHE, pathseg, path, fd, cap_rights_init(&rights, CAP_MKFIFOAT), td); if ((error = namei(&nd)) != 0) return (error); if (nd.ni_vp != NULL) { NDFREE(&nd, NDF_ONLY_PNBUF); if (nd.ni_vp == nd.ni_dvp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); vrele(nd.ni_vp); return (EEXIST); } if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0) return (error); goto restart; } VATTR_NULL(&vattr); vattr.va_type = VFIFO; vattr.va_mode = (mode & ALLPERMS) & ~td->td_proc->p_fd->fd_cmask; #ifdef MAC error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, &vattr); if (error != 0) goto out; #endif error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); if (error == 0) vput(nd.ni_vp); #ifdef MAC out: #endif vput(nd.ni_dvp); vn_finished_write(mp); NDFREE(&nd, NDF_ONLY_PNBUF); return (error); } /* * Make a hard file link. */ #ifndef _SYS_SYSPROTO_H_ struct link_args { char *path; char *link; }; #endif int sys_link(td, uap) struct thread *td; register struct link_args /* { char *path; char *link; } */ *uap; { return (kern_linkat(td, AT_FDCWD, AT_FDCWD, uap->path, uap->link, UIO_USERSPACE, FOLLOW)); } #ifndef _SYS_SYSPROTO_H_ struct linkat_args { int fd1; char *path1; int fd2; char *path2; int flag; }; #endif int sys_linkat(struct thread *td, struct linkat_args *uap) { int flag; flag = uap->flag; if (flag & ~AT_SYMLINK_FOLLOW) return (EINVAL); return (kern_linkat(td, uap->fd1, uap->fd2, uap->path1, uap->path2, UIO_USERSPACE, (flag & AT_SYMLINK_FOLLOW) ? FOLLOW : NOFOLLOW)); } int hardlink_check_uid = 0; SYSCTL_INT(_security_bsd, OID_AUTO, hardlink_check_uid, CTLFLAG_RW, &hardlink_check_uid, 0, "Unprivileged processes cannot create hard links to files owned by other " "users"); static int hardlink_check_gid = 0; SYSCTL_INT(_security_bsd, OID_AUTO, hardlink_check_gid, CTLFLAG_RW, &hardlink_check_gid, 0, "Unprivileged processes cannot create hard links to files owned by other " "groups"); static int can_hardlink(struct vnode *vp, struct ucred *cred) { struct vattr va; int error; if (!hardlink_check_uid && !hardlink_check_gid) return (0); error = VOP_GETATTR(vp, &va, cred); if (error != 0) return (error); if (hardlink_check_uid && cred->cr_uid != va.va_uid) { error = priv_check_cred(cred, PRIV_VFS_LINK, 0); if (error != 0) return (error); } if (hardlink_check_gid && !groupmember(va.va_gid, cred)) { error = priv_check_cred(cred, PRIV_VFS_LINK, 0); if (error != 0) return (error); } return (0); } int kern_linkat(struct thread *td, int fd1, int fd2, char *path1, char *path2, enum uio_seg segflg, int follow) { struct vnode *vp; struct mount *mp; struct nameidata nd; cap_rights_t rights; int error; again: bwillwrite(); NDINIT_ATRIGHTS(&nd, LOOKUP, follow | AUDITVNODE1, segflg, path1, fd1, cap_rights_init(&rights, CAP_LINKAT_SOURCE), td); if ((error = namei(&nd)) != 0) return (error); NDFREE(&nd, NDF_ONLY_PNBUF); vp = nd.ni_vp; if (vp->v_type == VDIR) { vrele(vp); return (EPERM); /* POSIX */ } NDINIT_ATRIGHTS(&nd, CREATE, LOCKPARENT | SAVENAME | AUDITVNODE2 | NOCACHE, segflg, path2, fd2, cap_rights_init(&rights, CAP_LINKAT_TARGET), td); if ((error = namei(&nd)) == 0) { if (nd.ni_vp != NULL) { NDFREE(&nd, NDF_ONLY_PNBUF); if (nd.ni_dvp == nd.ni_vp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); vrele(nd.ni_vp); vrele(vp); return (EEXIST); } else if (nd.ni_dvp->v_mount != vp->v_mount) { /* * Cross-device link. No need to recheck * vp->v_type, since it cannot change, except * to VBAD. */ NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); vrele(vp); return (EXDEV); } else if ((error = vn_lock(vp, LK_EXCLUSIVE)) == 0) { error = can_hardlink(vp, td->td_ucred); #ifdef MAC if (error == 0) error = mac_vnode_check_link(td->td_ucred, nd.ni_dvp, vp, &nd.ni_cnd); #endif if (error != 0) { vput(vp); vput(nd.ni_dvp); NDFREE(&nd, NDF_ONLY_PNBUF); return (error); } error = vn_start_write(vp, &mp, V_NOWAIT); if (error != 0) { vput(vp); vput(nd.ni_dvp); NDFREE(&nd, NDF_ONLY_PNBUF); error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); if (error != 0) return (error); goto again; } error = VOP_LINK(nd.ni_dvp, vp, &nd.ni_cnd); VOP_UNLOCK(vp, 0); vput(nd.ni_dvp); vn_finished_write(mp); NDFREE(&nd, NDF_ONLY_PNBUF); } else { vput(nd.ni_dvp); NDFREE(&nd, NDF_ONLY_PNBUF); vrele(vp); goto again; } } vrele(vp); return (error); } /* * Make a symbolic link. */ #ifndef _SYS_SYSPROTO_H_ struct symlink_args { char *path; char *link; }; #endif int sys_symlink(td, uap) struct thread *td; register struct symlink_args /* { char *path; char *link; } */ *uap; { return (kern_symlinkat(td, uap->path, AT_FDCWD, uap->link, UIO_USERSPACE)); } #ifndef _SYS_SYSPROTO_H_ struct symlinkat_args { char *path; int fd; char *path2; }; #endif int sys_symlinkat(struct thread *td, struct symlinkat_args *uap) { return (kern_symlinkat(td, uap->path1, uap->fd, uap->path2, UIO_USERSPACE)); } int kern_symlinkat(struct thread *td, char *path1, int fd, char *path2, enum uio_seg segflg) { struct mount *mp; struct vattr vattr; char *syspath; struct nameidata nd; int error; cap_rights_t rights; if (segflg == UIO_SYSSPACE) { syspath = path1; } else { syspath = uma_zalloc(namei_zone, M_WAITOK); if ((error = copyinstr(path1, syspath, MAXPATHLEN, NULL)) != 0) goto out; } AUDIT_ARG_TEXT(syspath); restart: bwillwrite(); NDINIT_ATRIGHTS(&nd, CREATE, LOCKPARENT | SAVENAME | AUDITVNODE1 | NOCACHE, segflg, path2, fd, cap_rights_init(&rights, CAP_SYMLINKAT), td); if ((error = namei(&nd)) != 0) goto out; if (nd.ni_vp) { NDFREE(&nd, NDF_ONLY_PNBUF); if (nd.ni_vp == nd.ni_dvp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); vrele(nd.ni_vp); error = EEXIST; goto out; } if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0) goto out; goto restart; } VATTR_NULL(&vattr); vattr.va_mode = ACCESSPERMS &~ td->td_proc->p_fd->fd_cmask; #ifdef MAC vattr.va_type = VLNK; error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, &vattr); if (error != 0) goto out2; #endif error = VOP_SYMLINK(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr, syspath); if (error == 0) vput(nd.ni_vp); #ifdef MAC out2: #endif NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); vn_finished_write(mp); out: if (segflg != UIO_SYSSPACE) uma_zfree(namei_zone, syspath); return (error); } /* * Delete a whiteout from the filesystem. */ int sys_undelete(td, uap) struct thread *td; register struct undelete_args /* { char *path; } */ *uap; { struct mount *mp; struct nameidata nd; int error; restart: bwillwrite(); NDINIT(&nd, DELETE, LOCKPARENT | DOWHITEOUT | AUDITVNODE1, UIO_USERSPACE, uap->path, td); error = namei(&nd); if (error != 0) return (error); if (nd.ni_vp != NULLVP || !(nd.ni_cnd.cn_flags & ISWHITEOUT)) { NDFREE(&nd, NDF_ONLY_PNBUF); if (nd.ni_vp == nd.ni_dvp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); if (nd.ni_vp) vrele(nd.ni_vp); return (EEXIST); } if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0) return (error); goto restart; } error = VOP_WHITEOUT(nd.ni_dvp, &nd.ni_cnd, DELETE); NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); vn_finished_write(mp); return (error); } /* * Delete a name from the filesystem. */ #ifndef _SYS_SYSPROTO_H_ struct unlink_args { char *path; }; #endif int sys_unlink(td, uap) struct thread *td; struct unlink_args /* { char *path; } */ *uap; { return (kern_unlinkat(td, AT_FDCWD, uap->path, UIO_USERSPACE, 0)); } #ifndef _SYS_SYSPROTO_H_ struct unlinkat_args { int fd; char *path; int flag; }; #endif int sys_unlinkat(struct thread *td, struct unlinkat_args *uap) { int flag = uap->flag; int fd = uap->fd; char *path = uap->path; if (flag & ~AT_REMOVEDIR) return (EINVAL); if (flag & AT_REMOVEDIR) return (kern_rmdirat(td, fd, path, UIO_USERSPACE)); else return (kern_unlinkat(td, fd, path, UIO_USERSPACE, 0)); } int kern_unlinkat(struct thread *td, int fd, char *path, enum uio_seg pathseg, ino_t oldinum) { struct mount *mp; struct vnode *vp; struct nameidata nd; struct stat sb; cap_rights_t rights; int error; restart: bwillwrite(); NDINIT_ATRIGHTS(&nd, DELETE, LOCKPARENT | LOCKLEAF | AUDITVNODE1, pathseg, path, fd, cap_rights_init(&rights, CAP_UNLINKAT), td); if ((error = namei(&nd)) != 0) return (error == EINVAL ? EPERM : error); vp = nd.ni_vp; if (vp->v_type == VDIR && oldinum == 0) { error = EPERM; /* POSIX */ } else if (oldinum != 0 && ((error = vn_stat(vp, &sb, td->td_ucred, NOCRED, td)) == 0) && sb.st_ino != oldinum) { error = EIDRM; /* Identifier removed */ } else { /* * The root of a mounted filesystem cannot be deleted. * * XXX: can this only be a VDIR case? */ if (vp->v_vflag & VV_ROOT) error = EBUSY; } if (error == 0) { if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); if (vp == nd.ni_dvp) vrele(vp); else vput(vp); if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0) return (error); goto restart; } #ifdef MAC error = mac_vnode_check_unlink(td->td_ucred, nd.ni_dvp, vp, &nd.ni_cnd); if (error != 0) goto out; #endif vfs_notify_upper(vp, VFS_NOTIFY_UPPER_UNLINK); error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd); #ifdef MAC out: #endif vn_finished_write(mp); } NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); if (vp == nd.ni_dvp) vrele(vp); else vput(vp); return (error); } /* * Reposition read/write file offset. */ #ifndef _SYS_SYSPROTO_H_ struct lseek_args { int fd; int pad; off_t offset; int whence; }; #endif int sys_lseek(td, uap) struct thread *td; register struct lseek_args /* { int fd; int pad; off_t offset; int whence; } */ *uap; { struct file *fp; cap_rights_t rights; int error; AUDIT_ARG_FD(uap->fd); error = fget(td, uap->fd, cap_rights_init(&rights, CAP_SEEK), &fp); if (error != 0) return (error); error = (fp->f_ops->fo_flags & DFLAG_SEEKABLE) != 0 ? fo_seek(fp, uap->offset, uap->whence, td) : ESPIPE; fdrop(fp, td); return (error); } #if defined(COMPAT_43) /* * Reposition read/write file offset. */ #ifndef _SYS_SYSPROTO_H_ struct olseek_args { int fd; long offset; int whence; }; #endif int olseek(td, uap) struct thread *td; register struct olseek_args /* { int fd; long offset; int whence; } */ *uap; { struct lseek_args /* { int fd; int pad; off_t offset; int whence; } */ nuap; nuap.fd = uap->fd; nuap.offset = uap->offset; nuap.whence = uap->whence; return (sys_lseek(td, &nuap)); } #endif /* COMPAT_43 */ #if defined(COMPAT_FREEBSD6) /* Version with the 'pad' argument */ int freebsd6_lseek(td, uap) struct thread *td; register struct freebsd6_lseek_args *uap; { struct lseek_args ouap; ouap.fd = uap->fd; ouap.offset = uap->offset; ouap.whence = uap->whence; return (sys_lseek(td, &ouap)); } #endif /* * Check access permissions using passed credentials. */ static int vn_access(vp, user_flags, cred, td) struct vnode *vp; int user_flags; struct ucred *cred; struct thread *td; { accmode_t accmode; int error; /* Flags == 0 means only check for existence. */ if (user_flags == 0) return (0); accmode = 0; if (user_flags & R_OK) accmode |= VREAD; if (user_flags & W_OK) accmode |= VWRITE; if (user_flags & X_OK) accmode |= VEXEC; #ifdef MAC error = mac_vnode_check_access(cred, vp, accmode); if (error != 0) return (error); #endif if ((accmode & VWRITE) == 0 || (error = vn_writechk(vp)) == 0) error = VOP_ACCESS(vp, accmode, cred, td); return (error); } /* * Check access permissions using "real" credentials. */ #ifndef _SYS_SYSPROTO_H_ struct access_args { char *path; int amode; }; #endif int sys_access(td, uap) struct thread *td; register struct access_args /* { char *path; int amode; } */ *uap; { return (kern_accessat(td, AT_FDCWD, uap->path, UIO_USERSPACE, 0, uap->amode)); } #ifndef _SYS_SYSPROTO_H_ struct faccessat_args { int dirfd; char *path; int amode; int flag; } #endif int sys_faccessat(struct thread *td, struct faccessat_args *uap) { return (kern_accessat(td, uap->fd, uap->path, UIO_USERSPACE, uap->flag, uap->amode)); } int kern_accessat(struct thread *td, int fd, char *path, enum uio_seg pathseg, int flag, int amode) { struct ucred *cred, *usecred; struct vnode *vp; struct nameidata nd; cap_rights_t rights; int error; if (flag & ~AT_EACCESS) return (EINVAL); if (amode != F_OK && (amode & ~(R_OK | W_OK | X_OK)) != 0) return (EINVAL); /* * Create and modify a temporary credential instead of one that * is potentially shared (if we need one). */ cred = td->td_ucred; if ((flag & AT_EACCESS) == 0 && ((cred->cr_uid != cred->cr_ruid || cred->cr_rgid != cred->cr_groups[0]))) { usecred = crdup(cred); usecred->cr_uid = cred->cr_ruid; usecred->cr_groups[0] = cred->cr_rgid; td->td_ucred = usecred; } else usecred = cred; AUDIT_ARG_VALUE(amode); NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF | AUDITVNODE1, pathseg, path, fd, cap_rights_init(&rights, CAP_FSTAT), td); if ((error = namei(&nd)) != 0) goto out; vp = nd.ni_vp; error = vn_access(vp, amode, usecred, td); NDFREE(&nd, NDF_ONLY_PNBUF); vput(vp); out: if (usecred != cred) { td->td_ucred = cred; crfree(usecred); } return (error); } /* * Check access permissions using "effective" credentials. */ #ifndef _SYS_SYSPROTO_H_ struct eaccess_args { char *path; int amode; }; #endif int sys_eaccess(td, uap) struct thread *td; register struct eaccess_args /* { char *path; int amode; } */ *uap; { return (kern_accessat(td, AT_FDCWD, uap->path, UIO_USERSPACE, AT_EACCESS, uap->amode)); } #if defined(COMPAT_43) /* * Get file status; this version follows links. */ #ifndef _SYS_SYSPROTO_H_ struct ostat_args { char *path; struct ostat *ub; }; #endif int ostat(td, uap) struct thread *td; register struct ostat_args /* { char *path; struct ostat *ub; } */ *uap; { struct stat sb; struct ostat osb; int error; error = kern_statat(td, 0, AT_FDCWD, uap->path, UIO_USERSPACE, &sb, NULL); if (error != 0) return (error); cvtstat(&sb, &osb); return (copyout(&osb, uap->ub, sizeof (osb))); } /* * Get file status; this version does not follow links. */ #ifndef _SYS_SYSPROTO_H_ struct olstat_args { char *path; struct ostat *ub; }; #endif int olstat(td, uap) struct thread *td; register struct olstat_args /* { char *path; struct ostat *ub; } */ *uap; { struct stat sb; struct ostat osb; int error; error = kern_statat(td, AT_SYMLINK_NOFOLLOW, AT_FDCWD, uap->path, UIO_USERSPACE, &sb, NULL); if (error != 0) return (error); cvtstat(&sb, &osb); return (copyout(&osb, uap->ub, sizeof (osb))); } /* * Convert from an old to a new stat structure. */ void cvtstat(st, ost) struct stat *st; struct ostat *ost; { ost->st_dev = st->st_dev; ost->st_ino = st->st_ino; ost->st_mode = st->st_mode; ost->st_nlink = st->st_nlink; ost->st_uid = st->st_uid; ost->st_gid = st->st_gid; ost->st_rdev = st->st_rdev; if (st->st_size < (quad_t)1 << 32) ost->st_size = st->st_size; else ost->st_size = -2; ost->st_atim = st->st_atim; ost->st_mtim = st->st_mtim; ost->st_ctim = st->st_ctim; ost->st_blksize = st->st_blksize; ost->st_blocks = st->st_blocks; ost->st_flags = st->st_flags; ost->st_gen = st->st_gen; } #endif /* COMPAT_43 */ /* * Get file status; this version follows links. */ #ifndef _SYS_SYSPROTO_H_ struct stat_args { char *path; struct stat *ub; }; #endif int sys_stat(td, uap) struct thread *td; register struct stat_args /* { char *path; struct stat *ub; } */ *uap; { struct stat sb; int error; error = kern_statat(td, 0, AT_FDCWD, uap->path, UIO_USERSPACE, &sb, NULL); if (error == 0) error = copyout(&sb, uap->ub, sizeof (sb)); return (error); } #ifndef _SYS_SYSPROTO_H_ struct fstatat_args { int fd; char *path; struct stat *buf; int flag; } #endif int sys_fstatat(struct thread *td, struct fstatat_args *uap) { struct stat sb; int error; error = kern_statat(td, uap->flag, uap->fd, uap->path, UIO_USERSPACE, &sb, NULL); if (error == 0) error = copyout(&sb, uap->buf, sizeof (sb)); return (error); } int kern_statat(struct thread *td, int flag, int fd, char *path, enum uio_seg pathseg, struct stat *sbp, void (*hook)(struct vnode *vp, struct stat *sbp)) { struct nameidata nd; struct stat sb; cap_rights_t rights; int error; if (flag & ~AT_SYMLINK_NOFOLLOW) return (EINVAL); NDINIT_ATRIGHTS(&nd, LOOKUP, ((flag & AT_SYMLINK_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKSHARED | LOCKLEAF | AUDITVNODE1, pathseg, path, fd, cap_rights_init(&rights, CAP_FSTAT), td); if ((error = namei(&nd)) != 0) return (error); error = vn_stat(nd.ni_vp, &sb, td->td_ucred, NOCRED, td); if (error == 0) { SDT_PROBE2(vfs, , stat, mode, path, sb.st_mode); if (S_ISREG(sb.st_mode)) SDT_PROBE2(vfs, , stat, reg, path, pathseg); if (__predict_false(hook != NULL)) hook(nd.ni_vp, &sb); } NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_vp); if (error != 0) return (error); *sbp = sb; #ifdef KTRACE if (KTRPOINT(td, KTR_STRUCT)) ktrstat(&sb); #endif return (0); } /* * Get file status; this version does not follow links. */ #ifndef _SYS_SYSPROTO_H_ struct lstat_args { char *path; struct stat *ub; }; #endif int sys_lstat(td, uap) struct thread *td; register struct lstat_args /* { char *path; struct stat *ub; } */ *uap; { struct stat sb; int error; error = kern_statat(td, AT_SYMLINK_NOFOLLOW, AT_FDCWD, uap->path, UIO_USERSPACE, &sb, NULL); if (error == 0) error = copyout(&sb, uap->ub, sizeof (sb)); return (error); } /* * Implementation of the NetBSD [l]stat() functions. */ void cvtnstat(sb, nsb) struct stat *sb; struct nstat *nsb; { bzero(nsb, sizeof *nsb); nsb->st_dev = sb->st_dev; nsb->st_ino = sb->st_ino; nsb->st_mode = sb->st_mode; nsb->st_nlink = sb->st_nlink; nsb->st_uid = sb->st_uid; nsb->st_gid = sb->st_gid; nsb->st_rdev = sb->st_rdev; nsb->st_atim = sb->st_atim; nsb->st_mtim = sb->st_mtim; nsb->st_ctim = sb->st_ctim; nsb->st_size = sb->st_size; nsb->st_blocks = sb->st_blocks; nsb->st_blksize = sb->st_blksize; nsb->st_flags = sb->st_flags; nsb->st_gen = sb->st_gen; nsb->st_birthtim = sb->st_birthtim; } #ifndef _SYS_SYSPROTO_H_ struct nstat_args { char *path; struct nstat *ub; }; #endif int sys_nstat(td, uap) struct thread *td; register struct nstat_args /* { char *path; struct nstat *ub; } */ *uap; { struct stat sb; struct nstat nsb; int error; error = kern_statat(td, 0, AT_FDCWD, uap->path, UIO_USERSPACE, &sb, NULL); if (error != 0) return (error); cvtnstat(&sb, &nsb); return (copyout(&nsb, uap->ub, sizeof (nsb))); } /* * NetBSD lstat. Get file status; this version does not follow links. */ #ifndef _SYS_SYSPROTO_H_ struct lstat_args { char *path; struct stat *ub; }; #endif int sys_nlstat(td, uap) struct thread *td; register struct nlstat_args /* { char *path; struct nstat *ub; } */ *uap; { struct stat sb; struct nstat nsb; int error; error = kern_statat(td, AT_SYMLINK_NOFOLLOW, AT_FDCWD, uap->path, UIO_USERSPACE, &sb, NULL); if (error != 0) return (error); cvtnstat(&sb, &nsb); return (copyout(&nsb, uap->ub, sizeof (nsb))); } /* * Get configurable pathname variables. */ #ifndef _SYS_SYSPROTO_H_ struct pathconf_args { char *path; int name; }; #endif int sys_pathconf(td, uap) struct thread *td; register struct pathconf_args /* { char *path; int name; } */ *uap; { return (kern_pathconf(td, uap->path, UIO_USERSPACE, uap->name, FOLLOW)); } #ifndef _SYS_SYSPROTO_H_ struct lpathconf_args { char *path; int name; }; #endif int sys_lpathconf(td, uap) struct thread *td; register struct lpathconf_args /* { char *path; int name; } */ *uap; { return (kern_pathconf(td, uap->path, UIO_USERSPACE, uap->name, NOFOLLOW)); } int kern_pathconf(struct thread *td, char *path, enum uio_seg pathseg, int name, u_long flags) { struct nameidata nd; int error; NDINIT(&nd, LOOKUP, LOCKSHARED | LOCKLEAF | AUDITVNODE1 | flags, pathseg, path, td); if ((error = namei(&nd)) != 0) return (error); NDFREE(&nd, NDF_ONLY_PNBUF); /* If asynchronous I/O is available, it works for all files. */ if (name == _PC_ASYNC_IO) td->td_retval[0] = async_io_version; else error = VOP_PATHCONF(nd.ni_vp, name, td->td_retval); vput(nd.ni_vp); return (error); } /* * Return target name of a symbolic link. */ #ifndef _SYS_SYSPROTO_H_ struct readlink_args { char *path; char *buf; size_t count; }; #endif int sys_readlink(td, uap) struct thread *td; register struct readlink_args /* { char *path; char *buf; size_t count; } */ *uap; { return (kern_readlinkat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->buf, UIO_USERSPACE, uap->count)); } #ifndef _SYS_SYSPROTO_H_ struct readlinkat_args { int fd; char *path; char *buf; size_t bufsize; }; #endif int sys_readlinkat(struct thread *td, struct readlinkat_args *uap) { return (kern_readlinkat(td, uap->fd, uap->path, UIO_USERSPACE, uap->buf, UIO_USERSPACE, uap->bufsize)); } int kern_readlinkat(struct thread *td, int fd, char *path, enum uio_seg pathseg, char *buf, enum uio_seg bufseg, size_t count) { struct vnode *vp; struct iovec aiov; struct uio auio; struct nameidata nd; int error; if (count > IOSIZE_MAX) return (EINVAL); NDINIT_AT(&nd, LOOKUP, NOFOLLOW | LOCKSHARED | LOCKLEAF | AUDITVNODE1, pathseg, path, fd, td); if ((error = namei(&nd)) != 0) return (error); NDFREE(&nd, NDF_ONLY_PNBUF); vp = nd.ni_vp; #ifdef MAC error = mac_vnode_check_readlink(td->td_ucred, vp); if (error != 0) { vput(vp); return (error); } #endif if (vp->v_type != VLNK) error = EINVAL; else { aiov.iov_base = buf; aiov.iov_len = count; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_offset = 0; auio.uio_rw = UIO_READ; auio.uio_segflg = bufseg; auio.uio_td = td; auio.uio_resid = count; error = VOP_READLINK(vp, &auio, td->td_ucred); td->td_retval[0] = count - auio.uio_resid; } vput(vp); return (error); } /* * Common implementation code for chflags() and fchflags(). */ static int setfflags(td, vp, flags) struct thread *td; struct vnode *vp; u_long flags; { struct mount *mp; struct vattr vattr; int error; /* We can't support the value matching VNOVAL. */ if (flags == VNOVAL) return (EOPNOTSUPP); /* * Prevent non-root users from setting flags on devices. When * a device is reused, users can retain ownership of the device * if they are allowed to set flags and programs assume that * chown can't fail when done as root. */ if (vp->v_type == VCHR || vp->v_type == VBLK) { error = priv_check(td, PRIV_VFS_CHFLAGS_DEV); if (error != 0) return (error); } if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) return (error); VATTR_NULL(&vattr); vattr.va_flags = flags; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); #ifdef MAC error = mac_vnode_check_setflags(td->td_ucred, vp, vattr.va_flags); if (error == 0) #endif error = VOP_SETATTR(vp, &vattr, td->td_ucred); VOP_UNLOCK(vp, 0); vn_finished_write(mp); return (error); } /* * Change flags of a file given a path name. */ #ifndef _SYS_SYSPROTO_H_ struct chflags_args { const char *path; u_long flags; }; #endif int sys_chflags(td, uap) struct thread *td; register struct chflags_args /* { const char *path; u_long flags; } */ *uap; { return (kern_chflagsat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->flags, 0)); } #ifndef _SYS_SYSPROTO_H_ struct chflagsat_args { int fd; const char *path; u_long flags; int atflag; } #endif int sys_chflagsat(struct thread *td, struct chflagsat_args *uap) { int fd = uap->fd; const char *path = uap->path; u_long flags = uap->flags; int atflag = uap->atflag; if (atflag & ~AT_SYMLINK_NOFOLLOW) return (EINVAL); return (kern_chflagsat(td, fd, path, UIO_USERSPACE, flags, atflag)); } /* * Same as chflags() but doesn't follow symlinks. */ int sys_lchflags(td, uap) struct thread *td; register struct lchflags_args /* { const char *path; u_long flags; } */ *uap; { return (kern_chflagsat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->flags, AT_SYMLINK_NOFOLLOW)); } static int kern_chflagsat(struct thread *td, int fd, const char *path, enum uio_seg pathseg, u_long flags, int atflag) { struct nameidata nd; cap_rights_t rights; int error, follow; AUDIT_ARG_FFLAGS(flags); follow = (atflag & AT_SYMLINK_NOFOLLOW) ? NOFOLLOW : FOLLOW; NDINIT_ATRIGHTS(&nd, LOOKUP, follow | AUDITVNODE1, pathseg, path, fd, cap_rights_init(&rights, CAP_FCHFLAGS), td); if ((error = namei(&nd)) != 0) return (error); NDFREE(&nd, NDF_ONLY_PNBUF); error = setfflags(td, nd.ni_vp, flags); vrele(nd.ni_vp); return (error); } /* * Change flags of a file given a file descriptor. */ #ifndef _SYS_SYSPROTO_H_ struct fchflags_args { int fd; u_long flags; }; #endif int sys_fchflags(td, uap) struct thread *td; register struct fchflags_args /* { int fd; u_long flags; } */ *uap; { struct file *fp; cap_rights_t rights; int error; AUDIT_ARG_FD(uap->fd); AUDIT_ARG_FFLAGS(uap->flags); error = getvnode(td, uap->fd, cap_rights_init(&rights, CAP_FCHFLAGS), &fp); if (error != 0) return (error); #ifdef AUDIT vn_lock(fp->f_vnode, LK_SHARED | LK_RETRY); AUDIT_ARG_VNODE1(fp->f_vnode); VOP_UNLOCK(fp->f_vnode, 0); #endif error = setfflags(td, fp->f_vnode, uap->flags); fdrop(fp, td); return (error); } /* * Common implementation code for chmod(), lchmod() and fchmod(). */ int setfmode(td, cred, vp, mode) struct thread *td; struct ucred *cred; struct vnode *vp; int mode; { struct mount *mp; struct vattr vattr; int error; if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) return (error); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); VATTR_NULL(&vattr); vattr.va_mode = mode & ALLPERMS; #ifdef MAC error = mac_vnode_check_setmode(cred, vp, vattr.va_mode); if (error == 0) #endif error = VOP_SETATTR(vp, &vattr, cred); VOP_UNLOCK(vp, 0); vn_finished_write(mp); return (error); } /* * Change mode of a file given path name. */ #ifndef _SYS_SYSPROTO_H_ struct chmod_args { char *path; int mode; }; #endif int sys_chmod(td, uap) struct thread *td; register struct chmod_args /* { char *path; int mode; } */ *uap; { return (kern_fchmodat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->mode, 0)); } #ifndef _SYS_SYSPROTO_H_ struct fchmodat_args { int dirfd; char *path; mode_t mode; int flag; } #endif int sys_fchmodat(struct thread *td, struct fchmodat_args *uap) { int flag = uap->flag; int fd = uap->fd; char *path = uap->path; mode_t mode = uap->mode; if (flag & ~AT_SYMLINK_NOFOLLOW) return (EINVAL); return (kern_fchmodat(td, fd, path, UIO_USERSPACE, mode, flag)); } /* * Change mode of a file given path name (don't follow links.) */ #ifndef _SYS_SYSPROTO_H_ struct lchmod_args { char *path; int mode; }; #endif int sys_lchmod(td, uap) struct thread *td; register struct lchmod_args /* { char *path; int mode; } */ *uap; { return (kern_fchmodat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->mode, AT_SYMLINK_NOFOLLOW)); } int kern_fchmodat(struct thread *td, int fd, char *path, enum uio_seg pathseg, mode_t mode, int flag) { struct nameidata nd; cap_rights_t rights; int error, follow; AUDIT_ARG_MODE(mode); follow = (flag & AT_SYMLINK_NOFOLLOW) ? NOFOLLOW : FOLLOW; NDINIT_ATRIGHTS(&nd, LOOKUP, follow | AUDITVNODE1, pathseg, path, fd, cap_rights_init(&rights, CAP_FCHMOD), td); if ((error = namei(&nd)) != 0) return (error); NDFREE(&nd, NDF_ONLY_PNBUF); error = setfmode(td, td->td_ucred, nd.ni_vp, mode); vrele(nd.ni_vp); return (error); } /* * Change mode of a file given a file descriptor. */ #ifndef _SYS_SYSPROTO_H_ struct fchmod_args { int fd; int mode; }; #endif int sys_fchmod(struct thread *td, struct fchmod_args *uap) { struct file *fp; cap_rights_t rights; int error; AUDIT_ARG_FD(uap->fd); AUDIT_ARG_MODE(uap->mode); error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FCHMOD), &fp); if (error != 0) return (error); error = fo_chmod(fp, uap->mode, td->td_ucred, td); fdrop(fp, td); return (error); } /* * Common implementation for chown(), lchown(), and fchown() */ int setfown(td, cred, vp, uid, gid) struct thread *td; struct ucred *cred; struct vnode *vp; uid_t uid; gid_t gid; { struct mount *mp; struct vattr vattr; int error; if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) return (error); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); VATTR_NULL(&vattr); vattr.va_uid = uid; vattr.va_gid = gid; #ifdef MAC error = mac_vnode_check_setowner(cred, vp, vattr.va_uid, vattr.va_gid); if (error == 0) #endif error = VOP_SETATTR(vp, &vattr, cred); VOP_UNLOCK(vp, 0); vn_finished_write(mp); return (error); } /* * Set ownership given a path name. */ #ifndef _SYS_SYSPROTO_H_ struct chown_args { char *path; int uid; int gid; }; #endif int sys_chown(td, uap) struct thread *td; register struct chown_args /* { char *path; int uid; int gid; } */ *uap; { return (kern_fchownat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->uid, uap->gid, 0)); } #ifndef _SYS_SYSPROTO_H_ struct fchownat_args { int fd; const char * path; uid_t uid; gid_t gid; int flag; }; #endif int sys_fchownat(struct thread *td, struct fchownat_args *uap) { int flag; flag = uap->flag; if (flag & ~AT_SYMLINK_NOFOLLOW) return (EINVAL); return (kern_fchownat(td, uap->fd, uap->path, UIO_USERSPACE, uap->uid, uap->gid, uap->flag)); } int kern_fchownat(struct thread *td, int fd, char *path, enum uio_seg pathseg, int uid, int gid, int flag) { struct nameidata nd; cap_rights_t rights; int error, follow; AUDIT_ARG_OWNER(uid, gid); follow = (flag & AT_SYMLINK_NOFOLLOW) ? NOFOLLOW : FOLLOW; NDINIT_ATRIGHTS(&nd, LOOKUP, follow | AUDITVNODE1, pathseg, path, fd, cap_rights_init(&rights, CAP_FCHOWN), td); if ((error = namei(&nd)) != 0) return (error); NDFREE(&nd, NDF_ONLY_PNBUF); error = setfown(td, td->td_ucred, nd.ni_vp, uid, gid); vrele(nd.ni_vp); return (error); } /* * Set ownership given a path name, do not cross symlinks. */ #ifndef _SYS_SYSPROTO_H_ struct lchown_args { char *path; int uid; int gid; }; #endif int sys_lchown(td, uap) struct thread *td; register struct lchown_args /* { char *path; int uid; int gid; } */ *uap; { return (kern_fchownat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->uid, uap->gid, AT_SYMLINK_NOFOLLOW)); } /* * Set ownership given a file descriptor. */ #ifndef _SYS_SYSPROTO_H_ struct fchown_args { int fd; int uid; int gid; }; #endif int sys_fchown(td, uap) struct thread *td; register struct fchown_args /* { int fd; int uid; int gid; } */ *uap; { struct file *fp; cap_rights_t rights; int error; AUDIT_ARG_FD(uap->fd); AUDIT_ARG_OWNER(uap->uid, uap->gid); error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FCHOWN), &fp); if (error != 0) return (error); error = fo_chown(fp, uap->uid, uap->gid, td->td_ucred, td); fdrop(fp, td); return (error); } /* * Common implementation code for utimes(), lutimes(), and futimes(). */ static int getutimes(usrtvp, tvpseg, tsp) const struct timeval *usrtvp; enum uio_seg tvpseg; struct timespec *tsp; { struct timeval tv[2]; const struct timeval *tvp; int error; if (usrtvp == NULL) { vfs_timestamp(&tsp[0]); tsp[1] = tsp[0]; } else { if (tvpseg == UIO_SYSSPACE) { tvp = usrtvp; } else { if ((error = copyin(usrtvp, tv, sizeof(tv))) != 0) return (error); tvp = tv; } if (tvp[0].tv_usec < 0 || tvp[0].tv_usec >= 1000000 || tvp[1].tv_usec < 0 || tvp[1].tv_usec >= 1000000) return (EINVAL); TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]); TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]); } return (0); } /* * Common implementation code for futimens(), utimensat(). */ #define UTIMENS_NULL 0x1 #define UTIMENS_EXIT 0x2 static int getutimens(const struct timespec *usrtsp, enum uio_seg tspseg, struct timespec *tsp, int *retflags) { struct timespec tsnow; int error; vfs_timestamp(&tsnow); *retflags = 0; if (usrtsp == NULL) { tsp[0] = tsnow; tsp[1] = tsnow; *retflags |= UTIMENS_NULL; return (0); } if (tspseg == UIO_SYSSPACE) { tsp[0] = usrtsp[0]; tsp[1] = usrtsp[1]; } else if ((error = copyin(usrtsp, tsp, sizeof(*tsp) * 2)) != 0) return (error); if (tsp[0].tv_nsec == UTIME_OMIT && tsp[1].tv_nsec == UTIME_OMIT) *retflags |= UTIMENS_EXIT; if (tsp[0].tv_nsec == UTIME_NOW && tsp[1].tv_nsec == UTIME_NOW) *retflags |= UTIMENS_NULL; if (tsp[0].tv_nsec == UTIME_OMIT) tsp[0].tv_sec = VNOVAL; else if (tsp[0].tv_nsec == UTIME_NOW) tsp[0] = tsnow; else if (tsp[0].tv_nsec < 0 || tsp[0].tv_nsec >= 1000000000L) return (EINVAL); if (tsp[1].tv_nsec == UTIME_OMIT) tsp[1].tv_sec = VNOVAL; else if (tsp[1].tv_nsec == UTIME_NOW) tsp[1] = tsnow; else if (tsp[1].tv_nsec < 0 || tsp[1].tv_nsec >= 1000000000L) return (EINVAL); return (0); } /* * Common implementation code for utimes(), lutimes(), futimes(), futimens(), * and utimensat(). */ static int setutimes(td, vp, ts, numtimes, nullflag) struct thread *td; struct vnode *vp; const struct timespec *ts; int numtimes; int nullflag; { struct mount *mp; struct vattr vattr; int error, setbirthtime; if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) return (error); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); setbirthtime = 0; if (numtimes < 3 && !VOP_GETATTR(vp, &vattr, td->td_ucred) && timespeccmp(&ts[1], &vattr.va_birthtime, < )) setbirthtime = 1; VATTR_NULL(&vattr); vattr.va_atime = ts[0]; vattr.va_mtime = ts[1]; if (setbirthtime) vattr.va_birthtime = ts[1]; if (numtimes > 2) vattr.va_birthtime = ts[2]; if (nullflag) vattr.va_vaflags |= VA_UTIMES_NULL; #ifdef MAC error = mac_vnode_check_setutimes(td->td_ucred, vp, vattr.va_atime, vattr.va_mtime); #endif if (error == 0) error = VOP_SETATTR(vp, &vattr, td->td_ucred); VOP_UNLOCK(vp, 0); vn_finished_write(mp); return (error); } /* * Set the access and modification times of a file. */ #ifndef _SYS_SYSPROTO_H_ struct utimes_args { char *path; struct timeval *tptr; }; #endif int sys_utimes(td, uap) struct thread *td; register struct utimes_args /* { char *path; struct timeval *tptr; } */ *uap; { return (kern_utimesat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->tptr, UIO_USERSPACE)); } #ifndef _SYS_SYSPROTO_H_ struct futimesat_args { int fd; const char * path; const struct timeval * times; }; #endif int sys_futimesat(struct thread *td, struct futimesat_args *uap) { return (kern_utimesat(td, uap->fd, uap->path, UIO_USERSPACE, uap->times, UIO_USERSPACE)); } int kern_utimesat(struct thread *td, int fd, char *path, enum uio_seg pathseg, struct timeval *tptr, enum uio_seg tptrseg) { struct nameidata nd; struct timespec ts[2]; cap_rights_t rights; int error; if ((error = getutimes(tptr, tptrseg, ts)) != 0) return (error); NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | AUDITVNODE1, pathseg, path, fd, cap_rights_init(&rights, CAP_FUTIMES), td); if ((error = namei(&nd)) != 0) return (error); NDFREE(&nd, NDF_ONLY_PNBUF); error = setutimes(td, nd.ni_vp, ts, 2, tptr == NULL); vrele(nd.ni_vp); return (error); } /* * Set the access and modification times of a file. */ #ifndef _SYS_SYSPROTO_H_ struct lutimes_args { char *path; struct timeval *tptr; }; #endif int sys_lutimes(td, uap) struct thread *td; register struct lutimes_args /* { char *path; struct timeval *tptr; } */ *uap; { return (kern_lutimes(td, uap->path, UIO_USERSPACE, uap->tptr, UIO_USERSPACE)); } int kern_lutimes(struct thread *td, char *path, enum uio_seg pathseg, struct timeval *tptr, enum uio_seg tptrseg) { struct timespec ts[2]; struct nameidata nd; int error; if ((error = getutimes(tptr, tptrseg, ts)) != 0) return (error); NDINIT(&nd, LOOKUP, NOFOLLOW | AUDITVNODE1, pathseg, path, td); if ((error = namei(&nd)) != 0) return (error); NDFREE(&nd, NDF_ONLY_PNBUF); error = setutimes(td, nd.ni_vp, ts, 2, tptr == NULL); vrele(nd.ni_vp); return (error); } /* * Set the access and modification times of a file. */ #ifndef _SYS_SYSPROTO_H_ struct futimes_args { int fd; struct timeval *tptr; }; #endif int sys_futimes(td, uap) struct thread *td; register struct futimes_args /* { int fd; struct timeval *tptr; } */ *uap; { return (kern_futimes(td, uap->fd, uap->tptr, UIO_USERSPACE)); } int kern_futimes(struct thread *td, int fd, struct timeval *tptr, enum uio_seg tptrseg) { struct timespec ts[2]; struct file *fp; cap_rights_t rights; int error; AUDIT_ARG_FD(fd); error = getutimes(tptr, tptrseg, ts); if (error != 0) return (error); error = getvnode(td, fd, cap_rights_init(&rights, CAP_FUTIMES), &fp); if (error != 0) return (error); #ifdef AUDIT vn_lock(fp->f_vnode, LK_SHARED | LK_RETRY); AUDIT_ARG_VNODE1(fp->f_vnode); VOP_UNLOCK(fp->f_vnode, 0); #endif error = setutimes(td, fp->f_vnode, ts, 2, tptr == NULL); fdrop(fp, td); return (error); } int sys_futimens(struct thread *td, struct futimens_args *uap) { return (kern_futimens(td, uap->fd, uap->times, UIO_USERSPACE)); } int kern_futimens(struct thread *td, int fd, struct timespec *tptr, enum uio_seg tptrseg) { struct timespec ts[2]; struct file *fp; cap_rights_t rights; int error, flags; AUDIT_ARG_FD(fd); error = getutimens(tptr, tptrseg, ts, &flags); if (error != 0) return (error); if (flags & UTIMENS_EXIT) return (0); error = getvnode(td, fd, cap_rights_init(&rights, CAP_FUTIMES), &fp); if (error != 0) return (error); #ifdef AUDIT vn_lock(fp->f_vnode, LK_SHARED | LK_RETRY); AUDIT_ARG_VNODE1(fp->f_vnode); VOP_UNLOCK(fp->f_vnode, 0); #endif error = setutimes(td, fp->f_vnode, ts, 2, flags & UTIMENS_NULL); fdrop(fp, td); return (error); } int sys_utimensat(struct thread *td, struct utimensat_args *uap) { return (kern_utimensat(td, uap->fd, uap->path, UIO_USERSPACE, uap->times, UIO_USERSPACE, uap->flag)); } int kern_utimensat(struct thread *td, int fd, char *path, enum uio_seg pathseg, struct timespec *tptr, enum uio_seg tptrseg, int flag) { struct nameidata nd; struct timespec ts[2]; cap_rights_t rights; int error, flags; if (flag & ~AT_SYMLINK_NOFOLLOW) return (EINVAL); if ((error = getutimens(tptr, tptrseg, ts, &flags)) != 0) return (error); NDINIT_ATRIGHTS(&nd, LOOKUP, ((flag & AT_SYMLINK_NOFOLLOW) ? NOFOLLOW : FOLLOW) | AUDITVNODE1, pathseg, path, fd, cap_rights_init(&rights, CAP_FUTIMES), td); if ((error = namei(&nd)) != 0) return (error); /* * We are allowed to call namei() regardless of 2xUTIME_OMIT. * POSIX states: * "If both tv_nsec fields are UTIME_OMIT... EACCESS may be detected." * "Search permission is denied by a component of the path prefix." */ NDFREE(&nd, NDF_ONLY_PNBUF); if ((flags & UTIMENS_EXIT) == 0) error = setutimes(td, nd.ni_vp, ts, 2, flags & UTIMENS_NULL); vrele(nd.ni_vp); return (error); } /* * Truncate a file given its path name. */ #ifndef _SYS_SYSPROTO_H_ struct truncate_args { char *path; int pad; off_t length; }; #endif int sys_truncate(td, uap) struct thread *td; register struct truncate_args /* { char *path; int pad; off_t length; } */ *uap; { return (kern_truncate(td, uap->path, UIO_USERSPACE, uap->length)); } int kern_truncate(struct thread *td, char *path, enum uio_seg pathseg, off_t length) { struct mount *mp; struct vnode *vp; void *rl_cookie; struct vattr vattr; struct nameidata nd; int error; if (length < 0) return(EINVAL); NDINIT(&nd, LOOKUP, FOLLOW | AUDITVNODE1, pathseg, path, td); if ((error = namei(&nd)) != 0) return (error); vp = nd.ni_vp; rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX); if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) { vn_rangelock_unlock(vp, rl_cookie); vrele(vp); return (error); } NDFREE(&nd, NDF_ONLY_PNBUF); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); if (vp->v_type == VDIR) error = EISDIR; #ifdef MAC else if ((error = mac_vnode_check_write(td->td_ucred, NOCRED, vp))) { } #endif else if ((error = vn_writechk(vp)) == 0 && (error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td)) == 0) { VATTR_NULL(&vattr); vattr.va_size = length; error = VOP_SETATTR(vp, &vattr, td->td_ucred); } VOP_UNLOCK(vp, 0); vn_finished_write(mp); vn_rangelock_unlock(vp, rl_cookie); vrele(vp); return (error); } #if defined(COMPAT_43) /* * Truncate a file given its path name. */ #ifndef _SYS_SYSPROTO_H_ struct otruncate_args { char *path; long length; }; #endif int otruncate(td, uap) struct thread *td; register struct otruncate_args /* { char *path; long length; } */ *uap; { struct truncate_args /* { char *path; int pad; off_t length; } */ nuap; nuap.path = uap->path; nuap.length = uap->length; return (sys_truncate(td, &nuap)); } #endif /* COMPAT_43 */ #if defined(COMPAT_FREEBSD6) /* Versions with the pad argument */ int freebsd6_truncate(struct thread *td, struct freebsd6_truncate_args *uap) { struct truncate_args ouap; ouap.path = uap->path; ouap.length = uap->length; return (sys_truncate(td, &ouap)); } int freebsd6_ftruncate(struct thread *td, struct freebsd6_ftruncate_args *uap) { struct ftruncate_args ouap; ouap.fd = uap->fd; ouap.length = uap->length; return (sys_ftruncate(td, &ouap)); } #endif /* * Sync an open file. */ #ifndef _SYS_SYSPROTO_H_ struct fsync_args { int fd; }; #endif int sys_fsync(td, uap) struct thread *td; struct fsync_args /* { int fd; } */ *uap; { struct vnode *vp; struct mount *mp; struct file *fp; cap_rights_t rights; int error, lock_flags; AUDIT_ARG_FD(uap->fd); error = getvnode(td, uap->fd, cap_rights_init(&rights, CAP_FSYNC), &fp); if (error != 0) return (error); vp = fp->f_vnode; error = vn_start_write(vp, &mp, V_WAIT | PCATCH); if (error != 0) goto drop; if (MNT_SHARED_WRITES(mp) || ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount))) { lock_flags = LK_SHARED; } else { lock_flags = LK_EXCLUSIVE; } vn_lock(vp, lock_flags | LK_RETRY); AUDIT_ARG_VNODE1(vp); if (vp->v_object != NULL) { VM_OBJECT_WLOCK(vp->v_object); vm_object_page_clean(vp->v_object, 0, 0, 0); VM_OBJECT_WUNLOCK(vp->v_object); } error = VOP_FSYNC(vp, MNT_WAIT, td); VOP_UNLOCK(vp, 0); vn_finished_write(mp); drop: fdrop(fp, td); return (error); } /* * Rename files. Source and destination must either both be directories, or * both not be directories. If target is a directory, it must be empty. */ #ifndef _SYS_SYSPROTO_H_ struct rename_args { char *from; char *to; }; #endif int sys_rename(td, uap) struct thread *td; register struct rename_args /* { char *from; char *to; } */ *uap; { return (kern_renameat(td, AT_FDCWD, uap->from, AT_FDCWD, uap->to, UIO_USERSPACE)); } #ifndef _SYS_SYSPROTO_H_ struct renameat_args { int oldfd; char *old; int newfd; char *new; }; #endif int sys_renameat(struct thread *td, struct renameat_args *uap) { return (kern_renameat(td, uap->oldfd, uap->old, uap->newfd, uap->new, UIO_USERSPACE)); } int kern_renameat(struct thread *td, int oldfd, char *old, int newfd, char *new, enum uio_seg pathseg) { struct mount *mp = NULL; struct vnode *tvp, *fvp, *tdvp; struct nameidata fromnd, tond; cap_rights_t rights; int error; again: bwillwrite(); #ifdef MAC NDINIT_ATRIGHTS(&fromnd, DELETE, LOCKPARENT | LOCKLEAF | SAVESTART | AUDITVNODE1, pathseg, old, oldfd, cap_rights_init(&rights, CAP_RENAMEAT_SOURCE), td); #else NDINIT_ATRIGHTS(&fromnd, DELETE, WANTPARENT | SAVESTART | AUDITVNODE1, pathseg, old, oldfd, cap_rights_init(&rights, CAP_RENAMEAT_SOURCE), td); #endif if ((error = namei(&fromnd)) != 0) return (error); #ifdef MAC error = mac_vnode_check_rename_from(td->td_ucred, fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd); VOP_UNLOCK(fromnd.ni_dvp, 0); if (fromnd.ni_dvp != fromnd.ni_vp) VOP_UNLOCK(fromnd.ni_vp, 0); #endif fvp = fromnd.ni_vp; NDINIT_ATRIGHTS(&tond, RENAME, LOCKPARENT | LOCKLEAF | NOCACHE | SAVESTART | AUDITVNODE2, pathseg, new, newfd, cap_rights_init(&rights, CAP_RENAMEAT_TARGET), td); if (fromnd.ni_vp->v_type == VDIR) tond.ni_cnd.cn_flags |= WILLBEDIR; if ((error = namei(&tond)) != 0) { /* Translate error code for rename("dir1", "dir2/."). */ if (error == EISDIR && fvp->v_type == VDIR) error = EINVAL; NDFREE(&fromnd, NDF_ONLY_PNBUF); vrele(fromnd.ni_dvp); vrele(fvp); goto out1; } tdvp = tond.ni_dvp; tvp = tond.ni_vp; error = vn_start_write(fvp, &mp, V_NOWAIT); if (error != 0) { NDFREE(&fromnd, NDF_ONLY_PNBUF); NDFREE(&tond, NDF_ONLY_PNBUF); if (tvp != NULL) vput(tvp); if (tdvp == tvp) vrele(tdvp); else vput(tdvp); vrele(fromnd.ni_dvp); vrele(fvp); vrele(tond.ni_startdir); if (fromnd.ni_startdir != NULL) vrele(fromnd.ni_startdir); error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); if (error != 0) return (error); goto again; } if (tvp != NULL) { if (fvp->v_type == VDIR && tvp->v_type != VDIR) { error = ENOTDIR; goto out; } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) { error = EISDIR; goto out; } #ifdef CAPABILITIES if (newfd != AT_FDCWD) { /* * If the target already exists we require CAP_UNLINKAT * from 'newfd'. */ error = cap_check(&tond.ni_filecaps.fc_rights, cap_rights_init(&rights, CAP_UNLINKAT)); if (error != 0) goto out; } #endif } if (fvp == tdvp) { error = EINVAL; goto out; } /* * If the source is the same as the destination (that is, if they * are links to the same vnode), then there is nothing to do. */ if (fvp == tvp) error = -1; #ifdef MAC else error = mac_vnode_check_rename_to(td->td_ucred, tdvp, tond.ni_vp, fromnd.ni_dvp == tdvp, &tond.ni_cnd); #endif out: if (error == 0) { error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd, tond.ni_dvp, tond.ni_vp, &tond.ni_cnd); NDFREE(&fromnd, NDF_ONLY_PNBUF); NDFREE(&tond, NDF_ONLY_PNBUF); } else { NDFREE(&fromnd, NDF_ONLY_PNBUF); NDFREE(&tond, NDF_ONLY_PNBUF); if (tvp != NULL) vput(tvp); if (tdvp == tvp) vrele(tdvp); else vput(tdvp); vrele(fromnd.ni_dvp); vrele(fvp); } vrele(tond.ni_startdir); vn_finished_write(mp); out1: if (fromnd.ni_startdir) vrele(fromnd.ni_startdir); if (error == -1) return (0); return (error); } /* * Make a directory file. */ #ifndef _SYS_SYSPROTO_H_ struct mkdir_args { char *path; int mode; }; #endif int sys_mkdir(td, uap) struct thread *td; register struct mkdir_args /* { char *path; int mode; } */ *uap; { return (kern_mkdirat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->mode)); } #ifndef _SYS_SYSPROTO_H_ struct mkdirat_args { int fd; char *path; mode_t mode; }; #endif int sys_mkdirat(struct thread *td, struct mkdirat_args *uap) { return (kern_mkdirat(td, uap->fd, uap->path, UIO_USERSPACE, uap->mode)); } int kern_mkdirat(struct thread *td, int fd, char *path, enum uio_seg segflg, int mode) { struct mount *mp; struct vnode *vp; struct vattr vattr; struct nameidata nd; cap_rights_t rights; int error; AUDIT_ARG_MODE(mode); restart: bwillwrite(); NDINIT_ATRIGHTS(&nd, CREATE, LOCKPARENT | SAVENAME | AUDITVNODE1 | NOCACHE, segflg, path, fd, cap_rights_init(&rights, CAP_MKDIRAT), td); nd.ni_cnd.cn_flags |= WILLBEDIR; if ((error = namei(&nd)) != 0) return (error); vp = nd.ni_vp; if (vp != NULL) { NDFREE(&nd, NDF_ONLY_PNBUF); /* * XXX namei called with LOCKPARENT but not LOCKLEAF has * the strange behaviour of leaving the vnode unlocked * if the target is the same vnode as the parent. */ if (vp == nd.ni_dvp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); vrele(vp); return (EEXIST); } if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0) return (error); goto restart; } VATTR_NULL(&vattr); vattr.va_type = VDIR; vattr.va_mode = (mode & ACCESSPERMS) &~ td->td_proc->p_fd->fd_cmask; #ifdef MAC error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, &vattr); if (error != 0) goto out; #endif error = VOP_MKDIR(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); #ifdef MAC out: #endif NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); if (error == 0) vput(nd.ni_vp); vn_finished_write(mp); return (error); } /* * Remove a directory file. */ #ifndef _SYS_SYSPROTO_H_ struct rmdir_args { char *path; }; #endif int sys_rmdir(td, uap) struct thread *td; struct rmdir_args /* { char *path; } */ *uap; { return (kern_rmdirat(td, AT_FDCWD, uap->path, UIO_USERSPACE)); } int kern_rmdirat(struct thread *td, int fd, char *path, enum uio_seg pathseg) { struct mount *mp; struct vnode *vp; struct nameidata nd; cap_rights_t rights; int error; restart: bwillwrite(); NDINIT_ATRIGHTS(&nd, DELETE, LOCKPARENT | LOCKLEAF | AUDITVNODE1, pathseg, path, fd, cap_rights_init(&rights, CAP_UNLINKAT), td); if ((error = namei(&nd)) != 0) return (error); vp = nd.ni_vp; if (vp->v_type != VDIR) { error = ENOTDIR; goto out; } /* * No rmdir "." please. */ if (nd.ni_dvp == vp) { error = EINVAL; goto out; } /* * The root of a mounted filesystem cannot be deleted. */ if (vp->v_vflag & VV_ROOT) { error = EBUSY; goto out; } #ifdef MAC error = mac_vnode_check_unlink(td->td_ucred, nd.ni_dvp, vp, &nd.ni_cnd); if (error != 0) goto out; #endif if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE(&nd, NDF_ONLY_PNBUF); vput(vp); if (nd.ni_dvp == vp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0) return (error); goto restart; } vfs_notify_upper(vp, VFS_NOTIFY_UPPER_UNLINK); error = VOP_RMDIR(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd); vn_finished_write(mp); out: NDFREE(&nd, NDF_ONLY_PNBUF); vput(vp); if (nd.ni_dvp == vp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); return (error); } #ifdef COMPAT_43 /* * Read a block of directory entries in a filesystem independent format. */ #ifndef _SYS_SYSPROTO_H_ struct ogetdirentries_args { int fd; char *buf; u_int count; long *basep; }; #endif int ogetdirentries(struct thread *td, struct ogetdirentries_args *uap) { long loff; int error; error = kern_ogetdirentries(td, uap, &loff); if (error == 0) error = copyout(&loff, uap->basep, sizeof(long)); return (error); } int kern_ogetdirentries(struct thread *td, struct ogetdirentries_args *uap, long *ploff) { struct vnode *vp; struct file *fp; struct uio auio, kuio; struct iovec aiov, kiov; struct dirent *dp, *edp; cap_rights_t rights; caddr_t dirbuf; int error, eofflag, readcnt; long loff; off_t foffset; /* XXX arbitrary sanity limit on `count'. */ if (uap->count > 64 * 1024) return (EINVAL); error = getvnode(td, uap->fd, cap_rights_init(&rights, CAP_READ), &fp); if (error != 0) return (error); if ((fp->f_flag & FREAD) == 0) { fdrop(fp, td); return (EBADF); } vp = fp->f_vnode; foffset = foffset_lock(fp, 0); unionread: if (vp->v_type != VDIR) { foffset_unlock(fp, foffset, 0); fdrop(fp, td); return (EINVAL); } aiov.iov_base = uap->buf; aiov.iov_len = uap->count; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_rw = UIO_READ; auio.uio_segflg = UIO_USERSPACE; auio.uio_td = td; auio.uio_resid = uap->count; vn_lock(vp, LK_SHARED | LK_RETRY); loff = auio.uio_offset = foffset; #ifdef MAC error = mac_vnode_check_readdir(td->td_ucred, vp); if (error != 0) { VOP_UNLOCK(vp, 0); foffset_unlock(fp, foffset, FOF_NOUPDATE); fdrop(fp, td); return (error); } #endif # if (BYTE_ORDER != LITTLE_ENDIAN) if (vp->v_mount->mnt_maxsymlinklen <= 0) { error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL, NULL); foffset = auio.uio_offset; } else # endif { kuio = auio; kuio.uio_iov = &kiov; kuio.uio_segflg = UIO_SYSSPACE; kiov.iov_len = uap->count; dirbuf = malloc(uap->count, M_TEMP, M_WAITOK); kiov.iov_base = dirbuf; error = VOP_READDIR(vp, &kuio, fp->f_cred, &eofflag, NULL, NULL); foffset = kuio.uio_offset; if (error == 0) { readcnt = uap->count - kuio.uio_resid; edp = (struct dirent *)&dirbuf[readcnt]; for (dp = (struct dirent *)dirbuf; dp < edp; ) { # if (BYTE_ORDER == LITTLE_ENDIAN) /* * The expected low byte of * dp->d_namlen is our dp->d_type. * The high MBZ byte of dp->d_namlen * is our dp->d_namlen. */ dp->d_type = dp->d_namlen; dp->d_namlen = 0; # else /* * The dp->d_type is the high byte * of the expected dp->d_namlen, * so must be zero'ed. */ dp->d_type = 0; # endif if (dp->d_reclen > 0) { dp = (struct dirent *) ((char *)dp + dp->d_reclen); } else { error = EIO; break; } } if (dp >= edp) error = uiomove(dirbuf, readcnt, &auio); } free(dirbuf, M_TEMP); } if (error != 0) { VOP_UNLOCK(vp, 0); foffset_unlock(fp, foffset, 0); fdrop(fp, td); return (error); } if (uap->count == auio.uio_resid && (vp->v_vflag & VV_ROOT) && (vp->v_mount->mnt_flag & MNT_UNION)) { struct vnode *tvp = vp; vp = vp->v_mount->mnt_vnodecovered; VREF(vp); fp->f_vnode = vp; fp->f_data = vp; foffset = 0; vput(tvp); goto unionread; } VOP_UNLOCK(vp, 0); foffset_unlock(fp, foffset, 0); fdrop(fp, td); td->td_retval[0] = uap->count - auio.uio_resid; if (error == 0) *ploff = loff; return (error); } #endif /* COMPAT_43 */ /* * Read a block of directory entries in a filesystem independent format. */ #ifndef _SYS_SYSPROTO_H_ struct getdirentries_args { int fd; char *buf; u_int count; long *basep; }; #endif int sys_getdirentries(td, uap) struct thread *td; register struct getdirentries_args /* { int fd; char *buf; u_int count; long *basep; } */ *uap; { long base; int error; error = kern_getdirentries(td, uap->fd, uap->buf, uap->count, &base, NULL, UIO_USERSPACE); if (error != 0) return (error); if (uap->basep != NULL) error = copyout(&base, uap->basep, sizeof(long)); return (error); } int kern_getdirentries(struct thread *td, int fd, char *buf, u_int count, long *basep, ssize_t *residp, enum uio_seg bufseg) { struct vnode *vp; struct file *fp; struct uio auio; struct iovec aiov; cap_rights_t rights; long loff; int error, eofflag; off_t foffset; AUDIT_ARG_FD(fd); if (count > IOSIZE_MAX) return (EINVAL); auio.uio_resid = count; error = getvnode(td, fd, cap_rights_init(&rights, CAP_READ), &fp); if (error != 0) return (error); if ((fp->f_flag & FREAD) == 0) { fdrop(fp, td); return (EBADF); } vp = fp->f_vnode; foffset = foffset_lock(fp, 0); unionread: if (vp->v_type != VDIR) { error = EINVAL; goto fail; } aiov.iov_base = buf; aiov.iov_len = count; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_rw = UIO_READ; auio.uio_segflg = bufseg; auio.uio_td = td; vn_lock(vp, LK_SHARED | LK_RETRY); AUDIT_ARG_VNODE1(vp); loff = auio.uio_offset = foffset; #ifdef MAC error = mac_vnode_check_readdir(td->td_ucred, vp); if (error == 0) #endif error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL, NULL); foffset = auio.uio_offset; if (error != 0) { VOP_UNLOCK(vp, 0); goto fail; } if (count == auio.uio_resid && (vp->v_vflag & VV_ROOT) && (vp->v_mount->mnt_flag & MNT_UNION)) { struct vnode *tvp = vp; vp = vp->v_mount->mnt_vnodecovered; VREF(vp); fp->f_vnode = vp; fp->f_data = vp; foffset = 0; vput(tvp); goto unionread; } VOP_UNLOCK(vp, 0); *basep = loff; if (residp != NULL) *residp = auio.uio_resid; td->td_retval[0] = count - auio.uio_resid; fail: foffset_unlock(fp, foffset, 0); fdrop(fp, td); return (error); } #ifndef _SYS_SYSPROTO_H_ struct getdents_args { int fd; char *buf; size_t count; }; #endif int sys_getdents(td, uap) struct thread *td; register struct getdents_args /* { int fd; char *buf; u_int count; } */ *uap; { struct getdirentries_args ap; ap.fd = uap->fd; ap.buf = uap->buf; ap.count = uap->count; ap.basep = NULL; return (sys_getdirentries(td, &ap)); } /* * Set the mode mask for creation of filesystem nodes. */ #ifndef _SYS_SYSPROTO_H_ struct umask_args { int newmask; }; #endif int sys_umask(td, uap) struct thread *td; struct umask_args /* { int newmask; } */ *uap; { struct filedesc *fdp; fdp = td->td_proc->p_fd; FILEDESC_XLOCK(fdp); td->td_retval[0] = fdp->fd_cmask; fdp->fd_cmask = uap->newmask & ALLPERMS; FILEDESC_XUNLOCK(fdp); return (0); } /* * Void all references to file by ripping underlying filesystem away from * vnode. */ #ifndef _SYS_SYSPROTO_H_ struct revoke_args { char *path; }; #endif int sys_revoke(td, uap) struct thread *td; register struct revoke_args /* { char *path; } */ *uap; { struct vnode *vp; struct vattr vattr; struct nameidata nd; int error; NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_USERSPACE, uap->path, td); if ((error = namei(&nd)) != 0) return (error); vp = nd.ni_vp; NDFREE(&nd, NDF_ONLY_PNBUF); if (vp->v_type != VCHR || vp->v_rdev == NULL) { error = EINVAL; goto out; } #ifdef MAC error = mac_vnode_check_revoke(td->td_ucred, vp); if (error != 0) goto out; #endif error = VOP_GETATTR(vp, &vattr, td->td_ucred); if (error != 0) goto out; if (td->td_ucred->cr_uid != vattr.va_uid) { error = priv_check(td, PRIV_VFS_ADMIN); if (error != 0) goto out; } if (vcount(vp) > 1) VOP_REVOKE(vp, REVOKEALL); out: vput(vp); return (error); } /* * Convert a user file descriptor to a kernel file entry and check that, if it * is a capability, the correct rights are present. A reference on the file * entry is held upon returning. */ int getvnode(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) { struct file *fp; int error; error = fget_unlocked(td->td_proc->p_fd, fd, rightsp, &fp, NULL); if (error != 0) return (error); /* * The file could be not of the vnode type, or it may be not * yet fully initialized, in which case the f_vnode pointer * may be set, but f_ops is still badfileops. E.g., * devfs_open() transiently create such situation to * facilitate csw d_fdopen(). * * Dupfdopen() handling in kern_openat() installs the * half-baked file into the process descriptor table, allowing * other thread to dereference it. Guard against the race by * checking f_ops. */ if (fp->f_vnode == NULL || fp->f_ops == &badfileops) { fdrop(fp, td); return (EINVAL); } *fpp = fp; return (0); } /* * Get an (NFS) file handle. */ #ifndef _SYS_SYSPROTO_H_ struct lgetfh_args { char *fname; fhandle_t *fhp; }; #endif int sys_lgetfh(td, uap) struct thread *td; register struct lgetfh_args *uap; { struct nameidata nd; fhandle_t fh; register struct vnode *vp; int error; error = priv_check(td, PRIV_VFS_GETFH); if (error != 0) return (error); NDINIT(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | AUDITVNODE1, UIO_USERSPACE, uap->fname, td); error = namei(&nd); if (error != 0) return (error); NDFREE(&nd, NDF_ONLY_PNBUF); vp = nd.ni_vp; bzero(&fh, sizeof(fh)); fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid; error = VOP_VPTOFH(vp, &fh.fh_fid); vput(vp); if (error == 0) error = copyout(&fh, uap->fhp, sizeof (fh)); return (error); } #ifndef _SYS_SYSPROTO_H_ struct getfh_args { char *fname; fhandle_t *fhp; }; #endif int sys_getfh(td, uap) struct thread *td; register struct getfh_args *uap; { struct nameidata nd; fhandle_t fh; register struct vnode *vp; int error; error = priv_check(td, PRIV_VFS_GETFH); if (error != 0) return (error); NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_USERSPACE, uap->fname, td); error = namei(&nd); if (error != 0) return (error); NDFREE(&nd, NDF_ONLY_PNBUF); vp = nd.ni_vp; bzero(&fh, sizeof(fh)); fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid; error = VOP_VPTOFH(vp, &fh.fh_fid); vput(vp); if (error == 0) error = copyout(&fh, uap->fhp, sizeof (fh)); return (error); } /* * syscall for the rpc.lockd to use to translate a NFS file handle into an * open descriptor. * * warning: do not remove the priv_check() call or this becomes one giant * security hole. */ #ifndef _SYS_SYSPROTO_H_ struct fhopen_args { const struct fhandle *u_fhp; int flags; }; #endif int sys_fhopen(td, uap) struct thread *td; struct fhopen_args /* { const struct fhandle *u_fhp; int flags; } */ *uap; { struct mount *mp; struct vnode *vp; struct fhandle fhp; struct file *fp; int fmode, error; int indx; error = priv_check(td, PRIV_VFS_FHOPEN); if (error != 0) return (error); indx = -1; fmode = FFLAGS(uap->flags); /* why not allow a non-read/write open for our lockd? */ if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT)) return (EINVAL); error = copyin(uap->u_fhp, &fhp, sizeof(fhp)); if (error != 0) return(error); /* find the mount point */ mp = vfs_busyfs(&fhp.fh_fsid); if (mp == NULL) return (ESTALE); /* now give me my vnode, it gets returned to me locked */ error = VFS_FHTOVP(mp, &fhp.fh_fid, LK_EXCLUSIVE, &vp); vfs_unbusy(mp); if (error != 0) return (error); error = falloc_noinstall(td, &fp); if (error != 0) { vput(vp); return (error); } /* * An extra reference on `fp' has been held for us by * falloc_noinstall(). */ #ifdef INVARIANTS td->td_dupfd = -1; #endif error = vn_open_vnode(vp, fmode, td->td_ucred, td, fp); if (error != 0) { KASSERT(fp->f_ops == &badfileops, ("VOP_OPEN in fhopen() set f_ops")); KASSERT(td->td_dupfd < 0, ("fhopen() encountered fdopen()")); vput(vp); goto bad; } #ifdef INVARIANTS td->td_dupfd = 0; #endif fp->f_vnode = vp; fp->f_seqcount = 1; finit(fp, (fmode & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE, vp, &vnops); VOP_UNLOCK(vp, 0); if ((fmode & O_TRUNC) != 0) { error = fo_truncate(fp, 0, td->td_ucred, td); if (error != 0) goto bad; } error = finstall(td, fp, &indx, fmode, NULL); bad: fdrop(fp, td); td->td_retval[0] = indx; return (error); } /* * Stat an (NFS) file handle. */ #ifndef _SYS_SYSPROTO_H_ struct fhstat_args { struct fhandle *u_fhp; struct stat *sb; }; #endif int sys_fhstat(td, uap) struct thread *td; register struct fhstat_args /* { struct fhandle *u_fhp; struct stat *sb; } */ *uap; { struct stat sb; struct fhandle fh; int error; error = copyin(uap->u_fhp, &fh, sizeof(fh)); if (error != 0) return (error); error = kern_fhstat(td, fh, &sb); if (error == 0) error = copyout(&sb, uap->sb, sizeof(sb)); return (error); } int kern_fhstat(struct thread *td, struct fhandle fh, struct stat *sb) { struct mount *mp; struct vnode *vp; int error; error = priv_check(td, PRIV_VFS_FHSTAT); if (error != 0) return (error); if ((mp = vfs_busyfs(&fh.fh_fsid)) == NULL) return (ESTALE); error = VFS_FHTOVP(mp, &fh.fh_fid, LK_EXCLUSIVE, &vp); vfs_unbusy(mp); if (error != 0) return (error); error = vn_stat(vp, sb, td->td_ucred, NOCRED, td); vput(vp); return (error); } /* * Implement fstatfs() for (NFS) file handles. */ #ifndef _SYS_SYSPROTO_H_ struct fhstatfs_args { struct fhandle *u_fhp; struct statfs *buf; }; #endif int sys_fhstatfs(td, uap) struct thread *td; struct fhstatfs_args /* { struct fhandle *u_fhp; struct statfs *buf; } */ *uap; { struct statfs sf; fhandle_t fh; int error; error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t)); if (error != 0) return (error); error = kern_fhstatfs(td, fh, &sf); if (error != 0) return (error); return (copyout(&sf, uap->buf, sizeof(sf))); } int kern_fhstatfs(struct thread *td, fhandle_t fh, struct statfs *buf) { struct statfs *sp; struct mount *mp; struct vnode *vp; int error; error = priv_check(td, PRIV_VFS_FHSTATFS); if (error != 0) return (error); if ((mp = vfs_busyfs(&fh.fh_fsid)) == NULL) return (ESTALE); error = VFS_FHTOVP(mp, &fh.fh_fid, LK_EXCLUSIVE, &vp); if (error != 0) { vfs_unbusy(mp); return (error); } vput(vp); error = prison_canseemount(td->td_ucred, mp); if (error != 0) goto out; #ifdef MAC error = mac_mount_check_stat(td->td_ucred, mp); if (error != 0) goto out; #endif /* * Set these in case the underlying filesystem fails to do so. */ sp = &mp->mnt_stat; sp->f_version = STATFS_VERSION; sp->f_namemax = NAME_MAX; sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK; error = VFS_STATFS(mp, sp); if (error == 0) *buf = *sp; out: vfs_unbusy(mp); return (error); } int kern_posix_fallocate(struct thread *td, int fd, off_t offset, off_t len) { struct file *fp; struct mount *mp; struct vnode *vp; cap_rights_t rights; off_t olen, ooffset; int error; if (offset < 0 || len <= 0) return (EINVAL); /* Check for wrap. */ if (offset > OFF_MAX - len) return (EFBIG); error = fget(td, fd, cap_rights_init(&rights, CAP_WRITE), &fp); if (error != 0) return (error); if ((fp->f_ops->fo_flags & DFLAG_SEEKABLE) == 0) { error = ESPIPE; goto out; } if ((fp->f_flag & FWRITE) == 0) { error = EBADF; goto out; } if (fp->f_type != DTYPE_VNODE) { error = ENODEV; goto out; } vp = fp->f_vnode; if (vp->v_type != VREG) { error = ENODEV; goto out; } /* Allocating blocks may take a long time, so iterate. */ for (;;) { olen = len; ooffset = offset; bwillwrite(); mp = NULL; error = vn_start_write(vp, &mp, V_WAIT | PCATCH); if (error != 0) break; error = vn_lock(vp, LK_EXCLUSIVE); if (error != 0) { vn_finished_write(mp); break; } #ifdef MAC error = mac_vnode_check_write(td->td_ucred, fp->f_cred, vp); if (error == 0) #endif error = VOP_ALLOCATE(vp, &offset, &len); VOP_UNLOCK(vp, 0); vn_finished_write(mp); if (olen + ooffset != offset + len) { panic("offset + len changed from %jx/%jx to %jx/%jx", ooffset, olen, offset, len); } if (error != 0 || len == 0) break; KASSERT(olen > len, ("Iteration did not make progress?")); maybe_yield(); } out: fdrop(fp, td); return (error); } int sys_posix_fallocate(struct thread *td, struct posix_fallocate_args *uap) { td->td_retval[0] = kern_posix_fallocate(td, uap->fd, uap->offset, uap->len); return (0); } /* * Unlike madvise(2), we do not make a best effort to remember every * possible caching hint. Instead, we remember the last setting with * the exception that we will allow POSIX_FADV_NORMAL to adjust the * region of any current setting. */ int kern_posix_fadvise(struct thread *td, int fd, off_t offset, off_t len, int advice) { struct fadvise_info *fa, *new; struct file *fp; struct vnode *vp; cap_rights_t rights; off_t end; int error; if (offset < 0 || len < 0 || offset > OFF_MAX - len) return (EINVAL); switch (advice) { case POSIX_FADV_SEQUENTIAL: case POSIX_FADV_RANDOM: case POSIX_FADV_NOREUSE: new = malloc(sizeof(*fa), M_FADVISE, M_WAITOK); break; case POSIX_FADV_NORMAL: case POSIX_FADV_WILLNEED: case POSIX_FADV_DONTNEED: new = NULL; break; default: return (EINVAL); } /* XXX: CAP_POSIX_FADVISE? */ error = fget(td, fd, cap_rights_init(&rights), &fp); if (error != 0) goto out; if ((fp->f_ops->fo_flags & DFLAG_SEEKABLE) == 0) { error = ESPIPE; goto out; } if (fp->f_type != DTYPE_VNODE) { error = ENODEV; goto out; } vp = fp->f_vnode; if (vp->v_type != VREG) { error = ENODEV; goto out; } if (len == 0) end = OFF_MAX; else end = offset + len - 1; switch (advice) { case POSIX_FADV_SEQUENTIAL: case POSIX_FADV_RANDOM: case POSIX_FADV_NOREUSE: /* * Try to merge any existing non-standard region with * this new region if possible, otherwise create a new * non-standard region for this request. */ mtx_pool_lock(mtxpool_sleep, fp); fa = fp->f_advice; if (fa != NULL && fa->fa_advice == advice && ((fa->fa_start <= end && fa->fa_end >= offset) || (end != OFF_MAX && fa->fa_start == end + 1) || (fa->fa_end != OFF_MAX && fa->fa_end + 1 == offset))) { if (offset < fa->fa_start) fa->fa_start = offset; if (end > fa->fa_end) fa->fa_end = end; } else { new->fa_advice = advice; new->fa_start = offset; new->fa_end = end; - new->fa_prevstart = 0; - new->fa_prevend = 0; fp->f_advice = new; new = fa; } mtx_pool_unlock(mtxpool_sleep, fp); break; case POSIX_FADV_NORMAL: /* * If a the "normal" region overlaps with an existing * non-standard region, trim or remove the * non-standard region. */ mtx_pool_lock(mtxpool_sleep, fp); fa = fp->f_advice; if (fa != NULL) { if (offset <= fa->fa_start && end >= fa->fa_end) { new = fa; fp->f_advice = NULL; } else if (offset <= fa->fa_start && end >= fa->fa_start) fa->fa_start = end + 1; else if (offset <= fa->fa_end && end >= fa->fa_end) fa->fa_end = offset - 1; else if (offset >= fa->fa_start && end <= fa->fa_end) { /* * If the "normal" region is a middle * portion of the existing * non-standard region, just remove * the whole thing rather than picking * one side or the other to * preserve. */ new = fa; fp->f_advice = NULL; } } mtx_pool_unlock(mtxpool_sleep, fp); break; case POSIX_FADV_WILLNEED: case POSIX_FADV_DONTNEED: error = VOP_ADVISE(vp, offset, end, advice); break; } out: if (fp != NULL) fdrop(fp, td); free(new, M_FADVISE); return (error); } int sys_posix_fadvise(struct thread *td, struct posix_fadvise_args *uap) { td->td_retval[0] = kern_posix_fadvise(td, uap->fd, uap->offset, uap->len, uap->advice); return (0); } Index: head/sys/kern/vfs_vnops.c =================================================================== --- head/sys/kern/vfs_vnops.c (revision 288430) +++ head/sys/kern/vfs_vnops.c (revision 288431) @@ -1,2537 +1,2469 @@ /*- * Copyright (c) 1982, 1986, 1989, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Copyright (c) 2012 Konstantin Belousov * Copyright (c) 2013, 2014 The FreeBSD Foundation * * Portions of this software were developed by Konstantin Belousov * under sponsorship from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)vfs_vnops.c 8.2 (Berkeley) 1/21/94 */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static fo_rdwr_t vn_read; static fo_rdwr_t vn_write; static fo_rdwr_t vn_io_fault; static fo_truncate_t vn_truncate; static fo_ioctl_t vn_ioctl; static fo_poll_t vn_poll; static fo_kqfilter_t vn_kqfilter; static fo_stat_t vn_statfile; static fo_close_t vn_closefile; static fo_mmap_t vn_mmap; struct fileops vnops = { .fo_read = vn_io_fault, .fo_write = vn_io_fault, .fo_truncate = vn_truncate, .fo_ioctl = vn_ioctl, .fo_poll = vn_poll, .fo_kqfilter = vn_kqfilter, .fo_stat = vn_statfile, .fo_close = vn_closefile, .fo_chmod = vn_chmod, .fo_chown = vn_chown, .fo_sendfile = vn_sendfile, .fo_seek = vn_seek, .fo_fill_kinfo = vn_fill_kinfo, .fo_mmap = vn_mmap, .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE }; static const int io_hold_cnt = 16; static int vn_io_fault_enable = 1; SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW, &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance"); static int vn_io_fault_prefault = 0; SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_prefault, CTLFLAG_RW, &vn_io_fault_prefault, 0, "Enable vn_io_fault prefaulting"); static u_long vn_io_faults_cnt; SYSCTL_ULONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD, &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers"); /* * Returns true if vn_io_fault mode of handling the i/o request should * be used. */ static bool do_vn_io_fault(struct vnode *vp, struct uio *uio) { struct mount *mp; return (uio->uio_segflg == UIO_USERSPACE && vp->v_type == VREG && (mp = vp->v_mount) != NULL && (mp->mnt_kern_flag & MNTK_NO_IOPF) != 0 && vn_io_fault_enable); } /* * Structure used to pass arguments to vn_io_fault1(), to do either * file- or vnode-based I/O calls. */ struct vn_io_fault_args { enum { VN_IO_FAULT_FOP, VN_IO_FAULT_VOP } kind; struct ucred *cred; int flags; union { struct fop_args_tag { struct file *fp; fo_rdwr_t *doio; } fop_args; struct vop_args_tag { struct vnode *vp; } vop_args; } args; }; static int vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args, struct thread *td); int vn_open(ndp, flagp, cmode, fp) struct nameidata *ndp; int *flagp, cmode; struct file *fp; { struct thread *td = ndp->ni_cnd.cn_thread; return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp)); } /* * Common code for vnode open operations via a name lookup. * Lookup the vnode and invoke VOP_CREATE if needed. * Check permissions, and call the VOP_OPEN or VOP_CREATE routine. * * Note that this does NOT free nameidata for the successful case, * due to the NDINIT being done elsewhere. */ int vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags, struct ucred *cred, struct file *fp) { struct vnode *vp; struct mount *mp; struct thread *td = ndp->ni_cnd.cn_thread; struct vattr vat; struct vattr *vap = &vat; int fmode, error; restart: fmode = *flagp; if ((fmode & (O_CREAT | O_EXCL | O_DIRECTORY)) == (O_CREAT | O_EXCL | O_DIRECTORY)) return (EINVAL); else if ((fmode & (O_CREAT | O_DIRECTORY)) == O_CREAT) { ndp->ni_cnd.cn_nameiop = CREATE; /* * Set NOCACHE to avoid flushing the cache when * rolling in many files at once. */ ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF | NOCACHE; if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0) ndp->ni_cnd.cn_flags |= FOLLOW; if (!(vn_open_flags & VN_OPEN_NOAUDIT)) ndp->ni_cnd.cn_flags |= AUDITVNODE1; if (vn_open_flags & VN_OPEN_NOCAPCHECK) ndp->ni_cnd.cn_flags |= NOCAPCHECK; bwillwrite(); if ((error = namei(ndp)) != 0) return (error); if (ndp->ni_vp == NULL) { VATTR_NULL(vap); vap->va_type = VREG; vap->va_mode = cmode; if (fmode & O_EXCL) vap->va_vaflags |= VA_EXCLUSIVE; if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE(ndp, NDF_ONLY_PNBUF); vput(ndp->ni_dvp); if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0) return (error); goto restart; } if ((vn_open_flags & VN_OPEN_NAMECACHE) != 0) ndp->ni_cnd.cn_flags |= MAKEENTRY; #ifdef MAC error = mac_vnode_check_create(cred, ndp->ni_dvp, &ndp->ni_cnd, vap); if (error == 0) #endif error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, vap); vput(ndp->ni_dvp); vn_finished_write(mp); if (error) { NDFREE(ndp, NDF_ONLY_PNBUF); return (error); } fmode &= ~O_TRUNC; vp = ndp->ni_vp; } else { if (ndp->ni_dvp == ndp->ni_vp) vrele(ndp->ni_dvp); else vput(ndp->ni_dvp); ndp->ni_dvp = NULL; vp = ndp->ni_vp; if (fmode & O_EXCL) { error = EEXIST; goto bad; } fmode &= ~O_CREAT; } } else { ndp->ni_cnd.cn_nameiop = LOOKUP; ndp->ni_cnd.cn_flags = ISOPEN | ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF; if (!(fmode & FWRITE)) ndp->ni_cnd.cn_flags |= LOCKSHARED; if (!(vn_open_flags & VN_OPEN_NOAUDIT)) ndp->ni_cnd.cn_flags |= AUDITVNODE1; if (vn_open_flags & VN_OPEN_NOCAPCHECK) ndp->ni_cnd.cn_flags |= NOCAPCHECK; if ((error = namei(ndp)) != 0) return (error); vp = ndp->ni_vp; } error = vn_open_vnode(vp, fmode, cred, td, fp); if (error) goto bad; *flagp = fmode; return (0); bad: NDFREE(ndp, NDF_ONLY_PNBUF); vput(vp); *flagp = fmode; ndp->ni_vp = NULL; return (error); } /* * Common code for vnode open operations once a vnode is located. * Check permissions, and call the VOP_OPEN routine. */ int vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred, struct thread *td, struct file *fp) { struct mount *mp; accmode_t accmode; struct flock lf; int error, have_flock, lock_flags, type; if (vp->v_type == VLNK) return (EMLINK); if (vp->v_type == VSOCK) return (EOPNOTSUPP); if (vp->v_type != VDIR && fmode & O_DIRECTORY) return (ENOTDIR); accmode = 0; if (fmode & (FWRITE | O_TRUNC)) { if (vp->v_type == VDIR) return (EISDIR); accmode |= VWRITE; } if (fmode & FREAD) accmode |= VREAD; if (fmode & FEXEC) accmode |= VEXEC; if ((fmode & O_APPEND) && (fmode & FWRITE)) accmode |= VAPPEND; #ifdef MAC if (fmode & O_CREAT) accmode |= VCREAT; if (fmode & O_VERIFY) accmode |= VVERIFY; error = mac_vnode_check_open(cred, vp, accmode); if (error) return (error); accmode &= ~(VCREAT | VVERIFY); #endif if ((fmode & O_CREAT) == 0) { if (accmode & VWRITE) { error = vn_writechk(vp); if (error) return (error); } if (accmode) { error = VOP_ACCESS(vp, accmode, cred, td); if (error) return (error); } } if (vp->v_type == VFIFO && VOP_ISLOCKED(vp) != LK_EXCLUSIVE) vn_lock(vp, LK_UPGRADE | LK_RETRY); if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0) return (error); if (fmode & (O_EXLOCK | O_SHLOCK)) { KASSERT(fp != NULL, ("open with flock requires fp")); lock_flags = VOP_ISLOCKED(vp); VOP_UNLOCK(vp, 0); lf.l_whence = SEEK_SET; lf.l_start = 0; lf.l_len = 0; if (fmode & O_EXLOCK) lf.l_type = F_WRLCK; else lf.l_type = F_RDLCK; type = F_FLOCK; if ((fmode & FNONBLOCK) == 0) type |= F_WAIT; error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type); have_flock = (error == 0); vn_lock(vp, lock_flags | LK_RETRY); if (error == 0 && vp->v_iflag & VI_DOOMED) error = ENOENT; /* * Another thread might have used this vnode as an * executable while the vnode lock was dropped. * Ensure the vnode is still able to be opened for * writing after the lock has been obtained. */ if (error == 0 && accmode & VWRITE) error = vn_writechk(vp); if (error) { VOP_UNLOCK(vp, 0); if (have_flock) { lf.l_whence = SEEK_SET; lf.l_start = 0; lf.l_len = 0; lf.l_type = F_UNLCK; (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK); } vn_start_write(vp, &mp, V_WAIT); vn_lock(vp, lock_flags | LK_RETRY); (void)VOP_CLOSE(vp, fmode, cred, td); vn_finished_write(mp); /* Prevent second close from fdrop()->vn_close(). */ if (fp != NULL) fp->f_ops= &badfileops; return (error); } fp->f_flag |= FHASLOCK; } if (fmode & FWRITE) { VOP_ADD_WRITECOUNT(vp, 1); CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d", __func__, vp, vp->v_writecount); } ASSERT_VOP_LOCKED(vp, "vn_open_vnode"); return (0); } /* * Check for write permissions on the specified vnode. * Prototype text segments cannot be written. */ int vn_writechk(vp) register struct vnode *vp; { ASSERT_VOP_LOCKED(vp, "vn_writechk"); /* * If there's shared text associated with * the vnode, try to free it up once. If * we fail, we can't allow writing. */ if (VOP_IS_TEXT(vp)) return (ETXTBSY); return (0); } /* * Vnode close call */ int vn_close(vp, flags, file_cred, td) register struct vnode *vp; int flags; struct ucred *file_cred; struct thread *td; { struct mount *mp; int error, lock_flags; if (vp->v_type != VFIFO && (flags & FWRITE) == 0 && MNT_EXTENDED_SHARED(vp->v_mount)) lock_flags = LK_SHARED; else lock_flags = LK_EXCLUSIVE; vn_start_write(vp, &mp, V_WAIT); vn_lock(vp, lock_flags | LK_RETRY); if (flags & FWRITE) { VNASSERT(vp->v_writecount > 0, vp, ("vn_close: negative writecount")); VOP_ADD_WRITECOUNT(vp, -1); CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d", __func__, vp, vp->v_writecount); } error = VOP_CLOSE(vp, flags, file_cred, td); vput(vp); vn_finished_write(mp); return (error); } /* * Heuristic to detect sequential operation. */ static int sequential_heuristic(struct uio *uio, struct file *fp) { ASSERT_VOP_LOCKED(fp->f_vnode, __func__); if (fp->f_flag & FRDAHEAD) return (fp->f_seqcount << IO_SEQSHIFT); /* * Offset 0 is handled specially. open() sets f_seqcount to 1 so * that the first I/O is normally considered to be slightly * sequential. Seeking to offset 0 doesn't change sequentiality * unless previous seeks have reduced f_seqcount to 0, in which * case offset 0 is not special. */ if ((uio->uio_offset == 0 && fp->f_seqcount > 0) || uio->uio_offset == fp->f_nextoff) { /* * f_seqcount is in units of fixed-size blocks so that it * depends mainly on the amount of sequential I/O and not * much on the number of sequential I/O's. The fixed size * of 16384 is hard-coded here since it is (not quite) just * a magic size that works well here. This size is more * closely related to the best I/O size for real disks than * to any block size used by software. */ fp->f_seqcount += howmany(uio->uio_resid, 16384); if (fp->f_seqcount > IO_SEQMAX) fp->f_seqcount = IO_SEQMAX; return (fp->f_seqcount << IO_SEQSHIFT); } /* Not sequential. Quickly draw-down sequentiality. */ if (fp->f_seqcount > 1) fp->f_seqcount = 1; else fp->f_seqcount = 0; return (0); } /* * Package up an I/O request on a vnode into a uio and do it. */ int vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset, enum uio_seg segflg, int ioflg, struct ucred *active_cred, struct ucred *file_cred, ssize_t *aresid, struct thread *td) { struct uio auio; struct iovec aiov; struct mount *mp; struct ucred *cred; void *rl_cookie; struct vn_io_fault_args args; int error, lock_flags; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; aiov.iov_base = base; aiov.iov_len = len; auio.uio_resid = len; auio.uio_offset = offset; auio.uio_segflg = segflg; auio.uio_rw = rw; auio.uio_td = td; error = 0; if ((ioflg & IO_NODELOCKED) == 0) { if ((ioflg & IO_RANGELOCKED) == 0) { if (rw == UIO_READ) { rl_cookie = vn_rangelock_rlock(vp, offset, offset + len); } else { rl_cookie = vn_rangelock_wlock(vp, offset, offset + len); } } else rl_cookie = NULL; mp = NULL; if (rw == UIO_WRITE) { if (vp->v_type != VCHR && (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) goto out; if (MNT_SHARED_WRITES(mp) || ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount))) lock_flags = LK_SHARED; else lock_flags = LK_EXCLUSIVE; } else lock_flags = LK_SHARED; vn_lock(vp, lock_flags | LK_RETRY); } else rl_cookie = NULL; ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); #ifdef MAC if ((ioflg & IO_NOMACCHECK) == 0) { if (rw == UIO_READ) error = mac_vnode_check_read(active_cred, file_cred, vp); else error = mac_vnode_check_write(active_cred, file_cred, vp); } #endif if (error == 0) { if (file_cred != NULL) cred = file_cred; else cred = active_cred; if (do_vn_io_fault(vp, &auio)) { args.kind = VN_IO_FAULT_VOP; args.cred = cred; args.flags = ioflg; args.args.vop_args.vp = vp; error = vn_io_fault1(vp, &auio, &args, td); } else if (rw == UIO_READ) { error = VOP_READ(vp, &auio, ioflg, cred); } else /* if (rw == UIO_WRITE) */ { error = VOP_WRITE(vp, &auio, ioflg, cred); } } if (aresid) *aresid = auio.uio_resid; else if (auio.uio_resid && error == 0) error = EIO; if ((ioflg & IO_NODELOCKED) == 0) { VOP_UNLOCK(vp, 0); if (mp != NULL) vn_finished_write(mp); } out: if (rl_cookie != NULL) vn_rangelock_unlock(vp, rl_cookie); return (error); } /* * Package up an I/O request on a vnode into a uio and do it. The I/O * request is split up into smaller chunks and we try to avoid saturating * the buffer cache while potentially holding a vnode locked, so we * check bwillwrite() before calling vn_rdwr(). We also call kern_yield() * to give other processes a chance to lock the vnode (either other processes * core'ing the same binary, or unrelated processes scanning the directory). */ int vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred, file_cred, aresid, td) enum uio_rw rw; struct vnode *vp; void *base; size_t len; off_t offset; enum uio_seg segflg; int ioflg; struct ucred *active_cred; struct ucred *file_cred; size_t *aresid; struct thread *td; { int error = 0; ssize_t iaresid; do { int chunk; /* * Force `offset' to a multiple of MAXBSIZE except possibly * for the first chunk, so that filesystems only need to * write full blocks except possibly for the first and last * chunks. */ chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE; if (chunk > len) chunk = len; if (rw != UIO_READ && vp->v_type == VREG) bwillwrite(); iaresid = 0; error = vn_rdwr(rw, vp, base, chunk, offset, segflg, ioflg, active_cred, file_cred, &iaresid, td); len -= chunk; /* aresid calc already includes length */ if (error) break; offset += chunk; base = (char *)base + chunk; kern_yield(PRI_USER); } while (len); if (aresid) *aresid = len + iaresid; return (error); } off_t foffset_lock(struct file *fp, int flags) { struct mtx *mtxp; off_t res; KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed")); #if OFF_MAX <= LONG_MAX /* * Caller only wants the current f_offset value. Assume that * the long and shorter integer types reads are atomic. */ if ((flags & FOF_NOLOCK) != 0) return (fp->f_offset); #endif /* * According to McKusick the vn lock was protecting f_offset here. * It is now protected by the FOFFSET_LOCKED flag. */ mtxp = mtx_pool_find(mtxpool_sleep, fp); mtx_lock(mtxp); if ((flags & FOF_NOLOCK) == 0) { while (fp->f_vnread_flags & FOFFSET_LOCKED) { fp->f_vnread_flags |= FOFFSET_LOCK_WAITING; msleep(&fp->f_vnread_flags, mtxp, PUSER -1, "vofflock", 0); } fp->f_vnread_flags |= FOFFSET_LOCKED; } res = fp->f_offset; mtx_unlock(mtxp); return (res); } void foffset_unlock(struct file *fp, off_t val, int flags) { struct mtx *mtxp; KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed")); #if OFF_MAX <= LONG_MAX if ((flags & FOF_NOLOCK) != 0) { if ((flags & FOF_NOUPDATE) == 0) fp->f_offset = val; if ((flags & FOF_NEXTOFF) != 0) fp->f_nextoff = val; return; } #endif mtxp = mtx_pool_find(mtxpool_sleep, fp); mtx_lock(mtxp); if ((flags & FOF_NOUPDATE) == 0) fp->f_offset = val; if ((flags & FOF_NEXTOFF) != 0) fp->f_nextoff = val; if ((flags & FOF_NOLOCK) == 0) { KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0, ("Lost FOFFSET_LOCKED")); if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING) wakeup(&fp->f_vnread_flags); fp->f_vnread_flags = 0; } mtx_unlock(mtxp); } void foffset_lock_uio(struct file *fp, struct uio *uio, int flags) { if ((flags & FOF_OFFSET) == 0) uio->uio_offset = foffset_lock(fp, flags); } void foffset_unlock_uio(struct file *fp, struct uio *uio, int flags) { if ((flags & FOF_OFFSET) == 0) foffset_unlock(fp, uio->uio_offset, flags); } static int get_advice(struct file *fp, struct uio *uio) { struct mtx *mtxp; int ret; ret = POSIX_FADV_NORMAL; if (fp->f_advice == NULL) return (ret); mtxp = mtx_pool_find(mtxpool_sleep, fp); mtx_lock(mtxp); if (uio->uio_offset >= fp->f_advice->fa_start && uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end) ret = fp->f_advice->fa_advice; mtx_unlock(mtxp); return (ret); } /* * File table vnode read routine. */ static int vn_read(fp, uio, active_cred, flags, td) struct file *fp; struct uio *uio; struct ucred *active_cred; int flags; struct thread *td; { struct vnode *vp; - struct mtx *mtxp; + off_t orig_offset; int error, ioflag; int advice; - off_t offset, start, end; KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td)); KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET")); vp = fp->f_vnode; ioflag = 0; if (fp->f_flag & FNONBLOCK) ioflag |= IO_NDELAY; if (fp->f_flag & O_DIRECT) ioflag |= IO_DIRECT; advice = get_advice(fp, uio); vn_lock(vp, LK_SHARED | LK_RETRY); switch (advice) { case POSIX_FADV_NORMAL: case POSIX_FADV_SEQUENTIAL: case POSIX_FADV_NOREUSE: ioflag |= sequential_heuristic(uio, fp); break; case POSIX_FADV_RANDOM: /* Disable read-ahead for random I/O. */ break; } - offset = uio->uio_offset; + orig_offset = uio->uio_offset; #ifdef MAC error = mac_vnode_check_read(active_cred, fp->f_cred, vp); if (error == 0) #endif error = VOP_READ(vp, uio, ioflag, fp->f_cred); fp->f_nextoff = uio->uio_offset; VOP_UNLOCK(vp, 0); if (error == 0 && advice == POSIX_FADV_NOREUSE && - offset != uio->uio_offset) { + orig_offset != uio->uio_offset) /* - * Use POSIX_FADV_DONTNEED to flush clean pages and - * buffers for the backing file after a - * POSIX_FADV_NOREUSE read(2). To optimize the common - * case of using POSIX_FADV_NOREUSE with sequential - * access, track the previous implicit DONTNEED - * request and grow this request to include the - * current read(2) in addition to the previous - * DONTNEED. With purely sequential access this will - * cause the DONTNEED requests to continously grow to - * cover all of the previously read regions of the - * file. This allows filesystem blocks that are - * accessed by multiple calls to read(2) to be flushed - * once the last read(2) finishes. + * Use POSIX_FADV_DONTNEED to flush pages and buffers + * for the backing file after a POSIX_FADV_NOREUSE + * read(2). */ - start = offset; - end = uio->uio_offset - 1; - mtxp = mtx_pool_find(mtxpool_sleep, fp); - mtx_lock(mtxp); - if (fp->f_advice != NULL && - fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) { - if (start != 0 && fp->f_advice->fa_prevend + 1 == start) - start = fp->f_advice->fa_prevstart; - else if (fp->f_advice->fa_prevstart != 0 && - fp->f_advice->fa_prevstart == end + 1) - end = fp->f_advice->fa_prevend; - fp->f_advice->fa_prevstart = start; - fp->f_advice->fa_prevend = end; - } - mtx_unlock(mtxp); - error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED); - } + error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1, + POSIX_FADV_DONTNEED); return (error); } /* * File table vnode write routine. */ static int vn_write(fp, uio, active_cred, flags, td) struct file *fp; struct uio *uio; struct ucred *active_cred; int flags; struct thread *td; { struct vnode *vp; struct mount *mp; - struct mtx *mtxp; + off_t orig_offset; int error, ioflag, lock_flags; int advice; - off_t offset, start, end; KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td)); KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET")); vp = fp->f_vnode; if (vp->v_type == VREG) bwillwrite(); ioflag = IO_UNIT; if (vp->v_type == VREG && (fp->f_flag & O_APPEND)) ioflag |= IO_APPEND; if (fp->f_flag & FNONBLOCK) ioflag |= IO_NDELAY; if (fp->f_flag & O_DIRECT) ioflag |= IO_DIRECT; if ((fp->f_flag & O_FSYNC) || (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))) ioflag |= IO_SYNC; mp = NULL; if (vp->v_type != VCHR && (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) goto unlock; advice = get_advice(fp, uio); if (MNT_SHARED_WRITES(mp) || (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) { lock_flags = LK_SHARED; } else { lock_flags = LK_EXCLUSIVE; } vn_lock(vp, lock_flags | LK_RETRY); switch (advice) { case POSIX_FADV_NORMAL: case POSIX_FADV_SEQUENTIAL: case POSIX_FADV_NOREUSE: ioflag |= sequential_heuristic(uio, fp); break; case POSIX_FADV_RANDOM: /* XXX: Is this correct? */ break; } - offset = uio->uio_offset; + orig_offset = uio->uio_offset; #ifdef MAC error = mac_vnode_check_write(active_cred, fp->f_cred, vp); if (error == 0) #endif error = VOP_WRITE(vp, uio, ioflag, fp->f_cred); fp->f_nextoff = uio->uio_offset; VOP_UNLOCK(vp, 0); if (vp->v_type != VCHR) vn_finished_write(mp); if (error == 0 && advice == POSIX_FADV_NOREUSE && - offset != uio->uio_offset) { + orig_offset != uio->uio_offset) /* - * Use POSIX_FADV_DONTNEED to flush clean pages and - * buffers for the backing file after a - * POSIX_FADV_NOREUSE write(2). To optimize the - * common case of using POSIX_FADV_NOREUSE with - * sequential access, track the previous implicit - * DONTNEED request and grow this request to include - * the current write(2) in addition to the previous - * DONTNEED. With purely sequential access this will - * cause the DONTNEED requests to continously grow to - * cover all of the previously written regions of the - * file. - * - * Note that the blocks just written are almost - * certainly still dirty, so this only works when - * VOP_ADVISE() calls from subsequent writes push out - * the data written by this write(2) once the backing - * buffers are clean. However, as compared to forcing - * IO_DIRECT, this gives much saner behavior. Write - * clustering is still allowed, and clean pages are - * merely moved to the cache page queue rather than - * outright thrown away. This means a subsequent - * read(2) can still avoid hitting the disk if the - * pages have not been reclaimed. - * - * This does make POSIX_FADV_NOREUSE largely useless - * with non-sequential access. However, sequential - * access is the more common use case and the flag is - * merely advisory. + * Use POSIX_FADV_DONTNEED to flush pages and buffers + * for the backing file after a POSIX_FADV_NOREUSE + * write(2). */ - start = offset; - end = uio->uio_offset - 1; - mtxp = mtx_pool_find(mtxpool_sleep, fp); - mtx_lock(mtxp); - if (fp->f_advice != NULL && - fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) { - if (start != 0 && fp->f_advice->fa_prevend + 1 == start) - start = fp->f_advice->fa_prevstart; - else if (fp->f_advice->fa_prevstart != 0 && - fp->f_advice->fa_prevstart == end + 1) - end = fp->f_advice->fa_prevend; - fp->f_advice->fa_prevstart = start; - fp->f_advice->fa_prevend = end; - } - mtx_unlock(mtxp); - error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED); - } - + error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1, + POSIX_FADV_DONTNEED); unlock: return (error); } /* * The vn_io_fault() is a wrapper around vn_read() and vn_write() to * prevent the following deadlock: * * Assume that the thread A reads from the vnode vp1 into userspace * buffer buf1 backed by the pages of vnode vp2. If a page in buf1 is * currently not resident, then system ends up with the call chain * vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] -> * vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2) * which establishes lock order vp1->vn_lock, then vp2->vn_lock. * If, at the same time, thread B reads from vnode vp2 into buffer buf2 * backed by the pages of vnode vp1, and some page in buf2 is not * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock. * * To prevent the lock order reversal and deadlock, vn_io_fault() does * not allow page faults to happen during VOP_READ() or VOP_WRITE(). * Instead, it first tries to do the whole range i/o with pagefaults * disabled. If all pages in the i/o buffer are resident and mapped, * VOP will succeed (ignoring the genuine filesystem errors). * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do * i/o in chunks, with all pages in the chunk prefaulted and held * using vm_fault_quick_hold_pages(). * * Filesystems using this deadlock avoidance scheme should use the * array of the held pages from uio, saved in the curthread->td_ma, * instead of doing uiomove(). A helper function * vn_io_fault_uiomove() converts uiomove request into * uiomove_fromphys() over td_ma array. * * Since vnode locks do not cover the whole i/o anymore, rangelocks * make the current i/o request atomic with respect to other i/os and * truncations. */ /* * Decode vn_io_fault_args and perform the corresponding i/o. */ static int vn_io_fault_doio(struct vn_io_fault_args *args, struct uio *uio, struct thread *td) { switch (args->kind) { case VN_IO_FAULT_FOP: return ((args->args.fop_args.doio)(args->args.fop_args.fp, uio, args->cred, args->flags, td)); case VN_IO_FAULT_VOP: if (uio->uio_rw == UIO_READ) { return (VOP_READ(args->args.vop_args.vp, uio, args->flags, args->cred)); } else if (uio->uio_rw == UIO_WRITE) { return (VOP_WRITE(args->args.vop_args.vp, uio, args->flags, args->cred)); } break; } panic("vn_io_fault_doio: unknown kind of io %d %d", args->kind, uio->uio_rw); } static int vn_io_fault_touch(char *base, const struct uio *uio) { int r; r = fubyte(base); if (r == -1 || (uio->uio_rw == UIO_READ && subyte(base, r) == -1)) return (EFAULT); return (0); } static int vn_io_fault_prefault_user(const struct uio *uio) { char *base; const struct iovec *iov; size_t len; ssize_t resid; int error, i; KASSERT(uio->uio_segflg == UIO_USERSPACE, ("vn_io_fault_prefault userspace")); error = i = 0; iov = uio->uio_iov; resid = uio->uio_resid; base = iov->iov_base; len = iov->iov_len; while (resid > 0) { error = vn_io_fault_touch(base, uio); if (error != 0) break; if (len < PAGE_SIZE) { if (len != 0) { error = vn_io_fault_touch(base + len - 1, uio); if (error != 0) break; resid -= len; } if (++i >= uio->uio_iovcnt) break; iov = uio->uio_iov + i; base = iov->iov_base; len = iov->iov_len; } else { len -= PAGE_SIZE; base += PAGE_SIZE; resid -= PAGE_SIZE; } } return (error); } /* * Common code for vn_io_fault(), agnostic to the kind of i/o request. * Uses vn_io_fault_doio() to make the call to an actual i/o function. * Used from vn_rdwr() and vn_io_fault(), which encode the i/o request * into args and call vn_io_fault1() to handle faults during the user * mode buffer accesses. */ static int vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args, struct thread *td) { vm_page_t ma[io_hold_cnt + 2]; struct uio *uio_clone, short_uio; struct iovec short_iovec[1]; vm_page_t *prev_td_ma; vm_prot_t prot; vm_offset_t addr, end; size_t len, resid; ssize_t adv; int error, cnt, save, saveheld, prev_td_ma_cnt; if (vn_io_fault_prefault) { error = vn_io_fault_prefault_user(uio); if (error != 0) return (error); /* Or ignore ? */ } prot = uio->uio_rw == UIO_READ ? VM_PROT_WRITE : VM_PROT_READ; /* * The UFS follows IO_UNIT directive and replays back both * uio_offset and uio_resid if an error is encountered during the * operation. But, since the iovec may be already advanced, * uio is still in an inconsistent state. * * Cache a copy of the original uio, which is advanced to the redo * point using UIO_NOCOPY below. */ uio_clone = cloneuio(uio); resid = uio->uio_resid; short_uio.uio_segflg = UIO_USERSPACE; short_uio.uio_rw = uio->uio_rw; short_uio.uio_td = uio->uio_td; save = vm_fault_disable_pagefaults(); error = vn_io_fault_doio(args, uio, td); if (error != EFAULT) goto out; atomic_add_long(&vn_io_faults_cnt, 1); uio_clone->uio_segflg = UIO_NOCOPY; uiomove(NULL, resid - uio->uio_resid, uio_clone); uio_clone->uio_segflg = uio->uio_segflg; saveheld = curthread_pflags_set(TDP_UIOHELD); prev_td_ma = td->td_ma; prev_td_ma_cnt = td->td_ma_cnt; while (uio_clone->uio_resid != 0) { len = uio_clone->uio_iov->iov_len; if (len == 0) { KASSERT(uio_clone->uio_iovcnt >= 1, ("iovcnt underflow")); uio_clone->uio_iov++; uio_clone->uio_iovcnt--; continue; } if (len > io_hold_cnt * PAGE_SIZE) len = io_hold_cnt * PAGE_SIZE; addr = (uintptr_t)uio_clone->uio_iov->iov_base; end = round_page(addr + len); if (end < addr) { error = EFAULT; break; } cnt = atop(end - trunc_page(addr)); /* * A perfectly misaligned address and length could cause * both the start and the end of the chunk to use partial * page. +2 accounts for such a situation. */ cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map, addr, len, prot, ma, io_hold_cnt + 2); if (cnt == -1) { error = EFAULT; break; } short_uio.uio_iov = &short_iovec[0]; short_iovec[0].iov_base = (void *)addr; short_uio.uio_iovcnt = 1; short_uio.uio_resid = short_iovec[0].iov_len = len; short_uio.uio_offset = uio_clone->uio_offset; td->td_ma = ma; td->td_ma_cnt = cnt; error = vn_io_fault_doio(args, &short_uio, td); vm_page_unhold_pages(ma, cnt); adv = len - short_uio.uio_resid; uio_clone->uio_iov->iov_base = (char *)uio_clone->uio_iov->iov_base + adv; uio_clone->uio_iov->iov_len -= adv; uio_clone->uio_resid -= adv; uio_clone->uio_offset += adv; uio->uio_resid -= adv; uio->uio_offset += adv; if (error != 0 || adv == 0) break; } td->td_ma = prev_td_ma; td->td_ma_cnt = prev_td_ma_cnt; curthread_pflags_restore(saveheld); out: vm_fault_enable_pagefaults(save); free(uio_clone, M_IOV); return (error); } static int vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td) { fo_rdwr_t *doio; struct vnode *vp; void *rl_cookie; struct vn_io_fault_args args; int error; doio = uio->uio_rw == UIO_READ ? vn_read : vn_write; vp = fp->f_vnode; foffset_lock_uio(fp, uio, flags); if (do_vn_io_fault(vp, uio)) { args.kind = VN_IO_FAULT_FOP; args.args.fop_args.fp = fp; args.args.fop_args.doio = doio; args.cred = active_cred; args.flags = flags | FOF_OFFSET; if (uio->uio_rw == UIO_READ) { rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset, uio->uio_offset + uio->uio_resid); } else if ((fp->f_flag & O_APPEND) != 0 || (flags & FOF_OFFSET) == 0) { /* For appenders, punt and lock the whole range. */ rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX); } else { rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset, uio->uio_offset + uio->uio_resid); } error = vn_io_fault1(vp, uio, &args, td); vn_rangelock_unlock(vp, rl_cookie); } else { error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td); } foffset_unlock_uio(fp, uio, flags); return (error); } /* * Helper function to perform the requested uiomove operation using * the held pages for io->uio_iov[0].iov_base buffer instead of * copyin/copyout. Access to the pages with uiomove_fromphys() * instead of iov_base prevents page faults that could occur due to * pmap_collect() invalidating the mapping created by * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or * object cleanup revoking the write access from page mappings. * * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove() * instead of plain uiomove(). */ int vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio) { struct uio transp_uio; struct iovec transp_iov[1]; struct thread *td; size_t adv; int error, pgadv; td = curthread; if ((td->td_pflags & TDP_UIOHELD) == 0 || uio->uio_segflg != UIO_USERSPACE) return (uiomove(data, xfersize, uio)); KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt)); transp_iov[0].iov_base = data; transp_uio.uio_iov = &transp_iov[0]; transp_uio.uio_iovcnt = 1; if (xfersize > uio->uio_resid) xfersize = uio->uio_resid; transp_uio.uio_resid = transp_iov[0].iov_len = xfersize; transp_uio.uio_offset = 0; transp_uio.uio_segflg = UIO_SYSSPACE; /* * Since transp_iov points to data, and td_ma page array * corresponds to original uio->uio_iov, we need to invert the * direction of the i/o operation as passed to * uiomove_fromphys(). */ switch (uio->uio_rw) { case UIO_WRITE: transp_uio.uio_rw = UIO_READ; break; case UIO_READ: transp_uio.uio_rw = UIO_WRITE; break; } transp_uio.uio_td = uio->uio_td; error = uiomove_fromphys(td->td_ma, ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK, xfersize, &transp_uio); adv = xfersize - transp_uio.uio_resid; pgadv = (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) - (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT); td->td_ma += pgadv; KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt, pgadv)); td->td_ma_cnt -= pgadv; uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv; uio->uio_iov->iov_len -= adv; uio->uio_resid -= adv; uio->uio_offset += adv; return (error); } int vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize, struct uio *uio) { struct thread *td; vm_offset_t iov_base; int cnt, pgadv; td = curthread; if ((td->td_pflags & TDP_UIOHELD) == 0 || uio->uio_segflg != UIO_USERSPACE) return (uiomove_fromphys(ma, offset, xfersize, uio)); KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt)); cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize; iov_base = (vm_offset_t)uio->uio_iov->iov_base; switch (uio->uio_rw) { case UIO_WRITE: pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma, offset, cnt); break; case UIO_READ: pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK, cnt); break; } pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT); td->td_ma += pgadv; KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt, pgadv)); td->td_ma_cnt -= pgadv; uio->uio_iov->iov_base = (char *)(iov_base + cnt); uio->uio_iov->iov_len -= cnt; uio->uio_resid -= cnt; uio->uio_offset += cnt; return (0); } /* * File table truncate routine. */ static int vn_truncate(struct file *fp, off_t length, struct ucred *active_cred, struct thread *td) { struct vattr vattr; struct mount *mp; struct vnode *vp; void *rl_cookie; int error; vp = fp->f_vnode; /* * Lock the whole range for truncation. Otherwise split i/o * might happen partly before and partly after the truncation. */ rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX); error = vn_start_write(vp, &mp, V_WAIT | PCATCH); if (error) goto out1; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); if (vp->v_type == VDIR) { error = EISDIR; goto out; } #ifdef MAC error = mac_vnode_check_write(active_cred, fp->f_cred, vp); if (error) goto out; #endif error = vn_writechk(vp); if (error == 0) { VATTR_NULL(&vattr); vattr.va_size = length; error = VOP_SETATTR(vp, &vattr, fp->f_cred); } out: VOP_UNLOCK(vp, 0); vn_finished_write(mp); out1: vn_rangelock_unlock(vp, rl_cookie); return (error); } /* * File table vnode stat routine. */ static int vn_statfile(fp, sb, active_cred, td) struct file *fp; struct stat *sb; struct ucred *active_cred; struct thread *td; { struct vnode *vp = fp->f_vnode; int error; vn_lock(vp, LK_SHARED | LK_RETRY); error = vn_stat(vp, sb, active_cred, fp->f_cred, td); VOP_UNLOCK(vp, 0); return (error); } /* * Stat a vnode; implementation for the stat syscall */ int vn_stat(vp, sb, active_cred, file_cred, td) struct vnode *vp; register struct stat *sb; struct ucred *active_cred; struct ucred *file_cred; struct thread *td; { struct vattr vattr; register struct vattr *vap; int error; u_short mode; #ifdef MAC error = mac_vnode_check_stat(active_cred, file_cred, vp); if (error) return (error); #endif vap = &vattr; /* * Initialize defaults for new and unusual fields, so that file * systems which don't support these fields don't need to know * about them. */ vap->va_birthtime.tv_sec = -1; vap->va_birthtime.tv_nsec = 0; vap->va_fsid = VNOVAL; vap->va_rdev = NODEV; error = VOP_GETATTR(vp, vap, active_cred); if (error) return (error); /* * Zero the spare stat fields */ bzero(sb, sizeof *sb); /* * Copy from vattr table */ if (vap->va_fsid != VNOVAL) sb->st_dev = vap->va_fsid; else sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0]; sb->st_ino = vap->va_fileid; mode = vap->va_mode; switch (vap->va_type) { case VREG: mode |= S_IFREG; break; case VDIR: mode |= S_IFDIR; break; case VBLK: mode |= S_IFBLK; break; case VCHR: mode |= S_IFCHR; break; case VLNK: mode |= S_IFLNK; break; case VSOCK: mode |= S_IFSOCK; break; case VFIFO: mode |= S_IFIFO; break; default: return (EBADF); }; sb->st_mode = mode; sb->st_nlink = vap->va_nlink; sb->st_uid = vap->va_uid; sb->st_gid = vap->va_gid; sb->st_rdev = vap->va_rdev; if (vap->va_size > OFF_MAX) return (EOVERFLOW); sb->st_size = vap->va_size; sb->st_atim = vap->va_atime; sb->st_mtim = vap->va_mtime; sb->st_ctim = vap->va_ctime; sb->st_birthtim = vap->va_birthtime; /* * According to www.opengroup.org, the meaning of st_blksize is * "a filesystem-specific preferred I/O block size for this * object. In some filesystem types, this may vary from file * to file" * Use miminum/default of PAGE_SIZE (e.g. for VCHR). */ sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize); sb->st_flags = vap->va_flags; if (priv_check(td, PRIV_VFS_GENERATION)) sb->st_gen = 0; else sb->st_gen = vap->va_gen; sb->st_blocks = vap->va_bytes / S_BLKSIZE; return (0); } /* * File table vnode ioctl routine. */ static int vn_ioctl(fp, com, data, active_cred, td) struct file *fp; u_long com; void *data; struct ucred *active_cred; struct thread *td; { struct vattr vattr; struct vnode *vp; int error; vp = fp->f_vnode; switch (vp->v_type) { case VDIR: case VREG: switch (com) { case FIONREAD: vn_lock(vp, LK_SHARED | LK_RETRY); error = VOP_GETATTR(vp, &vattr, active_cred); VOP_UNLOCK(vp, 0); if (error == 0) *(int *)data = vattr.va_size - fp->f_offset; return (error); case FIONBIO: case FIOASYNC: return (0); default: return (VOP_IOCTL(vp, com, data, fp->f_flag, active_cred, td)); } default: return (ENOTTY); } } /* * File table vnode poll routine. */ static int vn_poll(fp, events, active_cred, td) struct file *fp; int events; struct ucred *active_cred; struct thread *td; { struct vnode *vp; int error; vp = fp->f_vnode; #ifdef MAC vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); error = mac_vnode_check_poll(active_cred, fp->f_cred, vp); VOP_UNLOCK(vp, 0); if (!error) #endif error = VOP_POLL(vp, events, fp->f_cred, td); return (error); } /* * Acquire the requested lock and then check for validity. LK_RETRY * permits vn_lock to return doomed vnodes. */ int _vn_lock(struct vnode *vp, int flags, char *file, int line) { int error; VNASSERT((flags & LK_TYPE_MASK) != 0, vp, ("vn_lock called with no locktype.")); do { #ifdef DEBUG_VFS_LOCKS KASSERT(vp->v_holdcnt != 0, ("vn_lock %p: zero hold count", vp)); #endif error = VOP_LOCK1(vp, flags, file, line); flags &= ~LK_INTERLOCK; /* Interlock is always dropped. */ KASSERT((flags & LK_RETRY) == 0 || error == 0, ("LK_RETRY set with incompatible flags (0x%x) or an error occured (%d)", flags, error)); /* * Callers specify LK_RETRY if they wish to get dead vnodes. * If RETRY is not set, we return ENOENT instead. */ if (error == 0 && vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0) { VOP_UNLOCK(vp, 0); error = ENOENT; break; } } while (flags & LK_RETRY && error != 0); return (error); } /* * File table vnode close routine. */ static int vn_closefile(fp, td) struct file *fp; struct thread *td; { struct vnode *vp; struct flock lf; int error; vp = fp->f_vnode; fp->f_ops = &badfileops; if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) vref(vp); error = vn_close(vp, fp->f_flag, fp->f_cred, td); if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) { lf.l_whence = SEEK_SET; lf.l_start = 0; lf.l_len = 0; lf.l_type = F_UNLCK; (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK); vrele(vp); } return (error); } static bool vn_suspendable(struct mount *mp) { return (mp->mnt_op->vfs_susp_clean != NULL); } /* * Preparing to start a filesystem write operation. If the operation is * permitted, then we bump the count of operations in progress and * proceed. If a suspend request is in progress, we wait until the * suspension is over, and then proceed. */ static int vn_start_write_locked(struct mount *mp, int flags) { int error, mflags; mtx_assert(MNT_MTX(mp), MA_OWNED); error = 0; /* * Check on status of suspension. */ if ((curthread->td_pflags & TDP_IGNSUSP) == 0 || mp->mnt_susp_owner != curthread) { mflags = ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ? (flags & PCATCH) : 0) | (PUSER - 1); while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) { if (flags & V_NOWAIT) { error = EWOULDBLOCK; goto unlock; } error = msleep(&mp->mnt_flag, MNT_MTX(mp), mflags, "suspfs", 0); if (error) goto unlock; } } if (flags & V_XSLEEP) goto unlock; mp->mnt_writeopcount++; unlock: if (error != 0 || (flags & V_XSLEEP) != 0) MNT_REL(mp); MNT_IUNLOCK(mp); return (error); } int vn_start_write(struct vnode *vp, struct mount **mpp, int flags) { struct mount *mp; int error; KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL), ("V_MNTREF requires mp")); error = 0; /* * If a vnode is provided, get and return the mount point that * to which it will write. */ if (vp != NULL) { if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) { *mpp = NULL; if (error != EOPNOTSUPP) return (error); return (0); } } if ((mp = *mpp) == NULL) return (0); if (!vn_suspendable(mp)) { if (vp != NULL || (flags & V_MNTREF) != 0) vfs_rel(mp); return (0); } /* * VOP_GETWRITEMOUNT() returns with the mp refcount held through * a vfs_ref(). * As long as a vnode is not provided we need to acquire a * refcount for the provided mountpoint too, in order to * emulate a vfs_ref(). */ MNT_ILOCK(mp); if (vp == NULL && (flags & V_MNTREF) == 0) MNT_REF(mp); return (vn_start_write_locked(mp, flags)); } /* * Secondary suspension. Used by operations such as vop_inactive * routines that are needed by the higher level functions. These * are allowed to proceed until all the higher level functions have * completed (indicated by mnt_writeopcount dropping to zero). At that * time, these operations are halted until the suspension is over. */ int vn_start_secondary_write(struct vnode *vp, struct mount **mpp, int flags) { struct mount *mp; int error; KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL), ("V_MNTREF requires mp")); retry: if (vp != NULL) { if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) { *mpp = NULL; if (error != EOPNOTSUPP) return (error); return (0); } } /* * If we are not suspended or have not yet reached suspended * mode, then let the operation proceed. */ if ((mp = *mpp) == NULL) return (0); if (!vn_suspendable(mp)) { if (vp != NULL || (flags & V_MNTREF) != 0) vfs_rel(mp); return (0); } /* * VOP_GETWRITEMOUNT() returns with the mp refcount held through * a vfs_ref(). * As long as a vnode is not provided we need to acquire a * refcount for the provided mountpoint too, in order to * emulate a vfs_ref(). */ MNT_ILOCK(mp); if (vp == NULL && (flags & V_MNTREF) == 0) MNT_REF(mp); if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) { mp->mnt_secondary_writes++; mp->mnt_secondary_accwrites++; MNT_IUNLOCK(mp); return (0); } if (flags & V_NOWAIT) { MNT_REL(mp); MNT_IUNLOCK(mp); return (EWOULDBLOCK); } /* * Wait for the suspension to finish. */ error = msleep(&mp->mnt_flag, MNT_MTX(mp), (PUSER - 1) | PDROP | ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ? (flags & PCATCH) : 0), "suspfs", 0); vfs_rel(mp); if (error == 0) goto retry; return (error); } /* * Filesystem write operation has completed. If we are suspending and this * operation is the last one, notify the suspender that the suspension is * now in effect. */ void vn_finished_write(mp) struct mount *mp; { if (mp == NULL || !vn_suspendable(mp)) return; MNT_ILOCK(mp); MNT_REL(mp); mp->mnt_writeopcount--; if (mp->mnt_writeopcount < 0) panic("vn_finished_write: neg cnt"); if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 && mp->mnt_writeopcount <= 0) wakeup(&mp->mnt_writeopcount); MNT_IUNLOCK(mp); } /* * Filesystem secondary write operation has completed. If we are * suspending and this operation is the last one, notify the suspender * that the suspension is now in effect. */ void vn_finished_secondary_write(mp) struct mount *mp; { if (mp == NULL || !vn_suspendable(mp)) return; MNT_ILOCK(mp); MNT_REL(mp); mp->mnt_secondary_writes--; if (mp->mnt_secondary_writes < 0) panic("vn_finished_secondary_write: neg cnt"); if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 && mp->mnt_secondary_writes <= 0) wakeup(&mp->mnt_secondary_writes); MNT_IUNLOCK(mp); } /* * Request a filesystem to suspend write operations. */ int vfs_write_suspend(struct mount *mp, int flags) { int error; MPASS(vn_suspendable(mp)); MNT_ILOCK(mp); if (mp->mnt_susp_owner == curthread) { MNT_IUNLOCK(mp); return (EALREADY); } while (mp->mnt_kern_flag & MNTK_SUSPEND) msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0); /* * Unmount holds a write reference on the mount point. If we * own busy reference and drain for writers, we deadlock with * the reference draining in the unmount path. Callers of * vfs_write_suspend() must specify VS_SKIP_UNMOUNT if * vfs_busy() reference is owned and caller is not in the * unmount context. */ if ((flags & VS_SKIP_UNMOUNT) != 0 && (mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) { MNT_IUNLOCK(mp); return (EBUSY); } mp->mnt_kern_flag |= MNTK_SUSPEND; mp->mnt_susp_owner = curthread; if (mp->mnt_writeopcount > 0) (void) msleep(&mp->mnt_writeopcount, MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0); else MNT_IUNLOCK(mp); if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0) vfs_write_resume(mp, 0); return (error); } /* * Request a filesystem to resume write operations. */ void vfs_write_resume(struct mount *mp, int flags) { MPASS(vn_suspendable(mp)); MNT_ILOCK(mp); if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) { KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner")); mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 | MNTK_SUSPENDED); mp->mnt_susp_owner = NULL; wakeup(&mp->mnt_writeopcount); wakeup(&mp->mnt_flag); curthread->td_pflags &= ~TDP_IGNSUSP; if ((flags & VR_START_WRITE) != 0) { MNT_REF(mp); mp->mnt_writeopcount++; } MNT_IUNLOCK(mp); if ((flags & VR_NO_SUSPCLR) == 0) VFS_SUSP_CLEAN(mp); } else if ((flags & VR_START_WRITE) != 0) { MNT_REF(mp); vn_start_write_locked(mp, 0); } else { MNT_IUNLOCK(mp); } } /* * Helper loop around vfs_write_suspend() for filesystem unmount VFS * methods. */ int vfs_write_suspend_umnt(struct mount *mp) { int error; MPASS(vn_suspendable(mp)); KASSERT((curthread->td_pflags & TDP_IGNSUSP) == 0, ("vfs_write_suspend_umnt: recursed")); /* dounmount() already called vn_start_write(). */ for (;;) { vn_finished_write(mp); error = vfs_write_suspend(mp, 0); if (error != 0) { vn_start_write(NULL, &mp, V_WAIT); return (error); } MNT_ILOCK(mp); if ((mp->mnt_kern_flag & MNTK_SUSPENDED) != 0) break; MNT_IUNLOCK(mp); vn_start_write(NULL, &mp, V_WAIT); } mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2); wakeup(&mp->mnt_flag); MNT_IUNLOCK(mp); curthread->td_pflags |= TDP_IGNSUSP; return (0); } /* * Implement kqueues for files by translating it to vnode operation. */ static int vn_kqfilter(struct file *fp, struct knote *kn) { return (VOP_KQFILTER(fp->f_vnode, kn)); } /* * Simplified in-kernel wrapper calls for extended attribute access. * Both calls pass in a NULL credential, authorizing as "kernel" access. * Set IO_NODELOCKED in ioflg if the vnode is already locked. */ int vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace, const char *attrname, int *buflen, char *buf, struct thread *td) { struct uio auio; struct iovec iov; int error; iov.iov_len = *buflen; iov.iov_base = buf; auio.uio_iov = &iov; auio.uio_iovcnt = 1; auio.uio_rw = UIO_READ; auio.uio_segflg = UIO_SYSSPACE; auio.uio_td = td; auio.uio_offset = 0; auio.uio_resid = *buflen; if ((ioflg & IO_NODELOCKED) == 0) vn_lock(vp, LK_SHARED | LK_RETRY); ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); /* authorize attribute retrieval as kernel */ error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL, td); if ((ioflg & IO_NODELOCKED) == 0) VOP_UNLOCK(vp, 0); if (error == 0) { *buflen = *buflen - auio.uio_resid; } return (error); } /* * XXX failure mode if partially written? */ int vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace, const char *attrname, int buflen, char *buf, struct thread *td) { struct uio auio; struct iovec iov; struct mount *mp; int error; iov.iov_len = buflen; iov.iov_base = buf; auio.uio_iov = &iov; auio.uio_iovcnt = 1; auio.uio_rw = UIO_WRITE; auio.uio_segflg = UIO_SYSSPACE; auio.uio_td = td; auio.uio_offset = 0; auio.uio_resid = buflen; if ((ioflg & IO_NODELOCKED) == 0) { if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0) return (error); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); } ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); /* authorize attribute setting as kernel */ error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td); if ((ioflg & IO_NODELOCKED) == 0) { vn_finished_write(mp); VOP_UNLOCK(vp, 0); } return (error); } int vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace, const char *attrname, struct thread *td) { struct mount *mp; int error; if ((ioflg & IO_NODELOCKED) == 0) { if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0) return (error); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); } ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); /* authorize attribute removal as kernel */ error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td); if (error == EOPNOTSUPP) error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL, NULL, td); if ((ioflg & IO_NODELOCKED) == 0) { vn_finished_write(mp); VOP_UNLOCK(vp, 0); } return (error); } static int vn_get_ino_alloc_vget(struct mount *mp, void *arg, int lkflags, struct vnode **rvp) { return (VFS_VGET(mp, *(ino_t *)arg, lkflags, rvp)); } int vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp) { return (vn_vget_ino_gen(vp, vn_get_ino_alloc_vget, &ino, lkflags, rvp)); } int vn_vget_ino_gen(struct vnode *vp, vn_get_ino_t alloc, void *alloc_arg, int lkflags, struct vnode **rvp) { struct mount *mp; int ltype, error; ASSERT_VOP_LOCKED(vp, "vn_vget_ino_get"); mp = vp->v_mount; ltype = VOP_ISLOCKED(vp); KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED, ("vn_vget_ino: vp not locked")); error = vfs_busy(mp, MBF_NOWAIT); if (error != 0) { vfs_ref(mp); VOP_UNLOCK(vp, 0); error = vfs_busy(mp, 0); vn_lock(vp, ltype | LK_RETRY); vfs_rel(mp); if (error != 0) return (ENOENT); if (vp->v_iflag & VI_DOOMED) { vfs_unbusy(mp); return (ENOENT); } } VOP_UNLOCK(vp, 0); error = alloc(mp, alloc_arg, lkflags, rvp); vfs_unbusy(mp); if (*rvp != vp) vn_lock(vp, ltype | LK_RETRY); if (vp->v_iflag & VI_DOOMED) { if (error == 0) { if (*rvp == vp) vunref(vp); else vput(*rvp); } error = ENOENT; } return (error); } int vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio, struct thread *td) { if (vp->v_type != VREG || td == NULL) return (0); if ((uoff_t)uio->uio_offset + uio->uio_resid > lim_cur(td, RLIMIT_FSIZE)) { PROC_LOCK(td->td_proc); kern_psignal(td->td_proc, SIGXFSZ); PROC_UNLOCK(td->td_proc); return (EFBIG); } return (0); } int vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, struct thread *td) { struct vnode *vp; vp = fp->f_vnode; #ifdef AUDIT vn_lock(vp, LK_SHARED | LK_RETRY); AUDIT_ARG_VNODE1(vp); VOP_UNLOCK(vp, 0); #endif return (setfmode(td, active_cred, vp, mode)); } int vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, struct thread *td) { struct vnode *vp; vp = fp->f_vnode; #ifdef AUDIT vn_lock(vp, LK_SHARED | LK_RETRY); AUDIT_ARG_VNODE1(vp); VOP_UNLOCK(vp, 0); #endif return (setfown(td, active_cred, vp, uid, gid)); } void vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end) { vm_object_t object; if ((object = vp->v_object) == NULL) return; VM_OBJECT_WLOCK(object); vm_object_page_remove(object, start, end, 0); VM_OBJECT_WUNLOCK(object); } int vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred) { struct vattr va; daddr_t bn, bnp; uint64_t bsize; off_t noff; int error; KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA, ("Wrong command %lu", cmd)); if (vn_lock(vp, LK_SHARED) != 0) return (EBADF); if (vp->v_type != VREG) { error = ENOTTY; goto unlock; } error = VOP_GETATTR(vp, &va, cred); if (error != 0) goto unlock; noff = *off; if (noff >= va.va_size) { error = ENXIO; goto unlock; } bsize = vp->v_mount->mnt_stat.f_iosize; for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) { error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL); if (error == EOPNOTSUPP) { error = ENOTTY; goto unlock; } if ((bnp == -1 && cmd == FIOSEEKHOLE) || (bnp != -1 && cmd == FIOSEEKDATA)) { noff = bn * bsize; if (noff < *off) noff = *off; goto unlock; } } if (noff > va.va_size) noff = va.va_size; /* noff == va.va_size. There is an implicit hole at the end of file. */ if (cmd == FIOSEEKDATA) error = ENXIO; unlock: VOP_UNLOCK(vp, 0); if (error == 0) *off = noff; return (error); } int vn_seek(struct file *fp, off_t offset, int whence, struct thread *td) { struct ucred *cred; struct vnode *vp; struct vattr vattr; off_t foffset, size; int error, noneg; cred = td->td_ucred; vp = fp->f_vnode; foffset = foffset_lock(fp, 0); noneg = (vp->v_type != VCHR); error = 0; switch (whence) { case L_INCR: if (noneg && (foffset < 0 || (offset > 0 && foffset > OFF_MAX - offset))) { error = EOVERFLOW; break; } offset += foffset; break; case L_XTND: vn_lock(vp, LK_SHARED | LK_RETRY); error = VOP_GETATTR(vp, &vattr, cred); VOP_UNLOCK(vp, 0); if (error) break; /* * If the file references a disk device, then fetch * the media size and use that to determine the ending * offset. */ if (vattr.va_size == 0 && vp->v_type == VCHR && fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0) vattr.va_size = size; if (noneg && (vattr.va_size > OFF_MAX || (offset > 0 && vattr.va_size > OFF_MAX - offset))) { error = EOVERFLOW; break; } offset += vattr.va_size; break; case L_SET: break; case SEEK_DATA: error = fo_ioctl(fp, FIOSEEKDATA, &offset, cred, td); break; case SEEK_HOLE: error = fo_ioctl(fp, FIOSEEKHOLE, &offset, cred, td); break; default: error = EINVAL; } if (error == 0 && noneg && offset < 0) error = EINVAL; if (error != 0) goto drop; VFS_KNOTE_UNLOCKED(vp, 0); td->td_uretoff.tdu_off = offset; drop: foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0); return (error); } int vn_utimes_perm(struct vnode *vp, struct vattr *vap, struct ucred *cred, struct thread *td) { int error; /* * Grant permission if the caller is the owner of the file, or * the super-user, or has ACL_WRITE_ATTRIBUTES permission on * on the file. If the time pointer is null, then write * permission on the file is also sufficient. * * From NFSv4.1, draft 21, 6.2.1.3.1, Discussion of Mask Attributes: * A user having ACL_WRITE_DATA or ACL_WRITE_ATTRIBUTES * will be allowed to set the times [..] to the current * server time. */ error = VOP_ACCESSX(vp, VWRITE_ATTRIBUTES, cred, td); if (error != 0 && (vap->va_vaflags & VA_UTIMES_NULL) != 0) error = VOP_ACCESS(vp, VWRITE, cred, td); return (error); } int vn_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) { struct vnode *vp; int error; if (fp->f_type == DTYPE_FIFO) kif->kf_type = KF_TYPE_FIFO; else kif->kf_type = KF_TYPE_VNODE; vp = fp->f_vnode; vref(vp); FILEDESC_SUNLOCK(fdp); error = vn_fill_kinfo_vnode(vp, kif); vrele(vp); FILEDESC_SLOCK(fdp); return (error); } static inline void vn_fill_junk(struct kinfo_file *kif) { size_t len, olen; /* * Simulate vn_fullpath returning changing values for a given * vp during e.g. coredump. */ len = (arc4random() % (sizeof(kif->kf_path) - 2)) + 1; olen = strlen(kif->kf_path); if (len < olen) strcpy(&kif->kf_path[len - 1], "$"); else for (; olen < len; olen++) strcpy(&kif->kf_path[olen], "A"); } int vn_fill_kinfo_vnode(struct vnode *vp, struct kinfo_file *kif) { struct vattr va; char *fullpath, *freepath; int error; kif->kf_vnode_type = vntype_to_kinfo(vp->v_type); freepath = NULL; fullpath = "-"; error = vn_fullpath(curthread, vp, &fullpath, &freepath); if (error == 0) { strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path)); } if (freepath != NULL) free(freepath, M_TEMP); KFAIL_POINT_CODE(DEBUG_FP, fill_kinfo_vnode__random_path, vn_fill_junk(kif); ); /* * Retrieve vnode attributes. */ va.va_fsid = VNOVAL; va.va_rdev = NODEV; vn_lock(vp, LK_SHARED | LK_RETRY); error = VOP_GETATTR(vp, &va, curthread->td_ucred); VOP_UNLOCK(vp, 0); if (error != 0) return (error); if (va.va_fsid != VNOVAL) kif->kf_un.kf_file.kf_file_fsid = va.va_fsid; else kif->kf_un.kf_file.kf_file_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; kif->kf_un.kf_file.kf_file_fileid = va.va_fileid; kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode); kif->kf_un.kf_file.kf_file_size = va.va_size; kif->kf_un.kf_file.kf_file_rdev = va.va_rdev; return (0); } int vn_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff, struct thread *td) { #ifdef HWPMC_HOOKS struct pmckern_map_in pkm; #endif struct mount *mp; struct vnode *vp; vm_object_t object; vm_prot_t maxprot; boolean_t writecounted; int error; #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \ defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) /* * POSIX shared-memory objects are defined to have * kernel persistence, and are not defined to support * read(2)/write(2) -- or even open(2). Thus, we can * use MAP_ASYNC to trade on-disk coherence for speed. * The shm_open(3) library routine turns on the FPOSIXSHM * flag to request this behavior. */ if ((fp->f_flag & FPOSIXSHM) != 0) flags |= MAP_NOSYNC; #endif vp = fp->f_vnode; /* * Ensure that file and memory protections are * compatible. Note that we only worry about * writability if mapping is shared; in this case, * current and max prot are dictated by the open file. * XXX use the vnode instead? Problem is: what * credentials do we use for determination? What if * proc does a setuid? */ mp = vp->v_mount; if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) maxprot = VM_PROT_NONE; else maxprot = VM_PROT_EXECUTE; if ((fp->f_flag & FREAD) != 0) maxprot |= VM_PROT_READ; else if ((prot & VM_PROT_READ) != 0) return (EACCES); /* * If we are sharing potential changes via MAP_SHARED and we * are trying to get write permission although we opened it * without asking for it, bail out. */ if ((flags & MAP_SHARED) != 0) { if ((fp->f_flag & FWRITE) != 0) maxprot |= VM_PROT_WRITE; else if ((prot & VM_PROT_WRITE) != 0) return (EACCES); } else { maxprot |= VM_PROT_WRITE; cap_maxprot |= VM_PROT_WRITE; } maxprot &= cap_maxprot; writecounted = FALSE; error = vm_mmap_vnode(td, size, prot, &maxprot, &flags, vp, &foff, &object, &writecounted); if (error != 0) return (error); error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object, foff, writecounted, td); if (error != 0) { /* * If this mapping was accounted for in the vnode's * writecount, then undo that now. */ if (writecounted) vnode_pager_release_writecount(object, 0, size); vm_object_deallocate(object); } #ifdef HWPMC_HOOKS /* Inform hwpmc(4) if an executable is being mapped. */ if (error == 0 && (prot & VM_PROT_EXECUTE) != 0) { pkm.pm_file = vp; pkm.pm_address = (uintptr_t) addr; PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm); } #endif return (error); } Index: head/sys/sys/buf.h =================================================================== --- head/sys/sys/buf.h (revision 288430) +++ head/sys/sys/buf.h (revision 288431) @@ -1,553 +1,553 @@ /*- * Copyright (c) 1982, 1986, 1989, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)buf.h 8.9 (Berkeley) 3/30/95 * $FreeBSD$ */ #ifndef _SYS_BUF_H_ #define _SYS_BUF_H_ #include #include #include #include struct bio; struct buf; struct bufobj; struct mount; struct vnode; struct uio; /* * To avoid including */ LIST_HEAD(workhead, worklist); /* * These are currently used only by the soft dependency code, hence * are stored once in a global variable. If other subsystems wanted * to use these hooks, a pointer to a set of bio_ops could be added * to each buffer. */ extern struct bio_ops { void (*io_start)(struct buf *); void (*io_complete)(struct buf *); void (*io_deallocate)(struct buf *); int (*io_countdeps)(struct buf *, int); } bioops; struct vm_object; typedef unsigned char b_xflags_t; /* * The buffer header describes an I/O operation in the kernel. * * NOTES: * b_bufsize, b_bcount. b_bufsize is the allocation size of the * buffer, either DEV_BSIZE or PAGE_SIZE aligned. b_bcount is the * originally requested buffer size and can serve as a bounds check * against EOF. For most, but not all uses, b_bcount == b_bufsize. * * b_dirtyoff, b_dirtyend. Buffers support piecemeal, unaligned * ranges of dirty data that need to be written to backing store. * The range is typically clipped at b_bcount ( not b_bufsize ). * * b_resid. Number of bytes remaining in I/O. After an I/O operation * completes, b_resid is usually 0 indicating 100% success. * * All fields are protected by the buffer lock except those marked: * V - Protected by owning bufobj lock * Q - Protected by the buf queue lock * D - Protected by an dependency implementation specific lock */ struct buf { struct bufobj *b_bufobj; long b_bcount; void *b_caller1; caddr_t b_data; int b_error; uint8_t b_iocmd; uint8_t b_ioflags; off_t b_iooffset; long b_resid; void (*b_iodone)(struct buf *); daddr_t b_blkno; /* Underlying physical block number. */ off_t b_offset; /* Offset into file. */ TAILQ_ENTRY(buf) b_bobufs; /* (V) Buffer's associated vnode. */ uint32_t b_vflags; /* (V) BV_* flags */ unsigned short b_qindex; /* (Q) buffer queue index */ uint32_t b_flags; /* B_* flags. */ b_xflags_t b_xflags; /* extra flags */ struct lock b_lock; /* Buffer lock */ long b_bufsize; /* Allocated buffer size. */ int b_runningbufspace; /* when I/O is running, pipelining */ int b_kvasize; /* size of kva for buffer */ int b_dirtyoff; /* Offset in buffer of dirty region. */ int b_dirtyend; /* Offset of end of dirty region. */ caddr_t b_kvabase; /* base kva for buffer */ daddr_t b_lblkno; /* Logical block number. */ struct vnode *b_vp; /* Device vnode. */ struct ucred *b_rcred; /* Read credentials reference. */ struct ucred *b_wcred; /* Write credentials reference. */ union { TAILQ_ENTRY(buf) bu_freelist; /* (Q) */ struct { void (*pg_iodone)(void *, vm_page_t *, int, int); int pg_reqpage; } bu_pager; } b_union; #define b_freelist b_union.bu_freelist #define b_pager b_union.bu_pager union cluster_info { TAILQ_HEAD(cluster_list_head, buf) cluster_head; TAILQ_ENTRY(buf) cluster_entry; } b_cluster; struct vm_page *b_pages[btoc(MAXPHYS)]; int b_npages; struct workhead b_dep; /* (D) List of filesystem dependencies. */ void *b_fsprivate1; void *b_fsprivate2; void *b_fsprivate3; int b_pin_count; }; #define b_object b_bufobj->bo_object /* * These flags are kept in b_flags. * * Notes: * * B_ASYNC VOP calls on bp's are usually async whether or not * B_ASYNC is set, but some subsystems, such as NFS, like * to know what is best for the caller so they can * optimize the I/O. * * B_PAGING Indicates that bp is being used by the paging system or * some paging system and that the bp is not linked into * the b_vp's clean/dirty linked lists or ref counts. * Buffer vp reassignments are illegal in this case. * * B_CACHE This may only be set if the buffer is entirely valid. * The situation where B_DELWRI is set and B_CACHE is * clear MUST be committed to disk by getblk() so * B_DELWRI can also be cleared. See the comments for * getblk() in kern/vfs_bio.c. If B_CACHE is clear, * the caller is expected to clear BIO_ERROR and B_INVAL, * set BIO_READ, and initiate an I/O. * * The 'entire buffer' is defined to be the range from * 0 through b_bcount. * * B_MALLOC Request that the buffer be allocated from the malloc * pool, DEV_BSIZE aligned instead of PAGE_SIZE aligned. * * B_CLUSTEROK This flag is typically set for B_DELWRI buffers * by filesystems that allow clustering when the buffer * is fully dirty and indicates that it may be clustered * with other adjacent dirty buffers. Note the clustering * may not be used with the stage 1 data write under NFS * but may be used for the commit rpc portion. * * B_VMIO Indicates that the buffer is tied into an VM object. * The buffer's data is always PAGE_SIZE aligned even * if b_bufsize and b_bcount are not. ( b_bufsize is * always at least DEV_BSIZE aligned, though ). * * B_DIRECT Hint that we should attempt to completely free * the pages underlying the buffer. B_DIRECT is * sticky until the buffer is released and typically * only has an effect when B_RELBUF is also set. * */ #define B_AGE 0x00000001 /* Move to age queue when I/O done. */ #define B_NEEDCOMMIT 0x00000002 /* Append-write in progress. */ #define B_ASYNC 0x00000004 /* Start I/O, do not wait. */ #define B_DIRECT 0x00000008 /* direct I/O flag (pls free vmio) */ #define B_DEFERRED 0x00000010 /* Skipped over for cleaning */ #define B_CACHE 0x00000020 /* Bread found us in the cache. */ #define B_VALIDSUSPWRT 0x00000040 /* Valid write during suspension. */ #define B_DELWRI 0x00000080 /* Delay I/O until buffer reused. */ #define B_PERSISTENT 0x00000100 /* Perm. ref'ed while EXT2FS mounted. */ #define B_DONE 0x00000200 /* I/O completed. */ #define B_EINTR 0x00000400 /* I/O was interrupted */ -#define B_00000800 0x00000800 /* Available flag. */ +#define B_NOREUSE 0x00000800 /* Contents not reused once released. */ #define B_00001000 0x00001000 /* Available flag. */ #define B_INVAL 0x00002000 /* Does not contain valid info. */ #define B_BARRIER 0x00004000 /* Write this and all preceeding first. */ #define B_NOCACHE 0x00008000 /* Do not cache block after use. */ #define B_MALLOC 0x00010000 /* malloced b_data */ #define B_CLUSTEROK 0x00020000 /* Pagein op, so swap() can count it. */ #define B_000400000 0x00040000 /* Available flag. */ #define B_000800000 0x00080000 /* Available flag. */ #define B_001000000 0x00100000 /* Available flag. */ #define B_DIRTY 0x00200000 /* Needs writing later (in EXT2FS). */ #define B_RELBUF 0x00400000 /* Release VMIO buffer. */ #define B_FS_FLAG1 0x00800000 /* Available flag for FS use. */ #define B_NOCOPY 0x01000000 /* Don't copy-on-write this buf. */ #define B_INFREECNT 0x02000000 /* buf is counted in numfreebufs */ #define B_PAGING 0x04000000 /* volatile paging I/O -- bypass VMIO */ #define B_MANAGED 0x08000000 /* Managed by FS. */ #define B_RAM 0x10000000 /* Read ahead mark (flag) */ #define B_VMIO 0x20000000 /* VMIO flag */ #define B_CLUSTER 0x40000000 /* pagein op, so swap() can count it */ #define B_REMFREE 0x80000000 /* Delayed bremfree */ #define PRINT_BUF_FLAGS "\20\40remfree\37cluster\36vmio\35ram\34managed" \ "\33paging\32infreecnt\31nocopy\30b23\27relbuf\26dirty\25b20" \ "\24b19\23b18\22clusterok\21malloc\20nocache\17b14\16inval" \ - "\15b12\14b11\13eintr\12done\11persist\10delwri" \ + "\15b12\14noreuse\13eintr\12done\11persist\10delwri" \ "\7validsuspwrt\6cache\5deferred\4direct\3async\2needcommit\1age" /* * These flags are kept in b_xflags. */ #define BX_VNDIRTY 0x00000001 /* On vnode dirty list */ #define BX_VNCLEAN 0x00000002 /* On vnode clean list */ #define BX_BKGRDWRITE 0x00000010 /* Do writes in background */ #define BX_BKGRDMARKER 0x00000020 /* Mark buffer for splay tree */ #define BX_ALTDATA 0x00000040 /* Holds extended data */ #define PRINT_BUF_XFLAGS "\20\7altdata\6bkgrdmarker\5bkgrdwrite\2clean\1dirty" #define NOOFFSET (-1LL) /* No buffer offset calculated yet */ /* * These flags are kept in b_vflags. */ #define BV_SCANNED 0x00000001 /* VOP_FSYNC funcs mark written bufs */ #define BV_BKGRDINPROG 0x00000002 /* Background write in progress */ #define BV_BKGRDWAIT 0x00000004 /* Background write waiting */ #define BV_BKGRDERR 0x00000008 /* Error from background write */ #define PRINT_BUF_VFLAGS "\20\4bkgrderr\3bkgrdwait\2bkgrdinprog\1scanned" #ifdef _KERNEL /* * Buffer locking */ extern const char *buf_wmesg; /* Default buffer lock message */ #define BUF_WMESG "bufwait" #include /* XXX for curthread */ #include /* * Initialize a lock. */ #define BUF_LOCKINIT(bp) \ lockinit(&(bp)->b_lock, PRIBIO + 4, buf_wmesg, 0, 0) /* * * Get a lock sleeping non-interruptably until it becomes available. */ #define BUF_LOCK(bp, locktype, interlock) \ _lockmgr_args_rw(&(bp)->b_lock, (locktype), (interlock), \ LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, \ LOCK_FILE, LOCK_LINE) /* * Get a lock sleeping with specified interruptably and timeout. */ #define BUF_TIMELOCK(bp, locktype, interlock, wmesg, catch, timo) \ _lockmgr_args_rw(&(bp)->b_lock, (locktype) | LK_TIMELOCK, \ (interlock), (wmesg), (PRIBIO + 4) | (catch), (timo), \ LOCK_FILE, LOCK_LINE) /* * Release a lock. Only the acquiring process may free the lock unless * it has been handed off to biodone. */ #define BUF_UNLOCK(bp) do { \ KASSERT(((bp)->b_flags & B_REMFREE) == 0, \ ("BUF_UNLOCK %p while B_REMFREE is still set.", (bp))); \ \ (void)_lockmgr_args(&(bp)->b_lock, LK_RELEASE, NULL, \ LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, \ LOCK_FILE, LOCK_LINE); \ } while (0) /* * Check if a buffer lock is recursed. */ #define BUF_LOCKRECURSED(bp) \ lockmgr_recursed(&(bp)->b_lock) /* * Check if a buffer lock is currently held. */ #define BUF_ISLOCKED(bp) \ lockstatus(&(bp)->b_lock) /* * Free a buffer lock. */ #define BUF_LOCKFREE(bp) \ lockdestroy(&(bp)->b_lock) /* * Print informations on a buffer lock. */ #define BUF_LOCKPRINTINFO(bp) \ lockmgr_printinfo(&(bp)->b_lock) /* * Buffer lock assertions. */ #if defined(INVARIANTS) && defined(INVARIANT_SUPPORT) #define BUF_ASSERT_LOCKED(bp) \ _lockmgr_assert(&(bp)->b_lock, KA_LOCKED, LOCK_FILE, LOCK_LINE) #define BUF_ASSERT_SLOCKED(bp) \ _lockmgr_assert(&(bp)->b_lock, KA_SLOCKED, LOCK_FILE, LOCK_LINE) #define BUF_ASSERT_XLOCKED(bp) \ _lockmgr_assert(&(bp)->b_lock, KA_XLOCKED, LOCK_FILE, LOCK_LINE) #define BUF_ASSERT_UNLOCKED(bp) \ _lockmgr_assert(&(bp)->b_lock, KA_UNLOCKED, LOCK_FILE, LOCK_LINE) #define BUF_ASSERT_HELD(bp) #define BUF_ASSERT_UNHELD(bp) #else #define BUF_ASSERT_LOCKED(bp) #define BUF_ASSERT_SLOCKED(bp) #define BUF_ASSERT_XLOCKED(bp) #define BUF_ASSERT_UNLOCKED(bp) #define BUF_ASSERT_HELD(bp) #define BUF_ASSERT_UNHELD(bp) #endif #ifdef _SYS_PROC_H_ /* Avoid #include pollution */ /* * When initiating asynchronous I/O, change ownership of the lock to the * kernel. Once done, the lock may legally released by biodone. The * original owning process can no longer acquire it recursively, but must * wait until the I/O is completed and the lock has been freed by biodone. */ #define BUF_KERNPROC(bp) \ _lockmgr_disown(&(bp)->b_lock, LOCK_FILE, LOCK_LINE) #endif /* * Find out if the lock has waiters or not. */ #define BUF_LOCKWAITERS(bp) \ lockmgr_waiters(&(bp)->b_lock) #endif /* _KERNEL */ struct buf_queue_head { TAILQ_HEAD(buf_queue, buf) queue; daddr_t last_pblkno; struct buf *insert_point; struct buf *switch_point; }; /* * This structure describes a clustered I/O. */ struct cluster_save { long bs_bcount; /* Saved b_bcount. */ long bs_bufsize; /* Saved b_bufsize. */ int bs_nchildren; /* Number of associated buffers. */ struct buf **bs_children; /* List of associated buffers. */ }; #ifdef _KERNEL static __inline int bwrite(struct buf *bp) { KASSERT(bp->b_bufobj != NULL, ("bwrite: no bufobj bp=%p", bp)); KASSERT(bp->b_bufobj->bo_ops != NULL, ("bwrite: no bo_ops bp=%p", bp)); KASSERT(bp->b_bufobj->bo_ops->bop_write != NULL, ("bwrite: no bop_write bp=%p", bp)); return (BO_WRITE(bp->b_bufobj, bp)); } static __inline void bstrategy(struct buf *bp) { KASSERT(bp->b_bufobj != NULL, ("bstrategy: no bufobj bp=%p", bp)); KASSERT(bp->b_bufobj->bo_ops != NULL, ("bstrategy: no bo_ops bp=%p", bp)); KASSERT(bp->b_bufobj->bo_ops->bop_strategy != NULL, ("bstrategy: no bop_strategy bp=%p", bp)); BO_STRATEGY(bp->b_bufobj, bp); } static __inline void buf_start(struct buf *bp) { if (bioops.io_start) (*bioops.io_start)(bp); } static __inline void buf_complete(struct buf *bp) { if (bioops.io_complete) (*bioops.io_complete)(bp); } static __inline void buf_deallocate(struct buf *bp) { if (bioops.io_deallocate) (*bioops.io_deallocate)(bp); } static __inline int buf_countdeps(struct buf *bp, int i) { if (bioops.io_countdeps) return ((*bioops.io_countdeps)(bp, i)); else return (0); } #endif /* _KERNEL */ /* * Zero out the buffer's data area. */ #define clrbuf(bp) { \ bzero((bp)->b_data, (u_int)(bp)->b_bcount); \ (bp)->b_resid = 0; \ } /* * Flags for getblk's last parameter. */ #define GB_LOCK_NOWAIT 0x0001 /* Fail if we block on a buf lock. */ #define GB_NOCREAT 0x0002 /* Don't create a buf if not found. */ #define GB_NOWAIT_BD 0x0004 /* Do not wait for bufdaemon. */ #define GB_UNMAPPED 0x0008 /* Do not mmap buffer pages. */ #define GB_KVAALLOC 0x0010 /* But allocate KVA. */ #ifdef _KERNEL extern int nbuf; /* The number of buffer headers */ extern long maxswzone; /* Max KVA for swap structures */ extern long maxbcache; /* Max KVA for buffer cache */ extern long runningbufspace; extern long hibufspace; extern int dirtybufthresh; extern int bdwriteskip; extern int dirtybufferflushes; extern int altbufferflushes; extern int nswbuf; /* Number of swap I/O buffer headers. */ extern int cluster_pbuf_freecnt; /* Number of pbufs for clusters */ extern int vnode_pbuf_freecnt; /* Number of pbufs for vnode pager */ extern int vnode_async_pbuf_freecnt; /* Number of pbufs for vnode pager, asynchronous reads */ extern caddr_t unmapped_buf; /* Data address for unmapped buffers. */ static inline int buf_mapped(struct buf *bp) { return (bp->b_data != unmapped_buf); } void runningbufwakeup(struct buf *); void waitrunningbufspace(void); caddr_t kern_vfs_bio_buffer_alloc(caddr_t v, long physmem_est); void bufinit(void); void bufshutdown(int); void bdata2bio(struct buf *bp, struct bio *bip); void bwillwrite(void); int buf_dirty_count_severe(void); void bremfree(struct buf *); void bremfreef(struct buf *); /* XXX Force bremfree, only for nfs. */ #define bread(vp, blkno, size, cred, bpp) \ breadn_flags(vp, blkno, size, NULL, NULL, 0, cred, 0, bpp) #define bread_gb(vp, blkno, size, cred, gbflags, bpp) \ breadn_flags(vp, blkno, size, NULL, NULL, 0, cred, \ gbflags, bpp) #define breadn(vp, blkno, size, rablkno, rabsize, cnt, cred, bpp) \ breadn_flags(vp, blkno, size, rablkno, rabsize, cnt, cred, 0, bpp) int breadn_flags(struct vnode *, daddr_t, int, daddr_t *, int *, int, struct ucred *, int, struct buf **); void breada(struct vnode *, daddr_t *, int *, int, struct ucred *); void bdwrite(struct buf *); void bawrite(struct buf *); void babarrierwrite(struct buf *); int bbarrierwrite(struct buf *); void bdirty(struct buf *); void bundirty(struct buf *); void bufstrategy(struct bufobj *, struct buf *); void brelse(struct buf *); void bqrelse(struct buf *); int vfs_bio_awrite(struct buf *); void vfs_drain_busy_pages(struct buf *bp); struct buf * getpbuf(int *); struct buf *incore(struct bufobj *, daddr_t); struct buf *gbincore(struct bufobj *, daddr_t); struct buf *getblk(struct vnode *, daddr_t, int, int, int, int); struct buf *geteblk(int, int); int bufwait(struct buf *); int bufwrite(struct buf *); void bufdone(struct buf *); void bufdone_finish(struct buf *); void bd_speedup(void); int cluster_read(struct vnode *, u_quad_t, daddr_t, long, struct ucred *, long, int, int, struct buf **); int cluster_wbuild(struct vnode *, long, daddr_t, int, int); void cluster_write(struct vnode *, struct buf *, u_quad_t, int, int); void vfs_bio_bzero_buf(struct buf *bp, int base, int size); void vfs_bio_set_valid(struct buf *, int base, int size); void vfs_bio_clrbuf(struct buf *); void vfs_busy_pages(struct buf *, int clear_modify); void vfs_unbusy_pages(struct buf *); int vmapbuf(struct buf *, int); void vunmapbuf(struct buf *); void relpbuf(struct buf *, int *); void brelvp(struct buf *); void bgetvp(struct vnode *, struct buf *); void pbgetbo(struct bufobj *bo, struct buf *bp); void pbgetvp(struct vnode *, struct buf *); void pbrelbo(struct buf *); void pbrelvp(struct buf *); int allocbuf(struct buf *bp, int size); void reassignbuf(struct buf *); struct buf *trypbuf(int *); void bwait(struct buf *, u_char, const char *); void bdone(struct buf *); void bpin(struct buf *); void bunpin(struct buf *); void bunpin_wait(struct buf *); #endif /* _KERNEL */ #endif /* !_SYS_BUF_H_ */ Index: head/sys/sys/file.h =================================================================== --- head/sys/sys/file.h (revision 288430) +++ head/sys/sys/file.h (revision 288431) @@ -1,413 +1,411 @@ /*- * Copyright (c) 1982, 1986, 1989, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)file.h 8.3 (Berkeley) 1/9/95 * $FreeBSD$ */ #ifndef _SYS_FILE_H_ #define _SYS_FILE_H_ #ifndef _KERNEL #include /* XXX */ #include #include #else #include #include #include #include #include struct filedesc; struct stat; struct thread; struct uio; struct knote; struct vnode; struct socket; #endif /* _KERNEL */ #define DTYPE_VNODE 1 /* file */ #define DTYPE_SOCKET 2 /* communications endpoint */ #define DTYPE_PIPE 3 /* pipe */ #define DTYPE_FIFO 4 /* fifo (named pipe) */ #define DTYPE_KQUEUE 5 /* event queue */ #define DTYPE_CRYPTO 6 /* crypto */ #define DTYPE_MQUEUE 7 /* posix message queue */ #define DTYPE_SHM 8 /* swap-backed shared memory */ #define DTYPE_SEM 9 /* posix semaphore */ #define DTYPE_PTS 10 /* pseudo teletype master device */ #define DTYPE_DEV 11 /* Device specific fd type */ #define DTYPE_PROCDESC 12 /* process descriptor */ #define DTYPE_LINUXEFD 13 /* emulation eventfd type */ #ifdef _KERNEL struct file; struct filecaps; struct kinfo_file; struct ucred; #define FOF_OFFSET 0x01 /* Use the offset in uio argument */ #define FOF_NOLOCK 0x02 /* Do not take FOFFSET_LOCK */ #define FOF_NEXTOFF 0x04 /* Also update f_nextoff */ #define FOF_NOUPDATE 0x10 /* Do not update f_offset */ off_t foffset_lock(struct file *fp, int flags); void foffset_lock_uio(struct file *fp, struct uio *uio, int flags); void foffset_unlock(struct file *fp, off_t val, int flags); void foffset_unlock_uio(struct file *fp, struct uio *uio, int flags); static inline off_t foffset_get(struct file *fp) { return (foffset_lock(fp, FOF_NOLOCK)); } typedef int fo_rdwr_t(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td); typedef int fo_truncate_t(struct file *fp, off_t length, struct ucred *active_cred, struct thread *td); typedef int fo_ioctl_t(struct file *fp, u_long com, void *data, struct ucred *active_cred, struct thread *td); typedef int fo_poll_t(struct file *fp, int events, struct ucred *active_cred, struct thread *td); typedef int fo_kqfilter_t(struct file *fp, struct knote *kn); typedef int fo_stat_t(struct file *fp, struct stat *sb, struct ucred *active_cred, struct thread *td); typedef int fo_close_t(struct file *fp, struct thread *td); typedef int fo_chmod_t(struct file *fp, mode_t mode, struct ucred *active_cred, struct thread *td); typedef int fo_chown_t(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, struct thread *td); typedef int fo_sendfile_t(struct file *fp, int sockfd, struct uio *hdr_uio, struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, int kflags, struct thread *td); typedef int fo_seek_t(struct file *fp, off_t offset, int whence, struct thread *td); typedef int fo_fill_kinfo_t(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp); typedef int fo_mmap_t(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff, struct thread *td); typedef int fo_flags_t; struct fileops { fo_rdwr_t *fo_read; fo_rdwr_t *fo_write; fo_truncate_t *fo_truncate; fo_ioctl_t *fo_ioctl; fo_poll_t *fo_poll; fo_kqfilter_t *fo_kqfilter; fo_stat_t *fo_stat; fo_close_t *fo_close; fo_chmod_t *fo_chmod; fo_chown_t *fo_chown; fo_sendfile_t *fo_sendfile; fo_seek_t *fo_seek; fo_fill_kinfo_t *fo_fill_kinfo; fo_mmap_t *fo_mmap; fo_flags_t fo_flags; /* DFLAG_* below */ }; #define DFLAG_PASSABLE 0x01 /* may be passed via unix sockets. */ #define DFLAG_SEEKABLE 0x02 /* seekable / nonsequential */ #endif /* _KERNEL */ #if defined(_KERNEL) || defined(_WANT_FILE) /* * Kernel descriptor table. * One entry for each open kernel vnode and socket. * * Below is the list of locks that protects members in struct file. * * (a) f_vnode lock required (shared allows both reads and writes) * (f) protected with mtx_lock(mtx_pool_find(fp)) * (d) cdevpriv_mtx * none not locked */ struct fadvise_info { int fa_advice; /* (f) FADV_* type. */ off_t fa_start; /* (f) Region start. */ off_t fa_end; /* (f) Region end. */ - off_t fa_prevstart; /* (f) Previous NOREUSE start. */ - off_t fa_prevend; /* (f) Previous NOREUSE end. */ }; struct file { void *f_data; /* file descriptor specific data */ struct fileops *f_ops; /* File operations */ struct ucred *f_cred; /* associated credentials. */ struct vnode *f_vnode; /* NULL or applicable vnode */ short f_type; /* descriptor type */ short f_vnread_flags; /* (f) Sleep lock for f_offset */ volatile u_int f_flag; /* see fcntl.h */ volatile u_int f_count; /* reference count */ /* * DTYPE_VNODE specific fields. */ int f_seqcount; /* (a) Count of sequential accesses. */ off_t f_nextoff; /* next expected read/write offset. */ union { struct cdev_privdata *fvn_cdevpriv; /* (d) Private data for the cdev. */ struct fadvise_info *fvn_advice; } f_vnun; /* * DFLAG_SEEKABLE specific fields */ off_t f_offset; /* * Mandatory Access control information. */ void *f_label; /* Place-holder for MAC label. */ }; #define f_cdevpriv f_vnun.fvn_cdevpriv #define f_advice f_vnun.fvn_advice #define FOFFSET_LOCKED 0x1 #define FOFFSET_LOCK_WAITING 0x2 #define FDEVFS_VNODE 0x4 #endif /* _KERNEL || _WANT_FILE */ /* * Userland version of struct file, for sysctl */ struct xfile { size_t xf_size; /* size of struct xfile */ pid_t xf_pid; /* owning process */ uid_t xf_uid; /* effective uid of owning process */ int xf_fd; /* descriptor number */ void *xf_file; /* address of struct file */ short xf_type; /* descriptor type */ int xf_count; /* reference count */ int xf_msgcount; /* references from message queue */ off_t xf_offset; /* file offset */ void *xf_data; /* file descriptor specific data */ void *xf_vnode; /* vnode pointer */ u_int xf_flag; /* flags (see fcntl.h) */ }; #ifdef _KERNEL extern struct fileops vnops; extern struct fileops badfileops; extern struct fileops socketops; extern int maxfiles; /* kernel limit on number of open files */ extern int maxfilesperproc; /* per process limit on number of open files */ extern volatile int openfiles; /* actual number of open files */ int fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp); int fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp, struct file **fpp); int fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp); int fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp); int fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl, struct file **fpp); int _fdrop(struct file *fp, struct thread *td); fo_rdwr_t invfo_rdwr; fo_truncate_t invfo_truncate; fo_ioctl_t invfo_ioctl; fo_poll_t invfo_poll; fo_kqfilter_t invfo_kqfilter; fo_chmod_t invfo_chmod; fo_chown_t invfo_chown; fo_sendfile_t invfo_sendfile; fo_sendfile_t vn_sendfile; fo_seek_t vn_seek; fo_fill_kinfo_t vn_fill_kinfo; int vn_fill_kinfo_vnode(struct vnode *vp, struct kinfo_file *kif); void finit(struct file *, u_int, short, void *, struct fileops *); int fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp); int fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp); int fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp, struct filecaps *havecaps, struct vnode **vpp); int fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp); int fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp); int fgetsock(struct thread *td, int fd, cap_rights_t *rightsp, struct socket **spp, u_int *fflagp); void fputsock(struct socket *sp); static __inline int _fnoop(void) { return (0); } #define fhold(fp) \ (refcount_acquire(&(fp)->f_count)) #define fdrop(fp, td) \ (refcount_release(&(fp)->f_count) ? _fdrop((fp), (td)) : _fnoop()) static __inline fo_rdwr_t fo_read; static __inline fo_rdwr_t fo_write; static __inline fo_truncate_t fo_truncate; static __inline fo_ioctl_t fo_ioctl; static __inline fo_poll_t fo_poll; static __inline fo_kqfilter_t fo_kqfilter; static __inline fo_stat_t fo_stat; static __inline fo_close_t fo_close; static __inline fo_chmod_t fo_chmod; static __inline fo_chown_t fo_chown; static __inline fo_sendfile_t fo_sendfile; static __inline int fo_read(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td) { return ((*fp->f_ops->fo_read)(fp, uio, active_cred, flags, td)); } static __inline int fo_write(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td) { return ((*fp->f_ops->fo_write)(fp, uio, active_cred, flags, td)); } static __inline int fo_truncate(struct file *fp, off_t length, struct ucred *active_cred, struct thread *td) { return ((*fp->f_ops->fo_truncate)(fp, length, active_cred, td)); } static __inline int fo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, struct thread *td) { return ((*fp->f_ops->fo_ioctl)(fp, com, data, active_cred, td)); } static __inline int fo_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td) { return ((*fp->f_ops->fo_poll)(fp, events, active_cred, td)); } static __inline int fo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, struct thread *td) { return ((*fp->f_ops->fo_stat)(fp, sb, active_cred, td)); } static __inline int fo_close(struct file *fp, struct thread *td) { return ((*fp->f_ops->fo_close)(fp, td)); } static __inline int fo_kqfilter(struct file *fp, struct knote *kn) { return ((*fp->f_ops->fo_kqfilter)(fp, kn)); } static __inline int fo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, struct thread *td) { return ((*fp->f_ops->fo_chmod)(fp, mode, active_cred, td)); } static __inline int fo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, struct thread *td) { return ((*fp->f_ops->fo_chown)(fp, uid, gid, active_cred, td)); } static __inline int fo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, int kflags, struct thread *td) { return ((*fp->f_ops->fo_sendfile)(fp, sockfd, hdr_uio, trl_uio, offset, nbytes, sent, flags, kflags, td)); } static __inline int fo_seek(struct file *fp, off_t offset, int whence, struct thread *td) { return ((*fp->f_ops->fo_seek)(fp, offset, whence, td)); } static __inline int fo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) { return ((*fp->f_ops->fo_fill_kinfo)(fp, kif, fdp)); } static __inline int fo_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff, struct thread *td) { if (fp->f_ops->fo_mmap == NULL) return (ENODEV); return ((*fp->f_ops->fo_mmap)(fp, map, addr, size, prot, cap_maxprot, flags, foff, td)); } #endif /* _KERNEL */ #endif /* !SYS_FILE_H */ Index: head/sys/vm/vm_object.c =================================================================== --- head/sys/vm/vm_object.c (revision 288430) +++ head/sys/vm/vm_object.c (revision 288431) @@ -1,2649 +1,2649 @@ /*- * Copyright (c) 1991, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * The Mach Operating System project at Carnegie-Mellon University. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * from: @(#)vm_object.c 8.5 (Berkeley) 3/22/94 * * * Copyright (c) 1987, 1990 Carnegie-Mellon University. * All rights reserved. * * Authors: Avadis Tevanian, Jr., Michael Wayne Young * * Permission to use, copy, modify and distribute this software and * its documentation is hereby granted, provided that both the copyright * notice and this permission notice appear in all copies of the * software, derivative works or modified versions, and any portions * thereof, and that both notices appear in supporting documentation. * * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. * * Carnegie Mellon requests users of this software to return to * * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU * School of Computer Science * Carnegie Mellon University * Pittsburgh PA 15213-3890 * * any improvements or extensions that they make and grant Carnegie the * rights to redistribute these changes. */ /* * Virtual memory object module. */ #include __FBSDID("$FreeBSD$"); #include "opt_vm.h" #include #include #include #include #include #include #include #include #include /* for curproc, pageproc */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static int old_msync; SYSCTL_INT(_vm, OID_AUTO, old_msync, CTLFLAG_RW, &old_msync, 0, "Use old (insecure) msync behavior"); static int vm_object_page_collect_flush(vm_object_t object, vm_page_t p, int pagerflags, int flags, boolean_t *clearobjflags, boolean_t *eio); static boolean_t vm_object_page_remove_write(vm_page_t p, int flags, boolean_t *clearobjflags); static void vm_object_qcollapse(vm_object_t object); static void vm_object_vndeallocate(vm_object_t object); /* * Virtual memory objects maintain the actual data * associated with allocated virtual memory. A given * page of memory exists within exactly one object. * * An object is only deallocated when all "references" * are given up. Only one "reference" to a given * region of an object should be writeable. * * Associated with each object is a list of all resident * memory pages belonging to that object; this list is * maintained by the "vm_page" module, and locked by the object's * lock. * * Each object also records a "pager" routine which is * used to retrieve (and store) pages to the proper backing * storage. In addition, objects may be backed by other * objects from which they were virtual-copied. * * The only items within the object structure which are * modified after time of creation are: * reference count locked by object's lock * pager routine locked by object's lock * */ struct object_q vm_object_list; struct mtx vm_object_list_mtx; /* lock for object list and count */ struct vm_object kernel_object_store; struct vm_object kmem_object_store; static SYSCTL_NODE(_vm_stats, OID_AUTO, object, CTLFLAG_RD, 0, "VM object stats"); static long object_collapses; SYSCTL_LONG(_vm_stats_object, OID_AUTO, collapses, CTLFLAG_RD, &object_collapses, 0, "VM object collapses"); static long object_bypasses; SYSCTL_LONG(_vm_stats_object, OID_AUTO, bypasses, CTLFLAG_RD, &object_bypasses, 0, "VM object bypasses"); static uma_zone_t obj_zone; static int vm_object_zinit(void *mem, int size, int flags); #ifdef INVARIANTS static void vm_object_zdtor(void *mem, int size, void *arg); static void vm_object_zdtor(void *mem, int size, void *arg) { vm_object_t object; object = (vm_object_t)mem; KASSERT(object->ref_count == 0, ("object %p ref_count = %d", object, object->ref_count)); KASSERT(TAILQ_EMPTY(&object->memq), ("object %p has resident pages in its memq", object)); KASSERT(vm_radix_is_empty(&object->rtree), ("object %p has resident pages in its trie", object)); #if VM_NRESERVLEVEL > 0 KASSERT(LIST_EMPTY(&object->rvq), ("object %p has reservations", object)); #endif KASSERT(vm_object_cache_is_empty(object), ("object %p has cached pages", object)); KASSERT(object->paging_in_progress == 0, ("object %p paging_in_progress = %d", object, object->paging_in_progress)); KASSERT(object->resident_page_count == 0, ("object %p resident_page_count = %d", object, object->resident_page_count)); KASSERT(object->shadow_count == 0, ("object %p shadow_count = %d", object, object->shadow_count)); KASSERT(object->type == OBJT_DEAD, ("object %p has non-dead type %d", object, object->type)); } #endif static int vm_object_zinit(void *mem, int size, int flags) { vm_object_t object; object = (vm_object_t)mem; rw_init_flags(&object->lock, "vm object", RW_DUPOK | RW_NEW); /* These are true for any object that has been freed */ object->type = OBJT_DEAD; object->ref_count = 0; object->rtree.rt_root = 0; object->rtree.rt_flags = 0; object->paging_in_progress = 0; object->resident_page_count = 0; object->shadow_count = 0; object->cache.rt_root = 0; object->cache.rt_flags = 0; mtx_lock(&vm_object_list_mtx); TAILQ_INSERT_TAIL(&vm_object_list, object, object_list); mtx_unlock(&vm_object_list_mtx); return (0); } static void _vm_object_allocate(objtype_t type, vm_pindex_t size, vm_object_t object) { TAILQ_INIT(&object->memq); LIST_INIT(&object->shadow_head); object->type = type; switch (type) { case OBJT_DEAD: panic("_vm_object_allocate: can't create OBJT_DEAD"); case OBJT_DEFAULT: case OBJT_SWAP: object->flags = OBJ_ONEMAPPING; break; case OBJT_DEVICE: case OBJT_SG: object->flags = OBJ_FICTITIOUS | OBJ_UNMANAGED; break; case OBJT_MGTDEVICE: object->flags = OBJ_FICTITIOUS; break; case OBJT_PHYS: object->flags = OBJ_UNMANAGED; break; case OBJT_VNODE: object->flags = 0; break; default: panic("_vm_object_allocate: type %d is undefined", type); } object->size = size; object->generation = 1; object->ref_count = 1; object->memattr = VM_MEMATTR_DEFAULT; object->cred = NULL; object->charge = 0; object->handle = NULL; object->backing_object = NULL; object->backing_object_offset = (vm_ooffset_t) 0; #if VM_NRESERVLEVEL > 0 LIST_INIT(&object->rvq); #endif } /* * vm_object_init: * * Initialize the VM objects module. */ void vm_object_init(void) { TAILQ_INIT(&vm_object_list); mtx_init(&vm_object_list_mtx, "vm object_list", NULL, MTX_DEF); rw_init(&kernel_object->lock, "kernel vm object"); _vm_object_allocate(OBJT_PHYS, OFF_TO_IDX(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS), kernel_object); #if VM_NRESERVLEVEL > 0 kernel_object->flags |= OBJ_COLORED; kernel_object->pg_color = (u_short)atop(VM_MIN_KERNEL_ADDRESS); #endif rw_init(&kmem_object->lock, "kmem vm object"); _vm_object_allocate(OBJT_PHYS, OFF_TO_IDX(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS), kmem_object); #if VM_NRESERVLEVEL > 0 kmem_object->flags |= OBJ_COLORED; kmem_object->pg_color = (u_short)atop(VM_MIN_KERNEL_ADDRESS); #endif /* * The lock portion of struct vm_object must be type stable due * to vm_pageout_fallback_object_lock locking a vm object * without holding any references to it. */ obj_zone = uma_zcreate("VM OBJECT", sizeof (struct vm_object), NULL, #ifdef INVARIANTS vm_object_zdtor, #else NULL, #endif vm_object_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); vm_radix_init(); } void vm_object_clear_flag(vm_object_t object, u_short bits) { VM_OBJECT_ASSERT_WLOCKED(object); object->flags &= ~bits; } /* * Sets the default memory attribute for the specified object. Pages * that are allocated to this object are by default assigned this memory * attribute. * * Presently, this function must be called before any pages are allocated * to the object. In the future, this requirement may be relaxed for * "default" and "swap" objects. */ int vm_object_set_memattr(vm_object_t object, vm_memattr_t memattr) { VM_OBJECT_ASSERT_WLOCKED(object); switch (object->type) { case OBJT_DEFAULT: case OBJT_DEVICE: case OBJT_MGTDEVICE: case OBJT_PHYS: case OBJT_SG: case OBJT_SWAP: case OBJT_VNODE: if (!TAILQ_EMPTY(&object->memq)) return (KERN_FAILURE); break; case OBJT_DEAD: return (KERN_INVALID_ARGUMENT); default: panic("vm_object_set_memattr: object %p is of undefined type", object); } object->memattr = memattr; return (KERN_SUCCESS); } void vm_object_pip_add(vm_object_t object, short i) { VM_OBJECT_ASSERT_WLOCKED(object); object->paging_in_progress += i; } void vm_object_pip_subtract(vm_object_t object, short i) { VM_OBJECT_ASSERT_WLOCKED(object); object->paging_in_progress -= i; } void vm_object_pip_wakeup(vm_object_t object) { VM_OBJECT_ASSERT_WLOCKED(object); object->paging_in_progress--; if ((object->flags & OBJ_PIPWNT) && object->paging_in_progress == 0) { vm_object_clear_flag(object, OBJ_PIPWNT); wakeup(object); } } void vm_object_pip_wakeupn(vm_object_t object, short i) { VM_OBJECT_ASSERT_WLOCKED(object); if (i) object->paging_in_progress -= i; if ((object->flags & OBJ_PIPWNT) && object->paging_in_progress == 0) { vm_object_clear_flag(object, OBJ_PIPWNT); wakeup(object); } } void vm_object_pip_wait(vm_object_t object, char *waitid) { VM_OBJECT_ASSERT_WLOCKED(object); while (object->paging_in_progress) { object->flags |= OBJ_PIPWNT; VM_OBJECT_SLEEP(object, object, PVM, waitid, 0); } } /* * vm_object_allocate: * * Returns a new object with the given size. */ vm_object_t vm_object_allocate(objtype_t type, vm_pindex_t size) { vm_object_t object; object = (vm_object_t)uma_zalloc(obj_zone, M_WAITOK); _vm_object_allocate(type, size, object); return (object); } /* * vm_object_reference: * * Gets another reference to the given object. Note: OBJ_DEAD * objects can be referenced during final cleaning. */ void vm_object_reference(vm_object_t object) { if (object == NULL) return; VM_OBJECT_WLOCK(object); vm_object_reference_locked(object); VM_OBJECT_WUNLOCK(object); } /* * vm_object_reference_locked: * * Gets another reference to the given object. * * The object must be locked. */ void vm_object_reference_locked(vm_object_t object) { struct vnode *vp; VM_OBJECT_ASSERT_WLOCKED(object); object->ref_count++; if (object->type == OBJT_VNODE) { vp = object->handle; vref(vp); } } /* * Handle deallocating an object of type OBJT_VNODE. */ static void vm_object_vndeallocate(vm_object_t object) { struct vnode *vp = (struct vnode *) object->handle; VM_OBJECT_ASSERT_WLOCKED(object); KASSERT(object->type == OBJT_VNODE, ("vm_object_vndeallocate: not a vnode object")); KASSERT(vp != NULL, ("vm_object_vndeallocate: missing vp")); #ifdef INVARIANTS if (object->ref_count == 0) { vprint("vm_object_vndeallocate", vp); panic("vm_object_vndeallocate: bad object reference count"); } #endif /* * The test for text of vp vnode does not need a bypass to * reach right VV_TEXT there, since it is obtained from * object->handle. */ if (object->ref_count > 1 || (vp->v_vflag & VV_TEXT) == 0) { object->ref_count--; VM_OBJECT_WUNLOCK(object); /* vrele may need the vnode lock. */ vrele(vp); } else { vhold(vp); VM_OBJECT_WUNLOCK(object); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); vdrop(vp); VM_OBJECT_WLOCK(object); object->ref_count--; if (object->type == OBJT_DEAD) { VM_OBJECT_WUNLOCK(object); VOP_UNLOCK(vp, 0); } else { if (object->ref_count == 0) VOP_UNSET_TEXT(vp); VM_OBJECT_WUNLOCK(object); vput(vp); } } } /* * vm_object_deallocate: * * Release a reference to the specified object, * gained either through a vm_object_allocate * or a vm_object_reference call. When all references * are gone, storage associated with this object * may be relinquished. * * No object may be locked. */ void vm_object_deallocate(vm_object_t object) { vm_object_t temp; struct vnode *vp; while (object != NULL) { VM_OBJECT_WLOCK(object); if (object->type == OBJT_VNODE) { vm_object_vndeallocate(object); return; } KASSERT(object->ref_count != 0, ("vm_object_deallocate: object deallocated too many times: %d", object->type)); /* * If the reference count goes to 0 we start calling * vm_object_terminate() on the object chain. * A ref count of 1 may be a special case depending on the * shadow count being 0 or 1. */ object->ref_count--; if (object->ref_count > 1) { VM_OBJECT_WUNLOCK(object); return; } else if (object->ref_count == 1) { if (object->type == OBJT_SWAP && (object->flags & OBJ_TMPFS) != 0) { vp = object->un_pager.swp.swp_tmpfs; vhold(vp); VM_OBJECT_WUNLOCK(object); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); VM_OBJECT_WLOCK(object); if (object->type == OBJT_DEAD || object->ref_count != 1) { VM_OBJECT_WUNLOCK(object); VOP_UNLOCK(vp, 0); vdrop(vp); return; } if ((object->flags & OBJ_TMPFS) != 0) VOP_UNSET_TEXT(vp); VOP_UNLOCK(vp, 0); vdrop(vp); } if (object->shadow_count == 0 && object->handle == NULL && (object->type == OBJT_DEFAULT || (object->type == OBJT_SWAP && (object->flags & OBJ_TMPFS_NODE) == 0))) { vm_object_set_flag(object, OBJ_ONEMAPPING); } else if ((object->shadow_count == 1) && (object->handle == NULL) && (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) { vm_object_t robject; robject = LIST_FIRST(&object->shadow_head); KASSERT(robject != NULL, ("vm_object_deallocate: ref_count: %d, shadow_count: %d", object->ref_count, object->shadow_count)); KASSERT((robject->flags & OBJ_TMPFS_NODE) == 0, ("shadowed tmpfs v_object %p", object)); if (!VM_OBJECT_TRYWLOCK(robject)) { /* * Avoid a potential deadlock. */ object->ref_count++; VM_OBJECT_WUNLOCK(object); /* * More likely than not the thread * holding robject's lock has lower * priority than the current thread. * Let the lower priority thread run. */ pause("vmo_de", 1); continue; } /* * Collapse object into its shadow unless its * shadow is dead. In that case, object will * be deallocated by the thread that is * deallocating its shadow. */ if ((robject->flags & OBJ_DEAD) == 0 && (robject->handle == NULL) && (robject->type == OBJT_DEFAULT || robject->type == OBJT_SWAP)) { robject->ref_count++; retry: if (robject->paging_in_progress) { VM_OBJECT_WUNLOCK(object); vm_object_pip_wait(robject, "objde1"); temp = robject->backing_object; if (object == temp) { VM_OBJECT_WLOCK(object); goto retry; } } else if (object->paging_in_progress) { VM_OBJECT_WUNLOCK(robject); object->flags |= OBJ_PIPWNT; VM_OBJECT_SLEEP(object, object, PDROP | PVM, "objde2", 0); VM_OBJECT_WLOCK(robject); temp = robject->backing_object; if (object == temp) { VM_OBJECT_WLOCK(object); goto retry; } } else VM_OBJECT_WUNLOCK(object); if (robject->ref_count == 1) { robject->ref_count--; object = robject; goto doterm; } object = robject; vm_object_collapse(object); VM_OBJECT_WUNLOCK(object); continue; } VM_OBJECT_WUNLOCK(robject); } VM_OBJECT_WUNLOCK(object); return; } doterm: temp = object->backing_object; if (temp != NULL) { KASSERT((object->flags & OBJ_TMPFS_NODE) == 0, ("shadowed tmpfs v_object 2 %p", object)); VM_OBJECT_WLOCK(temp); LIST_REMOVE(object, shadow_list); temp->shadow_count--; VM_OBJECT_WUNLOCK(temp); object->backing_object = NULL; } /* * Don't double-terminate, we could be in a termination * recursion due to the terminate having to sync data * to disk. */ if ((object->flags & OBJ_DEAD) == 0) vm_object_terminate(object); else VM_OBJECT_WUNLOCK(object); object = temp; } } /* * vm_object_destroy removes the object from the global object list * and frees the space for the object. */ void vm_object_destroy(vm_object_t object) { /* * Release the allocation charge. */ if (object->cred != NULL) { swap_release_by_cred(object->charge, object->cred); object->charge = 0; crfree(object->cred); object->cred = NULL; } /* * Free the space for the object. */ uma_zfree(obj_zone, object); } /* * vm_object_terminate actually destroys the specified object, freeing * up all previously used resources. * * The object must be locked. * This routine may block. */ void vm_object_terminate(vm_object_t object) { vm_page_t p, p_next; VM_OBJECT_ASSERT_WLOCKED(object); /* * Make sure no one uses us. */ vm_object_set_flag(object, OBJ_DEAD); /* * wait for the pageout daemon to be done with the object */ vm_object_pip_wait(object, "objtrm"); KASSERT(!object->paging_in_progress, ("vm_object_terminate: pageout in progress")); /* * Clean and free the pages, as appropriate. All references to the * object are gone, so we don't need to lock it. */ if (object->type == OBJT_VNODE) { struct vnode *vp = (struct vnode *)object->handle; /* * Clean pages and flush buffers. */ vm_object_page_clean(object, 0, 0, OBJPC_SYNC); VM_OBJECT_WUNLOCK(object); vinvalbuf(vp, V_SAVE, 0, 0); VM_OBJECT_WLOCK(object); } KASSERT(object->ref_count == 0, ("vm_object_terminate: object with references, ref_count=%d", object->ref_count)); /* * Free any remaining pageable pages. This also removes them from the * paging queues. However, don't free wired pages, just remove them * from the object. Rather than incrementally removing each page from * the object, the page and object are reset to any empty state. */ TAILQ_FOREACH_SAFE(p, &object->memq, listq, p_next) { vm_page_assert_unbusied(p); vm_page_lock(p); /* * Optimize the page's removal from the object by resetting * its "object" field. Specifically, if the page is not * wired, then the effect of this assignment is that * vm_page_free()'s call to vm_page_remove() will return * immediately without modifying the page or the object. */ p->object = NULL; if (p->wire_count == 0) { vm_page_free(p); PCPU_INC(cnt.v_pfree); } vm_page_unlock(p); } /* * If the object contained any pages, then reset it to an empty state. * None of the object's fields, including "resident_page_count", were * modified by the preceding loop. */ if (object->resident_page_count != 0) { vm_radix_reclaim_allnodes(&object->rtree); TAILQ_INIT(&object->memq); object->resident_page_count = 0; if (object->type == OBJT_VNODE) vdrop(object->handle); } #if VM_NRESERVLEVEL > 0 if (__predict_false(!LIST_EMPTY(&object->rvq))) vm_reserv_break_all(object); #endif if (__predict_false(!vm_object_cache_is_empty(object))) vm_page_cache_free(object, 0, 0); KASSERT(object->cred == NULL || object->type == OBJT_DEFAULT || object->type == OBJT_SWAP, ("%s: non-swap obj %p has cred", __func__, object)); /* * Let the pager know object is dead. */ vm_pager_deallocate(object); VM_OBJECT_WUNLOCK(object); vm_object_destroy(object); } /* * Make the page read-only so that we can clear the object flags. However, if * this is a nosync mmap then the object is likely to stay dirty so do not * mess with the page and do not clear the object flags. Returns TRUE if the * page should be flushed, and FALSE otherwise. */ static boolean_t vm_object_page_remove_write(vm_page_t p, int flags, boolean_t *clearobjflags) { /* * If we have been asked to skip nosync pages and this is a * nosync page, skip it. Note that the object flags were not * cleared in this case so we do not have to set them. */ if ((flags & OBJPC_NOSYNC) != 0 && (p->oflags & VPO_NOSYNC) != 0) { *clearobjflags = FALSE; return (FALSE); } else { pmap_remove_write(p); return (p->dirty != 0); } } /* * vm_object_page_clean * * Clean all dirty pages in the specified range of object. Leaves page * on whatever queue it is currently on. If NOSYNC is set then do not * write out pages with VPO_NOSYNC set (originally comes from MAP_NOSYNC), * leaving the object dirty. * * When stuffing pages asynchronously, allow clustering. XXX we need a * synchronous clustering mode implementation. * * Odd semantics: if start == end, we clean everything. * * The object must be locked. * * Returns FALSE if some page from the range was not written, as * reported by the pager, and TRUE otherwise. */ boolean_t vm_object_page_clean(vm_object_t object, vm_ooffset_t start, vm_ooffset_t end, int flags) { vm_page_t np, p; vm_pindex_t pi, tend, tstart; int curgeneration, n, pagerflags; boolean_t clearobjflags, eio, res; VM_OBJECT_ASSERT_WLOCKED(object); /* * The OBJ_MIGHTBEDIRTY flag is only set for OBJT_VNODE * objects. The check below prevents the function from * operating on non-vnode objects. */ if ((object->flags & OBJ_MIGHTBEDIRTY) == 0 || object->resident_page_count == 0) return (TRUE); pagerflags = (flags & (OBJPC_SYNC | OBJPC_INVAL)) != 0 ? VM_PAGER_PUT_SYNC : VM_PAGER_CLUSTER_OK; pagerflags |= (flags & OBJPC_INVAL) != 0 ? VM_PAGER_PUT_INVAL : 0; tstart = OFF_TO_IDX(start); tend = (end == 0) ? object->size : OFF_TO_IDX(end + PAGE_MASK); clearobjflags = tstart == 0 && tend >= object->size; res = TRUE; rescan: curgeneration = object->generation; for (p = vm_page_find_least(object, tstart); p != NULL; p = np) { pi = p->pindex; if (pi >= tend) break; np = TAILQ_NEXT(p, listq); if (p->valid == 0) continue; if (vm_page_sleep_if_busy(p, "vpcwai")) { if (object->generation != curgeneration) { if ((flags & OBJPC_SYNC) != 0) goto rescan; else clearobjflags = FALSE; } np = vm_page_find_least(object, pi); continue; } if (!vm_object_page_remove_write(p, flags, &clearobjflags)) continue; n = vm_object_page_collect_flush(object, p, pagerflags, flags, &clearobjflags, &eio); if (eio) { res = FALSE; clearobjflags = FALSE; } if (object->generation != curgeneration) { if ((flags & OBJPC_SYNC) != 0) goto rescan; else clearobjflags = FALSE; } /* * If the VOP_PUTPAGES() did a truncated write, so * that even the first page of the run is not fully * written, vm_pageout_flush() returns 0 as the run * length. Since the condition that caused truncated * write may be permanent, e.g. exhausted free space, * accepting n == 0 would cause an infinite loop. * * Forwarding the iterator leaves the unwritten page * behind, but there is not much we can do there if * filesystem refuses to write it. */ if (n == 0) { n = 1; clearobjflags = FALSE; } np = vm_page_find_least(object, pi + n); } #if 0 VOP_FSYNC(vp, (pagerflags & VM_PAGER_PUT_SYNC) ? MNT_WAIT : 0); #endif if (clearobjflags) vm_object_clear_flag(object, OBJ_MIGHTBEDIRTY); return (res); } static int vm_object_page_collect_flush(vm_object_t object, vm_page_t p, int pagerflags, int flags, boolean_t *clearobjflags, boolean_t *eio) { vm_page_t ma[vm_pageout_page_count], p_first, tp; int count, i, mreq, runlen; vm_page_lock_assert(p, MA_NOTOWNED); VM_OBJECT_ASSERT_WLOCKED(object); count = 1; mreq = 0; for (tp = p; count < vm_pageout_page_count; count++) { tp = vm_page_next(tp); if (tp == NULL || vm_page_busied(tp)) break; if (!vm_object_page_remove_write(tp, flags, clearobjflags)) break; } for (p_first = p; count < vm_pageout_page_count; count++) { tp = vm_page_prev(p_first); if (tp == NULL || vm_page_busied(tp)) break; if (!vm_object_page_remove_write(tp, flags, clearobjflags)) break; p_first = tp; mreq++; } for (tp = p_first, i = 0; i < count; tp = TAILQ_NEXT(tp, listq), i++) ma[i] = tp; vm_pageout_flush(ma, count, pagerflags, mreq, &runlen, eio); return (runlen); } /* * Note that there is absolutely no sense in writing out * anonymous objects, so we track down the vnode object * to write out. * We invalidate (remove) all pages from the address space * for semantic correctness. * * If the backing object is a device object with unmanaged pages, then any * mappings to the specified range of pages must be removed before this * function is called. * * Note: certain anonymous maps, such as MAP_NOSYNC maps, * may start out with a NULL object. */ boolean_t vm_object_sync(vm_object_t object, vm_ooffset_t offset, vm_size_t size, boolean_t syncio, boolean_t invalidate) { vm_object_t backing_object; struct vnode *vp; struct mount *mp; int error, flags, fsync_after; boolean_t res; if (object == NULL) return (TRUE); res = TRUE; error = 0; VM_OBJECT_WLOCK(object); while ((backing_object = object->backing_object) != NULL) { VM_OBJECT_WLOCK(backing_object); offset += object->backing_object_offset; VM_OBJECT_WUNLOCK(object); object = backing_object; if (object->size < OFF_TO_IDX(offset + size)) size = IDX_TO_OFF(object->size) - offset; } /* * Flush pages if writing is allowed, invalidate them * if invalidation requested. Pages undergoing I/O * will be ignored by vm_object_page_remove(). * * We cannot lock the vnode and then wait for paging * to complete without deadlocking against vm_fault. * Instead we simply call vm_object_page_remove() and * allow it to block internally on a page-by-page * basis when it encounters pages undergoing async * I/O. */ if (object->type == OBJT_VNODE && (object->flags & OBJ_MIGHTBEDIRTY) != 0) { vp = object->handle; VM_OBJECT_WUNLOCK(object); (void) vn_start_write(vp, &mp, V_WAIT); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); if (syncio && !invalidate && offset == 0 && OFF_TO_IDX(size) == object->size) { /* * If syncing the whole mapping of the file, * it is faster to schedule all the writes in * async mode, also allowing the clustering, * and then wait for i/o to complete. */ flags = 0; fsync_after = TRUE; } else { flags = (syncio || invalidate) ? OBJPC_SYNC : 0; flags |= invalidate ? (OBJPC_SYNC | OBJPC_INVAL) : 0; fsync_after = FALSE; } VM_OBJECT_WLOCK(object); res = vm_object_page_clean(object, offset, offset + size, flags); VM_OBJECT_WUNLOCK(object); if (fsync_after) error = VOP_FSYNC(vp, MNT_WAIT, curthread); VOP_UNLOCK(vp, 0); vn_finished_write(mp); if (error != 0) res = FALSE; VM_OBJECT_WLOCK(object); } if ((object->type == OBJT_VNODE || object->type == OBJT_DEVICE) && invalidate) { if (object->type == OBJT_DEVICE) /* * The option OBJPR_NOTMAPPED must be passed here * because vm_object_page_remove() cannot remove * unmanaged mappings. */ flags = OBJPR_NOTMAPPED; else if (old_msync) flags = 0; else flags = OBJPR_CLEANONLY; vm_object_page_remove(object, OFF_TO_IDX(offset), OFF_TO_IDX(offset + size + PAGE_MASK), flags); } VM_OBJECT_WUNLOCK(object); return (res); } /* * vm_object_madvise: * * Implements the madvise function at the object/page level. * * MADV_WILLNEED (any object) * * Activate the specified pages if they are resident. * * MADV_DONTNEED (any object) * * Deactivate the specified pages if they are resident. * * MADV_FREE (OBJT_DEFAULT/OBJT_SWAP objects, * OBJ_ONEMAPPING only) * * Deactivate and clean the specified pages if they are * resident. This permits the process to reuse the pages * without faulting or the kernel to reclaim the pages * without I/O. */ void vm_object_madvise(vm_object_t object, vm_pindex_t pindex, vm_pindex_t end, int advise) { vm_pindex_t tpindex; vm_object_t backing_object, tobject; vm_page_t m; if (object == NULL) return; VM_OBJECT_WLOCK(object); /* * Locate and adjust resident pages */ for (; pindex < end; pindex += 1) { relookup: tobject = object; tpindex = pindex; shadowlookup: /* * MADV_FREE only operates on OBJT_DEFAULT or OBJT_SWAP pages * and those pages must be OBJ_ONEMAPPING. */ if (advise == MADV_FREE) { if ((tobject->type != OBJT_DEFAULT && tobject->type != OBJT_SWAP) || (tobject->flags & OBJ_ONEMAPPING) == 0) { goto unlock_tobject; } } else if ((tobject->flags & OBJ_UNMANAGED) != 0) goto unlock_tobject; m = vm_page_lookup(tobject, tpindex); if (m == NULL && advise == MADV_WILLNEED) { /* * If the page is cached, reactivate it. */ m = vm_page_alloc(tobject, tpindex, VM_ALLOC_IFCACHED | VM_ALLOC_NOBUSY); } if (m == NULL) { /* * There may be swap even if there is no backing page */ if (advise == MADV_FREE && tobject->type == OBJT_SWAP) swap_pager_freespace(tobject, tpindex, 1); /* * next object */ backing_object = tobject->backing_object; if (backing_object == NULL) goto unlock_tobject; VM_OBJECT_WLOCK(backing_object); tpindex += OFF_TO_IDX(tobject->backing_object_offset); if (tobject != object) VM_OBJECT_WUNLOCK(tobject); tobject = backing_object; goto shadowlookup; } else if (m->valid != VM_PAGE_BITS_ALL) goto unlock_tobject; /* * If the page is not in a normal state, skip it. */ vm_page_lock(m); if (m->hold_count != 0 || m->wire_count != 0) { vm_page_unlock(m); goto unlock_tobject; } KASSERT((m->flags & PG_FICTITIOUS) == 0, ("vm_object_madvise: page %p is fictitious", m)); KASSERT((m->oflags & VPO_UNMANAGED) == 0, ("vm_object_madvise: page %p is not managed", m)); if (vm_page_busied(m)) { if (advise == MADV_WILLNEED) { /* * Reference the page before unlocking and * sleeping so that the page daemon is less * likely to reclaim it. */ vm_page_aflag_set(m, PGA_REFERENCED); } if (object != tobject) VM_OBJECT_WUNLOCK(object); VM_OBJECT_WUNLOCK(tobject); vm_page_busy_sleep(m, "madvpo"); VM_OBJECT_WLOCK(object); goto relookup; } if (advise == MADV_WILLNEED) { vm_page_activate(m); } else { vm_page_advise(m, advise); } vm_page_unlock(m); if (advise == MADV_FREE && tobject->type == OBJT_SWAP) swap_pager_freespace(tobject, tpindex, 1); unlock_tobject: if (tobject != object) VM_OBJECT_WUNLOCK(tobject); } VM_OBJECT_WUNLOCK(object); } /* * vm_object_shadow: * * Create a new object which is backed by the * specified existing object range. The source * object reference is deallocated. * * The new object and offset into that object * are returned in the source parameters. */ void vm_object_shadow( vm_object_t *object, /* IN/OUT */ vm_ooffset_t *offset, /* IN/OUT */ vm_size_t length) { vm_object_t source; vm_object_t result; source = *object; /* * Don't create the new object if the old object isn't shared. */ if (source != NULL) { VM_OBJECT_WLOCK(source); if (source->ref_count == 1 && source->handle == NULL && (source->type == OBJT_DEFAULT || source->type == OBJT_SWAP)) { VM_OBJECT_WUNLOCK(source); return; } VM_OBJECT_WUNLOCK(source); } /* * Allocate a new object with the given length. */ result = vm_object_allocate(OBJT_DEFAULT, atop(length)); /* * The new object shadows the source object, adding a reference to it. * Our caller changes his reference to point to the new object, * removing a reference to the source object. Net result: no change * of reference count. * * Try to optimize the result object's page color when shadowing * in order to maintain page coloring consistency in the combined * shadowed object. */ result->backing_object = source; /* * Store the offset into the source object, and fix up the offset into * the new object. */ result->backing_object_offset = *offset; if (source != NULL) { VM_OBJECT_WLOCK(source); LIST_INSERT_HEAD(&source->shadow_head, result, shadow_list); source->shadow_count++; #if VM_NRESERVLEVEL > 0 result->flags |= source->flags & OBJ_COLORED; result->pg_color = (source->pg_color + OFF_TO_IDX(*offset)) & ((1 << (VM_NFREEORDER - 1)) - 1); #endif VM_OBJECT_WUNLOCK(source); } /* * Return the new things */ *offset = 0; *object = result; } /* * vm_object_split: * * Split the pages in a map entry into a new object. This affords * easier removal of unused pages, and keeps object inheritance from * being a negative impact on memory usage. */ void vm_object_split(vm_map_entry_t entry) { vm_page_t m, m_next; vm_object_t orig_object, new_object, source; vm_pindex_t idx, offidxstart; vm_size_t size; orig_object = entry->object.vm_object; if (orig_object->type != OBJT_DEFAULT && orig_object->type != OBJT_SWAP) return; if (orig_object->ref_count <= 1) return; VM_OBJECT_WUNLOCK(orig_object); offidxstart = OFF_TO_IDX(entry->offset); size = atop(entry->end - entry->start); /* * If swap_pager_copy() is later called, it will convert new_object * into a swap object. */ new_object = vm_object_allocate(OBJT_DEFAULT, size); /* * At this point, the new object is still private, so the order in * which the original and new objects are locked does not matter. */ VM_OBJECT_WLOCK(new_object); VM_OBJECT_WLOCK(orig_object); source = orig_object->backing_object; if (source != NULL) { VM_OBJECT_WLOCK(source); if ((source->flags & OBJ_DEAD) != 0) { VM_OBJECT_WUNLOCK(source); VM_OBJECT_WUNLOCK(orig_object); VM_OBJECT_WUNLOCK(new_object); vm_object_deallocate(new_object); VM_OBJECT_WLOCK(orig_object); return; } LIST_INSERT_HEAD(&source->shadow_head, new_object, shadow_list); source->shadow_count++; vm_object_reference_locked(source); /* for new_object */ vm_object_clear_flag(source, OBJ_ONEMAPPING); VM_OBJECT_WUNLOCK(source); new_object->backing_object_offset = orig_object->backing_object_offset + entry->offset; new_object->backing_object = source; } if (orig_object->cred != NULL) { new_object->cred = orig_object->cred; crhold(orig_object->cred); new_object->charge = ptoa(size); KASSERT(orig_object->charge >= ptoa(size), ("orig_object->charge < 0")); orig_object->charge -= ptoa(size); } retry: m = vm_page_find_least(orig_object, offidxstart); for (; m != NULL && (idx = m->pindex - offidxstart) < size; m = m_next) { m_next = TAILQ_NEXT(m, listq); /* * We must wait for pending I/O to complete before we can * rename the page. * * We do not have to VM_PROT_NONE the page as mappings should * not be changed by this operation. */ if (vm_page_busied(m)) { VM_OBJECT_WUNLOCK(new_object); vm_page_lock(m); VM_OBJECT_WUNLOCK(orig_object); vm_page_busy_sleep(m, "spltwt"); VM_OBJECT_WLOCK(orig_object); VM_OBJECT_WLOCK(new_object); goto retry; } /* vm_page_rename() will handle dirty and cache. */ if (vm_page_rename(m, new_object, idx)) { VM_OBJECT_WUNLOCK(new_object); VM_OBJECT_WUNLOCK(orig_object); VM_WAIT; VM_OBJECT_WLOCK(orig_object); VM_OBJECT_WLOCK(new_object); goto retry; } #if VM_NRESERVLEVEL > 0 /* * If some of the reservation's allocated pages remain with * the original object, then transferring the reservation to * the new object is neither particularly beneficial nor * particularly harmful as compared to leaving the reservation * with the original object. If, however, all of the * reservation's allocated pages are transferred to the new * object, then transferring the reservation is typically * beneficial. Determining which of these two cases applies * would be more costly than unconditionally renaming the * reservation. */ vm_reserv_rename(m, new_object, orig_object, offidxstart); #endif if (orig_object->type == OBJT_SWAP) vm_page_xbusy(m); } if (orig_object->type == OBJT_SWAP) { /* * swap_pager_copy() can sleep, in which case the orig_object's * and new_object's locks are released and reacquired. */ swap_pager_copy(orig_object, new_object, offidxstart, 0); TAILQ_FOREACH(m, &new_object->memq, listq) vm_page_xunbusy(m); /* * Transfer any cached pages from orig_object to new_object. * If swap_pager_copy() found swapped out pages within the * specified range of orig_object, then it changed * new_object's type to OBJT_SWAP when it transferred those * pages to new_object. Otherwise, new_object's type * should still be OBJT_DEFAULT and orig_object should not * contain any cached pages within the specified range. */ if (__predict_false(!vm_object_cache_is_empty(orig_object))) vm_page_cache_transfer(orig_object, offidxstart, new_object); } VM_OBJECT_WUNLOCK(orig_object); VM_OBJECT_WUNLOCK(new_object); entry->object.vm_object = new_object; entry->offset = 0LL; vm_object_deallocate(orig_object); VM_OBJECT_WLOCK(new_object); } #define OBSC_TEST_ALL_SHADOWED 0x0001 #define OBSC_COLLAPSE_NOWAIT 0x0002 #define OBSC_COLLAPSE_WAIT 0x0004 static int vm_object_backing_scan(vm_object_t object, int op) { int r = 1; vm_page_t p; vm_object_t backing_object; vm_pindex_t backing_offset_index; VM_OBJECT_ASSERT_WLOCKED(object); VM_OBJECT_ASSERT_WLOCKED(object->backing_object); backing_object = object->backing_object; backing_offset_index = OFF_TO_IDX(object->backing_object_offset); /* * Initial conditions */ if (op & OBSC_TEST_ALL_SHADOWED) { /* * We do not want to have to test for the existence of cache * or swap pages in the backing object. XXX but with the * new swapper this would be pretty easy to do. * * XXX what about anonymous MAP_SHARED memory that hasn't * been ZFOD faulted yet? If we do not test for this, the * shadow test may succeed! XXX */ if (backing_object->type != OBJT_DEFAULT) { return (0); } } if (op & OBSC_COLLAPSE_WAIT) { vm_object_set_flag(backing_object, OBJ_DEAD); } /* * Our scan */ p = TAILQ_FIRST(&backing_object->memq); while (p) { vm_page_t next = TAILQ_NEXT(p, listq); vm_pindex_t new_pindex = p->pindex - backing_offset_index; if (op & OBSC_TEST_ALL_SHADOWED) { vm_page_t pp; /* * Ignore pages outside the parent object's range * and outside the parent object's mapping of the * backing object. * * note that we do not busy the backing object's * page. */ if ( p->pindex < backing_offset_index || new_pindex >= object->size ) { p = next; continue; } /* * See if the parent has the page or if the parent's * object pager has the page. If the parent has the * page but the page is not valid, the parent's * object pager must have the page. * * If this fails, the parent does not completely shadow * the object and we might as well give up now. */ pp = vm_page_lookup(object, new_pindex); if ( (pp == NULL || pp->valid == 0) && !vm_pager_has_page(object, new_pindex, NULL, NULL) ) { r = 0; break; } } /* * Check for busy page */ if (op & (OBSC_COLLAPSE_WAIT | OBSC_COLLAPSE_NOWAIT)) { vm_page_t pp; if (op & OBSC_COLLAPSE_NOWAIT) { if (!p->valid || vm_page_busied(p)) { p = next; continue; } } else if (op & OBSC_COLLAPSE_WAIT) { if (vm_page_busied(p)) { VM_OBJECT_WUNLOCK(object); vm_page_lock(p); VM_OBJECT_WUNLOCK(backing_object); vm_page_busy_sleep(p, "vmocol"); VM_OBJECT_WLOCK(object); VM_OBJECT_WLOCK(backing_object); /* * If we slept, anything could have * happened. Since the object is * marked dead, the backing offset * should not have changed so we * just restart our scan. */ p = TAILQ_FIRST(&backing_object->memq); continue; } } KASSERT( p->object == backing_object, ("vm_object_backing_scan: object mismatch") ); if ( p->pindex < backing_offset_index || new_pindex >= object->size ) { if (backing_object->type == OBJT_SWAP) swap_pager_freespace(backing_object, p->pindex, 1); /* * Page is out of the parent object's range, we * can simply destroy it. */ vm_page_lock(p); KASSERT(!pmap_page_is_mapped(p), ("freeing mapped page %p", p)); if (p->wire_count == 0) vm_page_free(p); else vm_page_remove(p); vm_page_unlock(p); p = next; continue; } pp = vm_page_lookup(object, new_pindex); if ( (op & OBSC_COLLAPSE_NOWAIT) != 0 && (pp != NULL && pp->valid == 0) ) { if (backing_object->type == OBJT_SWAP) swap_pager_freespace(backing_object, p->pindex, 1); /* * The page in the parent is not (yet) valid. * We don't know anything about the state of * the original page. It might be mapped, * so we must avoid the next if here. * * This is due to a race in vm_fault() where * we must unbusy the original (backing_obj) * page before we can (re)lock the parent. * Hence we can get here. */ p = next; continue; } if ( pp != NULL || vm_pager_has_page(object, new_pindex, NULL, NULL) ) { if (backing_object->type == OBJT_SWAP) swap_pager_freespace(backing_object, p->pindex, 1); /* * page already exists in parent OR swap exists * for this location in the parent. Destroy * the original page from the backing object. * * Leave the parent's page alone */ vm_page_lock(p); KASSERT(!pmap_page_is_mapped(p), ("freeing mapped page %p", p)); if (p->wire_count == 0) vm_page_free(p); else vm_page_remove(p); vm_page_unlock(p); p = next; continue; } /* * Page does not exist in parent, rename the * page from the backing object to the main object. * * If the page was mapped to a process, it can remain * mapped through the rename. * vm_page_rename() will handle dirty and cache. */ if (vm_page_rename(p, object, new_pindex)) { if (op & OBSC_COLLAPSE_NOWAIT) { p = next; continue; } VM_OBJECT_WUNLOCK(backing_object); VM_OBJECT_WUNLOCK(object); VM_WAIT; VM_OBJECT_WLOCK(object); VM_OBJECT_WLOCK(backing_object); p = TAILQ_FIRST(&backing_object->memq); continue; } /* Use the old pindex to free the right page. */ if (backing_object->type == OBJT_SWAP) swap_pager_freespace(backing_object, new_pindex + backing_offset_index, 1); #if VM_NRESERVLEVEL > 0 /* * Rename the reservation. */ vm_reserv_rename(p, object, backing_object, backing_offset_index); #endif } p = next; } return (r); } /* * this version of collapse allows the operation to occur earlier and * when paging_in_progress is true for an object... This is not a complete * operation, but should plug 99.9% of the rest of the leaks. */ static void vm_object_qcollapse(vm_object_t object) { vm_object_t backing_object = object->backing_object; VM_OBJECT_ASSERT_WLOCKED(object); VM_OBJECT_ASSERT_WLOCKED(backing_object); if (backing_object->ref_count != 1) return; vm_object_backing_scan(object, OBSC_COLLAPSE_NOWAIT); } /* * vm_object_collapse: * * Collapse an object with the object backing it. * Pages in the backing object are moved into the * parent, and the backing object is deallocated. */ void vm_object_collapse(vm_object_t object) { VM_OBJECT_ASSERT_WLOCKED(object); while (TRUE) { vm_object_t backing_object; /* * Verify that the conditions are right for collapse: * * The object exists and the backing object exists. */ if ((backing_object = object->backing_object) == NULL) break; /* * we check the backing object first, because it is most likely * not collapsable. */ VM_OBJECT_WLOCK(backing_object); if (backing_object->handle != NULL || (backing_object->type != OBJT_DEFAULT && backing_object->type != OBJT_SWAP) || (backing_object->flags & OBJ_DEAD) || object->handle != NULL || (object->type != OBJT_DEFAULT && object->type != OBJT_SWAP) || (object->flags & OBJ_DEAD)) { VM_OBJECT_WUNLOCK(backing_object); break; } if ( object->paging_in_progress != 0 || backing_object->paging_in_progress != 0 ) { vm_object_qcollapse(object); VM_OBJECT_WUNLOCK(backing_object); break; } /* * We know that we can either collapse the backing object (if * the parent is the only reference to it) or (perhaps) have * the parent bypass the object if the parent happens to shadow * all the resident pages in the entire backing object. * * This is ignoring pager-backed pages such as swap pages. * vm_object_backing_scan fails the shadowing test in this * case. */ if (backing_object->ref_count == 1) { /* * If there is exactly one reference to the backing * object, we can collapse it into the parent. */ vm_object_backing_scan(object, OBSC_COLLAPSE_WAIT); #if VM_NRESERVLEVEL > 0 /* * Break any reservations from backing_object. */ if (__predict_false(!LIST_EMPTY(&backing_object->rvq))) vm_reserv_break_all(backing_object); #endif /* * Move the pager from backing_object to object. */ if (backing_object->type == OBJT_SWAP) { /* * swap_pager_copy() can sleep, in which case * the backing_object's and object's locks are * released and reacquired. * Since swap_pager_copy() is being asked to * destroy the source, it will change the * backing_object's type to OBJT_DEFAULT. */ swap_pager_copy( backing_object, object, OFF_TO_IDX(object->backing_object_offset), TRUE); /* * Free any cached pages from backing_object. */ if (__predict_false( !vm_object_cache_is_empty(backing_object))) vm_page_cache_free(backing_object, 0, 0); } /* * Object now shadows whatever backing_object did. * Note that the reference to * backing_object->backing_object moves from within * backing_object to within object. */ LIST_REMOVE(object, shadow_list); backing_object->shadow_count--; if (backing_object->backing_object) { VM_OBJECT_WLOCK(backing_object->backing_object); LIST_REMOVE(backing_object, shadow_list); LIST_INSERT_HEAD( &backing_object->backing_object->shadow_head, object, shadow_list); /* * The shadow_count has not changed. */ VM_OBJECT_WUNLOCK(backing_object->backing_object); } object->backing_object = backing_object->backing_object; object->backing_object_offset += backing_object->backing_object_offset; /* * Discard backing_object. * * Since the backing object has no pages, no pager left, * and no object references within it, all that is * necessary is to dispose of it. */ KASSERT(backing_object->ref_count == 1, ( "backing_object %p was somehow re-referenced during collapse!", backing_object)); backing_object->type = OBJT_DEAD; backing_object->ref_count = 0; VM_OBJECT_WUNLOCK(backing_object); vm_object_destroy(backing_object); object_collapses++; } else { vm_object_t new_backing_object; /* * If we do not entirely shadow the backing object, * there is nothing we can do so we give up. */ if (object->resident_page_count != object->size && vm_object_backing_scan(object, OBSC_TEST_ALL_SHADOWED) == 0) { VM_OBJECT_WUNLOCK(backing_object); break; } /* * Make the parent shadow the next object in the * chain. Deallocating backing_object will not remove * it, since its reference count is at least 2. */ LIST_REMOVE(object, shadow_list); backing_object->shadow_count--; new_backing_object = backing_object->backing_object; if ((object->backing_object = new_backing_object) != NULL) { VM_OBJECT_WLOCK(new_backing_object); LIST_INSERT_HEAD( &new_backing_object->shadow_head, object, shadow_list ); new_backing_object->shadow_count++; vm_object_reference_locked(new_backing_object); VM_OBJECT_WUNLOCK(new_backing_object); object->backing_object_offset += backing_object->backing_object_offset; } /* * Drop the reference count on backing_object. Since * its ref_count was at least 2, it will not vanish. */ backing_object->ref_count--; VM_OBJECT_WUNLOCK(backing_object); object_bypasses++; } /* * Try again with this object's new backing object. */ } } /* * vm_object_page_remove: * * For the given object, either frees or invalidates each of the * specified pages. In general, a page is freed. However, if a page is * wired for any reason other than the existence of a managed, wired * mapping, then it may be invalidated but not removed from the object. * Pages are specified by the given range ["start", "end") and the option * OBJPR_CLEANONLY. As a special case, if "end" is zero, then the range * extends from "start" to the end of the object. If the option * OBJPR_CLEANONLY is specified, then only the non-dirty pages within the * specified range are affected. If the option OBJPR_NOTMAPPED is * specified, then the pages within the specified range must have no * mappings. Otherwise, if this option is not specified, any mappings to * the specified pages are removed before the pages are freed or * invalidated. * * In general, this operation should only be performed on objects that * contain managed pages. There are, however, two exceptions. First, it * is performed on the kernel and kmem objects by vm_map_entry_delete(). * Second, it is used by msync(..., MS_INVALIDATE) to invalidate device- * backed pages. In both of these cases, the option OBJPR_CLEANONLY must * not be specified and the option OBJPR_NOTMAPPED must be specified. * * The object must be locked. */ void vm_object_page_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end, int options) { vm_page_t p, next; VM_OBJECT_ASSERT_WLOCKED(object); KASSERT((object->flags & OBJ_UNMANAGED) == 0 || (options & (OBJPR_CLEANONLY | OBJPR_NOTMAPPED)) == OBJPR_NOTMAPPED, ("vm_object_page_remove: illegal options for object %p", object)); if (object->resident_page_count == 0) goto skipmemq; vm_object_pip_add(object, 1); again: p = vm_page_find_least(object, start); /* * Here, the variable "p" is either (1) the page with the least pindex * greater than or equal to the parameter "start" or (2) NULL. */ for (; p != NULL && (p->pindex < end || end == 0); p = next) { next = TAILQ_NEXT(p, listq); /* * If the page is wired for any reason besides the existence * of managed, wired mappings, then it cannot be freed. For * example, fictitious pages, which represent device memory, * are inherently wired and cannot be freed. They can, * however, be invalidated if the option OBJPR_CLEANONLY is * not specified. */ vm_page_lock(p); if (vm_page_xbusied(p)) { VM_OBJECT_WUNLOCK(object); vm_page_busy_sleep(p, "vmopax"); VM_OBJECT_WLOCK(object); goto again; } if (p->wire_count != 0) { if ((options & OBJPR_NOTMAPPED) == 0) pmap_remove_all(p); if ((options & OBJPR_CLEANONLY) == 0) { p->valid = 0; vm_page_undirty(p); } goto next; } if (vm_page_busied(p)) { VM_OBJECT_WUNLOCK(object); vm_page_busy_sleep(p, "vmopar"); VM_OBJECT_WLOCK(object); goto again; } KASSERT((p->flags & PG_FICTITIOUS) == 0, ("vm_object_page_remove: page %p is fictitious", p)); if ((options & OBJPR_CLEANONLY) != 0 && p->valid != 0) { if ((options & OBJPR_NOTMAPPED) == 0) pmap_remove_write(p); if (p->dirty) goto next; } if ((options & OBJPR_NOTMAPPED) == 0) pmap_remove_all(p); vm_page_free(p); next: vm_page_unlock(p); } vm_object_pip_wakeup(object); skipmemq: if (__predict_false(!vm_object_cache_is_empty(object))) vm_page_cache_free(object, start, end); } /* - * vm_object_page_cache: + * vm_object_page_noreuse: * - * For the given object, attempt to move the specified clean - * pages to the cache queue. If a page is wired for any reason, - * then it will not be changed. Pages are specified by the given - * range ["start", "end"). As a special case, if "end" is zero, - * then the range extends from "start" to the end of the object. - * Any mappings to the specified pages are removed before the - * pages are moved to the cache queue. + * For the given object, attempt to move the specified pages to + * the head of the inactive queue. This bypasses regular LRU + * operation and allows the pages to be reused quickly under memory + * pressure. If a page is wired for any reason, then it will not + * be queued. Pages are specified by the range ["start", "end"). + * As a special case, if "end" is zero, then the range extends from + * "start" to the end of the object. * * This operation should only be performed on objects that * contain non-fictitious, managed pages. * * The object must be locked. */ void -vm_object_page_cache(vm_object_t object, vm_pindex_t start, vm_pindex_t end) +vm_object_page_noreuse(vm_object_t object, vm_pindex_t start, vm_pindex_t end) { struct mtx *mtx, *new_mtx; vm_page_t p, next; VM_OBJECT_ASSERT_WLOCKED(object); KASSERT((object->flags & (OBJ_FICTITIOUS | OBJ_UNMANAGED)) == 0, - ("vm_object_page_cache: illegal object %p", object)); + ("vm_object_page_noreuse: illegal object %p", object)); if (object->resident_page_count == 0) return; p = vm_page_find_least(object, start); /* * Here, the variable "p" is either (1) the page with the least pindex * greater than or equal to the parameter "start" or (2) NULL. */ mtx = NULL; for (; p != NULL && (p->pindex < end || end == 0); p = next) { next = TAILQ_NEXT(p, listq); /* * Avoid releasing and reacquiring the same page lock. */ new_mtx = vm_page_lockptr(p); if (mtx != new_mtx) { if (mtx != NULL) mtx_unlock(mtx); mtx = new_mtx; mtx_lock(mtx); } - vm_page_try_to_cache(p); + vm_page_deactivate_noreuse(p); } if (mtx != NULL) mtx_unlock(mtx); } /* * Populate the specified range of the object with valid pages. Returns * TRUE if the range is successfully populated and FALSE otherwise. * * Note: This function should be optimized to pass a larger array of * pages to vm_pager_get_pages() before it is applied to a non- * OBJT_DEVICE object. * * The object must be locked. */ boolean_t vm_object_populate(vm_object_t object, vm_pindex_t start, vm_pindex_t end) { vm_page_t m; vm_pindex_t pindex; int rv; VM_OBJECT_ASSERT_WLOCKED(object); for (pindex = start; pindex < end; pindex++) { m = vm_page_grab(object, pindex, VM_ALLOC_NORMAL); if (m->valid != VM_PAGE_BITS_ALL) { rv = vm_pager_get_pages(object, &m, 1, 0); if (rv != VM_PAGER_OK) { vm_page_lock(m); vm_page_free(m); vm_page_unlock(m); break; } } /* * Keep "m" busy because a subsequent iteration may unlock * the object. */ } if (pindex > start) { m = vm_page_lookup(object, start); while (m != NULL && m->pindex < pindex) { vm_page_xunbusy(m); m = TAILQ_NEXT(m, listq); } } return (pindex == end); } /* * Routine: vm_object_coalesce * Function: Coalesces two objects backing up adjoining * regions of memory into a single object. * * returns TRUE if objects were combined. * * NOTE: Only works at the moment if the second object is NULL - * if it's not, which object do we lock first? * * Parameters: * prev_object First object to coalesce * prev_offset Offset into prev_object * prev_size Size of reference to prev_object * next_size Size of reference to the second object * reserved Indicator that extension region has * swap accounted for * * Conditions: * The object must *not* be locked. */ boolean_t vm_object_coalesce(vm_object_t prev_object, vm_ooffset_t prev_offset, vm_size_t prev_size, vm_size_t next_size, boolean_t reserved) { vm_pindex_t next_pindex; if (prev_object == NULL) return (TRUE); VM_OBJECT_WLOCK(prev_object); if ((prev_object->type != OBJT_DEFAULT && prev_object->type != OBJT_SWAP) || (prev_object->flags & OBJ_TMPFS_NODE) != 0) { VM_OBJECT_WUNLOCK(prev_object); return (FALSE); } /* * Try to collapse the object first */ vm_object_collapse(prev_object); /* * Can't coalesce if: . more than one reference . paged out . shadows * another object . has a copy elsewhere (any of which mean that the * pages not mapped to prev_entry may be in use anyway) */ if (prev_object->backing_object != NULL) { VM_OBJECT_WUNLOCK(prev_object); return (FALSE); } prev_size >>= PAGE_SHIFT; next_size >>= PAGE_SHIFT; next_pindex = OFF_TO_IDX(prev_offset) + prev_size; if ((prev_object->ref_count > 1) && (prev_object->size != next_pindex)) { VM_OBJECT_WUNLOCK(prev_object); return (FALSE); } /* * Account for the charge. */ if (prev_object->cred != NULL) { /* * If prev_object was charged, then this mapping, * althought not charged now, may become writable * later. Non-NULL cred in the object would prevent * swap reservation during enabling of the write * access, so reserve swap now. Failed reservation * cause allocation of the separate object for the map * entry, and swap reservation for this entry is * managed in appropriate time. */ if (!reserved && !swap_reserve_by_cred(ptoa(next_size), prev_object->cred)) { return (FALSE); } prev_object->charge += ptoa(next_size); } /* * Remove any pages that may still be in the object from a previous * deallocation. */ if (next_pindex < prev_object->size) { vm_object_page_remove(prev_object, next_pindex, next_pindex + next_size, 0); if (prev_object->type == OBJT_SWAP) swap_pager_freespace(prev_object, next_pindex, next_size); #if 0 if (prev_object->cred != NULL) { KASSERT(prev_object->charge >= ptoa(prev_object->size - next_pindex), ("object %p overcharged 1 %jx %jx", prev_object, (uintmax_t)next_pindex, (uintmax_t)next_size)); prev_object->charge -= ptoa(prev_object->size - next_pindex); } #endif } /* * Extend the object if necessary. */ if (next_pindex + next_size > prev_object->size) prev_object->size = next_pindex + next_size; VM_OBJECT_WUNLOCK(prev_object); return (TRUE); } void vm_object_set_writeable_dirty(vm_object_t object) { VM_OBJECT_ASSERT_WLOCKED(object); if (object->type != OBJT_VNODE) { if ((object->flags & OBJ_TMPFS_NODE) != 0) { KASSERT(object->type == OBJT_SWAP, ("non-swap tmpfs")); vm_object_set_flag(object, OBJ_TMPFS_DIRTY); } return; } object->generation++; if ((object->flags & OBJ_MIGHTBEDIRTY) != 0) return; vm_object_set_flag(object, OBJ_MIGHTBEDIRTY); } /* * vm_object_unwire: * * For each page offset within the specified range of the given object, * find the highest-level page in the shadow chain and unwire it. A page * must exist at every page offset, and the highest-level page must be * wired. */ void vm_object_unwire(vm_object_t object, vm_ooffset_t offset, vm_size_t length, uint8_t queue) { vm_object_t tobject; vm_page_t m, tm; vm_pindex_t end_pindex, pindex, tpindex; int depth, locked_depth; KASSERT((offset & PAGE_MASK) == 0, ("vm_object_unwire: offset is not page aligned")); KASSERT((length & PAGE_MASK) == 0, ("vm_object_unwire: length is not a multiple of PAGE_SIZE")); /* The wired count of a fictitious page never changes. */ if ((object->flags & OBJ_FICTITIOUS) != 0) return; pindex = OFF_TO_IDX(offset); end_pindex = pindex + atop(length); locked_depth = 1; VM_OBJECT_RLOCK(object); m = vm_page_find_least(object, pindex); while (pindex < end_pindex) { if (m == NULL || pindex < m->pindex) { /* * The first object in the shadow chain doesn't * contain a page at the current index. Therefore, * the page must exist in a backing object. */ tobject = object; tpindex = pindex; depth = 0; do { tpindex += OFF_TO_IDX(tobject->backing_object_offset); tobject = tobject->backing_object; KASSERT(tobject != NULL, ("vm_object_unwire: missing page")); if ((tobject->flags & OBJ_FICTITIOUS) != 0) goto next_page; depth++; if (depth == locked_depth) { locked_depth++; VM_OBJECT_RLOCK(tobject); } } while ((tm = vm_page_lookup(tobject, tpindex)) == NULL); } else { tm = m; m = TAILQ_NEXT(m, listq); } vm_page_lock(tm); vm_page_unwire(tm, queue); vm_page_unlock(tm); next_page: pindex++; } /* Release the accumulated object locks. */ for (depth = 0; depth < locked_depth; depth++) { tobject = object->backing_object; VM_OBJECT_RUNLOCK(object); object = tobject; } } struct vnode * vm_object_vnode(vm_object_t object) { VM_OBJECT_ASSERT_LOCKED(object); if (object->type == OBJT_VNODE) return (object->handle); if (object->type == OBJT_SWAP && (object->flags & OBJ_TMPFS) != 0) return (object->un_pager.swp.swp_tmpfs); return (NULL); } static int sysctl_vm_object_list(SYSCTL_HANDLER_ARGS) { struct kinfo_vmobject kvo; char *fullpath, *freepath; struct vnode *vp; struct vattr va; vm_object_t obj; vm_page_t m; int count, error; if (req->oldptr == NULL) { /* * If an old buffer has not been provided, generate an * estimate of the space needed for a subsequent call. */ mtx_lock(&vm_object_list_mtx); count = 0; TAILQ_FOREACH(obj, &vm_object_list, object_list) { if (obj->type == OBJT_DEAD) continue; count++; } mtx_unlock(&vm_object_list_mtx); return (SYSCTL_OUT(req, NULL, sizeof(struct kinfo_vmobject) * count * 11 / 10)); } error = 0; /* * VM objects are type stable and are never removed from the * list once added. This allows us to safely read obj->object_list * after reacquiring the VM object lock. */ mtx_lock(&vm_object_list_mtx); TAILQ_FOREACH(obj, &vm_object_list, object_list) { if (obj->type == OBJT_DEAD) continue; VM_OBJECT_RLOCK(obj); if (obj->type == OBJT_DEAD) { VM_OBJECT_RUNLOCK(obj); continue; } mtx_unlock(&vm_object_list_mtx); kvo.kvo_size = ptoa(obj->size); kvo.kvo_resident = obj->resident_page_count; kvo.kvo_ref_count = obj->ref_count; kvo.kvo_shadow_count = obj->shadow_count; kvo.kvo_memattr = obj->memattr; kvo.kvo_active = 0; kvo.kvo_inactive = 0; TAILQ_FOREACH(m, &obj->memq, listq) { /* * A page may belong to the object but be * dequeued and set to PQ_NONE while the * object lock is not held. This makes the * reads of m->queue below racy, and we do not * count pages set to PQ_NONE. However, this * sysctl is only meant to give an * approximation of the system anyway. */ if (m->queue == PQ_ACTIVE) kvo.kvo_active++; else if (m->queue == PQ_INACTIVE) kvo.kvo_inactive++; } kvo.kvo_vn_fileid = 0; kvo.kvo_vn_fsid = 0; freepath = NULL; fullpath = ""; vp = NULL; switch (obj->type) { case OBJT_DEFAULT: kvo.kvo_type = KVME_TYPE_DEFAULT; break; case OBJT_VNODE: kvo.kvo_type = KVME_TYPE_VNODE; vp = obj->handle; vref(vp); break; case OBJT_SWAP: kvo.kvo_type = KVME_TYPE_SWAP; break; case OBJT_DEVICE: kvo.kvo_type = KVME_TYPE_DEVICE; break; case OBJT_PHYS: kvo.kvo_type = KVME_TYPE_PHYS; break; case OBJT_DEAD: kvo.kvo_type = KVME_TYPE_DEAD; break; case OBJT_SG: kvo.kvo_type = KVME_TYPE_SG; break; case OBJT_MGTDEVICE: kvo.kvo_type = KVME_TYPE_MGTDEVICE; break; default: kvo.kvo_type = KVME_TYPE_UNKNOWN; break; } VM_OBJECT_RUNLOCK(obj); if (vp != NULL) { vn_fullpath(curthread, vp, &fullpath, &freepath); vn_lock(vp, LK_SHARED | LK_RETRY); if (VOP_GETATTR(vp, &va, curthread->td_ucred) == 0) { kvo.kvo_vn_fileid = va.va_fileid; kvo.kvo_vn_fsid = va.va_fsid; } vput(vp); } strlcpy(kvo.kvo_path, fullpath, sizeof(kvo.kvo_path)); if (freepath != NULL) free(freepath, M_TEMP); /* Pack record size down */ kvo.kvo_structsize = offsetof(struct kinfo_vmobject, kvo_path) + strlen(kvo.kvo_path) + 1; kvo.kvo_structsize = roundup(kvo.kvo_structsize, sizeof(uint64_t)); error = SYSCTL_OUT(req, &kvo, kvo.kvo_structsize); mtx_lock(&vm_object_list_mtx); if (error) break; } mtx_unlock(&vm_object_list_mtx); return (error); } SYSCTL_PROC(_vm, OID_AUTO, objects, CTLTYPE_STRUCT | CTLFLAG_RW | CTLFLAG_SKIP | CTLFLAG_MPSAFE, NULL, 0, sysctl_vm_object_list, "S,kinfo_vmobject", "List of VM objects"); #include "opt_ddb.h" #ifdef DDB #include #include #include static int _vm_object_in_map(vm_map_t map, vm_object_t object, vm_map_entry_t entry) { vm_map_t tmpm; vm_map_entry_t tmpe; vm_object_t obj; int entcount; if (map == 0) return 0; if (entry == 0) { tmpe = map->header.next; entcount = map->nentries; while (entcount-- && (tmpe != &map->header)) { if (_vm_object_in_map(map, object, tmpe)) { return 1; } tmpe = tmpe->next; } } else if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { tmpm = entry->object.sub_map; tmpe = tmpm->header.next; entcount = tmpm->nentries; while (entcount-- && tmpe != &tmpm->header) { if (_vm_object_in_map(tmpm, object, tmpe)) { return 1; } tmpe = tmpe->next; } } else if ((obj = entry->object.vm_object) != NULL) { for (; obj; obj = obj->backing_object) if (obj == object) { return 1; } } return 0; } static int vm_object_in_map(vm_object_t object) { struct proc *p; /* sx_slock(&allproc_lock); */ FOREACH_PROC_IN_SYSTEM(p) { if (!p->p_vmspace /* || (p->p_flag & (P_SYSTEM|P_WEXIT)) */) continue; if (_vm_object_in_map(&p->p_vmspace->vm_map, object, 0)) { /* sx_sunlock(&allproc_lock); */ return 1; } } /* sx_sunlock(&allproc_lock); */ if (_vm_object_in_map(kernel_map, object, 0)) return 1; return 0; } DB_SHOW_COMMAND(vmochk, vm_object_check) { vm_object_t object; /* * make sure that internal objs are in a map somewhere * and none have zero ref counts. */ TAILQ_FOREACH(object, &vm_object_list, object_list) { if (object->handle == NULL && (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) { if (object->ref_count == 0) { db_printf("vmochk: internal obj has zero ref count: %ld\n", (long)object->size); } if (!vm_object_in_map(object)) { db_printf( "vmochk: internal obj is not in a map: " "ref: %d, size: %lu: 0x%lx, backing_object: %p\n", object->ref_count, (u_long)object->size, (u_long)object->size, (void *)object->backing_object); } } } } /* * vm_object_print: [ debug ] */ DB_SHOW_COMMAND(object, vm_object_print_static) { /* XXX convert args. */ vm_object_t object = (vm_object_t)addr; boolean_t full = have_addr; vm_page_t p; /* XXX count is an (unused) arg. Avoid shadowing it. */ #define count was_count int count; if (object == NULL) return; db_iprintf( "Object %p: type=%d, size=0x%jx, res=%d, ref=%d, flags=0x%x ruid %d charge %jx\n", object, (int)object->type, (uintmax_t)object->size, object->resident_page_count, object->ref_count, object->flags, object->cred ? object->cred->cr_ruid : -1, (uintmax_t)object->charge); db_iprintf(" sref=%d, backing_object(%d)=(%p)+0x%jx\n", object->shadow_count, object->backing_object ? object->backing_object->ref_count : 0, object->backing_object, (uintmax_t)object->backing_object_offset); if (!full) return; db_indent += 2; count = 0; TAILQ_FOREACH(p, &object->memq, listq) { if (count == 0) db_iprintf("memory:="); else if (count == 6) { db_printf("\n"); db_iprintf(" ..."); count = 0; } else db_printf(","); count++; db_printf("(off=0x%jx,page=0x%jx)", (uintmax_t)p->pindex, (uintmax_t)VM_PAGE_TO_PHYS(p)); } if (count != 0) db_printf("\n"); db_indent -= 2; } /* XXX. */ #undef count /* XXX need this non-static entry for calling from vm_map_print. */ void vm_object_print( /* db_expr_t */ long addr, boolean_t have_addr, /* db_expr_t */ long count, char *modif) { vm_object_print_static(addr, have_addr, count, modif); } DB_SHOW_COMMAND(vmopag, vm_object_print_pages) { vm_object_t object; vm_pindex_t fidx; vm_paddr_t pa; vm_page_t m, prev_m; int rcount, nl, c; nl = 0; TAILQ_FOREACH(object, &vm_object_list, object_list) { db_printf("new object: %p\n", (void *)object); if (nl > 18) { c = cngetc(); if (c != ' ') return; nl = 0; } nl++; rcount = 0; fidx = 0; pa = -1; TAILQ_FOREACH(m, &object->memq, listq) { if (m->pindex > 128) break; if ((prev_m = TAILQ_PREV(m, pglist, listq)) != NULL && prev_m->pindex + 1 != m->pindex) { if (rcount) { db_printf(" index(%ld)run(%d)pa(0x%lx)\n", (long)fidx, rcount, (long)pa); if (nl > 18) { c = cngetc(); if (c != ' ') return; nl = 0; } nl++; rcount = 0; } } if (rcount && (VM_PAGE_TO_PHYS(m) == pa + rcount * PAGE_SIZE)) { ++rcount; continue; } if (rcount) { db_printf(" index(%ld)run(%d)pa(0x%lx)\n", (long)fidx, rcount, (long)pa); if (nl > 18) { c = cngetc(); if (c != ' ') return; nl = 0; } nl++; } fidx = m->pindex; pa = VM_PAGE_TO_PHYS(m); rcount = 1; } if (rcount) { db_printf(" index(%ld)run(%d)pa(0x%lx)\n", (long)fidx, rcount, (long)pa); if (nl > 18) { c = cngetc(); if (c != ' ') return; nl = 0; } nl++; } } } #endif /* DDB */ Index: head/sys/vm/vm_object.h =================================================================== --- head/sys/vm/vm_object.h (revision 288430) +++ head/sys/vm/vm_object.h (revision 288431) @@ -1,327 +1,327 @@ /*- * Copyright (c) 1991, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * The Mach Operating System project at Carnegie-Mellon University. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * from: @(#)vm_object.h 8.3 (Berkeley) 1/12/94 * * * Copyright (c) 1987, 1990 Carnegie-Mellon University. * All rights reserved. * * Authors: Avadis Tevanian, Jr., Michael Wayne Young * * Permission to use, copy, modify and distribute this software and * its documentation is hereby granted, provided that both the copyright * notice and this permission notice appear in all copies of the * software, derivative works or modified versions, and any portions * thereof, and that both notices appear in supporting documentation. * * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. * * Carnegie Mellon requests users of this software to return to * * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU * School of Computer Science * Carnegie Mellon University * Pittsburgh PA 15213-3890 * * any improvements or extensions that they make and grant Carnegie the * rights to redistribute these changes. * * $FreeBSD$ */ /* * Virtual memory object module definitions. */ #ifndef _VM_OBJECT_ #define _VM_OBJECT_ #include #include #include #include #include /* * Types defined: * * vm_object_t Virtual memory object. * * The root of cached pages pool is protected by both the per-object lock * and the free pages queue mutex. * On insert in the cache radix trie, the per-object lock is expected * to be already held and the free pages queue mutex will be * acquired during the operation too. * On remove and lookup from the cache radix trie, only the free * pages queue mutex is expected to be locked. * These rules allow for reliably checking for the presence of cached * pages with only the per-object lock held, thereby reducing contention * for the free pages queue mutex. * * List of locks * (c) const until freed * (o) per-object lock * (f) free pages queue mutex * */ struct vm_object { struct rwlock lock; TAILQ_ENTRY(vm_object) object_list; /* list of all objects */ LIST_HEAD(, vm_object) shadow_head; /* objects that this is a shadow for */ LIST_ENTRY(vm_object) shadow_list; /* chain of shadow objects */ TAILQ_HEAD(respgs, vm_page) memq; /* list of resident pages */ struct vm_radix rtree; /* root of the resident page radix trie*/ vm_pindex_t size; /* Object size */ int generation; /* generation ID */ int ref_count; /* How many refs?? */ int shadow_count; /* how many objects that this is a shadow for */ vm_memattr_t memattr; /* default memory attribute for pages */ objtype_t type; /* type of pager */ u_short flags; /* see below */ u_short pg_color; /* (c) color of first page in obj */ u_int paging_in_progress; /* Paging (in or out) so don't collapse or destroy */ int resident_page_count; /* number of resident pages */ struct vm_object *backing_object; /* object that I'm a shadow of */ vm_ooffset_t backing_object_offset;/* Offset in backing object */ TAILQ_ENTRY(vm_object) pager_object_list; /* list of all objects of this pager type */ LIST_HEAD(, vm_reserv) rvq; /* list of reservations */ struct vm_radix cache; /* (o + f) root of the cache page radix trie */ void *handle; union { /* * VNode pager * * vnp_size - current size of file */ struct { off_t vnp_size; vm_ooffset_t writemappings; } vnp; /* * Device pager * * devp_pglist - list of allocated pages */ struct { TAILQ_HEAD(, vm_page) devp_pglist; struct cdev_pager_ops *ops; struct cdev *dev; } devp; /* * SG pager * * sgp_pglist - list of allocated pages */ struct { TAILQ_HEAD(, vm_page) sgp_pglist; } sgp; /* * Swap pager * * swp_tmpfs - back-pointer to the tmpfs vnode, * if any, which uses the vm object * as backing store. The handle * cannot be reused for linking, * because the vnode can be * reclaimed and recreated, making * the handle changed and hash-chain * invalid. * * swp_bcount - number of swap 'swblock' metablocks, each * contains up to 16 swapblk assignments. * see vm/swap_pager.h */ struct { void *swp_tmpfs; int swp_bcount; } swp; } un_pager; struct ucred *cred; vm_ooffset_t charge; }; /* * Flags */ #define OBJ_FICTITIOUS 0x0001 /* (c) contains fictitious pages */ #define OBJ_UNMANAGED 0x0002 /* (c) contains unmanaged pages */ #define OBJ_ACTIVE 0x0004 /* active objects */ #define OBJ_DEAD 0x0008 /* dead objects (during rundown) */ #define OBJ_NOSPLIT 0x0010 /* dont split this object */ #define OBJ_PIPWNT 0x0040 /* paging in progress wanted */ #define OBJ_MIGHTBEDIRTY 0x0100 /* object might be dirty, only for vnode */ #define OBJ_TMPFS_NODE 0x0200 /* object belongs to tmpfs VREG node */ #define OBJ_TMPFS_DIRTY 0x0400 /* dirty tmpfs obj */ #define OBJ_COLORED 0x1000 /* pg_color is defined */ #define OBJ_ONEMAPPING 0x2000 /* One USE (a single, non-forked) mapping flag */ #define OBJ_DISCONNECTWNT 0x4000 /* disconnect from vnode wanted */ #define OBJ_TMPFS 0x8000 /* has tmpfs vnode allocated */ #define IDX_TO_OFF(idx) (((vm_ooffset_t)(idx)) << PAGE_SHIFT) #define OFF_TO_IDX(off) ((vm_pindex_t)(((vm_ooffset_t)(off)) >> PAGE_SHIFT)) #ifdef _KERNEL #define OBJPC_SYNC 0x1 /* sync I/O */ #define OBJPC_INVAL 0x2 /* invalidate */ #define OBJPC_NOSYNC 0x4 /* skip if VPO_NOSYNC */ /* * The following options are supported by vm_object_page_remove(). */ #define OBJPR_CLEANONLY 0x1 /* Don't remove dirty pages. */ #define OBJPR_NOTMAPPED 0x2 /* Don't unmap pages. */ TAILQ_HEAD(object_q, vm_object); extern struct object_q vm_object_list; /* list of allocated objects */ extern struct mtx vm_object_list_mtx; /* lock for object list and count */ extern struct vm_object kernel_object_store; extern struct vm_object kmem_object_store; #define kernel_object (&kernel_object_store) #define kmem_object (&kmem_object_store) #define VM_OBJECT_ASSERT_LOCKED(object) \ rw_assert(&(object)->lock, RA_LOCKED) #define VM_OBJECT_ASSERT_RLOCKED(object) \ rw_assert(&(object)->lock, RA_RLOCKED) #define VM_OBJECT_ASSERT_WLOCKED(object) \ rw_assert(&(object)->lock, RA_WLOCKED) #define VM_OBJECT_ASSERT_UNLOCKED(object) \ rw_assert(&(object)->lock, RA_UNLOCKED) #define VM_OBJECT_LOCK_DOWNGRADE(object) \ rw_downgrade(&(object)->lock) #define VM_OBJECT_RLOCK(object) \ rw_rlock(&(object)->lock) #define VM_OBJECT_RUNLOCK(object) \ rw_runlock(&(object)->lock) #define VM_OBJECT_SLEEP(object, wchan, pri, wmesg, timo) \ rw_sleep((wchan), &(object)->lock, (pri), (wmesg), (timo)) #define VM_OBJECT_TRYRLOCK(object) \ rw_try_rlock(&(object)->lock) #define VM_OBJECT_TRYWLOCK(object) \ rw_try_wlock(&(object)->lock) #define VM_OBJECT_TRYUPGRADE(object) \ rw_try_upgrade(&(object)->lock) #define VM_OBJECT_WLOCK(object) \ rw_wlock(&(object)->lock) #define VM_OBJECT_WUNLOCK(object) \ rw_wunlock(&(object)->lock) /* * The object must be locked or thread private. */ static __inline void vm_object_set_flag(vm_object_t object, u_short bits) { object->flags |= bits; } /* * Conditionally set the object's color, which (1) enables the allocation * of physical memory reservations for anonymous objects and larger-than- * superpage-sized named objects and (2) determines the first page offset * within the object at which a reservation may be allocated. In other * words, the color determines the alignment of the object with respect * to the largest superpage boundary. When mapping named objects, like * files or POSIX shared memory objects, the color should be set to zero * before a virtual address is selected for the mapping. In contrast, * for anonymous objects, the color may be set after the virtual address * is selected. * * The object must be locked. */ static __inline void vm_object_color(vm_object_t object, u_short color) { if ((object->flags & OBJ_COLORED) == 0) { object->pg_color = color; object->flags |= OBJ_COLORED; } } void vm_object_clear_flag(vm_object_t object, u_short bits); void vm_object_pip_add(vm_object_t object, short i); void vm_object_pip_subtract(vm_object_t object, short i); void vm_object_pip_wakeup(vm_object_t object); void vm_object_pip_wakeupn(vm_object_t object, short i); void vm_object_pip_wait(vm_object_t object, char *waitid); static __inline boolean_t vm_object_cache_is_empty(vm_object_t object) { return (vm_radix_is_empty(&object->cache)); } vm_object_t vm_object_allocate (objtype_t, vm_pindex_t); boolean_t vm_object_coalesce(vm_object_t, vm_ooffset_t, vm_size_t, vm_size_t, boolean_t); void vm_object_collapse (vm_object_t); void vm_object_deallocate (vm_object_t); void vm_object_destroy (vm_object_t); void vm_object_terminate (vm_object_t); void vm_object_set_writeable_dirty (vm_object_t); void vm_object_init (void); void vm_object_madvise(vm_object_t, vm_pindex_t, vm_pindex_t, int); -void vm_object_page_cache(vm_object_t object, vm_pindex_t start, - vm_pindex_t end); boolean_t vm_object_page_clean(vm_object_t object, vm_ooffset_t start, vm_ooffset_t end, int flags); +void vm_object_page_noreuse(vm_object_t object, vm_pindex_t start, + vm_pindex_t end); void vm_object_page_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end, int options); boolean_t vm_object_populate(vm_object_t, vm_pindex_t, vm_pindex_t); void vm_object_print(long addr, boolean_t have_addr, long count, char *modif); void vm_object_reference (vm_object_t); void vm_object_reference_locked(vm_object_t); int vm_object_set_memattr(vm_object_t object, vm_memattr_t memattr); void vm_object_shadow (vm_object_t *, vm_ooffset_t *, vm_size_t); void vm_object_split(vm_map_entry_t); boolean_t vm_object_sync(vm_object_t, vm_ooffset_t, vm_size_t, boolean_t, boolean_t); void vm_object_unwire(vm_object_t object, vm_ooffset_t offset, vm_size_t length, uint8_t queue); struct vnode *vm_object_vnode(vm_object_t object); #endif /* _KERNEL */ #endif /* _VM_OBJECT_ */ Index: head/sys/vm/vm_page.c =================================================================== --- head/sys/vm/vm_page.c (revision 288430) +++ head/sys/vm/vm_page.c (revision 288431) @@ -1,3341 +1,3353 @@ /*- * Copyright (c) 1991 Regents of the University of California. * All rights reserved. * Copyright (c) 1998 Matthew Dillon. All Rights Reserved. * * This code is derived from software contributed to Berkeley by * The Mach Operating System project at Carnegie-Mellon University. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * from: @(#)vm_page.c 7.4 (Berkeley) 5/7/91 */ /*- * Copyright (c) 1987, 1990 Carnegie-Mellon University. * All rights reserved. * * Authors: Avadis Tevanian, Jr., Michael Wayne Young * * Permission to use, copy, modify and distribute this software and * its documentation is hereby granted, provided that both the copyright * notice and this permission notice appear in all copies of the * software, derivative works or modified versions, and any portions * thereof, and that both notices appear in supporting documentation. * * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. * * Carnegie Mellon requests users of this software to return to * * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU * School of Computer Science * Carnegie Mellon University * Pittsburgh PA 15213-3890 * * any improvements or extensions that they make and grant Carnegie the * rights to redistribute these changes. */ /* * GENERAL RULES ON VM_PAGE MANIPULATION * * - A page queue lock is required when adding or removing a page from a * page queue regardless of other locks or the busy state of a page. * * * In general, no thread besides the page daemon can acquire or * hold more than one page queue lock at a time. * * * The page daemon can acquire and hold any pair of page queue * locks in any order. * * - The object lock is required when inserting or removing * pages from an object (vm_page_insert() or vm_page_remove()). * */ /* * Resident memory management module. */ #include __FBSDID("$FreeBSD$"); #include "opt_vm.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * Associated with page of user-allocatable memory is a * page structure. */ struct vm_domain vm_dom[MAXMEMDOM]; struct mtx_padalign vm_page_queue_free_mtx; struct mtx_padalign pa_lock[PA_LOCK_COUNT]; vm_page_t vm_page_array; long vm_page_array_size; long first_page; int vm_page_zero_count; static int boot_pages = UMA_BOOT_PAGES; SYSCTL_INT(_vm, OID_AUTO, boot_pages, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &boot_pages, 0, "number of pages allocated for bootstrapping the VM system"); static int pa_tryrelock_restart; SYSCTL_INT(_vm, OID_AUTO, tryrelock_restart, CTLFLAG_RD, &pa_tryrelock_restart, 0, "Number of tryrelock restarts"); static TAILQ_HEAD(, vm_page) blacklist_head; static int sysctl_vm_page_blacklist(SYSCTL_HANDLER_ARGS); SYSCTL_PROC(_vm, OID_AUTO, page_blacklist, CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, sysctl_vm_page_blacklist, "A", "Blacklist pages"); static uma_zone_t fakepg_zone; static struct vnode *vm_page_alloc_init(vm_page_t m); static void vm_page_cache_turn_free(vm_page_t m); static void vm_page_clear_dirty_mask(vm_page_t m, vm_page_bits_t pagebits); static void vm_page_enqueue(uint8_t queue, vm_page_t m); static void vm_page_init_fakepg(void *dummy); static int vm_page_insert_after(vm_page_t m, vm_object_t object, vm_pindex_t pindex, vm_page_t mpred); static void vm_page_insert_radixdone(vm_page_t m, vm_object_t object, vm_page_t mpred); SYSINIT(vm_page, SI_SUB_VM, SI_ORDER_SECOND, vm_page_init_fakepg, NULL); static void vm_page_init_fakepg(void *dummy) { fakepg_zone = uma_zcreate("fakepg", sizeof(struct vm_page), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE | UMA_ZONE_VM); } /* Make sure that u_long is at least 64 bits when PAGE_SIZE is 32K. */ #if PAGE_SIZE == 32768 #ifdef CTASSERT CTASSERT(sizeof(u_long) >= 8); #endif #endif /* * Try to acquire a physical address lock while a pmap is locked. If we * fail to trylock we unlock and lock the pmap directly and cache the * locked pa in *locked. The caller should then restart their loop in case * the virtual to physical mapping has changed. */ int vm_page_pa_tryrelock(pmap_t pmap, vm_paddr_t pa, vm_paddr_t *locked) { vm_paddr_t lockpa; lockpa = *locked; *locked = pa; if (lockpa) { PA_LOCK_ASSERT(lockpa, MA_OWNED); if (PA_LOCKPTR(pa) == PA_LOCKPTR(lockpa)) return (0); PA_UNLOCK(lockpa); } if (PA_TRYLOCK(pa)) return (0); PMAP_UNLOCK(pmap); atomic_add_int(&pa_tryrelock_restart, 1); PA_LOCK(pa); PMAP_LOCK(pmap); return (EAGAIN); } /* * vm_set_page_size: * * Sets the page size, perhaps based upon the memory * size. Must be called before any use of page-size * dependent functions. */ void vm_set_page_size(void) { if (vm_cnt.v_page_size == 0) vm_cnt.v_page_size = PAGE_SIZE; if (((vm_cnt.v_page_size - 1) & vm_cnt.v_page_size) != 0) panic("vm_set_page_size: page size not a power of two"); } /* * vm_page_blacklist_next: * * Find the next entry in the provided string of blacklist * addresses. Entries are separated by space, comma, or newline. * If an invalid integer is encountered then the rest of the * string is skipped. Updates the list pointer to the next * character, or NULL if the string is exhausted or invalid. */ static vm_paddr_t vm_page_blacklist_next(char **list, char *end) { vm_paddr_t bad; char *cp, *pos; if (list == NULL || *list == NULL) return (0); if (**list =='\0') { *list = NULL; return (0); } /* * If there's no end pointer then the buffer is coming from * the kenv and we know it's null-terminated. */ if (end == NULL) end = *list + strlen(*list); /* Ensure that strtoq() won't walk off the end */ if (*end != '\0') { if (*end == '\n' || *end == ' ' || *end == ',') *end = '\0'; else { printf("Blacklist not terminated, skipping\n"); *list = NULL; return (0); } } for (pos = *list; *pos != '\0'; pos = cp) { bad = strtoq(pos, &cp, 0); if (*cp == '\0' || *cp == ' ' || *cp == ',' || *cp == '\n') { if (bad == 0) { if (++cp < end) continue; else break; } } else break; if (*cp == '\0' || ++cp >= end) *list = NULL; else *list = cp; return (trunc_page(bad)); } printf("Garbage in RAM blacklist, skipping\n"); *list = NULL; return (0); } /* * vm_page_blacklist_check: * * Iterate through the provided string of blacklist addresses, pulling * each entry out of the physical allocator free list and putting it * onto a list for reporting via the vm.page_blacklist sysctl. */ static void vm_page_blacklist_check(char *list, char *end) { vm_paddr_t pa; vm_page_t m; char *next; int ret; next = list; while (next != NULL) { if ((pa = vm_page_blacklist_next(&next, end)) == 0) continue; m = vm_phys_paddr_to_vm_page(pa); if (m == NULL) continue; mtx_lock(&vm_page_queue_free_mtx); ret = vm_phys_unfree_page(m); mtx_unlock(&vm_page_queue_free_mtx); if (ret == TRUE) { TAILQ_INSERT_TAIL(&blacklist_head, m, listq); if (bootverbose) printf("Skipping page with pa 0x%jx\n", (uintmax_t)pa); } } } /* * vm_page_blacklist_load: * * Search for a special module named "ram_blacklist". It'll be a * plain text file provided by the user via the loader directive * of the same name. */ static void vm_page_blacklist_load(char **list, char **end) { void *mod; u_char *ptr; u_int len; mod = NULL; ptr = NULL; mod = preload_search_by_type("ram_blacklist"); if (mod != NULL) { ptr = preload_fetch_addr(mod); len = preload_fetch_size(mod); } *list = ptr; if (ptr != NULL) *end = ptr + len; else *end = NULL; return; } static int sysctl_vm_page_blacklist(SYSCTL_HANDLER_ARGS) { vm_page_t m; struct sbuf sbuf; int error, first; first = 1; error = sysctl_wire_old_buffer(req, 0); if (error != 0) return (error); sbuf_new_for_sysctl(&sbuf, NULL, 128, req); TAILQ_FOREACH(m, &blacklist_head, listq) { sbuf_printf(&sbuf, "%s%#jx", first ? "" : ",", (uintmax_t)m->phys_addr); first = 0; } error = sbuf_finish(&sbuf); sbuf_delete(&sbuf); return (error); } static void vm_page_domain_init(struct vm_domain *vmd) { struct vm_pagequeue *pq; int i; *__DECONST(char **, &vmd->vmd_pagequeues[PQ_INACTIVE].pq_name) = "vm inactive pagequeue"; *__DECONST(int **, &vmd->vmd_pagequeues[PQ_INACTIVE].pq_vcnt) = &vm_cnt.v_inactive_count; *__DECONST(char **, &vmd->vmd_pagequeues[PQ_ACTIVE].pq_name) = "vm active pagequeue"; *__DECONST(int **, &vmd->vmd_pagequeues[PQ_ACTIVE].pq_vcnt) = &vm_cnt.v_active_count; vmd->vmd_page_count = 0; vmd->vmd_free_count = 0; vmd->vmd_segs = 0; vmd->vmd_oom = FALSE; vmd->vmd_pass = 0; for (i = 0; i < PQ_COUNT; i++) { pq = &vmd->vmd_pagequeues[i]; TAILQ_INIT(&pq->pq_pl); mtx_init(&pq->pq_mutex, pq->pq_name, "vm pagequeue", MTX_DEF | MTX_DUPOK); } } /* * vm_page_startup: * * Initializes the resident memory module. * * Allocates memory for the page cells, and * for the object/offset-to-page hash table headers. * Each page cell is initialized and placed on the free list. */ vm_offset_t vm_page_startup(vm_offset_t vaddr) { vm_offset_t mapped; vm_paddr_t page_range; vm_paddr_t new_end; int i; vm_paddr_t pa; vm_paddr_t last_pa; char *list, *listend; vm_paddr_t end; vm_paddr_t biggestsize; vm_paddr_t low_water, high_water; int biggestone; biggestsize = 0; biggestone = 0; vaddr = round_page(vaddr); for (i = 0; phys_avail[i + 1]; i += 2) { phys_avail[i] = round_page(phys_avail[i]); phys_avail[i + 1] = trunc_page(phys_avail[i + 1]); } low_water = phys_avail[0]; high_water = phys_avail[1]; for (i = 0; i < vm_phys_nsegs; i++) { if (vm_phys_segs[i].start < low_water) low_water = vm_phys_segs[i].start; if (vm_phys_segs[i].end > high_water) high_water = vm_phys_segs[i].end; } for (i = 0; phys_avail[i + 1]; i += 2) { vm_paddr_t size = phys_avail[i + 1] - phys_avail[i]; if (size > biggestsize) { biggestone = i; biggestsize = size; } if (phys_avail[i] < low_water) low_water = phys_avail[i]; if (phys_avail[i + 1] > high_water) high_water = phys_avail[i + 1]; } end = phys_avail[biggestone+1]; /* * Initialize the page and queue locks. */ mtx_init(&vm_page_queue_free_mtx, "vm page free queue", NULL, MTX_DEF); for (i = 0; i < PA_LOCK_COUNT; i++) mtx_init(&pa_lock[i], "vm page", NULL, MTX_DEF); for (i = 0; i < vm_ndomains; i++) vm_page_domain_init(&vm_dom[i]); /* * Allocate memory for use when boot strapping the kernel memory * allocator. * * CTFLAG_RDTUN doesn't work during the early boot process, so we must * manually fetch the value. */ TUNABLE_INT_FETCH("vm.boot_pages", &boot_pages); new_end = end - (boot_pages * UMA_SLAB_SIZE); new_end = trunc_page(new_end); mapped = pmap_map(&vaddr, new_end, end, VM_PROT_READ | VM_PROT_WRITE); bzero((void *)mapped, end - new_end); uma_startup((void *)mapped, boot_pages); #if defined(__aarch64__) || defined(__amd64__) || defined(__arm__) || \ defined(__i386__) || defined(__mips__) /* * Allocate a bitmap to indicate that a random physical page * needs to be included in a minidump. * * The amd64 port needs this to indicate which direct map pages * need to be dumped, via calls to dump_add_page()/dump_drop_page(). * * However, i386 still needs this workspace internally within the * minidump code. In theory, they are not needed on i386, but are * included should the sf_buf code decide to use them. */ last_pa = 0; for (i = 0; dump_avail[i + 1] != 0; i += 2) if (dump_avail[i + 1] > last_pa) last_pa = dump_avail[i + 1]; page_range = last_pa / PAGE_SIZE; vm_page_dump_size = round_page(roundup2(page_range, NBBY) / NBBY); new_end -= vm_page_dump_size; vm_page_dump = (void *)(uintptr_t)pmap_map(&vaddr, new_end, new_end + vm_page_dump_size, VM_PROT_READ | VM_PROT_WRITE); bzero((void *)vm_page_dump, vm_page_dump_size); #endif #ifdef __amd64__ /* * Request that the physical pages underlying the message buffer be * included in a crash dump. Since the message buffer is accessed * through the direct map, they are not automatically included. */ pa = DMAP_TO_PHYS((vm_offset_t)msgbufp->msg_ptr); last_pa = pa + round_page(msgbufsize); while (pa < last_pa) { dump_add_page(pa); pa += PAGE_SIZE; } #endif /* * Compute the number of pages of memory that will be available for * use (taking into account the overhead of a page structure per * page). */ first_page = low_water / PAGE_SIZE; #ifdef VM_PHYSSEG_SPARSE page_range = 0; for (i = 0; i < vm_phys_nsegs; i++) { page_range += atop(vm_phys_segs[i].end - vm_phys_segs[i].start); } for (i = 0; phys_avail[i + 1] != 0; i += 2) page_range += atop(phys_avail[i + 1] - phys_avail[i]); #elif defined(VM_PHYSSEG_DENSE) page_range = high_water / PAGE_SIZE - first_page; #else #error "Either VM_PHYSSEG_DENSE or VM_PHYSSEG_SPARSE must be defined." #endif end = new_end; /* * Reserve an unmapped guard page to trap access to vm_page_array[-1]. */ vaddr += PAGE_SIZE; /* * Initialize the mem entry structures now, and put them in the free * queue. */ new_end = trunc_page(end - page_range * sizeof(struct vm_page)); mapped = pmap_map(&vaddr, new_end, end, VM_PROT_READ | VM_PROT_WRITE); vm_page_array = (vm_page_t) mapped; #if VM_NRESERVLEVEL > 0 /* * Allocate memory for the reservation management system's data * structures. */ new_end = vm_reserv_startup(&vaddr, new_end, high_water); #endif #if defined(__aarch64__) || defined(__amd64__) || defined(__mips__) /* * pmap_map on arm64, amd64, and mips can come out of the direct-map, * not kvm like i386, so the pages must be tracked for a crashdump to * include this data. This includes the vm_page_array and the early * UMA bootstrap pages. */ for (pa = new_end; pa < phys_avail[biggestone + 1]; pa += PAGE_SIZE) dump_add_page(pa); #endif phys_avail[biggestone + 1] = new_end; /* * Add physical memory segments corresponding to the available * physical pages. */ for (i = 0; phys_avail[i + 1] != 0; i += 2) vm_phys_add_seg(phys_avail[i], phys_avail[i + 1]); /* * Clear all of the page structures */ bzero((caddr_t) vm_page_array, page_range * sizeof(struct vm_page)); for (i = 0; i < page_range; i++) vm_page_array[i].order = VM_NFREEORDER; vm_page_array_size = page_range; /* * Initialize the physical memory allocator. */ vm_phys_init(); /* * Add every available physical page that is not blacklisted to * the free lists. */ vm_cnt.v_page_count = 0; vm_cnt.v_free_count = 0; for (i = 0; phys_avail[i + 1] != 0; i += 2) { pa = phys_avail[i]; last_pa = phys_avail[i + 1]; while (pa < last_pa) { vm_phys_add_page(pa); pa += PAGE_SIZE; } } TAILQ_INIT(&blacklist_head); vm_page_blacklist_load(&list, &listend); vm_page_blacklist_check(list, listend); list = kern_getenv("vm.blacklist"); vm_page_blacklist_check(list, NULL); freeenv(list); #if VM_NRESERVLEVEL > 0 /* * Initialize the reservation management system. */ vm_reserv_init(); #endif return (vaddr); } void vm_page_reference(vm_page_t m) { vm_page_aflag_set(m, PGA_REFERENCED); } /* * vm_page_busy_downgrade: * * Downgrade an exclusive busy page into a single shared busy page. */ void vm_page_busy_downgrade(vm_page_t m) { u_int x; vm_page_assert_xbusied(m); for (;;) { x = m->busy_lock; x &= VPB_BIT_WAITERS; if (atomic_cmpset_rel_int(&m->busy_lock, VPB_SINGLE_EXCLUSIVER | x, VPB_SHARERS_WORD(1) | x)) break; } } /* * vm_page_sbusied: * * Return a positive value if the page is shared busied, 0 otherwise. */ int vm_page_sbusied(vm_page_t m) { u_int x; x = m->busy_lock; return ((x & VPB_BIT_SHARED) != 0 && x != VPB_UNBUSIED); } /* * vm_page_sunbusy: * * Shared unbusy a page. */ void vm_page_sunbusy(vm_page_t m) { u_int x; vm_page_assert_sbusied(m); for (;;) { x = m->busy_lock; if (VPB_SHARERS(x) > 1) { if (atomic_cmpset_int(&m->busy_lock, x, x - VPB_ONE_SHARER)) break; continue; } if ((x & VPB_BIT_WAITERS) == 0) { KASSERT(x == VPB_SHARERS_WORD(1), ("vm_page_sunbusy: invalid lock state")); if (atomic_cmpset_int(&m->busy_lock, VPB_SHARERS_WORD(1), VPB_UNBUSIED)) break; continue; } KASSERT(x == (VPB_SHARERS_WORD(1) | VPB_BIT_WAITERS), ("vm_page_sunbusy: invalid lock state for waiters")); vm_page_lock(m); if (!atomic_cmpset_int(&m->busy_lock, x, VPB_UNBUSIED)) { vm_page_unlock(m); continue; } wakeup(m); vm_page_unlock(m); break; } } /* * vm_page_busy_sleep: * * Sleep and release the page lock, using the page pointer as wchan. * This is used to implement the hard-path of busying mechanism. * * The given page must be locked. */ void vm_page_busy_sleep(vm_page_t m, const char *wmesg) { u_int x; vm_page_lock_assert(m, MA_OWNED); x = m->busy_lock; if (x == VPB_UNBUSIED) { vm_page_unlock(m); return; } if ((x & VPB_BIT_WAITERS) == 0 && !atomic_cmpset_int(&m->busy_lock, x, x | VPB_BIT_WAITERS)) { vm_page_unlock(m); return; } msleep(m, vm_page_lockptr(m), PVM | PDROP, wmesg, 0); } /* * vm_page_trysbusy: * * Try to shared busy a page. * If the operation succeeds 1 is returned otherwise 0. * The operation never sleeps. */ int vm_page_trysbusy(vm_page_t m) { u_int x; for (;;) { x = m->busy_lock; if ((x & VPB_BIT_SHARED) == 0) return (0); if (atomic_cmpset_acq_int(&m->busy_lock, x, x + VPB_ONE_SHARER)) return (1); } } /* * vm_page_xunbusy_hard: * * Called after the first try the exclusive unbusy of a page failed. * It is assumed that the waiters bit is on. */ void vm_page_xunbusy_hard(vm_page_t m) { vm_page_assert_xbusied(m); vm_page_lock(m); atomic_store_rel_int(&m->busy_lock, VPB_UNBUSIED); wakeup(m); vm_page_unlock(m); } /* * vm_page_flash: * * Wakeup anyone waiting for the page. * The ownership bits do not change. * * The given page must be locked. */ void vm_page_flash(vm_page_t m) { u_int x; vm_page_lock_assert(m, MA_OWNED); for (;;) { x = m->busy_lock; if ((x & VPB_BIT_WAITERS) == 0) return; if (atomic_cmpset_int(&m->busy_lock, x, x & (~VPB_BIT_WAITERS))) break; } wakeup(m); } /* * Keep page from being freed by the page daemon * much of the same effect as wiring, except much lower * overhead and should be used only for *very* temporary * holding ("wiring"). */ void vm_page_hold(vm_page_t mem) { vm_page_lock_assert(mem, MA_OWNED); mem->hold_count++; } void vm_page_unhold(vm_page_t mem) { vm_page_lock_assert(mem, MA_OWNED); KASSERT(mem->hold_count >= 1, ("vm_page_unhold: hold count < 0!!!")); --mem->hold_count; if (mem->hold_count == 0 && (mem->flags & PG_UNHOLDFREE) != 0) vm_page_free_toq(mem); } /* * vm_page_unhold_pages: * * Unhold each of the pages that is referenced by the given array. */ void vm_page_unhold_pages(vm_page_t *ma, int count) { struct mtx *mtx, *new_mtx; mtx = NULL; for (; count != 0; count--) { /* * Avoid releasing and reacquiring the same page lock. */ new_mtx = vm_page_lockptr(*ma); if (mtx != new_mtx) { if (mtx != NULL) mtx_unlock(mtx); mtx = new_mtx; mtx_lock(mtx); } vm_page_unhold(*ma); ma++; } if (mtx != NULL) mtx_unlock(mtx); } vm_page_t PHYS_TO_VM_PAGE(vm_paddr_t pa) { vm_page_t m; #ifdef VM_PHYSSEG_SPARSE m = vm_phys_paddr_to_vm_page(pa); if (m == NULL) m = vm_phys_fictitious_to_vm_page(pa); return (m); #elif defined(VM_PHYSSEG_DENSE) long pi; pi = atop(pa); if (pi >= first_page && (pi - first_page) < vm_page_array_size) { m = &vm_page_array[pi - first_page]; return (m); } return (vm_phys_fictitious_to_vm_page(pa)); #else #error "Either VM_PHYSSEG_DENSE or VM_PHYSSEG_SPARSE must be defined." #endif } /* * vm_page_getfake: * * Create a fictitious page with the specified physical address and * memory attribute. The memory attribute is the only the machine- * dependent aspect of a fictitious page that must be initialized. */ vm_page_t vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr) { vm_page_t m; m = uma_zalloc(fakepg_zone, M_WAITOK | M_ZERO); vm_page_initfake(m, paddr, memattr); return (m); } void vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr) { if ((m->flags & PG_FICTITIOUS) != 0) { /* * The page's memattr might have changed since the * previous initialization. Update the pmap to the * new memattr. */ goto memattr; } m->phys_addr = paddr; m->queue = PQ_NONE; /* Fictitious pages don't use "segind". */ m->flags = PG_FICTITIOUS; /* Fictitious pages don't use "order" or "pool". */ m->oflags = VPO_UNMANAGED; m->busy_lock = VPB_SINGLE_EXCLUSIVER; m->wire_count = 1; pmap_page_init(m); memattr: pmap_page_set_memattr(m, memattr); } /* * vm_page_putfake: * * Release a fictitious page. */ void vm_page_putfake(vm_page_t m) { KASSERT((m->oflags & VPO_UNMANAGED) != 0, ("managed %p", m)); KASSERT((m->flags & PG_FICTITIOUS) != 0, ("vm_page_putfake: bad page %p", m)); uma_zfree(fakepg_zone, m); } /* * vm_page_updatefake: * * Update the given fictitious page to the specified physical address and * memory attribute. */ void vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr) { KASSERT((m->flags & PG_FICTITIOUS) != 0, ("vm_page_updatefake: bad page %p", m)); m->phys_addr = paddr; pmap_page_set_memattr(m, memattr); } /* * vm_page_free: * * Free a page. */ void vm_page_free(vm_page_t m) { m->flags &= ~PG_ZERO; vm_page_free_toq(m); } /* * vm_page_free_zero: * * Free a page to the zerod-pages queue */ void vm_page_free_zero(vm_page_t m) { m->flags |= PG_ZERO; vm_page_free_toq(m); } /* * Unbusy and handle the page queueing for a page from the VOP_GETPAGES() * array which is not the request page. */ void vm_page_readahead_finish(vm_page_t m) { if (m->valid != 0) { /* * Since the page is not the requested page, whether * it should be activated or deactivated is not * obvious. Empirical results have shown that * deactivating the page is usually the best choice, * unless the page is wanted by another thread. */ vm_page_lock(m); if ((m->busy_lock & VPB_BIT_WAITERS) != 0) vm_page_activate(m); else vm_page_deactivate(m); vm_page_unlock(m); vm_page_xunbusy(m); } else { /* * Free the completely invalid page. Such page state * occurs due to the short read operation which did * not covered our page at all, or in case when a read * error happens. */ vm_page_lock(m); vm_page_free(m); vm_page_unlock(m); } } /* * vm_page_sleep_if_busy: * * Sleep and release the page queues lock if the page is busied. * Returns TRUE if the thread slept. * * The given page must be unlocked and object containing it must * be locked. */ int vm_page_sleep_if_busy(vm_page_t m, const char *msg) { vm_object_t obj; vm_page_lock_assert(m, MA_NOTOWNED); VM_OBJECT_ASSERT_WLOCKED(m->object); if (vm_page_busied(m)) { /* * The page-specific object must be cached because page * identity can change during the sleep, causing the * re-lock of a different object. * It is assumed that a reference to the object is already * held by the callers. */ obj = m->object; vm_page_lock(m); VM_OBJECT_WUNLOCK(obj); vm_page_busy_sleep(m, msg); VM_OBJECT_WLOCK(obj); return (TRUE); } return (FALSE); } /* * vm_page_dirty_KBI: [ internal use only ] * * Set all bits in the page's dirty field. * * The object containing the specified page must be locked if the * call is made from the machine-independent layer. * * See vm_page_clear_dirty_mask(). * * This function should only be called by vm_page_dirty(). */ void vm_page_dirty_KBI(vm_page_t m) { /* These assertions refer to this operation by its public name. */ KASSERT((m->flags & PG_CACHED) == 0, ("vm_page_dirty: page in cache!")); KASSERT(m->valid == VM_PAGE_BITS_ALL, ("vm_page_dirty: page is invalid!")); m->dirty = VM_PAGE_BITS_ALL; } /* * vm_page_insert: [ internal use only ] * * Inserts the given mem entry into the object and object list. * * The object must be locked. */ int vm_page_insert(vm_page_t m, vm_object_t object, vm_pindex_t pindex) { vm_page_t mpred; VM_OBJECT_ASSERT_WLOCKED(object); mpred = vm_radix_lookup_le(&object->rtree, pindex); return (vm_page_insert_after(m, object, pindex, mpred)); } /* * vm_page_insert_after: * * Inserts the page "m" into the specified object at offset "pindex". * * The page "mpred" must immediately precede the offset "pindex" within * the specified object. * * The object must be locked. */ static int vm_page_insert_after(vm_page_t m, vm_object_t object, vm_pindex_t pindex, vm_page_t mpred) { vm_pindex_t sidx; vm_object_t sobj; vm_page_t msucc; VM_OBJECT_ASSERT_WLOCKED(object); KASSERT(m->object == NULL, ("vm_page_insert_after: page already inserted")); if (mpred != NULL) { KASSERT(mpred->object == object, ("vm_page_insert_after: object doesn't contain mpred")); KASSERT(mpred->pindex < pindex, ("vm_page_insert_after: mpred doesn't precede pindex")); msucc = TAILQ_NEXT(mpred, listq); } else msucc = TAILQ_FIRST(&object->memq); if (msucc != NULL) KASSERT(msucc->pindex > pindex, ("vm_page_insert_after: msucc doesn't succeed pindex")); /* * Record the object/offset pair in this page */ sobj = m->object; sidx = m->pindex; m->object = object; m->pindex = pindex; /* * Now link into the object's ordered list of backed pages. */ if (vm_radix_insert(&object->rtree, m)) { m->object = sobj; m->pindex = sidx; return (1); } vm_page_insert_radixdone(m, object, mpred); return (0); } /* * vm_page_insert_radixdone: * * Complete page "m" insertion into the specified object after the * radix trie hooking. * * The page "mpred" must precede the offset "m->pindex" within the * specified object. * * The object must be locked. */ static void vm_page_insert_radixdone(vm_page_t m, vm_object_t object, vm_page_t mpred) { VM_OBJECT_ASSERT_WLOCKED(object); KASSERT(object != NULL && m->object == object, ("vm_page_insert_radixdone: page %p has inconsistent object", m)); if (mpred != NULL) { KASSERT(mpred->object == object, ("vm_page_insert_after: object doesn't contain mpred")); KASSERT(mpred->pindex < m->pindex, ("vm_page_insert_after: mpred doesn't precede pindex")); } if (mpred != NULL) TAILQ_INSERT_AFTER(&object->memq, mpred, m, listq); else TAILQ_INSERT_HEAD(&object->memq, m, listq); /* * Show that the object has one more resident page. */ object->resident_page_count++; /* * Hold the vnode until the last page is released. */ if (object->resident_page_count == 1 && object->type == OBJT_VNODE) vhold(object->handle); /* * Since we are inserting a new and possibly dirty page, * update the object's OBJ_MIGHTBEDIRTY flag. */ if (pmap_page_is_write_mapped(m)) vm_object_set_writeable_dirty(object); } /* * vm_page_remove: * * Removes the given mem entry from the object/offset-page * table and the object page list, but do not invalidate/terminate * the backing store. * * The object must be locked. The page must be locked if it is managed. */ void vm_page_remove(vm_page_t m) { vm_object_t object; boolean_t lockacq; if ((m->oflags & VPO_UNMANAGED) == 0) vm_page_lock_assert(m, MA_OWNED); if ((object = m->object) == NULL) return; VM_OBJECT_ASSERT_WLOCKED(object); if (vm_page_xbusied(m)) { lockacq = FALSE; if ((m->oflags & VPO_UNMANAGED) != 0 && !mtx_owned(vm_page_lockptr(m))) { lockacq = TRUE; vm_page_lock(m); } vm_page_flash(m); atomic_store_rel_int(&m->busy_lock, VPB_UNBUSIED); if (lockacq) vm_page_unlock(m); } /* * Now remove from the object's list of backed pages. */ vm_radix_remove(&object->rtree, m->pindex); TAILQ_REMOVE(&object->memq, m, listq); /* * And show that the object has one fewer resident page. */ object->resident_page_count--; /* * The vnode may now be recycled. */ if (object->resident_page_count == 0 && object->type == OBJT_VNODE) vdrop(object->handle); m->object = NULL; } /* * vm_page_lookup: * * Returns the page associated with the object/offset * pair specified; if none is found, NULL is returned. * * The object must be locked. */ vm_page_t vm_page_lookup(vm_object_t object, vm_pindex_t pindex) { VM_OBJECT_ASSERT_LOCKED(object); return (vm_radix_lookup(&object->rtree, pindex)); } /* * vm_page_find_least: * * Returns the page associated with the object with least pindex * greater than or equal to the parameter pindex, or NULL. * * The object must be locked. */ vm_page_t vm_page_find_least(vm_object_t object, vm_pindex_t pindex) { vm_page_t m; VM_OBJECT_ASSERT_LOCKED(object); if ((m = TAILQ_FIRST(&object->memq)) != NULL && m->pindex < pindex) m = vm_radix_lookup_ge(&object->rtree, pindex); return (m); } /* * Returns the given page's successor (by pindex) within the object if it is * resident; if none is found, NULL is returned. * * The object must be locked. */ vm_page_t vm_page_next(vm_page_t m) { vm_page_t next; VM_OBJECT_ASSERT_WLOCKED(m->object); if ((next = TAILQ_NEXT(m, listq)) != NULL && next->pindex != m->pindex + 1) next = NULL; return (next); } /* * Returns the given page's predecessor (by pindex) within the object if it is * resident; if none is found, NULL is returned. * * The object must be locked. */ vm_page_t vm_page_prev(vm_page_t m) { vm_page_t prev; VM_OBJECT_ASSERT_WLOCKED(m->object); if ((prev = TAILQ_PREV(m, pglist, listq)) != NULL && prev->pindex != m->pindex - 1) prev = NULL; return (prev); } /* * Uses the page mnew as a replacement for an existing page at index * pindex which must be already present in the object. * * The existing page must not be on a paging queue. */ vm_page_t vm_page_replace(vm_page_t mnew, vm_object_t object, vm_pindex_t pindex) { vm_page_t mold, mpred; VM_OBJECT_ASSERT_WLOCKED(object); /* * This function mostly follows vm_page_insert() and * vm_page_remove() without the radix, object count and vnode * dance. Double check such functions for more comments. */ mpred = vm_radix_lookup(&object->rtree, pindex); KASSERT(mpred != NULL, ("vm_page_replace: replacing page not present with pindex")); mpred = TAILQ_PREV(mpred, respgs, listq); if (mpred != NULL) KASSERT(mpred->pindex < pindex, ("vm_page_insert_after: mpred doesn't precede pindex")); mnew->object = object; mnew->pindex = pindex; mold = vm_radix_replace(&object->rtree, mnew); KASSERT(mold->queue == PQ_NONE, ("vm_page_replace: mold is on a paging queue")); /* Detach the old page from the resident tailq. */ TAILQ_REMOVE(&object->memq, mold, listq); mold->object = NULL; vm_page_xunbusy(mold); /* Insert the new page in the resident tailq. */ if (mpred != NULL) TAILQ_INSERT_AFTER(&object->memq, mpred, mnew, listq); else TAILQ_INSERT_HEAD(&object->memq, mnew, listq); if (pmap_page_is_write_mapped(mnew)) vm_object_set_writeable_dirty(object); return (mold); } /* * vm_page_rename: * * Move the given memory entry from its * current object to the specified target object/offset. * * Note: swap associated with the page must be invalidated by the move. We * have to do this for several reasons: (1) we aren't freeing the * page, (2) we are dirtying the page, (3) the VM system is probably * moving the page from object A to B, and will then later move * the backing store from A to B and we can't have a conflict. * * Note: we *always* dirty the page. It is necessary both for the * fact that we moved it, and because we may be invalidating * swap. If the page is on the cache, we have to deactivate it * or vm_page_dirty() will panic. Dirty pages are not allowed * on the cache. * * The objects must be locked. */ int vm_page_rename(vm_page_t m, vm_object_t new_object, vm_pindex_t new_pindex) { vm_page_t mpred; vm_pindex_t opidx; VM_OBJECT_ASSERT_WLOCKED(new_object); mpred = vm_radix_lookup_le(&new_object->rtree, new_pindex); KASSERT(mpred == NULL || mpred->pindex != new_pindex, ("vm_page_rename: pindex already renamed")); /* * Create a custom version of vm_page_insert() which does not depend * by m_prev and can cheat on the implementation aspects of the * function. */ opidx = m->pindex; m->pindex = new_pindex; if (vm_radix_insert(&new_object->rtree, m)) { m->pindex = opidx; return (1); } /* * The operation cannot fail anymore. The removal must happen before * the listq iterator is tainted. */ m->pindex = opidx; vm_page_lock(m); vm_page_remove(m); /* Return back to the new pindex to complete vm_page_insert(). */ m->pindex = new_pindex; m->object = new_object; vm_page_unlock(m); vm_page_insert_radixdone(m, new_object, mpred); vm_page_dirty(m); return (0); } /* * Convert all of the given object's cached pages that have a * pindex within the given range into free pages. If the value * zero is given for "end", then the range's upper bound is * infinity. If the given object is backed by a vnode and it * transitions from having one or more cached pages to none, the * vnode's hold count is reduced. */ void vm_page_cache_free(vm_object_t object, vm_pindex_t start, vm_pindex_t end) { vm_page_t m; boolean_t empty; mtx_lock(&vm_page_queue_free_mtx); if (__predict_false(vm_radix_is_empty(&object->cache))) { mtx_unlock(&vm_page_queue_free_mtx); return; } while ((m = vm_radix_lookup_ge(&object->cache, start)) != NULL) { if (end != 0 && m->pindex >= end) break; vm_radix_remove(&object->cache, m->pindex); vm_page_cache_turn_free(m); } empty = vm_radix_is_empty(&object->cache); mtx_unlock(&vm_page_queue_free_mtx); if (object->type == OBJT_VNODE && empty) vdrop(object->handle); } /* * Returns the cached page that is associated with the given * object and offset. If, however, none exists, returns NULL. * * The free page queue must be locked. */ static inline vm_page_t vm_page_cache_lookup(vm_object_t object, vm_pindex_t pindex) { mtx_assert(&vm_page_queue_free_mtx, MA_OWNED); return (vm_radix_lookup(&object->cache, pindex)); } /* * Remove the given cached page from its containing object's * collection of cached pages. * * The free page queue must be locked. */ static void vm_page_cache_remove(vm_page_t m) { mtx_assert(&vm_page_queue_free_mtx, MA_OWNED); KASSERT((m->flags & PG_CACHED) != 0, ("vm_page_cache_remove: page %p is not cached", m)); vm_radix_remove(&m->object->cache, m->pindex); m->object = NULL; vm_cnt.v_cache_count--; } /* * Transfer all of the cached pages with offset greater than or * equal to 'offidxstart' from the original object's cache to the * new object's cache. However, any cached pages with offset * greater than or equal to the new object's size are kept in the * original object. Initially, the new object's cache must be * empty. Offset 'offidxstart' in the original object must * correspond to offset zero in the new object. * * The new object must be locked. */ void vm_page_cache_transfer(vm_object_t orig_object, vm_pindex_t offidxstart, vm_object_t new_object) { vm_page_t m; /* * Insertion into an object's collection of cached pages * requires the object to be locked. In contrast, removal does * not. */ VM_OBJECT_ASSERT_WLOCKED(new_object); KASSERT(vm_radix_is_empty(&new_object->cache), ("vm_page_cache_transfer: object %p has cached pages", new_object)); mtx_lock(&vm_page_queue_free_mtx); while ((m = vm_radix_lookup_ge(&orig_object->cache, offidxstart)) != NULL) { /* * Transfer all of the pages with offset greater than or * equal to 'offidxstart' from the original object's * cache to the new object's cache. */ if ((m->pindex - offidxstart) >= new_object->size) break; vm_radix_remove(&orig_object->cache, m->pindex); /* Update the page's object and offset. */ m->object = new_object; m->pindex -= offidxstart; if (vm_radix_insert(&new_object->cache, m)) vm_page_cache_turn_free(m); } mtx_unlock(&vm_page_queue_free_mtx); } /* * Returns TRUE if a cached page is associated with the given object and * offset, and FALSE otherwise. * * The object must be locked. */ boolean_t vm_page_is_cached(vm_object_t object, vm_pindex_t pindex) { vm_page_t m; /* * Insertion into an object's collection of cached pages requires the * object to be locked. Therefore, if the object is locked and the * object's collection is empty, there is no need to acquire the free * page queues lock in order to prove that the specified page doesn't * exist. */ VM_OBJECT_ASSERT_WLOCKED(object); if (__predict_true(vm_object_cache_is_empty(object))) return (FALSE); mtx_lock(&vm_page_queue_free_mtx); m = vm_page_cache_lookup(object, pindex); mtx_unlock(&vm_page_queue_free_mtx); return (m != NULL); } /* * vm_page_alloc: * * Allocate and return a page that is associated with the specified * object and offset pair. By default, this page is exclusive busied. * * The caller must always specify an allocation class. * * allocation classes: * VM_ALLOC_NORMAL normal process request * VM_ALLOC_SYSTEM system *really* needs a page * VM_ALLOC_INTERRUPT interrupt time request * * optional allocation flags: * VM_ALLOC_COUNT(number) the number of additional pages that the caller * intends to allocate * VM_ALLOC_IFCACHED return page only if it is cached * VM_ALLOC_IFNOTCACHED return NULL, do not reactivate if the page * is cached * VM_ALLOC_NOBUSY do not exclusive busy the page * VM_ALLOC_NODUMP do not include the page in a kernel core dump * VM_ALLOC_NOOBJ page is not associated with an object and * should not be exclusive busy * VM_ALLOC_SBUSY shared busy the allocated page * VM_ALLOC_WIRED wire the allocated page * VM_ALLOC_ZERO prefer a zeroed page * * This routine may not sleep. */ vm_page_t vm_page_alloc(vm_object_t object, vm_pindex_t pindex, int req) { struct vnode *vp = NULL; vm_object_t m_object; vm_page_t m, mpred; int flags, req_class; mpred = 0; /* XXX: pacify gcc */ KASSERT((object != NULL) == ((req & VM_ALLOC_NOOBJ) == 0) && (object != NULL || (req & VM_ALLOC_SBUSY) == 0) && ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) != (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)), ("vm_page_alloc: inconsistent object(%p)/req(%x)", (void *)object, req)); if (object != NULL) VM_OBJECT_ASSERT_WLOCKED(object); req_class = req & VM_ALLOC_CLASS_MASK; /* * The page daemon is allowed to dig deeper into the free page list. */ if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT) req_class = VM_ALLOC_SYSTEM; if (object != NULL) { mpred = vm_radix_lookup_le(&object->rtree, pindex); KASSERT(mpred == NULL || mpred->pindex != pindex, ("vm_page_alloc: pindex already allocated")); } /* * The page allocation request can came from consumers which already * hold the free page queue mutex, like vm_page_insert() in * vm_page_cache(). */ mtx_lock_flags(&vm_page_queue_free_mtx, MTX_RECURSE); if (vm_cnt.v_free_count + vm_cnt.v_cache_count > vm_cnt.v_free_reserved || (req_class == VM_ALLOC_SYSTEM && vm_cnt.v_free_count + vm_cnt.v_cache_count > vm_cnt.v_interrupt_free_min) || (req_class == VM_ALLOC_INTERRUPT && vm_cnt.v_free_count + vm_cnt.v_cache_count > 0)) { /* * Allocate from the free queue if the number of free pages * exceeds the minimum for the request class. */ if (object != NULL && (m = vm_page_cache_lookup(object, pindex)) != NULL) { if ((req & VM_ALLOC_IFNOTCACHED) != 0) { mtx_unlock(&vm_page_queue_free_mtx); return (NULL); } if (vm_phys_unfree_page(m)) vm_phys_set_pool(VM_FREEPOOL_DEFAULT, m, 0); #if VM_NRESERVLEVEL > 0 else if (!vm_reserv_reactivate_page(m)) #else else #endif panic("vm_page_alloc: cache page %p is missing" " from the free queue", m); } else if ((req & VM_ALLOC_IFCACHED) != 0) { mtx_unlock(&vm_page_queue_free_mtx); return (NULL); #if VM_NRESERVLEVEL > 0 } else if (object == NULL || (object->flags & (OBJ_COLORED | OBJ_FICTITIOUS)) != OBJ_COLORED || (m = vm_reserv_alloc_page(object, pindex, mpred)) == NULL) { #else } else { #endif m = vm_phys_alloc_pages(object != NULL ? VM_FREEPOOL_DEFAULT : VM_FREEPOOL_DIRECT, 0); #if VM_NRESERVLEVEL > 0 if (m == NULL && vm_reserv_reclaim_inactive()) { m = vm_phys_alloc_pages(object != NULL ? VM_FREEPOOL_DEFAULT : VM_FREEPOOL_DIRECT, 0); } #endif } } else { /* * Not allocatable, give up. */ mtx_unlock(&vm_page_queue_free_mtx); atomic_add_int(&vm_pageout_deficit, max((u_int)req >> VM_ALLOC_COUNT_SHIFT, 1)); pagedaemon_wakeup(); return (NULL); } /* * At this point we had better have found a good page. */ KASSERT(m != NULL, ("vm_page_alloc: missing page")); KASSERT(m->queue == PQ_NONE, ("vm_page_alloc: page %p has unexpected queue %d", m, m->queue)); KASSERT(m->wire_count == 0, ("vm_page_alloc: page %p is wired", m)); KASSERT(m->hold_count == 0, ("vm_page_alloc: page %p is held", m)); KASSERT(!vm_page_sbusied(m), ("vm_page_alloc: page %p is busy", m)); KASSERT(m->dirty == 0, ("vm_page_alloc: page %p is dirty", m)); KASSERT(pmap_page_get_memattr(m) == VM_MEMATTR_DEFAULT, ("vm_page_alloc: page %p has unexpected memattr %d", m, pmap_page_get_memattr(m))); if ((m->flags & PG_CACHED) != 0) { KASSERT((m->flags & PG_ZERO) == 0, ("vm_page_alloc: cached page %p is PG_ZERO", m)); KASSERT(m->valid != 0, ("vm_page_alloc: cached page %p is invalid", m)); if (m->object == object && m->pindex == pindex) vm_cnt.v_reactivated++; else m->valid = 0; m_object = m->object; vm_page_cache_remove(m); if (m_object->type == OBJT_VNODE && vm_object_cache_is_empty(m_object)) vp = m_object->handle; } else { KASSERT(m->valid == 0, ("vm_page_alloc: free page %p is valid", m)); vm_phys_freecnt_adj(m, -1); if ((m->flags & PG_ZERO) != 0) vm_page_zero_count--; } mtx_unlock(&vm_page_queue_free_mtx); /* * Initialize the page. Only the PG_ZERO flag is inherited. */ flags = 0; if ((req & VM_ALLOC_ZERO) != 0) flags = PG_ZERO; flags &= m->flags; if ((req & VM_ALLOC_NODUMP) != 0) flags |= PG_NODUMP; m->flags = flags; m->aflags = 0; m->oflags = object == NULL || (object->flags & OBJ_UNMANAGED) != 0 ? VPO_UNMANAGED : 0; m->busy_lock = VPB_UNBUSIED; if ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_NOOBJ | VM_ALLOC_SBUSY)) == 0) m->busy_lock = VPB_SINGLE_EXCLUSIVER; if ((req & VM_ALLOC_SBUSY) != 0) m->busy_lock = VPB_SHARERS_WORD(1); if (req & VM_ALLOC_WIRED) { /* * The page lock is not required for wiring a page until that * page is inserted into the object. */ atomic_add_int(&vm_cnt.v_wire_count, 1); m->wire_count = 1; } m->act_count = 0; if (object != NULL) { if (vm_page_insert_after(m, object, pindex, mpred)) { /* See the comment below about hold count. */ if (vp != NULL) vdrop(vp); pagedaemon_wakeup(); if (req & VM_ALLOC_WIRED) { atomic_subtract_int(&vm_cnt.v_wire_count, 1); m->wire_count = 0; } m->object = NULL; m->oflags = VPO_UNMANAGED; vm_page_free(m); return (NULL); } /* Ignore device objects; the pager sets "memattr" for them. */ if (object->memattr != VM_MEMATTR_DEFAULT && (object->flags & OBJ_FICTITIOUS) == 0) pmap_page_set_memattr(m, object->memattr); } else m->pindex = pindex; /* * The following call to vdrop() must come after the above call * to vm_page_insert() in case both affect the same object and * vnode. Otherwise, the affected vnode's hold count could * temporarily become zero. */ if (vp != NULL) vdrop(vp); /* * Don't wakeup too often - wakeup the pageout daemon when * we would be nearly out of memory. */ if (vm_paging_needed()) pagedaemon_wakeup(); return (m); } static void vm_page_alloc_contig_vdrop(struct spglist *lst) { while (!SLIST_EMPTY(lst)) { vdrop((struct vnode *)SLIST_FIRST(lst)-> plinks.s.pv); SLIST_REMOVE_HEAD(lst, plinks.s.ss); } } /* * vm_page_alloc_contig: * * Allocate a contiguous set of physical pages of the given size "npages" * from the free lists. All of the physical pages must be at or above * the given physical address "low" and below the given physical address * "high". The given value "alignment" determines the alignment of the * first physical page in the set. If the given value "boundary" is * non-zero, then the set of physical pages cannot cross any physical * address boundary that is a multiple of that value. Both "alignment" * and "boundary" must be a power of two. * * If the specified memory attribute, "memattr", is VM_MEMATTR_DEFAULT, * then the memory attribute setting for the physical pages is configured * to the object's memory attribute setting. Otherwise, the memory * attribute setting for the physical pages is configured to "memattr", * overriding the object's memory attribute setting. However, if the * object's memory attribute setting is not VM_MEMATTR_DEFAULT, then the * memory attribute setting for the physical pages cannot be configured * to VM_MEMATTR_DEFAULT. * * The caller must always specify an allocation class. * * allocation classes: * VM_ALLOC_NORMAL normal process request * VM_ALLOC_SYSTEM system *really* needs a page * VM_ALLOC_INTERRUPT interrupt time request * * optional allocation flags: * VM_ALLOC_NOBUSY do not exclusive busy the page * VM_ALLOC_NODUMP do not include the page in a kernel core dump * VM_ALLOC_NOOBJ page is not associated with an object and * should not be exclusive busy * VM_ALLOC_SBUSY shared busy the allocated page * VM_ALLOC_WIRED wire the allocated page * VM_ALLOC_ZERO prefer a zeroed page * * This routine may not sleep. */ vm_page_t vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req, u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary, vm_memattr_t memattr) { struct vnode *drop; struct spglist deferred_vdrop_list; vm_page_t m, m_tmp, m_ret; u_int flags; int req_class; KASSERT((object != NULL) == ((req & VM_ALLOC_NOOBJ) == 0) && (object != NULL || (req & VM_ALLOC_SBUSY) == 0) && ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) != (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)), ("vm_page_alloc: inconsistent object(%p)/req(%x)", (void *)object, req)); if (object != NULL) { VM_OBJECT_ASSERT_WLOCKED(object); KASSERT(object->type == OBJT_PHYS, ("vm_page_alloc_contig: object %p isn't OBJT_PHYS", object)); } KASSERT(npages > 0, ("vm_page_alloc_contig: npages is zero")); req_class = req & VM_ALLOC_CLASS_MASK; /* * The page daemon is allowed to dig deeper into the free page list. */ if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT) req_class = VM_ALLOC_SYSTEM; SLIST_INIT(&deferred_vdrop_list); mtx_lock(&vm_page_queue_free_mtx); if (vm_cnt.v_free_count + vm_cnt.v_cache_count >= npages + vm_cnt.v_free_reserved || (req_class == VM_ALLOC_SYSTEM && vm_cnt.v_free_count + vm_cnt.v_cache_count >= npages + vm_cnt.v_interrupt_free_min) || (req_class == VM_ALLOC_INTERRUPT && vm_cnt.v_free_count + vm_cnt.v_cache_count >= npages)) { #if VM_NRESERVLEVEL > 0 retry: if (object == NULL || (object->flags & OBJ_COLORED) == 0 || (m_ret = vm_reserv_alloc_contig(object, pindex, npages, low, high, alignment, boundary)) == NULL) #endif m_ret = vm_phys_alloc_contig(npages, low, high, alignment, boundary); } else { mtx_unlock(&vm_page_queue_free_mtx); atomic_add_int(&vm_pageout_deficit, npages); pagedaemon_wakeup(); return (NULL); } if (m_ret != NULL) for (m = m_ret; m < &m_ret[npages]; m++) { drop = vm_page_alloc_init(m); if (drop != NULL) { /* * Enqueue the vnode for deferred vdrop(). */ m->plinks.s.pv = drop; SLIST_INSERT_HEAD(&deferred_vdrop_list, m, plinks.s.ss); } } else { #if VM_NRESERVLEVEL > 0 if (vm_reserv_reclaim_contig(npages, low, high, alignment, boundary)) goto retry; #endif } mtx_unlock(&vm_page_queue_free_mtx); if (m_ret == NULL) return (NULL); /* * Initialize the pages. Only the PG_ZERO flag is inherited. */ flags = 0; if ((req & VM_ALLOC_ZERO) != 0) flags = PG_ZERO; if ((req & VM_ALLOC_NODUMP) != 0) flags |= PG_NODUMP; if ((req & VM_ALLOC_WIRED) != 0) atomic_add_int(&vm_cnt.v_wire_count, npages); if (object != NULL) { if (object->memattr != VM_MEMATTR_DEFAULT && memattr == VM_MEMATTR_DEFAULT) memattr = object->memattr; } for (m = m_ret; m < &m_ret[npages]; m++) { m->aflags = 0; m->flags = (m->flags | PG_NODUMP) & flags; m->busy_lock = VPB_UNBUSIED; if (object != NULL) { if ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) == 0) m->busy_lock = VPB_SINGLE_EXCLUSIVER; if ((req & VM_ALLOC_SBUSY) != 0) m->busy_lock = VPB_SHARERS_WORD(1); } if ((req & VM_ALLOC_WIRED) != 0) m->wire_count = 1; /* Unmanaged pages don't use "act_count". */ m->oflags = VPO_UNMANAGED; if (object != NULL) { if (vm_page_insert(m, object, pindex)) { vm_page_alloc_contig_vdrop( &deferred_vdrop_list); if (vm_paging_needed()) pagedaemon_wakeup(); if ((req & VM_ALLOC_WIRED) != 0) atomic_subtract_int(&vm_cnt.v_wire_count, npages); for (m_tmp = m, m = m_ret; m < &m_ret[npages]; m++) { if ((req & VM_ALLOC_WIRED) != 0) m->wire_count = 0; if (m >= m_tmp) m->object = NULL; vm_page_free(m); } return (NULL); } } else m->pindex = pindex; if (memattr != VM_MEMATTR_DEFAULT) pmap_page_set_memattr(m, memattr); pindex++; } vm_page_alloc_contig_vdrop(&deferred_vdrop_list); if (vm_paging_needed()) pagedaemon_wakeup(); return (m_ret); } /* * Initialize a page that has been freshly dequeued from a freelist. * The caller has to drop the vnode returned, if it is not NULL. * * This function may only be used to initialize unmanaged pages. * * To be called with vm_page_queue_free_mtx held. */ static struct vnode * vm_page_alloc_init(vm_page_t m) { struct vnode *drop; vm_object_t m_object; KASSERT(m->queue == PQ_NONE, ("vm_page_alloc_init: page %p has unexpected queue %d", m, m->queue)); KASSERT(m->wire_count == 0, ("vm_page_alloc_init: page %p is wired", m)); KASSERT(m->hold_count == 0, ("vm_page_alloc_init: page %p is held", m)); KASSERT(!vm_page_sbusied(m), ("vm_page_alloc_init: page %p is busy", m)); KASSERT(m->dirty == 0, ("vm_page_alloc_init: page %p is dirty", m)); KASSERT(pmap_page_get_memattr(m) == VM_MEMATTR_DEFAULT, ("vm_page_alloc_init: page %p has unexpected memattr %d", m, pmap_page_get_memattr(m))); mtx_assert(&vm_page_queue_free_mtx, MA_OWNED); drop = NULL; if ((m->flags & PG_CACHED) != 0) { KASSERT((m->flags & PG_ZERO) == 0, ("vm_page_alloc_init: cached page %p is PG_ZERO", m)); m->valid = 0; m_object = m->object; vm_page_cache_remove(m); if (m_object->type == OBJT_VNODE && vm_object_cache_is_empty(m_object)) drop = m_object->handle; } else { KASSERT(m->valid == 0, ("vm_page_alloc_init: free page %p is valid", m)); vm_phys_freecnt_adj(m, -1); if ((m->flags & PG_ZERO) != 0) vm_page_zero_count--; } return (drop); } /* * vm_page_alloc_freelist: * * Allocate a physical page from the specified free page list. * * The caller must always specify an allocation class. * * allocation classes: * VM_ALLOC_NORMAL normal process request * VM_ALLOC_SYSTEM system *really* needs a page * VM_ALLOC_INTERRUPT interrupt time request * * optional allocation flags: * VM_ALLOC_COUNT(number) the number of additional pages that the caller * intends to allocate * VM_ALLOC_WIRED wire the allocated page * VM_ALLOC_ZERO prefer a zeroed page * * This routine may not sleep. */ vm_page_t vm_page_alloc_freelist(int flind, int req) { struct vnode *drop; vm_page_t m; u_int flags; int req_class; req_class = req & VM_ALLOC_CLASS_MASK; /* * The page daemon is allowed to dig deeper into the free page list. */ if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT) req_class = VM_ALLOC_SYSTEM; /* * Do not allocate reserved pages unless the req has asked for it. */ mtx_lock_flags(&vm_page_queue_free_mtx, MTX_RECURSE); if (vm_cnt.v_free_count + vm_cnt.v_cache_count > vm_cnt.v_free_reserved || (req_class == VM_ALLOC_SYSTEM && vm_cnt.v_free_count + vm_cnt.v_cache_count > vm_cnt.v_interrupt_free_min) || (req_class == VM_ALLOC_INTERRUPT && vm_cnt.v_free_count + vm_cnt.v_cache_count > 0)) m = vm_phys_alloc_freelist_pages(flind, VM_FREEPOOL_DIRECT, 0); else { mtx_unlock(&vm_page_queue_free_mtx); atomic_add_int(&vm_pageout_deficit, max((u_int)req >> VM_ALLOC_COUNT_SHIFT, 1)); pagedaemon_wakeup(); return (NULL); } if (m == NULL) { mtx_unlock(&vm_page_queue_free_mtx); return (NULL); } drop = vm_page_alloc_init(m); mtx_unlock(&vm_page_queue_free_mtx); /* * Initialize the page. Only the PG_ZERO flag is inherited. */ m->aflags = 0; flags = 0; if ((req & VM_ALLOC_ZERO) != 0) flags = PG_ZERO; m->flags &= flags; if ((req & VM_ALLOC_WIRED) != 0) { /* * The page lock is not required for wiring a page that does * not belong to an object. */ atomic_add_int(&vm_cnt.v_wire_count, 1); m->wire_count = 1; } /* Unmanaged pages don't use "act_count". */ m->oflags = VPO_UNMANAGED; if (drop != NULL) vdrop(drop); if (vm_paging_needed()) pagedaemon_wakeup(); return (m); } /* * vm_wait: (also see VM_WAIT macro) * * Sleep until free pages are available for allocation. * - Called in various places before memory allocations. */ void vm_wait(void) { mtx_lock(&vm_page_queue_free_mtx); if (curproc == pageproc) { vm_pageout_pages_needed = 1; msleep(&vm_pageout_pages_needed, &vm_page_queue_free_mtx, PDROP | PSWP, "VMWait", 0); } else { if (!vm_pages_needed) { vm_pages_needed = 1; wakeup(&vm_pages_needed); } msleep(&vm_cnt.v_free_count, &vm_page_queue_free_mtx, PDROP | PVM, "vmwait", 0); } } /* * vm_waitpfault: (also see VM_WAITPFAULT macro) * * Sleep until free pages are available for allocation. * - Called only in vm_fault so that processes page faulting * can be easily tracked. * - Sleeps at a lower priority than vm_wait() so that vm_wait()ing * processes will be able to grab memory first. Do not change * this balance without careful testing first. */ void vm_waitpfault(void) { mtx_lock(&vm_page_queue_free_mtx); if (!vm_pages_needed) { vm_pages_needed = 1; wakeup(&vm_pages_needed); } msleep(&vm_cnt.v_free_count, &vm_page_queue_free_mtx, PDROP | PUSER, "pfault", 0); } struct vm_pagequeue * vm_page_pagequeue(vm_page_t m) { return (&vm_phys_domain(m)->vmd_pagequeues[m->queue]); } /* * vm_page_dequeue: * * Remove the given page from its current page queue. * * The page must be locked. */ void vm_page_dequeue(vm_page_t m) { struct vm_pagequeue *pq; vm_page_assert_locked(m); KASSERT(m->queue < PQ_COUNT, ("vm_page_dequeue: page %p is not queued", m)); pq = vm_page_pagequeue(m); vm_pagequeue_lock(pq); m->queue = PQ_NONE; TAILQ_REMOVE(&pq->pq_pl, m, plinks.q); vm_pagequeue_cnt_dec(pq); vm_pagequeue_unlock(pq); } /* * vm_page_dequeue_locked: * * Remove the given page from its current page queue. * * The page and page queue must be locked. */ void vm_page_dequeue_locked(vm_page_t m) { struct vm_pagequeue *pq; vm_page_lock_assert(m, MA_OWNED); pq = vm_page_pagequeue(m); vm_pagequeue_assert_locked(pq); m->queue = PQ_NONE; TAILQ_REMOVE(&pq->pq_pl, m, plinks.q); vm_pagequeue_cnt_dec(pq); } /* * vm_page_enqueue: * * Add the given page to the specified page queue. * * The page must be locked. */ static void vm_page_enqueue(uint8_t queue, vm_page_t m) { struct vm_pagequeue *pq; vm_page_lock_assert(m, MA_OWNED); KASSERT(queue < PQ_COUNT, ("vm_page_enqueue: invalid queue %u request for page %p", queue, m)); pq = &vm_phys_domain(m)->vmd_pagequeues[queue]; vm_pagequeue_lock(pq); m->queue = queue; TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q); vm_pagequeue_cnt_inc(pq); vm_pagequeue_unlock(pq); } /* * vm_page_requeue: * * Move the given page to the tail of its current page queue. * * The page must be locked. */ void vm_page_requeue(vm_page_t m) { struct vm_pagequeue *pq; vm_page_lock_assert(m, MA_OWNED); KASSERT(m->queue != PQ_NONE, ("vm_page_requeue: page %p is not queued", m)); pq = vm_page_pagequeue(m); vm_pagequeue_lock(pq); TAILQ_REMOVE(&pq->pq_pl, m, plinks.q); TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q); vm_pagequeue_unlock(pq); } /* * vm_page_requeue_locked: * * Move the given page to the tail of its current page queue. * * The page queue must be locked. */ void vm_page_requeue_locked(vm_page_t m) { struct vm_pagequeue *pq; KASSERT(m->queue != PQ_NONE, ("vm_page_requeue_locked: page %p is not queued", m)); pq = vm_page_pagequeue(m); vm_pagequeue_assert_locked(pq); TAILQ_REMOVE(&pq->pq_pl, m, plinks.q); TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q); } /* * vm_page_activate: * * Put the specified page on the active list (if appropriate). * Ensure that act_count is at least ACT_INIT but do not otherwise * mess with it. * * The page must be locked. */ void vm_page_activate(vm_page_t m) { int queue; vm_page_lock_assert(m, MA_OWNED); if ((queue = m->queue) != PQ_ACTIVE) { if (m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0) { if (m->act_count < ACT_INIT) m->act_count = ACT_INIT; if (queue != PQ_NONE) vm_page_dequeue(m); vm_page_enqueue(PQ_ACTIVE, m); } else KASSERT(queue == PQ_NONE, ("vm_page_activate: wired page %p is queued", m)); } else { if (m->act_count < ACT_INIT) m->act_count = ACT_INIT; } } /* * vm_page_free_wakeup: * * Helper routine for vm_page_free_toq() and vm_page_cache(). This * routine is called when a page has been added to the cache or free * queues. * * The page queues must be locked. */ static inline void vm_page_free_wakeup(void) { mtx_assert(&vm_page_queue_free_mtx, MA_OWNED); /* * if pageout daemon needs pages, then tell it that there are * some free. */ if (vm_pageout_pages_needed && vm_cnt.v_cache_count + vm_cnt.v_free_count >= vm_cnt.v_pageout_free_min) { wakeup(&vm_pageout_pages_needed); vm_pageout_pages_needed = 0; } /* * wakeup processes that are waiting on memory if we hit a * high water mark. And wakeup scheduler process if we have * lots of memory. this process will swapin processes. */ if (vm_pages_needed && !vm_page_count_min()) { vm_pages_needed = 0; wakeup(&vm_cnt.v_free_count); } } /* * Turn a cached page into a free page, by changing its attributes. * Keep the statistics up-to-date. * * The free page queue must be locked. */ static void vm_page_cache_turn_free(vm_page_t m) { mtx_assert(&vm_page_queue_free_mtx, MA_OWNED); m->object = NULL; m->valid = 0; KASSERT((m->flags & PG_CACHED) != 0, ("vm_page_cache_turn_free: page %p is not cached", m)); m->flags &= ~PG_CACHED; vm_cnt.v_cache_count--; vm_phys_freecnt_adj(m, 1); } /* * vm_page_free_toq: * * Returns the given page to the free list, * disassociating it with any VM object. * * The object must be locked. The page must be locked if it is managed. */ void vm_page_free_toq(vm_page_t m) { if ((m->oflags & VPO_UNMANAGED) == 0) { vm_page_lock_assert(m, MA_OWNED); KASSERT(!pmap_page_is_mapped(m), ("vm_page_free_toq: freeing mapped page %p", m)); } else KASSERT(m->queue == PQ_NONE, ("vm_page_free_toq: unmanaged page %p is queued", m)); PCPU_INC(cnt.v_tfree); if (vm_page_sbusied(m)) panic("vm_page_free: freeing busy page %p", m); /* * Unqueue, then remove page. Note that we cannot destroy * the page here because we do not want to call the pager's * callback routine until after we've put the page on the * appropriate free queue. */ vm_page_remque(m); vm_page_remove(m); /* * If fictitious remove object association and * return, otherwise delay object association removal. */ if ((m->flags & PG_FICTITIOUS) != 0) { return; } m->valid = 0; vm_page_undirty(m); if (m->wire_count != 0) panic("vm_page_free: freeing wired page %p", m); if (m->hold_count != 0) { m->flags &= ~PG_ZERO; KASSERT((m->flags & PG_UNHOLDFREE) == 0, ("vm_page_free: freeing PG_UNHOLDFREE page %p", m)); m->flags |= PG_UNHOLDFREE; } else { /* * Restore the default memory attribute to the page. */ if (pmap_page_get_memattr(m) != VM_MEMATTR_DEFAULT) pmap_page_set_memattr(m, VM_MEMATTR_DEFAULT); /* * Insert the page into the physical memory allocator's * cache/free page queues. */ mtx_lock(&vm_page_queue_free_mtx); vm_phys_freecnt_adj(m, 1); #if VM_NRESERVLEVEL > 0 if (!vm_reserv_free_page(m)) #else if (TRUE) #endif vm_phys_free_pages(m, 0); if ((m->flags & PG_ZERO) != 0) ++vm_page_zero_count; else vm_page_zero_idle_wakeup(); vm_page_free_wakeup(); mtx_unlock(&vm_page_queue_free_mtx); } } /* * vm_page_wire: * * Mark this page as wired down by yet * another map, removing it from paging queues * as necessary. * * If the page is fictitious, then its wire count must remain one. * * The page must be locked. */ void vm_page_wire(vm_page_t m) { /* * Only bump the wire statistics if the page is not already wired, * and only unqueue the page if it is on some queue (if it is unmanaged * it is already off the queues). */ vm_page_lock_assert(m, MA_OWNED); if ((m->flags & PG_FICTITIOUS) != 0) { KASSERT(m->wire_count == 1, ("vm_page_wire: fictitious page %p's wire count isn't one", m)); return; } if (m->wire_count == 0) { KASSERT((m->oflags & VPO_UNMANAGED) == 0 || m->queue == PQ_NONE, ("vm_page_wire: unmanaged page %p is queued", m)); vm_page_remque(m); atomic_add_int(&vm_cnt.v_wire_count, 1); } m->wire_count++; KASSERT(m->wire_count != 0, ("vm_page_wire: wire_count overflow m=%p", m)); } /* * vm_page_unwire: * * Release one wiring of the specified page, potentially allowing it to be * paged out. Returns TRUE if the number of wirings transitions to zero and * FALSE otherwise. * * Only managed pages belonging to an object can be paged out. If the number * of wirings transitions to zero and the page is eligible for page out, then * the page is added to the specified paging queue (unless PQ_NONE is * specified). * * If a page is fictitious, then its wire count must always be one. * * A managed page must be locked. */ boolean_t vm_page_unwire(vm_page_t m, uint8_t queue) { KASSERT(queue < PQ_COUNT || queue == PQ_NONE, ("vm_page_unwire: invalid queue %u request for page %p", queue, m)); if ((m->oflags & VPO_UNMANAGED) == 0) vm_page_assert_locked(m); if ((m->flags & PG_FICTITIOUS) != 0) { KASSERT(m->wire_count == 1, ("vm_page_unwire: fictitious page %p's wire count isn't one", m)); return (FALSE); } if (m->wire_count > 0) { m->wire_count--; if (m->wire_count == 0) { atomic_subtract_int(&vm_cnt.v_wire_count, 1); if ((m->oflags & VPO_UNMANAGED) == 0 && m->object != NULL && queue != PQ_NONE) { if (queue == PQ_INACTIVE) m->flags &= ~PG_WINATCFLS; vm_page_enqueue(queue, m); } return (TRUE); } else return (FALSE); } else panic("vm_page_unwire: page %p's wire count is zero", m); } /* * Move the specified page to the inactive queue. * * Many pages placed on the inactive queue should actually go * into the cache, but it is difficult to figure out which. What * we do instead, if the inactive target is well met, is to put * clean pages at the head of the inactive queue instead of the tail. * This will cause them to be moved to the cache more quickly and * if not actively re-referenced, reclaimed more quickly. If we just * stick these pages at the end of the inactive queue, heavy filesystem * meta-data accesses can cause an unnecessary paging load on memory bound * processes. This optimization causes one-time-use metadata to be * reused more quickly. * * Normally athead is 0 resulting in LRU operation. athead is set * to 1 if we want this page to be 'as if it were placed in the cache', * except without unmapping it from the process address space. * * The page must be locked. */ static inline void _vm_page_deactivate(vm_page_t m, int athead) { struct vm_pagequeue *pq; int queue; vm_page_assert_locked(m); /* * Ignore if the page is already inactive, unless it is unlikely to be * reactivated. */ if ((queue = m->queue) == PQ_INACTIVE && !athead) return; if (m->wire_count == 0 && (m->oflags & VPO_UNMANAGED) == 0) { pq = &vm_phys_domain(m)->vmd_pagequeues[PQ_INACTIVE]; /* Avoid multiple acquisitions of the inactive queue lock. */ if (queue == PQ_INACTIVE) { vm_pagequeue_lock(pq); vm_page_dequeue_locked(m); } else { if (queue != PQ_NONE) vm_page_dequeue(m); m->flags &= ~PG_WINATCFLS; vm_pagequeue_lock(pq); } m->queue = PQ_INACTIVE; if (athead) TAILQ_INSERT_HEAD(&pq->pq_pl, m, plinks.q); else TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q); vm_pagequeue_cnt_inc(pq); vm_pagequeue_unlock(pq); } } /* * Move the specified page to the inactive queue. * * The page must be locked. */ void vm_page_deactivate(vm_page_t m) { _vm_page_deactivate(m, 0); } /* + * Move the specified page to the inactive queue with the expectation + * that it is unlikely to be reused. + * + * The page must be locked. + */ +void +vm_page_deactivate_noreuse(vm_page_t m) +{ + + _vm_page_deactivate(m, 1); +} + +/* * vm_page_try_to_cache: * * Returns 0 on failure, 1 on success */ int vm_page_try_to_cache(vm_page_t m) { vm_page_lock_assert(m, MA_OWNED); VM_OBJECT_ASSERT_WLOCKED(m->object); if (m->dirty || m->hold_count || m->wire_count || (m->oflags & VPO_UNMANAGED) != 0 || vm_page_busied(m)) return (0); pmap_remove_all(m); if (m->dirty) return (0); vm_page_cache(m); return (1); } /* * vm_page_try_to_free() * * Attempt to free the page. If we cannot free it, we do nothing. * 1 is returned on success, 0 on failure. */ int vm_page_try_to_free(vm_page_t m) { vm_page_lock_assert(m, MA_OWNED); if (m->object != NULL) VM_OBJECT_ASSERT_WLOCKED(m->object); if (m->dirty || m->hold_count || m->wire_count || (m->oflags & VPO_UNMANAGED) != 0 || vm_page_busied(m)) return (0); pmap_remove_all(m); if (m->dirty) return (0); vm_page_free(m); return (1); } /* * vm_page_cache * * Put the specified page onto the page cache queue (if appropriate). * * The object and page must be locked. */ void vm_page_cache(vm_page_t m) { vm_object_t object; boolean_t cache_was_empty; vm_page_lock_assert(m, MA_OWNED); object = m->object; VM_OBJECT_ASSERT_WLOCKED(object); if (vm_page_busied(m) || (m->oflags & VPO_UNMANAGED) || m->hold_count || m->wire_count) panic("vm_page_cache: attempting to cache busy page"); KASSERT(!pmap_page_is_mapped(m), ("vm_page_cache: page %p is mapped", m)); KASSERT(m->dirty == 0, ("vm_page_cache: page %p is dirty", m)); if (m->valid == 0 || object->type == OBJT_DEFAULT || (object->type == OBJT_SWAP && !vm_pager_has_page(object, m->pindex, NULL, NULL))) { /* * Hypothesis: A cache-eligible page belonging to a * default object or swap object but without a backing * store must be zero filled. */ vm_page_free(m); return; } KASSERT((m->flags & PG_CACHED) == 0, ("vm_page_cache: page %p is already cached", m)); /* * Remove the page from the paging queues. */ vm_page_remque(m); /* * Remove the page from the object's collection of resident * pages. */ vm_radix_remove(&object->rtree, m->pindex); TAILQ_REMOVE(&object->memq, m, listq); object->resident_page_count--; /* * Restore the default memory attribute to the page. */ if (pmap_page_get_memattr(m) != VM_MEMATTR_DEFAULT) pmap_page_set_memattr(m, VM_MEMATTR_DEFAULT); /* * Insert the page into the object's collection of cached pages * and the physical memory allocator's cache/free page queues. */ m->flags &= ~PG_ZERO; mtx_lock(&vm_page_queue_free_mtx); cache_was_empty = vm_radix_is_empty(&object->cache); if (vm_radix_insert(&object->cache, m)) { mtx_unlock(&vm_page_queue_free_mtx); if (object->resident_page_count == 0) vdrop(object->handle); m->object = NULL; vm_page_free(m); return; } /* * The above call to vm_radix_insert() could reclaim the one pre- * existing cached page from this object, resulting in a call to * vdrop(). */ if (!cache_was_empty) cache_was_empty = vm_radix_is_singleton(&object->cache); m->flags |= PG_CACHED; vm_cnt.v_cache_count++; PCPU_INC(cnt.v_tcached); #if VM_NRESERVLEVEL > 0 if (!vm_reserv_free_page(m)) { #else if (TRUE) { #endif vm_phys_free_pages(m, 0); } vm_page_free_wakeup(); mtx_unlock(&vm_page_queue_free_mtx); /* * Increment the vnode's hold count if this is the object's only * cached page. Decrement the vnode's hold count if this was * the object's only resident page. */ if (object->type == OBJT_VNODE) { if (cache_was_empty && object->resident_page_count != 0) vhold(object->handle); else if (!cache_was_empty && object->resident_page_count == 0) vdrop(object->handle); } } /* * vm_page_advise * - * Deactivate or do nothing, as appropriate. This routine is used - * by madvise() and vop_stdadvise(). + * Deactivate or do nothing, as appropriate. * * The object and page must be locked. */ void vm_page_advise(vm_page_t m, int advice) { vm_page_assert_locked(m); VM_OBJECT_ASSERT_WLOCKED(m->object); if (advice == MADV_FREE) /* * Mark the page clean. This will allow the page to be freed * up by the system. However, such pages are often reused * quickly by malloc() so we do not do anything that would * cause a page fault if we can help it. * * Specifically, we do not try to actually free the page now * nor do we try to put it in the cache (which would cause a * page fault on reuse). * * But we do make the page as freeable as we can without * actually taking the step of unmapping it. */ m->dirty = 0; else if (advice != MADV_DONTNEED) return; /* * Clear any references to the page. Otherwise, the page daemon will * immediately reactivate the page. */ vm_page_aflag_clear(m, PGA_REFERENCED); if (advice != MADV_FREE && m->dirty == 0 && pmap_is_modified(m)) vm_page_dirty(m); /* * Place clean pages at the head of the inactive queue rather than the * tail, thus defeating the queue's LRU operation and ensuring that the * page will be reused quickly. */ _vm_page_deactivate(m, m->dirty == 0); } /* * Grab a page, waiting until we are waken up due to the page * changing state. We keep on waiting, if the page continues * to be in the object. If the page doesn't exist, first allocate it * and then conditionally zero it. * * This routine may sleep. * * The object must be locked on entry. The lock will, however, be released * and reacquired if the routine sleeps. */ vm_page_t vm_page_grab(vm_object_t object, vm_pindex_t pindex, int allocflags) { vm_page_t m; int sleep; VM_OBJECT_ASSERT_WLOCKED(object); KASSERT((allocflags & VM_ALLOC_SBUSY) == 0 || (allocflags & VM_ALLOC_IGN_SBUSY) != 0, ("vm_page_grab: VM_ALLOC_SBUSY/VM_ALLOC_IGN_SBUSY mismatch")); retrylookup: if ((m = vm_page_lookup(object, pindex)) != NULL) { sleep = (allocflags & VM_ALLOC_IGN_SBUSY) != 0 ? vm_page_xbusied(m) : vm_page_busied(m); if (sleep) { if ((allocflags & VM_ALLOC_NOWAIT) != 0) return (NULL); /* * Reference the page before unlocking and * sleeping so that the page daemon is less * likely to reclaim it. */ vm_page_aflag_set(m, PGA_REFERENCED); vm_page_lock(m); VM_OBJECT_WUNLOCK(object); vm_page_busy_sleep(m, "pgrbwt"); VM_OBJECT_WLOCK(object); goto retrylookup; } else { if ((allocflags & VM_ALLOC_WIRED) != 0) { vm_page_lock(m); vm_page_wire(m); vm_page_unlock(m); } if ((allocflags & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) == 0) vm_page_xbusy(m); if ((allocflags & VM_ALLOC_SBUSY) != 0) vm_page_sbusy(m); return (m); } } m = vm_page_alloc(object, pindex, allocflags); if (m == NULL) { if ((allocflags & VM_ALLOC_NOWAIT) != 0) return (NULL); VM_OBJECT_WUNLOCK(object); VM_WAIT; VM_OBJECT_WLOCK(object); goto retrylookup; } else if (m->valid != 0) return (m); if (allocflags & VM_ALLOC_ZERO && (m->flags & PG_ZERO) == 0) pmap_zero_page(m); return (m); } /* * Mapping function for valid or dirty bits in a page. * * Inputs are required to range within a page. */ vm_page_bits_t vm_page_bits(int base, int size) { int first_bit; int last_bit; KASSERT( base + size <= PAGE_SIZE, ("vm_page_bits: illegal base/size %d/%d", base, size) ); if (size == 0) /* handle degenerate case */ return (0); first_bit = base >> DEV_BSHIFT; last_bit = (base + size - 1) >> DEV_BSHIFT; return (((vm_page_bits_t)2 << last_bit) - ((vm_page_bits_t)1 << first_bit)); } /* * vm_page_set_valid_range: * * Sets portions of a page valid. The arguments are expected * to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive * of any partial chunks touched by the range. The invalid portion of * such chunks will be zeroed. * * (base + size) must be less then or equal to PAGE_SIZE. */ void vm_page_set_valid_range(vm_page_t m, int base, int size) { int endoff, frag; VM_OBJECT_ASSERT_WLOCKED(m->object); if (size == 0) /* handle degenerate case */ return; /* * If the base is not DEV_BSIZE aligned and the valid * bit is clear, we have to zero out a portion of the * first block. */ if ((frag = base & ~(DEV_BSIZE - 1)) != base && (m->valid & (1 << (base >> DEV_BSHIFT))) == 0) pmap_zero_page_area(m, frag, base - frag); /* * If the ending offset is not DEV_BSIZE aligned and the * valid bit is clear, we have to zero out a portion of * the last block. */ endoff = base + size; if ((frag = endoff & ~(DEV_BSIZE - 1)) != endoff && (m->valid & (1 << (endoff >> DEV_BSHIFT))) == 0) pmap_zero_page_area(m, endoff, DEV_BSIZE - (endoff & (DEV_BSIZE - 1))); /* * Assert that no previously invalid block that is now being validated * is already dirty. */ KASSERT((~m->valid & vm_page_bits(base, size) & m->dirty) == 0, ("vm_page_set_valid_range: page %p is dirty", m)); /* * Set valid bits inclusive of any overlap. */ m->valid |= vm_page_bits(base, size); } /* * Clear the given bits from the specified page's dirty field. */ static __inline void vm_page_clear_dirty_mask(vm_page_t m, vm_page_bits_t pagebits) { uintptr_t addr; #if PAGE_SIZE < 16384 int shift; #endif /* * If the object is locked and the page is neither exclusive busy nor * write mapped, then the page's dirty field cannot possibly be * set by a concurrent pmap operation. */ VM_OBJECT_ASSERT_WLOCKED(m->object); if (!vm_page_xbusied(m) && !pmap_page_is_write_mapped(m)) m->dirty &= ~pagebits; else { /* * The pmap layer can call vm_page_dirty() without * holding a distinguished lock. The combination of * the object's lock and an atomic operation suffice * to guarantee consistency of the page dirty field. * * For PAGE_SIZE == 32768 case, compiler already * properly aligns the dirty field, so no forcible * alignment is needed. Only require existence of * atomic_clear_64 when page size is 32768. */ addr = (uintptr_t)&m->dirty; #if PAGE_SIZE == 32768 atomic_clear_64((uint64_t *)addr, pagebits); #elif PAGE_SIZE == 16384 atomic_clear_32((uint32_t *)addr, pagebits); #else /* PAGE_SIZE <= 8192 */ /* * Use a trick to perform a 32-bit atomic on the * containing aligned word, to not depend on the existence * of atomic_clear_{8, 16}. */ shift = addr & (sizeof(uint32_t) - 1); #if BYTE_ORDER == BIG_ENDIAN shift = (sizeof(uint32_t) - sizeof(m->dirty) - shift) * NBBY; #else shift *= NBBY; #endif addr &= ~(sizeof(uint32_t) - 1); atomic_clear_32((uint32_t *)addr, pagebits << shift); #endif /* PAGE_SIZE */ } } /* * vm_page_set_validclean: * * Sets portions of a page valid and clean. The arguments are expected * to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive * of any partial chunks touched by the range. The invalid portion of * such chunks will be zero'd. * * (base + size) must be less then or equal to PAGE_SIZE. */ void vm_page_set_validclean(vm_page_t m, int base, int size) { vm_page_bits_t oldvalid, pagebits; int endoff, frag; VM_OBJECT_ASSERT_WLOCKED(m->object); if (size == 0) /* handle degenerate case */ return; /* * If the base is not DEV_BSIZE aligned and the valid * bit is clear, we have to zero out a portion of the * first block. */ if ((frag = base & ~(DEV_BSIZE - 1)) != base && (m->valid & ((vm_page_bits_t)1 << (base >> DEV_BSHIFT))) == 0) pmap_zero_page_area(m, frag, base - frag); /* * If the ending offset is not DEV_BSIZE aligned and the * valid bit is clear, we have to zero out a portion of * the last block. */ endoff = base + size; if ((frag = endoff & ~(DEV_BSIZE - 1)) != endoff && (m->valid & ((vm_page_bits_t)1 << (endoff >> DEV_BSHIFT))) == 0) pmap_zero_page_area(m, endoff, DEV_BSIZE - (endoff & (DEV_BSIZE - 1))); /* * Set valid, clear dirty bits. If validating the entire * page we can safely clear the pmap modify bit. We also * use this opportunity to clear the VPO_NOSYNC flag. If a process * takes a write fault on a MAP_NOSYNC memory area the flag will * be set again. * * We set valid bits inclusive of any overlap, but we can only * clear dirty bits for DEV_BSIZE chunks that are fully within * the range. */ oldvalid = m->valid; pagebits = vm_page_bits(base, size); m->valid |= pagebits; #if 0 /* NOT YET */ if ((frag = base & (DEV_BSIZE - 1)) != 0) { frag = DEV_BSIZE - frag; base += frag; size -= frag; if (size < 0) size = 0; } pagebits = vm_page_bits(base, size & (DEV_BSIZE - 1)); #endif if (base == 0 && size == PAGE_SIZE) { /* * The page can only be modified within the pmap if it is * mapped, and it can only be mapped if it was previously * fully valid. */ if (oldvalid == VM_PAGE_BITS_ALL) /* * Perform the pmap_clear_modify() first. Otherwise, * a concurrent pmap operation, such as * pmap_protect(), could clear a modification in the * pmap and set the dirty field on the page before * pmap_clear_modify() had begun and after the dirty * field was cleared here. */ pmap_clear_modify(m); m->dirty = 0; m->oflags &= ~VPO_NOSYNC; } else if (oldvalid != VM_PAGE_BITS_ALL) m->dirty &= ~pagebits; else vm_page_clear_dirty_mask(m, pagebits); } void vm_page_clear_dirty(vm_page_t m, int base, int size) { vm_page_clear_dirty_mask(m, vm_page_bits(base, size)); } /* * vm_page_set_invalid: * * Invalidates DEV_BSIZE'd chunks within a page. Both the * valid and dirty bits for the effected areas are cleared. */ void vm_page_set_invalid(vm_page_t m, int base, int size) { vm_page_bits_t bits; vm_object_t object; object = m->object; VM_OBJECT_ASSERT_WLOCKED(object); if (object->type == OBJT_VNODE && base == 0 && IDX_TO_OFF(m->pindex) + size >= object->un_pager.vnp.vnp_size) bits = VM_PAGE_BITS_ALL; else bits = vm_page_bits(base, size); if (object->ref_count != 0 && m->valid == VM_PAGE_BITS_ALL && bits != 0) pmap_remove_all(m); KASSERT((bits == 0 && m->valid == VM_PAGE_BITS_ALL) || !pmap_page_is_mapped(m), ("vm_page_set_invalid: page %p is mapped", m)); m->valid &= ~bits; m->dirty &= ~bits; } /* * vm_page_zero_invalid() * * The kernel assumes that the invalid portions of a page contain * garbage, but such pages can be mapped into memory by user code. * When this occurs, we must zero out the non-valid portions of the * page so user code sees what it expects. * * Pages are most often semi-valid when the end of a file is mapped * into memory and the file's size is not page aligned. */ void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid) { int b; int i; VM_OBJECT_ASSERT_WLOCKED(m->object); /* * Scan the valid bits looking for invalid sections that * must be zeroed. Invalid sub-DEV_BSIZE'd areas ( where the * valid bit may be set ) have already been zeroed by * vm_page_set_validclean(). */ for (b = i = 0; i <= PAGE_SIZE / DEV_BSIZE; ++i) { if (i == (PAGE_SIZE / DEV_BSIZE) || (m->valid & ((vm_page_bits_t)1 << i))) { if (i > b) { pmap_zero_page_area(m, b << DEV_BSHIFT, (i - b) << DEV_BSHIFT); } b = i + 1; } } /* * setvalid is TRUE when we can safely set the zero'd areas * as being valid. We can do this if there are no cache consistancy * issues. e.g. it is ok to do with UFS, but not ok to do with NFS. */ if (setvalid) m->valid = VM_PAGE_BITS_ALL; } /* * vm_page_is_valid: * * Is (partial) page valid? Note that the case where size == 0 * will return FALSE in the degenerate case where the page is * entirely invalid, and TRUE otherwise. */ int vm_page_is_valid(vm_page_t m, int base, int size) { vm_page_bits_t bits; VM_OBJECT_ASSERT_LOCKED(m->object); bits = vm_page_bits(base, size); return (m->valid != 0 && (m->valid & bits) == bits); } /* * vm_page_ps_is_valid: * * Returns TRUE if the entire (super)page is valid and FALSE otherwise. */ boolean_t vm_page_ps_is_valid(vm_page_t m) { int i, npages; VM_OBJECT_ASSERT_LOCKED(m->object); npages = atop(pagesizes[m->psind]); /* * The physically contiguous pages that make up a superpage, i.e., a * page with a page size index ("psind") greater than zero, will * occupy adjacent entries in vm_page_array[]. */ for (i = 0; i < npages; i++) { if (m[i].valid != VM_PAGE_BITS_ALL) return (FALSE); } return (TRUE); } /* * Set the page's dirty bits if the page is modified. */ void vm_page_test_dirty(vm_page_t m) { VM_OBJECT_ASSERT_WLOCKED(m->object); if (m->dirty != VM_PAGE_BITS_ALL && pmap_is_modified(m)) vm_page_dirty(m); } void vm_page_lock_KBI(vm_page_t m, const char *file, int line) { mtx_lock_flags_(vm_page_lockptr(m), 0, file, line); } void vm_page_unlock_KBI(vm_page_t m, const char *file, int line) { mtx_unlock_flags_(vm_page_lockptr(m), 0, file, line); } int vm_page_trylock_KBI(vm_page_t m, const char *file, int line) { return (mtx_trylock_flags_(vm_page_lockptr(m), 0, file, line)); } #if defined(INVARIANTS) || defined(INVARIANT_SUPPORT) void vm_page_assert_locked_KBI(vm_page_t m, const char *file, int line) { vm_page_lock_assert_KBI(m, MA_OWNED, file, line); } void vm_page_lock_assert_KBI(vm_page_t m, int a, const char *file, int line) { mtx_assert_(vm_page_lockptr(m), a, file, line); } #endif #ifdef INVARIANTS void vm_page_object_lock_assert(vm_page_t m) { /* * Certain of the page's fields may only be modified by the * holder of the containing object's lock or the exclusive busy. * holder. Unfortunately, the holder of the write busy is * not recorded, and thus cannot be checked here. */ if (m->object != NULL && !vm_page_xbusied(m)) VM_OBJECT_ASSERT_WLOCKED(m->object); } void vm_page_assert_pga_writeable(vm_page_t m, uint8_t bits) { if ((bits & PGA_WRITEABLE) == 0) return; /* * The PGA_WRITEABLE flag can only be set if the page is * managed, is exclusively busied or the object is locked. * Currently, this flag is only set by pmap_enter(). */ KASSERT((m->oflags & VPO_UNMANAGED) == 0, ("PGA_WRITEABLE on unmanaged page")); if (!vm_page_xbusied(m)) VM_OBJECT_ASSERT_LOCKED(m->object); } #endif #include "opt_ddb.h" #ifdef DDB #include #include DB_SHOW_COMMAND(page, vm_page_print_page_info) { db_printf("vm_cnt.v_free_count: %d\n", vm_cnt.v_free_count); db_printf("vm_cnt.v_cache_count: %d\n", vm_cnt.v_cache_count); db_printf("vm_cnt.v_inactive_count: %d\n", vm_cnt.v_inactive_count); db_printf("vm_cnt.v_active_count: %d\n", vm_cnt.v_active_count); db_printf("vm_cnt.v_wire_count: %d\n", vm_cnt.v_wire_count); db_printf("vm_cnt.v_free_reserved: %d\n", vm_cnt.v_free_reserved); db_printf("vm_cnt.v_free_min: %d\n", vm_cnt.v_free_min); db_printf("vm_cnt.v_free_target: %d\n", vm_cnt.v_free_target); db_printf("vm_cnt.v_inactive_target: %d\n", vm_cnt.v_inactive_target); } DB_SHOW_COMMAND(pageq, vm_page_print_pageq_info) { int dom; db_printf("pq_free %d pq_cache %d\n", vm_cnt.v_free_count, vm_cnt.v_cache_count); for (dom = 0; dom < vm_ndomains; dom++) { db_printf( "dom %d page_cnt %d free %d pq_act %d pq_inact %d pass %d\n", dom, vm_dom[dom].vmd_page_count, vm_dom[dom].vmd_free_count, vm_dom[dom].vmd_pagequeues[PQ_ACTIVE].pq_cnt, vm_dom[dom].vmd_pagequeues[PQ_INACTIVE].pq_cnt, vm_dom[dom].vmd_pass); } } DB_SHOW_COMMAND(pginfo, vm_page_print_pginfo) { vm_page_t m; boolean_t phys; if (!have_addr) { db_printf("show pginfo addr\n"); return; } phys = strchr(modif, 'p') != NULL; if (phys) m = PHYS_TO_VM_PAGE(addr); else m = (vm_page_t)addr; db_printf( "page %p obj %p pidx 0x%jx phys 0x%jx q %d hold %d wire %d\n" " af 0x%x of 0x%x f 0x%x act %d busy %x valid 0x%x dirty 0x%x\n", m, m->object, (uintmax_t)m->pindex, (uintmax_t)m->phys_addr, m->queue, m->hold_count, m->wire_count, m->aflags, m->oflags, m->flags, m->act_count, m->busy_lock, m->valid, m->dirty); } #endif /* DDB */ Index: head/sys/vm/vm_page.h =================================================================== --- head/sys/vm/vm_page.h (revision 288430) +++ head/sys/vm/vm_page.h (revision 288431) @@ -1,673 +1,674 @@ /*- * Copyright (c) 1991, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * The Mach Operating System project at Carnegie-Mellon University. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * from: @(#)vm_page.h 8.2 (Berkeley) 12/13/93 * * * Copyright (c) 1987, 1990 Carnegie-Mellon University. * All rights reserved. * * Authors: Avadis Tevanian, Jr., Michael Wayne Young * * Permission to use, copy, modify and distribute this software and * its documentation is hereby granted, provided that both the copyright * notice and this permission notice appear in all copies of the * software, derivative works or modified versions, and any portions * thereof, and that both notices appear in supporting documentation. * * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. * * Carnegie Mellon requests users of this software to return to * * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU * School of Computer Science * Carnegie Mellon University * Pittsburgh PA 15213-3890 * * any improvements or extensions that they make and grant Carnegie the * rights to redistribute these changes. * * $FreeBSD$ */ /* * Resident memory system definitions. */ #ifndef _VM_PAGE_ #define _VM_PAGE_ #include /* * Management of resident (logical) pages. * * A small structure is kept for each resident * page, indexed by page number. Each structure * is an element of several collections: * * A radix tree used to quickly * perform object/offset lookups * * A list of all pages for a given object, * so they can be quickly deactivated at * time of deallocation. * * An ordered list of pages due for pageout. * * In addition, the structure contains the object * and offset to which this page belongs (for pageout), * and sundry status bits. * * In general, operations on this structure's mutable fields are * synchronized using either one of or a combination of the lock on the * object that the page belongs to (O), the pool lock for the page (P), * or the lock for either the free or paging queue (Q). If a field is * annotated below with two of these locks, then holding either lock is * sufficient for read access, but both locks are required for write * access. * * In contrast, the synchronization of accesses to the page's * dirty field is machine dependent (M). In the * machine-independent layer, the lock on the object that the * page belongs to must be held in order to operate on the field. * However, the pmap layer is permitted to set all bits within * the field without holding that lock. If the underlying * architecture does not support atomic read-modify-write * operations on the field's type, then the machine-independent * layer uses a 32-bit atomic on the aligned 32-bit word that * contains the dirty field. In the machine-independent layer, * the implementation of read-modify-write operations on the * field is encapsulated in vm_page_clear_dirty_mask(). */ #if PAGE_SIZE == 4096 #define VM_PAGE_BITS_ALL 0xffu typedef uint8_t vm_page_bits_t; #elif PAGE_SIZE == 8192 #define VM_PAGE_BITS_ALL 0xffffu typedef uint16_t vm_page_bits_t; #elif PAGE_SIZE == 16384 #define VM_PAGE_BITS_ALL 0xffffffffu typedef uint32_t vm_page_bits_t; #elif PAGE_SIZE == 32768 #define VM_PAGE_BITS_ALL 0xfffffffffffffffflu typedef uint64_t vm_page_bits_t; #endif struct vm_page { union { TAILQ_ENTRY(vm_page) q; /* page queue or free list (Q) */ struct { SLIST_ENTRY(vm_page) ss; /* private slists */ void *pv; } s; struct { u_long p; u_long v; } memguard; } plinks; TAILQ_ENTRY(vm_page) listq; /* pages in same object (O) */ vm_object_t object; /* which object am I in (O,P) */ vm_pindex_t pindex; /* offset into object (O,P) */ vm_paddr_t phys_addr; /* physical address of page */ struct md_page md; /* machine dependant stuff */ u_int wire_count; /* wired down maps refs (P) */ volatile u_int busy_lock; /* busy owners lock */ uint16_t hold_count; /* page hold count (P) */ uint16_t flags; /* page PG_* flags (P) */ uint8_t aflags; /* access is atomic */ uint8_t oflags; /* page VPO_* flags (O) */ uint8_t queue; /* page queue index (P,Q) */ int8_t psind; /* pagesizes[] index (O) */ int8_t segind; uint8_t order; /* index of the buddy queue */ uint8_t pool; u_char act_count; /* page usage count (P) */ /* NOTE that these must support one bit per DEV_BSIZE in a page */ /* so, on normal X86 kernels, they must be at least 8 bits wide */ vm_page_bits_t valid; /* map of valid DEV_BSIZE chunks (O) */ vm_page_bits_t dirty; /* map of dirty DEV_BSIZE chunks (M) */ }; /* * Page flags stored in oflags: * * Access to these page flags is synchronized by the lock on the object * containing the page (O). * * Note: VPO_UNMANAGED (used by OBJT_DEVICE, OBJT_PHYS and OBJT_SG) * indicates that the page is not under PV management but * otherwise should be treated as a normal page. Pages not * under PV management cannot be paged out via the * object/vm_page_t because there is no knowledge of their pte * mappings, and such pages are also not on any PQ queue. * */ #define VPO_UNUSED01 0x01 /* --available-- */ #define VPO_SWAPSLEEP 0x02 /* waiting for swap to finish */ #define VPO_UNMANAGED 0x04 /* no PV management for page */ #define VPO_SWAPINPROG 0x08 /* swap I/O in progress on page */ #define VPO_NOSYNC 0x10 /* do not collect for syncer */ /* * Busy page implementation details. * The algorithm is taken mostly by rwlock(9) and sx(9) locks implementation, * even if the support for owner identity is removed because of size * constraints. Checks on lock recursion are then not possible, while the * lock assertions effectiveness is someway reduced. */ #define VPB_BIT_SHARED 0x01 #define VPB_BIT_EXCLUSIVE 0x02 #define VPB_BIT_WAITERS 0x04 #define VPB_BIT_FLAGMASK \ (VPB_BIT_SHARED | VPB_BIT_EXCLUSIVE | VPB_BIT_WAITERS) #define VPB_SHARERS_SHIFT 3 #define VPB_SHARERS(x) \ (((x) & ~VPB_BIT_FLAGMASK) >> VPB_SHARERS_SHIFT) #define VPB_SHARERS_WORD(x) ((x) << VPB_SHARERS_SHIFT | VPB_BIT_SHARED) #define VPB_ONE_SHARER (1 << VPB_SHARERS_SHIFT) #define VPB_SINGLE_EXCLUSIVER VPB_BIT_EXCLUSIVE #define VPB_UNBUSIED VPB_SHARERS_WORD(0) #define PQ_NONE 255 #define PQ_INACTIVE 0 #define PQ_ACTIVE 1 #define PQ_COUNT 2 TAILQ_HEAD(pglist, vm_page); SLIST_HEAD(spglist, vm_page); struct vm_pagequeue { struct mtx pq_mutex; struct pglist pq_pl; int pq_cnt; int * const pq_vcnt; const char * const pq_name; } __aligned(CACHE_LINE_SIZE); struct vm_domain { struct vm_pagequeue vmd_pagequeues[PQ_COUNT]; u_int vmd_page_count; u_int vmd_free_count; long vmd_segs; /* bitmask of the segments */ boolean_t vmd_oom; int vmd_pass; /* local pagedaemon pass */ int vmd_last_active_scan; struct vm_page vmd_marker; /* marker for pagedaemon private use */ }; extern struct vm_domain vm_dom[MAXMEMDOM]; #define vm_pagequeue_assert_locked(pq) mtx_assert(&(pq)->pq_mutex, MA_OWNED) #define vm_pagequeue_lock(pq) mtx_lock(&(pq)->pq_mutex) #define vm_pagequeue_unlock(pq) mtx_unlock(&(pq)->pq_mutex) #ifdef _KERNEL static __inline void vm_pagequeue_cnt_add(struct vm_pagequeue *pq, int addend) { #ifdef notyet vm_pagequeue_assert_locked(pq); #endif pq->pq_cnt += addend; atomic_add_int(pq->pq_vcnt, addend); } #define vm_pagequeue_cnt_inc(pq) vm_pagequeue_cnt_add((pq), 1) #define vm_pagequeue_cnt_dec(pq) vm_pagequeue_cnt_add((pq), -1) #endif /* _KERNEL */ extern struct mtx_padalign vm_page_queue_free_mtx; extern struct mtx_padalign pa_lock[]; #if defined(__arm__) #define PDRSHIFT PDR_SHIFT #elif !defined(PDRSHIFT) #define PDRSHIFT 21 #endif #define pa_index(pa) ((pa) >> PDRSHIFT) #define PA_LOCKPTR(pa) ((struct mtx *)(&pa_lock[pa_index(pa) % PA_LOCK_COUNT])) #define PA_LOCKOBJPTR(pa) ((struct lock_object *)PA_LOCKPTR((pa))) #define PA_LOCK(pa) mtx_lock(PA_LOCKPTR(pa)) #define PA_TRYLOCK(pa) mtx_trylock(PA_LOCKPTR(pa)) #define PA_UNLOCK(pa) mtx_unlock(PA_LOCKPTR(pa)) #define PA_UNLOCK_COND(pa) \ do { \ if ((pa) != 0) { \ PA_UNLOCK((pa)); \ (pa) = 0; \ } \ } while (0) #define PA_LOCK_ASSERT(pa, a) mtx_assert(PA_LOCKPTR(pa), (a)) #ifdef KLD_MODULE #define vm_page_lock(m) vm_page_lock_KBI((m), LOCK_FILE, LOCK_LINE) #define vm_page_unlock(m) vm_page_unlock_KBI((m), LOCK_FILE, LOCK_LINE) #define vm_page_trylock(m) vm_page_trylock_KBI((m), LOCK_FILE, LOCK_LINE) #else /* !KLD_MODULE */ #define vm_page_lockptr(m) (PA_LOCKPTR(VM_PAGE_TO_PHYS((m)))) #define vm_page_lock(m) mtx_lock(vm_page_lockptr((m))) #define vm_page_unlock(m) mtx_unlock(vm_page_lockptr((m))) #define vm_page_trylock(m) mtx_trylock(vm_page_lockptr((m))) #endif #if defined(INVARIANTS) #define vm_page_assert_locked(m) \ vm_page_assert_locked_KBI((m), __FILE__, __LINE__) #define vm_page_lock_assert(m, a) \ vm_page_lock_assert_KBI((m), (a), __FILE__, __LINE__) #else #define vm_page_assert_locked(m) #define vm_page_lock_assert(m, a) #endif /* * The vm_page's aflags are updated using atomic operations. To set or clear * these flags, the functions vm_page_aflag_set() and vm_page_aflag_clear() * must be used. Neither these flags nor these functions are part of the KBI. * * PGA_REFERENCED may be cleared only if the page is locked. It is set by * both the MI and MD VM layers. However, kernel loadable modules should not * directly set this flag. They should call vm_page_reference() instead. * * PGA_WRITEABLE is set exclusively on managed pages by pmap_enter(). * When it does so, the object must be locked, or the page must be * exclusive busied. The MI VM layer must never access this flag * directly. Instead, it should call pmap_page_is_write_mapped(). * * PGA_EXECUTABLE may be set by pmap routines, and indicates that a page has * at least one executable mapping. It is not consumed by the MI VM layer. */ #define PGA_WRITEABLE 0x01 /* page may be mapped writeable */ #define PGA_REFERENCED 0x02 /* page has been referenced */ #define PGA_EXECUTABLE 0x04 /* page may be mapped executable */ /* * Page flags. If changed at any other time than page allocation or * freeing, the modification must be protected by the vm_page lock. */ #define PG_CACHED 0x0001 /* page is cached */ #define PG_FICTITIOUS 0x0004 /* physical page doesn't exist */ #define PG_ZERO 0x0008 /* page is zeroed */ #define PG_MARKER 0x0010 /* special queue marker page */ #define PG_WINATCFLS 0x0040 /* flush dirty page on inactive q */ #define PG_NODUMP 0x0080 /* don't include this page in a dump */ #define PG_UNHOLDFREE 0x0100 /* delayed free of a held page */ /* * Misc constants. */ #define ACT_DECLINE 1 #define ACT_ADVANCE 3 #define ACT_INIT 5 #define ACT_MAX 64 #ifdef _KERNEL #include #include /* * Each pageable resident page falls into one of four lists: * * free * Available for allocation now. * * cache * Almost available for allocation. Still associated with * an object, but clean and immediately freeable. * * The following lists are LRU sorted: * * inactive * Low activity, candidates for reclamation. * This is the list of pages that should be * paged out next. * * active * Pages that are "active" i.e. they have been * recently referenced. * */ extern int vm_page_zero_count; extern vm_page_t vm_page_array; /* First resident page in table */ extern long vm_page_array_size; /* number of vm_page_t's */ extern long first_page; /* first physical page number */ #define VM_PAGE_TO_PHYS(entry) ((entry)->phys_addr) vm_page_t PHYS_TO_VM_PAGE(vm_paddr_t pa); /* * Page allocation parameters for vm_page for the functions * vm_page_alloc(), vm_page_grab(), vm_page_alloc_contig() and * vm_page_alloc_freelist(). Some functions support only a subset * of the flags, and ignore others, see the flags legend. * * Bits 0 - 1 define class. * Bits 2 - 15 dedicated for flags. * Legend: * (a) - vm_page_alloc() supports the flag. * (c) - vm_page_alloc_contig() supports the flag. * (f) - vm_page_alloc_freelist() supports the flag. * (g) - vm_page_grab() supports the flag. * Bits above 15 define the count of additional pages that the caller * intends to allocate. */ #define VM_ALLOC_NORMAL 0 #define VM_ALLOC_INTERRUPT 1 #define VM_ALLOC_SYSTEM 2 #define VM_ALLOC_CLASS_MASK 3 #define VM_ALLOC_WIRED 0x0020 /* (acfg) Allocate non pageable page */ #define VM_ALLOC_ZERO 0x0040 /* (acfg) Try to obtain a zeroed page */ #define VM_ALLOC_NOOBJ 0x0100 /* (acg) No associated object */ #define VM_ALLOC_NOBUSY 0x0200 /* (acg) Do not busy the page */ #define VM_ALLOC_IFCACHED 0x0400 /* (ag) Fail if page is not cached */ #define VM_ALLOC_IFNOTCACHED 0x0800 /* (ag) Fail if page is cached */ #define VM_ALLOC_IGN_SBUSY 0x1000 /* (g) Ignore shared busy flag */ #define VM_ALLOC_NODUMP 0x2000 /* (ag) don't include in dump */ #define VM_ALLOC_SBUSY 0x4000 /* (acg) Shared busy the page */ #define VM_ALLOC_NOWAIT 0x8000 /* (g) Do not sleep, return NULL */ #define VM_ALLOC_COUNT_SHIFT 16 #define VM_ALLOC_COUNT(count) ((count) << VM_ALLOC_COUNT_SHIFT) #ifdef M_NOWAIT static inline int malloc2vm_flags(int malloc_flags) { int pflags; KASSERT((malloc_flags & M_USE_RESERVE) == 0 || (malloc_flags & M_NOWAIT) != 0, ("M_USE_RESERVE requires M_NOWAIT")); pflags = (malloc_flags & M_USE_RESERVE) != 0 ? VM_ALLOC_INTERRUPT : VM_ALLOC_SYSTEM; if ((malloc_flags & M_ZERO) != 0) pflags |= VM_ALLOC_ZERO; if ((malloc_flags & M_NODUMP) != 0) pflags |= VM_ALLOC_NODUMP; return (pflags); } #endif void vm_page_busy_downgrade(vm_page_t m); void vm_page_busy_sleep(vm_page_t m, const char *msg); void vm_page_flash(vm_page_t m); void vm_page_hold(vm_page_t mem); void vm_page_unhold(vm_page_t mem); void vm_page_free(vm_page_t m); void vm_page_free_zero(vm_page_t m); void vm_page_activate (vm_page_t); void vm_page_advise(vm_page_t m, int advice); vm_page_t vm_page_alloc (vm_object_t, vm_pindex_t, int); vm_page_t vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req, u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary, vm_memattr_t memattr); vm_page_t vm_page_alloc_freelist(int, int); vm_page_t vm_page_grab (vm_object_t, vm_pindex_t, int); void vm_page_cache(vm_page_t); void vm_page_cache_free(vm_object_t, vm_pindex_t, vm_pindex_t); void vm_page_cache_transfer(vm_object_t, vm_pindex_t, vm_object_t); int vm_page_try_to_cache (vm_page_t); int vm_page_try_to_free (vm_page_t); void vm_page_deactivate (vm_page_t); +void vm_page_deactivate_noreuse(vm_page_t); void vm_page_dequeue(vm_page_t m); void vm_page_dequeue_locked(vm_page_t m); vm_page_t vm_page_find_least(vm_object_t, vm_pindex_t); vm_page_t vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr); void vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr); int vm_page_insert (vm_page_t, vm_object_t, vm_pindex_t); boolean_t vm_page_is_cached(vm_object_t object, vm_pindex_t pindex); vm_page_t vm_page_lookup (vm_object_t, vm_pindex_t); vm_page_t vm_page_next(vm_page_t m); int vm_page_pa_tryrelock(pmap_t, vm_paddr_t, vm_paddr_t *); struct vm_pagequeue *vm_page_pagequeue(vm_page_t m); vm_page_t vm_page_prev(vm_page_t m); boolean_t vm_page_ps_is_valid(vm_page_t m); void vm_page_putfake(vm_page_t m); void vm_page_readahead_finish(vm_page_t m); void vm_page_reference(vm_page_t m); void vm_page_remove (vm_page_t); int vm_page_rename (vm_page_t, vm_object_t, vm_pindex_t); vm_page_t vm_page_replace(vm_page_t mnew, vm_object_t object, vm_pindex_t pindex); void vm_page_requeue(vm_page_t m); void vm_page_requeue_locked(vm_page_t m); int vm_page_sbusied(vm_page_t m); void vm_page_set_valid_range(vm_page_t m, int base, int size); int vm_page_sleep_if_busy(vm_page_t m, const char *msg); vm_offset_t vm_page_startup(vm_offset_t vaddr); void vm_page_sunbusy(vm_page_t m); int vm_page_trysbusy(vm_page_t m); void vm_page_unhold_pages(vm_page_t *ma, int count); boolean_t vm_page_unwire(vm_page_t m, uint8_t queue); void vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr); void vm_page_wire (vm_page_t); void vm_page_xunbusy_hard(vm_page_t m); void vm_page_set_validclean (vm_page_t, int, int); void vm_page_clear_dirty (vm_page_t, int, int); void vm_page_set_invalid (vm_page_t, int, int); int vm_page_is_valid (vm_page_t, int, int); void vm_page_test_dirty (vm_page_t); vm_page_bits_t vm_page_bits(int base, int size); void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid); void vm_page_free_toq(vm_page_t m); void vm_page_zero_idle_wakeup(void); void vm_page_dirty_KBI(vm_page_t m); void vm_page_lock_KBI(vm_page_t m, const char *file, int line); void vm_page_unlock_KBI(vm_page_t m, const char *file, int line); int vm_page_trylock_KBI(vm_page_t m, const char *file, int line); #if defined(INVARIANTS) || defined(INVARIANT_SUPPORT) void vm_page_assert_locked_KBI(vm_page_t m, const char *file, int line); void vm_page_lock_assert_KBI(vm_page_t m, int a, const char *file, int line); #endif #define vm_page_assert_sbusied(m) \ KASSERT(vm_page_sbusied(m), \ ("vm_page_assert_sbusied: page %p not shared busy @ %s:%d", \ (void *)m, __FILE__, __LINE__)); #define vm_page_assert_unbusied(m) \ KASSERT(!vm_page_busied(m), \ ("vm_page_assert_unbusied: page %p busy @ %s:%d", \ (void *)m, __FILE__, __LINE__)); #define vm_page_assert_xbusied(m) \ KASSERT(vm_page_xbusied(m), \ ("vm_page_assert_xbusied: page %p not exclusive busy @ %s:%d", \ (void *)m, __FILE__, __LINE__)); #define vm_page_busied(m) \ ((m)->busy_lock != VPB_UNBUSIED) #define vm_page_sbusy(m) do { \ if (!vm_page_trysbusy(m)) \ panic("%s: page %p failed shared busing", __func__, m); \ } while (0) #define vm_page_tryxbusy(m) \ (atomic_cmpset_acq_int(&m->busy_lock, VPB_UNBUSIED, \ VPB_SINGLE_EXCLUSIVER)) #define vm_page_xbusied(m) \ ((m->busy_lock & VPB_SINGLE_EXCLUSIVER) != 0) #define vm_page_xbusy(m) do { \ if (!vm_page_tryxbusy(m)) \ panic("%s: page %p failed exclusive busing", __func__, \ m); \ } while (0) #define vm_page_xunbusy(m) do { \ if (!atomic_cmpset_rel_int(&(m)->busy_lock, \ VPB_SINGLE_EXCLUSIVER, VPB_UNBUSIED)) \ vm_page_xunbusy_hard(m); \ } while (0) #ifdef INVARIANTS void vm_page_object_lock_assert(vm_page_t m); #define VM_PAGE_OBJECT_LOCK_ASSERT(m) vm_page_object_lock_assert(m) void vm_page_assert_pga_writeable(vm_page_t m, uint8_t bits); #define VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits) \ vm_page_assert_pga_writeable(m, bits) #else #define VM_PAGE_OBJECT_LOCK_ASSERT(m) (void)0 #define VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits) (void)0 #endif /* * We want to use atomic updates for the aflags field, which is 8 bits wide. * However, not all architectures support atomic operations on 8-bit * destinations. In order that we can easily use a 32-bit operation, we * require that the aflags field be 32-bit aligned. */ CTASSERT(offsetof(struct vm_page, aflags) % sizeof(uint32_t) == 0); /* * Clear the given bits in the specified page. */ static inline void vm_page_aflag_clear(vm_page_t m, uint8_t bits) { uint32_t *addr, val; /* * The PGA_REFERENCED flag can only be cleared if the page is locked. */ if ((bits & PGA_REFERENCED) != 0) vm_page_assert_locked(m); /* * Access the whole 32-bit word containing the aflags field with an * atomic update. Parallel non-atomic updates to the other fields * within this word are handled properly by the atomic update. */ addr = (void *)&m->aflags; KASSERT(((uintptr_t)addr & (sizeof(uint32_t) - 1)) == 0, ("vm_page_aflag_clear: aflags is misaligned")); val = bits; #if BYTE_ORDER == BIG_ENDIAN val <<= 24; #endif atomic_clear_32(addr, val); } /* * Set the given bits in the specified page. */ static inline void vm_page_aflag_set(vm_page_t m, uint8_t bits) { uint32_t *addr, val; VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits); /* * Access the whole 32-bit word containing the aflags field with an * atomic update. Parallel non-atomic updates to the other fields * within this word are handled properly by the atomic update. */ addr = (void *)&m->aflags; KASSERT(((uintptr_t)addr & (sizeof(uint32_t) - 1)) == 0, ("vm_page_aflag_set: aflags is misaligned")); val = bits; #if BYTE_ORDER == BIG_ENDIAN val <<= 24; #endif atomic_set_32(addr, val); } /* * vm_page_dirty: * * Set all bits in the page's dirty field. * * The object containing the specified page must be locked if the * call is made from the machine-independent layer. * * See vm_page_clear_dirty_mask(). */ static __inline void vm_page_dirty(vm_page_t m) { /* Use vm_page_dirty_KBI() under INVARIANTS to save memory. */ #if defined(KLD_MODULE) || defined(INVARIANTS) vm_page_dirty_KBI(m); #else m->dirty = VM_PAGE_BITS_ALL; #endif } /* * vm_page_remque: * * If the given page is in a page queue, then remove it from that page * queue. * * The page must be locked. */ static inline void vm_page_remque(vm_page_t m) { if (m->queue != PQ_NONE) vm_page_dequeue(m); } /* * vm_page_undirty: * * Set page to not be dirty. Note: does not clear pmap modify bits */ static __inline void vm_page_undirty(vm_page_t m) { VM_PAGE_OBJECT_LOCK_ASSERT(m); m->dirty = 0; } #endif /* _KERNEL */ #endif /* !_VM_PAGE_ */