Index: head/sys/kern/vfs_bio.c =================================================================== --- head/sys/kern/vfs_bio.c (revision 328913) +++ head/sys/kern/vfs_bio.c (revision 328914) @@ -1,5062 +1,5065 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * 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 #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; struct proc *bufspacedaemonproc; 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 void breada(struct vnode *, daddr_t *, int *, int, struct ucred *, int, void (*)(struct buf *)); static int buf_flush(struct vnode *vp, int); static int buf_recycle(bool); static int buf_scan(bool); 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); static void bufkva_reclaim(vmem_t *, int); static void bufkva_free(struct buf *); static int buf_import(void *, void **, int, int, int); static void buf_release(void *, void **, int); static void maxbcachebuf_adjust(void); #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_RW, &maxbufspace, 0, "Maximum allowed value of bufspace (including metadata)"); 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_RW, &lobufspace, 0, "Minimum amount of buffers we want to have"); long hibufspace; SYSCTL_LONG(_vfs, OID_AUTO, hibufspace, CTLFLAG_RW, &hibufspace, 0, "Maximum allowed value of bufspace (excluding metadata)"); long bufspacethresh; SYSCTL_LONG(_vfs, OID_AUTO, bufspacethresh, CTLFLAG_RW, &bufspacethresh, 0, "Bufspace consumed before waking the daemon to free some"); 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, "Target number of free buffers"); static int hifreebuffers; SYSCTL_INT(_vfs, OID_AUTO, hifreebuffers, CTLFLAG_RW, &hifreebuffers, 0, "Threshold for clean buffer recycling"); 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 acquisition"); 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 numbufallocfails; SYSCTL_INT(_vfs, OID_AUTO, numbufallocfails, CTLFLAG_RW, &numbufallocfails, 0, "Number of times buffer allocations failed"); 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"); int maxbcachebuf = MAXBCACHEBUF; SYSCTL_INT(_vfs, OID_AUTO, maxbcachebuf, CTLFLAG_RDTUN, &maxbcachebuf, 0, "Maximum size of a buffer cache block"); /* * This lock synchronizes access to bd_request. */ static struct mtx_padalign __exclusive_cache_line bdlock; /* * This lock protects the runningbufreq and synchronizes runningbufwakeup and * waitrunningbufspace(). */ static struct mtx_padalign __exclusive_cache_line rbreqlock; /* * Lock that protects needsbuffer and the sleeps/wakeups surrounding it. */ static struct rwlock_padalign __exclusive_cache_line nblock; /* * Lock that protects bdirtywait. */ static struct mtx_padalign __exclusive_cache_line 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/wakeup point for the bufspace daemon. */ static int bufspace_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; /* * 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(), bufspace_wakeup(), bwillwrite(), * getnewbuf(), and getblk(). */ static volatile int needsbuffer; /* * Synchronization for bwillwrite() waiters. */ static int bdirtywait; /* * Definitions for the buffer free lists. */ #define QUEUE_NONE 0 /* on no queue */ #define QUEUE_EMPTY 1 /* empty buffer headers */ #define QUEUE_DIRTY 2 /* B_DELWRI buffers */ #define QUEUE_CLEAN 3 /* non-B_DELWRI buffers */ #define QUEUE_SENTINEL 1024 /* not an queue index, but mark for sentinel */ /* Maximum number of clean buffer queues. */ #define CLEAN_QUEUES 16 /* Configured number of clean queues. */ static int clean_queues; /* Maximum number of buffer queues. */ #define BUFFER_QUEUES (QUEUE_CLEAN + CLEAN_QUEUES) /* 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 /* * Lock for each bufqueue */ static struct mtx_padalign __exclusive_cache_line bqlocks[BUFFER_QUEUES]; /* * per-cpu empty buffer cache. */ uma_zone_t buf_zone; /* * Single global constant for BUF_WMESG, to avoid getting multiple references. * buf_wmesg is referred from macros. */ const char *buf_wmesg = BUF_WMESG; 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 static int bqcleanq(void) { static int nextq; return ((atomic_fetchadd_int(&nextq, 1) % clean_queues) + QUEUE_CLEAN); } static int bqisclean(int qindex) { return (qindex >= QUEUE_CLEAN && qindex < QUEUE_CLEAN + CLEAN_QUEUES); } /* * bqlock: * * Return the appropriate queue lock based on the index. */ static inline struct mtx * bqlock(int qindex) { return (struct mtx *)&bqlocks[qindex]; } /* * 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(); } /* * bufspace_wakeup: * * 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 void bufspace_wakeup(void) { /* * If someone is waiting for bufspace, wake them up. * * Since needsbuffer is set prior to doing an additional queue * scan it is safe to check for the flag prior to acquiring the * lock. The thread that is preparing to scan again before * blocking would discover the buf we released. */ if (needsbuffer) { rw_rlock(&nblock); if (atomic_cmpset_int(&needsbuffer, 1, 0) == 1) wakeup(__DEVOLATILE(void *, &needsbuffer)); rw_runlock(&nblock); } } /* * bufspace_daemonwakeup: * * Wakeup the daemon responsible for freeing clean bufs. */ static void bufspace_daemonwakeup(void) { rw_rlock(&nblock); if (bufspace_request == 0) { bufspace_request = 1; wakeup(&bufspace_request); } rw_runlock(&nblock); } /* * bufspace_adjust: * * Adjust the reported bufspace for a KVA managed buffer, possibly * waking any waiters. */ static void bufspace_adjust(struct buf *bp, int bufsize) { long space; int diff; KASSERT((bp->b_flags & B_MALLOC) == 0, ("bufspace_adjust: malloc buf %p", bp)); diff = bufsize - bp->b_bufsize; if (diff < 0) { atomic_subtract_long(&bufspace, -diff); bufspace_wakeup(); } else { space = atomic_fetchadd_long(&bufspace, diff); /* Wake up the daemon on the transition. */ if (space < bufspacethresh && space + diff >= bufspacethresh) bufspace_daemonwakeup(); } bp->b_bufsize = bufsize; } /* * bufspace_reserve: * * Reserve bufspace before calling allocbuf(). metadata has a * different space limit than data. */ static int bufspace_reserve(int size, bool metadata) { long limit; long space; if (metadata) limit = maxbufspace; else limit = hibufspace; do { space = bufspace; if (space + size > limit) return (ENOSPC); } while (atomic_cmpset_long(&bufspace, space, space + size) == 0); /* Wake up the daemon on the transition. */ if (space < bufspacethresh && space + size >= bufspacethresh) bufspace_daemonwakeup(); return (0); } /* * bufspace_release: * * Release reserved bufspace after bufspace_adjust() has consumed it. */ static void bufspace_release(int size) { atomic_subtract_long(&bufspace, size); bufspace_wakeup(); } /* * bufspace_wait: * * Wait for bufspace, acting as the buf daemon if a locked vnode is * supplied. needsbuffer must be set in a safe fashion prior to * polling for space. The operation must be re-tried on return. */ static void bufspace_wait(struct vnode *vp, int gbflags, int slpflag, int slptimeo) { struct thread *td; int error, fl, norunbuf; if ((gbflags & GB_NOWAIT_BD) != 0) return; td = curthread; rw_wlock(&nblock); while (needsbuffer != 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 == 0) break; } error = rw_sleep(__DEVOLATILE(void *, &needsbuffer), &nblock, (PRIBIO + 4) | slpflag, "newbuf", slptimeo); if (error != 0) break; } rw_wunlock(&nblock); } /* * bufspace_daemon: * * buffer space management daemon. Tries to maintain some marginal * amount of free buffer space so that requesting processes neither * block nor work to reclaim buffers. */ static void bufspace_daemon(void) { for (;;) { kproc_suspend_check(bufspacedaemonproc); /* * Free buffers from the clean queue until we meet our * targets. * * Theory of operation: The buffer cache is most efficient * when some free buffer headers and space are always * available to getnewbuf(). This daemon attempts to prevent * the excessive blocking and synchronization associated * with shortfall. It goes through three phases according * demand: * * 1) The daemon wakes up voluntarily once per-second * during idle periods when the counters are below * the wakeup thresholds (bufspacethresh, lofreebuffers). * * 2) The daemon wakes up as we cross the thresholds * ahead of any potential blocking. This may bounce * slightly according to the rate of consumption and * release. * * 3) The daemon and consumers are starved for working * clean buffers. This is the 'bufspace' sleep below * which will inefficiently trade bufs with bqrelse * until we return to condition 2. */ while (bufspace > lobufspace || numfreebuffers < hifreebuffers) { if (buf_recycle(false) != 0) { atomic_set_int(&needsbuffer, 1); if (buf_recycle(false) != 0) { rw_wlock(&nblock); if (needsbuffer) rw_sleep(__DEVOLATILE(void *, &needsbuffer), &nblock, PRIBIO|PDROP, "bufspace", hz/10); else rw_wunlock(&nblock); } } maybe_yield(); } /* * Re-check our limits under the exclusive nblock. */ rw_wlock(&nblock); if (bufspace < bufspacethresh && numfreebuffers > lofreebuffers) { bufspace_request = 0; rw_sleep(&bufspace_request, &nblock, PRIBIO|PDROP, "-", hz); } else rw_wunlock(&nblock); } } static struct kproc_desc bufspace_kp = { "bufspacedaemon", bufspace_daemon, &bufspacedaemonproc }; SYSINIT(bufspacedaemon, SI_SUB_KTHREAD_BUF, SI_ORDER_FIRST, kproc_start, &bufspace_kp); /* * 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); 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(); } /* * 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); } /* * Adjust the maxbcachbuf tunable. */ static void maxbcachebuf_adjust(void) { int i; /* * maxbcachebuf must be a power of 2 >= MAXBSIZE. */ i = 2; while (i * 2 <= maxbcachebuf) i *= 2; maxbcachebuf = i; if (maxbcachebuf < MAXBSIZE) maxbcachebuf = MAXBSIZE; if (maxbcachebuf > MAXPHYS) maxbcachebuf = MAXPHYS; if (bootverbose != 0 && maxbcachebuf != MAXBCACHEBUF) printf("maxbcachebuf=%d\n", maxbcachebuf); } /* * 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); maxbcachebuf_adjust(); /* * 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; /* * Artificially 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; KASSERT(maxbcachebuf >= MAXBSIZE, ("maxbcachebuf (%d) must be >= MAXBSIZE (%d)\n", maxbcachebuf, MAXBSIZE)); mtx_init(&bqlocks[QUEUE_DIRTY], "bufq dirty lock", NULL, MTX_DEF); mtx_init(&bqlocks[QUEUE_EMPTY], "bufq empty lock", NULL, MTX_DEF); for (i = QUEUE_CLEAN; i < QUEUE_CLEAN + CLEAN_QUEUES; i++) mtx_init(&bqlocks[i], "bufq clean 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; 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 metadata. hibufspace is the nominal maximum * used by most other requests. The differential is required to * ensure that metadata deadlocks don't occur. * * maxbufspace is based on BKVASIZE. Allocating buffers larger then * this may result in KVM fragmentation which is not handled optimally * by the system. XXX This is less true with vmem. We could use * PAGE_SIZE. */ maxbufspace = (long)nbuf * BKVASIZE; hibufspace = lmax(3 * maxbufspace / 4, maxbufspace - maxbcachebuf * 10); lobufspace = (hibufspace / 20) * 19; /* 95% */ bufspacethresh = lobufspace + (hibufspace - lobufspace) / 2; /* * 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 occurring 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; /* * lofreebuffers should be sufficient to avoid stalling waiting on * buf headers under heavy utilization. The bufs in per-cpu caches * are counted as free but will be unavailable to threads executing * on other cpus. * * hifreebuffers is the free target for the bufspace daemon. This * should be set appropriately to limit work per-iteration. */ lofreebuffers = MIN((nbuf / 25) + (20 * mp_ncpus), 128 * mp_ncpus); hifreebuffers = (3 * lofreebuffers) / 2; numfreebuffers = nbuf; /* Setup the kva and free list allocators. */ vmem_set_reclaim(buffer_arena, bufkva_reclaim); buf_zone = uma_zcache_create("buf free cache", sizeof(struct buf), NULL, NULL, NULL, NULL, buf_import, buf_release, NULL, 0); /* * Size the clean queue according to the amount of buffer space. * One queue per-256mb up to the max. More queues gives better * concurrency but less accurate LRU. */ clean_queues = MIN(howmany(maxbufspace, 256*1024*1024), CLEAN_QUEUES); } #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) == 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 == NULL) 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; if (qindex != QUEUE_EMPTY) { BUF_ASSERT_XLOCKED(bp); } /* * Stick to the same clean queue for the lifetime of the buf to * limit locking below. Otherwise pick ont sequentially. */ if (qindex == QUEUE_CLEAN) { if (bqisclean(bp->b_qindex)) qindex = bp->b_qindex; else qindex = bqcleanq(); } /* * Handle delayed bremfree() processing. */ nlock = bqlock(qindex); 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); } /* * buf_free: * * Free a buffer to the buf zone once it no longer has valid contents. */ static void buf_free(struct buf *bp) { if (bp->b_flags & B_REMFREE) bremfreef(bp); if (bp->b_vflags & BV_BKGRDINPROG) panic("losing buffer 1"); 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); bufkva_free(bp); BUF_UNLOCK(bp); uma_zfree(buf_zone, bp); atomic_add_int(&numfreebuffers, 1); bufspace_wakeup(); } /* * buf_import: * * Import bufs into the uma cache from the buf list. The system still * expects a static array of bufs and much of the synchronization * around bufs assumes type stable storage. As a result, UMA is used * only as a per-cpu cache of bufs still maintained on a global list. */ static int buf_import(void *arg, void **store, int cnt, int domain, int flags) { struct buf *bp; int i; mtx_lock(&bqlocks[QUEUE_EMPTY]); for (i = 0; i < cnt; i++) { bp = TAILQ_FIRST(&bufqueues[QUEUE_EMPTY]); if (bp == NULL) break; bremfreel(bp); store[i] = bp; } mtx_unlock(&bqlocks[QUEUE_EMPTY]); return (i); } /* * buf_release: * * Release bufs from the uma cache back to the buffer queues. */ static void buf_release(void *arg, void **store, int cnt) { int i; for (i = 0; i < cnt; i++) binsfree(store[i], QUEUE_EMPTY); } /* * buf_alloc: * * Allocate an empty buffer header. */ static struct buf * buf_alloc(void) { struct buf *bp; bp = uma_zalloc(buf_zone, M_NOWAIT); if (bp == NULL) { bufspace_daemonwakeup(); atomic_add_int(&numbufallocfails, 1); return (NULL); } /* * Wake-up the bufspace daemon on transition. */ if (atomic_fetchadd_int(&numfreebuffers, -1) == lofreebuffers) bufspace_daemonwakeup(); if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL) != 0) panic("getnewbuf_empty: Locked buf %p on free queue.", bp); KASSERT(bp->b_vp == NULL, ("bp: %p still has vnode %p.", bp, bp->b_vp)); KASSERT((bp->b_flags & (B_DELWRI | B_NOREUSE)) == 0, ("invalid buffer %p flags %#x", bp, bp->b_flags)); 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)); KASSERT(bp->b_kvasize == 0, ("bp: %p still has kva\n", bp)); KASSERT(bp->b_bufsize == 0, ("bp: %p still has bufspace\n", bp)); bp->b_flags = 0; bp->b_ioflags = 0; bp->b_xflags = 0; 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_data = bp->b_kvabase = unmapped_buf; bp->b_fsprivate1 = NULL; bp->b_fsprivate2 = NULL; bp->b_fsprivate3 = NULL; LIST_INIT(&bp->b_dep); return (bp); } /* * buf_qrecycle: * * Free a buffer from the given bufqueue. kva controls whether the * freed buf must own some kva resources. This is used for * defragmenting. */ static int buf_qrecycle(int qindex, bool kva) { struct buf *bp, *nbp; if (kva) atomic_add_int(&bufdefragcnt, 1); nbp = NULL; mtx_lock(&bqlocks[qindex]); nbp = TAILQ_FIRST(&bufqueues[qindex]); /* * Run scan, possibly freeing data and/or kva mappings on the fly * depending. */ while ((bp = nbp) != NULL) { /* * Calculate next bp (we can only use it if we do not * release the bqlock). */ nbp = TAILQ_NEXT(bp, b_freelist); /* * If we are defragging then we need a buffer with * some kva to reclaim. */ if (kva && bp->b_kvasize == 0) continue; if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL) != 0) continue; /* * Skip buffers with background writes in progress. */ if ((bp->b_vflags & BV_BKGRDINPROG) != 0) { BUF_UNLOCK(bp); continue; } KASSERT(bp->b_qindex == qindex, ("getnewbuf: inconsistent queue %d bp %p", qindex, bp)); /* * NOTE: nbp is now entirely invalid. We can only restart * the scan from this point on. */ bremfreel(bp); mtx_unlock(&bqlocks[qindex]); /* * Requeue the background write buffer with error and * restart the scan. */ if ((bp->b_vflags & BV_BKGRDERR) != 0) { bqrelse(bp); mtx_lock(&bqlocks[qindex]); nbp = TAILQ_FIRST(&bufqueues[qindex]); continue; } bp->b_flags |= B_INVAL; brelse(bp); return (0); } mtx_unlock(&bqlocks[qindex]); return (ENOBUFS); } /* * buf_recycle: * * Iterate through all clean queues until we find a buf to recycle or * exhaust the search. */ static int buf_recycle(bool kva) { int qindex, first_qindex; qindex = first_qindex = bqcleanq(); do { if (buf_qrecycle(qindex, kva) == 0) return (0); if (++qindex == QUEUE_CLEAN + clean_queues) qindex = QUEUE_CLEAN; } while (qindex != first_qindex); return (ENOBUFS); } /* * buf_scan: * * Scan the clean queues looking for a buffer to recycle. needsbuffer * is set on failure so that the caller may optionally bufspace_wait() * in a race-free fashion. */ static int buf_scan(bool defrag) { int error; /* * To avoid heavy synchronization and wakeup races we set * needsbuffer and re-poll before failing. This ensures that * no frees can be missed between an unsuccessful poll and * going to sleep in a synchronized fashion. */ if ((error = buf_recycle(defrag)) != 0) { atomic_set_int(&needsbuffer, 1); bufspace_daemonwakeup(); error = buf_recycle(defrag); } if (error == 0) atomic_add_int(&getnewbufrestarts, 1); return (error); } /* * 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; } /* * 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)); if (bp->b_qindex != QUEUE_EMPTY) { 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; bp->b_flags &= ~B_REMFREE; } /* * bufkva_free: * * Free the kva allocation for a buffer. * */ static void bufkva_free(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; } /* * bufkva_alloc: * * Allocate the buffer KVA and set b_kvasize and b_kvabase. */ static int bufkva_alloc(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__)); bufkva_free(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. */ 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); } /* * bufkva_reclaim: * * Reclaim buffer kva by freeing buffers holding kva. This is a vmem * callback that fires to avoid returning failure. */ static void bufkva_reclaim(vmem_t *vmem, int flags) { int i; for (i = 0; i < 5; i++) if (buf_scan(true) != 0) break; return; } /* * 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. */ static void breada(struct vnode * vp, daddr_t * rablkno, int * rabsize, int cnt, struct ucred * cred, int flags, void (*ckhashfunc)(struct buf *)) { 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) { brelse(rabp); continue; } if (!TD_IS_IDLETHREAD(curthread)) { #ifdef RACCT if (racct_enable) { PROC_LOCK(curproc); racct_add_buf(curproc, rabp, 0); PROC_UNLOCK(curproc); } #endif /* RACCT */ curthread->td_ru.ru_inblock++; } rabp->b_flags |= B_ASYNC; rabp->b_flags &= ~B_INVAL; if ((flags & GB_CKHASH) != 0) { rabp->b_flags |= B_CKHASH; rabp->b_ckhashcalc = ckhashfunc; } 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); } } /* * 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. * * Always return a NULL buffer pointer (in bpp) when returning an error. */ int breadn_flags(struct vnode *vp, daddr_t blkno, int size, daddr_t *rablkno, int *rabsize, int cnt, struct ucred *cred, int flags, void (*ckhashfunc)(struct buf *), struct buf **bpp) { struct buf *bp; int readwait, rv; 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 */ readwait = 0; if ((bp->b_flags & B_CACHE) == 0) { if (!TD_IS_IDLETHREAD(curthread)) { #ifdef RACCT if (racct_enable) { PROC_LOCK(curproc); racct_add_buf(curproc, bp, 0); PROC_UNLOCK(curproc); } #endif /* RACCT */ curthread->td_ru.ru_inblock++; } bp->b_iocmd = BIO_READ; bp->b_flags &= ~B_INVAL; if ((flags & GB_CKHASH) != 0) { bp->b_flags |= B_CKHASH; bp->b_ckhashcalc = ckhashfunc; } 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; } /* * Attempt to initiate asynchronous I/O on read-ahead blocks. */ breada(vp, rablkno, rabsize, cnt, cred, flags, ckhashfunc); rv = 0; if (readwait) { rv = bufwait(bp); if (rv != 0) { brelse(bp); *bpp = NULL; } } 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); 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)) { #ifdef RACCT if (racct_enable) { PROC_LOCK(curproc); racct_add_buf(curproc, bp, 1); PROC_UNLOCK(curproc); } #endif /* RACCT */ curthread->td_ru.ru_oublock++; } if (oldflags & B_ASYNC) BUF_KERNPROC(bp); bp->b_iooffset = dbtob(bp->b_blkno); buf_track(bp, __func__); 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); } buf_track(bp, __func__); /* * 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; /* * Many functions erroneously call brelse with a NULL bp under rare * error conditions. Simply return when called with a NULL bp. */ if (bp == NULL) return; 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 != ENXIO || !LIST_EMPTY(&bp->b_dep)) && !(bp->b_flags & B_INVAL)) { /* * Failed write, redirty. All errors except ENXIO (which * means the device is gone) are treated as being * transient. * * XXX Treating EIO as transient is not correct; the * contract with the local storage device drivers is that * they will only return EIO once the I/O is no longer * retriable. Network I/O also respects this through the * guarantees of TCP and/or the internal retries of NFS. * ENOMEM might be transient, but we also have no way of * knowing when its ok to retry/reschedule. In general, * this entire case should be made obsolete through better * error handling/recovery and resource scheduling. * * Do this also for buffers that failed with ENXIO, but have * non-empty dependencies - the soft updates code might need * to access the buffer to untangle them. * * Must clear BIO_ERROR to prevent pages from being scrapped. */ 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 read I/O, or we were asked to free or not * cache the buffer, or we failed to write to a device that's * no longer present. */ 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 || (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); } buf_track(bp, __func__); /* buffers with no memory */ if (bp->b_bufsize == 0) { buf_free(bp); return; } /* buffers with junk contents */ 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); if (qindex == QUEUE_CLEAN) bufspace_wakeup(); } /* * 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)); qindex = QUEUE_NONE; 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: buf_track(bp, __func__); /* unlock */ BUF_UNLOCK(bp); if (qindex == QUEUE_CLEAN) bufspace_wakeup(); } /* * 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 i, iosize, resid; bool bogus; 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 = false; iosize = bp->b_bcount - bp->b_resid; VM_OBJECT_WLOCK(obj); for (i = 0; i < bp->b_npages; i++) { 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 = true; 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); foff = (foff + PAGE_SIZE) & ~(off_t)PAGE_MASK; iosize -= resid; } vm_object_pip_wakeupn(obj, bp->b_npages); 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) { /* * 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. */ 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", true); 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); if (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(). */ (void)vm_page_grab_pages(obj, OFF_TO_IDX(bp->b_offset) + bp->b_npages, VM_ALLOC_SYSTEM | VM_ALLOC_IGN_SBUSY | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED, &bp->b_pages[bp->b_npages], desiredpages - bp->b_npages); bp->b_npages = desiredpages; } /* * 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); } /* * getnewbuf_kva: * * Allocate KVA for an empty buf header according to gbflags. */ static int getnewbuf_kva(struct buf *bp, int gbflags, int maxsize) { if ((gbflags & (GB_UNMAPPED | GB_KVAALLOC)) != GB_UNMAPPED) { /* * In order to keep fragmentation sane we only allocate kva * in BKVASIZE chunks. XXX with vmem we can do page size. */ maxsize = (maxsize + BKVAMASK) & ~BKVAMASK; if (maxsize != bp->b_kvasize && bufkva_alloc(bp, maxsize, gbflags)) return (ENOSPC); } return (0); } /* * getnewbuf: * * Find and initialize a new buffer header, freeing up existing buffers * in the bufqueues as necessary. The new buffer is returned locked. * * 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 ) * * The caller is responsible for releasing the reserved bufspace after * allocbuf() is called. */ static struct buf * getnewbuf(struct vnode *vp, int slpflag, int slptimeo, int maxsize, int gbflags) { struct buf *bp; bool metadata, reserved; bp = NULL; 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); if (vp == NULL || (vp->v_vflag & (VV_MD | VV_SYSTEM)) != 0 || vp->v_type == VCHR) metadata = true; else metadata = false; atomic_add_int(&getnewbufcalls, 1); reserved = false; do { if (reserved == false && bufspace_reserve(maxsize, metadata) != 0) continue; reserved = true; if ((bp = buf_alloc()) == NULL) continue; if (getnewbuf_kva(bp, gbflags, maxsize) == 0) return (bp); break; } while(buf_scan(false) == 0); if (reserved) atomic_subtract_long(&bufspace, maxsize); if (bp != NULL) { bp->b_flags |= B_INVAL; brelse(bp); } bufspace_wait(vp, gbflags, slpflag, slptimeo); return (NULL); } /* * 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(&bqlocks[queue]); TAILQ_INSERT_HEAD(&bufqueues[queue], sentinel, b_freelist); mtx_unlock(&bqlocks[queue]); while (flushed != target) { maybe_yield(); mtx_lock(&bqlocks[queue]); 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(&bqlocks[queue]); 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(&bqlocks[queue]); continue; } error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL); mtx_unlock(&bqlocks[queue]); if (error != 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) != 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(&bqlocks[queue]); TAILQ_REMOVE(&bufqueues[queue], sentinel, b_freelist); mtx_unlock(&bqlocks[queue]); 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) { 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); while (bufkva_alloc(bp, maxsize, gbflags) != 0) { 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); bufspace_wait(bp->b_vp, gbflags, 0, 0); } 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 successful 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) { 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 inconsistent 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, maxsize, flags); if (bp == NULL) { if (slpflag || slptimeo) return NULL; /* * XXX This is here until the sleep path is diagnosed * enough to work under very low memory conditions. * * There's an issue on low memory, 4BSD+non-preempt * systems (eg MIPS routers with 32MB RAM) where buffer * exhaustion occurs without sleeping for buffer * reclaimation. This just sticks in a loop and * constantly attempts to allocate a buffer, which * hits exhaustion and tries to wakeup bufdaemon. * This never happens because we never yield. * * The real solution is to identify and fix these cases * so we aren't effectively busy-waiting in a loop * until the reclaimation path has cycles to run. */ kern_yield(PRI_USER); 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); bufspace_release(maxsize); 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); bufspace_release(maxsize); bp->b_flags &= ~B_DONE; } CTR4(KTR_BUF, "getblk(%p, %ld, %d) = %p", vp, (long)blkno, size, bp); BUF_ASSERT_HELD(bp); end: buf_track(bp, __func__); 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, maxsize, flags)) == NULL) { if ((flags & GB_NOWAIT_BD) && (curthread->td_pflags & TDP_BUFNEED) != 0) return (NULL); } allocbuf(bp, size); bufspace_release(maxsize); 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); bufspace_adjust(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); } bufspace_adjust(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 inconsistent 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 = roundup2(size, DEV_BSIZE); 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); bufspace_adjust(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; biotrack(bp, __func__); 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, atop(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 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); } #if defined(BUF_TRACKING) || defined(FULL_BUF_TRACKING) void biotrack_buf(struct bio *bp, const char *location) { buf_track(bp->bio_track_bp, location); } #endif /* * 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 occurred, 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 existence * in the biodone routine. */ void bufdone(struct buf *bp) { struct bufobj *dropobj; void (*biodone)(struct buf *); buf_track(bp, __func__); 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; - else if ((bp->b_flags & B_CKHASH) != 0) { - KASSERT(buf_mapped(bp), ("biodone: bp %p not mapped", bp)); - (*bp->b_ckhashcalc)(bp); - } /* 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 * occurred. 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); + } + if ((bp->b_flags & B_CKHASH) != 0) { + KASSERT(bp->b_iocmd == BIO_READ, + ("bufdone_finish: b_iocmd %d not BIO_READ", bp->b_iocmd)); + KASSERT(buf_mapped(bp), + ("bufdone_finish: bp %p not mapped", bp)); + (*bp->b_ckhashcalc)(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 * consistent. */ 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_page_sunbusy(m); } vm_object_pip_wakeupn(obj, bp->b_npages); 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 boundary 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", true); 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 * inconsistent. * * Since I/O has not been initiated yet, certain buffer flags * such as BIO_ERROR or B_INVAL may be in an inconsistent state * and should be ignored. */ void vfs_busy_pages(struct buf *bp, int clear_modify) { vm_object_t obj; vm_ooffset_t foff; vm_page_t m; int i; bool bogus; 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 = false; 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 = true; } 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; } } } /* * Update buffer flags based on I/O request parameters, optionally releasing the * buffer. If it's VMIO or direct I/O, the buffer pages are released to the VM, * where they may be placed on a page queue (VMIO) or freed immediately (direct * I/O). Otherwise the buffer is released to the cache. */ static void b_io_dismiss(struct buf *bp, int ioflag, bool release) { KASSERT((ioflag & IO_NOREUSE) == 0 || (ioflag & IO_VMIO) != 0, ("buf %p non-VMIO noreuse", bp)); if ((ioflag & IO_DIRECT) != 0) bp->b_flags |= B_DIRECT; if ((ioflag & (IO_VMIO | IO_DIRECT)) != 0 && LIST_EMPTY(&bp->b_dep)) { bp->b_flags |= B_RELBUF; if ((ioflag & IO_NOREUSE) != 0) bp->b_flags |= B_NOREUSE; if (release) brelse(bp); } else if (release) bqrelse(bp); } void vfs_bio_brelse(struct buf *bp, int ioflag) { b_io_dismiss(bp, ioflag, true); } void vfs_bio_set_flags(struct buf *bp, int ioflag) { b_io_dismiss(bp, ioflag, false); } /* * 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++) { /* * 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) | VM_ALLOC_WAITOK); 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; p->wire_count--; vm_page_free(p); } atomic_subtract_int(&vm_cnt.v_wire_count, bp->b_npages - newnpages); 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(bo2vnode(bo), 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); } /* * 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; } } /* * The MIPS pmap code currently doesn't handle aliased pages. * The VIPT caches may not handle page aliasing themselves, leading * to data corruption. * * As such, this code makes a system extremely unhappy if said * system doesn't support unaliasing the above situation in hardware. * Some "recent" systems (eg some mips24k/mips74k cores) don't enable * this feature at build time, so it has to be handled in software. * * Once the MIPS pmap/cache code grows to support this function on * earlier chips, it should be flipped back off. */ #ifdef __mips__ static int buf_pager_relbuf = 1; #else static int buf_pager_relbuf = 0; #endif SYSCTL_INT(_vfs, OID_AUTO, buf_pager_relbuf, CTLFLAG_RWTUN, &buf_pager_relbuf, 0, "Make buffer pager release buffers after reading"); /* * The buffer pager. It uses buffer reads to validate pages. * * In contrast to the generic local pager from vm/vnode_pager.c, this * pager correctly and easily handles volumes where the underlying * device block size is greater than the machine page size. The * buffer cache transparently extends the requested page run to be * aligned at the block boundary, and does the necessary bogus page * replacements in the addends to avoid obliterating already valid * pages. * * The only non-trivial issue is that the exclusive busy state for * pages, which is assumed by the vm_pager_getpages() interface, is * incompatible with the VMIO buffer cache's desire to share-busy the * pages. This function performs a trivial downgrade of the pages' * state before reading buffers, and a less trivial upgrade from the * shared-busy to excl-busy state after the read. */ int vfs_bio_getpages(struct vnode *vp, vm_page_t *ma, int count, int *rbehind, int *rahead, vbg_get_lblkno_t get_lblkno, vbg_get_blksize_t get_blksize) { vm_page_t m; vm_object_t object; struct buf *bp; struct mount *mp; daddr_t lbn, lbnp; vm_ooffset_t la, lb, poff, poffe; long bsize; int bo_bs, br_flags, error, i, pgsin, pgsin_a, pgsin_b; bool redo, lpart; object = vp->v_object; mp = vp->v_mount; la = IDX_TO_OFF(ma[count - 1]->pindex); if (la >= object->un_pager.vnp.vnp_size) return (VM_PAGER_BAD); /* * Change the meaning of la from where the last requested page starts * to where it ends, because that's the end of the requested region * and the start of the potential read-ahead region. */ la += PAGE_SIZE; lpart = la > object->un_pager.vnp.vnp_size; bo_bs = get_blksize(vp, get_lblkno(vp, IDX_TO_OFF(ma[0]->pindex))); /* * Calculate read-ahead, behind and total pages. */ pgsin = count; lb = IDX_TO_OFF(ma[0]->pindex); pgsin_b = OFF_TO_IDX(lb - rounddown2(lb, bo_bs)); pgsin += pgsin_b; if (rbehind != NULL) *rbehind = pgsin_b; pgsin_a = OFF_TO_IDX(roundup2(la, bo_bs) - la); if (la + IDX_TO_OFF(pgsin_a) >= object->un_pager.vnp.vnp_size) pgsin_a = OFF_TO_IDX(roundup2(object->un_pager.vnp.vnp_size, PAGE_SIZE) - la); pgsin += pgsin_a; if (rahead != NULL) *rahead = pgsin_a; VM_CNT_INC(v_vnodein); VM_CNT_ADD(v_vnodepgsin, pgsin); br_flags = (mp != NULL && (mp->mnt_kern_flag & MNTK_UNMAPPED_BUFS) != 0) ? GB_UNMAPPED : 0; VM_OBJECT_WLOCK(object); again: for (i = 0; i < count; i++) vm_page_busy_downgrade(ma[i]); VM_OBJECT_WUNLOCK(object); lbnp = -1; for (i = 0; i < count; i++) { m = ma[i]; /* * Pages are shared busy and the object lock is not * owned, which together allow for the pages' * invalidation. The racy test for validity avoids * useless creation of the buffer for the most typical * case when invalidation is not used in redo or for * parallel read. The shared->excl upgrade loop at * the end of the function catches the race in a * reliable way (protected by the object lock). */ if (m->valid == VM_PAGE_BITS_ALL) continue; poff = IDX_TO_OFF(m->pindex); poffe = MIN(poff + PAGE_SIZE, object->un_pager.vnp.vnp_size); for (; poff < poffe; poff += bsize) { lbn = get_lblkno(vp, poff); if (lbn == lbnp) goto next_page; lbnp = lbn; bsize = get_blksize(vp, lbn); error = bread_gb(vp, lbn, bsize, curthread->td_ucred, br_flags, &bp); if (error != 0) goto end_pages; if (LIST_EMPTY(&bp->b_dep)) { /* * Invalidation clears m->valid, but * may leave B_CACHE flag if the * buffer existed at the invalidation * time. In this case, recycle the * buffer to do real read on next * bread() after redo. * * Otherwise B_RELBUF is not strictly * necessary, enable to reduce buf * cache pressure. */ if (buf_pager_relbuf || m->valid != VM_PAGE_BITS_ALL) bp->b_flags |= B_RELBUF; bp->b_flags &= ~B_NOCACHE; brelse(bp); } else { bqrelse(bp); } } KASSERT(1 /* racy, enable for debugging */ || m->valid == VM_PAGE_BITS_ALL || i == count - 1, ("buf %d %p invalid", i, m)); if (i == count - 1 && lpart) { VM_OBJECT_WLOCK(object); if (m->valid != 0 && m->valid != VM_PAGE_BITS_ALL) vm_page_zero_invalid(m, TRUE); VM_OBJECT_WUNLOCK(object); } next_page:; } end_pages: VM_OBJECT_WLOCK(object); redo = false; for (i = 0; i < count; i++) { vm_page_sunbusy(ma[i]); ma[i] = vm_page_grab(object, ma[i]->pindex, VM_ALLOC_NORMAL); /* * Since the pages were only sbusy while neither the * buffer nor the object lock was held by us, or * reallocated while vm_page_grab() slept for busy * relinguish, they could have been invalidated. * Recheck the valid bits and re-read as needed. * * Note that the last page is made fully valid in the * read loop, and partial validity for the page at * index count - 1 could mean that the page was * invalidated or removed, so we must restart for * safety as well. */ if (ma[i]->valid != VM_PAGE_BITS_ALL) redo = true; } if (redo && error == 0) goto again; VM_OBJECT_WUNLOCK(object); return (error != 0 ? VM_PAGER_ERROR : VM_PAGER_OK); } #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; #ifdef FULL_BUF_TRACKING uint32_t i, j; #endif 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]; if (m != NULL) db_printf("(%p, 0x%lx, 0x%lx)", m->object, (u_long)m->pindex, (u_long)VM_PAGE_TO_PHYS(m)); else db_printf("( ??? )"); if ((i + 1) < bp->b_npages) db_printf(","); } db_printf("\n"); } #if defined(FULL_BUF_TRACKING) db_printf("b_io_tracking: b_io_tcnt = %u\n", bp->b_io_tcnt); i = bp->b_io_tcnt % BUF_TRACKING_SIZE; for (j = 1; j <= BUF_TRACKING_SIZE; j++) { if (bp->b_io_tracking[BUF_TRACKING_ENTRY(i - j)] == NULL) continue; db_printf(" %2u: %s\n", j, bp->b_io_tracking[BUF_TRACKING_ENTRY(i - j)]); } #elif defined(BUF_TRACKING) db_printf("b_io_tracking: %s\n", bp->b_io_tracking); #endif 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"); if (db_pager_quit) break; } } } 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_qindex == QUEUE_EMPTY) 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/ufs/ffs/ffs_alloc.c =================================================================== --- head/sys/ufs/ffs/ffs_alloc.c (revision 328913) +++ head/sys/ufs/ffs/ffs_alloc.c (revision 328914) @@ -1,3268 +1,3277 @@ /*- * SPDX-License-Identifier: (BSD-2-Clause-FreeBSD AND BSD-3-Clause) * * Copyright (c) 2002 Networks Associates Technology, Inc. * All rights reserved. * * This software was developed for the FreeBSD Project by Marshall * Kirk McKusick and Network Associates Laboratories, the Security * Research Division of Network Associates, Inc. under DARPA/SPAWAR * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS * research program * * 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. * * 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. * 3. 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. * * @(#)ffs_alloc.c 8.18 (Berkeley) 5/26/95 */ #include __FBSDID("$FreeBSD$"); #include "opt_quota.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 typedef ufs2_daddr_t allocfcn_t(struct inode *ip, u_int cg, ufs2_daddr_t bpref, int size, int rsize); static ufs2_daddr_t ffs_alloccg(struct inode *, u_int, ufs2_daddr_t, int, int); static ufs2_daddr_t ffs_alloccgblk(struct inode *, struct buf *, ufs2_daddr_t, int); static void ffs_blkfree_cg(struct ufsmount *, struct fs *, struct vnode *, ufs2_daddr_t, long, ino_t, struct workhead *); static void ffs_blkfree_trim_completed(struct bio *); static void ffs_blkfree_trim_task(void *ctx, int pending __unused); #ifdef INVARIANTS static int ffs_checkblk(struct inode *, ufs2_daddr_t, long); #endif static ufs2_daddr_t ffs_clusteralloc(struct inode *, u_int, ufs2_daddr_t, int); static ino_t ffs_dirpref(struct inode *); static ufs2_daddr_t ffs_fragextend(struct inode *, u_int, ufs2_daddr_t, int, int); static ufs2_daddr_t ffs_hashalloc (struct inode *, u_int, ufs2_daddr_t, int, int, allocfcn_t *); static ufs2_daddr_t ffs_nodealloccg(struct inode *, u_int, ufs2_daddr_t, int, int); static ufs1_daddr_t ffs_mapsearch(struct fs *, struct cg *, ufs2_daddr_t, int); static int ffs_reallocblks_ufs1(struct vop_reallocblks_args *); static int ffs_reallocblks_ufs2(struct vop_reallocblks_args *); static void ffs_ckhash_cg(struct buf *); /* * Allocate a block in the filesystem. * * The size of the requested block is given, which must be some * multiple of fs_fsize and <= fs_bsize. * A preference may be optionally specified. If a preference is given * the following hierarchy is used to allocate a block: * 1) allocate the requested block. * 2) allocate a rotationally optimal block in the same cylinder. * 3) allocate a block in the same cylinder group. * 4) quadradically rehash into other cylinder groups, until an * available block is located. * If no block preference is given the following hierarchy is used * to allocate a block: * 1) allocate a block in the cylinder group that contains the * inode for the file. * 2) quadradically rehash into other cylinder groups, until an * available block is located. */ int ffs_alloc(ip, lbn, bpref, size, flags, cred, bnp) struct inode *ip; ufs2_daddr_t lbn, bpref; int size, flags; struct ucred *cred; ufs2_daddr_t *bnp; { struct fs *fs; struct ufsmount *ump; ufs2_daddr_t bno; u_int cg, reclaimed; static struct timeval lastfail; static int curfail; int64_t delta; #ifdef QUOTA int error; #endif *bnp = 0; ump = ITOUMP(ip); fs = ump->um_fs; mtx_assert(UFS_MTX(ump), MA_OWNED); #ifdef INVARIANTS if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) { printf("dev = %s, bsize = %ld, size = %d, fs = %s\n", devtoname(ump->um_dev), (long)fs->fs_bsize, size, fs->fs_fsmnt); panic("ffs_alloc: bad size"); } if (cred == NOCRED) panic("ffs_alloc: missing credential"); #endif /* INVARIANTS */ reclaimed = 0; retry: #ifdef QUOTA UFS_UNLOCK(ump); error = chkdq(ip, btodb(size), cred, 0); if (error) return (error); UFS_LOCK(ump); #endif if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0) goto nospace; if (priv_check_cred(cred, PRIV_VFS_BLOCKRESERVE, 0) && freespace(fs, fs->fs_minfree) - numfrags(fs, size) < 0) goto nospace; if (bpref >= fs->fs_size) bpref = 0; if (bpref == 0) cg = ino_to_cg(fs, ip->i_number); else cg = dtog(fs, bpref); bno = ffs_hashalloc(ip, cg, bpref, size, size, ffs_alloccg); if (bno > 0) { delta = btodb(size); DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + delta); if (flags & IO_EXT) ip->i_flag |= IN_CHANGE; else ip->i_flag |= IN_CHANGE | IN_UPDATE; *bnp = bno; return (0); } nospace: #ifdef QUOTA UFS_UNLOCK(ump); /* * Restore user's disk quota because allocation failed. */ (void) chkdq(ip, -btodb(size), cred, FORCE); UFS_LOCK(ump); #endif if (reclaimed == 0 && (flags & IO_BUFLOCKED) == 0) { reclaimed = 1; softdep_request_cleanup(fs, ITOV(ip), cred, FLUSH_BLOCKS_WAIT); goto retry; } UFS_UNLOCK(ump); if (reclaimed > 0 && ppsratecheck(&lastfail, &curfail, 1)) { ffs_fserr(fs, ip->i_number, "filesystem full"); uprintf("\n%s: write failed, filesystem is full\n", fs->fs_fsmnt); } return (ENOSPC); } /* * Reallocate a fragment to a bigger size * * The number and size of the old block is given, and a preference * and new size is also specified. The allocator attempts to extend * the original block. Failing that, the regular block allocator is * invoked to get an appropriate block. */ int ffs_realloccg(ip, lbprev, bprev, bpref, osize, nsize, flags, cred, bpp) struct inode *ip; ufs2_daddr_t lbprev; ufs2_daddr_t bprev; ufs2_daddr_t bpref; int osize, nsize, flags; struct ucred *cred; struct buf **bpp; { struct vnode *vp; struct fs *fs; struct buf *bp; struct ufsmount *ump; u_int cg, request, reclaimed; int error, gbflags; ufs2_daddr_t bno; static struct timeval lastfail; static int curfail; int64_t delta; vp = ITOV(ip); ump = ITOUMP(ip); fs = ump->um_fs; bp = NULL; gbflags = (flags & BA_UNMAPPED) != 0 ? GB_UNMAPPED : 0; mtx_assert(UFS_MTX(ump), MA_OWNED); #ifdef INVARIANTS if (vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) panic("ffs_realloccg: allocation on suspended filesystem"); if ((u_int)osize > fs->fs_bsize || fragoff(fs, osize) != 0 || (u_int)nsize > fs->fs_bsize || fragoff(fs, nsize) != 0) { printf( "dev = %s, bsize = %ld, osize = %d, nsize = %d, fs = %s\n", devtoname(ump->um_dev), (long)fs->fs_bsize, osize, nsize, fs->fs_fsmnt); panic("ffs_realloccg: bad size"); } if (cred == NOCRED) panic("ffs_realloccg: missing credential"); #endif /* INVARIANTS */ reclaimed = 0; retry: if (priv_check_cred(cred, PRIV_VFS_BLOCKRESERVE, 0) && freespace(fs, fs->fs_minfree) - numfrags(fs, nsize - osize) < 0) { goto nospace; } if (bprev == 0) { printf("dev = %s, bsize = %ld, bprev = %jd, fs = %s\n", devtoname(ump->um_dev), (long)fs->fs_bsize, (intmax_t)bprev, fs->fs_fsmnt); panic("ffs_realloccg: bad bprev"); } UFS_UNLOCK(ump); /* * Allocate the extra space in the buffer. */ error = bread_gb(vp, lbprev, osize, NOCRED, gbflags, &bp); if (error) { brelse(bp); return (error); } if (bp->b_blkno == bp->b_lblkno) { if (lbprev >= UFS_NDADDR) panic("ffs_realloccg: lbprev out of range"); bp->b_blkno = fsbtodb(fs, bprev); } #ifdef QUOTA error = chkdq(ip, btodb(nsize - osize), cred, 0); if (error) { brelse(bp); return (error); } #endif /* * Check for extension in the existing location. */ *bpp = NULL; cg = dtog(fs, bprev); UFS_LOCK(ump); bno = ffs_fragextend(ip, cg, bprev, osize, nsize); if (bno) { if (bp->b_blkno != fsbtodb(fs, bno)) panic("ffs_realloccg: bad blockno"); delta = btodb(nsize - osize); DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + delta); if (flags & IO_EXT) ip->i_flag |= IN_CHANGE; else ip->i_flag |= IN_CHANGE | IN_UPDATE; allocbuf(bp, nsize); bp->b_flags |= B_DONE; vfs_bio_bzero_buf(bp, osize, nsize - osize); if ((bp->b_flags & (B_MALLOC | B_VMIO)) == B_VMIO) vfs_bio_set_valid(bp, osize, nsize - osize); *bpp = bp; return (0); } /* * Allocate a new disk location. */ if (bpref >= fs->fs_size) bpref = 0; switch ((int)fs->fs_optim) { case FS_OPTSPACE: /* * Allocate an exact sized fragment. Although this makes * best use of space, we will waste time relocating it if * the file continues to grow. If the fragmentation is * less than half of the minimum free reserve, we choose * to begin optimizing for time. */ request = nsize; if (fs->fs_minfree <= 5 || fs->fs_cstotal.cs_nffree > (off_t)fs->fs_dsize * fs->fs_minfree / (2 * 100)) break; log(LOG_NOTICE, "%s: optimization changed from SPACE to TIME\n", fs->fs_fsmnt); fs->fs_optim = FS_OPTTIME; break; case FS_OPTTIME: /* * At this point we have discovered a file that is trying to * grow a small fragment to a larger fragment. To save time, * we allocate a full sized block, then free the unused portion. * If the file continues to grow, the `ffs_fragextend' call * above will be able to grow it in place without further * copying. If aberrant programs cause disk fragmentation to * grow within 2% of the free reserve, we choose to begin * optimizing for space. */ request = fs->fs_bsize; if (fs->fs_cstotal.cs_nffree < (off_t)fs->fs_dsize * (fs->fs_minfree - 2) / 100) break; log(LOG_NOTICE, "%s: optimization changed from TIME to SPACE\n", fs->fs_fsmnt); fs->fs_optim = FS_OPTSPACE; break; default: printf("dev = %s, optim = %ld, fs = %s\n", devtoname(ump->um_dev), (long)fs->fs_optim, fs->fs_fsmnt); panic("ffs_realloccg: bad optim"); /* NOTREACHED */ } bno = ffs_hashalloc(ip, cg, bpref, request, nsize, ffs_alloccg); if (bno > 0) { bp->b_blkno = fsbtodb(fs, bno); if (!DOINGSOFTDEP(vp)) ffs_blkfree(ump, fs, ump->um_devvp, bprev, (long)osize, ip->i_number, vp->v_type, NULL); delta = btodb(nsize - osize); DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + delta); if (flags & IO_EXT) ip->i_flag |= IN_CHANGE; else ip->i_flag |= IN_CHANGE | IN_UPDATE; allocbuf(bp, nsize); bp->b_flags |= B_DONE; vfs_bio_bzero_buf(bp, osize, nsize - osize); if ((bp->b_flags & (B_MALLOC | B_VMIO)) == B_VMIO) vfs_bio_set_valid(bp, osize, nsize - osize); *bpp = bp; return (0); } #ifdef QUOTA UFS_UNLOCK(ump); /* * Restore user's disk quota because allocation failed. */ (void) chkdq(ip, -btodb(nsize - osize), cred, FORCE); UFS_LOCK(ump); #endif nospace: /* * no space available */ if (reclaimed == 0 && (flags & IO_BUFLOCKED) == 0) { reclaimed = 1; UFS_UNLOCK(ump); if (bp) { brelse(bp); bp = NULL; } UFS_LOCK(ump); softdep_request_cleanup(fs, vp, cred, FLUSH_BLOCKS_WAIT); goto retry; } UFS_UNLOCK(ump); if (bp) brelse(bp); if (reclaimed > 0 && ppsratecheck(&lastfail, &curfail, 1)) { ffs_fserr(fs, ip->i_number, "filesystem full"); uprintf("\n%s: write failed, filesystem is full\n", fs->fs_fsmnt); } return (ENOSPC); } /* * Reallocate a sequence of blocks into a contiguous sequence of blocks. * * The vnode and an array of buffer pointers for a range of sequential * logical blocks to be made contiguous is given. The allocator attempts * to find a range of sequential blocks starting as close as possible * from the end of the allocation for the logical block immediately * preceding the current range. If successful, the physical block numbers * in the buffer pointers and in the inode are changed to reflect the new * allocation. If unsuccessful, the allocation is left unchanged. The * success in doing the reallocation is returned. Note that the error * return is not reflected back to the user. Rather the previous block * allocation will be used. */ SYSCTL_NODE(_vfs, OID_AUTO, ffs, CTLFLAG_RW, 0, "FFS filesystem"); static int doasyncfree = 1; SYSCTL_INT(_vfs_ffs, OID_AUTO, doasyncfree, CTLFLAG_RW, &doasyncfree, 0, "do not force synchronous writes when blocks are reallocated"); static int doreallocblks = 1; SYSCTL_INT(_vfs_ffs, OID_AUTO, doreallocblks, CTLFLAG_RW, &doreallocblks, 0, "enable block reallocation"); static int maxclustersearch = 10; SYSCTL_INT(_vfs_ffs, OID_AUTO, maxclustersearch, CTLFLAG_RW, &maxclustersearch, 0, "max number of cylinder group to search for contigous blocks"); #ifdef DEBUG static volatile int prtrealloc = 0; #endif int ffs_reallocblks(ap) struct vop_reallocblks_args /* { struct vnode *a_vp; struct cluster_save *a_buflist; } */ *ap; { struct ufsmount *ump; /* * If the underlying device can do deletes, then skip reallocating * the blocks of this file into contiguous sequences. Devices that * benefit from BIO_DELETE also benefit from not moving the data. * These devices are flash and therefore work less well with this * optimization. Also skip if reallocblks has been disabled globally. */ ump = ap->a_vp->v_mount->mnt_data; if (ump->um_candelete || doreallocblks == 0) return (ENOSPC); /* * We can't wait in softdep prealloc as it may fsync and recurse * here. Instead we simply fail to reallocate blocks if this * rare condition arises. */ if (DOINGSOFTDEP(ap->a_vp)) if (softdep_prealloc(ap->a_vp, MNT_NOWAIT) != 0) return (ENOSPC); if (ump->um_fstype == UFS1) return (ffs_reallocblks_ufs1(ap)); return (ffs_reallocblks_ufs2(ap)); } static int ffs_reallocblks_ufs1(ap) struct vop_reallocblks_args /* { struct vnode *a_vp; struct cluster_save *a_buflist; } */ *ap; { struct fs *fs; struct inode *ip; struct vnode *vp; struct buf *sbp, *ebp; ufs1_daddr_t *bap, *sbap, *ebap; struct cluster_save *buflist; struct ufsmount *ump; ufs_lbn_t start_lbn, end_lbn; ufs1_daddr_t soff, newblk, blkno; ufs2_daddr_t pref; struct indir start_ap[UFS_NIADDR + 1], end_ap[UFS_NIADDR + 1], *idp; int i, cg, len, start_lvl, end_lvl, ssize; vp = ap->a_vp; ip = VTOI(vp); ump = ITOUMP(ip); fs = ump->um_fs; /* * If we are not tracking block clusters or if we have less than 4% * free blocks left, then do not attempt to cluster. Running with * less than 5% free block reserve is not recommended and those that * choose to do so do not expect to have good file layout. */ if (fs->fs_contigsumsize <= 0 || freespace(fs, 4) < 0) return (ENOSPC); buflist = ap->a_buflist; len = buflist->bs_nchildren; start_lbn = buflist->bs_children[0]->b_lblkno; end_lbn = start_lbn + len - 1; #ifdef INVARIANTS for (i = 0; i < len; i++) if (!ffs_checkblk(ip, dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize)) panic("ffs_reallocblks: unallocated block 1"); for (i = 1; i < len; i++) if (buflist->bs_children[i]->b_lblkno != start_lbn + i) panic("ffs_reallocblks: non-logical cluster"); blkno = buflist->bs_children[0]->b_blkno; ssize = fsbtodb(fs, fs->fs_frag); for (i = 1; i < len - 1; i++) if (buflist->bs_children[i]->b_blkno != blkno + (i * ssize)) panic("ffs_reallocblks: non-physical cluster %d", i); #endif /* * If the cluster crosses the boundary for the first indirect * block, leave space for the indirect block. Indirect blocks * are initially laid out in a position after the last direct * block. Block reallocation would usually destroy locality by * moving the indirect block out of the way to make room for * data blocks if we didn't compensate here. We should also do * this for other indirect block boundaries, but it is only * important for the first one. */ if (start_lbn < UFS_NDADDR && end_lbn >= UFS_NDADDR) return (ENOSPC); /* * If the latest allocation is in a new cylinder group, assume that * the filesystem has decided to move and do not force it back to * the previous cylinder group. */ if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) != dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno))) return (ENOSPC); if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) || ufs_getlbns(vp, end_lbn, end_ap, &end_lvl)) return (ENOSPC); /* * Get the starting offset and block map for the first block. */ if (start_lvl == 0) { sbap = &ip->i_din1->di_db[0]; soff = start_lbn; } else { idp = &start_ap[start_lvl - 1]; if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &sbp)) { brelse(sbp); return (ENOSPC); } sbap = (ufs1_daddr_t *)sbp->b_data; soff = idp->in_off; } /* * If the block range spans two block maps, get the second map. */ ebap = NULL; if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) { ssize = len; } else { #ifdef INVARIANTS if (start_lvl > 0 && start_ap[start_lvl - 1].in_lbn == idp->in_lbn) panic("ffs_reallocblk: start == end"); #endif ssize = len - (idp->in_off + 1); if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &ebp)) goto fail; ebap = (ufs1_daddr_t *)ebp->b_data; } /* * Find the preferred location for the cluster. If we have not * previously failed at this endeavor, then follow our standard * preference calculation. If we have failed at it, then pick up * where we last ended our search. */ UFS_LOCK(ump); if (ip->i_nextclustercg == -1) pref = ffs_blkpref_ufs1(ip, start_lbn, soff, sbap); else pref = cgdata(fs, ip->i_nextclustercg); /* * Search the block map looking for an allocation of the desired size. * To avoid wasting too much time, we limit the number of cylinder * groups that we will search. */ cg = dtog(fs, pref); for (i = min(maxclustersearch, fs->fs_ncg); i > 0; i--) { if ((newblk = ffs_clusteralloc(ip, cg, pref, len)) != 0) break; cg += 1; if (cg >= fs->fs_ncg) cg = 0; } /* * If we have failed in our search, record where we gave up for * next time. Otherwise, fall back to our usual search citerion. */ if (newblk == 0) { ip->i_nextclustercg = cg; UFS_UNLOCK(ump); goto fail; } ip->i_nextclustercg = -1; /* * We have found a new contiguous block. * * First we have to replace the old block pointers with the new * block pointers in the inode and indirect blocks associated * with the file. */ #ifdef DEBUG if (prtrealloc) printf("realloc: ino %ju, lbns %jd-%jd\n\told:", (uintmax_t)ip->i_number, (intmax_t)start_lbn, (intmax_t)end_lbn); #endif blkno = newblk; for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) { if (i == ssize) { bap = ebap; soff = -i; } #ifdef INVARIANTS if (!ffs_checkblk(ip, dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize)) panic("ffs_reallocblks: unallocated block 2"); if (dbtofsb(fs, buflist->bs_children[i]->b_blkno) != *bap) panic("ffs_reallocblks: alloc mismatch"); #endif #ifdef DEBUG if (prtrealloc) printf(" %d,", *bap); #endif if (DOINGSOFTDEP(vp)) { if (sbap == &ip->i_din1->di_db[0] && i < ssize) softdep_setup_allocdirect(ip, start_lbn + i, blkno, *bap, fs->fs_bsize, fs->fs_bsize, buflist->bs_children[i]); else softdep_setup_allocindir_page(ip, start_lbn + i, i < ssize ? sbp : ebp, soff + i, blkno, *bap, buflist->bs_children[i]); } *bap++ = blkno; } /* * Next we must write out the modified inode and indirect blocks. * For strict correctness, the writes should be synchronous since * the old block values may have been written to disk. In practise * they are almost never written, but if we are concerned about * strict correctness, the `doasyncfree' flag should be set to zero. * * The test on `doasyncfree' should be changed to test a flag * that shows whether the associated buffers and inodes have * been written. The flag should be set when the cluster is * started and cleared whenever the buffer or inode is flushed. * We can then check below to see if it is set, and do the * synchronous write only when it has been cleared. */ if (sbap != &ip->i_din1->di_db[0]) { if (doasyncfree) bdwrite(sbp); else bwrite(sbp); } else { ip->i_flag |= IN_CHANGE | IN_UPDATE; if (!doasyncfree) ffs_update(vp, 1); } if (ssize < len) { if (doasyncfree) bdwrite(ebp); else bwrite(ebp); } /* * Last, free the old blocks and assign the new blocks to the buffers. */ #ifdef DEBUG if (prtrealloc) printf("\n\tnew:"); #endif for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) { if (!DOINGSOFTDEP(vp)) ffs_blkfree(ump, fs, ump->um_devvp, dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize, ip->i_number, vp->v_type, NULL); buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno); #ifdef INVARIANTS if (!ffs_checkblk(ip, dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize)) panic("ffs_reallocblks: unallocated block 3"); #endif #ifdef DEBUG if (prtrealloc) printf(" %d,", blkno); #endif } #ifdef DEBUG if (prtrealloc) { prtrealloc--; printf("\n"); } #endif return (0); fail: if (ssize < len) brelse(ebp); if (sbap != &ip->i_din1->di_db[0]) brelse(sbp); return (ENOSPC); } static int ffs_reallocblks_ufs2(ap) struct vop_reallocblks_args /* { struct vnode *a_vp; struct cluster_save *a_buflist; } */ *ap; { struct fs *fs; struct inode *ip; struct vnode *vp; struct buf *sbp, *ebp; ufs2_daddr_t *bap, *sbap, *ebap; struct cluster_save *buflist; struct ufsmount *ump; ufs_lbn_t start_lbn, end_lbn; ufs2_daddr_t soff, newblk, blkno, pref; struct indir start_ap[UFS_NIADDR + 1], end_ap[UFS_NIADDR + 1], *idp; int i, cg, len, start_lvl, end_lvl, ssize; vp = ap->a_vp; ip = VTOI(vp); ump = ITOUMP(ip); fs = ump->um_fs; /* * If we are not tracking block clusters or if we have less than 4% * free blocks left, then do not attempt to cluster. Running with * less than 5% free block reserve is not recommended and those that * choose to do so do not expect to have good file layout. */ if (fs->fs_contigsumsize <= 0 || freespace(fs, 4) < 0) return (ENOSPC); buflist = ap->a_buflist; len = buflist->bs_nchildren; start_lbn = buflist->bs_children[0]->b_lblkno; end_lbn = start_lbn + len - 1; #ifdef INVARIANTS for (i = 0; i < len; i++) if (!ffs_checkblk(ip, dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize)) panic("ffs_reallocblks: unallocated block 1"); for (i = 1; i < len; i++) if (buflist->bs_children[i]->b_lblkno != start_lbn + i) panic("ffs_reallocblks: non-logical cluster"); blkno = buflist->bs_children[0]->b_blkno; ssize = fsbtodb(fs, fs->fs_frag); for (i = 1; i < len - 1; i++) if (buflist->bs_children[i]->b_blkno != blkno + (i * ssize)) panic("ffs_reallocblks: non-physical cluster %d", i); #endif /* * If the cluster crosses the boundary for the first indirect * block, do not move anything in it. Indirect blocks are * usually initially laid out in a position between the data * blocks. Block reallocation would usually destroy locality by * moving the indirect block out of the way to make room for * data blocks if we didn't compensate here. We should also do * this for other indirect block boundaries, but it is only * important for the first one. */ if (start_lbn < UFS_NDADDR && end_lbn >= UFS_NDADDR) return (ENOSPC); /* * If the latest allocation is in a new cylinder group, assume that * the filesystem has decided to move and do not force it back to * the previous cylinder group. */ if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) != dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno))) return (ENOSPC); if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) || ufs_getlbns(vp, end_lbn, end_ap, &end_lvl)) return (ENOSPC); /* * Get the starting offset and block map for the first block. */ if (start_lvl == 0) { sbap = &ip->i_din2->di_db[0]; soff = start_lbn; } else { idp = &start_ap[start_lvl - 1]; if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &sbp)) { brelse(sbp); return (ENOSPC); } sbap = (ufs2_daddr_t *)sbp->b_data; soff = idp->in_off; } /* * If the block range spans two block maps, get the second map. */ ebap = NULL; if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) { ssize = len; } else { #ifdef INVARIANTS if (start_lvl > 0 && start_ap[start_lvl - 1].in_lbn == idp->in_lbn) panic("ffs_reallocblk: start == end"); #endif ssize = len - (idp->in_off + 1); if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &ebp)) goto fail; ebap = (ufs2_daddr_t *)ebp->b_data; } /* * Find the preferred location for the cluster. If we have not * previously failed at this endeavor, then follow our standard * preference calculation. If we have failed at it, then pick up * where we last ended our search. */ UFS_LOCK(ump); if (ip->i_nextclustercg == -1) pref = ffs_blkpref_ufs2(ip, start_lbn, soff, sbap); else pref = cgdata(fs, ip->i_nextclustercg); /* * Search the block map looking for an allocation of the desired size. * To avoid wasting too much time, we limit the number of cylinder * groups that we will search. */ cg = dtog(fs, pref); for (i = min(maxclustersearch, fs->fs_ncg); i > 0; i--) { if ((newblk = ffs_clusteralloc(ip, cg, pref, len)) != 0) break; cg += 1; if (cg >= fs->fs_ncg) cg = 0; } /* * If we have failed in our search, record where we gave up for * next time. Otherwise, fall back to our usual search citerion. */ if (newblk == 0) { ip->i_nextclustercg = cg; UFS_UNLOCK(ump); goto fail; } ip->i_nextclustercg = -1; /* * We have found a new contiguous block. * * First we have to replace the old block pointers with the new * block pointers in the inode and indirect blocks associated * with the file. */ #ifdef DEBUG if (prtrealloc) printf("realloc: ino %ju, lbns %jd-%jd\n\told:", (uintmax_t)ip->i_number, (intmax_t)start_lbn, (intmax_t)end_lbn); #endif blkno = newblk; for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) { if (i == ssize) { bap = ebap; soff = -i; } #ifdef INVARIANTS if (!ffs_checkblk(ip, dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize)) panic("ffs_reallocblks: unallocated block 2"); if (dbtofsb(fs, buflist->bs_children[i]->b_blkno) != *bap) panic("ffs_reallocblks: alloc mismatch"); #endif #ifdef DEBUG if (prtrealloc) printf(" %jd,", (intmax_t)*bap); #endif if (DOINGSOFTDEP(vp)) { if (sbap == &ip->i_din2->di_db[0] && i < ssize) softdep_setup_allocdirect(ip, start_lbn + i, blkno, *bap, fs->fs_bsize, fs->fs_bsize, buflist->bs_children[i]); else softdep_setup_allocindir_page(ip, start_lbn + i, i < ssize ? sbp : ebp, soff + i, blkno, *bap, buflist->bs_children[i]); } *bap++ = blkno; } /* * Next we must write out the modified inode and indirect blocks. * For strict correctness, the writes should be synchronous since * the old block values may have been written to disk. In practise * they are almost never written, but if we are concerned about * strict correctness, the `doasyncfree' flag should be set to zero. * * The test on `doasyncfree' should be changed to test a flag * that shows whether the associated buffers and inodes have * been written. The flag should be set when the cluster is * started and cleared whenever the buffer or inode is flushed. * We can then check below to see if it is set, and do the * synchronous write only when it has been cleared. */ if (sbap != &ip->i_din2->di_db[0]) { if (doasyncfree) bdwrite(sbp); else bwrite(sbp); } else { ip->i_flag |= IN_CHANGE | IN_UPDATE; if (!doasyncfree) ffs_update(vp, 1); } if (ssize < len) { if (doasyncfree) bdwrite(ebp); else bwrite(ebp); } /* * Last, free the old blocks and assign the new blocks to the buffers. */ #ifdef DEBUG if (prtrealloc) printf("\n\tnew:"); #endif for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) { if (!DOINGSOFTDEP(vp)) ffs_blkfree(ump, fs, ump->um_devvp, dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize, ip->i_number, vp->v_type, NULL); buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno); #ifdef INVARIANTS if (!ffs_checkblk(ip, dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize)) panic("ffs_reallocblks: unallocated block 3"); #endif #ifdef DEBUG if (prtrealloc) printf(" %jd,", (intmax_t)blkno); #endif } #ifdef DEBUG if (prtrealloc) { prtrealloc--; printf("\n"); } #endif return (0); fail: if (ssize < len) brelse(ebp); if (sbap != &ip->i_din2->di_db[0]) brelse(sbp); return (ENOSPC); } /* * Allocate an inode in the filesystem. * * If allocating a directory, use ffs_dirpref to select the inode. * If allocating in a directory, the following hierarchy is followed: * 1) allocate the preferred inode. * 2) allocate an inode in the same cylinder group. * 3) quadradically rehash into other cylinder groups, until an * available inode is located. * If no inode preference is given the following hierarchy is used * to allocate an inode: * 1) allocate an inode in cylinder group 0. * 2) quadradically rehash into other cylinder groups, until an * available inode is located. */ int ffs_valloc(pvp, mode, cred, vpp) struct vnode *pvp; int mode; struct ucred *cred; struct vnode **vpp; { struct inode *pip; struct fs *fs; struct inode *ip; struct timespec ts; struct ufsmount *ump; ino_t ino, ipref; u_int cg; int error, error1, reclaimed; static struct timeval lastfail; static int curfail; *vpp = NULL; pip = VTOI(pvp); ump = ITOUMP(pip); fs = ump->um_fs; UFS_LOCK(ump); reclaimed = 0; retry: if (fs->fs_cstotal.cs_nifree == 0) goto noinodes; if ((mode & IFMT) == IFDIR) ipref = ffs_dirpref(pip); else ipref = pip->i_number; if (ipref >= fs->fs_ncg * fs->fs_ipg) ipref = 0; cg = ino_to_cg(fs, ipref); /* * Track number of dirs created one after another * in a same cg without intervening by files. */ if ((mode & IFMT) == IFDIR) { if (fs->fs_contigdirs[cg] < 255) fs->fs_contigdirs[cg]++; } else { if (fs->fs_contigdirs[cg] > 0) fs->fs_contigdirs[cg]--; } ino = (ino_t)ffs_hashalloc(pip, cg, ipref, mode, 0, (allocfcn_t *)ffs_nodealloccg); if (ino == 0) goto noinodes; error = ffs_vget(pvp->v_mount, ino, LK_EXCLUSIVE, vpp); if (error) { error1 = ffs_vgetf(pvp->v_mount, ino, LK_EXCLUSIVE, vpp, FFSV_FORCEINSMQ); ffs_vfree(pvp, ino, mode); if (error1 == 0) { ip = VTOI(*vpp); if (ip->i_mode) goto dup_alloc; ip->i_flag |= IN_MODIFIED; vput(*vpp); } return (error); } ip = VTOI(*vpp); if (ip->i_mode) { dup_alloc: printf("mode = 0%o, inum = %ju, fs = %s\n", ip->i_mode, (uintmax_t)ip->i_number, fs->fs_fsmnt); panic("ffs_valloc: dup alloc"); } if (DIP(ip, i_blocks) && (fs->fs_flags & FS_UNCLEAN) == 0) { /* XXX */ printf("free inode %s/%lu had %ld blocks\n", fs->fs_fsmnt, (u_long)ino, (long)DIP(ip, i_blocks)); DIP_SET(ip, i_blocks, 0); } ip->i_flags = 0; DIP_SET(ip, i_flags, 0); /* * Set up a new generation number for this inode. */ while (ip->i_gen == 0 || ++ip->i_gen == 0) ip->i_gen = arc4random(); DIP_SET(ip, i_gen, ip->i_gen); if (fs->fs_magic == FS_UFS2_MAGIC) { vfs_timestamp(&ts); ip->i_din2->di_birthtime = ts.tv_sec; ip->i_din2->di_birthnsec = ts.tv_nsec; } ufs_prepare_reclaim(*vpp); ip->i_flag = 0; (*vpp)->v_vflag = 0; (*vpp)->v_type = VNON; if (fs->fs_magic == FS_UFS2_MAGIC) { (*vpp)->v_op = &ffs_vnodeops2; ip->i_flag |= IN_UFS2; } else { (*vpp)->v_op = &ffs_vnodeops1; } return (0); noinodes: if (reclaimed == 0) { reclaimed = 1; softdep_request_cleanup(fs, pvp, cred, FLUSH_INODES_WAIT); goto retry; } UFS_UNLOCK(ump); if (ppsratecheck(&lastfail, &curfail, 1)) { ffs_fserr(fs, pip->i_number, "out of inodes"); uprintf("\n%s: create/symlink failed, no inodes free\n", fs->fs_fsmnt); } return (ENOSPC); } /* * Find a cylinder group to place a directory. * * The policy implemented by this algorithm is to allocate a * directory inode in the same cylinder group as its parent * directory, but also to reserve space for its files inodes * and data. Restrict the number of directories which may be * allocated one after another in the same cylinder group * without intervening allocation of files. * * If we allocate a first level directory then force allocation * in another cylinder group. */ static ino_t ffs_dirpref(pip) struct inode *pip; { struct fs *fs; int cg, prefcg, dirsize, cgsize; u_int avgifree, avgbfree, avgndir, curdirsize; u_int minifree, minbfree, maxndir; u_int mincg, minndir; u_int maxcontigdirs; mtx_assert(UFS_MTX(ITOUMP(pip)), MA_OWNED); fs = ITOFS(pip); avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg; avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg; avgndir = fs->fs_cstotal.cs_ndir / fs->fs_ncg; /* * Force allocation in another cg if creating a first level dir. */ ASSERT_VOP_LOCKED(ITOV(pip), "ffs_dirpref"); if (ITOV(pip)->v_vflag & VV_ROOT) { prefcg = arc4random() % fs->fs_ncg; mincg = prefcg; minndir = fs->fs_ipg; for (cg = prefcg; cg < fs->fs_ncg; cg++) if (fs->fs_cs(fs, cg).cs_ndir < minndir && fs->fs_cs(fs, cg).cs_nifree >= avgifree && fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { mincg = cg; minndir = fs->fs_cs(fs, cg).cs_ndir; } for (cg = 0; cg < prefcg; cg++) if (fs->fs_cs(fs, cg).cs_ndir < minndir && fs->fs_cs(fs, cg).cs_nifree >= avgifree && fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { mincg = cg; minndir = fs->fs_cs(fs, cg).cs_ndir; } return ((ino_t)(fs->fs_ipg * mincg)); } /* * Count various limits which used for * optimal allocation of a directory inode. */ maxndir = min(avgndir + fs->fs_ipg / 16, fs->fs_ipg); minifree = avgifree - avgifree / 4; if (minifree < 1) minifree = 1; minbfree = avgbfree - avgbfree / 4; if (minbfree < 1) minbfree = 1; cgsize = fs->fs_fsize * fs->fs_fpg; dirsize = fs->fs_avgfilesize * fs->fs_avgfpdir; curdirsize = avgndir ? (cgsize - avgbfree * fs->fs_bsize) / avgndir : 0; if (dirsize < curdirsize) dirsize = curdirsize; if (dirsize <= 0) maxcontigdirs = 0; /* dirsize overflowed */ else maxcontigdirs = min((avgbfree * fs->fs_bsize) / dirsize, 255); if (fs->fs_avgfpdir > 0) maxcontigdirs = min(maxcontigdirs, fs->fs_ipg / fs->fs_avgfpdir); if (maxcontigdirs == 0) maxcontigdirs = 1; /* * Limit number of dirs in one cg and reserve space for * regular files, but only if we have no deficit in * inodes or space. * * We are trying to find a suitable cylinder group nearby * our preferred cylinder group to place a new directory. * We scan from our preferred cylinder group forward looking * for a cylinder group that meets our criterion. If we get * to the final cylinder group and do not find anything, * we start scanning forwards from the beginning of the * filesystem. While it might seem sensible to start scanning * backwards or even to alternate looking forward and backward, * this approach fails badly when the filesystem is nearly full. * Specifically, we first search all the areas that have no space * and finally try the one preceding that. We repeat this on * every request and in the case of the final block end up * searching the entire filesystem. By jumping to the front * of the filesystem, our future forward searches always look * in new cylinder groups so finds every possible block after * one pass over the filesystem. */ prefcg = ino_to_cg(fs, pip->i_number); for (cg = prefcg; cg < fs->fs_ncg; cg++) if (fs->fs_cs(fs, cg).cs_ndir < maxndir && fs->fs_cs(fs, cg).cs_nifree >= minifree && fs->fs_cs(fs, cg).cs_nbfree >= minbfree) { if (fs->fs_contigdirs[cg] < maxcontigdirs) return ((ino_t)(fs->fs_ipg * cg)); } for (cg = 0; cg < prefcg; cg++) if (fs->fs_cs(fs, cg).cs_ndir < maxndir && fs->fs_cs(fs, cg).cs_nifree >= minifree && fs->fs_cs(fs, cg).cs_nbfree >= minbfree) { if (fs->fs_contigdirs[cg] < maxcontigdirs) return ((ino_t)(fs->fs_ipg * cg)); } /* * This is a backstop when we have deficit in space. */ for (cg = prefcg; cg < fs->fs_ncg; cg++) if (fs->fs_cs(fs, cg).cs_nifree >= avgifree) return ((ino_t)(fs->fs_ipg * cg)); for (cg = 0; cg < prefcg; cg++) if (fs->fs_cs(fs, cg).cs_nifree >= avgifree) break; return ((ino_t)(fs->fs_ipg * cg)); } /* * Select the desired position for the next block in a file. The file is * logically divided into sections. The first section is composed of the * direct blocks and the next fs_maxbpg blocks. Each additional section * contains fs_maxbpg blocks. * * If no blocks have been allocated in the first section, the policy is to * request a block in the same cylinder group as the inode that describes * the file. The first indirect is allocated immediately following the last * direct block and the data blocks for the first indirect immediately * follow it. * * If no blocks have been allocated in any other section, the indirect * block(s) are allocated in the same cylinder group as its inode in an * area reserved immediately following the inode blocks. The policy for * the data blocks is to place them in a cylinder group with a greater than * average number of free blocks. An appropriate cylinder group is found * by using a rotor that sweeps the cylinder groups. When a new group of * blocks is needed, the sweep begins in the cylinder group following the * cylinder group from which the previous allocation was made. The sweep * continues until a cylinder group with greater than the average number * of free blocks is found. If the allocation is for the first block in an * indirect block or the previous block is a hole, then the information on * the previous allocation is unavailable; here a best guess is made based * on the logical block number being allocated. * * If a section is already partially allocated, the policy is to * allocate blocks contiguously within the section if possible. */ ufs2_daddr_t ffs_blkpref_ufs1(ip, lbn, indx, bap) struct inode *ip; ufs_lbn_t lbn; int indx; ufs1_daddr_t *bap; { struct fs *fs; u_int cg, inocg; u_int avgbfree, startcg; ufs2_daddr_t pref; KASSERT(indx <= 0 || bap != NULL, ("need non-NULL bap")); mtx_assert(UFS_MTX(ITOUMP(ip)), MA_OWNED); fs = ITOFS(ip); /* * Allocation of indirect blocks is indicated by passing negative * values in indx: -1 for single indirect, -2 for double indirect, * -3 for triple indirect. As noted below, we attempt to allocate * the first indirect inline with the file data. For all later * indirect blocks, the data is often allocated in other cylinder * groups. However to speed random file access and to speed up * fsck, the filesystem reserves the first fs_metaspace blocks * (typically half of fs_minfree) of the data area of each cylinder * group to hold these later indirect blocks. */ inocg = ino_to_cg(fs, ip->i_number); if (indx < 0) { /* * Our preference for indirect blocks is the zone at the * beginning of the inode's cylinder group data area that * we try to reserve for indirect blocks. */ pref = cgmeta(fs, inocg); /* * If we are allocating the first indirect block, try to * place it immediately following the last direct block. */ if (indx == -1 && lbn < UFS_NDADDR + NINDIR(fs) && ip->i_din1->di_db[UFS_NDADDR - 1] != 0) pref = ip->i_din1->di_db[UFS_NDADDR - 1] + fs->fs_frag; return (pref); } /* * If we are allocating the first data block in the first indirect * block and the indirect has been allocated in the data block area, * try to place it immediately following the indirect block. */ if (lbn == UFS_NDADDR) { pref = ip->i_din1->di_ib[0]; if (pref != 0 && pref >= cgdata(fs, inocg) && pref < cgbase(fs, inocg + 1)) return (pref + fs->fs_frag); } /* * If we are at the beginning of a file, or we have already allocated * the maximum number of blocks per cylinder group, or we do not * have a block allocated immediately preceding us, then we need * to decide where to start allocating new blocks. */ if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) { /* * If we are allocating a directory data block, we want * to place it in the metadata area. */ if ((ip->i_mode & IFMT) == IFDIR) return (cgmeta(fs, inocg)); /* * Until we fill all the direct and all the first indirect's * blocks, we try to allocate in the data area of the inode's * cylinder group. */ if (lbn < UFS_NDADDR + NINDIR(fs)) return (cgdata(fs, inocg)); /* * Find a cylinder with greater than average number of * unused data blocks. */ if (indx == 0 || bap[indx - 1] == 0) startcg = inocg + lbn / fs->fs_maxbpg; else startcg = dtog(fs, bap[indx - 1]) + 1; startcg %= fs->fs_ncg; avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg; for (cg = startcg; cg < fs->fs_ncg; cg++) if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { fs->fs_cgrotor = cg; return (cgdata(fs, cg)); } for (cg = 0; cg <= startcg; cg++) if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { fs->fs_cgrotor = cg; return (cgdata(fs, cg)); } return (0); } /* * Otherwise, we just always try to lay things out contiguously. */ return (bap[indx - 1] + fs->fs_frag); } /* * Same as above, but for UFS2 */ ufs2_daddr_t ffs_blkpref_ufs2(ip, lbn, indx, bap) struct inode *ip; ufs_lbn_t lbn; int indx; ufs2_daddr_t *bap; { struct fs *fs; u_int cg, inocg; u_int avgbfree, startcg; ufs2_daddr_t pref; KASSERT(indx <= 0 || bap != NULL, ("need non-NULL bap")); mtx_assert(UFS_MTX(ITOUMP(ip)), MA_OWNED); fs = ITOFS(ip); /* * Allocation of indirect blocks is indicated by passing negative * values in indx: -1 for single indirect, -2 for double indirect, * -3 for triple indirect. As noted below, we attempt to allocate * the first indirect inline with the file data. For all later * indirect blocks, the data is often allocated in other cylinder * groups. However to speed random file access and to speed up * fsck, the filesystem reserves the first fs_metaspace blocks * (typically half of fs_minfree) of the data area of each cylinder * group to hold these later indirect blocks. */ inocg = ino_to_cg(fs, ip->i_number); if (indx < 0) { /* * Our preference for indirect blocks is the zone at the * beginning of the inode's cylinder group data area that * we try to reserve for indirect blocks. */ pref = cgmeta(fs, inocg); /* * If we are allocating the first indirect block, try to * place it immediately following the last direct block. */ if (indx == -1 && lbn < UFS_NDADDR + NINDIR(fs) && ip->i_din2->di_db[UFS_NDADDR - 1] != 0) pref = ip->i_din2->di_db[UFS_NDADDR - 1] + fs->fs_frag; return (pref); } /* * If we are allocating the first data block in the first indirect * block and the indirect has been allocated in the data block area, * try to place it immediately following the indirect block. */ if (lbn == UFS_NDADDR) { pref = ip->i_din2->di_ib[0]; if (pref != 0 && pref >= cgdata(fs, inocg) && pref < cgbase(fs, inocg + 1)) return (pref + fs->fs_frag); } /* * If we are at the beginning of a file, or we have already allocated * the maximum number of blocks per cylinder group, or we do not * have a block allocated immediately preceding us, then we need * to decide where to start allocating new blocks. */ if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) { /* * If we are allocating a directory data block, we want * to place it in the metadata area. */ if ((ip->i_mode & IFMT) == IFDIR) return (cgmeta(fs, inocg)); /* * Until we fill all the direct and all the first indirect's * blocks, we try to allocate in the data area of the inode's * cylinder group. */ if (lbn < UFS_NDADDR + NINDIR(fs)) return (cgdata(fs, inocg)); /* * Find a cylinder with greater than average number of * unused data blocks. */ if (indx == 0 || bap[indx - 1] == 0) startcg = inocg + lbn / fs->fs_maxbpg; else startcg = dtog(fs, bap[indx - 1]) + 1; startcg %= fs->fs_ncg; avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg; for (cg = startcg; cg < fs->fs_ncg; cg++) if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { fs->fs_cgrotor = cg; return (cgdata(fs, cg)); } for (cg = 0; cg <= startcg; cg++) if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { fs->fs_cgrotor = cg; return (cgdata(fs, cg)); } return (0); } /* * Otherwise, we just always try to lay things out contiguously. */ return (bap[indx - 1] + fs->fs_frag); } /* * Implement the cylinder overflow algorithm. * * The policy implemented by this algorithm is: * 1) allocate the block in its requested cylinder group. * 2) quadradically rehash on the cylinder group number. * 3) brute force search for a free block. * * Must be called with the UFS lock held. Will release the lock on success * and return with it held on failure. */ /*VARARGS5*/ static ufs2_daddr_t ffs_hashalloc(ip, cg, pref, size, rsize, allocator) struct inode *ip; u_int cg; ufs2_daddr_t pref; int size; /* Search size for data blocks, mode for inodes */ int rsize; /* Real allocated size. */ allocfcn_t *allocator; { struct fs *fs; ufs2_daddr_t result; u_int i, icg = cg; mtx_assert(UFS_MTX(ITOUMP(ip)), MA_OWNED); #ifdef INVARIANTS if (ITOV(ip)->v_mount->mnt_kern_flag & MNTK_SUSPENDED) panic("ffs_hashalloc: allocation on suspended filesystem"); #endif fs = ITOFS(ip); /* * 1: preferred cylinder group */ result = (*allocator)(ip, cg, pref, size, rsize); if (result) return (result); /* * 2: quadratic rehash */ for (i = 1; i < fs->fs_ncg; i *= 2) { cg += i; if (cg >= fs->fs_ncg) cg -= fs->fs_ncg; result = (*allocator)(ip, cg, 0, size, rsize); if (result) return (result); } /* * 3: brute force search * Note that we start at i == 2, since 0 was checked initially, * and 1 is always checked in the quadratic rehash. */ cg = (icg + 2) % fs->fs_ncg; for (i = 2; i < fs->fs_ncg; i++) { result = (*allocator)(ip, cg, 0, size, rsize); if (result) return (result); cg++; if (cg == fs->fs_ncg) cg = 0; } return (0); } /* * Determine whether a fragment can be extended. * * Check to see if the necessary fragments are available, and * if they are, allocate them. */ static ufs2_daddr_t ffs_fragextend(ip, cg, bprev, osize, nsize) struct inode *ip; u_int cg; ufs2_daddr_t bprev; int osize, nsize; { struct fs *fs; struct cg *cgp; struct buf *bp; struct ufsmount *ump; int nffree; long bno; int frags, bbase; int i, error; u_int8_t *blksfree; ump = ITOUMP(ip); fs = ump->um_fs; if (fs->fs_cs(fs, cg).cs_nffree < numfrags(fs, nsize - osize)) return (0); frags = numfrags(fs, nsize); bbase = fragnum(fs, bprev); if (bbase > fragnum(fs, (bprev + frags - 1))) { /* cannot extend across a block boundary */ return (0); } UFS_UNLOCK(ump); if ((error = ffs_getcg(fs, ump->um_devvp, cg, &bp, &cgp)) != 0) goto fail; bno = dtogd(fs, bprev); blksfree = cg_blksfree(cgp); for (i = numfrags(fs, osize); i < frags; i++) if (isclr(blksfree, bno + i)) goto fail; /* * the current fragment can be extended * deduct the count on fragment being extended into * increase the count on the remaining fragment (if any) * allocate the extended piece */ for (i = frags; i < fs->fs_frag - bbase; i++) if (isclr(blksfree, bno + i)) break; cgp->cg_frsum[i - numfrags(fs, osize)]--; if (i != frags) cgp->cg_frsum[i - frags]++; for (i = numfrags(fs, osize), nffree = 0; i < frags; i++) { clrbit(blksfree, bno + i); cgp->cg_cs.cs_nffree--; nffree++; } UFS_LOCK(ump); fs->fs_cstotal.cs_nffree -= nffree; fs->fs_cs(fs, cg).cs_nffree -= nffree; fs->fs_fmod = 1; ACTIVECLEAR(fs, cg); UFS_UNLOCK(ump); if (DOINGSOFTDEP(ITOV(ip))) softdep_setup_blkmapdep(bp, UFSTOVFS(ump), bprev, frags, numfrags(fs, osize)); bdwrite(bp); return (bprev); fail: brelse(bp); UFS_LOCK(ump); return (0); } /* * Determine whether a block can be allocated. * * Check to see if a block of the appropriate size is available, * and if it is, allocate it. */ static ufs2_daddr_t ffs_alloccg(ip, cg, bpref, size, rsize) struct inode *ip; u_int cg; ufs2_daddr_t bpref; int size; int rsize; { struct fs *fs; struct cg *cgp; struct buf *bp; struct ufsmount *ump; ufs1_daddr_t bno; ufs2_daddr_t blkno; int i, allocsiz, error, frags; u_int8_t *blksfree; ump = ITOUMP(ip); fs = ump->um_fs; if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize) return (0); UFS_UNLOCK(ump); if ((error = ffs_getcg(fs, ump->um_devvp, cg, &bp, &cgp)) != 0 || (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) goto fail; if (size == fs->fs_bsize) { UFS_LOCK(ump); blkno = ffs_alloccgblk(ip, bp, bpref, rsize); ACTIVECLEAR(fs, cg); UFS_UNLOCK(ump); bdwrite(bp); return (blkno); } /* * check to see if any fragments are already available * allocsiz is the size which will be allocated, hacking * it down to a smaller size if necessary */ blksfree = cg_blksfree(cgp); frags = numfrags(fs, size); for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++) if (cgp->cg_frsum[allocsiz] != 0) break; if (allocsiz == fs->fs_frag) { /* * no fragments were available, so a block will be * allocated, and hacked up */ if (cgp->cg_cs.cs_nbfree == 0) goto fail; UFS_LOCK(ump); blkno = ffs_alloccgblk(ip, bp, bpref, rsize); ACTIVECLEAR(fs, cg); UFS_UNLOCK(ump); bdwrite(bp); return (blkno); } KASSERT(size == rsize, ("ffs_alloccg: size(%d) != rsize(%d)", size, rsize)); bno = ffs_mapsearch(fs, cgp, bpref, allocsiz); if (bno < 0) goto fail; for (i = 0; i < frags; i++) clrbit(blksfree, bno + i); cgp->cg_cs.cs_nffree -= frags; cgp->cg_frsum[allocsiz]--; if (frags != allocsiz) cgp->cg_frsum[allocsiz - frags]++; UFS_LOCK(ump); fs->fs_cstotal.cs_nffree -= frags; fs->fs_cs(fs, cg).cs_nffree -= frags; fs->fs_fmod = 1; blkno = cgbase(fs, cg) + bno; ACTIVECLEAR(fs, cg); UFS_UNLOCK(ump); if (DOINGSOFTDEP(ITOV(ip))) softdep_setup_blkmapdep(bp, UFSTOVFS(ump), blkno, frags, 0); bdwrite(bp); return (blkno); fail: brelse(bp); UFS_LOCK(ump); return (0); } /* * Allocate a block in a cylinder group. * * This algorithm implements the following policy: * 1) allocate the requested block. * 2) allocate a rotationally optimal block in the same cylinder. * 3) allocate the next available block on the block rotor for the * specified cylinder group. * Note that this routine only allocates fs_bsize blocks; these * blocks may be fragmented by the routine that allocates them. */ static ufs2_daddr_t ffs_alloccgblk(ip, bp, bpref, size) struct inode *ip; struct buf *bp; ufs2_daddr_t bpref; int size; { struct fs *fs; struct cg *cgp; struct ufsmount *ump; ufs1_daddr_t bno; ufs2_daddr_t blkno; u_int8_t *blksfree; int i, cgbpref; ump = ITOUMP(ip); fs = ump->um_fs; mtx_assert(UFS_MTX(ump), MA_OWNED); cgp = (struct cg *)bp->b_data; blksfree = cg_blksfree(cgp); if (bpref == 0) { bpref = cgbase(fs, cgp->cg_cgx) + cgp->cg_rotor + fs->fs_frag; } else if ((cgbpref = dtog(fs, bpref)) != cgp->cg_cgx) { /* map bpref to correct zone in this cg */ if (bpref < cgdata(fs, cgbpref)) bpref = cgmeta(fs, cgp->cg_cgx); else bpref = cgdata(fs, cgp->cg_cgx); } /* * if the requested block is available, use it */ bno = dtogd(fs, blknum(fs, bpref)); if (ffs_isblock(fs, blksfree, fragstoblks(fs, bno))) goto gotit; /* * Take the next available block in this cylinder group. */ bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag); if (bno < 0) return (0); /* Update cg_rotor only if allocated from the data zone */ if (bno >= dtogd(fs, cgdata(fs, cgp->cg_cgx))) cgp->cg_rotor = bno; gotit: blkno = fragstoblks(fs, bno); ffs_clrblock(fs, blksfree, (long)blkno); ffs_clusteracct(fs, cgp, blkno, -1); cgp->cg_cs.cs_nbfree--; fs->fs_cstotal.cs_nbfree--; fs->fs_cs(fs, cgp->cg_cgx).cs_nbfree--; fs->fs_fmod = 1; blkno = cgbase(fs, cgp->cg_cgx) + bno; /* * If the caller didn't want the whole block free the frags here. */ size = numfrags(fs, size); if (size != fs->fs_frag) { bno = dtogd(fs, blkno); for (i = size; i < fs->fs_frag; i++) setbit(blksfree, bno + i); i = fs->fs_frag - size; cgp->cg_cs.cs_nffree += i; fs->fs_cstotal.cs_nffree += i; fs->fs_cs(fs, cgp->cg_cgx).cs_nffree += i; fs->fs_fmod = 1; cgp->cg_frsum[i]++; } /* XXX Fixme. */ UFS_UNLOCK(ump); if (DOINGSOFTDEP(ITOV(ip))) softdep_setup_blkmapdep(bp, UFSTOVFS(ump), blkno, size, 0); UFS_LOCK(ump); return (blkno); } /* * Determine whether a cluster can be allocated. * * We do not currently check for optimal rotational layout if there * are multiple choices in the same cylinder group. Instead we just * take the first one that we find following bpref. */ static ufs2_daddr_t ffs_clusteralloc(ip, cg, bpref, len) struct inode *ip; u_int cg; ufs2_daddr_t bpref; int len; { struct fs *fs; struct cg *cgp; struct buf *bp; struct ufsmount *ump; int i, run, bit, map, got, error; ufs2_daddr_t bno; u_char *mapp; int32_t *lp; u_int8_t *blksfree; ump = ITOUMP(ip); fs = ump->um_fs; if (fs->fs_maxcluster[cg] < len) return (0); UFS_UNLOCK(ump); if ((error = ffs_getcg(fs, ump->um_devvp, cg, &bp, &cgp)) != 0) { UFS_LOCK(ump); return (0); } /* * Check to see if a cluster of the needed size (or bigger) is * available in this cylinder group. */ lp = &cg_clustersum(cgp)[len]; for (i = len; i <= fs->fs_contigsumsize; i++) if (*lp++ > 0) break; if (i > fs->fs_contigsumsize) { /* * This is the first time looking for a cluster in this * cylinder group. Update the cluster summary information * to reflect the true maximum sized cluster so that * future cluster allocation requests can avoid reading * the cylinder group map only to find no clusters. */ lp = &cg_clustersum(cgp)[len - 1]; for (i = len - 1; i > 0; i--) if (*lp-- > 0) break; UFS_LOCK(ump); fs->fs_maxcluster[cg] = i; brelse(bp); return (0); } /* * Search the cluster map to find a big enough cluster. * We take the first one that we find, even if it is larger * than we need as we prefer to get one close to the previous * block allocation. We do not search before the current * preference point as we do not want to allocate a block * that is allocated before the previous one (as we will * then have to wait for another pass of the elevator * algorithm before it will be read). We prefer to fail and * be recalled to try an allocation in the next cylinder group. */ if (dtog(fs, bpref) != cg) bpref = cgdata(fs, cg); else bpref = blknum(fs, bpref); bpref = fragstoblks(fs, dtogd(fs, bpref)); mapp = &cg_clustersfree(cgp)[bpref / NBBY]; map = *mapp++; bit = 1 << (bpref % NBBY); for (run = 0, got = bpref; got < cgp->cg_nclusterblks; got++) { if ((map & bit) == 0) { run = 0; } else { run++; if (run == len) break; } if ((got & (NBBY - 1)) != (NBBY - 1)) { bit <<= 1; } else { map = *mapp++; bit = 1; } } if (got >= cgp->cg_nclusterblks) { UFS_LOCK(ump); brelse(bp); return (0); } /* * Allocate the cluster that we have found. */ blksfree = cg_blksfree(cgp); for (i = 1; i <= len; i++) if (!ffs_isblock(fs, blksfree, got - run + i)) panic("ffs_clusteralloc: map mismatch"); bno = cgbase(fs, cg) + blkstofrags(fs, got - run + 1); if (dtog(fs, bno) != cg) panic("ffs_clusteralloc: allocated out of group"); len = blkstofrags(fs, len); UFS_LOCK(ump); for (i = 0; i < len; i += fs->fs_frag) if (ffs_alloccgblk(ip, bp, bno + i, fs->fs_bsize) != bno + i) panic("ffs_clusteralloc: lost block"); ACTIVECLEAR(fs, cg); UFS_UNLOCK(ump); bdwrite(bp); return (bno); } static inline struct buf * getinobuf(struct inode *ip, u_int cg, u_int32_t cginoblk, int gbflags) { struct fs *fs; fs = ITOFS(ip); return (getblk(ITODEVVP(ip), fsbtodb(fs, ino_to_fsba(fs, cg * fs->fs_ipg + cginoblk)), (int)fs->fs_bsize, 0, 0, gbflags)); } /* * Synchronous inode initialization is needed only when barrier writes do not * work as advertised, and will impose a heavy cost on file creation in a newly * created filesystem. */ static int doasyncinodeinit = 1; SYSCTL_INT(_vfs_ffs, OID_AUTO, doasyncinodeinit, CTLFLAG_RWTUN, &doasyncinodeinit, 0, "Perform inode block initialization using asynchronous writes"); /* * Determine whether an inode can be allocated. * * Check to see if an inode is available, and if it is, * allocate it using the following policy: * 1) allocate the requested inode. * 2) allocate the next available inode after the requested * inode in the specified cylinder group. */ static ufs2_daddr_t ffs_nodealloccg(ip, cg, ipref, mode, unused) struct inode *ip; u_int cg; ufs2_daddr_t ipref; int mode; int unused; { struct fs *fs; struct cg *cgp; struct buf *bp, *ibp; struct ufsmount *ump; u_int8_t *inosused, *loc; struct ufs2_dinode *dp2; int error, start, len, i; u_int32_t old_initediblk; ump = ITOUMP(ip); fs = ump->um_fs; check_nifree: if (fs->fs_cs(fs, cg).cs_nifree == 0) return (0); UFS_UNLOCK(ump); if ((error = ffs_getcg(fs, ump->um_devvp, cg, &bp, &cgp)) != 0) { UFS_LOCK(ump); return (0); } restart: if (cgp->cg_cs.cs_nifree == 0) { brelse(bp); UFS_LOCK(ump); return (0); } inosused = cg_inosused(cgp); if (ipref) { ipref %= fs->fs_ipg; if (isclr(inosused, ipref)) goto gotit; } start = cgp->cg_irotor / NBBY; len = howmany(fs->fs_ipg - cgp->cg_irotor, NBBY); loc = memcchr(&inosused[start], 0xff, len); if (loc == NULL) { len = start + 1; start = 0; loc = memcchr(&inosused[start], 0xff, len); if (loc == NULL) { printf("cg = %d, irotor = %ld, fs = %s\n", cg, (long)cgp->cg_irotor, fs->fs_fsmnt); panic("ffs_nodealloccg: map corrupted"); /* NOTREACHED */ } } ipref = (loc - inosused) * NBBY + ffs(~*loc) - 1; gotit: /* * Check to see if we need to initialize more inodes. */ if (fs->fs_magic == FS_UFS2_MAGIC && ipref + INOPB(fs) > cgp->cg_initediblk && cgp->cg_initediblk < cgp->cg_niblk) { old_initediblk = cgp->cg_initediblk; /* * Free the cylinder group lock before writing the * initialized inode block. Entering the * babarrierwrite() with the cylinder group lock * causes lock order violation between the lock and * snaplk. * * Another thread can decide to initialize the same * inode block, but whichever thread first gets the * cylinder group lock after writing the newly * allocated inode block will update it and the other * will realize that it has lost and leave the * cylinder group unchanged. */ ibp = getinobuf(ip, cg, old_initediblk, GB_LOCK_NOWAIT); brelse(bp); if (ibp == NULL) { /* * The inode block buffer is already owned by * another thread, which must initialize it. * Wait on the buffer to allow another thread * to finish the updates, with dropped cg * buffer lock, then retry. */ ibp = getinobuf(ip, cg, old_initediblk, 0); brelse(ibp); UFS_LOCK(ump); goto check_nifree; } bzero(ibp->b_data, (int)fs->fs_bsize); dp2 = (struct ufs2_dinode *)(ibp->b_data); for (i = 0; i < INOPB(fs); i++) { while (dp2->di_gen == 0) dp2->di_gen = arc4random(); dp2++; } /* * Rather than adding a soft updates dependency to ensure * that the new inode block is written before it is claimed * by the cylinder group map, we just do a barrier write * here. The barrier write will ensure that the inode block * gets written before the updated cylinder group map can be * written. The barrier write should only slow down bulk * loading of newly created filesystems. */ if (doasyncinodeinit) babarrierwrite(ibp); else bwrite(ibp); /* * After the inode block is written, try to update the * cg initediblk pointer. If another thread beat us * to it, then leave it unchanged as the other thread * has already set it correctly. */ error = ffs_getcg(fs, ump->um_devvp, cg, &bp, &cgp); UFS_LOCK(ump); ACTIVECLEAR(fs, cg); UFS_UNLOCK(ump); if (error != 0) return (error); if (cgp->cg_initediblk == old_initediblk) cgp->cg_initediblk += INOPB(fs); goto restart; } cgp->cg_irotor = ipref; UFS_LOCK(ump); ACTIVECLEAR(fs, cg); setbit(inosused, ipref); cgp->cg_cs.cs_nifree--; fs->fs_cstotal.cs_nifree--; fs->fs_cs(fs, cg).cs_nifree--; fs->fs_fmod = 1; if ((mode & IFMT) == IFDIR) { cgp->cg_cs.cs_ndir++; fs->fs_cstotal.cs_ndir++; fs->fs_cs(fs, cg).cs_ndir++; } UFS_UNLOCK(ump); if (DOINGSOFTDEP(ITOV(ip))) softdep_setup_inomapdep(bp, ip, cg * fs->fs_ipg + ipref, mode); bdwrite(bp); return ((ino_t)(cg * fs->fs_ipg + ipref)); } /* * Free a block or fragment. * * The specified block or fragment is placed back in the * free map. If a fragment is deallocated, a possible * block reassembly is checked. */ static void ffs_blkfree_cg(ump, fs, devvp, bno, size, inum, dephd) struct ufsmount *ump; struct fs *fs; struct vnode *devvp; ufs2_daddr_t bno; long size; ino_t inum; struct workhead *dephd; { struct mount *mp; struct cg *cgp; struct buf *bp; ufs1_daddr_t fragno, cgbno; int i, blk, frags, bbase, error; u_int cg; u_int8_t *blksfree; struct cdev *dev; cg = dtog(fs, bno); if (devvp->v_type == VREG) { /* devvp is a snapshot */ MPASS(devvp->v_mount->mnt_data == ump); dev = ump->um_devvp->v_rdev; } else if (devvp->v_type == VCHR) { /* devvp is a normal disk device */ dev = devvp->v_rdev; ASSERT_VOP_LOCKED(devvp, "ffs_blkfree_cg"); } else return; #ifdef INVARIANTS if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0 || fragnum(fs, bno) + numfrags(fs, size) > fs->fs_frag) { printf("dev=%s, bno = %jd, bsize = %ld, size = %ld, fs = %s\n", devtoname(dev), (intmax_t)bno, (long)fs->fs_bsize, size, fs->fs_fsmnt); panic("ffs_blkfree_cg: bad size"); } #endif if ((u_int)bno >= fs->fs_size) { printf("bad block %jd, ino %lu\n", (intmax_t)bno, (u_long)inum); ffs_fserr(fs, inum, "bad block"); return; } if ((error = ffs_getcg(fs, devvp, cg, &bp, &cgp)) != 0) return; cgbno = dtogd(fs, bno); blksfree = cg_blksfree(cgp); UFS_LOCK(ump); if (size == fs->fs_bsize) { fragno = fragstoblks(fs, cgbno); if (!ffs_isfreeblock(fs, blksfree, fragno)) { if (devvp->v_type == VREG) { UFS_UNLOCK(ump); /* devvp is a snapshot */ brelse(bp); return; } printf("dev = %s, block = %jd, fs = %s\n", devtoname(dev), (intmax_t)bno, fs->fs_fsmnt); panic("ffs_blkfree_cg: freeing free block"); } ffs_setblock(fs, blksfree, fragno); ffs_clusteracct(fs, cgp, fragno, 1); cgp->cg_cs.cs_nbfree++; fs->fs_cstotal.cs_nbfree++; fs->fs_cs(fs, cg).cs_nbfree++; } else { bbase = cgbno - fragnum(fs, cgbno); /* * decrement the counts associated with the old frags */ blk = blkmap(fs, blksfree, bbase); ffs_fragacct(fs, blk, cgp->cg_frsum, -1); /* * deallocate the fragment */ frags = numfrags(fs, size); for (i = 0; i < frags; i++) { if (isset(blksfree, cgbno + i)) { printf("dev = %s, block = %jd, fs = %s\n", devtoname(dev), (intmax_t)(bno + i), fs->fs_fsmnt); panic("ffs_blkfree_cg: freeing free frag"); } setbit(blksfree, cgbno + i); } cgp->cg_cs.cs_nffree += i; fs->fs_cstotal.cs_nffree += i; fs->fs_cs(fs, cg).cs_nffree += i; /* * add back in counts associated with the new frags */ blk = blkmap(fs, blksfree, bbase); ffs_fragacct(fs, blk, cgp->cg_frsum, 1); /* * if a complete block has been reassembled, account for it */ fragno = fragstoblks(fs, bbase); if (ffs_isblock(fs, blksfree, fragno)) { cgp->cg_cs.cs_nffree -= fs->fs_frag; fs->fs_cstotal.cs_nffree -= fs->fs_frag; fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag; ffs_clusteracct(fs, cgp, fragno, 1); cgp->cg_cs.cs_nbfree++; fs->fs_cstotal.cs_nbfree++; fs->fs_cs(fs, cg).cs_nbfree++; } } fs->fs_fmod = 1; ACTIVECLEAR(fs, cg); UFS_UNLOCK(ump); mp = UFSTOVFS(ump); if (MOUNTEDSOFTDEP(mp) && devvp->v_type == VCHR) softdep_setup_blkfree(UFSTOVFS(ump), bp, bno, numfrags(fs, size), dephd); bdwrite(bp); } struct ffs_blkfree_trim_params { struct task task; struct ufsmount *ump; struct vnode *devvp; ufs2_daddr_t bno; long size; ino_t inum; struct workhead *pdephd; struct workhead dephd; }; static void ffs_blkfree_trim_task(ctx, pending) void *ctx; int pending; { struct ffs_blkfree_trim_params *tp; tp = ctx; ffs_blkfree_cg(tp->ump, tp->ump->um_fs, tp->devvp, tp->bno, tp->size, tp->inum, tp->pdephd); vn_finished_secondary_write(UFSTOVFS(tp->ump)); atomic_add_int(&tp->ump->um_trim_inflight, -1); free(tp, M_TEMP); } static void ffs_blkfree_trim_completed(bip) struct bio *bip; { struct ffs_blkfree_trim_params *tp; tp = bip->bio_caller2; g_destroy_bio(bip); TASK_INIT(&tp->task, 0, ffs_blkfree_trim_task, tp); taskqueue_enqueue(tp->ump->um_trim_tq, &tp->task); } void ffs_blkfree(ump, fs, devvp, bno, size, inum, vtype, dephd) struct ufsmount *ump; struct fs *fs; struct vnode *devvp; ufs2_daddr_t bno; long size; ino_t inum; enum vtype vtype; struct workhead *dephd; { struct mount *mp; struct bio *bip; struct ffs_blkfree_trim_params *tp; /* * Check to see if a snapshot wants to claim the block. * Check that devvp is a normal disk device, not a snapshot, * it has a snapshot(s) associated with it, and one of the * snapshots wants to claim the block. */ if (devvp->v_type == VCHR && (devvp->v_vflag & VV_COPYONWRITE) && ffs_snapblkfree(fs, devvp, bno, size, inum, vtype, dephd)) { return; } /* * Nothing to delay if TRIM is disabled, or the operation is * performed on the snapshot. */ if (!ump->um_candelete || devvp->v_type == VREG) { ffs_blkfree_cg(ump, fs, devvp, bno, size, inum, dephd); return; } /* * Postpone the set of the free bit in the cg bitmap until the * BIO_DELETE is completed. Otherwise, due to disk queue * reordering, TRIM might be issued after we reuse the block * and write some new data into it. */ atomic_add_int(&ump->um_trim_inflight, 1); tp = malloc(sizeof(struct ffs_blkfree_trim_params), M_TEMP, M_WAITOK); tp->ump = ump; tp->devvp = devvp; tp->bno = bno; tp->size = size; tp->inum = inum; if (dephd != NULL) { LIST_INIT(&tp->dephd); LIST_SWAP(dephd, &tp->dephd, worklist, wk_list); tp->pdephd = &tp->dephd; } else tp->pdephd = NULL; bip = g_alloc_bio(); bip->bio_cmd = BIO_DELETE; bip->bio_offset = dbtob(fsbtodb(fs, bno)); bip->bio_done = ffs_blkfree_trim_completed; bip->bio_length = size; bip->bio_caller2 = tp; mp = UFSTOVFS(ump); vn_start_secondary_write(NULL, &mp, 0); g_io_request(bip, (struct g_consumer *)devvp->v_bufobj.bo_private); } #ifdef INVARIANTS /* * Verify allocation of a block or fragment. Returns true if block or * fragment is allocated, false if it is free. */ static int ffs_checkblk(ip, bno, size) struct inode *ip; ufs2_daddr_t bno; long size; { struct fs *fs; struct cg *cgp; struct buf *bp; ufs1_daddr_t cgbno; int i, error, frags, free; u_int8_t *blksfree; fs = ITOFS(ip); if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) { printf("bsize = %ld, size = %ld, fs = %s\n", (long)fs->fs_bsize, size, fs->fs_fsmnt); panic("ffs_checkblk: bad size"); } if ((u_int)bno >= fs->fs_size) panic("ffs_checkblk: bad block %jd", (intmax_t)bno); error = ffs_getcg(fs, ITODEVVP(ip), dtog(fs, bno), &bp, &cgp); if (error) panic("ffs_checkblk: cylinder group read failed"); blksfree = cg_blksfree(cgp); cgbno = dtogd(fs, bno); if (size == fs->fs_bsize) { free = ffs_isblock(fs, blksfree, fragstoblks(fs, cgbno)); } else { frags = numfrags(fs, size); for (free = 0, i = 0; i < frags; i++) if (isset(blksfree, cgbno + i)) free++; if (free != 0 && free != frags) panic("ffs_checkblk: partially free fragment"); } brelse(bp); return (!free); } #endif /* INVARIANTS */ /* * Free an inode. */ int ffs_vfree(pvp, ino, mode) struct vnode *pvp; ino_t ino; int mode; { struct ufsmount *ump; if (DOINGSOFTDEP(pvp)) { softdep_freefile(pvp, ino, mode); return (0); } ump = VFSTOUFS(pvp->v_mount); return (ffs_freefile(ump, ump->um_fs, ump->um_devvp, ino, mode, NULL)); } /* * Do the actual free operation. * The specified inode is placed back in the free map. */ int ffs_freefile(ump, fs, devvp, ino, mode, wkhd) struct ufsmount *ump; struct fs *fs; struct vnode *devvp; ino_t ino; int mode; struct workhead *wkhd; { struct cg *cgp; struct buf *bp; ufs2_daddr_t cgbno; int error; u_int cg; u_int8_t *inosused; struct cdev *dev; cg = ino_to_cg(fs, ino); if (devvp->v_type == VREG) { /* devvp is a snapshot */ MPASS(devvp->v_mount->mnt_data == ump); dev = ump->um_devvp->v_rdev; cgbno = fragstoblks(fs, cgtod(fs, cg)); } else if (devvp->v_type == VCHR) { /* devvp is a normal disk device */ dev = devvp->v_rdev; cgbno = fsbtodb(fs, cgtod(fs, cg)); } else { bp = NULL; return (0); } if (ino >= fs->fs_ipg * fs->fs_ncg) panic("ffs_freefile: range: dev = %s, ino = %ju, fs = %s", devtoname(dev), (uintmax_t)ino, fs->fs_fsmnt); if ((error = ffs_getcg(fs, devvp, cg, &bp, &cgp)) != 0) return (error); inosused = cg_inosused(cgp); ino %= fs->fs_ipg; if (isclr(inosused, ino)) { printf("dev = %s, ino = %ju, fs = %s\n", devtoname(dev), (uintmax_t)(ino + cg * fs->fs_ipg), fs->fs_fsmnt); if (fs->fs_ronly == 0) panic("ffs_freefile: freeing free inode"); } clrbit(inosused, ino); if (ino < cgp->cg_irotor) cgp->cg_irotor = ino; cgp->cg_cs.cs_nifree++; UFS_LOCK(ump); fs->fs_cstotal.cs_nifree++; fs->fs_cs(fs, cg).cs_nifree++; if ((mode & IFMT) == IFDIR) { cgp->cg_cs.cs_ndir--; fs->fs_cstotal.cs_ndir--; fs->fs_cs(fs, cg).cs_ndir--; } fs->fs_fmod = 1; ACTIVECLEAR(fs, cg); UFS_UNLOCK(ump); if (MOUNTEDSOFTDEP(UFSTOVFS(ump)) && devvp->v_type == VCHR) softdep_setup_inofree(UFSTOVFS(ump), bp, ino + cg * fs->fs_ipg, wkhd); bdwrite(bp); return (0); } /* * Check to see if a file is free. * Used to check for allocated files in snapshots. */ int ffs_checkfreefile(fs, devvp, ino) struct fs *fs; struct vnode *devvp; ino_t ino; { struct cg *cgp; struct buf *bp; ufs2_daddr_t cgbno; int ret, error; u_int cg; u_int8_t *inosused; cg = ino_to_cg(fs, ino); if (devvp->v_type == VREG) { /* devvp is a snapshot */ cgbno = fragstoblks(fs, cgtod(fs, cg)); } else if (devvp->v_type == VCHR) { /* devvp is a normal disk device */ cgbno = fsbtodb(fs, cgtod(fs, cg)); } else { return (1); } if (ino >= fs->fs_ipg * fs->fs_ncg) return (1); if ((error = ffs_getcg(fs, devvp, cg, &bp, &cgp)) != 0) return (1); inosused = cg_inosused(cgp); ino %= fs->fs_ipg; ret = isclr(inosused, ino); brelse(bp); return (ret); } /* * Find a block of the specified size in the specified cylinder group. * * It is a panic if a request is made to find a block if none are * available. */ static ufs1_daddr_t ffs_mapsearch(fs, cgp, bpref, allocsiz) struct fs *fs; struct cg *cgp; ufs2_daddr_t bpref; int allocsiz; { ufs1_daddr_t bno; int start, len, loc, i; int blk, field, subfield, pos; u_int8_t *blksfree; /* * find the fragment by searching through the free block * map for an appropriate bit pattern */ if (bpref) start = dtogd(fs, bpref) / NBBY; else start = cgp->cg_frotor / NBBY; blksfree = cg_blksfree(cgp); len = howmany(fs->fs_fpg, NBBY) - start; loc = scanc((u_int)len, (u_char *)&blksfree[start], fragtbl[fs->fs_frag], (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY)))); if (loc == 0) { len = start + 1; start = 0; loc = scanc((u_int)len, (u_char *)&blksfree[0], fragtbl[fs->fs_frag], (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY)))); if (loc == 0) { printf("start = %d, len = %d, fs = %s\n", start, len, fs->fs_fsmnt); panic("ffs_alloccg: map corrupted"); /* NOTREACHED */ } } bno = (start + len - loc) * NBBY; cgp->cg_frotor = bno; /* * found the byte in the map * sift through the bits to find the selected frag */ for (i = bno + NBBY; bno < i; bno += fs->fs_frag) { blk = blkmap(fs, blksfree, bno); blk <<= 1; field = around[allocsiz]; subfield = inside[allocsiz]; for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) { if ((blk & field) == subfield) return (bno + pos); field <<= 1; subfield <<= 1; } } printf("bno = %lu, fs = %s\n", (u_long)bno, fs->fs_fsmnt); panic("ffs_alloccg: block not in map"); return (-1); } static const struct statfs * ffs_getmntstat(struct vnode *devvp) { if (devvp->v_type == VCHR) return (&devvp->v_rdev->si_mountpt->mnt_stat); return (ffs_getmntstat(VFSTOUFS(devvp->v_mount)->um_devvp)); } /* * Fetch and verify a cylinder group. */ int ffs_getcg(fs, devvp, cg, bpp, cgpp) struct fs *fs; struct vnode *devvp; u_int cg; struct buf **bpp; struct cg **cgpp; { struct buf *bp; struct cg *cgp; const struct statfs *sfs; int flags, error; *bpp = NULL; *cgpp = NULL; flags = 0; if ((fs->fs_metackhash & CK_CYLGRP) != 0) flags |= GB_CKHASH; error = breadn_flags(devvp, devvp->v_type == VREG ? fragstoblks(fs, cgtod(fs, cg)) : fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize, NULL, NULL, 0, NOCRED, flags, ffs_ckhash_cg, &bp); if (error != 0) return (error); cgp = (struct cg *)bp->b_data; if ((fs->fs_metackhash & CK_CYLGRP) != 0 && (bp->b_flags & B_CKHASH) != 0 && cgp->cg_ckhash != bp->b_ckhash) { sfs = ffs_getmntstat(devvp); printf("UFS %s%s (%s) cylinder checksum failed: cg %u, cgp: " "0x%x != bp: 0x%jx\n", devvp->v_type == VCHR ? "" : "snapshot of ", sfs->f_mntfromname, sfs->f_mntonname, cg, cgp->cg_ckhash, (uintmax_t)bp->b_ckhash); bp->b_flags &= ~B_CKHASH; bp->b_flags |= B_INVAL | B_NOCACHE; brelse(bp); return (EIO); } if (!cg_chkmagic(cgp) || cgp->cg_cgx != cg) { sfs = ffs_getmntstat(devvp); printf("UFS %s%s (%s)", devvp->v_type == VCHR ? "" : "snapshot of ", sfs->f_mntfromname, sfs->f_mntonname); if (!cg_chkmagic(cgp)) printf(" cg %u: bad magic number 0x%x should be 0x%x\n", cg, cgp->cg_magic, CG_MAGIC); else printf(": wrong cylinder group cg %u != cgx %u\n", cg, cgp->cg_cgx); bp->b_flags &= ~B_CKHASH; bp->b_flags |= B_INVAL | B_NOCACHE; brelse(bp); return (EIO); } bp->b_flags &= ~B_CKHASH; bp->b_xflags |= BX_BKGRDWRITE; + /* + * If we are using check hashes on the cylinder group then we want + * to limit changing the cylinder group time to when we are actually + * going to write it to disk so that its check hash remains correct + * in memory. If the CK_CYLGRP flag is set the time is updated in + * ffs_bufwrite() as the buffer is queued for writing. Otherwise we + * update the time here as we have done historically. + */ if ((fs->fs_metackhash & CK_CYLGRP) != 0) bp->b_xflags |= BX_CYLGRP; - cgp->cg_old_time = cgp->cg_time = time_second; + else + cgp->cg_old_time = cgp->cg_time = time_second; *bpp = bp; *cgpp = cgp; return (0); } static void ffs_ckhash_cg(bp) struct buf *bp; { uint32_t ckhash; struct cg *cgp; cgp = (struct cg *)bp->b_data; ckhash = cgp->cg_ckhash; cgp->cg_ckhash = 0; bp->b_ckhash = calculate_crc32c(~0L, bp->b_data, bp->b_bcount); cgp->cg_ckhash = ckhash; } /* * Fserr prints the name of a filesystem with an error diagnostic. * * The form of the error message is: * fs: error message */ void ffs_fserr(fs, inum, cp) struct fs *fs; ino_t inum; char *cp; { struct thread *td = curthread; /* XXX */ struct proc *p = td->td_proc; log(LOG_ERR, "pid %d (%s), uid %d inumber %ju on %s: %s\n", p->p_pid, p->p_comm, td->td_ucred->cr_uid, (uintmax_t)inum, fs->fs_fsmnt, cp); } /* * This function provides the capability for the fsck program to * update an active filesystem. Fourteen operations are provided: * * adjrefcnt(inode, amt) - adjusts the reference count on the * specified inode by the specified amount. Under normal * operation the count should always go down. Decrementing * the count to zero will cause the inode to be freed. * adjblkcnt(inode, amt) - adjust the number of blocks used by the * inode by the specified amount. * adjndir, adjbfree, adjifree, adjffree, adjnumclusters(amt) - * adjust the superblock summary. * freedirs(inode, count) - directory inodes [inode..inode + count - 1] * are marked as free. Inodes should never have to be marked * as in use. * freefiles(inode, count) - file inodes [inode..inode + count - 1] * are marked as free. Inodes should never have to be marked * as in use. * freeblks(blockno, size) - blocks [blockno..blockno + size - 1] * are marked as free. Blocks should never have to be marked * as in use. * setflags(flags, set/clear) - the fs_flags field has the specified * flags set (second parameter +1) or cleared (second parameter -1). * setcwd(dirinode) - set the current directory to dirinode in the * filesystem associated with the snapshot. * setdotdot(oldvalue, newvalue) - Verify that the inode number for ".." * in the current directory is oldvalue then change it to newvalue. * unlink(nameptr, oldvalue) - Verify that the inode number associated * with nameptr in the current directory is oldvalue then unlink it. * * The following functions may only be used on a quiescent filesystem * by the soft updates journal. They are not safe to be run on an active * filesystem. * * setinode(inode, dip) - the specified disk inode is replaced with the * contents pointed to by dip. * setbufoutput(fd, flags) - output associated with the specified file * descriptor (which must reference the character device supporting * the filesystem) switches from using physio to running through the * buffer cache when flags is set to 1. The descriptor reverts to * physio for output when flags is set to zero. */ static int sysctl_ffs_fsck(SYSCTL_HANDLER_ARGS); SYSCTL_PROC(_vfs_ffs, FFS_ADJ_REFCNT, adjrefcnt, CTLFLAG_WR|CTLTYPE_STRUCT, 0, 0, sysctl_ffs_fsck, "S,fsck", "Adjust Inode Reference Count"); static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_BLKCNT, adjblkcnt, CTLFLAG_WR, sysctl_ffs_fsck, "Adjust Inode Used Blocks Count"); static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NDIR, adjndir, CTLFLAG_WR, sysctl_ffs_fsck, "Adjust number of directories"); static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NBFREE, adjnbfree, CTLFLAG_WR, sysctl_ffs_fsck, "Adjust number of free blocks"); static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NIFREE, adjnifree, CTLFLAG_WR, sysctl_ffs_fsck, "Adjust number of free inodes"); static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NFFREE, adjnffree, CTLFLAG_WR, sysctl_ffs_fsck, "Adjust number of free frags"); static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NUMCLUSTERS, adjnumclusters, CTLFLAG_WR, sysctl_ffs_fsck, "Adjust number of free clusters"); static SYSCTL_NODE(_vfs_ffs, FFS_DIR_FREE, freedirs, CTLFLAG_WR, sysctl_ffs_fsck, "Free Range of Directory Inodes"); static SYSCTL_NODE(_vfs_ffs, FFS_FILE_FREE, freefiles, CTLFLAG_WR, sysctl_ffs_fsck, "Free Range of File Inodes"); static SYSCTL_NODE(_vfs_ffs, FFS_BLK_FREE, freeblks, CTLFLAG_WR, sysctl_ffs_fsck, "Free Range of Blocks"); static SYSCTL_NODE(_vfs_ffs, FFS_SET_FLAGS, setflags, CTLFLAG_WR, sysctl_ffs_fsck, "Change Filesystem Flags"); static SYSCTL_NODE(_vfs_ffs, FFS_SET_CWD, setcwd, CTLFLAG_WR, sysctl_ffs_fsck, "Set Current Working Directory"); static SYSCTL_NODE(_vfs_ffs, FFS_SET_DOTDOT, setdotdot, CTLFLAG_WR, sysctl_ffs_fsck, "Change Value of .. Entry"); static SYSCTL_NODE(_vfs_ffs, FFS_UNLINK, unlink, CTLFLAG_WR, sysctl_ffs_fsck, "Unlink a Duplicate Name"); static SYSCTL_NODE(_vfs_ffs, FFS_SET_INODE, setinode, CTLFLAG_WR, sysctl_ffs_fsck, "Update an On-Disk Inode"); static SYSCTL_NODE(_vfs_ffs, FFS_SET_BUFOUTPUT, setbufoutput, CTLFLAG_WR, sysctl_ffs_fsck, "Set Buffered Writing for Descriptor"); #define DEBUG 1 #ifdef DEBUG static int fsckcmds = 0; SYSCTL_INT(_debug, OID_AUTO, fsckcmds, CTLFLAG_RW, &fsckcmds, 0, ""); #endif /* DEBUG */ static int buffered_write(struct file *, struct uio *, struct ucred *, int, struct thread *); static int sysctl_ffs_fsck(SYSCTL_HANDLER_ARGS) { struct thread *td = curthread; struct fsck_cmd cmd; struct ufsmount *ump; struct vnode *vp, *dvp, *fdvp; struct inode *ip, *dp; struct mount *mp; struct fs *fs; ufs2_daddr_t blkno; long blkcnt, blksize; struct file *fp, *vfp; cap_rights_t rights; int filetype, error; static struct fileops *origops, bufferedops; if (req->newlen > sizeof cmd) return (EBADRPC); if ((error = SYSCTL_IN(req, &cmd, sizeof cmd)) != 0) return (error); if (cmd.version != FFS_CMD_VERSION) return (ERPCMISMATCH); if ((error = getvnode(td, cmd.handle, cap_rights_init(&rights, CAP_FSCK), &fp)) != 0) return (error); vp = fp->f_data; if (vp->v_type != VREG && vp->v_type != VDIR) { fdrop(fp, td); return (EINVAL); } vn_start_write(vp, &mp, V_WAIT); if (mp == NULL || strncmp(mp->mnt_stat.f_fstypename, "ufs", MFSNAMELEN)) { vn_finished_write(mp); fdrop(fp, td); return (EINVAL); } ump = VFSTOUFS(mp); if ((mp->mnt_flag & MNT_RDONLY) && ump->um_fsckpid != td->td_proc->p_pid) { vn_finished_write(mp); fdrop(fp, td); return (EROFS); } fs = ump->um_fs; filetype = IFREG; switch (oidp->oid_number) { case FFS_SET_FLAGS: #ifdef DEBUG if (fsckcmds) printf("%s: %s flags\n", mp->mnt_stat.f_mntonname, cmd.size > 0 ? "set" : "clear"); #endif /* DEBUG */ if (cmd.size > 0) fs->fs_flags |= (long)cmd.value; else fs->fs_flags &= ~(long)cmd.value; break; case FFS_ADJ_REFCNT: #ifdef DEBUG if (fsckcmds) { printf("%s: adjust inode %jd link count by %jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value, (intmax_t)cmd.size); } #endif /* DEBUG */ if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &vp))) break; ip = VTOI(vp); ip->i_nlink += cmd.size; DIP_SET(ip, i_nlink, ip->i_nlink); ip->i_effnlink += cmd.size; ip->i_flag |= IN_CHANGE | IN_MODIFIED; error = ffs_update(vp, 1); if (DOINGSOFTDEP(vp)) softdep_change_linkcnt(ip); vput(vp); break; case FFS_ADJ_BLKCNT: #ifdef DEBUG if (fsckcmds) { printf("%s: adjust inode %jd block count by %jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value, (intmax_t)cmd.size); } #endif /* DEBUG */ if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &vp))) break; ip = VTOI(vp); DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + cmd.size); ip->i_flag |= IN_CHANGE | IN_MODIFIED; error = ffs_update(vp, 1); vput(vp); break; case FFS_DIR_FREE: filetype = IFDIR; /* fall through */ case FFS_FILE_FREE: #ifdef DEBUG if (fsckcmds) { if (cmd.size == 1) printf("%s: free %s inode %ju\n", mp->mnt_stat.f_mntonname, filetype == IFDIR ? "directory" : "file", (uintmax_t)cmd.value); else printf("%s: free %s inodes %ju-%ju\n", mp->mnt_stat.f_mntonname, filetype == IFDIR ? "directory" : "file", (uintmax_t)cmd.value, (uintmax_t)(cmd.value + cmd.size - 1)); } #endif /* DEBUG */ while (cmd.size > 0) { if ((error = ffs_freefile(ump, fs, ump->um_devvp, cmd.value, filetype, NULL))) break; cmd.size -= 1; cmd.value += 1; } break; case FFS_BLK_FREE: #ifdef DEBUG if (fsckcmds) { if (cmd.size == 1) printf("%s: free block %jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); else printf("%s: free blocks %jd-%jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value, (intmax_t)cmd.value + cmd.size - 1); } #endif /* DEBUG */ blkno = cmd.value; blkcnt = cmd.size; blksize = fs->fs_frag - (blkno % fs->fs_frag); while (blkcnt > 0) { if (blksize > blkcnt) blksize = blkcnt; ffs_blkfree(ump, fs, ump->um_devvp, blkno, blksize * fs->fs_fsize, UFS_ROOTINO, VDIR, NULL); blkno += blksize; blkcnt -= blksize; blksize = fs->fs_frag; } break; /* * Adjust superblock summaries. fsck(8) is expected to * submit deltas when necessary. */ case FFS_ADJ_NDIR: #ifdef DEBUG if (fsckcmds) { printf("%s: adjust number of directories by %jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); } #endif /* DEBUG */ fs->fs_cstotal.cs_ndir += cmd.value; break; case FFS_ADJ_NBFREE: #ifdef DEBUG if (fsckcmds) { printf("%s: adjust number of free blocks by %+jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); } #endif /* DEBUG */ fs->fs_cstotal.cs_nbfree += cmd.value; break; case FFS_ADJ_NIFREE: #ifdef DEBUG if (fsckcmds) { printf("%s: adjust number of free inodes by %+jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); } #endif /* DEBUG */ fs->fs_cstotal.cs_nifree += cmd.value; break; case FFS_ADJ_NFFREE: #ifdef DEBUG if (fsckcmds) { printf("%s: adjust number of free frags by %+jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); } #endif /* DEBUG */ fs->fs_cstotal.cs_nffree += cmd.value; break; case FFS_ADJ_NUMCLUSTERS: #ifdef DEBUG if (fsckcmds) { printf("%s: adjust number of free clusters by %+jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); } #endif /* DEBUG */ fs->fs_cstotal.cs_numclusters += cmd.value; break; case FFS_SET_CWD: #ifdef DEBUG if (fsckcmds) { printf("%s: set current directory to inode %jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); } #endif /* DEBUG */ if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_SHARED, &vp))) break; AUDIT_ARG_VNODE1(vp); if ((error = change_dir(vp, td)) != 0) { vput(vp); break; } VOP_UNLOCK(vp, 0); pwd_chdir(td, vp); break; case FFS_SET_DOTDOT: #ifdef DEBUG if (fsckcmds) { printf("%s: change .. in cwd from %jd to %jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value, (intmax_t)cmd.size); } #endif /* DEBUG */ /* * First we have to get and lock the parent directory * to which ".." points. */ error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &fdvp); if (error) break; /* * Now we get and lock the child directory containing "..". */ FILEDESC_SLOCK(td->td_proc->p_fd); dvp = td->td_proc->p_fd->fd_cdir; FILEDESC_SUNLOCK(td->td_proc->p_fd); if ((error = vget(dvp, LK_EXCLUSIVE, td)) != 0) { vput(fdvp); break; } dp = VTOI(dvp); dp->i_offset = 12; /* XXX mastertemplate.dot_reclen */ error = ufs_dirrewrite(dp, VTOI(fdvp), (ino_t)cmd.size, DT_DIR, 0); cache_purge(fdvp); cache_purge(dvp); vput(dvp); vput(fdvp); break; case FFS_UNLINK: #ifdef DEBUG if (fsckcmds) { char buf[32]; if (copyinstr((char *)(intptr_t)cmd.value, buf,32,NULL)) strncpy(buf, "Name_too_long", 32); printf("%s: unlink %s (inode %jd)\n", mp->mnt_stat.f_mntonname, buf, (intmax_t)cmd.size); } #endif /* DEBUG */ /* * kern_unlinkat will do its own start/finish writes and * they do not nest, so drop ours here. Setting mp == NULL * indicates that vn_finished_write is not needed down below. */ vn_finished_write(mp); mp = NULL; error = kern_unlinkat(td, AT_FDCWD, (char *)(intptr_t)cmd.value, UIO_USERSPACE, (ino_t)cmd.size); break; case FFS_SET_INODE: if (ump->um_fsckpid != td->td_proc->p_pid) { error = EPERM; break; } #ifdef DEBUG if (fsckcmds) { printf("%s: update inode %jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); } #endif /* DEBUG */ if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &vp))) break; AUDIT_ARG_VNODE1(vp); ip = VTOI(vp); if (I_IS_UFS1(ip)) error = copyin((void *)(intptr_t)cmd.size, ip->i_din1, sizeof(struct ufs1_dinode)); else error = copyin((void *)(intptr_t)cmd.size, ip->i_din2, sizeof(struct ufs2_dinode)); if (error) { vput(vp); break; } ip->i_flag |= IN_CHANGE | IN_MODIFIED; error = ffs_update(vp, 1); vput(vp); break; case FFS_SET_BUFOUTPUT: if (ump->um_fsckpid != td->td_proc->p_pid) { error = EPERM; break; } if (ITOUMP(VTOI(vp)) != ump) { error = EINVAL; break; } #ifdef DEBUG if (fsckcmds) { printf("%s: %s buffered output for descriptor %jd\n", mp->mnt_stat.f_mntonname, cmd.size == 1 ? "enable" : "disable", (intmax_t)cmd.value); } #endif /* DEBUG */ if ((error = getvnode(td, cmd.value, cap_rights_init(&rights, CAP_FSCK), &vfp)) != 0) break; if (vfp->f_vnode->v_type != VCHR) { fdrop(vfp, td); error = EINVAL; break; } if (origops == NULL) { origops = vfp->f_ops; bcopy((void *)origops, (void *)&bufferedops, sizeof(bufferedops)); bufferedops.fo_write = buffered_write; } if (cmd.size == 1) atomic_store_rel_ptr((volatile uintptr_t *)&vfp->f_ops, (uintptr_t)&bufferedops); else atomic_store_rel_ptr((volatile uintptr_t *)&vfp->f_ops, (uintptr_t)origops); fdrop(vfp, td); break; default: #ifdef DEBUG if (fsckcmds) { printf("Invalid request %d from fsck\n", oidp->oid_number); } #endif /* DEBUG */ error = EINVAL; break; } fdrop(fp, td); vn_finished_write(mp); return (error); } /* * Function to switch a descriptor to use the buffer cache to stage * its I/O. This is needed so that writes to the filesystem device * will give snapshots a chance to copy modified blocks for which it * needs to retain copies. */ static int buffered_write(fp, uio, active_cred, flags, td) struct file *fp; struct uio *uio; struct ucred *active_cred; int flags; struct thread *td; { struct vnode *devvp, *vp; struct inode *ip; struct buf *bp; struct fs *fs; struct filedesc *fdp; int error; daddr_t lbn; /* * The devvp is associated with the /dev filesystem. To discover * the filesystem with which the device is associated, we depend * on the application setting the current directory to a location * within the filesystem being written. Yes, this is an ugly hack. */ devvp = fp->f_vnode; if (!vn_isdisk(devvp, NULL)) return (EINVAL); fdp = td->td_proc->p_fd; FILEDESC_SLOCK(fdp); vp = fdp->fd_cdir; vref(vp); FILEDESC_SUNLOCK(fdp); vn_lock(vp, LK_SHARED | LK_RETRY); /* * Check that the current directory vnode indeed belongs to * UFS before trying to dereference UFS-specific v_data fields. */ if (vp->v_op != &ffs_vnodeops1 && vp->v_op != &ffs_vnodeops2) { vput(vp); return (EINVAL); } ip = VTOI(vp); if (ITODEVVP(ip) != devvp) { vput(vp); return (EINVAL); } fs = ITOFS(ip); vput(vp); foffset_lock_uio(fp, uio, flags); vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); #ifdef DEBUG if (fsckcmds) { printf("%s: buffered write for block %jd\n", fs->fs_fsmnt, (intmax_t)btodb(uio->uio_offset)); } #endif /* DEBUG */ /* * All I/O must be contained within a filesystem block, start on * a fragment boundary, and be a multiple of fragments in length. */ if (uio->uio_resid > fs->fs_bsize - (uio->uio_offset % fs->fs_bsize) || fragoff(fs, uio->uio_offset) != 0 || fragoff(fs, uio->uio_resid) != 0) { error = EINVAL; goto out; } lbn = numfrags(fs, uio->uio_offset); bp = getblk(devvp, lbn, uio->uio_resid, 0, 0, 0); bp->b_flags |= B_RELBUF; if ((error = uiomove((char *)bp->b_data, uio->uio_resid, uio)) != 0) { brelse(bp); goto out; } error = bwrite(bp); out: VOP_UNLOCK(devvp, 0); foffset_unlock_uio(fp, uio, flags | FOF_NEXTOFF); return (error); } Index: head/sys/ufs/ffs/ffs_vfsops.c =================================================================== --- head/sys/ufs/ffs/ffs_vfsops.c (revision 328913) +++ head/sys/ufs/ffs/ffs_vfsops.c (revision 328914) @@ -1,2300 +1,2316 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1989, 1991, 1993, 1994 * 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. * 3. 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. * * @(#)ffs_vfsops.c 8.31 (Berkeley) 5/20/95 */ #include __FBSDID("$FreeBSD$"); #include "opt_quota.h" #include "opt_ufs.h" #include "opt_ffs.h" #include "opt_ddb.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 #include #include #include static uma_zone_t uma_inode, uma_ufs1, uma_ufs2; static int ffs_mountfs(struct vnode *, struct mount *, struct thread *); static void ffs_oldfscompat_read(struct fs *, struct ufsmount *, ufs2_daddr_t); static void ffs_ifree(struct ufsmount *ump, struct inode *ip); static int ffs_sync_lazy(struct mount *mp); static int ffs_use_bread(void *devfd, off_t loc, void **bufp, int size); static int ffs_use_bwrite(void *devfd, off_t loc, void *buf, int size); static vfs_init_t ffs_init; static vfs_uninit_t ffs_uninit; static vfs_extattrctl_t ffs_extattrctl; static vfs_cmount_t ffs_cmount; static vfs_unmount_t ffs_unmount; static vfs_mount_t ffs_mount; static vfs_statfs_t ffs_statfs; static vfs_fhtovp_t ffs_fhtovp; static vfs_sync_t ffs_sync; static struct vfsops ufs_vfsops = { .vfs_extattrctl = ffs_extattrctl, .vfs_fhtovp = ffs_fhtovp, .vfs_init = ffs_init, .vfs_mount = ffs_mount, .vfs_cmount = ffs_cmount, .vfs_quotactl = ufs_quotactl, .vfs_root = ufs_root, .vfs_statfs = ffs_statfs, .vfs_sync = ffs_sync, .vfs_uninit = ffs_uninit, .vfs_unmount = ffs_unmount, .vfs_vget = ffs_vget, .vfs_susp_clean = process_deferred_inactive, }; VFS_SET(ufs_vfsops, ufs, 0); MODULE_VERSION(ufs, 1); static b_strategy_t ffs_geom_strategy; static b_write_t ffs_bufwrite; static struct buf_ops ffs_ops = { .bop_name = "FFS", .bop_write = ffs_bufwrite, .bop_strategy = ffs_geom_strategy, .bop_sync = bufsync, #ifdef NO_FFS_SNAPSHOT .bop_bdflush = bufbdflush, #else .bop_bdflush = ffs_bdflush, #endif }; /* * Note that userquota and groupquota options are not currently used * by UFS/FFS code and generally mount(8) does not pass those options * from userland, but they can be passed by loader(8) via * vfs.root.mountfrom.options. */ static const char *ffs_opts[] = { "acls", "async", "noatime", "noclusterr", "noclusterw", "noexec", "export", "force", "from", "groupquota", "multilabel", "nfsv4acls", "fsckpid", "snapshot", "nosuid", "suiddir", "nosymfollow", "sync", "union", "userquota", NULL }; static int ffs_mount(struct mount *mp) { struct vnode *devvp; struct thread *td; struct ufsmount *ump = NULL; struct fs *fs; pid_t fsckpid = 0; int error, error1, flags; uint64_t mntorflags; accmode_t accmode; struct nameidata ndp; char *fspec; td = curthread; if (vfs_filteropt(mp->mnt_optnew, ffs_opts)) return (EINVAL); if (uma_inode == NULL) { uma_inode = uma_zcreate("FFS inode", sizeof(struct inode), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); uma_ufs1 = uma_zcreate("FFS1 dinode", sizeof(struct ufs1_dinode), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); uma_ufs2 = uma_zcreate("FFS2 dinode", sizeof(struct ufs2_dinode), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); } vfs_deleteopt(mp->mnt_optnew, "groupquota"); vfs_deleteopt(mp->mnt_optnew, "userquota"); fspec = vfs_getopts(mp->mnt_optnew, "from", &error); if (error) return (error); mntorflags = 0; if (vfs_getopt(mp->mnt_optnew, "acls", NULL, NULL) == 0) mntorflags |= MNT_ACLS; if (vfs_getopt(mp->mnt_optnew, "snapshot", NULL, NULL) == 0) { mntorflags |= MNT_SNAPSHOT; /* * Once we have set the MNT_SNAPSHOT flag, do not * persist "snapshot" in the options list. */ vfs_deleteopt(mp->mnt_optnew, "snapshot"); vfs_deleteopt(mp->mnt_opt, "snapshot"); } if (vfs_getopt(mp->mnt_optnew, "fsckpid", NULL, NULL) == 0 && vfs_scanopt(mp->mnt_optnew, "fsckpid", "%d", &fsckpid) == 1) { /* * Once we have set the restricted PID, do not * persist "fsckpid" in the options list. */ vfs_deleteopt(mp->mnt_optnew, "fsckpid"); vfs_deleteopt(mp->mnt_opt, "fsckpid"); if (mp->mnt_flag & MNT_UPDATE) { if (VFSTOUFS(mp)->um_fs->fs_ronly == 0 && vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) == 0) { vfs_mount_error(mp, "Checker enable: Must be read-only"); return (EINVAL); } } else if (vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) == 0) { vfs_mount_error(mp, "Checker enable: Must be read-only"); return (EINVAL); } /* Set to -1 if we are done */ if (fsckpid == 0) fsckpid = -1; } if (vfs_getopt(mp->mnt_optnew, "nfsv4acls", NULL, NULL) == 0) { if (mntorflags & MNT_ACLS) { vfs_mount_error(mp, "\"acls\" and \"nfsv4acls\" options " "are mutually exclusive"); return (EINVAL); } mntorflags |= MNT_NFS4ACLS; } MNT_ILOCK(mp); mp->mnt_flag |= mntorflags; MNT_IUNLOCK(mp); /* * If updating, check whether changing from read-only to * read/write; if there is no device name, that's all we do. */ if (mp->mnt_flag & MNT_UPDATE) { ump = VFSTOUFS(mp); fs = ump->um_fs; devvp = ump->um_devvp; if (fsckpid == -1 && ump->um_fsckpid > 0) { if ((error = ffs_flushfiles(mp, WRITECLOSE, td)) != 0 || (error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) return (error); g_topology_lock(); /* * Return to normal read-only mode. */ error = g_access(ump->um_cp, 0, -1, 0); g_topology_unlock(); ump->um_fsckpid = 0; } if (fs->fs_ronly == 0 && vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) { /* * Flush any dirty data and suspend filesystem. */ if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0) return (error); error = vfs_write_suspend_umnt(mp); if (error != 0) return (error); /* * Check for and optionally get rid of files open * for writing. */ flags = WRITECLOSE; if (mp->mnt_flag & MNT_FORCE) flags |= FORCECLOSE; if (MOUNTEDSOFTDEP(mp)) { error = softdep_flushfiles(mp, flags, td); } else { error = ffs_flushfiles(mp, flags, td); } if (error) { vfs_write_resume(mp, 0); return (error); } if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { printf("WARNING: %s Update error: blocks %jd " "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, fs->fs_pendinginodes); fs->fs_pendingblocks = 0; fs->fs_pendinginodes = 0; } if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0) fs->fs_clean = 1; if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) { fs->fs_ronly = 0; fs->fs_clean = 0; vfs_write_resume(mp, 0); return (error); } if (MOUNTEDSOFTDEP(mp)) softdep_unmount(mp); g_topology_lock(); /* * Drop our write and exclusive access. */ g_access(ump->um_cp, 0, -1, -1); g_topology_unlock(); fs->fs_ronly = 1; MNT_ILOCK(mp); mp->mnt_flag |= MNT_RDONLY; MNT_IUNLOCK(mp); /* * Allow the writers to note that filesystem * is ro now. */ vfs_write_resume(mp, 0); } if ((mp->mnt_flag & MNT_RELOAD) && (error = ffs_reload(mp, td, 0)) != 0) return (error); if (fs->fs_ronly && !vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) { /* * If we are running a checker, do not allow upgrade. */ if (ump->um_fsckpid > 0) { vfs_mount_error(mp, "Active checker, cannot upgrade to write"); return (EINVAL); } /* * If upgrade to read-write by non-root, then verify * that user has necessary permissions on the device. */ vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); error = VOP_ACCESS(devvp, VREAD | VWRITE, td->td_ucred, td); if (error) error = priv_check(td, PRIV_VFS_MOUNT_PERM); if (error) { VOP_UNLOCK(devvp, 0); return (error); } VOP_UNLOCK(devvp, 0); fs->fs_flags &= ~FS_UNCLEAN; if (fs->fs_clean == 0) { fs->fs_flags |= FS_UNCLEAN; if ((mp->mnt_flag & MNT_FORCE) || ((fs->fs_flags & (FS_SUJ | FS_NEEDSFSCK)) == 0 && (fs->fs_flags & FS_DOSOFTDEP))) { printf("WARNING: %s was not properly " "dismounted\n", fs->fs_fsmnt); } else { vfs_mount_error(mp, "R/W mount of %s denied. %s.%s", fs->fs_fsmnt, "Filesystem is not clean - run fsck", (fs->fs_flags & FS_SUJ) == 0 ? "" : " Forced mount will invalidate" " journal contents"); return (EPERM); } } g_topology_lock(); /* * Request exclusive write access. */ error = g_access(ump->um_cp, 0, 1, 1); g_topology_unlock(); if (error) return (error); if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0) return (error); fs->fs_ronly = 0; MNT_ILOCK(mp); mp->mnt_flag &= ~MNT_RDONLY; MNT_IUNLOCK(mp); fs->fs_mtime = time_second; /* check to see if we need to start softdep */ if ((fs->fs_flags & FS_DOSOFTDEP) && (error = softdep_mount(devvp, mp, fs, td->td_ucred))){ vn_finished_write(mp); return (error); } fs->fs_clean = 0; if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) { vn_finished_write(mp); return (error); } if (fs->fs_snapinum[0] != 0) ffs_snapshot_mount(mp); vn_finished_write(mp); } /* * Soft updates is incompatible with "async", * so if we are doing softupdates stop the user * from setting the async flag in an update. * Softdep_mount() clears it in an initial mount * or ro->rw remount. */ if (MOUNTEDSOFTDEP(mp)) { /* XXX: Reset too late ? */ MNT_ILOCK(mp); mp->mnt_flag &= ~MNT_ASYNC; MNT_IUNLOCK(mp); } /* * Keep MNT_ACLS flag if it is stored in superblock. */ if ((fs->fs_flags & FS_ACLS) != 0) { /* XXX: Set too late ? */ MNT_ILOCK(mp); mp->mnt_flag |= MNT_ACLS; MNT_IUNLOCK(mp); } if ((fs->fs_flags & FS_NFS4ACLS) != 0) { /* XXX: Set too late ? */ MNT_ILOCK(mp); mp->mnt_flag |= MNT_NFS4ACLS; MNT_IUNLOCK(mp); } /* * If this is a request from fsck to clean up the filesystem, * then allow the specified pid to proceed. */ if (fsckpid > 0) { if (ump->um_fsckpid != 0) { vfs_mount_error(mp, "Active checker already running on %s", fs->fs_fsmnt); return (EINVAL); } KASSERT(MOUNTEDSOFTDEP(mp) == 0, ("soft updates enabled on read-only file system")); g_topology_lock(); /* * Request write access. */ error = g_access(ump->um_cp, 0, 1, 0); g_topology_unlock(); if (error) { vfs_mount_error(mp, "Checker activation failed on %s", fs->fs_fsmnt); return (error); } ump->um_fsckpid = fsckpid; if (fs->fs_snapinum[0] != 0) ffs_snapshot_mount(mp); fs->fs_mtime = time_second; fs->fs_fmod = 1; fs->fs_clean = 0; (void) ffs_sbupdate(ump, MNT_WAIT, 0); } /* * If this is a snapshot request, take the snapshot. */ if (mp->mnt_flag & MNT_SNAPSHOT) return (ffs_snapshot(mp, fspec)); /* * Must not call namei() while owning busy ref. */ vfs_unbusy(mp); } /* * Not an update, or updating the name: look up the name * and verify that it refers to a sensible disk device. */ NDINIT(&ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec, td); error = namei(&ndp); if ((mp->mnt_flag & MNT_UPDATE) != 0) { /* * Unmount does not start if MNT_UPDATE is set. Mount * update busies mp before setting MNT_UPDATE. We * must be able to retain our busy ref succesfully, * without sleep. */ error1 = vfs_busy(mp, MBF_NOWAIT); MPASS(error1 == 0); } if (error != 0) return (error); NDFREE(&ndp, NDF_ONLY_PNBUF); devvp = ndp.ni_vp; if (!vn_isdisk(devvp, &error)) { vput(devvp); return (error); } /* * If mount by non-root, then verify that user has necessary * permissions on the device. */ accmode = VREAD; if ((mp->mnt_flag & MNT_RDONLY) == 0) accmode |= VWRITE; error = VOP_ACCESS(devvp, accmode, td->td_ucred, td); if (error) error = priv_check(td, PRIV_VFS_MOUNT_PERM); if (error) { vput(devvp); return (error); } if (mp->mnt_flag & MNT_UPDATE) { /* * Update only * * If it's not the same vnode, or at least the same device * then it's not correct. */ if (devvp->v_rdev != ump->um_devvp->v_rdev) error = EINVAL; /* needs translation */ vput(devvp); if (error) return (error); } else { /* * New mount * * We need the name for the mount point (also used for * "last mounted on") copied in. If an error occurs, * the mount point is discarded by the upper level code. * Note that vfs_mount_alloc() populates f_mntonname for us. */ if ((error = ffs_mountfs(devvp, mp, td)) != 0) { vrele(devvp); return (error); } if (fsckpid > 0) { KASSERT(MOUNTEDSOFTDEP(mp) == 0, ("soft updates enabled on read-only file system")); ump = VFSTOUFS(mp); fs = ump->um_fs; g_topology_lock(); /* * Request write access. */ error = g_access(ump->um_cp, 0, 1, 0); g_topology_unlock(); if (error) { printf("WARNING: %s: Checker activation " "failed\n", fs->fs_fsmnt); } else { ump->um_fsckpid = fsckpid; if (fs->fs_snapinum[0] != 0) ffs_snapshot_mount(mp); fs->fs_mtime = time_second; fs->fs_clean = 0; (void) ffs_sbupdate(ump, MNT_WAIT, 0); } } } vfs_mountedfrom(mp, fspec); return (0); } /* * Compatibility with old mount system call. */ static int ffs_cmount(struct mntarg *ma, void *data, uint64_t flags) { struct ufs_args args; struct export_args exp; int error; if (data == NULL) return (EINVAL); error = copyin(data, &args, sizeof args); if (error) return (error); vfs_oexport_conv(&args.export, &exp); ma = mount_argsu(ma, "from", args.fspec, MAXPATHLEN); ma = mount_arg(ma, "export", &exp, sizeof(exp)); error = kernel_mount(ma, flags); return (error); } /* * Reload all incore data for a filesystem (used after running fsck on * the root filesystem and finding things to fix). If the 'force' flag * is 0, the filesystem must be mounted read-only. * * Things to do to update the mount: * 1) invalidate all cached meta-data. * 2) re-read superblock from disk. * 3) re-read summary information from disk. * 4) invalidate all inactive vnodes. * 5) clear MNTK_SUSPEND2 and MNTK_SUSPENDED flags, allowing secondary * writers, if requested. * 6) invalidate all cached file data. * 7) re-read inode data for all active vnodes. */ int ffs_reload(struct mount *mp, struct thread *td, int flags) { struct vnode *vp, *mvp, *devvp; struct inode *ip; void *space; struct buf *bp; struct fs *fs, *newfs; struct ufsmount *ump; ufs2_daddr_t sblockloc; int i, blks, error; u_long size; int32_t *lp; ump = VFSTOUFS(mp); MNT_ILOCK(mp); if ((mp->mnt_flag & MNT_RDONLY) == 0 && (flags & FFSR_FORCE) == 0) { MNT_IUNLOCK(mp); return (EINVAL); } MNT_IUNLOCK(mp); /* * Step 1: invalidate all cached meta-data. */ devvp = VFSTOUFS(mp)->um_devvp; vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); if (vinvalbuf(devvp, 0, 0, 0) != 0) panic("ffs_reload: dirty1"); VOP_UNLOCK(devvp, 0); /* * Step 2: re-read superblock from disk. */ fs = VFSTOUFS(mp)->um_fs; if ((error = bread(devvp, btodb(fs->fs_sblockloc), fs->fs_sbsize, NOCRED, &bp)) != 0) return (error); newfs = (struct fs *)bp->b_data; if ((newfs->fs_magic != FS_UFS1_MAGIC && newfs->fs_magic != FS_UFS2_MAGIC) || newfs->fs_bsize > MAXBSIZE || newfs->fs_bsize < sizeof(struct fs)) { brelse(bp); return (EIO); /* XXX needs translation */ } /* * Copy pointer fields back into superblock before copying in XXX * new superblock. These should really be in the ufsmount. XXX * Note that important parameters (eg fs_ncg) are unchanged. */ newfs->fs_csp = fs->fs_csp; newfs->fs_maxcluster = fs->fs_maxcluster; newfs->fs_contigdirs = fs->fs_contigdirs; newfs->fs_active = fs->fs_active; newfs->fs_ronly = fs->fs_ronly; sblockloc = fs->fs_sblockloc; bcopy(newfs, fs, (u_int)fs->fs_sbsize); brelse(bp); mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen; ffs_oldfscompat_read(fs, VFSTOUFS(mp), sblockloc); UFS_LOCK(ump); if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { printf("WARNING: %s: reload pending error: blocks %jd " "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, fs->fs_pendinginodes); fs->fs_pendingblocks = 0; fs->fs_pendinginodes = 0; } UFS_UNLOCK(ump); /* * Step 3: re-read summary information from disk. */ size = fs->fs_cssize; blks = howmany(size, fs->fs_fsize); if (fs->fs_contigsumsize > 0) size += fs->fs_ncg * sizeof(int32_t); size += fs->fs_ncg * sizeof(u_int8_t); free(fs->fs_csp, M_UFSMNT); space = malloc(size, M_UFSMNT, M_WAITOK); fs->fs_csp = space; for (i = 0; i < blks; i += fs->fs_frag) { size = fs->fs_bsize; if (i + fs->fs_frag > blks) size = (blks - i) * fs->fs_fsize; error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size, NOCRED, &bp); if (error) return (error); bcopy(bp->b_data, space, (u_int)size); space = (char *)space + size; brelse(bp); } /* * We no longer know anything about clusters per cylinder group. */ if (fs->fs_contigsumsize > 0) { fs->fs_maxcluster = lp = space; for (i = 0; i < fs->fs_ncg; i++) *lp++ = fs->fs_contigsumsize; space = lp; } size = fs->fs_ncg * sizeof(u_int8_t); fs->fs_contigdirs = (u_int8_t *)space; bzero(fs->fs_contigdirs, size); if ((flags & FFSR_UNSUSPEND) != 0) { MNT_ILOCK(mp); mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2); wakeup(&mp->mnt_flag); MNT_IUNLOCK(mp); } loop: MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { /* * Skip syncer vnode. */ if (vp->v_type == VNON) { VI_UNLOCK(vp); continue; } /* * Step 4: invalidate all cached file data. */ if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) { MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); goto loop; } if (vinvalbuf(vp, 0, 0, 0)) panic("ffs_reload: dirty2"); /* * Step 5: re-read inode data for all active vnodes. */ ip = VTOI(vp); error = bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)), (int)fs->fs_bsize, NOCRED, &bp); if (error) { VOP_UNLOCK(vp, 0); vrele(vp); MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); return (error); } ffs_load_inode(bp, ip, fs, ip->i_number); ip->i_effnlink = ip->i_nlink; brelse(bp); VOP_UNLOCK(vp, 0); vrele(vp); } return (0); } /* * Common code for mount and mountroot */ static int ffs_mountfs(devvp, mp, td) struct vnode *devvp; struct mount *mp; struct thread *td; { struct ufsmount *ump; struct fs *fs; struct cdev *dev; int error, i, len, ronly; struct ucred *cred; struct g_consumer *cp; struct mount *nmp; fs = NULL; ump = NULL; cred = td ? td->td_ucred : NOCRED; ronly = (mp->mnt_flag & MNT_RDONLY) != 0; KASSERT(devvp->v_type == VCHR, ("reclaimed devvp")); dev = devvp->v_rdev; if (atomic_cmpset_acq_ptr((uintptr_t *)&dev->si_mountpt, 0, (uintptr_t)mp) == 0) { VOP_UNLOCK(devvp, 0); return (EBUSY); } g_topology_lock(); error = g_vfs_open(devvp, &cp, "ffs", ronly ? 0 : 1); g_topology_unlock(); if (error != 0) { atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0); VOP_UNLOCK(devvp, 0); return (error); } dev_ref(dev); devvp->v_bufobj.bo_ops = &ffs_ops; VOP_UNLOCK(devvp, 0); if (dev->si_iosize_max != 0) mp->mnt_iosize_max = dev->si_iosize_max; if (mp->mnt_iosize_max > MAXPHYS) mp->mnt_iosize_max = MAXPHYS; if ((SBLOCKSIZE % cp->provider->sectorsize) != 0) { error = EINVAL; vfs_mount_error(mp, "Invalid sectorsize %d for superblock size %d", cp->provider->sectorsize, SBLOCKSIZE); goto out; } /* fetch the superblock and summary information */ if ((error = ffs_sbget(devvp, &fs, -1, M_UFSMNT, ffs_use_bread)) != 0) goto out; fs->fs_fmod = 0; /* none of these types of check-hashes are maintained */ fs->fs_metackhash &= ~(CK_SUPERBLOCK | CK_INODE | CK_INDIR | CK_DIR); /* no support for directory indices or any other undefined flags */ fs->fs_flags &= ~FS_INDEXDIRS; fs->fs_flags &= FS_SUPPORTED; fs->fs_flags &= ~FS_UNCLEAN; if (fs->fs_clean == 0) { fs->fs_flags |= FS_UNCLEAN; if (ronly || (mp->mnt_flag & MNT_FORCE) || ((fs->fs_flags & (FS_SUJ | FS_NEEDSFSCK)) == 0 && (fs->fs_flags & FS_DOSOFTDEP))) { printf("WARNING: %s was not properly dismounted\n", fs->fs_fsmnt); } else { vfs_mount_error(mp, "R/W mount of %s denied. %s%s", fs->fs_fsmnt, "Filesystem is not clean - run fsck.", (fs->fs_flags & FS_SUJ) == 0 ? "" : " Forced mount will invalidate journal contents"); error = EPERM; goto out; } if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) && (mp->mnt_flag & MNT_FORCE)) { printf("WARNING: %s: lost blocks %jd files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, fs->fs_pendinginodes); fs->fs_pendingblocks = 0; fs->fs_pendinginodes = 0; } } if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { printf("WARNING: %s: mount pending error: blocks %jd " "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, fs->fs_pendinginodes); fs->fs_pendingblocks = 0; fs->fs_pendinginodes = 0; } if ((fs->fs_flags & FS_GJOURNAL) != 0) { #ifdef UFS_GJOURNAL /* * Get journal provider name. */ len = 1024; mp->mnt_gjprovider = malloc((u_long)len, M_UFSMNT, M_WAITOK); if (g_io_getattr("GJOURNAL::provider", cp, &len, mp->mnt_gjprovider) == 0) { mp->mnt_gjprovider = realloc(mp->mnt_gjprovider, len, M_UFSMNT, M_WAITOK); MNT_ILOCK(mp); mp->mnt_flag |= MNT_GJOURNAL; MNT_IUNLOCK(mp); } else { printf("WARNING: %s: GJOURNAL flag on fs " "but no gjournal provider below\n", mp->mnt_stat.f_mntonname); free(mp->mnt_gjprovider, M_UFSMNT); mp->mnt_gjprovider = NULL; } #else printf("WARNING: %s: GJOURNAL flag on fs but no " "UFS_GJOURNAL support\n", mp->mnt_stat.f_mntonname); #endif } else { mp->mnt_gjprovider = NULL; } ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO); ump->um_cp = cp; ump->um_bo = &devvp->v_bufobj; ump->um_fs = fs; if (fs->fs_magic == FS_UFS1_MAGIC) { ump->um_fstype = UFS1; ump->um_balloc = ffs_balloc_ufs1; } else { ump->um_fstype = UFS2; ump->um_balloc = ffs_balloc_ufs2; } ump->um_blkatoff = ffs_blkatoff; ump->um_truncate = ffs_truncate; ump->um_update = ffs_update; ump->um_valloc = ffs_valloc; ump->um_vfree = ffs_vfree; ump->um_ifree = ffs_ifree; ump->um_rdonly = ffs_rdonly; ump->um_snapgone = ffs_snapgone; mtx_init(UFS_MTX(ump), "FFS", "FFS Lock", MTX_DEF); ffs_oldfscompat_read(fs, ump, fs->fs_sblockloc); fs->fs_ronly = ronly; fs->fs_active = NULL; mp->mnt_data = ump; mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0]; mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1]; nmp = NULL; if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 || (nmp = vfs_getvfs(&mp->mnt_stat.f_fsid))) { if (nmp) vfs_rel(nmp); vfs_getnewfsid(mp); } mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen; MNT_ILOCK(mp); mp->mnt_flag |= MNT_LOCAL; MNT_IUNLOCK(mp); if ((fs->fs_flags & FS_MULTILABEL) != 0) { #ifdef MAC MNT_ILOCK(mp); mp->mnt_flag |= MNT_MULTILABEL; MNT_IUNLOCK(mp); #else printf("WARNING: %s: multilabel flag on fs but " "no MAC support\n", mp->mnt_stat.f_mntonname); #endif } if ((fs->fs_flags & FS_ACLS) != 0) { #ifdef UFS_ACL MNT_ILOCK(mp); if (mp->mnt_flag & MNT_NFS4ACLS) printf("WARNING: %s: ACLs flag on fs conflicts with " "\"nfsv4acls\" mount option; option ignored\n", mp->mnt_stat.f_mntonname); mp->mnt_flag &= ~MNT_NFS4ACLS; mp->mnt_flag |= MNT_ACLS; MNT_IUNLOCK(mp); #else printf("WARNING: %s: ACLs flag on fs but no ACLs support\n", mp->mnt_stat.f_mntonname); #endif } if ((fs->fs_flags & FS_NFS4ACLS) != 0) { #ifdef UFS_ACL MNT_ILOCK(mp); if (mp->mnt_flag & MNT_ACLS) printf("WARNING: %s: NFSv4 ACLs flag on fs conflicts " "with \"acls\" mount option; option ignored\n", mp->mnt_stat.f_mntonname); mp->mnt_flag &= ~MNT_ACLS; mp->mnt_flag |= MNT_NFS4ACLS; MNT_IUNLOCK(mp); #else printf("WARNING: %s: NFSv4 ACLs flag on fs but no " "ACLs support\n", mp->mnt_stat.f_mntonname); #endif } if ((fs->fs_flags & FS_TRIM) != 0) { len = sizeof(int); if (g_io_getattr("GEOM::candelete", cp, &len, &ump->um_candelete) == 0) { if (!ump->um_candelete) printf("WARNING: %s: TRIM flag on fs but disk " "does not support TRIM\n", mp->mnt_stat.f_mntonname); } else { printf("WARNING: %s: TRIM flag on fs but disk does " "not confirm that it supports TRIM\n", mp->mnt_stat.f_mntonname); ump->um_candelete = 0; } if (ump->um_candelete) { ump->um_trim_tq = taskqueue_create("trim", M_WAITOK, taskqueue_thread_enqueue, &ump->um_trim_tq); taskqueue_start_threads(&ump->um_trim_tq, 1, PVFS, "%s trim", mp->mnt_stat.f_mntonname); } } ump->um_mountp = mp; ump->um_dev = dev; ump->um_devvp = devvp; ump->um_nindir = fs->fs_nindir; ump->um_bptrtodb = fs->fs_fsbtodb; ump->um_seqinc = fs->fs_frag; for (i = 0; i < MAXQUOTAS; i++) ump->um_quotas[i] = NULLVP; #ifdef UFS_EXTATTR ufs_extattr_uepm_init(&ump->um_extattr); #endif /* * Set FS local "last mounted on" information (NULL pad) */ bzero(fs->fs_fsmnt, MAXMNTLEN); strlcpy(fs->fs_fsmnt, mp->mnt_stat.f_mntonname, MAXMNTLEN); mp->mnt_stat.f_iosize = fs->fs_bsize; if (mp->mnt_flag & MNT_ROOTFS) { /* * Root mount; update timestamp in mount structure. * this will be used by the common root mount code * to update the system clock. */ mp->mnt_time = fs->fs_time; } if (ronly == 0) { fs->fs_mtime = time_second; if ((fs->fs_flags & FS_DOSOFTDEP) && (error = softdep_mount(devvp, mp, fs, cred)) != 0) { ffs_flushfiles(mp, FORCECLOSE, td); goto out; } if (fs->fs_snapinum[0] != 0) ffs_snapshot_mount(mp); fs->fs_fmod = 1; fs->fs_clean = 0; (void) ffs_sbupdate(ump, MNT_WAIT, 0); } /* * Initialize filesystem state information in mount struct. */ MNT_ILOCK(mp); mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED | MNTK_NO_IOPF | MNTK_UNMAPPED_BUFS | MNTK_USES_BCACHE; MNT_IUNLOCK(mp); #ifdef UFS_EXTATTR #ifdef UFS_EXTATTR_AUTOSTART /* * * Auto-starting does the following: * - check for /.attribute in the fs, and extattr_start if so * - for each file in .attribute, enable that file with * an attribute of the same name. * Not clear how to report errors -- probably eat them. * This would all happen while the filesystem was busy/not * available, so would effectively be "atomic". */ (void) ufs_extattr_autostart(mp, td); #endif /* !UFS_EXTATTR_AUTOSTART */ #endif /* !UFS_EXTATTR */ return (0); out: if (fs != NULL) { free(fs->fs_csp, M_UFSMNT); free(fs, M_UFSMNT); } if (cp != NULL) { g_topology_lock(); g_vfs_close(cp); g_topology_unlock(); } if (ump) { mtx_destroy(UFS_MTX(ump)); if (mp->mnt_gjprovider != NULL) { free(mp->mnt_gjprovider, M_UFSMNT); mp->mnt_gjprovider = NULL; } free(ump, M_UFSMNT); mp->mnt_data = NULL; } atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0); dev_rel(dev); return (error); } /* * A read function for use by filesystem-layer routines. */ static int ffs_use_bread(void *devfd, off_t loc, void **bufp, int size) { struct buf *bp; int error; *bufp = malloc(size, M_UFSMNT, M_WAITOK); if ((error = bread((struct vnode *)devfd, btodb(loc), size, NOCRED, &bp)) != 0) { free(*bufp, M_UFSMNT); *bufp = NULL; return (error); } bcopy(bp->b_data, *bufp, size); bp->b_flags |= B_INVAL | B_NOCACHE; brelse(bp); return (0); } #include static int bigcgs = 0; SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, ""); /* * Sanity checks for loading old filesystem superblocks. * See ffs_oldfscompat_write below for unwound actions. * * XXX - Parts get retired eventually. * Unfortunately new bits get added. */ static void ffs_oldfscompat_read(fs, ump, sblockloc) struct fs *fs; struct ufsmount *ump; ufs2_daddr_t sblockloc; { off_t maxfilesize; /* * If not yet done, update fs_flags location and value of fs_sblockloc. */ if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) { fs->fs_flags = fs->fs_old_flags; fs->fs_old_flags |= FS_FLAGS_UPDATED; fs->fs_sblockloc = sblockloc; } /* * If not yet done, update UFS1 superblock with new wider fields. */ if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_maxbsize != fs->fs_bsize) { fs->fs_maxbsize = fs->fs_bsize; fs->fs_time = fs->fs_old_time; fs->fs_size = fs->fs_old_size; fs->fs_dsize = fs->fs_old_dsize; fs->fs_csaddr = fs->fs_old_csaddr; fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir; fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree; fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree; fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree; } if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_old_inodefmt < FS_44INODEFMT) { fs->fs_maxfilesize = ((uint64_t)1 << 31) - 1; fs->fs_qbmask = ~fs->fs_bmask; fs->fs_qfmask = ~fs->fs_fmask; } if (fs->fs_magic == FS_UFS1_MAGIC) { ump->um_savedmaxfilesize = fs->fs_maxfilesize; maxfilesize = (uint64_t)0x80000000 * fs->fs_bsize - 1; if (fs->fs_maxfilesize > maxfilesize) fs->fs_maxfilesize = maxfilesize; } /* Compatibility for old filesystems */ if (fs->fs_avgfilesize <= 0) fs->fs_avgfilesize = AVFILESIZ; if (fs->fs_avgfpdir <= 0) fs->fs_avgfpdir = AFPDIR; if (bigcgs) { fs->fs_save_cgsize = fs->fs_cgsize; fs->fs_cgsize = fs->fs_bsize; } } /* * Unwinding superblock updates for old filesystems. * See ffs_oldfscompat_read above for details. * * XXX - Parts get retired eventually. * Unfortunately new bits get added. */ void ffs_oldfscompat_write(fs, ump) struct fs *fs; struct ufsmount *ump; { /* * Copy back UFS2 updated fields that UFS1 inspects. */ if (fs->fs_magic == FS_UFS1_MAGIC) { fs->fs_old_time = fs->fs_time; fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir; fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree; fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree; fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree; fs->fs_maxfilesize = ump->um_savedmaxfilesize; } if (bigcgs) { fs->fs_cgsize = fs->fs_save_cgsize; fs->fs_save_cgsize = 0; } } /* * unmount system call */ static int ffs_unmount(mp, mntflags) struct mount *mp; int mntflags; { struct thread *td; struct ufsmount *ump = VFSTOUFS(mp); struct fs *fs; int error, flags, susp; #ifdef UFS_EXTATTR int e_restart; #endif flags = 0; td = curthread; fs = ump->um_fs; susp = 0; if (mntflags & MNT_FORCE) { flags |= FORCECLOSE; susp = fs->fs_ronly == 0; } #ifdef UFS_EXTATTR if ((error = ufs_extattr_stop(mp, td))) { if (error != EOPNOTSUPP) printf("WARNING: unmount %s: ufs_extattr_stop " "returned errno %d\n", mp->mnt_stat.f_mntonname, error); e_restart = 0; } else { ufs_extattr_uepm_destroy(&ump->um_extattr); e_restart = 1; } #endif if (susp) { error = vfs_write_suspend_umnt(mp); if (error != 0) goto fail1; } if (MOUNTEDSOFTDEP(mp)) error = softdep_flushfiles(mp, flags, td); else error = ffs_flushfiles(mp, flags, td); if (error != 0 && error != ENXIO) goto fail; UFS_LOCK(ump); if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { printf("WARNING: unmount %s: pending error: blocks %jd " "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, fs->fs_pendinginodes); fs->fs_pendingblocks = 0; fs->fs_pendinginodes = 0; } UFS_UNLOCK(ump); if (MOUNTEDSOFTDEP(mp)) softdep_unmount(mp); if (fs->fs_ronly == 0 || ump->um_fsckpid > 0) { fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1; error = ffs_sbupdate(ump, MNT_WAIT, 0); if (error && error != ENXIO) { fs->fs_clean = 0; goto fail; } } if (susp) vfs_write_resume(mp, VR_START_WRITE); if (ump->um_trim_tq != NULL) { while (ump->um_trim_inflight != 0) pause("ufsutr", hz); taskqueue_drain_all(ump->um_trim_tq); taskqueue_free(ump->um_trim_tq); } g_topology_lock(); if (ump->um_fsckpid > 0) { /* * Return to normal read-only mode. */ error = g_access(ump->um_cp, 0, -1, 0); ump->um_fsckpid = 0; } g_vfs_close(ump->um_cp); g_topology_unlock(); atomic_store_rel_ptr((uintptr_t *)&ump->um_dev->si_mountpt, 0); vrele(ump->um_devvp); dev_rel(ump->um_dev); mtx_destroy(UFS_MTX(ump)); if (mp->mnt_gjprovider != NULL) { free(mp->mnt_gjprovider, M_UFSMNT); mp->mnt_gjprovider = NULL; } free(fs->fs_csp, M_UFSMNT); free(fs, M_UFSMNT); free(ump, M_UFSMNT); mp->mnt_data = NULL; MNT_ILOCK(mp); mp->mnt_flag &= ~MNT_LOCAL; MNT_IUNLOCK(mp); if (td->td_su == mp) { td->td_su = NULL; vfs_rel(mp); } return (error); fail: if (susp) vfs_write_resume(mp, VR_START_WRITE); fail1: #ifdef UFS_EXTATTR if (e_restart) { ufs_extattr_uepm_init(&ump->um_extattr); #ifdef UFS_EXTATTR_AUTOSTART (void) ufs_extattr_autostart(mp, td); #endif } #endif return (error); } /* * Flush out all the files in a filesystem. */ int ffs_flushfiles(mp, flags, td) struct mount *mp; int flags; struct thread *td; { struct ufsmount *ump; int qerror, error; ump = VFSTOUFS(mp); qerror = 0; #ifdef QUOTA if (mp->mnt_flag & MNT_QUOTA) { int i; error = vflush(mp, 0, SKIPSYSTEM|flags, td); if (error) return (error); for (i = 0; i < MAXQUOTAS; i++) { error = quotaoff(td, mp, i); if (error != 0) { if ((flags & EARLYFLUSH) == 0) return (error); else qerror = error; } } /* * Here we fall through to vflush again to ensure that * we have gotten rid of all the system vnodes, unless * quotas must not be closed. */ } #endif ASSERT_VOP_LOCKED(ump->um_devvp, "ffs_flushfiles"); if (ump->um_devvp->v_vflag & VV_COPYONWRITE) { if ((error = vflush(mp, 0, SKIPSYSTEM | flags, td)) != 0) return (error); ffs_snapshot_unmount(mp); flags |= FORCECLOSE; /* * Here we fall through to vflush again to ensure * that we have gotten rid of all the system vnodes. */ } /* * Do not close system files if quotas were not closed, to be * able to sync the remaining dquots. The freeblks softupdate * workitems might hold a reference on a dquot, preventing * quotaoff() from completing. Next round of * softdep_flushworklist() iteration should process the * blockers, allowing the next run of quotaoff() to finally * flush held dquots. * * Otherwise, flush all the files. */ if (qerror == 0 && (error = vflush(mp, 0, flags, td)) != 0) return (error); /* * Flush filesystem metadata. */ vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY); error = VOP_FSYNC(ump->um_devvp, MNT_WAIT, td); VOP_UNLOCK(ump->um_devvp, 0); return (error); } /* * Get filesystem statistics. */ static int ffs_statfs(mp, sbp) struct mount *mp; struct statfs *sbp; { struct ufsmount *ump; struct fs *fs; ump = VFSTOUFS(mp); fs = ump->um_fs; if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC) panic("ffs_statfs"); sbp->f_version = STATFS_VERSION; sbp->f_bsize = fs->fs_fsize; sbp->f_iosize = fs->fs_bsize; sbp->f_blocks = fs->fs_dsize; UFS_LOCK(ump); sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag + fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks); sbp->f_bavail = freespace(fs, fs->fs_minfree) + dbtofsb(fs, fs->fs_pendingblocks); sbp->f_files = fs->fs_ncg * fs->fs_ipg - UFS_ROOTINO; sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes; UFS_UNLOCK(ump); sbp->f_namemax = UFS_MAXNAMLEN; return (0); } static bool sync_doupdate(struct inode *ip) { return ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) != 0); } /* * For a lazy sync, we only care about access times, quotas and the * superblock. Other filesystem changes are already converted to * cylinder group blocks or inode blocks updates and are written to * disk by syncer. */ static int ffs_sync_lazy(mp) struct mount *mp; { struct vnode *mvp, *vp; struct inode *ip; struct thread *td; int allerror, error; allerror = 0; td = curthread; if ((mp->mnt_flag & MNT_NOATIME) != 0) goto qupdate; MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) { if (vp->v_type == VNON) { VI_UNLOCK(vp); continue; } ip = VTOI(vp); /* * The IN_ACCESS flag is converted to IN_MODIFIED by * ufs_close() and ufs_getattr() by the calls to * ufs_itimes_locked(), without subsequent UFS_UPDATE(). * Test also all the other timestamp flags too, to pick up * any other cases that could be missed. */ if (!sync_doupdate(ip) && (vp->v_iflag & VI_OWEINACT) == 0) { VI_UNLOCK(vp); continue; } if ((error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK, td)) != 0) continue; if (sync_doupdate(ip)) error = ffs_update(vp, 0); if (error != 0) allerror = error; vput(vp); } qupdate: #ifdef QUOTA qsync(mp); #endif if (VFSTOUFS(mp)->um_fs->fs_fmod != 0 && (error = ffs_sbupdate(VFSTOUFS(mp), MNT_LAZY, 0)) != 0) allerror = error; return (allerror); } /* * Go through the disk queues to initiate sandbagged IO; * go through the inodes to write those that have been modified; * initiate the writing of the super block if it has been modified. * * Note: we are always called with the filesystem marked busy using * vfs_busy(). */ static int ffs_sync(mp, waitfor) struct mount *mp; int waitfor; { struct vnode *mvp, *vp, *devvp; struct thread *td; struct inode *ip; struct ufsmount *ump = VFSTOUFS(mp); struct fs *fs; int error, count, lockreq, allerror = 0; int suspend; int suspended; int secondary_writes; int secondary_accwrites; int softdep_deps; int softdep_accdeps; struct bufobj *bo; suspend = 0; suspended = 0; td = curthread; fs = ump->um_fs; if (fs->fs_fmod != 0 && fs->fs_ronly != 0 && ump->um_fsckpid == 0) panic("%s: ffs_sync: modification on read-only filesystem", fs->fs_fsmnt); if (waitfor == MNT_LAZY) { if (!rebooting) return (ffs_sync_lazy(mp)); waitfor = MNT_NOWAIT; } /* * Write back each (modified) inode. */ lockreq = LK_EXCLUSIVE | LK_NOWAIT; if (waitfor == MNT_SUSPEND) { suspend = 1; waitfor = MNT_WAIT; } if (waitfor == MNT_WAIT) lockreq = LK_EXCLUSIVE; lockreq |= LK_INTERLOCK | LK_SLEEPFAIL; loop: /* Grab snapshot of secondary write counts */ MNT_ILOCK(mp); secondary_writes = mp->mnt_secondary_writes; secondary_accwrites = mp->mnt_secondary_accwrites; MNT_IUNLOCK(mp); /* Grab snapshot of softdep dependency counts */ softdep_get_depcounts(mp, &softdep_deps, &softdep_accdeps); MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { /* * Depend on the vnode interlock to keep things stable enough * for a quick test. Since there might be hundreds of * thousands of vnodes, we cannot afford even a subroutine * call unless there's a good chance that we have work to do. */ if (vp->v_type == VNON) { VI_UNLOCK(vp); continue; } ip = VTOI(vp); if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 && vp->v_bufobj.bo_dirty.bv_cnt == 0) { VI_UNLOCK(vp); continue; } if ((error = vget(vp, lockreq, td)) != 0) { if (error == ENOENT || error == ENOLCK) { MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); goto loop; } continue; } if ((error = ffs_syncvnode(vp, waitfor, 0)) != 0) allerror = error; vput(vp); } /* * Force stale filesystem control information to be flushed. */ if (waitfor == MNT_WAIT || rebooting) { if ((error = softdep_flushworklist(ump->um_mountp, &count, td))) allerror = error; /* Flushed work items may create new vnodes to clean */ if (allerror == 0 && count) goto loop; } #ifdef QUOTA qsync(mp); #endif devvp = ump->um_devvp; bo = &devvp->v_bufobj; BO_LOCK(bo); if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) { BO_UNLOCK(bo); vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); error = VOP_FSYNC(devvp, waitfor, td); VOP_UNLOCK(devvp, 0); if (MOUNTEDSOFTDEP(mp) && (error == 0 || error == EAGAIN)) error = ffs_sbupdate(ump, waitfor, 0); if (error != 0) allerror = error; if (allerror == 0 && waitfor == MNT_WAIT) goto loop; } else if (suspend != 0) { if (softdep_check_suspend(mp, devvp, softdep_deps, softdep_accdeps, secondary_writes, secondary_accwrites) != 0) { MNT_IUNLOCK(mp); goto loop; /* More work needed */ } mtx_assert(MNT_MTX(mp), MA_OWNED); mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED; MNT_IUNLOCK(mp); suspended = 1; } else BO_UNLOCK(bo); /* * Write back modified superblock. */ if (fs->fs_fmod != 0 && (error = ffs_sbupdate(ump, waitfor, suspended)) != 0) allerror = error; return (allerror); } int ffs_vget(mp, ino, flags, vpp) struct mount *mp; ino_t ino; int flags; struct vnode **vpp; { return (ffs_vgetf(mp, ino, flags, vpp, 0)); } int ffs_vgetf(mp, ino, flags, vpp, ffs_flags) struct mount *mp; ino_t ino; int flags; struct vnode **vpp; int ffs_flags; { struct fs *fs; struct inode *ip; struct ufsmount *ump; struct buf *bp; struct vnode *vp; int error; error = vfs_hash_get(mp, ino, flags, curthread, vpp, NULL, NULL); if (error || *vpp != NULL) return (error); /* * We must promote to an exclusive lock for vnode creation. This * can happen if lookup is passed LOCKSHARED. */ if ((flags & LK_TYPE_MASK) == LK_SHARED) { flags &= ~LK_TYPE_MASK; flags |= LK_EXCLUSIVE; } /* * We do not lock vnode creation as it is believed to be too * expensive for such rare case as simultaneous creation of vnode * for same ino by different processes. We just allow them to race * and check later to decide who wins. Let the race begin! */ ump = VFSTOUFS(mp); fs = ump->um_fs; ip = uma_zalloc(uma_inode, M_WAITOK | M_ZERO); /* Allocate a new vnode/inode. */ error = getnewvnode("ufs", mp, fs->fs_magic == FS_UFS1_MAGIC ? &ffs_vnodeops1 : &ffs_vnodeops2, &vp); if (error) { *vpp = NULL; uma_zfree(uma_inode, ip); return (error); } /* * FFS supports recursive locking. */ lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL); VN_LOCK_AREC(vp); vp->v_data = ip; vp->v_bufobj.bo_bsize = fs->fs_bsize; ip->i_vnode = vp; ip->i_ump = ump; ip->i_number = ino; ip->i_ea_refs = 0; ip->i_nextclustercg = -1; ip->i_flag = fs->fs_magic == FS_UFS1_MAGIC ? 0 : IN_UFS2; #ifdef QUOTA { int i; for (i = 0; i < MAXQUOTAS; i++) ip->i_dquot[i] = NODQUOT; } #endif if (ffs_flags & FFSV_FORCEINSMQ) vp->v_vflag |= VV_FORCEINSMQ; error = insmntque(vp, mp); if (error != 0) { uma_zfree(uma_inode, ip); *vpp = NULL; return (error); } vp->v_vflag &= ~VV_FORCEINSMQ; error = vfs_hash_insert(vp, ino, flags, curthread, vpp, NULL, NULL); if (error || *vpp != NULL) return (error); /* Read in the disk contents for the inode, copy into the inode. */ error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)), (int)fs->fs_bsize, NOCRED, &bp); if (error) { /* * The inode does not contain anything useful, so it would * be misleading to leave it on its hash chain. With mode * still zero, it will be unlinked and returned to the free * list by vput(). */ brelse(bp); vput(vp); *vpp = NULL; return (error); } if (I_IS_UFS1(ip)) ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK); else ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK); ffs_load_inode(bp, ip, fs, ino); if (DOINGSOFTDEP(vp)) softdep_load_inodeblock(ip); else ip->i_effnlink = ip->i_nlink; bqrelse(bp); /* * Initialize the vnode from the inode, check for aliases. * Note that the underlying vnode may have changed. */ error = ufs_vinit(mp, I_IS_UFS1(ip) ? &ffs_fifoops1 : &ffs_fifoops2, &vp); if (error) { vput(vp); *vpp = NULL; return (error); } /* * Finish inode initialization. */ if (vp->v_type != VFIFO) { /* FFS supports shared locking for all files except fifos. */ VN_LOCK_ASHARE(vp); } /* * Set up a generation number for this inode if it does not * already have one. This should only happen on old filesystems. */ if (ip->i_gen == 0) { while (ip->i_gen == 0) ip->i_gen = arc4random(); if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { ip->i_flag |= IN_MODIFIED; DIP_SET(ip, i_gen, ip->i_gen); } } #ifdef MAC if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) { /* * If this vnode is already allocated, and we're running * multi-label, attempt to perform a label association * from the extended attributes on the inode. */ error = mac_vnode_associate_extattr(mp, vp); if (error) { /* ufs_inactive will release ip->i_devvp ref. */ vput(vp); *vpp = NULL; return (error); } } #endif *vpp = vp; return (0); } /* * File handle to vnode * * Have to be really careful about stale file handles: * - check that the inode number is valid * - for UFS2 check that the inode number is initialized * - call ffs_vget() to get the locked inode * - check for an unallocated inode (i_mode == 0) * - check that the given client host has export rights and return * those rights via. exflagsp and credanonp */ static int ffs_fhtovp(mp, fhp, flags, vpp) struct mount *mp; struct fid *fhp; int flags; struct vnode **vpp; { struct ufid *ufhp; struct ufsmount *ump; struct fs *fs; struct cg *cgp; struct buf *bp; ino_t ino; u_int cg; int error; ufhp = (struct ufid *)fhp; ino = ufhp->ufid_ino; ump = VFSTOUFS(mp); fs = ump->um_fs; if (ino < UFS_ROOTINO || ino >= fs->fs_ncg * fs->fs_ipg) return (ESTALE); /* * Need to check if inode is initialized because UFS2 does lazy * initialization and nfs_fhtovp can offer arbitrary inode numbers. */ if (fs->fs_magic != FS_UFS2_MAGIC) return (ufs_fhtovp(mp, ufhp, flags, vpp)); cg = ino_to_cg(fs, ino); if ((error = ffs_getcg(fs, ump->um_devvp, cg, &bp, &cgp)) != 0) return (error); if (ino >= cg * fs->fs_ipg + cgp->cg_initediblk) { brelse(bp); return (ESTALE); } brelse(bp); return (ufs_fhtovp(mp, ufhp, flags, vpp)); } /* * Initialize the filesystem. */ static int ffs_init(vfsp) struct vfsconf *vfsp; { ffs_susp_initialize(); softdep_initialize(); return (ufs_init(vfsp)); } /* * Undo the work of ffs_init(). */ static int ffs_uninit(vfsp) struct vfsconf *vfsp; { int ret; ret = ufs_uninit(vfsp); softdep_uninitialize(); ffs_susp_uninitialize(); return (ret); } /* * Structure used to pass information from ffs_sbupdate to its * helper routine ffs_use_bwrite. */ struct devfd { struct ufsmount *ump; struct buf *sbbp; int waitfor; int suspended; int error; }; /* * Write a superblock and associated information back to disk. */ int ffs_sbupdate(ump, waitfor, suspended) struct ufsmount *ump; int waitfor; int suspended; { struct fs *fs; struct buf *sbbp; struct devfd devfd; fs = ump->um_fs; if (fs->fs_ronly == 1 && (ump->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) != (MNT_RDONLY | MNT_UPDATE) && ump->um_fsckpid == 0) panic("ffs_sbupdate: write read-only filesystem"); /* * We use the superblock's buf to serialize calls to ffs_sbupdate(). */ sbbp = getblk(ump->um_devvp, btodb(fs->fs_sblockloc), (int)fs->fs_sbsize, 0, 0, 0); /* * Initialize info needed for write function. */ devfd.ump = ump; devfd.sbbp = sbbp; devfd.waitfor = waitfor; devfd.suspended = suspended; devfd.error = 0; return (ffs_sbput(&devfd, fs, fs->fs_sblockloc, ffs_use_bwrite)); } /* * Write function for use by filesystem-layer routines. */ static int ffs_use_bwrite(void *devfd, off_t loc, void *buf, int size) { struct devfd *devfdp; struct ufsmount *ump; struct buf *bp; struct fs *fs; int error; devfdp = devfd; ump = devfdp->ump; fs = ump->um_fs; /* * Writing the superblock summary information. */ if (loc != fs->fs_sblockloc) { bp = getblk(ump->um_devvp, btodb(loc), size, 0, 0, 0); bcopy(buf, bp->b_data, (u_int)size); if (devfdp->suspended) bp->b_flags |= B_VALIDSUSPWRT; if (devfdp->waitfor != MNT_WAIT) bawrite(bp); else if ((error = bwrite(bp)) != 0) devfdp->error = error; return (0); } /* * Writing the superblock itself. We need to do special checks for it. */ bp = devfdp->sbbp; if (devfdp->error != 0) { brelse(bp); return (devfdp->error); } if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 && (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) { printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n", fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1); fs->fs_sblockloc = SBLOCK_UFS1; } if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 && (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) { printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n", fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS2); fs->fs_sblockloc = SBLOCK_UFS2; } if (MOUNTEDSOFTDEP(ump->um_mountp)) softdep_setup_sbupdate(ump, (struct fs *)bp->b_data, bp); bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize); ffs_oldfscompat_write((struct fs *)bp->b_data, ump); if (devfdp->suspended) bp->b_flags |= B_VALIDSUSPWRT; if (devfdp->waitfor != MNT_WAIT) bawrite(bp); else if ((error = bwrite(bp)) != 0) devfdp->error = error; return (devfdp->error); } static int ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp, int attrnamespace, const char *attrname) { #ifdef UFS_EXTATTR return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace, attrname)); #else return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace, attrname)); #endif } static void ffs_ifree(struct ufsmount *ump, struct inode *ip) { if (ump->um_fstype == UFS1 && ip->i_din1 != NULL) uma_zfree(uma_ufs1, ip->i_din1); else if (ip->i_din2 != NULL) uma_zfree(uma_ufs2, ip->i_din2); uma_zfree(uma_inode, ip); } static int dobkgrdwrite = 1; SYSCTL_INT(_debug, OID_AUTO, dobkgrdwrite, CTLFLAG_RW, &dobkgrdwrite, 0, "Do background writes (honoring the BV_BKGRDWRITE flag)?"); /* * Complete a background write started from bwrite. */ static void ffs_backgroundwritedone(struct buf *bp) { struct bufobj *bufobj; struct buf *origbp; /* * Find the original buffer that we are writing. */ bufobj = bp->b_bufobj; BO_LOCK(bufobj); if ((origbp = gbincore(bp->b_bufobj, bp->b_lblkno)) == NULL) panic("backgroundwritedone: lost buffer"); /* * We should mark the cylinder group buffer origbp as * dirty, to not loose the failed write. */ if ((bp->b_ioflags & BIO_ERROR) != 0) origbp->b_vflags |= BV_BKGRDERR; BO_UNLOCK(bufobj); /* * Process dependencies then return any unfinished ones. */ if (!LIST_EMPTY(&bp->b_dep) && (bp->b_ioflags & BIO_ERROR) == 0) buf_complete(bp); #ifdef SOFTUPDATES if (!LIST_EMPTY(&bp->b_dep)) softdep_move_dependencies(bp, origbp); #endif /* * This buffer is marked B_NOCACHE so when it is released * by biodone it will be tossed. */ bp->b_flags |= B_NOCACHE; bp->b_flags &= ~B_CACHE; pbrelvp(bp); /* * Prevent brelse() from trying to keep and re-dirtying bp on * errors. It causes b_bufobj dereference in * bdirty()/reassignbuf(), and b_bufobj was cleared in * pbrelvp() above. */ if ((bp->b_ioflags & BIO_ERROR) != 0) bp->b_flags |= B_INVAL; bufdone(bp); BO_LOCK(bufobj); /* * Clear the BV_BKGRDINPROG flag in the original buffer * and awaken it if it is waiting for the write to complete. * If BV_BKGRDINPROG is not set in the original buffer it must * have been released and re-instantiated - which is not legal. */ KASSERT((origbp->b_vflags & BV_BKGRDINPROG), ("backgroundwritedone: lost buffer2")); origbp->b_vflags &= ~BV_BKGRDINPROG; if (origbp->b_vflags & BV_BKGRDWAIT) { origbp->b_vflags &= ~BV_BKGRDWAIT; wakeup(&origbp->b_xflags); } BO_UNLOCK(bufobj); } /* * 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. */ static int ffs_bufwrite(struct buf *bp) { struct buf *newbp; + struct cg *cgp; CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); if (bp->b_flags & B_INVAL) { brelse(bp); return (0); } if (!BUF_ISLOCKED(bp)) panic("bufwrite: buffer is not busy???"); /* * If a background write is already in progress, delay * writing this block if it is asynchronous. Otherwise * wait for the background write to complete. */ BO_LOCK(bp->b_bufobj); if (bp->b_vflags & BV_BKGRDINPROG) { if (bp->b_flags & B_ASYNC) { BO_UNLOCK(bp->b_bufobj); bdwrite(bp); return (0); } bp->b_vflags |= BV_BKGRDWAIT; msleep(&bp->b_xflags, BO_LOCKPTR(bp->b_bufobj), PRIBIO, "bwrbg", 0); if (bp->b_vflags & BV_BKGRDINPROG) panic("bufwrite: still writing"); } bp->b_vflags &= ~BV_BKGRDERR; BO_UNLOCK(bp->b_bufobj); /* * If this buffer is marked for background writing and we * do not have to wait for it, make a copy and write the * copy so as to leave this buffer ready for further use. * * This optimization eats a lot of memory. If we have a page * or buffer shortfall we can't do it. */ if (dobkgrdwrite && (bp->b_xflags & BX_BKGRDWRITE) && (bp->b_flags & B_ASYNC) && !vm_page_count_severe() && !buf_dirty_count_severe()) { KASSERT(bp->b_iodone == NULL, ("bufwrite: needs chained iodone (%p)", bp->b_iodone)); /* get a new block */ newbp = geteblk(bp->b_bufsize, GB_NOWAIT_BD); if (newbp == NULL) goto normal_write; KASSERT(buf_mapped(bp), ("Unmapped cg")); memcpy(newbp->b_data, bp->b_data, bp->b_bufsize); BO_LOCK(bp->b_bufobj); bp->b_vflags |= BV_BKGRDINPROG; BO_UNLOCK(bp->b_bufobj); newbp->b_xflags |= (bp->b_xflags & BX_FSPRIV) | BX_BKGRDMARKER; newbp->b_lblkno = bp->b_lblkno; newbp->b_blkno = bp->b_blkno; newbp->b_offset = bp->b_offset; newbp->b_iodone = ffs_backgroundwritedone; newbp->b_flags |= B_ASYNC; newbp->b_flags &= ~B_INVAL; pbgetvp(bp->b_vp, newbp); #ifdef SOFTUPDATES /* * Move over the dependencies. If there are rollbacks, * leave the parent buffer dirtied as it will need to * be written again. */ if (LIST_EMPTY(&bp->b_dep) || softdep_move_dependencies(bp, newbp) == 0) bundirty(bp); #else bundirty(bp); #endif /* * Initiate write on the copy, release the original. The * BKGRDINPROG flag prevents it from going away until - * the background write completes. + * the background write completes. We have to recalculate + * its check hash in case the buffer gets freed and then + * reconstituted from the buffer cache during a later read. */ + if ((bp->b_xflags & BX_CYLGRP) != 0) { + cgp = (struct cg *)bp->b_data; + cgp->cg_ckhash = 0; + cgp->cg_ckhash = + calculate_crc32c(~0L, bp->b_data, bp->b_bcount); + } bqrelse(bp); bp = newbp; } else /* Mark the buffer clean */ bundirty(bp); /* Let the normal bufwrite do the rest for us */ normal_write: + /* + * If we are writing a cylinder group, update its time. + */ + if ((bp->b_xflags & BX_CYLGRP) != 0) { + cgp = (struct cg *)bp->b_data; + cgp->cg_old_time = cgp->cg_time = time_second; + } return (bufwrite(bp)); } static void ffs_geom_strategy(struct bufobj *bo, struct buf *bp) { struct vnode *vp; struct buf *tbp; int error, nocopy; vp = bo2vnode(bo); if (bp->b_iocmd == BIO_WRITE) { if ((bp->b_flags & B_VALIDSUSPWRT) == 0 && bp->b_vp != NULL && bp->b_vp->v_mount != NULL && (bp->b_vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) != 0) panic("ffs_geom_strategy: bad I/O"); nocopy = bp->b_flags & B_NOCOPY; bp->b_flags &= ~(B_VALIDSUSPWRT | B_NOCOPY); if ((vp->v_vflag & VV_COPYONWRITE) && nocopy == 0 && vp->v_rdev->si_snapdata != NULL) { if ((bp->b_flags & B_CLUSTER) != 0) { runningbufwakeup(bp); TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head, b_cluster.cluster_entry) { error = ffs_copyonwrite(vp, tbp); if (error != 0 && error != EOPNOTSUPP) { bp->b_error = error; bp->b_ioflags |= BIO_ERROR; bufdone(bp); return; } } bp->b_runningbufspace = bp->b_bufsize; atomic_add_long(&runningbufspace, bp->b_runningbufspace); } else { error = ffs_copyonwrite(vp, bp); if (error != 0 && error != EOPNOTSUPP) { bp->b_error = error; bp->b_ioflags |= BIO_ERROR; bufdone(bp); return; } } } #ifdef SOFTUPDATES if ((bp->b_flags & B_CLUSTER) != 0) { TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head, b_cluster.cluster_entry) { if (!LIST_EMPTY(&tbp->b_dep)) buf_start(tbp); } } else { if (!LIST_EMPTY(&bp->b_dep)) buf_start(bp); } #endif /* * Check for metadata that needs check-hashes and update them. */ switch (bp->b_xflags & BX_FSPRIV) { case BX_CYLGRP: ((struct cg *)bp->b_data)->cg_ckhash = 0; ((struct cg *)bp->b_data)->cg_ckhash = calculate_crc32c(~0L, bp->b_data, bp->b_bcount); break; case BX_SUPERBLOCK: case BX_INODE: case BX_INDIR: case BX_DIR: printf("Check-hash write is unimplemented!!!\n"); break; case 0: break; default: printf("multiple buffer types 0x%b\n", (u_int)(bp->b_xflags & BX_FSPRIV), PRINT_UFS_BUF_XFLAGS); break; } } g_vfs_strategy(bo, bp); } int ffs_own_mount(const struct mount *mp) { if (mp->mnt_op == &ufs_vfsops) return (1); return (0); } #ifdef DDB #ifdef SOFTUPDATES /* defined in ffs_softdep.c */ extern void db_print_ffs(struct ufsmount *ump); DB_SHOW_COMMAND(ffs, db_show_ffs) { struct mount *mp; struct ufsmount *ump; if (have_addr) { ump = VFSTOUFS((struct mount *)addr); db_print_ffs(ump); return; } TAILQ_FOREACH(mp, &mountlist, mnt_list) { if (!strcmp(mp->mnt_stat.f_fstypename, ufs_vfsconf.vfc_name)) db_print_ffs(VFSTOUFS(mp)); } } #endif /* SOFTUPDATES */ #endif /* DDB */