Index: head/sys/fs/nfs/nfsport.h =================================================================== --- head/sys/fs/nfs/nfsport.h (revision 281959) +++ head/sys/fs/nfs/nfsport.h (revision 281960) @@ -1,969 +1,969 @@ /*- * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Rick Macklem at The University of Guelph. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #ifndef _NFS_NFSPORT_H_ #define _NFS_NFSPORT_H_ /* * In general, I'm not fond of #includes in .h files, but this seems * to be the cleanest way to handle #include files for the ports. */ #ifdef _KERNEL #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 #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * For Darwin, these functions should be "static" when built in a kext. * (This is always defined as nil otherwise.) */ #define APPLESTATIC #include #include #include #include #include #include #include #include #include #include #include "opt_nfs.h" #include "opt_ufs.h" /* * These types must be defined before the nfs includes. */ #define NFSSOCKADDR_T struct sockaddr * #define NFSPROC_T struct thread #define NFSDEV_T dev_t #define NFSSVCARGS nfssvc_args #define NFSACL_T struct acl /* * These should be defined as the types used for the corresponding VOP's * argument type. */ #define NFS_ACCESS_ARGS struct vop_access_args #define NFS_OPEN_ARGS struct vop_open_args #define NFS_GETATTR_ARGS struct vop_getattr_args #define NFS_LOOKUP_ARGS struct vop_lookup_args #define NFS_READDIR_ARGS struct vop_readdir_args /* * Allocate mbufs. Must succeed and never set the mbuf ptr to NULL. */ #define NFSMGET(m) do { \ MGET((m), M_WAITOK, MT_DATA); \ while ((m) == NULL ) { \ (void) nfs_catnap(PZERO, 0, "nfsmget"); \ MGET((m), M_WAITOK, MT_DATA); \ } \ } while (0) #define NFSMGETHDR(m) do { \ MGETHDR((m), M_WAITOK, MT_DATA); \ while ((m) == NULL ) { \ (void) nfs_catnap(PZERO, 0, "nfsmget"); \ MGETHDR((m), M_WAITOK, MT_DATA); \ } \ } while (0) #define NFSMCLGET(m, w) do { \ MGET((m), M_WAITOK, MT_DATA); \ while ((m) == NULL ) { \ (void) nfs_catnap(PZERO, 0, "nfsmget"); \ MGET((m), M_WAITOK, MT_DATA); \ } \ MCLGET((m), (w)); \ } while (0) #define NFSMCLGETHDR(m, w) do { \ MGETHDR((m), M_WAITOK, MT_DATA); \ while ((m) == NULL ) { \ (void) nfs_catnap(PZERO, 0, "nfsmget"); \ MGETHDR((m), M_WAITOK, MT_DATA); \ } \ } while (0) #define NFSMTOD mtod /* * Client side constant for size of a lockowner name. */ #define NFSV4CL_LOCKNAMELEN 12 /* * Type for a mutex lock. */ #define NFSMUTEX_T struct mtx #endif /* _KERNEL */ /* * NFSv4 Operation numbers. */ #define NFSV4OP_ACCESS 3 #define NFSV4OP_CLOSE 4 #define NFSV4OP_COMMIT 5 #define NFSV4OP_CREATE 6 #define NFSV4OP_DELEGPURGE 7 #define NFSV4OP_DELEGRETURN 8 #define NFSV4OP_GETATTR 9 #define NFSV4OP_GETFH 10 #define NFSV4OP_LINK 11 #define NFSV4OP_LOCK 12 #define NFSV4OP_LOCKT 13 #define NFSV4OP_LOCKU 14 #define NFSV4OP_LOOKUP 15 #define NFSV4OP_LOOKUPP 16 #define NFSV4OP_NVERIFY 17 #define NFSV4OP_OPEN 18 #define NFSV4OP_OPENATTR 19 #define NFSV4OP_OPENCONFIRM 20 #define NFSV4OP_OPENDOWNGRADE 21 #define NFSV4OP_PUTFH 22 #define NFSV4OP_PUTPUBFH 23 #define NFSV4OP_PUTROOTFH 24 #define NFSV4OP_READ 25 #define NFSV4OP_READDIR 26 #define NFSV4OP_READLINK 27 #define NFSV4OP_REMOVE 28 #define NFSV4OP_RENAME 29 #define NFSV4OP_RENEW 30 #define NFSV4OP_RESTOREFH 31 #define NFSV4OP_SAVEFH 32 #define NFSV4OP_SECINFO 33 #define NFSV4OP_SETATTR 34 #define NFSV4OP_SETCLIENTID 35 #define NFSV4OP_SETCLIENTIDCFRM 36 #define NFSV4OP_VERIFY 37 #define NFSV4OP_WRITE 38 #define NFSV4OP_RELEASELCKOWN 39 /* * Must be one greater than the last Operation#. */ #define NFSV4OP_NOPS 40 /* * Additional Ops for NFSv4.1. */ #define NFSV4OP_BACKCHANNELCTL 40 #define NFSV4OP_BINDCONNTOSESS 41 #define NFSV4OP_EXCHANGEID 42 #define NFSV4OP_CREATESESSION 43 #define NFSV4OP_DESTROYSESSION 44 #define NFSV4OP_FREESTATEID 45 #define NFSV4OP_GETDIRDELEG 46 #define NFSV4OP_GETDEVINFO 47 #define NFSV4OP_GETDEVLIST 48 #define NFSV4OP_LAYOUTCOMMIT 49 #define NFSV4OP_LAYOUTGET 50 #define NFSV4OP_LAYOUTRETURN 51 #define NFSV4OP_SECINFONONAME 52 #define NFSV4OP_SEQUENCE 53 #define NFSV4OP_SETSSV 54 #define NFSV4OP_TESTSTATEID 55 #define NFSV4OP_WANTDELEG 56 #define NFSV4OP_DESTROYCLIENTID 57 #define NFSV4OP_RECLAIMCOMPL 58 /* * Must be one more than last op#. */ #define NFSV41_NOPS 59 /* Quirky case if the illegal op code */ #define NFSV4OP_OPILLEGAL 10044 /* * Fake NFSV4OP_xxx used for nfsstat. Start at NFSV4OP_NOPS. */ #define NFSV4OP_SYMLINK (NFSV4OP_NOPS) #define NFSV4OP_MKDIR (NFSV4OP_NOPS + 1) #define NFSV4OP_RMDIR (NFSV4OP_NOPS + 2) #define NFSV4OP_READDIRPLUS (NFSV4OP_NOPS + 3) #define NFSV4OP_MKNOD (NFSV4OP_NOPS + 4) #define NFSV4OP_FSSTAT (NFSV4OP_NOPS + 5) #define NFSV4OP_FSINFO (NFSV4OP_NOPS + 6) #define NFSV4OP_PATHCONF (NFSV4OP_NOPS + 7) #define NFSV4OP_V3CREATE (NFSV4OP_NOPS + 8) /* * This is the count of the fake operations listed above. */ #define NFSV4OP_FAKENOPS 9 /* * and the Callback OPs */ #define NFSV4OP_CBGETATTR 3 #define NFSV4OP_CBRECALL 4 /* * Must be one greater than the last Callback Operation#. */ #define NFSV4OP_CBNOPS 5 /* * Additional Callback Ops for NFSv4.1 only. Not yet in nfsstats. */ #define NFSV4OP_CBLAYOUTRECALL 5 #define NFSV4OP_CBNOTIFY 6 #define NFSV4OP_CBPUSHDELEG 7 #define NFSV4OP_CBRECALLANY 8 #define NFSV4OP_CBRECALLOBJAVAIL 9 #define NFSV4OP_CBRECALLSLOT 10 #define NFSV4OP_CBSEQUENCE 11 #define NFSV4OP_CBWANTCANCELLED 12 #define NFSV4OP_CBNOTIFYLOCK 13 #define NFSV4OP_CBNOTIFYDEVID 14 /* * The lower numbers -> 21 are used by NFSv2 and v3. These define higher * numbers used by NFSv4. * NFS_V3NPROCS is one greater than the last V3 op and NFS_NPROCS is * one greater than the last number. */ #ifndef NFS_V3NPROCS #define NFS_V3NPROCS 22 #define NFSPROC_LOOKUPP 22 #define NFSPROC_SETCLIENTID 23 #define NFSPROC_SETCLIENTIDCFRM 24 #define NFSPROC_LOCK 25 #define NFSPROC_LOCKU 26 #define NFSPROC_OPEN 27 #define NFSPROC_CLOSE 28 #define NFSPROC_OPENCONFIRM 29 #define NFSPROC_LOCKT 30 #define NFSPROC_OPENDOWNGRADE 31 #define NFSPROC_RENEW 32 #define NFSPROC_PUTROOTFH 33 #define NFSPROC_RELEASELCKOWN 34 #define NFSPROC_DELEGRETURN 35 #define NFSPROC_RETDELEGREMOVE 36 #define NFSPROC_RETDELEGRENAME1 37 #define NFSPROC_RETDELEGRENAME2 38 #define NFSPROC_GETACL 39 #define NFSPROC_SETACL 40 /* * Must be defined as one higher than the last Proc# above. */ #define NFSV4_NPROCS 41 /* Additional procedures for NFSv4.1. */ #define NFSPROC_EXCHANGEID 41 #define NFSPROC_CREATESESSION 42 #define NFSPROC_DESTROYSESSION 43 #define NFSPROC_DESTROYCLIENT 44 #define NFSPROC_FREESTATEID 45 #define NFSPROC_LAYOUTGET 46 #define NFSPROC_GETDEVICEINFO 47 #define NFSPROC_LAYOUTCOMMIT 48 #define NFSPROC_LAYOUTRETURN 49 #define NFSPROC_RECLAIMCOMPL 50 #define NFSPROC_WRITEDS 51 #define NFSPROC_READDS 52 #define NFSPROC_COMMITDS 53 /* * Must be defined as one higher than the last NFSv4.1 Proc# above. */ #define NFSV41_NPROCS 54 #endif /* NFS_V3NPROCS */ /* * Stats structure */ struct ext_nfsstats { int attrcache_hits; int attrcache_misses; int lookupcache_hits; int lookupcache_misses; int direofcache_hits; int direofcache_misses; int accesscache_hits; int accesscache_misses; int biocache_reads; int read_bios; int read_physios; int biocache_writes; int write_bios; int write_physios; int biocache_readlinks; int readlink_bios; int biocache_readdirs; int readdir_bios; int rpccnt[NFSV4_NPROCS]; int rpcretries; int srvrpccnt[NFSV4OP_NOPS + NFSV4OP_FAKENOPS]; int srvrpc_errs; int srv_errs; int rpcrequests; int rpctimeouts; int rpcunexpected; int rpcinvalid; int srvcache_inproghits; int srvcache_idemdonehits; int srvcache_nonidemdonehits; int srvcache_misses; int srvcache_tcppeak; int srvcache_size; int srvclients; int srvopenowners; int srvopens; int srvlockowners; int srvlocks; int srvdelegates; int cbrpccnt[NFSV4OP_CBNOPS]; int clopenowners; int clopens; int cllockowners; int cllocks; int cldelegates; int cllocalopenowners; int cllocalopens; int cllocallockowners; int cllocallocks; }; #ifdef _KERNEL /* * Define the ext_nfsstats as nfsstats for the kernel code. */ #define nfsstats ext_nfsstats /* * Define NFS_NPROCS as NFSV4_NPROCS for the experimental kernel code. */ #ifndef NFS_NPROCS #define NFS_NPROCS NFSV4_NPROCS #endif #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * Just to keep nfs_var.h happy. */ struct nfs_vattr { int junk; }; struct nfsvattr { struct vattr na_vattr; nfsattrbit_t na_suppattr; u_int32_t na_mntonfileno; u_int64_t na_filesid[2]; }; #define na_type na_vattr.va_type #define na_mode na_vattr.va_mode #define na_nlink na_vattr.va_nlink #define na_uid na_vattr.va_uid #define na_gid na_vattr.va_gid #define na_fsid na_vattr.va_fsid #define na_fileid na_vattr.va_fileid #define na_size na_vattr.va_size #define na_blocksize na_vattr.va_blocksize #define na_atime na_vattr.va_atime #define na_mtime na_vattr.va_mtime #define na_ctime na_vattr.va_ctime #define na_gen na_vattr.va_gen #define na_flags na_vattr.va_flags #define na_rdev na_vattr.va_rdev #define na_bytes na_vattr.va_bytes #define na_filerev na_vattr.va_filerev #define na_vaflags na_vattr.va_vaflags #include /* * This is the header structure used for the lists, etc. (It has the * above record in it. */ struct nfsrv_stablefirst { LIST_HEAD(, nfsrv_stable) nsf_head; /* Head of nfsrv_stable list */ time_t nsf_eograce; /* Time grace period ends */ time_t *nsf_bootvals; /* Previous boottime values */ struct file *nsf_fp; /* File table pointer */ u_char nsf_flags; /* NFSNSF_ flags */ struct nfsf_rec nsf_rec; /* and above first record */ }; #define nsf_lease nsf_rec.lease #define nsf_numboots nsf_rec.numboots /* NFSNSF_xxx flags */ #define NFSNSF_UPDATEDONE 0x01 #define NFSNSF_GRACEOVER 0x02 #define NFSNSF_NEEDLOCK 0x04 #define NFSNSF_EXPIREDCLIENT 0x08 #define NFSNSF_NOOPENS 0x10 #define NFSNSF_OK 0x20 /* * Maximum number of boot times allowed in record. Although there is * really no need for a fixed upper bound, this serves as a sanity check * for a corrupted file. */ #define NFSNSF_MAXNUMBOOTS 10000 /* * This structure defines the other records in the file. The * nst_client array is actually the size of the client string name. */ struct nfst_rec { u_int16_t len; u_char flag; u_char client[1]; }; /* and the values for flag */ #define NFSNST_NEWSTATE 0x1 #define NFSNST_REVOKE 0x2 #define NFSNST_GOTSTATE 0x4 /* * This structure is linked onto nfsrv_stablefirst for the duration of * reclaim. */ struct nfsrv_stable { LIST_ENTRY(nfsrv_stable) nst_list; struct nfsclient *nst_clp; struct nfst_rec nst_rec; }; #define nst_timestamp nst_rec.timestamp #define nst_len nst_rec.len #define nst_flag nst_rec.flag #define nst_client nst_rec.client /* * At some point the server will run out of kernel storage for * state structures. For FreeBSD5.2, this results in a panic * kmem_map is full. It happens at well over 1000000 opens plus * locks on a PIII-800 with 256Mbytes, so that is where I've set * the limit. If your server panics due to too many opens/locks, * decrease the size of NFSRV_V4STATELIMIT. If you find the server * returning NFS4ERR_RESOURCE a lot and have lots of memory, try * increasing it. */ #define NFSRV_V4STATELIMIT 500000 /* Max # of Opens + Locks */ /* * The type required differs with BSDen (just the second arg). */ void nfsrvd_rcv(struct socket *, void *, int); /* * Macros for handling socket addresses. (Hopefully this makes the code * more portable, since I've noticed some 'BSD don't have sockaddrs in * mbufs any more.) */ #define NFSSOCKADDR(a, t) ((t)(a)) #define NFSSOCKADDRALLOC(a) \ do { \ MALLOC((a), struct sockaddr *, sizeof (struct sockaddr), \ M_SONAME, M_WAITOK); \ NFSBZERO((a), sizeof (struct sockaddr)); \ } while (0) #define NFSSOCKADDRSIZE(a, s) ((a)->sa_len = (s)) #define NFSSOCKADDRFREE(a) \ do { \ if (a) \ FREE((caddr_t)(a), M_SONAME); \ } while (0) /* * These should be defined as a process or thread structure, as required * for signal handling, etc. */ #define NFSNEWCRED(c) (crdup(c)) #define NFSPROCCRED(p) ((p)->td_ucred) #define NFSFREECRED(c) (crfree(c)) #define NFSUIOPROC(u, p) ((u)->uio_td = NULL) #define NFSPROCP(p) ((p)->td_proc) /* * Define these so that cn_hash and its length is ignored. */ #define NFSCNHASHZERO(c) #define NFSCNHASH(c, v) #define NCHNAMLEN 9999999 /* * These macros are defined to initialize and set the timer routine. */ #define NFS_TIMERINIT \ newnfs_timer(NULL) /* * Handle SMP stuff: */ #define NFSSTATESPINLOCK extern struct mtx nfs_state_mutex #define NFSLOCKSTATE() mtx_lock(&nfs_state_mutex) #define NFSUNLOCKSTATE() mtx_unlock(&nfs_state_mutex) #define NFSSTATEMUTEXPTR (&nfs_state_mutex) #define NFSREQSPINLOCK extern struct mtx nfs_req_mutex #define NFSLOCKREQ() mtx_lock(&nfs_req_mutex) #define NFSUNLOCKREQ() mtx_unlock(&nfs_req_mutex) #define NFSSOCKMUTEX extern struct mtx nfs_slock_mutex #define NFSSOCKMUTEXPTR (&nfs_slock_mutex) #define NFSLOCKSOCK() mtx_lock(&nfs_slock_mutex) #define NFSUNLOCKSOCK() mtx_unlock(&nfs_slock_mutex) #define NFSNAMEIDMUTEX extern struct mtx nfs_nameid_mutex #define NFSLOCKNAMEID() mtx_lock(&nfs_nameid_mutex) #define NFSUNLOCKNAMEID() mtx_unlock(&nfs_nameid_mutex) #define NFSNAMEIDREQUIRED() mtx_assert(&nfs_nameid_mutex, MA_OWNED) #define NFSCLSTATEMUTEX extern struct mtx nfs_clstate_mutex #define NFSCLSTATEMUTEXPTR (&nfs_clstate_mutex) #define NFSLOCKCLSTATE() mtx_lock(&nfs_clstate_mutex) #define NFSUNLOCKCLSTATE() mtx_unlock(&nfs_clstate_mutex) #define NFSDLOCKMUTEX extern struct mtx newnfsd_mtx #define NFSDLOCKMUTEXPTR (&newnfsd_mtx) #define NFSD_LOCK() mtx_lock(&newnfsd_mtx) #define NFSD_UNLOCK() mtx_unlock(&newnfsd_mtx) #define NFSD_LOCK_ASSERT() mtx_assert(&newnfsd_mtx, MA_OWNED) #define NFSD_UNLOCK_ASSERT() mtx_assert(&newnfsd_mtx, MA_NOTOWNED) #define NFSV4ROOTLOCKMUTEX extern struct mtx nfs_v4root_mutex #define NFSV4ROOTLOCKMUTEXPTR (&nfs_v4root_mutex) #define NFSLOCKV4ROOTMUTEX() mtx_lock(&nfs_v4root_mutex) #define NFSUNLOCKV4ROOTMUTEX() mtx_unlock(&nfs_v4root_mutex) #define NFSLOCKNODE(n) mtx_lock(&((n)->n_mtx)) #define NFSUNLOCKNODE(n) mtx_unlock(&((n)->n_mtx)) #define NFSLOCKMNT(m) mtx_lock(&((m)->nm_mtx)) #define NFSUNLOCKMNT(m) mtx_unlock(&((m)->nm_mtx)) #define NFSLOCKREQUEST(r) mtx_lock(&((r)->r_mtx)) #define NFSUNLOCKREQUEST(r) mtx_unlock(&((r)->r_mtx)) #define NFSPROCLISTLOCK() sx_slock(&allproc_lock) #define NFSPROCLISTUNLOCK() sx_sunlock(&allproc_lock) #define NFSLOCKSOCKREQ(r) mtx_lock(&((r)->nr_mtx)) #define NFSUNLOCKSOCKREQ(r) mtx_unlock(&((r)->nr_mtx)) #define NFSLOCKDS(d) mtx_lock(&((d)->nfsclds_mtx)) #define NFSUNLOCKDS(d) mtx_unlock(&((d)->nfsclds_mtx)) #define NFSSESSIONMUTEXPTR(s) (&((s)->mtx)) #define NFSLOCKSESSION(s) mtx_lock(&((s)->mtx)) #define NFSUNLOCKSESSION(s) mtx_unlock(&((s)->mtx)) /* * Use these macros to initialize/free a mutex. */ #define NFSINITSOCKMUTEX(m) mtx_init((m), "nfssock", NULL, MTX_DEF) #define NFSFREEMUTEX(m) mtx_destroy((m)) int nfsmsleep(void *, void *, int, const char *, struct timespec *); /* * And weird vm stuff in the nfs server. */ #define PDIRUNLOCK 0x0 #define MAX_COMMIT_COUNT (1024 * 1024) /* * Define these to handle the type of va_rdev. */ #define NFSMAKEDEV(m, n) makedev((m), (n)) #define NFSMAJOR(d) major(d) #define NFSMINOR(d) minor(d) /* * Define this to be the macro that returns the minimum size required * for a directory entry. */ #define DIRENT_SIZE(dp) GENERIC_DIRSIZ(dp) /* * The vnode tag for nfsv4root. */ #define VT_NFSV4ROOT "nfsv4root" /* * Define whatever it takes to do a vn_rdwr(). */ #define NFSD_RDWR(r, v, b, l, o, s, i, c, a, p) \ vn_rdwr((r), (v), (b), (l), (o), (s), (i), (c), NULL, (a), (p)) /* * Macros for handling memory for different BSDen. * NFSBCOPY(src, dst, len) - copies len bytes, non-overlapping * NFSOVBCOPY(src, dst, len) - ditto, but data areas might overlap * NFSBCMP(cp1, cp2, len) - compare len bytes, return 0 if same * NFSBZERO(cp, len) - set len bytes to 0x0 */ #define NFSBCOPY(s, d, l) bcopy((s), (d), (l)) #define NFSOVBCOPY(s, d, l) ovbcopy((s), (d), (l)) #define NFSBCMP(s, d, l) bcmp((s), (d), (l)) #define NFSBZERO(s, l) bzero((s), (l)) /* * Some queue.h files don't have these dfined in them. */ #define LIST_END(head) NULL #define SLIST_END(head) NULL #define TAILQ_END(head) NULL /* * This must be defined to be a global variable that increments once * per second, but never stops or goes backwards, even when a "date" * command changes the TOD clock. It is used for delta times for * leases, etc. */ #define NFSD_MONOSEC time_uptime /* * Declare the malloc types. */ MALLOC_DECLARE(M_NEWNFSRVCACHE); MALLOC_DECLARE(M_NEWNFSDCLIENT); MALLOC_DECLARE(M_NEWNFSDSTATE); MALLOC_DECLARE(M_NEWNFSDLOCK); MALLOC_DECLARE(M_NEWNFSDLOCKFILE); MALLOC_DECLARE(M_NEWNFSSTRING); MALLOC_DECLARE(M_NEWNFSUSERGROUP); MALLOC_DECLARE(M_NEWNFSDREQ); MALLOC_DECLARE(M_NEWNFSFH); MALLOC_DECLARE(M_NEWNFSCLOWNER); MALLOC_DECLARE(M_NEWNFSCLOPEN); MALLOC_DECLARE(M_NEWNFSCLDELEG); MALLOC_DECLARE(M_NEWNFSCLCLIENT); MALLOC_DECLARE(M_NEWNFSCLLOCKOWNER); MALLOC_DECLARE(M_NEWNFSCLLOCK); MALLOC_DECLARE(M_NEWNFSDIROFF); MALLOC_DECLARE(M_NEWNFSV4NODE); MALLOC_DECLARE(M_NEWNFSDIRECTIO); MALLOC_DECLARE(M_NEWNFSMNT); MALLOC_DECLARE(M_NEWNFSDROLLBACK); MALLOC_DECLARE(M_NEWNFSLAYOUT); MALLOC_DECLARE(M_NEWNFSFLAYOUT); MALLOC_DECLARE(M_NEWNFSDEVINFO); MALLOC_DECLARE(M_NEWNFSSOCKREQ); MALLOC_DECLARE(M_NEWNFSCLDS); MALLOC_DECLARE(M_NEWNFSLAYRECALL); MALLOC_DECLARE(M_NEWNFSDSESSION); #define M_NFSRVCACHE M_NEWNFSRVCACHE #define M_NFSDCLIENT M_NEWNFSDCLIENT #define M_NFSDSTATE M_NEWNFSDSTATE #define M_NFSDLOCK M_NEWNFSDLOCK #define M_NFSDLOCKFILE M_NEWNFSDLOCKFILE #define M_NFSSTRING M_NEWNFSSTRING #define M_NFSUSERGROUP M_NEWNFSUSERGROUP #define M_NFSDREQ M_NEWNFSDREQ #define M_NFSFH M_NEWNFSFH #define M_NFSCLOWNER M_NEWNFSCLOWNER #define M_NFSCLOPEN M_NEWNFSCLOPEN #define M_NFSCLDELEG M_NEWNFSCLDELEG #define M_NFSCLCLIENT M_NEWNFSCLCLIENT #define M_NFSCLLOCKOWNER M_NEWNFSCLLOCKOWNER #define M_NFSCLLOCK M_NEWNFSCLLOCK #define M_NFSDIROFF M_NEWNFSDIROFF #define M_NFSV4NODE M_NEWNFSV4NODE #define M_NFSDIRECTIO M_NEWNFSDIRECTIO #define M_NFSDROLLBACK M_NEWNFSDROLLBACK #define M_NFSLAYOUT M_NEWNFSLAYOUT #define M_NFSFLAYOUT M_NEWNFSFLAYOUT #define M_NFSDEVINFO M_NEWNFSDEVINFO #define M_NFSSOCKREQ M_NEWNFSSOCKREQ #define M_NFSCLDS M_NEWNFSCLDS #define M_NFSLAYRECALL M_NEWNFSLAYRECALL #define M_NFSDSESSION M_NEWNFSDSESSION #define NFSINT_SIGMASK(set) \ (SIGISMEMBER(set, SIGINT) || SIGISMEMBER(set, SIGTERM) || \ SIGISMEMBER(set, SIGHUP) || SIGISMEMBER(set, SIGKILL) || \ SIGISMEMBER(set, SIGQUIT)) /* * Convert a quota block count to byte count. */ #define NFSQUOTABLKTOBYTE(q, b) (q) *= (b) /* * Define this as the largest file size supported. (It should probably * be available via a VFS_xxx Op, but it isn't. */ #define NFSRV_MAXFILESIZE ((u_int64_t)0x800000000000) /* * Set this macro to index() or strchr(), whichever is supported. */ #define STRCHR(s, c) strchr((s), (c)) /* * Set the n_time in the client write rpc, as required. */ #define NFSWRITERPC_SETTIME(w, n, v4) \ do { \ if (w) { \ (n)->n_mtime = (n)->n_vattr.na_vattr.va_mtime; \ if (v4) \ (n)->n_change = (n)->n_vattr.na_vattr.va_filerev; \ } \ } while (0) /* * Fake value, just to make the client work. */ #define NFS_LATTR_NOSHRINK 1 /* * Prototypes for functions where the arguments vary for different ports. */ int nfscl_loadattrcache(struct vnode **, struct nfsvattr *, void *, void *, int, int); int newnfs_realign(struct mbuf **, int); /* * If the port runs on an SMP box that can enforce Atomic ops with low * overheads, define these as atomic increments/decrements. If not, * don't worry about it, since these are used for stats that can be * "out by one" without disastrous consequences. */ #define NFSINCRGLOBAL(a) ((a)++) /* * Assorted funky stuff to make things work under Darwin8. */ /* * These macros checks for a field in vattr being set. */ #define NFSATTRISSET(t, v, a) ((v)->a != (t)VNOVAL) #define NFSATTRISSETTIME(v, a) ((v)->a.tv_sec != VNOVAL) /* * Manipulate mount flags. */ #define NFSSTA_HASWRITEVERF 0x00040000 /* Has write verifier */ #define NFSSTA_GOTFSINFO 0x00100000 /* Got the fsinfo */ #define NFSSTA_NOLAYOUTCOMMIT 0x04000000 /* Don't do LayoutCommit */ #define NFSSTA_SESSPERSIST 0x08000000 /* Has a persistent session */ #define NFSSTA_TIMEO 0x10000000 /* Experiencing a timeout */ #define NFSSTA_LOCKTIMEO 0x20000000 /* Experiencing a lockd timeout */ #define NFSSTA_HASSETFSID 0x40000000 /* Has set the fsid */ #define NFSSTA_PNFS 0x80000000 /* pNFS is enabled */ #define NFSHASNFSV3(n) ((n)->nm_flag & NFSMNT_NFSV3) #define NFSHASNFSV4(n) ((n)->nm_flag & NFSMNT_NFSV4) #define NFSHASNFSV4N(n) ((n)->nm_minorvers > 0) #define NFSHASNFSV3OR4(n) ((n)->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) #define NFSHASGOTFSINFO(n) ((n)->nm_state & NFSSTA_GOTFSINFO) #define NFSHASHASSETFSID(n) ((n)->nm_state & NFSSTA_HASSETFSID) #define NFSHASSTRICT3530(n) ((n)->nm_flag & NFSMNT_STRICT3530) #define NFSHASWRITEVERF(n) ((n)->nm_state & NFSSTA_HASWRITEVERF) #define NFSHASINT(n) ((n)->nm_flag & NFSMNT_INT) #define NFSHASSOFT(n) ((n)->nm_flag & NFSMNT_SOFT) #define NFSHASINTORSOFT(n) ((n)->nm_flag & (NFSMNT_INT | NFSMNT_SOFT)) #define NFSHASDUMBTIMR(n) ((n)->nm_flag & NFSMNT_DUMBTIMR) #define NFSHASNOCONN(n) ((n)->nm_flag & NFSMNT_MNTD) #define NFSHASKERB(n) ((n)->nm_flag & NFSMNT_KERB) #define NFSHASALLGSSNAME(n) ((n)->nm_flag & NFSMNT_ALLGSSNAME) #define NFSHASINTEGRITY(n) ((n)->nm_flag & NFSMNT_INTEGRITY) #define NFSHASPRIVACY(n) ((n)->nm_flag & NFSMNT_PRIVACY) #define NFSSETWRITEVERF(n) ((n)->nm_state |= NFSSTA_HASWRITEVERF) #define NFSSETHASSETFSID(n) ((n)->nm_state |= NFSSTA_HASSETFSID) #define NFSHASPNFSOPT(n) ((n)->nm_flag & NFSMNT_PNFS) #define NFSHASNOLAYOUTCOMMIT(n) ((n)->nm_state & NFSSTA_NOLAYOUTCOMMIT) #define NFSHASSESSPERSIST(n) ((n)->nm_state & NFSSTA_SESSPERSIST) #define NFSHASPNFS(n) ((n)->nm_state & NFSSTA_PNFS) /* * Gets the stats field out of the mount structure. */ #define vfs_statfs(m) (&((m)->mnt_stat)) /* * Set boottime. */ #define NFSSETBOOTTIME(b) ((b) = boottime) /* * The size of directory blocks in the buffer cache. * MUST BE in the range of PAGE_SIZE <= NFS_DIRBLKSIZ <= MAXBSIZE!! */ #define NFS_DIRBLKSIZ (16 * DIRBLKSIZ) /* Must be a multiple of DIRBLKSIZ */ /* * Define these macros to access mnt_flag fields. */ #define NFSMNT_RDONLY(m) ((m)->mnt_flag & MNT_RDONLY) #endif /* _KERNEL */ /* * Define a structure similar to ufs_args for use in exporting the V4 root. */ struct nfsex_args { char *fspec; struct export_args export; }; /* * These export flags should be defined, but there are no bits left. * Maybe a separate mnt_exflag field could be added or the mnt_flag * field increased to 64 bits? */ #ifndef MNT_EXSTRICTACCESS #define MNT_EXSTRICTACCESS 0x0 #endif #ifndef MNT_EXV4ONLY #define MNT_EXV4ONLY 0x0 #endif #ifdef _KERNEL /* * Define this to invalidate the attribute cache for the nfs node. */ #define NFSINVALATTRCACHE(n) ((n)->n_attrstamp = 0) /* Used for FreeBSD only */ void nfsd_mntinit(void); /* * Define these for vnode lock/unlock ops. * * These are good abstractions to macro out, so that they can be added to * later, for debugging or stats, etc. */ #define NFSVOPLOCK(v, f) vn_lock((v), (f)) #define NFSVOPUNLOCK(v, f) VOP_UNLOCK((v), (f)) #define NFSVOPISLOCKED(v) VOP_ISLOCKED((v)) /* * Define ncl_hash(). */ #define ncl_hash(f, l) (fnv_32_buf((f), (l), FNV1_32_INIT)) int newnfs_iosize(struct nfsmount *); int newnfs_vncmpf(struct vnode *, void *); #ifndef NFS_MINDIRATTRTIMO #define NFS_MINDIRATTRTIMO 3 /* VDIR attrib cache timeout in sec */ #endif #ifndef NFS_MAXDIRATTRTIMO #define NFS_MAXDIRATTRTIMO 60 #endif /* * Nfs outstanding request list element */ struct nfsreq { TAILQ_ENTRY(nfsreq) r_chain; u_int32_t r_flags; /* flags on request, see below */ struct nfsmount *r_nmp; /* Client mnt ptr */ struct mtx r_mtx; /* Mutex lock for this structure */ }; #ifndef NFS_MAXBSIZE -#define NFS_MAXBSIZE MAXBSIZE +#define NFS_MAXBSIZE MAXBCACHEBUF #endif /* * This macro checks to see if issuing of delegations is allowed for this * vnode. */ #ifdef VV_DISABLEDELEG #define NFSVNO_DELEGOK(v) \ ((v) == NULL || ((v)->v_vflag & VV_DISABLEDELEG) == 0) #else #define NFSVNO_DELEGOK(v) (1) #endif #endif /* _KERNEL */ #endif /* _NFS_NFSPORT_H */ Index: head/sys/fs/nfsclient/nfs_clvfsops.c =================================================================== --- head/sys/fs/nfsclient/nfs_clvfsops.c (revision 281959) +++ head/sys/fs/nfsclient/nfs_clvfsops.c (revision 281960) @@ -1,1880 +1,1880 @@ /*- * Copyright (c) 1989, 1993, 1995 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Rick Macklem at The University of Guelph. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * from nfs_vfsops.c 8.12 (Berkeley) 5/20/95 */ #include __FBSDID("$FreeBSD$"); #include "opt_bootp.h" #include "opt_nfsroot.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 FEATURE(nfscl, "NFSv4 client"); extern int nfscl_ticks; extern struct timeval nfsboottime; extern struct nfsstats newnfsstats; extern int nfsrv_useacl; extern int nfscl_debuglevel; extern enum nfsiod_state ncl_iodwant[NFS_MAXASYNCDAEMON]; extern struct nfsmount *ncl_iodmount[NFS_MAXASYNCDAEMON]; extern struct mtx ncl_iod_mutex; NFSCLSTATEMUTEX; MALLOC_DEFINE(M_NEWNFSREQ, "newnfsclient_req", "NFS request header"); MALLOC_DEFINE(M_NEWNFSMNT, "newnfsmnt", "NFS mount struct"); SYSCTL_DECL(_vfs_nfs); static int nfs_ip_paranoia = 1; SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_ip_paranoia, CTLFLAG_RW, &nfs_ip_paranoia, 0, ""); static int nfs_tprintf_initial_delay = NFS_TPRINTF_INITIAL_DELAY; SYSCTL_INT(_vfs_nfs, NFS_TPRINTF_INITIAL_DELAY, downdelayinitial, CTLFLAG_RW, &nfs_tprintf_initial_delay, 0, ""); /* how long between console messages "nfs server foo not responding" */ static int nfs_tprintf_delay = NFS_TPRINTF_DELAY; SYSCTL_INT(_vfs_nfs, NFS_TPRINTF_DELAY, downdelayinterval, CTLFLAG_RW, &nfs_tprintf_delay, 0, ""); #ifdef NFS_DEBUG int nfs_debug; SYSCTL_INT(_vfs_nfs, OID_AUTO, debug, CTLFLAG_RW, &nfs_debug, 0, "Toggle debug flag"); #endif static int nfs_mountroot(struct mount *); static void nfs_sec_name(char *, int *); static void nfs_decode_args(struct mount *mp, struct nfsmount *nmp, struct nfs_args *argp, const char *, struct ucred *, struct thread *); static int mountnfs(struct nfs_args *, struct mount *, struct sockaddr *, char *, u_char *, int, u_char *, int, u_char *, int, struct vnode **, struct ucred *, struct thread *, int, int, int); static void nfs_getnlminfo(struct vnode *, uint8_t *, size_t *, struct sockaddr_storage *, int *, off_t *, struct timeval *); static vfs_mount_t nfs_mount; static vfs_cmount_t nfs_cmount; static vfs_unmount_t nfs_unmount; static vfs_root_t nfs_root; static vfs_statfs_t nfs_statfs; static vfs_sync_t nfs_sync; static vfs_sysctl_t nfs_sysctl; static vfs_purge_t nfs_purge; /* * nfs vfs operations. */ static struct vfsops nfs_vfsops = { .vfs_init = ncl_init, .vfs_mount = nfs_mount, .vfs_cmount = nfs_cmount, .vfs_root = nfs_root, .vfs_statfs = nfs_statfs, .vfs_sync = nfs_sync, .vfs_uninit = ncl_uninit, .vfs_unmount = nfs_unmount, .vfs_sysctl = nfs_sysctl, .vfs_purge = nfs_purge, }; VFS_SET(nfs_vfsops, nfs, VFCF_NETWORK | VFCF_SBDRY); /* So that loader and kldload(2) can find us, wherever we are.. */ MODULE_VERSION(nfs, 1); MODULE_DEPEND(nfs, nfscommon, 1, 1, 1); MODULE_DEPEND(nfs, krpc, 1, 1, 1); MODULE_DEPEND(nfs, nfssvc, 1, 1, 1); MODULE_DEPEND(nfs, nfslock, 1, 1, 1); /* * This structure is now defined in sys/nfs/nfs_diskless.c so that it * can be shared by both NFS clients. It is declared here so that it * will be defined for kernels built without NFS_ROOT, although it * isn't used in that case. */ #if !defined(NFS_ROOT) struct nfs_diskless nfs_diskless = { { { 0 } } }; struct nfsv3_diskless nfsv3_diskless = { { { 0 } } }; int nfs_diskless_valid = 0; #endif SYSCTL_INT(_vfs_nfs, OID_AUTO, diskless_valid, CTLFLAG_RD, &nfs_diskless_valid, 0, "Has the diskless struct been filled correctly"); SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_rootpath, CTLFLAG_RD, nfsv3_diskless.root_hostnam, 0, "Path to nfs root"); SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_rootaddr, CTLFLAG_RD, &nfsv3_diskless.root_saddr, sizeof(nfsv3_diskless.root_saddr), "%Ssockaddr_in", "Diskless root nfs address"); void newnfsargs_ntoh(struct nfs_args *); static int nfs_mountdiskless(char *, struct sockaddr_in *, struct nfs_args *, struct thread *, struct vnode **, struct mount *); static void nfs_convert_diskless(void); static void nfs_convert_oargs(struct nfs_args *args, struct onfs_args *oargs); int newnfs_iosize(struct nfsmount *nmp) { int iosize, maxio; /* First, set the upper limit for iosize */ if (nmp->nm_flag & NFSMNT_NFSV4) { maxio = NFS_MAXBSIZE; } else if (nmp->nm_flag & NFSMNT_NFSV3) { if (nmp->nm_sotype == SOCK_DGRAM) maxio = NFS_MAXDGRAMDATA; else maxio = NFS_MAXBSIZE; } else { maxio = NFS_V2MAXDATA; } if (nmp->nm_rsize > maxio || nmp->nm_rsize == 0) nmp->nm_rsize = maxio; - if (nmp->nm_rsize > MAXBSIZE) - nmp->nm_rsize = MAXBSIZE; + if (nmp->nm_rsize > NFS_MAXBSIZE) + nmp->nm_rsize = NFS_MAXBSIZE; if (nmp->nm_readdirsize > maxio || nmp->nm_readdirsize == 0) nmp->nm_readdirsize = maxio; if (nmp->nm_readdirsize > nmp->nm_rsize) nmp->nm_readdirsize = nmp->nm_rsize; if (nmp->nm_wsize > maxio || nmp->nm_wsize == 0) nmp->nm_wsize = maxio; - if (nmp->nm_wsize > MAXBSIZE) - nmp->nm_wsize = MAXBSIZE; + if (nmp->nm_wsize > NFS_MAXBSIZE) + nmp->nm_wsize = NFS_MAXBSIZE; /* * Calculate the size used for io buffers. Use the larger * of the two sizes to minimise nfs requests but make sure * that it is at least one VM page to avoid wasting buffer * space. */ iosize = imax(nmp->nm_rsize, nmp->nm_wsize); iosize = imax(iosize, PAGE_SIZE); nmp->nm_mountp->mnt_stat.f_iosize = iosize; return (iosize); } static void nfs_convert_oargs(struct nfs_args *args, struct onfs_args *oargs) { args->version = NFS_ARGSVERSION; args->addr = oargs->addr; args->addrlen = oargs->addrlen; args->sotype = oargs->sotype; args->proto = oargs->proto; args->fh = oargs->fh; args->fhsize = oargs->fhsize; args->flags = oargs->flags; args->wsize = oargs->wsize; args->rsize = oargs->rsize; args->readdirsize = oargs->readdirsize; args->timeo = oargs->timeo; args->retrans = oargs->retrans; args->readahead = oargs->readahead; args->hostname = oargs->hostname; } static void nfs_convert_diskless(void) { bcopy(&nfs_diskless.myif, &nfsv3_diskless.myif, sizeof(struct ifaliasreq)); bcopy(&nfs_diskless.mygateway, &nfsv3_diskless.mygateway, sizeof(struct sockaddr_in)); nfs_convert_oargs(&nfsv3_diskless.root_args,&nfs_diskless.root_args); if (nfsv3_diskless.root_args.flags & NFSMNT_NFSV3) { nfsv3_diskless.root_fhsize = NFSX_MYFH; bcopy(nfs_diskless.root_fh, nfsv3_diskless.root_fh, NFSX_MYFH); } else { nfsv3_diskless.root_fhsize = NFSX_V2FH; bcopy(nfs_diskless.root_fh, nfsv3_diskless.root_fh, NFSX_V2FH); } bcopy(&nfs_diskless.root_saddr,&nfsv3_diskless.root_saddr, sizeof(struct sockaddr_in)); bcopy(nfs_diskless.root_hostnam, nfsv3_diskless.root_hostnam, MNAMELEN); nfsv3_diskless.root_time = nfs_diskless.root_time; bcopy(nfs_diskless.my_hostnam, nfsv3_diskless.my_hostnam, MAXHOSTNAMELEN); nfs_diskless_valid = 3; } /* * nfs statfs call */ static int nfs_statfs(struct mount *mp, struct statfs *sbp) { struct vnode *vp; struct thread *td; struct nfsmount *nmp = VFSTONFS(mp); struct nfsvattr nfsva; struct nfsfsinfo fs; struct nfsstatfs sb; int error = 0, attrflag, gotfsinfo = 0, ret; struct nfsnode *np; td = curthread; error = vfs_busy(mp, MBF_NOWAIT); if (error) return (error); error = ncl_nget(mp, nmp->nm_fh, nmp->nm_fhsize, &np, LK_EXCLUSIVE); if (error) { vfs_unbusy(mp); return (error); } vp = NFSTOV(np); mtx_lock(&nmp->nm_mtx); if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) { mtx_unlock(&nmp->nm_mtx); error = nfsrpc_fsinfo(vp, &fs, td->td_ucred, td, &nfsva, &attrflag, NULL); if (!error) gotfsinfo = 1; } else mtx_unlock(&nmp->nm_mtx); if (!error) error = nfsrpc_statfs(vp, &sb, &fs, td->td_ucred, td, &nfsva, &attrflag, NULL); if (error != 0) NFSCL_DEBUG(2, "statfs=%d\n", error); if (attrflag == 0) { ret = nfsrpc_getattrnovp(nmp, nmp->nm_fh, nmp->nm_fhsize, 1, td->td_ucred, td, &nfsva, NULL, NULL); if (ret) { /* * Just set default values to get things going. */ NFSBZERO((caddr_t)&nfsva, sizeof (struct nfsvattr)); nfsva.na_vattr.va_type = VDIR; nfsva.na_vattr.va_mode = 0777; nfsva.na_vattr.va_nlink = 100; nfsva.na_vattr.va_uid = (uid_t)0; nfsva.na_vattr.va_gid = (gid_t)0; nfsva.na_vattr.va_fileid = 2; nfsva.na_vattr.va_gen = 1; nfsva.na_vattr.va_blocksize = NFS_FABLKSIZE; nfsva.na_vattr.va_size = 512 * 1024; } } (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); if (!error) { mtx_lock(&nmp->nm_mtx); if (gotfsinfo || (nmp->nm_flag & NFSMNT_NFSV4)) nfscl_loadfsinfo(nmp, &fs); nfscl_loadsbinfo(nmp, &sb, sbp); sbp->f_iosize = newnfs_iosize(nmp); mtx_unlock(&nmp->nm_mtx); if (sbp != &mp->mnt_stat) { bcopy(mp->mnt_stat.f_mntonname, sbp->f_mntonname, MNAMELEN); bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN); } strncpy(&sbp->f_fstypename[0], mp->mnt_vfc->vfc_name, MFSNAMELEN); } else if (NFS_ISV4(vp)) { error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); } vput(vp); vfs_unbusy(mp); return (error); } /* * nfs version 3 fsinfo rpc call */ int ncl_fsinfo(struct nfsmount *nmp, struct vnode *vp, struct ucred *cred, struct thread *td) { struct nfsfsinfo fs; struct nfsvattr nfsva; int error, attrflag; error = nfsrpc_fsinfo(vp, &fs, cred, td, &nfsva, &attrflag, NULL); if (!error) { if (attrflag) (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); mtx_lock(&nmp->nm_mtx); nfscl_loadfsinfo(nmp, &fs); mtx_unlock(&nmp->nm_mtx); } return (error); } /* * Mount a remote root fs via. nfs. This depends on the info in the * nfs_diskless structure that has been filled in properly by some primary * bootstrap. * It goes something like this: * - do enough of "ifconfig" by calling ifioctl() so that the system * can talk to the server * - If nfs_diskless.mygateway is filled in, use that address as * a default gateway. * - build the rootfs mount point and call mountnfs() to do the rest. * * It is assumed to be safe to read, modify, and write the nfsv3_diskless * structure, as well as other global NFS client variables here, as * nfs_mountroot() will be called once in the boot before any other NFS * client activity occurs. */ static int nfs_mountroot(struct mount *mp) { struct thread *td = curthread; struct nfsv3_diskless *nd = &nfsv3_diskless; struct socket *so; struct vnode *vp; struct ifreq ir; int error; u_long l; char buf[128]; char *cp; #if defined(BOOTP_NFSROOT) && defined(BOOTP) bootpc_init(); /* use bootp to get nfs_diskless filled in */ #elif defined(NFS_ROOT) nfs_setup_diskless(); #endif if (nfs_diskless_valid == 0) return (-1); if (nfs_diskless_valid == 1) nfs_convert_diskless(); /* * XXX splnet, so networks will receive... */ splnet(); /* * Do enough of ifconfig(8) so that the critical net interface can * talk to the server. */ error = socreate(nd->myif.ifra_addr.sa_family, &so, nd->root_args.sotype, 0, td->td_ucred, td); if (error) panic("nfs_mountroot: socreate(%04x): %d", nd->myif.ifra_addr.sa_family, error); #if 0 /* XXX Bad idea */ /* * We might not have been told the right interface, so we pass * over the first ten interfaces of the same kind, until we get * one of them configured. */ for (i = strlen(nd->myif.ifra_name) - 1; nd->myif.ifra_name[i] >= '0' && nd->myif.ifra_name[i] <= '9'; nd->myif.ifra_name[i] ++) { error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, td); if(!error) break; } #endif error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, td); if (error) panic("nfs_mountroot: SIOCAIFADDR: %d", error); if ((cp = kern_getenv("boot.netif.mtu")) != NULL) { ir.ifr_mtu = strtol(cp, NULL, 10); bcopy(nd->myif.ifra_name, ir.ifr_name, IFNAMSIZ); freeenv(cp); error = ifioctl(so, SIOCSIFMTU, (caddr_t)&ir, td); if (error) printf("nfs_mountroot: SIOCSIFMTU: %d", error); } soclose(so); /* * If the gateway field is filled in, set it as the default route. * Note that pxeboot will set a default route of 0 if the route * is not set by the DHCP server. Check also for a value of 0 * to avoid panicking inappropriately in that situation. */ if (nd->mygateway.sin_len != 0 && nd->mygateway.sin_addr.s_addr != 0) { struct sockaddr_in mask, sin; bzero((caddr_t)&mask, sizeof(mask)); sin = mask; sin.sin_family = AF_INET; sin.sin_len = sizeof(sin); /* XXX MRT use table 0 for this sort of thing */ CURVNET_SET(TD_TO_VNET(td)); error = rtrequest_fib(RTM_ADD, (struct sockaddr *)&sin, (struct sockaddr *)&nd->mygateway, (struct sockaddr *)&mask, RTF_UP | RTF_GATEWAY, NULL, RT_DEFAULT_FIB); CURVNET_RESTORE(); if (error) panic("nfs_mountroot: RTM_ADD: %d", error); } /* * Create the rootfs mount point. */ nd->root_args.fh = nd->root_fh; nd->root_args.fhsize = nd->root_fhsize; l = ntohl(nd->root_saddr.sin_addr.s_addr); snprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s", (l >> 24) & 0xff, (l >> 16) & 0xff, (l >> 8) & 0xff, (l >> 0) & 0xff, nd->root_hostnam); printf("NFS ROOT: %s\n", buf); nd->root_args.hostname = buf; if ((error = nfs_mountdiskless(buf, &nd->root_saddr, &nd->root_args, td, &vp, mp)) != 0) { return (error); } /* * This is not really an nfs issue, but it is much easier to * set hostname here and then let the "/etc/rc.xxx" files * mount the right /var based upon its preset value. */ mtx_lock(&prison0.pr_mtx); strlcpy(prison0.pr_hostname, nd->my_hostnam, sizeof(prison0.pr_hostname)); mtx_unlock(&prison0.pr_mtx); inittodr(ntohl(nd->root_time)); return (0); } /* * Internal version of mount system call for diskless setup. */ static int nfs_mountdiskless(char *path, struct sockaddr_in *sin, struct nfs_args *args, struct thread *td, struct vnode **vpp, struct mount *mp) { struct sockaddr *nam; int dirlen, error; char *dirpath; /* * Find the directory path in "path", which also has the server's * name/ip address in it. */ dirpath = strchr(path, ':'); if (dirpath != NULL) dirlen = strlen(++dirpath); else dirlen = 0; nam = sodupsockaddr((struct sockaddr *)sin, M_WAITOK); if ((error = mountnfs(args, mp, nam, path, NULL, 0, dirpath, dirlen, NULL, 0, vpp, td->td_ucred, td, NFS_DEFAULT_NAMETIMEO, NFS_DEFAULT_NEGNAMETIMEO, 0)) != 0) { printf("nfs_mountroot: mount %s on /: %d\n", path, error); return (error); } return (0); } static void nfs_sec_name(char *sec, int *flagsp) { if (!strcmp(sec, "krb5")) *flagsp |= NFSMNT_KERB; else if (!strcmp(sec, "krb5i")) *flagsp |= (NFSMNT_KERB | NFSMNT_INTEGRITY); else if (!strcmp(sec, "krb5p")) *flagsp |= (NFSMNT_KERB | NFSMNT_PRIVACY); } static void nfs_decode_args(struct mount *mp, struct nfsmount *nmp, struct nfs_args *argp, const char *hostname, struct ucred *cred, struct thread *td) { int s; int adjsock; char *p; s = splnet(); /* * Set read-only flag if requested; otherwise, clear it if this is * an update. If this is not an update, then either the read-only * flag is already clear, or this is a root mount and it was set * intentionally at some previous point. */ if (vfs_getopt(mp->mnt_optnew, "ro", NULL, NULL) == 0) { MNT_ILOCK(mp); mp->mnt_flag |= MNT_RDONLY; MNT_IUNLOCK(mp); } else if (mp->mnt_flag & MNT_UPDATE) { MNT_ILOCK(mp); mp->mnt_flag &= ~MNT_RDONLY; MNT_IUNLOCK(mp); } /* * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes * no sense in that context. Also, set up appropriate retransmit * and soft timeout behavior. */ if (argp->sotype == SOCK_STREAM) { nmp->nm_flag &= ~NFSMNT_NOCONN; nmp->nm_timeo = NFS_MAXTIMEO; if ((argp->flags & NFSMNT_NFSV4) != 0) nmp->nm_retry = INT_MAX; else nmp->nm_retry = NFS_RETRANS_TCP; } /* Also clear RDIRPLUS if NFSv2, it crashes some servers */ if ((argp->flags & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) == 0) { argp->flags &= ~NFSMNT_RDIRPLUS; nmp->nm_flag &= ~NFSMNT_RDIRPLUS; } /* Re-bind if rsrvd port requested and wasn't on one */ adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT) && (argp->flags & NFSMNT_RESVPORT); /* Also re-bind if we're switching to/from a connected UDP socket */ adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) != (argp->flags & NFSMNT_NOCONN)); /* Update flags atomically. Don't change the lock bits. */ nmp->nm_flag = argp->flags | nmp->nm_flag; splx(s); if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) { nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10; if (nmp->nm_timeo < NFS_MINTIMEO) nmp->nm_timeo = NFS_MINTIMEO; else if (nmp->nm_timeo > NFS_MAXTIMEO) nmp->nm_timeo = NFS_MAXTIMEO; } if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) { nmp->nm_retry = argp->retrans; if (nmp->nm_retry > NFS_MAXREXMIT) nmp->nm_retry = NFS_MAXREXMIT; } if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) { nmp->nm_wsize = argp->wsize; /* * Clip at the power of 2 below the size. There is an * issue (not isolated) that causes intermittent page * faults if this is not done. */ if (nmp->nm_wsize > NFS_FABLKSIZE) nmp->nm_wsize = 1 << (fls(nmp->nm_wsize) - 1); else nmp->nm_wsize = NFS_FABLKSIZE; } if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) { nmp->nm_rsize = argp->rsize; /* * Clip at the power of 2 below the size. There is an * issue (not isolated) that causes intermittent page * faults if this is not done. */ if (nmp->nm_rsize > NFS_FABLKSIZE) nmp->nm_rsize = 1 << (fls(nmp->nm_rsize) - 1); else nmp->nm_rsize = NFS_FABLKSIZE; } if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) { nmp->nm_readdirsize = argp->readdirsize; } if ((argp->flags & NFSMNT_ACREGMIN) && argp->acregmin >= 0) nmp->nm_acregmin = argp->acregmin; else nmp->nm_acregmin = NFS_MINATTRTIMO; if ((argp->flags & NFSMNT_ACREGMAX) && argp->acregmax >= 0) nmp->nm_acregmax = argp->acregmax; else nmp->nm_acregmax = NFS_MAXATTRTIMO; if ((argp->flags & NFSMNT_ACDIRMIN) && argp->acdirmin >= 0) nmp->nm_acdirmin = argp->acdirmin; else nmp->nm_acdirmin = NFS_MINDIRATTRTIMO; if ((argp->flags & NFSMNT_ACDIRMAX) && argp->acdirmax >= 0) nmp->nm_acdirmax = argp->acdirmax; else nmp->nm_acdirmax = NFS_MAXDIRATTRTIMO; if (nmp->nm_acdirmin > nmp->nm_acdirmax) nmp->nm_acdirmin = nmp->nm_acdirmax; if (nmp->nm_acregmin > nmp->nm_acregmax) nmp->nm_acregmin = nmp->nm_acregmax; if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0) { if (argp->readahead <= NFS_MAXRAHEAD) nmp->nm_readahead = argp->readahead; else nmp->nm_readahead = NFS_MAXRAHEAD; } if ((argp->flags & NFSMNT_WCOMMITSIZE) && argp->wcommitsize >= 0) { if (argp->wcommitsize < nmp->nm_wsize) nmp->nm_wcommitsize = nmp->nm_wsize; else nmp->nm_wcommitsize = argp->wcommitsize; } adjsock |= ((nmp->nm_sotype != argp->sotype) || (nmp->nm_soproto != argp->proto)); if (nmp->nm_client != NULL && adjsock) { int haslock = 0, error = 0; if (nmp->nm_sotype == SOCK_STREAM) { error = newnfs_sndlock(&nmp->nm_sockreq.nr_lock); if (!error) haslock = 1; } if (!error) { newnfs_disconnect(&nmp->nm_sockreq); if (haslock) newnfs_sndunlock(&nmp->nm_sockreq.nr_lock); nmp->nm_sotype = argp->sotype; nmp->nm_soproto = argp->proto; if (nmp->nm_sotype == SOCK_DGRAM) while (newnfs_connect(nmp, &nmp->nm_sockreq, cred, td, 0)) { printf("newnfs_args: retrying connect\n"); (void) nfs_catnap(PSOCK, 0, "nfscon"); } } } else { nmp->nm_sotype = argp->sotype; nmp->nm_soproto = argp->proto; } if (hostname != NULL) { strlcpy(nmp->nm_hostname, hostname, sizeof(nmp->nm_hostname)); p = strchr(nmp->nm_hostname, ':'); if (p != NULL) *p = '\0'; } } static const char *nfs_opts[] = { "from", "nfs_args", "noac", "noatime", "noexec", "suiddir", "nosuid", "nosymfollow", "union", "noclusterr", "noclusterw", "multilabel", "acls", "force", "update", "async", "noconn", "nolockd", "conn", "lockd", "intr", "rdirplus", "readdirsize", "soft", "hard", "mntudp", "tcp", "udp", "wsize", "rsize", "retrans", "actimeo", "acregmin", "acregmax", "acdirmin", "acdirmax", "resvport", "readahead", "hostname", "timeo", "timeout", "addr", "fh", "nfsv3", "sec", "principal", "nfsv4", "gssname", "allgssname", "dirpath", "minorversion", "nametimeo", "negnametimeo", "nocto", "noncontigwr", "pnfs", "wcommitsize", NULL }; /* * VFS Operations. * * mount system call * It seems a bit dumb to copyinstr() the host and path here and then * bcopy() them in mountnfs(), but I wanted to detect errors before * doing the sockargs() call because sockargs() allocates an mbuf and * an error after that means that I have to release the mbuf. */ /* ARGSUSED */ static int nfs_mount(struct mount *mp) { struct nfs_args args = { .version = NFS_ARGSVERSION, .addr = NULL, .addrlen = sizeof (struct sockaddr_in), .sotype = SOCK_STREAM, .proto = 0, .fh = NULL, .fhsize = 0, .flags = NFSMNT_RESVPORT, .wsize = NFS_WSIZE, .rsize = NFS_RSIZE, .readdirsize = NFS_READDIRSIZE, .timeo = 10, .retrans = NFS_RETRANS, .readahead = NFS_DEFRAHEAD, .wcommitsize = 0, /* was: NQ_DEFLEASE */ .hostname = NULL, .acregmin = NFS_MINATTRTIMO, .acregmax = NFS_MAXATTRTIMO, .acdirmin = NFS_MINDIRATTRTIMO, .acdirmax = NFS_MAXDIRATTRTIMO, }; int error = 0, ret, len; struct sockaddr *nam = NULL; struct vnode *vp; struct thread *td; char hst[MNAMELEN]; u_char nfh[NFSX_FHMAX], krbname[100], dirpath[100], srvkrbname[100]; char *opt, *name, *secname; int nametimeo = NFS_DEFAULT_NAMETIMEO; int negnametimeo = NFS_DEFAULT_NEGNAMETIMEO; int minvers = 0; int dirlen, has_nfs_args_opt, krbnamelen, srvkrbnamelen; size_t hstlen; has_nfs_args_opt = 0; if (vfs_filteropt(mp->mnt_optnew, nfs_opts)) { error = EINVAL; goto out; } td = curthread; if ((mp->mnt_flag & (MNT_ROOTFS | MNT_UPDATE)) == MNT_ROOTFS) { error = nfs_mountroot(mp); goto out; } nfscl_init(); /* * The old mount_nfs program passed the struct nfs_args * from userspace to kernel. The new mount_nfs program * passes string options via nmount() from userspace to kernel * and we populate the struct nfs_args in the kernel. */ if (vfs_getopt(mp->mnt_optnew, "nfs_args", NULL, NULL) == 0) { error = vfs_copyopt(mp->mnt_optnew, "nfs_args", &args, sizeof(args)); if (error != 0) goto out; if (args.version != NFS_ARGSVERSION) { error = EPROGMISMATCH; goto out; } has_nfs_args_opt = 1; } /* Handle the new style options. */ if (vfs_getopt(mp->mnt_optnew, "noac", NULL, NULL) == 0) { args.acdirmin = args.acdirmax = args.acregmin = args.acregmax = 0; args.flags |= NFSMNT_ACDIRMIN | NFSMNT_ACDIRMAX | NFSMNT_ACREGMIN | NFSMNT_ACREGMAX; } if (vfs_getopt(mp->mnt_optnew, "noconn", NULL, NULL) == 0) args.flags |= NFSMNT_NOCONN; if (vfs_getopt(mp->mnt_optnew, "conn", NULL, NULL) == 0) args.flags &= ~NFSMNT_NOCONN; if (vfs_getopt(mp->mnt_optnew, "nolockd", NULL, NULL) == 0) args.flags |= NFSMNT_NOLOCKD; if (vfs_getopt(mp->mnt_optnew, "lockd", NULL, NULL) == 0) args.flags &= ~NFSMNT_NOLOCKD; if (vfs_getopt(mp->mnt_optnew, "intr", NULL, NULL) == 0) args.flags |= NFSMNT_INT; if (vfs_getopt(mp->mnt_optnew, "rdirplus", NULL, NULL) == 0) args.flags |= NFSMNT_RDIRPLUS; if (vfs_getopt(mp->mnt_optnew, "resvport", NULL, NULL) == 0) args.flags |= NFSMNT_RESVPORT; if (vfs_getopt(mp->mnt_optnew, "noresvport", NULL, NULL) == 0) args.flags &= ~NFSMNT_RESVPORT; if (vfs_getopt(mp->mnt_optnew, "soft", NULL, NULL) == 0) args.flags |= NFSMNT_SOFT; if (vfs_getopt(mp->mnt_optnew, "hard", NULL, NULL) == 0) args.flags &= ~NFSMNT_SOFT; if (vfs_getopt(mp->mnt_optnew, "mntudp", NULL, NULL) == 0) args.sotype = SOCK_DGRAM; if (vfs_getopt(mp->mnt_optnew, "udp", NULL, NULL) == 0) args.sotype = SOCK_DGRAM; if (vfs_getopt(mp->mnt_optnew, "tcp", NULL, NULL) == 0) args.sotype = SOCK_STREAM; if (vfs_getopt(mp->mnt_optnew, "nfsv3", NULL, NULL) == 0) args.flags |= NFSMNT_NFSV3; if (vfs_getopt(mp->mnt_optnew, "nfsv4", NULL, NULL) == 0) { args.flags |= NFSMNT_NFSV4; args.sotype = SOCK_STREAM; } if (vfs_getopt(mp->mnt_optnew, "allgssname", NULL, NULL) == 0) args.flags |= NFSMNT_ALLGSSNAME; if (vfs_getopt(mp->mnt_optnew, "nocto", NULL, NULL) == 0) args.flags |= NFSMNT_NOCTO; if (vfs_getopt(mp->mnt_optnew, "noncontigwr", NULL, NULL) == 0) args.flags |= NFSMNT_NONCONTIGWR; if (vfs_getopt(mp->mnt_optnew, "pnfs", NULL, NULL) == 0) args.flags |= NFSMNT_PNFS; if (vfs_getopt(mp->mnt_optnew, "readdirsize", (void **)&opt, NULL) == 0) { if (opt == NULL) { vfs_mount_error(mp, "illegal readdirsize"); error = EINVAL; goto out; } ret = sscanf(opt, "%d", &args.readdirsize); if (ret != 1 || args.readdirsize <= 0) { vfs_mount_error(mp, "illegal readdirsize: %s", opt); error = EINVAL; goto out; } args.flags |= NFSMNT_READDIRSIZE; } if (vfs_getopt(mp->mnt_optnew, "readahead", (void **)&opt, NULL) == 0) { if (opt == NULL) { vfs_mount_error(mp, "illegal readahead"); error = EINVAL; goto out; } ret = sscanf(opt, "%d", &args.readahead); if (ret != 1 || args.readahead <= 0) { vfs_mount_error(mp, "illegal readahead: %s", opt); error = EINVAL; goto out; } args.flags |= NFSMNT_READAHEAD; } if (vfs_getopt(mp->mnt_optnew, "wsize", (void **)&opt, NULL) == 0) { if (opt == NULL) { vfs_mount_error(mp, "illegal wsize"); error = EINVAL; goto out; } ret = sscanf(opt, "%d", &args.wsize); if (ret != 1 || args.wsize <= 0) { vfs_mount_error(mp, "illegal wsize: %s", opt); error = EINVAL; goto out; } args.flags |= NFSMNT_WSIZE; } if (vfs_getopt(mp->mnt_optnew, "rsize", (void **)&opt, NULL) == 0) { if (opt == NULL) { vfs_mount_error(mp, "illegal rsize"); error = EINVAL; goto out; } ret = sscanf(opt, "%d", &args.rsize); if (ret != 1 || args.rsize <= 0) { vfs_mount_error(mp, "illegal wsize: %s", opt); error = EINVAL; goto out; } args.flags |= NFSMNT_RSIZE; } if (vfs_getopt(mp->mnt_optnew, "retrans", (void **)&opt, NULL) == 0) { if (opt == NULL) { vfs_mount_error(mp, "illegal retrans"); error = EINVAL; goto out; } ret = sscanf(opt, "%d", &args.retrans); if (ret != 1 || args.retrans <= 0) { vfs_mount_error(mp, "illegal retrans: %s", opt); error = EINVAL; goto out; } args.flags |= NFSMNT_RETRANS; } if (vfs_getopt(mp->mnt_optnew, "actimeo", (void **)&opt, NULL) == 0) { ret = sscanf(opt, "%d", &args.acregmin); if (ret != 1 || args.acregmin < 0) { vfs_mount_error(mp, "illegal actimeo: %s", opt); error = EINVAL; goto out; } args.acdirmin = args.acdirmax = args.acregmax = args.acregmin; args.flags |= NFSMNT_ACDIRMIN | NFSMNT_ACDIRMAX | NFSMNT_ACREGMIN | NFSMNT_ACREGMAX; } if (vfs_getopt(mp->mnt_optnew, "acregmin", (void **)&opt, NULL) == 0) { ret = sscanf(opt, "%d", &args.acregmin); if (ret != 1 || args.acregmin < 0) { vfs_mount_error(mp, "illegal acregmin: %s", opt); error = EINVAL; goto out; } args.flags |= NFSMNT_ACREGMIN; } if (vfs_getopt(mp->mnt_optnew, "acregmax", (void **)&opt, NULL) == 0) { ret = sscanf(opt, "%d", &args.acregmax); if (ret != 1 || args.acregmax < 0) { vfs_mount_error(mp, "illegal acregmax: %s", opt); error = EINVAL; goto out; } args.flags |= NFSMNT_ACREGMAX; } if (vfs_getopt(mp->mnt_optnew, "acdirmin", (void **)&opt, NULL) == 0) { ret = sscanf(opt, "%d", &args.acdirmin); if (ret != 1 || args.acdirmin < 0) { vfs_mount_error(mp, "illegal acdirmin: %s", opt); error = EINVAL; goto out; } args.flags |= NFSMNT_ACDIRMIN; } if (vfs_getopt(mp->mnt_optnew, "acdirmax", (void **)&opt, NULL) == 0) { ret = sscanf(opt, "%d", &args.acdirmax); if (ret != 1 || args.acdirmax < 0) { vfs_mount_error(mp, "illegal acdirmax: %s", opt); error = EINVAL; goto out; } args.flags |= NFSMNT_ACDIRMAX; } if (vfs_getopt(mp->mnt_optnew, "wcommitsize", (void **)&opt, NULL) == 0) { ret = sscanf(opt, "%d", &args.wcommitsize); if (ret != 1 || args.wcommitsize < 0) { vfs_mount_error(mp, "illegal wcommitsize: %s", opt); error = EINVAL; goto out; } args.flags |= NFSMNT_WCOMMITSIZE; } if (vfs_getopt(mp->mnt_optnew, "timeo", (void **)&opt, NULL) == 0) { ret = sscanf(opt, "%d", &args.timeo); if (ret != 1 || args.timeo <= 0) { vfs_mount_error(mp, "illegal timeo: %s", opt); error = EINVAL; goto out; } args.flags |= NFSMNT_TIMEO; } if (vfs_getopt(mp->mnt_optnew, "timeout", (void **)&opt, NULL) == 0) { ret = sscanf(opt, "%d", &args.timeo); if (ret != 1 || args.timeo <= 0) { vfs_mount_error(mp, "illegal timeout: %s", opt); error = EINVAL; goto out; } args.flags |= NFSMNT_TIMEO; } if (vfs_getopt(mp->mnt_optnew, "nametimeo", (void **)&opt, NULL) == 0) { ret = sscanf(opt, "%d", &nametimeo); if (ret != 1 || nametimeo < 0) { vfs_mount_error(mp, "illegal nametimeo: %s", opt); error = EINVAL; goto out; } } if (vfs_getopt(mp->mnt_optnew, "negnametimeo", (void **)&opt, NULL) == 0) { ret = sscanf(opt, "%d", &negnametimeo); if (ret != 1 || negnametimeo < 0) { vfs_mount_error(mp, "illegal negnametimeo: %s", opt); error = EINVAL; goto out; } } if (vfs_getopt(mp->mnt_optnew, "minorversion", (void **)&opt, NULL) == 0) { ret = sscanf(opt, "%d", &minvers); if (ret != 1 || minvers < 0 || minvers > 1 || (args.flags & NFSMNT_NFSV4) == 0) { vfs_mount_error(mp, "illegal minorversion: %s", opt); error = EINVAL; goto out; } } if (vfs_getopt(mp->mnt_optnew, "sec", (void **) &secname, NULL) == 0) nfs_sec_name(secname, &args.flags); if (mp->mnt_flag & MNT_UPDATE) { struct nfsmount *nmp = VFSTONFS(mp); if (nmp == NULL) { error = EIO; goto out; } /* * If a change from TCP->UDP is done and there are thread(s) * that have I/O RPC(s) in progress with a tranfer size * greater than NFS_MAXDGRAMDATA, those thread(s) will be * hung, retrying the RPC(s) forever. Usually these threads * will be seen doing an uninterruptible sleep on wait channel * "nfsreq". */ if (args.sotype == SOCK_DGRAM && nmp->nm_sotype == SOCK_STREAM) tprintf(td->td_proc, LOG_WARNING, "Warning: mount -u that changes TCP->UDP can result in hung threads\n"); /* * When doing an update, we can't change version, * security, switch lockd strategies or change cookie * translation */ args.flags = (args.flags & ~(NFSMNT_NFSV3 | NFSMNT_NFSV4 | NFSMNT_KERB | NFSMNT_INTEGRITY | NFSMNT_PRIVACY | NFSMNT_NOLOCKD /*|NFSMNT_XLATECOOKIE*/)) | (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4 | NFSMNT_KERB | NFSMNT_INTEGRITY | NFSMNT_PRIVACY | NFSMNT_NOLOCKD /*|NFSMNT_XLATECOOKIE*/)); nfs_decode_args(mp, nmp, &args, NULL, td->td_ucred, td); goto out; } /* * Make the nfs_ip_paranoia sysctl serve as the default connection * or no-connection mode for those protocols that support * no-connection mode (the flag will be cleared later for protocols * that do not support no-connection mode). This will allow a client * to receive replies from a different IP then the request was * sent to. Note: default value for nfs_ip_paranoia is 1 (paranoid), * not 0. */ if (nfs_ip_paranoia == 0) args.flags |= NFSMNT_NOCONN; if (has_nfs_args_opt != 0) { /* * In the 'nfs_args' case, the pointers in the args * structure are in userland - we copy them in here. */ if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX) { vfs_mount_error(mp, "Bad file handle"); error = EINVAL; goto out; } error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize); if (error != 0) goto out; error = copyinstr(args.hostname, hst, MNAMELEN - 1, &hstlen); if (error != 0) goto out; bzero(&hst[hstlen], MNAMELEN - hstlen); args.hostname = hst; /* sockargs() call must be after above copyin() calls */ error = getsockaddr(&nam, (caddr_t)args.addr, args.addrlen); if (error != 0) goto out; } else { if (vfs_getopt(mp->mnt_optnew, "fh", (void **)&args.fh, &args.fhsize) == 0) { if (args.fhsize < 0 || args.fhsize > NFSX_FHMAX) { vfs_mount_error(mp, "Bad file handle"); error = EINVAL; goto out; } bcopy(args.fh, nfh, args.fhsize); } else { args.fhsize = 0; } (void) vfs_getopt(mp->mnt_optnew, "hostname", (void **)&args.hostname, &len); if (args.hostname == NULL) { vfs_mount_error(mp, "Invalid hostname"); error = EINVAL; goto out; } bcopy(args.hostname, hst, MNAMELEN); hst[MNAMELEN - 1] = '\0'; } if (vfs_getopt(mp->mnt_optnew, "principal", (void **)&name, NULL) == 0) strlcpy(srvkrbname, name, sizeof (srvkrbname)); else snprintf(srvkrbname, sizeof (srvkrbname), "nfs@%s", hst); srvkrbnamelen = strlen(srvkrbname); if (vfs_getopt(mp->mnt_optnew, "gssname", (void **)&name, NULL) == 0) strlcpy(krbname, name, sizeof (krbname)); else krbname[0] = '\0'; krbnamelen = strlen(krbname); if (vfs_getopt(mp->mnt_optnew, "dirpath", (void **)&name, NULL) == 0) strlcpy(dirpath, name, sizeof (dirpath)); else dirpath[0] = '\0'; dirlen = strlen(dirpath); if (has_nfs_args_opt == 0) { if (vfs_getopt(mp->mnt_optnew, "addr", (void **)&args.addr, &args.addrlen) == 0) { if (args.addrlen > SOCK_MAXADDRLEN) { error = ENAMETOOLONG; goto out; } nam = malloc(args.addrlen, M_SONAME, M_WAITOK); bcopy(args.addr, nam, args.addrlen); nam->sa_len = args.addrlen; } else { vfs_mount_error(mp, "No server address"); error = EINVAL; goto out; } } args.fh = nfh; error = mountnfs(&args, mp, nam, hst, krbname, krbnamelen, dirpath, dirlen, srvkrbname, srvkrbnamelen, &vp, td->td_ucred, td, nametimeo, negnametimeo, minvers); out: if (!error) { MNT_ILOCK(mp); mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_NO_IOPF | MNTK_USES_BCACHE; MNT_IUNLOCK(mp); } return (error); } /* * VFS Operations. * * mount system call * It seems a bit dumb to copyinstr() the host and path here and then * bcopy() them in mountnfs(), but I wanted to detect errors before * doing the sockargs() call because sockargs() allocates an mbuf and * an error after that means that I have to release the mbuf. */ /* ARGSUSED */ static int nfs_cmount(struct mntarg *ma, void *data, uint64_t flags) { int error; struct nfs_args args; error = copyin(data, &args, sizeof (struct nfs_args)); if (error) return error; ma = mount_arg(ma, "nfs_args", &args, sizeof args); error = kernel_mount(ma, flags); return (error); } /* * Common code for mount and mountroot */ static int mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam, char *hst, u_char *krbname, int krbnamelen, u_char *dirpath, int dirlen, u_char *srvkrbname, int srvkrbnamelen, struct vnode **vpp, struct ucred *cred, struct thread *td, int nametimeo, int negnametimeo, int minvers) { struct nfsmount *nmp; struct nfsnode *np; int error, trycnt, ret; struct nfsvattr nfsva; struct nfsclclient *clp; struct nfsclds *dsp, *tdsp; uint32_t lease; static u_int64_t clval = 0; NFSCL_DEBUG(3, "in mnt\n"); clp = NULL; if (mp->mnt_flag & MNT_UPDATE) { nmp = VFSTONFS(mp); printf("%s: MNT_UPDATE is no longer handled here\n", __func__); FREE(nam, M_SONAME); return (0); } else { MALLOC(nmp, struct nfsmount *, sizeof (struct nfsmount) + krbnamelen + dirlen + srvkrbnamelen + 2, M_NEWNFSMNT, M_WAITOK | M_ZERO); TAILQ_INIT(&nmp->nm_bufq); if (clval == 0) clval = (u_int64_t)nfsboottime.tv_sec; nmp->nm_clval = clval++; nmp->nm_krbnamelen = krbnamelen; nmp->nm_dirpathlen = dirlen; nmp->nm_srvkrbnamelen = srvkrbnamelen; if (td->td_ucred->cr_uid != (uid_t)0) { /* * nm_uid is used to get KerberosV credentials for * the nfsv4 state handling operations if there is * no host based principal set. Use the uid of * this user if not root, since they are doing the * mount. I don't think setting this for root will * work, since root normally does not have user * credentials in a credentials cache. */ nmp->nm_uid = td->td_ucred->cr_uid; } else { /* * Just set to -1, so it won't be used. */ nmp->nm_uid = (uid_t)-1; } /* Copy and null terminate all the names */ if (nmp->nm_krbnamelen > 0) { bcopy(krbname, nmp->nm_krbname, nmp->nm_krbnamelen); nmp->nm_name[nmp->nm_krbnamelen] = '\0'; } if (nmp->nm_dirpathlen > 0) { bcopy(dirpath, NFSMNT_DIRPATH(nmp), nmp->nm_dirpathlen); nmp->nm_name[nmp->nm_krbnamelen + nmp->nm_dirpathlen + 1] = '\0'; } if (nmp->nm_srvkrbnamelen > 0) { bcopy(srvkrbname, NFSMNT_SRVKRBNAME(nmp), nmp->nm_srvkrbnamelen); nmp->nm_name[nmp->nm_krbnamelen + nmp->nm_dirpathlen + nmp->nm_srvkrbnamelen + 2] = '\0'; } nmp->nm_sockreq.nr_cred = crhold(cred); mtx_init(&nmp->nm_sockreq.nr_mtx, "nfssock", NULL, MTX_DEF); mp->mnt_data = nmp; nmp->nm_getinfo = nfs_getnlminfo; nmp->nm_vinvalbuf = ncl_vinvalbuf; } vfs_getnewfsid(mp); nmp->nm_mountp = mp; mtx_init(&nmp->nm_mtx, "NFSmount lock", NULL, MTX_DEF | MTX_DUPOK); /* * Since nfs_decode_args() might optionally set them, these * need to be set to defaults before the call, so that the * optional settings aren't overwritten. */ nmp->nm_nametimeo = nametimeo; nmp->nm_negnametimeo = negnametimeo; nmp->nm_timeo = NFS_TIMEO; nmp->nm_retry = NFS_RETRANS; nmp->nm_readahead = NFS_DEFRAHEAD; /* This is empirical approximation of sqrt(hibufspace) * 256. */ nmp->nm_wcommitsize = NFS_MAXBSIZE / 256; while ((long)nmp->nm_wcommitsize * nmp->nm_wcommitsize < hibufspace) nmp->nm_wcommitsize *= 2; nmp->nm_wcommitsize *= 256; if ((argp->flags & NFSMNT_NFSV4) != 0) nmp->nm_minorvers = minvers; else nmp->nm_minorvers = 0; nfs_decode_args(mp, nmp, argp, hst, cred, td); /* * V2 can only handle 32 bit filesizes. A 4GB-1 limit may be too * high, depending on whether we end up with negative offsets in * the client or server somewhere. 2GB-1 may be safer. * * For V3, ncl_fsinfo will adjust this as necessary. Assume maximum * that we can handle until we find out otherwise. */ if ((argp->flags & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) == 0) nmp->nm_maxfilesize = 0xffffffffLL; else nmp->nm_maxfilesize = OFF_MAX; if ((argp->flags & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) == 0) { nmp->nm_wsize = NFS_WSIZE; nmp->nm_rsize = NFS_RSIZE; nmp->nm_readdirsize = NFS_READDIRSIZE; } nmp->nm_numgrps = NFS_MAXGRPS; nmp->nm_tprintf_delay = nfs_tprintf_delay; if (nmp->nm_tprintf_delay < 0) nmp->nm_tprintf_delay = 0; nmp->nm_tprintf_initial_delay = nfs_tprintf_initial_delay; if (nmp->nm_tprintf_initial_delay < 0) nmp->nm_tprintf_initial_delay = 0; nmp->nm_fhsize = argp->fhsize; if (nmp->nm_fhsize > 0) bcopy((caddr_t)argp->fh, (caddr_t)nmp->nm_fh, argp->fhsize); bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN); nmp->nm_nam = nam; /* Set up the sockets and per-host congestion */ nmp->nm_sotype = argp->sotype; nmp->nm_soproto = argp->proto; nmp->nm_sockreq.nr_prog = NFS_PROG; if ((argp->flags & NFSMNT_NFSV4)) nmp->nm_sockreq.nr_vers = NFS_VER4; else if ((argp->flags & NFSMNT_NFSV3)) nmp->nm_sockreq.nr_vers = NFS_VER3; else nmp->nm_sockreq.nr_vers = NFS_VER2; if ((error = newnfs_connect(nmp, &nmp->nm_sockreq, cred, td, 0))) goto bad; /* For NFSv4.1, get the clientid now. */ if (nmp->nm_minorvers > 0) { NFSCL_DEBUG(3, "at getcl\n"); error = nfscl_getcl(mp, cred, td, 0, &clp); NFSCL_DEBUG(3, "aft getcl=%d\n", error); if (error != 0) goto bad; } if (nmp->nm_fhsize == 0 && (nmp->nm_flag & NFSMNT_NFSV4) && nmp->nm_dirpathlen > 0) { NFSCL_DEBUG(3, "in dirp\n"); /* * If the fhsize on the mount point == 0 for V4, the mount * path needs to be looked up. */ trycnt = 3; do { error = nfsrpc_getdirpath(nmp, NFSMNT_DIRPATH(nmp), cred, td); NFSCL_DEBUG(3, "aft dirp=%d\n", error); if (error) (void) nfs_catnap(PZERO, error, "nfsgetdirp"); } while (error && --trycnt > 0); if (error) { error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); goto bad; } } /* * A reference count is needed on the nfsnode representing the * remote root. If this object is not persistent, then backward * traversals of the mount point (i.e. "..") will not work if * the nfsnode gets flushed out of the cache. Ufs does not have * this problem, because one can identify root inodes by their * number == ROOTINO (2). */ if (nmp->nm_fhsize > 0) { /* * Set f_iosize to NFS_DIRBLKSIZ so that bo_bsize gets set * non-zero for the root vnode. f_iosize will be set correctly * by nfs_statfs() before any I/O occurs. */ mp->mnt_stat.f_iosize = NFS_DIRBLKSIZ; error = ncl_nget(mp, nmp->nm_fh, nmp->nm_fhsize, &np, LK_EXCLUSIVE); if (error) goto bad; *vpp = NFSTOV(np); /* * Get file attributes and transfer parameters for the * mountpoint. This has the side effect of filling in * (*vpp)->v_type with the correct value. */ ret = nfsrpc_getattrnovp(nmp, nmp->nm_fh, nmp->nm_fhsize, 1, cred, td, &nfsva, NULL, &lease); if (ret) { /* * Just set default values to get things going. */ NFSBZERO((caddr_t)&nfsva, sizeof (struct nfsvattr)); nfsva.na_vattr.va_type = VDIR; nfsva.na_vattr.va_mode = 0777; nfsva.na_vattr.va_nlink = 100; nfsva.na_vattr.va_uid = (uid_t)0; nfsva.na_vattr.va_gid = (gid_t)0; nfsva.na_vattr.va_fileid = 2; nfsva.na_vattr.va_gen = 1; nfsva.na_vattr.va_blocksize = NFS_FABLKSIZE; nfsva.na_vattr.va_size = 512 * 1024; lease = 60; } (void) nfscl_loadattrcache(vpp, &nfsva, NULL, NULL, 0, 1); if (nmp->nm_minorvers > 0) { NFSCL_DEBUG(3, "lease=%d\n", (int)lease); NFSLOCKCLSTATE(); clp->nfsc_renew = NFSCL_RENEW(lease); clp->nfsc_expire = NFSD_MONOSEC + clp->nfsc_renew; clp->nfsc_clientidrev++; if (clp->nfsc_clientidrev == 0) clp->nfsc_clientidrev++; NFSUNLOCKCLSTATE(); /* * Mount will succeed, so the renew thread can be * started now. */ nfscl_start_renewthread(clp); nfscl_clientrelease(clp); } if (argp->flags & NFSMNT_NFSV3) ncl_fsinfo(nmp, *vpp, cred, td); /* Mark if the mount point supports NFSv4 ACLs. */ if ((argp->flags & NFSMNT_NFSV4) != 0 && nfsrv_useacl != 0 && ret == 0 && NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL)) { MNT_ILOCK(mp); mp->mnt_flag |= MNT_NFS4ACLS; MNT_IUNLOCK(mp); } /* * Lose the lock but keep the ref. */ NFSVOPUNLOCK(*vpp, 0); return (0); } error = EIO; bad: if (clp != NULL) nfscl_clientrelease(clp); newnfs_disconnect(&nmp->nm_sockreq); crfree(nmp->nm_sockreq.nr_cred); if (nmp->nm_sockreq.nr_auth != NULL) AUTH_DESTROY(nmp->nm_sockreq.nr_auth); mtx_destroy(&nmp->nm_sockreq.nr_mtx); mtx_destroy(&nmp->nm_mtx); if (nmp->nm_clp != NULL) { NFSLOCKCLSTATE(); LIST_REMOVE(nmp->nm_clp, nfsc_list); NFSUNLOCKCLSTATE(); free(nmp->nm_clp, M_NFSCLCLIENT); } TAILQ_FOREACH_SAFE(dsp, &nmp->nm_sess, nfsclds_list, tdsp) nfscl_freenfsclds(dsp); FREE(nmp, M_NEWNFSMNT); FREE(nam, M_SONAME); return (error); } /* * unmount system call */ static int nfs_unmount(struct mount *mp, int mntflags) { struct thread *td; struct nfsmount *nmp; int error, flags = 0, i, trycnt = 0; struct nfsclds *dsp, *tdsp; td = curthread; if (mntflags & MNT_FORCE) flags |= FORCECLOSE; nmp = VFSTONFS(mp); /* * Goes something like this.. * - Call vflush() to clear out vnodes for this filesystem * - Close the socket * - Free up the data structures */ /* In the forced case, cancel any outstanding requests. */ if (mntflags & MNT_FORCE) { error = newnfs_nmcancelreqs(nmp); if (error) goto out; /* For a forced close, get rid of the renew thread now */ nfscl_umount(nmp, td); } /* We hold 1 extra ref on the root vnode; see comment in mountnfs(). */ do { error = vflush(mp, 1, flags, td); if ((mntflags & MNT_FORCE) && error != 0 && ++trycnt < 30) (void) nfs_catnap(PSOCK, error, "newndm"); } while ((mntflags & MNT_FORCE) && error != 0 && trycnt < 30); if (error) goto out; /* * We are now committed to the unmount. */ if ((mntflags & MNT_FORCE) == 0) nfscl_umount(nmp, td); /* Make sure no nfsiods are assigned to this mount. */ mtx_lock(&ncl_iod_mutex); for (i = 0; i < NFS_MAXASYNCDAEMON; i++) if (ncl_iodmount[i] == nmp) { ncl_iodwant[i] = NFSIOD_AVAILABLE; ncl_iodmount[i] = NULL; } mtx_unlock(&ncl_iod_mutex); newnfs_disconnect(&nmp->nm_sockreq); crfree(nmp->nm_sockreq.nr_cred); FREE(nmp->nm_nam, M_SONAME); if (nmp->nm_sockreq.nr_auth != NULL) AUTH_DESTROY(nmp->nm_sockreq.nr_auth); mtx_destroy(&nmp->nm_sockreq.nr_mtx); mtx_destroy(&nmp->nm_mtx); TAILQ_FOREACH_SAFE(dsp, &nmp->nm_sess, nfsclds_list, tdsp) nfscl_freenfsclds(dsp); FREE(nmp, M_NEWNFSMNT); out: return (error); } /* * Return root of a filesystem */ static int nfs_root(struct mount *mp, int flags, struct vnode **vpp) { struct vnode *vp; struct nfsmount *nmp; struct nfsnode *np; int error; nmp = VFSTONFS(mp); error = ncl_nget(mp, nmp->nm_fh, nmp->nm_fhsize, &np, flags); if (error) return error; vp = NFSTOV(np); /* * Get transfer parameters and attributes for root vnode once. */ mtx_lock(&nmp->nm_mtx); if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) { mtx_unlock(&nmp->nm_mtx); ncl_fsinfo(nmp, vp, curthread->td_ucred, curthread); } else mtx_unlock(&nmp->nm_mtx); if (vp->v_type == VNON) vp->v_type = VDIR; vp->v_vflag |= VV_ROOT; *vpp = vp; return (0); } /* * Flush out the buffer cache */ /* ARGSUSED */ static int nfs_sync(struct mount *mp, int waitfor) { struct vnode *vp, *mvp; struct thread *td; int error, allerror = 0; td = curthread; MNT_ILOCK(mp); /* * If a forced dismount is in progress, return from here so that * the umount(2) syscall doesn't get stuck in VFS_SYNC() before * calling VFS_UNMOUNT(). */ if ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) { MNT_IUNLOCK(mp); return (EBADF); } MNT_IUNLOCK(mp); /* * Force stale buffer cache information to be flushed. */ loop: MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { /* XXX Racy bv_cnt check. */ if (NFSVOPISLOCKED(vp) || vp->v_bufobj.bo_dirty.bv_cnt == 0 || waitfor == MNT_LAZY) { VI_UNLOCK(vp); continue; } if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) { MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); goto loop; } error = VOP_FSYNC(vp, waitfor, td); if (error) allerror = error; NFSVOPUNLOCK(vp, 0); vrele(vp); } return (allerror); } static int nfs_sysctl(struct mount *mp, fsctlop_t op, struct sysctl_req *req) { struct nfsmount *nmp = VFSTONFS(mp); struct vfsquery vq; int error; bzero(&vq, sizeof(vq)); switch (op) { #if 0 case VFS_CTL_NOLOCKS: val = (nmp->nm_flag & NFSMNT_NOLOCKS) ? 1 : 0; if (req->oldptr != NULL) { error = SYSCTL_OUT(req, &val, sizeof(val)); if (error) return (error); } if (req->newptr != NULL) { error = SYSCTL_IN(req, &val, sizeof(val)); if (error) return (error); if (val) nmp->nm_flag |= NFSMNT_NOLOCKS; else nmp->nm_flag &= ~NFSMNT_NOLOCKS; } break; #endif case VFS_CTL_QUERY: mtx_lock(&nmp->nm_mtx); if (nmp->nm_state & NFSSTA_TIMEO) vq.vq_flags |= VQ_NOTRESP; mtx_unlock(&nmp->nm_mtx); #if 0 if (!(nmp->nm_flag & NFSMNT_NOLOCKS) && (nmp->nm_state & NFSSTA_LOCKTIMEO)) vq.vq_flags |= VQ_NOTRESPLOCK; #endif error = SYSCTL_OUT(req, &vq, sizeof(vq)); break; case VFS_CTL_TIMEO: if (req->oldptr != NULL) { error = SYSCTL_OUT(req, &nmp->nm_tprintf_initial_delay, sizeof(nmp->nm_tprintf_initial_delay)); if (error) return (error); } if (req->newptr != NULL) { error = vfs_suser(mp, req->td); if (error) return (error); error = SYSCTL_IN(req, &nmp->nm_tprintf_initial_delay, sizeof(nmp->nm_tprintf_initial_delay)); if (error) return (error); if (nmp->nm_tprintf_initial_delay < 0) nmp->nm_tprintf_initial_delay = 0; } break; default: return (ENOTSUP); } return (0); } /* * Purge any RPCs in progress, so that they will all return errors. * This allows dounmount() to continue as far as VFS_UNMOUNT() for a * forced dismount. */ static void nfs_purge(struct mount *mp) { struct nfsmount *nmp = VFSTONFS(mp); newnfs_nmcancelreqs(nmp); } /* * Extract the information needed by the nlm from the nfs vnode. */ static void nfs_getnlminfo(struct vnode *vp, uint8_t *fhp, size_t *fhlenp, struct sockaddr_storage *sp, int *is_v3p, off_t *sizep, struct timeval *timeop) { struct nfsmount *nmp; struct nfsnode *np = VTONFS(vp); nmp = VFSTONFS(vp->v_mount); if (fhlenp != NULL) *fhlenp = (size_t)np->n_fhp->nfh_len; if (fhp != NULL) bcopy(np->n_fhp->nfh_fh, fhp, np->n_fhp->nfh_len); if (sp != NULL) bcopy(nmp->nm_nam, sp, min(nmp->nm_nam->sa_len, sizeof(*sp))); if (is_v3p != NULL) *is_v3p = NFS_ISV3(vp); if (sizep != NULL) *sizep = np->n_size; if (timeop != NULL) { timeop->tv_sec = nmp->nm_timeo / NFS_HZ; timeop->tv_usec = (nmp->nm_timeo % NFS_HZ) * (1000000 / NFS_HZ); } } /* * This function prints out an option name, based on the conditional * argument. */ static __inline void nfscl_printopt(struct nfsmount *nmp, int testval, char *opt, char **buf, size_t *blen) { int len; if (testval != 0 && *blen > strlen(opt)) { len = snprintf(*buf, *blen, "%s", opt); if (len != strlen(opt)) printf("EEK!!\n"); *buf += len; *blen -= len; } } /* * This function printf out an options integer value. */ static __inline void nfscl_printoptval(struct nfsmount *nmp, int optval, char *opt, char **buf, size_t *blen) { int len; if (*blen > strlen(opt) + 1) { /* Could result in truncated output string. */ len = snprintf(*buf, *blen, "%s=%d", opt, optval); if (len < *blen) { *buf += len; *blen -= len; } } } /* * Load the option flags and values into the buffer. */ void nfscl_retopts(struct nfsmount *nmp, char *buffer, size_t buflen) { char *buf; size_t blen; buf = buffer; blen = buflen; nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_NFSV4) != 0, "nfsv4", &buf, &blen); if ((nmp->nm_flag & NFSMNT_NFSV4) != 0) { nfscl_printoptval(nmp, nmp->nm_minorvers, ",minorversion", &buf, &blen); nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_PNFS) != 0, ",pnfs", &buf, &blen); } nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_NFSV3) != 0, "nfsv3", &buf, &blen); nfscl_printopt(nmp, (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) == 0, "nfsv2", &buf, &blen); nfscl_printopt(nmp, nmp->nm_sotype == SOCK_STREAM, ",tcp", &buf, &blen); nfscl_printopt(nmp, nmp->nm_sotype != SOCK_STREAM, ",udp", &buf, &blen); nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_RESVPORT) != 0, ",resvport", &buf, &blen); nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_NOCONN) != 0, ",noconn", &buf, &blen); nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_SOFT) == 0, ",hard", &buf, &blen); nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_SOFT) != 0, ",soft", &buf, &blen); nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_INT) != 0, ",intr", &buf, &blen); nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_NOCTO) == 0, ",cto", &buf, &blen); nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_NOCTO) != 0, ",nocto", &buf, &blen); nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_NONCONTIGWR) != 0, ",noncontigwr", &buf, &blen); nfscl_printopt(nmp, (nmp->nm_flag & (NFSMNT_NOLOCKD | NFSMNT_NFSV4)) == 0, ",lockd", &buf, &blen); nfscl_printopt(nmp, (nmp->nm_flag & (NFSMNT_NOLOCKD | NFSMNT_NFSV4)) == NFSMNT_NOLOCKD, ",nolockd", &buf, &blen); nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_RDIRPLUS) != 0, ",rdirplus", &buf, &blen); nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_KERB) == 0, ",sec=sys", &buf, &blen); nfscl_printopt(nmp, (nmp->nm_flag & (NFSMNT_KERB | NFSMNT_INTEGRITY | NFSMNT_PRIVACY)) == NFSMNT_KERB, ",sec=krb5", &buf, &blen); nfscl_printopt(nmp, (nmp->nm_flag & (NFSMNT_KERB | NFSMNT_INTEGRITY | NFSMNT_PRIVACY)) == (NFSMNT_KERB | NFSMNT_INTEGRITY), ",sec=krb5i", &buf, &blen); nfscl_printopt(nmp, (nmp->nm_flag & (NFSMNT_KERB | NFSMNT_INTEGRITY | NFSMNT_PRIVACY)) == (NFSMNT_KERB | NFSMNT_PRIVACY), ",sec=krb5p", &buf, &blen); nfscl_printoptval(nmp, nmp->nm_acdirmin, ",acdirmin", &buf, &blen); nfscl_printoptval(nmp, nmp->nm_acdirmax, ",acdirmax", &buf, &blen); nfscl_printoptval(nmp, nmp->nm_acregmin, ",acregmin", &buf, &blen); nfscl_printoptval(nmp, nmp->nm_acregmax, ",acregmax", &buf, &blen); nfscl_printoptval(nmp, nmp->nm_nametimeo, ",nametimeo", &buf, &blen); nfscl_printoptval(nmp, nmp->nm_negnametimeo, ",negnametimeo", &buf, &blen); nfscl_printoptval(nmp, nmp->nm_rsize, ",rsize", &buf, &blen); nfscl_printoptval(nmp, nmp->nm_wsize, ",wsize", &buf, &blen); nfscl_printoptval(nmp, nmp->nm_readdirsize, ",readdirsize", &buf, &blen); nfscl_printoptval(nmp, nmp->nm_readahead, ",readahead", &buf, &blen); nfscl_printoptval(nmp, nmp->nm_wcommitsize, ",wcommitsize", &buf, &blen); nfscl_printoptval(nmp, nmp->nm_timeo, ",timeout", &buf, &blen); nfscl_printoptval(nmp, nmp->nm_retry, ",retrans", &buf, &blen); } Index: head/sys/kern/vfs_bio.c =================================================================== --- head/sys/kern/vfs_bio.c (revision 281959) +++ head/sys/kern/vfs_bio.c (revision 281960) @@ -1,4635 +1,4637 @@ /*- * 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 "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, }; /* * XXX buf is global because kern_shutdown.c and ffs_checkoverlap has * carnal knowledge of buffers. This knowledge should be moved to vfs_bio.c. */ struct buf *buf; /* buffer header pool */ caddr_t unmapped_buf; /* Used below and for softdep flushing threads in ufs/ffs/ffs_softdep.c */ struct proc *bufdaemonproc; static int inmem(struct vnode *vp, daddr_t blkno); static void vm_hold_free_pages(struct buf *bp, int newbsize); static void vm_hold_load_pages(struct buf *bp, vm_offset_t from, vm_offset_t to); static void vfs_page_set_valid(struct buf *bp, vm_ooffset_t off, vm_page_t m); static void vfs_page_set_validclean(struct buf *bp, vm_ooffset_t off, vm_page_t m); static void vfs_clean_pages_dirty_buf(struct buf *bp); static void vfs_setdirty_locked_object(struct buf *bp); static void vfs_vmio_release(struct buf *bp); static int vfs_bio_clcheck(struct vnode *vp, int size, daddr_t lblkno, daddr_t blkno); static int buf_flush(int); static int flushbufqueues(int, int); static void buf_daemon(void); static void bremfreel(struct buf *bp); static __inline void bd_wakeup(void); static int sysctl_runningspace(SYSCTL_HANDLER_ARGS); #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) static int sysctl_bufspace(SYSCTL_HANDLER_ARGS); #endif int vmiodirenable = TRUE; SYSCTL_INT(_vfs, OID_AUTO, vmiodirenable, CTLFLAG_RW, &vmiodirenable, 0, "Use the VM system for directory writes"); long runningbufspace; SYSCTL_LONG(_vfs, OID_AUTO, runningbufspace, CTLFLAG_RD, &runningbufspace, 0, "Amount of presently outstanding async buffer io"); static long bufspace; #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) SYSCTL_PROC(_vfs, OID_AUTO, bufspace, CTLTYPE_LONG|CTLFLAG_MPSAFE|CTLFLAG_RD, &bufspace, 0, sysctl_bufspace, "L", "Virtual memory used for buffers"); #else SYSCTL_LONG(_vfs, OID_AUTO, bufspace, CTLFLAG_RD, &bufspace, 0, "Virtual memory used for buffers"); #endif static long unmapped_bufspace; SYSCTL_LONG(_vfs, OID_AUTO, unmapped_bufspace, CTLFLAG_RD, &unmapped_bufspace, 0, "Amount of unmapped buffers, inclusive in the bufspace"); static long maxbufspace; SYSCTL_LONG(_vfs, OID_AUTO, maxbufspace, CTLFLAG_RD, &maxbufspace, 0, "Maximum allowed value of bufspace (including buf_daemon)"); static long bufmallocspace; SYSCTL_LONG(_vfs, OID_AUTO, bufmallocspace, CTLFLAG_RD, &bufmallocspace, 0, "Amount of malloced memory for buffers"); static long maxbufmallocspace; SYSCTL_LONG(_vfs, OID_AUTO, maxmallocbufspace, CTLFLAG_RW, &maxbufmallocspace, 0, "Maximum amount of malloced memory for buffers"); static long lobufspace; SYSCTL_LONG(_vfs, OID_AUTO, lobufspace, CTLFLAG_RD, &lobufspace, 0, "Minimum amount of buffers we want to have"); long hibufspace; SYSCTL_LONG(_vfs, OID_AUTO, hibufspace, CTLFLAG_RD, &hibufspace, 0, "Maximum allowed value of bufspace (excluding buf_daemon)"); static int bufreusecnt; SYSCTL_INT(_vfs, OID_AUTO, bufreusecnt, CTLFLAG_RW, &bufreusecnt, 0, "Number of times we have reused a buffer"); static int buffreekvacnt; SYSCTL_INT(_vfs, OID_AUTO, buffreekvacnt, CTLFLAG_RW, &buffreekvacnt, 0, "Number of times we have freed the KVA space from some buffer"); static int bufdefragcnt; SYSCTL_INT(_vfs, OID_AUTO, bufdefragcnt, CTLFLAG_RW, &bufdefragcnt, 0, "Number of times we have had to repeat buffer allocation to defragment"); static long lorunningspace; SYSCTL_PROC(_vfs, OID_AUTO, lorunningspace, CTLTYPE_LONG | CTLFLAG_MPSAFE | CTLFLAG_RW, &lorunningspace, 0, sysctl_runningspace, "L", "Minimum preferred space used for in-progress I/O"); static long hirunningspace; SYSCTL_PROC(_vfs, OID_AUTO, hirunningspace, CTLTYPE_LONG | CTLFLAG_MPSAFE | CTLFLAG_RW, &hirunningspace, 0, sysctl_runningspace, "L", "Maximum amount of space to use for in-progress I/O"); int dirtybufferflushes; SYSCTL_INT(_vfs, OID_AUTO, dirtybufferflushes, CTLFLAG_RW, &dirtybufferflushes, 0, "Number of bdwrite to bawrite conversions to limit dirty buffers"); int bdwriteskip; SYSCTL_INT(_vfs, OID_AUTO, bdwriteskip, CTLFLAG_RW, &bdwriteskip, 0, "Number of buffers supplied to bdwrite with snapshot deadlock risk"); int altbufferflushes; SYSCTL_INT(_vfs, OID_AUTO, altbufferflushes, CTLFLAG_RW, &altbufferflushes, 0, "Number of fsync flushes to limit dirty buffers"); static int recursiveflushes; SYSCTL_INT(_vfs, OID_AUTO, recursiveflushes, CTLFLAG_RW, &recursiveflushes, 0, "Number of flushes skipped due to being recursive"); static int numdirtybuffers; SYSCTL_INT(_vfs, OID_AUTO, numdirtybuffers, CTLFLAG_RD, &numdirtybuffers, 0, "Number of buffers that are dirty (has unwritten changes) at the moment"); static int lodirtybuffers; SYSCTL_INT(_vfs, OID_AUTO, lodirtybuffers, CTLFLAG_RW, &lodirtybuffers, 0, "How many buffers we want to have free before bufdaemon can sleep"); static int hidirtybuffers; SYSCTL_INT(_vfs, OID_AUTO, hidirtybuffers, CTLFLAG_RW, &hidirtybuffers, 0, "When the number of dirty buffers is considered severe"); int dirtybufthresh; SYSCTL_INT(_vfs, OID_AUTO, dirtybufthresh, CTLFLAG_RW, &dirtybufthresh, 0, "Number of bdwrite to bawrite conversions to clear dirty buffers"); static int numfreebuffers; SYSCTL_INT(_vfs, OID_AUTO, numfreebuffers, CTLFLAG_RD, &numfreebuffers, 0, "Number of free buffers"); static int lofreebuffers; SYSCTL_INT(_vfs, OID_AUTO, lofreebuffers, CTLFLAG_RW, &lofreebuffers, 0, "XXX Unused"); static int hifreebuffers; SYSCTL_INT(_vfs, OID_AUTO, hifreebuffers, CTLFLAG_RW, &hifreebuffers, 0, "XXX Complicatedly unused"); static int getnewbufcalls; SYSCTL_INT(_vfs, OID_AUTO, getnewbufcalls, CTLFLAG_RW, &getnewbufcalls, 0, "Number of calls to getnewbuf"); static int getnewbufrestarts; SYSCTL_INT(_vfs, OID_AUTO, getnewbufrestarts, CTLFLAG_RW, &getnewbufrestarts, 0, "Number of times getnewbuf has had to restart a buffer aquisition"); static int mappingrestarts; SYSCTL_INT(_vfs, OID_AUTO, mappingrestarts, CTLFLAG_RW, &mappingrestarts, 0, "Number of times getblk has had to restart a buffer mapping for " "unmapped buffer"); static int flushbufqtarget = 100; SYSCTL_INT(_vfs, OID_AUTO, flushbufqtarget, CTLFLAG_RW, &flushbufqtarget, 0, "Amount of work to do in flushbufqueues when helping bufdaemon"); static long notbufdflushes; SYSCTL_LONG(_vfs, OID_AUTO, notbufdflushes, CTLFLAG_RD, ¬bufdflushes, 0, "Number of dirty buffer flushes done by the bufdaemon helpers"); static long barrierwrites; SYSCTL_LONG(_vfs, OID_AUTO, barrierwrites, CTLFLAG_RW, &barrierwrites, 0, "Number of barrier writes"); SYSCTL_INT(_vfs, OID_AUTO, unmapped_buf_allowed, CTLFLAG_RD, &unmapped_buf_allowed, 0, "Permit the use of the unmapped i/o"); /* * Lock for the non-dirty bufqueues */ static struct mtx_padalign bqclean; /* * Lock for the dirty queue. */ static struct mtx_padalign bqdirty; /* * This lock synchronizes access to bd_request. */ static struct mtx_padalign bdlock; /* * This lock protects the runningbufreq and synchronizes runningbufwakeup and * waitrunningbufspace(). */ static struct mtx_padalign rbreqlock; /* * Lock that protects needsbuffer and the sleeps/wakeups surrounding it. */ static struct rwlock_padalign nblock; /* * Lock that protects bdirtywait. */ static struct mtx_padalign bdirtylock; /* * Wakeup point for bufdaemon, as well as indicator of whether it is already * active. Set to 1 when the bufdaemon is already "on" the queue, 0 when it * is idling. */ static int bd_request; /* * Request for the buf daemon to write more buffers than is indicated by * lodirtybuf. This may be necessary to push out excess dependencies or * defragment the address space where a simple count of the number of dirty * buffers is insufficient to characterize the demand for flushing them. */ static int bd_speedupreq; /* * bogus page -- for I/O to/from partially complete buffers * this is a temporary solution to the problem, but it is not * really that bad. it would be better to split the buffer * for input in the case of buffers partially already in memory, * but the code is intricate enough already. */ vm_page_t bogus_page; /* * Synchronization (sleep/wakeup) variable for active buffer space requests. * Set when wait starts, cleared prior to wakeup(). * Used in runningbufwakeup() and waitrunningbufspace(). */ static int runningbufreq; /* * Synchronization (sleep/wakeup) variable for buffer requests. * Can contain the VFS_BIO_NEED flags defined below; setting/clearing is done * by and/or. * Used in numdirtywakeup(), bufspacewakeup(), bufcountadd(), bwillwrite(), * getnewbuf(), and getblk(). */ static volatile int needsbuffer; /* * Synchronization for bwillwrite() waiters. */ static int bdirtywait; /* * Definitions for the buffer free lists. */ #define BUFFER_QUEUES 5 /* number of free buffer queues */ #define QUEUE_NONE 0 /* on no queue */ #define QUEUE_CLEAN 1 /* non-B_DELWRI buffers */ #define QUEUE_DIRTY 2 /* B_DELWRI buffers */ #define QUEUE_EMPTYKVA 3 /* empty buffer headers w/KVA assignment */ #define QUEUE_EMPTY 4 /* empty buffer headers */ #define QUEUE_SENTINEL 1024 /* not an queue index, but mark for sentinel */ /* Queues for free buffers with various properties */ static TAILQ_HEAD(bqueues, buf) bufqueues[BUFFER_QUEUES] = { { 0 } }; #ifdef INVARIANTS static int bq_len[BUFFER_QUEUES]; #endif /* * Single global constant for BUF_WMESG, to avoid getting multiple references. * buf_wmesg is referred from macros. */ const char *buf_wmesg = BUF_WMESG; #define VFS_BIO_NEED_ANY 0x01 /* any freeable buffer */ #define VFS_BIO_NEED_FREE 0x04 /* wait for free bufs, hi hysteresis */ #define VFS_BIO_NEED_BUFSPACE 0x08 /* wait for buf space, lo hysteresis */ static int sysctl_runningspace(SYSCTL_HANDLER_ARGS) { long value; int error; value = *(long *)arg1; error = sysctl_handle_long(oidp, &value, 0, req); if (error != 0 || req->newptr == NULL) return (error); mtx_lock(&rbreqlock); if (arg1 == &hirunningspace) { if (value < lorunningspace) error = EINVAL; else hirunningspace = value; } else { KASSERT(arg1 == &lorunningspace, ("%s: unknown arg1", __func__)); if (value > hirunningspace) error = EINVAL; else lorunningspace = value; } mtx_unlock(&rbreqlock); return (error); } #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) static int sysctl_bufspace(SYSCTL_HANDLER_ARGS) { long lvalue; int ivalue; if (sizeof(int) == sizeof(long) || req->oldlen >= sizeof(long)) return (sysctl_handle_long(oidp, arg1, arg2, req)); lvalue = *(long *)arg1; if (lvalue > INT_MAX) /* On overflow, still write out a long to trigger ENOMEM. */ return (sysctl_handle_long(oidp, &lvalue, 0, req)); ivalue = lvalue; return (sysctl_handle_int(oidp, &ivalue, 0, req)); } #endif /* * bqlock: * * Return the appropriate queue lock based on the index. */ static inline struct mtx * bqlock(int qindex) { if (qindex == QUEUE_DIRTY) return (struct mtx *)(&bqdirty); return (struct mtx *)(&bqclean); } /* * bdirtywakeup: * * Wakeup any bwillwrite() waiters. */ static void bdirtywakeup(void) { mtx_lock(&bdirtylock); if (bdirtywait) { bdirtywait = 0; wakeup(&bdirtywait); } mtx_unlock(&bdirtylock); } /* * bdirtysub: * * Decrement the numdirtybuffers count by one and wakeup any * threads blocked in bwillwrite(). */ static void bdirtysub(void) { if (atomic_fetchadd_int(&numdirtybuffers, -1) == (lodirtybuffers + hidirtybuffers) / 2) bdirtywakeup(); } /* * bdirtyadd: * * Increment the numdirtybuffers count by one and wakeup the buf * daemon if needed. */ static void bdirtyadd(void) { /* * Only do the wakeup once as we cross the boundary. The * buf daemon will keep running until the condition clears. */ if (atomic_fetchadd_int(&numdirtybuffers, 1) == (lodirtybuffers + hidirtybuffers) / 2) bd_wakeup(); } /* * bufspacewakeup: * * Called when buffer space is potentially available for recovery. * getnewbuf() will block on this flag when it is unable to free * sufficient buffer space. Buffer space becomes recoverable when * bp's get placed back in the queues. */ static __inline void bufspacewakeup(void) { int need_wakeup, on; /* * If someone is waiting for BUF space, wake them up. Even * though we haven't freed the kva space yet, the waiting * process will be able to now. */ rw_rlock(&nblock); for (;;) { need_wakeup = 0; on = needsbuffer; if ((on & VFS_BIO_NEED_BUFSPACE) == 0) break; need_wakeup = 1; if (atomic_cmpset_rel_int(&needsbuffer, on, on & ~VFS_BIO_NEED_BUFSPACE)) break; } if (need_wakeup) wakeup(__DEVOLATILE(void *, &needsbuffer)); rw_runlock(&nblock); } /* * runningwakeup: * * Wake up processes that are waiting on asynchronous writes to fall * below lorunningspace. */ static void runningwakeup(void) { mtx_lock(&rbreqlock); if (runningbufreq) { runningbufreq = 0; wakeup(&runningbufreq); } mtx_unlock(&rbreqlock); } /* * runningbufwakeup: * * Decrement the outstanding write count according. */ void runningbufwakeup(struct buf *bp) { long space, bspace; bspace = bp->b_runningbufspace; if (bspace == 0) return; space = atomic_fetchadd_long(&runningbufspace, -bspace); KASSERT(space >= bspace, ("runningbufspace underflow %ld %ld", space, bspace)); bp->b_runningbufspace = 0; /* * Only acquire the lock and wakeup on the transition from exceeding * the threshold to falling below it. */ if (space < lorunningspace) return; if (space - bspace > lorunningspace) return; runningwakeup(); } /* * bufcountadd: * * Called when a buffer has been added to one of the free queues to * account for the buffer and to wakeup anyone waiting for free buffers. * This typically occurs when large amounts of metadata are being handled * by the buffer cache ( else buffer space runs out first, usually ). */ static __inline void bufcountadd(struct buf *bp) { int mask, need_wakeup, old, on; KASSERT((bp->b_flags & B_INFREECNT) == 0, ("buf %p already counted as free", bp)); bp->b_flags |= B_INFREECNT; old = atomic_fetchadd_int(&numfreebuffers, 1); KASSERT(old >= 0 && old < nbuf, ("numfreebuffers climbed to %d", old + 1)); mask = VFS_BIO_NEED_ANY; if (numfreebuffers >= hifreebuffers) mask |= VFS_BIO_NEED_FREE; rw_rlock(&nblock); for (;;) { need_wakeup = 0; on = needsbuffer; if (on == 0) break; need_wakeup = 1; if (atomic_cmpset_rel_int(&needsbuffer, on, on & ~mask)) break; } if (need_wakeup) wakeup(__DEVOLATILE(void *, &needsbuffer)); rw_runlock(&nblock); } /* * bufcountsub: * * Decrement the numfreebuffers count as needed. */ static void bufcountsub(struct buf *bp) { int old; /* * Fixup numfreebuffers count. If the buffer is invalid or not * delayed-write, the buffer was free and we must decrement * numfreebuffers. */ if ((bp->b_flags & B_INVAL) || (bp->b_flags & B_DELWRI) == 0) { KASSERT((bp->b_flags & B_INFREECNT) != 0, ("buf %p not counted in numfreebuffers", bp)); bp->b_flags &= ~B_INFREECNT; old = atomic_fetchadd_int(&numfreebuffers, -1); KASSERT(old > 0, ("numfreebuffers dropped to %d", old - 1)); } } /* * waitrunningbufspace() * * runningbufspace is a measure of the amount of I/O currently * running. This routine is used in async-write situations to * prevent creating huge backups of pending writes to a device. * Only asynchronous writes are governed by this function. * * This does NOT turn an async write into a sync write. It waits * for earlier writes to complete and generally returns before the * caller's write has reached the device. */ void waitrunningbufspace(void) { mtx_lock(&rbreqlock); while (runningbufspace > hirunningspace) { runningbufreq = 1; msleep(&runningbufreq, &rbreqlock, PVM, "wdrain", 0); } mtx_unlock(&rbreqlock); } /* * vfs_buf_test_cache: * * Called when a buffer is extended. This function clears the B_CACHE * bit if the newly extended portion of the buffer does not contain * valid data. */ static __inline void vfs_buf_test_cache(struct buf *bp, vm_ooffset_t foff, vm_offset_t off, vm_offset_t size, vm_page_t m) { VM_OBJECT_ASSERT_LOCKED(m->object); if (bp->b_flags & B_CACHE) { int base = (foff + off) & PAGE_MASK; if (vm_page_is_valid(m, base, size) == 0) bp->b_flags &= ~B_CACHE; } } /* Wake up the buffer daemon if necessary */ static __inline void bd_wakeup(void) { mtx_lock(&bdlock); if (bd_request == 0) { bd_request = 1; wakeup(&bd_request); } mtx_unlock(&bdlock); } /* * bd_speedup - speedup the buffer cache flushing code */ void bd_speedup(void) { int needwake; mtx_lock(&bdlock); needwake = 0; if (bd_speedupreq == 0 || bd_request == 0) needwake = 1; bd_speedupreq = 1; bd_request = 1; if (needwake) wakeup(&bd_request); mtx_unlock(&bdlock); } #ifndef NSWBUF_MIN #define NSWBUF_MIN 16 #endif #ifdef __i386__ #define TRANSIENT_DENOM 5 #else #define TRANSIENT_DENOM 10 #endif /* * Calculating buffer cache scaling values and reserve space for buffer * headers. This is called during low level kernel initialization and * may be called more then once. We CANNOT write to the memory area * being reserved at this time. */ caddr_t kern_vfs_bio_buffer_alloc(caddr_t v, long physmem_est) { int tuned_nbuf; long maxbuf, maxbuf_sz, buf_sz, biotmap_sz; /* * physmem_est is in pages. Convert it to kilobytes (assumes * PAGE_SIZE is >= 1K) */ physmem_est = physmem_est * (PAGE_SIZE / 1024); /* * The nominal buffer size (and minimum KVA allocation) is BKVASIZE. * For the first 64MB of ram nominally allocate sufficient buffers to * cover 1/4 of our ram. Beyond the first 64MB allocate additional * buffers to cover 1/10 of our ram over 64MB. When auto-sizing * the buffer cache we limit the eventual kva reservation to * maxbcache bytes. * * factor represents the 1/4 x ram conversion. */ if (nbuf == 0) { int factor = 4 * BKVASIZE / 1024; nbuf = 50; if (physmem_est > 4096) nbuf += min((physmem_est - 4096) / factor, 65536 / factor); if (physmem_est > 65536) nbuf += min((physmem_est - 65536) * 2 / (factor * 5), 32 * 1024 * 1024 / (factor * 5)); if (maxbcache && nbuf > maxbcache / BKVASIZE) nbuf = maxbcache / BKVASIZE; tuned_nbuf = 1; } else tuned_nbuf = 0; /* XXX Avoid unsigned long overflows later on with maxbufspace. */ maxbuf = (LONG_MAX / 3) / BKVASIZE; if (nbuf > maxbuf) { if (!tuned_nbuf) printf("Warning: nbufs lowered from %d to %ld\n", nbuf, maxbuf); nbuf = maxbuf; } /* * Ideal allocation size for the transient bio submap is 10% * of the maximal space buffer map. This roughly corresponds * to the amount of the buffer mapped for typical UFS load. * * Clip the buffer map to reserve space for the transient * BIOs, if its extent is bigger than 90% (80% on i386) of the * maximum buffer map extent on the platform. * * The fall-back to the maxbuf in case of maxbcache unset, * allows to not trim the buffer KVA for the architectures * with ample KVA space. */ if (bio_transient_maxcnt == 0 && unmapped_buf_allowed) { maxbuf_sz = maxbcache != 0 ? maxbcache : maxbuf * BKVASIZE; buf_sz = (long)nbuf * BKVASIZE; if (buf_sz < maxbuf_sz / TRANSIENT_DENOM * (TRANSIENT_DENOM - 1)) { /* * There is more KVA than memory. Do not * adjust buffer map size, and assign the rest * of maxbuf to transient map. */ biotmap_sz = maxbuf_sz - buf_sz; } else { /* * Buffer map spans all KVA we could afford on * this platform. Give 10% (20% on i386) of * the buffer map to the transient bio map. */ biotmap_sz = buf_sz / TRANSIENT_DENOM; buf_sz -= biotmap_sz; } if (biotmap_sz / INT_MAX > MAXPHYS) bio_transient_maxcnt = INT_MAX; else bio_transient_maxcnt = biotmap_sz / MAXPHYS; /* * Artifically limit to 1024 simultaneous in-flight I/Os * using the transient mapping. */ if (bio_transient_maxcnt > 1024) bio_transient_maxcnt = 1024; if (tuned_nbuf) nbuf = buf_sz / BKVASIZE; } /* * swbufs are used as temporary holders for I/O, such as paging I/O. * We have no less then 16 and no more then 256. */ nswbuf = min(nbuf / 4, 256); TUNABLE_INT_FETCH("kern.nswbuf", &nswbuf); if (nswbuf < NSWBUF_MIN) nswbuf = NSWBUF_MIN; /* * Reserve space for the buffer cache buffers */ swbuf = (void *)v; v = (caddr_t)(swbuf + nswbuf); buf = (void *)v; v = (caddr_t)(buf + nbuf); return(v); } /* Initialize the buffer subsystem. Called before use of any buffers. */ void bufinit(void) { struct buf *bp; int i; + CTASSERT(MAXBCACHEBUF >= MAXBSIZE); mtx_init(&bqclean, "bufq clean lock", NULL, MTX_DEF); mtx_init(&bqdirty, "bufq dirty lock", NULL, MTX_DEF); mtx_init(&rbreqlock, "runningbufspace lock", NULL, MTX_DEF); rw_init(&nblock, "needsbuffer lock"); mtx_init(&bdlock, "buffer daemon lock", NULL, MTX_DEF); mtx_init(&bdirtylock, "dirty buf lock", NULL, MTX_DEF); /* next, make a null set of free lists */ for (i = 0; i < BUFFER_QUEUES; i++) TAILQ_INIT(&bufqueues[i]); /* finally, initialize each buffer header and stick on empty q */ for (i = 0; i < nbuf; i++) { bp = &buf[i]; bzero(bp, sizeof *bp); bp->b_flags = B_INVAL | B_INFREECNT; bp->b_rcred = NOCRED; bp->b_wcred = NOCRED; bp->b_qindex = QUEUE_EMPTY; bp->b_xflags = 0; LIST_INIT(&bp->b_dep); BUF_LOCKINIT(bp); TAILQ_INSERT_TAIL(&bufqueues[QUEUE_EMPTY], bp, b_freelist); #ifdef INVARIANTS bq_len[QUEUE_EMPTY]++; #endif } /* * maxbufspace is the absolute maximum amount of buffer space we are * allowed to reserve in KVM and in real terms. The absolute maximum * is nominally used by buf_daemon. hibufspace is the nominal maximum * used by most other processes. The differential is required to * ensure that buf_daemon is able to run when other processes might * be blocked waiting for buffer space. * * maxbufspace is based on BKVASIZE. Allocating buffers larger then * this may result in KVM fragmentation which is not handled optimally * by the system. */ maxbufspace = (long)nbuf * BKVASIZE; - hibufspace = lmax(3 * maxbufspace / 4, maxbufspace - MAXBSIZE * 10); - lobufspace = hibufspace - MAXBSIZE; + hibufspace = lmax(3 * maxbufspace / 4, maxbufspace - MAXBCACHEBUF * 10); + lobufspace = hibufspace - MAXBCACHEBUF; /* * Note: The 16 MiB upper limit for hirunningspace was chosen * arbitrarily and may need further tuning. It corresponds to * 128 outstanding write IO requests (if IO size is 128 KiB), * which fits with many RAID controllers' tagged queuing limits. * The lower 1 MiB limit is the historical upper limit for * hirunningspace. */ - hirunningspace = lmax(lmin(roundup(hibufspace / 64, MAXBSIZE), + hirunningspace = lmax(lmin(roundup(hibufspace / 64, MAXBCACHEBUF), 16 * 1024 * 1024), 1024 * 1024); - lorunningspace = roundup((hirunningspace * 2) / 3, MAXBSIZE); + lorunningspace = roundup((hirunningspace * 2) / 3, MAXBCACHEBUF); /* * Limit the amount of malloc memory since it is wired permanently into * the kernel space. Even though this is accounted for in the buffer * allocation, we don't want the malloced region to grow uncontrolled. * The malloc scheme improves memory utilization significantly on average * (small) directories. */ maxbufmallocspace = hibufspace / 20; /* * Reduce the chance of a deadlock occuring by limiting the number * of delayed-write dirty buffers we allow to stack up. */ hidirtybuffers = nbuf / 4 + 20; dirtybufthresh = hidirtybuffers * 9 / 10; numdirtybuffers = 0; /* * To support extreme low-memory systems, make sure hidirtybuffers cannot * eat up all available buffer space. This occurs when our minimum cannot * be met. We try to size hidirtybuffers to 3/4 our buffer space assuming * BKVASIZE'd buffers. */ while ((long)hidirtybuffers * BKVASIZE > 3 * hibufspace / 4) { hidirtybuffers >>= 1; } lodirtybuffers = hidirtybuffers / 2; /* * Try to keep the number of free buffers in the specified range, * and give special processes (e.g. like buf_daemon) access to an * emergency reserve. */ lofreebuffers = nbuf / 18 + 5; hifreebuffers = 2 * lofreebuffers; numfreebuffers = nbuf; bogus_page = vm_page_alloc(NULL, 0, VM_ALLOC_NOOBJ | VM_ALLOC_NORMAL | VM_ALLOC_WIRED); unmapped_buf = (caddr_t)kva_alloc(MAXPHYS); } #ifdef INVARIANTS static inline void vfs_buf_check_mapped(struct buf *bp) { KASSERT((bp->b_flags & B_UNMAPPED) == 0, ("mapped buf %p %x", bp, bp->b_flags)); 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)); } static inline void vfs_buf_check_unmapped(struct buf *bp) { KASSERT((bp->b_flags & B_UNMAPPED) == B_UNMAPPED, ("unmapped buf %p %x", bp, bp->b_flags)); KASSERT(bp->b_kvabase == unmapped_buf, ("unmapped buf: corrupted b_kvabase %p", 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 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)); } /* * bfreekva() - free the kva allocation for a buffer. * * Since this call frees up buffer space, we call bufspacewakeup(). */ static void bfreekva(struct buf *bp) { if (bp->b_kvasize == 0) return; atomic_add_int(&buffreekvacnt, 1); atomic_subtract_long(&bufspace, bp->b_kvasize); if ((bp->b_flags & B_UNMAPPED) == 0) { BUF_CHECK_MAPPED(bp); vmem_free(buffer_arena, (vm_offset_t)bp->b_kvabase, bp->b_kvasize); } else { BUF_CHECK_UNMAPPED(bp); if ((bp->b_flags & B_KVAALLOC) != 0) { vmem_free(buffer_arena, (vm_offset_t)bp->b_kvaalloc, bp->b_kvasize); } atomic_subtract_long(&unmapped_bufspace, bp->b_kvasize); bp->b_flags &= ~(B_UNMAPPED | B_KVAALLOC); } bp->b_kvasize = 0; bufspacewakeup(); } /* * binsfree: * * Insert the buffer into the appropriate free list. */ static void binsfree(struct buf *bp, int qindex) { struct mtx *olock, *nlock; BUF_ASSERT_XLOCKED(bp); olock = bqlock(bp->b_qindex); nlock = bqlock(qindex); mtx_lock(olock); /* Handle delayed bremfree() processing. */ if (bp->b_flags & B_REMFREE) bremfreel(bp); if (bp->b_qindex != QUEUE_NONE) panic("binsfree: free buffer onto another queue???"); bp->b_qindex = qindex; if (olock != nlock) { mtx_unlock(olock); mtx_lock(nlock); } if (bp->b_flags & B_AGE) TAILQ_INSERT_HEAD(&bufqueues[bp->b_qindex], bp, b_freelist); else TAILQ_INSERT_TAIL(&bufqueues[bp->b_qindex], bp, b_freelist); #ifdef INVARIANTS bq_len[bp->b_qindex]++; #endif mtx_unlock(nlock); /* * Something we can maybe free or reuse. */ if (bp->b_bufsize && !(bp->b_flags & B_DELWRI)) bufspacewakeup(); if ((bp->b_flags & B_INVAL) || !(bp->b_flags & B_DELWRI)) bufcountadd(bp); } /* * bremfree: * * Mark the buffer for removal from the appropriate free list. * */ void bremfree(struct buf *bp) { CTR3(KTR_BUF, "bremfree(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT((bp->b_flags & B_REMFREE) == 0, ("bremfree: buffer %p already marked for delayed removal.", bp)); KASSERT(bp->b_qindex != QUEUE_NONE, ("bremfree: buffer %p not on a queue.", bp)); BUF_ASSERT_XLOCKED(bp); bp->b_flags |= B_REMFREE; bufcountsub(bp); } /* * bremfreef: * * Force an immediate removal from a free list. Used only in nfs when * it abuses the b_freelist pointer. */ void bremfreef(struct buf *bp) { struct mtx *qlock; qlock = bqlock(bp->b_qindex); mtx_lock(qlock); bremfreel(bp); mtx_unlock(qlock); } /* * bremfreel: * * Removes a buffer from the free list, must be called with the * correct qlock held. */ static void bremfreel(struct buf *bp) { CTR3(KTR_BUF, "bremfreel(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(bp->b_qindex != QUEUE_NONE, ("bremfreel: buffer %p not on a queue.", bp)); BUF_ASSERT_XLOCKED(bp); mtx_assert(bqlock(bp->b_qindex), MA_OWNED); TAILQ_REMOVE(&bufqueues[bp->b_qindex], bp, b_freelist); #ifdef INVARIANTS KASSERT(bq_len[bp->b_qindex] >= 1, ("queue %d underflow", bp->b_qindex)); bq_len[bp->b_qindex]--; #endif bp->b_qindex = QUEUE_NONE; /* * If this was a delayed bremfree() we only need to remove the buffer * from the queue and return the stats are already done. */ if (bp->b_flags & B_REMFREE) { bp->b_flags &= ~B_REMFREE; return; } bufcountsub(bp); } /* * Attempt to initiate asynchronous I/O on read-ahead blocks. We must * clear BIO_ERROR and B_INVAL prior to initiating I/O . If B_CACHE is set, * the buffer is valid and we do not have to do anything. */ void breada(struct vnode * vp, daddr_t * rablkno, int * rabsize, int cnt, struct ucred * cred) { struct buf *rabp; int i; for (i = 0; i < cnt; i++, rablkno++, rabsize++) { if (inmem(vp, *rablkno)) continue; rabp = getblk(vp, *rablkno, *rabsize, 0, 0, 0); if ((rabp->b_flags & B_CACHE) == 0) { if (!TD_IS_IDLETHREAD(curthread)) curthread->td_ru.ru_inblock++; rabp->b_flags |= B_ASYNC; rabp->b_flags &= ~B_INVAL; rabp->b_ioflags &= ~BIO_ERROR; rabp->b_iocmd = BIO_READ; if (rabp->b_rcred == NOCRED && cred != NOCRED) rabp->b_rcred = crhold(cred); vfs_busy_pages(rabp, 0); BUF_KERNPROC(rabp); rabp->b_iooffset = dbtob(rabp->b_blkno); bstrategy(rabp); } else { brelse(rabp); } } } /* * Entry point for bread() and breadn() via #defines in sys/buf.h. * * Get a buffer with the specified data. Look in the cache first. We * must clear BIO_ERROR and B_INVAL prior to initiating I/O. If B_CACHE * is set, the buffer is valid and we do not have to do anything, see * getblk(). Also starts asynchronous I/O on read-ahead blocks. */ int breadn_flags(struct vnode *vp, daddr_t blkno, int size, daddr_t *rablkno, int *rabsize, int cnt, struct ucred *cred, int flags, struct buf **bpp) { struct buf *bp; int rv = 0, readwait = 0; CTR3(KTR_BUF, "breadn(%p, %jd, %d)", vp, blkno, size); /* * Can only return NULL if GB_LOCK_NOWAIT flag is specified. */ *bpp = bp = getblk(vp, blkno, size, 0, 0, flags); if (bp == NULL) return (EBUSY); /* if not found in cache, do some I/O */ if ((bp->b_flags & B_CACHE) == 0) { if (!TD_IS_IDLETHREAD(curthread)) curthread->td_ru.ru_inblock++; bp->b_iocmd = BIO_READ; bp->b_flags &= ~B_INVAL; bp->b_ioflags &= ~BIO_ERROR; if (bp->b_rcred == NOCRED && cred != NOCRED) bp->b_rcred = crhold(cred); vfs_busy_pages(bp, 0); bp->b_iooffset = dbtob(bp->b_blkno); bstrategy(bp); ++readwait; } breada(vp, rablkno, rabsize, cnt, cred); if (readwait) { rv = bufwait(bp); } return (rv); } /* * Write, release buffer on completion. (Done by iodone * if async). Do not bother writing anything if the buffer * is invalid. * * Note that we set B_CACHE here, indicating that buffer is * fully valid and thus cacheable. This is true even of NFS * now so we set it generally. This could be set either here * or in biodone() since the I/O is synchronous. We put it * here. */ int bufwrite(struct buf *bp) { int oldflags; struct vnode *vp; long space; int vp_md; CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); if (bp->b_flags & B_INVAL) { brelse(bp); return (0); } if (bp->b_flags & B_BARRIER) barrierwrites++; oldflags = bp->b_flags; BUF_ASSERT_HELD(bp); if (bp->b_pin_count > 0) bunpin_wait(bp); KASSERT(!(bp->b_vflags & BV_BKGRDINPROG), ("FFS background buffer should not get here %p", bp)); vp = bp->b_vp; if (vp) vp_md = vp->v_vflag & VV_MD; else vp_md = 0; /* * Mark the buffer clean. Increment the bufobj write count * before bundirty() call, to prevent other thread from seeing * empty dirty list and zero counter for writes in progress, * falsely indicating that the bufobj is clean. */ bufobj_wref(bp->b_bufobj); bundirty(bp); bp->b_flags &= ~B_DONE; bp->b_ioflags &= ~BIO_ERROR; bp->b_flags |= B_CACHE; bp->b_iocmd = BIO_WRITE; vfs_busy_pages(bp, 1); /* * Normal bwrites pipeline writes */ bp->b_runningbufspace = bp->b_bufsize; space = atomic_fetchadd_long(&runningbufspace, bp->b_runningbufspace); if (!TD_IS_IDLETHREAD(curthread)) curthread->td_ru.ru_oublock++; if (oldflags & B_ASYNC) BUF_KERNPROC(bp); bp->b_iooffset = dbtob(bp->b_blkno); bstrategy(bp); if ((oldflags & B_ASYNC) == 0) { int rtval = bufwait(bp); brelse(bp); return (rtval); } else if (space > hirunningspace) { /* * don't allow the async write to saturate the I/O * system. We will not deadlock here because * we are blocking waiting for I/O that is already in-progress * to complete. We do not block here if it is the update * or syncer daemon trying to clean up as that can lead * to deadlock. */ if ((curthread->td_pflags & TDP_NORUNNINGBUF) == 0 && !vp_md) waitrunningbufspace(); } return (0); } void bufbdflush(struct bufobj *bo, struct buf *bp) { struct buf *nbp; if (bo->bo_dirty.bv_cnt > dirtybufthresh + 10) { (void) VOP_FSYNC(bp->b_vp, MNT_NOWAIT, curthread); altbufferflushes++; } else if (bo->bo_dirty.bv_cnt > dirtybufthresh) { BO_LOCK(bo); /* * Try to find a buffer to flush. */ TAILQ_FOREACH(nbp, &bo->bo_dirty.bv_hd, b_bobufs) { if ((nbp->b_vflags & BV_BKGRDINPROG) || BUF_LOCK(nbp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) continue; if (bp == nbp) panic("bdwrite: found ourselves"); BO_UNLOCK(bo); /* Don't countdeps with the bo lock held. */ if (buf_countdeps(nbp, 0)) { BO_LOCK(bo); BUF_UNLOCK(nbp); continue; } if (nbp->b_flags & B_CLUSTEROK) { vfs_bio_awrite(nbp); } else { bremfree(nbp); bawrite(nbp); } dirtybufferflushes++; break; } if (nbp == NULL) BO_UNLOCK(bo); } } /* * Delayed write. (Buffer is marked dirty). Do not bother writing * anything if the buffer is marked invalid. * * Note that since the buffer must be completely valid, we can safely * set B_CACHE. In fact, we have to set B_CACHE here rather then in * biodone() in order to prevent getblk from writing the buffer * out synchronously. */ void bdwrite(struct buf *bp) { struct thread *td = curthread; struct vnode *vp; struct bufobj *bo; CTR3(KTR_BUF, "bdwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp)); KASSERT((bp->b_flags & B_BARRIER) == 0, ("Barrier request in delayed write %p", bp)); BUF_ASSERT_HELD(bp); if (bp->b_flags & B_INVAL) { brelse(bp); return; } /* * If we have too many dirty buffers, don't create any more. * If we are wildly over our limit, then force a complete * cleanup. Otherwise, just keep the situation from getting * out of control. Note that we have to avoid a recursive * disaster and not try to clean up after our own cleanup! */ vp = bp->b_vp; bo = bp->b_bufobj; if ((td->td_pflags & (TDP_COWINPROGRESS|TDP_INBDFLUSH)) == 0) { td->td_pflags |= TDP_INBDFLUSH; BO_BDFLUSH(bo, bp); td->td_pflags &= ~TDP_INBDFLUSH; } else recursiveflushes++; bdirty(bp); /* * Set B_CACHE, indicating that the buffer is fully valid. This is * true even of NFS now. */ bp->b_flags |= B_CACHE; /* * This bmap keeps the system from needing to do the bmap later, * perhaps when the system is attempting to do a sync. Since it * is likely that the indirect block -- or whatever other datastructure * that the filesystem needs is still in memory now, it is a good * thing to do this. Note also, that if the pageout daemon is * requesting a sync -- there might not be enough memory to do * the bmap then... So, this is important to do. */ if (vp->v_type != VCHR && bp->b_lblkno == bp->b_blkno) { VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, NULL, NULL); } /* * Set the *dirty* buffer range based upon the VM system dirty * pages. * * Mark the buffer pages as clean. We need to do this here to * satisfy the vnode_pager and the pageout daemon, so that it * thinks that the pages have been "cleaned". Note that since * the pages are in a delayed write buffer -- the VFS layer * "will" see that the pages get written out on the next sync, * or perhaps the cluster will be completed. */ vfs_clean_pages_dirty_buf(bp); bqrelse(bp); /* * note: we cannot initiate I/O from a bdwrite even if we wanted to, * due to the softdep code. */ } /* * bdirty: * * Turn buffer into delayed write request. We must clear BIO_READ and * B_RELBUF, and we must set B_DELWRI. We reassign the buffer to * itself to properly update it in the dirty/clean lists. We mark it * B_DONE to ensure that any asynchronization of the buffer properly * clears B_DONE ( else a panic will occur later ). * * bdirty() is kinda like bdwrite() - we have to clear B_INVAL which * might have been set pre-getblk(). Unlike bwrite/bdwrite, bdirty() * should only be called if the buffer is known-good. * * Since the buffer is not on a queue, we do not update the numfreebuffers * count. * * The buffer must be on QUEUE_NONE. */ void bdirty(struct buf *bp) { CTR3(KTR_BUF, "bdirty(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp)); KASSERT(bp->b_flags & B_REMFREE || bp->b_qindex == QUEUE_NONE, ("bdirty: buffer %p still on queue %d", bp, bp->b_qindex)); BUF_ASSERT_HELD(bp); bp->b_flags &= ~(B_RELBUF); bp->b_iocmd = BIO_WRITE; if ((bp->b_flags & B_DELWRI) == 0) { bp->b_flags |= /* XXX B_DONE | */ B_DELWRI; reassignbuf(bp); bdirtyadd(); } } /* * bundirty: * * Clear B_DELWRI for buffer. * * Since the buffer is not on a queue, we do not update the numfreebuffers * count. * * The buffer must be on QUEUE_NONE. */ void bundirty(struct buf *bp) { CTR3(KTR_BUF, "bundirty(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp)); KASSERT(bp->b_flags & B_REMFREE || bp->b_qindex == QUEUE_NONE, ("bundirty: buffer %p still on queue %d", bp, bp->b_qindex)); BUF_ASSERT_HELD(bp); if (bp->b_flags & B_DELWRI) { bp->b_flags &= ~B_DELWRI; reassignbuf(bp); bdirtysub(); } /* * Since it is now being written, we can clear its deferred write flag. */ bp->b_flags &= ~B_DEFERRED; } /* * bawrite: * * Asynchronous write. Start output on a buffer, but do not wait for * it to complete. The buffer is released when the output completes. * * bwrite() ( or the VOP routine anyway ) is responsible for handling * B_INVAL buffers. Not us. */ void bawrite(struct buf *bp) { bp->b_flags |= B_ASYNC; (void) bwrite(bp); } /* * babarrierwrite: * * Asynchronous barrier write. Start output on a buffer, but do not * wait for it to complete. Place a write barrier after this write so * that this buffer and all buffers written before it are committed to * the disk before any buffers written after this write are committed * to the disk. The buffer is released when the output completes. */ void babarrierwrite(struct buf *bp) { bp->b_flags |= B_ASYNC | B_BARRIER; (void) bwrite(bp); } /* * bbarrierwrite: * * Synchronous barrier write. Start output on a buffer and wait for * it to complete. Place a write barrier after this write so that * this buffer and all buffers written before it are committed to * the disk before any buffers written after this write are committed * to the disk. The buffer is released when the output completes. */ int bbarrierwrite(struct buf *bp) { bp->b_flags |= B_BARRIER; return (bwrite(bp)); } /* * bwillwrite: * * Called prior to the locking of any vnodes when we are expecting to * write. We do not want to starve the buffer cache with too many * dirty buffers so we block here. By blocking prior to the locking * of any vnodes we attempt to avoid the situation where a locked vnode * prevents the various system daemons from flushing related buffers. */ void bwillwrite(void) { if (numdirtybuffers >= hidirtybuffers) { mtx_lock(&bdirtylock); while (numdirtybuffers >= hidirtybuffers) { bdirtywait = 1; msleep(&bdirtywait, &bdirtylock, (PRIBIO + 4), "flswai", 0); } mtx_unlock(&bdirtylock); } } /* * Return true if we have too many dirty buffers. */ int buf_dirty_count_severe(void) { return(numdirtybuffers >= hidirtybuffers); } static __noinline int buf_vm_page_count_severe(void) { KFAIL_POINT_CODE(DEBUG_FP, buf_pressure, return 1); return vm_page_count_severe(); } /* * brelse: * * Release a busy buffer and, if requested, free its resources. The * buffer will be stashed in the appropriate bufqueue[] allowing it * to be accessed later as a cache entity or reused for other purposes. */ void brelse(struct buf *bp) { int qindex; CTR3(KTR_BUF, "brelse(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(!(bp->b_flags & (B_CLUSTER|B_PAGING)), ("brelse: inappropriate B_PAGING or B_CLUSTER bp %p", bp)); 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_iocmd == BIO_WRITE && (bp->b_ioflags & BIO_ERROR) && bp->b_error == EIO && !(bp->b_flags & B_INVAL)) { /* * Failed write, redirty. Must clear BIO_ERROR to prevent * pages from being scrapped. If the error is anything * other than an I/O error (EIO), assume that retrying * is futile. */ bp->b_ioflags &= ~BIO_ERROR; bdirty(bp); } else if ((bp->b_flags & (B_NOCACHE | B_INVAL)) || (bp->b_ioflags & BIO_ERROR) || (bp->b_bufsize <= 0)) { /* * Either a failed I/O or we were asked to free or not * cache the buffer. */ bp->b_flags |= B_INVAL; if (!LIST_EMPTY(&bp->b_dep)) buf_deallocate(bp); if (bp->b_flags & B_DELWRI) bdirtysub(); bp->b_flags &= ~(B_DELWRI | B_CACHE); if ((bp->b_flags & B_VMIO) == 0) { if (bp->b_bufsize) allocbuf(bp, 0); if (bp->b_vp) brelvp(bp); } } /* * We must clear B_RELBUF if B_DELWRI is set. If vfs_vmio_release() * 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_release(), even * if B_DELWRI is set. * * If B_DELWRI is not set we may have to set B_RELBUF if we are low * on pages to return pages to the VM page queues. */ if (bp->b_flags & B_DELWRI) bp->b_flags &= ~B_RELBUF; else if (buf_vm_page_count_severe()) { /* * 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)) 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_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)) ) { int i, j, resid; vm_page_t m; off_t foff; vm_pindex_t poff; vm_object_t obj; obj = bp->b_bufobj->bo_object; /* * 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 */ resid = bp->b_bufsize; foff = bp->b_offset; for (i = 0; i < bp->b_npages; i++) { int had_bogus = 0; m = bp->b_pages[i]; /* * If we hit a bogus page, fixup *all* the bogus pages * now. */ if (m == bogus_page) { poff = OFF_TO_IDX(bp->b_offset); had_bogus = 1; VM_OBJECT_RLOCK(obj); for (j = i; j < bp->b_npages; j++) { vm_page_t mtmp; mtmp = bp->b_pages[j]; if (mtmp == bogus_page) { mtmp = vm_page_lookup(obj, poff + j); if (!mtmp) { panic("brelse: page missing\n"); } bp->b_pages[j] = mtmp; } } VM_OBJECT_RUNLOCK(obj); if ((bp->b_flags & (B_INVAL | B_UNMAPPED)) == 0) { BUF_CHECK_MAPPED(bp); pmap_qenter( trunc_page((vm_offset_t)bp->b_data), bp->b_pages, bp->b_npages); } m = bp->b_pages[i]; } if ((bp->b_flags & B_NOCACHE) || (bp->b_ioflags & BIO_ERROR && bp->b_iocmd == BIO_READ)) { int poffset = foff & PAGE_MASK; int presid = resid > (PAGE_SIZE - poffset) ? (PAGE_SIZE - poffset) : resid; KASSERT(presid >= 0, ("brelse: extra page")); VM_OBJECT_WLOCK(obj); while (vm_page_xbusied(m)) { vm_page_lock(m); VM_OBJECT_WUNLOCK(obj); vm_page_busy_sleep(m, "mbncsh"); VM_OBJECT_WLOCK(obj); } if (pmap_page_wired_mappings(m) == 0) vm_page_set_invalid(m, poffset, presid); VM_OBJECT_WUNLOCK(obj); if (had_bogus) printf("avoided corruption bug in bogus_page/brelse code\n"); } resid -= PAGE_SIZE - (foff & PAGE_MASK); foff = (foff + PAGE_SIZE) & ~(off_t)PAGE_MASK; } if (bp->b_flags & (B_INVAL | B_RELBUF)) vfs_vmio_release(bp); } else if (bp->b_flags & B_VMIO) { if (bp->b_flags & (B_INVAL | B_RELBUF)) { vfs_vmio_release(bp); } } else if ((bp->b_flags & (B_INVAL | B_RELBUF)) != 0) { if (bp->b_bufsize != 0) allocbuf(bp, 0); if (bp->b_vp != NULL) brelvp(bp); } /* * If the buffer has junk contents signal it and eventually * clean up B_DELWRI and diassociate the vnode so that gbincore() * doesn't find it. */ if (bp->b_bufsize == 0 || (bp->b_ioflags & BIO_ERROR) != 0 || (bp->b_flags & (B_INVAL | B_NOCACHE | B_RELBUF)) != 0) bp->b_flags |= B_INVAL; if (bp->b_flags & B_INVAL) { if (bp->b_flags & B_DELWRI) bundirty(bp); if (bp->b_vp) brelvp(bp); } /* buffers with no memory */ if (bp->b_bufsize == 0) { bp->b_xflags &= ~(BX_BKGRDWRITE | BX_ALTDATA); if (bp->b_vflags & BV_BKGRDINPROG) panic("losing buffer 1"); if (bp->b_kvasize) qindex = QUEUE_EMPTYKVA; else qindex = QUEUE_EMPTY; bp->b_flags |= B_AGE; /* buffers with junk contents */ } else if (bp->b_flags & (B_INVAL | B_NOCACHE | B_RELBUF) || (bp->b_ioflags & BIO_ERROR)) { bp->b_xflags &= ~(BX_BKGRDWRITE | BX_ALTDATA); if (bp->b_vflags & BV_BKGRDINPROG) panic("losing buffer 2"); qindex = QUEUE_CLEAN; bp->b_flags |= B_AGE; /* remaining buffers */ } else if (bp->b_flags & B_DELWRI) qindex = QUEUE_DIRTY; else qindex = QUEUE_CLEAN; binsfree(bp, qindex); bp->b_flags &= ~(B_ASYNC | B_NOCACHE | B_AGE | B_RELBUF | B_DIRECT); if ((bp->b_flags & B_DELWRI) == 0 && (bp->b_xflags & BX_VNDIRTY)) panic("brelse: not dirty"); /* unlock */ BUF_UNLOCK(bp); } /* * Release a buffer back to the appropriate queue but do not try to free * it. The buffer is expected to be used again soon. * * bqrelse() is used by bdwrite() to requeue a delayed write, and used by * biodone() to requeue an async I/O on completion. It is also used when * known good buffers need to be requeued but we think we may need the data * again soon. * * XXX we should be able to leave the B_RELBUF hint set on completion. */ void bqrelse(struct buf *bp) { int qindex; CTR3(KTR_BUF, "bqrelse(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(!(bp->b_flags & (B_CLUSTER|B_PAGING)), ("bqrelse: inappropriate B_PAGING or B_CLUSTER bp %p", bp)); if (BUF_LOCKRECURSED(bp)) { /* do not release to free list */ BUF_UNLOCK(bp); return; } bp->b_flags &= ~(B_ASYNC | B_NOCACHE | B_AGE | B_RELBUF); if (bp->b_flags & B_MANAGED) { if (bp->b_flags & B_REMFREE) bremfreef(bp); goto out; } /* buffers with stale but valid contents */ if (bp->b_flags & B_DELWRI) { qindex = QUEUE_DIRTY; } else { if ((bp->b_flags & B_DELWRI) == 0 && (bp->b_xflags & BX_VNDIRTY)) panic("bqrelse: not dirty"); /* * BKGRDINPROG can only be set with the buf and bufobj * locks both held. We tolerate a race to clear it here. */ if (buf_vm_page_count_severe() && (bp->b_vflags & BV_BKGRDINPROG) == 0) { /* * We are too low on memory, we have to try to free * the buffer (most importantly: the wired pages * making up its backing store) *now*. */ brelse(bp); return; } qindex = QUEUE_CLEAN; } binsfree(bp, qindex); out: /* unlock */ BUF_UNLOCK(bp); } /* Give pages used by the bp back to the VM system (where possible) */ static void vfs_vmio_release(struct buf *bp) { vm_object_t obj; vm_page_t m; int i; if ((bp->b_flags & B_UNMAPPED) == 0) { BUF_CHECK_MAPPED(bp); pmap_qremove(trunc_page((vm_offset_t)bp->b_data), bp->b_npages); } else BUF_CHECK_UNMAPPED(bp); obj = bp->b_bufobj->bo_object; if (obj != NULL) VM_OBJECT_WLOCK(obj); for (i = 0; i < bp->b_npages; i++) { m = bp->b_pages[i]; bp->b_pages[i] = NULL; /* * In order to keep page LRU ordering consistent, put * everything on the inactive queue. */ vm_page_lock(m); vm_page_unwire(m, PQ_INACTIVE); /* * Might as well free the page if we can and it has * no valid data. We also free the page if the * buffer was used for direct I/O */ if ((bp->b_flags & B_ASYNC) == 0 && !m->valid) { if (m->wire_count == 0 && !vm_page_busied(m)) vm_page_free(m); } else if (bp->b_flags & B_DIRECT) vm_page_try_to_free(m); else if (buf_vm_page_count_severe()) vm_page_try_to_cache(m); vm_page_unlock(m); } if (obj != NULL) VM_OBJECT_WUNLOCK(obj); if (bp->b_bufsize) { bufspacewakeup(); bp->b_bufsize = 0; } bp->b_npages = 0; bp->b_flags &= ~B_VMIO; if (bp->b_vp) brelvp(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_flags & B_UNMAPPED) != 0 ? 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); } static void setbufkva(struct buf *bp, vm_offset_t addr, int maxsize, int gbflags) { KASSERT((bp->b_flags & (B_UNMAPPED | B_KVAALLOC)) == 0 && bp->b_kvasize == 0, ("call bfreekva(%p)", bp)); if ((gbflags & GB_UNMAPPED) == 0) { bp->b_kvabase = (caddr_t)addr; } else if ((gbflags & GB_KVAALLOC) != 0) { KASSERT((gbflags & GB_UNMAPPED) != 0, ("GB_KVAALLOC without GB_UNMAPPED")); bp->b_kvaalloc = (caddr_t)addr; bp->b_flags |= B_UNMAPPED | B_KVAALLOC; atomic_add_long(&unmapped_bufspace, bp->b_kvasize); } bp->b_kvasize = maxsize; } /* * Allocate the buffer KVA and set b_kvasize. Also set b_kvabase if * needed. */ static int allocbufkva(struct buf *bp, int maxsize, int gbflags) { vm_offset_t addr; bfreekva(bp); addr = 0; if (vmem_alloc(buffer_arena, maxsize, M_BESTFIT | M_NOWAIT, &addr)) { /* * Buffer map is too fragmented. Request the caller * to defragment the map. */ atomic_add_int(&bufdefragcnt, 1); return (1); } setbufkva(bp, addr, maxsize, gbflags); atomic_add_long(&bufspace, bp->b_kvasize); return (0); } /* * Ask the bufdaemon for help, or act as bufdaemon itself, when a * locked vnode is supplied. */ static void getnewbuf_bufd_help(struct vnode *vp, int gbflags, int slpflag, int slptimeo, int defrag) { struct thread *td; char *waitmsg; int cnt, error, flags, norunbuf, wait; mtx_assert(&bqclean, MA_OWNED); if (defrag) { flags = VFS_BIO_NEED_BUFSPACE; waitmsg = "nbufkv"; } else if (bufspace >= hibufspace) { waitmsg = "nbufbs"; flags = VFS_BIO_NEED_BUFSPACE; } else { waitmsg = "newbuf"; flags = VFS_BIO_NEED_ANY; } atomic_set_int(&needsbuffer, flags); mtx_unlock(&bqclean); bd_speedup(); /* heeeelp */ if ((gbflags & GB_NOWAIT_BD) != 0) return; td = curthread; cnt = 0; wait = MNT_NOWAIT; rw_wlock(&nblock); while ((needsbuffer & flags) != 0) { if (vp != NULL && vp->v_type != VCHR && (td->td_pflags & TDP_BUFNEED) == 0) { rw_wunlock(&nblock); /* * getblk() is called with a vnode locked, and * some majority of the dirty buffers may as * well belong to the vnode. Flushing the * buffers there would make a progress that * cannot be achieved by the buf_daemon, that * cannot lock the vnode. */ if (cnt++ > 2) wait = MNT_WAIT; ASSERT_VOP_LOCKED(vp, "bufd_helper"); error = VOP_ISLOCKED(vp) == LK_EXCLUSIVE ? 0 : vn_lock(vp, LK_TRYUPGRADE); if (error == 0) { /* play bufdaemon */ norunbuf = curthread_pflags_set(TDP_BUFNEED | TDP_NORUNNINGBUF); VOP_FSYNC(vp, wait, td); atomic_add_long(¬bufdflushes, 1); curthread_pflags_restore(norunbuf); } rw_wlock(&nblock); if ((needsbuffer & flags) == 0) break; } error = rw_sleep(__DEVOLATILE(void *, &needsbuffer), &nblock, (PRIBIO + 4) | slpflag, waitmsg, slptimeo); if (error != 0) break; } rw_wunlock(&nblock); } static void getnewbuf_reuse_bp(struct buf *bp, int qindex) { CTR6(KTR_BUF, "getnewbuf(%p) vp %p flags %X kvasize %d bufsize %d " "queue %d (recycling)", bp, bp->b_vp, bp->b_flags, bp->b_kvasize, bp->b_bufsize, qindex); mtx_assert(&bqclean, MA_NOTOWNED); /* * Note: we no longer distinguish between VMIO and non-VMIO * buffers. */ KASSERT((bp->b_flags & B_DELWRI) == 0, ("delwri buffer %p found in queue %d", bp, qindex)); if (qindex == QUEUE_CLEAN) { if (bp->b_flags & B_VMIO) { bp->b_flags &= ~B_ASYNC; vfs_vmio_release(bp); } if (bp->b_vp != NULL) brelvp(bp); } /* * Get the rest of the buffer freed up. b_kva* is still valid * after this operation. */ if (bp->b_rcred != NOCRED) { crfree(bp->b_rcred); bp->b_rcred = NOCRED; } if (bp->b_wcred != NOCRED) { crfree(bp->b_wcred); bp->b_wcred = NOCRED; } if (!LIST_EMPTY(&bp->b_dep)) buf_deallocate(bp); if (bp->b_vflags & BV_BKGRDINPROG) panic("losing buffer 3"); KASSERT(bp->b_vp == NULL, ("bp: %p still has vnode %p. qindex: %d", bp, bp->b_vp, qindex)); KASSERT((bp->b_xflags & (BX_VNCLEAN|BX_VNDIRTY)) == 0, ("bp: %p still on a buffer list. xflags %X", bp, bp->b_xflags)); if (bp->b_bufsize) allocbuf(bp, 0); bp->b_flags &= B_UNMAPPED | B_KVAALLOC; bp->b_ioflags = 0; bp->b_xflags = 0; KASSERT((bp->b_flags & B_INFREECNT) == 0, ("buf %p still counted as free?", bp)); bp->b_vflags = 0; bp->b_vp = NULL; bp->b_blkno = bp->b_lblkno = 0; bp->b_offset = NOOFFSET; bp->b_iodone = 0; bp->b_error = 0; bp->b_resid = 0; bp->b_bcount = 0; bp->b_npages = 0; bp->b_dirtyoff = bp->b_dirtyend = 0; bp->b_bufobj = NULL; bp->b_pin_count = 0; bp->b_fsprivate1 = NULL; bp->b_fsprivate2 = NULL; bp->b_fsprivate3 = NULL; LIST_INIT(&bp->b_dep); } static int flushingbufs; static struct buf * getnewbuf_scan(int maxsize, int defrag, int unmapped, int metadata) { struct buf *bp, *nbp; int nqindex, qindex, pass; KASSERT(!unmapped || !defrag, ("both unmapped and defrag")); pass = 1; restart: atomic_add_int(&getnewbufrestarts, 1); /* * Setup for scan. If we do not have enough free buffers, * we setup a degenerate case that immediately fails. Note * that if we are specially marked process, we are allowed to * dip into our reserves. * * The scanning sequence is nominally: EMPTY->EMPTYKVA->CLEAN * for the allocation of the mapped buffer. For unmapped, the * easiest is to start with EMPTY outright. * * We start with EMPTYKVA. If the list is empty we backup to EMPTY. * However, there are a number of cases (defragging, reusing, ...) * where we cannot backup. */ nbp = NULL; mtx_lock(&bqclean); if (!defrag && unmapped) { nqindex = QUEUE_EMPTY; nbp = TAILQ_FIRST(&bufqueues[QUEUE_EMPTY]); } if (nbp == NULL) { nqindex = QUEUE_EMPTYKVA; nbp = TAILQ_FIRST(&bufqueues[QUEUE_EMPTYKVA]); } /* * If no EMPTYKVA buffers and we are either defragging or * reusing, locate a CLEAN buffer to free or reuse. If * bufspace useage is low skip this step so we can allocate a * new buffer. */ if (nbp == NULL && (defrag || bufspace >= lobufspace)) { nqindex = QUEUE_CLEAN; nbp = TAILQ_FIRST(&bufqueues[QUEUE_CLEAN]); } /* * If we could not find or were not allowed to reuse a CLEAN * buffer, check to see if it is ok to use an EMPTY buffer. * We can only use an EMPTY buffer if allocating its KVA would * not otherwise run us out of buffer space. No KVA is needed * for the unmapped allocation. */ if (nbp == NULL && defrag == 0 && (bufspace + maxsize < hibufspace || metadata)) { nqindex = QUEUE_EMPTY; nbp = TAILQ_FIRST(&bufqueues[QUEUE_EMPTY]); } /* * All available buffers might be clean, retry ignoring the * lobufspace as the last resort. */ if (nbp == NULL && !TAILQ_EMPTY(&bufqueues[QUEUE_CLEAN])) { nqindex = QUEUE_CLEAN; nbp = TAILQ_FIRST(&bufqueues[QUEUE_CLEAN]); } /* * Run scan, possibly freeing data and/or kva mappings on the fly * depending. */ while ((bp = nbp) != NULL) { qindex = nqindex; /* * Calculate next bp (we can only use it if we do not * block or do other fancy things). */ if ((nbp = TAILQ_NEXT(bp, b_freelist)) == NULL) { switch (qindex) { case QUEUE_EMPTY: nqindex = QUEUE_EMPTYKVA; nbp = TAILQ_FIRST(&bufqueues[QUEUE_EMPTYKVA]); if (nbp != NULL) break; /* FALLTHROUGH */ case QUEUE_EMPTYKVA: nqindex = QUEUE_CLEAN; nbp = TAILQ_FIRST(&bufqueues[QUEUE_CLEAN]); if (nbp != NULL) break; /* FALLTHROUGH */ case QUEUE_CLEAN: if (metadata && pass == 1) { pass = 2; nqindex = QUEUE_EMPTY; nbp = TAILQ_FIRST( &bufqueues[QUEUE_EMPTY]); } /* * nbp is NULL. */ break; } } /* * If we are defragging then we need a buffer with * b_kvasize != 0. XXX this situation should no longer * occur, if defrag is non-zero the buffer's b_kvasize * should also be non-zero at this point. XXX */ if (defrag && bp->b_kvasize == 0) { printf("Warning: defrag empty buffer %p\n", bp); continue; } /* * Start freeing the bp. This is somewhat involved. nbp * remains valid only for QUEUE_EMPTY[KVA] bp's. */ if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL) != 0) continue; /* * BKGRDINPROG can only be set with the buf and bufobj * locks both held. We tolerate a race to clear it here. */ if (bp->b_vflags & BV_BKGRDINPROG) { BUF_UNLOCK(bp); continue; } KASSERT(bp->b_qindex == qindex, ("getnewbuf: inconsistent queue %d bp %p", qindex, bp)); bremfreel(bp); mtx_unlock(&bqclean); /* * NOTE: nbp is now entirely invalid. We can only restart * the scan from this point on. */ getnewbuf_reuse_bp(bp, qindex); mtx_assert(&bqclean, MA_NOTOWNED); /* * If we are defragging then free the buffer. */ if (defrag) { bp->b_flags |= B_INVAL; bfreekva(bp); brelse(bp); defrag = 0; goto restart; } /* * Notify any waiters for the buffer lock about * identity change by freeing the buffer. */ if (qindex == QUEUE_CLEAN && BUF_LOCKWAITERS(bp)) { bp->b_flags |= B_INVAL; bfreekva(bp); brelse(bp); goto restart; } if (metadata) break; /* * If we are overcomitted then recover the buffer and its * KVM space. This occurs in rare situations when multiple * processes are blocked in getnewbuf() or allocbuf(). */ if (bufspace >= hibufspace) flushingbufs = 1; if (flushingbufs && bp->b_kvasize != 0) { bp->b_flags |= B_INVAL; bfreekva(bp); brelse(bp); goto restart; } if (bufspace < lobufspace) flushingbufs = 0; break; } return (bp); } /* * getnewbuf: * * Find and initialize a new buffer header, freeing up existing buffers * in the bufqueues as necessary. The new buffer is returned locked. * * Important: B_INVAL is not set. If the caller wishes to throw the * buffer away, the caller must set B_INVAL prior to calling brelse(). * * We block if: * We have insufficient buffer headers * We have insufficient buffer space * buffer_arena is too fragmented ( space reservation fails ) * If we have to flush dirty buffers ( but we try to avoid this ) */ static struct buf * getnewbuf(struct vnode *vp, int slpflag, int slptimeo, int size, int maxsize, int gbflags) { struct buf *bp; int defrag, metadata; KASSERT((gbflags & (GB_UNMAPPED | GB_KVAALLOC)) != GB_KVAALLOC, ("GB_KVAALLOC only makes sense with GB_UNMAPPED")); if (!unmapped_buf_allowed) gbflags &= ~(GB_UNMAPPED | GB_KVAALLOC); defrag = 0; if (vp == NULL || (vp->v_vflag & (VV_MD | VV_SYSTEM)) != 0 || vp->v_type == VCHR) metadata = 1; else metadata = 0; /* * We can't afford to block since we might be holding a vnode lock, * which may prevent system daemons from running. We deal with * low-memory situations by proactively returning memory and running * async I/O rather then sync I/O. */ atomic_add_int(&getnewbufcalls, 1); atomic_subtract_int(&getnewbufrestarts, 1); restart: bp = getnewbuf_scan(maxsize, defrag, (gbflags & (GB_UNMAPPED | GB_KVAALLOC)) == GB_UNMAPPED, metadata); if (bp != NULL) defrag = 0; /* * If we exhausted our list, sleep as appropriate. We may have to * wakeup various daemons and write out some dirty buffers. * * Generally we are sleeping due to insufficient buffer space. */ if (bp == NULL) { mtx_assert(&bqclean, MA_OWNED); getnewbuf_bufd_help(vp, gbflags, slpflag, slptimeo, defrag); mtx_assert(&bqclean, MA_NOTOWNED); } else if ((gbflags & (GB_UNMAPPED | GB_KVAALLOC)) == GB_UNMAPPED) { mtx_assert(&bqclean, MA_NOTOWNED); bfreekva(bp); bp->b_flags |= B_UNMAPPED; bp->b_kvabase = bp->b_data = unmapped_buf; bp->b_kvasize = maxsize; atomic_add_long(&bufspace, bp->b_kvasize); atomic_add_long(&unmapped_bufspace, bp->b_kvasize); atomic_add_int(&bufreusecnt, 1); } else { mtx_assert(&bqclean, MA_NOTOWNED); /* * We finally have a valid bp. We aren't quite out of the * woods, we still have to reserve kva space. In order * to keep fragmentation sane we only allocate kva in * BKVASIZE chunks. */ maxsize = (maxsize + BKVAMASK) & ~BKVAMASK; if (maxsize != bp->b_kvasize || (bp->b_flags & (B_UNMAPPED | B_KVAALLOC)) == B_UNMAPPED) { if (allocbufkva(bp, maxsize, gbflags)) { defrag = 1; bp->b_flags |= B_INVAL; brelse(bp); goto restart; } atomic_add_int(&bufreusecnt, 1); } else if ((bp->b_flags & B_KVAALLOC) != 0 && (gbflags & (GB_UNMAPPED | GB_KVAALLOC)) == 0) { /* * If the reused buffer has KVA allocated, * reassign b_kvaalloc to b_kvabase. */ bp->b_kvabase = bp->b_kvaalloc; bp->b_flags &= ~B_KVAALLOC; atomic_subtract_long(&unmapped_bufspace, bp->b_kvasize); atomic_add_int(&bufreusecnt, 1); } else if ((bp->b_flags & (B_UNMAPPED | B_KVAALLOC)) == 0 && (gbflags & (GB_UNMAPPED | GB_KVAALLOC)) == (GB_UNMAPPED | GB_KVAALLOC)) { /* * The case of reused buffer already have KVA * mapped, but the request is for unmapped * buffer with KVA allocated. */ bp->b_kvaalloc = bp->b_kvabase; bp->b_data = bp->b_kvabase = unmapped_buf; bp->b_flags |= B_UNMAPPED | B_KVAALLOC; atomic_add_long(&unmapped_bufspace, bp->b_kvasize); atomic_add_int(&bufreusecnt, 1); } if ((gbflags & GB_UNMAPPED) == 0) { bp->b_saveaddr = bp->b_kvabase; bp->b_data = bp->b_saveaddr; bp->b_flags &= ~B_UNMAPPED; BUF_CHECK_MAPPED(bp); } } return (bp); } /* * buf_daemon: * * buffer flushing daemon. Buffers are normally flushed by the * update daemon but if it cannot keep up this process starts to * take the load in an attempt to prevent getnewbuf() from blocking. */ static struct kproc_desc buf_kp = { "bufdaemon", buf_daemon, &bufdaemonproc }; SYSINIT(bufdaemon, SI_SUB_KTHREAD_BUF, SI_ORDER_FIRST, kproc_start, &buf_kp); static int buf_flush(int target) { int flushed; flushed = flushbufqueues(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. */ flushed = flushbufqueues(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(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(int target, int flushdeps) { struct buf *sentinel; struct vnode *vp; struct mount *mp; struct buf *bp; int hasdeps; int flushed; int queue; int error; flushed = 0; queue = QUEUE_DIRTY; bp = NULL; sentinel = malloc(sizeof(struct buf), M_TEMP, M_WAITOK | M_ZERO); sentinel->b_qindex = QUEUE_SENTINEL; mtx_lock(&bqdirty); TAILQ_INSERT_HEAD(&bufqueues[queue], sentinel, b_freelist); mtx_unlock(&bqdirty); while (flushed != target) { maybe_yield(); mtx_lock(&bqdirty); bp = TAILQ_NEXT(sentinel, b_freelist); if (bp != NULL) { TAILQ_REMOVE(&bufqueues[queue], sentinel, b_freelist); TAILQ_INSERT_AFTER(&bufqueues[queue], bp, sentinel, b_freelist); } else { mtx_unlock(&bqdirty); break; } KASSERT(bp->b_qindex != QUEUE_SENTINEL, ("parallel calls to flushbufqueues() bp %p", bp)); error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL); mtx_unlock(&bqdirty); if (error != 0) continue; if (bp->b_pin_count > 0) { BUF_UNLOCK(bp); continue; } /* * BKGRDINPROG can only be set with the buf and bufobj * locks both held. We tolerate a race to clear it here. */ if ((bp->b_vflags & BV_BKGRDINPROG) != 0 || (bp->b_flags & B_DELWRI) == 0) { BUF_UNLOCK(bp); continue; } if (bp->b_flags & B_INVAL) { bremfreef(bp); brelse(bp); flushed++; continue; } if (!LIST_EMPTY(&bp->b_dep) && buf_countdeps(bp, 0)) { if (flushdeps == 0) { BUF_UNLOCK(bp); continue; } hasdeps = 1; } else hasdeps = 0; /* * We must hold the lock on a vnode before writing * one of its buffers. Otherwise we may confuse, or * in the case of a snapshot vnode, deadlock the * system. * * The lock order here is the reverse of the normal * of vnode followed by buf lock. This is ok because * the NOWAIT will prevent deadlock. */ vp = bp->b_vp; if (vn_start_write(vp, &mp, V_NOWAIT) != 0) { BUF_UNLOCK(bp); continue; } error = vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT); if (error == 0) { CTR3(KTR_BUF, "flushbufqueue(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); vfs_bio_awrite(bp); vn_finished_write(mp); VOP_UNLOCK(vp, 0); flushwithdeps += hasdeps; flushed++; if (runningbufspace > hirunningspace) waitrunningbufspace(); continue; } vn_finished_write(mp); BUF_UNLOCK(bp); } mtx_lock(&bqdirty); TAILQ_REMOVE(&bufqueues[queue], sentinel, b_freelist); mtx_unlock(&bqdirty); free(sentinel, M_TEMP); return (flushed); } /* * Check to see if a block is currently memory resident. */ struct buf * incore(struct bufobj *bo, daddr_t blkno) { struct buf *bp; BO_RLOCK(bo); bp = gbincore(bo, blkno); BO_RUNLOCK(bo); return (bp); } /* * Returns true if no I/O is needed to access the * associated VM object. This is like incore except * it also hunts around in the VM system for the data. */ static int inmem(struct vnode * vp, daddr_t blkno) { vm_object_t obj; vm_offset_t toff, tinc, size; vm_page_t m; vm_ooffset_t off; ASSERT_VOP_LOCKED(vp, "inmem"); if (incore(&vp->v_bufobj, blkno)) return 1; if (vp->v_mount == NULL) return 0; obj = vp->v_object; if (obj == NULL) return (0); size = PAGE_SIZE; if (size > vp->v_mount->mnt_stat.f_iosize) size = vp->v_mount->mnt_stat.f_iosize; off = (vm_ooffset_t)blkno * (vm_ooffset_t)vp->v_mount->mnt_stat.f_iosize; VM_OBJECT_RLOCK(obj); for (toff = 0; toff < vp->v_mount->mnt_stat.f_iosize; toff += tinc) { m = vm_page_lookup(obj, OFF_TO_IDX(off + toff)); if (!m) goto notinmem; tinc = size; if (tinc > PAGE_SIZE - ((toff + off) & PAGE_MASK)) tinc = PAGE_SIZE - ((toff + off) & PAGE_MASK); if (vm_page_is_valid(m, (vm_offset_t) ((toff + off) & PAGE_MASK), tinc) == 0) goto notinmem; } VM_OBJECT_RUNLOCK(obj); return 1; notinmem: VM_OBJECT_RUNLOCK(obj); return (0); } /* * Set the dirty range for a buffer based on the status of the dirty * bits in the pages comprising the buffer. The range is limited * to the size of the buffer. * * Tell the VM system that the pages associated with this buffer * are clean. This is used for delayed writes where the data is * going to go to disk eventually without additional VM intevention. * * Note that while we only really need to clean through to b_bcount, we * just go ahead and clean through to b_bufsize. */ static void vfs_clean_pages_dirty_buf(struct buf *bp) { vm_ooffset_t foff, noff, eoff; vm_page_t m; int i; if ((bp->b_flags & B_VMIO) == 0 || bp->b_bufsize == 0) return; foff = bp->b_offset; KASSERT(bp->b_offset != NOOFFSET, ("vfs_clean_pages_dirty_buf: no buffer offset")); VM_OBJECT_WLOCK(bp->b_bufobj->bo_object); vfs_drain_busy_pages(bp); vfs_setdirty_locked_object(bp); for (i = 0; i < bp->b_npages; i++) { noff = (foff + PAGE_SIZE) & ~(off_t)PAGE_MASK; eoff = noff; if (eoff > bp->b_offset + bp->b_bufsize) eoff = bp->b_offset + bp->b_bufsize; m = bp->b_pages[i]; vfs_page_set_validclean(bp, foff, m); /* vm_page_clear_dirty(m, foff & PAGE_MASK, eoff - foff); */ foff = noff; } VM_OBJECT_WUNLOCK(bp->b_bufobj->bo_object); } static void vfs_setdirty_locked_object(struct buf *bp) { vm_object_t object; int i; object = bp->b_bufobj->bo_object; VM_OBJECT_ASSERT_WLOCKED(object); /* * We qualify the scan for modified pages on whether the * object has been flushed yet. */ if ((object->flags & OBJ_MIGHTBEDIRTY) != 0) { vm_offset_t boffset; vm_offset_t eoffset; /* * test the pages to see if they have been modified directly * by users through the VM system. */ for (i = 0; i < bp->b_npages; i++) vm_page_test_dirty(bp->b_pages[i]); /* * Calculate the encompassing dirty range, boffset and eoffset, * (eoffset - boffset) bytes. */ for (i = 0; i < bp->b_npages; i++) { if (bp->b_pages[i]->dirty) break; } boffset = (i << PAGE_SHIFT) - (bp->b_offset & PAGE_MASK); for (i = bp->b_npages - 1; i >= 0; --i) { if (bp->b_pages[i]->dirty) { break; } } eoffset = ((i + 1) << PAGE_SHIFT) - (bp->b_offset & PAGE_MASK); /* * Fit it to the buffer. */ if (eoffset > bp->b_bcount) eoffset = bp->b_bcount; /* * If we have a good dirty range, merge with the existing * dirty range. */ if (boffset < eoffset) { if (bp->b_dirtyoff > boffset) bp->b_dirtyoff = boffset; if (bp->b_dirtyend < eoffset) bp->b_dirtyend = eoffset; } } } /* * Allocate the KVA mapping for an existing buffer. It handles the * cases of both B_UNMAPPED buffer, and buffer with the preallocated * KVA which is not mapped (B_KVAALLOC). */ static void bp_unmapped_get_kva(struct buf *bp, daddr_t blkno, int size, int gbflags) { struct buf *scratch_bp; int bsize, maxsize, need_mapping, need_kva; off_t offset; need_mapping = (bp->b_flags & B_UNMAPPED) != 0 && (gbflags & GB_UNMAPPED) == 0; need_kva = (bp->b_flags & (B_KVAALLOC | B_UNMAPPED)) == B_UNMAPPED && (gbflags & GB_KVAALLOC) != 0; if (!need_mapping && !need_kva) return; BUF_CHECK_UNMAPPED(bp); if (need_mapping && (bp->b_flags & B_KVAALLOC) != 0) { /* * Buffer is not mapped, but the KVA was already * reserved at the time of the instantiation. Use the * allocated space. */ bp->b_flags &= ~B_KVAALLOC; KASSERT(bp->b_kvaalloc != 0, ("kvaalloc == 0")); bp->b_kvabase = bp->b_kvaalloc; atomic_subtract_long(&unmapped_bufspace, bp->b_kvasize); goto has_addr; } /* * Calculate the amount of the address space we would reserve * if the buffer was mapped. */ bsize = vn_isdisk(bp->b_vp, NULL) ? DEV_BSIZE : bp->b_bufobj->bo_bsize; KASSERT(bsize != 0, ("bsize == 0, check bo->bo_bsize")); offset = blkno * bsize; maxsize = size + (offset & PAGE_MASK); maxsize = imax(maxsize, bsize); mapping_loop: if (allocbufkva(bp, maxsize, gbflags)) { /* * Request defragmentation. getnewbuf() returns us the * allocated space by the scratch buffer KVA. */ scratch_bp = getnewbuf(bp->b_vp, 0, 0, size, maxsize, gbflags | (GB_UNMAPPED | GB_KVAALLOC)); if (scratch_bp == NULL) { if ((gbflags & GB_NOWAIT_BD) != 0) { /* * XXXKIB: defragmentation cannot * succeed, not sure what else to do. */ panic("GB_NOWAIT_BD and B_UNMAPPED %p", bp); } atomic_add_int(&mappingrestarts, 1); goto mapping_loop; } KASSERT((scratch_bp->b_flags & B_KVAALLOC) != 0, ("scratch bp !B_KVAALLOC %p", scratch_bp)); setbufkva(bp, (vm_offset_t)scratch_bp->b_kvaalloc, scratch_bp->b_kvasize, gbflags); /* Get rid of the scratch buffer. */ scratch_bp->b_kvasize = 0; scratch_bp->b_flags |= B_INVAL; scratch_bp->b_flags &= ~(B_UNMAPPED | B_KVAALLOC); brelse(scratch_bp); } if (!need_mapping) return; has_addr: bp->b_saveaddr = bp->b_kvabase; bp->b_data = bp->b_saveaddr; /* b_offset is handled by bpmap_qenter */ bp->b_flags &= ~B_UNMAPPED; BUF_CHECK_MAPPED(bp); bpmap_qenter(bp); } /* * getblk: * * Get a block given a specified block and offset into a file/device. * The buffers B_DONE bit will be cleared on return, making it almost * ready for an I/O initiation. B_INVAL may or may not be set on * return. The caller should clear B_INVAL prior to initiating a * READ. * * For a non-VMIO buffer, B_CACHE is set to the opposite of B_INVAL for * an existing buffer. * * For a VMIO buffer, B_CACHE is modified according to the backing VM. * If getblk()ing a previously 0-sized invalid buffer, B_CACHE is set * and then cleared based on the backing VM. If the previous buffer is * non-0-sized but invalid, B_CACHE will be cleared. * * If getblk() must create a new buffer, the new buffer is returned with * both B_INVAL and B_CACHE clear unless it is a VMIO buffer, in which * case it is returned with B_INVAL clear and B_CACHE set based on the * backing VM. * * getblk() also forces a bwrite() for any B_DELWRI buffer whos * B_CACHE bit is clear. * * What this means, basically, is that the caller should use B_CACHE to * determine whether the buffer is fully valid or not and should clear * B_INVAL prior to issuing a read. If the caller intends to validate * the buffer by loading its data area with something, the caller needs * to clear B_INVAL. If the caller does this without issuing an I/O, * the caller should set B_CACHE ( as an optimization ), else the caller * should issue the I/O and biodone() will set B_CACHE if the I/O was * a write attempt or if it was a successfull read. If the caller * intends to issue a READ, the caller must clear B_INVAL and BIO_ERROR * prior to issuing the READ. biodone() will *not* clear B_INVAL. */ struct buf * getblk(struct vnode *vp, daddr_t blkno, int size, int slpflag, int slptimeo, int flags) { struct buf *bp; struct bufobj *bo; int bsize, error, maxsize, vmio; off_t offset; CTR3(KTR_BUF, "getblk(%p, %ld, %d)", vp, (long)blkno, size); KASSERT((flags & (GB_UNMAPPED | GB_KVAALLOC)) != GB_KVAALLOC, ("GB_KVAALLOC only makes sense with GB_UNMAPPED")); ASSERT_VOP_LOCKED(vp, "getblk"); - if (size > MAXBSIZE) - panic("getblk: size(%d) > MAXBSIZE(%d)\n", size, MAXBSIZE); + if (size > MAXBCACHEBUF) + panic("getblk: size(%d) > MAXBCACHEBUF(%d)\n", size, + MAXBCACHEBUF); if (!unmapped_buf_allowed) flags &= ~(GB_UNMAPPED | GB_KVAALLOC); bo = &vp->v_bufobj; loop: BO_RLOCK(bo); bp = gbincore(bo, blkno); if (bp != NULL) { int lockflags; /* * Buffer is in-core. If the buffer is not busy nor managed, * it must be on a queue. */ lockflags = LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK; if (flags & GB_LOCK_NOWAIT) lockflags |= LK_NOWAIT; error = BUF_TIMELOCK(bp, lockflags, BO_LOCKPTR(bo), "getblk", slpflag, slptimeo); /* * If we slept and got the lock we have to restart in case * the buffer changed identities. */ if (error == ENOLCK) goto loop; /* We timed out or were interrupted. */ else if (error) return (NULL); /* If recursed, assume caller knows the rules. */ else if (BUF_LOCKRECURSED(bp)) goto end; /* * The buffer is locked. B_CACHE is cleared if the buffer is * invalid. Otherwise, for a non-VMIO buffer, B_CACHE is set * and for a VMIO buffer B_CACHE is adjusted according to the * backing VM cache. */ if (bp->b_flags & B_INVAL) bp->b_flags &= ~B_CACHE; else if ((bp->b_flags & (B_VMIO | B_INVAL)) == 0) bp->b_flags |= B_CACHE; if (bp->b_flags & B_MANAGED) MPASS(bp->b_qindex == QUEUE_NONE); else bremfree(bp); /* * check for size inconsistencies for non-VMIO case. */ if (bp->b_bcount != size) { if ((bp->b_flags & B_VMIO) == 0 || (size > bp->b_kvasize)) { if (bp->b_flags & B_DELWRI) { /* * If buffer is pinned and caller does * not want sleep waiting for it to be * unpinned, bail out * */ if (bp->b_pin_count > 0) { if (flags & GB_LOCK_NOWAIT) { bqrelse(bp); return (NULL); } else { bunpin_wait(bp); } } bp->b_flags |= B_NOCACHE; bwrite(bp); } else { if (LIST_EMPTY(&bp->b_dep)) { bp->b_flags |= B_RELBUF; brelse(bp); } else { bp->b_flags |= B_NOCACHE; bwrite(bp); } } goto loop; } } /* * Handle the case of unmapped buffer which should * become mapped, or the buffer for which KVA * reservation is requested. */ bp_unmapped_get_kva(bp, blkno, size, flags); /* * If the size is inconsistant in the VMIO case, we can resize * the buffer. This might lead to B_CACHE getting set or * cleared. If the size has not changed, B_CACHE remains * unchanged from its previous state. */ if (bp->b_bcount != size) 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; } maxsize = imax(maxsize, bsize); bp = getnewbuf(vp, slpflag, slptimeo, size, maxsize, flags); if (bp == NULL) { if (slpflag || slptimeo) return NULL; goto loop; } /* * This code is used to make sure that a buffer is not * created while the getnewbuf routine is blocked. * This can be a problem whether the vnode is locked or not. * If the buffer is created out from under us, we have to * throw away the one we just created. * * Note: this must occur before we associate the buffer * with the vp especially considering limitations in * the splay tree implementation when dealing with duplicate * lblkno's. */ BO_LOCK(bo); if (gbincore(bo, blkno)) { BO_UNLOCK(bo); bp->b_flags |= B_INVAL; brelse(bp); goto loop; } /* * Insert the buffer into the hash, so that it can * be found by incore. */ bp->b_blkno = bp->b_lblkno = blkno; bp->b_offset = offset; bgetvp(vp, bp); BO_UNLOCK(bo); /* * set B_VMIO bit. allocbuf() the buffer bigger. Since the * buffer size starts out as 0, B_CACHE will be set by * allocbuf() for the VMIO case prior to it testing the * backing store for validity. */ if (vmio) { bp->b_flags |= B_VMIO; KASSERT(vp->v_object == bp->b_bufobj->bo_object, ("ARGH! different b_bufobj->bo_object %p %p %p\n", bp, vp->v_object, bp->b_bufobj->bo_object)); } else { bp->b_flags &= ~B_VMIO; KASSERT(bp->b_bufobj->bo_object == NULL, ("ARGH! has b_bufobj->bo_object %p %p\n", bp, bp->b_bufobj->bo_object)); BUF_CHECK_MAPPED(bp); } allocbuf(bp, size); bp->b_flags &= ~B_DONE; } CTR4(KTR_BUF, "getblk(%p, %ld, %d) = %p", vp, (long)blkno, size, bp); BUF_ASSERT_HELD(bp); end: KASSERT(bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); return (bp); } /* * Get an empty, disassociated buffer of given size. The buffer is initially * set to B_INVAL. */ struct buf * geteblk(int size, int flags) { struct buf *bp; int maxsize; maxsize = (size + BKVAMASK) & ~BKVAMASK; while ((bp = getnewbuf(NULL, 0, 0, size, maxsize, flags)) == NULL) { if ((flags & GB_NOWAIT_BD) && (curthread->td_pflags & TDP_BUFNEED) != 0) return (NULL); } allocbuf(bp, size); bp->b_flags |= B_INVAL; /* b_dep cleared by getnewbuf() */ BUF_ASSERT_HELD(bp); return (bp); } /* * This code constitutes the buffer memory from either anonymous system * memory (in the case of non-VMIO operations) or from an associated * VM object (in the case of VMIO operations). This code is able to * resize a buffer up or down. * * Note that this code is tricky, and has many complications to resolve * deadlock or inconsistant data situations. Tread lightly!!! * There are B_CACHE and B_DELWRI interactions that must be dealt with by * the caller. Calling this code willy nilly can result in the loss of data. * * allocbuf() only adjusts B_CACHE for VMIO buffers. getblk() deals with * B_CACHE for the non-VMIO case. */ int allocbuf(struct buf *bp, int size) { int newbsize, mbsize; int i; BUF_ASSERT_HELD(bp); if (bp->b_kvasize < size) panic("allocbuf: buffer too small"); if ((bp->b_flags & B_VMIO) == 0) { caddr_t origbuf; int origbufsize; /* * Just get anonymous memory from the kernel. Don't * mess with B_CACHE. */ mbsize = (size + DEV_BSIZE - 1) & ~(DEV_BSIZE - 1); if (bp->b_flags & B_MALLOC) newbsize = mbsize; else newbsize = round_page(size); if (newbsize < bp->b_bufsize) { /* * malloced buffers are not shrunk */ if (bp->b_flags & B_MALLOC) { if (newbsize) { bp->b_bcount = size; } else { free(bp->b_data, M_BIOBUF); if (bp->b_bufsize) { atomic_subtract_long( &bufmallocspace, bp->b_bufsize); bufspacewakeup(); bp->b_bufsize = 0; } bp->b_saveaddr = bp->b_kvabase; bp->b_data = bp->b_saveaddr; bp->b_bcount = 0; bp->b_flags &= ~B_MALLOC; } return 1; } vm_hold_free_pages(bp, newbsize); } else if (newbsize > bp->b_bufsize) { /* * 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 ( (bufmallocspace < maxbufmallocspace) && (bp->b_bufsize == 0) && (mbsize <= PAGE_SIZE/2)) { bp->b_data = malloc(mbsize, M_BIOBUF, M_WAITOK); bp->b_bufsize = mbsize; bp->b_bcount = size; bp->b_flags |= B_MALLOC; atomic_add_long(&bufmallocspace, mbsize); return 1; } origbuf = NULL; origbufsize = 0; /* * If the buffer is growing on its other-than-first allocation, * then we revert to the page-allocation scheme. */ if (bp->b_flags & B_MALLOC) { origbuf = bp->b_data; origbufsize = bp->b_bufsize; bp->b_data = bp->b_kvabase; if (bp->b_bufsize) { atomic_subtract_long(&bufmallocspace, bp->b_bufsize); bufspacewakeup(); bp->b_bufsize = 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) { bcopy(origbuf, bp->b_data, origbufsize); free(origbuf, M_BIOBUF); } } } else { int desiredpages; newbsize = (size + DEV_BSIZE - 1) & ~(DEV_BSIZE - 1); 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) { /* * DEV_BSIZE aligned new buffer size is less then the * DEV_BSIZE aligned existing buffer size. Figure out * if we have to remove any pages. */ if (desiredpages < bp->b_npages) { vm_page_t m; if ((bp->b_flags & B_UNMAPPED) == 0) { 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); VM_OBJECT_WLOCK(bp->b_bufobj->bo_object); for (i = desiredpages; i < bp->b_npages; i++) { /* * the page is not freed here -- it * is the responsibility of * vnode_pager_setsize */ m = bp->b_pages[i]; KASSERT(m != bogus_page, ("allocbuf: bogus page found")); while (vm_page_sleep_if_busy(m, "biodep")) continue; bp->b_pages[i] = NULL; vm_page_lock(m); vm_page_unwire(m, PQ_INACTIVE); vm_page_unlock(m); } VM_OBJECT_WUNLOCK(bp->b_bufobj->bo_object); bp->b_npages = desiredpages; } } else if (size > bp->b_bcount) { /* * We are growing the buffer, possibly in a * byte-granular fashion. */ vm_object_t obj; vm_offset_t toff; vm_offset_t tinc; /* * Step 1, bring in the VM pages from the object, * allocating them if necessary. We must clear * B_CACHE if these pages are not valid for the * range covered by the buffer. */ obj = bp->b_bufobj->bo_object; VM_OBJECT_WLOCK(obj); while (bp->b_npages < desiredpages) { vm_page_t m; /* * We must allocate system pages since blocking * here could interfere with paging I/O, no * matter which process we are. * * Only exclusive busy can be tested here. * Blocking on shared busy might lead to * deadlocks once allocbuf() is called after * pages are vfs_busy_pages(). */ m = vm_page_grab(obj, OFF_TO_IDX(bp->b_offset) + bp->b_npages, VM_ALLOC_NOBUSY | VM_ALLOC_SYSTEM | VM_ALLOC_WIRED | VM_ALLOC_IGN_SBUSY | VM_ALLOC_COUNT(desiredpages - bp->b_npages)); if (m->valid == 0) bp->b_flags &= ~B_CACHE; bp->b_pages[bp->b_npages] = m; ++bp->b_npages; } /* * Step 2. We've loaded the pages into the buffer, * we have to figure out if we can still have B_CACHE * set. Note that B_CACHE is set according to the * byte-granular range ( bcount and size ), new 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; vfs_buf_test_cache( bp, bp->b_offset, toff, tinc, bp->b_pages[pi] ); toff += tinc; tinc = PAGE_SIZE; } VM_OBJECT_WUNLOCK(obj); /* * Step 3, fixup the KVM pmap. */ if ((bp->b_flags & B_UNMAPPED) == 0) bpmap_qenter(bp); else BUF_CHECK_UNMAPPED(bp); } } if (newbsize < bp->b_bufsize) bufspacewakeup(); bp->b_bufsize = newbsize; /* actual buffer allocation */ bp->b_bcount = size; /* requested buffer size */ return 1; } extern int inflight_transient_maps; void biodone(struct bio *bp) { struct mtx *mtxp; void (*done)(struct bio *); vm_offset_t start, end; if ((bp->bio_flags & BIO_TRANSIENT_MAPPING) != 0) { bp->bio_flags &= ~BIO_TRANSIENT_MAPPING; bp->bio_flags |= BIO_UNMAPPED; start = trunc_page((vm_offset_t)bp->bio_data); end = round_page((vm_offset_t)bp->bio_data + bp->bio_length); bp->bio_data = unmapped_buf; pmap_qremove(start, OFF_TO_IDX(end - start)); vmem_free(transient_arena, start, end - start); atomic_add_int(&inflight_transient_maps, -1); } done = bp->bio_done; if (done == NULL) { mtxp = mtx_pool_find(mtxpool_sleep, bp); mtx_lock(mtxp); bp->bio_flags |= BIO_DONE; wakeup(bp); mtx_unlock(mtxp); } else { bp->bio_flags |= BIO_DONE; done(bp); } } /* * Wait for a BIO to finish. */ int biowait(struct bio *bp, const char *wchan) { struct mtx *mtxp; mtxp = mtx_pool_find(mtxpool_sleep, bp); mtx_lock(mtxp); while ((bp->bio_flags & BIO_DONE) == 0) msleep(bp, mtxp, PRIBIO, wchan, 0); mtx_unlock(mtxp); if (bp->bio_error != 0) return (bp->bio_error); if (!(bp->bio_flags & BIO_ERROR)) return (0); return (EIO); } void biofinish(struct bio *bp, struct devstat *stat, int error) { if (error) { bp->bio_error = error; bp->bio_flags |= BIO_ERROR; } if (stat != NULL) devstat_end_transaction_bio(stat, bp); biodone(bp); } /* * bufwait: * * Wait for buffer I/O completion, returning error status. The buffer * is left locked and B_DONE on return. B_EINTR is converted into an EINTR * error and cleared. */ int bufwait(struct buf *bp) { if (bp->b_iocmd == BIO_READ) bwait(bp, PRIBIO, "biord"); else bwait(bp, PRIBIO, "biowr"); if (bp->b_flags & B_EINTR) { bp->b_flags &= ~B_EINTR; return (EINTR); } if (bp->b_ioflags & BIO_ERROR) { return (bp->b_error ? bp->b_error : EIO); } else { return (0); } } /* * Call back function from struct bio back up to struct buf. */ static void bufdonebio(struct bio *bip) { struct buf *bp; bp = bip->bio_caller2; bp->b_resid = bip->bio_resid; bp->b_ioflags = bip->bio_flags; bp->b_error = bip->bio_error; if (bp->b_error) bp->b_ioflags |= BIO_ERROR; bufdone(bp); g_destroy_bio(bip); } void dev_strategy(struct cdev *dev, struct buf *bp) { struct cdevsw *csw; int ref; KASSERT(dev->si_refcount > 0, ("dev_strategy on un-referenced struct cdev *(%s) %p", devtoname(dev), dev)); csw = dev_refthread(dev, &ref); dev_strategy_csw(dev, csw, bp); dev_relthread(dev, ref); } void dev_strategy_csw(struct cdev *dev, struct cdevsw *csw, struct buf *bp) { struct bio *bip; KASSERT(bp->b_iocmd == BIO_READ || bp->b_iocmd == BIO_WRITE, ("b_iocmd botch")); KASSERT(((dev->si_flags & SI_ETERNAL) != 0 && csw != NULL) || dev->si_threadcount > 0, ("dev_strategy_csw threadcount cdev *(%s) %p", devtoname(dev), dev)); if (csw == NULL) { bp->b_error = ENXIO; bp->b_ioflags = BIO_ERROR; bufdone(bp); return; } for (;;) { bip = g_new_bio(); if (bip != NULL) break; /* Try again later */ tsleep(&bp, PRIBIO, "dev_strat", hz/10); } bip->bio_cmd = bp->b_iocmd; bip->bio_offset = bp->b_iooffset; bip->bio_length = bp->b_bcount; bip->bio_bcount = bp->b_bcount; /* XXX: remove */ bdata2bio(bp, bip); bip->bio_done = bufdonebio; bip->bio_caller2 = bp; bip->bio_dev = dev; (*csw->d_strategy)(bip); } /* * bufdone: * * Finish I/O on a buffer, optionally calling a completion function. * This is usually called from an interrupt so process blocking is * not allowed. * * biodone is also responsible for setting B_CACHE in a B_VMIO bp. * In a non-VMIO bp, B_CACHE will be set on the next getblk() * assuming B_INVAL is clear. * * For the VMIO case, we set B_CACHE if the op was a read and no * read error occured, or if the op was a write. B_CACHE is never * set if the buffer is invalid or otherwise uncacheable. * * biodone does not mess with B_INVAL, allowing the I/O routine or the * initiator to leave B_INVAL set to brelse the buffer out of existance * in the biodone routine. */ void bufdone(struct buf *bp) { struct bufobj *dropobj; void (*biodone)(struct buf *); CTR3(KTR_BUF, "bufdone(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); dropobj = NULL; KASSERT(!(bp->b_flags & B_DONE), ("biodone: bp %p already done", bp)); BUF_ASSERT_HELD(bp); runningbufwakeup(bp); if (bp->b_iocmd == BIO_WRITE) dropobj = bp->b_bufobj; /* call optional completion function if requested */ if (bp->b_iodone != NULL) { biodone = bp->b_iodone; bp->b_iodone = NULL; (*biodone) (bp); if (dropobj) bufobj_wdrop(dropobj); return; } bufdone_finish(bp); if (dropobj) bufobj_wdrop(dropobj); } void bufdone_finish(struct buf *bp) { BUF_ASSERT_HELD(bp); if (!LIST_EMPTY(&bp->b_dep)) buf_complete(bp); if (bp->b_flags & B_VMIO) { vm_ooffset_t foff; vm_page_t m; vm_object_t obj; struct vnode *vp; int bogus, i, iosize; obj = bp->b_bufobj->bo_object; KASSERT(obj->paging_in_progress >= bp->b_npages, ("biodone_finish: paging in progress(%d) < b_npages(%d)", obj->paging_in_progress, bp->b_npages)); vp = bp->b_vp; KASSERT(vp->v_holdcnt > 0, ("biodone_finish: vnode %p has zero hold count", vp)); KASSERT(vp->v_object != NULL, ("biodone_finish: vnode %p has no vm_object", vp)); foff = bp->b_offset; KASSERT(bp->b_offset != NOOFFSET, ("biodone_finish: bp %p has no buffer offset", bp)); /* * Set B_CACHE if the op was a normal read and no error * occured. B_CACHE is set for writes in the b*write() * routines. */ iosize = bp->b_bcount - bp->b_resid; if (bp->b_iocmd == BIO_READ && !(bp->b_flags & (B_INVAL|B_NOCACHE)) && !(bp->b_ioflags & BIO_ERROR)) { bp->b_flags |= B_CACHE; } bogus = 0; VM_OBJECT_WLOCK(obj); for (i = 0; i < bp->b_npages; i++) { int bogusflag = 0; int resid; resid = ((foff + PAGE_SIZE) & ~(off_t)PAGE_MASK) - foff; if (resid > iosize) resid = iosize; /* * cleanup bogus pages, restoring the originals */ m = bp->b_pages[i]; if (m == bogus_page) { bogus = bogusflag = 1; m = vm_page_lookup(obj, OFF_TO_IDX(foff)); if (m == NULL) panic("biodone: page disappeared!"); bp->b_pages[i] = m; } KASSERT(OFF_TO_IDX(foff) == m->pindex, ("biodone_finish: foff(%jd)/pindex(%ju) mismatch", (intmax_t)foff, (uintmax_t)m->pindex)); /* * 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. */ if ((bp->b_iocmd == BIO_READ) && !bogusflag && resid > 0) { KASSERT((m->dirty & vm_page_bits(foff & PAGE_MASK, resid)) == 0, ("bufdone_finish:" " page %p has unexpected dirty bits", m)); vfs_page_set_valid(bp, foff, m); } vm_page_sunbusy(m); vm_object_pip_subtract(obj, 1); foff = (foff + PAGE_SIZE) & ~(off_t)PAGE_MASK; iosize -= resid; } vm_object_pip_wakeupn(obj, 0); VM_OBJECT_WUNLOCK(obj); if (bogus && (bp->b_flags & B_UNMAPPED) == 0) { BUF_CHECK_MAPPED(bp); pmap_qenter(trunc_page((vm_offset_t)bp->b_data), bp->b_pages, bp->b_npages); } } /* * For asynchronous completions, release the buffer now. The brelse * will do a wakeup there if necessary - so no need to do a wakeup * here in the async case. The sync case always needs to do a wakeup. */ if (bp->b_flags & B_ASYNC) { if ((bp->b_flags & (B_NOCACHE | B_INVAL | B_RELBUF)) || (bp->b_ioflags & BIO_ERROR)) brelse(bp); else bqrelse(bp); } else bdone(bp); } /* * This routine is called in lieu of iodone in the case of * incomplete I/O. This keeps the busy status for pages * consistant. */ void vfs_unbusy_pages(struct buf *bp) { int i; vm_object_t obj; vm_page_t m; runningbufwakeup(bp); if (!(bp->b_flags & B_VMIO)) return; obj = bp->b_bufobj->bo_object; VM_OBJECT_WLOCK(obj); for (i = 0; i < bp->b_npages; i++) { m = bp->b_pages[i]; if (m == bogus_page) { m = vm_page_lookup(obj, OFF_TO_IDX(bp->b_offset) + i); if (!m) panic("vfs_unbusy_pages: page missing\n"); bp->b_pages[i] = m; if ((bp->b_flags & B_UNMAPPED) == 0) { BUF_CHECK_MAPPED(bp); pmap_qenter(trunc_page((vm_offset_t)bp->b_data), bp->b_pages, bp->b_npages); } else BUF_CHECK_UNMAPPED(bp); } vm_object_pip_subtract(obj, 1); vm_page_sunbusy(m); } vm_object_pip_wakeupn(obj, 0); VM_OBJECT_WUNLOCK(obj); } /* * vfs_page_set_valid: * * Set the valid bits in a page based on the supplied offset. The * range is restricted to the buffer's size. * * This routine is typically called after a read completes. */ static void vfs_page_set_valid(struct buf *bp, vm_ooffset_t off, vm_page_t m) { vm_ooffset_t eoff; /* * Compute the end offset, eoff, such that [off, eoff) does not span a * page boundary and eoff is not greater than the end of the buffer. * The end of the buffer, in this case, is our file EOF, not the * allocation size of the buffer. */ eoff = (off + PAGE_SIZE) & ~(vm_ooffset_t)PAGE_MASK; if (eoff > bp->b_offset + bp->b_bcount) eoff = bp->b_offset + bp->b_bcount; /* * Set valid range. This is typically the entire buffer and thus the * entire page. */ if (eoff > off) vm_page_set_valid_range(m, off & PAGE_MASK, eoff - off); } /* * vfs_page_set_validclean: * * Set the valid bits and clear the dirty bits in a page based on the * supplied offset. The range is restricted to the buffer's size. */ static void vfs_page_set_validclean(struct buf *bp, vm_ooffset_t off, vm_page_t m) { vm_ooffset_t soff, eoff; /* * Start and end offsets in buffer. eoff - soff may not cross a * page boundry or cross the end of the buffer. The end of the * buffer, in this case, is our file EOF, not the allocation size * of the buffer. */ soff = off; eoff = (off + PAGE_SIZE) & ~(off_t)PAGE_MASK; if (eoff > bp->b_offset + bp->b_bcount) eoff = bp->b_offset + bp->b_bcount; /* * Set valid range. This is typically the entire buffer and thus the * entire page. */ if (eoff > soff) { vm_page_set_validclean( m, (vm_offset_t) (soff & PAGE_MASK), (vm_offset_t) (eoff - soff) ); } } /* * Ensure that all buffer pages are not exclusive busied. If any page is * exclusive busy, drain it. */ void vfs_drain_busy_pages(struct buf *bp) { vm_page_t m; int i, last_busied; VM_OBJECT_ASSERT_WLOCKED(bp->b_bufobj->bo_object); last_busied = 0; for (i = 0; i < bp->b_npages; i++) { m = bp->b_pages[i]; if (vm_page_xbusied(m)) { for (; last_busied < i; last_busied++) vm_page_sbusy(bp->b_pages[last_busied]); while (vm_page_xbusied(m)) { vm_page_lock(m); VM_OBJECT_WUNLOCK(bp->b_bufobj->bo_object); vm_page_busy_sleep(m, "vbpage"); VM_OBJECT_WLOCK(bp->b_bufobj->bo_object); } } } for (i = 0; i < last_busied; i++) vm_page_sunbusy(bp->b_pages[i]); } /* * This routine is called before a device strategy routine. * It is used to tell the VM system that paging I/O is in * progress, and treat the pages associated with the buffer * almost as being exclusive busy. Also the object paging_in_progress * flag is handled to make sure that the object doesn't become * inconsistant. * * Since I/O has not been initiated yet, certain buffer flags * such as BIO_ERROR or B_INVAL may be in an inconsistant state * and should be ignored. */ void vfs_busy_pages(struct buf *bp, int clear_modify) { int i, bogus; vm_object_t obj; vm_ooffset_t foff; vm_page_t m; if (!(bp->b_flags & B_VMIO)) return; obj = bp->b_bufobj->bo_object; foff = bp->b_offset; KASSERT(bp->b_offset != NOOFFSET, ("vfs_busy_pages: no buffer offset")); VM_OBJECT_WLOCK(obj); vfs_drain_busy_pages(bp); if (bp->b_bufsize != 0) vfs_setdirty_locked_object(bp); bogus = 0; for (i = 0; i < bp->b_npages; i++) { m = bp->b_pages[i]; if ((bp->b_flags & B_CLUSTER) == 0) { vm_object_pip_add(obj, 1); vm_page_sbusy(m); } /* * When readying a buffer for a read ( i.e * clear_modify == 0 ), it is important to do * bogus_page replacement for valid pages in * partially instantiated buffers. Partially * instantiated buffers can, in turn, occur when * reconstituting a buffer from its VM backing store * base. We only have to do this if B_CACHE is * clear ( which causes the I/O to occur in the * first place ). The replacement prevents the read * I/O from overwriting potentially dirty VM-backed * pages. XXX bogus page replacement is, uh, bogus. * It may not work properly with small-block devices. * We need to find a better way. */ if (clear_modify) { pmap_remove_write(m); vfs_page_set_validclean(bp, foff, m); } else if (m->valid == VM_PAGE_BITS_ALL && (bp->b_flags & B_CACHE) == 0) { bp->b_pages[i] = bogus_page; bogus++; } foff = (foff + PAGE_SIZE) & ~(off_t)PAGE_MASK; } VM_OBJECT_WUNLOCK(obj); if (bogus && (bp->b_flags & B_UNMAPPED) == 0) { 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 ((bp->b_flags & B_UNMAPPED) == 0) { BUF_CHECK_MAPPED(bp); bzero(bp->b_data + base, size); } else { BUF_CHECK_UNMAPPED(bp); n = PAGE_SIZE - (base & PAGE_MASK); for (i = base / PAGE_SIZE; size > 0 && i < bp->b_npages; ++i) { m = bp->b_pages[i]; if (n > size) n = size; pmap_zero_page_area(m, base & PAGE_MASK, n); base += n; size -= n; n = PAGE_SIZE; } } } /* * vm_hold_load_pages and vm_hold_free_pages get pages into * a buffers address space. The pages are anonymous and are * not associated with a file object. */ static void vm_hold_load_pages(struct buf *bp, vm_offset_t from, vm_offset_t to) { vm_offset_t pg; vm_page_t p; int index; BUF_CHECK_MAPPED(bp); to = round_page(to); from = round_page(from); index = (from - trunc_page((vm_offset_t)bp->b_data)) >> PAGE_SHIFT; for (pg = from; pg < to; pg += PAGE_SIZE, index++) { tryagain: /* * note: must allocate system pages since blocking here * could interfere with paging I/O, no matter which * process we are. */ p = vm_page_alloc(NULL, 0, VM_ALLOC_SYSTEM | VM_ALLOC_NOOBJ | VM_ALLOC_WIRED | VM_ALLOC_COUNT((to - pg) >> PAGE_SHIFT)); if (p == NULL) { VM_WAIT; goto tryagain; } pmap_qenter(pg, &p, 1); bp->b_pages[index] = p; } bp->b_npages = index; } /* Return pages associated with this buf to the vm system */ static void vm_hold_free_pages(struct buf *bp, int newbsize) { vm_offset_t from; vm_page_t p; int index, newnpages; BUF_CHECK_MAPPED(bp); from = round_page((vm_offset_t)bp->b_data + newbsize); newnpages = (from - trunc_page((vm_offset_t)bp->b_data)) >> PAGE_SHIFT; if (bp->b_npages > newnpages) pmap_qremove(from, bp->b_npages - newnpages); for (index = newnpages; index < bp->b_npages; index++) { p = bp->b_pages[index]; bp->b_pages[index] = NULL; if (vm_page_sbusied(p)) printf("vm_hold_free_pages: blkno: %jd, lblkno: %jd\n", (intmax_t)bp->b_blkno, (intmax_t)bp->b_lblkno); p->wire_count--; vm_page_free(p); atomic_subtract_int(&vm_cnt.v_wire_count, 1); } bp->b_npages = newnpages; } /* * Map an IO request into kernel virtual address space. * * All requests are (re)mapped into kernel VA space. * Notice that we use b_bufsize for the size of the buffer * to be mapped. b_bcount might be modified by the driver. * * Note that even if the caller determines that the address space should * be valid, a race or a smaller-file mapped into a larger space may * actually cause vmapbuf() to fail, so all callers of vmapbuf() MUST * check the return value. */ int vmapbuf(struct buf *bp, int mapbuf) { caddr_t kva; 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; if (mapbuf || !unmapped_buf_allowed) { pmap_qenter((vm_offset_t)bp->b_saveaddr, bp->b_pages, pidx); kva = bp->b_saveaddr; bp->b_saveaddr = bp->b_data; bp->b_data = kva + (((vm_offset_t)bp->b_data) & PAGE_MASK); bp->b_flags &= ~B_UNMAPPED; } else { bp->b_flags |= B_UNMAPPED; bp->b_offset = ((vm_offset_t)bp->b_data) & PAGE_MASK; bp->b_saveaddr = bp->b_data; 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. */ void vunmapbuf(struct buf *bp) { int npages; npages = bp->b_npages; if (bp->b_flags & B_UNMAPPED) bp->b_flags &= ~B_UNMAPPED; else pmap_qremove(trunc_page((vm_offset_t)bp->b_data), npages); vm_page_unhold_pages(bp->b_pages, npages); bp->b_data = bp->b_saveaddr; } void bdone(struct buf *bp) { struct mtx *mtxp; mtxp = mtx_pool_find(mtxpool_sleep, bp); mtx_lock(mtxp); bp->b_flags |= B_DONE; wakeup(bp); mtx_unlock(mtxp); } void bwait(struct buf *bp, u_char pri, const char *wchan) { struct mtx *mtxp; mtxp = mtx_pool_find(mtxpool_sleep, bp); mtx_lock(mtxp); while ((bp->b_flags & B_DONE) == 0) msleep(bp, mtxp, pri, wchan, 0); mtx_unlock(mtxp); } int bufsync(struct bufobj *bo, int waitfor) { return (VOP_FSYNC(bo->__bo_vnode, waitfor, curthread)); } void bufstrategy(struct bufobj *bo, struct buf *bp) { int i = 0; struct vnode *vp; vp = bp->b_vp; KASSERT(vp == bo->bo_private, ("Inconsistent vnode bufstrategy")); KASSERT(vp->v_type != VCHR && vp->v_type != VBLK, ("Wrong vnode in bufstrategy(bp=%p, vp=%p)", bp, vp)); i = VOP_STRATEGY(vp, bp); KASSERT(i == 0, ("VOP_STRATEGY failed bp=%p vp=%p", bp, bp->b_vp)); } void bufobj_wrefl(struct bufobj *bo) { KASSERT(bo != NULL, ("NULL bo in bufobj_wref")); ASSERT_BO_WLOCKED(bo); bo->bo_numoutput++; } void bufobj_wref(struct bufobj *bo) { KASSERT(bo != NULL, ("NULL bo in bufobj_wref")); BO_LOCK(bo); bo->bo_numoutput++; BO_UNLOCK(bo); } void bufobj_wdrop(struct bufobj *bo) { KASSERT(bo != NULL, ("NULL bo in bufobj_wdrop")); BO_LOCK(bo); KASSERT(bo->bo_numoutput > 0, ("bufobj_wdrop non-positive count")); if ((--bo->bo_numoutput == 0) && (bo->bo_flag & BO_WWAIT)) { bo->bo_flag &= ~BO_WWAIT; wakeup(&bo->bo_numoutput); } BO_UNLOCK(bo); } int bufobj_wwait(struct bufobj *bo, int slpflag, int timeo) { int error; KASSERT(bo != NULL, ("NULL bo in bufobj_wwait")); ASSERT_BO_WLOCKED(bo); error = 0; while (bo->bo_numoutput) { bo->bo_flag |= BO_WWAIT; error = msleep(&bo->bo_numoutput, BO_LOCKPTR(bo), slpflag | (PRIBIO + 1), "bo_wwait", timeo); if (error) break; } return (error); } void bpin(struct buf *bp) { struct mtx *mtxp; mtxp = mtx_pool_find(mtxpool_sleep, bp); mtx_lock(mtxp); bp->b_pin_count++; mtx_unlock(mtxp); } void bunpin(struct buf *bp) { struct mtx *mtxp; mtxp = mtx_pool_find(mtxpool_sleep, bp); mtx_lock(mtxp); if (--bp->b_pin_count == 0) wakeup(bp); mtx_unlock(mtxp); } void bunpin_wait(struct buf *bp) { struct mtx *mtxp; mtxp = mtx_pool_find(mtxpool_sleep, bp); mtx_lock(mtxp); while (bp->b_pin_count > 0) msleep(bp, mtxp, PRIBIO, "bwunpin", 0); mtx_unlock(mtxp); } /* * Set bio_data or bio_ma for struct bio from the struct buf. */ void bdata2bio(struct buf *bp, struct bio *bip) { if ((bp->b_flags & B_UNMAPPED) != 0) { KASSERT(unmapped_buf_allowed, ("unmapped")); bip->bio_ma = bp->b_pages; bip->bio_ma_n = bp->b_npages; bip->bio_data = unmapped_buf; bip->bio_ma_offset = (vm_offset_t)bp->b_offset & PAGE_MASK; bip->bio_flags |= BIO_UNMAPPED; KASSERT(round_page(bip->bio_ma_offset + bip->bio_length) / PAGE_SIZE == bp->b_npages, ("Buffer %p too short: %d %lld %d", bp, bip->bio_ma_offset, (long long)bip->bio_length, bip->bio_ma_n)); } else { bip->bio_data = bp->b_data; bip->bio_ma = NULL; } } #include "opt_ddb.h" #ifdef DDB #include /* DDB command to show buffer data */ DB_SHOW_COMMAND(buffer, db_show_buffer) { /* get args */ struct buf *bp = (struct buf *)addr; if (!have_addr) { db_printf("usage: show buffer \n"); return; } db_printf("buf at %p\n", bp); db_printf("b_flags = 0x%b, b_xflags=0x%b, b_vflags=0x%b\n", (u_int)bp->b_flags, PRINT_BUF_FLAGS, (u_int)bp->b_xflags, PRINT_BUF_XFLAGS, (u_int)bp->b_vflags, PRINT_BUF_VFLAGS); db_printf( "b_error = %d, b_bufsize = %ld, b_bcount = %ld, b_resid = %ld\n" "b_bufobj = (%p), b_data = %p, b_blkno = %jd, b_lblkno = %jd, " "b_dep = %p\n", bp->b_error, bp->b_bufsize, bp->b_bcount, bp->b_resid, bp->b_bufobj, bp->b_data, (intmax_t)bp->b_blkno, (intmax_t)bp->b_lblkno, bp->b_dep.lh_first); if (bp->b_npages) { int i; db_printf("b_npages = %d, pages(OBJ, IDX, PA): ", bp->b_npages); for (i = 0; i < bp->b_npages; i++) { vm_page_t m; m = bp->b_pages[i]; db_printf("(%p, 0x%lx, 0x%lx)", (void *)m->object, (u_long)m->pindex, (u_long)VM_PAGE_TO_PHYS(m)); if ((i + 1) < bp->b_npages) db_printf(","); } db_printf("\n"); } db_printf(" "); BUF_LOCKPRINTINFO(bp); } DB_SHOW_COMMAND(lockedbufs, lockedbufs) { struct buf *bp; int i; for (i = 0; i < nbuf; i++) { bp = &buf[i]; if (BUF_ISLOCKED(bp)) { db_show_buffer((uintptr_t)bp, 1, 0, NULL); db_printf("\n"); } } } DB_SHOW_COMMAND(vnodebufs, db_show_vnodebufs) { struct vnode *vp; struct buf *bp; if (!have_addr) { db_printf("usage: show vnodebufs \n"); return; } vp = (struct vnode *)addr; db_printf("Clean buffers:\n"); TAILQ_FOREACH(bp, &vp->v_bufobj.bo_clean.bv_hd, b_bobufs) { db_show_buffer((uintptr_t)bp, 1, 0, NULL); db_printf("\n"); } db_printf("Dirty buffers:\n"); TAILQ_FOREACH(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs) { db_show_buffer((uintptr_t)bp, 1, 0, NULL); db_printf("\n"); } } DB_COMMAND(countfreebufs, db_coundfreebufs) { struct buf *bp; int i, used = 0, nfree = 0; if (have_addr) { db_printf("usage: countfreebufs\n"); return; } for (i = 0; i < nbuf; i++) { bp = &buf[i]; if ((bp->b_flags & B_INFREECNT) != 0) nfree++; else used++; } db_printf("Counted %d free, %d used (%d tot)\n", nfree, used, nfree + used); db_printf("numfreebuffers is %d\n", numfreebuffers); } #endif /* DDB */ Index: head/sys/sys/param.h =================================================================== --- head/sys/sys/param.h (revision 281959) +++ head/sys/sys/param.h (revision 281960) @@ -1,348 +1,362 @@ /*- * Copyright (c) 1982, 1986, 1989, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)param.h 8.3 (Berkeley) 4/4/95 * $FreeBSD$ */ #ifndef _SYS_PARAM_H_ #define _SYS_PARAM_H_ #include #define BSD 199506 /* System version (year & month). */ #define BSD4_3 1 #define BSD4_4 1 /* * __FreeBSD_version numbers are documented in the Porter's Handbook. * If you bump the version for any reason, you should update the documentation * there. * Currently this lives here in the doc/ repository: * * head/en_US.ISO8859-1/books/porters-handbook/versions/chapter.xml * * scheme is: Rxx * 'R' is in the range 0 to 4 if this is a release branch or * x.0-CURRENT before RELENG_*_0 is created, otherwise 'R' is * in the range 5 to 9. */ #undef __FreeBSD_version #define __FreeBSD_version 1100070 /* Master, propagated to newvers */ /* * __FreeBSD_kernel__ indicates that this system uses the kernel of FreeBSD, * which by definition is always true on FreeBSD. This macro is also defined * on other systems that use the kernel of FreeBSD, such as GNU/kFreeBSD. * * It is tempting to use this macro in userland code when we want to enable * kernel-specific routines, and in fact it's fine to do this in code that * is part of FreeBSD itself. However, be aware that as presence of this * macro is still not widespread (e.g. older FreeBSD versions, 3rd party * compilers, etc), it is STRONGLY DISCOURAGED to check for this macro in * external applications without also checking for __FreeBSD__ as an * alternative. */ #undef __FreeBSD_kernel__ #define __FreeBSD_kernel__ #ifdef _KERNEL #define P_OSREL_SIGWAIT 700000 #define P_OSREL_SIGSEGV 700004 #define P_OSREL_MAP_ANON 800104 #define P_OSREL_MAP_FSTRICT 1100036 #define P_OSREL_MAJOR(x) ((x) / 100000) #endif #ifndef LOCORE #include #endif /* * Machine-independent constants (some used in following include files). * Redefined constants are from POSIX 1003.1 limits file. * * MAXCOMLEN should be >= sizeof(ac_comm) (see ) */ #include #define MAXCOMLEN 19 /* max command name remembered */ #define MAXINTERP PATH_MAX /* max interpreter file name length */ #define MAXLOGNAME 33 /* max login name length (incl. NUL) */ #define MAXUPRC CHILD_MAX /* max simultaneous processes */ #define NCARGS ARG_MAX /* max bytes for an exec function */ #define NGROUPS (NGROUPS_MAX+1) /* max number groups */ #define NOFILE OPEN_MAX /* max open files per process */ #define NOGROUP 65535 /* marker for empty group set member */ #define MAXHOSTNAMELEN 256 /* max hostname size */ #define SPECNAMELEN 63 /* max length of devicename */ /* More types and definitions used throughout the kernel. */ #ifdef _KERNEL #include #include #ifndef LOCORE #include #include #endif #ifndef FALSE #define FALSE 0 #endif #ifndef TRUE #define TRUE 1 #endif #endif #ifndef _KERNEL /* Signals. */ #include #endif /* Machine type dependent parameters. */ #include #ifndef _KERNEL #include #endif #ifndef DEV_BSHIFT #define DEV_BSHIFT 9 /* log2(DEV_BSIZE) */ #endif #define DEV_BSIZE (1<>PAGE_SHIFT) #endif /* * btodb() is messy and perhaps slow because `bytes' may be an off_t. We * want to shift an unsigned type to avoid sign extension and we don't * want to widen `bytes' unnecessarily. Assume that the result fits in * a daddr_t. */ #ifndef btodb #define btodb(bytes) /* calculates (bytes / DEV_BSIZE) */ \ (sizeof (bytes) > sizeof(long) \ ? (daddr_t)((unsigned long long)(bytes) >> DEV_BSHIFT) \ : (daddr_t)((unsigned long)(bytes) >> DEV_BSHIFT)) #endif #ifndef dbtob #define dbtob(db) /* calculates (db * DEV_BSIZE) */ \ ((off_t)(db) << DEV_BSHIFT) #endif #define PRIMASK 0x0ff #define PCATCH 0x100 /* OR'd with pri for tsleep to check signals */ #define PDROP 0x200 /* OR'd with pri to stop re-entry of interlock mutex */ #define NZERO 0 /* default "nice" */ #define NBBY 8 /* number of bits in a byte */ #define NBPW sizeof(int) /* number of bytes per word (integer) */ #define CMASK 022 /* default file mask: S_IWGRP|S_IWOTH */ #define NODEV (dev_t)(-1) /* non-existent device */ /* * File system parameters and macros. * * MAXBSIZE - Filesystems are made out of blocks of at most MAXBSIZE bytes * per block. MAXBSIZE may be made larger without effecting * any existing filesystems as long as it does not exceed MAXPHYS, * and may be made smaller at the risk of not being able to use * filesystems which require a block size exceeding MAXBSIZE. * + * MAXBCACHEBUF - Maximum size of a buffer in the buffer cache. This must + * be >= MAXBSIZE and can be set differently for different + * architectures by defining it in . + * Making this larger allows NFS to do larger reads/writes. + * * BKVASIZE - Nominal buffer space per buffer, in bytes. BKVASIZE is the * minimum KVM memory reservation the kernel is willing to make. * Filesystems can of course request smaller chunks. Actual * backing memory uses a chunk size of a page (PAGE_SIZE). + * The default value here can be overridden on a per-architecture + * basis by defining it in . This should + * probably be done to increase its value, when MAXBCACHEBUF is + * defined as a larger value in . * * If you make BKVASIZE too small you risk seriously fragmenting * the buffer KVM map which may slow things down a bit. If you * make it too big the kernel will not be able to optimally use * the KVM memory reserved for the buffer cache and will wind * up with too-few buffers. * * The default is 16384, roughly 2x the block size used by a * normal UFS filesystem. */ #define MAXBSIZE 65536 /* must be power of 2 */ +#ifndef MAXBCACHEBUF +#define MAXBCACHEBUF MAXBSIZE /* must be a power of 2 >= MAXBSIZE */ +#endif +#ifndef BKVASIZE #define BKVASIZE 16384 /* must be power of 2 */ +#endif #define BKVAMASK (BKVASIZE-1) /* * MAXPATHLEN defines the longest permissible path length after expanding * symbolic links. It is used to allocate a temporary buffer from the buffer * pool in which to do the name expansion, hence should be a power of two, * and must be less than or equal to MAXBSIZE. MAXSYMLINKS defines the * maximum number of symbolic links that may be expanded in a path name. * It should be set high enough to allow all legitimate uses, but halt * infinite loops reasonably quickly. */ #define MAXPATHLEN PATH_MAX #define MAXSYMLINKS 32 /* Bit map related macros. */ #define setbit(a,i) (((unsigned char *)(a))[(i)/NBBY] |= 1<<((i)%NBBY)) #define clrbit(a,i) (((unsigned char *)(a))[(i)/NBBY] &= ~(1<<((i)%NBBY))) #define isset(a,i) \ (((const unsigned char *)(a))[(i)/NBBY] & (1<<((i)%NBBY))) #define isclr(a,i) \ ((((const unsigned char *)(a))[(i)/NBBY] & (1<<((i)%NBBY))) == 0) /* Macros for counting and rounding. */ #ifndef howmany #define howmany(x, y) (((x)+((y)-1))/(y)) #endif #define nitems(x) (sizeof((x)) / sizeof((x)[0])) #define rounddown(x, y) (((x)/(y))*(y)) #define rounddown2(x, y) ((x)&(~((y)-1))) /* if y is power of two */ #define roundup(x, y) ((((x)+((y)-1))/(y))*(y)) /* to any y */ #define roundup2(x, y) (((x)+((y)-1))&(~((y)-1))) /* if y is powers of two */ #define powerof2(x) ((((x)-1)&(x))==0) /* Macros for min/max. */ #define MIN(a,b) (((a)<(b))?(a):(b)) #define MAX(a,b) (((a)>(b))?(a):(b)) #ifdef _KERNEL /* * Basic byte order function prototypes for non-inline functions. */ #ifndef LOCORE #ifndef _BYTEORDER_PROTOTYPED #define _BYTEORDER_PROTOTYPED __BEGIN_DECLS __uint32_t htonl(__uint32_t); __uint16_t htons(__uint16_t); __uint32_t ntohl(__uint32_t); __uint16_t ntohs(__uint16_t); __END_DECLS #endif #endif #ifndef lint #ifndef _BYTEORDER_FUNC_DEFINED #define _BYTEORDER_FUNC_DEFINED #define htonl(x) __htonl(x) #define htons(x) __htons(x) #define ntohl(x) __ntohl(x) #define ntohs(x) __ntohs(x) #endif /* !_BYTEORDER_FUNC_DEFINED */ #endif /* lint */ #endif /* _KERNEL */ /* * Scale factor for scaled integers used to count %cpu time and load avgs. * * The number of CPU `tick's that map to a unique `%age' can be expressed * by the formula (1 / (2 ^ (FSHIFT - 11))). The maximum load average that * can be calculated (assuming 32 bits) can be closely approximated using * the formula (2 ^ (2 * (16 - FSHIFT))) for (FSHIFT < 15). * * For the scheduler to maintain a 1:1 mapping of CPU `tick' to `%age', * FSHIFT must be at least 11; this gives us a maximum load avg of ~1024. */ #define FSHIFT 11 /* bits to right of fixed binary point */ #define FSCALE (1<> (PAGE_SHIFT - DEV_BSHIFT)) #define ctodb(db) /* calculates pages to devblks */ \ ((db) << (PAGE_SHIFT - DEV_BSHIFT)) /* * Old spelling of __containerof(). */ #define member2struct(s, m, x) \ ((struct s *)(void *)((char *)(x) - offsetof(struct s, m))) /* * Access a variable length array that has been declared as a fixed * length array. */ #define __PAST_END(array, offset) (((__typeof__(*(array)) *)(array))[offset]) #endif /* _SYS_PARAM_H_ */