Index: stable/10/sys/cddl/contrib/opensolaris/uts/common/dtrace/fasttrap.c =================================================================== --- stable/10/sys/cddl/contrib/opensolaris/uts/common/dtrace/fasttrap.c (revision 271000) +++ stable/10/sys/cddl/contrib/opensolaris/uts/common/dtrace/fasttrap.c (revision 271001) @@ -1,2724 +1,2724 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END * * Portions Copyright 2010 The FreeBSD Foundation * * $FreeBSD$ */ /* * Copyright 2008 Sun Microsystems, Inc. All rights reserved. * Use is subject to license terms. */ /* * Copyright (c) 2013, Joyent, Inc. All rights reserved. */ #include #include #include #include #include #include #if defined(sun) #include #endif #include #include #include #if defined(sun) #include #endif #include #include #include #include #include #include #include #include #if defined(sun) #include #endif #include #include #if !defined(sun) #include #include #include #include #include #include #include #include #include #endif /* * User-Land Trap-Based Tracing * ---------------------------- * * The fasttrap provider allows DTrace consumers to instrument any user-level * instruction to gather data; this includes probes with semantic * signifigance like entry and return as well as simple offsets into the * function. While the specific techniques used are very ISA specific, the * methodology is generalizable to any architecture. * * * The General Methodology * ----------------------- * * With the primary goal of tracing every user-land instruction and the * limitation that we can't trust user space so don't want to rely on much * information there, we begin by replacing the instructions we want to trace * with trap instructions. Each instruction we overwrite is saved into a hash * table keyed by process ID and pc address. When we enter the kernel due to * this trap instruction, we need the effects of the replaced instruction to * appear to have occurred before we proceed with the user thread's * execution. * * Each user level thread is represented by a ulwp_t structure which is * always easily accessible through a register. The most basic way to produce * the effects of the instruction we replaced is to copy that instruction out * to a bit of scratch space reserved in the user thread's ulwp_t structure * (a sort of kernel-private thread local storage), set the PC to that * scratch space and single step. When we reenter the kernel after single * stepping the instruction we must then adjust the PC to point to what would * normally be the next instruction. Of course, special care must be taken * for branches and jumps, but these represent such a small fraction of any * instruction set that writing the code to emulate these in the kernel is * not too difficult. * * Return probes may require several tracepoints to trace every return site, * and, conversely, each tracepoint may activate several probes (the entry * and offset 0 probes, for example). To solve this muliplexing problem, * tracepoints contain lists of probes to activate and probes contain lists * of tracepoints to enable. If a probe is activated, it adds its ID to * existing tracepoints or creates new ones as necessary. * * Most probes are activated _before_ the instruction is executed, but return * probes are activated _after_ the effects of the last instruction of the * function are visible. Return probes must be fired _after_ we have * single-stepped the instruction whereas all other probes are fired * beforehand. * * * Lock Ordering * ------------- * * The lock ordering below -- both internally and with respect to the DTrace * framework -- is a little tricky and bears some explanation. Each provider * has a lock (ftp_mtx) that protects its members including reference counts * for enabled probes (ftp_rcount), consumers actively creating probes * (ftp_ccount) and USDT consumers (ftp_mcount); all three prevent a provider * from being freed. A provider is looked up by taking the bucket lock for the * provider hash table, and is returned with its lock held. The provider lock * may be taken in functions invoked by the DTrace framework, but may not be * held while calling functions in the DTrace framework. * * To ensure consistency over multiple calls to the DTrace framework, the * creation lock (ftp_cmtx) should be held. Naturally, the creation lock may * not be taken when holding the provider lock as that would create a cyclic * lock ordering. In situations where one would naturally take the provider * lock and then the creation lock, we instead up a reference count to prevent * the provider from disappearing, drop the provider lock, and acquire the * creation lock. * * Briefly: * bucket lock before provider lock * DTrace before provider lock * creation lock before DTrace * never hold the provider lock and creation lock simultaneously */ static d_open_t fasttrap_open; static d_ioctl_t fasttrap_ioctl; static struct cdevsw fasttrap_cdevsw = { .d_version = D_VERSION, .d_open = fasttrap_open, .d_ioctl = fasttrap_ioctl, .d_name = "fasttrap", }; static struct cdev *fasttrap_cdev; static dtrace_meta_provider_id_t fasttrap_meta_id; static struct proc *fasttrap_cleanup_proc; static struct mtx fasttrap_cleanup_mtx; static uint_t fasttrap_cleanup_work, fasttrap_cleanup_drain, fasttrap_cleanup_cv; /* * Generation count on modifications to the global tracepoint lookup table. */ static volatile uint64_t fasttrap_mod_gen; /* * When the fasttrap provider is loaded, fasttrap_max is set to either * FASTTRAP_MAX_DEFAULT or the value for fasttrap-max-probes in the * fasttrap.conf file. Each time a probe is created, fasttrap_total is * incremented by the number of tracepoints that may be associated with that * probe; fasttrap_total is capped at fasttrap_max. */ #define FASTTRAP_MAX_DEFAULT 250000 static uint32_t fasttrap_max; static uint32_t fasttrap_total; /* * Copyright (c) 2011, Joyent, Inc. All rights reserved. */ #define FASTTRAP_TPOINTS_DEFAULT_SIZE 0x4000 #define FASTTRAP_PROVIDERS_DEFAULT_SIZE 0x100 #define FASTTRAP_PROCS_DEFAULT_SIZE 0x100 #define FASTTRAP_PID_NAME "pid" fasttrap_hash_t fasttrap_tpoints; static fasttrap_hash_t fasttrap_provs; static fasttrap_hash_t fasttrap_procs; static uint64_t fasttrap_pid_count; /* pid ref count */ static kmutex_t fasttrap_count_mtx; /* lock on ref count */ #define FASTTRAP_ENABLE_FAIL 1 #define FASTTRAP_ENABLE_PARTIAL 2 static int fasttrap_tracepoint_enable(proc_t *, fasttrap_probe_t *, uint_t); static void fasttrap_tracepoint_disable(proc_t *, fasttrap_probe_t *, uint_t); static fasttrap_provider_t *fasttrap_provider_lookup(pid_t, const char *, const dtrace_pattr_t *); static void fasttrap_provider_retire(pid_t, const char *, int); static void fasttrap_provider_free(fasttrap_provider_t *); static fasttrap_proc_t *fasttrap_proc_lookup(pid_t); static void fasttrap_proc_release(fasttrap_proc_t *); #if !defined(sun) static void fasttrap_thread_dtor(void *, struct thread *); #endif #define FASTTRAP_PROVS_INDEX(pid, name) \ ((fasttrap_hash_str(name) + (pid)) & fasttrap_provs.fth_mask) #define FASTTRAP_PROCS_INDEX(pid) ((pid) & fasttrap_procs.fth_mask) #if !defined(sun) static kmutex_t fasttrap_cpuc_pid_lock[MAXCPU]; static eventhandler_tag fasttrap_thread_dtor_tag; #endif static int fasttrap_highbit(ulong_t i) { int h = 1; if (i == 0) return (0); #ifdef _LP64 if (i & 0xffffffff00000000ul) { h += 32; i >>= 32; } #endif if (i & 0xffff0000) { h += 16; i >>= 16; } if (i & 0xff00) { h += 8; i >>= 8; } if (i & 0xf0) { h += 4; i >>= 4; } if (i & 0xc) { h += 2; i >>= 2; } if (i & 0x2) { h += 1; } return (h); } static uint_t fasttrap_hash_str(const char *p) { unsigned int g; uint_t hval = 0; while (*p) { hval = (hval << 4) + *p++; if ((g = (hval & 0xf0000000)) != 0) hval ^= g >> 24; hval &= ~g; } return (hval); } void fasttrap_sigtrap(proc_t *p, kthread_t *t, uintptr_t pc) { #if defined(sun) sigqueue_t *sqp = kmem_zalloc(sizeof (sigqueue_t), KM_SLEEP); sqp->sq_info.si_signo = SIGTRAP; sqp->sq_info.si_code = TRAP_DTRACE; sqp->sq_info.si_addr = (caddr_t)pc; mutex_enter(&p->p_lock); sigaddqa(p, t, sqp); mutex_exit(&p->p_lock); if (t != NULL) aston(t); #else ksiginfo_t *ksi = kmem_zalloc(sizeof (ksiginfo_t), KM_SLEEP); ksiginfo_init(ksi); ksi->ksi_signo = SIGTRAP; ksi->ksi_code = TRAP_DTRACE; ksi->ksi_addr = (caddr_t)pc; PROC_LOCK(p); (void) tdksignal(t, SIGTRAP, ksi); PROC_UNLOCK(p); #endif } #if !defined(sun) /* * Obtain a chunk of scratch space in the address space of the target process. */ fasttrap_scrspace_t * fasttrap_scraddr(struct thread *td, fasttrap_proc_t *fprc) { fasttrap_scrblock_t *scrblk; fasttrap_scrspace_t *scrspc; struct proc *p; vm_offset_t addr; int error, i; scrspc = NULL; if (td->t_dtrace_sscr != NULL) { /* If the thread already has scratch space, we're done. */ scrspc = (fasttrap_scrspace_t *)td->t_dtrace_sscr; return (scrspc); } p = td->td_proc; mutex_enter(&fprc->ftpc_mtx); if (LIST_EMPTY(&fprc->ftpc_fscr)) { /* * No scratch space is available, so we'll map a new scratch * space block into the traced process' address space. */ addr = 0; error = vm_map_find(&p->p_vmspace->vm_map, NULL, 0, &addr, FASTTRAP_SCRBLOCK_SIZE, 0, VMFS_ANY_SPACE, VM_PROT_ALL, VM_PROT_ALL, 0); if (error != KERN_SUCCESS) goto done; scrblk = malloc(sizeof(*scrblk), M_SOLARIS, M_WAITOK); scrblk->ftsb_addr = addr; LIST_INSERT_HEAD(&fprc->ftpc_scrblks, scrblk, ftsb_next); /* * Carve the block up into chunks and put them on the free list. */ for (i = 0; i < FASTTRAP_SCRBLOCK_SIZE / FASTTRAP_SCRSPACE_SIZE; i++) { scrspc = malloc(sizeof(*scrspc), M_SOLARIS, M_WAITOK); scrspc->ftss_addr = addr + i * FASTTRAP_SCRSPACE_SIZE; LIST_INSERT_HEAD(&fprc->ftpc_fscr, scrspc, ftss_next); } } /* * Take the first scratch chunk off the free list, put it on the * allocated list, and return its address. */ scrspc = LIST_FIRST(&fprc->ftpc_fscr); LIST_REMOVE(scrspc, ftss_next); LIST_INSERT_HEAD(&fprc->ftpc_ascr, scrspc, ftss_next); /* * This scratch space is reserved for use by td until the thread exits. */ td->t_dtrace_sscr = scrspc; done: mutex_exit(&fprc->ftpc_mtx); return (scrspc); } /* * Return any allocated per-thread scratch space chunks back to the process' * free list. */ static void fasttrap_thread_dtor(void *arg __unused, struct thread *td) { fasttrap_bucket_t *bucket; fasttrap_proc_t *fprc; fasttrap_scrspace_t *scrspc; pid_t pid; if (td->t_dtrace_sscr == NULL) return; pid = td->td_proc->p_pid; bucket = &fasttrap_procs.fth_table[FASTTRAP_PROCS_INDEX(pid)]; fprc = NULL; /* Look up the fasttrap process handle for this process. */ mutex_enter(&bucket->ftb_mtx); for (fprc = bucket->ftb_data; fprc != NULL; fprc = fprc->ftpc_next) { if (fprc->ftpc_pid == pid) { mutex_enter(&fprc->ftpc_mtx); mutex_exit(&bucket->ftb_mtx); break; } } if (fprc == NULL) { mutex_exit(&bucket->ftb_mtx); return; } scrspc = (fasttrap_scrspace_t *)td->t_dtrace_sscr; LIST_REMOVE(scrspc, ftss_next); LIST_INSERT_HEAD(&fprc->ftpc_fscr, scrspc, ftss_next); mutex_exit(&fprc->ftpc_mtx); } #endif /* * This function ensures that no threads are actively using the memory * associated with probes that were formerly live. */ static void fasttrap_mod_barrier(uint64_t gen) { int i; if (gen < fasttrap_mod_gen) return; fasttrap_mod_gen++; CPU_FOREACH(i) { mutex_enter(&fasttrap_cpuc_pid_lock[i]); mutex_exit(&fasttrap_cpuc_pid_lock[i]); } } /* * This function performs asynchronous cleanup of fasttrap providers. The * Solaris implementation of this mechanism use a timeout that's activated in * fasttrap_pid_cleanup(), but this doesn't work in FreeBSD: one may sleep while * holding the DTrace mutexes, but it is unsafe to sleep in a callout handler. * Thus we use a dedicated process to perform the cleanup when requested. */ /*ARGSUSED*/ static void fasttrap_pid_cleanup_cb(void *data) { fasttrap_provider_t **fpp, *fp; fasttrap_bucket_t *bucket; dtrace_provider_id_t provid; int i, later = 0, rval; mtx_lock(&fasttrap_cleanup_mtx); while (!fasttrap_cleanup_drain || later > 0) { fasttrap_cleanup_work = 0; mtx_unlock(&fasttrap_cleanup_mtx); later = 0; /* * Iterate over all the providers trying to remove the marked * ones. If a provider is marked but not retired, we just * have to take a crack at removing it -- it's no big deal if * we can't. */ for (i = 0; i < fasttrap_provs.fth_nent; i++) { bucket = &fasttrap_provs.fth_table[i]; mutex_enter(&bucket->ftb_mtx); fpp = (fasttrap_provider_t **)&bucket->ftb_data; while ((fp = *fpp) != NULL) { if (!fp->ftp_marked) { fpp = &fp->ftp_next; continue; } mutex_enter(&fp->ftp_mtx); /* * If this provider has consumers actively * creating probes (ftp_ccount) or is a USDT * provider (ftp_mcount), we can't unregister * or even condense. */ if (fp->ftp_ccount != 0 || fp->ftp_mcount != 0) { mutex_exit(&fp->ftp_mtx); fp->ftp_marked = 0; continue; } if (!fp->ftp_retired || fp->ftp_rcount != 0) fp->ftp_marked = 0; mutex_exit(&fp->ftp_mtx); /* * If we successfully unregister this * provider we can remove it from the hash * chain and free the memory. If our attempt * to unregister fails and this is a retired * provider, increment our flag to try again * pretty soon. If we've consumed more than * half of our total permitted number of * probes call dtrace_condense() to try to * clean out the unenabled probes. */ provid = fp->ftp_provid; if ((rval = dtrace_unregister(provid)) != 0) { if (fasttrap_total > fasttrap_max / 2) (void) dtrace_condense(provid); if (rval == EAGAIN) fp->ftp_marked = 1; later += fp->ftp_marked; fpp = &fp->ftp_next; } else { *fpp = fp->ftp_next; fasttrap_provider_free(fp); } } mutex_exit(&bucket->ftb_mtx); } mtx_lock(&fasttrap_cleanup_mtx); /* * If we were unable to retire a provider, try again after a * second. This situation can occur in certain circumstances * where providers cannot be unregistered even though they have * no probes enabled because of an execution of dtrace -l or * something similar. */ if (later > 0 || fasttrap_cleanup_work || fasttrap_cleanup_drain) { mtx_unlock(&fasttrap_cleanup_mtx); pause("ftclean", hz); mtx_lock(&fasttrap_cleanup_mtx); } else mtx_sleep(&fasttrap_cleanup_cv, &fasttrap_cleanup_mtx, 0, "ftcl", 0); } /* * Wake up the thread in fasttrap_unload() now that we're done. */ wakeup(&fasttrap_cleanup_drain); mtx_unlock(&fasttrap_cleanup_mtx); kthread_exit(); } /* * Activates the asynchronous cleanup mechanism. */ static void fasttrap_pid_cleanup(void) { mtx_lock(&fasttrap_cleanup_mtx); if (!fasttrap_cleanup_work) { fasttrap_cleanup_work = 1; wakeup(&fasttrap_cleanup_cv); } mtx_unlock(&fasttrap_cleanup_mtx); } /* * This is called from cfork() via dtrace_fasttrap_fork(). The child * process's address space is (roughly) a copy of the parent process's so * we have to remove all the instrumentation we had previously enabled in the * parent. */ static void fasttrap_fork(proc_t *p, proc_t *cp) { #if !defined(sun) fasttrap_scrblock_t *scrblk; fasttrap_proc_t *fprc = NULL; #endif pid_t ppid = p->p_pid; int i; #if defined(sun) ASSERT(curproc == p); ASSERT(p->p_proc_flag & P_PR_LOCK); #else PROC_LOCK_ASSERT(p, MA_OWNED); #endif #if defined(sun) ASSERT(p->p_dtrace_count > 0); #else if (p->p_dtrace_helpers) { /* * dtrace_helpers_duplicate() allocates memory. */ _PHOLD(cp); PROC_UNLOCK(p); PROC_UNLOCK(cp); dtrace_helpers_duplicate(p, cp); PROC_LOCK(cp); PROC_LOCK(p); _PRELE(cp); } /* * This check is purposely here instead of in kern_fork.c because, * for legal resons, we cannot include the dtrace_cddl.h header * inside kern_fork.c and insert if-clause there. */ if (p->p_dtrace_count == 0) return; #endif ASSERT(cp->p_dtrace_count == 0); /* * This would be simpler and faster if we maintained per-process * hash tables of enabled tracepoints. It could, however, potentially * slow down execution of a tracepoint since we'd need to go * through two levels of indirection. In the future, we should * consider either maintaining per-process ancillary lists of * enabled tracepoints or hanging a pointer to a per-process hash * table of enabled tracepoints off the proc structure. */ /* * We don't have to worry about the child process disappearing * because we're in fork(). */ #if defined(sun) mtx_lock_spin(&cp->p_slock); sprlock_proc(cp); mtx_unlock_spin(&cp->p_slock); #else /* * fasttrap_tracepoint_remove() expects the child process to be * unlocked and the VM then expects curproc to be unlocked. */ _PHOLD(cp); PROC_UNLOCK(cp); PROC_UNLOCK(p); #endif /* * Iterate over every tracepoint looking for ones that belong to the * parent process, and remove each from the child process. */ for (i = 0; i < fasttrap_tpoints.fth_nent; i++) { fasttrap_tracepoint_t *tp; fasttrap_bucket_t *bucket = &fasttrap_tpoints.fth_table[i]; mutex_enter(&bucket->ftb_mtx); for (tp = bucket->ftb_data; tp != NULL; tp = tp->ftt_next) { if (tp->ftt_pid == ppid && tp->ftt_proc->ftpc_acount != 0) { int ret = fasttrap_tracepoint_remove(cp, tp); ASSERT(ret == 0); /* * The count of active providers can only be * decremented (i.e. to zero) during exec, * exit, and removal of a meta provider so it * should be impossible to drop the count * mid-fork. */ ASSERT(tp->ftt_proc->ftpc_acount != 0); #if !defined(sun) fprc = tp->ftt_proc; #endif } } mutex_exit(&bucket->ftb_mtx); #if !defined(sun) /* * Unmap any scratch space inherited from the parent's address * space. */ if (fprc != NULL) { mutex_enter(&fprc->ftpc_mtx); LIST_FOREACH(scrblk, &fprc->ftpc_scrblks, ftsb_next) { vm_map_remove(&cp->p_vmspace->vm_map, scrblk->ftsb_addr, scrblk->ftsb_addr + FASTTRAP_SCRBLOCK_SIZE); } mutex_exit(&fprc->ftpc_mtx); } #endif } #if defined(sun) mutex_enter(&cp->p_lock); sprunlock(cp); #else PROC_LOCK(p); PROC_LOCK(cp); _PRELE(cp); #endif } /* * This is called from proc_exit() or from exec_common() if p_dtrace_probes * is set on the proc structure to indicate that there is a pid provider * associated with this process. */ static void fasttrap_exec_exit(proc_t *p) { #if !defined(sun) struct thread *td; #endif #if defined(sun) ASSERT(p == curproc); #else PROC_LOCK_ASSERT(p, MA_OWNED); _PHOLD(p); /* * Since struct threads may be recycled, we cannot rely on t_dtrace_sscr * fields to be zeroed by kdtrace_thread_ctor. Thus we must zero it * ourselves when a process exits. */ FOREACH_THREAD_IN_PROC(p, td) td->t_dtrace_sscr = NULL; PROC_UNLOCK(p); #endif /* * We clean up the pid provider for this process here; user-land * static probes are handled by the meta-provider remove entry point. */ fasttrap_provider_retire(p->p_pid, FASTTRAP_PID_NAME, 0); #if !defined(sun) if (p->p_dtrace_helpers) dtrace_helpers_destroy(p); PROC_LOCK(p); _PRELE(p); #endif } /*ARGSUSED*/ static void fasttrap_pid_provide(void *arg, dtrace_probedesc_t *desc) { /* * There are no "default" pid probes. */ } static int fasttrap_tracepoint_enable(proc_t *p, fasttrap_probe_t *probe, uint_t index) { fasttrap_tracepoint_t *tp, *new_tp = NULL; fasttrap_bucket_t *bucket; fasttrap_id_t *id; pid_t pid; uintptr_t pc; ASSERT(index < probe->ftp_ntps); pid = probe->ftp_pid; pc = probe->ftp_tps[index].fit_tp->ftt_pc; id = &probe->ftp_tps[index].fit_id; ASSERT(probe->ftp_tps[index].fit_tp->ftt_pid == pid); #if defined(sun) ASSERT(!(p->p_flag & SVFORK)); #endif /* * Before we make any modifications, make sure we've imposed a barrier * on the generation in which this probe was last modified. */ fasttrap_mod_barrier(probe->ftp_gen); bucket = &fasttrap_tpoints.fth_table[FASTTRAP_TPOINTS_INDEX(pid, pc)]; /* * If the tracepoint has already been enabled, just add our id to the * list of interested probes. This may be our second time through * this path in which case we'll have constructed the tracepoint we'd * like to install. If we can't find a match, and have an allocated * tracepoint ready to go, enable that one now. * * A tracepoint whose process is defunct is also considered defunct. */ again: mutex_enter(&bucket->ftb_mtx); for (tp = bucket->ftb_data; tp != NULL; tp = tp->ftt_next) { /* * Note that it's safe to access the active count on the * associated proc structure because we know that at least one * provider (this one) will still be around throughout this * operation. */ if (tp->ftt_pid != pid || tp->ftt_pc != pc || tp->ftt_proc->ftpc_acount == 0) continue; /* * Now that we've found a matching tracepoint, it would be * a decent idea to confirm that the tracepoint is still * enabled and the trap instruction hasn't been overwritten. * Since this is a little hairy, we'll punt for now. */ /* * This can't be the first interested probe. We don't have * to worry about another thread being in the midst of * deleting this tracepoint (which would be the only valid * reason for a tracepoint to have no interested probes) * since we're holding P_PR_LOCK for this process. */ ASSERT(tp->ftt_ids != NULL || tp->ftt_retids != NULL); switch (id->fti_ptype) { case DTFTP_ENTRY: case DTFTP_OFFSETS: case DTFTP_IS_ENABLED: id->fti_next = tp->ftt_ids; membar_producer(); tp->ftt_ids = id; membar_producer(); break; case DTFTP_RETURN: case DTFTP_POST_OFFSETS: id->fti_next = tp->ftt_retids; membar_producer(); tp->ftt_retids = id; membar_producer(); break; default: ASSERT(0); } mutex_exit(&bucket->ftb_mtx); if (new_tp != NULL) { new_tp->ftt_ids = NULL; new_tp->ftt_retids = NULL; } return (0); } /* * If we have a good tracepoint ready to go, install it now while * we have the lock held and no one can screw with us. */ if (new_tp != NULL) { int rc = 0; new_tp->ftt_next = bucket->ftb_data; membar_producer(); bucket->ftb_data = new_tp; membar_producer(); mutex_exit(&bucket->ftb_mtx); /* * Activate the tracepoint in the ISA-specific manner. * If this fails, we need to report the failure, but * indicate that this tracepoint must still be disabled * by calling fasttrap_tracepoint_disable(). */ if (fasttrap_tracepoint_install(p, new_tp) != 0) rc = FASTTRAP_ENABLE_PARTIAL; /* * Increment the count of the number of tracepoints active in * the victim process. */ #if defined(sun) ASSERT(p->p_proc_flag & P_PR_LOCK); #endif p->p_dtrace_count++; return (rc); } mutex_exit(&bucket->ftb_mtx); /* * Initialize the tracepoint that's been preallocated with the probe. */ new_tp = probe->ftp_tps[index].fit_tp; ASSERT(new_tp->ftt_pid == pid); ASSERT(new_tp->ftt_pc == pc); ASSERT(new_tp->ftt_proc == probe->ftp_prov->ftp_proc); ASSERT(new_tp->ftt_ids == NULL); ASSERT(new_tp->ftt_retids == NULL); switch (id->fti_ptype) { case DTFTP_ENTRY: case DTFTP_OFFSETS: case DTFTP_IS_ENABLED: id->fti_next = NULL; new_tp->ftt_ids = id; break; case DTFTP_RETURN: case DTFTP_POST_OFFSETS: id->fti_next = NULL; new_tp->ftt_retids = id; break; default: ASSERT(0); } /* * If the ISA-dependent initialization goes to plan, go back to the * beginning and try to install this freshly made tracepoint. */ if (fasttrap_tracepoint_init(p, new_tp, pc, id->fti_ptype) == 0) goto again; new_tp->ftt_ids = NULL; new_tp->ftt_retids = NULL; return (FASTTRAP_ENABLE_FAIL); } static void fasttrap_tracepoint_disable(proc_t *p, fasttrap_probe_t *probe, uint_t index) { fasttrap_bucket_t *bucket; fasttrap_provider_t *provider = probe->ftp_prov; fasttrap_tracepoint_t **pp, *tp; fasttrap_id_t *id, **idp = NULL; pid_t pid; uintptr_t pc; ASSERT(index < probe->ftp_ntps); pid = probe->ftp_pid; pc = probe->ftp_tps[index].fit_tp->ftt_pc; id = &probe->ftp_tps[index].fit_id; ASSERT(probe->ftp_tps[index].fit_tp->ftt_pid == pid); /* * Find the tracepoint and make sure that our id is one of the * ones registered with it. */ bucket = &fasttrap_tpoints.fth_table[FASTTRAP_TPOINTS_INDEX(pid, pc)]; mutex_enter(&bucket->ftb_mtx); for (tp = bucket->ftb_data; tp != NULL; tp = tp->ftt_next) { if (tp->ftt_pid == pid && tp->ftt_pc == pc && tp->ftt_proc == provider->ftp_proc) break; } /* * If we somehow lost this tracepoint, we're in a world of hurt. */ ASSERT(tp != NULL); switch (id->fti_ptype) { case DTFTP_ENTRY: case DTFTP_OFFSETS: case DTFTP_IS_ENABLED: ASSERT(tp->ftt_ids != NULL); idp = &tp->ftt_ids; break; case DTFTP_RETURN: case DTFTP_POST_OFFSETS: ASSERT(tp->ftt_retids != NULL); idp = &tp->ftt_retids; break; default: ASSERT(0); } while ((*idp)->fti_probe != probe) { idp = &(*idp)->fti_next; ASSERT(*idp != NULL); } id = *idp; *idp = id->fti_next; membar_producer(); ASSERT(id->fti_probe == probe); /* * If there are other registered enablings of this tracepoint, we're * all done, but if this was the last probe assocated with this * this tracepoint, we need to remove and free it. */ if (tp->ftt_ids != NULL || tp->ftt_retids != NULL) { /* * If the current probe's tracepoint is in use, swap it * for an unused tracepoint. */ if (tp == probe->ftp_tps[index].fit_tp) { fasttrap_probe_t *tmp_probe; fasttrap_tracepoint_t **tmp_tp; uint_t tmp_index; if (tp->ftt_ids != NULL) { tmp_probe = tp->ftt_ids->fti_probe; /* LINTED - alignment */ tmp_index = FASTTRAP_ID_INDEX(tp->ftt_ids); tmp_tp = &tmp_probe->ftp_tps[tmp_index].fit_tp; } else { tmp_probe = tp->ftt_retids->fti_probe; /* LINTED - alignment */ tmp_index = FASTTRAP_ID_INDEX(tp->ftt_retids); tmp_tp = &tmp_probe->ftp_tps[tmp_index].fit_tp; } ASSERT(*tmp_tp != NULL); ASSERT(*tmp_tp != probe->ftp_tps[index].fit_tp); ASSERT((*tmp_tp)->ftt_ids == NULL); ASSERT((*tmp_tp)->ftt_retids == NULL); probe->ftp_tps[index].fit_tp = *tmp_tp; *tmp_tp = tp; } mutex_exit(&bucket->ftb_mtx); /* * Tag the modified probe with the generation in which it was * changed. */ probe->ftp_gen = fasttrap_mod_gen; return; } mutex_exit(&bucket->ftb_mtx); /* * We can't safely remove the tracepoint from the set of active * tracepoints until we've actually removed the fasttrap instruction * from the process's text. We can, however, operate on this * tracepoint secure in the knowledge that no other thread is going to * be looking at it since we hold P_PR_LOCK on the process if it's * live or we hold the provider lock on the process if it's dead and * gone. */ /* * We only need to remove the actual instruction if we're looking * at an existing process */ if (p != NULL) { /* * If we fail to restore the instruction we need to kill * this process since it's in a completely unrecoverable * state. */ if (fasttrap_tracepoint_remove(p, tp) != 0) fasttrap_sigtrap(p, NULL, pc); /* * Decrement the count of the number of tracepoints active * in the victim process. */ #if defined(sun) ASSERT(p->p_proc_flag & P_PR_LOCK); #endif p->p_dtrace_count--; } /* * Remove the probe from the hash table of active tracepoints. */ mutex_enter(&bucket->ftb_mtx); pp = (fasttrap_tracepoint_t **)&bucket->ftb_data; ASSERT(*pp != NULL); while (*pp != tp) { pp = &(*pp)->ftt_next; ASSERT(*pp != NULL); } *pp = tp->ftt_next; membar_producer(); mutex_exit(&bucket->ftb_mtx); /* * Tag the modified probe with the generation in which it was changed. */ probe->ftp_gen = fasttrap_mod_gen; } static void fasttrap_enable_callbacks(void) { /* * We don't have to play the rw lock game here because we're * providing something rather than taking something away -- * we can be sure that no threads have tried to follow this * function pointer yet. */ mutex_enter(&fasttrap_count_mtx); if (fasttrap_pid_count == 0) { ASSERT(dtrace_pid_probe_ptr == NULL); ASSERT(dtrace_return_probe_ptr == NULL); dtrace_pid_probe_ptr = &fasttrap_pid_probe; dtrace_return_probe_ptr = &fasttrap_return_probe; } ASSERT(dtrace_pid_probe_ptr == &fasttrap_pid_probe); ASSERT(dtrace_return_probe_ptr == &fasttrap_return_probe); fasttrap_pid_count++; mutex_exit(&fasttrap_count_mtx); } static void fasttrap_disable_callbacks(void) { #if defined(sun) ASSERT(MUTEX_HELD(&cpu_lock)); #endif mutex_enter(&fasttrap_count_mtx); ASSERT(fasttrap_pid_count > 0); fasttrap_pid_count--; if (fasttrap_pid_count == 0) { #if defined(sun) cpu_t *cur, *cpu = CPU; for (cur = cpu->cpu_next_onln; cur != cpu; cur = cur->cpu_next_onln) { rw_enter(&cur->cpu_ft_lock, RW_WRITER); } #endif dtrace_pid_probe_ptr = NULL; dtrace_return_probe_ptr = NULL; #if defined(sun) for (cur = cpu->cpu_next_onln; cur != cpu; cur = cur->cpu_next_onln) { rw_exit(&cur->cpu_ft_lock); } #endif } mutex_exit(&fasttrap_count_mtx); } /*ARGSUSED*/ static void fasttrap_pid_enable(void *arg, dtrace_id_t id, void *parg) { fasttrap_probe_t *probe = parg; proc_t *p = NULL; int i, rc; ASSERT(probe != NULL); ASSERT(!probe->ftp_enabled); ASSERT(id == probe->ftp_id); #if defined(sun) ASSERT(MUTEX_HELD(&cpu_lock)); #endif /* * Increment the count of enabled probes on this probe's provider; * the provider can't go away while the probe still exists. We * must increment this even if we aren't able to properly enable * this probe. */ mutex_enter(&probe->ftp_prov->ftp_mtx); probe->ftp_prov->ftp_rcount++; mutex_exit(&probe->ftp_prov->ftp_mtx); /* * If this probe's provider is retired (meaning it was valid in a * previously exec'ed incarnation of this address space), bail out. The * provider can't go away while we're in this code path. */ if (probe->ftp_prov->ftp_retired) return; /* * If we can't find the process, it may be that we're in the context of * a fork in which the traced process is being born and we're copying * USDT probes. Otherwise, the process is gone so bail. */ #if defined(sun) if ((p = sprlock(probe->ftp_pid)) == NULL) { if ((curproc->p_flag & SFORKING) == 0) return; mutex_enter(&pidlock); p = prfind(probe->ftp_pid); /* * Confirm that curproc is indeed forking the process in which * we're trying to enable probes. */ ASSERT(p != NULL); ASSERT(p->p_parent == curproc); ASSERT(p->p_stat == SIDL); mutex_enter(&p->p_lock); mutex_exit(&pidlock); sprlock_proc(p); } ASSERT(!(p->p_flag & SVFORK)); mutex_exit(&p->p_lock); #else if ((p = pfind(probe->ftp_pid)) == NULL) return; #endif /* * We have to enable the trap entry point before any user threads have * the chance to execute the trap instruction we're about to place * in their process's text. */ #ifdef __FreeBSD__ /* * pfind() returns a locked process. */ _PHOLD(p); PROC_UNLOCK(p); #endif fasttrap_enable_callbacks(); /* * Enable all the tracepoints and add this probe's id to each * tracepoint's list of active probes. */ for (i = 0; i < probe->ftp_ntps; i++) { if ((rc = fasttrap_tracepoint_enable(p, probe, i)) != 0) { /* * If enabling the tracepoint failed completely, * we don't have to disable it; if the failure * was only partial we must disable it. */ if (rc == FASTTRAP_ENABLE_FAIL) i--; else ASSERT(rc == FASTTRAP_ENABLE_PARTIAL); /* * Back up and pull out all the tracepoints we've * created so far for this probe. */ while (i >= 0) { fasttrap_tracepoint_disable(p, probe, i); i--; } #if defined(sun) mutex_enter(&p->p_lock); sprunlock(p); #else PRELE(p); #endif /* * Since we're not actually enabling this probe, * drop our reference on the trap table entry. */ fasttrap_disable_callbacks(); return; } } #if defined(sun) mutex_enter(&p->p_lock); sprunlock(p); #else PRELE(p); #endif probe->ftp_enabled = 1; } /*ARGSUSED*/ static void fasttrap_pid_disable(void *arg, dtrace_id_t id, void *parg) { fasttrap_probe_t *probe = parg; fasttrap_provider_t *provider = probe->ftp_prov; proc_t *p; int i, whack = 0; ASSERT(id == probe->ftp_id); mutex_enter(&provider->ftp_mtx); /* * We won't be able to acquire a /proc-esque lock on the process * iff the process is dead and gone. In this case, we rely on the * provider lock as a point of mutual exclusion to prevent other * DTrace consumers from disabling this probe. */ if ((p = pfind(probe->ftp_pid)) != NULL) { #ifdef __FreeBSD__ _PHOLD(p); PROC_UNLOCK(p); #endif } /* * Disable all the associated tracepoints (for fully enabled probes). */ if (probe->ftp_enabled) { for (i = 0; i < probe->ftp_ntps; i++) { fasttrap_tracepoint_disable(p, probe, i); } } ASSERT(provider->ftp_rcount > 0); provider->ftp_rcount--; if (p != NULL) { /* * Even though we may not be able to remove it entirely, we * mark this retired provider to get a chance to remove some * of the associated probes. */ if (provider->ftp_retired && !provider->ftp_marked) whack = provider->ftp_marked = 1; mutex_exit(&provider->ftp_mtx); } else { /* * If the process is dead, we're just waiting for the * last probe to be disabled to be able to free it. */ if (provider->ftp_rcount == 0 && !provider->ftp_marked) whack = provider->ftp_marked = 1; mutex_exit(&provider->ftp_mtx); } if (whack) fasttrap_pid_cleanup(); #ifdef __FreeBSD__ if (p != NULL) PRELE(p); #endif if (!probe->ftp_enabled) return; probe->ftp_enabled = 0; #if defined(sun) ASSERT(MUTEX_HELD(&cpu_lock)); #endif fasttrap_disable_callbacks(); } /*ARGSUSED*/ static void fasttrap_pid_getargdesc(void *arg, dtrace_id_t id, void *parg, dtrace_argdesc_t *desc) { fasttrap_probe_t *probe = parg; char *str; int i, ndx; desc->dtargd_native[0] = '\0'; desc->dtargd_xlate[0] = '\0'; if (probe->ftp_prov->ftp_retired != 0 || desc->dtargd_ndx >= probe->ftp_nargs) { desc->dtargd_ndx = DTRACE_ARGNONE; return; } ndx = (probe->ftp_argmap != NULL) ? probe->ftp_argmap[desc->dtargd_ndx] : desc->dtargd_ndx; str = probe->ftp_ntypes; for (i = 0; i < ndx; i++) { str += strlen(str) + 1; } ASSERT(strlen(str + 1) < sizeof (desc->dtargd_native)); (void) strcpy(desc->dtargd_native, str); if (probe->ftp_xtypes == NULL) return; str = probe->ftp_xtypes; for (i = 0; i < desc->dtargd_ndx; i++) { str += strlen(str) + 1; } ASSERT(strlen(str + 1) < sizeof (desc->dtargd_xlate)); (void) strcpy(desc->dtargd_xlate, str); } /*ARGSUSED*/ static void fasttrap_pid_destroy(void *arg, dtrace_id_t id, void *parg) { fasttrap_probe_t *probe = parg; int i; size_t size; ASSERT(probe != NULL); ASSERT(!probe->ftp_enabled); ASSERT(fasttrap_total >= probe->ftp_ntps); atomic_add_32(&fasttrap_total, -probe->ftp_ntps); size = offsetof(fasttrap_probe_t, ftp_tps[probe->ftp_ntps]); if (probe->ftp_gen + 1 >= fasttrap_mod_gen) fasttrap_mod_barrier(probe->ftp_gen); for (i = 0; i < probe->ftp_ntps; i++) { kmem_free(probe->ftp_tps[i].fit_tp, sizeof (fasttrap_tracepoint_t)); } kmem_free(probe, size); } static const dtrace_pattr_t pid_attr = { { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, }; static dtrace_pops_t pid_pops = { fasttrap_pid_provide, NULL, fasttrap_pid_enable, fasttrap_pid_disable, NULL, NULL, fasttrap_pid_getargdesc, fasttrap_pid_getarg, NULL, fasttrap_pid_destroy }; static dtrace_pops_t usdt_pops = { fasttrap_pid_provide, NULL, fasttrap_pid_enable, fasttrap_pid_disable, NULL, NULL, fasttrap_pid_getargdesc, fasttrap_usdt_getarg, NULL, fasttrap_pid_destroy }; static fasttrap_proc_t * fasttrap_proc_lookup(pid_t pid) { fasttrap_bucket_t *bucket; fasttrap_proc_t *fprc, *new_fprc; bucket = &fasttrap_procs.fth_table[FASTTRAP_PROCS_INDEX(pid)]; mutex_enter(&bucket->ftb_mtx); for (fprc = bucket->ftb_data; fprc != NULL; fprc = fprc->ftpc_next) { if (fprc->ftpc_pid == pid && fprc->ftpc_acount != 0) { mutex_enter(&fprc->ftpc_mtx); mutex_exit(&bucket->ftb_mtx); fprc->ftpc_rcount++; - atomic_add_64(&fprc->ftpc_acount, 1); + atomic_inc_64(&fprc->ftpc_acount); ASSERT(fprc->ftpc_acount <= fprc->ftpc_rcount); mutex_exit(&fprc->ftpc_mtx); return (fprc); } } /* * Drop the bucket lock so we don't try to perform a sleeping * allocation under it. */ mutex_exit(&bucket->ftb_mtx); new_fprc = kmem_zalloc(sizeof (fasttrap_proc_t), KM_SLEEP); new_fprc->ftpc_pid = pid; new_fprc->ftpc_rcount = 1; new_fprc->ftpc_acount = 1; #if !defined(sun) mutex_init(&new_fprc->ftpc_mtx, "fasttrap proc mtx", MUTEX_DEFAULT, NULL); #endif mutex_enter(&bucket->ftb_mtx); /* * Take another lap through the list to make sure a proc hasn't * been created for this pid while we weren't under the bucket lock. */ for (fprc = bucket->ftb_data; fprc != NULL; fprc = fprc->ftpc_next) { if (fprc->ftpc_pid == pid && fprc->ftpc_acount != 0) { mutex_enter(&fprc->ftpc_mtx); mutex_exit(&bucket->ftb_mtx); fprc->ftpc_rcount++; - atomic_add_64(&fprc->ftpc_acount, 1); + atomic_inc_64(&fprc->ftpc_acount); ASSERT(fprc->ftpc_acount <= fprc->ftpc_rcount); mutex_exit(&fprc->ftpc_mtx); kmem_free(new_fprc, sizeof (fasttrap_proc_t)); return (fprc); } } new_fprc->ftpc_next = bucket->ftb_data; bucket->ftb_data = new_fprc; mutex_exit(&bucket->ftb_mtx); return (new_fprc); } static void fasttrap_proc_release(fasttrap_proc_t *proc) { fasttrap_bucket_t *bucket; fasttrap_proc_t *fprc, **fprcp; pid_t pid = proc->ftpc_pid; #if !defined(sun) fasttrap_scrblock_t *scrblk, *scrblktmp; fasttrap_scrspace_t *scrspc, *scrspctmp; struct proc *p; struct thread *td; #endif mutex_enter(&proc->ftpc_mtx); ASSERT(proc->ftpc_rcount != 0); ASSERT(proc->ftpc_acount <= proc->ftpc_rcount); if (--proc->ftpc_rcount != 0) { mutex_exit(&proc->ftpc_mtx); return; } #if !defined(sun) /* * Free all structures used to manage per-thread scratch space. */ LIST_FOREACH_SAFE(scrblk, &proc->ftpc_scrblks, ftsb_next, scrblktmp) { LIST_REMOVE(scrblk, ftsb_next); free(scrblk, M_SOLARIS); } LIST_FOREACH_SAFE(scrspc, &proc->ftpc_fscr, ftss_next, scrspctmp) { LIST_REMOVE(scrspc, ftss_next); free(scrspc, M_SOLARIS); } LIST_FOREACH_SAFE(scrspc, &proc->ftpc_ascr, ftss_next, scrspctmp) { LIST_REMOVE(scrspc, ftss_next); free(scrspc, M_SOLARIS); } if ((p = pfind(pid)) != NULL) { FOREACH_THREAD_IN_PROC(p, td) td->t_dtrace_sscr = NULL; PROC_UNLOCK(p); } #endif mutex_exit(&proc->ftpc_mtx); /* * There should definitely be no live providers associated with this * process at this point. */ ASSERT(proc->ftpc_acount == 0); bucket = &fasttrap_procs.fth_table[FASTTRAP_PROCS_INDEX(pid)]; mutex_enter(&bucket->ftb_mtx); fprcp = (fasttrap_proc_t **)&bucket->ftb_data; while ((fprc = *fprcp) != NULL) { if (fprc == proc) break; fprcp = &fprc->ftpc_next; } /* * Something strange has happened if we can't find the proc. */ ASSERT(fprc != NULL); *fprcp = fprc->ftpc_next; mutex_exit(&bucket->ftb_mtx); kmem_free(fprc, sizeof (fasttrap_proc_t)); } /* * Lookup a fasttrap-managed provider based on its name and associated pid. * If the pattr argument is non-NULL, this function instantiates the provider * if it doesn't exist otherwise it returns NULL. The provider is returned * with its lock held. */ static fasttrap_provider_t * fasttrap_provider_lookup(pid_t pid, const char *name, const dtrace_pattr_t *pattr) { fasttrap_provider_t *fp, *new_fp = NULL; fasttrap_bucket_t *bucket; char provname[DTRACE_PROVNAMELEN]; proc_t *p; cred_t *cred; ASSERT(strlen(name) < sizeof (fp->ftp_name)); ASSERT(pattr != NULL); bucket = &fasttrap_provs.fth_table[FASTTRAP_PROVS_INDEX(pid, name)]; mutex_enter(&bucket->ftb_mtx); /* * Take a lap through the list and return the match if we find it. */ for (fp = bucket->ftb_data; fp != NULL; fp = fp->ftp_next) { if (fp->ftp_pid == pid && strcmp(fp->ftp_name, name) == 0 && !fp->ftp_retired) { mutex_enter(&fp->ftp_mtx); mutex_exit(&bucket->ftb_mtx); return (fp); } } /* * Drop the bucket lock so we don't try to perform a sleeping * allocation under it. */ mutex_exit(&bucket->ftb_mtx); /* * Make sure the process exists, isn't a child created as the result * of a vfork(2), and isn't a zombie (but may be in fork). */ if ((p = pfind(pid)) == NULL) return (NULL); /* * Increment p_dtrace_probes so that the process knows to inform us * when it exits or execs. fasttrap_provider_free() decrements this * when we're done with this provider. */ p->p_dtrace_probes++; /* * Grab the credentials for this process so we have * something to pass to dtrace_register(). */ PROC_LOCK_ASSERT(p, MA_OWNED); crhold(p->p_ucred); cred = p->p_ucred; PROC_UNLOCK(p); new_fp = kmem_zalloc(sizeof (fasttrap_provider_t), KM_SLEEP); new_fp->ftp_pid = pid; new_fp->ftp_proc = fasttrap_proc_lookup(pid); #if !defined(sun) mutex_init(&new_fp->ftp_mtx, "provider mtx", MUTEX_DEFAULT, NULL); mutex_init(&new_fp->ftp_cmtx, "lock on creating", MUTEX_DEFAULT, NULL); #endif ASSERT(new_fp->ftp_proc != NULL); mutex_enter(&bucket->ftb_mtx); /* * Take another lap through the list to make sure a provider hasn't * been created for this pid while we weren't under the bucket lock. */ for (fp = bucket->ftb_data; fp != NULL; fp = fp->ftp_next) { if (fp->ftp_pid == pid && strcmp(fp->ftp_name, name) == 0 && !fp->ftp_retired) { mutex_enter(&fp->ftp_mtx); mutex_exit(&bucket->ftb_mtx); fasttrap_provider_free(new_fp); crfree(cred); return (fp); } } (void) strcpy(new_fp->ftp_name, name); /* * Fail and return NULL if either the provider name is too long * or we fail to register this new provider with the DTrace * framework. Note that this is the only place we ever construct * the full provider name -- we keep it in pieces in the provider * structure. */ if (snprintf(provname, sizeof (provname), "%s%u", name, (uint_t)pid) >= sizeof (provname) || dtrace_register(provname, pattr, DTRACE_PRIV_PROC | DTRACE_PRIV_OWNER | DTRACE_PRIV_ZONEOWNER, cred, pattr == &pid_attr ? &pid_pops : &usdt_pops, new_fp, &new_fp->ftp_provid) != 0) { mutex_exit(&bucket->ftb_mtx); fasttrap_provider_free(new_fp); crfree(cred); return (NULL); } new_fp->ftp_next = bucket->ftb_data; bucket->ftb_data = new_fp; mutex_enter(&new_fp->ftp_mtx); mutex_exit(&bucket->ftb_mtx); crfree(cred); return (new_fp); } static void fasttrap_provider_free(fasttrap_provider_t *provider) { pid_t pid = provider->ftp_pid; proc_t *p; /* * There need to be no associated enabled probes, no consumers * creating probes, and no meta providers referencing this provider. */ ASSERT(provider->ftp_rcount == 0); ASSERT(provider->ftp_ccount == 0); ASSERT(provider->ftp_mcount == 0); /* * If this provider hasn't been retired, we need to explicitly drop the * count of active providers on the associated process structure. */ if (!provider->ftp_retired) { - atomic_add_64(&provider->ftp_proc->ftpc_acount, -1); + atomic_dec_64(&provider->ftp_proc->ftpc_acount); ASSERT(provider->ftp_proc->ftpc_acount < provider->ftp_proc->ftpc_rcount); } fasttrap_proc_release(provider->ftp_proc); #if !defined(sun) mutex_destroy(&provider->ftp_mtx); mutex_destroy(&provider->ftp_cmtx); #endif kmem_free(provider, sizeof (fasttrap_provider_t)); /* * Decrement p_dtrace_probes on the process whose provider we're * freeing. We don't have to worry about clobbering somone else's * modifications to it because we have locked the bucket that * corresponds to this process's hash chain in the provider hash * table. Don't sweat it if we can't find the process. */ if ((p = pfind(pid)) == NULL) { return; } p->p_dtrace_probes--; #if !defined(sun) PROC_UNLOCK(p); #endif } static void fasttrap_provider_retire(pid_t pid, const char *name, int mprov) { fasttrap_provider_t *fp; fasttrap_bucket_t *bucket; dtrace_provider_id_t provid; ASSERT(strlen(name) < sizeof (fp->ftp_name)); bucket = &fasttrap_provs.fth_table[FASTTRAP_PROVS_INDEX(pid, name)]; mutex_enter(&bucket->ftb_mtx); for (fp = bucket->ftb_data; fp != NULL; fp = fp->ftp_next) { if (fp->ftp_pid == pid && strcmp(fp->ftp_name, name) == 0 && !fp->ftp_retired) break; } if (fp == NULL) { mutex_exit(&bucket->ftb_mtx); return; } mutex_enter(&fp->ftp_mtx); ASSERT(!mprov || fp->ftp_mcount > 0); if (mprov && --fp->ftp_mcount != 0) { mutex_exit(&fp->ftp_mtx); mutex_exit(&bucket->ftb_mtx); return; } /* * Mark the provider to be removed in our post-processing step, mark it * retired, and drop the active count on its proc. Marking it indicates * that we should try to remove it; setting the retired flag indicates * that we're done with this provider; dropping the active the proc * releases our hold, and when this reaches zero (as it will during * exit or exec) the proc and associated providers become defunct. * * We obviously need to take the bucket lock before the provider lock * to perform the lookup, but we need to drop the provider lock * before calling into the DTrace framework since we acquire the * provider lock in callbacks invoked from the DTrace framework. The * bucket lock therefore protects the integrity of the provider hash * table. */ - atomic_add_64(&fp->ftp_proc->ftpc_acount, -1); + atomic_dec_64(&fp->ftp_proc->ftpc_acount); ASSERT(fp->ftp_proc->ftpc_acount < fp->ftp_proc->ftpc_rcount); fp->ftp_retired = 1; fp->ftp_marked = 1; provid = fp->ftp_provid; mutex_exit(&fp->ftp_mtx); /* * We don't have to worry about invalidating the same provider twice * since fasttrap_provider_lookup() will ignore provider that have * been marked as retired. */ dtrace_invalidate(provid); mutex_exit(&bucket->ftb_mtx); fasttrap_pid_cleanup(); } static int fasttrap_uint32_cmp(const void *ap, const void *bp) { return (*(const uint32_t *)ap - *(const uint32_t *)bp); } static int fasttrap_uint64_cmp(const void *ap, const void *bp) { return (*(const uint64_t *)ap - *(const uint64_t *)bp); } static int fasttrap_add_probe(fasttrap_probe_spec_t *pdata) { fasttrap_provider_t *provider; fasttrap_probe_t *pp; fasttrap_tracepoint_t *tp; char *name; int i, aframes = 0, whack; /* * There needs to be at least one desired trace point. */ if (pdata->ftps_noffs == 0) return (EINVAL); switch (pdata->ftps_type) { case DTFTP_ENTRY: name = "entry"; aframes = FASTTRAP_ENTRY_AFRAMES; break; case DTFTP_RETURN: name = "return"; aframes = FASTTRAP_RETURN_AFRAMES; break; case DTFTP_OFFSETS: name = NULL; break; default: return (EINVAL); } if ((provider = fasttrap_provider_lookup(pdata->ftps_pid, FASTTRAP_PID_NAME, &pid_attr)) == NULL) return (ESRCH); /* * Increment this reference count to indicate that a consumer is * actively adding a new probe associated with this provider. This * prevents the provider from being deleted -- we'll need to check * for pending deletions when we drop this reference count. */ provider->ftp_ccount++; mutex_exit(&provider->ftp_mtx); /* * Grab the creation lock to ensure consistency between calls to * dtrace_probe_lookup() and dtrace_probe_create() in the face of * other threads creating probes. We must drop the provider lock * before taking this lock to avoid a three-way deadlock with the * DTrace framework. */ mutex_enter(&provider->ftp_cmtx); if (name == NULL) { for (i = 0; i < pdata->ftps_noffs; i++) { char name_str[17]; (void) sprintf(name_str, "%llx", (unsigned long long)pdata->ftps_offs[i]); if (dtrace_probe_lookup(provider->ftp_provid, pdata->ftps_mod, pdata->ftps_func, name_str) != 0) continue; - atomic_add_32(&fasttrap_total, 1); + atomic_inc_32(&fasttrap_total); if (fasttrap_total > fasttrap_max) { - atomic_add_32(&fasttrap_total, -1); + atomic_dec_32(&fasttrap_total); goto no_mem; } pp = kmem_zalloc(sizeof (fasttrap_probe_t), KM_SLEEP); pp->ftp_prov = provider; pp->ftp_faddr = pdata->ftps_pc; pp->ftp_fsize = pdata->ftps_size; pp->ftp_pid = pdata->ftps_pid; pp->ftp_ntps = 1; tp = kmem_zalloc(sizeof (fasttrap_tracepoint_t), KM_SLEEP); tp->ftt_proc = provider->ftp_proc; tp->ftt_pc = pdata->ftps_offs[i] + pdata->ftps_pc; tp->ftt_pid = pdata->ftps_pid; pp->ftp_tps[0].fit_tp = tp; pp->ftp_tps[0].fit_id.fti_probe = pp; pp->ftp_tps[0].fit_id.fti_ptype = pdata->ftps_type; pp->ftp_id = dtrace_probe_create(provider->ftp_provid, pdata->ftps_mod, pdata->ftps_func, name_str, FASTTRAP_OFFSET_AFRAMES, pp); } } else if (dtrace_probe_lookup(provider->ftp_provid, pdata->ftps_mod, pdata->ftps_func, name) == 0) { atomic_add_32(&fasttrap_total, pdata->ftps_noffs); if (fasttrap_total > fasttrap_max) { atomic_add_32(&fasttrap_total, -pdata->ftps_noffs); goto no_mem; } /* * Make sure all tracepoint program counter values are unique. * We later assume that each probe has exactly one tracepoint * for a given pc. */ qsort(pdata->ftps_offs, pdata->ftps_noffs, sizeof (uint64_t), fasttrap_uint64_cmp); for (i = 1; i < pdata->ftps_noffs; i++) { if (pdata->ftps_offs[i] > pdata->ftps_offs[i - 1]) continue; atomic_add_32(&fasttrap_total, -pdata->ftps_noffs); goto no_mem; } ASSERT(pdata->ftps_noffs > 0); pp = kmem_zalloc(offsetof(fasttrap_probe_t, ftp_tps[pdata->ftps_noffs]), KM_SLEEP); pp->ftp_prov = provider; pp->ftp_faddr = pdata->ftps_pc; pp->ftp_fsize = pdata->ftps_size; pp->ftp_pid = pdata->ftps_pid; pp->ftp_ntps = pdata->ftps_noffs; for (i = 0; i < pdata->ftps_noffs; i++) { tp = kmem_zalloc(sizeof (fasttrap_tracepoint_t), KM_SLEEP); tp->ftt_proc = provider->ftp_proc; tp->ftt_pc = pdata->ftps_offs[i] + pdata->ftps_pc; tp->ftt_pid = pdata->ftps_pid; pp->ftp_tps[i].fit_tp = tp; pp->ftp_tps[i].fit_id.fti_probe = pp; pp->ftp_tps[i].fit_id.fti_ptype = pdata->ftps_type; } pp->ftp_id = dtrace_probe_create(provider->ftp_provid, pdata->ftps_mod, pdata->ftps_func, name, aframes, pp); } mutex_exit(&provider->ftp_cmtx); /* * We know that the provider is still valid since we incremented the * creation reference count. If someone tried to clean up this provider * while we were using it (e.g. because the process called exec(2) or * exit(2)), take note of that and try to clean it up now. */ mutex_enter(&provider->ftp_mtx); provider->ftp_ccount--; whack = provider->ftp_retired; mutex_exit(&provider->ftp_mtx); if (whack) fasttrap_pid_cleanup(); return (0); no_mem: /* * If we've exhausted the allowable resources, we'll try to remove * this provider to free some up. This is to cover the case where * the user has accidentally created many more probes than was * intended (e.g. pid123:::). */ mutex_exit(&provider->ftp_cmtx); mutex_enter(&provider->ftp_mtx); provider->ftp_ccount--; provider->ftp_marked = 1; mutex_exit(&provider->ftp_mtx); fasttrap_pid_cleanup(); return (ENOMEM); } /*ARGSUSED*/ static void * fasttrap_meta_provide(void *arg, dtrace_helper_provdesc_t *dhpv, pid_t pid) { fasttrap_provider_t *provider; /* * A 32-bit unsigned integer (like a pid for example) can be * expressed in 10 or fewer decimal digits. Make sure that we'll * have enough space for the provider name. */ if (strlen(dhpv->dthpv_provname) + 10 >= sizeof (provider->ftp_name)) { printf("failed to instantiate provider %s: " "name too long to accomodate pid", dhpv->dthpv_provname); return (NULL); } /* * Don't let folks spoof the true pid provider. */ if (strcmp(dhpv->dthpv_provname, FASTTRAP_PID_NAME) == 0) { printf("failed to instantiate provider %s: " "%s is an invalid name", dhpv->dthpv_provname, FASTTRAP_PID_NAME); return (NULL); } /* * The highest stability class that fasttrap supports is ISA; cap * the stability of the new provider accordingly. */ if (dhpv->dthpv_pattr.dtpa_provider.dtat_class > DTRACE_CLASS_ISA) dhpv->dthpv_pattr.dtpa_provider.dtat_class = DTRACE_CLASS_ISA; if (dhpv->dthpv_pattr.dtpa_mod.dtat_class > DTRACE_CLASS_ISA) dhpv->dthpv_pattr.dtpa_mod.dtat_class = DTRACE_CLASS_ISA; if (dhpv->dthpv_pattr.dtpa_func.dtat_class > DTRACE_CLASS_ISA) dhpv->dthpv_pattr.dtpa_func.dtat_class = DTRACE_CLASS_ISA; if (dhpv->dthpv_pattr.dtpa_name.dtat_class > DTRACE_CLASS_ISA) dhpv->dthpv_pattr.dtpa_name.dtat_class = DTRACE_CLASS_ISA; if (dhpv->dthpv_pattr.dtpa_args.dtat_class > DTRACE_CLASS_ISA) dhpv->dthpv_pattr.dtpa_args.dtat_class = DTRACE_CLASS_ISA; if ((provider = fasttrap_provider_lookup(pid, dhpv->dthpv_provname, &dhpv->dthpv_pattr)) == NULL) { printf("failed to instantiate provider %s for " "process %u", dhpv->dthpv_provname, (uint_t)pid); return (NULL); } /* * Up the meta provider count so this provider isn't removed until * the meta provider has been told to remove it. */ provider->ftp_mcount++; mutex_exit(&provider->ftp_mtx); return (provider); } /*ARGSUSED*/ static void fasttrap_meta_create_probe(void *arg, void *parg, dtrace_helper_probedesc_t *dhpb) { fasttrap_provider_t *provider = parg; fasttrap_probe_t *pp; fasttrap_tracepoint_t *tp; int i, j; uint32_t ntps; /* * Since the meta provider count is non-zero we don't have to worry * about this provider disappearing. */ ASSERT(provider->ftp_mcount > 0); /* * The offsets must be unique. */ qsort(dhpb->dthpb_offs, dhpb->dthpb_noffs, sizeof (uint32_t), fasttrap_uint32_cmp); for (i = 1; i < dhpb->dthpb_noffs; i++) { if (dhpb->dthpb_base + dhpb->dthpb_offs[i] <= dhpb->dthpb_base + dhpb->dthpb_offs[i - 1]) return; } qsort(dhpb->dthpb_enoffs, dhpb->dthpb_nenoffs, sizeof (uint32_t), fasttrap_uint32_cmp); for (i = 1; i < dhpb->dthpb_nenoffs; i++) { if (dhpb->dthpb_base + dhpb->dthpb_enoffs[i] <= dhpb->dthpb_base + dhpb->dthpb_enoffs[i - 1]) return; } /* * Grab the creation lock to ensure consistency between calls to * dtrace_probe_lookup() and dtrace_probe_create() in the face of * other threads creating probes. */ mutex_enter(&provider->ftp_cmtx); if (dtrace_probe_lookup(provider->ftp_provid, dhpb->dthpb_mod, dhpb->dthpb_func, dhpb->dthpb_name) != 0) { mutex_exit(&provider->ftp_cmtx); return; } ntps = dhpb->dthpb_noffs + dhpb->dthpb_nenoffs; ASSERT(ntps > 0); atomic_add_32(&fasttrap_total, ntps); if (fasttrap_total > fasttrap_max) { atomic_add_32(&fasttrap_total, -ntps); mutex_exit(&provider->ftp_cmtx); return; } pp = kmem_zalloc(offsetof(fasttrap_probe_t, ftp_tps[ntps]), KM_SLEEP); pp->ftp_prov = provider; pp->ftp_pid = provider->ftp_pid; pp->ftp_ntps = ntps; pp->ftp_nargs = dhpb->dthpb_xargc; pp->ftp_xtypes = dhpb->dthpb_xtypes; pp->ftp_ntypes = dhpb->dthpb_ntypes; /* * First create a tracepoint for each actual point of interest. */ for (i = 0; i < dhpb->dthpb_noffs; i++) { tp = kmem_zalloc(sizeof (fasttrap_tracepoint_t), KM_SLEEP); tp->ftt_proc = provider->ftp_proc; tp->ftt_pc = dhpb->dthpb_base + dhpb->dthpb_offs[i]; tp->ftt_pid = provider->ftp_pid; pp->ftp_tps[i].fit_tp = tp; pp->ftp_tps[i].fit_id.fti_probe = pp; #ifdef __sparc pp->ftp_tps[i].fit_id.fti_ptype = DTFTP_POST_OFFSETS; #else pp->ftp_tps[i].fit_id.fti_ptype = DTFTP_OFFSETS; #endif } /* * Then create a tracepoint for each is-enabled point. */ for (j = 0; i < ntps; i++, j++) { tp = kmem_zalloc(sizeof (fasttrap_tracepoint_t), KM_SLEEP); tp->ftt_proc = provider->ftp_proc; tp->ftt_pc = dhpb->dthpb_base + dhpb->dthpb_enoffs[j]; tp->ftt_pid = provider->ftp_pid; pp->ftp_tps[i].fit_tp = tp; pp->ftp_tps[i].fit_id.fti_probe = pp; pp->ftp_tps[i].fit_id.fti_ptype = DTFTP_IS_ENABLED; } /* * If the arguments are shuffled around we set the argument remapping * table. Later, when the probe fires, we only remap the arguments * if the table is non-NULL. */ for (i = 0; i < dhpb->dthpb_xargc; i++) { if (dhpb->dthpb_args[i] != i) { pp->ftp_argmap = dhpb->dthpb_args; break; } } /* * The probe is fully constructed -- register it with DTrace. */ pp->ftp_id = dtrace_probe_create(provider->ftp_provid, dhpb->dthpb_mod, dhpb->dthpb_func, dhpb->dthpb_name, FASTTRAP_OFFSET_AFRAMES, pp); mutex_exit(&provider->ftp_cmtx); } /*ARGSUSED*/ static void fasttrap_meta_remove(void *arg, dtrace_helper_provdesc_t *dhpv, pid_t pid) { /* * Clean up the USDT provider. There may be active consumers of the * provider busy adding probes, no damage will actually befall the * provider until that count has dropped to zero. This just puts * the provider on death row. */ fasttrap_provider_retire(pid, dhpv->dthpv_provname, 1); } static dtrace_mops_t fasttrap_mops = { fasttrap_meta_create_probe, fasttrap_meta_provide, fasttrap_meta_remove }; /*ARGSUSED*/ static int fasttrap_open(struct cdev *dev __unused, int oflags __unused, int devtype __unused, struct thread *td __unused) { return (0); } /*ARGSUSED*/ static int fasttrap_ioctl(struct cdev *dev, u_long cmd, caddr_t arg, int fflag, struct thread *td) { #ifdef notyet struct kinfo_proc kp; const cred_t *cr = td->td_ucred; #endif if (!dtrace_attached()) return (EAGAIN); if (cmd == FASTTRAPIOC_MAKEPROBE) { fasttrap_probe_spec_t *uprobe = *(fasttrap_probe_spec_t **)arg; fasttrap_probe_spec_t *probe; uint64_t noffs; size_t size; int ret, err; if (copyin(&uprobe->ftps_noffs, &noffs, sizeof (uprobe->ftps_noffs))) return (EFAULT); /* * Probes must have at least one tracepoint. */ if (noffs == 0) return (EINVAL); size = sizeof (fasttrap_probe_spec_t) + sizeof (probe->ftps_offs[0]) * (noffs - 1); if (size > 1024 * 1024) return (ENOMEM); probe = kmem_alloc(size, KM_SLEEP); if (copyin(uprobe, probe, size) != 0 || probe->ftps_noffs != noffs) { kmem_free(probe, size); return (EFAULT); } /* * Verify that the function and module strings contain no * funny characters. */ if (u8_validate(probe->ftps_func, strlen(probe->ftps_func), NULL, U8_VALIDATE_ENTIRE, &err) < 0) { ret = EINVAL; goto err; } if (u8_validate(probe->ftps_mod, strlen(probe->ftps_mod), NULL, U8_VALIDATE_ENTIRE, &err) < 0) { ret = EINVAL; goto err; } #ifdef notyet if (!PRIV_POLICY_CHOICE(cr, PRIV_ALL, B_FALSE)) { proc_t *p; pid_t pid = probe->ftps_pid; #if defined(sun) mutex_enter(&pidlock); #endif /* * Report an error if the process doesn't exist * or is actively being birthed. */ p = pfind(pid); if (p) fill_kinfo_proc(p, &kp); if (p == NULL || kp.ki_stat == SIDL) { #if defined(sun) mutex_exit(&pidlock); #endif return (ESRCH); } #if defined(sun) mutex_enter(&p->p_lock); mutex_exit(&pidlock); #else PROC_LOCK_ASSERT(p, MA_OWNED); #endif #ifdef notyet if ((ret = priv_proc_cred_perm(cr, p, NULL, VREAD | VWRITE)) != 0) { #if defined(sun) mutex_exit(&p->p_lock); #else PROC_UNLOCK(p); #endif return (ret); } #endif /* notyet */ #if defined(sun) mutex_exit(&p->p_lock); #else PROC_UNLOCK(p); #endif } #endif /* notyet */ ret = fasttrap_add_probe(probe); err: kmem_free(probe, size); return (ret); } else if (cmd == FASTTRAPIOC_GETINSTR) { fasttrap_instr_query_t instr; fasttrap_tracepoint_t *tp; uint_t index; #if defined(sun) int ret; #endif #if defined(sun) if (copyin((void *)arg, &instr, sizeof (instr)) != 0) return (EFAULT); #endif #ifdef notyet if (!PRIV_POLICY_CHOICE(cr, PRIV_ALL, B_FALSE)) { proc_t *p; pid_t pid = instr.ftiq_pid; #if defined(sun) mutex_enter(&pidlock); #endif /* * Report an error if the process doesn't exist * or is actively being birthed. */ p = pfind(pid); if (p) fill_kinfo_proc(p, &kp); if (p == NULL || kp.ki_stat == SIDL) { #if defined(sun) mutex_exit(&pidlock); #endif return (ESRCH); } #if defined(sun) mutex_enter(&p->p_lock); mutex_exit(&pidlock); #else PROC_LOCK_ASSERT(p, MA_OWNED); #endif #ifdef notyet if ((ret = priv_proc_cred_perm(cr, p, NULL, VREAD)) != 0) { #if defined(sun) mutex_exit(&p->p_lock); #else PROC_UNLOCK(p); #endif return (ret); } #endif /* notyet */ #if defined(sun) mutex_exit(&p->p_lock); #else PROC_UNLOCK(p); #endif } #endif /* notyet */ index = FASTTRAP_TPOINTS_INDEX(instr.ftiq_pid, instr.ftiq_pc); mutex_enter(&fasttrap_tpoints.fth_table[index].ftb_mtx); tp = fasttrap_tpoints.fth_table[index].ftb_data; while (tp != NULL) { if (instr.ftiq_pid == tp->ftt_pid && instr.ftiq_pc == tp->ftt_pc && tp->ftt_proc->ftpc_acount != 0) break; tp = tp->ftt_next; } if (tp == NULL) { mutex_exit(&fasttrap_tpoints.fth_table[index].ftb_mtx); return (ENOENT); } bcopy(&tp->ftt_instr, &instr.ftiq_instr, sizeof (instr.ftiq_instr)); mutex_exit(&fasttrap_tpoints.fth_table[index].ftb_mtx); if (copyout(&instr, (void *)arg, sizeof (instr)) != 0) return (EFAULT); return (0); } return (EINVAL); } static int fasttrap_load(void) { ulong_t nent; int i, ret; /* Create the /dev/dtrace/fasttrap entry. */ fasttrap_cdev = make_dev(&fasttrap_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "dtrace/fasttrap"); mtx_init(&fasttrap_cleanup_mtx, "fasttrap clean", "dtrace", MTX_DEF); mutex_init(&fasttrap_count_mtx, "fasttrap count mtx", MUTEX_DEFAULT, NULL); #if defined(sun) fasttrap_max = ddi_getprop(DDI_DEV_T_ANY, devi, DDI_PROP_DONTPASS, "fasttrap-max-probes", FASTTRAP_MAX_DEFAULT); #else fasttrap_max = FASTTRAP_MAX_DEFAULT; #endif fasttrap_total = 0; /* * Conjure up the tracepoints hashtable... */ #if defined(sun) nent = ddi_getprop(DDI_DEV_T_ANY, devi, DDI_PROP_DONTPASS, "fasttrap-hash-size", FASTTRAP_TPOINTS_DEFAULT_SIZE); #else nent = FASTTRAP_TPOINTS_DEFAULT_SIZE; #endif if (nent == 0 || nent > 0x1000000) nent = FASTTRAP_TPOINTS_DEFAULT_SIZE; if ((nent & (nent - 1)) == 0) fasttrap_tpoints.fth_nent = nent; else fasttrap_tpoints.fth_nent = 1 << fasttrap_highbit(nent); ASSERT(fasttrap_tpoints.fth_nent > 0); fasttrap_tpoints.fth_mask = fasttrap_tpoints.fth_nent - 1; fasttrap_tpoints.fth_table = kmem_zalloc(fasttrap_tpoints.fth_nent * sizeof (fasttrap_bucket_t), KM_SLEEP); #if !defined(sun) for (i = 0; i < fasttrap_tpoints.fth_nent; i++) mutex_init(&fasttrap_tpoints.fth_table[i].ftb_mtx, "tracepoints bucket mtx", MUTEX_DEFAULT, NULL); #endif /* * ... and the providers hash table... */ nent = FASTTRAP_PROVIDERS_DEFAULT_SIZE; if ((nent & (nent - 1)) == 0) fasttrap_provs.fth_nent = nent; else fasttrap_provs.fth_nent = 1 << fasttrap_highbit(nent); ASSERT(fasttrap_provs.fth_nent > 0); fasttrap_provs.fth_mask = fasttrap_provs.fth_nent - 1; fasttrap_provs.fth_table = kmem_zalloc(fasttrap_provs.fth_nent * sizeof (fasttrap_bucket_t), KM_SLEEP); #if !defined(sun) for (i = 0; i < fasttrap_provs.fth_nent; i++) mutex_init(&fasttrap_provs.fth_table[i].ftb_mtx, "providers bucket mtx", MUTEX_DEFAULT, NULL); #endif ret = kproc_create(fasttrap_pid_cleanup_cb, NULL, &fasttrap_cleanup_proc, 0, 0, "ftcleanup"); if (ret != 0) { destroy_dev(fasttrap_cdev); #if !defined(sun) for (i = 0; i < fasttrap_provs.fth_nent; i++) mutex_destroy(&fasttrap_provs.fth_table[i].ftb_mtx); for (i = 0; i < fasttrap_tpoints.fth_nent; i++) mutex_destroy(&fasttrap_tpoints.fth_table[i].ftb_mtx); #endif kmem_free(fasttrap_provs.fth_table, fasttrap_provs.fth_nent * sizeof (fasttrap_bucket_t)); mtx_destroy(&fasttrap_cleanup_mtx); mutex_destroy(&fasttrap_count_mtx); return (ret); } /* * ... and the procs hash table. */ nent = FASTTRAP_PROCS_DEFAULT_SIZE; if ((nent & (nent - 1)) == 0) fasttrap_procs.fth_nent = nent; else fasttrap_procs.fth_nent = 1 << fasttrap_highbit(nent); ASSERT(fasttrap_procs.fth_nent > 0); fasttrap_procs.fth_mask = fasttrap_procs.fth_nent - 1; fasttrap_procs.fth_table = kmem_zalloc(fasttrap_procs.fth_nent * sizeof (fasttrap_bucket_t), KM_SLEEP); #if !defined(sun) for (i = 0; i < fasttrap_procs.fth_nent; i++) mutex_init(&fasttrap_procs.fth_table[i].ftb_mtx, "processes bucket mtx", MUTEX_DEFAULT, NULL); CPU_FOREACH(i) { mutex_init(&fasttrap_cpuc_pid_lock[i], "fasttrap barrier", MUTEX_DEFAULT, NULL); } /* * This event handler must run before kdtrace_thread_dtor() since it * accesses the thread's struct kdtrace_thread. */ fasttrap_thread_dtor_tag = EVENTHANDLER_REGISTER(thread_dtor, fasttrap_thread_dtor, NULL, EVENTHANDLER_PRI_FIRST); #endif /* * Install our hooks into fork(2), exec(2), and exit(2). */ dtrace_fasttrap_fork = &fasttrap_fork; dtrace_fasttrap_exit = &fasttrap_exec_exit; dtrace_fasttrap_exec = &fasttrap_exec_exit; (void) dtrace_meta_register("fasttrap", &fasttrap_mops, NULL, &fasttrap_meta_id); return (0); } static int fasttrap_unload(void) { int i, fail = 0; /* * Unregister the meta-provider to make sure no new fasttrap- * managed providers come along while we're trying to close up * shop. If we fail to detach, we'll need to re-register as a * meta-provider. We can fail to unregister as a meta-provider * if providers we manage still exist. */ if (fasttrap_meta_id != DTRACE_METAPROVNONE && dtrace_meta_unregister(fasttrap_meta_id) != 0) return (-1); /* * Iterate over all of our providers. If there's still a process * that corresponds to that pid, fail to detach. */ for (i = 0; i < fasttrap_provs.fth_nent; i++) { fasttrap_provider_t **fpp, *fp; fasttrap_bucket_t *bucket = &fasttrap_provs.fth_table[i]; mutex_enter(&bucket->ftb_mtx); fpp = (fasttrap_provider_t **)&bucket->ftb_data; while ((fp = *fpp) != NULL) { /* * Acquire and release the lock as a simple way of * waiting for any other consumer to finish with * this provider. A thread must first acquire the * bucket lock so there's no chance of another thread * blocking on the provider's lock. */ mutex_enter(&fp->ftp_mtx); mutex_exit(&fp->ftp_mtx); if (dtrace_unregister(fp->ftp_provid) != 0) { fail = 1; fpp = &fp->ftp_next; } else { *fpp = fp->ftp_next; fasttrap_provider_free(fp); } } mutex_exit(&bucket->ftb_mtx); } if (fail) { (void) dtrace_meta_register("fasttrap", &fasttrap_mops, NULL, &fasttrap_meta_id); return (-1); } /* * Stop new processes from entering these hooks now, before the * fasttrap_cleanup thread runs. That way all processes will hopefully * be out of these hooks before we free fasttrap_provs.fth_table */ ASSERT(dtrace_fasttrap_fork == &fasttrap_fork); dtrace_fasttrap_fork = NULL; ASSERT(dtrace_fasttrap_exec == &fasttrap_exec_exit); dtrace_fasttrap_exec = NULL; ASSERT(dtrace_fasttrap_exit == &fasttrap_exec_exit); dtrace_fasttrap_exit = NULL; mtx_lock(&fasttrap_cleanup_mtx); fasttrap_cleanup_drain = 1; /* Wait for the cleanup thread to finish up and signal us. */ wakeup(&fasttrap_cleanup_cv); mtx_sleep(&fasttrap_cleanup_drain, &fasttrap_cleanup_mtx, 0, "ftcld", 0); fasttrap_cleanup_proc = NULL; mtx_destroy(&fasttrap_cleanup_mtx); #ifdef DEBUG mutex_enter(&fasttrap_count_mtx); ASSERT(fasttrap_pid_count == 0); mutex_exit(&fasttrap_count_mtx); #endif #if !defined(sun) EVENTHANDLER_DEREGISTER(thread_dtor, fasttrap_thread_dtor_tag); for (i = 0; i < fasttrap_tpoints.fth_nent; i++) mutex_destroy(&fasttrap_tpoints.fth_table[i].ftb_mtx); for (i = 0; i < fasttrap_provs.fth_nent; i++) mutex_destroy(&fasttrap_provs.fth_table[i].ftb_mtx); for (i = 0; i < fasttrap_procs.fth_nent; i++) mutex_destroy(&fasttrap_procs.fth_table[i].ftb_mtx); #endif kmem_free(fasttrap_tpoints.fth_table, fasttrap_tpoints.fth_nent * sizeof (fasttrap_bucket_t)); fasttrap_tpoints.fth_nent = 0; kmem_free(fasttrap_provs.fth_table, fasttrap_provs.fth_nent * sizeof (fasttrap_bucket_t)); fasttrap_provs.fth_nent = 0; kmem_free(fasttrap_procs.fth_table, fasttrap_procs.fth_nent * sizeof (fasttrap_bucket_t)); fasttrap_procs.fth_nent = 0; #if !defined(sun) destroy_dev(fasttrap_cdev); mutex_destroy(&fasttrap_count_mtx); CPU_FOREACH(i) { mutex_destroy(&fasttrap_cpuc_pid_lock[i]); } #endif return (0); } /* ARGSUSED */ static int fasttrap_modevent(module_t mod __unused, int type, void *data __unused) { int error = 0; switch (type) { case MOD_LOAD: break; case MOD_UNLOAD: break; case MOD_SHUTDOWN: break; default: error = EOPNOTSUPP; break; } return (error); } SYSINIT(fasttrap_load, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, fasttrap_load, NULL); SYSUNINIT(fasttrap_unload, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, fasttrap_unload, NULL); DEV_MODULE(fasttrap, fasttrap_modevent, NULL); MODULE_VERSION(fasttrap, 1); MODULE_DEPEND(fasttrap, dtrace, 1, 1, 1); MODULE_DEPEND(fasttrap, opensolaris, 1, 1, 1); Index: stable/10/sys/cddl/contrib/opensolaris/uts/common/dtrace/profile.c =================================================================== --- stable/10/sys/cddl/contrib/opensolaris/uts/common/dtrace/profile.c (revision 271000) +++ stable/10/sys/cddl/contrib/opensolaris/uts/common/dtrace/profile.c (revision 271001) @@ -1,576 +1,576 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright 2007 Sun Microsystems, Inc. All rights reserved. * Use is subject to license terms. */ #pragma ident "%Z%%M% %I% %E% SMI" #include #include #include #include #include #include #include #include #include #include #include #include #include static dev_info_t *profile_devi; static dtrace_provider_id_t profile_id; /* * Regardless of platform, the stack frames look like this in the case of the * profile provider: * * profile_fire * cyclic_expire * cyclic_fire * [ cbe ] * [ interrupt code ] * * On x86, there are five frames from the generic interrupt code; further, the * interrupted instruction appears as its own stack frame, giving us a total of * 10. * * On SPARC, the picture is further complicated because the compiler * optimizes away tail-calls -- so the following frames are optimized away: * * profile_fire * cyclic_expire * * This gives three frames. However, on DEBUG kernels, the cyclic_expire * frame cannot be tail-call eliminated, yielding four frames in this case. * * All of the above constraints lead to the mess below. Yes, the profile * provider should ideally figure this out on-the-fly by hitting one of its own * probes and then walking its own stack trace. This is complicated, however, * and the static definition doesn't seem to be overly brittle. Still, we * allow for a manual override in case we get it completely wrong. */ #ifdef __x86 #define PROF_ARTIFICIAL_FRAMES 10 #else #ifdef __sparc #ifdef DEBUG #define PROF_ARTIFICIAL_FRAMES 4 #else #define PROF_ARTIFICIAL_FRAMES 3 #endif #endif #endif #define PROF_NAMELEN 15 #define PROF_PROFILE 0 #define PROF_TICK 1 #define PROF_PREFIX_PROFILE "profile-" #define PROF_PREFIX_TICK "tick-" typedef struct profile_probe { char prof_name[PROF_NAMELEN]; dtrace_id_t prof_id; int prof_kind; hrtime_t prof_interval; cyclic_id_t prof_cyclic; } profile_probe_t; typedef struct profile_probe_percpu { hrtime_t profc_expected; hrtime_t profc_interval; profile_probe_t *profc_probe; } profile_probe_percpu_t; hrtime_t profile_interval_min = NANOSEC / 5000; /* 5000 hz */ int profile_aframes = 0; /* override */ static int profile_rates[] = { 97, 199, 499, 997, 1999, 4001, 4999, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }; static int profile_ticks[] = { 1, 10, 100, 500, 1000, 5000, 0, 0, 0, 0, 0, 0, 0, 0, 0 }; /* * profile_max defines the upper bound on the number of profile probes that * can exist (this is to prevent malicious or clumsy users from exhausing * system resources by creating a slew of profile probes). At mod load time, * this gets its value from PROFILE_MAX_DEFAULT or profile-max-probes if it's * present in the profile.conf file. */ #define PROFILE_MAX_DEFAULT 1000 /* default max. number of probes */ static uint32_t profile_max; /* maximum number of profile probes */ static uint32_t profile_total; /* current number of profile probes */ static void profile_fire(void *arg) { profile_probe_percpu_t *pcpu = arg; profile_probe_t *prof = pcpu->profc_probe; hrtime_t late; late = dtrace_gethrtime() - pcpu->profc_expected; pcpu->profc_expected += pcpu->profc_interval; dtrace_probe(prof->prof_id, CPU->cpu_profile_pc, CPU->cpu_profile_upc, late, 0, 0); } static void profile_tick(void *arg) { profile_probe_t *prof = arg; dtrace_probe(prof->prof_id, CPU->cpu_profile_pc, CPU->cpu_profile_upc, 0, 0, 0); } static void profile_create(hrtime_t interval, const char *name, int kind) { profile_probe_t *prof; int nr_frames = PROF_ARTIFICIAL_FRAMES + dtrace_mach_aframes(); if (profile_aframes) nr_frames = profile_aframes; if (interval < profile_interval_min) return; if (dtrace_probe_lookup(profile_id, NULL, NULL, name) != 0) return; - atomic_add_32(&profile_total, 1); + atomic_inc_32(&profile_total); if (profile_total > profile_max) { - atomic_add_32(&profile_total, -1); + atomic_dec_32(&profile_total); return; } prof = kmem_zalloc(sizeof (profile_probe_t), KM_SLEEP); (void) strcpy(prof->prof_name, name); prof->prof_interval = interval; prof->prof_cyclic = CYCLIC_NONE; prof->prof_kind = kind; prof->prof_id = dtrace_probe_create(profile_id, NULL, NULL, name, nr_frames, prof); } /*ARGSUSED*/ static void profile_provide(void *arg, const dtrace_probedesc_t *desc) { int i, j, rate, kind; hrtime_t val = 0, mult = 1, len; const char *name, *suffix = NULL; const struct { char *prefix; int kind; } types[] = { { PROF_PREFIX_PROFILE, PROF_PROFILE }, { PROF_PREFIX_TICK, PROF_TICK }, { NULL, NULL } }; const struct { char *name; hrtime_t mult; } suffixes[] = { { "ns", NANOSEC / NANOSEC }, { "nsec", NANOSEC / NANOSEC }, { "us", NANOSEC / MICROSEC }, { "usec", NANOSEC / MICROSEC }, { "ms", NANOSEC / MILLISEC }, { "msec", NANOSEC / MILLISEC }, { "s", NANOSEC / SEC }, { "sec", NANOSEC / SEC }, { "m", NANOSEC * (hrtime_t)60 }, { "min", NANOSEC * (hrtime_t)60 }, { "h", NANOSEC * (hrtime_t)(60 * 60) }, { "hour", NANOSEC * (hrtime_t)(60 * 60) }, { "d", NANOSEC * (hrtime_t)(24 * 60 * 60) }, { "day", NANOSEC * (hrtime_t)(24 * 60 * 60) }, { "hz", 0 }, { NULL } }; if (desc == NULL) { char n[PROF_NAMELEN]; /* * If no description was provided, provide all of our probes. */ for (i = 0; i < sizeof (profile_rates) / sizeof (int); i++) { if ((rate = profile_rates[i]) == 0) continue; (void) snprintf(n, PROF_NAMELEN, "%s%d", PROF_PREFIX_PROFILE, rate); profile_create(NANOSEC / rate, n, PROF_PROFILE); } for (i = 0; i < sizeof (profile_ticks) / sizeof (int); i++) { if ((rate = profile_ticks[i]) == 0) continue; (void) snprintf(n, PROF_NAMELEN, "%s%d", PROF_PREFIX_TICK, rate); profile_create(NANOSEC / rate, n, PROF_TICK); } return; } name = desc->dtpd_name; for (i = 0; types[i].prefix != NULL; i++) { len = strlen(types[i].prefix); if (strncmp(name, types[i].prefix, len) != 0) continue; break; } if (types[i].prefix == NULL) return; kind = types[i].kind; j = strlen(name) - len; /* * We need to start before any time suffix. */ for (j = strlen(name); j >= len; j--) { if (name[j] >= '0' && name[j] <= '9') break; suffix = &name[j]; } ASSERT(suffix != NULL); /* * Now determine the numerical value present in the probe name. */ for (; j >= len; j--) { if (name[j] < '0' || name[j] > '9') return; val += (name[j] - '0') * mult; mult *= (hrtime_t)10; } if (val == 0) return; /* * Look-up the suffix to determine the multiplier. */ for (i = 0, mult = 0; suffixes[i].name != NULL; i++) { if (strcasecmp(suffixes[i].name, suffix) == 0) { mult = suffixes[i].mult; break; } } if (suffixes[i].name == NULL && *suffix != '\0') return; if (mult == 0) { /* * The default is frequency-per-second. */ val = NANOSEC / val; } else { val *= mult; } profile_create(val, name, kind); } /*ARGSUSED*/ static void profile_destroy(void *arg, dtrace_id_t id, void *parg) { profile_probe_t *prof = parg; ASSERT(prof->prof_cyclic == CYCLIC_NONE); kmem_free(prof, sizeof (profile_probe_t)); ASSERT(profile_total >= 1); - atomic_add_32(&profile_total, -1); + atomic_dec_32(&profile_total); } /*ARGSUSED*/ static void profile_online(void *arg, cpu_t *cpu, cyc_handler_t *hdlr, cyc_time_t *when) { profile_probe_t *prof = arg; profile_probe_percpu_t *pcpu; pcpu = kmem_zalloc(sizeof (profile_probe_percpu_t), KM_SLEEP); pcpu->profc_probe = prof; hdlr->cyh_func = profile_fire; hdlr->cyh_arg = pcpu; hdlr->cyh_level = CY_HIGH_LEVEL; when->cyt_interval = prof->prof_interval; when->cyt_when = dtrace_gethrtime() + when->cyt_interval; pcpu->profc_expected = when->cyt_when; pcpu->profc_interval = when->cyt_interval; } /*ARGSUSED*/ static void profile_offline(void *arg, cpu_t *cpu, void *oarg) { profile_probe_percpu_t *pcpu = oarg; ASSERT(pcpu->profc_probe == arg); kmem_free(pcpu, sizeof (profile_probe_percpu_t)); } /*ARGSUSED*/ static void profile_enable(void *arg, dtrace_id_t id, void *parg) { profile_probe_t *prof = parg; cyc_omni_handler_t omni; cyc_handler_t hdlr; cyc_time_t when; ASSERT(prof->prof_interval != 0); ASSERT(MUTEX_HELD(&cpu_lock)); if (prof->prof_kind == PROF_TICK) { hdlr.cyh_func = profile_tick; hdlr.cyh_arg = prof; hdlr.cyh_level = CY_HIGH_LEVEL; when.cyt_interval = prof->prof_interval; when.cyt_when = dtrace_gethrtime() + when.cyt_interval; } else { ASSERT(prof->prof_kind == PROF_PROFILE); omni.cyo_online = profile_online; omni.cyo_offline = profile_offline; omni.cyo_arg = prof; } if (prof->prof_kind == PROF_TICK) { prof->prof_cyclic = cyclic_add(&hdlr, &when); } else { prof->prof_cyclic = cyclic_add_omni(&omni); } } /*ARGSUSED*/ static void profile_disable(void *arg, dtrace_id_t id, void *parg) { profile_probe_t *prof = parg; ASSERT(prof->prof_cyclic != CYCLIC_NONE); ASSERT(MUTEX_HELD(&cpu_lock)); cyclic_remove(prof->prof_cyclic); prof->prof_cyclic = CYCLIC_NONE; } /*ARGSUSED*/ static int profile_usermode(void *arg, dtrace_id_t id, void *parg) { return (CPU->cpu_profile_pc == 0); } static dtrace_pattr_t profile_attr = { { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, { DTRACE_STABILITY_UNSTABLE, DTRACE_STABILITY_UNSTABLE, DTRACE_CLASS_UNKNOWN }, { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, }; static dtrace_pops_t profile_pops = { profile_provide, NULL, profile_enable, profile_disable, NULL, NULL, NULL, NULL, profile_usermode, profile_destroy }; static int profile_attach(dev_info_t *devi, ddi_attach_cmd_t cmd) { switch (cmd) { case DDI_ATTACH: break; case DDI_RESUME: return (DDI_SUCCESS); default: return (DDI_FAILURE); } if (ddi_create_minor_node(devi, "profile", S_IFCHR, 0, DDI_PSEUDO, NULL) == DDI_FAILURE || dtrace_register("profile", &profile_attr, DTRACE_PRIV_KERNEL | DTRACE_PRIV_USER, NULL, &profile_pops, NULL, &profile_id) != 0) { ddi_remove_minor_node(devi, NULL); return (DDI_FAILURE); } profile_max = ddi_getprop(DDI_DEV_T_ANY, devi, DDI_PROP_DONTPASS, "profile-max-probes", PROFILE_MAX_DEFAULT); ddi_report_dev(devi); profile_devi = devi; return (DDI_SUCCESS); } static int profile_detach(dev_info_t *devi, ddi_detach_cmd_t cmd) { switch (cmd) { case DDI_DETACH: break; case DDI_SUSPEND: return (DDI_SUCCESS); default: return (DDI_FAILURE); } if (dtrace_unregister(profile_id) != 0) return (DDI_FAILURE); ddi_remove_minor_node(devi, NULL); return (DDI_SUCCESS); } /*ARGSUSED*/ static int profile_info(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result) { int error; switch (infocmd) { case DDI_INFO_DEVT2DEVINFO: *result = (void *)profile_devi; error = DDI_SUCCESS; break; case DDI_INFO_DEVT2INSTANCE: *result = (void *)0; error = DDI_SUCCESS; break; default: error = DDI_FAILURE; } return (error); } /*ARGSUSED*/ static int profile_open(dev_t *devp, int flag, int otyp, cred_t *cred_p) { return (0); } static struct cb_ops profile_cb_ops = { profile_open, /* open */ nodev, /* close */ nulldev, /* strategy */ nulldev, /* print */ nodev, /* dump */ nodev, /* read */ nodev, /* write */ nodev, /* ioctl */ nodev, /* devmap */ nodev, /* mmap */ nodev, /* segmap */ nochpoll, /* poll */ ddi_prop_op, /* cb_prop_op */ 0, /* streamtab */ D_NEW | D_MP /* Driver compatibility flag */ }; static struct dev_ops profile_ops = { DEVO_REV, /* devo_rev, */ 0, /* refcnt */ profile_info, /* get_dev_info */ nulldev, /* identify */ nulldev, /* probe */ profile_attach, /* attach */ profile_detach, /* detach */ nodev, /* reset */ &profile_cb_ops, /* driver operations */ NULL, /* bus operations */ nodev /* dev power */ }; /* * Module linkage information for the kernel. */ static struct modldrv modldrv = { &mod_driverops, /* module type (this is a pseudo driver) */ "Profile Interrupt Tracing", /* name of module */ &profile_ops, /* driver ops */ }; static struct modlinkage modlinkage = { MODREV_1, (void *)&modldrv, NULL }; int _init(void) { return (mod_install(&modlinkage)); } int _info(struct modinfo *modinfop) { return (mod_info(&modlinkage, modinfop)); } int _fini(void) { return (mod_remove(&modlinkage)); } Index: stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dbuf.c =================================================================== --- stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dbuf.c (revision 271000) +++ stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dbuf.c (revision 271001) @@ -1,2858 +1,2858 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright 2011 Nexenta Systems, Inc. All rights reserved. * Copyright (c) 2012, 2014 by Delphix. All rights reserved. * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. * Copyright (c) 2013, Joyent, Inc. All rights reserved. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * Number of times that zfs_free_range() took the slow path while doing * a zfs receive. A nonzero value indicates a potential performance problem. */ uint64_t zfs_free_range_recv_miss; static void dbuf_destroy(dmu_buf_impl_t *db); static boolean_t dbuf_undirty(dmu_buf_impl_t *db, dmu_tx_t *tx); static void dbuf_write(dbuf_dirty_record_t *dr, arc_buf_t *data, dmu_tx_t *tx); /* * Global data structures and functions for the dbuf cache. */ static kmem_cache_t *dbuf_cache; /* ARGSUSED */ static int dbuf_cons(void *vdb, void *unused, int kmflag) { dmu_buf_impl_t *db = vdb; bzero(db, sizeof (dmu_buf_impl_t)); mutex_init(&db->db_mtx, NULL, MUTEX_DEFAULT, NULL); cv_init(&db->db_changed, NULL, CV_DEFAULT, NULL); refcount_create(&db->db_holds); return (0); } /* ARGSUSED */ static void dbuf_dest(void *vdb, void *unused) { dmu_buf_impl_t *db = vdb; mutex_destroy(&db->db_mtx); cv_destroy(&db->db_changed); refcount_destroy(&db->db_holds); } /* * dbuf hash table routines */ static dbuf_hash_table_t dbuf_hash_table; static uint64_t dbuf_hash_count; static uint64_t dbuf_hash(void *os, uint64_t obj, uint8_t lvl, uint64_t blkid) { uintptr_t osv = (uintptr_t)os; uint64_t crc = -1ULL; ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY); crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (lvl)) & 0xFF]; crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (osv >> 6)) & 0xFF]; crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 0)) & 0xFF]; crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 8)) & 0xFF]; crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (blkid >> 0)) & 0xFF]; crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (blkid >> 8)) & 0xFF]; crc ^= (osv>>14) ^ (obj>>16) ^ (blkid>>16); return (crc); } #define DBUF_HASH(os, obj, level, blkid) dbuf_hash(os, obj, level, blkid); #define DBUF_EQUAL(dbuf, os, obj, level, blkid) \ ((dbuf)->db.db_object == (obj) && \ (dbuf)->db_objset == (os) && \ (dbuf)->db_level == (level) && \ (dbuf)->db_blkid == (blkid)) dmu_buf_impl_t * dbuf_find(dnode_t *dn, uint8_t level, uint64_t blkid) { dbuf_hash_table_t *h = &dbuf_hash_table; objset_t *os = dn->dn_objset; uint64_t obj = dn->dn_object; uint64_t hv = DBUF_HASH(os, obj, level, blkid); uint64_t idx = hv & h->hash_table_mask; dmu_buf_impl_t *db; mutex_enter(DBUF_HASH_MUTEX(h, idx)); for (db = h->hash_table[idx]; db != NULL; db = db->db_hash_next) { if (DBUF_EQUAL(db, os, obj, level, blkid)) { mutex_enter(&db->db_mtx); if (db->db_state != DB_EVICTING) { mutex_exit(DBUF_HASH_MUTEX(h, idx)); return (db); } mutex_exit(&db->db_mtx); } } mutex_exit(DBUF_HASH_MUTEX(h, idx)); return (NULL); } /* * Insert an entry into the hash table. If there is already an element * equal to elem in the hash table, then the already existing element * will be returned and the new element will not be inserted. * Otherwise returns NULL. */ static dmu_buf_impl_t * dbuf_hash_insert(dmu_buf_impl_t *db) { dbuf_hash_table_t *h = &dbuf_hash_table; objset_t *os = db->db_objset; uint64_t obj = db->db.db_object; int level = db->db_level; uint64_t blkid = db->db_blkid; uint64_t hv = DBUF_HASH(os, obj, level, blkid); uint64_t idx = hv & h->hash_table_mask; dmu_buf_impl_t *dbf; mutex_enter(DBUF_HASH_MUTEX(h, idx)); for (dbf = h->hash_table[idx]; dbf != NULL; dbf = dbf->db_hash_next) { if (DBUF_EQUAL(dbf, os, obj, level, blkid)) { mutex_enter(&dbf->db_mtx); if (dbf->db_state != DB_EVICTING) { mutex_exit(DBUF_HASH_MUTEX(h, idx)); return (dbf); } mutex_exit(&dbf->db_mtx); } } mutex_enter(&db->db_mtx); db->db_hash_next = h->hash_table[idx]; h->hash_table[idx] = db; mutex_exit(DBUF_HASH_MUTEX(h, idx)); - atomic_add_64(&dbuf_hash_count, 1); + atomic_inc_64(&dbuf_hash_count); return (NULL); } /* * Remove an entry from the hash table. It must be in the EVICTING state. */ static void dbuf_hash_remove(dmu_buf_impl_t *db) { dbuf_hash_table_t *h = &dbuf_hash_table; uint64_t hv = DBUF_HASH(db->db_objset, db->db.db_object, db->db_level, db->db_blkid); uint64_t idx = hv & h->hash_table_mask; dmu_buf_impl_t *dbf, **dbp; /* * We musn't hold db_mtx to maintain lock ordering: * DBUF_HASH_MUTEX > db_mtx. */ ASSERT(refcount_is_zero(&db->db_holds)); ASSERT(db->db_state == DB_EVICTING); ASSERT(!MUTEX_HELD(&db->db_mtx)); mutex_enter(DBUF_HASH_MUTEX(h, idx)); dbp = &h->hash_table[idx]; while ((dbf = *dbp) != db) { dbp = &dbf->db_hash_next; ASSERT(dbf != NULL); } *dbp = db->db_hash_next; db->db_hash_next = NULL; mutex_exit(DBUF_HASH_MUTEX(h, idx)); - atomic_add_64(&dbuf_hash_count, -1); + atomic_dec_64(&dbuf_hash_count); } static arc_evict_func_t dbuf_do_evict; static void dbuf_evict_user(dmu_buf_impl_t *db) { ASSERT(MUTEX_HELD(&db->db_mtx)); if (db->db_level != 0 || db->db_evict_func == NULL) return; if (db->db_user_data_ptr_ptr) *db->db_user_data_ptr_ptr = db->db.db_data; db->db_evict_func(&db->db, db->db_user_ptr); db->db_user_ptr = NULL; db->db_user_data_ptr_ptr = NULL; db->db_evict_func = NULL; } boolean_t dbuf_is_metadata(dmu_buf_impl_t *db) { if (db->db_level > 0) { return (B_TRUE); } else { boolean_t is_metadata; DB_DNODE_ENTER(db); is_metadata = DMU_OT_IS_METADATA(DB_DNODE(db)->dn_type); DB_DNODE_EXIT(db); return (is_metadata); } } void dbuf_evict(dmu_buf_impl_t *db) { ASSERT(MUTEX_HELD(&db->db_mtx)); ASSERT(db->db_buf == NULL); ASSERT(db->db_data_pending == NULL); dbuf_clear(db); dbuf_destroy(db); } void dbuf_init(void) { uint64_t hsize = 1ULL << 16; dbuf_hash_table_t *h = &dbuf_hash_table; int i; /* * The hash table is big enough to fill all of physical memory * with an average 4K block size. The table will take up * totalmem*sizeof(void*)/4K (i.e. 2MB/GB with 8-byte pointers). */ while (hsize * 4096 < (uint64_t)physmem * PAGESIZE) hsize <<= 1; retry: h->hash_table_mask = hsize - 1; h->hash_table = kmem_zalloc(hsize * sizeof (void *), KM_NOSLEEP); if (h->hash_table == NULL) { /* XXX - we should really return an error instead of assert */ ASSERT(hsize > (1ULL << 10)); hsize >>= 1; goto retry; } dbuf_cache = kmem_cache_create("dmu_buf_impl_t", sizeof (dmu_buf_impl_t), 0, dbuf_cons, dbuf_dest, NULL, NULL, NULL, 0); for (i = 0; i < DBUF_MUTEXES; i++) mutex_init(&h->hash_mutexes[i], NULL, MUTEX_DEFAULT, NULL); } void dbuf_fini(void) { dbuf_hash_table_t *h = &dbuf_hash_table; int i; for (i = 0; i < DBUF_MUTEXES; i++) mutex_destroy(&h->hash_mutexes[i]); kmem_free(h->hash_table, (h->hash_table_mask + 1) * sizeof (void *)); kmem_cache_destroy(dbuf_cache); } /* * Other stuff. */ #ifdef ZFS_DEBUG static void dbuf_verify(dmu_buf_impl_t *db) { dnode_t *dn; dbuf_dirty_record_t *dr; ASSERT(MUTEX_HELD(&db->db_mtx)); if (!(zfs_flags & ZFS_DEBUG_DBUF_VERIFY)) return; ASSERT(db->db_objset != NULL); DB_DNODE_ENTER(db); dn = DB_DNODE(db); if (dn == NULL) { ASSERT(db->db_parent == NULL); ASSERT(db->db_blkptr == NULL); } else { ASSERT3U(db->db.db_object, ==, dn->dn_object); ASSERT3P(db->db_objset, ==, dn->dn_objset); ASSERT3U(db->db_level, <, dn->dn_nlevels); ASSERT(db->db_blkid == DMU_BONUS_BLKID || db->db_blkid == DMU_SPILL_BLKID || !avl_is_empty(&dn->dn_dbufs)); } if (db->db_blkid == DMU_BONUS_BLKID) { ASSERT(dn != NULL); ASSERT3U(db->db.db_size, >=, dn->dn_bonuslen); ASSERT3U(db->db.db_offset, ==, DMU_BONUS_BLKID); } else if (db->db_blkid == DMU_SPILL_BLKID) { ASSERT(dn != NULL); ASSERT3U(db->db.db_size, >=, dn->dn_bonuslen); ASSERT0(db->db.db_offset); } else { ASSERT3U(db->db.db_offset, ==, db->db_blkid * db->db.db_size); } for (dr = db->db_data_pending; dr != NULL; dr = dr->dr_next) ASSERT(dr->dr_dbuf == db); for (dr = db->db_last_dirty; dr != NULL; dr = dr->dr_next) ASSERT(dr->dr_dbuf == db); /* * We can't assert that db_size matches dn_datablksz because it * can be momentarily different when another thread is doing * dnode_set_blksz(). */ if (db->db_level == 0 && db->db.db_object == DMU_META_DNODE_OBJECT) { dr = db->db_data_pending; /* * It should only be modified in syncing context, so * make sure we only have one copy of the data. */ ASSERT(dr == NULL || dr->dt.dl.dr_data == db->db_buf); } /* verify db->db_blkptr */ if (db->db_blkptr) { if (db->db_parent == dn->dn_dbuf) { /* db is pointed to by the dnode */ /* ASSERT3U(db->db_blkid, <, dn->dn_nblkptr); */ if (DMU_OBJECT_IS_SPECIAL(db->db.db_object)) ASSERT(db->db_parent == NULL); else ASSERT(db->db_parent != NULL); if (db->db_blkid != DMU_SPILL_BLKID) ASSERT3P(db->db_blkptr, ==, &dn->dn_phys->dn_blkptr[db->db_blkid]); } else { /* db is pointed to by an indirect block */ int epb = db->db_parent->db.db_size >> SPA_BLKPTRSHIFT; ASSERT3U(db->db_parent->db_level, ==, db->db_level+1); ASSERT3U(db->db_parent->db.db_object, ==, db->db.db_object); /* * dnode_grow_indblksz() can make this fail if we don't * have the struct_rwlock. XXX indblksz no longer * grows. safe to do this now? */ if (RW_WRITE_HELD(&dn->dn_struct_rwlock)) { ASSERT3P(db->db_blkptr, ==, ((blkptr_t *)db->db_parent->db.db_data + db->db_blkid % epb)); } } } if ((db->db_blkptr == NULL || BP_IS_HOLE(db->db_blkptr)) && (db->db_buf == NULL || db->db_buf->b_data) && db->db.db_data && db->db_blkid != DMU_BONUS_BLKID && db->db_state != DB_FILL && !dn->dn_free_txg) { /* * If the blkptr isn't set but they have nonzero data, * it had better be dirty, otherwise we'll lose that * data when we evict this buffer. */ if (db->db_dirtycnt == 0) { uint64_t *buf = db->db.db_data; int i; for (i = 0; i < db->db.db_size >> 3; i++) { ASSERT(buf[i] == 0); } } } DB_DNODE_EXIT(db); } #endif static void dbuf_update_data(dmu_buf_impl_t *db) { ASSERT(MUTEX_HELD(&db->db_mtx)); if (db->db_level == 0 && db->db_user_data_ptr_ptr) { ASSERT(!refcount_is_zero(&db->db_holds)); *db->db_user_data_ptr_ptr = db->db.db_data; } } static void dbuf_set_data(dmu_buf_impl_t *db, arc_buf_t *buf) { ASSERT(MUTEX_HELD(&db->db_mtx)); db->db_buf = buf; if (buf != NULL) { ASSERT(buf->b_data != NULL); db->db.db_data = buf->b_data; if (!arc_released(buf)) arc_set_callback(buf, dbuf_do_evict, db); dbuf_update_data(db); } else { dbuf_evict_user(db); db->db.db_data = NULL; if (db->db_state != DB_NOFILL) db->db_state = DB_UNCACHED; } } /* * Loan out an arc_buf for read. Return the loaned arc_buf. */ arc_buf_t * dbuf_loan_arcbuf(dmu_buf_impl_t *db) { arc_buf_t *abuf; mutex_enter(&db->db_mtx); if (arc_released(db->db_buf) || refcount_count(&db->db_holds) > 1) { int blksz = db->db.db_size; spa_t *spa = db->db_objset->os_spa; mutex_exit(&db->db_mtx); abuf = arc_loan_buf(spa, blksz); bcopy(db->db.db_data, abuf->b_data, blksz); } else { abuf = db->db_buf; arc_loan_inuse_buf(abuf, db); dbuf_set_data(db, NULL); mutex_exit(&db->db_mtx); } return (abuf); } uint64_t dbuf_whichblock(dnode_t *dn, uint64_t offset) { if (dn->dn_datablkshift) { return (offset >> dn->dn_datablkshift); } else { ASSERT3U(offset, <, dn->dn_datablksz); return (0); } } static void dbuf_read_done(zio_t *zio, arc_buf_t *buf, void *vdb) { dmu_buf_impl_t *db = vdb; mutex_enter(&db->db_mtx); ASSERT3U(db->db_state, ==, DB_READ); /* * All reads are synchronous, so we must have a hold on the dbuf */ ASSERT(refcount_count(&db->db_holds) > 0); ASSERT(db->db_buf == NULL); ASSERT(db->db.db_data == NULL); if (db->db_level == 0 && db->db_freed_in_flight) { /* we were freed in flight; disregard any error */ arc_release(buf, db); bzero(buf->b_data, db->db.db_size); arc_buf_freeze(buf); db->db_freed_in_flight = FALSE; dbuf_set_data(db, buf); db->db_state = DB_CACHED; } else if (zio == NULL || zio->io_error == 0) { dbuf_set_data(db, buf); db->db_state = DB_CACHED; } else { ASSERT(db->db_blkid != DMU_BONUS_BLKID); ASSERT3P(db->db_buf, ==, NULL); VERIFY(arc_buf_remove_ref(buf, db)); db->db_state = DB_UNCACHED; } cv_broadcast(&db->db_changed); dbuf_rele_and_unlock(db, NULL); } static void dbuf_read_impl(dmu_buf_impl_t *db, zio_t *zio, uint32_t *flags) { dnode_t *dn; zbookmark_phys_t zb; uint32_t aflags = ARC_NOWAIT; DB_DNODE_ENTER(db); dn = DB_DNODE(db); ASSERT(!refcount_is_zero(&db->db_holds)); /* We need the struct_rwlock to prevent db_blkptr from changing. */ ASSERT(RW_LOCK_HELD(&dn->dn_struct_rwlock)); ASSERT(MUTEX_HELD(&db->db_mtx)); ASSERT(db->db_state == DB_UNCACHED); ASSERT(db->db_buf == NULL); if (db->db_blkid == DMU_BONUS_BLKID) { int bonuslen = MIN(dn->dn_bonuslen, dn->dn_phys->dn_bonuslen); ASSERT3U(bonuslen, <=, db->db.db_size); db->db.db_data = zio_buf_alloc(DN_MAX_BONUSLEN); arc_space_consume(DN_MAX_BONUSLEN, ARC_SPACE_OTHER); if (bonuslen < DN_MAX_BONUSLEN) bzero(db->db.db_data, DN_MAX_BONUSLEN); if (bonuslen) bcopy(DN_BONUS(dn->dn_phys), db->db.db_data, bonuslen); DB_DNODE_EXIT(db); dbuf_update_data(db); db->db_state = DB_CACHED; mutex_exit(&db->db_mtx); return; } /* * Recheck BP_IS_HOLE() after dnode_block_freed() in case dnode_sync() * processes the delete record and clears the bp while we are waiting * for the dn_mtx (resulting in a "no" from block_freed). */ if (db->db_blkptr == NULL || BP_IS_HOLE(db->db_blkptr) || (db->db_level == 0 && (dnode_block_freed(dn, db->db_blkid) || BP_IS_HOLE(db->db_blkptr)))) { arc_buf_contents_t type = DBUF_GET_BUFC_TYPE(db); DB_DNODE_EXIT(db); dbuf_set_data(db, arc_buf_alloc(db->db_objset->os_spa, db->db.db_size, db, type)); bzero(db->db.db_data, db->db.db_size); db->db_state = DB_CACHED; *flags |= DB_RF_CACHED; mutex_exit(&db->db_mtx); return; } DB_DNODE_EXIT(db); db->db_state = DB_READ; mutex_exit(&db->db_mtx); if (DBUF_IS_L2CACHEABLE(db)) aflags |= ARC_L2CACHE; if (DBUF_IS_L2COMPRESSIBLE(db)) aflags |= ARC_L2COMPRESS; SET_BOOKMARK(&zb, db->db_objset->os_dsl_dataset ? db->db_objset->os_dsl_dataset->ds_object : DMU_META_OBJSET, db->db.db_object, db->db_level, db->db_blkid); dbuf_add_ref(db, NULL); (void) arc_read(zio, db->db_objset->os_spa, db->db_blkptr, dbuf_read_done, db, ZIO_PRIORITY_SYNC_READ, (*flags & DB_RF_CANFAIL) ? ZIO_FLAG_CANFAIL : ZIO_FLAG_MUSTSUCCEED, &aflags, &zb); if (aflags & ARC_CACHED) *flags |= DB_RF_CACHED; } int dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags) { int err = 0; boolean_t havepzio = (zio != NULL); boolean_t prefetch; dnode_t *dn; /* * We don't have to hold the mutex to check db_state because it * can't be freed while we have a hold on the buffer. */ ASSERT(!refcount_is_zero(&db->db_holds)); if (db->db_state == DB_NOFILL) return (SET_ERROR(EIO)); DB_DNODE_ENTER(db); dn = DB_DNODE(db); if ((flags & DB_RF_HAVESTRUCT) == 0) rw_enter(&dn->dn_struct_rwlock, RW_READER); prefetch = db->db_level == 0 && db->db_blkid != DMU_BONUS_BLKID && (flags & DB_RF_NOPREFETCH) == 0 && dn != NULL && DBUF_IS_CACHEABLE(db); mutex_enter(&db->db_mtx); if (db->db_state == DB_CACHED) { mutex_exit(&db->db_mtx); if (prefetch) dmu_zfetch(&dn->dn_zfetch, db->db.db_offset, db->db.db_size, TRUE); if ((flags & DB_RF_HAVESTRUCT) == 0) rw_exit(&dn->dn_struct_rwlock); DB_DNODE_EXIT(db); } else if (db->db_state == DB_UNCACHED) { spa_t *spa = dn->dn_objset->os_spa; if (zio == NULL) zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL); dbuf_read_impl(db, zio, &flags); /* dbuf_read_impl has dropped db_mtx for us */ if (prefetch) dmu_zfetch(&dn->dn_zfetch, db->db.db_offset, db->db.db_size, flags & DB_RF_CACHED); if ((flags & DB_RF_HAVESTRUCT) == 0) rw_exit(&dn->dn_struct_rwlock); DB_DNODE_EXIT(db); if (!havepzio) err = zio_wait(zio); } else { /* * Another reader came in while the dbuf was in flight * between UNCACHED and CACHED. Either a writer will finish * writing the buffer (sending the dbuf to CACHED) or the * first reader's request will reach the read_done callback * and send the dbuf to CACHED. Otherwise, a failure * occurred and the dbuf went to UNCACHED. */ mutex_exit(&db->db_mtx); if (prefetch) dmu_zfetch(&dn->dn_zfetch, db->db.db_offset, db->db.db_size, TRUE); if ((flags & DB_RF_HAVESTRUCT) == 0) rw_exit(&dn->dn_struct_rwlock); DB_DNODE_EXIT(db); /* Skip the wait per the caller's request. */ mutex_enter(&db->db_mtx); if ((flags & DB_RF_NEVERWAIT) == 0) { while (db->db_state == DB_READ || db->db_state == DB_FILL) { ASSERT(db->db_state == DB_READ || (flags & DB_RF_HAVESTRUCT) == 0); cv_wait(&db->db_changed, &db->db_mtx); } if (db->db_state == DB_UNCACHED) err = SET_ERROR(EIO); } mutex_exit(&db->db_mtx); } ASSERT(err || havepzio || db->db_state == DB_CACHED); return (err); } static void dbuf_noread(dmu_buf_impl_t *db) { ASSERT(!refcount_is_zero(&db->db_holds)); ASSERT(db->db_blkid != DMU_BONUS_BLKID); mutex_enter(&db->db_mtx); while (db->db_state == DB_READ || db->db_state == DB_FILL) cv_wait(&db->db_changed, &db->db_mtx); if (db->db_state == DB_UNCACHED) { arc_buf_contents_t type = DBUF_GET_BUFC_TYPE(db); spa_t *spa = db->db_objset->os_spa; ASSERT(db->db_buf == NULL); ASSERT(db->db.db_data == NULL); dbuf_set_data(db, arc_buf_alloc(spa, db->db.db_size, db, type)); db->db_state = DB_FILL; } else if (db->db_state == DB_NOFILL) { dbuf_set_data(db, NULL); } else { ASSERT3U(db->db_state, ==, DB_CACHED); } mutex_exit(&db->db_mtx); } /* * This is our just-in-time copy function. It makes a copy of * buffers, that have been modified in a previous transaction * group, before we modify them in the current active group. * * This function is used in two places: when we are dirtying a * buffer for the first time in a txg, and when we are freeing * a range in a dnode that includes this buffer. * * Note that when we are called from dbuf_free_range() we do * not put a hold on the buffer, we just traverse the active * dbuf list for the dnode. */ static void dbuf_fix_old_data(dmu_buf_impl_t *db, uint64_t txg) { dbuf_dirty_record_t *dr = db->db_last_dirty; ASSERT(MUTEX_HELD(&db->db_mtx)); ASSERT(db->db.db_data != NULL); ASSERT(db->db_level == 0); ASSERT(db->db.db_object != DMU_META_DNODE_OBJECT); if (dr == NULL || (dr->dt.dl.dr_data != ((db->db_blkid == DMU_BONUS_BLKID) ? db->db.db_data : db->db_buf))) return; /* * If the last dirty record for this dbuf has not yet synced * and its referencing the dbuf data, either: * reset the reference to point to a new copy, * or (if there a no active holders) * just null out the current db_data pointer. */ ASSERT(dr->dr_txg >= txg - 2); if (db->db_blkid == DMU_BONUS_BLKID) { /* Note that the data bufs here are zio_bufs */ dr->dt.dl.dr_data = zio_buf_alloc(DN_MAX_BONUSLEN); arc_space_consume(DN_MAX_BONUSLEN, ARC_SPACE_OTHER); bcopy(db->db.db_data, dr->dt.dl.dr_data, DN_MAX_BONUSLEN); } else if (refcount_count(&db->db_holds) > db->db_dirtycnt) { int size = db->db.db_size; arc_buf_contents_t type = DBUF_GET_BUFC_TYPE(db); spa_t *spa = db->db_objset->os_spa; dr->dt.dl.dr_data = arc_buf_alloc(spa, size, db, type); bcopy(db->db.db_data, dr->dt.dl.dr_data->b_data, size); } else { dbuf_set_data(db, NULL); } } void dbuf_unoverride(dbuf_dirty_record_t *dr) { dmu_buf_impl_t *db = dr->dr_dbuf; blkptr_t *bp = &dr->dt.dl.dr_overridden_by; uint64_t txg = dr->dr_txg; ASSERT(MUTEX_HELD(&db->db_mtx)); ASSERT(dr->dt.dl.dr_override_state != DR_IN_DMU_SYNC); ASSERT(db->db_level == 0); if (db->db_blkid == DMU_BONUS_BLKID || dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN) return; ASSERT(db->db_data_pending != dr); /* free this block */ if (!BP_IS_HOLE(bp) && !dr->dt.dl.dr_nopwrite) zio_free(db->db_objset->os_spa, txg, bp); dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN; dr->dt.dl.dr_nopwrite = B_FALSE; /* * Release the already-written buffer, so we leave it in * a consistent dirty state. Note that all callers are * modifying the buffer, so they will immediately do * another (redundant) arc_release(). Therefore, leave * the buf thawed to save the effort of freezing & * immediately re-thawing it. */ arc_release(dr->dt.dl.dr_data, db); } /* * Evict (if its unreferenced) or clear (if its referenced) any level-0 * data blocks in the free range, so that any future readers will find * empty blocks. * * This is a no-op if the dataset is in the middle of an incremental * receive; see comment below for details. */ void dbuf_free_range(dnode_t *dn, uint64_t start_blkid, uint64_t end_blkid, dmu_tx_t *tx) { dmu_buf_impl_t *db, *db_next, db_search; uint64_t txg = tx->tx_txg; avl_index_t where; if (end_blkid > dn->dn_maxblkid && (end_blkid != DMU_SPILL_BLKID)) end_blkid = dn->dn_maxblkid; dprintf_dnode(dn, "start=%llu end=%llu\n", start_blkid, end_blkid); db_search.db_level = 0; db_search.db_blkid = start_blkid; db_search.db_state = DB_SEARCH; mutex_enter(&dn->dn_dbufs_mtx); if (start_blkid >= dn->dn_unlisted_l0_blkid) { /* There can't be any dbufs in this range; no need to search. */ #ifdef DEBUG db = avl_find(&dn->dn_dbufs, &db_search, &where); ASSERT3P(db, ==, NULL); db = avl_nearest(&dn->dn_dbufs, where, AVL_AFTER); ASSERT(db == NULL || db->db_level > 0); #endif mutex_exit(&dn->dn_dbufs_mtx); return; } else if (dmu_objset_is_receiving(dn->dn_objset)) { /* * If we are receiving, we expect there to be no dbufs in * the range to be freed, because receive modifies each * block at most once, and in offset order. If this is * not the case, it can lead to performance problems, * so note that we unexpectedly took the slow path. */ atomic_inc_64(&zfs_free_range_recv_miss); } db = avl_find(&dn->dn_dbufs, &db_search, &where); ASSERT3P(db, ==, NULL); db = avl_nearest(&dn->dn_dbufs, where, AVL_AFTER); for (; db != NULL; db = db_next) { db_next = AVL_NEXT(&dn->dn_dbufs, db); ASSERT(db->db_blkid != DMU_BONUS_BLKID); if (db->db_level != 0 || db->db_blkid > end_blkid) { break; } ASSERT3U(db->db_blkid, >=, start_blkid); /* found a level 0 buffer in the range */ mutex_enter(&db->db_mtx); if (dbuf_undirty(db, tx)) { /* mutex has been dropped and dbuf destroyed */ continue; } if (db->db_state == DB_UNCACHED || db->db_state == DB_NOFILL || db->db_state == DB_EVICTING) { ASSERT(db->db.db_data == NULL); mutex_exit(&db->db_mtx); continue; } if (db->db_state == DB_READ || db->db_state == DB_FILL) { /* will be handled in dbuf_read_done or dbuf_rele */ db->db_freed_in_flight = TRUE; mutex_exit(&db->db_mtx); continue; } if (refcount_count(&db->db_holds) == 0) { ASSERT(db->db_buf); dbuf_clear(db); continue; } /* The dbuf is referenced */ if (db->db_last_dirty != NULL) { dbuf_dirty_record_t *dr = db->db_last_dirty; if (dr->dr_txg == txg) { /* * This buffer is "in-use", re-adjust the file * size to reflect that this buffer may * contain new data when we sync. */ if (db->db_blkid != DMU_SPILL_BLKID && db->db_blkid > dn->dn_maxblkid) dn->dn_maxblkid = db->db_blkid; dbuf_unoverride(dr); } else { /* * This dbuf is not dirty in the open context. * Either uncache it (if its not referenced in * the open context) or reset its contents to * empty. */ dbuf_fix_old_data(db, txg); } } /* clear the contents if its cached */ if (db->db_state == DB_CACHED) { ASSERT(db->db.db_data != NULL); arc_release(db->db_buf, db); bzero(db->db.db_data, db->db.db_size); arc_buf_freeze(db->db_buf); } mutex_exit(&db->db_mtx); } mutex_exit(&dn->dn_dbufs_mtx); } static int dbuf_block_freeable(dmu_buf_impl_t *db) { dsl_dataset_t *ds = db->db_objset->os_dsl_dataset; uint64_t birth_txg = 0; /* * We don't need any locking to protect db_blkptr: * If it's syncing, then db_last_dirty will be set * so we'll ignore db_blkptr. * * This logic ensures that only block births for * filled blocks are considered. */ ASSERT(MUTEX_HELD(&db->db_mtx)); if (db->db_last_dirty && (db->db_blkptr == NULL || !BP_IS_HOLE(db->db_blkptr))) { birth_txg = db->db_last_dirty->dr_txg; } else if (db->db_blkptr != NULL && !BP_IS_HOLE(db->db_blkptr)) { birth_txg = db->db_blkptr->blk_birth; } /* * If this block don't exist or is in a snapshot, it can't be freed. * Don't pass the bp to dsl_dataset_block_freeable() since we * are holding the db_mtx lock and might deadlock if we are * prefetching a dedup-ed block. */ if (birth_txg != 0) return (ds == NULL || dsl_dataset_block_freeable(ds, NULL, birth_txg)); else return (B_FALSE); } void dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx) { arc_buf_t *buf, *obuf; int osize = db->db.db_size; arc_buf_contents_t type = DBUF_GET_BUFC_TYPE(db); dnode_t *dn; ASSERT(db->db_blkid != DMU_BONUS_BLKID); DB_DNODE_ENTER(db); dn = DB_DNODE(db); /* XXX does *this* func really need the lock? */ ASSERT(RW_WRITE_HELD(&dn->dn_struct_rwlock)); /* * This call to dmu_buf_will_dirty() with the dn_struct_rwlock held * is OK, because there can be no other references to the db * when we are changing its size, so no concurrent DB_FILL can * be happening. */ /* * XXX we should be doing a dbuf_read, checking the return * value and returning that up to our callers */ dmu_buf_will_dirty(&db->db, tx); /* create the data buffer for the new block */ buf = arc_buf_alloc(dn->dn_objset->os_spa, size, db, type); /* copy old block data to the new block */ obuf = db->db_buf; bcopy(obuf->b_data, buf->b_data, MIN(osize, size)); /* zero the remainder */ if (size > osize) bzero((uint8_t *)buf->b_data + osize, size - osize); mutex_enter(&db->db_mtx); dbuf_set_data(db, buf); VERIFY(arc_buf_remove_ref(obuf, db)); db->db.db_size = size; if (db->db_level == 0) { ASSERT3U(db->db_last_dirty->dr_txg, ==, tx->tx_txg); db->db_last_dirty->dt.dl.dr_data = buf; } mutex_exit(&db->db_mtx); dnode_willuse_space(dn, size-osize, tx); DB_DNODE_EXIT(db); } void dbuf_release_bp(dmu_buf_impl_t *db) { objset_t *os = db->db_objset; ASSERT(dsl_pool_sync_context(dmu_objset_pool(os))); ASSERT(arc_released(os->os_phys_buf) || list_link_active(&os->os_dsl_dataset->ds_synced_link)); ASSERT(db->db_parent == NULL || arc_released(db->db_parent->db_buf)); (void) arc_release(db->db_buf, db); } dbuf_dirty_record_t * dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx) { dnode_t *dn; objset_t *os; dbuf_dirty_record_t **drp, *dr; int drop_struct_lock = FALSE; boolean_t do_free_accounting = B_FALSE; int txgoff = tx->tx_txg & TXG_MASK; ASSERT(tx->tx_txg != 0); ASSERT(!refcount_is_zero(&db->db_holds)); DMU_TX_DIRTY_BUF(tx, db); DB_DNODE_ENTER(db); dn = DB_DNODE(db); /* * Shouldn't dirty a regular buffer in syncing context. Private * objects may be dirtied in syncing context, but only if they * were already pre-dirtied in open context. */ ASSERT(!dmu_tx_is_syncing(tx) || BP_IS_HOLE(dn->dn_objset->os_rootbp) || DMU_OBJECT_IS_SPECIAL(dn->dn_object) || dn->dn_objset->os_dsl_dataset == NULL); /* * We make this assert for private objects as well, but after we * check if we're already dirty. They are allowed to re-dirty * in syncing context. */ ASSERT(dn->dn_object == DMU_META_DNODE_OBJECT || dn->dn_dirtyctx == DN_UNDIRTIED || dn->dn_dirtyctx == (dmu_tx_is_syncing(tx) ? DN_DIRTY_SYNC : DN_DIRTY_OPEN)); mutex_enter(&db->db_mtx); /* * XXX make this true for indirects too? The problem is that * transactions created with dmu_tx_create_assigned() from * syncing context don't bother holding ahead. */ ASSERT(db->db_level != 0 || db->db_state == DB_CACHED || db->db_state == DB_FILL || db->db_state == DB_NOFILL); mutex_enter(&dn->dn_mtx); /* * Don't set dirtyctx to SYNC if we're just modifying this as we * initialize the objset. */ if (dn->dn_dirtyctx == DN_UNDIRTIED && !BP_IS_HOLE(dn->dn_objset->os_rootbp)) { dn->dn_dirtyctx = (dmu_tx_is_syncing(tx) ? DN_DIRTY_SYNC : DN_DIRTY_OPEN); ASSERT(dn->dn_dirtyctx_firstset == NULL); dn->dn_dirtyctx_firstset = kmem_alloc(1, KM_SLEEP); } mutex_exit(&dn->dn_mtx); if (db->db_blkid == DMU_SPILL_BLKID) dn->dn_have_spill = B_TRUE; /* * If this buffer is already dirty, we're done. */ drp = &db->db_last_dirty; ASSERT(*drp == NULL || (*drp)->dr_txg <= tx->tx_txg || db->db.db_object == DMU_META_DNODE_OBJECT); while ((dr = *drp) != NULL && dr->dr_txg > tx->tx_txg) drp = &dr->dr_next; if (dr && dr->dr_txg == tx->tx_txg) { DB_DNODE_EXIT(db); if (db->db_level == 0 && db->db_blkid != DMU_BONUS_BLKID) { /* * If this buffer has already been written out, * we now need to reset its state. */ dbuf_unoverride(dr); if (db->db.db_object != DMU_META_DNODE_OBJECT && db->db_state != DB_NOFILL) arc_buf_thaw(db->db_buf); } mutex_exit(&db->db_mtx); return (dr); } /* * Only valid if not already dirty. */ ASSERT(dn->dn_object == 0 || dn->dn_dirtyctx == DN_UNDIRTIED || dn->dn_dirtyctx == (dmu_tx_is_syncing(tx) ? DN_DIRTY_SYNC : DN_DIRTY_OPEN)); ASSERT3U(dn->dn_nlevels, >, db->db_level); ASSERT((dn->dn_phys->dn_nlevels == 0 && db->db_level == 0) || dn->dn_phys->dn_nlevels > db->db_level || dn->dn_next_nlevels[txgoff] > db->db_level || dn->dn_next_nlevels[(tx->tx_txg-1) & TXG_MASK] > db->db_level || dn->dn_next_nlevels[(tx->tx_txg-2) & TXG_MASK] > db->db_level); /* * We should only be dirtying in syncing context if it's the * mos or we're initializing the os or it's a special object. * However, we are allowed to dirty in syncing context provided * we already dirtied it in open context. Hence we must make * this assertion only if we're not already dirty. */ os = dn->dn_objset; ASSERT(!dmu_tx_is_syncing(tx) || DMU_OBJECT_IS_SPECIAL(dn->dn_object) || os->os_dsl_dataset == NULL || BP_IS_HOLE(os->os_rootbp)); ASSERT(db->db.db_size != 0); dprintf_dbuf(db, "size=%llx\n", (u_longlong_t)db->db.db_size); if (db->db_blkid != DMU_BONUS_BLKID) { /* * Update the accounting. * Note: we delay "free accounting" until after we drop * the db_mtx. This keeps us from grabbing other locks * (and possibly deadlocking) in bp_get_dsize() while * also holding the db_mtx. */ dnode_willuse_space(dn, db->db.db_size, tx); do_free_accounting = dbuf_block_freeable(db); } /* * If this buffer is dirty in an old transaction group we need * to make a copy of it so that the changes we make in this * transaction group won't leak out when we sync the older txg. */ dr = kmem_zalloc(sizeof (dbuf_dirty_record_t), KM_SLEEP); if (db->db_level == 0) { void *data_old = db->db_buf; if (db->db_state != DB_NOFILL) { if (db->db_blkid == DMU_BONUS_BLKID) { dbuf_fix_old_data(db, tx->tx_txg); data_old = db->db.db_data; } else if (db->db.db_object != DMU_META_DNODE_OBJECT) { /* * Release the data buffer from the cache so * that we can modify it without impacting * possible other users of this cached data * block. Note that indirect blocks and * private objects are not released until the * syncing state (since they are only modified * then). */ arc_release(db->db_buf, db); dbuf_fix_old_data(db, tx->tx_txg); data_old = db->db_buf; } ASSERT(data_old != NULL); } dr->dt.dl.dr_data = data_old; } else { mutex_init(&dr->dt.di.dr_mtx, NULL, MUTEX_DEFAULT, NULL); list_create(&dr->dt.di.dr_children, sizeof (dbuf_dirty_record_t), offsetof(dbuf_dirty_record_t, dr_dirty_node)); } if (db->db_blkid != DMU_BONUS_BLKID && os->os_dsl_dataset != NULL) dr->dr_accounted = db->db.db_size; dr->dr_dbuf = db; dr->dr_txg = tx->tx_txg; dr->dr_next = *drp; *drp = dr; /* * We could have been freed_in_flight between the dbuf_noread * and dbuf_dirty. We win, as though the dbuf_noread() had * happened after the free. */ if (db->db_level == 0 && db->db_blkid != DMU_BONUS_BLKID && db->db_blkid != DMU_SPILL_BLKID) { mutex_enter(&dn->dn_mtx); if (dn->dn_free_ranges[txgoff] != NULL) { range_tree_clear(dn->dn_free_ranges[txgoff], db->db_blkid, 1); } mutex_exit(&dn->dn_mtx); db->db_freed_in_flight = FALSE; } /* * This buffer is now part of this txg */ dbuf_add_ref(db, (void *)(uintptr_t)tx->tx_txg); db->db_dirtycnt += 1; ASSERT3U(db->db_dirtycnt, <=, 3); mutex_exit(&db->db_mtx); if (db->db_blkid == DMU_BONUS_BLKID || db->db_blkid == DMU_SPILL_BLKID) { mutex_enter(&dn->dn_mtx); ASSERT(!list_link_active(&dr->dr_dirty_node)); list_insert_tail(&dn->dn_dirty_records[txgoff], dr); mutex_exit(&dn->dn_mtx); dnode_setdirty(dn, tx); DB_DNODE_EXIT(db); return (dr); } else if (do_free_accounting) { blkptr_t *bp = db->db_blkptr; int64_t willfree = (bp && !BP_IS_HOLE(bp)) ? bp_get_dsize(os->os_spa, bp) : db->db.db_size; /* * This is only a guess -- if the dbuf is dirty * in a previous txg, we don't know how much * space it will use on disk yet. We should * really have the struct_rwlock to access * db_blkptr, but since this is just a guess, * it's OK if we get an odd answer. */ ddt_prefetch(os->os_spa, bp); dnode_willuse_space(dn, -willfree, tx); } if (!RW_WRITE_HELD(&dn->dn_struct_rwlock)) { rw_enter(&dn->dn_struct_rwlock, RW_READER); drop_struct_lock = TRUE; } if (db->db_level == 0) { dnode_new_blkid(dn, db->db_blkid, tx, drop_struct_lock); ASSERT(dn->dn_maxblkid >= db->db_blkid); } if (db->db_level+1 < dn->dn_nlevels) { dmu_buf_impl_t *parent = db->db_parent; dbuf_dirty_record_t *di; int parent_held = FALSE; if (db->db_parent == NULL || db->db_parent == dn->dn_dbuf) { int epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; parent = dbuf_hold_level(dn, db->db_level+1, db->db_blkid >> epbs, FTAG); ASSERT(parent != NULL); parent_held = TRUE; } if (drop_struct_lock) rw_exit(&dn->dn_struct_rwlock); ASSERT3U(db->db_level+1, ==, parent->db_level); di = dbuf_dirty(parent, tx); if (parent_held) dbuf_rele(parent, FTAG); mutex_enter(&db->db_mtx); /* * Since we've dropped the mutex, it's possible that * dbuf_undirty() might have changed this out from under us. */ if (db->db_last_dirty == dr || dn->dn_object == DMU_META_DNODE_OBJECT) { mutex_enter(&di->dt.di.dr_mtx); ASSERT3U(di->dr_txg, ==, tx->tx_txg); ASSERT(!list_link_active(&dr->dr_dirty_node)); list_insert_tail(&di->dt.di.dr_children, dr); mutex_exit(&di->dt.di.dr_mtx); dr->dr_parent = di; } mutex_exit(&db->db_mtx); } else { ASSERT(db->db_level+1 == dn->dn_nlevels); ASSERT(db->db_blkid < dn->dn_nblkptr); ASSERT(db->db_parent == NULL || db->db_parent == dn->dn_dbuf); mutex_enter(&dn->dn_mtx); ASSERT(!list_link_active(&dr->dr_dirty_node)); list_insert_tail(&dn->dn_dirty_records[txgoff], dr); mutex_exit(&dn->dn_mtx); if (drop_struct_lock) rw_exit(&dn->dn_struct_rwlock); } dnode_setdirty(dn, tx); DB_DNODE_EXIT(db); return (dr); } /* * Undirty a buffer in the transaction group referenced by the given * transaction. Return whether this evicted the dbuf. */ static boolean_t dbuf_undirty(dmu_buf_impl_t *db, dmu_tx_t *tx) { dnode_t *dn; uint64_t txg = tx->tx_txg; dbuf_dirty_record_t *dr, **drp; ASSERT(txg != 0); ASSERT(db->db_blkid != DMU_BONUS_BLKID); ASSERT0(db->db_level); ASSERT(MUTEX_HELD(&db->db_mtx)); /* * If this buffer is not dirty, we're done. */ for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next) if (dr->dr_txg <= txg) break; if (dr == NULL || dr->dr_txg < txg) return (B_FALSE); ASSERT(dr->dr_txg == txg); ASSERT(dr->dr_dbuf == db); DB_DNODE_ENTER(db); dn = DB_DNODE(db); dprintf_dbuf(db, "size=%llx\n", (u_longlong_t)db->db.db_size); ASSERT(db->db.db_size != 0); /* * Any space we accounted for in dp_dirty_* will be cleaned up by * dsl_pool_sync(). This is relatively rare so the discrepancy * is not a big deal. */ *drp = dr->dr_next; /* * Note that there are three places in dbuf_dirty() * where this dirty record may be put on a list. * Make sure to do a list_remove corresponding to * every one of those list_insert calls. */ if (dr->dr_parent) { mutex_enter(&dr->dr_parent->dt.di.dr_mtx); list_remove(&dr->dr_parent->dt.di.dr_children, dr); mutex_exit(&dr->dr_parent->dt.di.dr_mtx); } else if (db->db_blkid == DMU_SPILL_BLKID || db->db_level+1 == dn->dn_nlevels) { ASSERT(db->db_blkptr == NULL || db->db_parent == dn->dn_dbuf); mutex_enter(&dn->dn_mtx); list_remove(&dn->dn_dirty_records[txg & TXG_MASK], dr); mutex_exit(&dn->dn_mtx); } DB_DNODE_EXIT(db); if (db->db_state != DB_NOFILL) { dbuf_unoverride(dr); ASSERT(db->db_buf != NULL); ASSERT(dr->dt.dl.dr_data != NULL); if (dr->dt.dl.dr_data != db->db_buf) VERIFY(arc_buf_remove_ref(dr->dt.dl.dr_data, db)); } if (db->db_level != 0) { mutex_destroy(&dr->dt.di.dr_mtx); list_destroy(&dr->dt.di.dr_children); } kmem_free(dr, sizeof (dbuf_dirty_record_t)); ASSERT(db->db_dirtycnt > 0); db->db_dirtycnt -= 1; if (refcount_remove(&db->db_holds, (void *)(uintptr_t)txg) == 0) { arc_buf_t *buf = db->db_buf; ASSERT(db->db_state == DB_NOFILL || arc_released(buf)); dbuf_set_data(db, NULL); VERIFY(arc_buf_remove_ref(buf, db)); dbuf_evict(db); return (B_TRUE); } return (B_FALSE); } void dmu_buf_will_dirty(dmu_buf_t *db_fake, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; int rf = DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH; ASSERT(tx->tx_txg != 0); ASSERT(!refcount_is_zero(&db->db_holds)); DB_DNODE_ENTER(db); if (RW_WRITE_HELD(&DB_DNODE(db)->dn_struct_rwlock)) rf |= DB_RF_HAVESTRUCT; DB_DNODE_EXIT(db); (void) dbuf_read(db, NULL, rf); (void) dbuf_dirty(db, tx); } void dmu_buf_will_not_fill(dmu_buf_t *db_fake, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; db->db_state = DB_NOFILL; dmu_buf_will_fill(db_fake, tx); } void dmu_buf_will_fill(dmu_buf_t *db_fake, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; ASSERT(db->db_blkid != DMU_BONUS_BLKID); ASSERT(tx->tx_txg != 0); ASSERT(db->db_level == 0); ASSERT(!refcount_is_zero(&db->db_holds)); ASSERT(db->db.db_object != DMU_META_DNODE_OBJECT || dmu_tx_private_ok(tx)); dbuf_noread(db); (void) dbuf_dirty(db, tx); } #pragma weak dmu_buf_fill_done = dbuf_fill_done /* ARGSUSED */ void dbuf_fill_done(dmu_buf_impl_t *db, dmu_tx_t *tx) { mutex_enter(&db->db_mtx); DBUF_VERIFY(db); if (db->db_state == DB_FILL) { if (db->db_level == 0 && db->db_freed_in_flight) { ASSERT(db->db_blkid != DMU_BONUS_BLKID); /* we were freed while filling */ /* XXX dbuf_undirty? */ bzero(db->db.db_data, db->db.db_size); db->db_freed_in_flight = FALSE; } db->db_state = DB_CACHED; cv_broadcast(&db->db_changed); } mutex_exit(&db->db_mtx); } void dmu_buf_write_embedded(dmu_buf_t *dbuf, void *data, bp_embedded_type_t etype, enum zio_compress comp, int uncompressed_size, int compressed_size, int byteorder, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbuf; struct dirty_leaf *dl; dmu_object_type_t type; DB_DNODE_ENTER(db); type = DB_DNODE(db)->dn_type; DB_DNODE_EXIT(db); ASSERT0(db->db_level); ASSERT(db->db_blkid != DMU_BONUS_BLKID); dmu_buf_will_not_fill(dbuf, tx); ASSERT3U(db->db_last_dirty->dr_txg, ==, tx->tx_txg); dl = &db->db_last_dirty->dt.dl; encode_embedded_bp_compressed(&dl->dr_overridden_by, data, comp, uncompressed_size, compressed_size); BPE_SET_ETYPE(&dl->dr_overridden_by, etype); BP_SET_TYPE(&dl->dr_overridden_by, type); BP_SET_LEVEL(&dl->dr_overridden_by, 0); BP_SET_BYTEORDER(&dl->dr_overridden_by, byteorder); dl->dr_override_state = DR_OVERRIDDEN; dl->dr_overridden_by.blk_birth = db->db_last_dirty->dr_txg; } /* * Directly assign a provided arc buf to a given dbuf if it's not referenced * by anybody except our caller. Otherwise copy arcbuf's contents to dbuf. */ void dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx) { ASSERT(!refcount_is_zero(&db->db_holds)); ASSERT(db->db_blkid != DMU_BONUS_BLKID); ASSERT(db->db_level == 0); ASSERT(DBUF_GET_BUFC_TYPE(db) == ARC_BUFC_DATA); ASSERT(buf != NULL); ASSERT(arc_buf_size(buf) == db->db.db_size); ASSERT(tx->tx_txg != 0); arc_return_buf(buf, db); ASSERT(arc_released(buf)); mutex_enter(&db->db_mtx); while (db->db_state == DB_READ || db->db_state == DB_FILL) cv_wait(&db->db_changed, &db->db_mtx); ASSERT(db->db_state == DB_CACHED || db->db_state == DB_UNCACHED); if (db->db_state == DB_CACHED && refcount_count(&db->db_holds) - 1 > db->db_dirtycnt) { mutex_exit(&db->db_mtx); (void) dbuf_dirty(db, tx); bcopy(buf->b_data, db->db.db_data, db->db.db_size); VERIFY(arc_buf_remove_ref(buf, db)); xuio_stat_wbuf_copied(); return; } xuio_stat_wbuf_nocopy(); if (db->db_state == DB_CACHED) { dbuf_dirty_record_t *dr = db->db_last_dirty; ASSERT(db->db_buf != NULL); if (dr != NULL && dr->dr_txg == tx->tx_txg) { ASSERT(dr->dt.dl.dr_data == db->db_buf); if (!arc_released(db->db_buf)) { ASSERT(dr->dt.dl.dr_override_state == DR_OVERRIDDEN); arc_release(db->db_buf, db); } dr->dt.dl.dr_data = buf; VERIFY(arc_buf_remove_ref(db->db_buf, db)); } else if (dr == NULL || dr->dt.dl.dr_data != db->db_buf) { arc_release(db->db_buf, db); VERIFY(arc_buf_remove_ref(db->db_buf, db)); } db->db_buf = NULL; } ASSERT(db->db_buf == NULL); dbuf_set_data(db, buf); db->db_state = DB_FILL; mutex_exit(&db->db_mtx); (void) dbuf_dirty(db, tx); dmu_buf_fill_done(&db->db, tx); } /* * "Clear" the contents of this dbuf. This will mark the dbuf * EVICTING and clear *most* of its references. Unfortunately, * when we are not holding the dn_dbufs_mtx, we can't clear the * entry in the dn_dbufs list. We have to wait until dbuf_destroy() * in this case. For callers from the DMU we will usually see: * dbuf_clear()->arc_clear_callback()->dbuf_do_evict()->dbuf_destroy() * For the arc callback, we will usually see: * dbuf_do_evict()->dbuf_clear();dbuf_destroy() * Sometimes, though, we will get a mix of these two: * DMU: dbuf_clear()->arc_clear_callback() * ARC: dbuf_do_evict()->dbuf_destroy() * * This routine will dissociate the dbuf from the arc, by calling * arc_clear_callback(), but will not evict the data from the ARC. */ void dbuf_clear(dmu_buf_impl_t *db) { dnode_t *dn; dmu_buf_impl_t *parent = db->db_parent; dmu_buf_impl_t *dndb; boolean_t dbuf_gone = B_FALSE; ASSERT(MUTEX_HELD(&db->db_mtx)); ASSERT(refcount_is_zero(&db->db_holds)); dbuf_evict_user(db); if (db->db_state == DB_CACHED) { ASSERT(db->db.db_data != NULL); if (db->db_blkid == DMU_BONUS_BLKID) { zio_buf_free(db->db.db_data, DN_MAX_BONUSLEN); arc_space_return(DN_MAX_BONUSLEN, ARC_SPACE_OTHER); } db->db.db_data = NULL; db->db_state = DB_UNCACHED; } ASSERT(db->db_state == DB_UNCACHED || db->db_state == DB_NOFILL); ASSERT(db->db_data_pending == NULL); db->db_state = DB_EVICTING; db->db_blkptr = NULL; DB_DNODE_ENTER(db); dn = DB_DNODE(db); dndb = dn->dn_dbuf; if (db->db_blkid != DMU_BONUS_BLKID && MUTEX_HELD(&dn->dn_dbufs_mtx)) { avl_remove(&dn->dn_dbufs, db); (void) atomic_dec_32_nv(&dn->dn_dbufs_count); membar_producer(); DB_DNODE_EXIT(db); /* * Decrementing the dbuf count means that the hold corresponding * to the removed dbuf is no longer discounted in dnode_move(), * so the dnode cannot be moved until after we release the hold. * The membar_producer() ensures visibility of the decremented * value in dnode_move(), since DB_DNODE_EXIT doesn't actually * release any lock. */ dnode_rele(dn, db); db->db_dnode_handle = NULL; } else { DB_DNODE_EXIT(db); } if (db->db_buf) dbuf_gone = arc_clear_callback(db->db_buf); if (!dbuf_gone) mutex_exit(&db->db_mtx); /* * If this dbuf is referenced from an indirect dbuf, * decrement the ref count on the indirect dbuf. */ if (parent && parent != dndb) dbuf_rele(parent, db); } static int dbuf_findbp(dnode_t *dn, int level, uint64_t blkid, int fail_sparse, dmu_buf_impl_t **parentp, blkptr_t **bpp) { int nlevels, epbs; *parentp = NULL; *bpp = NULL; ASSERT(blkid != DMU_BONUS_BLKID); if (blkid == DMU_SPILL_BLKID) { mutex_enter(&dn->dn_mtx); if (dn->dn_have_spill && (dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR)) *bpp = &dn->dn_phys->dn_spill; else *bpp = NULL; dbuf_add_ref(dn->dn_dbuf, NULL); *parentp = dn->dn_dbuf; mutex_exit(&dn->dn_mtx); return (0); } if (dn->dn_phys->dn_nlevels == 0) nlevels = 1; else nlevels = dn->dn_phys->dn_nlevels; epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT; ASSERT3U(level * epbs, <, 64); ASSERT(RW_LOCK_HELD(&dn->dn_struct_rwlock)); if (level >= nlevels || (blkid > (dn->dn_phys->dn_maxblkid >> (level * epbs)))) { /* the buffer has no parent yet */ return (SET_ERROR(ENOENT)); } else if (level < nlevels-1) { /* this block is referenced from an indirect block */ int err = dbuf_hold_impl(dn, level+1, blkid >> epbs, fail_sparse, NULL, parentp); if (err) return (err); err = dbuf_read(*parentp, NULL, (DB_RF_HAVESTRUCT | DB_RF_NOPREFETCH | DB_RF_CANFAIL)); if (err) { dbuf_rele(*parentp, NULL); *parentp = NULL; return (err); } *bpp = ((blkptr_t *)(*parentp)->db.db_data) + (blkid & ((1ULL << epbs) - 1)); return (0); } else { /* the block is referenced from the dnode */ ASSERT3U(level, ==, nlevels-1); ASSERT(dn->dn_phys->dn_nblkptr == 0 || blkid < dn->dn_phys->dn_nblkptr); if (dn->dn_dbuf) { dbuf_add_ref(dn->dn_dbuf, NULL); *parentp = dn->dn_dbuf; } *bpp = &dn->dn_phys->dn_blkptr[blkid]; return (0); } } static dmu_buf_impl_t * dbuf_create(dnode_t *dn, uint8_t level, uint64_t blkid, dmu_buf_impl_t *parent, blkptr_t *blkptr) { objset_t *os = dn->dn_objset; dmu_buf_impl_t *db, *odb; ASSERT(RW_LOCK_HELD(&dn->dn_struct_rwlock)); ASSERT(dn->dn_type != DMU_OT_NONE); db = kmem_cache_alloc(dbuf_cache, KM_SLEEP); db->db_objset = os; db->db.db_object = dn->dn_object; db->db_level = level; db->db_blkid = blkid; db->db_last_dirty = NULL; db->db_dirtycnt = 0; db->db_dnode_handle = dn->dn_handle; db->db_parent = parent; db->db_blkptr = blkptr; db->db_user_ptr = NULL; db->db_user_data_ptr_ptr = NULL; db->db_evict_func = NULL; db->db_immediate_evict = 0; db->db_freed_in_flight = 0; if (blkid == DMU_BONUS_BLKID) { ASSERT3P(parent, ==, dn->dn_dbuf); db->db.db_size = DN_MAX_BONUSLEN - (dn->dn_nblkptr-1) * sizeof (blkptr_t); ASSERT3U(db->db.db_size, >=, dn->dn_bonuslen); db->db.db_offset = DMU_BONUS_BLKID; db->db_state = DB_UNCACHED; /* the bonus dbuf is not placed in the hash table */ arc_space_consume(sizeof (dmu_buf_impl_t), ARC_SPACE_OTHER); return (db); } else if (blkid == DMU_SPILL_BLKID) { db->db.db_size = (blkptr != NULL) ? BP_GET_LSIZE(blkptr) : SPA_MINBLOCKSIZE; db->db.db_offset = 0; } else { int blocksize = db->db_level ? 1 << dn->dn_indblkshift : dn->dn_datablksz; db->db.db_size = blocksize; db->db.db_offset = db->db_blkid * blocksize; } /* * Hold the dn_dbufs_mtx while we get the new dbuf * in the hash table *and* added to the dbufs list. * This prevents a possible deadlock with someone * trying to look up this dbuf before its added to the * dn_dbufs list. */ mutex_enter(&dn->dn_dbufs_mtx); db->db_state = DB_EVICTING; if ((odb = dbuf_hash_insert(db)) != NULL) { /* someone else inserted it first */ kmem_cache_free(dbuf_cache, db); mutex_exit(&dn->dn_dbufs_mtx); return (odb); } avl_add(&dn->dn_dbufs, db); if (db->db_level == 0 && db->db_blkid >= dn->dn_unlisted_l0_blkid) dn->dn_unlisted_l0_blkid = db->db_blkid + 1; db->db_state = DB_UNCACHED; mutex_exit(&dn->dn_dbufs_mtx); arc_space_consume(sizeof (dmu_buf_impl_t), ARC_SPACE_OTHER); if (parent && parent != dn->dn_dbuf) dbuf_add_ref(parent, db); ASSERT(dn->dn_object == DMU_META_DNODE_OBJECT || refcount_count(&dn->dn_holds) > 0); (void) refcount_add(&dn->dn_holds, db); (void) atomic_inc_32_nv(&dn->dn_dbufs_count); dprintf_dbuf(db, "db=%p\n", db); return (db); } static int dbuf_do_evict(void *private) { dmu_buf_impl_t *db = private; if (!MUTEX_HELD(&db->db_mtx)) mutex_enter(&db->db_mtx); ASSERT(refcount_is_zero(&db->db_holds)); if (db->db_state != DB_EVICTING) { ASSERT(db->db_state == DB_CACHED); DBUF_VERIFY(db); db->db_buf = NULL; dbuf_evict(db); } else { mutex_exit(&db->db_mtx); dbuf_destroy(db); } return (0); } static void dbuf_destroy(dmu_buf_impl_t *db) { ASSERT(refcount_is_zero(&db->db_holds)); if (db->db_blkid != DMU_BONUS_BLKID) { /* * If this dbuf is still on the dn_dbufs list, * remove it from that list. */ if (db->db_dnode_handle != NULL) { dnode_t *dn; DB_DNODE_ENTER(db); dn = DB_DNODE(db); mutex_enter(&dn->dn_dbufs_mtx); avl_remove(&dn->dn_dbufs, db); (void) atomic_dec_32_nv(&dn->dn_dbufs_count); mutex_exit(&dn->dn_dbufs_mtx); DB_DNODE_EXIT(db); /* * Decrementing the dbuf count means that the hold * corresponding to the removed dbuf is no longer * discounted in dnode_move(), so the dnode cannot be * moved until after we release the hold. */ dnode_rele(dn, db); db->db_dnode_handle = NULL; } dbuf_hash_remove(db); } db->db_parent = NULL; db->db_buf = NULL; ASSERT(db->db.db_data == NULL); ASSERT(db->db_hash_next == NULL); ASSERT(db->db_blkptr == NULL); ASSERT(db->db_data_pending == NULL); kmem_cache_free(dbuf_cache, db); arc_space_return(sizeof (dmu_buf_impl_t), ARC_SPACE_OTHER); } void dbuf_prefetch(dnode_t *dn, uint64_t blkid, zio_priority_t prio) { dmu_buf_impl_t *db = NULL; blkptr_t *bp = NULL; ASSERT(blkid != DMU_BONUS_BLKID); ASSERT(RW_LOCK_HELD(&dn->dn_struct_rwlock)); if (dnode_block_freed(dn, blkid)) return; /* dbuf_find() returns with db_mtx held */ if (db = dbuf_find(dn, 0, blkid)) { /* * This dbuf is already in the cache. We assume that * it is already CACHED, or else about to be either * read or filled. */ mutex_exit(&db->db_mtx); return; } if (dbuf_findbp(dn, 0, blkid, TRUE, &db, &bp) == 0) { if (bp && !BP_IS_HOLE(bp) && !BP_IS_EMBEDDED(bp)) { dsl_dataset_t *ds = dn->dn_objset->os_dsl_dataset; uint32_t aflags = ARC_NOWAIT | ARC_PREFETCH; zbookmark_phys_t zb; SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET, dn->dn_object, 0, blkid); (void) arc_read(NULL, dn->dn_objset->os_spa, bp, NULL, NULL, prio, ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, &aflags, &zb); } if (db) dbuf_rele(db, NULL); } } /* * Returns with db_holds incremented, and db_mtx not held. * Note: dn_struct_rwlock must be held. */ int dbuf_hold_impl(dnode_t *dn, uint8_t level, uint64_t blkid, int fail_sparse, void *tag, dmu_buf_impl_t **dbp) { dmu_buf_impl_t *db, *parent = NULL; ASSERT(blkid != DMU_BONUS_BLKID); ASSERT(RW_LOCK_HELD(&dn->dn_struct_rwlock)); ASSERT3U(dn->dn_nlevels, >, level); *dbp = NULL; top: /* dbuf_find() returns with db_mtx held */ db = dbuf_find(dn, level, blkid); if (db == NULL) { blkptr_t *bp = NULL; int err; ASSERT3P(parent, ==, NULL); err = dbuf_findbp(dn, level, blkid, fail_sparse, &parent, &bp); if (fail_sparse) { if (err == 0 && bp && BP_IS_HOLE(bp)) err = SET_ERROR(ENOENT); if (err) { if (parent) dbuf_rele(parent, NULL); return (err); } } if (err && err != ENOENT) return (err); db = dbuf_create(dn, level, blkid, parent, bp); } if (db->db_buf && refcount_is_zero(&db->db_holds)) { arc_buf_add_ref(db->db_buf, db); if (db->db_buf->b_data == NULL) { dbuf_clear(db); if (parent) { dbuf_rele(parent, NULL); parent = NULL; } goto top; } ASSERT3P(db->db.db_data, ==, db->db_buf->b_data); } ASSERT(db->db_buf == NULL || arc_referenced(db->db_buf)); /* * If this buffer is currently syncing out, and we are are * still referencing it from db_data, we need to make a copy * of it in case we decide we want to dirty it again in this txg. */ if (db->db_level == 0 && db->db_blkid != DMU_BONUS_BLKID && dn->dn_object != DMU_META_DNODE_OBJECT && db->db_state == DB_CACHED && db->db_data_pending) { dbuf_dirty_record_t *dr = db->db_data_pending; if (dr->dt.dl.dr_data == db->db_buf) { arc_buf_contents_t type = DBUF_GET_BUFC_TYPE(db); dbuf_set_data(db, arc_buf_alloc(dn->dn_objset->os_spa, db->db.db_size, db, type)); bcopy(dr->dt.dl.dr_data->b_data, db->db.db_data, db->db.db_size); } } (void) refcount_add(&db->db_holds, tag); dbuf_update_data(db); DBUF_VERIFY(db); mutex_exit(&db->db_mtx); /* NOTE: we can't rele the parent until after we drop the db_mtx */ if (parent) dbuf_rele(parent, NULL); ASSERT3P(DB_DNODE(db), ==, dn); ASSERT3U(db->db_blkid, ==, blkid); ASSERT3U(db->db_level, ==, level); *dbp = db; return (0); } dmu_buf_impl_t * dbuf_hold(dnode_t *dn, uint64_t blkid, void *tag) { dmu_buf_impl_t *db; int err = dbuf_hold_impl(dn, 0, blkid, FALSE, tag, &db); return (err ? NULL : db); } dmu_buf_impl_t * dbuf_hold_level(dnode_t *dn, int level, uint64_t blkid, void *tag) { dmu_buf_impl_t *db; int err = dbuf_hold_impl(dn, level, blkid, FALSE, tag, &db); return (err ? NULL : db); } void dbuf_create_bonus(dnode_t *dn) { ASSERT(RW_WRITE_HELD(&dn->dn_struct_rwlock)); ASSERT(dn->dn_bonus == NULL); dn->dn_bonus = dbuf_create(dn, 0, DMU_BONUS_BLKID, dn->dn_dbuf, NULL); } int dbuf_spill_set_blksz(dmu_buf_t *db_fake, uint64_t blksz, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; dnode_t *dn; if (db->db_blkid != DMU_SPILL_BLKID) return (SET_ERROR(ENOTSUP)); if (blksz == 0) blksz = SPA_MINBLOCKSIZE; if (blksz > SPA_MAXBLOCKSIZE) blksz = SPA_MAXBLOCKSIZE; else blksz = P2ROUNDUP(blksz, SPA_MINBLOCKSIZE); DB_DNODE_ENTER(db); dn = DB_DNODE(db); rw_enter(&dn->dn_struct_rwlock, RW_WRITER); dbuf_new_size(db, blksz, tx); rw_exit(&dn->dn_struct_rwlock); DB_DNODE_EXIT(db); return (0); } void dbuf_rm_spill(dnode_t *dn, dmu_tx_t *tx) { dbuf_free_range(dn, DMU_SPILL_BLKID, DMU_SPILL_BLKID, tx); } #pragma weak dmu_buf_add_ref = dbuf_add_ref void dbuf_add_ref(dmu_buf_impl_t *db, void *tag) { int64_t holds = refcount_add(&db->db_holds, tag); ASSERT(holds > 1); } /* * If you call dbuf_rele() you had better not be referencing the dnode handle * unless you have some other direct or indirect hold on the dnode. (An indirect * hold is a hold on one of the dnode's dbufs, including the bonus buffer.) * Without that, the dbuf_rele() could lead to a dnode_rele() followed by the * dnode's parent dbuf evicting its dnode handles. */ void dbuf_rele(dmu_buf_impl_t *db, void *tag) { mutex_enter(&db->db_mtx); dbuf_rele_and_unlock(db, tag); } void dmu_buf_rele(dmu_buf_t *db, void *tag) { dbuf_rele((dmu_buf_impl_t *)db, tag); } /* * dbuf_rele() for an already-locked dbuf. This is necessary to allow * db_dirtycnt and db_holds to be updated atomically. */ void dbuf_rele_and_unlock(dmu_buf_impl_t *db, void *tag) { int64_t holds; ASSERT(MUTEX_HELD(&db->db_mtx)); DBUF_VERIFY(db); /* * Remove the reference to the dbuf before removing its hold on the * dnode so we can guarantee in dnode_move() that a referenced bonus * buffer has a corresponding dnode hold. */ holds = refcount_remove(&db->db_holds, tag); ASSERT(holds >= 0); /* * We can't freeze indirects if there is a possibility that they * may be modified in the current syncing context. */ if (db->db_buf && holds == (db->db_level == 0 ? db->db_dirtycnt : 0)) arc_buf_freeze(db->db_buf); if (holds == db->db_dirtycnt && db->db_level == 0 && db->db_immediate_evict) dbuf_evict_user(db); if (holds == 0) { if (db->db_blkid == DMU_BONUS_BLKID) { mutex_exit(&db->db_mtx); /* * If the dnode moves here, we cannot cross this barrier * until the move completes. */ DB_DNODE_ENTER(db); (void) atomic_dec_32_nv(&DB_DNODE(db)->dn_dbufs_count); DB_DNODE_EXIT(db); /* * The bonus buffer's dnode hold is no longer discounted * in dnode_move(). The dnode cannot move until after * the dnode_rele(). */ dnode_rele(DB_DNODE(db), db); } else if (db->db_buf == NULL) { /* * This is a special case: we never associated this * dbuf with any data allocated from the ARC. */ ASSERT(db->db_state == DB_UNCACHED || db->db_state == DB_NOFILL); dbuf_evict(db); } else if (arc_released(db->db_buf)) { arc_buf_t *buf = db->db_buf; /* * This dbuf has anonymous data associated with it. */ dbuf_set_data(db, NULL); VERIFY(arc_buf_remove_ref(buf, db)); dbuf_evict(db); } else { VERIFY(!arc_buf_remove_ref(db->db_buf, db)); /* * A dbuf will be eligible for eviction if either the * 'primarycache' property is set or a duplicate * copy of this buffer is already cached in the arc. * * In the case of the 'primarycache' a buffer * is considered for eviction if it matches the * criteria set in the property. * * To decide if our buffer is considered a * duplicate, we must call into the arc to determine * if multiple buffers are referencing the same * block on-disk. If so, then we simply evict * ourselves. */ if (!DBUF_IS_CACHEABLE(db)) { if (db->db_blkptr != NULL && !BP_IS_HOLE(db->db_blkptr) && !BP_IS_EMBEDDED(db->db_blkptr)) { spa_t *spa = dmu_objset_spa(db->db_objset); blkptr_t bp = *db->db_blkptr; dbuf_clear(db); arc_freed(spa, &bp); } else { dbuf_clear(db); } } else if (arc_buf_eviction_needed(db->db_buf)) { dbuf_clear(db); } else { mutex_exit(&db->db_mtx); } } } else { mutex_exit(&db->db_mtx); } } #pragma weak dmu_buf_refcount = dbuf_refcount uint64_t dbuf_refcount(dmu_buf_impl_t *db) { return (refcount_count(&db->db_holds)); } void * dmu_buf_set_user(dmu_buf_t *db_fake, void *user_ptr, void *user_data_ptr_ptr, dmu_buf_evict_func_t *evict_func) { return (dmu_buf_update_user(db_fake, NULL, user_ptr, user_data_ptr_ptr, evict_func)); } void * dmu_buf_set_user_ie(dmu_buf_t *db_fake, void *user_ptr, void *user_data_ptr_ptr, dmu_buf_evict_func_t *evict_func) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; db->db_immediate_evict = TRUE; return (dmu_buf_update_user(db_fake, NULL, user_ptr, user_data_ptr_ptr, evict_func)); } void * dmu_buf_update_user(dmu_buf_t *db_fake, void *old_user_ptr, void *user_ptr, void *user_data_ptr_ptr, dmu_buf_evict_func_t *evict_func) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; ASSERT(db->db_level == 0); ASSERT((user_ptr == NULL) == (evict_func == NULL)); mutex_enter(&db->db_mtx); if (db->db_user_ptr == old_user_ptr) { db->db_user_ptr = user_ptr; db->db_user_data_ptr_ptr = user_data_ptr_ptr; db->db_evict_func = evict_func; dbuf_update_data(db); } else { old_user_ptr = db->db_user_ptr; } mutex_exit(&db->db_mtx); return (old_user_ptr); } void * dmu_buf_get_user(dmu_buf_t *db_fake) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; ASSERT(!refcount_is_zero(&db->db_holds)); return (db->db_user_ptr); } boolean_t dmu_buf_freeable(dmu_buf_t *dbuf) { boolean_t res = B_FALSE; dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbuf; if (db->db_blkptr) res = dsl_dataset_block_freeable(db->db_objset->os_dsl_dataset, db->db_blkptr, db->db_blkptr->blk_birth); return (res); } blkptr_t * dmu_buf_get_blkptr(dmu_buf_t *db) { dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; return (dbi->db_blkptr); } static void dbuf_check_blkptr(dnode_t *dn, dmu_buf_impl_t *db) { /* ASSERT(dmu_tx_is_syncing(tx) */ ASSERT(MUTEX_HELD(&db->db_mtx)); if (db->db_blkptr != NULL) return; if (db->db_blkid == DMU_SPILL_BLKID) { db->db_blkptr = &dn->dn_phys->dn_spill; BP_ZERO(db->db_blkptr); return; } if (db->db_level == dn->dn_phys->dn_nlevels-1) { /* * This buffer was allocated at a time when there was * no available blkptrs from the dnode, or it was * inappropriate to hook it in (i.e., nlevels mis-match). */ ASSERT(db->db_blkid < dn->dn_phys->dn_nblkptr); ASSERT(db->db_parent == NULL); db->db_parent = dn->dn_dbuf; db->db_blkptr = &dn->dn_phys->dn_blkptr[db->db_blkid]; DBUF_VERIFY(db); } else { dmu_buf_impl_t *parent = db->db_parent; int epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; ASSERT(dn->dn_phys->dn_nlevels > 1); if (parent == NULL) { mutex_exit(&db->db_mtx); rw_enter(&dn->dn_struct_rwlock, RW_READER); (void) dbuf_hold_impl(dn, db->db_level+1, db->db_blkid >> epbs, FALSE, db, &parent); rw_exit(&dn->dn_struct_rwlock); mutex_enter(&db->db_mtx); db->db_parent = parent; } db->db_blkptr = (blkptr_t *)parent->db.db_data + (db->db_blkid & ((1ULL << epbs) - 1)); DBUF_VERIFY(db); } } static void dbuf_sync_indirect(dbuf_dirty_record_t *dr, dmu_tx_t *tx) { dmu_buf_impl_t *db = dr->dr_dbuf; dnode_t *dn; zio_t *zio; ASSERT(dmu_tx_is_syncing(tx)); dprintf_dbuf_bp(db, db->db_blkptr, "blkptr=%p", db->db_blkptr); mutex_enter(&db->db_mtx); ASSERT(db->db_level > 0); DBUF_VERIFY(db); /* Read the block if it hasn't been read yet. */ if (db->db_buf == NULL) { mutex_exit(&db->db_mtx); (void) dbuf_read(db, NULL, DB_RF_MUST_SUCCEED); mutex_enter(&db->db_mtx); } ASSERT3U(db->db_state, ==, DB_CACHED); ASSERT(db->db_buf != NULL); DB_DNODE_ENTER(db); dn = DB_DNODE(db); /* Indirect block size must match what the dnode thinks it is. */ ASSERT3U(db->db.db_size, ==, 1<dn_phys->dn_indblkshift); dbuf_check_blkptr(dn, db); DB_DNODE_EXIT(db); /* Provide the pending dirty record to child dbufs */ db->db_data_pending = dr; mutex_exit(&db->db_mtx); dbuf_write(dr, db->db_buf, tx); zio = dr->dr_zio; mutex_enter(&dr->dt.di.dr_mtx); dbuf_sync_list(&dr->dt.di.dr_children, tx); ASSERT(list_head(&dr->dt.di.dr_children) == NULL); mutex_exit(&dr->dt.di.dr_mtx); zio_nowait(zio); } static void dbuf_sync_leaf(dbuf_dirty_record_t *dr, dmu_tx_t *tx) { arc_buf_t **datap = &dr->dt.dl.dr_data; dmu_buf_impl_t *db = dr->dr_dbuf; dnode_t *dn; objset_t *os; uint64_t txg = tx->tx_txg; ASSERT(dmu_tx_is_syncing(tx)); dprintf_dbuf_bp(db, db->db_blkptr, "blkptr=%p", db->db_blkptr); mutex_enter(&db->db_mtx); /* * To be synced, we must be dirtied. But we * might have been freed after the dirty. */ if (db->db_state == DB_UNCACHED) { /* This buffer has been freed since it was dirtied */ ASSERT(db->db.db_data == NULL); } else if (db->db_state == DB_FILL) { /* This buffer was freed and is now being re-filled */ ASSERT(db->db.db_data != dr->dt.dl.dr_data); } else { ASSERT(db->db_state == DB_CACHED || db->db_state == DB_NOFILL); } DBUF_VERIFY(db); DB_DNODE_ENTER(db); dn = DB_DNODE(db); if (db->db_blkid == DMU_SPILL_BLKID) { mutex_enter(&dn->dn_mtx); dn->dn_phys->dn_flags |= DNODE_FLAG_SPILL_BLKPTR; mutex_exit(&dn->dn_mtx); } /* * If this is a bonus buffer, simply copy the bonus data into the * dnode. It will be written out when the dnode is synced (and it * will be synced, since it must have been dirty for dbuf_sync to * be called). */ if (db->db_blkid == DMU_BONUS_BLKID) { dbuf_dirty_record_t **drp; ASSERT(*datap != NULL); ASSERT0(db->db_level); ASSERT3U(dn->dn_phys->dn_bonuslen, <=, DN_MAX_BONUSLEN); bcopy(*datap, DN_BONUS(dn->dn_phys), dn->dn_phys->dn_bonuslen); DB_DNODE_EXIT(db); if (*datap != db->db.db_data) { zio_buf_free(*datap, DN_MAX_BONUSLEN); arc_space_return(DN_MAX_BONUSLEN, ARC_SPACE_OTHER); } db->db_data_pending = NULL; drp = &db->db_last_dirty; while (*drp != dr) drp = &(*drp)->dr_next; ASSERT(dr->dr_next == NULL); ASSERT(dr->dr_dbuf == db); *drp = dr->dr_next; if (dr->dr_dbuf->db_level != 0) { list_destroy(&dr->dt.di.dr_children); mutex_destroy(&dr->dt.di.dr_mtx); } kmem_free(dr, sizeof (dbuf_dirty_record_t)); ASSERT(db->db_dirtycnt > 0); db->db_dirtycnt -= 1; dbuf_rele_and_unlock(db, (void *)(uintptr_t)txg); return; } os = dn->dn_objset; /* * This function may have dropped the db_mtx lock allowing a dmu_sync * operation to sneak in. As a result, we need to ensure that we * don't check the dr_override_state until we have returned from * dbuf_check_blkptr. */ dbuf_check_blkptr(dn, db); /* * If this buffer is in the middle of an immediate write, * wait for the synchronous IO to complete. */ while (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC) { ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT); cv_wait(&db->db_changed, &db->db_mtx); ASSERT(dr->dt.dl.dr_override_state != DR_NOT_OVERRIDDEN); } if (db->db_state != DB_NOFILL && dn->dn_object != DMU_META_DNODE_OBJECT && refcount_count(&db->db_holds) > 1 && dr->dt.dl.dr_override_state != DR_OVERRIDDEN && *datap == db->db_buf) { /* * If this buffer is currently "in use" (i.e., there * are active holds and db_data still references it), * then make a copy before we start the write so that * any modifications from the open txg will not leak * into this write. * * NOTE: this copy does not need to be made for * objects only modified in the syncing context (e.g. * DNONE_DNODE blocks). */ int blksz = arc_buf_size(*datap); arc_buf_contents_t type = DBUF_GET_BUFC_TYPE(db); *datap = arc_buf_alloc(os->os_spa, blksz, db, type); bcopy(db->db.db_data, (*datap)->b_data, blksz); } db->db_data_pending = dr; mutex_exit(&db->db_mtx); dbuf_write(dr, *datap, tx); ASSERT(!list_link_active(&dr->dr_dirty_node)); if (dn->dn_object == DMU_META_DNODE_OBJECT) { list_insert_tail(&dn->dn_dirty_records[txg&TXG_MASK], dr); DB_DNODE_EXIT(db); } else { /* * Although zio_nowait() does not "wait for an IO", it does * initiate the IO. If this is an empty write it seems plausible * that the IO could actually be completed before the nowait * returns. We need to DB_DNODE_EXIT() first in case * zio_nowait() invalidates the dbuf. */ DB_DNODE_EXIT(db); zio_nowait(dr->dr_zio); } } void dbuf_sync_list(list_t *list, dmu_tx_t *tx) { dbuf_dirty_record_t *dr; while (dr = list_head(list)) { if (dr->dr_zio != NULL) { /* * If we find an already initialized zio then we * are processing the meta-dnode, and we have finished. * The dbufs for all dnodes are put back on the list * during processing, so that we can zio_wait() * these IOs after initiating all child IOs. */ ASSERT3U(dr->dr_dbuf->db.db_object, ==, DMU_META_DNODE_OBJECT); break; } list_remove(list, dr); if (dr->dr_dbuf->db_level > 0) dbuf_sync_indirect(dr, tx); else dbuf_sync_leaf(dr, tx); } } /* ARGSUSED */ static void dbuf_write_ready(zio_t *zio, arc_buf_t *buf, void *vdb) { dmu_buf_impl_t *db = vdb; dnode_t *dn; blkptr_t *bp = zio->io_bp; blkptr_t *bp_orig = &zio->io_bp_orig; spa_t *spa = zio->io_spa; int64_t delta; uint64_t fill = 0; int i; ASSERT3P(db->db_blkptr, ==, bp); DB_DNODE_ENTER(db); dn = DB_DNODE(db); delta = bp_get_dsize_sync(spa, bp) - bp_get_dsize_sync(spa, bp_orig); dnode_diduse_space(dn, delta - zio->io_prev_space_delta); zio->io_prev_space_delta = delta; if (bp->blk_birth != 0) { ASSERT((db->db_blkid != DMU_SPILL_BLKID && BP_GET_TYPE(bp) == dn->dn_type) || (db->db_blkid == DMU_SPILL_BLKID && BP_GET_TYPE(bp) == dn->dn_bonustype) || BP_IS_EMBEDDED(bp)); ASSERT(BP_GET_LEVEL(bp) == db->db_level); } mutex_enter(&db->db_mtx); #ifdef ZFS_DEBUG if (db->db_blkid == DMU_SPILL_BLKID) { ASSERT(dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR); ASSERT(!(BP_IS_HOLE(db->db_blkptr)) && db->db_blkptr == &dn->dn_phys->dn_spill); } #endif if (db->db_level == 0) { mutex_enter(&dn->dn_mtx); if (db->db_blkid > dn->dn_phys->dn_maxblkid && db->db_blkid != DMU_SPILL_BLKID) dn->dn_phys->dn_maxblkid = db->db_blkid; mutex_exit(&dn->dn_mtx); if (dn->dn_type == DMU_OT_DNODE) { dnode_phys_t *dnp = db->db.db_data; for (i = db->db.db_size >> DNODE_SHIFT; i > 0; i--, dnp++) { if (dnp->dn_type != DMU_OT_NONE) fill++; } } else { if (BP_IS_HOLE(bp)) { fill = 0; } else { fill = 1; } } } else { blkptr_t *ibp = db->db.db_data; ASSERT3U(db->db.db_size, ==, 1<dn_phys->dn_indblkshift); for (i = db->db.db_size >> SPA_BLKPTRSHIFT; i > 0; i--, ibp++) { if (BP_IS_HOLE(ibp)) continue; fill += BP_GET_FILL(ibp); } } DB_DNODE_EXIT(db); if (!BP_IS_EMBEDDED(bp)) bp->blk_fill = fill; mutex_exit(&db->db_mtx); } /* * The SPA will call this callback several times for each zio - once * for every physical child i/o (zio->io_phys_children times). This * allows the DMU to monitor the progress of each logical i/o. For example, * there may be 2 copies of an indirect block, or many fragments of a RAID-Z * block. There may be a long delay before all copies/fragments are completed, * so this callback allows us to retire dirty space gradually, as the physical * i/os complete. */ /* ARGSUSED */ static void dbuf_write_physdone(zio_t *zio, arc_buf_t *buf, void *arg) { dmu_buf_impl_t *db = arg; objset_t *os = db->db_objset; dsl_pool_t *dp = dmu_objset_pool(os); dbuf_dirty_record_t *dr; int delta = 0; dr = db->db_data_pending; ASSERT3U(dr->dr_txg, ==, zio->io_txg); /* * The callback will be called io_phys_children times. Retire one * portion of our dirty space each time we are called. Any rounding * error will be cleaned up by dsl_pool_sync()'s call to * dsl_pool_undirty_space(). */ delta = dr->dr_accounted / zio->io_phys_children; dsl_pool_undirty_space(dp, delta, zio->io_txg); } /* ARGSUSED */ static void dbuf_write_done(zio_t *zio, arc_buf_t *buf, void *vdb) { dmu_buf_impl_t *db = vdb; blkptr_t *bp_orig = &zio->io_bp_orig; blkptr_t *bp = db->db_blkptr; objset_t *os = db->db_objset; dmu_tx_t *tx = os->os_synctx; dbuf_dirty_record_t **drp, *dr; ASSERT0(zio->io_error); ASSERT(db->db_blkptr == bp); /* * For nopwrites and rewrites we ensure that the bp matches our * original and bypass all the accounting. */ if (zio->io_flags & (ZIO_FLAG_IO_REWRITE | ZIO_FLAG_NOPWRITE)) { ASSERT(BP_EQUAL(bp, bp_orig)); } else { dsl_dataset_t *ds = os->os_dsl_dataset; (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE); dsl_dataset_block_born(ds, bp, tx); } mutex_enter(&db->db_mtx); DBUF_VERIFY(db); drp = &db->db_last_dirty; while ((dr = *drp) != db->db_data_pending) drp = &dr->dr_next; ASSERT(!list_link_active(&dr->dr_dirty_node)); ASSERT(dr->dr_dbuf == db); ASSERT(dr->dr_next == NULL); *drp = dr->dr_next; #ifdef ZFS_DEBUG if (db->db_blkid == DMU_SPILL_BLKID) { dnode_t *dn; DB_DNODE_ENTER(db); dn = DB_DNODE(db); ASSERT(dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR); ASSERT(!(BP_IS_HOLE(db->db_blkptr)) && db->db_blkptr == &dn->dn_phys->dn_spill); DB_DNODE_EXIT(db); } #endif if (db->db_level == 0) { ASSERT(db->db_blkid != DMU_BONUS_BLKID); ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN); if (db->db_state != DB_NOFILL) { if (dr->dt.dl.dr_data != db->db_buf) VERIFY(arc_buf_remove_ref(dr->dt.dl.dr_data, db)); else if (!arc_released(db->db_buf)) arc_set_callback(db->db_buf, dbuf_do_evict, db); } } else { dnode_t *dn; DB_DNODE_ENTER(db); dn = DB_DNODE(db); ASSERT(list_head(&dr->dt.di.dr_children) == NULL); ASSERT3U(db->db.db_size, ==, 1 << dn->dn_phys->dn_indblkshift); if (!BP_IS_HOLE(db->db_blkptr)) { int epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT; ASSERT3U(db->db_blkid, <=, dn->dn_phys->dn_maxblkid >> (db->db_level * epbs)); ASSERT3U(BP_GET_LSIZE(db->db_blkptr), ==, db->db.db_size); if (!arc_released(db->db_buf)) arc_set_callback(db->db_buf, dbuf_do_evict, db); } DB_DNODE_EXIT(db); mutex_destroy(&dr->dt.di.dr_mtx); list_destroy(&dr->dt.di.dr_children); } kmem_free(dr, sizeof (dbuf_dirty_record_t)); cv_broadcast(&db->db_changed); ASSERT(db->db_dirtycnt > 0); db->db_dirtycnt -= 1; db->db_data_pending = NULL; dbuf_rele_and_unlock(db, (void *)(uintptr_t)tx->tx_txg); } static void dbuf_write_nofill_ready(zio_t *zio) { dbuf_write_ready(zio, NULL, zio->io_private); } static void dbuf_write_nofill_done(zio_t *zio) { dbuf_write_done(zio, NULL, zio->io_private); } static void dbuf_write_override_ready(zio_t *zio) { dbuf_dirty_record_t *dr = zio->io_private; dmu_buf_impl_t *db = dr->dr_dbuf; dbuf_write_ready(zio, NULL, db); } static void dbuf_write_override_done(zio_t *zio) { dbuf_dirty_record_t *dr = zio->io_private; dmu_buf_impl_t *db = dr->dr_dbuf; blkptr_t *obp = &dr->dt.dl.dr_overridden_by; mutex_enter(&db->db_mtx); if (!BP_EQUAL(zio->io_bp, obp)) { if (!BP_IS_HOLE(obp)) dsl_free(spa_get_dsl(zio->io_spa), zio->io_txg, obp); arc_release(dr->dt.dl.dr_data, db); } mutex_exit(&db->db_mtx); dbuf_write_done(zio, NULL, db); } /* Issue I/O to commit a dirty buffer to disk. */ static void dbuf_write(dbuf_dirty_record_t *dr, arc_buf_t *data, dmu_tx_t *tx) { dmu_buf_impl_t *db = dr->dr_dbuf; dnode_t *dn; objset_t *os; dmu_buf_impl_t *parent = db->db_parent; uint64_t txg = tx->tx_txg; zbookmark_phys_t zb; zio_prop_t zp; zio_t *zio; int wp_flag = 0; DB_DNODE_ENTER(db); dn = DB_DNODE(db); os = dn->dn_objset; if (db->db_state != DB_NOFILL) { if (db->db_level > 0 || dn->dn_type == DMU_OT_DNODE) { /* * Private object buffers are released here rather * than in dbuf_dirty() since they are only modified * in the syncing context and we don't want the * overhead of making multiple copies of the data. */ if (BP_IS_HOLE(db->db_blkptr)) { arc_buf_thaw(data); } else { dbuf_release_bp(db); } } } if (parent != dn->dn_dbuf) { /* Our parent is an indirect block. */ /* We have a dirty parent that has been scheduled for write. */ ASSERT(parent && parent->db_data_pending); /* Our parent's buffer is one level closer to the dnode. */ ASSERT(db->db_level == parent->db_level-1); /* * We're about to modify our parent's db_data by modifying * our block pointer, so the parent must be released. */ ASSERT(arc_released(parent->db_buf)); zio = parent->db_data_pending->dr_zio; } else { /* Our parent is the dnode itself. */ ASSERT((db->db_level == dn->dn_phys->dn_nlevels-1 && db->db_blkid != DMU_SPILL_BLKID) || (db->db_blkid == DMU_SPILL_BLKID && db->db_level == 0)); if (db->db_blkid != DMU_SPILL_BLKID) ASSERT3P(db->db_blkptr, ==, &dn->dn_phys->dn_blkptr[db->db_blkid]); zio = dn->dn_zio; } ASSERT(db->db_level == 0 || data == db->db_buf); ASSERT3U(db->db_blkptr->blk_birth, <=, txg); ASSERT(zio); SET_BOOKMARK(&zb, os->os_dsl_dataset ? os->os_dsl_dataset->ds_object : DMU_META_OBJSET, db->db.db_object, db->db_level, db->db_blkid); if (db->db_blkid == DMU_SPILL_BLKID) wp_flag = WP_SPILL; wp_flag |= (db->db_state == DB_NOFILL) ? WP_NOFILL : 0; dmu_write_policy(os, dn, db->db_level, wp_flag, &zp); DB_DNODE_EXIT(db); if (db->db_level == 0 && dr->dt.dl.dr_override_state == DR_OVERRIDDEN) { /* * The BP for this block has been provided by open context * (by dmu_sync() or dmu_buf_write_embedded()). */ void *contents = (data != NULL) ? data->b_data : NULL; dr->dr_zio = zio_write(zio, os->os_spa, txg, db->db_blkptr, contents, db->db.db_size, &zp, dbuf_write_override_ready, NULL, dbuf_write_override_done, dr, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb); mutex_enter(&db->db_mtx); dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN; zio_write_override(dr->dr_zio, &dr->dt.dl.dr_overridden_by, dr->dt.dl.dr_copies, dr->dt.dl.dr_nopwrite); mutex_exit(&db->db_mtx); } else if (db->db_state == DB_NOFILL) { ASSERT(zp.zp_checksum == ZIO_CHECKSUM_OFF || zp.zp_checksum == ZIO_CHECKSUM_NOPARITY); dr->dr_zio = zio_write(zio, os->os_spa, txg, db->db_blkptr, NULL, db->db.db_size, &zp, dbuf_write_nofill_ready, NULL, dbuf_write_nofill_done, db, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED | ZIO_FLAG_NODATA, &zb); } else { ASSERT(arc_released(data)); dr->dr_zio = arc_write(zio, os->os_spa, txg, db->db_blkptr, data, DBUF_IS_L2CACHEABLE(db), DBUF_IS_L2COMPRESSIBLE(db), &zp, dbuf_write_ready, dbuf_write_physdone, dbuf_write_done, db, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb); } } Index: stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/spa.c =================================================================== --- stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/spa.c (revision 271000) +++ stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/spa.c (revision 271001) @@ -1,6944 +1,6944 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 2013 by Delphix. All rights reserved. * Copyright (c) 2013, 2014, Nexenta Systems, Inc. All rights reserved. * Copyright (c) 2013 Martin Matuska . All rights reserved. */ /* * SPA: Storage Pool Allocator * * This file contains all the routines used when modifying on-disk SPA state. * This includes opening, importing, destroying, exporting a pool, and syncing a * pool. */ #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 #ifdef _KERNEL #include #include #include #endif /* _KERNEL */ #include "zfs_prop.h" #include "zfs_comutil.h" /* Check hostid on import? */ static int check_hostid = 1; SYSCTL_DECL(_vfs_zfs); TUNABLE_INT("vfs.zfs.check_hostid", &check_hostid); SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RW, &check_hostid, 0, "Check hostid on import?"); /* * The interval, in seconds, at which failed configuration cache file writes * should be retried. */ static int zfs_ccw_retry_interval = 300; typedef enum zti_modes { ZTI_MODE_FIXED, /* value is # of threads (min 1) */ ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */ ZTI_MODE_NULL, /* don't create a taskq */ ZTI_NMODES } zti_modes_t; #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) } #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 } #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 } #define ZTI_N(n) ZTI_P(n, 1) #define ZTI_ONE ZTI_N(1) typedef struct zio_taskq_info { zti_modes_t zti_mode; uint_t zti_value; uint_t zti_count; } zio_taskq_info_t; static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = { "issue", "issue_high", "intr", "intr_high" }; /* * This table defines the taskq settings for each ZFS I/O type. When * initializing a pool, we use this table to create an appropriately sized * taskq. Some operations are low volume and therefore have a small, static * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE * macros. Other operations process a large amount of data; the ZTI_BATCH * macro causes us to create a taskq oriented for throughput. Some operations * are so high frequency and short-lived that the taskq itself can become a a * point of lock contention. The ZTI_P(#, #) macro indicates that we need an * additional degree of parallelism specified by the number of threads per- * taskq and the number of taskqs; when dispatching an event in this case, the * particular taskq is chosen at random. * * The different taskq priorities are to handle the different contexts (issue * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that * need to be handled with minimum delay. */ const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = { /* ISSUE ISSUE_HIGH INTR INTR_HIGH */ { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */ { ZTI_N(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL }, /* READ */ { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */ { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */ { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */ { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */ }; static void spa_sync_version(void *arg, dmu_tx_t *tx); static void spa_sync_props(void *arg, dmu_tx_t *tx); static boolean_t spa_has_active_shared_spare(spa_t *spa); static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config, spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, char **ereport); static void spa_vdev_resilver_done(spa_t *spa); uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */ #ifdef PSRSET_BIND id_t zio_taskq_psrset_bind = PS_NONE; #endif #ifdef SYSDC boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */ #endif uint_t zio_taskq_basedc = 80; /* base duty cycle */ boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */ extern int zfs_sync_pass_deferred_free; #ifndef illumos extern void spa_deadman(void *arg); #endif /* * This (illegal) pool name is used when temporarily importing a spa_t in order * to get the vdev stats associated with the imported devices. */ #define TRYIMPORT_NAME "$import" /* * ========================================================================== * SPA properties routines * ========================================================================== */ /* * Add a (source=src, propname=propval) list to an nvlist. */ static void spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval, uint64_t intval, zprop_source_t src) { const char *propname = zpool_prop_to_name(prop); nvlist_t *propval; VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0); VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0); if (strval != NULL) VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0); else VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0); VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0); nvlist_free(propval); } /* * Get property values from the spa configuration. */ static void spa_prop_get_config(spa_t *spa, nvlist_t **nvp) { vdev_t *rvd = spa->spa_root_vdev; dsl_pool_t *pool = spa->spa_dsl_pool; uint64_t size, alloc, cap, version; zprop_source_t src = ZPROP_SRC_NONE; spa_config_dirent_t *dp; metaslab_class_t *mc = spa_normal_class(spa); ASSERT(MUTEX_HELD(&spa->spa_props_lock)); if (rvd != NULL) { alloc = metaslab_class_get_alloc(spa_normal_class(spa)); size = metaslab_class_get_space(spa_normal_class(spa)); spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src); spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src); spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src); spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL, size - alloc, src); spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL, metaslab_class_fragmentation(mc), src); spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, metaslab_class_expandable_space(mc), src); spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL, (spa_mode(spa) == FREAD), src); cap = (size == 0) ? 0 : (alloc * 100 / size); spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src); spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL, ddt_get_pool_dedup_ratio(spa), src); spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL, rvd->vdev_state, src); version = spa_version(spa); if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) src = ZPROP_SRC_DEFAULT; else src = ZPROP_SRC_LOCAL; spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src); } if (pool != NULL) { /* * The $FREE directory was introduced in SPA_VERSION_DEADLISTS, * when opening pools before this version freedir will be NULL. */ if (pool->dp_free_dir != NULL) { spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL, pool->dp_free_dir->dd_phys->dd_used_bytes, src); } else { spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL, 0, src); } if (pool->dp_leak_dir != NULL) { spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL, pool->dp_leak_dir->dd_phys->dd_used_bytes, src); } else { spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL, 0, src); } } spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src); if (spa->spa_comment != NULL) { spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment, 0, ZPROP_SRC_LOCAL); } if (spa->spa_root != NULL) spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root, 0, ZPROP_SRC_LOCAL); if ((dp = list_head(&spa->spa_config_list)) != NULL) { if (dp->scd_path == NULL) { spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, "none", 0, ZPROP_SRC_LOCAL); } else if (strcmp(dp->scd_path, spa_config_path) != 0) { spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, dp->scd_path, 0, ZPROP_SRC_LOCAL); } } } /* * Get zpool property values. */ int spa_prop_get(spa_t *spa, nvlist_t **nvp) { objset_t *mos = spa->spa_meta_objset; zap_cursor_t zc; zap_attribute_t za; int err; VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0); mutex_enter(&spa->spa_props_lock); /* * Get properties from the spa config. */ spa_prop_get_config(spa, nvp); /* If no pool property object, no more prop to get. */ if (mos == NULL || spa->spa_pool_props_object == 0) { mutex_exit(&spa->spa_props_lock); return (0); } /* * Get properties from the MOS pool property object. */ for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object); (err = zap_cursor_retrieve(&zc, &za)) == 0; zap_cursor_advance(&zc)) { uint64_t intval = 0; char *strval = NULL; zprop_source_t src = ZPROP_SRC_DEFAULT; zpool_prop_t prop; if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL) continue; switch (za.za_integer_length) { case 8: /* integer property */ if (za.za_first_integer != zpool_prop_default_numeric(prop)) src = ZPROP_SRC_LOCAL; if (prop == ZPOOL_PROP_BOOTFS) { dsl_pool_t *dp; dsl_dataset_t *ds = NULL; dp = spa_get_dsl(spa); dsl_pool_config_enter(dp, FTAG); if (err = dsl_dataset_hold_obj(dp, za.za_first_integer, FTAG, &ds)) { dsl_pool_config_exit(dp, FTAG); break; } strval = kmem_alloc( MAXNAMELEN + strlen(MOS_DIR_NAME) + 1, KM_SLEEP); dsl_dataset_name(ds, strval); dsl_dataset_rele(ds, FTAG); dsl_pool_config_exit(dp, FTAG); } else { strval = NULL; intval = za.za_first_integer; } spa_prop_add_list(*nvp, prop, strval, intval, src); if (strval != NULL) kmem_free(strval, MAXNAMELEN + strlen(MOS_DIR_NAME) + 1); break; case 1: /* string property */ strval = kmem_alloc(za.za_num_integers, KM_SLEEP); err = zap_lookup(mos, spa->spa_pool_props_object, za.za_name, 1, za.za_num_integers, strval); if (err) { kmem_free(strval, za.za_num_integers); break; } spa_prop_add_list(*nvp, prop, strval, 0, src); kmem_free(strval, za.za_num_integers); break; default: break; } } zap_cursor_fini(&zc); mutex_exit(&spa->spa_props_lock); out: if (err && err != ENOENT) { nvlist_free(*nvp); *nvp = NULL; return (err); } return (0); } /* * Validate the given pool properties nvlist and modify the list * for the property values to be set. */ static int spa_prop_validate(spa_t *spa, nvlist_t *props) { nvpair_t *elem; int error = 0, reset_bootfs = 0; uint64_t objnum = 0; boolean_t has_feature = B_FALSE; elem = NULL; while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { uint64_t intval; char *strval, *slash, *check, *fname; const char *propname = nvpair_name(elem); zpool_prop_t prop = zpool_name_to_prop(propname); switch (prop) { case ZPROP_INVAL: if (!zpool_prop_feature(propname)) { error = SET_ERROR(EINVAL); break; } /* * Sanitize the input. */ if (nvpair_type(elem) != DATA_TYPE_UINT64) { error = SET_ERROR(EINVAL); break; } if (nvpair_value_uint64(elem, &intval) != 0) { error = SET_ERROR(EINVAL); break; } if (intval != 0) { error = SET_ERROR(EINVAL); break; } fname = strchr(propname, '@') + 1; if (zfeature_lookup_name(fname, NULL) != 0) { error = SET_ERROR(EINVAL); break; } has_feature = B_TRUE; break; case ZPOOL_PROP_VERSION: error = nvpair_value_uint64(elem, &intval); if (!error && (intval < spa_version(spa) || intval > SPA_VERSION_BEFORE_FEATURES || has_feature)) error = SET_ERROR(EINVAL); break; case ZPOOL_PROP_DELEGATION: case ZPOOL_PROP_AUTOREPLACE: case ZPOOL_PROP_LISTSNAPS: case ZPOOL_PROP_AUTOEXPAND: error = nvpair_value_uint64(elem, &intval); if (!error && intval > 1) error = SET_ERROR(EINVAL); break; case ZPOOL_PROP_BOOTFS: /* * If the pool version is less than SPA_VERSION_BOOTFS, * or the pool is still being created (version == 0), * the bootfs property cannot be set. */ if (spa_version(spa) < SPA_VERSION_BOOTFS) { error = SET_ERROR(ENOTSUP); break; } /* * Make sure the vdev config is bootable */ if (!vdev_is_bootable(spa->spa_root_vdev)) { error = SET_ERROR(ENOTSUP); break; } reset_bootfs = 1; error = nvpair_value_string(elem, &strval); if (!error) { objset_t *os; uint64_t compress; if (strval == NULL || strval[0] == '\0') { objnum = zpool_prop_default_numeric( ZPOOL_PROP_BOOTFS); break; } if (error = dmu_objset_hold(strval, FTAG, &os)) break; /* Must be ZPL and not gzip compressed. */ if (dmu_objset_type(os) != DMU_OST_ZFS) { error = SET_ERROR(ENOTSUP); } else if ((error = dsl_prop_get_int_ds(dmu_objset_ds(os), zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress)) == 0 && !BOOTFS_COMPRESS_VALID(compress)) { error = SET_ERROR(ENOTSUP); } else { objnum = dmu_objset_id(os); } dmu_objset_rele(os, FTAG); } break; case ZPOOL_PROP_FAILUREMODE: error = nvpair_value_uint64(elem, &intval); if (!error && (intval < ZIO_FAILURE_MODE_WAIT || intval > ZIO_FAILURE_MODE_PANIC)) error = SET_ERROR(EINVAL); /* * This is a special case which only occurs when * the pool has completely failed. This allows * the user to change the in-core failmode property * without syncing it out to disk (I/Os might * currently be blocked). We do this by returning * EIO to the caller (spa_prop_set) to trick it * into thinking we encountered a property validation * error. */ if (!error && spa_suspended(spa)) { spa->spa_failmode = intval; error = SET_ERROR(EIO); } break; case ZPOOL_PROP_CACHEFILE: if ((error = nvpair_value_string(elem, &strval)) != 0) break; if (strval[0] == '\0') break; if (strcmp(strval, "none") == 0) break; if (strval[0] != '/') { error = SET_ERROR(EINVAL); break; } slash = strrchr(strval, '/'); ASSERT(slash != NULL); if (slash[1] == '\0' || strcmp(slash, "/.") == 0 || strcmp(slash, "/..") == 0) error = SET_ERROR(EINVAL); break; case ZPOOL_PROP_COMMENT: if ((error = nvpair_value_string(elem, &strval)) != 0) break; for (check = strval; *check != '\0'; check++) { /* * The kernel doesn't have an easy isprint() * check. For this kernel check, we merely * check ASCII apart from DEL. Fix this if * there is an easy-to-use kernel isprint(). */ if (*check >= 0x7f) { error = SET_ERROR(EINVAL); break; } check++; } if (strlen(strval) > ZPROP_MAX_COMMENT) error = E2BIG; break; case ZPOOL_PROP_DEDUPDITTO: if (spa_version(spa) < SPA_VERSION_DEDUP) error = SET_ERROR(ENOTSUP); else error = nvpair_value_uint64(elem, &intval); if (error == 0 && intval != 0 && intval < ZIO_DEDUPDITTO_MIN) error = SET_ERROR(EINVAL); break; } if (error) break; } if (!error && reset_bootfs) { error = nvlist_remove(props, zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING); if (!error) { error = nvlist_add_uint64(props, zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum); } } return (error); } void spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync) { char *cachefile; spa_config_dirent_t *dp; if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), &cachefile) != 0) return; dp = kmem_alloc(sizeof (spa_config_dirent_t), KM_SLEEP); if (cachefile[0] == '\0') dp->scd_path = spa_strdup(spa_config_path); else if (strcmp(cachefile, "none") == 0) dp->scd_path = NULL; else dp->scd_path = spa_strdup(cachefile); list_insert_head(&spa->spa_config_list, dp); if (need_sync) spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); } int spa_prop_set(spa_t *spa, nvlist_t *nvp) { int error; nvpair_t *elem = NULL; boolean_t need_sync = B_FALSE; if ((error = spa_prop_validate(spa, nvp)) != 0) return (error); while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) { zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem)); if (prop == ZPOOL_PROP_CACHEFILE || prop == ZPOOL_PROP_ALTROOT || prop == ZPOOL_PROP_READONLY) continue; if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) { uint64_t ver; if (prop == ZPOOL_PROP_VERSION) { VERIFY(nvpair_value_uint64(elem, &ver) == 0); } else { ASSERT(zpool_prop_feature(nvpair_name(elem))); ver = SPA_VERSION_FEATURES; need_sync = B_TRUE; } /* Save time if the version is already set. */ if (ver == spa_version(spa)) continue; /* * In addition to the pool directory object, we might * create the pool properties object, the features for * read object, the features for write object, or the * feature descriptions object. */ error = dsl_sync_task(spa->spa_name, NULL, spa_sync_version, &ver, 6, ZFS_SPACE_CHECK_RESERVED); if (error) return (error); continue; } need_sync = B_TRUE; break; } if (need_sync) { return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props, nvp, 6, ZFS_SPACE_CHECK_RESERVED)); } return (0); } /* * If the bootfs property value is dsobj, clear it. */ void spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) { if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) { VERIFY(zap_remove(spa->spa_meta_objset, spa->spa_pool_props_object, zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0); spa->spa_bootfs = 0; } } /*ARGSUSED*/ static int spa_change_guid_check(void *arg, dmu_tx_t *tx) { uint64_t *newguid = arg; spa_t *spa = dmu_tx_pool(tx)->dp_spa; vdev_t *rvd = spa->spa_root_vdev; uint64_t vdev_state; spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); vdev_state = rvd->vdev_state; spa_config_exit(spa, SCL_STATE, FTAG); if (vdev_state != VDEV_STATE_HEALTHY) return (SET_ERROR(ENXIO)); ASSERT3U(spa_guid(spa), !=, *newguid); return (0); } static void spa_change_guid_sync(void *arg, dmu_tx_t *tx) { uint64_t *newguid = arg; spa_t *spa = dmu_tx_pool(tx)->dp_spa; uint64_t oldguid; vdev_t *rvd = spa->spa_root_vdev; oldguid = spa_guid(spa); spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); rvd->vdev_guid = *newguid; rvd->vdev_guid_sum += (*newguid - oldguid); vdev_config_dirty(rvd); spa_config_exit(spa, SCL_STATE, FTAG); spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu", oldguid, *newguid); } /* * Change the GUID for the pool. This is done so that we can later * re-import a pool built from a clone of our own vdevs. We will modify * the root vdev's guid, our own pool guid, and then mark all of our * vdevs dirty. Note that we must make sure that all our vdevs are * online when we do this, or else any vdevs that weren't present * would be orphaned from our pool. We are also going to issue a * sysevent to update any watchers. */ int spa_change_guid(spa_t *spa) { int error; uint64_t guid; mutex_enter(&spa->spa_vdev_top_lock); mutex_enter(&spa_namespace_lock); guid = spa_generate_guid(NULL); error = dsl_sync_task(spa->spa_name, spa_change_guid_check, spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED); if (error == 0) { spa_config_sync(spa, B_FALSE, B_TRUE); spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID); } mutex_exit(&spa_namespace_lock); mutex_exit(&spa->spa_vdev_top_lock); return (error); } /* * ========================================================================== * SPA state manipulation (open/create/destroy/import/export) * ========================================================================== */ static int spa_error_entry_compare(const void *a, const void *b) { spa_error_entry_t *sa = (spa_error_entry_t *)a; spa_error_entry_t *sb = (spa_error_entry_t *)b; int ret; ret = bcmp(&sa->se_bookmark, &sb->se_bookmark, sizeof (zbookmark_phys_t)); if (ret < 0) return (-1); else if (ret > 0) return (1); else return (0); } /* * Utility function which retrieves copies of the current logs and * re-initializes them in the process. */ void spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) { ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t)); bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t)); avl_create(&spa->spa_errlist_scrub, spa_error_entry_compare, sizeof (spa_error_entry_t), offsetof(spa_error_entry_t, se_avl)); avl_create(&spa->spa_errlist_last, spa_error_entry_compare, sizeof (spa_error_entry_t), offsetof(spa_error_entry_t, se_avl)); } static void spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q) { const zio_taskq_info_t *ztip = &zio_taskqs[t][q]; enum zti_modes mode = ztip->zti_mode; uint_t value = ztip->zti_value; uint_t count = ztip->zti_count; spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; char name[32]; uint_t flags = 0; boolean_t batch = B_FALSE; if (mode == ZTI_MODE_NULL) { tqs->stqs_count = 0; tqs->stqs_taskq = NULL; return; } ASSERT3U(count, >, 0); tqs->stqs_count = count; tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP); switch (mode) { case ZTI_MODE_FIXED: ASSERT3U(value, >=, 1); value = MAX(value, 1); break; case ZTI_MODE_BATCH: batch = B_TRUE; flags |= TASKQ_THREADS_CPU_PCT; value = zio_taskq_batch_pct; break; default: panic("unrecognized mode for %s_%s taskq (%u:%u) in " "spa_activate()", zio_type_name[t], zio_taskq_types[q], mode, value); break; } for (uint_t i = 0; i < count; i++) { taskq_t *tq; if (count > 1) { (void) snprintf(name, sizeof (name), "%s_%s_%u", zio_type_name[t], zio_taskq_types[q], i); } else { (void) snprintf(name, sizeof (name), "%s_%s", zio_type_name[t], zio_taskq_types[q]); } #ifdef SYSDC if (zio_taskq_sysdc && spa->spa_proc != &p0) { if (batch) flags |= TASKQ_DC_BATCH; tq = taskq_create_sysdc(name, value, 50, INT_MAX, spa->spa_proc, zio_taskq_basedc, flags); } else { #endif pri_t pri = maxclsyspri; /* * The write issue taskq can be extremely CPU * intensive. Run it at slightly lower priority * than the other taskqs. */ if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE) pri--; tq = taskq_create_proc(name, value, pri, 50, INT_MAX, spa->spa_proc, flags); #ifdef SYSDC } #endif tqs->stqs_taskq[i] = tq; } } static void spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q) { spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; if (tqs->stqs_taskq == NULL) { ASSERT0(tqs->stqs_count); return; } for (uint_t i = 0; i < tqs->stqs_count; i++) { ASSERT3P(tqs->stqs_taskq[i], !=, NULL); taskq_destroy(tqs->stqs_taskq[i]); } kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *)); tqs->stqs_taskq = NULL; } /* * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority. * Note that a type may have multiple discrete taskqs to avoid lock contention * on the taskq itself. In that case we choose which taskq at random by using * the low bits of gethrtime(). */ void spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q, task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent) { spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; taskq_t *tq; ASSERT3P(tqs->stqs_taskq, !=, NULL); ASSERT3U(tqs->stqs_count, !=, 0); if (tqs->stqs_count == 1) { tq = tqs->stqs_taskq[0]; } else { #ifdef _KERNEL tq = tqs->stqs_taskq[cpu_ticks() % tqs->stqs_count]; #else tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count]; #endif } taskq_dispatch_ent(tq, func, arg, flags, ent); } static void spa_create_zio_taskqs(spa_t *spa) { for (int t = 0; t < ZIO_TYPES; t++) { for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { spa_taskqs_init(spa, t, q); } } } #ifdef _KERNEL #ifdef SPA_PROCESS static void spa_thread(void *arg) { callb_cpr_t cprinfo; spa_t *spa = arg; user_t *pu = PTOU(curproc); CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr, spa->spa_name); ASSERT(curproc != &p0); (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs), "zpool-%s", spa->spa_name); (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm)); #ifdef PSRSET_BIND /* bind this thread to the requested psrset */ if (zio_taskq_psrset_bind != PS_NONE) { pool_lock(); mutex_enter(&cpu_lock); mutex_enter(&pidlock); mutex_enter(&curproc->p_lock); if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind, 0, NULL, NULL) == 0) { curthread->t_bind_pset = zio_taskq_psrset_bind; } else { cmn_err(CE_WARN, "Couldn't bind process for zfs pool \"%s\" to " "pset %d\n", spa->spa_name, zio_taskq_psrset_bind); } mutex_exit(&curproc->p_lock); mutex_exit(&pidlock); mutex_exit(&cpu_lock); pool_unlock(); } #endif #ifdef SYSDC if (zio_taskq_sysdc) { sysdc_thread_enter(curthread, 100, 0); } #endif spa->spa_proc = curproc; spa->spa_did = curthread->t_did; spa_create_zio_taskqs(spa); mutex_enter(&spa->spa_proc_lock); ASSERT(spa->spa_proc_state == SPA_PROC_CREATED); spa->spa_proc_state = SPA_PROC_ACTIVE; cv_broadcast(&spa->spa_proc_cv); CALLB_CPR_SAFE_BEGIN(&cprinfo); while (spa->spa_proc_state == SPA_PROC_ACTIVE) cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock); ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE); spa->spa_proc_state = SPA_PROC_GONE; spa->spa_proc = &p0; cv_broadcast(&spa->spa_proc_cv); CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */ mutex_enter(&curproc->p_lock); lwp_exit(); } #endif /* SPA_PROCESS */ #endif /* * Activate an uninitialized pool. */ static void spa_activate(spa_t *spa, int mode) { ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); spa->spa_state = POOL_STATE_ACTIVE; spa->spa_mode = mode; spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops); spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops); /* Try to create a covering process */ mutex_enter(&spa->spa_proc_lock); ASSERT(spa->spa_proc_state == SPA_PROC_NONE); ASSERT(spa->spa_proc == &p0); spa->spa_did = 0; #ifdef SPA_PROCESS /* Only create a process if we're going to be around a while. */ if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) { if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri, NULL, 0) == 0) { spa->spa_proc_state = SPA_PROC_CREATED; while (spa->spa_proc_state == SPA_PROC_CREATED) { cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); } ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); ASSERT(spa->spa_proc != &p0); ASSERT(spa->spa_did != 0); } else { #ifdef _KERNEL cmn_err(CE_WARN, "Couldn't create process for zfs pool \"%s\"\n", spa->spa_name); #endif } } #endif /* SPA_PROCESS */ mutex_exit(&spa->spa_proc_lock); /* If we didn't create a process, we need to create our taskqs. */ ASSERT(spa->spa_proc == &p0); if (spa->spa_proc == &p0) { spa_create_zio_taskqs(spa); } /* * Start TRIM thread. */ trim_thread_create(spa); list_create(&spa->spa_config_dirty_list, sizeof (vdev_t), offsetof(vdev_t, vdev_config_dirty_node)); list_create(&spa->spa_state_dirty_list, sizeof (vdev_t), offsetof(vdev_t, vdev_state_dirty_node)); txg_list_create(&spa->spa_vdev_txg_list, offsetof(struct vdev, vdev_txg_node)); avl_create(&spa->spa_errlist_scrub, spa_error_entry_compare, sizeof (spa_error_entry_t), offsetof(spa_error_entry_t, se_avl)); avl_create(&spa->spa_errlist_last, spa_error_entry_compare, sizeof (spa_error_entry_t), offsetof(spa_error_entry_t, se_avl)); } /* * Opposite of spa_activate(). */ static void spa_deactivate(spa_t *spa) { ASSERT(spa->spa_sync_on == B_FALSE); ASSERT(spa->spa_dsl_pool == NULL); ASSERT(spa->spa_root_vdev == NULL); ASSERT(spa->spa_async_zio_root == NULL); ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); /* * Stop TRIM thread in case spa_unload() wasn't called directly * before spa_deactivate(). */ trim_thread_destroy(spa); txg_list_destroy(&spa->spa_vdev_txg_list); list_destroy(&spa->spa_config_dirty_list); list_destroy(&spa->spa_state_dirty_list); for (int t = 0; t < ZIO_TYPES; t++) { for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { spa_taskqs_fini(spa, t, q); } } metaslab_class_destroy(spa->spa_normal_class); spa->spa_normal_class = NULL; metaslab_class_destroy(spa->spa_log_class); spa->spa_log_class = NULL; /* * If this was part of an import or the open otherwise failed, we may * still have errors left in the queues. Empty them just in case. */ spa_errlog_drain(spa); avl_destroy(&spa->spa_errlist_scrub); avl_destroy(&spa->spa_errlist_last); spa->spa_state = POOL_STATE_UNINITIALIZED; mutex_enter(&spa->spa_proc_lock); if (spa->spa_proc_state != SPA_PROC_NONE) { ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); spa->spa_proc_state = SPA_PROC_DEACTIVATE; cv_broadcast(&spa->spa_proc_cv); while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) { ASSERT(spa->spa_proc != &p0); cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); } ASSERT(spa->spa_proc_state == SPA_PROC_GONE); spa->spa_proc_state = SPA_PROC_NONE; } ASSERT(spa->spa_proc == &p0); mutex_exit(&spa->spa_proc_lock); #ifdef SPA_PROCESS /* * We want to make sure spa_thread() has actually exited the ZFS * module, so that the module can't be unloaded out from underneath * it. */ if (spa->spa_did != 0) { thread_join(spa->spa_did); spa->spa_did = 0; } #endif /* SPA_PROCESS */ } /* * Verify a pool configuration, and construct the vdev tree appropriately. This * will create all the necessary vdevs in the appropriate layout, with each vdev * in the CLOSED state. This will prep the pool before open/creation/import. * All vdev validation is done by the vdev_alloc() routine. */ static int spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, int atype) { nvlist_t **child; uint_t children; int error; if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) return (error); if ((*vdp)->vdev_ops->vdev_op_leaf) return (0); error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, &child, &children); if (error == ENOENT) return (0); if (error) { vdev_free(*vdp); *vdp = NULL; return (SET_ERROR(EINVAL)); } for (int c = 0; c < children; c++) { vdev_t *vd; if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, atype)) != 0) { vdev_free(*vdp); *vdp = NULL; return (error); } } ASSERT(*vdp != NULL); return (0); } /* * Opposite of spa_load(). */ static void spa_unload(spa_t *spa) { int i; ASSERT(MUTEX_HELD(&spa_namespace_lock)); /* * Stop TRIM thread. */ trim_thread_destroy(spa); /* * Stop async tasks. */ spa_async_suspend(spa); /* * Stop syncing. */ if (spa->spa_sync_on) { txg_sync_stop(spa->spa_dsl_pool); spa->spa_sync_on = B_FALSE; } /* * Wait for any outstanding async I/O to complete. */ if (spa->spa_async_zio_root != NULL) { (void) zio_wait(spa->spa_async_zio_root); spa->spa_async_zio_root = NULL; } bpobj_close(&spa->spa_deferred_bpobj); spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); /* * Close all vdevs. */ if (spa->spa_root_vdev) vdev_free(spa->spa_root_vdev); ASSERT(spa->spa_root_vdev == NULL); /* * Close the dsl pool. */ if (spa->spa_dsl_pool) { dsl_pool_close(spa->spa_dsl_pool); spa->spa_dsl_pool = NULL; spa->spa_meta_objset = NULL; } ddt_unload(spa); /* * Drop and purge level 2 cache */ spa_l2cache_drop(spa); for (i = 0; i < spa->spa_spares.sav_count; i++) vdev_free(spa->spa_spares.sav_vdevs[i]); if (spa->spa_spares.sav_vdevs) { kmem_free(spa->spa_spares.sav_vdevs, spa->spa_spares.sav_count * sizeof (void *)); spa->spa_spares.sav_vdevs = NULL; } if (spa->spa_spares.sav_config) { nvlist_free(spa->spa_spares.sav_config); spa->spa_spares.sav_config = NULL; } spa->spa_spares.sav_count = 0; for (i = 0; i < spa->spa_l2cache.sav_count; i++) { vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]); vdev_free(spa->spa_l2cache.sav_vdevs[i]); } if (spa->spa_l2cache.sav_vdevs) { kmem_free(spa->spa_l2cache.sav_vdevs, spa->spa_l2cache.sav_count * sizeof (void *)); spa->spa_l2cache.sav_vdevs = NULL; } if (spa->spa_l2cache.sav_config) { nvlist_free(spa->spa_l2cache.sav_config); spa->spa_l2cache.sav_config = NULL; } spa->spa_l2cache.sav_count = 0; spa->spa_async_suspended = 0; if (spa->spa_comment != NULL) { spa_strfree(spa->spa_comment); spa->spa_comment = NULL; } spa_config_exit(spa, SCL_ALL, FTAG); } /* * Load (or re-load) the current list of vdevs describing the active spares for * this pool. When this is called, we have some form of basic information in * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and * then re-generate a more complete list including status information. */ static void spa_load_spares(spa_t *spa) { nvlist_t **spares; uint_t nspares; int i; vdev_t *vd, *tvd; ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); /* * First, close and free any existing spare vdevs. */ for (i = 0; i < spa->spa_spares.sav_count; i++) { vd = spa->spa_spares.sav_vdevs[i]; /* Undo the call to spa_activate() below */ if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, B_FALSE)) != NULL && tvd->vdev_isspare) spa_spare_remove(tvd); vdev_close(vd); vdev_free(vd); } if (spa->spa_spares.sav_vdevs) kmem_free(spa->spa_spares.sav_vdevs, spa->spa_spares.sav_count * sizeof (void *)); if (spa->spa_spares.sav_config == NULL) nspares = 0; else VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); spa->spa_spares.sav_count = (int)nspares; spa->spa_spares.sav_vdevs = NULL; if (nspares == 0) return; /* * Construct the array of vdevs, opening them to get status in the * process. For each spare, there is potentially two different vdev_t * structures associated with it: one in the list of spares (used only * for basic validation purposes) and one in the active vdev * configuration (if it's spared in). During this phase we open and * validate each vdev on the spare list. If the vdev also exists in the * active configuration, then we also mark this vdev as an active spare. */ spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *), KM_SLEEP); for (i = 0; i < spa->spa_spares.sav_count; i++) { VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, VDEV_ALLOC_SPARE) == 0); ASSERT(vd != NULL); spa->spa_spares.sav_vdevs[i] = vd; if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, B_FALSE)) != NULL) { if (!tvd->vdev_isspare) spa_spare_add(tvd); /* * We only mark the spare active if we were successfully * able to load the vdev. Otherwise, importing a pool * with a bad active spare would result in strange * behavior, because multiple pool would think the spare * is actively in use. * * There is a vulnerability here to an equally bizarre * circumstance, where a dead active spare is later * brought back to life (onlined or otherwise). Given * the rarity of this scenario, and the extra complexity * it adds, we ignore the possibility. */ if (!vdev_is_dead(tvd)) spa_spare_activate(tvd); } vd->vdev_top = vd; vd->vdev_aux = &spa->spa_spares; if (vdev_open(vd) != 0) continue; if (vdev_validate_aux(vd) == 0) spa_spare_add(vd); } /* * Recompute the stashed list of spares, with status information * this time. */ VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), KM_SLEEP); for (i = 0; i < spa->spa_spares.sav_count; i++) spares[i] = vdev_config_generate(spa, spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE); VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0); for (i = 0; i < spa->spa_spares.sav_count; i++) nvlist_free(spares[i]); kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); } /* * Load (or re-load) the current list of vdevs describing the active l2cache for * this pool. When this is called, we have some form of basic information in * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and * then re-generate a more complete list including status information. * Devices which are already active have their details maintained, and are * not re-opened. */ static void spa_load_l2cache(spa_t *spa) { nvlist_t **l2cache; uint_t nl2cache; int i, j, oldnvdevs; uint64_t guid; vdev_t *vd, **oldvdevs, **newvdevs; spa_aux_vdev_t *sav = &spa->spa_l2cache; ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); if (sav->sav_config != NULL) { VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); } else { nl2cache = 0; newvdevs = NULL; } oldvdevs = sav->sav_vdevs; oldnvdevs = sav->sav_count; sav->sav_vdevs = NULL; sav->sav_count = 0; /* * Process new nvlist of vdevs. */ for (i = 0; i < nl2cache; i++) { VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID, &guid) == 0); newvdevs[i] = NULL; for (j = 0; j < oldnvdevs; j++) { vd = oldvdevs[j]; if (vd != NULL && guid == vd->vdev_guid) { /* * Retain previous vdev for add/remove ops. */ newvdevs[i] = vd; oldvdevs[j] = NULL; break; } } if (newvdevs[i] == NULL) { /* * Create new vdev */ VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, VDEV_ALLOC_L2CACHE) == 0); ASSERT(vd != NULL); newvdevs[i] = vd; /* * Commit this vdev as an l2cache device, * even if it fails to open. */ spa_l2cache_add(vd); vd->vdev_top = vd; vd->vdev_aux = sav; spa_l2cache_activate(vd); if (vdev_open(vd) != 0) continue; (void) vdev_validate_aux(vd); if (!vdev_is_dead(vd)) l2arc_add_vdev(spa, vd); } } /* * Purge vdevs that were dropped */ for (i = 0; i < oldnvdevs; i++) { uint64_t pool; vd = oldvdevs[i]; if (vd != NULL) { ASSERT(vd->vdev_isl2cache); if (spa_l2cache_exists(vd->vdev_guid, &pool) && pool != 0ULL && l2arc_vdev_present(vd)) l2arc_remove_vdev(vd); vdev_clear_stats(vd); vdev_free(vd); } } if (oldvdevs) kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); if (sav->sav_config == NULL) goto out; sav->sav_vdevs = newvdevs; sav->sav_count = (int)nl2cache; /* * Recompute the stashed list of l2cache devices, with status * information this time. */ VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0); l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); for (i = 0; i < sav->sav_count; i++) l2cache[i] = vdev_config_generate(spa, sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE); VERIFY(nvlist_add_nvlist_array(sav->sav_config, ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0); out: for (i = 0; i < sav->sav_count; i++) nvlist_free(l2cache[i]); if (sav->sav_count) kmem_free(l2cache, sav->sav_count * sizeof (void *)); } static int load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) { dmu_buf_t *db; char *packed = NULL; size_t nvsize = 0; int error; *value = NULL; error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db); if (error != 0) return (error); nvsize = *(uint64_t *)db->db_data; dmu_buf_rele(db, FTAG); packed = kmem_alloc(nvsize, KM_SLEEP); error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed, DMU_READ_PREFETCH); if (error == 0) error = nvlist_unpack(packed, nvsize, value, 0); kmem_free(packed, nvsize); return (error); } /* * Checks to see if the given vdev could not be opened, in which case we post a * sysevent to notify the autoreplace code that the device has been removed. */ static void spa_check_removed(vdev_t *vd) { for (int c = 0; c < vd->vdev_children; c++) spa_check_removed(vd->vdev_child[c]); if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) && !vd->vdev_ishole) { zfs_post_autoreplace(vd->vdev_spa, vd); spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK); } } /* * Validate the current config against the MOS config */ static boolean_t spa_config_valid(spa_t *spa, nvlist_t *config) { vdev_t *mrvd, *rvd = spa->spa_root_vdev; nvlist_t *nv; VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0); spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0); ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children); /* * If we're doing a normal import, then build up any additional * diagnostic information about missing devices in this config. * We'll pass this up to the user for further processing. */ if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) { nvlist_t **child, *nv; uint64_t idx = 0; child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **), KM_SLEEP); VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); for (int c = 0; c < rvd->vdev_children; c++) { vdev_t *tvd = rvd->vdev_child[c]; vdev_t *mtvd = mrvd->vdev_child[c]; if (tvd->vdev_ops == &vdev_missing_ops && mtvd->vdev_ops != &vdev_missing_ops && mtvd->vdev_islog) child[idx++] = vdev_config_generate(spa, mtvd, B_FALSE, 0); } if (idx) { VERIFY(nvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, child, idx) == 0); VERIFY(nvlist_add_nvlist(spa->spa_load_info, ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0); for (int i = 0; i < idx; i++) nvlist_free(child[i]); } nvlist_free(nv); kmem_free(child, rvd->vdev_children * sizeof (char **)); } /* * Compare the root vdev tree with the information we have * from the MOS config (mrvd). Check each top-level vdev * with the corresponding MOS config top-level (mtvd). */ for (int c = 0; c < rvd->vdev_children; c++) { vdev_t *tvd = rvd->vdev_child[c]; vdev_t *mtvd = mrvd->vdev_child[c]; /* * Resolve any "missing" vdevs in the current configuration. * If we find that the MOS config has more accurate information * about the top-level vdev then use that vdev instead. */ if (tvd->vdev_ops == &vdev_missing_ops && mtvd->vdev_ops != &vdev_missing_ops) { if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) continue; /* * Device specific actions. */ if (mtvd->vdev_islog) { spa_set_log_state(spa, SPA_LOG_CLEAR); } else { /* * XXX - once we have 'readonly' pool * support we should be able to handle * missing data devices by transitioning * the pool to readonly. */ continue; } /* * Swap the missing vdev with the data we were * able to obtain from the MOS config. */ vdev_remove_child(rvd, tvd); vdev_remove_child(mrvd, mtvd); vdev_add_child(rvd, mtvd); vdev_add_child(mrvd, tvd); spa_config_exit(spa, SCL_ALL, FTAG); vdev_load(mtvd); spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); vdev_reopen(rvd); } else if (mtvd->vdev_islog) { /* * Load the slog device's state from the MOS config * since it's possible that the label does not * contain the most up-to-date information. */ vdev_load_log_state(tvd, mtvd); vdev_reopen(tvd); } } vdev_free(mrvd); spa_config_exit(spa, SCL_ALL, FTAG); /* * Ensure we were able to validate the config. */ return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum); } /* * Check for missing log devices */ static boolean_t spa_check_logs(spa_t *spa) { boolean_t rv = B_FALSE; switch (spa->spa_log_state) { case SPA_LOG_MISSING: /* need to recheck in case slog has been restored */ case SPA_LOG_UNKNOWN: rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0); if (rv) spa_set_log_state(spa, SPA_LOG_MISSING); break; } return (rv); } static boolean_t spa_passivate_log(spa_t *spa) { vdev_t *rvd = spa->spa_root_vdev; boolean_t slog_found = B_FALSE; ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); if (!spa_has_slogs(spa)) return (B_FALSE); for (int c = 0; c < rvd->vdev_children; c++) { vdev_t *tvd = rvd->vdev_child[c]; metaslab_group_t *mg = tvd->vdev_mg; if (tvd->vdev_islog) { metaslab_group_passivate(mg); slog_found = B_TRUE; } } return (slog_found); } static void spa_activate_log(spa_t *spa) { vdev_t *rvd = spa->spa_root_vdev; ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); for (int c = 0; c < rvd->vdev_children; c++) { vdev_t *tvd = rvd->vdev_child[c]; metaslab_group_t *mg = tvd->vdev_mg; if (tvd->vdev_islog) metaslab_group_activate(mg); } } int spa_offline_log(spa_t *spa) { int error; error = dmu_objset_find(spa_name(spa), zil_vdev_offline, NULL, DS_FIND_CHILDREN); if (error == 0) { /* * We successfully offlined the log device, sync out the * current txg so that the "stubby" block can be removed * by zil_sync(). */ txg_wait_synced(spa->spa_dsl_pool, 0); } return (error); } static void spa_aux_check_removed(spa_aux_vdev_t *sav) { int i; for (i = 0; i < sav->sav_count; i++) spa_check_removed(sav->sav_vdevs[i]); } void spa_claim_notify(zio_t *zio) { spa_t *spa = zio->io_spa; if (zio->io_error) return; mutex_enter(&spa->spa_props_lock); /* any mutex will do */ if (spa->spa_claim_max_txg < zio->io_bp->blk_birth) spa->spa_claim_max_txg = zio->io_bp->blk_birth; mutex_exit(&spa->spa_props_lock); } typedef struct spa_load_error { uint64_t sle_meta_count; uint64_t sle_data_count; } spa_load_error_t; static void spa_load_verify_done(zio_t *zio) { blkptr_t *bp = zio->io_bp; spa_load_error_t *sle = zio->io_private; dmu_object_type_t type = BP_GET_TYPE(bp); int error = zio->io_error; spa_t *spa = zio->io_spa; if (error) { if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) && type != DMU_OT_INTENT_LOG) - atomic_add_64(&sle->sle_meta_count, 1); + atomic_inc_64(&sle->sle_meta_count); else - atomic_add_64(&sle->sle_data_count, 1); + atomic_inc_64(&sle->sle_data_count); } zio_data_buf_free(zio->io_data, zio->io_size); mutex_enter(&spa->spa_scrub_lock); spa->spa_scrub_inflight--; cv_broadcast(&spa->spa_scrub_io_cv); mutex_exit(&spa->spa_scrub_lock); } /* * Maximum number of concurrent scrub i/os to create while verifying * a pool while importing it. */ int spa_load_verify_maxinflight = 10000; boolean_t spa_load_verify_metadata = B_TRUE; boolean_t spa_load_verify_data = B_TRUE; SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_maxinflight, CTLFLAG_RWTUN, &spa_load_verify_maxinflight, 0, "Maximum number of concurrent scrub I/Os to create while verifying a " "pool while importing it"); SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_metadata, CTLFLAG_RWTUN, &spa_load_verify_metadata, 0, "Check metadata on import?"); SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_data, CTLFLAG_RWTUN, &spa_load_verify_data, 0, "Check user data on import?"); /*ARGSUSED*/ static int spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) { if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) return (0); /* * Note: normally this routine will not be called if * spa_load_verify_metadata is not set. However, it may be useful * to manually set the flag after the traversal has begun. */ if (!spa_load_verify_metadata) return (0); if (BP_GET_BUFC_TYPE(bp) == ARC_BUFC_DATA && !spa_load_verify_data) return (0); zio_t *rio = arg; size_t size = BP_GET_PSIZE(bp); void *data = zio_data_buf_alloc(size); mutex_enter(&spa->spa_scrub_lock); while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight) cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); spa->spa_scrub_inflight++; mutex_exit(&spa->spa_scrub_lock); zio_nowait(zio_read(rio, spa, bp, data, size, spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB, ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL | ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb)); return (0); } static int spa_load_verify(spa_t *spa) { zio_t *rio; spa_load_error_t sle = { 0 }; zpool_rewind_policy_t policy; boolean_t verify_ok = B_FALSE; int error = 0; zpool_get_rewind_policy(spa->spa_config, &policy); if (policy.zrp_request & ZPOOL_NEVER_REWIND) return (0); rio = zio_root(spa, NULL, &sle, ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); if (spa_load_verify_metadata) { error = traverse_pool(spa, spa->spa_verify_min_txg, TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, spa_load_verify_cb, rio); } (void) zio_wait(rio); spa->spa_load_meta_errors = sle.sle_meta_count; spa->spa_load_data_errors = sle.sle_data_count; if (!error && sle.sle_meta_count <= policy.zrp_maxmeta && sle.sle_data_count <= policy.zrp_maxdata) { int64_t loss = 0; verify_ok = B_TRUE; spa->spa_load_txg = spa->spa_uberblock.ub_txg; spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp; loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts; VERIFY(nvlist_add_uint64(spa->spa_load_info, ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0); VERIFY(nvlist_add_int64(spa->spa_load_info, ZPOOL_CONFIG_REWIND_TIME, loss) == 0); VERIFY(nvlist_add_uint64(spa->spa_load_info, ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0); } else { spa->spa_load_max_txg = spa->spa_uberblock.ub_txg; } if (error) { if (error != ENXIO && error != EIO) error = SET_ERROR(EIO); return (error); } return (verify_ok ? 0 : EIO); } /* * Find a value in the pool props object. */ static void spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val) { (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object, zpool_prop_to_name(prop), sizeof (uint64_t), 1, val); } /* * Find a value in the pool directory object. */ static int spa_dir_prop(spa_t *spa, const char *name, uint64_t *val) { return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, name, sizeof (uint64_t), 1, val)); } static int spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err) { vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux); return (err); } /* * Fix up config after a partly-completed split. This is done with the * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off * pool have that entry in their config, but only the splitting one contains * a list of all the guids of the vdevs that are being split off. * * This function determines what to do with that list: either rejoin * all the disks to the pool, or complete the splitting process. To attempt * the rejoin, each disk that is offlined is marked online again, and * we do a reopen() call. If the vdev label for every disk that was * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL) * then we call vdev_split() on each disk, and complete the split. * * Otherwise we leave the config alone, with all the vdevs in place in * the original pool. */ static void spa_try_repair(spa_t *spa, nvlist_t *config) { uint_t extracted; uint64_t *glist; uint_t i, gcount; nvlist_t *nvl; vdev_t **vd; boolean_t attempt_reopen; if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0) return; /* check that the config is complete */ if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, &glist, &gcount) != 0) return; vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP); /* attempt to online all the vdevs & validate */ attempt_reopen = B_TRUE; for (i = 0; i < gcount; i++) { if (glist[i] == 0) /* vdev is hole */ continue; vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE); if (vd[i] == NULL) { /* * Don't bother attempting to reopen the disks; * just do the split. */ attempt_reopen = B_FALSE; } else { /* attempt to re-online it */ vd[i]->vdev_offline = B_FALSE; } } if (attempt_reopen) { vdev_reopen(spa->spa_root_vdev); /* check each device to see what state it's in */ for (extracted = 0, i = 0; i < gcount; i++) { if (vd[i] != NULL && vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL) break; ++extracted; } } /* * If every disk has been moved to the new pool, or if we never * even attempted to look at them, then we split them off for * good. */ if (!attempt_reopen || gcount == extracted) { for (i = 0; i < gcount; i++) if (vd[i] != NULL) vdev_split(vd[i]); vdev_reopen(spa->spa_root_vdev); } kmem_free(vd, gcount * sizeof (vdev_t *)); } static int spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig) { nvlist_t *config = spa->spa_config; char *ereport = FM_EREPORT_ZFS_POOL; char *comment; int error; uint64_t pool_guid; nvlist_t *nvl; if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) return (SET_ERROR(EINVAL)); ASSERT(spa->spa_comment == NULL); if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0) spa->spa_comment = spa_strdup(comment); /* * Versioning wasn't explicitly added to the label until later, so if * it's not present treat it as the initial version. */ if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &spa->spa_ubsync.ub_version) != 0) spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &spa->spa_config_txg); if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) && spa_guid_exists(pool_guid, 0)) { error = SET_ERROR(EEXIST); } else { spa->spa_config_guid = pool_guid; if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) == 0) { VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting, KM_SLEEP) == 0); } nvlist_free(spa->spa_load_info); spa->spa_load_info = fnvlist_alloc(); gethrestime(&spa->spa_loaded_ts); error = spa_load_impl(spa, pool_guid, config, state, type, mosconfig, &ereport); } spa->spa_minref = refcount_count(&spa->spa_refcount); if (error) { if (error != EEXIST) { spa->spa_loaded_ts.tv_sec = 0; spa->spa_loaded_ts.tv_nsec = 0; } if (error != EBADF) { zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0); } } spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE; spa->spa_ena = 0; return (error); } /* * Load an existing storage pool, using the pool's builtin spa_config as a * source of configuration information. */ static int spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config, spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, char **ereport) { int error = 0; nvlist_t *nvroot = NULL; nvlist_t *label; vdev_t *rvd; uberblock_t *ub = &spa->spa_uberblock; uint64_t children, config_cache_txg = spa->spa_config_txg; int orig_mode = spa->spa_mode; int parse; uint64_t obj; boolean_t missing_feat_write = B_FALSE; /* * If this is an untrusted config, access the pool in read-only mode. * This prevents things like resilvering recently removed devices. */ if (!mosconfig) spa->spa_mode = FREAD; ASSERT(MUTEX_HELD(&spa_namespace_lock)); spa->spa_load_state = state; if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot)) return (SET_ERROR(EINVAL)); parse = (type == SPA_IMPORT_EXISTING ? VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT); /* * Create "The Godfather" zio to hold all async IOs */ spa->spa_async_zio_root = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER); /* * Parse the configuration into a vdev tree. We explicitly set the * value that will be returned by spa_version() since parsing the * configuration requires knowing the version number. */ spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse); spa_config_exit(spa, SCL_ALL, FTAG); if (error != 0) return (error); ASSERT(spa->spa_root_vdev == rvd); if (type != SPA_IMPORT_ASSEMBLE) { ASSERT(spa_guid(spa) == pool_guid); } /* * Try to open all vdevs, loading each label in the process. */ spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); error = vdev_open(rvd); spa_config_exit(spa, SCL_ALL, FTAG); if (error != 0) return (error); /* * We need to validate the vdev labels against the configuration that * we have in hand, which is dependent on the setting of mosconfig. If * mosconfig is true then we're validating the vdev labels based on * that config. Otherwise, we're validating against the cached config * (zpool.cache) that was read when we loaded the zfs module, and then * later we will recursively call spa_load() and validate against * the vdev config. * * If we're assembling a new pool that's been split off from an * existing pool, the labels haven't yet been updated so we skip * validation for now. */ if (type != SPA_IMPORT_ASSEMBLE) { spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); error = vdev_validate(rvd, mosconfig); spa_config_exit(spa, SCL_ALL, FTAG); if (error != 0) return (error); if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) return (SET_ERROR(ENXIO)); } /* * Find the best uberblock. */ vdev_uberblock_load(rvd, ub, &label); /* * If we weren't able to find a single valid uberblock, return failure. */ if (ub->ub_txg == 0) { nvlist_free(label); return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); } /* * If the pool has an unsupported version we can't open it. */ if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) { nvlist_free(label); return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); } if (ub->ub_version >= SPA_VERSION_FEATURES) { nvlist_t *features; /* * If we weren't able to find what's necessary for reading the * MOS in the label, return failure. */ if (label == NULL || nvlist_lookup_nvlist(label, ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) { nvlist_free(label); return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); } /* * Update our in-core representation with the definitive values * from the label. */ nvlist_free(spa->spa_label_features); VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0); } nvlist_free(label); /* * Look through entries in the label nvlist's features_for_read. If * there is a feature listed there which we don't understand then we * cannot open a pool. */ if (ub->ub_version >= SPA_VERSION_FEATURES) { nvlist_t *unsup_feat; VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) == 0); for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features, NULL); nvp != NULL; nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) { if (!zfeature_is_supported(nvpair_name(nvp))) { VERIFY(nvlist_add_string(unsup_feat, nvpair_name(nvp), "") == 0); } } if (!nvlist_empty(unsup_feat)) { VERIFY(nvlist_add_nvlist(spa->spa_load_info, ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0); nvlist_free(unsup_feat); return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP)); } nvlist_free(unsup_feat); } /* * If the vdev guid sum doesn't match the uberblock, we have an * incomplete configuration. We first check to see if the pool * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN). * If it is, defer the vdev_guid_sum check till later so we * can handle missing vdevs. */ if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE && rvd->vdev_guid_sum != ub->ub_guid_sum) return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) { spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); spa_try_repair(spa, config); spa_config_exit(spa, SCL_ALL, FTAG); nvlist_free(spa->spa_config_splitting); spa->spa_config_splitting = NULL; } /* * Initialize internal SPA structures. */ spa->spa_state = POOL_STATE_ACTIVE; spa->spa_ubsync = spa->spa_uberblock; spa->spa_verify_min_txg = spa->spa_extreme_rewind ? TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1; spa->spa_first_txg = spa->spa_last_ubsync_txg ? spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1; spa->spa_claim_max_txg = spa->spa_first_txg; spa->spa_prev_software_version = ub->ub_software_version; error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool); if (error) return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0) return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); if (spa_version(spa) >= SPA_VERSION_FEATURES) { boolean_t missing_feat_read = B_FALSE; nvlist_t *unsup_feat, *enabled_feat; if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ, &spa->spa_feat_for_read_obj) != 0) { return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); } if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE, &spa->spa_feat_for_write_obj) != 0) { return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); } if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS, &spa->spa_feat_desc_obj) != 0) { return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); } enabled_feat = fnvlist_alloc(); unsup_feat = fnvlist_alloc(); if (!spa_features_check(spa, B_FALSE, unsup_feat, enabled_feat)) missing_feat_read = B_TRUE; if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) { if (!spa_features_check(spa, B_TRUE, unsup_feat, enabled_feat)) { missing_feat_write = B_TRUE; } } fnvlist_add_nvlist(spa->spa_load_info, ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat); if (!nvlist_empty(unsup_feat)) { fnvlist_add_nvlist(spa->spa_load_info, ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); } fnvlist_free(enabled_feat); fnvlist_free(unsup_feat); if (!missing_feat_read) { fnvlist_add_boolean(spa->spa_load_info, ZPOOL_CONFIG_CAN_RDONLY); } /* * If the state is SPA_LOAD_TRYIMPORT, our objective is * twofold: to determine whether the pool is available for * import in read-write mode and (if it is not) whether the * pool is available for import in read-only mode. If the pool * is available for import in read-write mode, it is displayed * as available in userland; if it is not available for import * in read-only mode, it is displayed as unavailable in * userland. If the pool is available for import in read-only * mode but not read-write mode, it is displayed as unavailable * in userland with a special note that the pool is actually * available for open in read-only mode. * * As a result, if the state is SPA_LOAD_TRYIMPORT and we are * missing a feature for write, we must first determine whether * the pool can be opened read-only before returning to * userland in order to know whether to display the * abovementioned note. */ if (missing_feat_read || (missing_feat_write && spa_writeable(spa))) { return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP)); } /* * Load refcounts for ZFS features from disk into an in-memory * cache during SPA initialization. */ for (spa_feature_t i = 0; i < SPA_FEATURES; i++) { uint64_t refcount; error = feature_get_refcount_from_disk(spa, &spa_feature_table[i], &refcount); if (error == 0) { spa->spa_feat_refcount_cache[i] = refcount; } else if (error == ENOTSUP) { spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED; } else { return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); } } } if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) { if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG, &spa->spa_feat_enabled_txg_obj) != 0) return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); } spa->spa_is_initializing = B_TRUE; error = dsl_pool_open(spa->spa_dsl_pool); spa->spa_is_initializing = B_FALSE; if (error != 0) return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); if (!mosconfig) { uint64_t hostid; nvlist_t *policy = NULL, *nvconfig; if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig, ZPOOL_CONFIG_HOSTID, &hostid) == 0) { char *hostname; unsigned long myhostid = 0; VERIFY(nvlist_lookup_string(nvconfig, ZPOOL_CONFIG_HOSTNAME, &hostname) == 0); #ifdef _KERNEL myhostid = zone_get_hostid(NULL); #else /* _KERNEL */ /* * We're emulating the system's hostid in userland, so * we can't use zone_get_hostid(). */ (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid); #endif /* _KERNEL */ if (check_hostid && hostid != 0 && myhostid != 0 && hostid != myhostid) { nvlist_free(nvconfig); cmn_err(CE_WARN, "pool '%s' could not be " "loaded as it was last accessed by " "another system (host: %s hostid: 0x%lx). " "See: http://illumos.org/msg/ZFS-8000-EY", spa_name(spa), hostname, (unsigned long)hostid); return (SET_ERROR(EBADF)); } } if (nvlist_lookup_nvlist(spa->spa_config, ZPOOL_REWIND_POLICY, &policy) == 0) VERIFY(nvlist_add_nvlist(nvconfig, ZPOOL_REWIND_POLICY, policy) == 0); spa_config_set(spa, nvconfig); spa_unload(spa); spa_deactivate(spa); spa_activate(spa, orig_mode); return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE)); } if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0) return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj); if (error != 0) return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); /* * Load the bit that tells us to use the new accounting function * (raid-z deflation). If we have an older pool, this will not * be present. */ error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate); if (error != 0 && error != ENOENT) return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION, &spa->spa_creation_version); if (error != 0 && error != ENOENT) return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); /* * Load the persistent error log. If we have an older pool, this will * not be present. */ error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last); if (error != 0 && error != ENOENT) return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB, &spa->spa_errlog_scrub); if (error != 0 && error != ENOENT) return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); /* * Load the history object. If we have an older pool, this * will not be present. */ error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history); if (error != 0 && error != ENOENT) return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); /* * If we're assembling the pool from the split-off vdevs of * an existing pool, we don't want to attach the spares & cache * devices. */ /* * Load any hot spares for this pool. */ error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object); if (error != 0 && error != ENOENT) return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); if (load_nvlist(spa, spa->spa_spares.sav_object, &spa->spa_spares.sav_config) != 0) return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); spa_load_spares(spa); spa_config_exit(spa, SCL_ALL, FTAG); } else if (error == 0) { spa->spa_spares.sav_sync = B_TRUE; } /* * Load any level 2 ARC devices for this pool. */ error = spa_dir_prop(spa, DMU_POOL_L2CACHE, &spa->spa_l2cache.sav_object); if (error != 0 && error != ENOENT) return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); if (load_nvlist(spa, spa->spa_l2cache.sav_object, &spa->spa_l2cache.sav_config) != 0) return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); spa_load_l2cache(spa); spa_config_exit(spa, SCL_ALL, FTAG); } else if (error == 0) { spa->spa_l2cache.sav_sync = B_TRUE; } spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object); if (error && error != ENOENT) return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); if (error == 0) { uint64_t autoreplace; spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs); spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace); spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation); spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode); spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand); spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO, &spa->spa_dedup_ditto); spa->spa_autoreplace = (autoreplace != 0); } /* * If the 'autoreplace' property is set, then post a resource notifying * the ZFS DE that it should not issue any faults for unopenable * devices. We also iterate over the vdevs, and post a sysevent for any * unopenable vdevs so that the normal autoreplace handler can take * over. */ if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) { spa_check_removed(spa->spa_root_vdev); /* * For the import case, this is done in spa_import(), because * at this point we're using the spare definitions from * the MOS config, not necessarily from the userland config. */ if (state != SPA_LOAD_IMPORT) { spa_aux_check_removed(&spa->spa_spares); spa_aux_check_removed(&spa->spa_l2cache); } } /* * Load the vdev state for all toplevel vdevs. */ vdev_load(rvd); /* * Propagate the leaf DTLs we just loaded all the way up the tree. */ spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); vdev_dtl_reassess(rvd, 0, 0, B_FALSE); spa_config_exit(spa, SCL_ALL, FTAG); /* * Load the DDTs (dedup tables). */ error = ddt_load(spa); if (error != 0) return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); spa_update_dspace(spa); /* * Validate the config, using the MOS config to fill in any * information which might be missing. If we fail to validate * the config then declare the pool unfit for use. If we're * assembling a pool from a split, the log is not transferred * over. */ if (type != SPA_IMPORT_ASSEMBLE) { nvlist_t *nvconfig; if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); if (!spa_config_valid(spa, nvconfig)) { nvlist_free(nvconfig); return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); } nvlist_free(nvconfig); /* * Now that we've validated the config, check the state of the * root vdev. If it can't be opened, it indicates one or * more toplevel vdevs are faulted. */ if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) return (SET_ERROR(ENXIO)); if (spa_check_logs(spa)) { *ereport = FM_EREPORT_ZFS_LOG_REPLAY; return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO)); } } if (missing_feat_write) { ASSERT(state == SPA_LOAD_TRYIMPORT); /* * At this point, we know that we can open the pool in * read-only mode but not read-write mode. We now have enough * information and can return to userland. */ return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP)); } /* * We've successfully opened the pool, verify that we're ready * to start pushing transactions. */ if (state != SPA_LOAD_TRYIMPORT) { if (error = spa_load_verify(spa)) return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error)); } if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER || spa->spa_load_max_txg == UINT64_MAX)) { dmu_tx_t *tx; int need_update = B_FALSE; ASSERT(state != SPA_LOAD_TRYIMPORT); /* * Claim log blocks that haven't been committed yet. * This must all happen in a single txg. * Note: spa_claim_max_txg is updated by spa_claim_notify(), * invoked from zil_claim_log_block()'s i/o done callback. * Price of rollback is that we abandon the log. */ spa->spa_claiming = B_TRUE; tx = dmu_tx_create_assigned(spa_get_dsl(spa), spa_first_txg(spa)); (void) dmu_objset_find(spa_name(spa), zil_claim, tx, DS_FIND_CHILDREN); dmu_tx_commit(tx); spa->spa_claiming = B_FALSE; spa_set_log_state(spa, SPA_LOG_GOOD); spa->spa_sync_on = B_TRUE; txg_sync_start(spa->spa_dsl_pool); /* * Wait for all claims to sync. We sync up to the highest * claimed log block birth time so that claimed log blocks * don't appear to be from the future. spa_claim_max_txg * will have been set for us by either zil_check_log_chain() * (invoked from spa_check_logs()) or zil_claim() above. */ txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg); /* * If the config cache is stale, or we have uninitialized * metaslabs (see spa_vdev_add()), then update the config. * * If this is a verbatim import, trust the current * in-core spa_config and update the disk labels. */ if (config_cache_txg != spa->spa_config_txg || state == SPA_LOAD_IMPORT || state == SPA_LOAD_RECOVER || (spa->spa_import_flags & ZFS_IMPORT_VERBATIM)) need_update = B_TRUE; for (int c = 0; c < rvd->vdev_children; c++) if (rvd->vdev_child[c]->vdev_ms_array == 0) need_update = B_TRUE; /* * Update the config cache asychronously in case we're the * root pool, in which case the config cache isn't writable yet. */ if (need_update) spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); /* * Check all DTLs to see if anything needs resilvering. */ if (!dsl_scan_resilvering(spa->spa_dsl_pool) && vdev_resilver_needed(rvd, NULL, NULL)) spa_async_request(spa, SPA_ASYNC_RESILVER); /* * Log the fact that we booted up (so that we can detect if * we rebooted in the middle of an operation). */ spa_history_log_version(spa, "open"); /* * Delete any inconsistent datasets. */ (void) dmu_objset_find(spa_name(spa), dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN); /* * Clean up any stale temporary dataset userrefs. */ dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool); } return (0); } static int spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig) { int mode = spa->spa_mode; spa_unload(spa); spa_deactivate(spa); spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1; spa_activate(spa, mode); spa_async_suspend(spa); return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig)); } /* * If spa_load() fails this function will try loading prior txg's. If * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this * function will not rewind the pool and will return the same error as * spa_load(). */ static int spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig, uint64_t max_request, int rewind_flags) { nvlist_t *loadinfo = NULL; nvlist_t *config = NULL; int load_error, rewind_error; uint64_t safe_rewind_txg; uint64_t min_txg; if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) { spa->spa_load_max_txg = spa->spa_load_txg; spa_set_log_state(spa, SPA_LOG_CLEAR); } else { spa->spa_load_max_txg = max_request; if (max_request != UINT64_MAX) spa->spa_extreme_rewind = B_TRUE; } load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig); if (load_error == 0) return (0); if (spa->spa_root_vdev != NULL) config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg; spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp; if (rewind_flags & ZPOOL_NEVER_REWIND) { nvlist_free(config); return (load_error); } if (state == SPA_LOAD_RECOVER) { /* Price of rolling back is discarding txgs, including log */ spa_set_log_state(spa, SPA_LOG_CLEAR); } else { /* * If we aren't rolling back save the load info from our first * import attempt so that we can restore it after attempting * to rewind. */ loadinfo = spa->spa_load_info; spa->spa_load_info = fnvlist_alloc(); } spa->spa_load_max_txg = spa->spa_last_ubsync_txg; safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ? TXG_INITIAL : safe_rewind_txg; /* * Continue as long as we're finding errors, we're still within * the acceptable rewind range, and we're still finding uberblocks */ while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg && spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) { if (spa->spa_load_max_txg < safe_rewind_txg) spa->spa_extreme_rewind = B_TRUE; rewind_error = spa_load_retry(spa, state, mosconfig); } spa->spa_extreme_rewind = B_FALSE; spa->spa_load_max_txg = UINT64_MAX; if (config && (rewind_error || state != SPA_LOAD_RECOVER)) spa_config_set(spa, config); if (state == SPA_LOAD_RECOVER) { ASSERT3P(loadinfo, ==, NULL); return (rewind_error); } else { /* Store the rewind info as part of the initial load info */ fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO, spa->spa_load_info); /* Restore the initial load info */ fnvlist_free(spa->spa_load_info); spa->spa_load_info = loadinfo; return (load_error); } } /* * Pool Open/Import * * The import case is identical to an open except that the configuration is sent * down from userland, instead of grabbed from the configuration cache. For the * case of an open, the pool configuration will exist in the * POOL_STATE_UNINITIALIZED state. * * The stats information (gen/count/ustats) is used to gather vdev statistics at * the same time open the pool, without having to keep around the spa_t in some * ambiguous state. */ static int spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy, nvlist_t **config) { spa_t *spa; spa_load_state_t state = SPA_LOAD_OPEN; int error; int locked = B_FALSE; int firstopen = B_FALSE; *spapp = NULL; /* * As disgusting as this is, we need to support recursive calls to this * function because dsl_dir_open() is called during spa_load(), and ends * up calling spa_open() again. The real fix is to figure out how to * avoid dsl_dir_open() calling this in the first place. */ if (mutex_owner(&spa_namespace_lock) != curthread) { mutex_enter(&spa_namespace_lock); locked = B_TRUE; } if ((spa = spa_lookup(pool)) == NULL) { if (locked) mutex_exit(&spa_namespace_lock); return (SET_ERROR(ENOENT)); } if (spa->spa_state == POOL_STATE_UNINITIALIZED) { zpool_rewind_policy_t policy; firstopen = B_TRUE; zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config, &policy); if (policy.zrp_request & ZPOOL_DO_REWIND) state = SPA_LOAD_RECOVER; spa_activate(spa, spa_mode_global); if (state != SPA_LOAD_RECOVER) spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg, policy.zrp_request); if (error == EBADF) { /* * If vdev_validate() returns failure (indicated by * EBADF), it indicates that one of the vdevs indicates * that the pool has been exported or destroyed. If * this is the case, the config cache is out of sync and * we should remove the pool from the namespace. */ spa_unload(spa); spa_deactivate(spa); spa_config_sync(spa, B_TRUE, B_TRUE); spa_remove(spa); if (locked) mutex_exit(&spa_namespace_lock); return (SET_ERROR(ENOENT)); } if (error) { /* * We can't open the pool, but we still have useful * information: the state of each vdev after the * attempted vdev_open(). Return this to the user. */ if (config != NULL && spa->spa_config) { VERIFY(nvlist_dup(spa->spa_config, config, KM_SLEEP) == 0); VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, spa->spa_load_info) == 0); } spa_unload(spa); spa_deactivate(spa); spa->spa_last_open_failed = error; if (locked) mutex_exit(&spa_namespace_lock); *spapp = NULL; return (error); } } spa_open_ref(spa, tag); if (config != NULL) *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); /* * If we've recovered the pool, pass back any information we * gathered while doing the load. */ if (state == SPA_LOAD_RECOVER) { VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, spa->spa_load_info) == 0); } if (locked) { spa->spa_last_open_failed = 0; spa->spa_last_ubsync_txg = 0; spa->spa_load_txg = 0; mutex_exit(&spa_namespace_lock); #ifdef __FreeBSD__ #ifdef _KERNEL if (firstopen) zvol_create_minors(spa->spa_name); #endif #endif } *spapp = spa; return (0); } int spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy, nvlist_t **config) { return (spa_open_common(name, spapp, tag, policy, config)); } int spa_open(const char *name, spa_t **spapp, void *tag) { return (spa_open_common(name, spapp, tag, NULL, NULL)); } /* * Lookup the given spa_t, incrementing the inject count in the process, * preventing it from being exported or destroyed. */ spa_t * spa_inject_addref(char *name) { spa_t *spa; mutex_enter(&spa_namespace_lock); if ((spa = spa_lookup(name)) == NULL) { mutex_exit(&spa_namespace_lock); return (NULL); } spa->spa_inject_ref++; mutex_exit(&spa_namespace_lock); return (spa); } void spa_inject_delref(spa_t *spa) { mutex_enter(&spa_namespace_lock); spa->spa_inject_ref--; mutex_exit(&spa_namespace_lock); } /* * Add spares device information to the nvlist. */ static void spa_add_spares(spa_t *spa, nvlist_t *config) { nvlist_t **spares; uint_t i, nspares; nvlist_t *nvroot; uint64_t guid; vdev_stat_t *vs; uint_t vsc; uint64_t pool; ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); if (spa->spa_spares.sav_count == 0) return; VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); if (nspares != 0) { VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, spares, nspares) == 0); VERIFY(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); /* * Go through and find any spares which have since been * repurposed as an active spare. If this is the case, update * their status appropriately. */ for (i = 0; i < nspares; i++) { VERIFY(nvlist_lookup_uint64(spares[i], ZPOOL_CONFIG_GUID, &guid) == 0); if (spa_spare_exists(guid, &pool, NULL) && pool != 0ULL) { VERIFY(nvlist_lookup_uint64_array( spares[i], ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc) == 0); vs->vs_state = VDEV_STATE_CANT_OPEN; vs->vs_aux = VDEV_AUX_SPARED; } } } } /* * Add l2cache device information to the nvlist, including vdev stats. */ static void spa_add_l2cache(spa_t *spa, nvlist_t *config) { nvlist_t **l2cache; uint_t i, j, nl2cache; nvlist_t *nvroot; uint64_t guid; vdev_t *vd; vdev_stat_t *vs; uint_t vsc; ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); if (spa->spa_l2cache.sav_count == 0) return; VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); if (nl2cache != 0) { VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); VERIFY(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); /* * Update level 2 cache device stats. */ for (i = 0; i < nl2cache; i++) { VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID, &guid) == 0); vd = NULL; for (j = 0; j < spa->spa_l2cache.sav_count; j++) { if (guid == spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { vd = spa->spa_l2cache.sav_vdevs[j]; break; } } ASSERT(vd != NULL); VERIFY(nvlist_lookup_uint64_array(l2cache[i], ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc) == 0); vdev_get_stats(vd, vs); } } } static void spa_add_feature_stats(spa_t *spa, nvlist_t *config) { nvlist_t *features; zap_cursor_t zc; zap_attribute_t za; ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0); /* We may be unable to read features if pool is suspended. */ if (spa_suspended(spa)) goto out; if (spa->spa_feat_for_read_obj != 0) { for (zap_cursor_init(&zc, spa->spa_meta_objset, spa->spa_feat_for_read_obj); zap_cursor_retrieve(&zc, &za) == 0; zap_cursor_advance(&zc)) { ASSERT(za.za_integer_length == sizeof (uint64_t) && za.za_num_integers == 1); VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, za.za_first_integer)); } zap_cursor_fini(&zc); } if (spa->spa_feat_for_write_obj != 0) { for (zap_cursor_init(&zc, spa->spa_meta_objset, spa->spa_feat_for_write_obj); zap_cursor_retrieve(&zc, &za) == 0; zap_cursor_advance(&zc)) { ASSERT(za.za_integer_length == sizeof (uint64_t) && za.za_num_integers == 1); VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, za.za_first_integer)); } zap_cursor_fini(&zc); } out: VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS, features) == 0); nvlist_free(features); } int spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen) { int error; spa_t *spa; *config = NULL; error = spa_open_common(name, &spa, FTAG, NULL, config); if (spa != NULL) { /* * This still leaves a window of inconsistency where the spares * or l2cache devices could change and the config would be * self-inconsistent. */ spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); if (*config != NULL) { uint64_t loadtimes[2]; loadtimes[0] = spa->spa_loaded_ts.tv_sec; loadtimes[1] = spa->spa_loaded_ts.tv_nsec; VERIFY(nvlist_add_uint64_array(*config, ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0); VERIFY(nvlist_add_uint64(*config, ZPOOL_CONFIG_ERRCOUNT, spa_get_errlog_size(spa)) == 0); if (spa_suspended(spa)) VERIFY(nvlist_add_uint64(*config, ZPOOL_CONFIG_SUSPENDED, spa->spa_failmode) == 0); spa_add_spares(spa, *config); spa_add_l2cache(spa, *config); spa_add_feature_stats(spa, *config); } } /* * We want to get the alternate root even for faulted pools, so we cheat * and call spa_lookup() directly. */ if (altroot) { if (spa == NULL) { mutex_enter(&spa_namespace_lock); spa = spa_lookup(name); if (spa) spa_altroot(spa, altroot, buflen); else altroot[0] = '\0'; spa = NULL; mutex_exit(&spa_namespace_lock); } else { spa_altroot(spa, altroot, buflen); } } if (spa != NULL) { spa_config_exit(spa, SCL_CONFIG, FTAG); spa_close(spa, FTAG); } return (error); } /* * Validate that the auxiliary device array is well formed. We must have an * array of nvlists, each which describes a valid leaf vdev. If this is an * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be * specified, as long as they are well-formed. */ static int spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, spa_aux_vdev_t *sav, const char *config, uint64_t version, vdev_labeltype_t label) { nvlist_t **dev; uint_t i, ndev; vdev_t *vd; int error; ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); /* * It's acceptable to have no devs specified. */ if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) return (0); if (ndev == 0) return (SET_ERROR(EINVAL)); /* * Make sure the pool is formatted with a version that supports this * device type. */ if (spa_version(spa) < version) return (SET_ERROR(ENOTSUP)); /* * Set the pending device list so we correctly handle device in-use * checking. */ sav->sav_pending = dev; sav->sav_npending = ndev; for (i = 0; i < ndev; i++) { if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, mode)) != 0) goto out; if (!vd->vdev_ops->vdev_op_leaf) { vdev_free(vd); error = SET_ERROR(EINVAL); goto out; } /* * The L2ARC currently only supports disk devices in * kernel context. For user-level testing, we allow it. */ #ifdef _KERNEL if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) && strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) { error = SET_ERROR(ENOTBLK); vdev_free(vd); goto out; } #endif vd->vdev_top = vd; if ((error = vdev_open(vd)) == 0 && (error = vdev_label_init(vd, crtxg, label)) == 0) { VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, vd->vdev_guid) == 0); } vdev_free(vd); if (error && (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) goto out; else error = 0; } out: sav->sav_pending = NULL; sav->sav_npending = 0; return (error); } static int spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) { int error; ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, VDEV_LABEL_SPARE)) != 0) { return (error); } return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, VDEV_LABEL_L2CACHE)); } static void spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, const char *config) { int i; if (sav->sav_config != NULL) { nvlist_t **olddevs; uint_t oldndevs; nvlist_t **newdevs; /* * Generate new dev list by concatentating with the * current dev list. */ VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config, &olddevs, &oldndevs) == 0); newdevs = kmem_alloc(sizeof (void *) * (ndevs + oldndevs), KM_SLEEP); for (i = 0; i < oldndevs; i++) VERIFY(nvlist_dup(olddevs[i], &newdevs[i], KM_SLEEP) == 0); for (i = 0; i < ndevs; i++) VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs], KM_SLEEP) == 0); VERIFY(nvlist_remove(sav->sav_config, config, DATA_TYPE_NVLIST_ARRAY) == 0); VERIFY(nvlist_add_nvlist_array(sav->sav_config, config, newdevs, ndevs + oldndevs) == 0); for (i = 0; i < oldndevs + ndevs; i++) nvlist_free(newdevs[i]); kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); } else { /* * Generate a new dev list. */ VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME, KM_SLEEP) == 0); VERIFY(nvlist_add_nvlist_array(sav->sav_config, config, devs, ndevs) == 0); } } /* * Stop and drop level 2 ARC devices */ void spa_l2cache_drop(spa_t *spa) { vdev_t *vd; int i; spa_aux_vdev_t *sav = &spa->spa_l2cache; for (i = 0; i < sav->sav_count; i++) { uint64_t pool; vd = sav->sav_vdevs[i]; ASSERT(vd != NULL); if (spa_l2cache_exists(vd->vdev_guid, &pool) && pool != 0ULL && l2arc_vdev_present(vd)) l2arc_remove_vdev(vd); } } /* * Pool Creation */ int spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, nvlist_t *zplprops) { spa_t *spa; char *altroot = NULL; vdev_t *rvd; dsl_pool_t *dp; dmu_tx_t *tx; int error = 0; uint64_t txg = TXG_INITIAL; nvlist_t **spares, **l2cache; uint_t nspares, nl2cache; uint64_t version, obj; boolean_t has_features; /* * If this pool already exists, return failure. */ mutex_enter(&spa_namespace_lock); if (spa_lookup(pool) != NULL) { mutex_exit(&spa_namespace_lock); return (SET_ERROR(EEXIST)); } /* * Allocate a new spa_t structure. */ (void) nvlist_lookup_string(props, zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); spa = spa_add(pool, NULL, altroot); spa_activate(spa, spa_mode_global); if (props && (error = spa_prop_validate(spa, props))) { spa_deactivate(spa); spa_remove(spa); mutex_exit(&spa_namespace_lock); return (error); } has_features = B_FALSE; for (nvpair_t *elem = nvlist_next_nvpair(props, NULL); elem != NULL; elem = nvlist_next_nvpair(props, elem)) { if (zpool_prop_feature(nvpair_name(elem))) has_features = B_TRUE; } if (has_features || nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) { version = SPA_VERSION; } ASSERT(SPA_VERSION_IS_SUPPORTED(version)); spa->spa_first_txg = txg; spa->spa_uberblock.ub_txg = txg - 1; spa->spa_uberblock.ub_version = version; spa->spa_ubsync = spa->spa_uberblock; /* * Create "The Godfather" zio to hold all async IOs */ spa->spa_async_zio_root = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER); /* * Create the root vdev. */ spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); ASSERT(error != 0 || rvd != NULL); ASSERT(error != 0 || spa->spa_root_vdev == rvd); if (error == 0 && !zfs_allocatable_devs(nvroot)) error = SET_ERROR(EINVAL); if (error == 0 && (error = vdev_create(rvd, txg, B_FALSE)) == 0 && (error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) == 0) { for (int c = 0; c < rvd->vdev_children; c++) { vdev_ashift_optimize(rvd->vdev_child[c]); vdev_metaslab_set_size(rvd->vdev_child[c]); vdev_expand(rvd->vdev_child[c], txg); } } spa_config_exit(spa, SCL_ALL, FTAG); if (error != 0) { spa_unload(spa); spa_deactivate(spa); spa_remove(spa); mutex_exit(&spa_namespace_lock); return (error); } /* * Get the list of spares, if specified. */ if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0) { VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME, KM_SLEEP) == 0); VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, spares, nspares) == 0); spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); spa_load_spares(spa); spa_config_exit(spa, SCL_ALL, FTAG); spa->spa_spares.sav_sync = B_TRUE; } /* * Get the list of level 2 cache devices, if specified. */ if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0) { VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, NV_UNIQUE_NAME, KM_SLEEP) == 0); VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); spa_load_l2cache(spa); spa_config_exit(spa, SCL_ALL, FTAG); spa->spa_l2cache.sav_sync = B_TRUE; } spa->spa_is_initializing = B_TRUE; spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg); spa->spa_meta_objset = dp->dp_meta_objset; spa->spa_is_initializing = B_FALSE; /* * Create DDTs (dedup tables). */ ddt_create(spa); spa_update_dspace(spa); tx = dmu_tx_create_assigned(dp, txg); /* * Create the pool config object. */ spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); if (zap_add(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { cmn_err(CE_PANIC, "failed to add pool config"); } if (spa_version(spa) >= SPA_VERSION_FEATURES) spa_feature_create_zap_objects(spa, tx); if (zap_add(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, sizeof (uint64_t), 1, &version, tx) != 0) { cmn_err(CE_PANIC, "failed to add pool version"); } /* Newly created pools with the right version are always deflated. */ if (version >= SPA_VERSION_RAIDZ_DEFLATE) { spa->spa_deflate = TRUE; if (zap_add(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { cmn_err(CE_PANIC, "failed to add deflate"); } } /* * Create the deferred-free bpobj. Turn off compression * because sync-to-convergence takes longer if the blocksize * keeps changing. */ obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx); dmu_object_set_compress(spa->spa_meta_objset, obj, ZIO_COMPRESS_OFF, tx); if (zap_add(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ, sizeof (uint64_t), 1, &obj, tx) != 0) { cmn_err(CE_PANIC, "failed to add bpobj"); } VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj)); /* * Create the pool's history object. */ if (version >= SPA_VERSION_ZPOOL_HISTORY) spa_history_create_obj(spa, tx); /* * Set pool properties. */ spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND); if (props != NULL) { spa_configfile_set(spa, props, B_FALSE); spa_sync_props(props, tx); } dmu_tx_commit(tx); spa->spa_sync_on = B_TRUE; txg_sync_start(spa->spa_dsl_pool); /* * We explicitly wait for the first transaction to complete so that our * bean counters are appropriately updated. */ txg_wait_synced(spa->spa_dsl_pool, txg); spa_config_sync(spa, B_FALSE, B_TRUE); spa_history_log_version(spa, "create"); spa->spa_minref = refcount_count(&spa->spa_refcount); mutex_exit(&spa_namespace_lock); return (0); } #ifdef _KERNEL #if defined(sun) /* * Get the root pool information from the root disk, then import the root pool * during the system boot up time. */ extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **); static nvlist_t * spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid) { nvlist_t *config; nvlist_t *nvtop, *nvroot; uint64_t pgid; if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0) return (NULL); /* * Add this top-level vdev to the child array. */ VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0); VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pgid) == 0); VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0); /* * Put this pool's top-level vdevs into a root vdev. */ VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0); VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, &nvtop, 1) == 0); /* * Replace the existing vdev_tree with the new root vdev in * this pool's configuration (remove the old, add the new). */ VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); nvlist_free(nvroot); return (config); } /* * Walk the vdev tree and see if we can find a device with "better" * configuration. A configuration is "better" if the label on that * device has a more recent txg. */ static void spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg) { for (int c = 0; c < vd->vdev_children; c++) spa_alt_rootvdev(vd->vdev_child[c], avd, txg); if (vd->vdev_ops->vdev_op_leaf) { nvlist_t *label; uint64_t label_txg; if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid, &label) != 0) return; VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, &label_txg) == 0); /* * Do we have a better boot device? */ if (label_txg > *txg) { *txg = label_txg; *avd = vd; } nvlist_free(label); } } /* * Import a root pool. * * For x86. devpath_list will consist of devid and/or physpath name of * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a"). * The GRUB "findroot" command will return the vdev we should boot. * * For Sparc, devpath_list consists the physpath name of the booting device * no matter the rootpool is a single device pool or a mirrored pool. * e.g. * "/pci@1f,0/ide@d/disk@0,0:a" */ int spa_import_rootpool(char *devpath, char *devid) { spa_t *spa; vdev_t *rvd, *bvd, *avd = NULL; nvlist_t *config, *nvtop; uint64_t guid, txg; char *pname; int error; /* * Read the label from the boot device and generate a configuration. */ config = spa_generate_rootconf(devpath, devid, &guid); #if defined(_OBP) && defined(_KERNEL) if (config == NULL) { if (strstr(devpath, "/iscsi/ssd") != NULL) { /* iscsi boot */ get_iscsi_bootpath_phy(devpath); config = spa_generate_rootconf(devpath, devid, &guid); } } #endif if (config == NULL) { cmn_err(CE_NOTE, "Cannot read the pool label from '%s'", devpath); return (SET_ERROR(EIO)); } VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, &pname) == 0); VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0); mutex_enter(&spa_namespace_lock); if ((spa = spa_lookup(pname)) != NULL) { /* * Remove the existing root pool from the namespace so that we * can replace it with the correct config we just read in. */ spa_remove(spa); } spa = spa_add(pname, config, NULL); spa->spa_is_root = B_TRUE; spa->spa_import_flags = ZFS_IMPORT_VERBATIM; /* * Build up a vdev tree based on the boot device's label config. */ VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0); spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, VDEV_ALLOC_ROOTPOOL); spa_config_exit(spa, SCL_ALL, FTAG); if (error) { mutex_exit(&spa_namespace_lock); nvlist_free(config); cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", pname); return (error); } /* * Get the boot vdev. */ if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) { cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu", (u_longlong_t)guid); error = SET_ERROR(ENOENT); goto out; } /* * Determine if there is a better boot device. */ avd = bvd; spa_alt_rootvdev(rvd, &avd, &txg); if (avd != bvd) { cmn_err(CE_NOTE, "The boot device is 'degraded'. Please " "try booting from '%s'", avd->vdev_path); error = SET_ERROR(EINVAL); goto out; } /* * If the boot device is part of a spare vdev then ensure that * we're booting off the active spare. */ if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops && !bvd->vdev_isspare) { cmn_err(CE_NOTE, "The boot device is currently spared. Please " "try booting from '%s'", bvd->vdev_parent-> vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path); error = SET_ERROR(EINVAL); goto out; } error = 0; out: spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); vdev_free(rvd); spa_config_exit(spa, SCL_ALL, FTAG); mutex_exit(&spa_namespace_lock); nvlist_free(config); return (error); } #else extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs, uint64_t *count); static nvlist_t * spa_generate_rootconf(const char *name) { nvlist_t **configs, **tops; nvlist_t *config; nvlist_t *best_cfg, *nvtop, *nvroot; uint64_t *holes; uint64_t best_txg; uint64_t nchildren; uint64_t pgid; uint64_t count; uint64_t i; uint_t nholes; if (vdev_geom_read_pool_label(name, &configs, &count) != 0) return (NULL); ASSERT3U(count, !=, 0); best_txg = 0; for (i = 0; i < count; i++) { uint64_t txg; VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG, &txg) == 0); if (txg > best_txg) { best_txg = txg; best_cfg = configs[i]; } } /* * Multi-vdev root pool configuration discovery is not supported yet. */ nchildren = 1; nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren); holes = NULL; nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY, &holes, &nholes); tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP); for (i = 0; i < nchildren; i++) { if (i >= count) break; if (configs[i] == NULL) continue; VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0); nvlist_dup(nvtop, &tops[i], KM_SLEEP); } for (i = 0; holes != NULL && i < nholes; i++) { if (i >= nchildren) continue; if (tops[holes[i]] != NULL) continue; nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP); VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0); VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID, holes[i]) == 0); VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID, 0) == 0); } for (i = 0; i < nchildren; i++) { if (tops[i] != NULL) continue; nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP); VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE, VDEV_TYPE_MISSING) == 0); VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID, i) == 0); VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID, 0) == 0); } /* * Create pool config based on the best vdev config. */ nvlist_dup(best_cfg, &config, KM_SLEEP); /* * Put this pool's top-level vdevs into a root vdev. */ VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pgid) == 0); VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0); VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, tops, nchildren) == 0); /* * Replace the existing vdev_tree with the new root vdev in * this pool's configuration (remove the old, add the new). */ VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); /* * Drop vdev config elements that should not be present at pool level. */ nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64); nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64); for (i = 0; i < count; i++) nvlist_free(configs[i]); kmem_free(configs, count * sizeof(void *)); for (i = 0; i < nchildren; i++) nvlist_free(tops[i]); kmem_free(tops, nchildren * sizeof(void *)); nvlist_free(nvroot); return (config); } int spa_import_rootpool(const char *name) { spa_t *spa; vdev_t *rvd, *bvd, *avd = NULL; nvlist_t *config, *nvtop; uint64_t txg; char *pname; int error; /* * Read the label from the boot device and generate a configuration. */ config = spa_generate_rootconf(name); mutex_enter(&spa_namespace_lock); if (config != NULL) { VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, &pname) == 0 && strcmp(name, pname) == 0); VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0); if ((spa = spa_lookup(pname)) != NULL) { /* * Remove the existing root pool from the namespace so * that we can replace it with the correct config * we just read in. */ spa_remove(spa); } spa = spa_add(pname, config, NULL); /* * Set spa_ubsync.ub_version as it can be used in vdev_alloc() * via spa_version(). */ if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &spa->spa_ubsync.ub_version) != 0) spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; } else if ((spa = spa_lookup(name)) == NULL) { cmn_err(CE_NOTE, "Cannot find the pool label for '%s'", name); return (EIO); } else { VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0); } spa->spa_is_root = B_TRUE; spa->spa_import_flags = ZFS_IMPORT_VERBATIM; /* * Build up a vdev tree based on the boot device's label config. */ VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0); spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, VDEV_ALLOC_ROOTPOOL); spa_config_exit(spa, SCL_ALL, FTAG); if (error) { mutex_exit(&spa_namespace_lock); nvlist_free(config); cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", pname); return (error); } spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); vdev_free(rvd); spa_config_exit(spa, SCL_ALL, FTAG); mutex_exit(&spa_namespace_lock); nvlist_free(config); return (0); } #endif /* sun */ #endif /* * Import a non-root pool into the system. */ int spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags) { spa_t *spa; char *altroot = NULL; spa_load_state_t state = SPA_LOAD_IMPORT; zpool_rewind_policy_t policy; uint64_t mode = spa_mode_global; uint64_t readonly = B_FALSE; int error; nvlist_t *nvroot; nvlist_t **spares, **l2cache; uint_t nspares, nl2cache; /* * If a pool with this name exists, return failure. */ mutex_enter(&spa_namespace_lock); if (spa_lookup(pool) != NULL) { mutex_exit(&spa_namespace_lock); return (SET_ERROR(EEXIST)); } /* * Create and initialize the spa structure. */ (void) nvlist_lookup_string(props, zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); (void) nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly); if (readonly) mode = FREAD; spa = spa_add(pool, config, altroot); spa->spa_import_flags = flags; /* * Verbatim import - Take a pool and insert it into the namespace * as if it had been loaded at boot. */ if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) { if (props != NULL) spa_configfile_set(spa, props, B_FALSE); spa_config_sync(spa, B_FALSE, B_TRUE); mutex_exit(&spa_namespace_lock); return (0); } spa_activate(spa, mode); /* * Don't start async tasks until we know everything is healthy. */ spa_async_suspend(spa); zpool_get_rewind_policy(config, &policy); if (policy.zrp_request & ZPOOL_DO_REWIND) state = SPA_LOAD_RECOVER; /* * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig * because the user-supplied config is actually the one to trust when * doing an import. */ if (state != SPA_LOAD_RECOVER) spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg, policy.zrp_request); /* * Propagate anything learned while loading the pool and pass it * back to caller (i.e. rewind info, missing devices, etc). */ VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, spa->spa_load_info) == 0); spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); /* * Toss any existing sparelist, as it doesn't have any validity * anymore, and conflicts with spa_has_spare(). */ if (spa->spa_spares.sav_config) { nvlist_free(spa->spa_spares.sav_config); spa->spa_spares.sav_config = NULL; spa_load_spares(spa); } if (spa->spa_l2cache.sav_config) { nvlist_free(spa->spa_l2cache.sav_config); spa->spa_l2cache.sav_config = NULL; spa_load_l2cache(spa); } VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); if (error == 0) error = spa_validate_aux(spa, nvroot, -1ULL, VDEV_ALLOC_SPARE); if (error == 0) error = spa_validate_aux(spa, nvroot, -1ULL, VDEV_ALLOC_L2CACHE); spa_config_exit(spa, SCL_ALL, FTAG); if (props != NULL) spa_configfile_set(spa, props, B_FALSE); if (error != 0 || (props && spa_writeable(spa) && (error = spa_prop_set(spa, props)))) { spa_unload(spa); spa_deactivate(spa); spa_remove(spa); mutex_exit(&spa_namespace_lock); return (error); } spa_async_resume(spa); /* * Override any spares and level 2 cache devices as specified by * the user, as these may have correct device names/devids, etc. */ if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0) { if (spa->spa_spares.sav_config) VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); else VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME, KM_SLEEP) == 0); VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, spares, nspares) == 0); spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); spa_load_spares(spa); spa_config_exit(spa, SCL_ALL, FTAG); spa->spa_spares.sav_sync = B_TRUE; } if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0) { if (spa->spa_l2cache.sav_config) VERIFY(nvlist_remove(spa->spa_l2cache.sav_config, ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0); else VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, NV_UNIQUE_NAME, KM_SLEEP) == 0); VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); spa_load_l2cache(spa); spa_config_exit(spa, SCL_ALL, FTAG); spa->spa_l2cache.sav_sync = B_TRUE; } /* * Check for any removed devices. */ if (spa->spa_autoreplace) { spa_aux_check_removed(&spa->spa_spares); spa_aux_check_removed(&spa->spa_l2cache); } if (spa_writeable(spa)) { /* * Update the config cache to include the newly-imported pool. */ spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); } /* * It's possible that the pool was expanded while it was exported. * We kick off an async task to handle this for us. */ spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); mutex_exit(&spa_namespace_lock); spa_history_log_version(spa, "import"); #ifdef __FreeBSD__ #ifdef _KERNEL zvol_create_minors(pool); #endif #endif return (0); } nvlist_t * spa_tryimport(nvlist_t *tryconfig) { nvlist_t *config = NULL; char *poolname; spa_t *spa; uint64_t state; int error; if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) return (NULL); if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) return (NULL); /* * Create and initialize the spa structure. */ mutex_enter(&spa_namespace_lock); spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL); spa_activate(spa, FREAD); /* * Pass off the heavy lifting to spa_load(). * Pass TRUE for mosconfig because the user-supplied config * is actually the one to trust when doing an import. */ error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE); /* * If 'tryconfig' was at least parsable, return the current config. */ if (spa->spa_root_vdev != NULL) { config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, poolname) == 0); VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, state) == 0); VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, spa->spa_uberblock.ub_timestamp) == 0); VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, spa->spa_load_info) == 0); /* * If the bootfs property exists on this pool then we * copy it out so that external consumers can tell which * pools are bootable. */ if ((!error || error == EEXIST) && spa->spa_bootfs) { char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); /* * We have to play games with the name since the * pool was opened as TRYIMPORT_NAME. */ if (dsl_dsobj_to_dsname(spa_name(spa), spa->spa_bootfs, tmpname) == 0) { char *cp; char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); cp = strchr(tmpname, '/'); if (cp == NULL) { (void) strlcpy(dsname, tmpname, MAXPATHLEN); } else { (void) snprintf(dsname, MAXPATHLEN, "%s/%s", poolname, ++cp); } VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_BOOTFS, dsname) == 0); kmem_free(dsname, MAXPATHLEN); } kmem_free(tmpname, MAXPATHLEN); } /* * Add the list of hot spares and level 2 cache devices. */ spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); spa_add_spares(spa, config); spa_add_l2cache(spa, config); spa_config_exit(spa, SCL_CONFIG, FTAG); } spa_unload(spa); spa_deactivate(spa); spa_remove(spa); mutex_exit(&spa_namespace_lock); return (config); } /* * Pool export/destroy * * The act of destroying or exporting a pool is very simple. We make sure there * is no more pending I/O and any references to the pool are gone. Then, we * update the pool state and sync all the labels to disk, removing the * configuration from the cache afterwards. If the 'hardforce' flag is set, then * we don't sync the labels or remove the configuration cache. */ static int spa_export_common(char *pool, int new_state, nvlist_t **oldconfig, boolean_t force, boolean_t hardforce) { spa_t *spa; if (oldconfig) *oldconfig = NULL; if (!(spa_mode_global & FWRITE)) return (SET_ERROR(EROFS)); mutex_enter(&spa_namespace_lock); if ((spa = spa_lookup(pool)) == NULL) { mutex_exit(&spa_namespace_lock); return (SET_ERROR(ENOENT)); } /* * Put a hold on the pool, drop the namespace lock, stop async tasks, * reacquire the namespace lock, and see if we can export. */ spa_open_ref(spa, FTAG); mutex_exit(&spa_namespace_lock); spa_async_suspend(spa); mutex_enter(&spa_namespace_lock); spa_close(spa, FTAG); /* * The pool will be in core if it's openable, * in which case we can modify its state. */ if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) { /* * Objsets may be open only because they're dirty, so we * have to force it to sync before checking spa_refcnt. */ txg_wait_synced(spa->spa_dsl_pool, 0); /* * A pool cannot be exported or destroyed if there are active * references. If we are resetting a pool, allow references by * fault injection handlers. */ if (!spa_refcount_zero(spa) || (spa->spa_inject_ref != 0 && new_state != POOL_STATE_UNINITIALIZED)) { spa_async_resume(spa); mutex_exit(&spa_namespace_lock); return (SET_ERROR(EBUSY)); } /* * A pool cannot be exported if it has an active shared spare. * This is to prevent other pools stealing the active spare * from an exported pool. At user's own will, such pool can * be forcedly exported. */ if (!force && new_state == POOL_STATE_EXPORTED && spa_has_active_shared_spare(spa)) { spa_async_resume(spa); mutex_exit(&spa_namespace_lock); return (SET_ERROR(EXDEV)); } /* * We want this to be reflected on every label, * so mark them all dirty. spa_unload() will do the * final sync that pushes these changes out. */ if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); spa->spa_state = new_state; spa->spa_final_txg = spa_last_synced_txg(spa) + TXG_DEFER_SIZE + 1; vdev_config_dirty(spa->spa_root_vdev); spa_config_exit(spa, SCL_ALL, FTAG); } } spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY); if (spa->spa_state != POOL_STATE_UNINITIALIZED) { spa_unload(spa); spa_deactivate(spa); } if (oldconfig && spa->spa_config) VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0); if (new_state != POOL_STATE_UNINITIALIZED) { if (!hardforce) spa_config_sync(spa, B_TRUE, B_TRUE); spa_remove(spa); } mutex_exit(&spa_namespace_lock); return (0); } /* * Destroy a storage pool. */ int spa_destroy(char *pool) { return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, B_FALSE, B_FALSE)); } /* * Export a storage pool. */ int spa_export(char *pool, nvlist_t **oldconfig, boolean_t force, boolean_t hardforce) { return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, force, hardforce)); } /* * Similar to spa_export(), this unloads the spa_t without actually removing it * from the namespace in any way. */ int spa_reset(char *pool) { return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, B_FALSE, B_FALSE)); } /* * ========================================================================== * Device manipulation * ========================================================================== */ /* * Add a device to a storage pool. */ int spa_vdev_add(spa_t *spa, nvlist_t *nvroot) { uint64_t txg, id; int error; vdev_t *rvd = spa->spa_root_vdev; vdev_t *vd, *tvd; nvlist_t **spares, **l2cache; uint_t nspares, nl2cache; ASSERT(spa_writeable(spa)); txg = spa_vdev_enter(spa); if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, VDEV_ALLOC_ADD)) != 0) return (spa_vdev_exit(spa, NULL, txg, error)); spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, &nspares) != 0) nspares = 0; if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) != 0) nl2cache = 0; if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) return (spa_vdev_exit(spa, vd, txg, EINVAL)); if (vd->vdev_children != 0 && (error = vdev_create(vd, txg, B_FALSE)) != 0) return (spa_vdev_exit(spa, vd, txg, error)); /* * We must validate the spares and l2cache devices after checking the * children. Otherwise, vdev_inuse() will blindly overwrite the spare. */ if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) return (spa_vdev_exit(spa, vd, txg, error)); /* * Transfer each new top-level vdev from vd to rvd. */ for (int c = 0; c < vd->vdev_children; c++) { /* * Set the vdev id to the first hole, if one exists. */ for (id = 0; id < rvd->vdev_children; id++) { if (rvd->vdev_child[id]->vdev_ishole) { vdev_free(rvd->vdev_child[id]); break; } } tvd = vd->vdev_child[c]; vdev_remove_child(vd, tvd); tvd->vdev_id = id; vdev_add_child(rvd, tvd); vdev_config_dirty(tvd); } if (nspares != 0) { spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, ZPOOL_CONFIG_SPARES); spa_load_spares(spa); spa->spa_spares.sav_sync = B_TRUE; } if (nl2cache != 0) { spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, ZPOOL_CONFIG_L2CACHE); spa_load_l2cache(spa); spa->spa_l2cache.sav_sync = B_TRUE; } /* * We have to be careful when adding new vdevs to an existing pool. * If other threads start allocating from these vdevs before we * sync the config cache, and we lose power, then upon reboot we may * fail to open the pool because there are DVAs that the config cache * can't translate. Therefore, we first add the vdevs without * initializing metaslabs; sync the config cache (via spa_vdev_exit()); * and then let spa_config_update() initialize the new metaslabs. * * spa_load() checks for added-but-not-initialized vdevs, so that * if we lose power at any point in this sequence, the remaining * steps will be completed the next time we load the pool. */ (void) spa_vdev_exit(spa, vd, txg, 0); mutex_enter(&spa_namespace_lock); spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); mutex_exit(&spa_namespace_lock); return (0); } /* * Attach a device to a mirror. The arguments are the path to any device * in the mirror, and the nvroot for the new device. If the path specifies * a device that is not mirrored, we automatically insert the mirror vdev. * * If 'replacing' is specified, the new device is intended to replace the * existing device; in this case the two devices are made into their own * mirror using the 'replacing' vdev, which is functionally identical to * the mirror vdev (it actually reuses all the same ops) but has a few * extra rules: you can't attach to it after it's been created, and upon * completion of resilvering, the first disk (the one being replaced) * is automatically detached. */ int spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) { uint64_t txg, dtl_max_txg; vdev_t *rvd = spa->spa_root_vdev; vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; vdev_ops_t *pvops; char *oldvdpath, *newvdpath; int newvd_isspare; int error; ASSERT(spa_writeable(spa)); txg = spa_vdev_enter(spa); oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); if (oldvd == NULL) return (spa_vdev_exit(spa, NULL, txg, ENODEV)); if (!oldvd->vdev_ops->vdev_op_leaf) return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); pvd = oldvd->vdev_parent; if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, VDEV_ALLOC_ATTACH)) != 0) return (spa_vdev_exit(spa, NULL, txg, EINVAL)); if (newrootvd->vdev_children != 1) return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); newvd = newrootvd->vdev_child[0]; if (!newvd->vdev_ops->vdev_op_leaf) return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); if ((error = vdev_create(newrootvd, txg, replacing)) != 0) return (spa_vdev_exit(spa, newrootvd, txg, error)); /* * Spares can't replace logs */ if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare) return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); if (!replacing) { /* * For attach, the only allowable parent is a mirror or the root * vdev. */ if (pvd->vdev_ops != &vdev_mirror_ops && pvd->vdev_ops != &vdev_root_ops) return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); pvops = &vdev_mirror_ops; } else { /* * Active hot spares can only be replaced by inactive hot * spares. */ if (pvd->vdev_ops == &vdev_spare_ops && oldvd->vdev_isspare && !spa_has_spare(spa, newvd->vdev_guid)) return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); /* * If the source is a hot spare, and the parent isn't already a * spare, then we want to create a new hot spare. Otherwise, we * want to create a replacing vdev. The user is not allowed to * attach to a spared vdev child unless the 'isspare' state is * the same (spare replaces spare, non-spare replaces * non-spare). */ if (pvd->vdev_ops == &vdev_replacing_ops && spa_version(spa) < SPA_VERSION_MULTI_REPLACE) { return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); } else if (pvd->vdev_ops == &vdev_spare_ops && newvd->vdev_isspare != oldvd->vdev_isspare) { return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); } if (newvd->vdev_isspare) pvops = &vdev_spare_ops; else pvops = &vdev_replacing_ops; } /* * Make sure the new device is big enough. */ if (newvd->vdev_asize < vdev_get_min_asize(oldvd)) return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); /* * The new device cannot have a higher alignment requirement * than the top-level vdev. */ if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); /* * If this is an in-place replacement, update oldvd's path and devid * to make it distinguishable from newvd, and unopenable from now on. */ if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { spa_strfree(oldvd->vdev_path); oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, KM_SLEEP); (void) sprintf(oldvd->vdev_path, "%s/%s", newvd->vdev_path, "old"); if (oldvd->vdev_devid != NULL) { spa_strfree(oldvd->vdev_devid); oldvd->vdev_devid = NULL; } } /* mark the device being resilvered */ newvd->vdev_resilver_txg = txg; /* * If the parent is not a mirror, or if we're replacing, insert the new * mirror/replacing/spare vdev above oldvd. */ if (pvd->vdev_ops != pvops) pvd = vdev_add_parent(oldvd, pvops); ASSERT(pvd->vdev_top->vdev_parent == rvd); ASSERT(pvd->vdev_ops == pvops); ASSERT(oldvd->vdev_parent == pvd); /* * Extract the new device from its root and add it to pvd. */ vdev_remove_child(newrootvd, newvd); newvd->vdev_id = pvd->vdev_children; newvd->vdev_crtxg = oldvd->vdev_crtxg; vdev_add_child(pvd, newvd); tvd = newvd->vdev_top; ASSERT(pvd->vdev_top == tvd); ASSERT(tvd->vdev_parent == rvd); vdev_config_dirty(tvd); /* * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account * for any dmu_sync-ed blocks. It will propagate upward when * spa_vdev_exit() calls vdev_dtl_reassess(). */ dtl_max_txg = txg + TXG_CONCURRENT_STATES; vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL, dtl_max_txg - TXG_INITIAL); if (newvd->vdev_isspare) { spa_spare_activate(newvd); spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE); } oldvdpath = spa_strdup(oldvd->vdev_path); newvdpath = spa_strdup(newvd->vdev_path); newvd_isspare = newvd->vdev_isspare; /* * Mark newvd's DTL dirty in this txg. */ vdev_dirty(tvd, VDD_DTL, newvd, txg); /* * Schedule the resilver to restart in the future. We do this to * ensure that dmu_sync-ed blocks have been stitched into the * respective datasets. */ dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg); /* * Commit the config */ (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0); spa_history_log_internal(spa, "vdev attach", NULL, "%s vdev=%s %s vdev=%s", replacing && newvd_isspare ? "spare in" : replacing ? "replace" : "attach", newvdpath, replacing ? "for" : "to", oldvdpath); spa_strfree(oldvdpath); spa_strfree(newvdpath); if (spa->spa_bootfs) spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH); return (0); } /* * Detach a device from a mirror or replacing vdev. * * If 'replace_done' is specified, only detach if the parent * is a replacing vdev. */ int spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done) { uint64_t txg; int error; vdev_t *rvd = spa->spa_root_vdev; vdev_t *vd, *pvd, *cvd, *tvd; boolean_t unspare = B_FALSE; uint64_t unspare_guid = 0; char *vdpath; ASSERT(spa_writeable(spa)); txg = spa_vdev_enter(spa); vd = spa_lookup_by_guid(spa, guid, B_FALSE); if (vd == NULL) return (spa_vdev_exit(spa, NULL, txg, ENODEV)); if (!vd->vdev_ops->vdev_op_leaf) return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); pvd = vd->vdev_parent; /* * If the parent/child relationship is not as expected, don't do it. * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing * vdev that's replacing B with C. The user's intent in replacing * is to go from M(A,B) to M(A,C). If the user decides to cancel * the replace by detaching C, the expected behavior is to end up * M(A,B). But suppose that right after deciding to detach C, * the replacement of B completes. We would have M(A,C), and then * ask to detach C, which would leave us with just A -- not what * the user wanted. To prevent this, we make sure that the * parent/child relationship hasn't changed -- in this example, * that C's parent is still the replacing vdev R. */ if (pvd->vdev_guid != pguid && pguid != 0) return (spa_vdev_exit(spa, NULL, txg, EBUSY)); /* * Only 'replacing' or 'spare' vdevs can be replaced. */ if (replace_done && pvd->vdev_ops != &vdev_replacing_ops && pvd->vdev_ops != &vdev_spare_ops) return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); ASSERT(pvd->vdev_ops != &vdev_spare_ops || spa_version(spa) >= SPA_VERSION_SPARES); /* * Only mirror, replacing, and spare vdevs support detach. */ if (pvd->vdev_ops != &vdev_replacing_ops && pvd->vdev_ops != &vdev_mirror_ops && pvd->vdev_ops != &vdev_spare_ops) return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); /* * If this device has the only valid copy of some data, * we cannot safely detach it. */ if (vdev_dtl_required(vd)) return (spa_vdev_exit(spa, NULL, txg, EBUSY)); ASSERT(pvd->vdev_children >= 2); /* * If we are detaching the second disk from a replacing vdev, then * check to see if we changed the original vdev's path to have "/old" * at the end in spa_vdev_attach(). If so, undo that change now. */ if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 && vd->vdev_path != NULL) { size_t len = strlen(vd->vdev_path); for (int c = 0; c < pvd->vdev_children; c++) { cvd = pvd->vdev_child[c]; if (cvd == vd || cvd->vdev_path == NULL) continue; if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && strcmp(cvd->vdev_path + len, "/old") == 0) { spa_strfree(cvd->vdev_path); cvd->vdev_path = spa_strdup(vd->vdev_path); break; } } } /* * If we are detaching the original disk from a spare, then it implies * that the spare should become a real disk, and be removed from the * active spare list for the pool. */ if (pvd->vdev_ops == &vdev_spare_ops && vd->vdev_id == 0 && pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare) unspare = B_TRUE; /* * Erase the disk labels so the disk can be used for other things. * This must be done after all other error cases are handled, * but before we disembowel vd (so we can still do I/O to it). * But if we can't do it, don't treat the error as fatal -- * it may be that the unwritability of the disk is the reason * it's being detached! */ error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); /* * Remove vd from its parent and compact the parent's children. */ vdev_remove_child(pvd, vd); vdev_compact_children(pvd); /* * Remember one of the remaining children so we can get tvd below. */ cvd = pvd->vdev_child[pvd->vdev_children - 1]; /* * If we need to remove the remaining child from the list of hot spares, * do it now, marking the vdev as no longer a spare in the process. * We must do this before vdev_remove_parent(), because that can * change the GUID if it creates a new toplevel GUID. For a similar * reason, we must remove the spare now, in the same txg as the detach; * otherwise someone could attach a new sibling, change the GUID, and * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail. */ if (unspare) { ASSERT(cvd->vdev_isspare); spa_spare_remove(cvd); unspare_guid = cvd->vdev_guid; (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); cvd->vdev_unspare = B_TRUE; } /* * If the parent mirror/replacing vdev only has one child, * the parent is no longer needed. Remove it from the tree. */ if (pvd->vdev_children == 1) { if (pvd->vdev_ops == &vdev_spare_ops) cvd->vdev_unspare = B_FALSE; vdev_remove_parent(cvd); } /* * We don't set tvd until now because the parent we just removed * may have been the previous top-level vdev. */ tvd = cvd->vdev_top; ASSERT(tvd->vdev_parent == rvd); /* * Reevaluate the parent vdev state. */ vdev_propagate_state(cvd); /* * If the 'autoexpand' property is set on the pool then automatically * try to expand the size of the pool. For example if the device we * just detached was smaller than the others, it may be possible to * add metaslabs (i.e. grow the pool). We need to reopen the vdev * first so that we can obtain the updated sizes of the leaf vdevs. */ if (spa->spa_autoexpand) { vdev_reopen(tvd); vdev_expand(tvd, txg); } vdev_config_dirty(tvd); /* * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that * vd->vdev_detached is set and free vd's DTL object in syncing context. * But first make sure we're not on any *other* txg's DTL list, to * prevent vd from being accessed after it's freed. */ vdpath = spa_strdup(vd->vdev_path); for (int t = 0; t < TXG_SIZE; t++) (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); vd->vdev_detached = B_TRUE; vdev_dirty(tvd, VDD_DTL, vd, txg); spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE); /* hang on to the spa before we release the lock */ spa_open_ref(spa, FTAG); error = spa_vdev_exit(spa, vd, txg, 0); spa_history_log_internal(spa, "detach", NULL, "vdev=%s", vdpath); spa_strfree(vdpath); /* * If this was the removal of the original device in a hot spare vdev, * then we want to go through and remove the device from the hot spare * list of every other pool. */ if (unspare) { spa_t *altspa = NULL; mutex_enter(&spa_namespace_lock); while ((altspa = spa_next(altspa)) != NULL) { if (altspa->spa_state != POOL_STATE_ACTIVE || altspa == spa) continue; spa_open_ref(altspa, FTAG); mutex_exit(&spa_namespace_lock); (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE); mutex_enter(&spa_namespace_lock); spa_close(altspa, FTAG); } mutex_exit(&spa_namespace_lock); /* search the rest of the vdevs for spares to remove */ spa_vdev_resilver_done(spa); } /* all done with the spa; OK to release */ mutex_enter(&spa_namespace_lock); spa_close(spa, FTAG); mutex_exit(&spa_namespace_lock); return (error); } /* * Split a set of devices from their mirrors, and create a new pool from them. */ int spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config, nvlist_t *props, boolean_t exp) { int error = 0; uint64_t txg, *glist; spa_t *newspa; uint_t c, children, lastlog; nvlist_t **child, *nvl, *tmp; dmu_tx_t *tx; char *altroot = NULL; vdev_t *rvd, **vml = NULL; /* vdev modify list */ boolean_t activate_slog; ASSERT(spa_writeable(spa)); txg = spa_vdev_enter(spa); /* clear the log and flush everything up to now */ activate_slog = spa_passivate_log(spa); (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); error = spa_offline_log(spa); txg = spa_vdev_config_enter(spa); if (activate_slog) spa_activate_log(spa); if (error != 0) return (spa_vdev_exit(spa, NULL, txg, error)); /* check new spa name before going any further */ if (spa_lookup(newname) != NULL) return (spa_vdev_exit(spa, NULL, txg, EEXIST)); /* * scan through all the children to ensure they're all mirrors */ if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child, &children) != 0) return (spa_vdev_exit(spa, NULL, txg, EINVAL)); /* first, check to ensure we've got the right child count */ rvd = spa->spa_root_vdev; lastlog = 0; for (c = 0; c < rvd->vdev_children; c++) { vdev_t *vd = rvd->vdev_child[c]; /* don't count the holes & logs as children */ if (vd->vdev_islog || vd->vdev_ishole) { if (lastlog == 0) lastlog = c; continue; } lastlog = 0; } if (children != (lastlog != 0 ? lastlog : rvd->vdev_children)) return (spa_vdev_exit(spa, NULL, txg, EINVAL)); /* next, ensure no spare or cache devices are part of the split */ if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 || nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0) return (spa_vdev_exit(spa, NULL, txg, EINVAL)); vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP); glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP); /* then, loop over each vdev and validate it */ for (c = 0; c < children; c++) { uint64_t is_hole = 0; (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE, &is_hole); if (is_hole != 0) { if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole || spa->spa_root_vdev->vdev_child[c]->vdev_islog) { continue; } else { error = SET_ERROR(EINVAL); break; } } /* which disk is going to be split? */ if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID, &glist[c]) != 0) { error = SET_ERROR(EINVAL); break; } /* look it up in the spa */ vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE); if (vml[c] == NULL) { error = SET_ERROR(ENODEV); break; } /* make sure there's nothing stopping the split */ if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops || vml[c]->vdev_islog || vml[c]->vdev_ishole || vml[c]->vdev_isspare || vml[c]->vdev_isl2cache || !vdev_writeable(vml[c]) || vml[c]->vdev_children != 0 || vml[c]->vdev_state != VDEV_STATE_HEALTHY || c != spa->spa_root_vdev->vdev_child[c]->vdev_id) { error = SET_ERROR(EINVAL); break; } if (vdev_dtl_required(vml[c])) { error = SET_ERROR(EBUSY); break; } /* we need certain info from the top level */ VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY, vml[c]->vdev_top->vdev_ms_array) == 0); VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT, vml[c]->vdev_top->vdev_ms_shift) == 0); VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE, vml[c]->vdev_top->vdev_asize) == 0); VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT, vml[c]->vdev_top->vdev_ashift) == 0); } if (error != 0) { kmem_free(vml, children * sizeof (vdev_t *)); kmem_free(glist, children * sizeof (uint64_t)); return (spa_vdev_exit(spa, NULL, txg, error)); } /* stop writers from using the disks */ for (c = 0; c < children; c++) { if (vml[c] != NULL) vml[c]->vdev_offline = B_TRUE; } vdev_reopen(spa->spa_root_vdev); /* * Temporarily record the splitting vdevs in the spa config. This * will disappear once the config is regenerated. */ VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0); VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, glist, children) == 0); kmem_free(glist, children * sizeof (uint64_t)); mutex_enter(&spa->spa_props_lock); VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, nvl) == 0); mutex_exit(&spa->spa_props_lock); spa->spa_config_splitting = nvl; vdev_config_dirty(spa->spa_root_vdev); /* configure and create the new pool */ VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0); VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0); VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, spa_version(spa)) == 0); VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, spa->spa_config_txg) == 0); VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, spa_generate_guid(NULL)) == 0); (void) nvlist_lookup_string(props, zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); /* add the new pool to the namespace */ newspa = spa_add(newname, config, altroot); newspa->spa_config_txg = spa->spa_config_txg; spa_set_log_state(newspa, SPA_LOG_CLEAR); /* release the spa config lock, retaining the namespace lock */ spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); if (zio_injection_enabled) zio_handle_panic_injection(spa, FTAG, 1); spa_activate(newspa, spa_mode_global); spa_async_suspend(newspa); #ifndef sun /* mark that we are creating new spa by splitting */ newspa->spa_splitting_newspa = B_TRUE; #endif /* create the new pool from the disks of the original pool */ error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE); #ifndef sun newspa->spa_splitting_newspa = B_FALSE; #endif if (error) goto out; /* if that worked, generate a real config for the new pool */ if (newspa->spa_root_vdev != NULL) { VERIFY(nvlist_alloc(&newspa->spa_config_splitting, NV_UNIQUE_NAME, KM_SLEEP) == 0); VERIFY(nvlist_add_uint64(newspa->spa_config_splitting, ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0); spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL, B_TRUE)); } /* set the props */ if (props != NULL) { spa_configfile_set(newspa, props, B_FALSE); error = spa_prop_set(newspa, props); if (error) goto out; } /* flush everything */ txg = spa_vdev_config_enter(newspa); vdev_config_dirty(newspa->spa_root_vdev); (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG); if (zio_injection_enabled) zio_handle_panic_injection(spa, FTAG, 2); spa_async_resume(newspa); /* finally, update the original pool's config */ txg = spa_vdev_config_enter(spa); tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); error = dmu_tx_assign(tx, TXG_WAIT); if (error != 0) dmu_tx_abort(tx); for (c = 0; c < children; c++) { if (vml[c] != NULL) { vdev_split(vml[c]); if (error == 0) spa_history_log_internal(spa, "detach", tx, "vdev=%s", vml[c]->vdev_path); vdev_free(vml[c]); } } vdev_config_dirty(spa->spa_root_vdev); spa->spa_config_splitting = NULL; nvlist_free(nvl); if (error == 0) dmu_tx_commit(tx); (void) spa_vdev_exit(spa, NULL, txg, 0); if (zio_injection_enabled) zio_handle_panic_injection(spa, FTAG, 3); /* split is complete; log a history record */ spa_history_log_internal(newspa, "split", NULL, "from pool %s", spa_name(spa)); kmem_free(vml, children * sizeof (vdev_t *)); /* if we're not going to mount the filesystems in userland, export */ if (exp) error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL, B_FALSE, B_FALSE); return (error); out: spa_unload(newspa); spa_deactivate(newspa); spa_remove(newspa); txg = spa_vdev_config_enter(spa); /* re-online all offlined disks */ for (c = 0; c < children; c++) { if (vml[c] != NULL) vml[c]->vdev_offline = B_FALSE; } vdev_reopen(spa->spa_root_vdev); nvlist_free(spa->spa_config_splitting); spa->spa_config_splitting = NULL; (void) spa_vdev_exit(spa, NULL, txg, error); kmem_free(vml, children * sizeof (vdev_t *)); return (error); } static nvlist_t * spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid) { for (int i = 0; i < count; i++) { uint64_t guid; VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID, &guid) == 0); if (guid == target_guid) return (nvpp[i]); } return (NULL); } static void spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count, nvlist_t *dev_to_remove) { nvlist_t **newdev = NULL; if (count > 1) newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP); for (int i = 0, j = 0; i < count; i++) { if (dev[i] == dev_to_remove) continue; VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0); } VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0); VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0); for (int i = 0; i < count - 1; i++) nvlist_free(newdev[i]); if (count > 1) kmem_free(newdev, (count - 1) * sizeof (void *)); } /* * Evacuate the device. */ static int spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd) { uint64_t txg; int error = 0; ASSERT(MUTEX_HELD(&spa_namespace_lock)); ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); ASSERT(vd == vd->vdev_top); /* * Evacuate the device. We don't hold the config lock as writer * since we need to do I/O but we do keep the * spa_namespace_lock held. Once this completes the device * should no longer have any blocks allocated on it. */ if (vd->vdev_islog) { if (vd->vdev_stat.vs_alloc != 0) error = spa_offline_log(spa); } else { error = SET_ERROR(ENOTSUP); } if (error) return (error); /* * The evacuation succeeded. Remove any remaining MOS metadata * associated with this vdev, and wait for these changes to sync. */ ASSERT0(vd->vdev_stat.vs_alloc); txg = spa_vdev_config_enter(spa); vd->vdev_removing = B_TRUE; vdev_dirty_leaves(vd, VDD_DTL, txg); vdev_config_dirty(vd); spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); return (0); } /* * Complete the removal by cleaning up the namespace. */ static void spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd) { vdev_t *rvd = spa->spa_root_vdev; uint64_t id = vd->vdev_id; boolean_t last_vdev = (id == (rvd->vdev_children - 1)); ASSERT(MUTEX_HELD(&spa_namespace_lock)); ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); ASSERT(vd == vd->vdev_top); /* * Only remove any devices which are empty. */ if (vd->vdev_stat.vs_alloc != 0) return; (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); if (list_link_active(&vd->vdev_state_dirty_node)) vdev_state_clean(vd); if (list_link_active(&vd->vdev_config_dirty_node)) vdev_config_clean(vd); vdev_free(vd); if (last_vdev) { vdev_compact_children(rvd); } else { vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops); vdev_add_child(rvd, vd); } vdev_config_dirty(rvd); /* * Reassess the health of our root vdev. */ vdev_reopen(rvd); } /* * Remove a device from the pool - * * Removing a device from the vdev namespace requires several steps * and can take a significant amount of time. As a result we use * the spa_vdev_config_[enter/exit] functions which allow us to * grab and release the spa_config_lock while still holding the namespace * lock. During each step the configuration is synced out. * * Currently, this supports removing only hot spares, slogs, and level 2 ARC * devices. */ int spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare) { vdev_t *vd; metaslab_group_t *mg; nvlist_t **spares, **l2cache, *nv; uint64_t txg = 0; uint_t nspares, nl2cache; int error = 0; boolean_t locked = MUTEX_HELD(&spa_namespace_lock); ASSERT(spa_writeable(spa)); if (!locked) txg = spa_vdev_enter(spa); vd = spa_lookup_by_guid(spa, guid, B_FALSE); if (spa->spa_spares.sav_vdevs != NULL && nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 && (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) { /* * Only remove the hot spare if it's not currently in use * in this pool. */ if (vd == NULL || unspare) { spa_vdev_remove_aux(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, spares, nspares, nv); spa_load_spares(spa); spa->spa_spares.sav_sync = B_TRUE; } else { error = SET_ERROR(EBUSY); } } else if (spa->spa_l2cache.sav_vdevs != NULL && nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 && (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) { /* * Cache devices can always be removed. */ spa_vdev_remove_aux(spa->spa_l2cache.sav_config, ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv); spa_load_l2cache(spa); spa->spa_l2cache.sav_sync = B_TRUE; } else if (vd != NULL && vd->vdev_islog) { ASSERT(!locked); ASSERT(vd == vd->vdev_top); mg = vd->vdev_mg; /* * Stop allocating from this vdev. */ metaslab_group_passivate(mg); /* * Wait for the youngest allocations and frees to sync, * and then wait for the deferral of those frees to finish. */ spa_vdev_config_exit(spa, NULL, txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG); /* * Attempt to evacuate the vdev. */ error = spa_vdev_remove_evacuate(spa, vd); txg = spa_vdev_config_enter(spa); /* * If we couldn't evacuate the vdev, unwind. */ if (error) { metaslab_group_activate(mg); return (spa_vdev_exit(spa, NULL, txg, error)); } /* * Clean up the vdev namespace. */ spa_vdev_remove_from_namespace(spa, vd); } else if (vd != NULL) { /* * Normal vdevs cannot be removed (yet). */ error = SET_ERROR(ENOTSUP); } else { /* * There is no vdev of any kind with the specified guid. */ error = SET_ERROR(ENOENT); } if (!locked) return (spa_vdev_exit(spa, NULL, txg, error)); return (error); } /* * Find any device that's done replacing, or a vdev marked 'unspare' that's * currently spared, so we can detach it. */ static vdev_t * spa_vdev_resilver_done_hunt(vdev_t *vd) { vdev_t *newvd, *oldvd; for (int c = 0; c < vd->vdev_children; c++) { oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); if (oldvd != NULL) return (oldvd); } /* * Check for a completed replacement. We always consider the first * vdev in the list to be the oldest vdev, and the last one to be * the newest (see spa_vdev_attach() for how that works). In * the case where the newest vdev is faulted, we will not automatically * remove it after a resilver completes. This is OK as it will require * user intervention to determine which disk the admin wishes to keep. */ if (vd->vdev_ops == &vdev_replacing_ops) { ASSERT(vd->vdev_children > 1); newvd = vd->vdev_child[vd->vdev_children - 1]; oldvd = vd->vdev_child[0]; if (vdev_dtl_empty(newvd, DTL_MISSING) && vdev_dtl_empty(newvd, DTL_OUTAGE) && !vdev_dtl_required(oldvd)) return (oldvd); } /* * Check for a completed resilver with the 'unspare' flag set. */ if (vd->vdev_ops == &vdev_spare_ops) { vdev_t *first = vd->vdev_child[0]; vdev_t *last = vd->vdev_child[vd->vdev_children - 1]; if (last->vdev_unspare) { oldvd = first; newvd = last; } else if (first->vdev_unspare) { oldvd = last; newvd = first; } else { oldvd = NULL; } if (oldvd != NULL && vdev_dtl_empty(newvd, DTL_MISSING) && vdev_dtl_empty(newvd, DTL_OUTAGE) && !vdev_dtl_required(oldvd)) return (oldvd); /* * If there are more than two spares attached to a disk, * and those spares are not required, then we want to * attempt to free them up now so that they can be used * by other pools. Once we're back down to a single * disk+spare, we stop removing them. */ if (vd->vdev_children > 2) { newvd = vd->vdev_child[1]; if (newvd->vdev_isspare && last->vdev_isspare && vdev_dtl_empty(last, DTL_MISSING) && vdev_dtl_empty(last, DTL_OUTAGE) && !vdev_dtl_required(newvd)) return (newvd); } } return (NULL); } static void spa_vdev_resilver_done(spa_t *spa) { vdev_t *vd, *pvd, *ppvd; uint64_t guid, sguid, pguid, ppguid; spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { pvd = vd->vdev_parent; ppvd = pvd->vdev_parent; guid = vd->vdev_guid; pguid = pvd->vdev_guid; ppguid = ppvd->vdev_guid; sguid = 0; /* * If we have just finished replacing a hot spared device, then * we need to detach the parent's first child (the original hot * spare) as well. */ if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 && ppvd->vdev_children == 2) { ASSERT(pvd->vdev_ops == &vdev_replacing_ops); sguid = ppvd->vdev_child[1]->vdev_guid; } ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd)); spa_config_exit(spa, SCL_ALL, FTAG); if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0) return; if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0) return; spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); } spa_config_exit(spa, SCL_ALL, FTAG); } /* * Update the stored path or FRU for this vdev. */ int spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value, boolean_t ispath) { vdev_t *vd; boolean_t sync = B_FALSE; ASSERT(spa_writeable(spa)); spa_vdev_state_enter(spa, SCL_ALL); if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) return (spa_vdev_state_exit(spa, NULL, ENOENT)); if (!vd->vdev_ops->vdev_op_leaf) return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); if (ispath) { if (strcmp(value, vd->vdev_path) != 0) { spa_strfree(vd->vdev_path); vd->vdev_path = spa_strdup(value); sync = B_TRUE; } } else { if (vd->vdev_fru == NULL) { vd->vdev_fru = spa_strdup(value); sync = B_TRUE; } else if (strcmp(value, vd->vdev_fru) != 0) { spa_strfree(vd->vdev_fru); vd->vdev_fru = spa_strdup(value); sync = B_TRUE; } } return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0)); } int spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) { return (spa_vdev_set_common(spa, guid, newpath, B_TRUE)); } int spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru) { return (spa_vdev_set_common(spa, guid, newfru, B_FALSE)); } /* * ========================================================================== * SPA Scanning * ========================================================================== */ int spa_scan_stop(spa_t *spa) { ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); if (dsl_scan_resilvering(spa->spa_dsl_pool)) return (SET_ERROR(EBUSY)); return (dsl_scan_cancel(spa->spa_dsl_pool)); } int spa_scan(spa_t *spa, pool_scan_func_t func) { ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE) return (SET_ERROR(ENOTSUP)); /* * If a resilver was requested, but there is no DTL on a * writeable leaf device, we have nothing to do. */ if (func == POOL_SCAN_RESILVER && !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); return (0); } return (dsl_scan(spa->spa_dsl_pool, func)); } /* * ========================================================================== * SPA async task processing * ========================================================================== */ static void spa_async_remove(spa_t *spa, vdev_t *vd) { if (vd->vdev_remove_wanted) { vd->vdev_remove_wanted = B_FALSE; vd->vdev_delayed_close = B_FALSE; vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); /* * We want to clear the stats, but we don't want to do a full * vdev_clear() as that will cause us to throw away * degraded/faulted state as well as attempt to reopen the * device, all of which is a waste. */ vd->vdev_stat.vs_read_errors = 0; vd->vdev_stat.vs_write_errors = 0; vd->vdev_stat.vs_checksum_errors = 0; vdev_state_dirty(vd->vdev_top); } for (int c = 0; c < vd->vdev_children; c++) spa_async_remove(spa, vd->vdev_child[c]); } static void spa_async_probe(spa_t *spa, vdev_t *vd) { if (vd->vdev_probe_wanted) { vd->vdev_probe_wanted = B_FALSE; vdev_reopen(vd); /* vdev_open() does the actual probe */ } for (int c = 0; c < vd->vdev_children; c++) spa_async_probe(spa, vd->vdev_child[c]); } static void spa_async_autoexpand(spa_t *spa, vdev_t *vd) { sysevent_id_t eid; nvlist_t *attr; char *physpath; if (!spa->spa_autoexpand) return; for (int c = 0; c < vd->vdev_children; c++) { vdev_t *cvd = vd->vdev_child[c]; spa_async_autoexpand(spa, cvd); } if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL) return; physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath); VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP); nvlist_free(attr); kmem_free(physpath, MAXPATHLEN); } static void spa_async_thread(void *arg) { spa_t *spa = arg; int tasks; ASSERT(spa->spa_sync_on); mutex_enter(&spa->spa_async_lock); tasks = spa->spa_async_tasks; spa->spa_async_tasks &= SPA_ASYNC_REMOVE; mutex_exit(&spa->spa_async_lock); /* * See if the config needs to be updated. */ if (tasks & SPA_ASYNC_CONFIG_UPDATE) { uint64_t old_space, new_space; mutex_enter(&spa_namespace_lock); old_space = metaslab_class_get_space(spa_normal_class(spa)); spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); new_space = metaslab_class_get_space(spa_normal_class(spa)); mutex_exit(&spa_namespace_lock); /* * If the pool grew as a result of the config update, * then log an internal history event. */ if (new_space != old_space) { spa_history_log_internal(spa, "vdev online", NULL, "pool '%s' size: %llu(+%llu)", spa_name(spa), new_space, new_space - old_space); } } if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) { spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); spa_async_autoexpand(spa, spa->spa_root_vdev); spa_config_exit(spa, SCL_CONFIG, FTAG); } /* * See if any devices need to be probed. */ if (tasks & SPA_ASYNC_PROBE) { spa_vdev_state_enter(spa, SCL_NONE); spa_async_probe(spa, spa->spa_root_vdev); (void) spa_vdev_state_exit(spa, NULL, 0); } /* * If any devices are done replacing, detach them. */ if (tasks & SPA_ASYNC_RESILVER_DONE) spa_vdev_resilver_done(spa); /* * Kick off a resilver. */ if (tasks & SPA_ASYNC_RESILVER) dsl_resilver_restart(spa->spa_dsl_pool, 0); /* * Let the world know that we're done. */ mutex_enter(&spa->spa_async_lock); spa->spa_async_thread = NULL; cv_broadcast(&spa->spa_async_cv); mutex_exit(&spa->spa_async_lock); thread_exit(); } static void spa_async_thread_vd(void *arg) { spa_t *spa = arg; int tasks; ASSERT(spa->spa_sync_on); mutex_enter(&spa->spa_async_lock); tasks = spa->spa_async_tasks; retry: spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE; mutex_exit(&spa->spa_async_lock); /* * See if any devices need to be marked REMOVED. */ if (tasks & SPA_ASYNC_REMOVE) { spa_vdev_state_enter(spa, SCL_NONE); spa_async_remove(spa, spa->spa_root_vdev); for (int i = 0; i < spa->spa_l2cache.sav_count; i++) spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); for (int i = 0; i < spa->spa_spares.sav_count; i++) spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); (void) spa_vdev_state_exit(spa, NULL, 0); } /* * Let the world know that we're done. */ mutex_enter(&spa->spa_async_lock); tasks = spa->spa_async_tasks; if ((tasks & SPA_ASYNC_REMOVE) != 0) goto retry; spa->spa_async_thread_vd = NULL; cv_broadcast(&spa->spa_async_cv); mutex_exit(&spa->spa_async_lock); thread_exit(); } void spa_async_suspend(spa_t *spa) { mutex_enter(&spa->spa_async_lock); spa->spa_async_suspended++; while (spa->spa_async_thread != NULL && spa->spa_async_thread_vd != NULL) cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); mutex_exit(&spa->spa_async_lock); } void spa_async_resume(spa_t *spa) { mutex_enter(&spa->spa_async_lock); ASSERT(spa->spa_async_suspended != 0); spa->spa_async_suspended--; mutex_exit(&spa->spa_async_lock); } static boolean_t spa_async_tasks_pending(spa_t *spa) { uint_t non_config_tasks; uint_t config_task; boolean_t config_task_suspended; non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE | SPA_ASYNC_REMOVE); config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE; if (spa->spa_ccw_fail_time == 0) { config_task_suspended = B_FALSE; } else { config_task_suspended = (gethrtime() - spa->spa_ccw_fail_time) < (zfs_ccw_retry_interval * NANOSEC); } return (non_config_tasks || (config_task && !config_task_suspended)); } static void spa_async_dispatch(spa_t *spa) { mutex_enter(&spa->spa_async_lock); if (spa_async_tasks_pending(spa) && !spa->spa_async_suspended && spa->spa_async_thread == NULL && rootdir != NULL) spa->spa_async_thread = thread_create(NULL, 0, spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); mutex_exit(&spa->spa_async_lock); } static void spa_async_dispatch_vd(spa_t *spa) { mutex_enter(&spa->spa_async_lock); if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 && !spa->spa_async_suspended && spa->spa_async_thread_vd == NULL && rootdir != NULL) spa->spa_async_thread_vd = thread_create(NULL, 0, spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri); mutex_exit(&spa->spa_async_lock); } void spa_async_request(spa_t *spa, int task) { zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task); mutex_enter(&spa->spa_async_lock); spa->spa_async_tasks |= task; mutex_exit(&spa->spa_async_lock); spa_async_dispatch_vd(spa); } /* * ========================================================================== * SPA syncing routines * ========================================================================== */ static int bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) { bpobj_t *bpo = arg; bpobj_enqueue(bpo, bp, tx); return (0); } static int spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) { zio_t *zio = arg; zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp, BP_GET_PSIZE(bp), zio->io_flags)); return (0); } /* * Note: this simple function is not inlined to make it easier to dtrace the * amount of time spent syncing frees. */ static void spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx) { zio_t *zio = zio_root(spa, NULL, NULL, 0); bplist_iterate(bpl, spa_free_sync_cb, zio, tx); VERIFY(zio_wait(zio) == 0); } /* * Note: this simple function is not inlined to make it easier to dtrace the * amount of time spent syncing deferred frees. */ static void spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx) { zio_t *zio = zio_root(spa, NULL, NULL, 0); VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj, spa_free_sync_cb, zio, tx), ==, 0); VERIFY0(zio_wait(zio)); } static void spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) { char *packed = NULL; size_t bufsize; size_t nvsize = 0; dmu_buf_t *db; VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); /* * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration * information. This avoids the dmu_buf_will_dirty() path and * saves us a pre-read to get data we don't actually care about. */ bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE); packed = kmem_alloc(bufsize, KM_SLEEP); VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, KM_SLEEP) == 0); bzero(packed + nvsize, bufsize - nvsize); dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); kmem_free(packed, bufsize); VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); dmu_buf_will_dirty(db, tx); *(uint64_t *)db->db_data = nvsize; dmu_buf_rele(db, FTAG); } static void spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, const char *config, const char *entry) { nvlist_t *nvroot; nvlist_t **list; int i; if (!sav->sav_sync) return; /* * Update the MOS nvlist describing the list of available devices. * spa_validate_aux() will have already made sure this nvlist is * valid and the vdevs are labeled appropriately. */ if (sav->sav_object == 0) { sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); VERIFY(zap_update(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, &sav->sav_object, tx) == 0); } VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); if (sav->sav_count == 0) { VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0); } else { list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); for (i = 0; i < sav->sav_count; i++) list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], B_FALSE, VDEV_CONFIG_L2CACHE); VERIFY(nvlist_add_nvlist_array(nvroot, config, list, sav->sav_count) == 0); for (i = 0; i < sav->sav_count; i++) nvlist_free(list[i]); kmem_free(list, sav->sav_count * sizeof (void *)); } spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); nvlist_free(nvroot); sav->sav_sync = B_FALSE; } static void spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) { nvlist_t *config; if (list_is_empty(&spa->spa_config_dirty_list)) return; spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); config = spa_config_generate(spa, spa->spa_root_vdev, dmu_tx_get_txg(tx), B_FALSE); /* * If we're upgrading the spa version then make sure that * the config object gets updated with the correct version. */ if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version) fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, spa->spa_uberblock.ub_version); spa_config_exit(spa, SCL_STATE, FTAG); if (spa->spa_config_syncing) nvlist_free(spa->spa_config_syncing); spa->spa_config_syncing = config; spa_sync_nvlist(spa, spa->spa_config_object, config, tx); } static void spa_sync_version(void *arg, dmu_tx_t *tx) { uint64_t *versionp = arg; uint64_t version = *versionp; spa_t *spa = dmu_tx_pool(tx)->dp_spa; /* * Setting the version is special cased when first creating the pool. */ ASSERT(tx->tx_txg != TXG_INITIAL); ASSERT(SPA_VERSION_IS_SUPPORTED(version)); ASSERT(version >= spa_version(spa)); spa->spa_uberblock.ub_version = version; vdev_config_dirty(spa->spa_root_vdev); spa_history_log_internal(spa, "set", tx, "version=%lld", version); } /* * Set zpool properties. */ static void spa_sync_props(void *arg, dmu_tx_t *tx) { nvlist_t *nvp = arg; spa_t *spa = dmu_tx_pool(tx)->dp_spa; objset_t *mos = spa->spa_meta_objset; nvpair_t *elem = NULL; mutex_enter(&spa->spa_props_lock); while ((elem = nvlist_next_nvpair(nvp, elem))) { uint64_t intval; char *strval, *fname; zpool_prop_t prop; const char *propname; zprop_type_t proptype; spa_feature_t fid; switch (prop = zpool_name_to_prop(nvpair_name(elem))) { case ZPROP_INVAL: /* * We checked this earlier in spa_prop_validate(). */ ASSERT(zpool_prop_feature(nvpair_name(elem))); fname = strchr(nvpair_name(elem), '@') + 1; VERIFY0(zfeature_lookup_name(fname, &fid)); spa_feature_enable(spa, fid, tx); spa_history_log_internal(spa, "set", tx, "%s=enabled", nvpair_name(elem)); break; case ZPOOL_PROP_VERSION: intval = fnvpair_value_uint64(elem); /* * The version is synced seperatly before other * properties and should be correct by now. */ ASSERT3U(spa_version(spa), >=, intval); break; case ZPOOL_PROP_ALTROOT: /* * 'altroot' is a non-persistent property. It should * have been set temporarily at creation or import time. */ ASSERT(spa->spa_root != NULL); break; case ZPOOL_PROP_READONLY: case ZPOOL_PROP_CACHEFILE: /* * 'readonly' and 'cachefile' are also non-persisitent * properties. */ break; case ZPOOL_PROP_COMMENT: strval = fnvpair_value_string(elem); if (spa->spa_comment != NULL) spa_strfree(spa->spa_comment); spa->spa_comment = spa_strdup(strval); /* * We need to dirty the configuration on all the vdevs * so that their labels get updated. It's unnecessary * to do this for pool creation since the vdev's * configuratoin has already been dirtied. */ if (tx->tx_txg != TXG_INITIAL) vdev_config_dirty(spa->spa_root_vdev); spa_history_log_internal(spa, "set", tx, "%s=%s", nvpair_name(elem), strval); break; default: /* * Set pool property values in the poolprops mos object. */ if (spa->spa_pool_props_object == 0) { spa->spa_pool_props_object = zap_create_link(mos, DMU_OT_POOL_PROPS, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, tx); } /* normalize the property name */ propname = zpool_prop_to_name(prop); proptype = zpool_prop_get_type(prop); if (nvpair_type(elem) == DATA_TYPE_STRING) { ASSERT(proptype == PROP_TYPE_STRING); strval = fnvpair_value_string(elem); VERIFY0(zap_update(mos, spa->spa_pool_props_object, propname, 1, strlen(strval) + 1, strval, tx)); spa_history_log_internal(spa, "set", tx, "%s=%s", nvpair_name(elem), strval); } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { intval = fnvpair_value_uint64(elem); if (proptype == PROP_TYPE_INDEX) { const char *unused; VERIFY0(zpool_prop_index_to_string( prop, intval, &unused)); } VERIFY0(zap_update(mos, spa->spa_pool_props_object, propname, 8, 1, &intval, tx)); spa_history_log_internal(spa, "set", tx, "%s=%lld", nvpair_name(elem), intval); } else { ASSERT(0); /* not allowed */ } switch (prop) { case ZPOOL_PROP_DELEGATION: spa->spa_delegation = intval; break; case ZPOOL_PROP_BOOTFS: spa->spa_bootfs = intval; break; case ZPOOL_PROP_FAILUREMODE: spa->spa_failmode = intval; break; case ZPOOL_PROP_AUTOEXPAND: spa->spa_autoexpand = intval; if (tx->tx_txg != TXG_INITIAL) spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); break; case ZPOOL_PROP_DEDUPDITTO: spa->spa_dedup_ditto = intval; break; default: break; } } } mutex_exit(&spa->spa_props_lock); } /* * Perform one-time upgrade on-disk changes. spa_version() does not * reflect the new version this txg, so there must be no changes this * txg to anything that the upgrade code depends on after it executes. * Therefore this must be called after dsl_pool_sync() does the sync * tasks. */ static void spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) { dsl_pool_t *dp = spa->spa_dsl_pool; ASSERT(spa->spa_sync_pass == 1); rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { dsl_pool_create_origin(dp, tx); /* Keeping the origin open increases spa_minref */ spa->spa_minref += 3; } if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { dsl_pool_upgrade_clones(dp, tx); } if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES && spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) { dsl_pool_upgrade_dir_clones(dp, tx); /* Keeping the freedir open increases spa_minref */ spa->spa_minref += 3; } if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES && spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { spa_feature_create_zap_objects(spa, tx); } /* * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable * when possibility to use lz4 compression for metadata was added * Old pools that have this feature enabled must be upgraded to have * this feature active */ if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { boolean_t lz4_en = spa_feature_is_enabled(spa, SPA_FEATURE_LZ4_COMPRESS); boolean_t lz4_ac = spa_feature_is_active(spa, SPA_FEATURE_LZ4_COMPRESS); if (lz4_en && !lz4_ac) spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx); } rrw_exit(&dp->dp_config_rwlock, FTAG); } /* * Sync the specified transaction group. New blocks may be dirtied as * part of the process, so we iterate until it converges. */ void spa_sync(spa_t *spa, uint64_t txg) { dsl_pool_t *dp = spa->spa_dsl_pool; objset_t *mos = spa->spa_meta_objset; bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK]; vdev_t *rvd = spa->spa_root_vdev; vdev_t *vd; dmu_tx_t *tx; int error; VERIFY(spa_writeable(spa)); /* * Lock out configuration changes. */ spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); spa->spa_syncing_txg = txg; spa->spa_sync_pass = 0; /* * If there are any pending vdev state changes, convert them * into config changes that go out with this transaction group. */ spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); while (list_head(&spa->spa_state_dirty_list) != NULL) { /* * We need the write lock here because, for aux vdevs, * calling vdev_config_dirty() modifies sav_config. * This is ugly and will become unnecessary when we * eliminate the aux vdev wart by integrating all vdevs * into the root vdev tree. */ spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER); while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { vdev_state_clean(vd); vdev_config_dirty(vd); } spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); } spa_config_exit(spa, SCL_STATE, FTAG); tx = dmu_tx_create_assigned(dp, txg); spa->spa_sync_starttime = gethrtime(); #ifdef illumos VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, spa->spa_sync_starttime + spa->spa_deadman_synctime)); #else /* FreeBSD */ #ifdef _KERNEL callout_reset(&spa->spa_deadman_cycid, hz * spa->spa_deadman_synctime / NANOSEC, spa_deadman, spa); #endif #endif /* * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, * set spa_deflate if we have no raid-z vdevs. */ if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { int i; for (i = 0; i < rvd->vdev_children; i++) { vd = rvd->vdev_child[i]; if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) break; } if (i == rvd->vdev_children) { spa->spa_deflate = TRUE; VERIFY(0 == zap_add(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, sizeof (uint64_t), 1, &spa->spa_deflate, tx)); } } /* * If anything has changed in this txg, or if someone is waiting * for this txg to sync (eg, spa_vdev_remove()), push the * deferred frees from the previous txg. If not, leave them * alone so that we don't generate work on an otherwise idle * system. */ if (!txg_list_empty(&dp->dp_dirty_datasets, txg) || !txg_list_empty(&dp->dp_dirty_dirs, txg) || !txg_list_empty(&dp->dp_sync_tasks, txg) || ((dsl_scan_active(dp->dp_scan) || txg_sync_waiting(dp)) && !spa_shutting_down(spa))) { spa_sync_deferred_frees(spa, tx); } /* * Iterate to convergence. */ do { int pass = ++spa->spa_sync_pass; spa_sync_config_object(spa, tx); spa_sync_aux_dev(spa, &spa->spa_spares, tx, ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); spa_errlog_sync(spa, txg); dsl_pool_sync(dp, txg); if (pass < zfs_sync_pass_deferred_free) { spa_sync_frees(spa, free_bpl, tx); } else { bplist_iterate(free_bpl, bpobj_enqueue_cb, &spa->spa_deferred_bpobj, tx); } ddt_sync(spa, txg); dsl_scan_sync(dp, tx); while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) vdev_sync(vd, txg); if (pass == 1) spa_sync_upgrades(spa, tx); } while (dmu_objset_is_dirty(mos, txg)); /* * Rewrite the vdev configuration (which includes the uberblock) * to commit the transaction group. * * If there are no dirty vdevs, we sync the uberblock to a few * random top-level vdevs that are known to be visible in the * config cache (see spa_vdev_add() for a complete description). * If there *are* dirty vdevs, sync the uberblock to all vdevs. */ for (;;) { /* * We hold SCL_STATE to prevent vdev open/close/etc. * while we're attempting to write the vdev labels. */ spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); if (list_is_empty(&spa->spa_config_dirty_list)) { vdev_t *svd[SPA_DVAS_PER_BP]; int svdcount = 0; int children = rvd->vdev_children; int c0 = spa_get_random(children); for (int c = 0; c < children; c++) { vd = rvd->vdev_child[(c0 + c) % children]; if (vd->vdev_ms_array == 0 || vd->vdev_islog) continue; svd[svdcount++] = vd; if (svdcount == SPA_DVAS_PER_BP) break; } error = vdev_config_sync(svd, svdcount, txg, B_FALSE); if (error != 0) error = vdev_config_sync(svd, svdcount, txg, B_TRUE); } else { error = vdev_config_sync(rvd->vdev_child, rvd->vdev_children, txg, B_FALSE); if (error != 0) error = vdev_config_sync(rvd->vdev_child, rvd->vdev_children, txg, B_TRUE); } if (error == 0) spa->spa_last_synced_guid = rvd->vdev_guid; spa_config_exit(spa, SCL_STATE, FTAG); if (error == 0) break; zio_suspend(spa, NULL); zio_resume_wait(spa); } dmu_tx_commit(tx); #ifdef illumos VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); #else /* FreeBSD */ #ifdef _KERNEL callout_drain(&spa->spa_deadman_cycid); #endif #endif /* * Clear the dirty config list. */ while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) vdev_config_clean(vd); /* * Now that the new config has synced transactionally, * let it become visible to the config cache. */ if (spa->spa_config_syncing != NULL) { spa_config_set(spa, spa->spa_config_syncing); spa->spa_config_txg = txg; spa->spa_config_syncing = NULL; } spa->spa_ubsync = spa->spa_uberblock; dsl_pool_sync_done(dp, txg); /* * Update usable space statistics. */ while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) vdev_sync_done(vd, txg); spa_update_dspace(spa); /* * It had better be the case that we didn't dirty anything * since vdev_config_sync(). */ ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); spa->spa_sync_pass = 0; spa_config_exit(spa, SCL_CONFIG, FTAG); spa_handle_ignored_writes(spa); /* * If any async tasks have been requested, kick them off. */ spa_async_dispatch(spa); spa_async_dispatch_vd(spa); } /* * Sync all pools. We don't want to hold the namespace lock across these * operations, so we take a reference on the spa_t and drop the lock during the * sync. */ void spa_sync_allpools(void) { spa_t *spa = NULL; mutex_enter(&spa_namespace_lock); while ((spa = spa_next(spa)) != NULL) { if (spa_state(spa) != POOL_STATE_ACTIVE || !spa_writeable(spa) || spa_suspended(spa)) continue; spa_open_ref(spa, FTAG); mutex_exit(&spa_namespace_lock); txg_wait_synced(spa_get_dsl(spa), 0); mutex_enter(&spa_namespace_lock); spa_close(spa, FTAG); } mutex_exit(&spa_namespace_lock); } /* * ========================================================================== * Miscellaneous routines * ========================================================================== */ /* * Remove all pools in the system. */ void spa_evict_all(void) { spa_t *spa; /* * Remove all cached state. All pools should be closed now, * so every spa in the AVL tree should be unreferenced. */ mutex_enter(&spa_namespace_lock); while ((spa = spa_next(NULL)) != NULL) { /* * Stop async tasks. The async thread may need to detach * a device that's been replaced, which requires grabbing * spa_namespace_lock, so we must drop it here. */ spa_open_ref(spa, FTAG); mutex_exit(&spa_namespace_lock); spa_async_suspend(spa); mutex_enter(&spa_namespace_lock); spa_close(spa, FTAG); if (spa->spa_state != POOL_STATE_UNINITIALIZED) { spa_unload(spa); spa_deactivate(spa); } spa_remove(spa); } mutex_exit(&spa_namespace_lock); } vdev_t * spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux) { vdev_t *vd; int i; if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) return (vd); if (aux) { for (i = 0; i < spa->spa_l2cache.sav_count; i++) { vd = spa->spa_l2cache.sav_vdevs[i]; if (vd->vdev_guid == guid) return (vd); } for (i = 0; i < spa->spa_spares.sav_count; i++) { vd = spa->spa_spares.sav_vdevs[i]; if (vd->vdev_guid == guid) return (vd); } } return (NULL); } void spa_upgrade(spa_t *spa, uint64_t version) { ASSERT(spa_writeable(spa)); spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); /* * This should only be called for a non-faulted pool, and since a * future version would result in an unopenable pool, this shouldn't be * possible. */ ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version)); ASSERT3U(version, >=, spa->spa_uberblock.ub_version); spa->spa_uberblock.ub_version = version; vdev_config_dirty(spa->spa_root_vdev); spa_config_exit(spa, SCL_ALL, FTAG); txg_wait_synced(spa_get_dsl(spa), 0); } boolean_t spa_has_spare(spa_t *spa, uint64_t guid) { int i; uint64_t spareguid; spa_aux_vdev_t *sav = &spa->spa_spares; for (i = 0; i < sav->sav_count; i++) if (sav->sav_vdevs[i]->vdev_guid == guid) return (B_TRUE); for (i = 0; i < sav->sav_npending; i++) { if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, &spareguid) == 0 && spareguid == guid) return (B_TRUE); } return (B_FALSE); } /* * Check if a pool has an active shared spare device. * Note: reference count of an active spare is 2, as a spare and as a replace */ static boolean_t spa_has_active_shared_spare(spa_t *spa) { int i, refcnt; uint64_t pool; spa_aux_vdev_t *sav = &spa->spa_spares; for (i = 0; i < sav->sav_count; i++) { if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, &refcnt) && pool != 0ULL && pool == spa_guid(spa) && refcnt > 2) return (B_TRUE); } return (B_FALSE); } /* * Post a sysevent corresponding to the given event. The 'name' must be one of * the event definitions in sys/sysevent/eventdefs.h. The payload will be * filled in from the spa and (optionally) the vdev. This doesn't do anything * in the userland libzpool, as we don't want consumers to misinterpret ztest * or zdb as real changes. */ void spa_event_notify(spa_t *spa, vdev_t *vd, const char *name) { #ifdef _KERNEL sysevent_t *ev; sysevent_attr_list_t *attr = NULL; sysevent_value_t value; sysevent_id_t eid; ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs", SE_SLEEP); value.value_type = SE_DATA_TYPE_STRING; value.value.sv_string = spa_name(spa); if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0) goto done; value.value_type = SE_DATA_TYPE_UINT64; value.value.sv_uint64 = spa_guid(spa); if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0) goto done; if (vd) { value.value_type = SE_DATA_TYPE_UINT64; value.value.sv_uint64 = vd->vdev_guid; if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value, SE_SLEEP) != 0) goto done; if (vd->vdev_path) { value.value_type = SE_DATA_TYPE_STRING; value.value.sv_string = vd->vdev_path; if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH, &value, SE_SLEEP) != 0) goto done; } } if (sysevent_attach_attributes(ev, attr) != 0) goto done; attr = NULL; (void) log_sysevent(ev, SE_SLEEP, &eid); done: if (attr) sysevent_free_attr(attr); sysevent_free(ev); #endif } Index: stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/refcount.h =================================================================== --- stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/refcount.h (revision 271000) +++ stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/refcount.h (revision 271001) @@ -1,113 +1,113 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 2012 by Delphix. All rights reserved. */ #ifndef _SYS_REFCOUNT_H #define _SYS_REFCOUNT_H #include #include #include_next #include #include #ifdef __cplusplus extern "C" { #endif /* * If the reference is held only by the calling function and not any * particular object, use FTAG (which is a string) for the holder_tag. * Otherwise, use the object that holds the reference. */ #define FTAG ((char *)__func__) #ifdef ZFS_DEBUG typedef struct reference { list_node_t ref_link; void *ref_holder; uint64_t ref_number; uint8_t *ref_removed; } reference_t; typedef struct refcount { kmutex_t rc_mtx; boolean_t rc_tracked; list_t rc_list; list_t rc_removed; uint64_t rc_count; uint64_t rc_removed_count; } refcount_t; /* Note: refcount_t must be initialized with refcount_create[_untracked]() */ void refcount_create(refcount_t *rc); void refcount_create_untracked(refcount_t *rc); void refcount_destroy(refcount_t *rc); void refcount_destroy_many(refcount_t *rc, uint64_t number); int refcount_is_zero(refcount_t *rc); int64_t refcount_count(refcount_t *rc); int64_t refcount_add(refcount_t *rc, void *holder_tag); int64_t refcount_remove(refcount_t *rc, void *holder_tag); int64_t refcount_add_many(refcount_t *rc, uint64_t number, void *holder_tag); int64_t refcount_remove_many(refcount_t *rc, uint64_t number, void *holder_tag); void refcount_transfer(refcount_t *dst, refcount_t *src); void refcount_sysinit(void); void refcount_fini(void); #else /* ZFS_DEBUG */ typedef struct refcount { uint64_t rc_count; } refcount_t; #define refcount_create(rc) ((rc)->rc_count = 0) #define refcount_create_untracked(rc) ((rc)->rc_count = 0) #define refcount_destroy(rc) ((rc)->rc_count = 0) #define refcount_destroy_many(rc, number) ((rc)->rc_count = 0) #define refcount_is_zero(rc) ((rc)->rc_count == 0) #define refcount_count(rc) ((rc)->rc_count) -#define refcount_add(rc, holder) atomic_add_64_nv(&(rc)->rc_count, 1) -#define refcount_remove(rc, holder) atomic_add_64_nv(&(rc)->rc_count, -1) +#define refcount_add(rc, holder) atomic_inc_64_nv(&(rc)->rc_count) +#define refcount_remove(rc, holder) atomic_dec_64_nv(&(rc)->rc_count) #define refcount_add_many(rc, number, holder) \ atomic_add_64_nv(&(rc)->rc_count, number) #define refcount_remove_many(rc, number, holder) \ atomic_add_64_nv(&(rc)->rc_count, -number) #define refcount_transfer(dst, src) { \ uint64_t __tmp = (src)->rc_count; \ atomic_add_64(&(src)->rc_count, -__tmp); \ atomic_add_64(&(dst)->rc_count, __tmp); \ } #define refcount_sysinit() #define refcount_fini() #endif /* ZFS_DEBUG */ #ifdef __cplusplus } #endif #endif /* _SYS_REFCOUNT_H */ Index: stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_cache.c =================================================================== --- stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_cache.c (revision 271000) +++ stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_cache.c (revision 271001) @@ -1,438 +1,438 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright 2009 Sun Microsystems, Inc. All rights reserved. * Use is subject to license terms. */ /* * Copyright (c) 2013 by Delphix. All rights reserved. */ #include #include #include #include #include /* * Virtual device read-ahead caching. * * This file implements a simple LRU read-ahead cache. When the DMU reads * a given block, it will often want other, nearby blocks soon thereafter. * We take advantage of this by reading a larger disk region and caching * the result. In the best case, this can turn 128 back-to-back 512-byte * reads into a single 64k read followed by 127 cache hits; this reduces * latency dramatically. In the worst case, it can turn an isolated 512-byte * read into a 64k read, which doesn't affect latency all that much but is * terribly wasteful of bandwidth. A more intelligent version of the cache * could keep track of access patterns and not do read-ahead unless it sees * at least two temporally close I/Os to the same region. Currently, only * metadata I/O is inflated. A futher enhancement could take advantage of * more semantic information about the I/O. And it could use something * faster than an AVL tree; that was chosen solely for convenience. * * There are five cache operations: allocate, fill, read, write, evict. * * (1) Allocate. This reserves a cache entry for the specified region. * We separate the allocate and fill operations so that multiple threads * don't generate I/O for the same cache miss. * * (2) Fill. When the I/O for a cache miss completes, the fill routine * places the data in the previously allocated cache entry. * * (3) Read. Read data from the cache. * * (4) Write. Update cache contents after write completion. * * (5) Evict. When allocating a new entry, we evict the oldest (LRU) entry * if the total cache size exceeds zfs_vdev_cache_size. */ /* * These tunables are for performance analysis. */ /* * All i/os smaller than zfs_vdev_cache_max will be turned into * 1<ve_offset < ve2->ve_offset) return (-1); if (ve1->ve_offset > ve2->ve_offset) return (1); return (0); } static int vdev_cache_lastused_compare(const void *a1, const void *a2) { const vdev_cache_entry_t *ve1 = a1; const vdev_cache_entry_t *ve2 = a2; if (ve1->ve_lastused < ve2->ve_lastused) return (-1); if (ve1->ve_lastused > ve2->ve_lastused) return (1); /* * Among equally old entries, sort by offset to ensure uniqueness. */ return (vdev_cache_offset_compare(a1, a2)); } /* * Evict the specified entry from the cache. */ static void vdev_cache_evict(vdev_cache_t *vc, vdev_cache_entry_t *ve) { ASSERT(MUTEX_HELD(&vc->vc_lock)); ASSERT(ve->ve_fill_io == NULL); ASSERT(ve->ve_data != NULL); avl_remove(&vc->vc_lastused_tree, ve); avl_remove(&vc->vc_offset_tree, ve); zio_buf_free(ve->ve_data, VCBS); kmem_free(ve, sizeof (vdev_cache_entry_t)); } /* * Allocate an entry in the cache. At the point we don't have the data, * we're just creating a placeholder so that multiple threads don't all * go off and read the same blocks. */ static vdev_cache_entry_t * vdev_cache_allocate(zio_t *zio) { vdev_cache_t *vc = &zio->io_vd->vdev_cache; uint64_t offset = P2ALIGN(zio->io_offset, VCBS); vdev_cache_entry_t *ve; ASSERT(MUTEX_HELD(&vc->vc_lock)); if (zfs_vdev_cache_size == 0) return (NULL); /* * If adding a new entry would exceed the cache size, * evict the oldest entry (LRU). */ if ((avl_numnodes(&vc->vc_lastused_tree) << zfs_vdev_cache_bshift) > zfs_vdev_cache_size) { ve = avl_first(&vc->vc_lastused_tree); if (ve->ve_fill_io != NULL) return (NULL); ASSERT(ve->ve_hits != 0); vdev_cache_evict(vc, ve); } ve = kmem_zalloc(sizeof (vdev_cache_entry_t), KM_SLEEP); ve->ve_offset = offset; ve->ve_lastused = ddi_get_lbolt(); ve->ve_data = zio_buf_alloc(VCBS); avl_add(&vc->vc_offset_tree, ve); avl_add(&vc->vc_lastused_tree, ve); return (ve); } static void vdev_cache_hit(vdev_cache_t *vc, vdev_cache_entry_t *ve, zio_t *zio) { uint64_t cache_phase = P2PHASE(zio->io_offset, VCBS); ASSERT(MUTEX_HELD(&vc->vc_lock)); ASSERT(ve->ve_fill_io == NULL); if (ve->ve_lastused != ddi_get_lbolt()) { avl_remove(&vc->vc_lastused_tree, ve); ve->ve_lastused = ddi_get_lbolt(); avl_add(&vc->vc_lastused_tree, ve); } ve->ve_hits++; bcopy(ve->ve_data + cache_phase, zio->io_data, zio->io_size); } /* * Fill a previously allocated cache entry with data. */ static void vdev_cache_fill(zio_t *fio) { vdev_t *vd = fio->io_vd; vdev_cache_t *vc = &vd->vdev_cache; vdev_cache_entry_t *ve = fio->io_private; zio_t *pio; ASSERT(fio->io_size == VCBS); /* * Add data to the cache. */ mutex_enter(&vc->vc_lock); ASSERT(ve->ve_fill_io == fio); ASSERT(ve->ve_offset == fio->io_offset); ASSERT(ve->ve_data == fio->io_data); ve->ve_fill_io = NULL; /* * Even if this cache line was invalidated by a missed write update, * any reads that were queued up before the missed update are still * valid, so we can satisfy them from this line before we evict it. */ while ((pio = zio_walk_parents(fio)) != NULL) vdev_cache_hit(vc, ve, pio); if (fio->io_error || ve->ve_missed_update) vdev_cache_evict(vc, ve); mutex_exit(&vc->vc_lock); } /* * Read data from the cache. Returns B_TRUE cache hit, B_FALSE on miss. */ boolean_t vdev_cache_read(zio_t *zio) { vdev_cache_t *vc = &zio->io_vd->vdev_cache; vdev_cache_entry_t *ve, ve_search; uint64_t cache_offset = P2ALIGN(zio->io_offset, VCBS); uint64_t cache_phase = P2PHASE(zio->io_offset, VCBS); zio_t *fio; ASSERT(zio->io_type == ZIO_TYPE_READ); if (zio->io_flags & ZIO_FLAG_DONT_CACHE) return (B_FALSE); if (zio->io_size > zfs_vdev_cache_max) return (B_FALSE); /* * If the I/O straddles two or more cache blocks, don't cache it. */ if (P2BOUNDARY(zio->io_offset, zio->io_size, VCBS)) return (B_FALSE); ASSERT(cache_phase + zio->io_size <= VCBS); mutex_enter(&vc->vc_lock); ve_search.ve_offset = cache_offset; ve = avl_find(&vc->vc_offset_tree, &ve_search, NULL); if (ve != NULL) { if (ve->ve_missed_update) { mutex_exit(&vc->vc_lock); return (B_FALSE); } if ((fio = ve->ve_fill_io) != NULL) { zio_vdev_io_bypass(zio); zio_add_child(zio, fio); mutex_exit(&vc->vc_lock); VDCSTAT_BUMP(vdc_stat_delegations); return (B_TRUE); } vdev_cache_hit(vc, ve, zio); zio_vdev_io_bypass(zio); mutex_exit(&vc->vc_lock); VDCSTAT_BUMP(vdc_stat_hits); return (B_TRUE); } ve = vdev_cache_allocate(zio); if (ve == NULL) { mutex_exit(&vc->vc_lock); return (B_FALSE); } fio = zio_vdev_delegated_io(zio->io_vd, cache_offset, ve->ve_data, VCBS, ZIO_TYPE_READ, ZIO_PRIORITY_NOW, ZIO_FLAG_DONT_CACHE, vdev_cache_fill, ve); ve->ve_fill_io = fio; zio_vdev_io_bypass(zio); zio_add_child(zio, fio); mutex_exit(&vc->vc_lock); zio_nowait(fio); VDCSTAT_BUMP(vdc_stat_misses); return (B_TRUE); } /* * Update cache contents upon write completion. */ void vdev_cache_write(zio_t *zio) { vdev_cache_t *vc = &zio->io_vd->vdev_cache; vdev_cache_entry_t *ve, ve_search; uint64_t io_start = zio->io_offset; uint64_t io_end = io_start + zio->io_size; uint64_t min_offset = P2ALIGN(io_start, VCBS); uint64_t max_offset = P2ROUNDUP(io_end, VCBS); avl_index_t where; ASSERT(zio->io_type == ZIO_TYPE_WRITE); mutex_enter(&vc->vc_lock); ve_search.ve_offset = min_offset; ve = avl_find(&vc->vc_offset_tree, &ve_search, &where); if (ve == NULL) ve = avl_nearest(&vc->vc_offset_tree, where, AVL_AFTER); while (ve != NULL && ve->ve_offset < max_offset) { uint64_t start = MAX(ve->ve_offset, io_start); uint64_t end = MIN(ve->ve_offset + VCBS, io_end); if (ve->ve_fill_io != NULL) { ve->ve_missed_update = 1; } else { bcopy((char *)zio->io_data + start - io_start, ve->ve_data + start - ve->ve_offset, end - start); } ve = AVL_NEXT(&vc->vc_offset_tree, ve); } mutex_exit(&vc->vc_lock); } void vdev_cache_purge(vdev_t *vd) { vdev_cache_t *vc = &vd->vdev_cache; vdev_cache_entry_t *ve; mutex_enter(&vc->vc_lock); while ((ve = avl_first(&vc->vc_offset_tree)) != NULL) vdev_cache_evict(vc, ve); mutex_exit(&vc->vc_lock); } void vdev_cache_init(vdev_t *vd) { vdev_cache_t *vc = &vd->vdev_cache; mutex_init(&vc->vc_lock, NULL, MUTEX_DEFAULT, NULL); avl_create(&vc->vc_offset_tree, vdev_cache_offset_compare, sizeof (vdev_cache_entry_t), offsetof(struct vdev_cache_entry, ve_offset_node)); avl_create(&vc->vc_lastused_tree, vdev_cache_lastused_compare, sizeof (vdev_cache_entry_t), offsetof(struct vdev_cache_entry, ve_lastused_node)); } void vdev_cache_fini(vdev_t *vd) { vdev_cache_t *vc = &vd->vdev_cache; vdev_cache_purge(vd); avl_destroy(&vc->vc_offset_tree); avl_destroy(&vc->vc_lastused_tree); mutex_destroy(&vc->vc_lock); } void vdev_cache_stat_init(void) { vdc_ksp = kstat_create("zfs", 0, "vdev_cache_stats", "misc", KSTAT_TYPE_NAMED, sizeof (vdc_stats) / sizeof (kstat_named_t), KSTAT_FLAG_VIRTUAL); if (vdc_ksp != NULL) { vdc_ksp->ks_data = &vdc_stats; kstat_install(vdc_ksp); } } void vdev_cache_stat_fini(void) { if (vdc_ksp != NULL) { kstat_delete(vdc_ksp); vdc_ksp = NULL; } } Index: stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_label.c =================================================================== --- stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_label.c (revision 271000) +++ stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_label.c (revision 271001) @@ -1,1292 +1,1292 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 2013 by Delphix. All rights reserved. */ /* * Virtual Device Labels * --------------------- * * The vdev label serves several distinct purposes: * * 1. Uniquely identify this device as part of a ZFS pool and confirm its * identity within the pool. * * 2. Verify that all the devices given in a configuration are present * within the pool. * * 3. Determine the uberblock for the pool. * * 4. In case of an import operation, determine the configuration of the * toplevel vdev of which it is a part. * * 5. If an import operation cannot find all the devices in the pool, * provide enough information to the administrator to determine which * devices are missing. * * It is important to note that while the kernel is responsible for writing the * label, it only consumes the information in the first three cases. The * latter information is only consumed in userland when determining the * configuration to import a pool. * * * Label Organization * ------------------ * * Before describing the contents of the label, it's important to understand how * the labels are written and updated with respect to the uberblock. * * When the pool configuration is altered, either because it was newly created * or a device was added, we want to update all the labels such that we can deal * with fatal failure at any point. To this end, each disk has two labels which * are updated before and after the uberblock is synced. Assuming we have * labels and an uberblock with the following transaction groups: * * L1 UB L2 * +------+ +------+ +------+ * | | | | | | * | t10 | | t10 | | t10 | * | | | | | | * +------+ +------+ +------+ * * In this stable state, the labels and the uberblock were all updated within * the same transaction group (10). Each label is mirrored and checksummed, so * that we can detect when we fail partway through writing the label. * * In order to identify which labels are valid, the labels are written in the * following manner: * * 1. For each vdev, update 'L1' to the new label * 2. Update the uberblock * 3. For each vdev, update 'L2' to the new label * * Given arbitrary failure, we can determine the correct label to use based on * the transaction group. If we fail after updating L1 but before updating the * UB, we will notice that L1's transaction group is greater than the uberblock, * so L2 must be valid. If we fail after writing the uberblock but before * writing L2, we will notice that L2's transaction group is less than L1, and * therefore L1 is valid. * * Another added complexity is that not every label is updated when the config * is synced. If we add a single device, we do not want to have to re-write * every label for every device in the pool. This means that both L1 and L2 may * be older than the pool uberblock, because the necessary information is stored * on another vdev. * * * On-disk Format * -------------- * * The vdev label consists of two distinct parts, and is wrapped within the * vdev_label_t structure. The label includes 8k of padding to permit legacy * VTOC disk labels, but is otherwise ignored. * * The first half of the label is a packed nvlist which contains pool wide * properties, per-vdev properties, and configuration information. It is * described in more detail below. * * The latter half of the label consists of a redundant array of uberblocks. * These uberblocks are updated whenever a transaction group is committed, * or when the configuration is updated. When a pool is loaded, we scan each * vdev for the 'best' uberblock. * * * Configuration Information * ------------------------- * * The nvlist describing the pool and vdev contains the following elements: * * version ZFS on-disk version * name Pool name * state Pool state * txg Transaction group in which this label was written * pool_guid Unique identifier for this pool * vdev_tree An nvlist describing vdev tree. * features_for_read * An nvlist of the features necessary for reading the MOS. * * Each leaf device label also contains the following: * * top_guid Unique ID for top-level vdev in which this is contained * guid Unique ID for the leaf vdev * * The 'vs' configuration follows the format described in 'spa_config.c'. */ #include #include #include #include #include #include #include #include #include #include #include #include #include static boolean_t vdev_trim_on_init = B_TRUE; SYSCTL_DECL(_vfs_zfs_vdev); SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, trim_on_init, CTLFLAG_RW, &vdev_trim_on_init, 0, "Enable/disable full vdev trim on initialisation"); /* * Basic routines to read and write from a vdev label. * Used throughout the rest of this file. */ uint64_t vdev_label_offset(uint64_t psize, int l, uint64_t offset) { ASSERT(offset < sizeof (vdev_label_t)); ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0); return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ? 0 : psize - VDEV_LABELS * sizeof (vdev_label_t))); } /* * Returns back the vdev label associated with the passed in offset. */ int vdev_label_number(uint64_t psize, uint64_t offset) { int l; if (offset >= psize - VDEV_LABEL_END_SIZE) { offset -= psize - VDEV_LABEL_END_SIZE; offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t); } l = offset / sizeof (vdev_label_t); return (l < VDEV_LABELS ? l : -1); } static void vdev_label_read(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset, uint64_t size, zio_done_func_t *done, void *private, int flags) { ASSERT(spa_config_held(zio->io_spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); ASSERT(flags & ZIO_FLAG_CONFIG_WRITER); zio_nowait(zio_read_phys(zio, vd, vdev_label_offset(vd->vdev_psize, l, offset), size, buf, ZIO_CHECKSUM_LABEL, done, private, ZIO_PRIORITY_SYNC_READ, flags, B_TRUE)); } static void vdev_label_write(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset, uint64_t size, zio_done_func_t *done, void *private, int flags) { ASSERT(spa_config_held(zio->io_spa, SCL_ALL, RW_WRITER) == SCL_ALL || (spa_config_held(zio->io_spa, SCL_CONFIG | SCL_STATE, RW_READER) == (SCL_CONFIG | SCL_STATE) && dsl_pool_sync_context(spa_get_dsl(zio->io_spa)))); ASSERT(flags & ZIO_FLAG_CONFIG_WRITER); zio_nowait(zio_write_phys(zio, vd, vdev_label_offset(vd->vdev_psize, l, offset), size, buf, ZIO_CHECKSUM_LABEL, done, private, ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE)); } /* * Generate the nvlist representing this vdev's config. */ nvlist_t * vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats, vdev_config_flag_t flags) { nvlist_t *nv = NULL; nv = fnvlist_alloc(); fnvlist_add_string(nv, ZPOOL_CONFIG_TYPE, vd->vdev_ops->vdev_op_type); if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE))) fnvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id); fnvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid); if (vd->vdev_path != NULL) fnvlist_add_string(nv, ZPOOL_CONFIG_PATH, vd->vdev_path); if (vd->vdev_devid != NULL) fnvlist_add_string(nv, ZPOOL_CONFIG_DEVID, vd->vdev_devid); if (vd->vdev_physpath != NULL) fnvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH, vd->vdev_physpath); if (vd->vdev_fru != NULL) fnvlist_add_string(nv, ZPOOL_CONFIG_FRU, vd->vdev_fru); if (vd->vdev_nparity != 0) { ASSERT(strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_RAIDZ) == 0); /* * Make sure someone hasn't managed to sneak a fancy new vdev * into a crufty old storage pool. */ ASSERT(vd->vdev_nparity == 1 || (vd->vdev_nparity <= 2 && spa_version(spa) >= SPA_VERSION_RAIDZ2) || (vd->vdev_nparity <= 3 && spa_version(spa) >= SPA_VERSION_RAIDZ3)); /* * Note that we'll add the nparity tag even on storage pools * that only support a single parity device -- older software * will just ignore it. */ fnvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, vd->vdev_nparity); } if (vd->vdev_wholedisk != -1ULL) fnvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, vd->vdev_wholedisk); if (vd->vdev_not_present) fnvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1); if (vd->vdev_isspare) fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1); if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) && vd == vd->vdev_top) { fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, vd->vdev_ms_array); fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, vd->vdev_ms_shift); fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift); fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE, vd->vdev_asize); fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, vd->vdev_islog); if (vd->vdev_removing) fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING, vd->vdev_removing); } if (vd->vdev_dtl_sm != NULL) { fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL, space_map_object(vd->vdev_dtl_sm)); } if (vd->vdev_crtxg) fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg); if (getstats) { vdev_stat_t vs; pool_scan_stat_t ps; vdev_get_stats(vd, &vs); fnvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS, (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t)); /* provide either current or previous scan information */ if (spa_scan_get_stats(spa, &ps) == 0) { fnvlist_add_uint64_array(nv, ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps, sizeof (pool_scan_stat_t) / sizeof (uint64_t)); } } if (!vd->vdev_ops->vdev_op_leaf) { nvlist_t **child; int c, idx; ASSERT(!vd->vdev_ishole); child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *), KM_SLEEP); for (c = 0, idx = 0; c < vd->vdev_children; c++) { vdev_t *cvd = vd->vdev_child[c]; /* * If we're generating an nvlist of removing * vdevs then skip over any device which is * not being removed. */ if ((flags & VDEV_CONFIG_REMOVING) && !cvd->vdev_removing) continue; child[idx++] = vdev_config_generate(spa, cvd, getstats, flags); } if (idx) { fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, child, idx); } for (c = 0; c < idx; c++) nvlist_free(child[c]); kmem_free(child, vd->vdev_children * sizeof (nvlist_t *)); } else { const char *aux = NULL; if (vd->vdev_offline && !vd->vdev_tmpoffline) fnvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, B_TRUE); if (vd->vdev_resilver_txg != 0) fnvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG, vd->vdev_resilver_txg); if (vd->vdev_faulted) fnvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, B_TRUE); if (vd->vdev_degraded) fnvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, B_TRUE); if (vd->vdev_removed) fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, B_TRUE); if (vd->vdev_unspare) fnvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, B_TRUE); if (vd->vdev_ishole) fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, B_TRUE); switch (vd->vdev_stat.vs_aux) { case VDEV_AUX_ERR_EXCEEDED: aux = "err_exceeded"; break; case VDEV_AUX_EXTERNAL: aux = "external"; break; } if (aux != NULL) fnvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, aux); if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) { fnvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID, vd->vdev_orig_guid); } } return (nv); } /* * Generate a view of the top-level vdevs. If we currently have holes * in the namespace, then generate an array which contains a list of holey * vdevs. Additionally, add the number of top-level children that currently * exist. */ void vdev_top_config_generate(spa_t *spa, nvlist_t *config) { vdev_t *rvd = spa->spa_root_vdev; uint64_t *array; uint_t c, idx; array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP); for (c = 0, idx = 0; c < rvd->vdev_children; c++) { vdev_t *tvd = rvd->vdev_child[c]; if (tvd->vdev_ishole) array[idx++] = c; } if (idx) { VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY, array, idx) == 0); } VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, rvd->vdev_children) == 0); kmem_free(array, rvd->vdev_children * sizeof (uint64_t)); } /* * Returns the configuration from the label of the given vdev. For vdevs * which don't have a txg value stored on their label (i.e. spares/cache) * or have not been completely initialized (txg = 0) just return * the configuration from the first valid label we find. Otherwise, * find the most up-to-date label that does not exceed the specified * 'txg' value. */ nvlist_t * vdev_label_read_config(vdev_t *vd, uint64_t txg) { spa_t *spa = vd->vdev_spa; nvlist_t *config = NULL; vdev_phys_t *vp; zio_t *zio; uint64_t best_txg = 0; int error = 0; int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE; ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); if (!vdev_readable(vd)) return (NULL); vp = zio_buf_alloc(sizeof (vdev_phys_t)); retry: for (int l = 0; l < VDEV_LABELS; l++) { nvlist_t *label = NULL; zio = zio_root(spa, NULL, NULL, flags); vdev_label_read(zio, vd, l, vp, offsetof(vdev_label_t, vl_vdev_phys), sizeof (vdev_phys_t), NULL, NULL, flags); if (zio_wait(zio) == 0 && nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist), &label, 0) == 0) { uint64_t label_txg = 0; /* * Auxiliary vdevs won't have txg values in their * labels and newly added vdevs may not have been * completely initialized so just return the * configuration from the first valid label we * encounter. */ error = nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, &label_txg); if ((error || label_txg == 0) && !config) { config = label; break; } else if (label_txg <= txg && label_txg > best_txg) { best_txg = label_txg; nvlist_free(config); config = fnvlist_dup(label); } } if (label != NULL) { nvlist_free(label); label = NULL; } } if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) { flags |= ZIO_FLAG_TRYHARD; goto retry; } zio_buf_free(vp, sizeof (vdev_phys_t)); return (config); } /* * Determine if a device is in use. The 'spare_guid' parameter will be filled * in with the device guid if this spare is active elsewhere on the system. */ static boolean_t vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason, uint64_t *spare_guid, uint64_t *l2cache_guid) { spa_t *spa = vd->vdev_spa; uint64_t state, pool_guid, device_guid, txg, spare_pool; uint64_t vdtxg = 0; nvlist_t *label; if (spare_guid) *spare_guid = 0ULL; if (l2cache_guid) *l2cache_guid = 0ULL; /* * Read the label, if any, and perform some basic sanity checks. */ if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) return (B_FALSE); (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG, &vdtxg); if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0 || nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &device_guid) != 0) { nvlist_free(label); return (B_FALSE); } if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, &pool_guid) != 0 || nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, &txg) != 0)) { nvlist_free(label); return (B_FALSE); } nvlist_free(label); /* * Check to see if this device indeed belongs to the pool it claims to * be a part of. The only way this is allowed is if the device is a hot * spare (which we check for later on). */ if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && !spa_guid_exists(pool_guid, device_guid) && !spa_spare_exists(device_guid, NULL, NULL) && !spa_l2cache_exists(device_guid, NULL)) return (B_FALSE); /* * If the transaction group is zero, then this an initialized (but * unused) label. This is only an error if the create transaction * on-disk is the same as the one we're using now, in which case the * user has attempted to add the same vdev multiple times in the same * transaction. */ if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE && txg == 0 && vdtxg == crtxg) return (B_TRUE); /* * Check to see if this is a spare device. We do an explicit check for * spa_has_spare() here because it may be on our pending list of spares * to add. We also check if it is an l2cache device. */ if (spa_spare_exists(device_guid, &spare_pool, NULL) || spa_has_spare(spa, device_guid)) { if (spare_guid) *spare_guid = device_guid; switch (reason) { case VDEV_LABEL_CREATE: case VDEV_LABEL_L2CACHE: return (B_TRUE); case VDEV_LABEL_REPLACE: return (!spa_has_spare(spa, device_guid) || spare_pool != 0ULL); case VDEV_LABEL_SPARE: return (spa_has_spare(spa, device_guid)); } } /* * Check to see if this is an l2cache device. */ if (spa_l2cache_exists(device_guid, NULL)) return (B_TRUE); /* * We can't rely on a pool's state if it's been imported * read-only. Instead we look to see if the pools is marked * read-only in the namespace and set the state to active. */ if ((spa = spa_by_guid(pool_guid, device_guid)) != NULL && spa_mode(spa) == FREAD) state = POOL_STATE_ACTIVE; /* * If the device is marked ACTIVE, then this device is in use by another * pool on the system. */ return (state == POOL_STATE_ACTIVE); } /* * Initialize a vdev label. We check to make sure each leaf device is not in * use, and writable. We put down an initial label which we will later * overwrite with a complete label. Note that it's important to do this * sequentially, not in parallel, so that we catch cases of multiple use of the * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with * itself. */ int vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason) { spa_t *spa = vd->vdev_spa; nvlist_t *label; vdev_phys_t *vp; char *pad2; uberblock_t *ub; zio_t *zio; char *buf; size_t buflen; int error; uint64_t spare_guid, l2cache_guid; int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); for (int c = 0; c < vd->vdev_children; c++) if ((error = vdev_label_init(vd->vdev_child[c], crtxg, reason)) != 0) return (error); /* Track the creation time for this vdev */ vd->vdev_crtxg = crtxg; if (!vd->vdev_ops->vdev_op_leaf || !spa_writeable(spa)) return (0); /* * Dead vdevs cannot be initialized. */ if (vdev_is_dead(vd)) return (SET_ERROR(EIO)); /* * Determine if the vdev is in use. */ if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT && vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid)) return (SET_ERROR(EBUSY)); /* * If this is a request to add or replace a spare or l2cache device * that is in use elsewhere on the system, then we must update the * guid (which was initialized to a random value) to reflect the * actual GUID (which is shared between multiple pools). */ if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE && spare_guid != 0ULL) { uint64_t guid_delta = spare_guid - vd->vdev_guid; vd->vdev_guid += guid_delta; for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) pvd->vdev_guid_sum += guid_delta; /* * If this is a replacement, then we want to fallthrough to the * rest of the code. If we're adding a spare, then it's already * labeled appropriately and we can just return. */ if (reason == VDEV_LABEL_SPARE) return (0); ASSERT(reason == VDEV_LABEL_REPLACE || reason == VDEV_LABEL_SPLIT); } if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE && l2cache_guid != 0ULL) { uint64_t guid_delta = l2cache_guid - vd->vdev_guid; vd->vdev_guid += guid_delta; for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent) pvd->vdev_guid_sum += guid_delta; /* * If this is a replacement, then we want to fallthrough to the * rest of the code. If we're adding an l2cache, then it's * already labeled appropriately and we can just return. */ if (reason == VDEV_LABEL_L2CACHE) return (0); ASSERT(reason == VDEV_LABEL_REPLACE); } /* * TRIM the whole thing so that we start with a clean slate. * It's just an optimization, so we don't care if it fails. * Don't TRIM if removing so that we don't interfere with zpool * disaster recovery. */ if (zfs_trim_enabled && vdev_trim_on_init && (reason == VDEV_LABEL_CREATE || reason == VDEV_LABEL_SPARE || reason == VDEV_LABEL_L2CACHE)) zio_wait(zio_trim(NULL, spa, vd, 0, vd->vdev_psize)); /* * Initialize its label. */ vp = zio_buf_alloc(sizeof (vdev_phys_t)); bzero(vp, sizeof (vdev_phys_t)); /* * Generate a label describing the pool and our top-level vdev. * We mark it as being from txg 0 to indicate that it's not * really part of an active pool just yet. The labels will * be written again with a meaningful txg by spa_sync(). */ if (reason == VDEV_LABEL_SPARE || (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) { /* * For inactive hot spares, we generate a special label that * identifies as a mutually shared hot spare. We write the * label if we are adding a hot spare, or if we are removing an * active hot spare (in which case we want to revert the * labels). */ VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, spa_version(spa)) == 0); VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, POOL_STATE_SPARE) == 0); VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, vd->vdev_guid) == 0); } else if (reason == VDEV_LABEL_L2CACHE || (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) { /* * For level 2 ARC devices, add a special label. */ VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0); VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION, spa_version(spa)) == 0); VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, POOL_STATE_L2CACHE) == 0); VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID, vd->vdev_guid) == 0); } else { uint64_t txg = 0ULL; if (reason == VDEV_LABEL_SPLIT) txg = spa->spa_uberblock.ub_txg; label = spa_config_generate(spa, vd, txg, B_FALSE); /* * Add our creation time. This allows us to detect multiple * vdev uses as described above, and automatically expires if we * fail. */ VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG, crtxg) == 0); } buf = vp->vp_nvlist; buflen = sizeof (vp->vp_nvlist); error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP); if (error != 0) { nvlist_free(label); zio_buf_free(vp, sizeof (vdev_phys_t)); /* EFAULT means nvlist_pack ran out of room */ return (error == EFAULT ? ENAMETOOLONG : EINVAL); } /* * Initialize uberblock template. */ ub = zio_buf_alloc(VDEV_UBERBLOCK_RING); bzero(ub, VDEV_UBERBLOCK_RING); *ub = spa->spa_uberblock; ub->ub_txg = 0; /* Initialize the 2nd padding area. */ pad2 = zio_buf_alloc(VDEV_PAD_SIZE); bzero(pad2, VDEV_PAD_SIZE); /* * Write everything in parallel. */ retry: zio = zio_root(spa, NULL, NULL, flags); for (int l = 0; l < VDEV_LABELS; l++) { vdev_label_write(zio, vd, l, vp, offsetof(vdev_label_t, vl_vdev_phys), sizeof (vdev_phys_t), NULL, NULL, flags); /* * Skip the 1st padding area. * Zero out the 2nd padding area where it might have * left over data from previous filesystem format. */ vdev_label_write(zio, vd, l, pad2, offsetof(vdev_label_t, vl_pad2), VDEV_PAD_SIZE, NULL, NULL, flags); vdev_label_write(zio, vd, l, ub, offsetof(vdev_label_t, vl_uberblock), VDEV_UBERBLOCK_RING, NULL, NULL, flags); } error = zio_wait(zio); if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) { flags |= ZIO_FLAG_TRYHARD; goto retry; } nvlist_free(label); zio_buf_free(pad2, VDEV_PAD_SIZE); zio_buf_free(ub, VDEV_UBERBLOCK_RING); zio_buf_free(vp, sizeof (vdev_phys_t)); /* * If this vdev hasn't been previously identified as a spare, then we * mark it as such only if a) we are labeling it as a spare, or b) it * exists as a spare elsewhere in the system. Do the same for * level 2 ARC devices. */ if (error == 0 && !vd->vdev_isspare && (reason == VDEV_LABEL_SPARE || spa_spare_exists(vd->vdev_guid, NULL, NULL))) spa_spare_add(vd); if (error == 0 && !vd->vdev_isl2cache && (reason == VDEV_LABEL_L2CACHE || spa_l2cache_exists(vd->vdev_guid, NULL))) spa_l2cache_add(vd); return (error); } /* * ========================================================================== * uberblock load/sync * ========================================================================== */ /* * Consider the following situation: txg is safely synced to disk. We've * written the first uberblock for txg + 1, and then we lose power. When we * come back up, we fail to see the uberblock for txg + 1 because, say, * it was on a mirrored device and the replica to which we wrote txg + 1 * is now offline. If we then make some changes and sync txg + 1, and then * the missing replica comes back, then for a few seconds we'll have two * conflicting uberblocks on disk with the same txg. The solution is simple: * among uberblocks with equal txg, choose the one with the latest timestamp. */ static int vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2) { if (ub1->ub_txg < ub2->ub_txg) return (-1); if (ub1->ub_txg > ub2->ub_txg) return (1); if (ub1->ub_timestamp < ub2->ub_timestamp) return (-1); if (ub1->ub_timestamp > ub2->ub_timestamp) return (1); return (0); } struct ubl_cbdata { uberblock_t *ubl_ubbest; /* Best uberblock */ vdev_t *ubl_vd; /* vdev associated with the above */ }; static void vdev_uberblock_load_done(zio_t *zio) { vdev_t *vd = zio->io_vd; spa_t *spa = zio->io_spa; zio_t *rio = zio->io_private; uberblock_t *ub = zio->io_data; struct ubl_cbdata *cbp = rio->io_private; ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd)); if (zio->io_error == 0 && uberblock_verify(ub) == 0) { mutex_enter(&rio->io_lock); if (ub->ub_txg <= spa->spa_load_max_txg && vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) { /* * Keep track of the vdev in which this uberblock * was found. We will use this information later * to obtain the config nvlist associated with * this uberblock. */ *cbp->ubl_ubbest = *ub; cbp->ubl_vd = vd; } mutex_exit(&rio->io_lock); } zio_buf_free(zio->io_data, zio->io_size); } static void vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags, struct ubl_cbdata *cbp) { for (int c = 0; c < vd->vdev_children; c++) vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp); if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { for (int l = 0; l < VDEV_LABELS; l++) { for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) { vdev_label_read(zio, vd, l, zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)), VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd), vdev_uberblock_load_done, zio, flags); } } } } /* * Reads the 'best' uberblock from disk along with its associated * configuration. First, we read the uberblock array of each label of each * vdev, keeping track of the uberblock with the highest txg in each array. * Then, we read the configuration from the same vdev as the best uberblock. */ void vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config) { zio_t *zio; spa_t *spa = rvd->vdev_spa; struct ubl_cbdata cb; int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD; ASSERT(ub); ASSERT(config); bzero(ub, sizeof (uberblock_t)); *config = NULL; cb.ubl_ubbest = ub; cb.ubl_vd = NULL; spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); zio = zio_root(spa, NULL, &cb, flags); vdev_uberblock_load_impl(zio, rvd, flags, &cb); (void) zio_wait(zio); /* * It's possible that the best uberblock was discovered on a label * that has a configuration which was written in a future txg. * Search all labels on this vdev to find the configuration that * matches the txg for our uberblock. */ if (cb.ubl_vd != NULL) *config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg); spa_config_exit(spa, SCL_ALL, FTAG); } /* * On success, increment root zio's count of good writes. * We only get credit for writes to known-visible vdevs; see spa_vdev_add(). */ static void vdev_uberblock_sync_done(zio_t *zio) { uint64_t *good_writes = zio->io_private; if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0) - atomic_add_64(good_writes, 1); + atomic_inc_64(good_writes); } /* * Write the uberblock to all labels of all leaves of the specified vdev. */ static void vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, int flags) { uberblock_t *ubbuf; int n; for (int c = 0; c < vd->vdev_children; c++) vdev_uberblock_sync(zio, ub, vd->vdev_child[c], flags); if (!vd->vdev_ops->vdev_op_leaf) return; if (!vdev_writeable(vd)) return; n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1); ubbuf = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)); bzero(ubbuf, VDEV_UBERBLOCK_SIZE(vd)); *ubbuf = *ub; for (int l = 0; l < VDEV_LABELS; l++) vdev_label_write(zio, vd, l, ubbuf, VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd), vdev_uberblock_sync_done, zio->io_private, flags | ZIO_FLAG_DONT_PROPAGATE); zio_buf_free(ubbuf, VDEV_UBERBLOCK_SIZE(vd)); } /* Sync the uberblocks to all vdevs in svd[] */ int vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags) { spa_t *spa = svd[0]->vdev_spa; zio_t *zio; uint64_t good_writes = 0; zio = zio_root(spa, NULL, &good_writes, flags); for (int v = 0; v < svdcount; v++) vdev_uberblock_sync(zio, ub, svd[v], flags); (void) zio_wait(zio); /* * Flush the uberblocks to disk. This ensures that the odd labels * are no longer needed (because the new uberblocks and the even * labels are safely on disk), so it is safe to overwrite them. */ zio = zio_root(spa, NULL, NULL, flags); for (int v = 0; v < svdcount; v++) zio_flush(zio, svd[v]); (void) zio_wait(zio); return (good_writes >= 1 ? 0 : EIO); } /* * On success, increment the count of good writes for our top-level vdev. */ static void vdev_label_sync_done(zio_t *zio) { uint64_t *good_writes = zio->io_private; if (zio->io_error == 0) - atomic_add_64(good_writes, 1); + atomic_inc_64(good_writes); } /* * If there weren't enough good writes, indicate failure to the parent. */ static void vdev_label_sync_top_done(zio_t *zio) { uint64_t *good_writes = zio->io_private; if (*good_writes == 0) zio->io_error = SET_ERROR(EIO); kmem_free(good_writes, sizeof (uint64_t)); } /* * We ignore errors for log and cache devices, simply free the private data. */ static void vdev_label_sync_ignore_done(zio_t *zio) { kmem_free(zio->io_private, sizeof (uint64_t)); } /* * Write all even or odd labels to all leaves of the specified vdev. */ static void vdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg, int flags) { nvlist_t *label; vdev_phys_t *vp; char *buf; size_t buflen; for (int c = 0; c < vd->vdev_children; c++) vdev_label_sync(zio, vd->vdev_child[c], l, txg, flags); if (!vd->vdev_ops->vdev_op_leaf) return; if (!vdev_writeable(vd)) return; /* * Generate a label describing the top-level config to which we belong. */ label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE); vp = zio_buf_alloc(sizeof (vdev_phys_t)); bzero(vp, sizeof (vdev_phys_t)); buf = vp->vp_nvlist; buflen = sizeof (vp->vp_nvlist); if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) { for (; l < VDEV_LABELS; l += 2) { vdev_label_write(zio, vd, l, vp, offsetof(vdev_label_t, vl_vdev_phys), sizeof (vdev_phys_t), vdev_label_sync_done, zio->io_private, flags | ZIO_FLAG_DONT_PROPAGATE); } } zio_buf_free(vp, sizeof (vdev_phys_t)); nvlist_free(label); } int vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags) { list_t *dl = &spa->spa_config_dirty_list; vdev_t *vd; zio_t *zio; int error; /* * Write the new labels to disk. */ zio = zio_root(spa, NULL, NULL, flags); for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) { uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t), KM_SLEEP); ASSERT(!vd->vdev_ishole); zio_t *vio = zio_null(zio, spa, NULL, (vd->vdev_islog || vd->vdev_aux != NULL) ? vdev_label_sync_ignore_done : vdev_label_sync_top_done, good_writes, flags); vdev_label_sync(vio, vd, l, txg, flags); zio_nowait(vio); } error = zio_wait(zio); /* * Flush the new labels to disk. */ zio = zio_root(spa, NULL, NULL, flags); for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) zio_flush(zio, vd); (void) zio_wait(zio); return (error); } /* * Sync the uberblock and any changes to the vdev configuration. * * The order of operations is carefully crafted to ensure that * if the system panics or loses power at any time, the state on disk * is still transactionally consistent. The in-line comments below * describe the failure semantics at each stage. * * Moreover, vdev_config_sync() is designed to be idempotent: if it fails * at any time, you can just call it again, and it will resume its work. */ int vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg, boolean_t tryhard) { spa_t *spa = svd[0]->vdev_spa; uberblock_t *ub = &spa->spa_uberblock; vdev_t *vd; zio_t *zio; int error; int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL; /* * Normally, we don't want to try too hard to write every label and * uberblock. If there is a flaky disk, we don't want the rest of the * sync process to block while we retry. But if we can't write a * single label out, we should retry with ZIO_FLAG_TRYHARD before * bailing out and declaring the pool faulted. */ if (tryhard) flags |= ZIO_FLAG_TRYHARD; ASSERT(ub->ub_txg <= txg); /* * If this isn't a resync due to I/O errors, * and nothing changed in this transaction group, * and the vdev configuration hasn't changed, * then there's nothing to do. */ if (ub->ub_txg < txg && uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE && list_is_empty(&spa->spa_config_dirty_list)) return (0); if (txg > spa_freeze_txg(spa)) return (0); ASSERT(txg <= spa->spa_final_txg); /* * Flush the write cache of every disk that's been written to * in this transaction group. This ensures that all blocks * written in this txg will be committed to stable storage * before any uberblock that references them. */ zio = zio_root(spa, NULL, NULL, flags); for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd; vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg))) zio_flush(zio, vd); (void) zio_wait(zio); /* * Sync out the even labels (L0, L2) for every dirty vdev. If the * system dies in the middle of this process, that's OK: all of the * even labels that made it to disk will be newer than any uberblock, * and will therefore be considered invalid. The odd labels (L1, L3), * which have not yet been touched, will still be valid. We flush * the new labels to disk to ensure that all even-label updates * are committed to stable storage before the uberblock update. */ if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0) return (error); /* * Sync the uberblocks to all vdevs in svd[]. * If the system dies in the middle of this step, there are two cases * to consider, and the on-disk state is consistent either way: * * (1) If none of the new uberblocks made it to disk, then the * previous uberblock will be the newest, and the odd labels * (which had not yet been touched) will be valid with respect * to that uberblock. * * (2) If one or more new uberblocks made it to disk, then they * will be the newest, and the even labels (which had all * been successfully committed) will be valid with respect * to the new uberblocks. */ if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0) return (error); /* * Sync out odd labels for every dirty vdev. If the system dies * in the middle of this process, the even labels and the new * uberblocks will suffice to open the pool. The next time * the pool is opened, the first thing we'll do -- before any * user data is modified -- is mark every vdev dirty so that * all labels will be brought up to date. We flush the new labels * to disk to ensure that all odd-label updates are committed to * stable storage before the next transaction group begins. */ if ((error = vdev_label_sync_list(spa, 1, txg, flags)) != 0) return (error); trim_thread_wakeup(spa); return (0); } Index: stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_vfsops.c =================================================================== --- stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_vfsops.c (revision 271000) +++ stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_vfsops.c (revision 271001) @@ -1,2543 +1,2543 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 2011 Pawel Jakub Dawidek . * All rights reserved. * Copyright (c) 2013 by Delphix. All rights reserved. */ /* Portions Copyright 2010 Robert Milkowski */ #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 "zfs_comutil.h" struct mtx zfs_debug_mtx; MTX_SYSINIT(zfs_debug_mtx, &zfs_debug_mtx, "zfs_debug", MTX_DEF); SYSCTL_NODE(_vfs, OID_AUTO, zfs, CTLFLAG_RW, 0, "ZFS file system"); int zfs_super_owner; SYSCTL_INT(_vfs_zfs, OID_AUTO, super_owner, CTLFLAG_RW, &zfs_super_owner, 0, "File system owner can perform privileged operation on his file systems"); int zfs_debug_level; TUNABLE_INT("vfs.zfs.debug", &zfs_debug_level); SYSCTL_INT(_vfs_zfs, OID_AUTO, debug, CTLFLAG_RW, &zfs_debug_level, 0, "Debug level"); SYSCTL_NODE(_vfs_zfs, OID_AUTO, version, CTLFLAG_RD, 0, "ZFS versions"); static int zfs_version_acl = ZFS_ACL_VERSION; SYSCTL_INT(_vfs_zfs_version, OID_AUTO, acl, CTLFLAG_RD, &zfs_version_acl, 0, "ZFS_ACL_VERSION"); static int zfs_version_spa = SPA_VERSION; SYSCTL_INT(_vfs_zfs_version, OID_AUTO, spa, CTLFLAG_RD, &zfs_version_spa, 0, "SPA_VERSION"); static int zfs_version_zpl = ZPL_VERSION; SYSCTL_INT(_vfs_zfs_version, OID_AUTO, zpl, CTLFLAG_RD, &zfs_version_zpl, 0, "ZPL_VERSION"); static int zfs_mount(vfs_t *vfsp); static int zfs_umount(vfs_t *vfsp, int fflag); static int zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp); static int zfs_statfs(vfs_t *vfsp, struct statfs *statp); static int zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp); static int zfs_sync(vfs_t *vfsp, int waitfor); static int zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp, struct ucred **credanonp, int *numsecflavors, int **secflavors); static int zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp); static void zfs_objset_close(zfsvfs_t *zfsvfs); static void zfs_freevfs(vfs_t *vfsp); static struct vfsops zfs_vfsops = { .vfs_mount = zfs_mount, .vfs_unmount = zfs_umount, .vfs_root = zfs_root, .vfs_statfs = zfs_statfs, .vfs_vget = zfs_vget, .vfs_sync = zfs_sync, .vfs_checkexp = zfs_checkexp, .vfs_fhtovp = zfs_fhtovp, }; VFS_SET(zfs_vfsops, zfs, VFCF_JAIL | VFCF_DELEGADMIN); /* * We need to keep a count of active fs's. * This is necessary to prevent our module * from being unloaded after a umount -f */ static uint32_t zfs_active_fs_count = 0; /*ARGSUSED*/ static int zfs_sync(vfs_t *vfsp, int waitfor) { /* * Data integrity is job one. We don't want a compromised kernel * writing to the storage pool, so we never sync during panic. */ if (panicstr) return (0); if (vfsp != NULL) { /* * Sync a specific filesystem. */ zfsvfs_t *zfsvfs = vfsp->vfs_data; dsl_pool_t *dp; int error; error = vfs_stdsync(vfsp, waitfor); if (error != 0) return (error); ZFS_ENTER(zfsvfs); dp = dmu_objset_pool(zfsvfs->z_os); /* * If the system is shutting down, then skip any * filesystems which may exist on a suspended pool. */ if (sys_shutdown && spa_suspended(dp->dp_spa)) { ZFS_EXIT(zfsvfs); return (0); } if (zfsvfs->z_log != NULL) zil_commit(zfsvfs->z_log, 0); ZFS_EXIT(zfsvfs); } else { /* * Sync all ZFS filesystems. This is what happens when you * run sync(1M). Unlike other filesystems, ZFS honors the * request by waiting for all pools to commit all dirty data. */ spa_sync_allpools(); } return (0); } #ifndef __FreeBSD_kernel__ static int zfs_create_unique_device(dev_t *dev) { major_t new_major; do { ASSERT3U(zfs_minor, <=, MAXMIN32); minor_t start = zfs_minor; do { mutex_enter(&zfs_dev_mtx); if (zfs_minor >= MAXMIN32) { /* * If we're still using the real major * keep out of /dev/zfs and /dev/zvol minor * number space. If we're using a getudev()'ed * major number, we can use all of its minors. */ if (zfs_major == ddi_name_to_major(ZFS_DRIVER)) zfs_minor = ZFS_MIN_MINOR; else zfs_minor = 0; } else { zfs_minor++; } *dev = makedevice(zfs_major, zfs_minor); mutex_exit(&zfs_dev_mtx); } while (vfs_devismounted(*dev) && zfs_minor != start); if (zfs_minor == start) { /* * We are using all ~262,000 minor numbers for the * current major number. Create a new major number. */ if ((new_major = getudev()) == (major_t)-1) { cmn_err(CE_WARN, "zfs_mount: Can't get unique major " "device number."); return (-1); } mutex_enter(&zfs_dev_mtx); zfs_major = new_major; zfs_minor = 0; mutex_exit(&zfs_dev_mtx); } else { break; } /* CONSTANTCONDITION */ } while (1); return (0); } #endif /* !__FreeBSD_kernel__ */ static void atime_changed_cb(void *arg, uint64_t newval) { zfsvfs_t *zfsvfs = arg; if (newval == TRUE) { zfsvfs->z_atime = TRUE; zfsvfs->z_vfs->vfs_flag &= ~MNT_NOATIME; vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME); vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0); } else { zfsvfs->z_atime = FALSE; zfsvfs->z_vfs->vfs_flag |= MNT_NOATIME; vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME); vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0); } } static void xattr_changed_cb(void *arg, uint64_t newval) { zfsvfs_t *zfsvfs = arg; if (newval == TRUE) { /* XXX locking on vfs_flag? */ #ifdef TODO zfsvfs->z_vfs->vfs_flag |= VFS_XATTR; #endif vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR); vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_XATTR, NULL, 0); } else { /* XXX locking on vfs_flag? */ #ifdef TODO zfsvfs->z_vfs->vfs_flag &= ~VFS_XATTR; #endif vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_XATTR); vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR, NULL, 0); } } static void blksz_changed_cb(void *arg, uint64_t newval) { zfsvfs_t *zfsvfs = arg; if (newval < SPA_MINBLOCKSIZE || newval > SPA_MAXBLOCKSIZE || !ISP2(newval)) newval = SPA_MAXBLOCKSIZE; zfsvfs->z_max_blksz = newval; zfsvfs->z_vfs->mnt_stat.f_iosize = newval; } static void readonly_changed_cb(void *arg, uint64_t newval) { zfsvfs_t *zfsvfs = arg; if (newval) { /* XXX locking on vfs_flag? */ zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY; vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW); vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0); } else { /* XXX locking on vfs_flag? */ zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY; vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO); vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0); } } static void setuid_changed_cb(void *arg, uint64_t newval) { zfsvfs_t *zfsvfs = arg; if (newval == FALSE) { zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID; vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID); vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0); } else { zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID; vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID); vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0); } } static void exec_changed_cb(void *arg, uint64_t newval) { zfsvfs_t *zfsvfs = arg; if (newval == FALSE) { zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC; vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC); vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0); } else { zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC; vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC); vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0); } } /* * The nbmand mount option can be changed at mount time. * We can't allow it to be toggled on live file systems or incorrect * behavior may be seen from cifs clients * * This property isn't registered via dsl_prop_register(), but this callback * will be called when a file system is first mounted */ static void nbmand_changed_cb(void *arg, uint64_t newval) { zfsvfs_t *zfsvfs = arg; if (newval == FALSE) { vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND); vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0); } else { vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND); vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0); } } static void snapdir_changed_cb(void *arg, uint64_t newval) { zfsvfs_t *zfsvfs = arg; zfsvfs->z_show_ctldir = newval; } static void vscan_changed_cb(void *arg, uint64_t newval) { zfsvfs_t *zfsvfs = arg; zfsvfs->z_vscan = newval; } static void acl_mode_changed_cb(void *arg, uint64_t newval) { zfsvfs_t *zfsvfs = arg; zfsvfs->z_acl_mode = newval; } static void acl_inherit_changed_cb(void *arg, uint64_t newval) { zfsvfs_t *zfsvfs = arg; zfsvfs->z_acl_inherit = newval; } static int zfs_register_callbacks(vfs_t *vfsp) { struct dsl_dataset *ds = NULL; objset_t *os = NULL; zfsvfs_t *zfsvfs = NULL; uint64_t nbmand; boolean_t readonly = B_FALSE; boolean_t do_readonly = B_FALSE; boolean_t setuid = B_FALSE; boolean_t do_setuid = B_FALSE; boolean_t exec = B_FALSE; boolean_t do_exec = B_FALSE; #ifdef illumos boolean_t devices = B_FALSE; boolean_t do_devices = B_FALSE; #endif boolean_t xattr = B_FALSE; boolean_t do_xattr = B_FALSE; boolean_t atime = B_FALSE; boolean_t do_atime = B_FALSE; int error = 0; ASSERT(vfsp); zfsvfs = vfsp->vfs_data; ASSERT(zfsvfs); os = zfsvfs->z_os; /* * This function can be called for a snapshot when we update snapshot's * mount point, which isn't really supported. */ if (dmu_objset_is_snapshot(os)) return (EOPNOTSUPP); /* * The act of registering our callbacks will destroy any mount * options we may have. In order to enable temporary overrides * of mount options, we stash away the current values and * restore them after we register the callbacks. */ if (vfs_optionisset(vfsp, MNTOPT_RO, NULL) || !spa_writeable(dmu_objset_spa(os))) { readonly = B_TRUE; do_readonly = B_TRUE; } else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) { readonly = B_FALSE; do_readonly = B_TRUE; } if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) { setuid = B_FALSE; do_setuid = B_TRUE; } else { if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) { setuid = B_FALSE; do_setuid = B_TRUE; } else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) { setuid = B_TRUE; do_setuid = B_TRUE; } } if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) { exec = B_FALSE; do_exec = B_TRUE; } else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) { exec = B_TRUE; do_exec = B_TRUE; } if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) { xattr = B_FALSE; do_xattr = B_TRUE; } else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) { xattr = B_TRUE; do_xattr = B_TRUE; } if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) { atime = B_FALSE; do_atime = B_TRUE; } else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) { atime = B_TRUE; do_atime = B_TRUE; } /* * nbmand is a special property. It can only be changed at * mount time. * * This is weird, but it is documented to only be changeable * at mount time. */ if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) { nbmand = B_FALSE; } else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) { nbmand = B_TRUE; } else { char osname[MAXNAMELEN]; dmu_objset_name(os, osname); if (error = dsl_prop_get_integer(osname, "nbmand", &nbmand, NULL)) { return (error); } } /* * Register property callbacks. * * It would probably be fine to just check for i/o error from * the first prop_register(), but I guess I like to go * overboard... */ ds = dmu_objset_ds(os); dsl_pool_config_enter(dmu_objset_pool(os), FTAG); error = dsl_prop_register(ds, zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs); error = error ? error : dsl_prop_register(ds, zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs); error = error ? error : dsl_prop_register(ds, zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs); error = error ? error : dsl_prop_register(ds, zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs); #ifdef illumos error = error ? error : dsl_prop_register(ds, zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs); #endif error = error ? error : dsl_prop_register(ds, zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs); error = error ? error : dsl_prop_register(ds, zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs); error = error ? error : dsl_prop_register(ds, zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs); error = error ? error : dsl_prop_register(ds, zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs); error = error ? error : dsl_prop_register(ds, zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb, zfsvfs); error = error ? error : dsl_prop_register(ds, zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zfsvfs); dsl_pool_config_exit(dmu_objset_pool(os), FTAG); if (error) goto unregister; /* * Invoke our callbacks to restore temporary mount options. */ if (do_readonly) readonly_changed_cb(zfsvfs, readonly); if (do_setuid) setuid_changed_cb(zfsvfs, setuid); if (do_exec) exec_changed_cb(zfsvfs, exec); if (do_xattr) xattr_changed_cb(zfsvfs, xattr); if (do_atime) atime_changed_cb(zfsvfs, atime); nbmand_changed_cb(zfsvfs, nbmand); return (0); unregister: /* * We may attempt to unregister some callbacks that are not * registered, but this is OK; it will simply return ENOMSG, * which we will ignore. */ (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs); (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs); (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs); (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs); #ifdef illumos (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs); #endif (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs); (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs); (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs); (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs); (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb, zfsvfs); (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zfsvfs); return (error); } static int zfs_space_delta_cb(dmu_object_type_t bonustype, void *data, uint64_t *userp, uint64_t *groupp) { /* * Is it a valid type of object to track? */ if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA) return (SET_ERROR(ENOENT)); /* * If we have a NULL data pointer * then assume the id's aren't changing and * return EEXIST to the dmu to let it know to * use the same ids */ if (data == NULL) return (SET_ERROR(EEXIST)); if (bonustype == DMU_OT_ZNODE) { znode_phys_t *znp = data; *userp = znp->zp_uid; *groupp = znp->zp_gid; } else { int hdrsize; sa_hdr_phys_t *sap = data; sa_hdr_phys_t sa = *sap; boolean_t swap = B_FALSE; ASSERT(bonustype == DMU_OT_SA); if (sa.sa_magic == 0) { /* * This should only happen for newly created * files that haven't had the znode data filled * in yet. */ *userp = 0; *groupp = 0; return (0); } if (sa.sa_magic == BSWAP_32(SA_MAGIC)) { sa.sa_magic = SA_MAGIC; sa.sa_layout_info = BSWAP_16(sa.sa_layout_info); swap = B_TRUE; } else { VERIFY3U(sa.sa_magic, ==, SA_MAGIC); } hdrsize = sa_hdrsize(&sa); VERIFY3U(hdrsize, >=, sizeof (sa_hdr_phys_t)); *userp = *((uint64_t *)((uintptr_t)data + hdrsize + SA_UID_OFFSET)); *groupp = *((uint64_t *)((uintptr_t)data + hdrsize + SA_GID_OFFSET)); if (swap) { *userp = BSWAP_64(*userp); *groupp = BSWAP_64(*groupp); } } return (0); } static void fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr, char *domainbuf, int buflen, uid_t *ridp) { uint64_t fuid; const char *domain; fuid = strtonum(fuidstr, NULL); domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid)); if (domain) (void) strlcpy(domainbuf, domain, buflen); else domainbuf[0] = '\0'; *ridp = FUID_RID(fuid); } static uint64_t zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type) { switch (type) { case ZFS_PROP_USERUSED: return (DMU_USERUSED_OBJECT); case ZFS_PROP_GROUPUSED: return (DMU_GROUPUSED_OBJECT); case ZFS_PROP_USERQUOTA: return (zfsvfs->z_userquota_obj); case ZFS_PROP_GROUPQUOTA: return (zfsvfs->z_groupquota_obj); } return (0); } int zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type, uint64_t *cookiep, void *vbuf, uint64_t *bufsizep) { int error; zap_cursor_t zc; zap_attribute_t za; zfs_useracct_t *buf = vbuf; uint64_t obj; if (!dmu_objset_userspace_present(zfsvfs->z_os)) return (SET_ERROR(ENOTSUP)); obj = zfs_userquota_prop_to_obj(zfsvfs, type); if (obj == 0) { *bufsizep = 0; return (0); } for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep); (error = zap_cursor_retrieve(&zc, &za)) == 0; zap_cursor_advance(&zc)) { if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) > *bufsizep) break; fuidstr_to_sid(zfsvfs, za.za_name, buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid); buf->zu_space = za.za_first_integer; buf++; } if (error == ENOENT) error = 0; ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep); *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf; *cookiep = zap_cursor_serialize(&zc); zap_cursor_fini(&zc); return (error); } /* * buf must be big enough (eg, 32 bytes) */ static int id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid, char *buf, boolean_t addok) { uint64_t fuid; int domainid = 0; if (domain && domain[0]) { domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok); if (domainid == -1) return (SET_ERROR(ENOENT)); } fuid = FUID_ENCODE(domainid, rid); (void) sprintf(buf, "%llx", (longlong_t)fuid); return (0); } int zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type, const char *domain, uint64_t rid, uint64_t *valp) { char buf[32]; int err; uint64_t obj; *valp = 0; if (!dmu_objset_userspace_present(zfsvfs->z_os)) return (SET_ERROR(ENOTSUP)); obj = zfs_userquota_prop_to_obj(zfsvfs, type); if (obj == 0) return (0); err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_FALSE); if (err) return (err); err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp); if (err == ENOENT) err = 0; return (err); } int zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type, const char *domain, uint64_t rid, uint64_t quota) { char buf[32]; int err; dmu_tx_t *tx; uint64_t *objp; boolean_t fuid_dirtied; if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA) return (SET_ERROR(EINVAL)); if (zfsvfs->z_version < ZPL_VERSION_USERSPACE) return (SET_ERROR(ENOTSUP)); objp = (type == ZFS_PROP_USERQUOTA) ? &zfsvfs->z_userquota_obj : &zfsvfs->z_groupquota_obj; err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE); if (err) return (err); fuid_dirtied = zfsvfs->z_fuid_dirty; tx = dmu_tx_create(zfsvfs->z_os); dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL); if (*objp == 0) { dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE, zfs_userquota_prop_prefixes[type]); } if (fuid_dirtied) zfs_fuid_txhold(zfsvfs, tx); err = dmu_tx_assign(tx, TXG_WAIT); if (err) { dmu_tx_abort(tx); return (err); } mutex_enter(&zfsvfs->z_lock); if (*objp == 0) { *objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA, DMU_OT_NONE, 0, tx); VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ, zfs_userquota_prop_prefixes[type], 8, 1, objp, tx)); } mutex_exit(&zfsvfs->z_lock); if (quota == 0) { err = zap_remove(zfsvfs->z_os, *objp, buf, tx); if (err == ENOENT) err = 0; } else { err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, "a, tx); } ASSERT(err == 0); if (fuid_dirtied) zfs_fuid_sync(zfsvfs, tx); dmu_tx_commit(tx); return (err); } boolean_t zfs_fuid_overquota(zfsvfs_t *zfsvfs, boolean_t isgroup, uint64_t fuid) { char buf[32]; uint64_t used, quota, usedobj, quotaobj; int err; usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT; quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj; if (quotaobj == 0 || zfsvfs->z_replay) return (B_FALSE); (void) sprintf(buf, "%llx", (longlong_t)fuid); err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, "a); if (err != 0) return (B_FALSE); err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used); if (err != 0) return (B_FALSE); return (used >= quota); } boolean_t zfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup) { uint64_t fuid; uint64_t quotaobj; quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj; fuid = isgroup ? zp->z_gid : zp->z_uid; if (quotaobj == 0 || zfsvfs->z_replay) return (B_FALSE); return (zfs_fuid_overquota(zfsvfs, isgroup, fuid)); } int zfsvfs_create(const char *osname, zfsvfs_t **zfvp) { objset_t *os; zfsvfs_t *zfsvfs; uint64_t zval; int i, error; uint64_t sa_obj; zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP); /* * We claim to always be readonly so we can open snapshots; * other ZPL code will prevent us from writing to snapshots. */ error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os); if (error) { kmem_free(zfsvfs, sizeof (zfsvfs_t)); return (error); } /* * Initialize the zfs-specific filesystem structure. * Should probably make this a kmem cache, shuffle fields, * and just bzero up to z_hold_mtx[]. */ zfsvfs->z_vfs = NULL; zfsvfs->z_parent = zfsvfs; zfsvfs->z_max_blksz = SPA_MAXBLOCKSIZE; zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE; zfsvfs->z_os = os; error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version); if (error) { goto out; } else if (zfsvfs->z_version > zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) { (void) printf("Can't mount a version %lld file system " "on a version %lld pool\n. Pool must be upgraded to mount " "this file system.", (u_longlong_t)zfsvfs->z_version, (u_longlong_t)spa_version(dmu_objset_spa(os))); error = SET_ERROR(ENOTSUP); goto out; } if ((error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &zval)) != 0) goto out; zfsvfs->z_norm = (int)zval; if ((error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &zval)) != 0) goto out; zfsvfs->z_utf8 = (zval != 0); if ((error = zfs_get_zplprop(os, ZFS_PROP_CASE, &zval)) != 0) goto out; zfsvfs->z_case = (uint_t)zval; /* * Fold case on file systems that are always or sometimes case * insensitive. */ if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE || zfsvfs->z_case == ZFS_CASE_MIXED) zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER; zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os); zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os); if (zfsvfs->z_use_sa) { /* should either have both of these objects or none */ error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj); if (error) return (error); } else { /* * Pre SA versions file systems should never touch * either the attribute registration or layout objects. */ sa_obj = 0; } error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END, &zfsvfs->z_attr_table); if (error) goto out; if (zfsvfs->z_version >= ZPL_VERSION_SA) sa_register_update_callback(os, zfs_sa_upgrade); error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1, &zfsvfs->z_root); if (error) goto out; ASSERT(zfsvfs->z_root != 0); error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1, &zfsvfs->z_unlinkedobj); if (error) goto out; error = zap_lookup(os, MASTER_NODE_OBJ, zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA], 8, 1, &zfsvfs->z_userquota_obj); if (error && error != ENOENT) goto out; error = zap_lookup(os, MASTER_NODE_OBJ, zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA], 8, 1, &zfsvfs->z_groupquota_obj); if (error && error != ENOENT) goto out; error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1, &zfsvfs->z_fuid_obj); if (error && error != ENOENT) goto out; error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1, &zfsvfs->z_shares_dir); if (error && error != ENOENT) goto out; mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL); mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL); list_create(&zfsvfs->z_all_znodes, sizeof (znode_t), offsetof(znode_t, z_link_node)); rrm_init(&zfsvfs->z_teardown_lock, B_FALSE); rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL); rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL); for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL); *zfvp = zfsvfs; return (0); out: dmu_objset_disown(os, zfsvfs); *zfvp = NULL; kmem_free(zfsvfs, sizeof (zfsvfs_t)); return (error); } static int zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting) { int error; error = zfs_register_callbacks(zfsvfs->z_vfs); if (error) return (error); /* * Set the objset user_ptr to track its zfsvfs. */ mutex_enter(&zfsvfs->z_os->os_user_ptr_lock); dmu_objset_set_user(zfsvfs->z_os, zfsvfs); mutex_exit(&zfsvfs->z_os->os_user_ptr_lock); zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data); /* * If we are not mounting (ie: online recv), then we don't * have to worry about replaying the log as we blocked all * operations out since we closed the ZIL. */ if (mounting) { boolean_t readonly; /* * During replay we remove the read only flag to * allow replays to succeed. */ readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY; if (readonly != 0) zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY; else zfs_unlinked_drain(zfsvfs); /* * Parse and replay the intent log. * * Because of ziltest, this must be done after * zfs_unlinked_drain(). (Further note: ziltest * doesn't use readonly mounts, where * zfs_unlinked_drain() isn't called.) This is because * ziltest causes spa_sync() to think it's committed, * but actually it is not, so the intent log contains * many txg's worth of changes. * * In particular, if object N is in the unlinked set in * the last txg to actually sync, then it could be * actually freed in a later txg and then reallocated * in a yet later txg. This would write a "create * object N" record to the intent log. Normally, this * would be fine because the spa_sync() would have * written out the fact that object N is free, before * we could write the "create object N" intent log * record. * * But when we are in ziltest mode, we advance the "open * txg" without actually spa_sync()-ing the changes to * disk. So we would see that object N is still * allocated and in the unlinked set, and there is an * intent log record saying to allocate it. */ if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) { if (zil_replay_disable) { zil_destroy(zfsvfs->z_log, B_FALSE); } else { zfsvfs->z_replay = B_TRUE; zil_replay(zfsvfs->z_os, zfsvfs, zfs_replay_vector); zfsvfs->z_replay = B_FALSE; } } zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */ } return (0); } extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */ void zfsvfs_free(zfsvfs_t *zfsvfs) { int i; /* * This is a barrier to prevent the filesystem from going away in * zfs_znode_move() until we can safely ensure that the filesystem is * not unmounted. We consider the filesystem valid before the barrier * and invalid after the barrier. */ rw_enter(&zfsvfs_lock, RW_READER); rw_exit(&zfsvfs_lock); zfs_fuid_destroy(zfsvfs); mutex_destroy(&zfsvfs->z_znodes_lock); mutex_destroy(&zfsvfs->z_lock); list_destroy(&zfsvfs->z_all_znodes); rrm_destroy(&zfsvfs->z_teardown_lock); rw_destroy(&zfsvfs->z_teardown_inactive_lock); rw_destroy(&zfsvfs->z_fuid_lock); for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) mutex_destroy(&zfsvfs->z_hold_mtx[i]); kmem_free(zfsvfs, sizeof (zfsvfs_t)); } static void zfs_set_fuid_feature(zfsvfs_t *zfsvfs) { zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os); if (zfsvfs->z_vfs) { if (zfsvfs->z_use_fuids) { vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR); vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS); vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS); vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE); vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER); vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE); } else { vfs_clear_feature(zfsvfs->z_vfs, VFSFT_XVATTR); vfs_clear_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS); vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS); vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE); vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER); vfs_clear_feature(zfsvfs->z_vfs, VFSFT_REPARSE); } } zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os); } static int zfs_domount(vfs_t *vfsp, char *osname) { uint64_t recordsize, fsid_guid; int error = 0; zfsvfs_t *zfsvfs; vnode_t *vp; ASSERT(vfsp); ASSERT(osname); error = zfsvfs_create(osname, &zfsvfs); if (error) return (error); zfsvfs->z_vfs = vfsp; #ifdef illumos /* Initialize the generic filesystem structure. */ vfsp->vfs_bcount = 0; vfsp->vfs_data = NULL; if (zfs_create_unique_device(&mount_dev) == -1) { error = SET_ERROR(ENODEV); goto out; } ASSERT(vfs_devismounted(mount_dev) == 0); #endif if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize, NULL)) goto out; zfsvfs->z_vfs->vfs_bsize = SPA_MINBLOCKSIZE; zfsvfs->z_vfs->mnt_stat.f_iosize = recordsize; vfsp->vfs_data = zfsvfs; vfsp->mnt_flag |= MNT_LOCAL; vfsp->mnt_kern_flag |= MNTK_LOOKUP_SHARED; vfsp->mnt_kern_flag |= MNTK_SHARED_WRITES; vfsp->mnt_kern_flag |= MNTK_EXTENDED_SHARED; /* * The fsid is 64 bits, composed of an 8-bit fs type, which * separates our fsid from any other filesystem types, and a * 56-bit objset unique ID. The objset unique ID is unique to * all objsets open on this system, provided by unique_create(). * The 8-bit fs type must be put in the low bits of fsid[1] * because that's where other Solaris filesystems put it. */ fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os); ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0); vfsp->vfs_fsid.val[0] = fsid_guid; vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) | vfsp->mnt_vfc->vfc_typenum & 0xFF; /* * Set features for file system. */ zfs_set_fuid_feature(zfsvfs); if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) { vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS); vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE); vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE); } else if (zfsvfs->z_case == ZFS_CASE_MIXED) { vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS); vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE); } vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED); if (dmu_objset_is_snapshot(zfsvfs->z_os)) { uint64_t pval; atime_changed_cb(zfsvfs, B_FALSE); readonly_changed_cb(zfsvfs, B_TRUE); if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL)) goto out; xattr_changed_cb(zfsvfs, pval); zfsvfs->z_issnap = B_TRUE; zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED; mutex_enter(&zfsvfs->z_os->os_user_ptr_lock); dmu_objset_set_user(zfsvfs->z_os, zfsvfs); mutex_exit(&zfsvfs->z_os->os_user_ptr_lock); } else { error = zfsvfs_setup(zfsvfs, B_TRUE); } vfs_mountedfrom(vfsp, osname); /* Grab extra reference. */ VERIFY(VFS_ROOT(vfsp, LK_EXCLUSIVE, &vp) == 0); VOP_UNLOCK(vp, 0); if (!zfsvfs->z_issnap) zfsctl_create(zfsvfs); out: if (error) { dmu_objset_disown(zfsvfs->z_os, zfsvfs); zfsvfs_free(zfsvfs); } else { - atomic_add_32(&zfs_active_fs_count, 1); + atomic_inc_32(&zfs_active_fs_count); } return (error); } void zfs_unregister_callbacks(zfsvfs_t *zfsvfs) { objset_t *os = zfsvfs->z_os; struct dsl_dataset *ds; /* * Unregister properties. */ if (!dmu_objset_is_snapshot(os)) { ds = dmu_objset_ds(os); VERIFY(dsl_prop_unregister(ds, "atime", atime_changed_cb, zfsvfs) == 0); VERIFY(dsl_prop_unregister(ds, "xattr", xattr_changed_cb, zfsvfs) == 0); VERIFY(dsl_prop_unregister(ds, "recordsize", blksz_changed_cb, zfsvfs) == 0); VERIFY(dsl_prop_unregister(ds, "readonly", readonly_changed_cb, zfsvfs) == 0); VERIFY(dsl_prop_unregister(ds, "setuid", setuid_changed_cb, zfsvfs) == 0); VERIFY(dsl_prop_unregister(ds, "exec", exec_changed_cb, zfsvfs) == 0); VERIFY(dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb, zfsvfs) == 0); VERIFY(dsl_prop_unregister(ds, "aclmode", acl_mode_changed_cb, zfsvfs) == 0); VERIFY(dsl_prop_unregister(ds, "aclinherit", acl_inherit_changed_cb, zfsvfs) == 0); VERIFY(dsl_prop_unregister(ds, "vscan", vscan_changed_cb, zfsvfs) == 0); } } #ifdef SECLABEL /* * Convert a decimal digit string to a uint64_t integer. */ static int str_to_uint64(char *str, uint64_t *objnum) { uint64_t num = 0; while (*str) { if (*str < '0' || *str > '9') return (SET_ERROR(EINVAL)); num = num*10 + *str++ - '0'; } *objnum = num; return (0); } /* * The boot path passed from the boot loader is in the form of * "rootpool-name/root-filesystem-object-number'. Convert this * string to a dataset name: "rootpool-name/root-filesystem-name". */ static int zfs_parse_bootfs(char *bpath, char *outpath) { char *slashp; uint64_t objnum; int error; if (*bpath == 0 || *bpath == '/') return (SET_ERROR(EINVAL)); (void) strcpy(outpath, bpath); slashp = strchr(bpath, '/'); /* if no '/', just return the pool name */ if (slashp == NULL) { return (0); } /* if not a number, just return the root dataset name */ if (str_to_uint64(slashp+1, &objnum)) { return (0); } *slashp = '\0'; error = dsl_dsobj_to_dsname(bpath, objnum, outpath); *slashp = '/'; return (error); } /* * Check that the hex label string is appropriate for the dataset being * mounted into the global_zone proper. * * Return an error if the hex label string is not default or * admin_low/admin_high. For admin_low labels, the corresponding * dataset must be readonly. */ int zfs_check_global_label(const char *dsname, const char *hexsl) { if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0) return (0); if (strcasecmp(hexsl, ADMIN_HIGH) == 0) return (0); if (strcasecmp(hexsl, ADMIN_LOW) == 0) { /* must be readonly */ uint64_t rdonly; if (dsl_prop_get_integer(dsname, zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL)) return (SET_ERROR(EACCES)); return (rdonly ? 0 : EACCES); } return (SET_ERROR(EACCES)); } /* * Determine whether the mount is allowed according to MAC check. * by comparing (where appropriate) label of the dataset against * the label of the zone being mounted into. If the dataset has * no label, create one. * * Returns 0 if access allowed, error otherwise (e.g. EACCES) */ static int zfs_mount_label_policy(vfs_t *vfsp, char *osname) { int error, retv; zone_t *mntzone = NULL; ts_label_t *mnt_tsl; bslabel_t *mnt_sl; bslabel_t ds_sl; char ds_hexsl[MAXNAMELEN]; retv = EACCES; /* assume the worst */ /* * Start by getting the dataset label if it exists. */ error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL), 1, sizeof (ds_hexsl), &ds_hexsl, NULL); if (error) return (SET_ERROR(EACCES)); /* * If labeling is NOT enabled, then disallow the mount of datasets * which have a non-default label already. No other label checks * are needed. */ if (!is_system_labeled()) { if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) return (0); return (SET_ERROR(EACCES)); } /* * Get the label of the mountpoint. If mounting into the global * zone (i.e. mountpoint is not within an active zone and the * zoned property is off), the label must be default or * admin_low/admin_high only; no other checks are needed. */ mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE); if (mntzone->zone_id == GLOBAL_ZONEID) { uint64_t zoned; zone_rele(mntzone); if (dsl_prop_get_integer(osname, zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL)) return (SET_ERROR(EACCES)); if (!zoned) return (zfs_check_global_label(osname, ds_hexsl)); else /* * This is the case of a zone dataset being mounted * initially, before the zone has been fully created; * allow this mount into global zone. */ return (0); } mnt_tsl = mntzone->zone_slabel; ASSERT(mnt_tsl != NULL); label_hold(mnt_tsl); mnt_sl = label2bslabel(mnt_tsl); if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) { /* * The dataset doesn't have a real label, so fabricate one. */ char *str = NULL; if (l_to_str_internal(mnt_sl, &str) == 0 && dsl_prop_set_string(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL), ZPROP_SRC_LOCAL, str) == 0) retv = 0; if (str != NULL) kmem_free(str, strlen(str) + 1); } else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) { /* * Now compare labels to complete the MAC check. If the * labels are equal then allow access. If the mountpoint * label dominates the dataset label, allow readonly access. * Otherwise, access is denied. */ if (blequal(mnt_sl, &ds_sl)) retv = 0; else if (bldominates(mnt_sl, &ds_sl)) { vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0); retv = 0; } } label_rele(mnt_tsl); zone_rele(mntzone); return (retv); } #endif /* SECLABEL */ #ifdef OPENSOLARIS_MOUNTROOT static int zfs_mountroot(vfs_t *vfsp, enum whymountroot why) { int error = 0; static int zfsrootdone = 0; zfsvfs_t *zfsvfs = NULL; znode_t *zp = NULL; vnode_t *vp = NULL; char *zfs_bootfs; char *zfs_devid; ASSERT(vfsp); /* * The filesystem that we mount as root is defined in the * boot property "zfs-bootfs" with a format of * "poolname/root-dataset-objnum". */ if (why == ROOT_INIT) { if (zfsrootdone++) return (SET_ERROR(EBUSY)); /* * the process of doing a spa_load will require the * clock to be set before we could (for example) do * something better by looking at the timestamp on * an uberblock, so just set it to -1. */ clkset(-1); if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) { cmn_err(CE_NOTE, "spa_get_bootfs: can not get " "bootfs name"); return (SET_ERROR(EINVAL)); } zfs_devid = spa_get_bootprop("diskdevid"); error = spa_import_rootpool(rootfs.bo_name, zfs_devid); if (zfs_devid) spa_free_bootprop(zfs_devid); if (error) { spa_free_bootprop(zfs_bootfs); cmn_err(CE_NOTE, "spa_import_rootpool: error %d", error); return (error); } if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) { spa_free_bootprop(zfs_bootfs); cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d", error); return (error); } spa_free_bootprop(zfs_bootfs); if (error = vfs_lock(vfsp)) return (error); if (error = zfs_domount(vfsp, rootfs.bo_name)) { cmn_err(CE_NOTE, "zfs_domount: error %d", error); goto out; } zfsvfs = (zfsvfs_t *)vfsp->vfs_data; ASSERT(zfsvfs); if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) { cmn_err(CE_NOTE, "zfs_zget: error %d", error); goto out; } vp = ZTOV(zp); mutex_enter(&vp->v_lock); vp->v_flag |= VROOT; mutex_exit(&vp->v_lock); rootvp = vp; /* * Leave rootvp held. The root file system is never unmounted. */ vfs_add((struct vnode *)0, vfsp, (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0); out: vfs_unlock(vfsp); return (error); } else if (why == ROOT_REMOUNT) { readonly_changed_cb(vfsp->vfs_data, B_FALSE); vfsp->vfs_flag |= VFS_REMOUNT; /* refresh mount options */ zfs_unregister_callbacks(vfsp->vfs_data); return (zfs_register_callbacks(vfsp)); } else if (why == ROOT_UNMOUNT) { zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data); (void) zfs_sync(vfsp, 0, 0); return (0); } /* * if "why" is equal to anything else other than ROOT_INIT, * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it. */ return (SET_ERROR(ENOTSUP)); } #endif /* OPENSOLARIS_MOUNTROOT */ static int getpoolname(const char *osname, char *poolname) { char *p; p = strchr(osname, '/'); if (p == NULL) { if (strlen(osname) >= MAXNAMELEN) return (ENAMETOOLONG); (void) strcpy(poolname, osname); } else { if (p - osname >= MAXNAMELEN) return (ENAMETOOLONG); (void) strncpy(poolname, osname, p - osname); poolname[p - osname] = '\0'; } return (0); } /*ARGSUSED*/ static int zfs_mount(vfs_t *vfsp) { kthread_t *td = curthread; vnode_t *mvp = vfsp->mnt_vnodecovered; cred_t *cr = td->td_ucred; char *osname; int error = 0; int canwrite; #ifdef illumos if (mvp->v_type != VDIR) return (SET_ERROR(ENOTDIR)); mutex_enter(&mvp->v_lock); if ((uap->flags & MS_REMOUNT) == 0 && (uap->flags & MS_OVERLAY) == 0 && (mvp->v_count != 1 || (mvp->v_flag & VROOT))) { mutex_exit(&mvp->v_lock); return (SET_ERROR(EBUSY)); } mutex_exit(&mvp->v_lock); /* * ZFS does not support passing unparsed data in via MS_DATA. * Users should use the MS_OPTIONSTR interface; this means * that all option parsing is already done and the options struct * can be interrogated. */ if ((uap->flags & MS_DATA) && uap->datalen > 0) #else if (!prison_allow(td->td_ucred, PR_ALLOW_MOUNT_ZFS)) return (SET_ERROR(EPERM)); if (vfs_getopt(vfsp->mnt_optnew, "from", (void **)&osname, NULL)) return (SET_ERROR(EINVAL)); #endif /* ! illumos */ /* * If full-owner-access is enabled and delegated administration is * turned on, we must set nosuid. */ if (zfs_super_owner && dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != ECANCELED) { secpolicy_fs_mount_clearopts(cr, vfsp); } /* * Check for mount privilege? * * If we don't have privilege then see if * we have local permission to allow it */ error = secpolicy_fs_mount(cr, mvp, vfsp); if (error) { if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != 0) goto out; if (!(vfsp->vfs_flag & MS_REMOUNT)) { vattr_t vattr; /* * Make sure user is the owner of the mount point * or has sufficient privileges. */ vattr.va_mask = AT_UID; vn_lock(mvp, LK_SHARED | LK_RETRY); if (VOP_GETATTR(mvp, &vattr, cr)) { VOP_UNLOCK(mvp, 0); goto out; } if (secpolicy_vnode_owner(mvp, cr, vattr.va_uid) != 0 && VOP_ACCESS(mvp, VWRITE, cr, td) != 0) { VOP_UNLOCK(mvp, 0); goto out; } VOP_UNLOCK(mvp, 0); } secpolicy_fs_mount_clearopts(cr, vfsp); } /* * Refuse to mount a filesystem if we are in a local zone and the * dataset is not visible. */ if (!INGLOBALZONE(curthread) && (!zone_dataset_visible(osname, &canwrite) || !canwrite)) { error = SET_ERROR(EPERM); goto out; } #ifdef SECLABEL error = zfs_mount_label_policy(vfsp, osname); if (error) goto out; #endif vfsp->vfs_flag |= MNT_NFS4ACLS; /* * When doing a remount, we simply refresh our temporary properties * according to those options set in the current VFS options. */ if (vfsp->vfs_flag & MS_REMOUNT) { /* refresh mount options */ zfs_unregister_callbacks(vfsp->vfs_data); error = zfs_register_callbacks(vfsp); goto out; } /* Initial root mount: try hard to import the requested root pool. */ if ((vfsp->vfs_flag & MNT_ROOTFS) != 0 && (vfsp->vfs_flag & MNT_UPDATE) == 0) { char pname[MAXNAMELEN]; error = getpoolname(osname, pname); if (error == 0) error = spa_import_rootpool(pname); if (error) goto out; } DROP_GIANT(); error = zfs_domount(vfsp, osname); PICKUP_GIANT(); #ifdef sun /* * Add an extra VFS_HOLD on our parent vfs so that it can't * disappear due to a forced unmount. */ if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap) VFS_HOLD(mvp->v_vfsp); #endif /* sun */ out: return (error); } static int zfs_statfs(vfs_t *vfsp, struct statfs *statp) { zfsvfs_t *zfsvfs = vfsp->vfs_data; uint64_t refdbytes, availbytes, usedobjs, availobjs; statp->f_version = STATFS_VERSION; ZFS_ENTER(zfsvfs); dmu_objset_space(zfsvfs->z_os, &refdbytes, &availbytes, &usedobjs, &availobjs); /* * The underlying storage pool actually uses multiple block sizes. * We report the fragsize as the smallest block size we support, * and we report our blocksize as the filesystem's maximum blocksize. */ statp->f_bsize = SPA_MINBLOCKSIZE; statp->f_iosize = zfsvfs->z_vfs->mnt_stat.f_iosize; /* * The following report "total" blocks of various kinds in the * file system, but reported in terms of f_frsize - the * "fragment" size. */ statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT; statp->f_bfree = availbytes / statp->f_bsize; statp->f_bavail = statp->f_bfree; /* no root reservation */ /* * statvfs() should really be called statufs(), because it assumes * static metadata. ZFS doesn't preallocate files, so the best * we can do is report the max that could possibly fit in f_files, * and that minus the number actually used in f_ffree. * For f_ffree, report the smaller of the number of object available * and the number of blocks (each object will take at least a block). */ statp->f_ffree = MIN(availobjs, statp->f_bfree); statp->f_files = statp->f_ffree + usedobjs; /* * We're a zfs filesystem. */ (void) strlcpy(statp->f_fstypename, "zfs", sizeof(statp->f_fstypename)); strlcpy(statp->f_mntfromname, vfsp->mnt_stat.f_mntfromname, sizeof(statp->f_mntfromname)); strlcpy(statp->f_mntonname, vfsp->mnt_stat.f_mntonname, sizeof(statp->f_mntonname)); statp->f_namemax = ZFS_MAXNAMELEN; ZFS_EXIT(zfsvfs); return (0); } static int zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp) { zfsvfs_t *zfsvfs = vfsp->vfs_data; znode_t *rootzp; int error; ZFS_ENTER_NOERROR(zfsvfs); error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp); if (error == 0) *vpp = ZTOV(rootzp); ZFS_EXIT(zfsvfs); if (error == 0) { error = vn_lock(*vpp, flags); if (error == 0) (*vpp)->v_vflag |= VV_ROOT; } if (error != 0) *vpp = NULL; return (error); } /* * Teardown the zfsvfs::z_os. * * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock' * and 'z_teardown_inactive_lock' held. */ static int zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting) { znode_t *zp; rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG); if (!unmounting) { /* * We purge the parent filesystem's vfsp as the parent * filesystem and all of its snapshots have their vnode's * v_vfsp set to the parent's filesystem's vfsp. Note, * 'z_parent' is self referential for non-snapshots. */ (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0); #ifdef FREEBSD_NAMECACHE cache_purgevfs(zfsvfs->z_parent->z_vfs); #endif } /* * Close the zil. NB: Can't close the zil while zfs_inactive * threads are blocked as zil_close can call zfs_inactive. */ if (zfsvfs->z_log) { zil_close(zfsvfs->z_log); zfsvfs->z_log = NULL; } rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER); /* * If we are not unmounting (ie: online recv) and someone already * unmounted this file system while we were doing the switcheroo, * or a reopen of z_os failed then just bail out now. */ if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) { rw_exit(&zfsvfs->z_teardown_inactive_lock); rrm_exit(&zfsvfs->z_teardown_lock, FTAG); return (SET_ERROR(EIO)); } /* * At this point there are no vops active, and any new vops will * fail with EIO since we have z_teardown_lock for writer (only * relavent for forced unmount). * * Release all holds on dbufs. */ mutex_enter(&zfsvfs->z_znodes_lock); for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL; zp = list_next(&zfsvfs->z_all_znodes, zp)) if (zp->z_sa_hdl) { ASSERT(ZTOV(zp)->v_count >= 0); zfs_znode_dmu_fini(zp); } mutex_exit(&zfsvfs->z_znodes_lock); /* * If we are unmounting, set the unmounted flag and let new vops * unblock. zfs_inactive will have the unmounted behavior, and all * other vops will fail with EIO. */ if (unmounting) { zfsvfs->z_unmounted = B_TRUE; rrm_exit(&zfsvfs->z_teardown_lock, FTAG); rw_exit(&zfsvfs->z_teardown_inactive_lock); } /* * z_os will be NULL if there was an error in attempting to reopen * zfsvfs, so just return as the properties had already been * unregistered and cached data had been evicted before. */ if (zfsvfs->z_os == NULL) return (0); /* * Unregister properties. */ zfs_unregister_callbacks(zfsvfs); /* * Evict cached data */ if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs->z_os)) && !(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY)) txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0); dmu_objset_evict_dbufs(zfsvfs->z_os); return (0); } /*ARGSUSED*/ static int zfs_umount(vfs_t *vfsp, int fflag) { kthread_t *td = curthread; zfsvfs_t *zfsvfs = vfsp->vfs_data; objset_t *os; cred_t *cr = td->td_ucred; int ret; ret = secpolicy_fs_unmount(cr, vfsp); if (ret) { if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource), ZFS_DELEG_PERM_MOUNT, cr)) return (ret); } /* * We purge the parent filesystem's vfsp as the parent filesystem * and all of its snapshots have their vnode's v_vfsp set to the * parent's filesystem's vfsp. Note, 'z_parent' is self * referential for non-snapshots. */ (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0); /* * Unmount any snapshots mounted under .zfs before unmounting the * dataset itself. */ if (zfsvfs->z_ctldir != NULL) { if ((ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0) return (ret); ret = vflush(vfsp, 0, 0, td); ASSERT(ret == EBUSY); if (!(fflag & MS_FORCE)) { if (zfsvfs->z_ctldir->v_count > 1) return (EBUSY); ASSERT(zfsvfs->z_ctldir->v_count == 1); } zfsctl_destroy(zfsvfs); ASSERT(zfsvfs->z_ctldir == NULL); } if (fflag & MS_FORCE) { /* * Mark file system as unmounted before calling * vflush(FORCECLOSE). This way we ensure no future vnops * will be called and risk operating on DOOMED vnodes. */ rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG); zfsvfs->z_unmounted = B_TRUE; rrm_exit(&zfsvfs->z_teardown_lock, FTAG); } /* * Flush all the files. */ ret = vflush(vfsp, 1, (fflag & MS_FORCE) ? FORCECLOSE : 0, td); if (ret != 0) { if (!zfsvfs->z_issnap) { zfsctl_create(zfsvfs); ASSERT(zfsvfs->z_ctldir != NULL); } return (ret); } #ifdef sun if (!(fflag & MS_FORCE)) { /* * Check the number of active vnodes in the file system. * Our count is maintained in the vfs structure, but the * number is off by 1 to indicate a hold on the vfs * structure itself. * * The '.zfs' directory maintains a reference of its * own, and any active references underneath are * reflected in the vnode count. */ if (zfsvfs->z_ctldir == NULL) { if (vfsp->vfs_count > 1) return (SET_ERROR(EBUSY)); } else { if (vfsp->vfs_count > 2 || zfsvfs->z_ctldir->v_count > 1) return (SET_ERROR(EBUSY)); } } #endif VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0); os = zfsvfs->z_os; /* * z_os will be NULL if there was an error in * attempting to reopen zfsvfs. */ if (os != NULL) { /* * Unset the objset user_ptr. */ mutex_enter(&os->os_user_ptr_lock); dmu_objset_set_user(os, NULL); mutex_exit(&os->os_user_ptr_lock); /* * Finally release the objset */ dmu_objset_disown(os, zfsvfs); } /* * We can now safely destroy the '.zfs' directory node. */ if (zfsvfs->z_ctldir != NULL) zfsctl_destroy(zfsvfs); if (zfsvfs->z_issnap) { vnode_t *svp = vfsp->mnt_vnodecovered; if (svp->v_count >= 2) VN_RELE(svp); } zfs_freevfs(vfsp); return (0); } static int zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp) { zfsvfs_t *zfsvfs = vfsp->vfs_data; znode_t *zp; int err; /* * zfs_zget() can't operate on virtual entries like .zfs/ or * .zfs/snapshot/ directories, that's why we return EOPNOTSUPP. * This will make NFS to switch to LOOKUP instead of using VGET. */ if (ino == ZFSCTL_INO_ROOT || ino == ZFSCTL_INO_SNAPDIR || (zfsvfs->z_shares_dir != 0 && ino == zfsvfs->z_shares_dir)) return (EOPNOTSUPP); ZFS_ENTER(zfsvfs); err = zfs_zget(zfsvfs, ino, &zp); if (err == 0 && zp->z_unlinked) { VN_RELE(ZTOV(zp)); err = EINVAL; } if (err == 0) *vpp = ZTOV(zp); ZFS_EXIT(zfsvfs); if (err == 0) err = vn_lock(*vpp, flags); if (err != 0) *vpp = NULL; return (err); } static int zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp, struct ucred **credanonp, int *numsecflavors, int **secflavors) { zfsvfs_t *zfsvfs = vfsp->vfs_data; /* * If this is regular file system vfsp is the same as * zfsvfs->z_parent->z_vfs, but if it is snapshot, * zfsvfs->z_parent->z_vfs represents parent file system * which we have to use here, because only this file system * has mnt_export configured. */ return (vfs_stdcheckexp(zfsvfs->z_parent->z_vfs, nam, extflagsp, credanonp, numsecflavors, secflavors)); } CTASSERT(SHORT_FID_LEN <= sizeof(struct fid)); CTASSERT(LONG_FID_LEN <= sizeof(struct fid)); static int zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp) { zfsvfs_t *zfsvfs = vfsp->vfs_data; znode_t *zp; uint64_t object = 0; uint64_t fid_gen = 0; uint64_t gen_mask; uint64_t zp_gen; int i, err; *vpp = NULL; ZFS_ENTER(zfsvfs); /* * On FreeBSD we can get snapshot's mount point or its parent file * system mount point depending if snapshot is already mounted or not. */ if (zfsvfs->z_parent == zfsvfs && fidp->fid_len == LONG_FID_LEN) { zfid_long_t *zlfid = (zfid_long_t *)fidp; uint64_t objsetid = 0; uint64_t setgen = 0; for (i = 0; i < sizeof (zlfid->zf_setid); i++) objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i); for (i = 0; i < sizeof (zlfid->zf_setgen); i++) setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i); ZFS_EXIT(zfsvfs); err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs); if (err) return (SET_ERROR(EINVAL)); ZFS_ENTER(zfsvfs); } if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) { zfid_short_t *zfid = (zfid_short_t *)fidp; for (i = 0; i < sizeof (zfid->zf_object); i++) object |= ((uint64_t)zfid->zf_object[i]) << (8 * i); for (i = 0; i < sizeof (zfid->zf_gen); i++) fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i); } else { ZFS_EXIT(zfsvfs); return (SET_ERROR(EINVAL)); } /* * A zero fid_gen means we are in .zfs or the .zfs/snapshot * directory tree. If the object == zfsvfs->z_shares_dir, then * we are in the .zfs/shares directory tree. */ if ((fid_gen == 0 && (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) || (zfsvfs->z_shares_dir != 0 && object == zfsvfs->z_shares_dir)) { *vpp = zfsvfs->z_ctldir; ASSERT(*vpp != NULL); if (object == ZFSCTL_INO_SNAPDIR) { VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL, 0, NULL, NULL, NULL, NULL, NULL) == 0); } else if (object == zfsvfs->z_shares_dir) { VERIFY(zfsctl_root_lookup(*vpp, "shares", vpp, NULL, 0, NULL, NULL, NULL, NULL, NULL) == 0); } else { VN_HOLD(*vpp); } ZFS_EXIT(zfsvfs); err = vn_lock(*vpp, flags); if (err != 0) *vpp = NULL; return (err); } gen_mask = -1ULL >> (64 - 8 * i); dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask); if (err = zfs_zget(zfsvfs, object, &zp)) { ZFS_EXIT(zfsvfs); return (err); } (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen, sizeof (uint64_t)); zp_gen = zp_gen & gen_mask; if (zp_gen == 0) zp_gen = 1; if (zp->z_unlinked || zp_gen != fid_gen) { dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen); VN_RELE(ZTOV(zp)); ZFS_EXIT(zfsvfs); return (SET_ERROR(EINVAL)); } *vpp = ZTOV(zp); ZFS_EXIT(zfsvfs); err = vn_lock(*vpp, flags | LK_RETRY); if (err == 0) vnode_create_vobject(*vpp, zp->z_size, curthread); else *vpp = NULL; return (err); } /* * Block out VOPs and close zfsvfs_t::z_os * * Note, if successful, then we return with the 'z_teardown_lock' and * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying * dataset and objset intact so that they can be atomically handed off during * a subsequent rollback or recv operation and the resume thereafter. */ int zfs_suspend_fs(zfsvfs_t *zfsvfs) { int error; if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0) return (error); return (0); } /* * Rebuild SA and release VOPs. Note that ownership of the underlying dataset * is an invariant across any of the operations that can be performed while the * filesystem was suspended. Whether it succeeded or failed, the preconditions * are the same: the relevant objset and associated dataset are owned by * zfsvfs, held, and long held on entry. */ int zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname) { int err; znode_t *zp; uint64_t sa_obj = 0; ASSERT(RRM_WRITE_HELD(&zfsvfs->z_teardown_lock)); ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock)); /* * We already own this, so just hold and rele it to update the * objset_t, as the one we had before may have been evicted. */ VERIFY0(dmu_objset_hold(osname, zfsvfs, &zfsvfs->z_os)); VERIFY3P(zfsvfs->z_os->os_dsl_dataset->ds_owner, ==, zfsvfs); VERIFY(dsl_dataset_long_held(zfsvfs->z_os->os_dsl_dataset)); dmu_objset_rele(zfsvfs->z_os, zfsvfs); /* * Make sure version hasn't changed */ err = zfs_get_zplprop(zfsvfs->z_os, ZFS_PROP_VERSION, &zfsvfs->z_version); if (err) goto bail; err = zap_lookup(zfsvfs->z_os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj); if (err && zfsvfs->z_version >= ZPL_VERSION_SA) goto bail; if ((err = sa_setup(zfsvfs->z_os, sa_obj, zfs_attr_table, ZPL_END, &zfsvfs->z_attr_table)) != 0) goto bail; if (zfsvfs->z_version >= ZPL_VERSION_SA) sa_register_update_callback(zfsvfs->z_os, zfs_sa_upgrade); VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0); zfs_set_fuid_feature(zfsvfs); /* * Attempt to re-establish all the active znodes with * their dbufs. If a zfs_rezget() fails, then we'll let * any potential callers discover that via ZFS_ENTER_VERIFY_VP * when they try to use their znode. */ mutex_enter(&zfsvfs->z_znodes_lock); for (zp = list_head(&zfsvfs->z_all_znodes); zp; zp = list_next(&zfsvfs->z_all_znodes, zp)) { (void) zfs_rezget(zp); } mutex_exit(&zfsvfs->z_znodes_lock); bail: /* release the VOPs */ rw_exit(&zfsvfs->z_teardown_inactive_lock); rrm_exit(&zfsvfs->z_teardown_lock, FTAG); if (err) { /* * Since we couldn't setup the sa framework, try to force * unmount this file system. */ if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0) (void) dounmount(zfsvfs->z_vfs, MS_FORCE, curthread); } return (err); } static void zfs_freevfs(vfs_t *vfsp) { zfsvfs_t *zfsvfs = vfsp->vfs_data; #ifdef sun /* * If this is a snapshot, we have an extra VFS_HOLD on our parent * from zfs_mount(). Release it here. If we came through * zfs_mountroot() instead, we didn't grab an extra hold, so * skip the VFS_RELE for rootvfs. */ if (zfsvfs->z_issnap && (vfsp != rootvfs)) VFS_RELE(zfsvfs->z_parent->z_vfs); #endif /* sun */ zfsvfs_free(zfsvfs); - atomic_add_32(&zfs_active_fs_count, -1); + atomic_dec_32(&zfs_active_fs_count); } #ifdef __i386__ static int desiredvnodes_backup; #endif static void zfs_vnodes_adjust(void) { #ifdef __i386__ int newdesiredvnodes; desiredvnodes_backup = desiredvnodes; /* * We calculate newdesiredvnodes the same way it is done in * vntblinit(). If it is equal to desiredvnodes, it means that * it wasn't tuned by the administrator and we can tune it down. */ newdesiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * vm_kmem_size / (5 * (sizeof(struct vm_object) + sizeof(struct vnode)))); if (newdesiredvnodes == desiredvnodes) desiredvnodes = (3 * newdesiredvnodes) / 4; #endif } static void zfs_vnodes_adjust_back(void) { #ifdef __i386__ desiredvnodes = desiredvnodes_backup; #endif } void zfs_init(void) { printf("ZFS filesystem version: " ZPL_VERSION_STRING "\n"); /* * Initialize .zfs directory structures */ zfsctl_init(); /* * Initialize znode cache, vnode ops, etc... */ zfs_znode_init(); /* * Reduce number of vnodes. Originally number of vnodes is calculated * with UFS inode in mind. We reduce it here, because it's too big for * ZFS/i386. */ zfs_vnodes_adjust(); dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb); } void zfs_fini(void) { zfsctl_fini(); zfs_znode_fini(); zfs_vnodes_adjust_back(); } int zfs_busy(void) { return (zfs_active_fs_count != 0); } int zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers) { int error; objset_t *os = zfsvfs->z_os; dmu_tx_t *tx; if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION) return (SET_ERROR(EINVAL)); if (newvers < zfsvfs->z_version) return (SET_ERROR(EINVAL)); if (zfs_spa_version_map(newvers) > spa_version(dmu_objset_spa(zfsvfs->z_os))) return (SET_ERROR(ENOTSUP)); tx = dmu_tx_create(os); dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR); if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) { dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE, ZFS_SA_ATTRS); dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); } error = dmu_tx_assign(tx, TXG_WAIT); if (error) { dmu_tx_abort(tx); return (error); } error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR, 8, 1, &newvers, tx); if (error) { dmu_tx_commit(tx); return (error); } if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) { uint64_t sa_obj; ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=, SPA_VERSION_SA); sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE, DMU_OT_NONE, 0, tx); error = zap_add(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx); ASSERT0(error); VERIFY(0 == sa_set_sa_object(os, sa_obj)); sa_register_update_callback(os, zfs_sa_upgrade); } spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx, "from %llu to %llu", zfsvfs->z_version, newvers); dmu_tx_commit(tx); zfsvfs->z_version = newvers; zfs_set_fuid_feature(zfsvfs); return (0); } /* * Read a property stored within the master node. */ int zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value) { const char *pname; int error = ENOENT; /* * Look up the file system's value for the property. For the * version property, we look up a slightly different string. */ if (prop == ZFS_PROP_VERSION) pname = ZPL_VERSION_STR; else pname = zfs_prop_to_name(prop); if (os != NULL) error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value); if (error == ENOENT) { /* No value set, use the default value */ switch (prop) { case ZFS_PROP_VERSION: *value = ZPL_VERSION; break; case ZFS_PROP_NORMALIZE: case ZFS_PROP_UTF8ONLY: *value = 0; break; case ZFS_PROP_CASE: *value = ZFS_CASE_SENSITIVE; break; default: return (error); } error = 0; } return (error); } #ifdef _KERNEL void zfsvfs_update_fromname(const char *oldname, const char *newname) { char tmpbuf[MAXPATHLEN]; struct mount *mp; char *fromname; size_t oldlen; oldlen = strlen(oldname); mtx_lock(&mountlist_mtx); TAILQ_FOREACH(mp, &mountlist, mnt_list) { fromname = mp->mnt_stat.f_mntfromname; if (strcmp(fromname, oldname) == 0) { (void)strlcpy(fromname, newname, sizeof(mp->mnt_stat.f_mntfromname)); continue; } if (strncmp(fromname, oldname, oldlen) == 0 && (fromname[oldlen] == '/' || fromname[oldlen] == '@')) { (void)snprintf(tmpbuf, sizeof(tmpbuf), "%s%s", newname, fromname + oldlen); (void)strlcpy(fromname, tmpbuf, sizeof(mp->mnt_stat.f_mntfromname)); continue; } } mtx_unlock(&mountlist_mtx); } #endif Index: stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zio_inject.c =================================================================== --- stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zio_inject.c (revision 271000) +++ stable/10/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zio_inject.c (revision 271001) @@ -1,524 +1,524 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 2012, 2014 by Delphix. All rights reserved. */ /* * ZFS fault injection * * To handle fault injection, we keep track of a series of zinject_record_t * structures which describe which logical block(s) should be injected with a * fault. These are kept in a global list. Each record corresponds to a given * spa_t and maintains a special hold on the spa_t so that it cannot be deleted * or exported while the injection record exists. * * Device level injection is done using the 'zi_guid' field. If this is set, it * means that the error is destined for a particular device, not a piece of * data. * * This is a rather poor data structure and algorithm, but we don't expect more * than a few faults at any one time, so it should be sufficient for our needs. */ #include #include #include #include #include #include uint32_t zio_injection_enabled; typedef struct inject_handler { int zi_id; spa_t *zi_spa; zinject_record_t zi_record; list_node_t zi_link; } inject_handler_t; static list_t inject_handlers; static krwlock_t inject_lock; static int inject_next_id = 1; /* * Returns true if the given record matches the I/O in progress. */ static boolean_t zio_match_handler(zbookmark_phys_t *zb, uint64_t type, zinject_record_t *record, int error) { /* * Check for a match against the MOS, which is based on type */ if (zb->zb_objset == DMU_META_OBJSET && record->zi_objset == DMU_META_OBJSET && record->zi_object == DMU_META_DNODE_OBJECT) { if (record->zi_type == DMU_OT_NONE || type == record->zi_type) return (record->zi_freq == 0 || spa_get_random(100) < record->zi_freq); else return (B_FALSE); } /* * Check for an exact match. */ if (zb->zb_objset == record->zi_objset && zb->zb_object == record->zi_object && zb->zb_level == record->zi_level && zb->zb_blkid >= record->zi_start && zb->zb_blkid <= record->zi_end && error == record->zi_error) return (record->zi_freq == 0 || spa_get_random(100) < record->zi_freq); return (B_FALSE); } /* * Panic the system when a config change happens in the function * specified by tag. */ void zio_handle_panic_injection(spa_t *spa, char *tag, uint64_t type) { inject_handler_t *handler; rw_enter(&inject_lock, RW_READER); for (handler = list_head(&inject_handlers); handler != NULL; handler = list_next(&inject_handlers, handler)) { if (spa != handler->zi_spa) continue; if (handler->zi_record.zi_type == type && strcmp(tag, handler->zi_record.zi_func) == 0) panic("Panic requested in function %s\n", tag); } rw_exit(&inject_lock); } /* * Determine if the I/O in question should return failure. Returns the errno * to be returned to the caller. */ int zio_handle_fault_injection(zio_t *zio, int error) { int ret = 0; inject_handler_t *handler; /* * Ignore I/O not associated with any logical data. */ if (zio->io_logical == NULL) return (0); /* * Currently, we only support fault injection on reads. */ if (zio->io_type != ZIO_TYPE_READ) return (0); rw_enter(&inject_lock, RW_READER); for (handler = list_head(&inject_handlers); handler != NULL; handler = list_next(&inject_handlers, handler)) { if (zio->io_spa != handler->zi_spa || handler->zi_record.zi_cmd != ZINJECT_DATA_FAULT) continue; /* If this handler matches, return EIO */ if (zio_match_handler(&zio->io_logical->io_bookmark, zio->io_bp ? BP_GET_TYPE(zio->io_bp) : DMU_OT_NONE, &handler->zi_record, error)) { ret = error; break; } } rw_exit(&inject_lock); return (ret); } /* * Determine if the zio is part of a label update and has an injection * handler associated with that portion of the label. Currently, we * allow error injection in either the nvlist or the uberblock region of * of the vdev label. */ int zio_handle_label_injection(zio_t *zio, int error) { inject_handler_t *handler; vdev_t *vd = zio->io_vd; uint64_t offset = zio->io_offset; int label; int ret = 0; if (offset >= VDEV_LABEL_START_SIZE && offset < vd->vdev_psize - VDEV_LABEL_END_SIZE) return (0); rw_enter(&inject_lock, RW_READER); for (handler = list_head(&inject_handlers); handler != NULL; handler = list_next(&inject_handlers, handler)) { uint64_t start = handler->zi_record.zi_start; uint64_t end = handler->zi_record.zi_end; if (handler->zi_record.zi_cmd != ZINJECT_LABEL_FAULT) continue; /* * The injection region is the relative offsets within a * vdev label. We must determine the label which is being * updated and adjust our region accordingly. */ label = vdev_label_number(vd->vdev_psize, offset); start = vdev_label_offset(vd->vdev_psize, label, start); end = vdev_label_offset(vd->vdev_psize, label, end); if (zio->io_vd->vdev_guid == handler->zi_record.zi_guid && (offset >= start && offset <= end)) { ret = error; break; } } rw_exit(&inject_lock); return (ret); } int zio_handle_device_injection(vdev_t *vd, zio_t *zio, int error) { inject_handler_t *handler; int ret = 0; /* * We skip over faults in the labels unless it's during * device open (i.e. zio == NULL). */ if (zio != NULL) { uint64_t offset = zio->io_offset; if (offset < VDEV_LABEL_START_SIZE || offset >= vd->vdev_psize - VDEV_LABEL_END_SIZE) return (0); } rw_enter(&inject_lock, RW_READER); for (handler = list_head(&inject_handlers); handler != NULL; handler = list_next(&inject_handlers, handler)) { if (handler->zi_record.zi_cmd != ZINJECT_DEVICE_FAULT) continue; if (vd->vdev_guid == handler->zi_record.zi_guid) { if (handler->zi_record.zi_failfast && (zio == NULL || (zio->io_flags & (ZIO_FLAG_IO_RETRY | ZIO_FLAG_TRYHARD)))) { continue; } /* Handle type specific I/O failures */ if (zio != NULL && handler->zi_record.zi_iotype != ZIO_TYPES && handler->zi_record.zi_iotype != zio->io_type) continue; if (handler->zi_record.zi_error == error) { /* * For a failed open, pretend like the device * has gone away. */ if (error == ENXIO) vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED; /* * Treat these errors as if they had been * retried so that all the appropriate stats * and FMA events are generated. */ if (!handler->zi_record.zi_failfast && zio != NULL) zio->io_flags |= ZIO_FLAG_IO_RETRY; ret = error; break; } if (handler->zi_record.zi_error == ENXIO) { ret = SET_ERROR(EIO); break; } } } rw_exit(&inject_lock); return (ret); } /* * Simulate hardware that ignores cache flushes. For requested number * of seconds nix the actual writing to disk. */ void zio_handle_ignored_writes(zio_t *zio) { inject_handler_t *handler; rw_enter(&inject_lock, RW_READER); for (handler = list_head(&inject_handlers); handler != NULL; handler = list_next(&inject_handlers, handler)) { /* Ignore errors not destined for this pool */ if (zio->io_spa != handler->zi_spa || handler->zi_record.zi_cmd != ZINJECT_IGNORED_WRITES) continue; /* * Positive duration implies # of seconds, negative * a number of txgs */ if (handler->zi_record.zi_timer == 0) { if (handler->zi_record.zi_duration > 0) handler->zi_record.zi_timer = ddi_get_lbolt64(); else handler->zi_record.zi_timer = zio->io_txg; } /* Have a "problem" writing 60% of the time */ if (spa_get_random(100) < 60) zio->io_pipeline &= ~ZIO_VDEV_IO_STAGES; break; } rw_exit(&inject_lock); } void spa_handle_ignored_writes(spa_t *spa) { inject_handler_t *handler; if (zio_injection_enabled == 0) return; rw_enter(&inject_lock, RW_READER); for (handler = list_head(&inject_handlers); handler != NULL; handler = list_next(&inject_handlers, handler)) { if (spa != handler->zi_spa || handler->zi_record.zi_cmd != ZINJECT_IGNORED_WRITES) continue; if (handler->zi_record.zi_duration > 0) { VERIFY(handler->zi_record.zi_timer == 0 || handler->zi_record.zi_timer + handler->zi_record.zi_duration * hz > ddi_get_lbolt64()); } else { /* duration is negative so the subtraction here adds */ VERIFY(handler->zi_record.zi_timer == 0 || handler->zi_record.zi_timer - handler->zi_record.zi_duration >= spa_syncing_txg(spa)); } } rw_exit(&inject_lock); } uint64_t zio_handle_io_delay(zio_t *zio) { vdev_t *vd = zio->io_vd; inject_handler_t *handler; uint64_t seconds = 0; if (zio_injection_enabled == 0) return (0); rw_enter(&inject_lock, RW_READER); for (handler = list_head(&inject_handlers); handler != NULL; handler = list_next(&inject_handlers, handler)) { if (handler->zi_record.zi_cmd != ZINJECT_DELAY_IO) continue; if (vd->vdev_guid == handler->zi_record.zi_guid) { seconds = handler->zi_record.zi_timer; break; } } rw_exit(&inject_lock); return (seconds); } /* * Create a new handler for the given record. We add it to the list, adding * a reference to the spa_t in the process. We increment zio_injection_enabled, * which is the switch to trigger all fault injection. */ int zio_inject_fault(char *name, int flags, int *id, zinject_record_t *record) { inject_handler_t *handler; int error; spa_t *spa; /* * If this is pool-wide metadata, make sure we unload the corresponding * spa_t, so that the next attempt to load it will trigger the fault. * We call spa_reset() to unload the pool appropriately. */ if (flags & ZINJECT_UNLOAD_SPA) if ((error = spa_reset(name)) != 0) return (error); if (!(flags & ZINJECT_NULL)) { /* * spa_inject_ref() will add an injection reference, which will * prevent the pool from being removed from the namespace while * still allowing it to be unloaded. */ if ((spa = spa_inject_addref(name)) == NULL) return (SET_ERROR(ENOENT)); handler = kmem_alloc(sizeof (inject_handler_t), KM_SLEEP); rw_enter(&inject_lock, RW_WRITER); *id = handler->zi_id = inject_next_id++; handler->zi_spa = spa; handler->zi_record = *record; list_insert_tail(&inject_handlers, handler); - atomic_add_32(&zio_injection_enabled, 1); + atomic_inc_32(&zio_injection_enabled); rw_exit(&inject_lock); } /* * Flush the ARC, so that any attempts to read this data will end up * going to the ZIO layer. Note that this is a little overkill, but * we don't have the necessary ARC interfaces to do anything else, and * fault injection isn't a performance critical path. */ if (flags & ZINJECT_FLUSH_ARC) arc_flush(NULL); return (0); } /* * Returns the next record with an ID greater than that supplied to the * function. Used to iterate over all handlers in the system. */ int zio_inject_list_next(int *id, char *name, size_t buflen, zinject_record_t *record) { inject_handler_t *handler; int ret; mutex_enter(&spa_namespace_lock); rw_enter(&inject_lock, RW_READER); for (handler = list_head(&inject_handlers); handler != NULL; handler = list_next(&inject_handlers, handler)) if (handler->zi_id > *id) break; if (handler) { *record = handler->zi_record; *id = handler->zi_id; (void) strncpy(name, spa_name(handler->zi_spa), buflen); ret = 0; } else { ret = SET_ERROR(ENOENT); } rw_exit(&inject_lock); mutex_exit(&spa_namespace_lock); return (ret); } /* * Clear the fault handler with the given identifier, or return ENOENT if none * exists. */ int zio_clear_fault(int id) { inject_handler_t *handler; rw_enter(&inject_lock, RW_WRITER); for (handler = list_head(&inject_handlers); handler != NULL; handler = list_next(&inject_handlers, handler)) if (handler->zi_id == id) break; if (handler == NULL) { rw_exit(&inject_lock); return (SET_ERROR(ENOENT)); } list_remove(&inject_handlers, handler); rw_exit(&inject_lock); spa_inject_delref(handler->zi_spa); kmem_free(handler, sizeof (inject_handler_t)); - atomic_add_32(&zio_injection_enabled, -1); + atomic_dec_32(&zio_injection_enabled); return (0); } void zio_inject_init(void) { rw_init(&inject_lock, NULL, RW_DEFAULT, NULL); list_create(&inject_handlers, sizeof (inject_handler_t), offsetof(inject_handler_t, zi_link)); } void zio_inject_fini(void) { list_destroy(&inject_handlers); rw_destroy(&inject_lock); } Index: stable/10/sys/cddl/contrib/opensolaris/uts/common/os/fm.c =================================================================== --- stable/10/sys/cddl/contrib/opensolaris/uts/common/os/fm.c (revision 271000) +++ stable/10/sys/cddl/contrib/opensolaris/uts/common/os/fm.c (revision 271001) @@ -1,1402 +1,1399 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved. */ /* * Fault Management Architecture (FMA) Resource and Protocol Support * * The routines contained herein provide services to support kernel subsystems * in publishing fault management telemetry (see PSARC 2002/412 and 2003/089). * * Name-Value Pair Lists * * The embodiment of an FMA protocol element (event, fmri or authority) is a * name-value pair list (nvlist_t). FMA-specific nvlist construtor and * destructor functions, fm_nvlist_create() and fm_nvlist_destroy(), are used * to create an nvpair list using custom allocators. Callers may choose to * allocate either from the kernel memory allocator, or from a preallocated * buffer, useful in constrained contexts like high-level interrupt routines. * * Protocol Event and FMRI Construction * * Convenience routines are provided to construct nvlist events according to * the FMA Event Protocol and Naming Schema specification for ereports and * FMRIs for the dev, cpu, hc, mem, legacy hc and de schemes. * * ENA Manipulation * * Routines to generate ENA formats 0, 1 and 2 are available as well as * routines to increment formats 1 and 2. Individual fields within the * ENA are extractable via fm_ena_time_get(), fm_ena_id_get(), * fm_ena_format_get() and fm_ena_gen_get(). */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * URL and SUNW-MSG-ID value to display for fm_panic(), defined below. These * values must be kept in sync with the FMA source code in usr/src/cmd/fm. */ static const char *fm_url = "http://www.sun.com/msg"; static const char *fm_msgid = "SUNOS-8000-0G"; static char *volatile fm_panicstr = NULL; #ifdef sun errorq_t *ereport_errorq; #endif void *ereport_dumpbuf; size_t ereport_dumplen; static uint_t ereport_chanlen = ERPT_EVCH_MAX; static evchan_t *ereport_chan = NULL; static ulong_t ereport_qlen = 0; static size_t ereport_size = 0; static int ereport_cols = 80; extern void fastreboot_disable_highpil(void); /* * Common fault management kstats to record ereport generation * failures */ struct erpt_kstat { kstat_named_t erpt_dropped; /* num erpts dropped on post */ kstat_named_t erpt_set_failed; /* num erpt set failures */ kstat_named_t fmri_set_failed; /* num fmri set failures */ kstat_named_t payload_set_failed; /* num payload set failures */ }; static struct erpt_kstat erpt_kstat_data = { { "erpt-dropped", KSTAT_DATA_UINT64 }, { "erpt-set-failed", KSTAT_DATA_UINT64 }, { "fmri-set-failed", KSTAT_DATA_UINT64 }, { "payload-set-failed", KSTAT_DATA_UINT64 } }; #ifdef sun /*ARGSUSED*/ static void fm_drain(void *private, void *data, errorq_elem_t *eep) { nvlist_t *nvl = errorq_elem_nvl(ereport_errorq, eep); if (!panicstr) (void) fm_ereport_post(nvl, EVCH_TRYHARD); else fm_nvprint(nvl); } #endif void fm_init(void) { kstat_t *ksp; #ifdef sun (void) sysevent_evc_bind(FM_ERROR_CHAN, &ereport_chan, EVCH_CREAT | EVCH_HOLD_PEND); (void) sysevent_evc_control(ereport_chan, EVCH_SET_CHAN_LEN, &ereport_chanlen); #endif if (ereport_qlen == 0) ereport_qlen = ERPT_MAX_ERRS * MAX(max_ncpus, 4); if (ereport_size == 0) ereport_size = ERPT_DATA_SZ; #ifdef sun ereport_errorq = errorq_nvcreate("fm_ereport_queue", (errorq_func_t)fm_drain, NULL, ereport_qlen, ereport_size, FM_ERR_PIL, ERRORQ_VITAL); if (ereport_errorq == NULL) panic("failed to create required ereport error queue"); #endif ereport_dumpbuf = kmem_alloc(ereport_size, KM_SLEEP); ereport_dumplen = ereport_size; /* Initialize ereport allocation and generation kstats */ ksp = kstat_create("unix", 0, "fm", "misc", KSTAT_TYPE_NAMED, sizeof (struct erpt_kstat) / sizeof (kstat_named_t), KSTAT_FLAG_VIRTUAL); if (ksp != NULL) { ksp->ks_data = &erpt_kstat_data; kstat_install(ksp); } else { cmn_err(CE_NOTE, "failed to create fm/misc kstat\n"); } } #ifdef sun /* * Formatting utility function for fm_nvprintr. We attempt to wrap chunks of * output so they aren't split across console lines, and return the end column. */ /*PRINTFLIKE4*/ static int fm_printf(int depth, int c, int cols, const char *format, ...) { va_list ap; int width; char c1; va_start(ap, format); width = vsnprintf(&c1, sizeof (c1), format, ap); va_end(ap); if (c + width >= cols) { console_printf("\n\r"); c = 0; if (format[0] != ' ' && depth > 0) { console_printf(" "); c++; } } va_start(ap, format); console_vprintf(format, ap); va_end(ap); return ((c + width) % cols); } /* * Recursively print a nvlist in the specified column width and return the * column we end up in. This function is called recursively by fm_nvprint(), * below. We generically format the entire nvpair using hexadecimal * integers and strings, and elide any integer arrays. Arrays are basically * used for cache dumps right now, so we suppress them so as not to overwhelm * the amount of console output we produce at panic time. This can be further * enhanced as FMA technology grows based upon the needs of consumers. All * FMA telemetry is logged using the dump device transport, so the console * output serves only as a fallback in case this procedure is unsuccessful. */ static int fm_nvprintr(nvlist_t *nvl, int d, int c, int cols) { nvpair_t *nvp; for (nvp = nvlist_next_nvpair(nvl, NULL); nvp != NULL; nvp = nvlist_next_nvpair(nvl, nvp)) { data_type_t type = nvpair_type(nvp); const char *name = nvpair_name(nvp); boolean_t b; uint8_t i8; uint16_t i16; uint32_t i32; uint64_t i64; char *str; nvlist_t *cnv; if (strcmp(name, FM_CLASS) == 0) continue; /* already printed by caller */ c = fm_printf(d, c, cols, " %s=", name); switch (type) { case DATA_TYPE_BOOLEAN: c = fm_printf(d + 1, c, cols, " 1"); break; case DATA_TYPE_BOOLEAN_VALUE: (void) nvpair_value_boolean_value(nvp, &b); c = fm_printf(d + 1, c, cols, b ? "1" : "0"); break; case DATA_TYPE_BYTE: (void) nvpair_value_byte(nvp, &i8); c = fm_printf(d + 1, c, cols, "%x", i8); break; case DATA_TYPE_INT8: (void) nvpair_value_int8(nvp, (void *)&i8); c = fm_printf(d + 1, c, cols, "%x", i8); break; case DATA_TYPE_UINT8: (void) nvpair_value_uint8(nvp, &i8); c = fm_printf(d + 1, c, cols, "%x", i8); break; case DATA_TYPE_INT16: (void) nvpair_value_int16(nvp, (void *)&i16); c = fm_printf(d + 1, c, cols, "%x", i16); break; case DATA_TYPE_UINT16: (void) nvpair_value_uint16(nvp, &i16); c = fm_printf(d + 1, c, cols, "%x", i16); break; case DATA_TYPE_INT32: (void) nvpair_value_int32(nvp, (void *)&i32); c = fm_printf(d + 1, c, cols, "%x", i32); break; case DATA_TYPE_UINT32: (void) nvpair_value_uint32(nvp, &i32); c = fm_printf(d + 1, c, cols, "%x", i32); break; case DATA_TYPE_INT64: (void) nvpair_value_int64(nvp, (void *)&i64); c = fm_printf(d + 1, c, cols, "%llx", (u_longlong_t)i64); break; case DATA_TYPE_UINT64: (void) nvpair_value_uint64(nvp, &i64); c = fm_printf(d + 1, c, cols, "%llx", (u_longlong_t)i64); break; case DATA_TYPE_HRTIME: (void) nvpair_value_hrtime(nvp, (void *)&i64); c = fm_printf(d + 1, c, cols, "%llx", (u_longlong_t)i64); break; case DATA_TYPE_STRING: (void) nvpair_value_string(nvp, &str); c = fm_printf(d + 1, c, cols, "\"%s\"", str ? str : ""); break; case DATA_TYPE_NVLIST: c = fm_printf(d + 1, c, cols, "["); (void) nvpair_value_nvlist(nvp, &cnv); c = fm_nvprintr(cnv, d + 1, c, cols); c = fm_printf(d + 1, c, cols, " ]"); break; case DATA_TYPE_NVLIST_ARRAY: { nvlist_t **val; uint_t i, nelem; c = fm_printf(d + 1, c, cols, "["); (void) nvpair_value_nvlist_array(nvp, &val, &nelem); for (i = 0; i < nelem; i++) { c = fm_nvprintr(val[i], d + 1, c, cols); } c = fm_printf(d + 1, c, cols, " ]"); } break; case DATA_TYPE_BOOLEAN_ARRAY: case DATA_TYPE_BYTE_ARRAY: case DATA_TYPE_INT8_ARRAY: case DATA_TYPE_UINT8_ARRAY: case DATA_TYPE_INT16_ARRAY: case DATA_TYPE_UINT16_ARRAY: case DATA_TYPE_INT32_ARRAY: case DATA_TYPE_UINT32_ARRAY: case DATA_TYPE_INT64_ARRAY: case DATA_TYPE_UINT64_ARRAY: case DATA_TYPE_STRING_ARRAY: c = fm_printf(d + 1, c, cols, "[...]"); break; case DATA_TYPE_UNKNOWN: c = fm_printf(d + 1, c, cols, ""); break; } } return (c); } void fm_nvprint(nvlist_t *nvl) { char *class; int c = 0; console_printf("\r"); if (nvlist_lookup_string(nvl, FM_CLASS, &class) == 0) c = fm_printf(0, c, ereport_cols, "%s", class); if (fm_nvprintr(nvl, 0, c, ereport_cols) != 0) console_printf("\n"); console_printf("\n"); } /* * Wrapper for panic() that first produces an FMA-style message for admins. * Normally such messages are generated by fmd(1M)'s syslog-msgs agent: this * is the one exception to that rule and the only error that gets messaged. * This function is intended for use by subsystems that have detected a fatal * error and enqueued appropriate ereports and wish to then force a panic. */ /*PRINTFLIKE1*/ void fm_panic(const char *format, ...) { va_list ap; (void) atomic_cas_ptr((void *)&fm_panicstr, NULL, (void *)format); #if defined(__i386) || defined(__amd64) fastreboot_disable_highpil(); #endif /* __i386 || __amd64 */ va_start(ap, format); vpanic(format, ap); va_end(ap); } /* * Simply tell the caller if fm_panicstr is set, ie. an fma event has * caused the panic. If so, something other than the default panic * diagnosis method will diagnose the cause of the panic. */ int is_fm_panic() { if (fm_panicstr) return (1); else return (0); } /* * Print any appropriate FMA banner message before the panic message. This * function is called by panicsys() and prints the message for fm_panic(). * We print the message here so that it comes after the system is quiesced. * A one-line summary is recorded in the log only (cmn_err(9F) with "!" prefix). * The rest of the message is for the console only and not needed in the log, * so it is printed using console_printf(). We break it up into multiple * chunks so as to avoid overflowing any small legacy prom_printf() buffers. */ void fm_banner(void) { timespec_t tod; hrtime_t now; if (!fm_panicstr) return; /* panic was not initiated by fm_panic(); do nothing */ if (panicstr) { tod = panic_hrestime; now = panic_hrtime; } else { gethrestime(&tod); now = gethrtime_waitfree(); } cmn_err(CE_NOTE, "!SUNW-MSG-ID: %s, " "TYPE: Error, VER: 1, SEVERITY: Major\n", fm_msgid); console_printf( "\n\rSUNW-MSG-ID: %s, TYPE: Error, VER: 1, SEVERITY: Major\n" "EVENT-TIME: 0x%lx.0x%lx (0x%llx)\n", fm_msgid, tod.tv_sec, tod.tv_nsec, (u_longlong_t)now); console_printf( "PLATFORM: %s, CSN: -, HOSTNAME: %s\n" "SOURCE: %s, REV: %s %s\n", platform, utsname.nodename, utsname.sysname, utsname.release, utsname.version); console_printf( "DESC: Errors have been detected that require a reboot to ensure system\n" "integrity. See %s/%s for more information.\n", fm_url, fm_msgid); console_printf( "AUTO-RESPONSE: Solaris will attempt to save and diagnose the error telemetry\n" "IMPACT: The system will sync files, save a crash dump if needed, and reboot\n" "REC-ACTION: Save the error summary below in case telemetry cannot be saved\n"); console_printf("\n"); } /* * Utility function to write all of the pending ereports to the dump device. * This function is called at either normal reboot or panic time, and simply * iterates over the in-transit messages in the ereport sysevent channel. */ void fm_ereport_dump(void) { evchanq_t *chq; sysevent_t *sep; erpt_dump_t ed; timespec_t tod; hrtime_t now; char *buf; size_t len; if (panicstr) { tod = panic_hrestime; now = panic_hrtime; } else { if (ereport_errorq != NULL) errorq_drain(ereport_errorq); gethrestime(&tod); now = gethrtime_waitfree(); } /* * In the panic case, sysevent_evc_walk_init() will return NULL. */ if ((chq = sysevent_evc_walk_init(ereport_chan, NULL)) == NULL && !panicstr) return; /* event channel isn't initialized yet */ while ((sep = sysevent_evc_walk_step(chq)) != NULL) { if ((buf = sysevent_evc_event_attr(sep, &len)) == NULL) break; ed.ed_magic = ERPT_MAGIC; ed.ed_chksum = checksum32(buf, len); ed.ed_size = (uint32_t)len; ed.ed_pad = 0; ed.ed_hrt_nsec = SE_TIME(sep); ed.ed_hrt_base = now; ed.ed_tod_base.sec = tod.tv_sec; ed.ed_tod_base.nsec = tod.tv_nsec; dumpvp_write(&ed, sizeof (ed)); dumpvp_write(buf, len); } sysevent_evc_walk_fini(chq); } #endif /* * Post an error report (ereport) to the sysevent error channel. The error * channel must be established with a prior call to sysevent_evc_create() * before publication may occur. */ void fm_ereport_post(nvlist_t *ereport, int evc_flag) { size_t nvl_size = 0; evchan_t *error_chan; sysevent_id_t eid; (void) nvlist_size(ereport, &nvl_size, NV_ENCODE_NATIVE); if (nvl_size > ERPT_DATA_SZ || nvl_size == 0) { - atomic_add_64(&erpt_kstat_data.erpt_dropped.value.ui64, 1); + atomic_inc_64(&erpt_kstat_data.erpt_dropped.value.ui64); return; } #ifdef sun if (sysevent_evc_bind(FM_ERROR_CHAN, &error_chan, EVCH_CREAT|EVCH_HOLD_PEND) != 0) { - atomic_add_64(&erpt_kstat_data.erpt_dropped.value.ui64, 1); + atomic_inc_64(&erpt_kstat_data.erpt_dropped.value.ui64); return; } if (sysevent_evc_publish(error_chan, EC_FM, ESC_FM_ERROR, SUNW_VENDOR, FM_PUB, ereport, evc_flag) != 0) { - atomic_add_64(&erpt_kstat_data.erpt_dropped.value.ui64, 1); + atomic_inc_64(&erpt_kstat_data.erpt_dropped.value.ui64); (void) sysevent_evc_unbind(error_chan); return; } (void) sysevent_evc_unbind(error_chan); #else (void) ddi_log_sysevent(NULL, SUNW_VENDOR, EC_DEV_STATUS, ESC_DEV_DLE, ereport, &eid, DDI_SLEEP); #endif } /* * Wrapppers for FM nvlist allocators */ /* ARGSUSED */ static void * i_fm_alloc(nv_alloc_t *nva, size_t size) { return (kmem_zalloc(size, KM_SLEEP)); } /* ARGSUSED */ static void i_fm_free(nv_alloc_t *nva, void *buf, size_t size) { kmem_free(buf, size); } const nv_alloc_ops_t fm_mem_alloc_ops = { NULL, NULL, i_fm_alloc, i_fm_free, NULL }; /* * Create and initialize a new nv_alloc_t for a fixed buffer, buf. A pointer * to the newly allocated nv_alloc_t structure is returned upon success or NULL * is returned to indicate that the nv_alloc structure could not be created. */ nv_alloc_t * fm_nva_xcreate(char *buf, size_t bufsz) { nv_alloc_t *nvhdl = kmem_zalloc(sizeof (nv_alloc_t), KM_SLEEP); if (bufsz == 0 || nv_alloc_init(nvhdl, nv_fixed_ops, buf, bufsz) != 0) { kmem_free(nvhdl, sizeof (nv_alloc_t)); return (NULL); } return (nvhdl); } /* * Destroy a previously allocated nv_alloc structure. The fixed buffer * associated with nva must be freed by the caller. */ void fm_nva_xdestroy(nv_alloc_t *nva) { nv_alloc_fini(nva); kmem_free(nva, sizeof (nv_alloc_t)); } /* * Create a new nv list. A pointer to a new nv list structure is returned * upon success or NULL is returned to indicate that the structure could * not be created. The newly created nv list is created and managed by the * operations installed in nva. If nva is NULL, the default FMA nva * operations are installed and used. * * When called from the kernel and nva == NULL, this function must be called * from passive kernel context with no locks held that can prevent a * sleeping memory allocation from occurring. Otherwise, this function may * be called from other kernel contexts as long a valid nva created via * fm_nva_create() is supplied. */ nvlist_t * fm_nvlist_create(nv_alloc_t *nva) { int hdl_alloced = 0; nvlist_t *nvl; nv_alloc_t *nvhdl; if (nva == NULL) { nvhdl = kmem_zalloc(sizeof (nv_alloc_t), KM_SLEEP); if (nv_alloc_init(nvhdl, &fm_mem_alloc_ops, NULL, 0) != 0) { kmem_free(nvhdl, sizeof (nv_alloc_t)); return (NULL); } hdl_alloced = 1; } else { nvhdl = nva; } if (nvlist_xalloc(&nvl, NV_UNIQUE_NAME, nvhdl) != 0) { if (hdl_alloced) { nv_alloc_fini(nvhdl); kmem_free(nvhdl, sizeof (nv_alloc_t)); } return (NULL); } return (nvl); } /* * Destroy a previously allocated nvlist structure. flag indicates whether * or not the associated nva structure should be freed (FM_NVA_FREE) or * retained (FM_NVA_RETAIN). Retaining the nv alloc structure allows * it to be re-used for future nvlist creation operations. */ void fm_nvlist_destroy(nvlist_t *nvl, int flag) { nv_alloc_t *nva = nvlist_lookup_nv_alloc(nvl); nvlist_free(nvl); if (nva != NULL) { if (flag == FM_NVA_FREE) fm_nva_xdestroy(nva); } } int i_fm_payload_set(nvlist_t *payload, const char *name, va_list ap) { int nelem, ret = 0; data_type_t type; while (ret == 0 && name != NULL) { type = va_arg(ap, data_type_t); switch (type) { case DATA_TYPE_BYTE: ret = nvlist_add_byte(payload, name, va_arg(ap, uint_t)); break; case DATA_TYPE_BYTE_ARRAY: nelem = va_arg(ap, int); ret = nvlist_add_byte_array(payload, name, va_arg(ap, uchar_t *), nelem); break; case DATA_TYPE_BOOLEAN_VALUE: ret = nvlist_add_boolean_value(payload, name, va_arg(ap, boolean_t)); break; case DATA_TYPE_BOOLEAN_ARRAY: nelem = va_arg(ap, int); ret = nvlist_add_boolean_array(payload, name, va_arg(ap, boolean_t *), nelem); break; case DATA_TYPE_INT8: ret = nvlist_add_int8(payload, name, va_arg(ap, int)); break; case DATA_TYPE_INT8_ARRAY: nelem = va_arg(ap, int); ret = nvlist_add_int8_array(payload, name, va_arg(ap, int8_t *), nelem); break; case DATA_TYPE_UINT8: ret = nvlist_add_uint8(payload, name, va_arg(ap, uint_t)); break; case DATA_TYPE_UINT8_ARRAY: nelem = va_arg(ap, int); ret = nvlist_add_uint8_array(payload, name, va_arg(ap, uint8_t *), nelem); break; case DATA_TYPE_INT16: ret = nvlist_add_int16(payload, name, va_arg(ap, int)); break; case DATA_TYPE_INT16_ARRAY: nelem = va_arg(ap, int); ret = nvlist_add_int16_array(payload, name, va_arg(ap, int16_t *), nelem); break; case DATA_TYPE_UINT16: ret = nvlist_add_uint16(payload, name, va_arg(ap, uint_t)); break; case DATA_TYPE_UINT16_ARRAY: nelem = va_arg(ap, int); ret = nvlist_add_uint16_array(payload, name, va_arg(ap, uint16_t *), nelem); break; case DATA_TYPE_INT32: ret = nvlist_add_int32(payload, name, va_arg(ap, int32_t)); break; case DATA_TYPE_INT32_ARRAY: nelem = va_arg(ap, int); ret = nvlist_add_int32_array(payload, name, va_arg(ap, int32_t *), nelem); break; case DATA_TYPE_UINT32: ret = nvlist_add_uint32(payload, name, va_arg(ap, uint32_t)); break; case DATA_TYPE_UINT32_ARRAY: nelem = va_arg(ap, int); ret = nvlist_add_uint32_array(payload, name, va_arg(ap, uint32_t *), nelem); break; case DATA_TYPE_INT64: ret = nvlist_add_int64(payload, name, va_arg(ap, int64_t)); break; case DATA_TYPE_INT64_ARRAY: nelem = va_arg(ap, int); ret = nvlist_add_int64_array(payload, name, va_arg(ap, int64_t *), nelem); break; case DATA_TYPE_UINT64: ret = nvlist_add_uint64(payload, name, va_arg(ap, uint64_t)); break; case DATA_TYPE_UINT64_ARRAY: nelem = va_arg(ap, int); ret = nvlist_add_uint64_array(payload, name, va_arg(ap, uint64_t *), nelem); break; case DATA_TYPE_STRING: ret = nvlist_add_string(payload, name, va_arg(ap, char *)); break; case DATA_TYPE_STRING_ARRAY: nelem = va_arg(ap, int); ret = nvlist_add_string_array(payload, name, va_arg(ap, char **), nelem); break; case DATA_TYPE_NVLIST: ret = nvlist_add_nvlist(payload, name, va_arg(ap, nvlist_t *)); break; case DATA_TYPE_NVLIST_ARRAY: nelem = va_arg(ap, int); ret = nvlist_add_nvlist_array(payload, name, va_arg(ap, nvlist_t **), nelem); break; default: ret = EINVAL; } name = va_arg(ap, char *); } return (ret); } void fm_payload_set(nvlist_t *payload, ...) { int ret; const char *name; va_list ap; va_start(ap, payload); name = va_arg(ap, char *); ret = i_fm_payload_set(payload, name, ap); va_end(ap); if (ret) - atomic_add_64( - &erpt_kstat_data.payload_set_failed.value.ui64, 1); + atomic_inc_64(&erpt_kstat_data.payload_set_failed.value.ui64); } /* * Set-up and validate the members of an ereport event according to: * * Member name Type Value * ==================================================== * class string ereport * version uint8_t 0 * ena uint64_t * detector nvlist_t * ereport-payload nvlist_t * * We don't actually add a 'version' member to the payload. Really, * the version quoted to us by our caller is that of the category 1 * "ereport" event class (and we require FM_EREPORT_VERS0) but * the payload version of the actual leaf class event under construction * may be something else. Callers should supply a version in the varargs, * or (better) we could take two version arguments - one for the * ereport category 1 classification (expect FM_EREPORT_VERS0) and one * for the leaf class. */ void fm_ereport_set(nvlist_t *ereport, int version, const char *erpt_class, uint64_t ena, const nvlist_t *detector, ...) { char ereport_class[FM_MAX_CLASS]; const char *name; va_list ap; int ret; if (version != FM_EREPORT_VERS0) { - atomic_add_64(&erpt_kstat_data.erpt_set_failed.value.ui64, 1); + atomic_inc_64(&erpt_kstat_data.erpt_set_failed.value.ui64); return; } (void) snprintf(ereport_class, FM_MAX_CLASS, "%s.%s", FM_EREPORT_CLASS, erpt_class); if (nvlist_add_string(ereport, FM_CLASS, ereport_class) != 0) { - atomic_add_64(&erpt_kstat_data.erpt_set_failed.value.ui64, 1); + atomic_inc_64(&erpt_kstat_data.erpt_set_failed.value.ui64); return; } if (nvlist_add_uint64(ereport, FM_EREPORT_ENA, ena)) { - atomic_add_64(&erpt_kstat_data.erpt_set_failed.value.ui64, 1); + atomic_inc_64(&erpt_kstat_data.erpt_set_failed.value.ui64); } if (nvlist_add_nvlist(ereport, FM_EREPORT_DETECTOR, (nvlist_t *)detector) != 0) { - atomic_add_64(&erpt_kstat_data.erpt_set_failed.value.ui64, 1); + atomic_inc_64(&erpt_kstat_data.erpt_set_failed.value.ui64); } va_start(ap, detector); name = va_arg(ap, const char *); ret = i_fm_payload_set(ereport, name, ap); va_end(ap); if (ret) - atomic_add_64(&erpt_kstat_data.erpt_set_failed.value.ui64, 1); + atomic_inc_64(&erpt_kstat_data.erpt_set_failed.value.ui64); } /* * Set-up and validate the members of an hc fmri according to; * * Member name Type Value * =================================================== * version uint8_t 0 * auth nvlist_t * hc-name string * hc-id string * * Note that auth and hc-id are optional members. */ #define HC_MAXPAIRS 20 #define HC_MAXNAMELEN 50 static int fm_fmri_hc_set_common(nvlist_t *fmri, int version, const nvlist_t *auth) { if (version != FM_HC_SCHEME_VERSION) { - atomic_add_64(&erpt_kstat_data.fmri_set_failed.value.ui64, 1); + atomic_inc_64(&erpt_kstat_data.fmri_set_failed.value.ui64); return (0); } if (nvlist_add_uint8(fmri, FM_VERSION, version) != 0 || nvlist_add_string(fmri, FM_FMRI_SCHEME, FM_FMRI_SCHEME_HC) != 0) { - atomic_add_64(&erpt_kstat_data.fmri_set_failed.value.ui64, 1); + atomic_inc_64(&erpt_kstat_data.fmri_set_failed.value.ui64); return (0); } if (auth != NULL && nvlist_add_nvlist(fmri, FM_FMRI_AUTHORITY, (nvlist_t *)auth) != 0) { - atomic_add_64(&erpt_kstat_data.fmri_set_failed.value.ui64, 1); + atomic_inc_64(&erpt_kstat_data.fmri_set_failed.value.ui64); return (0); } return (1); } void fm_fmri_hc_set(nvlist_t *fmri, int version, const nvlist_t *auth, nvlist_t *snvl, int npairs, ...) { nv_alloc_t *nva = nvlist_lookup_nv_alloc(fmri); nvlist_t *pairs[HC_MAXPAIRS]; va_list ap; int i; if (!fm_fmri_hc_set_common(fmri, version, auth)) return; npairs = MIN(npairs, HC_MAXPAIRS); va_start(ap, npairs); for (i = 0; i < npairs; i++) { const char *name = va_arg(ap, const char *); uint32_t id = va_arg(ap, uint32_t); char idstr[11]; (void) snprintf(idstr, sizeof (idstr), "%u", id); pairs[i] = fm_nvlist_create(nva); if (nvlist_add_string(pairs[i], FM_FMRI_HC_NAME, name) != 0 || nvlist_add_string(pairs[i], FM_FMRI_HC_ID, idstr) != 0) { - atomic_add_64( - &erpt_kstat_data.fmri_set_failed.value.ui64, 1); + atomic_inc_64( + &erpt_kstat_data.fmri_set_failed.value.ui64); } } va_end(ap); if (nvlist_add_nvlist_array(fmri, FM_FMRI_HC_LIST, pairs, npairs) != 0) - atomic_add_64(&erpt_kstat_data.fmri_set_failed.value.ui64, 1); + atomic_inc_64(&erpt_kstat_data.fmri_set_failed.value.ui64); for (i = 0; i < npairs; i++) fm_nvlist_destroy(pairs[i], FM_NVA_RETAIN); if (snvl != NULL) { if (nvlist_add_nvlist(fmri, FM_FMRI_HC_SPECIFIC, snvl) != 0) { - atomic_add_64( - &erpt_kstat_data.fmri_set_failed.value.ui64, 1); + atomic_inc_64( + &erpt_kstat_data.fmri_set_failed.value.ui64); } } } /* * Set-up and validate the members of an dev fmri according to: * * Member name Type Value * ==================================================== * version uint8_t 0 * auth nvlist_t * devpath string * [devid] string * [target-port-l0id] string * * Note that auth and devid are optional members. */ void fm_fmri_dev_set(nvlist_t *fmri_dev, int version, const nvlist_t *auth, const char *devpath, const char *devid, const char *tpl0) { int err = 0; if (version != DEV_SCHEME_VERSION0) { - atomic_add_64(&erpt_kstat_data.fmri_set_failed.value.ui64, 1); + atomic_inc_64(&erpt_kstat_data.fmri_set_failed.value.ui64); return; } err |= nvlist_add_uint8(fmri_dev, FM_VERSION, version); err |= nvlist_add_string(fmri_dev, FM_FMRI_SCHEME, FM_FMRI_SCHEME_DEV); if (auth != NULL) { err |= nvlist_add_nvlist(fmri_dev, FM_FMRI_AUTHORITY, (nvlist_t *)auth); } err |= nvlist_add_string(fmri_dev, FM_FMRI_DEV_PATH, devpath); if (devid != NULL) err |= nvlist_add_string(fmri_dev, FM_FMRI_DEV_ID, devid); if (tpl0 != NULL) err |= nvlist_add_string(fmri_dev, FM_FMRI_DEV_TGTPTLUN0, tpl0); if (err) - atomic_add_64(&erpt_kstat_data.fmri_set_failed.value.ui64, 1); + atomic_inc_64(&erpt_kstat_data.fmri_set_failed.value.ui64); } /* * Set-up and validate the members of an cpu fmri according to: * * Member name Type Value * ==================================================== * version uint8_t 0 * auth nvlist_t * cpuid uint32_t * cpumask uint8_t * serial uint64_t * * Note that auth, cpumask, serial are optional members. * */ void fm_fmri_cpu_set(nvlist_t *fmri_cpu, int version, const nvlist_t *auth, uint32_t cpu_id, uint8_t *cpu_maskp, const char *serial_idp) { uint64_t *failedp = &erpt_kstat_data.fmri_set_failed.value.ui64; if (version < CPU_SCHEME_VERSION1) { - atomic_add_64(failedp, 1); + atomic_inc_64(failedp); return; } if (nvlist_add_uint8(fmri_cpu, FM_VERSION, version) != 0) { - atomic_add_64(failedp, 1); + atomic_inc_64(failedp); return; } if (nvlist_add_string(fmri_cpu, FM_FMRI_SCHEME, FM_FMRI_SCHEME_CPU) != 0) { - atomic_add_64(failedp, 1); + atomic_inc_64(failedp); return; } if (auth != NULL && nvlist_add_nvlist(fmri_cpu, FM_FMRI_AUTHORITY, (nvlist_t *)auth) != 0) - atomic_add_64(failedp, 1); + atomic_inc_64(failedp); if (nvlist_add_uint32(fmri_cpu, FM_FMRI_CPU_ID, cpu_id) != 0) - atomic_add_64(failedp, 1); + atomic_inc_64(failedp); if (cpu_maskp != NULL && nvlist_add_uint8(fmri_cpu, FM_FMRI_CPU_MASK, *cpu_maskp) != 0) - atomic_add_64(failedp, 1); + atomic_inc_64(failedp); if (serial_idp == NULL || nvlist_add_string(fmri_cpu, FM_FMRI_CPU_SERIAL_ID, (char *)serial_idp) != 0) - atomic_add_64(failedp, 1); + atomic_inc_64(failedp); } /* * Set-up and validate the members of a mem according to: * * Member name Type Value * ==================================================== * version uint8_t 0 * auth nvlist_t [optional] * unum string * serial string [optional*] * offset uint64_t [optional] * * * serial is required if offset is present */ void fm_fmri_mem_set(nvlist_t *fmri, int version, const nvlist_t *auth, const char *unum, const char *serial, uint64_t offset) { if (version != MEM_SCHEME_VERSION0) { - atomic_add_64(&erpt_kstat_data.fmri_set_failed.value.ui64, 1); + atomic_inc_64(&erpt_kstat_data.fmri_set_failed.value.ui64); return; } if (!serial && (offset != (uint64_t)-1)) { - atomic_add_64(&erpt_kstat_data.fmri_set_failed.value.ui64, 1); + atomic_inc_64(&erpt_kstat_data.fmri_set_failed.value.ui64); return; } if (nvlist_add_uint8(fmri, FM_VERSION, version) != 0) { - atomic_add_64(&erpt_kstat_data.fmri_set_failed.value.ui64, 1); + atomic_inc_64(&erpt_kstat_data.fmri_set_failed.value.ui64); return; } if (nvlist_add_string(fmri, FM_FMRI_SCHEME, FM_FMRI_SCHEME_MEM) != 0) { - atomic_add_64(&erpt_kstat_data.fmri_set_failed.value.ui64, 1); + atomic_inc_64(&erpt_kstat_data.fmri_set_failed.value.ui64); return; } if (auth != NULL) { if (nvlist_add_nvlist(fmri, FM_FMRI_AUTHORITY, (nvlist_t *)auth) != 0) { - atomic_add_64( - &erpt_kstat_data.fmri_set_failed.value.ui64, 1); + atomic_inc_64( + &erpt_kstat_data.fmri_set_failed.value.ui64); } } if (nvlist_add_string(fmri, FM_FMRI_MEM_UNUM, unum) != 0) { - atomic_add_64(&erpt_kstat_data.fmri_set_failed.value.ui64, 1); + atomic_inc_64(&erpt_kstat_data.fmri_set_failed.value.ui64); } if (serial != NULL) { if (nvlist_add_string_array(fmri, FM_FMRI_MEM_SERIAL_ID, (char **)&serial, 1) != 0) { - atomic_add_64( - &erpt_kstat_data.fmri_set_failed.value.ui64, 1); + atomic_inc_64( + &erpt_kstat_data.fmri_set_failed.value.ui64); } - if (offset != (uint64_t)-1) { - if (nvlist_add_uint64(fmri, FM_FMRI_MEM_OFFSET, - offset) != 0) { - atomic_add_64(&erpt_kstat_data. - fmri_set_failed.value.ui64, 1); - } + if (offset != (uint64_t)-1 && nvlist_add_uint64(fmri, + FM_FMRI_MEM_OFFSET, offset) != 0) { + atomic_inc_64( + &erpt_kstat_data.fmri_set_failed.value.ui64); } } } void fm_fmri_zfs_set(nvlist_t *fmri, int version, uint64_t pool_guid, uint64_t vdev_guid) { if (version != ZFS_SCHEME_VERSION0) { - atomic_add_64(&erpt_kstat_data.fmri_set_failed.value.ui64, 1); + atomic_inc_64(&erpt_kstat_data.fmri_set_failed.value.ui64); return; } if (nvlist_add_uint8(fmri, FM_VERSION, version) != 0) { - atomic_add_64(&erpt_kstat_data.fmri_set_failed.value.ui64, 1); + atomic_inc_64(&erpt_kstat_data.fmri_set_failed.value.ui64); return; } if (nvlist_add_string(fmri, FM_FMRI_SCHEME, FM_FMRI_SCHEME_ZFS) != 0) { - atomic_add_64(&erpt_kstat_data.fmri_set_failed.value.ui64, 1); + atomic_inc_64(&erpt_kstat_data.fmri_set_failed.value.ui64); return; } if (nvlist_add_uint64(fmri, FM_FMRI_ZFS_POOL, pool_guid) != 0) { - atomic_add_64(&erpt_kstat_data.fmri_set_failed.value.ui64, 1); + atomic_inc_64(&erpt_kstat_data.fmri_set_failed.value.ui64); } if (vdev_guid != 0) { if (nvlist_add_uint64(fmri, FM_FMRI_ZFS_VDEV, vdev_guid) != 0) { - atomic_add_64( - &erpt_kstat_data.fmri_set_failed.value.ui64, 1); + atomic_inc_64( + &erpt_kstat_data.fmri_set_failed.value.ui64); } } } uint64_t fm_ena_increment(uint64_t ena) { uint64_t new_ena; switch (ENA_FORMAT(ena)) { case FM_ENA_FMT1: new_ena = ena + (1 << ENA_FMT1_GEN_SHFT); break; case FM_ENA_FMT2: new_ena = ena + (1 << ENA_FMT2_GEN_SHFT); break; default: new_ena = 0; } return (new_ena); } uint64_t fm_ena_generate_cpu(uint64_t timestamp, processorid_t cpuid, uchar_t format) { uint64_t ena = 0; switch (format) { case FM_ENA_FMT1: if (timestamp) { ena = (uint64_t)((format & ENA_FORMAT_MASK) | ((cpuid << ENA_FMT1_CPUID_SHFT) & ENA_FMT1_CPUID_MASK) | ((timestamp << ENA_FMT1_TIME_SHFT) & ENA_FMT1_TIME_MASK)); } else { ena = (uint64_t)((format & ENA_FORMAT_MASK) | ((cpuid << ENA_FMT1_CPUID_SHFT) & ENA_FMT1_CPUID_MASK) | ((gethrtime_waitfree() << ENA_FMT1_TIME_SHFT) & ENA_FMT1_TIME_MASK)); } break; case FM_ENA_FMT2: ena = (uint64_t)((format & ENA_FORMAT_MASK) | ((timestamp << ENA_FMT2_TIME_SHFT) & ENA_FMT2_TIME_MASK)); break; default: break; } return (ena); } uint64_t fm_ena_generate(uint64_t timestamp, uchar_t format) { return (fm_ena_generate_cpu(timestamp, PCPU_GET(cpuid), format)); } uint64_t fm_ena_generation_get(uint64_t ena) { uint64_t gen; switch (ENA_FORMAT(ena)) { case FM_ENA_FMT1: gen = (ena & ENA_FMT1_GEN_MASK) >> ENA_FMT1_GEN_SHFT; break; case FM_ENA_FMT2: gen = (ena & ENA_FMT2_GEN_MASK) >> ENA_FMT2_GEN_SHFT; break; default: gen = 0; break; } return (gen); } uchar_t fm_ena_format_get(uint64_t ena) { return (ENA_FORMAT(ena)); } uint64_t fm_ena_id_get(uint64_t ena) { uint64_t id; switch (ENA_FORMAT(ena)) { case FM_ENA_FMT1: id = (ena & ENA_FMT1_ID_MASK) >> ENA_FMT1_ID_SHFT; break; case FM_ENA_FMT2: id = (ena & ENA_FMT2_ID_MASK) >> ENA_FMT2_ID_SHFT; break; default: id = 0; } return (id); } uint64_t fm_ena_time_get(uint64_t ena) { uint64_t time; switch (ENA_FORMAT(ena)) { case FM_ENA_FMT1: time = (ena & ENA_FMT1_TIME_MASK) >> ENA_FMT1_TIME_SHFT; break; case FM_ENA_FMT2: time = (ena & ENA_FMT2_TIME_MASK) >> ENA_FMT2_TIME_SHFT; break; default: time = 0; } return (time); } #ifdef sun /* * Convert a getpcstack() trace to symbolic name+offset, and add the resulting * string array to a Fault Management ereport as FM_EREPORT_PAYLOAD_NAME_STACK. */ void fm_payload_stack_add(nvlist_t *payload, const pc_t *stack, int depth) { int i; char *sym; ulong_t off; char *stkpp[FM_STK_DEPTH]; char buf[FM_STK_DEPTH * FM_SYM_SZ]; char *stkp = buf; for (i = 0; i < depth && i != FM_STK_DEPTH; i++, stkp += FM_SYM_SZ) { if ((sym = kobj_getsymname(stack[i], &off)) != NULL) (void) snprintf(stkp, FM_SYM_SZ, "%s+%lx", sym, off); else (void) snprintf(stkp, FM_SYM_SZ, "%lx", (long)stack[i]); stkpp[i] = stkp; } fm_payload_set(payload, FM_EREPORT_PAYLOAD_NAME_STACK, DATA_TYPE_STRING_ARRAY, depth, stkpp, NULL); } #endif #ifdef sun void print_msg_hwerr(ctid_t ct_id, proc_t *p) { uprintf("Killed process %d (%s) in contract id %d " "due to hardware error\n", p->p_pid, p->p_user.u_comm, ct_id); } #endif void fm_fmri_hc_create(nvlist_t *fmri, int version, const nvlist_t *auth, nvlist_t *snvl, nvlist_t *bboard, int npairs, ...) { nv_alloc_t *nva = nvlist_lookup_nv_alloc(fmri); nvlist_t *pairs[HC_MAXPAIRS]; nvlist_t **hcl; uint_t n; int i, j; va_list ap; char *hcname, *hcid; if (!fm_fmri_hc_set_common(fmri, version, auth)) return; /* * copy the bboard nvpairs to the pairs array */ if (nvlist_lookup_nvlist_array(bboard, FM_FMRI_HC_LIST, &hcl, &n) != 0) { - atomic_add_64(&erpt_kstat_data.fmri_set_failed.value.ui64, 1); + atomic_inc_64(&erpt_kstat_data.fmri_set_failed.value.ui64); return; } for (i = 0; i < n; i++) { if (nvlist_lookup_string(hcl[i], FM_FMRI_HC_NAME, &hcname) != 0) { - atomic_add_64( - &erpt_kstat_data.fmri_set_failed.value.ui64, 1); + atomic_inc_64( + &erpt_kstat_data.fmri_set_failed.value.ui64); return; } if (nvlist_lookup_string(hcl[i], FM_FMRI_HC_ID, &hcid) != 0) { - atomic_add_64( - &erpt_kstat_data.fmri_set_failed.value.ui64, 1); + atomic_inc_64( + &erpt_kstat_data.fmri_set_failed.value.ui64); return; } pairs[i] = fm_nvlist_create(nva); if (nvlist_add_string(pairs[i], FM_FMRI_HC_NAME, hcname) != 0 || nvlist_add_string(pairs[i], FM_FMRI_HC_ID, hcid) != 0) { for (j = 0; j <= i; j++) { if (pairs[j] != NULL) fm_nvlist_destroy(pairs[j], FM_NVA_RETAIN); } - atomic_add_64( - &erpt_kstat_data.fmri_set_failed.value.ui64, 1); + atomic_inc_64( + &erpt_kstat_data.fmri_set_failed.value.ui64); return; } } /* * create the pairs from passed in pairs */ npairs = MIN(npairs, HC_MAXPAIRS); va_start(ap, npairs); for (i = n; i < npairs + n; i++) { const char *name = va_arg(ap, const char *); uint32_t id = va_arg(ap, uint32_t); char idstr[11]; (void) snprintf(idstr, sizeof (idstr), "%u", id); pairs[i] = fm_nvlist_create(nva); if (nvlist_add_string(pairs[i], FM_FMRI_HC_NAME, name) != 0 || nvlist_add_string(pairs[i], FM_FMRI_HC_ID, idstr) != 0) { for (j = 0; j <= i; j++) { if (pairs[j] != NULL) fm_nvlist_destroy(pairs[j], FM_NVA_RETAIN); } - atomic_add_64( - &erpt_kstat_data.fmri_set_failed.value.ui64, 1); + atomic_inc_64( + &erpt_kstat_data.fmri_set_failed.value.ui64); return; } } va_end(ap); /* * Create the fmri hc list */ if (nvlist_add_nvlist_array(fmri, FM_FMRI_HC_LIST, pairs, npairs + n) != 0) { - atomic_add_64(&erpt_kstat_data.fmri_set_failed.value.ui64, 1); + atomic_inc_64(&erpt_kstat_data.fmri_set_failed.value.ui64); return; } for (i = 0; i < npairs + n; i++) { fm_nvlist_destroy(pairs[i], FM_NVA_RETAIN); } if (snvl != NULL) { if (nvlist_add_nvlist(fmri, FM_FMRI_HC_SPECIFIC, snvl) != 0) { - atomic_add_64( - &erpt_kstat_data.fmri_set_failed.value.ui64, 1); + atomic_inc_64( + &erpt_kstat_data.fmri_set_failed.value.ui64); return; } } } Index: stable/10/sys/cddl/contrib/opensolaris/uts/sparc/dtrace/fasttrap_isa.c =================================================================== --- stable/10/sys/cddl/contrib/opensolaris/uts/sparc/dtrace/fasttrap_isa.c (revision 271000) +++ stable/10/sys/cddl/contrib/opensolaris/uts/sparc/dtrace/fasttrap_isa.c (revision 271001) @@ -1,1597 +1,1595 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright 2007 Sun Microsystems, Inc. All rights reserved. * Use is subject to license terms. */ -#pragma ident "%Z%%M% %I% %E% SMI" - #include #include #include #include #include #include #include #include #include #include #include /* * Lossless User-Land Tracing on SPARC * ----------------------------------- * * The Basic Idea * * The most important design constraint is, of course, correct execution of * the user thread above all else. The next most important goal is rapid * execution. We combine execution of instructions in user-land with * emulation of certain instructions in the kernel to aim for complete * correctness and maximal performance. * * We take advantage of the split PC/NPC architecture to speed up logical * single-stepping; when we copy an instruction out to the scratch space in * the ulwp_t structure (held in the %g7 register on SPARC), we can * effectively single step by setting the PC to our scratch space and leaving * the NPC alone. This executes the replaced instruction and then continues * on without having to reenter the kernel as with single- stepping. The * obvious caveat is for instructions whose execution is PC dependant -- * branches, call and link instructions (call and jmpl), and the rdpc * instruction. These instructions cannot be executed in the manner described * so they must be emulated in the kernel. * * Emulation for this small set of instructions if fairly simple; the most * difficult part being emulating branch conditions. * * * A Cache Heavy Portfolio * * It's important to note at this time that copying an instruction out to the * ulwp_t scratch space in user-land is rather complicated. SPARC has * separate data and instruction caches so any writes to the D$ (using a * store instruction for example) aren't necessarily reflected in the I$. * The flush instruction can be used to synchronize the two and must be used * for any self-modifying code, but the flush instruction only applies to the * primary address space (the absence of a flusha analogue to the flush * instruction that accepts an ASI argument is an obvious omission from SPARC * v9 where the notion of the alternate address space was introduced on * SPARC). To correctly copy out the instruction we must use a block store * that doesn't allocate in the D$ and ensures synchronization with the I$; * see dtrace_blksuword32() for the implementation (this function uses * ASI_BLK_COMMIT_S to write a block through the secondary ASI in the manner * described). Refer to the UltraSPARC I/II manual for details on the * ASI_BLK_COMMIT_S ASI. * * * Return Subtleties * * When we're firing a return probe we need to expose the value returned by * the function being traced. Since the function can set the return value * in its last instruction, we need to fire the return probe only _after_ * the effects of the instruction are apparent. For instructions that we * emulate, we can call dtrace_probe() after we've performed the emulation; * for instructions that we execute after we return to user-land, we set * %pc to the instruction we copied out (as described above) and set %npc * to a trap instruction stashed in the ulwp_t structure. After the traced * instruction is executed, the trap instruction returns control to the * kernel where we can fire the return probe. * * This need for a second trap in cases where we execute the traced * instruction makes it all the more important to emulate the most common * instructions to avoid the second trip in and out of the kernel. * * * Making it Fast * * Since copying out an instruction is neither simple nor inexpensive for the * CPU, we should attempt to avoid doing it in as many cases as possible. * Since function entry and return are usually the most interesting probe * sites, we attempt to tune the performance of the fasttrap provider around * instructions typically in those places. * * Looking at a bunch of functions in libraries and executables reveals that * most functions begin with either a save or a sethi (to setup a larger * argument to the save) and end with a restore or an or (in the case of leaf * functions). To try to improve performance, we emulate all of these * instructions in the kernel. * * The save and restore instructions are a little tricky since they perform * register window maniplulation. Rather than trying to tinker with the * register windows from the kernel, we emulate the implicit add that takes * place as part of those instructions and set the %pc to point to a simple * save or restore we've hidden in the ulwp_t structure. If we're in a return * probe so want to make it seem as though the tracepoint has been completely * executed we need to remember that we've pulled this trick with restore and * pull registers from the previous window (the one that we'll switch to once * the simple store instruction is executed) rather than the current one. This * is why in the case of emulating a restore we set the DTrace CPU flag * CPU_DTRACE_FAKERESTORE before calling dtrace_probe() for the return probes * (see fasttrap_return_common()). */ #define OP(x) ((x) >> 30) #define OP2(x) (((x) >> 22) & 0x07) #define OP3(x) (((x) >> 19) & 0x3f) #define RCOND(x) (((x) >> 25) & 0x07) #define COND(x) (((x) >> 25) & 0x0f) #define A(x) (((x) >> 29) & 0x01) #define I(x) (((x) >> 13) & 0x01) #define RD(x) (((x) >> 25) & 0x1f) #define RS1(x) (((x) >> 14) & 0x1f) #define RS2(x) (((x) >> 0) & 0x1f) #define CC(x) (((x) >> 20) & 0x03) #define DISP16(x) ((((x) >> 6) & 0xc000) | ((x) & 0x3fff)) #define DISP22(x) ((x) & 0x3fffff) #define DISP19(x) ((x) & 0x7ffff) #define DISP30(x) ((x) & 0x3fffffff) #define SW_TRAP(x) ((x) & 0x7f) #define OP3_OR 0x02 #define OP3_RD 0x28 #define OP3_JMPL 0x38 #define OP3_RETURN 0x39 #define OP3_TCC 0x3a #define OP3_SAVE 0x3c #define OP3_RESTORE 0x3d #define OP3_PREFETCH 0x2d #define OP3_CASA 0x3c #define OP3_PREFETCHA 0x3d #define OP3_CASXA 0x3e #define OP2_ILLTRAP 0x0 #define OP2_BPcc 0x1 #define OP2_Bicc 0x2 #define OP2_BPr 0x3 #define OP2_SETHI 0x4 #define OP2_FBPfcc 0x5 #define OP2_FBfcc 0x6 #define R_G0 0 #define R_O0 8 #define R_SP 14 #define R_I0 24 #define R_I1 25 #define R_I2 26 #define R_I3 27 #define R_I4 28 /* * Check the comment in fasttrap.h when changing these offsets or adding * new instructions. */ #define FASTTRAP_OFF_SAVE 64 #define FASTTRAP_OFF_RESTORE 68 #define FASTTRAP_OFF_FTRET 72 #define FASTTRAP_OFF_RETURN 76 #define BREAKPOINT_INSTR 0x91d02001 /* ta 1 */ /* * Tunable to let users turn off the fancy save instruction optimization. * If a program is non-ABI compliant, there's a possibility that the save * instruction optimization could cause an error. */ int fasttrap_optimize_save = 1; static uint64_t fasttrap_anarg(struct regs *rp, int argno) { uint64_t value; if (argno < 6) return ((&rp->r_o0)[argno]); if (curproc->p_model == DATAMODEL_NATIVE) { struct frame *fr = (struct frame *)(rp->r_sp + STACK_BIAS); DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); value = dtrace_fulword(&fr->fr_argd[argno]); DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT | CPU_DTRACE_BADADDR | CPU_DTRACE_BADALIGN); } else { struct frame32 *fr = (struct frame32 *)rp->r_sp; DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); value = dtrace_fuword32(&fr->fr_argd[argno]); DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT | CPU_DTRACE_BADADDR | CPU_DTRACE_BADALIGN); } return (value); } static ulong_t fasttrap_getreg(struct regs *, uint_t); static void fasttrap_putreg(struct regs *, uint_t, ulong_t); static void fasttrap_usdt_args(fasttrap_probe_t *probe, struct regs *rp, uint_t fake_restore, int argc, uintptr_t *argv) { int i, x, cap = MIN(argc, probe->ftp_nargs); int inc = (fake_restore ? 16 : 0); /* * The only way we'll hit the fake_restore case is if a USDT probe is * invoked as a tail-call. While it wouldn't be incorrect, we can * avoid a call to fasttrap_getreg(), and safely use rp->r_sp * directly since a tail-call can't be made if the invoked function * would use the argument dump space (i.e. if there were more than * 6 arguments). We take this shortcut because unconditionally rooting * around for R_FP (R_SP + 16) would be unnecessarily painful. */ if (curproc->p_model == DATAMODEL_NATIVE) { struct frame *fr = (struct frame *)(rp->r_sp + STACK_BIAS); uintptr_t v; for (i = 0; i < cap; i++) { x = probe->ftp_argmap[i]; if (x < 6) argv[i] = fasttrap_getreg(rp, R_O0 + x + inc); else if (fasttrap_fulword(&fr->fr_argd[x], &v) != 0) argv[i] = 0; } } else { struct frame32 *fr = (struct frame32 *)rp->r_sp; uint32_t v; for (i = 0; i < cap; i++) { x = probe->ftp_argmap[i]; if (x < 6) argv[i] = fasttrap_getreg(rp, R_O0 + x + inc); else if (fasttrap_fuword32(&fr->fr_argd[x], &v) != 0) argv[i] = 0; } } for (; i < argc; i++) { argv[i] = 0; } } static void fasttrap_return_common(struct regs *rp, uintptr_t pc, pid_t pid, uint_t fake_restore) { fasttrap_tracepoint_t *tp; fasttrap_bucket_t *bucket; fasttrap_id_t *id; kmutex_t *pid_mtx; dtrace_icookie_t cookie; pid_mtx = &cpu_core[CPU->cpu_id].cpuc_pid_lock; mutex_enter(pid_mtx); bucket = &fasttrap_tpoints.fth_table[FASTTRAP_TPOINTS_INDEX(pid, pc)]; for (tp = bucket->ftb_data; tp != NULL; tp = tp->ftt_next) { if (pid == tp->ftt_pid && pc == tp->ftt_pc && tp->ftt_proc->ftpc_acount != 0) break; } /* * Don't sweat it if we can't find the tracepoint again; unlike * when we're in fasttrap_pid_probe(), finding the tracepoint here * is not essential to the correct execution of the process. */ if (tp == NULL || tp->ftt_retids == NULL) { mutex_exit(pid_mtx); return; } for (id = tp->ftt_retids; id != NULL; id = id->fti_next) { fasttrap_probe_t *probe = id->fti_probe; if (id->fti_ptype == DTFTP_POST_OFFSETS) { if (probe->ftp_argmap != NULL && fake_restore) { uintptr_t t[5]; fasttrap_usdt_args(probe, rp, fake_restore, sizeof (t) / sizeof (t[0]), t); cookie = dtrace_interrupt_disable(); DTRACE_CPUFLAG_SET(CPU_DTRACE_FAKERESTORE); dtrace_probe(probe->ftp_id, t[0], t[1], t[2], t[3], t[4]); DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_FAKERESTORE); dtrace_interrupt_enable(cookie); } else if (probe->ftp_argmap != NULL) { uintptr_t t[5]; fasttrap_usdt_args(probe, rp, fake_restore, sizeof (t) / sizeof (t[0]), t); dtrace_probe(probe->ftp_id, t[0], t[1], t[2], t[3], t[4]); } else if (fake_restore) { uintptr_t arg0 = fasttrap_getreg(rp, R_I0); uintptr_t arg1 = fasttrap_getreg(rp, R_I1); uintptr_t arg2 = fasttrap_getreg(rp, R_I2); uintptr_t arg3 = fasttrap_getreg(rp, R_I3); uintptr_t arg4 = fasttrap_getreg(rp, R_I4); cookie = dtrace_interrupt_disable(); DTRACE_CPUFLAG_SET(CPU_DTRACE_FAKERESTORE); dtrace_probe(probe->ftp_id, arg0, arg1, arg2, arg3, arg4); DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_FAKERESTORE); dtrace_interrupt_enable(cookie); } else { dtrace_probe(probe->ftp_id, rp->r_o0, rp->r_o1, rp->r_o2, rp->r_o3, rp->r_o4); } continue; } /* * If this is only a possible return point, we must * be looking at a potential tail call in leaf context. * If the %npc is still within this function, then we * must have misidentified a jmpl as a tail-call when it * is, in fact, part of a jump table. It would be nice to * remove this tracepoint, but this is neither the time * nor the place. */ if ((tp->ftt_flags & FASTTRAP_F_RETMAYBE) && rp->r_npc - probe->ftp_faddr < probe->ftp_fsize) continue; /* * It's possible for a function to branch to the delay slot * of an instruction that we've identified as a return site. * We can dectect this spurious return probe activation by * observing that in this case %npc will be %pc + 4 and %npc * will be inside the current function (unless the user is * doing _crazy_ instruction picking in which case there's * very little we can do). The second check is important * in case the last instructions of a function make a tail- * call to the function located immediately subsequent. */ if (rp->r_npc == rp->r_pc + 4 && rp->r_npc - probe->ftp_faddr < probe->ftp_fsize) continue; /* * The first argument is the offset of return tracepoint * in the function; the remaining arguments are the return * values. * * If fake_restore is set, we need to pull the return values * out of the %i's rather than the %o's -- a little trickier. */ if (!fake_restore) { dtrace_probe(probe->ftp_id, pc - probe->ftp_faddr, rp->r_o0, rp->r_o1, rp->r_o2, rp->r_o3); } else { uintptr_t arg0 = fasttrap_getreg(rp, R_I0); uintptr_t arg1 = fasttrap_getreg(rp, R_I1); uintptr_t arg2 = fasttrap_getreg(rp, R_I2); uintptr_t arg3 = fasttrap_getreg(rp, R_I3); cookie = dtrace_interrupt_disable(); DTRACE_CPUFLAG_SET(CPU_DTRACE_FAKERESTORE); dtrace_probe(probe->ftp_id, pc - probe->ftp_faddr, arg0, arg1, arg2, arg3); DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_FAKERESTORE); dtrace_interrupt_enable(cookie); } } mutex_exit(pid_mtx); } int fasttrap_pid_probe(struct regs *rp) { proc_t *p = curproc; fasttrap_tracepoint_t *tp, tp_local; fasttrap_id_t *id; pid_t pid; uintptr_t pc = rp->r_pc; uintptr_t npc = rp->r_npc; uintptr_t orig_pc = pc; fasttrap_bucket_t *bucket; kmutex_t *pid_mtx; uint_t fake_restore = 0, is_enabled = 0; dtrace_icookie_t cookie; /* * It's possible that a user (in a veritable orgy of bad planning) * could redirect this thread's flow of control before it reached the * return probe fasttrap. In this case we need to kill the process * since it's in a unrecoverable state. */ if (curthread->t_dtrace_step) { ASSERT(curthread->t_dtrace_on); fasttrap_sigtrap(p, curthread, pc); return (0); } /* * Clear all user tracing flags. */ curthread->t_dtrace_ft = 0; curthread->t_dtrace_pc = 0; curthread->t_dtrace_npc = 0; curthread->t_dtrace_scrpc = 0; curthread->t_dtrace_astpc = 0; /* * Treat a child created by a call to vfork(2) as if it were its * parent. We know that there's only one thread of control in such a * process: this one. */ while (p->p_flag & SVFORK) { p = p->p_parent; } pid = p->p_pid; pid_mtx = &cpu_core[CPU->cpu_id].cpuc_pid_lock; mutex_enter(pid_mtx); bucket = &fasttrap_tpoints.fth_table[FASTTRAP_TPOINTS_INDEX(pid, pc)]; /* * Lookup the tracepoint that the process just hit. */ for (tp = bucket->ftb_data; tp != NULL; tp = tp->ftt_next) { if (pid == tp->ftt_pid && pc == tp->ftt_pc && tp->ftt_proc->ftpc_acount != 0) break; } /* * If we couldn't find a matching tracepoint, either a tracepoint has * been inserted without using the pid ioctl interface (see * fasttrap_ioctl), or somehow we have mislaid this tracepoint. */ if (tp == NULL) { mutex_exit(pid_mtx); return (-1); } for (id = tp->ftt_ids; id != NULL; id = id->fti_next) { fasttrap_probe_t *probe = id->fti_probe; int isentry = (id->fti_ptype == DTFTP_ENTRY); if (id->fti_ptype == DTFTP_IS_ENABLED) { is_enabled = 1; continue; } /* * We note that this was an entry probe to help ustack() find * the first caller. */ if (isentry) { cookie = dtrace_interrupt_disable(); DTRACE_CPUFLAG_SET(CPU_DTRACE_ENTRY); } dtrace_probe(probe->ftp_id, rp->r_o0, rp->r_o1, rp->r_o2, rp->r_o3, rp->r_o4); if (isentry) { DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_ENTRY); dtrace_interrupt_enable(cookie); } } /* * We're about to do a bunch of work so we cache a local copy of * the tracepoint to emulate the instruction, and then find the * tracepoint again later if we need to light up any return probes. */ tp_local = *tp; mutex_exit(pid_mtx); tp = &tp_local; /* * If there's an is-enabled probe conntected to this tracepoint it * means that there was a 'mov %g0, %o0' instruction that was placed * there by DTrace when the binary was linked. As this probe is, in * fact, enabled, we need to stuff 1 into %o0. Accordingly, we can * bypass all the instruction emulation logic since we know the * inevitable result. It's possible that a user could construct a * scenario where the 'is-enabled' probe was on some other * instruction, but that would be a rather exotic way to shoot oneself * in the foot. */ if (is_enabled) { rp->r_o0 = 1; pc = rp->r_npc; npc = pc + 4; goto done; } /* * We emulate certain types of instructions to ensure correctness * (in the case of position dependent instructions) or optimize * common cases. The rest we have the thread execute back in user- * land. */ switch (tp->ftt_type) { case FASTTRAP_T_SAVE: { int32_t imm; /* * This an optimization to let us handle function entry * probes more efficiently. Many functions begin with a save * instruction that follows the pattern: * save %sp, , %sp * * Meanwhile, we've stashed the instruction: * save %g1, %g0, %sp * * off of %g7, so all we have to do is stick the right value * into %g1 and reset %pc to point to the instruction we've * cleverly hidden (%npc should not be touched). */ imm = tp->ftt_instr << 19; imm >>= 19; rp->r_g1 = rp->r_sp + imm; pc = rp->r_g7 + FASTTRAP_OFF_SAVE; break; } case FASTTRAP_T_RESTORE: { ulong_t value; uint_t rd; /* * This is an optimization to let us handle function * return probes more efficiently. Most non-leaf functions * end with the sequence: * ret * restore , , %oX * * We've stashed the instruction: * restore %g0, %g0, %g0 * * off of %g7 so we just need to place the correct value * in the right %i register (since after our fake-o * restore, the %i's will become the %o's) and set the %pc * to point to our hidden restore. We also set fake_restore to * let fasttrap_return_common() know that it will find the * return values in the %i's rather than the %o's. */ if (I(tp->ftt_instr)) { int32_t imm; imm = tp->ftt_instr << 19; imm >>= 19; value = fasttrap_getreg(rp, RS1(tp->ftt_instr)) + imm; } else { value = fasttrap_getreg(rp, RS1(tp->ftt_instr)) + fasttrap_getreg(rp, RS2(tp->ftt_instr)); } /* * Convert %o's to %i's; leave %g's as they are. */ rd = RD(tp->ftt_instr); fasttrap_putreg(rp, ((rd & 0x18) == 0x8) ? rd + 16 : rd, value); pc = rp->r_g7 + FASTTRAP_OFF_RESTORE; fake_restore = 1; break; } case FASTTRAP_T_RETURN: { uintptr_t target; /* * A return instruction is like a jmpl (without the link * part) that executes an implicit restore. We've stashed * the instruction: * return %o0 * * off of %g7 so we just need to place the target in %o0 * and set the %pc to point to the stashed return instruction. * We use %o0 since that register disappears after the return * executes, erasing any evidence of this tampering. */ if (I(tp->ftt_instr)) { int32_t imm; imm = tp->ftt_instr << 19; imm >>= 19; target = fasttrap_getreg(rp, RS1(tp->ftt_instr)) + imm; } else { target = fasttrap_getreg(rp, RS1(tp->ftt_instr)) + fasttrap_getreg(rp, RS2(tp->ftt_instr)); } fasttrap_putreg(rp, R_O0, target); pc = rp->r_g7 + FASTTRAP_OFF_RETURN; fake_restore = 1; break; } case FASTTRAP_T_OR: { ulong_t value; if (I(tp->ftt_instr)) { int32_t imm; imm = tp->ftt_instr << 19; imm >>= 19; value = fasttrap_getreg(rp, RS1(tp->ftt_instr)) | imm; } else { value = fasttrap_getreg(rp, RS1(tp->ftt_instr)) | fasttrap_getreg(rp, RS2(tp->ftt_instr)); } fasttrap_putreg(rp, RD(tp->ftt_instr), value); pc = rp->r_npc; npc = pc + 4; break; } case FASTTRAP_T_SETHI: if (RD(tp->ftt_instr) != R_G0) { uint32_t imm32 = tp->ftt_instr << 10; fasttrap_putreg(rp, RD(tp->ftt_instr), (ulong_t)imm32); } pc = rp->r_npc; npc = pc + 4; break; case FASTTRAP_T_CCR: { uint_t c, v, z, n, taken; uint_t ccr = rp->r_tstate >> TSTATE_CCR_SHIFT; if (tp->ftt_cc != 0) ccr >>= 4; c = (ccr >> 0) & 1; v = (ccr >> 1) & 1; z = (ccr >> 2) & 1; n = (ccr >> 3) & 1; switch (tp->ftt_code) { case 0x0: /* BN */ taken = 0; break; case 0x1: /* BE */ taken = z; break; case 0x2: /* BLE */ taken = z | (n ^ v); break; case 0x3: /* BL */ taken = n ^ v; break; case 0x4: /* BLEU */ taken = c | z; break; case 0x5: /* BCS (BLU) */ taken = c; break; case 0x6: /* BNEG */ taken = n; break; case 0x7: /* BVS */ taken = v; break; case 0x8: /* BA */ /* * We handle the BA case differently since the annul * bit means something slightly different. */ panic("fasttrap: mishandled a branch"); taken = 1; break; case 0x9: /* BNE */ taken = ~z; break; case 0xa: /* BG */ taken = ~(z | (n ^ v)); break; case 0xb: /* BGE */ taken = ~(n ^ v); break; case 0xc: /* BGU */ taken = ~(c | z); break; case 0xd: /* BCC (BGEU) */ taken = ~c; break; case 0xe: /* BPOS */ taken = ~n; break; case 0xf: /* BVC */ taken = ~v; break; } if (taken & 1) { pc = rp->r_npc; npc = tp->ftt_dest; } else if (tp->ftt_flags & FASTTRAP_F_ANNUL) { /* * Untaken annulled branches don't execute the * instruction in the delay slot. */ pc = rp->r_npc + 4; npc = pc + 4; } else { pc = rp->r_npc; npc = pc + 4; } break; } case FASTTRAP_T_FCC: { uint_t fcc; uint_t taken; uint64_t fsr; dtrace_getfsr(&fsr); if (tp->ftt_cc == 0) { fcc = (fsr >> 10) & 0x3; } else { uint_t shift; ASSERT(tp->ftt_cc <= 3); shift = 30 + tp->ftt_cc * 2; fcc = (fsr >> shift) & 0x3; } switch (tp->ftt_code) { case 0x0: /* FBN */ taken = (1 << fcc) & (0|0|0|0); break; case 0x1: /* FBNE */ taken = (1 << fcc) & (8|4|2|0); break; case 0x2: /* FBLG */ taken = (1 << fcc) & (0|4|2|0); break; case 0x3: /* FBUL */ taken = (1 << fcc) & (8|0|2|0); break; case 0x4: /* FBL */ taken = (1 << fcc) & (0|0|2|0); break; case 0x5: /* FBUG */ taken = (1 << fcc) & (8|4|0|0); break; case 0x6: /* FBG */ taken = (1 << fcc) & (0|4|0|0); break; case 0x7: /* FBU */ taken = (1 << fcc) & (8|0|0|0); break; case 0x8: /* FBA */ /* * We handle the FBA case differently since the annul * bit means something slightly different. */ panic("fasttrap: mishandled a branch"); taken = (1 << fcc) & (8|4|2|1); break; case 0x9: /* FBE */ taken = (1 << fcc) & (0|0|0|1); break; case 0xa: /* FBUE */ taken = (1 << fcc) & (8|0|0|1); break; case 0xb: /* FBGE */ taken = (1 << fcc) & (0|4|0|1); break; case 0xc: /* FBUGE */ taken = (1 << fcc) & (8|4|0|1); break; case 0xd: /* FBLE */ taken = (1 << fcc) & (0|0|2|1); break; case 0xe: /* FBULE */ taken = (1 << fcc) & (8|0|2|1); break; case 0xf: /* FBO */ taken = (1 << fcc) & (0|4|2|1); break; } if (taken) { pc = rp->r_npc; npc = tp->ftt_dest; } else if (tp->ftt_flags & FASTTRAP_F_ANNUL) { /* * Untaken annulled branches don't execute the * instruction in the delay slot. */ pc = rp->r_npc + 4; npc = pc + 4; } else { pc = rp->r_npc; npc = pc + 4; } break; } case FASTTRAP_T_REG: { int64_t value; uint_t taken; uint_t reg = RS1(tp->ftt_instr); /* * An ILP32 process shouldn't be using a branch predicated on * an %i or an %l since it would violate the ABI. It's a * violation of the ABI because we can't ensure deterministic * behavior. We should have identified this case when we * enabled the probe. */ ASSERT(p->p_model == DATAMODEL_LP64 || reg < 16); value = (int64_t)fasttrap_getreg(rp, reg); switch (tp->ftt_code) { case 0x1: /* BRZ */ taken = (value == 0); break; case 0x2: /* BRLEZ */ taken = (value <= 0); break; case 0x3: /* BRLZ */ taken = (value < 0); break; case 0x5: /* BRNZ */ taken = (value != 0); break; case 0x6: /* BRGZ */ taken = (value > 0); break; case 0x7: /* BRGEZ */ taken = (value >= 0); break; default: case 0x0: case 0x4: panic("fasttrap: mishandled a branch"); } if (taken) { pc = rp->r_npc; npc = tp->ftt_dest; } else if (tp->ftt_flags & FASTTRAP_F_ANNUL) { /* * Untaken annulled branches don't execute the * instruction in the delay slot. */ pc = rp->r_npc + 4; npc = pc + 4; } else { pc = rp->r_npc; npc = pc + 4; } break; } case FASTTRAP_T_ALWAYS: /* * BAs, BA,As... */ if (tp->ftt_flags & FASTTRAP_F_ANNUL) { /* * Annulled branch always instructions never execute * the instruction in the delay slot. */ pc = tp->ftt_dest; npc = tp->ftt_dest + 4; } else { pc = rp->r_npc; npc = tp->ftt_dest; } break; case FASTTRAP_T_RDPC: fasttrap_putreg(rp, RD(tp->ftt_instr), rp->r_pc); pc = rp->r_npc; npc = pc + 4; break; case FASTTRAP_T_CALL: /* * It's a call _and_ link remember... */ rp->r_o7 = rp->r_pc; pc = rp->r_npc; npc = tp->ftt_dest; break; case FASTTRAP_T_JMPL: pc = rp->r_npc; if (I(tp->ftt_instr)) { uint_t rs1 = RS1(tp->ftt_instr); int32_t imm; imm = tp->ftt_instr << 19; imm >>= 19; npc = fasttrap_getreg(rp, rs1) + imm; } else { uint_t rs1 = RS1(tp->ftt_instr); uint_t rs2 = RS2(tp->ftt_instr); npc = fasttrap_getreg(rp, rs1) + fasttrap_getreg(rp, rs2); } /* * Do the link part of the jump-and-link instruction. */ fasttrap_putreg(rp, RD(tp->ftt_instr), rp->r_pc); break; case FASTTRAP_T_COMMON: { curthread->t_dtrace_scrpc = rp->r_g7; curthread->t_dtrace_astpc = rp->r_g7 + FASTTRAP_OFF_FTRET; /* * Copy the instruction to a reserved location in the * user-land thread structure, then set the PC to that * location and leave the NPC alone. We take pains to ensure * consistency in the instruction stream (See SPARC * Architecture Manual Version 9, sections 8.4.7, A.20, and * H.1.6; UltraSPARC I/II User's Manual, sections 3.1.1.1, * and 13.6.4) by using the ASI ASI_BLK_COMMIT_S to copy the * instruction into the user's address space without * bypassing the I$. There's no AS_USER version of this ASI * (as exist for other ASIs) so we use the lofault * mechanism to catch faults. */ if (dtrace_blksuword32(rp->r_g7, &tp->ftt_instr, 1) == -1) { /* * If the copyout fails, then the process's state * is not consistent (the effects of the traced * instruction will never be seen). This process * cannot be allowed to continue execution. */ fasttrap_sigtrap(curproc, curthread, pc); return (0); } curthread->t_dtrace_pc = pc; curthread->t_dtrace_npc = npc; curthread->t_dtrace_on = 1; pc = curthread->t_dtrace_scrpc; if (tp->ftt_retids != NULL) { curthread->t_dtrace_step = 1; curthread->t_dtrace_ret = 1; npc = curthread->t_dtrace_astpc; } break; } default: panic("fasttrap: mishandled an instruction"); } /* * This bit me in the ass a couple of times, so lets toss this * in as a cursory sanity check. */ ASSERT(pc != rp->r_g7 + 4); ASSERT(pc != rp->r_g7 + 8); done: /* * If there were no return probes when we first found the tracepoint, * we should feel no obligation to honor any return probes that were * subsequently enabled -- they'll just have to wait until the next * time around. */ if (tp->ftt_retids != NULL) { /* * We need to wait until the results of the instruction are * apparent before invoking any return probes. If this * instruction was emulated we can just call * fasttrap_return_common(); if it needs to be executed, we * need to wait until we return to the kernel. */ if (tp->ftt_type != FASTTRAP_T_COMMON) { fasttrap_return_common(rp, orig_pc, pid, fake_restore); } else { ASSERT(curthread->t_dtrace_ret != 0); ASSERT(curthread->t_dtrace_pc == orig_pc); ASSERT(curthread->t_dtrace_scrpc == rp->r_g7); ASSERT(npc == curthread->t_dtrace_astpc); } } ASSERT(pc != 0); rp->r_pc = pc; rp->r_npc = npc; return (0); } int fasttrap_return_probe(struct regs *rp) { proc_t *p = ttoproc(curthread); pid_t pid; uintptr_t pc = curthread->t_dtrace_pc; uintptr_t npc = curthread->t_dtrace_npc; curthread->t_dtrace_pc = 0; curthread->t_dtrace_npc = 0; curthread->t_dtrace_scrpc = 0; curthread->t_dtrace_astpc = 0; /* * Treat a child created by a call to vfork(2) as if it were its * parent. We know there's only one thread of control in such a * process: this one. */ while (p->p_flag & SVFORK) { p = p->p_parent; } /* * We set the %pc and %npc to their values when the traced * instruction was initially executed so that it appears to * dtrace_probe() that we're on the original instruction, and so that * the user can't easily detect our complex web of lies. * dtrace_return_probe() (our caller) will correctly set %pc and %npc * after we return. */ rp->r_pc = pc; rp->r_npc = npc; pid = p->p_pid; fasttrap_return_common(rp, pc, pid, 0); return (0); } int fasttrap_tracepoint_install(proc_t *p, fasttrap_tracepoint_t *tp) { fasttrap_instr_t instr = FASTTRAP_INSTR; if (uwrite(p, &instr, 4, tp->ftt_pc) != 0) return (-1); return (0); } int fasttrap_tracepoint_remove(proc_t *p, fasttrap_tracepoint_t *tp) { fasttrap_instr_t instr; /* * Distinguish between read or write failures and a changed * instruction. */ if (uread(p, &instr, 4, tp->ftt_pc) != 0) return (0); if (instr != FASTTRAP_INSTR && instr != BREAKPOINT_INSTR) return (0); if (uwrite(p, &tp->ftt_instr, 4, tp->ftt_pc) != 0) return (-1); return (0); } int fasttrap_tracepoint_init(proc_t *p, fasttrap_tracepoint_t *tp, uintptr_t pc, fasttrap_probe_type_t type) { uint32_t instr; int32_t disp; /* * Read the instruction at the given address out of the process's * address space. We don't have to worry about a debugger * changing this instruction before we overwrite it with our trap * instruction since P_PR_LOCK is set. */ if (uread(p, &instr, 4, pc) != 0) return (-1); /* * Decode the instruction to fill in the probe flags. We can have * the process execute most instructions on its own using a pc/npc * trick, but pc-relative control transfer present a problem since * we're relocating the instruction. We emulate these instructions * in the kernel. We assume a default type and over-write that as * needed. * * pc-relative instructions must be emulated for correctness; * other instructions (which represent a large set of commonly traced * instructions) are emulated or otherwise optimized for performance. */ tp->ftt_type = FASTTRAP_T_COMMON; if (OP(instr) == 1) { /* * Call instructions. */ tp->ftt_type = FASTTRAP_T_CALL; disp = DISP30(instr) << 2; tp->ftt_dest = pc + (intptr_t)disp; } else if (OP(instr) == 0) { /* * Branch instructions. * * Unconditional branches need careful attention when they're * annulled: annulled unconditional branches never execute * the instruction in the delay slot. */ switch (OP2(instr)) { case OP2_ILLTRAP: case 0x7: /* * The compiler may place an illtrap after a call to * a function that returns a structure. In the case of * a returned structure, the compiler places an illtrap * whose const22 field is the size of the returned * structure immediately following the delay slot of * the call. To stay out of the way, we refuse to * place tracepoints on top of illtrap instructions. * * This is one of the dumbest architectural decisions * I've ever had to work around. * * We also identify the only illegal op2 value (See * SPARC Architecture Manual Version 9, E.2 table 31). */ return (-1); case OP2_BPcc: if (COND(instr) == 8) { tp->ftt_type = FASTTRAP_T_ALWAYS; } else { /* * Check for an illegal instruction. */ if (CC(instr) & 1) return (-1); tp->ftt_type = FASTTRAP_T_CCR; tp->ftt_cc = CC(instr); tp->ftt_code = COND(instr); } if (A(instr) != 0) tp->ftt_flags |= FASTTRAP_F_ANNUL; disp = DISP19(instr); disp <<= 13; disp >>= 11; tp->ftt_dest = pc + (intptr_t)disp; break; case OP2_Bicc: if (COND(instr) == 8) { tp->ftt_type = FASTTRAP_T_ALWAYS; } else { tp->ftt_type = FASTTRAP_T_CCR; tp->ftt_cc = 0; tp->ftt_code = COND(instr); } if (A(instr) != 0) tp->ftt_flags |= FASTTRAP_F_ANNUL; disp = DISP22(instr); disp <<= 10; disp >>= 8; tp->ftt_dest = pc + (intptr_t)disp; break; case OP2_BPr: /* * Check for an illegal instruction. */ if ((RCOND(instr) & 3) == 0) return (-1); /* * It's a violation of the v8plus ABI to use a * register-predicated branch in a 32-bit app if * the register used is an %l or an %i (%gs and %os * are legit because they're not saved to the stack * in 32-bit words when we take a trap). */ if (p->p_model == DATAMODEL_ILP32 && RS1(instr) >= 16) return (-1); tp->ftt_type = FASTTRAP_T_REG; if (A(instr) != 0) tp->ftt_flags |= FASTTRAP_F_ANNUL; disp = DISP16(instr); disp <<= 16; disp >>= 14; tp->ftt_dest = pc + (intptr_t)disp; tp->ftt_code = RCOND(instr); break; case OP2_SETHI: tp->ftt_type = FASTTRAP_T_SETHI; break; case OP2_FBPfcc: if (COND(instr) == 8) { tp->ftt_type = FASTTRAP_T_ALWAYS; } else { tp->ftt_type = FASTTRAP_T_FCC; tp->ftt_cc = CC(instr); tp->ftt_code = COND(instr); } if (A(instr) != 0) tp->ftt_flags |= FASTTRAP_F_ANNUL; disp = DISP19(instr); disp <<= 13; disp >>= 11; tp->ftt_dest = pc + (intptr_t)disp; break; case OP2_FBfcc: if (COND(instr) == 8) { tp->ftt_type = FASTTRAP_T_ALWAYS; } else { tp->ftt_type = FASTTRAP_T_FCC; tp->ftt_cc = 0; tp->ftt_code = COND(instr); } if (A(instr) != 0) tp->ftt_flags |= FASTTRAP_F_ANNUL; disp = DISP22(instr); disp <<= 10; disp >>= 8; tp->ftt_dest = pc + (intptr_t)disp; break; } } else if (OP(instr) == 2) { switch (OP3(instr)) { case OP3_RETURN: tp->ftt_type = FASTTRAP_T_RETURN; break; case OP3_JMPL: tp->ftt_type = FASTTRAP_T_JMPL; break; case OP3_RD: if (RS1(instr) == 5) tp->ftt_type = FASTTRAP_T_RDPC; break; case OP3_SAVE: /* * We optimize for save instructions at function * entry; see the comment in fasttrap_pid_probe() * (near FASTTRAP_T_SAVE) for details. */ if (fasttrap_optimize_save != 0 && type == DTFTP_ENTRY && I(instr) == 1 && RD(instr) == R_SP) tp->ftt_type = FASTTRAP_T_SAVE; break; case OP3_RESTORE: /* * We optimize restore instructions at function * return; see the comment in fasttrap_pid_probe() * (near FASTTRAP_T_RESTORE) for details. * * rd must be an %o or %g register. */ if ((RD(instr) & 0x10) == 0) tp->ftt_type = FASTTRAP_T_RESTORE; break; case OP3_OR: /* * A large proportion of instructions in the delay * slot of retl instructions are or's so we emulate * these downstairs as an optimization. */ tp->ftt_type = FASTTRAP_T_OR; break; case OP3_TCC: /* * Breakpoint instructions are effectively position- * dependent since the debugger uses the %pc value * to lookup which breakpoint was executed. As a * result, we can't actually instrument breakpoints. */ if (SW_TRAP(instr) == ST_BREAKPOINT) return (-1); break; case 0x19: case 0x1d: case 0x29: case 0x33: case 0x3f: /* * Identify illegal instructions (See SPARC * Architecture Manual Version 9, E.2 table 32). */ return (-1); } } else if (OP(instr) == 3) { uint32_t op3 = OP3(instr); /* * Identify illegal instructions (See SPARC Architecture * Manual Version 9, E.2 table 33). */ if ((op3 & 0x28) == 0x28) { if (op3 != OP3_PREFETCH && op3 != OP3_CASA && op3 != OP3_PREFETCHA && op3 != OP3_CASXA) return (-1); } else { if ((op3 & 0x0f) == 0x0c || (op3 & 0x3b) == 0x31) return (-1); } } tp->ftt_instr = instr; /* * We don't know how this tracepoint is going to be used, but in case * it's used as part of a function return probe, we need to indicate * whether it's always a return site or only potentially a return * site. If it's part of a return probe, it's always going to be a * return from that function if it's a restore instruction or if * the previous instruction was a return. If we could reliably * distinguish jump tables from return sites, this wouldn't be * necessary. */ if (tp->ftt_type != FASTTRAP_T_RESTORE && (uread(p, &instr, 4, pc - sizeof (instr)) != 0 || !(OP(instr) == 2 && OP3(instr) == OP3_RETURN))) tp->ftt_flags |= FASTTRAP_F_RETMAYBE; return (0); } /*ARGSUSED*/ uint64_t fasttrap_pid_getarg(void *arg, dtrace_id_t id, void *parg, int argno, int aframes) { return (fasttrap_anarg(ttolwp(curthread)->lwp_regs, argno)); } /*ARGSUSED*/ uint64_t fasttrap_usdt_getarg(void *arg, dtrace_id_t id, void *parg, int argno, int aframes) { return (fasttrap_anarg(ttolwp(curthread)->lwp_regs, argno)); } static uint64_t fasttrap_getreg_fast_cnt; static uint64_t fasttrap_getreg_mpcb_cnt; static uint64_t fasttrap_getreg_slow_cnt; static ulong_t fasttrap_getreg(struct regs *rp, uint_t reg) { ulong_t value; dtrace_icookie_t cookie; struct machpcb *mpcb; extern ulong_t dtrace_getreg_win(uint_t, uint_t); /* * We have the %os and %gs in our struct regs, but if we need to * snag a %l or %i we need to go scrounging around in the process's * address space. */ if (reg == 0) return (0); if (reg < 16) return ((&rp->r_g1)[reg - 1]); /* * Before we look at the user's stack, we'll check the register * windows to see if the information we want is in there. */ cookie = dtrace_interrupt_disable(); if (dtrace_getotherwin() > 0) { value = dtrace_getreg_win(reg, 1); dtrace_interrupt_enable(cookie); - atomic_add_64(&fasttrap_getreg_fast_cnt, 1); + atomic_inc_64(&fasttrap_getreg_fast_cnt); return (value); } dtrace_interrupt_enable(cookie); /* * First check the machpcb structure to see if we've already read * in the register window we're looking for; if we haven't, (and * we probably haven't) try to copy in the value of the register. */ /* LINTED - alignment */ mpcb = (struct machpcb *)((caddr_t)rp - REGOFF); if (get_udatamodel() == DATAMODEL_NATIVE) { struct frame *fr = (struct frame *)(rp->r_sp + STACK_BIAS); if (mpcb->mpcb_wbcnt > 0) { struct rwindow *rwin = (void *)mpcb->mpcb_wbuf; int i = mpcb->mpcb_wbcnt; do { i--; if ((long)mpcb->mpcb_spbuf[i] != rp->r_sp) continue; - atomic_add_64(&fasttrap_getreg_mpcb_cnt, 1); + atomic_inc_64(&fasttrap_getreg_mpcb_cnt); return (rwin[i].rw_local[reg - 16]); } while (i > 0); } if (fasttrap_fulword(&fr->fr_local[reg - 16], &value) != 0) goto err; } else { struct frame32 *fr = (struct frame32 *)(uintptr_t)(caddr32_t)rp->r_sp; uint32_t *v32 = (uint32_t *)&value; if (mpcb->mpcb_wbcnt > 0) { struct rwindow32 *rwin = (void *)mpcb->mpcb_wbuf; int i = mpcb->mpcb_wbcnt; do { i--; if ((long)mpcb->mpcb_spbuf[i] != rp->r_sp) continue; - atomic_add_64(&fasttrap_getreg_mpcb_cnt, 1); + atomic_inc_64(&fasttrap_getreg_mpcb_cnt); return (rwin[i].rw_local[reg - 16]); } while (i > 0); } if (fasttrap_fuword32(&fr->fr_local[reg - 16], &v32[1]) != 0) goto err; v32[0] = 0; } - atomic_add_64(&fasttrap_getreg_slow_cnt, 1); + atomic_inc_64(&fasttrap_getreg_slow_cnt); return (value); err: /* * If the copy in failed, the process will be in a irrecoverable * state, and we have no choice but to kill it. */ kern_psignal(ttoproc(curthread), SIGILL); return (0); } static uint64_t fasttrap_putreg_fast_cnt; static uint64_t fasttrap_putreg_mpcb_cnt; static uint64_t fasttrap_putreg_slow_cnt; static void fasttrap_putreg(struct regs *rp, uint_t reg, ulong_t value) { dtrace_icookie_t cookie; struct machpcb *mpcb; extern void dtrace_putreg_win(uint_t, ulong_t); if (reg == 0) return; if (reg < 16) { (&rp->r_g1)[reg - 1] = value; return; } /* * If the user process is still using some register windows, we * can just place the value in the correct window. */ cookie = dtrace_interrupt_disable(); if (dtrace_getotherwin() > 0) { dtrace_putreg_win(reg, value); dtrace_interrupt_enable(cookie); - atomic_add_64(&fasttrap_putreg_fast_cnt, 1); + atomic_inc_64(&fasttrap_putreg_fast_cnt); return; } dtrace_interrupt_enable(cookie); /* * First see if there's a copy of the register window in the * machpcb structure that we can modify; if there isn't try to * copy out the value. If that fails, we try to create a new * register window in the machpcb structure. While this isn't * _precisely_ the intended use of the machpcb structure, it * can't cause any problems since we know at this point in the * code that all of the user's data have been flushed out of the * register file (since %otherwin is 0). */ /* LINTED - alignment */ mpcb = (struct machpcb *)((caddr_t)rp - REGOFF); if (get_udatamodel() == DATAMODEL_NATIVE) { struct frame *fr = (struct frame *)(rp->r_sp + STACK_BIAS); /* LINTED - alignment */ struct rwindow *rwin = (struct rwindow *)mpcb->mpcb_wbuf; if (mpcb->mpcb_wbcnt > 0) { int i = mpcb->mpcb_wbcnt; do { i--; if ((long)mpcb->mpcb_spbuf[i] != rp->r_sp) continue; rwin[i].rw_local[reg - 16] = value; - atomic_add_64(&fasttrap_putreg_mpcb_cnt, 1); + atomic_inc_64(&fasttrap_putreg_mpcb_cnt); return; } while (i > 0); } if (fasttrap_sulword(&fr->fr_local[reg - 16], value) != 0) { if (mpcb->mpcb_wbcnt >= MAXWIN || copyin(fr, &rwin[mpcb->mpcb_wbcnt], sizeof (*rwin)) != 0) goto err; rwin[mpcb->mpcb_wbcnt].rw_local[reg - 16] = value; mpcb->mpcb_spbuf[mpcb->mpcb_wbcnt] = (caddr_t)rp->r_sp; mpcb->mpcb_wbcnt++; - atomic_add_64(&fasttrap_putreg_mpcb_cnt, 1); + atomic_inc_64(&fasttrap_putreg_mpcb_cnt); return; } } else { struct frame32 *fr = (struct frame32 *)(uintptr_t)(caddr32_t)rp->r_sp; /* LINTED - alignment */ struct rwindow32 *rwin = (struct rwindow32 *)mpcb->mpcb_wbuf; uint32_t v32 = (uint32_t)value; if (mpcb->mpcb_wbcnt > 0) { int i = mpcb->mpcb_wbcnt; do { i--; if ((long)mpcb->mpcb_spbuf[i] != rp->r_sp) continue; rwin[i].rw_local[reg - 16] = v32; - atomic_add_64(&fasttrap_putreg_mpcb_cnt, 1); + atomic_inc_64(&fasttrap_putreg_mpcb_cnt); return; } while (i > 0); } if (fasttrap_suword32(&fr->fr_local[reg - 16], v32) != 0) { if (mpcb->mpcb_wbcnt >= MAXWIN || copyin(fr, &rwin[mpcb->mpcb_wbcnt], sizeof (*rwin)) != 0) goto err; rwin[mpcb->mpcb_wbcnt].rw_local[reg - 16] = v32; mpcb->mpcb_spbuf[mpcb->mpcb_wbcnt] = (caddr_t)rp->r_sp; mpcb->mpcb_wbcnt++; - atomic_add_64(&fasttrap_putreg_mpcb_cnt, 1); + atomic_inc_64(&fasttrap_putreg_mpcb_cnt); return; } } - atomic_add_64(&fasttrap_putreg_slow_cnt, 1); + atomic_inc_64(&fasttrap_putreg_slow_cnt); return; err: /* * If we couldn't record this register's value, the process is in an * irrecoverable state and we have no choice but to euthanize it. */ kern_psignal(ttoproc(curthread), SIGILL); } Index: stable/10 =================================================================== --- stable/10 (revision 271000) +++ stable/10 (revision 271001) Property changes on: stable/10 ___________________________________________________________________ Modified: svn:mergeinfo ## -0,0 +0,1 ## Merged /head:r270247