diff options
Diffstat (limited to 'fs/nfs/pnfs.c')
| -rw-r--r-- | fs/nfs/pnfs.c | 1645 |
1 files changed, 1314 insertions, 331 deletions
diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c index 1b1bc1a0fb0..6fdcd233d6f 100644 --- a/fs/nfs/pnfs.c +++ b/fs/nfs/pnfs.c @@ -28,10 +28,15 @@ */ #include <linux/nfs_fs.h> +#include <linux/nfs_page.h> +#include <linux/module.h> #include "internal.h" #include "pnfs.h" +#include "iostat.h" +#include "nfs4trace.h" #define NFSDBG_FACILITY NFSDBG_PNFS +#define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ) /* Locking: * @@ -67,6 +72,10 @@ find_pnfs_driver(u32 id) spin_lock(&pnfs_spinlock); local = find_pnfs_driver_locked(id); + if (local != NULL && !try_module_get(local->owner)) { + dprintk("%s: Could not grab reference on module\n", __func__); + local = NULL; + } spin_unlock(&pnfs_spinlock); return local; } @@ -75,7 +84,11 @@ void unset_pnfs_layoutdriver(struct nfs_server *nfss) { if (nfss->pnfs_curr_ld) { - nfss->pnfs_curr_ld->clear_layoutdriver(nfss); + if (nfss->pnfs_curr_ld->clear_layoutdriver) + nfss->pnfs_curr_ld->clear_layoutdriver(nfss); + /* Decrement the MDS count. Purge the deviceid cache if zero */ + if (atomic_dec_and_test(&nfss->nfs_client->cl_mds_count)) + nfs4_deviceid_purge_client(nfss->nfs_client); module_put(nfss->pnfs_curr_ld->owner); } nfss->pnfs_curr_ld = NULL; @@ -88,7 +101,8 @@ unset_pnfs_layoutdriver(struct nfs_server *nfss) * @id layout type. Zero (illegal layout type) indicates pNFS not in use. */ void -set_pnfs_layoutdriver(struct nfs_server *server, u32 id) +set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh, + u32 id) { struct pnfs_layoutdriver_type *ld_type = NULL; @@ -96,8 +110,8 @@ set_pnfs_layoutdriver(struct nfs_server *server, u32 id) goto out_no_driver; if (!(server->nfs_client->cl_exchange_flags & (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) { - printk(KERN_ERR "%s: id %u cl_exchange_flags 0x%x\n", __func__, - id, server->nfs_client->cl_exchange_flags); + printk(KERN_ERR "NFS: %s: id %u cl_exchange_flags 0x%x\n", + __func__, id, server->nfs_client->cl_exchange_flags); goto out_no_driver; } ld_type = find_pnfs_driver(id); @@ -110,18 +124,17 @@ set_pnfs_layoutdriver(struct nfs_server *server, u32 id) goto out_no_driver; } } - if (!try_module_get(ld_type->owner)) { - dprintk("%s: Could not grab reference on module\n", __func__); - goto out_no_driver; - } server->pnfs_curr_ld = ld_type; - if (ld_type->set_layoutdriver(server)) { - printk(KERN_ERR - "%s: Error initializing mount point for layout driver %u.\n", - __func__, id); + if (ld_type->set_layoutdriver + && ld_type->set_layoutdriver(server, mntfh)) { + printk(KERN_ERR "NFS: %s: Error initializing pNFS layout " + "driver %u.\n", __func__, id); module_put(ld_type->owner); goto out_no_driver; } + /* Bump the MDS count */ + atomic_inc(&server->nfs_client->cl_mds_count); + dprintk("%s: pNFS module for %u set\n", __func__, id); return; @@ -137,11 +150,11 @@ pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type) struct pnfs_layoutdriver_type *tmp; if (ld_type->id == 0) { - printk(KERN_ERR "%s id 0 is reserved\n", __func__); + printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__); return status; } if (!ld_type->alloc_lseg || !ld_type->free_lseg) { - printk(KERN_ERR "%s Layout driver must provide " + printk(KERN_ERR "NFS: %s Layout driver must provide " "alloc_lseg and free_lseg.\n", __func__); return status; } @@ -154,7 +167,7 @@ pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type) dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id, ld_type->name); } else { - printk(KERN_ERR "%s Module with id %d already loaded!\n", + printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n", __func__, ld_type->id); } spin_unlock(&pnfs_spinlock); @@ -179,95 +192,240 @@ EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver); /* Need to hold i_lock if caller does not already hold reference */ void -get_layout_hdr(struct pnfs_layout_hdr *lo) +pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo) { atomic_inc(&lo->plh_refcount); } +static struct pnfs_layout_hdr * +pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags) +{ + struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld; + return ld->alloc_layout_hdr(ino, gfp_flags); +} + static void -destroy_layout_hdr(struct pnfs_layout_hdr *lo) +pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo) { - dprintk("%s: freeing layout cache %p\n", __func__, lo); - BUG_ON(!list_empty(&lo->plh_layouts)); - NFS_I(lo->plh_inode)->layout = NULL; - kfree(lo); + struct nfs_server *server = NFS_SERVER(lo->plh_inode); + struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld; + + if (!list_empty(&lo->plh_layouts)) { + struct nfs_client *clp = server->nfs_client; + + spin_lock(&clp->cl_lock); + list_del_init(&lo->plh_layouts); + spin_unlock(&clp->cl_lock); + } + put_rpccred(lo->plh_lc_cred); + return ld->free_layout_hdr(lo); } static void -put_layout_hdr_locked(struct pnfs_layout_hdr *lo) +pnfs_detach_layout_hdr(struct pnfs_layout_hdr *lo) { - if (atomic_dec_and_test(&lo->plh_refcount)) - destroy_layout_hdr(lo); + struct nfs_inode *nfsi = NFS_I(lo->plh_inode); + dprintk("%s: freeing layout cache %p\n", __func__, lo); + nfsi->layout = NULL; + /* Reset MDS Threshold I/O counters */ + nfsi->write_io = 0; + nfsi->read_io = 0; } void -put_layout_hdr(struct pnfs_layout_hdr *lo) +pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo) { struct inode *inode = lo->plh_inode; if (atomic_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) { - destroy_layout_hdr(lo); + pnfs_detach_layout_hdr(lo); spin_unlock(&inode->i_lock); + pnfs_free_layout_hdr(lo); } } +static int +pnfs_iomode_to_fail_bit(u32 iomode) +{ + return iomode == IOMODE_RW ? + NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED; +} + +static void +pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit) +{ + lo->plh_retry_timestamp = jiffies; + if (!test_and_set_bit(fail_bit, &lo->plh_flags)) + atomic_inc(&lo->plh_refcount); +} + +static void +pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit) +{ + if (test_and_clear_bit(fail_bit, &lo->plh_flags)) + atomic_dec(&lo->plh_refcount); +} + +static void +pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode) +{ + struct inode *inode = lo->plh_inode; + struct pnfs_layout_range range = { + .iomode = iomode, + .offset = 0, + .length = NFS4_MAX_UINT64, + }; + LIST_HEAD(head); + + spin_lock(&inode->i_lock); + pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode)); + pnfs_mark_matching_lsegs_invalid(lo, &head, &range); + spin_unlock(&inode->i_lock); + pnfs_free_lseg_list(&head); + dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__, + iomode == IOMODE_RW ? "RW" : "READ"); +} + +static bool +pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode) +{ + unsigned long start, end; + int fail_bit = pnfs_iomode_to_fail_bit(iomode); + + if (test_bit(fail_bit, &lo->plh_flags) == 0) + return false; + end = jiffies; + start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT; + if (!time_in_range(lo->plh_retry_timestamp, start, end)) { + /* It is time to retry the failed layoutgets */ + pnfs_layout_clear_fail_bit(lo, fail_bit); + return false; + } + return true; +} + static void init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg) { INIT_LIST_HEAD(&lseg->pls_list); + INIT_LIST_HEAD(&lseg->pls_lc_list); atomic_set(&lseg->pls_refcount, 1); smp_mb(); set_bit(NFS_LSEG_VALID, &lseg->pls_flags); lseg->pls_layout = lo; } -static void free_lseg(struct pnfs_layout_segment *lseg) +static void pnfs_free_lseg(struct pnfs_layout_segment *lseg) { struct inode *ino = lseg->pls_layout->plh_inode; NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg); - /* Matched by get_layout_hdr in pnfs_insert_layout */ - put_layout_hdr(NFS_I(ino)->layout); } -/* The use of tmp_list is necessary because pnfs_curr_ld->free_lseg - * could sleep, so must be called outside of the lock. - * Returns 1 if object was removed, otherwise return 0. - */ -static int -put_lseg_locked(struct pnfs_layout_segment *lseg, - struct list_head *tmp_list) +static void +pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo, + struct pnfs_layout_segment *lseg) +{ + struct inode *inode = lo->plh_inode; + + WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); + list_del_init(&lseg->pls_list); + /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */ + atomic_dec(&lo->plh_refcount); + if (list_empty(&lo->plh_segs)) + clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags); + rpc_wake_up(&NFS_SERVER(inode)->roc_rpcwaitq); +} + +void +pnfs_put_lseg(struct pnfs_layout_segment *lseg) { + struct pnfs_layout_hdr *lo; + struct inode *inode; + + if (!lseg) + return; + dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg, atomic_read(&lseg->pls_refcount), test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); - if (atomic_dec_and_test(&lseg->pls_refcount)) { - struct inode *ino = lseg->pls_layout->plh_inode; - - BUG_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags)); - list_del(&lseg->pls_list); - if (list_empty(&lseg->pls_layout->plh_segs)) { - struct nfs_client *clp; - - clp = NFS_SERVER(ino)->nfs_client; - spin_lock(&clp->cl_lock); - /* List does not take a reference, so no need for put here */ - list_del_init(&lseg->pls_layout->plh_layouts); - spin_unlock(&clp->cl_lock); - clear_bit(NFS_LAYOUT_BULK_RECALL, &lseg->pls_layout->plh_flags); - } - rpc_wake_up(&NFS_SERVER(ino)->roc_rpcwaitq); - list_add(&lseg->pls_list, tmp_list); - return 1; + lo = lseg->pls_layout; + inode = lo->plh_inode; + if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) { + pnfs_get_layout_hdr(lo); + pnfs_layout_remove_lseg(lo, lseg); + spin_unlock(&inode->i_lock); + pnfs_free_lseg(lseg); + pnfs_put_layout_hdr(lo); } - return 0; +} +EXPORT_SYMBOL_GPL(pnfs_put_lseg); + +static u64 +end_offset(u64 start, u64 len) +{ + u64 end; + + end = start + len; + return end >= start ? end : NFS4_MAX_UINT64; } +/* + * is l2 fully contained in l1? + * start1 end1 + * [----------------------------------) + * start2 end2 + * [----------------) + */ static bool -should_free_lseg(u32 lseg_iomode, u32 recall_iomode) +pnfs_lseg_range_contained(const struct pnfs_layout_range *l1, + const struct pnfs_layout_range *l2) { - return (recall_iomode == IOMODE_ANY || - lseg_iomode == recall_iomode); + u64 start1 = l1->offset; + u64 end1 = end_offset(start1, l1->length); + u64 start2 = l2->offset; + u64 end2 = end_offset(start2, l2->length); + + return (start1 <= start2) && (end1 >= end2); +} + +/* + * is l1 and l2 intersecting? + * start1 end1 + * [----------------------------------) + * start2 end2 + * [----------------) + */ +static bool +pnfs_lseg_range_intersecting(const struct pnfs_layout_range *l1, + const struct pnfs_layout_range *l2) +{ + u64 start1 = l1->offset; + u64 end1 = end_offset(start1, l1->length); + u64 start2 = l2->offset; + u64 end2 = end_offset(start2, l2->length); + + return (end1 == NFS4_MAX_UINT64 || end1 > start2) && + (end2 == NFS4_MAX_UINT64 || end2 > start1); +} + +static bool +should_free_lseg(const struct pnfs_layout_range *lseg_range, + const struct pnfs_layout_range *recall_range) +{ + return (recall_range->iomode == IOMODE_ANY || + lseg_range->iomode == recall_range->iomode) && + pnfs_lseg_range_intersecting(lseg_range, recall_range); +} + +static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg, + struct list_head *tmp_list) +{ + if (!atomic_dec_and_test(&lseg->pls_refcount)) + return false; + pnfs_layout_remove_lseg(lseg->pls_layout, lseg); + list_add(&lseg->pls_list, tmp_list); + return true; } /* Returns 1 if lseg is removed from list, 0 otherwise */ @@ -281,7 +439,10 @@ static int mark_lseg_invalid(struct pnfs_layout_segment *lseg, * list. It will now be removed when all * outstanding io is finished. */ - rv = put_lseg_locked(lseg, tmp_list); + dprintk("%s: lseg %p ref %d\n", __func__, lseg, + atomic_read(&lseg->pls_refcount)); + if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list)) + rv = 1; } return rv; } @@ -290,17 +451,20 @@ static int mark_lseg_invalid(struct pnfs_layout_segment *lseg, * after call. */ int -mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo, +pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo, struct list_head *tmp_list, - u32 iomode) + struct pnfs_layout_range *recall_range) { struct pnfs_layout_segment *lseg, *next; int invalid = 0, removed = 0; dprintk("%s:Begin lo %p\n", __func__, lo); + if (list_empty(&lo->plh_segs)) + return 0; list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) - if (should_free_lseg(lseg->pls_range.iomode, iomode)) { + if (!recall_range || + should_free_lseg(&lseg->pls_range, recall_range)) { dprintk("%s: freeing lseg %p iomode %d " "offset %llu length %llu\n", __func__, lseg, lseg->pls_range.iomode, lseg->pls_range.offset, @@ -312,14 +476,18 @@ mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo, return invalid - removed; } +/* note free_me must contain lsegs from a single layout_hdr */ void pnfs_free_lseg_list(struct list_head *free_me) { struct pnfs_layout_segment *lseg, *tmp; + if (list_empty(free_me)) + return; + list_for_each_entry_safe(lseg, tmp, free_me, pls_list) { list_del(&lseg->pls_list); - free_lseg(lseg); + pnfs_free_lseg(lseg); } } @@ -332,13 +500,147 @@ pnfs_destroy_layout(struct nfs_inode *nfsi) spin_lock(&nfsi->vfs_inode.i_lock); lo = nfsi->layout; if (lo) { - set_bit(NFS_LAYOUT_DESTROYED, &nfsi->layout->plh_flags); - mark_matching_lsegs_invalid(lo, &tmp_list, IOMODE_ANY); - /* Matched by refcount set to 1 in alloc_init_layout_hdr */ - put_layout_hdr_locked(lo); + lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */ + pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL); + pnfs_get_layout_hdr(lo); + pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED); + pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED); + spin_unlock(&nfsi->vfs_inode.i_lock); + pnfs_free_lseg_list(&tmp_list); + pnfs_put_layout_hdr(lo); + } else + spin_unlock(&nfsi->vfs_inode.i_lock); +} +EXPORT_SYMBOL_GPL(pnfs_destroy_layout); + +static bool +pnfs_layout_add_bulk_destroy_list(struct inode *inode, + struct list_head *layout_list) +{ + struct pnfs_layout_hdr *lo; + bool ret = false; + + spin_lock(&inode->i_lock); + lo = NFS_I(inode)->layout; + if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) { + pnfs_get_layout_hdr(lo); + list_add(&lo->plh_bulk_destroy, layout_list); + ret = true; } - spin_unlock(&nfsi->vfs_inode.i_lock); - pnfs_free_lseg_list(&tmp_list); + spin_unlock(&inode->i_lock); + return ret; +} + +/* Caller must hold rcu_read_lock and clp->cl_lock */ +static int +pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp, + struct nfs_server *server, + struct list_head *layout_list) +{ + struct pnfs_layout_hdr *lo, *next; + struct inode *inode; + + list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) { + inode = igrab(lo->plh_inode); + if (inode == NULL) + continue; + list_del_init(&lo->plh_layouts); + if (pnfs_layout_add_bulk_destroy_list(inode, layout_list)) + continue; + rcu_read_unlock(); + spin_unlock(&clp->cl_lock); + iput(inode); + spin_lock(&clp->cl_lock); + rcu_read_lock(); + return -EAGAIN; + } + return 0; +} + +static int +pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list, + bool is_bulk_recall) +{ + struct pnfs_layout_hdr *lo; + struct inode *inode; + struct pnfs_layout_range range = { + .iomode = IOMODE_ANY, + .offset = 0, + .length = NFS4_MAX_UINT64, + }; + LIST_HEAD(lseg_list); + int ret = 0; + + while (!list_empty(layout_list)) { + lo = list_entry(layout_list->next, struct pnfs_layout_hdr, + plh_bulk_destroy); + dprintk("%s freeing layout for inode %lu\n", __func__, + lo->plh_inode->i_ino); + inode = lo->plh_inode; + spin_lock(&inode->i_lock); + list_del_init(&lo->plh_bulk_destroy); + lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */ + if (is_bulk_recall) + set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags); + if (pnfs_mark_matching_lsegs_invalid(lo, &lseg_list, &range)) + ret = -EAGAIN; + spin_unlock(&inode->i_lock); + pnfs_free_lseg_list(&lseg_list); + pnfs_put_layout_hdr(lo); + iput(inode); + } + return ret; +} + +int +pnfs_destroy_layouts_byfsid(struct nfs_client *clp, + struct nfs_fsid *fsid, + bool is_recall) +{ + struct nfs_server *server; + LIST_HEAD(layout_list); + + spin_lock(&clp->cl_lock); + rcu_read_lock(); +restart: + list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { + if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0) + continue; + if (pnfs_layout_bulk_destroy_byserver_locked(clp, + server, + &layout_list) != 0) + goto restart; + } + rcu_read_unlock(); + spin_unlock(&clp->cl_lock); + + if (list_empty(&layout_list)) + return 0; + return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall); +} + +int +pnfs_destroy_layouts_byclid(struct nfs_client *clp, + bool is_recall) +{ + struct nfs_server *server; + LIST_HEAD(layout_list); + + spin_lock(&clp->cl_lock); + rcu_read_lock(); +restart: + list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { + if (pnfs_layout_bulk_destroy_byserver_locked(clp, + server, + &layout_list) != 0) + goto restart; + } + rcu_read_unlock(); + spin_unlock(&clp->cl_lock); + + if (list_empty(&layout_list)) + return 0; + return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall); } /* @@ -348,20 +650,30 @@ pnfs_destroy_layout(struct nfs_inode *nfsi) void pnfs_destroy_all_layouts(struct nfs_client *clp) { - struct pnfs_layout_hdr *lo; - LIST_HEAD(tmp_list); + nfs4_deviceid_mark_client_invalid(clp); + nfs4_deviceid_purge_client(clp); - spin_lock(&clp->cl_lock); - list_splice_init(&clp->cl_layouts, &tmp_list); - spin_unlock(&clp->cl_lock); + pnfs_destroy_layouts_byclid(clp, false); +} - while (!list_empty(&tmp_list)) { - lo = list_entry(tmp_list.next, struct pnfs_layout_hdr, - plh_layouts); - dprintk("%s freeing layout for inode %lu\n", __func__, - lo->plh_inode->i_ino); - pnfs_destroy_layout(NFS_I(lo->plh_inode)); - } +/* + * Compare 2 layout stateid sequence ids, to see which is newer, + * taking into account wraparound issues. + */ +static bool pnfs_seqid_is_newer(u32 s1, u32 s2) +{ + return (s32)(s1 - s2) > 0; +} + +static void +pnfs_verify_layout_stateid(struct pnfs_layout_hdr *lo, + const nfs4_stateid *new, + struct list_head *free_me_list) +{ + if (nfs4_stateid_match_other(&lo->plh_stateid, new)) + return; + /* Layout is new! Kill existing layout segments */ + pnfs_mark_matching_lsegs_invalid(lo, free_me_list, NULL); } /* update lo->plh_stateid with new if is more recent */ @@ -369,39 +681,39 @@ void pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new, bool update_barrier) { - u32 oldseq, newseq; + u32 oldseq, newseq, new_barrier; + int empty = list_empty(&lo->plh_segs); - oldseq = be32_to_cpu(lo->plh_stateid.stateid.seqid); - newseq = be32_to_cpu(new->stateid.seqid); - if ((int)(newseq - oldseq) > 0) { - memcpy(&lo->plh_stateid, &new->stateid, sizeof(new->stateid)); + oldseq = be32_to_cpu(lo->plh_stateid.seqid); + newseq = be32_to_cpu(new->seqid); + if (empty || pnfs_seqid_is_newer(newseq, oldseq)) { + nfs4_stateid_copy(&lo->plh_stateid, new); if (update_barrier) { - u32 new_barrier = be32_to_cpu(new->stateid.seqid); - - if ((int)(new_barrier - lo->plh_barrier)) - lo->plh_barrier = new_barrier; + new_barrier = be32_to_cpu(new->seqid); } else { /* Because of wraparound, we want to keep the barrier - * "close" to the current seqids. It needs to be - * within 2**31 to count as "behind", so if it - * gets too near that limit, give us a litle leeway - * and bring it to within 2**30. - * NOTE - and yes, this is all unsigned arithmetic. + * "close" to the current seqids. */ - if (unlikely((newseq - lo->plh_barrier) > (3 << 29))) - lo->plh_barrier = newseq - (1 << 30); + new_barrier = newseq - atomic_read(&lo->plh_outstanding); } + if (empty || pnfs_seqid_is_newer(new_barrier, lo->plh_barrier)) + lo->plh_barrier = new_barrier; } } +static bool +pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo, + const nfs4_stateid *stateid) +{ + u32 seqid = be32_to_cpu(stateid->seqid); + + return !pnfs_seqid_is_newer(seqid, lo->plh_barrier); +} + /* lget is set to 1 if called from inside send_layoutget call chain */ static bool -pnfs_layoutgets_blocked(struct pnfs_layout_hdr *lo, nfs4_stateid *stateid, - int lget) +pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo, int lget) { - if ((stateid) && - (int)(lo->plh_barrier - be32_to_cpu(stateid->stateid.seqid)) >= 0) - return true; return lo->plh_block_lgets || test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) || (list_empty(&lo->plh_segs) && @@ -416,18 +728,19 @@ pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo, dprintk("--> %s\n", __func__); spin_lock(&lo->plh_inode->i_lock); - if (pnfs_layoutgets_blocked(lo, NULL, 1)) { + if (pnfs_layoutgets_blocked(lo, 1)) { status = -EAGAIN; + } else if (!nfs4_valid_open_stateid(open_state)) { + status = -EBADF; } else if (list_empty(&lo->plh_segs)) { int seq; do { seq = read_seqbegin(&open_state->seqlock); - memcpy(dst->data, open_state->stateid.data, - sizeof(open_state->stateid.data)); + nfs4_stateid_copy(dst, &open_state->stateid); } while (read_seqretry(&open_state->seqlock, seq)); } else - memcpy(dst->data, lo->plh_stateid.data, sizeof(lo->plh_stateid.data)); + nfs4_stateid_copy(dst, &lo->plh_stateid); spin_unlock(&lo->plh_inode->i_lock); dprintk("<-- %s\n", __func__); return status; @@ -442,40 +755,160 @@ pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo, static struct pnfs_layout_segment * send_layoutget(struct pnfs_layout_hdr *lo, struct nfs_open_context *ctx, - u32 iomode) + struct pnfs_layout_range *range, + gfp_t gfp_flags) { struct inode *ino = lo->plh_inode; struct nfs_server *server = NFS_SERVER(ino); struct nfs4_layoutget *lgp; - struct pnfs_layout_segment *lseg = NULL; + struct pnfs_layout_segment *lseg; dprintk("--> %s\n", __func__); - BUG_ON(ctx == NULL); - lgp = kzalloc(sizeof(*lgp), GFP_KERNEL); + lgp = kzalloc(sizeof(*lgp), gfp_flags); if (lgp == NULL) return NULL; - lgp->args.minlength = NFS4_MAX_UINT64; + + lgp->args.minlength = PAGE_CACHE_SIZE; + if (lgp->args.minlength > range->length) + lgp->args.minlength = range->length; lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE; - lgp->args.range.iomode = iomode; - lgp->args.range.offset = 0; - lgp->args.range.length = NFS4_MAX_UINT64; + lgp->args.range = *range; lgp->args.type = server->pnfs_curr_ld->id; lgp->args.inode = ino; lgp->args.ctx = get_nfs_open_context(ctx); - lgp->lsegpp = &lseg; + lgp->gfp_flags = gfp_flags; + lgp->cred = lo->plh_lc_cred; /* Synchronously retrieve layout information from server and * store in lseg. */ - nfs4_proc_layoutget(lgp); - if (!lseg) { - /* remember that LAYOUTGET failed and suspend trying */ - set_bit(lo_fail_bit(iomode), &lo->plh_flags); + lseg = nfs4_proc_layoutget(lgp, gfp_flags); + if (IS_ERR(lseg)) { + switch (PTR_ERR(lseg)) { + case -ENOMEM: + case -ERESTARTSYS: + break; + default: + /* remember that LAYOUTGET failed and suspend trying */ + pnfs_layout_io_set_failed(lo, range->iomode); + } + return NULL; } + return lseg; } +static void pnfs_clear_layoutcommit(struct inode *inode, + struct list_head *head) +{ + struct nfs_inode *nfsi = NFS_I(inode); + struct pnfs_layout_segment *lseg, *tmp; + + if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) + return; + list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) { + if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) + continue; + pnfs_lseg_dec_and_remove_zero(lseg, head); + } +} + +/* + * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr + * when the layout segment list is empty. + * + * Note that a pnfs_layout_hdr can exist with an empty layout segment + * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the + * deviceid is marked invalid. + */ +int +_pnfs_return_layout(struct inode *ino) +{ + struct pnfs_layout_hdr *lo = NULL; + struct nfs_inode *nfsi = NFS_I(ino); + LIST_HEAD(tmp_list); + struct nfs4_layoutreturn *lrp; + nfs4_stateid stateid; + int status = 0, empty; + + dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino); + + spin_lock(&ino->i_lock); + lo = nfsi->layout; + if (!lo) { + spin_unlock(&ino->i_lock); + dprintk("NFS: %s no layout to return\n", __func__); + goto out; + } + stateid = nfsi->layout->plh_stateid; + /* Reference matched in nfs4_layoutreturn_release */ + pnfs_get_layout_hdr(lo); + empty = list_empty(&lo->plh_segs); + pnfs_clear_layoutcommit(ino, &tmp_list); + pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL); + /* Don't send a LAYOUTRETURN if list was initially empty */ + if (empty) { + spin_unlock(&ino->i_lock); + pnfs_put_layout_hdr(lo); + dprintk("NFS: %s no layout segments to return\n", __func__); + goto out; + } + lo->plh_block_lgets++; + spin_unlock(&ino->i_lock); + pnfs_free_lseg_list(&tmp_list); + + lrp = kzalloc(sizeof(*lrp), GFP_KERNEL); + if (unlikely(lrp == NULL)) { + status = -ENOMEM; + spin_lock(&ino->i_lock); + lo->plh_block_lgets--; + spin_unlock(&ino->i_lock); + pnfs_put_layout_hdr(lo); + goto out; + } + + lrp->args.stateid = stateid; + lrp->args.layout_type = NFS_SERVER(ino)->pnfs_curr_ld->id; + lrp->args.inode = ino; + lrp->args.layout = lo; + lrp->clp = NFS_SERVER(ino)->nfs_client; + lrp->cred = lo->plh_lc_cred; + + status = nfs4_proc_layoutreturn(lrp); +out: + dprintk("<-- %s status: %d\n", __func__, status); + return status; +} +EXPORT_SYMBOL_GPL(_pnfs_return_layout); + +int +pnfs_commit_and_return_layout(struct inode *inode) +{ + struct pnfs_layout_hdr *lo; + int ret; + + spin_lock(&inode->i_lock); + lo = NFS_I(inode)->layout; + if (lo == NULL) { + spin_unlock(&inode->i_lock); + return 0; + } + pnfs_get_layout_hdr(lo); + /* Block new layoutgets and read/write to ds */ + lo->plh_block_lgets++; + spin_unlock(&inode->i_lock); + filemap_fdatawait(inode->i_mapping); + ret = pnfs_layoutcommit_inode(inode, true); + if (ret == 0) + ret = _pnfs_return_layout(inode); + spin_lock(&inode->i_lock); + lo->plh_block_lgets--; + spin_unlock(&inode->i_lock); + pnfs_put_layout_hdr(lo); + return ret; +} + bool pnfs_roc(struct inode *ino) { struct pnfs_layout_hdr *lo; @@ -496,7 +929,7 @@ bool pnfs_roc(struct inode *ino) if (!found) goto out_nolayout; lo->plh_block_lgets++; - get_layout_hdr(lo); /* matched in pnfs_roc_release */ + pnfs_get_layout_hdr(lo); /* matched in pnfs_roc_release */ spin_unlock(&ino->i_lock); pnfs_free_lseg_list(&tmp_list); return true; @@ -513,8 +946,12 @@ void pnfs_roc_release(struct inode *ino) spin_lock(&ino->i_lock); lo = NFS_I(ino)->layout; lo->plh_block_lgets--; - put_layout_hdr_locked(lo); - spin_unlock(&ino->i_lock); + if (atomic_dec_and_test(&lo->plh_refcount)) { + pnfs_detach_layout_hdr(lo); + spin_unlock(&ino->i_lock); + pnfs_free_layout_hdr(lo); + } else + spin_unlock(&ino->i_lock); } void pnfs_roc_set_barrier(struct inode *ino, u32 barrier) @@ -523,32 +960,34 @@ void pnfs_roc_set_barrier(struct inode *ino, u32 barrier) spin_lock(&ino->i_lock); lo = NFS_I(ino)->layout; - if ((int)(barrier - lo->plh_barrier) > 0) + if (pnfs_seqid_is_newer(barrier, lo->plh_barrier)) lo->plh_barrier = barrier; spin_unlock(&ino->i_lock); } -bool pnfs_roc_drain(struct inode *ino, u32 *barrier) +bool pnfs_roc_drain(struct inode *ino, u32 *barrier, struct rpc_task *task) { struct nfs_inode *nfsi = NFS_I(ino); + struct pnfs_layout_hdr *lo; struct pnfs_layout_segment *lseg; + u32 current_seqid; bool found = false; spin_lock(&ino->i_lock); list_for_each_entry(lseg, &nfsi->layout->plh_segs, pls_list) if (test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) { + rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL); found = true; - break; + goto out; } - if (!found) { - struct pnfs_layout_hdr *lo = nfsi->layout; - u32 current_seqid = be32_to_cpu(lo->plh_stateid.stateid.seqid); + lo = nfsi->layout; + current_seqid = be32_to_cpu(lo->plh_stateid.seqid); - /* Since close does not return a layout stateid for use as - * a barrier, we choose the worst-case barrier. - */ - *barrier = current_seqid + atomic_read(&lo->plh_outstanding); - } + /* Since close does not return a layout stateid for use as + * a barrier, we choose the worst-case barrier. + */ + *barrier = current_seqid + atomic_read(&lo->plh_outstanding); +out: spin_unlock(&ino->i_lock); return found; } @@ -559,24 +998,35 @@ bool pnfs_roc_drain(struct inode *ino, u32 *barrier) * are seen first. */ static s64 -cmp_layout(u32 iomode1, u32 iomode2) +pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1, + const struct pnfs_layout_range *l2) { + s64 d; + + /* high offset > low offset */ + d = l1->offset - l2->offset; + if (d) + return d; + + /* short length > long length */ + d = l2->length - l1->length; + if (d) + return d; + /* read > read/write */ - return (int)(iomode2 == IOMODE_READ) - (int)(iomode1 == IOMODE_READ); + return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ); } static void -pnfs_insert_layout(struct pnfs_layout_hdr *lo, +pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg) { struct pnfs_layout_segment *lp; - int found = 0; dprintk("%s:Begin\n", __func__); - assert_spin_locked(&lo->plh_inode->i_lock); list_for_each_entry(lp, &lo->plh_segs, pls_list) { - if (cmp_layout(lp->pls_range.iomode, lseg->pls_range.iomode) > 0) + if (pnfs_lseg_range_cmp(&lseg->pls_range, &lp->pls_range) > 0) continue; list_add_tail(&lseg->pls_list, &lp->pls_list); dprintk("%s: inserted lseg %p " @@ -586,60 +1036,61 @@ pnfs_insert_layout(struct pnfs_layout_hdr *lo, lseg->pls_range.offset, lseg->pls_range.length, lp, lp->pls_range.iomode, lp->pls_range.offset, lp->pls_range.length); - found = 1; - break; - } - if (!found) { - list_add_tail(&lseg->pls_list, &lo->plh_segs); - dprintk("%s: inserted lseg %p " - "iomode %d offset %llu length %llu at tail\n", - __func__, lseg, lseg->pls_range.iomode, - lseg->pls_range.offset, lseg->pls_range.length); + goto out; } - get_layout_hdr(lo); + list_add_tail(&lseg->pls_list, &lo->plh_segs); + dprintk("%s: inserted lseg %p " + "iomode %d offset %llu length %llu at tail\n", + __func__, lseg, lseg->pls_range.iomode, + lseg->pls_range.offset, lseg->pls_range.length); +out: + pnfs_get_layout_hdr(lo); dprintk("%s:Return\n", __func__); } static struct pnfs_layout_hdr * -alloc_init_layout_hdr(struct inode *ino) +alloc_init_layout_hdr(struct inode *ino, + struct nfs_open_context *ctx, + gfp_t gfp_flags) { struct pnfs_layout_hdr *lo; - lo = kzalloc(sizeof(struct pnfs_layout_hdr), GFP_KERNEL); + lo = pnfs_alloc_layout_hdr(ino, gfp_flags); if (!lo) return NULL; atomic_set(&lo->plh_refcount, 1); INIT_LIST_HEAD(&lo->plh_layouts); INIT_LIST_HEAD(&lo->plh_segs); - INIT_LIST_HEAD(&lo->plh_bulk_recall); + INIT_LIST_HEAD(&lo->plh_bulk_destroy); lo->plh_inode = ino; + lo->plh_lc_cred = get_rpccred(ctx->cred); return lo; } static struct pnfs_layout_hdr * -pnfs_find_alloc_layout(struct inode *ino) +pnfs_find_alloc_layout(struct inode *ino, + struct nfs_open_context *ctx, + gfp_t gfp_flags) { struct nfs_inode *nfsi = NFS_I(ino); struct pnfs_layout_hdr *new = NULL; dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout); - assert_spin_locked(&ino->i_lock); - if (nfsi->layout) { - if (test_bit(NFS_LAYOUT_DESTROYED, &nfsi->layout->plh_flags)) - return NULL; - else - return nfsi->layout; - } + if (nfsi->layout != NULL) + goto out_existing; spin_unlock(&ino->i_lock); - new = alloc_init_layout_hdr(ino); + new = alloc_init_layout_hdr(ino, ctx, gfp_flags); spin_lock(&ino->i_lock); - if (likely(nfsi->layout == NULL)) /* Won the race? */ + if (likely(nfsi->layout == NULL)) { /* Won the race? */ nfsi->layout = new; - else - kfree(new); + return new; + } else if (new != NULL) + pnfs_free_layout_hdr(new); +out_existing: + pnfs_get_layout_hdr(nfsi->layout); return nfsi->layout; } @@ -654,30 +1105,41 @@ pnfs_find_alloc_layout(struct inode *ino) * READ READ true * READ RW true */ -static int -is_matching_lseg(struct pnfs_layout_segment *lseg, u32 iomode) +static bool +pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range, + const struct pnfs_layout_range *range) { - return (iomode != IOMODE_RW || lseg->pls_range.iomode == IOMODE_RW); + struct pnfs_layout_range range1; + + if ((range->iomode == IOMODE_RW && + ls_range->iomode != IOMODE_RW) || + !pnfs_lseg_range_intersecting(ls_range, range)) + return 0; + + /* range1 covers only the first byte in the range */ + range1 = *range; + range1.length = 1; + return pnfs_lseg_range_contained(ls_range, &range1); } /* * lookup range in layout */ static struct pnfs_layout_segment * -pnfs_find_lseg(struct pnfs_layout_hdr *lo, u32 iomode) +pnfs_find_lseg(struct pnfs_layout_hdr *lo, + struct pnfs_layout_range *range) { struct pnfs_layout_segment *lseg, *ret = NULL; dprintk("%s:Begin\n", __func__); - assert_spin_locked(&lo->plh_inode->i_lock); list_for_each_entry(lseg, &lo->plh_segs, pls_list) { if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) && - is_matching_lseg(lseg, iomode)) { - ret = lseg; + pnfs_lseg_range_match(&lseg->pls_range, range)) { + ret = pnfs_get_lseg(lseg); break; } - if (cmp_layout(iomode, lseg->pls_range.iomode) > 0) + if (lseg->pls_range.offset > range->offset) break; } @@ -687,104 +1149,188 @@ pnfs_find_lseg(struct pnfs_layout_hdr *lo, u32 iomode) } /* + * Use mdsthreshold hints set at each OPEN to determine if I/O should go + * to the MDS or over pNFS + * + * The nfs_inode read_io and write_io fields are cumulative counters reset + * when there are no layout segments. Note that in pnfs_update_layout iomode + * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a + * WRITE request. + * + * A return of true means use MDS I/O. + * + * From rfc 5661: + * If a file's size is smaller than the file size threshold, data accesses + * SHOULD be sent to the metadata server. If an I/O request has a length that + * is below the I/O size threshold, the I/O SHOULD be sent to the metadata + * server. If both file size and I/O size are provided, the client SHOULD + * reach or exceed both thresholds before sending its read or write + * requests to the data server. + */ +static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx, + struct inode *ino, int iomode) +{ + struct nfs4_threshold *t = ctx->mdsthreshold; + struct nfs_inode *nfsi = NFS_I(ino); + loff_t fsize = i_size_read(ino); + bool size = false, size_set = false, io = false, io_set = false, ret = false; + + if (t == NULL) + return ret; + + dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n", + __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz); + + switch (iomode) { + case IOMODE_READ: + if (t->bm & THRESHOLD_RD) { + dprintk("%s fsize %llu\n", __func__, fsize); + size_set = true; + if (fsize < t->rd_sz) + size = true; + } + if (t->bm & THRESHOLD_RD_IO) { + dprintk("%s nfsi->read_io %llu\n", __func__, + nfsi->read_io); + io_set = true; + if (nfsi->read_io < t->rd_io_sz) + io = true; + } + break; + case IOMODE_RW: + if (t->bm & THRESHOLD_WR) { + dprintk("%s fsize %llu\n", __func__, fsize); + size_set = true; + if (fsize < t->wr_sz) + size = true; + } + if (t->bm & THRESHOLD_WR_IO) { + dprintk("%s nfsi->write_io %llu\n", __func__, + nfsi->write_io); + io_set = true; + if (nfsi->write_io < t->wr_io_sz) + io = true; + } + break; + } + if (size_set && io_set) { + if (size && io) + ret = true; + } else if (size || io) + ret = true; + + dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret); + return ret; +} + +/* * Layout segment is retreived from the server if not cached. * The appropriate layout segment is referenced and returned to the caller. */ struct pnfs_layout_segment * pnfs_update_layout(struct inode *ino, struct nfs_open_context *ctx, - enum pnfs_iomode iomode) + loff_t pos, + u64 count, + enum pnfs_iomode iomode, + gfp_t gfp_flags) { - struct nfs_inode *nfsi = NFS_I(ino); - struct nfs_client *clp = NFS_SERVER(ino)->nfs_client; + struct pnfs_layout_range arg = { + .iomode = iomode, + .offset = pos, + .length = count, + }; + unsigned pg_offset; + struct nfs_server *server = NFS_SERVER(ino); + struct nfs_client *clp = server->nfs_client; struct pnfs_layout_hdr *lo; struct pnfs_layout_segment *lseg = NULL; + bool first; if (!pnfs_enabled_sb(NFS_SERVER(ino))) - return NULL; + goto out; + + if (pnfs_within_mdsthreshold(ctx, ino, iomode)) + goto out; + spin_lock(&ino->i_lock); - lo = pnfs_find_alloc_layout(ino); + lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags); if (lo == NULL) { - dprintk("%s ERROR: can't get pnfs_layout_hdr\n", __func__); - goto out_unlock; + spin_unlock(&ino->i_lock); + goto out; } /* Do we even need to bother with this? */ - if (test_bit(NFS4CLNT_LAYOUTRECALL, &clp->cl_state) || - test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { + if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { dprintk("%s matches recall, use MDS\n", __func__); goto out_unlock; } - /* Check to see if the layout for the given range already exists */ - lseg = pnfs_find_lseg(lo, iomode); - if (lseg) - goto out_unlock; /* if LAYOUTGET already failed once we don't try again */ - if (test_bit(lo_fail_bit(iomode), &nfsi->layout->plh_flags)) + if (pnfs_layout_io_test_failed(lo, iomode)) + goto out_unlock; + + /* Check to see if the layout for the given range already exists */ + lseg = pnfs_find_lseg(lo, &arg); + if (lseg) goto out_unlock; - if (pnfs_layoutgets_blocked(lo, NULL, 0)) + if (pnfs_layoutgets_blocked(lo, 0)) goto out_unlock; atomic_inc(&lo->plh_outstanding); - get_layout_hdr(lo); - if (list_empty(&lo->plh_segs)) { + first = list_empty(&lo->plh_layouts) ? true : false; + spin_unlock(&ino->i_lock); + + if (first) { /* The lo must be on the clp list if there is any * chance of a CB_LAYOUTRECALL(FILE) coming in. */ spin_lock(&clp->cl_lock); - BUG_ON(!list_empty(&lo->plh_layouts)); - list_add_tail(&lo->plh_layouts, &clp->cl_layouts); + list_add_tail(&lo->plh_layouts, &server->layouts); spin_unlock(&clp->cl_lock); } - spin_unlock(&ino->i_lock); - lseg = send_layoutget(lo, ctx, iomode); - if (!lseg) { - spin_lock(&ino->i_lock); - if (list_empty(&lo->plh_segs)) { - spin_lock(&clp->cl_lock); - list_del_init(&lo->plh_layouts); - spin_unlock(&clp->cl_lock); - clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags); - } - spin_unlock(&ino->i_lock); + pg_offset = arg.offset & ~PAGE_CACHE_MASK; + if (pg_offset) { + arg.offset -= pg_offset; + arg.length += pg_offset; } + if (arg.length != NFS4_MAX_UINT64) + arg.length = PAGE_CACHE_ALIGN(arg.length); + + lseg = send_layoutget(lo, ctx, &arg, gfp_flags); atomic_dec(&lo->plh_outstanding); - put_layout_hdr(lo); +out_put_layout_hdr: + pnfs_put_layout_hdr(lo); out: - dprintk("%s end, state 0x%lx lseg %p\n", __func__, - nfsi->layout->plh_flags, lseg); + dprintk("%s: inode %s/%llu pNFS layout segment %s for " + "(%s, offset: %llu, length: %llu)\n", + __func__, ino->i_sb->s_id, + (unsigned long long)NFS_FILEID(ino), + lseg == NULL ? "not found" : "found", + iomode==IOMODE_RW ? "read/write" : "read-only", + (unsigned long long)pos, + (unsigned long long)count); return lseg; out_unlock: spin_unlock(&ino->i_lock); - goto out; + goto out_put_layout_hdr; } +EXPORT_SYMBOL_GPL(pnfs_update_layout); -int +struct pnfs_layout_segment * pnfs_layout_process(struct nfs4_layoutget *lgp) { struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout; struct nfs4_layoutget_res *res = &lgp->res; struct pnfs_layout_segment *lseg; struct inode *ino = lo->plh_inode; - struct nfs_client *clp = NFS_SERVER(ino)->nfs_client; + LIST_HEAD(free_me); int status = 0; - /* Verify we got what we asked for. - * Note that because the xdr parsing only accepts a single - * element array, this can fail even if the server is behaving - * correctly. - */ - if (lgp->args.range.iomode > res->range.iomode || - res->range.offset != 0 || - res->range.length != NFS4_MAX_UINT64) { - status = -EINVAL; - goto out; - } /* Inject layout blob into I/O device driver */ - lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res); + lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags); if (!lseg || IS_ERR(lseg)) { if (!lseg) status = -ENOMEM; @@ -796,31 +1342,37 @@ pnfs_layout_process(struct nfs4_layoutget *lgp) } spin_lock(&ino->i_lock); - if (test_bit(NFS4CLNT_LAYOUTRECALL, &clp->cl_state) || - test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { + if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) { dprintk("%s forget reply due to recall\n", __func__); goto out_forget_reply; } - if (pnfs_layoutgets_blocked(lo, &res->stateid, 1)) { + if (pnfs_layoutgets_blocked(lo, 1) || + pnfs_layout_stateid_blocked(lo, &res->stateid)) { dprintk("%s forget reply due to state\n", __func__); goto out_forget_reply; } + + /* Check that the new stateid matches the old stateid */ + pnfs_verify_layout_stateid(lo, &res->stateid, &free_me); + /* Done processing layoutget. Set the layout stateid */ + pnfs_set_layout_stateid(lo, &res->stateid, false); + init_lseg(lo, lseg); lseg->pls_range = res->range; - *lgp->lsegpp = lseg; - pnfs_insert_layout(lo, lseg); + pnfs_get_lseg(lseg); + pnfs_layout_insert_lseg(lo, lseg); if (res->return_on_close) { set_bit(NFS_LSEG_ROC, &lseg->pls_flags); set_bit(NFS_LAYOUT_ROC, &lo->plh_flags); } - /* Done processing layoutget. Set the layout stateid */ - pnfs_set_layout_stateid(lo, &res->stateid, false); spin_unlock(&ino->i_lock); + pnfs_free_lseg_list(&free_me); + return lseg; out: - return status; + return ERR_PTR(status); out_forget_reply: spin_unlock(&ino->i_lock); @@ -829,137 +1381,568 @@ out_forget_reply: goto out; } +void +pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) +{ + u64 rd_size = req->wb_bytes; + + WARN_ON_ONCE(pgio->pg_lseg != NULL); + + if (pgio->pg_dreq == NULL) + rd_size = i_size_read(pgio->pg_inode) - req_offset(req); + else + rd_size = nfs_dreq_bytes_left(pgio->pg_dreq); + + pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, + req->wb_context, + req_offset(req), + rd_size, + IOMODE_READ, + GFP_KERNEL); + /* If no lseg, fall back to read through mds */ + if (pgio->pg_lseg == NULL) + nfs_pageio_reset_read_mds(pgio); + +} +EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read); + +void +pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio, + struct nfs_page *req, u64 wb_size) +{ + WARN_ON_ONCE(pgio->pg_lseg != NULL); + + pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, + req->wb_context, + req_offset(req), + wb_size, + IOMODE_RW, + GFP_NOFS); + /* If no lseg, fall back to write through mds */ + if (pgio->pg_lseg == NULL) + nfs_pageio_reset_write_mds(pgio); +} +EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write); + /* - * Device ID cache. Currently supports one layout type per struct nfs_client. - * Add layout type to the lookup key to expand to support multiple types. + * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number + * of bytes (maximum @req->wb_bytes) that can be coalesced. */ -int -pnfs_alloc_init_deviceid_cache(struct nfs_client *clp, - void (*free_callback)(struct pnfs_deviceid_node *)) +size_t +pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev, + struct nfs_page *req) +{ + unsigned int size; + u64 seg_end, req_start, seg_left; + + size = nfs_generic_pg_test(pgio, prev, req); + if (!size) + return 0; + + /* + * 'size' contains the number of bytes left in the current page (up + * to the original size asked for in @req->wb_bytes). + * + * Calculate how many bytes are left in the layout segment + * and if there are less bytes than 'size', return that instead. + * + * Please also note that 'end_offset' is actually the offset of the + * first byte that lies outside the pnfs_layout_range. FIXME? + * + */ + if (pgio->pg_lseg) { + seg_end = end_offset(pgio->pg_lseg->pls_range.offset, + pgio->pg_lseg->pls_range.length); + req_start = req_offset(req); + WARN_ON_ONCE(req_start > seg_end); + /* start of request is past the last byte of this segment */ + if (req_start >= seg_end) + return 0; + + /* adjust 'size' iff there are fewer bytes left in the + * segment than what nfs_generic_pg_test returned */ + seg_left = seg_end - req_start; + if (seg_left < size) + size = (unsigned int)seg_left; + } + + return size; +} +EXPORT_SYMBOL_GPL(pnfs_generic_pg_test); + +int pnfs_write_done_resend_to_mds(struct inode *inode, + struct list_head *head, + const struct nfs_pgio_completion_ops *compl_ops, + struct nfs_direct_req *dreq) +{ + struct nfs_pageio_descriptor pgio; + LIST_HEAD(failed); + + /* Resend all requests through the MDS */ + nfs_pageio_init_write(&pgio, inode, FLUSH_STABLE, true, compl_ops); + pgio.pg_dreq = dreq; + while (!list_empty(head)) { + struct nfs_page *req = nfs_list_entry(head->next); + + nfs_list_remove_request(req); + if (!nfs_pageio_add_request(&pgio, req)) + nfs_list_add_request(req, &failed); + } + nfs_pageio_complete(&pgio); + + if (!list_empty(&failed)) { + /* For some reason our attempt to resend pages. Mark the + * overall send request as having failed, and let + * nfs_writeback_release_full deal with the error. + */ + list_move(&failed, head); + return -EIO; + } + return 0; +} +EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds); + +static void pnfs_ld_handle_write_error(struct nfs_pgio_data *data) +{ + struct nfs_pgio_header *hdr = data->header; + + dprintk("pnfs write error = %d\n", hdr->pnfs_error); + if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & + PNFS_LAYOUTRET_ON_ERROR) { + pnfs_return_layout(hdr->inode); + } + if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) + data->task.tk_status = pnfs_write_done_resend_to_mds(hdr->inode, + &hdr->pages, + hdr->completion_ops, + hdr->dreq); +} + +/* + * Called by non rpc-based layout drivers + */ +void pnfs_ld_write_done(struct nfs_pgio_data *data) +{ + struct nfs_pgio_header *hdr = data->header; + + trace_nfs4_pnfs_write(data, hdr->pnfs_error); + if (!hdr->pnfs_error) { + pnfs_set_layoutcommit(data); + hdr->mds_ops->rpc_call_done(&data->task, data); + } else + pnfs_ld_handle_write_error(data); + hdr->mds_ops->rpc_release(data); +} +EXPORT_SYMBOL_GPL(pnfs_ld_write_done); + +static void +pnfs_write_through_mds(struct nfs_pageio_descriptor *desc, + struct nfs_pgio_data *data) { - struct pnfs_deviceid_cache *c; + struct nfs_pgio_header *hdr = data->header; - c = kzalloc(sizeof(struct pnfs_deviceid_cache), GFP_KERNEL); - if (!c) + if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { + list_splice_tail_init(&hdr->pages, &desc->pg_list); + nfs_pageio_reset_write_mds(desc); + desc->pg_recoalesce = 1; + } + nfs_pgio_data_release(data); +} + +static enum pnfs_try_status +pnfs_try_to_write_data(struct nfs_pgio_data *wdata, + const struct rpc_call_ops *call_ops, + struct pnfs_layout_segment *lseg, + int how) +{ + struct nfs_pgio_header *hdr = wdata->header; + struct inode *inode = hdr->inode; + enum pnfs_try_status trypnfs; + struct nfs_server *nfss = NFS_SERVER(inode); + + hdr->mds_ops = call_ops; + + dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__, + inode->i_ino, wdata->args.count, wdata->args.offset, how); + trypnfs = nfss->pnfs_curr_ld->write_pagelist(wdata, how); + if (trypnfs != PNFS_NOT_ATTEMPTED) + nfs_inc_stats(inode, NFSIOS_PNFS_WRITE); + dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); + return trypnfs; +} + +static void +pnfs_do_write(struct nfs_pageio_descriptor *desc, + struct nfs_pgio_header *hdr, int how) +{ + struct nfs_pgio_data *data = hdr->data; + const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; + struct pnfs_layout_segment *lseg = desc->pg_lseg; + enum pnfs_try_status trypnfs; + + desc->pg_lseg = NULL; + trypnfs = pnfs_try_to_write_data(data, call_ops, lseg, how); + if (trypnfs == PNFS_NOT_ATTEMPTED) + pnfs_write_through_mds(desc, data); + pnfs_put_lseg(lseg); +} + +static void pnfs_writehdr_free(struct nfs_pgio_header *hdr) +{ + pnfs_put_lseg(hdr->lseg); + nfs_rw_header_free(hdr); +} +EXPORT_SYMBOL_GPL(pnfs_writehdr_free); + +int +pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc) +{ + struct nfs_rw_header *whdr; + struct nfs_pgio_header *hdr; + int ret; + + whdr = nfs_rw_header_alloc(desc->pg_rw_ops); + if (!whdr) { + desc->pg_completion_ops->error_cleanup(&desc->pg_list); + pnfs_put_lseg(desc->pg_lseg); + desc->pg_lseg = NULL; return -ENOMEM; - spin_lock(&clp->cl_lock); - if (clp->cl_devid_cache != NULL) { - atomic_inc(&clp->cl_devid_cache->dc_ref); - dprintk("%s [kref [%d]]\n", __func__, - atomic_read(&clp->cl_devid_cache->dc_ref)); - kfree(c); - } else { - /* kzalloc initializes hlists */ - spin_lock_init(&c->dc_lock); - atomic_set(&c->dc_ref, 1); - c->dc_free_callback = free_callback; - clp->cl_devid_cache = c; - dprintk("%s [new]\n", __func__); } - spin_unlock(&clp->cl_lock); + hdr = &whdr->header; + nfs_pgheader_init(desc, hdr, pnfs_writehdr_free); + hdr->lseg = pnfs_get_lseg(desc->pg_lseg); + atomic_inc(&hdr->refcnt); + ret = nfs_generic_pgio(desc, hdr); + if (ret != 0) { + pnfs_put_lseg(desc->pg_lseg); + desc->pg_lseg = NULL; + } else + pnfs_do_write(desc, hdr, desc->pg_ioflags); + if (atomic_dec_and_test(&hdr->refcnt)) + hdr->completion_ops->completion(hdr); + return ret; +} +EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages); + +int pnfs_read_done_resend_to_mds(struct inode *inode, + struct list_head *head, + const struct nfs_pgio_completion_ops *compl_ops, + struct nfs_direct_req *dreq) +{ + struct nfs_pageio_descriptor pgio; + LIST_HEAD(failed); + + /* Resend all requests through the MDS */ + nfs_pageio_init_read(&pgio, inode, true, compl_ops); + pgio.pg_dreq = dreq; + while (!list_empty(head)) { + struct nfs_page *req = nfs_list_entry(head->next); + + nfs_list_remove_request(req); + if (!nfs_pageio_add_request(&pgio, req)) + nfs_list_add_request(req, &failed); + } + nfs_pageio_complete(&pgio); + + if (!list_empty(&failed)) { + list_move(&failed, head); + return -EIO; + } return 0; } -EXPORT_SYMBOL_GPL(pnfs_alloc_init_deviceid_cache); +EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds); + +static void pnfs_ld_handle_read_error(struct nfs_pgio_data *data) +{ + struct nfs_pgio_header *hdr = data->header; + + dprintk("pnfs read error = %d\n", hdr->pnfs_error); + if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags & + PNFS_LAYOUTRET_ON_ERROR) { + pnfs_return_layout(hdr->inode); + } + if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) + data->task.tk_status = pnfs_read_done_resend_to_mds(hdr->inode, + &hdr->pages, + hdr->completion_ops, + hdr->dreq); +} /* - * Called from pnfs_layoutdriver_type->free_lseg - * last layout segment reference frees deviceid + * Called by non rpc-based layout drivers */ -void -pnfs_put_deviceid(struct pnfs_deviceid_cache *c, - struct pnfs_deviceid_node *devid) +void pnfs_ld_read_done(struct nfs_pgio_data *data) { - struct nfs4_deviceid *id = &devid->de_id; - struct pnfs_deviceid_node *d; - struct hlist_node *n; - long h = nfs4_deviceid_hash(id); + struct nfs_pgio_header *hdr = data->header; - dprintk("%s [%d]\n", __func__, atomic_read(&devid->de_ref)); - if (!atomic_dec_and_lock(&devid->de_ref, &c->dc_lock)) - return; + trace_nfs4_pnfs_read(data, hdr->pnfs_error); + if (likely(!hdr->pnfs_error)) { + __nfs4_read_done_cb(data); + hdr->mds_ops->rpc_call_done(&data->task, data); + } else + pnfs_ld_handle_read_error(data); + hdr->mds_ops->rpc_release(data); +} +EXPORT_SYMBOL_GPL(pnfs_ld_read_done); - hlist_for_each_entry_rcu(d, n, &c->dc_deviceids[h], de_node) - if (!memcmp(&d->de_id, id, sizeof(*id))) { - hlist_del_rcu(&d->de_node); - spin_unlock(&c->dc_lock); - synchronize_rcu(); - c->dc_free_callback(devid); - return; - } - spin_unlock(&c->dc_lock); - /* Why wasn't it found in the list? */ - BUG(); +static void +pnfs_read_through_mds(struct nfs_pageio_descriptor *desc, + struct nfs_pgio_data *data) +{ + struct nfs_pgio_header *hdr = data->header; + + if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { + list_splice_tail_init(&hdr->pages, &desc->pg_list); + nfs_pageio_reset_read_mds(desc); + desc->pg_recoalesce = 1; + } + nfs_pgio_data_release(data); } -EXPORT_SYMBOL_GPL(pnfs_put_deviceid); -/* Find and reference a deviceid */ -struct pnfs_deviceid_node * -pnfs_find_get_deviceid(struct pnfs_deviceid_cache *c, struct nfs4_deviceid *id) +/* + * Call the appropriate parallel I/O subsystem read function. + */ +static enum pnfs_try_status +pnfs_try_to_read_data(struct nfs_pgio_data *rdata, + const struct rpc_call_ops *call_ops, + struct pnfs_layout_segment *lseg) { - struct pnfs_deviceid_node *d; - struct hlist_node *n; - long hash = nfs4_deviceid_hash(id); + struct nfs_pgio_header *hdr = rdata->header; + struct inode *inode = hdr->inode; + struct nfs_server *nfss = NFS_SERVER(inode); + enum pnfs_try_status trypnfs; - dprintk("--> %s hash %ld\n", __func__, hash); - rcu_read_lock(); - hlist_for_each_entry_rcu(d, n, &c->dc_deviceids[hash], de_node) { - if (!memcmp(&d->de_id, id, sizeof(*id))) { - if (!atomic_inc_not_zero(&d->de_ref)) { - goto fail; - } else { - rcu_read_unlock(); - return d; - } - } + hdr->mds_ops = call_ops; + + dprintk("%s: Reading ino:%lu %u@%llu\n", + __func__, inode->i_ino, rdata->args.count, rdata->args.offset); + + trypnfs = nfss->pnfs_curr_ld->read_pagelist(rdata); + if (trypnfs != PNFS_NOT_ATTEMPTED) + nfs_inc_stats(inode, NFSIOS_PNFS_READ); + dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs); + return trypnfs; +} + +static void +pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr) +{ + struct nfs_pgio_data *data = hdr->data; + const struct rpc_call_ops *call_ops = desc->pg_rpc_callops; + struct pnfs_layout_segment *lseg = desc->pg_lseg; + enum pnfs_try_status trypnfs; + + desc->pg_lseg = NULL; + trypnfs = pnfs_try_to_read_data(data, call_ops, lseg); + if (trypnfs == PNFS_NOT_ATTEMPTED) + pnfs_read_through_mds(desc, data); + pnfs_put_lseg(lseg); +} + +static void pnfs_readhdr_free(struct nfs_pgio_header *hdr) +{ + pnfs_put_lseg(hdr->lseg); + nfs_rw_header_free(hdr); +} +EXPORT_SYMBOL_GPL(pnfs_readhdr_free); + +int +pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc) +{ + struct nfs_rw_header *rhdr; + struct nfs_pgio_header *hdr; + int ret; + + rhdr = nfs_rw_header_alloc(desc->pg_rw_ops); + if (!rhdr) { + desc->pg_completion_ops->error_cleanup(&desc->pg_list); + ret = -ENOMEM; + pnfs_put_lseg(desc->pg_lseg); + desc->pg_lseg = NULL; + return ret; } -fail: - rcu_read_unlock(); - return NULL; + hdr = &rhdr->header; + nfs_pgheader_init(desc, hdr, pnfs_readhdr_free); + hdr->lseg = pnfs_get_lseg(desc->pg_lseg); + atomic_inc(&hdr->refcnt); + ret = nfs_generic_pgio(desc, hdr); + if (ret != 0) { + pnfs_put_lseg(desc->pg_lseg); + desc->pg_lseg = NULL; + } else + pnfs_do_read(desc, hdr); + if (atomic_dec_and_test(&hdr->refcnt)) + hdr->completion_ops->completion(hdr); + return ret; +} +EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages); + +static void pnfs_clear_layoutcommitting(struct inode *inode) +{ + unsigned long *bitlock = &NFS_I(inode)->flags; + + clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock); + smp_mb__after_atomic(); + wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING); } -EXPORT_SYMBOL_GPL(pnfs_find_get_deviceid); /* - * Add a deviceid to the cache. - * GETDEVICEINFOs for same deviceid can race. If deviceid is found, discard new + * There can be multiple RW segments. */ -struct pnfs_deviceid_node * -pnfs_add_deviceid(struct pnfs_deviceid_cache *c, struct pnfs_deviceid_node *new) -{ - struct pnfs_deviceid_node *d; - long hash = nfs4_deviceid_hash(&new->de_id); - - dprintk("--> %s hash %ld\n", __func__, hash); - spin_lock(&c->dc_lock); - d = pnfs_find_get_deviceid(c, &new->de_id); - if (d) { - spin_unlock(&c->dc_lock); - dprintk("%s [discard]\n", __func__); - c->dc_free_callback(new); - return d; +static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp) +{ + struct pnfs_layout_segment *lseg; + + list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) { + if (lseg->pls_range.iomode == IOMODE_RW && + test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) + list_add(&lseg->pls_lc_list, listp); + } +} + +static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp) +{ + struct pnfs_layout_segment *lseg, *tmp; + + /* Matched by references in pnfs_set_layoutcommit */ + list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) { + list_del_init(&lseg->pls_lc_list); + pnfs_put_lseg(lseg); } - INIT_HLIST_NODE(&new->de_node); - atomic_set(&new->de_ref, 1); - hlist_add_head_rcu(&new->de_node, &c->dc_deviceids[hash]); - spin_unlock(&c->dc_lock); - dprintk("%s [new]\n", __func__); - return new; + + pnfs_clear_layoutcommitting(inode); +} + +void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg) +{ + pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode); } -EXPORT_SYMBOL_GPL(pnfs_add_deviceid); +EXPORT_SYMBOL_GPL(pnfs_set_lo_fail); void -pnfs_put_deviceid_cache(struct nfs_client *clp) +pnfs_set_layoutcommit(struct nfs_pgio_data *wdata) { - struct pnfs_deviceid_cache *local = clp->cl_devid_cache; + struct nfs_pgio_header *hdr = wdata->header; + struct inode *inode = hdr->inode; + struct nfs_inode *nfsi = NFS_I(inode); + loff_t end_pos = wdata->mds_offset + wdata->res.count; + bool mark_as_dirty = false; + + spin_lock(&inode->i_lock); + if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) { + mark_as_dirty = true; + dprintk("%s: Set layoutcommit for inode %lu ", + __func__, inode->i_ino); + } + if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &hdr->lseg->pls_flags)) { + /* references matched in nfs4_layoutcommit_release */ + pnfs_get_lseg(hdr->lseg); + } + if (end_pos > nfsi->layout->plh_lwb) + nfsi->layout->plh_lwb = end_pos; + spin_unlock(&inode->i_lock); + dprintk("%s: lseg %p end_pos %llu\n", + __func__, hdr->lseg, nfsi->layout->plh_lwb); + + /* if pnfs_layoutcommit_inode() runs between inode locks, the next one + * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */ + if (mark_as_dirty) + mark_inode_dirty_sync(inode); +} +EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit); - dprintk("--> %s ({%d})\n", __func__, atomic_read(&local->dc_ref)); - if (atomic_dec_and_lock(&local->dc_ref, &clp->cl_lock)) { - int i; - /* Verify cache is empty */ - for (i = 0; i < NFS4_DEVICE_ID_HASH_SIZE; i++) - BUG_ON(!hlist_empty(&local->dc_deviceids[i])); - clp->cl_devid_cache = NULL; - spin_unlock(&clp->cl_lock); - kfree(local); +void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data) +{ + struct nfs_server *nfss = NFS_SERVER(data->args.inode); + + if (nfss->pnfs_curr_ld->cleanup_layoutcommit) + nfss->pnfs_curr_ld->cleanup_layoutcommit(data); + pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list); +} + +/* + * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and + * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough + * data to disk to allow the server to recover the data if it crashes. + * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag + * is off, and a COMMIT is sent to a data server, or + * if WRITEs to a data server return NFS_DATA_SYNC. + */ +int +pnfs_layoutcommit_inode(struct inode *inode, bool sync) +{ + struct nfs4_layoutcommit_data *data; + struct nfs_inode *nfsi = NFS_I(inode); + loff_t end_pos; + int status; + + if (!pnfs_layoutcommit_outstanding(inode)) + return 0; + + dprintk("--> %s inode %lu\n", __func__, inode->i_ino); + + status = -EAGAIN; + if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) { + if (!sync) + goto out; + status = wait_on_bit_lock(&nfsi->flags, + NFS_INO_LAYOUTCOMMITTING, + nfs_wait_bit_killable, + TASK_KILLABLE); + if (status) + goto out; + } + + status = -ENOMEM; + /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */ + data = kzalloc(sizeof(*data), GFP_NOFS); + if (!data) + goto clear_layoutcommitting; + + status = 0; + spin_lock(&inode->i_lock); + if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) + goto out_unlock; + + INIT_LIST_HEAD(&data->lseg_list); + pnfs_list_write_lseg(inode, &data->lseg_list); + + end_pos = nfsi->layout->plh_lwb; + nfsi->layout->plh_lwb = 0; + + nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid); + spin_unlock(&inode->i_lock); + + data->args.inode = inode; + data->cred = get_rpccred(nfsi->layout->plh_lc_cred); + nfs_fattr_init(&data->fattr); + data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask; + data->res.fattr = &data->fattr; + data->args.lastbytewritten = end_pos - 1; + data->res.server = NFS_SERVER(inode); + + status = nfs4_proc_layoutcommit(data, sync); +out: + if (status) + mark_inode_dirty_sync(inode); + dprintk("<-- %s status %d\n", __func__, status); + return status; +out_unlock: + spin_unlock(&inode->i_lock); + kfree(data); +clear_layoutcommitting: + pnfs_clear_layoutcommitting(inode); + goto out; +} + +struct nfs4_threshold *pnfs_mdsthreshold_alloc(void) +{ + struct nfs4_threshold *thp; + + thp = kzalloc(sizeof(*thp), GFP_NOFS); + if (!thp) { + dprintk("%s mdsthreshold allocation failed\n", __func__); + return NULL; } + return thp; } -EXPORT_SYMBOL_GPL(pnfs_put_deviceid_cache); |