X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fobdecho%2Fecho_client.c;h=3d5e1457d6ab70ead170f2d00c4655bedcf97df9;hb=30964bcc157e3ad76714b7e55491112c3ae8ebad;hp=763ddb80c2f69a34b6d79e24f87830e6c0eda842;hpb=7312616768bfed768ecc00ba20322c37568138d0;p=fs%2Flustre-release.git diff --git a/lustre/obdecho/echo_client.c b/lustre/obdecho/echo_client.c index 763ddb8..3d5e145 100644 --- a/lustre/obdecho/echo_client.c +++ b/lustre/obdecho/echo_client.c @@ -1,7 +1,7 @@ /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- * vim:expandtab:shiftwidth=8:tabstop=8: * - * Copyright (c) 2001, 2002 Cluster File Systems, Inc. + * Copyright (c) 2001-2003 Cluster File Systems, Inc. * * This file is part of Lustre, http://www.lustre.org. * @@ -19,216 +19,1349 @@ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */ +#define DEBUG_SUBSYSTEM S_ECHO +#ifdef __KERNEL__ #include #include #include +#include +#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0)) +#include +#endif +#include +#include +#else +#include +#endif -#define DEBUG_SUBSYSTEM S_ECHO - +#include #include #include #include #include #include -static int echo_iocontrol(long cmd, struct lustre_handle *obdconn, int len, - void *karg, void *uarg) +static obd_id last_object_id; + +#if 0 +static void +echo_printk_object (char *msg, struct ec_object *eco) { - struct obd_device *obd = class_conn2obd(obdconn); - struct echo_client_obd *ec = &obd->u.echo_client; - struct obd_ioctl_data *data = karg; - int rw = OBD_BRW_READ, rc = 0; - struct lov_stripe_md *lsm = NULL; - ENTRY; + struct lov_stripe_md *lsm = eco->eco_lsm; + int i; - if (obd == NULL) { - CERROR("ioctl: No device\n"); - GOTO(out, rc = -EINVAL); + printk (KERN_INFO "Lustre: %s: object %p: "LPX64", refs %d%s: "LPX64 + "=%u!%u\n", msg, eco, eco->eco_id, eco->eco_refcount, + eco->eco_deleted ? "(deleted) " : "", + lsm->lsm_object_id, lsm->lsm_stripe_size, + lsm->lsm_stripe_count); + + for (i = 0; i < lsm->lsm_stripe_count; i++) + printk (KERN_INFO "Lustre: @%2u:"LPX64"\n", + lsm->lsm_oinfo[i].loi_ost_idx, + lsm->lsm_oinfo[i].loi_id); +} +#endif + +static struct ec_object * +echo_find_object_locked (struct obd_device *obd, obd_id id) +{ + struct echo_client_obd *ec = &obd->u.echocli; + struct ec_object *eco = NULL; + struct list_head *el; + + list_for_each (el, &ec->ec_objects) { + eco = list_entry (el, struct ec_object, eco_obj_chain); + + if (eco->eco_id == id) + return (eco); } + return (NULL); +} - if (data->ioc_inllen1 == sizeof(*lsm)) { - lsm = (struct lov_stripe_md *)data->ioc_inlbuf1; - } else if (data->ioc_inllen1 != 0) { - CERROR("nonzero ioc_inllen1 != sizeof(struct lov_stripe_md)\n"); - GOTO(out, rc = -EINVAL); +static int +echo_copyout_lsm (struct lov_stripe_md *lsm, void *ulsm, int ulsm_nob) +{ + int nob; + + nob = offsetof (struct lov_stripe_md, lsm_oinfo[lsm->lsm_stripe_count]); + if (nob > ulsm_nob) + return (-EINVAL); + + if (copy_to_user (ulsm, lsm, nob)) + return (-EFAULT); + + return (0); +} + +static int +echo_copyin_lsm (struct obd_device *obd, struct lov_stripe_md *lsm, + void *ulsm, int ulsm_nob) +{ + struct echo_client_obd *ec = &obd->u.echocli; + int nob; + + if (ulsm_nob < sizeof (*lsm)) + return (-EINVAL); + + if (copy_from_user (lsm, ulsm, sizeof (*lsm))) + return (-EFAULT); + + nob = lsm->lsm_stripe_count * sizeof (lsm->lsm_oinfo[0]); + + if (ulsm_nob < nob || + lsm->lsm_stripe_count > ec->ec_nstripes || + lsm->lsm_magic != LOV_MAGIC || + (lsm->lsm_stripe_size & (PAGE_SIZE - 1)) != 0 || + ((__u64)lsm->lsm_stripe_size * lsm->lsm_stripe_count > ~0UL)) + return (-EINVAL); + + if (copy_from_user(lsm->lsm_oinfo, + ((struct lov_stripe_md *)ulsm)->lsm_oinfo, nob)) + return (-EFAULT); + + return (0); +} + +static struct ec_object * +echo_allocate_object (struct obd_device *obd) +{ + struct echo_client_obd *ec = &obd->u.echocli; + struct ec_object *eco; + int rc; + + OBD_ALLOC(eco, sizeof (*eco)); + if (eco == NULL) + return NULL; + + rc = obd_alloc_memmd(ec->ec_exp, &eco->eco_lsm); + if (rc < 0) { + OBD_FREE(eco, sizeof (*eco)); + return NULL; } - switch (cmd) { - case OBD_IOC_CREATE: { - struct lov_stripe_md *tmp_lsm = NULL; - rc = obd_create(&ec->conn, &data->ioc_obdo1, &tmp_lsm); - if (lsm) - memcpy(lsm, tmp_lsm, sizeof(*tmp_lsm)); + eco->eco_device = obd; + eco->eco_deleted = 0; + eco->eco_refcount = 0; + eco->eco_lsm->lsm_magic = LOV_MAGIC; + /* leave stripe count 0 by default */ - GOTO(out, rc); + return (eco); +} + +static void +echo_free_object (struct ec_object *eco) +{ + struct obd_device *obd = eco->eco_device; + struct echo_client_obd *ec = &obd->u.echocli; + + LASSERT (eco->eco_refcount == 0); + obd_free_memmd(ec->ec_exp, &eco->eco_lsm); + OBD_FREE (eco, sizeof (*eco)); +} + +static int echo_create_object(struct obd_device *obd, int on_target, + struct obdo *oa, void *ulsm, int ulsm_nob, + struct obd_trans_info *oti) +{ + struct echo_client_obd *ec = &obd->u.echocli; + struct ec_object *eco2; + struct ec_object *eco; + struct lov_stripe_md *lsm; + int rc; + int i, idx; + + if ((oa->o_valid & OBD_MD_FLID) == 0 && /* no obj id */ + (on_target || /* set_stripe */ + ec->ec_nstripes != 0)) { /* LOV */ + CERROR ("No valid oid\n"); + return (-EINVAL); } - case OBD_IOC_GETATTR: - rc = obd_getattr(&ec->conn, &data->ioc_obdo1, lsm); - GOTO(out, rc); + if (ulsm != NULL) { + eco = echo_allocate_object (obd); + if (eco == NULL) + return (-ENOMEM); - case OBD_IOC_SETATTR: - rc = obd_setattr(&ec->conn, &data->ioc_obdo1, lsm); - GOTO(out, rc); + lsm = eco->eco_lsm; - case OBD_IOC_DESTROY: - rc = obd_destroy(&ec->conn, &data->ioc_obdo1, lsm); - GOTO(out, rc); + rc = echo_copyin_lsm (obd, lsm, ulsm, ulsm_nob); + if (rc != 0) + goto failed; - case OBD_IOC_OPEN: - rc = obd_open(&ec->conn, &data->ioc_obdo1, lsm); - GOTO(out, rc); + /* setup object ID here for !on_target and LOV hint */ + if ((oa->o_valid & OBD_MD_FLID) != 0) + eco->eco_id = lsm->lsm_object_id = oa->o_id; - case OBD_IOC_CLOSE: - rc = obd_close(&ec->conn, &data->ioc_obdo1, lsm); - GOTO(out, rc); + if (lsm->lsm_stripe_count == 0) + lsm->lsm_stripe_count = ec->ec_nstripes; - case OBD_IOC_BRW_WRITE: - rw = OBD_BRW_WRITE; - case OBD_IOC_BRW_READ: { - struct lov_stripe_md tmp_lsm; - struct obd_brw_set *set; - obd_count pages = 0; - struct brw_page *pga, *pgp; - __u64 off, id = data->ioc_obdo1.o_id; - int gfp_mask = (id & 1) ? GFP_HIGHUSER : GFP_KERNEL; - int j, verify = (id != 0); - - if (lsm && lsm->lsm_object_id != id) { - CERROR("LSM object ID ("LPU64") != id ("LPU64")\n", - lsm->lsm_object_id, id); - GOTO(out, rc = -EINVAL); + if (lsm->lsm_stripe_size == 0) + lsm->lsm_stripe_size = PAGE_SIZE; + + idx = ll_insecure_random_int(); + + /* setup stripes: indices + default ids if required */ + for (i = 0; i < lsm->lsm_stripe_count; i++) { + if (lsm->lsm_oinfo[i].loi_id == 0) + lsm->lsm_oinfo[i].loi_id = lsm->lsm_object_id; + + lsm->lsm_oinfo[i].loi_ost_idx = + (idx + i) % ec->ec_nstripes; } + } else { + OBD_ALLOC(eco, sizeof(*eco)); + eco->eco_device = obd; + lsm = NULL; + } + + if (oa->o_id == 0) + oa->o_id = ++last_object_id; + + if (on_target) { + oa->o_gr = FILTER_GROUP_ECHO; + oa->o_valid |= OBD_MD_FLGROUP; + rc = obd_create(ec->ec_exp, oa, NULL, 0, &lsm, oti); + if (rc != 0) + goto failed; + + /* See what object ID we were given */ + eco->eco_id = oa->o_id = lsm->lsm_object_id; + oa->o_valid |= OBD_MD_FLID; + + LASSERT(eco->eco_lsm == NULL || eco->eco_lsm == lsm); + eco->eco_lsm = lsm; + } + + spin_lock (&ec->ec_lock); + + eco2 = echo_find_object_locked (obd, oa->o_id); + if (eco2 != NULL) { /* conflict */ + spin_unlock (&ec->ec_lock); + + CERROR ("Can't create object id "LPX64": id already exists%s\n", + oa->o_id, on_target ? " (undoing create)" : ""); + + if (on_target) + obd_destroy(ec->ec_exp, oa, lsm, oti); + + rc = -EEXIST; + goto failed; + } + + list_add (&eco->eco_obj_chain, &ec->ec_objects); + spin_unlock (&ec->ec_lock); + CDEBUG (D_INFO, + "created %p: "LPX64"=%u#%u@%u refs %d del %d\n", + eco, eco->eco_id, + eco->eco_lsm->lsm_stripe_size, + eco->eco_lsm->lsm_stripe_count, + eco->eco_lsm->lsm_oinfo[0].loi_ost_idx, + eco->eco_refcount, eco->eco_deleted); + return (0); + + failed: + echo_free_object (eco); + return (rc); +} + +static int +echo_get_object (struct ec_object **ecop, struct obd_device *obd, + struct obdo *oa) +{ + struct echo_client_obd *ec = &obd->u.echocli; + struct ec_object *eco; + struct ec_object *eco2; + int rc; + + if ((oa->o_valid & OBD_MD_FLID) == 0 || + oa->o_id == 0) /* disallow use of object id 0 */ + { + CERROR ("No valid oid\n"); + return (-EINVAL); + } + + spin_lock (&ec->ec_lock); + eco = echo_find_object_locked (obd, oa->o_id); + if (eco != NULL) { + if (eco->eco_deleted) /* being deleted */ + return (-EAGAIN); /* (see comment in cleanup) */ + + eco->eco_refcount++; + spin_unlock (&ec->ec_lock); + *ecop = eco; + CDEBUG (D_INFO, + "found %p: "LPX64"=%u#%u@%u refs %d del %d\n", + eco, eco->eco_id, + eco->eco_lsm->lsm_stripe_size, + eco->eco_lsm->lsm_stripe_count, + eco->eco_lsm->lsm_oinfo[0].loi_ost_idx, + eco->eco_refcount, eco->eco_deleted); + return (0); + } + spin_unlock (&ec->ec_lock); + + if (ec->ec_nstripes != 0) /* striping required */ + return (-ENOENT); + + eco = echo_allocate_object (obd); + if (eco == NULL) + return (-ENOMEM); + + eco->eco_id = eco->eco_lsm->lsm_object_id = oa->o_id; + + spin_lock (&ec->ec_lock); + + eco2 = echo_find_object_locked (obd, oa->o_id); + if (eco2 == NULL) { /* didn't race */ + list_add (&eco->eco_obj_chain, &ec->ec_objects); + spin_unlock (&ec->ec_lock); + eco->eco_refcount = 1; + *ecop = eco; + CDEBUG (D_INFO, + "created %p: "LPX64"=%u#%u@%d refs %d del %d\n", + eco, eco->eco_id, + eco->eco_lsm->lsm_stripe_size, + eco->eco_lsm->lsm_stripe_count, + eco->eco_lsm->lsm_oinfo[0].loi_ost_idx, + eco->eco_refcount, eco->eco_deleted); + return (0); + } + + if (eco2->eco_deleted) + rc = -EAGAIN; /* lose race */ + else { + eco2->eco_refcount++; /* take existing */ + *ecop = eco2; + rc = 0; + LASSERT (eco2->eco_id == eco2->eco_lsm->lsm_object_id); + CDEBUG (D_INFO, + "found(2) %p: "LPX64"=%u#%u@%d refs %d del %d\n", + eco2, eco2->eco_id, + eco2->eco_lsm->lsm_stripe_size, + eco2->eco_lsm->lsm_stripe_count, + eco2->eco_lsm->lsm_oinfo[0].loi_ost_idx, + eco2->eco_refcount, eco2->eco_deleted); + } + + spin_unlock (&ec->ec_lock); + + echo_free_object (eco); + return (rc); +} + +static void +echo_put_object (struct ec_object *eco) +{ + struct obd_device *obd = eco->eco_device; + struct echo_client_obd *ec = &obd->u.echocli; + + /* Release caller's ref on the object. + * delete => mark for deletion when last ref goes + */ - if (!lsm) { - memset(&tmp_lsm, 0, sizeof(tmp_lsm)); - lsm = &tmp_lsm; - lsm->lsm_object_id = id; + spin_lock (&ec->ec_lock); + + eco->eco_refcount--; + LASSERT (eco->eco_refcount >= 0); + + CDEBUG(D_INFO, "put %p: "LPX64"=%u#%u@%d refs %d del %d\n", + eco, eco->eco_id, + eco->eco_lsm->lsm_stripe_size, + eco->eco_lsm->lsm_stripe_count, + eco->eco_lsm->lsm_oinfo[0].loi_ost_idx, + eco->eco_refcount, eco->eco_deleted); + + if (eco->eco_refcount != 0 || !eco->eco_deleted) { + spin_unlock (&ec->ec_lock); + return; + } + + spin_unlock (&ec->ec_lock); + + /* NB leave obj in the object list. We must prevent anyone from + * attempting to enqueue on this object number until we can be + * sure there will be no more lock callbacks. + */ + obd_cancel_unused(ec->ec_exp, eco->eco_lsm, 0, NULL); + + /* now we can let it go */ + spin_lock (&ec->ec_lock); + list_del (&eco->eco_obj_chain); + spin_unlock (&ec->ec_lock); + + LASSERT (eco->eco_refcount == 0); + + echo_free_object (eco); +} + +static void +echo_get_stripe_off_id (struct lov_stripe_md *lsm, obd_off *offp, obd_id *idp) +{ + unsigned long stripe_count; + unsigned long stripe_size; + unsigned long width; + unsigned long woffset; + int stripe_index; + obd_off offset; + + if (lsm->lsm_stripe_count <= 1) + return; + + offset = *offp; + stripe_size = lsm->lsm_stripe_size; + stripe_count = lsm->lsm_stripe_count; + + /* width = # bytes in all stripes */ + width = stripe_size * stripe_count; + + /* woffset = offset within a width; offset = whole number of widths */ + woffset = do_div (offset, width); + + stripe_index = woffset / stripe_size; + + *idp = lsm->lsm_oinfo[stripe_index].loi_id; + *offp = offset * stripe_size + woffset % stripe_size; +} + +static void +echo_client_page_debug_setup(struct lov_stripe_md *lsm, + struct page *page, int rw, obd_id id, + obd_off offset, obd_off count) +{ + char *addr; + obd_off stripe_off; + obd_id stripe_id; + int delta; + + /* no partial pages on the client */ + LASSERT(count == PAGE_SIZE); + + addr = kmap(page); + + for (delta = 0; delta < PAGE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) { + if (rw == OBD_BRW_WRITE) { + stripe_off = offset + delta; + stripe_id = id; + echo_get_stripe_off_id(lsm, &stripe_off, &stripe_id); + } else { + stripe_off = 0xdeadbeef00c0ffeeULL; + stripe_id = 0xdeadbeef00c0ffeeULL; } + block_debug_setup(addr + delta, OBD_ECHO_BLOCK_SIZE, + stripe_off, stripe_id); + } + + kunmap(page); +} + +static int +echo_client_page_debug_check(struct lov_stripe_md *lsm, + struct page *page, obd_id id, + obd_off offset, obd_off count) +{ + obd_off stripe_off; + obd_id stripe_id; + char *addr; + int delta; + int rc; + int rc2; - if (data->ioc_count < 0) { - CERROR("invalid buffer size: "LPD64"\n", - data->ioc_count); - GOTO(out, rc = -EINVAL); + /* no partial pages on the client */ + LASSERT(count == PAGE_SIZE); + + addr = kmap(page); + + for (rc = delta = 0; delta < PAGE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) { + stripe_off = offset + delta; + stripe_id = id; + echo_get_stripe_off_id (lsm, &stripe_off, &stripe_id); + + rc2 = block_debug_check("test_brw", + addr + delta, OBD_ECHO_BLOCK_SIZE, + stripe_off, stripe_id); + if (rc2 != 0) { + CERROR ("Error in echo object "LPX64"\n", id); + rc = rc2; } + } + + kunmap(page); + return rc; +} + +static int echo_client_kbrw(struct obd_device *obd, int rw, struct obdo *oa, + struct lov_stripe_md *lsm, obd_off offset, + obd_size count, struct obd_trans_info *oti) +{ + struct echo_client_obd *ec = &obd->u.echocli; + obd_count npages; + struct brw_page *pga; + struct brw_page *pgp; + obd_off off; + int i; + int rc; + int verify = 0; + int gfp_mask; + + verify = ((oa->o_id) != ECHO_PERSISTENT_OBJID && + (oa->o_valid & OBD_MD_FLFLAGS) != 0 && + (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0); + + gfp_mask = ((oa->o_id & 2) == 0) ? GFP_KERNEL : GFP_HIGHUSER; + + LASSERT(rw == OBD_BRW_WRITE || rw == OBD_BRW_READ); + + if (count <= 0 || + (count & (PAGE_SIZE - 1)) != 0 || + (lsm != NULL && + lsm->lsm_object_id != oa->o_id)) + return (-EINVAL); + + /* XXX think again with misaligned I/O */ + npages = count >> PAGE_SHIFT; + + OBD_ALLOC(pga, npages * sizeof(*pga)); + if (pga == NULL) + return (-ENOMEM); + + for (i = 0, pgp = pga, off = offset; + i < npages; + i++, pgp++, off += PAGE_SIZE) { + + LASSERT (pgp->pg == NULL); /* for cleanup */ + + rc = -ENOMEM; + pgp->pg = alloc_pages (gfp_mask, 0); + if (pgp->pg == NULL) + goto out; + + pgp->count = PAGE_SIZE; + pgp->disk_offset = pgp->page_offset = off; + pgp->flag = 0; + + if (verify) + echo_client_page_debug_setup(lsm, pgp->pg, rw, + oa->o_id, off, pgp->count); + } + + rc = obd_brw(rw, ec->ec_exp, oa, lsm, npages, pga, oti); + + out: + if (rc != 0 || rw != OBD_BRW_READ) + verify = 0; + + for (i = 0, pgp = pga; i < npages; i++, pgp++) { + if (pgp->pg == NULL) + continue; + + if (verify) { + int vrc; + vrc = echo_client_page_debug_check(lsm, pgp->pg, oa->o_id, + pgp->page_offset, + pgp->count); + if (vrc != 0 && rc == 0) + rc = vrc; + } + __free_pages(pgp->pg, 0); + } + OBD_FREE(pga, npages * sizeof(*pga)); + return (rc); +} + +#ifdef __KERNEL__ +#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0)) +static int echo_client_ubrw(struct obd_device *obd, int rw, + struct obdo *oa, struct lov_stripe_md *lsm, + obd_off offset, obd_size count, char *buffer, + struct obd_trans_info *oti) +{ + struct echo_client_obd *ec = &obd->u.echocli; + obd_count npages; + struct brw_page *pga; + struct brw_page *pgp; + obd_off off; + struct kiobuf *kiobuf; + int i; + int rc; + + LASSERT (rw == OBD_BRW_WRITE || + rw == OBD_BRW_READ); + + /* NB: for now, only whole pages, page aligned */ + + if (count <= 0 || + ((long)buffer & (PAGE_SIZE - 1)) != 0 || + (count & (PAGE_SIZE - 1)) != 0 || + (lsm != NULL && lsm->lsm_object_id != oa->o_id)) + return (-EINVAL); + + /* XXX think again with misaligned I/O */ + npages = count >> PAGE_SHIFT; + + OBD_ALLOC(pga, npages * sizeof(*pga)); + if (pga == NULL) + return (-ENOMEM); + + rc = alloc_kiovec (1, &kiobuf); + if (rc != 0) + goto out_1; + + rc = map_user_kiobuf ((rw == OBD_BRW_READ) ? READ : WRITE, + kiobuf, (unsigned long)buffer, count); + if (rc != 0) + goto out_2; + + LASSERT (kiobuf->offset == 0); + LASSERT (kiobuf->nr_pages == npages); + + for (i = 0, off = offset, pgp = pga; + i < npages; + i++, off += PAGE_SIZE, pgp++) { + pgp->disk_offset = pgp->page_offset = off; + pgp->pg = kiobuf->maplist[i]; + pgp->count = PAGE_SIZE; + pgp->flag = 0; + } + + rc = obd_brw(rw, ec->ec_exp, oa, lsm, npages, pga, oti); + + // if (rw == OBD_BRW_READ) + // mark_dirty_kiobuf (kiobuf, count); + + unmap_kiobuf (kiobuf); + out_2: + free_kiovec (1, &kiobuf); + out_1: + OBD_FREE(pga, npages * sizeof(*pga)); + return (rc); +} +#else +static int echo_client_ubrw(struct obd_device *obd, int rw, + struct obdo *oa, struct lov_stripe_md *lsm, + obd_off offset, obd_size count, char *buffer, + struct obd_trans_info *oti) +{ +#warning "echo_client_ubrw() needs to be ported on 2.6 yet" + LBUG(); + return 0; +} +#endif +#endif + +struct echo_async_state; + +#define EAP_MAGIC 79277927 +struct echo_async_page { + int eap_magic; + struct page *eap_page; + void *eap_cookie; + obd_off eap_off; + struct echo_async_state *eap_eas; + struct list_head eap_item; +}; + +#define EAP_FROM_COOKIE(c) \ + (LASSERT(((struct echo_async_page *)(c))->eap_magic == EAP_MAGIC), \ + (struct echo_async_page *)(c)) + +struct echo_async_state { + spinlock_t eas_lock; + obd_off eas_next_offset; + obd_off eas_end_offset; + int eas_in_flight; + int eas_rc; + wait_queue_head_t eas_waitq; + struct list_head eas_avail; + struct obdo eas_oa; + struct lov_stripe_md *eas_lsm; +}; + +static int eas_should_wake(struct echo_async_state *eas) +{ + unsigned long flags; + int rc = 0; + spin_lock_irqsave(&eas->eas_lock, flags); + if (eas->eas_rc == 0 && !list_empty(&eas->eas_avail)) + rc = 1; + spin_unlock_irqrestore(&eas->eas_lock, flags); + return rc; +}; + +static int ec_ap_make_ready(void *data, int cmd) +{ + /* our pages are issued ready */ + LBUG(); + return 0; +} +static int ec_ap_refresh_count(void *data, int cmd) +{ + /* our pages are issued with a stable count */ + LBUG(); + return PAGE_SIZE; +} +static void ec_ap_fill_obdo(void *data, int cmd, struct obdo *oa) +{ + struct echo_async_page *eap = EAP_FROM_COOKIE(data); + + memcpy(oa, &eap->eap_eas->eas_oa, sizeof(*oa)); +} + +static void ec_ap_completion(void *data, int cmd, struct obdo *oa, int rc) +{ + struct echo_async_page *eap = EAP_FROM_COOKIE(data); + struct echo_async_state *eas; + unsigned long flags; + + eas = eap->eap_eas; + + if (cmd == OBD_BRW_READ && + eas->eas_oa.o_id != ECHO_PERSISTENT_OBJID && + (eas->eas_oa.o_valid & OBD_MD_FLFLAGS) != 0 && + (eas->eas_oa.o_flags & OBD_FL_DEBUG_CHECK) != 0) + echo_client_page_debug_check(eas->eas_lsm, eap->eap_page, + eas->eas_oa.o_id, eap->eap_off, + PAGE_SIZE); + + spin_lock_irqsave(&eas->eas_lock, flags); + if (rc && !eas->eas_rc) + eas->eas_rc = rc; + eas->eas_in_flight--; + list_add(&eap->eap_item, &eas->eas_avail); + wake_up(&eas->eas_waitq); + spin_unlock_irqrestore(&eas->eas_lock, flags); +} + +static struct obd_async_page_ops ec_async_page_ops = { + .ap_make_ready = ec_ap_make_ready, + .ap_refresh_count = ec_ap_refresh_count, + .ap_fill_obdo = ec_ap_fill_obdo, + .ap_completion = ec_ap_completion, +}; + +static int echo_client_async_page(struct obd_export *exp, int rw, + struct obdo *oa, struct lov_stripe_md *lsm, + obd_off offset, obd_size count, + obd_size batching) +{ + obd_count npages, i; + struct echo_async_page *eap; + struct echo_async_state eas; + struct list_head *pos, *n; + int rc = 0; + unsigned long flags; + LIST_HEAD(pages); +#if 0 + int verify; + int gfp_mask; + /* oa_id == 0 => speed test (no verification) else... + * oa & 1 => use HIGHMEM + */ + verify = (oa->o_id != 0); + gfp_mask = ((oa->o_id & 1) == 0) ? GFP_KERNEL : GFP_HIGHUSER; +#endif + + LASSERT(rw == OBD_BRW_WRITE || rw == OBD_BRW_READ); - set = obd_brw_set_new(); - if (set == NULL) + if (count <= 0 || + (count & (PAGE_SIZE - 1)) != 0 || + (lsm != NULL && + lsm->lsm_object_id != oa->o_id)) + return (-EINVAL); + + /* XXX think again with misaligned I/O */ + npages = batching >> PAGE_SHIFT; + + memcpy(&eas.eas_oa, oa, sizeof(*oa)); + eas.eas_next_offset = offset; + eas.eas_end_offset = offset + count; + spin_lock_init(&eas.eas_lock); + init_waitqueue_head(&eas.eas_waitq); + eas.eas_in_flight = 0; + eas.eas_rc = 0; + eas.eas_lsm = lsm; + INIT_LIST_HEAD(&eas.eas_avail); + + /* prepare the group of pages that we're going to be keeping + * in flight */ + for (i = 0; i < npages; i++) { + struct page *page = alloc_page(GFP_KERNEL); + if (page == NULL) GOTO(out, rc = -ENOMEM); - pages = data->ioc_count / PAGE_SIZE; - off = data->ioc_offset; + page->private = 0; + list_add_tail(&PAGE_LIST(page), &pages); - CDEBUG(D_INODE, "BRW %s with %d pages @ "LPX64"\n", - rw == OBD_BRW_READ ? "read" : "write", pages, off); - OBD_ALLOC(pga, pages * sizeof(*pga)); - if (!pga) { - CERROR("no memory for %d BRW per-page data\n", pages); - GOTO(brw_free, rc = -ENOMEM); + OBD_ALLOC(eap, sizeof(*eap)); + if (eap == NULL) + GOTO(out, rc = -ENOMEM); + + eap->eap_magic = EAP_MAGIC; + eap->eap_page = page; + eap->eap_eas = &eas; + page->private = (unsigned long)eap; + list_add_tail(&eap->eap_item, &eas.eas_avail); + } + + /* first we spin queueing io and being woken by its completion */ + spin_lock_irqsave(&eas.eas_lock, flags); + for(;;) { + int rc; + + /* sleep until we have a page to send */ + spin_unlock_irqrestore(&eas.eas_lock, flags); + rc = wait_event_interruptible(eas.eas_waitq, + eas_should_wake(&eas)); + spin_lock_irqsave(&eas.eas_lock, flags); + if (rc && !eas.eas_rc) + eas.eas_rc = rc; + if (eas.eas_rc) + break; + if (list_empty(&eas.eas_avail)) + continue; + eap = list_entry(eas.eas_avail.next, struct echo_async_page, + eap_item); + list_del(&eap->eap_item); + spin_unlock_irqrestore(&eas.eas_lock, flags); + + /* unbind the eap from its old page offset */ + if (eap->eap_cookie != NULL) { + obd_teardown_async_page(exp, lsm, NULL, + eap->eap_cookie); + eap->eap_cookie = NULL; } - for (j = 0, pgp = pga; j < pages; j++, off += PAGE_SIZE, pgp++){ - pgp->pg = alloc_pages(gfp_mask, 0); - if (!pgp->pg) { - CERROR("no memory for brw pages\n"); - GOTO(brw_cleanup, rc = -ENOMEM); - } - pgp->count = PAGE_SIZE; - pgp->off = off; - pgp->flag = 0; - - if (verify) { - void *addr = kmap(pgp->pg); - - if (rw == OBD_BRW_WRITE) - page_debug_setup(addr, pgp->count, - pgp->off, id); - else - page_debug_setup(addr, pgp->count, - 0xdeadbeef00c0ffee, - 0xdeadbeef00c0ffee); - kunmap(pgp->pg); - } + eas.eas_next_offset += PAGE_SIZE; + eap->eap_off = eas.eas_next_offset; + + rc = obd_prep_async_page(exp, lsm, NULL, eap->eap_page, + eap->eap_off, &ec_async_page_ops, + eap, &eap->eap_cookie); + if (rc) { + spin_lock_irqsave(&eas.eas_lock, flags); + eas.eas_rc = rc; + break; + } + + if (oa->o_id != ECHO_PERSISTENT_OBJID && + (oa->o_valid & OBD_MD_FLFLAGS) != 0 && + (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0) + echo_client_page_debug_setup(lsm, eap->eap_page, rw, + oa->o_id, + eap->eap_off, PAGE_SIZE); + + /* always asserts urgent, which isn't quite right */ + rc = obd_queue_async_io(exp, lsm, NULL, eap->eap_cookie, + rw, 0, PAGE_SIZE, 0, + ASYNC_READY | ASYNC_URGENT | + ASYNC_COUNT_STABLE); + spin_lock_irqsave(&eas.eas_lock, flags); + if (rc && !eas.eas_rc) { + eas.eas_rc = rc; + break; + } + eas.eas_in_flight++; + if (eas.eas_next_offset == eas.eas_end_offset) + break; + } + + /* still hold the eas_lock here.. */ + + /* now we just spin waiting for all the rpcs to complete */ + while(eas.eas_in_flight) { + spin_unlock_irqrestore(&eas.eas_lock, flags); + wait_event_interruptible(eas.eas_waitq, + eas.eas_in_flight == 0); + spin_lock_irqsave(&eas.eas_lock, flags); + } + spin_unlock_irqrestore(&eas.eas_lock, flags); + +out: + list_for_each_safe(pos, n, &pages) { + struct page *page = list_entry(pos, struct page, + PAGE_LIST_ENTRY); + + list_del(&PAGE_LIST(page)); + if (page->private != 0) { + eap = (struct echo_async_page *)page->private; + if (eap->eap_cookie != NULL) + obd_teardown_async_page(exp, lsm, NULL, + eap->eap_cookie); + OBD_FREE(eap, sizeof(*eap)); } + __free_page(page); + } + + RETURN(rc); +} + +static int echo_client_prep_commit(struct obd_export *exp, int rw, + struct obdo *oa, struct lov_stripe_md *lsm, + obd_off offset, obd_size count, + obd_size batch, struct obd_trans_info *oti) +{ + struct obd_ioobj ioo; + struct niobuf_local *lnb; + struct niobuf_remote *rnb; + obd_off off; + obd_size npages, tot_pages; + int i, ret = 0; + ENTRY; + + if (count <= 0 || (count & (PAGE_SIZE - 1)) != 0 || + (lsm != NULL && lsm->lsm_object_id != oa->o_id)) + RETURN(-EINVAL); + + npages = batch >> PAGE_SHIFT; + tot_pages = count >> PAGE_SHIFT; + + OBD_ALLOC(lnb, npages * sizeof(struct niobuf_local)); + OBD_ALLOC(rnb, npages * sizeof(struct niobuf_remote)); + + if (lnb == NULL || rnb == NULL) + GOTO(out, ret = -ENOMEM); + + obdo_to_ioobj(oa, &ioo); - set->brw_callback = ll_brw_sync_wait; - rc = obd_brw(rw, &ec->conn, lsm, j, pga, set); - if (rc) - CERROR("test_brw: error from obd_brw: rc = %d\n", rc); - else { - rc = ll_brw_sync_wait(set, CB_PHASE_START); - if (rc) - CERROR("test_brw: error from callback: rc = " - "%d\n", rc); + off = offset; + + for(; tot_pages; tot_pages -= npages) { + if (tot_pages < npages) + npages = tot_pages; + + for (i = 0; i < npages; i++, off += PAGE_SIZE) { + rnb[i].offset = off; + rnb[i].len = PAGE_SIZE; } - EXIT; - brw_cleanup: - for (j = 0, pgp = pga; j < pages; j++, pgp++) { - if (pgp->pg == NULL) + + /* XXX this can't be the best.. */ + memset(oti, 0, sizeof(*oti)); + ioo.ioo_bufcnt = npages; + + ret = obd_preprw(rw, exp, oa, 1, &ioo, npages, rnb, lnb, oti); + if (ret != 0) + GOTO(out, ret); + + for (i = 0; i < npages; i++) { + struct page *page = lnb[i].page; + + /* read past eof? */ + if (page == NULL && lnb[i].rc == 0) continue; - if (verify && !rc) { - void *addr = kmap(pgp->pg); + if (oa->o_id == ECHO_PERSISTENT_OBJID || + (oa->o_valid & OBD_MD_FLFLAGS) == 0 || + (oa->o_flags & OBD_FL_DEBUG_CHECK) == 0) + continue; - rc = page_debug_check("test_brw", addr, - PAGE_SIZE, pgp->off, id); - kunmap(pgp->pg); - } - __free_pages(pgp->pg, 0); + + if (rw == OBD_BRW_WRITE) + echo_client_page_debug_setup(lsm, page, rw, + oa->o_id, + rnb[i].offset, + rnb[i].len); + else + echo_client_page_debug_check(lsm, page, + oa->o_id, + rnb[i].offset, + rnb[i].len); } - brw_free: - obd_brw_set_free(set); - OBD_FREE(pga, pages * sizeof(*pga)); - GOTO(out, rc); + + ret = obd_commitrw(rw, exp, oa, 1, &ioo, npages, lnb, oti, ret); + if (ret != 0) + GOTO(out, ret); + } + +out: + if (lnb) + OBD_FREE(lnb, npages * sizeof(struct niobuf_local)); + if (rnb) + OBD_FREE(rnb, npages * sizeof(struct niobuf_remote)); + RETURN(ret); +} + +int echo_client_brw_ioctl(int rw, struct obd_export *exp, + struct obd_ioctl_data *data) +{ + struct obd_device *obd = class_exp2obd(exp); + struct echo_client_obd *ec = &obd->u.echocli; + struct obd_trans_info dummy_oti; + struct ec_object *eco; + int rc; + ENTRY; + + rc = echo_get_object(&eco, obd, &data->ioc_obdo1); + if (rc) + RETURN(rc); + + memset(&dummy_oti, 0, sizeof(dummy_oti)); + + data->ioc_obdo1.o_valid &= ~OBD_MD_FLHANDLE; + data->ioc_obdo1.o_valid |= OBD_MD_FLGROUP; + data->ioc_obdo1.o_gr = FILTER_GROUP_ECHO; + + switch((long)data->ioc_pbuf1) { + case 1: + if (data->ioc_pbuf2 == NULL) { // NULL user data pointer + rc = echo_client_kbrw(obd, rw, &data->ioc_obdo1, + eco->eco_lsm, data->ioc_offset, + data->ioc_count, &dummy_oti); + } else { +#ifdef __KERNEL__ + rc = echo_client_ubrw(obd, rw, &data->ioc_obdo1, + eco->eco_lsm, data->ioc_offset, + data->ioc_count, data->ioc_pbuf2, + &dummy_oti); +#endif + } + break; + case 2: + rc = echo_client_async_page(ec->ec_exp, rw, &data->ioc_obdo1, + eco->eco_lsm, data->ioc_offset, + data->ioc_count, data->ioc_plen1); + break; + case 3: + rc = echo_client_prep_commit(ec->ec_exp, rw, &data->ioc_obdo1, + eco->eco_lsm, data->ioc_offset, + data->ioc_count, data->ioc_plen1, + &dummy_oti); + break; + default: + rc = -EINVAL; + } + echo_put_object(eco); + RETURN(rc); +} + +static int +echo_ldlm_callback (struct ldlm_lock *lock, struct ldlm_lock_desc *new, + void *data, int flag) +{ + struct ec_object *eco = (struct ec_object *)data; + struct echo_client_obd *ec = &(eco->eco_device->u.echocli); + struct lustre_handle lockh; + struct list_head *el; + int found = 0; + int rc; + + ldlm_lock2handle (lock, &lockh); + + /* #ifdef this out if we're not feeling paranoid */ + spin_lock (&ec->ec_lock); + list_for_each (el, &ec->ec_objects) { + found = (eco == list_entry(el, struct ec_object, + eco_obj_chain)); + if (found) + break; } + spin_unlock (&ec->ec_lock); + LASSERT (found); + + switch (flag) { + case LDLM_CB_BLOCKING: + CDEBUG(D_INFO, "blocking callback on "LPX64", handle "LPX64"\n", + eco->eco_id, lockh.cookie); + rc = ldlm_cli_cancel (&lockh); + if (rc != ELDLM_OK) + CERROR ("ldlm_cli_cancel failed: %d\n", rc); + break; + + case LDLM_CB_CANCELING: + CDEBUG(D_INFO, "cancel callback on "LPX64", handle "LPX64"\n", + eco->eco_id, lockh.cookie); + break; + default: - return -ENOTTY; + LBUG (); } + return (0); +} + +static int +echo_client_enqueue(struct obd_export *exp, struct obdo *oa, + int mode, obd_off offset, obd_size nob) +{ + struct obd_device *obd = exp->exp_obd; + struct echo_client_obd *ec = &obd->u.echocli; + struct lustre_handle *ulh = obdo_handle (oa); + struct ec_object *eco; + struct ec_lock *ecl; + int flags; + int rc; + + if (!(mode == LCK_PR || mode == LCK_PW)) + return -EINVAL; + + if ((offset & (PAGE_SIZE - 1)) != 0 || + (nob & (PAGE_SIZE - 1)) != 0) + return -EINVAL; + + rc = echo_get_object (&eco, obd, oa); + if (rc != 0) + return rc; + + rc = -ENOMEM; + OBD_ALLOC (ecl, sizeof (*ecl)); + if (ecl == NULL) + goto failed_0; + + ecl->ecl_mode = mode; + ecl->ecl_object = eco; + ecl->ecl_policy.l_extent.start = offset; + ecl->ecl_policy.l_extent.end = + (nob == 0) ? ((obd_off) -1) : (offset + nob - 1); + + flags = 0; + rc = obd_enqueue(ec->ec_exp, eco->eco_lsm, LDLM_EXTENT, + &ecl->ecl_policy, mode, &flags, echo_ldlm_callback, + ldlm_completion_ast, NULL, eco, sizeof(struct ost_lvb), + lustre_swab_ost_lvb, &ecl->ecl_lock_handle); + if (rc != 0) + goto failed_1; + + CDEBUG(D_INFO, "enqueue handle "LPX64"\n", ecl->ecl_lock_handle.cookie); + + /* NB ecl takes object ref from echo_get_object() above */ + spin_lock(&ec->ec_lock); + + list_add(&ecl->ecl_exp_chain, &exp->exp_ec_data.eced_locks); + ulh->cookie = ecl->ecl_cookie = ec->ec_unique++; + + spin_unlock(&ec->ec_lock); + + oa->o_valid |= OBD_MD_FLHANDLE; + return 0; + + failed_1: + OBD_FREE (ecl, sizeof (*ecl)); + failed_0: + echo_put_object (eco); + return (rc); +} + +static int +echo_client_cancel(struct obd_export *exp, struct obdo *oa) +{ + struct obd_device *obd = exp->exp_obd; + struct echo_client_obd *ec = &obd->u.echocli; + struct lustre_handle *ulh = obdo_handle (oa); + struct ec_lock *ecl = NULL; + int found = 0; + struct list_head *el; + int rc; + + if ((oa->o_valid & OBD_MD_FLHANDLE) == 0) + return -EINVAL; + + spin_lock (&ec->ec_lock); + + list_for_each (el, &exp->exp_ec_data.eced_locks) { + ecl = list_entry (el, struct ec_lock, ecl_exp_chain); + found = (ecl->ecl_cookie == ulh->cookie); + if (found) { + list_del (&ecl->ecl_exp_chain); + break; + } + } + + spin_unlock (&ec->ec_lock); + + if (!found) + return (-ENOENT); + + rc = obd_cancel(ec->ec_exp, ecl->ecl_object->eco_lsm, ecl->ecl_mode, + &ecl->ecl_lock_handle); + + echo_put_object (ecl->ecl_object); + OBD_FREE (ecl, sizeof (*ecl)); + + return rc; +} + +static int +echo_client_iocontrol(unsigned int cmd, struct obd_export *exp, + int len, void *karg, void *uarg) +{ + struct obd_device *obd; + struct echo_client_obd *ec; + struct ec_object *eco; + struct obd_ioctl_data *data = karg; + struct obd_trans_info dummy_oti; + struct oti_req_ack_lock *ack_lock; + struct obdo *oa; + int rw = OBD_BRW_READ; + int rc = 0; + int i; + ENTRY; + + unlock_kernel(); + + memset(&dummy_oti, 0, sizeof(dummy_oti)); + + obd = exp->exp_obd; + ec = &obd->u.echocli; + + switch (cmd) { + case OBD_IOC_CREATE: /* may create echo object */ + if (!capable (CAP_SYS_ADMIN)) + GOTO (out, rc = -EPERM); + + rc = echo_create_object (obd, 1, &data->ioc_obdo1, + data->ioc_pbuf1, data->ioc_plen1, + &dummy_oti); + GOTO(out, rc); + + case OBD_IOC_DESTROY: + if (!capable (CAP_SYS_ADMIN)) + GOTO (out, rc = -EPERM); + + rc = echo_get_object (&eco, obd, &data->ioc_obdo1); + if (rc == 0) { + oa = &data->ioc_obdo1; + oa->o_gr = FILTER_GROUP_ECHO; + oa->o_valid |= OBD_MD_FLGROUP; + rc = obd_destroy(ec->ec_exp, oa, eco->eco_lsm, + &dummy_oti); + if (rc == 0) + eco->eco_deleted = 1; + echo_put_object(eco); + } + GOTO(out, rc); + + case OBD_IOC_GETATTR: + rc = echo_get_object (&eco, obd, &data->ioc_obdo1); + if (rc == 0) { + rc = obd_getattr(ec->ec_exp, &data->ioc_obdo1, + eco->eco_lsm); + echo_put_object(eco); + } + GOTO(out, rc); + + case OBD_IOC_SETATTR: + if (!capable (CAP_SYS_ADMIN)) + GOTO (out, rc = -EPERM); + + rc = echo_get_object (&eco, obd, &data->ioc_obdo1); + if (rc == 0) { + rc = obd_setattr(ec->ec_exp, &data->ioc_obdo1, + eco->eco_lsm, NULL); + echo_put_object(eco); + } + GOTO(out, rc); + + case OBD_IOC_BRW_WRITE: + if (!capable (CAP_SYS_ADMIN)) + GOTO (out, rc = -EPERM); + + rw = OBD_BRW_WRITE; + /* fall through */ + case OBD_IOC_BRW_READ: + rc = echo_client_brw_ioctl(rw, exp, data); + GOTO(out, rc); + + case ECHO_IOC_GET_STRIPE: + rc = echo_get_object(&eco, obd, &data->ioc_obdo1); + if (rc == 0) { + rc = echo_copyout_lsm(eco->eco_lsm, data->ioc_pbuf1, + data->ioc_plen1); + echo_put_object(eco); + } + GOTO(out, rc); + + case ECHO_IOC_SET_STRIPE: + if (!capable (CAP_SYS_ADMIN)) + GOTO (out, rc = -EPERM); + + if (data->ioc_pbuf1 == NULL) { /* unset */ + rc = echo_get_object(&eco, obd, &data->ioc_obdo1); + if (rc == 0) { + eco->eco_deleted = 1; + echo_put_object(eco); + } + } else { + rc = echo_create_object(obd, 0, &data->ioc_obdo1, + data->ioc_pbuf1, + data->ioc_plen1, &dummy_oti); + } + GOTO (out, rc); + + case ECHO_IOC_ENQUEUE: + if (!capable (CAP_SYS_ADMIN)) + GOTO (out, rc = -EPERM); + + rc = echo_client_enqueue(exp, &data->ioc_obdo1, + data->ioc_conn1, /* lock mode */ + data->ioc_offset, data->ioc_count);/*extent*/ + GOTO (out, rc); + + case ECHO_IOC_CANCEL: + rc = echo_client_cancel(exp, &data->ioc_obdo1); + GOTO (out, rc); + + default: + CERROR ("echo_ioctl(): unrecognised ioctl %#x\n", cmd); + GOTO (out, rc = -ENOTTY); + } + + EXIT; out: - RETURN(rc); + + /* XXX this should be in a helper also called by target_send_reply */ + for (ack_lock = dummy_oti.oti_ack_locks, i = 0; i < 4; + i++, ack_lock++) { + if (!ack_lock->mode) + break; + ldlm_lock_decref(&ack_lock->lock, ack_lock->mode); + } + + lock_kernel(); + + return rc; } -static int echo_setup(struct obd_device *obddev, obd_count len, void *buf) +static int +echo_client_setup(struct obd_device *obddev, obd_count len, void *buf) { - struct obd_ioctl_data* data = buf; - struct echo_client_obd *ec = &obddev->u.echo_client; + struct lustre_cfg* lcfg = buf; + struct echo_client_obd *ec = &obddev->u.echocli; struct obd_device *tgt; + struct lustre_handle conn = {0, }; + struct obd_uuid echo_uuid = { "ECHO_UUID" }; int rc; ENTRY; - if (data->ioc_inllen1 < 1) { - CERROR("requires a TARGET OBD UUID\n"); + if (lcfg->lcfg_inllen1 < 1) { + CERROR("requires a TARGET OBD name\n"); RETURN(-EINVAL); } - if (data->ioc_inllen1 > 37) { - CERROR("OBD UUID must be less than 38 characters\n"); + + tgt = class_name2obd(lcfg->lcfg_inlbuf1); + if (!tgt || !tgt->obd_attached || !tgt->obd_set_up) { + CERROR("device not attached or not set up (%s)\n", + lcfg->lcfg_inlbuf1); RETURN(-EINVAL); } - MOD_INC_USE_COUNT; - tgt = class_uuid2obd(data->ioc_inlbuf1); - if (!tgt || !(tgt->obd_flags & OBD_ATTACHED) || - !(tgt->obd_flags & OBD_SET_UP)) { - CERROR("device not attached or not set up (%d)\n", - data->ioc_dev); - GOTO(error_dec, rc = -EINVAL); - } + spin_lock_init (&ec->ec_lock); + INIT_LIST_HEAD (&ec->ec_objects); + ec->ec_unique = 0; - rc = obd_connect(&ec->conn, tgt, NULL, NULL, NULL); + rc = obd_connect(&conn, tgt, &echo_uuid, NULL, FILTER_GROUP_ECHO); if (rc) { - CERROR("fail to connect to device %d\n", data->ioc_dev); - GOTO(error_dec, rc = -EINVAL); + CERROR("fail to connect to device %s\n", lcfg->lcfg_inlbuf1); + return (rc); } - RETURN(rc); -error_dec: - MOD_DEC_USE_COUNT; + ec->ec_exp = class_conn2export(&conn); + RETURN(rc); } -static int echo_cleanup(struct obd_device * obddev) +static int echo_client_cleanup(struct obd_device *obddev, int flags) { - struct echo_client_obd *ec = &obddev->u.echo_client; + struct list_head *el; + struct ec_object *eco; + struct echo_client_obd *ec = &obddev->u.echocli; int rc; ENTRY; @@ -237,40 +1370,99 @@ static int echo_cleanup(struct obd_device * obddev) RETURN(-EBUSY); } - rc = obd_disconnect(&ec->conn); - if (rc) { + /* XXX assuming sole access */ + while (!list_empty(&ec->ec_objects)) { + el = ec->ec_objects.next; + eco = list_entry(el, struct ec_object, eco_obj_chain); + + LASSERT(eco->eco_refcount == 0); + eco->eco_refcount = 1; + eco->eco_deleted = 1; + echo_put_object(eco); + } + + rc = obd_disconnect(ec->ec_exp, 0); + if (rc != 0) CERROR("fail to disconnect device: %d\n", rc); - RETURN(-EINVAL); + + RETURN(rc); +} + +static int echo_client_connect(struct lustre_handle *conn, + struct obd_device *src, + struct obd_uuid *cluuid, + struct obd_connect_data *data, + unsigned long flags) +{ + struct obd_export *exp; + int rc; + + rc = class_connect(conn, src, cluuid); + if (rc == 0) { + exp = class_conn2export(conn); + INIT_LIST_HEAD(&exp->exp_ec_data.eced_locks); + class_export_put(exp); } - MOD_DEC_USE_COUNT; - RETURN(0); + RETURN (rc); } -static int echo_connect(struct lustre_handle *conn, struct obd_device *src, - obd_uuid_t cluuid, struct recovd_obd *recovd, - ptlrpc_recovery_cb_t recover) +static int echo_client_disconnect(struct obd_export *exp, + unsigned long flags) { - return class_connect(conn, src, cluuid); + struct obd_device *obd; + struct echo_client_obd *ec; + struct ec_lock *ecl; + int rc; + ENTRY; + + if (exp == NULL) + GOTO(out, rc = -EINVAL); + + obd = exp->exp_obd; + ec = &obd->u.echocli; + + /* no more contention on export's lock list */ + while (!list_empty (&exp->exp_ec_data.eced_locks)) { + ecl = list_entry (exp->exp_ec_data.eced_locks.next, + struct ec_lock, ecl_exp_chain); + list_del (&ecl->ecl_exp_chain); + + rc = obd_cancel(ec->ec_exp, ecl->ecl_object->eco_lsm, + ecl->ecl_mode, &ecl->ecl_lock_handle); + + CDEBUG (D_INFO, "Cancel lock on object "LPX64" on disconnect " + "(%d)\n", ecl->ecl_object->eco_id, rc); + + echo_put_object (ecl->ecl_object); + OBD_FREE (ecl, sizeof (*ecl)); + } + + rc = class_disconnect(exp, 0); + GOTO(out, rc); + out: + return rc; } static struct obd_ops echo_obd_ops = { - o_setup: echo_setup, - o_cleanup: echo_cleanup, - o_iocontrol: echo_iocontrol, - o_connect: echo_connect, - o_disconnect: class_disconnect + .o_owner = THIS_MODULE, + .o_setup = echo_client_setup, + .o_cleanup = echo_client_cleanup, + .o_iocontrol = echo_client_iocontrol, + .o_connect = echo_client_connect, + .o_disconnect = echo_client_disconnect }; int echo_client_init(void) { - extern struct lprocfs_vars status_class_var[]; + struct lprocfs_static_vars lvars; - return class_register_type(&echo_obd_ops, status_class_var, + lprocfs_init_vars(echo, &lvars); + return class_register_type(&echo_obd_ops, NULL, lvars.module_vars, OBD_ECHO_CLIENT_DEVICENAME); } -void echo_client_cleanup(void) +void echo_client_exit(void) { class_unregister_type(OBD_ECHO_CLIENT_DEVICENAME); }