CVE-2026-74591: mm/filemap: __filemap_add_folio() restore index before retrying

In the Linux kernel, the following vulnerability has been resolved:

mm/filemap: filemapaddfolio() restore index before retrying

In filemapaddfolio()'s split-a-conflict loop, xassetorder() is applied repeatedly: each application modifies xas.xaindex, rounding it down according to the splitorder attempted at that stage: and if all goes as intended, it eventually (or immediately) converges on an xastrysplit() to the required folioorder, with xas.xaindex now the same as index: then xasstore() puts the new folio into the xarray there.

But if a new node was needed, and GFPNOWAIT allocation did not get one, the lock is dropped, xasnomem() used to allocate, and sequence retried. If (that part of) the xarray is unchanged when the lock is reacquired, no problem. But what if the conflict was meanwhile resolved by another thread (perhaps even doing the same thing, inserting a folio at that same index)? Isn't there a danger of now putting our folio into the xarray at an intermediate rounded-down index? With !foliocontains() bug to follow, when CONFIGDEBUGVM=y is checking for that.

Fix this with an xassetorder() to restore the original xas.xaindex at the bottom of the loop, so the retry does a full re-evaluation after reacquiring the lock, and cannot reach xasstore() with the wrong index.

Production was suffering from rare SIGILLs and SIGSEGVs, executable text found a page away from where it belonged, !foliocontains() bug hit when debug enabled: symptoms not seen since this patch went in.